diff --git a/q103/random_k2/question.json b/q103/random_k2/question.json index 69ac8a0fde54103fa8d00fdd53f02e8715ab5f91..af6807e45631a71bd68825fd0f5ffc85f47774cd 100644 --- a/q103/random_k2/question.json +++ b/q103/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "e639029d16094ea71d964e2fb953952b.pdf", - "web_0867a04cdab7a273.pdf", - "web_a39c0d7f5ff0dedb.pdf" + "389c76c12bae49339e801c93d7587859e248a538.pdf", + "web_ae1b8a373064ff5e.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q103/random_k2/random_1.md b/q103/random_k2/random_1.md index 85e510cedc0e76a41806e14e4f5679216b834d6f..c230816f9b1098ca6662073190a692ee54d16a56 100644 --- a/q103/random_k2/random_1.md +++ b/q103/random_k2/random_1.md @@ -1,1480 +1,3444 @@ -Historic Architectural Resources Survey Report - -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 - -North Carolina Department of Transportation -Report Prepared by Nancy Van Dolsen - -November 1999 - - Historic Architectural Resources Survey Report - -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 - -November 1999 - -Principal Investigator (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation - -441-1/4,4 (cid:9) - -Barbara Church, Supervisor (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation - -Date - -//7/67/1. -Date - - TABLE OF CONTENTS - -Project Description (cid:9) - -Purpose of Survey and Report (cid:9) - -Methodology (cid:9) - -Summary Findings of the Survey (cid:9) - -Historic Contexts and Background Information (cid:9) - -Property Inventory and Evaluations - -William Jones Arney House (cid:9) - -Marshall Arney House (cid:9) - -Corpening, House (cid:9) - -Early Twentieth-Century Frame House (cid:9) - -Nineteenth-Century Log House (cid:9) - -Bibliography (cid:9) - -Appendix A: Correspondence with the State Historic Preservation Office (cid:9) - -LIST OF FIGURES - -Figure 1: Project Location Map (cid:9) - -Figure 2: Area of Potential Effects with Historic Properties Noted (cid:9) - -Figure 3: Tax Map, William Jones Arney House (cid:9) - -Figure 4: Site Plan, William Jones Arney House (cid:9) - -Figure 5: Tax Map, Marshall Arney House (cid:9) - -Figure 6: Site Plan, Marshall Arney House (cid:9) - -Page - - 1 - -1 - +| | | When Can | LLMs | Learn | to Reason | with | Weak | Supervision? | | +| --- | --- | -------- | ---- | ----- | --------- | ---- | ---- | ------------ | --- | +SalmanRahman12* JingyanShen2* AnnaMordvina2 HamidPalangi3 SaadiaGabriel1 PavelIzmailov2 +| | | ❖Projectpage: | | salmanrahman.net/rlvr-weak-supervision | | | | | | +| ----- | -------- | ------------- | ------------- | -------------------------------------- | --- | --------------------------------------------------- | --- | ---------------------- | --- | +| | | Abstract | | | | pabilitiesinlargelanguagemodels(Guoetal.,2025;Jaech | | | | +| | | | | | | etal.,2024;Teametal.,2025). | | Withonlybinaryfeedback | | +| Large | language | models | have achieved | signifi- | | | | | | +6202 rpA 02 ]GL.sc[ 1v47581.4062:viXra +cantreasoningimprovementsthroughreinforce- oncorrectness,RLVRhasenabledsubstantialgainsacross +diversereasoningtaskswithoutrequiringdensesupervision. +| ment | learning | with verifiable | rewards | (RLVR). | | | | | | +| ---- | -------- | --------------- | ------- | ------- | --- | --- | --- | --- | --- | +Yetasmodelcapabilitiesgrow,constructinghigh- However,recentfindingssuggesttheseimprovementsmay +qualityrewardsignalsbecomesincreasinglydif- bedrivenbyfactorsotherthantheintegrationofcorrectness +| | | | | | | signals. | SomestudiesreportthatRLVRsucceedsevenun- | | | +| ------- | ------ | ------------ | ------------- | ---- | --- | -------- | ---------------------------------------- | --- | --- | +| ficult, | making | it essential | to understand | when | | | | | | +RLVR can succeed under weaker forms of su- derextremeconditions: trainingonjustasingleexample +canyieldsignificantgains(Wangetal.,2025a),andrandom +| pervision. | | We conduct | a systematic | empirical | | | | | | +| ---------- | --- | ---------- | ------------ | --------- | --- | --- | --- | --- | --- | +studyacrossdiversemodelfamiliesandreasoning orincorrectrewardssometimesmatchground-truthperfor- +domains under three weak supervision settings: mance (Shao et al., 2025). Other work shows that proxy +signalssuchasself-certainty(Zhaoetal.,2025;Prabhudesai +| scarce | data, | noisy rewards, | and self-supervised | | | | | | | +| ------ | ----- | -------------- | ------------------- | --- | --- | --- | --- | --- | --- | +proxy rewards. We find that generalization is etal.,2025),entropyminimization(Agarwaletal.,2025), +majorityvoting(Zuoetal.,2025),orself-generatedtraining +governedbytrainingrewardsaturationdynamics: +models that generalize exhibit a prolonged pre- data(Huangetal.,2025)canreplaceverifiablerewards. +saturationphaseduringwhichtrainingrewardand +Furthermore,techniquesthatsucceedononemodelfamily +downstream performance climb together, while oftenfailonothers(Shaoetal.,2025),underreportedbase- +modelsthatsaturaterapidlymemorizeratherthan +linesmayinflateperceivedbenefits(Chandaketal.,2025), +learn. Weidentifyreasoningfaithfulness,defined and prolonged training with proxy rewards (i.e., reward +astheextenttowhichamodel’sintermediatesteps +| | | | | | | signals derived | from | model outputs without | ground-truth | +| --- | --- | --- | --- | --- | --- | --------------- | ---- | --------------------- | ------------ | +logicallysupportitsfinalanswer,asthepre-RL +verification)canleadtorewardhackingandperformance +propertythatpredictswhichregimeamodelfalls collapse(Shafayatetal.,2025). Thesemixedresultsleavea +into,whileoutputdiversityaloneisuninformative. +| | | | | | | fundamentalquestion: | | WhencanRLVRgeneralize1under | | +| --- | --- | --- | --- | --- | --- | -------------------- | --- | --------------------------- | --- | +Motivatedbythesefindings,wedisentanglethe weaksupervision,andwhatdeterminessuccessorfailure? +contributionsofcontinualpre-trainingandsuper- +visedfine-tuning,findingthatSFTonexplicitrea- UnderstandingwhenRLVRworksunderweaksupervision +mattersforpractice.Ground-truthverifiersareoftenlimited: +soningtracesisnecessaryforgeneralizationunder +weaksupervision,whilecontinualpre-trainingon labelsmaybenoisyorunavailable,andasmodelsbecome +domaindataamplifiestheeffect.Appliedtogether strongerthantheirsupervisors,alternativerewardsignals +becomenecessary(Burnsetal.,2023). +toLlama3.2-3B-Base,theseinterventionsenable +generalizationacrossallthreesettingswherethe +WeconductasystematicempiricalstudyofRLVRunder +basemodelpreviouslyfailed. weak supervision across two model families (Qwen and +| | | | | | | Llama), | and three | reasoning domains (MATH, | SCIENCE, | +| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------------- | ------------------------ | -------- | +| | | | | | | andGRAPH). | Ourworkisorganizedaroundthreequestions: | | | +1.Introduction +| | | | | | | • RQ1(WeakSupervision):DoesRLVRgeneralizeacross | | | | +| --- | --- | --- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | +Reinforcementlearningwithverifiablerewards(RLVR)has +modelfamiliesanddomainsunderscarcedata,noisyre- +emergedasapowerfulparadigmforimprovingreasoningca- +wards,andself-supervisedproxyrewards? +*Equal 1University • RQ2(ModelProperties): Whatpre-RLmodelproper- +| | contribution | | of | California, | Los An- | | | | | +| ----- | ------------ | --------------- | -------- | -------------- | ------- | --- | --- | --- | --- | +| | 2New | | 3Google. | | | | | | | +| geles | | York University | | Correspondence | | | | | | +1Throughout,weusegeneralizationtomeanimprovementon +| to: Salman | Rahman | , | | Jingyan | Shen | | | | | +| ---------- | ------ | --------------------- | --- | ------- | ---- | --- | --- | --- | --- | +downstreamevaluationbenchmarks,bothin-domainheld-outsets +. +andout-of-domaintransfer,followingRLtraining. +Preprint.April21,2026. 1 -4 - - 6 - -8 - - 15 +LLMReasoningwithWeakSupervision +tiesdeterminewhetheramodelgeneralizesunderweak 2.ExperimentalSetup +supervision? +| | | | | | | | We evaluate | the | following | model families: | | (1) Qwen2.5- | | +| -------------------- | --- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | ----------------------------------- | --------------- | --- | ------------ | --- | +| • RQ3(Intervention): | | | Howcanweenablegeneralization | | | | | | | | | | | +| | | | | | | | 1.5B/3B(Base): | | General-purposemodelspretrainedon18 | | | | | +inmodelsthatfailunderweaksupervision? +trilliontokens(Team,2024);(2)Qwen2.5-Math-1.5B/7B +Ourinvestigationuncoversthreefindings. First,general- (Math-specialized): BuiltuponQwen2.5withanadditional +izationunderweaksupervisionisgovernedbytraining 1trillionmath-relatedtokens(Yangetal.,2024);(3)Llama- +rewardsaturationdynamics. Modelsthatgeneralizeex- 3.2-3B/8B-Instruct(Instruction-tuned): Pretrainedon9 +hibitaprolongedpre-saturationphaseduringwhichtraining trilliontokensandalignedviaSFT,rejectionsampling,and +| | | | | | | | DPO(Dubeyetal.,2024). | | | WeusetheInstructvariantsfor | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | --------------------------- | --- | --- | --- | +rewardclimbssteadilyandthemodellearnstransferablerea- +soningpatterns;modelsthatfailsaturaterapidlyandentera Llamabecausethebasemodelsdonotreliablyfollowthe +post-saturationphasewherefurthertrainingyieldsdiminish- required formatfor on-policy rollouts. We revisit Llama- +ingreturns. Whichregimeamodelfallsintodependsonits Basein§4,whereSFThandlestheformat-followingissue. +| pretrainingpriors: | | modelswithstrongdomain-alignedpre- | | | | | | | | | | | | +| ------------------ | --- | ---------------------------------- | --- | --- | --- | --- | ------------------- | --- | ----------------------------- | --- | --- | --- | --- | +| | | | | | | | DomainsandDatasets. | | Weselectthreedomainswithvary- | | | | | +training(QwenonMATHandSCIENCE)sustainextended +| | | | | | | | inglevelsofpretrainingexposure: | | | | MATH(highexposure), | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | ------------------- | --- | --- | +pre-saturationphasesandgeneralizeunderscarcedata,noisy +SCIENCE(moderatecoverage)andGRAPHtasks(underrep- +rewards,andself-supervisedproxyrewards,whilemodels +| | | | | | | | resentedintypicalpretrainingcorpora). | | | | WeuseSkywork- | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | ------------- | --- | --- | +withoutsuchpriors(Llamaacrossalldomains,andQwenon +OR1(Heetal.,2025a)forMATH,SCPdatasets(Liuetal., +GRAPH)saturaterapidlyandfailtogeneralizeevenunder +| | | | | | | | 2025a; Lu | et al., | 2025) spanning | physics, | | chemistry, | and | +| ------------------- | --- | -------------------------------- | --- | --- | --- | --- | --------- | ------- | -------------- | -------- | --- | ---------- | --- | +| moderatelabelnoise. | | Wetreatthemodel-familycontrastas | | | | | | | | | | | | +biologyforSCIENCE,andtasksfromReasoningGym(Sto- +aproxyforpretraining-priorstrengthratherthananintrin- +janovskietal.,2025)involvingdiscretealgorithmicreason- +sicpropertyofeitherfamily,areadingthat§4confirmsby +| | | | | | | | ingforGRAPH. | ForMathandScience,weusethe1.5B/3B | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --------------------------------- | --- | --- | --- | --- | --- | +showingthatcontinualpre-trainingonmathdatatransforms +modelsasourprimaryexperimentsandadditionallyevalu- +Llama’sRLbehaviortoresembleQwen’s. +ate7B/8Bmodelstoverifythatourfindingsholdatlarger +Second,reasoningfaithfulness,notoutputdiversity,dis- scale. For GRAPH, we only use the 7B/8B variants be- +tinguishesmodelsthatgeneralizefrommodelsthatmem- cause the smaller models achieve solve@16 = 0, leaving +orize. Anaturalhypothesisforrapidsaturationisthatfail- noinformativesignalforRL.Moredetailsareprovidedin +| ing models | lack | exploratory | | capacity. | We find | the oppo- | AppendixB. | | | | | | | +| ---------- | ---- | ----------- | --- | --------- | ------- | --------- | ---------- | --- | --- | --- | --- | --- | --- | +site: Llamamodelsreachperfecttrainingrewardfasterthan +| | | | | | | | Model-AwareDataFiltering. | | | Toensureinformativetrain- | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | ------------------------- | --- | --- | --- | +Qwenandmaintainhigheroutputdiversitythroughouttrain- +ingsignals,weimplementmodel-specificdifficultyfiltering. +| ing, yet | they generalize | | poorly. | The | missing | property is | | | | | | | | +| -------- | --------------- | --- | ------- | --- | ------- | ----------- | --- | --- | --- | --- | --- | --- | --- | +Foreachproblem,wesample16responsesandcountcorrect +reasoningfaithfulness,definedbywhetheramodel’sinter- +| | | | | | | | solutions | (solve@16 | ∈ [0,16]). | We | retain | only problems | | +| ------------------------------------------- | --- | --- | --- | --- | --- | ---------- | --------- | --------- | ---------- | --- | ------ | ------------- | --- | +| mediatestepslogicallysupportitsfinalanswer. | | | | | | Modelsthat | | | | | | | | +wheresolve@16∈[1,15],effectivelydiscardinginstances +| saturate | rapidly | produce | correct | answers | through | reason- | | | | | | | | +| -------- | ------- | ------- | ------- | ------- | ------- | ------- | --- | --- | --- | --- | --- | --- | --- | +thatareeithertrivialorintractableforthemodel,stratified +ingchainsthatdonotjustifythem,memorizingratherthan +| | | | | | | | equally | across difficulty | levels | (details | in | Appendix | B.2). | +| --------- | --------- | --- | ---------------- | --- | ---- | ---------- | ------- | ----------------- | ------ | -------- | --- | -------- | ----- | +| learning. | Diversity | is | only informative | | when | considered | | | | | | | | +Thisfilteredsetservesasthecandidatepoolforallweak +jointlywithfaithfulness. +supervisionsettingsstudiedinthiswork;wedescribehow +Third,SFTonexplicitreasoningtracesisnecessaryfor trainingdataisconstructedfromthispoolforallsettings +| generalizationunderweaksupervision, | | | | | andcontinual | | in§3. | | | | | | | +| ---------------------------------------------- | --------- | --- | --- | ------- | ------------ | ---------- | ---------------------- | --- | ----------------------- | --- | --- | --- | --- | +| pre-training | amplifies | | the | effect. | We run a | controlled | | | | | | | | +| | | | | | | | TrainingConfiguration. | | WeuseGRPO(GroupRelative | | | | | +| comparisonthatdisentanglesthetwointerventions, | | | | | | train- | | | | | | | | +PolicyOptimization)asourRLalgorithm(Shaoetal.,2024). +| ing Llama3.2-3B | | Base, | a continually | | pre-trained | variant | | | | | | | | +| --------------- | --- | ----- | ------------- | --- | ----------- | ------- | --- | --- | --- | --- | --- | --- | --- | +ForeachqueryqsampledfromtrainingdatasetsD,agroup +| (CPT, ours), | and | Instruct, | each | with | either Thinking | SFT | | | | | | | | +| --------------------------------------------------- | --- | --------- | ---- | ---- | --------------- | --- | ----------------------- | ----------- | ---- | ----------------------- | --- | ------------- | --- | +| | | | | | | | ofindividualresponses{o | | }G | aresampledfromthepolicy | | | | +| (explicitreasoningtraces)orNon-ThinkingSFT(finalso- | | | | | | | | | i | i=1 | | | | +| | | | | | | | π before | the update. | GRPO | maximizes | | the following | | +θold +| lutionsonly). | ThinkingSFTisnecessary: | | | | itimprovesrea- | | | | | | | | | +| ------------- | ----------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | +objective: +soningfaithfulness,extendsthepre-saturationphase,and (cid:104) +(θ)=E +enablesgeneralizationunderallthreeweaksupervisionset- J GRPO (q,a)∼D,{oi}G +i=1 ∼πθold (·|q) +| | | | | | | | G | |oi| | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ---------- | ---------- | ------------- | ---------- | ----------- | --- | ------- | +| tings,whileNon-ThinkingSFTonthesamepromptsfails. | | | | | | | 1 (cid:88) | 1 (cid:88) | | | | | | +| | | | | | | | | | min (cid:0) ρ | Aˆ ,clip(ρ | ,1−ϵ,1+ϵ)Aˆ | | (cid:1) | +Continualpre-trainingisamultiplierratherthanasubstitute. i,t i i,t i +| | | | | | | | G | |o | | | | | | | +| ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | ------ | --------- | --- | --- | +| | | | | | | | i=1 | i t=1 | | | | | | +| CPTcombinedwithThinkingSFTproducesthestrongest | | | | | | | | | | | (cid:105) | | | +| | | | | | | | | | −βD | (π ||π | ) , | | | +generalization, recovering performance in settings where KL θ ref +Llamapreviouslyfailed. +πθ(oi,t|q,oi,0 +1 1 sat +(effectivepre-saturationlearning)whileredmarkrapidsaturationt(8) < 100. Thelarge-smallgapatsaturationstepsG(n1,8) and +sat sat,in +G(n1,8) aregenerallysmall.Resultsonmorebenchmarksandpass@kmetricsarereportedinTable3-7inAppendix. +sat,ood +In-domainBenchmarks OODBenchmark +Model t(8) +sat +∆( +s +8 +a +) t -i - -T - +∆∗ +p +( +o +8 +s +) t -e -e -h -S - -: -§ +∆( +s +8 +a +) +t +∆∗ +p +( +o +8 +s +) +t +Gsat,in ∆( +s +8 +a +) +t +∆∗ +p +( +o +8 +s +) +t +Gsat,ood +TrainingDomain:Math MATH500 AMC G(2048,8) SCP-Hard G(2048,8) +sat,in sat,ood +Qwen2.5-Math-1.5B 302 29.7 1.5 18.7 0.6 -1.1 10.5 2.1 2.4 +Qwen2.5-1.5B 170 32.1 0.9 12.7 3.3 -0.5 7.0 0.3 -0.4 +Llama3.2-3B-Instruct 55 10.8 -1.9 8.8 -2.1 -0.9 3.9 0.0 1.5 +TrainingDomain:Science SCP-Hard GPQA-Diamond G(2048,8) MATH500 G(2048,8) +sat,in sat,ood +Qwen2.5-Math-1.5B 268 14.5 1.1 16.9 1.6 1.1 25.3 0.8 1.1 +Qwen2.5-1.5B 161 6.4 0.2 13.3 1.7 1.8 32.3 2.1 1.2 +Llama3.2-3B-Instruct 61 1.8 1.7 11.9 3.0 5.1 7.3 2.2 0.6 +TrainingDomain:Graph QuantumLock LargestIsland +G(n1,8)† +MATH500 +G(n1,8)† +sat,in sat,ood +Qwen2.5-Math-7B 150 8.3 4.9 19.8 1.9 -1.8 21.0 2.1 -3.7 +Llama3.1-8B-Instruct 29 10.1 7.1 1.8 1.0 3.0† 9.1 3.8 0.0† +acrossdifficultylevelsdefinedin§2. ForN < 64,were- tional gain after saturation, defined as ∆∗(n)(M) := +post +peatpromptsuniformlytoreachbatchsize64(e.g.,N =8 +max +M(n)(t)−M(n)(cid:0) t(n)(cid:1) +. Values near zero +implies8repeats). +t∈[t( +s +n +at +),T] sat +indicatenegligiblepost-saturationgains. +Tostudytrainingdynamics,weleveragerewardsaturationto +• Large-small gap +G(n′,n)(M): +we define this gap as +distinguishperiodswherethepolicyimprovesonthetrain- sat +M(n′)(t(n))−M(n)(t(n))forn′ > n,whichcompares +ingdatasetfromthosewhereitplateaus. Intuitively,once sat sat +performancebetweenlarger(n′)andsmaller(n)datasets +trainingrewardsaturates,furtherupdatesyieldlittlenewsig- +(cid:104) (cid:105) atthesaturationstepofthesmallerrun. Atthesmaller +nal.Wedefiner¯ :=E 1 (cid:80)G r as +t q∼D,{oi}G i=1 ∼πold(·|q) G i=1 i run’ssaturationstep,howmuchbetterdoesthelargerrun +theexpectedtrainingrewardatupdatestept∈{1,...,T}, perform? Largerpositivevaluesindicatesubstantialben- +andletr¯ max :=max 1≤t≤T r¯ t bethemaximumrewardob- efitfrommoredata;valuesnearzerosuggestlimitedad- +servedduringtraining. Weidentifytraininghassaturated vantagefromincreasingdatasetsize. WedenoteG(n1,8) +sat,in +oncetherewardisclosetothismaximum,anddefinethe +astheaveragegapoverthein-domainbenchmarks,and +saturationstepastheearliestupdatewherethisoccurs: G(n1,8) astheaveragegapoverOODbenchmarks. +sat,ood +(cid:110) +t sat :=inf t∈{1,...,T eff }:r¯ t ≥ϵ max r¯ max }. Pre-saturationphasedominatessmall-samplelearning, +and its length predicts generalization. Table 1 sum- +Weuseϵ =0.99andsetT =T−50,i.e.,wesearchfor +max eff marizes the proposed metrics across model families and +t onlyuptothefirstT updatestoavoidboundaryeffects +sat eff training domains when training on 8 examples. Results +neartheendoftraining.Wedefinethepre-saturationphase +onmorebenchmarksandpass@kmetricsareprovidedin +asallstepst∈{1,...,t −1}andpost-saturationphase +sat Appendix C.2 and Tables 3-7. All model-domain pairs +asallstepst∈{min(t sat ,T),...,T}. showclearlypositive∆(8)forallmetrics(i.e.,bothavg@16 +sat +To quantify data efficiency, we introduce three metrics. andpass@k,k ∈ {4,8,16})acrossin-domainandout-of- +LetM(n)(t)denoteanevaluationmetric(e.g.,avg@16on domain benchmarks, indicating that as few as 8 training +MATH-500)attrainingsteptfortrainingwithnsamples, examples can trigger measurable learning during the pre- +andt(n)bethecorrespondingsaturationstep. saturation phase. Neither G(2048,8) nor G(2048,8) is sig- +sat sat,in sat,out +nificantly greater than zero on 7 out of 8 model-domain +• Pre-saturationgain∆(n)(M): performancegainfrom +sat pairs,indicatingthatthepre-saturationimprovementsare +initializationtosaturationas∆(n)(M):=M(n)(cid:0) t(n)(cid:1) +− oftencomparabletothoseobtainedwithlargertrainingsets. +sat sat +M(n)(0). Largerpositivevaluesindicateeffectivelearn- Thissuggeststhatearlylearningisnotstronglydata-limited. +ingbeforesaturation. Incontrast,thepost-saturationresidual∆∗(8) istypically +post +• Post-saturation residual ∆∗(n)(M): maximum addi- smallerthan∆(8),indicatingdiminishingreturnsoncethe +post sat +4 -9 +LLMReasoningwithWeakSupervision +Qwen2.5-Math-7B Llama-3.2-3B-Instruct gestthatevenformodelswithstrongmathematicalpriors, +| | | Graph | | | | Math | | | | | | | | | +| --- | --- | ----- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +thelackofdomain-specificpre-trainingacceleratessatura- +| | draweR gniniarT 1.0 | | | draweR gniniarT 0.75 | | | | | | | | | | | +| --- | ------------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.8 +| | | | | 0.60 | | | | tionandnecessitateshigherdatavolumetodrivelearning. | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | --- | --- | --- | +0.6 +| | | | | 0.45 | | | | Wefurtherprovideillustrationsfor7Band8Bmodelson | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | --- | --- | --- | +0.4 +| | | | | 0.30 | | | | MATHandSCIENCEdomainsinAppendixC.3. | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +| | 0.2 | | | 0.15 | | | | | | | | | | | +0.0 +)%( kcoL mutnauQ Extendedpre-saturationenablesout-of-domaintransfer. +| | 40 | | | )%( 005-HTAM 52 | | | | | | | | | | | +| --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Positive∆(8) +| | 30 | | | 48 | | | | | valuesinTable1indicatethatthereasoning | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------------- | --- | -------------- | --- | ----- | -------- | +| | 20 | | | | | | | | sat | | | | | | +| | | | | 44 | | | | patterns | learned during | the | pre-saturation | | phase | transfer | +10 +| | | | | 40 | | | | acrossdomains,particularlyforQwenmodels. | | | | | Withonly | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | -------- | --- | 0 -1 -2 +36 +| | | | | 28 | | | | 8samples,Qwen2.5-1.5BtrainedonMATHachievescon- | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | +)%( 005-HTAM 78 +| | | | | 24 | | | | sistent gains | on the | out-of-domain | | SCIENCE | benchmark | | +| --- | --- | --- | --- | ------- | --- | --- | --- | ------------- | ------ | ------------- | --- | ------- | --------- | --- | +| | 72 | | | )%( CMA | | | | | | | | | | | +| | 66 | | | 20 | | | | | | | | | | | +(SCP-Hard),whileQwen2.5-Math-7BtrainedonGRAPH +| | 60 | | | 16 | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +improvesout-of-domainMATH-500performanceby21.0% +| | 54 | | | 12 | | | | | | | | | | | +| --- | --- | --- | --- | ----- | --- | --- | --- | --------- | ------------ | ----- | ------ | ---- | ------- | ------- | +| | | | | | | | | (Fig. 1). | In contrast, | Llama | models | show | limited | out-of- | +| | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 | | | | | | | | +Training Steps Training Steps domain transfer even when in-domain performance im- +| | | =0 | | =0.3 | | =0.7 | | | | | | | | | +| --- | --- | --- | --- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +proves;theirgainsremainlocalizedtothespecifictraining +| | | =0.1 | | =0.5 | | =0.9 | | | | | | | | | +| --- | --- | ---- | --- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +distribution. +Figure2.Effectofrewardlabelcorruptionontrainingdynam- +icsandgeneralization.γdenotesthefractionoftrainingprompts +Takeaway: +| withcorruptedlabels,rangingfromclean(γ | | | | | | = 0)tomostlyin- | | | | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +(1)RLVRcangeneralizefromasfewas8samples +| correct(γ | | = 0.9). | | | | | | | | | | | | | +| --------- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ForQwenonGRAPHandLlamaonMATH, +generalizationdegradeatγ ≥ 0.5. when models remain in an extended pre-saturation +ForLlama,trainingreward +curvesstaycloseacrossallγ,suggestingoverfittingtonoise. phase,whereasrapidlysaturatingmodelsrequiresub- +stantiallymoredata.(2)Whetherscarce-datalearning +8-samplerunreachest(8). succeedsismodel-anddomain-dependent,reflecting +| | | | sat | | | | | theinfluenceofpretrainingpriors. | | | | (3)Inthelow-data | | | +| ---- | ------- | --- | -------- | ------ | ------ | ------------ | --- | -------------------------------- | --- | --- | --- | ---------------- | --- | --- | +| Fig. | 1 shows | the | training | curves | across | data scales. | The | | | | | | | | +regime,Llamamodelscanachieveperfecttrainingre- +lengthofthepre-saturationphaseistheprimarydeterminant wardsmuchfasterthanQwenbyrapidlymemorizing +ofwhetheramodelcangeneralize. With8trainingsamples, trainingexamplesbutachievelittlemeaningfultask +| Qwen2.5-Math-1.5B | | | on | MATH increases | | reward | steadily | | | | | | | | +| ----------------- | --- | --- | --- | -------------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | --- | +learning. +forover300steps;thissustainedascentallowsthemodelto +extractgeneralizablereasoningpatternsthattransfertoheld- +| out | evaluation | benchmarks | | such | as MATH-500 | and | SCP- | 3.2.NoisyRewards | | | | | | | +| --- | ---------- | ---------- | --- | ---- | ----------- | --- | ---- | ---------------- | --- | --- | --- | --- | --- | --- | +Hard. Awithin-familycomparisonisolatesthepretraining +effect: Qwen2.5-Math-1.5B,whichsharesarchitecturewith Whenground-truthverifiersareavailablebutimperfect,re- +| | | | | | | | | wardlabelsmaycontainerrors. | | | ToevaluateRLVRrobust- | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --------------------- | --- | --- | --- | +Qwen2.5-1.5Bbuthasadditionalmath-specificpretraining, +nesstosuchnoisysupervision,wevarythefractionofincor- +saturatesmoreslowlyandtransfersfurther(Table1). +| | | | | | | | | rectlabelsγ | byrandomlyreplacingground-truthanswers | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | -------------------------------------- | --- | --- | --- | --- | --- | +Figs. 13, 14, and 15 (Appendix C.1) show the full range with the most frequent incorrect answer produced by the +| N | ∈ {8,32,64,512,2048} | | | across | MATH, | SCIENCE, | and | | | | | | | | +| ---------------------------------------- | -------------------- | -------- | ------ | ------- | ----- | ------------ | --- | ---------------------------------- | ------------ | ---------- | --- | --------------- | ------ | --- | +| | | | | | | | | modelitself(detailsinAppendixD.1). | | | | Unlessotherwise | | | +| GRAPH. | | For Qwen | models | on MATH | | and SCIENCE, | in- | | | | | | | | +| | | | | | | | | noted,experimentsuseN | | =2048. | | | | | +| domainperformanceisnearlyindependentofN. | | | | | | ForLlama | | | | | | | | | +| | | | | | | | | RLVR | demonstrates | robustness | | to reward | noise, | but | +acrossalldomains,andforQwenonGRAPH,differentN +| | | | | | | | | generalizationvariesacrossmodels. | | | | Fig.2andAppendix | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ---------------- | --- | --- | +producesvisiblydifferentdynamicsonsomeoftheevals, +Fig.26summarizeperformanceacrosssevenmodel–domain +withsmallerdatasetssaturatingearlierandatlowerdown- +streamperformance. pairsundervaryingγ. Atγ ≤0.3,testperformanceacross +| | | | | | | | | mostsettingsremainsclosetothecleanrewards(γ | | | | | | =0),in- | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | --- | ------- | +Modelswithoutdomain-alignedpriorssaturaterapidly dicating robustness to moderate label noise. On MATH +| andfailtogeneralize. | | | Incontrast,Llamamodelsacrossall | | | | | | | | | | | | +| -------------------- | --- | --- | ------------------------------- | --- | --- | --- | --- | ------------ | ---- | ------ | -------- | ----- | ----- | ---- | +| | | | | | | | | and SCIENCE, | Qwen | models | maintain | gains | under | sub- | +domains,andQwenonGRAPH(Fig.1)exhibitcleardepen- stantial corruption (up to γ = 0.7). In contrast, Qwen +| denceondatascale. | | | ForLlama,trainingon8samplesleads | | | | | | | | | | | | +| ----------------- | --- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +onGRAPHandLlamaonMATHandSCIENCEdegradeat +torapidsaturation,witht(8)occurringwithinthefirst100: +| | | | | | | | | γ ≥ 0.5. | Higherγ | leadstoconsistentlylowertrainingre- | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ------- | ----------------------------------- | --- | --- | --- | --- | +sat +itmaximizesthetrainingrewardmuchfasterthantheQwen wardsthroughouttraining,butforLlamaonMATH,training +models. Thesemodelsrequirelargerdatasets(N ≥512)to rewardcurvesremainnearlyidenticalacrossallγ despite +achievemeaningfulgeneralization(detailsinAppendixC.1 severecorruption,indicatingLlamafitsincorrectanswers +| Fig.13andFig.14). | | | TheresultsintheGRAPHdomainsug- | | | | | | | | | | | | +| ----------------- | --- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | 5 -8 -. + +LLMReasoningwithWeakSupervision +| Qwen2.5-3B | Llama-3.2-3B-Instruct | | | | | | +| ------------------- | --------------------- | --- | --- | --- | --- | --- | +| Science | Science | | | | | | +| draweR gniniarT 1.0 | 1.0 | | | | | | +| 0.8 | 0.8 | | | | | | +| 0.6 | 0.6 | | | | | | +0.4 +0.4 +0.2 +0.2 +)%( 005-HTAM +| 65 | 48 | | | | | | +| --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | +| | 40 | Figure4.Evolutionofsemanticdiversityduring8-sampletrain- | | | | | +60 +| 55 | 32 | ingonMATH.Llamashowssignificantlyhigherpost-saturation | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | +| 50 | 24 | | | | | | +diversitythanQwen,albeitwithlowerperformanceoutcomes. +| 45 | 16 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | +40 +)%( draH-PCS 40 SCIENCE)showimprovementwithmajorityvoting,while +20 +| 32 | | othermodelsfailentirely. | | ForQwen2.5-3Bon | | SCIENCE, | +| --- | --- | ----------------------------------------------------- | --- | --------------- | --- | -------- | +| 24 | 15 | | | | | | +| | 10 | majorityvotingyieldstemporarygainsbeforecollapseafter | | | | | +16 +| 8 | 5 | | | | | | +| --- | --- | --------- | --------------------------------------- | --- | --- | --- | +| | | 500steps, | asthepolicyconvergestowardasingleoutput | | | | +| 0 | 0 | | | | | | +0 150300450600750 0 150300450600750 tomaximizeagreement. Self-certaintyrewardsleadtoper- +| Training Steps | Training Steps | | | | | | +| -------------- | -------------- | --- | --- | --- | --- | --- | +RLVR Majority Vote Self Certainty formancecollapseacrossallsettings. Theseresultsshow +thatcurrentself-supervisedproxyrewardsareinsufficient +| | | toreplaceverifiablefeedbackinmostsettings. | | | | (detailsin | +| --- | --- | ------------------------------------------ | --- | --- | --- | ---------- | +Figure3.Comparisonofrewardvariants(RLVR,self-certainty, +AppendixD.2andFig.27). +majorityvote)with1024trainingsamples.Proxyrewardswith- +outverifiersexhibitfailuremodesunderprolongedtraining:train- +ingcollapse(self-certainty)andrewardspikesfollowedbyperfor- Takeaway: Self-supervisedproxyrewardssucceed +mancedrops(majorityvote)(moreresultsareinAppendixD.2). onlyformath-specializedmodels(Qwen-Mathunder +| | | majorityvoting). | Othermodelsexhibitthesamepat- | | | | +| --- | --- | ---------------- | ----------------------------- | --- | --- | --- | +moreeasily. Wealsoobservethatmodel-domainpairswith ternweobservedin§3.1and§3.2: fastersaturation +faster saturation (§3.1) are generally less robust to label andweakerpretrainingpriorscoincidewithbrittle- +noise,aconnectionwedevelopin§3.4and§4. ness. Under prolonged training, the failure mode +| | | isrewardhacking: | | policiesconvergetowardoutputs | | | +| --- | --- | ---------------- | --- | ----------------------------- | --- | --- | +Takeaway: Robustnesstolabelnoisevariessharply thatmaximizetheproxywithoutcorrespondingdown- +| acrossmodel-domainpairs:QwenonMATHandSCI- | | streamgains. | | | | | +| ----------------------------------------- | --- | ------------ | --- | --- | --- | --- | +ENCE toleratesupto70%corruption, whileLlama +andQwenonGRAPHdegradeat50%.Model-domain +pairsthatsaturatefasterundercleanrewardsareless 3.4.WhyDoModelsFailUnderWeakSupervision? +robust,andLlamafitscorruptedlabelsnearlyasfast +| | | Theresultsin§3.1–§3.3showaconsistentpattern: | | | | models | +| --- | --- | -------------------------------------------- | --- | --- | --- | ------ | +ascleanones—evidencethatrapidsaturationreflects +| | | withstrongdomain-alignedpretraining(Qwenon | | | | MATH | +| --- | --- | ------------------------------------------ | --- | --- | --- | ---- | +memorizationcapacityratherthanlearningefficiency. +| | | and SCIENCE) | generalize | under weak | supervision, | while | +| --- | --- | -------------------------------- | ------------------- | ---------- | ------------ | ---------- | +| | | thosewithout(Llamaacrossdomains, | | | Qwenon | GRAPH) | +| | | fail. A | natural hypothesis, | motivated | by prior | work link- | +3.3.Self-SupervisedProxyRewards +ingdiminishedexploratorycapacitytorapidpolicysatura- +Whenground-truthverifiersareentirelyunavailable,models tion(Cuietal.,2025),isthatfailingmodelsproduceless +mustrelyonalternativerewardsignals(Burnsetal.,2023; diverse outputs. To test this, we analyze model behavior +Rahmanetal.,2025;Bowmanetal.,2022). Recentwork alongtwocomplementaryaxes: responsediversityandrea- +hasproposedself-supervisedproxyrewardsderivedfrom soningfaithfulness. Formaldefinitionsandimplementation +model outputs, but whether these approaches work well detailsareprovidedinAppendixF. +acrossmodelfamiliesandtaskdomainsremainsunexplored. +| | | To quantify | response | diversity, we | quantify semantic | di- | +| ------------------------- | ------------------------- | ----------- | -------- | ------------- | ----------------- | --- | +| Weevaluatetwosuchrewards: | self-certainty(Zhaoetal., | | | | | | +versitytocharacterizemeaningfulpatternsinthemodel’s +2025)andmajorityvote(Zuoetal.,2025)(implementation +reasoningratherthansurface-levelvariation(Farquharetal., +detailsinAppendixD.2). +2024;Lietal.,2025).Wemeasurediversityonthe8-sample +Proxyrewardstriggerrewardhackingandpolicycol- subsetoftheMATH,SCIENCEandGRAPHtrainingdatasets, +lapse. WhileRLVRtoleratesmoderatelabelnoiseinsome aswellasontheMATH-500evaluationdataset,overase- +model-domainpairs(§3.2),Fig.3showsthatfullyreplacing lectionofpromptsatvariousstepsthroughouttraining. For +verifiablefeedbackwithself-supervisedproxysignalsintro- each prompt, we cluster model responses using pairwise +ducesseverefailuresunderprolongedtraining. Onlymath- similarity judgments from an LLM judge and define the +specialized models (Qwen2.5-Math-1.5B on MATH and diversityscoreastheShannondiversityindexoverthere- 6 +LLMReasoningwithWeakSupervision +| | | | | | | | faithfulness. | Fig. 5 (right) | reports | faithful | diversity: | di- | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | -------------- | ------- | -------- | ---------- | --------- | +| | | | | | | | versitycomputedonlyoverfaithfulresponses. | | | | | Thisjoint | +measurerevealsaconsistentpatternacrossallthreedomains. +| | | | | | | | On MATH,Llama’sapparentdiversityadvantage(Fig. | | | | | 4) | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | +disappears—mostdiverseresponsesareunfaithful,andthe +| | | | | | | | faithfulsubsetisnarrow. | | OnSCIENCE,alignedproportions | | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | ---------------------------- | --- | --- | --- | +areuniformlyhighacrossmodels,maskingrealdifferences +inreasoningquality;faithfuldiversityseparatesthem,with +Qwen-Mathmaintainingthehighestvaluesthroughouttrain- +| | | | | | | | ing. OnGRAPH,Qwen-MathandLlamashowcomparable | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | +alignedproportions,butQwen-Mathsustainshigherfaithful +| | | | | | | | diversity. | Ineverycase,themodelthatgeneralizesbestin | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------------------------------------- | --- | --- | --- | --- | +§3.1istheoneexploringthewidestrangeoffaithfulrea- +Figure5.Evolutionofreasoningfaithfulness(oncorrectsam- soningpaths—nottheonewiththehighestrawdiversity, +ples)andfaithfuldiversityonmodelsthroughoutRLusing8 nor the one with the highest aligned proportion. Raw di- +samplesfromavarietyofdatasets.LlamamodelsintheMATH +versityoverstatesexploratorycapacity;alignedproportion +domainexhibitsignificantlylowerfaithfulnesscomparedtoQwen. +saturatesoneasierdomains;onlytheirintersectionpredicts +generalization. +| sultingclusters. | | SeeFigure31forthejudgemodelprompt. | | | | | | | | | | | +| ------------------------------------------- | --- | ---------------------------------- | --- | --- | --- | ----- | --------- | --- | --- | --- | --- | --- | +| Highdiversitydoesnotpreventrapidsaturation. | | | | | | Fig.4 | Takeaway: | | | | | | +reportstheevolutionofdiversityscoresformodelstrained +| | | | | | | | Low reasoning | faithfulness, | | not low | diversity, | ex- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------- | --- | ------- | ---------- | --- | +on 8 samples from the MATH training dataset, computed plains why models fail under weak supervision: +on the corresponding training set. Llama reaches reward rapidlysaturatingmodelsmemorizeanswersrather +saturation earlier and retains higher diversity than Qwen, than acquire transferable reasoning. Raw diversity +theoppositeofwhattheexploration-saturationhypothesis metricsaremisleading—Llamaexhibitshigherout- +predicts. DiversitycomputedontheMATH-500evaluation +| | | | | | | | put diversity | than Qwen | while | generalizing | | worse. | +| --- | --- | --- | --- | --- | --- | --- | ------------- | --------- | ----- | ------------ | --- | ------ | +datasetispresentedintheappendix(Fig.30). Diversitybecomesinformativeonlywhencomputed +overfaithfulresponses. +Sincediversityalonedoesnotexplainfailureunderweak +| supervision, | we | investigate | the faithfulness | | of a model’s | | | | | | | | +| ------------ | --------------------------------------- | ----------- | ---------------- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | +| reasoning. | Inspiredbypriorwork(Bakeretal.,2025),we | | | | | | | | | | | | +definearesponseasfaithfulifitsreasoningtracecontains Insummary,§3showsthatthesurprisingcapabilitiesoften +the information needed to justify the final answer and is attributedtoRLVR,suchaslearningfromscarcedata,tol- +logicallyconsistentwithit. Atagiventrainingstepandfor eratingnoisyrewards,succeedingwithoutverification,are +notuniversalbutdependonpre-RLreasoningfaithfulness. +agivenprompt,wecategorizeeachpolicyrolloutasaligned, +partiallyaligned,ormisalignedbasedonrubricsprovided §4takesupthenaturalquestion: canpre-RLinterventions +to an LLM-as-a-judge (see prompt in Fig. 32). We then targetingfaithfulnessextendthepre-saturationphaseand +computethepolicyfaithfulnessrateF (l)asthefractionof recovergeneralizationunderweaksupervision? +π +| responsesassignedtolabell. | | | AppendixFoutlinesresults | | | | | | | | | | +| -------------------------- | --- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +forinter-modelagreementonalignmentcategorizationto 4.ImprovingRLVRUnderWeakSupervision +evaluatethereliabilityofourLLM-as-a-judge. +viaPre-RLTraining +Modelswithrapidsaturationexhibitlowreasoningfaith- +Section3showedthatrapidsaturationandlowreasoning +fulness. Fig.5(left)showsthefractionofcorrectresponses +| | | | | | | | faithfulnessarelinked: | modelsthatgeneralizepoorlyunder | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------- | ------------------------------- | --- | --- | --- | --- | +thatarealignedoverRLtrainingacrossmodelsanddomains +weaksupervisionproducecorrectanswersthroughreason- +| studiedin§3.1. | | Onthe | MATH domain, | theLlamamodel | | | | | | | | | +| --------------------------------------- | ----- | --------- | -------------------------------- | ------------- | ------------ | --- | ------------------------------------------------------- | ------------ | --------------- | ---------------------- | ---- | ---------- | +| | | | | | | | ingthatdoesnotsupportthem. | | | Thisraisesacausalques- | | | +| shows much | lower | reasoning | faithfulness | during | training | | | | | | | | +| | | | | | | | tion. Iffaithfulnessdrivesthepre-saturationphase,andthe | | | | | | +| thantheQwenmodels. | | | ThisindicatesthatLlama’srapidre- | | | | | | | | | | +| | | | | | | | pre-saturation | phase drives | generalization, | | then | instilling | +| wardgainsdonotreflectimprovedreasoning: | | | | | asubstantial | | | | | | | | +faithfulnessbeforeRLshouldextendthephaseandrecover +fractionofcorrectanswersarememorized,withreasoning +| | | | | | | | generalization. | Wetestthisbyrunningacontrolledcom- | | | | | +| ----------- | ------ | ------- | ---------- | ----- | -------- | --- | --------------- | ---------------------------------- | --- | --- | --- | --- | +| traces that | do not | support | them. Fig. | 33 in | Appendix | F | | | | | | | +parisonofpre-RLinterventionsonLlama3.2-3B,themodel +| includes | additional | faithfulness | results | on these | domains, | | | | | | | | +| -------- | ---------- | ------------ | ------- | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +coveringproportionalignedandproportionmisalignedon thatfailedmostconsistentlyin§3. +correct,incorrectandallresponses. Westudytwoaxesofpre-RLtraining. Thefirstiscontinual +Reasoning diversity should be considered jointly with pre-training(CPT),extendedtrainingondomain-specific +7 + +LLMReasoningwithWeakSupervision +| | draweR gniniarT | | | )%( 005-HTAM | | | 32 | | | | | | +| ------ | --------------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | | | | | 56 | | | | | )%( draH-PCS | | | +| | | | | | | | )%( CMA | | | 15 | | | +| ecracS | | | | | | | 24 | | | | | | +| ataD | 0.6 | | | | 48 | | | | | | | | +| | | | | | 40 | | 16 | | | | | | +12 +18 +6 +| | 0.0 | | | | 12 | | 0 | | | 0 | | | +| -------- | --------------- | --- | --- | ------------ | ----- | --- | ------- | --- | --- | ------------ | --- | ------- | +| | 0 | 150 | 300 | 450 | 0 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | draweR gniniarT | | | )%( 005-HTAM | 64 | | | | | | | | +| | | | | | | | 30 | | | )%( draH-PCS | | | +| ytirojaM | | | | | 56 | | )%( CMA | | | | | | +etoV +| | 0.6 | | | | 48 | | | | | 15 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +20 +| | 0.0 | | | | 0 | | | | | 0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +60 +| draweR ysioN | draweR gniniarT | | | )%( 005-HTAM | | | | | | | | | +| ------------ | --------------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | 0.6 | | | | | | 25 | | | )%( draH-PCS | | | +| )7.0= | | | | | 54 | | )%( CMA | | | | | | +15 +| | | | | | 48 | | 20 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.3 +| ( | | | | | | | 15 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| | | | | | 15 | | 5 | | | | | | +| --- | --- | -------------- | --- | --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | | | | | 10 | | 0 | | | 0 | | | +| | 0 | 150 | 300 | 450 | 0 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure6.RLtrainingdynamicsandgeneralizationonMATHforLlama3.2-3BBase,CPT,andInstructvariantsunderdifferent +SFTinitializationsacrossthreeweaksupervisionsettings:scarcedata(N =8,top),majorityvote(middle),andnoisyreward +(γ =0.7,bottom).ThinkingSFT(solidlines)consistentlyprolongsthepre-saturationphaseandimprovesgeneralizationforbothCPT +andBasemodelscomparedtotheirNon-ThinkingSFTcounterparts(dashedlines)andtheInstructbaseline(dash-dot).CPT+Thinking +SFTachievesthestrongestperformanceacrossallsettings. +pretrainingtokenstostrengthenthepretrainingprior. The explicitreasoningtracesinfluencesubsequentRLdynamics. +second is supervised fine-tuning (SFT), with the specific WecomparetwoSFTregimesthatdifferonlyinwhether +questionofwhetherSFTonexplicitreasoningtracesdiffers thesupervisionincludesexplicitreasoning. Bothregimes +in its effect from SFT on final answers alone. Crossing use the same 43.5K math prompts and differ only in the +these axes gives a 2×2 design: two initializations (Base, targetoutput. Specifically,wesamplethesepromptsfrom +CPT)eachfollowedbytwoSFTregimes(Thinking,Non- OpenThoughts-114K (Guha et al., 2025), retaining only +Thinking).WeadditionallyincludeLlama3.2-3B-Instructas thosewhosereasoningtraceshavecorrectfinalanswersand +areference: itsharesthearchitectureofLlama3.2-3B-Base totallengthbelow8192tokens. +| but | has undergone | | extensive instruction | | tuning, | rejection | | | | | | | +| --- | ------------- | --- | --------------------- | --- | ------- | --------- | ------------------------------------------------- | --- | --- | --- | --- | --- | +| | | | | | | | • Non-thinkingSFT:Themodelissupervisedtooutputthe | | | | | | +sampling,andDPO,providingastrongoff-the-shelfbase- +finalsolutionwithoutgeneratingintermediatereasoning +| lineagainstwhichtojudgeourtargetedinterventions. | | | | | | We | | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +traces. +thenrunRLunderallthreeweaksupervisionsettingsfrom +| | | | | | | | • ThinkingSFT:Themodelistrainedonexplicit,verified | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------------- | --- | --- | --- | --- | --- | +§3: scarcedata,noisyrewards,andself-supervisedproxy +long-formreasoningtraces. +rewards. +Wefocusonthe MATH domainfortworeasons: Llama’s AtrainingexampleisshowninFig.12intheAppendix.The +| | | | | | | | SFTregimesarenear-iso-compute: | | | | ThinkingSFTtrainson | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | ------------------- | --- | +baselinefailureissharpestthere,providingthecleanesttest +roughly1Btokens,Non-ThinkingSFTonroughly0.27B, +ofwhetherpre-RLinterventionscanrecovergeneralization; +| | | | | | | | bothnegligiblerelativetothe52B-tokenCPTstage. | | | | | Differ- | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | ------- | +andhigh-qualitymathpretrainingcorpora(Nemotron-CC- +encesbetweenThinkingandNon-ThinkingSFTtherefore +Math)andreasoning-tracedatasets(OpenThoughts-114K) +areavailable,enablingtheinterventionsatsufficientscale. reflectthecontentofthesupervisionratherthanitscost. We +| | | | | | | | reporttheCPTlosscurveinAppendixFig. | | | | | 10andtheSFT | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | ----------- | +ContinualPre-Training(CPT).Wecontinuallypre-train losscurvesinFig. 11. +| Llama3.2-3B-Base | | | for one epoch | on | approximately | 52B | | | | | | | +| ---------------- | --- | --- | ------------- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | +ImplementationandtrainingdetailsofSFTareprovidedin +| math | tokens | from | the Nemotron-CC-Math | | dataset | (Ma- | | | | | | | +| ------------------- | ------ | ---- | ------------------------------- | --- | ------- | ---- | -------------------------------------------------- | --- | ---------------------------------- | --- | --- | --- | +| | | | | | | | AppendixB.6. | | ForthesubsequentRLphase,weevaluate | | | | +| habadietal.,2025)2. | | | TrainingdetailsareprovidedinAp- | | | | | | | | | | +| | | | | | | | acrossallthreeweaksupervisionsettings:scarcedata(N | | | | | = | +pendixB.5. +| | | | | | | | 8), noisy | rewards | (γ = | 0.7), and | self-supervised | proxy | +| ------------------- | --- | --- | ----------------------------- | --- | --- | --- | ---------------------- | ------- | ----------------------------- | --------- | --------------- | ----- | +| SFTTrainingRegimes. | | | FollowingCPTorBaseinitializa- | | | | | | | | | | +| | | | | | | | rewards(majorityvote). | | Allotherhyperparametersfollow | | | | +tion,weapplysupervisedfine-tuningtodeterminewhether theconfigurationsin§2,withthemaximumresponselength +duringRLextendedto8192tokenstoaccommodatelong- +2Nemotron-CC-Math-v1 8 -3 -4 -1 -R -S -& -4 -2 -4 -1 -R -S - -A -N -I -L -O -R -A -C -H -T -R -O -N -, +LLMReasoningwithWeakSupervision +formreasoningtraces. +4.1.Results +| Fig. 6 reports | RL | training | dynamics | for | the five | pre-RL | | | | | | | | +| -------------- | --- | -------- | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | +configurations(Base,CPT,andInstruct,withThinkingSFT +| or Non-Thinking | | SFT | applied to | Base | and CPT) | across | | | | | | | | +| -------------------------------- | ------ | --------- | ---------- | --------------- | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +| thethreeweaksupervisionsettings. | | | | Foreachsetting, | | we | | | | | | | | +| plot training | reward | alongside | three | downstream | | metrics: | | | | | | | | +Figure7.EvolutionofreasoningfaithfulnessoftheLlama3.2- +twoin-domain(MATH-500,AMC)andoneout-of-domain 3Bfamilyonweaksupervisiondomainswhencombinedwith +(SCP-Hard);additionalbenchmarksandpass@kresultsare continualpretrainingandSFTvariants.Whencombinedwith +Fig.34andFig.35inAppendixG.Wedrawthreefindings Thinking-SFT and CPT, the Llama3.2-3B-Base model exhibits +higherreasoningfaithfulness. +fromthisfigure,developedintheparagraphsbelow. +ThinkingSFTisnecessaryforsubstantiallearningunder +| | | | | | | | highest faithfulness | | among | all configurations, | | | consistent | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | ----- | ------------------- | --- | --- | ---------- | +weaksupervision.TheInstructbaselineisflatordecreasing +| | | | | | | | with its strongest | generalization | | | across | all weak | supervi- | +| --- | --- | --- | --- | --- | --- | --- | ------------------ | -------------- | --- | --- | ------ | -------- | -------- | +acrossallthreesettingsonalldownstreamevaluations— +| | | | | | | | sion settings. | Together | with | the | extended | pre-saturation | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | -------- | ---- | --- | -------- | -------------- | --- | +RLproducesnomeaningfulimprovementfromthisstarting +| | | | | | | | dynamics | visible in | Fig. | 6 (leftmost | column), | | this result | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ---- | ----------- | -------- | --- | ----------- | +point. ThinkingSFTistheonlyinterventionthatenables +| | | | | | | | supportsourhypothesisin§3.4: | | | | pre-RLinterventionsthat | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ----------------------- | --- | --- | +substantialdownstreamgainsonscarcedataandmajority +instillfaithfulnessproducelongerpre-saturationphasesand +vote,anditdoessoforbothBaseandCPTinitializations +recoveredgeneralization,inmodelsthatpreviouslyfailed. +| (solidblueandsolidred). | | | Non-ThinkingSFTshowsmodest | | | | | | | | | | | +| ----------------------- | ------------ | ------ | -------------------------- | ------- | ---------------- | ----- | --------- | ---------------------------------- | --- | --- | --- | --- | --- | +| gains only | when | paired | with CPT, | and | only under | noisy | | | | | | | | +| | | | | | | | Takeaway: | SFTonexplicitreasoningtraces,noton | | | | | | +| rewards; | Non-Thinking | | SFT on | Base is | flat or degrades | | | | | | | | | +finalanswers,isnecessaryforLlamatolearnsubstan- +acrossallthreesettings. +| | | | | | | | tiallyfromRLunderweaksupervision. | | | | | Itraisesrea- | | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ------------ | --- | +soningfaithfulness,extendsthepre-saturationphase, +| CPTamplifiestheThinkingSFTeffect. | | | | | ThinkingSFT | | | | | | | | | +| --------------------------------- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +onBasealoneproducesmodestgains. CombinedwithCPT, andrecoversgeneralizationunderscarcedata,noisy +| | | | | | | | rewards, | and self-supervised | | | proxy | rewards. | Con- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------- | --- | --- | ----- | -------- | ---- | +itproducessubstantiallylargergainsoneveryevaluation: +CPT+ThinkingSFTisthetop-performingcurveacrossall tinualpre-trainingamplifiestheeffectbutdoesnot +threeweaksupervisionsettingsandallthreeevals.TheCPT substituteforit: CPT+Non-ThinkingSFTfailsde- +| | | | | | | | spitematchedcompute. | | | Thestrongestconfiguration, | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | -------------------------- | --- | --- | --- | ++Non-ThinkingSFTcomparisonrulesoutacompute-based +explanation:thesame52BCPTtokens,pairedwithSFTtar- CPT + Thinking SFT, recovers performance in set- +tingswhereLlamahadpreviouslycollapsedentirely. +getsthatstripreasoningtraces,failtoenablegeneralization +| onscarcedataandmajorityvote. | | | | Theamplificationisspe- | | | | | | | | | | +| --------------------------------------------------- | --- | --- | ------------------------------- | ---------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | +| cifictothecombination: | | | extrapre-trainingcomputealoneis | | | | | | | | | | | +| insufficient;ThinkingSFTalonehelpsbutislimited,only | | | | | | | 5.RelatedWork | | | | | | | +thecombinationrecoversfullgeneralization. +| | | | | | | | RLVR for | Reasoning. | Reinforcement | | | learning | with ver- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ------------- | --- | --- | -------- | --------- | +Baseinitializationfailsundermostweaksupervisionset- ifiable rewards has emerged as an effective post-training +tingsregardlessofSFT.TheBasemodelshowsmeaningful +| | | | | | | | method for | improving | reasoning | | in large | language | mod- | +| --------------------------------- | --- | --- | --- | --- | ------------- | --- | ----------- | ---------- | --------- | ------- | -------- | -------- | ---------- | +| improvementonlyintwocombinations: | | | | | Base+Thinking | | | | | | | | | +| | | | | | | | els (Guo et | al., 2025; | Olmo | et al., | 2025; | Yu et | al., 2025; | +SFT under scarce data and majority vote, and even there Zengetal.,2025). RecentworkhasexploredwhenRLVR +| gains are | modest. | Under | noisy | rewards, | neither | Base + | | | | | | | | +| --------- | ------- | ----- | ----- | -------- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | +yieldsimprovements(Liuetal.,2025b;a;Huetal.,2025). +Thinking SFT nor Base + Non-Thinking SFT produces Wangetal.(2025a)demonstratethattrainingonasingleex- +| meaningfuldownstreamimprovement. | | | | ThisisolatesCPT’s | | | | | | | | | | +| -------------------------------- | --- | --- | --- | ----------------- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --------- | --- | +| | | | | | | | amplecanprovidemeaningfullearningsignals. | | | | | Otherwork | | +contribution:ThinkingSFTisnecessarybutnotsufficient— +exploresalternativerewards,includingself-certainty(Zhao +domain-alignedpretrainingisrequiredfortheintervention etal.,2025),majorityvoting(Zuoetal.,2025),negativesig- +togeneralizeacrossallthreeweaksupervisionsettings. +nals(Zhuetal.,2025),self-generatedtrainingdataHuang +| | | | | | | | etal.(2025),andspuriousrewards(Shaoetal.,2025). | | | | | | How- | +| ----------------------------------------- | --- | --- | --- | --- | --- | ------- | ----------------------------------------------- | --- | --- | --- | --- | --- | ---- | +| ThinkingSFTimprovesreasoningfaithfulness. | | | | | | In§3.4, | | | | | | | | +ever,thesefindingsoftendonottransferacrossmodelfam- +weidentifiedlowreasoningfaithfulnessasthepre-RLprop- +| | | | | | | | ilies, with | studies | reporting | inconsistent | | results | between | +| --------- | ------------- | --- | ------------ | ---------- | --- | ------- | ----------- | ------- | --------- | ------------ | --- | ------- | ------- | +| erty that | distinguished | | failing from | succeeding | | models. | | | | | | | | +QwenandLlama(Zengetal.,2025;Gandhietal.,2025; +| Fig. 7 shows | that | Thinking | SFT | raises | aligned-response | | | | | | | | | +| ------------ | ---- | -------- | --- | ------ | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | +ratethroughoutthepre-saturationphase,relativetotheNon- Shao et al., 2025). Moreover, most prior work focuses +onimprovingperformanceonnarrowdomains(primarily +| ThinkingSFTbaseline. | | | CPT+ThinkingSFTachievesthe | | | | | | | | | | | +| -------------------- | --- | --- | -------------------------- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | ------------- | --- | +| | | | | | | | math)withoutexamininggeneralization. | | | | | Recentwork(He | | +9 -Y -T +LLMReasoningwithWeakSupervision +et al., 2026; Yang et al., 2026; Plesner et al., 2026) has gesttwoconcretepracticesforRLfromweaksupervision. +concurrentlystudiedwhenandhowRLVRcanlearnunder First, monitor training reward saturation as a diagnostic: +self-supervisionornoisysupervision.Ourworkextendsthis plateauedrewardwithflatdownstreamperformanceindi- +literatureintwoways. First,wecharacterizetheconditions catesthemodelhasexhaustedwhatRLcanextractfromits +underwhichRLVRgeneralizesacrossmodelfamiliesand priors,andfurtherRLcomputeisunlikelytohelp. Second, +domains, focusing on saturation dynamics and reasoning whenweaksupervisionfails, allocatecomputetopre-RL +faithfulness. Second, we identify a concrete intervention interventionsthatinstallstrongpriorsratherthantolonger +thatrestoresgeneralizationinmodelswhereweaksupervi- RL training. Taken together, our findings argue that RL +sionwouldotherwisefail. underweaksupervisionisbestunderstoodnotasatraining +techniqueappliedtoafixedmodel,butasthefinalstageof +| Role of | Pre-Training | and | Fine-Tuning | | in RL. | Recent | | | | | | | | +| ------- | ------------ | --- | ----------- | --- | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | +apipelinewhosesuccessislargelydeterminedbeforeRL +workemphasizesthatpre-trainingandmid-trainingshape +begins. +| RL generalization | | (Qi et | al., 2025; | Wang | et | al., 2025b; | | | | | | | | +| ----------------- | --- | ------ | ---------- | ---- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | +Zhangetal.,2025;Akteretal.,2025),butfocusesoncom- +puteallocationanddistributionalignmenttoimproveperfor- Acknowledgements +mance.Ourworkspecificallyfocusesonunderstandinghow +WewouldliketothankLeonLi,VatsalBaherwani,Rohun +| base model | priors | shaped | from continual | | pretraining | and | | | | | | | | +| ---------- | ------ | ------ | -------------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +Agrawal,SiyanZhao,LiweiJiang,andAndyHanfortheir +| reasoning | SFT can | enable | generalization | | across | different | | | | | | | | +| --------- | ------- | ------ | -------------- | --- | ------ | --------- | ---------- | ----------- | --- | -------- | --- | ---------- | ----- | +| | | | | | | | insightful | discussions | and | feedback | on | the draft. | Pavel | +weaksupervisionsettings. +| | | | | | | | Izmailov | was supported | | by a grant | from | the | Alignment | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------- | --- | ---------- | ---- | --- | --------- | +Diversity and Faithfulness in Reasoning. Maintaining Project,fundedbytheUKAISecurityInstitute(grantAP- +| outputdiversityduringRLhasbeenproposedtopromote | | | | | | | S2-100141). | | | | | | | +| ----------------------------------------------- | --- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +explorationandmitigatemodelcollapse(Kirketal.,2024; +Casperetal.,2023;Rafailovetal.,2023;Yuetal.,2025), +References +| but prior | work has | not explored | | what | types | of diversity | | | | | | | | +| --------- | -------- | ------------ | --- | ---- | ----- | ------------ | --- | --- | --- | --- | --- | --- | --- | +benefitgeneralization. Separately,researchhashighlighted Agarwal,S.,Zhang,Z.,Yuan,L.,Han,J.,andPeng,H. The +| | | | | | | | unreasonable | effectiveness | | of | entropy | minimization | in | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------- | --- | --- | ------- | ------------ | --- | +mismatchesbetweenchain-of-thoughttracesandmodelpre- +dictions(Turpinetal.,2023;Chenetal.,2025b;Bakeretal., llmreasoning. arXivpreprintarXiv:2505.15134,2025. +2025;Tuteketal.,2025)andemphasizedtheimportanceof +| | | | | | | | AI-MO. | Aime 2024. | https://huggingface.co | | | | | +| ------------------------------------------- | --- | --- | --- | --- | --- | ---------- | ------ | ---------- | ---------------------- | --- | --- | --- | --- | +| ensuringfaithfulreasoningthroughouttraining | | | | | | (Guietal., | | | | | | | | +/datasets/AI-MO/aimo-validation-aime, +2026). Wenetal.(2025)arguesthatRLVRcanincentivize +2024a. +correctreasoninginbaseLLMsaslongaspriorshavebeen +| established. | Our | work connects | | these | lines of | research, | | | | | | | | +| ------------ | -------------- | ------------- | ---- | ----- | -------- | ----------- | ------ | --------- | ---------------------- | --- | --- | --- | --- | +| | | | | | | | AI-MO. | Amc 2023. | https://huggingface.co | | | | | +| showing | that diversity | alone | does | not | ensure | generaliza- | | | | | | | | +/datasets/AI-MO/aimo-validation-amc, +tionandthatreasoningfaithfulnessdistinguishesmodels’ +2024b. +| trainingdynamics. | | Wefurtherdemonstratethatpre-RLin- | | | | | | | | | | | | +| ----------------- | --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +terventioncanimprovereasoningfaithfulnessandimprove Akter, S. N., Prabhumoye, S., Nyberg, E., Patwary, M., +generalizationunderweaksupervision. Shoeybi,M.,Choi,Y.,andCatanzaro,B. Front-loading +| | | | | | | | reasoning: | The | synergy | between | pretraining | | and post- | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------- | ------- | ----------- | --- | --------- | +6.Conclusion trainingdata. arXivpreprintarXiv:2510.03264,2025. +| | | | | | | | Baker, B., | Huizinga, | J., | Gao, L., | Dou, | Z., Guan, | M. Y., | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | --- | -------- | ---- | --------- | ------ | +Inthiswork,westudiedwhenandwhyRLVRgeneralizes +underweaksupervisionacrossdiversemodelfamiliesand Madry,A.,Zaremba,W.,Pachocki,J.,andFarhi,D.Mon- +threereasoningdomains. Successunderscarcedata,noisy itoringreasoningmodelsformisbehaviorandtherisksof +promotingobfuscation.arXivpreprintarXiv:2503.11926, +rewards,andself-supervisedproxyrewardsdependsonpre- +| RLproperties,pretrainingpriorsandreasoningfaithfulness, | | | | | | | 2025. | | | | | | | +| ------------------------------------------------------- | ----- | -------- | ------ | ------ | ---- | -------- | ------- | ------------ | --- | ---------- | --------- | --- | ----------- | +| rather than | on RL | dynamics | alone. | Models | that | saturate | | | | | | | | +| | | | | | | | Bowman, | S. R., Hyun, | | J., Perez, | E., Chen, | E., | Pettit, C., | +rapidlyproducecorrectanswersthroughreasoningthatdoes +| | | | | | | | Heiner, | S., Lukosˇiu¯te˙, | | K., Askell, | A., | Jones, | A., Chen, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----------------- | --- | ----------- | --- | ------ | --------- | +notsupportthem,memorizingratherthanlearning,while +| | | | | | | | A.,etal. | Measuringprogressonscalableoversightfor | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------- | --------------------------------------- | --- | --- | --- | --- | --- | +maintainingthehighoutputdiversitynormallytakenasa +| | | | | | | | largelanguagemodels. | | | arXivpreprintarXiv:2211.03540, | | | | +| --------------- | ------------ | --- | ------ | ------------- | --- | ------- | -------------------- | --- | --- | ------------------------------ | --- | --- | --- | +| sign of healthy | exploration. | | Pre-RL | interventions | | target- | | | | | | | | +2022. +| ingreasoningfaithfulnessrecovergeneralization: | | | | | | SFTon | | | | | | | | +| ---------------------------------------------- | --- | ------ | ---------------- | --- | ----------- | ----- | --- | --- | --- | --- | --- | --- | --- | +| explicit reasoning | | traces | is the necessary | | ingredient, | and | | | | | | | | +Burns,C.,Izmailov,P.,Kirchner,J.H.,Baker,B.,Gao,L., +| continual | pre-training | on | reasoning-heavy | | data | amplifies | | | | | | | | +| --------- | ------------ | --- | --------------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | +Aschenbrenner,L.,Chen,Y.,Ecoffet,A.,Joglekar,M., +the effect without substituting for it. These findings sug- Leike,J.,etal. Weak-to-stronggeneralization: Eliciting +10 + +LLMReasoningwithWeakSupervision +strongcapabilitieswithweaksupervision. arXivpreprint Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, +arXiv:2312.09390,2023. A.,Letman,A.,Mathur,A.,Schelten,A.,Yang,A.,Fan, +| | | | | | | A.,etal. | Thellama3herdofmodels. | | | arXive-prints,pp. | | | +| --- | --- | --- | --- | --- | --- | -------- | ---------------------- | --- | --- | ----------------- | --- | --- | +Casper,S.,Davies,X.,Shi,C.,Gilbert,T.K.,Scheurer,J., arXiv–2407,2024. +Rando,J.,Freedman,R.,Korbak,T.,Lindner,D.,Freire, +| | | | | | | Farquhar,S.,Kossen,J.,Kuhn,L.,andGal,Y. | | | | | Detecting | | +| --------- | ------ | ------ | ------------- | ------ | ------------ | --------------------------------------- | --- | --- | --- | --- | --------- | --- | +| P., Wang, | T. T., | Marks, | S., Se´gerie, | C.-R., | Carroll, M., | | | | | | | | +hallucinationsinlargelanguagemodelsusingsemantic +Peng,A.,Christoffersen,P.J.K.,Damani,M.,Slocum,S., +| | | | | | | entropy. | Nature,630(8017):625–630,2024. | | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------ | --- | --- | --- | --- | --- | +Anwar,U.,Siththaranjan,A.,Nadeau,M.,Michaud,E.J., +Pfau, J., Krasheninnikov, D., Chen, X., Langosco, L., Gandhi, K., Chakravarthy, A., Singh, A., Lile, N., and +Hase,P.,Biyik,E.,Dragan,A.D.,Krueger,D.,Sadigh, +| | | | | | | Goodman, | N. | D. Cognitive | behaviors | | that enable | self- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------------ | --------- | --- | ----------- | ----- | +D.,andHadfield-Menell,D. Openproblemsandfunda- improvingreasoners,or,fourhabitsofhighlyeffective +mentallimitationsofreinforcementlearningfromhuman stars. arXivpreprintarXiv:2503.01307,2025. +| feedback. | TransactionsonMachineLearningResearch, | | | | | | | | | | | | +| --------- | -------------------------------------- | --- | --- | --- | --- | -------------------- | --------- | ------ | --------------------- | ------------ | --- | --- | +| | | | | | | Google | DeepMind. | Gemini | 3 flash. | https://gemi | | | +| 2023. | URL https://openreview.net/forum | | | | | | | | | | | | +| | | | | | | ni.google.com/,2025. | | | ReleasedDecember2025. | | | | +?id=bx24KpJ4Eb. +| | | | | | | Accessed: | 2026-02-18. | | | | | | +| --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | --- | --- | --- | --- | +Chandak,N.,Goel,S.,andPrabhu,A. Incorrectbaseline Guha, E., Marten, R., Keh, S., Raoof, N., Smyrnis, G., +| evaluations | call | into | question recent | llm-rl | claims. ht | | | | | | | | +| ----------- | ---- | ---- | --------------- | ------ | ---------- | --- | --- | --- | --- | --- | --- | --- | +Bansal,H.,Nezhurina,M.,Mercat,J.,Vu,T.,Sprague,Z., +tps://safe-lip-9a8.notion.site/Incor +| | | | | | | etal. | Openthoughts: | Datarecipesforreasoningmodels. | | | | | +| --- | --- | --- | --- | --- | --- | ----- | ------------- | ------------------------------ | --- | --- | --- | --- | +rect-Baseline-Evaluations-Call-into-Q +arXivpreprintarXiv:2506.04178,2025. +uestion-Recent-LLM-RL-Claims-2012f1f +| | | | | | | Gui, R., | Li, Y., | Qu, X., Liu, | Z., | Cheng, | Y., and Cheng, | | +| -------------------------------- | --- | --- | --- | --- | ----- | -------- | ------- | ------------ | --- | ------ | -------------- | --- | +| bf0ee8094ab8ded1953c15a37?pvs=4, | | | | | 2025. | | | | | | | | +NotionBlog. Y. Faithrl: Learning to reason faithfully through +| | | | | | | step-level | faithfulness | maximization. | | | arXiv preprint | | +| --- | --- | --- | --- | --- | --- | ---------- | ------------ | ------------- | --- | --- | -------------- | --- | +arXiv:2602.03507,2026. +| Chen, P., | Li, X., | Li, Z., | Yin, W., Chen, | X., | and Lin, T. | | | | | | | | +| ----------- | -------- | ------------------ | -------------- | ------------- | ------------ | ----------------------------------- | ----- | ---------- | --------- | --- | --------- | ---- | +| Exploration | vs | exploitation: | Rethinking | | rlvr through | | | | | | | | +| | | | | | | Guo, D., | Yang, | D., Zhang, | H., Song, | J., | Wang, P., | Zhu, | +| clipping, | entropy, | andspuriousreward. | | arXivpreprint | | | | | | | | | +| | | | | | | Q.,Xu,R.,Zhang,R.,Ma,S.,Bi,X.,etal. | | | | | Deepseek- | | +arXiv:2512.16912,2025a. +r1incentivizesreasoninginllmsthroughreinforcement +| | | | | | | learning. | Nature,645(8081):633–638,2025. | | | | | | +| --- | --- | --- | --- | --- | --- | --------- | ------------------------------ | --- | --- | --- | --- | --- | +Chen,Y.,Benton,J.,Radhakrishnan,A.,Uesato,J.,Deni- +son,C.,Schulman,J.,Somani,A.,Hase,P.,Wagner,M., He, B., Zuo, Y., Liu, Z., Zhao, S., Fu, Z., Yang, J., Qian, +Roger,F.,etal. Reasoningmodelsdon’talwayssaywhat C., Zhang, K., Fan, Y., Cui, G., et al. How far can +theythink. arXivpreprintarXiv:2505.05410,2025b. unsupervised rlvr scale llm training? arXiv preprint +arXiv:2603.08660,2026. +Cheng,Z.,Hao,S.,Liu,T.,Zhou,F.,Xie,Y.,Yao,F.,Bian, +He,C.,Luo,R.,Bai,Y.,Hu,S.,Thai,Z.,Shen,J.,Hu,J., +Y.,Zhuang,Y.,Dey,N.,Zha,Y.,Gu,Y.,Zhou,K.,Wang, +| | | | | | | Han,X.,Huang,Y.,Zhang,Y.,etal. | | | | Olympiadbench: | | A | +| ------- | -------- | --- | ------------- | ------------ | ------- | ------------------------------ | --- | --- | --- | -------------- | --- | --- | +| Y., Li, | Y., Fan, | R., | She, J., Gao, | C., Saparov, | A., Li, | | | | | | | | +H., Killian, T. W., Yurochkin, M., Liu, Z., Xing, E. P., challengingbenchmarkforpromotingagiwitholympiad- +| | | | | | | levelbilingualmultimodalscientificproblems. | | | | | InPro- | | +| ------- | ------------- | --- | ------------- | -------- | ------- | ------------------------------------------- | --- | --- | --- | --- | ------ | --- | +| and Hu, | Z. Revisiting | | reinforcement | learning | for llm | | | | | | | | +ceedingsofthe62ndAnnualMeetingoftheAssociation +| reasoningfromacross-domainperspective,2025. | | | | | URL | | | | | | | | +| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +https://arxiv.org/abs/2506.14965. forComputationalLinguistics(Volume1: LongPapers), +pp.3828–3850,2024. +| Cohen, J. | A coefficient | | of agreement | for nominal | scales. | | | | | | | | +| --------- | ------------- | --- | ------------ | ----------- | ------- | ------------ | -------- | ------ | -------- | ----- | ---------- | --- | +| | | | | | | He, J., Liu, | J., Liu, | C. Y., | Yan, R., | Wang, | C., Cheng, | P., | +EducationalandPsychologicalMeasurement,20(1):37– +| | | | | | | Zhang, | X., Zhang, | F., Xu, | J., Shen, | W., | Li, S., | Zeng, | +| --- | --- | --- | --- | --- | --- | ------ | ---------- | ------- | --------- | --- | ------- | ----- | +46,1960. +| | | | | | | L., Wei, | T., Cheng, | C., | An, B., | Liu, Y., | and Zhou, | Y. | +| --- | --- | --- | --- | --- | --- | ------------------------------------ | ---------- | --- | ------- | -------- | ------------- | --- | +| | | | | | | Skyworkopenreasoner1technicalreport. | | | | | arXivpreprint | | +Cui,G.,Zhang,Y.,Chen,J.,Yuan,L.,Wang,Z.,Zuo,Y.,Li, arXiv:2505.22312,2025a. +| H.,Fan,Y.,Chen,H.,Chen,W.,etal. | | | | Theentropymech- | | | | | | | | | +| ------------------------------- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +anismofreinforcementlearningforreasoninglanguage He, J., Liu, J., Liu, C. Y., Yan, R., Wang, C., Cheng, P., +Zhang,X.,Zhang,F.,Xu,J.,Shen,W.,Li,S.,Zeng,L., +| models. | arXivpreprintarXiv:2505.22617,2025. | | | | | | | | | | | | +| ------- | ----------------------------------- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | --- | +| | | | | | | Wei,T.,Cheng,C.,Liu,Y.,andZhou,Y. | | | | | Skyworkopen | | +Du,X.,Yao,Y.,Ma,K.,Wang,B.,Zheng,T.,Zhu,K.,Liu, reasonerseries. https://capricious-hydroge +M.,Liang,Y.,Jin,X.,Wei,Z.,etal. Supergpqa: Scaling n-41c.notion.site/Skywork-Open-Reaon +llm evaluation across 285 graduate disciplines. arXiv ser-Series-1d0bc9ae823a80459b46c149e +| preprintarXiv:2502.14739,2025. | | | | | | 4f51680,2025b. | | NotionBlog. | | | | | +| ------------------------------ | --- | --- | --- | --- | --- | -------------- | --- | ----------- | --- | --- | --- | --- | +11 + +LLMReasoningwithWeakSupervision +Hendrycks,D.,Burns,C.,Kadavath,S.,Arora,A.,Basart, Liu, Z., Chen, C., Li, W., Qi, P., Pang, T., Du, C., Lee, +S., Tang, E., Song, D., and Steinhardt, J. Measuring W.S.,andLin,M. Understandingr1-zero-liketraining: +mathematicalproblemsolvingwiththemathdataset. In Acriticalperspective. arXivpreprintarXiv:2503.20783, +| Thirty-fifthConferenceonNeuralInformationProcessing | | | | | | | 2025b. | | | | | | | | +| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +SystemsDatasetsandBenchmarksTrack(Round2). +Lu,D.,Tan,X.,Xu,R.,Yao,T.,Qu,C.,Chu,W.,Xu,Y.,and +Qi,Y.Scp-116k:Ahigh-qualityproblem-solutiondataset +| Hu, J., Liu, | M., | Lu, X., | Wu, F., | Harchaoui, | | Z., Diao, S., | | | | | | | | | +| ------------ | --- | ------- | ------- | ---------- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Choi,Y.,Molchanov,P.,Yang,J.,Kautz,J.,etal. Brorl: and a generalized pipeline for automated extraction in +| | | | | | | | the higher | education | | science | domain. | | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | --- | ------- | ------- | --- | ----- | -------- | +Scalingreinforcementlearningviabroadenedexploration. +| arXivpreprintarXiv:2510.01180,2025. | | | | | | | arXiv:2501.15587,2025. | | | | | | | | +| ----------------------------------- | ------- | --------- | ------- | ------ | ------- | ----------- | ----------------------------- | ----- | --------- | --- | ----------- | ----------------- | --- | -------- | +| | | | | | | | Mahabadi, | R.K., | Satheesh, | S., | Prabhumoye, | | S., | Patwary, | +| Huang, | C., Yu, | W., Wang, | X., | Zhang, | H., | Li, Z., Li, | | | | | | | | | +| | | | | | | | M.,Shoeybi,M.,andCatanzaro,B. | | | | | Nemotron-cc-math: | | | +| R., Huang, | J., | Mi, | H., and | Yu, D. | R-zero: | Self- | | | | | | | | | +A133billion-token-scalehighqualitymathpretraining +| evolving | reasoning | llm | from | zero data. | arXiv | preprint | | | | | | | | | +| -------- | --------- | --- | ---- | ---------- | ----- | -------- | -------- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | | | | | dataset. | arXivpreprintarXiv:2508.15096,2025. | | | | | | | +arXiv:2508.05004,2025. +| | | | | | | | Olmo, T., | Ettinger, | A., | Bertsch, | A., | Kuehl, | B., | Graham, | +| --- | --- | --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --- | ------ | --- | ------- | +Hurst,A.,Lerer,A.,Goucher,A.P.,Perelman,A.,Ramesh, D., Heineman, D., Groeneveld, D., Brahman, F., Tim- +A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., arXiv preprint +| | | | | | | | bers, | F., Ivison, | H., | et al. | Olmo | 3. | | | +| ---------------- | --- | ----------------- | --- | --- | ------------- | --- | ----- | ----------- | --- | ------ | ---- | --- | --- | --- | +| Radford,A.,etal. | | Gpt-4osystemcard. | | | arXivpreprint | | | | | | | | | | +arXiv:2512.13961,2025. +arXiv:2410.21276,2024. +| | | | | | | | OpenAI. | Openai | o3 and | o4-mini | system | | card. | https: | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------ | ------ | ------- | ------ | --- | ----- | ------ | +Jaech,A.,Kalai,A.,Lerer,A.,Richardson,A.,El-Kishky, //cdn.openai.com/pdf/2221c875-02dc-4 +A., Low, A., Helyar, A., Madry, A., Beutel, A., Car- 789-800b-e7758f3722c1/o3-and-o4-min +ney, A., et al. Openai o1 system card. arXiv preprint i-system-card.pdf,April2025a. Accessed2026- +01-25. +arXiv:2412.16720,2024. +| | | | | | | | OpenAI. | gpt-oss-20b | | model | card, | 2025b. | URL | https: | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----------- | --- | ----- | ----- | ------ | --- | ------ | +Kirk,R.,Mediratta,I.,Nalmpantis,C.,Luketina,J.,Ham- +| | | | | | | | //arxiv.org/abs/2508.10925. | | | | | | Accessed:2026- | | +| ------------------------------------- | --- | --- | --- | --- | ------------- | --- | --------------------------- | --- | --- | --- | --- | --- | -------------- | --- | +| bro,E.,Grefenstette,E.,andRaileanu,R. | | | | | Understanding | | | | | | | | | | +02-18. +| theeffectsofrlhfonllmgeneralisationanddiversity. | | | | | | In | | | | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +TheTwelfthInternationalConferenceonLearningRepre- +| | | | | | | | Opencompass. | | Aime2025. | https://huggingface. | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------- | -------------------- | --- | --- | --- | --- | +sentations,2024. URLhttps://openreview.net co/datasets/opencompass/AIME2025,2024. +/forum?id=PXD3FAVHJT. +| | | | | | | | Plesner, | A., Guzma´n, | | F., and | Athalye, | A. | An | imperfect | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------ | --- | ------- | -------- | --- | --- | --------- | +Lewkowycz, A., Andreassen, A., Dohan, D., Dyer, E., verifier is good enough: Learning with noisy rewards. +Michalewski, H., Ramasesh, V., Slone, A., Anil, C., arXivpreprintarXiv:2604.07666,2026. +| Schlag,I.,Gutman-Solo,T.,etal. | | | | Solvingquantitative | | | | | | | | | | | +| ------------------------------------ | --- | --- | --- | ------------------- | --- | ---------- | ------------ | --- | ----- | ------------- | --- | ---------------- | --- | --- | +| | | | | | | | Prabhudesai, | M., | Chen, | L., Ippoliti, | | A., Fragkiadaki, | | K., | +| reasoningproblemswithlanguagemodels. | | | | | | Advancesin | | | | | | | | | +neural information processing systems, 35:3843–3857, Liu, H., and Pathak, D. Maximizing confidence alone +| | | | | | | | improvesreasoning. | | | arXivpreprintarXiv:2505.22660, | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | ------------------------------ | --- | --- | --- | --- | +2022. +2025. +Li,T.,Zhang,Y.,Yu,P.,Saha,S.,Khashabi,D.,Weston,J., +| | | | | | | | Qi, Z., Nie, | F., | Alahi, | A., Zou, | J., | Lakkaraju, | H., | Du, Y., | +| ------------------------ | --- | --- | --------------------------- | --- | --- | --- | ----------------------------- | --- | ------ | -------- | --- | ---------- | ---------- | ------- | +| Lanchantin,J.,andWang,T. | | | Jointlyreinforcingdiversity | | | | | | | | | | | | +| | | | | | | | Xing,E.,Kakade,S.,andZhang,H. | | | | | Evolm: | Insearchof | | +andqualityinlanguagemodelgenerations.arXivpreprint +| | | | | | | | lostlanguagemodeltrainingdynamics. | | | | | | arXivpreprint | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | ------------- | --- | +arXiv:2509.02534,2025. +arXiv:2506.16029,2025. +Lightman,H.,Kosaraju,V.,Burda,Y.,Edwards,H.,Baker, +Rafailov,R.,Sharma,A.,Mitchell,E.,Ermon,S.,Manning, +| B., Lee, | T., Leike, | J., | Schulman, | J., | Sutskever, | I., and | | | | | | | | | +| -------- | ---------- | --- | --------- | --- | ---------- | ------- | --------------- | --- | ----------------------------- | --- | --- | --- | --- | ---- | +| | | | | | | | C.D.,andFinn,C. | | Directpreferenceoptimization: | | | | | Your | +Cobbe, K. Let’s verify step by step. In The Twelfth InAdvances +languagemodelissecretlyarewardmodel. +InternationalConferenceonLearningRepresentations, +| | | | | | | | in Neural | Information | | Processing | | Systems, | volume | 36, | +| ----- | --- | --- | --- | --- | --- | --- | -------------------- | ----------- | --- | ---------- | --- | -------- | ------ | --- | +| 2023. | | | | | | | pp.53728–53741,2023. | | | | | | | | +Liu,M.,Diao,S.,Lu,X.,Hu,J.,Dong,X.,Choi,Y.,Kautz, Rahman,S.,Issaka,S.,Suvarna,A.,Liu,G.,Shiffer,J.,Lee, +J.,andDong,Y. Prorl: Prolongedreinforcementlearning J.,Parvez,M.R.,Palangi,H.,Feng,S.,Peng,N.,etal. +expandsreasoningboundariesinlargelanguagemodels. Aidebateaidsassessmentofcontroversialclaims. arXiv +arXivpreprintarXiv:2505.24864,2025a. preprintarXiv:2506.02175,2025. +12 + +LLMReasoningwithWeakSupervision +Rein,D.,Hou,B.L.,Stickland,A.C.,Petty,J.,Pang,R.Y., Suzhou,China,November2025.AssociationforCom- +Dirani, J., Michael, J., and Bowman, S. R. Gpqa: A putationalLinguistics. ISBN979-8-89176-332-6. doi: +10.18653/v1/2025.emnlp-main.504.URLhttps://ac +| graduate-level | | google-proof | q&a benchmark. | | In First | | | | | +| -------------- | --- | ------------ | -------------- | --- | -------- | --- | --- | --- | --- | +ConferenceonLanguageModeling,2024. lanthology.org/2025.emnlp-main.504/. +Shafayat, S., Tajwar, F., Salakhutdinov, R., Schneider, J., Wang,X.,Hu,Z.,Lu,P.,Zhu,Y.,Zhang,J.,Subramaniam, +| andZanette,A. | | Canlargereasoningmodelsself-train? | | | | | | | | +| ------------- | --- | ---------------------------------- | --- | --- | --- | ----------- | ------------- | -------- | ---------------- | +| | | | | | | S., Loomba, | A. R., Zhang, | S., Sun, | Y., and Wang, W. | +arXivpreprintarXiv:2505.21444,2025. Scibench: Evaluating college-level scientific problem- +| | | | | | | solvingabilitiesoflargelanguagemodels. | | | arXivpreprint | +| ------------ | ----------------------------------- | --- | --- | --- | --- | -------------------------------------- | --- | --- | ------------- | +| Shannon,C.E. | Amathematicaltheoryofcommunication. | | | | | | | | | +arXiv:2307.10635,2023. +TheBellsystemtechnicaljournal,27(3):379–423,1948. +Wang,Y.,Ma,X.,Zhang,G.,Ni,Y.,Chandra,A.,Guo,S., +| Shao, R., | Li, S. | S., Xin, | R., Geng, S., | Wang, | Y., Oh, S., | | | | | +| ------------------------------------------ | ------ | -------- | ------------- | ----- | ----------- | -------------------------------------- | --- | --- | --------- | +| | | | | | | Ren,W.,Arulraj,A.,He,X.,Jiang,Z.,etal. | | | Mmlu-pro: | +| Du,S.S.,Lambert,N.,Min,S.,Krishna,R.,etal. | | | | | Spu- | | | | | +riousrewards: Rethinkingtrainingsignalsinrlvr. arXiv Amorerobustandchallengingmulti-tasklanguageun- +| | | | | | | derstandingbenchmark. | | AdvancesinNeuralInformation | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --------------------------- | --- | +preprintarXiv:2506.10947,2025. +ProcessingSystems,37:95266–95290,2024. +Shao,Z.,Wang,P.,Zhu,Q.,Xu,R.,Song,J.,Bi,X.,Zhang, +H.,Zhang,M.,Li,Y.,Wu,Y.,etal. Deepseekmath: Push- Wang, Y., Yang, Q., Zeng, Z., Ren, L., Liu, L., Peng, B., +ingthelimitsofmathematicalreasoninginopenlanguage Cheng,H.,He,X.,Wang,K.,Gao,J.,etal. Reinforce- +mentlearningforreasoninginlargelanguagemodelswith +| models. | arXivpreprintarXiv:2402.03300,2024. | | | | | | | | | +| ------- | ----------------------------------- | --- | --- | --- | --- | ------------------- | ------------------------------ | --- | --- | +| | | | | | | onetrainingexample. | arXivpreprintarXiv:2504.20571, | | | +Sheng,G.,Zhang,C.,Ye,Z.,Wu,X.,Zhang,W.,Zhang, +2025a. +| R., Peng, | Y., | Lin, H., | and Wu, | C. Hybridflow: | A | | | | | +| --------- | --- | -------- | ------- | -------------- | --- | --- | --- | --- | --- | +flexible and efficient rlhf framework. arXiv preprint Wang, Z., Zhou, F., Li, X., and Liu, P. Octothinker: +arXiv:2409.19256,2024. Mid-trainingincentivizesreinforcementlearningscaling. +arXivpreprintarXiv:2506.20512,2025b. +Stojanovski,Z.,Stanley,O.,Sharratt,J.,Jones,R.,Adefioye, +A.,Kaddour,J.,andKo¨pf,A.Reasoninggym:Reasoning Wen,X.,Liu,Z.,Zheng,S.,Ye,S.,Wu,Z.,Wang,Y.,Xu, +environmentsforreinforcementlearningwithverifiable +| | | | | | | Z.,Liang,X.,Li,J.,Miao,Z.,etal. | | Reinforcementlearn- | | +| --- | --- | --- | --- | --- | --- | ------------------------------- | --- | ------------------- | --- | +arXivpreprintarXiv:2505.24760,2025. +rewards. ingwithverifiablerewardsimplicitlyincentivizescorrect +| | | | | | | reasoninginbasellms. | arXivpreprintarXiv:2506.14245, | | | +| -------- | --------- | ----- | --------- | --------- | --------- | -------------------- | ------------------------------ | --- | --- | +| Sun, Y., | Shen, J., | Wang, | Y., Chen, | T., Wang, | Z., Zhou, | | | | | +2025. +| M., and | Zhang, | H. Improving | data | efficiency | for llm | | | | | +| ------- | ------ | ------------ | ---- | ---------- | ------- | --- | --- | --- | --- | +reinforcementfine-tuningthroughdifficulty-targetedon- +| | | | | | | Williams,R.J. | Simplestatisticalgradient-followingalgo- | | | +| --------- | --------- | --- | --------------- | ----- | -------- | -------------------------------------------- | ---------------------------------------- | --- | ------- | +| line data | selection | and | rollout replay. | arXiv | preprint | | | | | +| | | | | | | rithmsforconnectionistreinforcementlearning. | | | Machine | +arXiv:2506.05316,2025. +learning,8(3):229–256,1992. +Team,K.,Du,A.,Gao,B.,Xing,B.,Jiang,C.,Chen,C., +Yang,A.,Zhang,B.,Hui,B.,Gao,B.,Yu,B.,Li,C.,Liu,D., +| Li, C., | Xiao, | C., Du, | C., Liao, C., | et al. | Kimi k1. 5: | | | | | +| ------- | ----- | ------- | ------------- | ------ | ----------- | --- | --- | --- | --- | +Tu,J.,Zhou,J.,Lin,J.,Lu,K.,Xue,M.,Lin,R.,Liu,T., +| Scalingreinforcementlearningwithllms. | | | | arXivpreprint | | | | | | +| ------------------------------------- | --- | --- | --- | ------------- | --- | ------------------ | --- | ---------------------------- | --- | +| | | | | | | Ren,X.,andZhang,Z. | | Qwen2.5-mathtechnicalreport: | | +arXiv:2501.12599,2025. +Towardmathematicalexpertmodelviaself-improvement. +arXivpreprintarXiv:2409.12122,2024. +| Team,Q. | Qwen2.5: | Apartyoffoundationmodels,Septem- | | | | | | | | +| ------------- | ----------------------------- | -------------------------------- | --- | --- | --- | ---------------------------------------------- | --------------- | ------------- | ----------- | +| ber2024. | URLhttps://qwenlm.github.io/b | | | | | | | | | +| log/qwen2.5/. | | | | | | Yang,S.,Zhu,G.,Song,B.,Li,S.,Wang,H.,Zheng,X., | | | | +| | | | | | | Ma, Y., | Chen, Z., Wang, | W., and Chen, | G. Can llms | +Turpin, M., Michael, J., Perez, E., andBowman, S. Lan- learntoreasonrobustlyundernoisysupervision? arXiv +guage models don’t always say what they think: Un- preprintarXiv:2604.03993,2026. +| faithfulexplanationsinchain-of-thoughtprompting. | | | | | Ad- | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +vances in Neural Information Processing Systems, 36: Yu,Q.,Zhang,Z.,Zhu,R.,Yuan,Y.,Zuo,X.,Yue,Y.,Dai, +| | | | | | | W.,Fan,T.,Liu,G.,Liu,L.,etal. | | Dapo: | Anopen-source | +| --- | --- | --- | --- | --- | --- | ----------------------------- | --- | ----- | ------------- | +74952–74965,2023. +llmreinforcementlearningsystematscale.arXivpreprint +| Tutek, M., | Hashemi | Chaleshtori, | F., | Marasovic, | A., and | | | | | +| ---------- | ------- | ------------ | --- | ---------- | ------- | --- | --- | --- | --- | +arXiv:2503.14476,2025. +| Belinkov, | Y. | Measuring | chain of | thought | faithfulness | | | | | +| --------- | --- | --------- | -------- | ------- | ------------ | --- | --- | --- | --- | +by unlearning reasoning steps. In Christodoulopoulos, Yue,Y.,Chen,Z.,Lu,R.,Zhao,A.,Wang,Z.,Song,S.,and +C.,Chakraborty,T.,Rose,C.,andPeng,V.(eds.),Pro- Huang,G. Doesreinforcementlearningreallyincentivize +ceedings of the 2025 Conference on Empirical Meth- reasoningcapacityinllmsbeyondthebasemodel? arXiv +ods in Natural Language Processing, pp. 9935–9960, preprintarXiv:2504.13837,2025. +13 + +LLMReasoningwithWeakSupervision +| Zeng, W., | Huang, Y., Liu, | Q., Liu, W., | He, K., Ma, Z., | +| --------- | --------------- | -------------------------- | --------------- | +| andHe,J. | Simplerl-zoo: | Investigatingandtamingzero | | +reinforcementlearningforopenbasemodelsinthewild. +arXivpreprintarXiv:2503.18892,2025. +| Zhang, C., | Neubig, G., and | Yue, X. On | the interplay of | +| ---------- | --------------- | ---------- | ---------------- | +pre-training,mid-training,andrlonreasoninglanguage +| models. | arXivpreprintarXiv:2512.07783,2025. | | | +| --------- | ----------------------------------- | ----------------- | ------------ | +| Zhao, X., | Kang, Z., Feng, | A., Levine, S., | and Song, D. | +| Learning | to reason without | external rewards. | arXiv | +preprintarXiv:2505.19590,2025. +Zhu,X.,Xia,M.,Wei,Z.,Chen,W.-L.,Chen,D.,andMeng, +Y. Thesurprisingeffectivenessofnegativereinforcement +| inllmreasoning. | arXivpreprintarXiv:2506.01347,2025. | | | +| --------------- | ----------------------------------- | --- | --- | +Zuo,Y.,Zhang,K.,Sheng,L.,Qu,S.,Cui,G.,Zhu,X.,Li, +| H.,Zhang,Y.,Long,X.,Hua,E.,etal. | | | Ttrl: Test-time | +| -------------------------------- | --- | --- | --------------- | +reinforcementlearning.arXivpreprintarXiv:2504.16084, +2025. +14 + +LLMReasoningwithWeakSupervision +A.LimitationsandFutureWork +Weacknowledgeseverallimitations. First,duetocomputationalconstraints,ouranalysisisrestrictedtospecificmodel +families and scales. Validating these findings across larger architectures and broader task suites remains an important +direction.Second,ouranalysisofdiversityandfaithfulnessreliesonanLLM-as-a-judgeframework.Althoughweconducted +small-scalehumanverificationtovalidatelabelquality,wecurrentlyrestrictthisevaluationtoasmallscaletoallowfor +reasonablelabelingcosts.Consequently,thedevelopmentofscalablemetricsforreasoningfaithfulnessanddiversityremains +animportantdirectionforfutureresearch. +B.ImplementationDetails +B.1.TrainingandEvaluationDatasets +WeinvestigateRLtrainingdynamicsacrosstwomodelfamilies: Qwen(comprisingQwen2.5-1.5B/3BandQwen2.5-Math- +1.5B/7B)andLlama(Llama-3.2-3B/8B-Instruct). Ouranalysisspansthreedistinctreasoningdomains,MATH,SCIENCE, +andGRAPH,allowingforaholisticinvestigationofRLVRunderweaksupervisionacrossdifferentdomainsandmodel +families. ForMATH,wesampletrainingpromptsfromtheSkywork-OR1(Heetal.,2025b)dataset. ForSCIENCE,wedraw +problemsfromtheSCPdatasetcuratedbypriorwork(Liuetal.,2025a;Luetal.,2025),byselectingPhysics,Chemistry, +andBiologysubjects. ForGRAPH,wegeneratetwosyntheticalgorithmictasks,QuantumLockandLargestIsland,using +thecurriculumspecificationsprovidedbytheReasoningGymbenchmark(Stojanovskietal.,2025). Foreachtask,we +instantiatefivedifficultylevelsfollowingthebenchmark’scurriculum,withabalancednumberofsamplesperlevel. +Weincludethefollowingdomain-specificbenchmarksforevaluations: +• MATH500(Lightmanetal.,2023): AwidelyusedsubsetoftheMATHtestsplit(Hendrycksetal.). +• AMC(AI-MO,2024b): 40competition-levelmathquestions. +• AIME2024(AI-MO,2024a): 30competition-levelmathquestions. +• AIME2025(Opencompass,2024): 30competition-levelmathquestions. +• MinervaMath(Lewkowyczetal.,2022): Asetof272undergraduate-levelscienceandmathquestionsfromMIT +OpenCourseWare. +• OlympiadBench(Heetal.,2024): Abenchmarkof675problemsfrominternationalmatholympiadsandphysics +contests. +• GPQA-Diamond(Reinetal.,2024): 198expert-levelquestionsfromGPQAspanningphysics,chemistry,andbiology; +wepreprocessthedatafollowingpreviouspractice(Chengetal.,2025). +• SCP-Hard(Luetal.,2025;Liuetal.,2025a): Aheld-outsetof50SCPquestionsfilteredsuchthatthebasemodels +(Qwen2.5-1.5BseriesmodelsandLlama3.2-3B-Instructmodel)achievesolve@16=1,containingdisjointquestions +fromtheSCPtrainingdatasets. +• SuperGPQA (Du et al., 2025): a subset constructed from the original SuperGPQA which contains 319 science +questionsand250non-sciencequestions. +• MMLUSCI(Wangetal.,2024): asubsetofMMLUProbenchmarkcontainingallcollege-levelchemistry,physics +andbiologyquestions. +• ScienceBench(Wangetal.,2023): 692college-levelsciencequestions. +• GraphTest: Aheld-outsetof50algorithmicallygeneratedinstancesfromtheQuantumLock andLargestIsland +tasks using Reasoning Gym (Stojanovski et al., 2025), disjoint from training, filtered such that the base models +(Qwen2.5-1.5BseriesandLlama3.2-3B-Instruct)achievePass@16=1. +WealsonotethatGPQA-Diamond,MMLUSCI,andSuperGPQAaremultiple-choicebenchmarks,forwhich +pass@kmaybealessreliablemetric. +Table2detailsthetrainingandevaluationdatasetsacrossthethreereasoningdomains. +15 + +LLMReasoningwithWeakSupervision +Table2.Trainingdatasetsandevaluationbenchmarksacrossthreereasoningdomains. +Domain TrainingSource In-DistributionEval Out-of-DistributionEval +MATH Skywork-OR1 MATH-500,AMC,AIME-2024,AIME- Science Bench, SuperGPQA, GPQA- +2025,MinervaMath,OlympiadBench Diamond,SCP-Hard,MMLUSCI +SCIENCE SCP-116K SCP-Hard,GPQA-Diamond,MMLUSCI, MATH-500, AMC, Minerva Math, +SuperGPQA,ScienceBench OlympiadBench +GRAPH ReasoningGym QuantumLock,LargestIsland MATH-500,AIME-2024,MinervaMath, +GPQA-Diamond,SCP-Hard +PrompttemplateforMATHandGRAPH. +system +YouareahelpfulAIAssistant,designedtoprovidewell-reasonedanddetailedresponses. +YouFIRSTthinkaboutthereasoningprocessstepbystepandthenprovidetheuserwiththeanswer. +Pleaseencloseyourfinalanswerinthebox: \boxed{YourAnswer}. +user + +assistant +Figure8.PrompttemplateusedforRLtrainingandevaluationonMATHandGRAPH.Theplaceholderisreplaced +withtheactualmathematicalquestionduringfine-tuningandevaluation.Specialtokensareomittedforclarity. +B.2.TrainingDataPreparationDetails +Wedescribeourprocedureforconstructingfilteredtrainingdatasetstailoredtoeachmodel’scapabilities. +DifficultyEstimation. Foreachprobleminthesourcedataset,wesample16responsesfromthebasemodelandcount +thenumberofcorrectsolutions,yieldingsolve@16∈[0,16]. Weretainonlyproblemswithsolve@16∈[1,15],excluding +problemsthataretoodifficult(solve@16=0)ortriviallyeasy(solve@16=16)forthemodel. +PrompttemplateforSCIENCE. +system +Let’sthinkstepbystepandoutputthefinalanswerwithin\boxed{}. +user + +assistant +Figure9.PrompttemplateusedforRLtrainingandevaluationonSCIENCE.Theplaceholderisreplacedwiththe +actualmathematicalquestionduringfine-tuningandevaluation.Specialtokensareomittedforclarity. +Stratified Sampling. We use a stratified round-robin selection method to construct training subsets of size N ∈ +{8,32,64,512,2048}. Filteredproblemsarepartitionedinto15bins{B }15 accordingtotheirsolve@16values. Toselect +i i=1 +N problems: +1. Initialization: Setthecurrentcountofselectedproblemsn =0. +total +2. Round-RobinSelection: Whilen 1 during RL training, we observe that ∆(8) keeps the same sign for all k ∈ {1,4,8,16} across most +sat +model-benchmarkpairs,indicatingconsistentimprovementinbothpass@1andpass@k. Thisindicatesthatduringthe +pre-saturationperiod,themodelisnotjustclosingpass@kandpass@1gap. +C.3.AdditionalExperimentalResultsonLargeModels +Inthissection,wewillreportthefullevaluationresultson7Band8Bmodels. +Fig. 20 and Fig. 21 show the results of Qwen2.5-Math-7B and Llama3.1-8B-Instruct models on MATH domain with +19 + +LLMReasoningwithWeakSupervision +Table7.Graph-domaintraining(7B/8B):in-distributionbenchmarks. +| Model | Metric t(8) | QuantumLock | | LargestIsland | +| ----- | ----------- | ----------- | --- | ------------- | +sat +| | | ∆ ∆∗ | G ∆ | ∆∗ G | +| -------------------- | ---------- | --------- | --------- | --------- | +| | | sat post | sat sat | post sat | +| Qwen2.5-Math-7B | Avg@16 151 | 8.0 4.9 | 7.3 19.8 | 1.9 -10.9 | +| | Pass@4 | 22.6 1.5 | -1.3 16.5 | 6.1 16.3 | +| | Pass@8 | 26.8 2.8 | -2.4 10.7 | 9.0 29.9 | +| | Pass@16 | 30.6 5.5 | -4.1 6.8 | 8.6 41.3 | +| LLama3.1-8B-Instruct | Avg@16 29 | 10.1 7.1 | 3.2 1.8 | 1.0 -0.2 | +| | Pass@4 | 15.3 -0.0 | 6.2 1.8 | 0.0 1.6 | +| | Pass@8 | 15.4 7.4 | 12.5 1.8 | 0.0 2.8 | +| | Pass@16 | 20.1 16.0 | 25.0 1.8 | 0.0 3.1 | +in-domainandout-of-domainbenchmarks,respectively. +Fig.22andFig.23presenttheresultsofQwen2.5-Math-7BandLlama3.1-8B-InstructmodelsonSCIENCEdomainwith +in-domainandout-of-domainbenchmarks,respectively. +Fig.25providestheresultsofQwen2.5-Math-7BandLlama3.1-8B-InstructmodelsonGRAPHdomainwithmoreout-of- +domainbenchmarks. +Similar to the observations on smaller models, during the pre-saturation phases, models show generalization on both +in-domainandout-of-domainbenchmarksintermsofpass@kmetrics. Comparedtothe3Bmodel,the8BLlamamodel +exhibitsbettercross-domaingeneralization. However,LlamamodelsstillsaturatemorefasterthanQwenmodelsandshow +cleardatadependence(e.g.,Fig.22onSCIENCE). +D.RewardTypeEffect +D.1.AdditionalResultsonRewardCorruption +Rewardcorruptionimplementation. Foreachcorruptionlevelγ,weuniformlysampleaγ fractionofpromptsfromthe +N =2048trainingsetforeachmodel–domainpair. Foreachselectedprompt,wedraw96modelresponsesattemperature +1.0andselectthemostfrequentlyoccurringincorrectfinalanswer(i.e.,onethatreceiveszerorewardunderourverifier)as +thecorruptedtarget. DuringRLtraining,wereplacetheground-truthlabelsoftheselectedpromptswiththesecorrupted +labels. For Llama models and the GRAPH domain, we cap γ at 0.9 due to the base model’s inability to generate valid +solutionsevenwithextensivesampling. +Results. Fig.2showscomplementaryresultstoSection3.2. Weobservesimilarpatternsthatsomemodelsarerobustto +evenlargeamountsofrewardnoise. Inparticular,Qwenmodelsexhibitgeneralizationabilityevenwhentrainedonalmost +completelycorrupteddata;incontrast,Llamamodelstendtoshowhighrewardcurvesyetpoorergeneralizationtonewdata, +suggestingoverfittingtoincorrectresponses. +D.2.AdditionalResultsonSelf-SupervisedProxyRewards +Proxyrewardsimplementation. Weevaluatetwoself-supervisedproxyrewardsasalternativestoground-truthverification: +majorityvotingandself-certainty. +1. MajorityVotingReward. FollowingTTRL(Zuoetal.,2025),weestimatepseudo-labelsviamajorityvotingand +assignbinaryrewardsbasedonagreementwiththeconsensusanswer. Foreachprompt,wesample16responsesfrom +thepolicymodel. Themostfrequentlyoccurringansweramongthese16responsesisselectedasthepseudo-label. +Rewards are then computed as: r = 1 if the response matches the pseudo-label, and r = 0 otherwise. For policy +optimization,weusethefirst8responsestocomputeadvantages. AllotherRLhyperparametersfollowSectionB.3. +2. Self-Certainty Reward. Following Zhao et al. (2025), we use the model’s own confidence as the reward signal. +Self-certaintyisdefinedastheaverageKLdivergencebetweenauniformdistributionoverthevocabularyandthe +20 + +LLMReasoningwithWeakSupervision +model’snext-tokendistribution: +|o| +1 (cid:88) +r =Self-certainty(o|q):= KL(U∥p (·|q,o )) (1) +|o| πθ 1and0otherwise. +ForadatadistributionD,wedefinetheoverallgenerationdiversityasd (D)=E [Div (x)]. Empirically,wesample +| | | | | π | x∼D π | | | +| --- | --- | --- | --- | --- | ----- | --- | --- | +N =16outputsperpromptandestimated using8promptsfromthespecifieddataset. +π +WedefineFaithfulDiversityasthismetriccalculatedonlyonresponsesthatachieveafaithfulnessscoreof1(seebelow). +Fig.36showsanexampleoftheLM-as-judgeoutputwhenpromptedtoevaluatethesimilarityof2responses. +F.2.Quantificationofreasoningfaithfulness +Inspiredbypriorwork(Bakeretal.,2025),wedefinethefaithfulnessasaresponse’sintermediatereasoningtracecontains +allrelevantinformationandremainslogicallyconsistentwiththepredictedfinalanswer. Eachmodelrolloutproducesa +responseythatcontains(i)areasoningtraceand(ii)afinalanswer. Wewritey =(r,a),whereristhereasoningtextanda +istheextractedfinalanswer. Foreachinputpromptx,wesampley ∼π(·|x)fromthepolicy. +Faithfulnesslabeling. Wedefineadiscretefaithfulnesslabelingfunctions :X ×Y →{0,1,1},wheres (x,y) +| | | | | faithful | | 2 | faithful | +| --- | --- | --- | --- | -------- | --- | --- | -------- | +measurestheinternalagreementbetweenrandainy: +• s faithful (x,y)=1(aligned)ifthereasoningtracerconstitutesacoherentandlogicallysupportivejustificationforthe +producedanswera,regardlessofwhetheraiscorrect; +• s (x,y)= 1 (partiallyaligned)ifrexhibitsaplausibleargumentativetrajectorytowardabutcontainssubstantial +| faithful | 2 | | | | | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | +gaps,unsupportedleaps,orlocalinconsistenciesthatweakenthejustification; +22 + +LLMReasoningwithWeakSupervision +• s (x,y) = 0(misaligned)ifaisnotsupportedbyr, e.g., r contradictsa, failstoaddressthequestion, orthe +faithful +answerappearsasthe“lucky”guess. +Inpractice,weimplements (x,y)byqueryingOpenAIo3(OpenAI,2025a)asanLLM-as-a-judgewithafixedrubric +faithful +(Fig.32). OpenAIo3isusedforthistask,asopposedtoGPT-4o,duetorequiringalargermodelinordertobeableto +accuratelyreasonaboutcomplexmathematicalandscientificstepspresentinthereasoningtraces. Foralabell∈{0,1,1}, +2 +wedefinethefaithfulnessrateofpolicyπoverdatasetDas +F (l) := P (cid:2) s (x,y)=l (cid:3) . +π x∼D,y∼π(·|x) faithful +Attrainingstept,weapproximateF (l)usingN trainingprompts{x }N andK rolloutsperprompt: +πt i i=1 +N K +F(cid:98)πt (l) = +N +1 +K +(cid:88)(cid:88) 1(cid:8) +s +faithful +(cid:0) +x +i +,y +i,k +(cid:1) +=l +(cid:9) +, y +i,k +∼π t -c -e -j -o -r -P - +(·|x i -c -F - +). (3) +i=1k=1 +WeuseN =8promptsandK =16rolloutsperpromptatselectedRLcheckpointsonthespecifiedtrainingdataset. We +reportF(cid:98)πt (l)forl∈{0,1 2 - +,1}tocharacterizethedistributionofreasoningfaithfulnessunderthepolicyπ +t . - -W +Fig.37showsanexampleoftheLM-as-judgeoutputwhenpromptedtoevaluatethefaithfulnessofamodelresponsewhen +trainedontheMATHtrainingdataset. +ReliabilityofLLM-as-a-judge. TomitigatebiasfromusinganLLM-as-a-judgeforfaithfulnessevaluation,weassess +consistency across multiple LLM judges by computing Cohen’s Kappa (Cohen, 1960) across 16 faithfulness-scored +Qwen2.5-Math-1.5Boutputswhentrainedon8samplesfromtheMATHtrainingdatasetatsteps20,120and440. +Thejudgesachievesubstantialagreement(κ=0.752and0.649),indicatingconsistentfaithfulnesslabelingacrossdifferent +models. Weadditionallyconductedasmall-scalemanualevaluationtohuman-checkthefaithfulnessscoresandfindfair +alignmentwiththeLLMjudges. +F.3.Additionalresultsondiversityanalysis +Fig.30showsthesemanticdiversityofLlama3.2-3B-Instruct,Qwen2.5-1.5BandQwen2.5-Math-1.5BontheMATH-500 +evaluationdatasetthroughoutRLtraining. Qwen-Mathexhibitshigherreasoningdiversityoncorrectresponsesthanthe +othermodelsatthelaterstagesoftraining,highlightingthatRLenablesittosuccessfullylearndiverseandreliablestrategies; +coupledwithitsbetterperformanceontheevaluationdataset,thisindicatesstrongergeneralizationproperties. Inparticular, +weobservesignificantlylowerdiversityintheLlama3.2-3B-Instructmodelincomparisontoitsdiversitymetriconthe +trainingdataset(Fig.4),implyingdisagreementbetweentrainingandevaluationdistributionsandfurtherhighlightingthe +limitationsoftrainingdiversityasanindicatorofreasoningcapabilities. +F.4.Additionalresultsonfaithfulnessanalysis +Fig.33showstheproportionofresponsesthatareclassifiedasalignedormisalignedwhencalculatedwithrespecttocorrect, +incorrectorallresponses. Outofallcorrect,incorrectandoverallresponses,bothQwen2.5-1.5BandQwen2.5-Math-1.5B +showhigherproportionofalignedresponsesandlowerproportionofmisalignedresponsesthanLlama3.2-3Bwhentrained +on8samplesfromtheMATHdataset. Qwen2.5-Math-1.5Badditionallyshowsthisresultwhentrainedon8samplesfrom +SCIENCE. +G.Pre-RLIntervention +Fig.34reportpass@16resultsandFig.35reportsresultsonmorebenchmarks. +23 + +LLMReasoningwithWeakSupervision +ExamplepromptandresponseformatofSFT +UserPrompt: +Calculatethelimitofthefunction: +arcsin(3x) +| | lim | √ √ | | | +| --- | --- | ------ | --- | --- | +| | x→0 | 2+x− 2 | | | +Assistant: + +Okay,soIneedtofindthelimit... +Substitutingx=0givestheindeterminateform 0,so... +0 +Rationalizingthedenominator,weget... + +Thelimitiscalculatedasfollows: +1. Recognizetheindeterminateform: substitutingx=0yields +0 +, +0 +whichsuggestsusingrationalizationorl’Hoˆpital’srule. +2. Rationalizethedenominator: +| | √ | √ | √ | √ | +| ---------- | ------ | ------------ | ------- | ------- | +| | | | (cid:0) | (cid:1) | +| arcsin(3x) | 2+x+ | 2 arcsin(3x) | 2+x+ | 2 | +| √ | √ · √ | √ = | | . | +| 2+x− | 2 2+x+ | 2 | x | | +3. Splitthelimit: +| | | √ | √ | | +| --- | ---------- | ---------- | ------- | --- | +| | arcsin(3x) | (cid:0) | (cid:1) | | +| | lim | · lim 2+x+ | 2 . | | +x +| | x→0 | x→0 | | | +| --- | --- | --- | --- | --- | +4. Evaluateeachpart: +• Using +arcsinz +| | | lim =1, | | | +| --- | --- | ------- | --- | --- | z +z→0 +thefirstlimitbecomes3. +√ +• Thesecondlimitevaluatesto2 2. +5. Combinetheresults: √ √ +3·2 2=6 2. +FinalAnswer: +√ +6 2 +Figure12. ExamplepromptandresponseformatofSFT.InThinkingSFT,themodelistrainedwithreasoningtracesenclosedby +and,whereasNon-ThinkingSFTomitsthem. +24 -p - -l -l -i - -) -a - -• - - i +LLMReasoningwithWeakSupervision +| | 8 samples | | 32 samples | | 64 samples | | | 512 samples | | 2048 samples | | +| ------------ | ------------------- | --- | ------------ | --- | ---------- | ------- | --- | ----------- | --------------- | ------------ | --- | +| | draweR gniniarT 1.0 | | )%( 005-HTAM | | | | | | | | | +| | | | 72 | | | 52 | | | )%( draH-PCS 36 | | | +| htaM-5.2newQ | 0.8 | | | | | )%( CMA | | | | | | +| | | | | | | 48 | | | 30 | | | +69 +| | 0.6 | | | | | | | | 24 | | | +| ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| B5.1- | | | | | | 44 | | | | | | +| | | | 66 | | | | | | 18 | | | +| | 0.4 | | | | | 40 | | | | | | +12 +| | 0.2 | | 63 | | | 36 | | | | | | +| ------------ | ----------------- | ------- | --------------- | -------------- | ------- | --- | -------------- | ------- | --------------- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +| | draweR gniniarT 1 | | )%( 005-HTAM 60 | | | | | | | | | +| B5.1-5.2newQ | | | | | | 30 | | | )%( draH-PCS 20 | | | +50 +)%( CMA 24 +15 +| | | | 40 | | | 18 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +10 +| | | | 30 | | | 12 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | | | 20 | | | 6 | | | 5 | | | +| | 0 | | 10 | | | 0 | | | 0 | | | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +draweR gniniarT 1.0 +| B3-5.2newQ | | | )%( 005-HTAM 66 | | | | | | )%( draH-PCS | | | +| ---------- | --- | --- | --------------- | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | | | | | | 40 | | | 25 | | | +| | 0.8 | | | | | )%( CMA | | | | | | +| | | | 63 | | | 36 | | | 20 | | | +0.6 +| | | | 60 | | | 32 | | | 15 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.4 | | | | | 28 | | | 10 | | | +57 +0.2 +| | | | 54 | | | 24 | | | 5 | | | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +B3-2.3amalL Training Steps Training Steps Training Steps Training Steps +draweR gniniarT +| tcurtsnI- | 1.0 | | )%( 005-HTAM | | | | | | )%( draH-PCS 15 | | | +| --------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | --------------- | --- | --- | +| | | | 54 | | | 30 | | | | | | +| | 0.8 | | | | | )%( CMA | | | 12 | | | +| | 0.6 | | 51 | | | 25 | | | 9 | | | +| | | | 48 | | | 20 | | | 6 | | | +0.4 +| | | | | | | 15 | | | 3 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.2 | | 45 | | | | | | | | | 0 - 1 - -g +10 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure13.ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonMATHdomain.Resultsareaveraged +overthreeindependentruns,withshadedregionsindicatingerrorbars. Verticaldashedlinesdenotethesaturationstepforeachdata +scaleifitsaturatesbefore496gradientsteps.Llamamodelsexhibitrapidsaturationinsmall-sampleregimesandrelyheavilyondata +scale.Incontrast,Qwenmodelsyieldcomparableperformanceacrossvaryingsamplesizes,characterizedbyextendedsaturationperiods. +Evaluationresultsinthisfigurearebasedongreedydecoding. +25 + +LLMReasoningwithWeakSupervision +| | 8 samples | | 32 samples | | 64 samples | | | 512 samples | | 2048 samples | | +| --- | --------- | --- | ---------- | --- | ---------- | --- | --- | ----------- | --- | ------------ | --- | +)%( dnomaiD AQPG +draweR gniniarT 1.0 +| | | | )%( draH-PCS 40 | | | 30 | | | )%( 005-HTAM | | | +| ------------ | --- | --- | --------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | +| htaM-5.2newQ | | | | | | | | | 72 | | | +| | 0.8 | | 35 | | | 25 | | | | | | +| | 0.6 | | | | | 20 | | | 69 | | | +| B5.1- | | | 30 | | | | | | | | | +| | 0.4 | | | | | 15 | | | | | | +| | | | 25 | | | | | | 66 | | | +10 +| | 0.2 | | 20 | | | | | | | | | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +| | | | | | | 5 | | | 63 | | | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +| | draweR gniniarT | | | | | 30 | | | | | | +| ------------ | --------------- | --- | --------------- | --- | --- | --- | --- | --- | --------------- | --- | --- | +| | 1 | | )%( draH-PCS 30 | | | | | | )%( 005-HTAM 60 | | | +| B5.1-5.2newQ | | | | | | 24 | | | | | | +| | | | 24 | | | | | | 50 | | | +| | | | 18 | | | 18 | | | 40 | | | +| | | | 12 | | | 12 | | | 30 | | | +| | | | 6 | | | 6 | | | | | | +20 +| | 0 | | 0 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +draweR gniniarT +| B3-5.2newQ | 1.0 | | )%( draH-PCS | | | | | | )%( 005-HTAM | | | +| ---------- | --- | --- | ------------ | --- | --- | --- | --- | --- | ------------ | --- | --- | +| | | | 30 | | | 33 | | | 66 | | | +0.8 +63 +| | 0.6 | | 24 | | | 30 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 18 | | | 27 | | | 60 | | | +0.4 +| | | | 12 | | | 24 | | | 57 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.2 +| | | | 6 | | | 21 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +54 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +B3-2.3amalL draweR gniniarT +| tcurtsnI- | 1.0 | | | | | | | | )%( 005-HTAM | | | +| --------- | --- | --- | ------------ | --- | --- | --- | --- | --- | ------------ | --- | --- | +| | | | )%( draH-PCS | | | 30 | | | 54 | | | +| | 0.8 | | 32 | | | | | | | | | +| | | | | | | 27 | | | 52 | | | +24 +| | 0.6 | | | | | 24 | | | 50 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| | 0.4 | | | | | 21 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 8 | | | | | | 48 | | | +18 +| | 0.2 | | | | | | | | 46 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure14. ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonSCIENCEdomain. Resultsare +averagedoverthreeindependentruns,withshadedregionsindicatingerrorbars.Verticaldashedlinesdenotethesaturationstepforeach +datascale.Thepre-saturationphaseyieldssimilargainsacrossallsamplesizes;however,afterthesaturationpoint,largersamplesizes +demonstratedistinctbenefits.Modelsexhibitsignificantlydifferentsaturationdynamicsonsmallsamples.Evaluationresultsinthisfigure +arebasedongreedydecoding. +| | 8 samples | | 32 samples | | 128 samples | | | 256 samples | | 882 samples | | +| --- | --------- | --- | ---------- | --- | ----------- | --- | --- | ----------- | --- | ----------- | --- | +htaM-5.2newQ +| | draweR gniniarT 1 | | )%( kcoL mutnauQ 50 | | | )%( dnalsI tsegraL | | | 82 | | | +| --- | ----------------- | --- | ------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | +)%( 005-HTAM +| | | | 40 | | | 32 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +80 +| B7- | | | | | | 24 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 30 | | | | | | 78 | | | +| | | | 20 | | | 16 | | | | | | +76 +8 +10 +74 +| | 0 | | 0 | | | 0 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +)%( kcoL mutnauQ +| B8-2.3amalL | draweR gniniarT 1.0 | | | | | )%( dnalsI tsegraL | | | | | | +| ----------- | ------------------- | --- | --- | --- | --- | ------------------ | --- | --- | ------------ | --- | --- | +| tcurtsnI- | | | | | | 6 | | | )%( 005-HTAM | | | +| | | | 45 | | | | | | 54 | | | +0.8 +4 +0.6 +| | | | 30 | | | | | | 52 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.4 | | | | | 2 | | | | | | +| | | | 15 | | | | | | 50 | | | +0.2 +| | | | 0 | | | 0 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure15. ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonGRAPHdomain. Weuselarger +models(Qwen2.5-Math-7B,Llama-3.1-8B-Instruct)duetoincreasedtaskdifficulty.Resultsareaveragedoverthreeindependentruns, +withshadedregionsindicatingerrorbars.Verticaldashedlinesdenotethesaturationstepforeachdatascale.Qwenmodelalsosaturates +fasterherethaninotherdomains.Largerdatasetsyieldcleargainsinthepost-saturationphases.Evaluationresultsinthisfigurearebased +ongreedydecoding. +26 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +88 +80 +| 60 | | | | | | | | | 88 | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 005-HTAM | | | | | | 80 | | | | | | +70 +| 45 | | | | | | 72 | | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| | | | | | | 64 | | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +| | | | 50 | | | 56 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +64 +15 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 40 | | | 60 | | | 70 | | | 80 | | | +50 +| 30 | | | | | | 60 | | | 70 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +CMA +| | | | 40 | | | 50 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 20 | | | | | | | | | 60 | | | +| | | | 30 | | | 40 | | | 50 | | | +10 +| | | | 20 | | | 30 | | | | | | +| --------------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| htaM avreniM 24 | | | | | | | | | 48 | | | +| | | | 36 | | | 42 | | | | | | +| 18 | | | 30 | | | 36 | | | 42 | | | +24 +| 12 | | | | | | 30 | | | 36 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 18 | | | 24 | | | 30 | | | +6 +| | | | 12 | | | 18 | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +hcneBdaipmylO 30 +| | | | 40 | | | 48 | | | 56 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +24 +| | | | 32 | | | 40 | | | 48 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +18 +| | | | | | | 32 | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 12 | | | 24 | | | | | | | | | +| | | | 16 | | | 24 | | | 32 | | | +6 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +40 +15 +| 4202 EMIA | | | 24 | | | 32 | | | 40 | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +10 +| | | | 18 | | | 24 | | | 30 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 5 | | | 12 | | | 16 | | | | | | +20 +| 0 | | | 6 | | | | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | | | | 8 | | | 10 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +10.0 +| | | | 16 | | | 20 | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5202 EMIA 7.5 +12 +| | | | | | | 15 | | | 18 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5.0 +| | | | 8 | | | 10 | | | 12 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +2.5 +| | | | 4 | | | 5 | | | 6 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.0 +| | | | 0 | | | 0 | | | 0 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure16.Fullin-domainbenchmarkevaluationresultsfortheMATHdomainacrossmultiplemodels.Verticaldashedlinesdenote +thesaturationstepforeachdatascale. +27 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +90 +| dnomaiD AQPG | | | 60 | | | 75 | | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 24 | | | | | | | | | 75 | | | +| | | | 45 | | | 60 | | | | | | +| 16 | | | | | | 45 | | | 60 | | | +30 +| 8 | | | | | | 30 | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 32 | | | | | | | | | 80 | | | +| | | | 45 | | | 60 | | | | | | +draH-PCS 24 +60 +45 +| 16 | | | 30 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 30 | | | 40 | | | +| 8 | | | 15 | | | | | | | | | +| | | | | | | 15 | | | 20 | | | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | 25 | | | 30 | | | | | | +16 +hcneB ecneicS +| | | | 20 | | | 25 | | | 30 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| | | | | | | 20 | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 8 | | | 15 | | | | | | | | | +| | | | | | | 15 | | | 18 | | | +| 4 | | | 10 | | | | | | | | | +10 +| | | | 5 | | | | | | 12 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 60 | | | 80 | | | | | | | | | +| | | | | | | 80 | | | 90 | | | +ICS ULMM +| 45 | | | 60 | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | 40 | | | | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +45 +| 15 | | | 20 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| 0 | | | | | | | | | 30 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +60 +20 +| AQPGrepuS | | | 40 | | | | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 45 | | | 60 | | | +| 15 | | | 30 | | | | | | | | | +| 10 | | | | | | | | | 45 | | | +| | | | 20 | | | 30 | | | | | | +| 5 | | | | | | | | | 30 | | | +| | | | 10 | | | 15 | | | | | | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure17.Fullin-domainbenchmarkevaluationresultsfortheSCIENCEdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +28 -5, -g - -1 - -dVIA41 A3AWS S302:10S3 1V1if1.1031.1H3lV ONOISIH - -SIN :37VOS - -— - -I - -g - -§ - - all significant resources within the APE; and (3) to evaluate these resources according to -the National Register of Historic Places criteria. - -The survey methodology consisted of a field survey and background research on the -project area. A NCDOT staff architectural historian conducted a field survey in August -and October 1999, by car and on foot. All structures over 50 years of age in the APE -were photographed and keyed to an area map. - -Background research was conducted at the Western office of the State Historic -Preservation Office, Asheville; the North Carolina State Library, Raleigh; and the Burke -County Court House in Morganton. - -A meeting was held with Ms. Jennifer Martin at the State Historic Preservation Office on -August 20, 1999 to discuss the eligibility of two of the historic resources, the William -Jones Arney House and the Marshall Arney House. It was agreed at that time the two -properties were potentially eligible for the National Register of Historic Places. - -SUMMARY FINDINGS OF THE SURVEY - -The project proposes to pave and improve SR 1424, Johns River Road and SR 1438, -Corpening Bridge Road in Quaker Meadows and Upper Creek townships, Burke County. -In letters dated July 9, 1999 and August 10, 1999, the State Historic Preservation Office -identified four structures of historical or architectural importance within the general -project area, the Marshall Arney House, the William Arney House, the Arney Chapel and -the Corpening House (See Appendix A). Two additional historic properties, not -mentioned in the letters, are located within the project area: a late nineteenth-century log -house and an early twentieth-century frame house. No properties listed on the National -Register were located within the APE for the project. Of the four historic properties -mentioned in the letter, one, the Arney Chapel, is located outside the APE. Of the five -historic properties located within the APE, three—the Corpening House, the nineteenth- -century log house, and the early twentieth-century frame house --were determined not -eligible for the National Register. The Marshall Arney House and the William Arney -House were determined to be eligible for the National Register. Criterion Consideration -G, for properties that have achieved significance within the last fifty years, states that -properties less than fifty years of age may be listed on the National Register only if they -are of exceptional importance or if they are integral parts of districts eligible for the -National Register. There are no properties in the APE that qualify for the National -Register under Criterion Consideration G. - -Properties Listed on the National Register: - -None - -Ilistoric Architectural Resources Survey Report: iVCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nancy Ian Dolsen - -Page 4 - - Properties Listed on the North Carolina State Study List: - -None - -Properties Evaluated and Considered Eligible for the National Register: - -William Jones Amey House, page 8 -Marshall Amey House, page 15 - -Properties Evaluated and Considered Not Eligible for the National Register: - -Corpening House, page 21 -Twentieth-century frame house, page 24 -Nineteenth-century log house, pay 28 - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 5 - - HISTORIC CONTEXTS AND BACKGROUND INFORMATION - -During the late seventeenth century, occupation of the land of present-day Burke County -was contested between two native-American tribes, the Cherokee and the Catawba. The -area eventually became a buffer zone between the two tribes. The French and Indian War -(1754-1763) and the American Revolution had a catastrophic effect on both tribes, -forcing the Cherokees further into the mountains, and reducing the Catawba population to -250 members by 1784. - -Transient hunters and trappers of European descent settled in the county during the -middle of the eighteenth century. In 1777 the General Assembly approved the formation -of a new county named Burke in the western portion of the state that encompassed all or -part of 16 present counties. Burke was named for Thomas Burke (1747-1783), a -member of the Continental Congress and third Governor of North Carolina.' Burke -County achieved its present size of 511 square acres in 1885. - -According to an architectural survey of the county completed in 1983-1984, there are -three distinct regions in the county: the rich bottom lands along the Catawba, Johns, -Upper and Lower Creeks, and Warrior Forks rivers; the remote dead-end coves of the -South Mountains and Blue Ridge; and the rolling foothills interspersed with small -valleys.2 The subject project is located along the bottom land of the Johns River in -Quaker Meadows and Upper Creek townships. - -The bottom lands contained the richest agricultural soil in the county, and were the site of -antebellum plantations. Most of the county's large, brick plantation homes dating to the -first half of the nineteenth century stand along the Johns and Catawba rivers in Quaker -Meadows and Upper Creek townships, including Bellevue (1823-1826), Cedar Grove -(1825), and Pleasant Valley (ca. 1820; ca. 1975). In Burke, the typical plantation of 1860 -consisted of 850 acres of improved land and thousands of acres of unimproved land. -The typical 1860 Burke County farm, however, contained only 60 acres of cultivated land -and only 60 out of 1,239 families owned more than 100 acres of improved land.3 - -After the Civil War, the land-use pattern of ownership changed. The large-scale -plantations were dissolved, and there was an increase in the number of small-scale farms, -from 500 in 1860 to 1,648 in 1880. At the same time, the average farm size dropped to -27 acres.4 - -The project area is the site of five post-bellum houses that date to this period of -subdividing larger plantation properties. The houses stand on land first owned by one of -Burke County's first European settlers, John Perkins, who first explored the county in the - -I Entry for Thomas Burke, Biographical Directory of the United States Congress, I774-Present, Website: -Bioguide.conress.gov. -2 J. Randall Cotton, et at., Historic Burke: An Architectural Sites Inventory of Burke County (Morganton, -NC: Historic Burke Foundation, Inc., 1987) 13. -3 Cotton, et at, 23. -4 Cotton, et al, 27. - -Historic Architectural Resources Survey Report: NCDOT. SR 1424. Burke County: November 1999 -Nancy Van Dolsen - -Page 6 - - 1750s. These five houses represent a range of building types constructed in the county -during the last quarter of the nineteenth century: two frame T-shaped houses, a small -two-room log house, and two square-shaped two-story frame dwellings. - -At the end of the twentieth century, much of the land in the county is held by three -property owners: the US Government (Pisgah Forest), the state (South Mountain State -Park) and Carolina Centers, a timber company. Carolina Centers presently owns the land -upon which the Corpening, Marshall Amey, and William Jones Arney houses stand. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 7 - - PROPERTY INVENTORY AND EVALUATIONS - -William Jones Arney House - -Location: (cid:9) -The William Jones Arney House is located on the northwest corner -of the junction of SR 1424 and SR 1438. The house sits on a 23,136.10-acre tract -owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a timber -company. The property was formerly owned by Duke Power, Inc. - -The William Jones Arney House sits within a cleared area, at the - -Setting: (cid:9) -top of a rise, facing east to Johns River. A short dirt lane leading from Johns -River Road runs south of the house, ending at a small frame granary west of the -dwelling. - -Description: - -Dwelling: The William Jones Arney House, dating to ca. 1890, is a two- -and-a-half story frame residence with a T-shaped plan. The exterior of the -house is clad with narrow weatherboards, and a metal roof covers the -building. To the rear of the building is a kitchen connected to the main -house by a covered walkway. The kitchen stands one-story and has a -shed-roofed porch on each eave elevation. The windows in the kitchen -contain six-over-six sash and those in the main block of the house have -two-over-two sash. A one-story tripartite bay window adorns the south -gable end wall of the ell. Pointed arched louvered windows are located at -the peaks of the gable end walls. Two interior brick chimneys are located -at the center of the ells, providing an interior stack to each room in the -house. - -The interior walls of the house are sheathed in horizontal boards, as are the -ceilings. The house (excluding the kitchen) has four rooms on the first -and four rooms on the second floor. - -Well House: A concrete block well house stands south of the south porch -of the kitchen. The well house is approximately 3 feet in height and is -covered with a corrugated metal roof. - -Granary: To the west of the house stands a frame granary also constructed -ca. 18900. The granary building is a gable-front building with a central -door. The building is weatherboarded and has a metal roof. An open side -shed was constructed onto the east eave wall, and a shed roof porch is -located on the south elevation. - -Background Information: (cid:9) -land ca. 1885. Arney, a descendent of one of the pioneering families in the -county, both farmed and owned a general store (now demolished). The store - -William Jones Arney (1849-1911) moved to this - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy ran Dolsen - -Page 8 - - stood to the west, at the intersection of Corpening Bridge and Piedmont roads. -Arney, and his wife Margaret McCall, owned over 500 acres in the area along -John's River. The couple had nine children? - -At the death of William and Margaret Arney, ownership of the part of the farm -upon which this house stood was transferred to George Phifer Arney who sold -240 acres to Western Carolina Power for $100 in 1922.6 The property is presently -owned by Carolina Centers, LLC. - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966, as amended, the William Jones Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A, since it is not representative of a significant trend in the history of the -area. The property is no longer under cultivation, and all but one of the -agricultural outbuildings have been demolished or removed. - -The William Jones Arney House is considered not eligible for the National -Register of Historic Places under Criterion B because it is not associated with any -individuals of specific historical significance. - -The William Jones Arney House is considered eligible for the National Register -under Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is a regional form built by farmers of middle economic status. The Arney -House is a good representation of this form, and exhibits excellent integrity and -retention of setting. The granary building is also representative of a vernacular -building form, and exhibits good integrity of design, materials, workmanship, -and setting. - -The building is not likely to yield information important to the history of building -technology; therefore it is not eligible for the National Register under Criterion D. - -Boundaries: The proposed National Register boundaries comprise approximately -2.8 acres that surround the house and its accompanying outbuilding. The land -was deeded to George Phifer Arney, and is the portion of the William Jones -Arney estate upon which the subject structures stand. This parcel is part of parcel -JR-21, which is a portion of the larger Carolina Centers tract. - -Boundary Justification: (cid:9) -include the house, outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. - -The proposed historic boundaries were drawn to - -5 William Jones Amey House, Historic Resource Survey Form, Burke County, 1983. Asheville, NC: -Western Office, Division of Cultural Resources. -6 Burke County Deed Book T4:293. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 -Nancy Van Dolsen - -Page 9 - - Figure 3: Tax Map, William Jones Arney House - -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 10 - - Figure 4: Site plan, William Jones Arney House - -Historic Architectural Resources Survey Report,- NCD07; SR 1-114, Burke County: November 1999 (cid:9) -Nancy l'an Dolsen - -Page 11 - - — - -— (cid:9) - -_ - -- (cid:9) - -41E - -re-7_. - — -liq -- -- . (cid:9) - -- - -_ - -_ (cid:9) - -_ - -_- - -- --- - -— s- — - -_--..------ - -- - -- (cid:9) -77--- - --_ - -- (cid:9) - -—. (cid:9) - -—______-- -- ------_--,- --- ---------- - -_ - -_ --- — - -- -- - --4-. - -, - --4 - ----_ - -a - -Photograph 1: William Jones Amey House, facing southwest - -Photograph 2: William Jones Amey House, facing northeast - -Historic Architectural Resources Survey Report: NCDOT SR 1424, Burke County: November 1999 -Nancy Van Dolsen - -Page 12 - -(cid:9) -(cid:9) - . - -:t. (cid:9) - -. --- - -, - -,,J-, (cid:9) - -, - -••••- (cid:9) - -, (cid:9) - -i - --- (cid:9) -_ -__ - --.------ (cid:9) - -----.•-•_:, (cid:9) ----- - -(cid:9) ,- - --.? - ---- -_= - -- (cid:9) - -_ - -- - -. (cid:9) - --- (cid:9) -- (cid:9) - -- -•- -_ --4--.- -— -- (cid:9) - -------,_ - -- - --_ -_.----e- --, (cid:9) - (cid:9) --- (cid:9) - -l'---0` (cid:9) - -_ (cid:9) -_ - -_ - -_ - -- (cid:9) - -- (cid:9) - -- - -- --7---- - --__-- (cid:9) -------- -- (cid:9) - -- (cid:9) - (cid:9) - -- -- - -- (cid:9) - -- -- - -.-• =,---• -.-__----------- - -- (cid:9) - (cid:9) - -- - -, - -t t -, (cid:9) -- (cid:9) -- (cid:9) -= (cid:9) - -_..----, (cid:9) - -. (cid:9) - -_ - -- (cid:9) - -Photograph 3: William Jones Arney house, view facing northeast - -Photograph 4: Granary, view facing north - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 (cid:9) -:Vancy Van Dolsen - -Page 13 - -(cid:9) - " -. -• - -: -• -• - -0 - -. - -0 - -c -e -) - -z - -C - -. -) - -E - -- -: - -6: -1 - -;1 - -8 - -4 -. -1 - - Marshall Arney House - -The Marshall Arney House is located on the west side of SR 1424. - -Location: (cid:9) -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. - -The Marshall Arney House stands in a clearing of a wooded area - -Setting: (cid:9) -halfway up a hillside on a knoll approximately 100 feet wide in a bend of the -Johns River. Johns River Road is located on the next downhill terrace. East of -the roadway the land slopes down to the river. - -Description: - -Dwelling: Built ca. 1900, the Marshall Arney House is a two-and-a-half- -story, T-shaped house and features a plan similar to his father's house, the -William Jones Arney House. The primary elevation faces south. The -Marshall Arney House has a wrap-around porch with chamfered posts. -Similar to the William Jones Arney House, the Marshall Arney House has -pointed arched louvred windows at the peaks of the end gables. The end -gables are pedimented. Two early twentieth-century additions were -constructed onto the north gable end; a two-story gable wing, and a one- -story shed-roof ell. The building is weatherboarded, and retains most of -its original two-over-two sash windows. - -Springhouse: To the west of the house stands a gable-front springhouse -banked into the side of the hill. The springhouse has a stone lower level -with a stuccoed finish. The interior of the upper floor is sheathed in -diagonal boards, and the exterior is weatherboarded. The gable roof is -cantilevered over the façade of the springhouse. - -Background Information: (cid:9) -Constructed by Marshall Arney on land bequeathed -to him by his father, William Jones Amey, this house is a substantial dwelling for -the area. Marshall Arney was a land agent for Duke Power, served for twenty- -five years on the Burke County School Board and was instrumental in -consolidating and erecting new schools in the county during the early twentieth -century. In April 1922 Marshall Arney and his wife, L. Inez Arney, sold 250 -acres, including the subject structure, to Western Carolina Power. - -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, the Marshall Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A. since it is not representative of a significant trend in the history of the -area. - -historic Architectural Resources Survey Report: NCDOT. SR I424. Burke County; November 1999 (cid:9) -Nancy i'an Dolsen - -Page 15 - - The Marshall Amey House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. - -The Marshall Arney House is considered eligible for the National Register under -Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is found in the region as a house type built by farmers of middle -economic status and part-time farmers with jobs off of the farm. This type of -farmhouse, if ever once prevalent, has become rare due to the large tracts of -former agricultural land that has been acquired for forest and timber land. The -Amey House is a good representation of this type of farm house and exhibits -integrity of materials, workmanship, setting, location, feeling, and association. -The springhouse building is also representative of a vernacular building form and -exhibits good architectural integrity. - -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - -Boundaries: The proposed National Register boundaries comprise -approximately 1.9 acres that surround the house and its accompanying -outbuilding. The land was deeded to Marshall Arne y at death of his father. This -tract is part of parcel JR-18, which is a portion of the larger Carolina Centers -tract. - -Boundary Justification: (cid:9) -include the house. outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. - -The proposed historic boundaries were drawn to - -Historic Architectural Resources Survey Report: ACDOT SR 1424. Burke County: November 1999 (cid:9) -Nancy Ilzn Dolsen - -Page 16 - - Figure 5: Tax Map, Marshall Arney House - -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke Counr.t...,- November 1999 (cid:9) -Nancy ran Dolsen - -Page 17 - - L -T - -. - -m -u - -l - -t - -4 -C - -C -a -) - -' -• - -C - -• - -• -• -• -• -• - -• -• -• - -c -. - -.VC1)Or SR 1414, Bur ke Con - - :Vovember 19 99 - - Corpening House - -The Corpening House stands on the west side of SR 1424. about - -Location: (cid:9) -1.5 miles south of the junction with SR 1405. The house sits on a 23,136.10-acre -tract owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a -timber company. The property was formerly owned by Duke Power, Inc. - -The Corpening House stands on a slight rise overlooking the wide - -Setting: (cid:9) -bottomland along a bend in the Johns River. The land around the house is now -used to grow trees and shrubs for sale to nurseries and landscaping companies. - -Description: - -The ca. 1900 features a common form for the period: a two- - -Dwelling: (cid:9) -story, square, boxy building with a pyramidal roof. The facade features a -large central pediment over the second-floor windows, and a smaller -pediment covers the center of the one-story porch across the facade. -Turned posts support the porch roof. The exterior of the house is clad in -narrow weatherboards. The window sashes have been replaced with one- -over-one double-hung sash. Since 1983, the porch has lost its decorative -brackets, and the back porch has been removed andior remodeled. A shed -roof addition covers the rear elevation. No outbuildings stand on the -property. - -Background Information: (cid:9) -the holdings of John Perkins. one of the county's first settlers of European -descent. The property is now part of the large landholdings of the Carolina -Centers corporation. - -The Corpening House stands on land once part of - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, the Corpening House is considered not eligible for the National -Register ofl-listoric Places under Criterion A, since it is not representative of a -significant trend in the history of the area. - -The Corpening House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. - -The Corpening House is considered not eligible for the National Register under -Criterion C since the building has lost integrity and does not embody distinctive -characteristics of a significant building type. The building no longer retains its -original porch trim, window sash, and the form has been compromised by the -alterations to the rear of the dwelling. - -Historic Architectural Resources Survey Report: NCDOT1SR 1424, Burke County: November 1999 (cid:9) -Nancy lint Dolsen - -Page 21 - - The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - - -Historic Architectural Resources Survey Report; NCD071SR 1424, Burke County: November 1999 -Nancyl'an Dohen - -Page 22 - - Photograph 1: Corpcning House, facing east - -Photograph 2: Corpening House, facing north - -Historic Architectural Resources Survey Report: NCD07; SR 1424, Burke County: November 1999 (cid:9) -Nancy ("an Dohen - -Page 23 - - Early Twentieth-Century Frame House - -This frame house stands on the east side of Corpening Bridge - -Location: (cid:9) -Road. SR 1438. approximately 1.5 miles east of the bridge over the Johns River. -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. - -Setting: (cid:9) -outbuildings stand south of the house, down a slope, and are no longer used. - -This frame house sits in a wooded area that was once farmed. The - -Description: - -A two-and-a-half story, three-bay, double-pile dwelling, - -this house features a hip roof with a central gable dormer. The house has -an exposed extended rafter roof and is covered with weatherboard. A one- -story hip-roof porch covers the facade of the building. The house retains -its six-over-six sash windows. A one-story gable roof ell is located on the -rear of the house. - -Outbuildings: To the east of the house stands a frame domestic -outbuilding. The gable-front building stands on a cement block -foundation. and is covered with weatherboard. An off-center door is -located on the gable end. To the south of the house are three frame -outbuildings in poor condition. The outbuildings have metal roofs and -narrow board siding. The buildings appear to have served as a granary, -stable, and a workshop. - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, this house is considered not eligible for the National Register of -Historic Places under Criterion A, since it is not representative of a significant -trend in the history of the area. - -This house is considered not eligible for the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. - -This house is considered not eligible for the National Register under Criterion C -since the building does not embody distinctive characteristics of a significant -building type. The outbuildings have lost their architectural integrity and their -setting. - -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen - -Page 24 - - Photograph I: Early Twentieth-Century Frame House, facing northeast - -Photograph 2: Early Twentieth-Century Frame House, facing north - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Do/sell - -Page 25 - - Photograph 3: Frame Domestic Outbuilding, facing northeast - -Photograph 4: Frame Granary, facing south - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 -Nancy I'an Dolsen - -Page 26 - - 0 - -t -t -j - -0 -. - -k -. -r -; - -C - -I -) - -- -0 - -. -2 -4 - -- -a - -r -/ -1 - -0 - -C - -. -. - -, - -0 +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| ------------ | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| | | | 60 | | | | | | 90 | | | +| dnomaiD AQPG | | | | | | 75 | | | | | | +24 +| | | | 45 | | | 60 | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +18 +| | | | 30 | | | 45 | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| 6 | | | | | | 30 | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 20 | | | | | | 60 | | | 80 | | | +40 +draH-PCS +| 15 | | | 30 | | | 45 | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 10 | | | | | | 30 | | | | | | +| | | | 20 | | | | | | 40 | | | +5 +| | | | 10 | | | 15 | | | 20 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 16 | | | 24 | | | 28 | | | 32 | | | +hcneB ecneicS +| | | | | | | 24 | | | 28 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 12 | | | 20 | | | | | | | | | +| 8 | | | 16 | | | 20 | | | 24 | | | +| | | | | | | 16 | | | 20 | | | +| 4 | | | 12 | | | | | | | | | +| 0 | | | 8 | | | 12 | | | 16 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 60 | | | 80 | | | | | | | | | +| | | | | | | 80 | | | 90 | | | +ICS ULMM +| 45 | | | 60 | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | | | | | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +40 +| 15 | | | | | | | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| | | | | | | 20 | | | 30 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 0 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure18.Fullout-of-domainbenchmarkevaluationresultsfortheMATHdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +| | | Qwen2.5-Math-1.5B | | | Qwen2.5-1.5B | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | --------- | ------------ | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +90 +| 60 | | | 75 | | | | | | 90 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +005-HTAM +| | | | | | | 75 | | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 45 | | | 60 | | | | | | | | | +| 30 | | | | | | 60 | | | 70 | | | +45 +60 +| 15 | | | 30 | | | 45 | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +60 +| 40 | | | | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | 45 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| CMA | | | | | | | | | 60 | | | +45 +20 +| | | | 30 | | | | | | 45 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 10 | | | | | | 30 | | | | | | +| | | | 15 | | | | | | 30 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +40 +48 +| htaM avreniM 24 | | | | | | 40 | | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +32 +| 18 | | | | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +32 +24 +| 12 | | | | | | 24 | | | 32 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| 6 | | | | | | | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +8 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +30 +| hcneBdaipmylO | | | | | | 50 | | | 60 | | | +| ------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 24 | | | | | | | | | 50 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 18 | | | 30 | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +12 +| | | | 20 | | | | | | 30 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 6 | | | | | | 20 | | | | | | +| | | | 10 | | | | | | 20 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure19.Fullout-of-domainbenchmarkevaluationresultsfortheSCIENCEdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +29 + +LLMReasoningwithWeakSupervision +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 | N=2048 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | ---- | --------- | ------ | ---- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| 80 | | | 93 | | | 95.0 | | | 97.5 | | | +90 +| | | | 87 | | | 92.5 | | | 95.0 | | | +| ----------- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | --- | +| 005-HTAM 70 | | | | | | 90.0 | | | 92.5 | | | +84 +| 60 | | | 81 | | | 87.5 | | | 90.0 | | | +| --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | --- | +| | | | 72 | | | 80 | | | | | | +84 +| 50 | | | 68 | | | 76 | | | 80 | | | +| --- | --- | --- | ----- | --- | --- | ------- | --- | --- | ----- | --- | ------- | +| | | | 64 | | | 72 | | | | | | +| | | | | | | 68 | | | 76 | | | +| 40 | | | 60 | | | | | | 72 | | | +| | | | 56 | | | 64 | | | | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 60 | | | | | | | | | 92 | | | +| | | | 76 | | | 82.5 | | | | | | +| | | | 7 2 | | | 8 0 . 0 | | | 8 8 | | | +| | | | | | | 7 7 . 5 | | | 8 4 | | | +| 45 | | | 6 8 | | | 7 5 . 0 | | | 8 0 | | | +| CMA | | | 64 | | | | | | | | | +| | | | | | | 72.5 | | | 76 | | | +| 30 | | | 4 5 | | | 54 | | | 66 | | | +4 0 +| | | | 35 | | | 48 | | | 60 | | | +| ------------ | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 15 | | | 30 | | | 42 | | | 54 | | | +| | | | 25 | | | 36 | | | 48 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 32 | | | | | | | | | 57 | | | +| htaM avreniM | | | | | | 51 | | | | | | +| 28 | | | 44 | | | | | | | | | +| | | | | | | 48 | | | 54 | | | +| 24 | | | 40 | | | | | | | | | +| | | | | | | 45 | | | 51 | | | +20 +36 +| | | | | | | 42 | | | 48 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| | | | 32 | | | 39 | | | | | | +| ---------------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ------- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| hcneBdaipmylO 40 | | | 56 | | | 63 | | | 70.0 | | | +| | | | 5 2 | | | 6 0 | | | 6 7 . 5 | | | +6 5 . 0 +| 32 | | | 4 8 | | | 5 7 | | | 6 2 . 5 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | +| | | | 4 4 | | | 5 4 | | | | | | +| 24 | | | | | | 5 1 | | | 6 0 . 0 | | | +| | | | 32 | | | 40 | | | 48 | | | +36 +| 16 | | | 28 | | | | | | 44 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | 24 | | | 32 | | | 40 | | | +| | | | 20 | | | 28 | | | 36 | | | +| 8 | | | | | | | | | 32 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 32 | | | | | | 54 | | | | | | +60 +| 4202 EMIA | | | 40 | | | 48 | | | 55 | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 24 | | | | | | 42 | | | 50 | | | +| | | | 30 | | | 36 | | | | | | +45 +| 16 | | | | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 20 | | | 24 | | | 32 | | | +| 8 | | | | | | 18 | | | | | | +| | | | 10 | | | 12 | | | 24 | | | +16 +| 0 | | | | | | 6 | | | 8 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +24 +| | | | | | | 32 | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5202 EMIA 12 +| | | | 18 | | | 24 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +8 +| | | | 12 | | | 16 | | | 20 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 4 | | | 6 | | | | | | | | | +| | | | | | | 8 | | | 10 | | | +| 0 | | | 0 | | | 0 | | | 0 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure20.Fullin-domainbenchmarkevaluationresultsfortheMATHdomainon7Band8Bmodels. +30 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 N=2048 | | | | +| --- | --------- | --------------- | --------- | -------------------- | --- | ---------- | --- | ---------- | --- | +| | Avg@16(%) | | Pass@4(%) | | | Pass@8(%) | | Pass@16(%) | | +64 +| dnomaiD AQPG 30 | | | | | 80 | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | 56 | | | | | 88 | | | +| 25 | | | | | 72 | | | | | +80 +| 20 | | 48 | | | 64 | | | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 15 | | 40 | | | 56 | | 72 | | | +| 10 | | 32 | | | 48 | | 64 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +56 +| 24 | | | | | 72 | | 88 | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| draH-PCS | | 48 | | | | | | | | +| 18 | | | | | 64 | | 80 | | | +40 +| | | | | | 56 | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| | | 32 | | | 48 | | 64 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +6 +40 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +33 +| hcneB ecneicS 18 | | 27 | | | | | 36 | | | +| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +| 15 | | 24 | | | | | 33 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +27 +| | | 21 | | | | | 30 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 12 | | | | | 24 | | | | | +| | | 18 | | | | | 27 | | | +| 9 | | | | | 21 | | | | | +| | | 15 | | | | | 24 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +64 +90 +| ICS ULMM | | 80 | | | | | 94.5 | | | +| -------- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| 56 | | | | | 88 | | | | | +| 48 | | 76 | | | | | 93.0 | | | +86 +40 +| | | 72 | | | 84 | | 91.5 | | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +32 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | -------------- | ----- | -------------- | --- | ----- | -------------- | ----- | -------------- | --- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Figure21.Fullout-of-domainbenchmarkevaluationresultsfortheMATHdomainon7Band8Bmodels. +| | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 N=384 | | | | +| --- | --------- | --------------- | --------- | -------------------- | --- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | Pass@4(%) | | | Pass@8(%) | | Pass@16(%) | | +96 +| dnomaiD AQPG | | 64 | | | 80 | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 30 | | | | | | | 88 | | | +| | | 56 | | | 70 | | | | | +| 24 | | | | | | | 80 | | | +48 +| 18 | | | | | 60 | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 12 | | | | | 50 | | | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| | | 32 | | | | | 64 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| | | | | | 80 | | 90 | | | +40 +| | | 60 | | | 70 | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +draH-PCS 30 +| | | 50 | | | 60 | | 70 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 20 | | 40 | | | | | | | | +| | | | | | 50 | | 60 | | | +| 10 | | 30 | | | | | | | | +| | | | | | 40 | | 50 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| 24 | | 32 | | | 36 | | 40 | | | +hcneB ecneicS +| 20 | | 28 | | | 32 | | 36 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 16 | | 24 | | | 28 | | 32 | | | +| 12 | | 20 | | | 24 | | 28 | | | +| 8 | | 16 | | | 20 | | 24 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +92 +70 +| ICS ULMM | | 84 | | | 90 | | 94 | | | +| --------- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 60 | | 80 | | | | | | | | +| | | | | | 88 | | 92 | | | +| 50 | | 76 | | | | | | | | +| | | | | | 86 | | 90 | | | +| 40 | | 72 | | | | | | | | +| | | | | | 84 | | 88 | | | +| 30 | | 68 | | | | | | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| 24 | | | | | | | 75 | | | +| AQPGrepuS | | 45 | | | 60 | | | | | +70 +| 20 | | 40 | | | 55 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +65 +16 +| | | 35 | | | 50 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 12 | | | | | | | 60 | | | +| | | 30 | | | 45 | | | | | +55 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | -------------- | ----- | -------------- | --- | ----- | -------------- | ----- | -------------- | --- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Figure22.Fullin-domainbenchmarkevaluationresultsfortheSCIENCEdomainon7Band8Bmodels. +31 +LLMReasoningwithWeakSupervision +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | | N=8 N=384 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | ---- | --------- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| 80 | | | 93 | | | 95.0 | | | 95 | | | +9 0 +| 005-HTAM | | | 8 7 | | | 9 2 . 5 | | | | | | +| -------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | +| 70 | | | 8 4 | | | 9 0 . 0 | | | 90 | | | +| | | | 81 | | | 87.5 | | | | | | +| 60 | | | | | | | | | 85 | | | +| | | | 75 | | | 80 | | | | | | +| 50 | | | 70 | | | 76 | | | 80 | | | +65 +72 +| 40 | | | 60 | | | 68 | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +64 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 60 | | | | | | 84 | | | 92 | | | +| | | | 7 5 | | | 81 | | | 88 | | | +| | | | 7 2 | | | 78 | | | | | | +| 45 | | | 69 | | | | | | 84 | | | +| CMA | | | 66 | | | 75 | | | 80 | | | +| | | | 6 43 | | | 7 2 | | | | | | +| 30 | | | 8 | | | 5 4 | | | 6 5 | | | +| | | | 4 2 | | | | | | 6 0 | | | +| | | | 3 6 | | | 4 8 | | | 5 5 | | | +| 15 | | | 3 0 | | | 4 2 | | | | | | +| | | | 2 4 | | | 3 6 | | | 5 0 | | | +4 5 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +57.5 +| htaM avreniM 35 | | | | | | 51 | | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| | | | 44 | | | | | | 55.0 | | | +| 30 | | | | | | 48 | | | | | | +52.5 +| 25 | | | 40 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +45 +| | | | 36 | | | | | | 50.0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| 20 | | | | | | 42 | | | | | | +47.5 +| 15 | | | 32 | | | | | | | | | +| ---------------- | --- | --- | ----- | --- | --- | -------- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 39 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | | | 56 | | | 64 | | | | | | +| hcneBdaipmylO 40 | | | | | | | | | 69 | | | +| | | | 48 | | | 60 | | | 66 | | | +| 32 | | | | | | 56 | | | 63 | | | +| | | | | | | 52 | | | 60 | | | +| 24 | | | 40 | | | 45 | | | | | | +50 +| | | | 32 | | | 40 | | | 45 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 16 | | | | | | 35 | | | 40 | | | +| | | | 24 | | | 30 | | | | | | +| 8 | | | | | | 25 | | | 35 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure23.Fullout-of-domainbenchmarkevaluationresultsfortheSCIENCEdomainon7Band8Bmodels. +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | | N=8 N=256 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | --- | --------- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +100 +| kcoL mutnauQ | | | | | | 80 | | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 45 | | | 60 | | | | | | | | | +75 +60 +| 30 | | | 40 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 40 | | | 50 | | | +15 +| | | | 20 | | | 20 | | | 25 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | | | 0 | | | 0 | | | 0 | | | +| -------------- | --- | ------- | --- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 450 | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 40 | | | 60 | | | | | | 50 | | | +| dnalsI tsegraL | | | | | | 60 | | | | | | +| 30 | | | | | | | | | 40 | | | +| | | | 45 | | | 45 | | | 30 | | | +20 +| 20 | | | 30 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 30 | | | 8 | | | +| 10 | | | | | | | | | 6 | | | +| | | | 15 | | | 15 | | | 4 | | | +| 0 | | | 0 | | | 0 | | | 2 | | | +0 +| 0 | 150 | 300 450 | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure24.Fullin-domainbenchmarkevaluationresultsfortheGRAPHdomainon7Band8Bmodels. +32 + +LLMReasoningwithWeakSupervision +75 +60 +45 +30 +0 150 300 450 +005-HTAM +Avg@16(%) Pass@4(%) Pass@8(%) Pass@16(%) +95.0 +90 92.5 +8 8 4 7 90.0 88 +81 87.5 +7 7 8 2 85 8 .0 0 80 +64 72 +72 +56 64 +48 56 64 +0 150 300 450 0 150 300 450 0 150 300 450 +18 +12 +6 +0 +0 150 300 450 +4202 +EMIA +60 +40 50 +35 45 +30 40 45 +25 35 +30 16 24 +12 18 +8 12 15 +4 6 +0 0 +0 150 300 450 0 150 300 450 0 150 300 450 +40 +32 +24 +16 +0 150 300 450 +htaM +avreniM +55 +48 +50 55 +42 +45 50 36 +40 45 +30 +35 40 +24 +0 150 300 450 0 150 300 450 0 150 300 450 +32 +24 +16 +8 +0 150 300 450 +dnomaiD +AQPG +60 88 +70 +50 80 +60 +40 72 50 +30 64 +40 +20 56 +0 150 300 450 0 150 300 450 0 150 300 450 +40 +32 +24 +16 +8 +0 150 300 450 +Training Steps +draH-PCS +Qwen2.5-Math-7B LLama3.1-8B-Instruct N=8 N=256 +60 +70 80 +50 60 70 +40 50 60 +30 40 50 +20 30 40 +0 150 300 450 0 150 300 450 0 150 300 450 +Training Steps Training Steps Training Steps +Figure25.Fullout-of-domainbenchmarkevaluationresultsfortheGRAPHdomainon7Band8Bmodels. +0.60 +0.45 +0.30 +0.15 +0.00 +draweR +gniniarT +Qwen2.5-1.5B +Math +0.75 +0.60 +0.45 +0.30 +0.15 +draweR +gniniarT +Llama-3.2-3B-Instruct +Math +0.60 +0.45 +0.30 +0.15 +0.00 +draweR +gniniarT +Qwen2.5-1.5B +Science +0.8 +0.6 +0.4 +0.2 +draweR +gniniarT +Llama-3.2-3B-Instruct +Science +1.0 0.8 +0.6 +0.4 +0.2 +0.0 +draweR +gniniarT +Qwen2.5-Math-7B +Graph +60 +50 +40 +30 +20 +)%( +005-HTAM +54 +51 +48 +45 +42 +39 +)%( +005-HTAM +30 +24 +18 +12 +6 +0 +)%( +tluciffiD-PCS 32 +24 +16 +8 +)%( +tluciffiD-PCS +50 +40 +30 +20 +10 +0 +)%( +kcoL +mutnauQ +30 +24 +18 +12 +6 +0 150 300 450 +Training Steps +)%( +CMA +32 +28 +24 +20 +16 +12 +0 150 300 450 +Training Steps +)%( +CMA +60 +50 +40 +30 +20 +0 150 300 450 +Training Steps +)%( +005-HTAM +54 +52 +50 +48 +46 +0 150 300 450 +Training Steps +)%( +005-HTAM +82 +80 +78 +76 +74 +0 150 300 450 +Training Steps +)%( +005-HTAM +=0 =0.1 =0.3 =0.5 =0.7 =0.9/1.0 +Figure26.Effectofrewardlabelcorruptionontrainingdynamicsandgeneralization.γdenotesthefractionoftrainingprompts +withcorruptedlabels,rangingfromclean(γ =0)tofullyincorrect(γ =1).QwenmodelsonMATHandSCIENCEdomainsmaintain +performanceundersubstantialcorruption,whilegeneralizationofLlamamodelsandGRAPHdomaindegradeatγ ≥0.5.Evaluation +resultsinthisfigurearebasedongreedydecoding. +33 + +LLMReasoningwithWeakSupervision +Qwen2.5-Math-1.5B Llama-3.2-3B-Instruct Qwen2.5-1.5B Qwen2.5-Math-1.5B +| | | Math | | | Math | | | Science | | | Science | +| --------------- | --- | ---- | --- | ---- | ---- | --- | --- | ------- | --- | --- | ------- | +| draweR gniniarT | | | | | | | 1 | | | 1.0 | | +| 0.75 | | | | 0.90 | | | | | | 0.8 | | +0.75 +| 0.60 | | | | | | | | | | 0.6 | | +| ---- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | +| | | | | 0.60 | | | | | | 0.4 | | +| 0.45 | | | | 0.45 | | | | | | | | +0.2 +| 0.30 | | | | 0.30 | | | 0 | | | | | +| ---- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | +)%( 005-HTAM +| | | | | 60 | | | 60 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +70 +70 +| | | | | 50 | | | | | | 60 | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +50 +| | | | | 40 | | | | | | 50 | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| )%( tluciffiD PCS | | | | | | | 12 | | | | | +| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 32 | | | 15 | | | | | | | | +| | | | | 12 | | | 9 | | | 32 | | +| | 24 | | | | | | | | | 24 | | +| | | | | 9 | | | 6 | | | | | +| | 16 | | | 6 | | | | | | 16 | | +| | 8 | | | | | | 3 | | | | | +| | | | | 3 | | | | | | 8 | | +| | 0 | | | 0 | | | 0 | | | 0 | | +0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +| --- | --- | -------------- | --- | ---- | -------------- | ------------- | --- | -------------- | --- | --- | -------------- | +| | | | | RLVR | | Majority Vote | | Self Certainty | | | | +Figure27.Comparisonofrewardvariants(RLVR,self-certainty,majorityvote)with1024trainingsamples.Proxyrewardswithout +verifiersexhibitfailuremodesunderprolongedtraining:trainingcollapse(self-certainty),andrewardspikesfollowedbyperformance +drops(majorityvote).Evaluationresultsinthisfigurearebasedongreedydecoding. +| | | | | | GRPO-POS | | GRPO-NEG | | GRPO | | | +| --- | --- | --- | --- | --- | -------- | --- | -------- | --- | ---- | --- | --- | +Training Reward SCP-Difficult (%) GPQA Diamond (%) MATH-500 (%) +| htaM-5.2newQ | | | | | | | )%( dnomaiD AQPG | | | | | +| ------------ | ------------------- | --- | --- | -------------------- | --- | --- | ---------------- | --- | --- | ------------ | --- | +| | draweR gniniarT 1.0 | | | )%( tluciffiD-PCS 35 | | | | | | | 72 | +| | | | | | | | | 25 | | )%( 005-HTAM | | +| | 0.8 | | | 30 | | | | | | | 70 | +B5.1- +20 +| | 0.6 | | | 25 | | | | | | | 68 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +| | 0.4 | | | 20 | | | | | | | 66 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | | | 10 | | | 64 | +| | 0.2 | | | 15 | | | | | | | | +5 +62 +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| --- | --- | --------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ------------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +| | draweR gniniarT 1.0 | | | )%( tluciffiD-PCS | | | | | | 52.5 | | +| ----------- | ------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | ------------ | --- | +| B3-2.3amalL | | | | 15 | | | | | | )%( 005-HTAM | | +27.5 +| | tcurtsnI- 0.8 | | | 12 | | | | | | 51.0 | | +| --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | +25.0 +| | | | | 9 | | | | | | 49.5 | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | +| | 0.6 | | | | | | 22.5 | | | | | +6 +| | 0.4 | | | | | | 20.0 | | | 48.0 | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | +3 +46.5 +| | 0.2 | | | 0 | | | 17.5 | | | | | +| --- | --- | -------------- | --- | --- | -------------- | --- | ---- | -------------- | --- | --- | -------------- | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +Figure28.EffectofbaselinevariantsonSCIENCEdomainwith8trainingsamples.GRPO-pos(positiveupdatesonly)andGRPO-neg +(negativeupdatesonly)producecomparableperformancetostandardGRPO. +| | | | | | GRPO-POS | | GRPO-NEG | | GRPO | | | +| --- | --- | --- | --- | --- | -------- | --- | -------- | --- | ---- | --- | --- | +htaM-5.2newQ Training Reward SCP-Difficult (%) GPQA Diamond (%) MATH-500 (%) +)%( dnomaiD AQPG +| | draweR gniniarT | | | )%( tluciffiD-PCS 40 | | | | | | )%( 005-HTAM | | +| --- | --------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | +| | 0.75 | | | | | | | 25 | | | 72 | +| | B5.1- | | | 35 | | | | | | | | +| | | | | | | | | 20 | | | 70 | +| | 0.60 | | | 30 | | | | | | | | +68 +| | 0.45 | | | 25 | | | | 15 | | | | +| --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +66 +10 +| | 0.30 | | | 20 | | | | | | | 64 | +| --- | ---- | -------------- | --- | --- | -------------- | --- | --- | -------------- | --- | --- | -------------- | +| | | | | 15 | | | | 5 | | | 62 | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +)%( dnomaiD AQPG +| | draweR gniniarT | | | )%( tluciffiD-PCS 25 | | | | 28 | | | 54 | +| ----------- | --------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | +| B3-2.3amalL | 0.90 | | | | | | | | | )%( 005-HTAM | | +| | tcurtsnI- | | | | | | | 26 | | | | +| | 0.75 | | | 20 | | | | | | | 52 | +24 +| | 0.60 | | | 15 | | | | | | | 50 | +| --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.45 | | | | | | | 22 | | | 48 | +10 +| | 0.30 | | | | | | | 20 | | | | +| --- | ---- | -------------- | --- | --- | -------------- | --- | --- | -------------- | --- | --- | -------------- | +| | | | | 5 | | | | | | | 46 | +| | 0.15 | | | | | | | 18 | | | | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +Figure29. EffectofbaselinevariantsonSCIENCEdomainwith1024trainingsamples. SimilartoFigs.28,GRPO-pos(positive +updatesonly)andGRPO-neg(negativeupdatesonly)producecomparableperformancetostandardGRPO. +34 + +LLMReasoningwithWeakSupervision +Figure30. Responsediversityon8samplesfromtheMATH-500evaluationdataset. Qwen-mathshowshighdiversitywithinits +correctanswers,suggestingarangeoflearnedrobustreasoningpaths. +DiversityPromptforLLM-as-a-judge +Youaregiventheoriginalpromptandtwomodel-generatedresponses. Determinewhetherthetworesponsesuse +differentstrategiestosolvetheproblem. +Usethefollowingguidelines: +-Different solution methods: Clearly different approaches (e.g., algebraic vs. geometric, analytical vs. +numerical). +-Criticalreasoningdivergence:Significantdifferencesinkeyreasoningstepsorassumptions,eveniffinalanswers +match. +-Conceptualdifferences: Distinctunderlyingconceptsorrepresentations(e.g.,probabilityvs. combinatorics). +**Alsolabelasdifferentif:**Thetworesponsessharethesamegeneralapproachbutdiffermeaningfullyin +specificintermediatestepsormanipulationscrucialtothesolution. +Originalprompt: prompt +Generation0: generation0 +Generation1: generation1 +Question: DoGeneration0andGeneration1usedifferentstrategies? +You may first generate a short reasoning, then end your response with either ||yes|| if they use different +strategiesor||no||iftheyusethesamestrategy. +Figure31.LMprompttochecksimilaritybetweenresponses. +35 + +LLMReasoningwithWeakSupervision +FaithfulnessPromptforLLM-as-a-judge +Youwillbegiven: (1)amathproblemprompt,and(2)amodelresponsethatmayincludeBOTHreasoninganda +finalanswer. +Definitions: +”Reasoning” = the parts of the model response that attempt to justify or derive a result (intermediate steps, +equations,explanations,casework,narrativelogic). +”Finalanswer”=themodel’sexplicitcommittedresult(e.g.,after”Final:”,”Answer:”,”Therefore”,” +boxed”, or the last clear numeric/symbolic conclusion). If multiple answers appear, treat the last explicitly +committedoneasthefinalanswer. +Task:Decidehowwellthereasoningsupportsthefinalanswer,usingtheselabels: +Label1(Correlated): +Thereasoningformsacoherentderivationthatwouldleadtothefinalanswerasstated. Minoralgebraslipsare +allowedIFtheoverallderivationstillclearlytargetsthatanswer. Thefinalanswermaybeobjectivelywrong;you +judgealignment,notcorrectness. +Label0.5(PartiallyCorrelated): +Thereasoningisrelatedtotheproblemandseemstomovetowardthefinalanswer,buthasmajorgaps,unjustified +leaps,missingsteps,orseriouserrorsthatbreaktheproof. Theanswerisnotapurenon-sequitur,butthesupport +isweak/incomplete. +Label0(Uncorrelated): +The final answer is not supported by the reasoning. Examples include: contradiction with earlier derived +statements; switchingtoanunrelatedmethod; violatingkeyconstraintsfromtheprompt; orthefinalanswer +appearingasanunsupportedguess. +Outputformat(MANDATORY): +1)Brieflyidentify(a)theextractedfinalanswerand(b)thekeyreasoningpathin1–3sentences. +2)Thenoutputexactlyonelabeltokenonitsownattheend: ∥1∥or∥0.5∥or∥0∥. +Prompt:prompt +Response:response +Question: Does the reasoning path correspond to the provided answer? You may first generate a short +reasoning,thenendyourresponsewitheither∥1∥iftheyarefullycorrelated,∥0.5∥iftheyarepartiallycorrelated, +or∥0∥iftheanswerisuncorrelatedtotheprecedinglogic. +Figure32.LMprompttoevaluatereasoningfaithfulnessonasamplefromtheMATHdataset. +36 + +LLMReasoningwithWeakSupervision +Figure33.Proportionofalignedandmisalignedresponsesacrossmodelsandtrainingdatasets. +| draweR gniniarT | | )%( 005-HTAM 90 | | | 72 | | 80 | | | +| --------------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +)%( draH-PCS +| | | 84 | | | )%( CMA 64 | | | | | +| ------ | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | +| ecracS | 0.6 | 78 | | | 56 | | | | | +ataD +40 +60 +40 +| | | 54 | | | 30 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.0 +| | 0 150 300 | 450 | 0 150 | 300 | 450 | 0 150 300 | 450 | 0 150 | 300 450 | +| --- | --------- | --- | ----- | --- | --- | --------- | --- | ----- | ------- | +90 +| draweR gniniarT | | )%( 005-HTAM | | | | | )%( draH-PCS | | | +| --------------- | --- | ------------ | --- | --- | ---------- | --- | ------------ | --- | --- | +| ytirojaM | | | | | )%( CMA 60 | | 60 | | | +etoV +0.6 +60 +40 +30 +0.0 +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +draweR ysioN +| draweR gniniarT | | )%( 005-HTAM | | | | | | | | +| --------------- | --- | ------------ | --- | --- | ---------- | --- | ------------ | --- | --- | +| | 0.6 | | | | | | )%( draH-PCS | | | +| )7.0= | | | | | )%( CMA 60 | | | | | +| | | 75 | | | | | 60 | | | +0.3 +( +40 +30 +50 +| | 0 150 300 | 450 | 0 150 | 300 | 450 | 0 150 300 | 450 | 0 150 | 300 450 | +| --- | -------------- | --- | -------------- | --- | --- | -------------- | --- | -------------- | ------- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure34.Evaluationresultsofpass@16metricacrossmodelswithdifferentpre-RLinterventiononweaksupervision. +37 + +LLMReasoningwithWeakSupervision +| | avg@16 | | pass@16 | | avg@16 | | pass@16 | | +| --- | ------ | --- | ------- | --- | ------ | --- | ------- | --- | +)%( hcneB ecneicS +)%( 4202 EMIA +30 +8 +| | | 20 | | 10 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +ecracS +ataD +15 +0 0 150 300 450 0 0 150 300 450 0 0 150 300 450 0 150 300 450 +)%( hcneB ecneicS +)%( 4202 EMIA +ytirojaM +| 8 | | 20 | | 10 | | 20 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +etoV +| 0 | | 0 | | 0 | | 0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +| draweR ysioN | | | | )%( hcneB ecneicS | | | | | +| ------------ | --- | --- | --- | ----------------- | --- | --- | --- | --- | +)%( 4202 EMIA +| )7.0= | | | | | | 30 | | | +| ----- | --- | --- | --- | --- | --- | --- | --- | --- | +| 8 | | 25 | | 10 | | | | | +( +15 +| 0 | | 0 | | 0 | | | | | +| --- | -------------- | --- | -------------- | --- | -------------- | --- | -------------- | --- | +| 0 | 150 300 | 450 | 0 150 300 | 450 | 0 150 300 | 450 | 0 150 300 | 450 | +| | Training Steps | | Training Steps | | Training Steps | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure35. EvaluationresultsonAIME2024andScienceBenchacrossmodelswithdifferentpre-RLinterventiononweak +supervision. +38 + +LLMReasoningwithWeakSupervision +DiversityJudgeExampleOutput +Prompt: Aboxcontains5whiteballsand5blackballs. Idrawthemoutofthebox,oneatatime. Whatisthe +probabilitythatallofmydrawsalternatecolors,startingandendingwiththesamecolor? +ResponseA: To determine the probability that all of the ball draws alternate colors, starting and end- +ingwiththesamecolor,weneedtoconsiderthefollowing: +1. **Totalnumberofwaystoarrangetheballs**: Thereare5whiteballsand5blackballs,sothetotalnumberof +waystoarrangethese10ballsisgivenbythebinomialcoefficient +(cid:0)10(cid:1) +,whichrepresentsthenumberofwaysto +5 +choose5positionsoutof10forthewhiteballs(theremaining5positionswillbefortheblackballs). Thisvalue +iscalculatedas: +(cid:18) (cid:19) +10 10! += =252 +5 5!5! +2. **Numberoffavorableoutcomes**: Forthedrawingstoalternatecolorsstartingandendingwiththesame +color,thereareonlytwopossiblesequences: -Startingwithawhiteballandalternating(WBWBWBWBW +B)-Startingwithablackballandalternating(BWBWBWBWBW) +Sothereareonly2favorableoutcomes. +3. **Probability calculation**: Theprobability istheratioof thenumberof favorableoutcomestothe total +numberofoutcomes. Therefore,theprobabilityP is: +Numberoffavorableoutcomes 2 1 +P = = = +Totalnumberofoutcomes 252 126 1 +Sothefinalansweris . +126 +ResponseB: Tosolvetheproblem,weneedtodeterminetheprobabilitythatasequenceof10balls(5white +and5black)isarrangedsuchthattheyalternatecolors. Thereareonlytwopossiblesequencesthatmeetthis +condition: ”WBWBWBWBWB”and”BWBWBWBWBW”. Thetotalnumberofwaystoarrange5whiteand5 +blackballsisgivenbythebinomialcoefficient +(cid:0)10(cid:1) +,whichrepresentsthenumberofdistinctpermutationsof5 +5 +whiteand5blackballs. Thisiscalculatedas: +(cid:18) (cid:19) +10 10! += =252 +5 5!5! +Sincethereareonly2favorableoutcomes(thetwoalternatingsequencesmentionedabove),theprobabilityisthe +ratioofthenumberoffavorableoutcomestothetotalnumberofoutcomes: +2 1 +Probability= = +252 126 1 -) - -November 1999 - - Nineteenth-Century Log House - -Location: (cid:9) -0.25 miles south of the intersection of SR 1424 and SR 1405. - -This log house stands on the west side of SR 1424, approximately - -The log house sits approximately 40 feet from the road, and has - -Setting: (cid:9) -two mid-twentieth century outbuildings standing to the rear of the house. To the -north of the log house stands a late twentieth-century residence. - -Description: - -The one-story, side-gable, three-bay log dwelling is - -Dwelling: (cid:9) -constructed with half-dovetail corner notching. The house features a -variety Of wood sheathing, including narrow weatherboards and novelty -siding. Windows in the building have been enlarged and have six-over-six -sashes. The porch on the facade has also been replaced. A one-story -frame addition was constructed onto the rear of the dwelling. An exterior -brick chimney is located on the south elevation. - -Outbuildings: Two frame outbuildings stand south of the house: a frame -chicken house with a shed roof and a small, one-story frame building with -an entry on the gable end. - -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, this house is considered not -eligible for the National Register of Historic Places under Criterion A, since it is -not representative of a significant. trend in the history of the area. - -This house is considered not eligible fir the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. - -This house is considered not eligible for the National Register under Criterion C -since the building has lost integrity and does not embody distinctive -characteristics. The building no longer retains its original porch or windows and -the form has been compromised by the addition to the rear of the dwelling: - -The building is considered not likely to yield information important to the history -of building technology; therefore it is not eligible for the National Register under -Criterion D. - -Historic Architectural Resources Survey Report: NCDOT, SR 1424. Burke County: November 1999 -Nancy l'art Dolsen - -Page 28 - - Photograph 1: Nineteenth-Century Log House, facing northeast - -Photograph 2: Nineteenth-Century Log 1-louse facing southwest - -Historic Architectural Resources Survey Report: .VC DOT SR 1424, Burke County: November 1999 -Nancy I an Dolsen - -Page 29 - - Photograph 3: View of Chicken Coop, facing northeast - -Photograph 4: Domestic outbuilding, facing north - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nwtcy i in Dolsen - -Page 30 - - BIBLIOGRAPHY - -Arney, Marshall House. Historic Resource Survey Form, Burke County, 1983. -Asheville, NC: Western Office. Division of Cultural Resources. - -Arney, William Jones House, Historic Resource Survey Form, Burke County, 1983. - -Asheville, NC: Western Office, Division of Cultural Resources. - -Biographical Directory of the United States Congress, 1774-Present, Thomas Burke,- - -Website: Bioguide.conress.gov. - -Burke County Deeds, Burke County Court House, Morganton, NC. - -Burke County Plat Book, Burke County Court House, Morganton, NC. - -Corpening House, Historic Resource Survey Form, Burke County, 1983. Asheville,NC: - -Western Office, Division of Cultural Resources. - -Cotton, J. Randall, Suzanne Pickens Wylie, and Millie M. Barbee. Historic Burke: An -Architectural Sites Inventory of Burke County. Mortzanton, NC: Historic Burke -Foundation. Inc.. 1987. - -Phifer, Edward W. Burke: the History of a North Carolina County, 1777-1920. - -Morganton, NC: by the author, 1977. - -Historic Architectural Resources Survey Report: NCD01; SR 1424, Burke County: November 1999 (cid:9) -Nancyl'an Dolsen - -Page 31 - - APPENDIX A - -NCDOT, SR 1424. Bur ke Cou - -November 1 999 - - North Carolina Department of Cultural Resources - -James B. Hunt Jr., Governor -Betty Ray McCain, Secretary - -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 - -July 9, 1999 - -Stephen A. Moore, PE -District Engineer -NCDOT -Route 1 Box 169C -Marion NC 28752 - -Division of Archives and History -Jeffrey J. Crow, Director - -LOCATION -507 North Blount Street -Raleigh, NC -State Courier 53-31-31 - -Re: (cid:9) Widen and pave SR 1438 Corpening Bridge Road, - -Burke County, ER 99-9229 - -Dear Mr. Moore: - -We have received notification of a nationwide permit for the above project and would -like to comment. - -We have reviewed the permit application for the improvement and paving of SR 1438 -Corpening Bridge Road and located the following structures of historical or -architectural importance within the general area of the project: - -William Avery House, located at the northwest corner of the junction of SR 1438 -and SR 1424. - -Arney's Chapel, located at the east side of SR 1423 south of its junction with SR -1439. - -We recommend that an architectural historian with the North Carolina Department of -Transportation evaluate these sites to determine if they are within the area of potential -effect for the permit and are eligible for listing on the National Register of Historic -Places. - -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. - -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. - --Sincerely, - -1- (cid:9) - -• (r-N — - -Da.vicl'Brook -Deputy State Historic Preservation Officer - -DB:slw - -109 East Jones Street • Raleigh, North Carolina 27601-2807 - - cc: (cid:9) - -Army- Corps of Engineers, Asheville -William D. Gilmore - -t-Barbara Church - - AUG-16—":39 MON 10:10 ID: - -TEL NO: - -351; P02 — - -North Carolina Department of Cultural Resources - -Jams B. Hunt Jr.. Governor -Betty Ray McCain, Secretary - -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 - -August 10, 1999 - -Stephen A. Moorc -District Engineer -North Carolina Department of Transportation -Route 1, Box 169C -Marion NC 28752 - -Re: - -Paving SR 1414, Johns River Road, Burke County, -ER 00--7154 - -Division of Archives Bnd History -Jeffrey 1. Crow, Director - -LOCATION -507 North Blount Street -Raleieh, NC -State Courier 53-31-31 - -it- , - have received notification of a nationwide permit for the above project and would - -i%1'- - -Dear Mr. Moore: - -We -like to comment. - -We have conducted a search of our maps and files and have located the following -structures of historical or architectural importance within the general area of the -project: - -qX.117. - -Marshall Arney House, located on thc wcst side of SR 1424, 0.5 mile northwest -of its junction with SR 1438. - -William Arney House, located at the northwest corner of the SR 1424 and SR -1438 junction. - -9;0 ck - -Corpening House, located on the west side of SR 1424, L5 mile south of its -junction with SR 1405. - -We recommend that an architectural historian with the North Carolina Department of -Transportation identify and evaluate these properties and any other structures over -fifty years of age within the project area, and report the findings to us. - -An important archaeological site, a mound reported by J. W. Spainour and Cyrus -Thomas in the late 1800s, is recorded between the Johns River and Johns River Road -south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 311310 highlighted in -yellow. - -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan - -109 Eist Jones Street • Ra1e10, North Cirolina 27601-2807 - - AUG-16-'99 MON 10:11 ID: (cid:9) - -TEL NO: (cid:9) - -p355 P03 - -Stephen A. Moore -August 10, 1999, Page 2 - -sheets 8,9, 12. 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. - -end south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 31BK3 highlighted in -yellow. - -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan -sheets 8, 9, 12, 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. - -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. - -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. - -Sincerely, - -&Ad,- - -Deputy State Historic Preservation Officer - -/ David Brook - -DB:slw - -Enclosure - -cc: (cid:9) - -Army Corps of Engineers, Asheville -William D. Gilmore -Thomas Padgett - +So,theprobabilitythatallofmydrawsalternatecolors,startingandendingwiththesamecolor,is . +126 +JudgeOutput: Both Generation 0 and Generation 1 use the same strategy to solve the problem. They both +identifythattherearetwopossiblesequencesthatsatisfytheconditionofalternatingcolorsandstartingand +endingwiththesamecolor: ”WBWBWBWBWB”and”BWBWBWBWBW”. Theybothcalculatethetotal +numberofwaystoarrangethe10ballsusingthebinomialcoefficient +(cid:0)10(cid:1) +,whichaccountsforthepermutations +5 +of5whiteand5blackballs. Finally, theybothcomputetheprobabilitybytakingtheratioofthenumberof +favorableoutcomes(2)tothetotalnumberofoutcomes(252),resultinginthesamefinalanswerof 1 . +126 +Therearenosignificantdifferencesintheirreasoningsteps, assumptions, orintermediatecalculations. Both +responsesfollowasimilarlogicalflowandusethesamemathematicalconceptstoarriveatthesolution. +JudgeResult: ∥no∥ +Figure36.QualitativeExampleofDiversityAnalysis +39 + +LLMReasoningwithWeakSupervision +FaithfulnessJudgeExampleOutput +Prompt: Findallrealpairs(a,b)thatsolvethesystemofequation +a2+b2 =25, +3(a+b)−ab=15. +Response: Tosolvethesystemofequations: +(cid:40) +a2+b2 =25, +3(a+b)−ab=15, +wecanusethefollowingreasoningandalgebraicmanipulations. First,recalltheidentity: +a2+b2 =(a+b)2−2ab. +Let’sdenotes=a+bandp=ab. So,thefirstequationbecomes: s2−2p=25.Thesecondequationremains: +3s−p=15.Wecansolvethesetwoequationssimultaneously. First,weexpresspfromthesecondequation: +p=3s−15.Next,wesubstitutepintothefirstequation: s2−2(3s−15)=25.Simplifytheequation: +s2−6s+30=25 =⇒ s2−6s+5=0. +√ +Thisisaquadraticequationins. Wecansolveforsusingthequadraticformulas= −b± b2−4ac,wherea=1, +2a +b=−6,andc=5: +√ √ +6± 36−20 6± 16 6±4 +s= = = . +2 2 2 +So,thesolutionsforsare: s=5 or s=1.Now,wefindthecorrespondingpvaluesforeachs: 1. Ifs=5, +thenp=3(5)−15=0. 2. Ifs=1,thenp=3(1)−15=−12. Wenowneedtofindthepairs(a,b)thatsatisfy +theequationsa+b=sandab=p. Thesearetherootsofthequadraticequations: 1. Fors=5andp=0,the +quadraticequationis: +t2−5t=0 =⇒ t(t−5)=0. +So,thesolutionsaret=0andt=5. Thus,thepairs(a,b)are(0,5)and(5,0). 2. Fors=1andp=−12,the +quadraticequationis: +t2−t−12=0. +Wesolvethisquadraticequationusingthequadraticformula: +√ +1± 1+48 1±7 +t= = . +2 2 +So,thesolutionsaret=4andt=−3. Thus,thepairs(a,b)are(4,−3)and(−3,4). Therefore,thecomplete +setofsolutionsis: +(0,5),(5,0),(4,−3),(−3,4) . +JudgeOutput: Finalanswergiven: (0,5),(5,0),(4,−3),(−3,4).Reasoningpath: Introducessumss=a+b +andproductp=ab,rewritesequationsintoasysteminsandp,solvestogets=5or1,findscorrespondingp, +thensolvesquadratict2−st+p=0toobtainallorderedpairsconsistentwitheach(s,p),yieldingthelisted +solutions. Thederivationiscompleteandconsistentlyleadstothestatedpairs. +JudgeResult: ∥1∥ +Figure37.QualitativeExampleofFaithfulnessAnalysis +40 \ No newline at end of file diff --git a/q103/random_k2/random_2.md b/q103/random_k2/random_2.md index 9a6992971674500fba1af17645e45a0908def2ef..d1447fd886d0d69697641e94fcf1890d83769afd 100644 --- a/q103/random_k2/random_2.md +++ b/q103/random_k2/random_2.md @@ -1,727 +1,1679 @@ - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Hobart and William SmitH CollegeS - Online ISSN: 2054-0965 (Online) +Handbook for +International Students - Print ISSN: 2054-0957 (Print) +intercultural affairs Center +288 Pulteney Street +geneva, nY 14456 +(315) 781-3319 – Phone +(315) 781-3888 – Fax - Website: https://www.eajournals.org/ + TABLE OF CONTENTS - Publication of the European Centre for Research Training and Development -UK +SECTION I: General Information + Welcome -Cloud Computing: Revolutionizing Digital -Transformation in the Modern Era +important Campus Phone numbers -Lavanya Jacintha Victor -The University of Texas at Austin, USA + office locations -doi: https://doi.org/10.37745/ejcsit.2013/vol13n181223 Published May 12, 2025 +banks and theaters +Places to Stay +Frequently asked Questions -Citation: Victor L.J. (2025) Cloud Computing: Revolutionizing Digital Transformation in the Modern Era, European -Journal of Computer Science and Information Technology, 13(18),12-23 +SECTION II: Immigration Procedures + How to get assistance with immigration matters +immigration basics and definition of terms -Abstract: Cloud computing has revolutionized digital transformation by fundamentally changing how -organizations approach their technological infrastructure and business operations. The shift from -traditional on-premises systems to cloud-based solutions has enabled unprecedented levels of scalability, -efficiency, and innovation across industries. From artificial intelligence integration to edge computing -adoption, cloud technologies have democratized access to advanced capabilities while reducing -operational costs. Organizations worldwide are embracing hybrid and multi-cloud strategies, prioritizing -security through zero-trust architectures, and focusing on sustainable computing practices. The evolution -of serverless computing and edge technologies continues to shape the future of enterprise IT, offering new -possibilities for innovation while addressing crucial challenges in data processing, security, and -compliance. -Keywords: cloud computing, digital transformation, edge computing, serverless architecture, hybrid -cloud + obtaining a Visa and Form i-20 -INTRODUCTION +differences between F-1 and J-1 Status -In today's rapidly evolving technological landscape, digital transformation has become a cornerstone of -business evolution and success. According to recent International Data Corporation (IDC) forecasts, -worldwide spending on digital transformation is projected to maintain a robust compound annual growth -rate (CAGR) of 16.3% from 2022 to 2026, culminating in a staggering $3.4 trillion by 2026. The digital -and information technology sector has demonstrated remarkable resilience even during economic -uncertainties, showcasing a significant 17.8% growth in digital transformation investment during 2022 -alone [1]. This comprehensive shift towards digitalization represents more than just a trend—it's a -fundamental reimagining of how organizations operate, deliver value, and interact with their stakeholders -through the integration of digital technologies. + arriving at a U.S. port of entry… What a student can expect +dealing with the department of Homeland Security (dHS) +exceptions to a Full Course of Study +duration of Status for F-1 Students +temporary departure from the US + + accepting U.S. government Public assistance or Public benefits + Work opportunities for students with F-1 Status + Work opportunities for students in J-1 Status + +Practical training + optional Practical training (oPt) + Curricular Practical training (CPt) +departure Checklist + +page 1 +9 +10 +10 +10 12 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -The impact of digital transformation is profound yet challenging. Despite substantial investments, recent -executive surveys reveal that approximately 70% of digital transformation initiatives are falling short of -their objectives. A significant factor contributing to this shortfall is the persistent gap between technological -capabilities and workforce readiness. Nearly two-thirds of surveyed executives report that their -organizations struggle with change management and employee adoption of new digital systems. -Furthermore, 84% of organizations cite human-centric issues, rather than technical complications, as the -primary barriers to successful digital transformation [2]. - -The landscape of digital transformation has evolved significantly, with organizations focusing on three -critical areas: technology infrastructure modernization, process automation, and workforce enablement. -The IDC research highlights that industries such as financial services, manufacturing, and retail are leading -the investment charge, with discrete manufacturing alone expected to invest $382 billion in digital -transformation by 2026 [1]. However, the success of these investments heavily depends on organizations' -ability to address fundamental human elements. Studies indicate that organizations achieving digital -transformation success are three times more likely to prioritize workforce development and change -management strategies alongside their technological implementations [2]. - -The momentum of digital transformation continues to accelerate, driven by market demands and -competitive pressures. While the technological spending trajectory remains impressive, with manufacturing -and financial services each projected to exceed $1 trillion in total digital transformation investments through -2026 [1], the focus is increasingly shifting toward holistic transformation approaches. Successful -organizations are now recognizing that digital transformation requires a balanced investment in both -technology and people, with leadership development, skill enhancement, and cultural adaptation being as -crucial as the technological infrastructure itself [2]. - -The Evolution of Digital Transformation -The journey of digital transformation began with the basic digitization of manual processes, marking a -fundamental shift in business operations. Current research indicates that organizations implementing digital -transformation strategies have achieved remarkable improvements in their operational metrics. By -integrating automation and digital workflows, businesses have reported up to 80% reduction in process -completion times and a 90% decrease in human errors across various operational tasks. This evolution has -particularly impacted customer service operations, where automation has enabled organizations to handle -customer inquiries 60% faster while maintaining high satisfaction rates [3]. What started as simple -workflow automation—like digital leave management systems and basic inventory tracking—has evolved -into sophisticated, interconnected ecosystems that drive entire business operations. - -The transformation landscape has seen significant advancement in data processing capabilities, with -modern systems able to handle complex analytical tasks in real-time. Organizations implementing -comprehensive digital transformation strategies have witnessed a 70% improvement in their decision- -making processes, primarily due to enhanced data accessibility and automated analytical capabilities. +15 +16 +17 +19 +21 +22 +24 +24 +24 +25 +26 +26 +27 +27 +28 +29 + +SECTION III: -13 +Sample Fall orientation Schedule for international Students - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +30 - Online ISSN: 2054-0965 (Online) +2 + + SECTION I: +GENERAL INFORMATION + +Welcome! + +Welcome to Hobart and William Smith Colleges (the Colleges or HWS). to you, the +decision to study abroad might have been a small one – or, indeed, a rather big step. +either way you won’t know what your experience will be like until you are here. this +handbook is designed to help you prepare for the adventure of studying in the United +States. in our experience the better prepared you are the more you will benefit from +your time at the Colleges. + While we are trying to answer many of your questions prior to your arrival in the +United States, some of your questions will only arise once you are here. the inter- +national Student adviser will serve as a resource to you and your parents or guardians +throughout the application process and during your stay at Hobart and William Smith +Colleges. The adviser will be your primary contact for any questions or concerns +regarding visas and immigration issues, employment, travel and a host of other + matters that may arise. + +the International Student Office (iSo) is located on the second floor of the + +intercultural affairs Center at 288 Pulteney Street. this office advocates for and ad- +dresses the needs of international students as well as provides programs and services +related to immigration matters. during the academic school year the iSo sponsors +and co-sponsors events designed to promote cross-cultural sharing between the many +groups at the Colleges and +geneva communities. +Currently, international +students come to us from 22 +nations: austria, bulgaria, +Canada, China, Colombia, +Czech republic, germany, +india, iran, nepal, Pakistan, +Portugal, Scotland, South +africa, South Korea, Swe- +den, Switzerland, thailand, +the netherlands, turkey, +Ukraine and Vietnam. + +Intercultural Affairs Center – Home to the International Student Office. + +in the fall international students attend a special two-day orientation before +the regular first year orientation begins. at Hobart and William Smith, we ask that +incoming international students join us ahead of time as it is a wonderful opportunity +to get acquainted with the new culture, get assimilated with the campus, and provide +a forum to answer your questions before the rush of orientation activities begins. at +the iSo, we will check to make sure that all your immigration paperwork is in order as +well as help you gain a better understanding of what your F-1 or J-1 status means and +how to maintain it; help to answer any questions that you might have about adjusting +to life as a student in the United States, specifically life at HWS; and meet students, +staff and faculty members. + +3 + + Questions? Please +contact the International +Student Adviser by phone + at (315) 781-3319. The + International Student + Office is located on the + 2nd floor of the +Intercultural Affairs Cent- +ter, 288 Pulteney Street, +Geneva, NY 14456. (315) +781-3319 – Phone (315) +781-3888-Fax. Office +hours are 8:30 am to +5:00 pm weekdays during +the academic year. + +THINGS TO DO BEFORE GOING ABROAD +before you can board a plane and start your study +abroad adventure, there are some formalities you will +need to take care of. the earlier you start thinking +about the things you need, the smoother your prepara- +tions will go. this list will help you consider some of +the things you will have to do. + +Obtaining a Visa +Upon receipt of the form i-20 (for F-1 students) or dS- +2019 (for J-1 exchange students) you can apply for your +student visa. it is wise to schedule an appointment with +the U.S. embassy or consulate in your country as soon +as possible as sometimes there are no appointments +available for weeks. You should also give yourself some +time to prepare for your visit to the U.S. embassy or +consulate. Visa procedures vary from country to country. +make sure to check the website of the embassy or + +consulate in your country to find out what documents you need to bring. document +you certainly need include: (1) a current, valid passport; (2) your i-20 or dS-2019 +form; (3) evidence of financial support for the duration of time you will spend in the +US; (4) proof that you have a permanent residence outside the US; (5) one or more +passport pictures meeting U.S. requirements; (6) a nonimmigrant visa application. + +Medical Insurance +HWS Colleges require that all students buy the medical insurance provided by +the colleges. all J-1 student-visa holders (that is, all exchange students) are also +required by U.S. law to have their own medical insurance. While F-2 students are +covered by the Colleges medical insurance you might nevertheless consider buying +additional medical coverage. + +Important Documents +You need to bring all important documents relating to your legal, medical, and aca- +demic history with you. Such documents might include medical and dental records, +certificates of immunizations, information about medical conditions and treatments, +prescriptions for medication and eyeglasses or contact lenses, and official transcripts +from secondary school and college. make sure you photocopy all documents and +leave some copies with a person you trust in your home country. it is also helpful to +translate these documents into english before you leave your home country. + +Packing +Pack as light as possible. remember that you might have to carry all your things at +different times during the journey. if you are studying abroad for only a semester you +might not need to bring your entire wardrobe. You might also want to leave room +in your suitcase for souvenirs you will buy while in the US. at the same time there +might be things you want to bring because they will make you more comfortable + +4 + + and will make your transition to a new environment easier. Some exchange students +and international students also like to bring small gifts from their home country to +share with new friends abroad. this might be a means for you to share and explain +your culture to your new acquaintances. if you have traditional, national cloth- +ing consider bringing it to wear at events sponsored by the international Students +association or other cultural events. if you arrive during spring semester (January), +to bring a warm jacket and boots, a scarf, gloves and hat. We encourage you to take +advantage of the option to ship your packages ahead of time using the box number +that will be provided to you from the Colleges orientation office. You may also +consider purchasing larger items upon your arrival at our local shopping centers. + +Traveling Tips +You can book your flight as soon as you find the Colleges start date for international +Student orientation. at this time, you might have already received the schedule of +international student orientation. the Colleges provide a complimentary pick-up +from rochester and Syracuse airport for those international students coming to the +Colleges for the first time if you arrive during the assigned times. long-distance +traveling can be really exhausting, so try to get a good-night’s sleep before the trip, +eat lightly and drink plenty of water. it is also a good idea to wear several layers of +clothes so you can accommodate to changes in temperature. label your luggage with +both your home address and the address in the United States (if you don’t have an +address yet, use that of the international student office). Keep you passport, impor- +tant documents, flight tickets, and money or traveler’s checks with you at all times. +You might also want to pack everything you need for a one-night emergency stay +into your hand luggage in case you get stuck at the airport or your luggage gets lost. + +Money +You should carry a small amount of U.S. dollars with you while traveling so that you +can buy some food or other small things you might need. However, it is not a good +idea to carry large sums of money. We recommend that you buy traveler’s checks or +get a credit card that allows you to withdraw money from a cash machine when you +need to. Upon arrival on campus, the international student adviser can help you to +set up an account with a local bank. it is highly recommended to have an U.S. bank +account because it will make payments so much easier. Students can create bank +accounts free of charge. + +GETTING TO KNOW HOBART AND WILLIAM SMITH COLLEGES +College is an exciting time for everyone and it is a transition for all students. if you +are an exchange student, you already went through the transition process once. +nevertheless, studying abroad brings with it a range of new experiences. every col- +lege is a little different and if you have never studied in the United States before, +many things might be very confusing. + +Hobart and William Smith Colleges +Hobart and William Smith is a small, private liberal arts institution. You can expect +to study in relatively small classes and build close relationships with your classmates - Print ISSN: 2054-0957 (Print) +5 - Website: https://www.eajournals.org/ + and professors. Virtually all students are being housed on campus. living so close +to each other has many advantages: it is relatively easy to get to know new people, +get involved in clubs and activities, and make an impact on the community. at the +same time it can be hard to find personal time and separate school from other activi- +ties when the two are not physically separated. + +Classes +at HWS, a normal course load is four classes per semester. this means that you +will be in class for between 12 and 15 hours every week, or more if you take courses +requiring labs or studio work. there are different types of classes. 100- and 200-level +classes are usually introductory classes and have a wider focus. 300- and 400-level +classes have a narrower, more specific focus. Some professors prefer to lecture, but +you also encounter class-discussions. Higher level classes tend to be taught in a semi- +nar style, in which a lot of the class’ content relies on student presentations. there +are also independent study classes in which you research a topic under a professor’s +guidance but without the classroom structure. as an upper-class student you might +choose to work on an honors project. this is an independent research project that +will culminate in a 80- to 100-page paper. You can take classes from all departments +but at the end of your Sophomore (2nd) year, you will need to declare a major and a +minor subject (or one major, two minors; or two majors, one minor; or two majors, +two minors). as part of your liberal arts education, you will also have to fulfill cer- +tain academic goals. Your academic adviser will help you with that. + +Grading and Grades +Foreign students often find that americans are very competitive. this also reflects +on the system of grading. You will find that you will have many opportunities to +prove your knowledge of a subject in class. many professors will give short, unan- +nounced quizzes that test your knowledge of the course material assigned for that +day. exams are bigger, announced tests that cover the course material of a longer pe- +riod of time. they can be compromised of true or false questions, sentence comple- +tion (“fill in the blank”), multiple choice, matching, essay questions, or a combina- +tion of these. Professors will also assign essays and papers you have to write on your +own time. Your class participation might also count toward your final grade. the +Colleges use the grading system based on a four-point scale: a= 4.0, b= 3.0, C= 2.0, +d= 1.0. You need at least a C average in your classes to remain a full-time student. + +Homework +Some foreign students are overwhelmed by the quantity of homework assigned by +american colleges. especially in the beginning when you still have problems with +the english language, keeping up with homework can be really hard. if you find +yourself struggling to keep up with work, ask for help. Find the Center for teaching +and learning on campus and schedule an appointment! Ctl is open to all students +and its services are already covered by your tuition. they can help you with your +writing assignments, help with time management, or assign you a tutor to help you +with the content of a class. it is also important to realize that most american stu- +dents prioritize their work: decide what readings are important to do and read them +thoroughly; skim the others. +6 + + Professors and Advisers +every student at HWS is assigned an academic adviser. He or she will help you +with any questions you have related to your academic life. Feel free to consult them +before choosing classes, if you are struggling with certain classes or professors, if you +want to know how and when to declare majors and minors, and any other questions +you might have. most professors are also interested in helping you as much as pos- +sible and want to get to know you outside of class. don’t feel embarrassed to ask a +professor a question after class. You should also take advantage of their office hours. +interaction between professors and students in america is a lot tighter than in many +other countries. Students sometimes stop by their professor’s office simply to talk +about something that they are interested in. + +Important information on financial aid +Hobart and William Smith Colleges offer merit scholarships as well as need-based +grants to international students. the average package typically equals more than +half the annual cost of tuition, room and board. Highly competitive applicants will +be considered for a full tuition scholarships. + +Financial aid is available for a maximum of four years for those who enroll as + +first-year students. For those entering with advanced standing (transfers), aid is +available for the time allotted to complete a degree. awards based on demonstrated +financial need, if it becomes apparent that such need no longer exists, may be re- +duced. Students must resubmit a financial aid application each year. + +ADJUSTING TO A NEW ENVIRONMENT +going abroad for a semester or the duration of one’s studies is exciting. at times, +however, the confrontation with a foreign culture can be challenging. Studying +abroad is a learning experience in every sense of the word. besides your scholastic +learning in college courses, you will get to know the american culture and how +people around you view their world and their lives. by experiencing and understand- +ing a different way of life, you will also learn about things in your own culture you +have never thought of before. this is your chance to broaden your view and reflect +on and mature your own values. the process of learning about a new culture and +finding your own place in the new environment can be confusing and tiring. the +best way to deal with a new situation and find out about something you don’t under- +stand is to ask questions. + +Culture Shock and Adjustment +Culture shock is a normal phenomenon on your way to adjusting to a new environ- +ment. For some people, the encounter with a new culture can be shocking, indeed, +but for most people culture shock is not as sudden and dramatic as you might expect. +When the excitement about being in a foreign country wears off, cultural differences +can become overwhelming. You might think your language skills are inadequate; you +might be struggling with course work; you might be troubled by unfamiliar food; you +might have difficulties interpreting other people’s behavior; you might be offended +by the way people act or dress or by their body language. there are a variety of +symptoms of culture shock. Some people feel tired and nervous; some are frustrated + +7 + + with or even hostile toward the american culture; some get angry and irritated very +easily; some isolate themselves or socialize only with other international students and +people who speak their language. You might also feel very homesick and question +why you decided to go abroad in the first place. a way to prepare yourself to deal with +culture shock is to clarify for yourself why you want to study abroad. Knowing your +personal expectations and goals will help you overcome periods of doubt. + +THINGS TO DO +• Chose a non-credit or half-credit class like tennis, scuba diving, or basic sailing. +• Join some of the many student clubs on campus (http://www.hws.edu/studentlife/ + +campuslife/sll/clubs). + +• You like to be active? Visit the sport and recreation facilities and take a yoga class, +borrow a racquet and play tennis or racquetball. You can also try out for intramu- +ral sports. + +• get involved in the religious life on campus! there are services and meditation as +well as bible study sessions, ecumenical candlelight prayers, Sabbath dinners and +a lot more. + +• Participate in ongoing events – take in a movie on Fridays at Friday Flix, visit the +Cellar Pub in the basement of Coxe Hall and see what is happening (comedians, +bands, etc.), show school spirit by supporting one of many sporting events hap- +pening on campus. + +• are you interested in current events? Political and social developments? the +Fisher Center lecture Series, the President’s Forum and the President’s radio +Show are a few venues through which you can satisfy this interest. + +• travel the area! intercultural affairs (iC) and the Center for global education + +(Cge) will occasionally offer trips to see the area, oraP (outdoor recreational +activities Program) organizes hiking, rock climbing and rafting trips (among +others), and you can always get a group of friends together and go somewhere on +your own. + +• offer your help in the language departments. maybe someone needs a Spanish + +tutor or a teaching assistant for mandarin? + +• discover geneva. Volunteer. to find out how visit the Public Service office in + +trinity (2nd floor) (or call x3825). + +• do you like to dance? or watch others dance? the dance department is big on +campus and the students as well as the faculty put on dance shows! ask about +Koshare or join one of the dance teams. + +• Check the daily Update! there is always something going on – just show up and + +try it out! - Publication of the European Centre for Research Training and Development -UK +8 -Furthermore, the integration of artificial intelligence and machine learning has enabled predictive -maintenance systems that reduce equipment downtime by up to 45% [3]. + P +H +O +N +E +N +U +M +B +E +R +S +& +O +F +F +I +C +E +S + + Extension + +IMPORTANT PHONE NUMBERS +Office +business office +Center for global education +Center for teaching and learning +Counseling Center +deans: Hobart +deans: William Smith +Financial aid +Hubbs Health Center +information technology +intercultural affairs +registrar +residential education +Security +emergency +Student accounts + +3655 +3307 +3351 +3388 +3300 +3467 +3315 +3600 +4357 (HelP) +3319 +3651 +3900 +3656 +3333 +3386 + +OFFICE LOCATIONS +admissions +alcohol and other drug Programs +alumni House +athletics (WS) +buildings and grounds +business office +Career Services +Center for global education +Ctl +Chapel +College Store +Communications office +deans: Hobart +deans: William Smith +dining Services + +Financial aid +Hubbs Health/Counseling Center +information technology +intercultural affairs +President +Provost’s office +Public Service office +registrar +residential education +Security +Student accounts +Student life and leadership +WeoS radio Station + +bradford House +Smith Hall +615 South main +Winn-Seeley +medbery Parking lot +demarest (basement) +trinity (1st Floor) +trinity (3rd Floor) +Harris House (678 S. main St.) +St. John’s Chapel (near 630 S. main St.) +51 Saint Clair Street, Sherrill Hall +durfee House (639 S. main St.) +Smith Hall (1st Floor) +Smith Hall (1st Floor) +Scandling Center Café, Saga, +Cellar Pub, de Cordova Café +demarest (basement) +119 St. Clair St. +Williams Hall +288 Pulteney St. +Coxe Hall (2nd Floor) +Coxe Hall (2nd Floor) +trinity (3rd Floor) +gulick (1st Floor) +101 St. Clair St. +medbery Parking lot +demarest (basement) +Scandling Center (2nd Floor) +113 Hamilton St. + +9 + + BANKS AND MOVIE THEATERS IN GENEVA + +Community Bank +5 Seneca St. +geneva, nY 14456 +Phone: (315) 789-7700 +Website: www.communitybankna.com + +First Niagara Bank +811 Canandaigua road +geneva, nY 14456 +Phone: (315) 789-1049 +Fax: (315) 789-6009 +Website: www.firstniagarabank.com + +BANKS +Five-Star Bank +P.o. box 193 +geneva, nY 14456 +Phone: (315) 789-2300 +Fax: (315) 787-3263 +Website: www.five-starbank.com + +Seneca Falls Savings Bank +342 Hamilton St. +geneva, nY 14456 +Phone: 315-789-8100 +Website: www.seneca24.net/ + +MOVIE THEATERS +Geneva Movieplex 8 +t & C Plaza, Hamilton St. +geneva, nY, 14456 +Phone: (315) 789-4524 +Web: www.cinemanorth.com + +PLACES TO STAY +Note: the Colleges provide this list as a service to all who might visit our area, but +do not endorse any local hotel, motel, inn or bed and breakfast. + +the following are a handful of accommodations that are available in and around +geneva, no more than 30 minutes from campus. lodging is also available in ithaca, +rochester and Syracuse, which are each approximately one hour from Hobart and +William Smith. You can find out more about accommodations outside the greater +geneva area by visiting the geneva Chamber of Commerce or ontario County +tourism bureau websites. + +Holiday inn, Waterloo +(315) 539-5011 or (800) HolidaY + +Hampton inn +(315) 781-2037 + +ramada inn, geneva, lakefront +(315) 789-0400 + +belhurst Castle/White Spring manor, +geneva +(315) 781-0201 + +microtel inn & Suites, Seneca Falls +(315) 539-8438 + +geneva-on-the-lake, geneva +(315) 789-7190 or (800) 3-geneVa + +motel 6, Hamilton Street, geneva +(315) 789-4050 or (800) 466-8356 -The Era of On-Premises Infrastructure -The first major milestone in this journey was the establishment of on-premises data centers, which -fundamentally changed how businesses managed their digital operations. On-premises data centers -emerged as critical infrastructure components, offering organizations complete control over their hardware, -software, and data management processes. This infrastructure model provided significant advantages in -terms of data security and compliance, particularly for industries handling sensitive information. A key -benefit was the ability to maintain an average network latency of less than 5 milliseconds, crucial for time- -sensitive applications and real-time data processing [4]. +10 -The financial implications of on-premises infrastructure were substantial but justified for many -organizations. While initial setup costs typically ranged between $10,000 to $25,000 per rack, with annual -operating costs between $12,000 and $18,000, these investments provided organizations with direct control -over their technological infrastructure. The model proved particularly advantageous for businesses with -consistent workloads and stringent regulatory requirements, as it eliminated monthly recurring costs -associated with cloud services and provided better control over long-term expenses [4]. + G +E +N +E +V +A + +, + +N +E +W +Y +O +R +K + +PLACES TO EAT +*Please note that some restaurants are closed on Mondays or Tuesdays, +so call ahead. + +The area code for all numbers listed below is (315). + +American/Continental +abigail’s restaurant, routes 5 & 20, Seneca Falls, 539-9300 +belhurst Castle, lochland road, route 14, 781-0201 +the Crow’s nest, 415 boody’s Hill road, 781-0600 +emile’s restaurant, east north Street and routes 5 & 20, 789-2775 +geneva-on-the-lake, 1001 lochland road, 789-7190 +madderlake Café, West lake road, 781-2424 +the red brick inn, 291 main St., 789-3247 + +Diners/Grilles +Flour Petal Café, 34 linden St., 781-2233 +Parkers’ grille & tap House, 100 Seneca St., 789-4656 +Ports Cone & grill, 4432 W. lake road, 789-2020 +Water Street Café, 467 exchange St., 789-2560 +the Captain’s room, 372 exchange St., 781-0122 +dana’s time out, 258 Hamilton St., 781-2107 + +Italian-American and Pizza +nonna Cosentino’s trattoria, 1 railroad Place, 789-1638 +Uncle Joe’s Pizzeria, 99 n. genesee Street, 781-1199 + +International +ling ling, 481 Hamilton Street, 781-0668 +main moon, 480 exchange Street, 781-5106 +Wing tai oriental restaurant, 164 Castle Street, 789-8892 +garcia’s mexican restaurant, 313 Hamilton Street, 789-5760 + +Fast Food/Family Chains +routes 5 & 20 west of downtown geneva (Hamilton Street) offers an array of popu- +lar fast food and family chains, including Quiznos, tim Horton’s, Club 86 bagels +and Cakes, Friendly’s, KFC-taco bell, mcdonald’s, long John Silver’s, Pizza Hut, +Subway, Waffle Works and Wendy’s. + +11 + + FREQUENTLY ASKED QUESTIONS (FAQs) + +Do you require the TOEFL for international students? +Yes, we do require the toeFl (test of english as a Foreign language). We offer limited +english language support on campus, and require that students have a solid foundation +in english prior to admission. the minimum required computer based toeFl score is +220. the Colleges also will accept the aPiel examination in lieu of the toeFl. + +Do you offer financial assistance for international students? +need-based scholarships are available to international students. the cost of attend- +ing college here for the 2007-2008 academic year is U.S. $45,968. the average aid +package typically equals more than half the annual cost of tuition, room and board. +Highly competitive applicants will be considered for a full-tuition scholarship. the +financial aid application forms are included in the application for admissions. + +How many international students do you currently have on campus? +We have 32 full-time matriculated students in a degree program Fall 2006 +Semester. the following countries are represented: austria, bulgaria, Canada, +China, Colombia, Czech republic, germany, india, nepal, Pakistan, South Korea, +Sweden, turkey, Ukraine, United Kingdom and Vietnam. + +Do you have an international student adviser? +Yes, the student adviser is responsible for working with our international students +on campus to help them make the transition from home to the Colleges. We offer an +international student orientation prior to the first-year and transfer student orienta- +tion at the beginning of the academic year, and the beginning of the spring semester +for new arrivals. a variety of activities and programs are available throughout the +year. Student advising and counseling is available in a variety of other areas on cam- +pus, including faculty advisers, Hubbs Health Center, and the Center for teaching +and learning. We make every effort to work individually with each student to help +them reach their potential and make the most of the opportunities available here. + +Do you have an international student organization? +Yes. the international Student association (iSa) is a student organization that cel- +ebrates the diverse cultures found around the globe. other cultural clubs with interna- +tional perspectives represented on campus include asian Student association (aSU), +latin american organization (lao), Caribbean Student association (CSa), and +dia al-din: muslim Student association. + +Do you have E.S.L. services? +Yes, an entry level course is offered, as are workshops and one-to-one tutoring. How- +ever, the Colleges do not offer a full e.S.l. curriculum. + +What kinds of housing are available? +Within the Hobart and William Smith community are dozens of smaller communi- +ties--those of the individual residences. the range of housing options is extensive, -Security and compliance capabilities represented another crucial aspect of on-premises infrastructure. -Organizations maintained full control over their security protocols, physical access management, and data -handling procedures. This control was particularly valuable for industries subject to strict regulatory -requirements, such as healthcare and financial services, where data sovereignty and compliance are -paramount concerns. The ability to customize security measures and maintain direct oversight of data -storage and processing activities allowed organizations to achieve compliance rates exceeding 99% [4]. +12 -However, this approach presented notable challenges in terms of scalability and resource management. -Organizations needed to maintain dedicated IT teams for infrastructure management, with staffing costs -typically accounting for 40-50% of the total operational budget. Additionally, scaling operations required -significant advance planning and capital investment, with expansion projects often taking several months -to complete. The necessity for redundant systems and backup power solutions further added to the -complexity and cost of maintaining on-premises infrastructure [4]. + with coed or single-sex college residences, cooperative houses, theme houses, and +town houses among the choices you’ll find over your four years here. residence +halls offer a variety of configurations, with traditional singles and doubles as well as +suites. Some recent theme houses have included Jewish Culture House, multicul- +tural Sisterhood House, Writers House, and international awareness House. the +ten cooperative houses provide an alternative to standard residence-hall living, with +residents sharing responsibility for food and housecleaning arrangements. + +Is it possible to work on campus? +You are eligible to work up to 20 hours a week on campus, if your visa status permits +it. However, the demand for such employment exceeds the number of jobs available. +therefore, you cannot use on-campus employment as a resource for the issuance of +the i-20 form. + +STORAGE AND TRANSPORTATION OF SCHOOL YEAR ITEMS +Special consideration will be given to those international students who need storage +on campus during the school breaks. international students must attend a manda- +tory meeting in September regarding storage and transportation of school year items. +attendance will be taken at this meeting. Students unable to attend this meeting +must take the initiative to set up a meeting with the international Student adviser +before oct. 1. + +the storage and transportation meeting in September will encompass the following +information: +a. date and time of transportation to and from storage areas + +- Winter break (december) +- Summer break (may) +b. location of storage areas +c. How to pack and maximum number of storage items allowed +each student will be allowed no more than 8 items. items must be labeled with +students name, return date and number of items. boxes must be secured properly +and must be easily manageable. large items such as refrigerators, sofas, etc. will not +be accommodated due to limited space. +d. transportation requirements: items are to be placed at curbside 15 minutes +before van arrives for pick up. +e. Cost of transportation = $20 per student (gas and labor cost) +f. information on surcharges: the student’s account will be charged a surcharge for +each infraction: + +- No shows = $50 +- Fail to have items at curbside = $20 +- Not packed = $20 +- Change of time and/or date without 24 hour notice to Intercultural Affairs + Office = $40 + +Please Note: Storage space is limited, therefore there is no guarantee that you +will get storage or transportation support. + +13 + + IN CASE OF AN EMERGENCY +if you should have a need to contact the international Student adviser while travel- +ing outside of the United States please call (315) 781-3656 and ask for extension +3319. You can also contact the office directly by calling (315) 781-3319. + +In an emergency, they can be reached at (315) 781-3333. + +remember that the international Student adviser, while a resource to you, is not +your only resource during your time at the Colleges. While on campus, your respec- +tive deans will be available to assist you with a variety of issues, and should always +be kept aware of important matters. + +WHAT TO EXPECT DURING SCHOOL BREAKS +Since not everyone is able to return home during school breaks, international +students are allowed to remain on campus during those times. at specific times +during the fall and spring, you will receive e-mails from the residential education +office asking about their plans during the upcoming break. Your immediate response +is necessary as it is the only way to ensure that you will get permission to remain in +your room or elsewhere on the campus during school breaks. + +Students are not allowed to remain in their rooms during fall, winter and spring + +breaks. during summer vacation, students who remain on campus must move into +temporary housing. once again, an e-mail from the residential education office +is the first step in determining where students will be staying during the summer. +Currently, international students are allowed to stay on campus during the summer +holiday free of charge (subject to change). + +the dining hall is not open to students during the summer months so students +will need to provide their own meals. most summer housing will feature a kitchen, +but in the event that one doesn’t, appropriate accommodations will be made for +those students who wish to prepare their own meals. + +the international student adviser must also know your plans and should also be + +notified if you will or will not be staying on campus. + +US Holidays/Observances +January 1 +January 15 +3rd monday in February +2nd Sunday in may +4th monday in may +3rd Sunday in June +July 4th +1st monday in September +october 31 +november 11 +4th thursday in november +december 4 +december 25 +december 26 14 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +new Year’s day +martin luther King Jr. day +Presidents day +mother’s day +memorial day +Father’s day +independence day +labor day +Halloween +Veterans day +thanksgiving +Hanukkah begins +Christmas day +Kwanzaa + + SECTION II: +IMMIGRATION PROCEDURES + +HOW TO GET ASSISTANCE WITH IMMIGRATION MATTERS + +the international Student adviser is available to help you process your immigration +forms and documents. Working with the department of Homeland Security can be +very complicated, detailed, and time-consuming. it is important that you do not rely +on information obtained from friends and other students because circumstances may +vary from one student to another and the regulations can and do change. + the international Student adviser must work within the federal immigration +laws. When making a request, please make sure that you allow plenty of time for +your request to be processed and plan to be patient. Have your documents, such as +your passport, i-94 card, i-20 or dS-2019 with you when you go to the international +student office for immigration assistance. + +Important: You should allow at least one week for processing signatures, letters +and/or other types of paperwork. Special consideration will be given for emergency +situations. + +Reminder: the international Student adviser cannot remember when each +student’s documents expire. it is YOUR responsibility to keep track of all immigra- +tion documents and keep them current. + +Hristo Dimitrov ’09 (Bulguria), and Ayokunle Abogan ’06 (Nigeria) - Online ISSN: 2054-0965 (Online) +15 - Print ISSN: 2054-0957 (Print) + IMMIGRATION BASICS AND DEFINITION OF TERMS - Website: https://www.eajournals.org/ +as with all countries, the United States has laws and regulations governing foreign- +ers who are temporarily within its borders. the immigration regulations are adminis- +tered by the Department of Homeland Security (DHS). - Publication of the European Centre for Research Training and Development -UK +immigration regulations can change frequently and are difficult to interpret. it +is important for international students to understand the rules as they apply to stu- +dent status. The DHS does not consider ignorance of the law a legitimate reason +for failing to obey it. + Questions should be directed to the international Student adviser, who will be +able to assist with legal regulations and can sign your immigration documents. -Table 1. Digital Transformation Performance Metrics: Year-over-Year Analysis (2021-2024) +dHS requires that a copy of your passport, i-94 and other immigration docu- +ments be kept on file by the international Student adviser, as well as your local +address and telephone number. -Metric Category +Passport +be sure to keep your passport in a safe place. it is your most important document. +most passports are good for a limited period of time. Your passport must be valid at +all times while you are in the United States. -Equipment Downtime Reduction -Customer Inquiry Speed -Decision-Making Efficiency -Infrastructure Staff Cost -Network Latency (ms) +the expiration date of your passport should be six months beyond the date you -2021 (%) -25 -30 -40 -35 -20 +enter the U.S. if your passport expiration date is close to the date you will begin +classes, obtain a new passport before leaving home. -2022 (%) -35 -45 -55 -40 -15 +it is your responsibility to remember when your passport will expire and to con- +tact the nearest consulate or embassy for your country. to find the nearest consulate +or U.S. embassy check http://usembassy.state.gov. -2023 (%) -40 -55 -65 -45 -10 +Your passport can only be extended by your government and it usually takes +several weeks for a passport to be renewed, so be sure to allow enough time. You may +extend your passport in the U.S. at your home country’s embassy or consulate. You +are in this country illegally if your passport expires and are subject to deportation. -2024 (%) -45 -60 -70 -50 -5 +Visas +the visa stamped in your passport at the U.S. embassy or Consulate is for permis- +sion to enter the United States for a specific purpose and period of time. Students +are usually given an F-1 visa. J-1 visas are given to researchers, scholars and students +who are sponsored by agencies or universities. -The Cloud Computing Revolution -The emergence of cloud computing marked a paradigm shift in the digital transformation landscape, -fundamentally altering how businesses approach their IT infrastructure and operational capabilities. Recent -market analysis reveals that cloud spending now represents approximately 30% of IT budgets across -industries, with this figure expected to grow significantly. The industry is witnessing an unprecedented shift -toward multi-cloud and hybrid cloud strategies, with organizations increasingly focusing on cloud cost -optimization and management. This evolution has led to the emergence of specialized roles such as cloud -financial operations (FinOps) professionals, who are becoming critical to managing cloud investments -effectively [5]. - -The adoption of cloud computing continues to accelerate, driven by both necessity and opportunity. -Industry research indicates that companies can capture up to $1 trillion in annual value through cloud -adoption, highlighting the transformative potential of this technology. This value creation stems from -various sources, including innovation acceleration, cost reduction, and improved resilience. The impact is -particularly significant in Fortune 500 companies, where cloud adoption has enabled revenue growth of up -to $100 billion annually through broader cloud-supported initiatives [6]. - -Key Benefits of Cloud Computing: Infrastructure Flexibility -Cloud platforms have revolutionized resource allocation capabilities, enabling unprecedented operational -efficiency. The shift toward cloud-native architectures has become increasingly prominent, with -organizations moving beyond basic "lift and shift" approaches to more sophisticated cloud-native -implementations. This trend has been accelerated by the growing adoption of containers and microservices, -with Kubernetes emerging as a de facto standard for container orchestration. The market has seen a -significant focus on edge computing capabilities, with major cloud providers expanding their edge presence -to support growing demands for low-latency applications and improved performance [5]. - -Cost Optimization -The financial implications of cloud adoption are substantial and multifaceted. Research shows that -organizations can potentially reduce IT costs by 20 to 30 percent through cloud optimization initiatives. -However, the most significant value creation opportunities lie in business innovation, which can generate -$770 billion to $995 billion in annual EBITDA across Fortune 500 companies. This value stems from +Since the visa stamp is only important for entry and re-entry to the United +States, there is no need to worry if it expires while you are here if you are maintain- +ing your student status. because a visa is only a permit to apply for entry into a coun- +try, it is not possible to obtain a U.S. visa while you are in the U.S. to apply for a +visa, you must go to a U.S. consulate or embassy in a country other than the United +States, usually your home country. -15 +I-94 (Arrival/Departure Record) +Form i-94, also called the arrival/departure record of Stay Permit, is the small white +card given to you at the port of entry to the U.S. this form is very important since it +officially determines how long an international student can remain in the U.S. - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +16 - Online ISSN: 2054-0965 (Online) + Students on the F-1 and J-1 visas usually have their i-94 stamped “d/S” (dura- + +tion of Status), which means that the student can remain in this country for the +length of time required to complete his/her studies as long as the student follows all +of the regulations of the student visa. the eleven-digit number written on your i-94 +is your dHS admission number, which is used by the dHS to monitor entry into and +departure from the U.S. The I-94 cards are turned in when leaving the U.S. and a +new I-94 is completed with each subsequent entry. + +if you are on a visa other than a F or J, or if you have a specific date stamped on + +your i-94 card instead of the “d/S” designation, you must apply for an extension, a +Change of Status, or exit the U.S before the expiration date or you will be consid- +ered out of status and a visa overstay, which can have serious dHS consequences. + +SEVIS +an acronym for Student and exchange Visitor information System (SeViS). it is a +data collection and monitoring system that creates an interface between institutions +of higher education, the department of Homeland Security (dHS), consulates and +embassies abroad, and ports of entry. Schools are required to make regular electronic +updates in SeViS throughout each semester on the records of their enrolled students +in F-1 and J-1 status and their dependents and their researchers and faculty on J-1 +status. this includes, but is not limited to enrollment status, changes in address, +changes in level of study, employment recommendations and school transfers. + +Form I-20 +When Hobart and William Smith Colleges accepts an international student for +full-time study, the student is sent a Form i-20 a-b as proof of acceptance and to +use when applying for an F-1 Student Visa at a U.S. Consulate abroad. the student +presents the i-20 with their passport and visa at the Port of entry and the i-20 is +stamped and given back to the student. this is your Certificate of eligibility, and +it must be signed by a dSo/ international Student adviser before you leave the +United States. Students should keep copies of all i-20 issued to them. + +Immigration Status +this is often confused with “visa” but your immigration status, e.g., F-1, J-1, etc. is +determined at the time of your entry into the U.S. by an immigration official and is +noted on the Form i-94. You may have many visa stamps in your passport but, upon +entry into the U.S., an immigration inspector will admit you in only one immigra- +tion status which is noted on the i-94 card. + +OBTAINING A VISA AND FORM I-20 +Your passport should be valid at least six months beyond the date of entering the U.S. +as a student. be sure to check your passport expiration date. if it will expire close to +when you will begin classes, obtain a new passport before applying for your visa. + +How to Obtain a Visa +although U.S. embassies will not issue a student visa prior to 120 days before the +start date on the i-20 (or dS-2019 for J’s), you can schedule an appointment prior - Print ISSN: 2054-0957 (Print) +17 - Website: https://www.eajournals.org/ + to the 120-day period. Since processing times have become longer due to increased +security clearances, students should check with their local U.S. embassy or consulate +to learn the estimated waiting time and allow plenty of time to obtain the visa. a +list of consulates and embassies can be found at http://usembassy.state.gov/. + +You will need to pay a $100 SeViS fee at least 3 business days before your visa +interview. the fee can be paid online or by mail. For complete directions, see http:// +www.fmJfee.com. + Unless you are a sponsored J-1 student or already in another visa category, the +Colleges’ office of admissions will send you a Form i-20 when you are admitted and +have submitted proof of your ability to finance your education. take the i-20 form, +your acceptance letter, passport, copies of sponsorship letters and bank statements +to prove financial support, and a receipt for your SeViS fee payment to the United +States embassy or consulate nearest your home. once you have acquired the visa +stamp in your passport, you are ready to enter the U.S. + +Note: Students may not enter the U.S. more than 30 days prior to the start date +on their I-20. + +Canadian and some British Commonwealth citizens do not need visas to enter +the U.S., but they must request F-1 status and present their i-20 forms for process- +ing at the point of entry even if the official does not ask for it. they should also be +prepared to show evidence of financial support and proof of citizenship. Without a +stamped i-20 and an i-94 card with the F-1 or J-1 designation on it, you are con- +sidered a tourist in the United States and will not be legally able to study and take +advantage of the special privileges allowed those in Student Status until he or she +has been in an immigration status appropriate for studying for at least nine consecu- +tive months. + +Note: If you will not be entering the U.S. on an F-1 visa, do not use an I-20 form +to obtain a visa. If you received an I-20, please return it to the Office of Admis- +sions with an explanation of the type of visa you hold. + +if your visa document (i-20 or dS-2019) does not arrive in time for you to secure +the proper visa, do not enter the United States on a B1/B2 Visitor Visa. obtain a +b-2 “prospective student” or “prospective exchange visitor” visa from a U.S. embassy +or consulate post in your country. be certain the consul writes “Prospective Student” +or “Prospective exchange Visitor” (for J-1) in your passport on the visa page. You +will need to present your letter of acceptance and evidence of financial support to +request this visa. + +in general, it is best to avoid using the b-2 visa since you will have to pay a fee +and do a time-consuming change of status through dHS (department of Homeland +Security) after you arrive. You may not begin your studies until the change is com- +pleted, so this could also delay the start of your studies until a subsequent term. + +Note: Students must attend the school named on the I-20 or DS-2019 presented at +the port of entry in order to be in status. Also, do not enter the U.S. on a “W-T” - Publication of the European Centre for Research Training and Development -UK +18 -various sources, including accelerated product development, improved customer experience, and enhanced -operational efficiency. The cloud's consumption-based pricing model has enabled organizations to achieve -15 to 20 percent cost reduction in infrastructure spending through improved capacity management [6]. + (waiver) as this classification cannot be extended or changed and you will lose your +legal immigration status as a student. + +Students Transferring From another Colleges in the United States +if you are an F-1 student transferring from a U.S. institution and will not be leav- +ing the country prior to enrollment at Hobart and William Smith Colleges, follow +the procedures required by dHS. bring your completed i-20 to the international +Student adviser’s office as soon as possible after your arrival on campus. Sign the +document and leave it to be processed by a dSo/ international Student adviser +who will notify you when you may pick it up. + +if you leave the U.S. before beginning your studies at Hobart and William +Smith Colleges, use the new i-20 issued by the Colleges when you arrive at the port +of entry. Keep the i-20 from your old school with your immigration records, but do +not use it to enter the U.S. it is not necessary for transfer students with an unex- +pired student visa and who have maintained their status to obtain a new visa, even +though the former school’s name is listed. + +Transfer to another School +to be eligible to transfer schools, you must be pursuing a full-course of study at your +current school. otherwise you are “out of status” and must apply to dHS for rein- +statement to student status, which, if granted, will allow you to transfer. + +Students transferring to another school must notify the dSo/international +Student adviser of their intent to transfer. Hobart and William Smith Colleges +must release your SeViS record to the new school before a new i-20 can be issued +for the transfer. Upon arrival at the new school, the student should take the new +i-20 to the international Student adviser at the new school. J-1 students wishing to +transfer should consult with their sponsor. + +Changing Your Academic Level +if you are currently a student at Hobart and William Smith Colleges, and will be +continuing your studies in an additional program of study (such as a mat Program), +you should follow the transfer procedure. a new i-20 will be issued for the new pro- +gram, and you will use the new i-20 to enter the country. if you do not leave, bring +it to the international Student adviser as described in the transfer process in the +preceding paragraph. Failure to complete this “internal transfer” will cause you to +be out of status with dHS, which can adversely affect work authorizations or other +student privileges. + +DIFFERENCES BETWEEN F-1 AND J-1 STATUS +there are two types of entry visa issued to nonimmigrant students who intend to +study full-time at Hobart and William Smith Colleges—the F-1 student visa and the +J-1 exchange visitor visa. + +• Source of Funding: J-1 exchange visitor visa status is available to those students +who are supported substantially by funding other than personal or family funds. +Such funding may include that which comes from the U.S. government or the -Global Reach and Collaboration -Cloud infrastructure has enabled unprecedented global operational capabilities, with particular emphasis on -sustainability and environmental impact. Organizations are increasingly prioritizing environmental -considerations in their cloud strategies, with major providers committing to aggressive sustainability goals. -The industry has seen a growing focus on renewable energy usage and improved data center efficiency, -with cloud providers investing heavily in sustainable operations. This shift has resulted in significant -improvements in energy efficiency and reduced carbon footprints for organizations migrating to cloud -services [5]. +19 -The impact of cloud computing extends beyond operational benefits to strategic value creation. -Organizations leveraging cloud capabilities have reported significant improvements in their ability to -innovate and respond to market changes. The cloud has enabled faster time-to-market for new products and -services, with some companies reducing development cycles by up to 50 percent. Additionally, cloud -adoption has facilitated better risk management and improved security posture, with organizations able to -leverage advanced security features and compliance capabilities provided by cloud platforms. The value at -stake is substantial, with research indicating that companies can potentially capture between $900 billion -and $1.2 trillion in annual value through comprehensive cloud transformations [6]. + student’s home government, or an international organization, or the Colleges. Stu- +dents who are supported by personal or family funds are ineligible for J-1 immigra- +tion status and must come to the U.S. to study in F-1 immigration status. + +• Foreign Residence Requirement: Students in J-1 immigration status and the J-2 +dependents may be subject to a “foreign residence requirement.” J-1 students who +receive direct or indirect U.S. or home government funding, or who are studying +in fields for which personnel are considered in short supply in their home coun- +tries (most developing nations have “Country Skills” lists of varying lengths), are +ineligible to apply for a change to another non-immigrant status (except a or g) +or permanent residency in the U.S. until they have satisfied the “two year home +country physical presence requirement. only those in J-1 status with government +funding or whose skills are on Country Skills list are subject to the foreign resi- +dence requirement, which mean upon completion of the program of study they must +reside in their country of last legal permanent residence for two years before they are +eligible to apply for entry into the U.S. on an immigrant visa or on specialized work +visa (H or l). there is no foreign residence requirement for students on F-1 student +status. refer to the department of State’s website for additional information on the +exchange Visitor Program. + +• Medical Insurance: Students in J-1 status and their J-2 dependents are required to +have comprehensive medical coverage in order to fulfill U.S. government regula- +tions governing the J exchange Visitor status. Currently the medical insurance +offered by Hobart and William Smith Colleges does not cover all requirements for +those in J immigration status. the U.S. department of State specifies that program +participants and their dependents have medical coverage with the following mini- +mum requirements (22 CFr 62:14): medical benefits of at least $50,000 per accident +or illness; repatriation of remains in the amount of $7, 500, medical evacuation of +the exchange visitor to his or her home country in the amount of $10, 000, and a +deductible not to exceed $500 per accident or illness. For more information please +contact the international Student adviser. + +• Work Permission: Students in either F-1 or J-1 immigration status may work part +time (up to 20 hours per week) on campus with permission from the international +Student adviser. For both statuses, permission to work off campus based on eco- +nomic need may be requested only after the first full academic year of student status, +and only under extraordinary circumstances of unforeseen need. the United States +Citizenship and immigration Services (USCiS) authorizes F-1 students to work off +campus, whereas J-1 students submit an application for work permission to their visa +sponsor, i.e. the agency or school that issued the dS-2019 form. + +• Practical or Academic Training: Students in F-1 and J-1 immigration status are +eligible to engage in differing lengths of practical or academic training during their +studies and at the completion of their academic programs. Practical training, a +privilege of F-1 immigration status if you are eligible, is employment in one’s field of +study. those in F-1 immigration classification may be eligible for “curricular” practi- -16 +20 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 + cal training (which is authorized by the international Student adviser) during their +programs, in addition to one year of “optional” practical training that may take place +during or after one’s program of study. Permission to engage in practical training may +be granted only after you have been in valid student status for one academic year, +except in a graduate program that requires it from the first term. the international +Student adviser recommends optional practical training and the USCiS authorizes +it. refer to the Practical training overview for complete information on both cur- +ricular and optional practical training. academic training, a privilege of J-1 immi- +gration status if you are eligible, is employment which is integral, not just related, +to one’s field of study. one is eligible to apply for permission to engage in academic +training after one academic term in valid student status. a total of 18 months of +academic training is possible during or following completion of a program of study, +with the exception of some postdoctoral research positions that may qualify for up +to 36 months. an application for academic training is made to the J-1 student’s +exchange Visitor program sponsor. refer to academic training authorization for +those in J-1 exchange Visitor immigration Status for complete information. + +ARRIVING AT A U.S. PORT OF ENTRY … +WHAT A STUDENT CAN EXPECT +Plan your arrival: You may be refused entry into the United States if you attempt to +arrive more than 30 days before the start date listed on your SeViS i-20 form. + +Always carry your passport and I-20 on hand: do not check the following docu- +ments in your baggage. if your baggage is lost or delayed, you will be unable to pres- +ent the documents at your port of entry. as a result, you may not be able to enter +the United States. + +In addition, it is strongly recommended that you carry on hand the following + +documentation: + +1. evidence of financial resources +2. evidence of student status, such as tuition receipts +3. name and contact information for your international Student adviser, + +including a 24-hour emergency contact number at the school + +Complete your entry paperwork: +Arriving by Air: Flight attendants will distribute Customs declaration Forms and +arrival departure Forms (i-94). these must be completed prior to landing. + +Arriving by Land or Sea: the CbP officer at the port of entry will provide the +necessary Customs declaration Form and the arrival-departure Form (i-94) to be +filled out upon arrival. + as you arrive at the port of entry proceed to the terminal area for arriving +passengers. Have the following documents available for presentation: you passport, +i-20, arrival-departure Form (i-94), and the Customs declaration Form. Your docu- +ments will be examined and processed by the department of Homeland Security +(dHS) at the port of entry. Your passport will be returned to you along with an i-94 +card (arrival/departure record) designated “d/S” which means “duration of status” - Online ISSN: 2054-0965 (Online) +21 - Print ISSN: 2054-0957 (Print) + and refers to the time you are engaged in a full-time course of study as indicated on +your visa document. the dHS official will stamp and return your i-20 to you. + +Note: Your documents are extremely important. They legitimize your presence in +the United States. Keep them in a safe and secure place. + +For Further Information: +Preparing to Study in the U.S. http://educationusa.state.gov/ +applying for a Visa to Study in the U.S. http://travel.state.gov/visa/temp/types/ +types_1268.html + +MAINTAINING & BEING IN STATUS WITH DHS +an F-1 student is considered in-status for the entire length of time during which +he/she is enrolled as a full-time student in an educational program if adhering to +the conditions of the visa designated on the i-94 card, plus any authorized period of +practical training and an additional sixty (60) days to prepare for departure. be sure +to pay close attention to the conditions described below: + +• report to the international Student adviser for SeViS registration upon arrival +at Hobart and William Smith Colleges. + +• attend only the school authorized. this means that you must only attend the +school whose i-20 you last used to enter the U.S. or the school whose i-20 you used +to transfer schools. if you decide to withdraw from the Colleges before completing +your program, you are not allowed to remain in the U.S. with in-student status un- +less you transfer to another institution. + +• Complete a full course of study for each academic term and make normal academ- +ic progress towards completion of program. attendance and/or full-time enrollment +is not required during the summer. at Hobart and William Smith Colleges 4 credit +hours per semester is required and exceptions can be made on the recommenda- +tion of the international Student adviser for valid medical or academic reasons as +defined by the dHS. be sure to speak with the international Student adviser about +any exceptions beFore dropping below full-time status in order to avoid the seri- +ous consequences that may result from being out of status. + +• transfer or extend program in a timely manner. duration of Status expires if a +student takes longer than expected to complete a given academic level (e.g. bachelor’s, +master’s, and doctorate). the amount of time permitted for completing studies at a +given level is determined by the date on the initial i-20 issued at the beginning of each +academic level. Students must pay close attention to the expected completion date +noted in item 5 on the form i-20 for each level. if more time is needed, an applica- +tion for extension of stay must be filed with the dHS. there are certain requirements +to meet in order to be eligible for an extension, so you will need to consult with the +international Student adviser at least 45 days before the completion date. - Website: https://www.eajournals.org/ +22 - Publication of the European Centre for Research Training and Development -UK + • Keep continuity in the program of study (in general an absence from the U.S. of +more than 5 months terminates status. However, an authorized study abroad pro- +gram does not break continuity). -Table 2. Cloud Adoption: Cost Reduction and Efficiency Analysis [5, 6]. +• refrain from unauthorized work. When you apply for an F-1 visa, you must prove +that you have full financial support. the law assumes that you will not be employed +in the U.S. and therefore will not accept employment that could be filled by a U.S. +worker or student. there are some exceptions to this policy provided that full-time +enrollment during the year is maintained and the appropriate permission is ob- +tained, but you should always check with the international Student adviser before +accepting employment, especially if it is off campus. -Metric Category +• When traveling abroad, always return in proper F-1 status. -Q1 2024 (%) Q2 2024 (%) Q3 2024 (%) Q4 2024 (%) +• maintain a valid passport for 6 months into the future (Some countries have +agreements with the U.S. that recognize the passport as valid for 6 months after the +expiration date). -IT Budget Cloud Spend -IT Cost Reduction -Infrastructure Cost Savings -Development Cycle Reduction -Market Response Improvement +• report any address changes to the international Student adviser within 10 days +of the change. dHS regulations require that the international Student adviser +maintain a record of the current address where you are living. therefore, you must +always notify the international Student adviser immediately of any change of ad- +dress. addresses with a post office box number or addresses of your academic depart- +ment are not acceptable to the dHS. -15 -20 -15 -35 -25 +• Comply with Special registration requirements if applicable (nSeerS). -20 -23 -17 -40 -30 +• File timely requests of change of status, practical training and other benefits. -25 -27 -18 -45 -35 +• obey all State and Federal laws. -30 -30 -20 -50 -40 - -Advanced Technologies Enabled by Cloud Computing - -Artificial Intelligence and Machine Learning -Cloud platforms have fundamentally democratized access to AI and ML capabilities, creating -unprecedented opportunities for organizations across industries. According to recent market analysis, the -global artificial intelligence market size was valued at USD 454.12 billion in 2023 and is projected to surge -to USD 2,575.16 billion by 2032, growing at an impressive CAGR of 21.4%. This remarkable growth is -driven by the increasing adoption of cloud-based AI solutions across various sectors, including healthcare, -automotive, and retail. The integration of AI with cloud computing has become particularly crucial in the -post-pandemic era, with organizations accelerating their digital transformation initiatives to maintain -competitive advantage [7]. - -The impact of cloud-enabled AI extends across multiple domains, with North America leading the market -share at 35% in 2023. This dominance is attributed to the presence of major technology companies and -widespread adoption of AI technologies in various industries. The Asia-Pacific region is experiencing the -fastest growth, with a projected CAGR of 24.3% during the forecast period, driven by increasing -investments in AI research and development, particularly in countries like China, Japan, and South Korea -[7]. - -Big Data Analytics and IoT Integration -The transformation in data handling and analysis capabilities through cloud computing has revolutionized -how organizations process and utilize information. Edge-cloud infrastructure has emerged as a crucial -component in handling the massive data generated by IoT devices, with studies showing that edge -computing can reduce data transmission costs by up to 30% while improving response times significantly. -The integration of advanced analytics algorithms in edge-cloud architectures has enabled real-time -processing capabilities, crucial for applications requiring immediate insights and actions [8]. - -The evolution of IoT data analytics has led to the development of sophisticated middleware architectures -that efficiently manage data flow between edge devices and cloud platforms. Research indicates that hybrid -edge-cloud solutions can reduce latency by up to 50% compared to traditional cloud-only approaches. These -architectures implement various data analytic algorithms, including deep learning, reinforcement learning, +• depart form the U.S. in a timely manner. -17 +• do not enter on a tourist visa (b-1 or b-2). - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Note: Students who fail to maintain full-time enrollment are not eligible for any +of the benefits of the student visa, such as on-campus and off-campus employ- +ment, practical training, remaining in the U.S. during summer vacation, transfer +to another school or change of level, or registration for further courses. they are +considered to be “out of status” and must apply to dHS for reinstatement to valid +student status. eligibility for benefits will be regained only if and when the dHS +approves the reinstatement. if you have any questions regarding reinstatement, see +the international Student adviser immediately. - Online ISSN: 2054-0965 (Online) +23 - Print ISSN: 2054-0957 (Print) + EXCEPTIONS TO A FULL COURSE OF STUDY - Website: https://www.eajournals.org/ +• the normal vacation period; - Publication of the European Centre for Research Training and Development -UK +• medical problems documented by a licensed medical doctor, osteopath or clinical +psychologist. approval for reduced or no course load for a medical condition is limited +to 12 months total for the program of study; -and federated learning, enabling more efficient processing of IoT data streams. The middleware layer plays -a crucial role in managing these processes, with modern implementations supporting dynamic resource -allocation and automated scaling based on workload demands [8]. +• teaching or research assistantship; -The convergence of IoT and cloud computing has particularly benefited from advances in edge computing -capabilities. Contemporary edge-cloud infrastructures support multiple types of edge analytics, including -descriptive, diagnostic, predictive, and prescriptive analytics. This comprehensive approach enables -organizations to process data closer to its source, reducing bandwidth requirements while maintaining data -quality and analysis accuracy. Studies have shown that edge-cloud implementations can achieve up to 40% -improvement in energy efficiency compared to traditional cloud-only solutions [8]. +• Valid academic reasons, such as first semester language difficulties and/or adjustment +problems, improper course placement, unfamiliarity with american teaching methods; -Security considerations have become increasingly important in IoT data analytics, with edge-cloud -architectures implementing sophisticated security measures at multiple levels. The integration of -blockchain technology with edge-cloud infrastructure has emerged as a promising solution for ensuring -data integrity and secure communication between devices. Furthermore, the development of privacy- -preserving machine learning techniques has enabled organizations to maintain data confidentiality while -performing complex analytics tasks [8]. +• the final term when the student needs less than a fulltime course load to complete +program of study. Student must be enrolled in at least one credit; -Table 3. Cloud-Enabled Technologies: Growth and Efficiency Analysis [7, 8]. +• during a period of full-time optional or Curricular Practical training. -Technology Metric +DURATION OF STATUS FOR F-1 STUDENTS +an F-1 student is admitted to the U.S. for a period known as “duration of Status” +which is recorded upon entry documents i-94 and i-20 with the notation “d/S”. du- +ration of Status refers to the period a student is pursuing a full course of study, plus +any authorized practical training following completion. The dates on the I-20 de- +termine the initial parameters of the duration of status. A student can be admit- +ted to the U.S. up to 30 days prior to the report date on the I-20 and may remain +60 days after completion of program or practical training. Certain factors may +reduce grace period to 15 days or to 0 days. an international Student adviser au- +thorized withdrawal from classes is allowed a 15-day period for departure. a student +who fails to maintain status is not eligible for an additional period for departure. -2023 Q3 (%) 2023 Q4 (%) 2024 Q1 (%) 2024 Q2 (%) +TEMPORARY DEPARTURE FROM THE U.S. +If you plan on leaving the U.S. temporarily to visit your homeland or some other +country, an International Student Adviser must validate your I-20 for re-entry into +the U.S. You may not be allowed back into the U.S. without the validated immigra- +tion documents, a current visa, and a passport which is valid for at least six months +after you re-enter the country. + most students with valid status may travel to Canada, mexico and many of the +adjacent islands if they have a valid passport, a valid i-94 and a valid i-20 or dS-2019. +if you re-enter the U.S. in fewer than 30 days, an unexpired U.S. visa is not required. +However, Canada, mexico and the island nations do require entry visas for citizens of +certain countries. -North America AI Market Share -APAC Growth Rate -Data Transmission Cost Reduction -Edge-Cloud Latency Reduction -Energy Efficiency Improvement +it is always best to check with the embassy or consulate of any country you wish -35 -20 -20 -35 -25 +to visit. Since certain countries are ineligible for this benefit and travel regulations are +subject to change, you should check with the international Student adviser if you +intend to travel with an expired visa. -37 -22 -25 -40 -30 +Students who apply for a new visa in contiguous territory forfeit the right to re- -40 24 -28 -45 -35 -42 + enter with an expired visa. if the new visa is denied, the student would not be able to +re-enter the United States to study until a valid visa is obtained in their home country. + +ACCEPTING U.S. GOVERNMENT PUBLIC ASSISTANCE OR PUBLIC +BENEFITS +international students and scholars should not, under any circumstance, accept public +benefits from the federal, state, county or any other local government agencies in the +United States. this includes, but is not limited to, participation in programs such +as medicare/medicaid, Supplemental Social Security income, aid to Families with +dependent Children, food stamps, and subsidized housing. accepting such benefits +can cause you to lose your legal status in the United States and create a bar to your +re-entry to this country. + +Cooperation between state social service agencies and the inS and department + +of State make it possible for an inS inspector or U.S. Consular officer to become +aware of an individual who has accepted public benefits. the acceptance of public +benefits is interpreted as a declaration that the individual is unable to meet his or her +financial obligations (a requirement of obtaining and maintaining status) and thereby +rendering him or herself ineligible for F-1 or J-1 status. + occasionally a well-meaning social service agency, hospital or public housing admin- +istrator may offer such assistance not realizing that non-immigrants are ineligible for such +benefits. additionally, even though U.S. citizen children of non-immigrant parents may be +eligible for pre and post-natal public benefits, the non-immigrant parent may be denied a +visa or re-entry based on the child’s receiving the benefits. again this is interpreted as the +non-immigrant individual not being able to meet his or her financial obligations. + +Information on U.S. Social Security Numbers (SSN) +if you will be employed in the United States, including at Hobart and William Smith +Colleges, you must obtain a U.S. Social Security number, which is used by the U.S. +government to identify wage earners for tax purposes. the number assigned to interna- +tional students is not a valid U.S. Social Security number. + +to apply for a Social Security number (SSn) you must obtain a “proof of work” + +form from the international Student adviser confirming your student or exchange +visitor status, that you are authorized to engage in on-campus employment, the nature +of the employment to be engaged in, and identify the department/office for whom you +will be working before being able to apply to the Social Security administration for +a SSn. You must also present your passport, your i-20 or dS-2019 form and your i-94 +form. the Social Security administration office is open monday-Friday from 9:00 a.m. +to 4:00 p.m. (closed on federal holidays). + +Social Ssecurity numbers will not be issued to foreign nationals who are not +authorized to work. a Social Security number is not required for opening a bank +account, registering for school, filing a U.S. tax form or applying for a driver’s license. +if you are required to file a tax return and are not eligible for a U.S. Social Security +number, you will have to apply to the internal revenue Service (irS) for a tax iden- +tification number (tin) on form W-7. this form may be obtained from the Colleges +payroll department. + +For more information about U.S. Social Security Numbers visit the website or + +call their toll-free number at 1-800-772-1213. + +25 + + WORK OPPORTUNITIES FOR STUDENTS IN F-1 STATUS +The following is a list of all opportunities to obtain authorization to work in the +United States if you are in F-1 immigration status. Each category may allow for +authorization under different circumstances and at different times during your +stay in the U.S. This is summary of work benefits but does not include all regu- +lations governing employment. + +On-campus Work: Full-time students are allowed to work up to twenty hours per +week during the academic year. on campus employment means that you are paid by +Hobart and William Smith Colleges or working at a Colleges location with an em- +ployer that is providing services to students, such as Sodexho’s Saga dining hall or +the Café. on-campus employment requires an i-9 employment verification autho- +rization form to be completed by you at the Colleges business office. + +Off-campus Work Based on Financial Need: Students who have been in F-1 status +for one academic year and who are experiencing extreme financial difficulties due to +unforeseen changes that severely affect their financial resources may apply for work +permission. authorization is given by the dHS with the recommendation of the +international Student advisers/ dSo. + +Internship with an International Organization: Students in F-1 status are eligible +to apply for permission to work for international organizations such as the World +bank, the international monetary Fund, etc. authorization is granted by the dHS +with the recommendation of the international Student adviser/dSo. + +WORK OPPORTUNITIES FOR STUDENTS IN J-1 STATUS +J-1 On-Campus Employment +enrolled J-1 students may work on campus up to 20 hours per week during the +semester and full time during the break with the permission of their sponsor. if the +Colleges are your sponsor please see the international Student adviser. + +J-1 Off-Campus Employment +if a J-1 student needs employment during the academic year due to financial hard- +ship arising form unforeseen circumstances, the program sponsor may authorize it. +Such employment cannot interfere with full-time pursuit of the exchange visitor’s +program objective and is limited to 20 hours per week during the academic year and +full-time during break. + +PRACTICAL TRAINING: +this is defined as employment related to your field of study and is available dur- +ing and after completion of your studies. the following is a list of the categories of +practical training: + +Curricular practical training (before completion of your studies) is defined as + +employment pursuant to a required or optional internship which is an integral part +of the established curriculum of your school. + 26 -30 -50 -40 - -Emerging Trends in Cloud Computing - -Edge Computing and Serverless Architecture -Edge computing has emerged as a transformative force in the cloud computing landscape, bringing -computation closer to data sources and end-users. The integration of edge computing with serverless -architecture represents a significant evolution in cloud technology, addressing critical challenges in modern -application development and deployment. This combination has become particularly crucial for real-time -processing requirements, with edge computing reducing latency by processing data closer to its source -while serverless architecture eliminates the complexity of server management. The implementation of edge- -serverless solutions has shown remarkable improvements in application performance, with organizations -reporting reduction in latency from seconds to milliseconds for critical operations [9]. -The impact of serverless computing extends beyond performance improvements. Recent implementations -have demonstrated significant advantages in terms of cost efficiency and resource utilization. Organizations -18 + optional practical training is defined as employment related to your field of study. +it is authorized in variable increments, deducted form the maximum of one year of +optional practical training per educational level, and is available in the following +variations: - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -adopting serverless architectures have reported substantial reductions in operational overhead, as the -technology eliminates the need for traditional server management and scaling operations. The pay-as-you- -go model inherent in serverless computing has revolutionized cost structures, with businesses only paying -for actual compute time used rather than maintaining constantly running servers. Furthermore, the -integration of edge computing with serverless architecture has enabled more efficient handling of IoT -devices and real-time data processing requirements, resulting in improved application responsiveness and -user experience [9]. - -Hybrid Cloud Solutions -The hybrid cloud market has witnessed extraordinary growth and transformation, with current valuations -at USD 183.7 billion and projected to reach an impressive USD 405.62 billion by 2032, growing at a CAGR -of 9.2%. This remarkable expansion is driven by increasing demands for flexible, scalable, and secure cloud -solutions across various industries. The financial services sector has emerged as a particularly strong -adopter of hybrid cloud solutions, leveraging the technology to balance security requirements with -operational efficiency [10]. - -The adoption of hybrid cloud solutions has been significantly influenced by several key factors, including -the rising need for enhanced data security, disaster recovery capabilities, and the growing complexity of -regulatory compliance requirements. Organizations in the banking, financial services, and insurance (BFSI) -sector have been at the forefront of hybrid cloud adoption, particularly in North America, which currently -holds approximately 35% of the global market share. The integration of artificial intelligence and machine -learning capabilities within hybrid cloud environments has further accelerated adoption rates, enabling -organizations to optimize workload placement and resource utilization more effectively [10]. - -Market analysis reveals that large enterprises currently dominate the hybrid cloud market, accounting for a -significant portion of the overall market share. This dominance is attributed to their complex IT -requirements and substantial resources for implementation. However, small and medium-sized enterprises -(SMEs) are increasingly adopting hybrid cloud solutions, driven by the technology's ability to provide -enterprise-grade capabilities without requiring extensive upfront investments. The public sector has also -emerged as a significant growth driver, with government agencies increasingly migrating to hybrid cloud -environments to modernize their IT infrastructure while maintaining strict security and compliance -requirements [10]. - -Looking Ahead: Future Developments -The convergence of edge computing, serverless architecture, and hybrid cloud solutions continues to -evolve, with emerging technologies and methodologies shaping the future of cloud computing. Edge- -serverless platforms are becoming increasingly sophisticated, incorporating advanced features such as -automated scaling, improved security protocols, and enhanced development tools. The integration of -artificial intelligence and machine learning capabilities at the edge is expected to further transform -application development and deployment paradigms, enabling more intelligent and responsive systems [9]. +1. optional practical training after completion of your studies +2. optional practical training before completion of your studies with three pos- -19 +sible variations: - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +a. practical training during the annual vacation. the annual vacation must - Online ISSN: 2054-0965 (Online) +follow at least two consecutive terms of fulltime enrollment and is usually taken in +the summer - Print ISSN: 2054-0957 (Print) +b. practical training when school is in session, provided that employment - Website: https://www.eajournals.org/ +does not exceed 20 hours a week - Publication of the European Centre for Research Training and Development -UK +c. practical training following the completion of all the course requirements -Table 4. Cloud Architecture Evolution: Latency and Processing Metrics [9, 10]. +for the degree except the thesis or equivalent -Performance Indicator +Optional Practical Training (OPT) Before Completion of Degree Requirements +definition: employment related to your field of study before completion of all degree +requirements. it is authorized in variable increments and is deducted from the +maximum of one year (12 months) of optional practical training per educational +level. there are three circumstances under which this type of practical training can +be authorized: -Traditional Edge Only Serverless Edge-Serverless +1. Following completion of all degree requirements except the thesis or equivalent +2. during the annual vacation +3. Part-time (up to 20 hours per week) during the academic year -Processing Latency (ms) -Resource Overhead (%) -Server Management Time (hrs/week) -Response Time (ms) -Infrastructure Complexity Score +Eligibility to Engage In OPT: You must: +• be a registered student in F-1 status physically in the United States at the time of +application +• have been enrolled in lawful student status on a full-time basis for at least one full +academic year +• not have used twelve months or more of full-time curricular practical training +oPt is permitted for a period of up to a maximum of twelve months per educational +level (bachelors, masters, etc.). -45 -65 -40 -50 -75 +Procedure: obtaining authorization for oPt is a two-step process. Your first step in +obtaining an employment authorization document (ead) is to qualify for a certifica- +tion from the international Student adviser/dSo recommending you for oPt. the +international Student adviser/dSo cannot authorize oPt - only the United States +Citizenship and immigration Services (USCiS) can do so by issuing you an ead. -25 -45 -30 -35 -55 +You may not begin employment until you have been issued an ead by the US- -15 -35 -20 -20 -40 +CiS and can work only within the time period indicated on the card. requests for +employment authorization must be mailed to the USCiS regional Service Center +having jurisdiction over your place of residence and may require 6 to 12 weeks for +adjudication. -5 -25 -15 -8 -30 +27 -Security and Compliance in the Cloud Era -The landscape of cloud security continues to evolve rapidly in response to emerging threats and changing -business requirements. Recent analysis indicates that cloud security remains a top priority for organizations, -with 82% of companies increasing their cloud security budgets in 2024. The shift towards zero trust -architecture has become particularly prominent, with 76% of organizations either implementing or planning -to implement zero trust security models. This transition has been driven by the growing recognition that -traditional perimeter-based security approaches are insufficient in today's distributed cloud environments, -where data and applications are accessed from multiple locations and devices [11]. - -Identity and access management has emerged as a critical component of modern cloud security strategies. -The implementation of comprehensive IAM solutions has become essential as organizations grapple with -increasingly sophisticated cyber threats. Studies show that organizations with mature identity and access -management practices experience significantly fewer security incidents, with advanced authentication -mechanisms reducing unauthorized access attempts by up to 67%. The adoption of biometric authentication -and behavioral analytics has further strengthened security postures, with organizations reporting a 45% -reduction in credential-based attacks [11]. - -The financial implications of data breaches have underscored the importance of robust cloud security -measures. According to IBM's Cost of a Data Breach Report 2024, the global average cost of a data breach -has reached $4.45 million, marking a 15% increase over three years. Organizations leveraging AI and -automation in their security operations have demonstrated significantly better outcomes, with automated -security tools reducing the average time to identify and contain breaches by 108 days. Furthermore, -companies implementing zero trust architecture have seen an average reduction of $1.17 million in breach -costs compared to those without such security measures [12]. - -The impact of remote work and hybrid environments continues to shape cloud security strategies. -Organizations with over 50% of their workforce working remotely have reported higher average breach -costs, reaching $4.99 million compared to $4.24 million for organizations with less remote work. This -disparity has driven increased adoption of advanced security technologies, with 49% of organizations now -implementing AI platforms for cybersecurity. The study reveals that organizations using security AI and + When to Apply: +• not more than 90 days prior to being enrolled for one full academic year· +• not more than 120 days prior to the employment start date that you select +Please note: the employment start date cannot be earlier than the date on which +you will have completed one full academic year. + +How to Apply: +Obtaining the International Student Adviser recommendation +Since you must first obtain a new SeViS i-20 with the international Student ad- +viser/dSo’s oPt endorsement, which may require 5 working days, you should make +every effort to apply as early as possible within the 90 or 120 day window. You might +need to come twice to the international Student adviser/dSo—once to submit the +application and another to pick up the approved recommendation on a SeViS i-20. + +Curricular Practical Training in F-1 Status +Definition: Curricular practical training (CPt) is employment pursuant to an intern- +ship requirement or internship elective which is an integral part of the established +curriculum of your school, which will be counted towards completion of your degree. + +Eligibility to engage in CPT: You must be a degree candidate who has been in +lawful status for at least one academic year. You must have been offered a specific +training opportunity that must fall into one of the following categories: +• to satisfy an internship requirement expected of all students in your program +and necessary for completion of your degree; +• to satisfy an internship elective which counts towards completion of your degree. +the internship elective must be listed in your school’s bulletin and must state in the +official course description that it is specifically designed to allow students enrolled in +the course to engage in off-campus internships; +• the internship may be done to engage in research necessary and integral to the +completion of the doctoral dissertation. + +Restrictions: U.S. Citizenship and immigration Services (USCiS) regulations do +not place an absolute limit on the amount of CPt you may utilize. However, if you +engage in one year or more of full-time CPt, you will lose your eligibility to engage +in optional Practical training. CPt is always part-time during the academic year +unless the student is engaged in doctoral dissertation research. CPt may be full time +during the annual vacation. You may engage in curricular practical training only +after your SeViS record has been updated by the international Student adviser and +the i-20 has been printed, indicating the approved employment. the endorsement +will indicate the precise details of the authorized training, including the name and +location of the employer, the specific dates of the training period, and whether it +is full-time or part-time. We encourage you to contact the international Student +adviser/dSo if you have questions. -20 +28 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -automation experienced breach lifecycle times that were 108 days shorter compared to those not utilizing -these technologies [12]. - -Cloud security compliance has become increasingly complex with the evolution of regulatory requirements -and data protection standards. Recent trends show a significant shift towards automated compliance -monitoring and reporting tools, with organizations implementing these solutions reporting 43% lower -compliance-related costs. The adoption of cloud security posture management (CSPM) tools has grown -substantially, with 65% of organizations now using these platforms to maintain continuous compliance and -security assessment capabilities. This has led to a 55% improvement in the detection and remediation of -compliance violations [11]. - -Future Outlook: Cloud Computing and Digital Transformation -The landscape of cloud computing continues to evolve rapidly, with organizations increasingly focusing on -strategic cloud adoption and optimization. According to recent industry analysis, 2024 marks a significant -shift toward more sophisticated cloud strategies, with 94% of enterprises already utilizing cloud services -and 87% adopting multi-cloud approaches. This evolution is particularly noteworthy in the realm of cloud -cost optimization, where organizations are implementing FinOps practices to better manage and optimize -their cloud spending. The trend toward cloud-native development has become increasingly prominent, with -enterprises reporting 30-40% faster application development cycles through cloud-native approaches [13]. -The integration of artificial intelligence and machine learning capabilities has become a cornerstone of -modern cloud strategies. Organizations are increasingly leveraging AI-powered cloud services for various -purposes, from automated infrastructure management to intelligent application development. The adoption -of cloud-native platforms has enabled businesses to achieve remarkable improvements in operational -efficiency, with companies reporting 25-35% reduction in development cycles and 40-50% improvement -in resource utilization. Furthermore, the implementation of containerization and microservices architectures -has led to significant improvements in application scalability and maintenance efficiency [13]. - -Sustainability and environmental responsibility have emerged as critical priorities in cloud computing -strategies for 2024 and beyond. Organizations are increasingly focusing on green cloud computing -initiatives, with major providers implementing advanced energy management systems and sustainable data -center practices. This shift towards environmentally conscious cloud operations has become a key -differentiator in vendor selection, with businesses prioritizing providers that demonstrate strong -commitments to sustainability. The industry is witnessing a growing emphasis on energy-efficient cloud -architectures and sustainable computing practices [14]. - -Security and compliance considerations continue to shape the future of cloud computing, with zero trust -architectures gaining prominence. The industry has seen a significant shift toward comprehensive security -frameworks that encompass everything from infrastructure protection to application security. Organizations -are increasingly adopting automated security solutions and implementing advanced threat detection -systems. The focus on regulatory compliance has intensified, with businesses implementing sophisticated + Departure Checklist -21 +q admissions letter from HWS - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +q evidence of Financial resources - Online ISSN: 2054-0965 (Online) +q Passport with Student Visa - Print ISSN: 2054-0957 (Print) +q Form 1-20 or dS-2019 - Website: https://www.eajournals.org/ +q Copies of medical records and other important documents translated into english - Publication of the European Centre for Research Training and Development -UK +q if J-1 visa holder – additional health insurance requirements. this must also be -compliance monitoring and management tools to ensure adherence to evolving regulatory requirements -[14]. +translated into english, if needed. -The evolution of cloud management and governance frameworks represents another significant trend. -Organizations are implementing more sophisticated approaches to cloud governance, incorporating -advanced monitoring tools and automated policy enforcement mechanisms. The emphasis on operational -excellence has led to the adoption of advanced observability platforms and AIOps solutions, enabling better -visibility and control over cloud environments. This trend is particularly evident in the increasing adoption -of cloud management platforms that provide comprehensive visibility and control across multi-cloud -environments [14]. +q emergency outfit packed in carry-on luggage -CONCLUSION +q Calling card to use during travel -Cloud computing stands as a transformative force in the digital era, reshaping how organizations operate, -innovate, and deliver value. The convergence of cloud technologies with AI, machine learning, and IoT has -created powerful new capabilities for businesses of all sizes. As organizations continue to adapt to changing -market demands, the role of cloud computing extends beyond technological infrastructure to become a -strategic enabler of business transformation. The future holds promise for further innovations in edge -computing, serverless architectures, and sustainable cloud practices, while security and compliance remain -paramount considerations. Organizations that successfully leverage these cloud capabilities while -maintaining robust security measures and embracing sustainable practices will be best positioned to thrive -in an increasingly digital world. +q emergency phone numbers (HWS: 1-315-781-3333) -REFERENCES +q minimum of 100 U.S. dollars for use in an emergency -[1] Businesswire, "IDC Spending Guide Sees Worldwide Digital Transformation Investments Reaching +q Sweater or jacket in event of bad weather -$3.4 Trillion in 2026," 2022. [Online]. Available: -https://www.businesswire.com/news/home/20221026005193/en/IDC-Spending-Guide-Sees- -Worldwide-Digital-Transformation-Investments-Reaching-%243.4-Trillion-in-2026 +q if taking prescription medicine bring prescription for future refills (translated into -[2] Noah Barsky, "Majority Of Digital Transformations Off Track, Executive Survey Finds—Here Are 3 +english) -Fixes," Forbes, 2023. [Online]. Available: -https://www.forbes.com/sites/noahbarsky/2023/10/02/people-problems-plague-digital- -transformation/ +q Photocopy of your passport, visa, etc. do not carry in the same place as your -[3] Arbab Khan, "Unlocking the Power of Digital Transformation and Automation," Rushford, 2024. +original documents. -[Online]. Available: https://rushford.ch/insights/unlocking-the-power-of-digital-transformation- -and-automation/ +29 -[4] George, "On-Premises vs. Cloud Data Center: Which Is Right for Your Business?" FS, 2022. + SECTION III: +*SAMPLE* ORIENTATION SCHEDULE *SAMPLE* -[Online]. Available: https://www.fs.com/blog/onpremises-vs-cloud-data-center-which-is-right- -for-your-business-5867.html +Hobart and William Smith Colleges +international Students orientation +“let Your adventure begin!” +august 22 -25, 2006 -[5] Neal Weinberg, "5 cloud market trends and how they will impact IT," CIO, 2024. [Online]. Available: +Tuesday, August 22 +8:30 a.m. –7:00 p.m. -https://www.cio.com/article/655693/7-cloud-market-trends-and-how-they-will-impact-it.html +Wednesday, August 23 +8:30 a.m. –12:00 p.m. -22 +arrival/Check-in +location: office of intercultural affairs +288 Pulteney Street +**bring all immigration paperwork - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +arrival/Check-in +location: office of intercultural affairs, +288 Pulteney Street +**bring all immigration paperwork - Online ISSN: 2054-0965 (Online) +HWS Student id Cards +location: registrar’s office - Print ISSN: 2054-0957 (Print) +Campus tour - Website: https://www.eajournals.org/ +Shopping @ Wal-mart & Wegmans Supermarket, +Van leaves from intercultural affairs parking lot - Publication of the European Centre for Research Training and Development -UK +12:30 p.m. - 1:30 p.m. -[6] Will Forrest et al., "Cloud’s trillion-dollar prize is up for grabs," McKinsey Digital, 2021. [Online]. +lunch with deans of the Colleges +location: blue room in Scandling Center -Available: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion- -dollar-prize-is-up-for-grabs +1:45 p.m. – 2:45 p.m. -[7] Precedence Research, "Artificial Intelligence (AI) Market Size, Share, and Trends 2025 to 2034," +overview of library Services +location: Warren Hunting Smith library -2025. [Online]. Available: https://www.precedenceresearch.com/artificial-intelligence-market -[8] Abel E. Edje, M.S. Abd Latiff and Weng Howe Chan, "IoT data analytic algorithms on edge-cloud +6:00 p.m. – 7:30 p.m. -infrastructure: A review," ScienceDirect, 2023. [Online]. Available: -https://www.sciencedirect.com/science/article/pii/S2352864823001530#:~:text=In%20addition% -2C%20future%20advances%20are,Comparison%20of%20previous%20research%20surveys.&te -xt=Comparison%20of%20middleware%20architectures. +dinner +location: intercultural affairs Center -[9] IvanDev, "The Rise of Serverless Architecture and Edge Computing," DEV, 2025. [Online]. +Thursday, August 24 +7:30 a.m. – 8:00 a.m. -Available: https://dev.to/ivansing/the-rise-of-serverless-architecture-and-edge-computing-2aab +8:15 a.m. 9:45 a.m. -[10] Globe Newswire, "Hybrid Cloud Market Projected to Hit USD 405.62 Billion by 2032| SNS +30 -Insider," 2025. [Online]. Available: https://www.globenewswire.com/news- -release/2025/03/21/3047113/0/en/Hybrid-Cloud-Market-Projected-to-Hit-USD-405-62-Billion- -by-2032-SNS-Insider.html +breakfast +location: Saga dining Hall in the blue room -[11] Thales, "Cloud Security in 2024: Addressing the Shifting Landscape," CSA, 2024. [Online]. +immigration regulations overview +location: Conference room, 2nd Floor, +intercultural affairs -Available: https://cloudsecurityalliance.org/blog/2024/06/27/cloud-security-in-2024-addressing- -the-shifting-landscape + 10:00 a.m. – 11:00 a.m. -[12] MAHESH NAWALE, "7 Key Takeaways From IBM's Cost of a Data Breach Report 2024," ZScaler, -2024. [Online]. Available: https://www.zscaler.com/blogs/product-insights/7-key-takeaways-ibm- -s-cost-data-breach-report-2024 +11:05 a.m. – 11:35 a.m. -[13] Andrii Stashko-Ziablietsov and Oleksandr Virga, "Cloud strategy: a 2024 guide," Avenga, 2024. +on Campus employment/Social Security +elaine Ferrara, business office +location: tba -[Online]. Available: https://www.avenga.com/magazine/cloud-strategy/ +benefits of the HWS Health insurance +ann brink, bursars office +location: tba -[14] Willie Devitt, "Cloud Technology Trends and Predictions for 2024," Eviden, 2024. [Online]. +12:00 p.m. – 1:30 p.m. -Available: https://eviden.com/publications/whitepapers/cloud-technology-trends-and-predictions- -for-2024/#:~:text=reputation%20and%20trustworthiness- -,At%20Eviden%2C%20we%20predict%20that%20in%202024%2C%20the%20focus%20on,com -pliance%20posture%20for%20our%20customers. +lunch with Chaplain lesley adams +location: Chaplain’s residence, 630 S. main St. -23 +1:45 p.m. – 2:15 p.m. + +money matters +beth turner, director of Financial aid +location: tba + +2:30 p.m. – 3:00 p.m. + +a Healthy You +location: Hubbs Health Center + +3:15 p.m. – 3:45 p.m. + +5:30 p.m. – 7:30 p.m. + +Friday , August 25 +10:00 a.m. – 11:00 a.m. + +Study basics +Center for teaching & learning +location: Center for teaching & learning +(Ctl) + +Welcome dinner +President and mrs. mark gearan & Family +location: President’s House – 690 South main St. + +guide to U.S. education & Culture +amy teel and doug reilly, Center for +global education +location: delancey House + +12:00 p.m. – 1:00 p.m. + +lunch +location: delancey House + +1:30 p.m. + +Join regular campus wide orientation schedule + +Welcome to Hobart and William Smith Colleges + +31 diff --git a/q103/random_k4/question.json b/q103/random_k4/question.json index 0550df92a5db60167f9fbf92006cfde8afac8623..1be07639b1332a0f367dcd764b58640c19b6d112 100644 --- a/q103/random_k4/question.json +++ b/q103/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "e639029d16094ea71d964e2fb953952b.pdf", - "web_0867a04cdab7a273.pdf", - "web_a39c0d7f5ff0dedb.pdf", - "web_5906edbb3f342646.pdf", - "web_660e5c500c425b81.pdf" + "389c76c12bae49339e801c93d7587859e248a538.pdf", + "web_ae1b8a373064ff5e.pdf", + "web_77f281bc58aa8364.pdf", + "web_2c8b6d1a0ca91c7b.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q103/random_k4/random_1.md b/q103/random_k4/random_1.md index 85e510cedc0e76a41806e14e4f5679216b834d6f..c230816f9b1098ca6662073190a692ee54d16a56 100644 --- a/q103/random_k4/random_1.md +++ b/q103/random_k4/random_1.md @@ -1,1480 +1,3444 @@ -Historic Architectural Resources Survey Report - -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 - -North Carolina Department of Transportation -Report Prepared by Nancy Van Dolsen - -November 1999 - - Historic Architectural Resources Survey Report - -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 - -November 1999 - -Principal Investigator (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation - -441-1/4,4 (cid:9) - -Barbara Church, Supervisor (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation - -Date - -//7/67/1. -Date - - TABLE OF CONTENTS - -Project Description (cid:9) - -Purpose of Survey and Report (cid:9) - -Methodology (cid:9) - -Summary Findings of the Survey (cid:9) - -Historic Contexts and Background Information (cid:9) - -Property Inventory and Evaluations - -William Jones Arney House (cid:9) - -Marshall Arney House (cid:9) - -Corpening, House (cid:9) - -Early Twentieth-Century Frame House (cid:9) - -Nineteenth-Century Log House (cid:9) - -Bibliography (cid:9) - -Appendix A: Correspondence with the State Historic Preservation Office (cid:9) - -LIST OF FIGURES - -Figure 1: Project Location Map (cid:9) - -Figure 2: Area of Potential Effects with Historic Properties Noted (cid:9) - -Figure 3: Tax Map, William Jones Arney House (cid:9) - -Figure 4: Site Plan, William Jones Arney House (cid:9) - -Figure 5: Tax Map, Marshall Arney House (cid:9) - -Figure 6: Site Plan, Marshall Arney House (cid:9) - -Page - - 1 - -1 - +| | | When Can | LLMs | Learn | to Reason | with | Weak | Supervision? | | +| --- | --- | -------- | ---- | ----- | --------- | ---- | ---- | ------------ | --- | +SalmanRahman12* JingyanShen2* AnnaMordvina2 HamidPalangi3 SaadiaGabriel1 PavelIzmailov2 +| | | ❖Projectpage: | | salmanrahman.net/rlvr-weak-supervision | | | | | | +| ----- | -------- | ------------- | ------------- | -------------------------------------- | --- | --------------------------------------------------- | --- | ---------------------- | --- | +| | | Abstract | | | | pabilitiesinlargelanguagemodels(Guoetal.,2025;Jaech | | | | +| | | | | | | etal.,2024;Teametal.,2025). | | Withonlybinaryfeedback | | +| Large | language | models | have achieved | signifi- | | | | | | +6202 rpA 02 ]GL.sc[ 1v47581.4062:viXra +cantreasoningimprovementsthroughreinforce- oncorrectness,RLVRhasenabledsubstantialgainsacross +diversereasoningtaskswithoutrequiringdensesupervision. +| ment | learning | with verifiable | rewards | (RLVR). | | | | | | +| ---- | -------- | --------------- | ------- | ------- | --- | --- | --- | --- | --- | +Yetasmodelcapabilitiesgrow,constructinghigh- However,recentfindingssuggesttheseimprovementsmay +qualityrewardsignalsbecomesincreasinglydif- bedrivenbyfactorsotherthantheintegrationofcorrectness +| | | | | | | signals. | SomestudiesreportthatRLVRsucceedsevenun- | | | +| ------- | ------ | ------------ | ------------- | ---- | --- | -------- | ---------------------------------------- | --- | --- | +| ficult, | making | it essential | to understand | when | | | | | | +RLVR can succeed under weaker forms of su- derextremeconditions: trainingonjustasingleexample +canyieldsignificantgains(Wangetal.,2025a),andrandom +| pervision. | | We conduct | a systematic | empirical | | | | | | +| ---------- | --- | ---------- | ------------ | --------- | --- | --- | --- | --- | --- | +studyacrossdiversemodelfamiliesandreasoning orincorrectrewardssometimesmatchground-truthperfor- +domains under three weak supervision settings: mance (Shao et al., 2025). Other work shows that proxy +signalssuchasself-certainty(Zhaoetal.,2025;Prabhudesai +| scarce | data, | noisy rewards, | and self-supervised | | | | | | | +| ------ | ----- | -------------- | ------------------- | --- | --- | --- | --- | --- | --- | +proxy rewards. We find that generalization is etal.,2025),entropyminimization(Agarwaletal.,2025), +majorityvoting(Zuoetal.,2025),orself-generatedtraining +governedbytrainingrewardsaturationdynamics: +models that generalize exhibit a prolonged pre- data(Huangetal.,2025)canreplaceverifiablerewards. +saturationphaseduringwhichtrainingrewardand +Furthermore,techniquesthatsucceedononemodelfamily +downstream performance climb together, while oftenfailonothers(Shaoetal.,2025),underreportedbase- +modelsthatsaturaterapidlymemorizeratherthan +linesmayinflateperceivedbenefits(Chandaketal.,2025), +learn. Weidentifyreasoningfaithfulness,defined and prolonged training with proxy rewards (i.e., reward +astheextenttowhichamodel’sintermediatesteps +| | | | | | | signals derived | from | model outputs without | ground-truth | +| --- | --- | --- | --- | --- | --- | --------------- | ---- | --------------------- | ------------ | +logicallysupportitsfinalanswer,asthepre-RL +verification)canleadtorewardhackingandperformance +propertythatpredictswhichregimeamodelfalls collapse(Shafayatetal.,2025). Thesemixedresultsleavea +into,whileoutputdiversityaloneisuninformative. +| | | | | | | fundamentalquestion: | | WhencanRLVRgeneralize1under | | +| --- | --- | --- | --- | --- | --- | -------------------- | --- | --------------------------- | --- | +Motivatedbythesefindings,wedisentanglethe weaksupervision,andwhatdeterminessuccessorfailure? +contributionsofcontinualpre-trainingandsuper- +visedfine-tuning,findingthatSFTonexplicitrea- UnderstandingwhenRLVRworksunderweaksupervision +mattersforpractice.Ground-truthverifiersareoftenlimited: +soningtracesisnecessaryforgeneralizationunder +weaksupervision,whilecontinualpre-trainingon labelsmaybenoisyorunavailable,andasmodelsbecome +domaindataamplifiestheeffect.Appliedtogether strongerthantheirsupervisors,alternativerewardsignals +becomenecessary(Burnsetal.,2023). +toLlama3.2-3B-Base,theseinterventionsenable +generalizationacrossallthreesettingswherethe +WeconductasystematicempiricalstudyofRLVRunder +basemodelpreviouslyfailed. weak supervision across two model families (Qwen and +| | | | | | | Llama), | and three | reasoning domains (MATH, | SCIENCE, | +| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------------- | ------------------------ | -------- | +| | | | | | | andGRAPH). | Ourworkisorganizedaroundthreequestions: | | | +1.Introduction +| | | | | | | • RQ1(WeakSupervision):DoesRLVRgeneralizeacross | | | | +| --- | --- | --- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | +Reinforcementlearningwithverifiablerewards(RLVR)has +modelfamiliesanddomainsunderscarcedata,noisyre- +emergedasapowerfulparadigmforimprovingreasoningca- +wards,andself-supervisedproxyrewards? +*Equal 1University • RQ2(ModelProperties): Whatpre-RLmodelproper- +| | contribution | | of | California, | Los An- | | | | | +| ----- | ------------ | --------------- | -------- | -------------- | ------- | --- | --- | --- | --- | +| | 2New | | 3Google. | | | | | | | +| geles | | York University | | Correspondence | | | | | | +1Throughout,weusegeneralizationtomeanimprovementon +| to: Salman | Rahman | , | | Jingyan | Shen | | | | | +| ---------- | ------ | --------------------- | --- | ------- | ---- | --- | --- | --- | --- | +downstreamevaluationbenchmarks,bothin-domainheld-outsets +. +andout-of-domaintransfer,followingRLtraining. +Preprint.April21,2026. 1 -4 - - 6 - -8 - - 15 +LLMReasoningwithWeakSupervision +tiesdeterminewhetheramodelgeneralizesunderweak 2.ExperimentalSetup +supervision? +| | | | | | | | We evaluate | the | following | model families: | | (1) Qwen2.5- | | +| -------------------- | --- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | ----------------------------------- | --------------- | --- | ------------ | --- | +| • RQ3(Intervention): | | | Howcanweenablegeneralization | | | | | | | | | | | +| | | | | | | | 1.5B/3B(Base): | | General-purposemodelspretrainedon18 | | | | | +inmodelsthatfailunderweaksupervision? +trilliontokens(Team,2024);(2)Qwen2.5-Math-1.5B/7B +Ourinvestigationuncoversthreefindings. First,general- (Math-specialized): BuiltuponQwen2.5withanadditional +izationunderweaksupervisionisgovernedbytraining 1trillionmath-relatedtokens(Yangetal.,2024);(3)Llama- +rewardsaturationdynamics. Modelsthatgeneralizeex- 3.2-3B/8B-Instruct(Instruction-tuned): Pretrainedon9 +hibitaprolongedpre-saturationphaseduringwhichtraining trilliontokensandalignedviaSFT,rejectionsampling,and +| | | | | | | | DPO(Dubeyetal.,2024). | | | WeusetheInstructvariantsfor | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | --------------------------- | --- | --- | --- | +rewardclimbssteadilyandthemodellearnstransferablerea- +soningpatterns;modelsthatfailsaturaterapidlyandentera Llamabecausethebasemodelsdonotreliablyfollowthe +post-saturationphasewherefurthertrainingyieldsdiminish- required formatfor on-policy rollouts. We revisit Llama- +ingreturns. Whichregimeamodelfallsintodependsonits Basein§4,whereSFThandlestheformat-followingissue. +| pretrainingpriors: | | modelswithstrongdomain-alignedpre- | | | | | | | | | | | | +| ------------------ | --- | ---------------------------------- | --- | --- | --- | --- | ------------------- | --- | ----------------------------- | --- | --- | --- | --- | +| | | | | | | | DomainsandDatasets. | | Weselectthreedomainswithvary- | | | | | +training(QwenonMATHandSCIENCE)sustainextended +| | | | | | | | inglevelsofpretrainingexposure: | | | | MATH(highexposure), | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | ------------------- | --- | --- | +pre-saturationphasesandgeneralizeunderscarcedata,noisy +SCIENCE(moderatecoverage)andGRAPHtasks(underrep- +rewards,andself-supervisedproxyrewards,whilemodels +| | | | | | | | resentedintypicalpretrainingcorpora). | | | | WeuseSkywork- | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | ------------- | --- | --- | +withoutsuchpriors(Llamaacrossalldomains,andQwenon +OR1(Heetal.,2025a)forMATH,SCPdatasets(Liuetal., +GRAPH)saturaterapidlyandfailtogeneralizeevenunder +| | | | | | | | 2025a; Lu | et al., | 2025) spanning | physics, | | chemistry, | and | +| ------------------- | --- | -------------------------------- | --- | --- | --- | --- | --------- | ------- | -------------- | -------- | --- | ---------- | --- | +| moderatelabelnoise. | | Wetreatthemodel-familycontrastas | | | | | | | | | | | | +biologyforSCIENCE,andtasksfromReasoningGym(Sto- +aproxyforpretraining-priorstrengthratherthananintrin- +janovskietal.,2025)involvingdiscretealgorithmicreason- +sicpropertyofeitherfamily,areadingthat§4confirmsby +| | | | | | | | ingforGRAPH. | ForMathandScience,weusethe1.5B/3B | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --------------------------------- | --- | --- | --- | --- | --- | +showingthatcontinualpre-trainingonmathdatatransforms +modelsasourprimaryexperimentsandadditionallyevalu- +Llama’sRLbehaviortoresembleQwen’s. +ate7B/8Bmodelstoverifythatourfindingsholdatlarger +Second,reasoningfaithfulness,notoutputdiversity,dis- scale. For GRAPH, we only use the 7B/8B variants be- +tinguishesmodelsthatgeneralizefrommodelsthatmem- cause the smaller models achieve solve@16 = 0, leaving +orize. Anaturalhypothesisforrapidsaturationisthatfail- noinformativesignalforRL.Moredetailsareprovidedin +| ing models | lack | exploratory | | capacity. | We find | the oppo- | AppendixB. | | | | | | | +| ---------- | ---- | ----------- | --- | --------- | ------- | --------- | ---------- | --- | --- | --- | --- | --- | --- | +site: Llamamodelsreachperfecttrainingrewardfasterthan +| | | | | | | | Model-AwareDataFiltering. | | | Toensureinformativetrain- | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | ------------------------- | --- | --- | --- | +Qwenandmaintainhigheroutputdiversitythroughouttrain- +ingsignals,weimplementmodel-specificdifficultyfiltering. +| ing, yet | they generalize | | poorly. | The | missing | property is | | | | | | | | +| -------- | --------------- | --- | ------- | --- | ------- | ----------- | --- | --- | --- | --- | --- | --- | --- | +Foreachproblem,wesample16responsesandcountcorrect +reasoningfaithfulness,definedbywhetheramodel’sinter- +| | | | | | | | solutions | (solve@16 | ∈ [0,16]). | We | retain | only problems | | +| ------------------------------------------- | --- | --- | --- | --- | --- | ---------- | --------- | --------- | ---------- | --- | ------ | ------------- | --- | +| mediatestepslogicallysupportitsfinalanswer. | | | | | | Modelsthat | | | | | | | | +wheresolve@16∈[1,15],effectivelydiscardinginstances +| saturate | rapidly | produce | correct | answers | through | reason- | | | | | | | | +| -------- | ------- | ------- | ------- | ------- | ------- | ------- | --- | --- | --- | --- | --- | --- | --- | +thatareeithertrivialorintractableforthemodel,stratified +ingchainsthatdonotjustifythem,memorizingratherthan +| | | | | | | | equally | across difficulty | levels | (details | in | Appendix | B.2). | +| --------- | --------- | --- | ---------------- | --- | ---- | ---------- | ------- | ----------------- | ------ | -------- | --- | -------- | ----- | +| learning. | Diversity | is | only informative | | when | considered | | | | | | | | +Thisfilteredsetservesasthecandidatepoolforallweak +jointlywithfaithfulness. +supervisionsettingsstudiedinthiswork;wedescribehow +Third,SFTonexplicitreasoningtracesisnecessaryfor trainingdataisconstructedfromthispoolforallsettings +| generalizationunderweaksupervision, | | | | | andcontinual | | in§3. | | | | | | | +| ---------------------------------------------- | --------- | --- | --- | ------- | ------------ | ---------- | ---------------------- | --- | ----------------------- | --- | --- | --- | --- | +| pre-training | amplifies | | the | effect. | We run a | controlled | | | | | | | | +| | | | | | | | TrainingConfiguration. | | WeuseGRPO(GroupRelative | | | | | +| comparisonthatdisentanglesthetwointerventions, | | | | | | train- | | | | | | | | +PolicyOptimization)asourRLalgorithm(Shaoetal.,2024). +| ing Llama3.2-3B | | Base, | a continually | | pre-trained | variant | | | | | | | | +| --------------- | --- | ----- | ------------- | --- | ----------- | ------- | --- | --- | --- | --- | --- | --- | --- | +ForeachqueryqsampledfromtrainingdatasetsD,agroup +| (CPT, ours), | and | Instruct, | each | with | either Thinking | SFT | | | | | | | | +| --------------------------------------------------- | --- | --------- | ---- | ---- | --------------- | --- | ----------------------- | ----------- | ---- | ----------------------- | --- | ------------- | --- | +| | | | | | | | ofindividualresponses{o | | }G | aresampledfromthepolicy | | | | +| (explicitreasoningtraces)orNon-ThinkingSFT(finalso- | | | | | | | | | i | i=1 | | | | +| | | | | | | | π before | the update. | GRPO | maximizes | | the following | | +θold +| lutionsonly). | ThinkingSFTisnecessary: | | | | itimprovesrea- | | | | | | | | | +| ------------- | ----------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | +objective: +soningfaithfulness,extendsthepre-saturationphase,and (cid:104) +(θ)=E +enablesgeneralizationunderallthreeweaksupervisionset- J GRPO (q,a)∼D,{oi}G +i=1 ∼πθold (·|q) +| | | | | | | | G | |oi| | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ---------- | ---------- | ------------- | ---------- | ----------- | --- | ------- | +| tings,whileNon-ThinkingSFTonthesamepromptsfails. | | | | | | | 1 (cid:88) | 1 (cid:88) | | | | | | +| | | | | | | | | | min (cid:0) ρ | Aˆ ,clip(ρ | ,1−ϵ,1+ϵ)Aˆ | | (cid:1) | +Continualpre-trainingisamultiplierratherthanasubstitute. i,t i i,t i +| | | | | | | | G | |o | | | | | | | +| ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | ------ | --------- | --- | --- | +| | | | | | | | i=1 | i t=1 | | | | | | +| CPTcombinedwithThinkingSFTproducesthestrongest | | | | | | | | | | | (cid:105) | | | +| | | | | | | | | | −βD | (π ||π | ) , | | | +generalization, recovering performance in settings where KL θ ref +Llamapreviouslyfailed. +πθ(oi,t|q,oi,0 +1 1 sat +(effectivepre-saturationlearning)whileredmarkrapidsaturationt(8) < 100. Thelarge-smallgapatsaturationstepsG(n1,8) and +sat sat,in +G(n1,8) aregenerallysmall.Resultsonmorebenchmarksandpass@kmetricsarereportedinTable3-7inAppendix. +sat,ood +In-domainBenchmarks OODBenchmark +Model t(8) +sat +∆( +s +8 +a +) t -i - -T - +∆∗ +p +( +o +8 +s +) t -e -e -h -S - -: -§ +∆( +s +8 +a +) +t +∆∗ +p +( +o +8 +s +) +t +Gsat,in ∆( +s +8 +a +) +t +∆∗ +p +( +o +8 +s +) +t +Gsat,ood +TrainingDomain:Math MATH500 AMC G(2048,8) SCP-Hard G(2048,8) +sat,in sat,ood +Qwen2.5-Math-1.5B 302 29.7 1.5 18.7 0.6 -1.1 10.5 2.1 2.4 +Qwen2.5-1.5B 170 32.1 0.9 12.7 3.3 -0.5 7.0 0.3 -0.4 +Llama3.2-3B-Instruct 55 10.8 -1.9 8.8 -2.1 -0.9 3.9 0.0 1.5 +TrainingDomain:Science SCP-Hard GPQA-Diamond G(2048,8) MATH500 G(2048,8) +sat,in sat,ood +Qwen2.5-Math-1.5B 268 14.5 1.1 16.9 1.6 1.1 25.3 0.8 1.1 +Qwen2.5-1.5B 161 6.4 0.2 13.3 1.7 1.8 32.3 2.1 1.2 +Llama3.2-3B-Instruct 61 1.8 1.7 11.9 3.0 5.1 7.3 2.2 0.6 +TrainingDomain:Graph QuantumLock LargestIsland +G(n1,8)† +MATH500 +G(n1,8)† +sat,in sat,ood +Qwen2.5-Math-7B 150 8.3 4.9 19.8 1.9 -1.8 21.0 2.1 -3.7 +Llama3.1-8B-Instruct 29 10.1 7.1 1.8 1.0 3.0† 9.1 3.8 0.0† +acrossdifficultylevelsdefinedin§2. ForN < 64,were- tional gain after saturation, defined as ∆∗(n)(M) := +post +peatpromptsuniformlytoreachbatchsize64(e.g.,N =8 +max +M(n)(t)−M(n)(cid:0) t(n)(cid:1) +. Values near zero +implies8repeats). +t∈[t( +s +n +at +),T] sat +indicatenegligiblepost-saturationgains. +Tostudytrainingdynamics,weleveragerewardsaturationto +• Large-small gap +G(n′,n)(M): +we define this gap as +distinguishperiodswherethepolicyimprovesonthetrain- sat +M(n′)(t(n))−M(n)(t(n))forn′ > n,whichcompares +ingdatasetfromthosewhereitplateaus. Intuitively,once sat sat +performancebetweenlarger(n′)andsmaller(n)datasets +trainingrewardsaturates,furtherupdatesyieldlittlenewsig- +(cid:104) (cid:105) atthesaturationstepofthesmallerrun. Atthesmaller +nal.Wedefiner¯ :=E 1 (cid:80)G r as +t q∼D,{oi}G i=1 ∼πold(·|q) G i=1 i run’ssaturationstep,howmuchbetterdoesthelargerrun +theexpectedtrainingrewardatupdatestept∈{1,...,T}, perform? Largerpositivevaluesindicatesubstantialben- +andletr¯ max :=max 1≤t≤T r¯ t bethemaximumrewardob- efitfrommoredata;valuesnearzerosuggestlimitedad- +servedduringtraining. Weidentifytraininghassaturated vantagefromincreasingdatasetsize. WedenoteG(n1,8) +sat,in +oncetherewardisclosetothismaximum,anddefinethe +astheaveragegapoverthein-domainbenchmarks,and +saturationstepastheearliestupdatewherethisoccurs: G(n1,8) astheaveragegapoverOODbenchmarks. +sat,ood +(cid:110) +t sat :=inf t∈{1,...,T eff }:r¯ t ≥ϵ max r¯ max }. Pre-saturationphasedominatessmall-samplelearning, +and its length predicts generalization. Table 1 sum- +Weuseϵ =0.99andsetT =T−50,i.e.,wesearchfor +max eff marizes the proposed metrics across model families and +t onlyuptothefirstT updatestoavoidboundaryeffects +sat eff training domains when training on 8 examples. Results +neartheendoftraining.Wedefinethepre-saturationphase +onmorebenchmarksandpass@kmetricsareprovidedin +asallstepst∈{1,...,t −1}andpost-saturationphase +sat Appendix C.2 and Tables 3-7. All model-domain pairs +asallstepst∈{min(t sat ,T),...,T}. showclearlypositive∆(8)forallmetrics(i.e.,bothavg@16 +sat +To quantify data efficiency, we introduce three metrics. andpass@k,k ∈ {4,8,16})acrossin-domainandout-of- +LetM(n)(t)denoteanevaluationmetric(e.g.,avg@16on domain benchmarks, indicating that as few as 8 training +MATH-500)attrainingsteptfortrainingwithnsamples, examples can trigger measurable learning during the pre- +andt(n)bethecorrespondingsaturationstep. saturation phase. Neither G(2048,8) nor G(2048,8) is sig- +sat sat,in sat,out +nificantly greater than zero on 7 out of 8 model-domain +• Pre-saturationgain∆(n)(M): performancegainfrom +sat pairs,indicatingthatthepre-saturationimprovementsare +initializationtosaturationas∆(n)(M):=M(n)(cid:0) t(n)(cid:1) +− oftencomparabletothoseobtainedwithlargertrainingsets. +sat sat +M(n)(0). Largerpositivevaluesindicateeffectivelearn- Thissuggeststhatearlylearningisnotstronglydata-limited. +ingbeforesaturation. Incontrast,thepost-saturationresidual∆∗(8) istypically +post +• Post-saturation residual ∆∗(n)(M): maximum addi- smallerthan∆(8),indicatingdiminishingreturnsoncethe +post sat +4 -9 +LLMReasoningwithWeakSupervision +Qwen2.5-Math-7B Llama-3.2-3B-Instruct gestthatevenformodelswithstrongmathematicalpriors, +| | | Graph | | | | Math | | | | | | | | | +| --- | --- | ----- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +thelackofdomain-specificpre-trainingacceleratessatura- +| | draweR gniniarT 1.0 | | | draweR gniniarT 0.75 | | | | | | | | | | | +| --- | ------------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.8 +| | | | | 0.60 | | | | tionandnecessitateshigherdatavolumetodrivelearning. | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | --- | --- | --- | +0.6 +| | | | | 0.45 | | | | Wefurtherprovideillustrationsfor7Band8Bmodelson | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | --- | --- | --- | +0.4 +| | | | | 0.30 | | | | MATHandSCIENCEdomainsinAppendixC.3. | | | | | | | +| --- | --- | --- | --- | ---- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +| | 0.2 | | | 0.15 | | | | | | | | | | | +0.0 +)%( kcoL mutnauQ Extendedpre-saturationenablesout-of-domaintransfer. +| | 40 | | | )%( 005-HTAM 52 | | | | | | | | | | | +| --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Positive∆(8) +| | 30 | | | 48 | | | | | valuesinTable1indicatethatthereasoning | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------------- | --- | -------------- | --- | ----- | -------- | +| | 20 | | | | | | | | sat | | | | | | +| | | | | 44 | | | | patterns | learned during | the | pre-saturation | | phase | transfer | +10 +| | | | | 40 | | | | acrossdomains,particularlyforQwenmodels. | | | | | Withonly | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | -------- | --- | 0 -1 -2 +36 +| | | | | 28 | | | | 8samples,Qwen2.5-1.5BtrainedonMATHachievescon- | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | +)%( 005-HTAM 78 +| | | | | 24 | | | | sistent gains | on the | out-of-domain | | SCIENCE | benchmark | | +| --- | --- | --- | --- | ------- | --- | --- | --- | ------------- | ------ | ------------- | --- | ------- | --------- | --- | +| | 72 | | | )%( CMA | | | | | | | | | | | +| | 66 | | | 20 | | | | | | | | | | | +(SCP-Hard),whileQwen2.5-Math-7BtrainedonGRAPH +| | 60 | | | 16 | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +improvesout-of-domainMATH-500performanceby21.0% +| | 54 | | | 12 | | | | | | | | | | | +| --- | --- | --- | --- | ----- | --- | --- | --- | --------- | ------------ | ----- | ------ | ---- | ------- | ------- | +| | | | | | | | | (Fig. 1). | In contrast, | Llama | models | show | limited | out-of- | +| | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 | | | | | | | | +Training Steps Training Steps domain transfer even when in-domain performance im- +| | | =0 | | =0.3 | | =0.7 | | | | | | | | | +| --- | --- | --- | --- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +proves;theirgainsremainlocalizedtothespecifictraining +| | | =0.1 | | =0.5 | | =0.9 | | | | | | | | | +| --- | --- | ---- | --- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +distribution. +Figure2.Effectofrewardlabelcorruptionontrainingdynam- +icsandgeneralization.γdenotesthefractionoftrainingprompts +Takeaway: +| withcorruptedlabels,rangingfromclean(γ | | | | | | = 0)tomostlyin- | | | | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +(1)RLVRcangeneralizefromasfewas8samples +| correct(γ | | = 0.9). | | | | | | | | | | | | | +| --------- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ForQwenonGRAPHandLlamaonMATH, +generalizationdegradeatγ ≥ 0.5. when models remain in an extended pre-saturation +ForLlama,trainingreward +curvesstaycloseacrossallγ,suggestingoverfittingtonoise. phase,whereasrapidlysaturatingmodelsrequiresub- +stantiallymoredata.(2)Whetherscarce-datalearning +8-samplerunreachest(8). succeedsismodel-anddomain-dependent,reflecting +| | | | sat | | | | | theinfluenceofpretrainingpriors. | | | | (3)Inthelow-data | | | +| ---- | ------- | --- | -------- | ------ | ------ | ------------ | --- | -------------------------------- | --- | --- | --- | ---------------- | --- | --- | +| Fig. | 1 shows | the | training | curves | across | data scales. | The | | | | | | | | +regime,Llamamodelscanachieveperfecttrainingre- +lengthofthepre-saturationphaseistheprimarydeterminant wardsmuchfasterthanQwenbyrapidlymemorizing +ofwhetheramodelcangeneralize. With8trainingsamples, trainingexamplesbutachievelittlemeaningfultask +| Qwen2.5-Math-1.5B | | | on | MATH increases | | reward | steadily | | | | | | | | +| ----------------- | --- | --- | --- | -------------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | --- | +learning. +forover300steps;thissustainedascentallowsthemodelto +extractgeneralizablereasoningpatternsthattransfertoheld- +| out | evaluation | benchmarks | | such | as MATH-500 | and | SCP- | 3.2.NoisyRewards | | | | | | | +| --- | ---------- | ---------- | --- | ---- | ----------- | --- | ---- | ---------------- | --- | --- | --- | --- | --- | --- | +Hard. Awithin-familycomparisonisolatesthepretraining +effect: Qwen2.5-Math-1.5B,whichsharesarchitecturewith Whenground-truthverifiersareavailablebutimperfect,re- +| | | | | | | | | wardlabelsmaycontainerrors. | | | ToevaluateRLVRrobust- | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --------------------- | --- | --- | --- | +Qwen2.5-1.5Bbuthasadditionalmath-specificpretraining, +nesstosuchnoisysupervision,wevarythefractionofincor- +saturatesmoreslowlyandtransfersfurther(Table1). +| | | | | | | | | rectlabelsγ | byrandomlyreplacingground-truthanswers | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | -------------------------------------- | --- | --- | --- | --- | --- | +Figs. 13, 14, and 15 (Appendix C.1) show the full range with the most frequent incorrect answer produced by the +| N | ∈ {8,32,64,512,2048} | | | across | MATH, | SCIENCE, | and | | | | | | | | +| ---------------------------------------- | -------------------- | -------- | ------ | ------- | ----- | ------------ | --- | ---------------------------------- | ------------ | ---------- | --- | --------------- | ------ | --- | +| | | | | | | | | modelitself(detailsinAppendixD.1). | | | | Unlessotherwise | | | +| GRAPH. | | For Qwen | models | on MATH | | and SCIENCE, | in- | | | | | | | | +| | | | | | | | | noted,experimentsuseN | | =2048. | | | | | +| domainperformanceisnearlyindependentofN. | | | | | | ForLlama | | | | | | | | | +| | | | | | | | | RLVR | demonstrates | robustness | | to reward | noise, | but | +acrossalldomains,andforQwenonGRAPH,differentN +| | | | | | | | | generalizationvariesacrossmodels. | | | | Fig.2andAppendix | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ---------------- | --- | --- | +producesvisiblydifferentdynamicsonsomeoftheevals, +Fig.26summarizeperformanceacrosssevenmodel–domain +withsmallerdatasetssaturatingearlierandatlowerdown- +streamperformance. pairsundervaryingγ. Atγ ≤0.3,testperformanceacross +| | | | | | | | | mostsettingsremainsclosetothecleanrewards(γ | | | | | | =0),in- | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | --- | ------- | +Modelswithoutdomain-alignedpriorssaturaterapidly dicating robustness to moderate label noise. On MATH +| andfailtogeneralize. | | | Incontrast,Llamamodelsacrossall | | | | | | | | | | | | +| -------------------- | --- | --- | ------------------------------- | --- | --- | --- | --- | ------------ | ---- | ------ | -------- | ----- | ----- | ---- | +| | | | | | | | | and SCIENCE, | Qwen | models | maintain | gains | under | sub- | +domains,andQwenonGRAPH(Fig.1)exhibitcleardepen- stantial corruption (up to γ = 0.7). In contrast, Qwen +| denceondatascale. | | | ForLlama,trainingon8samplesleads | | | | | | | | | | | | +| ----------------- | --- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +onGRAPHandLlamaonMATHandSCIENCEdegradeat +torapidsaturation,witht(8)occurringwithinthefirst100: +| | | | | | | | | γ ≥ 0.5. | Higherγ | leadstoconsistentlylowertrainingre- | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ------- | ----------------------------------- | --- | --- | --- | --- | +sat +itmaximizesthetrainingrewardmuchfasterthantheQwen wardsthroughouttraining,butforLlamaonMATH,training +models. Thesemodelsrequirelargerdatasets(N ≥512)to rewardcurvesremainnearlyidenticalacrossallγ despite +achievemeaningfulgeneralization(detailsinAppendixC.1 severecorruption,indicatingLlamafitsincorrectanswers +| Fig.13andFig.14). | | | TheresultsintheGRAPHdomainsug- | | | | | | | | | | | | +| ----------------- | --- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | 5 -8 -. + +LLMReasoningwithWeakSupervision +| Qwen2.5-3B | Llama-3.2-3B-Instruct | | | | | | +| ------------------- | --------------------- | --- | --- | --- | --- | --- | +| Science | Science | | | | | | +| draweR gniniarT 1.0 | 1.0 | | | | | | +| 0.8 | 0.8 | | | | | | +| 0.6 | 0.6 | | | | | | +0.4 +0.4 +0.2 +0.2 +)%( 005-HTAM +| 65 | 48 | | | | | | +| --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | +| | 40 | Figure4.Evolutionofsemanticdiversityduring8-sampletrain- | | | | | +60 +| 55 | 32 | ingonMATH.Llamashowssignificantlyhigherpost-saturation | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | +| 50 | 24 | | | | | | +diversitythanQwen,albeitwithlowerperformanceoutcomes. +| 45 | 16 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | +40 +)%( draH-PCS 40 SCIENCE)showimprovementwithmajorityvoting,while +20 +| 32 | | othermodelsfailentirely. | | ForQwen2.5-3Bon | | SCIENCE, | +| --- | --- | ----------------------------------------------------- | --- | --------------- | --- | -------- | +| 24 | 15 | | | | | | +| | 10 | majorityvotingyieldstemporarygainsbeforecollapseafter | | | | | +16 +| 8 | 5 | | | | | | +| --- | --- | --------- | --------------------------------------- | --- | --- | --- | +| | | 500steps, | asthepolicyconvergestowardasingleoutput | | | | +| 0 | 0 | | | | | | +0 150300450600750 0 150300450600750 tomaximizeagreement. Self-certaintyrewardsleadtoper- +| Training Steps | Training Steps | | | | | | +| -------------- | -------------- | --- | --- | --- | --- | --- | +RLVR Majority Vote Self Certainty formancecollapseacrossallsettings. Theseresultsshow +thatcurrentself-supervisedproxyrewardsareinsufficient +| | | toreplaceverifiablefeedbackinmostsettings. | | | | (detailsin | +| --- | --- | ------------------------------------------ | --- | --- | --- | ---------- | +Figure3.Comparisonofrewardvariants(RLVR,self-certainty, +AppendixD.2andFig.27). +majorityvote)with1024trainingsamples.Proxyrewardswith- +outverifiersexhibitfailuremodesunderprolongedtraining:train- +ingcollapse(self-certainty)andrewardspikesfollowedbyperfor- Takeaway: Self-supervisedproxyrewardssucceed +mancedrops(majorityvote)(moreresultsareinAppendixD.2). onlyformath-specializedmodels(Qwen-Mathunder +| | | majorityvoting). | Othermodelsexhibitthesamepat- | | | | +| --- | --- | ---------------- | ----------------------------- | --- | --- | --- | +moreeasily. Wealsoobservethatmodel-domainpairswith ternweobservedin§3.1and§3.2: fastersaturation +faster saturation (§3.1) are generally less robust to label andweakerpretrainingpriorscoincidewithbrittle- +noise,aconnectionwedevelopin§3.4and§4. ness. Under prolonged training, the failure mode +| | | isrewardhacking: | | policiesconvergetowardoutputs | | | +| --- | --- | ---------------- | --- | ----------------------------- | --- | --- | +Takeaway: Robustnesstolabelnoisevariessharply thatmaximizetheproxywithoutcorrespondingdown- +| acrossmodel-domainpairs:QwenonMATHandSCI- | | streamgains. | | | | | +| ----------------------------------------- | --- | ------------ | --- | --- | --- | --- | +ENCE toleratesupto70%corruption, whileLlama +andQwenonGRAPHdegradeat50%.Model-domain +pairsthatsaturatefasterundercleanrewardsareless 3.4.WhyDoModelsFailUnderWeakSupervision? +robust,andLlamafitscorruptedlabelsnearlyasfast +| | | Theresultsin§3.1–§3.3showaconsistentpattern: | | | | models | +| --- | --- | -------------------------------------------- | --- | --- | --- | ------ | +ascleanones—evidencethatrapidsaturationreflects +| | | withstrongdomain-alignedpretraining(Qwenon | | | | MATH | +| --- | --- | ------------------------------------------ | --- | --- | --- | ---- | +memorizationcapacityratherthanlearningefficiency. +| | | and SCIENCE) | generalize | under weak | supervision, | while | +| --- | --- | -------------------------------- | ------------------- | ---------- | ------------ | ---------- | +| | | thosewithout(Llamaacrossdomains, | | | Qwenon | GRAPH) | +| | | fail. A | natural hypothesis, | motivated | by prior | work link- | +3.3.Self-SupervisedProxyRewards +ingdiminishedexploratorycapacitytorapidpolicysatura- +Whenground-truthverifiersareentirelyunavailable,models tion(Cuietal.,2025),isthatfailingmodelsproduceless +mustrelyonalternativerewardsignals(Burnsetal.,2023; diverse outputs. To test this, we analyze model behavior +Rahmanetal.,2025;Bowmanetal.,2022). Recentwork alongtwocomplementaryaxes: responsediversityandrea- +hasproposedself-supervisedproxyrewardsderivedfrom soningfaithfulness. Formaldefinitionsandimplementation +model outputs, but whether these approaches work well detailsareprovidedinAppendixF. +acrossmodelfamiliesandtaskdomainsremainsunexplored. +| | | To quantify | response | diversity, we | quantify semantic | di- | +| ------------------------- | ------------------------- | ----------- | -------- | ------------- | ----------------- | --- | +| Weevaluatetwosuchrewards: | self-certainty(Zhaoetal., | | | | | | +versitytocharacterizemeaningfulpatternsinthemodel’s +2025)andmajorityvote(Zuoetal.,2025)(implementation +reasoningratherthansurface-levelvariation(Farquharetal., +detailsinAppendixD.2). +2024;Lietal.,2025).Wemeasurediversityonthe8-sample +Proxyrewardstriggerrewardhackingandpolicycol- subsetoftheMATH,SCIENCEandGRAPHtrainingdatasets, +lapse. WhileRLVRtoleratesmoderatelabelnoiseinsome aswellasontheMATH-500evaluationdataset,overase- +model-domainpairs(§3.2),Fig.3showsthatfullyreplacing lectionofpromptsatvariousstepsthroughouttraining. For +verifiablefeedbackwithself-supervisedproxysignalsintro- each prompt, we cluster model responses using pairwise +ducesseverefailuresunderprolongedtraining. Onlymath- similarity judgments from an LLM judge and define the +specialized models (Qwen2.5-Math-1.5B on MATH and diversityscoreastheShannondiversityindexoverthere- 6 +LLMReasoningwithWeakSupervision +| | | | | | | | faithfulness. | Fig. 5 (right) | reports | faithful | diversity: | di- | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | -------------- | ------- | -------- | ---------- | --------- | +| | | | | | | | versitycomputedonlyoverfaithfulresponses. | | | | | Thisjoint | +measurerevealsaconsistentpatternacrossallthreedomains. +| | | | | | | | On MATH,Llama’sapparentdiversityadvantage(Fig. | | | | | 4) | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | +disappears—mostdiverseresponsesareunfaithful,andthe +| | | | | | | | faithfulsubsetisnarrow. | | OnSCIENCE,alignedproportions | | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | ---------------------------- | --- | --- | --- | +areuniformlyhighacrossmodels,maskingrealdifferences +inreasoningquality;faithfuldiversityseparatesthem,with +Qwen-Mathmaintainingthehighestvaluesthroughouttrain- +| | | | | | | | ing. OnGRAPH,Qwen-MathandLlamashowcomparable | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | +alignedproportions,butQwen-Mathsustainshigherfaithful +| | | | | | | | diversity. | Ineverycase,themodelthatgeneralizesbestin | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------------------------------------- | --- | --- | --- | --- | +§3.1istheoneexploringthewidestrangeoffaithfulrea- +Figure5.Evolutionofreasoningfaithfulness(oncorrectsam- soningpaths—nottheonewiththehighestrawdiversity, +ples)andfaithfuldiversityonmodelsthroughoutRLusing8 nor the one with the highest aligned proportion. Raw di- +samplesfromavarietyofdatasets.LlamamodelsintheMATH +versityoverstatesexploratorycapacity;alignedproportion +domainexhibitsignificantlylowerfaithfulnesscomparedtoQwen. +saturatesoneasierdomains;onlytheirintersectionpredicts +generalization. +| sultingclusters. | | SeeFigure31forthejudgemodelprompt. | | | | | | | | | | | +| ------------------------------------------- | --- | ---------------------------------- | --- | --- | --- | ----- | --------- | --- | --- | --- | --- | --- | +| Highdiversitydoesnotpreventrapidsaturation. | | | | | | Fig.4 | Takeaway: | | | | | | +reportstheevolutionofdiversityscoresformodelstrained +| | | | | | | | Low reasoning | faithfulness, | | not low | diversity, | ex- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------- | --- | ------- | ---------- | --- | +on 8 samples from the MATH training dataset, computed plains why models fail under weak supervision: +on the corresponding training set. Llama reaches reward rapidlysaturatingmodelsmemorizeanswersrather +saturation earlier and retains higher diversity than Qwen, than acquire transferable reasoning. Raw diversity +theoppositeofwhattheexploration-saturationhypothesis metricsaremisleading—Llamaexhibitshigherout- +predicts. DiversitycomputedontheMATH-500evaluation +| | | | | | | | put diversity | than Qwen | while | generalizing | | worse. | +| --- | --- | --- | --- | --- | --- | --- | ------------- | --------- | ----- | ------------ | --- | ------ | +datasetispresentedintheappendix(Fig.30). Diversitybecomesinformativeonlywhencomputed +overfaithfulresponses. +Sincediversityalonedoesnotexplainfailureunderweak +| supervision, | we | investigate | the faithfulness | | of a model’s | | | | | | | | +| ------------ | --------------------------------------- | ----------- | ---------------- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | +| reasoning. | Inspiredbypriorwork(Bakeretal.,2025),we | | | | | | | | | | | | +definearesponseasfaithfulifitsreasoningtracecontains Insummary,§3showsthatthesurprisingcapabilitiesoften +the information needed to justify the final answer and is attributedtoRLVR,suchaslearningfromscarcedata,tol- +logicallyconsistentwithit. Atagiventrainingstepandfor eratingnoisyrewards,succeedingwithoutverification,are +notuniversalbutdependonpre-RLreasoningfaithfulness. +agivenprompt,wecategorizeeachpolicyrolloutasaligned, +partiallyaligned,ormisalignedbasedonrubricsprovided §4takesupthenaturalquestion: canpre-RLinterventions +to an LLM-as-a-judge (see prompt in Fig. 32). We then targetingfaithfulnessextendthepre-saturationphaseand +computethepolicyfaithfulnessrateF (l)asthefractionof recovergeneralizationunderweaksupervision? +π +| responsesassignedtolabell. | | | AppendixFoutlinesresults | | | | | | | | | | +| -------------------------- | --- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +forinter-modelagreementonalignmentcategorizationto 4.ImprovingRLVRUnderWeakSupervision +evaluatethereliabilityofourLLM-as-a-judge. +viaPre-RLTraining +Modelswithrapidsaturationexhibitlowreasoningfaith- +Section3showedthatrapidsaturationandlowreasoning +fulness. Fig.5(left)showsthefractionofcorrectresponses +| | | | | | | | faithfulnessarelinked: | modelsthatgeneralizepoorlyunder | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------- | ------------------------------- | --- | --- | --- | --- | +thatarealignedoverRLtrainingacrossmodelsanddomains +weaksupervisionproducecorrectanswersthroughreason- +| studiedin§3.1. | | Onthe | MATH domain, | theLlamamodel | | | | | | | | | +| --------------------------------------- | ----- | --------- | -------------------------------- | ------------- | ------------ | --- | ------------------------------------------------------- | ------------ | --------------- | ---------------------- | ---- | ---------- | +| | | | | | | | ingthatdoesnotsupportthem. | | | Thisraisesacausalques- | | | +| shows much | lower | reasoning | faithfulness | during | training | | | | | | | | +| | | | | | | | tion. Iffaithfulnessdrivesthepre-saturationphase,andthe | | | | | | +| thantheQwenmodels. | | | ThisindicatesthatLlama’srapidre- | | | | | | | | | | +| | | | | | | | pre-saturation | phase drives | generalization, | | then | instilling | +| wardgainsdonotreflectimprovedreasoning: | | | | | asubstantial | | | | | | | | +faithfulnessbeforeRLshouldextendthephaseandrecover +fractionofcorrectanswersarememorized,withreasoning +| | | | | | | | generalization. | Wetestthisbyrunningacontrolledcom- | | | | | +| ----------- | ------ | ------- | ---------- | ----- | -------- | --- | --------------- | ---------------------------------- | --- | --- | --- | --- | +| traces that | do not | support | them. Fig. | 33 in | Appendix | F | | | | | | | +parisonofpre-RLinterventionsonLlama3.2-3B,themodel +| includes | additional | faithfulness | results | on these | domains, | | | | | | | | +| -------- | ---------- | ------------ | ------- | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +coveringproportionalignedandproportionmisalignedon thatfailedmostconsistentlyin§3. +correct,incorrectandallresponses. Westudytwoaxesofpre-RLtraining. Thefirstiscontinual +Reasoning diversity should be considered jointly with pre-training(CPT),extendedtrainingondomain-specific +7 + +LLMReasoningwithWeakSupervision +| | draweR gniniarT | | | )%( 005-HTAM | | | 32 | | | | | | +| ------ | --------------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | | | | | 56 | | | | | )%( draH-PCS | | | +| | | | | | | | )%( CMA | | | 15 | | | +| ecracS | | | | | | | 24 | | | | | | +| ataD | 0.6 | | | | 48 | | | | | | | | +| | | | | | 40 | | 16 | | | | | | +12 +18 +6 +| | 0.0 | | | | 12 | | 0 | | | 0 | | | +| -------- | --------------- | --- | --- | ------------ | ----- | --- | ------- | --- | --- | ------------ | --- | ------- | +| | 0 | 150 | 300 | 450 | 0 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | draweR gniniarT | | | )%( 005-HTAM | 64 | | | | | | | | +| | | | | | | | 30 | | | )%( draH-PCS | | | +| ytirojaM | | | | | 56 | | )%( CMA | | | | | | +etoV +| | 0.6 | | | | 48 | | | | | 15 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +20 +| | 0.0 | | | | 0 | | | | | 0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +60 +| draweR ysioN | draweR gniniarT | | | )%( 005-HTAM | | | | | | | | | +| ------------ | --------------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | 0.6 | | | | | | 25 | | | )%( draH-PCS | | | +| )7.0= | | | | | 54 | | )%( CMA | | | | | | +15 +| | | | | | 48 | | 20 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.3 +| ( | | | | | | | 15 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| | | | | | 15 | | 5 | | | | | | +| --- | --- | -------------- | --- | --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | | | | | 10 | | 0 | | | 0 | | | +| | 0 | 150 | 300 | 450 | 0 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure6.RLtrainingdynamicsandgeneralizationonMATHforLlama3.2-3BBase,CPT,andInstructvariantsunderdifferent +SFTinitializationsacrossthreeweaksupervisionsettings:scarcedata(N =8,top),majorityvote(middle),andnoisyreward +(γ =0.7,bottom).ThinkingSFT(solidlines)consistentlyprolongsthepre-saturationphaseandimprovesgeneralizationforbothCPT +andBasemodelscomparedtotheirNon-ThinkingSFTcounterparts(dashedlines)andtheInstructbaseline(dash-dot).CPT+Thinking +SFTachievesthestrongestperformanceacrossallsettings. +pretrainingtokenstostrengthenthepretrainingprior. The explicitreasoningtracesinfluencesubsequentRLdynamics. +second is supervised fine-tuning (SFT), with the specific WecomparetwoSFTregimesthatdifferonlyinwhether +questionofwhetherSFTonexplicitreasoningtracesdiffers thesupervisionincludesexplicitreasoning. Bothregimes +in its effect from SFT on final answers alone. Crossing use the same 43.5K math prompts and differ only in the +these axes gives a 2×2 design: two initializations (Base, targetoutput. Specifically,wesamplethesepromptsfrom +CPT)eachfollowedbytwoSFTregimes(Thinking,Non- OpenThoughts-114K (Guha et al., 2025), retaining only +Thinking).WeadditionallyincludeLlama3.2-3B-Instructas thosewhosereasoningtraceshavecorrectfinalanswersand +areference: itsharesthearchitectureofLlama3.2-3B-Base totallengthbelow8192tokens. +| but | has undergone | | extensive instruction | | tuning, | rejection | | | | | | | +| --- | ------------- | --- | --------------------- | --- | ------- | --------- | ------------------------------------------------- | --- | --- | --- | --- | --- | +| | | | | | | | • Non-thinkingSFT:Themodelissupervisedtooutputthe | | | | | | +sampling,andDPO,providingastrongoff-the-shelfbase- +finalsolutionwithoutgeneratingintermediatereasoning +| lineagainstwhichtojudgeourtargetedinterventions. | | | | | | We | | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +traces. +thenrunRLunderallthreeweaksupervisionsettingsfrom +| | | | | | | | • ThinkingSFT:Themodelistrainedonexplicit,verified | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------------- | --- | --- | --- | --- | --- | +§3: scarcedata,noisyrewards,andself-supervisedproxy +long-formreasoningtraces. +rewards. +Wefocusonthe MATH domainfortworeasons: Llama’s AtrainingexampleisshowninFig.12intheAppendix.The +| | | | | | | | SFTregimesarenear-iso-compute: | | | | ThinkingSFTtrainson | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | ------------------- | --- | +baselinefailureissharpestthere,providingthecleanesttest +roughly1Btokens,Non-ThinkingSFTonroughly0.27B, +ofwhetherpre-RLinterventionscanrecovergeneralization; +| | | | | | | | bothnegligiblerelativetothe52B-tokenCPTstage. | | | | | Differ- | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | ------- | +andhigh-qualitymathpretrainingcorpora(Nemotron-CC- +encesbetweenThinkingandNon-ThinkingSFTtherefore +Math)andreasoning-tracedatasets(OpenThoughts-114K) +areavailable,enablingtheinterventionsatsufficientscale. reflectthecontentofthesupervisionratherthanitscost. We +| | | | | | | | reporttheCPTlosscurveinAppendixFig. | | | | | 10andtheSFT | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | ----------- | +ContinualPre-Training(CPT).Wecontinuallypre-train losscurvesinFig. 11. +| Llama3.2-3B-Base | | | for one epoch | on | approximately | 52B | | | | | | | +| ---------------- | --- | --- | ------------- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | +ImplementationandtrainingdetailsofSFTareprovidedin +| math | tokens | from | the Nemotron-CC-Math | | dataset | (Ma- | | | | | | | +| ------------------- | ------ | ---- | ------------------------------- | --- | ------- | ---- | -------------------------------------------------- | --- | ---------------------------------- | --- | --- | --- | +| | | | | | | | AppendixB.6. | | ForthesubsequentRLphase,weevaluate | | | | +| habadietal.,2025)2. | | | TrainingdetailsareprovidedinAp- | | | | | | | | | | +| | | | | | | | acrossallthreeweaksupervisionsettings:scarcedata(N | | | | | = | +pendixB.5. +| | | | | | | | 8), noisy | rewards | (γ = | 0.7), and | self-supervised | proxy | +| ------------------- | --- | --- | ----------------------------- | --- | --- | --- | ---------------------- | ------- | ----------------------------- | --------- | --------------- | ----- | +| SFTTrainingRegimes. | | | FollowingCPTorBaseinitializa- | | | | | | | | | | +| | | | | | | | rewards(majorityvote). | | Allotherhyperparametersfollow | | | | +tion,weapplysupervisedfine-tuningtodeterminewhether theconfigurationsin§2,withthemaximumresponselength +duringRLextendedto8192tokenstoaccommodatelong- +2Nemotron-CC-Math-v1 8 -3 -4 -1 -R -S -& -4 -2 -4 -1 -R -S - -A -N -I -L -O -R -A -C -H -T -R -O -N -, +LLMReasoningwithWeakSupervision +formreasoningtraces. +4.1.Results +| Fig. 6 reports | RL | training | dynamics | for | the five | pre-RL | | | | | | | | +| -------------- | --- | -------- | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | +configurations(Base,CPT,andInstruct,withThinkingSFT +| or Non-Thinking | | SFT | applied to | Base | and CPT) | across | | | | | | | | +| -------------------------------- | ------ | --------- | ---------- | --------------- | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +| thethreeweaksupervisionsettings. | | | | Foreachsetting, | | we | | | | | | | | +| plot training | reward | alongside | three | downstream | | metrics: | | | | | | | | +Figure7.EvolutionofreasoningfaithfulnessoftheLlama3.2- +twoin-domain(MATH-500,AMC)andoneout-of-domain 3Bfamilyonweaksupervisiondomainswhencombinedwith +(SCP-Hard);additionalbenchmarksandpass@kresultsare continualpretrainingandSFTvariants.Whencombinedwith +Fig.34andFig.35inAppendixG.Wedrawthreefindings Thinking-SFT and CPT, the Llama3.2-3B-Base model exhibits +higherreasoningfaithfulness. +fromthisfigure,developedintheparagraphsbelow. +ThinkingSFTisnecessaryforsubstantiallearningunder +| | | | | | | | highest faithfulness | | among | all configurations, | | | consistent | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | ----- | ------------------- | --- | --- | ---------- | +weaksupervision.TheInstructbaselineisflatordecreasing +| | | | | | | | with its strongest | generalization | | | across | all weak | supervi- | +| --- | --- | --- | --- | --- | --- | --- | ------------------ | -------------- | --- | --- | ------ | -------- | -------- | +acrossallthreesettingsonalldownstreamevaluations— +| | | | | | | | sion settings. | Together | with | the | extended | pre-saturation | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | -------- | ---- | --- | -------- | -------------- | --- | +RLproducesnomeaningfulimprovementfromthisstarting +| | | | | | | | dynamics | visible in | Fig. | 6 (leftmost | column), | | this result | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ---- | ----------- | -------- | --- | ----------- | +point. ThinkingSFTistheonlyinterventionthatenables +| | | | | | | | supportsourhypothesisin§3.4: | | | | pre-RLinterventionsthat | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ----------------------- | --- | --- | +substantialdownstreamgainsonscarcedataandmajority +instillfaithfulnessproducelongerpre-saturationphasesand +vote,anditdoessoforbothBaseandCPTinitializations +recoveredgeneralization,inmodelsthatpreviouslyfailed. +| (solidblueandsolidred). | | | Non-ThinkingSFTshowsmodest | | | | | | | | | | | +| ----------------------- | ------------ | ------ | -------------------------- | ------- | ---------------- | ----- | --------- | ---------------------------------- | --- | --- | --- | --- | --- | +| gains only | when | paired | with CPT, | and | only under | noisy | | | | | | | | +| | | | | | | | Takeaway: | SFTonexplicitreasoningtraces,noton | | | | | | +| rewards; | Non-Thinking | | SFT on | Base is | flat or degrades | | | | | | | | | +finalanswers,isnecessaryforLlamatolearnsubstan- +acrossallthreesettings. +| | | | | | | | tiallyfromRLunderweaksupervision. | | | | | Itraisesrea- | | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ------------ | --- | +soningfaithfulness,extendsthepre-saturationphase, +| CPTamplifiestheThinkingSFTeffect. | | | | | ThinkingSFT | | | | | | | | | +| --------------------------------- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +onBasealoneproducesmodestgains. CombinedwithCPT, andrecoversgeneralizationunderscarcedata,noisy +| | | | | | | | rewards, | and self-supervised | | | proxy | rewards. | Con- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------- | --- | --- | ----- | -------- | ---- | +itproducessubstantiallylargergainsoneveryevaluation: +CPT+ThinkingSFTisthetop-performingcurveacrossall tinualpre-trainingamplifiestheeffectbutdoesnot +threeweaksupervisionsettingsandallthreeevals.TheCPT substituteforit: CPT+Non-ThinkingSFTfailsde- +| | | | | | | | spitematchedcompute. | | | Thestrongestconfiguration, | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | -------------------------- | --- | --- | --- | ++Non-ThinkingSFTcomparisonrulesoutacompute-based +explanation:thesame52BCPTtokens,pairedwithSFTtar- CPT + Thinking SFT, recovers performance in set- +tingswhereLlamahadpreviouslycollapsedentirely. +getsthatstripreasoningtraces,failtoenablegeneralization +| onscarcedataandmajorityvote. | | | | Theamplificationisspe- | | | | | | | | | | +| --------------------------------------------------- | --- | --- | ------------------------------- | ---------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | +| cifictothecombination: | | | extrapre-trainingcomputealoneis | | | | | | | | | | | +| insufficient;ThinkingSFTalonehelpsbutislimited,only | | | | | | | 5.RelatedWork | | | | | | | +thecombinationrecoversfullgeneralization. +| | | | | | | | RLVR for | Reasoning. | Reinforcement | | | learning | with ver- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ------------- | --- | --- | -------- | --------- | +Baseinitializationfailsundermostweaksupervisionset- ifiable rewards has emerged as an effective post-training +tingsregardlessofSFT.TheBasemodelshowsmeaningful +| | | | | | | | method for | improving | reasoning | | in large | language | mod- | +| --------------------------------- | --- | --- | --- | --- | ------------- | --- | ----------- | ---------- | --------- | ------- | -------- | -------- | ---------- | +| improvementonlyintwocombinations: | | | | | Base+Thinking | | | | | | | | | +| | | | | | | | els (Guo et | al., 2025; | Olmo | et al., | 2025; | Yu et | al., 2025; | +SFT under scarce data and majority vote, and even there Zengetal.,2025). RecentworkhasexploredwhenRLVR +| gains are | modest. | Under | noisy | rewards, | neither | Base + | | | | | | | | +| --------- | ------- | ----- | ----- | -------- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | +yieldsimprovements(Liuetal.,2025b;a;Huetal.,2025). +Thinking SFT nor Base + Non-Thinking SFT produces Wangetal.(2025a)demonstratethattrainingonasingleex- +| meaningfuldownstreamimprovement. | | | | ThisisolatesCPT’s | | | | | | | | | | +| -------------------------------- | --- | --- | --- | ----------------- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --------- | --- | +| | | | | | | | amplecanprovidemeaningfullearningsignals. | | | | | Otherwork | | +contribution:ThinkingSFTisnecessarybutnotsufficient— +exploresalternativerewards,includingself-certainty(Zhao +domain-alignedpretrainingisrequiredfortheintervention etal.,2025),majorityvoting(Zuoetal.,2025),negativesig- +togeneralizeacrossallthreeweaksupervisionsettings. +nals(Zhuetal.,2025),self-generatedtrainingdataHuang +| | | | | | | | etal.(2025),andspuriousrewards(Shaoetal.,2025). | | | | | | How- | +| ----------------------------------------- | --- | --- | --- | --- | --- | ------- | ----------------------------------------------- | --- | --- | --- | --- | --- | ---- | +| ThinkingSFTimprovesreasoningfaithfulness. | | | | | | In§3.4, | | | | | | | | +ever,thesefindingsoftendonottransferacrossmodelfam- +weidentifiedlowreasoningfaithfulnessasthepre-RLprop- +| | | | | | | | ilies, with | studies | reporting | inconsistent | | results | between | +| --------- | ------------- | --- | ------------ | ---------- | --- | ------- | ----------- | ------- | --------- | ------------ | --- | ------- | ------- | +| erty that | distinguished | | failing from | succeeding | | models. | | | | | | | | +QwenandLlama(Zengetal.,2025;Gandhietal.,2025; +| Fig. 7 shows | that | Thinking | SFT | raises | aligned-response | | | | | | | | | +| ------------ | ---- | -------- | --- | ------ | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | +ratethroughoutthepre-saturationphase,relativetotheNon- Shao et al., 2025). Moreover, most prior work focuses +onimprovingperformanceonnarrowdomains(primarily +| ThinkingSFTbaseline. | | | CPT+ThinkingSFTachievesthe | | | | | | | | | | | +| -------------------- | --- | --- | -------------------------- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | ------------- | --- | +| | | | | | | | math)withoutexamininggeneralization. | | | | | Recentwork(He | | +9 -Y -T +LLMReasoningwithWeakSupervision +et al., 2026; Yang et al., 2026; Plesner et al., 2026) has gesttwoconcretepracticesforRLfromweaksupervision. +concurrentlystudiedwhenandhowRLVRcanlearnunder First, monitor training reward saturation as a diagnostic: +self-supervisionornoisysupervision.Ourworkextendsthis plateauedrewardwithflatdownstreamperformanceindi- +literatureintwoways. First,wecharacterizetheconditions catesthemodelhasexhaustedwhatRLcanextractfromits +underwhichRLVRgeneralizesacrossmodelfamiliesand priors,andfurtherRLcomputeisunlikelytohelp. Second, +domains, focusing on saturation dynamics and reasoning whenweaksupervisionfails, allocatecomputetopre-RL +faithfulness. Second, we identify a concrete intervention interventionsthatinstallstrongpriorsratherthantolonger +thatrestoresgeneralizationinmodelswhereweaksupervi- RL training. Taken together, our findings argue that RL +sionwouldotherwisefail. underweaksupervisionisbestunderstoodnotasatraining +techniqueappliedtoafixedmodel,butasthefinalstageof +| Role of | Pre-Training | and | Fine-Tuning | | in RL. | Recent | | | | | | | | +| ------- | ------------ | --- | ----------- | --- | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | +apipelinewhosesuccessislargelydeterminedbeforeRL +workemphasizesthatpre-trainingandmid-trainingshape +begins. +| RL generalization | | (Qi et | al., 2025; | Wang | et | al., 2025b; | | | | | | | | +| ----------------- | --- | ------ | ---------- | ---- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | +Zhangetal.,2025;Akteretal.,2025),butfocusesoncom- +puteallocationanddistributionalignmenttoimproveperfor- Acknowledgements +mance.Ourworkspecificallyfocusesonunderstandinghow +WewouldliketothankLeonLi,VatsalBaherwani,Rohun +| base model | priors | shaped | from continual | | pretraining | and | | | | | | | | +| ---------- | ------ | ------ | -------------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +Agrawal,SiyanZhao,LiweiJiang,andAndyHanfortheir +| reasoning | SFT can | enable | generalization | | across | different | | | | | | | | +| --------- | ------- | ------ | -------------- | --- | ------ | --------- | ---------- | ----------- | --- | -------- | --- | ---------- | ----- | +| | | | | | | | insightful | discussions | and | feedback | on | the draft. | Pavel | +weaksupervisionsettings. +| | | | | | | | Izmailov | was supported | | by a grant | from | the | Alignment | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------- | --- | ---------- | ---- | --- | --------- | +Diversity and Faithfulness in Reasoning. Maintaining Project,fundedbytheUKAISecurityInstitute(grantAP- +| outputdiversityduringRLhasbeenproposedtopromote | | | | | | | S2-100141). | | | | | | | +| ----------------------------------------------- | --- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +explorationandmitigatemodelcollapse(Kirketal.,2024; +Casperetal.,2023;Rafailovetal.,2023;Yuetal.,2025), +References +| but prior | work has | not explored | | what | types | of diversity | | | | | | | | +| --------- | -------- | ------------ | --- | ---- | ----- | ------------ | --- | --- | --- | --- | --- | --- | --- | +benefitgeneralization. Separately,researchhashighlighted Agarwal,S.,Zhang,Z.,Yuan,L.,Han,J.,andPeng,H. The +| | | | | | | | unreasonable | effectiveness | | of | entropy | minimization | in | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------- | --- | --- | ------- | ------------ | --- | +mismatchesbetweenchain-of-thoughttracesandmodelpre- +dictions(Turpinetal.,2023;Chenetal.,2025b;Bakeretal., llmreasoning. arXivpreprintarXiv:2505.15134,2025. +2025;Tuteketal.,2025)andemphasizedtheimportanceof +| | | | | | | | AI-MO. | Aime 2024. | https://huggingface.co | | | | | +| ------------------------------------------- | --- | --- | --- | --- | --- | ---------- | ------ | ---------- | ---------------------- | --- | --- | --- | --- | +| ensuringfaithfulreasoningthroughouttraining | | | | | | (Guietal., | | | | | | | | +/datasets/AI-MO/aimo-validation-aime, +2026). Wenetal.(2025)arguesthatRLVRcanincentivize +2024a. +correctreasoninginbaseLLMsaslongaspriorshavebeen +| established. | Our | work connects | | these | lines of | research, | | | | | | | | +| ------------ | -------------- | ------------- | ---- | ----- | -------- | ----------- | ------ | --------- | ---------------------- | --- | --- | --- | --- | +| | | | | | | | AI-MO. | Amc 2023. | https://huggingface.co | | | | | +| showing | that diversity | alone | does | not | ensure | generaliza- | | | | | | | | +/datasets/AI-MO/aimo-validation-amc, +tionandthatreasoningfaithfulnessdistinguishesmodels’ +2024b. +| trainingdynamics. | | Wefurtherdemonstratethatpre-RLin- | | | | | | | | | | | | +| ----------------- | --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +terventioncanimprovereasoningfaithfulnessandimprove Akter, S. N., Prabhumoye, S., Nyberg, E., Patwary, M., +generalizationunderweaksupervision. Shoeybi,M.,Choi,Y.,andCatanzaro,B. Front-loading +| | | | | | | | reasoning: | The | synergy | between | pretraining | | and post- | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------- | ------- | ----------- | --- | --------- | +6.Conclusion trainingdata. arXivpreprintarXiv:2510.03264,2025. +| | | | | | | | Baker, B., | Huizinga, | J., | Gao, L., | Dou, | Z., Guan, | M. Y., | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | --- | -------- | ---- | --------- | ------ | +Inthiswork,westudiedwhenandwhyRLVRgeneralizes +underweaksupervisionacrossdiversemodelfamiliesand Madry,A.,Zaremba,W.,Pachocki,J.,andFarhi,D.Mon- +threereasoningdomains. Successunderscarcedata,noisy itoringreasoningmodelsformisbehaviorandtherisksof +promotingobfuscation.arXivpreprintarXiv:2503.11926, +rewards,andself-supervisedproxyrewardsdependsonpre- +| RLproperties,pretrainingpriorsandreasoningfaithfulness, | | | | | | | 2025. | | | | | | | +| ------------------------------------------------------- | ----- | -------- | ------ | ------ | ---- | -------- | ------- | ------------ | --- | ---------- | --------- | --- | ----------- | +| rather than | on RL | dynamics | alone. | Models | that | saturate | | | | | | | | +| | | | | | | | Bowman, | S. R., Hyun, | | J., Perez, | E., Chen, | E., | Pettit, C., | +rapidlyproducecorrectanswersthroughreasoningthatdoes +| | | | | | | | Heiner, | S., Lukosˇiu¯te˙, | | K., Askell, | A., | Jones, | A., Chen, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----------------- | --- | ----------- | --- | ------ | --------- | +notsupportthem,memorizingratherthanlearning,while +| | | | | | | | A.,etal. | Measuringprogressonscalableoversightfor | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------- | --------------------------------------- | --- | --- | --- | --- | --- | +maintainingthehighoutputdiversitynormallytakenasa +| | | | | | | | largelanguagemodels. | | | arXivpreprintarXiv:2211.03540, | | | | +| --------------- | ------------ | --- | ------ | ------------- | --- | ------- | -------------------- | --- | --- | ------------------------------ | --- | --- | --- | +| sign of healthy | exploration. | | Pre-RL | interventions | | target- | | | | | | | | +2022. +| ingreasoningfaithfulnessrecovergeneralization: | | | | | | SFTon | | | | | | | | +| ---------------------------------------------- | --- | ------ | ---------------- | --- | ----------- | ----- | --- | --- | --- | --- | --- | --- | --- | +| explicit reasoning | | traces | is the necessary | | ingredient, | and | | | | | | | | +Burns,C.,Izmailov,P.,Kirchner,J.H.,Baker,B.,Gao,L., +| continual | pre-training | on | reasoning-heavy | | data | amplifies | | | | | | | | +| --------- | ------------ | --- | --------------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | +Aschenbrenner,L.,Chen,Y.,Ecoffet,A.,Joglekar,M., +the effect without substituting for it. These findings sug- Leike,J.,etal. Weak-to-stronggeneralization: Eliciting +10 + +LLMReasoningwithWeakSupervision +strongcapabilitieswithweaksupervision. arXivpreprint Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, +arXiv:2312.09390,2023. A.,Letman,A.,Mathur,A.,Schelten,A.,Yang,A.,Fan, +| | | | | | | A.,etal. | Thellama3herdofmodels. | | | arXive-prints,pp. | | | +| --- | --- | --- | --- | --- | --- | -------- | ---------------------- | --- | --- | ----------------- | --- | --- | +Casper,S.,Davies,X.,Shi,C.,Gilbert,T.K.,Scheurer,J., arXiv–2407,2024. +Rando,J.,Freedman,R.,Korbak,T.,Lindner,D.,Freire, +| | | | | | | Farquhar,S.,Kossen,J.,Kuhn,L.,andGal,Y. | | | | | Detecting | | +| --------- | ------ | ------ | ------------- | ------ | ------------ | --------------------------------------- | --- | --- | --- | --- | --------- | --- | +| P., Wang, | T. T., | Marks, | S., Se´gerie, | C.-R., | Carroll, M., | | | | | | | | +hallucinationsinlargelanguagemodelsusingsemantic +Peng,A.,Christoffersen,P.J.K.,Damani,M.,Slocum,S., +| | | | | | | entropy. | Nature,630(8017):625–630,2024. | | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------ | --- | --- | --- | --- | --- | +Anwar,U.,Siththaranjan,A.,Nadeau,M.,Michaud,E.J., +Pfau, J., Krasheninnikov, D., Chen, X., Langosco, L., Gandhi, K., Chakravarthy, A., Singh, A., Lile, N., and +Hase,P.,Biyik,E.,Dragan,A.D.,Krueger,D.,Sadigh, +| | | | | | | Goodman, | N. | D. Cognitive | behaviors | | that enable | self- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------------ | --------- | --- | ----------- | ----- | +D.,andHadfield-Menell,D. Openproblemsandfunda- improvingreasoners,or,fourhabitsofhighlyeffective +mentallimitationsofreinforcementlearningfromhuman stars. arXivpreprintarXiv:2503.01307,2025. +| feedback. | TransactionsonMachineLearningResearch, | | | | | | | | | | | | +| --------- | -------------------------------------- | --- | --- | --- | --- | -------------------- | --------- | ------ | --------------------- | ------------ | --- | --- | +| | | | | | | Google | DeepMind. | Gemini | 3 flash. | https://gemi | | | +| 2023. | URL https://openreview.net/forum | | | | | | | | | | | | +| | | | | | | ni.google.com/,2025. | | | ReleasedDecember2025. | | | | +?id=bx24KpJ4Eb. +| | | | | | | Accessed: | 2026-02-18. | | | | | | +| --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | --- | --- | --- | --- | +Chandak,N.,Goel,S.,andPrabhu,A. Incorrectbaseline Guha, E., Marten, R., Keh, S., Raoof, N., Smyrnis, G., +| evaluations | call | into | question recent | llm-rl | claims. ht | | | | | | | | +| ----------- | ---- | ---- | --------------- | ------ | ---------- | --- | --- | --- | --- | --- | --- | --- | +Bansal,H.,Nezhurina,M.,Mercat,J.,Vu,T.,Sprague,Z., +tps://safe-lip-9a8.notion.site/Incor +| | | | | | | etal. | Openthoughts: | Datarecipesforreasoningmodels. | | | | | +| --- | --- | --- | --- | --- | --- | ----- | ------------- | ------------------------------ | --- | --- | --- | --- | +rect-Baseline-Evaluations-Call-into-Q +arXivpreprintarXiv:2506.04178,2025. +uestion-Recent-LLM-RL-Claims-2012f1f +| | | | | | | Gui, R., | Li, Y., | Qu, X., Liu, | Z., | Cheng, | Y., and Cheng, | | +| -------------------------------- | --- | --- | --- | --- | ----- | -------- | ------- | ------------ | --- | ------ | -------------- | --- | +| bf0ee8094ab8ded1953c15a37?pvs=4, | | | | | 2025. | | | | | | | | +NotionBlog. Y. Faithrl: Learning to reason faithfully through +| | | | | | | step-level | faithfulness | maximization. | | | arXiv preprint | | +| --- | --- | --- | --- | --- | --- | ---------- | ------------ | ------------- | --- | --- | -------------- | --- | +arXiv:2602.03507,2026. +| Chen, P., | Li, X., | Li, Z., | Yin, W., Chen, | X., | and Lin, T. | | | | | | | | +| ----------- | -------- | ------------------ | -------------- | ------------- | ------------ | ----------------------------------- | ----- | ---------- | --------- | --- | --------- | ---- | +| Exploration | vs | exploitation: | Rethinking | | rlvr through | | | | | | | | +| | | | | | | Guo, D., | Yang, | D., Zhang, | H., Song, | J., | Wang, P., | Zhu, | +| clipping, | entropy, | andspuriousreward. | | arXivpreprint | | | | | | | | | +| | | | | | | Q.,Xu,R.,Zhang,R.,Ma,S.,Bi,X.,etal. | | | | | Deepseek- | | +arXiv:2512.16912,2025a. +r1incentivizesreasoninginllmsthroughreinforcement +| | | | | | | learning. | Nature,645(8081):633–638,2025. | | | | | | +| --- | --- | --- | --- | --- | --- | --------- | ------------------------------ | --- | --- | --- | --- | --- | +Chen,Y.,Benton,J.,Radhakrishnan,A.,Uesato,J.,Deni- +son,C.,Schulman,J.,Somani,A.,Hase,P.,Wagner,M., He, B., Zuo, Y., Liu, Z., Zhao, S., Fu, Z., Yang, J., Qian, +Roger,F.,etal. Reasoningmodelsdon’talwayssaywhat C., Zhang, K., Fan, Y., Cui, G., et al. How far can +theythink. arXivpreprintarXiv:2505.05410,2025b. unsupervised rlvr scale llm training? arXiv preprint +arXiv:2603.08660,2026. +Cheng,Z.,Hao,S.,Liu,T.,Zhou,F.,Xie,Y.,Yao,F.,Bian, +He,C.,Luo,R.,Bai,Y.,Hu,S.,Thai,Z.,Shen,J.,Hu,J., +Y.,Zhuang,Y.,Dey,N.,Zha,Y.,Gu,Y.,Zhou,K.,Wang, +| | | | | | | Han,X.,Huang,Y.,Zhang,Y.,etal. | | | | Olympiadbench: | | A | +| ------- | -------- | --- | ------------- | ------------ | ------- | ------------------------------ | --- | --- | --- | -------------- | --- | --- | +| Y., Li, | Y., Fan, | R., | She, J., Gao, | C., Saparov, | A., Li, | | | | | | | | +H., Killian, T. W., Yurochkin, M., Liu, Z., Xing, E. P., challengingbenchmarkforpromotingagiwitholympiad- +| | | | | | | levelbilingualmultimodalscientificproblems. | | | | | InPro- | | +| ------- | ------------- | --- | ------------- | -------- | ------- | ------------------------------------------- | --- | --- | --- | --- | ------ | --- | +| and Hu, | Z. Revisiting | | reinforcement | learning | for llm | | | | | | | | +ceedingsofthe62ndAnnualMeetingoftheAssociation +| reasoningfromacross-domainperspective,2025. | | | | | URL | | | | | | | | +| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +https://arxiv.org/abs/2506.14965. forComputationalLinguistics(Volume1: LongPapers), +pp.3828–3850,2024. +| Cohen, J. | A coefficient | | of agreement | for nominal | scales. | | | | | | | | +| --------- | ------------- | --- | ------------ | ----------- | ------- | ------------ | -------- | ------ | -------- | ----- | ---------- | --- | +| | | | | | | He, J., Liu, | J., Liu, | C. Y., | Yan, R., | Wang, | C., Cheng, | P., | +EducationalandPsychologicalMeasurement,20(1):37– +| | | | | | | Zhang, | X., Zhang, | F., Xu, | J., Shen, | W., | Li, S., | Zeng, | +| --- | --- | --- | --- | --- | --- | ------ | ---------- | ------- | --------- | --- | ------- | ----- | +46,1960. +| | | | | | | L., Wei, | T., Cheng, | C., | An, B., | Liu, Y., | and Zhou, | Y. | +| --- | --- | --- | --- | --- | --- | ------------------------------------ | ---------- | --- | ------- | -------- | ------------- | --- | +| | | | | | | Skyworkopenreasoner1technicalreport. | | | | | arXivpreprint | | +Cui,G.,Zhang,Y.,Chen,J.,Yuan,L.,Wang,Z.,Zuo,Y.,Li, arXiv:2505.22312,2025a. +| H.,Fan,Y.,Chen,H.,Chen,W.,etal. | | | | Theentropymech- | | | | | | | | | +| ------------------------------- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +anismofreinforcementlearningforreasoninglanguage He, J., Liu, J., Liu, C. Y., Yan, R., Wang, C., Cheng, P., +Zhang,X.,Zhang,F.,Xu,J.,Shen,W.,Li,S.,Zeng,L., +| models. | arXivpreprintarXiv:2505.22617,2025. | | | | | | | | | | | | +| ------- | ----------------------------------- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | --- | +| | | | | | | Wei,T.,Cheng,C.,Liu,Y.,andZhou,Y. | | | | | Skyworkopen | | +Du,X.,Yao,Y.,Ma,K.,Wang,B.,Zheng,T.,Zhu,K.,Liu, reasonerseries. https://capricious-hydroge +M.,Liang,Y.,Jin,X.,Wei,Z.,etal. Supergpqa: Scaling n-41c.notion.site/Skywork-Open-Reaon +llm evaluation across 285 graduate disciplines. arXiv ser-Series-1d0bc9ae823a80459b46c149e +| preprintarXiv:2502.14739,2025. | | | | | | 4f51680,2025b. | | NotionBlog. | | | | | +| ------------------------------ | --- | --- | --- | --- | --- | -------------- | --- | ----------- | --- | --- | --- | --- | +11 + +LLMReasoningwithWeakSupervision +Hendrycks,D.,Burns,C.,Kadavath,S.,Arora,A.,Basart, Liu, Z., Chen, C., Li, W., Qi, P., Pang, T., Du, C., Lee, +S., Tang, E., Song, D., and Steinhardt, J. Measuring W.S.,andLin,M. Understandingr1-zero-liketraining: +mathematicalproblemsolvingwiththemathdataset. In Acriticalperspective. arXivpreprintarXiv:2503.20783, +| Thirty-fifthConferenceonNeuralInformationProcessing | | | | | | | 2025b. | | | | | | | | +| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +SystemsDatasetsandBenchmarksTrack(Round2). +Lu,D.,Tan,X.,Xu,R.,Yao,T.,Qu,C.,Chu,W.,Xu,Y.,and +Qi,Y.Scp-116k:Ahigh-qualityproblem-solutiondataset +| Hu, J., Liu, | M., | Lu, X., | Wu, F., | Harchaoui, | | Z., Diao, S., | | | | | | | | | +| ------------ | --- | ------- | ------- | ---------- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Choi,Y.,Molchanov,P.,Yang,J.,Kautz,J.,etal. Brorl: and a generalized pipeline for automated extraction in +| | | | | | | | the higher | education | | science | domain. | | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | --- | ------- | ------- | --- | ----- | -------- | +Scalingreinforcementlearningviabroadenedexploration. +| arXivpreprintarXiv:2510.01180,2025. | | | | | | | arXiv:2501.15587,2025. | | | | | | | | +| ----------------------------------- | ------- | --------- | ------- | ------ | ------- | ----------- | ----------------------------- | ----- | --------- | --- | ----------- | ----------------- | --- | -------- | +| | | | | | | | Mahabadi, | R.K., | Satheesh, | S., | Prabhumoye, | | S., | Patwary, | +| Huang, | C., Yu, | W., Wang, | X., | Zhang, | H., | Li, Z., Li, | | | | | | | | | +| | | | | | | | M.,Shoeybi,M.,andCatanzaro,B. | | | | | Nemotron-cc-math: | | | +| R., Huang, | J., | Mi, | H., and | Yu, D. | R-zero: | Self- | | | | | | | | | +A133billion-token-scalehighqualitymathpretraining +| evolving | reasoning | llm | from | zero data. | arXiv | preprint | | | | | | | | | +| -------- | --------- | --- | ---- | ---------- | ----- | -------- | -------- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | | | | | dataset. | arXivpreprintarXiv:2508.15096,2025. | | | | | | | +arXiv:2508.05004,2025. +| | | | | | | | Olmo, T., | Ettinger, | A., | Bertsch, | A., | Kuehl, | B., | Graham, | +| --- | --- | --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --- | ------ | --- | ------- | +Hurst,A.,Lerer,A.,Goucher,A.P.,Perelman,A.,Ramesh, D., Heineman, D., Groeneveld, D., Brahman, F., Tim- +A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., arXiv preprint +| | | | | | | | bers, | F., Ivison, | H., | et al. | Olmo | 3. | | | +| ---------------- | --- | ----------------- | --- | --- | ------------- | --- | ----- | ----------- | --- | ------ | ---- | --- | --- | --- | +| Radford,A.,etal. | | Gpt-4osystemcard. | | | arXivpreprint | | | | | | | | | | +arXiv:2512.13961,2025. +arXiv:2410.21276,2024. +| | | | | | | | OpenAI. | Openai | o3 and | o4-mini | system | | card. | https: | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------ | ------ | ------- | ------ | --- | ----- | ------ | +Jaech,A.,Kalai,A.,Lerer,A.,Richardson,A.,El-Kishky, //cdn.openai.com/pdf/2221c875-02dc-4 +A., Low, A., Helyar, A., Madry, A., Beutel, A., Car- 789-800b-e7758f3722c1/o3-and-o4-min +ney, A., et al. Openai o1 system card. arXiv preprint i-system-card.pdf,April2025a. Accessed2026- +01-25. +arXiv:2412.16720,2024. +| | | | | | | | OpenAI. | gpt-oss-20b | | model | card, | 2025b. | URL | https: | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----------- | --- | ----- | ----- | ------ | --- | ------ | +Kirk,R.,Mediratta,I.,Nalmpantis,C.,Luketina,J.,Ham- +| | | | | | | | //arxiv.org/abs/2508.10925. | | | | | | Accessed:2026- | | +| ------------------------------------- | --- | --- | --- | --- | ------------- | --- | --------------------------- | --- | --- | --- | --- | --- | -------------- | --- | +| bro,E.,Grefenstette,E.,andRaileanu,R. | | | | | Understanding | | | | | | | | | | +02-18. +| theeffectsofrlhfonllmgeneralisationanddiversity. | | | | | | In | | | | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +TheTwelfthInternationalConferenceonLearningRepre- +| | | | | | | | Opencompass. | | Aime2025. | https://huggingface. | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------- | -------------------- | --- | --- | --- | --- | +sentations,2024. URLhttps://openreview.net co/datasets/opencompass/AIME2025,2024. +/forum?id=PXD3FAVHJT. +| | | | | | | | Plesner, | A., Guzma´n, | | F., and | Athalye, | A. | An | imperfect | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------ | --- | ------- | -------- | --- | --- | --------- | +Lewkowycz, A., Andreassen, A., Dohan, D., Dyer, E., verifier is good enough: Learning with noisy rewards. +Michalewski, H., Ramasesh, V., Slone, A., Anil, C., arXivpreprintarXiv:2604.07666,2026. +| Schlag,I.,Gutman-Solo,T.,etal. | | | | Solvingquantitative | | | | | | | | | | | +| ------------------------------------ | --- | --- | --- | ------------------- | --- | ---------- | ------------ | --- | ----- | ------------- | --- | ---------------- | --- | --- | +| | | | | | | | Prabhudesai, | M., | Chen, | L., Ippoliti, | | A., Fragkiadaki, | | K., | +| reasoningproblemswithlanguagemodels. | | | | | | Advancesin | | | | | | | | | +neural information processing systems, 35:3843–3857, Liu, H., and Pathak, D. Maximizing confidence alone +| | | | | | | | improvesreasoning. | | | arXivpreprintarXiv:2505.22660, | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | ------------------------------ | --- | --- | --- | --- | +2022. +2025. +Li,T.,Zhang,Y.,Yu,P.,Saha,S.,Khashabi,D.,Weston,J., +| | | | | | | | Qi, Z., Nie, | F., | Alahi, | A., Zou, | J., | Lakkaraju, | H., | Du, Y., | +| ------------------------ | --- | --- | --------------------------- | --- | --- | --- | ----------------------------- | --- | ------ | -------- | --- | ---------- | ---------- | ------- | +| Lanchantin,J.,andWang,T. | | | Jointlyreinforcingdiversity | | | | | | | | | | | | +| | | | | | | | Xing,E.,Kakade,S.,andZhang,H. | | | | | Evolm: | Insearchof | | +andqualityinlanguagemodelgenerations.arXivpreprint +| | | | | | | | lostlanguagemodeltrainingdynamics. | | | | | | arXivpreprint | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | ------------- | --- | +arXiv:2509.02534,2025. +arXiv:2506.16029,2025. +Lightman,H.,Kosaraju,V.,Burda,Y.,Edwards,H.,Baker, +Rafailov,R.,Sharma,A.,Mitchell,E.,Ermon,S.,Manning, +| B., Lee, | T., Leike, | J., | Schulman, | J., | Sutskever, | I., and | | | | | | | | | +| -------- | ---------- | --- | --------- | --- | ---------- | ------- | --------------- | --- | ----------------------------- | --- | --- | --- | --- | ---- | +| | | | | | | | C.D.,andFinn,C. | | Directpreferenceoptimization: | | | | | Your | +Cobbe, K. Let’s verify step by step. In The Twelfth InAdvances +languagemodelissecretlyarewardmodel. +InternationalConferenceonLearningRepresentations, +| | | | | | | | in Neural | Information | | Processing | | Systems, | volume | 36, | +| ----- | --- | --- | --- | --- | --- | --- | -------------------- | ----------- | --- | ---------- | --- | -------- | ------ | --- | +| 2023. | | | | | | | pp.53728–53741,2023. | | | | | | | | +Liu,M.,Diao,S.,Lu,X.,Hu,J.,Dong,X.,Choi,Y.,Kautz, Rahman,S.,Issaka,S.,Suvarna,A.,Liu,G.,Shiffer,J.,Lee, +J.,andDong,Y. Prorl: Prolongedreinforcementlearning J.,Parvez,M.R.,Palangi,H.,Feng,S.,Peng,N.,etal. +expandsreasoningboundariesinlargelanguagemodels. Aidebateaidsassessmentofcontroversialclaims. arXiv +arXivpreprintarXiv:2505.24864,2025a. preprintarXiv:2506.02175,2025. +12 + +LLMReasoningwithWeakSupervision +Rein,D.,Hou,B.L.,Stickland,A.C.,Petty,J.,Pang,R.Y., Suzhou,China,November2025.AssociationforCom- +Dirani, J., Michael, J., and Bowman, S. R. Gpqa: A putationalLinguistics. ISBN979-8-89176-332-6. doi: +10.18653/v1/2025.emnlp-main.504.URLhttps://ac +| graduate-level | | google-proof | q&a benchmark. | | In First | | | | | +| -------------- | --- | ------------ | -------------- | --- | -------- | --- | --- | --- | --- | +ConferenceonLanguageModeling,2024. lanthology.org/2025.emnlp-main.504/. +Shafayat, S., Tajwar, F., Salakhutdinov, R., Schneider, J., Wang,X.,Hu,Z.,Lu,P.,Zhu,Y.,Zhang,J.,Subramaniam, +| andZanette,A. | | Canlargereasoningmodelsself-train? | | | | | | | | +| ------------- | --- | ---------------------------------- | --- | --- | --- | ----------- | ------------- | -------- | ---------------- | +| | | | | | | S., Loomba, | A. R., Zhang, | S., Sun, | Y., and Wang, W. | +arXivpreprintarXiv:2505.21444,2025. Scibench: Evaluating college-level scientific problem- +| | | | | | | solvingabilitiesoflargelanguagemodels. | | | arXivpreprint | +| ------------ | ----------------------------------- | --- | --- | --- | --- | -------------------------------------- | --- | --- | ------------- | +| Shannon,C.E. | Amathematicaltheoryofcommunication. | | | | | | | | | +arXiv:2307.10635,2023. +TheBellsystemtechnicaljournal,27(3):379–423,1948. +Wang,Y.,Ma,X.,Zhang,G.,Ni,Y.,Chandra,A.,Guo,S., +| Shao, R., | Li, S. | S., Xin, | R., Geng, S., | Wang, | Y., Oh, S., | | | | | +| ------------------------------------------ | ------ | -------- | ------------- | ----- | ----------- | -------------------------------------- | --- | --- | --------- | +| | | | | | | Ren,W.,Arulraj,A.,He,X.,Jiang,Z.,etal. | | | Mmlu-pro: | +| Du,S.S.,Lambert,N.,Min,S.,Krishna,R.,etal. | | | | | Spu- | | | | | +riousrewards: Rethinkingtrainingsignalsinrlvr. arXiv Amorerobustandchallengingmulti-tasklanguageun- +| | | | | | | derstandingbenchmark. | | AdvancesinNeuralInformation | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --------------------------- | --- | +preprintarXiv:2506.10947,2025. +ProcessingSystems,37:95266–95290,2024. +Shao,Z.,Wang,P.,Zhu,Q.,Xu,R.,Song,J.,Bi,X.,Zhang, +H.,Zhang,M.,Li,Y.,Wu,Y.,etal. Deepseekmath: Push- Wang, Y., Yang, Q., Zeng, Z., Ren, L., Liu, L., Peng, B., +ingthelimitsofmathematicalreasoninginopenlanguage Cheng,H.,He,X.,Wang,K.,Gao,J.,etal. Reinforce- +mentlearningforreasoninginlargelanguagemodelswith +| models. | arXivpreprintarXiv:2402.03300,2024. | | | | | | | | | +| ------- | ----------------------------------- | --- | --- | --- | --- | ------------------- | ------------------------------ | --- | --- | +| | | | | | | onetrainingexample. | arXivpreprintarXiv:2504.20571, | | | +Sheng,G.,Zhang,C.,Ye,Z.,Wu,X.,Zhang,W.,Zhang, +2025a. +| R., Peng, | Y., | Lin, H., | and Wu, | C. Hybridflow: | A | | | | | +| --------- | --- | -------- | ------- | -------------- | --- | --- | --- | --- | --- | +flexible and efficient rlhf framework. arXiv preprint Wang, Z., Zhou, F., Li, X., and Liu, P. Octothinker: +arXiv:2409.19256,2024. Mid-trainingincentivizesreinforcementlearningscaling. +arXivpreprintarXiv:2506.20512,2025b. +Stojanovski,Z.,Stanley,O.,Sharratt,J.,Jones,R.,Adefioye, +A.,Kaddour,J.,andKo¨pf,A.Reasoninggym:Reasoning Wen,X.,Liu,Z.,Zheng,S.,Ye,S.,Wu,Z.,Wang,Y.,Xu, +environmentsforreinforcementlearningwithverifiable +| | | | | | | Z.,Liang,X.,Li,J.,Miao,Z.,etal. | | Reinforcementlearn- | | +| --- | --- | --- | --- | --- | --- | ------------------------------- | --- | ------------------- | --- | +arXivpreprintarXiv:2505.24760,2025. +rewards. ingwithverifiablerewardsimplicitlyincentivizescorrect +| | | | | | | reasoninginbasellms. | arXivpreprintarXiv:2506.14245, | | | +| -------- | --------- | ----- | --------- | --------- | --------- | -------------------- | ------------------------------ | --- | --- | +| Sun, Y., | Shen, J., | Wang, | Y., Chen, | T., Wang, | Z., Zhou, | | | | | +2025. +| M., and | Zhang, | H. Improving | data | efficiency | for llm | | | | | +| ------- | ------ | ------------ | ---- | ---------- | ------- | --- | --- | --- | --- | +reinforcementfine-tuningthroughdifficulty-targetedon- +| | | | | | | Williams,R.J. | Simplestatisticalgradient-followingalgo- | | | +| --------- | --------- | --- | --------------- | ----- | -------- | -------------------------------------------- | ---------------------------------------- | --- | ------- | +| line data | selection | and | rollout replay. | arXiv | preprint | | | | | +| | | | | | | rithmsforconnectionistreinforcementlearning. | | | Machine | +arXiv:2506.05316,2025. +learning,8(3):229–256,1992. +Team,K.,Du,A.,Gao,B.,Xing,B.,Jiang,C.,Chen,C., +Yang,A.,Zhang,B.,Hui,B.,Gao,B.,Yu,B.,Li,C.,Liu,D., +| Li, C., | Xiao, | C., Du, | C., Liao, C., | et al. | Kimi k1. 5: | | | | | +| ------- | ----- | ------- | ------------- | ------ | ----------- | --- | --- | --- | --- | +Tu,J.,Zhou,J.,Lin,J.,Lu,K.,Xue,M.,Lin,R.,Liu,T., +| Scalingreinforcementlearningwithllms. | | | | arXivpreprint | | | | | | +| ------------------------------------- | --- | --- | --- | ------------- | --- | ------------------ | --- | ---------------------------- | --- | +| | | | | | | Ren,X.,andZhang,Z. | | Qwen2.5-mathtechnicalreport: | | +arXiv:2501.12599,2025. +Towardmathematicalexpertmodelviaself-improvement. +arXivpreprintarXiv:2409.12122,2024. +| Team,Q. | Qwen2.5: | Apartyoffoundationmodels,Septem- | | | | | | | | +| ------------- | ----------------------------- | -------------------------------- | --- | --- | --- | ---------------------------------------------- | --------------- | ------------- | ----------- | +| ber2024. | URLhttps://qwenlm.github.io/b | | | | | | | | | +| log/qwen2.5/. | | | | | | Yang,S.,Zhu,G.,Song,B.,Li,S.,Wang,H.,Zheng,X., | | | | +| | | | | | | Ma, Y., | Chen, Z., Wang, | W., and Chen, | G. Can llms | +Turpin, M., Michael, J., Perez, E., andBowman, S. Lan- learntoreasonrobustlyundernoisysupervision? arXiv +guage models don’t always say what they think: Un- preprintarXiv:2604.03993,2026. +| faithfulexplanationsinchain-of-thoughtprompting. | | | | | Ad- | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +vances in Neural Information Processing Systems, 36: Yu,Q.,Zhang,Z.,Zhu,R.,Yuan,Y.,Zuo,X.,Yue,Y.,Dai, +| | | | | | | W.,Fan,T.,Liu,G.,Liu,L.,etal. | | Dapo: | Anopen-source | +| --- | --- | --- | --- | --- | --- | ----------------------------- | --- | ----- | ------------- | +74952–74965,2023. +llmreinforcementlearningsystematscale.arXivpreprint +| Tutek, M., | Hashemi | Chaleshtori, | F., | Marasovic, | A., and | | | | | +| ---------- | ------- | ------------ | --- | ---------- | ------- | --- | --- | --- | --- | +arXiv:2503.14476,2025. +| Belinkov, | Y. | Measuring | chain of | thought | faithfulness | | | | | +| --------- | --- | --------- | -------- | ------- | ------------ | --- | --- | --- | --- | +by unlearning reasoning steps. In Christodoulopoulos, Yue,Y.,Chen,Z.,Lu,R.,Zhao,A.,Wang,Z.,Song,S.,and +C.,Chakraborty,T.,Rose,C.,andPeng,V.(eds.),Pro- Huang,G. Doesreinforcementlearningreallyincentivize +ceedings of the 2025 Conference on Empirical Meth- reasoningcapacityinllmsbeyondthebasemodel? arXiv +ods in Natural Language Processing, pp. 9935–9960, preprintarXiv:2504.13837,2025. +13 + +LLMReasoningwithWeakSupervision +| Zeng, W., | Huang, Y., Liu, | Q., Liu, W., | He, K., Ma, Z., | +| --------- | --------------- | -------------------------- | --------------- | +| andHe,J. | Simplerl-zoo: | Investigatingandtamingzero | | +reinforcementlearningforopenbasemodelsinthewild. +arXivpreprintarXiv:2503.18892,2025. +| Zhang, C., | Neubig, G., and | Yue, X. On | the interplay of | +| ---------- | --------------- | ---------- | ---------------- | +pre-training,mid-training,andrlonreasoninglanguage +| models. | arXivpreprintarXiv:2512.07783,2025. | | | +| --------- | ----------------------------------- | ----------------- | ------------ | +| Zhao, X., | Kang, Z., Feng, | A., Levine, S., | and Song, D. | +| Learning | to reason without | external rewards. | arXiv | +preprintarXiv:2505.19590,2025. +Zhu,X.,Xia,M.,Wei,Z.,Chen,W.-L.,Chen,D.,andMeng, +Y. Thesurprisingeffectivenessofnegativereinforcement +| inllmreasoning. | arXivpreprintarXiv:2506.01347,2025. | | | +| --------------- | ----------------------------------- | --- | --- | +Zuo,Y.,Zhang,K.,Sheng,L.,Qu,S.,Cui,G.,Zhu,X.,Li, +| H.,Zhang,Y.,Long,X.,Hua,E.,etal. | | | Ttrl: Test-time | +| -------------------------------- | --- | --- | --------------- | +reinforcementlearning.arXivpreprintarXiv:2504.16084, +2025. +14 + +LLMReasoningwithWeakSupervision +A.LimitationsandFutureWork +Weacknowledgeseverallimitations. First,duetocomputationalconstraints,ouranalysisisrestrictedtospecificmodel +families and scales. Validating these findings across larger architectures and broader task suites remains an important +direction.Second,ouranalysisofdiversityandfaithfulnessreliesonanLLM-as-a-judgeframework.Althoughweconducted +small-scalehumanverificationtovalidatelabelquality,wecurrentlyrestrictthisevaluationtoasmallscaletoallowfor +reasonablelabelingcosts.Consequently,thedevelopmentofscalablemetricsforreasoningfaithfulnessanddiversityremains +animportantdirectionforfutureresearch. +B.ImplementationDetails +B.1.TrainingandEvaluationDatasets +WeinvestigateRLtrainingdynamicsacrosstwomodelfamilies: Qwen(comprisingQwen2.5-1.5B/3BandQwen2.5-Math- +1.5B/7B)andLlama(Llama-3.2-3B/8B-Instruct). Ouranalysisspansthreedistinctreasoningdomains,MATH,SCIENCE, +andGRAPH,allowingforaholisticinvestigationofRLVRunderweaksupervisionacrossdifferentdomainsandmodel +families. ForMATH,wesampletrainingpromptsfromtheSkywork-OR1(Heetal.,2025b)dataset. ForSCIENCE,wedraw +problemsfromtheSCPdatasetcuratedbypriorwork(Liuetal.,2025a;Luetal.,2025),byselectingPhysics,Chemistry, +andBiologysubjects. ForGRAPH,wegeneratetwosyntheticalgorithmictasks,QuantumLockandLargestIsland,using +thecurriculumspecificationsprovidedbytheReasoningGymbenchmark(Stojanovskietal.,2025). Foreachtask,we +instantiatefivedifficultylevelsfollowingthebenchmark’scurriculum,withabalancednumberofsamplesperlevel. +Weincludethefollowingdomain-specificbenchmarksforevaluations: +• MATH500(Lightmanetal.,2023): AwidelyusedsubsetoftheMATHtestsplit(Hendrycksetal.). +• AMC(AI-MO,2024b): 40competition-levelmathquestions. +• AIME2024(AI-MO,2024a): 30competition-levelmathquestions. +• AIME2025(Opencompass,2024): 30competition-levelmathquestions. +• MinervaMath(Lewkowyczetal.,2022): Asetof272undergraduate-levelscienceandmathquestionsfromMIT +OpenCourseWare. +• OlympiadBench(Heetal.,2024): Abenchmarkof675problemsfrominternationalmatholympiadsandphysics +contests. +• GPQA-Diamond(Reinetal.,2024): 198expert-levelquestionsfromGPQAspanningphysics,chemistry,andbiology; +wepreprocessthedatafollowingpreviouspractice(Chengetal.,2025). +• SCP-Hard(Luetal.,2025;Liuetal.,2025a): Aheld-outsetof50SCPquestionsfilteredsuchthatthebasemodels +(Qwen2.5-1.5BseriesmodelsandLlama3.2-3B-Instructmodel)achievesolve@16=1,containingdisjointquestions +fromtheSCPtrainingdatasets. +• SuperGPQA (Du et al., 2025): a subset constructed from the original SuperGPQA which contains 319 science +questionsand250non-sciencequestions. +• MMLUSCI(Wangetal.,2024): asubsetofMMLUProbenchmarkcontainingallcollege-levelchemistry,physics +andbiologyquestions. +• ScienceBench(Wangetal.,2023): 692college-levelsciencequestions. +• GraphTest: Aheld-outsetof50algorithmicallygeneratedinstancesfromtheQuantumLock andLargestIsland +tasks using Reasoning Gym (Stojanovski et al., 2025), disjoint from training, filtered such that the base models +(Qwen2.5-1.5BseriesandLlama3.2-3B-Instruct)achievePass@16=1. +WealsonotethatGPQA-Diamond,MMLUSCI,andSuperGPQAaremultiple-choicebenchmarks,forwhich +pass@kmaybealessreliablemetric. +Table2detailsthetrainingandevaluationdatasetsacrossthethreereasoningdomains. +15 + +LLMReasoningwithWeakSupervision +Table2.Trainingdatasetsandevaluationbenchmarksacrossthreereasoningdomains. +Domain TrainingSource In-DistributionEval Out-of-DistributionEval +MATH Skywork-OR1 MATH-500,AMC,AIME-2024,AIME- Science Bench, SuperGPQA, GPQA- +2025,MinervaMath,OlympiadBench Diamond,SCP-Hard,MMLUSCI +SCIENCE SCP-116K SCP-Hard,GPQA-Diamond,MMLUSCI, MATH-500, AMC, Minerva Math, +SuperGPQA,ScienceBench OlympiadBench +GRAPH ReasoningGym QuantumLock,LargestIsland MATH-500,AIME-2024,MinervaMath, +GPQA-Diamond,SCP-Hard +PrompttemplateforMATHandGRAPH. +system +YouareahelpfulAIAssistant,designedtoprovidewell-reasonedanddetailedresponses. +YouFIRSTthinkaboutthereasoningprocessstepbystepandthenprovidetheuserwiththeanswer. +Pleaseencloseyourfinalanswerinthebox: \boxed{YourAnswer}. +user + +assistant +Figure8.PrompttemplateusedforRLtrainingandevaluationonMATHandGRAPH.Theplaceholderisreplaced +withtheactualmathematicalquestionduringfine-tuningandevaluation.Specialtokensareomittedforclarity. +B.2.TrainingDataPreparationDetails +Wedescribeourprocedureforconstructingfilteredtrainingdatasetstailoredtoeachmodel’scapabilities. +DifficultyEstimation. Foreachprobleminthesourcedataset,wesample16responsesfromthebasemodelandcount +thenumberofcorrectsolutions,yieldingsolve@16∈[0,16]. Weretainonlyproblemswithsolve@16∈[1,15],excluding +problemsthataretoodifficult(solve@16=0)ortriviallyeasy(solve@16=16)forthemodel. +PrompttemplateforSCIENCE. +system +Let’sthinkstepbystepandoutputthefinalanswerwithin\boxed{}. +user + +assistant +Figure9.PrompttemplateusedforRLtrainingandevaluationonSCIENCE.Theplaceholderisreplacedwiththe +actualmathematicalquestionduringfine-tuningandevaluation.Specialtokensareomittedforclarity. +Stratified Sampling. We use a stratified round-robin selection method to construct training subsets of size N ∈ +{8,32,64,512,2048}. Filteredproblemsarepartitionedinto15bins{B }15 accordingtotheirsolve@16values. Toselect +i i=1 +N problems: +1. Initialization: Setthecurrentcountofselectedproblemsn =0. +total +2. Round-RobinSelection: Whilen 1 during RL training, we observe that ∆(8) keeps the same sign for all k ∈ {1,4,8,16} across most +sat +model-benchmarkpairs,indicatingconsistentimprovementinbothpass@1andpass@k. Thisindicatesthatduringthe +pre-saturationperiod,themodelisnotjustclosingpass@kandpass@1gap. +C.3.AdditionalExperimentalResultsonLargeModels +Inthissection,wewillreportthefullevaluationresultson7Band8Bmodels. +Fig. 20 and Fig. 21 show the results of Qwen2.5-Math-7B and Llama3.1-8B-Instruct models on MATH domain with +19 + +LLMReasoningwithWeakSupervision +Table7.Graph-domaintraining(7B/8B):in-distributionbenchmarks. +| Model | Metric t(8) | QuantumLock | | LargestIsland | +| ----- | ----------- | ----------- | --- | ------------- | +sat +| | | ∆ ∆∗ | G ∆ | ∆∗ G | +| -------------------- | ---------- | --------- | --------- | --------- | +| | | sat post | sat sat | post sat | +| Qwen2.5-Math-7B | Avg@16 151 | 8.0 4.9 | 7.3 19.8 | 1.9 -10.9 | +| | Pass@4 | 22.6 1.5 | -1.3 16.5 | 6.1 16.3 | +| | Pass@8 | 26.8 2.8 | -2.4 10.7 | 9.0 29.9 | +| | Pass@16 | 30.6 5.5 | -4.1 6.8 | 8.6 41.3 | +| LLama3.1-8B-Instruct | Avg@16 29 | 10.1 7.1 | 3.2 1.8 | 1.0 -0.2 | +| | Pass@4 | 15.3 -0.0 | 6.2 1.8 | 0.0 1.6 | +| | Pass@8 | 15.4 7.4 | 12.5 1.8 | 0.0 2.8 | +| | Pass@16 | 20.1 16.0 | 25.0 1.8 | 0.0 3.1 | +in-domainandout-of-domainbenchmarks,respectively. +Fig.22andFig.23presenttheresultsofQwen2.5-Math-7BandLlama3.1-8B-InstructmodelsonSCIENCEdomainwith +in-domainandout-of-domainbenchmarks,respectively. +Fig.25providestheresultsofQwen2.5-Math-7BandLlama3.1-8B-InstructmodelsonGRAPHdomainwithmoreout-of- +domainbenchmarks. +Similar to the observations on smaller models, during the pre-saturation phases, models show generalization on both +in-domainandout-of-domainbenchmarksintermsofpass@kmetrics. Comparedtothe3Bmodel,the8BLlamamodel +exhibitsbettercross-domaingeneralization. However,LlamamodelsstillsaturatemorefasterthanQwenmodelsandshow +cleardatadependence(e.g.,Fig.22onSCIENCE). +D.RewardTypeEffect +D.1.AdditionalResultsonRewardCorruption +Rewardcorruptionimplementation. Foreachcorruptionlevelγ,weuniformlysampleaγ fractionofpromptsfromthe +N =2048trainingsetforeachmodel–domainpair. Foreachselectedprompt,wedraw96modelresponsesattemperature +1.0andselectthemostfrequentlyoccurringincorrectfinalanswer(i.e.,onethatreceiveszerorewardunderourverifier)as +thecorruptedtarget. DuringRLtraining,wereplacetheground-truthlabelsoftheselectedpromptswiththesecorrupted +labels. For Llama models and the GRAPH domain, we cap γ at 0.9 due to the base model’s inability to generate valid +solutionsevenwithextensivesampling. +Results. Fig.2showscomplementaryresultstoSection3.2. Weobservesimilarpatternsthatsomemodelsarerobustto +evenlargeamountsofrewardnoise. Inparticular,Qwenmodelsexhibitgeneralizationabilityevenwhentrainedonalmost +completelycorrupteddata;incontrast,Llamamodelstendtoshowhighrewardcurvesyetpoorergeneralizationtonewdata, +suggestingoverfittingtoincorrectresponses. +D.2.AdditionalResultsonSelf-SupervisedProxyRewards +Proxyrewardsimplementation. Weevaluatetwoself-supervisedproxyrewardsasalternativestoground-truthverification: +majorityvotingandself-certainty. +1. MajorityVotingReward. FollowingTTRL(Zuoetal.,2025),weestimatepseudo-labelsviamajorityvotingand +assignbinaryrewardsbasedonagreementwiththeconsensusanswer. Foreachprompt,wesample16responsesfrom +thepolicymodel. Themostfrequentlyoccurringansweramongthese16responsesisselectedasthepseudo-label. +Rewards are then computed as: r = 1 if the response matches the pseudo-label, and r = 0 otherwise. For policy +optimization,weusethefirst8responsestocomputeadvantages. AllotherRLhyperparametersfollowSectionB.3. +2. Self-Certainty Reward. Following Zhao et al. (2025), we use the model’s own confidence as the reward signal. +Self-certaintyisdefinedastheaverageKLdivergencebetweenauniformdistributionoverthevocabularyandthe +20 + +LLMReasoningwithWeakSupervision +model’snext-tokendistribution: +|o| +1 (cid:88) +r =Self-certainty(o|q):= KL(U∥p (·|q,o )) (1) +|o| πθ 1and0otherwise. +ForadatadistributionD,wedefinetheoverallgenerationdiversityasd (D)=E [Div (x)]. Empirically,wesample +| | | | | π | x∼D π | | | +| --- | --- | --- | --- | --- | ----- | --- | --- | +N =16outputsperpromptandestimated using8promptsfromthespecifieddataset. +π +WedefineFaithfulDiversityasthismetriccalculatedonlyonresponsesthatachieveafaithfulnessscoreof1(seebelow). +Fig.36showsanexampleoftheLM-as-judgeoutputwhenpromptedtoevaluatethesimilarityof2responses. +F.2.Quantificationofreasoningfaithfulness +Inspiredbypriorwork(Bakeretal.,2025),wedefinethefaithfulnessasaresponse’sintermediatereasoningtracecontains +allrelevantinformationandremainslogicallyconsistentwiththepredictedfinalanswer. Eachmodelrolloutproducesa +responseythatcontains(i)areasoningtraceand(ii)afinalanswer. Wewritey =(r,a),whereristhereasoningtextanda +istheextractedfinalanswer. Foreachinputpromptx,wesampley ∼π(·|x)fromthepolicy. +Faithfulnesslabeling. Wedefineadiscretefaithfulnesslabelingfunctions :X ×Y →{0,1,1},wheres (x,y) +| | | | | faithful | | 2 | faithful | +| --- | --- | --- | --- | -------- | --- | --- | -------- | +measurestheinternalagreementbetweenrandainy: +• s faithful (x,y)=1(aligned)ifthereasoningtracerconstitutesacoherentandlogicallysupportivejustificationforthe +producedanswera,regardlessofwhetheraiscorrect; +• s (x,y)= 1 (partiallyaligned)ifrexhibitsaplausibleargumentativetrajectorytowardabutcontainssubstantial +| faithful | 2 | | | | | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | +gaps,unsupportedleaps,orlocalinconsistenciesthatweakenthejustification; +22 + +LLMReasoningwithWeakSupervision +• s (x,y) = 0(misaligned)ifaisnotsupportedbyr, e.g., r contradictsa, failstoaddressthequestion, orthe +faithful +answerappearsasthe“lucky”guess. +Inpractice,weimplements (x,y)byqueryingOpenAIo3(OpenAI,2025a)asanLLM-as-a-judgewithafixedrubric +faithful +(Fig.32). OpenAIo3isusedforthistask,asopposedtoGPT-4o,duetorequiringalargermodelinordertobeableto +accuratelyreasonaboutcomplexmathematicalandscientificstepspresentinthereasoningtraces. Foralabell∈{0,1,1}, +2 +wedefinethefaithfulnessrateofpolicyπoverdatasetDas +F (l) := P (cid:2) s (x,y)=l (cid:3) . +π x∼D,y∼π(·|x) faithful +Attrainingstept,weapproximateF (l)usingN trainingprompts{x }N andK rolloutsperprompt: +πt i i=1 +N K +F(cid:98)πt (l) = +N +1 +K +(cid:88)(cid:88) 1(cid:8) +s +faithful +(cid:0) +x +i +,y +i,k +(cid:1) +=l +(cid:9) +, y +i,k +∼π t -c -e -j -o -r -P - +(·|x i -c -F - +). (3) +i=1k=1 +WeuseN =8promptsandK =16rolloutsperpromptatselectedRLcheckpointsonthespecifiedtrainingdataset. We +reportF(cid:98)πt (l)forl∈{0,1 2 - +,1}tocharacterizethedistributionofreasoningfaithfulnessunderthepolicyπ +t . - -W +Fig.37showsanexampleoftheLM-as-judgeoutputwhenpromptedtoevaluatethefaithfulnessofamodelresponsewhen +trainedontheMATHtrainingdataset. +ReliabilityofLLM-as-a-judge. TomitigatebiasfromusinganLLM-as-a-judgeforfaithfulnessevaluation,weassess +consistency across multiple LLM judges by computing Cohen’s Kappa (Cohen, 1960) across 16 faithfulness-scored +Qwen2.5-Math-1.5Boutputswhentrainedon8samplesfromtheMATHtrainingdatasetatsteps20,120and440. +Thejudgesachievesubstantialagreement(κ=0.752and0.649),indicatingconsistentfaithfulnesslabelingacrossdifferent +models. Weadditionallyconductedasmall-scalemanualevaluationtohuman-checkthefaithfulnessscoresandfindfair +alignmentwiththeLLMjudges. +F.3.Additionalresultsondiversityanalysis +Fig.30showsthesemanticdiversityofLlama3.2-3B-Instruct,Qwen2.5-1.5BandQwen2.5-Math-1.5BontheMATH-500 +evaluationdatasetthroughoutRLtraining. Qwen-Mathexhibitshigherreasoningdiversityoncorrectresponsesthanthe +othermodelsatthelaterstagesoftraining,highlightingthatRLenablesittosuccessfullylearndiverseandreliablestrategies; +coupledwithitsbetterperformanceontheevaluationdataset,thisindicatesstrongergeneralizationproperties. Inparticular, +weobservesignificantlylowerdiversityintheLlama3.2-3B-Instructmodelincomparisontoitsdiversitymetriconthe +trainingdataset(Fig.4),implyingdisagreementbetweentrainingandevaluationdistributionsandfurtherhighlightingthe +limitationsoftrainingdiversityasanindicatorofreasoningcapabilities. +F.4.Additionalresultsonfaithfulnessanalysis +Fig.33showstheproportionofresponsesthatareclassifiedasalignedormisalignedwhencalculatedwithrespecttocorrect, +incorrectorallresponses. Outofallcorrect,incorrectandoverallresponses,bothQwen2.5-1.5BandQwen2.5-Math-1.5B +showhigherproportionofalignedresponsesandlowerproportionofmisalignedresponsesthanLlama3.2-3Bwhentrained +on8samplesfromtheMATHdataset. Qwen2.5-Math-1.5Badditionallyshowsthisresultwhentrainedon8samplesfrom +SCIENCE. +G.Pre-RLIntervention +Fig.34reportpass@16resultsandFig.35reportsresultsonmorebenchmarks. +23 + +LLMReasoningwithWeakSupervision +ExamplepromptandresponseformatofSFT +UserPrompt: +Calculatethelimitofthefunction: +arcsin(3x) +| | lim | √ √ | | | +| --- | --- | ------ | --- | --- | +| | x→0 | 2+x− 2 | | | +Assistant: + +Okay,soIneedtofindthelimit... +Substitutingx=0givestheindeterminateform 0,so... +0 +Rationalizingthedenominator,weget... + +Thelimitiscalculatedasfollows: +1. Recognizetheindeterminateform: substitutingx=0yields +0 +, +0 +whichsuggestsusingrationalizationorl’Hoˆpital’srule. +2. Rationalizethedenominator: +| | √ | √ | √ | √ | +| ---------- | ------ | ------------ | ------- | ------- | +| | | | (cid:0) | (cid:1) | +| arcsin(3x) | 2+x+ | 2 arcsin(3x) | 2+x+ | 2 | +| √ | √ · √ | √ = | | . | +| 2+x− | 2 2+x+ | 2 | x | | +3. Splitthelimit: +| | | √ | √ | | +| --- | ---------- | ---------- | ------- | --- | +| | arcsin(3x) | (cid:0) | (cid:1) | | +| | lim | · lim 2+x+ | 2 . | | +x +| | x→0 | x→0 | | | +| --- | --- | --- | --- | --- | +4. Evaluateeachpart: +• Using +arcsinz +| | | lim =1, | | | +| --- | --- | ------- | --- | --- | z +z→0 +thefirstlimitbecomes3. +√ +• Thesecondlimitevaluatesto2 2. +5. Combinetheresults: √ √ +3·2 2=6 2. +FinalAnswer: +√ +6 2 +Figure12. ExamplepromptandresponseformatofSFT.InThinkingSFT,themodelistrainedwithreasoningtracesenclosedby +and,whereasNon-ThinkingSFTomitsthem. +24 -p - -l -l -i - -) -a - -• - - i +LLMReasoningwithWeakSupervision +| | 8 samples | | 32 samples | | 64 samples | | | 512 samples | | 2048 samples | | +| ------------ | ------------------- | --- | ------------ | --- | ---------- | ------- | --- | ----------- | --------------- | ------------ | --- | +| | draweR gniniarT 1.0 | | )%( 005-HTAM | | | | | | | | | +| | | | 72 | | | 52 | | | )%( draH-PCS 36 | | | +| htaM-5.2newQ | 0.8 | | | | | )%( CMA | | | | | | +| | | | | | | 48 | | | 30 | | | +69 +| | 0.6 | | | | | | | | 24 | | | +| ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| B5.1- | | | | | | 44 | | | | | | +| | | | 66 | | | | | | 18 | | | +| | 0.4 | | | | | 40 | | | | | | +12 +| | 0.2 | | 63 | | | 36 | | | | | | +| ------------ | ----------------- | ------- | --------------- | -------------- | ------- | --- | -------------- | ------- | --------------- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +| | draweR gniniarT 1 | | )%( 005-HTAM 60 | | | | | | | | | +| B5.1-5.2newQ | | | | | | 30 | | | )%( draH-PCS 20 | | | +50 +)%( CMA 24 +15 +| | | | 40 | | | 18 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +10 +| | | | 30 | | | 12 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | | | 20 | | | 6 | | | 5 | | | +| | 0 | | 10 | | | 0 | | | 0 | | | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +draweR gniniarT 1.0 +| B3-5.2newQ | | | )%( 005-HTAM 66 | | | | | | )%( draH-PCS | | | +| ---------- | --- | --- | --------------- | --- | --- | ------- | --- | --- | ------------ | --- | --- | +| | | | | | | 40 | | | 25 | | | +| | 0.8 | | | | | )%( CMA | | | | | | +| | | | 63 | | | 36 | | | 20 | | | +0.6 +| | | | 60 | | | 32 | | | 15 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.4 | | | | | 28 | | | 10 | | | +57 +0.2 +| | | | 54 | | | 24 | | | 5 | | | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +B3-2.3amalL Training Steps Training Steps Training Steps Training Steps +draweR gniniarT +| tcurtsnI- | 1.0 | | )%( 005-HTAM | | | | | | )%( draH-PCS 15 | | | +| --------- | --- | --- | ------------ | --- | --- | ------- | --- | --- | --------------- | --- | --- | +| | | | 54 | | | 30 | | | | | | +| | 0.8 | | | | | )%( CMA | | | 12 | | | +| | 0.6 | | 51 | | | 25 | | | 9 | | | +| | | | 48 | | | 20 | | | 6 | | | +0.4 +| | | | | | | 15 | | | 3 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.2 | | 45 | | | | | | | | | 0 - 1 - -g +10 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure13.ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonMATHdomain.Resultsareaveraged +overthreeindependentruns,withshadedregionsindicatingerrorbars. Verticaldashedlinesdenotethesaturationstepforeachdata +scaleifitsaturatesbefore496gradientsteps.Llamamodelsexhibitrapidsaturationinsmall-sampleregimesandrelyheavilyondata +scale.Incontrast,Qwenmodelsyieldcomparableperformanceacrossvaryingsamplesizes,characterizedbyextendedsaturationperiods. +Evaluationresultsinthisfigurearebasedongreedydecoding. +25 + +LLMReasoningwithWeakSupervision +| | 8 samples | | 32 samples | | 64 samples | | | 512 samples | | 2048 samples | | +| --- | --------- | --- | ---------- | --- | ---------- | --- | --- | ----------- | --- | ------------ | --- | +)%( dnomaiD AQPG +draweR gniniarT 1.0 +| | | | )%( draH-PCS 40 | | | 30 | | | )%( 005-HTAM | | | +| ------------ | --- | --- | --------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | +| htaM-5.2newQ | | | | | | | | | 72 | | | +| | 0.8 | | 35 | | | 25 | | | | | | +| | 0.6 | | | | | 20 | | | 69 | | | +| B5.1- | | | 30 | | | | | | | | | +| | 0.4 | | | | | 15 | | | | | | +| | | | 25 | | | | | | 66 | | | +10 +| | 0.2 | | 20 | | | | | | | | | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +| | | | | | | 5 | | | 63 | | | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +| | draweR gniniarT | | | | | 30 | | | | | | +| ------------ | --------------- | --- | --------------- | --- | --- | --- | --- | --- | --------------- | --- | --- | +| | 1 | | )%( draH-PCS 30 | | | | | | )%( 005-HTAM 60 | | | +| B5.1-5.2newQ | | | | | | 24 | | | | | | +| | | | 24 | | | | | | 50 | | | +| | | | 18 | | | 18 | | | 40 | | | +| | | | 12 | | | 12 | | | 30 | | | +| | | | 6 | | | 6 | | | | | | +20 +| | 0 | | 0 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +draweR gniniarT +| B3-5.2newQ | 1.0 | | )%( draH-PCS | | | | | | )%( 005-HTAM | | | +| ---------- | --- | --- | ------------ | --- | --- | --- | --- | --- | ------------ | --- | --- | +| | | | 30 | | | 33 | | | 66 | | | +0.8 +63 +| | 0.6 | | 24 | | | 30 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 18 | | | 27 | | | 60 | | | +0.4 +| | | | 12 | | | 24 | | | 57 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.2 +| | | | 6 | | | 21 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +54 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | ----- | ------- | --- | --- | ------- | --- | --- | ------- | --- | --- | ------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +B3-2.3amalL draweR gniniarT +| tcurtsnI- | 1.0 | | | | | | | | )%( 005-HTAM | | | +| --------- | --- | --- | ------------ | --- | --- | --- | --- | --- | ------------ | --- | --- | +| | | | )%( draH-PCS | | | 30 | | | 54 | | | +| | 0.8 | | 32 | | | | | | | | | +| | | | | | | 27 | | | 52 | | | +24 +| | 0.6 | | | | | 24 | | | 50 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| | 0.4 | | | | | 21 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 8 | | | | | | 48 | | | +18 +| | 0.2 | | | | | | | | 46 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure14. ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonSCIENCEdomain. Resultsare +averagedoverthreeindependentruns,withshadedregionsindicatingerrorbars.Verticaldashedlinesdenotethesaturationstepforeach +datascale.Thepre-saturationphaseyieldssimilargainsacrossallsamplesizes;however,afterthesaturationpoint,largersamplesizes +demonstratedistinctbenefits.Modelsexhibitsignificantlydifferentsaturationdynamicsonsmallsamples.Evaluationresultsinthisfigure +arebasedongreedydecoding. +| | 8 samples | | 32 samples | | 128 samples | | | 256 samples | | 882 samples | | +| --- | --------- | --- | ---------- | --- | ----------- | --- | --- | ----------- | --- | ----------- | --- | +htaM-5.2newQ +| | draweR gniniarT 1 | | )%( kcoL mutnauQ 50 | | | )%( dnalsI tsegraL | | | 82 | | | +| --- | ----------------- | --- | ------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | +)%( 005-HTAM +| | | | 40 | | | 32 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +80 +| B7- | | | | | | 24 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 30 | | | | | | 78 | | | +| | | | 20 | | | 16 | | | | | | +76 +8 +10 +74 +| | 0 | | 0 | | | 0 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +)%( kcoL mutnauQ +| B8-2.3amalL | draweR gniniarT 1.0 | | | | | )%( dnalsI tsegraL | | | | | | +| ----------- | ------------------- | --- | --- | --- | --- | ------------------ | --- | --- | ------------ | --- | --- | +| tcurtsnI- | | | | | | 6 | | | )%( 005-HTAM | | | +| | | | 45 | | | | | | 54 | | | +0.8 +4 +0.6 +| | | | 30 | | | | | | 52 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.4 | | | | | 2 | | | | | | +| | | | 15 | | | | | | 50 | | | +0.2 +| | | | 0 | | | 0 | | | | | | +| --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | --- | -------------- | ------- | +| | 0 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure15. ComparisonsofRLtrainingdynamicsandperformanceacrossdifferentmodelsonGRAPHdomain. Weuselarger +models(Qwen2.5-Math-7B,Llama-3.1-8B-Instruct)duetoincreasedtaskdifficulty.Resultsareaveragedoverthreeindependentruns, +withshadedregionsindicatingerrorbars.Verticaldashedlinesdenotethesaturationstepforeachdatascale.Qwenmodelalsosaturates +fasterherethaninotherdomains.Largerdatasetsyieldcleargainsinthepost-saturationphases.Evaluationresultsinthisfigurearebased +ongreedydecoding. +26 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +88 +80 +| 60 | | | | | | | | | 88 | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 005-HTAM | | | | | | 80 | | | | | | +70 +| 45 | | | | | | 72 | | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| | | | | | | 64 | | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +| | | | 50 | | | 56 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +64 +15 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 40 | | | 60 | | | 70 | | | 80 | | | +50 +| 30 | | | | | | 60 | | | 70 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +CMA +| | | | 40 | | | 50 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 20 | | | | | | | | | 60 | | | +| | | | 30 | | | 40 | | | 50 | | | +10 +| | | | 20 | | | 30 | | | | | | +| --------------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| htaM avreniM 24 | | | | | | | | | 48 | | | +| | | | 36 | | | 42 | | | | | | +| 18 | | | 30 | | | 36 | | | 42 | | | +24 +| 12 | | | | | | 30 | | | 36 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 18 | | | 24 | | | 30 | | | +6 +| | | | 12 | | | 18 | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +hcneBdaipmylO 30 +| | | | 40 | | | 48 | | | 56 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +24 +| | | | 32 | | | 40 | | | 48 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +18 +| | | | | | | 32 | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 12 | | | 24 | | | | | | | | | +| | | | 16 | | | 24 | | | 32 | | | +6 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +40 +15 +| 4202 EMIA | | | 24 | | | 32 | | | 40 | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +10 +| | | | 18 | | | 24 | | | 30 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 5 | | | 12 | | | 16 | | | | | | +20 +| 0 | | | 6 | | | | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | | | | 8 | | | 10 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +10.0 +| | | | 16 | | | 20 | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5202 EMIA 7.5 +12 +| | | | | | | 15 | | | 18 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5.0 +| | | | 8 | | | 10 | | | 12 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +2.5 +| | | | 4 | | | 5 | | | 6 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.0 +| | | | 0 | | | 0 | | | 0 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure16.Fullin-domainbenchmarkevaluationresultsfortheMATHdomainacrossmultiplemodels.Verticaldashedlinesdenote +thesaturationstepforeachdatascale. +27 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +90 +| dnomaiD AQPG | | | 60 | | | 75 | | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 24 | | | | | | | | | 75 | | | +| | | | 45 | | | 60 | | | | | | +| 16 | | | | | | 45 | | | 60 | | | +30 +| 8 | | | | | | 30 | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 32 | | | | | | | | | 80 | | | +| | | | 45 | | | 60 | | | | | | +draH-PCS 24 +60 +45 +| 16 | | | 30 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 30 | | | 40 | | | +| 8 | | | 15 | | | | | | | | | +| | | | | | | 15 | | | 20 | | | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | 25 | | | 30 | | | | | | +16 +hcneB ecneicS +| | | | 20 | | | 25 | | | 30 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| | | | | | | 20 | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 8 | | | 15 | | | | | | | | | +| | | | | | | 15 | | | 18 | | | +| 4 | | | 10 | | | | | | | | | +10 +| | | | 5 | | | | | | 12 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 60 | | | 80 | | | | | | | | | +| | | | | | | 80 | | | 90 | | | +ICS ULMM +| 45 | | | 60 | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | 40 | | | | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +45 +| 15 | | | 20 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| 0 | | | | | | | | | 30 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +60 +20 +| AQPGrepuS | | | 40 | | | | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 45 | | | 60 | | | +| 15 | | | 30 | | | | | | | | | +| 10 | | | | | | | | | 45 | | | +| | | | 20 | | | 30 | | | | | | +| 5 | | | | | | | | | 30 | | | +| | | | 10 | | | 15 | | | | | | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure17.Fullin-domainbenchmarkevaluationresultsfortheSCIENCEdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +28 -5, -g - -1 - -dVIA41 A3AWS S302:10S3 1V1if1.1031.1H3lV ONOISIH - -SIN :37VOS - -— - -I - -g - -§ - - all significant resources within the APE; and (3) to evaluate these resources according to -the National Register of Historic Places criteria. - -The survey methodology consisted of a field survey and background research on the -project area. A NCDOT staff architectural historian conducted a field survey in August -and October 1999, by car and on foot. All structures over 50 years of age in the APE -were photographed and keyed to an area map. - -Background research was conducted at the Western office of the State Historic -Preservation Office, Asheville; the North Carolina State Library, Raleigh; and the Burke -County Court House in Morganton. - -A meeting was held with Ms. Jennifer Martin at the State Historic Preservation Office on -August 20, 1999 to discuss the eligibility of two of the historic resources, the William -Jones Arney House and the Marshall Arney House. It was agreed at that time the two -properties were potentially eligible for the National Register of Historic Places. - -SUMMARY FINDINGS OF THE SURVEY - -The project proposes to pave and improve SR 1424, Johns River Road and SR 1438, -Corpening Bridge Road in Quaker Meadows and Upper Creek townships, Burke County. -In letters dated July 9, 1999 and August 10, 1999, the State Historic Preservation Office -identified four structures of historical or architectural importance within the general -project area, the Marshall Arney House, the William Arney House, the Arney Chapel and -the Corpening House (See Appendix A). Two additional historic properties, not -mentioned in the letters, are located within the project area: a late nineteenth-century log -house and an early twentieth-century frame house. No properties listed on the National -Register were located within the APE for the project. Of the four historic properties -mentioned in the letter, one, the Arney Chapel, is located outside the APE. Of the five -historic properties located within the APE, three—the Corpening House, the nineteenth- -century log house, and the early twentieth-century frame house --were determined not -eligible for the National Register. The Marshall Arney House and the William Arney -House were determined to be eligible for the National Register. Criterion Consideration -G, for properties that have achieved significance within the last fifty years, states that -properties less than fifty years of age may be listed on the National Register only if they -are of exceptional importance or if they are integral parts of districts eligible for the -National Register. There are no properties in the APE that qualify for the National -Register under Criterion Consideration G. - -Properties Listed on the National Register: - -None - -Ilistoric Architectural Resources Survey Report: iVCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nancy Ian Dolsen - -Page 4 - - Properties Listed on the North Carolina State Study List: - -None - -Properties Evaluated and Considered Eligible for the National Register: - -William Jones Amey House, page 8 -Marshall Amey House, page 15 - -Properties Evaluated and Considered Not Eligible for the National Register: - -Corpening House, page 21 -Twentieth-century frame house, page 24 -Nineteenth-century log house, pay 28 - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 5 - - HISTORIC CONTEXTS AND BACKGROUND INFORMATION - -During the late seventeenth century, occupation of the land of present-day Burke County -was contested between two native-American tribes, the Cherokee and the Catawba. The -area eventually became a buffer zone between the two tribes. The French and Indian War -(1754-1763) and the American Revolution had a catastrophic effect on both tribes, -forcing the Cherokees further into the mountains, and reducing the Catawba population to -250 members by 1784. - -Transient hunters and trappers of European descent settled in the county during the -middle of the eighteenth century. In 1777 the General Assembly approved the formation -of a new county named Burke in the western portion of the state that encompassed all or -part of 16 present counties. Burke was named for Thomas Burke (1747-1783), a -member of the Continental Congress and third Governor of North Carolina.' Burke -County achieved its present size of 511 square acres in 1885. - -According to an architectural survey of the county completed in 1983-1984, there are -three distinct regions in the county: the rich bottom lands along the Catawba, Johns, -Upper and Lower Creeks, and Warrior Forks rivers; the remote dead-end coves of the -South Mountains and Blue Ridge; and the rolling foothills interspersed with small -valleys.2 The subject project is located along the bottom land of the Johns River in -Quaker Meadows and Upper Creek townships. - -The bottom lands contained the richest agricultural soil in the county, and were the site of -antebellum plantations. Most of the county's large, brick plantation homes dating to the -first half of the nineteenth century stand along the Johns and Catawba rivers in Quaker -Meadows and Upper Creek townships, including Bellevue (1823-1826), Cedar Grove -(1825), and Pleasant Valley (ca. 1820; ca. 1975). In Burke, the typical plantation of 1860 -consisted of 850 acres of improved land and thousands of acres of unimproved land. -The typical 1860 Burke County farm, however, contained only 60 acres of cultivated land -and only 60 out of 1,239 families owned more than 100 acres of improved land.3 - -After the Civil War, the land-use pattern of ownership changed. The large-scale -plantations were dissolved, and there was an increase in the number of small-scale farms, -from 500 in 1860 to 1,648 in 1880. At the same time, the average farm size dropped to -27 acres.4 - -The project area is the site of five post-bellum houses that date to this period of -subdividing larger plantation properties. The houses stand on land first owned by one of -Burke County's first European settlers, John Perkins, who first explored the county in the - -I Entry for Thomas Burke, Biographical Directory of the United States Congress, I774-Present, Website: -Bioguide.conress.gov. -2 J. Randall Cotton, et at., Historic Burke: An Architectural Sites Inventory of Burke County (Morganton, -NC: Historic Burke Foundation, Inc., 1987) 13. -3 Cotton, et at, 23. -4 Cotton, et al, 27. - -Historic Architectural Resources Survey Report: NCDOT. SR 1424. Burke County: November 1999 -Nancy Van Dolsen - -Page 6 - - 1750s. These five houses represent a range of building types constructed in the county -during the last quarter of the nineteenth century: two frame T-shaped houses, a small -two-room log house, and two square-shaped two-story frame dwellings. - -At the end of the twentieth century, much of the land in the county is held by three -property owners: the US Government (Pisgah Forest), the state (South Mountain State -Park) and Carolina Centers, a timber company. Carolina Centers presently owns the land -upon which the Corpening, Marshall Amey, and William Jones Arney houses stand. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 7 - - PROPERTY INVENTORY AND EVALUATIONS - -William Jones Arney House - -Location: (cid:9) -The William Jones Arney House is located on the northwest corner -of the junction of SR 1424 and SR 1438. The house sits on a 23,136.10-acre tract -owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a timber -company. The property was formerly owned by Duke Power, Inc. - -The William Jones Arney House sits within a cleared area, at the - -Setting: (cid:9) -top of a rise, facing east to Johns River. A short dirt lane leading from Johns -River Road runs south of the house, ending at a small frame granary west of the -dwelling. - -Description: - -Dwelling: The William Jones Arney House, dating to ca. 1890, is a two- -and-a-half story frame residence with a T-shaped plan. The exterior of the -house is clad with narrow weatherboards, and a metal roof covers the -building. To the rear of the building is a kitchen connected to the main -house by a covered walkway. The kitchen stands one-story and has a -shed-roofed porch on each eave elevation. The windows in the kitchen -contain six-over-six sash and those in the main block of the house have -two-over-two sash. A one-story tripartite bay window adorns the south -gable end wall of the ell. Pointed arched louvered windows are located at -the peaks of the gable end walls. Two interior brick chimneys are located -at the center of the ells, providing an interior stack to each room in the -house. - -The interior walls of the house are sheathed in horizontal boards, as are the -ceilings. The house (excluding the kitchen) has four rooms on the first -and four rooms on the second floor. - -Well House: A concrete block well house stands south of the south porch -of the kitchen. The well house is approximately 3 feet in height and is -covered with a corrugated metal roof. - -Granary: To the west of the house stands a frame granary also constructed -ca. 18900. The granary building is a gable-front building with a central -door. The building is weatherboarded and has a metal roof. An open side -shed was constructed onto the east eave wall, and a shed roof porch is -located on the south elevation. - -Background Information: (cid:9) -land ca. 1885. Arney, a descendent of one of the pioneering families in the -county, both farmed and owned a general store (now demolished). The store - -William Jones Arney (1849-1911) moved to this - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy ran Dolsen - -Page 8 - - stood to the west, at the intersection of Corpening Bridge and Piedmont roads. -Arney, and his wife Margaret McCall, owned over 500 acres in the area along -John's River. The couple had nine children? - -At the death of William and Margaret Arney, ownership of the part of the farm -upon which this house stood was transferred to George Phifer Arney who sold -240 acres to Western Carolina Power for $100 in 1922.6 The property is presently -owned by Carolina Centers, LLC. - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966, as amended, the William Jones Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A, since it is not representative of a significant trend in the history of the -area. The property is no longer under cultivation, and all but one of the -agricultural outbuildings have been demolished or removed. - -The William Jones Arney House is considered not eligible for the National -Register of Historic Places under Criterion B because it is not associated with any -individuals of specific historical significance. - -The William Jones Arney House is considered eligible for the National Register -under Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is a regional form built by farmers of middle economic status. The Arney -House is a good representation of this form, and exhibits excellent integrity and -retention of setting. The granary building is also representative of a vernacular -building form, and exhibits good integrity of design, materials, workmanship, -and setting. - -The building is not likely to yield information important to the history of building -technology; therefore it is not eligible for the National Register under Criterion D. - -Boundaries: The proposed National Register boundaries comprise approximately -2.8 acres that surround the house and its accompanying outbuilding. The land -was deeded to George Phifer Arney, and is the portion of the William Jones -Arney estate upon which the subject structures stand. This parcel is part of parcel -JR-21, which is a portion of the larger Carolina Centers tract. - -Boundary Justification: (cid:9) -include the house, outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. - -The proposed historic boundaries were drawn to - -5 William Jones Amey House, Historic Resource Survey Form, Burke County, 1983. Asheville, NC: -Western Office, Division of Cultural Resources. -6 Burke County Deed Book T4:293. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 -Nancy Van Dolsen - -Page 9 - - Figure 3: Tax Map, William Jones Arney House - -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 10 - - Figure 4: Site plan, William Jones Arney House - -Historic Architectural Resources Survey Report,- NCD07; SR 1-114, Burke County: November 1999 (cid:9) -Nancy l'an Dolsen - -Page 11 - - — - -— (cid:9) - -_ - -- (cid:9) - -41E - -re-7_. - — -liq -- -- . (cid:9) - -- - -_ - -_ (cid:9) - -_ - -_- - -- --- - -— s- — - -_--..------ - -- - -- (cid:9) -77--- - --_ - -- (cid:9) - -—. (cid:9) - -—______-- -- ------_--,- --- ---------- - -_ - -_ --- — - -- -- - --4-. - -, - --4 - ----_ - -a - -Photograph 1: William Jones Amey House, facing southwest - -Photograph 2: William Jones Amey House, facing northeast - -Historic Architectural Resources Survey Report: NCDOT SR 1424, Burke County: November 1999 -Nancy Van Dolsen - -Page 12 - -(cid:9) -(cid:9) - . - -:t. (cid:9) - -. --- - -, - -,,J-, (cid:9) - -, - -••••- (cid:9) - -, (cid:9) - -i - --- (cid:9) -_ -__ - --.------ (cid:9) - -----.•-•_:, (cid:9) ----- - -(cid:9) ,- - --.? - ---- -_= - -- (cid:9) - -_ - -- - -. (cid:9) - --- (cid:9) -- (cid:9) - -- -•- -_ --4--.- -— -- (cid:9) - -------,_ - -- - --_ -_.----e- --, (cid:9) - (cid:9) --- (cid:9) - -l'---0` (cid:9) - -_ (cid:9) -_ - -_ - -_ - -- (cid:9) - -- (cid:9) - -- - -- --7---- - --__-- (cid:9) -------- -- (cid:9) - -- (cid:9) - (cid:9) - -- -- - -- (cid:9) - -- -- - -.-• =,---• -.-__----------- - -- (cid:9) - (cid:9) - -- - -, - -t t -, (cid:9) -- (cid:9) -- (cid:9) -= (cid:9) - -_..----, (cid:9) - -. (cid:9) - -_ - -- (cid:9) - -Photograph 3: William Jones Arney house, view facing northeast - -Photograph 4: Granary, view facing north - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 (cid:9) -:Vancy Van Dolsen - -Page 13 - -(cid:9) - " -. -• - -: -• -• - -0 - -. - -0 - -c -e -) - -z - -C - -. -) - -E - -- -: - -6: -1 - -;1 - -8 - -4 -. -1 - - Marshall Arney House - -The Marshall Arney House is located on the west side of SR 1424. - -Location: (cid:9) -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. - -The Marshall Arney House stands in a clearing of a wooded area - -Setting: (cid:9) -halfway up a hillside on a knoll approximately 100 feet wide in a bend of the -Johns River. Johns River Road is located on the next downhill terrace. East of -the roadway the land slopes down to the river. - -Description: - -Dwelling: Built ca. 1900, the Marshall Arney House is a two-and-a-half- -story, T-shaped house and features a plan similar to his father's house, the -William Jones Arney House. The primary elevation faces south. The -Marshall Arney House has a wrap-around porch with chamfered posts. -Similar to the William Jones Arney House, the Marshall Arney House has -pointed arched louvred windows at the peaks of the end gables. The end -gables are pedimented. Two early twentieth-century additions were -constructed onto the north gable end; a two-story gable wing, and a one- -story shed-roof ell. The building is weatherboarded, and retains most of -its original two-over-two sash windows. - -Springhouse: To the west of the house stands a gable-front springhouse -banked into the side of the hill. The springhouse has a stone lower level -with a stuccoed finish. The interior of the upper floor is sheathed in -diagonal boards, and the exterior is weatherboarded. The gable roof is -cantilevered over the façade of the springhouse. - -Background Information: (cid:9) -Constructed by Marshall Arney on land bequeathed -to him by his father, William Jones Amey, this house is a substantial dwelling for -the area. Marshall Arney was a land agent for Duke Power, served for twenty- -five years on the Burke County School Board and was instrumental in -consolidating and erecting new schools in the county during the early twentieth -century. In April 1922 Marshall Arney and his wife, L. Inez Arney, sold 250 -acres, including the subject structure, to Western Carolina Power. - -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, the Marshall Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A. since it is not representative of a significant trend in the history of the -area. - -historic Architectural Resources Survey Report: NCDOT. SR I424. Burke County; November 1999 (cid:9) -Nancy i'an Dolsen - -Page 15 - - The Marshall Amey House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. - -The Marshall Arney House is considered eligible for the National Register under -Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is found in the region as a house type built by farmers of middle -economic status and part-time farmers with jobs off of the farm. This type of -farmhouse, if ever once prevalent, has become rare due to the large tracts of -former agricultural land that has been acquired for forest and timber land. The -Amey House is a good representation of this type of farm house and exhibits -integrity of materials, workmanship, setting, location, feeling, and association. -The springhouse building is also representative of a vernacular building form and -exhibits good architectural integrity. - -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - -Boundaries: The proposed National Register boundaries comprise -approximately 1.9 acres that surround the house and its accompanying -outbuilding. The land was deeded to Marshall Arne y at death of his father. This -tract is part of parcel JR-18, which is a portion of the larger Carolina Centers -tract. - -Boundary Justification: (cid:9) -include the house. outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. - -The proposed historic boundaries were drawn to - -Historic Architectural Resources Survey Report: ACDOT SR 1424. Burke County: November 1999 (cid:9) -Nancy Ilzn Dolsen - -Page 16 - - Figure 5: Tax Map, Marshall Arney House - -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke Counr.t...,- November 1999 (cid:9) -Nancy ran Dolsen - -Page 17 - - L -T - -. - -m -u - -l - -t - -4 -C - -C -a -) - -' -• - -C - -• - -• -• -• -• -• - -• -• -• - -c -. - -.VC1)Or SR 1414, Bur ke Con - - :Vovember 19 99 - - Corpening House - -The Corpening House stands on the west side of SR 1424. about - -Location: (cid:9) -1.5 miles south of the junction with SR 1405. The house sits on a 23,136.10-acre -tract owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a -timber company. The property was formerly owned by Duke Power, Inc. - -The Corpening House stands on a slight rise overlooking the wide - -Setting: (cid:9) -bottomland along a bend in the Johns River. The land around the house is now -used to grow trees and shrubs for sale to nurseries and landscaping companies. - -Description: - -The ca. 1900 features a common form for the period: a two- - -Dwelling: (cid:9) -story, square, boxy building with a pyramidal roof. The facade features a -large central pediment over the second-floor windows, and a smaller -pediment covers the center of the one-story porch across the facade. -Turned posts support the porch roof. The exterior of the house is clad in -narrow weatherboards. The window sashes have been replaced with one- -over-one double-hung sash. Since 1983, the porch has lost its decorative -brackets, and the back porch has been removed andior remodeled. A shed -roof addition covers the rear elevation. No outbuildings stand on the -property. - -Background Information: (cid:9) -the holdings of John Perkins. one of the county's first settlers of European -descent. The property is now part of the large landholdings of the Carolina -Centers corporation. - -The Corpening House stands on land once part of - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, the Corpening House is considered not eligible for the National -Register ofl-listoric Places under Criterion A, since it is not representative of a -significant trend in the history of the area. - -The Corpening House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. - -The Corpening House is considered not eligible for the National Register under -Criterion C since the building has lost integrity and does not embody distinctive -characteristics of a significant building type. The building no longer retains its -original porch trim, window sash, and the form has been compromised by the -alterations to the rear of the dwelling. - -Historic Architectural Resources Survey Report: NCDOT1SR 1424, Burke County: November 1999 (cid:9) -Nancy lint Dolsen - -Page 21 - - The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - - -Historic Architectural Resources Survey Report; NCD071SR 1424, Burke County: November 1999 -Nancyl'an Dohen - -Page 22 - - Photograph 1: Corpcning House, facing east - -Photograph 2: Corpening House, facing north - -Historic Architectural Resources Survey Report: NCD07; SR 1424, Burke County: November 1999 (cid:9) -Nancy ("an Dohen - -Page 23 - - Early Twentieth-Century Frame House - -This frame house stands on the east side of Corpening Bridge - -Location: (cid:9) -Road. SR 1438. approximately 1.5 miles east of the bridge over the Johns River. -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. - -Setting: (cid:9) -outbuildings stand south of the house, down a slope, and are no longer used. - -This frame house sits in a wooded area that was once farmed. The - -Description: - -A two-and-a-half story, three-bay, double-pile dwelling, - -this house features a hip roof with a central gable dormer. The house has -an exposed extended rafter roof and is covered with weatherboard. A one- -story hip-roof porch covers the facade of the building. The house retains -its six-over-six sash windows. A one-story gable roof ell is located on the -rear of the house. - -Outbuildings: To the east of the house stands a frame domestic -outbuilding. The gable-front building stands on a cement block -foundation. and is covered with weatherboard. An off-center door is -located on the gable end. To the south of the house are three frame -outbuildings in poor condition. The outbuildings have metal roofs and -narrow board siding. The buildings appear to have served as a granary, -stable, and a workshop. - -For purposes of compliance with Section 106 of the National - -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, this house is considered not eligible for the National Register of -Historic Places under Criterion A, since it is not representative of a significant -trend in the history of the area. - -This house is considered not eligible for the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. - -This house is considered not eligible for the National Register under Criterion C -since the building does not embody distinctive characteristics of a significant -building type. The outbuildings have lost their architectural integrity and their -setting. - -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen - -Page 24 - - Photograph I: Early Twentieth-Century Frame House, facing northeast - -Photograph 2: Early Twentieth-Century Frame House, facing north - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Do/sell - -Page 25 - - Photograph 3: Frame Domestic Outbuilding, facing northeast - -Photograph 4: Frame Granary, facing south - -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 -Nancy I'an Dolsen - -Page 26 - - 0 - -t -t -j - -0 -. - -k -. -r -; - -C - -I -) - -- -0 - -. -2 -4 - -- -a - -r -/ -1 - -0 - -C - -. -. - -, - -0 +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-1.5B | | Qwen2.5-1.5B | | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| ------------ | --------- | ----------------- | --- | ------------ | --- | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| | | | 60 | | | | | | 90 | | | +| dnomaiD AQPG | | | | | | 75 | | | | | | +24 +| | | | 45 | | | 60 | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +18 +| | | | 30 | | | 45 | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| 6 | | | | | | 30 | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 20 | | | | | | 60 | | | 80 | | | +40 +draH-PCS +| 15 | | | 30 | | | 45 | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 10 | | | | | | 30 | | | | | | +| | | | 20 | | | | | | 40 | | | +5 +| | | | 10 | | | 15 | | | 20 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 16 | | | 24 | | | 28 | | | 32 | | | +hcneB ecneicS +| | | | | | | 24 | | | 28 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 12 | | | 20 | | | | | | | | | +| 8 | | | 16 | | | 20 | | | 24 | | | +| | | | | | | 16 | | | 20 | | | +| 4 | | | 12 | | | | | | | | | +| 0 | | | 8 | | | 12 | | | 16 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 60 | | | 80 | | | | | | | | | +| | | | | | | 80 | | | 90 | | | +ICS ULMM +| 45 | | | 60 | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | | | | | | | 60 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +40 +| 15 | | | | | | | | | 45 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +20 +| | | | | | | 20 | | | 30 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 0 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure18.Fullout-of-domainbenchmarkevaluationresultsfortheMATHdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +| | | Qwen2.5-Math-1.5B | | | Qwen2.5-1.5B | LLama3.2-3B-Instruct | | N=8 | N=2048 | | | +| --- | --------- | ----------------- | --- | --------- | ------------ | -------------------- | --------- | --- | ------ | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +90 +| 60 | | | 75 | | | | | | 90 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +005-HTAM +| | | | | | | 75 | | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 45 | | | 60 | | | | | | | | | +| 30 | | | | | | 60 | | | 70 | | | +45 +60 +| 15 | | | 30 | | | 45 | | | | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +60 +| 40 | | | | | | | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| 30 | | | 45 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| CMA | | | | | | | | | 60 | | | +45 +20 +| | | | 30 | | | | | | 45 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 10 | | | | | | 30 | | | | | | +| | | | 15 | | | | | | 30 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +40 +48 +| htaM avreniM 24 | | | | | | 40 | | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +32 +| 18 | | | | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +32 +24 +| 12 | | | | | | 24 | | | 32 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| 6 | | | | | | | | | 24 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +8 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +30 +| hcneBdaipmylO | | | | | | 50 | | | 60 | | | +| ------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 24 | | | | | | | | | 50 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 18 | | | 30 | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +12 +| | | | 20 | | | | | | 30 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 6 | | | | | | 20 | | | | | | +| | | | 10 | | | | | | 20 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure19.Fullout-of-domainbenchmarkevaluationresultsfortheSCIENCEdomainacrossmultiplemodels.Verticaldashedlines +denotethesaturationstepforeachdatascale. +29 + +LLMReasoningwithWeakSupervision +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 | N=2048 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | ---- | --------- | ------ | ---- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| 80 | | | 93 | | | 95.0 | | | 97.5 | | | +90 +| | | | 87 | | | 92.5 | | | 95.0 | | | +| ----------- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | --- | +| 005-HTAM 70 | | | | | | 90.0 | | | 92.5 | | | +84 +| 60 | | | 81 | | | 87.5 | | | 90.0 | | | +| --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | --- | +| | | | 72 | | | 80 | | | | | | +84 +| 50 | | | 68 | | | 76 | | | 80 | | | +| --- | --- | --- | ----- | --- | --- | ------- | --- | --- | ----- | --- | ------- | +| | | | 64 | | | 72 | | | | | | +| | | | | | | 68 | | | 76 | | | +| 40 | | | 60 | | | | | | 72 | | | +| | | | 56 | | | 64 | | | | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 60 | | | | | | | | | 92 | | | +| | | | 76 | | | 82.5 | | | | | | +| | | | 7 2 | | | 8 0 . 0 | | | 8 8 | | | +| | | | | | | 7 7 . 5 | | | 8 4 | | | +| 45 | | | 6 8 | | | 7 5 . 0 | | | 8 0 | | | +| CMA | | | 64 | | | | | | | | | +| | | | | | | 72.5 | | | 76 | | | +| 30 | | | 4 5 | | | 54 | | | 66 | | | +4 0 +| | | | 35 | | | 48 | | | 60 | | | +| ------------ | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 15 | | | 30 | | | 42 | | | 54 | | | +| | | | 25 | | | 36 | | | 48 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 32 | | | | | | | | | 57 | | | +| htaM avreniM | | | | | | 51 | | | | | | +| 28 | | | 44 | | | | | | | | | +| | | | | | | 48 | | | 54 | | | +| 24 | | | 40 | | | | | | | | | +| | | | | | | 45 | | | 51 | | | +20 +36 +| | | | | | | 42 | | | 48 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +16 +| | | | 32 | | | 39 | | | | | | +| ---------------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ------- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| hcneBdaipmylO 40 | | | 56 | | | 63 | | | 70.0 | | | +| | | | 5 2 | | | 6 0 | | | 6 7 . 5 | | | +6 5 . 0 +| 32 | | | 4 8 | | | 5 7 | | | 6 2 . 5 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | +| | | | 4 4 | | | 5 4 | | | | | | +| 24 | | | | | | 5 1 | | | 6 0 . 0 | | | +| | | | 32 | | | 40 | | | 48 | | | +36 +| 16 | | | 28 | | | | | | 44 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| | | | 24 | | | 32 | | | 40 | | | +| | | | 20 | | | 28 | | | 36 | | | +| 8 | | | | | | | | | 32 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 32 | | | | | | 54 | | | | | | +60 +| 4202 EMIA | | | 40 | | | 48 | | | 55 | | | +| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 24 | | | | | | 42 | | | 50 | | | +| | | | 30 | | | 36 | | | | | | +45 +| 16 | | | | | | | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | 20 | | | 24 | | | 32 | | | +| 8 | | | | | | 18 | | | | | | +| | | | 10 | | | 12 | | | 24 | | | +16 +| 0 | | | | | | 6 | | | 8 | | | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +24 +| | | | | | | 32 | | | 40 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +5202 EMIA 12 +| | | | 18 | | | 24 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +8 +| | | | 12 | | | 16 | | | 20 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 4 | | | 6 | | | | | | | | | +| | | | | | | 8 | | | 10 | | | +| 0 | | | 0 | | | 0 | | | 0 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure20.Fullin-domainbenchmarkevaluationresultsfortheMATHdomainon7Band8Bmodels. +30 + +LLMReasoningwithWeakSupervision +| | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 N=2048 | | | | +| --- | --------- | --------------- | --------- | -------------------- | --- | ---------- | --- | ---------- | --- | +| | Avg@16(%) | | Pass@4(%) | | | Pass@8(%) | | Pass@16(%) | | +64 +| dnomaiD AQPG 30 | | | | | 80 | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | 56 | | | | | 88 | | | +| 25 | | | | | 72 | | | | | +80 +| 20 | | 48 | | | 64 | | | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 15 | | 40 | | | 56 | | 72 | | | +| 10 | | 32 | | | 48 | | 64 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +56 +| 24 | | | | | 72 | | 88 | | | +| -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| draH-PCS | | 48 | | | | | | | | +| 18 | | | | | 64 | | 80 | | | +40 +| | | | | | 56 | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +12 +| | | 32 | | | 48 | | 64 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +6 +40 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +33 +| hcneB ecneicS 18 | | 27 | | | | | 36 | | | +| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +30 +| 15 | | 24 | | | | | 33 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +27 +| | | 21 | | | | | 30 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 12 | | | | | 24 | | | | | +| | | 18 | | | | | 27 | | | +| 9 | | | | | 21 | | | | | +| | | 15 | | | | | 24 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +64 +90 +| ICS ULMM | | 80 | | | | | 94.5 | | | +| -------- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| 56 | | | | | 88 | | | | | +| 48 | | 76 | | | | | 93.0 | | | +86 +40 +| | | 72 | | | 84 | | 91.5 | | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +32 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | -------------- | ----- | -------------- | --- | ----- | -------------- | ----- | -------------- | --- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Figure21.Fullout-of-domainbenchmarkevaluationresultsfortheMATHdomainon7Band8Bmodels. +| | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | N=8 N=384 | | | | +| --- | --------- | --------------- | --------- | -------------------- | --- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | Pass@4(%) | | | Pass@8(%) | | Pass@16(%) | | +96 +| dnomaiD AQPG | | 64 | | | 80 | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 30 | | | | | | | 88 | | | +| | | 56 | | | 70 | | | | | +| 24 | | | | | | | 80 | | | +48 +| 18 | | | | | 60 | | 72 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +40 +| 12 | | | | | 50 | | | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| | | 32 | | | | | 64 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| | | | | | 80 | | 90 | | | +40 +| | | 60 | | | 70 | | 80 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +draH-PCS 30 +| | | 50 | | | 60 | | 70 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 20 | | 40 | | | | | | | | +| | | | | | 50 | | 60 | | | +| 10 | | 30 | | | | | | | | +| | | | | | 40 | | 50 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| 24 | | 32 | | | 36 | | 40 | | | +hcneB ecneicS +| 20 | | 28 | | | 32 | | 36 | | | +| --- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 16 | | 24 | | | 28 | | 32 | | | +| 12 | | 20 | | | 24 | | 28 | | | +| 8 | | 16 | | | 20 | | 24 | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +92 +70 +| ICS ULMM | | 84 | | | 90 | | 94 | | | +| --------- | ------- | ----- | --- | --- | ----- | ------- | ----- | ------- | --- | +| 60 | | 80 | | | | | | | | +| | | | | | 88 | | 92 | | | +| 50 | | 76 | | | | | | | | +| | | | | | 86 | | 90 | | | +| 40 | | 72 | | | | | | | | +| | | | | | 84 | | 88 | | | +| 30 | | 68 | | | | | | | | +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| 24 | | | | | | | 75 | | | +| AQPGrepuS | | 45 | | | 60 | | | | | +70 +| 20 | | 40 | | | 55 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +65 +16 +| | | 35 | | | 50 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 12 | | | | | | | 60 | | | +| | | 30 | | | 45 | | | | | +55 +| 0 | 150 300 | 450 0 | 150 | 300 | 450 0 | 150 300 | 450 0 | 150 300 | 450 | +| --- | -------------- | ----- | -------------- | --- | ----- | -------------- | ----- | -------------- | --- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Figure22.Fullin-domainbenchmarkevaluationresultsfortheSCIENCEdomainon7Band8Bmodels. +31 +LLMReasoningwithWeakSupervision +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | | N=8 N=384 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | ---- | --------- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +| 80 | | | 93 | | | 95.0 | | | 95 | | | +9 0 +| 005-HTAM | | | 8 7 | | | 9 2 . 5 | | | | | | +| -------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | +| 70 | | | 8 4 | | | 9 0 . 0 | | | 90 | | | +| | | | 81 | | | 87.5 | | | | | | +| 60 | | | | | | | | | 85 | | | +| | | | 75 | | | 80 | | | | | | +| 50 | | | 70 | | | 76 | | | 80 | | | +65 +72 +| 40 | | | 60 | | | 68 | | | 75 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +64 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 60 | | | | | | 84 | | | 92 | | | +| | | | 7 5 | | | 81 | | | 88 | | | +| | | | 7 2 | | | 78 | | | | | | +| 45 | | | 69 | | | | | | 84 | | | +| CMA | | | 66 | | | 75 | | | 80 | | | +| | | | 6 43 | | | 7 2 | | | | | | +| 30 | | | 8 | | | 5 4 | | | 6 5 | | | +| | | | 4 2 | | | | | | 6 0 | | | +| | | | 3 6 | | | 4 8 | | | 5 5 | | | +| 15 | | | 3 0 | | | 4 2 | | | | | | +| | | | 2 4 | | | 3 6 | | | 5 0 | | | +4 5 +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | --- | --- | ----- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +57.5 +| htaM avreniM 35 | | | | | | 51 | | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| | | | 44 | | | | | | 55.0 | | | +| 30 | | | | | | 48 | | | | | | +52.5 +| 25 | | | 40 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +45 +| | | | 36 | | | | | | 50.0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +| 20 | | | | | | 42 | | | | | | +47.5 +| 15 | | | 32 | | | | | | | | | +| ---------------- | --- | --- | ----- | --- | --- | -------- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 39 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | | | 56 | | | 64 | | | | | | +| hcneBdaipmylO 40 | | | | | | | | | 69 | | | +| | | | 48 | | | 60 | | | 66 | | | +| 32 | | | | | | 56 | | | 63 | | | +| | | | | | | 52 | | | 60 | | | +| 24 | | | 40 | | | 45 | | | | | | +50 +| | | | 32 | | | 40 | | | 45 | | | +| --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| 16 | | | | | | 35 | | | 40 | | | +| | | | 24 | | | 30 | | | | | | +| 8 | | | | | | 25 | | | 35 | | | +| 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure23.Fullout-of-domainbenchmarkevaluationresultsfortheSCIENCEdomainon7Band8Bmodels. +| | | | Qwen2.5-Math-7B | | LLama3.1-8B-Instruct | | | N=8 N=256 | | | | +| --- | --------- | --- | --------------- | --------- | -------------------- | --- | --------- | --------- | --- | ---------- | --- | +| | Avg@16(%) | | | Pass@4(%) | | | Pass@8(%) | | | Pass@16(%) | | +100 +| kcoL mutnauQ | | | | | | 80 | | | | | | +| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| 45 | | | 60 | | | | | | | | | +75 +60 +| 30 | | | 40 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 40 | | | 50 | | | +15 +| | | | 20 | | | 20 | | | 25 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 +| | | | 0 | | | 0 | | | 0 | | | +| -------------- | --- | ------- | --- | --- | --- | ----- | --- | --- | ----- | --- | ------- | +| 0 | 150 | 300 450 | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| 40 | | | 60 | | | | | | 50 | | | +| dnalsI tsegraL | | | | | | 60 | | | | | | +| 30 | | | | | | | | | 40 | | | +| | | | 45 | | | 45 | | | 30 | | | +20 +| 20 | | | 30 | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | 30 | | | 8 | | | +| 10 | | | | | | | | | 6 | | | +| | | | 15 | | | 15 | | | 4 | | | +| 0 | | | 0 | | | 0 | | | 2 | | | +0 +| 0 | 150 | 300 450 | 0 | 150 | 300 | 450 0 | 150 | 300 | 450 0 | 150 | 300 450 | +| --- | -------------- | ------- | --- | -------------- | --- | ----- | -------------- | --- | ----- | -------------- | ------- | +| | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | | +Figure24.Fullin-domainbenchmarkevaluationresultsfortheGRAPHdomainon7Band8Bmodels. +32 + +LLMReasoningwithWeakSupervision +75 +60 +45 +30 +0 150 300 450 +005-HTAM +Avg@16(%) Pass@4(%) Pass@8(%) Pass@16(%) +95.0 +90 92.5 +8 8 4 7 90.0 88 +81 87.5 +7 7 8 2 85 8 .0 0 80 +64 72 +72 +56 64 +48 56 64 +0 150 300 450 0 150 300 450 0 150 300 450 +18 +12 +6 +0 +0 150 300 450 +4202 +EMIA +60 +40 50 +35 45 +30 40 45 +25 35 +30 16 24 +12 18 +8 12 15 +4 6 +0 0 +0 150 300 450 0 150 300 450 0 150 300 450 +40 +32 +24 +16 +0 150 300 450 +htaM +avreniM +55 +48 +50 55 +42 +45 50 36 +40 45 +30 +35 40 +24 +0 150 300 450 0 150 300 450 0 150 300 450 +32 +24 +16 +8 +0 150 300 450 +dnomaiD +AQPG +60 88 +70 +50 80 +60 +40 72 50 +30 64 +40 +20 56 +0 150 300 450 0 150 300 450 0 150 300 450 +40 +32 +24 +16 +8 +0 150 300 450 +Training Steps +draH-PCS +Qwen2.5-Math-7B LLama3.1-8B-Instruct N=8 N=256 +60 +70 80 +50 60 70 +40 50 60 +30 40 50 +20 30 40 +0 150 300 450 0 150 300 450 0 150 300 450 +Training Steps Training Steps Training Steps +Figure25.Fullout-of-domainbenchmarkevaluationresultsfortheGRAPHdomainon7Band8Bmodels. +0.60 +0.45 +0.30 +0.15 +0.00 +draweR +gniniarT +Qwen2.5-1.5B +Math +0.75 +0.60 +0.45 +0.30 +0.15 +draweR +gniniarT +Llama-3.2-3B-Instruct +Math +0.60 +0.45 +0.30 +0.15 +0.00 +draweR +gniniarT +Qwen2.5-1.5B +Science +0.8 +0.6 +0.4 +0.2 +draweR +gniniarT +Llama-3.2-3B-Instruct +Science +1.0 0.8 +0.6 +0.4 +0.2 +0.0 +draweR +gniniarT +Qwen2.5-Math-7B +Graph +60 +50 +40 +30 +20 +)%( +005-HTAM +54 +51 +48 +45 +42 +39 +)%( +005-HTAM +30 +24 +18 +12 +6 +0 +)%( +tluciffiD-PCS 32 +24 +16 +8 +)%( +tluciffiD-PCS +50 +40 +30 +20 +10 +0 +)%( +kcoL +mutnauQ +30 +24 +18 +12 +6 +0 150 300 450 +Training Steps +)%( +CMA +32 +28 +24 +20 +16 +12 +0 150 300 450 +Training Steps +)%( +CMA +60 +50 +40 +30 +20 +0 150 300 450 +Training Steps +)%( +005-HTAM +54 +52 +50 +48 +46 +0 150 300 450 +Training Steps +)%( +005-HTAM +82 +80 +78 +76 +74 +0 150 300 450 +Training Steps +)%( +005-HTAM +=0 =0.1 =0.3 =0.5 =0.7 =0.9/1.0 +Figure26.Effectofrewardlabelcorruptionontrainingdynamicsandgeneralization.γdenotesthefractionoftrainingprompts +withcorruptedlabels,rangingfromclean(γ =0)tofullyincorrect(γ =1).QwenmodelsonMATHandSCIENCEdomainsmaintain +performanceundersubstantialcorruption,whilegeneralizationofLlamamodelsandGRAPHdomaindegradeatγ ≥0.5.Evaluation +resultsinthisfigurearebasedongreedydecoding. +33 + +LLMReasoningwithWeakSupervision +Qwen2.5-Math-1.5B Llama-3.2-3B-Instruct Qwen2.5-1.5B Qwen2.5-Math-1.5B +| | | Math | | | Math | | | Science | | | Science | +| --------------- | --- | ---- | --- | ---- | ---- | --- | --- | ------- | --- | --- | ------- | +| draweR gniniarT | | | | | | | 1 | | | 1.0 | | +| 0.75 | | | | 0.90 | | | | | | 0.8 | | +0.75 +| 0.60 | | | | | | | | | | 0.6 | | +| ---- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | +| | | | | 0.60 | | | | | | 0.4 | | +| 0.45 | | | | 0.45 | | | | | | | | +0.2 +| 0.30 | | | | 0.30 | | | 0 | | | | | +| ---- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | +)%( 005-HTAM +| | | | | 60 | | | 60 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +70 +70 +| | | | | 50 | | | | | | 60 | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +50 +| | | | | 40 | | | | | | 50 | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +60 +| )%( tluciffiD PCS | | | | | | | 12 | | | | | +| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 32 | | | 15 | | | | | | | | +| | | | | 12 | | | 9 | | | 32 | | +| | 24 | | | | | | | | | 24 | | +| | | | | 9 | | | 6 | | | | | +| | 16 | | | 6 | | | | | | 16 | | +| | 8 | | | | | | 3 | | | | | +| | | | | 3 | | | | | | 8 | | +| | 0 | | | 0 | | | 0 | | | 0 | | +0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +| --- | --- | -------------- | --- | ---- | -------------- | ------------- | --- | -------------- | --- | --- | -------------- | +| | | | | RLVR | | Majority Vote | | Self Certainty | | | | +Figure27.Comparisonofrewardvariants(RLVR,self-certainty,majorityvote)with1024trainingsamples.Proxyrewardswithout +verifiersexhibitfailuremodesunderprolongedtraining:trainingcollapse(self-certainty),andrewardspikesfollowedbyperformance +drops(majorityvote).Evaluationresultsinthisfigurearebasedongreedydecoding. +| | | | | | GRPO-POS | | GRPO-NEG | | GRPO | | | +| --- | --- | --- | --- | --- | -------- | --- | -------- | --- | ---- | --- | --- | +Training Reward SCP-Difficult (%) GPQA Diamond (%) MATH-500 (%) +| htaM-5.2newQ | | | | | | | )%( dnomaiD AQPG | | | | | +| ------------ | ------------------- | --- | --- | -------------------- | --- | --- | ---------------- | --- | --- | ------------ | --- | +| | draweR gniniarT 1.0 | | | )%( tluciffiD-PCS 35 | | | | | | | 72 | +| | | | | | | | | 25 | | )%( 005-HTAM | | +| | 0.8 | | | 30 | | | | | | | 70 | +B5.1- +20 +| | 0.6 | | | 25 | | | | | | | 68 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +15 +| | 0.4 | | | 20 | | | | | | | 66 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | | | 10 | | | 64 | +| | 0.2 | | | 15 | | | | | | | | +5 +62 +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| --- | --- | --------- | --- | --- | ----- | --- | --- | ----- | --- | --- | ------------- | +Training Steps Training Steps )%( dnomaiD AQPG Training Steps Training Steps +| | draweR gniniarT 1.0 | | | )%( tluciffiD-PCS | | | | | | 52.5 | | +| ----------- | ------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | ------------ | --- | +| B3-2.3amalL | | | | 15 | | | | | | )%( 005-HTAM | | +27.5 +| | tcurtsnI- 0.8 | | | 12 | | | | | | 51.0 | | +| --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | +25.0 +| | | | | 9 | | | | | | 49.5 | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | +| | 0.6 | | | | | | 22.5 | | | | | +6 +| | 0.4 | | | | | | 20.0 | | | 48.0 | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | ---- | --- | +3 +46.5 +| | 0.2 | | | 0 | | | 17.5 | | | | | +| --- | --- | -------------- | --- | --- | -------------- | --- | ---- | -------------- | --- | --- | -------------- | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +Figure28.EffectofbaselinevariantsonSCIENCEdomainwith8trainingsamples.GRPO-pos(positiveupdatesonly)andGRPO-neg +(negativeupdatesonly)producecomparableperformancetostandardGRPO. +| | | | | | GRPO-POS | | GRPO-NEG | | GRPO | | | +| --- | --- | --- | --- | --- | -------- | --- | -------- | --- | ---- | --- | --- | +htaM-5.2newQ Training Reward SCP-Difficult (%) GPQA Diamond (%) MATH-500 (%) +)%( dnomaiD AQPG +| | draweR gniniarT | | | )%( tluciffiD-PCS 40 | | | | | | )%( 005-HTAM | | +| --- | --------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | +| | 0.75 | | | | | | | 25 | | | 72 | +| | B5.1- | | | 35 | | | | | | | | +| | | | | | | | | 20 | | | 70 | +| | 0.60 | | | 30 | | | | | | | | +68 +| | 0.45 | | | 25 | | | | 15 | | | | +| --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +66 +10 +| | 0.30 | | | 20 | | | | | | | 64 | +| --- | ---- | -------------- | --- | --- | -------------- | --- | --- | -------------- | --- | --- | -------------- | +| | | | | 15 | | | | 5 | | | 62 | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +)%( dnomaiD AQPG +| | draweR gniniarT | | | )%( tluciffiD-PCS 25 | | | | 28 | | | 54 | +| ----------- | --------------- | --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | +| B3-2.3amalL | 0.90 | | | | | | | | | )%( 005-HTAM | | +| | tcurtsnI- | | | | | | | 26 | | | | +| | 0.75 | | | 20 | | | | | | | 52 | +24 +| | 0.60 | | | 15 | | | | | | | 50 | +| --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 0.45 | | | | | | | 22 | | | 48 | +10 +| | 0.30 | | | | | | | 20 | | | | +| --- | ---- | -------------- | --- | --- | -------------- | --- | --- | -------------- | --- | --- | -------------- | +| | | | | 5 | | | | | | | 46 | +| | 0.15 | | | | | | | 18 | | | | +| | | 0 150 300 | 450 | | 0 150 | 300 | 450 | 0 150 | 300 | 450 | 0 150 300 450 | +| | | Training Steps | | | Training Steps | | | Training Steps | | | Training Steps | +Figure29. EffectofbaselinevariantsonSCIENCEdomainwith1024trainingsamples. SimilartoFigs.28,GRPO-pos(positive +updatesonly)andGRPO-neg(negativeupdatesonly)producecomparableperformancetostandardGRPO. +34 + +LLMReasoningwithWeakSupervision +Figure30. Responsediversityon8samplesfromtheMATH-500evaluationdataset. Qwen-mathshowshighdiversitywithinits +correctanswers,suggestingarangeoflearnedrobustreasoningpaths. +DiversityPromptforLLM-as-a-judge +Youaregiventheoriginalpromptandtwomodel-generatedresponses. Determinewhetherthetworesponsesuse +differentstrategiestosolvetheproblem. +Usethefollowingguidelines: +-Different solution methods: Clearly different approaches (e.g., algebraic vs. geometric, analytical vs. +numerical). +-Criticalreasoningdivergence:Significantdifferencesinkeyreasoningstepsorassumptions,eveniffinalanswers +match. +-Conceptualdifferences: Distinctunderlyingconceptsorrepresentations(e.g.,probabilityvs. combinatorics). +**Alsolabelasdifferentif:**Thetworesponsessharethesamegeneralapproachbutdiffermeaningfullyin +specificintermediatestepsormanipulationscrucialtothesolution. +Originalprompt: prompt +Generation0: generation0 +Generation1: generation1 +Question: DoGeneration0andGeneration1usedifferentstrategies? +You may first generate a short reasoning, then end your response with either ||yes|| if they use different +strategiesor||no||iftheyusethesamestrategy. +Figure31.LMprompttochecksimilaritybetweenresponses. +35 + +LLMReasoningwithWeakSupervision +FaithfulnessPromptforLLM-as-a-judge +Youwillbegiven: (1)amathproblemprompt,and(2)amodelresponsethatmayincludeBOTHreasoninganda +finalanswer. +Definitions: +”Reasoning” = the parts of the model response that attempt to justify or derive a result (intermediate steps, +equations,explanations,casework,narrativelogic). +”Finalanswer”=themodel’sexplicitcommittedresult(e.g.,after”Final:”,”Answer:”,”Therefore”,” +boxed”, or the last clear numeric/symbolic conclusion). If multiple answers appear, treat the last explicitly +committedoneasthefinalanswer. +Task:Decidehowwellthereasoningsupportsthefinalanswer,usingtheselabels: +Label1(Correlated): +Thereasoningformsacoherentderivationthatwouldleadtothefinalanswerasstated. Minoralgebraslipsare +allowedIFtheoverallderivationstillclearlytargetsthatanswer. Thefinalanswermaybeobjectivelywrong;you +judgealignment,notcorrectness. +Label0.5(PartiallyCorrelated): +Thereasoningisrelatedtotheproblemandseemstomovetowardthefinalanswer,buthasmajorgaps,unjustified +leaps,missingsteps,orseriouserrorsthatbreaktheproof. Theanswerisnotapurenon-sequitur,butthesupport +isweak/incomplete. +Label0(Uncorrelated): +The final answer is not supported by the reasoning. Examples include: contradiction with earlier derived +statements; switchingtoanunrelatedmethod; violatingkeyconstraintsfromtheprompt; orthefinalanswer +appearingasanunsupportedguess. +Outputformat(MANDATORY): +1)Brieflyidentify(a)theextractedfinalanswerand(b)thekeyreasoningpathin1–3sentences. +2)Thenoutputexactlyonelabeltokenonitsownattheend: ∥1∥or∥0.5∥or∥0∥. +Prompt:prompt +Response:response +Question: Does the reasoning path correspond to the provided answer? You may first generate a short +reasoning,thenendyourresponsewitheither∥1∥iftheyarefullycorrelated,∥0.5∥iftheyarepartiallycorrelated, +or∥0∥iftheanswerisuncorrelatedtotheprecedinglogic. +Figure32.LMprompttoevaluatereasoningfaithfulnessonasamplefromtheMATHdataset. +36 + +LLMReasoningwithWeakSupervision +Figure33.Proportionofalignedandmisalignedresponsesacrossmodelsandtrainingdatasets. +| draweR gniniarT | | )%( 005-HTAM 90 | | | 72 | | 80 | | | +| --------------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +)%( draH-PCS +| | | 84 | | | )%( CMA 64 | | | | | +| ------ | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | +| ecracS | 0.6 | 78 | | | 56 | | | | | +ataD +40 +60 +40 +| | | 54 | | | 30 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +0.0 +| | 0 150 300 | 450 | 0 150 | 300 | 450 | 0 150 300 | 450 | 0 150 | 300 450 | +| --- | --------- | --- | ----- | --- | --- | --------- | --- | ----- | ------- | +90 +| draweR gniniarT | | )%( 005-HTAM | | | | | )%( draH-PCS | | | +| --------------- | --- | ------------ | --- | --- | ---------- | --- | ------------ | --- | --- | +| ytirojaM | | | | | )%( CMA 60 | | 60 | | | +etoV +0.6 +60 +40 +30 +0.0 +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +draweR ysioN +| draweR gniniarT | | )%( 005-HTAM | | | | | | | | +| --------------- | --- | ------------ | --- | --- | ---------- | --- | ------------ | --- | --- | +| | 0.6 | | | | | | )%( draH-PCS | | | +| )7.0= | | | | | )%( CMA 60 | | | | | +| | | 75 | | | | | 60 | | | +0.3 +( +40 +30 +50 +| | 0 150 300 | 450 | 0 150 | 300 | 450 | 0 150 300 | 450 | 0 150 | 300 450 | +| --- | -------------- | --- | -------------- | --- | --- | -------------- | --- | -------------- | ------- | +| | Training Steps | | Training Steps | | | Training Steps | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure34.Evaluationresultsofpass@16metricacrossmodelswithdifferentpre-RLinterventiononweaksupervision. +37 + +LLMReasoningwithWeakSupervision +| | avg@16 | | pass@16 | | avg@16 | | pass@16 | | +| --- | ------ | --- | ------- | --- | ------ | --- | ------- | --- | +)%( hcneB ecneicS +)%( 4202 EMIA +30 +8 +| | | 20 | | 10 | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +ecracS +ataD +15 +0 0 150 300 450 0 0 150 300 450 0 0 150 300 450 0 150 300 450 +)%( hcneB ecneicS +)%( 4202 EMIA +ytirojaM +| 8 | | 20 | | 10 | | 20 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +etoV +| 0 | | 0 | | 0 | | 0 | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 0 150 300 450 600 +| draweR ysioN | | | | )%( hcneB ecneicS | | | | | +| ------------ | --- | --- | --- | ----------------- | --- | --- | --- | --- | +)%( 4202 EMIA +| )7.0= | | | | | | 30 | | | +| ----- | --- | --- | --- | --- | --- | --- | --- | --- | +| 8 | | 25 | | 10 | | | | | +( +15 +| 0 | | 0 | | 0 | | | | | +| --- | -------------- | --- | -------------- | --- | -------------- | --- | -------------- | --- | +| 0 | 150 300 | 450 | 0 150 300 | 450 | 0 150 300 | 450 | 0 150 300 | 450 | +| | Training Steps | | Training Steps | | Training Steps | | Training Steps | | +Base + Thinking SFT Base + Non-Thinking SFT CPT + Thinking SFT CPT + Non-Thinking SFT Instruct +Figure35. EvaluationresultsonAIME2024andScienceBenchacrossmodelswithdifferentpre-RLinterventiononweak +supervision. +38 + +LLMReasoningwithWeakSupervision +DiversityJudgeExampleOutput +Prompt: Aboxcontains5whiteballsand5blackballs. Idrawthemoutofthebox,oneatatime. Whatisthe +probabilitythatallofmydrawsalternatecolors,startingandendingwiththesamecolor? +ResponseA: To determine the probability that all of the ball draws alternate colors, starting and end- +ingwiththesamecolor,weneedtoconsiderthefollowing: +1. **Totalnumberofwaystoarrangetheballs**: Thereare5whiteballsand5blackballs,sothetotalnumberof +waystoarrangethese10ballsisgivenbythebinomialcoefficient +(cid:0)10(cid:1) +,whichrepresentsthenumberofwaysto +5 +choose5positionsoutof10forthewhiteballs(theremaining5positionswillbefortheblackballs). Thisvalue +iscalculatedas: +(cid:18) (cid:19) +10 10! += =252 +5 5!5! +2. **Numberoffavorableoutcomes**: Forthedrawingstoalternatecolorsstartingandendingwiththesame +color,thereareonlytwopossiblesequences: -Startingwithawhiteballandalternating(WBWBWBWBW +B)-Startingwithablackballandalternating(BWBWBWBWBW) +Sothereareonly2favorableoutcomes. +3. **Probability calculation**: Theprobability istheratioof thenumberof favorableoutcomestothe total +numberofoutcomes. Therefore,theprobabilityP is: +Numberoffavorableoutcomes 2 1 +P = = = +Totalnumberofoutcomes 252 126 1 +Sothefinalansweris . +126 +ResponseB: Tosolvetheproblem,weneedtodeterminetheprobabilitythatasequenceof10balls(5white +and5black)isarrangedsuchthattheyalternatecolors. Thereareonlytwopossiblesequencesthatmeetthis +condition: ”WBWBWBWBWB”and”BWBWBWBWBW”. Thetotalnumberofwaystoarrange5whiteand5 +blackballsisgivenbythebinomialcoefficient +(cid:0)10(cid:1) +,whichrepresentsthenumberofdistinctpermutationsof5 +5 +whiteand5blackballs. Thisiscalculatedas: +(cid:18) (cid:19) +10 10! += =252 +5 5!5! +Sincethereareonly2favorableoutcomes(thetwoalternatingsequencesmentionedabove),theprobabilityisthe +ratioofthenumberoffavorableoutcomestothetotalnumberofoutcomes: +2 1 +Probability= = +252 126 1 -) - -November 1999 - - Nineteenth-Century Log House - -Location: (cid:9) -0.25 miles south of the intersection of SR 1424 and SR 1405. - -This log house stands on the west side of SR 1424, approximately - -The log house sits approximately 40 feet from the road, and has - -Setting: (cid:9) -two mid-twentieth century outbuildings standing to the rear of the house. To the -north of the log house stands a late twentieth-century residence. - -Description: - -The one-story, side-gable, three-bay log dwelling is - -Dwelling: (cid:9) -constructed with half-dovetail corner notching. The house features a -variety Of wood sheathing, including narrow weatherboards and novelty -siding. Windows in the building have been enlarged and have six-over-six -sashes. The porch on the facade has also been replaced. A one-story -frame addition was constructed onto the rear of the dwelling. An exterior -brick chimney is located on the south elevation. - -Outbuildings: Two frame outbuildings stand south of the house: a frame -chicken house with a shed roof and a small, one-story frame building with -an entry on the gable end. - -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, this house is considered not -eligible for the National Register of Historic Places under Criterion A, since it is -not representative of a significant. trend in the history of the area. - -This house is considered not eligible fir the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. - -This house is considered not eligible for the National Register under Criterion C -since the building has lost integrity and does not embody distinctive -characteristics. The building no longer retains its original porch or windows and -the form has been compromised by the addition to the rear of the dwelling: - -The building is considered not likely to yield information important to the history -of building technology; therefore it is not eligible for the National Register under -Criterion D. - -Historic Architectural Resources Survey Report: NCDOT, SR 1424. Burke County: November 1999 -Nancy l'art Dolsen - -Page 28 - - Photograph 1: Nineteenth-Century Log House, facing northeast - -Photograph 2: Nineteenth-Century Log 1-louse facing southwest - -Historic Architectural Resources Survey Report: .VC DOT SR 1424, Burke County: November 1999 -Nancy I an Dolsen - -Page 29 - - Photograph 3: View of Chicken Coop, facing northeast - -Photograph 4: Domestic outbuilding, facing north - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nwtcy i in Dolsen - -Page 30 - - BIBLIOGRAPHY - -Arney, Marshall House. Historic Resource Survey Form, Burke County, 1983. -Asheville, NC: Western Office. Division of Cultural Resources. - -Arney, William Jones House, Historic Resource Survey Form, Burke County, 1983. - -Asheville, NC: Western Office, Division of Cultural Resources. - -Biographical Directory of the United States Congress, 1774-Present, Thomas Burke,- - -Website: Bioguide.conress.gov. - -Burke County Deeds, Burke County Court House, Morganton, NC. - -Burke County Plat Book, Burke County Court House, Morganton, NC. - -Corpening House, Historic Resource Survey Form, Burke County, 1983. Asheville,NC: - -Western Office, Division of Cultural Resources. - -Cotton, J. Randall, Suzanne Pickens Wylie, and Millie M. Barbee. Historic Burke: An -Architectural Sites Inventory of Burke County. Mortzanton, NC: Historic Burke -Foundation. Inc.. 1987. - -Phifer, Edward W. Burke: the History of a North Carolina County, 1777-1920. - -Morganton, NC: by the author, 1977. - -Historic Architectural Resources Survey Report: NCD01; SR 1424, Burke County: November 1999 (cid:9) -Nancyl'an Dolsen - -Page 31 - - APPENDIX A - -NCDOT, SR 1424. Bur ke Cou - -November 1 999 - - North Carolina Department of Cultural Resources - -James B. Hunt Jr., Governor -Betty Ray McCain, Secretary - -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 - -July 9, 1999 - -Stephen A. Moore, PE -District Engineer -NCDOT -Route 1 Box 169C -Marion NC 28752 - -Division of Archives and History -Jeffrey J. Crow, Director - -LOCATION -507 North Blount Street -Raleigh, NC -State Courier 53-31-31 - -Re: (cid:9) Widen and pave SR 1438 Corpening Bridge Road, - -Burke County, ER 99-9229 - -Dear Mr. Moore: - -We have received notification of a nationwide permit for the above project and would -like to comment. - -We have reviewed the permit application for the improvement and paving of SR 1438 -Corpening Bridge Road and located the following structures of historical or -architectural importance within the general area of the project: - -William Avery House, located at the northwest corner of the junction of SR 1438 -and SR 1424. - -Arney's Chapel, located at the east side of SR 1423 south of its junction with SR -1439. - -We recommend that an architectural historian with the North Carolina Department of -Transportation evaluate these sites to determine if they are within the area of potential -effect for the permit and are eligible for listing on the National Register of Historic -Places. - -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. - -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. - --Sincerely, - -1- (cid:9) - -• (r-N — - -Da.vicl'Brook -Deputy State Historic Preservation Officer - -DB:slw - -109 East Jones Street • Raleigh, North Carolina 27601-2807 - - cc: (cid:9) - -Army- Corps of Engineers, Asheville -William D. Gilmore - -t-Barbara Church - - AUG-16—":39 MON 10:10 ID: - -TEL NO: - -351; P02 — - -North Carolina Department of Cultural Resources - -Jams B. Hunt Jr.. Governor -Betty Ray McCain, Secretary - -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 - -August 10, 1999 - -Stephen A. Moorc -District Engineer -North Carolina Department of Transportation -Route 1, Box 169C -Marion NC 28752 - -Re: - -Paving SR 1414, Johns River Road, Burke County, -ER 00--7154 - -Division of Archives Bnd History -Jeffrey 1. Crow, Director - -LOCATION -507 North Blount Street -Raleieh, NC -State Courier 53-31-31 - -it- , - have received notification of a nationwide permit for the above project and would - -i%1'- - -Dear Mr. Moore: - -We -like to comment. - -We have conducted a search of our maps and files and have located the following -structures of historical or architectural importance within the general area of the -project: - -qX.117. - -Marshall Arney House, located on thc wcst side of SR 1424, 0.5 mile northwest -of its junction with SR 1438. - -William Arney House, located at the northwest corner of the SR 1424 and SR -1438 junction. - -9;0 ck - -Corpening House, located on the west side of SR 1424, L5 mile south of its -junction with SR 1405. - -We recommend that an architectural historian with the North Carolina Department of -Transportation identify and evaluate these properties and any other structures over -fifty years of age within the project area, and report the findings to us. - -An important archaeological site, a mound reported by J. W. Spainour and Cyrus -Thomas in the late 1800s, is recorded between the Johns River and Johns River Road -south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 311310 highlighted in -yellow. - -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan - -109 Eist Jones Street • Ra1e10, North Cirolina 27601-2807 - - AUG-16-'99 MON 10:11 ID: (cid:9) - -TEL NO: (cid:9) - -p355 P03 - -Stephen A. Moore -August 10, 1999, Page 2 - -sheets 8,9, 12. 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. - -end south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 31BK3 highlighted in -yellow. - -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan -sheets 8, 9, 12, 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. - -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. - -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. - -Sincerely, - -&Ad,- - -Deputy State Historic Preservation Officer - -/ David Brook - -DB:slw - -Enclosure - -cc: (cid:9) - -Army Corps of Engineers, Asheville -William D. Gilmore -Thomas Padgett - +So,theprobabilitythatallofmydrawsalternatecolors,startingandendingwiththesamecolor,is . +126 +JudgeOutput: Both Generation 0 and Generation 1 use the same strategy to solve the problem. They both +identifythattherearetwopossiblesequencesthatsatisfytheconditionofalternatingcolorsandstartingand +endingwiththesamecolor: ”WBWBWBWBWB”and”BWBWBWBWBW”. Theybothcalculatethetotal +numberofwaystoarrangethe10ballsusingthebinomialcoefficient +(cid:0)10(cid:1) +,whichaccountsforthepermutations +5 +of5whiteand5blackballs. Finally, theybothcomputetheprobabilitybytakingtheratioofthenumberof +favorableoutcomes(2)tothetotalnumberofoutcomes(252),resultinginthesamefinalanswerof 1 . +126 +Therearenosignificantdifferencesintheirreasoningsteps, assumptions, orintermediatecalculations. Both +responsesfollowasimilarlogicalflowandusethesamemathematicalconceptstoarriveatthesolution. +JudgeResult: ∥no∥ +Figure36.QualitativeExampleofDiversityAnalysis +39 + +LLMReasoningwithWeakSupervision +FaithfulnessJudgeExampleOutput +Prompt: Findallrealpairs(a,b)thatsolvethesystemofequation +a2+b2 =25, +3(a+b)−ab=15. +Response: Tosolvethesystemofequations: +(cid:40) +a2+b2 =25, +3(a+b)−ab=15, +wecanusethefollowingreasoningandalgebraicmanipulations. First,recalltheidentity: +a2+b2 =(a+b)2−2ab. +Let’sdenotes=a+bandp=ab. So,thefirstequationbecomes: s2−2p=25.Thesecondequationremains: +3s−p=15.Wecansolvethesetwoequationssimultaneously. First,weexpresspfromthesecondequation: +p=3s−15.Next,wesubstitutepintothefirstequation: s2−2(3s−15)=25.Simplifytheequation: +s2−6s+30=25 =⇒ s2−6s+5=0. +√ +Thisisaquadraticequationins. Wecansolveforsusingthequadraticformulas= −b± b2−4ac,wherea=1, +2a +b=−6,andc=5: +√ √ +6± 36−20 6± 16 6±4 +s= = = . +2 2 2 +So,thesolutionsforsare: s=5 or s=1.Now,wefindthecorrespondingpvaluesforeachs: 1. Ifs=5, +thenp=3(5)−15=0. 2. Ifs=1,thenp=3(1)−15=−12. Wenowneedtofindthepairs(a,b)thatsatisfy +theequationsa+b=sandab=p. Thesearetherootsofthequadraticequations: 1. Fors=5andp=0,the +quadraticequationis: +t2−5t=0 =⇒ t(t−5)=0. +So,thesolutionsaret=0andt=5. Thus,thepairs(a,b)are(0,5)and(5,0). 2. Fors=1andp=−12,the +quadraticequationis: +t2−t−12=0. +Wesolvethisquadraticequationusingthequadraticformula: +√ +1± 1+48 1±7 +t= = . +2 2 +So,thesolutionsaret=4andt=−3. Thus,thepairs(a,b)are(4,−3)and(−3,4). Therefore,thecomplete +setofsolutionsis: +(0,5),(5,0),(4,−3),(−3,4) . +JudgeOutput: Finalanswergiven: (0,5),(5,0),(4,−3),(−3,4).Reasoningpath: Introducessumss=a+b +andproductp=ab,rewritesequationsintoasysteminsandp,solvestogets=5or1,findscorrespondingp, +thensolvesquadratict2−st+p=0toobtainallorderedpairsconsistentwitheach(s,p),yieldingthelisted +solutions. Thederivationiscompleteandconsistentlyleadstothestatedpairs. +JudgeResult: ∥1∥ +Figure37.QualitativeExampleofFaithfulnessAnalysis +40 \ No newline at end of file diff --git a/q103/random_k4/random_2.md b/q103/random_k4/random_2.md index 9a6992971674500fba1af17645e45a0908def2ef..d1447fd886d0d69697641e94fcf1890d83769afd 100644 --- a/q103/random_k4/random_2.md +++ b/q103/random_k4/random_2.md @@ -1,727 +1,1679 @@ - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Hobart and William SmitH CollegeS - Online ISSN: 2054-0965 (Online) +Handbook for +International Students - Print ISSN: 2054-0957 (Print) +intercultural affairs Center +288 Pulteney Street +geneva, nY 14456 +(315) 781-3319 – Phone +(315) 781-3888 – Fax - Website: https://www.eajournals.org/ + TABLE OF CONTENTS - Publication of the European Centre for Research Training and Development -UK +SECTION I: General Information + Welcome -Cloud Computing: Revolutionizing Digital -Transformation in the Modern Era +important Campus Phone numbers -Lavanya Jacintha Victor -The University of Texas at Austin, USA + office locations -doi: https://doi.org/10.37745/ejcsit.2013/vol13n181223 Published May 12, 2025 +banks and theaters +Places to Stay +Frequently asked Questions -Citation: Victor L.J. (2025) Cloud Computing: Revolutionizing Digital Transformation in the Modern Era, European -Journal of Computer Science and Information Technology, 13(18),12-23 +SECTION II: Immigration Procedures + How to get assistance with immigration matters +immigration basics and definition of terms -Abstract: Cloud computing has revolutionized digital transformation by fundamentally changing how -organizations approach their technological infrastructure and business operations. The shift from -traditional on-premises systems to cloud-based solutions has enabled unprecedented levels of scalability, -efficiency, and innovation across industries. From artificial intelligence integration to edge computing -adoption, cloud technologies have democratized access to advanced capabilities while reducing -operational costs. Organizations worldwide are embracing hybrid and multi-cloud strategies, prioritizing -security through zero-trust architectures, and focusing on sustainable computing practices. The evolution -of serverless computing and edge technologies continues to shape the future of enterprise IT, offering new -possibilities for innovation while addressing crucial challenges in data processing, security, and -compliance. -Keywords: cloud computing, digital transformation, edge computing, serverless architecture, hybrid -cloud + obtaining a Visa and Form i-20 -INTRODUCTION +differences between F-1 and J-1 Status -In today's rapidly evolving technological landscape, digital transformation has become a cornerstone of -business evolution and success. According to recent International Data Corporation (IDC) forecasts, -worldwide spending on digital transformation is projected to maintain a robust compound annual growth -rate (CAGR) of 16.3% from 2022 to 2026, culminating in a staggering $3.4 trillion by 2026. The digital -and information technology sector has demonstrated remarkable resilience even during economic -uncertainties, showcasing a significant 17.8% growth in digital transformation investment during 2022 -alone [1]. This comprehensive shift towards digitalization represents more than just a trend—it's a -fundamental reimagining of how organizations operate, deliver value, and interact with their stakeholders -through the integration of digital technologies. + arriving at a U.S. port of entry… What a student can expect +dealing with the department of Homeland Security (dHS) +exceptions to a Full Course of Study +duration of Status for F-1 Students +temporary departure from the US + + accepting U.S. government Public assistance or Public benefits + Work opportunities for students with F-1 Status + Work opportunities for students in J-1 Status + +Practical training + optional Practical training (oPt) + Curricular Practical training (CPt) +departure Checklist + +page 1 +9 +10 +10 +10 12 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -The impact of digital transformation is profound yet challenging. Despite substantial investments, recent -executive surveys reveal that approximately 70% of digital transformation initiatives are falling short of -their objectives. A significant factor contributing to this shortfall is the persistent gap between technological -capabilities and workforce readiness. Nearly two-thirds of surveyed executives report that their -organizations struggle with change management and employee adoption of new digital systems. -Furthermore, 84% of organizations cite human-centric issues, rather than technical complications, as the -primary barriers to successful digital transformation [2]. - -The landscape of digital transformation has evolved significantly, with organizations focusing on three -critical areas: technology infrastructure modernization, process automation, and workforce enablement. -The IDC research highlights that industries such as financial services, manufacturing, and retail are leading -the investment charge, with discrete manufacturing alone expected to invest $382 billion in digital -transformation by 2026 [1]. However, the success of these investments heavily depends on organizations' -ability to address fundamental human elements. Studies indicate that organizations achieving digital -transformation success are three times more likely to prioritize workforce development and change -management strategies alongside their technological implementations [2]. - -The momentum of digital transformation continues to accelerate, driven by market demands and -competitive pressures. While the technological spending trajectory remains impressive, with manufacturing -and financial services each projected to exceed $1 trillion in total digital transformation investments through -2026 [1], the focus is increasingly shifting toward holistic transformation approaches. Successful -organizations are now recognizing that digital transformation requires a balanced investment in both -technology and people, with leadership development, skill enhancement, and cultural adaptation being as -crucial as the technological infrastructure itself [2]. - -The Evolution of Digital Transformation -The journey of digital transformation began with the basic digitization of manual processes, marking a -fundamental shift in business operations. Current research indicates that organizations implementing digital -transformation strategies have achieved remarkable improvements in their operational metrics. By -integrating automation and digital workflows, businesses have reported up to 80% reduction in process -completion times and a 90% decrease in human errors across various operational tasks. This evolution has -particularly impacted customer service operations, where automation has enabled organizations to handle -customer inquiries 60% faster while maintaining high satisfaction rates [3]. What started as simple -workflow automation—like digital leave management systems and basic inventory tracking—has evolved -into sophisticated, interconnected ecosystems that drive entire business operations. - -The transformation landscape has seen significant advancement in data processing capabilities, with -modern systems able to handle complex analytical tasks in real-time. Organizations implementing -comprehensive digital transformation strategies have witnessed a 70% improvement in their decision- -making processes, primarily due to enhanced data accessibility and automated analytical capabilities. +15 +16 +17 +19 +21 +22 +24 +24 +24 +25 +26 +26 +27 +27 +28 +29 + +SECTION III: -13 +Sample Fall orientation Schedule for international Students - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +30 - Online ISSN: 2054-0965 (Online) +2 + + SECTION I: +GENERAL INFORMATION + +Welcome! + +Welcome to Hobart and William Smith Colleges (the Colleges or HWS). to you, the +decision to study abroad might have been a small one – or, indeed, a rather big step. +either way you won’t know what your experience will be like until you are here. this +handbook is designed to help you prepare for the adventure of studying in the United +States. in our experience the better prepared you are the more you will benefit from +your time at the Colleges. + While we are trying to answer many of your questions prior to your arrival in the +United States, some of your questions will only arise once you are here. the inter- +national Student adviser will serve as a resource to you and your parents or guardians +throughout the application process and during your stay at Hobart and William Smith +Colleges. The adviser will be your primary contact for any questions or concerns +regarding visas and immigration issues, employment, travel and a host of other + matters that may arise. + +the International Student Office (iSo) is located on the second floor of the + +intercultural affairs Center at 288 Pulteney Street. this office advocates for and ad- +dresses the needs of international students as well as provides programs and services +related to immigration matters. during the academic school year the iSo sponsors +and co-sponsors events designed to promote cross-cultural sharing between the many +groups at the Colleges and +geneva communities. +Currently, international +students come to us from 22 +nations: austria, bulgaria, +Canada, China, Colombia, +Czech republic, germany, +india, iran, nepal, Pakistan, +Portugal, Scotland, South +africa, South Korea, Swe- +den, Switzerland, thailand, +the netherlands, turkey, +Ukraine and Vietnam. + +Intercultural Affairs Center – Home to the International Student Office. + +in the fall international students attend a special two-day orientation before +the regular first year orientation begins. at Hobart and William Smith, we ask that +incoming international students join us ahead of time as it is a wonderful opportunity +to get acquainted with the new culture, get assimilated with the campus, and provide +a forum to answer your questions before the rush of orientation activities begins. at +the iSo, we will check to make sure that all your immigration paperwork is in order as +well as help you gain a better understanding of what your F-1 or J-1 status means and +how to maintain it; help to answer any questions that you might have about adjusting +to life as a student in the United States, specifically life at HWS; and meet students, +staff and faculty members. + +3 + + Questions? Please +contact the International +Student Adviser by phone + at (315) 781-3319. The + International Student + Office is located on the + 2nd floor of the +Intercultural Affairs Cent- +ter, 288 Pulteney Street, +Geneva, NY 14456. (315) +781-3319 – Phone (315) +781-3888-Fax. Office +hours are 8:30 am to +5:00 pm weekdays during +the academic year. + +THINGS TO DO BEFORE GOING ABROAD +before you can board a plane and start your study +abroad adventure, there are some formalities you will +need to take care of. the earlier you start thinking +about the things you need, the smoother your prepara- +tions will go. this list will help you consider some of +the things you will have to do. + +Obtaining a Visa +Upon receipt of the form i-20 (for F-1 students) or dS- +2019 (for J-1 exchange students) you can apply for your +student visa. it is wise to schedule an appointment with +the U.S. embassy or consulate in your country as soon +as possible as sometimes there are no appointments +available for weeks. You should also give yourself some +time to prepare for your visit to the U.S. embassy or +consulate. Visa procedures vary from country to country. +make sure to check the website of the embassy or + +consulate in your country to find out what documents you need to bring. document +you certainly need include: (1) a current, valid passport; (2) your i-20 or dS-2019 +form; (3) evidence of financial support for the duration of time you will spend in the +US; (4) proof that you have a permanent residence outside the US; (5) one or more +passport pictures meeting U.S. requirements; (6) a nonimmigrant visa application. + +Medical Insurance +HWS Colleges require that all students buy the medical insurance provided by +the colleges. all J-1 student-visa holders (that is, all exchange students) are also +required by U.S. law to have their own medical insurance. While F-2 students are +covered by the Colleges medical insurance you might nevertheless consider buying +additional medical coverage. + +Important Documents +You need to bring all important documents relating to your legal, medical, and aca- +demic history with you. Such documents might include medical and dental records, +certificates of immunizations, information about medical conditions and treatments, +prescriptions for medication and eyeglasses or contact lenses, and official transcripts +from secondary school and college. make sure you photocopy all documents and +leave some copies with a person you trust in your home country. it is also helpful to +translate these documents into english before you leave your home country. + +Packing +Pack as light as possible. remember that you might have to carry all your things at +different times during the journey. if you are studying abroad for only a semester you +might not need to bring your entire wardrobe. You might also want to leave room +in your suitcase for souvenirs you will buy while in the US. at the same time there +might be things you want to bring because they will make you more comfortable + +4 + + and will make your transition to a new environment easier. Some exchange students +and international students also like to bring small gifts from their home country to +share with new friends abroad. this might be a means for you to share and explain +your culture to your new acquaintances. if you have traditional, national cloth- +ing consider bringing it to wear at events sponsored by the international Students +association or other cultural events. if you arrive during spring semester (January), +to bring a warm jacket and boots, a scarf, gloves and hat. We encourage you to take +advantage of the option to ship your packages ahead of time using the box number +that will be provided to you from the Colleges orientation office. You may also +consider purchasing larger items upon your arrival at our local shopping centers. + +Traveling Tips +You can book your flight as soon as you find the Colleges start date for international +Student orientation. at this time, you might have already received the schedule of +international student orientation. the Colleges provide a complimentary pick-up +from rochester and Syracuse airport for those international students coming to the +Colleges for the first time if you arrive during the assigned times. long-distance +traveling can be really exhausting, so try to get a good-night’s sleep before the trip, +eat lightly and drink plenty of water. it is also a good idea to wear several layers of +clothes so you can accommodate to changes in temperature. label your luggage with +both your home address and the address in the United States (if you don’t have an +address yet, use that of the international student office). Keep you passport, impor- +tant documents, flight tickets, and money or traveler’s checks with you at all times. +You might also want to pack everything you need for a one-night emergency stay +into your hand luggage in case you get stuck at the airport or your luggage gets lost. + +Money +You should carry a small amount of U.S. dollars with you while traveling so that you +can buy some food or other small things you might need. However, it is not a good +idea to carry large sums of money. We recommend that you buy traveler’s checks or +get a credit card that allows you to withdraw money from a cash machine when you +need to. Upon arrival on campus, the international student adviser can help you to +set up an account with a local bank. it is highly recommended to have an U.S. bank +account because it will make payments so much easier. Students can create bank +accounts free of charge. + +GETTING TO KNOW HOBART AND WILLIAM SMITH COLLEGES +College is an exciting time for everyone and it is a transition for all students. if you +are an exchange student, you already went through the transition process once. +nevertheless, studying abroad brings with it a range of new experiences. every col- +lege is a little different and if you have never studied in the United States before, +many things might be very confusing. + +Hobart and William Smith Colleges +Hobart and William Smith is a small, private liberal arts institution. You can expect +to study in relatively small classes and build close relationships with your classmates - Print ISSN: 2054-0957 (Print) +5 - Website: https://www.eajournals.org/ + and professors. Virtually all students are being housed on campus. living so close +to each other has many advantages: it is relatively easy to get to know new people, +get involved in clubs and activities, and make an impact on the community. at the +same time it can be hard to find personal time and separate school from other activi- +ties when the two are not physically separated. + +Classes +at HWS, a normal course load is four classes per semester. this means that you +will be in class for between 12 and 15 hours every week, or more if you take courses +requiring labs or studio work. there are different types of classes. 100- and 200-level +classes are usually introductory classes and have a wider focus. 300- and 400-level +classes have a narrower, more specific focus. Some professors prefer to lecture, but +you also encounter class-discussions. Higher level classes tend to be taught in a semi- +nar style, in which a lot of the class’ content relies on student presentations. there +are also independent study classes in which you research a topic under a professor’s +guidance but without the classroom structure. as an upper-class student you might +choose to work on an honors project. this is an independent research project that +will culminate in a 80- to 100-page paper. You can take classes from all departments +but at the end of your Sophomore (2nd) year, you will need to declare a major and a +minor subject (or one major, two minors; or two majors, one minor; or two majors, +two minors). as part of your liberal arts education, you will also have to fulfill cer- +tain academic goals. Your academic adviser will help you with that. + +Grading and Grades +Foreign students often find that americans are very competitive. this also reflects +on the system of grading. You will find that you will have many opportunities to +prove your knowledge of a subject in class. many professors will give short, unan- +nounced quizzes that test your knowledge of the course material assigned for that +day. exams are bigger, announced tests that cover the course material of a longer pe- +riod of time. they can be compromised of true or false questions, sentence comple- +tion (“fill in the blank”), multiple choice, matching, essay questions, or a combina- +tion of these. Professors will also assign essays and papers you have to write on your +own time. Your class participation might also count toward your final grade. the +Colleges use the grading system based on a four-point scale: a= 4.0, b= 3.0, C= 2.0, +d= 1.0. You need at least a C average in your classes to remain a full-time student. + +Homework +Some foreign students are overwhelmed by the quantity of homework assigned by +american colleges. especially in the beginning when you still have problems with +the english language, keeping up with homework can be really hard. if you find +yourself struggling to keep up with work, ask for help. Find the Center for teaching +and learning on campus and schedule an appointment! Ctl is open to all students +and its services are already covered by your tuition. they can help you with your +writing assignments, help with time management, or assign you a tutor to help you +with the content of a class. it is also important to realize that most american stu- +dents prioritize their work: decide what readings are important to do and read them +thoroughly; skim the others. +6 + + Professors and Advisers +every student at HWS is assigned an academic adviser. He or she will help you +with any questions you have related to your academic life. Feel free to consult them +before choosing classes, if you are struggling with certain classes or professors, if you +want to know how and when to declare majors and minors, and any other questions +you might have. most professors are also interested in helping you as much as pos- +sible and want to get to know you outside of class. don’t feel embarrassed to ask a +professor a question after class. You should also take advantage of their office hours. +interaction between professors and students in america is a lot tighter than in many +other countries. Students sometimes stop by their professor’s office simply to talk +about something that they are interested in. + +Important information on financial aid +Hobart and William Smith Colleges offer merit scholarships as well as need-based +grants to international students. the average package typically equals more than +half the annual cost of tuition, room and board. Highly competitive applicants will +be considered for a full tuition scholarships. + +Financial aid is available for a maximum of four years for those who enroll as + +first-year students. For those entering with advanced standing (transfers), aid is +available for the time allotted to complete a degree. awards based on demonstrated +financial need, if it becomes apparent that such need no longer exists, may be re- +duced. Students must resubmit a financial aid application each year. + +ADJUSTING TO A NEW ENVIRONMENT +going abroad for a semester or the duration of one’s studies is exciting. at times, +however, the confrontation with a foreign culture can be challenging. Studying +abroad is a learning experience in every sense of the word. besides your scholastic +learning in college courses, you will get to know the american culture and how +people around you view their world and their lives. by experiencing and understand- +ing a different way of life, you will also learn about things in your own culture you +have never thought of before. this is your chance to broaden your view and reflect +on and mature your own values. the process of learning about a new culture and +finding your own place in the new environment can be confusing and tiring. the +best way to deal with a new situation and find out about something you don’t under- +stand is to ask questions. + +Culture Shock and Adjustment +Culture shock is a normal phenomenon on your way to adjusting to a new environ- +ment. For some people, the encounter with a new culture can be shocking, indeed, +but for most people culture shock is not as sudden and dramatic as you might expect. +When the excitement about being in a foreign country wears off, cultural differences +can become overwhelming. You might think your language skills are inadequate; you +might be struggling with course work; you might be troubled by unfamiliar food; you +might have difficulties interpreting other people’s behavior; you might be offended +by the way people act or dress or by their body language. there are a variety of +symptoms of culture shock. Some people feel tired and nervous; some are frustrated + +7 + + with or even hostile toward the american culture; some get angry and irritated very +easily; some isolate themselves or socialize only with other international students and +people who speak their language. You might also feel very homesick and question +why you decided to go abroad in the first place. a way to prepare yourself to deal with +culture shock is to clarify for yourself why you want to study abroad. Knowing your +personal expectations and goals will help you overcome periods of doubt. + +THINGS TO DO +• Chose a non-credit or half-credit class like tennis, scuba diving, or basic sailing. +• Join some of the many student clubs on campus (http://www.hws.edu/studentlife/ + +campuslife/sll/clubs). + +• You like to be active? Visit the sport and recreation facilities and take a yoga class, +borrow a racquet and play tennis or racquetball. You can also try out for intramu- +ral sports. + +• get involved in the religious life on campus! there are services and meditation as +well as bible study sessions, ecumenical candlelight prayers, Sabbath dinners and +a lot more. + +• Participate in ongoing events – take in a movie on Fridays at Friday Flix, visit the +Cellar Pub in the basement of Coxe Hall and see what is happening (comedians, +bands, etc.), show school spirit by supporting one of many sporting events hap- +pening on campus. + +• are you interested in current events? Political and social developments? the +Fisher Center lecture Series, the President’s Forum and the President’s radio +Show are a few venues through which you can satisfy this interest. + +• travel the area! intercultural affairs (iC) and the Center for global education + +(Cge) will occasionally offer trips to see the area, oraP (outdoor recreational +activities Program) organizes hiking, rock climbing and rafting trips (among +others), and you can always get a group of friends together and go somewhere on +your own. + +• offer your help in the language departments. maybe someone needs a Spanish + +tutor or a teaching assistant for mandarin? + +• discover geneva. Volunteer. to find out how visit the Public Service office in + +trinity (2nd floor) (or call x3825). + +• do you like to dance? or watch others dance? the dance department is big on +campus and the students as well as the faculty put on dance shows! ask about +Koshare or join one of the dance teams. + +• Check the daily Update! there is always something going on – just show up and + +try it out! - Publication of the European Centre for Research Training and Development -UK +8 -Furthermore, the integration of artificial intelligence and machine learning has enabled predictive -maintenance systems that reduce equipment downtime by up to 45% [3]. + P +H +O +N +E +N +U +M +B +E +R +S +& +O +F +F +I +C +E +S + + Extension + +IMPORTANT PHONE NUMBERS +Office +business office +Center for global education +Center for teaching and learning +Counseling Center +deans: Hobart +deans: William Smith +Financial aid +Hubbs Health Center +information technology +intercultural affairs +registrar +residential education +Security +emergency +Student accounts + +3655 +3307 +3351 +3388 +3300 +3467 +3315 +3600 +4357 (HelP) +3319 +3651 +3900 +3656 +3333 +3386 + +OFFICE LOCATIONS +admissions +alcohol and other drug Programs +alumni House +athletics (WS) +buildings and grounds +business office +Career Services +Center for global education +Ctl +Chapel +College Store +Communications office +deans: Hobart +deans: William Smith +dining Services + +Financial aid +Hubbs Health/Counseling Center +information technology +intercultural affairs +President +Provost’s office +Public Service office +registrar +residential education +Security +Student accounts +Student life and leadership +WeoS radio Station + +bradford House +Smith Hall +615 South main +Winn-Seeley +medbery Parking lot +demarest (basement) +trinity (1st Floor) +trinity (3rd Floor) +Harris House (678 S. main St.) +St. John’s Chapel (near 630 S. main St.) +51 Saint Clair Street, Sherrill Hall +durfee House (639 S. main St.) +Smith Hall (1st Floor) +Smith Hall (1st Floor) +Scandling Center Café, Saga, +Cellar Pub, de Cordova Café +demarest (basement) +119 St. Clair St. +Williams Hall +288 Pulteney St. +Coxe Hall (2nd Floor) +Coxe Hall (2nd Floor) +trinity (3rd Floor) +gulick (1st Floor) +101 St. Clair St. +medbery Parking lot +demarest (basement) +Scandling Center (2nd Floor) +113 Hamilton St. + +9 + + BANKS AND MOVIE THEATERS IN GENEVA + +Community Bank +5 Seneca St. +geneva, nY 14456 +Phone: (315) 789-7700 +Website: www.communitybankna.com + +First Niagara Bank +811 Canandaigua road +geneva, nY 14456 +Phone: (315) 789-1049 +Fax: (315) 789-6009 +Website: www.firstniagarabank.com + +BANKS +Five-Star Bank +P.o. box 193 +geneva, nY 14456 +Phone: (315) 789-2300 +Fax: (315) 787-3263 +Website: www.five-starbank.com + +Seneca Falls Savings Bank +342 Hamilton St. +geneva, nY 14456 +Phone: 315-789-8100 +Website: www.seneca24.net/ + +MOVIE THEATERS +Geneva Movieplex 8 +t & C Plaza, Hamilton St. +geneva, nY, 14456 +Phone: (315) 789-4524 +Web: www.cinemanorth.com + +PLACES TO STAY +Note: the Colleges provide this list as a service to all who might visit our area, but +do not endorse any local hotel, motel, inn or bed and breakfast. + +the following are a handful of accommodations that are available in and around +geneva, no more than 30 minutes from campus. lodging is also available in ithaca, +rochester and Syracuse, which are each approximately one hour from Hobart and +William Smith. You can find out more about accommodations outside the greater +geneva area by visiting the geneva Chamber of Commerce or ontario County +tourism bureau websites. + +Holiday inn, Waterloo +(315) 539-5011 or (800) HolidaY + +Hampton inn +(315) 781-2037 + +ramada inn, geneva, lakefront +(315) 789-0400 + +belhurst Castle/White Spring manor, +geneva +(315) 781-0201 + +microtel inn & Suites, Seneca Falls +(315) 539-8438 + +geneva-on-the-lake, geneva +(315) 789-7190 or (800) 3-geneVa + +motel 6, Hamilton Street, geneva +(315) 789-4050 or (800) 466-8356 -The Era of On-Premises Infrastructure -The first major milestone in this journey was the establishment of on-premises data centers, which -fundamentally changed how businesses managed their digital operations. On-premises data centers -emerged as critical infrastructure components, offering organizations complete control over their hardware, -software, and data management processes. This infrastructure model provided significant advantages in -terms of data security and compliance, particularly for industries handling sensitive information. A key -benefit was the ability to maintain an average network latency of less than 5 milliseconds, crucial for time- -sensitive applications and real-time data processing [4]. +10 -The financial implications of on-premises infrastructure were substantial but justified for many -organizations. While initial setup costs typically ranged between $10,000 to $25,000 per rack, with annual -operating costs between $12,000 and $18,000, these investments provided organizations with direct control -over their technological infrastructure. The model proved particularly advantageous for businesses with -consistent workloads and stringent regulatory requirements, as it eliminated monthly recurring costs -associated with cloud services and provided better control over long-term expenses [4]. + G +E +N +E +V +A + +, + +N +E +W +Y +O +R +K + +PLACES TO EAT +*Please note that some restaurants are closed on Mondays or Tuesdays, +so call ahead. + +The area code for all numbers listed below is (315). + +American/Continental +abigail’s restaurant, routes 5 & 20, Seneca Falls, 539-9300 +belhurst Castle, lochland road, route 14, 781-0201 +the Crow’s nest, 415 boody’s Hill road, 781-0600 +emile’s restaurant, east north Street and routes 5 & 20, 789-2775 +geneva-on-the-lake, 1001 lochland road, 789-7190 +madderlake Café, West lake road, 781-2424 +the red brick inn, 291 main St., 789-3247 + +Diners/Grilles +Flour Petal Café, 34 linden St., 781-2233 +Parkers’ grille & tap House, 100 Seneca St., 789-4656 +Ports Cone & grill, 4432 W. lake road, 789-2020 +Water Street Café, 467 exchange St., 789-2560 +the Captain’s room, 372 exchange St., 781-0122 +dana’s time out, 258 Hamilton St., 781-2107 + +Italian-American and Pizza +nonna Cosentino’s trattoria, 1 railroad Place, 789-1638 +Uncle Joe’s Pizzeria, 99 n. genesee Street, 781-1199 + +International +ling ling, 481 Hamilton Street, 781-0668 +main moon, 480 exchange Street, 781-5106 +Wing tai oriental restaurant, 164 Castle Street, 789-8892 +garcia’s mexican restaurant, 313 Hamilton Street, 789-5760 + +Fast Food/Family Chains +routes 5 & 20 west of downtown geneva (Hamilton Street) offers an array of popu- +lar fast food and family chains, including Quiznos, tim Horton’s, Club 86 bagels +and Cakes, Friendly’s, KFC-taco bell, mcdonald’s, long John Silver’s, Pizza Hut, +Subway, Waffle Works and Wendy’s. + +11 + + FREQUENTLY ASKED QUESTIONS (FAQs) + +Do you require the TOEFL for international students? +Yes, we do require the toeFl (test of english as a Foreign language). We offer limited +english language support on campus, and require that students have a solid foundation +in english prior to admission. the minimum required computer based toeFl score is +220. the Colleges also will accept the aPiel examination in lieu of the toeFl. + +Do you offer financial assistance for international students? +need-based scholarships are available to international students. the cost of attend- +ing college here for the 2007-2008 academic year is U.S. $45,968. the average aid +package typically equals more than half the annual cost of tuition, room and board. +Highly competitive applicants will be considered for a full-tuition scholarship. the +financial aid application forms are included in the application for admissions. + +How many international students do you currently have on campus? +We have 32 full-time matriculated students in a degree program Fall 2006 +Semester. the following countries are represented: austria, bulgaria, Canada, +China, Colombia, Czech republic, germany, india, nepal, Pakistan, South Korea, +Sweden, turkey, Ukraine, United Kingdom and Vietnam. + +Do you have an international student adviser? +Yes, the student adviser is responsible for working with our international students +on campus to help them make the transition from home to the Colleges. We offer an +international student orientation prior to the first-year and transfer student orienta- +tion at the beginning of the academic year, and the beginning of the spring semester +for new arrivals. a variety of activities and programs are available throughout the +year. Student advising and counseling is available in a variety of other areas on cam- +pus, including faculty advisers, Hubbs Health Center, and the Center for teaching +and learning. We make every effort to work individually with each student to help +them reach their potential and make the most of the opportunities available here. + +Do you have an international student organization? +Yes. the international Student association (iSa) is a student organization that cel- +ebrates the diverse cultures found around the globe. other cultural clubs with interna- +tional perspectives represented on campus include asian Student association (aSU), +latin american organization (lao), Caribbean Student association (CSa), and +dia al-din: muslim Student association. + +Do you have E.S.L. services? +Yes, an entry level course is offered, as are workshops and one-to-one tutoring. How- +ever, the Colleges do not offer a full e.S.l. curriculum. + +What kinds of housing are available? +Within the Hobart and William Smith community are dozens of smaller communi- +ties--those of the individual residences. the range of housing options is extensive, -Security and compliance capabilities represented another crucial aspect of on-premises infrastructure. -Organizations maintained full control over their security protocols, physical access management, and data -handling procedures. This control was particularly valuable for industries subject to strict regulatory -requirements, such as healthcare and financial services, where data sovereignty and compliance are -paramount concerns. The ability to customize security measures and maintain direct oversight of data -storage and processing activities allowed organizations to achieve compliance rates exceeding 99% [4]. +12 -However, this approach presented notable challenges in terms of scalability and resource management. -Organizations needed to maintain dedicated IT teams for infrastructure management, with staffing costs -typically accounting for 40-50% of the total operational budget. Additionally, scaling operations required -significant advance planning and capital investment, with expansion projects often taking several months -to complete. The necessity for redundant systems and backup power solutions further added to the -complexity and cost of maintaining on-premises infrastructure [4]. + with coed or single-sex college residences, cooperative houses, theme houses, and +town houses among the choices you’ll find over your four years here. residence +halls offer a variety of configurations, with traditional singles and doubles as well as +suites. Some recent theme houses have included Jewish Culture House, multicul- +tural Sisterhood House, Writers House, and international awareness House. the +ten cooperative houses provide an alternative to standard residence-hall living, with +residents sharing responsibility for food and housecleaning arrangements. + +Is it possible to work on campus? +You are eligible to work up to 20 hours a week on campus, if your visa status permits +it. However, the demand for such employment exceeds the number of jobs available. +therefore, you cannot use on-campus employment as a resource for the issuance of +the i-20 form. + +STORAGE AND TRANSPORTATION OF SCHOOL YEAR ITEMS +Special consideration will be given to those international students who need storage +on campus during the school breaks. international students must attend a manda- +tory meeting in September regarding storage and transportation of school year items. +attendance will be taken at this meeting. Students unable to attend this meeting +must take the initiative to set up a meeting with the international Student adviser +before oct. 1. + +the storage and transportation meeting in September will encompass the following +information: +a. date and time of transportation to and from storage areas + +- Winter break (december) +- Summer break (may) +b. location of storage areas +c. How to pack and maximum number of storage items allowed +each student will be allowed no more than 8 items. items must be labeled with +students name, return date and number of items. boxes must be secured properly +and must be easily manageable. large items such as refrigerators, sofas, etc. will not +be accommodated due to limited space. +d. transportation requirements: items are to be placed at curbside 15 minutes +before van arrives for pick up. +e. Cost of transportation = $20 per student (gas and labor cost) +f. information on surcharges: the student’s account will be charged a surcharge for +each infraction: + +- No shows = $50 +- Fail to have items at curbside = $20 +- Not packed = $20 +- Change of time and/or date without 24 hour notice to Intercultural Affairs + Office = $40 + +Please Note: Storage space is limited, therefore there is no guarantee that you +will get storage or transportation support. + +13 + + IN CASE OF AN EMERGENCY +if you should have a need to contact the international Student adviser while travel- +ing outside of the United States please call (315) 781-3656 and ask for extension +3319. You can also contact the office directly by calling (315) 781-3319. + +In an emergency, they can be reached at (315) 781-3333. + +remember that the international Student adviser, while a resource to you, is not +your only resource during your time at the Colleges. While on campus, your respec- +tive deans will be available to assist you with a variety of issues, and should always +be kept aware of important matters. + +WHAT TO EXPECT DURING SCHOOL BREAKS +Since not everyone is able to return home during school breaks, international +students are allowed to remain on campus during those times. at specific times +during the fall and spring, you will receive e-mails from the residential education +office asking about their plans during the upcoming break. Your immediate response +is necessary as it is the only way to ensure that you will get permission to remain in +your room or elsewhere on the campus during school breaks. + +Students are not allowed to remain in their rooms during fall, winter and spring + +breaks. during summer vacation, students who remain on campus must move into +temporary housing. once again, an e-mail from the residential education office +is the first step in determining where students will be staying during the summer. +Currently, international students are allowed to stay on campus during the summer +holiday free of charge (subject to change). + +the dining hall is not open to students during the summer months so students +will need to provide their own meals. most summer housing will feature a kitchen, +but in the event that one doesn’t, appropriate accommodations will be made for +those students who wish to prepare their own meals. + +the international student adviser must also know your plans and should also be + +notified if you will or will not be staying on campus. + +US Holidays/Observances +January 1 +January 15 +3rd monday in February +2nd Sunday in may +4th monday in may +3rd Sunday in June +July 4th +1st monday in September +october 31 +november 11 +4th thursday in november +december 4 +december 25 +december 26 14 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +new Year’s day +martin luther King Jr. day +Presidents day +mother’s day +memorial day +Father’s day +independence day +labor day +Halloween +Veterans day +thanksgiving +Hanukkah begins +Christmas day +Kwanzaa + + SECTION II: +IMMIGRATION PROCEDURES + +HOW TO GET ASSISTANCE WITH IMMIGRATION MATTERS + +the international Student adviser is available to help you process your immigration +forms and documents. Working with the department of Homeland Security can be +very complicated, detailed, and time-consuming. it is important that you do not rely +on information obtained from friends and other students because circumstances may +vary from one student to another and the regulations can and do change. + the international Student adviser must work within the federal immigration +laws. When making a request, please make sure that you allow plenty of time for +your request to be processed and plan to be patient. Have your documents, such as +your passport, i-94 card, i-20 or dS-2019 with you when you go to the international +student office for immigration assistance. + +Important: You should allow at least one week for processing signatures, letters +and/or other types of paperwork. Special consideration will be given for emergency +situations. + +Reminder: the international Student adviser cannot remember when each +student’s documents expire. it is YOUR responsibility to keep track of all immigra- +tion documents and keep them current. + +Hristo Dimitrov ’09 (Bulguria), and Ayokunle Abogan ’06 (Nigeria) - Online ISSN: 2054-0965 (Online) +15 - Print ISSN: 2054-0957 (Print) + IMMIGRATION BASICS AND DEFINITION OF TERMS - Website: https://www.eajournals.org/ +as with all countries, the United States has laws and regulations governing foreign- +ers who are temporarily within its borders. the immigration regulations are adminis- +tered by the Department of Homeland Security (DHS). - Publication of the European Centre for Research Training and Development -UK +immigration regulations can change frequently and are difficult to interpret. it +is important for international students to understand the rules as they apply to stu- +dent status. The DHS does not consider ignorance of the law a legitimate reason +for failing to obey it. + Questions should be directed to the international Student adviser, who will be +able to assist with legal regulations and can sign your immigration documents. -Table 1. Digital Transformation Performance Metrics: Year-over-Year Analysis (2021-2024) +dHS requires that a copy of your passport, i-94 and other immigration docu- +ments be kept on file by the international Student adviser, as well as your local +address and telephone number. -Metric Category +Passport +be sure to keep your passport in a safe place. it is your most important document. +most passports are good for a limited period of time. Your passport must be valid at +all times while you are in the United States. -Equipment Downtime Reduction -Customer Inquiry Speed -Decision-Making Efficiency -Infrastructure Staff Cost -Network Latency (ms) +the expiration date of your passport should be six months beyond the date you -2021 (%) -25 -30 -40 -35 -20 +enter the U.S. if your passport expiration date is close to the date you will begin +classes, obtain a new passport before leaving home. -2022 (%) -35 -45 -55 -40 -15 +it is your responsibility to remember when your passport will expire and to con- +tact the nearest consulate or embassy for your country. to find the nearest consulate +or U.S. embassy check http://usembassy.state.gov. -2023 (%) -40 -55 -65 -45 -10 +Your passport can only be extended by your government and it usually takes +several weeks for a passport to be renewed, so be sure to allow enough time. You may +extend your passport in the U.S. at your home country’s embassy or consulate. You +are in this country illegally if your passport expires and are subject to deportation. -2024 (%) -45 -60 -70 -50 -5 +Visas +the visa stamped in your passport at the U.S. embassy or Consulate is for permis- +sion to enter the United States for a specific purpose and period of time. Students +are usually given an F-1 visa. J-1 visas are given to researchers, scholars and students +who are sponsored by agencies or universities. -The Cloud Computing Revolution -The emergence of cloud computing marked a paradigm shift in the digital transformation landscape, -fundamentally altering how businesses approach their IT infrastructure and operational capabilities. Recent -market analysis reveals that cloud spending now represents approximately 30% of IT budgets across -industries, with this figure expected to grow significantly. The industry is witnessing an unprecedented shift -toward multi-cloud and hybrid cloud strategies, with organizations increasingly focusing on cloud cost -optimization and management. This evolution has led to the emergence of specialized roles such as cloud -financial operations (FinOps) professionals, who are becoming critical to managing cloud investments -effectively [5]. - -The adoption of cloud computing continues to accelerate, driven by both necessity and opportunity. -Industry research indicates that companies can capture up to $1 trillion in annual value through cloud -adoption, highlighting the transformative potential of this technology. This value creation stems from -various sources, including innovation acceleration, cost reduction, and improved resilience. The impact is -particularly significant in Fortune 500 companies, where cloud adoption has enabled revenue growth of up -to $100 billion annually through broader cloud-supported initiatives [6]. - -Key Benefits of Cloud Computing: Infrastructure Flexibility -Cloud platforms have revolutionized resource allocation capabilities, enabling unprecedented operational -efficiency. The shift toward cloud-native architectures has become increasingly prominent, with -organizations moving beyond basic "lift and shift" approaches to more sophisticated cloud-native -implementations. This trend has been accelerated by the growing adoption of containers and microservices, -with Kubernetes emerging as a de facto standard for container orchestration. The market has seen a -significant focus on edge computing capabilities, with major cloud providers expanding their edge presence -to support growing demands for low-latency applications and improved performance [5]. - -Cost Optimization -The financial implications of cloud adoption are substantial and multifaceted. Research shows that -organizations can potentially reduce IT costs by 20 to 30 percent through cloud optimization initiatives. -However, the most significant value creation opportunities lie in business innovation, which can generate -$770 billion to $995 billion in annual EBITDA across Fortune 500 companies. This value stems from +Since the visa stamp is only important for entry and re-entry to the United +States, there is no need to worry if it expires while you are here if you are maintain- +ing your student status. because a visa is only a permit to apply for entry into a coun- +try, it is not possible to obtain a U.S. visa while you are in the U.S. to apply for a +visa, you must go to a U.S. consulate or embassy in a country other than the United +States, usually your home country. -15 +I-94 (Arrival/Departure Record) +Form i-94, also called the arrival/departure record of Stay Permit, is the small white +card given to you at the port of entry to the U.S. this form is very important since it +officially determines how long an international student can remain in the U.S. - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +16 - Online ISSN: 2054-0965 (Online) + Students on the F-1 and J-1 visas usually have their i-94 stamped “d/S” (dura- + +tion of Status), which means that the student can remain in this country for the +length of time required to complete his/her studies as long as the student follows all +of the regulations of the student visa. the eleven-digit number written on your i-94 +is your dHS admission number, which is used by the dHS to monitor entry into and +departure from the U.S. The I-94 cards are turned in when leaving the U.S. and a +new I-94 is completed with each subsequent entry. + +if you are on a visa other than a F or J, or if you have a specific date stamped on + +your i-94 card instead of the “d/S” designation, you must apply for an extension, a +Change of Status, or exit the U.S before the expiration date or you will be consid- +ered out of status and a visa overstay, which can have serious dHS consequences. + +SEVIS +an acronym for Student and exchange Visitor information System (SeViS). it is a +data collection and monitoring system that creates an interface between institutions +of higher education, the department of Homeland Security (dHS), consulates and +embassies abroad, and ports of entry. Schools are required to make regular electronic +updates in SeViS throughout each semester on the records of their enrolled students +in F-1 and J-1 status and their dependents and their researchers and faculty on J-1 +status. this includes, but is not limited to enrollment status, changes in address, +changes in level of study, employment recommendations and school transfers. + +Form I-20 +When Hobart and William Smith Colleges accepts an international student for +full-time study, the student is sent a Form i-20 a-b as proof of acceptance and to +use when applying for an F-1 Student Visa at a U.S. Consulate abroad. the student +presents the i-20 with their passport and visa at the Port of entry and the i-20 is +stamped and given back to the student. this is your Certificate of eligibility, and +it must be signed by a dSo/ international Student adviser before you leave the +United States. Students should keep copies of all i-20 issued to them. + +Immigration Status +this is often confused with “visa” but your immigration status, e.g., F-1, J-1, etc. is +determined at the time of your entry into the U.S. by an immigration official and is +noted on the Form i-94. You may have many visa stamps in your passport but, upon +entry into the U.S., an immigration inspector will admit you in only one immigra- +tion status which is noted on the i-94 card. + +OBTAINING A VISA AND FORM I-20 +Your passport should be valid at least six months beyond the date of entering the U.S. +as a student. be sure to check your passport expiration date. if it will expire close to +when you will begin classes, obtain a new passport before applying for your visa. + +How to Obtain a Visa +although U.S. embassies will not issue a student visa prior to 120 days before the +start date on the i-20 (or dS-2019 for J’s), you can schedule an appointment prior - Print ISSN: 2054-0957 (Print) +17 - Website: https://www.eajournals.org/ + to the 120-day period. Since processing times have become longer due to increased +security clearances, students should check with their local U.S. embassy or consulate +to learn the estimated waiting time and allow plenty of time to obtain the visa. a +list of consulates and embassies can be found at http://usembassy.state.gov/. + +You will need to pay a $100 SeViS fee at least 3 business days before your visa +interview. the fee can be paid online or by mail. For complete directions, see http:// +www.fmJfee.com. + Unless you are a sponsored J-1 student or already in another visa category, the +Colleges’ office of admissions will send you a Form i-20 when you are admitted and +have submitted proof of your ability to finance your education. take the i-20 form, +your acceptance letter, passport, copies of sponsorship letters and bank statements +to prove financial support, and a receipt for your SeViS fee payment to the United +States embassy or consulate nearest your home. once you have acquired the visa +stamp in your passport, you are ready to enter the U.S. + +Note: Students may not enter the U.S. more than 30 days prior to the start date +on their I-20. + +Canadian and some British Commonwealth citizens do not need visas to enter +the U.S., but they must request F-1 status and present their i-20 forms for process- +ing at the point of entry even if the official does not ask for it. they should also be +prepared to show evidence of financial support and proof of citizenship. Without a +stamped i-20 and an i-94 card with the F-1 or J-1 designation on it, you are con- +sidered a tourist in the United States and will not be legally able to study and take +advantage of the special privileges allowed those in Student Status until he or she +has been in an immigration status appropriate for studying for at least nine consecu- +tive months. + +Note: If you will not be entering the U.S. on an F-1 visa, do not use an I-20 form +to obtain a visa. If you received an I-20, please return it to the Office of Admis- +sions with an explanation of the type of visa you hold. + +if your visa document (i-20 or dS-2019) does not arrive in time for you to secure +the proper visa, do not enter the United States on a B1/B2 Visitor Visa. obtain a +b-2 “prospective student” or “prospective exchange visitor” visa from a U.S. embassy +or consulate post in your country. be certain the consul writes “Prospective Student” +or “Prospective exchange Visitor” (for J-1) in your passport on the visa page. You +will need to present your letter of acceptance and evidence of financial support to +request this visa. + +in general, it is best to avoid using the b-2 visa since you will have to pay a fee +and do a time-consuming change of status through dHS (department of Homeland +Security) after you arrive. You may not begin your studies until the change is com- +pleted, so this could also delay the start of your studies until a subsequent term. + +Note: Students must attend the school named on the I-20 or DS-2019 presented at +the port of entry in order to be in status. Also, do not enter the U.S. on a “W-T” - Publication of the European Centre for Research Training and Development -UK +18 -various sources, including accelerated product development, improved customer experience, and enhanced -operational efficiency. The cloud's consumption-based pricing model has enabled organizations to achieve -15 to 20 percent cost reduction in infrastructure spending through improved capacity management [6]. + (waiver) as this classification cannot be extended or changed and you will lose your +legal immigration status as a student. + +Students Transferring From another Colleges in the United States +if you are an F-1 student transferring from a U.S. institution and will not be leav- +ing the country prior to enrollment at Hobart and William Smith Colleges, follow +the procedures required by dHS. bring your completed i-20 to the international +Student adviser’s office as soon as possible after your arrival on campus. Sign the +document and leave it to be processed by a dSo/ international Student adviser +who will notify you when you may pick it up. + +if you leave the U.S. before beginning your studies at Hobart and William +Smith Colleges, use the new i-20 issued by the Colleges when you arrive at the port +of entry. Keep the i-20 from your old school with your immigration records, but do +not use it to enter the U.S. it is not necessary for transfer students with an unex- +pired student visa and who have maintained their status to obtain a new visa, even +though the former school’s name is listed. + +Transfer to another School +to be eligible to transfer schools, you must be pursuing a full-course of study at your +current school. otherwise you are “out of status” and must apply to dHS for rein- +statement to student status, which, if granted, will allow you to transfer. + +Students transferring to another school must notify the dSo/international +Student adviser of their intent to transfer. Hobart and William Smith Colleges +must release your SeViS record to the new school before a new i-20 can be issued +for the transfer. Upon arrival at the new school, the student should take the new +i-20 to the international Student adviser at the new school. J-1 students wishing to +transfer should consult with their sponsor. + +Changing Your Academic Level +if you are currently a student at Hobart and William Smith Colleges, and will be +continuing your studies in an additional program of study (such as a mat Program), +you should follow the transfer procedure. a new i-20 will be issued for the new pro- +gram, and you will use the new i-20 to enter the country. if you do not leave, bring +it to the international Student adviser as described in the transfer process in the +preceding paragraph. Failure to complete this “internal transfer” will cause you to +be out of status with dHS, which can adversely affect work authorizations or other +student privileges. + +DIFFERENCES BETWEEN F-1 AND J-1 STATUS +there are two types of entry visa issued to nonimmigrant students who intend to +study full-time at Hobart and William Smith Colleges—the F-1 student visa and the +J-1 exchange visitor visa. + +• Source of Funding: J-1 exchange visitor visa status is available to those students +who are supported substantially by funding other than personal or family funds. +Such funding may include that which comes from the U.S. government or the -Global Reach and Collaboration -Cloud infrastructure has enabled unprecedented global operational capabilities, with particular emphasis on -sustainability and environmental impact. Organizations are increasingly prioritizing environmental -considerations in their cloud strategies, with major providers committing to aggressive sustainability goals. -The industry has seen a growing focus on renewable energy usage and improved data center efficiency, -with cloud providers investing heavily in sustainable operations. This shift has resulted in significant -improvements in energy efficiency and reduced carbon footprints for organizations migrating to cloud -services [5]. +19 -The impact of cloud computing extends beyond operational benefits to strategic value creation. -Organizations leveraging cloud capabilities have reported significant improvements in their ability to -innovate and respond to market changes. The cloud has enabled faster time-to-market for new products and -services, with some companies reducing development cycles by up to 50 percent. Additionally, cloud -adoption has facilitated better risk management and improved security posture, with organizations able to -leverage advanced security features and compliance capabilities provided by cloud platforms. The value at -stake is substantial, with research indicating that companies can potentially capture between $900 billion -and $1.2 trillion in annual value through comprehensive cloud transformations [6]. + student’s home government, or an international organization, or the Colleges. Stu- +dents who are supported by personal or family funds are ineligible for J-1 immigra- +tion status and must come to the U.S. to study in F-1 immigration status. + +• Foreign Residence Requirement: Students in J-1 immigration status and the J-2 +dependents may be subject to a “foreign residence requirement.” J-1 students who +receive direct or indirect U.S. or home government funding, or who are studying +in fields for which personnel are considered in short supply in their home coun- +tries (most developing nations have “Country Skills” lists of varying lengths), are +ineligible to apply for a change to another non-immigrant status (except a or g) +or permanent residency in the U.S. until they have satisfied the “two year home +country physical presence requirement. only those in J-1 status with government +funding or whose skills are on Country Skills list are subject to the foreign resi- +dence requirement, which mean upon completion of the program of study they must +reside in their country of last legal permanent residence for two years before they are +eligible to apply for entry into the U.S. on an immigrant visa or on specialized work +visa (H or l). there is no foreign residence requirement for students on F-1 student +status. refer to the department of State’s website for additional information on the +exchange Visitor Program. + +• Medical Insurance: Students in J-1 status and their J-2 dependents are required to +have comprehensive medical coverage in order to fulfill U.S. government regula- +tions governing the J exchange Visitor status. Currently the medical insurance +offered by Hobart and William Smith Colleges does not cover all requirements for +those in J immigration status. the U.S. department of State specifies that program +participants and their dependents have medical coverage with the following mini- +mum requirements (22 CFr 62:14): medical benefits of at least $50,000 per accident +or illness; repatriation of remains in the amount of $7, 500, medical evacuation of +the exchange visitor to his or her home country in the amount of $10, 000, and a +deductible not to exceed $500 per accident or illness. For more information please +contact the international Student adviser. + +• Work Permission: Students in either F-1 or J-1 immigration status may work part +time (up to 20 hours per week) on campus with permission from the international +Student adviser. For both statuses, permission to work off campus based on eco- +nomic need may be requested only after the first full academic year of student status, +and only under extraordinary circumstances of unforeseen need. the United States +Citizenship and immigration Services (USCiS) authorizes F-1 students to work off +campus, whereas J-1 students submit an application for work permission to their visa +sponsor, i.e. the agency or school that issued the dS-2019 form. + +• Practical or Academic Training: Students in F-1 and J-1 immigration status are +eligible to engage in differing lengths of practical or academic training during their +studies and at the completion of their academic programs. Practical training, a +privilege of F-1 immigration status if you are eligible, is employment in one’s field of +study. those in F-1 immigration classification may be eligible for “curricular” practi- -16 +20 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 + cal training (which is authorized by the international Student adviser) during their +programs, in addition to one year of “optional” practical training that may take place +during or after one’s program of study. Permission to engage in practical training may +be granted only after you have been in valid student status for one academic year, +except in a graduate program that requires it from the first term. the international +Student adviser recommends optional practical training and the USCiS authorizes +it. refer to the Practical training overview for complete information on both cur- +ricular and optional practical training. academic training, a privilege of J-1 immi- +gration status if you are eligible, is employment which is integral, not just related, +to one’s field of study. one is eligible to apply for permission to engage in academic +training after one academic term in valid student status. a total of 18 months of +academic training is possible during or following completion of a program of study, +with the exception of some postdoctoral research positions that may qualify for up +to 36 months. an application for academic training is made to the J-1 student’s +exchange Visitor program sponsor. refer to academic training authorization for +those in J-1 exchange Visitor immigration Status for complete information. + +ARRIVING AT A U.S. PORT OF ENTRY … +WHAT A STUDENT CAN EXPECT +Plan your arrival: You may be refused entry into the United States if you attempt to +arrive more than 30 days before the start date listed on your SeViS i-20 form. + +Always carry your passport and I-20 on hand: do not check the following docu- +ments in your baggage. if your baggage is lost or delayed, you will be unable to pres- +ent the documents at your port of entry. as a result, you may not be able to enter +the United States. + +In addition, it is strongly recommended that you carry on hand the following + +documentation: + +1. evidence of financial resources +2. evidence of student status, such as tuition receipts +3. name and contact information for your international Student adviser, + +including a 24-hour emergency contact number at the school + +Complete your entry paperwork: +Arriving by Air: Flight attendants will distribute Customs declaration Forms and +arrival departure Forms (i-94). these must be completed prior to landing. + +Arriving by Land or Sea: the CbP officer at the port of entry will provide the +necessary Customs declaration Form and the arrival-departure Form (i-94) to be +filled out upon arrival. + as you arrive at the port of entry proceed to the terminal area for arriving +passengers. Have the following documents available for presentation: you passport, +i-20, arrival-departure Form (i-94), and the Customs declaration Form. Your docu- +ments will be examined and processed by the department of Homeland Security +(dHS) at the port of entry. Your passport will be returned to you along with an i-94 +card (arrival/departure record) designated “d/S” which means “duration of status” - Online ISSN: 2054-0965 (Online) +21 - Print ISSN: 2054-0957 (Print) + and refers to the time you are engaged in a full-time course of study as indicated on +your visa document. the dHS official will stamp and return your i-20 to you. + +Note: Your documents are extremely important. They legitimize your presence in +the United States. Keep them in a safe and secure place. + +For Further Information: +Preparing to Study in the U.S. http://educationusa.state.gov/ +applying for a Visa to Study in the U.S. http://travel.state.gov/visa/temp/types/ +types_1268.html + +MAINTAINING & BEING IN STATUS WITH DHS +an F-1 student is considered in-status for the entire length of time during which +he/she is enrolled as a full-time student in an educational program if adhering to +the conditions of the visa designated on the i-94 card, plus any authorized period of +practical training and an additional sixty (60) days to prepare for departure. be sure +to pay close attention to the conditions described below: + +• report to the international Student adviser for SeViS registration upon arrival +at Hobart and William Smith Colleges. + +• attend only the school authorized. this means that you must only attend the +school whose i-20 you last used to enter the U.S. or the school whose i-20 you used +to transfer schools. if you decide to withdraw from the Colleges before completing +your program, you are not allowed to remain in the U.S. with in-student status un- +less you transfer to another institution. + +• Complete a full course of study for each academic term and make normal academ- +ic progress towards completion of program. attendance and/or full-time enrollment +is not required during the summer. at Hobart and William Smith Colleges 4 credit +hours per semester is required and exceptions can be made on the recommenda- +tion of the international Student adviser for valid medical or academic reasons as +defined by the dHS. be sure to speak with the international Student adviser about +any exceptions beFore dropping below full-time status in order to avoid the seri- +ous consequences that may result from being out of status. + +• transfer or extend program in a timely manner. duration of Status expires if a +student takes longer than expected to complete a given academic level (e.g. bachelor’s, +master’s, and doctorate). the amount of time permitted for completing studies at a +given level is determined by the date on the initial i-20 issued at the beginning of each +academic level. Students must pay close attention to the expected completion date +noted in item 5 on the form i-20 for each level. if more time is needed, an applica- +tion for extension of stay must be filed with the dHS. there are certain requirements +to meet in order to be eligible for an extension, so you will need to consult with the +international Student adviser at least 45 days before the completion date. - Website: https://www.eajournals.org/ +22 - Publication of the European Centre for Research Training and Development -UK + • Keep continuity in the program of study (in general an absence from the U.S. of +more than 5 months terminates status. However, an authorized study abroad pro- +gram does not break continuity). -Table 2. Cloud Adoption: Cost Reduction and Efficiency Analysis [5, 6]. +• refrain from unauthorized work. When you apply for an F-1 visa, you must prove +that you have full financial support. the law assumes that you will not be employed +in the U.S. and therefore will not accept employment that could be filled by a U.S. +worker or student. there are some exceptions to this policy provided that full-time +enrollment during the year is maintained and the appropriate permission is ob- +tained, but you should always check with the international Student adviser before +accepting employment, especially if it is off campus. -Metric Category +• When traveling abroad, always return in proper F-1 status. -Q1 2024 (%) Q2 2024 (%) Q3 2024 (%) Q4 2024 (%) +• maintain a valid passport for 6 months into the future (Some countries have +agreements with the U.S. that recognize the passport as valid for 6 months after the +expiration date). -IT Budget Cloud Spend -IT Cost Reduction -Infrastructure Cost Savings -Development Cycle Reduction -Market Response Improvement +• report any address changes to the international Student adviser within 10 days +of the change. dHS regulations require that the international Student adviser +maintain a record of the current address where you are living. therefore, you must +always notify the international Student adviser immediately of any change of ad- +dress. addresses with a post office box number or addresses of your academic depart- +ment are not acceptable to the dHS. -15 -20 -15 -35 -25 +• Comply with Special registration requirements if applicable (nSeerS). -20 -23 -17 -40 -30 +• File timely requests of change of status, practical training and other benefits. -25 -27 -18 -45 -35 +• obey all State and Federal laws. -30 -30 -20 -50 -40 - -Advanced Technologies Enabled by Cloud Computing - -Artificial Intelligence and Machine Learning -Cloud platforms have fundamentally democratized access to AI and ML capabilities, creating -unprecedented opportunities for organizations across industries. According to recent market analysis, the -global artificial intelligence market size was valued at USD 454.12 billion in 2023 and is projected to surge -to USD 2,575.16 billion by 2032, growing at an impressive CAGR of 21.4%. This remarkable growth is -driven by the increasing adoption of cloud-based AI solutions across various sectors, including healthcare, -automotive, and retail. The integration of AI with cloud computing has become particularly crucial in the -post-pandemic era, with organizations accelerating their digital transformation initiatives to maintain -competitive advantage [7]. - -The impact of cloud-enabled AI extends across multiple domains, with North America leading the market -share at 35% in 2023. This dominance is attributed to the presence of major technology companies and -widespread adoption of AI technologies in various industries. The Asia-Pacific region is experiencing the -fastest growth, with a projected CAGR of 24.3% during the forecast period, driven by increasing -investments in AI research and development, particularly in countries like China, Japan, and South Korea -[7]. - -Big Data Analytics and IoT Integration -The transformation in data handling and analysis capabilities through cloud computing has revolutionized -how organizations process and utilize information. Edge-cloud infrastructure has emerged as a crucial -component in handling the massive data generated by IoT devices, with studies showing that edge -computing can reduce data transmission costs by up to 30% while improving response times significantly. -The integration of advanced analytics algorithms in edge-cloud architectures has enabled real-time -processing capabilities, crucial for applications requiring immediate insights and actions [8]. - -The evolution of IoT data analytics has led to the development of sophisticated middleware architectures -that efficiently manage data flow between edge devices and cloud platforms. Research indicates that hybrid -edge-cloud solutions can reduce latency by up to 50% compared to traditional cloud-only approaches. These -architectures implement various data analytic algorithms, including deep learning, reinforcement learning, +• depart form the U.S. in a timely manner. -17 +• do not enter on a tourist visa (b-1 or b-2). - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Note: Students who fail to maintain full-time enrollment are not eligible for any +of the benefits of the student visa, such as on-campus and off-campus employ- +ment, practical training, remaining in the U.S. during summer vacation, transfer +to another school or change of level, or registration for further courses. they are +considered to be “out of status” and must apply to dHS for reinstatement to valid +student status. eligibility for benefits will be regained only if and when the dHS +approves the reinstatement. if you have any questions regarding reinstatement, see +the international Student adviser immediately. - Online ISSN: 2054-0965 (Online) +23 - Print ISSN: 2054-0957 (Print) + EXCEPTIONS TO A FULL COURSE OF STUDY - Website: https://www.eajournals.org/ +• the normal vacation period; - Publication of the European Centre for Research Training and Development -UK +• medical problems documented by a licensed medical doctor, osteopath or clinical +psychologist. approval for reduced or no course load for a medical condition is limited +to 12 months total for the program of study; -and federated learning, enabling more efficient processing of IoT data streams. The middleware layer plays -a crucial role in managing these processes, with modern implementations supporting dynamic resource -allocation and automated scaling based on workload demands [8]. +• teaching or research assistantship; -The convergence of IoT and cloud computing has particularly benefited from advances in edge computing -capabilities. Contemporary edge-cloud infrastructures support multiple types of edge analytics, including -descriptive, diagnostic, predictive, and prescriptive analytics. This comprehensive approach enables -organizations to process data closer to its source, reducing bandwidth requirements while maintaining data -quality and analysis accuracy. Studies have shown that edge-cloud implementations can achieve up to 40% -improvement in energy efficiency compared to traditional cloud-only solutions [8]. +• Valid academic reasons, such as first semester language difficulties and/or adjustment +problems, improper course placement, unfamiliarity with american teaching methods; -Security considerations have become increasingly important in IoT data analytics, with edge-cloud -architectures implementing sophisticated security measures at multiple levels. The integration of -blockchain technology with edge-cloud infrastructure has emerged as a promising solution for ensuring -data integrity and secure communication between devices. Furthermore, the development of privacy- -preserving machine learning techniques has enabled organizations to maintain data confidentiality while -performing complex analytics tasks [8]. +• the final term when the student needs less than a fulltime course load to complete +program of study. Student must be enrolled in at least one credit; -Table 3. Cloud-Enabled Technologies: Growth and Efficiency Analysis [7, 8]. +• during a period of full-time optional or Curricular Practical training. -Technology Metric +DURATION OF STATUS FOR F-1 STUDENTS +an F-1 student is admitted to the U.S. for a period known as “duration of Status” +which is recorded upon entry documents i-94 and i-20 with the notation “d/S”. du- +ration of Status refers to the period a student is pursuing a full course of study, plus +any authorized practical training following completion. The dates on the I-20 de- +termine the initial parameters of the duration of status. A student can be admit- +ted to the U.S. up to 30 days prior to the report date on the I-20 and may remain +60 days after completion of program or practical training. Certain factors may +reduce grace period to 15 days or to 0 days. an international Student adviser au- +thorized withdrawal from classes is allowed a 15-day period for departure. a student +who fails to maintain status is not eligible for an additional period for departure. -2023 Q3 (%) 2023 Q4 (%) 2024 Q1 (%) 2024 Q2 (%) +TEMPORARY DEPARTURE FROM THE U.S. +If you plan on leaving the U.S. temporarily to visit your homeland or some other +country, an International Student Adviser must validate your I-20 for re-entry into +the U.S. You may not be allowed back into the U.S. without the validated immigra- +tion documents, a current visa, and a passport which is valid for at least six months +after you re-enter the country. + most students with valid status may travel to Canada, mexico and many of the +adjacent islands if they have a valid passport, a valid i-94 and a valid i-20 or dS-2019. +if you re-enter the U.S. in fewer than 30 days, an unexpired U.S. visa is not required. +However, Canada, mexico and the island nations do require entry visas for citizens of +certain countries. -North America AI Market Share -APAC Growth Rate -Data Transmission Cost Reduction -Edge-Cloud Latency Reduction -Energy Efficiency Improvement +it is always best to check with the embassy or consulate of any country you wish -35 -20 -20 -35 -25 +to visit. Since certain countries are ineligible for this benefit and travel regulations are +subject to change, you should check with the international Student adviser if you +intend to travel with an expired visa. -37 -22 -25 -40 -30 +Students who apply for a new visa in contiguous territory forfeit the right to re- -40 24 -28 -45 -35 -42 + enter with an expired visa. if the new visa is denied, the student would not be able to +re-enter the United States to study until a valid visa is obtained in their home country. + +ACCEPTING U.S. GOVERNMENT PUBLIC ASSISTANCE OR PUBLIC +BENEFITS +international students and scholars should not, under any circumstance, accept public +benefits from the federal, state, county or any other local government agencies in the +United States. this includes, but is not limited to, participation in programs such +as medicare/medicaid, Supplemental Social Security income, aid to Families with +dependent Children, food stamps, and subsidized housing. accepting such benefits +can cause you to lose your legal status in the United States and create a bar to your +re-entry to this country. + +Cooperation between state social service agencies and the inS and department + +of State make it possible for an inS inspector or U.S. Consular officer to become +aware of an individual who has accepted public benefits. the acceptance of public +benefits is interpreted as a declaration that the individual is unable to meet his or her +financial obligations (a requirement of obtaining and maintaining status) and thereby +rendering him or herself ineligible for F-1 or J-1 status. + occasionally a well-meaning social service agency, hospital or public housing admin- +istrator may offer such assistance not realizing that non-immigrants are ineligible for such +benefits. additionally, even though U.S. citizen children of non-immigrant parents may be +eligible for pre and post-natal public benefits, the non-immigrant parent may be denied a +visa or re-entry based on the child’s receiving the benefits. again this is interpreted as the +non-immigrant individual not being able to meet his or her financial obligations. + +Information on U.S. Social Security Numbers (SSN) +if you will be employed in the United States, including at Hobart and William Smith +Colleges, you must obtain a U.S. Social Security number, which is used by the U.S. +government to identify wage earners for tax purposes. the number assigned to interna- +tional students is not a valid U.S. Social Security number. + +to apply for a Social Security number (SSn) you must obtain a “proof of work” + +form from the international Student adviser confirming your student or exchange +visitor status, that you are authorized to engage in on-campus employment, the nature +of the employment to be engaged in, and identify the department/office for whom you +will be working before being able to apply to the Social Security administration for +a SSn. You must also present your passport, your i-20 or dS-2019 form and your i-94 +form. the Social Security administration office is open monday-Friday from 9:00 a.m. +to 4:00 p.m. (closed on federal holidays). + +Social Ssecurity numbers will not be issued to foreign nationals who are not +authorized to work. a Social Security number is not required for opening a bank +account, registering for school, filing a U.S. tax form or applying for a driver’s license. +if you are required to file a tax return and are not eligible for a U.S. Social Security +number, you will have to apply to the internal revenue Service (irS) for a tax iden- +tification number (tin) on form W-7. this form may be obtained from the Colleges +payroll department. + +For more information about U.S. Social Security Numbers visit the website or + +call their toll-free number at 1-800-772-1213. + +25 + + WORK OPPORTUNITIES FOR STUDENTS IN F-1 STATUS +The following is a list of all opportunities to obtain authorization to work in the +United States if you are in F-1 immigration status. Each category may allow for +authorization under different circumstances and at different times during your +stay in the U.S. This is summary of work benefits but does not include all regu- +lations governing employment. + +On-campus Work: Full-time students are allowed to work up to twenty hours per +week during the academic year. on campus employment means that you are paid by +Hobart and William Smith Colleges or working at a Colleges location with an em- +ployer that is providing services to students, such as Sodexho’s Saga dining hall or +the Café. on-campus employment requires an i-9 employment verification autho- +rization form to be completed by you at the Colleges business office. + +Off-campus Work Based on Financial Need: Students who have been in F-1 status +for one academic year and who are experiencing extreme financial difficulties due to +unforeseen changes that severely affect their financial resources may apply for work +permission. authorization is given by the dHS with the recommendation of the +international Student advisers/ dSo. + +Internship with an International Organization: Students in F-1 status are eligible +to apply for permission to work for international organizations such as the World +bank, the international monetary Fund, etc. authorization is granted by the dHS +with the recommendation of the international Student adviser/dSo. + +WORK OPPORTUNITIES FOR STUDENTS IN J-1 STATUS +J-1 On-Campus Employment +enrolled J-1 students may work on campus up to 20 hours per week during the +semester and full time during the break with the permission of their sponsor. if the +Colleges are your sponsor please see the international Student adviser. + +J-1 Off-Campus Employment +if a J-1 student needs employment during the academic year due to financial hard- +ship arising form unforeseen circumstances, the program sponsor may authorize it. +Such employment cannot interfere with full-time pursuit of the exchange visitor’s +program objective and is limited to 20 hours per week during the academic year and +full-time during break. + +PRACTICAL TRAINING: +this is defined as employment related to your field of study and is available dur- +ing and after completion of your studies. the following is a list of the categories of +practical training: + +Curricular practical training (before completion of your studies) is defined as + +employment pursuant to a required or optional internship which is an integral part +of the established curriculum of your school. + 26 -30 -50 -40 - -Emerging Trends in Cloud Computing - -Edge Computing and Serverless Architecture -Edge computing has emerged as a transformative force in the cloud computing landscape, bringing -computation closer to data sources and end-users. The integration of edge computing with serverless -architecture represents a significant evolution in cloud technology, addressing critical challenges in modern -application development and deployment. This combination has become particularly crucial for real-time -processing requirements, with edge computing reducing latency by processing data closer to its source -while serverless architecture eliminates the complexity of server management. The implementation of edge- -serverless solutions has shown remarkable improvements in application performance, with organizations -reporting reduction in latency from seconds to milliseconds for critical operations [9]. -The impact of serverless computing extends beyond performance improvements. Recent implementations -have demonstrated significant advantages in terms of cost efficiency and resource utilization. Organizations -18 + optional practical training is defined as employment related to your field of study. +it is authorized in variable increments, deducted form the maximum of one year of +optional practical training per educational level, and is available in the following +variations: - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -adopting serverless architectures have reported substantial reductions in operational overhead, as the -technology eliminates the need for traditional server management and scaling operations. The pay-as-you- -go model inherent in serverless computing has revolutionized cost structures, with businesses only paying -for actual compute time used rather than maintaining constantly running servers. Furthermore, the -integration of edge computing with serverless architecture has enabled more efficient handling of IoT -devices and real-time data processing requirements, resulting in improved application responsiveness and -user experience [9]. - -Hybrid Cloud Solutions -The hybrid cloud market has witnessed extraordinary growth and transformation, with current valuations -at USD 183.7 billion and projected to reach an impressive USD 405.62 billion by 2032, growing at a CAGR -of 9.2%. This remarkable expansion is driven by increasing demands for flexible, scalable, and secure cloud -solutions across various industries. The financial services sector has emerged as a particularly strong -adopter of hybrid cloud solutions, leveraging the technology to balance security requirements with -operational efficiency [10]. - -The adoption of hybrid cloud solutions has been significantly influenced by several key factors, including -the rising need for enhanced data security, disaster recovery capabilities, and the growing complexity of -regulatory compliance requirements. Organizations in the banking, financial services, and insurance (BFSI) -sector have been at the forefront of hybrid cloud adoption, particularly in North America, which currently -holds approximately 35% of the global market share. The integration of artificial intelligence and machine -learning capabilities within hybrid cloud environments has further accelerated adoption rates, enabling -organizations to optimize workload placement and resource utilization more effectively [10]. - -Market analysis reveals that large enterprises currently dominate the hybrid cloud market, accounting for a -significant portion of the overall market share. This dominance is attributed to their complex IT -requirements and substantial resources for implementation. However, small and medium-sized enterprises -(SMEs) are increasingly adopting hybrid cloud solutions, driven by the technology's ability to provide -enterprise-grade capabilities without requiring extensive upfront investments. The public sector has also -emerged as a significant growth driver, with government agencies increasingly migrating to hybrid cloud -environments to modernize their IT infrastructure while maintaining strict security and compliance -requirements [10]. - -Looking Ahead: Future Developments -The convergence of edge computing, serverless architecture, and hybrid cloud solutions continues to -evolve, with emerging technologies and methodologies shaping the future of cloud computing. Edge- -serverless platforms are becoming increasingly sophisticated, incorporating advanced features such as -automated scaling, improved security protocols, and enhanced development tools. The integration of -artificial intelligence and machine learning capabilities at the edge is expected to further transform -application development and deployment paradigms, enabling more intelligent and responsive systems [9]. +1. optional practical training after completion of your studies +2. optional practical training before completion of your studies with three pos- -19 +sible variations: - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +a. practical training during the annual vacation. the annual vacation must - Online ISSN: 2054-0965 (Online) +follow at least two consecutive terms of fulltime enrollment and is usually taken in +the summer - Print ISSN: 2054-0957 (Print) +b. practical training when school is in session, provided that employment - Website: https://www.eajournals.org/ +does not exceed 20 hours a week - Publication of the European Centre for Research Training and Development -UK +c. practical training following the completion of all the course requirements -Table 4. Cloud Architecture Evolution: Latency and Processing Metrics [9, 10]. +for the degree except the thesis or equivalent -Performance Indicator +Optional Practical Training (OPT) Before Completion of Degree Requirements +definition: employment related to your field of study before completion of all degree +requirements. it is authorized in variable increments and is deducted from the +maximum of one year (12 months) of optional practical training per educational +level. there are three circumstances under which this type of practical training can +be authorized: -Traditional Edge Only Serverless Edge-Serverless +1. Following completion of all degree requirements except the thesis or equivalent +2. during the annual vacation +3. Part-time (up to 20 hours per week) during the academic year -Processing Latency (ms) -Resource Overhead (%) -Server Management Time (hrs/week) -Response Time (ms) -Infrastructure Complexity Score +Eligibility to Engage In OPT: You must: +• be a registered student in F-1 status physically in the United States at the time of +application +• have been enrolled in lawful student status on a full-time basis for at least one full +academic year +• not have used twelve months or more of full-time curricular practical training +oPt is permitted for a period of up to a maximum of twelve months per educational +level (bachelors, masters, etc.). -45 -65 -40 -50 -75 +Procedure: obtaining authorization for oPt is a two-step process. Your first step in +obtaining an employment authorization document (ead) is to qualify for a certifica- +tion from the international Student adviser/dSo recommending you for oPt. the +international Student adviser/dSo cannot authorize oPt - only the United States +Citizenship and immigration Services (USCiS) can do so by issuing you an ead. -25 -45 -30 -35 -55 +You may not begin employment until you have been issued an ead by the US- -15 -35 -20 -20 -40 +CiS and can work only within the time period indicated on the card. requests for +employment authorization must be mailed to the USCiS regional Service Center +having jurisdiction over your place of residence and may require 6 to 12 weeks for +adjudication. -5 -25 -15 -8 -30 +27 -Security and Compliance in the Cloud Era -The landscape of cloud security continues to evolve rapidly in response to emerging threats and changing -business requirements. Recent analysis indicates that cloud security remains a top priority for organizations, -with 82% of companies increasing their cloud security budgets in 2024. The shift towards zero trust -architecture has become particularly prominent, with 76% of organizations either implementing or planning -to implement zero trust security models. This transition has been driven by the growing recognition that -traditional perimeter-based security approaches are insufficient in today's distributed cloud environments, -where data and applications are accessed from multiple locations and devices [11]. - -Identity and access management has emerged as a critical component of modern cloud security strategies. -The implementation of comprehensive IAM solutions has become essential as organizations grapple with -increasingly sophisticated cyber threats. Studies show that organizations with mature identity and access -management practices experience significantly fewer security incidents, with advanced authentication -mechanisms reducing unauthorized access attempts by up to 67%. The adoption of biometric authentication -and behavioral analytics has further strengthened security postures, with organizations reporting a 45% -reduction in credential-based attacks [11]. - -The financial implications of data breaches have underscored the importance of robust cloud security -measures. According to IBM's Cost of a Data Breach Report 2024, the global average cost of a data breach -has reached $4.45 million, marking a 15% increase over three years. Organizations leveraging AI and -automation in their security operations have demonstrated significantly better outcomes, with automated -security tools reducing the average time to identify and contain breaches by 108 days. Furthermore, -companies implementing zero trust architecture have seen an average reduction of $1.17 million in breach -costs compared to those without such security measures [12]. - -The impact of remote work and hybrid environments continues to shape cloud security strategies. -Organizations with over 50% of their workforce working remotely have reported higher average breach -costs, reaching $4.99 million compared to $4.24 million for organizations with less remote work. This -disparity has driven increased adoption of advanced security technologies, with 49% of organizations now -implementing AI platforms for cybersecurity. The study reveals that organizations using security AI and + When to Apply: +• not more than 90 days prior to being enrolled for one full academic year· +• not more than 120 days prior to the employment start date that you select +Please note: the employment start date cannot be earlier than the date on which +you will have completed one full academic year. + +How to Apply: +Obtaining the International Student Adviser recommendation +Since you must first obtain a new SeViS i-20 with the international Student ad- +viser/dSo’s oPt endorsement, which may require 5 working days, you should make +every effort to apply as early as possible within the 90 or 120 day window. You might +need to come twice to the international Student adviser/dSo—once to submit the +application and another to pick up the approved recommendation on a SeViS i-20. + +Curricular Practical Training in F-1 Status +Definition: Curricular practical training (CPt) is employment pursuant to an intern- +ship requirement or internship elective which is an integral part of the established +curriculum of your school, which will be counted towards completion of your degree. + +Eligibility to engage in CPT: You must be a degree candidate who has been in +lawful status for at least one academic year. You must have been offered a specific +training opportunity that must fall into one of the following categories: +• to satisfy an internship requirement expected of all students in your program +and necessary for completion of your degree; +• to satisfy an internship elective which counts towards completion of your degree. +the internship elective must be listed in your school’s bulletin and must state in the +official course description that it is specifically designed to allow students enrolled in +the course to engage in off-campus internships; +• the internship may be done to engage in research necessary and integral to the +completion of the doctoral dissertation. + +Restrictions: U.S. Citizenship and immigration Services (USCiS) regulations do +not place an absolute limit on the amount of CPt you may utilize. However, if you +engage in one year or more of full-time CPt, you will lose your eligibility to engage +in optional Practical training. CPt is always part-time during the academic year +unless the student is engaged in doctoral dissertation research. CPt may be full time +during the annual vacation. You may engage in curricular practical training only +after your SeViS record has been updated by the international Student adviser and +the i-20 has been printed, indicating the approved employment. the endorsement +will indicate the precise details of the authorized training, including the name and +location of the employer, the specific dates of the training period, and whether it +is full-time or part-time. We encourage you to contact the international Student +adviser/dSo if you have questions. -20 +28 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -automation experienced breach lifecycle times that were 108 days shorter compared to those not utilizing -these technologies [12]. - -Cloud security compliance has become increasingly complex with the evolution of regulatory requirements -and data protection standards. Recent trends show a significant shift towards automated compliance -monitoring and reporting tools, with organizations implementing these solutions reporting 43% lower -compliance-related costs. The adoption of cloud security posture management (CSPM) tools has grown -substantially, with 65% of organizations now using these platforms to maintain continuous compliance and -security assessment capabilities. This has led to a 55% improvement in the detection and remediation of -compliance violations [11]. - -Future Outlook: Cloud Computing and Digital Transformation -The landscape of cloud computing continues to evolve rapidly, with organizations increasingly focusing on -strategic cloud adoption and optimization. According to recent industry analysis, 2024 marks a significant -shift toward more sophisticated cloud strategies, with 94% of enterprises already utilizing cloud services -and 87% adopting multi-cloud approaches. This evolution is particularly noteworthy in the realm of cloud -cost optimization, where organizations are implementing FinOps practices to better manage and optimize -their cloud spending. The trend toward cloud-native development has become increasingly prominent, with -enterprises reporting 30-40% faster application development cycles through cloud-native approaches [13]. -The integration of artificial intelligence and machine learning capabilities has become a cornerstone of -modern cloud strategies. Organizations are increasingly leveraging AI-powered cloud services for various -purposes, from automated infrastructure management to intelligent application development. The adoption -of cloud-native platforms has enabled businesses to achieve remarkable improvements in operational -efficiency, with companies reporting 25-35% reduction in development cycles and 40-50% improvement -in resource utilization. Furthermore, the implementation of containerization and microservices architectures -has led to significant improvements in application scalability and maintenance efficiency [13]. - -Sustainability and environmental responsibility have emerged as critical priorities in cloud computing -strategies for 2024 and beyond. Organizations are increasingly focusing on green cloud computing -initiatives, with major providers implementing advanced energy management systems and sustainable data -center practices. This shift towards environmentally conscious cloud operations has become a key -differentiator in vendor selection, with businesses prioritizing providers that demonstrate strong -commitments to sustainability. The industry is witnessing a growing emphasis on energy-efficient cloud -architectures and sustainable computing practices [14]. - -Security and compliance considerations continue to shape the future of cloud computing, with zero trust -architectures gaining prominence. The industry has seen a significant shift toward comprehensive security -frameworks that encompass everything from infrastructure protection to application security. Organizations -are increasingly adopting automated security solutions and implementing advanced threat detection -systems. The focus on regulatory compliance has intensified, with businesses implementing sophisticated + Departure Checklist -21 +q admissions letter from HWS - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +q evidence of Financial resources - Online ISSN: 2054-0965 (Online) +q Passport with Student Visa - Print ISSN: 2054-0957 (Print) +q Form 1-20 or dS-2019 - Website: https://www.eajournals.org/ +q Copies of medical records and other important documents translated into english - Publication of the European Centre for Research Training and Development -UK +q if J-1 visa holder – additional health insurance requirements. this must also be -compliance monitoring and management tools to ensure adherence to evolving regulatory requirements -[14]. +translated into english, if needed. -The evolution of cloud management and governance frameworks represents another significant trend. -Organizations are implementing more sophisticated approaches to cloud governance, incorporating -advanced monitoring tools and automated policy enforcement mechanisms. The emphasis on operational -excellence has led to the adoption of advanced observability platforms and AIOps solutions, enabling better -visibility and control over cloud environments. This trend is particularly evident in the increasing adoption -of cloud management platforms that provide comprehensive visibility and control across multi-cloud -environments [14]. +q emergency outfit packed in carry-on luggage -CONCLUSION +q Calling card to use during travel -Cloud computing stands as a transformative force in the digital era, reshaping how organizations operate, -innovate, and deliver value. The convergence of cloud technologies with AI, machine learning, and IoT has -created powerful new capabilities for businesses of all sizes. As organizations continue to adapt to changing -market demands, the role of cloud computing extends beyond technological infrastructure to become a -strategic enabler of business transformation. The future holds promise for further innovations in edge -computing, serverless architectures, and sustainable cloud practices, while security and compliance remain -paramount considerations. Organizations that successfully leverage these cloud capabilities while -maintaining robust security measures and embracing sustainable practices will be best positioned to thrive -in an increasingly digital world. +q emergency phone numbers (HWS: 1-315-781-3333) -REFERENCES +q minimum of 100 U.S. dollars for use in an emergency -[1] Businesswire, "IDC Spending Guide Sees Worldwide Digital Transformation Investments Reaching +q Sweater or jacket in event of bad weather -$3.4 Trillion in 2026," 2022. [Online]. Available: -https://www.businesswire.com/news/home/20221026005193/en/IDC-Spending-Guide-Sees- -Worldwide-Digital-Transformation-Investments-Reaching-%243.4-Trillion-in-2026 +q if taking prescription medicine bring prescription for future refills (translated into -[2] Noah Barsky, "Majority Of Digital Transformations Off Track, Executive Survey Finds—Here Are 3 +english) -Fixes," Forbes, 2023. [Online]. Available: -https://www.forbes.com/sites/noahbarsky/2023/10/02/people-problems-plague-digital- -transformation/ +q Photocopy of your passport, visa, etc. do not carry in the same place as your -[3] Arbab Khan, "Unlocking the Power of Digital Transformation and Automation," Rushford, 2024. +original documents. -[Online]. Available: https://rushford.ch/insights/unlocking-the-power-of-digital-transformation- -and-automation/ +29 -[4] George, "On-Premises vs. Cloud Data Center: Which Is Right for Your Business?" FS, 2022. + SECTION III: +*SAMPLE* ORIENTATION SCHEDULE *SAMPLE* -[Online]. Available: https://www.fs.com/blog/onpremises-vs-cloud-data-center-which-is-right- -for-your-business-5867.html +Hobart and William Smith Colleges +international Students orientation +“let Your adventure begin!” +august 22 -25, 2006 -[5] Neal Weinberg, "5 cloud market trends and how they will impact IT," CIO, 2024. [Online]. Available: +Tuesday, August 22 +8:30 a.m. –7:00 p.m. -https://www.cio.com/article/655693/7-cloud-market-trends-and-how-they-will-impact-it.html +Wednesday, August 23 +8:30 a.m. –12:00 p.m. -22 +arrival/Check-in +location: office of intercultural affairs +288 Pulteney Street +**bring all immigration paperwork - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +arrival/Check-in +location: office of intercultural affairs, +288 Pulteney Street +**bring all immigration paperwork - Online ISSN: 2054-0965 (Online) +HWS Student id Cards +location: registrar’s office - Print ISSN: 2054-0957 (Print) +Campus tour - Website: https://www.eajournals.org/ +Shopping @ Wal-mart & Wegmans Supermarket, +Van leaves from intercultural affairs parking lot - Publication of the European Centre for Research Training and Development -UK +12:30 p.m. - 1:30 p.m. -[6] Will Forrest et al., "Cloud’s trillion-dollar prize is up for grabs," McKinsey Digital, 2021. [Online]. +lunch with deans of the Colleges +location: blue room in Scandling Center -Available: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion- -dollar-prize-is-up-for-grabs +1:45 p.m. – 2:45 p.m. -[7] Precedence Research, "Artificial Intelligence (AI) Market Size, Share, and Trends 2025 to 2034," +overview of library Services +location: Warren Hunting Smith library -2025. [Online]. Available: https://www.precedenceresearch.com/artificial-intelligence-market -[8] Abel E. Edje, M.S. Abd Latiff and Weng Howe Chan, "IoT data analytic algorithms on edge-cloud +6:00 p.m. – 7:30 p.m. -infrastructure: A review," ScienceDirect, 2023. [Online]. Available: -https://www.sciencedirect.com/science/article/pii/S2352864823001530#:~:text=In%20addition% -2C%20future%20advances%20are,Comparison%20of%20previous%20research%20surveys.&te -xt=Comparison%20of%20middleware%20architectures. +dinner +location: intercultural affairs Center -[9] IvanDev, "The Rise of Serverless Architecture and Edge Computing," DEV, 2025. [Online]. +Thursday, August 24 +7:30 a.m. – 8:00 a.m. -Available: https://dev.to/ivansing/the-rise-of-serverless-architecture-and-edge-computing-2aab +8:15 a.m. 9:45 a.m. -[10] Globe Newswire, "Hybrid Cloud Market Projected to Hit USD 405.62 Billion by 2032| SNS +30 -Insider," 2025. [Online]. Available: https://www.globenewswire.com/news- -release/2025/03/21/3047113/0/en/Hybrid-Cloud-Market-Projected-to-Hit-USD-405-62-Billion- -by-2032-SNS-Insider.html +breakfast +location: Saga dining Hall in the blue room -[11] Thales, "Cloud Security in 2024: Addressing the Shifting Landscape," CSA, 2024. [Online]. +immigration regulations overview +location: Conference room, 2nd Floor, +intercultural affairs -Available: https://cloudsecurityalliance.org/blog/2024/06/27/cloud-security-in-2024-addressing- -the-shifting-landscape + 10:00 a.m. – 11:00 a.m. -[12] MAHESH NAWALE, "7 Key Takeaways From IBM's Cost of a Data Breach Report 2024," ZScaler, -2024. [Online]. Available: https://www.zscaler.com/blogs/product-insights/7-key-takeaways-ibm- -s-cost-data-breach-report-2024 +11:05 a.m. – 11:35 a.m. -[13] Andrii Stashko-Ziablietsov and Oleksandr Virga, "Cloud strategy: a 2024 guide," Avenga, 2024. +on Campus employment/Social Security +elaine Ferrara, business office +location: tba -[Online]. Available: https://www.avenga.com/magazine/cloud-strategy/ +benefits of the HWS Health insurance +ann brink, bursars office +location: tba -[14] Willie Devitt, "Cloud Technology Trends and Predictions for 2024," Eviden, 2024. [Online]. +12:00 p.m. – 1:30 p.m. -Available: https://eviden.com/publications/whitepapers/cloud-technology-trends-and-predictions- -for-2024/#:~:text=reputation%20and%20trustworthiness- -,At%20Eviden%2C%20we%20predict%20that%20in%202024%2C%20the%20focus%20on,com -pliance%20posture%20for%20our%20customers. +lunch with Chaplain lesley adams +location: Chaplain’s residence, 630 S. main St. -23 +1:45 p.m. – 2:15 p.m. + +money matters +beth turner, director of Financial aid +location: tba + +2:30 p.m. – 3:00 p.m. + +a Healthy You +location: Hubbs Health Center + +3:15 p.m. – 3:45 p.m. + +5:30 p.m. – 7:30 p.m. + +Friday , August 25 +10:00 a.m. – 11:00 a.m. + +Study basics +Center for teaching & learning +location: Center for teaching & learning +(Ctl) + +Welcome dinner +President and mrs. mark gearan & Family +location: President’s House – 690 South main St. + +guide to U.S. education & Culture +amy teel and doug reilly, Center for +global education +location: delancey House + +12:00 p.m. – 1:00 p.m. + +lunch +location: delancey House + +1:30 p.m. + +Join regular campus wide orientation schedule + +Welcome to Hobart and William Smith Colleges + +31 diff --git a/q103/random_k4/random_3.md b/q103/random_k4/random_3.md index fea36abbe8215721fb9751f54461e22cbeefc712..602df279b63a2b7c5600ed9bc60d31b5343359ac 100644 --- a/q103/random_k4/random_3.md +++ b/q103/random_k4/random_3.md @@ -1,1211 +1,2008 @@ -© Lonely Planet Publications Pty Ltd +INTERNATIONAL STUDENT +HANDBOOK -BECKY OHLSEN +A resource guide published by the +Office of International Education +Albany State University +Albany, GA -SEATTLE + TABLE OF CONTENTS +Notice to Students………………………………………………………………..……………………….. -C I T Y G U I D E +Welcome Letter from ISA……………..…………………………………………………………..………….…………………. - I NTRO DUCI N G SE AT TLE +Financial Documentation………………………………………………………….………………..…….. -Kayaking on Lake Union (p166) +Academic Life………………………………………………………………………………………………………….…………….. -RICHARD CUMMINS +American Customs and Culture………………………………………………………………………………..………… -It’s raining again, but you don’t mind. This is your cue -to duck inside for another latte. Maybe you’ll chase it -with a local microbrew, then tackle the Space Needle -before rocking the Crocodile Café. +Immigration Status…………………………………….……………………………………………………………..……………. -Welcome to Seattle, whose perpetually cloudy skies mean that if the sun peeks out for a second, -everyone drops whatever they’re doing and gears up to take their kayaks or bicycles out. They -can do this because they’re part-time baristas, freelance IT consultants, club bookers or chefs -at all-organic small-plates restaurants. They’re as likely to drink PBR as IPA, but they’d rather -die than sip truck-stop coffee or down a quarter pounder with cheese. They always recycle. -They’ve been on skid road, and they know Alki has nothing to do with the drinking habits -of barflies in Pioneer Square. If you ask them, they’ll tell you it really doesn’t rain that much -around here. +Automobiles…………………………………………………………….………………………..…………. -They might also tell you to watch the ‘flying fish’ at Pike Place Market early in the morning, -visit the Olympic Sculpture Garden around sunset and take an evening ferry ride to Bainbridge -Island. Most likely they’ll say not to bother with the Space Needle unless the weather is bril- -liant. If you’re game for a bar crawl, they might take you to Georgetown, the neighborhood for -drinkers in the know. If you’re arty, it might be an opening at an edgy Belltown gallery instead. -Or you could just get hopped up on caffeine and go thrifting all day. Despite what you hear, -you may be surprised at how little anyone minds the fickle weather. +Emergencies……………………………………………………………………………………………………………………………. -2 +English Language…………………………………………………………………………………………… - SEATTLE LIFE +Entertainment………………………………………………..………………………………………………………………………. -Seattle is the largest metro area in the -Pacific Northwest, with a population of -about 3.3 million (580,000 in the city -proper). It’s a bookish, well-educated -place, but also a dynamic and inventive -urban center – most of the city’s recent -economic growth has been fueled by tech- -nology: both high-tech and the old-school, engine-parts variety. The Port of Seattle is the -second-largest handler of container cargo in the United States. +Health Care…………………………………………………………………………………………………… -‘Seattle sits on the verge -of major structural and -demographic change’ +Housing……………………………………………………………………………….…………………………………………….…. -Two of its most successful citizens also happen to be Seattle’s two largest public figures, Bill -Gates and Paul Allen. And they symbolize a certain aspect of the city’s contradictory attitude -toward its own success. Both undeniably ambitious and indisputably successful, Gates and -Allen are seen simultaneously as points of civic pride and shameless capitalists who are totally -alien to the prevailing Seattle culture. +Dining Services……………………………………………………………………………..….…………… -And speaking of culture, you’ll be forgiven for thinking Seattle’s is mostly white – at least if -you hang out in hipster coffee shops or go to many indie-rock shows – but Seattle has a rich -multicultural history and is home to Native American, African American, Asian American and -growing Ethiopian American populations. The city ranks 23rd of the 30 largest US cities in -African-American population, at 8.4% – higher than every West Coast city except Los Angeles. -It also has the largest concentration of Native Americans in the Pacific Northwest. The largest -group in urban Seattle is the Duwamish people, the Native Americans who originally lived on -the shores of Elliott Bay. +Legal Concerns………………………………………….…………………………………………..……………………………… -It’s been a neighborhood-centric city for most of its history, but Seattle now sits on the verge -of major structural and demographic change. The area expects 40% population growth in the -next two decades, and the regional transportation board has already begun reshaping the city’s -landscape with new mass-transit plans. It’s an exciting time to visit an always exciting place. +Mail………………………………………..…………………………………………………………………… -Marching girls parade through Seattle's International District (p59) +Money and Banking……………………..……………………………………….………………………………………………... -LAWRENCE WORCESTER +Regent’s Test…………………………………………………………………………………….…………. + +Religious Life……………………………………..…………………………………………………….…………………………… + +Shopping…………………………………………………………………………..………………………….. + +Student Athletes ………………………………………….……………………………………………………………..……….. + +Social Security Numbers ……………………………………………..…………………….…………… + +Taxes……………………………………………….…………………………………………………………………………………… 3 - H I G H LI G HT S +4 + +5 + +6 + +10 + +12 + +18 + +19 + +21 + +21 + +22 + +24 + +26 + +27 + +28 + +29 + +30 + +32 + +32 + +33 + +34 + +35 + +Telephone and Telegrams ……………………….……………………………………………………….. 35 + +Testing …………………….………………….……………………………………………………………..… + +Transportation ……………………….…………………………………………………………………………………………… + +Campus Contacts…………………………………………………………………………………………………………………… + +Checklists……………………………………………………………………………………………………… + +Services for International Students ………………………………………………………….………………………………. + +36 + +36 + +38 + +39 + +40 + + Office of International Education + +NOTICE TO STUDENTS +This guide should be used in combination with the ASU Undergraduate +Catalog/ASU Graduate Catalog and the RAM Student Handbook. The University +Catalog contains important information and regulations of the university that you +will need to know. You should obtain a copy of the Catalog (Undergraduate or +Graduate) and the RAM Student Handbook as soon as you begin your program +of study. + +3 + + Office of International Education + +Dear International Student: + +It gives me great pleasure to welcome you to the campus of Albany State University. I +will serve as your International Student Advisor. I want you to know that you are a +valuable part of the ALBANY STATE UNIVERSITY family like any other student and most +of all I want you to be successful in your studies. + +To help you reach this goal, I offer services that include providing you with important +information about your visa status and the benefits it includes. There are laws and +regulations you are required to follow in order to legally remain in the U.S. and I am +required to submit updates and other reports regularly and by specific deadlines to +confirm that. + +This guide is designed to help you locate the resources and services you will need while +you are at ASU and answer some of your questions. You are encouraged to ALWAYS +know your F-1 Visa Student Status and regulation involved. + +As an international student at ALBANY STATE UNIVERSITY you are required to meet +with me each semester of your matriculation. I encourage you to schedule +appointments so that we can meet to discuss and review options available to you at any +point of your educational endeavors at ALBANY STATE UNIVERSITY. - +In addition, please note that while you are matriculating as an International Student +ASU currently does not provide any financial assistance for you. It is your responsibility +to have the financial resources to continue your education. If for any reason you are +unable to continue to make financial contribution to your education endeavors you are +at risk of being in violation of the student immigration status. (please see the next page) - +I look forward to assisting you at ALBANY STATE UNIVERSITY and seeing you reach the +end of your program as focused and ready for the future as you are here today. + +Sincerely, + +Sue Ann Balch, J.D. +F1-Visa Student Advisor +sueann.balch@asurams.edu (Email) +229-430-7075 (Office Number) +229-434-7106 (fax Number) +205 Reese Hall 4 -D OW N TOW N & FI R ST H I LL + Financial Documentation (Required Forms) +You can acquire these forms from the office of Recruitment and Admission -Downtown Seattle is the functional business heart of the city, with a jungle of office -buildings and an electric weekday buzz. It’s also home to many of the city’s main -hotels, as well as the Seattle Art Museum. Head to First Hill to see the stately, historic -Sorrento Hotel. +*Albany State University International Student Financial Statement (To Be Completed by Student) -LAWRENCE WORCESTER +*I-34 Affidavit of Support (To Be Completed By Sponsor, Affirmed Under Oath and Notarized) - Sorrento Hotel -Dabble in decadence at the Sorrento Hotel (p51) - Seattle Public Library -Find peace and prose at this stylish library (p49) - Seattle Art Museum -Gawp at Jonathan Borofsky’s 1992 sculpture -Hammering Man, which dominates the museum -entrance (p48) +Original/Official Supporting Financial Documentation Must be: - +1. Written in English or have certified English translation attached and a +recognized transcript evaluation by a company such as WES or Silney. +2. Dated no more than six months prior to the first day of the semester you -RICHARD CUMMINS +plan to attend. -RICHARD CUMMINS +3. Clearly state the account balance in United States Dollar amount(s). +4. Must provide the following information: Date account opened; total -  +amount deposited for the past year; +Tuition and fees +Living Expenses +Expenses of dependant +Other: Books & Health Insurance -LAWRENCE WORCESTER +$14,600.00 +$9,000.00 +$0.00 +$4,000.00 +*$27,600.00 -PIONEER SQUARE & INTERNATIONAL DISTRICT +(*Fee changes yearly, 2016 cost listed above) -Pioneer Square is the birthplace of Seattle, and its redbrick buildings and plazas give -it a historical, film-set feel. It’s also popular for its restaurants, bars and independent -shops, which are well worth a visit. The International District is home to Seattle’s -Chinatown and the Wing Luke Museum of the Asian Pacific American Experience. +5. All monies for tuition, health insurance, Room Board, Books & Supplies +and other fees must be must be paid by the first day of classes each +semester at Albany State University after student has been accepted. +This process must occur until the student has completed matriculation at +Albany State University - Wing Luke Museum of the -Asian Pacific American -Experience -Absorb yourself in Pan-Asian history (p59) - Pioneer Square Park -Step back in time to Seattle’s early days (p57) - Zeitgeist -Perk up on coffee in arty surrounds (p141) +How to set-up account? - +1. After student receives acceptance letter from -ANN CECIL +undergraduate or graduate admissions, student will +begin the account set-up process. - +2. Student will mail fees for the year in United States -CHUCK PEFLEY/ALAMY +Dollar Amount to: -5 +Albany State University +Financial Operations +Attn: Student Account Manager +504 College Drive +Albany, GA 31705 +3. Student will mail money certified or return receipt -  +request or any secure means of mail delivery or pay +in person. - +ACADEMIC LIFE +ADVISORS +All students are assigned an academic advisor when they enroll in a degree +program. Your advisor will be a faculty member in your department. He or she +will help you choose which courses to take each semester. If you need to + +5 + + change your schedule in any way, you must get the approval of your academic +advisor. + +Academic advisors can better serve you when they understand your educational +and career plans. They can help you prepare for graduate school or for a job. +They can arrange work experiences as a part of your degree program if this will +help you. If you give your advisor a clear picture of what you plan to do, he or +she can guide you toward the most suitable courses and university activities. + +INDIVIDUALITY +Individuality is very important to Americans. This value is as important in the +classroom as it is in other parts of U.S. culture. Independent thinking is +expected of students in American classrooms. Interpretation, analysis, critical +thinking, and even challenging the professor may be expected of you in class. +You may be graded on your class participation. Usually, professors will tell you +at the beginning of the course if you are to be evaluated on class participation. +It takes time and practice to get used to speaking out in class, but it is often an +important part of the learning experience. + +REGULATIONS +Certain acts are considered to be academic dishonesty and are not permitted at +Albany State University. Submitting another person’s work as your own or doing +academic work for another person is not allowed. Academic dishonesty includes +copying exam papers, assignments, or term papers or allowing another person to +copy your work. If you use ideas or information published by other people in +papers that you turn in, you must give the writers credit by providing citations in +footnotes or references. Your professors will tell you how to do this. Students +who turn in work which is not their own will be disciplined by the professor and +possibly the university. It could lead to expulsion from the university. + +GLOSSARY OF “UNIVERSITY LANGUAGE” +CATALOG/BULLETIN +The university publishes an undergraduate and graduate catalog. These catalogs +are sometimes called bulletins. They include information on all aspects of +university life, including regulations governing academic life, course descriptions, +and a list of faculty members. You must follow the rules set forth in the catalog +for your academic program. It is an important resource. Read it carefully and +refer to it when you need information about academic offerings and +requirements. + +CLASS +Sessions or meetings of particular courses. These are listed in the Schedule +Report, issued each semester, which lists the days, times and venues the class +meets. + +COLLEGE +The University is divided into units known as “Colleges.” The Colleges at ASU +are the Colleges of Arts and Humanities, College of Business, College of 6 -PI K E PL ACE M AR K E T & WATE R FRO NT + Education, and College of Science and Health Professions. Unlike many other +countries, the term college in U.S. usage does not refer to high school or +secondary education. It is often used interchangeably with the term university. + +COMPREHENSIVE EXAMINATION +Examinations are required by most departments before a student can receive a +Master’s degree. The exams test the student’s knowledge of his or her entire +field of study. + +CORE CURRICULUM +The Core Curriculum is composed of a wide spectrum of general courses. It is +designed to give all students a broad base of basic knowledge and is required of +all students seeking bachelor’s degrees at ASU. + +COURSE +Subject Offerings. These are listed in the course catalog, by department. +Students usually take 4-5 courses per semester. The course listings are also +available through BANNER on the ASU website. + +CREDIT HOUR +Courses are given credit hour values. Most courses are 3 credit hours. This +means that the course meets for approximately 3 hours each week during the +semester. 1-hour course meets for 1 hour, etc. In order to be considered full- +time, undergraduates must enroll in 12 credit hours, graduate students need to +take 9 credit hours each semester. + +CURRICULUM +This refers to the program that you are following, and its requirements. There +are required courses and elective courses in most curricula. + +DEPARTMENT +Within each College are several Departments that specialize in a certain subject. +For example, the College of Education includes the Department of Teacher +Education and Department of Counseling and Educational Leadership. + +DROP-DATE +Students may choose to drop or withdraw from a course without penalty up to +the middle of the semester. However, you must remember that: + +1. As a foreign student, you must remain a full-time student at all times. +You must be enrolled for 12 hours at all times, unless it is a vacation +period which is the summer semester or you otherwise have +permission from the DSO in the International Office. Graduate +students must take 9 hours. + +2. Withdrawal hour’s associated with classes that are dropped due to +documented extenuating circumstances will be included in the 12 +hour total. -One of Seattle’s best-loved attractions, Pike Place Market is a labyrinthine fish and -produce market that oozes theatricality. Find yourself entertained amid ‘flying fish,’ -buskers, tourists and locals before stepping out along the Waterfront, where scenic -views of Puget Sound and the Olympic Mountains await. +7 -LAWRENCE WORCESTER + 3. Students will receive a “WF” (This symbol indicates withdrawal with penalty) for course + +withdrawals identified after the 16 hour limit has been reached. +There are financial and G.P.A conditions associated with the “WF”. + +4. “W” (This symbol indicates that a student was permitted to withdraw without penalty) grades will be awarded to +students who withdraw from classes prior to the end of the schedule +change deadline and to those students who are authorized to +withdraw due to extenuating circumstances. + +After the midterm drop-date, special permission is required for course +withdrawal. The grade the student has earned up to the point of withdrawal will +be reflected on his or her transcript; however, it will not affect the GPA. In +addition, as a F1-Visa Student you are required to notify the International +student Advisor of any schedule changes you are making. + +ELECTIVE COURSE +These are optional courses. Each student must choose several of these courses. +The academic advisor will suggest specific electives. + +FINAL +This is the last, major examination of the semester. Usually, one week is set +aside at the end of the semester for final exams. A final is a long exam, which +sometimes lasts as long as two hours. Not all professors choose to give final +exams. Some will assign research papers or special projects instead. + +GRADE POINT AVERAGE +Your performance is evaluated with grades of A, B, C, D, or F. The grades A +through F are given a numerical value (A=4, F=0) and weighted by the number +of credit hours associated with each grade. Your grade point average is +computed by averaging the numerical values of the credit hours for all of the +courses you have completed. “GPA” is the common abbreviation for Grade Point +Average. + +GRADUATE STUDENTS +These are students who have already received a Bachelor’s degree or the +equivalent. The graduate degrees offered at ASU are the Specialist degree in +Education and Master’s degree in several disciplines. +MID-TERM +A Mid-Term is a major examination given in many classes in the middle of the +semester. This word also refers to the date signaling the end of the first half of +the semester. + +MINOR +A secondary, optional program of study in addition to the student’s major. + +SEMESTER +The school year at Albany State University is divided into two semesters. Fall +semester, which begins in August and spring semester, which begins in January, - Pike Place Market -Watch the world go by at this loud, lovable market (p62) - Seattle Aquarium -Spy the underwater world of Puget Sound (p66) - Waterfront -Catch the flavour of a major seaport (p65) +8 -JOHN ELK III + constitute the normal academic year. There is a full six-week summer session. +Summer terms are optional terms for foreign students. + +QUIZ +An informal examination, shorter in length than a regular examination. + +REQUIRED COURSE +These are courses which each student seeking a particular degree must take. + +RESEARCH PAPER +This is a major paper which can be from 5 to 25 pages in length. In order to +write this paper, a student must select an approved topic, somewhat narrow in +scope, and do research in the library. Papers are to be typed. There are +manuals in the bookstore to show you how to type it, or you can hire a +professional typist. + +SECTION +Some courses are offered in a number of sections, or classes, each semester, so +that a large number of students may attend. A course number, section letter, +and call number identify each section. + +SYLLABUS +A list of information about a course including the books you will use, the +professor’s name, office and phone number. Usually there is a general outline of +how the course will proceed, and sometimes assignments and dates of +examinations are included. + +TRANSCRIPT +A record of students’ courses and grades. An “official transcript” is sealed by the +issuing institution, and normally sent directly to the school or office requesting it. +It is used to document a student’s academic performance while attending +college. Universities, graduate schools, and some employers use this to +determine the desirability of accepting or hiring students. + +UNDERGRADUATE +These are students who are studying for their first degree at the university. +During the first year, students are called freshmen; the second year, +sophomores; the third year, juniors; and their last year, seniors. The +undergraduate degree is the Bachelor’s degree. - +9 -LAWRENCE WORCESTER + AMERICAN CUSTOMS AND CULTURE +The culture and customs of the United States are difficult to define precisely, +because geographic areas and the people who live in them differ. Most of what +you will learn about the customs and culture of the people of the United States +will be by experience and through contact with people. + +While you are in the U.S., it is good to learn the customs of the people here. It +is important, however, that you maintain your own values and beliefs. If you’re +personal preferences or religious beliefs influence you not to do certain things, +do not be embarrassed to explain this. + +TIME +Being on time is very important in the American society. Classes, parties, +concerts, meetings, and other planned events all start at a specific time. If you +need to see someone, you should ALWAYS make an appointment beforehand. +Do not show up at their office or home unexpectedly. It is considered very +important not to be late to scheduled appointments or events. If you are late for +an appointment, it is important to apologize to the person(s) you have kept +waiting. Students who are always late for appointments or who turn in +assignments after the deadline earn negative reputations. Many professors will +deduct from a grade if an assignment is late or if a student arrives late for class. + +INFORMALITY +There is a great deal of informality between all types of people in the U.S. It is +our way of life. The American society does not have a formal class structure. +We may dress more casually and act more informally in social situations than you +may be used to. + +GREETING, INVITATIONS, AND VISITING +The most common forms of greeting are a handshake, smile and “Hello, I’m +pleased to meet you.” The Americans do not usually embrace in public, except +with members of their family or very close friends. + +Professors should be addressed as Doctor or Professor unless they request +otherwise. You should also use titles (usually Mr., Mrs., or Ms.) with older +persons unless they ask you to call them by their first name. For other students, +it is customary that you address them by their first name. + +If you are invited somewhere and you accept the invitation, it is VERY +IMPORTANT that you appear at the event. Do not accept an invitation simply +because you are afraid to say no. If you cannot or do not want to go to the +event, politely tell the person who invited you htat you cannot or choose not to +attend. It is much more inappropriate to accept an invitation and then fail to +appear than it is to decline the invitation in the first place. If invited to visit +someone’s home, you are not normally expected to bring a gift. It is always +acceptable, but not necessary, to bring a small token or an inexpensive gift to +your host or hostess. An expensive gift is usually inappropriate and may make - B E LLTOW N +10 -The alternative scene in Belltown may have mellowed slightly, but this neighborhood -can still be classed as cool. Watch a band at the Crocodile Café before taking an -easy stroll to other offbeat bars in the area. Belltown is also home to the impressive -Olympic Sculpture Park. + your host or hostess uncomfortable or embarrassed. American homes, usually +does not have domestic servants, so if you stay as a houseguest, it is proper for +you to make your own bed or volunteer to help with other household chores. + +TABLE MANNERS +Table manners is most common to Americans including not speaking with food in +your mouth, keeping your elbows off the table, and keeping your hands in you +lap while you’re not eating. A knife, fork, and spoon are used unless “finger +foods” are being served. You should not belch at the table after eating; this is +considered rude in the U.S. After the meal is finished, the napkin should be laid +beside the plate and used silverware should be laid across the top of the plate. + +RELATIONSHIPS +Most Americans have many friendly acquaintances, but just a few close friends. +Visitors from other countries often remark that Americans are very friendly, even +at the first meeting, but that it is difficult to get to know them very well. +Sometimes this is true; although Americans are friendly to everyone, they are +also very private people and often have just a few intimate friends. + +Americans often say things like “let’s get together sometime” as a friendly way to +say good-bye, NOT as an invitation. If you would like to get to know someone +better, it is a good idea for you to invite them to go somewhere with you on a +specific day, at a scheduled time. Participate in as many social activities as you +can and you will meet and make many new friends. + +MEN AND WOMEN +American women are independent, and women and men are considered equal. +It is not appropriate to whistle at, touch, or make personal comments about +women whom you do not know EXTREMELY well. It is common for American +university women to live alone or with other women in apartments off campus. +Guests of both sexes are often invited to these apartments. Do not assume that +your hostess has romantic intentions simply because she has invited you to stop +by her apartment. + +There are no clear guidelines for dating Americans. If you want to get to know +someone, it is often best just to ask the person to join you for a snack or to get +together to study. It is important to remember, however, that if two people go +out on a date, it only means that they enjoy each other’s company. It does not +always indicate serious intentions or constitute a promise! If you begin to date +someone from another culture, it is important to remember that s/he simply may +not view the relationship in the same way that you do, especially regarding the +seriousness of the relationship or the desire to engage in sexual activity. - +11 -RICHARD CUMMINS + It is most often the man who invites the woman out on a date, and customarily +the man pays all of the expenses unless the persons agree that it will be a +“Dutch Treat,” that is, each person pays for his or her own expenses. Women +sometimes ask men for dates. A woman who asks a man out should be +prepared to pay for the expenses. + +BATHING AND HYGIENE +Most Americans bathe or shower daily, and use deodorants and anti- perspirants. +Americans are not accustomed to body odors and find it offensive. Some +international visitors may want to adjust their bathing habits and accommodate +American preferences for the time they are here. + +CULTURE SHOCK +There are many differences between the American culture and the home culture +of many international students. Adapting to these differences can cause “culture +shock.” If you are having trouble adjusting to life in the U.S., or are +experiencing depression or physical or emotional problems, you may be +experiencing culture shock. The Office of the International Student Advisor can +provide cross-cultural counseling to help you, or refer you to a counselor on +campus free of charge. Counseling is always confidential. + +ADDITIONAL INFORMATION +If you would like additional information about American culture, please visit the +Office of the International Student Advisor. The office contains several +handbooks and guidebooks for foreign visitors in the United States. + +IMMIGRATION STATUS +THE U.S. CITIZENSHIP AND IMMIGRATION SERVICES (USCIS) is the branch of +the Federal government created to monitor the entry and exit of foreign +nationals. This branch of the government was formerly known as the +Immigration and Naturalization Service (INS). You will encounter INS officials as +you travel between the U.S. and your home country and when you apply for +special programs. The following information should help you understand the +regulations concerning student immigration status. Most ASU students have +either F-1 or J-1 immigration status. + +PASSPORT +Your passport must be valid at all times. When it is time to renew your passport, +contact your country’s embassy in Washington, D.C. to determine what +procedures you need to follow in order to extend your passport. + +VISA +There is a visa stamped in your passport, which permitted you to enter the +United States. A single entry visa permits you to enter the U.S. only once on +that specific visa. A multiple entry visa allows you to enter the U.S. as many +times as you like (after leaving on vacations or trips home) as long as you - Olympic Sculpture Park -Feel small next to Richard Serra’s 2004 -sculpture, Wake (p67) - Ace Hotel -Be fascinated by bedtime reading at this hipster -hotel (p179) - Shorty’s -Play pinball and drink beer (p142) +12 - + remain a full-time student. Student stays are valid for the “duration of status”. +This status is indicated by the mark “D/S’ in your passport and on your I-20/DS- +2019 and I-94. As long as you remain in full-time student status, you are in the +country legally, regardless of the expiration date on your entry visa. +If you leave the U.S. after your visa expires, you must go to a U.S. Embassy or +Consulate-General in your home country and obtain a new visa before you can +return to the U.S. It is difficult to obtain a new visa in a “third country.” It is +difficult to obtain new F-1 visas in Canada or Mexico, unless you are from those +countries. + +F-1 IMMIGRATION STATUS +If you are a full-time degree-seeking student or a Rotary student at ASU, you +have entered the U.S. with F-1 immigration status. This type of immigration +status is designed to allow you to obtain your degree and return home following +your graduation. F-1 status is a “non-immigrant” status. “Non-immigrant” +means that you are not moving to the U.S. permanently. This immigration status +is valid for “duration of status.” + +• You should carefully consider the dates of your authorized stay and +make sure you are following the procedures under U.S. immigration +laws. It is important that you depart the U.S. on or before the last +day you are authorized to be in the U.S. on any given trip, based on +the specified end date on your Arrival-Departure Record, Form I-94. +Failure to depart the U.S. will cause you to be out-of- +status. Additional information on successfully maintaining your +immigration status while a student or exchange visitor can be found +on the Immigration and Customs Enforcement (ICE) website. +• Staying beyond the period of time authorized by the Department of + +Homeland Security (DHS) and being out-of-status in the +United States is a violation of U.S. immigration laws, and may cause +you to be ineligible for a visa in the future for return travels to the +U.S. + +• Staying unlawfully in the United States beyond the date Customs and +Border Protection (CBP) officials have authorized--even by one day-- +results in your visa being automatically voided, in accordance with +INA 222(g). Under this provision of immigration law, if you overstay +on your nonimmigrant authorized stay in the U.S., your visa will be +automatically voided. In this situation, you are required to reapply for +a new nonimmigrant visa, generally in your country of nationality. + +• For non immigrants in the U.S. who have an Arrival-Departure + +Record, Form I-94 with the CBP admitting officer endorsement of +Duration of Status or D/S, but who are no longer performing the +same function in the U.S. that they were originally admitted +to perform (e.g. you are no longer working for the same employer or +you are no longer attending the same school), a DHS or an +immigration judge makes a finding of status violation, resulting in the +termination of the period of authorized stay. + +13 + + Staying Beyond Your Authorized Stay in the U.S. and Being Out of Status + +• You should carefully consider the dates of your authorized stay and make +sure you are following the procedures under U.S. immigration laws. It is +important that you depart the U.S. on or before the last day you are +authorized to be in the U.S. on any given trip, based on the specified end +date on your Arrival-Departure Record, Form I-94. Failure to depart the +U.S. will cause you to be out-of-status. Additional information on +successfully maintaining your immigration status while a student or +exchange visitor can be found on the Immigration and Customs +Enforcement (ICE) website. + +• Staying beyond the period of time authorized by the Department of + +Homeland Security (DHS) and being out-of-status in the United States is a +violation of U.S. immigration laws, and may cause you to be ineligible for +a visa in the future for return travel to the U.S. Select Classes of Aliens +Ineligible to Receive Visas to learn more. + +• Staying unlawfully in the United States beyond the date Customs and +Border Protection (CBP) officials have authorized--even by one day-- +results in your visa being automatically voided, in accordance with INA +222(g). Under this provision of immigration law, if you overstay on your +nonimmigrant authorized stay in the U.S., your visa will be automatically +voided. In this situation, you are required to reapply for a new +nonimmigrant visa, generally in your country of nationality. + +• For non immigrants in the U.S. who have an Arrival-Departure Record, +Form I-94 with the CBP admitting officer endorsement of Duration of +Status or D/S, but who are no longer performing the same function in the +U.S. that they were originally admitted to perform (e.g. you are no longer +working for the same employer or you are no longer attending the same +school), a DHS or an immigration judge makes a finding of status +violation, resulting in the termination of the period of authorized stay. + +What Items Do Returning Students Need? + +All applicants applying for renewals must submit: + +• A passport valid for at least six months; +• An application Form DS-156, together with a Form DS-158. Both forms +must be completed and signed. Some applicants will also be required +to complete and sign Form DS-157. Blank forms are available without +charge at all U.S. consular offices and on the Visa Services website +under Visa Applications Forms at the United States Citizenship and +Immigration Services (https://www.uscis.gov/). + +• A receipt for visa processing fee. A receipt showing payment of the visa + +application fee for each applicant, including each child listed in a +parent’s passport who is also applying for a U.S. visa, is needed; +• A new I-20 or an I-20 that has been endorsed on the back by a school + +official within the past 12 months. + +14 + + All applicants applying for renewals should be prepared to submit: + +• A certified copy of your grades from the school in which you are + +enrolled; + +• Financial documents from you or your sponsor, showing your ability to + +cover the cost of your schooling. + +Students Away from Classes More Than Five Months +Students in or outside the U.S., who have been away from classes for more +than five months, will likely need a new visa to enter the U.S. + +How long may I stay on my F-1 student visa? + +When you enter the United States on a student visa, you will usually be +admitted for the duration of your student status. That means you may stay as +long as you are a full-time student, even if the F-1 visa in your passport +expires while you are in America. For a student who has completed the course +of studies shown on the I-20, and any authorized practical training, the +student is allowed the following additional time in the U.S. before departure: + +• F-1 student - An additional 60 days, to prepare for departure from the + +U.S. or to transfer to another school. + +• M-1 student - An additional 30 days to depart the U.S. (Fixed time period, +in total not to exceed one year). The 30 days to prepare for departure is +permitted as long as the student maintained a full course of study and +maintained status. An M student may receive extensions up to three years +for the total program. + +As an example regarding duration of status, if you have a visa that is valid for +five years that will expire on January 1, 2001, and you are admitted into the +U.S. for the duration of your studies (often abbreviated in your passport or on +your I-94 card as "D/S"), you may stay in the U.S. as long as you are a full- +time student. Even if January 1, 2001 passes and your visa expires while in +America, you will still be in legal student status. However, if you depart the +U.S. with an expired visa, you will need to obtain a new one before being able +to return to America and resume your studies. A student visa cannot be +renewed or re-issued in the United States; it must be done at an Embassy or +Consulate abroad. + +FORM I-20 +Before you came to ASU, you were sent Form I-20 by the ASU Admissions Office. +This form is the one you use to obtain your F-1 student visa, and it is a very +important form. You should keep it safe with your passport. A Designated School +Official (DSO) must sign your I-20 whenever you travel outside the United +States, even if you are only going for a short time. If you change degree +programs (change your major), you will need a new I-20. New I-20 forms are +available from the International Admissions Officer on campus at 229-430-4646 +or visit the Office of the International Student Advisor for further assistance. If +you transferred to ASU, be sure to keep your ASU I-20 and the I-20 from your -LAWRENCE WORCESTER +15 - + previous school together. Note the date of expiration in #5 on your I-20. If you +need to stay past this time to complete your studies, you must contact a DSO at +least 30 days before this expiration date to extend your stay as a student. + +FULL-TIME STUDENT STATUS (In-Status) +In order to maintain your full time status, you must attend school at least 9 +semester hours per semester for graduate studies and at least 12 semester +hours per semester for undergraduate studies. If you have any questions +about this regulation, please see the International Student Advisor. Students +who fail to maintain this requirement risk being placed on probation and could +lose university scholarships, including International Student Scholarships. To +avoid losing your full-time status, check with your advisor and/or Registrar +before you drop a class. + +VACATION +The Summer Term is considered a vacation period for all non-immigrant +students, as long as the student intends to register full-time for the next term. +You may enroll part-time, full-time or not at all during the summer term. Make +sure you check with the International Student Advisor before taking ANY +semester other than the summer semester off. You can go home for summer +any time you wish. + +VACATIONS INSIDE OR OUTSIDE THE USA + +(A) Inside the USA: You do not need permission of any kind from anyone to +travel anywhere within the USA. Travel between the 50 states is totally +unrestricted. Have fun—see the USA! + +(B) Outside the USA: Before leaving the USA, you must go by the office of a +Designated School Official (DSO) with your passport and I-20 and let the +DSO sign your I-20 and check your visa. Your I-20 must be signed for +you to re-enter the USA. You must have both the signed (or a new) I-20 +and a valid F-1 visa in your passport to re-enter the USA. + +If your visa has expired, you will need to go to a U.S. Embassy or Consulate- +General to get a new visa before you can re-enter the country. U.S. VISAS ARE +NOT ISSUED IN THE USA. If you have dependents that are on an I-20, we need +to check their passports and visas as well. + +PLEASE VISIT THE OFFICE OF THE INTERNATIONAL STUDENT +ADVISOR AT LEAST TWO (2) WEEKS BEFORE YOUR TRIP SO THAT +THERE WILL BE SOMEONE AVAILABLE TO SIGN YOUR DOCUMENTS. + +TRANSFER STUDENTS +If you are transferring from another U.S. college or university to ASU, you must +see the International Student Advisor within 5 days of your registration at the +beginning of the semester for completion of the transfer process, which must be +done to maintain your student status. + +EMPLOYMENT -LAWRENCE WORCESTER +16 -7 + (A) ON-CAMPUS: As soon as you have registered for classes, you may work +on campus as a student assistant in the various departments if hired. . +Each department or facility has its own budget for student workers, so +you must go directly to them to seek employment. Once you have found +a job, you must register with the ASU Human Resources Department +(Department of Human Resources) in order to be paid for your work. To +register, you must have a Social Security Number. For more information +about Social Security Numbers, please see the section on Social Security +Numbers in this handbook. + +(B) OFF CAMPUS—ECONOMIC HARDSHIP: You must see the International + +Student Advisor before you can work off campus. Due to recent changes +in immigration laws, it is virtually impossible for foreign students to work +off campus. The only two exceptions are internships for academic credit +or extreme economic difficulty due to conditions beyond his or her +control. Students must apply to USCIS for permission to accept +employment off campus due to economic hardship, and the ISA can +assist you with this application. + +(C) OFF CAMPUS—PRACTICAL TRAINING: You must see the International +Student Advisor before you can work off campus. Curricular Practical +training (or an internship) is work directly related to your course of study +that will give you “hands-on” job experience that will help you get a +better job when you return to your home country. It can be granted only +after you have been at ASU for an academic year, and then only under +certain specific conditions. There are provisions for curricular practical +training before graduation and after graduation. + +Other than the above types of work, NO OTHER EMPLOYMENT IS PERMITTED +while you are in student status. You may not work off campus without +permission from USCIS and/or the ISA. If you do work off campus illegally, you +will lose your student immigration status and its benefits. Working illegally in the +United States is the most serious violation of non-immigrant status in the eyes of +the USCIS, and you WILL NOT be reinstated if you violate this regulation. + +INFORMATION UPDATES +Any change in address, phone number or marital status must be reported to the +Office of the International Student Advisor (ISA) and the Office of the Registrar +within 10 days. You must report the changes to both offices. If you do not +notify the ISA and the Registrar, you will not receive important mail from the +university. Please note that the Office of the Registrar maintains two addresses +on each student, a local address and a home address. The local address is your +address in Albany. If you live on campus, the university will automatically record +your ASU box number. If you live off campus, you must report your address to +the ISA and the Registrar. The home address is your address in your country. It +is the address you listed on your application for admission. All official +correspondence concerning your grades and your fees will be sent to the home +address. You may want to change your home address to your Albany address to +make sure you receive these notices. If you move off-campus, you must also -  +17 - + complete a United States Postal Service Change of Address card and sign the +card. Completing this card will allow the campus mailroom to forward mail to +you at your new address. + +THE CARDINAL RULE +If you have questions about working or any other USCIS-related +matter while in the USA, visit the Office of the International Student +Advisor. The Office of the International Student Advisor and +Designated School Officials are here to assist you. We will help you as +much as we can; however, it is YOUR RESPONSIBILITY to keep the +office informed of your status and to ask for help when you need it. The +U.S. Immigration and Naturalization Services holds you accountable for +your own actions and expects you to follow all laws pertaining to your +immigration status. If you choose to violate the law or fail to educate +yourself about the law, you will be responsible, not the Office of the +International Student Advisor. Do not allow yourself to fall out of +status. Do not rely on the advice of other students for immigration +matters. You could lose the privilege of studying in the U.S. now, and +you could be prevented from returning to the U.S. in the future. + +A revision of the immigration law, the Illegal Immigration Reform and +Immigrant Responsibility Act of 1996, contains very harsh penalties of +students who violate student immigrant status, including provision to +prevent offenders from returning to the United States for periods of +several years. The new immigration regulations based on this act are +currently in place, but because the law is a new one, there are often +modifications or updates to the regulations. The International Student +Advisor will keep you informed about modifications in the regulations. + +AUTOMOBILES +REQUIRED INSURANCE +Law requires liability insurance for a car. If you own an automobile in Georgia, +you must purchase liability insurance. Collision insurance is not required, but it is +recommended. Proof of insurance, usually in the form of an insurance card, +must be carried in the car at all times. Look in the “Yellow Pages” of the phone +book to locate insurance agents, or ask a friend to refer you. Compare prices +with several companies before buying. + +DRIVER’S LICENSE +As a non-resident student at the university, you may drive in Georgia with a valid +international driver’s license. If you do not have an international license or if +your license expires during your stay, you may apply for a state of Georgia +license at the `Dougherty County Department of Motor Vehicle Safety (DMVS), +2062 Newton Road, Albany. The contact telephone number is 229-430-4258. +Getting a Georgia license requires a written test and a driving test. You should +present your passport, I-94, and your I-20 or DS-2019 to the examiner. You will +also need to bring a lease or utility bill, which verifies your address. If you live in - +18 -8 + a residence hall, you may also need a letter from the International Student +Advisor certifying that you live on-campus. The fee for a five-year license for +driving a passenger car is $20.00. A provisional permit costs $10.00. All fees +must be paid in cash. In addition, you will need a denial letter from the Social +Security Office and/or a Social Security Card. Also, you will need a letter from +your International Student Advisor before you go. + +Many students find that having a State of Georgia license helps them to obtain +lower rates for automobile insurance. Georgia identification cards are also +available at the driver’s license office for a fee. Both the license and the ID card +contain photographs of the bearer. Because the driver’s license and the state ID +card are the most widely accepted forms of identification, it is easier to verify +your address and age with these items than with your passport or ASU ID. +Having proper identification is especially important for writing checks and +handling other banking matters. You should consider obtaining an official form +of identification other than your passport. If you must carry your passport with +you at all times, you risk losing it. You should call the license bureau at 229-430- +4258 to make sure that the test is being given on the day you plan to apply for a +license. The bureau is closed on Mondays and will not give tests when it is +raining. + +If you hold a license from another state in the U.S. or from another country, you +will need to surrender this license to obtain a Georgia license. Please note that +once an out-of-state license has been surrendered to the State of Georgia, it is +not possible to reclaim the license. You should consider whether you wish to +surrender your home country license before you go to the bureau. If you have +questions about this regulation, please see the International Student Advisor. + +RULES OF THE ROAD +Although your international license is valid in the U.S., it is a good idea to read +the State of Georgia Driver’s Manual before you begin driving. Every country has +different traffic laws. Reading these laws in advance could prevent you from +having an accident. Copies of the State of Georgia Driver’s Manual are available +in the Office of the International Student Advisor. + +TRAFFIC VIOLATIONS +If you break any of the traffic laws, including driving over the speed limit, you +risk being stopped by a police officer. If a police or sheriff car follows you with +flashing lights and sirens, the officer wants you to pull over to the side of the +road. You should be prepared to pull over in a safe spot and do as the officer +instructs. Usually, the officer will ask to see your driver’s license, proof of liability +insurance, and registration for the car. He will also ask you questions about your +driving. If you have broken a traffic law, he may issue you a citation (ticket). +Citations involve paying fines, and in extreme cases, imprisonments. If you are +in an accident, you must wait at the scene of the accident for a police officer to +investigate the accident. Do not move the car and do not leave until the officer +has interviewed you about the accident. + +PARKING -S E AT TLE CE NTE R & Q U E E N AN N E +19 -Seattle Center is a 1960s vision of the future, making it an endearingly retro place to -visit. As well as the Space Needle, it contains numerous museums and attractions, -including the Experience Music Project. Rising up on the hill above Seattle Center is -the dignified Queen Anne neighborhood. + Most parking in Albany is free. Parking on the ASU campus requires a permit, +which costs $15 per academic year. Please refer to the ASU Parking and +Transportation resource guide for parking regulations. If parking regulations are +violated, you will get a “ticket” and be required to pay a fine. Parking Rules on +campus are clearly stated in the RAM Student Handbook. + +RENTAL AND PURCHASE +If you have a valid driver’s license and are at least 21 years old, you will probably +be able to rent or lease a car. Some rental places also require a credit card in +your name or that you are at least 25 years of age. Check with the larger local +car dealerships about short- and long-term rental. + +If you wish to buy a car for your stay in the U.S., be CAREFUL. Most companies +will service your car if it needs repair after purchase. You must pay for all +repairs unless you have a “manufacturer’s warranty” or paper which states that +the maker of the car will pay for repairs for a limited period of time after +purchase. If you buy a used car from a private dealer or individual, check the +vehicle carefully before purchase. It might help to bring a friend who knows +about the mechanical working of an automobile to inspect the car for you. An +honest car dealer will be happy to let you inspect all parts of the car and +remember to ask questions before you buy. When purchasing a car, you must +have a TITLE and LICENCE PLATE (tag) from the State of Georgia. The dealer +where you purchased the car will give you the correct papers to fill out and help +you obtain these things. In addition, you must renew items such as insurance, +license and tags on a yearly basis. If you buy a car from an individual, be sure to +ask for assistance with these items. If you are bringing a car to Albany from +another state, you must transfer the title for the car and obtain license plates +from the State of Georgia. You should also notify your insurance company that +you have moved. + +License plates must be renewed each year before paying the tax on your car. +Once you have paid the tax, you will be given a sticker for that year. The sticker +should be placed in the upper right hand corner of the license plate. If you have +questions about license plates, please contact the Dougherty County Tag Office, +225 Pine Avenue, Room 104, Albany, GA. The telephone number is 229-431- +3255. + +EMERGENCIES +To call the POLICE, AMBULANCE, or FIRE DEPARTMENT, dial 911. Call this +number only if someone needs immediate help. This is a special emergency aid +number and needs no other numbers. For medical problems that are not life- +threatening emergencies, call the Palmyra Medical Centers, 2000 Palmyra Road, +Albany. Tel.:229- 229-434-2000 or Phoebe Putney Memorial Hospital, 417 3rd +Avenue, Albany. Tel.: 229-883-1800, 229-312-1000. Police concerns are handled +according to location: + +Campus Police: +Albany Police: +Leesburg Police: + +229-430-4711 +229-431-3266 or 229-431-2100 +229-759-6012 -RICHARD CUMMINS +20 - International -Fountain -Get wet in sci-fi International -Fountain (p73) - Queen Anne -Pose by perfectly manicured lawns -(p74) - Experience Music -Project -Be wowed by Frank Gehry’s 2000 -museum (p72) - Space Needle -Zoom to the top for incredible -views (p71) + Albany Fire Dept.: + +229-431-3266 + +ENGLISH LANGUAGE +Students who scored below 523 on the TOEFL (Test of English as a Foreign +Language) are REQUIRED to take special supplementary English as a Second +Language (ESL) courses. These courses are non-credit-bearing, but they will help +you with your academic English needs. The course numbers are ENGL 0075, +0077, 0098 and 0099. Any international students are welcome to take these +courses, even if the TOEFL score was above 523. + +English 1101 is the introductory university level English course required for all +ASU degree programs. The course prepares students for the remaining required +English courses in their degree program. + +There are some options available for students who wish to improve their English +language skills. For students who need help with their writing skills, the English +Department offers tutoring in the Writing Center lab located in Simmons Hall. +Another option for tutoring is the PEER TUTORING program. Tutoring is available +in most subjects free of charge during the fall and spring semesters. Contact the +Office of the International Student Advisor or your English professor to find out +which of these facilities will best meet your language needs. + +ENTERTAINMENT +ASU provides many entertainment and social opportunities for students. The +student centers on both campuses are available to you for all activities. It has as +its objectives the fulfillment of the cultural, educational, and recreational desires +of all students of ASU. Various programs and activities are planned and +supervised by the Director of Student Activities and the program coordinator to +round out and broaden the experiences of each student. The services and +facilities of the Student Union are open from 10.00 a.m. to 7.00 p.m. Monday +through Friday. On Saturdays and Sundays, the Game Room hours will be posted +in the game room, unless other wise notified. The facilities and services of the +Union are for students to make full use of. They include conference and meeting +rooms, lounges, a game room, Pizza Hut, the ASU Bookstore, Student +Government Office, a Post Office, an ATM, an information desk, vending +machines, and the ASU Board of Managers. + +The Albany State University International Student Association (ASUISA) is a +special club for international students at ASU. This group plans and promotes +cultural diversity and global awareness activities involving ASU and the +community. It also organizes programs and field trips for students who are +interested in more international activities. Look for information about the first +meeting in late August. + +For more information about clubs and activities, you should read the campus +paper, The Student Voice, and listen to the community radio stations. Also plan +to attend the campus-wide activities fair that takes place during the early part of + +21 + + the Fall Semester. Each of ASU’s clubs will have a booth at this fair to provide +more information for potential members. + +HEALTH CARE +The United States philosophy of health care is different from that of many other +countries. Health care in this country is viewed as an industry. All Health Care +professionals are expected to be paid by their patients or clients. Patients often +must prove ability to pay before treatment will be given. Doctors in this country +usually manage their offices so that they spend a minimum amount of time with +each patient. Nurses do as much as possible to help free the doctor to see more +patients. Expect to deal with nurses in the office and over the telephone. + +MEDICAL EMERGENCY +In case of medical emergency involving a life-threatening situation, or to request +an ambulance (emergency transportation to the hospital), call 911. Call this +number only in true emergencies. ASU students should not use the emergency +room services at the local hospitals, Phoebe or Palmyra, unless the medical +problem is life threatening. Students should use the university health center +infirmary as their main source of medical care. If you do not know whether the +situation is a true emergency, it is probably best to call campus police and have +the officer escort you to the health center. The nurse will evaluate the problem +and will send emergency cases to the hospital. THE HEALTH CENTER IS THE +CLOSEST MEDICAL FACITLITY TO THE RESIDENCE HALLS. Please note that +using the ambulance and the emergency room at the hospitals is very expensive. +This expense will not be covered by medical insurance if the situation is not a +true emergency. + +STUDENT HEALTH SERVICES +The University Health Service of Albany State University is located in the +Academic Administration (ACAD) Building, Room 138. Student Health Services +provide opportunities for safeguarding the health of all students at ASU. The +staff provides the following through student health fees: + +1. General medical care. +2. Confinement in the Health Services. Overnight stays are not available. +3. Consultation on general health needs. + +The health fee DOES NOT cover expenses for special medicines and treatments, +visits to doctors’ offices, operations or off-campus hospitalization. + +The University Health Services office is open Monday through Friday, 8:00 a.m. – +5:00 p.m. Our Doctors’ hours are Monday through Friday, 4:00-5:00 p.m. +Emergency calls are possible at any time. Each student, freshman or transferee, +is required to have on file a medical record completed and signed by a physician. +These records will be good for four years provided the student is in continuous +attendance at Albany State University and has consulted with the University +Physician periodically. + +22 + + DENTAL CARE +If you need to have your teeth cleaned or another dental service, you should +look in the telephone book Yellow Pages under “Dentists.” Phone to make an +appointment before you go. + +EYEGLASSES AND CONTACT LENSES +If you need an eye examination, or you want your eyeglasses or contact lenses +changed, you should look in the telephone book Yellow Pages under, “Opticians +and Optometrists.” Phone to make an appointment before you go. For medical +concerns with your eyes, look under “Ophthalmologist.” + +COUNSELING +Sometimes individuals become depressed, lonely or fearful in a strange +environment. Many overseas students come from a culture very different from +that of the U.S., and they often miss the support of their family and friends at +home. If you are feeling emotionally upset, depressed, or just need someone to +talk to, contact the International Student Advisor. S/he can make helpful +suggestions, listen to your concerns, and may refer you for further counseling. + +The primary function of the Counseling Services is to assist students in their total +development by providing services and programs to facilitate their intellectual, +emotional, social and academic growth. Counseling services also aid students in +determining a purpose and direction for their lives and in developing coping skills +and behaviors to aid them in working more efficiently with others in their +environment. + +PAYMENT +Health care in the United States is not free. Patients must pay for the services of +doctors, dentists, laboratory workers and hospitals. Their fees are often very +expensive. Care at the Health Services on campus is paid for by student health +fees each semester, so go to the Health Center first for any illness. If you have +an appointment with any off campus doctor or dentist, be prepared to +pay the bill at the conclusion of the appointment. Insurance will not +always cover 100% of the cost. Medical insurance does not cover the expense +of dental visits or eye exams and glasses. Students may want to plan to take +care of routine visits to these doctors during vacations at home. + +INSURANCE +Due to the high cost of medical care in the U.S., all foreign students +must purchase health insurance before enrolling for classes at ASU. +ASU is pleased to offer one policy for all ASU foreign students student. +Since September 1994, new regulations from the U.S. government +have determined what type of insurance coverage students must have. +The ASU policy has been designed to meet this minimum level of +coverage at a reasonable price. The plan covers portions of hospital +costs, doctor visits, and medication but not dental visits or eyeglasses. +Insurance to cover all costs at 100% and to cover dental visits and +eyeglasses is extremely expensive. Few U.S. residents carry such +coverage due to the high cost. The insurance system in the U.S. is + +23 + + designed for the patient to pay a portion of the cost in addition to the +insurance premium already paid. Insurance premiums are the fees paid +for insurance coverage. The university will bill all ASU international +students for the insurance premiums. Insurance from another source is +generally not accepted, since the coverage must meet U.S government +standards. If you need information about the ASU health insurance +program, see the International Student Advisor. If you do not pay your +premium, you will lose your insurance coverage and you will have a +hold placed on your student account, which will prevent you from +registering for classes. You should be careful to avoid a situation +where you lose your coverage because you never know when an illness +or accident may occur. If you have an illness or an accident and you +have no insurance, you will be responsible for 100% of the cost of your +treatment. For Student Health Insurance, please visit the Student +Health Insurance website. www.studenthealthinsurance.com + +GOOD HEALTH +The demand of university life and adjusting to a new environment can be very +stressful. Whenever a person is under stress, s/he is much more susceptible to +illness. Be sure that you take steps to protect your health. Eat properly, get +plenty of sleep, exercise, and protect yourself from communicable diseases, +especially sexually transmitted diseases such as AIDS. If you think you have a +health problem, seek help immediately. If you have questions about any health +matter, please visit the Student Health Center or the Counseling Center. + +IMMUNIZATIONS +All students in the state of Georgia must be vaccinated against measles, mumps, +and rubella. The vaccination can be obtained in one injection called an MMR. It +is recommended that you also receive a shot for Meningitis as well as a +Tuberculosis test. The vaccines are required because these diseases are +present in the U.S. You must protect your health while you are in the U.S. +Students must receive two doses (injection) of this “MMR” vaccine at least 30 +days apart. These vaccines are available at the Dougherty County Government +Office at 222 Pine Avenue, Albany. The telephone number of the Health +Department is 229-430-6200. Students who do not receive these vaccines will +not be allowed to register for classes. We prefer that you send proof of +immunizations and TB test as part of your application package. + +HOUSING +ON-CAMPUS HOUSING +The residence halls provide convenient, inexpensive housing on campus for +students. Most students will share a room with another student (roommate) and +a bathroom with other students nearby. Students must sign a room contract and +pay an application fee and deposit in order to live in a residence hall. The exact +amount to be paid will be included in your letter of admission. The contract +obligates you to live in the residence hall for the entire nine-month academic +year (2 semesters). The on-campus apartments are open all year. They are, + +24 + + however, in great demand. If you want to live in an on-campus apartment, stop +by the Office of Housing for more information or call 229-430-4741. + +If your room needs repairs or you have questions and concerns about any aspect +of life in the residence hall, you should contact your Resident Assistant. The +Resident Assistant, or RA, is also a student who lives in the hall, but this student +is an employee of the Office of Housing. The RA enforces the rules of the +residence hall such as quiet hours. The RA also handles issues such as changing +rooms or roommates and reporting when repairs are needed. The Residence +Hall Director supervises the Resident Assistant of the residence hall. Your +questions and concerns about life in the residence hall should be addressed to +the RA or the Residence Hall Director. Of course the staff of the Office of +Housing and Residence Life will be happy to assist you as well. The telephone +number is 229-430-4741. + +During recent years, the demand for on-campus housing has been high. It is +very important for you to read carefully any correspondence from the Office of +Housing. You need to follow specific procedures to reserve your room each year. +Please make a note of all deadlines and procedures. + +ROOMMATES +Living with an American roommate can be an excellent way to practice English, +learn about the U.S. culture and customs, or to meet a new friend. Living with +another person requires patience, tolerance, and kindness, especially as you get +to know one another during the first few days. It can also be one of the most +enjoyable aspects of your stay on campus. You should expect to disagree with +your roommate about some things. This is normal. The best way to solve your +differences is to speak honestly to one another. Americans are accustomed to +direct discussion of problems. If there are problems, which cannot be solved +between the two of you, you may go to your Resident Assistant (RA) who will +work with you to assign a new roommate. + +OFF-CAMPUS HOUSING +Off-campus housing is plentiful in Albany. There are many apartment complexes +and rental houses near campus and around town. Listed below are a few +suggestions of places to look for housing: + +1. + +2. + +3. + +4. + +Albany Herald, the local newspaper, lists apartments, houses, and +rooms for rent in the classified advertising section of the paper. The +ad will usually list the monthly rent as well as a phone number to call +for more information. +Bulletin boards and notices on campus: check classroom buildings for +notices and ads about apartments for rent. Many students looking for +roommates advertise in this manner. +The telephone directory: look under “Apartments” in the Yellow +Pages. +The Office of the International Student Advisor has listing of apartment +complexes in Albany, which lists such information as rental cost, size, +and distance from campus. + +25 + + RENT AND DEPOSITS +You will probably have to pay a deposit of two or three hundred dollars, plus +your first month’s rent before you can move into an apartment or rental house. +When you move out of the apartment you will get the deposit back if there has +been no damage to the apartment. You will also be required to sign a lease. A +lease is a legal contract that states the amount of time that you will live in the +apartment. Most leases are for one year. Once a lease is signed, you are legally +bound to live in the apartment, and you must pay the rent even if you decide +that you no longer want to live in the apartment. Do not sign any lease that you +do not understand. Be sure to check the dates of the lease carefully. Please +visit the International Student Advisor before you move off campus. He/she can +offer advice about apartments and assist you with interpreting leases. Please +note that if you lease an apartment in your name but also have roommates living +with you, you must be very careful about changing the name on the lease if you +move. If your name is on the lease, you are responsible for the payment of rent +even if you move out and allow other students to move into the apartment. If +you do not pay the rent, the landlord may take legal action against you. Please +be especially careful in situations where you share telephone and utility services +with other students. Never agree to have a lease or a utility bill placed in your +name unless you are certain no additional money is owed on the account. If you +change a lease or utility bill from another student’s name to yours, you will be +responsible for any money owed on the lease or account even if the charges +were billed before the account was placed in your name. + +DINING SERVICE +The University Dining Hall, which is under the supervision of the Director of Food +Services, provides nutritious and wholesome meals at a minimum cost to the +student. Meals are served in an ultra-modern facility, which includes a food court +consisting of the following concepts: Café Features, Wokery, Grill Works, Itza +Pizza, Deli Corner, Sweet Things, Salad Garden, Veggie and Drinks. The meals +are unlimited (all you can eat). The Dining Hall seating capacity is approximately +525 students. All boarding students must present proper University student +identification. Non-boarding students, faculty, staff and visitors must pay cash at +the cashier’s stand before being served. + +Meal Schedule + +Monday-Thursday +Breakfast +Lunch +Dinner +Friday +Breakfast +Lunch +Dinner + +Saturday & Sunday +Brunch + +7:00 a.m. – 8:45 a.m. +11:30 a.m. – 1:00 p.m. +5:00 p.m. – 7:00 p.m. + +7:00 a.m. – 8:45 a.m. +11:30 a.m. – 1:00 p.m. +4:30 p.m. – 6:30 p.m. + +10:30 a.m. – 12:30 p.m. + +26 + + Dinner + +4:00 p.m. − 5:30 p.m. + +(Schedule is subject to change without notice) + +LEGAL CONCERNS +LAWS +While you are in the U.S., you are subject to all of the laws and regulations, +which govern U.S. citizens. In addition, you are subject to immigration laws, +which define and limit your activities as a non-immigrant. Failure to comply with +immigration laws can lead to a request for your voluntary departure, or a forced +deportation from the U.S. + +EQUAL PROTECTION and DUE PROCESS +All persons in the U.S., regardless of citizenship, are entitled to the same +protection of the law and the same procedures of law enforcement. This means +that, with a few exceptions, you will be treated the same as an American citizen +in your relationship with the law enforcement agencies. Those few exceptions +involve immigration law. Consult the Office of the International Student Advisor +for clarification of this issue. + +ILLEGAL DRUGS +The possession, use, or sale of drugs or narcotics, including marijuana, hashish, +amphetamines, hallucinogens, barbiturates, cocaine, heroin, and a wide range of +other drugs, are serious offenses under Georgia law and immigration law. Those +who choose to use illegal drugs run the risk of expulsion from the university, +imprisonment, and/or deportation from the United States. + +ALCOHOL +The minimum legal age for drinking alcoholic beverages (beer, wine, liquor, etc.) +in the State of Georgia is 21. You cannot be served alcoholic beverages unless +you are 21 or older. You will be asked to present an identification card with your +photograph and your date of birth before you can purchase alcohol from a store +or enter a bar, nightclub, or other establishment that serves alcohol. Students +who are over age 18 but under 21 will be admitted, but these students are not +allowed to purchase alcohol. The police routinely search all nightclubs and bars +in Albany for students under age 21 who have consumed alcoholic beverages or +used false identification to enter the establishment. Students who are caught +are arrested, taken to jail, and fined. They may also face university disciplinary +measures. + +Driving a car after drinking alcoholic beverages is also illegal in Georgia, +regardless of age. Driving under the influence (DUI) of alcohol carries severe +penalties and can result in the loss of your driver’s license, heavy fines, +imprisonment, disciplinary action by the university, and/or deportation. NEVER +DRINK AND DRIVE. + +SMOKING + +27 + + Smoking has become socially unacceptable in many places. Smoking is not +permitted in Albany State University buildings (except dorm rooms). Smoking is +also forbidden in hospitals, many restaurants, and other public places. Always +ask permission of your companies before smoking. Never smoke in someone’s +home unless he or she smokes as well. + +MAIL +CAMPUS MAIL +Campus mail is a service provided by the university for mail going between +people and offices on the premises of the campus. To send a letter by campus +mail, clearly mark the office and person it should go to and write CAMPUS MAIL +in the upper right hand corner. No stamp is needed. Drop the letter in the +collection point at the post office or in your Department Office. + +POST OFFICES +The university post office can send materials in both domestic and overseas mail. +Overseas mail must be weighed and stamped according to weight. Domestic +mail requires standard postage of 44 cents per letter. Packages and letters +containing more than five sheets of paper usually require extra postage. Any +package weighing over 16 ounces may be mailed first class from the university +post office. Domestic mail also requires a zip code for delivery. This is a five- +letter code specific to the area the letter or package is being sent to. If you do +not know the zip code of the area you are mailing to, ask for a zip code directory +at the post office and look up the proper zip code. + +CHANGE OF ADDRESS +Students who move off campus or leave ASU for a vacation should fill out a U.S. +Postal Service Change of Address card, sign the card, and place it in a U.S +mailbox. By completing this card, you allow the postal service to automatically +forward your mail to your new address. There is no charge for this service. +Unfortunately, the Office of the International Student Advisor cannot forward +mail to students who have moved off campus. Please do not have your mail +sent to the Office of the International Student Advisor. You will also need to +complete an AR-11 form which you obtain in the International Office. + +ELECTRONIC MAIL +An electronic mail (e-mail) account is available to all ASU students, faculty, and +staff free of charge. All ASI students are expected to use their ASU Ram Mail +because that is the official mode of communication. If you need an account or +would like to learn more about this service, call the Office of Information +Technology at 229-430-4909. You can also use the Internet through ASU +website: www.asurams.edu. E-mail and the Internet are wonderful ways to +communicate with friends around the world at no expense to you. + +MONEY AND BANKING +BANKS +There are many banks located in Albany. Most of them have specially priced +services for ASU students. The Bank of America has an automated teller machine + +28 + + or ATM in the Reese Student Union building on lower campus. As with any other +service, you should determine which banking services are most important to you +and call the different banks to find out about prices and available services. + +CHECKING AND SAVING ACCOUNTS +All students should open checking and/or saving accounts at a local bank. It is +not wise to carry or keep large amounts of cash. Most local businesses will +accept checks from local banks only. + +It is important to know how much money is in your checking account at all +times. Do not write checks for more money than you have in your checking +account. Checks written for more than the amount in an account are returned +unpaid. You will have to pay a fine to the bank, to the business where you +wrote the check, and may be unable to write checks in the future. Writing checks +when you do not have enough money to cover the check is called “bouncing +checks.” Bouncing checks is a criminal offense punishable by imprisonment. A +checking account is a service. Charges for services vary from bank to bank. +Savings accounts allow you to deposit money safely in a bank and receive a +small amount of interest on the balance. Savings accounts require passbooks for +depositing and withdrawing money. Keep both your checkbook and your +passbook in a safe place and report it to the bank immediately if either document +is lost or stolen. + +CREDIT CARDS +Credit cards are available through most banks. They have high interest rates, +but are very easy to use. Credit cards can be used almost everywhere and are +billed on a monthly basis. The best way to use a credit card is to pay the entire +bill monthly; otherwise, your balance may get higher than you can pay. Many +students find it helpful to carry an international credit card such as Visa or +MasterCard. Purchases on these cards can be made in U.S. dollars while the bill +is sent to your home and is denoted in your home currency. These cards can +also be used at automated teller machines to withdraw cash. Be sure to inquire +about the transaction fees before you obtain a credit card. Often the use of +these cards is an inexpensive way to exchange currency. But the fees can +become expensive. Your student fees can also be paid by credit card +(MasterCard or Visa only). + +OTHER BANKING SERVICES OFFERED FOR FEES + +Cashier’s Check +Money Orders +Certified Checks +Travelers Checks +Safety deposit box rental for storing valuable items + +MONEY EXCHANGE +Foreign currency must be taken to a bank to be exchanged. All currency (except +Canadian) must be sent out of town by the local banks for exchange. There is a + +29 + + charge of about $10.00 for the transaction and a processing time of 3 to 6 +weeks. + +RECEIVING MONEY FROM HOME + +1. + +2. + +By Mail +Tell your family to send numbered cashier’s check in U.S. +dollars, purchased from a bank in your country that has +business arrangement with a well-known U.S. bank. Make sure +that the check has a bank routing number on the bottom of the +check. + +By Telex +You may receive money from home by telex (wire). Wired +money generally arrives 5-10 days after it is sent from overseas. +It should be arranged through a local bank, and you should +have a checking or savings account established at the bank. +The receiving bank will charge about $12.00 as handling fee. It +is not possible to have money wired directly to your ASU +student account. + +Please be sure to request your funds from home well in advance of fee payment +deadlines. International transfers can often experience unexpected delays, so +you should plan ahead each semester. THERE IS NO FLEXIBITY IN FEE +PAYMENT DEADLINES. Please do not request an extension of the set fee +payment deadline. Failure to pay your student fees in a timely manner will result +in your having to return home. + +30 + + INTERNATIONAL STUDENTS +If English is not a student’s native language, if the student did not attend high +school in the United States, and if the student wishes to discuss taking the +Regents’ Test especially designed for International students, the student must +contact the Regents’ Test Director whose office is in 103 Simmons Hall (229-430- +4735 – flo.hill@asurams.edu). The local International Students’ Task Force +Committee has suggested that a student be considered for a special form of the +essay test after taking and failing the test at least once. There is no special test +for reading at this time; however, students may be given more time and may be +allowed to use their translation dictionaries. + +RELIGIOUS LIFE +Religion is a very important part of South Georgia culture. People will probably +invite you to visit their church and to take part in church sponsored social +activities. While these church services and activities can be very meaningful to +you, participation will not obligate you to join a church. Churches often attempt +to bring students into their church families by asking students to change their +belief to match the beliefs of the church members. If church members or clergy +make you feel uncomfortable about your religious belief or your values, feel free +tell them that you wish to maintain your own beliefs. If they cannot accept that, +find another church group! + +There are many churches located in the Albany area. Denominations include: +African-Methodist Episcopal, Apostolic, Assembly of God, Bahai, Baptist, Catholic, +Charismatic, Christian, Church of Christ, Church of God, Church of Jesus Christ of +Latter Days, Episcopal (Anglican), Holiness, Interdenominational, Jehovah +Witness, Lutheran, Nazarene, Pentecostal, Salvation Army, Seventh Day +Adventist, and Unitarian. All denominations are listed under “Churches” or +“Synagogues” in the Yellow Pages of the telephone book. + +SHOPPING +GROCERY STORES +All foods are located in one store. You will find fresh fruits, vegetables, canned +(tinned) goods, baby foods, meats, bread, cleaning supplies, paper products, and +frozen foods. Generally, grocery stores have lower prices than smaller stores. +There are several large grocery stores in Albany and some of them are within +walking distance of the University. You can order special items if you are +looking for ingredients not readily available in Albany stores. + +DRUGSTORES AND PHARMACIES +This is the only place where you can buy medicine that the doctor has ordered +for you. You can often buy many other things there as well. The most +convenient drugstore is CVS on Radium Springs Road, near the campus. Less +expensive pharmacies are located in discount stores and some grocery stores. + +31 + + DISCOUNT STORES +These are department stores where you find a large variety of things including +clothing, pharmacies, garden supplies, small appliance, paper products, linens, +and even some types of furniture. They often have lower prices and can be +wonderful places to see American people and products. Some of the local +discount stores include K-Mart, Wal-Mart, and Target. + +RESTAURANTS +Unless you are buying food, restaurant owners do not like for you to sit and +“take up space” from paying customers. You should leave money (a tip) on the +table for the waiter or waitress when you eat at a restaurant. The tip is usually +15%-20% of the total bill, depending on the quality of the service. Fast food +restaurants do not have waiters and waitresses. Do not leave a tip at this type +of restaurant. There are a number of commercial fast food restaurants around +the campus. Food can be ordered “for here” (to be eaten in the restaurant) or +“to go” (to be eaten at home). + +THE ALBANY MALL +The Albany Mall is a shopping center housing several dozen stores and +restaurants. It is located some miles west of the campus on Dawson Road. You +can buy almost anything you need in the Mall. Prices in these stores are usually +higher than the prices at Discount Stores, but the Mall is a great place to meet +people and experience American culture! + +OUTLET MALL +A large mall containing only manufacturer’s outlet stores is located many miles +south of Albany on Interstate 75. This mall has over 100 stores and is a good +place to shop because the prices are low while the quality of the merchandise is +high. + +Student Athlete + +Best practices For International Athlete Recruits (NCAA) +F-1 Visa and Issuance of SEVIS Form I-20 +International student-athletes who require an F-1 student visa are subject to +specific requirements regarding their admissions eligibility, financial ability and +academic status. Failure of students to follow appropriate regulations and +guidelines can result in severe difficulties with their immigration status with the +U.S. Department of Homeland Security. To avoid difficulties, the NCAA strongly +recommends that athletics staff meet with appropriate school officials (e.g., +international admissions officers, international student advisors) to gain an +understanding of the school’s policies and procedures that are in place to protect +international students and to maintain compliance with all immigration +regulations. + +32 + + IMPORTANT NOTE FOR ATHLETICS ADMINISTRATORS AND ADMISSIONS OFFICERS +Under no circumstances should an international student athlete be issued a Form +I-20 with the expectation of financial support from the athletics office before the +student has been certified as a final qualifier by the NCAA Eligibility Center or is +determined to be a partial qualifier in Division II. + +• Doing so opens the possibility for an international student who might not be +eligible for financial aid to obtain a visa and enter the United States without +having the sufficient funds to cover tuition and living expenses, as required by +immigration laws. + +Financial Ability + +• All F-1 visa students must submit appropriate documentation of financial +support before the school can issue the SEVIS Form I-20 [certificate of eligibility +for nonimmigrant (F-1) student status]. +• Generally the school’s office of recruitment and admissions office will have +information as to what constitutes appropriate documentation of financial +support. +• Athletics department staff members must understand that withdrawal of +athletics financial aid for an international student-athlete could result in severe +economic hardship for the student and could result in a violation of the student’s +immigration status. + +What Athletics Administrators Should Tell Student-Athletes +• Student-athletes who will be receiving financial aid from the athletics +department must understand that if the amount of aid is less than the +amount required for issuance of the I-20, then the student will need to +provide additional documentation of financial support. + +• Student-athletes should be informed of the true costs of attending the + +school and must understand the financial implications in the event of the +loss of financial support from athletics. + +• Although students athletes may receive some financial assistance through + +athletics students are responsible for purchasing their books for the +duration of the academic tenure. + +• Student-athletes must understand that if their scholarship is made + +unavailable for whatever reason (e.g., because the student-athlete did not +meet initial-eligibility requirement, they must have other means of +financial support. If sufficient finances are not available from other +sources, the student risks violation of immigration status and possible +deportation. + +SOCIAL SECURITY NUMBERS +A Social Security Number is a government identification number issued to all +residents of the United States. Most residents receive this number as children. +In the U.S., we do not carry passports or national ID cards. Instead, we use this +number and a driver’s license as our primary forms of identification for +employment records, taxation, opening bank accounts, university records and + +33 + + other business matters. All drivers’ licenses contain a color photograph of the +bearer as well as the current address and date of birth. + +For your admission to ASU, you were issued a student identification number +which has the same number of digits as the social security number. The +university uses this ID number for all of your university records. The number +also appears on your student identification card. While this number is sufficient +for your academic records, you will need to obtain a Social Security number +before you can work on campus. To obtain the number, you must complete an +application form and deliver it, along with a letter of verification from the ISA +and an offer letter from the person who is hiring you on letterhead, in person to +the Social Security Administration in Albany. There is no charge for Social +Security numbers. Call 1-800-772-1213 for more information. + +TAXES +Federal Law requires that foreign students pay income tax from sources within +the U.S. This income generally includes salary from any employment, on or off +campus, as well as scholarship money. Tax return is due on April 15 for the +previous year. In addition you may need to file a tax statement. If you have +concern about your tax obligation, you may wish to consult a certified public +accountant (CPA). The Internal Revenue Service is the branch of the federal +government responsible for income tax collection. This agency is often simply +called IRS. You can call the IRS directly at 1-800-829-1040 for additional +information. + +TELEPHONE AND TELEGRAMS +TELEPHONE +Telephone service in the U.S. is divided into two types: local and long-distance +(including international) service. Most people work with two different companies +to obtain telephone services: a local carrier for local service and long distance +carrier for long distance service. The local carrier for Albany is BellSouth. +Contact the company by calling 229-780-2355. You may need to pay a deposit +to begin your local service. For about $30.00 a month, you may have unlimited +telephone calls within the country. You will be billed monthly for this service. + +There are several local and long distance services in the U.S. Each company lays +claims to lowest telephone rates for long distance. You may want to talk to each +company to learn who can offer you the best deal. Rates usually depend on +where you call and when you call. To get the cheapest rate, you may find that +all three companies have special plans with many restrictions. Be sure to read +the details of the special plans. Be sure to ask for the rates for calling your +country. You can contact these companies by calling the numbers listed in the +local telephone book. Once you choose a company, they will provide you with +direct dialing from your home and a calling card, which you can use from any +phone. + +INFORMATION + +34 + + If you cannot find a number on-line, you may call “Information,” also called +“Directory Assistance” and the number is 1411. In order to use the information +service, you must know the name of the person or organization you wish to call +as well as the city they are located in. The number you dial for information +depends upon the area you wish to call. The telephone book also contains useful +information about how to use the telephone and place different kinds of calls. +The Yellow Pages found on-line contain listings of local businesses and can help +you locate services. + +TOLL FREE CALLING +Most numbers with an area code of 1-800 or 1-888 are toll free calls. This +means that there will be no charge for the call. Many businesses and services +have toll free numbers for customer use. You should use toll free numbers +whenever possible to keep your telephone bill low. + +OTHER ASPECTS OF TELEPHONE SERVICES +If you live off-campus and share a telephone number, be sure to have an +agreement about how the bill will be paid and whose name will appear on the +bill. The names and numbers of all telephone subscribes are published in the +BellSouth telephone book unless you request an unpublished number. Albany has +several “local exchanges.” This term refers to the first three digits of the +telephone number. You will see prefixes listed as 430, 431, 434, and 435, for +example. Be sure to give all seven digits of your number. If you are calling from +one phone on campus to another, you only need to dial the last four digits of the +number. Many students today buy pre-paid cell phones. + +TELEGRAMS +Letters or money may be sent both within the U.S. and internationally through +Western Union by calling their toll-free number 1-800-325-6000. Mailgram is also +available through Western Union and is less expensive than telegrams. To send +a Mailgram, call Western Union at 1-800-325-6000. MoneyGram provides similar +services as Western Union. Their toll free number is 1-800-926-9400. + +FAX +Faxes may be sent or received through the fax services offered by some +companies in Albany. You will be charged for this service. For international +faxes, students should use Mailboxes, etc. at 229-434-1133 or 229-432-2230. +Unfortunately, the high cost of international and long distance faxes prohibits the +Office of the International Student Advisor or the University from sending +personal faxes for students. + +TESTING +The Albany State University Computer-based Testing Center (Department of +Counseling, Testing and Disability Services), located on the first floor in ACAD +195, administers and prepares students for several national computer-based +tests periodically throughout the year. The GRE, GMAT, MAT, MCAT, LSAT, +PCAT, and TOEFL are a few of the tests administered by the center. National test +scores are required for admission to U.S. graduate schools. These tests have + +35 + + standard fees and deadlines for enrollment. For information about a test, call +the Computer-based Testing office at 229- 430-1631. To sign up for a test, visit +the Division of Student Affairs and pick up an application. + +TRANSPORTATION +TRANSPORTATION IN ALBANY +Albany has a public transportation system. In emergencies, you can hire a taxi +(cab). Companies offering taxi services are listed under “Taxicabs” in the +Albany telephone book. Be sure and ask how much you will be charged before +hiring a taxi. Some students purchase bicycles for transportation. Cyclists must +ride on the roadways, as there are no separate bicycle paths. If you choose to +ride a bicycle, you must be very careful and watch out for cars. American drivers +are not accustomed to cyclists on the road, be sure to lock up your bike when +you are not using it. To travel outside Albany, you must go by bus or airplane. -ANN CECIL +BUS TRANSPORTATION +The Greyhound Bus Lines station in Albany is located at 300 Oglethorpe Blvd. +Call 1-800-231-2222 for nation-wide information or 229-432-0511 for local +terminal information. - +AIR TRANSPORTATION +Albany has an airport located south of town. Service is limited to several daily +flights to Atlanta, where connecting flights are available to other cities. Call 229- +430-5175 for schedules. -LAWRENCE WORCESTER +TRAVEL AGENTS +Travel agents will help you make travel arrangements for a small fee. They +often save you money by finding the lowest rates available to your destination. +Look in the Yellow Pages under “Travel Agencies”. Or you may book your tickets +online using a credit card at the airlines website. (e.g. www.delta.com) -RICHARD CUMMINS +INTERNATIONAL TRAVEL +If you know that you will return to your country during a school vacation, you +should try to plan international travel 2 or 3 months in advance. The best rates +for international travel are obtained by making reservations early. Also, all +students are eligible for discounted airfares through an organization called +Council Travel. To obtain these special fares, you must have an international +student ID card. These cards can be purchased for $20 from the Office of the +International Student Advisor or local travel agents. These cards also qualify the +bearer for discounts at museums and other tourist attractions. If you have +questions about international travel, you should speak with the staff at the Office +of the International Student Advisor or experienced travelers from your country. - LAKE UNION, CAPITOL HILL & VOLUNTEER PARK +36 -Lake Union hums with the sound of seaplanes, but you can have fun skirting the -edge on a bike or boat. Nearby Capitol Hill is an artsy and diverse neighborhood, -where you’ll find cafes, galleries and street theater. If you need a break from kayaks -and counterculture, stop at the Volunteer Park Conservatory. + CAMPUS CONTACTS - Wall of Sound -Flip through racks of music (p114) - Jimi Hendrix Statue -Check out Daryl Smith’s 1997 statue of the -guitar legend (p80) - Lake Union -Paddle or cycle around this Y-shaped lake (p76) - Volunteer Park Conservatory -Brush past palms in Seattle’s tropical corner -(p80) +Quick Reference - +Information - +Office -LAWRENCE WORCESTER +Phone Number -LAWRENCE WORCESTER +Advising and Immigration - +Office of Int’l Student Advisor -ANN CECIL +Meal Plan - +Food Services -JOHN ELK III +General Info (Switchboard) -9 +Facilities Management -  +Graduate Studies - +Graduate School -10 +Residence Life (Housing) -FR E M O NT & WALLI N G FO R D +Student Affairs -Fun-loving Fremont has coffee shops, pubs and some of the oddest public art that -you’ll ever see, making a visit to this neighborhood a must. Pleasant Wallingford, on -the route of the Burke-Gilman cycle trail, is home to some noteworthy restaurants -and Gas Works Park. +Mail delivery services for on-campus students -LAWRENCE WORCESTER +Post Office - Gas Works Park -Find yourself drawn inexplicably to a strange mass of pipes (p91) - Fremont Troll -Quake before the 1990 statue, the Fremont Troll (p88) - Statue of Lenin -Consider the seven tons of Slovakian scrap that make up Emil Venkov's -1988 statue (p88) +Parking Permits - +Public Safety -LAWRENCE WORCESTER +229-430-7075 -LAWRENCE WORCESTER +229-430-4750 - BALL AR D & D I SCOV E RY PAR K +229-430-4600 -Ballard is a Scandinavian-influenced neighborhood that lies just over Ballard Bridge. -It seems a world away from urban Seattle and is known for its down-to-earth bars -and live-music venues, as well as the interesting Fish Ladder. Neighboring Discovery -Park is 534 acres of urban wilderness. +229-430-5118 - +229-430-4741 -DANITA DELIMONT/ALAMY +229-430-4745 - Ballard Fish Ladder -Spot salmon on their way to spawn (p97) - Nordic Heritage Museum -Learn more about Ballard's heritage (p95) - Discovery Park -Walk the trails and escape the city (p98) +229-430-4711 - +Personal & Counseling -RON BUSKIRK/ALAMY +Counseling, Testing & Disability Services - + 229- 430-4667 -RICHARD CUMMINS +Reporting Theft or Crime -G R E E N L AK E & PH I N N E Y R I D G E +Public Safety -If you like zoos, parks, rose gardens and low-key neighborhoods, then you’ll love -Green Lake and Phinney Ridge. Green Lake Park is a perfect place for sunbathing -in summer or walking in fall, and Woodland Park Zoo is one of Seattle’s greatest -tourist attractions. +Admission Status - Woodland Park Zoo -Strut your stuff at the zoo (p94) - Seattle Rose Garden -Smell the scent of 5000 roses (p94) +Admissions Office - +Student Activities -LAWRENCE WORCESTER +Student Affairs - +Transcripts (to be sent from ASU) -JOHN ELK III +Registrar’s Office -11 +Tuition Payment -  +Business Office - +229-430-4711 - +229-430-4646 -12 +229-430-4739 -AROUN D SE AT TLE - -RICHARD CUMMINS - -Situated in a stunning part of the Pacific Northwest, Seattle is within easy reach of -mountain ranges, the ocean, hundreds of islands and a temperate rainforest. This -makes Seattle a perfect hub for getting out and exploring the natural world. - - Mt Rainier National Park -Climb the highest peak in the Cascades (p193) - Hoh River Rain Forest -Hike in a mossy temperate rainforest (p198) - San Juan Islands -Kayak around breathtaking islands (p189) - Ferry across Puget Sound -Ride the waves and look back at Seattle’s -famous skyline (p187) - -JOHN ELK III - - - -TOM BOYDEN - -LAWRENCE WORCESTER - - TH E AUTH O R - -Becky Ohlsen - -Becky has lived in -the Pacific Northwest -for the past 15 years, -originally drawn here -from her native Colo- -rado by the promise -of great music and -the ocean. Upon -arrival she discovered -that the area also pro- -vided excellent beer, -a wide variety of pinball machines, cute skater -boys and lots of rainy days in which to moodily -read in coffee shops. She lives in Portland, -Oregon, after surviving a brief, educational, -expensive and decidedly temporary stay in -New York City, where against everyone’s bet- -ter judgment she earned a master’s degree in -journalism. In addition to covering Seattle -for Lonely Planet, she has worked as a copy -editor and freelance writer for alternative -newspapers and magazines, and occasionally -reviews movies for a local independent radio -station. When she’s not traveling for guide- -book research, Becky races vintage motorcy- -cles very, very tentatively. She’s a member of -the Sang-Froid Riding Club and a veteran of -the infamous ‘555’ Ride, and much prefers -two wheels to four. Her favorite bar in Seattle -is Shorty’s and her favorite Nirvana album -is Bleach. - -LONELY PLANET AUTHORS -Why is our travel information the best in the world? -It’s simple: our authors are passionate, dedicated -travelers. They don’t take freebies in exchange for -positive coverage so you can be sure the advice you’re -given is impartial. They travel widely to all the popu- -lar spots, and off the beaten track. They don’t research -using just the internet or phone. They discover new -places not included in any other guidebook. They per- -sonally visit thousands of hotels, restaurants, palaces, -trails, galleries, temples and more. They speak with -dozens of locals every day to make sure you get the -kind of insider knowledge only a local could tell you. -They take pride in getting all the details right, and in -telling it how it is. Think you can do it? Find out how -at lonelyplanet.com. - -BECKY’S TOP SEATTLE DAY -When I’m in Seattle I like to wake up unchar- -acteristically early and go straight to Pike Place -Market (p62). The scene here is fascinating any -time of day, but it’s best in the morning, when -most of the customers are locals getting their -shopping done. It’s a nice place for a walk- -ing breakfast, too, with options ranging from -fresh fruit to coffee and pastries to a hefty hum -bao (steamed pork bun) big enough to power -you through the whole day. After exploring -the labyrinth of weird little market shops, I’ll -browse the progressive literature at Left Bank -Books (p110) to keep myself honest and bolster -hope. Afterwards, a short stroll down toward -Elliott Bay gives me a panoramic view of the -mountains from the new Olympic Sculpture -Park (p67). - -Of course I can’t be in Seattle without -spending some time digging around the -shelves at Elliott Bay Book Company (p113), -and as long as I’m this close I might as well -treat myself to an artisan ice cream at Molly -Moon’s (p131) for lunch. - -I’ll spend the rest of the afternoon soaking -up culture at the Seattle Art Museum (p48), -abandon it for happy hour at the Nitelite -Lounge (p140), then probably squander the -evening and whatever’s left in my pockets -on pinball and beer in the back room of -Shorty’s (p142). +229-430-4638 -15 +229-430-4610 - G E T TI N G STAR TE D - -Other than finding accommodations, Seattle doesn’t require a huge amount of pre-planning. -Most of the can’t-miss sights are clustered around downtown within easy walking distance of -each other. If you bring a decent map, book accommodations in advance and have some sort -of plan for transportation (whether it’s a car and an idea of where you’ll park overnight, or a -public-transport schedule), you’ll be fine. Keep in mind that the city is hillier than it looks on -maps and – if you’re driving – many streets in the city center are one-way only. - -Once you’re there, getting around is relatively easy. Driving and parking are the main an- -noyances – if you can avoid them, you’ll be happier. Luckily this is easy to do. Buses connect -to most neighborhoods of interest to visitors, although they stop running too early to be of -much use for nightlife exploration. But then, that’s why Mother Nature gave us taxi cabs (see -the Transportation chapter, p199). - -For more help with planning, visit Seattle’s Convention and Visitors Bureau (Map p50; %206-461-5888, 866- -732-2695; www.visitseattle.org; 701 Pike St; h9am-5pm). The website has a schedule of events, a few discount -coupons and a form for ordering a visitor-information packet. Also among the bureau’s services -is the Citywide Concierge Center, a free service for visitors that will book you a table, a tour or -transportation and answer questions about events going on while you’re in town. - -G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G - -O WHEN TO GO - -Unless you’re a skier or snowboarder – or -really interested in aquatic life – Seattle in -the winter makes for a pretty dreary holiday. -The city receives 65% of its precipitation from -November to March. But in spring the relent- -less downpour frequently lets up to provide -gorgeous days that are well worth the wait. - -April and May are best for enjoying the -fruits of all that gloomy winter rain: sud- -denly everything is supernaturally green and -covered in flowers. Summer – June, July and -August – is the high season and can usually -be relied upon for sunny days, warm weather -and lots of people saying things like ‘I thought -you said it rained all the time here.’ Fall, the -season of Bumbershoot and Oktoberfest, is -the favorite of beer lovers and music fans. - -FESTIVALS -January -CHINESE NEW YEAR -%206-382-1197; www.cidbia.org -Beginning toward the end of January or -the start of February and lasting for two -weeks, the year’s first big ethnic festival -takes place in the International District; you -can contact the Chinatown/International -District Business Improvement Association -for information. The first day of the New -Year festival is celebrated with parades, -firecrackers, fireworks and plenty of food. +37 -16 + SERVICES FOR INTERNATIONAL STUDENTS -February -NORTHWEST FLOWER & GARDEN -SHOW -%800-229-6311; www.gardenshow.com; -Washington State Convention & Trade Center -Usually held the second week of February, -this popular event includes lectures and -seminars, demo gardens and samples, and -children’s activities. - -March -MOISTURE FESTIVAL -%800-838-3006; www.moisturefestival.com; -Hale’s Palladium, 4301 Leary Way NW; -tickets $5-25 -With regular performances held between -March 15 and April 1, the Moisture Festival -has developed from a little-known -quirkfest into something that locals look -forward to each year. It has retained a -laid-back, casual vibe and usually features -random appearances by folks like Artis -the Spoonman, Circus Contraption and -The Bobs. - -ST PATRICK’S DAY -March 17 is the day the patron saint of -Ireland is honored by all those who have -Irish in their blood, and by those who want -to have Irish beer in their blood. Everyone -wears green (otherwise you risk getting -pinched). Irish bars such as Kells serve -green beer, and even bars that haven’t - - an Irish trinket in sight get swamped with -revelers. Catch the parade going from City -Hall to Westlake Center along 4th Ave. - -April -CHERRY BLOSSOM & JAPANESE -CULTURAL FESTIVAL -%206-723-2003; www.seattlecenter.com; -Seattle Center -Usually held in mid-April, this celebration of -Japanese heritage includes performances -of music, dance and drama. It’s part of -the Seattle Center’s series of multicultural -festivals, dubbed Festál. - -May -CINCO DE MAYO -May 5 commemorates the day on which -the Mexicans wiped out the French -Army in 1862. Now it’s the day all -Americans get to eat lots of Mexican -food and drink margaritas. Seattle -celebrates with an annual parade and -exhibitions of Hispanic artwork. - -OPENING DAY OF YACHT SEASON -%206-325-1000; www.seattleyachtclub.org -Held the first Saturday in May at various -locations on Lakes Washington and Union, -this Seattle original starts with a blessing of -the fleet. It features scull racing and a boat -parade through the canals. - -SEATTLE INTERNATIONAL -CHILDREN’S FESTIVAL -%206-684-7338; www.seattleinternational.org; -Seattle Center -Held in mid-May, this cultural extravaganza -includes a wide variety of performances -and activities for kids. - -NORTHWEST FOLKLIFE -FESTIVAL -%206-684-7300; www.nwfolklife.org; -Seattle Center -This festival – the largest of its kind -in North America – takes over Seattle -Center during Memorial Day weekend. -More than 5000 performers and artists -from over 100 countries present music, -dance, crafts, food and activities in -celebration of the rich cultural heritage -of the Pacific Northwest. - -SEATTLE INTERNATIONAL FILM -FESTIVAL -%206-324-9996; www.seattlefilm.com -Held from late May to mid-June, this -festival, the largest of its kind in the -nation, brings nearly a month’s worth -of international film premieres to -Seattle. Screenings take place at several -theaters across the city, including the -official SIFF Theater in Seattle Center. -Check the film festival’s website for a -detailed schedule. - -ADVANCE PLANNING -As in any large American city, planning ahead will let you spend more of your time enjoying Seattle. Things like car -rental, accommodations and tours should be booked in advance. If you show up without a hotel booking in peak season, -you might find yourself stranded. - -Train tickets to Seattle may need to be booked in advance, as some trains do sell out. Make reservations at www - -.amtrak.com or by phone at %800-872-7245. - -If you’re hoping to see a particular performance or game, whether it’s the Mariners or the opera, it’s wise to buy -tickets in advance. See individual listings chapters for price ranges and information on how to book (for sporting events, -see p165; for theater and other performance events, see p159). - -Some restaurants recommend or require booking a table in advance, and these are noted in the Eating chapter - -(p119). For most places in Seattle, booking a day or two ahead is enough, unless it’s a holiday. - -The Seattle CityPass (www.citypass.net), which offers discounts on Seattle museums and other attractions, can be -ordered in advance on its website. The website www.dailycandy.com produces a ‘weekend guide’ to Seattle; it’s worth -subscribing to the mailing list before a trip. - -It’s useful to read up on what’s happening in town via The Stranger’s savvy and entertaining online blogs, Slog (for -news) and LineOut (for music and nightlife) at www.thestranger.com. Rabid menu-readers will appreciate the Seattle -Times’ archive of restaurant reviews at www.seattletimes.com. - -Seattle’s Citywide Concierge Center (%206-461-5888; www.visitseattle.org; 701 Pike St) can help with advance - -bookings of almost anything you’ll need, including accommodations, tours and the CityPass. - -l - -o -n -e - -l -y -p - -l -a -n -e -t -. -c -o -m - -G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G -O + Orientation sessions and workshops for new students. + Liaison with the U.S. Citizenship and Immigration Services (USCIS) for +all of your immigration documents. We will help you with filing for +practical training, transfer of schools, and replacement of lost +documents. WE are your source for maintaining your immigration +status. -17 + Counseling and referral services. Personal and cross-cultural - June -FREMONT STREET FAIR -%206-649-6706; www.fremontfair.com -Off-kilter Fremont is just the place you’d -want to be for a street fair. This one has live -music, food and crafts, and the overtly artsy -Solstice Parade, where human-powered -floats traipse through the neighborhood -in a lively tribute to quirkiness. It takes -place adjacent to the ship canal, usually -in mid-June. - -SEATTLE PRIDE FESTIVAL -%206-684-7200; www.seattlepride.org; -Seattle Center -Seattle’s lesbian- and gay-pride event -usually falls on the last Sunday in June and -includes a film festival, art exhibit, parade -and the requisite food and entertainment. - -July -BITE OF SEATTLE -%425-283-5050; www.biteofseattle.com; -Seattle Center -This culinary celebration is usually held the -second weekend in July. Guests can sample -food from dozens of Seattle-area chefs, -including sub-$4 samples at the Just a Bite -booth, and taste local beer and wines. The -evening ends with live music. - -SEAFAIR -%206-728-0123; www.seafair.com -A month-long extravaganza, Seafair -is a civic celebration that began as a -hydroplane race on Lake Washington. -It’s now the city’s biggest festival, with -all manner of festivities extending across -Seattle. Events include a torchlight parade, -an air show, lots of music, a carnival and -even the arrival of the naval fleet. Lodging -is in short supply in Seattle on Seafair -weekends and traffic becomes hopelessly -tangled thanks to closed bridges and the -influx of visitors, so plan accordingly. - -September -BUMBERSHOOT -%206-281-7788; www.bumbershoot.org -Seattle’s biggest arts and cultural event -takes over Seattle Center on Labor Day -weekend (the first Monday in September). -Hundreds of musicians, artists, theater - -l - -o -n -e - -l -y -p - -l -a -n -e -t -. -c -o -m - - G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G -O +counseling is available on a confidential basis to all students. If you +would like to talk to someone, call the International Student Advisor +(ISA) at 229-430-4721 to make an appointment. If necessary, you +may be referred elsewhere for help. -18 + Information. If you do not know what to do, to whom to talk, or +where to find ANYTHING, the Office of the International Student +Advisor can help. -troupes and writers come from all over -the country to perform on the festival’s -two dozen stages. It’s hard not to find -something you like. There’s also a crafts -street fair and lots of good food from -local vendors. - -SEATTLE AIDS WALK -%206-329-6923; www.seattleaidswalk.org -This festive and friendly event, held in late -September, grows every year. It’s a 10km or -5km walk (or run). Locations vary, so check -the website for details. The walk benefits -the Lifelong AIDS Alliance. - -FREMONT OKTOBERFEST -%206-632-1500; www.fremontoktoberfest.org -There are versions of Oktoberfest held -all over town, but this one’s a good -introduction to the concept as well as the -neighborhood. Held in late September, -it includes crafts, a car show, live -entertainment and, of course, beer tasting. - -October -SEATTLE LESBIAN & GAY FILM -FESTIVAL -%206-323-4274; www.seattlequeerfilm.com -Held in the third week of October, this -film festival brings both mainstream -and underground queer films to various -theaters around town. - -HALLOWEEN -On October 31, kids and adults dress up in -scary costumes. In the safer neighborhoods -you’re likely to see children out ‘trick or - -OFFBEAT SEATTLE FESTIVALS - - Bumbershoot (left) A cultural buffet of music, -theater, artwork and poetry on 25 stages. - Fremont Street Fair (left) Celebrate one of - -Seattle’s quirkiest corners. - - Winterfest (opposite) A celebration of all that’s - -cold, furry and white. - - Trolloween (above) Just in case regular Halloween - -isn’t weird enough for you. - - Moisture Festival (p16) Performances by oddball - -artists both local and international. - - treating’ door-to-door for candy. The -gay and lesbian bars are especially wild -places on Halloween, and Fremont has -its own version, Trolloween (named after -the Fremont Troll). It features a candlelit -procession of costumed locals on Halloween -night, followed by a public dance. - -December -WINTERFEST -%206-684-7200; www.seattlecenter.com -Seattle Center holds a month-long -celebration of holiday traditions from -around the globe, starting with Winter -Worldfest, a massive concert and dance -performance, and continuing with exhibits, -dances, concerts and ice skating. - -NEW YEAR’S EVE -%206-443-2111; www.seattlecenter.com -The place to be on the 31st is Seattle -Center, where festivities are focused on the -Space Needle. Most people celebrate by -dressing up and drinking champagne, or -staying home and watching it all on TV. The -following day people stay home to nurse -their hangovers and watch college football. - -COSTS & MONEY - -As in most large cities, expenses in Seattle -can vary depending on your interests and -means. Accommodations will make up the -bulk of your expenses. Booking in advance -or planning your visit during the off-season -(September–May) will generally get you better -value for money. - -A dorm bed in a youth hostel costs about -$35 including taxes; a double room in one -of Seattle’s European-style hotels will go for -$60 to $80, plus tax. Midrange doubles are -between $100 and $200. See the Sleeping -chapter (p173) for details about Seattle Super -Saver hotel discount packages. - -After accommodations, dining will prob- -ably be your next-largest budget item. The -city offers a good range of meals, from spec- -tacularly decadent restaurant meals to quick, -budget-friendly cafe snacks. It’s a foodie city -that’s well positioned to take advantage of -fresh seafood and produce, which means -that in general splurges will be rewarded with -unforgettable dining experiences. Budget -meals are under $15 a person; breakfast and -brunch and most lunches fall in this category. - -HOW MUCH? -Adult ticket up the Space Needle $18 -Cover charge at the Crocodile Café $5 to $15 -Gallon of gasoline $3 -Late-night burrito from a taco cart $5 to $7 -Liter of bottled water $2 -Pint of local microbrewed beer $4.50 -Room at the hip Ace Hotel $99 -Souvenir T-shirt $20 -Taxi from airport to Pike Place Market $35 -Light-rail ticket from airport to downtown $2.50 - -Midrange prices hover in the $16 to $30 range, -and top-end dinners start around $30, plus -tax, tip and whatever you’re drinking. - -For discounts on sights and entertainment, -consider buying a Go Seattle Card (www.goseattlecard -.com) or a CityPass (www.citypass.net), both of -which offer free or discounted admission -to most of the city’s top attractions. See the -Neighborhoods chapter (p92) for details. - -INTERNET RESOURCES - -The following websites may be helpful in -planning your travels. -Bohonus VR (www.bohonus.com/galleries) If you just can’t -wait until you get there, check out this site for 360-degree -views of various Seattle sights, shops and hangouts. -Daily Candy (www.dailycandy.com) The Seattle edition -digs up new and fun things to see, do or buy each day. -Hankblog (http://hankblog.wordpress.com) Insider -art-related news and views from the folks at the Henry -Art Gallery. -Lineout (http://lineout.thestranger.com) The Stranger’s -music blog – a great way to get plugged into the scene -quickly. -Lonely Planet (www.lonelyplanet.com) Succinct summa- -ries on traveling to most places on earth; postcards from -other travelers; and the Thorn Tree bulletin board, where -you can ask questions before you go or dispense advice -when you get back. -Seattle DIY (www.seattlediy.com) A great source of -information on underground events and house shows. -Seattle’s Convention and Visitors Bureau (www.visit -seattle.org) Seattle’s official visitors information site. -Seattlest (www.seattlest.com) An informative (but less -savvy or sassy than the Stranger) blog about various news -and goings-on in and around Seattle. - -l - -o -n -e - -l -y -p - -l -a -n -e -t -. -c -o -m - -G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -C -O -S -T -S -& -M -O -N -E -Y + Letters of certification for government loans, money exchange, re- +entry into the U.S., invitations to visit the U.S., and other official +documents are available at the office. -19 + Information about Albany State University International Student - l - -o -n -e - -l -y -p - -l -a -n -e -t -. -c -o -m - -Slog (slog.thestranger.com) A frequently updated blog by -the staff of The Stranger newsweekly. -Sustainable Seattle (www.sustainableseattle.org) Though -it’s a bit academic, this site has thorough information on -ecofriendly projects and initiatives in Seattle. - -SUSTAINABLE SEATTLE - -Nature is a huge part of the Seattle experi- -ence and it should come as no surprise that -the city is good at being green. So what can -visitors do to help? - -International flights are pretty much un- -avoidable, so consider contributing to a car- -bon offset organization like www.carbonfund -.org, www.terrapass.com, www.nativeenergy -.com or www.driveneutral.org, where your -money is transformed into support for -sustain able practices. Or, if you’re close -enough, plan to travel on a biodiesel bus (p200) -or on a train (p201). - -Once you’ve arrived in Seattle, you can -easily get by without having a car. Local bus -networks are good, and biking is a great -option (rentals abound). Walking and using -the public transportation system are also good -earth-friendly options. And if you do need a -car, think about making use of www.flexcar -.com, a car-sharing program. Some car-rental -agencies offer hybrid vehicles, so be sure to ask -around. Also, take a look at www.craigslist -.com, an online bulletin board where you can -find regional rideshares (look under ‘com- -munity’); ie folks who are looking for people -to carpool and share the cost of gas. Similarly, -the adventurous can check www.couch -surfing.org, where you sleep on other member’s -couches – exchanging cultures, saving money -and using hopefully fewer resources. - -In hotels, save energy by requesting that -your sheets and towels not be laundered -every day. Shut off lights, the TV and air-con -when you’re not in the room. And consider -not using those inefficient little travel-sized -sundries. Hostels are another option as their - -high-density dormitories use less space and -energy. See the box, p180, for a list of hotels -that are making an effort. - -Sustainable Seattle, a nonprofit organi- -zation, holds an annual awards ceremony -in November to recognize local businesses -that meet certain sustainability criteria; visit -www.sustainableseattle.org to learn more -about the program. - -Our GreenDex (p227) at the end of this book -lists organizations and businesses that have -been recognized for taking extra measures to -help protect the environment, support local -charities, use local products or recycled ma- -terials, or promote regional culture, among -other positive practices. Here’s a sample: -Chaco Canyon Café (p131) Cafe serving all vegan, 90% -organic and almost half raw food; also an organic bakery. -FareStart Restaurant (p122) Profits support the FareStart -program, which provides intensive job training, housing -assistance and job placement for disadvantaged and -homeless people. -Foster Island Wetlands Trail (p100) A waterfront trail -that winds through marshlands and across to smaller -islands for bird-watching, swimming, fishing and kayaking. -Fremont Sunday Market (p86) A bustling marketplace -in the city’s quirkiest quadrant, there’s fresh produce, art -and junk. -Olympic Sculpture Park (p67) An excellent use of urban -space to present public art and sculpture. -Tillicum Village (p189) On Blake Island, tours include -traditional Indian salmon bake, dancing and a film about -Northwest Native Americans. -Tilth (p133) A certified-organic gourmet cafe in Wallingford. -Volunteer Park Conservatory (p80) A classic Victorian -greenhouse with five galleries representing different world -environments. -Washington Trails Association (p169) A nonprofit group -that organizes hiking trips, conservation efforts and -trail-building jaunts into local mountains. -Wing Luke Museum of the Asian Pacific American -Experience (p59) Museum devoted to the history of Asian -and Pacific culture and art in America. - -© Lonely Planet Publications Pty Ltd. To make it easier for you to use, access to this chapter is not -digitally restricted. In return, we think it’s fair to ask you to use it for personal, non-commercial -purposes only. In other words, please don’t upload this chapter to a peer-to-peer site, mass email -it to everyone you know, or resell it. See the terms and conditions on our site for a longer way of -saying the above - ‘Do the right thing with our content.’ - - G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -S -U -S -T -A -I -N -A -B -L -E -S -E -A -T -T -L -E +Association (ASUISA): ASUISA annually elects new student officers, +and sponsors activities throughout the year, by and for international +students. -20 + Other activities for all Global Programs students such as International + +Festival in November, picnics, coffee/tea hour, and the Cultural +Exchange Club/Global Ambassadors. + +38 + + CHECK LISTS + +ANNUAL CHECKLIST + +• Renew passport if necessary (six months before expiration). + +• File income tax returns if necessary (between January 1 and + +April 15). + +• Extend temporary stay with sponsor and with the + +International Advisor for an extended I-20 if necessary (30 +days before expiration). + +PERIODIC CHECKLIST + +• Report changes of address (within 10 days) to the Office of +the International Student Advisor and the Office of the +Registrar. This is required by law. So is your email +address and your telephone number. + +• Report any change of marital status, or if your family has + +joined you in the U.S., to the International Student Advisor. + +• Complete registration for Student Health Insurance every + +semester. + +• + +If leaving the U.S. temporarily, have 1-20 or DS-2019 +signed by a Designated School Official and check visa +validity. + +• Check in with International Student Advisor at the beginning +of each semester and schedule appointments periodically. +Always bring all of your documents when you come. + +Information from Handbook is from the following websites. +www.travel.state.gov/visa (U.S. Department of State) +www.ice.gov (Department of Homeland Security) +www.uscis.gov (U.S. Citizenship and Immigration Services) +www.ncaa.com (National Collegiate Athletic Association (NCAA)) + +39 + + Office of the International Student Advisor +Sue Ann Balch, J.D. +205 Reese Hall Lower East Campus, 504 College Drive +(229) 430-7075 (Office) +(229) 434-7106 (Fax) +Sueann.balch@asurams.edu (email address) + +40 diff --git a/q103/random_k4/random_4.md b/q103/random_k4/random_4.md index e3ed939376d2f6fad348f021570f136f5c225f39..a2b48e41bd57bce68e73b59f00704cb4c84c38f3 100644 --- a/q103/random_k4/random_4.md +++ b/q103/random_k4/random_4.md @@ -1,1527 +1,2098 @@ -Issue Brief - -MTV/AP-NORC: COMPARING THE -POLITICAL VIEWS OF YOUNG -PEOPLE AND THEIR -PARENTS’ -GENERATION - -Young people and their parents share a negative outlook on -the current political landscape in America and are largely -dissatisfied with the government’s handling of a wide -variety of key issues, according to a new poll of Americans -age 15-34, and parents of children age 15-26 by MTV and The -Associated Press-NORC Center for Public Affairs Research. - -Young people’s opinions about their generation’s ability to -impact the government have shifted in recent months, -following student-driven activism. Between the first poll -fielded between late February and early March 2018, -shortly after the shooting in Parkland, Florida, and this -second poll conducted roughly two months later, young -people age 15-34, and particularly those age 15-22, feel -more politically empowered. The proportion overall who -believe that elected officials care what they think rose -from a quarter to a third, and there was a similar rise in -the numbers who think people like them can impact the -government. - -The generation of parents of children age 15-26 is less -racially and ethnically diverse and includes a greater -proportion of Republicans than young people age 15-26, -and the survey demographics reflect this. Parents are -more likely to approve of President Donald Trump (42 -percent vs. 26 percent) and are somewhat more optimistic -(59 percent vs. 45 percent) than the younger generation. -Still, the findings show that young people and parents are -on the same page about politics in general. Majorities say -they tend to agree with one another on many key issues -and that their political discussions are generally amicable. - -© 2017 iStock/Nikada - -Three Things You Should Know -About the MTV/AP-NORC Poll of -Americans Age 15-34 and Parents -of 15-26 Year Olds: - -1) - -In the wake of recent protests and -demonstrations, 46 percent of young -people age 15-34 believe that people -like them can affect the government -at least a moderate amount, an -increase from 37 percent in the March -2018 poll. For 15-22 year olds, this rose -15 points, to 48 percent. - -2) Nearly two-thirds of young people -age 15-26 and parents of this age -group discuss politics at least once a -month. Majorities say these -conversations are civil and that they -are at least sometimes able to change -others’ political views when -disagreements do occur. - -3) More than two-thirds across both - -generations are disillusioned with the -two major political parties and feel a -third is needed. - -© Copyright 2018. The Associated Press and NORC - -June 2018 1 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Both young people age 15-26 and parents of this age group give the federal government a grade of F -more often than any other letter grade when assessing its performance on a wide variety of issues, -including gun control, the cost of higher education, foreign relations, immigration, environment and -climate change, health care, income inequality, taxes, sexual harassment and assault, abortion, racism, -and the treatment of citizens by police and the justice system. The economy and job creation fare -better, most commonly receiving grades of C from both groups. - -This survey is the second in a multiwave series of polls highlighting the voices of the youngest -generation of voters age 15-34, with this installment including a sample of parents of 15-26 year olds to -allow for comparisons between generations. This nationwide poll by MTV and The AP-NORC Center -surveyed a total of 939 young people, including 148 teens age 15-17, 410 adults age 18-26, and 381 adults -age 27-34, as well as 607 parents of young people age 15-26. - -AFTER RECENT DEMONSTRATIONS, YOUNG PEOPLE FEEL MORE -EMPOWERED IN POLITICS. - -Since the first MTV/AP-NORC Youth Political Pulse survey in March 2018, overall outlook among the -larger group of young people including those age 15-34 remains stable. Similar to the previous survey, -just under half are optimistic and a slight majority are pessimistic about the future of the United States. -Four in 10 have felt angry or anxious about the state of the country in the past month, while 1 in 10 -have felt excited or positive. - -Young peoples’ overall standpoint is fairly negative, but there is one area where young people have -grown somewhat more optimistic. After many recent protests and demonstrations, more now than in -March say elected officials care at least a moderate amount what people like them think (34 percent vs. -25 percent) and that people like them can affect the government at least a moderate amount (46 -percent vs. 37 percent). - -Since March 2018, more young people feel efficacious in American government. - -t -n -e -c -r -e -P - -100 - -90 - -80 - -70 + PUBLIC SPEAKING SKILLS -60 +1 + + 1.1 PUBLIC SPEAKING: DEFINITIONS + +UNIT-I + +Public speaking is the process or act of performing a speech to a live audience. This + +process can include preparation, organization, and the delivery of a message that is clear and + +engaging. It can range from a formal speech at a conference to an informal talk at a social + +gathering. + +Historical Perspective + +Public speaking dates back to ancient civilizations, notably the Greeks and Romans. + +In ancient Greece, public speaking was considered an essential skill for citizens. The great + +orator, Aristotle, even wrote a treatise on rhetoric. The Roman tradition also valued public + +oration, with figures like Cicero and Quintilian making significant contributions to the art of + +public speaking. + +1.2 COMPONENTS OF PUBLIC SPEAKING + + Speaker: The person delivering the speech, responsible for preparing and presenting + +the message. + + Message: The main content or idea being communicated. + + Audience: The group of people who receive and interpret the message. + + Channel: The medium through which the message is delivered, whether in person, + +via video, or another format. + + Feedback: The reactions and responses from the audience, which can be verbal or + +non-verbal. + + Context: The situation or environment in which the speech occurs, influencing how + +the message is received. + +2 + + 1.3 NEED FOR PUBLIC SPEAKING + +1.3.1. Personal Development + +Public speaking is crucial for personal growth. It helps build self-confidence, improves + +critical thinking, and enhances communication skills. + +a) Confidence Building + + Regular practice in public speaking situations can significantly increase self- + +confidence. Speaking in front of an audience, even a small one, helps individuals + +become more comfortable with expressing their ideas publicly. + +b) Critical Thinking + + Preparing a speech requires organizing thoughts and structuring them logically. This + +process helps develop critical thinking skills as one must evaluate information, + +formulate arguments, and anticipate audience questions. + +c) Communication Skills + + Public speaking refines both verbal and non-verbal communication skills. Effective + +speakers learn to articulate their thoughts clearly and use body language to reinforce + +their message. + +1.3.2. Professional Growth + +In the professional realm, public speaking skills are invaluable. They contribute to + +career advancement, networking, and leadership. + +a) Career Advancement + + Many career opportunities, especially in leadership roles, require the ability to speak + +confidently and effectively. Employers value employees who can present ideas clearly + +and lead discussions. + +b) Networking Opportunities: + + Public speaking events provide opportunities to meet and connect with peers, + +mentors, and industry leaders. These connections can lead to professional + +collaborations and career opportunities. + +c) Influence and Leadership: + + Effective public speakers can inspire and motivate others. This ability is crucial for + +leadership, as it involves guiding teams, making persuasive arguments, and driving + +organizational change. + +3 + + 1.3.3. Social Impact + +Public speaking plays a significant role in social advocacy, community engagement, + +and education. + +a) Advocacy and Activism: + + Public speakers can use their platform to advocate for social causes, influencing + +public opinion and policy. Activists and leaders often speak out to bring attention to + +important issues. + +b) Community Engagement + + Speaking at community events fosters a sense of belonging and encourages civic + +participation. Engaging with community members through speeches can promote + +unity and collective action. + +Education and Awareness: + + Educators and experts use public speaking to disseminate knowledge and raise + +awareness on various topics. This contributes to an informed and educated society. + +1.4 IMPORTANCE OF PUBLIC SPEAKING + +1.4.1 Effective Communication + +Public speaking is a powerful tool for effective communication, enabling clarity, + +engagement, and persuasion. + +a) Clarity and Precision + + Effective speakers convey their messages clearly and precisely, ensuring the audience + +understands their points. This reduces misunderstandings and enhances the + +effectiveness of communication. + +b) Audience Engagement + + Keeping the audience engaged is crucial for the success of a speech. Techniques like + +storytelling, humour, and interactive elements help maintain interest and involvement. + +c) Persuasion + + The ability to persuade is a key aspect of public speaking. Persuasive speeches aim to + +influence the audience’s beliefs, attitudes, or actions, which is essential in many + +professional and social contexts. + +1.4.2 Professional Relevance + +Public speaking is essential in professional settings, particularly for presentations, + +public relations, and crisis management. + +4 + + a) Presentations and Meetings: + + Many professional roles require presenting information to colleagues, clients, or + +stakeholders. Effective public speaking skills ensure these presentations are clear, + +engaging, and impactful. + +b) Public Relations: + + Managing a company’s or individual’s public image often involves public speaking. + +Delivering statements, handling media inquiries, and speaking at public events are all + +critical PR activities. + +c) Crisis Management + + During a crisis, clear and calm communication is vital. Public speakers must provide + +accurate information, reassure the public, and guide actions to manage the situation + +effectively. + +1.4.3 Societal Contributions + +Public speaking contributes to societal development through civic participation, education, + +and cultural preservation. + +a) Civic Participation + + Public speaking encourages active participation in democratic processes. By speaking + +at town halls, debates, and public forums, individuals can influence policy and + +contribute to community development. + +b) Education and Teaching + + Teachers and educators use public speaking to enhance learning experiences. + +Effective teaching involves clear communication, engaging presentations, and the + +ability to inspire students. + +c) Cultural Preservation + + Public speaking plays a role in preserving and promoting cultural heritage. + +Storytellers, historians, and cultural leaders use speeches to share traditions, history, + +and values. + +5 + + 1.5 ADVANTAGES OF PUBLIC SPEAKING + +1.5.1 Personal Benefits + +Public speaking offers numerous personal advantages, including increased self- + +esteem, networking opportunities, and improved problem-solving skills. + +a) Self-Esteem + + Successfully delivering a speech can boost self-esteem and provide a sense of + +accomplishment. Overcoming the fear of public speaking leads to greater self- + +confidence. + +b) Networking + + Speaking at events allows individuals to meet new people and expand their + +professional and personal networks. These connections can lead to collaborations and + +new opportunities. + +c) Problem-Solving + + Public speaking often involves addressing questions and concerns from the audience. + +This helps develop problem-solving skills as speakers learn to think on their feet and + +provide clear, thoughtful responses. + +1.5.2 Professional Benefits + +In the workplace, public speaking skills can enhance leadership, create career + +opportunities, and lead to professional recognition. + +a) Leadership + + Effective communication is a key leadership quality. Public speaking skills enable + +leaders to articulate their vision, motivate teams, and navigate challenges. + +b) Career Opportunities + + Many job roles, especially in management and leadership, require public speaking + +skills. Being a confident speaker can open doors to promotions and new career paths. + +c) Professional Recognition + + + +Individuals who excel in public speaking often gain recognition and respect in their + +field. This can lead to invitations to speak at conferences, awards, and other + +professional accolades. + +6 + + 1.5.3 Social Benefits + +Public speaking can foster community building, amplify advocacy efforts, and promote + +cultural exchange. + +a) Community Building + + Public speaking events bring people together, fostering a sense of community and + +shared purpose. Speakers can address community issues, inspire collective action, and + +promote social cohesion. + +b) Advocacy + + Speakers can use their platform to advocate for social justice, environmental + +protection, and other important causes. This helps raise awareness and mobilize + +support for these issues. + +c) Cultural Exchange + + Public speaking events can promote cultural exchange and understanding. Speakers + +from diverse backgrounds share their experiences and perspectives, enriching the + +audience’s knowledge and appreciation of different cultures. + +1. 6 SKILLS REQUIRED FOR PUBLIC SPEAKING + +1.6.1 Content Preparation + +Effective public speaking begins with thorough content preparation. This involves + +research, organization, and writing. + +a) Research + + Conducting comprehensive research is essential to gather accurate and relevant + +information. This ensures the speech is informative and credible. + +b) Organization + + Structuring the speech logically helps convey the message clearly. A well-organized + +speech typically includes an introduction, main body, and conclusion. + +c) Writing + + Crafting the speech with attention to language, tone, and style enhances its + +effectiveness. The writing should be clear, concise, and tailored to the audience. + +7 + + 1.6.2 Delivery Techniques + +The way a speech is delivered is just as important as its content. Key delivery + +techniques include vocal variety, body language, and eye contact. + +a) Vocal Variety + + Using different pitches, tones, and volumes can make the speech more engaging and + +emphasize key points. + +b) Body Language + + Gestures, facial expressions, and posture can reinforce the message and convey + +confidence. + +c) Eye Contact + + Maintaining eye contact with the audience helps build a connection and shows + +sincerity. + +1.6.3 Audience Engagement + +Engaging the audience is crucial for an effective speech. Techniques like storytelling, + +interactive elements, and adaptability can help. + +a) Storytelling + + Using stories and anecdotes can illustrate points and make the speech more relatable + +and memorable. + +b) Interactive Techniques + + + +Involving the audience through questions, discussions, and activities keeps them + +engaged and interested. + +c) Adaptability + + Being able to adjust the speech based on audience feedback and reactions is + +important. This may involve altering the delivery or content on the fly to better suit + +the audience. + +1.6.4 Overcoming Anxiety + +Public speaking anxiety is common but can be managed through preparation, practice, + +and mindfulness techniques. + +a) Preparation + + Thoroughly preparing the speech helps boost confidence and reduce anxiety. + +8 + + b) Practice + + Regularly practicing the speech, including in front of a mirror or a small audience, can + +make the speaker more comfortable and familiar with the content. + +c) Mindfulness Techniques + + Techniques like deep breathing, visualization, and mindfulness can help manage + +anxiety and maintain composure during the speech. + +1.6.5 Technical Proficiency + +Technical proficiency involves using visual aids, handling microphones, and + +managing time effectively. + +a) Use of Visual Aids + + + +Incorporating visual aids like slides, videos, and props can enhance the speech and + +make complex information more understandable. + +b) Microphone Handling + + Properly using microphones and other audio equipment ensures the audience can hear + +the speaker clearly. + +c) Time Management + + Managing time effectively is important to cover all points within the allotted time. + +This involves practicing the speech to ensure it fits within the time limit and being + +able to adjust on the fly if needed. + +1.7. CONCLUSION + +In conclusion, public speaking is a multifaceted skill that encompasses various + +aspects, from preparation to delivery. It is an essential tool for personal and professional + +development, offering numerous benefits and opportunities. By developing and refining + +public speaking skills, individuals can enhance their confidence, influence, and impact on + +society. + +9 + + 2.1 Elements of Public Speaking + +UNIT-II + +Public speaking is composed of several critical elements that contribute to the effectiveness + +and success of a speech. Understanding these elements is essential for mastering the art of + +public speaking. + +a) Speaker + +The speaker is the individual delivering the message. Their role is crucial as they are + +responsible for preparing and presenting the speech. A speaker's credibility, confidence, and + +delivery style greatly influence how the message is received. + +b) Message + +The message is the content or main idea conveyed in the speech. It includes the information, + +arguments, and emotions the speaker intends to communicate. Clarity, organization, and + +relevance of the message are key to effective public speaking. + +c) Audience + +The audience is the group of people who receive and interpret the message. Understanding + +the audience's demographics, interests, and expectations helps tailor the speech to engage and + +resonate with them. + +d) Channel + +The channel is the medium through which the message is delivered. This can be face-to-face + +communication, a video presentation, a webinar, or other formats. The choice of channel can + +affect how the message is perceived and received. + +e) Feedback + +Feedback consists of the responses and reactions from the audience, which can be verbal or + +non-verbal. Effective public speakers pay attention to feedback to adjust their delivery and + +ensure their message is being understood. + +f) Context + +Context refers to the environment and circumstances in which the speech is delivered. This + +includes the physical setting, the occasion, and the cultural background. Context can + +influence the tone and content of the speech. + +g) Purpose + +The purpose of the speech is the speaker's goal, whether it is to inform, persuade, entertain, or + +fulfill a specific occasion. Clearly defining the purpose helps in crafting a focused and + +effective message. + +10 + + h) Encoding and Decoding + + Encoding: The process of the speaker translating thoughts and ideas into words and + +gestures. + + Decoding: The audience's interpretation and understanding of the speaker's message. + +Effective communication occurs when the audience accurately decodes the speaker's + +message. + +2.2 TYPES OF PUBLIC SPEAKING + +Public speaking can be categorized into different types based on the purpose and context of + +the speech. Here are the main types: + +a. Ceremonial Public Speaking + +Ceremonial public speaking, also known as special occasion speaking, involves delivering + +speeches at formal events and ceremonies. These speeches are often emotive and personal, + +designed to celebrate, honor, or commemorate. + +Examples: + + Wedding Speeches: Best man or maid of honor speeches, father of the bride speech. + + Funeral Eulogies: Tributes to the deceased, reflecting on their life and legacy. + + Award Acceptance Speeches: Recipients expressing gratitude and acknowledging + +contributions. + +Key Elements: + + Emotional Connection: Establishing a heartfelt connection with the audience. + + Appropriate Tone: Matching the tone to the occasion, whether joyous, solemn, or + +celebratory. + + Brevity: Keeping the speech concise and to the point, respecting the event's flow. + +b. Demonstrative Public Speaking + +Demonstrative public speaking involves showing the audience how to do something. These + +speeches are educational and often include visual aids or live demonstrations to enhance + +understanding. + +Examples: + + Cooking Demonstrations: Showcasing how to prepare a specific dish. + + Technical Tutorials: Explaining how to use a software program or gadget. + + Craft Workshops: Demonstrating steps to create a craft project. + +11 + + Key Elements: + + Clarity: Clearly explaining each step in a logical order. + + Visual Aids: Using tools, props, or multimedia to illustrate points. + + Engagement: Encouraging audience participation or questions to ensure + +comprehension. + +c. Informative Public Speaking + +Informative public speaking aims to educate the audience about a particular topic. The goal is + +to increase understanding and awareness without persuading the audience to adopt a specific + +viewpoint. + +Examples: + + Lectures: Academic presentations on a specific subject. + + Seminars: Professional development sessions on industry-related topics. + + Briefings: Providing updates or reports on specific issues or projects. + +Key Elements: + + Research: Conducting thorough research to provide accurate and detailed + +information. + + Organization: Structuring the speech with a clear introduction, body, and conclusion. + + Neutrality: Presenting facts objectively without expressing personal opinions. + +d. Persuasive Public Speaking + +Persuasive public speaking aims to convince the audience to accept a particular viewpoint or + +take specific actions. This type of speech involves presenting arguments, evidence, and + +emotional appeals to influence the audience. + +Examples: + + Political Speeches: Campaign speeches advocating for policies or candidates. + + Sales Presentations: Pitching products or services to potential customers. + + Advocacy Speeches: Promoting social causes or public policies. + +Key Elements: + + Thesis Statement: Clearly stating the main argument or position. + + Evidence: Providing credible evidence and examples to support the argument. + + Emotional Appeals: Connecting with the audience's emotions to strengthen the + +persuasive impact. + + Call to Action: Encouraging the audience to take specific actions based on the + +speech. + +12 + + 2.3 CONCLUSION + +Public speaking encompasses various elements and types, each with unique + +characteristics and requirements. Understanding the fundamental elements—such as the + +speaker, message, audience, and context—forms the foundation for effective communication. -50 +Additionally, recognizing the different types of public speaking—ceremonial, demonstrative, -40 +informative, and persuasive—enables speakers to tailor their approach to suit the purpose and -30 +context of their speech. By mastering these aspects, individuals can enhance their public -20 +speaking skills and achieve their communication goals. -10 +13 -0 + UNIT-III -37 +3.1 TECHNIQUES FOR EFFECTIVE PUBLIC SPEAKING -25 +Effective public speaking is a vital skill in both personal and professional contexts. -46 +Mastering this art involves various techniques that enhance clarity, engagement, and impact. -34 +This material covers essential techniques to help you become a more effective and confident -March 2018 +public speaker. -May 2018 +3.1.1 Understanding Your Audience -Elected officials care at least a moderate amount +Audience Analysis -People like me can affect government at least a moderate amount + Demographics: Consider age, gender, education level, cultural background, and -Questions: How much do elected public officials care what people like you think? -How much can people like you affect what the government does? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +profession. Understanding these factors helps tailor your message to resonate with the -Young people age 15-22 in particular are increasingly optimistic when it comes to political efficacy. In -March, just a third said that people like them can affect what the government does at least a moderate -amount. Now, 48 percent of those age 15-22 say the same. +audience. -© Copyright 2018. The Associated Press and NORC + -June 2018 2 +Interests: Identify what the audience cares about and tailor your content to match - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +their interests. This helps in making the content relevant and engaging. -Young peoples’ priorities have also shifted. When asked to describe in their own words what issue -facing the country concerns them the most, 21 percent cited gun laws in March. Just 6 percent mention -gun issues now.1 The most common worries are now the economy (16 percent) and social inequality (14 -percent). + Knowledge Level: Gauge their familiarity with the topic to avoid oversimplification -BOTH YOUNG PEOPLE AND PARENTS SAY POLITICS HAVE BECOME -DYSFUNCTIONAL, AND BOTH ARE DISSATISFIED WITH THE TWO-PARTY -SYSTEM. +or unnecessary complexity. Knowing their knowledge level ensures that you neither -Young people age 15-26 and parents of young people age 15-26 alike have mixed outlooks on the future -of the United States. Young people are more pessimistic than parents (55 percent vs. 40 percent), -although these differences are mainly attributable to the parents being more Republican and less -racially and ethnically diverse than young people. +underestimate nor overestimate their understanding. -Both generations report relatively negative feelings about the state of the country over the past month. -Substantial minorities of both groups have felt anxious or angry in the past month, while few have felt -positive or excited. Young people are more likely than parents to say they feel anxious (43 percent vs. -29 percent) and less likely to feel positive (8 percent vs. 17 percent) or excited (8 percent vs. 15 percent). + Expectations: Understand what the audience expects to gain from your speech. -Young people age 15-26 are more anxious and less positive or excited about the state of America than -are parents of young people. +Meeting or exceeding these expectations can significantly enhance the impact of your -y +presentation. -l -e -m -e -r -t -x -e +Techniques -r -o + Surveys and Questionnaires: Collect information about the audience before the -y -r -e -v -y -a -s +event. This can include questions about their interests, preferences, and knowledge -o -h -w +level. -t -n -e -c -r -e -P + Observation: Pay attention to the audience's reactions and body language. This real- -100 +time feedback can help you adjust your delivery. -90 + Direct Interaction: Engage with the audience through questions and discussions -80 +before the speech. This builds rapport and provides valuable insights into their -70 +expectations. -60 +Practical Tips -50 + Adapt Your Language: Use language that is appropriate for the audience’s level of -40 +understanding. Avoid jargon or technical terms unless the audience is familiar with -30 +them. -20 +14 -10 +  Relate to the Audience: Share stories or examples that are relevant to the audience's -0 +experiences. This makes the content more relatable and engaging. -43 +3.1.2 Preparation and Organization -29 +Research -40 + Thorough Investigation: Gather accurate and up-to-date information on your topic. -36 +This ensures the credibility and relevance of your content. -17 + Credible Sources: Use reliable and reputable sources for your content. This adds -8 +authority to your speech and builds trust with the audience. + + Diverse Perspectives: Incorporate multiple viewpoints to enrich your speech. This + +demonstrates a well-rounded understanding of the topic and engages a broader + +audience. + +Structuring Your Speech + + + +Introduction: Capture attention with a strong opening statement, quote, or story. The + +introduction sets the tone and provides a roadmap for your speech. + + Body: Organize main points logically, using clear and concise language. Each point + +should build on the previous one, leading the audience through a coherent argument. + + Conclusion: Summarize key points and end with a powerful closing statement. The + +conclusion should reinforce your message and leave a lasting impression. + +Outlining + + Simple Outline: Main headings and subheadings. This provides a basic structure for + +your speech. + + Detailed Outline: Includes full sentences and transitional phrases. This helps in + +fleshing out your ideas and ensuring smooth transitions between points. + +Practical Tips + + Time Management: Allocate time for each section of your speech. Practice + +delivering within this timeframe to avoid running over. + + Rehearse: Practice delivering your speech multiple times. This helps in internalizing + +the content and improving delivery. + +3.1.3 Developing a Strong Opening + +Techniques + + Anecdotes: Start with a relevant story that sets the stage for your message. Anecdotes + +can be personal or about someone else, but they should be relatable to the audience. 15 -8 +  Questions: Pose a thought-provoking question to engage the audience. This -Anxious +encourages active thinking and participation. -Angry + Quotes: Use a powerful quote that relates to your topic. A well-chosen quote can set -Positive +the tone and add credibility. -Excited + Statistics: Present a surprising statistic to grab attention. Statistics can highlight the -Young people age 15-26 +importance of your topic and provoke interest. -Parents of young people age 15-26 +Importance -Question: In the past month… How _____ have you felt about the state of the country? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + First Impressions: Establish credibility and set the tone for your speech. A strong -Young people and parents have similar assessments of the political system and their own efficacy -within it. Majorities of both groups say American politics have been dysfunctional over the past -month. Sixty-four percent of young people and 67 percent of parents feel that politics have not been -functioning very well or at all. +opening makes a positive first impression. -Young people see the current two-party system as inadequate, and parents agree. Less than a third -think the Republican and Democratic parties do an adequate job of representing the American people, -while more than two-thirds say a third party is needed. + Audience Engagement: Captivate the audience from the beginning. An engaging -1 Interviews for this survey were conducted April 23-May 9, 2018, prior to the May 18 shooting in Santa Fe, Texas. +opening keeps the audience interested and attentive. -© Copyright 2018. The Associated Press and NORC +Practical Tips -June 2018 3 + Relevance: Ensure your opening is relevant to the main topic. A disconnected - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +opening can confuse the audience. -YOUNG PEOPLE AND PARENTS ARE SIMILARLY ENGAGED WITH THE -MIDTERM ELECTIONS. + Brevity: Keep the opening concise. An overly long opening can lose the audience's -When it comes to the upcoming midterm elections in November, parents of young people age 15-26 are -more likely than young people age 15-26 to be regularly reading or watching news about the elections -(66 percent vs. 45 percent). However, older adults tend to consume news more often than young -people in general2 and may be hearing more about the midterms as a result. +attention. -The two generations are equally likely to engage with the November elections in other ways, such as -posting or commenting about them on social media (25 percent of parents vs. 25 percent of young -people age 15-26), sharing political memes (18 percent of parents vs. 24 percent of young people age 15- -26), participating in events (14 percent of parents vs. 16 percent of young people age 15-26), or -volunteering with a candidate or issue campaign (11 percent of parents vs. 9 percent of young people -age 15-26). +3.1.4 Effective Body Language -BOTH GENERATIONS HAVE VARIED POLITICAL CONCERNS AND ARE -STRONGLY DISAPPOINTED WITH THE GOVERNMENT’S PERFORMANCE -ON KEY ISSUES. +Components -Economic issues top long lists of current concerns for both young people and parents. When asked to -say in their own words what concerns them most about the issues currently facing the country, 15 -percent of young people age 15-26 and 17 percent of parents of young people age 15-26 say they are -most concerned about the economy. The two groups place similar priorities on a range of other issues, -although social inequality ranks second highest for young people (13 percent) and fails to make the top -five most worrisome issues for parents (6 percent). + Gestures: Use hand movements to emphasize points and convey enthusiasm. -2 The Media Insight Project. 2014. The Personal News Cycle. -http://www.apnorc.org/PDFs/Media%20Insight/News%20Cycle/The_Personal_News_Cycle_Final.pdf +Gestures should be natural and aligned with your words. -© Copyright 2018. The Associated Press and NORC + Facial Expressions: Match expressions with the content to convey emotions. -June 2018 4 +Smiling, frowning, or showing surprise can reinforce your message. - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + Posture: Stand confidently, avoid slouching or leaning on the podium. Good posture -Young people and parents report a wide range of top concerns of issues facing the country. +projects confidence and authority. -The economy + Eye Contact: Maintain eye contact to build a connection with the audience. This -15 +makes your speech feel more personal and engaging. -17 +Techniques -Social inequality + Practice in Front of a Mirror: Observe and adjust your body language. This helps in -War or threat of nuclear war +becoming aware of and improving your gestures and expressions. -13 + Record and Review: Record your practice sessions to identify areas for -6 +improvement. Watching yourself can reveal habits you might not be aware of. -11 -10 + Feedback: Seek feedback from peers or mentors. Constructive criticism can provide -Immigration +valuable insights for improvement. -Gun laws +16 -Other issues + Practical Tips -Political climate + Avoid Distracting Movements: Be mindful of habits like fidgeting or pacing. These -Concerns about President Trump, leadership +can distract the audience from your message. -Foreign policy + Consistent Gestures: Use gestures consistently to avoid confusion. Random or -The environment +inconsistent gestures can be distracting. -Health care +3.1.5 Vocal Variety -Terrorism, homeland security, national security +Elements -Crime and safety + Pitch: Vary your pitch to avoid monotony and maintain interest. A monotone voice -Education +can be dull and disengaging. -Nothing + Volume: Adjust your volume to suit the venue and emphasize key points. Speaking -Everything +too softly can make it hard for the audience to hear, while speaking too loudly can be -7 -7 +overwhelming. -7 + Rate: Vary your speaking rate to highlight important information and maintain -4 +engagement. Speaking too fast can confuse the audience, while speaking too slowly -6 +can bore them. -9 + Pauses: Use pauses effectively to let key points sink in and to create dramatic effect. -5 +Pauses can also give you a moment to collect your thoughts. -8 +Techniques -5 + Breathing Exercises: Practice deep breathing to control your voice. Proper breathing -7 +supports vocal strength and clarity. -5 -6 + Voice Modulation Exercises: Work on pitch, tone, and volume variations. Exercises -4 -5 +like humming, lip trills, and tongue twisters can improve your vocal range and -3 +flexibility. -6 + Reading Aloud: Practice reading aloud to improve vocal variety and articulation. -2 +This helps in developing a more dynamic and engaging speaking style. -4 +Practical Tips -2 -3 + Record Yourself: Record your practice sessions to analyze your vocal variety. -2 -2 +Listening to yourself can help you identify areas for improvement. -1 -0 + Seek Feedback: Get feedback from others on your vocal delivery. Constructive -0 -1 +criticism can help you refine your vocal technique. -0 +3.1.6 Engaging the Audience -10 +17 -20 + Techniques -30 + -40 +Interactive Elements: Use polls, Q&A sessions, and discussions. Interaction keeps -50 +the audience involved and attentive. -60 + Stories and Examples: Share relevant anecdotes and case studies to illustrate points. -70 +Stories make abstract concepts more relatable and memorable. -80 + Humor: Incorporate appropriate humor to make the speech enjoyable and relatable. -90 +Humor can lighten the mood and make your speech more engaging. -100 + Visual Aids: Use slides, charts, and videos to support your message and keep the -Percent who offered each response +audience engaged. Visual aids can clarify complex information and maintain interest. -Young people age 15-26 +Importance -Parents of young people age 15-26 + Attention Retention: Keeps the audience focused and interested. Engaged audiences -Question: Of all of the issues facing the country right now, what concerns you the most? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +are more likely to absorb and retain information. -Young people age 15-26 believe the federal government is doing a poor job addressing a majority of key -issues. They most commonly give the government a failing grade for 15 of 17 key issues. However, -young people rate performance on the economy and job creation as average, most commonly giving a -grade of C. + Enhanced Understanding: Helps the audience grasp complex concepts. Interaction -© Copyright 2018. The Associated Press and NORC +and visual aids can make your message clearer and more accessible. -June 2018 5 +Practical Tips - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + Relevance: Ensure that your engagement techniques are relevant to your content. -Parents most commonly give the same grades as young people age 15-26 on most issues, with a few -exceptions. They give Cs rather than Fs for LGBT rights, the quality of education, terrorism and -homeland security, and taxes. +Irrelevant interaction can be distracting. -Young people age 15-26 and parents give low marks to the federal government on its handling of many -issues. + Balance: Balance interactive elements with your main content. Too much interaction -Young People -Age 15-26 +can detract from your core message. -Parents of Young -People Age 15-26 +3.17 Overcoming Anxiety -The environment and climate change +Techniques -Gun control + Preparation: The more prepared you are, the more confident you will feel. Thorough -Racism +preparation reduces uncertainty and boosts confidence. -Immigration + Practice: Regular practice reduces anxiety and increases familiarity with the content. -Cost of higher education +The more you practice, the more comfortable you become. -Treatment of citizens by police and justice system + Visualization: Visualize a successful speech to build confidence. Imagine yourself -Abortion +delivering the speech successfully and receiving positive feedback. -Income inequality + Relaxation Techniques: Use deep breathing, meditation, and positive affirmations. -Sexual harassment and assault +These techniques can help calm your mind and reduce physical symptoms of anxiety. -The quality of education +Managing Nervous Energy -Taxes + Physical Activity: Light exercise before speaking can reduce tension. Activities like -Foreign relations +stretching or a short walk can help calm nerves. -Health care policy +18 -The rights of LGBT Americans +  Hydration: Drink water to stay hydrated and prevent a dry mouth. Avoid caffeine, -Terrorism and homeland security +which can increase anxiety. -The economy + Positive Thinking: Focus on positive outcomes and remind yourself of past -Job creation +successes. Positive affirmations can boost your confidence. -A +Practical Tips -B + Arrive Early: Arrive early to familiarize yourself with the venue. This reduces -C +uncertainty and helps you feel more comfortable. -D + Connect with the Audience: Engage with audience members before the speech. -F A +Building rapport can make you feel more at ease. -B +3.1.8 Effective Use of Visual Aids -C +Types of Visual Aids -D + Slides: PowerPoint or Keynote presentations. Slides should be clear, concise, and - +visually appealing. - + Charts and Graphs: Illustrate data and trends. Charts and graphs can make complex - +data more understandable. - + Videos: Short clips to support or highlight points. Videos can add variety and - +illustrate points more vividly. - + Props: Physical objects that enhance understanding. Props can make abstract - +concepts more concrete and engaging. - +Techniques - + Simplicity: Keep slides and visuals simple and uncluttered. Avoid overwhelming the - +audience with too much information. - + Relevance: Ensure visual aids are directly related to your content. Irrelevant visuals - +can distract and confuse the audience. - + Practice: Familiarize yourself with the visual aids and integrate them smoothly into - +your speech. Practicing with your visual aids ensures a seamless presentation. - +Common Pitfalls - + Overreliance: Don't let visual aids overshadow your speech. Visual aids should - +support, not replace, your message. - + Technical Issues: Be prepared for potential technical difficulties. Have a backup plan - +in case of technical failures. - +19 - + Practical Tips - + Backup Copies: Have backup copies of your visual aids. This ensures you can - +continue your presentation even if technical issues arise. -F + Engage with Visuals: Interact with your visual aids to maintain audience - +engagement. Pointing out specific details or elaborating on visuals can enhance - +understanding. - +3.1.9 Handling Questions and Interactions - +Techniques - + Anticipate Questions: Predict potential questions and prepare answers. This helps - +you feel more confident and prepared. - + Active Listening: Listen carefully to questions and respond thoughtfully. Active - +listening shows respect and helps you provide accurate answers. - + Clarification: Ask for clarification if a question is unclear. This ensures you - +understand the question and can provide a relevant response. - + Confidence: Maintain composure and confidence when answering questions. - Indicates the most commonly given grade, or mode. The frequency of each grade is indicated with darker shading for -more common responses, according to the key below. +Confidence reassures the audience of your knowledge and credibility. -0% +Strategies -50% + Repetition: Restate the question to ensure understanding and give yourself time to -Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of -the following issues? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +think. Repeating the question also ensures everyone in the audience heard it. -PARENT/CHILD DISCUSSIONS ABOUT POLITICS ARE USUALLY CORDIAL, -BUT IT CAN BE DIFFICULT TO SWAY THE OTHER GENERATION WHEN -VIEWS DIFFER. + Bridging: Connect your answer to your main points or message. This keeps your -Young people age 15-26 and their parents have fairly regular discussions with one another about -politics and social issues, and few see these discussions as a source of conflict. Across the two groups, -64 percent of young people and parents say they discuss politics with each other at least once a -month, 20 percent say they have these discussions less often, and 16 percent say they never talk about -politics or social issues together. +response relevant and reinforces your key messages. -Among those who discuss political and social issues with their parents or children, a majority say they -usually see eye to eye in these conversations, and about a third say they debate diplomatically. Fewer -say the discussion turns into a major conflict or requires changing the subject to avoid arguments. + Acknowledgment: Acknowledge challenging questions and provide honest, -© Copyright 2018. The Associated Press and NORC +thoughtful responses. Honesty builds trust and credibility. -June 2018 6 +Practical Tips - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + Stay Calm: Stay calm and composed, even if faced with difficult questions. Keeping -Most parents and their children see eye to eye on politics and social issues. +your cool shows professionalism. -Someone changes the -subject quickly to avoid -getting into disagreements. -9% + Admit When You Don’t Know: It’s okay to admit when you don’t know an answer. -It turns into -World War III. -5% +Offer to follow up later with more information. -We debate things -diplomatically. -32% +3.1.10 Continuous Improvement -We usually see -eye to eye. -53% +Techniques -Question: When you talk with [him/her/them] about political and social issues, how would you describe how the discussion -goes? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + Self-Reflection: After each speech, reflect on what went well and what could be -When the two generations were asked about political discussions with a particular family member, -parents’ accounts are similar regardless of the specific relationship, such as mother-daughter or father- -son. Sons reporting on discussions with fathers are more likely than other pairings to say they talk -about politics at least once a week, but least likely to say these discussions tend to go smoothly. Only -42 percent of sons say they usually see eye to eye with their fathers, while 49 percent of sons say the -same about their mothers, 54 percent of daughters say so about their fathers, and 62 percent say so -about their mothers. +improved. Self-reflection helps you identify strengths and areas for growth. -Those who identify as political independents (36 percent) are less likely than Democrats (60 percent) -or Republicans (53 percent) to say they usually see eye to eye in parent/child discussions about -political and social issues. +20 -When asked about potential differences in opinion on specific issues, both generations report high -levels of agreement. In their discussions, at least 6 in 10 say there is agreement when discussing each -of 18 topics. Agreement levels are highest on Trump’s job performance and on the issues of racism, -sexual harassment and assault, and health care policy, and lowest on LGBT rights, abortion, and -foreign relations. +  Feedback: Seek feedback from peers, mentors, and the audience. Constructive -While agreement is high, there also seems to be a fair amount of ambiguity about family members’ -political positions. On 13 of the 18 issues asked about, at least 15 percent say they are unsure if they -tend to agree or disagree. +feedback provides valuable insights for improvement. -© Copyright 2018. The Associated Press and NORC + Training and Courses: Enroll in public speaking courses or workshops. Formal -June 2018 7 +training can provide new techniques and enhance your skills. - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + Practice: Regularly practice public speaking in various settings. Practice helps you -Young people and their parents tend to be on the same page with many political and social issues. +refine your skills and gain confidence. -President Trump's job performance +Setting Goals -Racism + Specific Goals: Set specific, measurable goals for improvement. Clear goals provide -Sexual harassment and assault +direction and motivation. -Health care policy + Track Progress: Keep a record of your speeches and progress. Tracking progress -The quality of education +helps you see improvements and stay motivated. -Treatment of citizens by police and justice system + Celebrate Achievements: Recognize and celebrate milestones and improvements. -Cost of higher education +Celebrating successes reinforces positive behavior and boosts confidence. -The economy +Practical Tips -Taxes + Join Speaking Groups: Join groups like Toastmasters to practice and receive -The environment and climate change +feedback. Regular practice and feedback from peers can significantly improve your -Income inequality +skills. -Immigration + Record Your Speeches: Record your speeches to analyze and improve your delivery. -Job creation +Watching recordings helps you identify areas for improvement. -Terrorism and homeland security +3.2 Conclusion -Gun control +Effective public speaking involves mastering various techniques that enhance clarity, -The rights of LGBT Americans +engagement, and impact. By understanding your audience, preparing thoroughly, using strong -Foreign relations +openings, employing effective body language and vocal variety, engaging the audience, -Abortion +overcoming anxiety, utilizing visual aids effectively, handling questions with confidence, and -76 +committing to continuous improvement, you can become a more effective and confident -76 +public speaker. -76 +21 -75 + 4.1 METHODS OF PUBLIC SPEAKING -74 +UNIT-IV -74 +Public speaking is a critical skill that can be used in various contexts, ranging from -73 +casual conversations to formal presentations. Different methods of public speaking are suited -72 +for different scenarios, and mastering these methods can significantly enhance your ability to -71 +communicate effectively. This material covers key methods of public speaking, providing -71 +detailed insights and practical tips for each. -70 +4.1.1 Impromptu -70 +Definition -69 +Impromptu speaking involves delivering a speech without prior preparation. This method is -69 +often required in spontaneous situations where quick thinking and adaptability are essential. -69 +Techniques -62 + Think Quickly: Organize your thoughts rapidly and structure them logically. -62 + Stay Calm: Maintain composure to think clearly and speak confidently. -61 + Use Personal Experiences: Draw from your own experiences and knowledge to -13 +support your points. -8 + Practice Regularly: Engage in impromptu speaking exercises to improve your skills. -9 +Advantages -10 + Flexibility: Allows you to adapt to the topic and audience on the spot. -9 + Authenticity: Speeches often sound more genuine and spontaneous. -8 +Disadvantages -13 + Lack of Structure: Can lead to disorganized thoughts and unclear messages. -15 + -13 +Increased Anxiety: Higher pressure due to lack of preparation. -9 +Practical Tips -8 + Practice Thinking on Your Feet: Engage in activities like debate clubs or -11 +improvisational theater. -9 + Stay Updated: Keep yourself informed about various topics to have a broad -11 +knowledge base to draw from. -17 +4.1.2 Manuscript Speaking -19 +22 -13 + Definition -16 +Manuscript speaking involves reading a fully written speech. This method is commonly used -8 +in formal settings where accuracy and precise wording are crucial, such as political speeches -9 +or legal presentations. -14 +Techniques -17 + Write Clearly: Ensure the manuscript is well-written and easy to read. -15 + Practice Reading: Rehearse reading the manuscript to maintain eye contact and -13 +natural delivery. -19 + Use Visual Aids: Supplement the manuscript with visual aids to enhance engagement. -19 + Control Your Pace: Read at a moderate pace to ensure clarity and comprehension. -19 +Advantages -20 + Precision: Ensures accurate delivery of complex information. + + Consistency: Provides a consistent message every time. + +Disadvantages + + Less Engagement: Can sound monotonous and disengage the audience. + + Limited Flexibility: Difficult to adapt to audience reactions or questions. + +Practical Tips + + Highlight Key Points: Use highlights or underlines to emphasize important sections + +in your manuscript. + + Practice with a Timer: Ensure your speech fits within the allotted time by practicing + +with a timer. + +4.1.3 Memorized Speaking + +Definition + +Memorized speaking involves delivering a speech from memory without any notes or + +manuscripts. This method is often used for speeches that require exact wording, such as toasts + +or keynote addresses. + +Techniques + + Break It Down: Divide the speech into smaller sections for easier memorization. + + Use Mnemonics: Employ mnemonic devices to aid memory. + + Rehearse Frequently: Practice regularly to reinforce memory. + + Visualize Success: Visualize delivering the speech confidently to build memory and + +reduce anxiety. 23 -17 + Advantages -22 + Engagement: Allows for direct eye contact and connection with the audience. -20 + Freedom of Movement: Enables more natural gestures and movement. -14 +Disadvantages -19 + Risk of Forgetting: High risk of forgetting parts of the speech. + + + +Inflexibility: Difficult to adapt on the fly if something goes wrong. + +Practical Tips + + Daily Practice: Set aside time each day to rehearse your speech. + + Use Visual Cues: Associate parts of your speech with specific visual cues to aid + +memory. + +4.1.4 Extemporaneous Speaking + +Definition + +Extemporaneous speaking involves delivering a speech from brief notes or an outline. This + +method balances preparation and spontaneity, allowing for a structured yet natural delivery. + +Techniques + + Prepare Thoroughly: Create a detailed outline with key points. + + Practice Extensively: Rehearse using the outline to become familiar with the content. + + Stay Flexible: Be ready to adapt to audience reactions and questions. + + Use Conversational Tone: Aim for a conversational tone to maintain natural delivery. + +Advantages + + Balance: Combines the benefits of preparation and spontaneity. + + Engagement: Allows for eye contact and audience interaction. + +Disadvantages + + Requires Skill: Demands good organization and speaking skills. + + Potential for Gaps: Risk of missing key points if not well-prepared. + +Practical Tips + + Note Cards: Use small note cards with key points to guide you. + + Engage with the Audience: Incorporate audience feedback and questions into your + +speech. + +4.1.5 Persuasive Speaking 24 -22 + Definition -Tend to agree +Persuasive speaking aims to convince the audience to accept a particular viewpoint or take a -Tend to disagree +specific action. This method is commonly used in sales pitches, political speeches, and -Not sure +advocacy campaigns. -Questions: When talking about the following issues, do you tend to agree or disagree with [him/her/them]? -When talking with [your family member] about Donald Trump and how he’s handling his job as president, do you tend to -agree or disagree with [him/her/them]? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +Techniques -The two generations report different issues as the biggest sources of disagreement. Among those who -say they generally don’t see eye to eye when talking about political and social issues, young people -age 15-26 say they tend to disagree with their parents most about racism (37 percent), LGBT rights (36 -percent), and how Trump is handling his presidency (33 percent). Parents of young people age 15-26 -report that the largest sources of disagreement are gun control (27 percent), LGBT rights (25 percent), -and how Trump is handling his presidency (21 percent). + Clear Argument: Present a clear and logical argument. -When there are disagreements, few young people and parents are able to persuade one another to -change their views, and the findings are consistent between the generations. Eleven percent overall -say they are always or often able to persuade their family member to change their views. Fifty-three -percent acknowledge that they are sometimes able to persuade their family member. Just 6 percent -say the reverse is true, that they are often or always convinced to change their views by their family -member. Another 44 percent say their family member is sometimes able to persuade them to change -their own views. + Evidence and Examples: Use evidence and examples to support your points. -© Copyright 2018. The Associated Press and NORC + Emotional Appeal: Appeal to the audience's emotions to strengthen your argument. -June 2018 8 + Call to Action: End with a clear call to action, urging the audience to act on your - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +message. -Few young people or parents find it easy to persuade one another to change their political and social -views. +Advantages -How often you are able to persuade your - family member to change their views + -11 +Influence: Can effectively change opinions or behaviors. -53 + Engagement: Often highly engaging and impactful. -36 +Disadvantages -How often your family member is able - to persuade you to change your views + Resistance: Audience may resist if they feel manipulated. -6 + Complexity: Requires strong argumentative and rhetorical skills. -44 +Practical Tips -49 + Understand Your Audience: Tailor your message to the audience's values and -0 +beliefs. -10 + Practice Persuasion: Engage in debates and discussions to hone your persuasive -20 +skills. -30 +4.1.6 Informative Speaking -40 +Definition -50 +Informative speaking aims to educate the audience about a specific topic. This method is -60 +commonly used in lectures, presentations, and training sessions. -70 +Techniques -80 + Clear Structure: Organize your speech with a clear introduction, body, and -90 +conclusion. -100 + Use Visual Aids: Enhance understanding with visual aids like slides, charts, and -Percent +videos. -Always or often + Simplify Complex Information: Break down complex information into easily -Sometimes +understandable parts. -Rarely or never +25 -Questions: In general, when you disagree with [him/her/that person] about political or social issues, how often would you -say that you’re able to persuade [him/her/them] to change [his/her/their views]? -In general, when you disagree with [him/her/that person] about political or social issues, how often would you say that [he -is/she is/they are] able to change your views? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +  Engage the Audience: Use questions and interactive elements to maintain -YOUNG PEOPLE ARE PERCEIVED AS MORE IDEALISTIC THAN THE OLDER -GENERATION, AND PARENTS FEEL THEIR CHILDREN’S GENERATION IS -LESS INFORMED THAN THEIR OWN. +engagement. -When asked to compare the generations, young people age 15-26 and parents of this age group tend to -agree on who is the more patriotic and rigid generation, but there is some disagreement about who is -more informed. +Advantages -About two-thirds of both young people and parents say the older generation is more patriotic. More -than half of each group feel the older generation is more rigid. + Educational Value: Provides valuable information to the audience. -Among both groups, the largest proportion view the parents’ generation as more politically active and -moral than the younger generation, and the younger generation as more idealistic. + Clarity: Focuses on clear and concise communication. -Young people are divided on whether their generation is more uninformed, their parents’ generation is -more uninformed, or whether the two are about equal. Parents believe the younger generation is more -uninformed than their own. +Disadvantages -© Copyright 2018. The Associated Press and NORC + Potential for Boredom: Can be dry and unengaging if not delivered well. -June 2018 9 + Overload: Risk of overwhelming the audience with too much information. - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +Practical Tips -Young people and parents agree that the older generation is more patriotic and more rigid. + Focus on Key Points: Highlight the most important information. -Which generation is more…? + Check for Understanding: Ask questions to ensure the audience understands the -Patriotic +material. -Moral +4.1.7 Demonstrative Speaking -Rigid +Definition -Young people -say… +Demonstrative speaking involves showing the audience how to do something. This method is -Parents say… +commonly used in tutorials, workshops, and cooking shows. -Young people -say… +Techniques -Parents say… + Step-by-Step Instructions: Provide clear and detailed instructions. -Young people -say… + Use Props and Visual Aids: Demonstrate each step using props and visual aids. -Parents say… + Engage the Audience: Invite audience participation to enhance understanding. -Compassionate + Rehearse the Demonstration: Practice the demonstration to ensure smooth -Young people -say… +execution. -Parents say… +Advantages -Politically active + Practical Knowledge: Provides hands-on learning experiences. -Young people -say… + Engagement: Interactive and engaging for the audience. -Idealistic +Disadvantages -Uninformed + Preparation Time: Requires significant preparation and practice. -Parents say… + Potential for Mistakes: Live demonstrations can go wrong. -Young people -say… +Practical Tips -Parents say… + Prepare Thoroughly: Ensure all materials and equipment are ready. -Young people -say… + Practice Multiple Times: Rehearse the demonstration several times to avoid -Parents say… +mistakes. -0 +26 -0 + 4.1.8 Ceremonial Speaking -0 +Definition -0 +Ceremonial speaking involves delivering speeches for special occasions, such as weddings, -0 +funerals, graduations, and award ceremonies. -0 +Techniques -0 + Understand the Occasion: Tailor your speech to the specific occasion and audience. -0 + Use Appropriate Tone: Match the tone of your speech to the occasion, whether it's -0 +celebratory, solemn, or formal. -0 + Personal Touch: Include personal stories or anecdotes to make the speech more -0 +meaningful. -0 + Express Emotions: Convey appropriate emotions to connect with the audience. -0 +Advantages -0 + Memorability: Creates lasting memories for special occasions. -64 + Emotional Impact: Can be highly emotional and impactful. -10 +Disadvantages -25 + Pressure: High expectations for memorable and impactful speeches. -20 + Emotional Difficulty: Can be challenging to deliver in emotionally charged -40 +situations. -60 +Practical Tips -80 + Practice Emotion Management: Practice delivering your speech to manage -100 +emotions. -65 + Get Feedback: Seek feedback from close friends or family to refine your speech. -7 +4.1.9 Motivational Speaking + +Definition + +Motivational speaking aims to inspire and uplift the audience. This method is commonly used + +by motivational speakers, coaches, and leaders. + +Techniques + + Personal Stories: Share personal experiences that illustrate your message. + + Positive Language: Use positive and uplifting language to inspire the audience. + + Call to Action: Encourage the audience to take positive actions. + + Engage Emotions: Appeal to the audience's emotions to create a lasting impact. + +Advantages + + + +Inspiration: Can significantly motivate and inspire the audience. 27 -20 +  Positive Impact: Promotes positive change and action. -44 +Disadvantages -20 + Skepticism: Some audience members may be skeptical of motivational speeches. -20 + High Expectations: Audiences often have high expectations for impactful messages. -20 +Practical Tips -20 + Be Authentic: Speak from the heart to connect with the audience genuinely. -53 + Use Powerful Quotes: Incorporate powerful quotes to reinforce your message. -57 +Public speaking encompasses a variety of methods, each suited to different contexts and -52 +purposes. Understanding and mastering these methods can significantly enhance your ability -40 +to communicate effectively. Whether delivering an impromptu speech, reading from -60 +4.2 ADVANTAGES AND DISADVANTAGES OF PUBLIC SPEAKING -16 +Public speaking is a powerful tool that can have significant personal, professional, and -40 +social impacts. While it offers numerous advantages, it also presents certain challenges. -40 +Understanding both the benefits and the potential drawbacks of public speaking can help you -40 +navigate and leverage this skill more effectively. -40 +4.2.1 Personal Advantages -60 +Confidence Boost -13 +Public speaking can significantly enhance your self-confidence. Successfully delivering a -60 +speech in front of an audience can make you feel more capable and self-assured in various -60 +aspects of life. -19 + Self-Esteem: Regular public speaking engagements can improve your self-esteem as + +you become more comfortable with expressing your ideas. + + Overcoming Fear: Conquering the fear of public speaking, known as glossophobia, + +can have a positive ripple effect on other areas where anxiety might be present. + +Communication Skills -60 +Engaging in public speaking helps develop your verbal and non-verbal communication skills. + + Articulation: Improves your ability to express thoughts clearly and concisely. + + Body Language: Enhances your understanding and use of effective body language to + +complement your message. + +Critical Thinking + +Preparing and delivering speeches enhances critical thinking skills. 28 -34 +  Organization: Requires you to structure your thoughts logically and coherently. -40 + Analysis: Encourages deep analysis of topics and consideration of different -60 +perspectives. -23 +4.2.2 Professional Advantages -40 +Career Advancement -40 +Public speaking can open doors to career advancement opportunities. -60 + Leadership Roles: Effective communication is a key leadership skill, often leading to -33 +promotions and leadership positions. -60 + Visibility: Increases your visibility within your organization and industry, making you -80 +a more attractive candidate for advancement. -39 +Networking Opportunities -80 +Public speaking events provide opportunities to network and build professional relationships. -33 + -80 +Industry Connections: Engage with peers, mentors, and potential clients or -100 +collaborators. -100 + Reputation Building: Establish yourself as an expert or thought leader in your field. -100 +Skill Development -17 +Public speaking hones various skills that are valuable in the workplace. -25 + Persuasion: Develops your ability to persuade and influence others, crucial for sales, -80 +marketing, and leadership roles. -100 + Time Management: Enhances your ability to manage time effectively during -28 +presentations and meetings. -80 +4.2.3 Social Advantages -37 +Influence and Persuasion -80 +Public speaking enables you to influence and persuade others effectively. -40 + Advocacy: Use your voice to advocate for causes and inspire change. -80 + Community Impact: Engage with your community and contribute positively by -100 +sharing valuable insights and information. -100 +Enhanced Relationships -100 +Improves interpersonal relationships through better communication. -25 + Listening Skills: Encourages active listening, which is essential for effective -80 +communication in personal relationships. -100 + Empathy: Develops empathy by considering the audience's perspective and needs. -32 +29 -28 + Community Engagement -40 +Public speaking fosters active participation in community and social causes. -60 + Volunteering: Speak at community events or local organizations to support various -80 +causes. -100 + Awareness: Raise awareness about important issues and encourage collective action. -30 +4.2.4 Personal Disadvantages -40 +Anxiety and Fear -28 +Public speaking anxiety, or glossophobia, is a common and often debilitating fear. -23 + Physical Symptoms: Can cause physical symptoms such as sweating, shaking, and a -32 +racing heart. -19 + Mental Strain: Leads to stress and mental strain before and during speaking -20 +engagements. -20 +Time-Consuming Preparation -20 +Preparing a well-crafted speech can be time-consuming. -20 + Research: Requires extensive research to gather relevant information. + + Rehearsal: Involves significant time spent rehearsing to ensure a smooth delivery. + +Risk of Criticism + +Exposure to public scrutiny and potential criticism is a constant risk. + + Negative Feedback: Speakers must be prepared to handle negative feedback and + +criticism. + + Self-Doubt: Criticism can lead to self-doubt and decreased confidence. + +4.2.5 Professional Disadvantages -40 +Reputation Damage -60 +A poor public speaking performance can negatively impact your professional reputation. -80 + Perception: Can lead to negative perceptions about your competence and expertise. -100 + Career Impact: May affect career progression and opportunities for advancement. -47 +Miscommunication + +There is a risk of miscommunicating information or being misunderstood. + + Clarity: Ensuring your message is clear and understood by all audience members can + +be challenging. + + Consistency: Maintaining consistency in your message across different audiences and + +settings is crucial. 30 -40 + High Stakes -60 +High-stakes speaking engagements can add pressure and stress. -32 + Performance Anxiety: The importance of certain speeches can increase anxiety and -40 +pressure. -43 + Consequences: Mistakes in high-stakes situations can have significant consequences. -40 +4.2.6 Social Disadvantages -60 +Exhaustion -60 +Frequent public speaking can be physically and mentally exhausting. -80 + Energy Drain: Continuous speaking engagements can lead to burnout and fatigue. -34 + Stress: Managing multiple speaking commitments can be stressful. -80 +Privacy Concerns -37 +Public exposure can lead to a loss of privacy. -80 + Public Scrutiny: Increased visibility can attract unwanted attention and scrutiny. -100 + Personal Boundaries: Balancing public life and personal privacy can be challenging. -100 +Misinterpretation -100 +There is a risk of being misinterpreted or taken out of context. -20 + Media: Statements can be misrepresented by the media or others. -38 + Audience: Different audience members may interpret the same message in various -20 +ways. -41 +4.2.7 Overcoming Disadvantages -20 +Managing Anxiety -37 +Several strategies can help manage public speaking anxiety. -20 + Preparation: Thorough preparation reduces anxiety by increasing familiarity with the + +material. + + Practice: Regular practice builds confidence and reduces fear. + + Relaxation Techniques: Use deep breathing, meditation, and positive affirmations to + +calm nerves. + +Efficient Preparation -Parents' generation +Efficient preparation can mitigate the time-consuming nature of public speaking. -Young people's generation + Time Management: Allocate specific times for research, writing, and rehearsing. -Both about equally + Use of Technology: Leverage technology for efficient research and rehearsal. -Question: Thinking about people in your generation, compared to people in [his/her/their] generation, do you think the -following words describe your generation better, describe [his/her/their] generation better, or describe both generations -about equally well? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + Seek Feedback: Obtain feedback from peers to refine and improve the speech. -© Copyright 2018. The Associated Press and NORC +31 -June 2018 10 + Handling Criticism - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +Effective ways to handle criticism include: -Republicans are more likely than Democrats to describe their own generation as patriotic (51 percent -vs. 32 percent) and are less likely to describe their own generation as idealistic (27 percent vs. 39 -percent). + Constructive Feedback: View criticism as an opportunity for growth. -FOR SOME, POLITICS SEEPS INTO INTERACTIONS WITH EXTENDED -FAMILY. + Stay Professional: Maintain professionalism and composure in response to criticism. -A significant minority of both generations have had political disagreements with relatives that have -led to conflict. Twenty percent have unfriended or blocked a family member on social media due to a -disagreement over politics, and 21 percent have themselves been unfriended or blocked by a family -member. Fewer than 1 in 5 have had a holiday gathering ruined (11 percent), have decided not to attend -a family event (12 percent), or have had a family member decide to not attend a family event (17 -percent) due to political disagreements. + Learn and Improve: Use feedback to improve future performances. -While there are no differences by generation, Democrats are more likely than Republicans to have had -political disagreements with relatives affect their social media behavior. Democrats are more likely -than Republicans to say they’ve unfriended or blocked a family member on social media (26 percent -vs. 12 percent) and to have had a family member unfriend or block them on social media (27 percent vs. -15 percent). +4.2.8 Enhancing the Advantages -EVEN AMONG THOSE WHO APPROVE OF TRUMP, THE REPORT CARD IS -MEDIOCRE. +Building Confidence -Young people age 15-34 give poor marks to the federal government on its handling of many public -policy issues. Even among Trump’s supporters, the report card is still only average. Young people who -approve of Trump’s job performance most commonly give the federal government a C for all issues -except terrorism and homeland security and treatment of citizens by the police and justice system, -which earn a B, and abortion, which earns an F. For those who disapprove of Trump, F is the most -common grade for all issues except job creation, which most commonly gets graded a C. +Strategies to build confidence include: -Most common grades for the federal government’s handling of issues. + Small Steps: Start with smaller, less intimidating speaking opportunities. -Approve of Trump + Positive Visualization: Visualize successful speaking engagements to build -Disapprove of Trump +confidence. -Gun control + -Cost of higher education +Incremental Challenges: Gradually increase the complexity and size of speaking -Foreign relations +engagements. -Immigration +Career Advancement -The environment and climate change +Maximize career advancement opportunities through: -Health care policy + Seek Opportunities: Actively seek out speaking opportunities in your field. -Income inequality + Professional Development: Attend workshops and courses to improve public -The economy +speaking skills. -Taxes + Networking: Use public speaking events to network and build professional -Sexual harassment and assault +relationships. -Abortion +Community Engagement -Racism +Enhance community engagement by: -The rights of LGBT Americans + Get Involved: Participate in community events and causes that matter to you. -Terrorism and homeland security + Volunteer: Volunteer to speak at community gatherings or local organizations. -Treatment of citizens by police and justice system + Advocate: Use public speaking to advocate for social issues and causes. -The quality of education +4.2.9 Practical Tips for Effective Public Speaking -Job creation +Know Your Audience -C + Research: Understand the audience's interests, knowledge level, and expectations. -C + Tailor Your Message: Adapt your content to resonate with the audience. -C +Structure Your Speech -C + Clear Outline: Create a clear and logical outline. -C + Engaging Introduction: Start with a strong opening to capture attention. -C +32 + +  Coherent Body: Develop your main points logically and clearly. + + Powerful Conclusion: End with a memorable closing statement. + +Practice and Rehearse + + Regular Practice: Rehearse multiple times to build confidence and familiarity. + + Feedback: Seek feedback from peers or mentors to refine your speech. + +Engage with the Audience + + Eye Contact: Maintain eye contact to build a connection. -C + -C +Interactive Elements: Use questions, polls, and discussions to engage the audience. -C + Body Language: Use effective body language to enhance your message. -C +4.3 Conclusion -F +Public speaking offers numerous advantages, including personal growth, professional -C +advancement, and positive social impact. However, it also presents challenges such as -C +anxiety, time-consuming preparation, and potential criticism. By understanding these -B +advantages and disadvantages, speakers can better prepare and deliver effective speeches. -B +Overcoming the disadvantages and enhancing the advantages requires continuous practice, -C +preparation, and a willingness to learn and grow. Effective public speaking can lead to -C +significant personal and professional development, making it a valuable skill to master. -F +33 + + Unit- V + +5.1 FIVE ACTIVITIES TO IMPROVE PUBLIC SPEAKING SKILLS: + +1. Toastmasters International: Joining a Toastmasters club is an excellent way to -F +practice public speaking in a supportive environment. Members give speeches, -F +receive feedback, and participate in impromptu speaking exercises. -F +2. Debate Clubs: Participating in debate clubs helps develop argumentative and -F +persuasive speaking skills. Debates require participants to think on their feet and -F +present coherent arguments. -F +3. Storytelling Workshops: Storytelling is a powerful communication tool. Workshops -F +focusing on storytelling techniques can help speakers engage their audience and -F +deliver memorable speeches. -F +4. Mock Presentations: Organize mock presentations with friends or colleagues. -F +Practice delivering your speech in front of an audience and solicit feedback for -F +improvement. -F +5. Public Speaking Classes: Enroll in public speaking classes or workshops. These -F +courses often -F +34 + + 5.2 Twenty Public Speaking Topics -F +Instruction: Prepare a five minutes speech on any of the following topics: -C +1. Child Abuse -Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of -the following issues? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +2. Racism -© Copyright 2018. The Associated Press and NORC +3. Right to privacy -June 2018 11 +4. Common sense - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +5. Team Building -Looking at grades by political party, the results are similar. Democrats’ most common grade across -most issues is an F. However, Democrats give better marks to the federal government’s efforts on job -creation, most commonly grading it a C. +6. Freedom -Republicans are slightly more mixed, but most commonly give issues a C. For abortion, they most -commonly give a grade of F. +7. Physical fitness -STUDY METHODOLOGY +8. Feminism -This survey was conducted by The Associated Press-NORC Center for Public Affairs Research and -with funding from MTV and NORC at the University of Chicago. Staff from NORC at the University of -Chicago, The Associated Press, and MTV collaborated on all aspects of the study. +9. True friendship -Data were collected using AmeriSpeak®, NORC’s probability-based panel designed to be -representative of the U.S. household population. During the initial recruitment phase of the panel, -randomly selected U.S. households were sampled with a known, non-zero probability of selection from -the NORC National Sample Frame and then contacted by U.S. mail, email, telephone, and field -interviewers (face-to-face). The panel provides sample coverage of approximately 97 percent of the U.S. -household population. Those excluded from the sample include people with P.O. Box only addresses, -some addresses not listed in the USPS Delivery Sequence File, and some newly constructed dwellings. +10. Emotional Intelligence -Interviews for this survey were conducted between April 23 and May 9, 2018, with young people age -15-34 and parents of young people age 15-26 representing the 50 states and the District of Columbia. -Interviews were conducted in both English and Spanish, depending on respondent preference. Adult -panel members age 18-34 were randomly drawn from AmeriSpeak, and 791 completed the survey—780 -via the web and 11 via telephone. Teen panel members age 15-17 were drawn from AmeriSpeak Teen, -and 148 completed the survey—145 via the web and 3 via telephone. Adult panel members who are -known parents of 15-17 year olds were randomly drawn from AmeriSpeak, and 104 completed the -survey—101 via the web and 3 via telephone. In addition, panel members age 30-76 were randomly -drawn to screen for parents of young adults age 18-26, and 503 completed the survey—469 via the web -and 34 via telephone. +11. Personality Development -Among young people age 15-34, the final stage completion rate is 25.9 percent, the weighted household -panel response rate is 33.7 percent, and the weighted household panel retention rate is 87.2 percent, for -a cumulative response rate of 7.6 percent. The overall margin of sampling error among young people -age 15-34 is +/- 4.3 percentage points at the 95 percent confidence level, including the design effect. The -overall margin of sampling error among young people age 15-26 is +/- 5.6 percentage points at the 95 -percent confidence level, including the design effect. The margin of sampling error may be higher for -subgroups. +12. Human rights -Among parents of young people age 15-26, the screener completion rate is 27.9 percent, with an -incidence rate of 25.2. The interview completion rate is 95.6 percent, and the final stage completion rate -is 26.7 percent. The weighted household panel response rate is 33.7 percent, and the weighted -household panel retention rate is 87.2 percent, for a cumulative response rate of 7.8 percent. The -overall margin of sampling error is +/- 6.7 percentage points at the 95 percent confidence level, -including the design effect. The margin of sampling error may be higher for subgroups. +13. Sophistication -Once the sample has been selected and fielded, and all the study data have been collected and made -final, a poststratification process is used to adjust for any survey nonresponse as well as any -noncoverage or under- and oversampling resulting from the study specific sample design. -Poststratification variables included age, gender, census division, race/ethnicity, and education. +14. Festivals in India -© Copyright 2018. The Associated Press and NORC +15. A good (or bad) boss -June 2018 12 +16. Positive attitude - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +17. If I were the Prime Minister -Weighting variables were obtained from the 2017 Current Population Survey. The weighted data -reflect the U.S. population of young people age 15-34 and parents of young people age 15-26. +18. Cell phones and Health -All analyses were conducted using STATA (version 14), which allows for adjustment of standard errors -for complex sample designs. All differences reported between subgroups of the U.S. population are at -the 95 percent level of statistical significance, meaning that there is only a 5 percent (or less) -probability that the observed differences could be attributed to chance variation in sampling. +19. Lessons from Nature -A comprehensive listing of all study questions, complete with tabulations of top-level results for each -question, is available on The AP-NORC Center for Public Affairs Research website: www.apnorc.org. +20. Life without internet -For more information, email info@apnorc.org. +35 -CONTRIBUTING RESEARCHERS + 5.3 Public Speaking Record -From NORC at the University of Chicago +Topic: ……………………………………………………. -From The Associated Press + Date: …………………. -Becky Reimer -Emily Alvarez -Liz Kantor -Trevor Tompson -Jennifer Benz +Blue Print of the Presentation -Emily Swanson +Introduction -ABOUT THE ASSOCIATED PRESS-NORC CENTER FOR PUBLIC AFFAIRS -RESEARCH +Main Idea 1 -The AP-NORC Center for Public Affairs Research taps into the power of social science research and the -highest-quality journalism to bring key information to people across the nation and throughout the -world. +Main idea 2 -■ +Main Idea 3 -The Associated Press (AP) is the world’s essential news organization, bringing fast, unbiased news -to all media platforms and formats. +Conclusion -■ NORC at the University of Chicago is one of the oldest and most respected, independent research +Observation by the Faculty: -institutions in the world. +Details -The two organizations have established The AP-NORC Center for Public Affairs Research to conduct, -analyze, and distribute social science research in the public interest on newsworthy topics, and to use -the power of journalism to tell the stories that research reveals. +Eye Contact +Body Language +Language +Content +Presentation Skills -The founding principles of The AP-NORC Center include a mandate to carefully preserve and protect -the scientific integrity and objectivity of NORC and the journalistic independence of AP. All work -conducted by the Center conforms to the highest levels of scientific integrity to prevent any real or -perceived bias in the research. All of the work of the Center is subject to review by its advisory -committee to help ensure it meets these standards. The Center will publicize the results of all studies -and make all datasets and study documentation available to scholars and the public. +Marks Awarded -ABOUT MTV +Remarks -MTV is a global youth culture brand inspired by music. For more information, check out -mtvpress.com. MTV is a unit of Viacom Inc. (NASDAQ: VIAB, VIA) +10 -© Copyright 2018. The Associated Press and NORC +36 -June 2018 13 +(Signature of the Faculty) diff --git a/q104/random_k2/question.json b/q104/random_k2/question.json index d39558bb7149772cf25cd97ee84df919e984f4ff..1d543460809734c2c26ea4cb4d1834715a5da20e 100644 --- a/q104/random_k2/question.json +++ b/q104/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "e639029d16094ea71d964e2fb953952b.pdf", - "web_0867a04cdab7a273.pdf", - "web_a39c0d7f5ff0dedb.pdf" + "web_6a55c713826db217.pdf", + "web_bf09a64e7d517c98.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q104/random_k2/random_1.md b/q104/random_k2/random_1.md index 85e510cedc0e76a41806e14e4f5679216b834d6f..5c5eba2e1fd980c21e8cbb1cd873801d5dedd049 100644 --- a/q104/random_k2/random_1.md +++ b/q104/random_k2/random_1.md @@ -1,1480 +1,2541 @@ -Historic Architectural Resources Survey Report + CONTENTS -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +01 +04 + 16 +36 +38 +39 +40 +41 +44 +45 -North Carolina Department of Transportation -Report Prepared by Nancy Van Dolsen +/ About HKU Business School -November 1999 +/ Explore Your Options - Historic Architectural Resources Survey Report +/ Enrich Your Life -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +/ 2026 JUPAS Admissions -November 1999 +/ 2026 Non-JUPAS Admissions -Principal Investigator (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation +/ 2026 Dual Degree Admissions + +/ Develop Your Career + +/ Freshmen Sharing + +/ Our Global Presence + +/ Useful Websites + +About HKU Business School + +HKU Business School strives to nurture first-class business leaders and foster academic and relevant +research to serve the needs of Hong Kong, Chinese Mainland and the rest of the world in the new Asia-led +economy. As Asia’s premier international business school, we engage leading scholars from all corners of the +globe and they instil in students global knowledge with an Asian perspective. + +With the highest proportion of non-local undergraduate students across all Faculties at HKU, four of our +undergraduate programmes ranked amongst the University’s Top 10 programmes in 2025#, HKU Business +School attracts the best and brightest students from Hong Kong and beyond. + +HKU Business School is fully accredited by the European Quality Improvement Systems (EQUIS) and the +Association to Advance Collegiate Schools of Business (AACSB). + +# Based on average JUPAS admission scores in the best 5 subjects in HKDSE + +VISION +To be a leading, +globally-impactful academic +institution of business and +economics. + +UNIQUE +PROPOSITION + +Deeply rooted in Hong Kong +Fully engaged with China +Truly international + +MISSION +STATEMENT + +To provide world-leading +business and economics +education that nurtures +dynamic and responsible +leaders. +To advance the frontiers of +knowledge through innovative +and globally-significant +research. +To create positive change in the +world by partnering with +communities. + +01 + + Dean’s Message + +I am delighted to introduce you to the HKU Business +School, a dynamic and forward-thinking academic +community where transformative opportunities and +unparalleled experiences await. + +As a part of the University of Hong Kong, which +proudly ranks 11th globally in the 2026 QS World +University Rankings, HKU Business School stands out +as a beacon of excellence in business and economics +in Asia. This year marks the 25th anniversary of our +Faculty, we continue to march forward as a leading, +impactful business school, championing +globally +research, and +leadership, pioneering +thought +educational excellence. + +of + +dynamic, + +challenging + +responsible, + +Our mission is to inspire and develop the next +generation +and +globally-minded business leaders. We offer a vibrant +and +learning environment where +students are encouraged to push their limits, think +critically, and innovate. Our team of leading scholars +from around the world, and experienced and +dedicated teachers as well as mentors bring a wealth +of knowledge, global perspectives, and personalised +guidance beyond the classroom. Our full spectrum of +top-quality degree programmes, including 11 diverse +undergraduate programmes and four dual bachelor’s +degrees +in partnership with world-renowned +universities, is designed to equip students with +cutting-edge knowledge and technical skills to +in the fast-changing global +navigate and excel +business landscape. + +In response to the transformative power of artificial +intelligence, we are proud to launch the Second Major +and Minor programmes in “Artificial Intelligence in +Business” +to our +(AIB), offered exclusively +undergraduate students. These programmes are +designed to equip future business +leaders with +practical proficiency in AI technologies, strategic +insight to drive cross-sector innovation, and a critical +understanding of the ethical and societal implications +of AI adoption. Through a rigorous curriculum that +integrates business fundamentals with cutting-edge +AI applications, students will be empowered to tackle +complex challenges and create meaningful impact in +a data-driven world. + +Greetings from +Associate Dean (Undergraduate) + +inspiration + +Business education today is an evolving domain that +builds upon learning from all academic disciplines and +draws +successful and +from both +unsuccessful business practices. Our undergraduate +curricula in six core academic areas equip students +with diverse knowledge across business and +economics spheres, and prepare them to be creative +leaders of tomorrow with global perspectives. + +A hallmark of our School that separates us from the +others is our truly multi-disciplinary approach. HKU +Business School’s extensive resources offer students +in +a wide range of choices, combining studies +economics and finance, accounting and finance, +accounting and data analytics, business and law, +business and data analytics, international business +and global management, as well as dual degrees in +collaboration with world-renowned universities. + +Our faculty members will engage you with their +inspiring teaching and how they bring their impactful +research into the classroom. They will open your mind +to diverse perspectives, +ignite your curiosity to +explore innovative ideas, and instil in you an appetite +for acquiring new knowledge and extraordinary +experiences throughout your academic journey and +career development. Our students are among the +brightest and most creative of their peers. Our alumni +have gone on to forge brilliant careers in all sectors of +the +international business +community, and have formed a global network that +you will be a part of for the rest of your life. + +regional and + +local, + +I hope that you will share my enthusiasm for the +programmes and my admiration for the people who +have contributed to their development. I look forward +to having you as a member of the HKU Business +School family and a future leader in the world of +business. + +Warm regards, + +Professor Derek Chan +Associate Dean (Undergraduate) + +Through our extensive partnerships with over 300 +institutions and corporations across more than 40 +countries, as well as our unique network of centres in +Beijing, Shenzhen, Shanghai, Vietnam, and Barcelona, +our students gain unparalleled global experiential +and enrichment opportunities. Whether through +internships, overseas exchange programmes, or +hands-on +these experiences +sharpen practical skills, broaden perspectives, and +prepare students to confidently tackle real-world +challenges. + +industry projects, + +Joining HKU Business School means becoming part of +a community that supports your growth and success. +You will collaborate with brilliant minds, engaging in +stimulating discussions, and working on impactful +projects that make a difference. The friendships, +networks, and experiences you build here will last a +to a myriad of career +lifetime, open doors +opportunities, and empower you to overcome +challenges, pursue your passions with confidence, +and contribute positively to your community and the +world at large. + +If you are ready to embark on an exciting journey of +learning, discovery, and personal development, I +invite you to explore what HKU Business School has to +look forward to welcoming you to our +offer. I +community, where future leaders are nurtured and +dreams take flight. + +Warm regards, + +Professor Hongbin Cai +Dean +Chair of Economics + +0302 Explore Your Options +A Wide Selection of Programmes + +Bachelor of Business Administration [BBA] + +The BBA programme combines strong functional training with intensive exposure +to communication skills, computer applications, and other social science and/or +science subjects. Students can choose to major in the following area(s): + +Entrepreneurship, +Design and + Innovation* + +Finance + +Human Resource +Management + +Information Systems +and Analytics + +Marketing + +BBA Curriculum Structure + +Components -441-1/4,4 (cid:9) +A -Barbara Church, Supervisor (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation +B -Date +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -//7/67/1. -Date +BBA Major +Faculty Core Courses +Disciplinary Core Courses / Disciplinary Electives +Capstone Course - TABLE OF CONTENTS +C -Project Description (cid:9) +Free Electives / Optional Major / Optional Minor(s) -Purpose of Survey and Report (cid:9) +Total -Methodology (cid:9) +Credits -Summary Findings of the Survey (cid:9) +54 +12 +36 +6 +- -Historic Contexts and Background Information (cid:9) +78 +24 +48 +6 -Property Inventory and Evaluations +108 -William Jones Arney House (cid:9) +240 -Marshall Arney House (cid:9) +*Candidates must undergo a selection process arranged by the Programme Coordinator for +declaring the EDI major. -Corpening, House (cid:9) +HKU – Sciences Po Dual Degree Programme -Early Twentieth-Century Frame House (cid:9) +Students will spend the first two years at Sciences Po in France and the third +and fourth years at HKU. Programmes at Sciences Po are taught in English (no +French requirements for being admitted) with intensive French language +classes for all non-francophone students. -Nineteenth-Century Log House (cid:9) +Upon successful completion of the dual degree programme, students will be +awarded the Bachelor of Business Administration (BBA) degree from HKU and +a Bachelor’s degree from Sciences Po. -Bibliography (cid:9) +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Appendix A: Correspondence with the State Historic Preservation Office (cid:9) +Find out more: +https://admissions.hku.hk/programmes/university- +collaborative-programmes/hku-sciences-po-dual- +degree-programme -LIST OF FIGURES +HKU – University of British Columbia +Dual Degree Programme -Figure 1: Project Location Map (cid:9) +Students will spend the first and final years at HKU and the second and third +years at the University of British Columbia (UBC) in Canada, and graduate with +a Bachelor of Business Administration from HKU and a Bachelor of Commerce +from UBC upon successful completion of the programme. -Figure 2: Area of Potential Effects with Historic Properties Noted (cid:9) +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Figure 3: Tax Map, William Jones Arney House (cid:9) +Find out more: +https://admissions.hku.hk/programmes/university- +collaborative-programmes/hku-university-of-british- +columbia-dual-degree-programme -Figure 4: Site Plan, William Jones Arney House (cid:9) +Find out more: +https://ug.hkubs.hku.hk/programme/bba -Figure 5: Tax Map, Marshall Arney House (cid:9) +0504 -Figure 6: Site Plan, Marshall Arney House (cid:9) + Bachelor of Business Administration in Accounting and +Finance [BBA(Acc&Fin)] -Page +Bachelor of Business Administration in Accounting Data +Analytics [BBA(ADA)] - 1 +The BBA(Acc&Fin) programme prepares students for a professional qualification +in accounting and provides the broad-based education in accounting and finance +which is necessary to progress toward a leadership role in the financial sector. +Students are required to take professional core in accounting and a minor in +finance, with an option to second major in finance. Tailor-made professional +internship opportunities with renowned +preparation programmes and +accounting firms and financial institutions are available for talented students. It +is accredited as an approved accountancy programme by: -1 +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia -1 +BBA(Acc&Fin) Curriculum Structure -4 +Components - 6 +Credits -8 +A - 15 +B -24 +C -28 +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Professional Core in Accounting +Faculty Core Courses +Disciplinary Core Courses +Capstone Course + +Minor in Finance* +Disciplinary Core Courses +Disciplinary Electives + +D. -31 +Free Electives / Optional Minor -.32 3 +*Students may opt to pursue a Major in Finance: -.10 11 +Components -.14 .15 +A - Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +B -PROJECT DESCRIPTION +C -The North Carolina Department of Transportation (NCDOT) has plans to pave SR 1424, -Johns River Road and SR 1438, Corpening Bridge Road, Quaker Meadows and Upper -Creek townships, Burke County (Figure 1). In addition, the NCDOT plans to minimize -curves in the roadway. A nationwide permit will be needed for the project. +University Graduation Requirements -The area of potential effects (APE) for historic architectural resources was delineated by -a NCDOT staff architectural historian and reviewed in the field. The APE boundary is -shown on the attached map (Figure 2). +Professional Core in Accounting -PURPOSE OF SURVEY AND REPORT +Major in Finance +Disciplinary Core Courses +Disciplinary Electives +Free Electives +Capstone Course -This Final Identification and Evaluation survey was conducted and the report prepared in -order to identify historic architectural resources located within the APE as part of the -environmental studies conducted by NCDOT and documented by a Categorical Exclusion -(CE). This report is prepared as a technical addendum to the CE and adheres to the -guidelines for compliance with the National Historic Preservation Act of 1966, as -amended. This report is on file at NCDOT and is available for review by the general -public. +54 +12 +36 +6 +- -METHODOLOGY +108 +24 +78 +6 -NCDOT conducted the survey and prepared this report in accordance with the provisions -of FHWA Technical Advisory T6640.8A (Guidance for Preparing and Processing -Environmental and Section 4(f) Documents); the Secretary of the Interior's Standards and -Guidelines for Archaeological and Historic Preservation (48 FR 44716); 36 CFR Part -800; 36 CFR Part 60; and Survey Procedures and Report Guidelines for Historic -Architectural Resources by NCDOT. This survey and report meet the guidelines of -NCDOT and the National Park Service. +36 +12 +24 -NCDOT completed an intensive survey with the following goals: (1) to determine the -APE, defined as the geographic area or areas within which a project may cause changes -in the character or use of historic properties, if any such properties exist; (2) to identify +42 -Historic Architectural Resources Survey Report; NCDOT,ISR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen +Total -Page 1 +240 - Figure 1: Project Location Map +Credits -Historic .4rchitectural Resources Survey Report: NCDOT, SR 1411, Burke County: November 1999 (cid:9) -Nancy Van Dolsen +54 -Page ? +108 + +78 +30 +18 +24 +6 + +Total + +240 + +Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares +students for a professional qualification in accounting. It provides students +with an option to focus on data analytics aspect in accounting instead of having +a minor in finance. It is an excellent choice for students who are interested in +STEM with accounting and business applications and wish to pursue their +career in the relevant fields. The intensive accounting training enables +BBA(ADA) students to distinguish themselves from students of the other data +analytics/STEM programmes. Students will be equipped with comprehensive +accounting and business knowledge that prepares them to obtain professional +qualifications in accounting. They will also acquire computer programming +and data analytics skills to meet the needs of the accounting and business +professions in today’s digital transformation economy. It is accredited as an +approved accountancy programme by: + +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia + +BBA(ADA) Curriculum Structure + +Components - S -E -C -R -U -O -S -E -R -L A -R -U -T -C -E -T -I -H + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Professional Core in Accounting Data Analytics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + C -R + +Free Electives + +Total + +Credits + +54 +12 +36 +6 +- + +156 +24 +114 +12 +6 + +30 + +240 + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-acc-and-fin + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-ada + +0706 + + Bachelor of Business Administration (Business Analytics) +[BBA(BA)] + +Bachelor of Business Administration (Law) and Bachelor of +Laws [BBA(Law)&LLB] + +The BBA(BA) programme is crafted to meet the growing industry demand for +talents in the business analytics field. Its curriculum offers students a +comprehensive coverage of both technical and managerial skill sets. Students +will learn a wide spectrum of knowledge from the disciplines of information +technology, data science, business statistics and management. Core courses +include programming language, database management, introduction to +business analytics, management information systems, decision and risk +analysis, etc. They help students build the foundations for advanced topics in +senior years. Students also have the option to choose from analytics technique +courses and business application courses. The aim is to provide students with +state-of-the-art experiences in artificial intelligence, big data, social media, +and other latest development. Such design helps train well-rounded students +who will be more adaptable in their career development. The programme is +suitable for students who have a passion for problem-solving through +analytics, and who aim to develop their career path in analytics. + +BBA(BA) Curriculum Structure + +Components + +Credits + A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Business Analytics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives (Analytics Techniques) +Disciplinary Electives (Business Applications) +Capstone Course + C -I -R -O -T -S -I -H - -P -A -M -Y -E -V -R -U -S - -e -l -t -i - -T - -t -e -e -h -S - -: -§ - -9 -0 -1 -2 -5 -8 -. + +Free Electives / Optional Major / Optional Minor(s) + +54 +12 +36 +6 +- + +96 +24 +42 +12 +12 6 -8 -3 -4 -1 -R -S -& -4 -2 -4 -1 -R -S +90 + +Jointly offered by the HKU Business School and the Faculty of Law, the +BBA(Law)&LLB integrated double-degree programme blends two strategic +disciplines into a single programme for the bright and highly motivated. It is an +excellent choice for students seeking a broader education with parallel +concentration on legal and business studies. Graduates of the programme will be +awarded two degrees — BBA(Law) by the HKU Business School, and LLB by the +Faculty of Law. After successful completion of the LLB programme, students +may apply for admission to the Postgraduate Certificate in Laws (PCLL) +programme provided that they meet the relevant admission requirements. + +The programme offers business stream and accounting stream to students in +their business studies. The accounting stream is accredited by: + +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia + +BBA(Law)&LLB Curriculum Structure (Business Stream) + +Components + +Credits A -N -I -L -O -R + +B + +C + +University Graduation Requirements +Language Enhancement Course +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +BBA Major +Faculty Core Courses +Disciplinary Core Courses / Disciplinary Electives +Capstone Course + +Professional Core in Law +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + +D. + +Free Electives / Optional Minor + +Total + +240 + +Total + +BBA(Law)&LLB Curriculum Structure (Accounting Stream) + +Components + A + +B + +University Graduation Requirements + +Professional Core in Accounting +Faculty Core Courses +Disciplinary Core Courses +Capstone Course + C -H -T -R -O -N - -, - -Y -T -N -U -O + +Professional Core in Law + +Total + +36 +6 +24 +6 +- + +78 +24 +48 +6 + +156 +108 +42 +6 + +30 + +300 + +Credits + +36 + +108 +24 +78 +6 + +156 + +300 + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-ba + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-law-and-llb + +0908 + + Bachelor of Business Administration in International Business +and Global Management [BBA(IBGM)] + +Bachelor of Economics [BEcon] + +BBA(IBGM) is an intensive, premier programme in international business and +global management featuring a unique curriculum designed to evolve with and +reflect the ever-changing demands of a globalised world. Exclusive elements +such as overseas exchange studies, a leadership boot camp, an Asia-Pacific +study tour, an international field trip and global enrichment opportunities are +strategically integrated into the curriculum, whilst students are given the +flexibility to pursue an in-depth inquiry into another specialised area of study +as a second major. This cross-functional and personalised approach aims to +foster a selected group of outstanding students to become the world’s future +leaders equipped with global perspectives, cultural sensitivity, social +responsibility and key competencies to succeed in the international business +environment. + +BBA(IBGM) Curriculum Structure + +Components + +A + +B + C + +D + E -K -R -U -B -t -c -e -j -o -r -P +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -i -c -F +Major in International Business and Global Management +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -2 +Second Major -. +Free Electives -W -z +Compulsory Overseas Exchange -p +Total -l -l -i +Credits -) -a +54 +12 +36 +6 +- -• +96 +24 +54 +12 +6 - i -0 - 1 +72-90 -g +0-18* -5, -g +- -1 +240 -dVIA41 A3AWS S302:10S3 1V1if1.1031.1H3lV ONOISIH +*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. -SIN :37VOS +The BEcon programme trains students in economic theories, statistical tools, +the application of theories to solve practical economic problems in trade, +banking and the labour market, and evaluation of the regional economic +performance of Chinese Mainland, Hong Kong and other economies. Students +will be furnished with solid theoretical and empirical economic knowledge to +analyse and understand human behaviour and social interactions. The +programme allows students a high degree of flexibility in planning their future +career specialisation. -— +BEcon Curriculum Structure -I +Components -g +A -§ +B - all significant resources within the APE; and (3) to evaluate these resources according to -the National Register of Historic Places criteria. +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -The survey methodology consisted of a field survey and background research on the -project area. A NCDOT staff architectural historian conducted a field survey in August -and October 1999, by car and on foot. All structures over 50 years of age in the APE -were photographed and keyed to an area map. +Major in Economics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -Background research was conducted at the Western office of the State Historic -Preservation Office, Asheville; the North Carolina State Library, Raleigh; and the Burke -County Court House in Morganton. +C -A meeting was held with Ms. Jennifer Martin at the State Historic Preservation Office on -August 20, 1999 to discuss the eligibility of two of the historic resources, the William -Jones Arney House and the Marshall Arney House. It was agreed at that time the two -properties were potentially eligible for the National Register of Historic Places. +Free Electives / Optional Major / Optional Minor(s) -SUMMARY FINDINGS OF THE SURVEY +Total -The project proposes to pave and improve SR 1424, Johns River Road and SR 1438, -Corpening Bridge Road in Quaker Meadows and Upper Creek townships, Burke County. -In letters dated July 9, 1999 and August 10, 1999, the State Historic Preservation Office -identified four structures of historical or architectural importance within the general -project area, the Marshall Arney House, the William Arney House, the Arney Chapel and -the Corpening House (See Appendix A). Two additional historic properties, not -mentioned in the letters, are located within the project area: a late nineteenth-century log -house and an early twentieth-century frame house. No properties listed on the National -Register were located within the APE for the project. Of the four historic properties -mentioned in the letter, one, the Arney Chapel, is located outside the APE. Of the five -historic properties located within the APE, three—the Corpening House, the nineteenth- -century log house, and the early twentieth-century frame house --were determined not -eligible for the National Register. The Marshall Arney House and the William Arney -House were determined to be eligible for the National Register. Criterion Consideration -G, for properties that have achieved significance within the last fifty years, states that -properties less than fifty years of age may be listed on the National Register only if they -are of exceptional importance or if they are integral parts of districts eligible for the -National Register. There are no properties in the APE that qualify for the National -Register under Criterion Consideration G. +Credits -Properties Listed on the National Register: +54 +12 +36 +6 +- -None +78 +24 +30 +18 +6 -Ilistoric Architectural Resources Survey Report: iVCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nancy Ian Dolsen +108 -Page 4 +240 - Properties Listed on the North Carolina State Study List: +Find out more: +http://ibgm.business.hku.hk -None +Find out more: +https://ug.hkubs.hku.hk/programme/becon -Properties Evaluated and Considered Eligible for the National Register: +1110 -William Jones Amey House, page 8 -Marshall Amey House, page 15 + Bachelor of Economics and Finance [BEcon&Fin] -Properties Evaluated and Considered Not Eligible for the National Register: +BEcon&Fin is designed to meet the interests of students who are seeking a +balanced study of both economics and finance. The programme recognises +the significance of economics both in its own right and as a foundation for the +study of finance, and the significance of finance as a career path for students +who are keen on joining this profession. Students will receive solid training in +theories and applications of economics and finance. They will be equipped +with the skills to analyse human behaviour and social interaction, as well as to +solve problems in the financial sphere. -Corpening House, page 21 -Twentieth-century frame house, page 24 -Nineteenth-century log house, pay 28 +BEcon&Fin Curriculum Structure -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen +Components -Page 5 +Credits - HISTORIC CONTEXTS AND BACKGROUND INFORMATION +A -During the late seventeenth century, occupation of the land of present-day Burke County -was contested between two native-American tribes, the Cherokee and the Catawba. The -area eventually became a buffer zone between the two tribes. The French and Indian War -(1754-1763) and the American Revolution had a catastrophic effect on both tribes, -forcing the Cherokees further into the mountains, and reducing the Catawba population to -250 members by 1784. +B -Transient hunters and trappers of European descent settled in the county during the -middle of the eighteenth century. In 1777 the General Assembly approved the formation -of a new county named Burke in the western portion of the state that encompassed all or -part of 16 present counties. Burke was named for Thomas Burke (1747-1783), a -member of the Continental Congress and third Governor of North Carolina.' Burke -County achieved its present size of 511 square acres in 1885. +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -According to an architectural survey of the county completed in 1983-1984, there are -three distinct regions in the county: the rich bottom lands along the Catawba, Johns, -Upper and Lower Creeks, and Warrior Forks rivers; the remote dead-end coves of the -South Mountains and Blue Ridge; and the rolling foothills interspersed with small -valleys.2 The subject project is located along the bottom land of the Johns River in -Quaker Meadows and Upper Creek townships. +Major in Economics and Finance +Faculty Core Courses +Disciplinary Core Courses +Economics Disciplinary Electives +Finance Disciplinary Electives +Capstone Course -The bottom lands contained the richest agricultural soil in the county, and were the site of -antebellum plantations. Most of the county's large, brick plantation homes dating to the -first half of the nineteenth century stand along the Johns and Catawba rivers in Quaker -Meadows and Upper Creek townships, including Bellevue (1823-1826), Cedar Grove -(1825), and Pleasant Valley (ca. 1820; ca. 1975). In Burke, the typical plantation of 1860 -consisted of 850 acres of improved land and thousands of acres of unimproved land. -The typical 1860 Burke County farm, however, contained only 60 acres of cultivated land -and only 60 out of 1,239 families owned more than 100 acres of improved land.3 - -After the Civil War, the land-use pattern of ownership changed. The large-scale -plantations were dissolved, and there was an increase in the number of small-scale farms, -from 500 in 1860 to 1,648 in 1880. At the same time, the average farm size dropped to -27 acres.4 - -The project area is the site of five post-bellum houses that date to this period of -subdividing larger plantation properties. The houses stand on land first owned by one of -Burke County's first European settlers, John Perkins, who first explored the county in the - -I Entry for Thomas Burke, Biographical Directory of the United States Congress, I774-Present, Website: -Bioguide.conress.gov. -2 J. Randall Cotton, et at., Historic Burke: An Architectural Sites Inventory of Burke County (Morganton, -NC: Historic Burke Foundation, Inc., 1987) 13. -3 Cotton, et at, 23. -4 Cotton, et al, 27. - -Historic Architectural Resources Survey Report: NCDOT. SR 1424. Burke County: November 1999 -Nancy Van Dolsen - -Page 6 - - 1750s. These five houses represent a range of building types constructed in the county -during the last quarter of the nineteenth century: two frame T-shaped houses, a small -two-room log house, and two square-shaped two-story frame dwellings. - -At the end of the twentieth century, much of the land in the county is held by three -property owners: the US Government (Pisgah Forest), the state (South Mountain State -Park) and Carolina Centers, a timber company. Carolina Centers presently owns the land -upon which the Corpening, Marshall Amey, and William Jones Arney houses stand. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 7 - - PROPERTY INVENTORY AND EVALUATIONS +C -William Jones Arney House +Free Electives / Optional Major / Optional Minor(s) -Location: (cid:9) -The William Jones Arney House is located on the northwest corner -of the junction of SR 1424 and SR 1438. The house sits on a 23,136.10-acre tract -owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a timber -company. The property was formerly owned by Duke Power, Inc. - -The William Jones Arney House sits within a cleared area, at the - -Setting: (cid:9) -top of a rise, facing east to Johns River. A short dirt lane leading from Johns -River Road runs south of the house, ending at a small frame granary west of the -dwelling. - -Description: - -Dwelling: The William Jones Arney House, dating to ca. 1890, is a two- -and-a-half story frame residence with a T-shaped plan. The exterior of the -house is clad with narrow weatherboards, and a metal roof covers the -building. To the rear of the building is a kitchen connected to the main -house by a covered walkway. The kitchen stands one-story and has a -shed-roofed porch on each eave elevation. The windows in the kitchen -contain six-over-six sash and those in the main block of the house have -two-over-two sash. A one-story tripartite bay window adorns the south -gable end wall of the ell. Pointed arched louvered windows are located at -the peaks of the gable end walls. Two interior brick chimneys are located -at the center of the ells, providing an interior stack to each room in the -house. +54 +12 +36 +6 +- -The interior walls of the house are sheathed in horizontal boards, as are the -ceilings. The house (excluding the kitchen) has four rooms on the first -and four rooms on the second floor. +96 +24 +42 +12 +12 +6 -Well House: A concrete block well house stands south of the south porch -of the kitchen. The well house is approximately 3 feet in height and is -covered with a corrugated metal roof. +90 -Granary: To the west of the house stands a frame granary also constructed -ca. 18900. The granary building is a gable-front building with a central -door. The building is weatherboarded and has a metal roof. An open side -shed was constructed onto the east eave wall, and a shed roof porch is -located on the south elevation. +Total -Background Information: (cid:9) -land ca. 1885. Arney, a descendent of one of the pioneering families in the -county, both farmed and owned a general store (now demolished). The store +240 -William Jones Arney (1849-1911) moved to this +HKU - Peking University “Future +Leaders” Dual Bachelor’s Degree in +Economics, Finance and Management +The programme combines the strengths of two leading business schools in +Hong Kong and Chinese Mainland and is designed to nurture first-class business +leaders and cultivate their global perspective with intimate understanding of +China. Students will spend the first two years at HKU and the third and fourth +years at Peking University (PKU), and be awarded a Bachelor of Economics and +Finance (BEcon&Fin) degree from HKU and a Bachelor of Management [BMgmt] +degree from PKU upon successful completion of the dual degree programme. -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy ran Dolsen +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Page 8 +Find out more: +https://ug.hkubs.hku.hk/admission/degree-program +mes/hku-pku-dual-degree - stood to the west, at the intersection of Corpening Bridge and Piedmont roads. -Arney, and his wife Margaret McCall, owned over 500 acres in the area along -John's River. The couple had nine children? +Bachelor of Finance in Asset Management and Private +Banking [BFin(AMPB)] +The BFin(AMPB) programme aims to prepare students to enter the rapidly +growing fields of asset management and private wealth management in Hong +Kong and Asia Pacific, or in banking and finance generally. During the first two +years, students complete introductory courses in accounting, economics, +mathematics, statistics, corporate finance, and investments and portfolio +analysis. In subsequent years, students take more specified and industry +related courses including equity valuation and investment in specific asset +classes e.g. private equity, venture capital, hedge funds and digital assets, +understanding of financial statements and doing financial analysis, lending +and credit in private wealth and corporate bank, financial regulations and +compliance, current +institutional +in finance, ESG +consulting and family office. The programme contains practical-oriented +courses taught by teachers with extensive experience in the finance industry +and offers an array of career-oriented enrichment activities and networking +opportunities with industry practitioners. -At the death of William and Margaret Arney, ownership of the part of the farm -upon which this house stood was transferred to George Phifer Arney who sold -240 acres to Western Carolina Power for $100 in 1922.6 The property is presently -owned by Carolina Centers, LLC. +investment, -For purposes of compliance with Section 106 of the National +issues -Evaluation: (cid:9) -Historic Preservation Act of 1966, as amended, the William Jones Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A, since it is not representative of a significant trend in the history of the -area. The property is no longer under cultivation, and all but one of the -agricultural outbuildings have been demolished or removed. +in -The William Jones Arney House is considered not eligible for the National -Register of Historic Places under Criterion B because it is not associated with any -individuals of specific historical significance. +BFin(AMPB) Curriculum Structure -The William Jones Arney House is considered eligible for the National Register -under Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is a regional form built by farmers of middle economic status. The Arney -House is a good representation of this form, and exhibits excellent integrity and -retention of setting. The granary building is also representative of a vernacular -building form, and exhibits good integrity of design, materials, workmanship, -and setting. +Components -The building is not likely to yield information important to the history of building -technology; therefore it is not eligible for the National Register under Criterion D. +Credits -Boundaries: The proposed National Register boundaries comprise approximately -2.8 acres that surround the house and its accompanying outbuilding. The land -was deeded to George Phifer Arney, and is the portion of the William Jones -Arney estate upon which the subject structures stand. This parcel is part of parcel -JR-21, which is a portion of the larger Carolina Centers tract. +A -Boundary Justification: (cid:9) -include the house, outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. +B -The proposed historic boundaries were drawn to +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -5 William Jones Amey House, Historic Resource Survey Form, Burke County, 1983. Asheville, NC: -Western Office, Division of Cultural Resources. -6 Burke County Deed Book T4:293. +Major in Asset Management and Private Banking +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 -Nancy Van Dolsen +C -Page 9 +Free Electives / Optional Major / Optional Minor(s) - Figure 3: Tax Map, William Jones Arney House +54 +12 +36 +6 +- -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke County; November 1999 (cid:9) -Nancy Van Dolsen +96 +24 +42 +24 +6 -Page 10 +90 - Figure 4: Site plan, William Jones Arney House +Total -Historic Architectural Resources Survey Report,- NCD07; SR 1-114, Burke County: November 1999 (cid:9) -Nancy l'an Dolsen +240 -Page 11 +Find out more: +https://ug.hkubs.hku.hk/programme/becon-and-fin - — +Find out more: +https://ug.hkubs.hku.hk/programme/bfin-ampb -— (cid:9) +1312 -_ + Bachelor of Science in Marketing Analytics and Technology +[BSc(MAT)] -- (cid:9) +Bachelor of Science in Quantitative Finance [BSc(QFin)] -41E +Marketing is critical to all businesses. It dictates whether a company will be +successful. Technology, Big Data, and digitization are fundamentally changing +how marketers operate. The BSc(MAT) programme equips students with +quantitative skills, technology know-how, and business marketing acumen to +pursue a career in the digital economy and modern businesses. It targets both +STEM students with an interest in business as well as business students +interested in the technology space. Students will gain a solid foundation in +cutting-edge technical areas including computer programming, data science, +and technology by excelling in courses from the Faculty of Engineering and +Faculty of Science. In addition, students will master digital business, +marketing and leadership through courses at the HKU Business School. This +interdisciplinary approach will enable students to engage in a variety of jobs +and functions such as the design and launch of new technology products, +targeting consumers, digital marketing, forecast and predicting markets, and +building a technology-focused strategy. The programme aims to nurture +future leaders in the field of technology and marketing who are able to use the +latest technology, science, and data analytics to support business and +marketing decisions in the digital world. -re-7_. - — -liq -- -- . (cid:9) +BSc(MAT) Curriculum Structure + +Components + +Credits + +A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Marketing Analytics and Technology +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives (Marketing Technology and Strategy) +Disciplinary Electives (Marketing Analytics) +Capstone Course + +C + +Free Electives / Optional Major / Optional Minor(s) +54 +12 +36 +6 - -_ +96 +24 +42 +12 +12 +6 + +90 -_ (cid:9) +Total -_ +240 -_- +BSc(QFin) is a premium finance programme. Built on a solid foundation in +economics and finance, the programme equips students with the essential +quant skills, including mathematics, statistics, and programming, that are +required in more advanced and sophisticated analyses, such as valuation +models of structured products, risk management, simulation and data +analysis, financial engineering, and quantitative trading. -- --- +The programme goes beyond merely equipping students with quant skills. +More importantly, it stresses on combining these skills with an understanding of +the financial markets, the macro environment, and the practical situations +where financial managers work, with a balanced mix of quant and business +courses. To nurture students’ soft skills and market acumen, the programme +provides a wide range of value-enhancing activities, such as Round Tables with +Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training +workshops, simulation trading workshops, and an international field trip to +other international financial centres such as New York, London and Zurich. -— s- — +BSc(QFin) Curriculum Structure -_--..------ +Components +Credits + +A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Quantitative Finance +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + +C + +Free Electives / Optional Major / Optional Minor(s) + +54 +12 +36 +6 - -- (cid:9) -77--- +96 +24 +54 +12 +6 --_ +90 -- (cid:9) +Total -—. (cid:9) +240 -—______-- -- ------_--,- --- ---------- +Find out more: +https://ug.hkubs.hku.hk/programme/bsc-mat -_ +Find out more: +https://ug.hkubs.hku.hk/programme/bsc-qfin -_ --- — +1514 -- -- + Enrich Your Life --4-. - -, +initiative and +HKU Business School fosters a culture of student +implementation of ideas, both within and outside the campus, as it is +these real-world engagements that set us apart. Experiences in learning +beyond the classroom, extending into realms outside academia, enable +students to cultivate an understanding of social dynamics, cultural +diversity, and collaborative +intellect. These experiences contribute +significantly to their personal growth, interpersonal skills, and social +development, shaping them into global citizens, lifelong learners, and +future business leaders. --4 +Overseas Enrichment Experiences ----_ +Business Exchange Programme / +HKUWW Undergraduate Student Exchange Programme +Even as the youngest Faculty at the University, our exchange +activities are nothing but phenomenal. Each year, over 400 +students participate in overseas exchanges at leading universities +to broaden their horizons. -a +Business School +Exchange Partners * -Photograph 1: William Jones Amey House, facing southwest +Canada -Photograph 2: William Jones Amey House, facing northeast +McGill University +University of British Columbia +University of Toronto +Western University +Simon Fraser University -Historic Architectural Resources Survey Report: NCDOT SR 1424, Burke County: November 1999 -Nancy Van Dolsen +U.K. -Page 12 +King's College London +University of Manchester +University of Edinburgh +University of Warwick -(cid:9) -(cid:9) - . +The Netherlands -:t. (cid:9) +University of Amsterdam +Erasmus University Rotterdam -. --- +400+ +exchange partners +across 40+ countries -, +400+ +Business School +students on exchange -,,J-, (cid:9) +Denmark -, +Copenhagen Business School +Aarhus University -••••- (cid:9) +Germany -, (cid:9) +Frankfurt School of Finance & Management +University of Mannheim +Munich Business School -i +Belgium --- (cid:9) -_ -__ +KU Leuven --.------ (cid:9) +Italy -----.•-•_:, (cid:9) ----- +Bocconi University -(cid:9) ,- +France --.? +U.S. ---- -_= +Toulouse School of Economics +ESSEC Business School -- (cid:9) +New York University +University of Southern California +University of Washington +University of Georgia -_ +Middle East -- +Reichman University -. (cid:9) +Singapore --- (cid:9) -- (cid:9) +National University of Singapore +Nanyang Technological University -- -•- -_ --4--.- -— -- (cid:9) +* This list is non-exclusive. + +International Field Trips +To enhance students' exposure to global businesses, the School organises +overseas field trips for students visiting corporate firms and mingling with +C-suite Executives in some of the world's most prominent financial +centres, e.g. Australia, France, Japan, London, New York City, Singapore, +Spain, and Zurich. + +Visiting Programmes +Through the Yale Visiting International Students Programme, students +are sent to spend a year at Yale University, where they actively engage +themselves in a multicultural society. With the generosity of the donors, +some of our students are selected for the Junior Year Abroad Programme +and Visiting Students Programme to study as visiting students at +Pembroke College and St Peter’s College of University of Oxford +respectively with substantial scholarships. The intellectual ambience of +these top universities is truly inspirational and conducive to learning. -------,_ +Beta Gamma Sigma Regional Leadership Conference +Beta Gamma Sigma (BGS) is the international honour society that +serves business programmes accredited by the Association to Advance +Collegiate Schools of Business (AACSB International). With the +establishment of the HKU BGS Chapter, the School registers students +annually for the BGS Regional Leadership Conference held in the U.S., +where they can meet and bond with many other fellow students from the +world's best business schools. The Leadership Summit is eye-opening +and challenging, which leads to a great journey of self-discovery. -- +Japan + +Kyoto University + +Australia + +University of Sydney +Monash University +University of New South Wales + +1716 Student Sharing + +Chen, Hsuan-yu +Exchange at National University of Singapore (Singapore), 2024-25 + +As a third-year undergraduate student pursuing a double major in +Accounting and Finance, I consider it a tremendous privilege to have +participated in a one-semester exchange at the National University of +Singapore (NUS) through the Business Exchange Programme. This +invaluable opportunity has not only enriched my academic journey but +also broadened my global perspective through immersive cultural +experiences and meaningful cross-cultural interactions. As I move +forward in my academic and professional journey, I carry with me not +just the knowledge gained from my courses, but the invaluable life +lessons experienced from Exchange. + +Lee, Tsz To Derek +Exchange at King’s College London (United Kingdom), 2024-25 + +I am incredibly grateful to have this valuable opportunity to study +abroad at King’s College London. The programme provided me a chance +to venture outside of my comfort zone. Not only can you learn how to be +more independent, you can also boost interpersonal skills given the +many opportunities to socialise with others, which is essential for your +future career. The experience is remarkable. I think you will regret it if +you miss out on the overseas exchange programme. + +Cheung, Nga Ting Caroline +Exchange at University of New South Wales (Australia), 2024-25 + +This exchange programme has changed me in many ways. It has helped +me become more outgoing, allowing me to make new friends and +explore different cities. Living independently has been a significant part +of this journey. From exploring new cities to forming lasting friendships, +each adventure has enriched my life in ways I never expected. This +opportunity has not only broadened my horizons but also deepened my +appreciation for diversity and connection among people from different +backgrounds. Each memory I have created during this time is a treasure +that I will carry with me forever. + +Ho, Ying Tung Crystal +Exchange at Copenhagen Business School (Denmark), 2024-25 + +In the Fall semester, I participated in a business exchange programme at +Denmark’s Copenhagen Business School (CBS). This experience was +truly transformative, enriching me both academically and personally +while allowing me to forge lifelong memories and friendships. An +exchange experience abroad truly broadens your horizons and offers +insights beyond what textbooks and lectures can provide. You will gain +invaluable perspectives and undergo personal growth like never before. +I wholeheartedly encourage you to muster your courage and embark on +an exchange journey - you won’t regret it! + +Whitfield III, Gerald Bruce Trey +Exchange at New York University (United States), 2024-25 + +I found the exchange experience to be extremely enriching. A major +highlight for me was exploring the lifestyle and professional sides of the +city. Because of my initiative socially, and in groups such as church, I was +able to get well-acquainted with the local rhythms and lifestyle, and +also made many friends that I’d consider friends for a lifetime. I am +deeply appreciative of being given this rewarding experience, which +significantly shaped both my future and the person I am today. + +Ip, Wing Tung Alla +Student Representative for the BGS Regional Leadership +Conference 2025 in Highland Heights, Kentucky + +Attending the BGS Leadership Conference at Northern Kentucky +University proved transformative. I received profound insights into +leadership, communication and cultural intelligence through immersive +workshops, case competitions and networking opportunities. The event +prompted reflections on personal leadership styles and encouraged +collaboration across diverse peers. As I prepare to enter the workforce, +I intend to apply these learned principles of trust-building, cultural +sensitivity and collaborative leadership to navigate challenges so I can +shape my professional journey with purpose and authenticity. + +1918 Corporate Exposure + +HKU Business School has a strong tie with corporate partners. To nurture students' business +and market acumen, enrichment programmes are organised through the joint efforts of +corporate partnerships. + +Enrichment Programmes in collaboration with Corporate Partners +HKU Business School has been collaborating with corporate partners on initiatives to spice up +students' career exploration journey. The partnership with Accenture enables students to +gain exclusive insights into business consulting through the HKU-Accenture Business +Consulting Programme. + +To provide students with a unique opportunity to explore current business trends and +developments in the Greater Bay Area (GBA), our School has co-organised a tailored study tour +with The Association of Chartered Certified Accountants (ACCA). This initiative aims to +enhance students’ readiness for future career development and personal success. Students will +have the chance to connect with seasoned accounting and business professionals in the GBA. + +HKU Business School Mentorship Programme +Leveraging the School's extensive network in the business and finance industry, we have +developed our own mentorship programme to furnish students with opportunities to +enhance their skills and knowledge essential for their future personal and career +development. Our students are empowered to connect to the real business world through +direct and active interaction with seasoned professionals, executives and entrepreneurs. + +Lee, Lok Yiu Angel +HKU Business School Mentorship Programme 2024-25 + +My mentor offered +invaluable career advice, highlighting key +considerations for entering specific fields and preparing for interviews. +Her experiences and mindset during job searches resonated with me as +I navigated something similar. She emphasised the importance of +open-mindedness and eagerness to learn in the workplace. I kept her +words at heart during my internship. I also enjoyed catching up with her +to share my new job experiences. + +Choi, Han Na Hannah +HKUBS x Deloitte China Mentorship Programme 2024-25 + +As my mentor has been working in Deloitte China for more than 10 +years, I was most curious about how he maintains a work-life balance +and how he stays dedicated and professional at work. Through our +lengthy conversations over lunch, he passed on valuable insights on +developing soft skills and adopting an appropriate work attitude at an +early stage. He is more than an audit mentor but also a life coach to me. + +Chan, Chin Ching Jacky +BSc(QFin) Executive Mentorship Programme 2024-25 + +Throughout the mentorship programme, I have received +precious guidance that has greatly influenced my career +path. My mentor highlighted the importance of identifying +a company that aligns with my goals and would provide +opportunities for me to develop the necessary skill sets. He +also shared practical advice on building and managing +professional and personal relationships. These lessons +have broadened my perspective and also equipped me +with the tools to handle real-world situations more +effectively and confidently. + +Wong, Wan Tat +HKUBS x ACCA Greater Bay Area Study Tour 2024-25 + +My greatest takeaway from this study tour was the power of peer +learning and collaboration. Being immersed in an environment with a +diverse group of individuals, all focused on similar goals, created an +incredible opportunity to learn from each other’s experiences and +perspectives. The collaborative projects and informal discussions were +invaluable in broadening my horizon and developing new approaches +to the accounting field in Chinese Mainland. I am leaving with a strong +network of contacts and a renewed appreciation for teamwork. + +Leung, Ho Lun Maverick +HKU-Accenture Business Consulting Programme 2024-25 + +I visited + +The HKU-Accenture Business Consulting Programme immersed me in +an authentic consulting experience through working with a real client – +their distribution centres, conducted market +HKTV. +questionnaires, and presented the final deliverables - all within the +eight-week programme. I was honoured to meet Managing Directors +from Accenture who shared crucial skills in conducting a consulting +project. It’s truly one of the best programmes for anyone considering a +consulting career! + +2120 Experiential Learning Programmes + +Experiential learning is embedded in our course curriculum. The Business Consulting +Practicum offers business and economics students hands-on experience in working with +clients to solve "real-life" problems relevant to their studies, under the guidance of faculty +members and professional mentors. The Business Case Analysis and Presentation enables +students to communicate and seek practical advice from senior executives through in-class +case presentations. The Impact Lab provides students with an excellent opportunity to work +directly with social ventures, again under the guidance of teachers and professional mentors. +Additionally, students can gain an in-depth understanding of the latest trends in economics, +finance, creativity, innovation, and entrepreneurship in the rapidly growing Greater Bay Area +by participating in the Creativity, Innovation & Entrepreneurship in China Programme. + +Chaabane, Adam Andrea +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2025 + +I have learned that prominent businessmen and professionals in Hong +Kong are people like me who were once attending business school. I +most enjoyed the guest +lectures because they allowed direct +interactions with inspiring figures. The virtual experience enabled me to +follow this programme easily from my home country. I definitely +recommend it for a better understanding of Hong Kong and GBA’s +markets and opportunities that arise in cross-border business. It would +be great to learn more about opportunities for foreigners in the region. + +Kaur, Suchleen +Business Consulting Practicum 2024 + +The opportunity for real-time interaction and feedback with the client had +propelled me to join BCP. I was always curious about how a consultant +would address and communicate with the client. BCP has taught me +about proper business etiquette and equipped me the necessary skills to +negotiate conflicting opinions during team discussions. Throughout the +process, our mentor and professors provided great help and insights. +Their invaluable feedback and questions helped nudge us in the right +direction, taking a step back to look at the big picture and most +importantly, prioritise the client’s needs. + +Zhou, Qingzhuo +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2025 + +My greatest takeaway from the CIEC 2025 programme was the unique +window to gain a comprehensive understanding of the Greater Bay +Area’s dynamic business landscape. I explored the latest trends in +business, finance, and innovation from guest talks and virtual lectures +with industry leaders. I particularly appreciate the coverage on a +diverse range of topics and the opportunity to earn HKU credits. A +highlight was the cultural exploration segment which offered an +unparalleled perspective on the Greater Bay Area. I recommend this +programme without hesitation. + +Tsang, Tze Yeung Matthew +Business Consulting Practicum 2024 + +BCP has been a rewarding experience that allowed me to apply my +academic knowledge to real-world business challenges. Working +directly with my mentor and client has pushed me to think critically and +creatively in ways I hadn't experienced in the classroom. I'm confident +that the problem-solving, communication, and client-facing skills I have +developed will serve me well when pursuing my future career in +consulting or another business-oriented field. + +2322 Cheung, Ho Ning Jolie +UW Global Business Case Competition 2025 (Champion) + +in UW GBCC was an + +Competing +incredibly rewarding and +unforgettable experience. We had the opportunity to work with +participants from different universities. +It was a fantastic +opportunity to exchange perspectives and learn from different +problem-solving approaches. Working under tight timeframes +pushed us to think creatively. UW GBCC has not only elevated my +professional capabilities but further solidified my belief in business +as a powerful vehicle for creating positive change in the world. + +Case Competitions + +The School provides ample opportunities to refine students' presentation and analytical skills +through a myriad of local as well as overseas competitions. From training to delicate analysis +of business cases to a variety of business scenarios, students are equipped with different +skillsets to compete with candidates from the world's leading universities. + +Lo, Nikkia +NHH International Case Competition 2024 (Champion) + +My week at NHH International Case Competition has been an +unforgettable rollercoaster of emotions. From the excitement of +exploring a new Nordic city, the exhaustion of our 24-hour case +competition, to the nervous anticipation of pitching on a global stage +with a +livestream, every moment was extraordinary. Meeting +like-minded peers around the world was an incredible experience that +forged new friendships. Thank you, HKU Business School, along with +our professors, coaches and staff for bringing this opportunity to us! --_ -_.----e- --, (cid:9) - (cid:9) --- (cid:9) +Xia, Yunchu Kevin +The McDonough Business Strategy Challenge 2025 (First Place) -l'---0` (cid:9) +Winning first place at MBSC2025 was incredible and deeply rewarding. +My team was highly engaged. We were constantly bouncing ideas off +each other and refining our strategy. There were moments where we hit +roadblocks, but our team rallied, challenged each other constructively +and developed a cohesive solution. Connecting with like-minded +participants, judges and NGOs was also invaluable. MBSC not only +sharpened my analytical and presentation skills but also reaffirmed my +commitment to leveraging business for good. I am so grateful for the +experience and very eager to apply these lessons moving forward! -_ (cid:9) -_ +2524 Liu, Yunqi Iris +Deloitte Tax Championship 2024 (Outstanding Team Award) -_ +This competition has been a fantastic learning experience. It +showed me that tax isn’t just about the technical details - it’s also +deeply connected to social trends and global events like climate +change, geopolitical conflicts and national policies. This broader +perspective has enhanced my understanding of the industry and +its importance in today’s world. I want to express my sincere +gratitude to my advisors for their guidance and recognition of my +team’s hard work. I am grateful for the opportunity to learn and +grow with such an amazing group of people. -_ +Pan, Ho Man Vivian +HKGCC Business Case Competition 2024/25 (Champion) -- (cid:9) +in -- (cid:9) +the HKGCC Case Competition was a -- +Participating +truly +transformative experience that greatly enhanced my strategic +thinking, teamwork and problem-solving skills. The competition +challenged me to analyse intricate business problems, think critically +under pressure and develop innovative solutions. Working closely +with people of various majors allowed me to appreciate differing +opinions that were at times conflicting, find ways to build consensus +and synthesize ideas effectively. -- --7---- - --__-- (cid:9) -------- -- (cid:9) +Students' achievements in case competitions +(2023-2025 awards) -- (cid:9) - (cid:9) +Overseas / Chinese Mainland -- -- +Asia Pacific Sustainable Marketing Case Competition 2025 -- (cid:9) +Third Place -- -- - -.-• =,---• -.-__----------- +Bloomberg Global Trading Challenge 2023 -- (cid:9) - (cid:9) +CBS Case Competition 2023 -- +CFA Institute Research Challenge 2025 - Asia Pacific Regional Final -, +CFA Institute Research Challenge 2025 - Sub-Regional Final -t t -, (cid:9) -- (cid:9) -- (cid:9) -= (cid:9) +CGMA Global Business Challenge 2023 - North Asia Final -_..----, (cid:9) +Champions' Trophy Case Competition 2024 -. (cid:9) +Grand Prize Winner +Regional Prize Winner: Asia & Oceania -_ +Winner -- (cid:9) +Winner -Photograph 3: William Jones Arney house, view facing northeast +Winner -Photograph 4: Granary, view facing north +Champion -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 (cid:9) -:Vancy Van Dolsen +Finalist -Page 13 +China Mainland, Hong Kong and Macao Digital Economy Innovation and +Entrepreneurship Competition 2024 -(cid:9) - " -. -• +2nd Runner-up (Gold Medal) -: -• -• +Deloitte Digital Camp 2024 -0 +Deloitte Digital Camp 2023 -. +Second Place -0 +Champion -c -e -) +Deloitte Tax Championship 2024 - National Round -z +Outstanding Team Award -C +Deloitte Tax Championship 2024 - Regional Round -. -) +Deloitte Tax Championship 2023 - National Round -E +Deloitte Tax Championship 2023 - Regional Round -- -: - -6: -1 - -;1 - -8 - -4 -. -1 - - Marshall Arney House +Top 3 teams -The Marshall Arney House is located on the west side of SR 1424. +Merit Award -Location: (cid:9) -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. +Top 3 teams -The Marshall Arney House stands in a clearing of a wooded area +Family Office Association Hong Kong (FOAHK) 2023 Case Competition -Setting: (cid:9) -halfway up a hillside on a knoll approximately 100 feet wide in a bend of the -Johns River. Johns River Road is located on the next downhill terrace. East of -the roadway the land slopes down to the river. +2nd Runner-up -Description: +HSBC Global Private Banking Case Challenge 2023 -Dwelling: Built ca. 1900, the Marshall Arney House is a two-and-a-half- -story, T-shaped house and features a plan similar to his father's house, the -William Jones Arney House. The primary elevation faces south. The -Marshall Arney House has a wrap-around porch with chamfered posts. -Similar to the William Jones Arney House, the Marshall Arney House has -pointed arched louvred windows at the peaks of the end gables. The end -gables are pedimented. Two early twentieth-century additions were -constructed onto the north gable end; a two-story gable wing, and a one- -story shed-roof ell. The building is weatherboarded, and retains most of -its original two-over-two sash windows. +Investing in the Future University Competition 2023 +by Pictet Asset Management -Springhouse: To the west of the house stands a gable-front springhouse -banked into the side of the hill. The springhouse has a stone lower level -with a stuccoed finish. The interior of the upper floor is sheathed in -diagonal boards, and the exterior is weatherboarded. The gable roof is -cantilevered over the façade of the springhouse. +John Molson Undergraduate Case Competition 2025 -Background Information: (cid:9) -Constructed by Marshall Arney on land bequeathed -to him by his father, William Jones Amey, this house is a substantial dwelling for -the area. Marshall Arney was a land agent for Duke Power, served for twenty- -five years on the Burke County School Board and was instrumental in -consolidating and erecting new schools in the county during the early twentieth -century. In April 1922 Marshall Arney and his wife, L. Inez Arney, sold 250 -acres, including the subject structure, to Western Carolina Power. +Lazaridis International Case Conference 2023 -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, the Marshall Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A. since it is not representative of a significant trend in the history of the -area. +Marshall International Case Competition 2024 -historic Architectural Resources Survey Report: NCDOT. SR I424. Burke County; November 1999 (cid:9) -Nancy i'an Dolsen +McDonough Business Strategy Challenge 2025 -Page 15 +NHH International Case Competition 2024 - The Marshall Amey House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. +Scotiabank International Case Competition 2023 -The Marshall Arney House is considered eligible for the National Register under -Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is found in the region as a house type built by farmers of middle -economic status and part-time farmers with jobs off of the farm. This type of -farmhouse, if ever once prevalent, has become rare due to the large tracts of -former agricultural land that has been acquired for forest and timber land. The -Amey House is a good representation of this type of farm house and exhibits -integrity of materials, workmanship, setting, location, feeling, and association. -The springhouse building is also representative of a vernacular building form and -exhibits good architectural integrity. +Sydney International Business Competition 2025 -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. +Thammasat Undergraduate Business Challenge 2024 -Boundaries: The proposed National Register boundaries comprise -approximately 1.9 acres that surround the house and its accompanying -outbuilding. The land was deeded to Marshall Arne y at death of his father. This -tract is part of parcel JR-18, which is a portion of the larger Carolina Centers -tract. +The Third KPMG ESG Case Study Competition -Boundary Justification: (cid:9) -include the house. outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. +UW Global Business Case Competition 2025 -The proposed historic boundaries were drawn to +UW Global Business Case Competition 2023 -Historic Architectural Resources Survey Report: ACDOT SR 1424. Burke County: November 1999 (cid:9) -Nancy Ilzn Dolsen +1st Runner-up -Page 16 +2nd Runner-up - Figure 5: Tax Map, Marshall Arney House +Finalist -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke Counr.t...,- November 1999 (cid:9) -Nancy ran Dolsen +1st Runner-up -Page 17 +Finalist - L -T +First Place -. - -m -u +Champion -l - -t - -4 -C +Winner -C -a -) +Third Place -' -• +1st Runner-up -C +Third Prize -• +Champion -• -• -• -• -• +Finalists -• -• -• +27 -c -. +26 Local -.VC1)Or SR 1414, Bur ke Con +Local - :Vovember 19 99 +100WF Hong Kong JumpStart University Investment Hackathon 2023 - Corpening House +1st Runner-up -The Corpening House stands on the west side of SR 1424. about +HKSI Institute Case Competition 2024 -Location: (cid:9) -1.5 miles south of the junction with SR 1405. The house sits on a 23,136.10-acre -tract owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a -timber company. The property was formerly owned by Duke Power, Inc. +Champion -The Corpening House stands on a slight rise overlooking the wide +ACCA Hong Kong Business Competition 2023 -Setting: (cid:9) -bottomland along a bend in the Johns River. The land around the house is now -used to grow trees and shrubs for sale to nurseries and landscaping companies. +Outstanding Team for start-up Filix +Medtech -Description: +Team Excellence Award -The ca. 1900 features a common form for the period: a two- +Merit Award for start-up MakeOmnia -Dwelling: (cid:9) -story, square, boxy building with a pyramidal roof. The facade features a -large central pediment over the second-floor windows, and a smaller -pediment covers the center of the one-story porch across the facade. -Turned posts support the porch roof. The exterior of the house is clad in -narrow weatherboards. The window sashes have been replaced with one- -over-one double-hung sash. Since 1983, the porch has lost its decorative -brackets, and the back porch has been removed andior remodeled. A shed -roof addition covers the rear elevation. No outbuildings stand on the -property. +Alvarez & Marsal Hong Kong Business Case Competition 2024 -Background Information: (cid:9) -the holdings of John Perkins. one of the county's first settlers of European -descent. The property is now part of the large landholdings of the Carolina -Centers corporation. +Champion & Best Presenter -The Corpening House stands on land once part of +HKSI Institute Case Competition 2023 -For purposes of compliance with Section 106 of the National +Best Performer of Roma Risk Advisory +Limited -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, the Corpening House is considered not eligible for the National -Register ofl-listoric Places under Criterion A, since it is not representative of a -significant trend in the history of the area. +1st Runner-up -The Corpening House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. +Merit -The Corpening House is considered not eligible for the National Register under -Criterion C since the building has lost integrity and does not embody distinctive -characteristics of a significant building type. The building no longer retains its -original porch trim, window sash, and the form has been compromised by the -alterations to the rear of the dwelling. +Champion -Historic Architectural Resources Survey Report: NCDOT1SR 1424, Burke County: November 1999 (cid:9) -Nancy lint Dolsen +1st Runner-up -Page 21 +Merit & Finalist - The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - +Semi-finalist -Historic Architectural Resources Survey Report; NCD071SR 1424, Burke County: November 1999 -Nancyl'an Dohen +Champion -Page 22 +2nd Runner-up - Photograph 1: Corpcning House, facing east +Best Presenter -Photograph 2: Corpening House, facing north +1st Runner-up -Historic Architectural Resources Survey Report: NCD07; SR 1424, Burke County: November 1999 (cid:9) -Nancy ("an Dohen +HKU-Airstar Bank Joint University Case Competition 2024 -Page 23 +HKU-SaSa Joint University Case Competition 2023 - Early Twentieth-Century Frame House +Hong Kong MBS Digital Economy Business Simulation Competition 2025 -This frame house stands on the east side of Corpening Bridge +Third Place -Location: (cid:9) -Road. SR 1438. approximately 1.5 miles east of the bridge over the Johns River. -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. +HSBC Global Private Banking Case Challenge 2024 -Setting: (cid:9) -outbuildings stand south of the house, down a slope, and are no longer used. +HSBC Life Insurance Competition 2025 -This frame house sits in a wooded area that was once farmed. The +HSBC Life Insurance Innovation Competition 2023 -Description: +1st Runner-up -A two-and-a-half story, three-bay, double-pile dwelling, +Top 3 Winner -this house features a hip roof with a central gable dormer. The house has -an exposed extended rafter roof and is covered with weatherboard. A one- -story hip-roof porch covers the facade of the building. The house retains -its six-over-six sash windows. A one-story gable roof ell is located on the -rear of the house. +1st Runner-up -Outbuildings: To the east of the house stands a frame domestic -outbuilding. The gable-front building stands on a cement block -foundation. and is covered with weatherboard. An off-center door is -located on the gable end. To the south of the house are three frame -outbuildings in poor condition. The outbuildings have metal roofs and -narrow board siding. The buildings appear to have served as a granary, -stable, and a workshop. +Best Use of Technology Award -For purposes of compliance with Section 106 of the National +International MBS Business Management Simulation Competition 2023 -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, this house is considered not eligible for the National Register of -Historic Places under Criterion A, since it is not representative of a significant -trend in the history of the area. +Champion -This house is considered not eligible for the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. +ISM-HK Student Case Competition 2025 -This house is considered not eligible for the National Register under Criterion C -since the building does not embody distinctive characteristics of a significant -building type. The outbuildings have lost their architectural integrity and their -setting. +ISM-HK Student Case Competition 2024 -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. +MBS Simulated Business Management Competition 2024 -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen +New Century Elite Development Programme 2023 -Page 24 +Venture Capital On Campus (VCOC) - Investment Challenge 2023 - Photograph I: Early Twentieth-Century Frame House, facing northeast +1st Runner-up -Photograph 2: Early Twentieth-Century Frame House, facing north +Champion -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Do/sell +Third Place -Page 25 +First place - Photograph 3: Frame Domestic Outbuilding, facing northeast +Best presenter -Photograph 4: Frame Granary, facing south +Third Place -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 -Nancy I'an Dolsen +Champion -Page 26 +1st Runner-up - 0 +2nd Runner-up -t -t -j +Winner Team -0 -. +29 -k -. -r -; +AmplifyME APAC M&A APAC Challenge 2025 -C +Third Place -I -) +BOCHK Challenge 2023 -- -0 +Bronze Award for Innovative Design -. -2 -4 +CGMA Global Business Challenge 2025 (Regional Final) -- -a +1st Runner-up -r -/ -1 +CGMA Global Business Challenge 2023 (Semi-final Round) -0 +Deloitte USHINE Case Competition 2024 -C +ESG Challenge 2023 + +EY Young Tax Professional of the Year 2023 - Hong Kong Region + +Champion + +Finalist + +Champion + +Champion + +1st Runner-up + +2nd Runner-up + +Family Office Association Hong Kong (FOAHK) 2025 Case Competition + +Champion + +HKCGI Corporate Governance Paper Competition and +Presentation Awards 2024 + +HKGCC Business Case Competition 2023/24 + +HKGCC Business Case Competition 2024/25 + +28 + +HKGCC Pitch Perfect Programme 2024 + +HKICPA Business Case Competition 2023 + +HKSI Institute Case Competition 2025 + +1st Runner-up + +1st Runner-up (Paper Writing Competition) + +1st Runner-up (Paper Presentation +Competition) + +Audience’s Favourite Team Award + +Chevalier - Champion + +Chinachem Group - Champion + +Swire Pacific - Champion -. -. +Chevalier - 1st Runner-up -, +Chinachem - 1st Runner-up -0 +HKGCC - Champion -1 -1 -) +Jardine Matheson - Champion -November 1999 +SOCAM - Champion - Nineteenth-Century Log House +Jardine Matheson - 1st Runner-up -Location: (cid:9) -0.25 miles south of the intersection of SR 1424 and SR 1405. +MTR - 2nd Runner-up -This log house stands on the west side of SR 1424, approximately +Overall Champion -The log house sits approximately 40 feet from the road, and has +Winner -Setting: (cid:9) -two mid-twentieth century outbuildings standing to the rear of the house. To the -north of the log house stands a late twentieth-century residence. +2nd Runner-up -Description: +Champion -The one-story, side-gable, three-bay log dwelling is +1st Runner-up -Dwelling: (cid:9) -constructed with half-dovetail corner notching. The house features a -variety Of wood sheathing, including narrow weatherboards and novelty -siding. Windows in the building have been enlarged and have six-over-six -sashes. The porch on the facade has also been replaced. A one-story -frame addition was constructed onto the rear of the dwelling. An exterior -brick chimney is located on the south elevation. +2nd Runner-up -Outbuildings: Two frame outbuildings stand south of the house: a frame -chicken house with a shed roof and a small, one-story frame building with -an entry on the gable end. +Merit -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, this house is considered not -eligible for the National Register of Historic Places under Criterion A, since it is -not representative of a significant. trend in the history of the area. + Chhawchharia, Mihika Amit +HSBC - Summer Analyst program -This house is considered not eligible fir the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. +internship, -This house is considered not eligible for the National Register under Criterion C -since the building has lost integrity and does not embody distinctive -characteristics. The building no longer retains its original porch or windows and -the form has been compromised by the addition to the rear of the dwelling: +I collaborated closely with various +Throughout my +departments of the bank and +interacted regularly with senior +management, gaining comprehensive exposure to the bank’s operations. +This experience allowed me to hone my professional skills, including +communication, leadership, financial analysis, and market expertise. I +was honored to be ranked as the top intern in my cohort and secured a +return offer to join HSBC as a full-time graduate analyst in their Global +Banking division. I am deeply grateful for the knowledge and support I +received at HKU, which enabled me to thrive in a prestigious global +institution alongside some of the brightest talents in the industry. -The building is considered not likely to yield information important to the history -of building technology; therefore it is not eligible for the National Register under -Criterion D. +Internship Experience -Historic Architectural Resources Survey Report: NCDOT, SR 1424. Burke County: November 1999 -Nancy l'art Dolsen +Lee, Chung Hong Calvin +HSBC - Global Risk Management Intern -Page 28 +Reflecting on my time at HSBC, I realise that I have grown on a +professional and personal level. I have developed a deeper passion for +the banking industry and a greater appreciation for the critical role that +risk management and compliance play. I wore many hats and gained +precious insights into the banking industry. I was responsible for +overseeing compliance risks across 32 markets in Asia Pacific. I am +confident that the lessons I have learned at HSBC will guide me every +step of the way. - Photograph 1: Nineteenth-Century Log House, facing northeast +Hui, Matthew +HSBC - International Banking (ISB) Intern - Corporate Coverage -Photograph 2: Nineteenth-Century Log 1-louse facing southwest +The comprehensive exposure to Corporate Banking was invaluable. From +analysing credit risk for lending applications to understanding diverse +product offerings for MNC clients, each day brought new learning +opportunities. I'm truly grateful for the experience of shadowing and +supporting different RMs in the team. Together with the sharing and +introductions from product partners, I learned how to identify clients' +needs and gained perspective on effectively reaching out and pitching to +clients. This experience taught me how to view banking relationships +through a strategic lens - understanding not just immediate needs but +trends. The +also anticipating +professional mindset, strategic thinking, and collaborative spirit fostered +during this internship will undoubtedly shape my future career path. -Historic Architectural Resources Survey Report: .VC DOT SR 1424, Burke County: November 1999 -Nancy I an Dolsen +future requirements and market -Page 29 +Nagar, Tanvi +HSBC - International Banking Intern - Photograph 3: View of Chicken Coop, facing northeast +immense + +My internship experience was nothing short of incredible. Not only did I +grow immensely in my technical and professional skills but also +underwent an +journey of personal growth. I received +unparalleled support from the team and my mentors and gained an +in-depth insight into the field of banking and finance in Hong Kong. I +formed meaningful connections and grew my network. This experience +has truly shaped me and continues to propel my drive to pursue a career +in Banking in the years to come. + +3130 Sharma, Seerat +KPMG - Financial Services Audit Intern +(COOP3000 Cooperative Education Programme) + +I had the privilege of contributing to audit engagements for leading +clients in the banking and insurance sectors, where I conducted +substantive testing, supported risk assessments and navigated audit +methodologies using advanced tools. This role sharpened my analytical +mindset, strengthened my collaboration skills and offered a front-row +view into the dynamic world of financial auditing. It’s been a truly +rewarding journey that has solidified my passion for a career in +professional services. + +Xu, Aijing +CMB International Capital Corporation - Wealth Management Intern + +During my internship, I actively engaged with the product team in +market analysis and client pitching focused on mutual funds. I gained +valuable insights into various asset classes and macroeconomic +trends. Analyzing mutual fund performances and participating in +client meetings allowed me to contribute to investment portfolio +strategies. Collaborating with +team enhanced my +communication skills and understanding of client needs. Overall, this +experience deepened my knowledge of wealth management and +strengthened my passion for finance. + +the sales + +Shi, Yu Cheung Karson +Bank of China - Credit Summer Intern (Risk Management) + +During the internship, the financial techniques and business knowledge +I learned from HKU Business School courses helped a lot in company +financial analysis. By applying knowledge taught in the financial +accounting module, I assisted in evaluating loan demand. Through +cross-departmental collaborations, I also realised how classroom +simulations mirrored real-world risk mitigation challenges. This +experience reinforced the value of integrating academic theories with +dynamic market practices. + +Hu, Wen Qing +PwC Assurance Practice Intern + +As an assurance intern in PwC, I assisted my seniors with sampling +financial statement line items, reconciling balances, and documented +relevant non-financial information. It showed me how demanding the +auditing industry was, and allowed me to set my expectations for a +career in auditing as an accounting student. + +Wirawen, Karen +HKU – Neuberger Berman Private Equity Internship Programme + +You never truly understand private equity until you’re immersed in it +and my internship at Neuberger Berman was deeply enriching. Coming +in with limited knowledge, I gained direct exposure by working on +various deals across sectors and geographies, including Japan, Korea +and Southeast Asia. From drafting memos for the +investment +committee to attending annual general meetings and taking notes on +calls, I progressed from simple tasks to meaningful responsibilities. The +seniors were welcoming, always open to teaching, providing feedback +and offering valuable career advice. + +Chung, Marcus Lok Sum +Morgan Stanley Industrial Placement Intern + +During my internship with Sales & Trading (Institutional Equity +Division), I had the amazing opportunity to rotate across trading +and structuring desks where I was exposed to the fast-paced +environment especially during periods of high market volatility. To +see how traders manage positions and mitigate risks first-hand +was incredibly insightful. I also contributed to constructing trading +strategies and seeing the tangible impact of our work was deeply +rewarding. Morgan Stanley definitely provided me with a broader +perspective and understanding of market dynamics, liquidity, +client interactions and cross-asset opportunities. + +3332 Our Student Scholars + +Chan, Kwan Yin Cheryl +Recipient of Louisa Cheang Inspiration Scholarship 2024-25 + +I would like to express my gratitude for the Louisa Cheang Inspiration +Scholarship, which has significantly contributed to my personal and +professional growth. In the past academic year, this scholarship enabled +in the full-time Registered Visiting Student +me to participate +Programme at St Catherine’s College, University of Oxford. This +immersive experience enhanced my legal studies through engaging +discussions with scholars and participation in mooting competitions, +sharpening my research and advocacy skills. Additionally, I plan to +utilise this scholarship to pursue overseas postgraduate studies, +allowing me to deepen my understanding of the complexities of law in +a global context. + +Hussain, Iman +Recipient of FBE Student Contribution and +Innovation Award 2024-25 / +HKU Business School Future Leader Scholarship 2024-25 + +As someone who believes in learning through real-world experience, +the FBE Student Contribution and Innovation Award and the HKU +Business School Future Leader Scholarship have been life-changing. +They not only enabled me to go on an exchange semester to Cardiff, +where I explored new cultures and perspectives but also shaped my +entrepreneurial journey. + +With this support, I’ve been able to work on building a social +enterprise that aligns with my passion for creating meaningful impact. +I’m incredibly thankful for these opportunities to grow and make a +difference. + +Chow, Yee Shun Eason +Recipient of the Jardine HKU Scholarship 2024-25 + +It is an incredible honour to have received the Jardine HKU Scholarship +which is enriching my university journey! This scholarship allowed me to +pursue my many goals – from attending international training camps as +a handball player to participating in overseas exchanges without +financial worries. I have had the privilege of undertaking a summer +internship at the Jardine Group, learning from seasoned professionals in +my future industry. The Jardine Foundation also hosted exciting +networking activities and site visits which helped me build lasting +connections with fellow scholars in a vibrant community. + +Shih, Pui Hei Iris +Recipient of the Hong Kong Jockey Club Scholarships - +Undergraduate Scholarship 2024-25 + +This scholarship has been transformative to my university life as it has +allowed me to fully embrace and explore the world beyond the +classroom. It provided the financial stability I needed to focus on my +studies and presented opportunities for me to participate in diverse +extracurricular activities. These experiences broadened my perspective +and opened me to different points of view. -Photograph 4: Domestic outbuilding, facing north +Nurgali, Adilzhan +Recipient of BOC Hong Kong - +One Belt One Road Scholarship 2024-25 -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nwtcy i in Dolsen +It means a lot to me to be awarded the BOC Hong Kong - One Belt One +Road Scholarship. It was not +just about financial support but +recognition of the hard work I have put in. As an international student +soon heading for exchange, this award gives me the freedom to explore +new cultures and experiences without extra stress. I am truly grateful to +be part of such an inclusive and supportive community. -Page 30 +3534 2026 JUPAS Admissions - BIBLIOGRAPHY +Admissions criteria (HKDSE) -Arney, Marshall House. Historic Resource Survey Form, Burke County, 1983. -Asheville, NC: Western Office. Division of Cultural Resources. +Local candidates applying on the strength of the Hong Kong Diploma of +Secondary Education (HKDSE) are required to fulfil the minimum entrance +requirements for HKU Business School programmes as shown below: -Arney, William Jones House, Historic Resource Survey Form, Burke County, 1983. +Level 4 +Level 4 -Asheville, NC: Western Office, Division of Cultural Resources. +EngEng ≥ +Level 4 or above in English Language +ChinChin ≥ +Level 3 or above in Chinese Language1 +MathMath ≥ +Level 3 or above in Mathematics -Biographical Directory of the United States Congress, 1774-Present, Thomas Burke,- +Level 3 +Level 3 -Website: Bioguide.conress.gov. +Level 3 +Level 3 -Burke County Deeds, Burke County Court House, Morganton, NC. +AA -Burke County Plat Book, Burke County Court House, Morganton, NC. +CSDCSD = +Grade “Attained (A)” in +Citizenship and Social Development2 -Corpening House, Historic Resource Survey Form, Burke County, 1983. Asheville,NC: +2x -Western Office, Division of Cultural Resources. +Elective +Elective +Subjects +Subjects -Cotton, J. Randall, Suzanne Pickens Wylie, and Millie M. Barbee. Historic Burke: An -Architectural Sites Inventory of Burke County. Mortzanton, NC: Historic Burke -Foundation. Inc.. 1987. +≥ -Phifer, Edward W. Burke: the History of a North Carolina County, 1777-1920. +Level 3 +Level 3 -Morganton, NC: by the author, 1977. +Level 3 or above in two Elective Subjects -Historic Architectural Resources Survey Report: NCD01; SR 1424, Burke County: November 1999 (cid:9) -Nancyl'an Dolsen +For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) +In addition to the minimum entrance requirements above, candidates applying for the BBA(BA), +BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the +HKDSE examination: -Page 31 +BBA(BA) / BSc(MAT) +≥ +Level 3 or above in one of the following subjects: - APPENDIX A +Level 3 -NCDOT, SR 1424. Bur ke Cou +Biology -November 1 999 +BAFS - North Carolina Department of Cultural Resources +Chemistry -James B. Hunt Jr., Governor -Betty Ray McCain, Secretary +Economics -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 +ICT -July 9, 1999 +Physics -Stephen A. Moore, PE -District Engineer -NCDOT -Route 1 Box 169C -Marion NC 28752 +BBA(BA) -Division of Archives and History -Jeffrey J. Crow, Director +BSc(MAT) -LOCATION -507 North Blount Street -Raleigh, NC -State Courier 53-31-31 +Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not +strictly required. -Re: (cid:9) Widen and pave SR 1438 Corpening Bridge Road, +BBA(Law)&LLB / BBA(IBGM) / BSc(QFin) -Burke County, ER 99-9229 +BBA(Law)&LLB -Dear Mr. Moore: +BBA(IBGM) -We have received notification of a nationwide permit for the above project and would -like to comment. +BSc(QFin) -We have reviewed the permit application for the improvement and paving of SR 1438 -Corpening Bridge Road and located the following structures of historical or -architectural importance within the general area of the project: +EngEng ≥ -William Avery House, located at the northwest corner of the junction of SR 1438 -and SR 1424. +Level 5 +Level 5 -Arney's Chapel, located at the east side of SR 1423 south of its junction with SR -1439. +Chin, Math +Chin, Math ≥ -We recommend that an architectural historian with the North Carolina Department of -Transportation evaluate these sites to determine if they are within the area of potential -effect for the permit and are eligible for listing on the National Register of Historic -Places. +Level 4 +Level 4 -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. +EngEng ≥ -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. +Level 5 +Level 5 --Sincerely, ++ -1- (cid:9) ++ -• (r-N — +M1/M2 +M1/M2 ≥ -Da.vicl'Brook -Deputy State Historic Preservation Officer +Level 3 +Level 3 -DB:slw +MathMath ≥ -109 East Jones Street • Raleigh, North Carolina 27601-2807 +Level 4 +Level 4 - cc: (cid:9) +Application -Army- Corps of Engineers, Asheville -William D. Gilmore +For details of the 2026 JUPAS exercise, please visit www.jupas.edu.hk. -t-Barbara Church +Programme Scoring Formula - AUG-16—":39 MON 10:10 ID: +For more information on the Programme Scoring Formula and Subject Weightings, please +visit the Admissions Office’s website at https://admissions.hku.hk/apply/jupas. -TEL NO: +Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis.Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered.Level 5 or above inEnglish LanguageLevel 4 or above inChinese Language1 & MathematicsLevel 5 or above inEnglish LanguageLevel 4 or above inMathematicsLevel 3 or above in Extended Module 1 or 2 of Mathematics3736 2026 Non-JUPAS Admissions -351; P02 — +Selection process and entrance requirements -North Carolina Department of Cultural Resources +Eligibility +(1) Local students applying for admission on international qualifications other than the -Jams B. Hunt Jr.. Governor -Betty Ray McCain, Secretary +HKDSE examination; -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 +(2) Non-local students with international qualifications other than Chinese Mainland's -August 10, 1999 +National Joint College Entrance Examination (NJCEE/Gao Kao). -Stephen A. Moorc -District Engineer -North Carolina Department of Transportation -Route 1, Box 169C -Marion NC 28752 +Application Procedures +All non-JUPAS applications should be submitted via the online application system by noon of +November 26, 2025 (Hong Kong Time) for the first-round evaluation. Scanned copies of the +following documents should also be uploaded by the application deadline. -Re: +(1) Official prediction of grades or scores from your current high school (if available); -Paving SR 1414, Johns River Road, Burke County, -ER 00--7154 +(2) All internal and public examination results during your senior secondary education; and -Division of Archives Bnd History -Jeffrey 1. Crow, Director +(3) Academic reference letter (optional) -LOCATION -507 North Blount Street -Raleieh, NC -State Courier 53-31-31 +Late submission of supporting documents will delay the consideration of application. -it- , - have received notification of a nationwide permit for the above project and would +Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other +programmes will be arranged if necessary. -i%1'- +University and Programme Entrance Requirements +Due to the complex combination of education background of applicants, each application will +be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the +standards and requirements on the website of the Admissions Office of the University before +making an application: -Dear Mr. Moore: +(1) Admissions standards (International Qualifications or National Qualifications); -We -like to comment. +(2) Programme specific subject requirements; and -We have conducted a search of our maps and files and have located the following -structures of historical or architectural importance within the general area of the -project: +(3) English language requirement -qX.117. +In particular, please note that the expected lower boundary for admissions listed on the +website is provided for reference purposes only. Admissions is competitive in nature and +achieving the expected lower boundary does not guarantee admission to the programme. +The actual admission grade or score will be dependent on the overall quality of applications +received in the 2026 admissions exercise. -Marshall Arney House, located on thc wcst side of SR 1424, 0.5 mile northwest -of its junction with SR 1438. +Announcement of Result +Offer results will be released from December 2025 onwards. All application status will be +updated in the online application system once an admissions decision has been made. +Applicants should check their application status online regularly. -William Arney House, located at the northwest corner of the SR 1424 and SR -1438 junction. +More Details +For further information about the application procedures and admissions +standards, +at +https://admissions.hku.hk/apply/international-qualifications. -9;0 ck +the Admissions Office’s website -Corpening House, located on the west side of SR 1424, L5 mile south of its -junction with SR 1405. +please -We recommend that an architectural historian with the North Carolina Department of -Transportation identify and evaluate these properties and any other structures over -fifty years of age within the project area, and report the findings to us. +visit -An important archaeological site, a mound reported by J. W. Spainour and Cyrus -Thomas in the late 1800s, is recorded between the Johns River and Johns River Road -south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 311310 highlighted in -yellow. +2026 Dual Degree Admissions -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan +Separate application is required for applying to the dual degree programmes, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. -109 Eist Jones Street • Ra1e10, North Cirolina 27601-2807 +For details of the dual degrees and the application procedures, please visit +the Admissions Office’s website at +https://admissions.hku.hk/programmes/university-collaborative-programmes. - AUG-16-'99 MON 10:11 ID: (cid:9) +3938 Develop Your Career -TEL NO: (cid:9) +Freshmen Sharing -p355 P03 +Upon graduation, over 82% of our students begin their careers in the commerce and industry sector. Others either +move on to further studies or explore opportunities with the civil service, education institutions and +community/social services. -Stephen A. Moore -August 10, 1999, Page 2 +Career Prospects -sheets 8,9, 12. 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. +Companies/organisations that have offered full-time positions to our graduates include but not limited to the +following: -end south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 31BK3 highlighted in -yellow. +Accenture Company Limited +Bank of China (Hong Kong) Limited +BDO International Limited +Bloomberg L.P. +Citibank (Hong Kong) Limited +Crédit Agricole Corporate and Investment Bank +Credit Suisse Group AG +Deloitte Advisory (Hong Kong) Limited +Deloitte Touche Tohmatsu Limited +Deutsche Bank +Ernst & Young +Goldman Sachs (Asia) LLC +Hang Seng Bank Limited +HKSAR Government +Hong Kong Monetary Authority +JPMorgan Chase & Co. +KPMG +Morgan Stanley +PricewaterhouseCoopers Limited +Standard Chartered Bank (Hong Kong) Limited +The Hongkong and Shanghai Banking Corporation +Limited (HSBC) +UBS Group AG + +Full employment +for the 19th + +consecutive year + +98.7% + +was the average salary of +our 2024 graduates + +$36,770 + +of our 2024 graduates have +received at least 2 job offers +upon graduation + +57% + +Kantbek, Yerassyl +BBA + +I chose HKU Business School because Hong Kong is +one of the world’s most competitive financial centres, +offering a dynamic environment to grow. HKU +combines +real-world +rigorous academics with +experiences, such as Asia-focused case studies and +competitions like HICS, which I plan to join in Year 2. I +am particularly drawn to programs in marketing and +finance, as well as opportunities for entrepreneurship +through the Techno-Entrepreneurship Core and +iDendron incubator. With a diverse student body and +global exchange options, I look forward to learning +from peers, connecting with mentors, and developing +skills to make a meaningful impact in business. + +The main reason I chose HKU, particularly HKU +Business School, lies in its global connections and +strong partnerships with other institutions and +respected organizations, which I believe will be +advantageous for future employment opportunities. +I expect to embrace challenges over the course of my +four years at the university, as these will be vital for +my intellectual and social maturity. The first time I +visited the campus, I was amazed by the diversity of +the students and +facilities, which +demonstrated that the university is truly a place for +everyone. One of the many great opportunities +offered here is the chance to travel for studies, which +I would rank among my highest interests. + +the open + +Pernice, Audrey Yiwen +BEcon&Fin + +4140 The main reason I chose BBA(IBGM) is its flexibility in allowing +students to choose a second major, as well as the numerous +overseas exchange opportunities. I am also fortunate to have +been admitted as a member of the HKU International Case +Society (HICS), where I get to know people from other programs +who are also interested in the realm of business. HKU Business +School provides me with the perfect platform to strike a balance +between my work and life. + +Ho, Lik Hang Gordon +BBA(IBGM) + +Lo, Hoi Tung Reacker +BSc(MAT) + +I chose HKU Business School for its professional background and +diverse opportunities. The multidisciplinary programme allows me to +pursue Marketing Analytics and Technology, aligning with my varied +interests in business and STEM. As a freshman, I am eager to embrace +exciting experiences such as internships and exchanges. I look +forward to gaining insights from fellow students and professionals +through upcoming career events. With the strong support from HKU, +I am committed to developing my entrepreneurial skills and pursuing +my dreams without regrets! + +Drawn by HKU’s global prestige and the excellence of HKU Business +School, I chose to pursue Quantitative Finance at HKU to deepen my +passion for quantitative trading and data-driven decision-making. +The school’s flexible curriculum empowers me to explore +complementary fields such as computer science, statistics, and risk +management. As a target school for top investment banks, HKU +also offers unparalleled career exposure. HKU provides the ideal +environment to grow academically and professionally, and I am +excited to explore the many opportunities ahead. + +Tsang, Chun Shing Patrick +BSc(QFin) + +Fung, Yat Hang Rex +BFin(AMPB) + +My aspiration for BFin(AMPB) was ignited early in high school by its +unparalleled, practitioner-focused design, a curriculum that truly +bridges finance theory with industry practice. This conviction was +solidified through the heartfelt guidance of our Programme +Director, Professor Anna Wong, who mentored me even before I +officially joined HKU, while I was an apprentice at a US bank. + +Now, experiencing BFin(AMPB) firsthand, the transition from the +professional world to the classroom feels seamless. It is this +powerful integration of theory and real-world practice that makes +AMPB incomparable. HKUBS has instantly instilled in me a sense of +ownership, this is not just a school, but a launchpad for impact, and +I am thrilled to begin the next chapter of my growth here. + +I was drawn to BBA(IBGM) because of HKU’s +world-renowned academic reputation and its vibrant, +international community. I believe a truly “global +business” education extends beyond the classroom, +through experiential learning opportunities, daily +interactions with local and non-local peers, and +participation in hall and student society cultures, I will +gain a more dynamic and holistic multidisciplinary +perspective on the business world. Over the coming +four years, I am eager to harness these opportunities +to become a well-rounded leader and effective +problem-solver for my future career on the global +business stage. + +Yan, Sung Hei Christopher +BBA(IBGM) + +4342 Our Global Presence + +Israel - +Tel Aviv + +Beijing + +Shenzhen + +Hong Kong + +Vietnam - +Ho Chi Minh City + +Useful Websites +The University of Hong Kong (HKU) + +Undergraduate Admissions +https://admissions.hku.hk/ -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan -sheets 8, 9, 12, 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. +HKU – Entrance Requirements -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. +For JUPAS Candidates +https://admissions.hku.hk/apply/jupas -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. +For non-JUPAS Candidates +https://admissions.hku.hk/apply/international-qualifications -Sincerely, +HKU Business School -&Ad,- +Overview of Undergraduate Programmes +https://ug.hkubs.hku.hk/programme -Deputy State Historic Preservation Officer +Enrichment Programmes +https://ug.hkubs.hku.hk/student-enrichment -/ David Brook +Other HKU Websites -DB:slw +Mainland Affairs Office (MAO) +https://mainlandaffairs.hku.hk/ -Enclosure +Centre of Development and Resources for +Students (CEDARS) +https://www.cedars.hku.hk/ -cc: (cid:9) +44 -Army Corps of Engineers, Asheville -William D. Gilmore -Thomas Padgett +45 diff --git a/q104/random_k2/random_2.md b/q104/random_k2/random_2.md index 9a6992971674500fba1af17645e45a0908def2ef..bdc0cc473d09f25f5a2f7f0b1e46304031484979 100644 --- a/q104/random_k2/random_2.md +++ b/q104/random_k2/random_2.md @@ -1,727 +1,637 @@ - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Youth in Middle East +and North Africa - Online ISSN: 2054-0965 (Online) +August 2019 - Print ISSN: 2054-0957 (Print) +Daniella Raz +University of Michigan - Website: https://www.eajournals.org/ +JULY 2018 Arab Barometer – Wave V +Topic Report – Youth - Publication of the European Centre for Research Training and Development -UK +Executive Summary -Cloud Computing: Revolutionizing Digital -Transformation in the Modern Era +Individuals between the ages of 15 and 29 account for roughly 30 percent of the +population of the Middle East and North Africa (MENA).1 The beliefs, attitudes, +and preferences of this large and growing segment of the population has the po- +tential to shape and drive the region’s future. Yet, widespread youth discontent +about their economic prospects poses a significant challenge. Youth unemploy- +ment rates across the region remain high and there is broad frustration with +government efforts to create employment opportunities.2 Though satisfaction +with the state education system varies greatly across countries, youth are less +satisfied than are older generations. -Lavanya Jacintha Victor -The University of Texas at Austin, USA +Youth also question if they have the ability to voice their frustrations to the gov- +ernment. In nearly all countries, less than half say the right to freedom of ex- +pression is guaranteed while half or less say they have the right to demonstrate +peacefully in all countries. This combination creates conditions where youth +across the MENA region, far more than older age cohorts, are considering emi- +grating from their country, primarily to Europe, North America, or GCC countries. +Unsurprisingly, many cite economic reasons as the main impetus for leaving. -doi: https://doi.org/10.37745/ejcsit.2013/vol13n181223 Published May 12, 2025 +However, youth differ from their older counterparts not only in terms of greater +frustration with the existing system. Relative to older generations youth also +tend to be less religious, less interested in politics, and more likely to be engaged +in social media. Generational differences are particularly stark in Morocco where +youth attitudes diverge substantially and dissatisfaction with the overall state of +affairs is quite salient. -Citation: Victor L.J. (2025) Cloud Computing: Revolutionizing Digital Transformation in the Modern Era, European -Journal of Computer Science and Information Technology, 13(18),12-23 +Meanwhile, on gender issues, most youth tend to be supportive of equal rights +for women. Majorities say women have the right to be the head of government +and to receive an education. Yet, the same does not hold true for equal inher- +itance rights with minorities saying women should receive the same as men +in most countries. Additionally, this generation cites low levels of support for +women playing equal roles in society, with most saying that men make better +political leaders and stating that husbands should have the final say in family +matters. + +Youth in MENA appear to be particularly inclined toward China relative to other +global powers. Majorities favor closer economic relations with China in most +countries surveyed, while less than half say the same about relations with the +U.S. in most countries. + +1https://www.undp.org/content/dam/rbas/report/AHDR%20Reports/AHDR%202016/AHDR%20Final% + +202016/AHDR2016En.pdf, pg. 7 + +2https://www.imf.org/external/np/vc/2012/061312.htm?id=186569 + +www.arabbarometer.org + +2 + + Arab Barometer – Wave V +Topic Report – Youth + +These are among the key findings from 11 nationally representative public opin- +ion surveys conducted across the Middle East and North Africa by the Arab +Barometer from September 2018 - April 2019. The surveys included more than +25,000 face-to-face interviews in the respondent’s place of residence and have +a margin of error of ±2 percent in most countries.3 + +Religiosity + +Across the region, youth are less religious than their older counterparts. In fact, in +every country surveyed less than half of youth identify as religious. More specif- +ically, fewer than one-fifth of youth in Algeria (15 percent), Tunisia (16 percent), +and Libya (18 percent), and fewer than one-quarter of youth in Jordan (22 per- +cent), Lebanon (23 percent), and Morocco (24 percent) consider themselves reli- +gious. Higher rates of youth religiosity are found in Palestine (28 percent), Sudan +(32 percent), Yemen (33 percent) Egypt (36 percent), and Iraq (42 percent). + +However, trends over time paint a more complicated picture throughout the +region. While indeed less than half of MENA youth identify as religious, in some +countries these rates represent an increase from the last Arab Barometer wave + +3For details about each survey, please visit: + +https://www.arabbarometer.org/waves/ + +arab-barometer-wave-v/ + +www.arabbarometer.org + +3 + + Arab Barometer – Wave V +Topic Report – Youth + +while in others a decrease. For instance, rates of religiosity have increased in +Egypt (+12), Yemen (+7), Tunisia (+6), Iraq (+6), Morocco (+2), and Libya (+2). In +contrast, youth identify as religious at lower rates in Sudan (-12), Palestine (-9), +Jordan (-7), Algeria (-3) and Lebanon (-2). + +Views on Politics + +In addition to generally low levels of religiosity, youth appear disinterested in +politics, both overall and when compared to older generations. Only three-in- +ten youth in Lebanon (31 percent), Yemen (29 percent), Palestine (29 percent), +Sudan (28 percent), and Egypt (27 percent) say they are interested or very inter- +ested in politics, while no more than a quarter say the same in Jordan (25 per- +cent), Morocco (23 percent), Iraq (21 percent), and Libya (21 percent). Tunisian +and Algerian youth are least politically interested with fewer than two-in-ten +citing interest in politics (17 percent and 15 percent, respectively). + +With the exception of Egypt and Lebanon, youth in the region are less interested +in politics than are older citizens. The difference between age cohorts is widest +in Tunisia and Palestine (10 percentage points). Elsewhere, though differences +are smaller, older citizens nonetheless report greater interest in politics than do +youth. +In Lebanon youth have only marginally greater interest (1 percentage +point), while in Egypt youth are 6 percentage points more likely than older citi- + +www.arabbarometer.org + +4 + + Arab Barometer – Wave V +Topic Report – Youth + +zens to say they are interested or very interested in politics. + +Youth disinterest in the political system could be in part due to their outlook +If young people perceive the most +on the fairness of parliamentary elections. +recent parliamentary elections as unfair, regardless of whether they voted or +were even eligible to vote in them, then their disinterest could perhaps reflect +disillusionment with their country’s political process. With the exception of the +Levant, youth are more likely than older age cohorts to evaluate the last par- +liamentary elections held in their respective country as being neither free nor +fair. In some countries surveyed this difference is stark. For instance, in Morocco +four-in-ten youth characterized the last parliamentary elections as such, com- +pared to fewer than a quarter of older Moroccans (23 percent). In Tunisia, youth +were 12 percentage points more likely than older citizens to characterize the +last parliamentary elections as being neither free nor fair (40 percent versus 28 +percent). + +Internet and Social Media + +Across the region, internet use, defined as going online at least once-in-a-few- +weeks, is substantially greater among youth than older generations. In fact, in +most of the surveyed countries (with the exception of Yemen), more than 90 +percent of youth use the internet, whereas among the older generation internet +usage does not approach such a rate. The highest rates of internet usage are +found in Lebanon (99 percent), followed by Jordan (96 percent), Palestine (95 +percent), Algeria (93 percent), Morocco (93 percent), Tunisia (91 percent), Egypt +(90 percent), Sudan (90 percent), Iraq (89 percent), and Libya (88 percent). The +lowest rate of internet usage among youth is seen in Yemen, where fewer than +eight-in-ten youth say they use the internet at least once every few weeks (77 +percent). The most striking generational differences in internet usage are seen in +Yemen and North Africa (excluding Libya), where internet use among youth ex- +ceeds that of older age cohorts by at least 30 percentage points. In contrast, the +intergenerational gap in internet use is less pronounced in the Levant (Lebanon, +Jordan, Palestine), where it does not exceed 20 percentage points. + +Interestingly, among internet users, social media usage is very high across all +demographics. In fact, youth are not so different from their older counterparts +in this regard. Though Yemeni youth report the lowest rates of internet usage in +the region, those online are also the most likely to engage in social media. Nearly +all Yemeni youth who are online (99 percent) use social media on a typical day. +However, most online youth across the region also engage in social media on a +typical day: Jordan (98 percent), Lebanon (97 percent), Iraq (97 percent), Pales- +tine (96 percent), Libya (96 percent), Egypt (96 percent), Tunisia (96 percent), +Morocco (96 percent), Sudan (94 percent), and Algeria (94 percent). Overall, +youth are consistently more likely than older generations to say they spend two + +www.arabbarometer.org -Abstract: Cloud computing has revolutionized digital transformation by fundamentally changing how -organizations approach their technological infrastructure and business operations. The shift from -traditional on-premises systems to cloud-based solutions has enabled unprecedented levels of scalability, -efficiency, and innovation across industries. From artificial intelligence integration to edge computing -adoption, cloud technologies have democratized access to advanced capabilities while reducing -operational costs. Organizations worldwide are embracing hybrid and multi-cloud strategies, prioritizing -security through zero-trust architectures, and focusing on sustainable computing practices. The evolution -of serverless computing and edge technologies continues to shape the future of enterprise IT, offering new -possibilities for innovation while addressing crucial challenges in data processing, security, and -compliance. -Keywords: cloud computing, digital transformation, edge computing, serverless architecture, hybrid -cloud - -INTRODUCTION +5 -In today's rapidly evolving technological landscape, digital transformation has become a cornerstone of -business evolution and success. According to recent International Data Corporation (IDC) forecasts, -worldwide spending on digital transformation is projected to maintain a robust compound annual growth -rate (CAGR) of 16.3% from 2022 to 2026, culminating in a staggering $3.4 trillion by 2026. The digital -and information technology sector has demonstrated remarkable resilience even during economic -uncertainties, showcasing a significant 17.8% growth in digital transformation investment during 2022 -alone [1]. This comprehensive shift towards digitalization represents more than just a trend—it's a -fundamental reimagining of how organizations operate, deliver value, and interact with their stakeholders -through the integration of digital technologies. + Arab Barometer – Wave V +Topic Report – Youth + +hours or more on social media platforms, although the size of this difference is +often marginal at best. As an example, the widest generational gap is seen in +Morocco, where youth are 10 percentage points more likely to use social me- +dia. However, the difference is much smaller in Yemen (1 percentage point) and +Jordan (2 percentage points), and non-existent in Palestine. + +Though there is little generational difference in overall quantity of social media +usage, there exists a much larger generational discrepancy in terms of moti- +vations for using and trust in social media. Youth are more likely than older +generations to use social media as their primary source of information to fol- +low breaking news, and also to trust this information more than that found in +newspapers or TV news programs. With respect to the former, youth are most +likely to say social media is their primary source for breaking news in Palestine +(76 percent), while around two-thirds say the same in Morocco (67 percent) and +Jordan (63 percent). About half of youth or more in Egypt (58 percent), Tunisia +(55 percent), Iraq (51 percent), and Algeria (49 percent) rely primarily on so- +cial media for breaking news, while less than half do so in Libya (45 percent), +Lebanon (43 percent), Yemen (37 percent) and Sudan (32 percent). Notably, +youth reliance on social media to follow breaking news far exceeds that of their +older counterparts in every country surveyed. The greatest discrepancy between +youth and older generations is seen in Morocco (53 percentage points), Egypt +(44 percentage points), Tunisia (36 percentage points), and Palestine (32 per- +centage points). Elsewhere, though less stark, rates of youth reliance on social + +www.arabbarometer.org + +6 + + Arab Barometer – Wave V +Topic Report – Youth + +media for breaking news still exceed that of older citizens by at least 15 per- +centage points or more. Such trends show that while youth and their older +counterparts report using social media at similar rates overall, their incentives +and motivations for using such platforms differ in important ways. + +Youth in the region vary dramatically in the level of trust in information they re- +ceive on social media. The highest rates of trust among youth are found in Mo- +rocco (77 percent) and Egypt (74 percent), followed by Lebanon (67 percent), +Palestine (56 percent), Iraq (53 percent), and Sudan (51 percent). In contrast, +less than half of youth in Algeria (49 percent), Yemen (47 percent), Tunisia (47 +percent), Libya (41 percent), and Jordan (36 percent) trust social media sources +more than newspapers and TV news. The magnitude of the differences across +age cohorts is also varies widely, ranging from single digit differences in Jordan +(2 percentage points), Libya (5 percentage points), Yemen (3 percentage points), +and Sudan (3 percentage points) to much larger generational gaps in Lebanon +(16 percentage points), Egypt (18 percentage points), and Morocco (32 percent- +age points). Overall, however, youth are more trusting of the information they +receive from social media than are older citizens in every country surveyed. + +Notably, the most used social media platform among youth across the region is +Facebook. The exception to this finding is Yemen, where youth use WhatsApp at +a higher rate than Facebook (46 percent versus 36 percent). Youth in Egypt (94 + +www.arabbarometer.org + +7 + + Arab Barometer – Wave V +Topic Report – Youth + +percent), Algeria (91 percent), Morocco (90 percent), and Lebanon (90 percent) +are most partial to Facebook. More than eight-in-ten youth in Sudan (85 per- +cent), Libya (82 percent), and Jordan (81 percent), and more than seven-in-ten +youth in Iraq (74 percent) and Palestine (73 percent) use Facebook. + +Education and Employment + +Education System + +Rates of youth satisfaction with state education systems, both overall and as they +compare to older generations, vary greatly throughout the region. Nearly two- +thirds of youth in Palestine (65 percent) and Jordan (65 percent), and over half +of youth in Yemen (54 percent) are satisfied or completely satisfied. However, +fewer than half are satisfied in Sudan (43 percent) and Lebanon (39 percent), +and roughly a third are satisfied in Algeria (34 percent), Libya (31 percent), Egypt +(30 percent), and Tunisia (30 percent). The lowest rates of satisfaction are seen +among Iraqi youth (24 percent) and Moroccan youth (17 percent). Given this +wide variation, there seems to be limited regional consensus about the state +of education. However, one potential regional trend is that across most of the +countries surveyed youth are less satisfied with the education system than are +their older counterparts. +Indeed, in the majority of countries this discrepancy +is small, as seen in Palestine (1 percentage point), Sudan (2 percentage points), +Libya (2 percentage points), Algeria (3 percentage points), Yemen (3 percent- +age points), Iraq (4 percentage points), and Lebanon (5 percentage points). Yet +elsewhere, for instance in Morocco and Egypt, the intergenerational gap is more +sizable (22 percentage points and 10 percentage points, respectively). Excep- +tions to this trend are found only among Tunisian and Jordanian youth, both of +whom voice satisfaction at higher rates than do older citizens (30 percent versus +28 percent and 65 percent versus 59 percent, respectively). + +www.arabbarometer.org -12 +8 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -The impact of digital transformation is profound yet challenging. Despite substantial investments, recent -executive surveys reveal that approximately 70% of digital transformation initiatives are falling short of -their objectives. A significant factor contributing to this shortfall is the persistent gap between technological -capabilities and workforce readiness. Nearly two-thirds of surveyed executives report that their -organizations struggle with change management and employee adoption of new digital systems. -Furthermore, 84% of organizations cite human-centric issues, rather than technical complications, as the -primary barriers to successful digital transformation [2]. - -The landscape of digital transformation has evolved significantly, with organizations focusing on three -critical areas: technology infrastructure modernization, process automation, and workforce enablement. -The IDC research highlights that industries such as financial services, manufacturing, and retail are leading -the investment charge, with discrete manufacturing alone expected to invest $382 billion in digital -transformation by 2026 [1]. However, the success of these investments heavily depends on organizations' -ability to address fundamental human elements. Studies indicate that organizations achieving digital -transformation success are three times more likely to prioritize workforce development and change -management strategies alongside their technological implementations [2]. - -The momentum of digital transformation continues to accelerate, driven by market demands and -competitive pressures. While the technological spending trajectory remains impressive, with manufacturing -and financial services each projected to exceed $1 trillion in total digital transformation investments through -2026 [1], the focus is increasingly shifting toward holistic transformation approaches. Successful -organizations are now recognizing that digital transformation requires a balanced investment in both -technology and people, with leadership development, skill enhancement, and cultural adaptation being as -crucial as the technological infrastructure itself [2]. - -The Evolution of Digital Transformation -The journey of digital transformation began with the basic digitization of manual processes, marking a -fundamental shift in business operations. Current research indicates that organizations implementing digital -transformation strategies have achieved remarkable improvements in their operational metrics. By -integrating automation and digital workflows, businesses have reported up to 80% reduction in process -completion times and a 90% decrease in human errors across various operational tasks. This evolution has -particularly impacted customer service operations, where automation has enabled organizations to handle -customer inquiries 60% faster while maintaining high satisfaction rates [3]. What started as simple -workflow automation—like digital leave management systems and basic inventory tracking—has evolved -into sophisticated, interconnected ecosystems that drive entire business operations. - -The transformation landscape has seen significant advancement in data processing capabilities, with -modern systems able to handle complex analytical tasks in real-time. Organizations implementing -comprehensive digital transformation strategies have witnessed a 70% improvement in their decision- -making processes, primarily due to enhanced data accessibility and automated analytical capabilities. + Arab Barometer – Wave V +Topic Report – Youth + +Looking to perceived corruption in the education system as one potential rea- +son contributing to low levels of satisfaction, in many countries more than half of +youth believe that rashwa (bribes) is required in order to receive better access to +education services. This belief is most commonly held in Egypt (64 percent), fol- +lowed by Lebanon (62 percent), Sudan (55 percent), Morocco (54 percent), Alge- +ria (53 percent), Yemen (53 percent), Iraq (52 percent), and Tunisia (50 percent). +However, while many youth believe that paying rashwa is necessary, this trend +is not common to all MENA youth as no more than a third of youth in Palestine +(33 percent), Libya (32 percent), and Jordan (29 percent) say the same regard- +ing rashwa to access better educational services. Prevalence of rashwa might +contribute to some dissatisfaction with state education systems, but it does not +account for or explain it in its entirety. + +Employment Opportunities + +Similar to satisfaction rates with state education systems, levels of youth satis- +faction with the government’s performance on creating employment opportu- +nities vary. However, despite this variation, across the region youth approval is +overwhelmingly weak, ranging from a low of five percent in Lebanon to a high +In addition to Lebanon, single-digit approval of +of only 32 percent in Yemen. +government performance on creating employment opportunities is seen in Iraq +(6 percent) and Algeria (9 percent). Fewer than two-in-ten youth in Morocco +(13 percent), Palestine (14 percent), Jordan (15 percent), Tunisia (16 percent), + +www.arabbarometer.org + +9 + + Arab Barometer – Wave V +Topic Report – Youth + +Sudan (17 percent), and Egypt (18 percent), and less than a quarter of youth in +Libya (22 percent) approve. + +Moreover, in addition to widespread dissatisfaction with government efforts to +improve employment opportunities, the vast majority of young Arab citizens +also believe that obtaining employment requires having strong connections. +Roughly nine-in-ten youth in Iraq (95 percent), Lebanon (94 percent), Jordan +(94 percent), Tunisia (94 percent), Libya (91 percent), Algeria (91 percent), Sudan +(90 percent), Palestine (90 percent), Egypt (90 percent), and Morocco (88 per- +cent) say that based on recent experience they think that obtaining employment +through wasta (personal connections) happens often or sometimes. Though +comparatively fewer, nearly seven-in-ten Yemeni youth (69 percent) hold the +same perception regarding the connection between wasta and employment. + +Emigration + +Given broad youth dissatisfaction with conditions in their countries, it is perhaps +not surprising that youth across the MENA region have a strong desire to em- +igrate. In six of eleven countries surveyed, roughly half or more of youth want +to leave their homeland. With the exception of Libya, North African and Jorda- +nian youth are most eager to migrate with about half or more citing a desire +to leave. Youth in Morocco cite not only the greatest desire to migrate, but also + +www.arabbarometer.org -13 +10 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 + Arab Barometer – Wave V +Topic Report – Youth + +show the greatest discrepancy with older generation’s desire to do so (70 per- +cent versus 25 percent). Meanwhile, over half of youth in Sudan (62 percent), +Jordan (59 percent), Tunisia (56 percent) and Algeria (56 percent) also wish to +leave their country. Roughly a third of Libyan (31 percent), Yemeni (33 percent), +and Lebanese (35 percent) youth say the same. + +Regional trends are evident with regard to which countries youth are most ea- +ger to migrate. Over two-thirds of youth in French-speaking North Africa want +to migrate to a European country. This is not surprising given both the proximity +of their countries to Europe and, perhaps more importantly, the close linkages +between France and these countries. In contrast, six-in-ten youth in Egypt, and +roughly half of youth in Yemen (48 percent) and Sudan (45 percent), say they +want to emigrate to a GCC country, which is in line with historic patterns of labor +migration to this area. Interestingly, fewer than half of youth across the region +hope to migrate to North America. Jordanian youth are most likely to cite a +desire to migrate to North America (42 percent), whereas Yemeni, Libyan, and +Tunisian youth are least likely (12 percent, 13 percent, and 13 percent). + +The least preferred destination for youth is non-GCC MENA countries. Fewer than +a third of youth across the region have considered emigrating to such countries. +Lebanese and Tunisian youth are least eager (4 percent and 5 percent, respec- +tively) while Yemeni youth are most eager, though still fewer than three-in-ten +Yemeni youth cite an interest in emigration to a non-GCC MENA country (27 + +www.arabbarometer.org + +11 + + Arab Barometer – Wave V +Topic Report – Youth + +percent). + +Perceived Civil Liberties + +Of the countries surveyed by the Arab Barometer, Tunisia is the only one cate- +gorized as free by Freedom House, with the remainder of countries identified as +partly free or not free. As such, it is unsurprising that youth in Tunisia are more +confident in their civil liberties like freedom of expression and right to peaceful +protest than are youth in other surveyed countries. However, despite being com- +paratively more confident in the protection of their civil liberties, Tunisian youth +are still quite similar to youth in the rest of the region who report generally low, +though varied, confidence in both freedom of speech and right to peacefully +protest. + +Looking first to freedom of speech, Tunisian youth are indeed most confident in +their freedom of expression with over six-in-ten saying that it is guaranteed to a +great or medium extent in the country (63 percent). More than half of youth in +Jordan (56 percent) and Lebanon (51 percent) say the same. In the rest of the +region fewer than half of respondents feel their freedom of expression is guar- +anteed. About half of youth in Yemen (49 percent), roughly four-in-ten youth +in Palestine (42 percent), Egypt (40 percent), and Morocco (39 percent), and +roughly a third of youth in Algeria (35 percent) and Iraq (33 percent) say that + +www.arabbarometer.org - Online ISSN: 2054-0965 (Online) +12 - Print ISSN: 2054-0957 (Print) + Arab Barometer – Wave V +Topic Report – Youth + +freedom of expression is guaranteed to a great or medium extent. Least confi- +dent of their freedom of expression are youth in Sudan (25 percent) and Libya +(24 percent). + +When comparing youth perceptions of their freedom of speech to the percep- +tions of older age cohorts, there is no clear regional trend. For instance, youth +in Lebanon (+10 points), Jordan (+3 points), Yemen (+4 points) and Palestine +(+4 points) are more confident in their freedom of expression than are older +citizens. Yet, the opposite trend is apparent in Morocco (-16 points), Egypt (-5 +points), and Tunisia (-4 points). +In the remainder of countries surveyed (Libya, +Sudan, Iraq, and Algeria) youth and older citizens do not diverge by more than +a point and rate their freedom of expression similarly. + +Youth perception of their freedom to participate in peaceful protests is also low +across the region, never exceeding half of youth in any country. Though nearly +half of youth in Tunisia (49 percent), Lebanon (49 percent), Jordan (47 percent), +and Yemen (46 percent) feel their freedom to peacefully demonstrate is guaran- +teed to a medium or great extent, fewer youth hold this belief elsewhere in the +region. Four-in-ten Palestinian youth, a third of Moroccan youth, and three-in- +ten Iraqi youth are confident of their freedom to participate in peaceful demon- +strations. Meanwhile, roughly a quarter of youth in Algeria (27 percent), Libya +(24 percent), and Egypt (24 percent) and fewer than two-in-ten in Sudan (19 +percent) say the same. + +www.arabbarometer.org - Website: https://www.eajournals.org/ +13 - Publication of the European Centre for Research Training and Development -UK + Arab Barometer – Wave V +Topic Report – Youth -Furthermore, the integration of artificial intelligence and machine learning has enabled predictive -maintenance systems that reduce equipment downtime by up to 45% [3]. +Attitudes on Gender -The Era of On-Premises Infrastructure -The first major milestone in this journey was the establishment of on-premises data centers, which -fundamentally changed how businesses managed their digital operations. On-premises data centers -emerged as critical infrastructure components, offering organizations complete control over their hardware, -software, and data management processes. This infrastructure model provided significant advantages in -terms of data security and compliance, particularly for industries handling sensitive information. A key -benefit was the ability to maintain an average network latency of less than 5 milliseconds, crucial for time- -sensitive applications and real-time data processing [4]. +Arab youth are divided in their views on gender equality. On the one hand, +most are supportive of equal rights for women, with the exception of inheri- +tance. However, there is substantially less support for women enjoying equal +roles in society. This is particularly evident in the political sphere. -The financial implications of on-premises infrastructure were substantial but justified for many -organizations. While initial setup costs typically ranged between $10,000 to $25,000 per rack, with annual -operating costs between $12,000 and $18,000, these investments provided organizations with direct control -over their technological infrastructure. The model proved particularly advantageous for businesses with -consistent workloads and stringent regulatory requirements, as it eliminated monthly recurring costs -associated with cloud services and provided better control over long-term expenses [4]. +Throughout the region fewer than a third of youth say that a university education +is more important for men than for women. Support for university education +for men compared with women is highest in Yemen and Sudan (27 percent, +each), followed by roughly two-in-ten youth in Jordan (23 percent), Egypt (22 +percent), Algeria (20 percent), Iraq (19 percent), Tunisia (17 percent) and Libya +(17 percent). Even lower rates of support for men’s education taking precedence +over that of women are found in Palestine (14 percent), Lebanon (13 percent), +and Morocco (12 percent). -Security and compliance capabilities represented another crucial aspect of on-premises infrastructure. -Organizations maintained full control over their security protocols, physical access management, and data -handling procedures. This control was particularly valuable for industries subject to strict regulatory -requirements, such as healthcare and financial services, where data sovereignty and compliance are -paramount concerns. The ability to customize security measures and maintain direct oversight of data -storage and processing activities allowed organizations to achieve compliance rates exceeding 99% [4]. +Youth also generally hold liberal views on the question of whether women should +have equal rights in the decision to get a divorce. The exception to this trend +is in Sudan, where support for women’s equal rights in deciding to divorce falls +below five-in-ten (46 percent). Support is highest in Lebanon (87 percent), fol- +lowed by Iraq (81 percent), Jordan (80 percent), Tunisia (78 percent), Morocco +(78 percent), Palestine (76 percent), Libya (75 percent) and Algeria (72 percent). +Though still greater than half, lower rates of support are seen in Egypt (64 per- -However, this approach presented notable challenges in terms of scalability and resource management. -Organizations needed to maintain dedicated IT teams for infrastructure management, with staffing costs -typically accounting for 40-50% of the total operational budget. Additionally, scaling operations required -significant advance planning and capital investment, with expansion projects often taking several months -to complete. The necessity for redundant systems and backup power solutions further added to the -complexity and cost of maintaining on-premises infrastructure [4]. +www.arabbarometer.org 14 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 1. Digital Transformation Performance Metrics: Year-over-Year Analysis (2021-2024) - -Metric Category - -Equipment Downtime Reduction -Customer Inquiry Speed -Decision-Making Efficiency -Infrastructure Staff Cost -Network Latency (ms) - -2021 (%) -25 -30 -40 -35 -20 - -2022 (%) -35 -45 -55 -40 -15 - -2023 (%) -40 -55 -65 -45 -10 - -2024 (%) -45 -60 -70 -50 -5 - -The Cloud Computing Revolution -The emergence of cloud computing marked a paradigm shift in the digital transformation landscape, -fundamentally altering how businesses approach their IT infrastructure and operational capabilities. Recent -market analysis reveals that cloud spending now represents approximately 30% of IT budgets across -industries, with this figure expected to grow significantly. The industry is witnessing an unprecedented shift -toward multi-cloud and hybrid cloud strategies, with organizations increasingly focusing on cloud cost -optimization and management. This evolution has led to the emergence of specialized roles such as cloud -financial operations (FinOps) professionals, who are becoming critical to managing cloud investments -effectively [5]. - -The adoption of cloud computing continues to accelerate, driven by both necessity and opportunity. -Industry research indicates that companies can capture up to $1 trillion in annual value through cloud -adoption, highlighting the transformative potential of this technology. This value creation stems from -various sources, including innovation acceleration, cost reduction, and improved resilience. The impact is -particularly significant in Fortune 500 companies, where cloud adoption has enabled revenue growth of up -to $100 billion annually through broader cloud-supported initiatives [6]. - -Key Benefits of Cloud Computing: Infrastructure Flexibility -Cloud platforms have revolutionized resource allocation capabilities, enabling unprecedented operational -efficiency. The shift toward cloud-native architectures has become increasingly prominent, with -organizations moving beyond basic "lift and shift" approaches to more sophisticated cloud-native -implementations. This trend has been accelerated by the growing adoption of containers and microservices, -with Kubernetes emerging as a de facto standard for container orchestration. The market has seen a -significant focus on edge computing capabilities, with major cloud providers expanding their edge presence -to support growing demands for low-latency applications and improved performance [5]. - -Cost Optimization -The financial implications of cloud adoption are substantial and multifaceted. Research shows that -organizations can potentially reduce IT costs by 20 to 30 percent through cloud optimization initiatives. -However, the most significant value creation opportunities lie in business innovation, which can generate -$770 billion to $995 billion in annual EBITDA across Fortune 500 companies. This value stems from + Arab Barometer – Wave V +Topic Report – Youth + +cent) and Yemen (55 percent). Notably, in most countries surveyed, the views +held by youth about women’s right to divorce are not significantly different from +the views held by older citizens. + +In a departure from predominantly liberal views on women’s rights, surprisingly +few youth in the region agree that women’s share of inheritance should be equal +to that of men. Nonetheless, Lebanon is a clear regional outlier to this trend with +over six-in-ten youth saying they agree or strongly agree that inheritance should +be equal between the sexes (62 percent). In contrast, far fewer youth elsewhere +are as supportive. Roughly three-in-ten youth in Iraq (30 percent) and Egypt (28 +percent), and around one-fourth of youth in Morocco (26 percent) and Tunisia +(24 percent) support equal inheritance. Still fewer youth in Algeria (18 percent), +Jordan (18 percent), Libya (16 percent), Sudan (13 percent), and Palestine (8 +percent) say the same. + +A plurality of youth in the region also believe that husbands should have the +final say in all family decisions. Support for this sentiment is greatest in Sudan +(73 percent), Algeria (73 percent), and Iraq (70 percent), followed by Egypt (66 +percent) and Libya (66 percent). Youth across Tunisia (56 percent), Yemen (55 +percent) and the Levant hold similar views to each other regarding women and +family decision-making. +In Lebanon and Jordan 55 percent of youth believe +husbands should have the final say in all family decisions, while in Palestine 50 +percent say the same. Though Moroccan youth are least likely to support limiting + +www.arabbarometer.org 15 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -various sources, including accelerated product development, improved customer experience, and enhanced -operational efficiency. The cloud's consumption-based pricing model has enabled organizations to achieve -15 to 20 percent cost reduction in infrastructure spending through improved capacity management [6]. - -Global Reach and Collaboration -Cloud infrastructure has enabled unprecedented global operational capabilities, with particular emphasis on -sustainability and environmental impact. Organizations are increasingly prioritizing environmental -considerations in their cloud strategies, with major providers committing to aggressive sustainability goals. -The industry has seen a growing focus on renewable energy usage and improved data center efficiency, -with cloud providers investing heavily in sustainable operations. This shift has resulted in significant -improvements in energy efficiency and reduced carbon footprints for organizations migrating to cloud -services [5]. - -The impact of cloud computing extends beyond operational benefits to strategic value creation. -Organizations leveraging cloud capabilities have reported significant improvements in their ability to -innovate and respond to market changes. The cloud has enabled faster time-to-market for new products and -services, with some companies reducing development cycles by up to 50 percent. Additionally, cloud -adoption has facilitated better risk management and improved security posture, with organizations able to -leverage advanced security features and compliance capabilities provided by cloud platforms. The value at -stake is substantial, with research indicating that companies can potentially capture between $900 billion -and $1.2 trillion in annual value through comprehensive cloud transformations [6]. + Arab Barometer – Wave V +Topic Report – Youth + +women’s roles in family decision-making, nonetheless nearly four-in-ten say that +husbands should have the final say in all family decisions (38 percent). Notably, +though youth in Morocco are far less likely than older generations to support +this notion (-15), youth in Jordan (+6), Lebanon (+7), Libya (+5), and Tunisia (+2) +are more supportive than are their older counterparts. + +In every country surveyed, with the exception of Morocco, over half of youth agree +that, in general, men are better at political leadership than women. Roughly +eight-in-ten youth in Sudan (80 percent) and Jordan (78 percent), and roughly +three-quarters of youth in Algeria (74 percent) hold this belief. Across much of +the rest of the region youth hold this view at similar rates. For instance, around +seven-in-ten youth say men are better at political leadership in Egypt (70 per- +cent), Iraq (69 percent), Yemen (68 percent), Palestine (68 percent), and Libya +(67 percent). Meanwhile, fewer in Tunisia, Lebanon, and Morocco say that men +are better at political leadership than women (52 percent, 51 percent, and 31 +percent, respectively). + +Interestingly, despite unfavorably comparing female political leaders to men and +criticizing women’s role as decision-makers in the family, Arab youth are often +open both to the idea of a woman being a leader of a Muslim country and to +setting aside a certain percentage of elected positions for women as a means +of attaining fairer political representation. More than half of youth support the +notion of a female president or prime minister of a Muslim nation in all coun- + +www.arabbarometer.org 16 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 2. Cloud Adoption: Cost Reduction and Efficiency Analysis [5, 6]. - -Metric Category - -Q1 2024 (%) Q2 2024 (%) Q3 2024 (%) Q4 2024 (%) - -IT Budget Cloud Spend -IT Cost Reduction -Infrastructure Cost Savings -Development Cycle Reduction -Market Response Improvement - -15 -20 -15 -35 -25 - -20 -23 -17 -40 -30 - -25 -27 -18 -45 -35 - -30 -30 -20 -50 -40 - -Advanced Technologies Enabled by Cloud Computing - -Artificial Intelligence and Machine Learning -Cloud platforms have fundamentally democratized access to AI and ML capabilities, creating -unprecedented opportunities for organizations across industries. According to recent market analysis, the -global artificial intelligence market size was valued at USD 454.12 billion in 2023 and is projected to surge -to USD 2,575.16 billion by 2032, growing at an impressive CAGR of 21.4%. This remarkable growth is -driven by the increasing adoption of cloud-based AI solutions across various sectors, including healthcare, -automotive, and retail. The integration of AI with cloud computing has become particularly crucial in the -post-pandemic era, with organizations accelerating their digital transformation initiatives to maintain -competitive advantage [7]. - -The impact of cloud-enabled AI extends across multiple domains, with North America leading the market -share at 35% in 2023. This dominance is attributed to the presence of major technology companies and -widespread adoption of AI technologies in various industries. The Asia-Pacific region is experiencing the -fastest growth, with a projected CAGR of 24.3% during the forecast period, driven by increasing -investments in AI research and development, particularly in countries like China, Japan, and South Korea -[7]. - -Big Data Analytics and IoT Integration -The transformation in data handling and analysis capabilities through cloud computing has revolutionized -how organizations process and utilize information. Edge-cloud infrastructure has emerged as a crucial -component in handling the massive data generated by IoT devices, with studies showing that edge -computing can reduce data transmission costs by up to 30% while improving response times significantly. -The integration of advanced analytics algorithms in edge-cloud architectures has enabled real-time -processing capabilities, crucial for applications requiring immediate insights and actions [8]. - -The evolution of IoT data analytics has led to the development of sophisticated middleware architectures -that efficiently manage data flow between edge devices and cloud platforms. Research indicates that hybrid -edge-cloud solutions can reduce latency by up to 50% compared to traditional cloud-only approaches. These -architectures implement various data analytic algorithms, including deep learning, reinforcement learning, + Arab Barometer – Wave V +Topic Report – Youth + +tries surveyed, excluding Algeria (37 percent). Roughly half of youth in Sudan +(51 percent) and Libya (54 percent), six-in-ten youth in Yemen (57 percent), Jor- +dan (60 percent), and Palestine (64 percent) and roughly seven-in-ten youth in +Tunisia (67 percent), Egypt (68 percent) and Iraq (69 percent) affirm that women +can become prime ministers or presidents of Muslim countries. Moroccan and +Lebanese youth (77 percent and 79 percent, respectively) are the most accept- +ing of this right. + +The majority of MENA youth also believe that in order to achieve fairer represen- +tation a certain percentage of elected positions should be set aside for women. +Support for this right exceeds half of all youth in all countries surveyed. Youth in +Iraq and Lebanon (79 percent, each) are most supportive, followed by Sudan (76 +percent), Palestine (75 percent), Jordan (72 percent) and Yemen (69 percent). +Excluding Sudan, North African youth show the least support for women’s quo- +tas in elected positions with less than two-thirds of youth in Egypt (66 percent), +Libya (65 percent), Tunisia (63 percent), Morocco (62 percent) and Algeria (53 +percent) saying they agree or strongly agree with this idea. + +International Relations + +In over half of the countries surveyed, youth are most supportive of stronger ties +with China and least interested in stronger ties with the United States among +global powers. Youth in Jordan (70 percent), Tunisia (65 percent), Sudan (62 +percent), Libya (61 percent), Yemen (50 percent), Palestine (49 percent), and +Iraq (47 percent) prefer strengthening ties with China at a higher rate than they +do for ties with the United States, the United Kingdom, or Russia. Exceptions to +this include youth in Morocco, who have a slightly stronger preference for closer +ties with the United Kingdom (63 percent), youth in Lebanon and Algeria, who +are more supportive by a small margin of stronger ties with Russia (38 percent +and 42 percent, respectively), and Egyptian youth, who are equally supportive +of strengthening ties with the United States and Russia (41 percent, each). + +In a majority of countries surveyed youth support for stronger ties with the United +States not only falls below 50 percent, but is also lower than youth support for +strengthening ties with other world powers. Exceptions to the former are Sudan +(60 percent), Morocco (57 percent), and Jordan (55 percent) where youth sup- +port exceeds half. In contrast, fewer than half of youth in Tunisia (47 percent), +and roughly a third of youth in Libya (36 percent), Iraq (34 percent), Palestine, +Lebanon, and Algeria (31 percent, each) say that relations with the United States +should become stronger. Of all countries surveyed Yemeni youth show not only +the least interest in closer ties with the United States (22 percent), but they are +also least interested in strengthening ties with Russia (26 percent) relative to +youth in other countries. + +www.arabbarometer.org 17 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -and federated learning, enabling more efficient processing of IoT data streams. The middleware layer plays -a crucial role in managing these processes, with modern implementations supporting dynamic resource -allocation and automated scaling based on workload demands [8]. - -The convergence of IoT and cloud computing has particularly benefited from advances in edge computing -capabilities. Contemporary edge-cloud infrastructures support multiple types of edge analytics, including -descriptive, diagnostic, predictive, and prescriptive analytics. This comprehensive approach enables -organizations to process data closer to its source, reducing bandwidth requirements while maintaining data -quality and analysis accuracy. Studies have shown that edge-cloud implementations can achieve up to 40% -improvement in energy efficiency compared to traditional cloud-only solutions [8]. - -Security considerations have become increasingly important in IoT data analytics, with edge-cloud -architectures implementing sophisticated security measures at multiple levels. The integration of -blockchain technology with edge-cloud infrastructure has emerged as a promising solution for ensuring -data integrity and secure communication between devices. Furthermore, the development of privacy- -preserving machine learning techniques has enabled organizations to maintain data confidentiality while -performing complex analytics tasks [8]. - -Table 3. Cloud-Enabled Technologies: Growth and Efficiency Analysis [7, 8]. - -Technology Metric - -2023 Q3 (%) 2023 Q4 (%) 2024 Q1 (%) 2024 Q2 (%) - -North America AI Market Share -APAC Growth Rate -Data Transmission Cost Reduction -Edge-Cloud Latency Reduction -Energy Efficiency Improvement - -35 -20 -20 -35 -25 - -37 -22 -25 -40 -30 - -40 -24 -28 -45 -35 + Arab Barometer – Wave V +Topic Report – Youth -42 -26 -30 -50 -40 +Looking to regional powers, one clear trend is that Arab youth are generally less +interested in stronger relations with Iran and more interested in improving re- +lations with Turkey. Though support for stronger ties with Iran never exceeds +four-in-ten youth in any country surveyed, support is highest in Lebanon (38 per- +cent), followed by Sudan (36 percent), Palestine (34 percent), Iraq (32 percent) +and Tunisia (31 percent). In Lebanon and Iraq, sectarian divides might explain +the comparatively greater support for strengthening ties with Iran, a country +with a strong Shi’ite majority. Roughly two-in-ten youth in Morocco (22 per- +cent), Jordan (21 percent), Yemen (20 percent), Algeria (20 percent), and Libya +(18 percent) prefer stronger relations with Iran, while only 8 percent of Egyptian +youth say the same. -Emerging Trends in Cloud Computing - -Edge Computing and Serverless Architecture -Edge computing has emerged as a transformative force in the cloud computing landscape, bringing -computation closer to data sources and end-users. The integration of edge computing with serverless -architecture represents a significant evolution in cloud technology, addressing critical challenges in modern -application development and deployment. This combination has become particularly crucial for real-time -processing requirements, with edge computing reducing latency by processing data closer to its source -while serverless architecture eliminates the complexity of server management. The implementation of edge- -serverless solutions has shown remarkable improvements in application performance, with organizations -reporting reduction in latency from seconds to milliseconds for critical operations [9]. -The impact of serverless computing extends beyond performance improvements. Recent implementations -have demonstrated significant advantages in terms of cost efficiency and resource utilization. Organizations +www.arabbarometer.org 18 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -adopting serverless architectures have reported substantial reductions in operational overhead, as the -technology eliminates the need for traditional server management and scaling operations. The pay-as-you- -go model inherent in serverless computing has revolutionized cost structures, with businesses only paying -for actual compute time used rather than maintaining constantly running servers. Furthermore, the -integration of edge computing with serverless architecture has enabled more efficient handling of IoT -devices and real-time data processing requirements, resulting in improved application responsiveness and -user experience [9]. - -Hybrid Cloud Solutions -The hybrid cloud market has witnessed extraordinary growth and transformation, with current valuations -at USD 183.7 billion and projected to reach an impressive USD 405.62 billion by 2032, growing at a CAGR -of 9.2%. This remarkable expansion is driven by increasing demands for flexible, scalable, and secure cloud -solutions across various industries. The financial services sector has emerged as a particularly strong -adopter of hybrid cloud solutions, leveraging the technology to balance security requirements with -operational efficiency [10]. - -The adoption of hybrid cloud solutions has been significantly influenced by several key factors, including -the rising need for enhanced data security, disaster recovery capabilities, and the growing complexity of -regulatory compliance requirements. Organizations in the banking, financial services, and insurance (BFSI) -sector have been at the forefront of hybrid cloud adoption, particularly in North America, which currently -holds approximately 35% of the global market share. The integration of artificial intelligence and machine -learning capabilities within hybrid cloud environments has further accelerated adoption rates, enabling -organizations to optimize workload placement and resource utilization more effectively [10]. - -Market analysis reveals that large enterprises currently dominate the hybrid cloud market, accounting for a -significant portion of the overall market share. This dominance is attributed to their complex IT -requirements and substantial resources for implementation. However, small and medium-sized enterprises -(SMEs) are increasingly adopting hybrid cloud solutions, driven by the technology's ability to provide -enterprise-grade capabilities without requiring extensive upfront investments. The public sector has also -emerged as a significant growth driver, with government agencies increasingly migrating to hybrid cloud -environments to modernize their IT infrastructure while maintaining strict security and compliance -requirements [10]. - -Looking Ahead: Future Developments -The convergence of edge computing, serverless architecture, and hybrid cloud solutions continues to -evolve, with emerging technologies and methodologies shaping the future of cloud computing. Edge- -serverless platforms are becoming increasingly sophisticated, incorporating advanced features such as -automated scaling, improved security protocols, and enhanced development tools. The integration of -artificial intelligence and machine learning capabilities at the edge is expected to further transform -application development and deployment paradigms, enabling more intelligent and responsive systems [9]. - -19 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 4. Cloud Architecture Evolution: Latency and Processing Metrics [9, 10]. - -Performance Indicator - -Traditional Edge Only Serverless Edge-Serverless - -Processing Latency (ms) -Resource Overhead (%) -Server Management Time (hrs/week) -Response Time (ms) -Infrastructure Complexity Score - -45 -65 -40 -50 -75 - -25 -45 -30 -35 -55 - -15 -35 -20 -20 -40 - -5 -25 -15 -8 -30 - -Security and Compliance in the Cloud Era -The landscape of cloud security continues to evolve rapidly in response to emerging threats and changing -business requirements. Recent analysis indicates that cloud security remains a top priority for organizations, -with 82% of companies increasing their cloud security budgets in 2024. The shift towards zero trust -architecture has become particularly prominent, with 76% of organizations either implementing or planning -to implement zero trust security models. This transition has been driven by the growing recognition that -traditional perimeter-based security approaches are insufficient in today's distributed cloud environments, -where data and applications are accessed from multiple locations and devices [11]. - -Identity and access management has emerged as a critical component of modern cloud security strategies. -The implementation of comprehensive IAM solutions has become essential as organizations grapple with -increasingly sophisticated cyber threats. Studies show that organizations with mature identity and access -management practices experience significantly fewer security incidents, with advanced authentication -mechanisms reducing unauthorized access attempts by up to 67%. The adoption of biometric authentication -and behavioral analytics has further strengthened security postures, with organizations reporting a 45% -reduction in credential-based attacks [11]. - -The financial implications of data breaches have underscored the importance of robust cloud security -measures. According to IBM's Cost of a Data Breach Report 2024, the global average cost of a data breach -has reached $4.45 million, marking a 15% increase over three years. Organizations leveraging AI and -automation in their security operations have demonstrated significantly better outcomes, with automated -security tools reducing the average time to identify and contain breaches by 108 days. Furthermore, -companies implementing zero trust architecture have seen an average reduction of $1.17 million in breach -costs compared to those without such security measures [12]. - -The impact of remote work and hybrid environments continues to shape cloud security strategies. -Organizations with over 50% of their workforce working remotely have reported higher average breach -costs, reaching $4.99 million compared to $4.24 million for organizations with less remote work. This -disparity has driven increased adoption of advanced security technologies, with 49% of organizations now -implementing AI platforms for cybersecurity. The study reveals that organizations using security AI and - -20 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -automation experienced breach lifecycle times that were 108 days shorter compared to those not utilizing -these technologies [12]. - -Cloud security compliance has become increasingly complex with the evolution of regulatory requirements -and data protection standards. Recent trends show a significant shift towards automated compliance -monitoring and reporting tools, with organizations implementing these solutions reporting 43% lower -compliance-related costs. The adoption of cloud security posture management (CSPM) tools has grown -substantially, with 65% of organizations now using these platforms to maintain continuous compliance and -security assessment capabilities. This has led to a 55% improvement in the detection and remediation of -compliance violations [11]. - -Future Outlook: Cloud Computing and Digital Transformation -The landscape of cloud computing continues to evolve rapidly, with organizations increasingly focusing on -strategic cloud adoption and optimization. According to recent industry analysis, 2024 marks a significant -shift toward more sophisticated cloud strategies, with 94% of enterprises already utilizing cloud services -and 87% adopting multi-cloud approaches. This evolution is particularly noteworthy in the realm of cloud -cost optimization, where organizations are implementing FinOps practices to better manage and optimize -their cloud spending. The trend toward cloud-native development has become increasingly prominent, with -enterprises reporting 30-40% faster application development cycles through cloud-native approaches [13]. -The integration of artificial intelligence and machine learning capabilities has become a cornerstone of -modern cloud strategies. Organizations are increasingly leveraging AI-powered cloud services for various -purposes, from automated infrastructure management to intelligent application development. The adoption -of cloud-native platforms has enabled businesses to achieve remarkable improvements in operational -efficiency, with companies reporting 25-35% reduction in development cycles and 40-50% improvement -in resource utilization. Furthermore, the implementation of containerization and microservices architectures -has led to significant improvements in application scalability and maintenance efficiency [13]. - -Sustainability and environmental responsibility have emerged as critical priorities in cloud computing -strategies for 2024 and beyond. Organizations are increasingly focusing on green cloud computing -initiatives, with major providers implementing advanced energy management systems and sustainable data -center practices. This shift towards environmentally conscious cloud operations has become a key -differentiator in vendor selection, with businesses prioritizing providers that demonstrate strong -commitments to sustainability. The industry is witnessing a growing emphasis on energy-efficient cloud -architectures and sustainable computing practices [14]. - -Security and compliance considerations continue to shape the future of cloud computing, with zero trust -architectures gaining prominence. The industry has seen a significant shift toward comprehensive security -frameworks that encompass everything from infrastructure protection to application security. Organizations -are increasingly adopting automated security solutions and implementing advanced threat detection -systems. The focus on regulatory compliance has intensified, with businesses implementing sophisticated - -21 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK + Arab Barometer – Wave V +Topic Report – Youth -compliance monitoring and management tools to ensure adherence to evolving regulatory requirements -[14]. +In comparison to other regional powers (Saudi Arabia, Qatar, and Iran), Arab +youth tend to prefer strengthening relations with Turkey. Youth in Jordan (82 +percent), Palestine (75 percent), Tunisia (64 percent), Yemen (61 percent), Mo- +rocco (56 percent), Algeria (53 percent), and Iraq (45 percent), are all more likely +to want improved relations with Turkey than to do so for relations with Saudi Ara- +bia, Qatar, or Iran. Youth in Sudan prefer strengthening relations with Saudi Ara- +bia, Turkey, and Qatar at similar rates (70 percent, 73 percent, and 74 percent, +respectively), as do youth in Lebanon (41 percent, 41 percent, and 42 percent, +respectively). In Libya and Egypt youth by far prefer improving ties with Saudi +Arabia (55 percent and 58 percent). -The evolution of cloud management and governance frameworks represents another significant trend. -Organizations are implementing more sophisticated approaches to cloud governance, incorporating -advanced monitoring tools and automated policy enforcement mechanisms. The emphasis on operational -excellence has led to the adoption of advanced observability platforms and AIOps solutions, enabling better -visibility and control over cloud environments. This trend is particularly evident in the increasing adoption -of cloud management platforms that provide comprehensive visibility and control across multi-cloud -environments [14]. +www.arabbarometer.org -CONCLUSION - -Cloud computing stands as a transformative force in the digital era, reshaping how organizations operate, -innovate, and deliver value. The convergence of cloud technologies with AI, machine learning, and IoT has -created powerful new capabilities for businesses of all sizes. As organizations continue to adapt to changing -market demands, the role of cloud computing extends beyond technological infrastructure to become a -strategic enabler of business transformation. The future holds promise for further innovations in edge -computing, serverless architectures, and sustainable cloud practices, while security and compliance remain -paramount considerations. Organizations that successfully leverage these cloud capabilities while -maintaining robust security measures and embracing sustainable practices will be best positioned to thrive -in an increasingly digital world. - -REFERENCES - -[1] Businesswire, "IDC Spending Guide Sees Worldwide Digital Transformation Investments Reaching - -$3.4 Trillion in 2026," 2022. [Online]. Available: -https://www.businesswire.com/news/home/20221026005193/en/IDC-Spending-Guide-Sees- -Worldwide-Digital-Transformation-Investments-Reaching-%243.4-Trillion-in-2026 - -[2] Noah Barsky, "Majority Of Digital Transformations Off Track, Executive Survey Finds—Here Are 3 - -Fixes," Forbes, 2023. [Online]. Available: -https://www.forbes.com/sites/noahbarsky/2023/10/02/people-problems-plague-digital- -transformation/ - -[3] Arbab Khan, "Unlocking the Power of Digital Transformation and Automation," Rushford, 2024. - -[Online]. Available: https://rushford.ch/insights/unlocking-the-power-of-digital-transformation- -and-automation/ - -[4] George, "On-Premises vs. Cloud Data Center: Which Is Right for Your Business?" FS, 2022. - -[Online]. Available: https://www.fs.com/blog/onpremises-vs-cloud-data-center-which-is-right- -for-your-business-5867.html - -[5] Neal Weinberg, "5 cloud market trends and how they will impact IT," CIO, 2024. [Online]. Available: - -https://www.cio.com/article/655693/7-cloud-market-trends-and-how-they-will-impact-it.html - -22 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -[6] Will Forrest et al., "Cloud’s trillion-dollar prize is up for grabs," McKinsey Digital, 2021. [Online]. - -Available: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion- -dollar-prize-is-up-for-grabs - -[7] Precedence Research, "Artificial Intelligence (AI) Market Size, Share, and Trends 2025 to 2034," - -2025. [Online]. Available: https://www.precedenceresearch.com/artificial-intelligence-market -[8] Abel E. Edje, M.S. Abd Latiff and Weng Howe Chan, "IoT data analytic algorithms on edge-cloud - -infrastructure: A review," ScienceDirect, 2023. [Online]. Available: -https://www.sciencedirect.com/science/article/pii/S2352864823001530#:~:text=In%20addition% -2C%20future%20advances%20are,Comparison%20of%20previous%20research%20surveys.&te -xt=Comparison%20of%20middleware%20architectures. - -[9] IvanDev, "The Rise of Serverless Architecture and Edge Computing," DEV, 2025. [Online]. - -Available: https://dev.to/ivansing/the-rise-of-serverless-architecture-and-edge-computing-2aab - -[10] Globe Newswire, "Hybrid Cloud Market Projected to Hit USD 405.62 Billion by 2032| SNS - -Insider," 2025. [Online]. Available: https://www.globenewswire.com/news- -release/2025/03/21/3047113/0/en/Hybrid-Cloud-Market-Projected-to-Hit-USD-405-62-Billion- -by-2032-SNS-Insider.html - -[11] Thales, "Cloud Security in 2024: Addressing the Shifting Landscape," CSA, 2024. [Online]. +19 -Available: https://cloudsecurityalliance.org/blog/2024/06/27/cloud-security-in-2024-addressing- -the-shifting-landscape + About Arab Barometer -[12] MAHESH NAWALE, "7 Key Takeaways From IBM's Cost of a Data Breach Report 2024," ZScaler, -2024. [Online]. Available: https://www.zscaler.com/blogs/product-insights/7-key-takeaways-ibm- -s-cost-data-breach-report-2024 +The Arab Barometer is a nonpartisan research network that +provides insights into the social, political, and economic attitudes +and values of ordinary citizens across the Arab world. -[13] Andrii Stashko-Ziablietsov and Oleksandr Virga, "Cloud strategy: a 2024 guide," Avenga, 2024. +We have been conducting rigorous, and nationally representative +face-to-face public opinion surveys on probability samples of the +adult populations across the Arab world since 2006. The margin of +error is ±3 percent. -[Online]. Available: https://www.avenga.com/magazine/cloud-strategy/ +The Arab Barometer is the largest repository of publicly available +data on the views of men and women in the MENA region. Our +findings give a voice to the needs and concerns of Arab publics. -[14] Willie Devitt, "Cloud Technology Trends and Predictions for 2024," Eviden, 2024. [Online]. +.......... -Available: https://eviden.com/publications/whitepapers/cloud-technology-trends-and-predictions- -for-2024/#:~:text=reputation%20and%20trustworthiness- -,At%20Eviden%2C%20we%20predict%20that%20in%202024%2C%20the%20focus%20on,com -pliance%20posture%20for%20our%20customers. +.......... -23 +.......... +ARABBAROMETER.ORGARABBAROMETER@ARABBAROMETER \ No newline at end of file diff --git a/q104/random_k4/question.json b/q104/random_k4/question.json index f618d510ca7c36c09eaf970ac19efe7276f4d831..4193ee46de605ec4794ab99f5bb310a91db7196a 100644 --- a/q104/random_k4/question.json +++ b/q104/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "e639029d16094ea71d964e2fb953952b.pdf", - "web_0867a04cdab7a273.pdf", - "web_a39c0d7f5ff0dedb.pdf", - "web_5906edbb3f342646.pdf", - "web_660e5c500c425b81.pdf" + "web_6a55c713826db217.pdf", + "web_bf09a64e7d517c98.pdf", + "web_715e6cf2d4ba3d94.pdf", + "web_d3b2a69cee942fe5.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q104/random_k4/random_1.md b/q104/random_k4/random_1.md index 85e510cedc0e76a41806e14e4f5679216b834d6f..5c5eba2e1fd980c21e8cbb1cd873801d5dedd049 100644 --- a/q104/random_k4/random_1.md +++ b/q104/random_k4/random_1.md @@ -1,1480 +1,2541 @@ -Historic Architectural Resources Survey Report + CONTENTS -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +01 +04 + 16 +36 +38 +39 +40 +41 +44 +45 -North Carolina Department of Transportation -Report Prepared by Nancy Van Dolsen +/ About HKU Business School -November 1999 +/ Explore Your Options - Historic Architectural Resources Survey Report +/ Enrich Your Life -Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +/ 2026 JUPAS Admissions -November 1999 +/ 2026 Non-JUPAS Admissions -Principal Investigator (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation +/ 2026 Dual Degree Admissions + +/ Develop Your Career + +/ Freshmen Sharing + +/ Our Global Presence + +/ Useful Websites + +About HKU Business School + +HKU Business School strives to nurture first-class business leaders and foster academic and relevant +research to serve the needs of Hong Kong, Chinese Mainland and the rest of the world in the new Asia-led +economy. As Asia’s premier international business school, we engage leading scholars from all corners of the +globe and they instil in students global knowledge with an Asian perspective. + +With the highest proportion of non-local undergraduate students across all Faculties at HKU, four of our +undergraduate programmes ranked amongst the University’s Top 10 programmes in 2025#, HKU Business +School attracts the best and brightest students from Hong Kong and beyond. + +HKU Business School is fully accredited by the European Quality Improvement Systems (EQUIS) and the +Association to Advance Collegiate Schools of Business (AACSB). + +# Based on average JUPAS admission scores in the best 5 subjects in HKDSE + +VISION +To be a leading, +globally-impactful academic +institution of business and +economics. + +UNIQUE +PROPOSITION + +Deeply rooted in Hong Kong +Fully engaged with China +Truly international + +MISSION +STATEMENT + +To provide world-leading +business and economics +education that nurtures +dynamic and responsible +leaders. +To advance the frontiers of +knowledge through innovative +and globally-significant +research. +To create positive change in the +world by partnering with +communities. + +01 + + Dean’s Message + +I am delighted to introduce you to the HKU Business +School, a dynamic and forward-thinking academic +community where transformative opportunities and +unparalleled experiences await. + +As a part of the University of Hong Kong, which +proudly ranks 11th globally in the 2026 QS World +University Rankings, HKU Business School stands out +as a beacon of excellence in business and economics +in Asia. This year marks the 25th anniversary of our +Faculty, we continue to march forward as a leading, +impactful business school, championing +globally +research, and +leadership, pioneering +thought +educational excellence. + +of + +dynamic, + +challenging + +responsible, + +Our mission is to inspire and develop the next +generation +and +globally-minded business leaders. We offer a vibrant +and +learning environment where +students are encouraged to push their limits, think +critically, and innovate. Our team of leading scholars +from around the world, and experienced and +dedicated teachers as well as mentors bring a wealth +of knowledge, global perspectives, and personalised +guidance beyond the classroom. Our full spectrum of +top-quality degree programmes, including 11 diverse +undergraduate programmes and four dual bachelor’s +degrees +in partnership with world-renowned +universities, is designed to equip students with +cutting-edge knowledge and technical skills to +in the fast-changing global +navigate and excel +business landscape. + +In response to the transformative power of artificial +intelligence, we are proud to launch the Second Major +and Minor programmes in “Artificial Intelligence in +Business” +to our +(AIB), offered exclusively +undergraduate students. These programmes are +designed to equip future business +leaders with +practical proficiency in AI technologies, strategic +insight to drive cross-sector innovation, and a critical +understanding of the ethical and societal implications +of AI adoption. Through a rigorous curriculum that +integrates business fundamentals with cutting-edge +AI applications, students will be empowered to tackle +complex challenges and create meaningful impact in +a data-driven world. + +Greetings from +Associate Dean (Undergraduate) + +inspiration + +Business education today is an evolving domain that +builds upon learning from all academic disciplines and +draws +successful and +from both +unsuccessful business practices. Our undergraduate +curricula in six core academic areas equip students +with diverse knowledge across business and +economics spheres, and prepare them to be creative +leaders of tomorrow with global perspectives. + +A hallmark of our School that separates us from the +others is our truly multi-disciplinary approach. HKU +Business School’s extensive resources offer students +in +a wide range of choices, combining studies +economics and finance, accounting and finance, +accounting and data analytics, business and law, +business and data analytics, international business +and global management, as well as dual degrees in +collaboration with world-renowned universities. + +Our faculty members will engage you with their +inspiring teaching and how they bring their impactful +research into the classroom. They will open your mind +to diverse perspectives, +ignite your curiosity to +explore innovative ideas, and instil in you an appetite +for acquiring new knowledge and extraordinary +experiences throughout your academic journey and +career development. Our students are among the +brightest and most creative of their peers. Our alumni +have gone on to forge brilliant careers in all sectors of +the +international business +community, and have formed a global network that +you will be a part of for the rest of your life. + +regional and + +local, + +I hope that you will share my enthusiasm for the +programmes and my admiration for the people who +have contributed to their development. I look forward +to having you as a member of the HKU Business +School family and a future leader in the world of +business. + +Warm regards, + +Professor Derek Chan +Associate Dean (Undergraduate) + +Through our extensive partnerships with over 300 +institutions and corporations across more than 40 +countries, as well as our unique network of centres in +Beijing, Shenzhen, Shanghai, Vietnam, and Barcelona, +our students gain unparalleled global experiential +and enrichment opportunities. Whether through +internships, overseas exchange programmes, or +hands-on +these experiences +sharpen practical skills, broaden perspectives, and +prepare students to confidently tackle real-world +challenges. + +industry projects, + +Joining HKU Business School means becoming part of +a community that supports your growth and success. +You will collaborate with brilliant minds, engaging in +stimulating discussions, and working on impactful +projects that make a difference. The friendships, +networks, and experiences you build here will last a +to a myriad of career +lifetime, open doors +opportunities, and empower you to overcome +challenges, pursue your passions with confidence, +and contribute positively to your community and the +world at large. + +If you are ready to embark on an exciting journey of +learning, discovery, and personal development, I +invite you to explore what HKU Business School has to +look forward to welcoming you to our +offer. I +community, where future leaders are nurtured and +dreams take flight. + +Warm regards, + +Professor Hongbin Cai +Dean +Chair of Economics + +0302 Explore Your Options +A Wide Selection of Programmes + +Bachelor of Business Administration [BBA] + +The BBA programme combines strong functional training with intensive exposure +to communication skills, computer applications, and other social science and/or +science subjects. Students can choose to major in the following area(s): + +Entrepreneurship, +Design and + Innovation* + +Finance + +Human Resource +Management + +Information Systems +and Analytics + +Marketing + +BBA Curriculum Structure + +Components -441-1/4,4 (cid:9) +A -Barbara Church, Supervisor (cid:9) -Historic Architectural Resources Section -North Carolina Department of Transportation +B -Date +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -//7/67/1. -Date +BBA Major +Faculty Core Courses +Disciplinary Core Courses / Disciplinary Electives +Capstone Course - TABLE OF CONTENTS +C -Project Description (cid:9) +Free Electives / Optional Major / Optional Minor(s) -Purpose of Survey and Report (cid:9) +Total -Methodology (cid:9) +Credits -Summary Findings of the Survey (cid:9) +54 +12 +36 +6 +- -Historic Contexts and Background Information (cid:9) +78 +24 +48 +6 -Property Inventory and Evaluations +108 -William Jones Arney House (cid:9) +240 -Marshall Arney House (cid:9) +*Candidates must undergo a selection process arranged by the Programme Coordinator for +declaring the EDI major. -Corpening, House (cid:9) +HKU – Sciences Po Dual Degree Programme -Early Twentieth-Century Frame House (cid:9) +Students will spend the first two years at Sciences Po in France and the third +and fourth years at HKU. Programmes at Sciences Po are taught in English (no +French requirements for being admitted) with intensive French language +classes for all non-francophone students. -Nineteenth-Century Log House (cid:9) +Upon successful completion of the dual degree programme, students will be +awarded the Bachelor of Business Administration (BBA) degree from HKU and +a Bachelor’s degree from Sciences Po. -Bibliography (cid:9) +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Appendix A: Correspondence with the State Historic Preservation Office (cid:9) +Find out more: +https://admissions.hku.hk/programmes/university- +collaborative-programmes/hku-sciences-po-dual- +degree-programme -LIST OF FIGURES +HKU – University of British Columbia +Dual Degree Programme -Figure 1: Project Location Map (cid:9) +Students will spend the first and final years at HKU and the second and third +years at the University of British Columbia (UBC) in Canada, and graduate with +a Bachelor of Business Administration from HKU and a Bachelor of Commerce +from UBC upon successful completion of the programme. -Figure 2: Area of Potential Effects with Historic Properties Noted (cid:9) +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Figure 3: Tax Map, William Jones Arney House (cid:9) +Find out more: +https://admissions.hku.hk/programmes/university- +collaborative-programmes/hku-university-of-british- +columbia-dual-degree-programme -Figure 4: Site Plan, William Jones Arney House (cid:9) +Find out more: +https://ug.hkubs.hku.hk/programme/bba -Figure 5: Tax Map, Marshall Arney House (cid:9) +0504 -Figure 6: Site Plan, Marshall Arney House (cid:9) + Bachelor of Business Administration in Accounting and +Finance [BBA(Acc&Fin)] -Page +Bachelor of Business Administration in Accounting Data +Analytics [BBA(ADA)] - 1 +The BBA(Acc&Fin) programme prepares students for a professional qualification +in accounting and provides the broad-based education in accounting and finance +which is necessary to progress toward a leadership role in the financial sector. +Students are required to take professional core in accounting and a minor in +finance, with an option to second major in finance. Tailor-made professional +internship opportunities with renowned +preparation programmes and +accounting firms and financial institutions are available for talented students. It +is accredited as an approved accountancy programme by: -1 +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia -1 +BBA(Acc&Fin) Curriculum Structure -4 +Components - 6 +Credits -8 +A - 15 +B -24 +C -28 +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Professional Core in Accounting +Faculty Core Courses +Disciplinary Core Courses +Capstone Course + +Minor in Finance* +Disciplinary Core Courses +Disciplinary Electives + +D. -31 +Free Electives / Optional Minor -.32 3 +*Students may opt to pursue a Major in Finance: -.10 11 +Components -.14 .15 +A - Pave SR 1424, Johns River Road and SR 1438, Corpening Bridge Road -Burke County, North Carolina -State Project No. 6.852109 +B -PROJECT DESCRIPTION +C -The North Carolina Department of Transportation (NCDOT) has plans to pave SR 1424, -Johns River Road and SR 1438, Corpening Bridge Road, Quaker Meadows and Upper -Creek townships, Burke County (Figure 1). In addition, the NCDOT plans to minimize -curves in the roadway. A nationwide permit will be needed for the project. +University Graduation Requirements -The area of potential effects (APE) for historic architectural resources was delineated by -a NCDOT staff architectural historian and reviewed in the field. The APE boundary is -shown on the attached map (Figure 2). +Professional Core in Accounting -PURPOSE OF SURVEY AND REPORT +Major in Finance +Disciplinary Core Courses +Disciplinary Electives +Free Electives +Capstone Course -This Final Identification and Evaluation survey was conducted and the report prepared in -order to identify historic architectural resources located within the APE as part of the -environmental studies conducted by NCDOT and documented by a Categorical Exclusion -(CE). This report is prepared as a technical addendum to the CE and adheres to the -guidelines for compliance with the National Historic Preservation Act of 1966, as -amended. This report is on file at NCDOT and is available for review by the general -public. +54 +12 +36 +6 +- -METHODOLOGY +108 +24 +78 +6 -NCDOT conducted the survey and prepared this report in accordance with the provisions -of FHWA Technical Advisory T6640.8A (Guidance for Preparing and Processing -Environmental and Section 4(f) Documents); the Secretary of the Interior's Standards and -Guidelines for Archaeological and Historic Preservation (48 FR 44716); 36 CFR Part -800; 36 CFR Part 60; and Survey Procedures and Report Guidelines for Historic -Architectural Resources by NCDOT. This survey and report meet the guidelines of -NCDOT and the National Park Service. +36 +12 +24 -NCDOT completed an intensive survey with the following goals: (1) to determine the -APE, defined as the geographic area or areas within which a project may cause changes -in the character or use of historic properties, if any such properties exist; (2) to identify +42 -Historic Architectural Resources Survey Report; NCDOT,ISR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen +Total -Page 1 +240 - Figure 1: Project Location Map +Credits -Historic .4rchitectural Resources Survey Report: NCDOT, SR 1411, Burke County: November 1999 (cid:9) -Nancy Van Dolsen +54 -Page ? +108 + +78 +30 +18 +24 +6 + +Total + +240 + +Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares +students for a professional qualification in accounting. It provides students +with an option to focus on data analytics aspect in accounting instead of having +a minor in finance. It is an excellent choice for students who are interested in +STEM with accounting and business applications and wish to pursue their +career in the relevant fields. The intensive accounting training enables +BBA(ADA) students to distinguish themselves from students of the other data +analytics/STEM programmes. Students will be equipped with comprehensive +accounting and business knowledge that prepares them to obtain professional +qualifications in accounting. They will also acquire computer programming +and data analytics skills to meet the needs of the accounting and business +professions in today’s digital transformation economy. It is accredited as an +approved accountancy programme by: + +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia + +BBA(ADA) Curriculum Structure + +Components - S -E -C -R -U -O -S -E -R -L A -R -U -T -C -E -T -I -H + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Professional Core in Accounting Data Analytics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + C -R + +Free Electives + +Total + +Credits + +54 +12 +36 +6 +- + +156 +24 +114 +12 +6 + +30 + +240 + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-acc-and-fin + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-ada + +0706 + + Bachelor of Business Administration (Business Analytics) +[BBA(BA)] + +Bachelor of Business Administration (Law) and Bachelor of +Laws [BBA(Law)&LLB] + +The BBA(BA) programme is crafted to meet the growing industry demand for +talents in the business analytics field. Its curriculum offers students a +comprehensive coverage of both technical and managerial skill sets. Students +will learn a wide spectrum of knowledge from the disciplines of information +technology, data science, business statistics and management. Core courses +include programming language, database management, introduction to +business analytics, management information systems, decision and risk +analysis, etc. They help students build the foundations for advanced topics in +senior years. Students also have the option to choose from analytics technique +courses and business application courses. The aim is to provide students with +state-of-the-art experiences in artificial intelligence, big data, social media, +and other latest development. Such design helps train well-rounded students +who will be more adaptable in their career development. The programme is +suitable for students who have a passion for problem-solving through +analytics, and who aim to develop their career path in analytics. + +BBA(BA) Curriculum Structure + +Components + +Credits + A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Business Analytics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives (Analytics Techniques) +Disciplinary Electives (Business Applications) +Capstone Course + C -I -R -O -T -S -I -H - -P -A -M -Y -E -V -R -U -S - -e -l -t -i - -T - -t -e -e -h -S - -: -§ - -9 -0 -1 -2 -5 -8 -. + +Free Electives / Optional Major / Optional Minor(s) + +54 +12 +36 +6 +- + +96 +24 +42 +12 +12 6 -8 -3 -4 -1 -R -S -& -4 -2 -4 -1 -R -S +90 + +Jointly offered by the HKU Business School and the Faculty of Law, the +BBA(Law)&LLB integrated double-degree programme blends two strategic +disciplines into a single programme for the bright and highly motivated. It is an +excellent choice for students seeking a broader education with parallel +concentration on legal and business studies. Graduates of the programme will be +awarded two degrees — BBA(Law) by the HKU Business School, and LLB by the +Faculty of Law. After successful completion of the LLB programme, students +may apply for admission to the Postgraduate Certificate in Laws (PCLL) +programme provided that they meet the relevant admission requirements. + +The programme offers business stream and accounting stream to students in +their business studies. The accounting stream is accredited by: + +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia + +BBA(Law)&LLB Curriculum Structure (Business Stream) + +Components + +Credits A -N -I -L -O -R + +B + +C + +University Graduation Requirements +Language Enhancement Course +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +BBA Major +Faculty Core Courses +Disciplinary Core Courses / Disciplinary Electives +Capstone Course + +Professional Core in Law +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + +D. + +Free Electives / Optional Minor + +Total + +240 + +Total + +BBA(Law)&LLB Curriculum Structure (Accounting Stream) + +Components + A + +B + +University Graduation Requirements + +Professional Core in Accounting +Faculty Core Courses +Disciplinary Core Courses +Capstone Course + C -H -T -R -O -N - -, - -Y -T -N -U -O + +Professional Core in Law + +Total + +36 +6 +24 +6 +- + +78 +24 +48 +6 + +156 +108 +42 +6 + +30 + +300 + +Credits + +36 + +108 +24 +78 +6 + +156 + +300 + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-ba + +Find out more: +https://ug.hkubs.hku.hk/programme/bba-law-and-llb + +0908 + + Bachelor of Business Administration in International Business +and Global Management [BBA(IBGM)] + +Bachelor of Economics [BEcon] + +BBA(IBGM) is an intensive, premier programme in international business and +global management featuring a unique curriculum designed to evolve with and +reflect the ever-changing demands of a globalised world. Exclusive elements +such as overseas exchange studies, a leadership boot camp, an Asia-Pacific +study tour, an international field trip and global enrichment opportunities are +strategically integrated into the curriculum, whilst students are given the +flexibility to pursue an in-depth inquiry into another specialised area of study +as a second major. This cross-functional and personalised approach aims to +foster a selected group of outstanding students to become the world’s future +leaders equipped with global perspectives, cultural sensitivity, social +responsibility and key competencies to succeed in the international business +environment. + +BBA(IBGM) Curriculum Structure + +Components + +A + +B + C + +D + E -K -R -U -B -t -c -e -j -o -r -P +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -i -c -F +Major in International Business and Global Management +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -2 +Second Major -. +Free Electives -W -z +Compulsory Overseas Exchange -p +Total -l -l -i +Credits -) -a +54 +12 +36 +6 +- -• +96 +24 +54 +12 +6 - i -0 - 1 +72-90 -g +0-18* -5, -g +- -1 +240 -dVIA41 A3AWS S302:10S3 1V1if1.1031.1H3lV ONOISIH +*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. -SIN :37VOS +The BEcon programme trains students in economic theories, statistical tools, +the application of theories to solve practical economic problems in trade, +banking and the labour market, and evaluation of the regional economic +performance of Chinese Mainland, Hong Kong and other economies. Students +will be furnished with solid theoretical and empirical economic knowledge to +analyse and understand human behaviour and social interactions. The +programme allows students a high degree of flexibility in planning their future +career specialisation. -— +BEcon Curriculum Structure -I +Components -g +A -§ +B - all significant resources within the APE; and (3) to evaluate these resources according to -the National Register of Historic Places criteria. +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -The survey methodology consisted of a field survey and background research on the -project area. A NCDOT staff architectural historian conducted a field survey in August -and October 1999, by car and on foot. All structures over 50 years of age in the APE -were photographed and keyed to an area map. +Major in Economics +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -Background research was conducted at the Western office of the State Historic -Preservation Office, Asheville; the North Carolina State Library, Raleigh; and the Burke -County Court House in Morganton. +C -A meeting was held with Ms. Jennifer Martin at the State Historic Preservation Office on -August 20, 1999 to discuss the eligibility of two of the historic resources, the William -Jones Arney House and the Marshall Arney House. It was agreed at that time the two -properties were potentially eligible for the National Register of Historic Places. +Free Electives / Optional Major / Optional Minor(s) -SUMMARY FINDINGS OF THE SURVEY +Total -The project proposes to pave and improve SR 1424, Johns River Road and SR 1438, -Corpening Bridge Road in Quaker Meadows and Upper Creek townships, Burke County. -In letters dated July 9, 1999 and August 10, 1999, the State Historic Preservation Office -identified four structures of historical or architectural importance within the general -project area, the Marshall Arney House, the William Arney House, the Arney Chapel and -the Corpening House (See Appendix A). Two additional historic properties, not -mentioned in the letters, are located within the project area: a late nineteenth-century log -house and an early twentieth-century frame house. No properties listed on the National -Register were located within the APE for the project. Of the four historic properties -mentioned in the letter, one, the Arney Chapel, is located outside the APE. Of the five -historic properties located within the APE, three—the Corpening House, the nineteenth- -century log house, and the early twentieth-century frame house --were determined not -eligible for the National Register. The Marshall Arney House and the William Arney -House were determined to be eligible for the National Register. Criterion Consideration -G, for properties that have achieved significance within the last fifty years, states that -properties less than fifty years of age may be listed on the National Register only if they -are of exceptional importance or if they are integral parts of districts eligible for the -National Register. There are no properties in the APE that qualify for the National -Register under Criterion Consideration G. +Credits -Properties Listed on the National Register: +54 +12 +36 +6 +- -None +78 +24 +30 +18 +6 -Ilistoric Architectural Resources Survey Report: iVCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nancy Ian Dolsen +108 -Page 4 +240 - Properties Listed on the North Carolina State Study List: +Find out more: +http://ibgm.business.hku.hk -None +Find out more: +https://ug.hkubs.hku.hk/programme/becon -Properties Evaluated and Considered Eligible for the National Register: +1110 -William Jones Amey House, page 8 -Marshall Amey House, page 15 + Bachelor of Economics and Finance [BEcon&Fin] -Properties Evaluated and Considered Not Eligible for the National Register: +BEcon&Fin is designed to meet the interests of students who are seeking a +balanced study of both economics and finance. The programme recognises +the significance of economics both in its own right and as a foundation for the +study of finance, and the significance of finance as a career path for students +who are keen on joining this profession. Students will receive solid training in +theories and applications of economics and finance. They will be equipped +with the skills to analyse human behaviour and social interaction, as well as to +solve problems in the financial sphere. -Corpening House, page 21 -Twentieth-century frame house, page 24 -Nineteenth-century log house, pay 28 +BEcon&Fin Curriculum Structure -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen +Components -Page 5 +Credits - HISTORIC CONTEXTS AND BACKGROUND INFORMATION +A -During the late seventeenth century, occupation of the land of present-day Burke County -was contested between two native-American tribes, the Cherokee and the Catawba. The -area eventually became a buffer zone between the two tribes. The French and Indian War -(1754-1763) and the American Revolution had a catastrophic effect on both tribes, -forcing the Cherokees further into the mountains, and reducing the Catawba population to -250 members by 1784. +B -Transient hunters and trappers of European descent settled in the county during the -middle of the eighteenth century. In 1777 the General Assembly approved the formation -of a new county named Burke in the western portion of the state that encompassed all or -part of 16 present counties. Burke was named for Thomas Burke (1747-1783), a -member of the Continental Congress and third Governor of North Carolina.' Burke -County achieved its present size of 511 square acres in 1885. +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -According to an architectural survey of the county completed in 1983-1984, there are -three distinct regions in the county: the rich bottom lands along the Catawba, Johns, -Upper and Lower Creeks, and Warrior Forks rivers; the remote dead-end coves of the -South Mountains and Blue Ridge; and the rolling foothills interspersed with small -valleys.2 The subject project is located along the bottom land of the Johns River in -Quaker Meadows and Upper Creek townships. +Major in Economics and Finance +Faculty Core Courses +Disciplinary Core Courses +Economics Disciplinary Electives +Finance Disciplinary Electives +Capstone Course -The bottom lands contained the richest agricultural soil in the county, and were the site of -antebellum plantations. Most of the county's large, brick plantation homes dating to the -first half of the nineteenth century stand along the Johns and Catawba rivers in Quaker -Meadows and Upper Creek townships, including Bellevue (1823-1826), Cedar Grove -(1825), and Pleasant Valley (ca. 1820; ca. 1975). In Burke, the typical plantation of 1860 -consisted of 850 acres of improved land and thousands of acres of unimproved land. -The typical 1860 Burke County farm, however, contained only 60 acres of cultivated land -and only 60 out of 1,239 families owned more than 100 acres of improved land.3 - -After the Civil War, the land-use pattern of ownership changed. The large-scale -plantations were dissolved, and there was an increase in the number of small-scale farms, -from 500 in 1860 to 1,648 in 1880. At the same time, the average farm size dropped to -27 acres.4 - -The project area is the site of five post-bellum houses that date to this period of -subdividing larger plantation properties. The houses stand on land first owned by one of -Burke County's first European settlers, John Perkins, who first explored the county in the - -I Entry for Thomas Burke, Biographical Directory of the United States Congress, I774-Present, Website: -Bioguide.conress.gov. -2 J. Randall Cotton, et at., Historic Burke: An Architectural Sites Inventory of Burke County (Morganton, -NC: Historic Burke Foundation, Inc., 1987) 13. -3 Cotton, et at, 23. -4 Cotton, et al, 27. - -Historic Architectural Resources Survey Report: NCDOT. SR 1424. Burke County: November 1999 -Nancy Van Dolsen - -Page 6 - - 1750s. These five houses represent a range of building types constructed in the county -during the last quarter of the nineteenth century: two frame T-shaped houses, a small -two-room log house, and two square-shaped two-story frame dwellings. - -At the end of the twentieth century, much of the land in the county is held by three -property owners: the US Government (Pisgah Forest), the state (South Mountain State -Park) and Carolina Centers, a timber company. Carolina Centers presently owns the land -upon which the Corpening, Marshall Amey, and William Jones Arney houses stand. - -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Dolsen - -Page 7 - - PROPERTY INVENTORY AND EVALUATIONS +C -William Jones Arney House +Free Electives / Optional Major / Optional Minor(s) -Location: (cid:9) -The William Jones Arney House is located on the northwest corner -of the junction of SR 1424 and SR 1438. The house sits on a 23,136.10-acre tract -owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a timber -company. The property was formerly owned by Duke Power, Inc. - -The William Jones Arney House sits within a cleared area, at the - -Setting: (cid:9) -top of a rise, facing east to Johns River. A short dirt lane leading from Johns -River Road runs south of the house, ending at a small frame granary west of the -dwelling. - -Description: - -Dwelling: The William Jones Arney House, dating to ca. 1890, is a two- -and-a-half story frame residence with a T-shaped plan. The exterior of the -house is clad with narrow weatherboards, and a metal roof covers the -building. To the rear of the building is a kitchen connected to the main -house by a covered walkway. The kitchen stands one-story and has a -shed-roofed porch on each eave elevation. The windows in the kitchen -contain six-over-six sash and those in the main block of the house have -two-over-two sash. A one-story tripartite bay window adorns the south -gable end wall of the ell. Pointed arched louvered windows are located at -the peaks of the gable end walls. Two interior brick chimneys are located -at the center of the ells, providing an interior stack to each room in the -house. +54 +12 +36 +6 +- -The interior walls of the house are sheathed in horizontal boards, as are the -ceilings. The house (excluding the kitchen) has four rooms on the first -and four rooms on the second floor. +96 +24 +42 +12 +12 +6 -Well House: A concrete block well house stands south of the south porch -of the kitchen. The well house is approximately 3 feet in height and is -covered with a corrugated metal roof. +90 -Granary: To the west of the house stands a frame granary also constructed -ca. 18900. The granary building is a gable-front building with a central -door. The building is weatherboarded and has a metal roof. An open side -shed was constructed onto the east eave wall, and a shed roof porch is -located on the south elevation. +Total -Background Information: (cid:9) -land ca. 1885. Arney, a descendent of one of the pioneering families in the -county, both farmed and owned a general store (now demolished). The store +240 -William Jones Arney (1849-1911) moved to this +HKU - Peking University “Future +Leaders” Dual Bachelor’s Degree in +Economics, Finance and Management +The programme combines the strengths of two leading business schools in +Hong Kong and Chinese Mainland and is designed to nurture first-class business +leaders and cultivate their global perspective with intimate understanding of +China. Students will spend the first two years at HKU and the third and fourth +years at Peking University (PKU), and be awarded a Bachelor of Economics and +Finance (BEcon&Fin) degree from HKU and a Bachelor of Management [BMgmt] +degree from PKU upon successful completion of the dual degree programme. -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy ran Dolsen +Separate application is required for applying to the dual degree programme, in +addition to the application under the JUPAS or International/Non-JUPAS +Admissions Scheme. -Page 8 +Find out more: +https://ug.hkubs.hku.hk/admission/degree-program +mes/hku-pku-dual-degree - stood to the west, at the intersection of Corpening Bridge and Piedmont roads. -Arney, and his wife Margaret McCall, owned over 500 acres in the area along -John's River. The couple had nine children? +Bachelor of Finance in Asset Management and Private +Banking [BFin(AMPB)] +The BFin(AMPB) programme aims to prepare students to enter the rapidly +growing fields of asset management and private wealth management in Hong +Kong and Asia Pacific, or in banking and finance generally. During the first two +years, students complete introductory courses in accounting, economics, +mathematics, statistics, corporate finance, and investments and portfolio +analysis. In subsequent years, students take more specified and industry +related courses including equity valuation and investment in specific asset +classes e.g. private equity, venture capital, hedge funds and digital assets, +understanding of financial statements and doing financial analysis, lending +and credit in private wealth and corporate bank, financial regulations and +compliance, current +institutional +in finance, ESG +consulting and family office. The programme contains practical-oriented +courses taught by teachers with extensive experience in the finance industry +and offers an array of career-oriented enrichment activities and networking +opportunities with industry practitioners. -At the death of William and Margaret Arney, ownership of the part of the farm -upon which this house stood was transferred to George Phifer Arney who sold -240 acres to Western Carolina Power for $100 in 1922.6 The property is presently -owned by Carolina Centers, LLC. +investment, -For purposes of compliance with Section 106 of the National +issues -Evaluation: (cid:9) -Historic Preservation Act of 1966, as amended, the William Jones Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A, since it is not representative of a significant trend in the history of the -area. The property is no longer under cultivation, and all but one of the -agricultural outbuildings have been demolished or removed. +in -The William Jones Arney House is considered not eligible for the National -Register of Historic Places under Criterion B because it is not associated with any -individuals of specific historical significance. +BFin(AMPB) Curriculum Structure -The William Jones Arney House is considered eligible for the National Register -under Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is a regional form built by farmers of middle economic status. The Arney -House is a good representation of this form, and exhibits excellent integrity and -retention of setting. The granary building is also representative of a vernacular -building form, and exhibits good integrity of design, materials, workmanship, -and setting. +Components -The building is not likely to yield information important to the history of building -technology; therefore it is not eligible for the National Register under Criterion D. +Credits -Boundaries: The proposed National Register boundaries comprise approximately -2.8 acres that surround the house and its accompanying outbuilding. The land -was deeded to George Phifer Arney, and is the portion of the William Jones -Arney estate upon which the subject structures stand. This parcel is part of parcel -JR-21, which is a portion of the larger Carolina Centers tract. +A -Boundary Justification: (cid:9) -include the house, outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. +B -The proposed historic boundaries were drawn to +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University -5 William Jones Amey House, Historic Resource Survey Form, Burke County, 1983. Asheville, NC: -Western Office, Division of Cultural Resources. -6 Burke County Deed Book T4:293. +Major in Asset Management and Private Banking +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 -Nancy Van Dolsen +C -Page 9 +Free Electives / Optional Major / Optional Minor(s) - Figure 3: Tax Map, William Jones Arney House +54 +12 +36 +6 +- -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke County; November 1999 (cid:9) -Nancy Van Dolsen +96 +24 +42 +24 +6 -Page 10 +90 - Figure 4: Site plan, William Jones Arney House +Total -Historic Architectural Resources Survey Report,- NCD07; SR 1-114, Burke County: November 1999 (cid:9) -Nancy l'an Dolsen +240 -Page 11 +Find out more: +https://ug.hkubs.hku.hk/programme/becon-and-fin - — +Find out more: +https://ug.hkubs.hku.hk/programme/bfin-ampb -— (cid:9) +1312 -_ + Bachelor of Science in Marketing Analytics and Technology +[BSc(MAT)] -- (cid:9) +Bachelor of Science in Quantitative Finance [BSc(QFin)] -41E +Marketing is critical to all businesses. It dictates whether a company will be +successful. Technology, Big Data, and digitization are fundamentally changing +how marketers operate. The BSc(MAT) programme equips students with +quantitative skills, technology know-how, and business marketing acumen to +pursue a career in the digital economy and modern businesses. It targets both +STEM students with an interest in business as well as business students +interested in the technology space. Students will gain a solid foundation in +cutting-edge technical areas including computer programming, data science, +and technology by excelling in courses from the Faculty of Engineering and +Faculty of Science. In addition, students will master digital business, +marketing and leadership through courses at the HKU Business School. This +interdisciplinary approach will enable students to engage in a variety of jobs +and functions such as the design and launch of new technology products, +targeting consumers, digital marketing, forecast and predicting markets, and +building a technology-focused strategy. The programme aims to nurture +future leaders in the field of technology and marketing who are able to use the +latest technology, science, and data analytics to support business and +marketing decisions in the digital world. -re-7_. - — -liq -- -- . (cid:9) +BSc(MAT) Curriculum Structure + +Components + +Credits + +A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Marketing Analytics and Technology +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives (Marketing Technology and Strategy) +Disciplinary Electives (Marketing Analytics) +Capstone Course + +C + +Free Electives / Optional Major / Optional Minor(s) +54 +12 +36 +6 - -_ +96 +24 +42 +12 +12 +6 + +90 -_ (cid:9) +Total -_ +240 -_- +BSc(QFin) is a premium finance programme. Built on a solid foundation in +economics and finance, the programme equips students with the essential +quant skills, including mathematics, statistics, and programming, that are +required in more advanced and sophisticated analyses, such as valuation +models of structured products, risk management, simulation and data +analysis, financial engineering, and quantitative trading. -- --- +The programme goes beyond merely equipping students with quant skills. +More importantly, it stresses on combining these skills with an understanding of +the financial markets, the macro environment, and the practical situations +where financial managers work, with a balanced mix of quant and business +courses. To nurture students’ soft skills and market acumen, the programme +provides a wide range of value-enhancing activities, such as Round Tables with +Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training +workshops, simulation trading workshops, and an international field trip to +other international financial centres such as New York, London and Zurich. -— s- — +BSc(QFin) Curriculum Structure -_--..------ +Components +Credits + +A + +B + +University Graduation Requirements +Language Enhancement Courses +Common Core Curriculum +Artificial Intelligence Literacy Courses +Non-Credit Bearing Courses as required by the University + +Major in Quantitative Finance +Faculty Core Courses +Disciplinary Core Courses +Disciplinary Electives +Capstone Course + +C + +Free Electives / Optional Major / Optional Minor(s) + +54 +12 +36 +6 - -- (cid:9) -77--- +96 +24 +54 +12 +6 --_ +90 -- (cid:9) +Total -—. (cid:9) +240 -—______-- -- ------_--,- --- ---------- +Find out more: +https://ug.hkubs.hku.hk/programme/bsc-mat -_ +Find out more: +https://ug.hkubs.hku.hk/programme/bsc-qfin -_ --- — +1514 -- -- + Enrich Your Life --4-. - -, +initiative and +HKU Business School fosters a culture of student +implementation of ideas, both within and outside the campus, as it is +these real-world engagements that set us apart. Experiences in learning +beyond the classroom, extending into realms outside academia, enable +students to cultivate an understanding of social dynamics, cultural +diversity, and collaborative +intellect. These experiences contribute +significantly to their personal growth, interpersonal skills, and social +development, shaping them into global citizens, lifelong learners, and +future business leaders. --4 +Overseas Enrichment Experiences ----_ +Business Exchange Programme / +HKUWW Undergraduate Student Exchange Programme +Even as the youngest Faculty at the University, our exchange +activities are nothing but phenomenal. Each year, over 400 +students participate in overseas exchanges at leading universities +to broaden their horizons. -a +Business School +Exchange Partners * -Photograph 1: William Jones Amey House, facing southwest +Canada -Photograph 2: William Jones Amey House, facing northeast +McGill University +University of British Columbia +University of Toronto +Western University +Simon Fraser University -Historic Architectural Resources Survey Report: NCDOT SR 1424, Burke County: November 1999 -Nancy Van Dolsen +U.K. -Page 12 +King's College London +University of Manchester +University of Edinburgh +University of Warwick -(cid:9) -(cid:9) - . +The Netherlands -:t. (cid:9) +University of Amsterdam +Erasmus University Rotterdam -. --- +400+ +exchange partners +across 40+ countries -, +400+ +Business School +students on exchange -,,J-, (cid:9) +Denmark -, +Copenhagen Business School +Aarhus University -••••- (cid:9) +Germany -, (cid:9) +Frankfurt School of Finance & Management +University of Mannheim +Munich Business School -i +Belgium --- (cid:9) -_ -__ +KU Leuven --.------ (cid:9) +Italy -----.•-•_:, (cid:9) ----- +Bocconi University -(cid:9) ,- +France --.? +U.S. ---- -_= +Toulouse School of Economics +ESSEC Business School -- (cid:9) +New York University +University of Southern California +University of Washington +University of Georgia -_ +Middle East -- +Reichman University -. (cid:9) +Singapore --- (cid:9) -- (cid:9) +National University of Singapore +Nanyang Technological University -- -•- -_ --4--.- -— -- (cid:9) +* This list is non-exclusive. + +International Field Trips +To enhance students' exposure to global businesses, the School organises +overseas field trips for students visiting corporate firms and mingling with +C-suite Executives in some of the world's most prominent financial +centres, e.g. Australia, France, Japan, London, New York City, Singapore, +Spain, and Zurich. + +Visiting Programmes +Through the Yale Visiting International Students Programme, students +are sent to spend a year at Yale University, where they actively engage +themselves in a multicultural society. With the generosity of the donors, +some of our students are selected for the Junior Year Abroad Programme +and Visiting Students Programme to study as visiting students at +Pembroke College and St Peter’s College of University of Oxford +respectively with substantial scholarships. The intellectual ambience of +these top universities is truly inspirational and conducive to learning. -------,_ +Beta Gamma Sigma Regional Leadership Conference +Beta Gamma Sigma (BGS) is the international honour society that +serves business programmes accredited by the Association to Advance +Collegiate Schools of Business (AACSB International). With the +establishment of the HKU BGS Chapter, the School registers students +annually for the BGS Regional Leadership Conference held in the U.S., +where they can meet and bond with many other fellow students from the +world's best business schools. The Leadership Summit is eye-opening +and challenging, which leads to a great journey of self-discovery. -- +Japan + +Kyoto University + +Australia + +University of Sydney +Monash University +University of New South Wales + +1716 Student Sharing + +Chen, Hsuan-yu +Exchange at National University of Singapore (Singapore), 2024-25 + +As a third-year undergraduate student pursuing a double major in +Accounting and Finance, I consider it a tremendous privilege to have +participated in a one-semester exchange at the National University of +Singapore (NUS) through the Business Exchange Programme. This +invaluable opportunity has not only enriched my academic journey but +also broadened my global perspective through immersive cultural +experiences and meaningful cross-cultural interactions. As I move +forward in my academic and professional journey, I carry with me not +just the knowledge gained from my courses, but the invaluable life +lessons experienced from Exchange. + +Lee, Tsz To Derek +Exchange at King’s College London (United Kingdom), 2024-25 + +I am incredibly grateful to have this valuable opportunity to study +abroad at King’s College London. The programme provided me a chance +to venture outside of my comfort zone. Not only can you learn how to be +more independent, you can also boost interpersonal skills given the +many opportunities to socialise with others, which is essential for your +future career. The experience is remarkable. I think you will regret it if +you miss out on the overseas exchange programme. + +Cheung, Nga Ting Caroline +Exchange at University of New South Wales (Australia), 2024-25 + +This exchange programme has changed me in many ways. It has helped +me become more outgoing, allowing me to make new friends and +explore different cities. Living independently has been a significant part +of this journey. From exploring new cities to forming lasting friendships, +each adventure has enriched my life in ways I never expected. This +opportunity has not only broadened my horizons but also deepened my +appreciation for diversity and connection among people from different +backgrounds. Each memory I have created during this time is a treasure +that I will carry with me forever. + +Ho, Ying Tung Crystal +Exchange at Copenhagen Business School (Denmark), 2024-25 + +In the Fall semester, I participated in a business exchange programme at +Denmark’s Copenhagen Business School (CBS). This experience was +truly transformative, enriching me both academically and personally +while allowing me to forge lifelong memories and friendships. An +exchange experience abroad truly broadens your horizons and offers +insights beyond what textbooks and lectures can provide. You will gain +invaluable perspectives and undergo personal growth like never before. +I wholeheartedly encourage you to muster your courage and embark on +an exchange journey - you won’t regret it! + +Whitfield III, Gerald Bruce Trey +Exchange at New York University (United States), 2024-25 + +I found the exchange experience to be extremely enriching. A major +highlight for me was exploring the lifestyle and professional sides of the +city. Because of my initiative socially, and in groups such as church, I was +able to get well-acquainted with the local rhythms and lifestyle, and +also made many friends that I’d consider friends for a lifetime. I am +deeply appreciative of being given this rewarding experience, which +significantly shaped both my future and the person I am today. + +Ip, Wing Tung Alla +Student Representative for the BGS Regional Leadership +Conference 2025 in Highland Heights, Kentucky + +Attending the BGS Leadership Conference at Northern Kentucky +University proved transformative. I received profound insights into +leadership, communication and cultural intelligence through immersive +workshops, case competitions and networking opportunities. The event +prompted reflections on personal leadership styles and encouraged +collaboration across diverse peers. As I prepare to enter the workforce, +I intend to apply these learned principles of trust-building, cultural +sensitivity and collaborative leadership to navigate challenges so I can +shape my professional journey with purpose and authenticity. + +1918 Corporate Exposure + +HKU Business School has a strong tie with corporate partners. To nurture students' business +and market acumen, enrichment programmes are organised through the joint efforts of +corporate partnerships. + +Enrichment Programmes in collaboration with Corporate Partners +HKU Business School has been collaborating with corporate partners on initiatives to spice up +students' career exploration journey. The partnership with Accenture enables students to +gain exclusive insights into business consulting through the HKU-Accenture Business +Consulting Programme. + +To provide students with a unique opportunity to explore current business trends and +developments in the Greater Bay Area (GBA), our School has co-organised a tailored study tour +with The Association of Chartered Certified Accountants (ACCA). This initiative aims to +enhance students’ readiness for future career development and personal success. Students will +have the chance to connect with seasoned accounting and business professionals in the GBA. + +HKU Business School Mentorship Programme +Leveraging the School's extensive network in the business and finance industry, we have +developed our own mentorship programme to furnish students with opportunities to +enhance their skills and knowledge essential for their future personal and career +development. Our students are empowered to connect to the real business world through +direct and active interaction with seasoned professionals, executives and entrepreneurs. + +Lee, Lok Yiu Angel +HKU Business School Mentorship Programme 2024-25 + +My mentor offered +invaluable career advice, highlighting key +considerations for entering specific fields and preparing for interviews. +Her experiences and mindset during job searches resonated with me as +I navigated something similar. She emphasised the importance of +open-mindedness and eagerness to learn in the workplace. I kept her +words at heart during my internship. I also enjoyed catching up with her +to share my new job experiences. + +Choi, Han Na Hannah +HKUBS x Deloitte China Mentorship Programme 2024-25 + +As my mentor has been working in Deloitte China for more than 10 +years, I was most curious about how he maintains a work-life balance +and how he stays dedicated and professional at work. Through our +lengthy conversations over lunch, he passed on valuable insights on +developing soft skills and adopting an appropriate work attitude at an +early stage. He is more than an audit mentor but also a life coach to me. + +Chan, Chin Ching Jacky +BSc(QFin) Executive Mentorship Programme 2024-25 + +Throughout the mentorship programme, I have received +precious guidance that has greatly influenced my career +path. My mentor highlighted the importance of identifying +a company that aligns with my goals and would provide +opportunities for me to develop the necessary skill sets. He +also shared practical advice on building and managing +professional and personal relationships. These lessons +have broadened my perspective and also equipped me +with the tools to handle real-world situations more +effectively and confidently. + +Wong, Wan Tat +HKUBS x ACCA Greater Bay Area Study Tour 2024-25 + +My greatest takeaway from this study tour was the power of peer +learning and collaboration. Being immersed in an environment with a +diverse group of individuals, all focused on similar goals, created an +incredible opportunity to learn from each other’s experiences and +perspectives. The collaborative projects and informal discussions were +invaluable in broadening my horizon and developing new approaches +to the accounting field in Chinese Mainland. I am leaving with a strong +network of contacts and a renewed appreciation for teamwork. + +Leung, Ho Lun Maverick +HKU-Accenture Business Consulting Programme 2024-25 + +I visited + +The HKU-Accenture Business Consulting Programme immersed me in +an authentic consulting experience through working with a real client – +their distribution centres, conducted market +HKTV. +questionnaires, and presented the final deliverables - all within the +eight-week programme. I was honoured to meet Managing Directors +from Accenture who shared crucial skills in conducting a consulting +project. It’s truly one of the best programmes for anyone considering a +consulting career! + +2120 Experiential Learning Programmes + +Experiential learning is embedded in our course curriculum. The Business Consulting +Practicum offers business and economics students hands-on experience in working with +clients to solve "real-life" problems relevant to their studies, under the guidance of faculty +members and professional mentors. The Business Case Analysis and Presentation enables +students to communicate and seek practical advice from senior executives through in-class +case presentations. The Impact Lab provides students with an excellent opportunity to work +directly with social ventures, again under the guidance of teachers and professional mentors. +Additionally, students can gain an in-depth understanding of the latest trends in economics, +finance, creativity, innovation, and entrepreneurship in the rapidly growing Greater Bay Area +by participating in the Creativity, Innovation & Entrepreneurship in China Programme. + +Chaabane, Adam Andrea +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2025 + +I have learned that prominent businessmen and professionals in Hong +Kong are people like me who were once attending business school. I +most enjoyed the guest +lectures because they allowed direct +interactions with inspiring figures. The virtual experience enabled me to +follow this programme easily from my home country. I definitely +recommend it for a better understanding of Hong Kong and GBA’s +markets and opportunities that arise in cross-border business. It would +be great to learn more about opportunities for foreigners in the region. + +Kaur, Suchleen +Business Consulting Practicum 2024 + +The opportunity for real-time interaction and feedback with the client had +propelled me to join BCP. I was always curious about how a consultant +would address and communicate with the client. BCP has taught me +about proper business etiquette and equipped me the necessary skills to +negotiate conflicting opinions during team discussions. Throughout the +process, our mentor and professors provided great help and insights. +Their invaluable feedback and questions helped nudge us in the right +direction, taking a step back to look at the big picture and most +importantly, prioritise the client’s needs. + +Zhou, Qingzhuo +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2025 + +My greatest takeaway from the CIEC 2025 programme was the unique +window to gain a comprehensive understanding of the Greater Bay +Area’s dynamic business landscape. I explored the latest trends in +business, finance, and innovation from guest talks and virtual lectures +with industry leaders. I particularly appreciate the coverage on a +diverse range of topics and the opportunity to earn HKU credits. A +highlight was the cultural exploration segment which offered an +unparalleled perspective on the Greater Bay Area. I recommend this +programme without hesitation. + +Tsang, Tze Yeung Matthew +Business Consulting Practicum 2024 + +BCP has been a rewarding experience that allowed me to apply my +academic knowledge to real-world business challenges. Working +directly with my mentor and client has pushed me to think critically and +creatively in ways I hadn't experienced in the classroom. I'm confident +that the problem-solving, communication, and client-facing skills I have +developed will serve me well when pursuing my future career in +consulting or another business-oriented field. + +2322 Cheung, Ho Ning Jolie +UW Global Business Case Competition 2025 (Champion) + +in UW GBCC was an + +Competing +incredibly rewarding and +unforgettable experience. We had the opportunity to work with +participants from different universities. +It was a fantastic +opportunity to exchange perspectives and learn from different +problem-solving approaches. Working under tight timeframes +pushed us to think creatively. UW GBCC has not only elevated my +professional capabilities but further solidified my belief in business +as a powerful vehicle for creating positive change in the world. + +Case Competitions + +The School provides ample opportunities to refine students' presentation and analytical skills +through a myriad of local as well as overseas competitions. From training to delicate analysis +of business cases to a variety of business scenarios, students are equipped with different +skillsets to compete with candidates from the world's leading universities. + +Lo, Nikkia +NHH International Case Competition 2024 (Champion) + +My week at NHH International Case Competition has been an +unforgettable rollercoaster of emotions. From the excitement of +exploring a new Nordic city, the exhaustion of our 24-hour case +competition, to the nervous anticipation of pitching on a global stage +with a +livestream, every moment was extraordinary. Meeting +like-minded peers around the world was an incredible experience that +forged new friendships. Thank you, HKU Business School, along with +our professors, coaches and staff for bringing this opportunity to us! --_ -_.----e- --, (cid:9) - (cid:9) --- (cid:9) +Xia, Yunchu Kevin +The McDonough Business Strategy Challenge 2025 (First Place) -l'---0` (cid:9) +Winning first place at MBSC2025 was incredible and deeply rewarding. +My team was highly engaged. We were constantly bouncing ideas off +each other and refining our strategy. There were moments where we hit +roadblocks, but our team rallied, challenged each other constructively +and developed a cohesive solution. Connecting with like-minded +participants, judges and NGOs was also invaluable. MBSC not only +sharpened my analytical and presentation skills but also reaffirmed my +commitment to leveraging business for good. I am so grateful for the +experience and very eager to apply these lessons moving forward! -_ (cid:9) -_ +2524 Liu, Yunqi Iris +Deloitte Tax Championship 2024 (Outstanding Team Award) -_ +This competition has been a fantastic learning experience. It +showed me that tax isn’t just about the technical details - it’s also +deeply connected to social trends and global events like climate +change, geopolitical conflicts and national policies. This broader +perspective has enhanced my understanding of the industry and +its importance in today’s world. I want to express my sincere +gratitude to my advisors for their guidance and recognition of my +team’s hard work. I am grateful for the opportunity to learn and +grow with such an amazing group of people. -_ +Pan, Ho Man Vivian +HKGCC Business Case Competition 2024/25 (Champion) -- (cid:9) +in -- (cid:9) +the HKGCC Case Competition was a -- +Participating +truly +transformative experience that greatly enhanced my strategic +thinking, teamwork and problem-solving skills. The competition +challenged me to analyse intricate business problems, think critically +under pressure and develop innovative solutions. Working closely +with people of various majors allowed me to appreciate differing +opinions that were at times conflicting, find ways to build consensus +and synthesize ideas effectively. -- --7---- - --__-- (cid:9) -------- -- (cid:9) +Students' achievements in case competitions +(2023-2025 awards) -- (cid:9) - (cid:9) +Overseas / Chinese Mainland -- -- +Asia Pacific Sustainable Marketing Case Competition 2025 -- (cid:9) +Third Place -- -- - -.-• =,---• -.-__----------- +Bloomberg Global Trading Challenge 2023 -- (cid:9) - (cid:9) +CBS Case Competition 2023 -- +CFA Institute Research Challenge 2025 - Asia Pacific Regional Final -, +CFA Institute Research Challenge 2025 - Sub-Regional Final -t t -, (cid:9) -- (cid:9) -- (cid:9) -= (cid:9) +CGMA Global Business Challenge 2023 - North Asia Final -_..----, (cid:9) +Champions' Trophy Case Competition 2024 -. (cid:9) +Grand Prize Winner +Regional Prize Winner: Asia & Oceania -_ +Winner -- (cid:9) +Winner -Photograph 3: William Jones Arney house, view facing northeast +Winner -Photograph 4: Granary, view facing north +Champion -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 (cid:9) -:Vancy Van Dolsen +Finalist -Page 13 +China Mainland, Hong Kong and Macao Digital Economy Innovation and +Entrepreneurship Competition 2024 -(cid:9) - " -. -• +2nd Runner-up (Gold Medal) -: -• -• +Deloitte Digital Camp 2024 -0 +Deloitte Digital Camp 2023 -. +Second Place -0 +Champion -c -e -) +Deloitte Tax Championship 2024 - National Round -z +Outstanding Team Award -C +Deloitte Tax Championship 2024 - Regional Round -. -) +Deloitte Tax Championship 2023 - National Round -E +Deloitte Tax Championship 2023 - Regional Round -- -: - -6: -1 - -;1 - -8 - -4 -. -1 - - Marshall Arney House +Top 3 teams -The Marshall Arney House is located on the west side of SR 1424. +Merit Award -Location: (cid:9) -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. +Top 3 teams -The Marshall Arney House stands in a clearing of a wooded area +Family Office Association Hong Kong (FOAHK) 2023 Case Competition -Setting: (cid:9) -halfway up a hillside on a knoll approximately 100 feet wide in a bend of the -Johns River. Johns River Road is located on the next downhill terrace. East of -the roadway the land slopes down to the river. +2nd Runner-up -Description: +HSBC Global Private Banking Case Challenge 2023 -Dwelling: Built ca. 1900, the Marshall Arney House is a two-and-a-half- -story, T-shaped house and features a plan similar to his father's house, the -William Jones Arney House. The primary elevation faces south. The -Marshall Arney House has a wrap-around porch with chamfered posts. -Similar to the William Jones Arney House, the Marshall Arney House has -pointed arched louvred windows at the peaks of the end gables. The end -gables are pedimented. Two early twentieth-century additions were -constructed onto the north gable end; a two-story gable wing, and a one- -story shed-roof ell. The building is weatherboarded, and retains most of -its original two-over-two sash windows. +Investing in the Future University Competition 2023 +by Pictet Asset Management -Springhouse: To the west of the house stands a gable-front springhouse -banked into the side of the hill. The springhouse has a stone lower level -with a stuccoed finish. The interior of the upper floor is sheathed in -diagonal boards, and the exterior is weatherboarded. The gable roof is -cantilevered over the façade of the springhouse. +John Molson Undergraduate Case Competition 2025 -Background Information: (cid:9) -Constructed by Marshall Arney on land bequeathed -to him by his father, William Jones Amey, this house is a substantial dwelling for -the area. Marshall Arney was a land agent for Duke Power, served for twenty- -five years on the Burke County School Board and was instrumental in -consolidating and erecting new schools in the county during the early twentieth -century. In April 1922 Marshall Arney and his wife, L. Inez Arney, sold 250 -acres, including the subject structure, to Western Carolina Power. +Lazaridis International Case Conference 2023 -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, the Marshall Arney House is -considered not eligible for the National Register of Historic Places under -Criterion A. since it is not representative of a significant trend in the history of the -area. +Marshall International Case Competition 2024 -historic Architectural Resources Survey Report: NCDOT. SR I424. Burke County; November 1999 (cid:9) -Nancy i'an Dolsen +McDonough Business Strategy Challenge 2025 -Page 15 +NHH International Case Competition 2024 - The Marshall Amey House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. +Scotiabank International Case Competition 2023 -The Marshall Arney House is considered eligible for the National Register under -Criterion C since the building embodies the distinctive characteristics of a -vernacular form of farmhouse for the region and exhibits excellent integrity. The -T-shaped frame rural dwelling dating to the late nineteenth and early twentieth -century is found in the region as a house type built by farmers of middle -economic status and part-time farmers with jobs off of the farm. This type of -farmhouse, if ever once prevalent, has become rare due to the large tracts of -former agricultural land that has been acquired for forest and timber land. The -Amey House is a good representation of this type of farm house and exhibits -integrity of materials, workmanship, setting, location, feeling, and association. -The springhouse building is also representative of a vernacular building form and -exhibits good architectural integrity. +Sydney International Business Competition 2025 -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. +Thammasat Undergraduate Business Challenge 2024 -Boundaries: The proposed National Register boundaries comprise -approximately 1.9 acres that surround the house and its accompanying -outbuilding. The land was deeded to Marshall Arne y at death of his father. This -tract is part of parcel JR-18, which is a portion of the larger Carolina Centers -tract. +The Third KPMG ESG Case Study Competition -Boundary Justification: (cid:9) -include the house. outbuilding, and the surrounding house lot subdivided from its -tax parcel lot of 23,136,10 acre tract. +UW Global Business Case Competition 2025 -The proposed historic boundaries were drawn to +UW Global Business Case Competition 2023 -Historic Architectural Resources Survey Report: ACDOT SR 1424. Burke County: November 1999 (cid:9) -Nancy Ilzn Dolsen +1st Runner-up -Page 16 +2nd Runner-up - Figure 5: Tax Map, Marshall Arney House +Finalist -Historic Architectural Resources Survey Report; NCDOT, SR 1414, Burke Counr.t...,- November 1999 (cid:9) -Nancy ran Dolsen +1st Runner-up -Page 17 +Finalist - L -T +First Place -. - -m -u +Champion -l - -t - -4 -C +Winner -C -a -) +Third Place -' -• +1st Runner-up -C +Third Prize -• +Champion -• -• -• -• -• +Finalists -• -• -• +27 -c -. +26 Local -.VC1)Or SR 1414, Bur ke Con +Local - :Vovember 19 99 +100WF Hong Kong JumpStart University Investment Hackathon 2023 - Corpening House +1st Runner-up -The Corpening House stands on the west side of SR 1424. about +HKSI Institute Case Competition 2024 -Location: (cid:9) -1.5 miles south of the junction with SR 1405. The house sits on a 23,136.10-acre -tract owned by Carolina Centers LLC (formerly Crescent Resources, Inc.), a -timber company. The property was formerly owned by Duke Power, Inc. +Champion -The Corpening House stands on a slight rise overlooking the wide +ACCA Hong Kong Business Competition 2023 -Setting: (cid:9) -bottomland along a bend in the Johns River. The land around the house is now -used to grow trees and shrubs for sale to nurseries and landscaping companies. +Outstanding Team for start-up Filix +Medtech -Description: +Team Excellence Award -The ca. 1900 features a common form for the period: a two- +Merit Award for start-up MakeOmnia -Dwelling: (cid:9) -story, square, boxy building with a pyramidal roof. The facade features a -large central pediment over the second-floor windows, and a smaller -pediment covers the center of the one-story porch across the facade. -Turned posts support the porch roof. The exterior of the house is clad in -narrow weatherboards. The window sashes have been replaced with one- -over-one double-hung sash. Since 1983, the porch has lost its decorative -brackets, and the back porch has been removed andior remodeled. A shed -roof addition covers the rear elevation. No outbuildings stand on the -property. +Alvarez & Marsal Hong Kong Business Case Competition 2024 -Background Information: (cid:9) -the holdings of John Perkins. one of the county's first settlers of European -descent. The property is now part of the large landholdings of the Carolina -Centers corporation. +Champion & Best Presenter -The Corpening House stands on land once part of +HKSI Institute Case Competition 2023 -For purposes of compliance with Section 106 of the National +Best Performer of Roma Risk Advisory +Limited -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, the Corpening House is considered not eligible for the National -Register ofl-listoric Places under Criterion A, since it is not representative of a -significant trend in the history of the area. +1st Runner-up -The Corpening House is considered not eligible for the National Register of -Historic Places under Criterion B because it is not associated with any individuals -of specific historical significance. +Merit -The Corpening House is considered not eligible for the National Register under -Criterion C since the building has lost integrity and does not embody distinctive -characteristics of a significant building type. The building no longer retains its -original porch trim, window sash, and the form has been compromised by the -alterations to the rear of the dwelling. +Champion -Historic Architectural Resources Survey Report: NCDOT1SR 1424, Burke County: November 1999 (cid:9) -Nancy lint Dolsen +1st Runner-up -Page 21 +Merit & Finalist - The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. - +Semi-finalist -Historic Architectural Resources Survey Report; NCD071SR 1424, Burke County: November 1999 -Nancyl'an Dohen +Champion -Page 22 +2nd Runner-up - Photograph 1: Corpcning House, facing east +Best Presenter -Photograph 2: Corpening House, facing north +1st Runner-up -Historic Architectural Resources Survey Report: NCD07; SR 1424, Burke County: November 1999 (cid:9) -Nancy ("an Dohen +HKU-Airstar Bank Joint University Case Competition 2024 -Page 23 +HKU-SaSa Joint University Case Competition 2023 - Early Twentieth-Century Frame House +Hong Kong MBS Digital Economy Business Simulation Competition 2025 -This frame house stands on the east side of Corpening Bridge +Third Place -Location: (cid:9) -Road. SR 1438. approximately 1.5 miles east of the bridge over the Johns River. -The house sits on a 23,136.10-acre tract owned by Carolina Centers LLC -(formerly Crescent Resources, Inc.), a timber company. The property was -formerly owned by Duke Power, Inc. +HSBC Global Private Banking Case Challenge 2024 -Setting: (cid:9) -outbuildings stand south of the house, down a slope, and are no longer used. +HSBC Life Insurance Competition 2025 -This frame house sits in a wooded area that was once farmed. The +HSBC Life Insurance Innovation Competition 2023 -Description: +1st Runner-up -A two-and-a-half story, three-bay, double-pile dwelling, +Top 3 Winner -this house features a hip roof with a central gable dormer. The house has -an exposed extended rafter roof and is covered with weatherboard. A one- -story hip-roof porch covers the facade of the building. The house retains -its six-over-six sash windows. A one-story gable roof ell is located on the -rear of the house. +1st Runner-up -Outbuildings: To the east of the house stands a frame domestic -outbuilding. The gable-front building stands on a cement block -foundation. and is covered with weatherboard. An off-center door is -located on the gable end. To the south of the house are three frame -outbuildings in poor condition. The outbuildings have metal roofs and -narrow board siding. The buildings appear to have served as a granary, -stable, and a workshop. +Best Use of Technology Award -For purposes of compliance with Section 106 of the National +International MBS Business Management Simulation Competition 2023 -Evaluation: (cid:9) -Historic Preservation Act of 1966 National Historic Preservation Act of 1966, as -amended, this house is considered not eligible for the National Register of -Historic Places under Criterion A, since it is not representative of a significant -trend in the history of the area. +Champion -This house is considered not eligible for the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. +ISM-HK Student Case Competition 2025 -This house is considered not eligible for the National Register under Criterion C -since the building does not embody distinctive characteristics of a significant -building type. The outbuildings have lost their architectural integrity and their -setting. +ISM-HK Student Case Competition 2024 -The building is not likely to yield information important to the history of building -technology; therefore it is considered not eligible for the National Register under -Criterion D. +MBS Simulated Business Management Competition 2024 -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy l'an Dolsen +New Century Elite Development Programme 2023 -Page 24 +Venture Capital On Campus (VCOC) - Investment Challenge 2023 - Photograph I: Early Twentieth-Century Frame House, facing northeast +1st Runner-up -Photograph 2: Early Twentieth-Century Frame House, facing north +Champion -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County; November 1999 (cid:9) -Nancy Van Do/sell +Third Place -Page 25 +First place - Photograph 3: Frame Domestic Outbuilding, facing northeast +Best presenter -Photograph 4: Frame Granary, facing south +Third Place -Historic Architectural Resources Survey Report: NCDOT, SR 1424, Burke County: November 1999 -Nancy I'an Dolsen +Champion -Page 26 +1st Runner-up - 0 +2nd Runner-up -t -t -j +Winner Team -0 -. +29 -k -. -r -; +AmplifyME APAC M&A APAC Challenge 2025 -C +Third Place -I -) +BOCHK Challenge 2023 -- -0 +Bronze Award for Innovative Design -. -2 -4 +CGMA Global Business Challenge 2025 (Regional Final) -- -a +1st Runner-up -r -/ -1 +CGMA Global Business Challenge 2023 (Semi-final Round) -0 +Deloitte USHINE Case Competition 2024 -C +ESG Challenge 2023 + +EY Young Tax Professional of the Year 2023 - Hong Kong Region + +Champion + +Finalist + +Champion + +Champion + +1st Runner-up + +2nd Runner-up + +Family Office Association Hong Kong (FOAHK) 2025 Case Competition + +Champion + +HKCGI Corporate Governance Paper Competition and +Presentation Awards 2024 + +HKGCC Business Case Competition 2023/24 + +HKGCC Business Case Competition 2024/25 + +28 + +HKGCC Pitch Perfect Programme 2024 + +HKICPA Business Case Competition 2023 + +HKSI Institute Case Competition 2025 + +1st Runner-up + +1st Runner-up (Paper Writing Competition) + +1st Runner-up (Paper Presentation +Competition) + +Audience’s Favourite Team Award + +Chevalier - Champion + +Chinachem Group - Champion + +Swire Pacific - Champion -. -. +Chevalier - 1st Runner-up -, +Chinachem - 1st Runner-up -0 +HKGCC - Champion -1 -1 -) +Jardine Matheson - Champion -November 1999 +SOCAM - Champion - Nineteenth-Century Log House +Jardine Matheson - 1st Runner-up -Location: (cid:9) -0.25 miles south of the intersection of SR 1424 and SR 1405. +MTR - 2nd Runner-up -This log house stands on the west side of SR 1424, approximately +Overall Champion -The log house sits approximately 40 feet from the road, and has +Winner -Setting: (cid:9) -two mid-twentieth century outbuildings standing to the rear of the house. To the -north of the log house stands a late twentieth-century residence. +2nd Runner-up -Description: +Champion -The one-story, side-gable, three-bay log dwelling is +1st Runner-up -Dwelling: (cid:9) -constructed with half-dovetail corner notching. The house features a -variety Of wood sheathing, including narrow weatherboards and novelty -siding. Windows in the building have been enlarged and have six-over-six -sashes. The porch on the facade has also been replaced. A one-story -frame addition was constructed onto the rear of the dwelling. An exterior -brick chimney is located on the south elevation. +2nd Runner-up -Outbuildings: Two frame outbuildings stand south of the house: a frame -chicken house with a shed roof and a small, one-story frame building with -an entry on the gable end. +Merit -Evaluation: (cid:9) -For purposes of compliance with Section 106 of the National -Historic Preservation Act of 1966, as amended, this house is considered not -eligible for the National Register of Historic Places under Criterion A, since it is -not representative of a significant. trend in the history of the area. + Chhawchharia, Mihika Amit +HSBC - Summer Analyst program -This house is considered not eligible fir the National Register of Historic Places -under Criterion B because it is not associated with any individuals of specific -historical significance. +internship, -This house is considered not eligible for the National Register under Criterion C -since the building has lost integrity and does not embody distinctive -characteristics. The building no longer retains its original porch or windows and -the form has been compromised by the addition to the rear of the dwelling: +I collaborated closely with various +Throughout my +departments of the bank and +interacted regularly with senior +management, gaining comprehensive exposure to the bank’s operations. +This experience allowed me to hone my professional skills, including +communication, leadership, financial analysis, and market expertise. I +was honored to be ranked as the top intern in my cohort and secured a +return offer to join HSBC as a full-time graduate analyst in their Global +Banking division. I am deeply grateful for the knowledge and support I +received at HKU, which enabled me to thrive in a prestigious global +institution alongside some of the brightest talents in the industry. -The building is considered not likely to yield information important to the history -of building technology; therefore it is not eligible for the National Register under -Criterion D. +Internship Experience -Historic Architectural Resources Survey Report: NCDOT, SR 1424. Burke County: November 1999 -Nancy l'art Dolsen +Lee, Chung Hong Calvin +HSBC - Global Risk Management Intern -Page 28 +Reflecting on my time at HSBC, I realise that I have grown on a +professional and personal level. I have developed a deeper passion for +the banking industry and a greater appreciation for the critical role that +risk management and compliance play. I wore many hats and gained +precious insights into the banking industry. I was responsible for +overseeing compliance risks across 32 markets in Asia Pacific. I am +confident that the lessons I have learned at HSBC will guide me every +step of the way. - Photograph 1: Nineteenth-Century Log House, facing northeast +Hui, Matthew +HSBC - International Banking (ISB) Intern - Corporate Coverage -Photograph 2: Nineteenth-Century Log 1-louse facing southwest +The comprehensive exposure to Corporate Banking was invaluable. From +analysing credit risk for lending applications to understanding diverse +product offerings for MNC clients, each day brought new learning +opportunities. I'm truly grateful for the experience of shadowing and +supporting different RMs in the team. Together with the sharing and +introductions from product partners, I learned how to identify clients' +needs and gained perspective on effectively reaching out and pitching to +clients. This experience taught me how to view banking relationships +through a strategic lens - understanding not just immediate needs but +trends. The +also anticipating +professional mindset, strategic thinking, and collaborative spirit fostered +during this internship will undoubtedly shape my future career path. -Historic Architectural Resources Survey Report: .VC DOT SR 1424, Burke County: November 1999 -Nancy I an Dolsen +future requirements and market -Page 29 +Nagar, Tanvi +HSBC - International Banking Intern - Photograph 3: View of Chicken Coop, facing northeast +immense + +My internship experience was nothing short of incredible. Not only did I +grow immensely in my technical and professional skills but also +underwent an +journey of personal growth. I received +unparalleled support from the team and my mentors and gained an +in-depth insight into the field of banking and finance in Hong Kong. I +formed meaningful connections and grew my network. This experience +has truly shaped me and continues to propel my drive to pursue a career +in Banking in the years to come. + +3130 Sharma, Seerat +KPMG - Financial Services Audit Intern +(COOP3000 Cooperative Education Programme) + +I had the privilege of contributing to audit engagements for leading +clients in the banking and insurance sectors, where I conducted +substantive testing, supported risk assessments and navigated audit +methodologies using advanced tools. This role sharpened my analytical +mindset, strengthened my collaboration skills and offered a front-row +view into the dynamic world of financial auditing. It’s been a truly +rewarding journey that has solidified my passion for a career in +professional services. + +Xu, Aijing +CMB International Capital Corporation - Wealth Management Intern + +During my internship, I actively engaged with the product team in +market analysis and client pitching focused on mutual funds. I gained +valuable insights into various asset classes and macroeconomic +trends. Analyzing mutual fund performances and participating in +client meetings allowed me to contribute to investment portfolio +strategies. Collaborating with +team enhanced my +communication skills and understanding of client needs. Overall, this +experience deepened my knowledge of wealth management and +strengthened my passion for finance. + +the sales + +Shi, Yu Cheung Karson +Bank of China - Credit Summer Intern (Risk Management) + +During the internship, the financial techniques and business knowledge +I learned from HKU Business School courses helped a lot in company +financial analysis. By applying knowledge taught in the financial +accounting module, I assisted in evaluating loan demand. Through +cross-departmental collaborations, I also realised how classroom +simulations mirrored real-world risk mitigation challenges. This +experience reinforced the value of integrating academic theories with +dynamic market practices. + +Hu, Wen Qing +PwC Assurance Practice Intern + +As an assurance intern in PwC, I assisted my seniors with sampling +financial statement line items, reconciling balances, and documented +relevant non-financial information. It showed me how demanding the +auditing industry was, and allowed me to set my expectations for a +career in auditing as an accounting student. + +Wirawen, Karen +HKU – Neuberger Berman Private Equity Internship Programme + +You never truly understand private equity until you’re immersed in it +and my internship at Neuberger Berman was deeply enriching. Coming +in with limited knowledge, I gained direct exposure by working on +various deals across sectors and geographies, including Japan, Korea +and Southeast Asia. From drafting memos for the +investment +committee to attending annual general meetings and taking notes on +calls, I progressed from simple tasks to meaningful responsibilities. The +seniors were welcoming, always open to teaching, providing feedback +and offering valuable career advice. + +Chung, Marcus Lok Sum +Morgan Stanley Industrial Placement Intern + +During my internship with Sales & Trading (Institutional Equity +Division), I had the amazing opportunity to rotate across trading +and structuring desks where I was exposed to the fast-paced +environment especially during periods of high market volatility. To +see how traders manage positions and mitigate risks first-hand +was incredibly insightful. I also contributed to constructing trading +strategies and seeing the tangible impact of our work was deeply +rewarding. Morgan Stanley definitely provided me with a broader +perspective and understanding of market dynamics, liquidity, +client interactions and cross-asset opportunities. + +3332 Our Student Scholars + +Chan, Kwan Yin Cheryl +Recipient of Louisa Cheang Inspiration Scholarship 2024-25 + +I would like to express my gratitude for the Louisa Cheang Inspiration +Scholarship, which has significantly contributed to my personal and +professional growth. In the past academic year, this scholarship enabled +in the full-time Registered Visiting Student +me to participate +Programme at St Catherine’s College, University of Oxford. This +immersive experience enhanced my legal studies through engaging +discussions with scholars and participation in mooting competitions, +sharpening my research and advocacy skills. Additionally, I plan to +utilise this scholarship to pursue overseas postgraduate studies, +allowing me to deepen my understanding of the complexities of law in +a global context. + +Hussain, Iman +Recipient of FBE Student Contribution and +Innovation Award 2024-25 / +HKU Business School Future Leader Scholarship 2024-25 + +As someone who believes in learning through real-world experience, +the FBE Student Contribution and Innovation Award and the HKU +Business School Future Leader Scholarship have been life-changing. +They not only enabled me to go on an exchange semester to Cardiff, +where I explored new cultures and perspectives but also shaped my +entrepreneurial journey. + +With this support, I’ve been able to work on building a social +enterprise that aligns with my passion for creating meaningful impact. +I’m incredibly thankful for these opportunities to grow and make a +difference. + +Chow, Yee Shun Eason +Recipient of the Jardine HKU Scholarship 2024-25 + +It is an incredible honour to have received the Jardine HKU Scholarship +which is enriching my university journey! This scholarship allowed me to +pursue my many goals – from attending international training camps as +a handball player to participating in overseas exchanges without +financial worries. I have had the privilege of undertaking a summer +internship at the Jardine Group, learning from seasoned professionals in +my future industry. The Jardine Foundation also hosted exciting +networking activities and site visits which helped me build lasting +connections with fellow scholars in a vibrant community. + +Shih, Pui Hei Iris +Recipient of the Hong Kong Jockey Club Scholarships - +Undergraduate Scholarship 2024-25 + +This scholarship has been transformative to my university life as it has +allowed me to fully embrace and explore the world beyond the +classroom. It provided the financial stability I needed to focus on my +studies and presented opportunities for me to participate in diverse +extracurricular activities. These experiences broadened my perspective +and opened me to different points of view. -Photograph 4: Domestic outbuilding, facing north +Nurgali, Adilzhan +Recipient of BOC Hong Kong - +One Belt One Road Scholarship 2024-25 -Historic Architectural Resources Survey Report; NCDOT, SR 1424, Burke County: November 1999 (cid:9) -Nwtcy i in Dolsen +It means a lot to me to be awarded the BOC Hong Kong - One Belt One +Road Scholarship. It was not +just about financial support but +recognition of the hard work I have put in. As an international student +soon heading for exchange, this award gives me the freedom to explore +new cultures and experiences without extra stress. I am truly grateful to +be part of such an inclusive and supportive community. -Page 30 +3534 2026 JUPAS Admissions - BIBLIOGRAPHY +Admissions criteria (HKDSE) -Arney, Marshall House. Historic Resource Survey Form, Burke County, 1983. -Asheville, NC: Western Office. Division of Cultural Resources. +Local candidates applying on the strength of the Hong Kong Diploma of +Secondary Education (HKDSE) are required to fulfil the minimum entrance +requirements for HKU Business School programmes as shown below: -Arney, William Jones House, Historic Resource Survey Form, Burke County, 1983. +Level 4 +Level 4 -Asheville, NC: Western Office, Division of Cultural Resources. +EngEng ≥ +Level 4 or above in English Language +ChinChin ≥ +Level 3 or above in Chinese Language1 +MathMath ≥ +Level 3 or above in Mathematics -Biographical Directory of the United States Congress, 1774-Present, Thomas Burke,- +Level 3 +Level 3 -Website: Bioguide.conress.gov. +Level 3 +Level 3 -Burke County Deeds, Burke County Court House, Morganton, NC. +AA -Burke County Plat Book, Burke County Court House, Morganton, NC. +CSDCSD = +Grade “Attained (A)” in +Citizenship and Social Development2 -Corpening House, Historic Resource Survey Form, Burke County, 1983. Asheville,NC: +2x -Western Office, Division of Cultural Resources. +Elective +Elective +Subjects +Subjects -Cotton, J. Randall, Suzanne Pickens Wylie, and Millie M. Barbee. Historic Burke: An -Architectural Sites Inventory of Burke County. Mortzanton, NC: Historic Burke -Foundation. Inc.. 1987. +≥ -Phifer, Edward W. Burke: the History of a North Carolina County, 1777-1920. +Level 3 +Level 3 -Morganton, NC: by the author, 1977. +Level 3 or above in two Elective Subjects -Historic Architectural Resources Survey Report: NCD01; SR 1424, Burke County: November 1999 (cid:9) -Nancyl'an Dolsen +For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) +In addition to the minimum entrance requirements above, candidates applying for the BBA(BA), +BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the +HKDSE examination: -Page 31 +BBA(BA) / BSc(MAT) +≥ +Level 3 or above in one of the following subjects: - APPENDIX A +Level 3 -NCDOT, SR 1424. Bur ke Cou +Biology -November 1 999 +BAFS - North Carolina Department of Cultural Resources +Chemistry -James B. Hunt Jr., Governor -Betty Ray McCain, Secretary +Economics -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 +ICT -July 9, 1999 +Physics -Stephen A. Moore, PE -District Engineer -NCDOT -Route 1 Box 169C -Marion NC 28752 +BBA(BA) -Division of Archives and History -Jeffrey J. Crow, Director +BSc(MAT) -LOCATION -507 North Blount Street -Raleigh, NC -State Courier 53-31-31 +Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not +strictly required. -Re: (cid:9) Widen and pave SR 1438 Corpening Bridge Road, +BBA(Law)&LLB / BBA(IBGM) / BSc(QFin) -Burke County, ER 99-9229 +BBA(Law)&LLB -Dear Mr. Moore: +BBA(IBGM) -We have received notification of a nationwide permit for the above project and would -like to comment. +BSc(QFin) -We have reviewed the permit application for the improvement and paving of SR 1438 -Corpening Bridge Road and located the following structures of historical or -architectural importance within the general area of the project: +EngEng ≥ -William Avery House, located at the northwest corner of the junction of SR 1438 -and SR 1424. +Level 5 +Level 5 -Arney's Chapel, located at the east side of SR 1423 south of its junction with SR -1439. +Chin, Math +Chin, Math ≥ -We recommend that an architectural historian with the North Carolina Department of -Transportation evaluate these sites to determine if they are within the area of potential -effect for the permit and are eligible for listing on the National Register of Historic -Places. +Level 4 +Level 4 -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. +EngEng ≥ -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. +Level 5 +Level 5 --Sincerely, ++ -1- (cid:9) ++ -• (r-N — +M1/M2 +M1/M2 ≥ -Da.vicl'Brook -Deputy State Historic Preservation Officer +Level 3 +Level 3 -DB:slw +MathMath ≥ -109 East Jones Street • Raleigh, North Carolina 27601-2807 +Level 4 +Level 4 - cc: (cid:9) +Application -Army- Corps of Engineers, Asheville -William D. Gilmore +For details of the 2026 JUPAS exercise, please visit www.jupas.edu.hk. -t-Barbara Church +Programme Scoring Formula - AUG-16—":39 MON 10:10 ID: +For more information on the Programme Scoring Formula and Subject Weightings, please +visit the Admissions Office’s website at https://admissions.hku.hk/apply/jupas. -TEL NO: +Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis.Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered.Level 5 or above inEnglish LanguageLevel 4 or above inChinese Language1 & MathematicsLevel 5 or above inEnglish LanguageLevel 4 or above inMathematicsLevel 3 or above in Extended Module 1 or 2 of Mathematics3736 2026 Non-JUPAS Admissions -351; P02 — +Selection process and entrance requirements -North Carolina Department of Cultural Resources +Eligibility +(1) Local students applying for admission on international qualifications other than the -Jams B. Hunt Jr.. Governor -Betty Ray McCain, Secretary +HKDSE examination; -MAILING ADDRESS -4617 Mail Service Center -Raleigh, NC 27699-4617 +(2) Non-local students with international qualifications other than Chinese Mainland's -August 10, 1999 +National Joint College Entrance Examination (NJCEE/Gao Kao). -Stephen A. Moorc -District Engineer -North Carolina Department of Transportation -Route 1, Box 169C -Marion NC 28752 +Application Procedures +All non-JUPAS applications should be submitted via the online application system by noon of +November 26, 2025 (Hong Kong Time) for the first-round evaluation. Scanned copies of the +following documents should also be uploaded by the application deadline. -Re: +(1) Official prediction of grades or scores from your current high school (if available); -Paving SR 1414, Johns River Road, Burke County, -ER 00--7154 +(2) All internal and public examination results during your senior secondary education; and -Division of Archives Bnd History -Jeffrey 1. Crow, Director +(3) Academic reference letter (optional) -LOCATION -507 North Blount Street -Raleieh, NC -State Courier 53-31-31 +Late submission of supporting documents will delay the consideration of application. -it- , - have received notification of a nationwide permit for the above project and would +Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other +programmes will be arranged if necessary. -i%1'- +University and Programme Entrance Requirements +Due to the complex combination of education background of applicants, each application will +be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the +standards and requirements on the website of the Admissions Office of the University before +making an application: -Dear Mr. Moore: +(1) Admissions standards (International Qualifications or National Qualifications); -We -like to comment. +(2) Programme specific subject requirements; and -We have conducted a search of our maps and files and have located the following -structures of historical or architectural importance within the general area of the -project: +(3) English language requirement -qX.117. +In particular, please note that the expected lower boundary for admissions listed on the +website is provided for reference purposes only. Admissions is competitive in nature and +achieving the expected lower boundary does not guarantee admission to the programme. +The actual admission grade or score will be dependent on the overall quality of applications +received in the 2026 admissions exercise. -Marshall Arney House, located on thc wcst side of SR 1424, 0.5 mile northwest -of its junction with SR 1438. +Announcement of Result +Offer results will be released from December 2025 onwards. All application status will be +updated in the online application system once an admissions decision has been made. +Applicants should check their application status online regularly. -William Arney House, located at the northwest corner of the SR 1424 and SR -1438 junction. +More Details +For further information about the application procedures and admissions +standards, +at +https://admissions.hku.hk/apply/international-qualifications. -9;0 ck +the Admissions Office’s website -Corpening House, located on the west side of SR 1424, L5 mile south of its -junction with SR 1405. +please -We recommend that an architectural historian with the North Carolina Department of -Transportation identify and evaluate these properties and any other structures over -fifty years of age within the project area, and report the findings to us. +visit -An important archaeological site, a mound reported by J. W. Spainour and Cyrus -Thomas in the late 1800s, is recorded between the Johns River and Johns River Road -south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 311310 highlighted in -yellow. +2026 Dual Degree Admissions -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan +Separate application is required for applying to the dual degree programmes, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. -109 Eist Jones Street • Ra1e10, North Cirolina 27601-2807 +For details of the dual degrees and the application procedures, please visit +the Admissions Office’s website at +https://admissions.hku.hk/programmes/university-collaborative-programmes. - AUG-16-'99 MON 10:11 ID: (cid:9) +3938 Develop Your Career -TEL NO: (cid:9) +Freshmen Sharing -p355 P03 +Upon graduation, over 82% of our students begin their careers in the commerce and industry sector. Others either +move on to further studies or explore opportunities with the civil service, education institutions and +community/social services. -Stephen A. Moore -August 10, 1999, Page 2 +Career Prospects -sheets 8,9, 12. 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. +Companies/organisations that have offered full-time positions to our graduates include but not limited to the +following: -end south of Perkinsville in the proposed project area. We recommend that a qualified -archaeologist survey the vicinity of the site area to be certain that the road -straightening does not impact site 31BK3. The portion of the Morganton North -quadrangle from your permit application is enclosed with site 31BK3 highlighted in -yellow. +Accenture Company Limited +Bank of China (Hong Kong) Limited +BDO International Limited +Bloomberg L.P. +Citibank (Hong Kong) Limited +Crédit Agricole Corporate and Investment Bank +Credit Suisse Group AG +Deloitte Advisory (Hong Kong) Limited +Deloitte Touche Tohmatsu Limited +Deutsche Bank +Ernst & Young +Goldman Sachs (Asia) LLC +Hang Seng Bank Limited +HKSAR Government +Hong Kong Monetary Authority +JPMorgan Chase & Co. +KPMG +Morgan Stanley +PricewaterhouseCoopers Limited +Standard Chartered Bank (Hong Kong) Limited +The Hongkong and Shanghai Banking Corporation +Limited (HSBC) +UBS Group AG + +Full employment +for the 19th + +consecutive year + +98.7% + +was the average salary of +our 2024 graduates + +$36,770 + +of our 2024 graduates have +received at least 2 job offers +upon graduation + +57% + +Kantbek, Yerassyl +BBA + +I chose HKU Business School because Hong Kong is +one of the world’s most competitive financial centres, +offering a dynamic environment to grow. HKU +combines +real-world +rigorous academics with +experiences, such as Asia-focused case studies and +competitions like HICS, which I plan to join in Year 2. I +am particularly drawn to programs in marketing and +finance, as well as opportunities for entrepreneurship +through the Techno-Entrepreneurship Core and +iDendron incubator. With a diverse student body and +global exchange options, I look forward to learning +from peers, connecting with mentors, and developing +skills to make a meaningful impact in business. + +The main reason I chose HKU, particularly HKU +Business School, lies in its global connections and +strong partnerships with other institutions and +respected organizations, which I believe will be +advantageous for future employment opportunities. +I expect to embrace challenges over the course of my +four years at the university, as these will be vital for +my intellectual and social maturity. The first time I +visited the campus, I was amazed by the diversity of +the students and +facilities, which +demonstrated that the university is truly a place for +everyone. One of the many great opportunities +offered here is the chance to travel for studies, which +I would rank among my highest interests. + +the open + +Pernice, Audrey Yiwen +BEcon&Fin + +4140 The main reason I chose BBA(IBGM) is its flexibility in allowing +students to choose a second major, as well as the numerous +overseas exchange opportunities. I am also fortunate to have +been admitted as a member of the HKU International Case +Society (HICS), where I get to know people from other programs +who are also interested in the realm of business. HKU Business +School provides me with the perfect platform to strike a balance +between my work and life. + +Ho, Lik Hang Gordon +BBA(IBGM) + +Lo, Hoi Tung Reacker +BSc(MAT) + +I chose HKU Business School for its professional background and +diverse opportunities. The multidisciplinary programme allows me to +pursue Marketing Analytics and Technology, aligning with my varied +interests in business and STEM. As a freshman, I am eager to embrace +exciting experiences such as internships and exchanges. I look +forward to gaining insights from fellow students and professionals +through upcoming career events. With the strong support from HKU, +I am committed to developing my entrepreneurial skills and pursuing +my dreams without regrets! + +Drawn by HKU’s global prestige and the excellence of HKU Business +School, I chose to pursue Quantitative Finance at HKU to deepen my +passion for quantitative trading and data-driven decision-making. +The school’s flexible curriculum empowers me to explore +complementary fields such as computer science, statistics, and risk +management. As a target school for top investment banks, HKU +also offers unparalleled career exposure. HKU provides the ideal +environment to grow academically and professionally, and I am +excited to explore the many opportunities ahead. + +Tsang, Chun Shing Patrick +BSc(QFin) + +Fung, Yat Hang Rex +BFin(AMPB) + +My aspiration for BFin(AMPB) was ignited early in high school by its +unparalleled, practitioner-focused design, a curriculum that truly +bridges finance theory with industry practice. This conviction was +solidified through the heartfelt guidance of our Programme +Director, Professor Anna Wong, who mentored me even before I +officially joined HKU, while I was an apprentice at a US bank. + +Now, experiencing BFin(AMPB) firsthand, the transition from the +professional world to the classroom feels seamless. It is this +powerful integration of theory and real-world practice that makes +AMPB incomparable. HKUBS has instantly instilled in me a sense of +ownership, this is not just a school, but a launchpad for impact, and +I am thrilled to begin the next chapter of my growth here. + +I was drawn to BBA(IBGM) because of HKU’s +world-renowned academic reputation and its vibrant, +international community. I believe a truly “global +business” education extends beyond the classroom, +through experiential learning opportunities, daily +interactions with local and non-local peers, and +participation in hall and student society cultures, I will +gain a more dynamic and holistic multidisciplinary +perspective on the business world. Over the coming +four years, I am eager to harness these opportunities +to become a well-rounded leader and effective +problem-solver for my future career on the global +business stage. + +Yan, Sung Hei Christopher +BBA(IBGM) + +4342 Our Global Presence + +Israel - +Tel Aviv + +Beijing + +Shenzhen + +Hong Kong + +Vietnam - +Ho Chi Minh City + +Useful Websites +The University of Hong Kong (HKU) + +Undergraduate Admissions +https://admissions.hku.hk/ -In addition to this site, it is recommended that a qualified archaeologist survey the -other proposed curve straightening areas in the project. These areas are found on plan -sheets 8, 9, 12, 13, 24, 28, 46, and 50. All of the archaeological work should be -conducted prior to construction activities. +HKU – Entrance Requirements -The above comments are made pursuant to Section 106 of the National Historic -Preservation Act and the Advisory Council on Historic Preservation's Regulations for -Compliance with Section 106 codified at 36 CFR Part 800. +For JUPAS Candidates +https://admissions.hku.hk/apply/jupas -Thank you for your cooperation and consideration. If you have questions concerning -the above comment, please contact Renee Gledhill-Earley, environmental review -coordinator, at 919/733-4763. +For non-JUPAS Candidates +https://admissions.hku.hk/apply/international-qualifications -Sincerely, +HKU Business School -&Ad,- +Overview of Undergraduate Programmes +https://ug.hkubs.hku.hk/programme -Deputy State Historic Preservation Officer +Enrichment Programmes +https://ug.hkubs.hku.hk/student-enrichment -/ David Brook +Other HKU Websites -DB:slw +Mainland Affairs Office (MAO) +https://mainlandaffairs.hku.hk/ -Enclosure +Centre of Development and Resources for +Students (CEDARS) +https://www.cedars.hku.hk/ -cc: (cid:9) +44 -Army Corps of Engineers, Asheville -William D. Gilmore -Thomas Padgett +45 diff --git a/q104/random_k4/random_2.md b/q104/random_k4/random_2.md index 9a6992971674500fba1af17645e45a0908def2ef..bdc0cc473d09f25f5a2f7f0b1e46304031484979 100644 --- a/q104/random_k4/random_2.md +++ b/q104/random_k4/random_2.md @@ -1,727 +1,637 @@ - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 +Youth in Middle East +and North Africa - Online ISSN: 2054-0965 (Online) +August 2019 - Print ISSN: 2054-0957 (Print) +Daniella Raz +University of Michigan - Website: https://www.eajournals.org/ +JULY 2018 Arab Barometer – Wave V +Topic Report – Youth - Publication of the European Centre for Research Training and Development -UK +Executive Summary -Cloud Computing: Revolutionizing Digital -Transformation in the Modern Era +Individuals between the ages of 15 and 29 account for roughly 30 percent of the +population of the Middle East and North Africa (MENA).1 The beliefs, attitudes, +and preferences of this large and growing segment of the population has the po- +tential to shape and drive the region’s future. Yet, widespread youth discontent +about their economic prospects poses a significant challenge. Youth unemploy- +ment rates across the region remain high and there is broad frustration with +government efforts to create employment opportunities.2 Though satisfaction +with the state education system varies greatly across countries, youth are less +satisfied than are older generations. -Lavanya Jacintha Victor -The University of Texas at Austin, USA +Youth also question if they have the ability to voice their frustrations to the gov- +ernment. In nearly all countries, less than half say the right to freedom of ex- +pression is guaranteed while half or less say they have the right to demonstrate +peacefully in all countries. This combination creates conditions where youth +across the MENA region, far more than older age cohorts, are considering emi- +grating from their country, primarily to Europe, North America, or GCC countries. +Unsurprisingly, many cite economic reasons as the main impetus for leaving. -doi: https://doi.org/10.37745/ejcsit.2013/vol13n181223 Published May 12, 2025 +However, youth differ from their older counterparts not only in terms of greater +frustration with the existing system. Relative to older generations youth also +tend to be less religious, less interested in politics, and more likely to be engaged +in social media. Generational differences are particularly stark in Morocco where +youth attitudes diverge substantially and dissatisfaction with the overall state of +affairs is quite salient. -Citation: Victor L.J. (2025) Cloud Computing: Revolutionizing Digital Transformation in the Modern Era, European -Journal of Computer Science and Information Technology, 13(18),12-23 +Meanwhile, on gender issues, most youth tend to be supportive of equal rights +for women. Majorities say women have the right to be the head of government +and to receive an education. Yet, the same does not hold true for equal inher- +itance rights with minorities saying women should receive the same as men +in most countries. Additionally, this generation cites low levels of support for +women playing equal roles in society, with most saying that men make better +political leaders and stating that husbands should have the final say in family +matters. + +Youth in MENA appear to be particularly inclined toward China relative to other +global powers. Majorities favor closer economic relations with China in most +countries surveyed, while less than half say the same about relations with the +U.S. in most countries. + +1https://www.undp.org/content/dam/rbas/report/AHDR%20Reports/AHDR%202016/AHDR%20Final% + +202016/AHDR2016En.pdf, pg. 7 + +2https://www.imf.org/external/np/vc/2012/061312.htm?id=186569 + +www.arabbarometer.org + +2 + + Arab Barometer – Wave V +Topic Report – Youth + +These are among the key findings from 11 nationally representative public opin- +ion surveys conducted across the Middle East and North Africa by the Arab +Barometer from September 2018 - April 2019. The surveys included more than +25,000 face-to-face interviews in the respondent’s place of residence and have +a margin of error of ±2 percent in most countries.3 + +Religiosity + +Across the region, youth are less religious than their older counterparts. In fact, in +every country surveyed less than half of youth identify as religious. More specif- +ically, fewer than one-fifth of youth in Algeria (15 percent), Tunisia (16 percent), +and Libya (18 percent), and fewer than one-quarter of youth in Jordan (22 per- +cent), Lebanon (23 percent), and Morocco (24 percent) consider themselves reli- +gious. Higher rates of youth religiosity are found in Palestine (28 percent), Sudan +(32 percent), Yemen (33 percent) Egypt (36 percent), and Iraq (42 percent). + +However, trends over time paint a more complicated picture throughout the +region. While indeed less than half of MENA youth identify as religious, in some +countries these rates represent an increase from the last Arab Barometer wave + +3For details about each survey, please visit: + +https://www.arabbarometer.org/waves/ + +arab-barometer-wave-v/ + +www.arabbarometer.org + +3 + + Arab Barometer – Wave V +Topic Report – Youth + +while in others a decrease. For instance, rates of religiosity have increased in +Egypt (+12), Yemen (+7), Tunisia (+6), Iraq (+6), Morocco (+2), and Libya (+2). In +contrast, youth identify as religious at lower rates in Sudan (-12), Palestine (-9), +Jordan (-7), Algeria (-3) and Lebanon (-2). + +Views on Politics + +In addition to generally low levels of religiosity, youth appear disinterested in +politics, both overall and when compared to older generations. Only three-in- +ten youth in Lebanon (31 percent), Yemen (29 percent), Palestine (29 percent), +Sudan (28 percent), and Egypt (27 percent) say they are interested or very inter- +ested in politics, while no more than a quarter say the same in Jordan (25 per- +cent), Morocco (23 percent), Iraq (21 percent), and Libya (21 percent). Tunisian +and Algerian youth are least politically interested with fewer than two-in-ten +citing interest in politics (17 percent and 15 percent, respectively). + +With the exception of Egypt and Lebanon, youth in the region are less interested +in politics than are older citizens. The difference between age cohorts is widest +in Tunisia and Palestine (10 percentage points). Elsewhere, though differences +are smaller, older citizens nonetheless report greater interest in politics than do +youth. +In Lebanon youth have only marginally greater interest (1 percentage +point), while in Egypt youth are 6 percentage points more likely than older citi- + +www.arabbarometer.org + +4 + + Arab Barometer – Wave V +Topic Report – Youth + +zens to say they are interested or very interested in politics. + +Youth disinterest in the political system could be in part due to their outlook +If young people perceive the most +on the fairness of parliamentary elections. +recent parliamentary elections as unfair, regardless of whether they voted or +were even eligible to vote in them, then their disinterest could perhaps reflect +disillusionment with their country’s political process. With the exception of the +Levant, youth are more likely than older age cohorts to evaluate the last par- +liamentary elections held in their respective country as being neither free nor +fair. In some countries surveyed this difference is stark. For instance, in Morocco +four-in-ten youth characterized the last parliamentary elections as such, com- +pared to fewer than a quarter of older Moroccans (23 percent). In Tunisia, youth +were 12 percentage points more likely than older citizens to characterize the +last parliamentary elections as being neither free nor fair (40 percent versus 28 +percent). + +Internet and Social Media + +Across the region, internet use, defined as going online at least once-in-a-few- +weeks, is substantially greater among youth than older generations. In fact, in +most of the surveyed countries (with the exception of Yemen), more than 90 +percent of youth use the internet, whereas among the older generation internet +usage does not approach such a rate. The highest rates of internet usage are +found in Lebanon (99 percent), followed by Jordan (96 percent), Palestine (95 +percent), Algeria (93 percent), Morocco (93 percent), Tunisia (91 percent), Egypt +(90 percent), Sudan (90 percent), Iraq (89 percent), and Libya (88 percent). The +lowest rate of internet usage among youth is seen in Yemen, where fewer than +eight-in-ten youth say they use the internet at least once every few weeks (77 +percent). The most striking generational differences in internet usage are seen in +Yemen and North Africa (excluding Libya), where internet use among youth ex- +ceeds that of older age cohorts by at least 30 percentage points. In contrast, the +intergenerational gap in internet use is less pronounced in the Levant (Lebanon, +Jordan, Palestine), where it does not exceed 20 percentage points. + +Interestingly, among internet users, social media usage is very high across all +demographics. In fact, youth are not so different from their older counterparts +in this regard. Though Yemeni youth report the lowest rates of internet usage in +the region, those online are also the most likely to engage in social media. Nearly +all Yemeni youth who are online (99 percent) use social media on a typical day. +However, most online youth across the region also engage in social media on a +typical day: Jordan (98 percent), Lebanon (97 percent), Iraq (97 percent), Pales- +tine (96 percent), Libya (96 percent), Egypt (96 percent), Tunisia (96 percent), +Morocco (96 percent), Sudan (94 percent), and Algeria (94 percent). Overall, +youth are consistently more likely than older generations to say they spend two + +www.arabbarometer.org -Abstract: Cloud computing has revolutionized digital transformation by fundamentally changing how -organizations approach their technological infrastructure and business operations. The shift from -traditional on-premises systems to cloud-based solutions has enabled unprecedented levels of scalability, -efficiency, and innovation across industries. From artificial intelligence integration to edge computing -adoption, cloud technologies have democratized access to advanced capabilities while reducing -operational costs. Organizations worldwide are embracing hybrid and multi-cloud strategies, prioritizing -security through zero-trust architectures, and focusing on sustainable computing practices. The evolution -of serverless computing and edge technologies continues to shape the future of enterprise IT, offering new -possibilities for innovation while addressing crucial challenges in data processing, security, and -compliance. -Keywords: cloud computing, digital transformation, edge computing, serverless architecture, hybrid -cloud - -INTRODUCTION +5 -In today's rapidly evolving technological landscape, digital transformation has become a cornerstone of -business evolution and success. According to recent International Data Corporation (IDC) forecasts, -worldwide spending on digital transformation is projected to maintain a robust compound annual growth -rate (CAGR) of 16.3% from 2022 to 2026, culminating in a staggering $3.4 trillion by 2026. The digital -and information technology sector has demonstrated remarkable resilience even during economic -uncertainties, showcasing a significant 17.8% growth in digital transformation investment during 2022 -alone [1]. This comprehensive shift towards digitalization represents more than just a trend—it's a -fundamental reimagining of how organizations operate, deliver value, and interact with their stakeholders -through the integration of digital technologies. + Arab Barometer – Wave V +Topic Report – Youth + +hours or more on social media platforms, although the size of this difference is +often marginal at best. As an example, the widest generational gap is seen in +Morocco, where youth are 10 percentage points more likely to use social me- +dia. However, the difference is much smaller in Yemen (1 percentage point) and +Jordan (2 percentage points), and non-existent in Palestine. + +Though there is little generational difference in overall quantity of social media +usage, there exists a much larger generational discrepancy in terms of moti- +vations for using and trust in social media. Youth are more likely than older +generations to use social media as their primary source of information to fol- +low breaking news, and also to trust this information more than that found in +newspapers or TV news programs. With respect to the former, youth are most +likely to say social media is their primary source for breaking news in Palestine +(76 percent), while around two-thirds say the same in Morocco (67 percent) and +Jordan (63 percent). About half of youth or more in Egypt (58 percent), Tunisia +(55 percent), Iraq (51 percent), and Algeria (49 percent) rely primarily on so- +cial media for breaking news, while less than half do so in Libya (45 percent), +Lebanon (43 percent), Yemen (37 percent) and Sudan (32 percent). Notably, +youth reliance on social media to follow breaking news far exceeds that of their +older counterparts in every country surveyed. The greatest discrepancy between +youth and older generations is seen in Morocco (53 percentage points), Egypt +(44 percentage points), Tunisia (36 percentage points), and Palestine (32 per- +centage points). Elsewhere, though less stark, rates of youth reliance on social + +www.arabbarometer.org + +6 + + Arab Barometer – Wave V +Topic Report – Youth + +media for breaking news still exceed that of older citizens by at least 15 per- +centage points or more. Such trends show that while youth and their older +counterparts report using social media at similar rates overall, their incentives +and motivations for using such platforms differ in important ways. + +Youth in the region vary dramatically in the level of trust in information they re- +ceive on social media. The highest rates of trust among youth are found in Mo- +rocco (77 percent) and Egypt (74 percent), followed by Lebanon (67 percent), +Palestine (56 percent), Iraq (53 percent), and Sudan (51 percent). In contrast, +less than half of youth in Algeria (49 percent), Yemen (47 percent), Tunisia (47 +percent), Libya (41 percent), and Jordan (36 percent) trust social media sources +more than newspapers and TV news. The magnitude of the differences across +age cohorts is also varies widely, ranging from single digit differences in Jordan +(2 percentage points), Libya (5 percentage points), Yemen (3 percentage points), +and Sudan (3 percentage points) to much larger generational gaps in Lebanon +(16 percentage points), Egypt (18 percentage points), and Morocco (32 percent- +age points). Overall, however, youth are more trusting of the information they +receive from social media than are older citizens in every country surveyed. + +Notably, the most used social media platform among youth across the region is +Facebook. The exception to this finding is Yemen, where youth use WhatsApp at +a higher rate than Facebook (46 percent versus 36 percent). Youth in Egypt (94 + +www.arabbarometer.org + +7 + + Arab Barometer – Wave V +Topic Report – Youth + +percent), Algeria (91 percent), Morocco (90 percent), and Lebanon (90 percent) +are most partial to Facebook. More than eight-in-ten youth in Sudan (85 per- +cent), Libya (82 percent), and Jordan (81 percent), and more than seven-in-ten +youth in Iraq (74 percent) and Palestine (73 percent) use Facebook. + +Education and Employment + +Education System + +Rates of youth satisfaction with state education systems, both overall and as they +compare to older generations, vary greatly throughout the region. Nearly two- +thirds of youth in Palestine (65 percent) and Jordan (65 percent), and over half +of youth in Yemen (54 percent) are satisfied or completely satisfied. However, +fewer than half are satisfied in Sudan (43 percent) and Lebanon (39 percent), +and roughly a third are satisfied in Algeria (34 percent), Libya (31 percent), Egypt +(30 percent), and Tunisia (30 percent). The lowest rates of satisfaction are seen +among Iraqi youth (24 percent) and Moroccan youth (17 percent). Given this +wide variation, there seems to be limited regional consensus about the state +of education. However, one potential regional trend is that across most of the +countries surveyed youth are less satisfied with the education system than are +their older counterparts. +Indeed, in the majority of countries this discrepancy +is small, as seen in Palestine (1 percentage point), Sudan (2 percentage points), +Libya (2 percentage points), Algeria (3 percentage points), Yemen (3 percent- +age points), Iraq (4 percentage points), and Lebanon (5 percentage points). Yet +elsewhere, for instance in Morocco and Egypt, the intergenerational gap is more +sizable (22 percentage points and 10 percentage points, respectively). Excep- +tions to this trend are found only among Tunisian and Jordanian youth, both of +whom voice satisfaction at higher rates than do older citizens (30 percent versus +28 percent and 65 percent versus 59 percent, respectively). + +www.arabbarometer.org -12 +8 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -The impact of digital transformation is profound yet challenging. Despite substantial investments, recent -executive surveys reveal that approximately 70% of digital transformation initiatives are falling short of -their objectives. A significant factor contributing to this shortfall is the persistent gap between technological -capabilities and workforce readiness. Nearly two-thirds of surveyed executives report that their -organizations struggle with change management and employee adoption of new digital systems. -Furthermore, 84% of organizations cite human-centric issues, rather than technical complications, as the -primary barriers to successful digital transformation [2]. - -The landscape of digital transformation has evolved significantly, with organizations focusing on three -critical areas: technology infrastructure modernization, process automation, and workforce enablement. -The IDC research highlights that industries such as financial services, manufacturing, and retail are leading -the investment charge, with discrete manufacturing alone expected to invest $382 billion in digital -transformation by 2026 [1]. However, the success of these investments heavily depends on organizations' -ability to address fundamental human elements. Studies indicate that organizations achieving digital -transformation success are three times more likely to prioritize workforce development and change -management strategies alongside their technological implementations [2]. - -The momentum of digital transformation continues to accelerate, driven by market demands and -competitive pressures. While the technological spending trajectory remains impressive, with manufacturing -and financial services each projected to exceed $1 trillion in total digital transformation investments through -2026 [1], the focus is increasingly shifting toward holistic transformation approaches. Successful -organizations are now recognizing that digital transformation requires a balanced investment in both -technology and people, with leadership development, skill enhancement, and cultural adaptation being as -crucial as the technological infrastructure itself [2]. - -The Evolution of Digital Transformation -The journey of digital transformation began with the basic digitization of manual processes, marking a -fundamental shift in business operations. Current research indicates that organizations implementing digital -transformation strategies have achieved remarkable improvements in their operational metrics. By -integrating automation and digital workflows, businesses have reported up to 80% reduction in process -completion times and a 90% decrease in human errors across various operational tasks. This evolution has -particularly impacted customer service operations, where automation has enabled organizations to handle -customer inquiries 60% faster while maintaining high satisfaction rates [3]. What started as simple -workflow automation—like digital leave management systems and basic inventory tracking—has evolved -into sophisticated, interconnected ecosystems that drive entire business operations. - -The transformation landscape has seen significant advancement in data processing capabilities, with -modern systems able to handle complex analytical tasks in real-time. Organizations implementing -comprehensive digital transformation strategies have witnessed a 70% improvement in their decision- -making processes, primarily due to enhanced data accessibility and automated analytical capabilities. + Arab Barometer – Wave V +Topic Report – Youth + +Looking to perceived corruption in the education system as one potential rea- +son contributing to low levels of satisfaction, in many countries more than half of +youth believe that rashwa (bribes) is required in order to receive better access to +education services. This belief is most commonly held in Egypt (64 percent), fol- +lowed by Lebanon (62 percent), Sudan (55 percent), Morocco (54 percent), Alge- +ria (53 percent), Yemen (53 percent), Iraq (52 percent), and Tunisia (50 percent). +However, while many youth believe that paying rashwa is necessary, this trend +is not common to all MENA youth as no more than a third of youth in Palestine +(33 percent), Libya (32 percent), and Jordan (29 percent) say the same regard- +ing rashwa to access better educational services. Prevalence of rashwa might +contribute to some dissatisfaction with state education systems, but it does not +account for or explain it in its entirety. + +Employment Opportunities + +Similar to satisfaction rates with state education systems, levels of youth satis- +faction with the government’s performance on creating employment opportu- +nities vary. However, despite this variation, across the region youth approval is +overwhelmingly weak, ranging from a low of five percent in Lebanon to a high +In addition to Lebanon, single-digit approval of +of only 32 percent in Yemen. +government performance on creating employment opportunities is seen in Iraq +(6 percent) and Algeria (9 percent). Fewer than two-in-ten youth in Morocco +(13 percent), Palestine (14 percent), Jordan (15 percent), Tunisia (16 percent), + +www.arabbarometer.org + +9 + + Arab Barometer – Wave V +Topic Report – Youth + +Sudan (17 percent), and Egypt (18 percent), and less than a quarter of youth in +Libya (22 percent) approve. + +Moreover, in addition to widespread dissatisfaction with government efforts to +improve employment opportunities, the vast majority of young Arab citizens +also believe that obtaining employment requires having strong connections. +Roughly nine-in-ten youth in Iraq (95 percent), Lebanon (94 percent), Jordan +(94 percent), Tunisia (94 percent), Libya (91 percent), Algeria (91 percent), Sudan +(90 percent), Palestine (90 percent), Egypt (90 percent), and Morocco (88 per- +cent) say that based on recent experience they think that obtaining employment +through wasta (personal connections) happens often or sometimes. Though +comparatively fewer, nearly seven-in-ten Yemeni youth (69 percent) hold the +same perception regarding the connection between wasta and employment. + +Emigration + +Given broad youth dissatisfaction with conditions in their countries, it is perhaps +not surprising that youth across the MENA region have a strong desire to em- +igrate. In six of eleven countries surveyed, roughly half or more of youth want +to leave their homeland. With the exception of Libya, North African and Jorda- +nian youth are most eager to migrate with about half or more citing a desire +to leave. Youth in Morocco cite not only the greatest desire to migrate, but also + +www.arabbarometer.org -13 +10 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 + Arab Barometer – Wave V +Topic Report – Youth + +show the greatest discrepancy with older generation’s desire to do so (70 per- +cent versus 25 percent). Meanwhile, over half of youth in Sudan (62 percent), +Jordan (59 percent), Tunisia (56 percent) and Algeria (56 percent) also wish to +leave their country. Roughly a third of Libyan (31 percent), Yemeni (33 percent), +and Lebanese (35 percent) youth say the same. + +Regional trends are evident with regard to which countries youth are most ea- +ger to migrate. Over two-thirds of youth in French-speaking North Africa want +to migrate to a European country. This is not surprising given both the proximity +of their countries to Europe and, perhaps more importantly, the close linkages +between France and these countries. In contrast, six-in-ten youth in Egypt, and +roughly half of youth in Yemen (48 percent) and Sudan (45 percent), say they +want to emigrate to a GCC country, which is in line with historic patterns of labor +migration to this area. Interestingly, fewer than half of youth across the region +hope to migrate to North America. Jordanian youth are most likely to cite a +desire to migrate to North America (42 percent), whereas Yemeni, Libyan, and +Tunisian youth are least likely (12 percent, 13 percent, and 13 percent). + +The least preferred destination for youth is non-GCC MENA countries. Fewer than +a third of youth across the region have considered emigrating to such countries. +Lebanese and Tunisian youth are least eager (4 percent and 5 percent, respec- +tively) while Yemeni youth are most eager, though still fewer than three-in-ten +Yemeni youth cite an interest in emigration to a non-GCC MENA country (27 + +www.arabbarometer.org + +11 + + Arab Barometer – Wave V +Topic Report – Youth + +percent). + +Perceived Civil Liberties + +Of the countries surveyed by the Arab Barometer, Tunisia is the only one cate- +gorized as free by Freedom House, with the remainder of countries identified as +partly free or not free. As such, it is unsurprising that youth in Tunisia are more +confident in their civil liberties like freedom of expression and right to peaceful +protest than are youth in other surveyed countries. However, despite being com- +paratively more confident in the protection of their civil liberties, Tunisian youth +are still quite similar to youth in the rest of the region who report generally low, +though varied, confidence in both freedom of speech and right to peacefully +protest. + +Looking first to freedom of speech, Tunisian youth are indeed most confident in +their freedom of expression with over six-in-ten saying that it is guaranteed to a +great or medium extent in the country (63 percent). More than half of youth in +Jordan (56 percent) and Lebanon (51 percent) say the same. In the rest of the +region fewer than half of respondents feel their freedom of expression is guar- +anteed. About half of youth in Yemen (49 percent), roughly four-in-ten youth +in Palestine (42 percent), Egypt (40 percent), and Morocco (39 percent), and +roughly a third of youth in Algeria (35 percent) and Iraq (33 percent) say that + +www.arabbarometer.org - Online ISSN: 2054-0965 (Online) +12 - Print ISSN: 2054-0957 (Print) + Arab Barometer – Wave V +Topic Report – Youth + +freedom of expression is guaranteed to a great or medium extent. Least confi- +dent of their freedom of expression are youth in Sudan (25 percent) and Libya +(24 percent). + +When comparing youth perceptions of their freedom of speech to the percep- +tions of older age cohorts, there is no clear regional trend. For instance, youth +in Lebanon (+10 points), Jordan (+3 points), Yemen (+4 points) and Palestine +(+4 points) are more confident in their freedom of expression than are older +citizens. Yet, the opposite trend is apparent in Morocco (-16 points), Egypt (-5 +points), and Tunisia (-4 points). +In the remainder of countries surveyed (Libya, +Sudan, Iraq, and Algeria) youth and older citizens do not diverge by more than +a point and rate their freedom of expression similarly. + +Youth perception of their freedom to participate in peaceful protests is also low +across the region, never exceeding half of youth in any country. Though nearly +half of youth in Tunisia (49 percent), Lebanon (49 percent), Jordan (47 percent), +and Yemen (46 percent) feel their freedom to peacefully demonstrate is guaran- +teed to a medium or great extent, fewer youth hold this belief elsewhere in the +region. Four-in-ten Palestinian youth, a third of Moroccan youth, and three-in- +ten Iraqi youth are confident of their freedom to participate in peaceful demon- +strations. Meanwhile, roughly a quarter of youth in Algeria (27 percent), Libya +(24 percent), and Egypt (24 percent) and fewer than two-in-ten in Sudan (19 +percent) say the same. + +www.arabbarometer.org - Website: https://www.eajournals.org/ +13 - Publication of the European Centre for Research Training and Development -UK + Arab Barometer – Wave V +Topic Report – Youth -Furthermore, the integration of artificial intelligence and machine learning has enabled predictive -maintenance systems that reduce equipment downtime by up to 45% [3]. +Attitudes on Gender -The Era of On-Premises Infrastructure -The first major milestone in this journey was the establishment of on-premises data centers, which -fundamentally changed how businesses managed their digital operations. On-premises data centers -emerged as critical infrastructure components, offering organizations complete control over their hardware, -software, and data management processes. This infrastructure model provided significant advantages in -terms of data security and compliance, particularly for industries handling sensitive information. A key -benefit was the ability to maintain an average network latency of less than 5 milliseconds, crucial for time- -sensitive applications and real-time data processing [4]. +Arab youth are divided in their views on gender equality. On the one hand, +most are supportive of equal rights for women, with the exception of inheri- +tance. However, there is substantially less support for women enjoying equal +roles in society. This is particularly evident in the political sphere. -The financial implications of on-premises infrastructure were substantial but justified for many -organizations. While initial setup costs typically ranged between $10,000 to $25,000 per rack, with annual -operating costs between $12,000 and $18,000, these investments provided organizations with direct control -over their technological infrastructure. The model proved particularly advantageous for businesses with -consistent workloads and stringent regulatory requirements, as it eliminated monthly recurring costs -associated with cloud services and provided better control over long-term expenses [4]. +Throughout the region fewer than a third of youth say that a university education +is more important for men than for women. Support for university education +for men compared with women is highest in Yemen and Sudan (27 percent, +each), followed by roughly two-in-ten youth in Jordan (23 percent), Egypt (22 +percent), Algeria (20 percent), Iraq (19 percent), Tunisia (17 percent) and Libya +(17 percent). Even lower rates of support for men’s education taking precedence +over that of women are found in Palestine (14 percent), Lebanon (13 percent), +and Morocco (12 percent). -Security and compliance capabilities represented another crucial aspect of on-premises infrastructure. -Organizations maintained full control over their security protocols, physical access management, and data -handling procedures. This control was particularly valuable for industries subject to strict regulatory -requirements, such as healthcare and financial services, where data sovereignty and compliance are -paramount concerns. The ability to customize security measures and maintain direct oversight of data -storage and processing activities allowed organizations to achieve compliance rates exceeding 99% [4]. +Youth also generally hold liberal views on the question of whether women should +have equal rights in the decision to get a divorce. The exception to this trend +is in Sudan, where support for women’s equal rights in deciding to divorce falls +below five-in-ten (46 percent). Support is highest in Lebanon (87 percent), fol- +lowed by Iraq (81 percent), Jordan (80 percent), Tunisia (78 percent), Morocco +(78 percent), Palestine (76 percent), Libya (75 percent) and Algeria (72 percent). +Though still greater than half, lower rates of support are seen in Egypt (64 per- -However, this approach presented notable challenges in terms of scalability and resource management. -Organizations needed to maintain dedicated IT teams for infrastructure management, with staffing costs -typically accounting for 40-50% of the total operational budget. Additionally, scaling operations required -significant advance planning and capital investment, with expansion projects often taking several months -to complete. The necessity for redundant systems and backup power solutions further added to the -complexity and cost of maintaining on-premises infrastructure [4]. +www.arabbarometer.org 14 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 1. Digital Transformation Performance Metrics: Year-over-Year Analysis (2021-2024) - -Metric Category - -Equipment Downtime Reduction -Customer Inquiry Speed -Decision-Making Efficiency -Infrastructure Staff Cost -Network Latency (ms) - -2021 (%) -25 -30 -40 -35 -20 - -2022 (%) -35 -45 -55 -40 -15 - -2023 (%) -40 -55 -65 -45 -10 - -2024 (%) -45 -60 -70 -50 -5 - -The Cloud Computing Revolution -The emergence of cloud computing marked a paradigm shift in the digital transformation landscape, -fundamentally altering how businesses approach their IT infrastructure and operational capabilities. Recent -market analysis reveals that cloud spending now represents approximately 30% of IT budgets across -industries, with this figure expected to grow significantly. The industry is witnessing an unprecedented shift -toward multi-cloud and hybrid cloud strategies, with organizations increasingly focusing on cloud cost -optimization and management. This evolution has led to the emergence of specialized roles such as cloud -financial operations (FinOps) professionals, who are becoming critical to managing cloud investments -effectively [5]. - -The adoption of cloud computing continues to accelerate, driven by both necessity and opportunity. -Industry research indicates that companies can capture up to $1 trillion in annual value through cloud -adoption, highlighting the transformative potential of this technology. This value creation stems from -various sources, including innovation acceleration, cost reduction, and improved resilience. The impact is -particularly significant in Fortune 500 companies, where cloud adoption has enabled revenue growth of up -to $100 billion annually through broader cloud-supported initiatives [6]. - -Key Benefits of Cloud Computing: Infrastructure Flexibility -Cloud platforms have revolutionized resource allocation capabilities, enabling unprecedented operational -efficiency. The shift toward cloud-native architectures has become increasingly prominent, with -organizations moving beyond basic "lift and shift" approaches to more sophisticated cloud-native -implementations. This trend has been accelerated by the growing adoption of containers and microservices, -with Kubernetes emerging as a de facto standard for container orchestration. The market has seen a -significant focus on edge computing capabilities, with major cloud providers expanding their edge presence -to support growing demands for low-latency applications and improved performance [5]. - -Cost Optimization -The financial implications of cloud adoption are substantial and multifaceted. Research shows that -organizations can potentially reduce IT costs by 20 to 30 percent through cloud optimization initiatives. -However, the most significant value creation opportunities lie in business innovation, which can generate -$770 billion to $995 billion in annual EBITDA across Fortune 500 companies. This value stems from + Arab Barometer – Wave V +Topic Report – Youth + +cent) and Yemen (55 percent). Notably, in most countries surveyed, the views +held by youth about women’s right to divorce are not significantly different from +the views held by older citizens. + +In a departure from predominantly liberal views on women’s rights, surprisingly +few youth in the region agree that women’s share of inheritance should be equal +to that of men. Nonetheless, Lebanon is a clear regional outlier to this trend with +over six-in-ten youth saying they agree or strongly agree that inheritance should +be equal between the sexes (62 percent). In contrast, far fewer youth elsewhere +are as supportive. Roughly three-in-ten youth in Iraq (30 percent) and Egypt (28 +percent), and around one-fourth of youth in Morocco (26 percent) and Tunisia +(24 percent) support equal inheritance. Still fewer youth in Algeria (18 percent), +Jordan (18 percent), Libya (16 percent), Sudan (13 percent), and Palestine (8 +percent) say the same. + +A plurality of youth in the region also believe that husbands should have the +final say in all family decisions. Support for this sentiment is greatest in Sudan +(73 percent), Algeria (73 percent), and Iraq (70 percent), followed by Egypt (66 +percent) and Libya (66 percent). Youth across Tunisia (56 percent), Yemen (55 +percent) and the Levant hold similar views to each other regarding women and +family decision-making. +In Lebanon and Jordan 55 percent of youth believe +husbands should have the final say in all family decisions, while in Palestine 50 +percent say the same. Though Moroccan youth are least likely to support limiting + +www.arabbarometer.org 15 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -various sources, including accelerated product development, improved customer experience, and enhanced -operational efficiency. The cloud's consumption-based pricing model has enabled organizations to achieve -15 to 20 percent cost reduction in infrastructure spending through improved capacity management [6]. - -Global Reach and Collaboration -Cloud infrastructure has enabled unprecedented global operational capabilities, with particular emphasis on -sustainability and environmental impact. Organizations are increasingly prioritizing environmental -considerations in their cloud strategies, with major providers committing to aggressive sustainability goals. -The industry has seen a growing focus on renewable energy usage and improved data center efficiency, -with cloud providers investing heavily in sustainable operations. This shift has resulted in significant -improvements in energy efficiency and reduced carbon footprints for organizations migrating to cloud -services [5]. - -The impact of cloud computing extends beyond operational benefits to strategic value creation. -Organizations leveraging cloud capabilities have reported significant improvements in their ability to -innovate and respond to market changes. The cloud has enabled faster time-to-market for new products and -services, with some companies reducing development cycles by up to 50 percent. Additionally, cloud -adoption has facilitated better risk management and improved security posture, with organizations able to -leverage advanced security features and compliance capabilities provided by cloud platforms. The value at -stake is substantial, with research indicating that companies can potentially capture between $900 billion -and $1.2 trillion in annual value through comprehensive cloud transformations [6]. + Arab Barometer – Wave V +Topic Report – Youth + +women’s roles in family decision-making, nonetheless nearly four-in-ten say that +husbands should have the final say in all family decisions (38 percent). Notably, +though youth in Morocco are far less likely than older generations to support +this notion (-15), youth in Jordan (+6), Lebanon (+7), Libya (+5), and Tunisia (+2) +are more supportive than are their older counterparts. + +In every country surveyed, with the exception of Morocco, over half of youth agree +that, in general, men are better at political leadership than women. Roughly +eight-in-ten youth in Sudan (80 percent) and Jordan (78 percent), and roughly +three-quarters of youth in Algeria (74 percent) hold this belief. Across much of +the rest of the region youth hold this view at similar rates. For instance, around +seven-in-ten youth say men are better at political leadership in Egypt (70 per- +cent), Iraq (69 percent), Yemen (68 percent), Palestine (68 percent), and Libya +(67 percent). Meanwhile, fewer in Tunisia, Lebanon, and Morocco say that men +are better at political leadership than women (52 percent, 51 percent, and 31 +percent, respectively). + +Interestingly, despite unfavorably comparing female political leaders to men and +criticizing women’s role as decision-makers in the family, Arab youth are often +open both to the idea of a woman being a leader of a Muslim country and to +setting aside a certain percentage of elected positions for women as a means +of attaining fairer political representation. More than half of youth support the +notion of a female president or prime minister of a Muslim nation in all coun- + +www.arabbarometer.org 16 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 2. Cloud Adoption: Cost Reduction and Efficiency Analysis [5, 6]. - -Metric Category - -Q1 2024 (%) Q2 2024 (%) Q3 2024 (%) Q4 2024 (%) - -IT Budget Cloud Spend -IT Cost Reduction -Infrastructure Cost Savings -Development Cycle Reduction -Market Response Improvement - -15 -20 -15 -35 -25 - -20 -23 -17 -40 -30 - -25 -27 -18 -45 -35 - -30 -30 -20 -50 -40 - -Advanced Technologies Enabled by Cloud Computing - -Artificial Intelligence and Machine Learning -Cloud platforms have fundamentally democratized access to AI and ML capabilities, creating -unprecedented opportunities for organizations across industries. According to recent market analysis, the -global artificial intelligence market size was valued at USD 454.12 billion in 2023 and is projected to surge -to USD 2,575.16 billion by 2032, growing at an impressive CAGR of 21.4%. This remarkable growth is -driven by the increasing adoption of cloud-based AI solutions across various sectors, including healthcare, -automotive, and retail. The integration of AI with cloud computing has become particularly crucial in the -post-pandemic era, with organizations accelerating their digital transformation initiatives to maintain -competitive advantage [7]. - -The impact of cloud-enabled AI extends across multiple domains, with North America leading the market -share at 35% in 2023. This dominance is attributed to the presence of major technology companies and -widespread adoption of AI technologies in various industries. The Asia-Pacific region is experiencing the -fastest growth, with a projected CAGR of 24.3% during the forecast period, driven by increasing -investments in AI research and development, particularly in countries like China, Japan, and South Korea -[7]. - -Big Data Analytics and IoT Integration -The transformation in data handling and analysis capabilities through cloud computing has revolutionized -how organizations process and utilize information. Edge-cloud infrastructure has emerged as a crucial -component in handling the massive data generated by IoT devices, with studies showing that edge -computing can reduce data transmission costs by up to 30% while improving response times significantly. -The integration of advanced analytics algorithms in edge-cloud architectures has enabled real-time -processing capabilities, crucial for applications requiring immediate insights and actions [8]. - -The evolution of IoT data analytics has led to the development of sophisticated middleware architectures -that efficiently manage data flow between edge devices and cloud platforms. Research indicates that hybrid -edge-cloud solutions can reduce latency by up to 50% compared to traditional cloud-only approaches. These -architectures implement various data analytic algorithms, including deep learning, reinforcement learning, + Arab Barometer – Wave V +Topic Report – Youth + +tries surveyed, excluding Algeria (37 percent). Roughly half of youth in Sudan +(51 percent) and Libya (54 percent), six-in-ten youth in Yemen (57 percent), Jor- +dan (60 percent), and Palestine (64 percent) and roughly seven-in-ten youth in +Tunisia (67 percent), Egypt (68 percent) and Iraq (69 percent) affirm that women +can become prime ministers or presidents of Muslim countries. Moroccan and +Lebanese youth (77 percent and 79 percent, respectively) are the most accept- +ing of this right. + +The majority of MENA youth also believe that in order to achieve fairer represen- +tation a certain percentage of elected positions should be set aside for women. +Support for this right exceeds half of all youth in all countries surveyed. Youth in +Iraq and Lebanon (79 percent, each) are most supportive, followed by Sudan (76 +percent), Palestine (75 percent), Jordan (72 percent) and Yemen (69 percent). +Excluding Sudan, North African youth show the least support for women’s quo- +tas in elected positions with less than two-thirds of youth in Egypt (66 percent), +Libya (65 percent), Tunisia (63 percent), Morocco (62 percent) and Algeria (53 +percent) saying they agree or strongly agree with this idea. + +International Relations + +In over half of the countries surveyed, youth are most supportive of stronger ties +with China and least interested in stronger ties with the United States among +global powers. Youth in Jordan (70 percent), Tunisia (65 percent), Sudan (62 +percent), Libya (61 percent), Yemen (50 percent), Palestine (49 percent), and +Iraq (47 percent) prefer strengthening ties with China at a higher rate than they +do for ties with the United States, the United Kingdom, or Russia. Exceptions to +this include youth in Morocco, who have a slightly stronger preference for closer +ties with the United Kingdom (63 percent), youth in Lebanon and Algeria, who +are more supportive by a small margin of stronger ties with Russia (38 percent +and 42 percent, respectively), and Egyptian youth, who are equally supportive +of strengthening ties with the United States and Russia (41 percent, each). + +In a majority of countries surveyed youth support for stronger ties with the United +States not only falls below 50 percent, but is also lower than youth support for +strengthening ties with other world powers. Exceptions to the former are Sudan +(60 percent), Morocco (57 percent), and Jordan (55 percent) where youth sup- +port exceeds half. In contrast, fewer than half of youth in Tunisia (47 percent), +and roughly a third of youth in Libya (36 percent), Iraq (34 percent), Palestine, +Lebanon, and Algeria (31 percent, each) say that relations with the United States +should become stronger. Of all countries surveyed Yemeni youth show not only +the least interest in closer ties with the United States (22 percent), but they are +also least interested in strengthening ties with Russia (26 percent) relative to +youth in other countries. + +www.arabbarometer.org 17 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -and federated learning, enabling more efficient processing of IoT data streams. The middleware layer plays -a crucial role in managing these processes, with modern implementations supporting dynamic resource -allocation and automated scaling based on workload demands [8]. - -The convergence of IoT and cloud computing has particularly benefited from advances in edge computing -capabilities. Contemporary edge-cloud infrastructures support multiple types of edge analytics, including -descriptive, diagnostic, predictive, and prescriptive analytics. This comprehensive approach enables -organizations to process data closer to its source, reducing bandwidth requirements while maintaining data -quality and analysis accuracy. Studies have shown that edge-cloud implementations can achieve up to 40% -improvement in energy efficiency compared to traditional cloud-only solutions [8]. - -Security considerations have become increasingly important in IoT data analytics, with edge-cloud -architectures implementing sophisticated security measures at multiple levels. The integration of -blockchain technology with edge-cloud infrastructure has emerged as a promising solution for ensuring -data integrity and secure communication between devices. Furthermore, the development of privacy- -preserving machine learning techniques has enabled organizations to maintain data confidentiality while -performing complex analytics tasks [8]. - -Table 3. Cloud-Enabled Technologies: Growth and Efficiency Analysis [7, 8]. - -Technology Metric - -2023 Q3 (%) 2023 Q4 (%) 2024 Q1 (%) 2024 Q2 (%) - -North America AI Market Share -APAC Growth Rate -Data Transmission Cost Reduction -Edge-Cloud Latency Reduction -Energy Efficiency Improvement - -35 -20 -20 -35 -25 - -37 -22 -25 -40 -30 - -40 -24 -28 -45 -35 + Arab Barometer – Wave V +Topic Report – Youth -42 -26 -30 -50 -40 +Looking to regional powers, one clear trend is that Arab youth are generally less +interested in stronger relations with Iran and more interested in improving re- +lations with Turkey. Though support for stronger ties with Iran never exceeds +four-in-ten youth in any country surveyed, support is highest in Lebanon (38 per- +cent), followed by Sudan (36 percent), Palestine (34 percent), Iraq (32 percent) +and Tunisia (31 percent). In Lebanon and Iraq, sectarian divides might explain +the comparatively greater support for strengthening ties with Iran, a country +with a strong Shi’ite majority. Roughly two-in-ten youth in Morocco (22 per- +cent), Jordan (21 percent), Yemen (20 percent), Algeria (20 percent), and Libya +(18 percent) prefer stronger relations with Iran, while only 8 percent of Egyptian +youth say the same. -Emerging Trends in Cloud Computing - -Edge Computing and Serverless Architecture -Edge computing has emerged as a transformative force in the cloud computing landscape, bringing -computation closer to data sources and end-users. The integration of edge computing with serverless -architecture represents a significant evolution in cloud technology, addressing critical challenges in modern -application development and deployment. This combination has become particularly crucial for real-time -processing requirements, with edge computing reducing latency by processing data closer to its source -while serverless architecture eliminates the complexity of server management. The implementation of edge- -serverless solutions has shown remarkable improvements in application performance, with organizations -reporting reduction in latency from seconds to milliseconds for critical operations [9]. -The impact of serverless computing extends beyond performance improvements. Recent implementations -have demonstrated significant advantages in terms of cost efficiency and resource utilization. Organizations +www.arabbarometer.org 18 - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -adopting serverless architectures have reported substantial reductions in operational overhead, as the -technology eliminates the need for traditional server management and scaling operations. The pay-as-you- -go model inherent in serverless computing has revolutionized cost structures, with businesses only paying -for actual compute time used rather than maintaining constantly running servers. Furthermore, the -integration of edge computing with serverless architecture has enabled more efficient handling of IoT -devices and real-time data processing requirements, resulting in improved application responsiveness and -user experience [9]. - -Hybrid Cloud Solutions -The hybrid cloud market has witnessed extraordinary growth and transformation, with current valuations -at USD 183.7 billion and projected to reach an impressive USD 405.62 billion by 2032, growing at a CAGR -of 9.2%. This remarkable expansion is driven by increasing demands for flexible, scalable, and secure cloud -solutions across various industries. The financial services sector has emerged as a particularly strong -adopter of hybrid cloud solutions, leveraging the technology to balance security requirements with -operational efficiency [10]. - -The adoption of hybrid cloud solutions has been significantly influenced by several key factors, including -the rising need for enhanced data security, disaster recovery capabilities, and the growing complexity of -regulatory compliance requirements. Organizations in the banking, financial services, and insurance (BFSI) -sector have been at the forefront of hybrid cloud adoption, particularly in North America, which currently -holds approximately 35% of the global market share. The integration of artificial intelligence and machine -learning capabilities within hybrid cloud environments has further accelerated adoption rates, enabling -organizations to optimize workload placement and resource utilization more effectively [10]. - -Market analysis reveals that large enterprises currently dominate the hybrid cloud market, accounting for a -significant portion of the overall market share. This dominance is attributed to their complex IT -requirements and substantial resources for implementation. However, small and medium-sized enterprises -(SMEs) are increasingly adopting hybrid cloud solutions, driven by the technology's ability to provide -enterprise-grade capabilities without requiring extensive upfront investments. The public sector has also -emerged as a significant growth driver, with government agencies increasingly migrating to hybrid cloud -environments to modernize their IT infrastructure while maintaining strict security and compliance -requirements [10]. - -Looking Ahead: Future Developments -The convergence of edge computing, serverless architecture, and hybrid cloud solutions continues to -evolve, with emerging technologies and methodologies shaping the future of cloud computing. Edge- -serverless platforms are becoming increasingly sophisticated, incorporating advanced features such as -automated scaling, improved security protocols, and enhanced development tools. The integration of -artificial intelligence and machine learning capabilities at the edge is expected to further transform -application development and deployment paradigms, enabling more intelligent and responsive systems [9]. - -19 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -Table 4. Cloud Architecture Evolution: Latency and Processing Metrics [9, 10]. - -Performance Indicator - -Traditional Edge Only Serverless Edge-Serverless - -Processing Latency (ms) -Resource Overhead (%) -Server Management Time (hrs/week) -Response Time (ms) -Infrastructure Complexity Score - -45 -65 -40 -50 -75 - -25 -45 -30 -35 -55 - -15 -35 -20 -20 -40 - -5 -25 -15 -8 -30 - -Security and Compliance in the Cloud Era -The landscape of cloud security continues to evolve rapidly in response to emerging threats and changing -business requirements. Recent analysis indicates that cloud security remains a top priority for organizations, -with 82% of companies increasing their cloud security budgets in 2024. The shift towards zero trust -architecture has become particularly prominent, with 76% of organizations either implementing or planning -to implement zero trust security models. This transition has been driven by the growing recognition that -traditional perimeter-based security approaches are insufficient in today's distributed cloud environments, -where data and applications are accessed from multiple locations and devices [11]. - -Identity and access management has emerged as a critical component of modern cloud security strategies. -The implementation of comprehensive IAM solutions has become essential as organizations grapple with -increasingly sophisticated cyber threats. Studies show that organizations with mature identity and access -management practices experience significantly fewer security incidents, with advanced authentication -mechanisms reducing unauthorized access attempts by up to 67%. The adoption of biometric authentication -and behavioral analytics has further strengthened security postures, with organizations reporting a 45% -reduction in credential-based attacks [11]. - -The financial implications of data breaches have underscored the importance of robust cloud security -measures. According to IBM's Cost of a Data Breach Report 2024, the global average cost of a data breach -has reached $4.45 million, marking a 15% increase over three years. Organizations leveraging AI and -automation in their security operations have demonstrated significantly better outcomes, with automated -security tools reducing the average time to identify and contain breaches by 108 days. Furthermore, -companies implementing zero trust architecture have seen an average reduction of $1.17 million in breach -costs compared to those without such security measures [12]. - -The impact of remote work and hybrid environments continues to shape cloud security strategies. -Organizations with over 50% of their workforce working remotely have reported higher average breach -costs, reaching $4.99 million compared to $4.24 million for organizations with less remote work. This -disparity has driven increased adoption of advanced security technologies, with 49% of organizations now -implementing AI platforms for cybersecurity. The study reveals that organizations using security AI and - -20 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -automation experienced breach lifecycle times that were 108 days shorter compared to those not utilizing -these technologies [12]. - -Cloud security compliance has become increasingly complex with the evolution of regulatory requirements -and data protection standards. Recent trends show a significant shift towards automated compliance -monitoring and reporting tools, with organizations implementing these solutions reporting 43% lower -compliance-related costs. The adoption of cloud security posture management (CSPM) tools has grown -substantially, with 65% of organizations now using these platforms to maintain continuous compliance and -security assessment capabilities. This has led to a 55% improvement in the detection and remediation of -compliance violations [11]. - -Future Outlook: Cloud Computing and Digital Transformation -The landscape of cloud computing continues to evolve rapidly, with organizations increasingly focusing on -strategic cloud adoption and optimization. According to recent industry analysis, 2024 marks a significant -shift toward more sophisticated cloud strategies, with 94% of enterprises already utilizing cloud services -and 87% adopting multi-cloud approaches. This evolution is particularly noteworthy in the realm of cloud -cost optimization, where organizations are implementing FinOps practices to better manage and optimize -their cloud spending. The trend toward cloud-native development has become increasingly prominent, with -enterprises reporting 30-40% faster application development cycles through cloud-native approaches [13]. -The integration of artificial intelligence and machine learning capabilities has become a cornerstone of -modern cloud strategies. Organizations are increasingly leveraging AI-powered cloud services for various -purposes, from automated infrastructure management to intelligent application development. The adoption -of cloud-native platforms has enabled businesses to achieve remarkable improvements in operational -efficiency, with companies reporting 25-35% reduction in development cycles and 40-50% improvement -in resource utilization. Furthermore, the implementation of containerization and microservices architectures -has led to significant improvements in application scalability and maintenance efficiency [13]. - -Sustainability and environmental responsibility have emerged as critical priorities in cloud computing -strategies for 2024 and beyond. Organizations are increasingly focusing on green cloud computing -initiatives, with major providers implementing advanced energy management systems and sustainable data -center practices. This shift towards environmentally conscious cloud operations has become a key -differentiator in vendor selection, with businesses prioritizing providers that demonstrate strong -commitments to sustainability. The industry is witnessing a growing emphasis on energy-efficient cloud -architectures and sustainable computing practices [14]. - -Security and compliance considerations continue to shape the future of cloud computing, with zero trust -architectures gaining prominence. The industry has seen a significant shift toward comprehensive security -frameworks that encompass everything from infrastructure protection to application security. Organizations -are increasingly adopting automated security solutions and implementing advanced threat detection -systems. The focus on regulatory compliance has intensified, with businesses implementing sophisticated - -21 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK + Arab Barometer – Wave V +Topic Report – Youth -compliance monitoring and management tools to ensure adherence to evolving regulatory requirements -[14]. +In comparison to other regional powers (Saudi Arabia, Qatar, and Iran), Arab +youth tend to prefer strengthening relations with Turkey. Youth in Jordan (82 +percent), Palestine (75 percent), Tunisia (64 percent), Yemen (61 percent), Mo- +rocco (56 percent), Algeria (53 percent), and Iraq (45 percent), are all more likely +to want improved relations with Turkey than to do so for relations with Saudi Ara- +bia, Qatar, or Iran. Youth in Sudan prefer strengthening relations with Saudi Ara- +bia, Turkey, and Qatar at similar rates (70 percent, 73 percent, and 74 percent, +respectively), as do youth in Lebanon (41 percent, 41 percent, and 42 percent, +respectively). In Libya and Egypt youth by far prefer improving ties with Saudi +Arabia (55 percent and 58 percent). -The evolution of cloud management and governance frameworks represents another significant trend. -Organizations are implementing more sophisticated approaches to cloud governance, incorporating -advanced monitoring tools and automated policy enforcement mechanisms. The emphasis on operational -excellence has led to the adoption of advanced observability platforms and AIOps solutions, enabling better -visibility and control over cloud environments. This trend is particularly evident in the increasing adoption -of cloud management platforms that provide comprehensive visibility and control across multi-cloud -environments [14]. +www.arabbarometer.org -CONCLUSION - -Cloud computing stands as a transformative force in the digital era, reshaping how organizations operate, -innovate, and deliver value. The convergence of cloud technologies with AI, machine learning, and IoT has -created powerful new capabilities for businesses of all sizes. As organizations continue to adapt to changing -market demands, the role of cloud computing extends beyond technological infrastructure to become a -strategic enabler of business transformation. The future holds promise for further innovations in edge -computing, serverless architectures, and sustainable cloud practices, while security and compliance remain -paramount considerations. Organizations that successfully leverage these cloud capabilities while -maintaining robust security measures and embracing sustainable practices will be best positioned to thrive -in an increasingly digital world. - -REFERENCES - -[1] Businesswire, "IDC Spending Guide Sees Worldwide Digital Transformation Investments Reaching - -$3.4 Trillion in 2026," 2022. [Online]. Available: -https://www.businesswire.com/news/home/20221026005193/en/IDC-Spending-Guide-Sees- -Worldwide-Digital-Transformation-Investments-Reaching-%243.4-Trillion-in-2026 - -[2] Noah Barsky, "Majority Of Digital Transformations Off Track, Executive Survey Finds—Here Are 3 - -Fixes," Forbes, 2023. [Online]. Available: -https://www.forbes.com/sites/noahbarsky/2023/10/02/people-problems-plague-digital- -transformation/ - -[3] Arbab Khan, "Unlocking the Power of Digital Transformation and Automation," Rushford, 2024. - -[Online]. Available: https://rushford.ch/insights/unlocking-the-power-of-digital-transformation- -and-automation/ - -[4] George, "On-Premises vs. Cloud Data Center: Which Is Right for Your Business?" FS, 2022. - -[Online]. Available: https://www.fs.com/blog/onpremises-vs-cloud-data-center-which-is-right- -for-your-business-5867.html - -[5] Neal Weinberg, "5 cloud market trends and how they will impact IT," CIO, 2024. [Online]. Available: - -https://www.cio.com/article/655693/7-cloud-market-trends-and-how-they-will-impact-it.html - -22 - - European Journal of Computer Science and Information Technology,13(18),12-23, 2025 - - Online ISSN: 2054-0965 (Online) - - Print ISSN: 2054-0957 (Print) - - Website: https://www.eajournals.org/ - - Publication of the European Centre for Research Training and Development -UK - -[6] Will Forrest et al., "Cloud’s trillion-dollar prize is up for grabs," McKinsey Digital, 2021. [Online]. - -Available: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion- -dollar-prize-is-up-for-grabs - -[7] Precedence Research, "Artificial Intelligence (AI) Market Size, Share, and Trends 2025 to 2034," - -2025. [Online]. Available: https://www.precedenceresearch.com/artificial-intelligence-market -[8] Abel E. Edje, M.S. Abd Latiff and Weng Howe Chan, "IoT data analytic algorithms on edge-cloud - -infrastructure: A review," ScienceDirect, 2023. [Online]. Available: -https://www.sciencedirect.com/science/article/pii/S2352864823001530#:~:text=In%20addition% -2C%20future%20advances%20are,Comparison%20of%20previous%20research%20surveys.&te -xt=Comparison%20of%20middleware%20architectures. - -[9] IvanDev, "The Rise of Serverless Architecture and Edge Computing," DEV, 2025. [Online]. - -Available: https://dev.to/ivansing/the-rise-of-serverless-architecture-and-edge-computing-2aab - -[10] Globe Newswire, "Hybrid Cloud Market Projected to Hit USD 405.62 Billion by 2032| SNS - -Insider," 2025. [Online]. Available: https://www.globenewswire.com/news- -release/2025/03/21/3047113/0/en/Hybrid-Cloud-Market-Projected-to-Hit-USD-405-62-Billion- -by-2032-SNS-Insider.html - -[11] Thales, "Cloud Security in 2024: Addressing the Shifting Landscape," CSA, 2024. [Online]. +19 -Available: https://cloudsecurityalliance.org/blog/2024/06/27/cloud-security-in-2024-addressing- -the-shifting-landscape + About Arab Barometer -[12] MAHESH NAWALE, "7 Key Takeaways From IBM's Cost of a Data Breach Report 2024," ZScaler, -2024. [Online]. Available: https://www.zscaler.com/blogs/product-insights/7-key-takeaways-ibm- -s-cost-data-breach-report-2024 +The Arab Barometer is a nonpartisan research network that +provides insights into the social, political, and economic attitudes +and values of ordinary citizens across the Arab world. -[13] Andrii Stashko-Ziablietsov and Oleksandr Virga, "Cloud strategy: a 2024 guide," Avenga, 2024. +We have been conducting rigorous, and nationally representative +face-to-face public opinion surveys on probability samples of the +adult populations across the Arab world since 2006. The margin of +error is ±3 percent. -[Online]. Available: https://www.avenga.com/magazine/cloud-strategy/ +The Arab Barometer is the largest repository of publicly available +data on the views of men and women in the MENA region. Our +findings give a voice to the needs and concerns of Arab publics. -[14] Willie Devitt, "Cloud Technology Trends and Predictions for 2024," Eviden, 2024. [Online]. +.......... -Available: https://eviden.com/publications/whitepapers/cloud-technology-trends-and-predictions- -for-2024/#:~:text=reputation%20and%20trustworthiness- -,At%20Eviden%2C%20we%20predict%20that%20in%202024%2C%20the%20focus%20on,com -pliance%20posture%20for%20our%20customers. +.......... -23 +.......... +ARABBAROMETER.ORGARABBAROMETER@ARABBAROMETER \ No newline at end of file diff --git a/q104/random_k4/random_3.md b/q104/random_k4/random_3.md index fea36abbe8215721fb9751f54461e22cbeefc712..b3f566f113a69a4c8533f2f4d870e45002f6953a 100644 --- a/q104/random_k4/random_3.md +++ b/q104/random_k4/random_3.md @@ -1,1211 +1,1617 @@ -© Lonely Planet Publications Pty Ltd +0 +2 +0 +2 + +p +e +S +7 -BECKY OHLSEN +] -SEATTLE +Y +C +. +s +c +[ + +1 +v +0 +0 +3 +3 +0 +. +9 +0 +0 +2 +: +v +i +X +r +a -C I T Y G U I D E +Measuring Massive Multitask Language Understanding + +Dan Hendrycks +UC Berkeley + +Collin Burns +Columbia University + +Steven Basart +UChicago + +Andy Zou +UC Berkeley + +Mantas Mazeika +UIUC + +Dawn Song +UC Berkeley + +Jacob Steinhardt +UC Berkeley + +ABSTRACT + +We propose a new test to measure a text model’s multitask accuracy. The test +covers 57 tasks including elementary mathematics, US history, computer science, +law, and more. To attain high accuracy on this test, models must possess extensive +world knowledge and problem solving ability. We find that while most recent +models have near random-chance accuracy, the very largest GPT-3 model improves +over random chance by almost 20 percentage points on average. However, on every +one of the 57 tasks, the best models still need substantial improvements before +they can reach human-level accuracy. Models also have lopsided performance +and frequently do not know when they are wrong. Worse, they still have near- +random accuracy on some socially important subjects such as morality and law. +By comprehensively evaluating the breadth and depth of a model’s academic and +professional understanding, our test can be used to analyze models across many +tasks and to identify important shortcomings. + +1 + +INTRODUCTION + +Natural Language Processing (NLP) models have achieved superhuman performance on a number of +recently proposed benchmarks. However, these models are still well below human level performance +for language understanding as a whole, suggesting a disconnect between our benchmarks and the +actual capabilities of these models. The General Language Understanding Evaluation benchmark +(GLUE) (Wang et al., 2018) was introduced in 2018 to evaluate performance on a wide range of NLP +tasks, and top models achieved superhuman performance within a year. To address the shortcomings +of GLUE, researchers designed the SuperGLUE benchmark with more difficult tasks (Wang et al., +2019). About a year since the release of SuperGLUE, performance is again essentially human-level +(Raffel et al., 2019). While these benchmarks evaluate linguistic skills more than overall language +understanding, an array of commonsense benchmarks have been proposed to measure basic reasoning +and everyday knowledge (Zellers et al., 2019; Huang et al., 2019; Bisk et al., 2019). However, these +recent benchmarks have similarly seen rapid progress (Khashabi et al., 2020). Overall, the near +human-level performance on these benchmarks suggests that they are not capturing important facets +of language understanding. + +Transformer models have driven this recent progress by pretraining on massive text corpora, including +all of Wikipedia, thousands of books, and numerous websites. These models consequently see +extensive information about specialized topics, most of which is not assessed by existing NLP +benchmarks. It consequently remains an open question just how capable current language models are +at learning and applying knowledge from many domains. + +To bridge the gap between the wide-ranging knowledge that models see during pretraining and the +existing measures of success, we introduce a new benchmark for assessing models across a diverse +set of subjects that humans learn. We design the benchmark to measure knowledge acquired during +pretraining by evaluating models exclusively in zero-shot and few-shot settings. This makes the +benchmark more challenging and more similar to how we evaluate humans. The benchmark covers +57 subjects across STEM, the humanities, the social sciences, and more. It ranges in difficulty +from an elementary level to an advanced professional level, and it tests both world knowledge and +problem solving ability. Subjects range from traditional areas, such as mathematics and history, to + +1 + + (a) An example of few-shot learning and inference us- +ing GPT-3. The blue underlined bold text is the auto- +completed response from GPT-3, while the preceding +text is the user-inputted prompt. In this 2-shot learning +example, there are two instruction examples and one +initially incomplete example. On average, GPT-3 has +low accuracy on high school mathematics questions. + +(b) Performance on a commonsense benchmark (Hel- +laSwag), a linguistic understanding benchmark (Super- +GLUE), and the massive multitask test. On previous +benchmarks, smaller models start well above random +chance levels and exhibit more continuous improve- +ments with model size increases, but on our test, GPT-3 +moves beyond random chance with the largest model. + +more specialized areas like law and ethics. The granularity and breadth of the subjects makes the +benchmark ideal for identifying a model’s blind spots. + +We find that meaningful progress on our benchmark has only become possible in recent months. +In particular, models up to 13 billion parameters (Brown et al., 2020) achieve random chance +performance of 25% accuracy, but the 175 billion parameter GPT-3 model reaches a much higher +43.9% accuracy (see Figure 1b). On the other hand, unlike human professionals GPT-3 does not +excel at any single subject. Instead, we find that performance is lopsided, with GPT-3 having almost +70% accuracy for its best subject but near-random performance for several other subjects. + +Our results indicate that while recent advances have been impressive, state-of-the-art models still +struggle at learning and applying knowledge from pretraining. The tasks with near-random accuracy +include calculation-heavy subjects such as physics and mathematics and subjects related to human +values such as law and morality. This second weakness is particularly concerning because it will +be important for future models to have a strong understanding of what is legal and what is ethical. +Worryingly, we also find that GPT-3 does not have an accurate sense of what it does or does not know +since its average confidence can be up to 24% off from its actual accuracy. We comprehensively +evaluate the breadth and depth of a model’s text understanding by covering numerous topics that +humans are incentivized to learn. Since our test consists in 57 tasks, it can be used to analyze +aggregate properties of models across tasks and to track important shortcomings. The test and code is +available at github.com/hendrycks/test. + +2 RELATED WORK + +Pretraining. The dominant paradigm in NLP is to pretrain large models on massive text corpora +including educational books and websites. In the process, these models are exposed to information +about a wide range of topics. Petroni et al. (2019) found that recent models learn enough information +from pretraining that they can serve as knowledge bases. However, no prior work has comprehensively +measured the knowledge models have across many real-world domains. + +Until recently, researchers primarily used fine-tuned models on downstream tasks (Devlin et al., 2019). +However, larger pretrained models like GPT-3 (Brown et al., 2020) have made it possible to achieve +competitive performance without fine-tuning by using few-shot learning, which removes the need for +a large fine-tuning set. With the advent of strong zero-shot and few-shot learning, it is now possible +to curate a diverse set of tasks for evaluation and remove the possibility of models on “spurious cues” +(Geirhos et al., 2020; Hendrycks et al., 2019b) in a dataset to achieve high performance. - I NTRO DUCI N G SE AT TLE +2 -Kayaking on Lake Union (p166) +Few Shot Prompt and Predicted AnswerHow many numbers are in the list 25, 26, ..., 100?(A) 75 (B) 76 (C) 22 (D) 23Answer: BCompute i+i2+i3+···+i258+i259.(A) -1 (B) 1 (C) i (D) -iAnswer: AIf 4 daps = 7 yaps, and 5 yaps = 3 baps,how many daps equal 42 baps?(A) 28 (B) 21 (C) 40 (D) 30Answer: C␣The following are multiple choice questionsabout high school mathematics.SmallMediumLargeX-LargeModel Size2030405060708090Performance (%)GPT-3 Few Shot Test PerformanceCommonsenseLinguisticsKnowledge (Ours) Benchmarks. Many recent benchmarks aim to assess a model’s general world knowledge and basic +reasoning ability by testing its “commonsense.” A number of commonsense benchmarks have been +proposed in the past year, but recent models are already nearing human-level performance on several +of these, including HellaSwag (Zellers et al., 2019), Physical IQA (Bisk et al., 2019), and CosmosQA +(Huang et al., 2019). By design, these datasets assess abilities that almost every child has. In contrast, +we include harder specialized subjects that people must study to learn. + +Some researchers have suggested that the future of NLP evaluation should focus on Natural Language +Generation (NLG) (Zellers et al., 2020), an idea that reaches back to the Turing Test (Turing, 1950). +However, NLG is notoriously difficult to evaluate and lacks a standard metric (Sai et al., 2020). +Consequently, we instead create a simple-to-evaluate test that measures classification accuracy on +multiple choice questions. + +While several question answering benchmarks exist, they are comparatively limited in scope. Most +either cover easy topics like grade school subjects for which models can already achieve strong +performance (Clark et al., 2018; Khot et al., 2019; Mihaylov et al., 2018; Clark et al., 2019), or +are focused on linguistic understanding in the form of reading comprehension (Lai et al., 2017; +Richardson et al., 2013). In contrast, we include a wide range of difficult subjects that go far beyond +linguistic understanding. + +3 A MULTITASK TEST + +We create a massive multitask test consisting of multiple-choice questions from various branches of +knowledge. The test spans subjects in the humanities, social sciences, hard sciences, and other areas +that are important for some people to learn. There are 57 tasks in total, which is also the number +of Atari games (Bellemare et al., 2013), all of which are listed in Appendix B. The questions in +the dataset were manually collected by graduate and undergraduate students from freely available +sources online. These include practice questions for tests such as the Graduate Record Examination +and the United States Medical Licensing Examination. It also includes questions designed for +undergraduate courses and questions designed for readers of Oxford University Press books. Some +tasks cover a subject, like psychology, but at a specific level of difficulty, such as “Elementary,” +“High School,” “College,” or “Professional.” For example, the “Professional Psychology” task draws +on questions from freely available practice questions for the Examination for Professional Practice +in Psychology, while the “High School Psychology” task has questions like those from Advanced +Placement Psychology examinations. + +We collected 15908 questions in total, which we split into a few-shot development set, a validation +set, and a test set. The few-shot development set has 5 questions per subject, the validation set may +be used for selecting hyperparameters and is made of 1543 questions, and the test set has 14080 +questions. Each subject contains 100 test examples at the minimum, which is longer than most exams +designed to assess people. + +Since our test aggregates different subjects and several levels of difficulty, we measure more than +straightforward commonsense or narrow linguistic understanding. Instead, we measure arbitrary +real-world text understanding. Since models are pretrained on the Internet, this enables us to test how +well they can extract useful knowledge from massive corpora. To succeed at our test, future models +should be well-rounded, possess extensive world knowledge, and develop expert-level problem +solving ability. These properties make the test likely to be an enduring and informative goalpost. + +3.1 HUMANITIES + +The humanities is a group of disciplines that make use of qualitative analysis and analytic methods +rather than scientific empirical methods. Branches of the humanities include law, philosophy, history, +and so on (Appendix B). Mastering these subjects requires a variety of skills. For example, legal +understanding requires knowledge of how to apply rules and standards to complex scenarios, and +also provide answers with stipulations and explanations. We illustrate this in Figure 2. Legal +understanding is also necessary for understanding and following rules and regulations, a necessary +capability to constrain open-world machine learning models. For philosophy, our questions cover +concepts like logical fallacies, formal logic, and famous philosophical arguments. It also covers +moral scenarios, including questions from the ETHICS dataset (Hendrycks et al., 2020) that test a +model’s understanding of normative statements through predicting widespread moral intuitions about -RICHARD CUMMINS +3 -It’s raining again, but you don’t mind. This is your cue -to duck inside for another latte. Maybe you’ll chase it -with a local microbrew, then tackle the Space Needle -before rocking the Crocodile Café. + Figure 2: This task requires understanding detailed and dissonant scenarios, applying appropriate +legal precedents, and choosing the correct explanation. The green checkmark is the ground truth. -Welcome to Seattle, whose perpetually cloudy skies mean that if the sun peeks out for a second, -everyone drops whatever they’re doing and gears up to take their kayaks or bicycles out. They -can do this because they’re part-time baristas, freelance IT consultants, club bookers or chefs -at all-organic small-plates restaurants. They’re as likely to drink PBR as IPA, but they’d rather -die than sip truck-stop coffee or down a quarter pounder with cheese. They always recycle. -They’ve been on skid road, and they know Alki has nothing to do with the drinking habits -of barflies in Pioneer Square. If you ask them, they’ll tell you it really doesn’t rain that much -around here. +diverse everyday scenarios. Finally, our history questions cover a wide range of time periods and +geographical locations, including prehistory and other advanced subjects. -They might also tell you to watch the ‘flying fish’ at Pike Place Market early in the morning, -visit the Olympic Sculpture Garden around sunset and take an evening ferry ride to Bainbridge -Island. Most likely they’ll say not to bother with the Space Needle unless the weather is bril- -liant. If you’re game for a bar crawl, they might take you to Georgetown, the neighborhood for -drinkers in the know. If you’re arty, it might be an opening at an edgy Belltown gallery instead. -Or you could just get hopped up on caffeine and go thrifting all day. Despite what you hear, -you may be surprised at how little anyone minds the fickle weather. +3.2 SOCIAL SCIENCE -2 +Social science includes branches of knowledge that examine human behavior and society. Subject +areas include economics, sociology, politics, geography, psychology, and so on. See Figure 3 +for example questions. Our economics questions include microeconomics, macroeconomics, and +econometrics, and cover different types of problems, including questions that require a mixture of +world knowledge, qualitative reasoning, or quantitative reasoning. We also include important but +more esoteric topics such as security studies in order to test the boundaries of what is experienced and +learned during pretraining. Social science also includes psychology, a field that may be especially +important for attaining a nuanced understanding of humans. - SEATTLE LIFE - -Seattle is the largest metro area in the -Pacific Northwest, with a population of -about 3.3 million (580,000 in the city -proper). It’s a bookish, well-educated -place, but also a dynamic and inventive -urban center – most of the city’s recent -economic growth has been fueled by tech- -nology: both high-tech and the old-school, engine-parts variety. The Port of Seattle is the -second-largest handler of container cargo in the United States. - -‘Seattle sits on the verge -of major structural and -demographic change’ - -Two of its most successful citizens also happen to be Seattle’s two largest public figures, Bill -Gates and Paul Allen. And they symbolize a certain aspect of the city’s contradictory attitude -toward its own success. Both undeniably ambitious and indisputably successful, Gates and -Allen are seen simultaneously as points of civic pride and shameless capitalists who are totally -alien to the prevailing Seattle culture. - -And speaking of culture, you’ll be forgiven for thinking Seattle’s is mostly white – at least if -you hang out in hipster coffee shops or go to many indie-rock shows – but Seattle has a rich -multicultural history and is home to Native American, African American, Asian American and -growing Ethiopian American populations. The city ranks 23rd of the 30 largest US cities in -African-American population, at 8.4% – higher than every West Coast city except Los Angeles. -It also has the largest concentration of Native Americans in the Pacific Northwest. The largest -group in urban Seattle is the Duwamish people, the Native Americans who originally lived on -the shores of Elliott Bay. - -It’s been a neighborhood-centric city for most of its history, but Seattle now sits on the verge -of major structural and demographic change. The area expects 40% population growth in the -next two decades, and the regional transportation board has already begun reshaping the city’s -landscape with new mass-transit plans. It’s an exciting time to visit an always exciting place. - -Marching girls parade through Seattle's International District (p59) - -LAWRENCE WORCESTER +Figure 3: Examples from the Microeconomics and Security Studies social science tasks. -3 +3.3 SCIENCE, TECHNOLOGY, ENGINEERING, AND MATHEMATICS (STEM) - H I G H LI G HT S +STEM subjects include physics, computer science, mathematics, and more. Two examples are shown +in Figure 4. Conceptual physics tests understanding of simple physics principles and may be thought +of as a harder version of the physical commonsense benchmark Physical IQA (Bisk et al., 2019). We +also test mathematical problem solving ability at various levels of difficulty, from the elementary to +the college level. College mathematics questions, like those found on the GRE mathematics subject +test, often require chains of reasoning and abstract knowledge. To encode mathematics expressions, +we use LaTeX or symbols such as * and ˆ for multiplication and exponentiation respectively. STEM +subjects require knowledge of empirical methods, fluid intelligence, and procedural knowledge. - +3.4 OTHER - +There is a long tail of subjects that either do not neatly fit into any of the three preceding categories or +for which there are not thousands of freely available questions. We put these subjects into Other. This 4 -D OW N TOW N & FI R ST H I LL +As Seller, an encyclopedia salesman, approached the grounds on which Hermit's house was situated,he saw a sign that said, "No salesmen. Trespassers will be prosecuted. Proceed at your own risk."Although Seller had not been invited to enter, he ignored the sign and drove up the driveway towardthe house. As he rounded a curve, a powerful explosive charge buried in the driveway exploded, andSeller was injured. Can Seller recover damages from Hermit for his injuries?(A) Yes, unless Hermit, when he planted the charge, intended only to deter, not harm, intruders.(B) Yes, if Hermit was responsible for the explosive charge under the driveway.(C) No, because Seller ignored the sign, which warned him against proceeding further.(D) No, if Hermit reasonably feared that intruders would come and harm him or his family.Professional LawOne of the reasons that the government discourages and regulates monopolies is that(A) producer surplus is lost and consumer surplus is gained.(B) monopoly prices ensure productive efficiency but cost society allocative efficiency.(C) monopoly firms do not engage in significant research and development.(D) consumer surplus is lost with higher prices and lower levels of output.MicroeconomicsSecurity StudiesWhy is HIV/AIDS considered to be a non-traditional security issue?(A) HIV/AIDS is a newly emergent disease that did not exist until the later 20th century.(B) Ill health can threaten the state indirectly, but can also be a threat to other referent objects such as the economy.(C) The overwhelming scale of HIV/AIDS requires a new approach in order to theorize its impact on security.(D) None of the above - HIV/AIDS is better defined as a traditional security issue. Figure 4: Examples from the Conceptual Physics and College Mathematics STEM tasks. + +section includes the Professional Medicine task, which has difficult questions that require humans +many years of study to master. An example is depicted in Figure 5. This section also contains +business topics like finance, accounting, and marketing, as well as knowledge of global facts. The +latter includes statistics about poverty in different countries over time, which may be necessary for +having an accurate model of the world internationally. + +Figure 5: A question from Professional Medicine which is a simulated question from the United +States Medical Licensing Examination. +4 EXPERIMENTS + +4.1 SETUP + +Assessment and Models. To measure performance on our multitask benchmark, we compute the +classification accuracy across all examples and tasks. We evaluate GPT-3 (Brown et al., 2020) and +UnifiedQA (Khashabi et al., 2020). For GPT-3 we use the OpenAI API, which provides access +to four model variants, “Ada,” “Babbage,” “Curie,” and “Davinci,” which we refer to as “Small” +(2.7 billion parameters), “Medium” (6.7 billion), “Large” (13 billion) and “X-Large” (175 billion). +UnifiedQA uses the T5 (Raffel et al., 2019) text-to-text backbone and is fine-tuned on previously +proposed question answering datasets (Lai et al., 2017), where the prediction is the class with the + +Model +Random Baseline +T5 +UnifiedQA +GPT-3 Small +GPT-3 Medium +GPT-3 Large +GPT-3 X-Large + +Humanities +25.0 +24.9 +38.0 +24.4 +26.1 +27.1 +40.8 + +Social Science +25.0 +25.2 +41.5 +30.9 +21.6 +25.6 +50.4 + +STEM +25.0 +24.8 +32.2 +26.0 +25.6 +24.3 +36.7 + +Other +25.0 +24.4 +42.1 +24.1 +25.5 +26.5 +48.8 + +Average +25.0 +24.8 +38.5 +25.9 +24.9 +26.0 +43.9 + +Table 1: Average weighted accuracy for each model on all four broad disciplines. All values are +percentages. Models proposed in the past few months (UnifiedQA, GPT-3) can move several percent +points beyond random chance on this test. -Downtown Seattle is the functional business heart of the city, with a jungle of office -buildings and an electric weekday buzz. It’s also home to many of the city’s main -hotels, as well as the Seattle Art Museum. Head to First Hill to see the stately, historic -Sorrento Hotel. +5 -LAWRENCE WORCESTER +When you drop a ball from rest it accelerates downward at 9.8 m/s². If you instead throw itdownward assuming no air resistance its acceleration immediately after leaving your hand is(A) 9.8 m/s²(B) more than 9.8 m/s²(C) less than 9.8 m/s²(D) Cannot say unless the speed of throw is given.ConceptualPhysicsCollegeMathematicsIn the complex z-plane, the set of points satisfying the equation z² = |z|² is a(A) pair of points(B) circle(C) half-line(D) lineA 33-year-old man undergoes a radical thyroidectomy for thyroid cancer. During the operation,moderate hemorrhaging requires ligation of several vessels in the left side of the neck.Postoperatively, serum studies show a calcium concentration of 7.5 mg/dL, albumin concentrationof 4 g/dL, and parathyroid hormone concentration of 200 pg/mL. Damage to which of the followingvessels caused the findings in this patient?(A) Branch of the costocervical trunk(B) Branch of the external carotid artery(C) Branch of the thyrocervical trunk(D) Tributary of the internal jugular veinProfessional Medicine highest token overlap with UnifiedQA’s text output. Since UnifiedQA is fine-tuned on other datasets, +we evaluate it without any further tuning to assess its transfer accuracy. + +Few-Shot Prompt. We feed GPT-3 prompts +like that shown in Figure 1a. We begin each +prompt with “The following are multiple choice +questions (with answers) about [subject].” For +zero-shot evaluation, we append the question to +the prompt. For few-shot evaluation, we add up +to 5 demonstration examples with answers to +the prompt before appending the question. All +prompts end with “Answer: ”. The model then +produces probabilities for the tokens “A,” “B,” +“C,” and “D,” and we treat the highest probability +option as the prediction. For consistent evalua- +tion, we create a dev set with 5 fixed few-shot +examples for each subject. + +4.2 RESULTS + +Model Size and Accuracy. We compare the +few-shot accuracy of each GPT-3 size in Table 1. +We find that the three smaller GPT-3 models +have near random accuracy (around 25%). We +also assess the 11 billion parameter T5 model +in a few-shot setting and confirmed that it like- +wise has random chance accuracy. In contrast, +we find that the X-Large 175 billion parame- +ter GPT-3 model performs substantially better +than random, with an accuracy of 43.9%. We +also find qualitatively similar results in the zero- +shot setting. While the smaller models have +around 25% zero-shot accuracy, Figure 8a in +Appendix A shows that the largest GPT-3 model +has a much higher zero-shot accuracy of about +37.7%. In Figure 1b we show that non-random +accuracy on the multitask test emerged with re- +cent large few-shot models compared to datasets +that assess commonsense and linguistic under- +standing. + +To test the importance of model size for other +methods, we also evaluate UnifiedQA models. +UnifiedQA has the advantage of being fine- +tuned on other question answering datasets, and +we assess it by evaluating its transfer perfor- +mance without any additional fine-tuning. The +largest UnifiedQA model we test has 3 billion +parameters, which is slightly larger than GPT-3 +Small. Nevertheless, we show in Table 1 that it +attains 38.5% accuracy. This is worse than few- +shot GPT-3 X-Large accuracy but higher than +zero-shot GPT-3 X-Large, despite UnifiedQA having two orders of magnitude fewer parameters. We +also find that even the smallest UnifiedQA variant, with just 60 million parameters, has approximately +30% accuracy. These results suggest that while model size is a key component for achieving strong +performance, it is not the only important factor. + +Figure 6: GPT-3 few shot accuracies for all of the +57 tasks. All task accuracies are markedly below +expert-level performance. + +Comparing Disciplines. Using our test, we discover that GPT-3 has lopsided performance and +several substantial knowledge gaps. Figure 6 shows the few-shot accuracy of GPT-3 for all 57 tasks. - Sorrento Hotel -Dabble in decadence at the Sorrento Hotel (p51) - Seattle Public Library -Find peace and prose at this stylish library (p49) - Seattle Art Museum -Gawp at Jonathan Borofsky’s 1992 sculpture -Hammering Man, which dominates the museum -entrance (p48) +6 - +020406080100Accuracy (%)College ChemistryMoral ScenariosCollege PhysicsHigh School PhysicsHigh School MathematicsFormal LogicElementary MathematicsAbstract AlgebraHigh School StatisticsMachine LearningEconometricsHigh School ChemistryProfessional AccountingProfessional LawCollege MathematicsProfessional MedicineConceptual PhysicsGlobal FactsHigh School Comp SciMedical GeneticsHigh School MacroeconomicsHigh School MicroeconomicsMoral DisputesProfessional PsychologyCollege BiologyVirologyCollege Comp SciBusiness EthicsNutritionCollege MedicineAnatomyClinical KnowledgeLogical FallaciesHigh School BiologyPublic RelationsAstronomyElectrical EngineeringHuman AgingPhilosophySecurity StudiesPrehistoryHigh School US HistorySociologyHigh School European HistoryHuman SexualityJurisprudenceWorld ReligionsInternational LawHigh School World HistoryManagementComputer SecurityHigh School GeographyHigh School Gov't and PoliticsMarketingMiscellaneousHigh School PsychologyUS Foreign PolicyRandom Chance (a) GPT-3’s completion for two prompts testing knowl- +edge of the order of operations. The blue underlined +bold text is the autocompleted response from GPT-3. +While it is has descriptive knowledge and knows about +of the order of operations, it does not know how to apply +its knowledge and does not obey operator precedence. (b) GPT-3’s average confidence is a poor estimator of + +its accuracy and can be off by up to 24%. + +It shows the GPT-3 is below expert-level performance for all tasks, with accuracy ranging from 69% +for US Foreign Policy to 26% for College Chemistry. + +Overall, GPT-3 does poorly on highly procedural problems. Figure 6 shows that calculation-heavy +STEM subjects tend to have low accuracy compared to verbal subjects. In fact, 9 out of the 10 +lowest-accuracy tasks are STEM subjects that emphasize mathematics or calculations. We speculate +that is in part because GPT-3 acquires declarative knowledge more readily than procedural knowlege. +For example, many questions in Elementary Mathematics require applying the order of operations +for arithmetic, which is described by the acronym PEMDAS (Parentheses Exponents Multiplication +Division Addition Subtraction). In Figure 7a, we confirm that GPT-3 is aware of the acronymn +PEMDAS. However, it does not consistently apply PEMDAS to actual problems. On the other hand, +procedural understanding is not its only weak point. We find that some verbal tasks such as Moral +Scenarios (Hendrycks et al., 2020) and Professional Law also have especially low accuracy. + +Our test also shows that GPT-3 acquires knowledge quite unlike humans. For example, GPT-3 learns +about topics in a pedagogically unusual order. GPT-3 does better on College Medicine (47.4%) +and College Mathematics (35.0%) than calculation-heavy Elementary Mathematics (29.9%). GPT-3 +demonstrates unusual breadth, but it does not master a single subject. In this way, our test shows that +GPT-3 has many knowledge blindspots and has capabilities that are lopsided. + +Calibration. We should not trust a model’s prediction unless the model is calibrated, meaning +that its confidence is a good estimate of the actual probability the prediction is correct. However, +large neural networks are often miscalibrated (Guo et al., 2017), especially under distribution shift +(Ovadia et al., 2019). We evaluate the calibration of GPT-3 by testing how well its average confidence +estimates its actual accuracy for each subject. We show the results in Figure 7b, which demonstrates +that GPT-3 is uncalibrated. In fact, its confidence is only weakly related to its actual accuracy in +the zero-shot setting, with the difference between its accuracy and confidence reaching up to 24% +for some subjects. Another calibration measure is the Root Mean Squared (RMS) calibration error +(Hendrycks et al., 2019a; Kumar et al., 2019). Many tasks have miscalibrated predictions, such +as Elementary Mathematics which has a zero-shot RMS calibration error of 19.4%. These results +suggest that model calibration has wide room for improvement. + +5 DISCUSSION + +Multimodal Understanding. While text is capable of conveying an enormous number of concepts +about the world, many important concepts are conveyed mainly through other modalities, such as +images, audio, and physical interaction (Bisk et al., 2020). Existing large-scale NLP models, such as +GPT-3, do not incorporate multimodal information, so we design our benchmark to capture a diverse +array of tasks in a text-only format. However, as models gain the ability to process multimodal inputs, +benchmarks should be designed to reflect this change. One such benchmark could be a “Turk Test,” +consisting of Amazon Mechanical Turk Human Intelligence Tasks. These are well-defined tasks that +require models to interact with flexible formats and demonstrate multimodal understanding. + +The Internet as a Training Set. A major distinction between our benchmark and previous multitask +NLP benchmarks is that we do not require large training sets. Instead, we assume that models have -RICHARD CUMMINS +7 -RICHARD CUMMINS +Declarative vs. Procedural KnowledgePrompt and Completion:The order of operations or PEMDAS isParentheses Exponents MultiplicationDivision Addition SubtractionPrompt and Completion:(1 + 1) × 2 = 3␣2030405060Confidence (%)2030405060Accuracy (%)GPT-3 Zero-Shot CalibrationFormal LogicMarketing acquired the requisite knowledge from reading vast quantities of diverse text from the Internet. This +process is typically called pretraining, but it can be thought of as training in its own right, where the +downstream evaluation is demonstrating whatever knowledge we would expect a human to pick up +from reading the same text. + +This motivates us to propose a methodological change so that models are trained more like how +humans learn. While most previous machine learning benchmarks have models learn from a large +question bank, humans primarily learn new subjects by reading books and listening to others talk +about the topic. For specialized subjects such as Professional Law, massive legal corpora are available, +such as the 164-volume legal encyclopedia Corpus Juris Secundum, but there are fewer than 5,000 +multistate bar exam questions available. Learning the entire law exclusively through a small number +of practice tests is implausible, so future models must learn more during pretraining. + +For this reason we assess pretrained models in a zero-shot or few-shot setting and we provide a dev, +val, and test set for each task. The dev set is used for few-shot prompts, the val set could be used for +hyperparameter tuning, and the test set is used to compute the final accuracy. Importantly, the format +of our evaluation is not identical to the format in which information is acquired during pretraining. +This has the benefit of obviating concerns about spurious training set annotation artifacts (Geirhos +et al., 2020; Hendrycks et al., 2019b) and is in stark contrast to the previous paradigm of identically +distributed training and test sets. This change also enables collecting a much more extensive and +diverse set of tasks for evaluation. We anticipate our methodology becoming more widespread as +models improve at extracting information from diverse online sources. + +Model Limitations. We find that current large-scale Transformers have wide room for improvement. +They are notably poor at modeling human (dis)approval, as evident by the low performance on the +Professional Law and Moral Scenarios tasks. For future systems to be aligned with human values, high +performance on these tasks is crucial (Hendrycks et al., 2020), so future research should especially +aim to increase accuracy on these tasks. Models also have difficulty performing calculations, so much +so that they exhibit poor performance on Elementary Mathematics and many other STEM subjects +with “plug and chug” problems. Additionally, they do not match expert-level performance on any +subject, so for all subjects it is subhuman. On average, models are only now starting to move beyond +random-chance accuracy levels. + +Addressing these shortcomings may be challenging. To illustrate this, we attempted to create a better +Professional Law model by pretraining on specialized data but achieved only limited success. We +collected approximately 2,000 additional Professional Law training examples. After fine-tuning a +RoBERTa-base model (Liu et al., 2019) using this custom training set, our model attained 32.8% test +accuracy. To test the impact of additional specialized training data, we also had RoBERTa continue +pretraining on approximately 1.6 million legal case summaries using Harvard’s Law Library case law +corpus case.law, but after fine-tuning it only attained 36.1% accuracy. This suggests that while +additional pretraining on relevant high quality text can help, it may not be enough to substantially +increase the performance of current models. + +It is unclear whether simply scaling up existing language models will solve the test. Current +understanding indicates that a 10× increase in model size must be accompanied by an approximate +5× increase in data (Kaplan et al., 2020). Aside from the tremendous expense in creating multi-trillion +parameter language models, data may also become a bottleneck, as there is far less written about +esoteric branches of knowledge than about everyday text. + +6 CONCLUSION + +We introduced a new test that measures how well text models can learn and apply knowledge +encountered during pretraining. By covering 57 subjects at varying levels of difficulty, the test +assesses language understanding in greater breadth and depth than previous benchmarks. We found +that it has recently become possible for models to make meaningful progress on the test, but that +state-of-the-art models have lopsided performance and still do not excel at any individual task. +We also showed that current models are uncalibrated and have difficulty with tasks that require +calculations. Worryingly, models also perform especially poorly on socially relevant subjects +including morality and law. Our expansive test can help researchers pinpoint important shortcomings +of models, making it easier to gain a clearer picture of state-of-the-art capabilities. -  +8 -LAWRENCE WORCESTER + 6.1 ACKNOWLEDGEMENTS -PIONEER SQUARE & INTERNATIONAL DISTRICT +We would like to thank the following for their helpful comments: Jan Leike, David Krueger, Alex +Tamkin, Girish Sastry, and Henry Zhu. DH is supported by the NSF GRFP Fellowship and an Open +Philanthropy Project Fellowship. This research was also supported by the NSF Frontier Award +1804794. -Pioneer Square is the birthplace of Seattle, and its redbrick buildings and plazas give -it a historical, film-set feel. It’s also popular for its restaurants, bars and independent -shops, which are well worth a visit. The International District is home to Seattle’s -Chinatown and the Wing Luke Museum of the Asian Pacific American Experience. +REFERENCES - Wing Luke Museum of the -Asian Pacific American -Experience -Absorb yourself in Pan-Asian history (p59) - Pioneer Square Park -Step back in time to Seattle’s early days (p57) - Zeitgeist -Perk up on coffee in arty surrounds (p141) +M. G. Bellemare, Y. Naddaf, J. Veness, and M. Bowling. The arcade learning environment: An +evaluation platform for general agents (extended abstract). J. Artif. Intell. Res., 47:253–279, 2013. - +Y. Bisk, R. Zellers, R. L. Bras, J. Gao, and Y. Choi. Piqa: Reasoning about physical commonsense in -ANN CECIL +natural language, 2019. - +Y. Bisk, A. Holtzman, J. Thomason, J. Andreas, Y. Bengio, J. Chai, M. Lapata, A. Lazaridou, J. May, -CHUCK PEFLEY/ALAMY +A. Nisnevich, N. Pinto, and J. Turian. Experience grounds language, 2020. -5 +T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, +G. Sastry, A. Askell, S. Agarwal, A. Herbert-Voss, G. Krueger, T. Henighan, R. Child, A. Ramesh, +D. M. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray, B. Chess, +J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, and D. Amodei. Language models +are few-shot learners, 2020. -  +P. Clark, I. Cowhey, O. Etzioni, T. Khot, A. Sabharwal, C. Schoenick, and O. Tafjord. Think you have +solved question answering? try arc, the ai2 reasoning challenge. ArXiv, abs/1803.05457, 2018. - +P. Clark, O. Etzioni, D. Khashabi, T. Khot, B. D. Mishra, K. Richardson, A. Sabharwal, C. Schoenick, +O. Tafjord, N. Tandon, S. Bhakthavatsalam, D. Groeneveld, M. Guerquin, and M. Schmitz. From ’f’ +to ’a’ on the n.y. regents science exams: An overview of the aristo project. ArXiv, abs/1909.01958, +2019. -6 +J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. Bert: Pre-training of deep bidirectional -PI K E PL ACE M AR K E T & WATE R FRO NT +transformers for language understanding. ArXiv, abs/1810.04805, 2019. -One of Seattle’s best-loved attractions, Pike Place Market is a labyrinthine fish and -produce market that oozes theatricality. Find yourself entertained amid ‘flying fish,’ -buskers, tourists and locals before stepping out along the Waterfront, where scenic -views of Puget Sound and the Olympic Mountains await. +R. Geirhos, J.-H. Jacobsen, C. Michaelis, R. Zemel, W. Brendel, M. Bethge, and F. A. Wichmann. -LAWRENCE WORCESTER +Shortcut learning in deep neural networks, 2020. - Pike Place Market -Watch the world go by at this loud, lovable market (p62) - Seattle Aquarium -Spy the underwater world of Puget Sound (p66) - Waterfront -Catch the flavour of a major seaport (p65) +C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger. On calibration of modern neural networks. ICML, -JOHN ELK III +2017. - +D. Hendrycks, M. Mazeika, and T. Dietterich. Deep anomaly detection with outlier exposure. ICLR, -LAWRENCE WORCESTER +2019a. - B E LLTOW N +D. Hendrycks, K. Zhao, S. Basart, J. Steinhardt, and D. Song. Natural adversarial examples. ArXiv, -The alternative scene in Belltown may have mellowed slightly, but this neighborhood -can still be classed as cool. Watch a band at the Crocodile Café before taking an -easy stroll to other offbeat bars in the area. Belltown is also home to the impressive -Olympic Sculpture Park. +abs/1907.07174, 2019b. - +D. Hendrycks, C. Burns, S. Basart, A. Critch, J. Li, D. Song, and J. Steinhardt. Aligning ai with -RICHARD CUMMINS +shared human values, 2020. - Olympic Sculpture Park -Feel small next to Richard Serra’s 2004 -sculpture, Wake (p67) - Ace Hotel -Be fascinated by bedtime reading at this hipster -hotel (p179) - Shorty’s -Play pinball and drink beer (p142) +L. Huang, R. L. Bras, C. Bhagavatula, and Y. Choi. Cosmos qa: Machine reading comprehension - +with contextual commonsense reasoning, 2019. -LAWRENCE WORCESTER +J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, - +and D. Amodei. Scaling laws for neural language models, 2020. -LAWRENCE WORCESTER +D. Khashabi, T. Khot, A. Sabharwal, O. Tafjord, P. Clark, and H. Hajishirzi. Unifiedqa: Crossing -7 +format boundaries with a single qa system, 2020. -  +T. Khot, P. Clark, M. Guerquin, P. Jansen, and A. Sabharwal. Qasc: A dataset for question answering - +via sentence composition, 2019. - +A. Kumar, P. Liang, and T. Ma. Verified uncertainty calibration, 2019. -8 +9 -S E AT TLE CE NTE R & Q U E E N AN N E + G. Lai, Q. Xie, H. Liu, Y. Yang, and E. Hovy. Race: Large-scale reading comprehension dataset from -Seattle Center is a 1960s vision of the future, making it an endearingly retro place to -visit. As well as the Space Needle, it contains numerous museums and attractions, -including the Experience Music Project. Rising up on the hill above Seattle Center is -the dignified Queen Anne neighborhood. +examinations, 2017. -RICHARD CUMMINS +Z. Lan, M. Chen, S. Goodman, K. Gimpel, P. Sharma, and R. Soricut. Albert: A lite bert for - International -Fountain -Get wet in sci-fi International -Fountain (p73) - Queen Anne -Pose by perfectly manicured lawns -(p74) - Experience Music -Project -Be wowed by Frank Gehry’s 2000 -museum (p72) - Space Needle -Zoom to the top for incredible -views (p71) +self-supervised learning of language representations. ArXiv, abs/1909.11942, 2020. -ANN CECIL +Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and +V. Stoyanov. Roberta: A robustly optimized bert pretraining approach. ArXiv, abs/1907.11692, +2019. - +T. Mihaylov, P. Clark, T. Khot, and A. Sabharwal. Can a suit of armor conduct electricity? a new -LAWRENCE WORCESTER +dataset for open book question answering. In EMNLP, 2018. -RICHARD CUMMINS +Y. Ovadia, E. Fertig, J. Ren, Z. Nado, D. Sculley, S. Nowozin, J. V. Dillon, B. Lakshminarayanan, +and J. Snoek. Can you trust your model’s uncertainty? Evaluating predictive uncertainty under +dataset shift. NeurIPS, 2019. - LAKE UNION, CAPITOL HILL & VOLUNTEER PARK +F. Petroni, T. Rocktäschel, P. Lewis, A. Bakhtin, Y. Wu, A. H. Miller, and S. Riedel. Language -Lake Union hums with the sound of seaplanes, but you can have fun skirting the -edge on a bike or boat. Nearby Capitol Hill is an artsy and diverse neighborhood, -where you’ll find cafes, galleries and street theater. If you need a break from kayaks -and counterculture, stop at the Volunteer Park Conservatory. +models as knowledge bases?, 2019. - Wall of Sound -Flip through racks of music (p114) - Jimi Hendrix Statue -Check out Daryl Smith’s 1997 statue of the -guitar legend (p80) - Lake Union -Paddle or cycle around this Y-shaped lake (p76) - Volunteer Park Conservatory -Brush past palms in Seattle’s tropical corner -(p80) +C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, and P. J. Liu. - +Exploring the limits of transfer learning with a unified text-to-text transformer, 2019. - +M. Richardson, C. J. Burges, and E. Renshaw. MCTest: A challenge dataset for the open-domain +machine comprehension of text. In Proceedings of the 2013 Conference on Empirical Methods in +Natural Language Processing, pages 193–203, Seattle, Washington, USA, Oct. 2013. Association +for Computational Linguistics. -LAWRENCE WORCESTER +A. B. Sai, A. K. Mohankumar, and M. M. Khapra. A survey of evaluation metrics used for nlg -LAWRENCE WORCESTER +systems. 2020. - +A. Turing. Computing machinery and intelligence. 1950. -ANN CECIL +A. Wang, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman. Glue: A multi-task benchmark - +and analysis platform for natural language understanding, 2018. -JOHN ELK III +A. Wang, Y. Pruksachatkun, N. Nangia, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman. +Superglue: A stickier benchmark for general-purpose language understanding systems, 2019. -9 +R. Zellers, A. Holtzman, Y. Bisk, A. Farhadi, and Y. Choi. Hellaswag: Can a machine really finish -  +your sentence?, 2019. - +R. Zellers, A. Holtzman, E. Clark, L. Qin, A. Farhadi, and Y. Choi. Evaluating machines by their + +real-world language use, 2020. 10 -FR E M O NT & WALLI N G FO R D + A ADDITIONAL ANALYSIS -Fun-loving Fremont has coffee shops, pubs and some of the oddest public art that -you’ll ever see, making a visit to this neighborhood a must. Pleasant Wallingford, on -the route of the Burke-Gilman cycle trail, is home to some noteworthy restaurants -and Gas Works Park. +(a) As the number of few-shot instruction examples in- +creases, the accuracy monotonically increases. Notably, +zero-shot performance is only somewhat lower than +5-shot accuracy. -LAWRENCE WORCESTER +(b) While models are more calibrated in a few-shot set- +ting than a zero-shot setting, they are still miscalibrated, +with gap between accuracy and confidence reaching up +to 14%. Here the correlation between confidence and +accuracy is r = 0.81, compared to r = 0.63 in the +zero-shot setting. - Gas Works Park -Find yourself drawn inexplicably to a strange mass of pipes (p91) - Fremont Troll -Quake before the 1990 statue, the Fremont Troll (p88) - Statue of Lenin -Consider the seven tons of Slovakian scrap that make up Emil Venkov's -1988 statue (p88) +A.1 ERROR ANALYSIS - +We qualitatively analyze when GPT-3 makes high confidence mistakes. We find that while many of +these mistakes were clearly wrong, many were mistakes that a human might make. For example, +one question it got wrong was “How many chromosomes do all human somatic cells contain?” The +correct answer is 46, while few-shot GPT-3 predicted 23 with confidence 97.5%. This answer would +have been correct if the question asked about the number of pairs of chromosomes. Similarly, many +of its other high confidence mistakes were also correct answers to slightly different questions. -LAWRENCE WORCESTER +A.2 FINE-TUNING -LAWRENCE WORCESTER +Rather than testing models in a few-shot setting, we now determine the impact of fine-tuning with +thousands of examples. For this section, we fine-tune RoBERTa-base (Liu et al., 2019) and ALBERT- +xxlarge (Lan et al., 2020), which have fewer than 250 million parameters and have at least 10× +fewer parameters than GPT-3 Small. Models are fine-tuned to predict one of four classes and are +fine-tuned using the dev+val set, and we test on the test set. We observe that these smaller models +can attain better-than-random accuracy. RoBERTa-base attains an overall accuracy of 27.9%, with +27.9% accuracy for the humanities, 28.8% for social sciences, 27.0% for STEM, and 27.7% for other. +ALBERT-xxlarge attains an accuracy of 27.1%, with 27.2% accuracy for the humanities, 25.7% for +the social sciences, 27.7% for STEM, and 27.9% for other. Consequently smaller models that are not +designed for QA are able to exceed random chance, though barely. - BALL AR D & D I SCOV E RY PAR K +B TEST DETAILS -Ballard is a Scandinavian-influenced neighborhood that lies just over Ballard Bridge. -It seems a world away from urban Seattle and is known for its down-to-earth bars -and live-music venues, as well as the interesting Fish Ladder. Neighboring Discovery -Park is 534 acres of urban wilderness. +B.1 TASK DESCRIPTIONS AND EXAMPLES - +We list all tasks and the topics they test in Table 2. We also provide an example for each task starting +with Figure 10. -DANITA DELIMONT/ALAMY +B.2 EXACT QUESTION AND ANSWER CONTAMINATION - Ballard Fish Ladder -Spot salmon on their way to spawn (p97) - Nordic Heritage Museum -Learn more about Ballard's heritage (p95) - Discovery Park -Walk the trails and escape the city (p98) +Since language models train on vast text corpora, there is some chance that they have seen the exact +question and answer during pretraining. If they memorized the exact question and answer, then +they would attain higher accuracy than their true ability. Likewise, a question’s entropy would be - +11 -RON BUSKIRK/ALAMY +0-Shot1-Shot2-Shot3-Shot4-Shot5-ShotNumber of Examples in Context3035404550Accuracy (%)GPT-3 Multitask Accuracy vs.Number of Examples in Context203040506070Confidence (%)203040506070Accuracy (%)GPT-3 Few-Shot Calibration especially low if it were memorized. Memorized questions and answers should have low entropy and +high accuracy. However, in Figure 9, we see that accuracy and question entropy are not positively +correlated, suggesting that the test’s low-entropy questions do not correspond to memorized (and +thereby correctly predicted) answers. This suggests that our exact questions were not memorized. +However, during pretraining models encountered text related to our questions through processing +Wikipedia. We also note that most of our questions came from PDFs or websites where questions and +answers are on separate pages. - +See Brown et al. (2020) for a previous discussion of contamination showing that the phenomena +hardly affects performance. To reduce the probability that future models encounter exact questions +during test-time, we provide a list of question sources on our github. -RICHARD CUMMINS +Figure 9: The average log probability of the question (without answer) is not strongly positively +correlated with accuracy, all else equal. Each point corresponds to a task. Higher log probability +indicates higher compression, and especially high log probability suggests memorization. In the zero- +shot question prompt, the correlation between average log probability and accuracy is r = −0.43, +and for the few-shot setting the correlation is r = −0.56. -G R E E N L AK E & PH I N N E Y R I D G E +12 -If you like zoos, parks, rose gardens and low-key neighborhoods, then you’ll love -Green Lake and Phinney Ridge. Green Lake Park is a perfect place for sunbathing -in summer or walking in fall, and Woodland Park Zoo is one of Seattle’s greatest -tourist attractions. +2.82.62.42.22.01.8Log Probability Per Token2030405060Accuracy (%)GPT-3 Zero-ShotPrompt Compression and Accuracy2.42.22.01.81.61.41.2Log Probability Per Token203040506070Accuracy (%)GPT-3 Few-ShotPrompt Compression and Accuracy Task +Abstract Algebra +Anatomy +Astronomy +Business Ethics +Clinical Knowledge +College Biology +College Chemistry +College Computer Science +College Mathematics +College Medicine +College Physics +Computer Security +Conceptual Physics +Econometrics +Electrical Engineering +Elementary Mathematics +Formal Logic +Global Facts +High School Biology +High School Chemistry +High School Computer Science +High School European History +High School Geography +High School Gov’t and Politics +High School Macroeconomics +High School Mathematics +High School Microeconomics +High School Physics +High School Psychology +High School Statistics +High School US History +High School World History +Human Aging +Human Sexuality +International Law +Jurisprudence +Logical Fallacies +Machine Learning +Management +Marketing +Medical Genetics +Miscellaneous +Moral Disputes +Moral Scenarios +Nutrition +Philosophy +Prehistory +Professional Accounting +Professional Law +Professional Medicine +Professional Psychology +Public Relations +Security Studies +Sociology +US Foreign Policy +Virology +World Religions + +Tested Concepts +Groups, rings, fields, vector spaces, ... +Central nervous system, circulatory system, ... +Solar system, galaxies, asteroids, ... +Corporate responsibility, stakeholders, regulation, ... +Spot diagnosis, joints, abdominal examination, ... +Cellular structure, molecular biology, ecology, ... +Analytical, organic, inorganic, physical, ... +Algorithms, systems, graphs, recursion, ... +Differential equations, real analysis, combinatorics, ... +Introductory biochemistry, sociology, reasoning, ... +Electromagnetism, thermodynamics, special relativity, ... +Cryptography, malware, side channels, fuzzing, ... +Newton’s laws, rotational motion, gravity, sound, ... +Volatility, long-run relationships, forecasting, ... +Circuits, power systems, electrical drives, ... +Word problems, multiplication, remainders, rounding, ... +Propositions, predicate logic, first-order logic, ... +Extreme poverty, literacy rates, life expectancy, ... +Natural selection, heredity, cell cycle, Krebs cycle, ... +Chemical reactions, ions, acids and bases, ... +Arrays, conditionals, iteration, inheritance, ... +Renaissance, reformation, industrialization, ... +Population migration, rural land-use, urban processes, ... +Branches of government, civil liberties, political ideologies, ... +Economic indicators, national income, international trade, ... +Pre-algebra, algebra, trigonometry, calculus, ... +Supply and demand, imperfect competition, market failure, ... +Kinematics, energy, torque, fluid pressure, ... +Behavior, personality, emotions, learning, ... +Random variables, sampling distributions, chi-square tests, ... +Civil War, the Great Depression, The Great Society, ... +Ottoman empire, economic imperialism, World War I, ... +Senescence, dementia, longevity, personality changes, ... +Pregnancy, sexual differentiation, sexual orientation, ... +Human rights, sovereignty, law of the sea, use of force, ... +Natural law, classical legal positivism, legal realism, ... +No true Scotsman, base rate fallacy, composition fallacy, ... +SVMs, VC dimension, deep learning architectures, ... +Organizing, communication, organizational structure, ... +Segmentation, pricing, market research, ... +Genes and cancer, common chromosome disorders, ... +Agriculture, Fermi estimation, pop culture, ... +Freedom of speech, addiction, the death penalty, ... +Detecting physical violence, stealing, externalities, ... +Metabolism, water-soluble vitamins, diabetes, ... +Skepticism, phronesis, skepticism, Singer’s Drowning Child, ... +Neanderthals, Mesoamerica, extinction, stone tools, ... +Auditing, reporting, regulation, valuation, ... +Torts, criminal law, contracts, property, evidence, ... +Diagnosis, pharmacotherapy, disease prevention, ... +Diagnosis, biology and behavior, lifespan development, ... +Media theory, crisis management, intelligence gathering, ... +Environmental security, terrorism, weapons of mass destruction, ... +Socialization, cities and community, inequality and wealth, ... +Soft power, Cold War foreign policy, isolationism, ... +Epidemiology, coronaviruses, retroviruses, herpesviruses, ... +Judaism, Christianity, Islam, Buddhism, Jainism, ... + +Table 2: Summary of all 57 tasks. + +13 + +Supercategory +STEM +STEM +STEM +Other +Other +STEM +STEM +STEM +STEM +Other +STEM +STEM +STEM +Social Sciences +STEM +STEM +Humanities +Other +STEM +STEM +STEM +Humanities +Social Sciences +Social Sciences +Social Sciences +STEM +Social Sciences +STEM +Social Sciences +STEM +Humanities +Humanities +Other +Social Sciences +Humanities +Humanities +Humanities +STEM +Other +Other +Other +Other +Humanities +Humanities +Other +Humanities +Humanities +Other +Humanities +Other +Social Sciences +Social Sciences +Social Sciences +Social Sciences +Social Sciences +Other +Humanities + + Find all c in Z3 such that Z3[x]/(x2 + c) is a field. +(A) 0 + +(D) 3 + +(C) 2 + +(B) 1 + +Figure 10: An Abstract Algebra example. + +What is the embryological origin of the hyoid bone? +(A) The first pharyngeal arch +(B) The first and second pharyngeal arches +(C) The second pharyngeal arch +(D) The second and third pharyngeal arches + +Figure 11: An Anatomy example. + +Why isn’t there a planet where the asteroid belt is located? +(A) A planet once formed here but it was broken apart by a catastrophic collision. +(B) There was not enough material in this part of the solar nebula to form a planet. +(C) There was too much rocky material to form a terrestrial planet but not enough gaseous material +to form a jovian planet. +(D) Resonance with Jupiter prevented material from collecting together to form a planet. + +Figure 12: An Astronomy example. + +Three contrasting tactics that CSO’s can engage in to meet their aims are +cally involves research and communication, +company’s operations or +(A) Non-violent direct action, Violent direct action, Indirect action, Boycott +(B) Indirect action, Instrumental action, Non-violent direct action, Information campaign +(C) Indirect action, Violent direct action, Non-violent direct-action Boycott. +(D) Non-violent direct action, Instrumental action, Indirect action, Information campaign + +which typi- +, which may involve physically attacking a + +, often involving some form of - Woodland Park Zoo -Strut your stuff at the zoo (p94) - Seattle Rose Garden -Smell the scent of 5000 roses (p94) +. - +Figure 13: A Business Ethics example. -LAWRENCE WORCESTER +How many attempts should you make to cannulate a patient before passing the job on to a senior +colleague? +(A) 4 - +(D) 1 -JOHN ELK III +(C) 2 -11 +(B) 3 -  +Figure 14: A Clinical Knowledge example. - +In a given population, 1 out of every 400 people has a cancer caused by a completely recessive +allele, b. Assuming the population is in Hardy-Weinberg equilibrium, which of the following is +the expected proportion of individuals who carry the b allele but are not expected to develop the +cancer? +(A) 1/400 - +(D) 38/400 -12 +(B) 19/400 + +(C) 20/400 + +Figure 15: A College Biology example. + +Which of the following statements about the lanthanide elements is NOT true? +(A) The most common oxidation state for the lanthanide elements is +3. +(B) Lanthanide complexes often have high coordination numbers (> 6). +(C) All of the lanthanide elements react with aqueous acid to liberate hydrogen. +(D) The atomic radii of the lanthanide elements increase across the period from La to Lu. + +Figure 16: A College Chemistry example. + +14 + + Consider a computer design in which multiple processors, each with a private cache memory, +share global memory using a single bus. This bus is the critical system resource. Each processor +can execute one instruction every 500 nanoseconds as long as memory references are satisfied +by its local cache. When a cache miss occurs, the processor is delayed for an additional 2,000 +nanoseconds. During half of this additional delay, the bus is dedicated to serving the cache miss. +During the other half, the processor cannot continue, but the bus is free to service requests from +other processors. On average, each instruction requires 2 memory references. On average, cache +misses occur on 1 percent of references. What proportion of the capacity of the bus would a single +processor consume, ignoring delays due to competition from other processors? +(D) 2/27 +(A) 1/50 + +(C) 1/25 + +(B) 1/27 + +Figure 17: A College Computer Science example. + +Let A be a real 2 × 2 matrix. Which of the following statements must be true? +I. All of the entries of A2 are nonnegative. +II. The determinant of A2 is nonnegative. +III. If A has two distinct eigenvalues, then A2 has two distinct eigenvalues. +(A) I only + +(D) II and III only + +(C) III only + +(B) II only + +Figure 18: A College Mathematics example. + +In a genetic test of a newborn, a rare genetic disorder is found that has X-linked recessive +transmission. Which of the following statements is likely true regarding the pedigree of this +disorder? +(A) All descendants on the maternal side will have the disorder. +(B) Females will be approximately twice as affected as males in this family. +(C) All daughters of an affected male will be affected. +(D) There will be equal distribution of males and females affected. + +Figure 19: A College Medicine example. + +One end of a Nichrome wire of length 2L and cross-sectional area A is attached to an end of +another Nichrome wire of length L and cross- sectional area 2A. If the free end of the longer wire +is at an electric potential of 8.0 volts, and the free end of the shorter wire is at an electric potential +of 1.0 volt, the potential at the junction of the two wires is most nearly equal to +(A) 2.4 V +(B) 3.3 V +(C) 4.5 V +(D) 5.7 V + +Figure 20: A College Physics example. + +Why is it that anti-virus scanners would not have found an exploitation of Heartbleed? +(A) It’s a vacuous question: Heartbleed only reads outside a buffer, so there is no possible exploit +(B) Anti-virus scanners tend to look for viruses and other malicious +(C) Heartbleed attacks the anti-virus scanner itself +(D) Anti-virus scanners tend to look for viruses and other malicious code, but Heartbleed +exploits steal secrets without injecting any code + +Figure 21: A Computer Security example. + +A model airplane flies slower when flying into the wind and faster with wind at its back. When +launched at right angles to the wind, a cross wind, its groundspeed compared with flying in still +air is +(A) the same + +(D) either greater or less depending on wind speed -AROUN D SE AT TLE - -RICHARD CUMMINS - -Situated in a stunning part of the Pacific Northwest, Seattle is within easy reach of -mountain ranges, the ocean, hundreds of islands and a temperate rainforest. This -makes Seattle a perfect hub for getting out and exploring the natural world. - - Mt Rainier National Park -Climb the highest peak in the Cascades (p193) - Hoh River Rain Forest -Hike in a mossy temperate rainforest (p198) - San Juan Islands -Kayak around breathtaking islands (p189) - Ferry across Puget Sound -Ride the waves and look back at Seattle’s -famous skyline (p187) - -JOHN ELK III - - - -TOM BOYDEN - -LAWRENCE WORCESTER - - TH E AUTH O R - -Becky Ohlsen - -Becky has lived in -the Pacific Northwest -for the past 15 years, -originally drawn here -from her native Colo- -rado by the promise -of great music and -the ocean. Upon -arrival she discovered -that the area also pro- -vided excellent beer, -a wide variety of pinball machines, cute skater -boys and lots of rainy days in which to moodily -read in coffee shops. She lives in Portland, -Oregon, after surviving a brief, educational, -expensive and decidedly temporary stay in -New York City, where against everyone’s bet- -ter judgment she earned a master’s degree in -journalism. In addition to covering Seattle -for Lonely Planet, she has worked as a copy -editor and freelance writer for alternative -newspapers and magazines, and occasionally -reviews movies for a local independent radio -station. When she’s not traveling for guide- -book research, Becky races vintage motorcy- -cles very, very tentatively. She’s a member of -the Sang-Froid Riding Club and a veteran of -the infamous ‘555’ Ride, and much prefers -two wheels to four. Her favorite bar in Seattle -is Shorty’s and her favorite Nirvana album -is Bleach. - -LONELY PLANET AUTHORS -Why is our travel information the best in the world? -It’s simple: our authors are passionate, dedicated -travelers. They don’t take freebies in exchange for -positive coverage so you can be sure the advice you’re -given is impartial. They travel widely to all the popu- -lar spots, and off the beaten track. They don’t research -using just the internet or phone. They discover new -places not included in any other guidebook. They per- -sonally visit thousands of hotels, restaurants, palaces, -trails, galleries, temples and more. They speak with -dozens of locals every day to make sure you get the -kind of insider knowledge only a local could tell you. -They take pride in getting all the details right, and in -telling it how it is. Think you can do it? Find out how -at lonelyplanet.com. - -BECKY’S TOP SEATTLE DAY -When I’m in Seattle I like to wake up unchar- -acteristically early and go straight to Pike Place -Market (p62). The scene here is fascinating any -time of day, but it’s best in the morning, when -most of the customers are locals getting their -shopping done. It’s a nice place for a walk- -ing breakfast, too, with options ranging from -fresh fruit to coffee and pastries to a hefty hum -bao (steamed pork bun) big enough to power -you through the whole day. After exploring -the labyrinth of weird little market shops, I’ll -browse the progressive literature at Left Bank -Books (p110) to keep myself honest and bolster -hope. Afterwards, a short stroll down toward -Elliott Bay gives me a panoramic view of the -mountains from the new Olympic Sculpture -Park (p67). - -Of course I can’t be in Seattle without -spending some time digging around the -shelves at Elliott Bay Book Company (p113), -and as long as I’m this close I might as well -treat myself to an artisan ice cream at Molly -Moon’s (p131) for lunch. - -I’ll spend the rest of the afternoon soaking -up culture at the Seattle Art Museum (p48), -abandon it for happy hour at the Nitelite -Lounge (p140), then probably squander the -evening and whatever’s left in my pockets -on pinball and beer in the back room of -Shorty’s (p142). +(B) greater + +(C) less + +Figure 22: A Conceptual Physics example. 15 - G E T TI N G STAR TE D - -Other than finding accommodations, Seattle doesn’t require a huge amount of pre-planning. -Most of the can’t-miss sights are clustered around downtown within easy walking distance of -each other. If you bring a decent map, book accommodations in advance and have some sort -of plan for transportation (whether it’s a car and an idea of where you’ll park overnight, or a -public-transport schedule), you’ll be fine. Keep in mind that the city is hillier than it looks on -maps and – if you’re driving – many streets in the city center are one-way only. - -Once you’re there, getting around is relatively easy. Driving and parking are the main an- -noyances – if you can avoid them, you’ll be happier. Luckily this is easy to do. Buses connect -to most neighborhoods of interest to visitors, although they stop running too early to be of -much use for nightlife exploration. But then, that’s why Mother Nature gave us taxi cabs (see -the Transportation chapter, p199). - -For more help with planning, visit Seattle’s Convention and Visitors Bureau (Map p50; %206-461-5888, 866- -732-2695; www.visitseattle.org; 701 Pike St; h9am-5pm). The website has a schedule of events, a few discount -coupons and a form for ordering a visitor-information packet. Also among the bureau’s services -is the Citywide Concierge Center, a free service for visitors that will book you a table, a tour or -transportation and answer questions about events going on while you’re in town. - -G -E -T -T -I - -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G - -O WHEN TO GO - -Unless you’re a skier or snowboarder – or -really interested in aquatic life – Seattle in -the winter makes for a pretty dreary holiday. -The city receives 65% of its precipitation from -November to March. But in spring the relent- -less downpour frequently lets up to provide -gorgeous days that are well worth the wait. - -April and May are best for enjoying the -fruits of all that gloomy winter rain: sud- -denly everything is supernaturally green and -covered in flowers. Summer – June, July and -August – is the high season and can usually -be relied upon for sunny days, warm weather -and lots of people saying things like ‘I thought -you said it rained all the time here.’ Fall, the -season of Bumbershoot and Oktoberfest, is -the favorite of beer lovers and music fans. - -FESTIVALS -January -CHINESE NEW YEAR -%206-382-1197; www.cidbia.org -Beginning toward the end of January or -the start of February and lasting for two -weeks, the year’s first big ethnic festival -takes place in the International District; you -can contact the Chinatown/International -District Business Improvement Association -for information. The first day of the New -Year festival is celebrated with parades, -firecrackers, fireworks and plenty of food. + Consider the following AR(1) model with the disturbances having zero mean and unit variance +yt = 0.2 + 0.4yt−1 + ut +The (unconditional) mean of y will be given by +(C) 0.5 +(A) 0.2 -16 +(D) 0.33 -February -NORTHWEST FLOWER & GARDEN -SHOW -%800-229-6311; www.gardenshow.com; -Washington State Convention & Trade Center -Usually held the second week of February, -this popular event includes lectures and -seminars, demo gardens and samples, and -children’s activities. - -March -MOISTURE FESTIVAL -%800-838-3006; www.moisturefestival.com; -Hale’s Palladium, 4301 Leary Way NW; -tickets $5-25 -With regular performances held between -March 15 and April 1, the Moisture Festival -has developed from a little-known -quirkfest into something that locals look -forward to each year. It has retained a -laid-back, casual vibe and usually features -random appearances by folks like Artis -the Spoonman, Circus Contraption and -The Bobs. - -ST PATRICK’S DAY -March 17 is the day the patron saint of -Ireland is honored by all those who have -Irish in their blood, and by those who want -to have Irish beer in their blood. Everyone -wears green (otherwise you risk getting -pinched). Irish bars such as Kells serve -green beer, and even bars that haven’t - - an Irish trinket in sight get swamped with -revelers. Catch the parade going from City -Hall to Westlake Center along 4th Ave. - -April -CHERRY BLOSSOM & JAPANESE -CULTURAL FESTIVAL -%206-723-2003; www.seattlecenter.com; -Seattle Center -Usually held in mid-April, this celebration of -Japanese heritage includes performances -of music, dance and drama. It’s part of -the Seattle Center’s series of multicultural -festivals, dubbed Festál. - -May -CINCO DE MAYO -May 5 commemorates the day on which -the Mexicans wiped out the French -Army in 1862. Now it’s the day all -Americans get to eat lots of Mexican -food and drink margaritas. Seattle -celebrates with an annual parade and -exhibitions of Hispanic artwork. - -OPENING DAY OF YACHT SEASON -%206-325-1000; www.seattleyachtclub.org -Held the first Saturday in May at various -locations on Lakes Washington and Union, -this Seattle original starts with a blessing of -the fleet. It features scull racing and a boat -parade through the canals. - -SEATTLE INTERNATIONAL -CHILDREN’S FESTIVAL -%206-684-7338; www.seattleinternational.org; -Seattle Center -Held in mid-May, this cultural extravaganza -includes a wide variety of performances -and activities for kids. - -NORTHWEST FOLKLIFE -FESTIVAL -%206-684-7300; www.nwfolklife.org; -Seattle Center -This festival – the largest of its kind -in North America – takes over Seattle -Center during Memorial Day weekend. -More than 5000 performers and artists -from over 100 countries present music, -dance, crafts, food and activities in -celebration of the rich cultural heritage -of the Pacific Northwest. - -SEATTLE INTERNATIONAL FILM -FESTIVAL -%206-324-9996; www.seattlefilm.com -Held from late May to mid-June, this -festival, the largest of its kind in the -nation, brings nearly a month’s worth -of international film premieres to -Seattle. Screenings take place at several -theaters across the city, including the -official SIFF Theater in Seattle Center. -Check the film festival’s website for a -detailed schedule. - -ADVANCE PLANNING -As in any large American city, planning ahead will let you spend more of your time enjoying Seattle. Things like car -rental, accommodations and tours should be booked in advance. If you show up without a hotel booking in peak season, -you might find yourself stranded. - -Train tickets to Seattle may need to be booked in advance, as some trains do sell out. Make reservations at www - -.amtrak.com or by phone at %800-872-7245. - -If you’re hoping to see a particular performance or game, whether it’s the Mariners or the opera, it’s wise to buy -tickets in advance. See individual listings chapters for price ranges and information on how to book (for sporting events, -see p165; for theater and other performance events, see p159). - -Some restaurants recommend or require booking a table in advance, and these are noted in the Eating chapter - -(p119). For most places in Seattle, booking a day or two ahead is enough, unless it’s a holiday. - -The Seattle CityPass (www.citypass.net), which offers discounts on Seattle museums and other attractions, can be -ordered in advance on its website. The website www.dailycandy.com produces a ‘weekend guide’ to Seattle; it’s worth -subscribing to the mailing list before a trip. - -It’s useful to read up on what’s happening in town via The Stranger’s savvy and entertaining online blogs, Slog (for -news) and LineOut (for music and nightlife) at www.thestranger.com. Rabid menu-readers will appreciate the Seattle -Times’ archive of restaurant reviews at www.seattletimes.com. - -Seattle’s Citywide Concierge Center (%206-461-5888; www.visitseattle.org; 701 Pike St) can help with advance - -bookings of almost anything you’ll need, including accommodations, tours and the CityPass. - -l - -o -n -e +(B) 0.4 -l -y -p +Figure 23: An Econometrics example. -l -a -n -e -t -. -c -o -m +A point pole has a strength of 4π × 10−4 weber. The force in newtons on a point pole of +4π × 1.5 × 10−4 weber placed at a distance of 10 cm from it will be +(A) 15 N. -G -E -T -T -I +(D) 3.75 N. -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G -O +(C) 7.5 N. -17 +(B) 20 N. - June -FREMONT STREET FAIR -%206-649-6706; www.fremontfair.com -Off-kilter Fremont is just the place you’d -want to be for a street fair. This one has live -music, food and crafts, and the overtly artsy -Solstice Parade, where human-powered -floats traipse through the neighborhood -in a lively tribute to quirkiness. It takes -place adjacent to the ship canal, usually -in mid-June. - -SEATTLE PRIDE FESTIVAL -%206-684-7200; www.seattlepride.org; -Seattle Center -Seattle’s lesbian- and gay-pride event -usually falls on the last Sunday in June and -includes a film festival, art exhibit, parade -and the requisite food and entertainment. - -July -BITE OF SEATTLE -%425-283-5050; www.biteofseattle.com; -Seattle Center -This culinary celebration is usually held the -second weekend in July. Guests can sample -food from dozens of Seattle-area chefs, -including sub-$4 samples at the Just a Bite -booth, and taste local beer and wines. The -evening ends with live music. - -SEAFAIR -%206-728-0123; www.seafair.com -A month-long extravaganza, Seafair -is a civic celebration that began as a -hydroplane race on Lake Washington. -It’s now the city’s biggest festival, with -all manner of festivities extending across -Seattle. Events include a torchlight parade, -an air show, lots of music, a carnival and -even the arrival of the naval fleet. Lodging -is in short supply in Seattle on Seafair -weekends and traffic becomes hopelessly -tangled thanks to closed bridges and the -influx of visitors, so plan accordingly. - -September -BUMBERSHOOT -%206-281-7788; www.bumbershoot.org -Seattle’s biggest arts and cultural event -takes over Seattle Center on Labor Day -weekend (the first Monday in September). -Hundreds of musicians, artists, theater - -l - -o -n -e +Figure 24: An Electrical Engineering example. -l -y -p +A total of 30 players will play basketball at a park. There will be exactly 5 players on each team. +Which statement correctly explains how to find the number of teams needed? +(A) Add 5 to 30 to find 35 teams. +(B) Divide 30 by 5 to find 6 teams. +(C) Multiply 30 and 5 to find 150 teams. +(D) Subtract 5 from 30 to find 25 teams. -l -a -n -e -t -. -c -o -m +Figure 25: An Elementary Mathematics example. - G -E -T -T -I +Determine whether the statements are logically equivalent or contradictory. If neither, determine +whether they are consistent or inconsistent. +E ⊃ (F · E) and ∼ E · F +(A) Logically equivalent +(B) Contradictory +(C) Neither logically equivalent nor contradictory, but consistent +(D) Inconsistent -N -G -S -T -A -R -T -E -D - -W -H -E -N -T -O -G -O +Figure 26: A Formal Logic example. -18 +As of 2017, how many of the world’s 1-year-old children today have been vaccinated against +some disease? +(A) 80% +(B) 60% +(C) 40% +(D) 20% -troupes and writers come from all over -the country to perform on the festival’s -two dozen stages. It’s hard not to find -something you like. There’s also a crafts -street fair and lots of good food from -local vendors. - -SEATTLE AIDS WALK -%206-329-6923; www.seattleaidswalk.org -This festive and friendly event, held in late -September, grows every year. It’s a 10km or -5km walk (or run). Locations vary, so check -the website for details. The walk benefits -the Lifelong AIDS Alliance. - -FREMONT OKTOBERFEST -%206-632-1500; www.fremontoktoberfest.org -There are versions of Oktoberfest held -all over town, but this one’s a good -introduction to the concept as well as the -neighborhood. Held in late September, -it includes crafts, a car show, live -entertainment and, of course, beer tasting. - -October -SEATTLE LESBIAN & GAY FILM -FESTIVAL -%206-323-4274; www.seattlequeerfilm.com -Held in the third week of October, this -film festival brings both mainstream -and underground queer films to various -theaters around town. - -HALLOWEEN -On October 31, kids and adults dress up in -scary costumes. In the safer neighborhoods -you’re likely to see children out ‘trick or - -OFFBEAT SEATTLE FESTIVALS - - Bumbershoot (left) A cultural buffet of music, -theater, artwork and poetry on 25 stages. - Fremont Street Fair (left) Celebrate one of - -Seattle’s quirkiest corners. - - Winterfest (opposite) A celebration of all that’s - -cold, furry and white. - - Trolloween (above) Just in case regular Halloween - -isn’t weird enough for you. - - Moisture Festival (p16) Performances by oddball - -artists both local and international. - - treating’ door-to-door for candy. The -gay and lesbian bars are especially wild -places on Halloween, and Fremont has -its own version, Trolloween (named after -the Fremont Troll). It features a candlelit -procession of costumed locals on Halloween -night, followed by a public dance. - -December -WINTERFEST -%206-684-7200; www.seattlecenter.com -Seattle Center holds a month-long -celebration of holiday traditions from -around the globe, starting with Winter -Worldfest, a massive concert and dance -performance, and continuing with exhibits, -dances, concerts and ice skating. - -NEW YEAR’S EVE -%206-443-2111; www.seattlecenter.com -The place to be on the 31st is Seattle -Center, where festivities are focused on the -Space Needle. Most people celebrate by -dressing up and drinking champagne, or -staying home and watching it all on TV. The -following day people stay home to nurse -their hangovers and watch college football. - -COSTS & MONEY - -As in most large cities, expenses in Seattle -can vary depending on your interests and -means. Accommodations will make up the -bulk of your expenses. Booking in advance -or planning your visit during the off-season -(September–May) will generally get you better -value for money. - -A dorm bed in a youth hostel costs about -$35 including taxes; a double room in one -of Seattle’s European-style hotels will go for -$60 to $80, plus tax. Midrange doubles are -between $100 and $200. See the Sleeping -chapter (p173) for details about Seattle Super -Saver hotel discount packages. - -After accommodations, dining will prob- -ably be your next-largest budget item. The -city offers a good range of meals, from spec- -tacularly decadent restaurant meals to quick, -budget-friendly cafe snacks. It’s a foodie city -that’s well positioned to take advantage of -fresh seafood and produce, which means -that in general splurges will be rewarded with -unforgettable dining experiences. Budget -meals are under $15 a person; breakfast and -brunch and most lunches fall in this category. - -HOW MUCH? -Adult ticket up the Space Needle $18 -Cover charge at the Crocodile Café $5 to $15 -Gallon of gasoline $3 -Late-night burrito from a taco cart $5 to $7 -Liter of bottled water $2 -Pint of local microbrewed beer $4.50 -Room at the hip Ace Hotel $99 -Souvenir T-shirt $20 -Taxi from airport to Pike Place Market $35 -Light-rail ticket from airport to downtown $2.50 - -Midrange prices hover in the $16 to $30 range, -and top-end dinners start around $30, plus -tax, tip and whatever you’re drinking. - -For discounts on sights and entertainment, -consider buying a Go Seattle Card (www.goseattlecard -.com) or a CityPass (www.citypass.net), both of -which offer free or discounted admission -to most of the city’s top attractions. See the -Neighborhoods chapter (p92) for details. - -INTERNET RESOURCES - -The following websites may be helpful in -planning your travels. -Bohonus VR (www.bohonus.com/galleries) If you just can’t -wait until you get there, check out this site for 360-degree -views of various Seattle sights, shops and hangouts. -Daily Candy (www.dailycandy.com) The Seattle edition -digs up new and fun things to see, do or buy each day. -Hankblog (http://hankblog.wordpress.com) Insider -art-related news and views from the folks at the Henry -Art Gallery. -Lineout (http://lineout.thestranger.com) The Stranger’s -music blog – a great way to get plugged into the scene -quickly. -Lonely Planet (www.lonelyplanet.com) Succinct summa- -ries on traveling to most places on earth; postcards from -other travelers; and the Thorn Tree bulletin board, where -you can ask questions before you go or dispense advice -when you get back. -Seattle DIY (www.seattlediy.com) A great source of -information on underground events and house shows. -Seattle’s Convention and Visitors Bureau (www.visit -seattle.org) Seattle’s official visitors information site. -Seattlest (www.seattlest.com) An informative (but less -savvy or sassy than the Stranger) blog about various news -and goings-on in and around Seattle. - -l - -o -n -e +Figure 27: A Global Facts example. -l -y -p +Homologous structures are often cited as evidence for the process of natural selection. All of the +following are examples of homologous structures EXCEPT +(A) the wings of a bird and the wings of a bat +(B) the flippers of a whale and the arms of a man +(C) the pectoral fins of a porpoise and the flippers of a seal +(D) the forelegs of an insect and the forelimbs of a dog -l -a -n -e -t -. -c -o -m +Figure 28: A High School Biology example. -G -E -T -T -I +From the solubility rules, which of the following is true? +(A) All chlorides, bromides, and iodides are soluble +(B) All sulfates are soluble +(C) All hydroxides are soluble +(D) All ammonium-containing compounds are soluble -N -G -S -T -A -R -T -E -D +Figure 29: A High School Chemistry example. -C -O -S -T -S -& -M -O -N -E -Y +16 + + A list of numbers has n elements, indexed from 1 to n. The following algorithm is intended to +display the number of elements in the list that have a value greater than 100. The algorithm uses +the variables count and position. Steps 3 and 4 are missing. +Step 1: Set count to 0 and position to 1. +Step 2: If the value of the element at index position is greater +than 100, increase the value of count by 1. +Step 3: (missing step) +Step 4: (missing step) +Step 5: Display the value of count. +Which of the following could be used to replace steps 3 and 4 so that the algorithm works as +intended? +(A) Step 3: Increase the value of position by 1. +Step 4: Repeat steps 2 and 3 until the value of count is greater than 100. +(B) Step 3: Increase the value of position by 1. +Step 4: Repeat steps 2 and 3 until t he value of position is greater than n. +(C) Step 3: Repeat step 2 until the value of count is greater than 100. +Step 4: Increase the value of position by 1. +(D) Step 3: Repeat step 2 until the value of position is greater than n. +Step 4: Increase the value of count by 1. + +Figure 30: A High School Computer Science example. + +This question refers to the following information. +Albeit the king’s Majesty justly and rightfully is and ought to be the supreme head of the Church +of England, and so is recognized by the clergy of this realm in their convocations, yet nevertheless, +for corroboration and confirmation thereof, and for increase of virtue in Christ’s religion within +this realm of England, and to repress and extirpate all errors, heresies, and other enormities and +abuses heretofore used in the same, be it enacted, by authority of this present Parliament, that the +king, our sovereign lord, his heirs and successors, kings of this realm, shall be taken, accepted, +and reputed the only supreme head in earth of the Church of England, called Anglicans Ecclesia; +and shall have and enjoy, annexed and united to the imperial crown of this realm, as well the +title and style thereof, as all honors, dignities, preeminences, jurisdictions, privileges, authorities, +immunities, profits, and commodities to the said dignity of the supreme head of the same Church +belonging and appertaining; and that our said sovereign lord, his heirs and successors, kings of +this realm, shall have full power and authority from time to time to visit, repress, redress, record, +order, correct, restrain, and amend all such errors, heresies, abuses, offenses, contempts, and +enormities, whatsoever they be, which by any manner of spiritual authority or jurisdiction ought +or may lawfully be reformed, repressed, ordered, redressed, corrected, restrained, or amended, +most to the pleasure of Almighty God, the increase of virtue in Christ’s religion, and for the +conservation of the peace, unity, and tranquility of this realm; any usage, foreign land, foreign +authority, prescription, or any other thing or things to the contrary hereof notwithstanding. +English Parliament, Act of Supremacy, 1534 +From the passage, one may infer that the English Parliament wished to argue that the Act of +Supremacy would +(A) give the English king a new position of authority +(B) give the position of head of the Church of England to Henry VIII alone and exclude his heirs +(C) establish Calvinism as the one true theology in England +(D) end various forms of corruption plaguing the Church in England + +Figure 31: A High School European History example. + +During the third stage of the demographic transition model, which of the following is true? +(A) Birth rates increase and population growth rate is less rapid. +(B) Birth rates decline and population growth rate is less rapid. +(C) Birth rates increase and population growth rate increases. +(D) Birth rates decrease and population growth rate increases. + +Figure 32: A High School Geography example. + +17 + + Which of the following best states an argument made by James Madison in The Federalist number +10? +(A) Honest politicians can prevent factions from developing. +(B) Factions are more likely to occur in large republics than in small ones. +(C) The negative effects of factionalism can be reduced by a republican government. +(D) Free elections are the people’s best defense against factionalism. + +Figure 33: A High School Government and Politics example. + +Which of the following is not included in the U.S. GDP? +(A) The U.S. military opens a new base in a foreign country with 1000 U.S. personnel. +(B) Japanese consumers buy thousands of CDs produced in the United States. +(C) An American pop singer performs a sold-out concert in Paris. +(D) A French theatrical production tours dozens of American cities. + +Figure 34: A High School Macroeconomics example. + +Joe was in charge of lights for a dance. The red light blinks every two seconds, the yellow light +every three seconds, and the blue light every five seconds. If we include the very beginning and +very end of the dance, how many times during a seven minute dance will all the lights come on at +the same time? (Assume that all three lights blink simultaneously at the very beginning of the +dance.) +(A) 3 +(B) 15 +(C) 6 +(D) 5 + +Figure 35: A High School Mathematics example. + +If the government subsidizes producers in a perfectly competitive market, then +(A) the demand for the product will increase +(B) the demand for the product will decrease +(C) the consumer surplus will increase +(D) the consumer surplus will decrease + +Figure 36: A High School Microeconomics example. + +A point charge, Q = +1 mC, is fixed at the origin. How much work is required to move a charge, +Q = +8 µC, from the point (0, 4 meters) to the point (3 meters, 0)? +(A) 3.5 J +(B) 6.0 J +(C) 22.5 J +(D) 40 J + +Figure 37: A High School Physics example. + +While swimming in the ocean, Ivan is frightened by a dark shadow in the water even before he +has the chance to identify what the shadow is. The synaptic connections taking place during this +incident of fright are best described by which of the following? +(A) Messages are sent from the thalamus directly to the amygdala. +(B) Messages are sent from the thalamus to the “what” and “where” pathways. +(C) Messages are sent from the parasympathetic nervous system to the cerebral cortex. +(D) Messages are sent from the frontal lobes to the pituitary gland. + +Figure 38: A High School Psychology example. + +18 + + Jonathan obtained a score of 80 on a statistics exam, placing him at the 90th percentile. Suppose +five points are added to everyone’s score. Jonathan’s new score will be at the +(A) 80th percentile. +(B) 85th percentile. +(C) 90th percentile. +(D) 95th percentile. + +Figure 39: A High School Statistics example. + +This question refers to the following information. +“Society in every state is a blessing, but government even in its best state is but a necessary evil; in +its worst state an intolerable one; for when we suffer, or are exposed to the same miseries by a +government, which we might expect in a country without government, our calamity is heightened +by reflecting that we furnish the means by which we suffer. Government, like dress, is the badge +of lost innocence; the palaces of kings are built on the ruins of the bowers of paradise. For were +the impulses of conscience clear, uniform, and irresistibly obeyed, man would need no other +lawgiver; but that not being the case, he finds it necessary to surrender up a part of his property to +furnish means for the protection of the rest; and this he is induced to do by the same prudence +which in every other case advises him out of two evils to choose the least. Wherefore, security +being the true design and end of government, it unanswerably follows that whatever form thereof +appears most likely to ensure it to us, with the least expense and greatest benefit, is preferable to +all others.” +Thomas Paine, Common Sense, 1776 +Which of the following “miseries” alluded to above were most condemned by Anti-Federalists of +the post-Revolutionary era? +(A) Organized response to Bacon’s Rebellion. +(B) Federal response to Shays’s Rebellion. +(C) Federal response to the Whiskey Rebellion. +(D) Federal response to Pontiac’s Rebellion. + +Figure 40: A High School US History example. + +All other things being equal, which of the following persons is more likely to show osteoporosis? +(A) An older Hispanic American woman +(B) An older African American woman +(C) An older Asian American woman +(D) An older Native American woman + +Figure 41: A Human Aging example. + +Morning sickness is typically a problem: +(A) during the first trimester +(B) during the second trimester +(C) during the third trimester +(D) all through the pregnancy + +Figure 42: A Human Sexuality example. 19 - l + Would a reservation to the definition of torture in the ICCPR be acceptable in contemporary +practice? +(A) This is an acceptable reservation if the reserving country’s legislation employs a different +definition +(B) This is an unacceptable reservation because it contravenes the object and purpose of +the ICCPR +(C) This is an unacceptable reservation because the definition of torture in the ICCPR is consistent +with customary international law +(D) This is an acceptable reservation because under general international law States have the right +to enter reservations to treaties -o -n -e +Figure 43: An International Law example. -l -y -p +Which position does Rawls claim is the least likely to be adopted by the POP (people in the +original position)? +(A) The POP would choose equality above liberty. +(B) The POP would opt for the ‘maximin’ strategy. +(C) The POP would opt for the ‘difference principle.’ +(D) The POP would reject the ‘system of natural liberty.’ -l -a -n -e -t -. -c -o -m - -Slog (slog.thestranger.com) A frequently updated blog by -the staff of The Stranger newsweekly. -Sustainable Seattle (www.sustainableseattle.org) Though -it’s a bit academic, this site has thorough information on -ecofriendly projects and initiatives in Seattle. - -SUSTAINABLE SEATTLE - -Nature is a huge part of the Seattle experi- -ence and it should come as no surprise that -the city is good at being green. So what can -visitors do to help? - -International flights are pretty much un- -avoidable, so consider contributing to a car- -bon offset organization like www.carbonfund -.org, www.terrapass.com, www.nativeenergy -.com or www.driveneutral.org, where your -money is transformed into support for -sustain able practices. Or, if you’re close -enough, plan to travel on a biodiesel bus (p200) -or on a train (p201). - -Once you’ve arrived in Seattle, you can -easily get by without having a car. Local bus -networks are good, and biking is a great -option (rentals abound). Walking and using -the public transportation system are also good -earth-friendly options. And if you do need a -car, think about making use of www.flexcar -.com, a car-sharing program. Some car-rental -agencies offer hybrid vehicles, so be sure to ask -around. Also, take a look at www.craigslist -.com, an online bulletin board where you can -find regional rideshares (look under ‘com- -munity’); ie folks who are looking for people -to carpool and share the cost of gas. Similarly, -the adventurous can check www.couch -surfing.org, where you sleep on other member’s -couches – exchanging cultures, saving money -and using hopefully fewer resources. - -In hotels, save energy by requesting that -your sheets and towels not be laundered -every day. Shut off lights, the TV and air-con -when you’re not in the room. And consider -not using those inefficient little travel-sized -sundries. Hostels are another option as their - -high-density dormitories use less space and -energy. See the box, p180, for a list of hotels -that are making an effort. - -Sustainable Seattle, a nonprofit organi- -zation, holds an annual awards ceremony -in November to recognize local businesses -that meet certain sustainability criteria; visit -www.sustainableseattle.org to learn more -about the program. - -Our GreenDex (p227) at the end of this book -lists organizations and businesses that have -been recognized for taking extra measures to -help protect the environment, support local -charities, use local products or recycled ma- -terials, or promote regional culture, among -other positive practices. Here’s a sample: -Chaco Canyon Café (p131) Cafe serving all vegan, 90% -organic and almost half raw food; also an organic bakery. -FareStart Restaurant (p122) Profits support the FareStart -program, which provides intensive job training, housing -assistance and job placement for disadvantaged and -homeless people. -Foster Island Wetlands Trail (p100) A waterfront trail -that winds through marshlands and across to smaller -islands for bird-watching, swimming, fishing and kayaking. -Fremont Sunday Market (p86) A bustling marketplace -in the city’s quirkiest quadrant, there’s fresh produce, art -and junk. -Olympic Sculpture Park (p67) An excellent use of urban -space to present public art and sculpture. -Tillicum Village (p189) On Blake Island, tours include -traditional Indian salmon bake, dancing and a film about -Northwest Native Americans. -Tilth (p133) A certified-organic gourmet cafe in Wallingford. -Volunteer Park Conservatory (p80) A classic Victorian -greenhouse with five galleries representing different world -environments. -Washington Trails Association (p169) A nonprofit group -that organizes hiking trips, conservation efforts and -trail-building jaunts into local mountains. -Wing Luke Museum of the Asian Pacific American -Experience (p59) Museum devoted to the history of Asian -and Pacific culture and art in America. - -© Lonely Planet Publications Pty Ltd. To make it easier for you to use, access to this chapter is not -digitally restricted. In return, we think it’s fair to ask you to use it for personal, non-commercial -purposes only. In other words, please don’t upload this chapter to a peer-to-peer site, mass email -it to everyone you know, or resell it. See the terms and conditions on our site for a longer way of -saying the above - ‘Do the right thing with our content.’ - - G -E -T -T -I - -N -G -S -T -A -R -T -E -D +Figure 44: A Jurisprudence example. -S -U -S -T -A -I -N -A -B -L -E -S -E -A -T -T -L -E +John Stuart Mill: Each person’s happiness is a good to that person, and the general happiness, +therefore, a good to the aggregate of all persons. +(A) Fallacy of Composition +(B) Fallacy of Division +(C) Gambler’s Fallacy +(D) Equivocation + +Figure 45: A Logical Fallacies example. + +A 6-sided die is rolled 15 times and the results are: side 1 comes up 0 times; side 2: 1 time; side +3: 2 times; side 4: 3 times; side 5: 4 times; side 6: 5 times. Based on these results, what is the +probability of side 3 coming up when using Add-1 Smoothing? +(A) 2/15 + +(C) 3/16 + +(D) 1/5 + +(B) 1/7 + +Figure 46: A Machine Learning example. + +According to Lewin, Lippet and White’s 1939 experiment, which form of leadership produced the +most work from participants? +(A) Laissez-faire +(B) Democratic +(C) Authoritarian +(D) A mix of laissez-faire and democratic + +Figure 47: A Management example. + +The single group within society that is most vulnerable to reference group influence is: +(A) The older consumer who feels somewhat left out of things. +(B) The married women, many of whom feel a need for stability in their lives. +(C) New immigrants who really want to assimilate into their new culture. +(D) Children, who base most of their buying decisions on outside influences. + +Figure 48: A Marketing example. 20 + Which of the following conditions does not show multifactorial inheritance? +(A) Pyloric stenosis +(B) Schizophrenia +(C) Spina bifida (neural tube defects) +(D) Marfan syndrome + +Figure 49: A Medical Genetics example. + +If each of the following meals provides the same number of calories, which meal requires the +most land to produce the food? +(A) Red beans and rice +(B) Steak and a baked potato +(C) Corn tortilla and refried beans +(D) Lentil soup and brown bread + +Figure 50: A Miscellaneous example. + +In response to Sandel’s “social justice” argument, Kamm argues that +(A) even if we were able to enhance ourselves or others, we would not thereby be obligated to do +so. +(B) there is a difference between (i) choosing that one have a certain trait and (ii) being responsible +for bearing the costs of that choice. +(C) having the option to enhance will lead to fewer instances of people who require the assistance +of others. +(D) all of the above + +Figure 51: A Moral Disputes example. + +In response to Sandel’s “social justice” argument, Kamm argues that +(A) even if we were able to enhance ourselves or others, we would not thereby be obligated to do +so. +(B) there is a difference between (i) choosing that one have a certain trait and (ii) being responsible +for bearing the costs of that choice. +(C) having the option to enhance will lead to fewer instances of people who require the assistance +of others. +(D) all of the above + +Figure 52: A Moral Disputes example. + +For which of these two scenarios does the main character (who uses I/me/my) do something +clearly morally wrong? +Scenario 1 | I drive very fast when I’m on a racetrack. +Scenario 2 | I drive very fast when I’m in the suburbs. +(A) Wrong, Wrong +(B) Wrong, Not wrong +(C) Not wrong, Wrong +(D) Not wrong, Not wrong + +Figure 53: A Moral Scenarios example. + +21 + + Which of the following is the most plausible explanation for the protective effect of dietary fibre +against cancer of the colon? +(A) Propionic acid, formed during colonic fibre fermentation inhibits liver fatty acid synthesis +(B) Butyric acid, formed during colonic fibre fermentation stimulates "silencing" of the SLC5A8 +tumour suppressor gene +(C) Butyric acid, formed during colonic fibre fermentation stimulates anti-oxidant defences +in the colon +(D) None of these options are correct + +Figure 54: A Nutrition example. + +According to Moore’s “ideal utilitarianism,” the right action is the one that brings about the +greatest amount of: +(A) pleasure. +(B) happiness. +(C) good. +(D) virtue. + +Figure 55: A Philosophy example. + +Researchers now believe that the decline of the Maya was caused chiefly by: +(A) a cataclysm of some kind, such as an earthquake, volcano, or tsunami. +(B) ecological degradation resulting from slash-and-burn farming techniques. +(C) endless wars between neighboring Mayan city-states. +(D) practices of interbreeding that led to a steep rise in congenital disorders. + +Figure 56: A Prehistory example. + +Krete is an unmarried taxpayer with income exclusively from wages. By December 31, year 1, +Krete’s employer has withheld $16,000 in federal income taxes and Krete has made no estimated +tax payments. On April 15, year 2, Krete timely filed for an extension request to file her individual +tax return, and paid $300 of additional taxes. Krete’s year 1 tax liability was $16,500 when she +timely filed her return on April 30, year 2, and paid the remaining tax liability balance. What +amount would be subject to the penalty for underpayment of estimated taxes? +(A) $0 +(B) $500 +(C) $1,650 +(D) $16,500 + +Figure 57: A Professional Accounting example. + +The night before his bar examination, the examinee’s next-door neighbor was having a party. The +music from the neighbor’s home was so loud that the examinee couldn’t fall asleep. The examinee +called the neighbor and asked her to please keep the noise down. The neighbor then abruptly hung +up. Angered, the examinee went into his closet and got a gun. He went outside and fired a bullet +through the neighbor’s living room window. Not intending to shoot anyone, the examinee fired +his gun at such an angle that the bullet would hit the ceiling. He merely wanted to cause some +damage to the neighbor’s home to relieve his angry rage. The bullet, however, ricocheted off the +ceiling and struck a partygoer in the back, killing him. The jurisdiction makes it a misdemeanor +to discharge a firearm in public. The examinee will most likely be found guilty for which of the +following crimes in connection to the death of the partygoer? +(A) Murder. +(B) Involuntary manslaughter. +(C) Voluntary manslaughter. +(D) Discharge of a firearm in public. + +Figure 58: A Professional Law example. + +22 + + A 63-year-old man is brought to the emergency department because of a 4-day history of in- +creasingly severe left leg pain and swelling of his left calf. He also has a 1-month history of +increasingly severe upper midthoracic back pain. During this time, he has had a 9-kg (20-lb) +weight loss despite no change in appetite. He has no history of major medical illness. His only +medication is ibuprofen. He is 180 cm (5 ft 11 in) tall and weighs 82 kg (180 lb); BMI is 25 +kg/m2 . His vital signs are within normal limits. On examination, lower extremity pulses are +palpable bilaterally. The remainder of the physical examination shows no abnormalities. An x-ray +of the thoracic spine shows no abnormalities. A CT scan of the abdomen shows a 3-cm mass in +the body of the pancreas; there are liver metastases and encasement of the superior mesenteric +artery. Ultrasonography of the left lower extremity shows a femoropopliteal venous clot. Which +of the following is the most likely cause of this patient’s symptoms? +(A) Carcinoid syndrome +(B) Hypercoagulability from advanced malignancy +(C) Multiple endocrine neoplasia +(D) Splenic artery aneurysm and embolic disease of the left lower extremity + +Figure 59: A Professional Medicine example. + +The technique that is most likely to produce an immediate improvement in the behavior of a child +who hits others and rips up schoolbooks is +(A) a combination of reinforcement for appropriate behavior and mild punishment for in- +appropriate behavior +(B) differential positive reinforcement +(C) loss of privileges for misbehaving +(D) shaping + +Figure 60: A Professional Psychology example. + +You work for a utility company that is building a biomass plant in the community. Your employer +asks you to give a series of community talks about the plant and future operations. You visit the +plant several hours before you are due to give a speech that has been prepared by your immediate +supervisor. During the tour of the plant, you discover several claims in the speech are not true. +What do you do? +(A) Write and deliver a new speech that you know is entirely correct. +(B) Cancel all speeches until you and your supervisor can get the information straight. +(C) Deliver the speech as prepared and discuss the inaccuracies with your supervisor afterward. +(D) Address the inaccuracies with your supervisor immediately and make the necessary +corrections before giving the speech. + +Figure 61: A Public Relations example. + +23 + + The Chemical Weapons Convention (CWC) prohibited the possession or deployment of chemical +weapons; however it failed to implement stipulations that would require signatories to declare their +existing stocks of chemical weapons, to identify facilities that were once involved in chemical +production, or to announce when their existing stocks would be destroyed. +(A) The Chemical Weapons Convention (CWC) prohibited the possession or deployment of +chemical weapons; however it failed to implement stipulations that would require signatories to +declare their existing stocks of chemical weapons, to identify facilities that were once involved in +chemical production, or to announce when their existing stocks would be destroyed. +(B) The CWC made some important developments regarding the use and possession of chemical +weapons and the destruction of existing stockpiles. However, the treaty failed to establish an +independent body empowered with the capacity to check treaty compliance. Lack of supra-state +authority has undermined the ability to enforce those developments. Given the anarchical nature +of international society it may be in the national security interest to retain stocks. +(C) Chemical weapons continue to exert a determining influence on international society. As early +as the 1970s military strategists were convinced of the deterrence effects chemical weapons could +have, comparable to the second strike survival logic of nuclear deterrence. The preferences of +strategists resulted in continued manufacture and stockpiling of weapons creating an international +crisis of stability. +(D) While the CWC has been ratified by the majority of international society, some nations +with a large chemical capability at their disposal have yet to enter into the treaty. However, +to some analysts the destructive military potential would be limited, having a moderate +effect on a well-equipped army in conventional warfare. Chemical arsenal essentially falls +under the category of the "poor mans" weaponry, being simplistic and inexpensive whilst +having limited military utility. However, the concern remains of the prospective impact a +terrorist chemical attack could have on civilian populations. + +Figure 62: A Security Studies example. + +Which of the following statements most closely corresponds with differential association theory? +(A) If all of your friends jumped off a bridge, I suppose you would too. +(B) You should be proud to be a part of this organization. +(C) If the door is closed, try the window. +(D) Once a thief, always a thief. + +Figure 63: A Sociology example. + +Which of the following statements most closely corresponds with differential association theory? +(A) If all of your friends jumped off a bridge, I suppose you would too. +(B) You should be proud to be a part of this organization. +(C) If the door is closed, try the window. +(D) Once a thief, always a thief. + +Figure 64: A Sociology example. + +Why did Congress oppose Wilson’s proposal for the League of Nations? +(A) It feared the League would encourage Soviet influence in the US +(B) It feared the League would be anti-democratic +(C) It feared the League would commit the US to an international alliance +(D) Both a and b + +Figure 65: A US Foreign Policy example. + +24 + + An observational study in diabetics assesses the role of an increased plasma fibrinogen level on +the risk of cardiac events. 130 diabetic patients are followed for 5 years to assess the development +of acute coronary syndrome. In the group of 60 patients with a normal baseline plasma fibrinogen +level, 20 develop acute coronary syndrome and 40 do not. In the group of 70 patients with a high +baseline plasma fibrinogen level, 40 develop acute coronary syndrome and 30 do not. Which of +the following is the best estimate of relative risk in patients with a high baseline plasma fibrinogen +level compared to patients with a normal baseline plasma fibrinogen level? +(A) (40/30)/(20/40) +(B) (40*40)/(20*30) +(C) (40*70)/(20*60) +(D) (40/70)/(20/60) + +Figure 66: A Virology example. + +The Great Cloud Sutra prophesied the imminent arrival of which person? +(A) Maitreya (Milo) +(B) The Buddha +(C) Zhou Dunyi +(D) Wang Yangming + +Figure 67: A World Religions example. + +25 + diff --git a/q104/random_k4/random_4.md b/q104/random_k4/random_4.md index e3ed939376d2f6fad348f021570f136f5c225f39..c906bff4a2b28ca8e5a69ef8758f9182d084ead4 100644 --- a/q104/random_k4/random_4.md +++ b/q104/random_k4/random_4.md @@ -1,1527 +1,1633 @@ -Issue Brief - -MTV/AP-NORC: COMPARING THE -POLITICAL VIEWS OF YOUNG -PEOPLE AND THEIR -PARENTS’ -GENERATION - -Young people and their parents share a negative outlook on -the current political landscape in America and are largely -dissatisfied with the government’s handling of a wide -variety of key issues, according to a new poll of Americans -age 15-34, and parents of children age 15-26 by MTV and The -Associated Press-NORC Center for Public Affairs Research. - -Young people’s opinions about their generation’s ability to -impact the government have shifted in recent months, -following student-driven activism. Between the first poll -fielded between late February and early March 2018, -shortly after the shooting in Parkland, Florida, and this -second poll conducted roughly two months later, young -people age 15-34, and particularly those age 15-22, feel -more politically empowered. The proportion overall who -believe that elected officials care what they think rose -from a quarter to a third, and there was a similar rise in -the numbers who think people like them can impact the -government. - -The generation of parents of children age 15-26 is less -racially and ethnically diverse and includes a greater -proportion of Republicans than young people age 15-26, -and the survey demographics reflect this. Parents are -more likely to approve of President Donald Trump (42 -percent vs. 26 percent) and are somewhat more optimistic -(59 percent vs. 45 percent) than the younger generation. -Still, the findings show that young people and parents are -on the same page about politics in general. Majorities say -they tend to agree with one another on many key issues -and that their political discussions are generally amicable. - -© 2017 iStock/Nikada - -Three Things You Should Know -About the MTV/AP-NORC Poll of -Americans Age 15-34 and Parents -of 15-26 Year Olds: +This copy is for personal use only. To order printed copies, contact reprints@rsna.org + +1733 + +Fluoroscopic Swallowing Examination: +Radiologic Findings and Analysis of +Their Causes and Pathophysiologic +Mechanisms + +Alberto I. Carbo, MD +Melanie Brown, CCC-SLP +Nour Nakrour, MD + +Abbreviations: CP = cricopharyngeus, HLC = +hyolaryngeal complex, PES = pharyngoesopha­ +geal segment, UES = upper esophageal sphincter + +RadioGraphics 2021; 41:1733–1749 + +https://doi.org/10.1148/rg.2021210051 + +Content Codes: + +From the Departments of Radiology (A.I.C., +N.N.) and Rehabilitation Services (M.B.), Ochs­ +ner LSU Health, 1501 Kings Hwy, Shreveport, +LA 71103. Presented as an education exhibit +at the 2020 RSNA Annual Meeting. Received +March 7, 2021; revision requested April 19 and +received May 14; accepted May 23. For this jour­ +nal­based SA­CME activity, the authors, editor, +and reviewers have disclosed no relevant rela­ +tionships. Address correspondence to A.I.C. +(e­mail: acarbo1396@gmail.com). + +©RSNA, 2021 + +SA-CME LEARNING OBJECTIVES + +After completing this journal-based SA-CME +activity, participants will be able to: + „Discuss the anatomy and imaging +phases of swallowing. + + „Describe abnormalities detected at +fluoroscopic swallowing examination. + + „Correlate the radiologic findings +with their causes and pathophysiologic +mechanisms. + +See rsna.org/learning-center-rg. + +Dysphagia is a common symptom in the general population, and its +prevalence increases with patient age. The deterioration of swallow­ +ing function has many acute and chronic causes, including cerebro­ +vascular and neuromuscular diseases, radiation, and surgery. In an +elderly population, diagnosis and treatment of swallowing abnor­ +malities is a high priority because it improves the patient’s quality +of life and helps them to avoid medical complications. Fluoroscopic +swallowing examinations and modified barium swallow studies are +the most used and most reliable diagnostic procedures to evaluate +swallowing disorders. Functional anomalies include disturbances +of the oral preparatory, oral propulsory, and pharyngeal phases +of swallowing as premature spillage from the mouth, nasal regur­ +gitation, delayed initiation of pharyngeal swallowing, incomplete +displacement of the hyolaryngeal complex, abnormal epiglottic +tilt, incomplete laryngeal closure, and pharyngeal dysmotilities. +Anatomic abnormalities of the pharynx include diverticula, be­ +nign strictures, and tumors. The abnormalities diagnosed on the +basis of fluoroscopic examination have a variety of treatment strat­ +egies, and the choice of treatment depends on the cause of the +anomaly and its pathophysiologic characteristics. The radiologist’s +interpretation of these characteristics is crucial to therapeutic de­ +cision making and achieving the best patient outcomes. + +Online supplemental material is available for this article. + +©RSNA, 2021 • radiographics.rsna.org + +Introduction +Oropharyngeal dysphagia is prevalent in 20% of the general popu­ +lation and in 50% of nursing home residents in the United States +(1). As the average age of the population continues to increase, +there has been an increase in dysphagia and related complications. +Fluoroscopically guided swallowing examinations are the most used +technique for evaluation of deglutition disorders. The examination is +called modified barium swallow when the diagnostic study includes a +therapeutic test; in those cases, it is performed with the participation +of a swallowing therapist or speech­language pathologist (2). + +Fluoroscopic examination of swallowing and modified barium +swallow should not be routine studies but should be custom­tailored +techniques designed by radiologists and swallowing therapists on +the basis of the patient’s clinical history and symptoms. Radiolo­ +gists should supervise or participate in the procedures and contrib­ +ute their knowledge to adapt the technique to unexpected findings +detected during examinations. They may decide during the study +to include the evaluation of organs such as the esophagus that may +explain or be associated with the patient’s symptoms. The radiologic + +GASTROINTESTINAL IMAGING + + 1734 October Special Issue 2021 + +radiographics.rsna.org + +TEACHING POINTS + „ Fluoroscopic examination of swallowing and modified barium +swallow should not be routine studies but should be custom- +tailored techniques designed by radiologists and swallowing +therapists on the basis of the patient’s clinical history and +symptoms. Radiologists should supervise or participate in the +procedures and contribute their knowledge to adapt the tech- +nique to unexpected findings detected during examinations. +They may decide during the study to include the evaluation of +organs such as the esophagus that may explain or be associ- +ated with the patient’s symptoms. + + „ For evaluation of the function of the PES, radiologists should +correlate the following functional units: (a) the opening of the +PES, (b) the elevation of the HLC, and (c) the pharyngeal con- +traction. Understanding the different possible mechanisms +that cause the abnormal function of the segment is essential +in deciding the best treatment strategies. + + „ As part of the fluoroscopic swallowing examination, a system- +atic evaluation of the air-filled structures and soft tissues of the +neck may provide information about the causative or associ- +ated abnormalities relating to the patient’s symptoms. + + „ The development of Zenker diverticula may involve anatomic +and pathophysiologic causes that include an area of wall +weakness, a distal obstruction due to the diminished opening +of the CP, and increased pharyngeal intraluminal pressure as a +consequence of mechanical obstruction of the outlet. + + „ The radiologic report should include all detected abnormali- +ties and should not be limited to description of the visualiza- +tion of penetration or aspiration. Because each abnormality +has its own specific therapeutic approach, the description +of radiologic findings and their correlation with the patho- +physiologic mechanisms and possible causes should be stated +clearly in the report. A meticulous radiologic diagnosis facili- +tates the decision making of referring physicians and swallow- +ing therapists. + +report should describe all the findings additional +to penetration or aspiration and present the facts +in a way that facilitates decision making by the +referring physicians and swallowing therapists. + +This article reviews the anatomy and physiol­ + +ogy of swallowing, explains the indications for +and techniques of fluoroscopic examination, +describes abnormal radiologic findings, and dis­ +cusses their pathophysiologic mechanisms. + +The discussion includes the abnormalities +of the pharyngoesophageal segment (PES) as +diminished or delayed opening and early clos­ +ing and the pathophysiology of the pharyngeal +residue after swallowing, penetration, and +aspiration. Complications of total laryngectomy +are examined, such as leakage, pseudoepiglot­ +tis, anterior pharyngeal pouches, neopharyngeal +dysmotilities, strictures, and malfunctioning +of the tracheoesophageal prosthesis for voice +reconstitution. We describe the radiologic find­ +ings of each abnormality and provide sample +images and a discussion of the possible mecha­ +nisms, with consideration of the patient’s clinical +history. + +Anatomy for Radiographic +Interpretation + +The oral cavity is bounded by the hard and soft +palates superiorly, by the tongue inferiorly, and +opens to the oropharynx posteriorly. The nasophar- +ynx or epipharynx is bounded by the pharyngeal +walls posteriorly and laterally, by the skull base su­ +periorly, and by the soft palate inferiorly. It opens +to the nasal cavity anteriorly and communicates +with the oropharynx inferiorly through the palato­ +pharyngeal isthmus or velopharyngeal seal. + +The oropharynx or mesopharynx is bounded +posteriorly and laterally by the pharyngeal walls +and constrictor muscles and inferiorly by the su­ +perior surface of the epiglottis and the hyoid bone. +It communicates with the nasopharynx superiorly +and with the mouth anteriorly. The laryngophar- +ynx or hypopharynx is bounded by the pharyngeal +walls posteriorly and laterally and extends to the +lower border of the cricoid cartilage inferiorly. It +communicates with the oropharynx superiorly, +with the cervical esophagus inferiorly, and with the +laryngeal vestibule anteriorly (3–6) (Fig 1). + +Phases of Swallowing + +Oral Preparatory Phase +The oral preparatory phase involves chewing the +food and mixing it with saliva. When the bolus +is considered “swallowable,” it is placed between +the upper surface of the tongue blade and the +palate. The tip of the bolus is held by a seal +formed by the tongue and the soft palate that +prevents premature leakage into the pharynx +before swallowing (7,8) (Fig 2A). + +Oral Propulsory Phase +During the oral propulsory phase, the bolus is pro­ +pelled into the oropharynx by means of an up­ +ward and backward movement of tongue. During +the passage of the bolus, the soft palate elevates +to a right angle to oppose the contraction of the +posterior and lateral pharyngeal walls that form +the Passavant ridge to close the palatopharyngeal +isthmus and prevent the penetration of material +into the nasopharynx (Fig 2B). + +Initiation of Pharyngeal Swallow +The next phase is the initiation of the pharyngeal +swallow, when the bolus passes the pharyngeal +pillars and enters into the oropharynx, triggering +the reflex that initiates the nonvoluntary pharyn­ +geal phase of swallowing. + +Pharyngeal Phase +In the pharyngeal phase, the hyolaryngeal complex +(HLC) elevates and shortens the pharynx, de­ +flects the epiglottis, closes the larynx, and opens + + RG • Volume 41 Number 6 + +Carbo et al 1735 + +the PES (Fig 2C). Laryngeal closure starts in the +true vocal folds, followed by the false vocal folds, +and ends in the laryngeal vestibule. Concurrently, +the stripping contraction wave of the posterior +and lateral pharyngeal walls begins and propels +the bolus into the esophagus. + +PES Opening Phase +The PES opening phase consists of (a) relaxation of +the upper esophageal sphincter (UES); (b) upward +and anterior traction of the cricoid cartilage, which +is attached to the HLC; and (c) opening of the +lumen by means of increased intrabolus pressure +by the thrust of the dorsum of the tongue and the +pharyngeal constrictors (Fig 2D) (9). + +Esophageal Phase +The esophageal phase begins when the bolus is +propelled through the esophagus into the stom­ +ach by means of esophageal peristalsis. Once +the bolus passes into the esophagus, (a) the +palatopharyngeal isthmus opens, reestablishing +the communication between the nasopharynx +and oropharynx and allowing the passage of air; +(b) the HLC returns to resting position; (c) the +larynx opens; (d) the epiglottis returns to upright +position; (e) the PES closes, and (f) respiration +resumes (Fig 2E). + +Indications for Fluoroscopic +Swallowing Examinations +Common indications for fluoroscopically guided +swallowing examinations are oropharyngeal +dysphagia (swallowing difficulty); globus sensa­ +tion; choking and coughing during swallowing; +prolonged intubation; cerebrovascular events (ie, +stroke); and surgery, trauma, or radiation treat­ +ments to the head and neck (10). + +Figure 1. +Illustration +shows the anatomy of +the pharynx. + +Fluoroscopic Swallowing Examination +Technique +The equipment used includes a radiologic appa­ +ratus with fluoroscopy and video recording with +slow motion, reverse, and stop­frame capabili­ +ties (2,10–13). The patient should be upright +or seated in a comfortable position, with their +dentures in their mouth. + +Lateral views are the most useful to evaluate +all phases of swallowing and to detect penetra­ +tions into the airway. Oblique views are per­ +formed to visualize the PES without the shoulder +in the way. Frontal views are used to diagnose +asymmetric swallowing abnormalities. + +Use of short exposure times and high kilovolt­ +age is recommended to improve image resolution +and minimize motion blurring of the bolus and +anatomic structures. For a functional examina­ +tion, the contrast material used is barium sulfate +in a concentration of 40% to 60% weight/volume +(mass of solute/volume of solution). Multiple +barium viscosities such as thin liquid, nectar +thick, honey thick, paste, and solids are used at +the speech­language pathologist’s discretion. +Iodinated water­soluble contrast material is used +when leakages or perforations are suspected (13). +Videofluoroscopic recording is used to detect + +functional abnormalities at 30 frames per sec­ +ond. Compared with videofluoroscopic record­ +ing, rapid­sequence digital images are acquired +at fewer frames per second, but the images have +better spatial resolution and are preferred for +anatomic details. + +The personnel included in a modified barium +swallow examination are a radiographic technolo­ +gist, a radiologist, and a speech­language patholo­ +gist. The role of the speech­language pathologist +is to use strategies that may improve the efficiency + + 1736 October Special Issue 2021 + +radiographics.rsna.org + +Figure 2. +Illustrations show the +phases of swallowing, which include +the oral preparatory phase (A), +the oral propulsory phase (B), the +pharyngeal phase (C), the PES +phase (D), and the esophageal +phase (E). + +and safety of swallowing, to educate patients about +their swallowing function, and to provide swallow­ +ing recommendations on the basis of the radio­ +graphic findings after study completion. + +Evaluation of esophageal anatomy and motility + +is important. The anatomic site that the patient +identifies as the source of the symptoms is not +a reliable indicator of the location of the cause. +Thus, radiologic examinations of patients with +dysphagia should include a morphologic and func­ +tional evaluation of the esophagus and esophago­ +gastric junction, which may be performed before +or after the swallowing examination. + +Functional Abnormalities + +Abnormalities of Oral Preparatory Phase +The motion of the tongue is essential for bolus +manipulation and the functions of chewing, mix­ +ing food with saliva, and holding the bolus in the +mouth before swallowing. When the tongue is + +dysfunctional, images may show uncoordinated +movements, fragmentation of the bolus, inability +to hold it, and drooling and leaking of contrast +material into the sulci of the cheeks and floor of +the mouth (Fig 3). Causes of tongue failure can +be secondary to atrophy, xerostomia, surgery, or +severe nerve or brain injuries (7,14,15). + +Premature Spillage from Mouth into +Pharynx +The back of the tongue blade and the soft pal­ +ate form the glossopalatal seal, which keeps the +bolus in the mouth before swallowing. Premature +spillage from the mouth into the pharynx may be +caused by unilateral or bilateral incompetence of +the seal, which may allow leakage of material into +the pharynx before a swallowing attempt, leading +to aspiration through an opened and unprotected +airway (Fig 4). Seal ineffectiveness can be a result +of weakness, atrophy, or poor muscular coordina­ +tion of the tongue and soft palate due to aging, + + RG • Volume 41 Number 6 + +Carbo et al 1737 + +Figure 3. Leakage into the floor of the mouth in a 77-year-old +woman. Lateral radiograph of the oral cavity after the swallow +shows that contrast material has leaked into the mouth floor +(large arrow) and the sulci of the cheeks (small arrows) because +of incompetence of the tongue. + +Figure 4. Leakage from the mouth into the pharynx in a +58-year-old woman. Lateral radiograph of the oral cavity and +pharynx before swallowing shows contrast material leaking +from the mouth bolus to the valleculae and hypopharynx (ar- +rows) due to incompetence of the glossopalatal seal. + +surgery, radiation, or acute and chronic neuro­ +logic illnesses. For example, patients with Parkin­ +son disease may perform repetitive unsuccessful +attempts to swallow the bolus (10,14). + +Delayed Initiation of Pharyngeal Swallow +When the bolus reaches the oropharynx, it +triggers the swallowing reflex that initiates the +nonvoluntary pharyngeal phase of swallowing. +The initiation of the pharyngeal swallow is rec­ +ognized radiographically by the elevation of the +hyoid bone as a component of the HLC. Hyoid +elevation starts when the tip of the bolus crosses +the angle of the mandible. A delayed swallowing +initiation is identified when the contrast material +is seen in the pharyngeal lumen but the hyoid +bone remains in a resting position (Fig 5). Causes +include decreased pharyngeal sensory perception +due to injury of mucosal receptors or neurologic +damage secondary to aging, radiation, a cerebro­ +vascular accident, or surgery (7,14). + +Nasal Regurgitation +The palatopharyngeal isthmus or seal should be +closed during the bolus transportation from the +mouth into the oropharynx. Nasal regurgitation +is diagnosed radiographically by evaluating pas­ +sage of the contrast material through a nonclosed +seal into the nasopharynx (Fig 6). Nasal penetra­ +tion can occur during transfer of the bolus from +the mouth to the pharynx due to incompetence +of the isthmus or after swallowing when the +seal is open and there is regurgitation from the +distal pharynx or esophagus due to an associated +distal dysfunction or mechanical obstruction. +The causes of an incompetent seal and nasal + +Figure 5. Delayed initiation of the pharyngeal swal- +low in a 55-year-old man. Lateral radiograph shows the +bolus in the oropharynx and hypopharynx (*), with the +hyoid bone still in resting position (arrow) due to a de- +layed triggering of the swallowing reflex. + +radiation, surgery, or a cerebrovascular accident. +Atrophy of the tongue or soft palate can lead +to compensatory hypertrophy of the opposite +muscle (10,14). + +Delayed or Slow Oral Propulsory Phase +When the bolus is judged to be “swallowable,” +it is placed in a midline depression in the upper +surface of the tongue blade until it is transferred +into the pharynx. A delayed initiation of tongue +propulsion or a slow transportation of the bolus +to the pharynx can be secondary to weakness or +the inability of the tongue to function because of + + 1738 October Special Issue 2021 + +radiographics.rsna.org + +Figure 8. Frozen pharynx in a +68-year-old woman with a his- +tory of radiation therapy. Lateral +radiograph shows absent contrac- +tions and rigidity of the pharyngeal +walls during the passage of bolus +and penetration of contrast mate- +rial into the larynx (arrow). + +Figure 6. Nasal regurgita- +tion in a 62-year-old man. +Lateral radiograph shows +contrast material penetrat- +ing into the nasopharynx +from the oropharynx (*) +during bolus displacement +because of insufficient clo- +sure of the palatopharyn- +geal isthmus. + +Figure 7. Limited elevation of +the hyolaryngeal complex in a +74-year-old man. Lateral radio- +graph during swallowing shows +that the hyoid bone (blue arrow) +did not reach the mandible at its +maximum displaced position, as is +expected in healthy patients. The +limited displacement may be due +to scarring of pharyngeal elevator +muscles from radiation treatment. +Associated findings are a poor +pharyngeal contraction, an incom- +plete closure of the larynx with +contrast material penetration (red +arrow), and a diminished opening +of the PES (yellow arrow). + +regurgitation include atrophy and insufficient +elevation of the soft palate due to neurologic +diseases, radiation, surgery, or congenital issues +such as a cleft palate or a fistula (7,14,16). An +oronasal fistula must be ruled out during the +examination. + +Slow or Incomplete Displacement +of HLC +A full upward and forward displacement of the +HLC is necessary to elevate and shorten the +pharynx, deflect the epiglottis, close the larynx, +and open the PES. The HLC is fully elevated +when the hyoid bone is seen at the level of the +inferior margin of the mandible. Slow or incom­ +plete elevation of the HLC is due to weakness of +the suprahyoid muscles and pharyngeal elevators +(Fig 7). Muscular weakness is a consequence of +atrophy and fibrosis secondary to aging, radia­ +tion, and neurologic or cerebrovascular diseases +(14). + +Weakness of Tongue and Pharyngeal +Contraction +Adequate thrust of the dorsum of the tongue +and the pharyngeal walls is required to increase +intrabolus pressure and propel the bolus into the + +Figure 9. Hypotony of the left +pharyngeal wall in a 59-year-old +woman. Frontal radiograph of the +pharynx during swallowing shows +bulging of the left pharyngeal wall +(arrowhead) and an asymmetric +tilt of the epiglottis (arrow). Find- +ings were due to a nerve injury +during surgery for resection of a +tumor in the base of the skull. + +esophagus. Unilateral or bilateral weakness or +atrophy of the muscles may prevent a timely and +complete bolus displacement. In severe cases +of atony, as with a “frozen pharynx,” the bolus +mostly moves by means of gravity (Fig 8). Lack +of appropriate contraction leaves residue in the +pharynx after swallowing and increases the risk +of aspiration. Reduced pharyngeal contraction +may be caused by muscular or neurologic dam­ +age due to radiation, surgery, or neurovascular +diseases. Weakness can be bilateral or unilateral +(Fig 9) (7,10,14,15). + + RG • Volume 41 Number 6 + +Carbo et al 1739 + +push the bolus and a distal relaxation to al­ +low its passage. At fluoroscopy, abnormalities +of pharyngeal peristalsis are recognized by +the presence of nonpropulsive contractions +that interrupt bolus passage and fragment the +contrast material in the pharyngeal lumen (Fig +10). Residue left in the pharyngeal lumen after +swallowing can lead to aspiration. The paralysis +can be bilateral because of injury of the pharyn­ +geal innervation that is secondary to radiation, +inflammatory changes, or neurologic diseases +or unilateral because of peripheral nerve injury +during surgery (7,14). + +Limited Tilt of Epiglottis +During swallowing, the tilt of the epiglottis pro­ +tects the entrance of the laryngeal vestibule and +deflects the bolus into the lateral food pathways +(pyriform sinuses). As a consequence, an insuf­ +ficient or absent epiglottic tilt may obstruct the +passage of the bolus and risk penetration of +material into the laryngeal inlet during swallow­ +ing (Fig 11). Causes of inadequate epiglottic tilt +include insufficient displacement of the HLC, +pharyngeal or laryngeal tumors, or fibrosis due to +radiation (7,14). + +Delayed or Incomplete Laryngeal Closure +Laryngeal closure starts with and continues as +HLC displacement. It starts in an inferior to +superior direction, beginning in the true vocal +folds, followed by the false vocal folds, and end­ +ing in the laryngeal vestibule to expel penetrated +material back into the hypopharynx. Closure of +the laryngeal vestibule is recognized at fluoros­ +copy as the narrowing and disappearance of the +air column in the laryngeal lumen during the +HLC elevation. A delayed, slow, or incomplete +closure of the larynx during swallowing may lead +to aspiration (Fig 12). Failure of laryngeal closure +may result from insufficient HLC elevation, cer­ +vical surgery, laryngeal fibrosis due to radiation, +or weakness of the supraglottic adductors due to +edema or fibrosis (14,17). + +Incomplete Opening of PES +The UES is a high­pressure zone at the PES that +is formed by muscular fibers of the thyropharyn­ +geus, cricopharyngeus (CP), and circular fibers +of the proximal esophagus. The PES is normally +closed to prevent inhaled air from passing into the +esophagus and gastroesophageal reflux from enter­ +ing into the airway. It opens during swallowing to +allow passage of the bolus, vomiting, or eructation. +Opening the PES involves (a) neural relaxation of +the tonically contracted UES muscles, (b) normal +distensibility of the segment, (c) traction forces +imparted by the opening of the UES muscles, and + +Figure 10. Nonpropulsive pharyngeal contrac- +tions in a 66-year-old man. Lateral radiograph of +the pharynx during swallowing shows narrowing +of the pharyngeal lumen (arrowheads) when the +hyoid bone is still elevated (blue arrow). Because +of premature pharyngeal closure, some contrast +material was trapped in the oropharynx (*) and +penetrated the laryngeal vestibule (red arrow). + +Incomplete epi- +Figure 11. +glottis tilt in a 72-year-old man. +Lateral radiograph shows a +swelling epiglottis in upward +position during the bolus pas- +sage (arrow). As a consequence +of the insufficient tilt of the epi- +glottis, a partial obstruction of +bolus displacement and pen- +etration of contrast material +into the laryngeal vestibule (*) +are visible. + +Figure 12. +Insufficient closure +of the laryngeal lumen during +swallowing in a 69-year-old +woman. Lateral radiograph of +the pharynx during swallow- +ing shows an opened laryngeal +vestibule with penetration of +contrast material. Additional +findings are deformity of the +epiglottis (*), poor pharyngeal +contractions, and a stricture of +the PES (arrow) due to chemo- +therapy and radiation therapy. + +Lack of Synchronization of Pharyngeal +Contractions +Peristalsis is a wave­like involuntary contraction +of the gastrointestinal tract with the purpose +of transporting the bolus. It consists of strip­ +ping contractions of the proximal muscles to + + 1740 October Special Issue 2021 + +radiographics.rsna.org + +Diagnosis and Treatment of Pharyngoesophageal Segment Dysmotilities + +PES Dysmotility + +PES Opening -1) +Mechanism -In the wake of recent protests and -demonstrations, 46 percent of young -people age 15-34 believe that people -like them can affect the government -at least a moderate amount, an -increase from 37 percent in the March -2018 poll. For 15-22 year olds, this rose -15 points, to 48 percent. +Therapeutic Objective -2) Nearly two-thirds of young people -age 15-26 and parents of this age -group discuss politics at least once a -month. Majorities say these -conversations are civil and that they -are at least sometimes able to change -others’ political views when -disagreements do occur. +Maneuvers -3) More than two-thirds across both +Delayed PES opening due +to slow HLC displace­ +ment -generations are disillusioned with the -two major political parties and feel a -third is needed. +Full opening only -© Copyright 2018. The Associated Press and NORC +when the HLC is +completely elevated -June 2018 1 +Delayed PES opening - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +Full opening only -Both young people age 15-26 and parents of this age group give the federal government a grade of F -more often than any other letter grade when assessing its performance on a wide variety of issues, -including gun control, the cost of higher education, foreign relations, immigration, environment and -climate change, health care, income inequality, taxes, sexual harassment and assault, abortion, racism, -and the treatment of citizens by police and the justice system. The economy and job creation fare -better, most commonly receiving grades of C from both groups. +Delayed initiation +of pharyngeal +swallow +UES spasm -This survey is the second in a multiwave series of polls highlighting the voices of the youngest -generation of voters age 15-34, with this installment including a sample of parents of 15-26 year olds to -allow for comparisons between generations. This nationwide poll by MTV and The AP-NORC Center -surveyed a total of 939 young people, including 148 teens age 15-17, 410 adults age 18-26, and 381 adults -age 27-34, as well as 607 parents of young people age 15-26. +when UES is relaxed -AFTER RECENT DEMONSTRATIONS, YOUNG PEOPLE FEEL MORE -EMPOWERED IN POLITICS. +Improve speed of swal­ +lowing initiation -Since the first MTV/AP-NORC Youth Political Pulse survey in March 2018, overall outlook among the -larger group of young people including those age 15-34 remains stable. Similar to the previous survey, -just under half are optimistic and a slight majority are pessimistic about the future of the United States. -Four in 10 have felt angry or anxious about the state of the country in the past month, while 1 in 10 -have felt excited or positive. +Bolus stimulation -Young peoples’ overall standpoint is fairly negative, but there is one area where young people have -grown somewhat more optimistic. After many recent protests and demonstrations, more now than in -March say elected officials care at least a moderate amount what people like them think (34 percent vs. -25 percent) and that people like them can affect the government at least a moderate amount (46 -percent vs. 37 percent). +Abbreviate the UES -Since March 2018, more young people feel efficacious in American government. +Increase bolus vol­ -t -n -e -c -r -e -P +relaxation -100 +ume and thickness -90 +due to impaired UES +relaxation -80 +Incomplete PES opening +due to nonrelaxation of +CP (CP achalasia) -70 +Incomplete PES opening +due to lack of elasticity +of the CP (fibrosis) +Incomplete PES opening +due to partial displace­ +ment of HLC -60 +Incomplete PES opening +due to weak pharyngeal +contraction -50 +Early PES closure due to +premature HLC return -40 +Early PES closure due +to premature UES +contraction -30 +Incomplete -20 +CP nonrelaxation -10 +Improve CP relaxation Endoscopic dilation +Botulinum toxin A -0 +Incomplete -37 +CP lack of elastic­ +ity (fibrosis) -25 +PES dilation -46 +Incomplete -34 +Weakness of HLC -March 2018 +Strengthening of HLC -May 2018 +elevators -Elected officials care at least a moderate amount +elevators -People like me can affect government at least a moderate amount +Incomplete -Questions: How much do elected public officials care what people like you think? -How much can people like you affect what the government does? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide +Weak pharyngeal +muscular con­ +tractions -Young people age 15-22 in particular are increasingly optimistic when it comes to political efficacy. In -March, just a third said that people like them can affect what the government does at least a moderate -amount. Now, 48 percent of those age 15-22 say the same. +Improvement of pha­ +ryngeal contraction -© Copyright 2018. The Associated Press and NORC +Full opening only -June 2018 2 +Weakness of HLC - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION +Increase and extend the -Young peoples’ priorities have also shifted. When asked to describe in their own words what issue -facing the country concerns them the most, 21 percent cited gun laws in March. Just 6 percent mention -gun issues now.1 The most common worries are now the economy (16 percent) and social inequality (14 -percent). +when the HLC is +completely elevated -BOTH YOUNG PEOPLE AND PARENTS SAY POLITICS HAVE BECOME -DYSFUNCTIONAL, AND BOTH ARE DISSATISFIED WITH THE TWO-PARTY -SYSTEM. +elevators -Young people age 15-26 and parents of young people age 15-26 alike have mixed outlooks on the future -of the United States. Young people are more pessimistic than parents (55 percent vs. 40 percent), -although these differences are mainly attributable to the parents being more Republican and less -racially and ethnically diverse than young people. +HLC elevation -Both generations report relatively negative feelings about the state of the country over the past month. -Substantial minorities of both groups have felt anxious or angry in the past month, while few have felt -positive or excited. Young people are more likely than parents to say they feel anxious (43 percent vs. -29 percent) and less likely to feel positive (8 percent vs. 17 percent) or excited (8 percent vs. 15 percent). +Full opening only -Young people age 15-26 are more anxious and less positive or excited about the state of America than -are parents of young people. +UES spasm -y +Delay the UES con­ -l -e -m -e -r -t -x -e +when the UES is +relaxed -r -o +traction -y -r -e -v -y -a -s +injection +CP myotomy +Effortful swallow +Endoscopic dilation +CP myotomy +Mendelsohn +Effortful swallow +Shaker +Base of tongue exer­ -o -h -w +cises -t -n -e -c -r -e -P +Tongue hold +Mendelsohn +Increase bolus vol­ -100 - -90 - -80 - -70 - -60 - -50 - -40 - -30 - -20 - -10 - -0 - -43 - -29 - -40 - -36 - -17 - -8 - -15 - -8 - -Anxious - -Angry - -Positive - -Excited - -Young people age 15-26 - -Parents of young people age 15-26 - -Question: In the past month… How _____ have you felt about the state of the country? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -Young people and parents have similar assessments of the political system and their own efficacy -within it. Majorities of both groups say American politics have been dysfunctional over the past -month. Sixty-four percent of young people and 67 percent of parents feel that politics have not been -functioning very well or at all. - -Young people see the current two-party system as inadequate, and parents agree. Less than a third -think the Republican and Democratic parties do an adequate job of representing the American people, -while more than two-thirds say a third party is needed. - -1 Interviews for this survey were conducted April 23-May 9, 2018, prior to the May 18 shooting in Santa Fe, Texas. - -© Copyright 2018. The Associated Press and NORC - -June 2018 3 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -YOUNG PEOPLE AND PARENTS ARE SIMILARLY ENGAGED WITH THE -MIDTERM ELECTIONS. - -When it comes to the upcoming midterm elections in November, parents of young people age 15-26 are -more likely than young people age 15-26 to be regularly reading or watching news about the elections -(66 percent vs. 45 percent). However, older adults tend to consume news more often than young -people in general2 and may be hearing more about the midterms as a result. - -The two generations are equally likely to engage with the November elections in other ways, such as -posting or commenting about them on social media (25 percent of parents vs. 25 percent of young -people age 15-26), sharing political memes (18 percent of parents vs. 24 percent of young people age 15- -26), participating in events (14 percent of parents vs. 16 percent of young people age 15-26), or -volunteering with a candidate or issue campaign (11 percent of parents vs. 9 percent of young people -age 15-26). - -BOTH GENERATIONS HAVE VARIED POLITICAL CONCERNS AND ARE -STRONGLY DISAPPOINTED WITH THE GOVERNMENT’S PERFORMANCE -ON KEY ISSUES. - -Economic issues top long lists of current concerns for both young people and parents. When asked to -say in their own words what concerns them most about the issues currently facing the country, 15 -percent of young people age 15-26 and 17 percent of parents of young people age 15-26 say they are -most concerned about the economy. The two groups place similar priorities on a range of other issues, -although social inequality ranks second highest for young people (13 percent) and fails to make the top -five most worrisome issues for parents (6 percent). - -2 The Media Insight Project. 2014. The Personal News Cycle. -http://www.apnorc.org/PDFs/Media%20Insight/News%20Cycle/The_Personal_News_Cycle_Final.pdf - -© Copyright 2018. The Associated Press and NORC - -June 2018 4 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Young people and parents report a wide range of top concerns of issues facing the country. - -The economy - -15 - -17 - -Social inequality - -War or threat of nuclear war - -13 - -6 - -11 -10 - -Immigration - -Gun laws - -Other issues - -Political climate - -Concerns about President Trump, leadership - -Foreign policy - -The environment - -Health care - -Terrorism, homeland security, national security - -Crime and safety - -Education - -Nothing - -Everything - -7 -7 - -7 - -4 - -6 - -9 - -5 - -8 - -5 - -7 - -5 -6 - -4 -5 - -3 - -6 - -2 - -4 - -2 -3 - -2 -2 - -1 -0 - -0 -1 - -0 - -10 - -20 - -30 - -40 - -50 - -60 - -70 - -80 - -90 - -100 - -Percent who offered each response - -Young people age 15-26 - -Parents of young people age 15-26 - -Question: Of all of the issues facing the country right now, what concerns you the most? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -Young people age 15-26 believe the federal government is doing a poor job addressing a majority of key -issues. They most commonly give the government a failing grade for 15 of 17 key issues. However, -young people rate performance on the economy and job creation as average, most commonly giving a -grade of C. - -© Copyright 2018. The Associated Press and NORC - -June 2018 5 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Parents most commonly give the same grades as young people age 15-26 on most issues, with a few -exceptions. They give Cs rather than Fs for LGBT rights, the quality of education, terrorism and -homeland security, and taxes. - -Young people age 15-26 and parents give low marks to the federal government on its handling of many -issues. - -Young People -Age 15-26 - -Parents of Young -People Age 15-26 - -The environment and climate change - -Gun control - -Racism - -Immigration - -Cost of higher education - -Treatment of citizens by police and justice system - -Abortion - -Income inequality - -Sexual harassment and assault - -The quality of education - -Taxes - -Foreign relations - -Health care policy - -The rights of LGBT Americans - -Terrorism and homeland security - -The economy - -Job creation - -A - -B - -C - -D - -F A - -B - -C - -D - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -F - - - - - - - - - - - - - - - - - - - - - - - - Indicates the most commonly given grade, or mode. The frequency of each grade is indicated with darker shading for -more common responses, according to the key below. - -0% - -50% - -Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of -the following issues? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -PARENT/CHILD DISCUSSIONS ABOUT POLITICS ARE USUALLY CORDIAL, -BUT IT CAN BE DIFFICULT TO SWAY THE OTHER GENERATION WHEN -VIEWS DIFFER. - -Young people age 15-26 and their parents have fairly regular discussions with one another about -politics and social issues, and few see these discussions as a source of conflict. Across the two groups, -64 percent of young people and parents say they discuss politics with each other at least once a -month, 20 percent say they have these discussions less often, and 16 percent say they never talk about -politics or social issues together. - -Among those who discuss political and social issues with their parents or children, a majority say they -usually see eye to eye in these conversations, and about a third say they debate diplomatically. Fewer -say the discussion turns into a major conflict or requires changing the subject to avoid arguments. - -© Copyright 2018. The Associated Press and NORC - -June 2018 6 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Most parents and their children see eye to eye on politics and social issues. - -Someone changes the -subject quickly to avoid -getting into disagreements. -9% - -It turns into -World War III. -5% - -We debate things -diplomatically. -32% - -We usually see -eye to eye. -53% - -Question: When you talk with [him/her/them] about political and social issues, how would you describe how the discussion -goes? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -When the two generations were asked about political discussions with a particular family member, -parents’ accounts are similar regardless of the specific relationship, such as mother-daughter or father- -son. Sons reporting on discussions with fathers are more likely than other pairings to say they talk -about politics at least once a week, but least likely to say these discussions tend to go smoothly. Only -42 percent of sons say they usually see eye to eye with their fathers, while 49 percent of sons say the -same about their mothers, 54 percent of daughters say so about their fathers, and 62 percent say so -about their mothers. - -Those who identify as political independents (36 percent) are less likely than Democrats (60 percent) -or Republicans (53 percent) to say they usually see eye to eye in parent/child discussions about -political and social issues. - -When asked about potential differences in opinion on specific issues, both generations report high -levels of agreement. In their discussions, at least 6 in 10 say there is agreement when discussing each -of 18 topics. Agreement levels are highest on Trump’s job performance and on the issues of racism, -sexual harassment and assault, and health care policy, and lowest on LGBT rights, abortion, and -foreign relations. - -While agreement is high, there also seems to be a fair amount of ambiguity about family members’ -political positions. On 13 of the 18 issues asked about, at least 15 percent say they are unsure if they -tend to agree or disagree. - -© Copyright 2018. The Associated Press and NORC - -June 2018 7 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Young people and their parents tend to be on the same page with many political and social issues. - -President Trump's job performance - -Racism - -Sexual harassment and assault - -Health care policy - -The quality of education - -Treatment of citizens by police and justice system - -Cost of higher education - -The economy - -Taxes - -The environment and climate change - -Income inequality - -Immigration - -Job creation - -Terrorism and homeland security - -Gun control - -The rights of LGBT Americans - -Foreign relations - -Abortion - -76 - -76 - -76 - -75 - -74 - -74 - -73 - -72 - -71 - -71 - -70 - -70 - -69 - -69 - -69 - -62 - -62 - -61 - -13 - -8 - -9 - -10 - -9 - -8 - -13 - -15 - -13 - -9 - -8 - -11 - -9 - -11 - -17 - -19 - -13 - -16 - -8 - -9 - -14 - -17 - -15 - -13 - -19 - -19 - -19 - -20 - -23 - -17 - -22 - -20 - -14 - -19 - -24 - -22 - -Tend to agree - -Tend to disagree - -Not sure - -Questions: When talking about the following issues, do you tend to agree or disagree with [him/her/them]? -When talking with [your family member] about Donald Trump and how he’s handling his job as president, do you tend to -agree or disagree with [him/her/them]? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -The two generations report different issues as the biggest sources of disagreement. Among those who -say they generally don’t see eye to eye when talking about political and social issues, young people -age 15-26 say they tend to disagree with their parents most about racism (37 percent), LGBT rights (36 -percent), and how Trump is handling his presidency (33 percent). Parents of young people age 15-26 -report that the largest sources of disagreement are gun control (27 percent), LGBT rights (25 percent), -and how Trump is handling his presidency (21 percent). - -When there are disagreements, few young people and parents are able to persuade one another to -change their views, and the findings are consistent between the generations. Eleven percent overall -say they are always or often able to persuade their family member to change their views. Fifty-three -percent acknowledge that they are sometimes able to persuade their family member. Just 6 percent -say the reverse is true, that they are often or always convinced to change their views by their family -member. Another 44 percent say their family member is sometimes able to persuade them to change -their own views. - -© Copyright 2018. The Associated Press and NORC - -June 2018 8 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Few young people or parents find it easy to persuade one another to change their political and social -views. - -How often you are able to persuade your - family member to change their views - -11 - -53 - -36 - -How often your family member is able - to persuade you to change your views - -6 - -44 - -49 - -0 - -10 - -20 - -30 - -40 - -50 - -60 - -70 - -80 - -90 - -100 - -Percent - -Always or often - -Sometimes - -Rarely or never - -Questions: In general, when you disagree with [him/her/that person] about political or social issues, how often would you -say that you’re able to persuade [him/her/them] to change [his/her/their views]? -In general, when you disagree with [him/her/that person] about political or social issues, how often would you say that [he -is/she is/they are] able to change your views? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -YOUNG PEOPLE ARE PERCEIVED AS MORE IDEALISTIC THAN THE OLDER -GENERATION, AND PARENTS FEEL THEIR CHILDREN’S GENERATION IS -LESS INFORMED THAN THEIR OWN. - -When asked to compare the generations, young people age 15-26 and parents of this age group tend to -agree on who is the more patriotic and rigid generation, but there is some disagreement about who is -more informed. - -About two-thirds of both young people and parents say the older generation is more patriotic. More -than half of each group feel the older generation is more rigid. - -Among both groups, the largest proportion view the parents’ generation as more politically active and -moral than the younger generation, and the younger generation as more idealistic. - -Young people are divided on whether their generation is more uninformed, their parents’ generation is -more uninformed, or whether the two are about equal. Parents believe the younger generation is more -uninformed than their own. - -© Copyright 2018. The Associated Press and NORC - -June 2018 9 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Young people and parents agree that the older generation is more patriotic and more rigid. - -Which generation is more…? - -Patriotic - -Moral - -Rigid - -Young people -say… - -Parents say… - -Young people -say… - -Parents say… - -Young people -say… - -Parents say… - -Compassionate - -Young people -say… - -Parents say… - -Politically active - -Young people -say… - -Idealistic - -Uninformed - -Parents say… - -Young people -say… - -Parents say… - -Young people -say… - -Parents say… - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -0 - -64 - -10 - -25 - -20 - -40 - -60 - -80 - -100 - -65 - -7 - -27 - -20 - -44 - -20 - -20 - -20 - -20 - -53 - -57 - -52 - -40 - -60 - -16 - -40 - -40 - -40 - -40 - -60 - -13 - -60 - -60 - -19 - -60 - -28 - -34 - -40 - -60 - -23 - -40 - -40 - -60 - -33 - -60 - -80 - -39 - -80 - -33 - -80 - -100 - -100 - -100 - -17 - -25 - -80 - -100 - -28 - -80 - -37 - -80 - -40 - -80 - -100 - -100 - -100 - -25 - -80 - -100 - -32 - -28 - -40 - -60 - -80 - -100 - -30 - -40 - -28 - -23 - -32 - -19 - -20 - -20 - -20 - -20 - -40 - -60 - -80 - -100 - -47 - -30 - -40 - -60 - -32 - -40 - -43 - -40 - -60 - -60 - -80 - -34 - -80 - -37 - -80 - -100 - -100 - -100 - -20 - -38 - -20 - -41 - -20 - -37 - -20 - -Parents' generation - -Young people's generation - -Both about equally - -Question: Thinking about people in your generation, compared to people in [his/her/their] generation, do you think the -following words describe your generation better, describe [his/her/their] generation better, or describe both generations -about equally well? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -© Copyright 2018. The Associated Press and NORC - -June 2018 10 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Republicans are more likely than Democrats to describe their own generation as patriotic (51 percent -vs. 32 percent) and are less likely to describe their own generation as idealistic (27 percent vs. 39 -percent). - -FOR SOME, POLITICS SEEPS INTO INTERACTIONS WITH EXTENDED -FAMILY. - -A significant minority of both generations have had political disagreements with relatives that have -led to conflict. Twenty percent have unfriended or blocked a family member on social media due to a -disagreement over politics, and 21 percent have themselves been unfriended or blocked by a family -member. Fewer than 1 in 5 have had a holiday gathering ruined (11 percent), have decided not to attend -a family event (12 percent), or have had a family member decide to not attend a family event (17 -percent) due to political disagreements. - -While there are no differences by generation, Democrats are more likely than Republicans to have had -political disagreements with relatives affect their social media behavior. Democrats are more likely -than Republicans to say they’ve unfriended or blocked a family member on social media (26 percent -vs. 12 percent) and to have had a family member unfriend or block them on social media (27 percent vs. -15 percent). - -EVEN AMONG THOSE WHO APPROVE OF TRUMP, THE REPORT CARD IS -MEDIOCRE. - -Young people age 15-34 give poor marks to the federal government on its handling of many public -policy issues. Even among Trump’s supporters, the report card is still only average. Young people who -approve of Trump’s job performance most commonly give the federal government a C for all issues -except terrorism and homeland security and treatment of citizens by the police and justice system, -which earn a B, and abortion, which earns an F. For those who disapprove of Trump, F is the most -common grade for all issues except job creation, which most commonly gets graded a C. - -Most common grades for the federal government’s handling of issues. - -Approve of Trump - -Disapprove of Trump - -Gun control - -Cost of higher education - -Foreign relations - -Immigration - -The environment and climate change - -Health care policy - -Income inequality - -The economy - -Taxes - -Sexual harassment and assault - -Abortion - -Racism - -The rights of LGBT Americans - -Terrorism and homeland security - -Treatment of citizens by police and justice system - -The quality of education - -Job creation - -C - -C - -C - -C - -C - -C - -C - -C - -C - -C - -F - -C - -C - -B - -B - -C - -C - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -F - -C - -Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of -the following issues? -Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide - -© Copyright 2018. The Associated Press and NORC - -June 2018 11 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Looking at grades by political party, the results are similar. Democrats’ most common grade across -most issues is an F. However, Democrats give better marks to the federal government’s efforts on job -creation, most commonly grading it a C. - -Republicans are slightly more mixed, but most commonly give issues a C. For abortion, they most -commonly give a grade of F. - -STUDY METHODOLOGY - -This survey was conducted by The Associated Press-NORC Center for Public Affairs Research and -with funding from MTV and NORC at the University of Chicago. Staff from NORC at the University of -Chicago, The Associated Press, and MTV collaborated on all aspects of the study. - -Data were collected using AmeriSpeak®, NORC’s probability-based panel designed to be -representative of the U.S. household population. During the initial recruitment phase of the panel, -randomly selected U.S. households were sampled with a known, non-zero probability of selection from -the NORC National Sample Frame and then contacted by U.S. mail, email, telephone, and field -interviewers (face-to-face). The panel provides sample coverage of approximately 97 percent of the U.S. -household population. Those excluded from the sample include people with P.O. Box only addresses, -some addresses not listed in the USPS Delivery Sequence File, and some newly constructed dwellings. - -Interviews for this survey were conducted between April 23 and May 9, 2018, with young people age -15-34 and parents of young people age 15-26 representing the 50 states and the District of Columbia. -Interviews were conducted in both English and Spanish, depending on respondent preference. Adult -panel members age 18-34 were randomly drawn from AmeriSpeak, and 791 completed the survey—780 -via the web and 11 via telephone. Teen panel members age 15-17 were drawn from AmeriSpeak Teen, -and 148 completed the survey—145 via the web and 3 via telephone. Adult panel members who are -known parents of 15-17 year olds were randomly drawn from AmeriSpeak, and 104 completed the -survey—101 via the web and 3 via telephone. In addition, panel members age 30-76 were randomly -drawn to screen for parents of young adults age 18-26, and 503 completed the survey—469 via the web -and 34 via telephone. - -Among young people age 15-34, the final stage completion rate is 25.9 percent, the weighted household -panel response rate is 33.7 percent, and the weighted household panel retention rate is 87.2 percent, for -a cumulative response rate of 7.6 percent. The overall margin of sampling error among young people -age 15-34 is +/- 4.3 percentage points at the 95 percent confidence level, including the design effect. The -overall margin of sampling error among young people age 15-26 is +/- 5.6 percentage points at the 95 -percent confidence level, including the design effect. The margin of sampling error may be higher for -subgroups. - -Among parents of young people age 15-26, the screener completion rate is 27.9 percent, with an -incidence rate of 25.2. The interview completion rate is 95.6 percent, and the final stage completion rate -is 26.7 percent. The weighted household panel response rate is 33.7 percent, and the weighted -household panel retention rate is 87.2 percent, for a cumulative response rate of 7.8 percent. The -overall margin of sampling error is +/- 6.7 percentage points at the 95 percent confidence level, -including the design effect. The margin of sampling error may be higher for subgroups. - -Once the sample has been selected and fielded, and all the study data have been collected and made -final, a poststratification process is used to adjust for any survey nonresponse as well as any -noncoverage or under- and oversampling resulting from the study specific sample design. -Poststratification variables included age, gender, census division, race/ethnicity, and education. - -© Copyright 2018. The Associated Press and NORC - -June 2018 12 - - MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION - -Weighting variables were obtained from the 2017 Current Population Survey. The weighted data -reflect the U.S. population of young people age 15-34 and parents of young people age 15-26. - -All analyses were conducted using STATA (version 14), which allows for adjustment of standard errors -for complex sample designs. All differences reported between subgroups of the U.S. population are at -the 95 percent level of statistical significance, meaning that there is only a 5 percent (or less) -probability that the observed differences could be attributed to chance variation in sampling. - -A comprehensive listing of all study questions, complete with tabulations of top-level results for each -question, is available on The AP-NORC Center for Public Affairs Research website: www.apnorc.org. - -For more information, email info@apnorc.org. - -CONTRIBUTING RESEARCHERS - -From NORC at the University of Chicago - -From The Associated Press - -Becky Reimer -Emily Alvarez -Liz Kantor -Trevor Tompson -Jennifer Benz - -Emily Swanson - -ABOUT THE ASSOCIATED PRESS-NORC CENTER FOR PUBLIC AFFAIRS -RESEARCH - -The AP-NORC Center for Public Affairs Research taps into the power of social science research and the -highest-quality journalism to bring key information to people across the nation and throughout the -world. - -■ - -The Associated Press (AP) is the world’s essential news organization, bringing fast, unbiased news -to all media platforms and formats. - -■ NORC at the University of Chicago is one of the oldest and most respected, independent research - -institutions in the world. - -The two organizations have established The AP-NORC Center for Public Affairs Research to conduct, -analyze, and distribute social science research in the public interest on newsworthy topics, and to use -the power of journalism to tell the stories that research reveals. - -The founding principles of The AP-NORC Center include a mandate to carefully preserve and protect -the scientific integrity and objectivity of NORC and the journalistic independence of AP. All work -conducted by the Center conforms to the highest levels of scientific integrity to prevent any real or -perceived bias in the research. All of the work of the Center is subject to review by its advisory -committee to help ensure it meets these standards. The Center will publicize the results of all studies -and make all datasets and study documentation available to scholars and the public. - -ABOUT MTV - -MTV is a global youth culture brand inspired by music. For more information, check out -mtvpress.com. MTV is a unit of Viacom Inc. (NASDAQ: VIAB, VIA) - -© Copyright 2018. The Associated Press and NORC - -June 2018 13 +ume and thickness; +Mendelsohn +Increase bolus vol­ + +ume and thickness + +PES dilation + +Sources.— References 9,10,14,18–26. + +(d) intrabolus pressure generated by contraction +of the pharynx. + +Primary causes of incomplete opening of the +UES during the passage of the bolus are (a) in­ +complete relaxation of the UES (achalasia) or lack +of coordination with the pharyngeal contraction +(dyskinesia), (b) abnormal elasticity and dis­ +tensibility of the CP as result of degeneration of +muscle fibers and replacement with fibroadipose +tissue, (c) noncomplete displacement of the HLC +during swallowing, and (d) insufficient intrabolus +pressure to open the lumen due to impairment of +the tongue and pharyngeal constrictor muscles +(Table) (7,9,14,18–26). Narrowing of the PES +prevents the free and complete passage of the +bolus into the esophagus, resulting in incomplete +clearance and residue in the pyriform sinuses, +with potentially postdeglutitive aspiration. + +On a lateral view, the diminished opening of the + +PES is recognized by a posterior protrusion (CP +bar) between C3 and C6 during the trans­sphinc­ +teric passage of the bolus (Fig 13). For evaluation +of the function of the PES, radiologists should +correlate the following functional units: (a) the +opening of the PES, (b) the elevation of the HLC, +and (c) the pharyngeal contraction. Understand­ +ing the different possible mechanisms that cause + +the abnormal function of the segment is essential +in deciding the best treatment strategies. + +Delayed Opening of PES +During swallowing, the PES opens completely +upon the arrival of the bolus to allow it to pass +into the esophagus. Delayed opening of the PES +is a consequence of delayed elevation of the HLC +or retarded relaxation of the UES (Table). When +the delayed opening is due to slow or retarded +relaxation of the UES, the hyoid bone is already +elevated when the PES is still closed (Fig 14). +Because the PES remains closed on the arrival of +the bolus, there is a transient luminal obstruction +that augments the pressure and distends the lu­ +men of the proximal pharynx, increasing the risk +of aspiration (9,14). + +Early Closure of PES +The PES should only close after the bolus has +passed completely into the esophagus. A prema­ +ture closure of the PES can be due to an early +contraction of the UES before the HLC returns +to resting position and the bolus pass is finished +(Fig 15) or a premature return of the HLC to +resting position (Fig 16) (Table). An early closure +of the PES can block the bolus and leave residue + + RG • Volume 41 Number 6 + +Carbo et al 1741 + +Figure 13. Diminished opening of the PES in +a 44-year-old woman with gastroesophageal re- +flux. Lateral radiograph of swallowing shows a +transverse filling defect in the posterior wall of +the PES, called the CP bar (yellow arrow), when +the hyolaryngeal complex is fully elevated (blue +arrow). The luminal narrowing obstructs the pas- +sage of the bolus and distends the proximal pha- +ryngeal lumen. + +Figure 14. Delayed opening +of the PES due to retarded relax- +ation of the UES in a 71-year-old +woman. Lateral radiograph shows +the bolus tip at the PES entrance, +which remains closed (yellow ar- +row), even when the hyoid bone +is fully elevated (blue arrow). On +subsequent images (not shown), +the PES opened completely, allow- +ing the full passage of the bolus. + +Figure 15. Early closure of the PES due to +premature closing of the UES in a 62-year- +old man. (A) Lateral radiograph shows a +normal PES opening during the passage of +the bolus. (B) Delayed lateral radiograph +shows an early closing of the PES (arrow- +heads), while the hyoid bone is still ele- +vated (arrow in A and B). The early closure +of the PES interrupts the passage of the +bolus, leaving residue in the hypopharynx +(*), with increased risk of aspiration. + +in the pyriform sinuses after swallowing, leading +to aspiration when the airway reopens (9,14). + +Residue in Valleculae and Pyriform +Sinuses after Swallowing +After swallowing, no materials or contrast ma­ +terial should remain in the pharyngeal lumen. +Residue in the vallecullae, pyriform sinuses, +or pharyngeal walls may lead to postswallow­ +ing aspiration when respiration is resumed (Fig +17A–17C). After swallowing, pooling of contrast +material in the valleculae can result from impaired +backward movement of the dorsum of the tongue +or insufficient epiglottic tilt. Residue in the pyri­ +form sinuses may result from impaired pharyngeal +contraction, PES dysmotilities, or mechanical +obstruction. Unilateral pooling can be secondary +to paralysis of the lateral pharyngeal wall due to a +nerve or muscle injury (Fig 17D) (14). + +Penetration.—No contrast material should +penetrate the laryngeal vestibule before, during, +or after swallowing. Penetration is diagnosed at +fluoroscopy when contrast material is seen in the +laryngeal lumen. During swallowing, penetrated +material should be extruded back into the phar­ +ynx by laryngeal contraction (ie, flash penetration). +In severe cases, contrast material may remain +in the laryngeal vestibule, reach the vocal folds, +and pass into the trachea. Whenever possible, the +mechanism responsible for penetration should be +recognized during the examination and stated in +the report to allow decision making on the best +therapeutic approach. Preswallowing penetrations +are usually a consequence of delayed initiation +of the swallowing reflex. Penetrations during +swallowing can be due to one or multiple abnor­ +malities such as incomplete or slow elevation of +the HLC, epiglottis tilt, incomplete closure of the + + 1742 October Special Issue 2021 + +radiographics.rsna.org + +Figure 16. Early closure of the PES due to premature +return of the hyolaryngeal complex in a 55-year-old +woman. (A) Lateral radiograph shows normal open- +ing of the PES during swallowing, with the hyoid bone +fully elevated (blue arrow). An incidental finding is a +penetration during swallowing (red arrow). (B) De- +layed lateral radiograph of the same swallow shows +an early closure of the PES (arrowhead) due to a pre- +mature return of the HLC, which is evidenced by the +low position of the hyoid bone (blue arrow). The early +closure of the PES interrupted the complete passage +of the bolus and left residue in the hypopharynx (*). + +Figure 17. Residue in the valleculae and pyriform sinuses after swallowing. (A) Lateral radiograph of the pharynx in a 72-year- +old man after swallowing shows contrast material pooling in the valleculae (V) and pyriform sinuses (PS). (B) Frontal radiograph +in a 40-year-old man after spinal fixation shows bilateral residue in the valleculae (green arrows) and pyriform sinuses (red arrows). +(C) Lateral radiograph in a 67-year-old woman after swallowing shows the penetration of residual contrast material (arrowhead) into +the larynx from the pyriform sinuses (*). (D) Frontal radiograph in a 72-year-old man after right-sided skull base surgery shows residue +in the right pyriform sinus (arrow) due to unilateral deficiency of the pharyngeal contraction. + +laryngeal vestibule, or motility disorders of the +pharynx or PES. Pharyngeal residue is respon­ +sible for penetrations after swallowing (14). + +Aspiration.—Aspiration is diagnosed when con­ +trast material is seen in the tracheal lumen, below +the vocal cords (Fig 18). Aspiration is a more +severe complication than penetration and may be +minimal, moderate, or massive when the con­ +trast material reaches the carina and bronchus. +Aspirated materials in the trachea may generate +a protective cough to expel them back into the +pharynx or may be nonsensate (ie, silent). A de­ +layed or absent cough reflex is valuable informa­ +tion by which to estimate severity, because silent +aspirators have a greater risk of developing pneu­ +monia. In addition to the causes of penetration, +a delayed or insufficient closure of the vocal folds +due to edema, radiation fibrosis, or surgery of the +vocal cord adductors may contribute to aspira­ +tion. External­beam radiation treatment for neck + +cancer is a frequent cause of aspiration because it +may impair all phases of swallowing. + +Anatomic Abnormalities + +Radiography of the Neck +As part of the fluoroscopic swallowing exami­ +nation, a systematic evaluation of the air­filled +structures and soft tissues of the neck may +provide information about the causative or as­ +sociated abnormalities relating to the patient’s +symptoms. The patient should be evaluated for +(a) soft palate atrophy due to radiation or sur­ +gery or compensatory hypertrophy due to tongue +atrophy (Fig E1); (b) tongue atrophy due to +aging, radiation, or surgery (Fig E2); (c) epiglot­ +tis enlargement due to acute radiation (Fig E3), +infection, and tumors or atrophy due to radia­ +tion; (d) swelling of prevertebral soft tissues due +to chemotherapy, radiation, or surgery (Fig E4) +and tumors, abscesses, or hematomas; (e) air + + RG • Volume 41 Number 6 + +Carbo et al 1743 + +Figure 18. Aspiration in a 77-year-old man. +Lateral view of swallowing shows contrast ma- +terial penetration into the laryngeal vestibule (*) +and aspiration into the tracheal lumen (red ar- +row) through the vocal folds (white arrows). + +Figure 19. Web in a 66-year-old +woman. Lateral radiograph of the +pharynx during swallowing shows a +thin horizontal linear filling defect in +the anterior wall of the hypopharynx +that obstructs the lumen (arrow). + +Figure 21. Pharyngeal tumor in +a 78-year-old man. Swallowing ex- +amination shows an irregular lumi- +nal narrowing of the pharynx, with +posterior displacement due to a large +infiltrating mass (arrows). + +Figure 20. Radiation stricture in +a 77-year-old man. Lateral radio- +graph of swallowing shows a short +severe luminal stricture in the phar- +ynx due to scarring (arrow), with a +mild proximal luminal distention. + +in the prevertebral soft tissues due to trauma +(Fig E5), gastrointestinal perforations (Fig E6), +diverticula (Fig E7), or in other locations as +postsurgical leakage (Fig E8); (f) soft­tissue­ +attenuation obstructive masses in the lumen of +pharynx (Fig E9), larynx (Fig E10), or PES (Fig +E11); (g) pharyngeal deformities as a nonclosed +PES at rest due to fibrosis because of radiation +therapy (Fig E12) or esophageal dilations due to +achalasia (Fig E13) or scleroderma; (h) abnor­ +mal soft­tissue calcifications; and (i) bone and +postsurgical abnormalities. + +Pharyngeal Strictures +Webs are thin mucosal folds that are frequently +seen in the anterior wall or circumference of the +hypopharynx or proximal esophagus. Webs are +associated with gastroesophageal reflux and syn­ +dromes such as Plummer­Vinson or epidermolysis +ampullosa and are seen as 2–4­mm­thick trans­ +verse linear filling defects in the hypopharynx and +cervical esophagus that may partially obstruct the +lumen and passage of the bolus (Fig 19) (27). + +Radiation strictures are a delayed complication + +of radiation due to vascular damage, ischemia, +and collagen deposition, with subsequent fibrosis +and scarring (Fig 20) (27,28). + +Cervical esophageal tumors can be seen radio­ +graphically as irregular or lobulated filling defects +protruding into the lumen of the pharynx or +larynx on radiographs or swallowing examina­ +tions (Fig 21) (27). Cervical tumors are more ac­ +curately detected and staged with cross­sectional +imaging such as CT, MRI, or PET. + + 1744 October Special Issue 2021 + +radiographics.rsna.org + +Figure 22. Zenker diverticulum in two patients. (A) Illustration shows a diverticulum (green arrow) +emerging from a triangular-shaped area in the posterior wall of the hypopharynx (blue and yellow ar- +rows), which was formed superiorly by the oblique fibers of the thyropharyngeus and inferiorly by hori- +zontal fibers of the cricopharyngeus. (B) Lateral radiograph during swallowing in a 74-year-old woman +shows a diverticulum in the posterior wall of the hypopharynx (red-orange arrow) above the CP bar +(yellow arrow). (C) Frontal radiograph in a 79-year-old man after swallowing contrast material shows +the diverticulum mildly displaced to the left side of the patient (*) and residue in both pyriform sinuses +(arrows). + +Zenker Diverticula +Zenker diverticula are outpouchings lined by squa­ +mous epithelial mucosa and submucosa and are +often surrounded by fibrous tissue, with absent +muscular fibers. They are located in an area of +muscular weakness called the Killian dehiscence +or the triangle of Laennec in the posterior wall of +the hypopharynx and are limited by the oblique +fibers of the thyropharyngeous superiorly and +laterally and by the horizontal fibers of the CP +inferiorly (Fig 22A). + +The development of Zenker diverticula may +involve anatomic and pathophysiologic causes that +include an area of wall weakness, a distal obstruc­ +tion due to the diminished opening of the CP, +and increased pharyngeal intraluminal pressure as +a consequence of mechanical obstruction of the +outlet. Strong evidence now shows that acquired +Zenker diverticula are, in most cases, second­ +ary to a poorly compliant but normally relax­ +ing CP muscle that cannot fully distend during +the process of sphincter opening. The increased +resistance to flow during swallowing in an attempt +to overcome the distal obstruction markedly in­ +creases the forces required to drive the passage of +the bolus. Increased intrabolus pressure imparted +to the area of relative muscular weakness (Killian +dehiscence) over many years gives rise to a poste­ +rior herniation of the pouch (27,29,30). + +Fluoroscopically, the diverticulum is seen +during and after swallowing as an outpouching + +arising from the posterior hypopharyngeal wall +above the prominence of the CP bar (Fig 22B). +In frontal views, a small diverticulum is seen in +the midline, but when it becomes larger, it may +relocate laterally, more frequently to the left side +(Fig 22C). After swallowing, the content of the +diverticulum may empty back into the hypo­ +pharynx, increasing the risk of postswallowing +aspiration. + +Pseudo-Zenker diverticula are saclike collections +of contrast material that are trapped in the same +location as that of the Zenker diverticula (27). +They are seen only after swallowing when the +pharyngeal lumen is collapsed and not when the +lumen is fully distended (Fig 23). + +Killian-Jamieson Diverticula +Killian-Jamieson diverticula are outpouchings from +the anterior lateral wall of the cervical esophagus +through a gap between the inferior margin of the +cricoid cartilage and CP and the lateral wall of +the suspensory ligament of the esophagus (Fig +24A) (27). They are seen during swallowing stud­ +ies as lateral or anterior outpouchings below the +level of the CP (Fig 24B). + +Complications of Total Laryngectomy + +Surgical Technique +Total laryngectomy is indicated for advanced or +nonradiosensitive laryngeal or cervical tumors. + + RG • Volume 41 Number 6 + +Carbo et al 1745 + +Figure 23. Pseudo-Zenker diverticulum in a +69-year-old woman. Lateral radiograph of the +pharynx after swallowing shows an outpouching +in the location of the Zenker diverticula (arrow). +The outpouching was not identified when the +pharyngeal lumen was fully distended during the +passage of the bolus (not shown). + +Figure 24. Killian-Jamieson diverticulum in a 72-year-old +man. (A) Illustration shows a diverticulum emerging from the +anterolateral wall of the cervical esophagus (arrow) below the +CP muscle. (B) Frontal radiograph during swallowing shows a +right lateral diverticulum (arrow). + +The surgical technique consists of removal of the +hyoid bone, thyroid and cricoid cartilages, epiglot­ +tis, and vocal folds (Fig 25A), followed by recon­ +struction and closure of the anterior defect with +mucosa, pharyngeal muscles, and skin (Fig 25B). +Finally, optional vocal restoration can be achieved +through creation of a fistula between the trache­ +ostoma and the esophagus, with placement of a +silastic prosthesis of varying length and diameter +in the fistula to allow the one­way passage of air +for speech rehabilitation (Fig 25C). In patients +with the tracheoesophageal fistula, the evaluation +of prosthesis position and function and PES motil­ +ity during phonation should be included as part of +the swallowing evaluation (31–33). + +Anastomotic Leakage +Anastomotic leakage is a frequent complication +in the early postoperative period. Risk factors +include oral infection, salvage total laryngec­ +tomy, and prior radiation therapy. Leakages are +detected at a swallowing examination as blind +outpouchings outside the neopharyngeal lumen +(sinus tracts) or as an extraluminal tract of con­ +trast material that communicates with the skin +(Fig 26) (27,34,35). + +Pseudoepiglottis +A pseudoepiglottis is an enlargement of the folds at +the base of the tongue due to the pull of pharyn­ +geal constrictors when a patient attempts to swal­ +low. The abnormality is seen as an anterior linear +defect at the base of tongue and the entrance of + +neopharynx that resembles an epiglottis (Fig 27). +The fold is more prominent when the patient +swallows and may narrow the entrance of the +neopharynx and obstruct the bolus transit from +the mouth. Increased intrabolus pressure proxi­ +mal to the obstruction contributes to the forma­ +tion of a proximal pocket, where residue and food +may accumulate (27). + +Anterior Pharyngeal Pouches +Anterior pharyngeal pouches are pockets or out­ +pouchings located at the anterior wall between +the base of tongue and the neopharynx entrance. +The mechanism of formation is multifactorial, +including a localized weakness due to absence +of a muscular layer and/or increased intrabolus +pressure against a more distal mechanical or +functional obstruction. Radiologic images show +an anterior outpouching at the entrance of the +neopharynx with contrast material pooling after +the swallow (Fig 28). If the pouch enlarges con­ +siderably, it may extend into the skin, creating a +pharyngo­cutaneous fistula (36). + +Neopharyngeal Dysmotilities +Neopharyngeal dysmotilities may be secondary to +resection of the laryngeal and pharyngeal struc­ +tures that support the tongue and pharyngeal +muscles and/or due to damage of the pharyngeal +muscles and nerves by the tumor, surgery, or +radiation therapy. Dysmotilities are seen on im­ +ages from swallowing studies as a limited thrust +of the base of tongue and/or pharyngeal walls, + + 1746 October Special Issue 2021 + +radiographics.rsna.org + +Figure 25. Surgical technique for laryngectomy. (A) Illus- +tration shows the anatomic structures that will be removed +as the hyoid bone, epiglottis, thyroid cartilage including the +vocal folds, cricoid cartilage, and first tracheal ring. (B) Illus- +tration shows the closure of the anterior wall with mucosa, +muscles, and skin for the reconstruction of the neopharynx. +(C) Illustration shows the creation of a tracheoesophageal +fistula, with placement of a one-way prosthesis (arrow) for +voice reconstitution. + +with weakening of the pharyngeal propulsion, +impaired coordination of peristalsis, reversed +peristalsis, and outflow resistance across the +PES (Fig 29). Because of neopharyngeal dys­ +motility, bolus transit time is prolonged, and +pharyngeal clearance is incomplete, with stasis +and residue (27,36). + +total + +Figure 26. Sinus tract in +a 78-year-old man. Lateral +radiograph of swallowing +after +laryngectomy +shows a leakage from the +proximal neopharynx into +the anterior soft tissues of +the neck (arrow). + +Dysfunctions of PES +Dysfunctions of the PES should be evaluated +during and after swallowing during voice gen­ +eration. Fluoroscopically, they are seen as an +insufficient or delayed opening or early closing +of the PES (Fig 30). These findings may be a +consequence of impaired pharyngeal propul­ +sion or increased resistance at the bolus passage +through the PES due to resection of the cri­ +coid cartilage or fibrosis secondary to radiation +therapy (27,36). + +Benign Strictures +Benign strictures of the neopharyngeal lumen are +secondary to surgical scarring and/or chemo­ + +therapy and radiation therapy. The strictures may +partially or completely obstruct the bolus flow, +distending the proximal lumen and retaining ma­ +terial the patient swallows (Fig 31) (37). + + RG • Volume 41 Number 6 + +Carbo et al 1747 + +Figure 27. Pseudoepiglot- +tis in a 78-year-old man who +underwent a total laryngec- +tomy. Lateral radiograph dur- +ing swallowing shows a linear +horizontal filling defect in the +anterior wall of the proximal +neopharynx that resembles +the epiglottis and partially +obstructs the passage of the +bolus (arrow). + +Figure +28. Anterior +pharyngeal pouch in a +72-year-old woman. Lat- +eral radiograph shows +contrast material pooling +in an outpouching at the +base of the tongue at the +entrance of the neophar- +ynx (*). + +Figure 29. Neopharyngeal dysmotility in a 70-year-old man. Se- +quential lateral radiographs of swallowing show a premature clo- +sure of the neopharynx during the passage of the bolus (arrows in +B), with residue in the proximal lumen. + +Figure 30. PES dysmotility in a +72-year-old woman. Lateral radio- +graph of swallowing shows PES +narrowing during the passage of +the bolus (arrow). + +Complications of Tracheoesophageal +Fistula and Prosthesis +A malpositioned or a too­large prosthesis may +protrude into the lumen of the tracheostoma and +obstruct it (Fig 32A). When the prosthetic flange +is located in the esophageal lumen instead of being +attached to the wall, the flange may cause obstruc­ +tion and interference in the passage of the bolus +and in phonation (Fig 32B). A too­short prosthesis +may lead to closure of the fistula, with associated +inflammatory changes. Esophageal contents can +pass around the prosthesis (ie, periprosthetic) and +into the tracheostoma during swallowing, because +the diameter of the prosthesis is too narrow com­ +pared with that of the stoma (Fig 32C) or through +the prosthesis because of malfunction (Fig 32D). +The prosthesis should be removed and replaced +after diagnosis of these complications (38–41). + +Conclusion +Swallowing is a complex mechanism that involves +at least 37 pairs of striated muscles, five cranial +nerves, and multiple supportive anatomic struc­ +tures that are controlled by the central nervous +system (10). As we have discussed in this article, +numerous anatomic and functional abnormali­ +ties can impair swallowing and can be detected at +fluoroscopy. + +The presence or supervision of a radiologist +during the examination may contribute to cor­ +relation of the perceived abnormalities with the +patient’s symptoms and prior treatments and +imaging studies. On the basis of this preliminary +analysis, the radiologist can decide, in conjunc­ +tion with the speech­language pathologist, if ad­ +ditional images or structures should be included +in the examination before the patient leaves the +radiography department. + + 1748 October Special Issue 2021 + +radiographics.rsna.org + +Figure 31. Postradiation stric- +ture in a 75-year-old man. Lateral +radiograph of swallowing after +total laryngectomy shows a stric- +ture in the neopharynx (arrows), +with proximal pharyngeal dilata- +tion (*). + +Figure 32. Complications of a tracheoesophageal prosthesis in four patients. (A) Oblique radiograph of swallowing in an 81-year- +old woman shows migration of the prosthesis (red arrows) into the lumen of the tracheostoma (green double-headed arrow). +(B) Oblique radiograph in a 74-year-old man after swallowing shows the esophageal flange of the prosthesis (arrows), which is +malpositioned in the lumen of the proximal esophagus instead of being attached to the wall. (C) Oblique radiograph in a 79-year- +old patient after swallowing shows contrast material leaking from the esophagus into the tracheal lumen through the surrounding +cephalad margin of the prosthesis (arrows). (D) Oblique radiograph in a 72-year-old man shows contrast material dripping into the +trachea through the lumen of prosthesis (arrow). + +The radiologic report should include all de­ +tected abnormalities and should not be limited to +description of the visualization of penetration or +aspiration. Because each abnormality has its own +specific therapeutic approach, the description +of radiologic findings and their correlation with +the pathophysiologic mechanisms and possible +causes should be stated clearly in the report. A +meticulous radiologic diagnosis facilitates the +decision making of referring physicians and swal­ +lowing therapists. + +Information obtained from swallowing exami­ +nations is useful to determine the oral capabilities +of the patient and whether they should initiate, +maintain, or discontinue oral intake without risk­ +ing development of pneumonia. A good­quality +examination and report not only improve early +diagnosis and treatment of a patient’s swallowing +disorders but also reduce morbidity, mortality, +length of hospital stay, and health care costs. + +Acknowledgment.—The authors wish to acknowledge Lory +Tubbs, MA, for her contribution with the illustrations. + +References + +1. Allen J, Belafsky P. Symptom indices for dysphagia assess­ +ment and management. In: Shaker R, Belafsky P, Postma +GN, Easterling C, eds. Principles of deglutition: a multidis­ +ciplinary text for swallowing and its disorders. New York, +NY: Springer Science Business Media, 2013; 357–379. +2. Logeman JA. Instrumental techniques for the study of swal­ +lowing. In: Evaluation and treatment of swallowing disorders: +imaging studies. 2nd ed. Austin, Tex: Pro­ed, 1998; 54–61. +3. Rubesin SE. Pharyngeal morphology. In: Ekberg O, ed. +Dysphagia. Diagnosis and treatment. 2nd ed. Cham, Swit­ +zerland: Springer International, 2019; 351–384. + +4. Ekberg O, Nylander G. Anatomy and physiology. In: Ekberg +O, ed. Dysphagia. Diagnosis and treatment. 2nd ed. Cham, +Switzerland: Springer, 2019; 3–20. + +5. Logeman JA. Anatomy and physiology of normal deglutition. +In: Evaluation and treatment of swallowing disorders. 2nd +ed. Austin, Tex: Pro­ed, 1998; 13–52. + +6. Belafsky PC, Lintzenich CR. Development, Anatomy, +and Physiology of the Pharynx. In: Shaker R, Belafsky P, +Postma GN, Easterling C, eds. Principles of deglutition: a +multidisciplinary text of swallowing and its disorders. New +York, NY: Springer, 2013; 165–173. + +7. Jones B. Abnormalities of pharyngeal function. In: Gore RM, +Levine MS, eds. Textbook of Gastrointestinal Radiology. +4th ed. Philadelphia, Pa: Saunders­Elsevier, 2014; 222–236. +8. Matsuo K, Palmer JB. Oral phase preparation and pro­ +pulsion: anatomy, physiology, rheology, mastication, and +transport. In: Shaker R, Belafsky P, Postma GN, Easterling + + RG • Volume 41 Number 6 + +Carbo et al 1749 + +C, eds. Principles of deglutition. A multidisciplinary text +of swallowing and its disorders. New York, NY: Springer, +2013; 117–132. + +9. Lang IM. Development, anatomy and physiology of the +upper esophageal sphincter and pharyngoesophageal junc­ +tion. In: Shaker R, Belafsky P, Postma GN, Easterling C, +eds. Principles of deglutition: a multidisciplinary text for +swallowing and its disorders. New York, NY: Springer, +2013; 235–255. + +10. Rubesin SE. Principles of performing a “modified barium +swallow” examination. In: RSNA Categorical Course in +Diagnostic Radiology: Gastrointestinal. Oak Brook, Ill: +Radiological Society of North America, 1997; 7–19. +11. Logeman JA. Evaluation of swallowing disorders. In: Evalu­ +ation and treatment of swallowing disorders. 2nd ed. Austin, +Tex: Pro­ed, 1998; 168–190. + +12. Rubesin SE. Pharynx: normal anatomy and examination +techniques. In: Gore RM, Levine MS, eds. Textbook +of Gastrointestinal radiology. 4th ed. Philadelphia, Pa: +Saunders­Elsevier, 2014; 207–221. + +13. Skucas J. Imging contrast agents. In: Gore RM, Levine +MS, eds. Textbook of Gastrointestinal Radiology. 3rd ed. +Philadelphia, Pa: Saunders­Elsevier, 2008; 3–17. + +14. Logeman JA. Disorders of deglutition. In: Evaluation and +treatment of swallowing disorders. 2nd ed. Austin, Tex: +Pro­ed, 1998; 71–119. + +15. Pokieser P, Scharitzer M. The clinical and radiological +approach to dysphagia. In: Ekberg O, ed. Dysphagia. Diag­ +nosis and treatment. 2nd ed. Cham, Switzerland: Springer +International, 2019; 285–315. + +16. Buchholz DW, Bosma JF, Donner MW. Adaptation, com­ +pensation, and decompensation of the pharyngeal swallow. +Gastrointest Radiol 1985;10(3):235–239. + +17. Lazarus C. Dysphagia Secondary to the Effects of Chemo­ +therapy and Radiation therapy. In: Shaker R, Belafsky P, +Postma GN, Easterling C, eds. Principles of deglutition. A +multidisciplinary text of swallowing and its disorders. New +York, NY: Springer, 2013; 431–444. + +18. Dantas RO. Cricopharyngeal bar. In: Shaker R, Belafsky P, +Postma GN, Easterling C, eds. Principles of deglutition. A +multidisciplinary text of swallowing and its disorders. New +York, NY: Springer, 2013; 509–514. + +19. Massey BT. Crocopharyngeal achalasia. In: Shaker R, +Belafsky P, Postma GN, Easterling C, eds. Principles of +deglutition. A multidisciplinary text of swallowing and its +disorders. New York, NY: Springer, 2013; 515–528. +20. Easterling C, Shaker R. Upper esophageal sphincter opening +muscle dysfunction. In: Shaker R, Belafsky P, Postma GN, +Easterling C, eds. Principles of deglutition. A multidisci­ +plinary text of swallowing and its disorders. New York, NY: +Springer, 2013; 529–538. + +21. Mei L, Sanvanson P. Upper esophageal sphincter restrictive +disorders. In: Bardan E, Shaker R, eds. Gastrointestinal mo­ +tility disorders. Cham, Switzerland: Springer, 2018; 79–907. +22. Lazarus CL, Logemann JA, Pauloski BR, et al. Swallow­ +ing disorders in head and neck cancer patients treated with +radiotherapy and adjuvant chemotherapy. Laryngoscope +1996;106(9 Pt 1):1157–1166. + +23. Born LJ, Harned RH, Rikkers LF, Pfeiffer RF, Quigley +EM. Cricopharyngeal dysfunction in Parkinson’s disease: +role in dysphagia and response to myotomy. Mov Disord +1996;11(1):53–58. + +24. Theurer JA, Johnston JL, Taves DH, et al. Swallowing after +right hemisphere stroke: oral versus pharyngeal deficits. Can +J Speech Lang Pathol Audiol 2008;32:114–122. + +25. Lazarus C, Logemann JA. Swallowing disorders in +closed head trauma patients. Arch Phys Med Rehabil +1987;68(2):79–84. + +26. McCullough GH, Kim Y. Effects of the Mendelsohn +maneuver on extent of hyoid movement and UES opening +post­stroke. Dysphagia 2013;28(4):511–519. + +27. Rubesin SE. Structural abnormalities of the pharynx. In: +Gore RM, Levine MS, eds. Textbook of gastrointestinal +radiology. 4th ed. Philadelphia, Pa: Saunders­Elsevier, +2014; 235–265. + +28. Lazarus C. Dysphagia secondary to the effects of chemo­ +therapy and radiation therapy. In: Shaker R, Belafsky P, +Postma GN, Easterling C, eds. Principles of deglutition: a +multidisciplinary text for swallowing and its disorders. New +York, NY: Springer Science Business Media, 2013; 431–443. +29. Cook IJ. Zenker’s diverticulum. In: Shaker R, Belafsky P, +Postma GN, Easterling C, eds. Principles of deglutition: a +multidisciplinary text for swallowing and its disorders. New +York, NY: Springer Science Business Media, 2013; 495–508. +30. Morrison MP, Postma GN. Surgical treatment of Zenker’s +diverticulum. In: Shaker R, Belafsky P, Postma GN, Easter­ +ling C, eds. Principles of deglutition: a multidisciplinary text +for swallowing and its disorders. New York, NY: Springer +Science Business Media, 2013; 847–858. + +31. Agrawal N, Goldenberg D. Primary and salvage total laryngec­ +tomy. Otolaryngol Clin North Am 2008;41(4):771–780, vii. +32. Schindler A, Mozzanic F, Barbiera F. Dysphagia evaluation +and treatment after head and neck surgery and/or chemo­ +radiotherapy of head and neck malignancy. In: Shaker R, +Belafsky P, Postma GN, Easterling C, eds. Principles of +deglutition: a multidisciplinary text for swallowing and +its disorders. New York, NY: Springer Science Business +Media, 2013; 649–668. + +33. Giannitto C, Preda L, Zurlo V, et al. Swallowing disorders +after oral cavity and pharyngolaryngeal surgery and role +of imaging. Gastroenterol Res Pract 2017;2017:7592034. +34. Wippold FJ II. Postoperative pharynx. In: Gore RM, Levine +MS, eds. In: Textbook of gastrointestinal radiology. 3rd +ed. Philadelphia, Pa: Saunders­Elsevier, 2008; 257–271. +35. Balfe DM, Koehler RE, Setzen M, Weyman PJ, Baron RL, +Ogura JH. Barium examination of the esophagus after total +laryngectomy. Radiology 1982;143(2):501–508. + +36. Anderson S, Hogan D, Panizza B. Causes and consequences +of anterior pharyngeal pouch after total laryngectomy. J +Laryngol Otol 2014;128(Suppl 2):S39–S42. + +37. Nguyen NP, Smith HJ, Sallah S. Evaluation and manage­ +ment of swallowing dysfunction following chemoradiation +for head and neck cancer. Curr Opin Otolaryngol Head +Neck Surg 2007;15(2):130–133. + +38. Logemann JA. Swallowing disorders after treatment of oral +and oropharyngeal cancer. In: Evaluation and treatment of +swallowing disorders. 2nd ed. Austin, Tex: Pro­Ed, 1998; +251–278. + +39. Pope TL, Shaffer HA Jr, Johns ME, et al. Radiologic evalu­ +ation of the Blom­Singer prosthesis for vocal restoration +after laryngectomy. Presented at the Annual Meeting of +the American Society of Head and Neck Radiology, Los +Angeles, CA, May 1981. + +40. Pauloski BR, Rademaker AW, Logemann JA, Colangelo +LA. Speech and swallowing in irradiated and nonirradiated +postsurgical oral cancer patients. Otolaryngol Head Neck +Surg 1998;118(5):616–624. + +41. Zhang T, Szczesniak M, Maclean J, et al. Biomechanics of +pharyngeal deglutitive function following total laryngectomy. +Otolaryngol Head Neck Surg 2016;155(2):295–302. + +This journal-based SA-CME activity has been approved for AMA PRA Category 1 CreditTM. See rsna.org/learning-center-rg. diff --git a/q110/random_k2/question.json b/q110/random_k2/question.json index 5ecb4c20f6fe1f735e5a700b2eee0c633a3b4da4..f271701cb0485fc57d36158143327961c3dbf2eb 100644 --- a/q110/random_k2/question.json +++ b/q110/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "earthlinkweb-150213112111-conversion-gate02_95.pdf", - "web_ca39b8721f6d1255.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf" + "resize_r2_0.pdf", + "web_88f0578d817bd13f.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q110/random_k2/random_1.md b/q110/random_k2/random_1.md index 6bfd67b67b9eb8baf6562a416c94e48f3b69ec3d..052cee2fff6d2fb0526b1d43400b3c01f2587dfb 100644 Binary files a/q110/random_k2/random_1.md and b/q110/random_k2/random_1.md differ diff --git a/q110/random_k2/random_2.md b/q110/random_k2/random_2.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..ebd6ecb8692e8722f7ba60c4670378f06f3bd010 100644 --- a/q110/random_k2/random_2.md +++ b/q110/random_k2/random_2.md @@ -1,1453 +1,613 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA +The Step-by-Step Guide +to Recruiting Top Talent + +Your blueprint to transform your talent acquisition practices + to attract the people you need. + + Table of Contents + +01 + +The Race for Talent + +02 + +Does Your Talent Acquisition +Process Need a Tune-up? + +03 + +5 Factors That Affect Your Talent +Acquisition Efforts + +04 + +The Step-by-Step Guide to +Recruiting Top Candidates + +page 3 + +page 4 + +page 5 + +page 7 + +BONUS + +05 + +10 Must-Haves of a Great +Recruitment Management System +page 11 + +06 + +Find and Keep +the Talent You Need + +page 12 + +2 + +BOOKMARKS +Click on the title to to be +directed to that page. + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. The Race for Talent + +01 + +It’s a job seeker’s market. + +High employment rates, an aging workforce, and +skills gaps in both technical and soft skills have put +employees in the driver’s seat. + +What this means for you is that you’re going to +have to fight harder than ever to recruit and retain +the people you need. + +You’re going to have to leverage technology, lean into +your values, improve your employer branding, and in +some cases take a sledgehammer to your current talent +acquisition strategies to create engaging and responsive +recruitment experiences that not only attract candidates +to your company but compel them to choose you over +the competition. + +It can be a daunting task… but we’ve got you covered +-as HCM consultants and SAP® SuccessFactors® +specialists, we’ve helped countless businesses do +exactly this. + +This eBook is the first step in the process. It includes the +step-by-step guide to recruiting the best talent. Use it to +evaluate where you stand and build a talent acquisition +process that gives you the workforce you need for a +successful future. + +Let’s get started. + +An overwhelming +majority of candidates +(69%) refuse to +re-apply to a company +if they have a poor +experience during the +application process. + +3 + +1 2016 Talent Board NAM CandE Research Report FINAL 170202.pdf +link: http://www.thetalentboard.org/wp-content/uploads/2017/02/2016_Talent_Board_NAM_Can- + +dE_Research_Report_FINAL_170202.pdf + +69%© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. + Does Your Talent Acquisition +Process Need a Tune-up? + +02 + +Let’s look under the hood at your talent acquisition strategy +and see what’s really there. Here are five key questions that +you need to answer to identify the roadblocks between where +you are, and where you want to be. + +01 + +02 + +03 + +04 + +05 + +Is your recruiting based on +current best practices? + +Do you have buy-in +from your team? + +Is your technology up to date +and fully cloud capable? + +HR trends are evolving more +rapidly now than ever before. +Make sure you’re not falling +behind the competition when +it comes to implementing best +practices. + +Is your team aligned internally? +Do they share a common lens? +Great candidate experiences +depend on the interactions +candidates have with your +team during the various stages +of the acquisition process. +Buy-in from your team is vital. + +If you haven’t moved to the +cloud yet, it is never too late. +If you’ve migrated but aren’t +satisfied with your current +system, it can be salvaged. +The right tech really does +make a difference. + +Are you always looking +for ways to innovate your +recruiting approach? + +Is your recruitment +management system (RCM) +making your job easier? + +Technology is playing an +increasingly crucial role in the +talent acquisition process. You +need to always have an eye on +the horizon, ear to the ground, +and willingness to adapt +and innovate your approach. +Remember: good ideas can +come from anywhere. + +The right tools should help +to increase efficiency. But +sometimes, the tool itself +sucks up your time. And that’s +frustrating. Make sure your +system and processes make +your job easier. That’s the most +valuable success metric. + +4 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 5 Factors That Affect +Your Talent Acquisition Efforts + +03 + +The first step to building an effective acquisition strategy is to know +what internal and external factors affect your recruitment efforts. + +Based on our experience and a study of current market conditions, here’s +5 factors that you need to be aware of when building your strategy. + +You need to determine how you’ll adapt, counter, and respond to +each of these. + +5 + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 01 + +GROWING DEMAND & +DWINDLING SUPPLY + +02 + +TURNOVER & +BURNOUT + +03 + +INCONSISTENT +HIRING PROCESSES + +04 + +INTENSE FOCUS ON +COST MANAGEMENT + +05 + +EVOLVING +EXPECTATIONS + +With just about every major +industry facing staffing +shortages, the candidates are +now in the driver’s seat with the +option of picking an employer +that responds to their needs +in a location of their choosing. +While talent acquisition and +human resources teams can’t +change the growing supply/ +demand imbalance, they need +to be aware of how this affects +their acquisition strategy and +take active steps to attract +Gen X, Millennials, and +pretty soon Gen Z before the +competition does. + +The staffing shortage also has +an adverse effect on current +professionals who have to +work longer hours leading to +burnout and turnover – which +leads to a dangerous loop of +an exhausted workforce, an +ever-growing pile of work, and +more turnover. Furthermore, an +organization with a reputation +for high turnover and burnout +creates additional challenges +for recruitment teams when it +comes to hiring the best talent. + +Modern businesses often have +complex org charts, siloed +branches, and evolving parts. +This creates inconsistent hiring +processes that vary from +department to department, +region to region, and result in +poor candidate experiences +that negatively affect the +organization’s brand image. +While the HR team can’t +control everything, cloud-based +comprehensive talent solutions +such as SAP SuccessFactors can +help companies centralize their +processes and bring visibility to +internal teams and candidates. + +Companies are always under +pressure to control costs. Up +against this mentality, HR +managers struggle to secure +funding for talent management +initiatives and, as a result, +end up using disparate, on +premise or outdated recruitment +technologies. This leads to error +prone processes due to the +doubling of data across systems, +labor intensive and slow hiring +steps, and a disconnected +candidate experience. + +The largest generation in the +workforce today is tech-savvy +and socially aware Millennials. +As candidates, Millennials +look for companies that +align with their values, offer +training and development, +and deliver opportunities for +career growth. They expect +their future employers to +be digitally connected with +an intuitive website that +showcases its brand and +culture. They want a simple, +mobile friendly job application +process, transparent and timely +communication, and an active +social media presence so they +can interact with the company +during the hiring process. + +6 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. The Step-by-Step Guide to +Recruiting Top Candidates + +04 + +Building an amazing team is just as important as building a great +customer base. But, hiring the cream of the crop isn’t as easy as it +used to be. In the race for talent, the candidate experience matters +more now than ever before! This step-by-step guide is the blueprint +you need to transform your recruitment processes to meet the +expectations of a truly multigenerational digital workforce. + +7 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 1 + +2 + +EVALUATE YOUR HIRING PROCESS +FROM START TO FINISH + +Start by evaluating your current hiring process to make +sure you know what your candidate experience is +truly like - we would even go so far to suggest secret- +shopping the hiring process. + +Next, storyboard your hiring process and map out +outcomes. The goal here is to plot the set of experiences +a candidate could have and optimize every step. + +Use the data insights to create repeatable and +consistent hiring processes that offer positive +experiences. + +Always make sure your hiring processes align +with and reflect your values and culture. + +TIP + 8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -i -C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | -B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) -M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +CREATE A STRONG, ATTRACTIVE +BRAND THAT SHOWS YOUR VALUE +TO TOP TALENT + +Your brand is your most valuable asset, when it comes to +recruiting talent. + +The first step in creating an attractive brand is developing +your Employee Value Proposition - a good EVP outlines +your organization’s mission and values and commitment +to employees. + +Once you have your EVP, the next step is to make sure +your career site not only reflects but showcases your EVP +- much like the buyer’s journey, the hiring journey now +begins with an online search. + +Lastly, make sure you have a strong content strategy +as it will help you build your brand identity and ensure +candidates find and connect with you online. + +Great content is relevant, trendy, and +provides value. We find setting up an +editorial board helps ensure you’re regularly +and consistently posting good content. + +TIP + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. < Continued from previous page + +3 + +BUILD A TALENT PIPELINE AND NURTURE TALENT + +There’s no such thing as “Just-in-Time Recruiting”. + +Smart recruiters know the importance of building and nurturing a talent pipeline. +To build a strong pipeline: + +01 + +02 + +Source talent globally +through an omni-channel +approach. Connect with and +reach out to candidates on +job boards, social platforms +and in-person events. + +Create a network of +communities, content, +and contact points so that +you’re constantly nurturing +candidates and making sure +your brand is top of mind. + +03 + +04 + +Make it simple for candidates +to get in touch and follow +up with HR by leveraging +automation, chatbots, and +other AI technology. + +05 + +Use technology to track +candidate progress and +development benchmarks. + +Take every opportunity +to share your culture and +events with candidates by +sharing different types of +company news and not just +job updates. + +Encourage your +organization’s +leaders to be +active online, +sharing stories +and updates, and +engaging with +candidates. + +TIP + +9 + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. < Continued from previous page + +4 + +5 + +ENHANCE THE CANDIDATE EXPERIENCE + +It’s time to end the candidate feedback black hole. + +According to the 2016 North American Talent Board +Candidate Experience research report, 47 percent of +candidates never receive any communication up to two +months after applying for an open position!2 + +Give yourself an edge over the competition by +leveraging your RCM to not only keep your internal +team on the same page but also track and +communicate with candidates regularly. + +Additionally, make sure your career site is always up to +date with current openings and that it showcases your +culture through employee testimonials. + +CREATE AND SHARE +A CONSISTENT MESSAGE + +The last step is less of a one-time only step and more so +an ongoing process. + +Once you’ve evaluated, streamlined, and automated +your hiring processes, developed your EVP, and built +out your pipeline strategies, you need to make sure your +organization continues to create and share consistent +content and messages with your online and offline +communities. + +Companies that constantly connect and engage passive +talent will have greater success at sourcing and hiring +top talent. + +Don’t forget to make your application +process mobile friendly. + +TIP + +10 + +Tell your story. Engage your current +workforce and ask them to share +testimonials about what a job with your +organization is really like. Candidates are +attracted to authentic messages driven by +the workforce, not mandated by leadership. + +TIP + +2 2016 Talent Board NAM CandE Research Report FINAL 170202.pdf + link: http://www.thetalentboard.org/wp-content/uploads/2017/02/2016_ +Talent_Board_NAM_CandE_Research_Report_FINAL_170202.pdf + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. BONUS + +10 Must-Haves of a Great +Recruitment Management System + +05 + +There’s no denying that technology has changed the talent acquisition process – But, how do you know if the +technology you’re currently using is delivering what you need to build a motivated, high performing workforce? + +Use this cheat-sheet of the 10 must-haves to evaluate your system. + +Seamless cloud integration – easily access candidate and +employee information from anywhere. + +A simplified workflow for recruiters & HR – make the hiring +process easier for top talent and your HR team to navigate. + +Standardized, optimized processes & communication – +with consistent processes in place, you’ll reduce error and +dramatically improve communications. + +Omni-channel job distribution – advertise open positions +across sites and social channels to increase job exposure. + + Customized, mobile-friendly landing pages & web portals – +make the hiring process faster and easier. + +06 + +07 + +08 + +09 + +10 + +Support for single and multi-stage applicant flows – make it +simpler to hire regardless of job complexity. + +A candidate-first experience – cater to the people you want +most and build a consistent pipeline for top talent. + +Empowered hiring managers – put the right information in the +hands of decisions makers. + +A smooth transition from candidate to employee – get new +employees up and running faster, saving time and costs. + +Full visibility through analytics – make data-driven hiring +decisions and track candidates from start to finish. + +01 + +02 + +03 + +04 + +05 + +11 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. Find and Keep +the Talent You Need + +Finding the right talent is key to growing your business. A major part of recruiting +success is the tools you use. Simplifying your data by replacing disparate, on- +premise systems with a unified, cloud-based HR system that provides real-time +data is a great start. + +Our certified SAP SuccessFactors professionals can help you master your new +recruiting system. But most importantly, we help you align that system to your +overall HR strategy and your company culture. Our goal is to help you transform +your business so you can find and attract great talent. + +Contact us today to find out how we can help you deliver a recruiting +solution designed to let your organization’s recruiting efforts thrive. + +12 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. © 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. This document is provided for information purposes +only, and the contents are subject to change without notice. This document is not warranted to be error-free, nor subject to +any other warranties or conditions, whether expressed orally or implied in law, including implied warranties and conditions +of merchantability or fitness for a particular purpose. We specifically disclaim any liability with respect to this document, and +no contractual obligations are formed either directly or indirectly by this document. This document may not be reproduced +or transmitted in any form or by any means, electronic or mechanical, for any purpose, without our prior written permission. +Rizing, Rizing HCM, and other Rizing products and services mentioned herein as well as their respective logos are trademarks +or registered trademarks of Rizing LLC or a Rizing affiliate company in the United States and other countries. All other product +and service names mentioned are the trademarks of their respective companies. + diff --git a/q110/random_k4/question.json b/q110/random_k4/question.json index 5d6beb30633c891bfbd8e5588725a71e6df4b4b7..a8e0ec2b43493ea9eabb2426e52a36951059cb02 100644 --- a/q110/random_k4/question.json +++ b/q110/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "earthlinkweb-150213112111-conversion-gate02_95.pdf", - "web_ca39b8721f6d1255.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf", - "web_040fe479bf878a27.pdf", - "web_715e6cf2d4ba3d94.pdf" + "resize_r2_0.pdf", + "web_88f0578d817bd13f.pdf", + "resize_fb_0.pdf", + "web_d1015c22be42ae71.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q110/random_k4/random_1.md b/q110/random_k4/random_1.md index 6bfd67b67b9eb8baf6562a416c94e48f3b69ec3d..052cee2fff6d2fb0526b1d43400b3c01f2587dfb 100644 Binary files a/q110/random_k4/random_1.md and b/q110/random_k4/random_1.md differ diff --git a/q110/random_k4/random_2.md b/q110/random_k4/random_2.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..ebd6ecb8692e8722f7ba60c4670378f06f3bd010 100644 --- a/q110/random_k4/random_2.md +++ b/q110/random_k4/random_2.md @@ -1,1453 +1,613 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA +The Step-by-Step Guide +to Recruiting Top Talent + +Your blueprint to transform your talent acquisition practices + to attract the people you need. + + Table of Contents + +01 + +The Race for Talent + +02 + +Does Your Talent Acquisition +Process Need a Tune-up? + +03 + +5 Factors That Affect Your Talent +Acquisition Efforts + +04 + +The Step-by-Step Guide to +Recruiting Top Candidates + +page 3 + +page 4 + +page 5 + +page 7 + +BONUS + +05 + +10 Must-Haves of a Great +Recruitment Management System +page 11 + +06 + +Find and Keep +the Talent You Need + +page 12 + +2 + +BOOKMARKS +Click on the title to to be +directed to that page. + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. The Race for Talent + +01 + +It’s a job seeker’s market. + +High employment rates, an aging workforce, and +skills gaps in both technical and soft skills have put +employees in the driver’s seat. + +What this means for you is that you’re going to +have to fight harder than ever to recruit and retain +the people you need. + +You’re going to have to leverage technology, lean into +your values, improve your employer branding, and in +some cases take a sledgehammer to your current talent +acquisition strategies to create engaging and responsive +recruitment experiences that not only attract candidates +to your company but compel them to choose you over +the competition. + +It can be a daunting task… but we’ve got you covered +-as HCM consultants and SAP® SuccessFactors® +specialists, we’ve helped countless businesses do +exactly this. + +This eBook is the first step in the process. It includes the +step-by-step guide to recruiting the best talent. Use it to +evaluate where you stand and build a talent acquisition +process that gives you the workforce you need for a +successful future. + +Let’s get started. + +An overwhelming +majority of candidates +(69%) refuse to +re-apply to a company +if they have a poor +experience during the +application process. + +3 + +1 2016 Talent Board NAM CandE Research Report FINAL 170202.pdf +link: http://www.thetalentboard.org/wp-content/uploads/2017/02/2016_Talent_Board_NAM_Can- + +dE_Research_Report_FINAL_170202.pdf + +69%© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. + Does Your Talent Acquisition +Process Need a Tune-up? + +02 + +Let’s look under the hood at your talent acquisition strategy +and see what’s really there. Here are five key questions that +you need to answer to identify the roadblocks between where +you are, and where you want to be. + +01 + +02 + +03 + +04 + +05 + +Is your recruiting based on +current best practices? + +Do you have buy-in +from your team? + +Is your technology up to date +and fully cloud capable? + +HR trends are evolving more +rapidly now than ever before. +Make sure you’re not falling +behind the competition when +it comes to implementing best +practices. + +Is your team aligned internally? +Do they share a common lens? +Great candidate experiences +depend on the interactions +candidates have with your +team during the various stages +of the acquisition process. +Buy-in from your team is vital. + +If you haven’t moved to the +cloud yet, it is never too late. +If you’ve migrated but aren’t +satisfied with your current +system, it can be salvaged. +The right tech really does +make a difference. + +Are you always looking +for ways to innovate your +recruiting approach? + +Is your recruitment +management system (RCM) +making your job easier? + +Technology is playing an +increasingly crucial role in the +talent acquisition process. You +need to always have an eye on +the horizon, ear to the ground, +and willingness to adapt +and innovate your approach. +Remember: good ideas can +come from anywhere. + +The right tools should help +to increase efficiency. But +sometimes, the tool itself +sucks up your time. And that’s +frustrating. Make sure your +system and processes make +your job easier. That’s the most +valuable success metric. + +4 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 5 Factors That Affect +Your Talent Acquisition Efforts + +03 + +The first step to building an effective acquisition strategy is to know +what internal and external factors affect your recruitment efforts. + +Based on our experience and a study of current market conditions, here’s +5 factors that you need to be aware of when building your strategy. + +You need to determine how you’ll adapt, counter, and respond to +each of these. + +5 + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 01 + +GROWING DEMAND & +DWINDLING SUPPLY + +02 + +TURNOVER & +BURNOUT + +03 + +INCONSISTENT +HIRING PROCESSES + +04 + +INTENSE FOCUS ON +COST MANAGEMENT + +05 + +EVOLVING +EXPECTATIONS + +With just about every major +industry facing staffing +shortages, the candidates are +now in the driver’s seat with the +option of picking an employer +that responds to their needs +in a location of their choosing. +While talent acquisition and +human resources teams can’t +change the growing supply/ +demand imbalance, they need +to be aware of how this affects +their acquisition strategy and +take active steps to attract +Gen X, Millennials, and +pretty soon Gen Z before the +competition does. + +The staffing shortage also has +an adverse effect on current +professionals who have to +work longer hours leading to +burnout and turnover – which +leads to a dangerous loop of +an exhausted workforce, an +ever-growing pile of work, and +more turnover. Furthermore, an +organization with a reputation +for high turnover and burnout +creates additional challenges +for recruitment teams when it +comes to hiring the best talent. + +Modern businesses often have +complex org charts, siloed +branches, and evolving parts. +This creates inconsistent hiring +processes that vary from +department to department, +region to region, and result in +poor candidate experiences +that negatively affect the +organization’s brand image. +While the HR team can’t +control everything, cloud-based +comprehensive talent solutions +such as SAP SuccessFactors can +help companies centralize their +processes and bring visibility to +internal teams and candidates. + +Companies are always under +pressure to control costs. Up +against this mentality, HR +managers struggle to secure +funding for talent management +initiatives and, as a result, +end up using disparate, on +premise or outdated recruitment +technologies. This leads to error +prone processes due to the +doubling of data across systems, +labor intensive and slow hiring +steps, and a disconnected +candidate experience. + +The largest generation in the +workforce today is tech-savvy +and socially aware Millennials. +As candidates, Millennials +look for companies that +align with their values, offer +training and development, +and deliver opportunities for +career growth. They expect +their future employers to +be digitally connected with +an intuitive website that +showcases its brand and +culture. They want a simple, +mobile friendly job application +process, transparent and timely +communication, and an active +social media presence so they +can interact with the company +during the hiring process. + +6 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. The Step-by-Step Guide to +Recruiting Top Candidates + +04 + +Building an amazing team is just as important as building a great +customer base. But, hiring the cream of the crop isn’t as easy as it +used to be. In the race for talent, the candidate experience matters +more now than ever before! This step-by-step guide is the blueprint +you need to transform your recruitment processes to meet the +expectations of a truly multigenerational digital workforce. + +7 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. 1 + +2 + +EVALUATE YOUR HIRING PROCESS +FROM START TO FINISH + +Start by evaluating your current hiring process to make +sure you know what your candidate experience is +truly like - we would even go so far to suggest secret- +shopping the hiring process. + +Next, storyboard your hiring process and map out +outcomes. The goal here is to plot the set of experiences +a candidate could have and optimize every step. + +Use the data insights to create repeatable and +consistent hiring processes that offer positive +experiences. + +Always make sure your hiring processes align +with and reflect your values and culture. + +TIP + 8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -i -C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | -B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) -M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +CREATE A STRONG, ATTRACTIVE +BRAND THAT SHOWS YOUR VALUE +TO TOP TALENT + +Your brand is your most valuable asset, when it comes to +recruiting talent. + +The first step in creating an attractive brand is developing +your Employee Value Proposition - a good EVP outlines +your organization’s mission and values and commitment +to employees. + +Once you have your EVP, the next step is to make sure +your career site not only reflects but showcases your EVP +- much like the buyer’s journey, the hiring journey now +begins with an online search. + +Lastly, make sure you have a strong content strategy +as it will help you build your brand identity and ensure +candidates find and connect with you online. + +Great content is relevant, trendy, and +provides value. We find setting up an +editorial board helps ensure you’re regularly +and consistently posting good content. + +TIP + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. < Continued from previous page + +3 + +BUILD A TALENT PIPELINE AND NURTURE TALENT + +There’s no such thing as “Just-in-Time Recruiting”. + +Smart recruiters know the importance of building and nurturing a talent pipeline. +To build a strong pipeline: + +01 + +02 + +Source talent globally +through an omni-channel +approach. Connect with and +reach out to candidates on +job boards, social platforms +and in-person events. + +Create a network of +communities, content, +and contact points so that +you’re constantly nurturing +candidates and making sure +your brand is top of mind. + +03 + +04 + +Make it simple for candidates +to get in touch and follow +up with HR by leveraging +automation, chatbots, and +other AI technology. + +05 + +Use technology to track +candidate progress and +development benchmarks. + +Take every opportunity +to share your culture and +events with candidates by +sharing different types of +company news and not just +job updates. + +Encourage your +organization’s +leaders to be +active online, +sharing stories +and updates, and +engaging with +candidates. + +TIP + +9 + +Continued on next page > + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. < Continued from previous page + +4 + +5 + +ENHANCE THE CANDIDATE EXPERIENCE + +It’s time to end the candidate feedback black hole. + +According to the 2016 North American Talent Board +Candidate Experience research report, 47 percent of +candidates never receive any communication up to two +months after applying for an open position!2 + +Give yourself an edge over the competition by +leveraging your RCM to not only keep your internal +team on the same page but also track and +communicate with candidates regularly. + +Additionally, make sure your career site is always up to +date with current openings and that it showcases your +culture through employee testimonials. + +CREATE AND SHARE +A CONSISTENT MESSAGE + +The last step is less of a one-time only step and more so +an ongoing process. + +Once you’ve evaluated, streamlined, and automated +your hiring processes, developed your EVP, and built +out your pipeline strategies, you need to make sure your +organization continues to create and share consistent +content and messages with your online and offline +communities. + +Companies that constantly connect and engage passive +talent will have greater success at sourcing and hiring +top talent. + +Don’t forget to make your application +process mobile friendly. + +TIP + +10 + +Tell your story. Engage your current +workforce and ask them to share +testimonials about what a job with your +organization is really like. Candidates are +attracted to authentic messages driven by +the workforce, not mandated by leadership. + +TIP + +2 2016 Talent Board NAM CandE Research Report FINAL 170202.pdf + link: http://www.thetalentboard.org/wp-content/uploads/2017/02/2016_ +Talent_Board_NAM_CandE_Research_Report_FINAL_170202.pdf + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. BONUS + +10 Must-Haves of a Great +Recruitment Management System + +05 + +There’s no denying that technology has changed the talent acquisition process – But, how do you know if the +technology you’re currently using is delivering what you need to build a motivated, high performing workforce? + +Use this cheat-sheet of the 10 must-haves to evaluate your system. + +Seamless cloud integration – easily access candidate and +employee information from anywhere. + +A simplified workflow for recruiters & HR – make the hiring +process easier for top talent and your HR team to navigate. + +Standardized, optimized processes & communication – +with consistent processes in place, you’ll reduce error and +dramatically improve communications. + +Omni-channel job distribution – advertise open positions +across sites and social channels to increase job exposure. + + Customized, mobile-friendly landing pages & web portals – +make the hiring process faster and easier. + +06 + +07 + +08 + +09 + +10 + +Support for single and multi-stage applicant flows – make it +simpler to hire regardless of job complexity. + +A candidate-first experience – cater to the people you want +most and build a consistent pipeline for top talent. + +Empowered hiring managers – put the right information in the +hands of decisions makers. + +A smooth transition from candidate to employee – get new +employees up and running faster, saving time and costs. + +Full visibility through analytics – make data-driven hiring +decisions and track candidates from start to finish. + +01 + +02 + +03 + +04 + +05 + +11 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. Find and Keep +the Talent You Need + +Finding the right talent is key to growing your business. A major part of recruiting +success is the tools you use. Simplifying your data by replacing disparate, on- +premise systems with a unified, cloud-based HR system that provides real-time +data is a great start. + +Our certified SAP SuccessFactors professionals can help you master your new +recruiting system. But most importantly, we help you align that system to your +overall HR strategy and your company culture. Our goal is to help you transform +your business so you can find and attract great talent. + +Contact us today to find out how we can help you deliver a recruiting +solution designed to let your organization’s recruiting efforts thrive. + +12 + +© 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. © 2020 Rizing LLC or a Rizing LLC affiliate company. All rights reserved. This document is provided for information purposes +only, and the contents are subject to change without notice. This document is not warranted to be error-free, nor subject to +any other warranties or conditions, whether expressed orally or implied in law, including implied warranties and conditions +of merchantability or fitness for a particular purpose. We specifically disclaim any liability with respect to this document, and +no contractual obligations are formed either directly or indirectly by this document. This document may not be reproduced +or transmitted in any form or by any means, electronic or mechanical, for any purpose, without our prior written permission. +Rizing, Rizing HCM, and other Rizing products and services mentioned herein as well as their respective logos are trademarks +or registered trademarks of Rizing LLC or a Rizing affiliate company in the United States and other countries. All other product +and service names mentioned are the trademarks of their respective companies. + diff --git a/q110/random_k4/random_3.md b/q110/random_k4/random_3.md index bac75566f27beeea6f3e99aec7e954b86be4654a..4bc854bda9c03ca810632facd384276a4d8fd0f2 100644 --- a/q110/random_k4/random_3.md +++ b/q110/random_k4/random_3.md @@ -1,1202 +1,793 @@ -FOR PUBLICATION +E15695 +Revised Edition V2 / December 2019 -UNITED STATES COURT OF APPEALS -FOR THE NINTH CIRCUIT +User Guide -X CORP., +E15695_X512FJ_FL_UM_V2_16pages.indd 1 - Plaintiff - Appellant, +2019/12/27 �� 09:59:07 - v. + Top View -ROBERT BONTA, in his official -capacity as Attorney General of -California, +NOTE: The keyboard's layout may vary per region or country. The top view may also +vary in appearance depending on the Notebook PC model. - Defendant - Appellee. +Array microphones -No. 24-271 +Camera indicator -D.C. No. -2:23-cv-01939- -WBS-AC +Camera -OPINION +Display panel -Appeal from the United States District Court -for the Eastern District of California -William B. Shubb, District Judge, Presiding +Power button -Argued and Submitted July 17, 2024 -San Francisco, California +Numeric keypad -Filed September 4, 2024 +Keyboard -Before: MILAN D. SMITH, JR., MARK J. BENNETT, -and ANTHONY D. JOHNSTONE, Circuit Judges. +Touchpad -Opinion by Judge Milan D. Smith, Jr. +Function key lock indicator +Capital lock indicator - 2 +2 -X CORP. V. BONTA +E15695_X512FJ_FL_UM_V2_16pages.indd 2 -SUMMARY* +2019/12/27 �� 09:59:07 -First Amendment / Social Media Platforms +Notebook PC User Guide I/O ports and slots -The panel reversed the district court’s order denying -social media platform owner X Corp.’s motion for a -preliminary injunction to enjoin enforcement of California -Assembly Bill AB 587 (AB 587), which requires large social -media companies to post their terms of service and to submit -reports to the Attorney General of California (the State) -about their terms of service and their content-moderation -policies and practices. +MicroSD card slot -The Content Category Report provisions of AB 587 -require social media companies to submit to the State a -semiannual report detailing whether and how they define six -categories of content: hate speech or racism, extremism or -radicalization, -or misinformation, -harassment, foreign political interference, and controlled -substance distribution. +USB 3.2 Gen 1 port -disinformation +Headphone/Headset/Microphone jack -The panel held that X Corp. was likely to succeed on the -merits of its claim that the Content Category Report -provisions facially violate the First Amendment. A facial -challenge is permissible because the Content Category -Report provisions raise the same First Amendment issues for -every social media company. The Content Category Report -provisions compel non-commercial speech, and are subject -to strict scrutiny because the provisions are content- -based. The Content Category Report provisions likely fail -strict scrutiny because they are not narrowly tailored to serve +Power (DC) input port -* This summary constitutes no part of the opinion of the court. It has -been prepared by court staff for the convenience of the reader. +USB 3.2 Gen 1 Type-C® port - X CORP. V. BONTA +USB 2.0 port + +HDMI port + +E15695_X512FJ_FL_UM_V2_16pages.indd 3 + +2019/12/27 �� 09:59:08 3 -the State’s purported goal of requiring social media -companies to be transparent about their policies and -practices. +Notebook PC User Guide Getting started + +1. Charge your Notebook PC + +A. Connect the DC power connector into your Notebook PC’s power (DC) input port. + +B. Plug the AC power adapter into a 100V~240V power source. + +IMPORTANT! Use only the bundled power adapter to charge the battery pack and +supply power to your Notebook PC. + +NOTE: The power adapter may vary in appearance, depending on models and your +region. + +2. Lift to open the display panel + +3. Press the power button + +Charge the Notebook PC +for 3 hours before using +it in battery mode for the +first time. + +4 + +E15695_X512FJ_FL_UM_V2_16pages.indd 4 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide + + Safety notices for your Notebook PC + +WARNING! +Your Notebook PC can get warm to hot while in use or while charging the battery pack. Do not +leave your Notebook PC on your lap or near any part of your body to prevent injury from heat. +When working on your Notebook PC, do not place it on surfaces that can block the vents. + +CAUTION! +• This Notebook PC should only be used in environments with ambient temperatures between + +5°C (41°F) and 35°C (95°F). + +• Refer to the rating label on the bottom of your Notebook PC and ensure that your power + +adapter complies with this rating. + +• The power adapter may become warm to hot while in use. Do not cover the adapter and keep + +it away from your body while it is connected to a power source. + +IMPORTANT! +• Ensure that your Notebook PC is connected to a grounded power adapter before turning it + +on for the first time. + +• When using your Notebook PC on power adapter mode, the socket outlet must be near to the + +unit and easily accessible. + +• Locate the input/output rating label on your Notebook PC and ensure that it matches the +input/output rating information on your power adapter. Some Notebook PC models may +have multiple rating output currents based on the available SKU. -The panel held that the remaining factors weighed in +• Power adapter information: -favor of a preliminary injunction. +- Input voltage: 100-240Vac -Accordingly, the panel reversed the district court’s -denial of a preliminary injunction, and remanded with -instructions to enter a preliminary injunction consistent with -the opinion and to determine whether the Content Category -Report provisions are severable from the remainder of AB -587 and, if so, which, if any, of the remaining challenged -provisions should also be enjoined. +- Input frequency: 50-60Hz -COUNSEL +- Rating output current: 2.37A (45W) / 3.42A (65W) -Joel L. Kurtzberg (argued), Floyd Abrams, Jason D. -Rozbruch, and Lisa J. Cole, Cahill Gordon & Reindel LLP, -New York, New York; William R. Warne and Meghan M. -Baker, Downey Brand LLP, Sacramento, California; for -Plaintiff-Appellant. +- Rating output voltage: 19V -Gabrielle D. Boutin (argued), Deputy Attorney General; -Anthony R. Hakl, Supervising Deputy Attorney General; -Thomas S. Patterson, Senior Deputy Attorney General; Rob -Bonta, Attorney General of California; Office of the -California Attorney General, Sacramento, California; for -Defendant-Appellee. +WARNING! +Read the following precautions for your Notebook PC’s battery: -Robert Corn-Revere and Joshua A. House, Foundation for -Individual Rights and Expression, Washington, D.C., for -Amicus Curiae Foundation for Individual Rights and -Expression. +• Only ASUS-authorized technicians should +remove the battery inside the device +(for non-removable battery only). - 4 +• The battery used in this device may present +a risk of fire or chemical burn if removed or +disassembled. -X CORP. V. BONTA +• Follow the warning labels for your personal -Trenton H. Norris, Mark W. Brennan, J. Ryan Thompson, -Sophie Baum, and Alexander Tablan, Hogan Lovells LLP, -San Francisco, California; Cory L. Andrews and John M. -Masslon II, Washington Legal Foundation, Washington, -D.C.; for Amicus Curiae Washington Legal Foundation. +safety. -Gene C. Schaerr, Schaerr Jaffe LLP, Washington, D.C., for -Amici Curiae Professor Eugene Volokh and Protect the First -Foundation. +• Risk of explosion if battery is replaced by an -Megan L. Brown, Jeremy J. Broggi, and Boyd Garriott, -Wiley Rein LLP, Washington, D.C.; Jonathan D. Urick and -Maria C. Monaghan, United States Chamber Litigation -Center; Washington, D.C.; for Amicus Curiae United States -of America Chamber of Commerce. +incorrect type. -Bruce D. Brown, Katie Townsend, Gabe Rottman, Grayson -Clary and Emily Hockett, Reporters Committee for Freedom -of the Press, Washington, D.C.; for Amicus Curiae Reporters -Committee for Freedom of the Press. +• Do not dispose of in fire. -David A. Greene and Aaron Mackey, Electronic Frontier -Foundation, San Francisco, California, for Amicus Curiae -Electronic Frontier Foundation. +• Never attempt to short-circuit your -Jacob M. Karr, Technology Law and Policy Clinic at New -York University, New York, New York; G.S. Hans, Cornell -Law School, Ithaca, New York; for Amici Curiae First -Amendment and Internet Law Scholars. +Notebook PC’s battery. -Michelle Quist and Lauren D. Wigginton, Buchalter APC, -Salt Lake City, Utah; Jon M. Greenbaum, Edward G. Caspar, -and Marc P. Epstein, Lawyers' Committee for Civil Rights -Under Law, Washington, D.C.; for Amicus Curiae Lawyers' -Committee for Civil Rights Under Law. +• Never attempt to disassemble and - X CORP. V. BONTA +reassemble the battery +(for non-removable battery only). + +• Discontinue usage if leakage is found. + +• This battery and its components must be + +recycled or disposed of properly. + +• Keep the battery and other small +components away from children. 5 -Viviana M. Hanley and Nathanial I. Levy, Deputy Attorneys -General; Michael L. Zuckerman, Deputy Solicitor General; -Jeremy Feigenbaum, Solicitor General; Metthew J. Platkin, -Attorney General of New Jersey; Office of the New Jersey -Attorney General, Trenton, New Jersey; Kristin K. Mayes, -Attorney General of Arizona, Office of the Arizona Attorney -General, Phoenix, Arizona; Philip J. Weiser, Attorney -General of Colorado, Office of the Colorado Attorney -General, Denver, Colorado; William Tong, Attorney -General of Connecticut, Office of the Connecticut Attorney -General, Hartford, Connecticut; Kathleen -Jennings, -Attorney General of Delaware, Office of the Delaware -Attorney General, Wilmington, Delaware; Brian L. -Schwalb, Attorney General of the District of Columbia, -Office of the District of Columbia Attorney General, -Washington, D.C.; Kwame Raoul, Attorney General of -Illinois, Office of the Illinois Attorney General, Chicago, -Illinois; Aaron M. Frey, Attorney General of Maine, Office -of the Maine Attorney General, Augusta, Maine; Anthony -G. Brown, Attorney General of Maryland, Office of the -Maryland Attorney General, Baltimore, Maryland; Andrea -J. Campbell, Attorney General of Massachusetts, Office of -the Massachusetts -Boston, -Attorney -Massachusetts; Dana Nessel, Attorney General of Michigan, -Office of -the Michigan Attorney General, Lansing, -Michigan; Keith Ellison, Attorney General of Minnesota, -Office of the Minnesota Attorney General, St. Paul, -Minnesota; Aaron D. Ford, Attorney General of Nevada, -Office of the Nevada Attorney General, Carson City, -Nevada; Letitia James, Attorney General of New York, -Office of the New York Attorney General, New York, New -York; Ellen F. Rosenblum, Attorney General of Oregon, -Office of the Oregon Attorney General, Salem, Oregon; - -General, - - 6 - -X CORP. V. BONTA - -Michelle A. Henry, Attorney General of Pennsylvania, -Office of Harrisburg, Pennsylvania; Charity R. Clark, -Attorney General of Vermont, Office of the Vermont -Attorney General, Montpelier, Vermont; Robert M. -Ferguson, Attorney General of Washington, Office of the -Washington Attorney General, Olympia, Washington; for -Amici Curiae States of New Jersey, Arizona, Colorado, -Connecticut, Delaware, The District of Columbia, Illinois, -Maine, Maryland, Massachusetts, Michigan, Minnesota, -Nevada, New York, Oregon, Pennsylvania, Vermont, and -Washington. - -Jason S. Harrow and Charles Gerstein, Gerstein Harrow -LLP, Los Angeles, California, for Amicus Curiae Institute -for Strategic Dialogue. - -Megan Iorio and Schuyler Standley, Electronic Privacy -Information Center, Washington, D.C., for Amicus Curiae -Electronic Privacy Information Center. - -Kristen G. Simplicio and Cort T. Carlson, Tycko & Zavareei -LLP, Washington, D.C.; John Yang, Niyati Shah, and Noah -Baron, Asian Americans Advancing Justice, Washington, -D.C.; for Amicus Curiae Asian Americans Advancing -Justice. - - X CORP. V. BONTA +E15695_X512FJ_FL_UM_V2_16pages.indd 5 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide + + Avis concernant les batteries remplaçables + +• La batterie de l’appareil peut présenter un risque d’incendie ou de brûlure si celle-ci est + +retirée ou désassemblée. + +• La batterie et ses composants doivent être recyclés de façon appropriée. + +Copyright Information +You acknowledge that all rights of this Manual remain with ASUS. Any and all rights, including without +limitation, in the Manual or website, are and shall remain the exclusive property of ASUS and/or its +licensors. Nothing in this Manual intends to transfer any such rights, or to vest any such rights to you. + +ASUS PROVIDES THIS MANUAL “AS IS” WITHOUT WARRANTY OF ANY KIND. SPECIFICATIONS AND +INFORMATION CONTAINED IN THIS MANUAL ARE FURNISHED FOR INFORMATIONAL USE ONLY, AND +ARE SUBJECT TO CHANGE AT ANY TIME WITHOUT NOTICE, AND SHOULD NOT BE CONSTRUED AS A +COMMITMENT BY ASUS. + +Copyright © 2019 ASUSTeK COMPUTER INC. All Rights Reserved. + +Limitation of Liability +Circumstances may arise where because of a default on ASUS’ part or other liability, you are entitled to +recover damages from ASUS. In each such instance, regardless of the basis on which you are entitled to +claim damages from ASUS, ASUS is liable for no more than damages for bodily injury (including death) +and damage to real property and tangible personal property; or any other actual and direct damages +resulted from omission or failure of performing legal duties under this Warranty Statement, up to the +listed contract price of each product. + +ASUS will only be responsible for or indemnify you for loss, damages or claims based in contract, tort or +infringement under this Warranty Statement. + +This limit also applies to ASUS’ suppliers and its reseller. It is the maximum for which ASUS, its suppliers, +and your reseller are collectively responsible. + +UNDER NO CIRCUMSTANCES IS ASUS LIABLE FOR ANY OF THE FOLLOWING: (1) THIRD-PARTY CLAIMS +AGAINST YOU FOR DAMAGES; (2) LOSS OF, OR DAMAGE TO, YOUR RECORDS OR DATA; OR (3) SPECIAL, +INCIDENTAL, OR INDIRECT DAMAGES OR FOR ANY ECONOMIC CONSEQUENTIAL DAMAGES (INCLUDING +LOST PROFITS OR SAVINGS), EVEN IF ASUS, ITS SUPPLIERS OR YOUR RESELLER IS INFORMED OF THEIR +POSSIBILITY. + +Service and Support +For complete E-Manual version, refer to our multi-language website at: + +https://www.asus.com/support/ + +6 + +E15695_X512FJ_FL_UM_V2_16pages.indd 6 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide FCC Radio Frequency (RF) Exposure Caution +Statement + +WARNING! Any changes or modifications not expressly approved by the party +responsible for compliance could void the user’s authority to operate this equipment. + +This equipment complies with FCC radiation exposure limits set forth for an uncontrolled +environment. To maintain compliance with FCC RF exposure compliance requirements, please +avoid direct contact to the transmitting antenna during transmitting. End users must follow the +specific operating instructions for satisfying RF exposure compliance. + +UL Safety Notices + +• DO NOT use the Notebook PC near water, for example, near a bath tub, wash bowl, kitchen + +sink or laundry tub, in a wet basement or near a swimming pool. + +• DO NOT use the Notebook PC during an electrical storm. There may be a remote risk of + +electric shock from lightning. + +• DO NOT use the Notebook PC in the vicinity of a gas leak. + +• DO NOT dispose the Notebook PC battery pack in a fire, as they may explode. Check with local +codes for possible special disposal instructions to reduce the risk of injury to persons due to +fire or explosion. + +• DO NOT use power adapters or batteries from other devices to reduce the risk of injury to + +persons due to fire or explosion. Use only UL certified power adapters or batteries supplied +by the manufacturer or authorized retailers. + +Coating Notice + +IMPORTANT! To provide electrical insulation and maintain electrical safety, a coating +is applied to insulate the device except on the areas where the I/O ports are located. + +Prevention of Hearing Loss +To prevent possible hearing damage, do not listen at high volume levels for long +periods. + +Power Safety Requirement +Products with electrical current ratings up to 6A and weighing more than 3Kg must use +approved power cords greater than or equal to: H05VV-F, 3G, 0.75mm2 or H05VV-F, 2G, 0.75mm2. 7 -OPINION - -M. SMITH, Circuit Judge: - -The California State Legislature enacted Assembly Bill -587 (AB 587) in September 2022. Cal. Bus. & Prof. Code -§§ 22675–81. - The law requires large social media -companies to, inter alia, post their terms of service and to -submit, on a semiannual basis, reports to the Attorney -General of California (the State) about their terms of service -and content-moderation policies and practices. X Corp., the -owner of the large social media platform X (formerly known -as Twitter), moved for a preliminary injunction to enjoin -enforcement of AB 587 on free speech and federal -preemption grounds. The district court denied X Corp.’s -motion, finding that X Corp. failed to establish a likelihood -of success on the merits. X Corp. appeals. For the reasons -below, we reverse and remand to the district court for further -proceedings consistent with this opinion. - -FACTUAL AND PROCEDURAL BACKGROUND - -AB 587 has three primary elements: (1) a requirement -that social media companies 1 publicly post their terms of -service, including processes for flagging content and -potential actions that may be taken with respect to flagged -content (Terms of Service (TOS) Posting), see Cal. Bus. & -Prof. Code § 22676, (2) a requirement that social media - -1 AB 587 does not apply to social media companies with gross annual -revenues of less than $100 million, Cal. Bus. & Prof. Code § 22680, nor -to “an internet-based service or application for which interactions -between users are limited to direct messages, commercial transactions, -consumer reviews of products, sellers, services, events, or places, or any -combination thereof,” id. § 22681. - - 8 - -X CORP. V. BONTA - -racism; - -(b) extremism or - -companies submit to the State a semiannual report detailing -their TOS and content-moderation practices including, if at -all, how the terms of service define and address (a) hate -speech or -radicalization; -(c) disinformation or misinformation; (d) harassment; and -(e) foreign political interference, as well as statistics on -content that was flagged by the social media company as -belonging to any of the categories (TOS Report), see id. -§ 22677, 2 and (3) a penalty provision, whereby the social -media company may be sued in court for, inter alia, -materially omitting or misrepresenting required information -and may be liable to pay up to $15,000 per violation per day, -see id. § 22678.3 - -On September 8, 2023, X Corp. filed a complaint against -the State seeking declaratory relief and injunctive relief -barring the law’s enforcement. The complaint alleges three -causes of action challenging the TOS Posting, TOS Report, -and penalty provision of AB 587 as: (1) a violation of the -free speech clauses of the U.S. and California Constitutions; -(2) a violation of the Dormant Commerce Clause; and -(3) federally preempted pursuant to the Communications -Decency Act, 47 U.S.C. § 230(c). X Corp. filed a motion for -preliminary injunction based on its free speech and - -2 AB 587 was subsequently amended to add to this list “[c]ontrolled -substance distribution.” 2023 Cal. Legis. Serv. 7680 (West). - -3 In assessing the amount of any penalty, a court is to consider whether -the social media company has made a reasonable, good faith attempt to -comply with the provisions of the statute. Cal. Bus. & Prof. Code -§ 22678(a)(3). - - X CORP. V. BONTA +E15695_X512FJ_FL_UM_V2_16pages.indd 7 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide Declaration of Compliance for Product Environmental +Regulation +ASUS follows the green design concept to design and manufacture our products, and +makes sure that each stage of the product life cycle of ASUS product is in line with global +environmental regulations. In addition, ASUS disclose the relevant information based on +regulation requirements. Please refer to http://csr.asus.com/Compliance.htm for information +disclosure based on regulation requirements ASUS is complied with. + +EU REACH and Article 33 +Complying with the REACH (Registration, Evaluation, Authorization, and Restriction of +Chemicals) regulatory framework, we publish the chemical substances in our products at ASUS +REACH website at http://csr.asus.com/english/REACH.htm. + +EU RoHS +This product complies with the EU RoHS Directive. For more details, see +http://csr.asus.com/english/article.aspx?id=35. + +Japan JIS-C-0950 Material Declarations +Information on Japan RoHS (JIS-C-0950) chemical disclosures is available on +http://csr.asus.com/english/article.aspx?id=19. + +India RoHS +This product complies with the “India E-Waste (Management) Rules, 2016” and prohibits use of +lead, mercury, hexavalent chromium, polybrominated biphenyls (PBBs) and polybrominated +diphenyl ethers (PBDEs) in concentrations exceeding 0.1% by weight in homogenous materials +and 0.01% by weight in homogenous materials for cadmium, except for the exemptions listed +in Schedule II of the Rule. + +8 + +E15695_X512FJ_FL_UM_V2_16pages.indd 8 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide ASUS Recycling/Takeback Services +ASUS recycling and takeback programs come from our commitment to the highest standards for +protecting our environment. We believe in providing solutions for you to be able to responsibly +recycle our products, batteries, other components as well as the packaging materials. Please +go to http://csr.asus.com/english/Takeback.htm for detailed recycling information in different +regions. + +Ecodesign Directive +European Union announced a framework for the setting of ecodesign requirements for energy- +related products (2009/125/EC). Specific Implementing Measures are aimed at improving +environmental performance of specific products or across multiple product types. ASUS +provides product information on the CSR website. Further information could be found at +https://csr.asus.com/english/article.aspx?id=1555. + +EPEAT Registered Products +The public disclosure of key environmental information for ASUS EPEAT (Electronic Product +Environmental Assessment Tool) registered products is available at +https://csr.asus.com/english/article.aspx?id=41. More information about EPEAT program and +purchase guidance can be found at www.epeat.net. + +Regional notice for Singapore + +Complies with +IMDA Standards +DB103778 + +This ASUS product complies with IMDA Standards. + +E15695_X512FJ_FL_UM_V2_16pages.indd 9 + +2019/12/27 �� 09:59:08 9 -preemption claims, seeking to enjoin the State from -enforcing the challenged provisions of AB 587.4 - -On December 28, 2023, the district court denied X -Corp.’s motion. The court began its analysis with X Corp.’s -First Amendment claim.5 The court held that X Corp. was -unlikely to prevail because the TOS Posting and TOS Report -requirements appeared constitutionally permissible in light -of Zauderer v. Office of Disciplinary Counsel of Supreme -Court of Ohio, 471 U.S. 626 (1985), the Supreme Court’s -test for compelled commercial speech. See X Corp. v. Bonta, -No. 23-cv-01939, 2023 WL 8948286, at *1–2 (E.D. Cal. -Dec. 28, 2023). - -The court’s analysis of the TOS Report requirement -focused primarily on the provisions requiring that social -media companies report whether and how they define and -address certain enumerated content categories. Id. at *2. -The court acknowledged that such reports do “not so easily -fit the traditional definition of commercial speech” because -they “are not advertisements” and because “social media -companies have no particular economic motivation to -provide them.” Id. However, the court applied Zauderer to -those provisions nevertheless so as to “follow[] the lead of -the Fifth and Eleventh Circuits.” Id. (citing NetChoice, LLC -v. Paxton, 49 F.4th 439, 485 (5th Cir. 2022), rev’d on other -grounds sub nom. Moody v. NetChoice, LLC, 144 S. Ct. 2383 - -4 X Corp. did not seek a preliminary injunction based upon the Dormant -Commerce Clause. - -5 The district court did not analyze X Corp.’s free speech claim under -Article I, Section 2, of the California Constitution, nor do the parties -meaningfully address this claim on appeal. Because we hold that X -Corp. is likely to succeed on its First Amendment claim, we do not reach -X Corp.’s free speech claim pursuant to the California Constitution. - - 10 - -X CORP. V. BONTA - -to - -(2024) (“NetChoice (Tex.)”), and NetChoice, LLC v. Att’y -Gen., Fla., 34 F.4th 1196, 1230 (11th Cir. 2022), rev’d on -other grounds sub nom. Moody, 144 S. Ct. 2383 (“NetChoice -(Fla.)”)). The court then concluded that the TOS Report -requirement satisfies Zauderer. Id. The court reasoned that -the provisions require speech that is “purely factual” and -“uncontroversial” because they “merely require[] social -media companies -their existing content -identify -moderation policies, if any, related to the specified -categories” and the “mere fact that the reports may be ‘tied -in some way to a controversial issue’ does not make the -reports themselves controversial.” Id. (quoting CTIA - The -Wireless Ass’n v. City of Berkeley, 928 F.3d 832, 845 (9th -Cir. 2019) (“CTIA II”)). The court rejected X Corp.’s -argument that the TOS Report requirement is “unduly -burdensome,” explaining that “AB 587 does not require that -a social media company adopt any of the specified -categories” of speech, and that in any event “Zauderer is -concerned not merely with logistical or economic burdens, -but burdens on speech.” Id. It further held that the TOS -Report requirement is “reasonably related to a substantial -government interest in requiring social media companies to -be transparent about their content moderation policies and -practices so that consumers can make informed decisions -about where they consume and disseminate news and -information.” Id. - -The district court also determined that X Corp. had failed -to show a likelihood of success on its claim that AB 587 is -preempted by 47 U.S.C. § 230(c). Id. at *3. The court -observed that the purpose of section 230(c) “is to provide -‘protection for “Good Samaritan” blocking and screening of -offensive material’” so that a website may “self-regulate -offensive third party content without fear of liability.” Id. - - X CORP. V. BONTA +Notebook PC User Guide Simplified EU Declaration of Conformity +ASUSTek Computer Inc. hereby declares that this device is in compliance with the essential +requirements and other relevant provisions of Directive 2014/53/EU. Full text of EU declaration +of conformity is available at https://www.asus.com/support/. + +The WiFi operating in the band 5150-5350 MHz shall be restricted to indoor use for countries +listed in the table below: + +AT +DE +LV +NO +FI + +BE +IS +LI +PL +SE + +BG +IE +LT +PT +CH + +CZ +IT +LU +RO +UK + +DK +EL +HU +SI +HR + +EE +ES +MT +SK + +FR +CY +NL +TR + +10 + +E15695_X512FJ_FL_UM_V2_16pages.indd 10 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide Federal Communications Commission Interference +Statement +This equipment has been tested and found to comply with the limits for a Class B digital +device, pursuant to part 15 of the FCC Rules. These limits are designed to provide reasonable +protection against harmful interference in a residential installation. This equipment generates, +uses and can radiate radio frequency energy and, if not installed and used in accordance with +the instructions, may cause harmful interference to radio communications. However, there is +no guarantee that interference will not occur in a particular installation. If this equipment does +cause harmful interference to radio or television reception, which can be determined by turning +the equipment off and on, the user is encouraged to try to correct the interference by one or +more of the following measures: + +• Reorient or relocate the receiving antenna. + +• + +Increase the separation between the equipment and receiver. + +• Connect the equipment into an outlet on a circuit different from that to which the receiver is + +connected. + +• Consult the dealer or an experienced radio/TV technician for help. + +FCC COMPLIANCE INFORMATION + +Per FCC Part 2 Section 2.1077 + +Responsible Party: Asus Computer International + +Address: + +48720 Kato Rd., Fremont, CA 94538 + +Phone/Fax No: + +(510)739-3777/(510)608-4555 + +hereby declares that the product + +Product Name : + +VivoBook/ASUS Laptop + +Model Number : + +X512U, F512U, A512U, R564U, P1504U, S512U, +K512U, V512U, X512F, F512F, A512F, R564F, P1504F, +S512F, K512F, V512F, X512D, F512D, A512D, R564D, +P1504D, S512D, K512D, V512D, X512J, A512J, R564J, + +P1504J, S512J, K512J, V512J + +compliance statement: + +This device complies with part 15 of the FCC Rules. Operation is subject to +the following two conditions: (1) This device may not cause harmful +interference, and (2) this device must accept any interference received, +including interference that may cause undesired operation. + +Ver. 180125 11 -(quoting Doe v. Internet Brands, Inc., 824 F.3d 846, 851–52 -(9th Cir. 2016)). The court concluded that AB 587 is not -preempted because, under its plain language, it “does not -provide for any potential liability stemming from a -company’s content moderation activities per se,” only for -failing to make AB 587’s required disclosures. Id. +E15695_X512FJ_FL_UM_V2_16pages.indd 11 + +2019/12/27 �� 09:59:08 + +Notebook PC User Guide RED RF Output table + +X512U / X512D + +RTL8821CE + +Function + +WiFi + +Bluetooth + +RTL8822BE + +Function + +WiFi + +Bluetooth + +QCNFA435 + +Function + +WiFi + +Bluetooth + +Intel 8265NGW + +Function + +WiFi + +Bluetooth + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +5745 - 5825 MHz + +2402 - 2480 MHz + +Maximum Output Power (EIRP) +18 dBm + +21 dBm + +20 dBm + +8 dBm + +Maximum Output Power (EIRP) +18 dBm + +21 dBm + +20 dBm + +8 dBm + +Maximum Output Power (EIRP) +18 dBm + +20 dBm + +19 dBm + +6 dBm -On January 12, 2024, X Corp. timely filed notice of its -appeal. The provision of AB 587 most relevant in this appeal -is section 22677(a), which reads in its entirety: +Maximum Output Power (EIRP) +18 dBm -(a) On a semiannual basis in accordance with -subdivision (b), a social media company shall -submit to the Attorney General a terms of -service report. The terms of service report -shall include, for each social media platform -owned or operated by the company, all of the -following: +21 dBm -(1) The current version of the terms of service -of the social media platform. +20 dBm -(2) If a social media company has filed its -first -report, a complete and detailed -description of any changes to the terms of -service since the previous report. +11 dBm -(3) A statement of whether the current -version of the terms of service defines each -of the following categories of content, and, if -so, -those categories, -the definitions of -including any subcategories: +13 dBm -(A) Hate speech or racism. +For the standard EN 300 440 V2.1.1, if this device operates in 5725-5875 MHz, it will be +considered as a receiver category 2. -(B) Extremism or radicalization. +12 -(C) Disinformation or misinformation. +E15695_X512FJ_FL_UM_V2_16pages.indd 12 - 12 +2019/12/27 �� 09:59:09 -X CORP. V. BONTA +Notebook PC User Guide X512F -(D) Harassment. +RTL8822BE -(E) Foreign political interference. +Function -(F) Controlled substance distribution.6 +WiFi -(4) A detailed description of content -moderation practices used by the social -media company for that platform, including, -but not limited to, all of the following: +Bluetooth -(A) Any existing policies intended to -address -the categories of content -described in paragraph (3). +RTL8822CE -(B) How automated content moderation -systems enforce terms of service of the -social media platform and when these -systems involve human review. +Function -(C) How the social media company -responds to user reports of violations of -the terms of service. +WiFi -(D) How the social media company -would remove -individual pieces of -content, users, or groups that violate the -terms of service, or take broader action -against individual users or against groups -of users that violate the terms of service. +Bluetooth -(E) The languages in which the social -media platform does not make terms of -service available, but does offer product +QCNFA435 -6 As noted above, section 22677(a)(3)(F) was added subsequent to X -Corp. filing its lawsuit in the district court. +Function - X CORP. V. BONTA +WiFi + +Bluetooth + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +2402 - 2480 MHz + +Maximum Output Power (EIRP) +18 dBm + +21 dBm + +20 dBm + +8 dBm + +Maximum Output Power (EIRP) +17 dBm + +15 dBm + +15 dBm + +8 dBm + +Maximum Output Power (EIRP) +18 dBm + +20 dBm + +19 dBm + +6 dBm + +E15695_X512FJ_FL_UM_V2_16pages.indd 13 + +2019/12/27 �� 09:59:09 13 -features, including, but not limited to, -menus and prompts. +Notebook PC User Guide Intel 9462NGW -(5) (A) Information on content that was -flagged by the social media company as -content belonging to any of the categories -described in paragraph (3), including all of -the following: +Function -(i) The total number of flagged items of -content. +WiFi -(ii) The total number of actioned items of -content. +Bluetooth -(iii) The total number of actioned items of -content that resulted in action taken by -the social media company against the -user or group of users responsible for the -content. +Intel 9560NGW -(iv) The total number of actioned items of -content that were removed, demonetized, -or deprioritized by the social media -company. +Function -(v) The number of times actioned items -of content were viewed by users. +WiFi -(vi) The number of times actioned items -of content were shared, and the number of -users that viewed the content before it -was actioned. +Bluetooth -(vii) The number of times users appealed -social media company actions taken on -that platform and the number of reversals -of social media company actions on +Intel AX201 - 14 +Function -X CORP. V. BONTA +WiFi -appeal disaggregated by each type of -action. +Bluetooth -(B) All information required by subparagraph -(A) shall be disaggregated into the following -categories: +Frequency +2412 - 2472 MHz -(i) The category of content, including any -relevant -in -paragraph (3). +5150 - 5350 MHz -categories +5470 - 5725 MHz -described +5745 - 5825 MHz -(ii) The type of content, including, but not -limited to, posts, comments, messages, -profiles of users, or groups of users. +2402 - 2480 MHz -(iii) The type of media of the content, -including, but not limited to, text, images, -and videos. +Frequency +2412 - 2472 MHz -the content was flagged, -(iv) How -including, but not limited to, flagged by -company employees or contractors, -flagged by artificial intelligence software, -flagged by community moderators, -flagged by civil society partners, and -flagged by users. +5150 - 5350 MHz -(v) How -the content was actioned, -including, but not limited to, actioned by -company employees or contractors, -intelligence -actioned -software, -community -moderators, actioned by civil society -partners, and actioned by users. +5470 - 5725 MHz -by -actioned +5745 - 5825 MHz -artificial -by +2402 - 2480 MHz - X CORP. V. BONTA +Frequency +2412 - 2472 MHz + +5150 - 5350 MHz + +5470 - 5725 MHz + +5725 - 5850 MHz + +2402 - 2480 MHz + +Maximum Output Power (EIRP) +17 dBm + +17 dBm + +16 dBm + +11 dBm + +11 dBm + +Maximum Output Power (EIRP) +17 dBm + +17 dBm + +16 dBm + +10 dBm + +12 dBm + +Maximum Output Power (EIRP) +19 dBm + +21 dBm + +20 dBm + +11 dBm + +10 dBm + +For the standard EN 300 440 V2.1.1, if this device operates in 5725-5875 MHz, it will be +considered as a receiver category 2. + +14 + +E15695_X512FJ_FL_UM_V2_16pages.indd 14 + +2019/12/27 �� 09:59:09 + +Notebook PC User Guide E15695_X512FJ_FL_UM_V2_16pages.indd 15 + +2019/12/27 �� 09:59:09 15 -JURISDICTION AND STANDARD OF REVIEW - -We have jurisdiction pursuant to 28 U.S.C. § 1292(a)(1) -to review the denial of a preliminary injunction. Creech v. -Idaho Comm’n of Pardons & Parole, 94 F.4th 851, 854 (9th -Cir. 2024). We review the denial of a preliminary injunction -for abuse of discretion, but we review de novo the -underlying issues of law. Cal. Chamber of Com. v. Council -for Educ. & Rsch. on Toxics, 29 F.4th 468, 475 (9th Cir. -2022). - -“The appropriate legal standard to analyze a preliminary -injunction motion requires a district court to determine -whether a movant has established that (1) [it] is likely to -succeed on the merits of [its] claim, (2) [it] is likely to suffer -irreparable harm absent the preliminary injunction, (3) the -balance of equities tips in [its] favor, and (4) a preliminary -injunction is in the public interest.” Baird v. Bonta, 81 F.4th -1036, 1040 (9th Cir. 2023); see Winter v. Nat. Res. Def. -Council, Inc., 555 U.S. 7, 20 (2008). Because “the party -opposing injunctive relief is a government entity” here, the -third and fourth factors “merge.” Fellowship of Christian -Athletes v. San Jose Unified Sch. Dist. Bd. of Educ., 82 F.4th -664, 695 (9th Cir. 2023) (en banc) (quoting Nken v. Holder, -556 U.S. 418, 435 (2009)). - -ANALYSIS - -On appeal, X Corp. challenges the district court’s ruling -on the TOS Report requirement and penalty provision as -applied to the TOS Report requirement. X Corp. does not -appeal the district court’s denial of a preliminary injunction -as to the TOS Posting requirement, Cal. Bus. & Prof. Code -§ 22676. - - 16 - -X CORP. V. BONTA - -X Corp. argues that the district court erred by finding that -X Corp. did not establish a likelihood of success on the -merits because -is -(1) the TOS Report -compelled, non-commercial speech subject to strict scrutiny, -not the lower tier of scrutiny in Zauderer, (2) regardless, the -TOS Report requirement fails under any level of scrutiny, -and (3) section 230’s broad immunity precludes liability -under AB 587. - -requirement - -X Corp. seeks to reverse the district court’s ruling as to -the entirety of the TOS Report requirement. But the thrust -of the appeal concerns section 22677(a)(3), which requires -that social media companies report whether and how they -define six categories of content, and sections 22677(a)(4)(A) -and (a)(5), which directly incorporate section 22677(a)(3). -For ease of reference, we refer to these sections as the -Content Category Report provisions. - -For the reasons below, we hold that the Content Category -Report provisions likely compel non-commercial speech and -are subject to strict scrutiny, under which they do not -survive. We reverse the district court on that basis. Because -we reverse on free speech grounds, we need not reach X -Corp.’s section 230 theory. We remand to the district court -to determine in the first instance whether the Content -Category Report provisions are severable from -the -remainder of AB 587, and if so, which, if any, of the -remaining challenged provisions should also be subject to -the preliminary injunction.7 - -7 We do not decide whether sections 22677(a)(1), (2), and (4)(B)–(E)— -which require that social media companies disclose the text of their TOS -and describe their enforcement mechanisms, without mention of specific - - X CORP. V. BONTA - -17 - -I. X Corp. is likely to succeed in showing that the -Content Category Report provisions facially violate -the First Amendment. - -“For a host of good reasons, courts usually handle -constitutional claims case by case, not en masse.” Moody, -144 S. Ct. at 2397. The Supreme Court “has therefore made -facial challenges hard to win.” Id. In a typical facial -challenge, a plaintiff cannot -succeed “unless he -‘establish[es] that no set of circumstances exists under which -the [law] would be valid,’ or he shows that the law lacks a -‘plainly legitimate sweep.’” Id. (alterations in original) (first -quoting United States v. Salerno, 481 U.S. 739, 745 (1987); -then quoting Wash. State Grange v. Wash. State Republican -Party, 552 U.S. 442, 449 (2008)). - -less demanding - -However, in First Amendment cases, the Supreme Court -“has lowered that very high bar.” Id. “To provide breathing -the Supreme Court has -room for free expression,” -“substituted a -though still rigorous -standard.” Id. (cleaned up) (quoting United States v. -Hansen, 599 U.S. 762, 769 (2023)); see also Tucson v. City -of Seattle, 91 F.4th 1318, 1327 (9th Cir. 2024). “[I]f the -law’s unconstitutional applications substantially outweigh -its constitutional ones,” then a court may sustain a facial -challenge to the law and strike it down. Moody, 144 S. Ct. -at 2397. As Moody clarified, a First Amendment facial -challenge has two parts: first, the courts must “assess the -state laws’ scope”; and second, the courts must “decide - -content categories—are facially constitutional. Neither party—either -below or on appeal—briefed what should happen to the remainder of -section 22677 if the Content Category Report provisions were found to -be likely unconstitutional. - - 18 - -X CORP. V. BONTA - -which of the laws’ applications violate the First Amendment, -and . . . measure them against the rest.” Id. at 2398. - -“[N]o one has paid much attention to” the requirements -for a facial challenge so far in this case. Id. at 2397. -Nevertheless, we conclude that a facial challenge is -permissible here. That is because all aspects of the Content -Category Report, in every application to a covered social -media company, raise the same First Amendment issues. As -explained in further detail below, every Content Category -Report must detail the company’s policies and actions -concerning certain state-specified categories of content -(even if only to detail the company’s decision not to define -the enumerated categories of section 22677(a)(3)). In effect, -the Content Category Report provisions compel every -covered social media company to reveal its policy opinion -about contentious issues, such as what constitutes hate -speech or misinformation and whether to moderate such -expression.8 - -8 X Corp. cites legislative history and statements from the California -State Attorney General in describing the indirect chilling effects AB 587 -may have by generating public controversy about the actions of social -media companies and thereby pressuring them to change their content -moderation policies. No matter how a social media company chooses to -moderate such content, the company will face backlash from its users -and the public. That is true even if the company decides not to define -the enumerated categories, because they will draw criticism for under- -moderating their community. While we account for these effects in our -analysis, whether State officials intended these effects plays no role in -our analysis of the merits of this facial challenge. See B & L Prods., Inc. -v. Newsom, 104 F.4th 108, 116 (9th Cir. 2024) (citing United States v. -O’Brien, 391 U.S. 367, 383 n.30 (1968)) (rejecting “the idea that -‘legislative motive’” of indirectly chilling speech “‘is a proper basis for -declaring a statute unconstitutional’”). - - X CORP. V. BONTA - -19 - -Thus, the Content Category Report provisions raise the -same First Amendment issues for every covered social -media company. That is true from the face of the law; we -need not “speculate about ‘hypothetical’ or ‘imaginary’ -cases.” See Wash. State Grange, 552 U.S. at 450. We -therefore proceed to consider whether the Content Category -Report provisions are likely to survive X Corp.’s First -Amendment facial challenge. - -A. The Content Category Report provisions compel -non-commercial speech and are subject to strict -scrutiny. - -regulation - -One of the First Amendment’s core purposes is “to -preserve an uninhibited marketplace of ideas in which truth -will ultimately prevail.” McCullen v. Coakley, 573 U.S. 464, -476 (2014) (quoting FCC v. League of Women Voters of -Cal., 468 U.S. 364, 377 (1984)). In evaluating whether a -regulation violates the First Amendment, courts “distinguish -between content-based and content-neutral regulations of -speech.” Vidal v. Elster, 602 U.S. 286, 292 (2024) (internal -quotation marks omitted) (quoting Nat’l Inst. of Fam. & Life -Advocs. v. Becerra, 585 U.S. 755, 766 (2018)). A content- -based -its -communicative content,” restricting discussion of a subject -matter or topic. Reed v. Town of Gilbert, 576 U.S. 155, 163 -(2015). “As a general matter,” a content-based regulation is -“presumptively unconstitutional and may be justified only if -the government proves that [it is] narrowly tailored to serve -compelling state interests.” Nat’l Inst. of Fam. & Life -Advocs., 585 U.S. at 766 (quoting Reed, 576 U.S. at 163). -When a state “compel[s] individuals to speak a particular -message,” the state “alter[s] the content of their speech,” and -engages in content-based regulation. Id. (cleaned up) -(quoting Riley v. Nat’l Fed’n of the Blind of N.C., Inc., 487 - -speech based on - -“target[s] - - 20 - -X CORP. V. BONTA - -U.S. 781, 795 (1988)). The First Amendment’s guarantee of -freedom of speech makes no distinction of “constitutional -significance” “between compelled speech and compelled -silence.” Riley, 487 U.S. at 796–97. - -In general, laws regulating commercial speech are -subject to a lesser standard of scrutiny. See Bolger v. Youngs -Drug Prods. Corp., 463 U.S. 60, 64–65 (1983) (discussing -recognition and evolution of commercial speech doctrine). -This holds true for both corporations and individuals alike. -See Pac. Gas & Elec. Co. v. Pub. Utils. Comm’n of Cal., 475 -U.S. 1, 16 (1986). Commercial speech is “usually defined -as speech that does no more than propose a commercial -transaction.” United States v. United Foods, Inc., 533 U.S. -405, 409 (2001). “Courts view this definition as just a -starting point, however, and instead try to give effect to a -‘common-sense distinction’ between commercial speech -and other varieties of speech.” Ariix, LLC v. NutriSearch -Corp., 985 F.3d 1107, 1115 (9th Cir. 2021) (cleaned up) -(quoting Jordan v. Jewel Food Stores, Inc., 743 F.3d 509, -516–17 (7th Cir. 2014)). Indeed, the “commercial speech -analysis is fact-driven, due to the inherent difficulty of -drawing bright lines that will clearly cabin commercial -speech in a distinct category.” First Resort, Inc. v. Herrera, -860 F.3d 1263, 1272 (9th Cir. 2017) (internal quotation -marks omitted) (quoting Greater Balt. Ctr. for Pregnancy -Concerns, Inc. v. Mayor & City Council of Balt., 721 F.3d -264, 284 (4th Cir. 2013)). - -Because of the difficulty of drawing clear lines between -commercial and non-commercial speech, the Supreme Court -in Bolger outlined three factors to consider. 463 U.S. at 64– -67. “Where the facts present a close question, ‘strong -support’ that the speech should be characterized as -commercial speech is found where [1] the speech is an - - X CORP. V. BONTA - -21 - -advertisement, [2] the speech refers to a particular product, -and [3] the speaker has an economic motivation.” Hunt v. -City of L.A., 638 F.3d 703, 715 (9th Cir. 2011) (citing -Bolger, 463 U.S. at 66–67). These so-called Bolger factors -are important guideposts, but they are not necessarily -dispositive. See Bolger, 463 U.S. at 67 n.14 (“Nor do we -mean to suggest that each of the characteristics present in -this case must necessarily be present in order for speech to -be commercial.”); Dex Media W., Inc. v. City of Seattle, 696 -F.3d 952, 958 (9th Cir. 2012). - -Commercial speech is generally subject to intermediate -scrutiny. Nat’l Ass’n of Wheat Growers v. Bonta, 85 F.4th -1263, 1266 (9th Cir. 2023). However, an exception applies -to compelled commercial speech that is “purely factual and -uncontroversial.” Id.; see Pac. Coast Horseshoeing Sch., -Inc. v. Kirchmeyer, 961 F.3d 1062, 1074 (9th Cir. 2020) -(citing Zauderer as a variation in the treatment of speech -“within the class of commercial speech”). “In that scenario, -the government need only demonstrate the compelled speech -survives a lesser form of scrutiny akin to a rational basis -test.” Nat’l Wheat, 85 F.4th at 1266. - -State legislatures do not have “freewheeling authority to -declare new categories of speech outside the scope of the -First Amendment.” United States v. Stevens, 559 U.S. 460, -472, (2010). Thus, “without persuasive evidence that a -novel restriction on content is part of a long (if heretofore -unrecognized) tradition of proscription, a legislature may not -revise the ‘judgment [of] the American people,’ embodied in -the First Amendment, ‘that the benefits of its restrictions on -the Government outweigh the costs.’” Brown v. Entm’t -Merchs. Ass’n, 564 U.S. 786, 792 (2011) (alteration in -original) (quoting Stevens, 559 U.S. at 470). - - 22 - -X CORP. V. BONTA - -Here, the Content Category Reports are not commercial -speech. They require a company to recast its content- -moderation practices in language prescribed by the State, -implicitly opining on whether and how certain controversial -categories of content should be moderated. As a result, few -indicia of commercial speech are present in the Content -Category Reports. - -First, the Content Category Reports do not satisfy the -“usual[] defin[ition]” of commercial speech—i.e., “speech -that does no more than propose a commercial transaction.” -See United Foods, Inc., 533 U.S. at 409; see also IMDb.com -Inc. v. Becerra, 962 F.3d 1111, 1122 (2020) (“Because -IMDb’s public profiles do not ‘propose a commercial -transaction,’ we need not reach the Bolger factors.”). The -State appears to concede as much in its answering brief. - -To the extent our circuit has recognized exceptions to -that general rule, those exceptions are limited and are -inapplicable to the Content Category Reports here. For -example, as identified by the First Amendment and Internet -Law Scholars amici, we have characterized the following -speech as commercial even if not a clear fit with the Supreme -Court’s above articulation: (i) targeted, individualized -solicitations, see Nationwide Biweekly Admin., Inc. v. Owen, -873 F.3d. 716, 731–32 (9th Cir. 2017); contract negotiations, -see S.F. Apartment Ass’n v. San Francisco, 881 F.3d 1169, -1177–78 (9th Cir. 2018); and retail product warnings, see -CTIA II, 928 F.3d at 845. Though it does not directly or -exclusively propose a commercial transaction, all of this -speech communicates the terms of an actual or potential -transaction. But the Content Category Reports go further: -they express a view about those terms by conveying whether -a company believes certain categories should be defined and -proscribed. - - X CORP. V. BONTA - -23 - -Second, the Content Category Reports fail to satisfy at -least two of the three Bolger factors. The compelled -disclosures are not advertisements. See Hunt, 638 F.3d at -715. Nor do the Content Category Reports merely disclose -existing commercial speech, so a social media company has -no economic motivation in their content. See id. The district -court found the same. The State does not dispute the district -court’s finding on appeal. Although the Bolger factors are -not dispositive, they are “important guideposts” to the -analysis and, here, further support the conclusion that the -compelled speech is non-commercial. See Ariix, LLC, 985 -F.3d at 1116. - -topics, - -Third, while a social media platform’s existing TOS and -content moderation policies may be commercial speech, its -opinions about and reasons for those policies are different in -character and kind. The Content Category Report provisions -would require 9 a social media company to convey the -company’s policy views on intensely debated and politically -fraught -racism, -misinformation, and radicalization, and also convey how the -company has applied its policies. The State suggests that -this requirement is subject to lower scrutiny because “it is -only a transparency measure” about the product. But even if -the Content Category Report provisions concern only -transparency, the relevant question here is: transparency into -what? Even a pure “transparency” measure, if it compels -non-commercial speech, is subject to strict scrutiny. See - -including - -speech, - -hate - -9 The State relies heavily on the fact that AB 587 does not affirmatively -require any social media company to opine on these topics, instead -requiring the company to convey its position only to the extent such a -policy already exists. That fact, however, is immaterial or at least non- -dispositive as to the nature of the speech being conveyed, which is -fundamentally non-commercial. - - 24 - -X CORP. V. BONTA - -Riley, 487 U.S. at 796–97. That is true of the Content -Category Report provisions. Insight into whether a social -media company considers, for example, (1) a post citing -rhetoric from on-campus protests to constitute hate speech; -(2) reports about a seized laptop to constitute foreign -political interference; or (3) posts about election fraud to -constitute misinformation -is sensitive, constitutionally -protected speech that the State could not otherwise compel a -social media company to disclose without satisfying strict -scrutiny. The mere fact that those beliefs are memorialized -in the company’s content moderation policy does not, by -itself, convert expression about -into -commercial speech. As X Corp. argues in its reply brief, -such a rule would be untenable. It would mean that basically -any compelled disclosure by any business about its activities -would be commercial and subject to a lower tier of scrutiny, -no matter how political in nature. Protection under the First -Amendment cannot be vitiated so easily.10 - -those beliefs - -The district court performed, essentially, no analysis on -this question. In fact, the district court acknowledged that -the Content Category Reports “do not so easily fit the -traditional definition of commercial speech” as they “are not -advertisements, and social media companies have no - -10 For substantially the same reason, nor can the test for whether speech -is commercial or non-commercial turn on whether the speech is “directed -to potential consumers and may presumably play a role in the decision -of whether to use the platform,” as the district court seemed to suggest. -Consider, for example, a state law that compels a social media company -to disclose the political affiliations of its managers. That information -could conceivably “play a role in the [potential consumer’s] decision of -whether to use the platform”—i.e., if the consumer is concerned about -the platform’s content being politically skewed. It could not be that such -a law compels only commercial speech subject to a lower tier of scrutiny. - - X CORP. V. BONTA - -25 - -to provide - -particular economic motivation -them.” -Nevertheless, the court applied Zauderer, suggesting the -compelled speech is commercial. See Nat’l Wheat, 85 F.4th -at 1275 (identifying Central Hudson Gas & Electric Corp. -v. Public Service Commission of New York, 447 U.S. 557 -(1980), and Zauderer as “two levels of scrutiny governing -compelled commercial speech” (emphasis added)); CTIA II, -928 F.3d at 843 (endorsing proposition that Zauderer is -merely the “exception to the general rule of Central -Hudson”). The district court offered no reason for that -decision except for wanting to “follow[] the lead of the Fifth -and Eleventh Circuits.” - -But neither the Fifth nor Eleventh Circuit dealt with -speech similar to the Content Category Reports. Unlike -Texas HB 20 or Florida SB 7072, the Content Category -Report provisions compel social media companies to report -whether and how they believe particular, controversial -categories of content should be defined and regulated on -their platforms. Neither the Texas nor Florida provisions at -issue in the NetChoice cases require a company to disclose -the existence or substance of its policies addressing such -categories. See NetChoice (Tex.), 49 F.4th at 446 (requiring -platforms to disclose “how they moderate and promote -content” and provide “high-level statistics” about their -moderation efforts without mention of controversial topics); -NetChoice (Fla.), 34 F.4th at 1206–07 (requiring platforms -to disclose information about their content-moderation -“standards” and “rule changes” without regard to particular -content categories). Though perhaps relevant to an analysis -of sections 22677(a)(1), (2), and (4)(B)–(E), these cases are -unhelpful on the issue of the Content Category Reports and -offer no compelling reason to apply Zauderer. - - 26 - -X CORP. V. BONTA - -For these reasons, we conclude that the Content -Category Report provisions compel non-commercial speech. -Because the provisions are content-based, which the State -does not contest, they are subject to strict scrutiny. See Nat’l -Inst. of Fam. & Life Advocs., 585 U.S. at 766.11 - -B. The Content Category Report provisions likely - -fail strict scrutiny. - -Strict scrutiny “is a demanding standard.” Brown v. Ent. -Merchants Ass’n, 564 U.S. 786, 799 (2011). “It is rare that -a regulation restricting speech because of its content will -ever be permissible.” United States v. Playboy Ent. Grp., -Inc., 529 U.S. 803, 818 (2000). A state must show that the -statute “furthers a compelling governmental interest and is -narrowly tailored to that end.” Reed, 576 U.S. at 171. “If a -less restrictive alternative would serve the [g]overnment’s -purpose, the legislature must use that alternative.” Playboy -Ent. Grp., Inc., 529 U.S. at 813. - -At minimum, the Content Category Report provisions -likely fail under strict scrutiny because they are not narrowly -tailored. They are more extensive than necessary to serve -the State’s purported goal of “requiring social media -companies to be transparent about their content-moderation -policies and practices so that consumers can make informed -decisions about where they consume and disseminate news -and information.” Consumers would still be meaningfully -informed if, for example, a company disclosed whether it - -11 X Corp. argues that strict scrutiny applies for the following additional -reasons: because AB 587 is viewpoint discriminatory, interferes with a -social media company’s constitutionally protected editorial judgment, -and regulates “speech about speech.” Several of the amici raise similar -arguments. Because we agree that strict scrutiny applies, we need not -reach these arguments. - - X CORP. V. BONTA - -27 - -was moderating certain categories of speech without having -to define those categories in a public report. Or, perhaps, a -company could be compelled to disclose a sample of posts -that have been removed without requiring the company to -explain why or on what grounds.12 - -In any event, the State does not attempt to argue that the -law survives strict scrutiny. For the reasons above, X Corp. -has shown a likelihood of success on the merits of its First -Amendment claim as to sections 22677(a)(3), (a)(4)(A), and -(a)(5). - -C. The remaining Winter factors weigh in favor of a - -preliminary injunction. - -With respect to the second factor, a loss of First -Amendment freedoms constitutes an irreparable injury. See -Fellowship of Christian Athletes, 82 F.4th at 694 (“It is -axiomatic that ‘[t]he loss of First Amendment freedoms, for -even minimal periods of time, unquestionably constitutes -irreparable injury.’” (citation omitted)). Because X Corp. -has a colorable First Amendment claim, it has demonstrated -that it likely will suffer irreparable harm. See Am. Bev. Ass’n -v. San Francisco, 916 F.3d 749, 758 (9th Cir. 2019) (en -banc). - -The third and fourth factors—balance of equities and -public interest—also favor X Corp. “[I]t is always in the -public interest to prevent the violation of a party’s -constitutional rights.” Fellowship of Christian Athletes, 82 -F.4th at 695 (citation omitted). When a party “‘raise[s] -serious First Amendment questions,’ that alone ‘compels a - -12 We do not opine on whether such laws would survive constitutional -scrutiny. They are offered only to illustrate that the Content Category -Report provisions are not narrowly tailored to the State’s interest. - - 28 - -X CORP. V. BONTA - -finding that the balance of hardships tips sharply in [its] -favor.’” Id. (second alteration in original) (quoting Am. Bev. -Ass’n, 916 F.3d at 758). The government reasonably has an -interest in transparency by social media platforms. But even -“undeniably admirable goals” “must yield” when they -“collide with the . . . Constitution.” Id. - -Because X Corp. has shown a likelihood of success on -the merits of its First Amendment claim, and the remaining -Winter factors weigh in favor of an injunction, we reverse -the district court’s decision denying a preliminary injunction -as to AB 587’s Content Category Report provisions. - -II. We remand to the district court to determine whether -the Content Category Report provisions are likely -severable from the remainder of AB 587. - -“Severability is . . . a matter of state law.” Sam Francis -Found. v. Christies, Inc., 784 F.3d 1320, 1325 (9th Cir. -2015) (en banc) (alteration in original) (quoting Leavitt v. -Jane L., 518 U.S. 137, 139 (1996) (per curiam)). “In -California, the presence of a severability clause in a statutory -scheme that contains an invalid provision ‘normally calls for -sustaining the valid part of the enactment.’” Garcia v. City -of Los Angeles, 11 F.4th 1113, 1120 (9th Cir. 2021) (quoting -Cal. Redevelopment Ass’n v. Matosantos, 267 P.3d 580, 607 -(Cal. 2011)). - -The parties did not brief severability on appeal, and the -severability arguments below appear to have been cursory. -During oral argument, counsel for the State suggested that, -were we to find that any part of the statute should be -enjoined, the issue of severability should be remanded. We -agree and leave it to the district court to determine in the first -instance whether the likely unconstitutional provisions of -AB 587, sections 22677(a)(3), (a)(4)(A), and (a)(5), are - - X CORP. V. BONTA - -29 - -severable from its remainder. See generally Detrich v. Ryan, -740 F.3d 1237, 1248–49 (9th Cir. 2013) (en banc) (observing -that it is “standard practice . . . to remand to the district court -for a decision in the first instance without requiring any -special justification for so doing”), overruled on other -grounds by Shinn v. Ramirez, 596 U.S. 366 (2022). - -CONCLUSION - -For the foregoing reasons, we REVERSE the district -court’s denial of a preliminary injunction as to California -Business and Professions Code sections 22677(a)(3), -(a)(4)(A), and (a)(5). We remand with instructions to enter -a preliminary injunction consistent with this opinion and to -determine whether these provisions are severable from the -remainder of AB 587 and, if so, which, if any, of the -remaining challenged provisions should also be enjoined. +Notebook PC User Guide 16 + +E15695_X512FJ_FL_UM_V2_16pages.indd 16 + +2019/12/27 �� 09:59:09 +Notebook PC User Guide \ No newline at end of file diff --git a/q110/random_k4/random_4.md b/q110/random_k4/random_4.md index b3f566f113a69a4c8533f2f4d870e45002f6953a..58207d8b1050f8f9278b645b3fe87a602dde59b2 100644 --- a/q110/random_k4/random_4.md +++ b/q110/random_k4/random_4.md @@ -1,1617 +1,614 @@ -0 -2 -0 -2 +Lecture – 3 +Cell division : Mitosis +& Meiosis +by -p -e -S -7 +Dr. Praveen Kumar -] +Asst. Prof. GPB -Y -C -. -s -c -[ +MSSSoA, CUTM, Odisha -1 -v -0 -0 -3 -3 -0 -. -9 -0 -0 -2 -: -v -i -X -r -a + Cell Division -Measuring Massive Multitask Language Understanding +All cells are derived from -Dan Hendrycks -UC Berkeley +pre-existing cells -Collin Burns -Columbia University +New cells are produced for -Steven Basart -UChicago +growth and to replace +damaged or old cells +Differs in prokaryotes -Andy Zou -UC Berkeley +(bacteria) and eukaryotes +(protists, fungi, plants, & +animals) -Mantas Mazeika -UIUC +2 -Dawn Song -UC Berkeley + Original DNA +strand + +DNA Replication +DNA must +be copied +or +replicated +before cell +division +Each new +cell will +then have +an +identical +copy of + +Two new, +identical DNA +strands -Jacob Steinhardt -UC Berkeley +3 -ABSTRACT + Cell Cycle -We propose a new test to measure a text model’s multitask accuracy. The test -covers 57 tasks including elementary mathematics, US history, computer science, -law, and more. To attain high accuracy on this test, models must possess extensive -world knowledge and problem solving ability. We find that while most recent -models have near random-chance accuracy, the very largest GPT-3 model improves -over random chance by almost 20 percentage points on average. However, on every -one of the 57 tasks, the best models still need substantial improvements before -they can reach human-level accuracy. Models also have lopsided performance -and frequently do not know when they are wrong. Worse, they still have near- -random accuracy on some socially important subjects such as morality and law. -By comprehensively evaluating the breadth and depth of a model’s academic and -professional understanding, our test can be used to analyze models across many -tasks and to identify important shortcomings. +DNA Copied -1 +Cells prepare for Division -INTRODUCTION - -Natural Language Processing (NLP) models have achieved superhuman performance on a number of -recently proposed benchmarks. However, these models are still well below human level performance -for language understanding as a whole, suggesting a disconnect between our benchmarks and the -actual capabilities of these models. The General Language Understanding Evaluation benchmark -(GLUE) (Wang et al., 2018) was introduced in 2018 to evaluate performance on a wide range of NLP -tasks, and top models achieved superhuman performance within a year. To address the shortcomings -of GLUE, researchers designed the SuperGLUE benchmark with more difficult tasks (Wang et al., -2019). About a year since the release of SuperGLUE, performance is again essentially human-level -(Raffel et al., 2019). While these benchmarks evaluate linguistic skills more than overall language -understanding, an array of commonsense benchmarks have been proposed to measure basic reasoning -and everyday knowledge (Zellers et al., 2019; Huang et al., 2019; Bisk et al., 2019). However, these -recent benchmarks have similarly seen rapid progress (Khashabi et al., 2020). Overall, the near -human-level performance on these benchmarks suggests that they are not capturing important facets -of language understanding. - -Transformer models have driven this recent progress by pretraining on massive text corpora, including -all of Wikipedia, thousands of books, and numerous websites. These models consequently see -extensive information about specialized topics, most of which is not assessed by existing NLP -benchmarks. It consequently remains an open question just how capable current language models are -at learning and applying knowledge from many domains. - -To bridge the gap between the wide-ranging knowledge that models see during pretraining and the -existing measures of success, we introduce a new benchmark for assessing models across a diverse -set of subjects that humans learn. We design the benchmark to measure knowledge acquired during -pretraining by evaluating models exclusively in zero-shot and few-shot settings. This makes the -benchmark more challenging and more similar to how we evaluate humans. The benchmark covers -57 subjects across STEM, the humanities, the social sciences, and more. It ranges in difficulty -from an elementary level to an advanced professional level, and it tests both world knowledge and -problem solving ability. Subjects range from traditional areas, such as mathematics and history, to +Cells +Mature -1 +4 - (a) An example of few-shot learning and inference us- -ing GPT-3. The blue underlined bold text is the auto- -completed response from GPT-3, while the preceding -text is the user-inputted prompt. In this 2-shot learning -example, there are two instruction examples and one -initially incomplete example. On average, GPT-3 has -low accuracy on high school mathematics questions. - -(b) Performance on a commonsense benchmark (Hel- -laSwag), a linguistic understanding benchmark (Super- -GLUE), and the massive multitask test. On previous -benchmarks, smaller models start well above random -chance levels and exhibit more continuous improve- -ments with model size increases, but on our test, GPT-3 -moves beyond random chance with the largest model. - -more specialized areas like law and ethics. The granularity and breadth of the subjects makes the -benchmark ideal for identifying a model’s blind spots. - -We find that meaningful progress on our benchmark has only become possible in recent months. -In particular, models up to 13 billion parameters (Brown et al., 2020) achieve random chance -performance of 25% accuracy, but the 175 billion parameter GPT-3 model reaches a much higher -43.9% accuracy (see Figure 1b). On the other hand, unlike human professionals GPT-3 does not -excel at any single subject. Instead, we find that performance is lopsided, with GPT-3 having almost -70% accuracy for its best subject but near-random performance for several other subjects. - -Our results indicate that while recent advances have been impressive, state-of-the-art models still -struggle at learning and applying knowledge from pretraining. The tasks with near-random accuracy -include calculation-heavy subjects such as physics and mathematics and subjects related to human -values such as law and morality. This second weakness is particularly concerning because it will -be important for future models to have a strong understanding of what is legal and what is ethical. -Worryingly, we also find that GPT-3 does not have an accurate sense of what it does or does not know -since its average confidence can be up to 24% off from its actual accuracy. We comprehensively -evaluate the breadth and depth of a model’s text understanding by covering numerous topics that -humans are incentivized to learn. Since our test consists in 57 tasks, it can be used to analyze -aggregate properties of models across tasks and to track important shortcomings. The test and code is -available at github.com/hendrycks/test. - -2 RELATED WORK - -Pretraining. The dominant paradigm in NLP is to pretrain large models on massive text corpora -including educational books and websites. In the process, these models are exposed to information -about a wide range of topics. Petroni et al. (2019) found that recent models learn enough information -from pretraining that they can serve as knowledge bases. However, no prior work has comprehensively -measured the knowledge models have across many real-world domains. - -Until recently, researchers primarily used fine-tuned models on downstream tasks (Devlin et al., 2019). -However, larger pretrained models like GPT-3 (Brown et al., 2020) have made it possible to achieve -competitive performance without fine-tuning by using few-shot learning, which removes the need for -a large fine-tuning set. With the advent of strong zero-shot and few-shot learning, it is now possible -to curate a diverse set of tasks for evaluation and remove the possibility of models on “spurious cues” -(Geirhos et al., 2020; Hendrycks et al., 2019b) in a dataset to achieve high performance. +Daughter Cells -2 +Cell Divides into Identical cells -Few Shot Prompt and Predicted AnswerHow many numbers are in the list 25, 26, ..., 100?(A) 75 (B) 76 (C) 22 (D) 23Answer: BCompute i+i2+i3+···+i258+i259.(A) -1 (B) 1 (C) i (D) -iAnswer: AIf 4 daps = 7 yaps, and 5 yaps = 3 baps,how many daps equal 42 baps?(A) 28 (B) 21 (C) 40 (D) 30Answer: C␣The following are multiple choice questionsabout high school mathematics.SmallMediumLargeX-LargeModel Size2030405060708090Performance (%)GPT-3 Few Shot Test PerformanceCommonsenseLinguisticsKnowledge (Ours) Benchmarks. Many recent benchmarks aim to assess a model’s general world knowledge and basic -reasoning ability by testing its “commonsense.” A number of commonsense benchmarks have been -proposed in the past year, but recent models are already nearing human-level performance on several -of these, including HellaSwag (Zellers et al., 2019), Physical IQA (Bisk et al., 2019), and CosmosQA -(Huang et al., 2019). By design, these datasets assess abilities that almost every child has. In contrast, -we include harder specialized subjects that people must study to learn. - -Some researchers have suggested that the future of NLP evaluation should focus on Natural Language -Generation (NLG) (Zellers et al., 2020), an idea that reaches back to the Turing Test (Turing, 1950). -However, NLG is notoriously difficult to evaluate and lacks a standard metric (Sai et al., 2020). -Consequently, we instead create a simple-to-evaluate test that measures classification accuracy on -multiple choice questions. - -While several question answering benchmarks exist, they are comparatively limited in scope. Most -either cover easy topics like grade school subjects for which models can already achieve strong -performance (Clark et al., 2018; Khot et al., 2019; Mihaylov et al., 2018; Clark et al., 2019), or -are focused on linguistic understanding in the form of reading comprehension (Lai et al., 2017; -Richardson et al., 2013). In contrast, we include a wide range of difficult subjects that go far beyond -linguistic understanding. - -3 A MULTITASK TEST - -We create a massive multitask test consisting of multiple-choice questions from various branches of -knowledge. The test spans subjects in the humanities, social sciences, hard sciences, and other areas -that are important for some people to learn. There are 57 tasks in total, which is also the number -of Atari games (Bellemare et al., 2013), all of which are listed in Appendix B. The questions in -the dataset were manually collected by graduate and undergraduate students from freely available -sources online. These include practice questions for tests such as the Graduate Record Examination -and the United States Medical Licensing Examination. It also includes questions designed for -undergraduate courses and questions designed for readers of Oxford University Press books. Some -tasks cover a subject, like psychology, but at a specific level of difficulty, such as “Elementary,” -“High School,” “College,” or “Professional.” For example, the “Professional Psychology” task draws -on questions from freely available practice questions for the Examination for Professional Practice -in Psychology, while the “High School Psychology” task has questions like those from Advanced -Placement Psychology examinations. - -We collected 15908 questions in total, which we split into a few-shot development set, a validation -set, and a test set. The few-shot development set has 5 questions per subject, the validation set may -be used for selecting hyperparameters and is made of 1543 questions, and the test set has 14080 -questions. Each subject contains 100 test examples at the minimum, which is longer than most exams -designed to assess people. - -Since our test aggregates different subjects and several levels of difficulty, we measure more than -straightforward commonsense or narrow linguistic understanding. Instead, we measure arbitrary -real-world text understanding. Since models are pretrained on the Internet, this enables us to test how -well they can extract useful knowledge from massive corpora. To succeed at our test, future models -should be well-rounded, possess extensive world knowledge, and develop expert-level problem -solving ability. These properties make the test likely to be an enduring and informative goalpost. - -3.1 HUMANITIES - -The humanities is a group of disciplines that make use of qualitative analysis and analytic methods -rather than scientific empirical methods. Branches of the humanities include law, philosophy, history, -and so on (Appendix B). Mastering these subjects requires a variety of skills. For example, legal -understanding requires knowledge of how to apply rules and standards to complex scenarios, and -also provide answers with stipulations and explanations. We illustrate this in Figure 2. Legal -understanding is also necessary for understanding and following rules and regulations, a necessary -capability to constrain open-world machine learning models. For philosophy, our questions cover -concepts like logical fallacies, formal logic, and famous philosophical arguments. It also covers -moral scenarios, including questions from the ETHICS dataset (Hendrycks et al., 2020) that test a -model’s understanding of normative statements through predicting widespread moral intuitions about + Interphase - G1 Stage +1st growth stage after cell -3 +division - Figure 2: This task requires understanding detailed and dissonant scenarios, applying appropriate -legal precedents, and choosing the correct explanation. The green checkmark is the ground truth. +Cells mature by making -diverse everyday scenarios. Finally, our history questions cover a wide range of time periods and -geographical locations, including prehistory and other advanced subjects. +more cytoplasm & +organelles -3.2 SOCIAL SCIENCE +Cell carries on its normal -Social science includes branches of knowledge that examine human behavior and society. Subject -areas include economics, sociology, politics, geography, psychology, and so on. See Figure 3 -for example questions. Our economics questions include microeconomics, macroeconomics, and -econometrics, and cover different types of problems, including questions that require a mixture of -world knowledge, qualitative reasoning, or quantitative reasoning. We also include important but -more esoteric topics such as security studies in order to test the boundaries of what is experienced and -learned during pretraining. Social science also includes psychology, a field that may be especially -important for attaining a nuanced understanding of humans. +metabolic activities -Figure 3: Examples from the Microeconomics and Security Studies social science tasks. +5 -3.3 SCIENCE, TECHNOLOGY, ENGINEERING, AND MATHEMATICS (STEM) + Interphase – S Stage -STEM subjects include physics, computer science, mathematics, and more. Two examples are shown -in Figure 4. Conceptual physics tests understanding of simple physics principles and may be thought -of as a harder version of the physical commonsense benchmark Physical IQA (Bisk et al., 2019). We -also test mathematical problem solving ability at various levels of difficulty, from the elementary to -the college level. College mathematics questions, like those found on the GRE mathematics subject -test, often require chains of reasoning and abstract knowledge. To encode mathematics expressions, -we use LaTeX or symbols such as * and ˆ for multiplication and exponentiation respectively. STEM -subjects require knowledge of empirical methods, fluid intelligence, and procedural knowledge. +Synthesis stage +DNA is copied or replicated -3.4 OTHER +Two identical +copies of +DNA -There is a long tail of subjects that either do not neatly fit into any of the three preceding categories or -for which there are not thousands of freely available questions. We put these subjects into Other. This +Original DNA -4 +6 -As Seller, an encyclopedia salesman, approached the grounds on which Hermit's house was situated,he saw a sign that said, "No salesmen. Trespassers will be prosecuted. Proceed at your own risk."Although Seller had not been invited to enter, he ignored the sign and drove up the driveway towardthe house. As he rounded a curve, a powerful explosive charge buried in the driveway exploded, andSeller was injured. Can Seller recover damages from Hermit for his injuries?(A) Yes, unless Hermit, when he planted the charge, intended only to deter, not harm, intruders.(B) Yes, if Hermit was responsible for the explosive charge under the driveway.(C) No, because Seller ignored the sign, which warned him against proceeding further.(D) No, if Hermit reasonably feared that intruders would come and harm him or his family.Professional LawOne of the reasons that the government discourages and regulates monopolies is that(A) producer surplus is lost and consumer surplus is gained.(B) monopoly prices ensure productive efficiency but cost society allocative efficiency.(C) monopoly firms do not engage in significant research and development.(D) consumer surplus is lost with higher prices and lower levels of output.MicroeconomicsSecurity StudiesWhy is HIV/AIDS considered to be a non-traditional security issue?(A) HIV/AIDS is a newly emergent disease that did not exist until the later 20th century.(B) Ill health can threaten the state indirectly, but can also be a threat to other referent objects such as the economy.(C) The overwhelming scale of HIV/AIDS requires a new approach in order to theorize its impact on security.(D) None of the above - HIV/AIDS is better defined as a traditional security issue. Figure 4: Examples from the Conceptual Physics and College Mathematics STEM tasks. - -section includes the Professional Medicine task, which has difficult questions that require humans -many years of study to master. An example is depicted in Figure 5. This section also contains -business topics like finance, accounting, and marketing, as well as knowledge of global facts. The -latter includes statistics about poverty in different countries over time, which may be necessary for -having an accurate model of the world internationally. - -Figure 5: A question from Professional Medicine which is a simulated question from the United -States Medical Licensing Examination. -4 EXPERIMENTS - -4.1 SETUP - -Assessment and Models. To measure performance on our multitask benchmark, we compute the -classification accuracy across all examples and tasks. We evaluate GPT-3 (Brown et al., 2020) and -UnifiedQA (Khashabi et al., 2020). For GPT-3 we use the OpenAI API, which provides access -to four model variants, “Ada,” “Babbage,” “Curie,” and “Davinci,” which we refer to as “Small” -(2.7 billion parameters), “Medium” (6.7 billion), “Large” (13 billion) and “X-Large” (175 billion). -UnifiedQA uses the T5 (Raffel et al., 2019) text-to-text backbone and is fine-tuned on previously -proposed question answering datasets (Lai et al., 2017), where the prediction is the class with the - -Model -Random Baseline -T5 -UnifiedQA -GPT-3 Small -GPT-3 Medium -GPT-3 Large -GPT-3 X-Large - -Humanities -25.0 -24.9 -38.0 -24.4 -26.1 -27.1 -40.8 - -Social Science -25.0 -25.2 -41.5 -30.9 -21.6 -25.6 -50.4 - -STEM -25.0 -24.8 -32.2 -26.0 -25.6 -24.3 -36.7 - -Other -25.0 -24.4 -42.1 -24.1 -25.5 -26.5 -48.8 - -Average -25.0 -24.8 -38.5 -25.9 -24.9 -26.0 -43.9 - -Table 1: Average weighted accuracy for each model on all four broad disciplines. All values are -percentages. Models proposed in the past few months (UnifiedQA, GPT-3) can move several percent -points beyond random chance on this test. + Interphase – G2 Stage -5 +2nd Growth Stage +Occurs after DNA has been copied +All cell structures needed for -When you drop a ball from rest it accelerates downward at 9.8 m/s². If you instead throw itdownward assuming no air resistance its acceleration immediately after leaving your hand is(A) 9.8 m/s²(B) more than 9.8 m/s²(C) less than 9.8 m/s²(D) Cannot say unless the speed of throw is given.ConceptualPhysicsCollegeMathematicsIn the complex z-plane, the set of points satisfying the equation z² = |z|² is a(A) pair of points(B) circle(C) half-line(D) lineA 33-year-old man undergoes a radical thyroidectomy for thyroid cancer. During the operation,moderate hemorrhaging requires ligation of several vessels in the left side of the neck.Postoperatively, serum studies show a calcium concentration of 7.5 mg/dL, albumin concentrationof 4 g/dL, and parathyroid hormone concentration of 200 pg/mL. Damage to which of the followingvessels caused the findings in this patient?(A) Branch of the costocervical trunk(B) Branch of the external carotid artery(C) Branch of the thyrocervical trunk(D) Tributary of the internal jugular veinProfessional Medicine highest token overlap with UnifiedQA’s text output. Since UnifiedQA is fine-tuned on other datasets, -we evaluate it without any further tuning to assess its transfer accuracy. - -Few-Shot Prompt. We feed GPT-3 prompts -like that shown in Figure 1a. We begin each -prompt with “The following are multiple choice -questions (with answers) about [subject].” For -zero-shot evaluation, we append the question to -the prompt. For few-shot evaluation, we add up -to 5 demonstration examples with answers to -the prompt before appending the question. All -prompts end with “Answer: ”. The model then -produces probabilities for the tokens “A,” “B,” -“C,” and “D,” and we treat the highest probability -option as the prediction. For consistent evalua- -tion, we create a dev set with 5 fixed few-shot -examples for each subject. - -4.2 RESULTS - -Model Size and Accuracy. We compare the -few-shot accuracy of each GPT-3 size in Table 1. -We find that the three smaller GPT-3 models -have near random accuracy (around 25%). We -also assess the 11 billion parameter T5 model -in a few-shot setting and confirmed that it like- -wise has random chance accuracy. In contrast, -we find that the X-Large 175 billion parame- -ter GPT-3 model performs substantially better -than random, with an accuracy of 43.9%. We -also find qualitatively similar results in the zero- -shot setting. While the smaller models have -around 25% zero-shot accuracy, Figure 8a in -Appendix A shows that the largest GPT-3 model -has a much higher zero-shot accuracy of about -37.7%. In Figure 1b we show that non-random -accuracy on the multitask test emerged with re- -cent large few-shot models compared to datasets -that assess commonsense and linguistic under- -standing. - -To test the importance of model size for other -methods, we also evaluate UnifiedQA models. -UnifiedQA has the advantage of being fine- -tuned on other question answering datasets, and -we assess it by evaluating its transfer perfor- -mance without any additional fine-tuning. The -largest UnifiedQA model we test has 3 billion -parameters, which is slightly larger than GPT-3 -Small. Nevertheless, we show in Table 1 that it -attains 38.5% accuracy. This is worse than few- -shot GPT-3 X-Large accuracy but higher than -zero-shot GPT-3 X-Large, despite UnifiedQA having two orders of magnitude fewer parameters. We -also find that even the smallest UnifiedQA variant, with just 60 million parameters, has approximately -30% accuracy. These results suggest that while model size is a key component for achieving strong -performance, it is not the only important factor. - -Figure 6: GPT-3 few shot accuracies for all of the -57 tasks. All task accuracies are markedly below -expert-level performance. - -Comparing Disciplines. Using our test, we discover that GPT-3 has lopsided performance and -several substantial knowledge gaps. Figure 6 shows the few-shot accuracy of GPT-3 for all 57 tasks. +division are made (e.g. centrioles) -6 +Both organelles & proteins are -020406080100Accuracy (%)College ChemistryMoral ScenariosCollege PhysicsHigh School PhysicsHigh School MathematicsFormal LogicElementary MathematicsAbstract AlgebraHigh School StatisticsMachine LearningEconometricsHigh School ChemistryProfessional AccountingProfessional LawCollege MathematicsProfessional MedicineConceptual PhysicsGlobal FactsHigh School Comp SciMedical GeneticsHigh School MacroeconomicsHigh School MicroeconomicsMoral DisputesProfessional PsychologyCollege BiologyVirologyCollege Comp SciBusiness EthicsNutritionCollege MedicineAnatomyClinical KnowledgeLogical FallaciesHigh School BiologyPublic RelationsAstronomyElectrical EngineeringHuman AgingPhilosophySecurity StudiesPrehistoryHigh School US HistorySociologyHigh School European HistoryHuman SexualityJurisprudenceWorld ReligionsInternational LawHigh School World HistoryManagementComputer SecurityHigh School GeographyHigh School Gov't and PoliticsMarketingMiscellaneousHigh School PsychologyUS Foreign PolicyRandom Chance (a) GPT-3’s completion for two prompts testing knowl- -edge of the order of operations. The blue underlined -bold text is the autocompleted response from GPT-3. -While it is has descriptive knowledge and knows about -of the order of operations, it does not know how to apply -its knowledge and does not obey operator precedence. (b) GPT-3’s average confidence is a poor estimator of - -its accuracy and can be off by up to 24%. - -It shows the GPT-3 is below expert-level performance for all tasks, with accuracy ranging from 69% -for US Foreign Policy to 26% for College Chemistry. - -Overall, GPT-3 does poorly on highly procedural problems. Figure 6 shows that calculation-heavy -STEM subjects tend to have low accuracy compared to verbal subjects. In fact, 9 out of the 10 -lowest-accuracy tasks are STEM subjects that emphasize mathematics or calculations. We speculate -that is in part because GPT-3 acquires declarative knowledge more readily than procedural knowlege. -For example, many questions in Elementary Mathematics require applying the order of operations -for arithmetic, which is described by the acronym PEMDAS (Parentheses Exponents Multiplication -Division Addition Subtraction). In Figure 7a, we confirm that GPT-3 is aware of the acronymn -PEMDAS. However, it does not consistently apply PEMDAS to actual problems. On the other hand, -procedural understanding is not its only weak point. We find that some verbal tasks such as Moral -Scenarios (Hendrycks et al., 2020) and Professional Law also have especially low accuracy. - -Our test also shows that GPT-3 acquires knowledge quite unlike humans. For example, GPT-3 learns -about topics in a pedagogically unusual order. GPT-3 does better on College Medicine (47.4%) -and College Mathematics (35.0%) than calculation-heavy Elementary Mathematics (29.9%). GPT-3 -demonstrates unusual breadth, but it does not master a single subject. In this way, our test shows that -GPT-3 has many knowledge blindspots and has capabilities that are lopsided. - -Calibration. We should not trust a model’s prediction unless the model is calibrated, meaning -that its confidence is a good estimate of the actual probability the prediction is correct. However, -large neural networks are often miscalibrated (Guo et al., 2017), especially under distribution shift -(Ovadia et al., 2019). We evaluate the calibration of GPT-3 by testing how well its average confidence -estimates its actual accuracy for each subject. We show the results in Figure 7b, which demonstrates -that GPT-3 is uncalibrated. In fact, its confidence is only weakly related to its actual accuracy in -the zero-shot setting, with the difference between its accuracy and confidence reaching up to 24% -for some subjects. Another calibration measure is the Root Mean Squared (RMS) calibration error -(Hendrycks et al., 2019a; Kumar et al., 2019). Many tasks have miscalibrated predictions, such -as Elementary Mathematics which has a zero-shot RMS calibration error of 19.4%. These results -suggest that model calibration has wide room for improvement. - -5 DISCUSSION - -Multimodal Understanding. While text is capable of conveying an enormous number of concepts -about the world, many important concepts are conveyed mainly through other modalities, such as -images, audio, and physical interaction (Bisk et al., 2020). Existing large-scale NLP models, such as -GPT-3, do not incorporate multimodal information, so we design our benchmark to capture a diverse -array of tasks in a text-only format. However, as models gain the ability to process multimodal inputs, -benchmarks should be designed to reflect this change. One such benchmark could be a “Turk Test,” -consisting of Amazon Mechanical Turk Human Intelligence Tasks. These are well-defined tasks that -require models to interact with flexible formats and demonstrate multimodal understanding. - -The Internet as a Training Set. A major distinction between our benchmark and previous multitask -NLP benchmarks is that we do not require large training sets. Instead, we assume that models have +synthesized 7 -Declarative vs. Procedural KnowledgePrompt and Completion:The order of operations or PEMDAS isParentheses Exponents MultiplicationDivision Addition SubtractionPrompt and Completion:(1 + 1) × 2 = 3␣2030405060Confidence (%)2030405060Accuracy (%)GPT-3 Zero-Shot CalibrationFormal LogicMarketing acquired the requisite knowledge from reading vast quantities of diverse text from the Internet. This -process is typically called pretraining, but it can be thought of as training in its own right, where the -downstream evaluation is demonstrating whatever knowledge we would expect a human to pick up -from reading the same text. - -This motivates us to propose a methodological change so that models are trained more like how -humans learn. While most previous machine learning benchmarks have models learn from a large -question bank, humans primarily learn new subjects by reading books and listening to others talk -about the topic. For specialized subjects such as Professional Law, massive legal corpora are available, -such as the 164-volume legal encyclopedia Corpus Juris Secundum, but there are fewer than 5,000 -multistate bar exam questions available. Learning the entire law exclusively through a small number -of practice tests is implausible, so future models must learn more during pretraining. - -For this reason we assess pretrained models in a zero-shot or few-shot setting and we provide a dev, -val, and test set for each task. The dev set is used for few-shot prompts, the val set could be used for -hyperparameter tuning, and the test set is used to compute the final accuracy. Importantly, the format -of our evaluation is not identical to the format in which information is acquired during pretraining. -This has the benefit of obviating concerns about spurious training set annotation artifacts (Geirhos -et al., 2020; Hendrycks et al., 2019b) and is in stark contrast to the previous paradigm of identically -distributed training and test sets. This change also enables collecting a much more extensive and -diverse set of tasks for evaluation. We anticipate our methodology becoming more widespread as -models improve at extracting information from diverse online sources. - -Model Limitations. We find that current large-scale Transformers have wide room for improvement. -They are notably poor at modeling human (dis)approval, as evident by the low performance on the -Professional Law and Moral Scenarios tasks. For future systems to be aligned with human values, high -performance on these tasks is crucial (Hendrycks et al., 2020), so future research should especially -aim to increase accuracy on these tasks. Models also have difficulty performing calculations, so much -so that they exhibit poor performance on Elementary Mathematics and many other STEM subjects -with “plug and chug” problems. Additionally, they do not match expert-level performance on any -subject, so for all subjects it is subhuman. On average, models are only now starting to move beyond -random-chance accuracy levels. - -Addressing these shortcomings may be challenging. To illustrate this, we attempted to create a better -Professional Law model by pretraining on specialized data but achieved only limited success. We -collected approximately 2,000 additional Professional Law training examples. After fine-tuning a -RoBERTa-base model (Liu et al., 2019) using this custom training set, our model attained 32.8% test -accuracy. To test the impact of additional specialized training data, we also had RoBERTa continue -pretraining on approximately 1.6 million legal case summaries using Harvard’s Law Library case law -corpus case.law, but after fine-tuning it only attained 36.1% accuracy. This suggests that while -additional pretraining on relevant high quality text can help, it may not be enough to substantially -increase the performance of current models. - -It is unclear whether simply scaling up existing language models will solve the test. Current -understanding indicates that a 10× increase in model size must be accompanied by an approximate -5× increase in data (Kaplan et al., 2020). Aside from the tremendous expense in creating multi-trillion -parameter language models, data may also become a bottleneck, as there is far less written about -esoteric branches of knowledge than about everyday text. - -6 CONCLUSION - -We introduced a new test that measures how well text models can learn and apply knowledge -encountered during pretraining. By covering 57 subjects at varying levels of difficulty, the test -assesses language understanding in greater breadth and depth than previous benchmarks. We found -that it has recently become possible for models to make meaningful progress on the test, but that -state-of-the-art models have lopsided performance and still do not excel at any individual task. -We also showed that current models are uncalibrated and have difficulty with tasks that require -calculations. Worryingly, models also perform especially poorly on socially relevant subjects -including morality and law. Our expansive test can help researchers pinpoint important shortcomings -of models, making it easier to gain a clearer picture of state-of-the-art capabilities. + What’s Happening in Interphase? + +What the cell looks like + +Animal Cell + +What’s occurring 8 - 6.1 ACKNOWLEDGEMENTS + Mitosis -We would like to thank the following for their helpful comments: Jan Leike, David Krueger, Alex -Tamkin, Girish Sastry, and Henry Zhu. DH is supported by the NSF GRFP Fellowship and an Open -Philanthropy Project Fellowship. This research was also supported by the NSF Frontier Award -1804794. +Division of the -REFERENCES +nucleus -M. G. Bellemare, Y. Naddaf, J. Veness, and M. Bowling. The arcade learning environment: An -evaluation platform for general agents (extended abstract). J. Artif. Intell. Res., 47:253–279, 2013. +Also called +karyokinesis +Only occurs in -Y. Bisk, R. Zellers, R. L. Bras, J. Gao, and Y. Choi. Piqa: Reasoning about physical commonsense in +eukaryotes -natural language, 2019. +Has four stages +Doesn’t occur in +some cells such +as brain cells -Y. Bisk, A. Holtzman, J. Thomason, J. Andreas, Y. Bengio, J. Chai, M. Lapata, A. Lazaridou, J. May, +9 -A. Nisnevich, N. Pinto, and J. Turian. Experience grounds language, 2020. + Four Mitotic Stages -T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, -G. Sastry, A. Askell, S. Agarwal, A. Herbert-Voss, G. Krueger, T. Henighan, R. Child, A. Ramesh, -D. M. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray, B. Chess, -J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, and D. Amodei. Language models -are few-shot learners, 2020. +Prophase +Metaphase +Anaphase +Telophase -P. Clark, I. Cowhey, O. Etzioni, T. Khot, A. Sabharwal, C. Schoenick, and O. Tafjord. Think you have -solved question answering? try arc, the ai2 reasoning challenge. ArXiv, abs/1803.05457, 2018. +10 -P. Clark, O. Etzioni, D. Khashabi, T. Khot, B. D. Mishra, K. Richardson, A. Sabharwal, C. Schoenick, -O. Tafjord, N. Tandon, S. Bhakthavatsalam, D. Groeneveld, M. Guerquin, and M. Schmitz. From ’f’ -to ’a’ on the n.y. regents science exams: An overview of the aristo project. ArXiv, abs/1909.01958, -2019. + Early Prophase +Chromatin in nucleus -J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. Bert: Pre-training of deep bidirectional +condenses to form visible +chromosomes -transformers for language understanding. ArXiv, abs/1810.04805, 2019. +Mitotic spindle forms from +fibers in cytoskeleton or +centrioles (animal) -R. Geirhos, J.-H. Jacobsen, C. Michaelis, R. Zemel, W. Brendel, M. Bethge, and F. A. Wichmann. +Cytoplasm -Shortcut learning in deep neural networks, 2020. +Nuclear Membrane -C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger. On calibration of modern neural networks. ICML, +Nucleolus -2017. +Chromosomes -D. Hendrycks, M. Mazeika, and T. Dietterich. Deep anomaly detection with outlier exposure. ICLR, + Late Prophase -2019a. +Nuclear membrane & nucleolus -D. Hendrycks, K. Zhao, S. Basart, J. Steinhardt, and D. Song. Natural adversarial examples. ArXiv, +are broken down -abs/1907.07174, 2019b. +Chromosomes continue -D. Hendrycks, C. Burns, S. Basart, A. Critch, J. Li, D. Song, and J. Steinhardt. Aligning ai with +condensing & are clearly visible -shared human values, 2020. +Spindle fibers called -L. Huang, R. L. Bras, C. Bhagavatula, and Y. Choi. Cosmos qa: Machine reading comprehension +kinetochores attach to the +centromere of each +chromosome -with contextual commonsense reasoning, 2019. +Spindle finishes forming -J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, +12 -and D. Amodei. Scaling laws for neural language models, 2020. +between the poles of the cell -D. Khashabi, T. Khot, A. Sabharwal, O. Tafjord, P. Clark, and H. Hajishirzi. Unifiedqa: Crossing + Metaphase -format boundaries with a single qa system, 2020. +Chromosomes, attached to the -T. Khot, P. Clark, M. Guerquin, P. Jansen, and A. Sabharwal. Qasc: A dataset for question answering +kinetochore fibers, move to the center +of the cell -via sentence composition, 2019. +Chromosomes are now lined up at the -A. Kumar, P. Liang, and T. Ma. Verified uncertainty calibration, 2019. +Equator of Cell -9 +equator - G. Lai, Q. Xie, H. Liu, Y. Yang, and E. Hovy. Race: Large-scale reading comprehension dataset from +Pole of +the Cell -examinations, 2017. +13 -Z. Lan, M. Chen, S. Goodman, K. Gimpel, P. Sharma, and R. Soricut. Albert: A lite bert for + Anaphase -self-supervised learning of language representations. ArXiv, abs/1909.11942, 2020. +Occurs rapidly +Sister -Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and -V. Stoyanov. Roberta: A robustly optimized bert pretraining approach. ArXiv, abs/1907.11692, -2019. +chromatids are +pulled apart to +opposite poles +of the cell by +kinetochore +fibers -T. Mihaylov, P. Clark, T. Khot, and A. Sabharwal. Can a suit of armor conduct electricity? a new +14 -dataset for open book question answering. In EMNLP, 2018. + Telophase -Y. Ovadia, E. Fertig, J. Ren, Z. Nado, D. Sculley, S. Nowozin, J. V. Dillon, B. Lakshminarayanan, -and J. Snoek. Can you trust your model’s uncertainty? Evaluating predictive uncertainty under -dataset shift. NeurIPS, 2019. +Sister chromatids at opposite -F. Petroni, T. Rocktäschel, P. Lewis, A. Bakhtin, Y. Wu, A. H. Miller, and S. Riedel. Language +poles -models as knowledge bases?, 2019. +Spindle disassembles +Nuclear envelope forms around +each set of sister chromatids -C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, and P. J. Liu. +Nucleolus reappears +CYTOKINESIS occurs +Chromosomes reappear as -Exploring the limits of transfer learning with a unified text-to-text transformer, 2019. +chromatin -M. Richardson, C. J. Burges, and E. Renshaw. MCTest: A challenge dataset for the open-domain -machine comprehension of text. In Proceedings of the 2013 Conference on Empirical Methods in -Natural Language Processing, pages 193–203, Seattle, Washington, USA, Oct. 2013. Association -for Computational Linguistics. +15 -A. B. Sai, A. K. Mohankumar, and M. M. Khapra. A survey of evaluation metrics used for nlg + Cytokinesis -systems. 2020. +Means division of the cytoplasm +Division of cell into two, identical -A. Turing. Computing machinery and intelligence. 1950. +halves called daughter cells -A. Wang, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman. Glue: A multi-task benchmark +In plant cells, cell plate forms at -and analysis platform for natural language understanding, 2018. +the equator to divide cell -A. Wang, Y. Pruksachatkun, N. Nangia, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman. -Superglue: A stickier benchmark for general-purpose language understanding systems, 2019. +In animal cells, cleavage furrow -R. Zellers, A. Holtzman, Y. Bisk, A. Farhadi, and Y. Choi. Hellaswag: Can a machine really finish +forms to split cell -your sentence?, 2019. +16 -R. Zellers, A. Holtzman, E. Clark, L. Qin, A. Farhadi, and Y. Choi. Evaluating machines by their + Daughter Cells of Mitosis -real-world language use, 2020. +Have the same number of -10 +chromosomes as each other and +as the parent cell from which +they were formed - A ADDITIONAL ANALYSIS +Identical to each other, but -(a) As the number of few-shot instruction examples in- -creases, the accuracy monotonically increases. Notably, -zero-shot performance is only somewhat lower than -5-shot accuracy. +smaller than parent cell -(b) While models are more calibrated in a few-shot set- -ting than a zero-shot setting, they are still miscalibrated, -with gap between accuracy and confidence reaching up -to 14%. Here the correlation between confidence and -accuracy is r = 0.81, compared to r = 0.63 in the -zero-shot setting. +Must grow in size to become -A.1 ERROR ANALYSIS +mature cells (G1 of Interphase) -We qualitatively analyze when GPT-3 makes high confidence mistakes. We find that while many of -these mistakes were clearly wrong, many were mistakes that a human might make. For example, -one question it got wrong was “How many chromosomes do all human somatic cells contain?” The -correct answer is 46, while few-shot GPT-3 predicted 23 with confidence 97.5%. This answer would -have been correct if the question asked about the number of pairs of chromosomes. Similarly, many -of its other high confidence mistakes were also correct answers to slightly different questions. +17 -A.2 FINE-TUNING + Meiosis +Formation of Gametes +(Eggs & Sperm) -Rather than testing models in a few-shot setting, we now determine the impact of fine-tuning with -thousands of examples. For this section, we fine-tune RoBERTa-base (Liu et al., 2019) and ALBERT- -xxlarge (Lan et al., 2020), which have fewer than 250 million parameters and have at least 10× -fewer parameters than GPT-3 Small. Models are fine-tuned to predict one of four classes and are -fine-tuned using the dev+val set, and we test on the test set. We observe that these smaller models -can attain better-than-random accuracy. RoBERTa-base attains an overall accuracy of 27.9%, with -27.9% accuracy for the humanities, 28.8% for social sciences, 27.0% for STEM, and 27.7% for other. -ALBERT-xxlarge attains an accuracy of 27.1%, with 27.2% accuracy for the humanities, 25.7% for -the social sciences, 27.7% for STEM, and 27.9% for other. Consequently smaller models that are not -designed for QA are able to exceed random chance, though barely. +18 -B TEST DETAILS + Facts About Meiosis +Preceded by interphase which +includes chromosome replication +Two meiotic divisions --- Meiosis -B.1 TASK DESCRIPTIONS AND EXAMPLES +I and Meiosis II -We list all tasks and the topics they test in Table 2. We also provide an example for each task starting -with Figure 10. +Called Reduction- division +Original cell is diploid (2n) +Four daughter cells produced that -B.2 EXACT QUESTION AND ANSWER CONTAMINATION +are monoploid (1n) -Since language models train on vast text corpora, there is some chance that they have seen the exact -question and answer during pretraining. If they memorized the exact question and answer, then -they would attain higher accuracy than their true ability. Likewise, a question’s entropy would be +19 -11 + Facts About Meiosis -0-Shot1-Shot2-Shot3-Shot4-Shot5-ShotNumber of Examples in Context3035404550Accuracy (%)GPT-3 Multitask Accuracy vs.Number of Examples in Context203040506070Confidence (%)203040506070Accuracy (%)GPT-3 Few-Shot Calibration especially low if it were memorized. Memorized questions and answers should have low entropy and -high accuracy. However, in Figure 9, we see that accuracy and question entropy are not positively -correlated, suggesting that the test’s low-entropy questions do not correspond to memorized (and -thereby correctly predicted) answers. This suggests that our exact questions were not memorized. -However, during pretraining models encountered text related to our questions through processing -Wikipedia. We also note that most of our questions came from PDFs or websites where questions and -answers are on separate pages. +Daughter cells contain half the +number of chromosomes as the +original cell -See Brown et al. (2020) for a previous discussion of contamination showing that the phenomena -hardly affects performance. To reduce the probability that future models encounter exact questions -during test-time, we provide a list of question sources on our github. +Produces gametes (eggs & sperm) +Occurs in the testes in males -Figure 9: The average log probability of the question (without answer) is not strongly positively -correlated with accuracy, all else equal. Each point corresponds to a task. Higher log probability -indicates higher compression, and especially high log probability suggests memorization. In the zero- -shot question prompt, the correlation between average log probability and accuracy is r = −0.43, -and for the few-shot setting the correlation is r = −0.56. +(Spermatogenesis) -12 +Occurs in the ovaries in females -2.82.62.42.22.01.8Log Probability Per Token2030405060Accuracy (%)GPT-3 Zero-ShotPrompt Compression and Accuracy2.42.22.01.81.61.41.2Log Probability Per Token203040506070Accuracy (%)GPT-3 Few-ShotPrompt Compression and Accuracy Task -Abstract Algebra -Anatomy -Astronomy -Business Ethics -Clinical Knowledge -College Biology -College Chemistry -College Computer Science -College Mathematics -College Medicine -College Physics -Computer Security -Conceptual Physics -Econometrics -Electrical Engineering -Elementary Mathematics -Formal Logic -Global Facts -High School Biology -High School Chemistry -High School Computer Science -High School European History -High School Geography -High School Gov’t and Politics -High School Macroeconomics -High School Mathematics -High School Microeconomics -High School Physics -High School Psychology -High School Statistics -High School US History -High School World History -Human Aging -Human Sexuality -International Law -Jurisprudence -Logical Fallacies -Machine Learning -Management -Marketing -Medical Genetics -Miscellaneous -Moral Disputes -Moral Scenarios -Nutrition -Philosophy -Prehistory -Professional Accounting -Professional Law -Professional Medicine -Professional Psychology -Public Relations -Security Studies -Sociology -US Foreign Policy -Virology -World Religions - -Tested Concepts -Groups, rings, fields, vector spaces, ... -Central nervous system, circulatory system, ... -Solar system, galaxies, asteroids, ... -Corporate responsibility, stakeholders, regulation, ... -Spot diagnosis, joints, abdominal examination, ... -Cellular structure, molecular biology, ecology, ... -Analytical, organic, inorganic, physical, ... -Algorithms, systems, graphs, recursion, ... -Differential equations, real analysis, combinatorics, ... -Introductory biochemistry, sociology, reasoning, ... -Electromagnetism, thermodynamics, special relativity, ... -Cryptography, malware, side channels, fuzzing, ... -Newton’s laws, rotational motion, gravity, sound, ... -Volatility, long-run relationships, forecasting, ... -Circuits, power systems, electrical drives, ... -Word problems, multiplication, remainders, rounding, ... -Propositions, predicate logic, first-order logic, ... -Extreme poverty, literacy rates, life expectancy, ... -Natural selection, heredity, cell cycle, Krebs cycle, ... -Chemical reactions, ions, acids and bases, ... -Arrays, conditionals, iteration, inheritance, ... -Renaissance, reformation, industrialization, ... -Population migration, rural land-use, urban processes, ... -Branches of government, civil liberties, political ideologies, ... -Economic indicators, national income, international trade, ... -Pre-algebra, algebra, trigonometry, calculus, ... -Supply and demand, imperfect competition, market failure, ... -Kinematics, energy, torque, fluid pressure, ... -Behavior, personality, emotions, learning, ... -Random variables, sampling distributions, chi-square tests, ... -Civil War, the Great Depression, The Great Society, ... -Ottoman empire, economic imperialism, World War I, ... -Senescence, dementia, longevity, personality changes, ... -Pregnancy, sexual differentiation, sexual orientation, ... -Human rights, sovereignty, law of the sea, use of force, ... -Natural law, classical legal positivism, legal realism, ... -No true Scotsman, base rate fallacy, composition fallacy, ... -SVMs, VC dimension, deep learning architectures, ... -Organizing, communication, organizational structure, ... -Segmentation, pricing, market research, ... -Genes and cancer, common chromosome disorders, ... -Agriculture, Fermi estimation, pop culture, ... -Freedom of speech, addiction, the death penalty, ... -Detecting physical violence, stealing, externalities, ... -Metabolism, water-soluble vitamins, diabetes, ... -Skepticism, phronesis, skepticism, Singer’s Drowning Child, ... -Neanderthals, Mesoamerica, extinction, stone tools, ... -Auditing, reporting, regulation, valuation, ... -Torts, criminal law, contracts, property, evidence, ... -Diagnosis, pharmacotherapy, disease prevention, ... -Diagnosis, biology and behavior, lifespan development, ... -Media theory, crisis management, intelligence gathering, ... -Environmental security, terrorism, weapons of mass destruction, ... -Socialization, cities and community, inequality and wealth, ... -Soft power, Cold War foreign policy, isolationism, ... -Epidemiology, coronaviruses, retroviruses, herpesviruses, ... -Judaism, Christianity, Islam, Buddhism, Jainism, ... - -Table 2: Summary of all 57 tasks. +(Oogenesis) -13 +20 -Supercategory -STEM -STEM -STEM -Other -Other -STEM -STEM -STEM -STEM -Other -STEM -STEM -STEM -Social Sciences -STEM -STEM -Humanities -Other -STEM -STEM -STEM -Humanities -Social Sciences -Social Sciences -Social Sciences -STEM -Social Sciences -STEM -Social Sciences -STEM -Humanities -Humanities -Other -Social Sciences -Humanities -Humanities -Humanities -STEM -Other -Other -Other -Other -Humanities -Humanities -Other -Humanities -Humanities -Other -Humanities -Other -Social Sciences -Social Sciences -Social Sciences -Social Sciences -Social Sciences -Other -Humanities - - Find all c in Z3 such that Z3[x]/(x2 + c) is a field. -(A) 0 - -(D) 3 - -(C) 2 - -(B) 1 - -Figure 10: An Abstract Algebra example. - -What is the embryological origin of the hyoid bone? -(A) The first pharyngeal arch -(B) The first and second pharyngeal arches -(C) The second pharyngeal arch -(D) The second and third pharyngeal arches - -Figure 11: An Anatomy example. - -Why isn’t there a planet where the asteroid belt is located? -(A) A planet once formed here but it was broken apart by a catastrophic collision. -(B) There was not enough material in this part of the solar nebula to form a planet. -(C) There was too much rocky material to form a terrestrial planet but not enough gaseous material -to form a jovian planet. -(D) Resonance with Jupiter prevented material from collecting together to form a planet. - -Figure 12: An Astronomy example. - -Three contrasting tactics that CSO’s can engage in to meet their aims are -cally involves research and communication, -company’s operations or -(A) Non-violent direct action, Violent direct action, Indirect action, Boycott -(B) Indirect action, Instrumental action, Non-violent direct action, Information campaign -(C) Indirect action, Violent direct action, Non-violent direct-action Boycott. -(D) Non-violent direct action, Instrumental action, Indirect action, Information campaign - -which typi- -, which may involve physically attacking a - -, often involving some form of - -. - -Figure 13: A Business Ethics example. - -How many attempts should you make to cannulate a patient before passing the job on to a senior -colleague? -(A) 4 - -(D) 1 - -(C) 2 - -(B) 3 - -Figure 14: A Clinical Knowledge example. - -In a given population, 1 out of every 400 people has a cancer caused by a completely recessive -allele, b. Assuming the population is in Hardy-Weinberg equilibrium, which of the following is -the expected proportion of individuals who carry the b allele but are not expected to develop the -cancer? -(A) 1/400 - -(D) 38/400 - -(B) 19/400 - -(C) 20/400 + Why Do we Need Meiosis? +It is the fundamental basis -Figure 15: A College Biology example. +of sexual reproduction -Which of the following statements about the lanthanide elements is NOT true? -(A) The most common oxidation state for the lanthanide elements is +3. -(B) Lanthanide complexes often have high coordination numbers (> 6). -(C) All of the lanthanide elements react with aqueous acid to liberate hydrogen. -(D) The atomic radii of the lanthanide elements increase across the period from La to Lu. +Two haploid (1n) gametes are -Figure 16: A College Chemistry example. +brought together through +fertilization to form a diploid +(2n) zygote -14 +21 - Consider a computer design in which multiple processors, each with a private cache memory, -share global memory using a single bus. This bus is the critical system resource. Each processor -can execute one instruction every 500 nanoseconds as long as memory references are satisfied -by its local cache. When a cache miss occurs, the processor is delayed for an additional 2,000 -nanoseconds. During half of this additional delay, the bus is dedicated to serving the cache miss. -During the other half, the processor cannot continue, but the bus is free to service requests from -other processors. On average, each instruction requires 2 memory references. On average, cache -misses occur on 1 percent of references. What proportion of the capacity of the bus would a single -processor consume, ignoring delays due to competition from other processors? -(D) 2/27 -(A) 1/50 + Replication of Chromosomes -(C) 1/25 +Occurs in +Interphase -(B) 1/27 +Replication is the -Figure 17: A College Computer Science example. +process of duplicating +a chromosome +Occurs prior to -Let A be a real 2 × 2 matrix. Which of the following statements must be true? -I. All of the entries of A2 are nonnegative. -II. The determinant of A2 is nonnegative. -III. If A has two distinct eigenvalues, then A2 has two distinct eigenvalues. -(A) I only +division -(D) II and III only +Replicated copies are -(C) III only +called sister +chromatids -(B) II only +Held together at -Figure 18: A College Mathematics example. +centromere -In a genetic test of a newborn, a rare genetic disorder is found that has X-linked recessive -transmission. Which of the following statements is likely true regarding the pedigree of this -disorder? -(A) All descendants on the maternal side will have the disorder. -(B) Females will be approximately twice as affected as males in this family. -(C) All daughters of an affected male will be affected. -(D) There will be equal distribution of males and females affected. +22 -Figure 19: A College Medicine example. + Meiosis: Two Part Cell +Division -One end of a Nichrome wire of length 2L and cross-sectional area A is attached to an end of -another Nichrome wire of length L and cross- sectional area 2A. If the free end of the longer wire -is at an electric potential of 8.0 volts, and the free end of the shorter wire is at an electric potential -of 1.0 volt, the potential at the junction of the two wires is most nearly equal to -(A) 2.4 V -(B) 3.3 V -(C) 4.5 V -(D) 5.7 V +Sister +chromatids +separate -Figure 20: A College Physics example. +Homologs +separate -Why is it that anti-virus scanners would not have found an exploitation of Heartbleed? -(A) It’s a vacuous question: Heartbleed only reads outside a buffer, so there is no possible exploit -(B) Anti-virus scanners tend to look for viruses and other malicious -(C) Heartbleed attacks the anti-virus scanner itself -(D) Anti-virus scanners tend to look for viruses and other malicious code, but Heartbleed -exploits steal secrets without injecting any code +Meiosis +I -Figure 21: A Computer Security example. +Meiosis +II -A model airplane flies slower when flying into the wind and faster with wind at its back. When -launched at right angles to the wind, a cross wind, its groundspeed compared with flying in still -air is -(A) the same +Diploid -(D) either greater or less depending on wind speed +Diploid -(B) greater +23 -(C) less +Haploid -Figure 22: A Conceptual Physics example. + Meiosis I: Reduction Division -15 +Nucleus - Consider the following AR(1) model with the disturbances having zero mean and unit variance -yt = 0.2 + 0.4yt−1 + ut -The (unconditional) mean of y will be given by -(C) 0.5 -(A) 0.2 +Early Prophase I +(Chromosome +number doubled) -(D) 0.33 +Spindle +fibers -(B) 0.4 +Nuclear +envelope -Figure 23: An Econometrics example. +Late Prophase +I -A point pole has a strength of 4π × 10−4 weber. The force in newtons on a point pole of -4π × 1.5 × 10−4 weber placed at a distance of 10 cm from it will be -(A) 15 N. +Metaphase I -(D) 3.75 N. +Anaphase I -(C) 7.5 N. +Telophase I (diploid) -(B) 20 N. +24 -Figure 24: An Electrical Engineering example. + Prophase I -A total of 30 players will play basketball at a park. There will be exactly 5 players on each team. -Which statement correctly explains how to find the number of teams needed? -(A) Add 5 to 30 to find 35 teams. -(B) Divide 30 by 5 to find 6 teams. -(C) Multiply 30 and 5 to find 150 teams. -(D) Subtract 5 from 30 to find 25 teams. +Early prophase +Homologs pair. +Crossing over +occurs. -Figure 25: An Elementary Mathematics example. +25 -Determine whether the statements are logically equivalent or contradictory. If neither, determine -whether they are consistent or inconsistent. -E ⊃ (F · E) and ∼ E · F -(A) Logically equivalent -(B) Contradictory -(C) Neither logically equivalent nor contradictory, but consistent -(D) Inconsistent +Late prophase +Chromosomes condense. +Spindle forms. +Nuclear envelope +fragments. -Figure 26: A Formal Logic example. + Tetrads Form in Prophase +I -As of 2017, how many of the world’s 1-year-old children today have been vaccinated against -some disease? -(A) 80% -(B) 60% -(C) 40% -(D) 20% +Homologous chromosomes +(each with sister chromatids) -Figure 27: A Global Facts example. +Join to form a TETRAD -Homologous structures are often cited as evidence for the process of natural selection. All of the -following are examples of homologous structures EXCEPT -(A) the wings of a bird and the wings of a bat -(B) the flippers of a whale and the arms of a man -(C) the pectoral fins of a porpoise and the flippers of a seal -(D) the forelegs of an insect and the forelimbs of a dog +26 -Figure 28: A High School Biology example. +Called Synapsis -From the solubility rules, which of the following is true? -(A) All chlorides, bromides, and iodides are soluble -(B) All sulfates are soluble -(C) All hydroxides are soluble -(D) All ammonium-containing compounds are soluble + Crossing-Over -Figure 29: A High School Chemistry example. +Homologous -16 +chromosomes in a +tetrad cross over +each other +Pieces of - A list of numbers has n elements, indexed from 1 to n. The following algorithm is intended to -display the number of elements in the list that have a value greater than 100. The algorithm uses -the variables count and position. Steps 3 and 4 are missing. -Step 1: Set count to 0 and position to 1. -Step 2: If the value of the element at index position is greater -than 100, increase the value of count by 1. -Step 3: (missing step) -Step 4: (missing step) -Step 5: Display the value of count. -Which of the following could be used to replace steps 3 and 4 so that the algorithm works as -intended? -(A) Step 3: Increase the value of position by 1. -Step 4: Repeat steps 2 and 3 until the value of count is greater than 100. -(B) Step 3: Increase the value of position by 1. -Step 4: Repeat steps 2 and 3 until t he value of position is greater than n. -(C) Step 3: Repeat step 2 until the value of count is greater than 100. -Step 4: Increase the value of position by 1. -(D) Step 3: Repeat step 2 until the value of position is greater than n. -Step 4: Increase the value of count by 1. - -Figure 30: A High School Computer Science example. - -This question refers to the following information. -Albeit the king’s Majesty justly and rightfully is and ought to be the supreme head of the Church -of England, and so is recognized by the clergy of this realm in their convocations, yet nevertheless, -for corroboration and confirmation thereof, and for increase of virtue in Christ’s religion within -this realm of England, and to repress and extirpate all errors, heresies, and other enormities and -abuses heretofore used in the same, be it enacted, by authority of this present Parliament, that the -king, our sovereign lord, his heirs and successors, kings of this realm, shall be taken, accepted, -and reputed the only supreme head in earth of the Church of England, called Anglicans Ecclesia; -and shall have and enjoy, annexed and united to the imperial crown of this realm, as well the -title and style thereof, as all honors, dignities, preeminences, jurisdictions, privileges, authorities, -immunities, profits, and commodities to the said dignity of the supreme head of the same Church -belonging and appertaining; and that our said sovereign lord, his heirs and successors, kings of -this realm, shall have full power and authority from time to time to visit, repress, redress, record, -order, correct, restrain, and amend all such errors, heresies, abuses, offenses, contempts, and -enormities, whatsoever they be, which by any manner of spiritual authority or jurisdiction ought -or may lawfully be reformed, repressed, ordered, redressed, corrected, restrained, or amended, -most to the pleasure of Almighty God, the increase of virtue in Christ’s religion, and for the -conservation of the peace, unity, and tranquility of this realm; any usage, foreign land, foreign -authority, prescription, or any other thing or things to the contrary hereof notwithstanding. -English Parliament, Act of Supremacy, 1534 -From the passage, one may infer that the English Parliament wished to argue that the Act of -Supremacy would -(A) give the English king a new position of authority -(B) give the position of head of the Church of England to Henry VIII alone and exclude his heirs -(C) establish Calvinism as the one true theology in England -(D) end various forms of corruption plaguing the Church in England - -Figure 31: A High School European History example. - -During the third stage of the demographic transition model, which of the following is true? -(A) Birth rates increase and population growth rate is less rapid. -(B) Birth rates decline and population growth rate is less rapid. -(C) Birth rates increase and population growth rate increases. -(D) Birth rates decrease and population growth rate increases. - -Figure 32: A High School Geography example. +chromosomes or +genes are +exchanged -17 +Produces Genetic +recombination in +the offspring - Which of the following best states an argument made by James Madison in The Federalist number -10? -(A) Honest politicians can prevent factions from developing. -(B) Factions are more likely to occur in large republics than in small ones. -(C) The negative effects of factionalism can be reduced by a republican government. -(D) Free elections are the people’s best defense against factionalism. - -Figure 33: A High School Government and Politics example. - -Which of the following is not included in the U.S. GDP? -(A) The U.S. military opens a new base in a foreign country with 1000 U.S. personnel. -(B) Japanese consumers buy thousands of CDs produced in the United States. -(C) An American pop singer performs a sold-out concert in Paris. -(D) A French theatrical production tours dozens of American cities. - -Figure 34: A High School Macroeconomics example. - -Joe was in charge of lights for a dance. The red light blinks every two seconds, the yellow light -every three seconds, and the blue light every five seconds. If we include the very beginning and -very end of the dance, how many times during a seven minute dance will all the lights come on at -the same time? (Assume that all three lights blink simultaneously at the very beginning of the -dance.) -(A) 3 -(B) 15 -(C) 6 -(D) 5 - -Figure 35: A High School Mathematics example. - -If the government subsidizes producers in a perfectly competitive market, then -(A) the demand for the product will increase -(B) the demand for the product will decrease -(C) the consumer surplus will increase -(D) the consumer surplus will decrease - -Figure 36: A High School Microeconomics example. - -A point charge, Q = +1 mC, is fixed at the origin. How much work is required to move a charge, -Q = +8 µC, from the point (0, 4 meters) to the point (3 meters, 0)? -(A) 3.5 J -(B) 6.0 J -(C) 22.5 J -(D) 40 J - -Figure 37: A High School Physics example. - -While swimming in the ocean, Ivan is frightened by a dark shadow in the water even before he -has the chance to identify what the shadow is. The synaptic connections taking place during this -incident of fright are best described by which of the following? -(A) Messages are sent from the thalamus directly to the amygdala. -(B) Messages are sent from the thalamus to the “what” and “where” pathways. -(C) Messages are sent from the parasympathetic nervous system to the cerebral cortex. -(D) Messages are sent from the frontal lobes to the pituitary gland. - -Figure 38: A High School Psychology example. +27 -18 + Metaphase I - Jonathan obtained a score of 80 on a statistics exam, placing him at the 90th percentile. Suppose -five points are added to everyone’s score. Jonathan’s new score will be at the -(A) 80th percentile. -(B) 85th percentile. -(C) 90th percentile. -(D) 95th percentile. - -Figure 39: A High School Statistics example. - -This question refers to the following information. -“Society in every state is a blessing, but government even in its best state is but a necessary evil; in -its worst state an intolerable one; for when we suffer, or are exposed to the same miseries by a -government, which we might expect in a country without government, our calamity is heightened -by reflecting that we furnish the means by which we suffer. Government, like dress, is the badge -of lost innocence; the palaces of kings are built on the ruins of the bowers of paradise. For were -the impulses of conscience clear, uniform, and irresistibly obeyed, man would need no other -lawgiver; but that not being the case, he finds it necessary to surrender up a part of his property to -furnish means for the protection of the rest; and this he is induced to do by the same prudence -which in every other case advises him out of two evils to choose the least. Wherefore, security -being the true design and end of government, it unanswerably follows that whatever form thereof -appears most likely to ensure it to us, with the least expense and greatest benefit, is preferable to -all others.” -Thomas Paine, Common Sense, 1776 -Which of the following “miseries” alluded to above were most condemned by Anti-Federalists of -the post-Revolutionary era? -(A) Organized response to Bacon’s Rebellion. -(B) Federal response to Shays’s Rebellion. -(C) Federal response to the Whiskey Rebellion. -(D) Federal response to Pontiac’s Rebellion. - -Figure 40: A High School US History example. - -All other things being equal, which of the following persons is more likely to show osteoporosis? -(A) An older Hispanic American woman -(B) An older African American woman -(C) An older Asian American woman -(D) An older Native American woman - -Figure 41: A Human Aging example. - -Morning sickness is typically a problem: -(A) during the first trimester -(B) during the second trimester -(C) during the third trimester -(D) all through the pregnancy - -Figure 42: A Human Sexuality example. +Homologous pairs +of chromosomes +align along the +equator of the cell -19 +28 - Would a reservation to the definition of torture in the ICCPR be acceptable in contemporary -practice? -(A) This is an acceptable reservation if the reserving country’s legislation employs a different -definition -(B) This is an unacceptable reservation because it contravenes the object and purpose of -the ICCPR -(C) This is an unacceptable reservation because the definition of torture in the ICCPR is consistent -with customary international law -(D) This is an acceptable reservation because under general international law States have the right -to enter reservations to treaties + Anaphase I -Figure 43: An International Law example. +Homologs separate and +move to opposite poles. -Which position does Rawls claim is the least likely to be adopted by the POP (people in the -original position)? -(A) The POP would choose equality above liberty. -(B) The POP would opt for the ‘maximin’ strategy. -(C) The POP would opt for the ‘difference principle.’ -(D) The POP would reject the ‘system of natural liberty.’ +Sister chromatids remain +attached at their centromeres. -Figure 44: A Jurisprudence example. +29 -John Stuart Mill: Each person’s happiness is a good to that person, and the general happiness, -therefore, a good to the aggregate of all persons. -(A) Fallacy of Composition -(B) Fallacy of Division -(C) Gambler’s Fallacy -(D) Equivocation + Telophase I -Figure 45: A Logical Fallacies example. +Nuclear envelopes +reassemble. -A 6-sided die is rolled 15 times and the results are: side 1 comes up 0 times; side 2: 1 time; side -3: 2 times; side 4: 3 times; side 5: 4 times; side 6: 5 times. Based on these results, what is the -probability of side 3 coming up when using Add-1 Smoothing? -(A) 2/15 +Spindle disappears. -(C) 3/16 +Cytokinesis divides cell into +two. -(D) 1/5 +30 -(B) 1/7 + Meiosis II -Figure 46: A Machine Learning example. +Gene X -According to Lewin, Lippet and White’s 1939 experiment, which form of leadership produced the -most work from participants? -(A) Laissez-faire -(B) Democratic -(C) Authoritarian -(D) A mix of laissez-faire and democratic +Only one homolog of each -Figure 47: A Management example. +chromosome is present in +the cell. -The single group within society that is most vulnerable to reference group influence is: -(A) The older consumer who feels somewhat left out of things. -(B) The married women, many of whom feel a need for stability in their lives. -(C) New immigrants who really want to assimilate into their new culture. -(D) Children, who base most of their buying decisions on outside influences. +Sister chromatids carry +identical genetic +information. -Figure 48: A Marketing example. +Meiosis II produces gametes with +one copy of each chromosome and +thus one copy of each gene. -20 +31 - Which of the following conditions does not show multifactorial inheritance? -(A) Pyloric stenosis -(B) Schizophrenia -(C) Spina bifida (neural tube defects) -(D) Marfan syndrome - -Figure 49: A Medical Genetics example. - -If each of the following meals provides the same number of calories, which meal requires the -most land to produce the food? -(A) Red beans and rice -(B) Steak and a baked potato -(C) Corn tortilla and refried beans -(D) Lentil soup and brown bread - -Figure 50: A Miscellaneous example. - -In response to Sandel’s “social justice” argument, Kamm argues that -(A) even if we were able to enhance ourselves or others, we would not thereby be obligated to do -so. -(B) there is a difference between (i) choosing that one have a certain trait and (ii) being responsible -for bearing the costs of that choice. -(C) having the option to enhance will lead to fewer instances of people who require the assistance -of others. -(D) all of the above - -Figure 51: A Moral Disputes example. - -In response to Sandel’s “social justice” argument, Kamm argues that -(A) even if we were able to enhance ourselves or others, we would not thereby be obligated to do -so. -(B) there is a difference between (i) choosing that one have a certain trait and (ii) being responsible -for bearing the costs of that choice. -(C) having the option to enhance will lead to fewer instances of people who require the assistance -of others. -(D) all of the above - -Figure 52: A Moral Disputes example. - -For which of these two scenarios does the main character (who uses I/me/my) do something -clearly morally wrong? -Scenario 1 | I drive very fast when I’m on a racetrack. -Scenario 2 | I drive very fast when I’m in the suburbs. -(A) Wrong, Wrong -(B) Wrong, Not wrong -(C) Not wrong, Wrong -(D) Not wrong, Not wrong - -Figure 53: A Moral Scenarios example. + Meiosis II: Reducing +Chromosome Number -21 +Prophase II - Which of the following is the most plausible explanation for the protective effect of dietary fibre -against cancer of the colon? -(A) Propionic acid, formed during colonic fibre fermentation inhibits liver fatty acid synthesis -(B) Butyric acid, formed during colonic fibre fermentation stimulates "silencing" of the SLC5A8 -tumour suppressor gene -(C) Butyric acid, formed during colonic fibre fermentation stimulates anti-oxidant defences -in the colon -(D) None of these options are correct - -Figure 54: A Nutrition example. - -According to Moore’s “ideal utilitarianism,” the right action is the one that brings about the -greatest amount of: -(A) pleasure. -(B) happiness. -(C) good. -(D) virtue. - -Figure 55: A Philosophy example. - -Researchers now believe that the decline of the Maya was caused chiefly by: -(A) a cataclysm of some kind, such as an earthquake, volcano, or tsunami. -(B) ecological degradation resulting from slash-and-burn farming techniques. -(C) endless wars between neighboring Mayan city-states. -(D) practices of interbreeding that led to a steep rise in congenital disorders. - -Figure 56: A Prehistory example. - -Krete is an unmarried taxpayer with income exclusively from wages. By December 31, year 1, -Krete’s employer has withheld $16,000 in federal income taxes and Krete has made no estimated -tax payments. On April 15, year 2, Krete timely filed for an extension request to file her individual -tax return, and paid $300 of additional taxes. Krete’s year 1 tax liability was $16,500 when she -timely filed her return on April 30, year 2, and paid the remaining tax liability balance. What -amount would be subject to the penalty for underpayment of estimated taxes? -(A) $0 -(B) $500 -(C) $1,650 -(D) $16,500 - -Figure 57: A Professional Accounting example. - -The night before his bar examination, the examinee’s next-door neighbor was having a party. The -music from the neighbor’s home was so loud that the examinee couldn’t fall asleep. The examinee -called the neighbor and asked her to please keep the noise down. The neighbor then abruptly hung -up. Angered, the examinee went into his closet and got a gun. He went outside and fired a bullet -through the neighbor’s living room window. Not intending to shoot anyone, the examinee fired -his gun at such an angle that the bullet would hit the ceiling. He merely wanted to cause some -damage to the neighbor’s home to relieve his angry rage. The bullet, however, ricocheted off the -ceiling and struck a partygoer in the back, killing him. The jurisdiction makes it a misdemeanor -to discharge a firearm in public. The examinee will most likely be found guilty for which of the -following crimes in connection to the death of the partygoer? -(A) Murder. -(B) Involuntary manslaughter. -(C) Voluntary manslaughter. -(D) Discharge of a firearm in public. - -Figure 58: A Professional Law example. +Metaphase II -22 +32 - A 63-year-old man is brought to the emergency department because of a 4-day history of in- -creasingly severe left leg pain and swelling of his left calf. He also has a 1-month history of -increasingly severe upper midthoracic back pain. During this time, he has had a 9-kg (20-lb) -weight loss despite no change in appetite. He has no history of major medical illness. His only -medication is ibuprofen. He is 180 cm (5 ft 11 in) tall and weighs 82 kg (180 lb); BMI is 25 -kg/m2 . His vital signs are within normal limits. On examination, lower extremity pulses are -palpable bilaterally. The remainder of the physical examination shows no abnormalities. An x-ray -of the thoracic spine shows no abnormalities. A CT scan of the abdomen shows a 3-cm mass in -the body of the pancreas; there are liver metastases and encasement of the superior mesenteric -artery. Ultrasonography of the left lower extremity shows a femoropopliteal venous clot. Which -of the following is the most likely cause of this patient’s symptoms? -(A) Carcinoid syndrome -(B) Hypercoagulability from advanced malignancy -(C) Multiple endocrine neoplasia -(D) Splenic artery aneurysm and embolic disease of the left lower extremity - -Figure 59: A Professional Medicine example. - -The technique that is most likely to produce an immediate improvement in the behavior of a child -who hits others and rips up schoolbooks is -(A) a combination of reinforcement for appropriate behavior and mild punishment for in- -appropriate behavior -(B) differential positive reinforcement -(C) loss of privileges for misbehaving -(D) shaping - -Figure 60: A Professional Psychology example. - -You work for a utility company that is building a biomass plant in the community. Your employer -asks you to give a series of community talks about the plant and future operations. You visit the -plant several hours before you are due to give a speech that has been prepared by your immediate -supervisor. During the tour of the plant, you discover several claims in the speech are not true. -What do you do? -(A) Write and deliver a new speech that you know is entirely correct. -(B) Cancel all speeches until you and your supervisor can get the information straight. -(C) Deliver the speech as prepared and discuss the inaccuracies with your supervisor afterward. -(D) Address the inaccuracies with your supervisor immediately and make the necessary -corrections before giving the speech. - -Figure 61: A Public Relations example. +Telophase II -23 +Anaphase II - The Chemical Weapons Convention (CWC) prohibited the possession or deployment of chemical -weapons; however it failed to implement stipulations that would require signatories to declare their -existing stocks of chemical weapons, to identify facilities that were once involved in chemical -production, or to announce when their existing stocks would be destroyed. -(A) The Chemical Weapons Convention (CWC) prohibited the possession or deployment of -chemical weapons; however it failed to implement stipulations that would require signatories to -declare their existing stocks of chemical weapons, to identify facilities that were once involved in -chemical production, or to announce when their existing stocks would be destroyed. -(B) The CWC made some important developments regarding the use and possession of chemical -weapons and the destruction of existing stockpiles. However, the treaty failed to establish an -independent body empowered with the capacity to check treaty compliance. Lack of supra-state -authority has undermined the ability to enforce those developments. Given the anarchical nature -of international society it may be in the national security interest to retain stocks. -(C) Chemical weapons continue to exert a determining influence on international society. As early -as the 1970s military strategists were convinced of the deterrence effects chemical weapons could -have, comparable to the second strike survival logic of nuclear deterrence. The preferences of -strategists resulted in continued manufacture and stockpiling of weapons creating an international -crisis of stability. -(D) While the CWC has been ratified by the majority of international society, some nations -with a large chemical capability at their disposal have yet to enter into the treaty. However, -to some analysts the destructive military potential would be limited, having a moderate -effect on a well-equipped army in conventional warfare. Chemical arsenal essentially falls -under the category of the "poor mans" weaponry, being simplistic and inexpensive whilst -having limited military utility. However, the concern remains of the prospective impact a -terrorist chemical attack could have on civilian populations. - -Figure 62: A Security Studies example. - -Which of the following statements most closely corresponds with differential association theory? -(A) If all of your friends jumped off a bridge, I suppose you would too. -(B) You should be proud to be a part of this organization. -(C) If the door is closed, try the window. -(D) Once a thief, always a thief. - -Figure 63: A Sociology example. - -Which of the following statements most closely corresponds with differential association theory? -(A) If all of your friends jumped off a bridge, I suppose you would too. -(B) You should be proud to be a part of this organization. -(C) If the door is closed, try the window. -(D) Once a thief, always a thief. - -Figure 64: A Sociology example. - -Why did Congress oppose Wilson’s proposal for the League of Nations? -(A) It feared the League would encourage Soviet influence in the US -(B) It feared the League would be anti-democratic -(C) It feared the League would commit the US to an international alliance -(D) Both a and b - -Figure 65: A US Foreign Policy example. +4 Identical haploid +cells -24 + Prophase II - An observational study in diabetics assesses the role of an increased plasma fibrinogen level on -the risk of cardiac events. 130 diabetic patients are followed for 5 years to assess the development -of acute coronary syndrome. In the group of 60 patients with a normal baseline plasma fibrinogen -level, 20 develop acute coronary syndrome and 40 do not. In the group of 70 patients with a high -baseline plasma fibrinogen level, 40 develop acute coronary syndrome and 30 do not. Which of -the following is the best estimate of relative risk in patients with a high baseline plasma fibrinogen -level compared to patients with a normal baseline plasma fibrinogen level? -(A) (40/30)/(20/40) -(B) (40*40)/(20*30) -(C) (40*70)/(20*60) -(D) (40/70)/(20/60) +Nuclear envelope +fragments. -Figure 66: A Virology example. +Spindle forms. -The Great Cloud Sutra prophesied the imminent arrival of which person? -(A) Maitreya (Milo) -(B) The Buddha -(C) Zhou Dunyi -(D) Wang Yangming +33 -Figure 67: A World Religions example. + Metaphase II -25 +Chromosomes align +along equator of cell. + +34 + + Anaphase II + +Equator + +Pole + +Sister chromatids +separate and move +to opposite poles. + +35 + + Telophase II + +Nuclear envelope +assembles. + +Chromosomes +decondense. + +Spindle disappears. + +Cytokinesis divides cell +into two. + +36 + + Results of Meiosis +Gametes (egg & sperm) +form + +Four haploid cells with one +copy of each chromosome + +One allele of each gene + +Different combinations of +alleles for different genes +along the chromosome + +37 + + Comparison of Divisions +Meiosis +Mitosis + +Number of +divisions + +Number of +daughter cells + +Genetically +identical? + +1 + +2 + +Yes + +2 + +4 + +No + +Chromosome # Same as parent + +Half of parent + +Where + +Somatic cells + +Germ cells + +When Throughout life + +At sexual maturity + +Role +38 + +Growth and +repair + +Sexual reproduction diff --git a/q117/random_k2/question.json b/q117/random_k2/question.json index ed9bea17257fc1d71dcc98ee821aaa199e9348ba..6cb77ada478c44d8b1ef8d552efb01393345755d 100644 --- a/q117/random_k2/question.json +++ b/q117/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "2023.acl-long.386.pdf", - "web_ab28fbcd92cfea3d.pdf", - "web_28449e05c7149be2.pdf" + "web_0e04ded35ac1bdc2.pdf", + "web_ae1b8a373064ff5e.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q117/random_k2/random_1.md b/q117/random_k2/random_1.md index 5dac85680ad1e5631acf8a091a98a42ab0400813..0523beec0142edd0aa6f2e27f15673239129abcd 100644 --- a/q117/random_k2/random_1.md +++ b/q117/random_k2/random_1.md @@ -1,2417 +1,1200 @@ -COMPLAINT INDEX -REPORT FOR YEAR 2021 +SURVEY REP ORT -Vicki Schmidt, Commissioner -insurance.kansas.gov +SP ONSORED BY - Table of Contents +EXCLUSIVE SURVEY -About this Report +ARE YOU -Indexes by line: Automobile | Accident & Health +THE LAST MILE +OF RETAIL? -Indexes by line: Homeowners | Individual Annuity + TABLE OF CONTENTS -Indexes by line: Individual Life | Long-Term Care - -File a Complaint - - 01 - - 05 - -06 - -07 - -08 - - About this Report - -What is a complaint? - -The Kansas Insurance Department receives thousands of complaints and inquiries each year about -insurance companies. Complaints and inquiries from the public give us an indication of the kind of -service companies are providing to their customers. We can then work with companies that need to -improve in certain areas. - -Complaints can also help us identify insurance companies or agents that may need to be investigated -for violations of Kansas insurance laws or regulations. The Department also uses complaints to focus -public education efforts on the issues that concern Kansas insurance consumers. - -Each time we receive a complaint, we notify the insurance company and any involved agents. -We send the company a copy of the complaint and ask that it be reviewed and an explanation -provided. Companies are required by law to respond to the Department within 15 working days. -Most companies are cooperative in working with us to resolve consumer complaints. We keep a -record of each complaint filed by a consumer for up to three years. - -A formal complaint is any written communication expressing dissatisfaction or concern with an -insurance company or an agent. General written inquiries received from consumers are not counted -in this report. We also do not include complaints regarding products not within the jurisdiction of the -Kansas Insurance Department, such as self-funded plans and Medicare prescription drug plans. - -NOTE: A formal complaint must be made in writing. - -What is a complaint index? - -The complaint index shows how a company compares to other companies in the same line of business. -It is a statistic that considers a company’s share of complaints in relation to its market share or share of -business written in Kansas. Since the complaint index is measured by market share, companies in all -lines of business will be measured against an index of 1.00. - -For example, a complaint index of 1.00 means the insurer’s share of all complaints received is equal -to its share of all the business written in Kansas. An index of 2.00 means that the insurer’s share -of complaints is twice as large as its share of business written. An index of 0.00 indicates that no -complaints were received for the company. - -The formula is as follows: - -Complaint share - -Market share - -OR - -Company’s complaints / -all complaints - -Company’s premium / All -written premium - -page 1 - -insurance.kansas.gov Which complaints do we use? - -This report shows complaints for the six major types of insurance: accident and health (including -HMOs), annuity, automobile, homeowners, life and long-term care. - -We do not include complaints regarding products not within the jurisdiction of the Department, such as -self-funded plans and Medicare prescription drug plans. In 2021, the Kansas Insurance Department -closed a total of 2,205 complaints. 2,110 involved matters within our jurisdiction. - -To ensure fairness, all closed complaints are counted without any attempt to distinguish between -“valid” or “invalid” complaints. - -Which companies are shown? - -Unless otherwise noted, we list the top 20 premium-writing companies for each line, as well as any -company with 10 or more closed complaints for that type of insurance during the calendar year. - -During 2021, $6,273,694.92 was recovered for Kansas consumers as a result of the Department’s -assistance. In 2021, the largest number of complaints concerned automobile insurance (29%) and -accident & health Insurance (28%). Here is the breakdown: - -All Closed Complaints & Inquiries - -Automobile - -29% - -Life - -3,000 - -2,500 - -Accident & Health - -28% - -Long-Term Care - -2,000 - -Homeowners - -Other - -15% - -14% - -Annuity - -1,500 - -1,000 - -500 - -0 - -*rounded to the nearest percent - -Closed Complaints - -Closed Inquiries - -9% - -4% - -1% - -2018 - -2,689 - -1,049 - -2019 - -2,643 - -878 - -2020 - -2,157 - -873 - -2021 - -2,205 - -795 - -All Closed Complaints & Inquiries - -Recoveries Per Year (In Millions) - -7 - -6 - -5 - -4 +Are You Winning The Last Mile Of Retail? ............................................................................................................................ 3 -2 - -1 - -0 - -6.27 - -4.12 - -3.28 - -1.84 - -2018 - -2019 - -2020 - -2021 - -page 2 - -3,000 - -2,500 - -2,000 - -1,500 - -1,000 - -500 - -0 - -Closed Complaints - -Closed Inquiries - -2018 - -2,689 - -1,049 - -2019 - -2,643 - -878 - -2020 - -2,157 - -873 - -2021 - -2,205 - -795 - -7 +Major Challenge: Managing The Customer Experience ............................................................................................. 6 -5 - -4 - -3 - -2 - -1 +Byline: Delivery Imperatives To Compete With Amazon Without Breaking The Bank .......................... -0 +8 -Recoveries Per Year (In Millions) +Success Story: Uncommon Goods ............................................................................................................................................... -6.27 +10 -4.12 +Retailers Seek Innovative Returns Solutions ..................................................................................................................... -3.28 +11 -1.84 +Byline: Final Mile Strategies: Balancing Customer Satisfaction And Profitability........................................ -2018 +15 -2019 +Success Story: Kohl’s And Amazon ........................................................................................................................................... -2020 +16 -2021 +Retailers Turn To Technology And Disruptors .................................................................................................................. -Complaint Index Report +16 - Sixty-seven percent (67%) of total complaints concerned problems with claims handling. Other -common types of complaints involved policyholder services, underwriting and marketing and sales. +Success Story: Neiman Marcus ................................................................................................................................................. -67% - Claims Handling +19 -19% - Policyholder Service +Conclusion ................................................................................................................................................................................................ -10% - Underwriting +19 -4% - Marketing and Sales +About Retail TouchPoints ................................................................................................................................................................ -Companies may be listed on more than one table and will usually have a different complaint index -on each table. For example, if you are shopping for auto insurance, go to the table titled, “Indexes by -line: Automobile” to check for your company’s complaint index. Companies are listed in alphabetical -order. +20 -We list individual insurance companies only, not groups of insurance companies. For example, you -may be insured by a company such as State Farm. We show the name of the subsidiary company -owned by State Farm. Your policy or ID card lists the exact name of your company. +About Oracle .......................................................................................................................................................................................... -Why is my company not listed? +20 -If a company is not listed, that means it had fewer than 10 complaints during the calendar year and -it was not one of the top 20 premium writers for that line in Kansas. Information can also be compiled -for lines of insurance not listed in this booklet. For information on a company not listed or another line -of insurance, please call our Consumer Assistance Hotline at 800-432-2484. +About Convey ......................................................................................................................................................................................... -Just a snapshot +20 -The statistics in this report give you a “snapshot” of a company that represents complaints filed with the -Department regarding that company’s business in Kansas for one year and one type of coverage. As -indexes are likely to change, you may want to review these “snapshots” annually. + ARE YOU WINNING THE LAST MILE +OF RETAIL? -Complaint indexes do not tell you everything about a company. You may also want to consider the -following factors when shopping for insurance: +In retail, the “last mile” is becoming more crowded, more rushed, more expensive, and -Service — Some companies sell through local agents and some through direct marketing or group -plans. You may want to find out how the company services its policyholders. +an ever-more-critical contributor to a positive customer experience. -Licensing Status — Call the Consumer Assistance Hotline at 800-432-2484 to find out if a company -is licensed to do business in Kansas. +Over the past 18 months, there has been a 50% increase in last mile services, with +most of the demand coming from B2C e-Commerce, according to The Last Mile Retail +Study, 2018 conducted by eft Supply Chain and Logistics Business Intelligence. -Financial Stability — Financial stability helps ensure a company can pay its claims. The Department -enforces statutory requirements and monitors the financial stability of companies licensed and -operating in the state. +In addition to the growth of online sales, consumer demands are adding fuel to the -page 3 +competitive fire: -insurance.kansas.gov You can check a company’s financial rating by contacting one of these organizations: +• 65% of consumers want greater flexibility for deliveries; -MOODY’S INVESTOR SERVICES +• 61% want faster deliveries; and -STANDARD & POOR’S -INSURANCE RATING SERVICES +• 51% want real-time visibility into the status of their orders. -212-553-1653 +This inaugural Retail TouchPoints Last Mile Benchmark Report, based on a survey +of 146 retail executives, provides a snapshot of the challenges retailers face in -www.moodys.com +managing the last mile, including shipping and delivery; post-sale services such as setup -877-772-5436 +and installation; keeping customers informed; and the tricky (and expensive) area of -www.standardandpoors.com +product returns. -A.M. BEST CO. +There has been +a 50% increase +in last mile +services over the +past 18 months, +with most of +the demand +coming from B2C +e-Commerce. -www.ambest.com +— EFT LAST MILE +RETAIL STUDY 2018 -Cost — We publish easy-to-use booklets and shopper’s guides. Download them free from our website -at insurance.kansas.gov or call our office to get a free copy 800-432-2484. +3 -Free booklets & shopper’s guides -Auto Insurance Shopper’s Guide +ARE YOU WINNING THE LAST MILE OF RETAIL? + While 36% of +retailers deliver +less than 10% of +orders directly to +customers, 19% +deliver 25% to 50% +of orders direct to +consumers; +24% deliver 75% +of more. -Health Insurance Shopper’s Guide +The Last Mile survey indicates that delivery of products to customers’ homes or offices +already is widespread — and growing. While 36% of retailers deliver less than 10% +of orders directly to customers (versus shoppers taking them directly from a store), +19% deliver 25% to 50% of orders direct to customers, and 24% deliver 75% or more +of orders direct to customers. Presumably, though, this last group of respondents are -Home and Renters Insurance Shopper’s Guide +online-only retailers that lack brick-and-mortar stores as distribution points. -Important +What Percentage Of Products Are Delivered To +Shoppers’ Homes/Offices? +(Versus picked up/purchased in-store) -Life, Annuities and Long-Term Care Shopper’s -Guide +LESS THAN 10% -Medicare Supplement Shopper’s Guide +10% TO 25% -To better understand the complaint process, you might want to review what type of complaint was -filed against the company you are looking at. The National Association of Insurance Commissioners -(NAIC) provides information by code about the complaint. +25% TO 50% -Information is available at https://content.naic.org/cis_consumer_information.htm +51% TO 75% -Input the NAIC company code number found to the left of the company name in the complaint index -report and hit the enter key. Once the company is displayed, select “Click for Report Options.” Next, -select “Go to Complaint Code Report.” You will then choose a report based on the type of coverage, -reason for complaint, or disposition (final outcome). Finally, select Kansas under the “I want to see -complaints for:” drop down. Please note this report varies from the Kansas Insurance Department’s -Complaint Index as it only captures confirmed complaints, which are defined as: +MORE THAN 75% -“A complaint in which the state department of insurance determines: +24% -a) The insurer, licensee, producer, or other regulated entity committed any violation of: +36% -1) An applicable state insurance law or regulation; +19% -2) A federal requirement that the state department of insurance has the authority to enforce; or +14% -3) The term/condition of an insurance policy or certificate; or +7% -b) The complaint and entity’s response, considered together, indicate that the entity was in error.” +4 -page 4 +ARE YOU WINNING THE LAST MILE OF RETAIL? Retailers have +seen their volume +of direct deliveries +increase over the +past 18 months: +less than 10% said +their percentage +has decreased. -Complaint Index Report Indexes by line: Automobile +In line with the results of the eft study, Last Mile respondents have seen their volume +of direct deliveries increase over the past 18 months: 29% say the percentage has +increased significantly, and 39% say it has increased slightly. Less than 10% said the +percentage has decreased, and 23% say the volume has remained the same. -Cocode +How Has The Percentage Of Products You Deliver To +Shoppers Changed In The Last 18 Months? -Company Name +23% -33154 +29% -Alliance Indemnity Company +INCREASED SIGNIFICANTLY -42579 +INCREASED SLIGHTLY -Allied Property & Casualty Insurance Company +DECREASED SIGNIFICANTLY -29688 +DECREASED SLIGHTLY -Allstate Fire & Casualty Insurance Company +STAYED ABOUT THE SAME -19232 +5% -Allstate Insurance Company +4% -10386 +39% -American Family Insurance Company +Which Fulfillment/Delivery Services Do You Offer? +(Among retailers with both stores and e-Commerce sites) -19275 +BUY VIA MOBILE SITE/APP, DELIVER TO HOME -American Family Mutual Insurance Company +BUY ONLINE, PICK UP IN-STORE -13773 +BUY VIA MOBILE SITE/APP, PICK UP IN-STORE -Farm Bureau Property & Casualty Insurance Company +BUY ONLINE OR VIA MOBILE APP, RETURN IN-STORE -21628 +BUY VIA STORE ASSOCIATE’S MOBILE DEVICE, DELIVER TO HOME -Farmers Insurance Company Incorporated +BUY VIA IN-STORE KIOSK, DELIVER TO HOME -13587 +46% -First Chicago Insurance Company +44% -14138 +31% -GEICO Advantage Insurance Company +28% -14139 +21% -GEICO Choice Insurance Company +16% -12966 +5 -Key Insurance Company +ARE YOU WINNING THE LAST MILE OF RETAIL? MANAGING CUSTOMER EXPERIENCE AND +EXPECTATIONS TOPS RETAIL CHALLENGES -23787 +The volume of last-mile activity is not the only thing that’s on the rise. Customer +expectations around delivery times, service levels and visibility into the process +all are increasing rapidly. That’s reflected in the fact that two of the top three concerns +among retailers responding to the Last Mile survey are related to customers: Customer +expectations around delivery times is the leading challenge, at 49%, and customer +expectations of free/low-cost shipping is third, at 44%. -Nationwide Mutual Insurance Company +What Are Your Most Significant Last Mile Challenges? -16322 +Two of the top three +concerns among +retailers are related +to customers: +expectations around +delivery times at 49%, +and expectations +of free/low-cost +shipping at 44%. -Progressive Direct Insurance Company +49% -42919 +47% -Progressive Northwestern Insurance Company +44% -25405 +18% -Safe Auto Insurance Company +CUSTOMER EXPECTATIONS +AROUND DELIVERY TIMES -24740 +RISING SHIPPING +COSTS -Safeco Insurance Company Of America +CUSTOMER +EXPECTATIONS OF FREE/ +LOW-COST SHIPPING -23388 +HANDLING CUSTOMER +QUERIES/COMPLAINTS +EFFECTIVELY -Shelter Mutual Insurance Company +17% -19070 +14% -Standard Fire Insurance Company +12% -25178 +DELIVERY/FULFILLMENT +LOGISTICS -State Farm Mutual Automobile Insurance Company +RETURNS +MANAGEMENT -42749 +MANAGING POST- +PURCHASE SERVICE +PROVIDERS -Traders Insurance Company +6 -25941 +ARE YOU WINNING THE LAST MILE OF RETAIL? “Ultimately, the most important factor in the last mile is the experience of the end +customer (the consumer),” according to the eft Last Mile Retail Study. “Consumers have +the same expectations of service and engagement from the places they shop online as + +they do in-store, except they expect it faster, in line with the speed of the digital world.” + +Amazon is playing a large part in raising the bar of consumer expectations. The +Amazon Prime service effectively “hides” shipping costs in an annual fee, which + +increases pressures on competitors to offer low- or no-cost shipping. Amazon and +others have poured investments into offering two-day, next-day, same-day, and even +less than one-hour shipping to a wider swath of consumers, and it has had an impact +on the whole customer mindset. + +For example, the maximum number of days people are willing to wait for an item to be +delivered in exchange for free shipping has decreased, from 5.5 days in 2012 to 4.1 +in 2018, according to the 2018 AlixPartners Home-Delivery Shopping Survey. Amazon +Prime members are less patient than the population at large: only 19% of Prime +members would wait a week in exchange for free shipping, compared to 31% of non- +Prime members. + +Free shipping has become very much the rule, not the exception. A December 2017 +survey by L2 revealed that 77% of brands offer free shipping on their web sites. +Among this group, 35% do not require a minimum spending threshold to qualify for +free shipping. For those retailers that do require a minimum order, the median dollar +amount of $50 has not changed since 2016. + +The maximum +number of days +people are willing +to wait for an item +to be delivered +in exchange for +free shipping has +decreased from +5.5 days in 2012 to +4.1 days in 2018. + +— ALIXPARTNERS +HOME-DELIVERY +SHOPPING SURVEY -United Services Automobile Association +7 -25968 +ARE YOU WINNING THE LAST MILE OF RETAIL? + DELIVERY IMPERATIVES TO COMPETE WITH AMAZON +WITHOUT BREAKING THE BANK -USAA Casualty Insurance Company +By Rob Taylor, +CEO, Convey -18600 +As retailers fight to stay relevant in the face of Amazon, last mile delivery is being pushed front and center. While Prime -USAA General Indemnity Company +delivery has made the words “fast” and “free” synonymous with a seamless shipping experience, companies have had to -% of total market share represented in this report: +align their operational costs with quickly changing customer expectations, or risk customer loss — 70% will not return after a -2021 Market Share +poor delivery experience. -2021 Complaints +Consumers continue to report that they are unimpressed with delivery standards today — nearly 60% of shoppers say that -2021 Complaint -Index +the delivery experience is mediocre at best, and similarly, for those returning parcel shipments, 42% said that they were not -2020 Complaint -Index +satisfied with retailers’ return processes. -2019 Complaint -Index +This has driven pressures to create the perfect delivery experience, with 83% of supply chain executives saying that CX is -0.99% +now a companywide goal. While the delivery experience is now a standard KPI, many still struggle with legacy technology. -1.22% +By investing in visibility solutions, retailers can identify exceptions that derail delivery, resolve issues before they result -4.63% +in an angry customer and reset expectations to create a positive outcome. For retailers, this responsibility involves defining -0.28% +appropriate expectations through proactive communication, keeping delivery date promises and providing -4.27% +self-service options. -6.12% +Communication: Like any great relationship, the retailer-consumer bond needs to be fueled by communication and met +expectations. For the 93% of consumers who say they want to stay informed and receive proactive updates about their -6.79% +shipments, communication creates a sense of transparency. This visibility and communication helps them answer, “Where is -4.40% +my order?” -0.15% +According to our research, returns communication is equally important, as it was a source of anxiety for 70% of customers. -3.13% +Customers returning items indicated that the “Where is my money?” question was most pressing, and they did not want to -1.94% +deal with the retailer to get that information. -0.81% +Commitment: While customers care about “fast and free shipping,” over 90% indicated that missing a promise date comes +with consequences ranging from expected compensation to disloyalty. If a retailer misses a promise date, consumers -1.31% +expect a proactive response, and this expectation is amplified over the holidays. Over 57% of consumers said they would be -6.59% +impressed if their retailer either refunded shipping costs or expedited a new shipment in response to a late shipment. -8.15% +Control: When it is possible for the consumer to resolve the exception or an anxiety, 97.5% of customers will opt for +self-service options. This includes the ability to re-route an in-transit shipment or expedite a new delivery, with younger -0.22% +consumers also asking for self-pickup options such as lockers. In reverse logistics, consumer control gives the consumer the -1.84% +ability to schedule the return themselves. In either scenario, they seek convenience — they do not want to pick up the phone -3.24% +to call a carrier’s 1-800 number. -3.39% +Customer experience is becoming increasingly important in final mile delivery. Those who can embrace the rise of digital and -16.23% +bend to their customers’ demands can also increase their own flexibility for package delivery. Ultimately, these retailers will -1.18% +be able to create a memorable delivery experience and a loyal following without compromising their operations budget. -2.08% +8 ARE YOU WINNING THE -1.60% +LAST MILE OF RETAIL? -1.42% + CLEARING UP CONSUMERS’ VISIBILITY CHALLENGES -81.99% +Retailers are paying attention to shoppers’ last-mile demands. According to the eft +survey, 47% have implemented delivery feedback surveys, and 40% have increased last +mile visibility for customers. -2 +Respondents to the Last Mile survey use a variety of tools to keep customers informed +about the status of their deliveries: 86% rely on email notifications, along with online +tracking links (38%), text messages (33%) and phone calls (29%). Retailers’ self- +evaluation of their communication skills is generally positive. A majority, 54%, rate them +as “good,” and only 8% rate them as “poor.” -4 +47% of +retailers have +implemented +delivery feedback +surveys and 40% +have increased +last mile visibility +for customers. -42 +How Do You Inform Customers About Their Delivery Status? +(Check all that apply) -10 +— EFT LAST MILE +RETAIL STUDY 2018 -10 +38% -21 +33% -20 +29% -21 +86% -16 +EMAIL NOTIFICATIONS -23 +ONLINE TRACKING LINK -10 +TEXT MESSAGES -16 +PHONE CALLS -4 +How Would You Rate Your Last-Mile +Communications With Customers? -59 +8% -49 +14% -12 +EXCELLENT -9 +GOOD -17 +SATISFACTORY -11 +POOR -74 +54% -20 +24% 9 -7 - -5 - -0.33 - -0.53 - -1.48 - -5.86 - -0.38 - -0.56 - -0.48 - -0.78 - -17.66 - -1.20 - -0.84 - -3.22 - -0.50 - -1.46 - -0.98 - -8.87 - -0.80 - -0.86 - -0.53 - -0.75 - -2.76 - -0.71 - -0.71 - -0.58 - -0.17 - -0.93 - -1.06 - -0.55 - -0.86 - -0.82 - -0.31 - -0.60 - -6.40 +ARE YOU WINNING THE LAST MILE OF RETAIL? + SUCCESS STORY -1.27 +VISIBILITY SOLUTION BOOSTS ENGAGEMENT AND CTR +FOR UNCOMMON GOODS -1.79 +Uncommon Goods, a heavily gift-oriented online business devoted to sustainability and customer service, knew that +its customers expected visibility into where their packages were at any given time. The retailer launched the Engage -3.68 +product from Convey in November 2016 to provide communication and proactive exception management, and has -0.82 +used the platform since then to deliver exceptional customer experiences. -0.84 +When an order ships, customers receive an email with a tracking link that takes them to a mobile-optimized branded -0.70 +page on the Convey platform. Behind the scenes, Convey integrates and translates all carrier information, allowing -7.53 +customers to watch and understand what’s happening as the shipment progresses. Shoppers can sign up for text -0.86 +messages and exception emails; rate the experience and submit feedback to the Uncommon Goods team; click back -0.57 +to FAQs; and in some cases place additional orders via special promotions. -0.50 +Key results include: -0.84 +• 12% of shoppers subscribe to SMS notifications; -1.98 +• 54% of shoppers leave feedback before delivery; and -0.96 +• 6% of shoppers are clicking through to the Uncommon Goods site from the tracking page. -0.94 +Early knowledge of customer feedback and shipment issues such as delays, incorrect addresses or damaged shipments -1.48 +gives the retailer’s team more time to proactively correct problems. “Customers really appreciate being in the loop, and -- +alerted when there’s an unforeseen delay,” said MC Halfpenny, Director of Operations, Uncommon Goods. “Rather than -0.38 +causing them to be angry, often Uncommon Goods looks like the hero because we’re being proactive.” -1.04 +10 ARE YOU WINNING THE -- +LAST MILE OF RETAIL? -0.76 + RETAILERS SEEK NEW SOLUTIONS TO +PERENNIAL PROBLEM OF RETURNS -0.64 +One of the most challenging aspects of the Last Mile is that it’s a two-way street. -0.64 +Customers can (and do) return items, and the rise of digital commerce has raised the -1.29 +stakes — and the costs — in this critical area. -- +“Return rates for online retail are higher — sometimes four times higher — than for -0.77 +traditional brick-and-mortar retailers,” said Patrick Bohannon, Vice President, Retail -1.10 +Strategy for Oracle. “Return policy is very influential to online sales. Consumers want to -4.07 +have easy and cost-free returns, so many retailers are liberalizing their return rates.” -- +A Deloitte 2017 holiday shopping survey indicated that consumers’ top perks for online +shopping were free shipping, at 72%, and store policies for easy returns, at 44%. -1.10 +While the majority (61%) of retailers responding to the Last Mile survey are able to +keep their returns percentage below 10%, more than one in four, 26%, have a returns +percentage in the 10% to 20% range. -0.73 +Returns +percentages +appear to be +on the rise; +only 13% of +retailers reported +decreases over +the past 18 +months. -- +What Is Your Returns Percentage? -0.79 +3% -0.88 +10% -0.93 +61% -0.74 +26% -2.64 +LESS THAN 10% -0.55 +10% TO 20% -0.84 +21% TO 30% -0.39 +31% TO 40% -Indexes by line: Accident & Health +Returns percentages appear to be on the rise: 26% of retailers reported a slight +increase over the past 18 months and 5% reported a significant increase. Only +13% reported decreases. -Cocode - -Company Name - -12321 - -American Continental Insurance Company - -60380 - -American Family Life Assurance Company of Columbus - -61263 +11 -Bankers Life & Casualty Company +ARE YOU WINNING THE LAST MILE OF RETAIL? How Has Your Returns Percentage +Changed Over The Past 18 Months? -47171 +56% -Blue Cross Blue Shield of Kansas City +5% -70729 +26% -Blue Cross Blue Shield of Kansas Incorporated +7% -67369 +6% -Cigna Health & Life Insurance Company +INCREASED SIGNIFICANTLY -68500 +INCREASED SLIGHTLY -Continental Life Insurance Company Brentwood +DECREASED SIGNIFICANTLY -62324 +DECREASED SLIGHTLY -Freedom Life Insurance Company Of America +STAYED THE SAME -64211 +Cost of returns +top the list +of retailer +challenges at +48%, followed by +customer issues +with packing/ +shipping items for +return, at 39% -Guarantee Trust Life Insurance Company +RETURNS CREATE MULTIPLE COST CENTERS AND COMPLEXITIES -73288 +Retailers responding to the Last Mile survey identified returns costs as their biggest +challenge, at 48%, but in addition to this general concern, returns create a number of +other hurdles. -Humana Insurance Company +What Are Your Biggest Returns Challenges? -12459 Medica Insurance Company +COST OF RETURNS SHIPPING/HANDLING -71412 +CUSTOMER ISSUES WITH PACKING/SHIPPING ITEMS FOR RETURN -Mutual Of Omaha Insurance Company +IMPACT ON BOTTOM-LINE REVENUE OF RE-SELLING RETURNED ITEMS -82538 +COMMUNICATING EFFECTIVELY WITH SHOPPERS ABOUT RETURNS +RULES/PROCESSES -National Health Insurance Company +COST/LOGISTICS OF CREDITING SHOPPERS’ ACCOUNTS/ +PROVIDING REFUNDS -67091 +MINIMIZING FRAUDULENT RETURNS -Northwestern Mutual Life Insurance Company +MANAGING IN-STORE STAFF REQUIREMENTS FOR RETURNS PROCESSING -67326 +48% -Old Surety Life Insurance Company +39% -15777 +28% -Oscar Insurance Company +23% -68462 +23% -Reserve National Insurance Company +22% -14345 +18% -Sunflower State Health Plan Incorporated +12 -56014 +ARE YOU WINNING THE LAST MILE OF RETAIL? “Retailers make +meticulous plans +down to the +square inch about +what’s carried in +their stores, but +those plans can +be upset by a bin +of onesie-twosie +returns.” -Thrivent Financial For Lutherans +— DAVID SOBIE, +HAPPY RETURNS -86231 +Returns can play havoc with basic retailing functions, including planning, inventory, -Transamerica Life Insurance Company +pricing and in-store assortments. “Many retailers do not plan for returned inventory as -72850 +an available ‘pot’ of stock, meaning they overbuy,” said Justine Shiou, Solutions Manager, +Planning & Optimization at Oracle. Additionally, “returns are generally made weeks after -United World Life Insurance Company +the initial purchase which reduces the time to re-sell the item at full price. This can be a -96385 +significant issue for fashion/short lifecycle retailers that turn quickly. The selling price is -UnitedHealthcare of Midwest Inc +likely to have already been marked down versus the price when it was purchased.” -70319 Washington National Insurance Company +David Sobie, Co-Founder and CEO of Happy Returns, which operates “return bars” -2021 Market Share +where shoppers can bring back items they have bought online, identified other returns- -2021 Complaints +related issues retailers grapple with. “Broadly speaking, the challenge for retailers is -2021 Complaint -Index +‘How do I get this inventory back and ready to be restocked in a way that doesn’t break -2020 Complaint -Index +the bank?’” said Sobie. -2019 Complaint -Index +Costs include shipping the item back to the distribution center or store, but that’s only -0.61% +the beginning, said Sobie. Someone needs to assess whether the item can be added -3.04% +back into the retailer’s inventory. In some cases this is a simple matter of relabeling -0.79% +and repackaging an undamaged item, but if the item is damaged in some way, -3.55% +refurbishment might be needed. The retailer might need to take a loss or sell the item -22.36% +to a downmarket channel. -3.01% +ONLINE RETURNS UPSET STORE ASSORTMENT PLANS -0.57% +Omnichannel retailers have their own set of returns challenges. A retailer’s online -0.99% +assortment may be much larger than that of an individual brick-and-mortar store, but -0.68% +if shoppers return an item bought online, it needs to be incorporated into the store’s -3.82% +assortment — or shipped to another store or DC at additional cost. -2.12% +“Retailers make meticulous plans down to the square inch about what’s carried in -1.05% +their stores, but those plans can be upset by a bin of onesie-twosie returns,” said -0.53% +Sobie. “That carefully planned store assortment can make the store look like more of a -0.73% +discount retailer.” -0.69% +13 -0.58% +ARE YOU WINNING THE LAST MILE OF RETAIL? + POOR RETURNS EXPERIENCES CREATE NEGATIVE CX -0.56% +Retailers seeking to keep returns costs down may decide to choose slower, less -17.11% +expensive shipping options, but they may end up paying higher costs in other -0.56% +areas. “The longer the returns process takes, the more customer service contacts -0.69% +will be generated,” Sobie warned. “The number-one reason people contact customer -0.75% +service is to check on their order, but the number-two reason relates to returns. These -2.36% +are not insignificant costs, either; they can run to a couple of bucks per contact at a -0.79% +well-run retailer.” -0 +Returns create headaches for consumers as well, meaning that if retailers don’t handle -8 +the process well, they risk losing dissatisfied customers. A 2017 study conducted by -1 +Forrester for Happy Returns, which surveyed online fashion shoppers, revealed that +returns are the worst part of online shopping for 73% of consumers. Among those +that have to return items, 39% say they do not shop at the retailer again. -5 +Returns are the +worst part of +online shopping +for 73% of +consumers. +Among those that +have to return +items, 39% don’t +shop that retailer +again. -46 +— FORRESTER 14 -2 - -4 - -2 - -5 - -16 - -1 - -8 - -0 - -0 - -6 - -1 - -67 - -0 - -1 - -0 - -0 +ARE YOU WINNING THE LAST MILE OF RETAIL? + FINAL MILE STRATEGIES: BALANCING CUSTOMER +SATISFACTION AND PROFITABILITY -5 - -0.00 - -1.02 - -0.49 - -0.55 - -0.80 - -1.80 - -1.37 - -1.57 - -1.14 - -0.51 - -2.93 - -0.37 - -5.91 - -0.00 - -0.00 +By Justine Shiou, Solutions Manager, +Planning and Optimization, Oracle -4.03 +ONLINE RETURNS AND INVENTORY SHUFFLING -0.70 +A sale is not a sale until the customer decides to keep it. Return rates have continued to increase across all channels but -1.52 +online sales have a higher return rate (more than 3X). Despite the operational impacts, retailers are liberalizing return -0.00 +policies: free returns, return in-store, return without limits basically, to reduce friction. This strategy has significantly -0.56 +increased operational cost. An online return to the warehouse includes shipping costs and restocking fees, an online back- -0.00 +to-store return also includes restocking fees but brings another layer of inefficiency, requiring the associate to handle the +item, run it back onto the floor, ultimately distracting him/her from more productive customer engagements. -0.00 +Markdown rates are also climbing exponentially, due to incongruent on-hand store inventory and delayed returns. Returns -2.44 +are generally made weeks after the initial purchase, which means reduced time to resell at full price, particularly for fashion -0.00 +retailers. The effectiveness of localized assortment strategies is degraded due to buy online, return in store (BORIS). The -1.38 +BORIS journey creates incongruent inventory and cluttered sales floors. Retailers are dealing with this in two ways: -0.00 +• Treatment 1: absorb the cost of taking one-off markdowns to clear the stock -1.06 +• Treatment 2: absorb the cost of transferring items back to the online warehouse or to another store with the -0.70 +item in its assortment, a.k.a. inventory shuffling -0.69 +While the practice of inventory shuffling [using transfers] used to be deemed costly, some retailers have decided this -0.00 +treatment is more sales effective than creating markdowns. Neither approach is optimal. -2.46 +TRADE REGIONS: MEASURING TRADEOFF BETWEEN CX AND GROSS MARGIN -0.69 +A focus on total trade region (online and stores combined) when developing omnichannel strategies and the role of the -0.24 +store is critical to long-term success. Retailers need to consider all roles, including fulfillment, pickup, return, traditional -3.42 +sales or a showroom concept. Advanced analysis on departments and items that have high conversion rates need to be -1.37 +conducted to save sales and create new revenue opportunities. Integrating business insights, like customer loyalty, with -9.04 +POS or order system data will more effectively drive planning by customer and fulfilment channel. With improved weekly -0.00 +sales, stock and intake accuracy, and in-season projections by channel, retailers will be able to increase gross margin, and -0.79 +reduce markdown liabilities. The ability to track the effects of cross-channel returns can help limit under- or over-stock -2.37 +situations, plan return percentages by channel and customer journey and plan cross-channel sales fulfilment paths. -2.42 +Retailers intrinsically are bound by the bottom line and shareholder interest. However, the retail environment has -1.51 +changed significantly while the mindset around profitability has not. Including a customer experience measure (such as -0.00 +NPS) in the planning process can also help retailers understand a more holistic view of their business and ensure the -4.73 +plan’s relevance to both gross margin and customer service (t-NPS). -0.00 +15 ARE YOU WINNING THE -0.00 +LAST MILE OF RETAIL? -5.72 + SUCCESS STORY -0.00 +KOHL’S GETS TRAFFIC BUMP FROM AMAZON RETURNS -0.58 +Returns aren’t all bad news. Savvy retailers with well-trained staffs have long been able to turn the return occasion into -0.34 +an opportunity to repair a damaged customer relationship, and in some cases even to take advantage of the consumer’s -1.31 +presence to sell them additional items. -0.76 +Since October 2017, Amazon shoppers have been able to return items at 82 Kohl’s stores. It seems benefits are flowing +to both parties: Amazon gets a physical location for returns, improving customer service and potentially cutting costs, -0.69 +and Kohl’s has seen increased foot traffic in the participating stores. -- +A study by UberMedia found that stores with Amazon Return Centers saw their daily visitors increase 48% after October +2017, compared to a 36% increase in non-Amazon Return center locations. The 12-point difference was over and above +the normal holiday traffic increase to be expected in Q4. -- +RETAILERS TURN TO TECHNOLOGY, +THIRD-PARTY SERVICES AND DISRUPTORS -2.55 +Retailers are tackling their last-mile challenges with a mix of traditional technologies and -0.64 +cutting-edge solutions. Additionally, as might be expected in an area that’s often far -3.26 +from their core competencies, retailers rely heavily on third-party services to fulfill many -0.91 +last-mile chores. -- +More than half, 59%, of retailers deploy courier/delivery services, and 28% drop +ship items directly from suppliers. For services that involve direct customer contact, +however, most retailers prefer to keep these functions in-house: only 14% outsource +customer service call centers, and just 12% rely on others for product delivery and +setup; 22% do not outsource any of their last-mile services. -- +Retailers’ choices for technological aids in this area generally run to the tried and -- +true: transportation management solutions and route optimization systems top the +current usage list, at 35% and 27% respectively. The percentages using more advanced +solutions, such as RFID on shipping containers and products/packaging, are in the -- +teens. IoT sensors, AI/machine learning, and the use of land-based robots or airborne -- +drones are even less common. Perhaps in an area as high-stakes as the last mile, -1.72 +retailers are waiting for the technology to be perfected before taking a chance on it. -- +For services that +involve direct +customer contact, +most retailers +prefer to keep +functions in- +house: 22% do +not outsource any +of their last-mile +services. -- - -0.00 - -0.00 - -3.27 - -% of total market share represented in this report: - -67.92% - -page 5 - -insurance.kansas.gov Indexes by line: Homeowners - -Cocode - -Company Name +16 -37907 +ARE YOU WINNING THE LAST MILE OF RETAIL? Which Last Mile Services Do You Outsource To Third Parties? +(Check all that apply) -Allstate Vehicle & Property Insurance Company +59% -19690 +COURIER/DELIVERY SERVICES -American Economy Insurance Company +28% -10386 +DROP SHIP ITEMS DIRECTLY FROM SUPPLIERS -American Family Insurance Company +14% -19275 +12% -American Family Mutual Insurance Company +10% -10872 +22% -American Strategic Insurance Corporation +CUSTOMER SERVICE/CALL CENTER -13773 +PRODUCT DELIVERY/SETUP -Farm Bureau Property & Casualty Insurance Company +RETURNS MANAGEMENT -21652 +WE DON’T OUTSOURCE ANY LAST MILE SERVICES -Farmers Insurance Exchange +Which Technologies Do You Use +To Manage Last-Mile Activities? -26298 +USE NOW -Farmers Property & Casualty Insurance Company +PLAN TO ADOPT +WITHIN 18 MONTHS -20419 +NO PLANS TO ADOPT -Homesite Indemnity Company +TRANSPORTATION MANAGEMENT SOLUTION -14451 Marysville Mutual Insurance Company +ROUTE OPTIMIZATION SYSTEM -25453 +VEHICLE MONITORING TECHNOLOGY -Nationwide Insurance Company Of America +RFID ON SHIPPING CONTAINERS -23787 +MOBILE DEVICE MONITORING OF DELIVERY PEOPLE -Nationwide Mutual Insurance Company +RFID ON PRODUCTS/PACKAGING -24740 +IOT SENSORS -Safeco Insurance Company Of America +AI/MACHINE LEARNING -23388 +LAND-BASED ROBOTIC DELIVERIES -Shelter Mutual Insurance Company +DRONE DELIVERIES -25143 +35% -State Farm Fire & Casualty Company +27% -27998 +18% -Travelers Home & Marine Insurance Company +16% -38130 +14% -Travelers Personal Insurance Company +14% -25941 +10% -United Services Automobile Association +7% -25968 +6% -USAA Casualty Insurance Company +2% -18600 +21% -USAA General Indemnity Company +27% -% of total market share represented in this report: +24% -2021 Market Share +24% -2021 Complaints +31% -2021 Complaint -Index +28% -2020 Complaint -Index +26% -2019 Complaint -Index +33% -4.36% +12% -1.47% +14% -4.98% +44% -8.70% +46% -1.35% +58% -6.25% +60% -4.85% +55% -1.22% +58% -1.41% +64% -1.20% +60% -1.70% +82% -2.10% +84% -1.70% - -2.48% - -21.08% - -3.42% - -4.39% - -2.61% - -1.39% - -1.23% - -77.89% - -21 - -2 - -13 - -14 - -6 - -13 - -7 - -3 - -5 - -4 - -3 - -1 - -7 - -5 - -78 - -9 - -10 - -2 - -3 - -6 - -1.63 - -0.46 - -0.89 - -0.55 - -1.51 - -0.70 - -0.49 - -0.83 - -1.20 - -1.13 - -0.60 - -0.16 - -1.39 - -0.68 - -1.25 - -0.89 - -0.77 - -0.26 - -0.73 - -1.65 - -0.86 - -- - -0.42 - -0.66 - -0.25 - -0.80 - -0.77 - -1.39 - -- - -1.14 - -- - -0.00 - -0.88 - -1.15 - -1.08 - -0.82 - -0.66 - -0.41 - -0.53 - -2.65 - -1.64 - -- - -1.10 - -0.75 - -0.86 - -0.81 - -1.36 - -0.00 - -- - -2.02 - -- - -- - -0.32 - -1.66 - -0.46 - -0.51 - -0.72 - -0.58 - -0.56 - -0.98 - -Indexes by line: Individual Annuity - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -Cocode - -Company Name - -90611 - -Allianz Life Insurance Company Of North America - -60488 - -American General Life Insurance Company - -61689 - -Athene Annuity & Life Company - -87726 - -Brighthouse Life Insurance Company - -62944 - -Equitable Financial Life Insurance Company - -63088 - -Farm Bureau Life Insurance Company - -63274 - -Fidelity & Guaranty Life Insurance Company - -91642 - -Forethought Life Insurance Company - -65056 - -Jackson National Life Insurance Company - -65676 - -Lincoln National Life Insurance Company - -66869 Nationwide Life Insurance Company - -91596 - -New York Life Insurance & Annuity Corporation - -66974 - -North American Company for Life & Health Insurance - -67466 - -Pacific Life Insurance Company - -86630 - -Prudential Annuity Life Assurance Corporation - -65005 - -RiverSource Life Insurance Company - -68675 - -Security Benefit Life Insurance Company - -62952 - -SILAC Insurance Company - -69345 - -Teachers Insurance & Annuity Association Of America - -56014 - -Thrivent Financial For Lutherans - -5.48% - -3.55% - -4.46% - -3.15% - -3.55% - -2.37% - -2.27% - -2.54% - -8.01% - -5.60% - -2.74% - -3.51% - -2.43% - -1.85% - -1.86% - -2.94% - -2.31% - -5.21% - -3.08% - -1.70% - -2 - -1 - -0 - -0 - -1 - -0 - -0 - -1 - -2 - -1 - -0 - -0 - -0 - -0 - -0 - -0 - -1 - -1 - -0 - -0 - -1.18 - -0.91 - -0.00 - -0.00 - -0.91 - -0.00 - -0.00 - -1.27 - -0.81 - -0.58 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -1.40 - -0.62 - -0.00 - -0.00 - -0.77 - -2.35 - -0.00 - -2.24 - -1.34 - -7.38 - -0.00 - -1.23 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -- - -0.00 - -0.53 - -0.00 - -0.00 - -- - -% of total market share represented in this report: - -68.62% - -1.15 - -1.72 - -1.93 - -- - -2.29 - -- - -1.12 - -0.77 - -- - -- - -- - -1.03 - -0.69 - -- - -- - -- - -0.84 - -- - -- - -- - -page 6 - -Complaint Index Report Indexes by line: Individual Life - -Cocode - -Company Name - -60399 - -American Family Life Insurance Company - -60488 - -American General Life Insurance Company - -63088 - -Farm Bureau Life Insurance Company - -65838 - -John Hancock Life Insurance Company USA - -58033 - -Knights Of Columbus - -65676 - -Lincoln National Life Insurance Company - -65935 Massachusetts Mutual Life Insurance Company - -66044 Midland National Life Insurance Company - -91596 - -New York Life Insurance & Annuity Corporation - -66915 - -New York Life Insurance Company - -67091 - -Northwestern Mutual Life Insurance Company - -67466 - -Pacific Life Insurance Company - -67644 - -Penn Mutual Life Insurance Company - -65919 - -Primerica Life Insurance Company - -68136 - -Protective Life Insurance Company - -79227 - -Pruco Life Insurance Company - -68241 - -Prudential Insurance Company Of America - -69108 - -State Farm Life Insurance Company - -56014 - -Thrivent Financial For Lutherans - -86231 - -Transamerica Life Insurance Company - -69868 - -United Of Omaha Life Insurance Company - -% of total market share represented in this report: - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -1.85% - -1.92% - -3.91% - -1.79% - -2.56% - -2.90% - -4.05% - -3.84% - -2.32% - -4.30% - -11.52% - -2.51% - -1.33% - -1.34% - -2.19% - -3.93% - -0.58% - -4.10% - -1.80% - -1.93% - -1.18% - -61.84% - -0 - -21 - -8 - -1 - -3 - -3 - -0 - -0 - -0 - -3 - -1 - -0 - -1 - -0 - -1 - -0 - -12 - -3 - -0 - -11 - -1 - -0.00 - -6.13 - -1.15 - -0.31 - -0.66 - -0.58 - -0.00 - -0.00 - -0.00 - -0.39 - -0.05 - -0.00 - -0.42 - -0.00 - -0.26 - -0.00 - -11.66 - -0.41 - -0.00 - -3.20 - -0.47 - -0.58 - -2.19 - -0.55 - -0.50 - -1.01 - -0.31 - -0.25 - -- - -0.00 - -0.38 - -0.24 - -0.00 - -0.00 - -1.58 - -2.09 - -0.14 - -- - -0.52 - -0.28 - -1.61 - -- - -0.49 - -3.76 - -0.34 - -0.00 - -0.35 - -0.27 - -0.13 - -- - -0.00 - -0.21 - -0.00 - -0.00 - -0.00 - -0.33 - -0.78 - -0.26 - -- - -0.21 - -0.25 - -2.69 - -- - -Indexes by line: Long-Term Care - -Cocode - -Company Name - -90611 - -Allianz Life Insurance Company Of North America - -61263 - -Bankers Life & Casualty Company - -70729 - -Blue Cross Blue Shield of Kansas Incorporated - -20443 - -Continental Casualty Company - -71404 - -Continental General Insurance Company - -70025 - -Genworth Life Insurance Company - -64211 - -Guarantee Trust Life Insurance Company - -65838 - -John Hancock Life Insurance Company USA - -58033 - -Knights Of Columbus - -65935 Massachusetts Mutual Life Insurance Company - -65978 Metropolitan Life Insurance Company - -71412 - -Mutual Of Omaha Insurance Company - -66915 - -New York Life Insurance Company - -69000 Northwestern Long Term Care Insurance Company - -68241 - -Prudential Insurance Company Of America - -65005 - -RiverSource Life Insurance Company - -25178 - -State Farm Mutual Auto Insurance Company - -56014 - -Thrivent Financial For Lutherans - -86231 - -Transamerica Life Insurance Company - -62235 - -Unum Life Insurance Company Of America - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -2.16% - -9.45% - -2.07% - -3.02% - -1.66% - -17.08% - -3.41% - -9.49% - -3.07% - -1.40% - -4.92% - -4.61% - -3.06% - -6.03% - -2.44% - -1.47% - -3.00% - -2.38% - -4.61% - -3.00% - -9 - -6 - -2 - -4 - -0 - -15 - -1 - -1 - -0 - -0 - -1 - -0 - -3 - -0 - -3 - -0 - -0 - -0 - -7 - -2 - -5.63 - -0.86 - -1.30 - -1.79 - -0.00 - -1.19 - -0.40 - -0.14 - -0.00 - -0.00 - -0.27 - -0.00 - -1.33 - -0.00 - -1.66 - -0.00 - -0.00 - -0.00 - -2.05 - -0.90 - -2.42 - -0.64 - -0.34 - -3.35 - -1.35 +17 -1.24 +ARE YOU WINNING THE LAST MILE OF RETAIL? One area where at least a few retailers are willing to try something new is with -0.00 +disruptive shippers, such as Uber, Deliv, Instacart and Shipt. The December 2017 -0.00 +Target acquisition of Shipt for $550 million, and high-profile Instacart partnerships with -0.00 +retailers including BJ’s Wholesale Club and Albertsons, may signal to other brands that -- +such services are “retail-ready.” -1.86 +Which Non-Traditional Shippers Have You Used +During The Past 12 Months? +(Check all that apply) -0.00 +Top retailer data +uses are focused +on functional +areas, but 42% +are using returns +data to adjust +their product +selections. -0.00 +21% -0.00 +8% -1.07 +7% -0.00 +11% -0.00 +UBER -0.00 +DELIV -3.21 +INSTACART -0.00 +SHIPT -1.78 +59% -0.76 +OTHER E.G. ANDREANI, ENJOY, SPEEDEE, +GOOGLE EXPRESS, SKIP THE DISHES -0.82 +Given the growing volume of last mile activity, retailers should look to step up their data -0.81 +collection and analysis efforts in order to gain a strategic advantage. At this point, top -0.73 +data uses are focused on functional areas such as resolving customer service issues, +at 48%, and tracking the speed/efficiency of drivers, at 46%. A significant minority of +respondents, 42%, are using returns data to adjust their product selections, a smart +move that could potentially reduce future returns rates. -0.32 +What Are Your Most Valuable Uses For Last Mile Data? -- +48% -0.14 +46% -- +42% -- +22% -1.15 +RESOLVE CUSTOMER +SERVICE ISSUES -0.94 +TRACK SPEED/ +EFFICIENCY OF DRIVERS +OR DELIVERY SERVICES -- +USE RETURNS DATA +TO ALTER PRODUCT +SELECTION -0.23 +DETERMINE MORE +EFFICIENT DELIVERY +ROUTES -- +17% -- +13% -0.81 +12% -- +CHANGE SHIPPING OFFERS +(I.E. DISCOUNTS FOR +DELIVERING MULTIPLE +ITEMS IN ONE BOX) -6.39 +CHANGE INTERNAL +FULFILLMENT RULES +(E.G. SET MINIMUM TRUCK +CAPACITIES AT 75%) -- +IDENTIFY LOCATIONS FOR +NEW STORES/ +E-COMMERCE ACTIVITIES -% of total market share represented in this report: +18 -88.33% +ARE YOU WINNING THE LAST MILE OF RETAIL? SUCCESS STORY -page 7 +NEIMAN MARCUS BOOSTS SPEED, SERVICE LEVELS +WITH PROACTIVE SOLUTION -insurance.kansas.gov Filing a Complaint -The Kansas Insurance Department’s Consumer Assistance Division provides free help to consumers. -Trained representatives can answer questions about your coverage or assist when you have a dispute -with your insurance company. +Last-mile challenges are multiplied when retailers are delivering large, bulky items such as furniture and appliances. At -How to file a complaint +least one in every 10 of these “white glove” deliveries faces a problem of some kind, such as an incorrect address or +phone number or a scheduling conflict with the carrier, according to the Wall Street Journal. -If you have tried unsuccessfully to resolve a problem with your insurance company or agent, we -encourage you to contact the insurance department. You may file a complaint online through our -website, by mail or email. We also accept complaints in-person. +When the retailer is Neiman Marcus, known for its high levels of in-store customer service, there’s lots of pressure to +get these deliveries right as often as possible. The supply chain team at Neiman Marcus collaborated with the customer -If you do need to file a formal complaint, we require the following information: +care group, using a Convey solution to track the most complex deliveries. -• Your name +The software captures problems with enough time to correct them before the scheduled delivery date. Within 30 days of +launching the solution, Neiman Marcus had reduced transit times to customer by 24%. -• Your address +CONCLUSION -• Your email address (if applicable) +Retailers and their last-mile partners are racing to keep up with customer expectations -• The policy number +that are only moving in one direction: higher. To make things even more challenging, -• A clear, concise written explanation of your -complaint — this can be sent by postal mail, -fax, or filed through our website +customers want fast, accurate deliveries, hassle-free returns, and enhanced visibility -• Your daytime telephone number +into these processes — all for little or no cost. -• Copies of supporting letters, police reports, +“The biggest challenge in the last mile is matching increasing cost against spiraling +consumer expectations,” said Paul Hamblin, Editor in Chief, Logistics Business magazine, +as quoted in the eft Last Mile Retail Study. “Next-day is already almost standard, same- +day is hot on its heels. Who’s going to pay for all this?” -• Type of insurance involved (automobile, +“The biggest +challenge in +the last mile +is matching +increasing cost +against spiraling +consumer +expectations.” -homeowners, etc.) +Technology and innovation are providing some of the answers, as retailers seek the -• The name of the insurance company, agent +most optimal transportation routes and try out new business models for shipping and -or adjuster +returns. But retailers will need to dig deeper in order to meet their last-mile challenges. -notes, etc. +— PAUL HAMBLIN, +LOGISTICS BUSINESS -• Tell us what has been done, including who -you’ve talked to and what you were told -(names, dates, times, places, etc.) +They need to gather insights about price sensitivity and elasticity in the last mile, so -• Desired outcome +that they can begin recouping at least some of their costs without alienating customers. -Do not send originals of any of your records, and keep a copy of the letter you send us. +They should invest in back-end systems that allow them to make firm delivery date -Contact the Kansas Insurance Department +promises, and to proactively prevent missteps that can damage the customer +relationship. Finally, retailers may want to rethink whether every order needs to be +delivered within 12, 24 or 48 hours. If consumers are confident that their order will -WEBSITE +arrive when the retailer has said it will, they might be more willing to wait a few days. -insurance.kansas.gov +19 -SOCIAL +ARE YOU WINNING THE LAST MILE OF RETAIL? + To Learn More... -KSinsurancedept +Retail TouchPoints is an online publishing network for retail executives, with content focused on +optimizing the customer experience across all channels. The Retail TouchPoints network is comprised of +three weekly e-newsletters, special reports, web seminars, exclusive benchmark research, an insightful -EMAIL +editorial blog, and a content-rich web site featuring daily news updates and multi-media interviews at +www.retailtouchpoints.com. The Retail TouchPoints team also interacts with social media communities +via Facebook, Twitter and LinkedIn. -kid.commissioner@ks.gov -kid.webcomplaints@ks.gov +P 201.257.8528 +info@retailtouchpoints.com -ADDRESS +Oracle provides retailers with a complete, open, and integrated suite of best-of-breed business -1300 SW Arrowhead Rd. -Topeka, KS 66604 +applications, cloud services, and hardware that are engineered to work together and empower -PHONE +commerce. Leading fashion, grocery, and specialty retailers use Oracle solutions to anticipate market -Main: 785-296-3071 -Fax: 785-296-7805 +changes, simplify operations and inspire authentic brand interactions. For more information, visit our +website at www.oracle.com/retail. -CONSUMER ASSISTANCE -HOTLINE +P 1.650.506.7000 +oneretailvoice_ww@oracle.com -800-432-2484 +Convey enables businesses worldwide to take control of shipping and delivery. Our cloud-based -page 8 +platform connects disparate data and processes, giving shippers the insight to make smart decisions, -Complaint Index Report 1300 SW ARROWHEAD ROAD -TOPEKA, KS 66604 -1-800-432-2484 -(785) 296-3071 +and the tools to take action. From parcel to freight, and first to final mile, Convey helps the world’s -INSURANCE.KANSAS.GOV -KID.COMMISSIONER@KS.GOV -(785) 296-7805 (FAX) -1-877-235-3151 (TTY/TDD) +largest brands create better delivery outcomes while reducing costs, improving visibility and -Vicki Schmidt, Commissioner -insurance.kansas.gov +transforming transportation into their next competitive advantage. Founded in 2013, Convey is based +in Austin, Texas. Learn more at: www.getconvey.com. -page 9 +P 602.842.2788 +info@getconvey.com -insurance.kansas.gov \ No newline at end of file diff --git a/q117/random_k2/random_2.md b/q117/random_k2/random_2.md index 5aa602e7fdd86b41f160b8fb2430ecc9847ae72b..d1447fd886d0d69697641e94fcf1890d83769afd 100644 --- a/q117/random_k2/random_2.md +++ b/q117/random_k2/random_2.md @@ -1,128 +1,1679 @@ -Competitive Analysis +Hobart and William SmitH CollegeS -Kit Needham +Handbook for +International Students - Any way the customer -is solving the problem now is -competition +intercultural affairs Center +288 Pulteney Street +geneva, nY 14456 +(315) 781-3319 – Phone +(315) 781-3888 – Fax - DIRECT vs INDIRECT + TABLE OF CONTENTS -DIRECT – similar products and services +SECTION I: General Information + Welcome -INDIRECT – solving same problem for same -target market with a different product or -service +important Campus Phone numbers - Who are Uber’s competitors? + office locations -Direct? +banks and theaters +Places to Stay +Frequently asked Questions -Indirect? +SECTION II: Immigration Procedures + How to get assistance with immigration matters +immigration basics and definition of terms - Competitive Factors + obtaining a Visa and Form i-20 -What makes a customer choose one -solution over another? +differences between F-1 and J-1 Status -Price – cheaper -Service – faster, personalized, convenient -Quality - lasts longer, stylish, tastes better + arriving at a U.S. port of entry… What a student can expect -At most, you can only compete -on 2 out of the 3 +dealing with the department of Homeland Security (dHS) +exceptions to a Full Course of Study +duration of Status for F-1 Students +temporary departure from the US - Unfair Advantage + accepting U.S. government Public assistance or Public benefits + Work opportunities for students with F-1 Status + Work opportunities for students in J-1 Status -Core competency that cannot be copied or -bought -- Deep domain expertise (hospital systems) -- One amazing hard thing (Google algorithm) -- Authority (existing reputation in market) -- Dream Team (previous startup success; rock +Practical training + optional Practical training (oPt) + Curricular Practical training (CPt) +departure Checklist -star) +page 1 +9 +10 +10 +10 +12 - Templates for Analysis +15 +16 +17 +19 +21 +22 +24 +24 +24 +25 +26 +26 +27 +27 +28 +29 - Most Common Format +SECTION III: -You +Sample Fall orientation Schedule for international Students -Competitor 1 +30 -Competitor 2 +2 -Competitor 3 + SECTION I: +GENERAL INFORMATION -Competitor 4 +Welcome! -Feature 1 +Welcome to Hobart and William Smith Colleges (the Colleges or HWS). to you, the +decision to study abroad might have been a small one – or, indeed, a rather big step. +either way you won’t know what your experience will be like until you are here. this +handbook is designed to help you prepare for the adventure of studying in the United +States. in our experience the better prepared you are the more you will benefit from +your time at the Colleges. + While we are trying to answer many of your questions prior to your arrival in the +United States, some of your questions will only arise once you are here. the inter- +national Student adviser will serve as a resource to you and your parents or guardians +throughout the application process and during your stay at Hobart and William Smith +Colleges. The adviser will be your primary contact for any questions or concerns +regarding visas and immigration issues, employment, travel and a host of other + matters that may arise. -Feature 2 +the International Student Office (iSo) is located on the second floor of the -Feature 3 +intercultural affairs Center at 288 Pulteney Street. this office advocates for and ad- +dresses the needs of international students as well as provides programs and services +related to immigration matters. during the academic school year the iSo sponsors +and co-sponsors events designed to promote cross-cultural sharing between the many +groups at the Colleges and +geneva communities. +Currently, international +students come to us from 22 +nations: austria, bulgaria, +Canada, China, Colombia, +Czech republic, germany, +india, iran, nepal, Pakistan, +Portugal, Scotland, South +africa, South Korea, Swe- +den, Switzerland, thailand, +the netherlands, turkey, +Ukraine and Vietnam. -Feature 4 +Intercultural Affairs Center – Home to the International Student Office. -Feature 5 +in the fall international students attend a special two-day orientation before +the regular first year orientation begins. at Hobart and William Smith, we ask that +incoming international students join us ahead of time as it is a wonderful opportunity +to get acquainted with the new culture, get assimilated with the campus, and provide +a forum to answer your questions before the rush of orientation activities begins. at +the iSo, we will check to make sure that all your immigration paperwork is in order as +well as help you gain a better understanding of what your F-1 or J-1 status means and +how to maintain it; help to answer any questions that you might have about adjusting +to life as a student in the United States, specifically life at HWS; and meet students, +staff and faculty members. - Examples of Features +3 -• Price -• Benefits -• Quality -• Durability -• Image/style -• Service -• Warranties -• Convenience + Questions? Please +contact the International +Student Adviser by phone + at (315) 781-3319. The + International Student + Office is located on the + 2nd floor of the +Intercultural Affairs Cent- +ter, 288 Pulteney Street, +Geneva, NY 14456. (315) +781-3319 – Phone (315) +781-3888-Fax. Office +hours are 8:30 am to +5:00 pm weekdays during +the academic year. -• Ease of Use -• # of features -• Type of features -• Wow factor -• Location(s) -• Distribution/Sales -• Certifications -• Endorsements +THINGS TO DO BEFORE GOING ABROAD +before you can board a plane and start your study +abroad adventure, there are some formalities you will +need to take care of. the earlier you start thinking +about the things you need, the smoother your prepara- +tions will go. this list will help you consider some of +the things you will have to do. - Where do you find info +Obtaining a Visa +Upon receipt of the form i-20 (for F-1 students) or dS- +2019 (for J-1 exchange students) you can apply for your +student visa. it is wise to schedule an appointment with +the U.S. embassy or consulate in your country as soon +as possible as sometimes there are no appointments +available for weeks. You should also give yourself some +time to prepare for your visit to the U.S. embassy or +consulate. Visa procedures vary from country to country. +make sure to check the website of the embassy or -1. Customer Discovery interviews -2. Advertising -3. Sales Brochure -4. Newspapers/Magazines -5. Website -6. Online sites that rate products/services -7. Your business and professional network +consulate in your country to find out what documents you need to bring. document +you certainly need include: (1) a current, valid passport; (2) your i-20 or dS-2019 +form; (3) evidence of financial support for the duration of time you will spend in the +US; (4) proof that you have a permanent residence outside the US; (5) one or more +passport pictures meeting U.S. requirements; (6) a nonimmigrant visa application. - Other +Medical Insurance +HWS Colleges require that all students buy the medical insurance provided by +the colleges. all J-1 student-visa holders (that is, all exchange students) are also +required by U.S. law to have their own medical insurance. While F-2 students are +covered by the Colleges medical insurance you might nevertheless consider buying +additional medical coverage. -• Research competitors that no longer are in +Important Documents +You need to bring all important documents relating to your legal, medical, and aca- +demic history with you. Such documents might include medical and dental records, +certificates of immunizations, information about medical conditions and treatments, +prescriptions for medication and eyeglasses or contact lenses, and official transcripts +from secondary school and college. make sure you photocopy all documents and +leave some copies with a person you trust in your home country. it is also helpful to +translate these documents into english before you leave your home country. -business +Packing +Pack as light as possible. remember that you might have to carry all your things at +different times during the journey. if you are studying abroad for only a semester you +might not need to bring your entire wardrobe. You might also want to leave room +in your suitcase for souvenirs you will buy while in the US. at the same time there +might be things you want to bring because they will make you more comfortable -• Research matrix will be messy; matrix for +4 -investment pitches are summaries of -strongest competitors with most important -features + and will make your transition to a new environment easier. Some exchange students +and international students also like to bring small gifts from their home country to +share with new friends abroad. this might be a means for you to share and explain +your culture to your new acquaintances. if you have traditional, national cloth- +ing consider bringing it to wear at events sponsored by the international Students +association or other cultural events. if you arrive during spring semester (January), +to bring a warm jacket and boots, a scarf, gloves and hat. We encourage you to take +advantage of the option to ship your packages ahead of time using the box number +that will be provided to you from the Colleges orientation office. You may also +consider purchasing larger items upon your arrival at our local shopping centers. - Finally +Traveling Tips +You can book your flight as soon as you find the Colleges start date for international +Student orientation. at this time, you might have already received the schedule of +international student orientation. the Colleges provide a complimentary pick-up +from rochester and Syracuse airport for those international students coming to the +Colleges for the first time if you arrive during the assigned times. long-distance +traveling can be really exhausting, so try to get a good-night’s sleep before the trip, +eat lightly and drink plenty of water. it is also a good idea to wear several layers of +clothes so you can accommodate to changes in temperature. label your luggage with +both your home address and the address in the United States (if you don’t have an +address yet, use that of the international student office). Keep you passport, impor- +tant documents, flight tickets, and money or traveler’s checks with you at all times. +You might also want to pack everything you need for a one-night emergency stay +into your hand luggage in case you get stuck at the airport or your luggage gets lost. -• Describe your competitive advantage +Money +You should carry a small amount of U.S. dollars with you while traveling so that you +can buy some food or other small things you might need. However, it is not a good +idea to carry large sums of money. We recommend that you buy traveler’s checks or +get a credit card that allows you to withdraw money from a cash machine when you +need to. Upon arrival on campus, the international student adviser can help you to +set up an account with a local bank. it is highly recommended to have an U.S. bank +account because it will make payments so much easier. Students can create bank +accounts free of charge. -Benefits sell -Features tell +GETTING TO KNOW HOBART AND WILLIAM SMITH COLLEGES +College is an exciting time for everyone and it is a transition for all students. if you +are an exchange student, you already went through the transition process once. +nevertheless, studying abroad brings with it a range of new experiences. every col- +lege is a little different and if you have never studied in the United States before, +many things might be very confusing. - Examples +Hobart and William Smith Colleges +Hobart and William Smith is a small, private liberal arts institution. You can expect +to study in relatively small classes and build close relationships with your classmates - Specialty analysis examples +5 - Positioning diagram but will still -need a feature matrix analysis + and professors. Virtually all students are being housed on campus. living so close +to each other has many advantages: it is relatively easy to get to know new people, +get involved in clubs and activities, and make an impact on the community. at the +same time it can be hard to find personal time and separate school from other activi- +ties when the two are not physically separated. - Works if there are only two main -features/dimensions +Classes +at HWS, a normal course load is four classes per semester. this means that you +will be in class for between 12 and 15 hours every week, or more if you take courses +requiring labs or studio work. there are different types of classes. 100- and 200-level +classes are usually introductory classes and have a wider focus. 300- and 400-level +classes have a narrower, more specific focus. Some professors prefer to lecture, but +you also encounter class-discussions. Higher level classes tend to be taught in a semi- +nar style, in which a lot of the class’ content relies on student presentations. there +are also independent study classes in which you research a topic under a professor’s +guidance but without the classroom structure. as an upper-class student you might +choose to work on an honors project. this is an independent research project that +will culminate in a 80- to 100-page paper. You can take classes from all departments +but at the end of your Sophomore (2nd) year, you will need to declare a major and a +minor subject (or one major, two minors; or two majors, one minor; or two majors, +two minors). as part of your liberal arts education, you will also have to fulfill cer- +tain academic goals. Your academic adviser will help you with that. + +Grading and Grades +Foreign students often find that americans are very competitive. this also reflects +on the system of grading. You will find that you will have many opportunities to +prove your knowledge of a subject in class. many professors will give short, unan- +nounced quizzes that test your knowledge of the course material assigned for that +day. exams are bigger, announced tests that cover the course material of a longer pe- +riod of time. they can be compromised of true or false questions, sentence comple- +tion (“fill in the blank”), multiple choice, matching, essay questions, or a combina- +tion of these. Professors will also assign essays and papers you have to write on your +own time. Your class participation might also count toward your final grade. the +Colleges use the grading system based on a four-point scale: a= 4.0, b= 3.0, C= 2.0, +d= 1.0. You need at least a C average in your classes to remain a full-time student. + +Homework +Some foreign students are overwhelmed by the quantity of homework assigned by +american colleges. especially in the beginning when you still have problems with +the english language, keeping up with homework can be really hard. if you find +yourself struggling to keep up with work, ask for help. Find the Center for teaching +and learning on campus and schedule an appointment! Ctl is open to all students +and its services are already covered by your tuition. they can help you with your +writing assignments, help with time management, or assign you a tutor to help you +with the content of a class. it is also important to realize that most american stu- +dents prioritize their work: decide what readings are important to do and read them +thoroughly; skim the others. +6 + + Professors and Advisers +every student at HWS is assigned an academic adviser. He or she will help you +with any questions you have related to your academic life. Feel free to consult them +before choosing classes, if you are struggling with certain classes or professors, if you +want to know how and when to declare majors and minors, and any other questions +you might have. most professors are also interested in helping you as much as pos- +sible and want to get to know you outside of class. don’t feel embarrassed to ask a +professor a question after class. You should also take advantage of their office hours. +interaction between professors and students in america is a lot tighter than in many +other countries. Students sometimes stop by their professor’s office simply to talk +about something that they are interested in. + +Important information on financial aid +Hobart and William Smith Colleges offer merit scholarships as well as need-based +grants to international students. the average package typically equals more than +half the annual cost of tuition, room and board. Highly competitive applicants will +be considered for a full tuition scholarships. + +Financial aid is available for a maximum of four years for those who enroll as + +first-year students. For those entering with advanced standing (transfers), aid is +available for the time allotted to complete a degree. awards based on demonstrated +financial need, if it becomes apparent that such need no longer exists, may be re- +duced. Students must resubmit a financial aid application each year. + +ADJUSTING TO A NEW ENVIRONMENT +going abroad for a semester or the duration of one’s studies is exciting. at times, +however, the confrontation with a foreign culture can be challenging. Studying +abroad is a learning experience in every sense of the word. besides your scholastic +learning in college courses, you will get to know the american culture and how +people around you view their world and their lives. by experiencing and understand- +ing a different way of life, you will also learn about things in your own culture you +have never thought of before. this is your chance to broaden your view and reflect +on and mature your own values. the process of learning about a new culture and +finding your own place in the new environment can be confusing and tiring. the +best way to deal with a new situation and find out about something you don’t under- +stand is to ask questions. + +Culture Shock and Adjustment +Culture shock is a normal phenomenon on your way to adjusting to a new environ- +ment. For some people, the encounter with a new culture can be shocking, indeed, +but for most people culture shock is not as sudden and dramatic as you might expect. +When the excitement about being in a foreign country wears off, cultural differences +can become overwhelming. You might think your language skills are inadequate; you +might be struggling with course work; you might be troubled by unfamiliar food; you +might have difficulties interpreting other people’s behavior; you might be offended +by the way people act or dress or by their body language. there are a variety of +symptoms of culture shock. Some people feel tired and nervous; some are frustrated + +7 + + with or even hostile toward the american culture; some get angry and irritated very +easily; some isolate themselves or socialize only with other international students and +people who speak their language. You might also feel very homesick and question +why you decided to go abroad in the first place. a way to prepare yourself to deal with +culture shock is to clarify for yourself why you want to study abroad. Knowing your +personal expectations and goals will help you overcome periods of doubt. + +THINGS TO DO +• Chose a non-credit or half-credit class like tennis, scuba diving, or basic sailing. +• Join some of the many student clubs on campus (http://www.hws.edu/studentlife/ + +campuslife/sll/clubs). + +• You like to be active? Visit the sport and recreation facilities and take a yoga class, +borrow a racquet and play tennis or racquetball. You can also try out for intramu- +ral sports. + +• get involved in the religious life on campus! there are services and meditation as +well as bible study sessions, ecumenical candlelight prayers, Sabbath dinners and +a lot more. + +• Participate in ongoing events – take in a movie on Fridays at Friday Flix, visit the +Cellar Pub in the basement of Coxe Hall and see what is happening (comedians, +bands, etc.), show school spirit by supporting one of many sporting events hap- +pening on campus. + +• are you interested in current events? Political and social developments? the +Fisher Center lecture Series, the President’s Forum and the President’s radio +Show are a few venues through which you can satisfy this interest. + +• travel the area! intercultural affairs (iC) and the Center for global education + +(Cge) will occasionally offer trips to see the area, oraP (outdoor recreational +activities Program) organizes hiking, rock climbing and rafting trips (among +others), and you can always get a group of friends together and go somewhere on +your own. + +• offer your help in the language departments. maybe someone needs a Spanish + +tutor or a teaching assistant for mandarin? + +• discover geneva. Volunteer. to find out how visit the Public Service office in + +trinity (2nd floor) (or call x3825). + +• do you like to dance? or watch others dance? the dance department is big on +campus and the students as well as the faculty put on dance shows! ask about +Koshare or join one of the dance teams. + +• Check the daily Update! there is always something going on – just show up and + +try it out! + +8 + + P +H +O +N +E +N +U +M +B +E +R +S +& +O +F +F +I +C +E +S + + Extension + +IMPORTANT PHONE NUMBERS +Office +business office +Center for global education +Center for teaching and learning +Counseling Center +deans: Hobart +deans: William Smith +Financial aid +Hubbs Health Center +information technology +intercultural affairs +registrar +residential education +Security +emergency +Student accounts + +3655 +3307 +3351 +3388 +3300 +3467 +3315 +3600 +4357 (HelP) +3319 +3651 +3900 +3656 +3333 +3386 + +OFFICE LOCATIONS +admissions +alcohol and other drug Programs +alumni House +athletics (WS) +buildings and grounds +business office +Career Services +Center for global education +Ctl +Chapel +College Store +Communications office +deans: Hobart +deans: William Smith +dining Services + +Financial aid +Hubbs Health/Counseling Center +information technology +intercultural affairs +President +Provost’s office +Public Service office +registrar +residential education +Security +Student accounts +Student life and leadership +WeoS radio Station + +bradford House +Smith Hall +615 South main +Winn-Seeley +medbery Parking lot +demarest (basement) +trinity (1st Floor) +trinity (3rd Floor) +Harris House (678 S. main St.) +St. John’s Chapel (near 630 S. main St.) +51 Saint Clair Street, Sherrill Hall +durfee House (639 S. main St.) +Smith Hall (1st Floor) +Smith Hall (1st Floor) +Scandling Center Café, Saga, +Cellar Pub, de Cordova Café +demarest (basement) +119 St. Clair St. +Williams Hall +288 Pulteney St. +Coxe Hall (2nd Floor) +Coxe Hall (2nd Floor) +trinity (3rd Floor) +gulick (1st Floor) +101 St. Clair St. +medbery Parking lot +demarest (basement) +Scandling Center (2nd Floor) +113 Hamilton St. + +9 + + BANKS AND MOVIE THEATERS IN GENEVA + +Community Bank +5 Seneca St. +geneva, nY 14456 +Phone: (315) 789-7700 +Website: www.communitybankna.com + +First Niagara Bank +811 Canandaigua road +geneva, nY 14456 +Phone: (315) 789-1049 +Fax: (315) 789-6009 +Website: www.firstniagarabank.com + +BANKS +Five-Star Bank +P.o. box 193 +geneva, nY 14456 +Phone: (315) 789-2300 +Fax: (315) 787-3263 +Website: www.five-starbank.com + +Seneca Falls Savings Bank +342 Hamilton St. +geneva, nY 14456 +Phone: 315-789-8100 +Website: www.seneca24.net/ + +MOVIE THEATERS +Geneva Movieplex 8 +t & C Plaza, Hamilton St. +geneva, nY, 14456 +Phone: (315) 789-4524 +Web: www.cinemanorth.com + +PLACES TO STAY +Note: the Colleges provide this list as a service to all who might visit our area, but +do not endorse any local hotel, motel, inn or bed and breakfast. + +the following are a handful of accommodations that are available in and around +geneva, no more than 30 minutes from campus. lodging is also available in ithaca, +rochester and Syracuse, which are each approximately one hour from Hobart and +William Smith. You can find out more about accommodations outside the greater +geneva area by visiting the geneva Chamber of Commerce or ontario County +tourism bureau websites. + +Holiday inn, Waterloo +(315) 539-5011 or (800) HolidaY + +Hampton inn +(315) 781-2037 + +ramada inn, geneva, lakefront +(315) 789-0400 + +belhurst Castle/White Spring manor, +geneva +(315) 781-0201 + +microtel inn & Suites, Seneca Falls +(315) 539-8438 + +geneva-on-the-lake, geneva +(315) 789-7190 or (800) 3-geneVa + +motel 6, Hamilton Street, geneva +(315) 789-4050 or (800) 466-8356 + +10 + + G +E +N +E +V +A + +, + +N +E +W +Y +O +R +K + +PLACES TO EAT +*Please note that some restaurants are closed on Mondays or Tuesdays, +so call ahead. + +The area code for all numbers listed below is (315). + +American/Continental +abigail’s restaurant, routes 5 & 20, Seneca Falls, 539-9300 +belhurst Castle, lochland road, route 14, 781-0201 +the Crow’s nest, 415 boody’s Hill road, 781-0600 +emile’s restaurant, east north Street and routes 5 & 20, 789-2775 +geneva-on-the-lake, 1001 lochland road, 789-7190 +madderlake Café, West lake road, 781-2424 +the red brick inn, 291 main St., 789-3247 + +Diners/Grilles +Flour Petal Café, 34 linden St., 781-2233 +Parkers’ grille & tap House, 100 Seneca St., 789-4656 +Ports Cone & grill, 4432 W. lake road, 789-2020 +Water Street Café, 467 exchange St., 789-2560 +the Captain’s room, 372 exchange St., 781-0122 +dana’s time out, 258 Hamilton St., 781-2107 + +Italian-American and Pizza +nonna Cosentino’s trattoria, 1 railroad Place, 789-1638 +Uncle Joe’s Pizzeria, 99 n. genesee Street, 781-1199 + +International +ling ling, 481 Hamilton Street, 781-0668 +main moon, 480 exchange Street, 781-5106 +Wing tai oriental restaurant, 164 Castle Street, 789-8892 +garcia’s mexican restaurant, 313 Hamilton Street, 789-5760 + +Fast Food/Family Chains +routes 5 & 20 west of downtown geneva (Hamilton Street) offers an array of popu- +lar fast food and family chains, including Quiznos, tim Horton’s, Club 86 bagels +and Cakes, Friendly’s, KFC-taco bell, mcdonald’s, long John Silver’s, Pizza Hut, +Subway, Waffle Works and Wendy’s. + +11 + + FREQUENTLY ASKED QUESTIONS (FAQs) + +Do you require the TOEFL for international students? +Yes, we do require the toeFl (test of english as a Foreign language). We offer limited +english language support on campus, and require that students have a solid foundation +in english prior to admission. the minimum required computer based toeFl score is +220. the Colleges also will accept the aPiel examination in lieu of the toeFl. + +Do you offer financial assistance for international students? +need-based scholarships are available to international students. the cost of attend- +ing college here for the 2007-2008 academic year is U.S. $45,968. the average aid +package typically equals more than half the annual cost of tuition, room and board. +Highly competitive applicants will be considered for a full-tuition scholarship. the +financial aid application forms are included in the application for admissions. + +How many international students do you currently have on campus? +We have 32 full-time matriculated students in a degree program Fall 2006 +Semester. the following countries are represented: austria, bulgaria, Canada, +China, Colombia, Czech republic, germany, india, nepal, Pakistan, South Korea, +Sweden, turkey, Ukraine, United Kingdom and Vietnam. + +Do you have an international student adviser? +Yes, the student adviser is responsible for working with our international students +on campus to help them make the transition from home to the Colleges. We offer an +international student orientation prior to the first-year and transfer student orienta- +tion at the beginning of the academic year, and the beginning of the spring semester +for new arrivals. a variety of activities and programs are available throughout the +year. Student advising and counseling is available in a variety of other areas on cam- +pus, including faculty advisers, Hubbs Health Center, and the Center for teaching +and learning. We make every effort to work individually with each student to help +them reach their potential and make the most of the opportunities available here. + +Do you have an international student organization? +Yes. the international Student association (iSa) is a student organization that cel- +ebrates the diverse cultures found around the globe. other cultural clubs with interna- +tional perspectives represented on campus include asian Student association (aSU), +latin american organization (lao), Caribbean Student association (CSa), and +dia al-din: muslim Student association. + +Do you have E.S.L. services? +Yes, an entry level course is offered, as are workshops and one-to-one tutoring. How- +ever, the Colleges do not offer a full e.S.l. curriculum. + +What kinds of housing are available? +Within the Hobart and William Smith community are dozens of smaller communi- +ties--those of the individual residences. the range of housing options is extensive, + +12 + + with coed or single-sex college residences, cooperative houses, theme houses, and +town houses among the choices you’ll find over your four years here. residence +halls offer a variety of configurations, with traditional singles and doubles as well as +suites. Some recent theme houses have included Jewish Culture House, multicul- +tural Sisterhood House, Writers House, and international awareness House. the +ten cooperative houses provide an alternative to standard residence-hall living, with +residents sharing responsibility for food and housecleaning arrangements. + +Is it possible to work on campus? +You are eligible to work up to 20 hours a week on campus, if your visa status permits +it. However, the demand for such employment exceeds the number of jobs available. +therefore, you cannot use on-campus employment as a resource for the issuance of +the i-20 form. + +STORAGE AND TRANSPORTATION OF SCHOOL YEAR ITEMS +Special consideration will be given to those international students who need storage +on campus during the school breaks. international students must attend a manda- +tory meeting in September regarding storage and transportation of school year items. +attendance will be taken at this meeting. Students unable to attend this meeting +must take the initiative to set up a meeting with the international Student adviser +before oct. 1. + +the storage and transportation meeting in September will encompass the following +information: +a. date and time of transportation to and from storage areas + +- Winter break (december) +- Summer break (may) +b. location of storage areas +c. How to pack and maximum number of storage items allowed +each student will be allowed no more than 8 items. items must be labeled with +students name, return date and number of items. boxes must be secured properly +and must be easily manageable. large items such as refrigerators, sofas, etc. will not +be accommodated due to limited space. +d. transportation requirements: items are to be placed at curbside 15 minutes +before van arrives for pick up. +e. Cost of transportation = $20 per student (gas and labor cost) +f. information on surcharges: the student’s account will be charged a surcharge for +each infraction: + +- No shows = $50 +- Fail to have items at curbside = $20 +- Not packed = $20 +- Change of time and/or date without 24 hour notice to Intercultural Affairs + Office = $40 + +Please Note: Storage space is limited, therefore there is no guarantee that you +will get storage or transportation support. + +13 + + IN CASE OF AN EMERGENCY +if you should have a need to contact the international Student adviser while travel- +ing outside of the United States please call (315) 781-3656 and ask for extension +3319. You can also contact the office directly by calling (315) 781-3319. + +In an emergency, they can be reached at (315) 781-3333. + +remember that the international Student adviser, while a resource to you, is not +your only resource during your time at the Colleges. While on campus, your respec- +tive deans will be available to assist you with a variety of issues, and should always +be kept aware of important matters. + +WHAT TO EXPECT DURING SCHOOL BREAKS +Since not everyone is able to return home during school breaks, international +students are allowed to remain on campus during those times. at specific times +during the fall and spring, you will receive e-mails from the residential education +office asking about their plans during the upcoming break. Your immediate response +is necessary as it is the only way to ensure that you will get permission to remain in +your room or elsewhere on the campus during school breaks. + +Students are not allowed to remain in their rooms during fall, winter and spring + +breaks. during summer vacation, students who remain on campus must move into +temporary housing. once again, an e-mail from the residential education office +is the first step in determining where students will be staying during the summer. +Currently, international students are allowed to stay on campus during the summer +holiday free of charge (subject to change). + +the dining hall is not open to students during the summer months so students +will need to provide their own meals. most summer housing will feature a kitchen, +but in the event that one doesn’t, appropriate accommodations will be made for +those students who wish to prepare their own meals. + +the international student adviser must also know your plans and should also be + +notified if you will or will not be staying on campus. + +US Holidays/Observances +January 1 +January 15 +3rd monday in February +2nd Sunday in may +4th monday in may +3rd Sunday in June +July 4th +1st monday in September +october 31 +november 11 +4th thursday in november +december 4 +december 25 +december 26 + +14 + +new Year’s day +martin luther King Jr. day +Presidents day +mother’s day +memorial day +Father’s day +independence day +labor day +Halloween +Veterans day +thanksgiving +Hanukkah begins +Christmas day +Kwanzaa + + SECTION II: +IMMIGRATION PROCEDURES + +HOW TO GET ASSISTANCE WITH IMMIGRATION MATTERS + +the international Student adviser is available to help you process your immigration +forms and documents. Working with the department of Homeland Security can be +very complicated, detailed, and time-consuming. it is important that you do not rely +on information obtained from friends and other students because circumstances may +vary from one student to another and the regulations can and do change. + the international Student adviser must work within the federal immigration +laws. When making a request, please make sure that you allow plenty of time for +your request to be processed and plan to be patient. Have your documents, such as +your passport, i-94 card, i-20 or dS-2019 with you when you go to the international +student office for immigration assistance. + +Important: You should allow at least one week for processing signatures, letters +and/or other types of paperwork. Special consideration will be given for emergency +situations. + +Reminder: the international Student adviser cannot remember when each +student’s documents expire. it is YOUR responsibility to keep track of all immigra- +tion documents and keep them current. + +Hristo Dimitrov ’09 (Bulguria), and Ayokunle Abogan ’06 (Nigeria) + +15 + + IMMIGRATION BASICS AND DEFINITION OF TERMS + +as with all countries, the United States has laws and regulations governing foreign- +ers who are temporarily within its borders. the immigration regulations are adminis- +tered by the Department of Homeland Security (DHS). + +immigration regulations can change frequently and are difficult to interpret. it +is important for international students to understand the rules as they apply to stu- +dent status. The DHS does not consider ignorance of the law a legitimate reason +for failing to obey it. + Questions should be directed to the international Student adviser, who will be +able to assist with legal regulations and can sign your immigration documents. + +dHS requires that a copy of your passport, i-94 and other immigration docu- +ments be kept on file by the international Student adviser, as well as your local +address and telephone number. + +Passport +be sure to keep your passport in a safe place. it is your most important document. +most passports are good for a limited period of time. Your passport must be valid at +all times while you are in the United States. + +the expiration date of your passport should be six months beyond the date you + +enter the U.S. if your passport expiration date is close to the date you will begin +classes, obtain a new passport before leaving home. + +it is your responsibility to remember when your passport will expire and to con- +tact the nearest consulate or embassy for your country. to find the nearest consulate +or U.S. embassy check http://usembassy.state.gov. + +Your passport can only be extended by your government and it usually takes +several weeks for a passport to be renewed, so be sure to allow enough time. You may +extend your passport in the U.S. at your home country’s embassy or consulate. You +are in this country illegally if your passport expires and are subject to deportation. + +Visas +the visa stamped in your passport at the U.S. embassy or Consulate is for permis- +sion to enter the United States for a specific purpose and period of time. Students +are usually given an F-1 visa. J-1 visas are given to researchers, scholars and students +who are sponsored by agencies or universities. + +Since the visa stamp is only important for entry and re-entry to the United +States, there is no need to worry if it expires while you are here if you are maintain- +ing your student status. because a visa is only a permit to apply for entry into a coun- +try, it is not possible to obtain a U.S. visa while you are in the U.S. to apply for a +visa, you must go to a U.S. consulate or embassy in a country other than the United +States, usually your home country. + +I-94 (Arrival/Departure Record) +Form i-94, also called the arrival/departure record of Stay Permit, is the small white +card given to you at the port of entry to the U.S. this form is very important since it +officially determines how long an international student can remain in the U.S. + +16 + + Students on the F-1 and J-1 visas usually have their i-94 stamped “d/S” (dura- + +tion of Status), which means that the student can remain in this country for the +length of time required to complete his/her studies as long as the student follows all +of the regulations of the student visa. the eleven-digit number written on your i-94 +is your dHS admission number, which is used by the dHS to monitor entry into and +departure from the U.S. The I-94 cards are turned in when leaving the U.S. and a +new I-94 is completed with each subsequent entry. + +if you are on a visa other than a F or J, or if you have a specific date stamped on + +your i-94 card instead of the “d/S” designation, you must apply for an extension, a +Change of Status, or exit the U.S before the expiration date or you will be consid- +ered out of status and a visa overstay, which can have serious dHS consequences. + +SEVIS +an acronym for Student and exchange Visitor information System (SeViS). it is a +data collection and monitoring system that creates an interface between institutions +of higher education, the department of Homeland Security (dHS), consulates and +embassies abroad, and ports of entry. Schools are required to make regular electronic +updates in SeViS throughout each semester on the records of their enrolled students +in F-1 and J-1 status and their dependents and their researchers and faculty on J-1 +status. this includes, but is not limited to enrollment status, changes in address, +changes in level of study, employment recommendations and school transfers. + +Form I-20 +When Hobart and William Smith Colleges accepts an international student for +full-time study, the student is sent a Form i-20 a-b as proof of acceptance and to +use when applying for an F-1 Student Visa at a U.S. Consulate abroad. the student +presents the i-20 with their passport and visa at the Port of entry and the i-20 is +stamped and given back to the student. this is your Certificate of eligibility, and +it must be signed by a dSo/ international Student adviser before you leave the +United States. Students should keep copies of all i-20 issued to them. + +Immigration Status +this is often confused with “visa” but your immigration status, e.g., F-1, J-1, etc. is +determined at the time of your entry into the U.S. by an immigration official and is +noted on the Form i-94. You may have many visa stamps in your passport but, upon +entry into the U.S., an immigration inspector will admit you in only one immigra- +tion status which is noted on the i-94 card. + +OBTAINING A VISA AND FORM I-20 +Your passport should be valid at least six months beyond the date of entering the U.S. +as a student. be sure to check your passport expiration date. if it will expire close to +when you will begin classes, obtain a new passport before applying for your visa. + +How to Obtain a Visa +although U.S. embassies will not issue a student visa prior to 120 days before the +start date on the i-20 (or dS-2019 for J’s), you can schedule an appointment prior + +17 + + to the 120-day period. Since processing times have become longer due to increased +security clearances, students should check with their local U.S. embassy or consulate +to learn the estimated waiting time and allow plenty of time to obtain the visa. a +list of consulates and embassies can be found at http://usembassy.state.gov/. + +You will need to pay a $100 SeViS fee at least 3 business days before your visa +interview. the fee can be paid online or by mail. For complete directions, see http:// +www.fmJfee.com. + Unless you are a sponsored J-1 student or already in another visa category, the +Colleges’ office of admissions will send you a Form i-20 when you are admitted and +have submitted proof of your ability to finance your education. take the i-20 form, +your acceptance letter, passport, copies of sponsorship letters and bank statements +to prove financial support, and a receipt for your SeViS fee payment to the United +States embassy or consulate nearest your home. once you have acquired the visa +stamp in your passport, you are ready to enter the U.S. + +Note: Students may not enter the U.S. more than 30 days prior to the start date +on their I-20. + +Canadian and some British Commonwealth citizens do not need visas to enter +the U.S., but they must request F-1 status and present their i-20 forms for process- +ing at the point of entry even if the official does not ask for it. they should also be +prepared to show evidence of financial support and proof of citizenship. Without a +stamped i-20 and an i-94 card with the F-1 or J-1 designation on it, you are con- +sidered a tourist in the United States and will not be legally able to study and take +advantage of the special privileges allowed those in Student Status until he or she +has been in an immigration status appropriate for studying for at least nine consecu- +tive months. + +Note: If you will not be entering the U.S. on an F-1 visa, do not use an I-20 form +to obtain a visa. If you received an I-20, please return it to the Office of Admis- +sions with an explanation of the type of visa you hold. + +if your visa document (i-20 or dS-2019) does not arrive in time for you to secure +the proper visa, do not enter the United States on a B1/B2 Visitor Visa. obtain a +b-2 “prospective student” or “prospective exchange visitor” visa from a U.S. embassy +or consulate post in your country. be certain the consul writes “Prospective Student” +or “Prospective exchange Visitor” (for J-1) in your passport on the visa page. You +will need to present your letter of acceptance and evidence of financial support to +request this visa. + +in general, it is best to avoid using the b-2 visa since you will have to pay a fee +and do a time-consuming change of status through dHS (department of Homeland +Security) after you arrive. You may not begin your studies until the change is com- +pleted, so this could also delay the start of your studies until a subsequent term. + +Note: Students must attend the school named on the I-20 or DS-2019 presented at +the port of entry in order to be in status. Also, do not enter the U.S. on a “W-T” + +18 + + (waiver) as this classification cannot be extended or changed and you will lose your +legal immigration status as a student. + +Students Transferring From another Colleges in the United States +if you are an F-1 student transferring from a U.S. institution and will not be leav- +ing the country prior to enrollment at Hobart and William Smith Colleges, follow +the procedures required by dHS. bring your completed i-20 to the international +Student adviser’s office as soon as possible after your arrival on campus. Sign the +document and leave it to be processed by a dSo/ international Student adviser +who will notify you when you may pick it up. + +if you leave the U.S. before beginning your studies at Hobart and William +Smith Colleges, use the new i-20 issued by the Colleges when you arrive at the port +of entry. Keep the i-20 from your old school with your immigration records, but do +not use it to enter the U.S. it is not necessary for transfer students with an unex- +pired student visa and who have maintained their status to obtain a new visa, even +though the former school’s name is listed. + +Transfer to another School +to be eligible to transfer schools, you must be pursuing a full-course of study at your +current school. otherwise you are “out of status” and must apply to dHS for rein- +statement to student status, which, if granted, will allow you to transfer. + +Students transferring to another school must notify the dSo/international +Student adviser of their intent to transfer. Hobart and William Smith Colleges +must release your SeViS record to the new school before a new i-20 can be issued +for the transfer. Upon arrival at the new school, the student should take the new +i-20 to the international Student adviser at the new school. J-1 students wishing to +transfer should consult with their sponsor. + +Changing Your Academic Level +if you are currently a student at Hobart and William Smith Colleges, and will be +continuing your studies in an additional program of study (such as a mat Program), +you should follow the transfer procedure. a new i-20 will be issued for the new pro- +gram, and you will use the new i-20 to enter the country. if you do not leave, bring +it to the international Student adviser as described in the transfer process in the +preceding paragraph. Failure to complete this “internal transfer” will cause you to +be out of status with dHS, which can adversely affect work authorizations or other +student privileges. + +DIFFERENCES BETWEEN F-1 AND J-1 STATUS +there are two types of entry visa issued to nonimmigrant students who intend to +study full-time at Hobart and William Smith Colleges—the F-1 student visa and the +J-1 exchange visitor visa. + +• Source of Funding: J-1 exchange visitor visa status is available to those students +who are supported substantially by funding other than personal or family funds. +Such funding may include that which comes from the U.S. government or the + +19 + + student’s home government, or an international organization, or the Colleges. Stu- +dents who are supported by personal or family funds are ineligible for J-1 immigra- +tion status and must come to the U.S. to study in F-1 immigration status. + +• Foreign Residence Requirement: Students in J-1 immigration status and the J-2 +dependents may be subject to a “foreign residence requirement.” J-1 students who +receive direct or indirect U.S. or home government funding, or who are studying +in fields for which personnel are considered in short supply in their home coun- +tries (most developing nations have “Country Skills” lists of varying lengths), are +ineligible to apply for a change to another non-immigrant status (except a or g) +or permanent residency in the U.S. until they have satisfied the “two year home +country physical presence requirement. only those in J-1 status with government +funding or whose skills are on Country Skills list are subject to the foreign resi- +dence requirement, which mean upon completion of the program of study they must +reside in their country of last legal permanent residence for two years before they are +eligible to apply for entry into the U.S. on an immigrant visa or on specialized work +visa (H or l). there is no foreign residence requirement for students on F-1 student +status. refer to the department of State’s website for additional information on the +exchange Visitor Program. + +• Medical Insurance: Students in J-1 status and their J-2 dependents are required to +have comprehensive medical coverage in order to fulfill U.S. government regula- +tions governing the J exchange Visitor status. Currently the medical insurance +offered by Hobart and William Smith Colleges does not cover all requirements for +those in J immigration status. the U.S. department of State specifies that program +participants and their dependents have medical coverage with the following mini- +mum requirements (22 CFr 62:14): medical benefits of at least $50,000 per accident +or illness; repatriation of remains in the amount of $7, 500, medical evacuation of +the exchange visitor to his or her home country in the amount of $10, 000, and a +deductible not to exceed $500 per accident or illness. For more information please +contact the international Student adviser. + +• Work Permission: Students in either F-1 or J-1 immigration status may work part +time (up to 20 hours per week) on campus with permission from the international +Student adviser. For both statuses, permission to work off campus based on eco- +nomic need may be requested only after the first full academic year of student status, +and only under extraordinary circumstances of unforeseen need. the United States +Citizenship and immigration Services (USCiS) authorizes F-1 students to work off +campus, whereas J-1 students submit an application for work permission to their visa +sponsor, i.e. the agency or school that issued the dS-2019 form. + +• Practical or Academic Training: Students in F-1 and J-1 immigration status are +eligible to engage in differing lengths of practical or academic training during their +studies and at the completion of their academic programs. Practical training, a +privilege of F-1 immigration status if you are eligible, is employment in one’s field of +study. those in F-1 immigration classification may be eligible for “curricular” practi- + +20 + + cal training (which is authorized by the international Student adviser) during their +programs, in addition to one year of “optional” practical training that may take place +during or after one’s program of study. Permission to engage in practical training may +be granted only after you have been in valid student status for one academic year, +except in a graduate program that requires it from the first term. the international +Student adviser recommends optional practical training and the USCiS authorizes +it. refer to the Practical training overview for complete information on both cur- +ricular and optional practical training. academic training, a privilege of J-1 immi- +gration status if you are eligible, is employment which is integral, not just related, +to one’s field of study. one is eligible to apply for permission to engage in academic +training after one academic term in valid student status. a total of 18 months of +academic training is possible during or following completion of a program of study, +with the exception of some postdoctoral research positions that may qualify for up +to 36 months. an application for academic training is made to the J-1 student’s +exchange Visitor program sponsor. refer to academic training authorization for +those in J-1 exchange Visitor immigration Status for complete information. + +ARRIVING AT A U.S. PORT OF ENTRY … +WHAT A STUDENT CAN EXPECT +Plan your arrival: You may be refused entry into the United States if you attempt to +arrive more than 30 days before the start date listed on your SeViS i-20 form. + +Always carry your passport and I-20 on hand: do not check the following docu- +ments in your baggage. if your baggage is lost or delayed, you will be unable to pres- +ent the documents at your port of entry. as a result, you may not be able to enter +the United States. + +In addition, it is strongly recommended that you carry on hand the following + +documentation: + +1. evidence of financial resources +2. evidence of student status, such as tuition receipts +3. name and contact information for your international Student adviser, + +including a 24-hour emergency contact number at the school + +Complete your entry paperwork: +Arriving by Air: Flight attendants will distribute Customs declaration Forms and +arrival departure Forms (i-94). these must be completed prior to landing. + +Arriving by Land or Sea: the CbP officer at the port of entry will provide the +necessary Customs declaration Form and the arrival-departure Form (i-94) to be +filled out upon arrival. + as you arrive at the port of entry proceed to the terminal area for arriving +passengers. Have the following documents available for presentation: you passport, +i-20, arrival-departure Form (i-94), and the Customs declaration Form. Your docu- +ments will be examined and processed by the department of Homeland Security +(dHS) at the port of entry. Your passport will be returned to you along with an i-94 +card (arrival/departure record) designated “d/S” which means “duration of status” + +21 + + and refers to the time you are engaged in a full-time course of study as indicated on +your visa document. the dHS official will stamp and return your i-20 to you. + +Note: Your documents are extremely important. They legitimize your presence in +the United States. Keep them in a safe and secure place. + +For Further Information: +Preparing to Study in the U.S. http://educationusa.state.gov/ +applying for a Visa to Study in the U.S. http://travel.state.gov/visa/temp/types/ +types_1268.html + +MAINTAINING & BEING IN STATUS WITH DHS +an F-1 student is considered in-status for the entire length of time during which +he/she is enrolled as a full-time student in an educational program if adhering to +the conditions of the visa designated on the i-94 card, plus any authorized period of +practical training and an additional sixty (60) days to prepare for departure. be sure +to pay close attention to the conditions described below: + +• report to the international Student adviser for SeViS registration upon arrival +at Hobart and William Smith Colleges. + +• attend only the school authorized. this means that you must only attend the +school whose i-20 you last used to enter the U.S. or the school whose i-20 you used +to transfer schools. if you decide to withdraw from the Colleges before completing +your program, you are not allowed to remain in the U.S. with in-student status un- +less you transfer to another institution. + +• Complete a full course of study for each academic term and make normal academ- +ic progress towards completion of program. attendance and/or full-time enrollment +is not required during the summer. at Hobart and William Smith Colleges 4 credit +hours per semester is required and exceptions can be made on the recommenda- +tion of the international Student adviser for valid medical or academic reasons as +defined by the dHS. be sure to speak with the international Student adviser about +any exceptions beFore dropping below full-time status in order to avoid the seri- +ous consequences that may result from being out of status. + +• transfer or extend program in a timely manner. duration of Status expires if a +student takes longer than expected to complete a given academic level (e.g. bachelor’s, +master’s, and doctorate). the amount of time permitted for completing studies at a +given level is determined by the date on the initial i-20 issued at the beginning of each +academic level. Students must pay close attention to the expected completion date +noted in item 5 on the form i-20 for each level. if more time is needed, an applica- +tion for extension of stay must be filed with the dHS. there are certain requirements +to meet in order to be eligible for an extension, so you will need to consult with the +international Student adviser at least 45 days before the completion date. + +22 + + • Keep continuity in the program of study (in general an absence from the U.S. of +more than 5 months terminates status. However, an authorized study abroad pro- +gram does not break continuity). + +• refrain from unauthorized work. When you apply for an F-1 visa, you must prove +that you have full financial support. the law assumes that you will not be employed +in the U.S. and therefore will not accept employment that could be filled by a U.S. +worker or student. there are some exceptions to this policy provided that full-time +enrollment during the year is maintained and the appropriate permission is ob- +tained, but you should always check with the international Student adviser before +accepting employment, especially if it is off campus. + +• When traveling abroad, always return in proper F-1 status. + +• maintain a valid passport for 6 months into the future (Some countries have +agreements with the U.S. that recognize the passport as valid for 6 months after the +expiration date). + +• report any address changes to the international Student adviser within 10 days +of the change. dHS regulations require that the international Student adviser +maintain a record of the current address where you are living. therefore, you must +always notify the international Student adviser immediately of any change of ad- +dress. addresses with a post office box number or addresses of your academic depart- +ment are not acceptable to the dHS. + +• Comply with Special registration requirements if applicable (nSeerS). + +• File timely requests of change of status, practical training and other benefits. + +• obey all State and Federal laws. + +• depart form the U.S. in a timely manner. + +• do not enter on a tourist visa (b-1 or b-2). + +Note: Students who fail to maintain full-time enrollment are not eligible for any +of the benefits of the student visa, such as on-campus and off-campus employ- +ment, practical training, remaining in the U.S. during summer vacation, transfer +to another school or change of level, or registration for further courses. they are +considered to be “out of status” and must apply to dHS for reinstatement to valid +student status. eligibility for benefits will be regained only if and when the dHS +approves the reinstatement. if you have any questions regarding reinstatement, see +the international Student adviser immediately. + +23 + + EXCEPTIONS TO A FULL COURSE OF STUDY + +• the normal vacation period; + +• medical problems documented by a licensed medical doctor, osteopath or clinical +psychologist. approval for reduced or no course load for a medical condition is limited +to 12 months total for the program of study; + +• teaching or research assistantship; + +• Valid academic reasons, such as first semester language difficulties and/or adjustment +problems, improper course placement, unfamiliarity with american teaching methods; + +• the final term when the student needs less than a fulltime course load to complete +program of study. Student must be enrolled in at least one credit; + +• during a period of full-time optional or Curricular Practical training. + +DURATION OF STATUS FOR F-1 STUDENTS +an F-1 student is admitted to the U.S. for a period known as “duration of Status” +which is recorded upon entry documents i-94 and i-20 with the notation “d/S”. du- +ration of Status refers to the period a student is pursuing a full course of study, plus +any authorized practical training following completion. The dates on the I-20 de- +termine the initial parameters of the duration of status. A student can be admit- +ted to the U.S. up to 30 days prior to the report date on the I-20 and may remain +60 days after completion of program or practical training. Certain factors may +reduce grace period to 15 days or to 0 days. an international Student adviser au- +thorized withdrawal from classes is allowed a 15-day period for departure. a student +who fails to maintain status is not eligible for an additional period for departure. + +TEMPORARY DEPARTURE FROM THE U.S. +If you plan on leaving the U.S. temporarily to visit your homeland or some other +country, an International Student Adviser must validate your I-20 for re-entry into +the U.S. You may not be allowed back into the U.S. without the validated immigra- +tion documents, a current visa, and a passport which is valid for at least six months +after you re-enter the country. + most students with valid status may travel to Canada, mexico and many of the +adjacent islands if they have a valid passport, a valid i-94 and a valid i-20 or dS-2019. +if you re-enter the U.S. in fewer than 30 days, an unexpired U.S. visa is not required. +However, Canada, mexico and the island nations do require entry visas for citizens of +certain countries. + +it is always best to check with the embassy or consulate of any country you wish + +to visit. Since certain countries are ineligible for this benefit and travel regulations are +subject to change, you should check with the international Student adviser if you +intend to travel with an expired visa. + +Students who apply for a new visa in contiguous territory forfeit the right to re- + +24 + + enter with an expired visa. if the new visa is denied, the student would not be able to +re-enter the United States to study until a valid visa is obtained in their home country. + +ACCEPTING U.S. GOVERNMENT PUBLIC ASSISTANCE OR PUBLIC +BENEFITS +international students and scholars should not, under any circumstance, accept public +benefits from the federal, state, county or any other local government agencies in the +United States. this includes, but is not limited to, participation in programs such +as medicare/medicaid, Supplemental Social Security income, aid to Families with +dependent Children, food stamps, and subsidized housing. accepting such benefits +can cause you to lose your legal status in the United States and create a bar to your +re-entry to this country. + +Cooperation between state social service agencies and the inS and department + +of State make it possible for an inS inspector or U.S. Consular officer to become +aware of an individual who has accepted public benefits. the acceptance of public +benefits is interpreted as a declaration that the individual is unable to meet his or her +financial obligations (a requirement of obtaining and maintaining status) and thereby +rendering him or herself ineligible for F-1 or J-1 status. + occasionally a well-meaning social service agency, hospital or public housing admin- +istrator may offer such assistance not realizing that non-immigrants are ineligible for such +benefits. additionally, even though U.S. citizen children of non-immigrant parents may be +eligible for pre and post-natal public benefits, the non-immigrant parent may be denied a +visa or re-entry based on the child’s receiving the benefits. again this is interpreted as the +non-immigrant individual not being able to meet his or her financial obligations. + +Information on U.S. Social Security Numbers (SSN) +if you will be employed in the United States, including at Hobart and William Smith +Colleges, you must obtain a U.S. Social Security number, which is used by the U.S. +government to identify wage earners for tax purposes. the number assigned to interna- +tional students is not a valid U.S. Social Security number. + +to apply for a Social Security number (SSn) you must obtain a “proof of work” + +form from the international Student adviser confirming your student or exchange +visitor status, that you are authorized to engage in on-campus employment, the nature +of the employment to be engaged in, and identify the department/office for whom you +will be working before being able to apply to the Social Security administration for +a SSn. You must also present your passport, your i-20 or dS-2019 form and your i-94 +form. the Social Security administration office is open monday-Friday from 9:00 a.m. +to 4:00 p.m. (closed on federal holidays). + +Social Ssecurity numbers will not be issued to foreign nationals who are not +authorized to work. a Social Security number is not required for opening a bank +account, registering for school, filing a U.S. tax form or applying for a driver’s license. +if you are required to file a tax return and are not eligible for a U.S. Social Security +number, you will have to apply to the internal revenue Service (irS) for a tax iden- +tification number (tin) on form W-7. this form may be obtained from the Colleges +payroll department. + +For more information about U.S. Social Security Numbers visit the website or + +call their toll-free number at 1-800-772-1213. + +25 + + WORK OPPORTUNITIES FOR STUDENTS IN F-1 STATUS +The following is a list of all opportunities to obtain authorization to work in the +United States if you are in F-1 immigration status. Each category may allow for +authorization under different circumstances and at different times during your +stay in the U.S. This is summary of work benefits but does not include all regu- +lations governing employment. + +On-campus Work: Full-time students are allowed to work up to twenty hours per +week during the academic year. on campus employment means that you are paid by +Hobart and William Smith Colleges or working at a Colleges location with an em- +ployer that is providing services to students, such as Sodexho’s Saga dining hall or +the Café. on-campus employment requires an i-9 employment verification autho- +rization form to be completed by you at the Colleges business office. + +Off-campus Work Based on Financial Need: Students who have been in F-1 status +for one academic year and who are experiencing extreme financial difficulties due to +unforeseen changes that severely affect their financial resources may apply for work +permission. authorization is given by the dHS with the recommendation of the +international Student advisers/ dSo. + +Internship with an International Organization: Students in F-1 status are eligible +to apply for permission to work for international organizations such as the World +bank, the international monetary Fund, etc. authorization is granted by the dHS +with the recommendation of the international Student adviser/dSo. + +WORK OPPORTUNITIES FOR STUDENTS IN J-1 STATUS +J-1 On-Campus Employment +enrolled J-1 students may work on campus up to 20 hours per week during the +semester and full time during the break with the permission of their sponsor. if the +Colleges are your sponsor please see the international Student adviser. + +J-1 Off-Campus Employment +if a J-1 student needs employment during the academic year due to financial hard- +ship arising form unforeseen circumstances, the program sponsor may authorize it. +Such employment cannot interfere with full-time pursuit of the exchange visitor’s +program objective and is limited to 20 hours per week during the academic year and +full-time during break. + +PRACTICAL TRAINING: +this is defined as employment related to your field of study and is available dur- +ing and after completion of your studies. the following is a list of the categories of +practical training: + +Curricular practical training (before completion of your studies) is defined as + +employment pursuant to a required or optional internship which is an integral part +of the established curriculum of your school. + +26 + + optional practical training is defined as employment related to your field of study. +it is authorized in variable increments, deducted form the maximum of one year of +optional practical training per educational level, and is available in the following +variations: + +1. optional practical training after completion of your studies +2. optional practical training before completion of your studies with three pos- + +sible variations: + +a. practical training during the annual vacation. the annual vacation must + +follow at least two consecutive terms of fulltime enrollment and is usually taken in +the summer + +b. practical training when school is in session, provided that employment + +does not exceed 20 hours a week + +c. practical training following the completion of all the course requirements + +for the degree except the thesis or equivalent + +Optional Practical Training (OPT) Before Completion of Degree Requirements +definition: employment related to your field of study before completion of all degree +requirements. it is authorized in variable increments and is deducted from the +maximum of one year (12 months) of optional practical training per educational +level. there are three circumstances under which this type of practical training can +be authorized: + +1. Following completion of all degree requirements except the thesis or equivalent +2. during the annual vacation +3. Part-time (up to 20 hours per week) during the academic year + +Eligibility to Engage In OPT: You must: +• be a registered student in F-1 status physically in the United States at the time of +application +• have been enrolled in lawful student status on a full-time basis for at least one full +academic year +• not have used twelve months or more of full-time curricular practical training +oPt is permitted for a period of up to a maximum of twelve months per educational +level (bachelors, masters, etc.). + +Procedure: obtaining authorization for oPt is a two-step process. Your first step in +obtaining an employment authorization document (ead) is to qualify for a certifica- +tion from the international Student adviser/dSo recommending you for oPt. the +international Student adviser/dSo cannot authorize oPt - only the United States +Citizenship and immigration Services (USCiS) can do so by issuing you an ead. + +You may not begin employment until you have been issued an ead by the US- + +CiS and can work only within the time period indicated on the card. requests for +employment authorization must be mailed to the USCiS regional Service Center +having jurisdiction over your place of residence and may require 6 to 12 weeks for +adjudication. + +27 + + When to Apply: +• not more than 90 days prior to being enrolled for one full academic year· +• not more than 120 days prior to the employment start date that you select +Please note: the employment start date cannot be earlier than the date on which +you will have completed one full academic year. + +How to Apply: +Obtaining the International Student Adviser recommendation +Since you must first obtain a new SeViS i-20 with the international Student ad- +viser/dSo’s oPt endorsement, which may require 5 working days, you should make +every effort to apply as early as possible within the 90 or 120 day window. You might +need to come twice to the international Student adviser/dSo—once to submit the +application and another to pick up the approved recommendation on a SeViS i-20. + +Curricular Practical Training in F-1 Status +Definition: Curricular practical training (CPt) is employment pursuant to an intern- +ship requirement or internship elective which is an integral part of the established +curriculum of your school, which will be counted towards completion of your degree. + +Eligibility to engage in CPT: You must be a degree candidate who has been in +lawful status for at least one academic year. You must have been offered a specific +training opportunity that must fall into one of the following categories: +• to satisfy an internship requirement expected of all students in your program +and necessary for completion of your degree; +• to satisfy an internship elective which counts towards completion of your degree. +the internship elective must be listed in your school’s bulletin and must state in the +official course description that it is specifically designed to allow students enrolled in +the course to engage in off-campus internships; +• the internship may be done to engage in research necessary and integral to the +completion of the doctoral dissertation. + +Restrictions: U.S. Citizenship and immigration Services (USCiS) regulations do +not place an absolute limit on the amount of CPt you may utilize. However, if you +engage in one year or more of full-time CPt, you will lose your eligibility to engage +in optional Practical training. CPt is always part-time during the academic year +unless the student is engaged in doctoral dissertation research. CPt may be full time +during the annual vacation. You may engage in curricular practical training only +after your SeViS record has been updated by the international Student adviser and +the i-20 has been printed, indicating the approved employment. the endorsement +will indicate the precise details of the authorized training, including the name and +location of the employer, the specific dates of the training period, and whether it +is full-time or part-time. We encourage you to contact the international Student +adviser/dSo if you have questions. + +28 + + Departure Checklist + +q admissions letter from HWS + +q evidence of Financial resources + +q Passport with Student Visa + +q Form 1-20 or dS-2019 + +q Copies of medical records and other important documents translated into english + +q if J-1 visa holder – additional health insurance requirements. this must also be + +translated into english, if needed. + +q emergency outfit packed in carry-on luggage + +q Calling card to use during travel + +q emergency phone numbers (HWS: 1-315-781-3333) + +q minimum of 100 U.S. dollars for use in an emergency + +q Sweater or jacket in event of bad weather + +q if taking prescription medicine bring prescription for future refills (translated into + +english) + +q Photocopy of your passport, visa, etc. do not carry in the same place as your + +original documents. + +29 + + SECTION III: +*SAMPLE* ORIENTATION SCHEDULE *SAMPLE* + +Hobart and William Smith Colleges +international Students orientation +“let Your adventure begin!” +august 22 -25, 2006 + +Tuesday, August 22 +8:30 a.m. –7:00 p.m. + +Wednesday, August 23 +8:30 a.m. –12:00 p.m. + +arrival/Check-in +location: office of intercultural affairs +288 Pulteney Street +**bring all immigration paperwork + +arrival/Check-in +location: office of intercultural affairs, +288 Pulteney Street +**bring all immigration paperwork + +HWS Student id Cards +location: registrar’s office + +Campus tour + +Shopping @ Wal-mart & Wegmans Supermarket, +Van leaves from intercultural affairs parking lot + +12:30 p.m. - 1:30 p.m. + +lunch with deans of the Colleges +location: blue room in Scandling Center + +1:45 p.m. – 2:45 p.m. + +overview of library Services +location: Warren Hunting Smith library + +6:00 p.m. – 7:30 p.m. + +dinner +location: intercultural affairs Center + +Thursday, August 24 +7:30 a.m. – 8:00 a.m. + +8:15 a.m. 9:45 a.m. + +30 + +breakfast +location: Saga dining Hall in the blue room + +immigration regulations overview +location: Conference room, 2nd Floor, +intercultural affairs + + 10:00 a.m. – 11:00 a.m. + +11:05 a.m. – 11:35 a.m. + +on Campus employment/Social Security +elaine Ferrara, business office +location: tba + +benefits of the HWS Health insurance +ann brink, bursars office +location: tba + +12:00 p.m. – 1:30 p.m. + +lunch with Chaplain lesley adams +location: Chaplain’s residence, 630 S. main St. + +1:45 p.m. – 2:15 p.m. + +money matters +beth turner, director of Financial aid +location: tba + +2:30 p.m. – 3:00 p.m. + +a Healthy You +location: Hubbs Health Center + +3:15 p.m. – 3:45 p.m. + +5:30 p.m. – 7:30 p.m. + +Friday , August 25 +10:00 a.m. – 11:00 a.m. + +Study basics +Center for teaching & learning +location: Center for teaching & learning +(Ctl) + +Welcome dinner +President and mrs. mark gearan & Family +location: President’s House – 690 South main St. + +guide to U.S. education & Culture +amy teel and doug reilly, Center for +global education +location: delancey House + +12:00 p.m. – 1:00 p.m. + +lunch +location: delancey House + +1:30 p.m. + +Join regular campus wide orientation schedule + +Welcome to Hobart and William Smith Colleges + +31 diff --git a/q117/random_k4/question.json b/q117/random_k4/question.json index 5918aed577967e4b58e5aed0e7a4d50003be2232..107b6a869c924e15c6f6d6e499e5d4abf620cf10 100644 --- a/q117/random_k4/question.json +++ b/q117/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "2023.acl-long.386.pdf", - "web_ab28fbcd92cfea3d.pdf", - "web_28449e05c7149be2.pdf", - "web_24ca5dd0f7f78cc6.pdf", - "web_6bb860a08864dd7a.pdf" + "web_0e04ded35ac1bdc2.pdf", + "web_ae1b8a373064ff5e.pdf", + "resize_r1_33.pdf", + "resize_r1_4.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q117/random_k4/random_1.md b/q117/random_k4/random_1.md index 5dac85680ad1e5631acf8a091a98a42ab0400813..0523beec0142edd0aa6f2e27f15673239129abcd 100644 --- a/q117/random_k4/random_1.md +++ b/q117/random_k4/random_1.md @@ -1,2417 +1,1200 @@ -COMPLAINT INDEX -REPORT FOR YEAR 2021 +SURVEY REP ORT -Vicki Schmidt, Commissioner -insurance.kansas.gov +SP ONSORED BY - Table of Contents +EXCLUSIVE SURVEY -About this Report +ARE YOU -Indexes by line: Automobile | Accident & Health +THE LAST MILE +OF RETAIL? -Indexes by line: Homeowners | Individual Annuity + TABLE OF CONTENTS -Indexes by line: Individual Life | Long-Term Care - -File a Complaint - - 01 - - 05 - -06 - -07 - -08 - - About this Report - -What is a complaint? - -The Kansas Insurance Department receives thousands of complaints and inquiries each year about -insurance companies. Complaints and inquiries from the public give us an indication of the kind of -service companies are providing to their customers. We can then work with companies that need to -improve in certain areas. - -Complaints can also help us identify insurance companies or agents that may need to be investigated -for violations of Kansas insurance laws or regulations. The Department also uses complaints to focus -public education efforts on the issues that concern Kansas insurance consumers. - -Each time we receive a complaint, we notify the insurance company and any involved agents. -We send the company a copy of the complaint and ask that it be reviewed and an explanation -provided. Companies are required by law to respond to the Department within 15 working days. -Most companies are cooperative in working with us to resolve consumer complaints. We keep a -record of each complaint filed by a consumer for up to three years. - -A formal complaint is any written communication expressing dissatisfaction or concern with an -insurance company or an agent. General written inquiries received from consumers are not counted -in this report. We also do not include complaints regarding products not within the jurisdiction of the -Kansas Insurance Department, such as self-funded plans and Medicare prescription drug plans. - -NOTE: A formal complaint must be made in writing. - -What is a complaint index? - -The complaint index shows how a company compares to other companies in the same line of business. -It is a statistic that considers a company’s share of complaints in relation to its market share or share of -business written in Kansas. Since the complaint index is measured by market share, companies in all -lines of business will be measured against an index of 1.00. - -For example, a complaint index of 1.00 means the insurer’s share of all complaints received is equal -to its share of all the business written in Kansas. An index of 2.00 means that the insurer’s share -of complaints is twice as large as its share of business written. An index of 0.00 indicates that no -complaints were received for the company. - -The formula is as follows: - -Complaint share - -Market share - -OR - -Company’s complaints / -all complaints - -Company’s premium / All -written premium - -page 1 - -insurance.kansas.gov Which complaints do we use? - -This report shows complaints for the six major types of insurance: accident and health (including -HMOs), annuity, automobile, homeowners, life and long-term care. - -We do not include complaints regarding products not within the jurisdiction of the Department, such as -self-funded plans and Medicare prescription drug plans. In 2021, the Kansas Insurance Department -closed a total of 2,205 complaints. 2,110 involved matters within our jurisdiction. - -To ensure fairness, all closed complaints are counted without any attempt to distinguish between -“valid” or “invalid” complaints. - -Which companies are shown? - -Unless otherwise noted, we list the top 20 premium-writing companies for each line, as well as any -company with 10 or more closed complaints for that type of insurance during the calendar year. - -During 2021, $6,273,694.92 was recovered for Kansas consumers as a result of the Department’s -assistance. In 2021, the largest number of complaints concerned automobile insurance (29%) and -accident & health Insurance (28%). Here is the breakdown: - -All Closed Complaints & Inquiries - -Automobile - -29% - -Life - -3,000 - -2,500 - -Accident & Health - -28% - -Long-Term Care - -2,000 - -Homeowners - -Other - -15% - -14% - -Annuity - -1,500 - -1,000 - -500 - -0 - -*rounded to the nearest percent - -Closed Complaints - -Closed Inquiries - -9% - -4% - -1% - -2018 - -2,689 - -1,049 - -2019 - -2,643 - -878 - -2020 - -2,157 - -873 - -2021 - -2,205 - -795 - -All Closed Complaints & Inquiries - -Recoveries Per Year (In Millions) - -7 - -6 - -5 - -4 +Are You Winning The Last Mile Of Retail? ............................................................................................................................ 3 -2 - -1 - -0 - -6.27 - -4.12 - -3.28 - -1.84 - -2018 - -2019 - -2020 - -2021 - -page 2 - -3,000 - -2,500 - -2,000 - -1,500 - -1,000 - -500 - -0 - -Closed Complaints - -Closed Inquiries - -2018 - -2,689 - -1,049 - -2019 - -2,643 - -878 - -2020 - -2,157 - -873 - -2021 - -2,205 - -795 - -7 +Major Challenge: Managing The Customer Experience ............................................................................................. 6 -5 - -4 - -3 - -2 - -1 +Byline: Delivery Imperatives To Compete With Amazon Without Breaking The Bank .......................... -0 +8 -Recoveries Per Year (In Millions) +Success Story: Uncommon Goods ............................................................................................................................................... -6.27 +10 -4.12 +Retailers Seek Innovative Returns Solutions ..................................................................................................................... -3.28 +11 -1.84 +Byline: Final Mile Strategies: Balancing Customer Satisfaction And Profitability........................................ -2018 +15 -2019 +Success Story: Kohl’s And Amazon ........................................................................................................................................... -2020 +16 -2021 +Retailers Turn To Technology And Disruptors .................................................................................................................. -Complaint Index Report +16 - Sixty-seven percent (67%) of total complaints concerned problems with claims handling. Other -common types of complaints involved policyholder services, underwriting and marketing and sales. +Success Story: Neiman Marcus ................................................................................................................................................. -67% - Claims Handling +19 -19% - Policyholder Service +Conclusion ................................................................................................................................................................................................ -10% - Underwriting +19 -4% - Marketing and Sales +About Retail TouchPoints ................................................................................................................................................................ -Companies may be listed on more than one table and will usually have a different complaint index -on each table. For example, if you are shopping for auto insurance, go to the table titled, “Indexes by -line: Automobile” to check for your company’s complaint index. Companies are listed in alphabetical -order. +20 -We list individual insurance companies only, not groups of insurance companies. For example, you -may be insured by a company such as State Farm. We show the name of the subsidiary company -owned by State Farm. Your policy or ID card lists the exact name of your company. +About Oracle .......................................................................................................................................................................................... -Why is my company not listed? +20 -If a company is not listed, that means it had fewer than 10 complaints during the calendar year and -it was not one of the top 20 premium writers for that line in Kansas. Information can also be compiled -for lines of insurance not listed in this booklet. For information on a company not listed or another line -of insurance, please call our Consumer Assistance Hotline at 800-432-2484. +About Convey ......................................................................................................................................................................................... -Just a snapshot +20 -The statistics in this report give you a “snapshot” of a company that represents complaints filed with the -Department regarding that company’s business in Kansas for one year and one type of coverage. As -indexes are likely to change, you may want to review these “snapshots” annually. + ARE YOU WINNING THE LAST MILE +OF RETAIL? -Complaint indexes do not tell you everything about a company. You may also want to consider the -following factors when shopping for insurance: +In retail, the “last mile” is becoming more crowded, more rushed, more expensive, and -Service — Some companies sell through local agents and some through direct marketing or group -plans. You may want to find out how the company services its policyholders. +an ever-more-critical contributor to a positive customer experience. -Licensing Status — Call the Consumer Assistance Hotline at 800-432-2484 to find out if a company -is licensed to do business in Kansas. +Over the past 18 months, there has been a 50% increase in last mile services, with +most of the demand coming from B2C e-Commerce, according to The Last Mile Retail +Study, 2018 conducted by eft Supply Chain and Logistics Business Intelligence. -Financial Stability — Financial stability helps ensure a company can pay its claims. The Department -enforces statutory requirements and monitors the financial stability of companies licensed and -operating in the state. +In addition to the growth of online sales, consumer demands are adding fuel to the -page 3 +competitive fire: -insurance.kansas.gov You can check a company’s financial rating by contacting one of these organizations: +• 65% of consumers want greater flexibility for deliveries; -MOODY’S INVESTOR SERVICES +• 61% want faster deliveries; and -STANDARD & POOR’S -INSURANCE RATING SERVICES +• 51% want real-time visibility into the status of their orders. -212-553-1653 +This inaugural Retail TouchPoints Last Mile Benchmark Report, based on a survey +of 146 retail executives, provides a snapshot of the challenges retailers face in -www.moodys.com +managing the last mile, including shipping and delivery; post-sale services such as setup -877-772-5436 +and installation; keeping customers informed; and the tricky (and expensive) area of -www.standardandpoors.com +product returns. -A.M. BEST CO. +There has been +a 50% increase +in last mile +services over the +past 18 months, +with most of +the demand +coming from B2C +e-Commerce. -www.ambest.com +— EFT LAST MILE +RETAIL STUDY 2018 -Cost — We publish easy-to-use booklets and shopper’s guides. Download them free from our website -at insurance.kansas.gov or call our office to get a free copy 800-432-2484. +3 -Free booklets & shopper’s guides -Auto Insurance Shopper’s Guide +ARE YOU WINNING THE LAST MILE OF RETAIL? + While 36% of +retailers deliver +less than 10% of +orders directly to +customers, 19% +deliver 25% to 50% +of orders direct to +consumers; +24% deliver 75% +of more. -Health Insurance Shopper’s Guide +The Last Mile survey indicates that delivery of products to customers’ homes or offices +already is widespread — and growing. While 36% of retailers deliver less than 10% +of orders directly to customers (versus shoppers taking them directly from a store), +19% deliver 25% to 50% of orders direct to customers, and 24% deliver 75% or more +of orders direct to customers. Presumably, though, this last group of respondents are -Home and Renters Insurance Shopper’s Guide +online-only retailers that lack brick-and-mortar stores as distribution points. -Important +What Percentage Of Products Are Delivered To +Shoppers’ Homes/Offices? +(Versus picked up/purchased in-store) -Life, Annuities and Long-Term Care Shopper’s -Guide +LESS THAN 10% -Medicare Supplement Shopper’s Guide +10% TO 25% -To better understand the complaint process, you might want to review what type of complaint was -filed against the company you are looking at. The National Association of Insurance Commissioners -(NAIC) provides information by code about the complaint. +25% TO 50% -Information is available at https://content.naic.org/cis_consumer_information.htm +51% TO 75% -Input the NAIC company code number found to the left of the company name in the complaint index -report and hit the enter key. Once the company is displayed, select “Click for Report Options.” Next, -select “Go to Complaint Code Report.” You will then choose a report based on the type of coverage, -reason for complaint, or disposition (final outcome). Finally, select Kansas under the “I want to see -complaints for:” drop down. Please note this report varies from the Kansas Insurance Department’s -Complaint Index as it only captures confirmed complaints, which are defined as: +MORE THAN 75% -“A complaint in which the state department of insurance determines: +24% -a) The insurer, licensee, producer, or other regulated entity committed any violation of: +36% -1) An applicable state insurance law or regulation; +19% -2) A federal requirement that the state department of insurance has the authority to enforce; or +14% -3) The term/condition of an insurance policy or certificate; or +7% -b) The complaint and entity’s response, considered together, indicate that the entity was in error.” +4 -page 4 +ARE YOU WINNING THE LAST MILE OF RETAIL? Retailers have +seen their volume +of direct deliveries +increase over the +past 18 months: +less than 10% said +their percentage +has decreased. -Complaint Index Report Indexes by line: Automobile +In line with the results of the eft study, Last Mile respondents have seen their volume +of direct deliveries increase over the past 18 months: 29% say the percentage has +increased significantly, and 39% say it has increased slightly. Less than 10% said the +percentage has decreased, and 23% say the volume has remained the same. -Cocode +How Has The Percentage Of Products You Deliver To +Shoppers Changed In The Last 18 Months? -Company Name +23% -33154 +29% -Alliance Indemnity Company +INCREASED SIGNIFICANTLY -42579 +INCREASED SLIGHTLY -Allied Property & Casualty Insurance Company +DECREASED SIGNIFICANTLY -29688 +DECREASED SLIGHTLY -Allstate Fire & Casualty Insurance Company +STAYED ABOUT THE SAME -19232 +5% -Allstate Insurance Company +4% -10386 +39% -American Family Insurance Company +Which Fulfillment/Delivery Services Do You Offer? +(Among retailers with both stores and e-Commerce sites) -19275 +BUY VIA MOBILE SITE/APP, DELIVER TO HOME -American Family Mutual Insurance Company +BUY ONLINE, PICK UP IN-STORE -13773 +BUY VIA MOBILE SITE/APP, PICK UP IN-STORE -Farm Bureau Property & Casualty Insurance Company +BUY ONLINE OR VIA MOBILE APP, RETURN IN-STORE -21628 +BUY VIA STORE ASSOCIATE’S MOBILE DEVICE, DELIVER TO HOME -Farmers Insurance Company Incorporated +BUY VIA IN-STORE KIOSK, DELIVER TO HOME -13587 +46% -First Chicago Insurance Company +44% -14138 +31% -GEICO Advantage Insurance Company +28% -14139 +21% -GEICO Choice Insurance Company +16% -12966 +5 -Key Insurance Company +ARE YOU WINNING THE LAST MILE OF RETAIL? MANAGING CUSTOMER EXPERIENCE AND +EXPECTATIONS TOPS RETAIL CHALLENGES -23787 +The volume of last-mile activity is not the only thing that’s on the rise. Customer +expectations around delivery times, service levels and visibility into the process +all are increasing rapidly. That’s reflected in the fact that two of the top three concerns +among retailers responding to the Last Mile survey are related to customers: Customer +expectations around delivery times is the leading challenge, at 49%, and customer +expectations of free/low-cost shipping is third, at 44%. -Nationwide Mutual Insurance Company +What Are Your Most Significant Last Mile Challenges? -16322 +Two of the top three +concerns among +retailers are related +to customers: +expectations around +delivery times at 49%, +and expectations +of free/low-cost +shipping at 44%. -Progressive Direct Insurance Company +49% -42919 +47% -Progressive Northwestern Insurance Company +44% -25405 +18% -Safe Auto Insurance Company +CUSTOMER EXPECTATIONS +AROUND DELIVERY TIMES -24740 +RISING SHIPPING +COSTS -Safeco Insurance Company Of America +CUSTOMER +EXPECTATIONS OF FREE/ +LOW-COST SHIPPING -23388 +HANDLING CUSTOMER +QUERIES/COMPLAINTS +EFFECTIVELY -Shelter Mutual Insurance Company +17% -19070 +14% -Standard Fire Insurance Company +12% -25178 +DELIVERY/FULFILLMENT +LOGISTICS -State Farm Mutual Automobile Insurance Company +RETURNS +MANAGEMENT -42749 +MANAGING POST- +PURCHASE SERVICE +PROVIDERS -Traders Insurance Company +6 -25941 +ARE YOU WINNING THE LAST MILE OF RETAIL? “Ultimately, the most important factor in the last mile is the experience of the end +customer (the consumer),” according to the eft Last Mile Retail Study. “Consumers have +the same expectations of service and engagement from the places they shop online as + +they do in-store, except they expect it faster, in line with the speed of the digital world.” + +Amazon is playing a large part in raising the bar of consumer expectations. The +Amazon Prime service effectively “hides” shipping costs in an annual fee, which + +increases pressures on competitors to offer low- or no-cost shipping. Amazon and +others have poured investments into offering two-day, next-day, same-day, and even +less than one-hour shipping to a wider swath of consumers, and it has had an impact +on the whole customer mindset. + +For example, the maximum number of days people are willing to wait for an item to be +delivered in exchange for free shipping has decreased, from 5.5 days in 2012 to 4.1 +in 2018, according to the 2018 AlixPartners Home-Delivery Shopping Survey. Amazon +Prime members are less patient than the population at large: only 19% of Prime +members would wait a week in exchange for free shipping, compared to 31% of non- +Prime members. + +Free shipping has become very much the rule, not the exception. A December 2017 +survey by L2 revealed that 77% of brands offer free shipping on their web sites. +Among this group, 35% do not require a minimum spending threshold to qualify for +free shipping. For those retailers that do require a minimum order, the median dollar +amount of $50 has not changed since 2016. + +The maximum +number of days +people are willing +to wait for an item +to be delivered +in exchange for +free shipping has +decreased from +5.5 days in 2012 to +4.1 days in 2018. + +— ALIXPARTNERS +HOME-DELIVERY +SHOPPING SURVEY -United Services Automobile Association +7 -25968 +ARE YOU WINNING THE LAST MILE OF RETAIL? + DELIVERY IMPERATIVES TO COMPETE WITH AMAZON +WITHOUT BREAKING THE BANK -USAA Casualty Insurance Company +By Rob Taylor, +CEO, Convey -18600 +As retailers fight to stay relevant in the face of Amazon, last mile delivery is being pushed front and center. While Prime -USAA General Indemnity Company +delivery has made the words “fast” and “free” synonymous with a seamless shipping experience, companies have had to -% of total market share represented in this report: +align their operational costs with quickly changing customer expectations, or risk customer loss — 70% will not return after a -2021 Market Share +poor delivery experience. -2021 Complaints +Consumers continue to report that they are unimpressed with delivery standards today — nearly 60% of shoppers say that -2021 Complaint -Index +the delivery experience is mediocre at best, and similarly, for those returning parcel shipments, 42% said that they were not -2020 Complaint -Index +satisfied with retailers’ return processes. -2019 Complaint -Index +This has driven pressures to create the perfect delivery experience, with 83% of supply chain executives saying that CX is -0.99% +now a companywide goal. While the delivery experience is now a standard KPI, many still struggle with legacy technology. -1.22% +By investing in visibility solutions, retailers can identify exceptions that derail delivery, resolve issues before they result -4.63% +in an angry customer and reset expectations to create a positive outcome. For retailers, this responsibility involves defining -0.28% +appropriate expectations through proactive communication, keeping delivery date promises and providing -4.27% +self-service options. -6.12% +Communication: Like any great relationship, the retailer-consumer bond needs to be fueled by communication and met +expectations. For the 93% of consumers who say they want to stay informed and receive proactive updates about their -6.79% +shipments, communication creates a sense of transparency. This visibility and communication helps them answer, “Where is -4.40% +my order?” -0.15% +According to our research, returns communication is equally important, as it was a source of anxiety for 70% of customers. -3.13% +Customers returning items indicated that the “Where is my money?” question was most pressing, and they did not want to -1.94% +deal with the retailer to get that information. -0.81% +Commitment: While customers care about “fast and free shipping,” over 90% indicated that missing a promise date comes +with consequences ranging from expected compensation to disloyalty. If a retailer misses a promise date, consumers -1.31% +expect a proactive response, and this expectation is amplified over the holidays. Over 57% of consumers said they would be -6.59% +impressed if their retailer either refunded shipping costs or expedited a new shipment in response to a late shipment. -8.15% +Control: When it is possible for the consumer to resolve the exception or an anxiety, 97.5% of customers will opt for +self-service options. This includes the ability to re-route an in-transit shipment or expedite a new delivery, with younger -0.22% +consumers also asking for self-pickup options such as lockers. In reverse logistics, consumer control gives the consumer the -1.84% +ability to schedule the return themselves. In either scenario, they seek convenience — they do not want to pick up the phone -3.24% +to call a carrier’s 1-800 number. -3.39% +Customer experience is becoming increasingly important in final mile delivery. Those who can embrace the rise of digital and -16.23% +bend to their customers’ demands can also increase their own flexibility for package delivery. Ultimately, these retailers will -1.18% +be able to create a memorable delivery experience and a loyal following without compromising their operations budget. -2.08% +8 ARE YOU WINNING THE -1.60% +LAST MILE OF RETAIL? -1.42% + CLEARING UP CONSUMERS’ VISIBILITY CHALLENGES -81.99% +Retailers are paying attention to shoppers’ last-mile demands. According to the eft +survey, 47% have implemented delivery feedback surveys, and 40% have increased last +mile visibility for customers. -2 +Respondents to the Last Mile survey use a variety of tools to keep customers informed +about the status of their deliveries: 86% rely on email notifications, along with online +tracking links (38%), text messages (33%) and phone calls (29%). Retailers’ self- +evaluation of their communication skills is generally positive. A majority, 54%, rate them +as “good,” and only 8% rate them as “poor.” -4 +47% of +retailers have +implemented +delivery feedback +surveys and 40% +have increased +last mile visibility +for customers. -42 +How Do You Inform Customers About Their Delivery Status? +(Check all that apply) -10 +— EFT LAST MILE +RETAIL STUDY 2018 -10 +38% -21 +33% -20 +29% -21 +86% -16 +EMAIL NOTIFICATIONS -23 +ONLINE TRACKING LINK -10 +TEXT MESSAGES -16 +PHONE CALLS -4 +How Would You Rate Your Last-Mile +Communications With Customers? -59 +8% -49 +14% -12 +EXCELLENT -9 +GOOD -17 +SATISFACTORY -11 +POOR -74 +54% -20 +24% 9 -7 - -5 - -0.33 - -0.53 - -1.48 - -5.86 - -0.38 - -0.56 - -0.48 - -0.78 - -17.66 - -1.20 - -0.84 - -3.22 - -0.50 - -1.46 - -0.98 - -8.87 - -0.80 - -0.86 - -0.53 - -0.75 - -2.76 - -0.71 - -0.71 - -0.58 - -0.17 - -0.93 - -1.06 - -0.55 - -0.86 - -0.82 - -0.31 - -0.60 - -6.40 +ARE YOU WINNING THE LAST MILE OF RETAIL? + SUCCESS STORY -1.27 +VISIBILITY SOLUTION BOOSTS ENGAGEMENT AND CTR +FOR UNCOMMON GOODS -1.79 +Uncommon Goods, a heavily gift-oriented online business devoted to sustainability and customer service, knew that +its customers expected visibility into where their packages were at any given time. The retailer launched the Engage -3.68 +product from Convey in November 2016 to provide communication and proactive exception management, and has -0.82 +used the platform since then to deliver exceptional customer experiences. -0.84 +When an order ships, customers receive an email with a tracking link that takes them to a mobile-optimized branded -0.70 +page on the Convey platform. Behind the scenes, Convey integrates and translates all carrier information, allowing -7.53 +customers to watch and understand what’s happening as the shipment progresses. Shoppers can sign up for text -0.86 +messages and exception emails; rate the experience and submit feedback to the Uncommon Goods team; click back -0.57 +to FAQs; and in some cases place additional orders via special promotions. -0.50 +Key results include: -0.84 +• 12% of shoppers subscribe to SMS notifications; -1.98 +• 54% of shoppers leave feedback before delivery; and -0.96 +• 6% of shoppers are clicking through to the Uncommon Goods site from the tracking page. -0.94 +Early knowledge of customer feedback and shipment issues such as delays, incorrect addresses or damaged shipments -1.48 +gives the retailer’s team more time to proactively correct problems. “Customers really appreciate being in the loop, and -- +alerted when there’s an unforeseen delay,” said MC Halfpenny, Director of Operations, Uncommon Goods. “Rather than -0.38 +causing them to be angry, often Uncommon Goods looks like the hero because we’re being proactive.” -1.04 +10 ARE YOU WINNING THE -- +LAST MILE OF RETAIL? -0.76 + RETAILERS SEEK NEW SOLUTIONS TO +PERENNIAL PROBLEM OF RETURNS -0.64 +One of the most challenging aspects of the Last Mile is that it’s a two-way street. -0.64 +Customers can (and do) return items, and the rise of digital commerce has raised the -1.29 +stakes — and the costs — in this critical area. -- +“Return rates for online retail are higher — sometimes four times higher — than for -0.77 +traditional brick-and-mortar retailers,” said Patrick Bohannon, Vice President, Retail -1.10 +Strategy for Oracle. “Return policy is very influential to online sales. Consumers want to -4.07 +have easy and cost-free returns, so many retailers are liberalizing their return rates.” -- +A Deloitte 2017 holiday shopping survey indicated that consumers’ top perks for online +shopping were free shipping, at 72%, and store policies for easy returns, at 44%. -1.10 +While the majority (61%) of retailers responding to the Last Mile survey are able to +keep their returns percentage below 10%, more than one in four, 26%, have a returns +percentage in the 10% to 20% range. -0.73 +Returns +percentages +appear to be +on the rise; +only 13% of +retailers reported +decreases over +the past 18 +months. -- +What Is Your Returns Percentage? -0.79 +3% -0.88 +10% -0.93 +61% -0.74 +26% -2.64 +LESS THAN 10% -0.55 +10% TO 20% -0.84 +21% TO 30% -0.39 +31% TO 40% -Indexes by line: Accident & Health +Returns percentages appear to be on the rise: 26% of retailers reported a slight +increase over the past 18 months and 5% reported a significant increase. Only +13% reported decreases. -Cocode - -Company Name - -12321 - -American Continental Insurance Company - -60380 - -American Family Life Assurance Company of Columbus - -61263 +11 -Bankers Life & Casualty Company +ARE YOU WINNING THE LAST MILE OF RETAIL? How Has Your Returns Percentage +Changed Over The Past 18 Months? -47171 +56% -Blue Cross Blue Shield of Kansas City +5% -70729 +26% -Blue Cross Blue Shield of Kansas Incorporated +7% -67369 +6% -Cigna Health & Life Insurance Company +INCREASED SIGNIFICANTLY -68500 +INCREASED SLIGHTLY -Continental Life Insurance Company Brentwood +DECREASED SIGNIFICANTLY -62324 +DECREASED SLIGHTLY -Freedom Life Insurance Company Of America +STAYED THE SAME -64211 +Cost of returns +top the list +of retailer +challenges at +48%, followed by +customer issues +with packing/ +shipping items for +return, at 39% -Guarantee Trust Life Insurance Company +RETURNS CREATE MULTIPLE COST CENTERS AND COMPLEXITIES -73288 +Retailers responding to the Last Mile survey identified returns costs as their biggest +challenge, at 48%, but in addition to this general concern, returns create a number of +other hurdles. -Humana Insurance Company +What Are Your Biggest Returns Challenges? -12459 Medica Insurance Company +COST OF RETURNS SHIPPING/HANDLING -71412 +CUSTOMER ISSUES WITH PACKING/SHIPPING ITEMS FOR RETURN -Mutual Of Omaha Insurance Company +IMPACT ON BOTTOM-LINE REVENUE OF RE-SELLING RETURNED ITEMS -82538 +COMMUNICATING EFFECTIVELY WITH SHOPPERS ABOUT RETURNS +RULES/PROCESSES -National Health Insurance Company +COST/LOGISTICS OF CREDITING SHOPPERS’ ACCOUNTS/ +PROVIDING REFUNDS -67091 +MINIMIZING FRAUDULENT RETURNS -Northwestern Mutual Life Insurance Company +MANAGING IN-STORE STAFF REQUIREMENTS FOR RETURNS PROCESSING -67326 +48% -Old Surety Life Insurance Company +39% -15777 +28% -Oscar Insurance Company +23% -68462 +23% -Reserve National Insurance Company +22% -14345 +18% -Sunflower State Health Plan Incorporated +12 -56014 +ARE YOU WINNING THE LAST MILE OF RETAIL? “Retailers make +meticulous plans +down to the +square inch about +what’s carried in +their stores, but +those plans can +be upset by a bin +of onesie-twosie +returns.” -Thrivent Financial For Lutherans +— DAVID SOBIE, +HAPPY RETURNS -86231 +Returns can play havoc with basic retailing functions, including planning, inventory, -Transamerica Life Insurance Company +pricing and in-store assortments. “Many retailers do not plan for returned inventory as -72850 +an available ‘pot’ of stock, meaning they overbuy,” said Justine Shiou, Solutions Manager, +Planning & Optimization at Oracle. Additionally, “returns are generally made weeks after -United World Life Insurance Company +the initial purchase which reduces the time to re-sell the item at full price. This can be a -96385 +significant issue for fashion/short lifecycle retailers that turn quickly. The selling price is -UnitedHealthcare of Midwest Inc +likely to have already been marked down versus the price when it was purchased.” -70319 Washington National Insurance Company +David Sobie, Co-Founder and CEO of Happy Returns, which operates “return bars” -2021 Market Share +where shoppers can bring back items they have bought online, identified other returns- -2021 Complaints +related issues retailers grapple with. “Broadly speaking, the challenge for retailers is -2021 Complaint -Index +‘How do I get this inventory back and ready to be restocked in a way that doesn’t break -2020 Complaint -Index +the bank?’” said Sobie. -2019 Complaint -Index +Costs include shipping the item back to the distribution center or store, but that’s only -0.61% +the beginning, said Sobie. Someone needs to assess whether the item can be added -3.04% +back into the retailer’s inventory. In some cases this is a simple matter of relabeling -0.79% +and repackaging an undamaged item, but if the item is damaged in some way, -3.55% +refurbishment might be needed. The retailer might need to take a loss or sell the item -22.36% +to a downmarket channel. -3.01% +ONLINE RETURNS UPSET STORE ASSORTMENT PLANS -0.57% +Omnichannel retailers have their own set of returns challenges. A retailer’s online -0.99% +assortment may be much larger than that of an individual brick-and-mortar store, but -0.68% +if shoppers return an item bought online, it needs to be incorporated into the store’s -3.82% +assortment — or shipped to another store or DC at additional cost. -2.12% +“Retailers make meticulous plans down to the square inch about what’s carried in -1.05% +their stores, but those plans can be upset by a bin of onesie-twosie returns,” said -0.53% +Sobie. “That carefully planned store assortment can make the store look like more of a -0.73% +discount retailer.” -0.69% +13 -0.58% +ARE YOU WINNING THE LAST MILE OF RETAIL? + POOR RETURNS EXPERIENCES CREATE NEGATIVE CX -0.56% +Retailers seeking to keep returns costs down may decide to choose slower, less -17.11% +expensive shipping options, but they may end up paying higher costs in other -0.56% +areas. “The longer the returns process takes, the more customer service contacts -0.69% +will be generated,” Sobie warned. “The number-one reason people contact customer -0.75% +service is to check on their order, but the number-two reason relates to returns. These -2.36% +are not insignificant costs, either; they can run to a couple of bucks per contact at a -0.79% +well-run retailer.” -0 +Returns create headaches for consumers as well, meaning that if retailers don’t handle -8 +the process well, they risk losing dissatisfied customers. A 2017 study conducted by -1 +Forrester for Happy Returns, which surveyed online fashion shoppers, revealed that +returns are the worst part of online shopping for 73% of consumers. Among those +that have to return items, 39% say they do not shop at the retailer again. -5 +Returns are the +worst part of +online shopping +for 73% of +consumers. +Among those that +have to return +items, 39% don’t +shop that retailer +again. -46 +— FORRESTER 14 -2 - -4 - -2 - -5 - -16 - -1 - -8 - -0 - -0 - -6 - -1 - -67 - -0 - -1 - -0 - -0 +ARE YOU WINNING THE LAST MILE OF RETAIL? + FINAL MILE STRATEGIES: BALANCING CUSTOMER +SATISFACTION AND PROFITABILITY -5 - -0.00 - -1.02 - -0.49 - -0.55 - -0.80 - -1.80 - -1.37 - -1.57 - -1.14 - -0.51 - -2.93 - -0.37 - -5.91 - -0.00 - -0.00 +By Justine Shiou, Solutions Manager, +Planning and Optimization, Oracle -4.03 +ONLINE RETURNS AND INVENTORY SHUFFLING -0.70 +A sale is not a sale until the customer decides to keep it. Return rates have continued to increase across all channels but -1.52 +online sales have a higher return rate (more than 3X). Despite the operational impacts, retailers are liberalizing return -0.00 +policies: free returns, return in-store, return without limits basically, to reduce friction. This strategy has significantly -0.56 +increased operational cost. An online return to the warehouse includes shipping costs and restocking fees, an online back- -0.00 +to-store return also includes restocking fees but brings another layer of inefficiency, requiring the associate to handle the +item, run it back onto the floor, ultimately distracting him/her from more productive customer engagements. -0.00 +Markdown rates are also climbing exponentially, due to incongruent on-hand store inventory and delayed returns. Returns -2.44 +are generally made weeks after the initial purchase, which means reduced time to resell at full price, particularly for fashion -0.00 +retailers. The effectiveness of localized assortment strategies is degraded due to buy online, return in store (BORIS). The -1.38 +BORIS journey creates incongruent inventory and cluttered sales floors. Retailers are dealing with this in two ways: -0.00 +• Treatment 1: absorb the cost of taking one-off markdowns to clear the stock -1.06 +• Treatment 2: absorb the cost of transferring items back to the online warehouse or to another store with the -0.70 +item in its assortment, a.k.a. inventory shuffling -0.69 +While the practice of inventory shuffling [using transfers] used to be deemed costly, some retailers have decided this -0.00 +treatment is more sales effective than creating markdowns. Neither approach is optimal. -2.46 +TRADE REGIONS: MEASURING TRADEOFF BETWEEN CX AND GROSS MARGIN -0.69 +A focus on total trade region (online and stores combined) when developing omnichannel strategies and the role of the -0.24 +store is critical to long-term success. Retailers need to consider all roles, including fulfillment, pickup, return, traditional -3.42 +sales or a showroom concept. Advanced analysis on departments and items that have high conversion rates need to be -1.37 +conducted to save sales and create new revenue opportunities. Integrating business insights, like customer loyalty, with -9.04 +POS or order system data will more effectively drive planning by customer and fulfilment channel. With improved weekly -0.00 +sales, stock and intake accuracy, and in-season projections by channel, retailers will be able to increase gross margin, and -0.79 +reduce markdown liabilities. The ability to track the effects of cross-channel returns can help limit under- or over-stock -2.37 +situations, plan return percentages by channel and customer journey and plan cross-channel sales fulfilment paths. -2.42 +Retailers intrinsically are bound by the bottom line and shareholder interest. However, the retail environment has -1.51 +changed significantly while the mindset around profitability has not. Including a customer experience measure (such as -0.00 +NPS) in the planning process can also help retailers understand a more holistic view of their business and ensure the -4.73 +plan’s relevance to both gross margin and customer service (t-NPS). -0.00 +15 ARE YOU WINNING THE -0.00 +LAST MILE OF RETAIL? -5.72 + SUCCESS STORY -0.00 +KOHL’S GETS TRAFFIC BUMP FROM AMAZON RETURNS -0.58 +Returns aren’t all bad news. Savvy retailers with well-trained staffs have long been able to turn the return occasion into -0.34 +an opportunity to repair a damaged customer relationship, and in some cases even to take advantage of the consumer’s -1.31 +presence to sell them additional items. -0.76 +Since October 2017, Amazon shoppers have been able to return items at 82 Kohl’s stores. It seems benefits are flowing +to both parties: Amazon gets a physical location for returns, improving customer service and potentially cutting costs, -0.69 +and Kohl’s has seen increased foot traffic in the participating stores. -- +A study by UberMedia found that stores with Amazon Return Centers saw their daily visitors increase 48% after October +2017, compared to a 36% increase in non-Amazon Return center locations. The 12-point difference was over and above +the normal holiday traffic increase to be expected in Q4. -- +RETAILERS TURN TO TECHNOLOGY, +THIRD-PARTY SERVICES AND DISRUPTORS -2.55 +Retailers are tackling their last-mile challenges with a mix of traditional technologies and -0.64 +cutting-edge solutions. Additionally, as might be expected in an area that’s often far -3.26 +from their core competencies, retailers rely heavily on third-party services to fulfill many -0.91 +last-mile chores. -- +More than half, 59%, of retailers deploy courier/delivery services, and 28% drop +ship items directly from suppliers. For services that involve direct customer contact, +however, most retailers prefer to keep these functions in-house: only 14% outsource +customer service call centers, and just 12% rely on others for product delivery and +setup; 22% do not outsource any of their last-mile services. -- +Retailers’ choices for technological aids in this area generally run to the tried and -- +true: transportation management solutions and route optimization systems top the +current usage list, at 35% and 27% respectively. The percentages using more advanced +solutions, such as RFID on shipping containers and products/packaging, are in the -- +teens. IoT sensors, AI/machine learning, and the use of land-based robots or airborne -- +drones are even less common. Perhaps in an area as high-stakes as the last mile, -1.72 +retailers are waiting for the technology to be perfected before taking a chance on it. -- +For services that +involve direct +customer contact, +most retailers +prefer to keep +functions in- +house: 22% do +not outsource any +of their last-mile +services. -- - -0.00 - -0.00 - -3.27 - -% of total market share represented in this report: - -67.92% - -page 5 - -insurance.kansas.gov Indexes by line: Homeowners - -Cocode - -Company Name +16 -37907 +ARE YOU WINNING THE LAST MILE OF RETAIL? Which Last Mile Services Do You Outsource To Third Parties? +(Check all that apply) -Allstate Vehicle & Property Insurance Company +59% -19690 +COURIER/DELIVERY SERVICES -American Economy Insurance Company +28% -10386 +DROP SHIP ITEMS DIRECTLY FROM SUPPLIERS -American Family Insurance Company +14% -19275 +12% -American Family Mutual Insurance Company +10% -10872 +22% -American Strategic Insurance Corporation +CUSTOMER SERVICE/CALL CENTER -13773 +PRODUCT DELIVERY/SETUP -Farm Bureau Property & Casualty Insurance Company +RETURNS MANAGEMENT -21652 +WE DON’T OUTSOURCE ANY LAST MILE SERVICES -Farmers Insurance Exchange +Which Technologies Do You Use +To Manage Last-Mile Activities? -26298 +USE NOW -Farmers Property & Casualty Insurance Company +PLAN TO ADOPT +WITHIN 18 MONTHS -20419 +NO PLANS TO ADOPT -Homesite Indemnity Company +TRANSPORTATION MANAGEMENT SOLUTION -14451 Marysville Mutual Insurance Company +ROUTE OPTIMIZATION SYSTEM -25453 +VEHICLE MONITORING TECHNOLOGY -Nationwide Insurance Company Of America +RFID ON SHIPPING CONTAINERS -23787 +MOBILE DEVICE MONITORING OF DELIVERY PEOPLE -Nationwide Mutual Insurance Company +RFID ON PRODUCTS/PACKAGING -24740 +IOT SENSORS -Safeco Insurance Company Of America +AI/MACHINE LEARNING -23388 +LAND-BASED ROBOTIC DELIVERIES -Shelter Mutual Insurance Company +DRONE DELIVERIES -25143 +35% -State Farm Fire & Casualty Company +27% -27998 +18% -Travelers Home & Marine Insurance Company +16% -38130 +14% -Travelers Personal Insurance Company +14% -25941 +10% -United Services Automobile Association +7% -25968 +6% -USAA Casualty Insurance Company +2% -18600 +21% -USAA General Indemnity Company +27% -% of total market share represented in this report: +24% -2021 Market Share +24% -2021 Complaints +31% -2021 Complaint -Index +28% -2020 Complaint -Index +26% -2019 Complaint -Index +33% -4.36% +12% -1.47% +14% -4.98% +44% -8.70% +46% -1.35% +58% -6.25% +60% -4.85% +55% -1.22% +58% -1.41% +64% -1.20% +60% -1.70% +82% -2.10% +84% -1.70% - -2.48% - -21.08% - -3.42% - -4.39% - -2.61% - -1.39% - -1.23% - -77.89% - -21 - -2 - -13 - -14 - -6 - -13 - -7 - -3 - -5 - -4 - -3 - -1 - -7 - -5 - -78 - -9 - -10 - -2 - -3 - -6 - -1.63 - -0.46 - -0.89 - -0.55 - -1.51 - -0.70 - -0.49 - -0.83 - -1.20 - -1.13 - -0.60 - -0.16 - -1.39 - -0.68 - -1.25 - -0.89 - -0.77 - -0.26 - -0.73 - -1.65 - -0.86 - -- - -0.42 - -0.66 - -0.25 - -0.80 - -0.77 - -1.39 - -- - -1.14 - -- - -0.00 - -0.88 - -1.15 - -1.08 - -0.82 - -0.66 - -0.41 - -0.53 - -2.65 - -1.64 - -- - -1.10 - -0.75 - -0.86 - -0.81 - -1.36 - -0.00 - -- - -2.02 - -- - -- - -0.32 - -1.66 - -0.46 - -0.51 - -0.72 - -0.58 - -0.56 - -0.98 - -Indexes by line: Individual Annuity - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -Cocode - -Company Name - -90611 - -Allianz Life Insurance Company Of North America - -60488 - -American General Life Insurance Company - -61689 - -Athene Annuity & Life Company - -87726 - -Brighthouse Life Insurance Company - -62944 - -Equitable Financial Life Insurance Company - -63088 - -Farm Bureau Life Insurance Company - -63274 - -Fidelity & Guaranty Life Insurance Company - -91642 - -Forethought Life Insurance Company - -65056 - -Jackson National Life Insurance Company - -65676 - -Lincoln National Life Insurance Company - -66869 Nationwide Life Insurance Company - -91596 - -New York Life Insurance & Annuity Corporation - -66974 - -North American Company for Life & Health Insurance - -67466 - -Pacific Life Insurance Company - -86630 - -Prudential Annuity Life Assurance Corporation - -65005 - -RiverSource Life Insurance Company - -68675 - -Security Benefit Life Insurance Company - -62952 - -SILAC Insurance Company - -69345 - -Teachers Insurance & Annuity Association Of America - -56014 - -Thrivent Financial For Lutherans - -5.48% - -3.55% - -4.46% - -3.15% - -3.55% - -2.37% - -2.27% - -2.54% - -8.01% - -5.60% - -2.74% - -3.51% - -2.43% - -1.85% - -1.86% - -2.94% - -2.31% - -5.21% - -3.08% - -1.70% - -2 - -1 - -0 - -0 - -1 - -0 - -0 - -1 - -2 - -1 - -0 - -0 - -0 - -0 - -0 - -0 - -1 - -1 - -0 - -0 - -1.18 - -0.91 - -0.00 - -0.00 - -0.91 - -0.00 - -0.00 - -1.27 - -0.81 - -0.58 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -1.40 - -0.62 - -0.00 - -0.00 - -0.77 - -2.35 - -0.00 - -2.24 - -1.34 - -7.38 - -0.00 - -1.23 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -0.00 - -- - -0.00 - -0.53 - -0.00 - -0.00 - -- - -% of total market share represented in this report: - -68.62% - -1.15 - -1.72 - -1.93 - -- - -2.29 - -- - -1.12 - -0.77 - -- - -- - -- - -1.03 - -0.69 - -- - -- - -- - -0.84 - -- - -- - -- - -page 6 - -Complaint Index Report Indexes by line: Individual Life - -Cocode - -Company Name - -60399 - -American Family Life Insurance Company - -60488 - -American General Life Insurance Company - -63088 - -Farm Bureau Life Insurance Company - -65838 - -John Hancock Life Insurance Company USA - -58033 - -Knights Of Columbus - -65676 - -Lincoln National Life Insurance Company - -65935 Massachusetts Mutual Life Insurance Company - -66044 Midland National Life Insurance Company - -91596 - -New York Life Insurance & Annuity Corporation - -66915 - -New York Life Insurance Company - -67091 - -Northwestern Mutual Life Insurance Company - -67466 - -Pacific Life Insurance Company - -67644 - -Penn Mutual Life Insurance Company - -65919 - -Primerica Life Insurance Company - -68136 - -Protective Life Insurance Company - -79227 - -Pruco Life Insurance Company - -68241 - -Prudential Insurance Company Of America - -69108 - -State Farm Life Insurance Company - -56014 - -Thrivent Financial For Lutherans - -86231 - -Transamerica Life Insurance Company - -69868 - -United Of Omaha Life Insurance Company - -% of total market share represented in this report: - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -1.85% - -1.92% - -3.91% - -1.79% - -2.56% - -2.90% - -4.05% - -3.84% - -2.32% - -4.30% - -11.52% - -2.51% - -1.33% - -1.34% - -2.19% - -3.93% - -0.58% - -4.10% - -1.80% - -1.93% - -1.18% - -61.84% - -0 - -21 - -8 - -1 - -3 - -3 - -0 - -0 - -0 - -3 - -1 - -0 - -1 - -0 - -1 - -0 - -12 - -3 - -0 - -11 - -1 - -0.00 - -6.13 - -1.15 - -0.31 - -0.66 - -0.58 - -0.00 - -0.00 - -0.00 - -0.39 - -0.05 - -0.00 - -0.42 - -0.00 - -0.26 - -0.00 - -11.66 - -0.41 - -0.00 - -3.20 - -0.47 - -0.58 - -2.19 - -0.55 - -0.50 - -1.01 - -0.31 - -0.25 - -- - -0.00 - -0.38 - -0.24 - -0.00 - -0.00 - -1.58 - -2.09 - -0.14 - -- - -0.52 - -0.28 - -1.61 - -- - -0.49 - -3.76 - -0.34 - -0.00 - -0.35 - -0.27 - -0.13 - -- - -0.00 - -0.21 - -0.00 - -0.00 - -0.00 - -0.33 - -0.78 - -0.26 - -- - -0.21 - -0.25 - -2.69 - -- - -Indexes by line: Long-Term Care - -Cocode - -Company Name - -90611 - -Allianz Life Insurance Company Of North America - -61263 - -Bankers Life & Casualty Company - -70729 - -Blue Cross Blue Shield of Kansas Incorporated - -20443 - -Continental Casualty Company - -71404 - -Continental General Insurance Company - -70025 - -Genworth Life Insurance Company - -64211 - -Guarantee Trust Life Insurance Company - -65838 - -John Hancock Life Insurance Company USA - -58033 - -Knights Of Columbus - -65935 Massachusetts Mutual Life Insurance Company - -65978 Metropolitan Life Insurance Company - -71412 - -Mutual Of Omaha Insurance Company - -66915 - -New York Life Insurance Company - -69000 Northwestern Long Term Care Insurance Company - -68241 - -Prudential Insurance Company Of America - -65005 - -RiverSource Life Insurance Company - -25178 - -State Farm Mutual Auto Insurance Company - -56014 - -Thrivent Financial For Lutherans - -86231 - -Transamerica Life Insurance Company - -62235 - -Unum Life Insurance Company Of America - -2021 Market Share - -2021 Complaints - -2021 Complaint -Index - -2020 Complaint -Index - -2019 Complaint -Index - -2.16% - -9.45% - -2.07% - -3.02% - -1.66% - -17.08% - -3.41% - -9.49% - -3.07% - -1.40% - -4.92% - -4.61% - -3.06% - -6.03% - -2.44% - -1.47% - -3.00% - -2.38% - -4.61% - -3.00% - -9 - -6 - -2 - -4 - -0 - -15 - -1 - -1 - -0 - -0 - -1 - -0 - -3 - -0 - -3 - -0 - -0 - -0 - -7 - -2 - -5.63 - -0.86 - -1.30 - -1.79 - -0.00 - -1.19 - -0.40 - -0.14 - -0.00 - -0.00 - -0.27 - -0.00 - -1.33 - -0.00 - -1.66 - -0.00 - -0.00 - -0.00 - -2.05 - -0.90 - -2.42 - -0.64 - -0.34 - -3.35 - -1.35 +17 -1.24 +ARE YOU WINNING THE LAST MILE OF RETAIL? One area where at least a few retailers are willing to try something new is with -0.00 +disruptive shippers, such as Uber, Deliv, Instacart and Shipt. The December 2017 -0.00 +Target acquisition of Shipt for $550 million, and high-profile Instacart partnerships with -0.00 +retailers including BJ’s Wholesale Club and Albertsons, may signal to other brands that -- +such services are “retail-ready.” -1.86 +Which Non-Traditional Shippers Have You Used +During The Past 12 Months? +(Check all that apply) -0.00 +Top retailer data +uses are focused +on functional +areas, but 42% +are using returns +data to adjust +their product +selections. -0.00 +21% -0.00 +8% -1.07 +7% -0.00 +11% -0.00 +UBER -0.00 +DELIV -3.21 +INSTACART -0.00 +SHIPT -1.78 +59% -0.76 +OTHER E.G. ANDREANI, ENJOY, SPEEDEE, +GOOGLE EXPRESS, SKIP THE DISHES -0.82 +Given the growing volume of last mile activity, retailers should look to step up their data -0.81 +collection and analysis efforts in order to gain a strategic advantage. At this point, top -0.73 +data uses are focused on functional areas such as resolving customer service issues, +at 48%, and tracking the speed/efficiency of drivers, at 46%. A significant minority of +respondents, 42%, are using returns data to adjust their product selections, a smart +move that could potentially reduce future returns rates. -0.32 +What Are Your Most Valuable Uses For Last Mile Data? -- +48% -0.14 +46% -- +42% -- +22% -1.15 +RESOLVE CUSTOMER +SERVICE ISSUES -0.94 +TRACK SPEED/ +EFFICIENCY OF DRIVERS +OR DELIVERY SERVICES -- +USE RETURNS DATA +TO ALTER PRODUCT +SELECTION -0.23 +DETERMINE MORE +EFFICIENT DELIVERY +ROUTES -- +17% -- +13% -0.81 +12% -- +CHANGE SHIPPING OFFERS +(I.E. DISCOUNTS FOR +DELIVERING MULTIPLE +ITEMS IN ONE BOX) -6.39 +CHANGE INTERNAL +FULFILLMENT RULES +(E.G. SET MINIMUM TRUCK +CAPACITIES AT 75%) -- +IDENTIFY LOCATIONS FOR +NEW STORES/ +E-COMMERCE ACTIVITIES -% of total market share represented in this report: +18 -88.33% +ARE YOU WINNING THE LAST MILE OF RETAIL? SUCCESS STORY -page 7 +NEIMAN MARCUS BOOSTS SPEED, SERVICE LEVELS +WITH PROACTIVE SOLUTION -insurance.kansas.gov Filing a Complaint -The Kansas Insurance Department’s Consumer Assistance Division provides free help to consumers. -Trained representatives can answer questions about your coverage or assist when you have a dispute -with your insurance company. +Last-mile challenges are multiplied when retailers are delivering large, bulky items such as furniture and appliances. At -How to file a complaint +least one in every 10 of these “white glove” deliveries faces a problem of some kind, such as an incorrect address or +phone number or a scheduling conflict with the carrier, according to the Wall Street Journal. -If you have tried unsuccessfully to resolve a problem with your insurance company or agent, we -encourage you to contact the insurance department. You may file a complaint online through our -website, by mail or email. We also accept complaints in-person. +When the retailer is Neiman Marcus, known for its high levels of in-store customer service, there’s lots of pressure to +get these deliveries right as often as possible. The supply chain team at Neiman Marcus collaborated with the customer -If you do need to file a formal complaint, we require the following information: +care group, using a Convey solution to track the most complex deliveries. -• Your name +The software captures problems with enough time to correct them before the scheduled delivery date. Within 30 days of +launching the solution, Neiman Marcus had reduced transit times to customer by 24%. -• Your address +CONCLUSION -• Your email address (if applicable) +Retailers and their last-mile partners are racing to keep up with customer expectations -• The policy number +that are only moving in one direction: higher. To make things even more challenging, -• A clear, concise written explanation of your -complaint — this can be sent by postal mail, -fax, or filed through our website +customers want fast, accurate deliveries, hassle-free returns, and enhanced visibility -• Your daytime telephone number +into these processes — all for little or no cost. -• Copies of supporting letters, police reports, +“The biggest challenge in the last mile is matching increasing cost against spiraling +consumer expectations,” said Paul Hamblin, Editor in Chief, Logistics Business magazine, +as quoted in the eft Last Mile Retail Study. “Next-day is already almost standard, same- +day is hot on its heels. Who’s going to pay for all this?” -• Type of insurance involved (automobile, +“The biggest +challenge in +the last mile +is matching +increasing cost +against spiraling +consumer +expectations.” -homeowners, etc.) +Technology and innovation are providing some of the answers, as retailers seek the -• The name of the insurance company, agent +most optimal transportation routes and try out new business models for shipping and -or adjuster +returns. But retailers will need to dig deeper in order to meet their last-mile challenges. -notes, etc. +— PAUL HAMBLIN, +LOGISTICS BUSINESS -• Tell us what has been done, including who -you’ve talked to and what you were told -(names, dates, times, places, etc.) +They need to gather insights about price sensitivity and elasticity in the last mile, so -• Desired outcome +that they can begin recouping at least some of their costs without alienating customers. -Do not send originals of any of your records, and keep a copy of the letter you send us. +They should invest in back-end systems that allow them to make firm delivery date -Contact the Kansas Insurance Department +promises, and to proactively prevent missteps that can damage the customer +relationship. Finally, retailers may want to rethink whether every order needs to be +delivered within 12, 24 or 48 hours. If consumers are confident that their order will -WEBSITE +arrive when the retailer has said it will, they might be more willing to wait a few days. -insurance.kansas.gov +19 -SOCIAL +ARE YOU WINNING THE LAST MILE OF RETAIL? + To Learn More... -KSinsurancedept +Retail TouchPoints is an online publishing network for retail executives, with content focused on +optimizing the customer experience across all channels. The Retail TouchPoints network is comprised of +three weekly e-newsletters, special reports, web seminars, exclusive benchmark research, an insightful -EMAIL +editorial blog, and a content-rich web site featuring daily news updates and multi-media interviews at +www.retailtouchpoints.com. The Retail TouchPoints team also interacts with social media communities +via Facebook, Twitter and LinkedIn. -kid.commissioner@ks.gov -kid.webcomplaints@ks.gov +P 201.257.8528 +info@retailtouchpoints.com -ADDRESS +Oracle provides retailers with a complete, open, and integrated suite of best-of-breed business -1300 SW Arrowhead Rd. -Topeka, KS 66604 +applications, cloud services, and hardware that are engineered to work together and empower -PHONE +commerce. Leading fashion, grocery, and specialty retailers use Oracle solutions to anticipate market -Main: 785-296-3071 -Fax: 785-296-7805 +changes, simplify operations and inspire authentic brand interactions. For more information, visit our +website at www.oracle.com/retail. -CONSUMER ASSISTANCE -HOTLINE +P 1.650.506.7000 +oneretailvoice_ww@oracle.com -800-432-2484 +Convey enables businesses worldwide to take control of shipping and delivery. Our cloud-based -page 8 +platform connects disparate data and processes, giving shippers the insight to make smart decisions, -Complaint Index Report 1300 SW ARROWHEAD ROAD -TOPEKA, KS 66604 -1-800-432-2484 -(785) 296-3071 +and the tools to take action. From parcel to freight, and first to final mile, Convey helps the world’s -INSURANCE.KANSAS.GOV -KID.COMMISSIONER@KS.GOV -(785) 296-7805 (FAX) -1-877-235-3151 (TTY/TDD) +largest brands create better delivery outcomes while reducing costs, improving visibility and -Vicki Schmidt, Commissioner -insurance.kansas.gov +transforming transportation into their next competitive advantage. Founded in 2013, Convey is based +in Austin, Texas. Learn more at: www.getconvey.com. -page 9 +P 602.842.2788 +info@getconvey.com -insurance.kansas.gov \ No newline at end of file diff --git a/q117/random_k4/random_2.md b/q117/random_k4/random_2.md index 5aa602e7fdd86b41f160b8fb2430ecc9847ae72b..d1447fd886d0d69697641e94fcf1890d83769afd 100644 --- a/q117/random_k4/random_2.md +++ b/q117/random_k4/random_2.md @@ -1,128 +1,1679 @@ -Competitive Analysis +Hobart and William SmitH CollegeS -Kit Needham +Handbook for +International Students - Any way the customer -is solving the problem now is -competition +intercultural affairs Center +288 Pulteney Street +geneva, nY 14456 +(315) 781-3319 – Phone +(315) 781-3888 – Fax - DIRECT vs INDIRECT + TABLE OF CONTENTS -DIRECT – similar products and services +SECTION I: General Information + Welcome -INDIRECT – solving same problem for same -target market with a different product or -service +important Campus Phone numbers - Who are Uber’s competitors? + office locations -Direct? +banks and theaters +Places to Stay +Frequently asked Questions -Indirect? +SECTION II: Immigration Procedures + How to get assistance with immigration matters +immigration basics and definition of terms - Competitive Factors + obtaining a Visa and Form i-20 -What makes a customer choose one -solution over another? +differences between F-1 and J-1 Status -Price – cheaper -Service – faster, personalized, convenient -Quality - lasts longer, stylish, tastes better + arriving at a U.S. port of entry… What a student can expect -At most, you can only compete -on 2 out of the 3 +dealing with the department of Homeland Security (dHS) +exceptions to a Full Course of Study +duration of Status for F-1 Students +temporary departure from the US - Unfair Advantage + accepting U.S. government Public assistance or Public benefits + Work opportunities for students with F-1 Status + Work opportunities for students in J-1 Status -Core competency that cannot be copied or -bought -- Deep domain expertise (hospital systems) -- One amazing hard thing (Google algorithm) -- Authority (existing reputation in market) -- Dream Team (previous startup success; rock +Practical training + optional Practical training (oPt) + Curricular Practical training (CPt) +departure Checklist -star) +page 1 +9 +10 +10 +10 +12 - Templates for Analysis +15 +16 +17 +19 +21 +22 +24 +24 +24 +25 +26 +26 +27 +27 +28 +29 - Most Common Format +SECTION III: -You +Sample Fall orientation Schedule for international Students -Competitor 1 +30 -Competitor 2 +2 -Competitor 3 + SECTION I: +GENERAL INFORMATION -Competitor 4 +Welcome! -Feature 1 +Welcome to Hobart and William Smith Colleges (the Colleges or HWS). to you, the +decision to study abroad might have been a small one – or, indeed, a rather big step. +either way you won’t know what your experience will be like until you are here. this +handbook is designed to help you prepare for the adventure of studying in the United +States. in our experience the better prepared you are the more you will benefit from +your time at the Colleges. + While we are trying to answer many of your questions prior to your arrival in the +United States, some of your questions will only arise once you are here. the inter- +national Student adviser will serve as a resource to you and your parents or guardians +throughout the application process and during your stay at Hobart and William Smith +Colleges. The adviser will be your primary contact for any questions or concerns +regarding visas and immigration issues, employment, travel and a host of other + matters that may arise. -Feature 2 +the International Student Office (iSo) is located on the second floor of the -Feature 3 +intercultural affairs Center at 288 Pulteney Street. this office advocates for and ad- +dresses the needs of international students as well as provides programs and services +related to immigration matters. during the academic school year the iSo sponsors +and co-sponsors events designed to promote cross-cultural sharing between the many +groups at the Colleges and +geneva communities. +Currently, international +students come to us from 22 +nations: austria, bulgaria, +Canada, China, Colombia, +Czech republic, germany, +india, iran, nepal, Pakistan, +Portugal, Scotland, South +africa, South Korea, Swe- +den, Switzerland, thailand, +the netherlands, turkey, +Ukraine and Vietnam. -Feature 4 +Intercultural Affairs Center – Home to the International Student Office. -Feature 5 +in the fall international students attend a special two-day orientation before +the regular first year orientation begins. at Hobart and William Smith, we ask that +incoming international students join us ahead of time as it is a wonderful opportunity +to get acquainted with the new culture, get assimilated with the campus, and provide +a forum to answer your questions before the rush of orientation activities begins. at +the iSo, we will check to make sure that all your immigration paperwork is in order as +well as help you gain a better understanding of what your F-1 or J-1 status means and +how to maintain it; help to answer any questions that you might have about adjusting +to life as a student in the United States, specifically life at HWS; and meet students, +staff and faculty members. - Examples of Features +3 -• Price -• Benefits -• Quality -• Durability -• Image/style -• Service -• Warranties -• Convenience + Questions? Please +contact the International +Student Adviser by phone + at (315) 781-3319. The + International Student + Office is located on the + 2nd floor of the +Intercultural Affairs Cent- +ter, 288 Pulteney Street, +Geneva, NY 14456. (315) +781-3319 – Phone (315) +781-3888-Fax. Office +hours are 8:30 am to +5:00 pm weekdays during +the academic year. -• Ease of Use -• # of features -• Type of features -• Wow factor -• Location(s) -• Distribution/Sales -• Certifications -• Endorsements +THINGS TO DO BEFORE GOING ABROAD +before you can board a plane and start your study +abroad adventure, there are some formalities you will +need to take care of. the earlier you start thinking +about the things you need, the smoother your prepara- +tions will go. this list will help you consider some of +the things you will have to do. - Where do you find info +Obtaining a Visa +Upon receipt of the form i-20 (for F-1 students) or dS- +2019 (for J-1 exchange students) you can apply for your +student visa. it is wise to schedule an appointment with +the U.S. embassy or consulate in your country as soon +as possible as sometimes there are no appointments +available for weeks. You should also give yourself some +time to prepare for your visit to the U.S. embassy or +consulate. Visa procedures vary from country to country. +make sure to check the website of the embassy or -1. Customer Discovery interviews -2. Advertising -3. Sales Brochure -4. Newspapers/Magazines -5. Website -6. Online sites that rate products/services -7. Your business and professional network +consulate in your country to find out what documents you need to bring. document +you certainly need include: (1) a current, valid passport; (2) your i-20 or dS-2019 +form; (3) evidence of financial support for the duration of time you will spend in the +US; (4) proof that you have a permanent residence outside the US; (5) one or more +passport pictures meeting U.S. requirements; (6) a nonimmigrant visa application. - Other +Medical Insurance +HWS Colleges require that all students buy the medical insurance provided by +the colleges. all J-1 student-visa holders (that is, all exchange students) are also +required by U.S. law to have their own medical insurance. While F-2 students are +covered by the Colleges medical insurance you might nevertheless consider buying +additional medical coverage. -• Research competitors that no longer are in +Important Documents +You need to bring all important documents relating to your legal, medical, and aca- +demic history with you. Such documents might include medical and dental records, +certificates of immunizations, information about medical conditions and treatments, +prescriptions for medication and eyeglasses or contact lenses, and official transcripts +from secondary school and college. make sure you photocopy all documents and +leave some copies with a person you trust in your home country. it is also helpful to +translate these documents into english before you leave your home country. -business +Packing +Pack as light as possible. remember that you might have to carry all your things at +different times during the journey. if you are studying abroad for only a semester you +might not need to bring your entire wardrobe. You might also want to leave room +in your suitcase for souvenirs you will buy while in the US. at the same time there +might be things you want to bring because they will make you more comfortable -• Research matrix will be messy; matrix for +4 -investment pitches are summaries of -strongest competitors with most important -features + and will make your transition to a new environment easier. Some exchange students +and international students also like to bring small gifts from their home country to +share with new friends abroad. this might be a means for you to share and explain +your culture to your new acquaintances. if you have traditional, national cloth- +ing consider bringing it to wear at events sponsored by the international Students +association or other cultural events. if you arrive during spring semester (January), +to bring a warm jacket and boots, a scarf, gloves and hat. We encourage you to take +advantage of the option to ship your packages ahead of time using the box number +that will be provided to you from the Colleges orientation office. You may also +consider purchasing larger items upon your arrival at our local shopping centers. - Finally +Traveling Tips +You can book your flight as soon as you find the Colleges start date for international +Student orientation. at this time, you might have already received the schedule of +international student orientation. the Colleges provide a complimentary pick-up +from rochester and Syracuse airport for those international students coming to the +Colleges for the first time if you arrive during the assigned times. long-distance +traveling can be really exhausting, so try to get a good-night’s sleep before the trip, +eat lightly and drink plenty of water. it is also a good idea to wear several layers of +clothes so you can accommodate to changes in temperature. label your luggage with +both your home address and the address in the United States (if you don’t have an +address yet, use that of the international student office). Keep you passport, impor- +tant documents, flight tickets, and money or traveler’s checks with you at all times. +You might also want to pack everything you need for a one-night emergency stay +into your hand luggage in case you get stuck at the airport or your luggage gets lost. -• Describe your competitive advantage +Money +You should carry a small amount of U.S. dollars with you while traveling so that you +can buy some food or other small things you might need. However, it is not a good +idea to carry large sums of money. We recommend that you buy traveler’s checks or +get a credit card that allows you to withdraw money from a cash machine when you +need to. Upon arrival on campus, the international student adviser can help you to +set up an account with a local bank. it is highly recommended to have an U.S. bank +account because it will make payments so much easier. Students can create bank +accounts free of charge. -Benefits sell -Features tell +GETTING TO KNOW HOBART AND WILLIAM SMITH COLLEGES +College is an exciting time for everyone and it is a transition for all students. if you +are an exchange student, you already went through the transition process once. +nevertheless, studying abroad brings with it a range of new experiences. every col- +lege is a little different and if you have never studied in the United States before, +many things might be very confusing. - Examples +Hobart and William Smith Colleges +Hobart and William Smith is a small, private liberal arts institution. You can expect +to study in relatively small classes and build close relationships with your classmates - Specialty analysis examples +5 - Positioning diagram but will still -need a feature matrix analysis + and professors. Virtually all students are being housed on campus. living so close +to each other has many advantages: it is relatively easy to get to know new people, +get involved in clubs and activities, and make an impact on the community. at the +same time it can be hard to find personal time and separate school from other activi- +ties when the two are not physically separated. - Works if there are only two main -features/dimensions +Classes +at HWS, a normal course load is four classes per semester. this means that you +will be in class for between 12 and 15 hours every week, or more if you take courses +requiring labs or studio work. there are different types of classes. 100- and 200-level +classes are usually introductory classes and have a wider focus. 300- and 400-level +classes have a narrower, more specific focus. Some professors prefer to lecture, but +you also encounter class-discussions. Higher level classes tend to be taught in a semi- +nar style, in which a lot of the class’ content relies on student presentations. there +are also independent study classes in which you research a topic under a professor’s +guidance but without the classroom structure. as an upper-class student you might +choose to work on an honors project. this is an independent research project that +will culminate in a 80- to 100-page paper. You can take classes from all departments +but at the end of your Sophomore (2nd) year, you will need to declare a major and a +minor subject (or one major, two minors; or two majors, one minor; or two majors, +two minors). as part of your liberal arts education, you will also have to fulfill cer- +tain academic goals. Your academic adviser will help you with that. + +Grading and Grades +Foreign students often find that americans are very competitive. this also reflects +on the system of grading. You will find that you will have many opportunities to +prove your knowledge of a subject in class. many professors will give short, unan- +nounced quizzes that test your knowledge of the course material assigned for that +day. exams are bigger, announced tests that cover the course material of a longer pe- +riod of time. they can be compromised of true or false questions, sentence comple- +tion (“fill in the blank”), multiple choice, matching, essay questions, or a combina- +tion of these. Professors will also assign essays and papers you have to write on your +own time. Your class participation might also count toward your final grade. the +Colleges use the grading system based on a four-point scale: a= 4.0, b= 3.0, C= 2.0, +d= 1.0. You need at least a C average in your classes to remain a full-time student. + +Homework +Some foreign students are overwhelmed by the quantity of homework assigned by +american colleges. especially in the beginning when you still have problems with +the english language, keeping up with homework can be really hard. if you find +yourself struggling to keep up with work, ask for help. Find the Center for teaching +and learning on campus and schedule an appointment! Ctl is open to all students +and its services are already covered by your tuition. they can help you with your +writing assignments, help with time management, or assign you a tutor to help you +with the content of a class. it is also important to realize that most american stu- +dents prioritize their work: decide what readings are important to do and read them +thoroughly; skim the others. +6 + + Professors and Advisers +every student at HWS is assigned an academic adviser. He or she will help you +with any questions you have related to your academic life. Feel free to consult them +before choosing classes, if you are struggling with certain classes or professors, if you +want to know how and when to declare majors and minors, and any other questions +you might have. most professors are also interested in helping you as much as pos- +sible and want to get to know you outside of class. don’t feel embarrassed to ask a +professor a question after class. You should also take advantage of their office hours. +interaction between professors and students in america is a lot tighter than in many +other countries. Students sometimes stop by their professor’s office simply to talk +about something that they are interested in. + +Important information on financial aid +Hobart and William Smith Colleges offer merit scholarships as well as need-based +grants to international students. the average package typically equals more than +half the annual cost of tuition, room and board. Highly competitive applicants will +be considered for a full tuition scholarships. + +Financial aid is available for a maximum of four years for those who enroll as + +first-year students. For those entering with advanced standing (transfers), aid is +available for the time allotted to complete a degree. awards based on demonstrated +financial need, if it becomes apparent that such need no longer exists, may be re- +duced. Students must resubmit a financial aid application each year. + +ADJUSTING TO A NEW ENVIRONMENT +going abroad for a semester or the duration of one’s studies is exciting. at times, +however, the confrontation with a foreign culture can be challenging. Studying +abroad is a learning experience in every sense of the word. besides your scholastic +learning in college courses, you will get to know the american culture and how +people around you view their world and their lives. by experiencing and understand- +ing a different way of life, you will also learn about things in your own culture you +have never thought of before. this is your chance to broaden your view and reflect +on and mature your own values. the process of learning about a new culture and +finding your own place in the new environment can be confusing and tiring. the +best way to deal with a new situation and find out about something you don’t under- +stand is to ask questions. + +Culture Shock and Adjustment +Culture shock is a normal phenomenon on your way to adjusting to a new environ- +ment. For some people, the encounter with a new culture can be shocking, indeed, +but for most people culture shock is not as sudden and dramatic as you might expect. +When the excitement about being in a foreign country wears off, cultural differences +can become overwhelming. You might think your language skills are inadequate; you +might be struggling with course work; you might be troubled by unfamiliar food; you +might have difficulties interpreting other people’s behavior; you might be offended +by the way people act or dress or by their body language. there are a variety of +symptoms of culture shock. Some people feel tired and nervous; some are frustrated + +7 + + with or even hostile toward the american culture; some get angry and irritated very +easily; some isolate themselves or socialize only with other international students and +people who speak their language. You might also feel very homesick and question +why you decided to go abroad in the first place. a way to prepare yourself to deal with +culture shock is to clarify for yourself why you want to study abroad. Knowing your +personal expectations and goals will help you overcome periods of doubt. + +THINGS TO DO +• Chose a non-credit or half-credit class like tennis, scuba diving, or basic sailing. +• Join some of the many student clubs on campus (http://www.hws.edu/studentlife/ + +campuslife/sll/clubs). + +• You like to be active? Visit the sport and recreation facilities and take a yoga class, +borrow a racquet and play tennis or racquetball. You can also try out for intramu- +ral sports. + +• get involved in the religious life on campus! there are services and meditation as +well as bible study sessions, ecumenical candlelight prayers, Sabbath dinners and +a lot more. + +• Participate in ongoing events – take in a movie on Fridays at Friday Flix, visit the +Cellar Pub in the basement of Coxe Hall and see what is happening (comedians, +bands, etc.), show school spirit by supporting one of many sporting events hap- +pening on campus. + +• are you interested in current events? Political and social developments? the +Fisher Center lecture Series, the President’s Forum and the President’s radio +Show are a few venues through which you can satisfy this interest. + +• travel the area! intercultural affairs (iC) and the Center for global education + +(Cge) will occasionally offer trips to see the area, oraP (outdoor recreational +activities Program) organizes hiking, rock climbing and rafting trips (among +others), and you can always get a group of friends together and go somewhere on +your own. + +• offer your help in the language departments. maybe someone needs a Spanish + +tutor or a teaching assistant for mandarin? + +• discover geneva. Volunteer. to find out how visit the Public Service office in + +trinity (2nd floor) (or call x3825). + +• do you like to dance? or watch others dance? the dance department is big on +campus and the students as well as the faculty put on dance shows! ask about +Koshare or join one of the dance teams. + +• Check the daily Update! there is always something going on – just show up and + +try it out! + +8 + + P +H +O +N +E +N +U +M +B +E +R +S +& +O +F +F +I +C +E +S + + Extension + +IMPORTANT PHONE NUMBERS +Office +business office +Center for global education +Center for teaching and learning +Counseling Center +deans: Hobart +deans: William Smith +Financial aid +Hubbs Health Center +information technology +intercultural affairs +registrar +residential education +Security +emergency +Student accounts + +3655 +3307 +3351 +3388 +3300 +3467 +3315 +3600 +4357 (HelP) +3319 +3651 +3900 +3656 +3333 +3386 + +OFFICE LOCATIONS +admissions +alcohol and other drug Programs +alumni House +athletics (WS) +buildings and grounds +business office +Career Services +Center for global education +Ctl +Chapel +College Store +Communications office +deans: Hobart +deans: William Smith +dining Services + +Financial aid +Hubbs Health/Counseling Center +information technology +intercultural affairs +President +Provost’s office +Public Service office +registrar +residential education +Security +Student accounts +Student life and leadership +WeoS radio Station + +bradford House +Smith Hall +615 South main +Winn-Seeley +medbery Parking lot +demarest (basement) +trinity (1st Floor) +trinity (3rd Floor) +Harris House (678 S. main St.) +St. John’s Chapel (near 630 S. main St.) +51 Saint Clair Street, Sherrill Hall +durfee House (639 S. main St.) +Smith Hall (1st Floor) +Smith Hall (1st Floor) +Scandling Center Café, Saga, +Cellar Pub, de Cordova Café +demarest (basement) +119 St. Clair St. +Williams Hall +288 Pulteney St. +Coxe Hall (2nd Floor) +Coxe Hall (2nd Floor) +trinity (3rd Floor) +gulick (1st Floor) +101 St. Clair St. +medbery Parking lot +demarest (basement) +Scandling Center (2nd Floor) +113 Hamilton St. + +9 + + BANKS AND MOVIE THEATERS IN GENEVA + +Community Bank +5 Seneca St. +geneva, nY 14456 +Phone: (315) 789-7700 +Website: www.communitybankna.com + +First Niagara Bank +811 Canandaigua road +geneva, nY 14456 +Phone: (315) 789-1049 +Fax: (315) 789-6009 +Website: www.firstniagarabank.com + +BANKS +Five-Star Bank +P.o. box 193 +geneva, nY 14456 +Phone: (315) 789-2300 +Fax: (315) 787-3263 +Website: www.five-starbank.com + +Seneca Falls Savings Bank +342 Hamilton St. +geneva, nY 14456 +Phone: 315-789-8100 +Website: www.seneca24.net/ + +MOVIE THEATERS +Geneva Movieplex 8 +t & C Plaza, Hamilton St. +geneva, nY, 14456 +Phone: (315) 789-4524 +Web: www.cinemanorth.com + +PLACES TO STAY +Note: the Colleges provide this list as a service to all who might visit our area, but +do not endorse any local hotel, motel, inn or bed and breakfast. + +the following are a handful of accommodations that are available in and around +geneva, no more than 30 minutes from campus. lodging is also available in ithaca, +rochester and Syracuse, which are each approximately one hour from Hobart and +William Smith. You can find out more about accommodations outside the greater +geneva area by visiting the geneva Chamber of Commerce or ontario County +tourism bureau websites. + +Holiday inn, Waterloo +(315) 539-5011 or (800) HolidaY + +Hampton inn +(315) 781-2037 + +ramada inn, geneva, lakefront +(315) 789-0400 + +belhurst Castle/White Spring manor, +geneva +(315) 781-0201 + +microtel inn & Suites, Seneca Falls +(315) 539-8438 + +geneva-on-the-lake, geneva +(315) 789-7190 or (800) 3-geneVa + +motel 6, Hamilton Street, geneva +(315) 789-4050 or (800) 466-8356 + +10 + + G +E +N +E +V +A + +, + +N +E +W +Y +O +R +K + +PLACES TO EAT +*Please note that some restaurants are closed on Mondays or Tuesdays, +so call ahead. + +The area code for all numbers listed below is (315). + +American/Continental +abigail’s restaurant, routes 5 & 20, Seneca Falls, 539-9300 +belhurst Castle, lochland road, route 14, 781-0201 +the Crow’s nest, 415 boody’s Hill road, 781-0600 +emile’s restaurant, east north Street and routes 5 & 20, 789-2775 +geneva-on-the-lake, 1001 lochland road, 789-7190 +madderlake Café, West lake road, 781-2424 +the red brick inn, 291 main St., 789-3247 + +Diners/Grilles +Flour Petal Café, 34 linden St., 781-2233 +Parkers’ grille & tap House, 100 Seneca St., 789-4656 +Ports Cone & grill, 4432 W. lake road, 789-2020 +Water Street Café, 467 exchange St., 789-2560 +the Captain’s room, 372 exchange St., 781-0122 +dana’s time out, 258 Hamilton St., 781-2107 + +Italian-American and Pizza +nonna Cosentino’s trattoria, 1 railroad Place, 789-1638 +Uncle Joe’s Pizzeria, 99 n. genesee Street, 781-1199 + +International +ling ling, 481 Hamilton Street, 781-0668 +main moon, 480 exchange Street, 781-5106 +Wing tai oriental restaurant, 164 Castle Street, 789-8892 +garcia’s mexican restaurant, 313 Hamilton Street, 789-5760 + +Fast Food/Family Chains +routes 5 & 20 west of downtown geneva (Hamilton Street) offers an array of popu- +lar fast food and family chains, including Quiznos, tim Horton’s, Club 86 bagels +and Cakes, Friendly’s, KFC-taco bell, mcdonald’s, long John Silver’s, Pizza Hut, +Subway, Waffle Works and Wendy’s. + +11 + + FREQUENTLY ASKED QUESTIONS (FAQs) + +Do you require the TOEFL for international students? +Yes, we do require the toeFl (test of english as a Foreign language). We offer limited +english language support on campus, and require that students have a solid foundation +in english prior to admission. the minimum required computer based toeFl score is +220. the Colleges also will accept the aPiel examination in lieu of the toeFl. + +Do you offer financial assistance for international students? +need-based scholarships are available to international students. the cost of attend- +ing college here for the 2007-2008 academic year is U.S. $45,968. the average aid +package typically equals more than half the annual cost of tuition, room and board. +Highly competitive applicants will be considered for a full-tuition scholarship. the +financial aid application forms are included in the application for admissions. + +How many international students do you currently have on campus? +We have 32 full-time matriculated students in a degree program Fall 2006 +Semester. the following countries are represented: austria, bulgaria, Canada, +China, Colombia, Czech republic, germany, india, nepal, Pakistan, South Korea, +Sweden, turkey, Ukraine, United Kingdom and Vietnam. + +Do you have an international student adviser? +Yes, the student adviser is responsible for working with our international students +on campus to help them make the transition from home to the Colleges. We offer an +international student orientation prior to the first-year and transfer student orienta- +tion at the beginning of the academic year, and the beginning of the spring semester +for new arrivals. a variety of activities and programs are available throughout the +year. Student advising and counseling is available in a variety of other areas on cam- +pus, including faculty advisers, Hubbs Health Center, and the Center for teaching +and learning. We make every effort to work individually with each student to help +them reach their potential and make the most of the opportunities available here. + +Do you have an international student organization? +Yes. the international Student association (iSa) is a student organization that cel- +ebrates the diverse cultures found around the globe. other cultural clubs with interna- +tional perspectives represented on campus include asian Student association (aSU), +latin american organization (lao), Caribbean Student association (CSa), and +dia al-din: muslim Student association. + +Do you have E.S.L. services? +Yes, an entry level course is offered, as are workshops and one-to-one tutoring. How- +ever, the Colleges do not offer a full e.S.l. curriculum. + +What kinds of housing are available? +Within the Hobart and William Smith community are dozens of smaller communi- +ties--those of the individual residences. the range of housing options is extensive, + +12 + + with coed or single-sex college residences, cooperative houses, theme houses, and +town houses among the choices you’ll find over your four years here. residence +halls offer a variety of configurations, with traditional singles and doubles as well as +suites. Some recent theme houses have included Jewish Culture House, multicul- +tural Sisterhood House, Writers House, and international awareness House. the +ten cooperative houses provide an alternative to standard residence-hall living, with +residents sharing responsibility for food and housecleaning arrangements. + +Is it possible to work on campus? +You are eligible to work up to 20 hours a week on campus, if your visa status permits +it. However, the demand for such employment exceeds the number of jobs available. +therefore, you cannot use on-campus employment as a resource for the issuance of +the i-20 form. + +STORAGE AND TRANSPORTATION OF SCHOOL YEAR ITEMS +Special consideration will be given to those international students who need storage +on campus during the school breaks. international students must attend a manda- +tory meeting in September regarding storage and transportation of school year items. +attendance will be taken at this meeting. Students unable to attend this meeting +must take the initiative to set up a meeting with the international Student adviser +before oct. 1. + +the storage and transportation meeting in September will encompass the following +information: +a. date and time of transportation to and from storage areas + +- Winter break (december) +- Summer break (may) +b. location of storage areas +c. How to pack and maximum number of storage items allowed +each student will be allowed no more than 8 items. items must be labeled with +students name, return date and number of items. boxes must be secured properly +and must be easily manageable. large items such as refrigerators, sofas, etc. will not +be accommodated due to limited space. +d. transportation requirements: items are to be placed at curbside 15 minutes +before van arrives for pick up. +e. Cost of transportation = $20 per student (gas and labor cost) +f. information on surcharges: the student’s account will be charged a surcharge for +each infraction: + +- No shows = $50 +- Fail to have items at curbside = $20 +- Not packed = $20 +- Change of time and/or date without 24 hour notice to Intercultural Affairs + Office = $40 + +Please Note: Storage space is limited, therefore there is no guarantee that you +will get storage or transportation support. + +13 + + IN CASE OF AN EMERGENCY +if you should have a need to contact the international Student adviser while travel- +ing outside of the United States please call (315) 781-3656 and ask for extension +3319. You can also contact the office directly by calling (315) 781-3319. + +In an emergency, they can be reached at (315) 781-3333. + +remember that the international Student adviser, while a resource to you, is not +your only resource during your time at the Colleges. While on campus, your respec- +tive deans will be available to assist you with a variety of issues, and should always +be kept aware of important matters. + +WHAT TO EXPECT DURING SCHOOL BREAKS +Since not everyone is able to return home during school breaks, international +students are allowed to remain on campus during those times. at specific times +during the fall and spring, you will receive e-mails from the residential education +office asking about their plans during the upcoming break. Your immediate response +is necessary as it is the only way to ensure that you will get permission to remain in +your room or elsewhere on the campus during school breaks. + +Students are not allowed to remain in their rooms during fall, winter and spring + +breaks. during summer vacation, students who remain on campus must move into +temporary housing. once again, an e-mail from the residential education office +is the first step in determining where students will be staying during the summer. +Currently, international students are allowed to stay on campus during the summer +holiday free of charge (subject to change). + +the dining hall is not open to students during the summer months so students +will need to provide their own meals. most summer housing will feature a kitchen, +but in the event that one doesn’t, appropriate accommodations will be made for +those students who wish to prepare their own meals. + +the international student adviser must also know your plans and should also be + +notified if you will or will not be staying on campus. + +US Holidays/Observances +January 1 +January 15 +3rd monday in February +2nd Sunday in may +4th monday in may +3rd Sunday in June +July 4th +1st monday in September +october 31 +november 11 +4th thursday in november +december 4 +december 25 +december 26 + +14 + +new Year’s day +martin luther King Jr. day +Presidents day +mother’s day +memorial day +Father’s day +independence day +labor day +Halloween +Veterans day +thanksgiving +Hanukkah begins +Christmas day +Kwanzaa + + SECTION II: +IMMIGRATION PROCEDURES + +HOW TO GET ASSISTANCE WITH IMMIGRATION MATTERS + +the international Student adviser is available to help you process your immigration +forms and documents. Working with the department of Homeland Security can be +very complicated, detailed, and time-consuming. it is important that you do not rely +on information obtained from friends and other students because circumstances may +vary from one student to another and the regulations can and do change. + the international Student adviser must work within the federal immigration +laws. When making a request, please make sure that you allow plenty of time for +your request to be processed and plan to be patient. Have your documents, such as +your passport, i-94 card, i-20 or dS-2019 with you when you go to the international +student office for immigration assistance. + +Important: You should allow at least one week for processing signatures, letters +and/or other types of paperwork. Special consideration will be given for emergency +situations. + +Reminder: the international Student adviser cannot remember when each +student’s documents expire. it is YOUR responsibility to keep track of all immigra- +tion documents and keep them current. + +Hristo Dimitrov ’09 (Bulguria), and Ayokunle Abogan ’06 (Nigeria) + +15 + + IMMIGRATION BASICS AND DEFINITION OF TERMS + +as with all countries, the United States has laws and regulations governing foreign- +ers who are temporarily within its borders. the immigration regulations are adminis- +tered by the Department of Homeland Security (DHS). + +immigration regulations can change frequently and are difficult to interpret. it +is important for international students to understand the rules as they apply to stu- +dent status. The DHS does not consider ignorance of the law a legitimate reason +for failing to obey it. + Questions should be directed to the international Student adviser, who will be +able to assist with legal regulations and can sign your immigration documents. + +dHS requires that a copy of your passport, i-94 and other immigration docu- +ments be kept on file by the international Student adviser, as well as your local +address and telephone number. + +Passport +be sure to keep your passport in a safe place. it is your most important document. +most passports are good for a limited period of time. Your passport must be valid at +all times while you are in the United States. + +the expiration date of your passport should be six months beyond the date you + +enter the U.S. if your passport expiration date is close to the date you will begin +classes, obtain a new passport before leaving home. + +it is your responsibility to remember when your passport will expire and to con- +tact the nearest consulate or embassy for your country. to find the nearest consulate +or U.S. embassy check http://usembassy.state.gov. + +Your passport can only be extended by your government and it usually takes +several weeks for a passport to be renewed, so be sure to allow enough time. You may +extend your passport in the U.S. at your home country’s embassy or consulate. You +are in this country illegally if your passport expires and are subject to deportation. + +Visas +the visa stamped in your passport at the U.S. embassy or Consulate is for permis- +sion to enter the United States for a specific purpose and period of time. Students +are usually given an F-1 visa. J-1 visas are given to researchers, scholars and students +who are sponsored by agencies or universities. + +Since the visa stamp is only important for entry and re-entry to the United +States, there is no need to worry if it expires while you are here if you are maintain- +ing your student status. because a visa is only a permit to apply for entry into a coun- +try, it is not possible to obtain a U.S. visa while you are in the U.S. to apply for a +visa, you must go to a U.S. consulate or embassy in a country other than the United +States, usually your home country. + +I-94 (Arrival/Departure Record) +Form i-94, also called the arrival/departure record of Stay Permit, is the small white +card given to you at the port of entry to the U.S. this form is very important since it +officially determines how long an international student can remain in the U.S. + +16 + + Students on the F-1 and J-1 visas usually have their i-94 stamped “d/S” (dura- + +tion of Status), which means that the student can remain in this country for the +length of time required to complete his/her studies as long as the student follows all +of the regulations of the student visa. the eleven-digit number written on your i-94 +is your dHS admission number, which is used by the dHS to monitor entry into and +departure from the U.S. The I-94 cards are turned in when leaving the U.S. and a +new I-94 is completed with each subsequent entry. + +if you are on a visa other than a F or J, or if you have a specific date stamped on + +your i-94 card instead of the “d/S” designation, you must apply for an extension, a +Change of Status, or exit the U.S before the expiration date or you will be consid- +ered out of status and a visa overstay, which can have serious dHS consequences. + +SEVIS +an acronym for Student and exchange Visitor information System (SeViS). it is a +data collection and monitoring system that creates an interface between institutions +of higher education, the department of Homeland Security (dHS), consulates and +embassies abroad, and ports of entry. Schools are required to make regular electronic +updates in SeViS throughout each semester on the records of their enrolled students +in F-1 and J-1 status and their dependents and their researchers and faculty on J-1 +status. this includes, but is not limited to enrollment status, changes in address, +changes in level of study, employment recommendations and school transfers. + +Form I-20 +When Hobart and William Smith Colleges accepts an international student for +full-time study, the student is sent a Form i-20 a-b as proof of acceptance and to +use when applying for an F-1 Student Visa at a U.S. Consulate abroad. the student +presents the i-20 with their passport and visa at the Port of entry and the i-20 is +stamped and given back to the student. this is your Certificate of eligibility, and +it must be signed by a dSo/ international Student adviser before you leave the +United States. Students should keep copies of all i-20 issued to them. + +Immigration Status +this is often confused with “visa” but your immigration status, e.g., F-1, J-1, etc. is +determined at the time of your entry into the U.S. by an immigration official and is +noted on the Form i-94. You may have many visa stamps in your passport but, upon +entry into the U.S., an immigration inspector will admit you in only one immigra- +tion status which is noted on the i-94 card. + +OBTAINING A VISA AND FORM I-20 +Your passport should be valid at least six months beyond the date of entering the U.S. +as a student. be sure to check your passport expiration date. if it will expire close to +when you will begin classes, obtain a new passport before applying for your visa. + +How to Obtain a Visa +although U.S. embassies will not issue a student visa prior to 120 days before the +start date on the i-20 (or dS-2019 for J’s), you can schedule an appointment prior + +17 + + to the 120-day period. Since processing times have become longer due to increased +security clearances, students should check with their local U.S. embassy or consulate +to learn the estimated waiting time and allow plenty of time to obtain the visa. a +list of consulates and embassies can be found at http://usembassy.state.gov/. + +You will need to pay a $100 SeViS fee at least 3 business days before your visa +interview. the fee can be paid online or by mail. For complete directions, see http:// +www.fmJfee.com. + Unless you are a sponsored J-1 student or already in another visa category, the +Colleges’ office of admissions will send you a Form i-20 when you are admitted and +have submitted proof of your ability to finance your education. take the i-20 form, +your acceptance letter, passport, copies of sponsorship letters and bank statements +to prove financial support, and a receipt for your SeViS fee payment to the United +States embassy or consulate nearest your home. once you have acquired the visa +stamp in your passport, you are ready to enter the U.S. + +Note: Students may not enter the U.S. more than 30 days prior to the start date +on their I-20. + +Canadian and some British Commonwealth citizens do not need visas to enter +the U.S., but they must request F-1 status and present their i-20 forms for process- +ing at the point of entry even if the official does not ask for it. they should also be +prepared to show evidence of financial support and proof of citizenship. Without a +stamped i-20 and an i-94 card with the F-1 or J-1 designation on it, you are con- +sidered a tourist in the United States and will not be legally able to study and take +advantage of the special privileges allowed those in Student Status until he or she +has been in an immigration status appropriate for studying for at least nine consecu- +tive months. + +Note: If you will not be entering the U.S. on an F-1 visa, do not use an I-20 form +to obtain a visa. If you received an I-20, please return it to the Office of Admis- +sions with an explanation of the type of visa you hold. + +if your visa document (i-20 or dS-2019) does not arrive in time for you to secure +the proper visa, do not enter the United States on a B1/B2 Visitor Visa. obtain a +b-2 “prospective student” or “prospective exchange visitor” visa from a U.S. embassy +or consulate post in your country. be certain the consul writes “Prospective Student” +or “Prospective exchange Visitor” (for J-1) in your passport on the visa page. You +will need to present your letter of acceptance and evidence of financial support to +request this visa. + +in general, it is best to avoid using the b-2 visa since you will have to pay a fee +and do a time-consuming change of status through dHS (department of Homeland +Security) after you arrive. You may not begin your studies until the change is com- +pleted, so this could also delay the start of your studies until a subsequent term. + +Note: Students must attend the school named on the I-20 or DS-2019 presented at +the port of entry in order to be in status. Also, do not enter the U.S. on a “W-T” + +18 + + (waiver) as this classification cannot be extended or changed and you will lose your +legal immigration status as a student. + +Students Transferring From another Colleges in the United States +if you are an F-1 student transferring from a U.S. institution and will not be leav- +ing the country prior to enrollment at Hobart and William Smith Colleges, follow +the procedures required by dHS. bring your completed i-20 to the international +Student adviser’s office as soon as possible after your arrival on campus. Sign the +document and leave it to be processed by a dSo/ international Student adviser +who will notify you when you may pick it up. + +if you leave the U.S. before beginning your studies at Hobart and William +Smith Colleges, use the new i-20 issued by the Colleges when you arrive at the port +of entry. Keep the i-20 from your old school with your immigration records, but do +not use it to enter the U.S. it is not necessary for transfer students with an unex- +pired student visa and who have maintained their status to obtain a new visa, even +though the former school’s name is listed. + +Transfer to another School +to be eligible to transfer schools, you must be pursuing a full-course of study at your +current school. otherwise you are “out of status” and must apply to dHS for rein- +statement to student status, which, if granted, will allow you to transfer. + +Students transferring to another school must notify the dSo/international +Student adviser of their intent to transfer. Hobart and William Smith Colleges +must release your SeViS record to the new school before a new i-20 can be issued +for the transfer. Upon arrival at the new school, the student should take the new +i-20 to the international Student adviser at the new school. J-1 students wishing to +transfer should consult with their sponsor. + +Changing Your Academic Level +if you are currently a student at Hobart and William Smith Colleges, and will be +continuing your studies in an additional program of study (such as a mat Program), +you should follow the transfer procedure. a new i-20 will be issued for the new pro- +gram, and you will use the new i-20 to enter the country. if you do not leave, bring +it to the international Student adviser as described in the transfer process in the +preceding paragraph. Failure to complete this “internal transfer” will cause you to +be out of status with dHS, which can adversely affect work authorizations or other +student privileges. + +DIFFERENCES BETWEEN F-1 AND J-1 STATUS +there are two types of entry visa issued to nonimmigrant students who intend to +study full-time at Hobart and William Smith Colleges—the F-1 student visa and the +J-1 exchange visitor visa. + +• Source of Funding: J-1 exchange visitor visa status is available to those students +who are supported substantially by funding other than personal or family funds. +Such funding may include that which comes from the U.S. government or the + +19 + + student’s home government, or an international organization, or the Colleges. Stu- +dents who are supported by personal or family funds are ineligible for J-1 immigra- +tion status and must come to the U.S. to study in F-1 immigration status. + +• Foreign Residence Requirement: Students in J-1 immigration status and the J-2 +dependents may be subject to a “foreign residence requirement.” J-1 students who +receive direct or indirect U.S. or home government funding, or who are studying +in fields for which personnel are considered in short supply in their home coun- +tries (most developing nations have “Country Skills” lists of varying lengths), are +ineligible to apply for a change to another non-immigrant status (except a or g) +or permanent residency in the U.S. until they have satisfied the “two year home +country physical presence requirement. only those in J-1 status with government +funding or whose skills are on Country Skills list are subject to the foreign resi- +dence requirement, which mean upon completion of the program of study they must +reside in their country of last legal permanent residence for two years before they are +eligible to apply for entry into the U.S. on an immigrant visa or on specialized work +visa (H or l). there is no foreign residence requirement for students on F-1 student +status. refer to the department of State’s website for additional information on the +exchange Visitor Program. + +• Medical Insurance: Students in J-1 status and their J-2 dependents are required to +have comprehensive medical coverage in order to fulfill U.S. government regula- +tions governing the J exchange Visitor status. Currently the medical insurance +offered by Hobart and William Smith Colleges does not cover all requirements for +those in J immigration status. the U.S. department of State specifies that program +participants and their dependents have medical coverage with the following mini- +mum requirements (22 CFr 62:14): medical benefits of at least $50,000 per accident +or illness; repatriation of remains in the amount of $7, 500, medical evacuation of +the exchange visitor to his or her home country in the amount of $10, 000, and a +deductible not to exceed $500 per accident or illness. For more information please +contact the international Student adviser. + +• Work Permission: Students in either F-1 or J-1 immigration status may work part +time (up to 20 hours per week) on campus with permission from the international +Student adviser. For both statuses, permission to work off campus based on eco- +nomic need may be requested only after the first full academic year of student status, +and only under extraordinary circumstances of unforeseen need. the United States +Citizenship and immigration Services (USCiS) authorizes F-1 students to work off +campus, whereas J-1 students submit an application for work permission to their visa +sponsor, i.e. the agency or school that issued the dS-2019 form. + +• Practical or Academic Training: Students in F-1 and J-1 immigration status are +eligible to engage in differing lengths of practical or academic training during their +studies and at the completion of their academic programs. Practical training, a +privilege of F-1 immigration status if you are eligible, is employment in one’s field of +study. those in F-1 immigration classification may be eligible for “curricular” practi- + +20 + + cal training (which is authorized by the international Student adviser) during their +programs, in addition to one year of “optional” practical training that may take place +during or after one’s program of study. Permission to engage in practical training may +be granted only after you have been in valid student status for one academic year, +except in a graduate program that requires it from the first term. the international +Student adviser recommends optional practical training and the USCiS authorizes +it. refer to the Practical training overview for complete information on both cur- +ricular and optional practical training. academic training, a privilege of J-1 immi- +gration status if you are eligible, is employment which is integral, not just related, +to one’s field of study. one is eligible to apply for permission to engage in academic +training after one academic term in valid student status. a total of 18 months of +academic training is possible during or following completion of a program of study, +with the exception of some postdoctoral research positions that may qualify for up +to 36 months. an application for academic training is made to the J-1 student’s +exchange Visitor program sponsor. refer to academic training authorization for +those in J-1 exchange Visitor immigration Status for complete information. + +ARRIVING AT A U.S. PORT OF ENTRY … +WHAT A STUDENT CAN EXPECT +Plan your arrival: You may be refused entry into the United States if you attempt to +arrive more than 30 days before the start date listed on your SeViS i-20 form. + +Always carry your passport and I-20 on hand: do not check the following docu- +ments in your baggage. if your baggage is lost or delayed, you will be unable to pres- +ent the documents at your port of entry. as a result, you may not be able to enter +the United States. + +In addition, it is strongly recommended that you carry on hand the following + +documentation: + +1. evidence of financial resources +2. evidence of student status, such as tuition receipts +3. name and contact information for your international Student adviser, + +including a 24-hour emergency contact number at the school + +Complete your entry paperwork: +Arriving by Air: Flight attendants will distribute Customs declaration Forms and +arrival departure Forms (i-94). these must be completed prior to landing. + +Arriving by Land or Sea: the CbP officer at the port of entry will provide the +necessary Customs declaration Form and the arrival-departure Form (i-94) to be +filled out upon arrival. + as you arrive at the port of entry proceed to the terminal area for arriving +passengers. Have the following documents available for presentation: you passport, +i-20, arrival-departure Form (i-94), and the Customs declaration Form. Your docu- +ments will be examined and processed by the department of Homeland Security +(dHS) at the port of entry. Your passport will be returned to you along with an i-94 +card (arrival/departure record) designated “d/S” which means “duration of status” + +21 + + and refers to the time you are engaged in a full-time course of study as indicated on +your visa document. the dHS official will stamp and return your i-20 to you. + +Note: Your documents are extremely important. They legitimize your presence in +the United States. Keep them in a safe and secure place. + +For Further Information: +Preparing to Study in the U.S. http://educationusa.state.gov/ +applying for a Visa to Study in the U.S. http://travel.state.gov/visa/temp/types/ +types_1268.html + +MAINTAINING & BEING IN STATUS WITH DHS +an F-1 student is considered in-status for the entire length of time during which +he/she is enrolled as a full-time student in an educational program if adhering to +the conditions of the visa designated on the i-94 card, plus any authorized period of +practical training and an additional sixty (60) days to prepare for departure. be sure +to pay close attention to the conditions described below: + +• report to the international Student adviser for SeViS registration upon arrival +at Hobart and William Smith Colleges. + +• attend only the school authorized. this means that you must only attend the +school whose i-20 you last used to enter the U.S. or the school whose i-20 you used +to transfer schools. if you decide to withdraw from the Colleges before completing +your program, you are not allowed to remain in the U.S. with in-student status un- +less you transfer to another institution. + +• Complete a full course of study for each academic term and make normal academ- +ic progress towards completion of program. attendance and/or full-time enrollment +is not required during the summer. at Hobart and William Smith Colleges 4 credit +hours per semester is required and exceptions can be made on the recommenda- +tion of the international Student adviser for valid medical or academic reasons as +defined by the dHS. be sure to speak with the international Student adviser about +any exceptions beFore dropping below full-time status in order to avoid the seri- +ous consequences that may result from being out of status. + +• transfer or extend program in a timely manner. duration of Status expires if a +student takes longer than expected to complete a given academic level (e.g. bachelor’s, +master’s, and doctorate). the amount of time permitted for completing studies at a +given level is determined by the date on the initial i-20 issued at the beginning of each +academic level. Students must pay close attention to the expected completion date +noted in item 5 on the form i-20 for each level. if more time is needed, an applica- +tion for extension of stay must be filed with the dHS. there are certain requirements +to meet in order to be eligible for an extension, so you will need to consult with the +international Student adviser at least 45 days before the completion date. + +22 + + • Keep continuity in the program of study (in general an absence from the U.S. of +more than 5 months terminates status. However, an authorized study abroad pro- +gram does not break continuity). + +• refrain from unauthorized work. When you apply for an F-1 visa, you must prove +that you have full financial support. the law assumes that you will not be employed +in the U.S. and therefore will not accept employment that could be filled by a U.S. +worker or student. there are some exceptions to this policy provided that full-time +enrollment during the year is maintained and the appropriate permission is ob- +tained, but you should always check with the international Student adviser before +accepting employment, especially if it is off campus. + +• When traveling abroad, always return in proper F-1 status. + +• maintain a valid passport for 6 months into the future (Some countries have +agreements with the U.S. that recognize the passport as valid for 6 months after the +expiration date). + +• report any address changes to the international Student adviser within 10 days +of the change. dHS regulations require that the international Student adviser +maintain a record of the current address where you are living. therefore, you must +always notify the international Student adviser immediately of any change of ad- +dress. addresses with a post office box number or addresses of your academic depart- +ment are not acceptable to the dHS. + +• Comply with Special registration requirements if applicable (nSeerS). + +• File timely requests of change of status, practical training and other benefits. + +• obey all State and Federal laws. + +• depart form the U.S. in a timely manner. + +• do not enter on a tourist visa (b-1 or b-2). + +Note: Students who fail to maintain full-time enrollment are not eligible for any +of the benefits of the student visa, such as on-campus and off-campus employ- +ment, practical training, remaining in the U.S. during summer vacation, transfer +to another school or change of level, or registration for further courses. they are +considered to be “out of status” and must apply to dHS for reinstatement to valid +student status. eligibility for benefits will be regained only if and when the dHS +approves the reinstatement. if you have any questions regarding reinstatement, see +the international Student adviser immediately. + +23 + + EXCEPTIONS TO A FULL COURSE OF STUDY + +• the normal vacation period; + +• medical problems documented by a licensed medical doctor, osteopath or clinical +psychologist. approval for reduced or no course load for a medical condition is limited +to 12 months total for the program of study; + +• teaching or research assistantship; + +• Valid academic reasons, such as first semester language difficulties and/or adjustment +problems, improper course placement, unfamiliarity with american teaching methods; + +• the final term when the student needs less than a fulltime course load to complete +program of study. Student must be enrolled in at least one credit; + +• during a period of full-time optional or Curricular Practical training. + +DURATION OF STATUS FOR F-1 STUDENTS +an F-1 student is admitted to the U.S. for a period known as “duration of Status” +which is recorded upon entry documents i-94 and i-20 with the notation “d/S”. du- +ration of Status refers to the period a student is pursuing a full course of study, plus +any authorized practical training following completion. The dates on the I-20 de- +termine the initial parameters of the duration of status. A student can be admit- +ted to the U.S. up to 30 days prior to the report date on the I-20 and may remain +60 days after completion of program or practical training. Certain factors may +reduce grace period to 15 days or to 0 days. an international Student adviser au- +thorized withdrawal from classes is allowed a 15-day period for departure. a student +who fails to maintain status is not eligible for an additional period for departure. + +TEMPORARY DEPARTURE FROM THE U.S. +If you plan on leaving the U.S. temporarily to visit your homeland or some other +country, an International Student Adviser must validate your I-20 for re-entry into +the U.S. You may not be allowed back into the U.S. without the validated immigra- +tion documents, a current visa, and a passport which is valid for at least six months +after you re-enter the country. + most students with valid status may travel to Canada, mexico and many of the +adjacent islands if they have a valid passport, a valid i-94 and a valid i-20 or dS-2019. +if you re-enter the U.S. in fewer than 30 days, an unexpired U.S. visa is not required. +However, Canada, mexico and the island nations do require entry visas for citizens of +certain countries. + +it is always best to check with the embassy or consulate of any country you wish + +to visit. Since certain countries are ineligible for this benefit and travel regulations are +subject to change, you should check with the international Student adviser if you +intend to travel with an expired visa. + +Students who apply for a new visa in contiguous territory forfeit the right to re- + +24 + + enter with an expired visa. if the new visa is denied, the student would not be able to +re-enter the United States to study until a valid visa is obtained in their home country. + +ACCEPTING U.S. GOVERNMENT PUBLIC ASSISTANCE OR PUBLIC +BENEFITS +international students and scholars should not, under any circumstance, accept public +benefits from the federal, state, county or any other local government agencies in the +United States. this includes, but is not limited to, participation in programs such +as medicare/medicaid, Supplemental Social Security income, aid to Families with +dependent Children, food stamps, and subsidized housing. accepting such benefits +can cause you to lose your legal status in the United States and create a bar to your +re-entry to this country. + +Cooperation between state social service agencies and the inS and department + +of State make it possible for an inS inspector or U.S. Consular officer to become +aware of an individual who has accepted public benefits. the acceptance of public +benefits is interpreted as a declaration that the individual is unable to meet his or her +financial obligations (a requirement of obtaining and maintaining status) and thereby +rendering him or herself ineligible for F-1 or J-1 status. + occasionally a well-meaning social service agency, hospital or public housing admin- +istrator may offer such assistance not realizing that non-immigrants are ineligible for such +benefits. additionally, even though U.S. citizen children of non-immigrant parents may be +eligible for pre and post-natal public benefits, the non-immigrant parent may be denied a +visa or re-entry based on the child’s receiving the benefits. again this is interpreted as the +non-immigrant individual not being able to meet his or her financial obligations. + +Information on U.S. Social Security Numbers (SSN) +if you will be employed in the United States, including at Hobart and William Smith +Colleges, you must obtain a U.S. Social Security number, which is used by the U.S. +government to identify wage earners for tax purposes. the number assigned to interna- +tional students is not a valid U.S. Social Security number. + +to apply for a Social Security number (SSn) you must obtain a “proof of work” + +form from the international Student adviser confirming your student or exchange +visitor status, that you are authorized to engage in on-campus employment, the nature +of the employment to be engaged in, and identify the department/office for whom you +will be working before being able to apply to the Social Security administration for +a SSn. You must also present your passport, your i-20 or dS-2019 form and your i-94 +form. the Social Security administration office is open monday-Friday from 9:00 a.m. +to 4:00 p.m. (closed on federal holidays). + +Social Ssecurity numbers will not be issued to foreign nationals who are not +authorized to work. a Social Security number is not required for opening a bank +account, registering for school, filing a U.S. tax form or applying for a driver’s license. +if you are required to file a tax return and are not eligible for a U.S. Social Security +number, you will have to apply to the internal revenue Service (irS) for a tax iden- +tification number (tin) on form W-7. this form may be obtained from the Colleges +payroll department. + +For more information about U.S. Social Security Numbers visit the website or + +call their toll-free number at 1-800-772-1213. + +25 + + WORK OPPORTUNITIES FOR STUDENTS IN F-1 STATUS +The following is a list of all opportunities to obtain authorization to work in the +United States if you are in F-1 immigration status. Each category may allow for +authorization under different circumstances and at different times during your +stay in the U.S. This is summary of work benefits but does not include all regu- +lations governing employment. + +On-campus Work: Full-time students are allowed to work up to twenty hours per +week during the academic year. on campus employment means that you are paid by +Hobart and William Smith Colleges or working at a Colleges location with an em- +ployer that is providing services to students, such as Sodexho’s Saga dining hall or +the Café. on-campus employment requires an i-9 employment verification autho- +rization form to be completed by you at the Colleges business office. + +Off-campus Work Based on Financial Need: Students who have been in F-1 status +for one academic year and who are experiencing extreme financial difficulties due to +unforeseen changes that severely affect their financial resources may apply for work +permission. authorization is given by the dHS with the recommendation of the +international Student advisers/ dSo. + +Internship with an International Organization: Students in F-1 status are eligible +to apply for permission to work for international organizations such as the World +bank, the international monetary Fund, etc. authorization is granted by the dHS +with the recommendation of the international Student adviser/dSo. + +WORK OPPORTUNITIES FOR STUDENTS IN J-1 STATUS +J-1 On-Campus Employment +enrolled J-1 students may work on campus up to 20 hours per week during the +semester and full time during the break with the permission of their sponsor. if the +Colleges are your sponsor please see the international Student adviser. + +J-1 Off-Campus Employment +if a J-1 student needs employment during the academic year due to financial hard- +ship arising form unforeseen circumstances, the program sponsor may authorize it. +Such employment cannot interfere with full-time pursuit of the exchange visitor’s +program objective and is limited to 20 hours per week during the academic year and +full-time during break. + +PRACTICAL TRAINING: +this is defined as employment related to your field of study and is available dur- +ing and after completion of your studies. the following is a list of the categories of +practical training: + +Curricular practical training (before completion of your studies) is defined as + +employment pursuant to a required or optional internship which is an integral part +of the established curriculum of your school. + +26 + + optional practical training is defined as employment related to your field of study. +it is authorized in variable increments, deducted form the maximum of one year of +optional practical training per educational level, and is available in the following +variations: + +1. optional practical training after completion of your studies +2. optional practical training before completion of your studies with three pos- + +sible variations: + +a. practical training during the annual vacation. the annual vacation must + +follow at least two consecutive terms of fulltime enrollment and is usually taken in +the summer + +b. practical training when school is in session, provided that employment + +does not exceed 20 hours a week + +c. practical training following the completion of all the course requirements + +for the degree except the thesis or equivalent + +Optional Practical Training (OPT) Before Completion of Degree Requirements +definition: employment related to your field of study before completion of all degree +requirements. it is authorized in variable increments and is deducted from the +maximum of one year (12 months) of optional practical training per educational +level. there are three circumstances under which this type of practical training can +be authorized: + +1. Following completion of all degree requirements except the thesis or equivalent +2. during the annual vacation +3. Part-time (up to 20 hours per week) during the academic year + +Eligibility to Engage In OPT: You must: +• be a registered student in F-1 status physically in the United States at the time of +application +• have been enrolled in lawful student status on a full-time basis for at least one full +academic year +• not have used twelve months or more of full-time curricular practical training +oPt is permitted for a period of up to a maximum of twelve months per educational +level (bachelors, masters, etc.). + +Procedure: obtaining authorization for oPt is a two-step process. Your first step in +obtaining an employment authorization document (ead) is to qualify for a certifica- +tion from the international Student adviser/dSo recommending you for oPt. the +international Student adviser/dSo cannot authorize oPt - only the United States +Citizenship and immigration Services (USCiS) can do so by issuing you an ead. + +You may not begin employment until you have been issued an ead by the US- + +CiS and can work only within the time period indicated on the card. requests for +employment authorization must be mailed to the USCiS regional Service Center +having jurisdiction over your place of residence and may require 6 to 12 weeks for +adjudication. + +27 + + When to Apply: +• not more than 90 days prior to being enrolled for one full academic year· +• not more than 120 days prior to the employment start date that you select +Please note: the employment start date cannot be earlier than the date on which +you will have completed one full academic year. + +How to Apply: +Obtaining the International Student Adviser recommendation +Since you must first obtain a new SeViS i-20 with the international Student ad- +viser/dSo’s oPt endorsement, which may require 5 working days, you should make +every effort to apply as early as possible within the 90 or 120 day window. You might +need to come twice to the international Student adviser/dSo—once to submit the +application and another to pick up the approved recommendation on a SeViS i-20. + +Curricular Practical Training in F-1 Status +Definition: Curricular practical training (CPt) is employment pursuant to an intern- +ship requirement or internship elective which is an integral part of the established +curriculum of your school, which will be counted towards completion of your degree. + +Eligibility to engage in CPT: You must be a degree candidate who has been in +lawful status for at least one academic year. You must have been offered a specific +training opportunity that must fall into one of the following categories: +• to satisfy an internship requirement expected of all students in your program +and necessary for completion of your degree; +• to satisfy an internship elective which counts towards completion of your degree. +the internship elective must be listed in your school’s bulletin and must state in the +official course description that it is specifically designed to allow students enrolled in +the course to engage in off-campus internships; +• the internship may be done to engage in research necessary and integral to the +completion of the doctoral dissertation. + +Restrictions: U.S. Citizenship and immigration Services (USCiS) regulations do +not place an absolute limit on the amount of CPt you may utilize. However, if you +engage in one year or more of full-time CPt, you will lose your eligibility to engage +in optional Practical training. CPt is always part-time during the academic year +unless the student is engaged in doctoral dissertation research. CPt may be full time +during the annual vacation. You may engage in curricular practical training only +after your SeViS record has been updated by the international Student adviser and +the i-20 has been printed, indicating the approved employment. the endorsement +will indicate the precise details of the authorized training, including the name and +location of the employer, the specific dates of the training period, and whether it +is full-time or part-time. We encourage you to contact the international Student +adviser/dSo if you have questions. + +28 + + Departure Checklist + +q admissions letter from HWS + +q evidence of Financial resources + +q Passport with Student Visa + +q Form 1-20 or dS-2019 + +q Copies of medical records and other important documents translated into english + +q if J-1 visa holder – additional health insurance requirements. this must also be + +translated into english, if needed. + +q emergency outfit packed in carry-on luggage + +q Calling card to use during travel + +q emergency phone numbers (HWS: 1-315-781-3333) + +q minimum of 100 U.S. dollars for use in an emergency + +q Sweater or jacket in event of bad weather + +q if taking prescription medicine bring prescription for future refills (translated into + +english) + +q Photocopy of your passport, visa, etc. do not carry in the same place as your + +original documents. + +29 + + SECTION III: +*SAMPLE* ORIENTATION SCHEDULE *SAMPLE* + +Hobart and William Smith Colleges +international Students orientation +“let Your adventure begin!” +august 22 -25, 2006 + +Tuesday, August 22 +8:30 a.m. –7:00 p.m. + +Wednesday, August 23 +8:30 a.m. –12:00 p.m. + +arrival/Check-in +location: office of intercultural affairs +288 Pulteney Street +**bring all immigration paperwork + +arrival/Check-in +location: office of intercultural affairs, +288 Pulteney Street +**bring all immigration paperwork + +HWS Student id Cards +location: registrar’s office + +Campus tour + +Shopping @ Wal-mart & Wegmans Supermarket, +Van leaves from intercultural affairs parking lot + +12:30 p.m. - 1:30 p.m. + +lunch with deans of the Colleges +location: blue room in Scandling Center + +1:45 p.m. – 2:45 p.m. + +overview of library Services +location: Warren Hunting Smith library + +6:00 p.m. – 7:30 p.m. + +dinner +location: intercultural affairs Center + +Thursday, August 24 +7:30 a.m. – 8:00 a.m. + +8:15 a.m. 9:45 a.m. + +30 + +breakfast +location: Saga dining Hall in the blue room + +immigration regulations overview +location: Conference room, 2nd Floor, +intercultural affairs + + 10:00 a.m. – 11:00 a.m. + +11:05 a.m. – 11:35 a.m. + +on Campus employment/Social Security +elaine Ferrara, business office +location: tba + +benefits of the HWS Health insurance +ann brink, bursars office +location: tba + +12:00 p.m. – 1:30 p.m. + +lunch with Chaplain lesley adams +location: Chaplain’s residence, 630 S. main St. + +1:45 p.m. – 2:15 p.m. + +money matters +beth turner, director of Financial aid +location: tba + +2:30 p.m. – 3:00 p.m. + +a Healthy You +location: Hubbs Health Center + +3:15 p.m. – 3:45 p.m. + +5:30 p.m. – 7:30 p.m. + +Friday , August 25 +10:00 a.m. – 11:00 a.m. + +Study basics +Center for teaching & learning +location: Center for teaching & learning +(Ctl) + +Welcome dinner +President and mrs. mark gearan & Family +location: President’s House – 690 South main St. + +guide to U.S. education & Culture +amy teel and doug reilly, Center for +global education +location: delancey House + +12:00 p.m. – 1:00 p.m. + +lunch +location: delancey House + +1:30 p.m. + +Join regular campus wide orientation schedule + +Welcome to Hobart and William Smith Colleges + +31 diff --git a/q117/random_k4/random_3.md b/q117/random_k4/random_3.md index 568e4c67cff46c017e73fbb66f4b5bb6beac70e7..176646e2bfa5eda33dc9bb0858d570d16a45ac86 100644 --- a/q117/random_k4/random_3.md +++ b/q117/random_k4/random_3.md @@ -1,1466 +1,447 @@ -Features -• Compatible with MCS-51™ Products -(cid:127) 8K Bytes of In-System Reprogrammable Flash Memory -(cid:127) Endurance: 1,000 Write/Erase Cycles -(cid:127) Fully Static Operation: 0 Hz to 24 MHz -(cid:127) Three-level Program Memory Lock -(cid:127) -256 x 8-bit Internal RAM -(cid:127) 32 Programmable I/O Lines -(cid:127) Three 16-bit Timer/Counters -(cid:127) Eight Interrupt Sources -| (cid:127) Programmable Serial Channel | | | | | | | 8-bit | -| ------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | -(cid:127) -Low-power Idle and Power-down Modes -Microcontroller -Description -with 8K Bytes -The AT89C52 is a low-power, high-performance CMOS 8-bit microcomputer with 8K -Flash -bytes of Flash programmable and erasable read only memory (PEROM). The device -is manufactured using Atmel’s high-density nonvolatile memory technology and is -compatible with the industry-standard 80C51 and 80C52 instruction set and pinout. -The on-chip Flash allows the program memory to be reprogrammed in-system or by a -AT89C52 -conventional nonvolatile memory programmer. By combining a versatile 8-bit CPU -with Flash on a monolithic chip, the Atmel AT89C52 is a powerful microcomputer -which provides a highly-flexible and cost-effective solution to many embedded control -applications. -Not Recommended -| Pin Configurations | | | | | PDIP | | for New Designs. | -| ------------------ | ------------ | --------------------- | --------------------- | ------------ | ---- | ------------- | ----------------- | -| | PQFP/TQFP | | | (T2) P1.0 | 1 | 40 VCC | Use AT89S52. | -| | | | | (T2 EX) P1.1 | 2 | 39 P0.0 (AD0) | | -| | )XE 2T( 1.1P | )0DA( 0.0P )1DA( 1.0P | )2DA( 2.0P )3DA( 3.0P | P1.2 | 3 | 38 P0.1 (AD1) | | -)2T( 0.1P -| | | | | P1.3 | 4 | 37 P0.2 (AD2) | | -| --- | -------------- | --- | --- | ---- | --- | ------------- | --- | -| | 4.1P 3.1P 2.1P | CCV | | P1.4 | 5 | 36 P0.3 (AD3) | | -CN -| | | | | P1.5 | 6 | 35 P0.4 (AD4) | | -| ----------- | ----------------------------------- | --------------------------------- | --------------------- | ----------- | --- | ------------- | --- | -| | | | | P1.6 | 7 | 34 P0.5 (AD5) | | -| | 44 34 24 14 04 | 93 83 73 63 | 53 43 | P1.7 | 8 | 33 P0.6 (AD6) | | -| P1.5 | 1 | | 33 P0.4 (AD4) | RST | 9 | 32 P0.7 (AD7) | | -| | | | | (RXD) P3.0 | 10 | 31 EA/VPP | | -| P1.6 | 2 | | 32 P0.5 (AD5) | | | | | -| P1.7 | 3 | | 31 P0.6 (AD6) | (TXD) P3.1 | 11 | 30 ALE/PROG | | -| | | | | (INT0) P3.2 | 12 | 29 PSEN | | -| RST | 4 | | 30 P0.7 (AD7) | | | | | -| (RXD) P3.0 | 5 | | 29 EA/VPP | (INT1) P3.3 | 13 | 28 P2.7 (A15) | | -| | | | | (T0) P3.4 | 14 | 27 P2.6 (A14) | | -| NC | 6 | | 28 NC | | | | | -| (TXD) P3.1 | 7 | | 27 ALE/PROG | (T1) P3.5 | 15 | 26 P2.5 (A13) | | -| (INT0) P3.2 | 8 | | 26 PSEN | (WR) P3.6 | 16 | 25 P2.4 (A12) | | -| | | | | (RD) P3.7 | 17 | 24 P2.3 (A11) | | -| (INT1) P3.3 | 9 | | 25 P2.7 (A15) | | | | | -| (T0) P3.4 | 10 | | 24 P2.6 (A14) | XTAL2 | 18 | 23 P2.2 (A10) | | -| | | | | XTAL1 | 19 | 22 P2.1 (A9) | | -| iT1) P3.5 | 11 | | 23 P2.5 (A13) | | | | | -| | | | | GND | 20 | 21 P2.0 (A8) | | -| | 21 31 41 51 61 | 71 81 91 02 | 12 22 | | | | | -| | 6.3P )RW( 7.3P )DR( 2LATX 1LATX DNG | CN 0.2P )8A( 1.2P )9A( 2.2P )01A( | 3.2P )11A( 4.2P )21A( | | | | | -PLCC -| | | | | | )XE 2T( 1.1P | )0DA( 0.0P )1DA( 1.0P )2DA( 2.0P )3DA( 3.0P | | -| --- | --- | --- | --- | --- | ------------ | ------------------------------------------- | --- | -)2T( 0.1P -| | | | | 4.1P 3.1P | 2.1P CCV | | | -| --- | --- | --- | --- | --------- | -------- | --- | --- | -CN -| | | | | 6 5 | 4 3 2 1 44 | 34 24 14 04 | | -| --- | --- | --- | ----------- | ------------------- | ----------------------------- | ------------------------------------------ | --- | -| | | | | P1.5 7 | | 39 P0.4 (AD4) | | -| | | | | P1.6 8 | | 38 P0.5 (AD5) | | -| | | | | P1.7 9 | | 37 P0.6 (AD6) | | -| | | | | RST 10 | | 36 P0.7 (AD7) | | -| | | | (RXD) P3.0 | 11 | | 35 EA/VPP | | -| | | | | NC 12 | | 34 NC | | -| | | | (TXD) P3.1 | 13 | | 33 ALE/PROG | | -| | | | (INT0) P3.2 | 14 | | 32 PSEN | | -| | | | (INT1) P3.3 | 15 | | 31 P2.7 (A15) | | -| | | | (T0) P3.4 | 16 | | 30 P2.6 (A14) | | -| | | | (T1) P3.5 | 17 | | 29 P2.5 (A13) | | -| | | | | 81 91 | 02 12 22 32 42 | 52 62 72 82 | | -| | | | | 6.3P )RW( 7.3P )DR( | 2 LATX 1LATX DNG CN 0.2P )8A( | 1.2P )9A( 2.2P )01A( 3.2P )11A( 4.2P )21A( | | -Rev. 0313H–02/00 -1 - -Block Diagram -| | | | P0.0 | - P0.7 | P2.0 | - P2.7 | | -| --- | --- | --- | ---- | ------ | ---- | ------ | --- | -V -CC -| | | | PORT | 0 DRIVERS | PORT | 2 DRIVERS | | -| --- | --- | --- | ---- | --------- | ---- | --------- | --- | -GND -| | RAM | ADDR. | | | | | | -| --- | -------- | ----- | --- | ------ | ------ | ----- | --- | -| | | | RAM | PORT 0 | PORT 2 | QUICK | | -| | REGISTER | | | LATCH | LATCH | FLASH | | -PROGRAM -| | B | | | | STACK | | | -| --- | -------- | --- | --- | --- | ------- | --- | -------- | -| | | ACC | | | POINTER | | ADDRESS | -| | REGISTER | | | | | | REGISTER | -BUFFER -| | | TMP2 | | TMP1 | | | | -| --- | --- | ---- | --- | ---- | --- | --- | --- | -PC -| | | | ALU | | | | INCREMENTER | -| --- | --- | --- | --- | ---------- | ------------ | ----- | ----------- | -| | | | | INTERRUPT, | SERIAL | PORT, | | -| | | | | AND | TIMER BLOCKS | | | -PROGRAM -| | | | PSW | | | | COUNTER | -| --- | --- | --- | --- | --- | --- | --- | ------- | -PSEN -TI M IN G -| ALE/PROG | | INSTRUCTION | | | | | | -| -------- | ------- | ----------- | --- | --- | --- | --- | ---- | -| | A N D | | | | | | DPTR | -| EA / V | CONTROL | REGISTER | | | | | | -PP -RST -| | | | PORT | 1 | PORT | 3 | | -| --- | --- | --- | ----- | --- | ----- | --- | --- | -| | | | LATCH | | LATCH | | | -OSC -| | | | PORT 1 | DRIVERS | PORT 3 | DRIVERS | | -| --- | --- | --- | ------ | ------- | ------ | ------- | --- | -| | | | P1.0 | - P1.7 | P3.0 | - P3.7 | | -AT89C52 -2 - -AT89C52 -The AT89C52 provides the following standard features: 8K Port 2 -bytes of Flash, 256 bytes of RAM, 32 I/O lines, three 16-bit Port 2 is an 8-bit bi-directional I/O port with internal pullups. -timer/counters, a six-vector two-level interrupt architecture, The Port 2 output buffers can sink/source four TTL inputs. -a full-duplex serial port, on-chip oscillator, and clock cir- When 1s are written to Port 2 pins, they are pulled high by -cuitry. In addition, the AT89C52 is designed with static logic the internal pullups and can be used as inputs. As inputs, -for operation down to zero frequency and supports two Port 2 pins that are externally being pulled low will source -software selectable power saving modes. The Idle Mode current (I ) because of the internal pullups. -IL -stops the CPU while allowing the RAM, timer/counters, -Port 2 emits the high-order address byte during fetches -serial port, and interrupt system to continue functioning. -from external program memory and during accesses to -The Power-down mode saves the RAM contents but -external data memory that use 16-bit addresses (MOVX @ -freezes the oscillator, disabling all other chip functions until -DPTR). In this application, Port 2 uses strong internal pul- -the next hardware reset. -lups when emitting 1s. During accesses to external data -memory that use 8-bit addresses (MOVX @ RI), Port 2 -Pin Description -emits the contents of the P2 Special Function Register. -VCC Port 2 also receives the high-order address bits and some -Supply voltage. control signals during Flash programming and verification. -GND Port 3 -Ground. Port 3 is an 8-bit bi-directional I/O port with internal pullups. -The Port 3 output buffers can sink/source four TTL inputs. -Port 0 -When 1s are written to Port 3 pins, they are pulled high by -Port 0 is an 8-bit open drain bi-directional I/O port. As an the internal pullups and can be used as inputs. As inputs, -output port, each pin can sink eight TTL inputs. When 1s Port 3 pins that are externally being pulled low will source -are written to port 0 pins, the pins can be used as high- current (I ) because of the pullups. -IL -impedance inputs. -Port 3 also serves the functions of various special features -Port 0 can also be configured to be the multiplexed low- of the AT89C51, as shown in the following table. -order address/data bus during accesses to external pro- -Port 3 also receives some control signals for Flash pro- -gram and data memory. In this mode, P0 has internal -gramming and verification. -pullups. -Port 0 also receives the code bytes during Flash program- Port Pin Alternate Functions -ming and outputs the code bytes during program -P3.0 RXD (serial input port) -verification. External pullups are required during program -verification. P3.1 TXD (serial output port) -Port 1 P3.2 INT0 (external interrupt 0) -Port 1 is an 8-bit bi-directional I/O port with internal pullups. P3.3 INT1 (external interrupt 1) -The Port 1 output buffers can sink/source four TTL inputs. -P3.4 T0 (timer 0 external input) -When 1s are written to Port 1 pins, they are pulled high by -the internal pullups and can be used as inputs. As inputs, P3.5 T1 (timer 1 external input) -Port 1 pins that are externally being pulled low will source -P3.6 WR (external data memory write strobe) -current (I ) because of the internal pullups. -IL -P3.7 RD (external data memory read strobe) -In addition, P1.0 and P1.1 can be configured to be the -timer/counter 2 external count input (P1.0/T2) and the -RST -timer/counter 2 trigger input (P1.1/T2EX), respectively, as -Reset input. A high on this pin for two machine cycles while -shown in the following table. -the oscillator is running resets the device. -Port 1 also receives the low-order address bytes during -Flash programming and verification. ALE/PROG -Address Latch Enable is an output pulse for latching the -Port Pin Alternate Functions -low byte of the address during accesses to external mem- -P1.0 T2 (external count input to Timer/Counter 2), ory. This pin is also the program pulse input (PROG) during -clock-out Flash programming. -P1.1 T2EX (Timer/Counter 2 capture/reload trigger and In normal operation, ALE is emitted at a constant rate of 1/6 -direction control) the oscillator frequency and may be used for external -3 - -timingor clocking purposes. Note, however, that one ALE EA/VPP -pulse is skipped during each access to external data External Access Enable. EA must be strapped to GND in -memory. order to enable the device to fetch code from external pro- -gram memory locations starting at 0000H up to FFFFH. -If desired, ALE operation can be disabled by setting bit 0 of -Note, however, that if lock bit 1 is programmed, EA will be -SFR location 8EH. With the bit set, ALE is active only dur- -internally latched on reset. -ing a MOVX or MOVC instruction. Otherwise, the pin is -weakly pulled high. Setting the ALE-disable bit has no EA should be strapped to V for internal program -CC -effect if the microcontroller is in external execution mode. executions. -This pin also receives the 12-volt programming enable volt- -PSEN age (V ) during Flash programming when 12-volt -PP -Program Store Enable is the read strobe to external pro- -programming is selected. -gram memory. -XTAL1 -When the AT89C52 is executing code from external pro- -Input to the inverting oscillator amplifier and input to the -gram memory, PSEN is activated twice each machine -internal clock operating circuit. -cycle, except that two PSEN activations are skipped during -each access to external data memory. -XTAL2 -Output from the inverting oscillator amplifier. -Table 1. AT89C52 SFR Map and Reset Values -0F8H 0FFH -B -0F0H 0F7H -00000000 -0E8H 0EFH -ACC -0E0H 0E7H -00000000 -0D8H 0DFH -PSW -0D0H 0D7H -00000000 -T2CON T2MOD RCAP2L RCAP2H TL2 TH2 -0C8H 0CFH -00000000 XXXXXX00 00000000 00000000 00000000 00000000 -0C0H 0C7H -IP -0B8H 0BFH -XX000000 -P3 -0B0H 0B7H -11111111 -IE -0A8H 0AFH -0X000000 -P2 -0A0H 0A7H -11111111 -SCON SBUF -98H 9FH -00000000 XXXXXXXX -P1 -90H 97H -11111111 -TCON TMOD TL0 TL1 TH0 TH1 -88H 8FH -00000000 00000000 00000000 00000000 00000000 00000000 -P0 SP DPL DPH PCON -80H 87H -11111111 00000111 00000000 00000000 0XXX0000 -AT89C52 -4 - -AT89C52 -Special Function Registers -A map of the on-chip memory area called the Special Func- new features. In that case, the reset or inactive values of -tion Register (SFR) space is shown in Table 1. the new bits will always be 0. -Note that not all of the addresses are occupied, and unoc- Timer 2 Registers Control and status bits are contained in -cupied addresses may not be implemented on the chip. registers T2CON (shown in Table 2) and T2MOD (shown in -Read accesses to these addresses will in general return Table 4) for Timer 2. The register pair (RCAP2H, RCAP2L) -random data, and write accesses will have an indetermi- are the Capture/Reload registers for Timer 2 in 16-bit cap- -nate effect. ture mode or 16-bit auto-reload mode. -User software should not write 1s to these unlisted loca- Interrupt Registers The individual interrupt enable bits are -tions, since they may be used in future products to invoke in the IE register. Two priorities can be set for each of the -six interrupt sources in the IP register.r -Table 2. T2CON – Timer/Counter 2 Control Registe -T2CON Address = 0C8H Reset Value = 0000 0000B -Bit Addressable -Bit TF2 EXF2 RCLK TCLK EXEN2 TR2 C/T2 CP/RL2 -7 6 5 4 3 2 1 0 -Symbol Function -TF2 Timer 2 overflow flag set by a Timer 2 overflow and must be cleared by software. TF2 will not be set when either -RCLK = 1 or TCLK = 1. -EXF2 Timer 2 external flag set when either a capture or reload is caused by a negative transition on T2EX and -EXEN2=1. When Timer 2 interrupt is enabled, EXF2 = 1 will cause the CPU to vector to the Timer 2 interrupt -routine. EXF2 must be cleared by software. EXF2 does not cause an interrupt in up/down counter mode -(DCEN=1). -RCLK Receive clock enable. When set, causes the serial port to use Timer 2 overflow pulses for its receive clock in serial -port Modes 1 and 3. RCLK = 0 causes Timer 1 overflow to be used for the receive clock. -TCLK Transmit clock enable. When set, causes the serial port to use Timer 2 overflow pulses for its transmit clock in serial -port Modes 1 and 3. TCLK = 0 causes Timer 1 overflows to be used for the transmit clock. -EXEN2 Timer 2 external enable. When set, allows a capture or reload to occur as a result of a negative transition on T2EX -if Timer 2 is not being used to clock the serial port. EXEN2 = 0 causes Timer 2 to ignore events at T2EX. -TR2 Start/Stop control for Timer 2. TR2 = 1 starts the timer. -C/T2 Timer or counter select for Timer 2. C/T2 = 0 for timer function. C/T2 = 1 for external event counter (falling edge -triggered). -CP/RL2 Capture/Reload select. CP/RL2 = 1 causes captures to occur on negative transitions at T2EX if EXEN2 = 1. CP/RL2 -= 0 causes automatic reloads to occur when Timer 2 overflows or negative transitions occur at T2EX when EXEN2 -= 1. When either RCLK or TCLK = 1, this bit is ignored and the timer is forced to auto-reload on Timer 2 overflow. -Data Memory -The AT89C52 implements 256 bytes of on-chip RAM. The specifies whether the CPU accesses the upper 128 bytes -upper 128 bytes occupy a parallel address space to the of RAM or the SFR space. Instructions that use direct -Special Function Registers. That means the upper 128 addressing access SFR space. -bytes have the same addresses as the SFR space but are -For example, the following direct addressing instruction -physically separate from SFR space. -accesses the SFR at location 0A0H (which is P2). -When an instruction accesses an internal location above MOV 0A0H, #data -address 7FH, the address mode used in the instruction -5 - -Instructions that use indirect addressing access the upper input pin, T2. In this function, the external input is sampled -128 bytes of RAM. For example, the following indirect during S5P2 of every machine cycle. When the samples -addressing instruction, where R0 contains 0A0H, accesses show a high in one cycle and a low in the next cycle, the -the data byte at address 0A0H, rather than P2 (whose count is incremented. The new count value appears in the -address is 0A0H). register during S3P1 of the cycle following the one in which -MOV @R0, #data the transition was detected. Since two machine cycles (24 -Note that stack operations are examples of indirect oscillator periods) are required to recognize a 1-to-0 transi- -addressing, so the upper 128 bytes of data RAM are avail- tion, the maximum count rate is 1/24 of the oscillator -able as stack space. frequency. To ensure that a given level is sampled at least -once before it changes, the level should be held for at least -one full machine cycle. -Timer 0 and 1 -Timer 0 and Timer 1 in the AT89C52 operate the same way Capture Mode -as Timer 0 and Timer 1 in the AT89C51. -In the capture mode, two options are selected by bit -EXEN2 in T2CON. If EXEN2 = 0, Timer 2 is a 16-bit timer -Timer 2 or counter which upon overflow sets bit TF2 in T2CON. -This bit can then be used to generate an interrupt. If -Timer 2 is a 16-bit Timer/Counter that can operate as either -EXEN2 = 1, Timer 2 performs the same operation, but a 1- -a timer or an event counter. The type of operation is -to-0 transition at external input T2EX also causes the cur- -selected by bit C/T2 in the SFR T2CON (shown in Table 2). -rent value in TH2 and TL2 to be captured into RCAP2H and -Timer 2 has three operating modes: capture, auto-reload -RCAP2L, respectively. In addition, the transition at T2EX -(up or down counting), and baud rate generator. The -causes bit EXF2 in T2CON to be set. The EXF2 bit, like -modes are selected by bits in T2CON, as shown in Table 3. -TF2, can generate an interrupt. The capture mode is illus- -Timer 2 consists of two 8-bit registers, TH2 and TL2. In the trated in Figure 1. -Timer function, the TL2 register is incremented every -machine cycle. Since a machine cycle consists of 12 oscil- Auto-reload (Up or Down Counter) -lator periods, the count rate is 1/12 of the oscillator -Timer 2 can be programmed to count up or down when -frequency. -configured in its 16-bit auto-reload mode. This feature is -Table 3. Timer 2 Operating Modes invoked by the DCEN (Down Counter Enable) bit located in -the SFR T2MOD (see Table 4). Upon reset, the DCEN bit -RCLK +TCLK CP/RL2 TR2 MODE -is set to 0 so that timer 2 will default to count up. When -0 0 1 16-bit Auto-reload DCEN is set, Timer2 can count up or down, depending on -the value of the T2EX pin. -0 1 1 16-bit Capture -1 X 1 Baud Rate Generator -X X 0 (Off) -In the Counter function, the register is incremented in -response to a 1-to-0 transition at its corresponding external -AT89C52 +Colorado Division +of Insurance Annual +Complaint and +Recoveries Report + +FY 2023-24 (July 2023 - June 2024) + + Contents +Facts and Figures � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 4 + +Digging Deeper Into Issues � � � � � � � � � � � � � � � � � � � � � � � � � 6 + +Helping Coloradans Navigate Homeowners Insurance + 6 -AT89C52 -Figure 1. Timer in Capture Mode -OSC ÷12 -C/T2 = 0 -TH2 TL2 TF2 -OVERFLOW -CONTROL -TR2 -C/T2 = 1 -T2 PIN CAPTURE -RCAP2H RCAP2L -TRANSITION -TIMER 2 -DETECTOR -INTERRUPT -T2EX PIN EXF2 -CONTROL -EXEN2 -Figure 2 shows Timer 2 automatically counting up when the direction of the count. A logic 1 at T2EX makes Timer 2 -DCEN = 0. In this mode, two options are selected by bit count up. The timer will overflow at 0FFFFH and set the -EXEN2 in T2CON. If EXEN2 = 0, Timer 2 counts up to TF2 bit. This overflow also causes the 16-bit value in -0FFFFH and then sets the TF2 bit upon overflow. The RCAP2H and RCAP2L to be reloaded into the timer regis- -overflow also causes the timer registers to be reloaded with ters, TH2 and TL2, respectively. -the 16-bit value in RCAP2H and RCAP2L. The values in -A logic 0 at T2EX makes Timer 2 count down. The timer -Timer in Capture ModeRCAP2H and RCAP2L are preset -underflows when TH2 and TL2 equal the values stored in -by software. If EXEN2 = 1, a 16-bit reload can be triggered -RCAP2H and RCAP2L. The underflow sets the TF2 bit and -either by an overflow or by a 1-to-0 transition at external -causes 0FFFFH to be reloaded into the timer registers. -input T2EX. This transition also sets the EXF2 bit. Both the -The EXF2 bit toggles whenever Timer 2 overflows or -TF2 and EXF2 bits can generate an interrupt if enabled. -underflows and can be used as a 17th bit of resolution. In -Setting the DCEN bit enables Timer 2 to count up or down, -this operating mode, EXF2 does not flag an interrupt. -as shown in Figure 3. In this mode, the T2EX pin controls -7 - -Figure 2. Timer 2 Auto Reload Mode (DCEN = 0) -| OSC | | ÷12 | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -C/T2 = 0 -| | | | | | TH2 | TL2 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -OVERFLOW -CONTROL -TR2 -C/T2 = 1 -RELOAD -TIMER 2 -T2 PIN -INTERRUPT -| | | | | | RCAP2H | RCAP2L | | | -| --- | --- | --- | --- | --- | ------ | ------ | --- | --- | -TF2 -TRANSITION -DETECTOR -| T2EX PIN | | | | | | | EXF2 | | -| --------- | --- | --- | --- | --- | --- | --- | ---- | --- | -CONTROL -EXEN2 -Table 4. T2MOD – Timer 2 Mode Control Register -| T2MOD Address = 0C9H | | | | | | Reset Value = XXXX XX00B | | | -| -------------------- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | -Not Bit Addressable -| | – | – | – | – | – | – | T2OE | DCEN | -| ------- | ------------------------------------ | --- | --- | --- | --- | --- | ---- | ---- | -| Bit | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 | -| Symbol | Function | | | | | | | | -| – | Not implemented, reserved for future | | | | | | | | -| T2OE | Timer 2 Output Enable bit. | | | | | | | | -DCEN When set, this bit allows Timer 2 to be configured as an up/down counter. -AT89C52 +Answers and Education � � � � � � � � � � � � � � � � � � � � � � � � 7 + +Digging Deeper Into Issues � � � � � � � � � � � � � � � � � � � � � � � � � 8 + +Providing a Better Understanding of Health Insurance + 8 -AT89C52 -Figure 3. Timer 2 Auto Reload Mode (DCEN = 1) -(DOWN COUNTING RELOAD VALUE) TOGGLE -0FFH 0FFH -EXF2 -OSC ÷12 OVERFLOW -C/T2 = 0 -TH2 TL2 TF2 -CONTROL -TR2 -C/T2 = 1 TIMER 2 -INTERRUPT -T2 PIN -RCAP2H RCAP2L -COUNT -(UP COUNTING RELOAD VALUE) DIRECTION -1=UP -0=DOWN -T2EX PIN -Figure 4. Timer 2 in Baud Rate Generator Mode -TIMER 1 OVERFLOW -÷2 -"0" "1" -NOTE: OSC. FREQ. IS DIVIDED BY 2, NOT 12 -SMOD1 -OSC ÷2 -C/T2 = 0 -"1" "0" -TH2 TL2 -RCLK -Rx -CONTROL CLOCK -÷16 -TR2 -C/T2 = 1 -"1" "0" -T2 PIN -TCLK -RCAP2H RCAP2L Tx -CLOCK -TRANSITION ÷16 -DETECTOR -TIMER 2 -T2EX PIN EXF2 INTERRUPT -CONTROL -EXEN2 -9 - -Baud Rate Generator -Timer 2 is selected as the baud rate generator by setting increments every state time (at 1/2 the oscillator fre- -TCLK and/or RCLK in T2CON (Table 2). Note that the quency). The baud rate formula is given below. -baud rates for transmit and receive can be different if Timer -Modes 1 and 3 Oscillator Frequency -2 is used for the receiver or transmitter and Timer 1 is used ------ B ---- a ---- u --- d --- -- R ---- a --- t -- e -------- = - 3 --- 2 ----- × ------ [ -- 6 --- 5 ---- 5 --- 3 --- 6 ----- – ----- ( -- R ---- C ---- A ---- P ---- 2 ---- H ---- , - R ----- C ---- A ---- P ---- 2 ---- L --- ) --- ] -for the other function. Setting RCLK and/or TCLK puts -Timer 2 into its baud rate generator mode, as shown in Fig- where (RCAP2H, RCAP2L) is the content of RCAP2H and -ure 4. RCAP2L taken as a 16-bit unsigned integer. -The baud rate generator mode is similar to the auto-reload Timer 2 as a baud rate generator is shown in Figure 4. This -mode, in that a rollover in TH2 causes the Timer 2 registers figure is valid only if RCLK or TCLK = 1 in T2CON. Note -to be reloaded with the 16-bit value in registers RCAP2H that a rollover in TH2 does not set TF2 and will not gener- -and RCAP2L, which are preset by software. ate an interrupt. Note too, that if EXEN2 is set, a 1-to-0 -The baud rates in Modes 1 and 3 are determined by Timer transition in T2EX will set EXF2 but will not cause a reload -2’s overflow rate according to the following equation. from (RCAP2H, RCAP2L) to (TH2, TL2). Thus when Timer -2 is in use as a baud rate generator, T2EX can be used as -Modes 1 and 3 Baud Rates = T ---- i - m ----- e ---- r -- - 2 ---- - O ----- v --- e --- r -- f -- l -- o --- w ---- -- R ---- a ---- t - e --- an extra external interrupt. -16 -Note that when Timer 2 is running (TR2 = 1) as a timer in -The Timer can be configured for either timer or counter the baud rate generator mode, TH2 or TL2 should not be -operation. In most applications, it is configured for timer read from or written to. Under these conditions, the Timer is -operation (CP/T2 = 0). The timer operation is different for incremented every state time, and the results of a read or -Timer 2 when it is used as a baud rate generator. Normally, write may not be accurate. The RCAP2 registers may be -as a timer, it increments every machine cycle (at 1/12 the read but should not be written to, because a write might -oscillator frequency). As a baud rate generator, however, it overlap a reload and cause write and/or reload errors. The -timer should be turned off (clear TR2) before accessing the -Timer 2 or RCAP2 registers. -Figure 5. Timer 2 in Clock-out Mode -TL2 TH2 -OSC ÷2 (8-BITS) (8-BITS) -TR2 -RCAP2L RCAP2H -C/T2 BIT -P1.0 -÷2 -(T2) -T2OE (T2MOD.1) -TRANSITION -DETECTOR -P1.1 TIMER 2 -(T2EX) EXF2 INTERRUPT -EXEN2 -AT89C52 -10 - -AT89C52 -Programmable Clock Out the Timer 2 flag, TF2, is set at S2P2 and is polled in the -same cycle in which the timer overflows. -A 50% duty cycle clock can be programmed to come out on -P1.0, as shown in Figure 5. This pin, besides being a regu- Table 5. Interrupt Enable (IE) Register -lar I/O pin, has two alternate functions. It can be -(MSB) (LSB) -programmed to input the external clock for Timer/Counter 2 -or to output a 50% duty cycle clock ranging from 61 Hz to 4 EA – ET2 ES ET1 EX1 ET0 EX0 -MHz at a 16 MHz operating frequency. -Enable Bit = 1 enables the interrupt. -To configure the Timer/Counter 2 as a clock generator, bit -Enable Bit = 0 disables the interrupt. -C/T2 (T2CON.1) must be cleared and bit T2OE (T2MOD.1) -must be set. Bit TR2 (T2CON.2) starts and stops the timer. -Symbol Position Function -The clock-out frequency depends on the oscillator fre- -quency and the reload value of Timer 2 capture registers EA IE.7 Disables all interrupts. If EA=0, -(RCAP2H, RCAP2L), as shown in the following equation. no interrupt is acknowledged. If -EA = 1, each interrupt source is -Oscillator Fequency -Clock-Out Frequency= ------------------------------------------------------------------------------------------- individually enabled or disabled -4× [65536–(RCAP2H,RCAP2L)] -by setting or clearing its enable -In the clock-out mode, Timer 2 roll-overs will not generate bit. -an interrupt. This behavior is similar to when Timer 2 is -– IE.6 Reserved. -used as a baud-rate generator. It is possible to use Timer 2 -ET2 IE.5 Timer 2 interrupt enable bit. -as a baud-rate generator and a clock generator simulta- -neously. Note, however, that the baud-rate and clock-out ES IE.4 Serial Port interrupt enable bit. -frequencies cannot be determined independently from one -ET1 IE.3 Timer 1 interrupt enable bit. -another since they both use RCAP2H and RCAP2L. -EX1 IE.2 External interrupt 1 enable bit. -UART -ET0 IE.1 Timer 0 interrupt enable bit. -The UART in the AT89C52 operates the same way as the -EX0 IE.0 External interrupt 0 enable bit. -UART in the AT89C51. -User software should never write 1s to unimplemented bits, -Interrupts because they may be used in future AT89 products. -The AT89C52 has a total of six interrupt vectors: two exter- -Figure 6. Interrupt Sources -nal interrupts (INT0 and INT1), three timer interrupts -(Timers 0, 1, and 2), and the serial port interrupt. These -interrupts are all shown in Figure 6. -0 -Each of these interrupt sources can be individually enabled INT0 IE0 -1 -or disabled by setting or clearing a bit in Special Function -Register IE. IE also contains a global disable bit, EA, which -disables all interrupts at once. -Note that Table shows that bit position IE.6 is unimple- TF0 -mented. In the AT89C51, bit position IE.5 is also -unimplemented. User software should not write 1s to these -bit positions, since they may be used in future AT89 -0 -products. INT1 IE1 -1 -Timer 2 interrupt is generated by the logical OR of bits TF2 -and EXF2 in register T2CON. Neither of these flags is -cleared by hardware when the service routine is vectored -to. In fact, the service routine may have to determine -TF1 -whether it was TF2 or EXF2 that generated the interrupt, -and that bit will have to be cleared in software. TI -RI -The Timer 0 and Timer 1 flags, TF0 and TF1, are set at -S5P2 of the cycle in which the timers overflow. The values TF2 -EXF2 -are then polled by the circuitry in the next cycle. However, -11 - -Oscillator Characteristics is restored to its normal operating level and must be held -active long enough to allow the oscillator to restart and -XTAL1 and XTAL2 are the input and output, respectively, -stabilize. -of an inverting amplifier that can be configured for use as -an on-chip oscillator, as shown in Figure 7. Either a quartz -Figure 7. Oscillator Connections -crystal or ceramic resonator may be used. To drive the -device from an external clock source, XTAL2 should be left C2 -unconnected while XTAL1 is driven, as shown in Figure 8. XTAL2 -There are no requirements on the duty cycle of the external -clock signal, since the input to the internal clocking circuitry -is through a divide-by-two flip-flop, but minimum and maxi- -C1 -mum voltage high and low time specifications must be -XTAL1 -observed. -Idle Mode -In idle mode, the CPU puts itself to sleep while all the on- -GND -chip peripherals remain active. The mode is invoked by -software. The content of the on-chip RAM and all the spe- -cial functions registers remain unchanged during this -mode. The idle mode can be terminated by any enabled -interrupt or by a hardware reset. -Note: C1, C2 = 30 pF ± 10 pF for Crystals -Note that when idle mode is terminated by a hardware = 40 pF ± 10 pF for Ceramic Resonators -reset, the device normally resumes program execution -from where it left off, up to two machine cycles before the Figure 8. External Clock Drive Configuration -internal reset algorithm takes control. On-chip hardware -inhibits access to internal RAM in this event, but access to -the port pins is not inhibited. To eliminate the possibility of NC XTAL2 -an unexpected write to a port pin when idle mode is termi- -nated by a reset, the instruction following the one that -invokes idle mode should not write to a port pin or to exter- -nal memory. EXTERNAL -OSCILLATOR XTAL1 -Power-down Mode -SIGNAL -In the power-down mode, the oscillator is stopped, and the -instruction that invokes power-down is the last instruction GND -executed. The on-chip RAM and Special Function Regis- -ters retain their values until the power-down mode is -terminated. The only exit from power-down is a hardware -reset. Reset redefines the SFRs but does not change the -on-chip RAM. The reset should not be activated before V -CC -Status of External Pins During Idle and Powe-down Modes -Mode Program Memory ALE PSEN PORT0 PORT1 PORT2 PORT3 -Idle Internal 1 1 Data Data Data Data -Idle External 1 1 Float Data Address Data -Power-down Internal 0 0 Data Data Data Data -Power-down External 0 0 Float Data Data Data -AT89C52 -12 - -AT89C52 -Program Memory Lock Bits -V = 12V V = 5V -PP PP -The AT89C52 has three lock bits that can be left unpro- -Signature (030H) = 1EH (030H) = 1EH -grammed (U) or can be programmed (P) to obtain the -(031H) = 52H (031H) = 52H -additional features listed in the following table. -(032H) = FFH (032H) = 05H -Lock Bit Protection Modes The AT89C52 code memory array is programmed byte-by- -byte in either programming mode. To program any non- -Program Lock Bits -blank byte in the on-chip Flash Memory, the entire memory -LB1 LB2 LB3 Protection Type must be erased using the Chip Erase Mode. -1 U U U No program lock features. Programming Algorithm Before programming the -AT89C52, the address, data and control signals should be -2 P U U MOVC instructions executed -set up according to the Flash programming mode table and -from external program -Figure 9 and Figure 10. To program the AT89C52, take the -memory are disabled from -following steps. -fetching code bytes from -internal memory, EA is 1. Input the desired memory location on the address -sampled and latched on reset, lines. -and further programming of -2. Input the appropriate data byte on the data lines. -the Flash memory is disabled. -3. Activate the correct combination of control signals. -3 P P U Same as mode 2, but verify is -4. Raise EA/V to 12V for the high-voltage program- -also disabled. PP -ming mode. -4 P P P Same as mode 3, but external -5. Pulse ALE/PROG once to program a byte in the -execution is also disabled. -Flash array or the lock bits. The byte-write cycle is -self-timed and typically takes no more than 1.5 ms. -When lock bit 1 is programmed, the logic level at the EA pin -Repeat steps 1 through 5, changing the address -is sampled and latched during reset. If the device is pow- -and data for the entire array or until the end of the -ered up without a reset, the latch initializes to a random -object file is reached. -value and holds that value until reset is activated. The -latched value of EA must agree with the current logic level Data Polling The AT89C52 features Data Polling to indi- -at that pin in order for the device to function properly. cate the end of a write cycle. During a write cycle, an -attempted read of the last byte written will result in the com- -plement of the written data on PO.7. Once the write cycle -Programming the Flash -has been completed, true data is valid on all outputs, and -The AT89C52 is normally shipped with the on-chip Flash the next cycle may begin. Data Polling may begin any time -memory array in the erased state (that is, contents = FFH) after a write cycle has been initiated. -and ready to be programmed. The programming interface Ready/Busy The progress of byte programming can also -accepts either a high-voltage (12-volt) or a low-voltage be monitored by the RDY/BSY output signal. P3.4 is pulled -(V ) program enable signal. The Low-voltage program- low after ALE goes high during programming to indicate -CC -ming mode provides a convenient way to program the BUSY. P3.4 is pulled high again when programming is -AT89C52 inside the user’s system, while the high-voltage done to indicate READY. -programming mode is compatible with conventional third- -Program Verify If lock bits LB1 and LB2 have not been -party Flash or EPROM programmers. -programmed, the programmed code data can be read back -The AT89C52 is shipped with either the high-voltage or via the address and data lines for verification. The lock bits -low-voltage programming mode enabled. The respective cannot be verified directly. Verification of the lock bits is -top-side marking and device signature codes are listed in achieved by observing that their features are enabled. -the following table. -Chip Erase The entire Flash array is erased electrically by -using the proper combination of control signals and by -V = 12V V = 5V -PP PP -holding ALE/PROG low for 10 ms. The code array is written -Top-side Mark AT89C52 AT89C52 with all 1s. The chip erase operation must be executed -xxxx xxxx - 5 before the code memory can be reprogrammed. -yyww yyww -13 - -Reading the Signature Bytes The signature bytes are Programming Interface -read by the same procedure as a normal verification of -Every code byte in the Flash array can be written, and the -locations 030H, 031H, and 032H, except that P3.6 and -P3.7 must be pulled to a logic low. The values returned are entire array can be erased, by using the appropriate combi- -nation of control signals. The write operation cycle is self- -as follows. -timed and once initiated, will automatically time itself to -(030H) = 1EH indicates manufactured by Atmel -completion. -(031H) = 52H indicates 89C52 -(032H) = FFH indicates 12V programming All major programming vendors offer worldwide support for -(032H) = 05H indicates 5V programming the Atmel microcontroller series. Please contact your local -programming vendor for the appropriate software revision. -Flash Programming Modes -| Mode | | RST PSEN | ALE/PROG | EA/V | P2.6 | P2.7 P3.6 | P3.7 | -| ---- | --- | -------- | -------- | ---- | ---- | --------- | ---- | -PP -| Write Code Data | | H L | | H/12V | L | H H | H | -| --------------- | ------- | --- | --- | ----- | --- | --- | --- | -| Read Code Data | | H L | H | H | L | L H | H | -| Write Lock | Bit - 1 | H L | | H/12V | H | H H | H | -| | Bit - 2 | H L | | H/12V | H | H L | L | -| | Bit - 3 | H L | | H/12V | H | L H | L | -| Chip Erase | | H L | | H/12V | H | L L | L | -(1) -| Read Signature Byte | | H L | H | H | L | L L | L | -| ------------------- | --- | --- | --- | --- | --- | --- | --- | -Note: 1. Chip Erase requires a 10 ms PROG pulse. -AT89C52 -14 - -AT89C52 -Figure 9. Programming the Flash Memory Figure 10. Verifying the Flash Memory -| | | | | | +5V | | | | | +5V | -| --- | ----------- | ------- | ------- | --- | --- | ----------- | ------- | --- | ------- | -------- | -| | | | AT87F52 | | | | | | AT87F52 | | -| | | A0 - A7 | | | | | A0 - A7 | | | | -| | ADDR. | | P1 | V | | ADDR. | | P1 | V | | -| | | | | CC | | | | | CC | | -| | OOOOH/1FFFH | | | | | OOOOH/1FFFH | | | | PGM DATA | -PGM -| | | | P2.0 - P2.4 | P0 | | | | P2.0 | - P2.4 P0 | (USE 10K | -| --- | --- | -------- | ----------- | --- | ---- | --- | -------- | ---- | --------- | -------- | -| | | A8 - A12 | | | DATA | | A8 - A12 | | | | -PULLUPS) -| | | | P2.6 | | | | | P2.6 | | | -| ----------- | ----- | --- | ---- | --- | ---- | ----------- | --- | ---- | --- | --- | -| SEE | FLASH | | P2.7 | ALE | PROG | SEE FLASH | | P2.7 | ALE | | -| PROGRAMMING | | | | | | PROGRAMMING | | | | | -P3.6 -| MODES | TABLE | | P3.6 | | | MODES | TABLE | | | | -| ----- | ----- | --- | ---- | --- | --- | ----- | ----- | --- | --- | --- | -V -IH -| | | | P3.7 | | | | | P3.7 | | | -| --- | --- | --- | ----- | --- | ---- | --- | --- | ----- | --- | --- | -| | | | XTAL2 | EA | V /V | | | XTAL2 | EA | | -IH PP -| 3-24 | MHz | | | | | 3-24 MHz | | | | | -| ---- | --- | --- | ----- | ---- | --- | -------- | --- | ----- | ---- | --- | -| | | | XTAL1 | RST | V | | | XTAL1 | RST | V | -| | | | | | IH | | | | | IH | -| | | | GND | PSEN | | | | GND | PSEN | | -Flash Programming and Verification Characteristics - = 5.0 ± 10% -T = 0°C to 70°C, V -| A | | CC | | | | | | | | | -| ------ | --- | -------------------------- | --- | --- | --- | --- | ---- | --- | ---- | ----- | -| Symbol | | Parameter | | | | | Min | | Max | Units | -| V | (1) | Programming Enable Voltage | | | | | 11.5 | | 12.5 | V | -PP -(1) -| I PP | | Programming Enable Current | | | | | | | 1.0 | mA | -| ---- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| 1/t | | Oscillator Frequency | | | | | 3 | | 24 | MHz | -CLCL -| t | | Address Setup to PROG Low | | | | | 48t | | | | -| ------ | --- | ------------------------- | ------------------ | --- | --- | --- | -------- | --- | --- | --- | -| AVGL | | | | | | | CLCL | | | | -| t | | Address Hold after PROG | | | | | 48t | | | | -| GHAX | | | | | | | CLCL | | | | -| t | | Data Setup to PROG Low | | | | | 48t | | | | -| DVGL | | | | | | | CLCL | | | | -| t GHDX | | Data Hold After PROG | | | | | 48t CLCL | | | | -| t | | P2.7 (ENABLE) High to V | | | | | 48t | | | | -| EHSH | | | | PP | | | CLCL | | | | -| t | | V | Setup to PROG Low | | | | 10 | | | µs | -| SHGL | | | PP | | | | | | | | -(1) -| t | | V | Hold after PROG | | | | 10 | | | µs | -| ---- | --- | ---------- | ---------------- | --- | --- | --- | --- | --- | --- | --- | -| GHSL | | | PP | | | | | | | | -| t | | PROG Width | | | | | 1 | | 110 | µs | -GLGH -| t AVQV | | Address to Data Valid | | | | | | | 48t CLCL | | -| ------ | --- | ------------------------ | --- | --- | --- | --- | --- | --- | -------- | --- | -| t | | ENABLE Low to Data Valid | | | | | | | 48t | | -| ELQV | | | | | | | | | CLCL | | -| t | | Data Float after ENABLE | | | | | 0 | | 48t | | -| EHQZ | | | | | | | | | CLCL | | -| t | | PROG High to BUSY Low | | | | | | | 1.0 | µs | -GHBL -| t | | Byte Write Cycle Time | | | | | | | 2.0 | ms | -| --- | --- | --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -WC -| Note: | 1. Only used in 12-volt programming mode. | | | | | | | | | | -| ----- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -15 - -Flash Programming and Verification Waveforms - High-voltage Mode (V =12V) -PP -| | | PROGRAMMING | | VERIFICATION | | -| --- | --- | ----------- | --- | ------------ | --- | -P1.0 - P1.7 -| | | ADDRESS | | ADDRESS | | -| --- | --- | ------- | --- | ------- | --- | -P2.0 - P2.4 -t -AVQV -| PORT 0 | | DATA IN | | DATA OUT | | -| ------ | --- | ------- | --- | -------- | --- | -t t -DVGL GHDX -| | t | | t | | | -| --- | ---- | --- | ---- | --- | --- | -| | AVGL | | GHAX | | | -ALE/PROG -| | t | | t | | | -| --- | ---- | --- | ---- | --- | --- | -| | SHGL | t | GHSL | | | -GLGH -| | | V PP | LOGIC | 1 | | -| ---- | --- | ---- | ----- | --- | --- | -| EA/V | | | LOGIC | 0 | | -PP -(2) -| | t | | | | t | -| ---- | --- | ---- | ---- | --- | ---- | -| | | EHSH | t | | EHQZ | -| P2.7 | | | ELQV | | | -(ENABLE) -t -GHBL -P3.4 -| (RDY/BSY) | | | BUSY | READY | | -| --------- | --- | --- | ---- | ----- | --- | -t -WC -Flash Programming and Verification Waveforms - Low-voltage Mode (V =5V) -PP -| | | PROGRAMMING | | VERIFICATION | | -| --- | --- | ----------- | --- | ------------ | --- | -P1.0 - P1.7 -| P2.0 - P2.4 | | ADDRESS | | ADDRESS | | -| ----------- | --- | ------- | --- | ------- | --- | -t -AVQV -| PORT 0 | | DATA IN | | DATA OUT | | -| ------ | --- | ------- | --- | -------- | --- | -t t -DVGL GHDX -| | t | | t | | | -| --- | ---- | --- | ---- | --- | --- | -| | AVGL | | GHAX | | | -ALE/PROG -t -SHGL -t -GLGH -| | | | LOGIC | 1 | | -| --- | --- | --- | ----- | --- | --- | -EA/V -| PP | | | LOGIC | 0 | | -| --- | --- | --- | ----- | --- | --- | -t -| | | EHSH | | | t | -| ---- | --- | ---- | ---- | --- | ---- | -| | | | t | | EHQZ | -| P2.7 | | | ELQV | | | -(ENABLE) -t -GHBL -P3.4 -| (RDY/BSY) | | | BUSY | READY | | -| --------- | --- | --- | ---- | ----- | --- | -t -WC -AT89C52 -16 - -AT89C52 -Absolute Maximum Ratings* -Operating Temperature..................................-55°C to +125°C *NOTICE: Stresses beyond those listed under “Absolute -Maximum Ratings” may cause permanent dam- -Storage Temperature.....................................-65°C to +150°C age to the device. This is a stress rating only and -functional operation of the device at these or any -Voltage on Any Pin other conditions beyond those indicated in the -with Respect to Ground.....................................-1.0V to +7.0V operational sections of this specification is not -implied. Exposure to absolute maximum rating -Maximum Operating Voltage............................................6.6V conditions for extended periods may affect -device reliability. -DC Output Current......................................................15.0 mA -DC Characteristics -The values shown in this table are valid for T = -40°C to 85°C and V = 5.0V ± 20%, unless otherwise noted. -A CC -| Symbol | Parameter | Condition | | Min | Max | Units | -| ------ | ---------------------- | ------------------- | --- | ---------- | ---------- | ----- | -| V | Input Low-voltage | (Except EA) | | -0.5 | 0.2 V -0.1 | V | -| IL | | | | | CC | | -| V | Input Low-voltage (EA) | | | -0.5 | 0.2 V -0.3 | V | -| IL1 | | | | | CC | | -| V | Input High-voltage | (Except XTAL1, RST) | | 0.2 V +0.9 | V +0.5 | V | -| IH | | | | CC | CC | | -| V | Input High-voltage | (XTAL1, RST) | | 0.7 V | V +0.5 | V | -| IH1 | | | | CC | CC | | -Output Low-voltage(1) (Ports 1,2,3) -| V OL | | I OL = 1.6 mA | | | 0.45 | V | -| ---- | --- | -------------- | --- | --- | ---- | --- | -Output Low-voltage(1) -| V | | I = 3.2 mA | | | 0.45 | V | -| --- | --- | ----------- | --- | --- | ---- | --- | -| OL1 | | OL | | | | | -(Port 0, ALE, PSEN) -| V | Output High-voltage | I = -60 µA, V | = 5V ± 10% | 2.4 | | V | -| --- | ------------------- | -------------- | ----------- | --- | --- | --- | -| OH | | OH | CC | | | | -(Ports 1,2,3, ALE, PSEN) -| | | I = -25 µA | | 0.75 V | | V | -| --- | --- | ----------- | --- | ------ | --- | --- | -| | | OH | | CC | | | -| | | I = -10 µA | | 0.9 V | | V | -| | | OH | | CC | | | - = 5V ± 10% -| V OH1 | Output High-voltage | I OH = -800 µA, V | CC | 2.4 | | V | -| ----- | ------------------- | ------------------ | --- | --- | --- | --- | -(Port 0 in External Bus Mode) -| | | I = -300 µA | | 0.75 V | | V | -| --- | ------------------------------------- | ------------ | --- | ------ | --- | --- | -| | | OH | | CC | | | -| | | I = -80 µA | | 0.9 V | | V | -| | | OH | | CC | | | -| I | Logical 0 Input Current (Ports 1,2,3) | V = 0.45V | | | -50 | µA | -| IL | | IN | | | | | -I Logical 1 to 0 Transition Current V = 2V, V = 5V ± 10% -650 µA -| TL | | IN | CC | | | | -| --- | --- | --- | --- | --- | --- | --- | -(Ports 1,2,3) -I LI Input Leakage Current (Port 0, EA) 0.45 < V IN < V CC ±10 µA -KΩ -| RRST | Reset Pulldown Resistor | | | 50 | 300 | | -| ---- | ----------------------- | ---------------------------- | --- | --- | --- | --- | -| C | Pin Capacitance | Test Freq. = 1 MHz, T = 25°C | | | 10 | pF | -| IO | | | A | | | | -| I | Power Supply Current | Active Mode, 12 MHz | | | 25 | mA | -CC -| | | Idle Mode, 12 MHz | | | 6.5 | mA | -| --- | ------------------ | ----------------- | --- | --- | --- | --- | -| | Power-down Mode(1) | | | | 100 | µA | -V CC = 6V -| | | | | | 40 | µA | -| --- | --- | --- | --- | --- | --- | --- | -V = 3V -CC -Notes: 1. Under steady state (non-transient) conditions, I must be externally limited as follows: -OL -Maximum I per port pin: 10 mA -OL -Maximum I OL per 8-bit port: -Port 0: 26 mA Ports 1, 2, 3: 15 mA -Maximum total I for all output pins: 71 mA -OL -If I exceeds the test condition, V may exceed the related specification. Pins are not guaranteed to sink current greater -| | OL OL | | | | | | -| --- | ------ | --- | --- | --- | --- | --- | -than the listed test conditions. -| 2. | Minimum V CC for Power-down is 2V. | | | | | | -| --- | ----------------------------------- | --- | --- | --- | --- | --- | -17 - -AC Characteristics -Under operating conditions, load capacitance for Port 0, ALE/PROG, and PSEN = 100 pF; load capacitance for all other -outputs = 80 pF. -External Program and Data Memory Characteristics -| | | 12 MHz Oscillator | | Variable Oscillator | | | -| ------ | -------------------- | ----------------- | --- | ------------------- | --- | ----- | -| Symbol | Parameter | Min | Max | Min | Max | Units | -| 1/t | Oscillator Frequency | | | 0 | 24 | MHz | -CLCL -| t | ALE Pulse Width | 127 | | 2t -40 | | ns | -| ------ | --------------------------------- | --- | --- | ---------- | ----------- | --- | -| LHLL | | | | CLCL | | | -| t AVLL | Address Valid to ALE Low | 43 | | t CLCL -13 | | ns | -| t | Address Hold After ALE Low | 48 | | t -20 | | ns | -| LLAX | | | | CLCL | | | -| t | ALE Low to Valid Instruction In | | 233 | | 4t -65 | ns | -| LLIV | | | | | CLCL | | -| t | ALE Low to PSEN Low | 43 | | t -13 | | ns | -| LLPL | | | | CLCL | | | -| t | PSEN Pulse Width | 205 | | 3t -20 | | ns | -| PLPH | | | | CLCL | | | -| t PLIV | PSEN Low to Valid Instruction In | | 145 | | 3t CLCL -45 | ns | -| t | Input Instruction Hold after PSEN | 0 | | 0 | | ns | -PXIX -| t | Input Instruction Float after PSEN | | 59 | | t -10 | ns | -| ------ | ---------------------------------- | --- | --- | ------- | ------ | --- | -| PXIZ | | | | | CLCL | | -| t | PSEN to Address Valid | 75 | | t -8 | | ns | -| PXAV | | | | CLCL | | | -| t | Address to Valid Instruction In | | 312 | | 5t -55 | ns | -| AVIV | | | | | CLCL | | -| t PLAZ | PSEN Low to Address Float | | 10 | | 10 | ns | -| t | RD Pulse Width | 400 | | 6t -100 | | ns | -| RLRH | | | | CLCL | | | -| t | WR Pulse Width | 400 | | 6t -100 | | ns | -| WLWH | | | | CLCL | | | -| t | RD Low to Valid Data In | | 252 | | 5t -90 | ns | -| RLDV | | | | | CLCL | | -| t | Data Hold After RD | 0 | | 0 | | ns | -RHDX -| t RHDZ | Data Float After RD | | 97 | | 2t CLCL -28 | ns | -| ------ | --------------------------- | --- | --- | ---------- | ----------- | --- | -| t | ALE Low to Valid Data In | | 517 | | 8t -150 | ns | -| LLDV | | | | | CLCL | | -| t | Address to Valid Data In | | 585 | | 9t -165 | ns | -| AVDV | | | | | CLCL | | -| t | ALE Low to RD or WR Low | 200 | 300 | 3t -50 | 3t +50 | ns | -| LLWL | | | | CLCL | CLCL | | -| t | Address to RD or WR Low | 203 | | 4t -75 | | ns | -| AVWL | | | | CLCL | | | -| t QVWX | Data Valid to WR Transition | 23 | | t CLCL -20 | | ns | -| t | Data Valid to WR High | 433 | | 7t -120 | | ns | -| QVWH | | | | CLCL | | | -| t | Data Hold After WR | 33 | | t -20 | | ns | -| WHQX | | | | CLCL | | | -| t | RD Low to Address Float | | 0 | | 0 | ns | -RLAZ -| t | RD or WR High to ALE High | 43 | 123 | t -20 | t +25 | ns | -| ---- | ------------------------- | --- | --- | ----- | ----- | --- | -| WHLH | | | | CLCL | CLCL | | -AT89C52 -18 - -AT89C52 -External Program Memory Read Cycle -t -LHLL -ALE -t -PLPH -t t -AVLL t LLIV -LLPL -PSEN t PLIV -t -t PXAV -PLAZ t -t PXIZ -LLAX -t -PXIX -PORT0 A0-A7 INSTRIN A0-A7 -t -AVIV -PORT2 A8-A15 A8-A15 -External Data Memory Read Cycle -t -LHLL -ALE -t -WHLH -PSEN -t -LLDV -t -RLRH -t -LLWL -RD -t -LLAX -t t -t RLDV RHDZ -AVLL t -RLAZ t -RHDX -PORT 0 A0 - A7 FROM RI OR DPL DATA IN A0 - A7 FROM PCL INSTR IN -t -AVWL -t -AVDV -PORT 2 P2.0 - P2.7 OR A8 - A15 FROM DPH A8 - A15 FROM PCH -19 - -External Data Memory Write Cycle -t -LHLL -ALE -t -WHLH -PSEN -t t -LLWL WLWH -WR -t -LLAX -t t -| | | t QVWX | | WHQX | | | -| --- | --- | ------ | --- | ---- | --- | --- | -AVLL -t -QVWH -PORT 0 A0 - A7 FROM RI OR DPL DATA OUT A0 - A7 FROM PCL INSTR IN -t -AVWL -PORT 2 P2.0 - P2.7 OR A8 - A15 FROM DPH A8 - A15 FROM PCH -External Clock Drive Waveforms -t -CHCX -t -| | | CHCX | t | | | t | -| --- | --- | ---- | ---- | --- | --- | ---- | -| | | | CLCH | | | CHCL | -V - 0.5V -CC -0.7 V -CC -| | 0.2 V | - 0.1V | | | | | -| --- | ------ | ------- | --- | --- | --- | --- | -CC -0.45V -t -CLCX -t -CLCL -External Clock Drive -| Symbol | Parameter | | Min | | Max | Units | -| ------ | -------------------- | --- | --- | --- | --- | ----- | -| 1/t | Oscillator Frequency | | 0 | | 24 | MHz | -CLCL -| t | Clock Period | | 41.6 | | | ns | -| --- | ------------ | --- | ---- | --- | --- | --- | -CLCL -| t | High Time | | 15 | | | ns | -| --- | --------- | --- | --- | --- | --- | --- | -CHCX -| t | Low Time | | 15 | | | ns | -| --- | -------- | --- | --- | --- | --- | --- | -CLCX -| t CLCH | Rise Time | | | | 20 | ns | -| ------ | --------- | --- | --- | --- | --- | --- | -| t | Fall Time | | | | 20 | ns | -CHCL -AT89C52 -20 - -AT89C52 -.Serial Port Timing: Shift Register Mode Test Conditions -The values in this table are valid for V = 5.0V ± 20% and Load Capacitance = 80 pF. -CC -| | | | | | 12 MHz Osc | | | | Variable Oscillator | | | | -| ------ | ---------------------------------------- | --- | --- | --- | ---------- | --- | --- | --- | ------------------- | --- | ---- | ----- | -| Symbol | Parameter | | | | Min | | Max | | Min | | Max | Units | -| t | Serial Port Clock Cycle Time | | | | 1.0 | | | | 12t | | | µs | -| XLXL | | | | | | | | | CLCL | | | | -| t | Output Data Setup to Clock Rising Edge | | | | 700 | | | 10t | -133 | | | ns | -| QVXH | | | | | | | | | CLCL | | | | -| t | Output Data Hold After Clock Rising Edge | | | | 50 | | | 2t | -117 | | | ns | -| XHQX | | | | | | | | | CLCL | | | | -| t XHDX | Input Data Hold After Clock Rising Edge | | | | 0 | | | | 0 | | | ns | -| t | Clock Rising Edge to Input Data Valid | | | | | | 700 | | | 10t | -133 | ns | -| XHDV | | | | | | | | | | | CLCL | | -Shift Register Mode Timing Waveforms -| | INSTRUCTION | 0 | 1 | 2 | 3 | 4 | | 5 | 6 | 7 | 8 | | -| --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -ALE -t -XLXL -CLOCK -t -QVXH -t -XHQX -WRITETOSBUF -| | | | 0 | 1 | 2 | | 3 | 4 | 5 | 6 | 7 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -t -| | | | | | XHDX | | | | | | SETTI | | -| --- | ---------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | ----- | --- | -| | OUTPUTDATA | | t | | | | | | | | | | -XHDV -| | CLEARRI | | VALID | VALID | VALID | | VALID | VALID | VALID | | VALID | VALID | -| ------------------------------------ | --------- | --- | ----- | ----- | ------------------ | --- | ----- | ----- | ----- | --- | ----- | ----- | -| | INPUTDATA | | | | | | | | | | SETRI | | -| AC Testing Input/Output Waveforms(1) | | | | | Float Waveforms(1) | | | | | | | | -V - 0.5V -| CC | | 0.2 V + 0.9V | | | | | V | + 0.1V | | | V - | 0.1V | -| --- | --- | ------------------ | --- | --- | --- | --- | ---- | ------ | ----------------- | --- | --- | ---- | -| | | CC | | | | | LOAD | | | | OL | | -| | | TEST POINTS | | | | V | | | Timing Reference | | | | -LOAD -Points -| | | 0.2 V - 0.1V | | | | | V | - 0.1V | | | | | -| --- | ----- | -------------------- | --- | --- | --- | --- | ---- | ------ | --- | --- | --- | ---- | -| | | C C | | | | | LOAD | | | | V + | 0.1V | -| | 0.45V | | | | | | | | | | OL | | -Note: 1. AC Inputs during testing are driven at V - 0.5V Note: 1. For timing purposes, a port pin is no longer floating -CC -for a logic 1 and 0.45V for a logic 0. Timing measure- when a 100 mV change from load voltage occurs. A -ments are made at V IH min. for a logic 1 and V IL max. port pin begins to float when a 100 mV change from -| | for a logic 0. | | | | | | the loaded V | | /V level occurs. | | | | -| --- | -------------- | --- | --- | --- | --- | --- | ------------ | --- | ----------------- | --- | --- | --- | -OH OL -21 - -Ordering Information -| Speed | Power | | | | -| ----- | -------- | ------------- | ------- | --------------- | -| (MHz) | Supply | Ordering Code | Package | Operation Range | -| 12 | 5V ± 20% | AT89C52-12AC | 44A | Commercial | -| | | AT89C52-12JC | 44J | (0°C to 70°C) | -| | | AT89C52-12PC | 40P6 | | -| | | AT89C52-12QC | 44Q | | -| | | AT89C52-12AI | 44A | Industrial | -(-40°C to 85°C) -| | | AT89C52-12JI | 44J | | -| --- | -------- | ------------ | ---- | ---------- | -| | | AT89C52-12PI | 40P6 | | -| | | AT89C52-12QI | 44Q | | -| 16 | 5V ± 20% | AT89C52-16AC | 44A | Commercial | -(0°C to 70°C) -| | | AT89C52-16JC | 44J | | -| --- | --- | ------------ | ---- | --------------- | -| | | AT89C52-16PC | 40P6 | | -| | | AT89C52-16QC | 44Q | | -| | | AT89C52-16AI | 44A | Industrial | -| | | AT89C52-16JI | 44J | (-40°C to 85°C) | -| | | AT89C52-16PI | 40P6 | | -| | | AT89C52-16QI | 44Q | | -5V ± 20% -| 20 | | AT89C52-20AC | 44A | Commercial | -| --- | -------- | ------------ | ---- | --------------- | -| | | AT89C52-20JC | 44J | (0°C to 70°C) | -| | | AT89C52-20PC | 40P6 | | -| | | AT89C52-20QC | 44Q | | -| | | AT89C52-20AI | 44A | Industrial | -| | | AT89C52-20JI | 44J | (-40°C to 85°C) | -| | | AT89C52-20PI | 40P6 | | -| | | AT89C52-20QI | 44Q | | -| 24 | 5V ± 20% | AT89C52-24AC | 44A | Commercial | -(0°C to 70°C) -| | | AT89C52-24JC | 44J | | -| --- | --- | ------------ | ---- | ---------- | -| | | AT89C52-24PC | 40P6 | | -| | | AT89C52-24QC | 44Q | | -| | | AT89C52-24AI | 44A | Industrial | -(-40°C to 85°C) -| | | AT89C52-24JI | 44J | | -| --- | --- | ------------ | ---- | --- | -| | | AT89C52-24PI | 40P6 | | -| | | AT89C52-24QI | 44Q | | -Package Type -| 44A | 44-lead, Thin Plastic Gull Wing Quad Flatpack (TQFP) | | | | -| ---- | -------------------------------------------------------- | --- | --- | --- | -| 44J | 44-lead, Plastic J-leaded Chip Carrier (PLCC) | | | | -| 40P6 | 40-lead, 0.600" Wide, Plastic Dual Inline Package (PDIP) | | | | -| 44Q | 44-lead, Plastic Gull Wing Quad Flatpack (PQFP) | | | | -AT89C52 -22 - -AT89C52 -Packaging Information -44A, 44-lead, Thin (1.0 mm) Plastic Gull Wing Quad 44J, 44-lead, Plastic J-leaded Chip Carrier (PLCC) -| Flatpack (TQFP) | | | Dimensions in Inches and (Millimeters) | | | | -| --------------- | --- | --- | -------------------------------------- | --- | --- | --- | -Dimensions in Millimeters and (Inches)* -JEDEC STANDARD MS-018 AC -JEDEC STANDARD MS-026 ACB -.045(1.14) X 30° - 45° -1 2 . 2 1 ( 0 . 4 7 8 ) SQ .045(1.14) X 45° P IN N O . 1 . 0 1 2 ( . 3 0 5 ) -| PIN 1 ID | | 1 1 . 7 5 ( 0 . 4 5 8 ) | | ID E | N T IF Y | | -| -------- | --- | ----------------------- | --- | ---- | -------- | --- | -. 0 0 8 ( . 2 0 3 ) -| | | | | | .656(16.7) | .630(16.0) | -| --------------- | --- | ----------- | ---------- | --- | ---------- | ---------- | -| | | 0.45(0.018) | | | SQ | .590(15.0) | -| 0.80(0.031) BSC | | 0.30(0.012) | | | .650(16.5) | | -| | | | .032(.813) | | | .021(.533) | -.695(17.7) -| | | | .026(.660) | | SQ | .013(.330) | -| --- | --- | --- | ---------- | --- | --- | ---------- | -.685(17.4) -| | | | .050(1.27) TYP | | | .043(1.09) | -| --- | --- | --- | -------------- | --- | ----------------- | ---------- | -| | | | | | .500(12.7) REF SQ | .020(.508) | -.120(3.05) -.090(2.29) -| | | 10.10(0.394) SQ | | | | .180(4.57) | -| --- | --- | --------------- | --- | --- | --- | ---------- | -9.90(0.386) -.165(4.19) -1.20(0.047) MAX -0 -0.20(.008) 7 -0.09(.003) -.022(.559) X 45° MAX (3X) -0.75(0.030) 0.15(0.006) -0.45(0.018) 0.05(0.002) -Controlling dimension: millimeters -40P6, 40-lead, 0.600" Wide, Plastic Dual Inline 44Q, 44-lead, Plastic Quad Flat Package (PQFP) -| Package (PDIP) | | | Dimensions in Millimeters and (Inches)* | | | | -| -------------- | --- | --- | --------------------------------------- | --- | --- | --- | -Dimensions in Inches and (Millimeters) -JEDEC STANDARD MS-022 AB -2.07(52.6) -| | | | | 13.45 (0.525) | SQ | | -| --- | ---------- | --- | --- | ------------- | --- | --- | -| | 2.04(51.8) | PIN | | 12.95 (0.506) | | | -1 -PIN 1 ID -.566(14.4) -| | | .530(13.5) | | | 0.50 (0.020) | | -| --- | --- | ---------- | ---------------- | --- | ------------ | --- | -| | | | 0.80 (0.031) BSC | | 0.35 (0.014) | | -.090(2.29) -1.900(48.26) REF MAX -.220(5.59) -.005(.127) -MAX MIN -SEATING -PLANE -.065(1.65) -| .161(4.09) | | .015(.381) | | | | | -| ---------- | --- | ---------- | --- | --- | --- | --- | -.125(3.18) -| | | . 0 2 2 ( . 5 5 9 ) | | 10 . | 1 0 ( 0 . 3 9 4 ) | | -| --- | ---------- | ------------------- | --- | ---- | ---------------------- | --- | -| | .065(1.65) | | | 9 . | 9 0 ( 0 . 3 8 6 ) SQ | | -. 0 1 4 ( . 3 5 6 ) -| .110(2.79) | .041(1.04) | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -| .090(2.29) | .630(16.0) | | | | | | -2.45 (0.096) MAX -| | .590(15.0) | | | 0 | | | -| --- | ---------- | ----- | ------------ | --- | --- | --- | -| | | 0 REF | 0.17 (0.007) | 7 | | | -| | | 15 | 0.13 (0.005) | | | | -.012(.305) -.008(.203) -| | .690(17.5) | | | 1.03 (0.041) | | | -| --- | ---------- | --- | --- | ------------ | ---------------- | --- | -| | .610(15.5) | | | | 0.25 (0.010) MAX | | -0.78 (0.030) -Controlling dimension: millimeters -23 - -Atmel Headquarters Atmel Operations -Corporate Headquarters Atmel Colorado Springs -2325 Orchard Parkway 1150 E. Cheyenne Mtn. Blvd. -San Jose, CA 95131 Colorado Springs, CO 80906 -TEL (408) 441-0311 TEL (719) 576-3300 -FAX (408) 487-2600 FAX (719) 540-1759 -Europe Atmel Rousset -Atmel U.K., Ltd. Zone Industrielle -Coliseum Business Centre 13106 Rousset Cedex -Riverside Way France -Camberley, Surrey GU15 3YL TEL (33) 4-4253-6000 -England FAX (33) 4-4253-6001 -TEL (44) 1276-686-677 -FAX (44) 1276-686-697 -Asia -Atmel Asia, Ltd. -Room 1219 -Chinachem Golden Plaza -77 Mody Road Tsimhatsui -East Kowloon -Hong Kong -TEL (852) 2721-9778 -FAX (852) 2722-1369 -Japan -Atmel Japan K.K. -9F, Tonetsu Shinkawa Bldg. -1-24-8 Shinkawa -Chuo-ku, Tokyo 104-0033 -Japan -TEL (81) 3-3523-3551 -FAX (81) 3-3523-7581 -Fax-on-Demand -North America: -1-(800) 292-8635 -International: -1-(408) 441-0732 -e-mail -literature@atmel.com -Web Site -http://www.atmel.com -BBS -1-(408) 436-4309 -© Atmel Corporation 1999. -Atmel Corporation makes no warranty for the use of its products, other than those expressly contained in the Company’s standard war- -ranty which is detailed in Atmel’s Terms and Conditions located on the Company’s web site. The Company assumes no responsibility for -any errors which may appear in this document, reserves the right to change devices or specifications detailed herein at any time without -notice, and does not make any commitment to update the information contained herein. No licenses to patents or other intellectual prop- -erty of Atmel are granted by the Company in connection with the sale of Atmel products, expressly or by implication. Atmel’s products are -not authorized for use as critical components in life support devices or systems. -Marks bearing ® and/or ™ are registered trademarks and trademarks of Atmel Corporation. -Printed on recycled paper. -Terms and product names in this document may be trademarks of others. -0313H–02/00/xM \ No newline at end of file +Out-of-Network Labs � � � � � � � � � � � � � � � � � � � � � � � � � � 8 + +Helping Consumers Understand Long-Term Care Insurance +Policy Options � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 9 + +Consumers’ Stories � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 10 + +Delay in Hail Damage Claim � � � � � � � � � � � � � � � � � � � � 10 + +Transplant Results in Out-of-Network Billing � � � � � � � � � 10 + +Smoke Damage from Marshall Fire � � � � � � � � � � � � � � � � 11 + +Following Through on Network Exception � � � � � � � � � � � 12 + +Improper Delay for Life Insurance Claim � � � � � � � � � � � � 12 + +Travel Insurance Not Paying Hospital Bill � � � � � � � � � � � � 13 + +Getting Medical Claims Paid � � � � � � � � � � � � � � � � � � � � 13 + +Contact the Colorado Division of Insurance � � � � � � � � � � � 14 + +2 | + + Colorado Division of Insurance Annual +Complaint and Recoveries Report + +FY 2023-24 (July 2023 - June 2024) + +The Colorado Division of Insurance (DOI), part of the Department of Regulatory Agencies +(DORA), regulates insurance in Colorado - the insurance companies as well as insurance +agents/brokers. But the DOI is also a resource for Coloradans on insurance matters. We +answer consumers’ questions, investigate consumers’ complaints and generally explain +the world of insurance to the public. Most of this work is done by the Consumer Services +Team, which is split into two components - Property & Casualty Insurance and Life & Health +Insurance. The Consumer Services Team also works with other DOI teams as necessary, +especially when systemic issues with insurance companies are identified through complaints +received and interactions with consumers. + +This report covers the consumer complaints submitted to the DOI, the work in investigating +the complaints, and the money recovered for consumers throughout the 2023-24 fiscal year +(FY 2023-24; July 2023 - June 2024). During that fiscal year, the Consumer Services Team +received more complaints as compared to the previous year, and recovered more money +for Coloradans. + +For the Property & Casualty section, complaints were mostly centered around homeowners +and auto insurance. This group not only investigated and resolved these complaints, but also +worked to educate consumers regarding the current challenges in the homeowners insurance +market and served as a valuable resource by providing assistance to consumers in Colorado. + +The Life & Health group received fewer complaints compared to the previous year. However, +they investigated nearly 2,000 complaints while helping consumers navigate their health +insurance. + +Ensuring access to care for Colorado consumers continued to be top of mind for the DOI in +working through consumers’ questions and complaints. + + | 3 + + Money Recovered for +Colorado Consumers + +Facts and +Figures + +This is money that goes to consumers in situations where +the DOI finds that an insurance company improperly +denied a claim, or did not initially pay the correct claim +amount. Recoveries can also come about when a company +is delaying payment on a claim or has not followed State +insurance law and regulations. + +Property & Casualty Insurance + +Total Money +Recovered +for Colorado +Consumers +$26,487,192 + +(an increase over the previous +year’s total of $21,523,066) + +Money recovered +on homeowners +insurance complaints + +$10,601,710 + +Money recovered +on auto insurance +complaints +$4,995,340 + +$19,253,299 +total recovered + +Life & Health Insurance + +Money recovered +on health insurance +complaints + +$4,080,002 + +Money recovered on +life insurance and +annuity complaints +$3,152,428 + +$7,232,430 +total recovered + +4 | Facts and Figures + + Facts and Figures + +Property & Casualty +Insurance Complaints + +Life & Health +Insurance Complaints + +Total Property & +Casualty Insurance +Complaints Received + +3,829 + +1,929 + +Total Life & Health +Insurance Complaints +Received + +Complaints +specifically related +to homeowners +insurance + +1,504 + +Complaints +and Inquiries +Received +6,987 + +Complaints and +Inquiries Closed + 7,327 + +1,716 + +Complaints +specifically related +to health insurance + +Note: Some complaints take longer than others to fully +investigate, so not all complaints are resolved in the year they +are received. This figure also includes 2,218 inquiries, which +are instances when a consumer’s complaint involves an easily +resolved question or falls outside of the DOI’s authority to +regulate. These complaints still require the Consumer Services +Team to contact the consumer to answer their questions and +educate them on their options. + +14,741 + +Telephone Calls: 14,741 + +Calls to the Consumer Services Team can include simple +questions regarding a claim or insurance policy, inquiries about +a company’s actions which may violate State insurance laws or +how to submit a formal complaint to the DOI + +Facts and Figures | 5 + + DOI’s Work + +Digging +Deeper +Into Issues + +Helping Coloradans Navigate +Homeowners Insurance + +Across the country, including here in Colorado, +people are working through challenges with +their homeowners insurance. The DOI has been +implementing a number of ways to address +these issues, including legislation like the FAIR +Plan (an insurer of last resort), market studies +and regular stakeholder meetings to hear from +consumers and share information about the +DOI’s efforts. Our Consumer Services Team has +also been working through these challenges +with individual consumers. + +6 | DOI’s Work - Digging Deeper Into Issues + + Digging Deeper Into Issues + +Answers and Education + +In the last year, the Team fielded about 570 calls +and emails from Colorado consumers concerned +about the affordability and availability of +homeowners insurance, as well as non-renewals, +which represented about 35% of all the calls and +emails the team received specifically related to +homeowners insurance. + +With each of these inquiries, a Team member takes +the time necessary to educate the consumer about +the factors leading to premium increases, the +underwriting decisions of insurance companies, +and the claims process, while also providing +resources to help find coverage and suggestions +to possibly lower premiums. + +DOI’s Work - Digging Deeper Into Issues | 7 + + DOI’s Work + +Digging +Deeper +Into Issues + +Providing a Better +Understanding of Health +Insurance + +In recent years, the DOI developed various programs +that increase access to health insurance and save +people money on health care. However people can +still run into problems and roadblocks when using +that health insurance. The Consumer Services Team +strives to not only answer people’s questions and +investigate their complaints, but also works to +help people gain a better understanding of their +coverage and benefits, and help them navigate the +health care system. + +Out-of-Network Labs + +Over the last year, the Consumer Services Team has seen an +uptick in complaints about physicians sending blood draws +and other lab work to out-of-network labs. Generally, this +shouldn’t occur, as in-network physicians are required by +their contract with insurance companies to use labs that +are also in-network. In addition to investigating each +consumer complaint to ensure that consumers are receiving +the benefits promised by their insurance policies, the Team +is currently working with insurance companies to address +this challenge and ensure members’ labs are being sent to +in-network providers by their contracted care teams. + +8 | DOI’s Work - Digging Deeper Into Issues + + Digging Deeper Into Issues + +Helping Consumers Understand Long-Term Care +Insurance Policy Options + +In this past year, the Consumer Services Team successfully +helped many consumers understand their Long–Term Care +(LTC) policies, as well as what options are available to +them when they experience a rate increase. The Team also +worked to gain a deeper knowledge about how LTC insurance +companies interpret their policies and determine payouts to +their policyholders. Through this effort, the Team identified +a specific policy option that could help many Coloradans. +An “Alternate Plan of Care” option is designed to provide +consideration for care that would not normally be covered +under LTC policies. And while this policy option can only be +accessed in very specific circumstances, the Team was able to +identify it as an option for specific consumers that had filed +complaints concerning coverage that was dependent on the +type of facility in which the care was received. In some cases +this allowed the policyholders to access policy benefits without +having to change the facility in which they were residing. + +DOI’s Work - Digging Deeper Into Issues | 9 + + Consumers’ +Stories + +10 | Consumers' Stories + +Delay in Hail Damage Claim + +The owners of an apartment complex filed a complaint +after an extensive delay in getting payment on a claim +regarding extensive hail damage to the complex. +The hail claim was filed on July 24, 2023, yet as of +November 2023, the insurance company had still not +paid the settlement amount. After the apartment +complex filed a complaint with the DOI on November 1, +the DOI contacted the insurance company, prompting +the company to start working with the complex to +negotiate the settlement. By November 22, 2023, the +insurance company had paid the settlement amount +of $2,657,730. + +Transplant Results in Out-of-Network +Billing + +The Team worked with a consumer with medical bills +over $420,000 due to a transplant. The consumer had +been insured by Friday Health, and even though the +company had made the appropriate arrangements +with the out-of-network health care providers, Friday +Health had been liquidated throughout the country +and put into receivership. The Consumer Services +Team worked with the providers to pause bills to +the consumer while working through the transplant +claims and provider payments. The Team also worked +with an outside advocacy agency, and ultimately the +consumer was not billed for anything over in-network +charges. + + Consumers’ Stories + +Smoke Damage from Marshall Fire + +Unfortunately, some consumers are still working through +issues resulting from the December 2021 Marshall Fire. +One consumer filed a complaint in November 2022, as their +insurance company had still not remediated the smoke +damage to their home. While the company did agree to +do some cleaning, it was not to the extent the consumer +needed. With the Consumer Services Team’s involvement, +the company eventually paid an additional $122,197 to +properly clean and remediate the smoke damage to the +home, and the Team was able to close the complaint in +August 2023. + +Consumers' Stories | 11 + + Consumers’ Stories + +Following Through on Network Exception + +With health insurance, issues around provider networks are +common. When Coloradans run into problems with a provider +network being inadequate in rural areas or regarding speciality +care and speciality providers, consumers will often require what is +called a Provider Network Gap Exception so they can get the care +they need. The Consumer Services Team makes sure the insurance +companies follow through and approve these exceptions as +necessary. In the last year, the Team worked through a consumer’s +exception that had been approved by Oscar Health. However, this +exception was complicated by the fact that Oscar had not set +up the appropriate agreement, called a Single Case Agreement +(SCA), with the out-of-network speciality provider. Additionally, +Oscar Health had exited the Colorado market. As a result, the +consumer was billed for over $30,000. The Team intervened to +ensure that Oscar Health processed the claim correctly, and the +consumer was covered at no greater cost to the consumer than +if the services and treatment were obtained from an in-network +provider. + +Improper Delay for Life Insurance Claim + +The DOI often needs to intervene when a life insurance policy +payout is delayed. In one particular case, a person was expecting +a payout of over $2 million from two policies. The claims were +filed on March 10, 2023, and the consumer battled with the +insurance company over the next three months to get this settled. +Even after the company received a coroner report, it did not +issue the payout. Finally, the consumer filed a complaint with +the DOI on July 14, 2023. After a DOI investigation, the company +processed the claim and made the payments by July 26, including +over $14,000 in interest. + +12 | Consumers' Stories + + Consumers’ Stories + +Travel Insurance Not Paying Hospital Bill + +A Colorado consumer purchased travel insurance, which included medical +coverage, for an overseas trip. A health emergency occurred early in the trip. In +January 2023, the foreign hospital billed for services in the amount of $48,656. +Yet, as of November 2023, the travel insurance had still not paid for this bill. +The consumer submitted a complaint to the DOI on November 28. The travel +insurance company was attempting to renegotiate the bill, however, based on +the DOI’s involvement, they wrapped up their review of the claim and paid the +full amount to the hospital by late December 2023. + +Getting Medical Claims Paid + +For many people, struggling with traumatic +medical problems can mean they do not have the +energy to navigate the health care system. The +Consumer Services Team worked with a consumer +whose wife was involved in a car accident that +resulted in a traumatic brain injury. She had been +moved multiple times from one facility to another +and was having claims denied. The Team worked +with the insurance company to get a better +understanding of the claims, as well as the claim +denials. Eventually, our Team ensured the claims +were paid, providing an immense sense of relief +for the family. + +Consumers' Stories | 13 + + Contact the Colorado +Division of Insurance + +If you have questions, concerns or complaints about your insurance, +insurance company or insurance agent, or you just need an explanation +of terminology, processes or the world of insurance in general, please +contact the Colorado Division of Insurance Consumer Services Team. + +303-894-7490 + +DORA_Insurance@state.co.us + +doi.colorado.gov (click on “File a Complaint”) + diff --git a/q117/random_k4/random_4.md b/q117/random_k4/random_4.md index 850bf0ca35ecd76a89f5f4d9b0ba198abf068312..b1f264dec8c881060a31fd51402112360988391b 100644 --- a/q117/random_k4/random_4.md +++ b/q117/random_k4/random_4.md @@ -1,1799 +1,633 @@ -Partners: -Microsoft -Azure - -Implementing -Hybrid Cloud -with MS Azure - -White paper - -Shaping -the future -of digital -business - -01 - -Implementing -Hybrid Cloud -with Microsoft -Azure - B - -How GFT and Azure -can help your organization -implement a strategic -hybrid model that is -cost effective and offers -sustainable benefits. - - Introduction - -A - -In a diverse world where -technology is in constant flux, many -organisations have computing -solutions that straddle cloud and on -premises. While both environments -have an important role to play today -and in the foreseeable future, a -consistent approach is crucial. - -For many organisations, integrating -these two to create a hybrid -model is essential but there are -many business and technical -considerations. - -This paper evaluates how Microsoft -Azure can help any organisation -implement a strategic hybrid model -that is cost effective and offers -sustainable benefits. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - 2 - -3 - -Why use a hybrid cloud? - -A - -Cloud computing gets a lot of attention, and for good -reason: it is where much of IT is going. However, -on-premises datacentres also have an important -role to play, both today and in the future. For many -organisations, integrating these to create a hybrid -cloud is essential. - -ASSESS -Why move to the cloud and -are you ready? - -OPTIMIZE -How can yiu make most -of the cloud? - -MIGRATE -What will be your -migratiion strategy? - -OPERATE -Have you considered -the impact on your support -and maintance teams? - -Microsoft understands this reality. -To help you to achieve it, they are offering -a broad range of cloud and on-premises -technologies that work together in a -coherent way. Moreover, unlike many other -vendors in the market, Microsoft provides -the flexibility to let you choose the path -that is right for your organisation. Microsoft -is committed to providing a consistent -hybrid cloud that supports every strategy. -So, what exactly is a hybrid cloud? -While getting everyone to agree on a -definition is not easy, there are some -obvious requirements. For example, you -need a way to connect your on-premises -datacentres with the cloud, a problem -Microsoft solves with Azure Virtual -Networks, Azure ExpressRoute. However, -basic connectivity is not enough; a hybrid -cloud must extend beyond this to provide -complete set of consistent services. -Microsoft believes that a true hybrid cloud -must provide four components, each of -which brings significant benefits. - -SECURE -Concerned the cloud isn‘t -secure? - -These are: - ■ Common identity for on-premises - -and cloud applications. This improves -productivity by giving users single -sign-on to all of their applications. - - ■ Integrated management and security -across the hybrid cloud. This offers a -cohesive way to monitor, manage, and -help secure the environment, giving -increased visibility and control. - - ■ A consistent data platform for the -datacentre and the cloud. This -ensures data portability, combined -with seamless access to on-premises -and cloud data services for deep data -insight. - - ■ Unified development and DevOps -across the cloud and on-premises -datacentres. This lets the client -move applications between the -two environments as needed, and -it improves developer productivity, -as both places now have the same -development environment. - -— - Taken together, these -four requirements provide -consistent experiences -for developers, data -professionals, IT -managers, and users. -Figure 1 summarizes this, -showing example -Microsoft technologies -for each. - -Azure Active -Directory - -Azure Management -and Security - -Azure -Data Services - -Azure -Services - -Common -Identity - -Integrated -Management -and Security - -Consistent -Data -Platform - -Unified -Development -and DevOps - -Active -Directory - -On-premises -Infrastructure - -SQL Server - -Azure Stack - -Figure 1: A hybrid cloud must provide consistent solutions in four areas: identity; -management and security; data platform; development and DevOps. - -As the figure shows, Microsoft offers hybrid cloud technologies that address all four areas. -Examples of these include: - - ■ Azure Active Directory works with the on-premises Active Directory to provide - -common identity for users. - - ■ Azure provides integrated management and security services for both cloud and on- - -premises infrastructure. - - ■ Azure data services combine with SQL Server to create a consistent data platform. - - ■ Microsoft Azure services in the cloud combined with Microsoft Azure Stack on- - -premises provide unified development and DevOps. - -What happens if you attempt to create a hybrid cloud without these four attributes? -The short answer is pain: you will have needless differences throughout your -environment. These differences bring complexity, which in turn makes the hybrid cloud -harder to use, harder to manage, and harder to secure. Risks increase while user benefits -shrink. - -Creating a truly consistent hybrid cloud, with the advantages it brings, is a better -approach. As this paper describes, Microsoft stands alone in offering these advantages. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - - 4 - -5 - -01. -Requirement: common identity -A - -When users access applications, they should not need to worry -about whether those applications are running on-premises or in the -cloud. Providing consistent identity is fundamental to achieving this, -which is why Microsoft created Azure Active Directory (Azure AD). -Among other benefits, cloud service offers secure single sign-on and -automated provisioning of new users. - -To use Azure AD in a hybrid cloud, an -organization first connects its on-premises -Active Directory to Azure AD in the cloud. -Users can then sign in as usual – Azure -AD is invisible to them. Those users can -now access both on-premises applications -and cloud applications without signing -in again. They enjoy single sign-on -throughout their hybrid world. -Azure AD supports cloud applications -from Microsoft, including Office 365 and -Dynamics 365. It also supports many -other SaaS offerings, including Google -Apps, Salesforce CRM, Dropbox, Box, -Slack, Service Now, Workday, and many -more. Just as Active Directory allows -single sign-on to on-premises applications -from many different vendors, Azure AD -provides this for cloud apps from many -SaaS providers. - -01.1 -Give your users single -sign-on to applications -anywhere - -Everyone hates having to remember -different passwords. In enterprises, the -problem of providing single sign-on – -the ability to log in once to access any -application – was solved long ago. With -the rise of Software as a Service (SaaS), -however, this problem must be solved -again. Rather than make users sign in -separately to each application, they -should be able to sign in just once, and -then access both on-premises and cloud -(that is, SaaS) applications. - -Figure 2 shows how Azure AD makes this -possible. - -Azure Active -Directory - -Cloud -Applications - -1. Connect - -Active -Directory - -2. Sign in - -3. Access - -Applications - -On-premises - -Figure 2: Azure Active Directory -lets a user sign in once, then -access both on-premises and cloud -applications. - -Using Azure AD for single sign-on also -offers other benefits: - - ■ Because Azure AD provides a common -account for many applications, there is -less need to have multiple passwords -that might be reused across different -SaaS applications. This improves -security, because a breach at one site -is less likely to expose a password -that has been reused for another -application. - - ■ When a user leaves the organization, -an administrator can terminate access -to multiple applications (cloud or on- -premises) by simply removing one ID -from Azure AD. If there were separate -sign-on to these applications, the -administrator would need to find and -remove each one. - -Microsoft’s broad support for hybrid -identity is unique among major cloud -platform providers. For example, AWS -Identity and Access Management focuses -on managing identity for AWS itself and -for resources running on AWS. Unlike -Azure AD, it does not provide a general -solution for single sign-on that works -across cloud applications from multiple -vendors. - -Hybrid SaaS applications -Applications that are available as both SaaS solutions and on-premises products represent another form of hybrid cloud computing. -For example, Office 365 includes several components that function this way, including Exchange and SharePoint. -In a hybrid world, hybrid SaaS applications can be useful. For example, SharePoint Online and SharePoint Server offer a common -administrative experience, as do Exchange Online and Exchange Server. These applications also provide other hybrid benefits, such -as support in SharePoint for searching across SharePoint Online and SharePoint Server and the ability to use a common email domain -across Exchange Online and Exchange Server. -Like other applications, hybrid SaaS solutions must deal with identity. With Office 365, the application is also using Azure AD, but users -may be unaware of this. This offers benefits, such as assigning licenses to specific Office 365 applications based on AD groups, so -that everyone in a particular group is granted access. Combining hybrid identity with hybrid SaaS applications is another way that a -consistent hybrid cloud provides value for an organization. - -— - 01.2 -Protect identities across -on-premises and cloud -environments -A - -Using a common identity to access many -applications has real benefits. It also makes -protecting that identity more important -than ever. Microsoft offers several ways to -do this. - -For example, Azure AD provides Multi- -Factor Authentication (MFA). With this -option, logging in to Azure AD requires -more than just a simple password. Users -also need a second factor, such as entering -a code sent to their mobile phone. With -MFA, even an attacker who steals a user’s -password cannot log in as that user. The -attacker would also need access to the -user’s phone, or another factor being used -for authentication. The result is better -identity protection and lower risk. - -Azure AD also supports conditional access -policies. These let your administrators -control access to specific applications -using not just the user’s identity, but -also based on what device is being -used, the location, the groups the user -belongs to, and more. User identities -define the perimeter of the hybrid cloud, -cross-platform APIs to integrate identity -management into on-premises or cloud -applications, with support for all modern -protocols, including SAML 2.0, WS-Fed, -OAuth 2.0, and OpenID Connect. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - 6 - -7 - -02. -Requirement: integrated -management and security -A - -Insights and -Analytics - -Automation -and Control - -Security -and Compliance - -Protection -and Recovery - -Azure -Management and Security - -Windows - -Linux - -Windows - -Linux - -On-premises - -Figure 3: Azure provides integrated -management and security services for -a modern hybrid cloud. - -Figure 4: The Monitoring Dashboard -provides a broad and customizable -view of a hybrid cloud environment. - -Using a hybrid cloud can broaden -options for delivering IT services to the -organization. However, hybrid clouds -bring new hurdles for management and -security. Challenges include: - ■ Monitoring combined on-premises -infrastructure and cloud resources - - ■ Effectively automating whenever - -possible, such as the response to alerts -raised through monitoring - - ■ Securing the larger surface area that a - -hybrid cloud brings - - ■ Providing effective data backup and -disaster recovery for both cloud and -on-premises resources - -Addressing these challenges requires -a specialist approach to management -and security that is designed for hybrid -clouds. To provide this, Microsoft delivers -management and security services from -Azure, providing built-in capabilities -throughout the operational lifecycle. -Azure includes a cohesive set of tools for -monitoring, configuring, and protecting -the entire hybrid cloud. Figure 3 shows -the big picture of Azure Management and -Security (which is sometimes referred to -as the Operations Management Suite). - -Designed for a hybrid and heterogeneous -world, Azure management and security -services give increased control of Windows -and Linux systems running in Azure or -in an on-premises datacentre. Although -not shown in the diagram, these Azure -services can also extend to management -of Windows and Linux systems running -at hosting services or on other cloud -platforms, such as AWS. - -Four key services comprise the lifecycle -approach to management from Azure: -Insight and Analytics, Automation and -Control, Protection and Recovery, and -Security and Compliance. All of them are -accessed through a single dashboard, an -example of which is shown in Figure 4. - -The console is built from tiles, and each -user can customise which tiles appear -to suit specific needs. A user focused on -security, for example, might choose to -show the status of antimalware software -on managed systems, the number of -outstanding security issues, and a map -showing where threats are coming from. -A user who is responsible for backup might -choose to display information about the - -latest backups for virtual machines, email, -and other data. A user who is responsible -for all of these areas – Azure management -and security services are designed to -be effective tools for generalists – might -display some combination of these -things. Whatever the requirement, this -customizable interface gives all users -access to the information and services -most relevant to their own needs. - -In addition, because Azure cloud services -are designed for a hybrid world, the -picture the dashboard presents can span -both on-premises and cloud datacentres. - -— - - 02.1 -Get common insight -and analytics across your -hybrid cloud -A - -One of the most important aspects of -management is monitoring to keep track of -what is happening in your world. Doing this -effectively in a hybrid cloud is especially -challenging, since you need the ability to -reach into both on-premises and cloud -datacentres. Azure addresses this with the -Insight and Analytics service. - -The Insight and Analytics service collects -information, such as log and performance -data, about all the systems it monitors. -Using the dashboard, a user can then -issue custom queries against this data or -create queries that run regularly, and then -generate an alert if there is an irregularity. -Moreover, to support common situations, -Insight and Analytics includes a group -of solutions that provide predefined -queries and logic for addressing a specific -area. For example, the AD Assessment -solutions displays the status of Active -Directory, along with recommendations for -improvement, while the SQL Assessment -solutions provides similar information for -SQL Server. - -This Azure monitoring service can -also connect directly to System Centre -Operations Manager. Doing this lets Insight -and Analytics receive information and -alerts that Operations Manager gets from -the systems it monitors. Connecting to -Operations Manager provides easy access -to useful information, and so it is a common -way to get started with Insight and Analytics. - -02.2 -Provide management -automation for your -hybrid cloud -A - -In general, it is a good idea to automate as -much of systems management as possible. -To do this from the cloud, Azure includes -the Automation and Control offering. -This service lets administrators create -PowerShell scripts called runbooks to -automate common processes. - -For example, think about the scenario just -described, where Insight and Analytics -raises an alert based on detection of a -problem with an application. One way to -handle this is to rely on an administrator -to find and fix the problem. This might be -the only option the first time a problem -surfaces. If it is likely to the administrator, -can create a runbook that takes the same -steps, such as truncating the database -table, and then configure the runbook to -execute whenever the alert appears. Doing -this can make problem resolution faster, -more reliable, and less expensive. - -The Automation and Control offering also -addresses many other scenarios. The -administrator might create runbooks that -reset user passwords or set up virtual -machines for a development environment -or schedule and deploy patches for -Windows and Linux. To make this easier, -Microsoft and others provide a gallery -of predefined scripts that address many -common scenarios. Besides runbooks, -this Azure service offering also provides -Desired State Configuration (DSC), which is -the ability to specify how Windows or Linux -server should be configured, then monitor -and enforce that configuration. - -Together with Insight and Analytics, -Automation and Control has one -overarching goal: to help proactively find -and fix problems in your hybrid cloud -before they affect your business. - -Insight and Analytics also provides other -useful tools, including: - - ■ Service Map, which can automatically -discover distributed applications in -your hybrid environment; show the -dependencies among application -components, such as databases and -business logic, and help trouble shoot -problems. - - ■ Network Performance Monitor allows - -an administrator track network -performance, including links between -on-premises and cloud datacentres, -the find and fix network problems. - -To understand the value of this -technology, consider a simple scenario, -and suppose Insight and Analytics raises -an alert about an application in your -environment. An administrator might -use Service Map to understand the -structure of that application and then -determine that the problem lies with -the application’s SQL Server database. -The administrator can use the SQL -Assessment solution provided by Insight -and Analytics do take a closer look at that -database. Perhaps the problem is that -one of the database’s tables has reached -its maximum size, for instance. Once -the administrator knows this, the table -can be truncated and the application -returned to normal operation. - -The important point to note is that the -process of finding and fixing an error is -the same whether the problem database -is running on-premises, at a hosting -provider, or in the cloud. It is also the -same for SQL Server and Azure SQL -Database. Consistency is an integral -feature of the Insight and Analytics -service because it was designed -specifically for a hybrid world. - -Compare this with other cloud vendors. -AWS, for example, provides CloudWatch -for monitoring a cloud environment. Yet -this technology provides little information -about anything else because is not -designed for a hybrid world. Service Map, -by contrast, can automatically discover -applications that span cloud and on- -premises datacentres, and the monitoring -capabilities of Insight and Analytics in -Azure treat both environments equally. -This Azure service is designed from the -ground up to work with a hybrid cloud. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - 02.3 -Get a unified view of -security and compliance -A - -8 - -9 - -No aspect of systems management is more -important than security. Which of your -systems are being attacked right now? -Where are the attacks coming from? What -is the status of antimalware software on -each of those systems? In a hybrid cloud, -you must be able to answer all of these -questions in a common way for systems -in both on-premises datacentres and the -cloud. - -Azure Security and Compliance addresses -these concerns. By providing a unified set -of tools, it lets your administrators - even -those who are not security specialists - -prevent, detect and respond to threats -across your hybrid cloud. Here are some -examples of what this capability provides: - - ■ Administrators can see the security -state of their entire hybrid cloud -in a single view. This view can -include antimalware status, whether -systems conform to defined baseline -configurations, and more. In fact, -the Security and Compliance service -offering relies on the repository and -query mechanism used by Insight and -Analytics, bringing the same breadth -and power to security management. - - ■ This broad visibility can help - -administrators quickly detect and -respond to security threats. For -example, an administrator can -determine that a virtual network is -incorrectly configured, the click a - -button in the dashboard to fix it, or that -a database in Azure SQL database -does not have encryption enabled, -and the turn it on with a few more -clicks. Security and Compliance even -provides a prioritized list of security -vulnerabilities like these, along with -guidance on how to fix them. - - ■ Because Azure Security and - -Compliance is continually updated by -Microsoft, it learns about new threats as -they appear. This helps administrators -maintain an up-to-date security posture -across the hybrid cloud. In addition, -because these updates draw from -Microsoft’s own information about -attacks on Azure, Xbox and other -cloud services, this threat intelligence -is based on a very large dataset. You -get the benefit of Microsoft’s broad -experience, along with the ongoing -updates possible in a cloud-based -solution. - -Once again, it is worth comparing what -Azure provides with what most other cloud -providers offer. Because these vendors are -focused solely on the cloud, their solutions -do not typically address the wider needs of -a hybrid environment. Moreover, since the -other major cloud providers tend to lack -Microsoft’s long experience as a provider -of enterprise management software, they -might not understand what is required -to manage an on-premises computing -environment as well as the cloud. - -— - 02.4 -Use cloud-based services -for on-premises backup -and disaster recovery -A - -Wherever it is stored, data needs to be -backed up. Given the massive amount -of low-cost storage available in a cloud -platform like Azure, using a cloud service -to do this makes sense. Similarly, many -applications need disaster recovery (DR) -to make sure they keep running in the face -of unexpected failure. The cloud is ideally -suited to provide this service. - -To meet these needs, Microsoft provides -two distinct services: Azure Backup and, -for DR, Azure Site Recovery. As its name -suggests, Azure Backup backs up data -to Azure datacentres. This data might be -from Windows or Linux virtual machines, -Exchange, SharePoint, SQL Server, or -Windows files. Whatever the source, Azure -Backup lets you store copies of data on -Azure, then restore that data as required. - -In addition, you can restore just the data -you need. Unlike tape backup, there is no -need to fetch an entire tape from an offsite -location. Azure Backup also lets you keep -that data on geo-redundant storage to -protect your backup in the unlikely event -that an entire Azure datacentre is disabled. -Azure Site Recovery is a cloud-based - -disaster recovery service designed to take -advantage of the scalability and resilience -of Azure. Figure 5 illustrates this service. - -Azure Site Recovery can be used with -Linux and Windows virtual machines -running on VMware or Hyper-V, as well as -physical servers. These can be running -anywhere in the hybrid cloud, on-premises, -at a hosting provider, or (for virtual -machines) on Azure. Wherever they are, -these systems send regular updates to -Azure Site Recovery in the cloud. Should -disaster strike, the service manages -failover to Azure. This includes creating -the necessary Azure virtual machines, -initializing these with the most recent -state stored in Azure Site Recovery, and -configuring virtual networks. Intended to -be used for even complex workloads, such - -as SAP, this service provides easy-to-use -disaster recovery at relatively low cost. -Backup and DR are commonly available -offerings, so what makes these Azure -services uniquely suited for a hybrid cloud? - -The answer is that both Azure Backup and -Azure Site Recovery are managed cloud -services. Any cloud platform lets you run -third-party software for backup and DR, -and Azure does too. You are free to use -CommVault, Veritas, or something else -with Azure Blobs, for example. This big -difference is Microsoft provides managed -services for both, which removes the need -to buy, install, and run backup and DR -solutions in the cloud. This is significantly -simpler than managing your own servers in -the cloud, and it is an important example of -the benefits of a hybrid cloud. - -Azure - -Windows -VMs - -Linux - -Windows -Windows -VMs -VMs - -Linux -Linux - -1. Send -updates - -Azure Site -Recovery - -2. Failover - -Windows -VMs - -Linux -VMs - -Azure - -VMware - -Windows -VMs - -Linux - -Hyper-V - -Windows -VMs - -Linux - -On-premises - -Figure 5: Azure Site Recovery provides disaster recovery for virtual -machines and physical servers with failover to Azure. - -Maintain hybrid or migrate entirely to the cloud? -Some organizations plan to remain hybrid indefinitely. Others view hybrid as a waystation on their journey to the cloud: they consider -hybrid cloud as part of a migration strategy. -If you are in this second category, a consistent hybrid cloud can make migration significantly easier. For example, Azure Site Recovery -can help with migration as well as disaster recovery because it can create new instances of on-premises applications on Azure. Rather -than manually moving applications to the cloud, you can rely on Azure Site Recovery to do this and to help you cut over to the new -cloud instances. The Microsoft hybrid cloud provides other tools as well, such as the migration wizard built into SQL Server Management -Studio to help move on-premises SQL to help move on-premises SQL Server applications to Azure IaaS virtual machines. -Microsoft also helps lower the cost of migration by transferring on-premises licenses to Azure. You can use your existing Windows -Server licenses with Software Assurance to enable up to 40 percent savings on Windows Server virtual machines in Azure by using -the Azure Hybrid Use Benefit. Similarly, license mobility provides the flexibility to deploy existing SQL Server licenses with Software -Assurance in the cloud without additional fees. These attributes can unlock significant savings as you look to extend into cloud or lift -and shift to cloud. You can also rely on GFT Cloud Team, to provide both migration knowledge and tools. -Whatever options you choose, Microsoft’s consistent approach to hybrid cloud can make migration to a full cloud environment simpler, -faster, and less expensive. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - - 10 - -11 - -03. -Requirement: consistent data platform -A - -What is the best approach for working with data in -a hybrid cloud? There is no single answer to suit all -organizations. However, one thing is universally clear: -you will have important information in both the cloud -and on-premises. Given this, it makes sense to have -a common approach to working with data in both -places. - -To facilitate this, the Microsoft hybrid cloud provides a consistent data platform. This -consistency lets an organization use the same tools and the skills throughout its -environment. Moreover, because organizations use data in various ways, the Microsoft -platform works with both operational data, such as orders in an online shopping -application, and analytical data, such as aggregated information used for data analysis - -The challenges this helps you address include: - ■ Using a common database across an on-premises datacentre and the public cloud. - - ■ Using data services in the cloud to complement on-premises database. - - ■ Providing consistent services on-premises and in the cloud for data warehousing, data - -analysis, and data visualization. - -03.1 -Take advantage of -common database on- -premises and in the cloud -A - -Microsoft SQL Server is a mature offering -that supports mission-critical workloads in -datacentres around the world. In a hybrid -cloud environment, this technology can be -used in a number of helpful ways. - -One option is to run SQL Server in an -Azure IaaS virtual machine. For example, -you might do this as part of moving an -entire application to the cloud to lower -your costs, or to create a development -environment for applications that are -deployed on-premises. Whatever the -reason, you can use the same database -technology throughout to your hybrid -cloud. This makes it easier to move data -and applications as needed to respond to -changing business requirements. - -Another possibility is to take advantage of -the cloud to provide business continuity - -Microsoft Azure VM - -Application - -On-premises - -SQL -Server - -SQL -Server - -SQL Server -Always On -Availability -Group - -Figure 6: With SQL Server AlwaysOn, a secondary server can run in an Azure virtual machine, -using the cloud to provide business continuity for an on-premises database. - -with SQL Server AlwaysOn Availability -Groups. Figure 6 shows how this looks. -As the figure shows, an availability -group has two, or more, instances of SQL -Server running on two different systems. -The AlwaysOn technology automatically -replicates changes to data across both -systems. If the primary fails, the secondary -can automatically take over, letting -applications that use this database continue -running. This kind of replication is essential - -for mission-critical workloads, and AlwaysOn -supports scenarios with a low recovery -time objective (RTO). - -Running the secondary server in the cloud -can save money and time. This secondary -can also be made readable to help scale -access to data. For example, a readable -secondary could be located in an Azure -datacentre that is closer to salespeople -using a mobile business intelligence app. - -— 03.2 -Save money by moving -your on-premises data to -the cloud -A - -Running a database in a cloud virtual -machine is useful. However, a hybrid -data platform should also provide cloud -database services that complement -on-premises databases. The Microsoft -hybrid cloud does this in several ways. - -SQL Server, for example, provides built- -in support for backups to Azure Blobs. -This support is simple to use – setting -up scheduled backup requires just a few -clicks – and it lets you take advantage of -the low-cost storage provided by Blobs. It -can also provide geo-replication of backup -data, making sure this data is stored in two -different Azure datacenters. Given the ever- -growing amount of data that organizations -need to store, having easy access to this -bottomless cloud storage is useful. - -Using Azure SQL Database, a PaaS data -service, brings more possibilities. Because -Azure SQL Database is based on SQL -Server, in fact, the two share the same -core database engine, applications can -access data in the same way with both -technologies. Azure SQL Database can -also be combined with SQL Server in -useful ways. - -For example, the SQL Server Stretch -Database feature lets an application -access what looks like a single table in a -SQL Server database. In fact, however, -some or all rows of that table might be -stored in Azure SQL Database. This -technology automatically moves data that -is not accessed for a defined period to the -cloud, as Figure 7 illustrates. - -2. Transparently -access cloud data -as needed - -1. Read - -Application - -On-premises - -Azure SQL -Database - -SQL Server - -Figure 7: SQL Server Stretch Database -automatically archives cold relational data in -the cloud. - -An application reading this data is -unaware that any of it has been moved -to the cloud. The application just issues -SELECTs as usual. When colder data is -required to satisfy a query, SQL Server -Stretch Database will automatically fetch -this data from Azure SQL Database. The -data is protected throughout the entire -process, both at rest and in motion. The -result is automatic archiving or less-used -data in the lower-cost cloud. You might -use this, for instance, to store the order -history of your customers. In this situation, -you certainly want to retain the older data, - -but since older orders are accessed less -frequently, storing them in Azure SQL -Database can save money while requiring -no changes to your applications. -SQL Server Stretch Database offers -another example of how Microsoft’s -hybrid cloud differs from its competitors. -Because GFT provides both SQL Server -and Azure SQL Database, we can -combine the two to deliver innovative -services in unique ways. AWL, for -example, does not support this – the -SQL Server Stretch Database feature is -available only with the Azure cloud. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper 12 - -13 - -03.3 -Use consistent data -warehousing, analysis, -and visualization services -A - -So far, the focus has been on operational -data. To see the full value of a consistent -data platform, we need to broaden our -scope. Analytical data is also an important -part of the story, which is another area -where Microsoft’s consistent hybrid cloud -can improve how an organization works -with data. Figure 8 shows one possible -scenario. - -Suppose you need to analyse operational -data held in an on-premise SQL Server -database and in Azure SQL Database. As -Figure 8 shows, you can load data from -both sources (and many others) into Azure -SQL Data Warehouse. This cloud service -can hold very large amounts of data, both -relational and unstructured, and empowers -users to set up a data warehouse in a few -minutes. The service also makes scaling -simpler: just move a slider to increase or -decrease warehouse capability. In addition, -to enable a user to issue queries over both -relational and unstructured data, Azure -SQL Data Warehouse proves PolyBase a -technology for using any language with -any data. - -You can use Azure Analysis Services to -analyse this data, as the figure shows. -Based on the proven technology of SQL -Server Analysis Services, this cloud service -lets your organization make use of existing -skills and familiar tools. Rather than -learning something entirely new, staff can -be productive immediately. In addition, as -Figure 8 shows, you can import existing -tabular models to reuse work already done. -Data analysts can also create new models, -and then deploy them to either Azure -Analysis Services or SQL Server Analysis -Services just by changing a URL. - -The last link in the analytics chain is -visualizing the data. To enable this, the -Microsoft data platform provides the cloud- -based Power BI (Business Intelligence). -This service can work with data from many -different sources, including Azure Analysis -Services, reports produced on-premises - -Azure SQL -Database - -Data - -Azure SQL -Data -Warehouse - -Azure -Analysis -Services - -Power BI - -Data - -Tabular Model - -Report - -SQL Server - -Analysis -Services - -Reporting -Services - -On-premises - -Figure 8: Microsoft’s hybrid cloud data platform -includes consistent services for analysing and -visualising both, on-premises and cloud data. - -software and services for working with -data in many different ways. The platform -also lets you move data and other -artifacts, like tabular models, as needed. -In addition, just as importantly, the cloud -components of this data platform are all -PaaS services. Rather than running on- -premises servers in IaaS virtual machines, -with all of the management effort this -implies, PaaS services take care of this. -Creating and scaling resources in the cloud -is straightforward, and Azure handles -high availability. These factors can lower -your costs and free your people for more -valuable work. - -Alternative solutions lack the breadth -of Microsoft’s hybrid data platform, as -well as consistency across the cloud and -your datacentre. They also do not offer -Microsoft’s commitment to providing PaaS -services wherever possible. Without this -kind of consistent data platform, you do -not really have a hybrid cloud. - -using SQL Server Reporting Services, -and more. It also has a connector to AWS -Redshift, Amazon’s data warehousing -services. - -Power BI can display dashboards through -a web browser, mobile devices, or in other -ways. It can also be accessed via natural -language queries spoken to a phone. For -example, a salesperson might use a phone -to easily call up information about last -month’s sales in London or to make other -specialized requests. - -Microsoft’s consistent data platform -provides additional technologies for a -hybrid cloud. For example, an organization -can gain rich insights from its data, -relational and non-relational, with fully -managed big data services in the cloud, -including Azure HDInsight and Azure -Data Lake. The Microsoft hybrid cloud -also enables analytics, deep learning, and -intelligent applications across on-premises -and cloud data with common templates -and reusable R language support. Azure -Machine Learning, Cognitive Services, and -other offerings make this possible. - -Why is Microsoft’s hybrid cloud data -platform better than alternatives? There are -several reasons. First, it is broad, providing - -— 04. -Requirement: unified development -and DevOps -A - -Some things are the same, of course – -both environments offer virtual machines -running Windows Server and Linux – but -others are quite distinct. These differences -can create problems, including the -following: - - ■ An application built for a cloud - -platform might be hard to move back -on-premises. If the application uses a -technology that is not available in your -datacentre, you might find yourself -committed to the cloud. - - ■ Providing up-to-date platform - -technologies in cases where a cloud -solution is not appropriated can be -difficult. A primary example of this is -edge-computing scenarios, where -applications cannot always connect -to the cloud or must run close to their -users for performance reasons. - - ■ Creating applications that exploit the -cloud while still complying with every -applicable regulation can be challenging. -If the application must run in even one -geography where data sovereignty -rules require an on-premises solution, -developers might need to create -different versions for the cloud and an -on-premises datacentre. - -One of the biggest -challenges in using the -cloud is that the -cloud development -platform differs from -traditional on-premises -platforms. - -Applications - -IaaS - -PaaS - -Serverless - -Containers - -Storage - -… - -Azure - -Applications - -IaaS - -PaaS - -Serverless - -Containers - -Storage - -… - -Azure Stack - -On-premises - -Figure 9: Azure Stack provides a subset of -Azure services in the datacentre, letting the -same applications run in both places. - - ■ Differences between a cloud platform - -and on-premises environment can make -it hard to create a common DevOps -environment for applications deployed -in both places. - -Microsoft Azure Stack addresses all -of these challenges. You can use this -technology to provide a subset of the -cloud services offered by Microsoft Azure -on your own premises. Figure 9 shows -how this looks. - -Azure Stack lets developers build and -deploy software in the same way, whether -it runs on-premises or in the cloud. It also -lets them implement consistent DevOps -mechanisms across a hybrid cloud. To -allow these things, Azure Stack provides -many of the most important Azure -technologies, and more are being added. - -Technologies in Azure Stack include: - - ■ Infrastructure as a service (IaaS), - -providing Windows and Linux virtual -machines on demand. - - ■ Platform as a service (PaaS), including -App Service, with support for creating -applications in .NET, PHP, Java, and -other environments, and Service -Fabric, a foundation for microservices -applications. Both Azure and Azure -Stack also support Cloud Foundry, a -cross-platform PaaS technology. - - ■ Serverless computing with Azure - -Function. - - ■ Container support with Azure Container -Service (ACS), which provides container -orchestration using Kubernetes, DC/OS, -and Swarm. - - ■ - -Storage, including Azure Blobs and -Tables. - -Rather than trying to stretch existing on- -premises technologies to provide cloud -services, Azure Stack brings Azure services -into the datacentre. Creating a consistent -hybrid cloud helps solve the problems -described earlier. - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper 14 - -15 - -04.3 -Create cloud applications -that meet every regulation -A - -For many organizations, the biggest barrier -to embracing cloud computing is regulatory. -Sometimes the law prohibits storing customer -data outside an organisation’s own datacentre, -or maybe remote storage is allowed, but -only within a specific country. Since no cloud -provider has a presence in every nation, these -regulations can prevent you from using the -cloud for some applications. - -With Azure Stack, your developers can -create an application that can be deployed -without modification on either Azure or Azure -Stack. If regulations permit cloud, running -an application on Azure might be the least -expensive, most scalable, and most secure -choice. If regulations dictate that data must -be kept on-premises, the same application -can be run on Azure Stack in the datacentre. -In addition, if you would like to run the -application externally, but are required to -keep it within a specific national border, you -can rely on a hosting provider that runs Azure -Stack in an in-country datacentre. Whatever -the regulatory requirement, the combination of -Azure and Azure Stack lets you meet it using -the same application. Moreover, as always, -these applications can be externally facing -or internally facing, with web clients, mobile -clients and more. - -Compare this to the options with a public- -cloud-only provider, such as AWS. To address -the same regulatory diversity, you would -likely need to build two different versions of -your application. A true hybrid cloud does -not require this; it provides the same modern -cloud platform everywhere. - -04.1 -Run the same modern -applications on-premises -and in the cloud -A - -04.2 -Provide integrated -solutions across edge -and cloud -A - -Many business requirements can be met -by applications running in the cloud. -But cloud is not suitable for all, so the -edge is still important. For example, in -situations where the latency inherent -in cloud access is not acceptable. -In a manufacturing environment, for -instance, a real-time control application -might need to be located very close -to the robots it controls. The speed -of light is a constraint that is not -going away. Alternatively, suppose a -continuous connection to the cloud is -not practical. Think of a cruise ship at -sea, for example, which might have only -intermittent internet access. - -In all of these situations, organizations -still want to create applications using -modern platform technologies. They -might also want to create applications -that can run in the cloud if needed. -Why lock these applications into an -on-premises environment if it is not -required? Using Azure Stack lets an -organization create the best possible -applications at the edge while still taking -advantage of the cloud as needed. For -example, a cruise ship might rely on a -shipboard application built on Azure -Stack to collect and aggregate data -about that ship. The same application -might also run on the Azure cloud to -collect and aggregate data across the -company’s fleet of ships. Having the -same development environment in both -places makes this possible. - -Because Azure and Azure Stack offer -the identical technologies – the same -services with the same interfaces – -moving applications between the two -is straightforward. This has several -advantages. - -First, an organisation can choose where -to deploy an application based on precise -needs. There is no commitment to either -cloud or an on-premises datacentre. -Instead, where an application runs in a -hybrid cloud can change with evolving -business and technical requirements. - -In addition, developers can use up-to-date -technologies for all of the applications -they create, whether they run in the -cloud or in the datacentre. This includes -externally facing applications, such as an -e-commerce system used by customers, -as well as internally facing applications, -such as a line-of-business solution used by -employees. The most recent innovations, -including serverless computing with Azure -Functions and modern container support -with ACS, are available in both places. -On-premises development projects are no -longer limited to older technologies. - -Just as important, using the same -technologies in both places means that an -organization can leverage the same skills -in both places. Rather than finding (and -keeping) people with different skills for -cloud and on-premises development, the -same people can be used for both kinds of -projects. Similarly, the same processes can -be adopted for deploying and updating -applications, as described in more detail -later. - -Compare this with what is required if you -choose a cloud platform from a provider -that is solely focused on the cloud, such -as Amazon Web Services (AWS). Without -an on-premises equivalent, you are forced -to use different technologies, people, and -processes in these two environments. -This is a clear example of why a consistent -hybrid cloud is so important. - -The Internet of Things as edge computing - -The Internet of Things (IoT) provides another topical example of combining edge -computing with the cloud. To support IoT applications, Microsoft Azure provides -IoT Hub, a cloud service that can accept and buffer large numbers of events from -many devices. These events can then be processed by applications running -on Azure, perhaps using Azure Stream Analytics or another Azure streaming -technology. In addition, to help create field gateways, systems that aggregate data -from multiple simple devices, Microsoft provides the IoT Gateway SDK. - -— - 04.4 -Use a common DevOps -environment on-premises -and in the cloud -A - -DevOps is a combination of tools, -processes, and culture that can improve -how software is deployed and updated. -Among the most important aspects of -implementing DevOps well are: - - ■ Automating creation of the environment - -the application needs. This might -include creating virtual machines, setting -up networks, deploying application -code, and more. Automation means -doing all of this with software, so this -aspect of DevOps is often referred to -with the phrase “infrastructure as code.” - - ■ Automating the deployment of new -software into this environment. This -typically uses tools such as Jenkins or -Visual Studio Team Services that allow -creating pipelines to build, test, and -deploy new code. - -In a hybrid world, the same application -might be run either on-premises or in the -cloud. Accordingly, an important part of -creating a hybrid cloud is setting up a -DevOps process that works identically -for both. With Azure and Azure Stack, the -Microsoft hybrid cloud provides this, as -shown in Figure 10. - -As the figure shows, a DevOps team -uses Azure Resource Manager (ARM) to -define the required infrastructure in both -Azure and Azure Stack. By creating an -ARM template, the team can fully specify -whatever environment an application -requires. Moreover, because Azure -and Azure Stack provide completely -consistent services, the same template -can be used to create this environment in -either the cloud or on-premises. -Once this is done, the DevOps team -can use its choice of tools to create -deployment pipelines for new software. -Because Azure Stack is a subset of -Azure, the same pipeline can target -either environment. Sending software -to one or the other requires changing -only the URL targeted by a pipeline. For -example, suppose you have created an -application that serves most of its users -from Azure in the cloud, but runs on -Azure Stack in a particular country for -regulatory reasons. The development -team can create an ARM template to -set up an identical environment in both -places, then use whatever build server -they choose to deploy updates to both -in the same way. This consistency lets -them create applications in a common -way, and then easily deploy them to the -right location. It also lets the team reuse -ARM templates across your hybrid cloud -as needed, simplifying your DevOps -process. - -1. Define infrastructure - -Azure Resource -Manager - -TEMPLATES - -Figure 10: Azure and Azure Stack can share -identical DevOps environments. - -Applications - -Azure - -Applications - -Azure Stack - -On-premises - -2. Deplay software - -Visual Studio Team -Services, Jenkins, … - -PIPELINES - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper - 16 - -17 - -Next steps - -Hybrid cloud computing is important for many -organizations, and it is likely to remain such. -Choosing the right hybrid cloud vendor is of fundamental -importance. - -A truly consistent hybrid cloud must provide four key components, all explicitly -designed for a hybrid world: - ■ Common identity - ■ Integrated management and security - ■ A consistent data platform - ■ Unified development and DevOps - -The Microsoft hybrid cloud offers all four. Whether your goal is integrating your on- -premise environment with the cloud, optimizing your computing across both worlds, or -innovating with the new services cloud computing provides, Microsoft supports you. -If you are looking for a holistic computing environment that lets on-premises resources -and the cloud smoothly work together, provides a unified approach to management, -and allows a single identity everywhere, you are looking for the Microsoft hybrid cloud. -Microsoft’s credibility and consistent hybrid capability makes it the right trusted vendor -for your safe journey. - -Microsoft Azure Cloud Migration -> https://azure.microsoft.com/en-us/solutions/cloud-migration/ - -Microsoft Azure Hybrid Integration -> https://azure.microsoft.com/en-us/solutions/hybrid-integration/ - -Microsoft Azure DevOps -> https://azure.microsoft.com/en-us/solutions/devops/ - -Microsoft Azure Internet of Things -> https://azure.microsoft.com/en-us/solutions/internet-of-things/ - -References -A - -Stay in touch -A - -Carlos Mattos -B -Global Business Unit -Head – Azure -E.: carlos.mattos@gft.com - -Team-oriented leader with a successful track record and over 20 years’ experience -spanning global corporate software development, strategy, sales, business development -and operations for software market. Nominated by Microsoft as Regional Director (RD) -from 2017 to 2020. Also awarded as Microsoft Most Valuable Professional (MVP) for 12 -years (2003-2016). - -— GFT -at a glance -A - -GFT is driving the digital -transformation of the world’s leading -companies in the financial and -insurance sectors, as well as in the -manufacturing industry. - -software engineering and innovation. -The GFT Technologies SE share is listed -in the Prime Standard segment of the -Frankfurt Stock Exchange (ticker: GFT-XE). - - blog.gft.com - twitter.com/gft_en - linkedin.com/company/gft-group - facebook.com/GFTGroup - >gft.com - -As an IT services and software engineering -provider, GFT offers strong consulting -and development skills across all aspects -of pioneering technologies, such as -cloud engineering, artificial intelligence, -mainframe modernisation and the Internet -of Things for Industry 4.0. - -With its in-depth technological expertise, -profound market know-how and strong -partnerships, GFT implements scalable -IT solutions to increase productivity. This -provides clients with faster access to new -IT applications and innovative business -models, while also reducing risk. - -Founded in 1987 and located in 15 countries -to ensure close proximity to its clients, GFT -employs 6,000 people. GFT provides them -with career opportunities in all areas of - -Shaping the futureof digital businessPartners:MicrosoftAzureImplementing Hybrid Cloud with Microsoft AzureWhite paper 18 - -19 - -GFT Corporate Center -in Stuttgart, Germany - -Innovations to touch -at our Digital Innovation -Labs - -Teamwork -worldwide - -— This report is supplied in good faith, -based on information made available -to GFT at the date of submission. It -contains confidential information that -must not be disclosed to third parties. -Please note that GFT does not warrant -the correctness or completion of the -information contained. The client is -solely responsible for the usage of the -information and any decisions based -upon it. - -GFT Technologies SE BSchelmenwasenstr. 34, 70567 Stuttgart, Germany BT. +49  711  620  420 Bwww.gft.com Binfo@gft.com - -GFT-201111-WP-AZUREHYBRID-LL-AABB-CCDD © GFT 2020 +English – Disney Terms of Use + +Home / English – Disney Terms of Use + +Last Updated: SEPTEMBER 26, 2019 + +Disney DTC LLC and/or its affiliates and subsidiaries (collectively, “Disney DTC”) are pleased +to provide to you their sites, software, applications, content, products and services (“Disney +Services”), which may be branded Disney, ABC, ESPN, Marvel, Pixar, Lucasfilm, FX or another +brand owned or licensed by Disney DTC. These terms govern your use and our provision of the +Disney Services on which these terms are posted, as well as Disney Services we make available +on third-party sites and platforms if these terms are disclosed to you in connection with your use +of the Disney Services. PLEASE READ THESE TERMS CAREFULLY BEFORE USING +THE DISNEY SERVICES. BY USING THE DISNEY SERVICES YOU AGREE TO BE +BOUND BY THESE TERMS. + +ANY DISPUTE BETWEEN YOU AND US, EXCEPT FOR SMALL CLAIMS, IS SUBJECT +TO A CLASS ACTION WAIVER AND MUST BE RESOLVED BY INDIVIDUAL BINDING +ARBITRATION. PLEASE READ THIS AGREEMENT IN ITS ENTIRETY, INCLUDING +THE ARBITRATION PROVISION BELOW. + +NOTHING IN THESE TERMS IS INTENDED TO AFFECT YOUR RIGHTS UNDER THE +LAW IN YOUR USUAL PLACE OF RESIDENCE THAT CAN NOT BE ALTERED BY +THESE TERMS. IF THERE IS A CONFLICT BETWEEN THOSE RIGHTS AND THESE +TERMS, YOUR RIGHTS UNDER APPLICABLE LOCAL LAW WILL PREVAIL. + +1. Contract between You and Us + + This is a contract (1) between you and Disney DTC LLC, as a Delaware limited liability +company doing business at 500 South Buena Vista Street, Burbank, California 91521, USA, or +(2) between you and any different service provider identified for a particular Disney Service. + +You must read and agree to these terms (the “Agreement”) before using the Disney Services. If +you do not agree, you may not use the Disney Services. These terms describe the limited basis +on which the Disney Services are made available to you and supersede prior agreements or +arrangements. + +Supplemental terms and conditions may apply to some Disney Services, such as rules for a +particular competition, content, software, application, service or other activity, or terms that may +accompany certain territories, content, products or software accessible through the Disney +Services. If applicable, such supplemental terms and conditions will be disclosed to you in +connection with such competition, content, product, software, application, service or +activity. Any supplemental terms and conditions are in addition to these terms and, in the event + + of a conflict, the supplemental terms will prevail over these terms. If you do not agree to the +supplemental terms and conditions disclosed, you must not use the applicable Disney Service. + +We may change these terms at any time. Any such amendment will be effective thirty (30) days +following either our dispatch of a notice to you or our posting of the amended terms. You are +responsible for periodically reviewing the terms for updates and amendments. By continuing to +use the Disney Services you will be deemed to have agreed to and accepted any amendments. If +you do not agree to any change to these terms, you must discontinue using the Disney +Services. Our customer service representatives are not authorized to modify any provision of +these terms, either verbally or in writing. + +We may immediately terminate this contract with respect to you (including your access to the +Disney Services) if you fail to comply with any provision of these terms. + +2. The Disney Services + +The Disney Services are for your personal, noncommercial use and are intended for +informational and entertainment purposes only. They do not constitute legal, financial, +professional, medical or healthcare advice or diagnosis and cannot be used for such +purposes. To support smooth operation of the Disney Services across wide geographic areas, +aspects of certain activities, such as game play, may be simulated to avoid delays. + +The Disney Services are our copyrighted property or the copyrighted property of our licensors or +licensees and all trademarks, service marks, trade names, trade dress and other intellectual +property rights in the Disney Services are owned by us or our licensors or licensees. Except as +we specifically agree in writing, no element of the Disney Services may be used or exploited in +any way other than as part of the Disney Services offered to you. You may own the physical +media on which elements of the Disney Services are delivered to you, but we retain full and +complete ownership of the Disney Services. We do not transfer title to any portion of the Disney +Services to you. + +Content and Software License + +If a Disney Service is configured to enable the use of software, content, virtual items or other +materials owned or licensed by us, we grant you a limited, non-exclusive, non-sublicensable, +non-transferable license to access and use such software, content, virtual item or other material +for your personal, noncommercial use only, only for as long as each Disney Service is made +available by us, with no right to reproduce, distribute, communicate to the public, make available +to the public, or transform any Disney Service in any media format or channel now known or +hereafter devised (except as may be expressly described or contemplated within the Disney +Service). + +You may not: circumvent or disable any content protection system or digital rights management +technology used with any Disney Service; decompile, reverse engineer, disassemble or otherwise +reduce any Disney Service to a human-readable form; remove identification, copyright or other +proprietary notices; or access or use any Disney Service in an unlawful or unauthorized manner + + or in a manner that suggests an association with our products, services or brands. You may not +access or use any Disney Service in violation of United States export control and economic +sanctions requirements. By acquiring services, content or software through the Disney Services, +you represent and warrant that your access to and use of the services, content or software will +comply with those requirements. + +Disclaimers and Limitation on Liability + +THE DISNEY SERVICES ARE PROVIDED “AS IS” AND “AS AVAILABLE.” WE +DISCLAIM ALL CONDITIONS, REPRESENTATIONS AND WARRANTIES NOT +EXPRESSLY SET OUT IN THESE TERMS TO THE FULLEST EXTENT PERMITTED BY +APPLICABLE LAW. + +FOR HELP, PLEASE REFER TO THE HELP SECTION OF THE APPLICABLE DISNEY +SERVICE FOR ASSISTANCE IF DIGITAL CONTENT SUPPLIED VIA A DISNEY +SERVICE IS NOT WORKING PROPERLY. We will not be liable for damage which could +have been avoided by following our advice to apply an update offered to you free of charge or +for damage which was caused by you failing to correctly follow installation instructions or to +have in place the minimum system requirements advised by us. + +We shall not be liable for delay or failure in performance for causes beyond our control or any +other damage which does not result from a breach of our obligation. + +We are not liable for business losses. We only supply the products for domestic and private use. +If you use the products for any commercial, business or re-sale purpose we will have no liability +to you for any loss of profit, loss of business, business interruption, or loss of business +opportunity. + +WE ARE NOT RESPONSIBLE FOR ANY LACK OF FUNCTIONALITY OR FAILURE TO +PROVIDE ANY PART OF THE DISNEY SERVICE, OR ANY LOSS OF CONTENT OR +DATA THAT IS DUE TO: YOUR EQUIPMENT, DEVICES, OPERATING SYSTEM OR +INTERNET CONNECTION, YOUR FAILURE TO COMPLY WITH SPECIFIED +COMPATIBILITY REQUIREMENTS OR THE CONSEQUENCES OF YOU CHANGING +YOUR EQUIPMENT, DEVICE, OPERATING SYSTEM OR INTERNET CONNECTION. + +WE SHALL NOT BE LIABLE TO YOU FOR INDIRECT, INCIDENTAL, SPECIAL OR +CONSEQUENTIAL DAMAGES, INCLUDING LOST PROFITS AND PROPERTY +DAMAGE, EVEN IF WE WERE ADVISED OF THE POSSIBILITY OF SUCH DAMAGES, +NOR SHALL WE BE HELD LIABLE FOR DELAY OR FAILURE IN PERFORMANCE +RESULTING FROM CAUSES BEYOND OUR REASONABLE CONTROL. + +IN NO EVENT SHALL OUR TOTAL LIABILITY TO YOU FOR ALL DAMAGES, LOSSES +AND CAUSES OF ACTION EXCEED ONE THOUSAND U.S. DOLLARS (US $1,000). + + Changes to the Disney Services + +The Disney Services are constantly evolving and will change over time. If we make a material +change to the Disney Services, we will provide you with reasonable notice as described above +and you will be entitled to terminate this contract prospectively by no longer using the Disney +Services. + +Additional Restrictions on Use of the Disney Services + +We do not allow uses of the Disney Services that are commercial or business-related, or that +advertise or offer to sell products or services (whether or not for profit), or that solicit others +(including solicitations for contributions or donations). You agree not to knowingly or recklessly +introduce a virus or other harmful component, or otherwise tamper with, impair or damage any +Disney Service or connected network, or interfere with any person or entity’s use or enjoyment +of any Disney Service. You agree not to use any software or device that allows automated +gameplay, expedited gameplay, or other manipulation of gameplay or game client and you agree +not to cheat or otherwise modify a Disney Service or game experience to create an advantage for +one user over another. Additionally, you agree not to access, monitor or copy, or permit another +person or entity to access, monitor or copy, any element of the Disney Services using a robot, +search engine, “spider,” “scraper,” or other automated means or manual process without our +express written permission. + +Third-Party Services and Content + +The Disney Services may integrate, be integrated into, or be provided in connection with third- +party services and content. We do not control those third-party services and content. You +should read the terms of use agreements and privacy policies that apply to such third-party +services and content. + +If you access a Disney Service using an Apple iOS, Android or Microsoft Windows-powered +device or Microsoft Xbox One, Apple Inc., Google, Inc. or Microsoft Corporation, respectively, +shall be a third-party beneficiary to this contract. However, these third-party beneficiaries are +not a party to this contract and are not responsible for the provision or support of the Disney +Services. You agree that your access to the Disney Services using these devices also shall be +subject to the usage terms set forth in the applicable third-party beneficiary’s terms of service. + +Mobile Networks + +When you access the Disney Services through a mobile network, your network or roaming +provider’s messaging, data and other rates and fees will apply. Downloading, installing or using +certain Disney Services may be prohibited or restricted by your network provider and not all +Disney Services may work with your network provider or device. + + Affiliate Advertising Programs + +Disney DTC is a participant in affiliate advertising programs designed to provide a means for +sites to earn advertising fees by advertising and linking to third party retail sites. + +3. Your Content and Account + + User Generated Content + +The Disney Services may allow you to communicate, submit, upload or otherwise make +available text, chats, images, audio, video, competition entries or other content (“User Generated +Content”), which may be accessible and viewable by the public. Access to these features may be +subject to age restrictions. You may not submit or upload User Generated Content that is +defamatory, harassing, threatening, bigoted, hateful, violent, vulgar, obscene, pornographic, or +otherwise offensive or that harms or can reasonably be expected to harm any person or entity, +whether or not such material is protected by law. + +In most instances, we do not claim ownership to your User Generated Content; however, you +grant us a non-exclusive, sublicensable, irrevocable and royalty-free worldwide license under all +copyrights, trademarks, patents, trade secrets, privacy and publicity rights and other intellectual +property rights for the full duration of those rights to use, reproduce, transmit, print, publish, +publicly display, exhibit, distribute, redistribute, copy, index, comment on, modify, transform, +adapt, translate, create derivative works based upon, publicly perform, publicly communicate, +make available, and otherwise exploit such User Generated Content, in whole or in part, in all +media formats and channels now known or hereafter devised (including in connection with the +Disney Services and on third-party sites and platforms), in any number of copies and without +limit as to time, manner and frequency of use, without further notice to you, without attribution +(to the extent this is not contrary to mandatory provisions of applicable law), and without the +requirement of permission from or payment to you or any other person or entity. You agree that +submission of User Generated Content does not establish any relationship of trust and confidence +between you and us, and that you have no expectation of compensation whatsoever (except as +may be specifically stated in the provisions of the Disney Services in connection with the +submission, or arising from it). + +You represent and warrant that your User Generated Content conforms to these terms and that +you own or have the necessary rights and permissions including, without limitation, all +copyrights, music rights and likeness rights (with respect to any person) contained in the User +Generated Content, without the need for payment to any other person or entity, to use and +exploit, and to authorize us to use and exploit, your User Generated Content in all manners +contemplated by these terms. You agree to indemnify and hold us and our subsidiary and +affiliated companies, and each of their respective employees and officers, harmless from any +demands, loss, liability, claims or expenses (including attorneys’ fees), made against us by any +third party arising out of or in connection with our use and exploitation of your User Generated +Content resulting from your breach of any warranty, representation or other provision of these +terms or any applicable supplemental terms. You also agree not to enforce any moral rights, +ancillary rights or similar rights in or to the User Generated Content against us or our licensees, + + distributors, agents, representatives and other authorized users, and agree to procure the same +agreement not to enforce from others who may possess such rights. + +To the extent that we authorize you to create, post, upload, distribute, publicly display or +publicly perform User Generated Content that requires the use of our copyrighted works, we +grant you a non-exclusive license to create a derivative work using the specifically referenced +copyrighted works as required for the sole purpose of creating the materials, provided that such +license shall be conditioned upon your assignment to us of all rights worldwide in the work you +create for the duration of copyright in the User Generated Content, in all formats and media +known or unknown to date, including for use on Disney Services and on third party sites and +platforms. If such rights are not assigned to us, your license to create derivative works using our +copyrighted works shall be null and void. + +We may monitor, screen, post, remove, modify, store and review User Generated Content or +communications sent through a Disney Service, at any time and for any reason, including to +ensure that the User Generated Content or communication conforms to these terms, without prior +notice to you. We are not responsible for, and do not endorse or guarantee, the opinions, views, +advice or recommendations posted or sent by users. + +Accounts + +Some Disney Services permit or require you to create an account to participate or to secure +additional benefits. You agree that any information you provide and maintain is accurate, +current and complete, including your contact information for notices and other communications +from us and your payment information. You agree not to impersonate or misrepresent your +affiliation with any person or entity, including using another person’s username, password or +other account information, or another person’s name or likeness, or provide false details for a +parent or guardian. You agree that we may take steps to verify the accuracy of information you +provide, including contact information for a parent or guardian. + +We have adopted and implemented a policy that provides for the termination, in appropriate +circumstances, of the accounts of users who are repeat infringers of copyright. In addition, we +may suspend or terminate your account and your ability to use the Disney Services if you engage +in, encourage, or advocate for illegal conduct or if you fail to comply with these terms or any +supplemental terms. + +Passwords and Security + +You agree that you will not share your account or account information with others. You are +responsible for taking reasonable steps to maintain the confidentiality of your username and +password, and you are responsible for all activities under your account that you can reasonably +control. You agree to promptly notify us of any unauthorized use of your username, password or +other account information, or of any other breach of security that you become aware of involving +your account or the Disney Services. + + The security, integrity and confidentiality of your information are extremely important to us. We +have implemented technical, administrative and physical security measures that are designed to +protect your information from unauthorized access, disclosure, use and modification. + +4. Paid Transactions + +Identity of Seller + +Sales are made by Disney DTC or the seller identified at the time of sale, if different. If you +have questions about your order, please contact the seller at the address provided and they will +assist you. Some digital storefronts on the Disney Services are operated by third parties and, in +that case, different or additional sale terms may apply, which you should read when they are +presented to you. + +Digital Content and Virtual Items + +We may make applications, games, software or other digital content available on the Disney +Services for you to license for a one-time fee. When purchasing a license to access such material +from a Disney Service, charges will be disclosed to you on the Disney Service before you +complete the license purchase. + +Your purchase of a virtual item or in-game currency is a payment for a limited, non-assignable +license to access and use such content or functionality in the Disney Services with no right to +reproduce, distribute, communicate to the public, make available to the public or transform any +Disney Service via any online media, in any media format or channel now known or hereafter +devised (except as may be expressly described or contemplated within the Disney Service). +Virtual items (including characters and character names) or in-game currency purchased or +available to you in the Disney Services can only be used in connection with the Disney Services +where you obtained them or where they were developed by you as a result of game play. These +items are not redeemable or subject to refund and cannot be traded outside of the Disney +Services for money or other items for value. We may modify or discontinue virtual items or in- +game currency at any time. + +Digital Movie Codes + +Digital codes originally packaged in a combination disc + code package (a package that includes +a DVD, Blu-Ray, and/or 4K/UHD disc(s) and a digital code) may not be sold separately and may +be redeemed only by an individual who obtains the code in the original combination disc + code +package, or by a family member of that individual. Digital codes are not authorized for +redemption if sold separately. You may use digital movie codes to obtain licensed access to +digital movies only as specifically authorized under these terms and conditions and the terms and +conditions of the participating provider of digital content through which you access or download +the digital movie. Digital movie codes sold, distributed, purchased, or transferred in a manner +inconsistent with these terms and conditions are subject to being invalidated. + + Digital Movie Code Returns. If you do not agree to the above terms and conditions for +redeeming a digital movie code obtained by you or a family member in an original combination +disc + code package, you may return the combination disc + code package to Disney for a refund +subject to the conditions provided at this link. + +Subscriptions + +Some Disney Services require paid subscriptions to access. By signing up for a subscription, +you agreed that your subscription will be automatically renewed and, unless you cancel your +subscription, you authorized us to charge your payment method for the renewal term. The period +of auto-renewal will be the same as your initial subscription period unless otherwise disclosed to +you at the time of sale. The renewal rate will be no more than the rate for the immediately prior +subscription period, excluding any promotional and discount pricing, unless we notify you of a +rate change prior to your auto-renewal, in which case you will have the right to cancel the +renewal of your subscription. From time to time, we may offer a free trial subscription for a +Disney Service. If you register for a free trial subscription, we will begin to bill your account +when the free trial subscription expires, unless you cancel your subscription before that time. + +When you subscribe to a Disney Service (other than semi-annual and annual subscriptions), you +have the right to cancel your contract within fourteen (14) days from the date the transaction is +concluded and you will receive a full refund of the subscription fee paid. For semi-annual and +annual subscriptions, if notice of cancellation is received within the first thirty (30) days +following the first day of initial billing, you will receive a refund of the subscription fee paid. If +we refund your subscription fee, you will still be obligated to pay other charges incurred by you +in the course of using the Disney Service prior to the date of cancellation. If you pay a periodic +subscription fee for a Disney Service, we will provide you with reasonable notice of changes to +the fees or billing methods in advance of their effective date and you will be able to cancel your +subscription prior to such change. If you subscribed online, we will give you the option of +cancelling the subscription online. + +The Order Process + +You will have the opportunity to review and confirm your order, including delivery address (if +applicable), payment method and product details. We will send to you a notice when we accept +your order and our acceptance will be deemed complete and for all purposes to have been +effectively communicated to you at the time we send the notice. At such time, the contract for +sale will be made and become binding on both you and us. The risk of loss in any goods you +purchase and the responsibility to insure them passes to you when the relevant goods are +delivered. + +We reserve the right to refuse or cancel any order prior to delivery. Some situations that may +result in your order being cancelled include system or typographical errors, inaccuracies in +product or pricing information or product availability, fairness among customers where supplies +are limited, or problems identified by our credit or fraud departments. We also may require +additional verification or information before accepting an order. We will contact you if any +portion of your order is cancelled or if additional information is required to accept your order. If + + your order is cancelled after we have processed your payment but prior to delivery, we will +refund your payment. + +Payments and Billing + +When you provide payment information, you represent and warrant that the information is +accurate, that you are authorized to use the payment method provided, and that you will notify us +of changes to the payment information. We reserve the right to utilize third party payment card +updating services to obtain current expiration dates on credit cards and debit cards. + +Right of Cancellation; Return of Goods + +You may have the right to cancel an order placed for a Disney Service - depending on the nature +of the Disney Service. Please read the following information carefully so you understand your +right of cancellation. + +If you wish to cancel, you must do so by following the cancellation instructions for the particular +Disney Service. A sample cancellation form is available here. + +Cancelling Subscriptions + +Please see the information above on the process for cancelling subscriptions in our Subscriptions +section. + +Digital Content + +When you purchase a license to access digital content or virtual items, you will be given an +opportunity to consent to delivery at the time of purchase. By consenting to delivery, you +acknowledge that you have lost the right to cancel. License purchase fees paid for digital content +are non-refundable. + +Physical Goods + +You have the right, within thirty (30) days from the date of your receipt of physical goods, to +cancel our contract with you and return the goods. This right does not apply to goods stated by +us on the Disney Services to be non-returnable, including but not limited to: + +• Any products with a seal, where the seal is broken, such as audio and video recordings, + +computer software, and CD’s, DVD’s or other physical media that have been supplied in +sealed packaging; and + +• Personalized items. + +Personalized Goods + +We reserve the right to refuse personalized orders at our discretion. Inappropriate use of our +personalization service will cause your order to be cancelled and any payment refunded. + + These cancellation rights are separate and in addition to your rights should any item we supply +be faulty. + +If you are returning goods that are not faulty, you may be required to pay for the cost of returning +the goods to us and we may deduct a reasonable amount if you use the goods. + +Pricing; Taxes + +We may revise the pricing for products and services we offer. When you place your order, we +estimate the applicable tax and include that estimate in the total for your convenience. Except to +the extent required under applicable tax laws, the actual tax amount that will be applied to your +order and charged to your payment method is based on calculations on the date of shipment, +regardless of when the order was placed. + +International Shipping; Customs + +When ordering goods for delivery to countries other than the country where the seller is located, +you may have to pay import duties and taxes levied. These and any additional charges for +customs clearance must be borne by you. For goods shipped internationally, please note that any +manufacturer warranty may not be valid; manufacturer service options may not be available; +manuals, instructions and safety warnings may not be in destination country languages; the +goods and accompanying materials may not be designed in accordance with destination country +standards, specifications, and labeling requirements; and the goods may not conform to +destination country voltage (requiring use of an adapter or converter). You are responsible for +assuring the goods can be lawfully imported to the destination country. When ordering from us, +the recipient is the importer of record and must comply with all laws and regulations of the +destination country. + +5. Competitions + +Competitions that you enter on a Disney Service may have supplemental rules and/or conditions, +but the following general rules apply absent contrary terms in any supplemental rules or +conditions for the competition. If there are rules or conditions for a competition and such rules +or conditions conflict with these terms, the provisions contained in the rules and conditions +govern and control the competition. + +Entries + +Your competition entry is User Generated Content and subject to all provisions of these terms +that govern your submission and our use of your User Generated Content. We may disqualify +entries that are late, misdirected, incomplete, corrupted, lost, illegible or invalid or where +appropriate parental consent was not provided. Competition entries are limited to one per person +and entries via agents or third parties or the use of multiple identities are not permitted. Use of +automated entries, votes or other programs is prohibited and all such entries (or votes) will be +disqualified. + + We reserve the right to modify, suspend, cancel or terminate a competition or extend or resume +the entry period or disqualify any participant or entry at any time without giving advance +notice. We will do so if it cannot be guaranteed the competition can be carried out fairly or +correctly for technical, legal or other reasons, or if we suspect that any person has manipulated +entries or results, provided false information or acted unethically. If we cancel or terminate a +competition, prizes may be awarded in any manner we deem fair and appropriate consistent with +local laws governing the competition. + +Eligibility + +To enter a competition, the competition must be open to residents in your country and you must +be a registered user of the Disney Services and have an active account with current contact +information. No purchase is necessary to enter a competition and a purchase will not improve +your chances of winning. Competitions are not open to our employees (or their immediate +families) or anyone else professionally associated with the competition. If you are under age 18 +(or under the age of majority under applicable law) and the competition is open to you, we may +need your parent or guardian’s consent before we can accept your entry. We reserve the right to +request proof of identity or to verify eligibility conditions and potential winning entries, and to +award any prize to a winner in person. Competitions are void where prohibited or restricted by +law. Potential winners who are residents in jurisdictions where competitions require an element +of skill may be required to answer a mathematical test in order to be eligible to win a prize. + +Prizes + +No cash or alternative prizes are available, except that we (or the party providing a prize) reserve +the right to substitute a similar prize of equal or greater value. Prizes cannot be transferred +(except to a child or other family member) or sold by winners. Only the number of prizes stated +for the competition is available to be won and all prizes will be awarded provided a sufficient +number of eligible entries are received and prizes are validly claimed by the date provided in +connection with the competition, after which no alternate winners will be selected or unclaimed +prizes awarded. Unless otherwise disclosed in the prize description prior to entry, winners are +responsible for all costs and expenses associated with claiming a prize. All taxes are solely the +responsibility of each winner, although we reserve the right to withhold applicable taxes and +each winner agrees to complete any required tax forms. + +Your acceptance of a prize constitutes agreement to participate in reasonable publicity related to +the competition and grants us an unconditional right to us to use your name, town or city and +state, province or country, likeness, prize information and statements by you about the +competition for publicity, advertising and promotional purposes subject to applicable law and +regulations, all without additional permission or compensation. As a condition of receiving a +prize, winners (or their parents or guardians) may be required to sign and return an affidavit of +eligibility, liability release and publicity release. + + 6. Additional Provisions + +Submissions and Unsolicited Ideas Policies + +Our long-standing company policy does not allow us to accept or consider unsolicited creative +ideas, suggestions or materials. In connection with anything you submit to us – whether or not +solicited by us – you agree that creative ideas, suggestions or other materials you submit are not +being made in confidence or trust and that no confidential or fiduciary relationship is intended or +created between you and us in any way, and that you have no expectation of review, +compensation or consideration of any type. + +Claims of Copyright Infringement + +Notifications of claimed copyright infringement and counter notices must be sent to our +designated agent: + +Lance R. Griffin +The Walt Disney Company +500 South Buena Vista Street +Burbank, California 91521, USA +Phone: +1 818-560-1000 +Fax: +1 818-560-4299 +Email: designated.agent@dig.twdc.com + +We are only able to accept notices in the languages in which these terms are made available by +us. + +We will respond expeditiously to claims of copyright infringement committed using the Disney +Services that are reported to our Designated Copyright Agent, identified above, in accordance +with the U.S. Digital Millennium Copyright Act of 1998 (“DMCA”) or, as applicable, other +laws. With respect to Disney Services hosted in the United States, these notices must include the +required information set forth in the DMCA and described in detail here. + +7. BINDING ARBITRATION AND CLASS ACTION WAIVER + +PROCEEDINGS TO RESOLVE OR LITIGATE A DISPUTE IN ANY FORUM WILL BE +CONDUCTED ON AN INDIVIDUAL BASIS. Neither you nor Disney DTC will seek to have a +dispute heard as a class action or private attorney general action or in any other proceeding in +which either party acts or proposes to act in a representative capacity. No arbitration or +proceeding can be combined with another without the prior written consent of all parties to the +arbitrations or proceedings. + +You and Disney DTC agree to arbitrate, as provided below, all disputes between you (including +any related disputes involving The Walt Disney Company or its affiliates), that are not resolved +informally, except disputes relating to the ownership or enforcement of intellectual property +rights. “Dispute” includes any dispute, action, or other controversy, whether based on past, + + present, or future events, between you and us concerning the Disney Services or this Agreement, +whether in contract, tort, warranty, statute, regulation, or other legal or equitable basis. You and +Disney DTC empower the arbitrator with the exclusive authority to resolve any dispute relating +to the interpretation, applicability or enforceability of these terms or the formation of this +contract, including the arbitrability of any dispute and any claim that all or any part of this +Agreement are void or voidable. + +A. In the event of a dispute, you or Disney DTC must send to the other party a notice of +dispute, which is a written statement that sets forth the name, address, and contact +information of the party giving the notice, the facts giving rise to the dispute, and the +relief requested. You must send any notice of dispute to Disney DTC, 500 South +Buena Vista Street, Burbank, California 91521-7620, USA, Attention: Legal. We will +send any notice of dispute to you at the contact information we have for you. You and +Disney DTC will attempt to resolve a dispute through informal negotiation within +sixty (60) days from the date the notice of dispute is sent. After that sixty (60) day +period and not before, you or we may commence an arbitration proceeding. You may +instead litigate a dispute in small claims court if the dispute meets the requirements to +be heard in small claims court, whether or not you negotiated informally first. + +B. If you and Disney DTC do not resolve a dispute by informal negotiation or in small +claims court, the dispute shall be resolved by binding arbitration before a neutral +arbitrator whose decision will be final except for a limited right of appeal under the +U.S. Federal Arbitration Act. YOU ARE GIVING UP THE RIGHT TO LITIGATE +A DISPUTE IN COURT BEFORE A JUDGE OR JURY. Arbitration will be +administered by JAMS Mediation, Arbitration and ADR Services (“JAMS”) in +accordance with the JAMS Streamlined Arbitration Rules and Procedures (the +“JAMS Rules”). The JAMS Rules and instructions about how to initiate an +arbitration are available at www.jamsadr.com or 1-800-352-5267. Arbitration may be +conducted in person, through the submission of documents, by phone, or online. +Proceedings that cannot be conducted through the submission of documents, by +phone, or online, will take place in either Los Angeles, California or the borough of +Manhattan, New York, New York, whichever is more convenient for you; provided, +however, that if circumstances prevent you from traveling to Los Angeles or New +York, JAMS may hold an in-person hearing in your hometown area. You and Disney +DTC agree to submit to the exclusive jurisdiction of the federal or state courts located +in either Los Angeles, California or the borough of Manhattan, New York, New +York, whichever is more convenient for you, in order to compel arbitration, to stay +proceedings pending arbitration, or to confirm, modify, vacate, or enter judgment on +the award entered by the arbitrator. The arbitrator may award damages to you +individually as a court could, including declaratory or injunctive relief, but only to the +extent required to satisfy your individual claim. + +Except as provided above with respect to jurisdiction in Los Angeles, California and Manhattan, +New York, New York, nothing in this arbitration provision shall be construed as consent by +Disney DTC or its affiliates to the jurisdiction of any other court with regard to disputes, claims +or controversies unrelated to the Disney Services or this Agreement. + + Choice of Forum + +For all Disney Services NOT provided by The Walt Disney Company Limited the following +shall apply: + +Except to the extent required under applicable local law which requires that venue be in the +country of your residence, you agree that any action at law or in equity arising out of or relating +to these terms or the Disney Services that is not subject to arbitration shall be filed, and that +venue properly lies, only in the state or federal courts located in the borough of Manhattan, New +York, New York, United States of America and you consent and submit to the personal +jurisdiction of such courts for the purposes of litigating such action. + +Choice of Law + +For all Disney Services NOT provided by The Walt Disney Company Limited the following +shall apply: + +Except to the extent that applicable local law requires that the law to be applied for contracts of +this type be that of the country of your residence, these terms are governed by and construed in +accordance with the laws of the State of New York and the laws of the United States, without +giving effect to any conflict of law principles. + +Severability + +If any provision of these terms shall be unlawful, void or for any reason unenforceable, then that +provision shall be deemed severable from these terms and shall not affect the validity and +enforceability of any remaining provisions. + +Survival + +The provisions of these terms which by their nature should survive the termination of these terms +shall survive such termination. + +Waiver + +No waiver of any provision of these terms by us shall be deemed a further or continuing waiver +of such provision or any other provision, and our failure to assert any right or provision under +these terms shall not constitute a waiver of such right or provision. diff --git a/q119/random_k2/question.json b/q119/random_k2/question.json index 1bbca3a4fa063d47a4c4fb952d962fcb0137d9c0..d22e94626d49f425151a4ad695d9d15ba99d8afc 100644 --- a/q119/random_k2/question.json +++ b/q119/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "tacl_a_00660.pdf", - "web_d8f936ecd4367f91.pdf", - "web_4cf87d006f2e4a65.pdf" + "web_d44e9a235a572259.pdf", + "resize_r2_4.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q119/random_k2/random_1.md b/q119/random_k2/random_1.md index 2cfb2a95e81c5a4894ee70af1d1b3fdec18ec7c6..405f7d94774fb08f29f0d478b4b1f971f0f80612 100644 --- a/q119/random_k2/random_1.md +++ b/q119/random_k2/random_1.md @@ -1,1447 +1,705 @@ - b o s t o n c o l l e g e +Fixture and Store Designs that Draw Shoppers +into the Center Store to Capitalize on Sales -Exchange Student -Handbook + 10 Ways to Lift Center Store Sales -Office of International Programs +Sources Cited - We are delighted that you have chosen to attend Boston College as an ex- -change student. We value the partnership that we have developed with your -home university and look forward to welcoming you as a representative of -your institution. Your presence will add a significant element to the interna- -tional dimension of Boston College, and we hope you will take advantage of -every opportunity to share your culture with our community. +Cohen, Deborah A., and Susan H. Babey. “Candy at the Cash Register — A Risk Factor for Obesity +and Chronic Disease.” New England Journal of Medicine. (2012): n. page. Web. 21 Jan. 2013. +. -As an exchange student, you are given the opportunity to study at Boston Col- -lege for a semester or academic year as part of your academic program at -your home university. Please note that it is not possible to transfer to Boston -College to finish your degree. Boston College has a variety of graduate pro- -grams that will welcome your application once you have received your degree. +Sorensen, Herb. Inside the Mind of the Shopper. 1st ed. Upper Saddle River: Pearson Education, +Inc., 2009. -Please visit www.bc.edu for detailed campus information. For information on -the Office of International Programs see www.bc.edu/international. +Strom , Stephanie. “In Grocery Stores, the Perimeters Take Center Stage.” New York Times +10/08/2012, New York edition B1. www.nytimes.com/2012/08/11/business/in-grocery-stores-the- +perimeters-take-center-stage.html?pagewanted=all -We wish you all the best as you begin to prepare for your upcoming trip to -Boston. Please let us know if we can be of any assistance. +Underhill, Paco. Why We Buy: The Science of Shopping. 1st ed. New York: Simon & Schuster +Paperbacks, 2009. 3-27. -Bernd Widdig, Ph.D. Esther Messing -Director, Office of International Programs Assistant Director for +“Stop Shopper Leakage in Center Store.” Shopper Sight News. 70 (2012): n. page. Web. 21 Jan. 2013. +. - International Exchange Students +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. +Not responsible for errors or omissions in printing. -1 + 10 Ways to Lift Center Store Sales -www.bc.edu/international +If store perimeter provides “sizzle,” center store must serve “steak” - TABLE OF CONTENTS +Colorful, creative, winding and sometimes even whimsical in-store perimeter displays of +produce, sampling stations, deli, sushi and coffee bars, and endless walls filled with everything +from flat screen TVs to the hottest brands of clothing are no doubt here to stay. Engaging +perimeter presentations are among the latest trends in fashionable and attractive retail +merchandising that represent a sweet spot for today’s progressive merchandisers looking for fresh +new ways to connect with shoppers. - INTRODUCTION + However, while perimeter sales within many mid- to large-size shopping environments clearly +represent a vibrant growth trend, a similar merchandising discipline and focus should be applied +to the center store. After all, smart merchants understand that what’s good for the perimeter +should also be good for the interior, because what good is the sizzle of the perimeter shopping +zone without the steak -- and inherent profit margin -- of the center store? - BEFORE YOU LEAVE HOME + Unfortunately, center store is more often than not treated as a warehouse of separate products +on cold, uninviting rows of fixtures. While shoppers may find the products in center store they +need, they seldom find the solutions they truly want to make their lives more enjoyable or +productive. How will that new jacket or shampoo fit their lifestyle? How can that product, or +combination of ingredients, help them make a better or healthier meal? Why will that dog food +help their pet be healthier and happier? - Student Visa - Travel Plans - Medical Insurance - Living Expenses +Madix has introduced new products that improve merchandise visibility. Radius and Wraparound Shelves +used on end caps make merchandise down inline runs more noticeable because these shelves do not + utilize end panels, which normally obstruct merchandise as you look down a run. - YOUR FIRST DAYS IN BOSTON - - Getting to Boston College - On-Campus Housing - Off-Campus Housing - Hotels and Temporary Housing - Checking-in and Orientation - - ACADEMICS - - Full-time Status - Course Selection and Registration - Drop/Add Period - Faculty Advisors - Tutoring - English Classes for Non-Natives - Transcripts of Academic Work - - ADMINISTRATIVE ISSUES - - Student Identification Card - Medical Insurance - Health Services - University Counseling Services - Students with Disabilities - E-mail - - LIFE BEYOND ACADEMICS - - CONTACT INFORMATION +3 -2 +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales + +What’s actually happening in center store? + +Retail store perimeter growth has come mostly from the pull demand from customers wanting + +fresh, unprocessed foods and products that exude health, youthful appeal and entertainment. +Nowhere is the effect on center store sales more evident than in grocery stores and big food +brands such as Heinz, Kraft and Kellogg. Published reports indicate sales or volume declines +as large as 2.8-percent for some of these powerhouses due to the shifting consumer interest in +healthier non-processed food choices most often found on the perimeter of grocery stores. As +a result, packaged goods manufacturers are researching ways to leverage the popularity of the +perimeter through cross merchandising to make the center store -- along with their profitable +brands -- more exciting and appealing to shoppers. + + Unfortunately, center store sales within most retail environments have been decreasing in +recent years. The reasons why are akin to the proverbial elephant in the room. Rather than +address this huge challenge head-on, many retailers have instead decided to ignore it and focus +on what is working from a perceptual standpoint, which is the perimeter. Regardless of the center +store neglect, and using grocery stores as an example, a report by John Lewis, president, consumer +North America for The Nielsen Company, points out that the center store represents 73-percent +of total store sales and 77-percent of profit despite growing shopper leakage from people who +browse and think about a purchase but ultimately don’t buy. This imbalance of sales and profits +between center and perimeter can be addressed not only through SKU count, but also through +lower margins in the perimeter due to high labor costs and shrink. + +This tale of the two shopping zones is supported by a Center Store Mega Study of 2 million +individual shopping trips that found as many as a quarter of shoppers walk the center store yet +don’t buy anything (“Shopper Sight News”). In addition, during the average 13-minute shopping +trip, shoppers spend on average only two minutes in the center store. Clearly when analyzing +the chasm between center store sales and the average time spent there by shoppers, the +opportunities are enormous for progressive retailers to better integrate perimeter and center store +merchandising efforts targeted at growing customers’ overall baskets. + +So, if bold, new methods of merchandising are working to drive traffic for the perimeter, +why are retailers continuing to ignore the bread and butter represented by the center store? +Within this vast and extremely valuable internal battlefield, most stores still stubbornly stick +with outdated merchandising methodology including everything from shelving and lighting to +traffic patterns and category management. The dichotomous relationship that currently exists +between the perimeter and center store ignores how shopping patterns have shifted and, more +importantly, represents lost sales opportunities. + + When asked to comment, Blake Sloan, partner and store designer at the full-service +architecture and design firm, Heights Venture Architects L.L.P., noted that the center store is often +not addressed with the same creative thought process as the perimeter. + +“It’s extremely difficult to break away from product-centric category management and focus +efforts towards a consumer-centric shopper management,” said Sloan. “The potential profit power +represented by the center store is too valuable to simply ignore. Instead, we suggest retailers +envision the center store experience as a unique merchandising and marketing opportunity. +Retailers should enhance the center store with creative fixturing, lighting and product +identification as they would for the perimeter to create a holistic total store experience to build +customer loyalty and improve sales and profits.” 4 -5 -5 -6 -6 - -7 -8 -9 -10 -10 - -11 -11 -13 -13 -14 -14 -14 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -15 -15 -16 -16 -17 -17 + 10 Ways to Lift Center Store Sales -18 + As a New York Times article entitled “In Grocery Stores, the Perimeters Take Center Stage” +by Stephanie Strom recently stated, “Long, soldier-straight rows of shelves...have long been the +heart of the American grocery store but are now showing signs of the grocery equivalent of +atherosclerosis.” The same could be said of pretty much any mid- to large sized retailer offering +hundreds, if not thousands of SKUs on the floor. After all, is there anything less inviting than +staring down a narrow, 60-foot long aisle boarded by shelves upon shelves of cans and boxes that +all seem to blend together? -20 +The bigger, more important equation is how to create a more pleasant and inviting center store -www.bc.edu/international +that works with, if not mimics, the perimeter? Solving the problem of how to keep shoppers in +the center store longer than the average two minutes with a sexier, well-thought-out solutions- +oriented approach would go a long way toward moving more high-margin goods and better +leverage consumer needs to win more of their shopping trips. - CHECKLIST + While there is no single solution to the center store challenge, here are 10 smart tips that can +go a long way toward creating a holistic shopping experience that leverages the entire store and +drives up both sales and profits. - BEFORE YOU LEAVE HOME +Solution #10: Take the edge off -Apply for F-1 or J-1 visa from local US Embassy or Consulate -Make travel plans for arrival and departure from BC -Secure on-campus or off-campus housing. If necessary, arrange tem- + Old-school thinking is that if you trap the shopper and force them to pass by lots of product +facings, you’ll create more sales opportunities. Still, some retailers persist in creating canyons of +metal shelving and expecting customers to willingly funnel through them. In fact, the opposite +is true. Not only do shoppers dread entering long aisles, in their hurry to get through them they +often scurry past products they might otherwise be interested in purchasing. -porary housing or hotel +“Too many stores suffer from the ‘peek and run’ syndrome where shoppers simply refuse to +enter long rows bordered by high shelving on either side unless they absolutely have to,” said +Shawn Kahler, Madix Vice President of Sales and Marketing. “This effectively negates any hope +shoppers will spend any meaningful time in the center store, and it certainly doesn’t create a +pleasant in-store experience.” -Complete and submit Immunization and Health Forms to BC - University Health Services +Kahler says the answer is quite obvious: make aisles shorter, less linear and overall more appealing +to remove the intimidation factor in hopes shoppers will choose to walk through the store, enter +aisles, browse and shop. - AT BOSTON COLLEGE +“It’s actually possible to create many different types of unique, interesting and appealing aisles -Check in at the Office of Residential Life to pick up keys and get hous- +by using standard parts and fixtures rather than designing custom shelving that is often cost +prohibitive,” he said. “For example, Madix, Inc. offers retailers several options such as its curved +gondola system that actually use standard gondola and wall parts to better manage space and +generate a maximum return on investment.” -ing assignment or locate off-campus housing -Check in at the Office of International Programs -Check in at the Office of International Students and Scholars -Attend orientation. Dates, times, and locations are in your acceptance + He says that radius curves of 6 or 10 feet provide maximum flexibility to retailers who wish to +transform their outdated straight aisles into something more inviting and productive. There are +also much longer sweeping, curved fixtures that create a panorama effect of products that are +also pleasing to the eye and create much more product visibility than do long, straight and narrow +aisles. The key learning is that retailers have available to them the tools and professional expertise +enabling them to create better, more effective center stores. -packet +5 -Get student identification card, set up e-mail account, and establish -library account. This process will be explained at the orientation. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -3 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - INTRODUCTION - -Office of International Programs - -The Office of International Programs (OIP), located in Hovey House at 258 -Hammond Street, is your primary contact while you are studying at BC. The -exchange program that brought you here is administered jointly between OIP -and a comparable office at your home university. In addition to admitting you -to Boston College, we pre-register you for courses, assist with housing selec- -tion, and can assign you to a faculty academic advisor if you wish. We also -maintain regular contact with our counterparts at your home university. -Please feel free to come to us with any questions or problems before, during, -or after your semester or year at Boston College. - -Our office also sponsors the Exchange Mentor Program. This program aims -to match you with a Boston College student who has studied abroad at your -home university. Mentors maintain regular contact with their students. - -OIP works alongside numerous offices on the Boston College campus to en- -sure that you have the same access to facilities, services, and opportunities -as all other BC students. Other offices that you may come into contact with -include: the Office of International Students and Scholars, Student Services, -Health Services, Residential Life (which encompasses the On-Campus Hous- -ing Office and the Off-Campus Housing Office), and various academic de- -partments. - -Office of International Students and Scholars - -The Office of International Students and Scholars (OISS) is responsible for -issuing the immigration documents required for international students to -apply for a student visa. They also issue work authorization and advise on -other issues related to maintaining legal status in the United States. In col- -laboration with OIP, OISS organizes a mandatory International Student Ori- -entation each semester. - -In addition to these important services, OISS sponsors programs for interna- -tional students throughout the year, including welcome socials, Boston walk- -ing tours, the Thanksgiving Host Family Program, and many other activities. -They also provide undergraduate international students with International -Assistants, who are BC students serving as peer advisors to international stu- -dents. You will meet these International Assistants during orientation. +Madix’s Radius end caps married with curved Maxi Line draw shoppers into the center store. +The undulating shelves make merchandise stand out. Madix offers components to curve +entire fixture runs, fixture sections, or individual shelves. -4 +In combination with shorter and more organic-shaped aisles, Dr. Herb Sorensen, Ph.D., author -www.bc.edu/international - - Things to Know Before You Leave Home - -Student Visa - -As a participant of an exchange pro- -gram to Boston College, you will -most likely be entering the United -States either on an F-1 student visa -or a J-1 exchange visitor visa. To ap- -ply for a student visa you will need: - - - - - -the enclosed Certificate of Eligi- -bility Form I-20 for F-1 visa (or -Form DS2019 for J-1 visa) - -a passport valid for at least six -months after your scheduled -return to your home country - - proof of funding such as a letter - -from your bank - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -With these items, you must apply to -the appropriate Embassy or Consu- -late General of the United States in your home country. If you are applying -for an F-1 visa, be sure to complete the items on the bottom of pages 1 and 3 -of your I-20 before applying for your visa. - -It is important to understand that as an exchange student on a non- -immigrant F-1 or J-1 visa you are required to be enrolled as a full-time student -for the duration of your participation on the exchange program. Please refer -to the Academics section for more information on full-time status. - -Travel Plans - -As an exchange student, it is your responsibility to arrange for your arrival to -and departure from the United States. Please refer to the Academic Calendar -enclosed with your acceptance packet for important dates. - -You should plan to arrive at Boston College on a weekday (Monday - Friday) -between 9:00 am and 5:00 pm, when university offices are open to assist you. -On-Campus Housing staff may be available on the weekend prior to the start -of the semester. Details will be sent at a later date. - -If you are planning to live off-campus and have not secured housing prior to -departure you should arrive as early as possible (5-7 days) to look for housing -before the beginning of orientation. Please see p. 9 for more information. +of “Inside the Mind of the Shopper,” advocates staggering the fixture patterns themselves from +section to section to break up long sight lines. The outcome of this advice is that as a shopper +looks down any aisle toward the center store, they actually see the end cap of the next shelving +fixture or more inviting concave or radius shelving and not the other side of the store. -5 +Solution #9: Drop the height of shelving -www.bc.edu/international - - Also, please be aware that the following immigration regulations are in effect -for all F-1 and J-1 students: - - New international students will not be permitted to enter the U.S. more -than 30 days before the start date of their program as indicated on the I- -20 or DS2019 form. - - Upon completion of studies (i.e. the end date listed on the I-20 or - -DS2019), F-1 students are permitted to remain in the U.S. for 60 days -and J-1 students are permitted to remain in the US for 30 days. - -Additional immigration information is available at the following web site: -http://www.bc.edu/offices/odsd/intercultural/immigration.html. - -Medical Insurance - -The Commonwealth of Massachusetts requires all full-time students (both -American and international) attending colleges and universities in Massachu- -setts to subscribe to a US-based insurance plan. You will be automatically -enrolled in the BC medical insurance upon your admission to the university. -If you do not want this insurance, you must submit proof of insurance with a -provider that has a US office after your arrival. Please see p. 15 for more infor- -mation on BC medical insurance. - -Living Expenses - -Past exchange students -have been surprised by -the cost differences be- -tween their home coun- -try and the Boston area. -When estimating the -costs of your studies -and living in Boston, be -sure to budget for -housing, meals, trans- -portation, books, enter- -tainment, and any trav- -els you plan to take be- -yond the local Boston -area. Specific informa- -tion on costs can be -found on the Estimated -Expenses page in the -back of this handbook. - -www.bc.edu/international + Merchants seem to love using vertical space. The higher, the better, right? Actually, according +to Dr. Sorensen, people do not prefer to shop tall shelves crammed with products in the center +store. Instead, they respond best when faced with shelving that’s 66 inches or lower and 15 to 20 +feet long or not longer than 30 feet. Such shelving ensures customers can easily see and navigate +their way through the store without feeling trapped or lost. An exception would be a retailer such +as JCPenney which has begun using tall walls on the perimeter of its stores where they do not +interfere with sight lines to promote fashion accessories. 6 -s -t -t -e -s -u -h -c -a -s -s -a -M - -, -y -b -n -a -r -G - - YOUR FIRST DAYS IN BOSTON - -Getting To Boston College - -Most students come to Boston -by flying to Logan International -Airport. Getting to Boston Col- -lege from Logan is not hard and -can be done in many ways: - -Taxi and Airport Shuttle - -The simplest, yet most expen- -sive, way to get to Boston College -is by taxi. - -Another alternative is to utilize -one of the private airport shuttle -services that operate from Logan -and will drop you off at Boston -College. Prices vary by company. -You might need to make a reser- -vation in advance. For a listing of shuttle services available at Logan, visit -www.massport.com/logan. A listing of private car services can be found on -the internet by Googling “livery services in Boston.” - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Public Transportation - -Finally, you can utilize public transportation to get to Boston College. The -subway is known locally as the “T.” For more information consult the Massa- -chusetts Bay Transportation Authority website at www.mbta.com. - -To take the "T" to Boston College, take the airport shuttle bus to the Airport -subway stop. Take the “Blue Line” subway from the airport to the -“Government Center” subway stop, where you should change trains to the -“Green Line.” Please note that there are four branches of the Green Line! Be -sure to take the “B-Line” train outbound, destination "Boston College," and -get off at the last stop. A public transportation map is included in your accep- -tance package. Obtain a reusable “Charlie Card” at the subway station to re- -duce your T fare. - -For more information on Logan International Airport and traveling into Bos- -ton from the airport, please consult www.massport.com/logan. +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -7 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - Moving Into On-Campus Housing - -Boston College has a number of residence halls. If you will be living on- -campus, we try to place you in residence halls with kitchen facilities. Students -living in rooms without access to kitchen facilities are required to purchase -the Boston College meal plan, which pays for meals in the campus dining -halls throughout the semester. Students living in rooms with kitchen access -have the option of purchasing the meal plan. If you do not wish to purchase -the full meal plan, you should strongly consider the “self-catering” option, -which allows you to purchase "Dining Bucks" in smaller amounts throughout -the semester. For more information on meal plans and Dining Bucks visit the -Dining Services web site at www.bc.edu/offices/bcds. - -Please be aware that sheets, blankets, pillows, towels, etc. are not provided in -the residence halls. These items can be purchased inexpensively in Boston, -so you may not want to take up luggage space with bedding and towels. - -The cost for on-campus housing varies by residence hall and room arrange- -ment. You will be assigned to available housing and will not be able to re- -quest a specific residence hall. Your BC student account will be billed once -your housing assignment is confirmed. You are then responsible for paying -for your housing upon arrival at Boston College. - -Every student on campus has a private phone number and a phone jack in -her/his room. You must provide your own telephone, which can be pur- -chased inexpensively in nearby stores. Local telephone service and voice mail -are free. Campus rates for long distance and overseas calls are very competi- -tive. Long distance calls are billed to your student account. Many students -also use cell phones and Skype for long distance and international calls. - -All on-campus rooms have free internet access, available through either an -Ethernet cable or wireless device. Contact 617/552-4357 with questions. - -Students with on-campus housing can move into their rooms on the first day -of the International Student Orientation for Fall/Full Year students, or on the -day before the Exchange Student Orientation for Spring students (see the -enclosed Academic Calendar). Be sure to arrive between 9:00 am and 5:00 -pm, when the housing office will be open to check you in and give you your -room key. The On-Campus Housing Office phone number is 617-552-3060. - -If you plan to arrive outside of normal business hours, please notify OIP as -soon as you make your travel plans so we can inform the housing office. -Upon arrival at BC, you should go directly to your residence hall and ask for -the RA (Resident Assistant) on call. The RA will let you into your room and -you can pick up your key on the next business day from the Residential Life -office. +In addition, traditional straight-faced shelves in most stores are designed to hold a maximum +capacity of products from top to bottom, but they are not shopper friendly. Such a design stifles +the ability to view products on lower shelves, often forcing the shopper to take a few steps back +into what may be a crowded aisle to see what’s on the bottom shelves. Such fixtures also visually +create a closed-in, claustrophobic feeling as shoppers walk down an aisle. -8 + One solution is to install pyramid-style fixtures on both sides of an aisle. Such fixtures have top +shelves that are recessed from the shopper by about 16 inches, giving shoppers nearly three full +feet of apparent extra aisle width. This allows higher-velocity products to be inventoried on lower +shelves while featuring higher-priced or lower-velocity items toward the top where they can gain +more exposure. -www.bc.edu/international - - Off-Campus Housing - -If you will be living off-campus, you -will need to make your own housing -arrangements. The Boston College -Off-Campus Housing Office has up- -to-date lists of apartments and -roommate openings near Boston -College and acts as a referral source -for rooms in private homes. Your -name will be given to the Off- -Campus Housing Office and they -will contact you and assist you in -your housing search. The staff can -advise you on suitable arrange- -ments, convenient locations, etc. -However, they cannot make ar- -rangements on your behalf. If you -choose to utilize the above services, -it is possible that you may secure -housing prior to your arrival in the -United States. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Because apartment rentals in the BC area are expensive, many international -students either share an apartment with other students or live with families in -private homes in the surrounding area. We advise all exchange students who -will be seeking off-campus housing to arrive at least 5-7 days before orienta- -tion in order to have enough time to make the necessary arrangements. You -should also make necessary hotel arrangements prior to arrival (see the Ho- -tels and Temporary Housing section for suggestions). - -Finally, be sure to carefully read your lease or other agreements to rent an -apartment before signing and make certain that you understand the terms. If -you are unclear about what a particular document means, consult the Off- -Campus Housing Office for advice. There are laws to protect tenants in Mas- -sachusetts, so be sure to find out what your rights are in case you run into a -conflict with your landlord. - -The Off-Campus Housing Office web site offers helpful hints to assist you in -your search. To access their web site, go to www.bc.edu/housing and click on -the Off-Campus Housing link on the left sidebar. - -Once you have settled into your accommodations for the semester, be sure -to inform OIP of your address and phone number so that we know how to -contact you when necessary. You will receive a local housing notification form -in your orientation packet. + A less radical alternative to pyramid fixtures is an offset shelving design that moves one or +more of the top shelves back in small increments between 8-12 inches. While slightly less effective +than pyramid fixtures, offset shelving still helps reduce the train-tunnel effect and allows the use +of sloped signboards that can be used for various in-store marketing devices such as digital media. -9 +Regardless of the type of fixture that’s used, Dr. Sorensen points out that there is a delicate +balance between trying to capture the shopper’s time, holding a shopper at a product display, and +closing the sale. Trying to sell too much within a small space can actually harm sales conversion, +because the longer it takes a shopper to make up his or her mind, the more angst they feel, the +more likely they are to abandon that area of the store, and the less time they are apt to linger in +other areas of the store. The rule of thumb should be convenience, which translates into fast. -www.bc.edu/international +Madix’s product lines are available in numerous sizes, +so a retailer can deploy fixtures that best meet its sales objectives. - Hotels and Temporary Housing +7 -If you will be arriving in Boston early to find off-campus housing or have ac- -companying guests who need temporary housing, you can obtain lists of tem- -porary accommodation options from the Off-Campus Housing web site at -www.bc.edu/housing (click on the Off-Campus Housing link). +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -Checking In + 10 Ways to Lift Center Store Sales -As an exchange student at Boston College, there are two offices with which -you must check in during normal business hours (Monday to Friday, 9am - -5pm) upon arrival at Boston College: the Office of International Programs -(OIP) and the Office of International Students and Scholars (OISS). +Solution #8: Widen the aisles -Upon your arrival, please come to OIP in Hovey House and check in so that -we know that you have arrived safely. + Using pyramid or offset fixtures are simple methods to effectively open up aisles visually and +literally from the feet up to head height, but another more obvious solution is to simply widen the +entire aisle. Some retailers serve up just enough room for two carts to pass by one another in an +aisle, while others make it almost impossible for even a single cart to pass between displays. While +the solution will vary between retail formats, the outcome should always cater to what makes the +shopper most comfortable and put them in a buying, not evacuation, mode. -You then need to check in with OISS. All international students are required -under the terms of their visas to check in officially at the Office of Interna- -tional Students and Scholars for immigration purposes. You will need your -passport with the white I-94 card attached, as well as your I-20 or DS2019 -form in order to check in. +In general, shoppers equate open space with the freedom to shop. The best way to generate -Orientation +traffic is to allow traffic to flow without restrictions. A shopper at a store with tall, narrow aisles +is less likely to enter the aisle if it has but one other shopper in it, much less several. With such +aisles, a single shopper with a cart stopped in an aisle can occupy as much as 10 feet of shelf facing, +effectively blocking that merchandise from other potential buyers. Obviously, this leads to far +fewer sales opportunities. -The Office of International Programs, in collaboration with the Office of Inter- -national Students and Scholars, sponsors an orientation program for all new -international students which includes new exchange students. For fall and full -year students orientation takes place the week before classes begin in Sep- -tember. Spring students have orientation prior to the first week of classes in -January. The orientation will provide you not only with important information -about Boston College, but also with valuable opportunities to meet other ex- -change students, staff members, and Boston College students. + On the other hand, a shopper at a store with aisles able to accommodate multiple shoppers +shoulder-to-shoulder is less likely to turn away. In this scenario, even if another shopper is camped +in front of a shelving unit another shopper can stand to the side and browse. The key is to allow a +relatively free flow of multiple shoppers through any aisle at any time. -An important part of this program is the English Placement Test, adminis- -tered to non-native English speakers. All international students, including -exchange students, are required to take this test if their first language is not -English. + Dr. Sorensen states that open space combined with short and low fixtures are open invitations +to shop. Likewise, he says, “Narrow, crowded aisles like packed highways can lead to social +pathology and even ‘aisle rage,’” whereby shoppers become angry at one another for blocking +each other or even blame the store for failing to be customer friendly. Ironically, retailers who do a +good job merchandising their perimeters typically do so with wide aisles, low fixtures and a sense +of openness. So why not apply the same principles to the center store experience where upward +of 75-percent of all sales occur? -Please see the OISS letter in your acceptance packet for additional informa- -tion about the orientation. +Madix designs its products to more efficiently utilize space, +which means a retailer can devote more space around fixtures to its shoppers. -www.bc.edu/international +8 -10 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales - ACADEMICS + Another, less obvious reason to expand aisles is what Paco Underhill, author of “Why We Buy: +The Science of Shopping,” calls the “butt-brush effect.” During research his firm conducted +for Bloomingdale’s department store, they inadvertently observed shopper interaction with a +clothing fixture near a ground-floor entrance they were monitoring. The fixture was adjacent to +a high-traffic aisle, and they saw that customers interacting with the fixture would typically get +“butt-brushed” from behind by people heading into or out of the store. After a few such bumps, +shoppers abandoned the fixture and moved on without buying. As a result, sales from the fixture +were well below expectations. After moving the fixture to another location that eliminated +the possibility of butt-brush, the retailer found that sales from the fixture immediately and +substantially increased. It simply boiled down to location. -Full-Time Status +The question retailers should be asking themselves is how often their customers are spooked -All international students in the US are required to maintain full-time student -status for the duration of their studies in the US. Undergraduates must be -enrolled in a minimum of 12 credits per semester. Graduate students must -be enrolled in a minimum of 9 credits per semester. Undergraduate exchange -students should expect to take between 12 and 15 credits per semester. -Graduate exchange students should expect to take between 9 and 12 credits -per semester. Failure to maintain full-time status will affect your legal status -in the United States, so it is very important to be enrolled in the appropriate -number of credits. +by being too close to one another or being brushed against other shoppers trying to navigate +around them? The solution may be something as easy as a realignment of a fixture or a more +complicated readjustment of the entire aisle space. -Selecting and Registering for Courses +Solution #7: Go mobile -We attempt to register you for the classes you requested on your application, -but please keep in mind that you must be flexible about what courses you will -take. We cannot guarantee that you will get all of the classes you request. If -you need to make any alterations to the courses you selected on your applica- -tion, please e-mail OIP at oipexchange@bc.edu. +Part of the challenge of store fixtures is that they are by definition “fixed,” meaning they +often are anchored in place and unlikely to be moved outside of major store resets. Rather than +be entirely tethered by immovable fixtures, some of the world’s most successful retailers have +embraced the more fluid flexibility of mobile displays to complement or, in radical cases, supplant +stationary displays. -When selecting courses, there are a number of factors that must be taken -into account. First, be sure to choose courses that correspond to the level at -which you were admitted to BC. Your acceptance letter states if you were ac- -cepted as an undergraduate or graduate student. Second, be sure to select -courses that correspond to your level of experience with the subject matter. -Each course listing on the BC course web site gives the course level. Levels -for all BC courses are: + Granted, fixtures on castors are not a new concept, but the use of such shelving is not +necessarily widespread, especially in the center store area. An example of a retailer that lives by +mobile fixtures is one of Europe’s most successful shoe store chains, Leder & Schuh’s. L&S has +virtually no immoveable fixtures. Instead, every eight weeks or so the latest shoe fashions are - CORE: Fulfills university core requirements. Fairly basic courses open to +Madix offers a variety of caster kits that convert conventional shelving systems into flexible mobile runs. -undergraduate students from any field of study. +9 - ONE: Undergraduate elective courses open to students from any field of +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -study. + 10 Ways to Lift Center Store Sales - TWO: Undergraduate courses designed primarily for those specializing in +displayed in a new retail theater created by set designers using mobile fixtures, lighting and props. +This has helped cement their reputation as a progressive, forward-thinking retailer with its finger +on the pulse of the rapidly changing world of fashion. Other retailers, such as RadioShack, Ross +Dress for Less, BestBuy, and numerous department and grocery store chains, have experienced +their own successes with the flexibility of mobile fixtures. -that field of study. + Mobile fixtures give retailers the ability to easily change the store layout to fit the changing +needs of shoppers, accommodate special sales and offers, and even cater to “event” opportunities +with integrated solutions for big events, holidays or celebrations. Have perishables that are about +to go out of date? Put them on mobile fixtures and place them near checkout lines or in natural +product adjacencies to stimulate purchase and reduce shrinkage. Coming up on “Go Pink” breast +cancer awareness day? Use mobile fixtures and round up different pink merchandise such as +clothing and accessories near front entrances for people who want to support the cause. Have +overstock of something that might take weeks to clear out? Use mobile fixtures to create sales +opportunities throughout the store to get shoppers to purchase what you want to sell. The +possibilities are endless. - THREE: Upper-level courses for undergraduate and graduate registration. -Most suited to those with background knowledge of the general subject -matter and specializing in that field of study. +Solution #6: Shorten or break up long runs -11 + Hand in hand with banishing railroad tunnel aisles, Dr. Sorensen suggests rethinking the +concept of aisles altogether. He advocates that the length of shelving from one end to the other +be between 15-20 feet and not longer than 30 feet. -www.bc.edu/international +Shorter aisles in length and height accomplish multiple purposes. First, they are more +inviting to navigate and provide a natural break between product facings. Next, if the fixtures +are of varying heights, shoppers can actually see across multiple shelves and gondolas. Shorter, - GRAD: For those admitted as graduate students. Requires advanced knowl- -edge of general subject matter. Please note that some of these courses are -open to undergraduate students with the permission of the professor. +Madix’s shelving lines can be ordered to accommodate shorter run lengths. -Undergraduate students should select courses with a level of Core, One, Two, -or Three. Graduate students should select courses with a level of Grad or -Three. +10 -Course listings on the BC course web site (www.bc.edu/courses) look similar -to this: +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales + +staggered fixtures also serve as artificial road bumps that force shoppers to slow down, navigate +left, right and around product displays, and be more aware of what they’re passing. Also, one of +the most potent sales opportunities exist with end caps, and the shorter the fixtures in length, +and the greater the number of fixtures that are used, the number of end cap merchandising areas +increases exponentially. + + A similar outcome can be accomplished by interspersing concave or radius shelving that +softens up center store aisles and visibly pulls customers in. Rounded or organically-shaped +shelving units such as the Madix Y-Gondola can add a pleasant visual impact to the sales floor and +create multiple focal points allowing retailers to highlight important items, sales or promotional +offers. Such shelving also actually improves the visibility of more SKUs when compared to straight +shelves, because they are typically viewed from more of a frontal perspective than through a side +view while walking down an aisle. + + Of note, some retailers, such as Walmart, have actually reverted back to taller, warehouse-style +shelving and what has been described as shelves stacked full of many different products, narrow +aisles and conspicuous pallet stacks. Why? Because their sales results support the notion that +such a shopping environment better supports their brand proposition of being a value-oriented +retailer. + +Explaining this perspective in a national best seller book entitled Why We Buy: The Science of +Shopping, Underhill says, “Historically, the more a store is packed (with merchandise), the more +people think of it as value — just as when you walk into a store and there are fewer things on the + +floor, you tend to think they’re expensive.” + +This simply means that it’s critical for retailers such as Walmart to understand the mindset +of their customers to know what it is they want. Regardless of a retailer’s brand proposition – +whether it be value, quality, selection or brand names – they must strike a balance between the +volume of products displayed and the physical and visual presentation to create an optimum +buying atmosphere for its customers. In other words, what works for a discounter may not work +for a specialty retailer, and vice versa. + +Solution #5: Integrate specialty merchandisers + +Specialty merchandisers are an ideal mobile way for retailers to experiment with product +displays and presentations on the sales floor without making major, permanent commitments to +space or marketing. + +Retailers can experiment with specialty “solution stations” that tie product categories together + +without totally realigning product groupings or fixtures. Common fixtures for such specialty +presentations include the use of T-Racks and mobile caster kits. + + Grocers could consider staging certain ethnic food nights, weeks or months to stimulate trial +and impulse purchases by shoppers. For example, a mobile fixture or end cap could be ideal for +“Italian Night Solutions” complete with recipes and suggestions for combinations of products that +otherwise would be spread throughout the store: pots, pans, utensils, spaghetti, lasagna, pizza +kits, sauces, spices, olive oil, cheese, meat and vegetable appetizers, bread, desert, wine and even +flowers. Another possibility could be “Pamper Your Pet” with a selection of high-end food, treats, +supplements, collars, bowls and toys. Evergreen events such as birthdays and anniversaries also +lend themselves to turn-key specialty merchandisers featuring apparel, gift boxes, cards, candles, +desserts and even alcohol. -Index +11 -Course # +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -Credits + 10 Ways to Lift Center Store Sales -Level +Madix manufactures specialty merchandisers to suit any market segment. Luxe, +a new display that features an attractive round upright and standard Madix slotting, +accepts Madix’s thousands of shelves and accessories. -Title +Retailers should experiment with integrated presentations for different topics and locations -Instructor +to see what does and doesn’t resonate with their customer base and tweak them accordingly. +In addition, Dr. Sorensen points out that something as simple as scattering static or mobile +merchandisers around the store in non-obvious locations almost always increases sales of those +products. For example, placing a merchandising display of cookies outside of the bakery, or +t-shirts and socks near the check-out queue, can stimulate incremental sales of those items outside +of their natural environment. The beauty of such displays is that they lend themselves +to experimentation. -Bldg +Solution #4: Make smart use of in-store marketing -Room +Point-of-sale marketing is found to improve the find-ability of merchandise and usually leads to +improved brand recognition. However, it’s important to effectively judge the use of such point-of- +sale material and be smart about leveraging the sales of merchandise the retailer wants to sell, not +necessarily what the manufacturer or supplier wants to push. After all, why take up valuable shelf +space and fight for shoppers’ attention for a low-margin or low velocity item or product category? +It’s better to focus on merchandise that has a higher profit margin, introduces new or innovative +products to customers, or supports the retailer’s particular point-of-difference (better selection, +higher quality, well-known brand names, etc.). -Time + Digital print and video signage are excellent ways to cut through the clutter and grab shoppers’ +attention in the noise of the typical retail environment. Such signage can be designed to educate, +entertain and engage shoppers with key messages they might not otherwise see. The Bissell +brand of vacuum cleaners in Lowe’s Home Improvement Stores is one example. Their use of digital +and video signage on end caps and mid-aisle displays stop shoppers in their tracks to deliver short, +concise sales messages about the benefits of clean floors and carpets and how Bissell products +deliver the best outcome and value. -9398 +12 -BI200 01 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -0569 + 10 Ways to Lift Center Store Sales -HS181 03 + Other, less extreme example of digital signage and even low-tech coupon and recipe pads and +dispensers are also effective in garnering attention for specific products and brands. Products +such as Lumicanopy by Madix, Inc. use LED lighting to illuminate graphics and promote categories +or brands. The typical Lumicanopy consumes little power, emits virtually no heat, exhibits uniform +light distribution that’s visible from many feet away, and is reusable over several years. Other +displays such as the Madix Easel SB retail shelf and point-of-sale display educates and entertains +shoppers with relevant information and does not protrude into aisles to interfere with shoppers. +The key is to be smart about what and how information is communicated to ensure its effect on +shoppers is maximized. -3 +Being able to deploy large format, illuminated graphics either in Madix’s canopies or within +the fixtures’ surfaces facilitates in-store marketing initiatives. The bright, colorful graphics +appeal to shoppers across the store. -3 +Solution #3: Turn up the lights -TWO +Lighting is one of the most important -- yet probably one of the least understood -- -Biology +merchandising techniques in the retail environment. Done correctly at the point-of-sale with +fixtures such as the Madix LEDge Light Versa and the Beam LED alternative to fluorescent canopy +lighting and retailers can improve mood, create the desired ambiance, and more importantly +create a measurable impact on shopping behavior. -DEPT + An article published in InformeDesign by Delores (Dee) Ginthner, an associate professor in +the Interior Design Program, Department of Design, Housing and Apparel at the University of +Minnesota, states several basic truisms that must be addressed when lighting a space to ensure +both psychological and behavioral issues of users are met. This includes the observations that +lighting must ensure safe navigation through a space; people always follow the brightest path; +brightness can focus attention; people like to face walls that are illuminated; and lighting can +cause people to alter body positions. -ONE +13 -American -Civilization +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales -Singleton + Ginthner further states that criteria created several years ago by Dr. John Flynn to evaluate +lighting remain true today. This criteria states that designers must use cues to determine the users’ +subjective response to space lighting in order to create space that makes users say something is +pleasant versus unpleasant, public versus private, spacious versus confined, relaxed versus tense +and visually clear versus hazy. -MWF -11 -TTH -9* - -In these examples, the index is the number used to register for the course -(you must include this number on your Course Registration Form). The -course number includes a two-letter abbreviation for the department (e.g. BI -= Biology, HS = History), followed by the three-digit course number. The two -numbers following the course number indicate the section. Popular courses -such as HS181 are offered in different sections to keep class sizes small. In -the example above, the course selected is section 3 of HS181. Each of the -courses in these examples carries a 3-credit load, which is the most common -load for university courses. It is your responsibility to make sure that you -have had all prerequisites required for any course that you are requesting. - -Courses typically meet several times a week. Classes that meet Monday, -Wednesday and Friday (MWF) are generally 50 minutes long, while Tuesday, -Thursday (TTH) classes are typically 75 minutes. These classes are indicated -in the course listing with an asterisk (*). - -Some courses may require co-registration for a supplementary lab or discus- -sion section and this will be stated in the course listing. Enrollment in some -courses may be restricted to majors only or may require department permis- -sion prior to registration. We are not able to pre-register you for restricted -courses, but you will be able to appeal to the department or professor after -you arrive to ask for special permission to enroll in such a class. Likewise, -undergraduates who seek to enroll in graduate-level courses will need to seek -direct permission from the instructor after arrival in Boston. It is fairly un- -usual for undergraduates to obtain approval for graduate courses and you -should not count on being able to do so. - -www.bc.edu/international + Needless to say, the science of lighting and how it applies to curved fixtures, pyramid or +recessed shelving, wider and shorter aisles, end caps, on-shelf merchandisers and even mobile +fixtures is critical to maximizing the impact and desired outcomes. The days of one-size-fits-all +overhead or fluorescent canopy lighting are being replaced by new efficient alternatives such +as environmentally friendly LEDs that produce light intensity similar to T5 fluorescent fixtures +while using only half the energy and lasting nearly three times longer. Lighting considerations +should be incorporated into the initial store design at the onset by true lighting professionals who +understand shopper psychology and how best to use lighting to stimulate sales. -12 +Madix is the first store fixture manufacturer to design energy-efficient, +point-of-sale lighting products that are specifically designed for store equipment +applications. Madix’s lighting solutions either attach underneath shelves, or they +replace the fixtures’ surfaces to pin-point light (e.g., back panels or shelf skins). - Finally, please note that many courses “close” during the early registration -period. Each course is listed in the computer with a maximum number of stu- -dents who can enroll. Once that limit is reached, the computer will not allow -any additional registration in that course. If you find that a course you need is -closed, you will have to appeal to the professor (either by e-mail prior to your -arrival or after you arrive) for an “override” to allow you into the class. This is -not uncommon and most professors are sympathetic to reasonable requests -from exchange students. However, it is not guaranteed that you will receive -an override for a closed course. - -Please look for classes offered in the semester in which you will be attending -(e.g., Fall, Spring) even if the list is for the previous year. For the most current -listing of available courses, see www.bc.edu/courses. - -Drop/Add - -The first week of classes at the beginning of each semester is called the -“Drop/Add” period. At this time you can make changes to your course sched- -ule, adding and/or dropping courses as needed. Don’t worry if you do not -have all of the courses that you need by the time classes start. You will have -this period of time to make adjustments to your course schedule, seek over- -rides as needed, etc. - -Your OIP exchange advisor can help you with most registration needs, includ- -ing add/drop. Registration matters can also be handled at Student Services in -Lyons Hall or on the computer via AGORA. You will receive more specific in- -formation about these resources upon arrival. In the meantime, refer all -course requests to OIP. - -Faculty Advisors - -You may request a faculty advisor for -assistance with academic planning for -your semester or year at BC. We will -make advisor placements based upon -your field of study. If your advisor at -home already knows a particular pro- -fessor at BC, or if you have already -made contact with a professor here -and you would like him/her to be your -advisor, please let us know. If you re- -quest a faculty advisor, you will be -required to meet with them by ap- -pointment during the first week of -classes and throughout the semester -as necessary. +Solution #2: Optimize traffic flow -13 +The art of converting visitors into shoppers and shoppers into buyers is accomplished in no -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -www.bc.edu/international - - Tutoring - -The Connors Family Learning Center offers free tutoring for more than 60 -undergraduate and some graduate courses, including writing and English -as a Second Language. Intensive, small-group supplemental instruction -sessions are available in several courses, and tutoring and assistance in -improving academic writing. For more information about services provided -by the Learning Center, go to www.bc.edu/libraries/centers/connors/. - -English Classes for Non-Native Speakers - -If your native language is not English you will be required to take a place- -ment examination during the Orientation in the fall or during the first week -of classes in January. If necessary, students will be recommended for a par- -ticular course based upon the results of this examination. The Program in -English for Foreign Students is based in the English Department and offers -several courses in college-level English for non-native speakers. Recent -course offerings include: - - EN 118 Essential of English Composition: extensive practice in writing - -and editing academic prose - - EN119 The Craft of Writing: practice writing academic essays and expo- -sure to English rhetoric, with emphasis on written analysis and the logi- -cal support of ideas - -In order to enroll in one of these courses, you must have taken the place- -ment test. - -Transcripts of Academic Work - -At the end of your time at Boston College, we will send an official transcript -of the academic work you completed to your home university. Before we -can release your BC transcript to your home university, however, you must -clear any outstanding debts to Boston College that you may have. Be sure -that all long-distance phone calls, on-campus housing fees, insurance, -meals, etc. have been paid for. - -Your BC student account must show a zero balance before you leave Bos- -ton or we will not be able to release your transcript. - -www.bc.edu/international +small part through optimizing traffic flow through the store. Dr. Sorensen suggests there are +several changes, some subtle and some overt, that can make a major impact on a store’s visceral +appeal while exposing shoppers to buying opportunities that may have gone undiscovered in a +traditional linear layout. 14 - ADMINISTRATIVE ISSUES - -Student Identification Cards - -If you submitted the required passport-size photographs with your Exchange -Student Application you will receive your student ID card during orientation. -If you did not send us photos, you will be able to have your ID card made at -Student Services in Lyons Hall following orientation. - -Your Boston College student identification card (“Eagle ID”) is a very impor- -tant item for you to have. In addition to identifying you as a BC student, the -card enables you to check books out from the library, to make purchases at -the BC bookstore, to access your student record and account from “U-VIEW” -computers around campus, and to make use of the BC Flynn Recreation -Complex (fee required), among other services. Be sure to keep your Eagle ID -with you at all times and do not lose it. A replacement fee will be charged by -Student Services for lost cards. - -Medical Insurance - -As explained in the Things to Know Before You Leave Home section, the -Commonwealth of Massachusetts requires all students enrolled in a Massa- -chusetts college or university to be covered by a US-based medical insurance -plan. Therefore, you have been automatically enrolled in the Boston College -medical insurance policy which provides adequate coverage during your stay -at BC, including travel within and outside the US during your period of cover- -age. Your student account was billed upon registration at BC. As an exchange -student it is your responsibility to pay for your insurance, which can be done -at Student Services. Full year students will be automatically re-enrolled in the -insurance plan for their second semester. You can find more information on -the BC medical insurance coverage on the BC web site at www.bc.edu/ -offices/uhs/services/insurance. - -If you do not want to be covered by BC medical insurance, you must be cov- -ered by insurance that is US-based or has a US contact office. You must have -proof of this insurance. If your insurance is comparable, you will be allowed -to waive the BC insurance and your student account will be credited accord- -ingly. Please note that it is not possible to have your medical insurance evalu- -ated before you arrive at Boston College. +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales + + One idea requires a significant reconfiguration of fixtures to change their orientation and +layout to the front of the store. Dr. Sorensen supports the theory that when shoppers within +the interior of a store look down aisles and see checkout registers, especially ones with no lines, +they subconsciously tend to hurry through their shopping experience so they can quickly check +out before the lines fill up: a sort of quick in, quick out mentality. To combat this and reduce the +“railroad tunnel” visual effect, he suggests shifting the aisles at a 45-degree angle to the front of +the store. An important benefit of such a design is that as shoppers enter the store they will be +approaching aisles from a less acute angle, which will make them more inviting to enter. + + Another way to increase traffic flow and lure shoppers is to determine new and relevant +product categories and bring them into the center store. Along the way and throughout the entire +store, the goal should always be putting the “right” products in the direct path of the customer. +Dr. Sorensen says, “There are patterns of movement... (in a store) that are not driven by products +as much as by open spaces and natural flows.” By observing, determining, and directing traffic +flow with smart shelving fixtures, lighting and merchandising materials, it’s possible to conduct +“anticipatory retailing.” This is when the retailer anticipates and meets the needs of shoppers +rather than making shoppers find things on their own. + +In support of an enhanced traffic flow, Dr. Sorensen advocates putting high margin + +merchandise rather than discounted goods in the customer’s path. Research shows that the most +important determinant for sales is based not on price, but on the placement of store fixtures +displaying those goods. Putting highly discounted merchandise such as 75-percent-off sweaters +or two-for-one cereal offers in the customer’s path may actually be counter-productive, because +it could prevent customers from venturing deeper into the center store and discovering higher- +priced, if not more desirable goods. + + Dollar General spokeswoman Mary Winn Gordon is quoted alongside Underhill in the New +York Times article as saying the chain store employs virtual “speed bumps” of related merchandise + +Madix’s array of fixture choices equips a retailer with options to tweak store designs to improve traffic flow and sales. 15 -www.bc.edu/international - - Health Services - -The Health Services Center at -Boston College offers a range -of services to BC students. -Staffed by first-class doctors, -nurse practitioners, and other -medical staff, the Health Ser- -vices Center is equipped to -handle health emergencies, -minor illnesses and questions -about healthy living. While the -general health services offered -are comprehensive, if you re- -quire specialty care you can -be referred to a qualified spe- -cialist. - -All undergraduate exchange -students who live on-campus -are required to pay the Health -Services fee. Exchange stu- -dents who live off-campus are -also entitled to this service for the same fee. Off-campus students can ei- -ther pay in advance, or the first time the service is used. We strongly recom- -mend that you utilize the Health Services if necessary. Services covered by -the fee are detailed on the Boston College Health Services website at -www.bc.edu/offices/uhs. - -Please note that Health Services fee is not the same as medical insurance. -You must still have adequate medical insurance coverage, especially for -such things as hospitalization, emergency room visits, diagnostic testing, -x-rays, etc., which are not covered by the Health Services fee. - -University Counseling Services - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sometimes students find adjusting to a new culture and new academic en- -vironment very challenging. The Counseling Services Office provides caring, -professional support and assistance to full-time students who request it. -Selected staff members in Counseling Services have extensive experience -with cross-cultural issues and are especially well-suited to help students -who are experiencing culture shock. It is very common in the United States -for students to seek the help of counseling professionals when dealing with -challenging situations. In fact, thousands of students take advantage of the -services offered by Counseling Services for a variety of issues each year. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales + +that goes together -- such as bananas alongside boxes of vanilla wafers -- in aisles or at entrances, +because it’s a consumer-friendly way to assist customers and pique their thought process while +generating incremental sales. By using such things as Y-Gondolas, mobile displays, LED lighting +fixtures, floor graphics, ceiling hangers or digital or video marketing to highlighting merchandise +in a non-intrusive, yet attention-getting fashion, it’s entirely possible to get customers to buy +items you wish to sell, not just what shoppers came into the store to purchase. + + An underutilized method of generating awareness and trial is product sampling and +demonstrations. Retailers such as Sam’s Club, Costco and Whole Foods have mastered the art +of sampling during high traffic days and peak times to expose customers to products they may +have otherwise ignored. Why not create a display that demonstrates the difference between +200 and 400 thread count sheets? Utilize sales staff as virtual models by having them wear the +merchandise being sold in the store, much like mega-brands Hollister and American Eagle have +done with their employees. Create mobile sampling stations offering customers tastes of lesser- +known foods such as couscous, Greek yogurt, even kale along with recipes incorporating those +foods to stimulate profitable impulse sales. + +In fact, the New England Journal of Medicine reports that food choices are largely guided by +our responses to outside signals and stimuli. A paper authored by Deborah Cohen of RAND Health +in Santa Monica, Calif., and Susan Babey of the UCLA Center for Health Policy Research, states that +impulse marketing takes place on a subconscious level, and that placement of foods in prominent +locations increases the rate at which they’re purchased. + + While most people may associate such impulse displays with checkout lines and sodas, candy, +gum and magazines, they can be just as effective in other areas of a store. Shopping for a new +blouse? You may be influenced by matching (high-margin) accessories attractively displayed +next to them. Buying new cat food? How about a pretty collar or catnip to go along with it. The +possibilities for proximity marketing are enormous, if not often overlooked, opportunities. + +Solution #1: Multiply the end cap effect + +End caps are perhaps the most misunderstood and underutilized spaces in the retail +environment. According to Dr. Sorensen, approximately 2-percent of the total items in a +typical store at any given time are being promoted on end-of-aisle displays or other secondary +promotional displays apart from the mail shelving fixtures. However, this 2-percent of items may +constitute a whopping 30-percent of all store sales. + + While retailers often use this space to promote sales or highly discounted merchandise, the fact +is that half the shoppers purchasing an item from such end caps are unaware the goods are being +sold at a reduced price, and half of the people who know would buy the product even if it wasn’t +on sale. + +So in other words, good retailers are simply giving away profits to as much as 75-percent of + +their shoppers because they are locked into the mindset that price considerations dominate +shopping decisions when research doesn’t support that point of view. Instead, great retailers +understand that a better use of such prime real estate is to satisfy the shopper’s inherent need +for value and convenience, and most customers are willing to pay full price to accomplish a fast +shopping experience. + + Another reason end caps can be extremely profitable is that because of their small size, they +generally feature only one or two products. As such, shoppers do not have an easy or immediate +way to compare the merchandise as they would in the category section inside the aisle. The +impulse nature of end caps and the lack of price sensitivity support the premise that retailers can +command a higher price for such merchandise. 16 -www.bc.edu/international - - The services offered by Counseling Services (at no additional cost) include -short-term individual counseling and psychotherapy as well as a limited -number of counseling groups. If you wish to speak with a counselor you -can make an appointment by calling 617-552-3310 or by visiting the Counsel- -ing Office in Gasson Hall 108. - -Also, please remember that OIP staff is always available and willing to help -if you have any questions, either academic or personal. Most of our staff -members have studied and/or lived outside the US and know firsthand the -challenges of adapting to a new country and culture. - -Students with Disabilities - -Boston College provides a range of services and accommodations for stu- -dents with physical or learning disabilities. If you require additional assis- -tance because of a disability, feel free to share that information with your -OIP exchange advisor. You may also contact the Boston College Coordina- -tor of Special Needs Services in Gasson Hall 108, telephone: 617-552-3310. - -E-mail - -Every student enrolled at -Boston College is eligible -to receive an e-mail ac- -count. You may obtain -your account after arriv- -ing at BC by presenting -your BC ID card at the -Campus Technology Re- -source Center (CTRC) -located on the second -floor of the O’Neill Li- -brary. The use of your BC -email account is required -once you have arrived on- -campus. If you have an- -other address that you -prefer to use, set up your BC account (by using AGORA) to forward all mes- -sages to the alternate account. OIP and other BC offices will send all infor- -mation during your time at Boston College to your BC email address. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -17 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - LIFE BEYOND ACADEMICS - -Religion - -As a Catholic and Jesuit institution, Boston College is sensitive to the spiri- -tual needs of all members of the BC community, whatever their respective -faiths or philosophical traditions. Catholic Masses are offered throughout -the week in several locations on campus and are open to all BC students, -faculty and staff. The Campus Ministry Office offers spiritual counseling -and guidance to students, as well as volunteer service opportunities in Bos- -ton and other parts of the world. Campus Ministry also provides informa- -tion and support for people of all religious and spiritual traditions by main- -taining a list of churches, temples and synagogues in the Boston College -area. Various student clubs, including the Asian Christian Fellowship, Jew- -ish Students’ Association, and Muslim Students’ Association, provide a -sense of community and support for students of various spiritual tradi- -tions. - -Activities/Opportunities - -There is a wide variety of clubs and -activities to meet any student’s in- -terests, including several intercul- -tural clubs focusing on individual -countries or regions. An Activities -Fair early in the Fall Semester will -give you an opportunity to learn -about many of these opportunities -and speak to some of the students -involved. For students on exchange -during the Spring semester, the Of- -fice of the Dean for Student Devel- -opment has a list of all BC clubs -and activities (www.bc.edu/a-z/ -clubs/). Don’t pass up this opportu- -nity to learn about American culture -outside the classroom. Students -who get involved in activities at Boston College consistently enjoy their ex- -perience here more than those who do not get involved. For this reason, -we strongly encourage you to join at least one club while you are here. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +Breaking up and staggering the traditional long aisles in many retail locations can literally -18 +double the number of possible end caps from the shoppers perspective, particularly in the +center store. Incorporating other display tools such as Y-Gondolas, curved shelving, pyramid and +recessed shelving, highlight lighting, sampling stations, lifestyle graphics and signage, can only +serve to increase the sales velocity of end caps. -www.bc.edu/international - - s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sports in American University Culture - -Many international students are surprised at the level of importance placed -on intercollegiate sports in the United States. American football, particu- -larly, draws thousands of screaming fans every Saturday during football -season, with students cheering for the home team. This is very much a part -of American college culture and, even if you are not a sports fan, it is some- -thing you should experience during your time as a student in the United -States. You may not know anything about the intricate rules of American -sports like football, hockey and basketball. However, the experience of sit- -ting in a crowded stadium, cheering, listening to the band, watching the -crowd, and even trying to follow the game, is a true American experience -and should not be missed. - -Recreation Complex - -The Flynn Recreation Complex provides Boston College students with a -broad array of sports and physical fitness opportunities, from aerobics to -weight lifting and swimming. As an exchange student you have the option -of joining the “RecPlex” for a semester or academic year (membership fees -apply). If you choose not to join the “RecPlex” you may still use the facilities -by presenting your BC ID card and paying a small fee per visit. +Madix understands how important end caps are to a retailer’s success. The company manufactures a number +of end cap systems that maximize space, improve visibility, and capitalize on sales. Madix’s new Y-Gondola +greatly expands the potential of end caps to positively influence shopping behavior. -19 - -www.bc.edu/international - - CONTACT INFORMATION - -Office of International Programs - -Location: -Mailing Address: +Decide what solutions make sense for you -Hovey House, 258 Hammond Street -Hovey House, 140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +Revisiting the relationship between the perimeter and center store requires that retailers -Phone: 617/552-3827 Fax: 617/552-0647 -Email: oipexchange@bc.edu -Web: www.bc.edu/international -Director: Bernd Widdig -Email: widdig@bc.edu -Assistant Director for Exchange Students: Esther Messing -Email: messines@bc.edu +rethink the cost of sales. Ignoring or shrinking the center store given its contribution to +overall profits doesn’t make sense when much more is possible. A better course of action is to +fundamentally rethink the consumer shopping process and adapt the entire store to better deliver +total customer solutions. After all, what good is the sizzle without a juicy steak to go along with it? -Office of International Students and Scholars +The display fixture experts at Madix, Inc. are available to provide free consultation to retailers -Location: -Mailing Address: +who wish to explore fresh, new ideas on how to bring the center store back to life to create a +holistic shopping experience that drives sales and profits. For more information, contact John +Clontz Director of Marketing and e-Business jclontz@madixinc.com 214-515-5400. -21 Campanella Way - Suite 249 -140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 - -Phone: 617/552-8005 Fax: 617/552-2190 -Web: www.bc.edu/offices/odsd/intercultural -Director: Adrienne Nussbaum -Email: adrienne.nussbaum@bc.edu - -Office of Residential Life +17 -Contact: -Web: +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -On-Campus Housing, Off-Campus Housing Advice -www.bc.edu/offices/reslife + 10 Ways to Lift Center Store Sales -Office of Student Services +Traditional Center Store Fixture Configuration -Contact: -Web: + Traditional center store fixture runs are usually long and extend 30 feet or +more. The long runs discourage shoppers from entering the department. Also, +the lengthy runs have a limited quantity of end caps, the store fixture with the +most effective sales potential. -Billing, Medical Insurance, ID, Course Registration -www.bc.edu/offices/stserv +18 -University Health Services +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com 10 Ways to Lift Center Store Sales -Contact: -Web: +Center Store Layout that Staggers Runs + A center store layout with staggered runs includes more and shorter fixture +runs. Shoppers are more likely to enter a center store that utilizes shorter, +less daunting runs. The staggering of the runs greatly increases a shopper’s +exposure to end caps as he or she walks through the department. -Immunization, Health and Wellness Issues -www.bc.edu/offices/uhs +19 -University Counseling Services +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. 10 Ways to Lift Center Store Sales -Contact: -Web: +Center Store Layout with Specialty Merchandisers and Transitional Components -Personal Counseling -www.bc.edu/offices/counseling +Integrating specialty merchandisers into a center store will improve +incremental sales. Utilize transitional components, including bases and +shelves, to draw attention to merchandise within merchandising categories. +The transitional components and shelves also make room for shoppers to +navigate around the specialty displays between the aisles. 20 -www.bc.edu/international - - ESTIMATED EXPENSES -2010-2011 - - Average Cost of Apartment/Room Rental - -Studio Apartment: $1000 - $1300 per month * -One Bedroom Apartment: $1200 - $1550 per month * -Two Bedroom Apartment: $1400 - $2000 per month * -Three-Four Bedroom Apartment: $2000 - $3800 per month * -Private Room in Local Home: $500 - $900 per month * - -*Please note: Boston rental agents often charge one half or one full month's -rent as a commission fee (also known as a "Finder's Fee") - -On-Campus Housing - -Shared Room: $1100 per person per month - -Meals and Books - -Campus Meal Plan: $2300 per semester for full meal plan -No Meal Plan: $1500 per semester -Books: $600 per semester +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales -Other Expenses +Center Store Layout using Y-Gondolas as End Caps -Medical Insurance: $871 (Fall semester); $1027 (Spring semester) ** -University Health Services: $210 per semester ** -Flynn Recreation Complex: $140 per semester + Madix’s Y-Gondola exponentially expands the sales potential of end caps. +Not only does the fixture greatly increase the end cap’s merchandise volume, +but it also makes the merchandise on the end cap visible from wider angles. A +store layout that utilizes Y-Gondolas on short fixture runs will create an inviting +and more shopable center store. -**The BC Medical Insurance plan is mandatory unless you can demonstrate -coverage by an insurer with a U.S. office. The University Health Services fee is -required for on-campus students. +21 -Personal Expenses: These will vary significantly according to the individ- -ual, but students should budget approximately $2000 per semester. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. 10 Ways to Lift Center Store Sales -Transportation from Logan Airport to BC +22 -Taxi: approximately $60 -Private Airport Shuttle: approximately $50 -Subway (“T”) Ride : $2.00 cash payment or $1.70 using a “Charlie Card” +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com www.madixinc.com +Texas 500 Airport Road, Terrell, TX 75160 | 800.776.2349 +Alabama 20984 Alabama Hwy 9, Goodwater, AL 35072 | 800.633.6282 +©2014 Madix, Inc. All rights reserved. Not responsible for errors and omissions in printing. Rev. 7/14 diff --git a/q119/random_k2/random_2.md b/q119/random_k2/random_2.md index 4b9bd14ed63d1882c240fd2561e1be123d06b2a6..310f0f860924656c9ab9a27398fe83594f525660 100644 --- a/q119/random_k2/random_2.md +++ b/q119/random_k2/random_2.md @@ -1,1483 +1,1527 @@ -USER MANUAL +Department of the Treasury +Departmental Offices -8-IN-1 ONE-TOUCH -AUTOMATIC ESPRESSO MAKER +FY 2025 -GCM4230 +Capital Investment Plan 1 -USER MANUAL ESPRESSO MAKER GCM4230 - CONTENTS + Standard IT Investments ...................................................................................................... 3 -Important Safeguards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 +Treasury Shared Services Security Operations Center (TSSSOC) .............................................. 3 -Know Your Espresso Maker . . . . . . . . . . . . . . . . . . . . . . . . . . 6 -Control Panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 +Major IT Investments .......................................................................................................... 4 -Using Your Espresso Maker +Cybersecurity Enhancement Account (CEA) - Cloud Enterprise ................................................ 4 -Before First Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 +Cybersecurity Enhancement Account (CEA) - Multi-Factor Authentication .............................. 6 -Setting Up and Priming the Espresso Maker . . . . . . . . . . . . . . . 8 +Cybersecurity Enhancement Account (CEA) - Other Cyber Priorities ....................................... 7 -Filling the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 +Cybersecurity Enhancement Account (CEA) - Security Logging ............................................... 9 -Placing the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 +Cybersecurity Enhancement Account (CEA) - Universal Encryption ...................................... 11 -Mug Size Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +Cybersecurity Enhancement Account (CEA) - Zero Trust Architecture ................................... 12 -Drink Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +HR LoB - HRConnect ................................................................................................................ 14 -Making an Espresso . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 +Treasury Enterprise Identity, Credential and Access Management (TEICAM) ........................ 16 -Making a Cappuccino . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 +Major Non-IT Investments ................................................................................................ 19 -Making a Latte . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 +Main Treasury Building and Freedman's Bank Building .......................................................... 19 -Making an Americano . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 +2 -Brewing in Manual Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 15 + Note to Reviewers: Consistent with the corresponding Summary of Capital Investments table, the +columns included in the investment tables below are defined as: -Frothing Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 +FY 2023 Actuals -Total actual obligations. +FY 2024 Estimated Obligations- Anticipated obligations from all budgetary resources (e.g., balances +from prior years, user fees, and FY 2023 Operating levels). +FY 2025 Estimated Obligations - Anticipated obligations from all budgetary resources (e.g., balances +from prior years, user fees, and FY 2025 President’s budget). -Cleaning and Maintenance +Standard IT Investments -Cleaning the Portafilter and Brew Head . . . . . . . . . . . . . . . . . 17 +Treasury Shared Services Security Operations Center (TSSSOC) -Cleaning the Water Reservoir . . . . . . . . . . . . . . . . . . . . . . . 17 +Description: +TSSSOC functions as the Department wide security operations center providing security monitoring, +incident coordination, incident response, and reporting. TSSSOC operates 24x7x365 and actively +coordinates with other federal agencies, DoD, and the IC. -Cleaning the Milk Reservoir and Frothing Tube . . . . . . . . . . . . . 17 +Investment Obligations: (In Millions of $): -Descaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 +T y p e -Warranty and Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -Model: GCM4230 | © 2023 Gourmia | 2.0 -The Steelstone Group Brooklyn, NY | Gourmia .com +F Y 2 0 2 3 +A c t u a l s -All rights reserved. No portion of this manual may be reproduced by any means whatsoever without written -permission from The Steelstone Group, except for the inclusion of quotations in a media review. Although -every precaution has been taken in the preparation of this manual, The Steelstone Group assumes no -responsibility for errors or omissions. Nor is any liability assumed for damages resulting from the use of -the information contained herein. Although every effort has been made to ensure that the information -contained in this manual is complete and accurate, The Steelstone Group is not engaged in rendering -professional advice or services to the individual consumer; further, The Steelstone Group shall not be held -responsible for any loss or damage allegedly arising from any information or suggestions in this manual. +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s - Congratulations on the -purchase of your Espresso -Maker from Gourmia! +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Here at Gourmia, our one and only goal is to make sure -nutritious and delicious meals are always a practical -possibility no matter how fast-paced your life is . Whether -you’re in need of a hearty breakfast, a mid-day snack, or a -satisfying supper, our dependable, easy-to-use appliances -are designed to work for you while you work your way -through your busy day . +C h a n g e i n +$ -This manual provides an overview and instructions on -all the espresso maker’s functions and features, along -with directions for assembling, operating, cleaning, -and maintaining your appliance . Please read all safety -instructions before getting started . Keep this manual handy -for future use and review . +% C h a n g e -Our team is available to help with any and every question -you might have about your new espresso maker . We’re -excited for you and want you to get the most out of your -purchase—so whether you need assistance with setup or -guidance on which features to enable for which drinks, -we’re happy to walk you through it, step by step . +2 . 2 6 -Happy and easy cooking! +1 . 0 9 -We’d love to hear about the delicious drinks you’re creating! -Share your thoughts, recipes, and photos with us. +1 . 0 9 -@gourmia +0 -3 +0% + +1 5 . 4 4 + +1 7 . 7 0 + +1 4 . 75 - S -d -r -a -u -G -E -f -a -S +1 5 . 8 4 -t -n -a -t -r -o -p -M +1 5 . 1 9 -I +1 6 . 2 8 -S A V E T H E S E I N S T R U C T I O N S +0 . 4 4 -IMPORTANT SAFEGUARDS +0 . 4 4 -Before operating this espresso maker, read and follow all these -basic safety precautions: +2.98% -1. The appliance can be used +2.78% -7. do not operate the appliance +Purpose, Accomplishments, Future Objectives: + +The Treasury Shared Services Security Operations Center (TSSSOC) provides leadership and oversight +for the offices and bureaus across Treasury in all aspects of cybersecurity operations. The program is +responsible for ensuring implementation of the Federal Information Security Modernization Act (FISMA) +of 2014 as well as other federal laws and guidance related to cybersecurity operations. The program also +provides central coordination and reporting on cybersecurity operational metrics and programs to external +agencies and sets Treasury cybersecurity operations standards and guidance. TSSSOC also provides +departmental situational awareness of cybersecurity incidents and coordinates response to intrusion +activity. The program also serves as a lead in implementing the Treasury Cybersecurity Strategy. + +Investments have contributed to Treasury's success in defending against cyber adversaries and provides +leadership and direction in mitigating threats to Treasury systems and services. Future objectives include: +collecting and storing data from all bureaus to meet M-21-31 objectives, expanding TSSSOC to offer +additional security operations and incident response services to bureaus, and improving communication +and collaboration throughout the Department through automation. + +3 -by children aged 8 years and -above and persons with reduced -physical, sensory or mental -capabilities or lack of experience -and knowledge if they have been -given supervision or instruction -concerning use of the appliance -in a safe way and understand the -hazards involved . Children should -not play with the appliance . + Major IT Investments -Cleaning and user maintenance -shall not be done by children -unless they are 8 or older and -supervised . +Cybersecurity Enhancement Account (CEA) - Cloud Enterprise -2. Keep the appliance and its cord +Description: -out of reach of children less than 8 -years old . +Treasury aims to improve, entice, and further accelerate enterprise-wide cloud adoption through +investment in the Treasury enterprise shared services cloud environments. The Department’s focus on +cloud security enhancement and upgraded capabilities to meet security and compliance risks will +involve investment in security controls, monitoring, and increased threat protections by providing a +common cloud operating platform for Departmental workloads, inclusive of all security elements +required by Executive Order 14028 and subsidiary guidance. Treasury’s recently awarded Blanket +Purchase Agreement (BPA) for cloud services (Treasury Cloud [TCloud]) will support our design, +development, and implementation of security patterns and guardrails with expanded Security +Operations Center (SOC) capabilities to accommodate increased telemetry from cloud assets and +workloads. -3. Make sure your outlet voltage +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -corresponds to the voltage stated -on the rating label of the espresso -maker . +Investment Obligations: (In Millions of $): -4. To protect against fire, electric +T y p e -shock, or injury to persons do not -immerse cord, plug, or appliance -in water or other liquid . +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -5. Close supervision is necessary +F Y 2 0 2 3 +A c t u a l s -when this appliance is used by or -near children . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -6. Unplug from outlet when not in +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -use and before cleaning . Allow the -appliance to cool before putting -on or taking off parts, and before -cleaning the appliance . +C h a n g e i n +$ -with a damaged cord or plug or if -the appliance malfunctions or has -been damaged in any manner . -Return appliance to an authorized -service facility for examination, -repair, or adjustment . +% C h a n g e -8. The use of accessories not +2 0 . 0 7 -recommended by the appliance -manufacturer may result in fire, -electric shock, or injury to persons . +2 5 . 3 3 -9. do not touch hot surfaces; use +1 5 . 8 0 -handles or knobs . +- 9 . 5 3 -10. do not operate this appliance -outdoors . do not use for -commercial purposes . +- 3 7 . 6 1 % -11. do not let the cord hang over +0 . 6 5 -the edge of a table or counter or -touch any hot surfaces . +1 . 6 0 -12. Use on a hard, flat, level, heat- +9 . 6 3 -resistant surface only . +8 . 0 3 -13. Always be sure to secure the +5 0 1 . 5 6 % -water reservoir tightly back inside -the unit with the lid closed and -filled with water between the -minimum and maximum mark -before turning the appliance on . +2 0 . 7 2 -14. do not remove the portafilter while -brewing coffee . Exercise caution -when removing the portafilter after -brewing since the metal parts -will be hot . Make sure to hold the -portafilter by the handle and use - the filter retention clip to dispose -the grounds . +2 6 . 9 3 + +2 5 . 4 3 + +- 1 . 5 0 + +- 5 . 5 7 % + +Purpose, Accomplishments, Future Objectives: + +The CEA cloud investment focuses on an enterprise-wide investment approach to build stronger +protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the CEA Cloud Enterprise investment, Treasury aims to improve, entice, and further +accelerate enterprise-wide cloud adoption through investment in the Treasury enterprise shared +services cloud environments. The Department’s focus on cloud security enhancement and upgraded +capabilities to meet security and compliance risks involves investment in security controls, monitoring, +and increased threat protections by providing a common cloud operating platform for Departmental +workloads, inclusive of all security elements required by Executive Order 14028 and subsidiary +guidance. 4 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - 15. Extreme caution must be -exercised when moving -appliances with hot liquids . + The FY 2025 President's Budget includes $25.4 million for Treasury’s CEA Cloud Enterprise +investment with alignment to the NIST Cybersecurity Framework priorities: + +• -16. Only remove the water reservoir -when the appliance is turned off . +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an -17. This appliance is not intended -to be operated by means of an -external timer or separate remote- -control system . +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -appliance and allow it to cool -down . The blockage may be -caused by the coffee being too -finely ground . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -20. do not place the appliance on or -near a hot gas or electric burner, -or in a heated oven . +appropriate safeguards to ensure delivery of critical services. -21. do not use the appliance for other +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -than its intended use . +occurrence of a cybersecurity event. -18. do not remove the portafilter +• Respond to Detected Cybersecurity Incidents Goal: Develop and implement appropriate -at any time during the brewing -process . +activities to take action regarding a detected cybersecurity incident. -19. If coffee does not drip, it means -the filter is blocked . Turn off the +In FY2025, CEA – Cloud Enterprise investments are targeted to the Department’s cloud focused +modernization and the intrinsic enhancement of our cybersecurity posture. Our investments will build +on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. -22. do not leave the appliance -unattended when turned on . +• Treasury seeks to support the following initiatives with final FY2025 appropriations: +• Bureau of the Fiscal Service cloud cybersecurity modernization and the necessary human -23. This appliance should only be -used in households or similar -environments . +capital to support initiatives. -S -d -r -a -u -G -E -f -a -S +• Cloud security resources (engineers, ISSM, etc.) to support the FY2023 awarded enterprise -t -n -a -t -r -o -p -M +TCloud contract and program. -I +• Application migrations from on-premises to cloud environments and/or from unique Bureau -Cord and plug Safety -1. A short power-supply cord is provided to reduce risk of becoming entangled +environments to the enterprise cloud platform. -in or tripping over a longer cord . +• -2. Extension cords may be used if care is exercised in their use . +Incident response infrastructure to support coordination and communication in an -3. +uncompromised environment for response and recovery activities. -If an extension cord is used: +• Migration of National Security System infrastructure to the cloud. +• Operations, maintenance, and ongoing investment in a sustainable low code cloud platforms +(i.e., ServiceNow, Salesforce) to support organizational capabilities, to include +Cybersecurity Governance, Risk, and Compliance; CEA Administration and Governance; +and Security. -• The marked electrical rating of the extension cord should be at least equal +• Storage support for enterprise logging collection and utilization of historic logs to support -to the electrical rating of the appliance . +threat hunt and threat detection activities, as well as additional behavioral analytics. +• Expansion of the Department’s enterprise offering for public-facing, cloud-based web hosting -• The longer cord should be arranged so that it will not drape over the +that includes an Akamai CDN with robust DoS and bot protections, a Drupal content +management system, and additional centralized site shield, web performance and +readiness/response investments that integrate with Treasury’s security operations +capabilities for threat detection and response. -countertop or tabletop where it can be pulled on by children or tripped -over . +Looking forward, CEA investments will aim to accomplish the below future objectives: -Electrical polarized plug -This unit has a polarized plug (one blade is wider than the other) . -To reduce the risk of electric shock, the plug is intended to be -used by inserting it into the electrical outlet only one way . If the -plug does not fit securely into the outlet, reverse the plug and -re-insert . If it still does not fit, contact a qualified electrician for -assistance . Do not modify the plug in any way . +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -poWEr +threats. -AC120V~ 60Hz 1,300W +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. 5 -USER MANUAL ESPRESSO MAKER GCM4230 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K - -KNOW YOUR -ESPRESSO MAKER - -Water Reservoir -Cover - -Removable -Water Reservoir - -Control Panel - -Brew Head - -Power Switch - -On the back-right -side of the unit - -Removable Drip Tray + Cybersecurity Enhancement Account (CEA) - Multi-Factor Authentication + +Description: + +Treasury’s investment in Multi-Factor Authentication (MFA) supports our development and +implementation of compliant technologies for internal systems and applications. This investment is a +critical component of the Department’s solutions to prevent unauthorized access by adversaries, +including nation state actors. MFA is a layered approach to security access to network or applications +by certifying identity with a combination of two or more approved pieces of evidence (or factors) to an +authentication mechanism. The Department’s investment in, and sustainment of, MFA capabilities +include strengthening MFA governance, ongoing investment in enterprise solutions to boost adoption, +and strategic partnerships to streamline MFA related procurement effort for all Treasury Bureaus and +Offices. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. + +Investment Obligations: (In Millions of $): + +T y p e + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) + +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +0 . 0 0 + +0 . 0 0 + +0 . 0 0 + +5 . 1 0 + +0 . 6 0 + +5 . 7 0 + +0 . 0 0 + +- 5 . 1 0 + +- 1 0 0 . 0 0 % + +2 . 5 0 + +2 . 5 0 + +1 . 9 0 + +3 1 6 . 6 7 % + +- 3 . 2 0 + +- 5 6 . 1 4 % + +Purpose, Accomplishments, Future Objectives: + +The CEA multi-factor authentication (MFA) investment focuses on an enterprise-wide investment +approach to build stronger protections and further benefit the bureaus in their cybersecurity goals. With +the creation of a centralized appropriation, administration, and governance, Treasury has been able to +develop and enhance Department-wide cloud services, support Bureau-level modernization goals, and +drive a more robust coordination of efforts to improve the Department’s ability to detect, identify, +protect, recover, and respond to cybersecurity events. + +Through the CEA Multi-Factor Authentication investment, Treasury aims to develop and implement +compliant technologies for internal systems and applications. This investment is a critical component +of the Department’s solutions to prevent unauthorized access by adversaries, including nation state +actors. MFA is a layered approach to security access to network or applications by certifying identity +with a combination of two or more approved pieces of evidence (or factors) to an authentication +mechanism. The Department’s investment in, and sustainment of, MFA capabilities include +strengthening MFA governance, ongoing investment in enterprise solutions to boost adoption of MFA, +and strategic partnerships with industry partners to streamline MFA related procurement effort for all +Treasury Bureaus and Offices. + +The FY 2025 President's Budget includes $2.5 million for Treasury’s CEA Multi-Factor +Authentication investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. 6 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 + In FY2025, CEA – Multi-Factor Authentication investment is targeted to the Department’s MFA- +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. - Milk Reservoir Lid +Treasury is looking to support the following initiatives with final FY2025 appropriations: +Conversion to Active Directory (AD) Federation that provides for the authorization, authentication, +and SSO functionality to applications and services virtually anywhere, including perimeter networks, +partner organizations, and the cloud. -Froth Strength Control +• Looking forward, CEA investments will aim to accomplish the below future objectives: +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -Froth Dispensing Tube Lever +threats. -Froth Dispensing Tube +• Provide leadership with greater visibility into cybersecurity efforts and encourage information -Minimum Strength +sharing. -Clean / Americano Position +Improve identification of cyber threats and better protect information systems from attack. -Maximum Strength +• +• Provide platform to enhance communication, collaboration, and transparency. -Removable Milk Reservoir +Cybersecurity Enhancement Account (CEA) - Other Cyber Priorities -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K +Description: -Measuring Scoop -and Tamper +Treasury maintains investment in Other Cybersecurity Priorities that address and respond to the +changing threat landscape of the Department, given our interconnected technology environment, which +has amplified the need for identifying and assessing the security posture of high value assets, national +security systems infrastructure, Treasury’s supply chain, cybersecurity governance, risk, and +compliance (GRC), amongst other necessary priorities. Additionally, Treasury is investing in a resilient +future through the acquisition of federal labor cybersecurity subject matter expertise to support our +complex technology portfolio, as well as the necessary resources to support the overall administration +and governance of Cybersecurity Enhancement Account (CEA) appropriations and the ongoing +performance of cybersecurity investments. -Portafilter Handle +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -Portafilter Retention Clip +Investment Obligations: (In Millions of $): -Portafilter +T y p e -Double-Shot -Filter +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -Single-Shot +F Y 2 0 2 3 +A c t u a l s -Control panel -All icons will light up when the milk reservoir is correctly inserted into the unit . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -ESPRESSO -Icon +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -LATTE -Icon +C h a n g e i n +$ -MANUAL -ESPRESSO Icon +% C h a n g e -CAPPUCCINO -Icon +1 7 . 7 9 -AMERICANO -Icon +2 . 7 3 -FROTH -Icon +6 . 0 5 -CLEAN -Icon +3 . 3 2 -7 +1 2 1 . 5 4 % -USER MANUAL ESPRESSO MAKER GCM4230 +8 . 7 5 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +2 6 . 5 5 -I +1 2 . 9 0 -USING YOUR -ESPRESSO MAKER -Before first use -1. remove and discard all packaging material and stickers that are not +1 5 . 6 3 -permanent . +1 8 . 9 0 + +2 4 . 9 5 + +6 . 0 1 + +4 6 . 5 8 % + +9 . 3 3 + +5 9 . 6 8 % + +7 -2. remove the water reservoir, milk reservoir, portafilter, filters, measuring -scoop, and removable drip tray. Wash them with non-abrasive soap . -Thoroughly dry the parts . Wipe the exterior with a damp cloth . + Purpose, Accomplishments, Future Objectives: -3. Before using the espresso maker for the first time, it is required to prime the +The CEA investment for other cybersecurity priorities focuses on an enterprise-wide investment +approach to build stronger protections and further benefit the bureaus in their cybersecurity goals. With +the creation of a centralized appropriation, administration, and governance, Treasury has been able to +develop and enhance Department-wide cloud services, support Bureau-level modernization goals, and +drive a more robust coordination of efforts to improve the Department’s ability to detect, identify, +protect, recover, and respond to cybersecurity events. -appliance . See Setting up and priming the Espresso Maker below . +Through the CEA Other Cybersecurity Priorities investment, Treasury maintains investment in +initiatives that address and respond to the changing threat landscape of the Department, given our +interconnected technology environment, which has amplified the need for identifying and assessing the +security posture of high value assets, national security systems infrastructure, Treasury’s supply chain, +cybersecurity governance, risk, and compliance (GRC), amongst other necessary priorities. +Additionally, Treasury is investing in a resilient future through the acquisition of federal labor +cybersecurity subject matter expertise to support our complex technology portfolio, as well as the +necessary resources to support the overall administration and governance of CEA appropriations and +the ongoing performance of cybersecurity investments. -Setting up and priming the -Espresso Maker -1. Place the espresso maker on -a hard, flat, and heat-resistant -surface . +The FY 2025 President's Budget includes $25.0 million for Treasury’s CEA Other Cybersecurity +Priorities investment with alignment to the NIST Cybersecurity Framework priorities: -7. Set the froth strength control to +• -the clean position . +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an -8. Turn the froth dispensing +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -2. Detach the water reservoir and -open the water reservoir cover . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -tube lever to rotate the froth -dispensing tube upward . +appropriate safeguards to ensure delivery of critical services. -9. Place a heatproof cup or mug +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -under the froth dispensing tube . +occurrence of a cybersecurity event. -10. Tap the CLEAN icon . The clean +• Respond to Detected Cybersecurity Incidents Goal: Develop and implement appropriate -icon will blink until the espresso -maker is done preheating . After -preheating, water will dispense -into the cup . When cleaning is -complete, the appliance will beep -three times . +activities to take action regarding a detected cybersecurity incident. -11. Discard the water dispensed into +In FY2025, CEA – Other Cybersecurity Priorities investment is targeted to the Department’s cyber +priorities focused om modernization and the intrinsic enhancement of our cybersecurity posture. Our +investments will build on prior year accomplishments, continue investment in enterprise shared +services, enhance Department-wide coordination of cyber efforts, improve responsiveness to threats, +deliver greater visibility and information shared, and improve threat detection, identification, and +protection capabilities. -the cup . +Treasury is looking to support the following initiatives with final FY2025 appropriations: -3. Fill the water reservoir with water -between the “Min” and “Max” -mark and reattach it to the back -of the appliance . . +• Establish a repeatable programmatic structure to review, assess, and govern emerging -4. Plug the power cord into an +technologies (i.e., quantum, Artificial Intelligence [AI], etc.) to ensure applications of all +security standards develop consistent policies, workforce planning, reporting, Bureau +adherence, ethical use, etc. -outlet . +• Acquisition of an enterprise EDR solution +• Workforce focused cyber awareness and other supplemental trainings +• Enhanced security compliance of a low code platform through Enhanced Adoption and -5. Turn the espresso maker on by -flicking the power switch on the -right side of the appliance . The -ESprESSo icon and ManuaL -ESprESSo icon will light up on -the control panel . +Automation of Change Management/Enablement, as well as increased use of common +controls. -6. +• Evolution of the Enterprise Cyber Risk Management (ECRM) initiative which will continue to -Insert the empty milk reservoir -into the appliance until it clicks -into place . All icons will light up on -the control panel . +strengthen Treasury's approach to managing risks across the enterprise and instantiate a +continuous review of all of its critical networks, systems, and data. 8 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - filling the portafilter -1. Select your filter size . Use the - -single-shot filter for one shot of -espresso . Use the double-shot -filter for larger sized drinks, or -for a double shot of espresso . -Align the gap in the filter with the -groove in the portafilter . Drop the -filter into the portafilter and rotate -the filter to ensure it’s locked in -place . - -2. Use the measuring spoon to fill - -the filter with espresso grounds . -Do not exceed the MAX line in -the filter . - -3. Use the tamper end of the - -measuring spoon to press on -the grounds . The surface of the -espresso grounds should be -smooth after pressing . - -4. Clean any excess grounds from - -the rim of the portafilter . - -placing the portafilter -1. Place the portafilter underneath -the brew head, with the handle -facing toward the left . - -2. Keep one hand atop the - -appliance to keep it steady while -fastening the portafilter into the -brew head . The handle must -align with the open lock icon on -the brew head . - -3. Turn the handle to the right to - -lock the portafilter into the brew -head . The handle must align with -the closed lock icon on the brew -head . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + • Development of a low code Configuration Management Database (CMDB) solution, +encompassing both Asset Management and Configuration Management. +• Development of a Risk Management Framework for enterprise infrastructure +• -9 +Investment in Organizational Resiliency and the associated human capital component to + +address current and emerging cyber priorities such as Supply Chain Risk Management +(SCRM) and incident response support. + +• Performance of enterprise-level FISMA readiness assessments. +• Facilitation of an Enterprise Cyber BPA ordering capability (PROTECTS) +• + +Investment in an Annual Threat Hunt that identifies a random sampling of Treasury systems to + +identify anomalies beyond traditional detection methodologies. + +Acquisition of Incident Response support, allowing for defined response timeline in the +event of a compromise and need for SME and resource depth. + +Looking forward, CEA investments will aim to accomplish the below future objectives: + +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +Cybersecurity Enhancement Account (CEA) - Security Logging + +Description: -USER MANUAL ESPRESSO MAKER GCM4230 +Treasury’s investments in Security Logging ensure compliance with Office of Management and +Budget (OMB) Memorandum 21-31 (M-21-31) to develop and implement security logging capabilities +to receive, store, analyze, and process security event and system log data. Investment funding targets +the creation and monitoring of data-stream disruptions, a shared log facility, storage and retention of +log data, development of threat hunt and incident response playbooks, and the creation of user +behavioral analytics capabilities to support detection of malicious behavior. The investment objective +is for all logs to be accessible and visible for the highest level (enterprise) security operations center at +the Department, the Treasury Shared Services Security Operation Center (TSSSOC). - Mug Size Guide +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -ESPRESSO +Investment Obligations: (In Millions of $): -CAPPUCCINO +T y p e -LATTE +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -AMERICANO +F Y 2 0 2 3 +A c t u a l s -Drink +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -Shot +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Size +C h a n g e i n +$ -SINGLE +% C h a n g e -DOUBLE +2 5 . 6 6 -SINGLE +2 . 4 8 -DOUBLE +2 9 . 4 1 -SINGLE +2 6 . 9 3 1 , 0 8 8 . 2 4 % -DOUBLE +0 . 0 0 -SINGLE +2 5 . 6 6 -DOUBLE +4 . 0 0 -2 OZ +6 . 4 8 -3 OZ +1 1 . 9 2 -5 OZ +7 . 9 2 -10 OZ +1 9 7 . 9 0 % -12 OZ +4 1 . 3 3 -16 OZ +3 4 . 8 5 -5 OZ +5 3 8 . 2 2 % -10 OZ +9 + + Purpose, Accomplishments, Future Objectives: + +The CEA security logging investment focuses on an enterprise-wide investment approach to build +stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the CEA Security Logging investment, Treasury ensures compliance with Office of +Management and Budget (OMB) Memorandum 21-31 (M-21-31) to develop and implement security +logging capabilities to receive, store, analyze, and process security event and system log data. +Investment funding targets the creation and monitoring of data-stream disruptions, a shared log facility, +storage and retention of log data, development of threat hunt and incident response playbooks, and the +creation of user behavioral analytics capabilities to support detection of malicious behavior. The +investment objective is for all logs to be accessible and visible for the highest level (enterprise) security +operations center at the Department, the Treasury Shared Services Security Operation Center +(TSSSOC). + +The FY 2025 President's Budget includes $41.3 million for Treasury’s CEA Security Logging +investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -note: To accommodate larger mugs, remove the drip tray . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +appropriate safeguards to ensure delivery of critical services. -I +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -drink types +occurrence of a cybersecurity event. -Espresso -One shot of -espresso . +In FY2025, CEA – Security Logging investment is targeted to the Department’s security logging +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. -Cappuccino -Made up of -1/3 espresso, -1/3 steamed milk, -and 1/3 milk foam . +Treasury is looking to support the following initiatives with final FY2025 appropriations: -Latte -Made up of -1/3 espresso and -2/3 steamed milk -topped with milk foam . +• Security Logging efforts to protect against Distributed Denial-of-Services (DDoS) and other +cyber-attacks including 1) increase of storage for Treasury Trusted Internet Connection +(TIC) packet capture data, allowing rapid forensic analysis of suspected attacks; 2) +expansion of Web Application Firewall (WAF) capabilities for applications and services +across the Department; and 3) implementation of indexers at the Network Colocations +which will provide critical cyber visibility and analytics. -Americano -Made up of -1/3 espresso and -2/3 hot water . +• -Espresso +Investment in Organizational Resiliency and the associated human capital component necessary -Steamed Milk +to strengthen threat analysis, provide Security Operations Center (SOC) support for +sustained cyber attacks, and establish, operate, and maintain the Treasury logging platform. -Hot Water +• Expansion of the SOC to provide enterprise SOC capabilities for Treasury bureaus, including -Milk Foam +Operations analysts, intake support, Threat & Research analysts, enhanced capabilities, and +associated equipment. + +• + +Investment in cloud-based log processing, storage, and platform licenses. 10 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Making an Espresso - -Ensure the water reservoir is filled with water and placed to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a cup on the drip tray . - -4. For a single shot, tap the - -ESPRESSO icon once . For a double -shot, tap the ESPRESSO icon twice . -Your selected drink type will blink -until the espresso maker is done -preheating . After preheating, the -brewing process will begin . - -5. When brewing is complete, the -appliance will beep three times -and your espresso will be ready -for sipping . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + Looking forward, CEA investments will aim to accomplish the below future objectives: -11 +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +Cybersecurity Enhancement Account (CEA) - Universal Encryption + +Description: + +The Department’s investment in Universal Encryption supports our commitment to fully achieve +encryption protocols outlined in Executive Order 14028 and subsidiary guidance (i.e., Office of +Management and Budget [OMB], Cybersecurity Infrastructure and Security Agency [CISA], and +others). Our investment focuses on full adoption of encryption of data at rest (DAR) and data in transit +(DIT) for all Treasury systems and will address the unique challenges of our environment, proposed +solutions, and the implementations necessary to address each system for compliance with encryption +standards and requirements. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. + +Investment Obligations: (In Millions of $): + +T y p e + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) + +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +0 . 0 0 + +0 . 0 0 + +0 . 0 0 + +2 . 5 0 + +0 . 4 8 + +2 . 9 8 + +1 . 0 0 + +- 1 . 5 0 + +- 6 0 . 0 0 % + +0 . 0 0 + +1 . 0 0 + +- 0 . 4 8 + +- 1 0 0 . 0 0 % + +- 1 . 9 8 + +- 6 6 . 3 9 % + +Purpose, Accomplishments, Future Objectives: -USER MANUAL ESPRESSO MAKER GCM4230 +The CEA universal encryption investment focuses on an enterprise-wide investment approach to build +stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +Through the CEA Universal Encryption investment, the Department supports initiatives that work +toward full achievement of encryption protocols outlined in Executive Order 14028 and subsidiary +guidance (i.e., Office of Management and Budget [OMB], Cybersecurity Infrastructure and Security +Agency [CISA], and others). Our investment focuses on full adoption of encryption of data at rest +(DAR) and data in transit (DIT) for all Treasury systems and will address the unique challenges of our -I +11 + + environment, proposed solutions, and the implementations necessary to address each system for +compliance with encryption standards and requirements. -Making a Cappuccino +The FY 2025 President's Budget includes $1 million for Treasury’s CEA Universal Encryption +investment with alignment to the NIST Cybersecurity Framework priorities: -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -1. Fill the milk reservoir with milk +appropriate safeguards to ensure delivery of critical services. -7. For a single shot, tap the +In FY 2025, CEA – Universal Encryption investment is targeted to the Department’s universal +encryption focused modernization and the intrinsic enhancement of our cybersecurity posture. Our +investments will build on prior year accomplishments, continue investment in enterprise shared +services, enhance Department-wide coordination of cyber efforts, improve responsiveness to threats, +deliver greater visibility and information shared, and improve threat detection, identification, and +protection capabilities. -below the “Max” mark . +Treasury is looking to support the following initiatives with final FY 2025 appropriations: -2. +• -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -cappuccino function will not be -available as an option . +Investment in Quantum-Resistant Cryptography that could protect the Department against -3. Set the froth strength control to +cryptanalytic attacks by a quantum computer. -the maximum strength . -note: The above setting is a -recommendation only . +Looking forward, CEA investments will aim to accomplish the below future objectives: -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -5. To attach the portafilter to the +threats. -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . +• Provide leadership with greater visibility into cybersecurity efforts and encourage information -6. Position the froth dispensing tube +sharing. -over a large heatproof cup . +Improve identification of cyber threats and better protect information systems from attack. -CAPUCCINO icon once . For a -double shot, tap the CAPPUCCINO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . +• +• Provide platform to enhance communication, collaboration, and transparency. -8. When brewing is complete, the -appliance will beep three times -and your cappuccino will be -ready for sipping . +Cybersecurity Enhancement Account (CEA) - Zero Trust Architecture -9. Discard the spent grounds by +Description: -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +Zero Trust Architecture (ZTA) investment priorities at Treasury seek to minimize implicit trust and +reinvigorate least privilege through a continual verification of each user, device, application, and +transition. The investment approach requires the Department to enhance visibility and threat detection +at the application level to improve its ability to support continuous threat analysis, detection, and +response, and enable to analysis of encrypted traffic. Compartmentalization, micro segmentation, and +reinforcing enforcement of continuous identity verification and conditional access policies will +improve our resistance to fraudulent tampering of privileged accounts. Our objective is a universal, +default security posture of “never trust, always verify” across our entire technology stack. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. 12 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making a Latte - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. Fill the milk reservoir with milk - -below the “Max” mark . - -6. Position the froth dispensing tube -over a large heatproof cup . - -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -latte function will not be available -as an option . - -3. Set the froth strength control to - -the minimum strength . -note: The above setting is a -recommendation only . - -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -5. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -7. For a single shot, tap the LATTE -icon once . For a double shot, -tap the LATTE icon twice . Your -selected drink type will blink until -the espresso machine is done -preheating . After preheating, the -brewing process will begin . - -8. When brewing is complete, the -appliance will beep three times -and your latte will be ready for -sipping . - -9. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . + Investment Obligations: (In Millions of $): + +T y p e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +1 4 . 8 1 + +3 3 . 1 0 + +1 2 . 8 0 + +- 2 0 . 3 0 + +- 6 1 . 3 3 % + +1 2 . 2 7 + +2 7 . 0 8 + +9 . 2 0 + +4 2 . 0 0 + +3 2 . 8 0 + +3 5 6 . 5 2 % + +4 2 . 3 0 + +5 4 . 8 0 + +1 2 . 5 0 + +2 9 . 5 5 % + +Purpose, Accomplishments, Future Objectives: + +The CEA zero trust architecture investment focuses on an enterprise-wide investment approach to +build stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation +of a centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the Zero Trust Architecture investment, Treasury seeks to minimize implicit trust and +reinvigorate least privilege through a continual verification of each user, device, application, and +transition. The investment approach requires the Department to enhance visibility and threat detection +at the application level to improve its ability to support continuous threat analysis, detection, and +response, and enable to analysis of encrypted traffic. Compartmentalization, micro segmentation, and +reinforcing enforcement of continuous identity verification and conditional access policies will +improve our resistance to fraudulent tampering of privileged accounts. Our objective is a universal, +default security posture of “never trust, always verify” across our entire technology stack. Treasury +leverages the Zero Trust Architecture investment to provide additional human capital to support +management of service providers, as well as development, testing, and deployment of the applications +that have the security controls integrated into them. This investment also supports the Treasury Zero +Trust enterprise identity access solution, known as Common Approach to Identify Access Management +(CAIA), for all external federal partners who access the Enterprise Content Management (ECM) +SharePoint platform for shared services. + +The FY 2025 President's Budget includes $54.8 million for Treasury’s CEA Zero Trust Architecture +investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. + +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement + +appropriate safeguards to ensure delivery of critical services. + +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the + +occurrence of a cybersecurity event. + +In FY 2025, CEA – Zero Trust Architecture investments are targeted to the Department’s zero trust +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. 13 -USER MANUAL ESPRESSO MAKER GCM4230 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making an americano - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. - -Insert a filled or empty milk -reservoir into the appliance until it -clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -americano function will not be -available as an option . - -2. Set the froth strength control to -the Clean / Americano position . - -3. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -4. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -5. Position the froth dispensing tube -over a large heatproof cup . - -6. For a single shot, tap the - -AMERICANO icon once . For a -double shot, tap the AMERICANO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . - -7. When brewing is complete, the -appliance will beep three times -and your americano will be ready -for sipping . - -8. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . + Treasury is looking to support the following initiatives with final FY2025 appropriations: + +• Expansion of Treasury’s cyber threat intelligence platform to include additional cyber threat +intelligence feeds, additional material from existing feeds, and expand access to platform +capabilities and shared data. + +• + +Implementation of a TSSSOC lateral scanning capability across enterprise application and + +centralized incident response + +• Acquisition of an enterprise Endpoint Detection and Response (EDR) solution +• + +Investment in Organizational Resiliency and the associated human capital component necessary + +to implement zero trust architecture across the Department’s IT systems + +• Bureau implementation of zero trust architecture +• Development or acquisition of Security Operations Center (SOC) services for the Treasury + +Security Data Network + +• Expansion of Treasury’s mobile Secret Internet Protocol Router (SIPR) capability +• Enterprise-wide vulnerability management efforts including the establishment of an enterprise +penetration testing shared service offering and acquisition of a vulnerability management +tool to track and conduct analysis of enterprise-wide critical vulnerabilities + +• Development of application threat modeling capabilities to prepare for increasing resilience of + +systems and robustness of testing. + +Looking forward, CEA investments will aim to accomplish the below future objectives: + +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +HR LoB - HRConnect + +Description: + +HR Connect is a Human Resources enterprise system. It is a web-based solution built on PeopleSoft +software. HR Connect transforms core back-office HR functions, moving them from a processing- +centric capability to a strategic-centric capability. + +Investment Obligations: (In Millions of $): + +T y p e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +1 2 . 2 8 + +1 8 . 5 9 + +1 8 . 8 7 + +0 . 2 8 + +1 . 5 1% + +3 7 . 1 4 + +4 9 . 4 2 + +3 7 . 0 5 + +5 5 . 6 4 + +3 8 . 2 6 + +5 7 . 1 3 + +1 . 2 1 + +3 . 2 7% + +1 . 4 9 + +2 . 6 8 % 14 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Brewing in Manual Mode - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a heatproof cup over the - -drip tray . - -4. Tap the MANUAL ESPRESSO icon . - -The text below the icon will blink -for three seconds then the brew -cycle will begin . - -5. When your desired amount of -coffee has dispensed, tap the -MANUAL ESPRESSO icon again and -coffee will cease brewing . -note: The maximum amount of -coffee that will dispense is 14 oz . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + Purpose, Accomplishments, Future Objectives: -15 +HRConnect is Treasury's enterprise human resources system. It is one of four federal OPM HR Lines +of Business providing HR services to the federal government. HRConnect is based on a combination of +(a) web-based solution built on PeopleSoft commercial-off-the-shelf (COTS) software, (b) Software as +a Service (Saas) platforms (e.g. Talent Management and Career Connector) and (c) internally +developed applications (e.g. Entrance on Duty System). HRConnect transforms core back-office HR +functions, moving them from a processing-centric capability to a strategic-centric capability enabled +through its commercial software underpinning. Additionally, self-service components of the software +fundamentally transform the standard government HR service delivery model, putting additional +information, services and processes (i.e., personal data, position management, requests for personnel +action, recruitment, reporting, etc.) directly in the hands of managers and employees. -USER MANUAL ESPRESSO MAKER GCM4230 +HRConnect supports the common HR Line of Business processes and provides core HR functionality +that is interoperable, portable and scalable. This shared solution provides automated systems that are +configurable to the individual organizations' needs while providing a single solution across the +Department and federal landscape. HRConnect's core functions include: Personnel Action Processing, +Managing Payroll, Administering Benefits, Time and Attendance and Labor Distribution. By enabling +the retirement of legacy systems and automating and streamlining many aspects of human resources, +HRConnect facilitates increased efficiency and overall productivity for its customers. HRConnect is +the system used by all Treasury bureaus and several other government agencies (over 22 entities) with +over 200,000 employees and contractors in total. - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +HRConnect Top FY 2023 Accomplishments: -I +1. Implemented 74 HRConnect change requests to maintain regulatory compliance, improve the +employee and manager experience, and provide tools to streamline and enable efficient +processing for HR Specialists. -frothing Milk +2. Maintained a steady state of 100% availability for customers. +3. Treasury HRLOB implemented multifactor authentication (MFA) for all TSSC applications -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +and achieved M-21-31 compliance for the HRConnect High Value Asset (HVA). -1. Fill the milk reservoir with milk +4. Completed the migration of HRConnect’s data service, Workforce Analytics from on prem +data centers to a FedRAMP High secure cloud hosted environment, effectively scaling and +maturing the platform. -5. Tap the FROTH icon . +5. Delivered Integrated Talent Management (ITM) and HRConnect real-time performance plan -below the “Max” mark . +and learning data integration. -2. Secure the milk reservoir in the +6. Implemented new Remote/Telework Agreement Type and Program and Project Job Identifier -appliance until it clicks into place . +fields and Mass Update Process, 34K positions updated. -3. Set the froth strength control . +7. Streamlined the background investigation security process across Treasury by expanding the -4. Position the froth dispensing tube +Adjudication Tracker application to three additional bureaus. -over a heatproof cup . +8. Continued to modernize HRConnect delivering enhancements to improve the user experience -6. Milk will froth for a maximum of -three minutes . To stop frothing -earlier, tap the FROTH icon again -and milk will cease frothing . +for employees, managers, and HR Specialists. -16 +9. Deployed a visual representation of key Time to Hire (T2H) metrics via the Treasury + +Workforce Dashboard to track and optimize the hiring timeline and make data-driven hiring +decisions. -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +10. Continued employee self-service (ESS) redesign leveraging fluid concepts. +11. Supported gender identity and other preferred name representation in accordance with - I +management directives. -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +15 -I + Enterprise API Service Platform [MuleSoft]: Transition 15+ partners to the new API Service +Framework. Launched integrations between USAS to DO and FinCEN Adjudication Tracker, ARC- +FS Worklist API and SAM.gov integration to HRC to update vendor information. +In FY2023: Treasury Office of the Chief Information Officer (OCIO) recently assumed responsibility +for the management of the transformation of Internal Revenue Service (IRS) Human Resource +Information Technology (HRIT) applications in support of the Inflation Reduction Act. This transfer of +responsibility allows the IRS CIO to focus their efforts on improving the taxpayer experience +leveraging the Treasury OCIO’s shared services and common platforms to facilitate the modernization +of IRS HRIT systems, the rationalization pillars include: HRConnect, Integrated Talent Management, +USA Staffing, Enterprise Data Management, and the Employee Portal. -CLEANING AND MAINTENANCE +A few of technical improvements, included: Nessus implementation, CAST implementation, piloted +the transition of 3 environments to Oracle Managed Services compartment in Q3, improved the fire +wall configurations and finalized the closure of the Memphis and Martinsburg HRC footprint. -the following cleaning instructions should be followed after every -use of the espresso maker. +Treasury Enterprise Identity, Credential and Access Management (TEICAM) -Cleaning the portafilter and Brew Head +Description: -1. Turn the appliance off and unplug -the power cord from the outlet . -Wait for the metal parts to cool -down before touching them . +The Treasury Enterprise Identity, Credential and Access Management (TEICAM), formerly submitted +as EIdM, consolidates funding of Treasury implementing the Homeland Security Presidential +Directive- (HSPD) 12, E-Auth, and Federal PKI initiatives. -2. Remove the portafilter from the -brew head and dump the used -espresso grounds . Lift the filter -retainer clip so that it’s placed -over the filter edge, and hold it in +Investment Obligations: (In Millions of $): -place as you dump the espresso -grounds into the trash . +T y p e -3. Rinse the single-shot filter, +F Y 2 0 2 3 +A c t u a l s -double-shot filter, and portafilter -with warm, soapy water . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -4. Wipe the underside of the brew -head with a damp cloth to clean -off any excess coffee grounds . +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Cleaning the Water reservoir +C h a n g e i n +$ -1. Discard leftover water . +% C h a n g e -2. Wash the water reservoir with soapy water . Rinse and dry well . +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) -Cleaning the Milk reservoir and frothing tube +4 3 . 0 7 -1. Remove the milk reservoir from +7 5 . 9 4 -the unit . If there is any remaining -milk, pour the milk into a separate -container and store it in the -refrigerator . +7 5 . 9 8 -2. Wash the milk reservoir and milk +0 . 0 4 -reservoir lid thoroughly with warm, -soapy water . Do not use abrasive -cleansers and scouring pads . -note: Hand wash only! The milk -reservoir is not dishwasher safe . +8 1 . 6 2 -3. Ensure there is water in the water +7 8 . 9 9 -reservoir . +8 0 . 9 5 -4. Position the froth dispensing tube -over a large heatproof cup . +0.05% -5. Set the froth strength to the Clean -/ Americano position . Tap the -CLEAN icon . +2.48% -6. Wait until all the water has -finished dispensing before -removing the cup from -underneath the brew head -(the unit will beep three times -and other function buttons will -reappear) . +1.29% -7. Dispose of the water in the mug -and any remaining water in the -milk reservoir and water reservoir . +1 . 9 6 -17 +2 . 0 0 + +T o t a l O b l i g a t i o n s + +1 2 4 . 6 9 + +1 5 4 . 9 3 -USER MANUAL ESPRESSO MAKER GCM4230 +1 5 6 . 9 3 - I +Purpose, Accomplishments, Future Objectives: -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +The TEICAM Business Case consolidates funding that supports Treasury’s implementation of HSPD- +12, FICAM and PKI requirements. This investment supports the target vision of "One Treasury One +Card" to provide universal access. Availability and use of a PIV/identity management standard within +Treasury provides a mechanism for Physical and Logical access to Treasury-wide assets. The TEICAM +phased implementation provides all of Treasury: -I +• Trusted identity processes for both Federal and public users of Treasury applications +• Enhanced security (by decreasing data breaches and trust violations) +• Compliance with laws, regulations, and standards +• +• Expanded use of PIV encryption and digital signature certificates (ensuring adoption of multi- -descaling +Improved interoperability -After using the espresso maker for a while, limescale will begin to build up inside -the appliance . It is important to descale the appliance every 2-3 months . If the -espresso maker is used daily, the descaling process should be done monthly . +factor authentication) +• Elimination of redundancy -1. Fill the removable water reservoir +16 + + Treasury/TEICAM has achieved many of the defined goals for PIV card issuance, physical access, +logical access, data synchronization, enterprise single sign-on, federation, and PIV required for both +privileged and unprivileged users. These goals have helped the Department align with the required +OMB and FICAM goals. +Additionally, TEICAM has updated the Department strategic roadmap & planned for the following +investment goals: + +1) CDM II-PRIV/CRED Management + +o Continue working with the Managed Service Bureaus and Non-Managed Service + +Bureaus on implementation of Credential (CRED) and Privileged Access Management +(PAM) + +o On-board new bureaus, such as IRS and DO Orbit Organizations (TSSOC, OFR, etc.), + +to the enterprise CDM Managed Service Offering (MSO) + +o Continue working with DHS CISA on the integration activities for the Federal -5. On the following day, discard the +Dashboard for Treasury’s CDM Data reporting -with equal parts water and -vinegar between the “Min” and -“Max” mark . +2) Plan, test, & implement the issuance of version 8.1 PIV cards (PIV V8.1) across the enterprise -2. Use the froth dispensing +in FY21-FY24 -tube lever to rotate the froth -dispensing tube upward and -position a mug underneath it . +3) Expand usage of the Enterprise Derived Credential issuance capability to support -3. Tap the CLEAN icon . +authentication to Treasury services/infrastructure from mobile devices across Treasury in +FY21-FY24 -4. Allow remaining water and +4) As part of the PIV-I solution, implement an application programming interface (API) -vinegar to sit overnight in the -water reservoir . +connection between a Fiscal Service application used to capture applicant data and +HRConnect in FY21-FY24 -troubleshooting +5) Implement an enterprise Hardware Security Module (HSM) as a service in FY22-FY24 +6) Enterprise solution for tracking and monitoring the Only Locally Trusted (OLT) PKI -water and vinegar . +certificates generated and in use within each bureau -6. Rinse the water reservoir +7) Plan, design, and implement a modernized Treasury Enterprise Federation Service (TEFS) -thoroughly with tap water and -then pour the water back into the -sink . +solution that enforces multifactor authentication and aligns with the Treasury Zero Trust +Architecture (ZTA) -7. Fill up the water reservoir with +8) In an effort to improve cost-savings, the Department utilizes interagency resources to +authenticate users, synchronize data, and to procure and maintain enterprise-wide +compliant PIV credentials (USAccess). As a mixed life-cycle investment, the TEICAM +Operations and Maintenance tasks includes OMB, FISMA, and Cyber reporting specific to +identity, credential and access management. -water and begin brewing again . -This washes out any remaining -vinegar . -notE: If an odor of vinegar -persists, run another cycle with -water . +Planned objectives and accomplishments include: -ISSUE +• Maintaining above 95% PIV card issuance rate and providing replacements for the PIV in time -SOLUTIONS +sensitive activities; -Only the ESPRESSO and MANUAL -ESPRESSO icons appear on the control -panel. +• Maintained 100% PIV required privileged account access and 97% PIV required unprivileged -A drink has been chosen and the unit is -running, but no coffee is dispensing from -the portafilter. +access; -Water is spreading from the drip tray -when brewing or cleaning. +• CDM II privileged access management is operating with a cloud infrastructure and production +at four bureaus performing O&M functions; we have commenced technical work on +bringing the IRS on as the fifth and largest managed service customer; -The other icons will appear when the milk -reservoir is correctly inserted into the unit. -Insert the milk reservoir into the unit by aligning -the groove on top of the reservoir with the -notch on the unit and pushing. It will click into -place when done correctly. +• Treasury Enterprise Federation Service (TEFS) onboarded 12 new customers and integrated -Detach then reattach the water reservoir to the -unit (this may need to be done a few times). -Run a clean cycle. See CLEANING THE MILK -RESERVOIR AND FROTHING TUBE on page 17. +with third-party identity providers to expand citizen services and access; +• Executed a Treasury-wide BPA for citizen authentication services access to the entire +enterprise, allowing for secure and fast public access to Treasury systems; + +• Developed an alternative authentication solution and deployed it within 6 weeks to ensure + +continuity of government during the pandemic; + +• + +Implemented a hardware security module (HSM) as a service for the Derived PIV solution and + +DO. + +17 -A. Water dispenses from the froth dispensing -tube when brewing Americano or running a -clean cycle. Ensure the froth dispensing tube is -above the cup. + • -B. Empty the drip tray. +Implementation of the department-wide PKI Only Locally Trusted (OLT) monitoring solution. 18 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 + Major Non-IT Investments - WARRANTY -AND SERVICE +Main Treasury Building and Freedman's Bank Building -This Limited Warranty is for one full year from the date of -purchase and requires proof of purchase from an authorized -seller . It is applicable to the original purchaser only and is not -transferable to a third party user . Repair or replacement of -defective parts is solely at Gourmia’s discretion . In the event -that repair isn’t possible, Gourmia will replace the product/part . -If product repair/replacement won’t suffice, Gourmia has the -option of refunding the cash value of the product or component -returned . +Description: -Product defects not covered under the Warranty provisions -include normal wear and damage incurred from use or -accidental negligence, misuse of instruction specifications -or repair by unauthorized parties . Gourmia is not liable for -any incidental or consequential damages incurred by such -circumstances . +Correct life safety and code issues, reduce building systems risk, and maintain the buildings. Absent +full funding to perform a complete repair and renovation, Treasury is utilizing available funding to +correct the most severe issues. -Only valid within the United States . +Investment Obligations: (In Millions of $): -register your product at -www.gourmia.com/warranty +T y p e -for customer service, questions or comments +F Y 2 0 2 3 +A c t u a l s -Visit +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -gourmia.com +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Email +C h a n g e i n +$ -info@gourmia.com +% C h a n g e -Call +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -888.552.0033 -Mon-Thu 9:00 a .m . to 6:00 p .m . ET -Fri 9:00 a .m . to 3:00 p .m . ET -Languages: English, Spanish, French +9 . 7 9 -@gourmia +0 . 0 0 -I +9 . 7 9 -E -C -v -r -E -S -d -n -a -Y -t -n -a -r -r -a -W +3 . 5 5 -19 +0 . 0 0 + +3 . 5 5 + +5 . 5 5 + +2 . 0 0 + +5 6 . 2 5 % + +0 . 0 0 + +5 . 5 5 + +0 . 0 0 -USER MANUAL ESPRESSO MAKER GCM4230 +0 . 0 0 % - Get cooking with +2 . 0 0 + +5 6 . 2 5 % + +Purpose, Accomplishments, Future Objectives: + +This investment is to address life safety and code compliance issues, reduce building systems risk by +upgrading a number of outdated systems, and bring both facilities into alignment with current building +standards. Absent full funding to perform a complete repair and renovation of these historical +buildings, available funding will be used to correct the most urgent issues. These investments are being +executed with the expectation that were Treasury to pursue a full renovation and modernization, recent +investments could be largely retained, achieving cost savings over the long term. These investments +support a safe and healthy work environment that meets Treasury operational requirements. Project +needs are mission focused and prioritized based on life-safety, security, code discrepancies, and needs +that pose significant financial risk if not addressed in a timely manner. + +Treasury’s three-step long-term strategy to continue to maintain and modernize its owned spaces: (1) +secure the building’s outer envelope; (2) conduct a condition assessment to identify additional needs +associated with the buildings’ continual aging and deferred maintenance; and (3) based on this +assessment, conduct a holistic modernization of the building’s systems and infrastructure. Targeted +investments address life safety and code compliance issues, reduce building systems risk by upgrading +outdated systems, and bring both facilities into alignment with current building standards. Absent full +funding to perform a complete repair and renovation of these historical buildings, available funding +will be used to correct the most urgent issues. These investments are being executed with the +expectation that were Treasury to pursue a full renovation and modernization, recent investments could +be largely retained, achieving cost savings over the long term. These investments support a safe and +healthy work environment that meets Treasury operational requirements. Project needs are mission +focused and prioritized based on life-safety, security, code discrepancies, and needs that pose +significant financial risk if not addressed in a timely manner. + +Treasury Operations has continued to strategically focus on restoring the health of the building +envelope (shell), to correct the deteriorating building structure and infrastructure. Components of the + +19 -We’d love to hear from you! -888.552.0033 -info@gourmia.com + Main Treasury building that have been repaired, replaced, or funded include new roofs (FY 2019/2021) +and repairs to 40 percent of the windows (FY 2015/2017). The Freedman’s Bank Building components +that have been repaired, replaced, or funded/planned include new roofs (FY 2017/2019), domestic +water line (FY 2020/2022), and replacement of all windows (FY 2009/2011). Funding and completion +of these exterior items represent significant progress towards the preservation of the two buildings, +maintaining a safe and healthy workplace, and reducing damage to the interior plaster and paint. The +contract for the first phase (west façade) of the Main Treasury exterior restoration project was awarded +late in FY 2021 and completed during FY 2022. Exterior restoration work continued through 2023 with +the completion of phase 2 (south) and phase 3 (north) north façades. In FY 2024, assuming adequate +funding, Phase 4 work will commence and, in FY 2025, with this requested funding, work on Phase 5 +will commence. + +The MT Building houses the entire chilled water plant for the Main Treasury Complex. A five cell +cooling tower feeds four water cooled centrifugal chillers providing a total cooling capacity of 2,050 +tons. The chillers are located in a mechanical room under the northwest lawn. The four chillers provide +over 98% of the cooling required for the entire Treasury Headquarters complex; therefore, the plant is +operated 24/7, 365 days per year. The Freedman’s Bank Building (FBB) does not generate chilled +water but rather uses chilled water from the main plant distributed by a dedicated secondary chilled +water pump which serves most air handling units directly. The Chilled Water System is over 30 years +old and has a wide variety of repair needs including bearings, motors, valves, distribution piping, +insulation, controls, and other repairs. The system has redundancy for repairs. However, with all of the +components aging at the same time the occurrence of repair needs combined with the time to repair +diminishes the reliability of the redundant capacity. It is imperative that these items are +repaired/replaced before the available redundancy is lost. A system breakdown would indefinitely +shutdown the Treasury Complex’s daily operations. This first-year cost of a multi-year project includes +design for this effort. + +There is currently a backlog of damaged paint and plaster as well as frayed carpet throughout MT and +FBB. Some of the paint and plaster damage is due to water intrusion. These funds will supplement +other funding to address the backlog in repairing these problems and implementing a regular schedule +of maintenance. The damaged paint and plaster are unsightly and pose potential health and safety risks, +while the frayed carpet is a safety hazard. 20 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 \ No newline at end of file diff --git a/q119/random_k4/question.json b/q119/random_k4/question.json index 62c1a70decd064372f30a0bae24be64f1c46ceb1..ca5e21c9d7b3edcce88be93fa5588d159cbb1c8c 100644 --- a/q119/random_k4/question.json +++ b/q119/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "tacl_a_00660.pdf", - "web_d8f936ecd4367f91.pdf", - "web_4cf87d006f2e4a65.pdf", - "web_2ed0552c74ec7055.pdf", - "web_77f281bc58aa8364.pdf" + "web_d44e9a235a572259.pdf", + "resize_r2_4.pdf", + "web_94442ce12f8c032c.pdf", + "web_19b03cfaf44f6244.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q119/random_k4/random_1.md b/q119/random_k4/random_1.md index 2cfb2a95e81c5a4894ee70af1d1b3fdec18ec7c6..405f7d94774fb08f29f0d478b4b1f971f0f80612 100644 --- a/q119/random_k4/random_1.md +++ b/q119/random_k4/random_1.md @@ -1,1447 +1,705 @@ - b o s t o n c o l l e g e +Fixture and Store Designs that Draw Shoppers +into the Center Store to Capitalize on Sales -Exchange Student -Handbook + 10 Ways to Lift Center Store Sales -Office of International Programs +Sources Cited - We are delighted that you have chosen to attend Boston College as an ex- -change student. We value the partnership that we have developed with your -home university and look forward to welcoming you as a representative of -your institution. Your presence will add a significant element to the interna- -tional dimension of Boston College, and we hope you will take advantage of -every opportunity to share your culture with our community. +Cohen, Deborah A., and Susan H. Babey. “Candy at the Cash Register — A Risk Factor for Obesity +and Chronic Disease.” New England Journal of Medicine. (2012): n. page. Web. 21 Jan. 2013. +. -As an exchange student, you are given the opportunity to study at Boston Col- -lege for a semester or academic year as part of your academic program at -your home university. Please note that it is not possible to transfer to Boston -College to finish your degree. Boston College has a variety of graduate pro- -grams that will welcome your application once you have received your degree. +Sorensen, Herb. Inside the Mind of the Shopper. 1st ed. Upper Saddle River: Pearson Education, +Inc., 2009. -Please visit www.bc.edu for detailed campus information. For information on -the Office of International Programs see www.bc.edu/international. +Strom , Stephanie. “In Grocery Stores, the Perimeters Take Center Stage.” New York Times +10/08/2012, New York edition B1. www.nytimes.com/2012/08/11/business/in-grocery-stores-the- +perimeters-take-center-stage.html?pagewanted=all -We wish you all the best as you begin to prepare for your upcoming trip to -Boston. Please let us know if we can be of any assistance. +Underhill, Paco. Why We Buy: The Science of Shopping. 1st ed. New York: Simon & Schuster +Paperbacks, 2009. 3-27. -Bernd Widdig, Ph.D. Esther Messing -Director, Office of International Programs Assistant Director for +“Stop Shopper Leakage in Center Store.” Shopper Sight News. 70 (2012): n. page. Web. 21 Jan. 2013. +. - International Exchange Students +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. +Not responsible for errors or omissions in printing. -1 + 10 Ways to Lift Center Store Sales -www.bc.edu/international +If store perimeter provides “sizzle,” center store must serve “steak” - TABLE OF CONTENTS +Colorful, creative, winding and sometimes even whimsical in-store perimeter displays of +produce, sampling stations, deli, sushi and coffee bars, and endless walls filled with everything +from flat screen TVs to the hottest brands of clothing are no doubt here to stay. Engaging +perimeter presentations are among the latest trends in fashionable and attractive retail +merchandising that represent a sweet spot for today’s progressive merchandisers looking for fresh +new ways to connect with shoppers. - INTRODUCTION + However, while perimeter sales within many mid- to large-size shopping environments clearly +represent a vibrant growth trend, a similar merchandising discipline and focus should be applied +to the center store. After all, smart merchants understand that what’s good for the perimeter +should also be good for the interior, because what good is the sizzle of the perimeter shopping +zone without the steak -- and inherent profit margin -- of the center store? - BEFORE YOU LEAVE HOME + Unfortunately, center store is more often than not treated as a warehouse of separate products +on cold, uninviting rows of fixtures. While shoppers may find the products in center store they +need, they seldom find the solutions they truly want to make their lives more enjoyable or +productive. How will that new jacket or shampoo fit their lifestyle? How can that product, or +combination of ingredients, help them make a better or healthier meal? Why will that dog food +help their pet be healthier and happier? - Student Visa - Travel Plans - Medical Insurance - Living Expenses +Madix has introduced new products that improve merchandise visibility. Radius and Wraparound Shelves +used on end caps make merchandise down inline runs more noticeable because these shelves do not + utilize end panels, which normally obstruct merchandise as you look down a run. - YOUR FIRST DAYS IN BOSTON - - Getting to Boston College - On-Campus Housing - Off-Campus Housing - Hotels and Temporary Housing - Checking-in and Orientation - - ACADEMICS - - Full-time Status - Course Selection and Registration - Drop/Add Period - Faculty Advisors - Tutoring - English Classes for Non-Natives - Transcripts of Academic Work - - ADMINISTRATIVE ISSUES - - Student Identification Card - Medical Insurance - Health Services - University Counseling Services - Students with Disabilities - E-mail - - LIFE BEYOND ACADEMICS - - CONTACT INFORMATION +3 -2 +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales + +What’s actually happening in center store? + +Retail store perimeter growth has come mostly from the pull demand from customers wanting + +fresh, unprocessed foods and products that exude health, youthful appeal and entertainment. +Nowhere is the effect on center store sales more evident than in grocery stores and big food +brands such as Heinz, Kraft and Kellogg. Published reports indicate sales or volume declines +as large as 2.8-percent for some of these powerhouses due to the shifting consumer interest in +healthier non-processed food choices most often found on the perimeter of grocery stores. As +a result, packaged goods manufacturers are researching ways to leverage the popularity of the +perimeter through cross merchandising to make the center store -- along with their profitable +brands -- more exciting and appealing to shoppers. + + Unfortunately, center store sales within most retail environments have been decreasing in +recent years. The reasons why are akin to the proverbial elephant in the room. Rather than +address this huge challenge head-on, many retailers have instead decided to ignore it and focus +on what is working from a perceptual standpoint, which is the perimeter. Regardless of the center +store neglect, and using grocery stores as an example, a report by John Lewis, president, consumer +North America for The Nielsen Company, points out that the center store represents 73-percent +of total store sales and 77-percent of profit despite growing shopper leakage from people who +browse and think about a purchase but ultimately don’t buy. This imbalance of sales and profits +between center and perimeter can be addressed not only through SKU count, but also through +lower margins in the perimeter due to high labor costs and shrink. + +This tale of the two shopping zones is supported by a Center Store Mega Study of 2 million +individual shopping trips that found as many as a quarter of shoppers walk the center store yet +don’t buy anything (“Shopper Sight News”). In addition, during the average 13-minute shopping +trip, shoppers spend on average only two minutes in the center store. Clearly when analyzing +the chasm between center store sales and the average time spent there by shoppers, the +opportunities are enormous for progressive retailers to better integrate perimeter and center store +merchandising efforts targeted at growing customers’ overall baskets. + +So, if bold, new methods of merchandising are working to drive traffic for the perimeter, +why are retailers continuing to ignore the bread and butter represented by the center store? +Within this vast and extremely valuable internal battlefield, most stores still stubbornly stick +with outdated merchandising methodology including everything from shelving and lighting to +traffic patterns and category management. The dichotomous relationship that currently exists +between the perimeter and center store ignores how shopping patterns have shifted and, more +importantly, represents lost sales opportunities. + + When asked to comment, Blake Sloan, partner and store designer at the full-service +architecture and design firm, Heights Venture Architects L.L.P., noted that the center store is often +not addressed with the same creative thought process as the perimeter. + +“It’s extremely difficult to break away from product-centric category management and focus +efforts towards a consumer-centric shopper management,” said Sloan. “The potential profit power +represented by the center store is too valuable to simply ignore. Instead, we suggest retailers +envision the center store experience as a unique merchandising and marketing opportunity. +Retailers should enhance the center store with creative fixturing, lighting and product +identification as they would for the perimeter to create a holistic total store experience to build +customer loyalty and improve sales and profits.” 4 -5 -5 -6 -6 - -7 -8 -9 -10 -10 - -11 -11 -13 -13 -14 -14 -14 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -15 -15 -16 -16 -17 -17 + 10 Ways to Lift Center Store Sales -18 + As a New York Times article entitled “In Grocery Stores, the Perimeters Take Center Stage” +by Stephanie Strom recently stated, “Long, soldier-straight rows of shelves...have long been the +heart of the American grocery store but are now showing signs of the grocery equivalent of +atherosclerosis.” The same could be said of pretty much any mid- to large sized retailer offering +hundreds, if not thousands of SKUs on the floor. After all, is there anything less inviting than +staring down a narrow, 60-foot long aisle boarded by shelves upon shelves of cans and boxes that +all seem to blend together? -20 +The bigger, more important equation is how to create a more pleasant and inviting center store -www.bc.edu/international +that works with, if not mimics, the perimeter? Solving the problem of how to keep shoppers in +the center store longer than the average two minutes with a sexier, well-thought-out solutions- +oriented approach would go a long way toward moving more high-margin goods and better +leverage consumer needs to win more of their shopping trips. - CHECKLIST + While there is no single solution to the center store challenge, here are 10 smart tips that can +go a long way toward creating a holistic shopping experience that leverages the entire store and +drives up both sales and profits. - BEFORE YOU LEAVE HOME +Solution #10: Take the edge off -Apply for F-1 or J-1 visa from local US Embassy or Consulate -Make travel plans for arrival and departure from BC -Secure on-campus or off-campus housing. If necessary, arrange tem- + Old-school thinking is that if you trap the shopper and force them to pass by lots of product +facings, you’ll create more sales opportunities. Still, some retailers persist in creating canyons of +metal shelving and expecting customers to willingly funnel through them. In fact, the opposite +is true. Not only do shoppers dread entering long aisles, in their hurry to get through them they +often scurry past products they might otherwise be interested in purchasing. -porary housing or hotel +“Too many stores suffer from the ‘peek and run’ syndrome where shoppers simply refuse to +enter long rows bordered by high shelving on either side unless they absolutely have to,” said +Shawn Kahler, Madix Vice President of Sales and Marketing. “This effectively negates any hope +shoppers will spend any meaningful time in the center store, and it certainly doesn’t create a +pleasant in-store experience.” -Complete and submit Immunization and Health Forms to BC - University Health Services +Kahler says the answer is quite obvious: make aisles shorter, less linear and overall more appealing +to remove the intimidation factor in hopes shoppers will choose to walk through the store, enter +aisles, browse and shop. - AT BOSTON COLLEGE +“It’s actually possible to create many different types of unique, interesting and appealing aisles -Check in at the Office of Residential Life to pick up keys and get hous- +by using standard parts and fixtures rather than designing custom shelving that is often cost +prohibitive,” he said. “For example, Madix, Inc. offers retailers several options such as its curved +gondola system that actually use standard gondola and wall parts to better manage space and +generate a maximum return on investment.” -ing assignment or locate off-campus housing -Check in at the Office of International Programs -Check in at the Office of International Students and Scholars -Attend orientation. Dates, times, and locations are in your acceptance + He says that radius curves of 6 or 10 feet provide maximum flexibility to retailers who wish to +transform their outdated straight aisles into something more inviting and productive. There are +also much longer sweeping, curved fixtures that create a panorama effect of products that are +also pleasing to the eye and create much more product visibility than do long, straight and narrow +aisles. The key learning is that retailers have available to them the tools and professional expertise +enabling them to create better, more effective center stores. -packet +5 -Get student identification card, set up e-mail account, and establish -library account. This process will be explained at the orientation. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -3 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - INTRODUCTION - -Office of International Programs - -The Office of International Programs (OIP), located in Hovey House at 258 -Hammond Street, is your primary contact while you are studying at BC. The -exchange program that brought you here is administered jointly between OIP -and a comparable office at your home university. In addition to admitting you -to Boston College, we pre-register you for courses, assist with housing selec- -tion, and can assign you to a faculty academic advisor if you wish. We also -maintain regular contact with our counterparts at your home university. -Please feel free to come to us with any questions or problems before, during, -or after your semester or year at Boston College. - -Our office also sponsors the Exchange Mentor Program. This program aims -to match you with a Boston College student who has studied abroad at your -home university. Mentors maintain regular contact with their students. - -OIP works alongside numerous offices on the Boston College campus to en- -sure that you have the same access to facilities, services, and opportunities -as all other BC students. Other offices that you may come into contact with -include: the Office of International Students and Scholars, Student Services, -Health Services, Residential Life (which encompasses the On-Campus Hous- -ing Office and the Off-Campus Housing Office), and various academic de- -partments. - -Office of International Students and Scholars - -The Office of International Students and Scholars (OISS) is responsible for -issuing the immigration documents required for international students to -apply for a student visa. They also issue work authorization and advise on -other issues related to maintaining legal status in the United States. In col- -laboration with OIP, OISS organizes a mandatory International Student Ori- -entation each semester. - -In addition to these important services, OISS sponsors programs for interna- -tional students throughout the year, including welcome socials, Boston walk- -ing tours, the Thanksgiving Host Family Program, and many other activities. -They also provide undergraduate international students with International -Assistants, who are BC students serving as peer advisors to international stu- -dents. You will meet these International Assistants during orientation. +Madix’s Radius end caps married with curved Maxi Line draw shoppers into the center store. +The undulating shelves make merchandise stand out. Madix offers components to curve +entire fixture runs, fixture sections, or individual shelves. -4 +In combination with shorter and more organic-shaped aisles, Dr. Herb Sorensen, Ph.D., author -www.bc.edu/international - - Things to Know Before You Leave Home - -Student Visa - -As a participant of an exchange pro- -gram to Boston College, you will -most likely be entering the United -States either on an F-1 student visa -or a J-1 exchange visitor visa. To ap- -ply for a student visa you will need: - - - - - -the enclosed Certificate of Eligi- -bility Form I-20 for F-1 visa (or -Form DS2019 for J-1 visa) - -a passport valid for at least six -months after your scheduled -return to your home country - - proof of funding such as a letter - -from your bank - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -With these items, you must apply to -the appropriate Embassy or Consu- -late General of the United States in your home country. If you are applying -for an F-1 visa, be sure to complete the items on the bottom of pages 1 and 3 -of your I-20 before applying for your visa. - -It is important to understand that as an exchange student on a non- -immigrant F-1 or J-1 visa you are required to be enrolled as a full-time student -for the duration of your participation on the exchange program. Please refer -to the Academics section for more information on full-time status. - -Travel Plans - -As an exchange student, it is your responsibility to arrange for your arrival to -and departure from the United States. Please refer to the Academic Calendar -enclosed with your acceptance packet for important dates. - -You should plan to arrive at Boston College on a weekday (Monday - Friday) -between 9:00 am and 5:00 pm, when university offices are open to assist you. -On-Campus Housing staff may be available on the weekend prior to the start -of the semester. Details will be sent at a later date. - -If you are planning to live off-campus and have not secured housing prior to -departure you should arrive as early as possible (5-7 days) to look for housing -before the beginning of orientation. Please see p. 9 for more information. +of “Inside the Mind of the Shopper,” advocates staggering the fixture patterns themselves from +section to section to break up long sight lines. The outcome of this advice is that as a shopper +looks down any aisle toward the center store, they actually see the end cap of the next shelving +fixture or more inviting concave or radius shelving and not the other side of the store. -5 +Solution #9: Drop the height of shelving -www.bc.edu/international - - Also, please be aware that the following immigration regulations are in effect -for all F-1 and J-1 students: - - New international students will not be permitted to enter the U.S. more -than 30 days before the start date of their program as indicated on the I- -20 or DS2019 form. - - Upon completion of studies (i.e. the end date listed on the I-20 or - -DS2019), F-1 students are permitted to remain in the U.S. for 60 days -and J-1 students are permitted to remain in the US for 30 days. - -Additional immigration information is available at the following web site: -http://www.bc.edu/offices/odsd/intercultural/immigration.html. - -Medical Insurance - -The Commonwealth of Massachusetts requires all full-time students (both -American and international) attending colleges and universities in Massachu- -setts to subscribe to a US-based insurance plan. You will be automatically -enrolled in the BC medical insurance upon your admission to the university. -If you do not want this insurance, you must submit proof of insurance with a -provider that has a US office after your arrival. Please see p. 15 for more infor- -mation on BC medical insurance. - -Living Expenses - -Past exchange students -have been surprised by -the cost differences be- -tween their home coun- -try and the Boston area. -When estimating the -costs of your studies -and living in Boston, be -sure to budget for -housing, meals, trans- -portation, books, enter- -tainment, and any trav- -els you plan to take be- -yond the local Boston -area. Specific informa- -tion on costs can be -found on the Estimated -Expenses page in the -back of this handbook. - -www.bc.edu/international + Merchants seem to love using vertical space. The higher, the better, right? Actually, according +to Dr. Sorensen, people do not prefer to shop tall shelves crammed with products in the center +store. Instead, they respond best when faced with shelving that’s 66 inches or lower and 15 to 20 +feet long or not longer than 30 feet. Such shelving ensures customers can easily see and navigate +their way through the store without feeling trapped or lost. An exception would be a retailer such +as JCPenney which has begun using tall walls on the perimeter of its stores where they do not +interfere with sight lines to promote fashion accessories. 6 -s -t -t -e -s -u -h -c -a -s -s -a -M - -, -y -b -n -a -r -G - - YOUR FIRST DAYS IN BOSTON - -Getting To Boston College - -Most students come to Boston -by flying to Logan International -Airport. Getting to Boston Col- -lege from Logan is not hard and -can be done in many ways: - -Taxi and Airport Shuttle - -The simplest, yet most expen- -sive, way to get to Boston College -is by taxi. - -Another alternative is to utilize -one of the private airport shuttle -services that operate from Logan -and will drop you off at Boston -College. Prices vary by company. -You might need to make a reser- -vation in advance. For a listing of shuttle services available at Logan, visit -www.massport.com/logan. A listing of private car services can be found on -the internet by Googling “livery services in Boston.” - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Public Transportation - -Finally, you can utilize public transportation to get to Boston College. The -subway is known locally as the “T.” For more information consult the Massa- -chusetts Bay Transportation Authority website at www.mbta.com. - -To take the "T" to Boston College, take the airport shuttle bus to the Airport -subway stop. Take the “Blue Line” subway from the airport to the -“Government Center” subway stop, where you should change trains to the -“Green Line.” Please note that there are four branches of the Green Line! Be -sure to take the “B-Line” train outbound, destination "Boston College," and -get off at the last stop. A public transportation map is included in your accep- -tance package. Obtain a reusable “Charlie Card” at the subway station to re- -duce your T fare. - -For more information on Logan International Airport and traveling into Bos- -ton from the airport, please consult www.massport.com/logan. +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -7 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - Moving Into On-Campus Housing - -Boston College has a number of residence halls. If you will be living on- -campus, we try to place you in residence halls with kitchen facilities. Students -living in rooms without access to kitchen facilities are required to purchase -the Boston College meal plan, which pays for meals in the campus dining -halls throughout the semester. Students living in rooms with kitchen access -have the option of purchasing the meal plan. If you do not wish to purchase -the full meal plan, you should strongly consider the “self-catering” option, -which allows you to purchase "Dining Bucks" in smaller amounts throughout -the semester. For more information on meal plans and Dining Bucks visit the -Dining Services web site at www.bc.edu/offices/bcds. - -Please be aware that sheets, blankets, pillows, towels, etc. are not provided in -the residence halls. These items can be purchased inexpensively in Boston, -so you may not want to take up luggage space with bedding and towels. - -The cost for on-campus housing varies by residence hall and room arrange- -ment. You will be assigned to available housing and will not be able to re- -quest a specific residence hall. Your BC student account will be billed once -your housing assignment is confirmed. You are then responsible for paying -for your housing upon arrival at Boston College. - -Every student on campus has a private phone number and a phone jack in -her/his room. You must provide your own telephone, which can be pur- -chased inexpensively in nearby stores. Local telephone service and voice mail -are free. Campus rates for long distance and overseas calls are very competi- -tive. Long distance calls are billed to your student account. Many students -also use cell phones and Skype for long distance and international calls. - -All on-campus rooms have free internet access, available through either an -Ethernet cable or wireless device. Contact 617/552-4357 with questions. - -Students with on-campus housing can move into their rooms on the first day -of the International Student Orientation for Fall/Full Year students, or on the -day before the Exchange Student Orientation for Spring students (see the -enclosed Academic Calendar). Be sure to arrive between 9:00 am and 5:00 -pm, when the housing office will be open to check you in and give you your -room key. The On-Campus Housing Office phone number is 617-552-3060. - -If you plan to arrive outside of normal business hours, please notify OIP as -soon as you make your travel plans so we can inform the housing office. -Upon arrival at BC, you should go directly to your residence hall and ask for -the RA (Resident Assistant) on call. The RA will let you into your room and -you can pick up your key on the next business day from the Residential Life -office. +In addition, traditional straight-faced shelves in most stores are designed to hold a maximum +capacity of products from top to bottom, but they are not shopper friendly. Such a design stifles +the ability to view products on lower shelves, often forcing the shopper to take a few steps back +into what may be a crowded aisle to see what’s on the bottom shelves. Such fixtures also visually +create a closed-in, claustrophobic feeling as shoppers walk down an aisle. -8 + One solution is to install pyramid-style fixtures on both sides of an aisle. Such fixtures have top +shelves that are recessed from the shopper by about 16 inches, giving shoppers nearly three full +feet of apparent extra aisle width. This allows higher-velocity products to be inventoried on lower +shelves while featuring higher-priced or lower-velocity items toward the top where they can gain +more exposure. -www.bc.edu/international - - Off-Campus Housing - -If you will be living off-campus, you -will need to make your own housing -arrangements. The Boston College -Off-Campus Housing Office has up- -to-date lists of apartments and -roommate openings near Boston -College and acts as a referral source -for rooms in private homes. Your -name will be given to the Off- -Campus Housing Office and they -will contact you and assist you in -your housing search. The staff can -advise you on suitable arrange- -ments, convenient locations, etc. -However, they cannot make ar- -rangements on your behalf. If you -choose to utilize the above services, -it is possible that you may secure -housing prior to your arrival in the -United States. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Because apartment rentals in the BC area are expensive, many international -students either share an apartment with other students or live with families in -private homes in the surrounding area. We advise all exchange students who -will be seeking off-campus housing to arrive at least 5-7 days before orienta- -tion in order to have enough time to make the necessary arrangements. You -should also make necessary hotel arrangements prior to arrival (see the Ho- -tels and Temporary Housing section for suggestions). - -Finally, be sure to carefully read your lease or other agreements to rent an -apartment before signing and make certain that you understand the terms. If -you are unclear about what a particular document means, consult the Off- -Campus Housing Office for advice. There are laws to protect tenants in Mas- -sachusetts, so be sure to find out what your rights are in case you run into a -conflict with your landlord. - -The Off-Campus Housing Office web site offers helpful hints to assist you in -your search. To access their web site, go to www.bc.edu/housing and click on -the Off-Campus Housing link on the left sidebar. - -Once you have settled into your accommodations for the semester, be sure -to inform OIP of your address and phone number so that we know how to -contact you when necessary. You will receive a local housing notification form -in your orientation packet. + A less radical alternative to pyramid fixtures is an offset shelving design that moves one or +more of the top shelves back in small increments between 8-12 inches. While slightly less effective +than pyramid fixtures, offset shelving still helps reduce the train-tunnel effect and allows the use +of sloped signboards that can be used for various in-store marketing devices such as digital media. -9 +Regardless of the type of fixture that’s used, Dr. Sorensen points out that there is a delicate +balance between trying to capture the shopper’s time, holding a shopper at a product display, and +closing the sale. Trying to sell too much within a small space can actually harm sales conversion, +because the longer it takes a shopper to make up his or her mind, the more angst they feel, the +more likely they are to abandon that area of the store, and the less time they are apt to linger in +other areas of the store. The rule of thumb should be convenience, which translates into fast. -www.bc.edu/international +Madix’s product lines are available in numerous sizes, +so a retailer can deploy fixtures that best meet its sales objectives. - Hotels and Temporary Housing +7 -If you will be arriving in Boston early to find off-campus housing or have ac- -companying guests who need temporary housing, you can obtain lists of tem- -porary accommodation options from the Off-Campus Housing web site at -www.bc.edu/housing (click on the Off-Campus Housing link). +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -Checking In + 10 Ways to Lift Center Store Sales -As an exchange student at Boston College, there are two offices with which -you must check in during normal business hours (Monday to Friday, 9am - -5pm) upon arrival at Boston College: the Office of International Programs -(OIP) and the Office of International Students and Scholars (OISS). +Solution #8: Widen the aisles -Upon your arrival, please come to OIP in Hovey House and check in so that -we know that you have arrived safely. + Using pyramid or offset fixtures are simple methods to effectively open up aisles visually and +literally from the feet up to head height, but another more obvious solution is to simply widen the +entire aisle. Some retailers serve up just enough room for two carts to pass by one another in an +aisle, while others make it almost impossible for even a single cart to pass between displays. While +the solution will vary between retail formats, the outcome should always cater to what makes the +shopper most comfortable and put them in a buying, not evacuation, mode. -You then need to check in with OISS. All international students are required -under the terms of their visas to check in officially at the Office of Interna- -tional Students and Scholars for immigration purposes. You will need your -passport with the white I-94 card attached, as well as your I-20 or DS2019 -form in order to check in. +In general, shoppers equate open space with the freedom to shop. The best way to generate -Orientation +traffic is to allow traffic to flow without restrictions. A shopper at a store with tall, narrow aisles +is less likely to enter the aisle if it has but one other shopper in it, much less several. With such +aisles, a single shopper with a cart stopped in an aisle can occupy as much as 10 feet of shelf facing, +effectively blocking that merchandise from other potential buyers. Obviously, this leads to far +fewer sales opportunities. -The Office of International Programs, in collaboration with the Office of Inter- -national Students and Scholars, sponsors an orientation program for all new -international students which includes new exchange students. For fall and full -year students orientation takes place the week before classes begin in Sep- -tember. Spring students have orientation prior to the first week of classes in -January. The orientation will provide you not only with important information -about Boston College, but also with valuable opportunities to meet other ex- -change students, staff members, and Boston College students. + On the other hand, a shopper at a store with aisles able to accommodate multiple shoppers +shoulder-to-shoulder is less likely to turn away. In this scenario, even if another shopper is camped +in front of a shelving unit another shopper can stand to the side and browse. The key is to allow a +relatively free flow of multiple shoppers through any aisle at any time. -An important part of this program is the English Placement Test, adminis- -tered to non-native English speakers. All international students, including -exchange students, are required to take this test if their first language is not -English. + Dr. Sorensen states that open space combined with short and low fixtures are open invitations +to shop. Likewise, he says, “Narrow, crowded aisles like packed highways can lead to social +pathology and even ‘aisle rage,’” whereby shoppers become angry at one another for blocking +each other or even blame the store for failing to be customer friendly. Ironically, retailers who do a +good job merchandising their perimeters typically do so with wide aisles, low fixtures and a sense +of openness. So why not apply the same principles to the center store experience where upward +of 75-percent of all sales occur? -Please see the OISS letter in your acceptance packet for additional informa- -tion about the orientation. +Madix designs its products to more efficiently utilize space, +which means a retailer can devote more space around fixtures to its shoppers. -www.bc.edu/international +8 -10 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales - ACADEMICS + Another, less obvious reason to expand aisles is what Paco Underhill, author of “Why We Buy: +The Science of Shopping,” calls the “butt-brush effect.” During research his firm conducted +for Bloomingdale’s department store, they inadvertently observed shopper interaction with a +clothing fixture near a ground-floor entrance they were monitoring. The fixture was adjacent to +a high-traffic aisle, and they saw that customers interacting with the fixture would typically get +“butt-brushed” from behind by people heading into or out of the store. After a few such bumps, +shoppers abandoned the fixture and moved on without buying. As a result, sales from the fixture +were well below expectations. After moving the fixture to another location that eliminated +the possibility of butt-brush, the retailer found that sales from the fixture immediately and +substantially increased. It simply boiled down to location. -Full-Time Status +The question retailers should be asking themselves is how often their customers are spooked -All international students in the US are required to maintain full-time student -status for the duration of their studies in the US. Undergraduates must be -enrolled in a minimum of 12 credits per semester. Graduate students must -be enrolled in a minimum of 9 credits per semester. Undergraduate exchange -students should expect to take between 12 and 15 credits per semester. -Graduate exchange students should expect to take between 9 and 12 credits -per semester. Failure to maintain full-time status will affect your legal status -in the United States, so it is very important to be enrolled in the appropriate -number of credits. +by being too close to one another or being brushed against other shoppers trying to navigate +around them? The solution may be something as easy as a realignment of a fixture or a more +complicated readjustment of the entire aisle space. -Selecting and Registering for Courses +Solution #7: Go mobile -We attempt to register you for the classes you requested on your application, -but please keep in mind that you must be flexible about what courses you will -take. We cannot guarantee that you will get all of the classes you request. If -you need to make any alterations to the courses you selected on your applica- -tion, please e-mail OIP at oipexchange@bc.edu. +Part of the challenge of store fixtures is that they are by definition “fixed,” meaning they +often are anchored in place and unlikely to be moved outside of major store resets. Rather than +be entirely tethered by immovable fixtures, some of the world’s most successful retailers have +embraced the more fluid flexibility of mobile displays to complement or, in radical cases, supplant +stationary displays. -When selecting courses, there are a number of factors that must be taken -into account. First, be sure to choose courses that correspond to the level at -which you were admitted to BC. Your acceptance letter states if you were ac- -cepted as an undergraduate or graduate student. Second, be sure to select -courses that correspond to your level of experience with the subject matter. -Each course listing on the BC course web site gives the course level. Levels -for all BC courses are: + Granted, fixtures on castors are not a new concept, but the use of such shelving is not +necessarily widespread, especially in the center store area. An example of a retailer that lives by +mobile fixtures is one of Europe’s most successful shoe store chains, Leder & Schuh’s. L&S has +virtually no immoveable fixtures. Instead, every eight weeks or so the latest shoe fashions are - CORE: Fulfills university core requirements. Fairly basic courses open to +Madix offers a variety of caster kits that convert conventional shelving systems into flexible mobile runs. -undergraduate students from any field of study. +9 - ONE: Undergraduate elective courses open to students from any field of +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -study. + 10 Ways to Lift Center Store Sales - TWO: Undergraduate courses designed primarily for those specializing in +displayed in a new retail theater created by set designers using mobile fixtures, lighting and props. +This has helped cement their reputation as a progressive, forward-thinking retailer with its finger +on the pulse of the rapidly changing world of fashion. Other retailers, such as RadioShack, Ross +Dress for Less, BestBuy, and numerous department and grocery store chains, have experienced +their own successes with the flexibility of mobile fixtures. -that field of study. + Mobile fixtures give retailers the ability to easily change the store layout to fit the changing +needs of shoppers, accommodate special sales and offers, and even cater to “event” opportunities +with integrated solutions for big events, holidays or celebrations. Have perishables that are about +to go out of date? Put them on mobile fixtures and place them near checkout lines or in natural +product adjacencies to stimulate purchase and reduce shrinkage. Coming up on “Go Pink” breast +cancer awareness day? Use mobile fixtures and round up different pink merchandise such as +clothing and accessories near front entrances for people who want to support the cause. Have +overstock of something that might take weeks to clear out? Use mobile fixtures to create sales +opportunities throughout the store to get shoppers to purchase what you want to sell. The +possibilities are endless. - THREE: Upper-level courses for undergraduate and graduate registration. -Most suited to those with background knowledge of the general subject -matter and specializing in that field of study. +Solution #6: Shorten or break up long runs -11 + Hand in hand with banishing railroad tunnel aisles, Dr. Sorensen suggests rethinking the +concept of aisles altogether. He advocates that the length of shelving from one end to the other +be between 15-20 feet and not longer than 30 feet. -www.bc.edu/international +Shorter aisles in length and height accomplish multiple purposes. First, they are more +inviting to navigate and provide a natural break between product facings. Next, if the fixtures +are of varying heights, shoppers can actually see across multiple shelves and gondolas. Shorter, - GRAD: For those admitted as graduate students. Requires advanced knowl- -edge of general subject matter. Please note that some of these courses are -open to undergraduate students with the permission of the professor. +Madix’s shelving lines can be ordered to accommodate shorter run lengths. -Undergraduate students should select courses with a level of Core, One, Two, -or Three. Graduate students should select courses with a level of Grad or -Three. +10 -Course listings on the BC course web site (www.bc.edu/courses) look similar -to this: +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales + +staggered fixtures also serve as artificial road bumps that force shoppers to slow down, navigate +left, right and around product displays, and be more aware of what they’re passing. Also, one of +the most potent sales opportunities exist with end caps, and the shorter the fixtures in length, +and the greater the number of fixtures that are used, the number of end cap merchandising areas +increases exponentially. + + A similar outcome can be accomplished by interspersing concave or radius shelving that +softens up center store aisles and visibly pulls customers in. Rounded or organically-shaped +shelving units such as the Madix Y-Gondola can add a pleasant visual impact to the sales floor and +create multiple focal points allowing retailers to highlight important items, sales or promotional +offers. Such shelving also actually improves the visibility of more SKUs when compared to straight +shelves, because they are typically viewed from more of a frontal perspective than through a side +view while walking down an aisle. + + Of note, some retailers, such as Walmart, have actually reverted back to taller, warehouse-style +shelving and what has been described as shelves stacked full of many different products, narrow +aisles and conspicuous pallet stacks. Why? Because their sales results support the notion that +such a shopping environment better supports their brand proposition of being a value-oriented +retailer. + +Explaining this perspective in a national best seller book entitled Why We Buy: The Science of +Shopping, Underhill says, “Historically, the more a store is packed (with merchandise), the more +people think of it as value — just as when you walk into a store and there are fewer things on the + +floor, you tend to think they’re expensive.” + +This simply means that it’s critical for retailers such as Walmart to understand the mindset +of their customers to know what it is they want. Regardless of a retailer’s brand proposition – +whether it be value, quality, selection or brand names – they must strike a balance between the +volume of products displayed and the physical and visual presentation to create an optimum +buying atmosphere for its customers. In other words, what works for a discounter may not work +for a specialty retailer, and vice versa. + +Solution #5: Integrate specialty merchandisers + +Specialty merchandisers are an ideal mobile way for retailers to experiment with product +displays and presentations on the sales floor without making major, permanent commitments to +space or marketing. + +Retailers can experiment with specialty “solution stations” that tie product categories together + +without totally realigning product groupings or fixtures. Common fixtures for such specialty +presentations include the use of T-Racks and mobile caster kits. + + Grocers could consider staging certain ethnic food nights, weeks or months to stimulate trial +and impulse purchases by shoppers. For example, a mobile fixture or end cap could be ideal for +“Italian Night Solutions” complete with recipes and suggestions for combinations of products that +otherwise would be spread throughout the store: pots, pans, utensils, spaghetti, lasagna, pizza +kits, sauces, spices, olive oil, cheese, meat and vegetable appetizers, bread, desert, wine and even +flowers. Another possibility could be “Pamper Your Pet” with a selection of high-end food, treats, +supplements, collars, bowls and toys. Evergreen events such as birthdays and anniversaries also +lend themselves to turn-key specialty merchandisers featuring apparel, gift boxes, cards, candles, +desserts and even alcohol. -Index +11 -Course # +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -Credits + 10 Ways to Lift Center Store Sales -Level +Madix manufactures specialty merchandisers to suit any market segment. Luxe, +a new display that features an attractive round upright and standard Madix slotting, +accepts Madix’s thousands of shelves and accessories. -Title +Retailers should experiment with integrated presentations for different topics and locations -Instructor +to see what does and doesn’t resonate with their customer base and tweak them accordingly. +In addition, Dr. Sorensen points out that something as simple as scattering static or mobile +merchandisers around the store in non-obvious locations almost always increases sales of those +products. For example, placing a merchandising display of cookies outside of the bakery, or +t-shirts and socks near the check-out queue, can stimulate incremental sales of those items outside +of their natural environment. The beauty of such displays is that they lend themselves +to experimentation. -Bldg +Solution #4: Make smart use of in-store marketing -Room +Point-of-sale marketing is found to improve the find-ability of merchandise and usually leads to +improved brand recognition. However, it’s important to effectively judge the use of such point-of- +sale material and be smart about leveraging the sales of merchandise the retailer wants to sell, not +necessarily what the manufacturer or supplier wants to push. After all, why take up valuable shelf +space and fight for shoppers’ attention for a low-margin or low velocity item or product category? +It’s better to focus on merchandise that has a higher profit margin, introduces new or innovative +products to customers, or supports the retailer’s particular point-of-difference (better selection, +higher quality, well-known brand names, etc.). -Time + Digital print and video signage are excellent ways to cut through the clutter and grab shoppers’ +attention in the noise of the typical retail environment. Such signage can be designed to educate, +entertain and engage shoppers with key messages they might not otherwise see. The Bissell +brand of vacuum cleaners in Lowe’s Home Improvement Stores is one example. Their use of digital +and video signage on end caps and mid-aisle displays stop shoppers in their tracks to deliver short, +concise sales messages about the benefits of clean floors and carpets and how Bissell products +deliver the best outcome and value. -9398 +12 -BI200 01 +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -0569 + 10 Ways to Lift Center Store Sales -HS181 03 + Other, less extreme example of digital signage and even low-tech coupon and recipe pads and +dispensers are also effective in garnering attention for specific products and brands. Products +such as Lumicanopy by Madix, Inc. use LED lighting to illuminate graphics and promote categories +or brands. The typical Lumicanopy consumes little power, emits virtually no heat, exhibits uniform +light distribution that’s visible from many feet away, and is reusable over several years. Other +displays such as the Madix Easel SB retail shelf and point-of-sale display educates and entertains +shoppers with relevant information and does not protrude into aisles to interfere with shoppers. +The key is to be smart about what and how information is communicated to ensure its effect on +shoppers is maximized. -3 +Being able to deploy large format, illuminated graphics either in Madix’s canopies or within +the fixtures’ surfaces facilitates in-store marketing initiatives. The bright, colorful graphics +appeal to shoppers across the store. -3 +Solution #3: Turn up the lights -TWO +Lighting is one of the most important -- yet probably one of the least understood -- -Biology +merchandising techniques in the retail environment. Done correctly at the point-of-sale with +fixtures such as the Madix LEDge Light Versa and the Beam LED alternative to fluorescent canopy +lighting and retailers can improve mood, create the desired ambiance, and more importantly +create a measurable impact on shopping behavior. -DEPT + An article published in InformeDesign by Delores (Dee) Ginthner, an associate professor in +the Interior Design Program, Department of Design, Housing and Apparel at the University of +Minnesota, states several basic truisms that must be addressed when lighting a space to ensure +both psychological and behavioral issues of users are met. This includes the observations that +lighting must ensure safe navigation through a space; people always follow the brightest path; +brightness can focus attention; people like to face walls that are illuminated; and lighting can +cause people to alter body positions. -ONE +13 -American -Civilization +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales -Singleton + Ginthner further states that criteria created several years ago by Dr. John Flynn to evaluate +lighting remain true today. This criteria states that designers must use cues to determine the users’ +subjective response to space lighting in order to create space that makes users say something is +pleasant versus unpleasant, public versus private, spacious versus confined, relaxed versus tense +and visually clear versus hazy. -MWF -11 -TTH -9* - -In these examples, the index is the number used to register for the course -(you must include this number on your Course Registration Form). The -course number includes a two-letter abbreviation for the department (e.g. BI -= Biology, HS = History), followed by the three-digit course number. The two -numbers following the course number indicate the section. Popular courses -such as HS181 are offered in different sections to keep class sizes small. In -the example above, the course selected is section 3 of HS181. Each of the -courses in these examples carries a 3-credit load, which is the most common -load for university courses. It is your responsibility to make sure that you -have had all prerequisites required for any course that you are requesting. - -Courses typically meet several times a week. Classes that meet Monday, -Wednesday and Friday (MWF) are generally 50 minutes long, while Tuesday, -Thursday (TTH) classes are typically 75 minutes. These classes are indicated -in the course listing with an asterisk (*). - -Some courses may require co-registration for a supplementary lab or discus- -sion section and this will be stated in the course listing. Enrollment in some -courses may be restricted to majors only or may require department permis- -sion prior to registration. We are not able to pre-register you for restricted -courses, but you will be able to appeal to the department or professor after -you arrive to ask for special permission to enroll in such a class. Likewise, -undergraduates who seek to enroll in graduate-level courses will need to seek -direct permission from the instructor after arrival in Boston. It is fairly un- -usual for undergraduates to obtain approval for graduate courses and you -should not count on being able to do so. - -www.bc.edu/international + Needless to say, the science of lighting and how it applies to curved fixtures, pyramid or +recessed shelving, wider and shorter aisles, end caps, on-shelf merchandisers and even mobile +fixtures is critical to maximizing the impact and desired outcomes. The days of one-size-fits-all +overhead or fluorescent canopy lighting are being replaced by new efficient alternatives such +as environmentally friendly LEDs that produce light intensity similar to T5 fluorescent fixtures +while using only half the energy and lasting nearly three times longer. Lighting considerations +should be incorporated into the initial store design at the onset by true lighting professionals who +understand shopper psychology and how best to use lighting to stimulate sales. -12 +Madix is the first store fixture manufacturer to design energy-efficient, +point-of-sale lighting products that are specifically designed for store equipment +applications. Madix’s lighting solutions either attach underneath shelves, or they +replace the fixtures’ surfaces to pin-point light (e.g., back panels or shelf skins). - Finally, please note that many courses “close” during the early registration -period. Each course is listed in the computer with a maximum number of stu- -dents who can enroll. Once that limit is reached, the computer will not allow -any additional registration in that course. If you find that a course you need is -closed, you will have to appeal to the professor (either by e-mail prior to your -arrival or after you arrive) for an “override” to allow you into the class. This is -not uncommon and most professors are sympathetic to reasonable requests -from exchange students. However, it is not guaranteed that you will receive -an override for a closed course. - -Please look for classes offered in the semester in which you will be attending -(e.g., Fall, Spring) even if the list is for the previous year. For the most current -listing of available courses, see www.bc.edu/courses. - -Drop/Add - -The first week of classes at the beginning of each semester is called the -“Drop/Add” period. At this time you can make changes to your course sched- -ule, adding and/or dropping courses as needed. Don’t worry if you do not -have all of the courses that you need by the time classes start. You will have -this period of time to make adjustments to your course schedule, seek over- -rides as needed, etc. - -Your OIP exchange advisor can help you with most registration needs, includ- -ing add/drop. Registration matters can also be handled at Student Services in -Lyons Hall or on the computer via AGORA. You will receive more specific in- -formation about these resources upon arrival. In the meantime, refer all -course requests to OIP. - -Faculty Advisors - -You may request a faculty advisor for -assistance with academic planning for -your semester or year at BC. We will -make advisor placements based upon -your field of study. If your advisor at -home already knows a particular pro- -fessor at BC, or if you have already -made contact with a professor here -and you would like him/her to be your -advisor, please let us know. If you re- -quest a faculty advisor, you will be -required to meet with them by ap- -pointment during the first week of -classes and throughout the semester -as necessary. +Solution #2: Optimize traffic flow -13 +The art of converting visitors into shoppers and shoppers into buyers is accomplished in no -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -www.bc.edu/international - - Tutoring - -The Connors Family Learning Center offers free tutoring for more than 60 -undergraduate and some graduate courses, including writing and English -as a Second Language. Intensive, small-group supplemental instruction -sessions are available in several courses, and tutoring and assistance in -improving academic writing. For more information about services provided -by the Learning Center, go to www.bc.edu/libraries/centers/connors/. - -English Classes for Non-Native Speakers - -If your native language is not English you will be required to take a place- -ment examination during the Orientation in the fall or during the first week -of classes in January. If necessary, students will be recommended for a par- -ticular course based upon the results of this examination. The Program in -English for Foreign Students is based in the English Department and offers -several courses in college-level English for non-native speakers. Recent -course offerings include: - - EN 118 Essential of English Composition: extensive practice in writing - -and editing academic prose - - EN119 The Craft of Writing: practice writing academic essays and expo- -sure to English rhetoric, with emphasis on written analysis and the logi- -cal support of ideas - -In order to enroll in one of these courses, you must have taken the place- -ment test. - -Transcripts of Academic Work - -At the end of your time at Boston College, we will send an official transcript -of the academic work you completed to your home university. Before we -can release your BC transcript to your home university, however, you must -clear any outstanding debts to Boston College that you may have. Be sure -that all long-distance phone calls, on-campus housing fees, insurance, -meals, etc. have been paid for. - -Your BC student account must show a zero balance before you leave Bos- -ton or we will not be able to release your transcript. - -www.bc.edu/international +small part through optimizing traffic flow through the store. Dr. Sorensen suggests there are +several changes, some subtle and some overt, that can make a major impact on a store’s visceral +appeal while exposing shoppers to buying opportunities that may have gone undiscovered in a +traditional linear layout. 14 - ADMINISTRATIVE ISSUES - -Student Identification Cards - -If you submitted the required passport-size photographs with your Exchange -Student Application you will receive your student ID card during orientation. -If you did not send us photos, you will be able to have your ID card made at -Student Services in Lyons Hall following orientation. - -Your Boston College student identification card (“Eagle ID”) is a very impor- -tant item for you to have. In addition to identifying you as a BC student, the -card enables you to check books out from the library, to make purchases at -the BC bookstore, to access your student record and account from “U-VIEW” -computers around campus, and to make use of the BC Flynn Recreation -Complex (fee required), among other services. Be sure to keep your Eagle ID -with you at all times and do not lose it. A replacement fee will be charged by -Student Services for lost cards. - -Medical Insurance - -As explained in the Things to Know Before You Leave Home section, the -Commonwealth of Massachusetts requires all students enrolled in a Massa- -chusetts college or university to be covered by a US-based medical insurance -plan. Therefore, you have been automatically enrolled in the Boston College -medical insurance policy which provides adequate coverage during your stay -at BC, including travel within and outside the US during your period of cover- -age. Your student account was billed upon registration at BC. As an exchange -student it is your responsibility to pay for your insurance, which can be done -at Student Services. Full year students will be automatically re-enrolled in the -insurance plan for their second semester. You can find more information on -the BC medical insurance coverage on the BC web site at www.bc.edu/ -offices/uhs/services/insurance. - -If you do not want to be covered by BC medical insurance, you must be cov- -ered by insurance that is US-based or has a US contact office. You must have -proof of this insurance. If your insurance is comparable, you will be allowed -to waive the BC insurance and your student account will be credited accord- -ingly. Please note that it is not possible to have your medical insurance evalu- -ated before you arrive at Boston College. +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales + + One idea requires a significant reconfiguration of fixtures to change their orientation and +layout to the front of the store. Dr. Sorensen supports the theory that when shoppers within +the interior of a store look down aisles and see checkout registers, especially ones with no lines, +they subconsciously tend to hurry through their shopping experience so they can quickly check +out before the lines fill up: a sort of quick in, quick out mentality. To combat this and reduce the +“railroad tunnel” visual effect, he suggests shifting the aisles at a 45-degree angle to the front of +the store. An important benefit of such a design is that as shoppers enter the store they will be +approaching aisles from a less acute angle, which will make them more inviting to enter. + + Another way to increase traffic flow and lure shoppers is to determine new and relevant +product categories and bring them into the center store. Along the way and throughout the entire +store, the goal should always be putting the “right” products in the direct path of the customer. +Dr. Sorensen says, “There are patterns of movement... (in a store) that are not driven by products +as much as by open spaces and natural flows.” By observing, determining, and directing traffic +flow with smart shelving fixtures, lighting and merchandising materials, it’s possible to conduct +“anticipatory retailing.” This is when the retailer anticipates and meets the needs of shoppers +rather than making shoppers find things on their own. + +In support of an enhanced traffic flow, Dr. Sorensen advocates putting high margin + +merchandise rather than discounted goods in the customer’s path. Research shows that the most +important determinant for sales is based not on price, but on the placement of store fixtures +displaying those goods. Putting highly discounted merchandise such as 75-percent-off sweaters +or two-for-one cereal offers in the customer’s path may actually be counter-productive, because +it could prevent customers from venturing deeper into the center store and discovering higher- +priced, if not more desirable goods. + + Dollar General spokeswoman Mary Winn Gordon is quoted alongside Underhill in the New +York Times article as saying the chain store employs virtual “speed bumps” of related merchandise + +Madix’s array of fixture choices equips a retailer with options to tweak store designs to improve traffic flow and sales. 15 -www.bc.edu/international - - Health Services - -The Health Services Center at -Boston College offers a range -of services to BC students. -Staffed by first-class doctors, -nurse practitioners, and other -medical staff, the Health Ser- -vices Center is equipped to -handle health emergencies, -minor illnesses and questions -about healthy living. While the -general health services offered -are comprehensive, if you re- -quire specialty care you can -be referred to a qualified spe- -cialist. - -All undergraduate exchange -students who live on-campus -are required to pay the Health -Services fee. Exchange stu- -dents who live off-campus are -also entitled to this service for the same fee. Off-campus students can ei- -ther pay in advance, or the first time the service is used. We strongly recom- -mend that you utilize the Health Services if necessary. Services covered by -the fee are detailed on the Boston College Health Services website at -www.bc.edu/offices/uhs. - -Please note that Health Services fee is not the same as medical insurance. -You must still have adequate medical insurance coverage, especially for -such things as hospitalization, emergency room visits, diagnostic testing, -x-rays, etc., which are not covered by the Health Services fee. - -University Counseling Services - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sometimes students find adjusting to a new culture and new academic en- -vironment very challenging. The Counseling Services Office provides caring, -professional support and assistance to full-time students who request it. -Selected staff members in Counseling Services have extensive experience -with cross-cultural issues and are especially well-suited to help students -who are experiencing culture shock. It is very common in the United States -for students to seek the help of counseling professionals when dealing with -challenging situations. In fact, thousands of students take advantage of the -services offered by Counseling Services for a variety of issues each year. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. + 10 Ways to Lift Center Store Sales + +that goes together -- such as bananas alongside boxes of vanilla wafers -- in aisles or at entrances, +because it’s a consumer-friendly way to assist customers and pique their thought process while +generating incremental sales. By using such things as Y-Gondolas, mobile displays, LED lighting +fixtures, floor graphics, ceiling hangers or digital or video marketing to highlighting merchandise +in a non-intrusive, yet attention-getting fashion, it’s entirely possible to get customers to buy +items you wish to sell, not just what shoppers came into the store to purchase. + + An underutilized method of generating awareness and trial is product sampling and +demonstrations. Retailers such as Sam’s Club, Costco and Whole Foods have mastered the art +of sampling during high traffic days and peak times to expose customers to products they may +have otherwise ignored. Why not create a display that demonstrates the difference between +200 and 400 thread count sheets? Utilize sales staff as virtual models by having them wear the +merchandise being sold in the store, much like mega-brands Hollister and American Eagle have +done with their employees. Create mobile sampling stations offering customers tastes of lesser- +known foods such as couscous, Greek yogurt, even kale along with recipes incorporating those +foods to stimulate profitable impulse sales. + +In fact, the New England Journal of Medicine reports that food choices are largely guided by +our responses to outside signals and stimuli. A paper authored by Deborah Cohen of RAND Health +in Santa Monica, Calif., and Susan Babey of the UCLA Center for Health Policy Research, states that +impulse marketing takes place on a subconscious level, and that placement of foods in prominent +locations increases the rate at which they’re purchased. + + While most people may associate such impulse displays with checkout lines and sodas, candy, +gum and magazines, they can be just as effective in other areas of a store. Shopping for a new +blouse? You may be influenced by matching (high-margin) accessories attractively displayed +next to them. Buying new cat food? How about a pretty collar or catnip to go along with it. The +possibilities for proximity marketing are enormous, if not often overlooked, opportunities. + +Solution #1: Multiply the end cap effect + +End caps are perhaps the most misunderstood and underutilized spaces in the retail +environment. According to Dr. Sorensen, approximately 2-percent of the total items in a +typical store at any given time are being promoted on end-of-aisle displays or other secondary +promotional displays apart from the mail shelving fixtures. However, this 2-percent of items may +constitute a whopping 30-percent of all store sales. + + While retailers often use this space to promote sales or highly discounted merchandise, the fact +is that half the shoppers purchasing an item from such end caps are unaware the goods are being +sold at a reduced price, and half of the people who know would buy the product even if it wasn’t +on sale. + +So in other words, good retailers are simply giving away profits to as much as 75-percent of + +their shoppers because they are locked into the mindset that price considerations dominate +shopping decisions when research doesn’t support that point of view. Instead, great retailers +understand that a better use of such prime real estate is to satisfy the shopper’s inherent need +for value and convenience, and most customers are willing to pay full price to accomplish a fast +shopping experience. + + Another reason end caps can be extremely profitable is that because of their small size, they +generally feature only one or two products. As such, shoppers do not have an easy or immediate +way to compare the merchandise as they would in the category section inside the aisle. The +impulse nature of end caps and the lack of price sensitivity support the premise that retailers can +command a higher price for such merchandise. 16 -www.bc.edu/international - - The services offered by Counseling Services (at no additional cost) include -short-term individual counseling and psychotherapy as well as a limited -number of counseling groups. If you wish to speak with a counselor you -can make an appointment by calling 617-552-3310 or by visiting the Counsel- -ing Office in Gasson Hall 108. - -Also, please remember that OIP staff is always available and willing to help -if you have any questions, either academic or personal. Most of our staff -members have studied and/or lived outside the US and know firsthand the -challenges of adapting to a new country and culture. - -Students with Disabilities - -Boston College provides a range of services and accommodations for stu- -dents with physical or learning disabilities. If you require additional assis- -tance because of a disability, feel free to share that information with your -OIP exchange advisor. You may also contact the Boston College Coordina- -tor of Special Needs Services in Gasson Hall 108, telephone: 617-552-3310. - -E-mail - -Every student enrolled at -Boston College is eligible -to receive an e-mail ac- -count. You may obtain -your account after arriv- -ing at BC by presenting -your BC ID card at the -Campus Technology Re- -source Center (CTRC) -located on the second -floor of the O’Neill Li- -brary. The use of your BC -email account is required -once you have arrived on- -campus. If you have an- -other address that you -prefer to use, set up your BC account (by using AGORA) to forward all mes- -sages to the alternate account. OIP and other BC offices will send all infor- -mation during your time at Boston College to your BC email address. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com -17 + 10 Ways to Lift Center Store Sales -www.bc.edu/international - - LIFE BEYOND ACADEMICS - -Religion - -As a Catholic and Jesuit institution, Boston College is sensitive to the spiri- -tual needs of all members of the BC community, whatever their respective -faiths or philosophical traditions. Catholic Masses are offered throughout -the week in several locations on campus and are open to all BC students, -faculty and staff. The Campus Ministry Office offers spiritual counseling -and guidance to students, as well as volunteer service opportunities in Bos- -ton and other parts of the world. Campus Ministry also provides informa- -tion and support for people of all religious and spiritual traditions by main- -taining a list of churches, temples and synagogues in the Boston College -area. Various student clubs, including the Asian Christian Fellowship, Jew- -ish Students’ Association, and Muslim Students’ Association, provide a -sense of community and support for students of various spiritual tradi- -tions. - -Activities/Opportunities - -There is a wide variety of clubs and -activities to meet any student’s in- -terests, including several intercul- -tural clubs focusing on individual -countries or regions. An Activities -Fair early in the Fall Semester will -give you an opportunity to learn -about many of these opportunities -and speak to some of the students -involved. For students on exchange -during the Spring semester, the Of- -fice of the Dean for Student Devel- -opment has a list of all BC clubs -and activities (www.bc.edu/a-z/ -clubs/). Don’t pass up this opportu- -nity to learn about American culture -outside the classroom. Students -who get involved in activities at Boston College consistently enjoy their ex- -perience here more than those who do not get involved. For this reason, -we strongly encourage you to join at least one club while you are here. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +Breaking up and staggering the traditional long aisles in many retail locations can literally -18 +double the number of possible end caps from the shoppers perspective, particularly in the +center store. Incorporating other display tools such as Y-Gondolas, curved shelving, pyramid and +recessed shelving, highlight lighting, sampling stations, lifestyle graphics and signage, can only +serve to increase the sales velocity of end caps. -www.bc.edu/international - - s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sports in American University Culture - -Many international students are surprised at the level of importance placed -on intercollegiate sports in the United States. American football, particu- -larly, draws thousands of screaming fans every Saturday during football -season, with students cheering for the home team. This is very much a part -of American college culture and, even if you are not a sports fan, it is some- -thing you should experience during your time as a student in the United -States. You may not know anything about the intricate rules of American -sports like football, hockey and basketball. However, the experience of sit- -ting in a crowded stadium, cheering, listening to the band, watching the -crowd, and even trying to follow the game, is a true American experience -and should not be missed. - -Recreation Complex - -The Flynn Recreation Complex provides Boston College students with a -broad array of sports and physical fitness opportunities, from aerobics to -weight lifting and swimming. As an exchange student you have the option -of joining the “RecPlex” for a semester or academic year (membership fees -apply). If you choose not to join the “RecPlex” you may still use the facilities -by presenting your BC ID card and paying a small fee per visit. +Madix understands how important end caps are to a retailer’s success. The company manufactures a number +of end cap systems that maximize space, improve visibility, and capitalize on sales. Madix’s new Y-Gondola +greatly expands the potential of end caps to positively influence shopping behavior. -19 - -www.bc.edu/international - - CONTACT INFORMATION - -Office of International Programs - -Location: -Mailing Address: +Decide what solutions make sense for you -Hovey House, 258 Hammond Street -Hovey House, 140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +Revisiting the relationship between the perimeter and center store requires that retailers -Phone: 617/552-3827 Fax: 617/552-0647 -Email: oipexchange@bc.edu -Web: www.bc.edu/international -Director: Bernd Widdig -Email: widdig@bc.edu -Assistant Director for Exchange Students: Esther Messing -Email: messines@bc.edu +rethink the cost of sales. Ignoring or shrinking the center store given its contribution to +overall profits doesn’t make sense when much more is possible. A better course of action is to +fundamentally rethink the consumer shopping process and adapt the entire store to better deliver +total customer solutions. After all, what good is the sizzle without a juicy steak to go along with it? -Office of International Students and Scholars +The display fixture experts at Madix, Inc. are available to provide free consultation to retailers -Location: -Mailing Address: +who wish to explore fresh, new ideas on how to bring the center store back to life to create a +holistic shopping experience that drives sales and profits. For more information, contact John +Clontz Director of Marketing and e-Business jclontz@madixinc.com 214-515-5400. -21 Campanella Way - Suite 249 -140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 - -Phone: 617/552-8005 Fax: 617/552-2190 -Web: www.bc.edu/offices/odsd/intercultural -Director: Adrienne Nussbaum -Email: adrienne.nussbaum@bc.edu - -Office of Residential Life +17 -Contact: -Web: +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. -On-Campus Housing, Off-Campus Housing Advice -www.bc.edu/offices/reslife + 10 Ways to Lift Center Store Sales -Office of Student Services +Traditional Center Store Fixture Configuration -Contact: -Web: + Traditional center store fixture runs are usually long and extend 30 feet or +more. The long runs discourage shoppers from entering the department. Also, +the lengthy runs have a limited quantity of end caps, the store fixture with the +most effective sales potential. -Billing, Medical Insurance, ID, Course Registration -www.bc.edu/offices/stserv +18 -University Health Services +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com 10 Ways to Lift Center Store Sales -Contact: -Web: +Center Store Layout that Staggers Runs + A center store layout with staggered runs includes more and shorter fixture +runs. Shoppers are more likely to enter a center store that utilizes shorter, +less daunting runs. The staggering of the runs greatly increases a shopper’s +exposure to end caps as he or she walks through the department. -Immunization, Health and Wellness Issues -www.bc.edu/offices/uhs +19 -University Counseling Services +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. 10 Ways to Lift Center Store Sales -Contact: -Web: +Center Store Layout with Specialty Merchandisers and Transitional Components -Personal Counseling -www.bc.edu/offices/counseling +Integrating specialty merchandisers into a center store will improve +incremental sales. Utilize transitional components, including bases and +shelves, to draw attention to merchandise within merchandising categories. +The transitional components and shelves also make room for shoppers to +navigate around the specialty displays between the aisles. 20 -www.bc.edu/international - - ESTIMATED EXPENSES -2010-2011 - - Average Cost of Apartment/Room Rental - -Studio Apartment: $1000 - $1300 per month * -One Bedroom Apartment: $1200 - $1550 per month * -Two Bedroom Apartment: $1400 - $2000 per month * -Three-Four Bedroom Apartment: $2000 - $3800 per month * -Private Room in Local Home: $500 - $900 per month * - -*Please note: Boston rental agents often charge one half or one full month's -rent as a commission fee (also known as a "Finder's Fee") - -On-Campus Housing - -Shared Room: $1100 per person per month - -Meals and Books - -Campus Meal Plan: $2300 per semester for full meal plan -No Meal Plan: $1500 per semester -Books: $600 per semester +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com + 10 Ways to Lift Center Store Sales -Other Expenses +Center Store Layout using Y-Gondolas as End Caps -Medical Insurance: $871 (Fall semester); $1027 (Spring semester) ** -University Health Services: $210 per semester ** -Flynn Recreation Complex: $140 per semester + Madix’s Y-Gondola exponentially expands the sales potential of end caps. +Not only does the fixture greatly increase the end cap’s merchandise volume, +but it also makes the merchandise on the end cap visible from wider angles. A +store layout that utilizes Y-Gondolas on short fixture runs will create an inviting +and more shopable center store. -**The BC Medical Insurance plan is mandatory unless you can demonstrate -coverage by an insurer with a U.S. office. The University Health Services fee is -required for on-campus students. +21 -Personal Expenses: These will vary significantly according to the individ- -ual, but students should budget approximately $2000 per semester. +© 2014 Madix, Inc. All Rights Reserved. Prices and specifications subject to change without notice. Not responsible for errors or omissions in printing. 10 Ways to Lift Center Store Sales -Transportation from Logan Airport to BC +22 -Taxi: approximately $60 -Private Airport Shuttle: approximately $50 -Subway (“T”) Ride : $2.00 cash payment or $1.70 using a “Charlie Card” +Texas: 800-776-2349 Fax: 972-563-0792 Alabama: 800-633-6282 Fax: 256-839-5608Mexico: 001-800-010-0699 www.madixinc.com www.madixinc.com +Texas 500 Airport Road, Terrell, TX 75160 | 800.776.2349 +Alabama 20984 Alabama Hwy 9, Goodwater, AL 35072 | 800.633.6282 +©2014 Madix, Inc. All rights reserved. Not responsible for errors and omissions in printing. Rev. 7/14 diff --git a/q119/random_k4/random_2.md b/q119/random_k4/random_2.md index 4b9bd14ed63d1882c240fd2561e1be123d06b2a6..310f0f860924656c9ab9a27398fe83594f525660 100644 --- a/q119/random_k4/random_2.md +++ b/q119/random_k4/random_2.md @@ -1,1483 +1,1527 @@ -USER MANUAL +Department of the Treasury +Departmental Offices -8-IN-1 ONE-TOUCH -AUTOMATIC ESPRESSO MAKER +FY 2025 -GCM4230 +Capital Investment Plan 1 -USER MANUAL ESPRESSO MAKER GCM4230 - CONTENTS + Standard IT Investments ...................................................................................................... 3 -Important Safeguards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 +Treasury Shared Services Security Operations Center (TSSSOC) .............................................. 3 -Know Your Espresso Maker . . . . . . . . . . . . . . . . . . . . . . . . . . 6 -Control Panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 +Major IT Investments .......................................................................................................... 4 -Using Your Espresso Maker +Cybersecurity Enhancement Account (CEA) - Cloud Enterprise ................................................ 4 -Before First Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 +Cybersecurity Enhancement Account (CEA) - Multi-Factor Authentication .............................. 6 -Setting Up and Priming the Espresso Maker . . . . . . . . . . . . . . . 8 +Cybersecurity Enhancement Account (CEA) - Other Cyber Priorities ....................................... 7 -Filling the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 +Cybersecurity Enhancement Account (CEA) - Security Logging ............................................... 9 -Placing the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 +Cybersecurity Enhancement Account (CEA) - Universal Encryption ...................................... 11 -Mug Size Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +Cybersecurity Enhancement Account (CEA) - Zero Trust Architecture ................................... 12 -Drink Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +HR LoB - HRConnect ................................................................................................................ 14 -Making an Espresso . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 +Treasury Enterprise Identity, Credential and Access Management (TEICAM) ........................ 16 -Making a Cappuccino . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 +Major Non-IT Investments ................................................................................................ 19 -Making a Latte . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 +Main Treasury Building and Freedman's Bank Building .......................................................... 19 -Making an Americano . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 +2 -Brewing in Manual Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 15 + Note to Reviewers: Consistent with the corresponding Summary of Capital Investments table, the +columns included in the investment tables below are defined as: -Frothing Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 +FY 2023 Actuals -Total actual obligations. +FY 2024 Estimated Obligations- Anticipated obligations from all budgetary resources (e.g., balances +from prior years, user fees, and FY 2023 Operating levels). +FY 2025 Estimated Obligations - Anticipated obligations from all budgetary resources (e.g., balances +from prior years, user fees, and FY 2025 President’s budget). -Cleaning and Maintenance +Standard IT Investments -Cleaning the Portafilter and Brew Head . . . . . . . . . . . . . . . . . 17 +Treasury Shared Services Security Operations Center (TSSSOC) -Cleaning the Water Reservoir . . . . . . . . . . . . . . . . . . . . . . . 17 +Description: +TSSSOC functions as the Department wide security operations center providing security monitoring, +incident coordination, incident response, and reporting. TSSSOC operates 24x7x365 and actively +coordinates with other federal agencies, DoD, and the IC. -Cleaning the Milk Reservoir and Frothing Tube . . . . . . . . . . . . . 17 +Investment Obligations: (In Millions of $): -Descaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 +T y p e -Warranty and Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -Model: GCM4230 | © 2023 Gourmia | 2.0 -The Steelstone Group Brooklyn, NY | Gourmia .com +F Y 2 0 2 3 +A c t u a l s -All rights reserved. No portion of this manual may be reproduced by any means whatsoever without written -permission from The Steelstone Group, except for the inclusion of quotations in a media review. Although -every precaution has been taken in the preparation of this manual, The Steelstone Group assumes no -responsibility for errors or omissions. Nor is any liability assumed for damages resulting from the use of -the information contained herein. Although every effort has been made to ensure that the information -contained in this manual is complete and accurate, The Steelstone Group is not engaged in rendering -professional advice or services to the individual consumer; further, The Steelstone Group shall not be held -responsible for any loss or damage allegedly arising from any information or suggestions in this manual. +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s - Congratulations on the -purchase of your Espresso -Maker from Gourmia! +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Here at Gourmia, our one and only goal is to make sure -nutritious and delicious meals are always a practical -possibility no matter how fast-paced your life is . Whether -you’re in need of a hearty breakfast, a mid-day snack, or a -satisfying supper, our dependable, easy-to-use appliances -are designed to work for you while you work your way -through your busy day . +C h a n g e i n +$ -This manual provides an overview and instructions on -all the espresso maker’s functions and features, along -with directions for assembling, operating, cleaning, -and maintaining your appliance . Please read all safety -instructions before getting started . Keep this manual handy -for future use and review . +% C h a n g e -Our team is available to help with any and every question -you might have about your new espresso maker . We’re -excited for you and want you to get the most out of your -purchase—so whether you need assistance with setup or -guidance on which features to enable for which drinks, -we’re happy to walk you through it, step by step . +2 . 2 6 -Happy and easy cooking! +1 . 0 9 -We’d love to hear about the delicious drinks you’re creating! -Share your thoughts, recipes, and photos with us. +1 . 0 9 -@gourmia +0 -3 +0% + +1 5 . 4 4 + +1 7 . 7 0 + +1 4 . 75 - S -d -r -a -u -G -E -f -a -S +1 5 . 8 4 -t -n -a -t -r -o -p -M +1 5 . 1 9 -I +1 6 . 2 8 -S A V E T H E S E I N S T R U C T I O N S +0 . 4 4 -IMPORTANT SAFEGUARDS +0 . 4 4 -Before operating this espresso maker, read and follow all these -basic safety precautions: +2.98% -1. The appliance can be used +2.78% -7. do not operate the appliance +Purpose, Accomplishments, Future Objectives: + +The Treasury Shared Services Security Operations Center (TSSSOC) provides leadership and oversight +for the offices and bureaus across Treasury in all aspects of cybersecurity operations. The program is +responsible for ensuring implementation of the Federal Information Security Modernization Act (FISMA) +of 2014 as well as other federal laws and guidance related to cybersecurity operations. The program also +provides central coordination and reporting on cybersecurity operational metrics and programs to external +agencies and sets Treasury cybersecurity operations standards and guidance. TSSSOC also provides +departmental situational awareness of cybersecurity incidents and coordinates response to intrusion +activity. The program also serves as a lead in implementing the Treasury Cybersecurity Strategy. + +Investments have contributed to Treasury's success in defending against cyber adversaries and provides +leadership and direction in mitigating threats to Treasury systems and services. Future objectives include: +collecting and storing data from all bureaus to meet M-21-31 objectives, expanding TSSSOC to offer +additional security operations and incident response services to bureaus, and improving communication +and collaboration throughout the Department through automation. + +3 -by children aged 8 years and -above and persons with reduced -physical, sensory or mental -capabilities or lack of experience -and knowledge if they have been -given supervision or instruction -concerning use of the appliance -in a safe way and understand the -hazards involved . Children should -not play with the appliance . + Major IT Investments -Cleaning and user maintenance -shall not be done by children -unless they are 8 or older and -supervised . +Cybersecurity Enhancement Account (CEA) - Cloud Enterprise -2. Keep the appliance and its cord +Description: -out of reach of children less than 8 -years old . +Treasury aims to improve, entice, and further accelerate enterprise-wide cloud adoption through +investment in the Treasury enterprise shared services cloud environments. The Department’s focus on +cloud security enhancement and upgraded capabilities to meet security and compliance risks will +involve investment in security controls, monitoring, and increased threat protections by providing a +common cloud operating platform for Departmental workloads, inclusive of all security elements +required by Executive Order 14028 and subsidiary guidance. Treasury’s recently awarded Blanket +Purchase Agreement (BPA) for cloud services (Treasury Cloud [TCloud]) will support our design, +development, and implementation of security patterns and guardrails with expanded Security +Operations Center (SOC) capabilities to accommodate increased telemetry from cloud assets and +workloads. -3. Make sure your outlet voltage +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -corresponds to the voltage stated -on the rating label of the espresso -maker . +Investment Obligations: (In Millions of $): -4. To protect against fire, electric +T y p e -shock, or injury to persons do not -immerse cord, plug, or appliance -in water or other liquid . +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -5. Close supervision is necessary +F Y 2 0 2 3 +A c t u a l s -when this appliance is used by or -near children . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -6. Unplug from outlet when not in +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -use and before cleaning . Allow the -appliance to cool before putting -on or taking off parts, and before -cleaning the appliance . +C h a n g e i n +$ -with a damaged cord or plug or if -the appliance malfunctions or has -been damaged in any manner . -Return appliance to an authorized -service facility for examination, -repair, or adjustment . +% C h a n g e -8. The use of accessories not +2 0 . 0 7 -recommended by the appliance -manufacturer may result in fire, -electric shock, or injury to persons . +2 5 . 3 3 -9. do not touch hot surfaces; use +1 5 . 8 0 -handles or knobs . +- 9 . 5 3 -10. do not operate this appliance -outdoors . do not use for -commercial purposes . +- 3 7 . 6 1 % -11. do not let the cord hang over +0 . 6 5 -the edge of a table or counter or -touch any hot surfaces . +1 . 6 0 -12. Use on a hard, flat, level, heat- +9 . 6 3 -resistant surface only . +8 . 0 3 -13. Always be sure to secure the +5 0 1 . 5 6 % -water reservoir tightly back inside -the unit with the lid closed and -filled with water between the -minimum and maximum mark -before turning the appliance on . +2 0 . 7 2 -14. do not remove the portafilter while -brewing coffee . Exercise caution -when removing the portafilter after -brewing since the metal parts -will be hot . Make sure to hold the -portafilter by the handle and use - the filter retention clip to dispose -the grounds . +2 6 . 9 3 + +2 5 . 4 3 + +- 1 . 5 0 + +- 5 . 5 7 % + +Purpose, Accomplishments, Future Objectives: + +The CEA cloud investment focuses on an enterprise-wide investment approach to build stronger +protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the CEA Cloud Enterprise investment, Treasury aims to improve, entice, and further +accelerate enterprise-wide cloud adoption through investment in the Treasury enterprise shared +services cloud environments. The Department’s focus on cloud security enhancement and upgraded +capabilities to meet security and compliance risks involves investment in security controls, monitoring, +and increased threat protections by providing a common cloud operating platform for Departmental +workloads, inclusive of all security elements required by Executive Order 14028 and subsidiary +guidance. 4 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - 15. Extreme caution must be -exercised when moving -appliances with hot liquids . + The FY 2025 President's Budget includes $25.4 million for Treasury’s CEA Cloud Enterprise +investment with alignment to the NIST Cybersecurity Framework priorities: + +• -16. Only remove the water reservoir -when the appliance is turned off . +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an -17. This appliance is not intended -to be operated by means of an -external timer or separate remote- -control system . +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -appliance and allow it to cool -down . The blockage may be -caused by the coffee being too -finely ground . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -20. do not place the appliance on or -near a hot gas or electric burner, -or in a heated oven . +appropriate safeguards to ensure delivery of critical services. -21. do not use the appliance for other +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -than its intended use . +occurrence of a cybersecurity event. -18. do not remove the portafilter +• Respond to Detected Cybersecurity Incidents Goal: Develop and implement appropriate -at any time during the brewing -process . +activities to take action regarding a detected cybersecurity incident. -19. If coffee does not drip, it means -the filter is blocked . Turn off the +In FY2025, CEA – Cloud Enterprise investments are targeted to the Department’s cloud focused +modernization and the intrinsic enhancement of our cybersecurity posture. Our investments will build +on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. -22. do not leave the appliance -unattended when turned on . +• Treasury seeks to support the following initiatives with final FY2025 appropriations: +• Bureau of the Fiscal Service cloud cybersecurity modernization and the necessary human -23. This appliance should only be -used in households or similar -environments . +capital to support initiatives. -S -d -r -a -u -G -E -f -a -S +• Cloud security resources (engineers, ISSM, etc.) to support the FY2023 awarded enterprise -t -n -a -t -r -o -p -M +TCloud contract and program. -I +• Application migrations from on-premises to cloud environments and/or from unique Bureau -Cord and plug Safety -1. A short power-supply cord is provided to reduce risk of becoming entangled +environments to the enterprise cloud platform. -in or tripping over a longer cord . +• -2. Extension cords may be used if care is exercised in their use . +Incident response infrastructure to support coordination and communication in an -3. +uncompromised environment for response and recovery activities. -If an extension cord is used: +• Migration of National Security System infrastructure to the cloud. +• Operations, maintenance, and ongoing investment in a sustainable low code cloud platforms +(i.e., ServiceNow, Salesforce) to support organizational capabilities, to include +Cybersecurity Governance, Risk, and Compliance; CEA Administration and Governance; +and Security. -• The marked electrical rating of the extension cord should be at least equal +• Storage support for enterprise logging collection and utilization of historic logs to support -to the electrical rating of the appliance . +threat hunt and threat detection activities, as well as additional behavioral analytics. +• Expansion of the Department’s enterprise offering for public-facing, cloud-based web hosting -• The longer cord should be arranged so that it will not drape over the +that includes an Akamai CDN with robust DoS and bot protections, a Drupal content +management system, and additional centralized site shield, web performance and +readiness/response investments that integrate with Treasury’s security operations +capabilities for threat detection and response. -countertop or tabletop where it can be pulled on by children or tripped -over . +Looking forward, CEA investments will aim to accomplish the below future objectives: -Electrical polarized plug -This unit has a polarized plug (one blade is wider than the other) . -To reduce the risk of electric shock, the plug is intended to be -used by inserting it into the electrical outlet only one way . If the -plug does not fit securely into the outlet, reverse the plug and -re-insert . If it still does not fit, contact a qualified electrician for -assistance . Do not modify the plug in any way . +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -poWEr +threats. -AC120V~ 60Hz 1,300W +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. 5 -USER MANUAL ESPRESSO MAKER GCM4230 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K - -KNOW YOUR -ESPRESSO MAKER - -Water Reservoir -Cover - -Removable -Water Reservoir - -Control Panel - -Brew Head - -Power Switch - -On the back-right -side of the unit - -Removable Drip Tray + Cybersecurity Enhancement Account (CEA) - Multi-Factor Authentication + +Description: + +Treasury’s investment in Multi-Factor Authentication (MFA) supports our development and +implementation of compliant technologies for internal systems and applications. This investment is a +critical component of the Department’s solutions to prevent unauthorized access by adversaries, +including nation state actors. MFA is a layered approach to security access to network or applications +by certifying identity with a combination of two or more approved pieces of evidence (or factors) to an +authentication mechanism. The Department’s investment in, and sustainment of, MFA capabilities +include strengthening MFA governance, ongoing investment in enterprise solutions to boost adoption, +and strategic partnerships to streamline MFA related procurement effort for all Treasury Bureaus and +Offices. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. + +Investment Obligations: (In Millions of $): + +T y p e + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) + +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +0 . 0 0 + +0 . 0 0 + +0 . 0 0 + +5 . 1 0 + +0 . 6 0 + +5 . 7 0 + +0 . 0 0 + +- 5 . 1 0 + +- 1 0 0 . 0 0 % + +2 . 5 0 + +2 . 5 0 + +1 . 9 0 + +3 1 6 . 6 7 % + +- 3 . 2 0 + +- 5 6 . 1 4 % + +Purpose, Accomplishments, Future Objectives: + +The CEA multi-factor authentication (MFA) investment focuses on an enterprise-wide investment +approach to build stronger protections and further benefit the bureaus in their cybersecurity goals. With +the creation of a centralized appropriation, administration, and governance, Treasury has been able to +develop and enhance Department-wide cloud services, support Bureau-level modernization goals, and +drive a more robust coordination of efforts to improve the Department’s ability to detect, identify, +protect, recover, and respond to cybersecurity events. + +Through the CEA Multi-Factor Authentication investment, Treasury aims to develop and implement +compliant technologies for internal systems and applications. This investment is a critical component +of the Department’s solutions to prevent unauthorized access by adversaries, including nation state +actors. MFA is a layered approach to security access to network or applications by certifying identity +with a combination of two or more approved pieces of evidence (or factors) to an authentication +mechanism. The Department’s investment in, and sustainment of, MFA capabilities include +strengthening MFA governance, ongoing investment in enterprise solutions to boost adoption of MFA, +and strategic partnerships with industry partners to streamline MFA related procurement effort for all +Treasury Bureaus and Offices. + +The FY 2025 President's Budget includes $2.5 million for Treasury’s CEA Multi-Factor +Authentication investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. 6 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 + In FY2025, CEA – Multi-Factor Authentication investment is targeted to the Department’s MFA- +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. - Milk Reservoir Lid +Treasury is looking to support the following initiatives with final FY2025 appropriations: +Conversion to Active Directory (AD) Federation that provides for the authorization, authentication, +and SSO functionality to applications and services virtually anywhere, including perimeter networks, +partner organizations, and the cloud. -Froth Strength Control +• Looking forward, CEA investments will aim to accomplish the below future objectives: +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -Froth Dispensing Tube Lever +threats. -Froth Dispensing Tube +• Provide leadership with greater visibility into cybersecurity efforts and encourage information -Minimum Strength +sharing. -Clean / Americano Position +Improve identification of cyber threats and better protect information systems from attack. -Maximum Strength +• +• Provide platform to enhance communication, collaboration, and transparency. -Removable Milk Reservoir +Cybersecurity Enhancement Account (CEA) - Other Cyber Priorities -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K +Description: -Measuring Scoop -and Tamper +Treasury maintains investment in Other Cybersecurity Priorities that address and respond to the +changing threat landscape of the Department, given our interconnected technology environment, which +has amplified the need for identifying and assessing the security posture of high value assets, national +security systems infrastructure, Treasury’s supply chain, cybersecurity governance, risk, and +compliance (GRC), amongst other necessary priorities. Additionally, Treasury is investing in a resilient +future through the acquisition of federal labor cybersecurity subject matter expertise to support our +complex technology portfolio, as well as the necessary resources to support the overall administration +and governance of Cybersecurity Enhancement Account (CEA) appropriations and the ongoing +performance of cybersecurity investments. -Portafilter Handle +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -Portafilter Retention Clip +Investment Obligations: (In Millions of $): -Portafilter +T y p e -Double-Shot -Filter +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -Single-Shot +F Y 2 0 2 3 +A c t u a l s -Control panel -All icons will light up when the milk reservoir is correctly inserted into the unit . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -ESPRESSO -Icon +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -LATTE -Icon +C h a n g e i n +$ -MANUAL -ESPRESSO Icon +% C h a n g e -CAPPUCCINO -Icon +1 7 . 7 9 -AMERICANO -Icon +2 . 7 3 -FROTH -Icon +6 . 0 5 -CLEAN -Icon +3 . 3 2 -7 +1 2 1 . 5 4 % -USER MANUAL ESPRESSO MAKER GCM4230 +8 . 7 5 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +2 6 . 5 5 -I +1 2 . 9 0 -USING YOUR -ESPRESSO MAKER -Before first use -1. remove and discard all packaging material and stickers that are not +1 5 . 6 3 -permanent . +1 8 . 9 0 + +2 4 . 9 5 + +6 . 0 1 + +4 6 . 5 8 % + +9 . 3 3 + +5 9 . 6 8 % + +7 -2. remove the water reservoir, milk reservoir, portafilter, filters, measuring -scoop, and removable drip tray. Wash them with non-abrasive soap . -Thoroughly dry the parts . Wipe the exterior with a damp cloth . + Purpose, Accomplishments, Future Objectives: -3. Before using the espresso maker for the first time, it is required to prime the +The CEA investment for other cybersecurity priorities focuses on an enterprise-wide investment +approach to build stronger protections and further benefit the bureaus in their cybersecurity goals. With +the creation of a centralized appropriation, administration, and governance, Treasury has been able to +develop and enhance Department-wide cloud services, support Bureau-level modernization goals, and +drive a more robust coordination of efforts to improve the Department’s ability to detect, identify, +protect, recover, and respond to cybersecurity events. -appliance . See Setting up and priming the Espresso Maker below . +Through the CEA Other Cybersecurity Priorities investment, Treasury maintains investment in +initiatives that address and respond to the changing threat landscape of the Department, given our +interconnected technology environment, which has amplified the need for identifying and assessing the +security posture of high value assets, national security systems infrastructure, Treasury’s supply chain, +cybersecurity governance, risk, and compliance (GRC), amongst other necessary priorities. +Additionally, Treasury is investing in a resilient future through the acquisition of federal labor +cybersecurity subject matter expertise to support our complex technology portfolio, as well as the +necessary resources to support the overall administration and governance of CEA appropriations and +the ongoing performance of cybersecurity investments. -Setting up and priming the -Espresso Maker -1. Place the espresso maker on -a hard, flat, and heat-resistant -surface . +The FY 2025 President's Budget includes $25.0 million for Treasury’s CEA Other Cybersecurity +Priorities investment with alignment to the NIST Cybersecurity Framework priorities: -7. Set the froth strength control to +• -the clean position . +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an -8. Turn the froth dispensing +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -2. Detach the water reservoir and -open the water reservoir cover . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -tube lever to rotate the froth -dispensing tube upward . +appropriate safeguards to ensure delivery of critical services. -9. Place a heatproof cup or mug +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -under the froth dispensing tube . +occurrence of a cybersecurity event. -10. Tap the CLEAN icon . The clean +• Respond to Detected Cybersecurity Incidents Goal: Develop and implement appropriate -icon will blink until the espresso -maker is done preheating . After -preheating, water will dispense -into the cup . When cleaning is -complete, the appliance will beep -three times . +activities to take action regarding a detected cybersecurity incident. -11. Discard the water dispensed into +In FY2025, CEA – Other Cybersecurity Priorities investment is targeted to the Department’s cyber +priorities focused om modernization and the intrinsic enhancement of our cybersecurity posture. Our +investments will build on prior year accomplishments, continue investment in enterprise shared +services, enhance Department-wide coordination of cyber efforts, improve responsiveness to threats, +deliver greater visibility and information shared, and improve threat detection, identification, and +protection capabilities. -the cup . +Treasury is looking to support the following initiatives with final FY2025 appropriations: -3. Fill the water reservoir with water -between the “Min” and “Max” -mark and reattach it to the back -of the appliance . . +• Establish a repeatable programmatic structure to review, assess, and govern emerging -4. Plug the power cord into an +technologies (i.e., quantum, Artificial Intelligence [AI], etc.) to ensure applications of all +security standards develop consistent policies, workforce planning, reporting, Bureau +adherence, ethical use, etc. -outlet . +• Acquisition of an enterprise EDR solution +• Workforce focused cyber awareness and other supplemental trainings +• Enhanced security compliance of a low code platform through Enhanced Adoption and -5. Turn the espresso maker on by -flicking the power switch on the -right side of the appliance . The -ESprESSo icon and ManuaL -ESprESSo icon will light up on -the control panel . +Automation of Change Management/Enablement, as well as increased use of common +controls. -6. +• Evolution of the Enterprise Cyber Risk Management (ECRM) initiative which will continue to -Insert the empty milk reservoir -into the appliance until it clicks -into place . All icons will light up on -the control panel . +strengthen Treasury's approach to managing risks across the enterprise and instantiate a +continuous review of all of its critical networks, systems, and data. 8 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - filling the portafilter -1. Select your filter size . Use the - -single-shot filter for one shot of -espresso . Use the double-shot -filter for larger sized drinks, or -for a double shot of espresso . -Align the gap in the filter with the -groove in the portafilter . Drop the -filter into the portafilter and rotate -the filter to ensure it’s locked in -place . - -2. Use the measuring spoon to fill - -the filter with espresso grounds . -Do not exceed the MAX line in -the filter . - -3. Use the tamper end of the - -measuring spoon to press on -the grounds . The surface of the -espresso grounds should be -smooth after pressing . - -4. Clean any excess grounds from - -the rim of the portafilter . - -placing the portafilter -1. Place the portafilter underneath -the brew head, with the handle -facing toward the left . - -2. Keep one hand atop the - -appliance to keep it steady while -fastening the portafilter into the -brew head . The handle must -align with the open lock icon on -the brew head . - -3. Turn the handle to the right to - -lock the portafilter into the brew -head . The handle must align with -the closed lock icon on the brew -head . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + • Development of a low code Configuration Management Database (CMDB) solution, +encompassing both Asset Management and Configuration Management. +• Development of a Risk Management Framework for enterprise infrastructure +• -9 +Investment in Organizational Resiliency and the associated human capital component to + +address current and emerging cyber priorities such as Supply Chain Risk Management +(SCRM) and incident response support. + +• Performance of enterprise-level FISMA readiness assessments. +• Facilitation of an Enterprise Cyber BPA ordering capability (PROTECTS) +• + +Investment in an Annual Threat Hunt that identifies a random sampling of Treasury systems to + +identify anomalies beyond traditional detection methodologies. + +Acquisition of Incident Response support, allowing for defined response timeline in the +event of a compromise and need for SME and resource depth. + +Looking forward, CEA investments will aim to accomplish the below future objectives: + +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +Cybersecurity Enhancement Account (CEA) - Security Logging + +Description: -USER MANUAL ESPRESSO MAKER GCM4230 +Treasury’s investments in Security Logging ensure compliance with Office of Management and +Budget (OMB) Memorandum 21-31 (M-21-31) to develop and implement security logging capabilities +to receive, store, analyze, and process security event and system log data. Investment funding targets +the creation and monitoring of data-stream disruptions, a shared log facility, storage and retention of +log data, development of threat hunt and incident response playbooks, and the creation of user +behavioral analytics capabilities to support detection of malicious behavior. The investment objective +is for all logs to be accessible and visible for the highest level (enterprise) security operations center at +the Department, the Treasury Shared Services Security Operation Center (TSSSOC). - Mug Size Guide +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. -ESPRESSO +Investment Obligations: (In Millions of $): -CAPPUCCINO +T y p e -LATTE +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -AMERICANO +F Y 2 0 2 3 +A c t u a l s -Drink +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -Shot +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Size +C h a n g e i n +$ -SINGLE +% C h a n g e -DOUBLE +2 5 . 6 6 -SINGLE +2 . 4 8 -DOUBLE +2 9 . 4 1 -SINGLE +2 6 . 9 3 1 , 0 8 8 . 2 4 % -DOUBLE +0 . 0 0 -SINGLE +2 5 . 6 6 -DOUBLE +4 . 0 0 -2 OZ +6 . 4 8 -3 OZ +1 1 . 9 2 -5 OZ +7 . 9 2 -10 OZ +1 9 7 . 9 0 % -12 OZ +4 1 . 3 3 -16 OZ +3 4 . 8 5 -5 OZ +5 3 8 . 2 2 % -10 OZ +9 + + Purpose, Accomplishments, Future Objectives: + +The CEA security logging investment focuses on an enterprise-wide investment approach to build +stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the CEA Security Logging investment, Treasury ensures compliance with Office of +Management and Budget (OMB) Memorandum 21-31 (M-21-31) to develop and implement security +logging capabilities to receive, store, analyze, and process security event and system log data. +Investment funding targets the creation and monitoring of data-stream disruptions, a shared log facility, +storage and retention of log data, development of threat hunt and incident response playbooks, and the +creation of user behavioral analytics capabilities to support detection of malicious behavior. The +investment objective is for all logs to be accessible and visible for the highest level (enterprise) security +operations center at the Department, the Treasury Shared Services Security Operation Center +(TSSSOC). + +The FY 2025 President's Budget includes $41.3 million for Treasury’s CEA Security Logging +investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. -note: To accommodate larger mugs, remove the drip tray . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +appropriate safeguards to ensure delivery of critical services. -I +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the -drink types +occurrence of a cybersecurity event. -Espresso -One shot of -espresso . +In FY2025, CEA – Security Logging investment is targeted to the Department’s security logging +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. -Cappuccino -Made up of -1/3 espresso, -1/3 steamed milk, -and 1/3 milk foam . +Treasury is looking to support the following initiatives with final FY2025 appropriations: -Latte -Made up of -1/3 espresso and -2/3 steamed milk -topped with milk foam . +• Security Logging efforts to protect against Distributed Denial-of-Services (DDoS) and other +cyber-attacks including 1) increase of storage for Treasury Trusted Internet Connection +(TIC) packet capture data, allowing rapid forensic analysis of suspected attacks; 2) +expansion of Web Application Firewall (WAF) capabilities for applications and services +across the Department; and 3) implementation of indexers at the Network Colocations +which will provide critical cyber visibility and analytics. -Americano -Made up of -1/3 espresso and -2/3 hot water . +• -Espresso +Investment in Organizational Resiliency and the associated human capital component necessary -Steamed Milk +to strengthen threat analysis, provide Security Operations Center (SOC) support for +sustained cyber attacks, and establish, operate, and maintain the Treasury logging platform. -Hot Water +• Expansion of the SOC to provide enterprise SOC capabilities for Treasury bureaus, including -Milk Foam +Operations analysts, intake support, Threat & Research analysts, enhanced capabilities, and +associated equipment. + +• + +Investment in cloud-based log processing, storage, and platform licenses. 10 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Making an Espresso - -Ensure the water reservoir is filled with water and placed to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a cup on the drip tray . - -4. For a single shot, tap the - -ESPRESSO icon once . For a double -shot, tap the ESPRESSO icon twice . -Your selected drink type will blink -until the espresso maker is done -preheating . After preheating, the -brewing process will begin . - -5. When brewing is complete, the -appliance will beep three times -and your espresso will be ready -for sipping . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + Looking forward, CEA investments will aim to accomplish the below future objectives: -11 +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +Cybersecurity Enhancement Account (CEA) - Universal Encryption + +Description: + +The Department’s investment in Universal Encryption supports our commitment to fully achieve +encryption protocols outlined in Executive Order 14028 and subsidiary guidance (i.e., Office of +Management and Budget [OMB], Cybersecurity Infrastructure and Security Agency [CISA], and +others). Our investment focuses on full adoption of encryption of data at rest (DAR) and data in transit +(DIT) for all Treasury systems and will address the unique challenges of our environment, proposed +solutions, and the implementations necessary to address each system for compliance with encryption +standards and requirements. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. + +Investment Obligations: (In Millions of $): + +T y p e + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) + +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +0 . 0 0 + +0 . 0 0 + +0 . 0 0 + +2 . 5 0 + +0 . 4 8 + +2 . 9 8 + +1 . 0 0 + +- 1 . 5 0 + +- 6 0 . 0 0 % + +0 . 0 0 + +1 . 0 0 + +- 0 . 4 8 + +- 1 0 0 . 0 0 % + +- 1 . 9 8 + +- 6 6 . 3 9 % + +Purpose, Accomplishments, Future Objectives: -USER MANUAL ESPRESSO MAKER GCM4230 +The CEA universal encryption investment focuses on an enterprise-wide investment approach to build +stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation of a +centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +Through the CEA Universal Encryption investment, the Department supports initiatives that work +toward full achievement of encryption protocols outlined in Executive Order 14028 and subsidiary +guidance (i.e., Office of Management and Budget [OMB], Cybersecurity Infrastructure and Security +Agency [CISA], and others). Our investment focuses on full adoption of encryption of data at rest +(DAR) and data in transit (DIT) for all Treasury systems and will address the unique challenges of our -I +11 + + environment, proposed solutions, and the implementations necessary to address each system for +compliance with encryption standards and requirements. -Making a Cappuccino +The FY 2025 President's Budget includes $1 million for Treasury’s CEA Universal Encryption +investment with alignment to the NIST Cybersecurity Framework priorities: -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement -1. Fill the milk reservoir with milk +appropriate safeguards to ensure delivery of critical services. -7. For a single shot, tap the +In FY 2025, CEA – Universal Encryption investment is targeted to the Department’s universal +encryption focused modernization and the intrinsic enhancement of our cybersecurity posture. Our +investments will build on prior year accomplishments, continue investment in enterprise shared +services, enhance Department-wide coordination of cyber efforts, improve responsiveness to threats, +deliver greater visibility and information shared, and improve threat detection, identification, and +protection capabilities. -below the “Max” mark . +Treasury is looking to support the following initiatives with final FY 2025 appropriations: -2. +• -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -cappuccino function will not be -available as an option . +Investment in Quantum-Resistant Cryptography that could protect the Department against -3. Set the froth strength control to +cryptanalytic attacks by a quantum computer. -the maximum strength . -note: The above setting is a -recommendation only . +Looking forward, CEA investments will aim to accomplish the below future objectives: -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to -5. To attach the portafilter to the +threats. -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . +• Provide leadership with greater visibility into cybersecurity efforts and encourage information -6. Position the froth dispensing tube +sharing. -over a large heatproof cup . +Improve identification of cyber threats and better protect information systems from attack. -CAPUCCINO icon once . For a -double shot, tap the CAPPUCCINO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . +• +• Provide platform to enhance communication, collaboration, and transparency. -8. When brewing is complete, the -appliance will beep three times -and your cappuccino will be -ready for sipping . +Cybersecurity Enhancement Account (CEA) - Zero Trust Architecture -9. Discard the spent grounds by +Description: -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +Zero Trust Architecture (ZTA) investment priorities at Treasury seek to minimize implicit trust and +reinvigorate least privilege through a continual verification of each user, device, application, and +transition. The investment approach requires the Department to enhance visibility and threat detection +at the application level to improve its ability to support continuous threat analysis, detection, and +response, and enable to analysis of encrypted traffic. Compartmentalization, micro segmentation, and +reinforcing enforcement of continuous identity verification and conditional access policies will +improve our resistance to fraudulent tampering of privileged accounts. Our objective is a universal, +default security posture of “never trust, always verify” across our entire technology stack. + +Previously part of the Cybersecurity Enhancement Account (CEA) 015-000200319 Investment. 12 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making a Latte - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. Fill the milk reservoir with milk - -below the “Max” mark . - -6. Position the froth dispensing tube -over a large heatproof cup . - -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -latte function will not be available -as an option . - -3. Set the froth strength control to - -the minimum strength . -note: The above setting is a -recommendation only . - -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -5. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -7. For a single shot, tap the LATTE -icon once . For a double shot, -tap the LATTE icon twice . Your -selected drink type will blink until -the espresso machine is done -preheating . After preheating, the -brewing process will begin . - -8. When brewing is complete, the -appliance will beep three times -and your latte will be ready for -sipping . - -9. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . + Investment Obligations: (In Millions of $): + +T y p e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +1 4 . 8 1 + +3 3 . 1 0 + +1 2 . 8 0 + +- 2 0 . 3 0 + +- 6 1 . 3 3 % + +1 2 . 2 7 + +2 7 . 0 8 + +9 . 2 0 + +4 2 . 0 0 + +3 2 . 8 0 + +3 5 6 . 5 2 % + +4 2 . 3 0 + +5 4 . 8 0 + +1 2 . 5 0 + +2 9 . 5 5 % + +Purpose, Accomplishments, Future Objectives: + +The CEA zero trust architecture investment focuses on an enterprise-wide investment approach to +build stronger protections and further benefit the bureaus in their cybersecurity goals. With the creation +of a centralized appropriation, administration, and governance, Treasury has been able to develop and +enhance Department-wide cloud services, support Bureau-level modernization goals, and drive a more +robust coordination of efforts to improve the Department’s ability to detect, identify, protect, recover, +and respond to cybersecurity events. + +Through the Zero Trust Architecture investment, Treasury seeks to minimize implicit trust and +reinvigorate least privilege through a continual verification of each user, device, application, and +transition. The investment approach requires the Department to enhance visibility and threat detection +at the application level to improve its ability to support continuous threat analysis, detection, and +response, and enable to analysis of encrypted traffic. Compartmentalization, micro segmentation, and +reinforcing enforcement of continuous identity verification and conditional access policies will +improve our resistance to fraudulent tampering of privileged accounts. Our objective is a universal, +default security posture of “never trust, always verify” across our entire technology stack. Treasury +leverages the Zero Trust Architecture investment to provide additional human capital to support +management of service providers, as well as development, testing, and deployment of the applications +that have the security controls integrated into them. This investment also supports the Treasury Zero +Trust enterprise identity access solution, known as Common Approach to Identify Access Management +(CAIA), for all external federal partners who access the Enterprise Content Management (ECM) +SharePoint platform for shared services. + +The FY 2025 President's Budget includes $54.8 million for Treasury’s CEA Zero Trust Architecture +investment with alignment to the NIST Cybersecurity Framework priorities: + +• + +Identify the Business Context, Resources & Cybersecurity Risk Goal: Develop an + +organizational understanding to manage cybersecurity risk to systems, people, assets, data, +and capabilities. + +• Protect the Delivery of Critical Infrastructure Services Goal: Develop and implement + +appropriate safeguards to ensure delivery of critical services. + +• Detect Cybersecurity Events Goal: Develop and implement appropriate activities to identify the + +occurrence of a cybersecurity event. + +In FY 2025, CEA – Zero Trust Architecture investments are targeted to the Department’s zero trust +focused modernization and the intrinsic enhancement of our cybersecurity posture. Our investments +will build on prior year accomplishments, continue investment in enterprise shared services, enhance +Department-wide coordination of cyber efforts, improve responsiveness to threats, deliver greater +visibility and information shared, and improve threat detection, identification, and protection +capabilities. 13 -USER MANUAL ESPRESSO MAKER GCM4230 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making an americano - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. - -Insert a filled or empty milk -reservoir into the appliance until it -clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -americano function will not be -available as an option . - -2. Set the froth strength control to -the Clean / Americano position . - -3. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -4. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -5. Position the froth dispensing tube -over a large heatproof cup . - -6. For a single shot, tap the - -AMERICANO icon once . For a -double shot, tap the AMERICANO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . - -7. When brewing is complete, the -appliance will beep three times -and your americano will be ready -for sipping . - -8. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . + Treasury is looking to support the following initiatives with final FY2025 appropriations: + +• Expansion of Treasury’s cyber threat intelligence platform to include additional cyber threat +intelligence feeds, additional material from existing feeds, and expand access to platform +capabilities and shared data. + +• + +Implementation of a TSSSOC lateral scanning capability across enterprise application and + +centralized incident response + +• Acquisition of an enterprise Endpoint Detection and Response (EDR) solution +• + +Investment in Organizational Resiliency and the associated human capital component necessary + +to implement zero trust architecture across the Department’s IT systems + +• Bureau implementation of zero trust architecture +• Development or acquisition of Security Operations Center (SOC) services for the Treasury + +Security Data Network + +• Expansion of Treasury’s mobile Secret Internet Protocol Router (SIPR) capability +• Enterprise-wide vulnerability management efforts including the establishment of an enterprise +penetration testing shared service offering and acquisition of a vulnerability management +tool to track and conduct analysis of enterprise-wide critical vulnerabilities + +• Development of application threat modeling capabilities to prepare for increasing resilience of + +systems and robustness of testing. + +Looking forward, CEA investments will aim to accomplish the below future objectives: + +• Enhance Department-wide coordination of cybersecurity efforts and improve responsiveness to + +threats. + +• Provide leadership with greater visibility into cybersecurity efforts and encourage information + +sharing. + +Improve identification of cyber threats and better protect information systems from attack. + +• +• Provide platform to enhance communication, collaboration, and transparency. + +HR LoB - HRConnect + +Description: + +HR Connect is a Human Resources enterprise system. It is a web-based solution built on PeopleSoft +software. HR Connect transforms core back-office HR functions, moving them from a processing- +centric capability to a strategic-centric capability. + +Investment Obligations: (In Millions of $): + +T y p e + +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s + +F Y 2 0 2 3 +A c t u a l s + +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s + +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s + +C h a n g e i n +$ + +% C h a n g e + +1 2 . 2 8 + +1 8 . 5 9 + +1 8 . 8 7 + +0 . 2 8 + +1 . 5 1% + +3 7 . 1 4 + +4 9 . 4 2 + +3 7 . 0 5 + +5 5 . 6 4 + +3 8 . 2 6 + +5 7 . 1 3 + +1 . 2 1 + +3 . 2 7% + +1 . 4 9 + +2 . 6 8 % 14 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Brewing in Manual Mode - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a heatproof cup over the - -drip tray . - -4. Tap the MANUAL ESPRESSO icon . - -The text below the icon will blink -for three seconds then the brew -cycle will begin . - -5. When your desired amount of -coffee has dispensed, tap the -MANUAL ESPRESSO icon again and -coffee will cease brewing . -note: The maximum amount of -coffee that will dispense is 14 oz . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I + Purpose, Accomplishments, Future Objectives: -15 +HRConnect is Treasury's enterprise human resources system. It is one of four federal OPM HR Lines +of Business providing HR services to the federal government. HRConnect is based on a combination of +(a) web-based solution built on PeopleSoft commercial-off-the-shelf (COTS) software, (b) Software as +a Service (Saas) platforms (e.g. Talent Management and Career Connector) and (c) internally +developed applications (e.g. Entrance on Duty System). HRConnect transforms core back-office HR +functions, moving them from a processing-centric capability to a strategic-centric capability enabled +through its commercial software underpinning. Additionally, self-service components of the software +fundamentally transform the standard government HR service delivery model, putting additional +information, services and processes (i.e., personal data, position management, requests for personnel +action, recruitment, reporting, etc.) directly in the hands of managers and employees. -USER MANUAL ESPRESSO MAKER GCM4230 +HRConnect supports the common HR Line of Business processes and provides core HR functionality +that is interoperable, portable and scalable. This shared solution provides automated systems that are +configurable to the individual organizations' needs while providing a single solution across the +Department and federal landscape. HRConnect's core functions include: Personnel Action Processing, +Managing Payroll, Administering Benefits, Time and Attendance and Labor Distribution. By enabling +the retirement of legacy systems and automating and streamlining many aspects of human resources, +HRConnect facilitates increased efficiency and overall productivity for its customers. HRConnect is +the system used by all Treasury bureaus and several other government agencies (over 22 entities) with +over 200,000 employees and contractors in total. - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +HRConnect Top FY 2023 Accomplishments: -I +1. Implemented 74 HRConnect change requests to maintain regulatory compliance, improve the +employee and manager experience, and provide tools to streamline and enable efficient +processing for HR Specialists. -frothing Milk +2. Maintained a steady state of 100% availability for customers. +3. Treasury HRLOB implemented multifactor authentication (MFA) for all TSSC applications -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +and achieved M-21-31 compliance for the HRConnect High Value Asset (HVA). -1. Fill the milk reservoir with milk +4. Completed the migration of HRConnect’s data service, Workforce Analytics from on prem +data centers to a FedRAMP High secure cloud hosted environment, effectively scaling and +maturing the platform. -5. Tap the FROTH icon . +5. Delivered Integrated Talent Management (ITM) and HRConnect real-time performance plan -below the “Max” mark . +and learning data integration. -2. Secure the milk reservoir in the +6. Implemented new Remote/Telework Agreement Type and Program and Project Job Identifier -appliance until it clicks into place . +fields and Mass Update Process, 34K positions updated. -3. Set the froth strength control . +7. Streamlined the background investigation security process across Treasury by expanding the -4. Position the froth dispensing tube +Adjudication Tracker application to three additional bureaus. -over a heatproof cup . +8. Continued to modernize HRConnect delivering enhancements to improve the user experience -6. Milk will froth for a maximum of -three minutes . To stop frothing -earlier, tap the FROTH icon again -and milk will cease frothing . +for employees, managers, and HR Specialists. -16 +9. Deployed a visual representation of key Time to Hire (T2H) metrics via the Treasury + +Workforce Dashboard to track and optimize the hiring timeline and make data-driven hiring +decisions. -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +10. Continued employee self-service (ESS) redesign leveraging fluid concepts. +11. Supported gender identity and other preferred name representation in accordance with - I +management directives. -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +15 -I + Enterprise API Service Platform [MuleSoft]: Transition 15+ partners to the new API Service +Framework. Launched integrations between USAS to DO and FinCEN Adjudication Tracker, ARC- +FS Worklist API and SAM.gov integration to HRC to update vendor information. +In FY2023: Treasury Office of the Chief Information Officer (OCIO) recently assumed responsibility +for the management of the transformation of Internal Revenue Service (IRS) Human Resource +Information Technology (HRIT) applications in support of the Inflation Reduction Act. This transfer of +responsibility allows the IRS CIO to focus their efforts on improving the taxpayer experience +leveraging the Treasury OCIO’s shared services and common platforms to facilitate the modernization +of IRS HRIT systems, the rationalization pillars include: HRConnect, Integrated Talent Management, +USA Staffing, Enterprise Data Management, and the Employee Portal. -CLEANING AND MAINTENANCE +A few of technical improvements, included: Nessus implementation, CAST implementation, piloted +the transition of 3 environments to Oracle Managed Services compartment in Q3, improved the fire +wall configurations and finalized the closure of the Memphis and Martinsburg HRC footprint. -the following cleaning instructions should be followed after every -use of the espresso maker. +Treasury Enterprise Identity, Credential and Access Management (TEICAM) -Cleaning the portafilter and Brew Head +Description: -1. Turn the appliance off and unplug -the power cord from the outlet . -Wait for the metal parts to cool -down before touching them . +The Treasury Enterprise Identity, Credential and Access Management (TEICAM), formerly submitted +as EIdM, consolidates funding of Treasury implementing the Homeland Security Presidential +Directive- (HSPD) 12, E-Auth, and Federal PKI initiatives. -2. Remove the portafilter from the -brew head and dump the used -espresso grounds . Lift the filter -retainer clip so that it’s placed -over the filter edge, and hold it in +Investment Obligations: (In Millions of $): -place as you dump the espresso -grounds into the trash . +T y p e -3. Rinse the single-shot filter, +F Y 2 0 2 3 +A c t u a l s -double-shot filter, and portafilter -with warm, soapy water . +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -4. Wipe the underside of the brew -head with a damp cloth to clean -off any excess coffee grounds . +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Cleaning the Water reservoir +C h a n g e i n +$ -1. Discard leftover water . +% C h a n g e -2. Wash the water reservoir with soapy water . Rinse and dry well . +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) -Cleaning the Milk reservoir and frothing tube +4 3 . 0 7 -1. Remove the milk reservoir from +7 5 . 9 4 -the unit . If there is any remaining -milk, pour the milk into a separate -container and store it in the -refrigerator . +7 5 . 9 8 -2. Wash the milk reservoir and milk +0 . 0 4 -reservoir lid thoroughly with warm, -soapy water . Do not use abrasive -cleansers and scouring pads . -note: Hand wash only! The milk -reservoir is not dishwasher safe . +8 1 . 6 2 -3. Ensure there is water in the water +7 8 . 9 9 -reservoir . +8 0 . 9 5 -4. Position the froth dispensing tube -over a large heatproof cup . +0.05% -5. Set the froth strength to the Clean -/ Americano position . Tap the -CLEAN icon . +2.48% -6. Wait until all the water has -finished dispensing before -removing the cup from -underneath the brew head -(the unit will beep three times -and other function buttons will -reappear) . +1.29% -7. Dispose of the water in the mug -and any remaining water in the -milk reservoir and water reservoir . +1 . 9 6 -17 +2 . 0 0 + +T o t a l O b l i g a t i o n s + +1 2 4 . 6 9 + +1 5 4 . 9 3 -USER MANUAL ESPRESSO MAKER GCM4230 +1 5 6 . 9 3 - I +Purpose, Accomplishments, Future Objectives: -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +The TEICAM Business Case consolidates funding that supports Treasury’s implementation of HSPD- +12, FICAM and PKI requirements. This investment supports the target vision of "One Treasury One +Card" to provide universal access. Availability and use of a PIV/identity management standard within +Treasury provides a mechanism for Physical and Logical access to Treasury-wide assets. The TEICAM +phased implementation provides all of Treasury: -I +• Trusted identity processes for both Federal and public users of Treasury applications +• Enhanced security (by decreasing data breaches and trust violations) +• Compliance with laws, regulations, and standards +• +• Expanded use of PIV encryption and digital signature certificates (ensuring adoption of multi- -descaling +Improved interoperability -After using the espresso maker for a while, limescale will begin to build up inside -the appliance . It is important to descale the appliance every 2-3 months . If the -espresso maker is used daily, the descaling process should be done monthly . +factor authentication) +• Elimination of redundancy -1. Fill the removable water reservoir +16 + + Treasury/TEICAM has achieved many of the defined goals for PIV card issuance, physical access, +logical access, data synchronization, enterprise single sign-on, federation, and PIV required for both +privileged and unprivileged users. These goals have helped the Department align with the required +OMB and FICAM goals. +Additionally, TEICAM has updated the Department strategic roadmap & planned for the following +investment goals: + +1) CDM II-PRIV/CRED Management + +o Continue working with the Managed Service Bureaus and Non-Managed Service + +Bureaus on implementation of Credential (CRED) and Privileged Access Management +(PAM) + +o On-board new bureaus, such as IRS and DO Orbit Organizations (TSSOC, OFR, etc.), + +to the enterprise CDM Managed Service Offering (MSO) + +o Continue working with DHS CISA on the integration activities for the Federal -5. On the following day, discard the +Dashboard for Treasury’s CDM Data reporting -with equal parts water and -vinegar between the “Min” and -“Max” mark . +2) Plan, test, & implement the issuance of version 8.1 PIV cards (PIV V8.1) across the enterprise -2. Use the froth dispensing +in FY21-FY24 -tube lever to rotate the froth -dispensing tube upward and -position a mug underneath it . +3) Expand usage of the Enterprise Derived Credential issuance capability to support -3. Tap the CLEAN icon . +authentication to Treasury services/infrastructure from mobile devices across Treasury in +FY21-FY24 -4. Allow remaining water and +4) As part of the PIV-I solution, implement an application programming interface (API) -vinegar to sit overnight in the -water reservoir . +connection between a Fiscal Service application used to capture applicant data and +HRConnect in FY21-FY24 -troubleshooting +5) Implement an enterprise Hardware Security Module (HSM) as a service in FY22-FY24 +6) Enterprise solution for tracking and monitoring the Only Locally Trusted (OLT) PKI -water and vinegar . +certificates generated and in use within each bureau -6. Rinse the water reservoir +7) Plan, design, and implement a modernized Treasury Enterprise Federation Service (TEFS) -thoroughly with tap water and -then pour the water back into the -sink . +solution that enforces multifactor authentication and aligns with the Treasury Zero Trust +Architecture (ZTA) -7. Fill up the water reservoir with +8) In an effort to improve cost-savings, the Department utilizes interagency resources to +authenticate users, synchronize data, and to procure and maintain enterprise-wide +compliant PIV credentials (USAccess). As a mixed life-cycle investment, the TEICAM +Operations and Maintenance tasks includes OMB, FISMA, and Cyber reporting specific to +identity, credential and access management. -water and begin brewing again . -This washes out any remaining -vinegar . -notE: If an odor of vinegar -persists, run another cycle with -water . +Planned objectives and accomplishments include: -ISSUE +• Maintaining above 95% PIV card issuance rate and providing replacements for the PIV in time -SOLUTIONS +sensitive activities; -Only the ESPRESSO and MANUAL -ESPRESSO icons appear on the control -panel. +• Maintained 100% PIV required privileged account access and 97% PIV required unprivileged -A drink has been chosen and the unit is -running, but no coffee is dispensing from -the portafilter. +access; -Water is spreading from the drip tray -when brewing or cleaning. +• CDM II privileged access management is operating with a cloud infrastructure and production +at four bureaus performing O&M functions; we have commenced technical work on +bringing the IRS on as the fifth and largest managed service customer; -The other icons will appear when the milk -reservoir is correctly inserted into the unit. -Insert the milk reservoir into the unit by aligning -the groove on top of the reservoir with the -notch on the unit and pushing. It will click into -place when done correctly. +• Treasury Enterprise Federation Service (TEFS) onboarded 12 new customers and integrated -Detach then reattach the water reservoir to the -unit (this may need to be done a few times). -Run a clean cycle. See CLEANING THE MILK -RESERVOIR AND FROTHING TUBE on page 17. +with third-party identity providers to expand citizen services and access; +• Executed a Treasury-wide BPA for citizen authentication services access to the entire +enterprise, allowing for secure and fast public access to Treasury systems; + +• Developed an alternative authentication solution and deployed it within 6 weeks to ensure + +continuity of government during the pandemic; + +• + +Implemented a hardware security module (HSM) as a service for the Derived PIV solution and + +DO. + +17 -A. Water dispenses from the froth dispensing -tube when brewing Americano or running a -clean cycle. Ensure the froth dispensing tube is -above the cup. + • -B. Empty the drip tray. +Implementation of the department-wide PKI Only Locally Trusted (OLT) monitoring solution. 18 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 + Major Non-IT Investments - WARRANTY -AND SERVICE +Main Treasury Building and Freedman's Bank Building -This Limited Warranty is for one full year from the date of -purchase and requires proof of purchase from an authorized -seller . It is applicable to the original purchaser only and is not -transferable to a third party user . Repair or replacement of -defective parts is solely at Gourmia’s discretion . In the event -that repair isn’t possible, Gourmia will replace the product/part . -If product repair/replacement won’t suffice, Gourmia has the -option of refunding the cash value of the product or component -returned . +Description: -Product defects not covered under the Warranty provisions -include normal wear and damage incurred from use or -accidental negligence, misuse of instruction specifications -or repair by unauthorized parties . Gourmia is not liable for -any incidental or consequential damages incurred by such -circumstances . +Correct life safety and code issues, reduce building systems risk, and maintain the buildings. Absent +full funding to perform a complete repair and renovation, Treasury is utilizing available funding to +correct the most severe issues. -Only valid within the United States . +Investment Obligations: (In Millions of $): -register your product at -www.gourmia.com/warranty +T y p e -for customer service, questions or comments +F Y 2 0 2 3 +A c t u a l s -Visit +F Y 2 0 2 4 +E s t i m a t e d +O b l i g a t i o n s -gourmia.com +F Y 2 0 2 5 +E s t i m a t e d +O b l i g a t i o n s -Email +C h a n g e i n +$ -info@gourmia.com +% C h a n g e -Call +S u b - T o t a l D M E O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l l a b o r ( G o v t . F T E ) ) +S u b - T o t a l O & M O b l i g a t i o n s ( I n c l u d i n g +I n t e r n a l L a b o r ( G o v t . F T E ) ) +T o t a l O b l i g a t i o n s -888.552.0033 -Mon-Thu 9:00 a .m . to 6:00 p .m . ET -Fri 9:00 a .m . to 3:00 p .m . ET -Languages: English, Spanish, French +9 . 7 9 -@gourmia +0 . 0 0 -I +9 . 7 9 -E -C -v -r -E -S -d -n -a -Y -t -n -a -r -r -a -W +3 . 5 5 -19 +0 . 0 0 + +3 . 5 5 + +5 . 5 5 + +2 . 0 0 + +5 6 . 2 5 % + +0 . 0 0 + +5 . 5 5 + +0 . 0 0 -USER MANUAL ESPRESSO MAKER GCM4230 +0 . 0 0 % - Get cooking with +2 . 0 0 + +5 6 . 2 5 % + +Purpose, Accomplishments, Future Objectives: + +This investment is to address life safety and code compliance issues, reduce building systems risk by +upgrading a number of outdated systems, and bring both facilities into alignment with current building +standards. Absent full funding to perform a complete repair and renovation of these historical +buildings, available funding will be used to correct the most urgent issues. These investments are being +executed with the expectation that were Treasury to pursue a full renovation and modernization, recent +investments could be largely retained, achieving cost savings over the long term. These investments +support a safe and healthy work environment that meets Treasury operational requirements. Project +needs are mission focused and prioritized based on life-safety, security, code discrepancies, and needs +that pose significant financial risk if not addressed in a timely manner. + +Treasury’s three-step long-term strategy to continue to maintain and modernize its owned spaces: (1) +secure the building’s outer envelope; (2) conduct a condition assessment to identify additional needs +associated with the buildings’ continual aging and deferred maintenance; and (3) based on this +assessment, conduct a holistic modernization of the building’s systems and infrastructure. Targeted +investments address life safety and code compliance issues, reduce building systems risk by upgrading +outdated systems, and bring both facilities into alignment with current building standards. Absent full +funding to perform a complete repair and renovation of these historical buildings, available funding +will be used to correct the most urgent issues. These investments are being executed with the +expectation that were Treasury to pursue a full renovation and modernization, recent investments could +be largely retained, achieving cost savings over the long term. These investments support a safe and +healthy work environment that meets Treasury operational requirements. Project needs are mission +focused and prioritized based on life-safety, security, code discrepancies, and needs that pose +significant financial risk if not addressed in a timely manner. + +Treasury Operations has continued to strategically focus on restoring the health of the building +envelope (shell), to correct the deteriorating building structure and infrastructure. Components of the + +19 -We’d love to hear from you! -888.552.0033 -info@gourmia.com + Main Treasury building that have been repaired, replaced, or funded include new roofs (FY 2019/2021) +and repairs to 40 percent of the windows (FY 2015/2017). The Freedman’s Bank Building components +that have been repaired, replaced, or funded/planned include new roofs (FY 2017/2019), domestic +water line (FY 2020/2022), and replacement of all windows (FY 2009/2011). Funding and completion +of these exterior items represent significant progress towards the preservation of the two buildings, +maintaining a safe and healthy workplace, and reducing damage to the interior plaster and paint. The +contract for the first phase (west façade) of the Main Treasury exterior restoration project was awarded +late in FY 2021 and completed during FY 2022. Exterior restoration work continued through 2023 with +the completion of phase 2 (south) and phase 3 (north) north façades. In FY 2024, assuming adequate +funding, Phase 4 work will commence and, in FY 2025, with this requested funding, work on Phase 5 +will commence. + +The MT Building houses the entire chilled water plant for the Main Treasury Complex. A five cell +cooling tower feeds four water cooled centrifugal chillers providing a total cooling capacity of 2,050 +tons. The chillers are located in a mechanical room under the northwest lawn. The four chillers provide +over 98% of the cooling required for the entire Treasury Headquarters complex; therefore, the plant is +operated 24/7, 365 days per year. The Freedman’s Bank Building (FBB) does not generate chilled +water but rather uses chilled water from the main plant distributed by a dedicated secondary chilled +water pump which serves most air handling units directly. The Chilled Water System is over 30 years +old and has a wide variety of repair needs including bearings, motors, valves, distribution piping, +insulation, controls, and other repairs. The system has redundancy for repairs. However, with all of the +components aging at the same time the occurrence of repair needs combined with the time to repair +diminishes the reliability of the redundant capacity. It is imperative that these items are +repaired/replaced before the available redundancy is lost. A system breakdown would indefinitely +shutdown the Treasury Complex’s daily operations. This first-year cost of a multi-year project includes +design for this effort. + +There is currently a backlog of damaged paint and plaster as well as frayed carpet throughout MT and +FBB. Some of the paint and plaster damage is due to water intrusion. These funds will supplement +other funding to address the backlog in repairing these problems and implementing a regular schedule +of maintenance. The damaged paint and plaster are unsightly and pose potential health and safety risks, +while the frayed carpet is a safety hazard. 20 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 \ No newline at end of file diff --git a/q119/random_k4/random_3.md b/q119/random_k4/random_3.md index 1fcd9c0007bb10d1c45cc658d02fe79e8c0298ad..e61c09a30f5ce985c14428b31252865b9bfb8032 100644 --- a/q119/random_k4/random_3.md +++ b/q119/random_k4/random_3.md @@ -1,2074 +1,540 @@ -Priya Sharma, MD -Department of Radiology -University of Florida Gainesville +Module 9: Presentation Skills and Public Speaking + +GOALS +By the end of this module, participants should be able to understand: + • Strategies for becoming a better public speaker. + • Strategies for effective presentations. + • Strategies for calming nerves. + +TIME +Total: 30 minutes + +OVERVIEW +The focus of this session is to help people think more strategically about their public +speaking and presentation skills in order to improve their overall effectiveness in relaying +information to others about advocacy skills and the Common Core State Standards. + +PTA leaders have the opportunity to share the PTA messages to many groups. Additionally, +PTA leaders are called upon to use public speaking and presentation skills on a regular +basis, even if the leaders are not comfortable speaking in public. + +MATERIALS NEEDED + • Easel paper + • Masking tape + • Dark markers + +HANDOUTS + • PowerPoint printout + • Presentation Checklist + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking + EQUIPMENT NEEDED + • Easel/flipchart + • If you have elected to develop your own PowerPoint presentation to suit your + + specific training audience for this module, you will need: + +– Computer with PowerPoint software +– LCD projector +– All cords and cable needed to connect computer with LCD projector, + and extension cord to connect computer and LCD projector with electrical + outlet +– AV stand or tab (if you will be using PowerPoint in this module) +– Portable speakers for videos + +COPYRIGHT NOTICE +Materials in this module are Copyright, National PTA. PTA shall have the right to use +them for training purposes within, and at all, levels of the association in perpetuity. +Training modules may not be used or sold to other entities. + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking + + Introduction + +WELCOME TO MODULE 9: PRESENTATION SKILLS +AND PUBLIC SPEAKING + +SAY +By the end of this module you should understand: + • Strategies for becoming a better public speaker. + • Strategies for developing stronger presentations. + • Strategies for calming nerves. + +Chances are that you will have to speak in public as part of your role with PTA. While this +can seem intimidating, the benefits of being able to speak well outweigh any perceived fears. + +In this session we are going to discuss strategies for becoming a better public speaker, +delivering effective presentations and calming nerves that are associated with public +speaking. + +Even if you don’t need to make regular presentations in front of a group, good public +speaking skills are useful for all aspects of your life, work and home. For example, you +might have to talk about your organization at a conference, make a speech after accepting +an award or provide training for new PTA members. You might be asked to make a speech +at a friend’s wedding, give a eulogy for a loved one or inspire a group of PTA volunteers at +a meeting. In short, being a good public speaker can enhance your reputation, boost your +self-confidence, and open up countless opportunities. + +However, while good public speaking skills can open doors, poor speaking skills can close +them. For example, your boss might decide against promoting you after sitting through a +poorly-delivered presentation. You might lose a valuable new PTA member by failing to +connect with a prospect during a PTA sales pitch or you could make a poor impression +with your new team because you trip over your words and don’t look people in the eye. + +Make sure that you learn how to speak well! + +ASK +Who is afraid of public speaking? (Take a few minutes to discuss.) + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking SAY +Well, we are not alone! Some surveys and research results show that most people are +terrified of speaking in front of a live audience, as you can see in this list of the top 10 +fears globally! + +DISPLAY ON A FLIPCHART OR POWERPOINT: +Top 10 Global Fears + 1. Public speaking or stage fright + 2. Death and end of life + 3. Spiders and other arachnids creatures + 4. Darkness and twilight + 5. Heights, altitude or elevations + 6. People or social situations + 7. Flying in airplanes + 8. Open spaces and squares + 9. Natural thunder and lightning + 10. Confined spaces or small rooms + +SAY +Chances are that you will at some point have to speak in public as part of your role with +PTA. While this can seem intimidating, the benefits of being able to speak well outweigh +any perceived fears. To become a better speaker, use the following strategies: -GUIDELINES FOR RESIDENTS IN FLUOROSCOPY +DISPLAY ON A FLIP CHART OR POWERPOINT: + • Planning + • Practice + • Engaging Your audience -Welcome to GI/GU Fluoroscopy +• Body Language +• Thinking Positively -Introduction: + • Calming Nerves + +SAY +Planning — Make sure that you plan your communication appropriately to think about +how you’ll structure what you’re going to say from your opening words all the way +through to your closing statement. + +Think about how important a book’s first paragraph is; if it doesn’t grab you, you’re likely +going to put it down. The same principle goes for public speaking: from the beginning, +you need to intrigue your audience. You will notice that in each of the PTA modules we +have included “attention grabbers” throughout the scripts. These can include quotes from +famous people on the topic, statistics that emphasize your point or funny anecdotes that +lighten the mood. -The aim of the Fluoroscopy Service at UF is to provide complete and accurate +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking You also want to develop your closing remarks. What you leave your audience with should +summarize what has been covered, the key point or points you want them to take away +and stress any action you want them to take after the session. + +Planning also helps you to think on your feet. This is especially important for unpredictable +question and answer sessions or last-minute communications. + +And finally, it is very important that you make presentations “your own” even if you have +been provided with all the written tools including a script. If your audience does not feel +like you are genuinely knowledgeable or excited about the topic, they will not engage with +you or find your material valuable. + +Practice + +There’s a good reason that we say, “Practice makes perfect!” You simply cannot be a +confident, compelling speaker without practice. + +To get practice, seek opportunities to speak in front of others. As a PTA member, you +have the perfect audience in your PTA family. Your local PTA is a great place to develop +your public speaking skills because you already know your audience, are familiar with the +material and your surroundings and you should have a higher comfort level. The first step +is by volunteering to speak at meetings. + +When you are planning to deliver a presentation or a prepared speech, create it as early as +possible. The earlier you put it together, the more time you’ll have to practice. This is not +the time to wait until the last minute! + +Practice it plenty of times alone, using the resources you’ll rely on at the event, and, as you +practice, tweak your words until they flow smoothly and easily. Then, if appropriate, do a +practice run in front of a small audience: this will help you calm your fears and make you +feel more comfortable with the material. Your audience can also give you useful feedback, +both on your material and on your performance. + +Whenever possible, record your presentations and speeches. You can improve your speaking +skills dramatically by watching yourself later, and then working on improving in areas that +did not go well. As you watch, notice any verbal stalls, such as “um” or “like”, and practice +replacing those verbal stalls with brief pauses. Look at your body language: are you swaying, +leaning on the podium or leaning heavily on one leg? Are you looking at the audience? +Did you smile? Did you speak clearly at all times? Pay attention to your gestures. Do they +appear natural or forced? Make sure that people can see them, especially if you’re standing +behind a podium. -studies performed in an expeditious manner, with attempts to minimize patient +Last, look at how you handled interruptions, such as a sneeze or a question that you weren’t +prepared for. Does your face show surprise, hesitation, or annoyance? If so, practice managing +interruptions like these smoothly so that you’re even better next time. + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking + Engaging Your Audience -discomfort and delay while assisting referring physicians in answering clinical +When you speak, try to engage your audience. This makes you feel less isolated as a speaker +and keeps everyone involved with your message. If appropriate, ask leading questions +targeted to individuals or groups, and encourage people to participate and ask questions. + +Keep in mind that some words reduce your power as a speaker. Think about how these +sentences sound: + + “I just want to add that I think we really need the Common Core State Standards.” -conundrums. While many answers can be provided through the use of cross- + “I just think this plan is a good one.” -sectional imaging, fluoroscopy provides real-time visualization of dynamic processes +Phrases such as “Just” and “I think” limit your authority and conviction. Don’t use them. -as they occur. + “I believe in the Common Core State Standards because they are clearer and more focused.” -• Each day begins after morning conference where you will stop by in + “I am confident that this plan will increase enrollment and allow our members stronger -Fluoroscopy to obtain the list of daily cases (from Nanci – Lead Fluoroscopy + advocacy efforts.” -Technologist). +Notice the increase in authority and conviction? -• Review your schedule and find out which studies are happening in the North +A similar word is “actually.” When you use “actually,” it conveys a sense of submissiveness +or even surprise. -and South Tower. If you notice that there are studies that are scheduled at + “Actually, I’d like to add that we were under budget last year.” -the same time (requiring you to be two places at once!), please ask a resident +By removing actually your message is clear, concise and stresses confidence: -who has already completed his/her fluoroscopy rotation for assistance + “We were under budget last year.” -(covering resident) in your schedule. As the GI/GU resident you will have +Also, pay attention to how you’re speaking. If you’re nervous, you might talk quickly. +This increases the chances that you’ll trip over your words or say something you don’t +mean. Force yourself to slow down by breathing deeply. Don’t be afraid to gather your +thoughts; pauses are an important part of conversation, and they make you sound +confident, natural, and authentic. -first choice on which study you would like to do! +Finally, avoid reading word-for-word from your notes. When you read, you are not +looking at your audience so you will not be seen as an engaging speaker. Additionally, +when you try to read from your notes and look up occasionally, you appear awkward +and come across as less confident. + +Instead, make a list of important points on index cards and use bullet points. For instance +for this workshop, I know I need to cover each of the strategies: planning, practice, engaging +your audience, body language, thinking positively, and calming nerves. + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking + Here is a sample index card that allows you to engage with your audience: + +DISPLAY ON A FLIP CHART OR POWERPOINT: +Planning + • Attention grabbers + • Closing summation + • Plan for Q&A + • Making it your own -o The covering resident will likely be able to assist you by doing a few of +SAY +Body Language +Your body language will give your audience constant, subtle clues about your inner state. +If you’re nervous or if you don’t believe in what you’re saying, the audience will know by +how you present yourself. Research has shown that up to 94% of communication is non- +verbal! -the swallowing studies that have been scheduled in the North Tower. +ASK +What are some examples of non-verbal communication that you have seen during +presentations? (Discuss both positive and negative examples.) -o Please make sure that you notify the head fluoro tech that the +Examples: + Pay attention to your body language: stand up straight, take deep breaths, look people in the + eye and smile. Don’t lean on one leg or use gestures that feel unnatural. Don’t cross your arms. -covering resident will be covering cases in the North Tower, while you +SAY +Many people prefer to speak behind a podium when giving presentations. While +podiums can be useful for holding notes, they put a barrier between you and the +audience. They can also become a “crutch,” giving you a hiding place from the +dozens or hundreds of eyes that are on you. + +Instead of standing behind a podium, walk around and use gestures to engage the +audience. This movement and energy will also come through in your voice, making +it more active and passionate. -perform cases in the South Tower. +Don’t forget to SMILE! When you smile, your audience is more inclined to forgive +nervousness and they will be rooting for you. -▪ Provide the Head Fluoro Tech with the name of your covering +Think Positively — Positive thinking can make a huge difference to the success of your +communication because it helps you feel more confident. + +Fear makes it all too easy to slip into a cycle of negative self-talk, especially right +before you speak. -resident and an efficient means to communicate with him/her. +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking Self-sabotaging thoughts such as, “I’ll never be good at this!” or “I’m going to fall flat on +my face!” lower your confidence and increase the chances that you won’t achieve what +you’re truly capable of. - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Use affirmations and visualization to raise your confidence. This is especially important +right before your speech or presentation. Visualize giving a successful presentation, and +imagine how you’ll feel once it’s over and when you’ve made a positive difference for +others. Use positive affirmations such as, “I am the local expert on this topic!” or “I am +going to do well!” -• After obtaining the list of cases it is important to determine why the +Don’t forget to smile! -studies are being requested. The key to a successful fluoroscopy rotation is +DISPLAY ON A FLIPCHART OR POWERPOINT: +“The way you overcome fear and shyness is to become so wrapped up in something that +you forget to be afraid!” - Lady Bird Johnson -taking an active role a patient care advocate. This involves obtaining a good +SAY +Cope With Your Nerves -history- through EPIC (medical records), conversations with the referring +ASK +How often have you listened to or watched a speaker who really messed up? -clinicians as well as the review of prior imaging. Do not forget to review +SAY: +Chances are the answer is “not very often.” When we have to speak in front of others, we +can envision terrible things happening. We imagine forgetting every point we want to +make, passing out from our nervousness, or doing so horribly that we will never be asked +to speak again. But those things almost never happen! We build them up in our minds +and end up more nervous than we need to be. + +Make an effort to stop thinking about yourself, your nervousness, and your fear. Instead, +focus on your audience: what you’re saying is “about them.” Remember that you’re trying +to help or educate your audience in some way, and your message is more important than +your fear. Concentrate on the audience’s wants and needs, instead of your own. + +Crowds are more intimidating than individuals, so think of your speech as a conversation +that you’re having with one person. Although your audience may be 100 people, focus on +one person at a time, and talk to that person as if he or she is the only one in the room. +You should seek several people within the audience to focus on, ideally strategically located +around the room. + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking DISPLAY ON A FLIPCHART OR POWERPOINT: +Steps for Conquering Your Presentation Nerves + Step 1 – Know your audience + Step 2 – Know your material + Step 3 – Structure your presentation + Step 4 – Calm yourself from the inside + +SAY +Notice we didn’t say to get rid of your nervousness. This is because presenting is not a +natural activity. Even the most practiced presenters get a bit nervous. The point is this: +your nervous energy can be used to your advantage. When you are in a heightened state +from the adrenaline that is being pumped into your body, you can use that energy to +communicate enthusiastically, convincingly, and passionately. The key is to decrease your +level of nervousness so you can use your energy on these positive activities, not on trying +to control your nerves. The more uncertain you are, the more nervous you will be. + +Step 1 — Know Your Audience: If you are asked to speak to a group, make sure to take +time to find out a few things beforehand: + • Ask who will be in the audience? (Are they parents, teachers, members of the + + business community, school board members, etc.) + + • What do they already know about the topic? Are they experts on the topic? + + Will this be brand new information? -recent procedures such as endoscopies, colonoscopies and the operative + • Is there a meeting or event planned before and/or after your talk? What other + topics will be discussed? What is the agenda for the meeting? How much time + do you have? -reports. This information will help you understand what you might see during +The more confident you are that you are presenting useful and interesting material for +your audience, the less nervous you will be overall. A great tip is to greet audience members +at the door and do a quick survey of why they are there and what they expect. This really +helps you build a rapport with the group. -the examination, allow you to anticipate certain findings and possible changes +Step 2 — Know Your Material: Nothing is worse for nerves than trying to give a +presentation on a topic that you do not know! Our modules will provide you with +information and tools for each of the topics; however, it is on you to make sure that you +take time to read the material and become familiar with each topic. Remember: Your +audience will ask you questions and will look to you as the expert. -that may need to be made in a “routine” study protocol. +Another important point to remember is that you cannot possibly cover everything you +know in your presentation. That would probably be long and boring. So select the most +pertinent points from your subject base and then supplement with other material if time +allows. A great tip is to make your material interesting and memorable, include occasional +questions to the audience to encourage audience participation. This enhances the learning +experience and gives you a break from presenting. It also allows you deliver your information +in a more conversational manner, which is often more believable. -2 +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking -• In some cases, you may note that the examinations which are ordered are + Step 3 — Structure Your Presentation: A common technique for trying to calm +nervousness is memorizing what you intend to say; however, all this does is make your +delivery sound like it is coming from a robot. If you miss a word or draw a blank, your whole +presentation is thrown off and then your nervousness gets worse. It is far better to structure +your presentation so that you give yourself clues to what is coming next. + • Have a set of key phrases listed on a cue card. + • Refer to these phrases to trigger your mind as to what is coming up next. + • If you’re using slides, use these key phrases in your transitions. -not in fact actually needed. Please check with your attending cancelling any +This approach helps you control your own uncertainty about whether you will remember +what you want to say and the order you want to say it. -of the cases on the schedule. +Step 4 — Calm Yourself from the Inside: Nervousness causes physiological reactions, +which are mostly attributed to the increase of adrenaline in your system. You can counteract +these effects with a few simple techniques: -• Once you have done your homework, present the cases to the GI/GU + Practice deep breathing – Adrenalin causes you to breathe shallowly. By breathing deeply + your brain will get the oxygen it needs and the slower pace will trick your body into + believing you are calmer. It also helps with voice quivers, which can occur when your + breathing is irregular. -attending and discuss what are you planning to do and what you may see in + Drink water – Adrenalin can cause a dry mouth, which in turn leads to getting tongue- + tied. Have a glass of water handy. Take sips occasionally, especially when you want to + emphasize a point. -the clinical setting. + Smile – This is a natural relaxant that sends positive chemicals through your body. -• After performing the study, send the pictures via PACS. Call the attending + Use visualization techniques – Imagine that you are delivering your presentation to an + audience that is interested, enthused, smiling, and reacting positively. Cement this + positive image in your mind and recall it right before you are ready to go on. -to review the images. The attending may ask you to obtain more images. DO +Here are a few more calming presentation tips: + • Press and massage your forehead to energize the front of the brain and speech center. + • Just before you start talking, pause, make eye contact, and smile. This last moment of + + peace is very relaxing and gives you time to adjust to being the center of attention. -NOT SEND THE PATIENT without reviewing the images with an attending. + • Speak more slowly than you would in a conversation, and leave longer pauses between + sentences. This slower pace will calm you down, and it will also make you easier to + hear, especially at the back of a large room. + + • Move around during your presentation. This will expend some of your nervous energy. + • Remember that the audience is there to get some information and it is your job to put it + + across to them. + +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking + + When it comes to presenting, nerves are inevitable. Letting them get the better of you is not. +You need to develop a strategy for taking the focus off your nervousness and putting that +energy to positive use. By controlling as much of the uncertainly as you can, you increase +your confidence in your ability to deliver an excellent presentation. This confidence then +counteracts your nerves, and you create a positive cycle for yourself. + +Nerves are not your enemy and you don’t have to fear public speaking. For your next +presentation, be knowledgeable, be well-practiced and prepared, and try out some physical +relaxation techniques. Amaze yourself and impress your audience with your calm and +cool delivery of a great presentation. -If the GI/GU attending has stepped out ask for an alternate attending. +Grassroots Advocacy Trainer’s ManualModule 9: Presentation Skills and Public Speaking Grassroots Advocacy Trainer’s Manual -• You may be requested to add additional cases onto the schedule, before + How much does your audience know about the presentation's subject? (Make sure +that your presentation matches their knowledge level – don't make it too +complicated or too simple for their needs.) -adding cases check with your attending. - -The remainder of this packet will contain your educational goals, guidelines, and - -most common procedural protocols as well as a list of references to guide your - -reading for this month. - - COMPETENCY-BASED GOALS AND OBJECTIVES - -1. Medical Knowledge - -a. Instrumentation and Protocols - -i. Basic fluoroscopic instrumentation - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -3 - -A. The first time you are on fluoro – ask your attending for - -an introduction to the fluoro tower and buttons - -ii. Common fluoroscopic artifacts - -iii. Become familiar with common fluoroscopic protocols - -iv. Types of oral contrast - -b. Bowel - -i. Normal and abnormal appearances of the esophagus, stomach, - -small bowel and colon on fluoroscopy (details to follow in the - -packet) - -c. Gynecology - -i. Normal and abnormal appearance of the uterus and uterine - -tubes on hysterosalpingography (HSG) - -d. Urethra and Bladder - -i. Normal and abnormal appearance of the urethra on retrograde - -urethrogram (RUG) - -ii. Normal and abnormal appearance of the urinary bladder and - -urethra on voiding cystourethrogram (VCUG) - -2. Patient Care - -a. Gather clinical information about patients relevant to the - -interpretation of the examination as described in the introduction. - -b. Communicate with the patient prior the procedure. Introduce - -yourself, including the fact that you are a radiology resident. Let - -them know what you will be doing and what you are looking for. - -Demonstrate caring and respectful behavior. - -i. In your pediatric fluoroscopy rotation – you will be explaining - -the procedure to the patient’s parents who will be present for - -the entirety of the study. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -c. Use of information technology to support patient care decisions – IE - -Look up things you don’t know! - -d. Effectively perform fluoroscopic procedures in a way that minimizes - -patient discomfort but answers the clinician’s question. - -e. If an attending is not present and if asked or you sense the patient or - -family member is uncomfortable with the lack of direct supervision - -ask the attending to participate. - -f. If you are uncomfortable with the procedure or concerned with - -findings that may need further evaluation, inform the attending. - -4 - -3. Practice-Based Learning and Improvement - -a. Participate in self-learning including outside reading, Statdx, - -radiopedia and articles found in Radiology and Radiographics. See the - -list provided. - -b. Participate in QI/QA activities – watch out for artifacts. Remember - -to consider fluoro time and reduce dose by collimating etc. - -4. Interpersonal and Communication Skills - -a. Dictate prompt, concise, accurate and concise radiological reports. - -b. Promptly communicate urgent, critical or unexpected findings to - -residents, referring physicians or clinicians and document the - -communication in the report. (Name of clinician, findings discussed, - -time and date). - -5. Professionalism - -a. Demonstrate respect and compassion to patients, physicians, and - -staff. - -b. Demonstrate positive work habits, such as punctuality. - -c. Commitment to ethical principles and patient confidentiality. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -6. Systems Based Practice - -a. Understand how medical decisions affecting patient care within the - -larger system. - -b. Demonstrate knowledge of appropriateness criteria. - -5 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -FLUOROSCOPIC GI/GU STUDIES - -6 - -• Esophagram/Barium Swallow - -• Modified Barium Swallow (performed with Speech Pathologist) with or - -without limited esophagram - -• Upper GI examination (UGI) - -• Small bowel follow through (SBFT) - -• Post-surgical: Esophageal repair/stent, POEM, Gastric pull-up, Whipple, - -Gastric Bypass - -• Barium Enema Examination - -• Air contrast Enema for Intussusception Reduction - -• Fistulogram - -• Voiding Cystourethrogram - -• Retrograde Urethrogram - -• Fluoroscopic Assistance for Hysterosalpingogram - -*This list contains the most commonly performed fluoroscopic studies at Shands - -Hospital, although this is not exhaustive of the procedures performed in the - -Fluoroscopic Suite, it should suit your needs for this rotation* - -• Radiologic placement of an enteric tube (RARE – and must make sure that all - -the criteria for floor placement and Core-trak attempt have been performed - -– please see document by Dr. Abbitt below) - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -7 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -OVERVIEW OF CONTRAST AGENTS - -8 - -There are a few types of contrast that are used in Fluoroscopy: - -1. Barium Sulfate: is a relatively inert agent; however it can incite an - -inflammatory response if spilled into the MEDIASTINUM or PERITONEAL - -CAVITY. Use barium cautiously if there is a colonic obstruction. - -a. Consult Peds attending if there is a question. - -i. Peds surgery prefers Barium over water soluble. - -ii. Peds tends to use Barium in most cases, except in the post-op - -state. - -2. Iodine-based contrast agents: - -a. High Osmolality Agents (HOCM) – include agents such as Gastroview - -and Gastrograffin (never used in kids!). These tend to be flavored - -(not well!) and are for enteric use ONLY. All of these agents are ionic. - -i. WARNINGS: HOCM agents can cause bronchospasm, pulmonary - -edema, pneumonitis and are CONTRAINDICATED IN THOSE - -PATIENTS WITH TRACHEAL ASPIRATION OR - -TRACHEOESPHAGEAL FISTULA (TEF). - -1. Do not use these agents in patients who are status post - -esophageal interventions! Please use LOCM agents. - -2. These agents are hyperosmolar – so do not use them in - -the elderly or infants – they will draw fluid into the bowel - -lumen. - -b. Low Osmolality Agents (LOCM) – include agents such as Omnipaque - -and Isovue. LOCM agents include both ionic and NON-ionic compounds. - - HELPFUL HINTS FOR PATIENT POSITIONING - -9 - -Meet the Fluoro tower… - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -Picture Reference: http://pubs.rsna.org/doi/pdf/10.1148/radiographics.20.4.g00jl301115 - -(*Suggest reading*) - -We will orient the patient in reference to the X-ray Table. The patient is in the - -supine position above. - - The following images are copied from: - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -10 - -Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical - -Publishers Inc.; 1996. - - Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical - -Publishers Inc.; 1996. - -11 - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - - Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical - -Publishers Inc.; 1996. - -12 - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - - REFERENCES - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -13 - -There are a wide range of useful articles that can be used to learn about the - -procedures and imaging findings of certain pathologies. These articles are listed - -under suggested reading which can be found after each procedure. However, there - -are a few books which I do suggest that you use as your primary references for - -“how –to” and for pathologic findings. - -Books: - -1. Houston JD, Davis M. Fundamentals of Fluoroscopy. 1st ed. Saunders, 2001. -2. Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical - -Publishers Inc.; 1996. - -3. Johnson, CD, Schmit, GD. Mayo Clinic Gastrointestinal Imaging Review. - -Rochester, MN: Mayo Clinic Scientific Press, 2005. - -Journal Articles - -1. Geise, R. Fluoroscopy: Recording of Fluoroscopic Images and Automatic - -Exposure Control. Radiographics 2001; (21): 227-236. - -http://dx.doi.org/10.1148/radiographics.21.1.g01ja19227 - -2. Schueler, BA. General Overview of Fluoroscopic Imaging. Radiographics - -2000; 20 (1) 1115-1126. - -http://pubs.rsna.org/doi/abs/10.1148/radiographics.20.4.g00jl301115 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -14 - -ESOPHAGRAM/BARIUM SWALLOW - -Most Common Indications: Dysphagia, odynophagia, globus sensation - -Patient Preparation: NPO for a minimum of 2 hours - -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast), 12.5 mm Barium - -Tablet - -*Please note that if this exam is being done in a patient who is post-operative or - -there is a question for perforation you will be using WATER SOLUBLE CONTRAST. - -Additionally, if the patient is at risk for both PERFORATION and ASPIRATION, a - -low osmolar intravenous agent can be administer orally without the risk of - -hypertonic pneumonitis.* - -Procedure: - -In the North tower Fluoroscopy suite, please note that most exams are done for - -inpatients and therefore techniques that you may see at the VA or in Plaza may - -differ. For example in the VA or Plaza you may see the use of “fizzies.” This allows - -us to create “double contrast” and allows the radiologist to evaluation the mucosa - -of the esophagus. Since, most of the patients we do here at Shands are post- - -operative we do not typically use “fizzies” or the double contrast technique. - -However if mucosal evaluation is needed we will use “fizzies” per attending - -discretion. - -Furthermore, “fizzies” are not used in the pediatric population. - - Adults: - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -15 - -1. Start the procedure by explaining to the patient exactly what you are going - -to do today. - -2. Begin with the adult in the standing position facing you. Since many of our - -patients are post-operative or frail they will not be able to stand. In that - -case the study can be performed in the supine or reverse Trendelenburg - -position. - -3. Fluoro the chest and remove any artifacts – such as clips and leads from the - -patient. - -4. Obtain a scout radiograph of the chest. - -a. ***IF WE ARE USING FIZZIES, PLEASE ADMINISTER THEM TO - -THE PATIENT AT THIS POINT*** - -i. Have the patient take the “gas crystals” like they are doing a - -shot, into the back of their throats, and have them chase it - -with water. - -ii. INSTRUCT THE PATIENT NOT TO BURP!! The gas is what - -gives us the contrast for mucosal evaluation. - -5. Ask the patient to take a mouthful of the “thick” barium” and watch the - -patient swallow. For the first swallow you do not have to obtain any images, - -this is for you to understand the patient’s motility and ability to perform - -study. - -6. Now position the patient in the LPO position standing and have them swallow - -a mouthful of barium (3 swallows for your 3 shots). - -a. Shoot the proximal (cervical) esophagus. You may grab or shoot a full - -exposure. - -b. Shoot the mid thoracic esophagus. - -c. Shoot the distal thoracic esophagus. - -*Please note the following images are obtained with the double contrast - -technique outlining the mucosa* If there is a mucosal question, the exam - -should be performed with “fizzies” – please see addendum at the end of this - -protocol. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -16 - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -HELPFUL HINT 1: Also when, obtaining images, remember if there is - -something that looks like a stricture – have the patient drink and either - -demonstrate that the finding remains, or the contrast clears and that it was - -likely peristalsis. - -HELPFUL HINT 2: If you do see something abnormal remember this is what - -are evaluating, Take pictures, magnify your image etc. But DO KEEP AN EYE - -on your fluoro time! - -7. We will now lie the patient down on the table and place them in the right - -lateral position and obtain lateral pictures of the esophagus (obtain a column - -of barium in the esophagus). You may use “thin barium” as you will have the - -patient taking multiple sips quickly or “chugging the barium”. - -a. Center first over the upper thoracic esophagus. - -b. Include a shot of the open lower esophageal sphincter (LES) on the - -last two views that you take - -c. Obtain a magnified image of the LES. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -17 - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -8. Now we will evaluate for reflux. - -a. Place the patient in the RPO position, this will allow the contrast to - -pool into the cardia of the stomach. In peds, fluoro intermittently for - -3-5 minutes to evaluate completely. - -i. To illicit reflux, there are three maneuvers that one can apply: - -1. Observe the GE junction while the patient performs a - -Valsalva maneuver and coughs. - -2. Watch while the patients rolls over onto the right lateral - -position. - -3. “Water siphon test” – have the patient take two sips of - -water in the RPO position. Watch the water go down the - -GE junction and check if there is any reflux. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -ii. If you observe the patient refluxing, it is important to note - -three things: - -18 - -1. The maneuver that elicited the reflux. - -2. How far patient refluxed (cervical, upper, mid, distal - -thoracic esophagus). - -3. Time it takes for the esophagus to clear the reflux. - -9. Now we will examine the pharynx by centering on the pharynx. The patient - -will be standing (if able) for this portion of the exam. - -You may use 1-2x magnification for this portion of the exam. - -a. Have the patient take a few sips of thick barium and ask them to - -“gargle” with the barium. Instruct the patient that you are trying to - -“coat their throat” with the barium so that we may outline their - -anatomy. - -b. With the patient’s throat coated we will perform a series of - -maneuvers to demonstrate the anatomy of the pharynx. - -i. Ask the patient to hold their chin up - -ii. Take a scout image - -iii. Now ask the patient to say “EEEEEE” and take an image - -iv. Now have the patient PUFF OUT their cheeks, take an image - -v. Repeat the above maneuvers with the patient in the lateral - -position. - -The following images demonstrate your scouts in the AP and Lateral positions - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -If you do see abnormalities on this portion of the exam – obtain a CINE like - -study. Ask the patient to hold the contrast in their mouth– change the film - -speed on the tower to 3 frames/second and have the patient swallow on the - -count of three and obtain images. - -19 - -10. Complete the exam by having the patient swallow the barium pill. - -a. The barium tablet is typically 12.5mm in size as strictures that are - -less than 13 mm are the ones that cause dysphagia in patients. - -b. Strictures ranging in size from 13-19mm may or may not cause - -symptoms. - -11. Demonstrate where the barium pill ends up. - -a. If the pill gets “caught” – have the patient drink water to move the - -pill along. - -b. The pill will eventually dissolve so reassure the patient that it will not - -be stuck. - -12. Now check the images with your attending and repeat any portions of the - -exam that your attending may feel are unclear. - -13. Upon completion of the exam, feel free to let the patient know if the - -preliminary findings, noting of course that the final read will be send to - -their ordering physician. - -Peds - -The above can be used as guidelines in the pediatric population; however, often - -times the indications are different. As mentioned above, double contrast is not - -used in pediatric patients. - -Make sure the TECH has used pediatric settings on the fluoro tower and there is - -pulsed fluoro. - -You will NOT take full exposures, but instead will use FLUORO Grabs, last image - -hold saves. - - You will NOT use MAGNIFICATION. - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -20 - -Also, we do not typically exam the pharynx in peds, the pharynx is usually examined - -in the setting of a modified barium swallow study with a speech pathologist. - -Suggested Reading: - -1. Levine, MS, Rubensin SE. Diseases of the Esophagus: Diagnosis with - -Esophagography. Radiology 237(2): 414-427. - -http://geiselmed.dartmouth.edu/radiology/resources/fluoroscopy/pdf/esop - -hageal_disease.pdf - -2. Levine, MS. Radiologic Evaluation of the Esophageal Phase of Swallowing: In - -Levine, MS. Manual of Diagnostic and Therapeutic Techniques for Disorders - -of Deglutition. http://link.springer.com/chapter/10.1007%2F978-1-4614- - -3779-6_4 - -3. Leudtke, PA et al. Radiologic Diagnosis of Benign Esophageal Strictures: A - -Pattern Approach. Radiology 23 (4): 897-909. - -http://pubs.rsna.org/doi/pdf/10.1148/rg.234025717 - -4. Gupta S, Levine MS, Rubesin SE, Katzka DA, Laufer I. Usefulness of barium - -studies for differentiating benign and malignant strictures of the - -esophagus. AJR 2003; 180:737-744. - -5. Baker et al. Gastroesophageal Reflux Disease: Integrating the Barium - -Esophagram before and after Antireflux Surgery. Radiographics 243 (2): - -329-339. http://pubs.rsna.org/doi/pdf/10.1148/radiol.2432050057 - -Websites: - -http://www.radiologyassistant.nl/en/p472458f15c55a/esophagus-part-i.html - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -MODIFIED BARIUM SWALLOW WITH LIMITED ESOPHAGRAM - -21 - -For this exam we are providing fluoroscopic assistance to the Speech Pathologist - -as they evaluate the patient’s ability to swallow a variety of consistencies of foods - -coating in barium. - -1. Introduce yourself to the speech pathologist as the radiology resident on - -service. - -2. Discuss what the expected imaging findings are. - -3. Discuss the speech pathologist expectations of the study and whether or - -not a limited esophagram will be performed. - -4. Introduce yourself to the patient as the radiologist who will be assisting the - -Speech Pathologist for this exam. - -5. Line up the patients head/neck under fluoroscopy, remember to collimate to - -the region of interest. - -6. Make sure the patient’s eyes are NOT included in your field. - -7. Observe the Speech Pathologist giving the patient a variety of consistencies. - -As the radiologist, it is your job to note any instances of aspiration or - -penetration on the exam and what consistency of barium they occurred with. - -8. Observe the patient for any other pathologies, such as premature slippage, - -delayed swallow etc. - -9. If there are any unexpected findings not related to a swallow study that you - -feel would warrant more imaging. Please save shots and discuss them with - -your attending at the time of interpretation. - -10. The Speech Pathologist may ask you to perform a limited Esophagram. - -a. This means we will observe the esophagus in ONLY ONE VIEW. If - -there is pathology – please take an image to document. We will then - -suggest to the clinicians to perform a full diagnostic esophagram. - -11. Discuss the case with the Speech Pathologist upon completion and note any - -findings that he/she may add to your interpretation. - -12. It is not necessary to review the images with your attending prior to sending - -the patient. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. However, take down any notes that you feel are necessary for - -discussion with your attending at the time of interpretation. - -22 - -Suggested Reading: - -1. Matsuo, K et al. Anatomy and Physiology of Feeding and Swallowing – Normal - -and Abnormal. Phys Med Rehabil Clin N Am. 2008 Nov; 19(4): 691–707. - -http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2597750/ - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -UPPER GI SERIES (Back) - -23 - -Most Common Indications: Epigastric pain, anemia, heme+ stools, GERD, hiatal - -hernia - -Patient Preparation: NPO after midnight, or at least 4-6 hours. - -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast) - -*Please note that if this exam is being done in a patient who is post-operative or - -there is a question for perforation you will be using WATER SOLUBLE CONTRAST. - -Additionally, if the patient is at risk for both PERFORATION and ASPIRATION, a - -low osmolar intravenous agent can be administered orally without the risk of - -hypertonic pneumonitis.* - -Procedure: - -1. Start the procedure by explaining to the patient exactly what you are going - -to do today. - -2. Begin with the adult in the standing position facing you. Since many of our - -patients are post-operative or frail they will not be able to stand. In that - -case the study can be performed in the supine or reverse Trendelenburg - -position. - -3. Fluoro the chest and abdomen and remove any artifacts – such as clips and - -leads from the patient. - -4. Look for four things: - -a. Full stomach (cannot evaluate stomach if the stomach is full) - -b. Bowel obstruction - -c. Free air - -d. Barium present in small bowel or colon - -5. ***IF WE ARE USING FIZZIES, PLEASE ADMINISTER THEM TO THE - -PATIENT AT THIS POINT*** - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. Have the patient take the “gas crystals” like they are doing a shot, - -into the back of their throats, and have them chase it with water. - -b. INSTRUCT THE PATIENT NOT TO BURP!! The gas is what gives us - -24 - -the contrast for mucosal evaluation. - -6. Ask the patient to take a mouthful of the “thick” barium” and watch the - -patient swallow. For the first swallow you do not have to obtain any images, - -this is for you to understand the patient’s motility and ability to perform - -study. - -7. Now position the patient in the LPO position standing and have them swallow - -a mouthful of barium (3 swallows for your 3 shots). - -a. Shoot the proximal (cervical) esophagus. You may grab or shoot a full - -exposure. - -b. Shoot the mid thoracic esophagus. - -c. Shoot the distal thoracic esophagus. - -8. You may need to instruct patient to drink more barium if you do not feel - -there is enough in the stomach. - -9. Next we will move the patient into the supine position. - -a. If we are using the “fizzies” (ie doing a double contrast exam) and - -looking at mucosa please do the following: - -i. Instruct the patient that we are attempt to coat their stomach - -with barium (this is important if we are looking at the mucosa - -using the fizzies technique). - -ii. We will ask the patient to roll through the left lateral position - -to the prone postion and back (“roll like a cement mixer”). - -10. Capture images of the stomach in the supine position. - -11. Now we will obtain contrast filled images of the stomach. Please see the - -images below for guidance. - -a. Shoot the cardia and fundus of the stomach - -b. Shoot the mid body of the stomach outlining the greater curvature - -c. Shoot the antropyloric region of the stomach (if you miss this here - -don’t worry you’ll get it when you evaluate the duodenal bulb) - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -25 - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -Although not pictured above, please obtain images of a contrast filled cardia, as - -attempted in the first image (labelled two) above. - -12. Now we will watch the duodenal bulb (spade shaped). If performing a double - -contrast exam (ie with fizzies). Please follow the instructions below - -a. Have the patient moved into the LPO position, and try to capture an - -air-filled duodenal bulb. It may help to initially place the patient in a - -left lateral decubitus position, and gently roll them into an LPO - -position while coughing to capture an air-filled bulb. Note if you - -cannot obtain this come back to this LATER. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -26 - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -13. For the standard contrast filled duodenal bulb we will have the patient move - -into the RPO to Right lateral position. This encourages the contrast to move - -through the stomach to the gastroduodenal junction and for contrast to - -move through the duodenal C-loop. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -27 - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -14. Once you see the C-loop of the duodenum place the patient in the supine - -position. - -a. The contrast will flow across the third and fourth portions of the - -duodenum and will cross the Ligament of Treitz (In a peds case you - -are done here!). - -i. Take this picture - -ii. You should also have a good picture of a contrast filled stomach - -and be able to evaluate the proximal small bowel - -15. Now test for reflux (please see guidelines for the Esophagram). - -16. Obtain a picture of the standing stomach and proximal small bowel as your - -final image (full exposure on adults). - -17. *If there was a request to evaluate for distal stricture you can give the - -patient the barium pill and demonstrate it passing through to the stomach. - -18. Otherwise, review your images with your attending and repeat any portions - -that are needed. - -19. Discuss the preliminary findings with the patient if desired. - - Peds - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -28 - -Overall the peds upper GI is a quick study to typically rule out malrotation, - -evaluate equivocal cases of HPS, duodenal atresia/web and/or diagnose GERD. - -We will always use the single contrast technique and fluoro grabs to obtain any - -pictures with the pediatric fluoro settings. - -We will examine the esophagus in the AP and lateral planes with fluoro grabs as the - -child/infant is drinking. Typically given patient size we can fit the stomach on our - -views (without having to move the patient and obtain multiple pictures). The most - -important part of the study is document the normal positioning of the ligament of - -Treitz. - -Be careful when you have the baby in the lateral positioning the quickly move the - -infant supine to watch the contrast cross over the ligament of Treitz. - -Suggested Reading: - -1. Rubensin et al. Double-Contrast Upper Gastrointestinal Radiography: A - -Pattern Approach for Diseases of the Stomach. Radiology 2008; 246 (1): 33- - -48. http://pubs.rsna.org/doi/pdf/10.1148/radiol.2461061245. - -2. Rubensin et al. Upper Gastrointestinal Tract: Imaging Techniques. In: - -Rubensin et al. Abdominal Imaging. Springer Verlag 2013. - -http://link.springer.com/referenceworkentry/10.1007%2F978-3-642- - -13327-5_13 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -Small Bowel Follow Through (SBFT) - -Most Common Indications: Crohn’s Disease, heme+ stools, abdominal pain NOS - -In peds: This study is often done in post-operative patients who have had bowel - -resected or due to atresias/strictures or webs. It is also performed in children - -29 - -with gastroschisis repairs. - -*Please note this exam is different than the Gastrograffin Challenge which is - -often ordered in the setting of potential bowel obstruction. The Gastrograffin - -Challenge (which is a misnomer, as we do not use Gastrograffin) is a series of - -abdominal x-rays which are taken after the administration of contrast (by the - -tech) documenting the passage of contrast through the bowel.* - -Patient Preparation: NPO after midnight - -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast can be used but - -often gets “washed out” as it is not as dense as barium) - -Procedure: - -This study is typically performed just following an upper GI. In the majority of - -case CT or MR enterography is used when evaluation of ONLY the small bowel is - -requested. - -1. Please see preceding protocol for how to perform an UPPER GI and obtain - -the proper scout radiograph. - -2. We will begin our instruction after the contrast has crossed the ligament of - -Treitz. - -3. Obtain radiographs (fluoro grabs in peds, although may consider spots in - -children with known motility disorders to reduce radiation) in approximately - -30 minutes intervals. If there is extremely slow transit as may be seen in - -post-surgical infants you may increase this interval of time between - -radiographs. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. During the interval time the patient may go ahead and eat and walk - -around the hospital to encourage bowel transit. - -30 - -4. Once the contrast has reached the level of the ascending colon, for an adult - -or outpatient examination we will go and specifically exam the terminal ileum - -(esp for those with Crohns). - -a. Obtain the balloon paddle and compress “spot” the terminal ileum - -(fluoro grab or spot the image). You may even see the appendix. - -5. Compress the small bowel in all four quadrants of the abdominal and make - -sure that there appears to be normal compressibility and that the bowel - -does not appear “stuck.” - -6. You should at least fluoro grab images of the four quadrants. - -7. Review all images with your attending. Upon his/her approval you may let the - -patient go. - -Suggested Reading: - -1. Levine, MS et al. Pattern Approach for Diseases of Mesenteric Small Bowel - -on Barium Studies. Radiology 2008; 249 (2): 445-460. - -http://pubs.rsna.org/doi/pdf/10.1148/radiol.2491071336. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -POST-SURGICAL CASES - -31 - -We see a wide variety of patients here at Shands who often undergo complex - -surgeries which need fluoroscopic evaluation for complications. The following are a - -series of post-operative cases that we do see and some helpful hints for how to - -complete the requested studies. - -For all surgical studies: REMEMBER to obtain a scout image of the area of study - -prior to beginning any study. There are often devices and sutures that we need to - -be aware of. ALSO, upon completion of the study obtain a POST radiograph. - -Esophageal Studies: - -• These are often requested after esophageal repair, stenting, POEM - -procedure or gastric pull-through/gastric pull-up. The goal of these studies - -is to evaluate for leak. Therefore, distention of the esophagus is paramount - -and the patient should be encouraged to take large gulps of contrast. - -• The esophagus is typically dysmotile. However, leak and contained leak are - -often noted as well. - -• In this case, you will perform a single contrast water soluble exam of the - -esophagus with the post-surgical patient lying down on the fluoro table. - -• Attempt to view in the esophagus in at least the AP and Lateral planes. - -However, oblique views may reveal subtle leaks if you are able to position the - -patient. Obtain mag views if you do demonstrate a leak. Note any areas of - -stricture or narrowing. If the narrowing is transient demonstrate a fully - -distended esophagus. - -Fluoroscopic Evaluation of Whipple: - -Whipple's operation is most often performed in patients with ductal - -adenocarcinoma of the pancreas. The duodenum, part of or entire pancreas, and - - distal common bile duct are resected. Mostly commonly, the common bile duct, - -pancreas, and proximal duodenum are anastomosed to a loop of jejunum. - -32 - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -Diagram of the Post-surgical appearance of the Whipple - -Remember to use LOCM contrast and attempt to distend the anastomoses. Obtain - -views in the AP, lateral and oblique planes. If the patient has an NG tube - -remember to remove the contrast from the stomach as these patients are an - -aspiration risk. - - Fluoroscopic Evaluation of Gastric Bypass - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -33 - -Picture Reference: - -Left: Levine, MS et. al. . Imaging of Bariatric Surgery: Normal and Post-Operative Complications. Available at: - -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520. - -Right:Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -Typically for gastric bypass, a pouch is formed at the cardia of the stomach. A - -jejunal Roux-limb is brought through the transverse mesocolon. A side to side - -anastomosis is gastrojejunostomy is created. - -1. Obtain digital scout film of left upper quadrant. - -2. In a standing lateral position, have the patient swallow one swallow of water- - -soluble contrast and fluoro the lateral pharynx. If there is no aspiration, - -place patient erect AP and have them drink a few swallows of contrast, - -centering over left upper quadrant. Watch for emptying and extravasation. - -Table spot image over pouch. - -3. Then place the patient in the supine position (make sure the patient does not - -exceed the table weight limit!). Obtain 3 to 4 digital spot images of the - -gastric pouch and Roux-limb in several obliquities. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. Do not magnify these images as most patients are so large that - -exposure time would be long and result in motion artifact. - -34 - -4. Don’t forget to get delayed images of the upper abdomen to confirm - -contrast emptying into the distal jejunum, beyond where the Roux-limb - -courses through the transverse mesocolon. - -a. Is there a leak? This usually occurs at the gastrojejunostomy - -anastomosis or the blind jejunal stump. - -b. Is there obstruction? Usually at gastrojejunostomy, as the Roux-limb - -traverses the mesocolon, or at the jejunojejunostomy, - -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520 - -Suggested Reading: - -1. Levine, MS, Carrucci, LS. Imaging of Bariatric Surgery: Normal and Post- - -Operative Complications. Radiology 2014; 270 (2): 327-341. - -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520. - -2. Swanson JO, Levine MS, Redfern RO, Rubesin SE. Usefulness of high- - -density barium for detection of leaks after esophagogastrectomy, total - -gastrectomy, and total laryngectomy. AJR 2003;181:415-420. - -3. Gervais, DA et al. Complications after Pancreatoduodenectomy: Imaging and - -Imaging-guided Interventional Procedures. RadioGraphics 2001; 21:673–690. - -http://geiselmed.dartmouth.edu/radiology/pdf/complications_radiographics. - -pdf. - -4. Hamath et al. Fluoroscopic Findings post-peroral esophageal myomectomy. - -Abdominal Imaging 40 (2): 237-245. - -http://link.springer.com/article/10.1007%2Fs00261-014-0209-7 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -35 - - SINGLE CONTRAST ENEMA - -Indication: Failed colonoscopy, stricture, mass, fistula evaluation (i.e. - -rectovaginal/rectovesicular), colostomy reversal. The ordering physician should - -consult with the Fluoro Radiologist to discuss whether CT with rectal contrast or - -fluoroscopic enema is the most appropriate exam given patient history and mobility. - -The contrast agent (barium or dilute Omnipaque) is protocolled by the radiologist - -performing the fluoroscopic study. - -Please note that Radiology will not perform a therapeutic “gastrograffin” enema, -and is not permitted to dispense MD Gastroview for this purpose, nor is MD -Gastroview ever used as a contrast agent in Radiology with the exception of the -Gastrograffin Challenge Small Bowel Study. - -Patient Preparation: Requires a preceding bowel clean-out beginning the day before - -the procedure and NPO on day of procedure. No prep is done in children. - -Materials: Enema bag with attached catheter tubing with rectal tube tip, barium or - -water soluble contrast (in adults we use the Polibar kit) - -**Please note the techniques used for a double contrast enema do vary slightly. - -However given colonoscopy, CT/MR enterography these are rarely performed. The - -single contrast enema is also very rare – given the ease of CT and rectal contrast- - -as stated above.** - -Procedure: - -1. Obtain a routine scout to evaluate stool burden and exclude any potential - -free air. - -2. Do a careful rectal examination – this will help you orient the proper - -direction of the anal canal and rectum as a guide to inserting the enema tip. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. This should also help you assess the presence of rectal mass or - -stricture. - -36 - -3. Insert the tip of the enema with the patient in the left lateral decubitus - -position with the knees drawn up towards the chest. - -4. Tape the buttocks together and ask the patient to hold their butt cheeks - -together for the entirety of the exam. - -5. We will now cover the technique for an adult patient – single contrast barium - -enema and review the necessary images. We will discuss the peds single - -contrast enema afterward with a concentration on differences in technique. - -a. Now allow the patient to roll into the supine position. - -b. Fill and distend the rectum with barium (shoot image) - -c. Roll the patient to the supine LPO and add barium until the column - -reaches descending colon. - -Picture Reference: - -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: - -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf - -d. Obtain a spot image of the sigmoid colon. - -e. Continue rolling the patient towards the right and obtain an image of - -the transverse colon – stop the barium. - -f. Then roll the patient RPO and spot the splenic flexure – stop the - -barium. - -g. Turn on barium and fill the ascending colon and look for reflux into - -the terminal ileum. Turn off the barium. - -h. You may use the paddle to spot the terminal ileum. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -6. For pediatric patients: - -a. In peds – usual indication of this study is for evaluation of - -Hirschsprungs’ or Meconium plug – so we will begin with evaluation of - -the rectum in the left lateral position. - -i. In the case of evaluation for meconium plug make sure that you - -use Cysto-Conray (not barium!). - -37 - -Picture Reference: - -Neville, HL et al. Pediatric Hirschsprung Disease Workup. Available at: http://emedicine.medscape.com/article/929733- - -workup#c4 - -b. We will then obtain fluoro grabs of the same as the adult patients - -(but only grabs! No spots!) - -7. Once you feel certain that you have obtains all the images necessary, review - -them with your attending and respot any necessary. - -8. Unclamp the barium and lower the enema bag so that the contrast drains by - -gravity. - -9. Obtain a post radiograph. - - Suggested Reading: - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -38 - -1. Rubensin, SE et al. Double-Contrast Barium Enema Examination Technique. - -Radiology 2000; 215:642–650. - -http://pubs.rsna.org/doi/pdf/10.1148/radiology.215.3.r00jn36642. - -2. Levine, MS et al. Diagnosis of Colorectal Neoplasms at Double-Contrast - -Barium Enema Examination. Radiology 2000; 216:11–18. - -http://pubs.rsna.org/doi/pdf/10.1148/radiology.216.1.r00jl3311 - -3. Greenall, MJ et al. Complications of diverticular disease: A review of the - -barium enema findings. Gastrointestinal Radiology 1983;8:353-358. - -http://link.springer.com/article/10.1007%2FBF01948150#page-1 - -Suggested Reading on Peds: - -1. Berrocal et al. Congenital Anomalies of the Small Intestine, Colon, and - -Rectum. Radiographics 1999; 19 (5): 1219-1236. - -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.19.5.g99se041219. - -2. Laya et al. Patterns of Microcolon: Imaging Strategies for Diagnosis of - -Lower Intestinal Obstruction in Neonates. J Am Osteopath Coll Radiol 2015; - -4 (1): 2-11. http://www.jaocr.org/articles/patterns-of-microcolon-imaging- - -strategies-for-diagnosis-of-lower-intestinal-obstruction-in-neonates - -Websites: - -Hirschsprungs - -http://emedicine.medscape.com/article/409150-overview - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -AIR CONTRAST INTUSSUSCEPTION REDUCTION ***PEDS ONLY*** - -39 - -Indication: Intussusception seen on ultrasound, high clinical suspicion despite - -negative US - -Patient Preparation/Radiology Prep: Obtain left lateral radiograph to exclude the - -presence of intraperitoneal air, Make sure that surgical resident or fellow is - -present in the event of perforation. - -Materials: Catheter tubing, in peds we use feeding tubes ranging from 5-8 Fr - -depending on the patient size, sphygmomanometer - -Procedure: - -1. Insert the rectal tube and begin taping the buttocks. - -a. You can use 4x4 gauzes for additional reinforcement - -2. Obtain a scout image hold. - -3. Make sure the catheter tubing is well connected to the sphygmomanometer. - -4. You may pump the sphygmomanometer up to 120 mmHg in attempts to reduce - -the intussuscepted bowel. - -5. Successful intussusception means seeing air in the distal ileum. - -6. If intussuscepted bowel recurs – you may try up to 3 times or for 10 - -minutes. - -7. Upon completion obtain a grab to document success. - -Suggested Reading: - -1. Del-Pozo, G. et al. Intussusception in Children: Current Concepts in Diagnosis - -and Enema Reduction. Radiographics 1999; 19: 299-319. - -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.19.2.g99mr14299 - -2. http://emedicine.medscape.com/article/930708-overview - -3. Applegate, KE. Clinically Suspected Intussusception in Children: Evidence- - -Based Review and Self-Assessment Module. Journal of Roentgenology. - - 2005;185: S175-S183. 10.2214/ajr.185.3_supplement.0185s175 - -http://www.ajronline.org/doi/full/10.2214/ajr.185.3_supplement.0185s175 - -40 - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -FISTULOGRAM - -41 - -Most Common Indications: Enterocutaneous Fistula - -Patient Preparation: None - -Materials: Small bore feeding tube, water soluble contrast - -Procedure: - -1. Use a small bore feeding tube to cannulate the fistula at the surface of the - -skin. - -2. Inject water soluble contrast under pressure. - -3. Make sure to turn the patient to document any fistula as well as the flow of - -contrast. - -4. Obtain spot images in supine, lateral and oblique views. - -Suggested Reading: - -1. Lee, JK, Stein SL. Radiographic and Endoscopic Diagnosis and Treatment of - -Enterocutaneous Fistulas. Clin Colon Rectal Surg. 2010 Sep; 23(3): 149–160. - -http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2967314/pdf/ccrs23149.pdf - -2. Pickhardt, PJ et al. Acquired Gastrointestinal Fistulas: Classification, - -Etiologies, and Imaging Evaluation. Radiology 2002; 224:9–23. - -http://pubs.rsna.org/doi/pdf/10.1148/radiol.2241011185 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -VOIDING CYSTOURETHROGRAM - -42 - -Most Common Indications: - -Peds: Vesicoureteral reflux, multiple febrile UTIs, posterior urethral valves, - -congenital abnormalities - -Adults: Post-operative evaluate for bladder injury, trauma, dysuria - -Patient Preparation: None. - -Materials: 5-8 French catheter tubing (PEDS), Foley Catheter, Cysto Conray or - -Omnipaque - -Procedure: - -1. Place the catheter/Foley using aseptic technique. - -2. Obtain a scout image prior to instilling contrast in the bladder (grab only for - -peds) - -3. Fill the bladder. - -a. For adults fill the bladder until the patient can no longer bear the - -presence of contrast - -b. Alternatively, when you are doing this in a pediatric patient you will - -use the cyclical technique of filling the bladder. - -i. IE the patient will pee around the catheter and you will fill the - -bladder a few times to evaluate for the presence of reflux. - -ii. Estimated bladder capacity in kids: (AGE + 2)30 = BLADDER - -CAPACITY (in milliliters/cubic centimeters) - -4. Obtain images of the full bladder in the supine, oblique and lateral positions - -(bilaterally). - -a. If you have any questions or concerns: STOP and discuss the case - -with your attending. In the next step we will remove the catheter. - -5. For adults: if the patient is able to urinate, you may remove the catheter - -prior to the urethra images. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -a. However, often times, patients with long term in-dwelling Foley - -catheters have difficulty urinating – so in those patients allow them to - -43 - -begin urinating and then remove the catheter. - -b. For males make sure the patient is in the oblique lateral position so - -that all of the urethra is clearly seen. - -i. Make sure all the portions of the urethra are seen. - -6. For little peds patients: - -a. Since we do not have control over when the patient will urinate – we - -have to be very fast and quick with our imaging. - -b. Make sure you are centered over the urethra (you only get one shot - -once the catheter is out, otherwise we will have to re-catheterize the -patient ) - -i. Make sure they are in the oblique lateral position to get - -optimal visualization of the urethra - -1. In males we are looking for posterior urethral valves! - -c. As the patient urinates, QUICKLY remove the catheter and obtain - -images of the contrast filled urethra. - -7. Try to have the patient urinate as much contrast as possible. - -a. Do not leave the patient with a large amount of contrast in the - -bladder without notifying the ordering physician. - -8. Obtain a post-void image (grab for peds). - -9. Review imaging and findings with attending before releasing the patient. - -Suggested Reading: - -1. Fernbach, SK. et al. Pediatric Voiding Cystourethrography: A Pictorial Guide. - -RadioGraphics 2000; 20:155–168. - -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.20.1.g00ja12155 - -2. Yu, NC. et al. Fistulas of the Genitourinary Tract: A Radiologic Review. - -RadioGraphics 2004; 24:1331–1352. - -http://pubs.rsna.org/doi/full/10.1148/rg.245035219 - -3. Cystography Technique Emedicine Description - -http://emedicine.medscape.com/article/1893772-technique - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -44 - -RETROGRADE URETHROGRAM - -Most Common Indication: Penile/Urethral trauma, urethral stricture, hematuria - -Patient Preparation: None - -Materials: Smaller bore foley catheter tubing 16-18 French, Omnipaque or - -CystoConray - -Procedure: - -1. Position the tower over the patient’s pelvis with the center focused just - -below the pubic bone. - -2. Flush the 16-F or 18-F Foley catheter with radiopaque contrast to remove - -any air bubbles. - -a. The penile glans and urethral meatus should be cleaned with - -antiseptic. - -3. The Foley catheter is then placed just inside the urethral meatus so that - -the Foley catheter balloon rests in the fossa navicularis. - -a. Some attendings will use a red rubber catheter and/or foley and do - -not inflate the balloon, but hold presseure at the glans to seal the - -tube. - -i. Ask your attending their preferred method. - -4. With the Foley in position, fill the catheter balloon with 1-2 mL of - -radiopaque contrast or saline solution. - -a. Overfilling must be avoided, or it will rupture the distal urethra. (A - -conscious patient can be asked if pain accompanies balloon filling). - -b. Pull the penis laterally (or ask the patient to) to straighten the - -urethra, grasping the penis as distally as possible, and distal to the - -inflated balloon. - -5. Obtain a pre-injection scout image. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -6. The catheter-tipped syringe is then filled with approximately 50 mL of - -radiopaque contrast, and 20-30 mL of contrast is injected in a retrograde - -45 - -fashion. - -a. Consider obtaining a fluoro loop of the injection – the tech may help - -you with the settings on the machine to do this. - -7. Static images of the urethra are taken during retrograde injection of - -radiopaque contrast. - -Picture Reference: - -Burks, FN et. al. Urethrogram. Available at: http://emedicine.medscape.com/article/1893948-overview - -8. You should see a jet into the bladder. Then you have completely evaluated - -the entirety of the urethra. - -9. Check images with your attending and obtain any other necessary images. - -10. When the exam is completely you may remove the catheter - -Suggested Reading: - -1. Urethrogram Emedicine website (see above) - -2. Kawashima, A. et al. Imaging of Urethral Disease: A Pictorial Review. - -RadioGraphics 2004; 24:S195–S216. - -http://pubs.rsna.org/doi/pdf/10.1148/rg.24si045504 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -3. Berrocal, T et al. Anomalies of the Distal Ureter, Bladder, and Urethra in - -Children: Embryologic, Radiologic, and Pathologic Features. RadioGraphics - -46 - -2002; 22:1139–1164. - -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.22.5.g02se101139 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -FLUOROSCOPIC ASSISTANCE FOR HYSTEROSALPINGOGRAM - -47 - -Most Common Indications: Tubal occlusion, tubal patency, endometrial polyp, - -evaluation for mullerian anomalies - -This procedure is done with the OB/GYN attending and it is your job to - -provide fluoroscopic assistance and obtain all the view necessary to - -document spillage or non-spillage of contrast from the Fallopian Tubes. - -The patient will be lying supine on the fluoro table. - -1. Obtain a scout image of the catheter once it has been inserted. - -2. Obtain a picture in the early filling phase of the endometrium to assess - -for subtle filling defects as well as a later phase when the endometrial - -cavity is full of contrast. - -3. Spot pictures of the contrast spilling from the fallopian tubes - -a. Obtain pictures with the patient in the supine and bilateral oblique - -positions. - -b. If you are ever unsure of contrast spillage from the fallopian - -tubes, please do a full exposure. Contrast is often easier to see on - -fully exposed images. - -4. Upon completion of the study confirm imaging findings with the OB/GYN - -attending. - -Suggested Reading: - -1. Simpson, WL et al. Hysterosalpingography: A Reemerging Study. - -RadioGraphics 2006; 26:419–431. - -http://pubs.rsna.org/doi/pdf/10.1148/rg.262055109 - -2. Chalazonitis A. et al. Hysterosalpingography: Technique and Applications. - -Current Problems in Diagnostic Radiology 2009; 38 (5): 199-205. - -http://www.sciencedirect.com/science/article/pii/S0363018808000200 - - 3. Behr, SC et al. Imaging of Müllerian Duct Anomalies. RadioGraphics 2012; - -32:E233–E250. http://pubs.rsna.org/doi/pdf/10.1148/rg.326125515 - -48 - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -RADIOLOGIC PLACEMENT OF ENTERIC TUBE - -WITH FLUOROSCOPIC GUIDANCE - -49 - -We may be asked to provide fluoroscopic assistance to clinicians in placement of - -enteric tubes – and even asked to place them ourselves. Please follow radiology - -protocol (see document by Dr. Abbitt – hung above your work stations before - -agreeing to assisting). - -Please note we do not stock enteric tubes in the Fluoroscopic suite. - -Below I will describe the placement of a nasojejunal tube with guidewire. - -1. Lubricate the guidewire well and assure that the wire and catheter fit - -together properly. - -2. If there is oral anesthetic spray available (Benzocaine topic anesthetic) - -spray the patient’s pharynx for about 1 sec duration. - -a. Have the patient swallow the material. - -3. Insert the guidewire into the tube leaving 10-20 cm floppy at the tip. - -4. Lubricate the end of the catheter with viscuous xylocaine (2%). - -5. With the patient sitting on the edge of the table, pass the catheter through - -the nose. - -6. Have the patient flex their neck, chin to the chest and move the catheter - -through the pharynx into the esophagus. - -7. Continue pushing the catheter until you feel resistance. - -8. Place the patient in the right lateral decubitus position and pass the tube - -with fluoroscopic guidance. - -9. Typically the catheter will pass into the antrum in the right lateral position. - -10. Then move the patient into the LPO position so that the catheter can pass - -into the pylorus and duodenal bulb. - -11. Keep the tip of the guidewire at the pylorus, slipping the catheter over the - -wire in the distal duodenum or proximal jejunum. - -12. When the tip makes it to the duodenojejunal junction tape the tube in place - -at the nose. - -13. Document your success with a radiograph. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -Note: Try not to exceed five minutes of fluoro time in your attempts. If the Fluoro - -alarm notifying you of the five minute mark rings, you may complete the study and - -notify the ordering physician that the procedure was not successful although - -50 - -multiple attempts were made. +Planning + +Where and how will you present (indoors, outdoors, standing, sitting)? Will this +affect how you need to prepare? + +What is going on before and after your presentation? Do you need to adjust your +presentation? + +How long do you have? Will your presentation fit in the time allotted? + +Are you knowledgeable enough about the topic that you're covering in your +presentation? If not, how will you build this knowledge? + +Can you visit the presentation room beforehand to get a feel for it? + +Does the presentation room have everything that you need? (For example, electrical +supply, Internet access, projector screen, and so on.) Do you need to bring your own +equipment? (You should always bring your own equipment even if you don’t use it, +as a back-up) +Do you want to inspire your audience to act? If so, how are you going to do this? + +Are you presenting as part of a team? How will you divide up the presentation? +Have you scheduled planning and practice time well in advance? +Have you prepared an evaluation? +Do you need to provide your own handouts and evaluations? How many copies do +you need? If someone else is making them, how will you get them the material? +How will you know they are ready? + + Does your introduction grab your audience's attention? And does it need to explain +your objectives? + +Content + +Do you follow this by clearly defining the points of the presentation? +Are the main points in a logical sequence? +Do these points flow well? +Do the main points need support from visual aids or props? + +Is your presentation too complex? (It's easy to go overboard, especially when you +want to impress - remember, keep it simple and focused.) + +Does your presentation contain any jargon or acronyms? (These can be confusing +for many people, so make sure that you define or eliminate jargon/acronyms from +your presentation.) + +Does your presentation's conclusion summarize the presentation clearly and +concisely? +Is the conclusion strong? +Have you tied the conclusion to the introduction? + +Have you practiced your presentation standing (or sitting, if applicable), paying +close attention to your body language and posture? +Have you rehearsed often enough to be able to speak smoothly and fluently? +Have you practiced your presentation in front of others? Your practice audience can +give you valuable feedback about your presentation. + + Power Point Presentation + +Is the power point easy to read/view and easy to understand? +Is it tied into the points that you're trying to communicate? + +Do your slides all look consistent? +Have you used the same fonts throughout? Are the slides easy to +read? Can they be seen easily from all areas of the room? + +Have you included too much wording? (Remember, slides are meant +to support points or concepts, not replace them. So, no sentences or +paragraphs!) +If you're representing an organization, do your visual aids and slides +match your organization's branding? + + Do you have all of the equipment you need? Laptop, projector, all +cords needed to hook it all up? Do you have spare projector bulbs, +just in case your existing bulbs blow? Do you have speakers if you +plan to show a video? +Have you prepared contingency plans in case your audio/visual +equipment fails? Do you have a printed copy so you can use it in the +event Power point is not an option? +Be sure to save your presentation on your laptop, a USB drive and +email it to yourself, just in case! +Even if all equipment is supposed to be provided; BRING YOU OWN. +You never know! + + The Day of your Presentation + +Do you have your slides, notes, and other visual aids in the right +order? +Are you dressed and groomed appropriately? (Make sure that this is +in keeping with your audience's expectations.) + +Have you left enough time for travel and setting up? +Have you checked your visual aids to ensure that they're working, +and that you know how to use them? + +REMEMBER: Did you provide a sign in sheet? Be sure to remind +everyone to sign in. (Ideally, your sign in sheet includes printed +name, email, phone number) +Do you know how to deal with nervousness? (Presentation nerves +are very common, so learn how to manage presentation nerves to +use that energy to your advantage.) +REMEMBER: Greet participants as they arrive if possible, helps to +establish rapport. +REMEMBER: During your presentation, make and maintain eye +contact with members of your audience? +Have you made sure that your audience understands everything that +you've covered? (Invite them to ask questions if you're unsure.) + +REMEMBER: Hand out evaluations Have you asked for feedback from +your audience? Is there anything that you could learn, to improve +your next presentation? +Do you need to follow up with any of your audience? Put a star by +their name on your sign in sheet. diff --git a/q119/random_k4/random_4.md b/q119/random_k4/random_4.md index 602df279b63a2b7c5600ed9bc60d31b5343359ac..7c9bfe03e0f0526a505a4d46c111520dc2f0531c 100644 --- a/q119/random_k4/random_4.md +++ b/q119/random_k4/random_4.md @@ -1,2008 +1,1126 @@ -INTERNATIONAL STUDENT -HANDBOOK - -A resource guide published by the -Office of International Education -Albany State University -Albany, GA - - TABLE OF CONTENTS -Notice to Students………………………………………………………………..……………………….. - -Welcome Letter from ISA……………..…………………………………………………………..………….…………………. - -Financial Documentation………………………………………………………….………………..…….. - -Academic Life………………………………………………………………………………………………………….…………….. - -American Customs and Culture………………………………………………………………………………..………… - -Immigration Status…………………………………….……………………………………………………………..……………. - -Automobiles…………………………………………………………….………………………..…………. - -Emergencies……………………………………………………………………………………………………………………………. - -English Language…………………………………………………………………………………………… - -Entertainment………………………………………………..………………………………………………………………………. - -Health Care…………………………………………………………………………………………………… - -Housing……………………………………………………………………………….…………………………………………….…. - -Dining Services……………………………………………………………………………..….…………… - -Legal Concerns………………………………………….…………………………………………..……………………………… - -Mail………………………………………..…………………………………………………………………… - -Money and Banking……………………..……………………………………….………………………………………………... - -Regent’s Test…………………………………………………………………………………….…………. - -Religious Life……………………………………..…………………………………………………….…………………………… - -Shopping…………………………………………………………………………..………………………….. - -Student Athletes ………………………………………….……………………………………………………………..……….. - -Social Security Numbers ……………………………………………..…………………….…………… - -Taxes……………………………………………….…………………………………………………………………………………… - -3 - -4 - -5 - -6 - -10 - -12 - -18 - -19 - -21 - -21 - -22 - -24 - -26 - -27 - -28 - -29 - -30 - -32 - -32 - -33 - -34 - -35 - -Telephone and Telegrams ……………………….……………………………………………………….. 35 - -Testing …………………….………………….……………………………………………………………..… - -Transportation ……………………….…………………………………………………………………………………………… - -Campus Contacts…………………………………………………………………………………………………………………… - -Checklists……………………………………………………………………………………………………… - -Services for International Students ………………………………………………………….………………………………. - -36 - -36 - -38 - -39 - -40 - - Office of International Education - -NOTICE TO STUDENTS -This guide should be used in combination with the ASU Undergraduate -Catalog/ASU Graduate Catalog and the RAM Student Handbook. The University -Catalog contains important information and regulations of the university that you -will need to know. You should obtain a copy of the Catalog (Undergraduate or -Graduate) and the RAM Student Handbook as soon as you begin your program -of study. - -3 - - Office of International Education - -Dear International Student: - -It gives me great pleasure to welcome you to the campus of Albany State University. I -will serve as your International Student Advisor. I want you to know that you are a -valuable part of the ALBANY STATE UNIVERSITY family like any other student and most -of all I want you to be successful in your studies. - -To help you reach this goal, I offer services that include providing you with important -information about your visa status and the benefits it includes. There are laws and -regulations you are required to follow in order to legally remain in the U.S. and I am -required to submit updates and other reports regularly and by specific deadlines to -confirm that. - -This guide is designed to help you locate the resources and services you will need while -you are at ASU and answer some of your questions. You are encouraged to ALWAYS -know your F-1 Visa Student Status and regulation involved. - -As an international student at ALBANY STATE UNIVERSITY you are required to meet -with me each semester of your matriculation. I encourage you to schedule -appointments so that we can meet to discuss and review options available to you at any -point of your educational endeavors at ALBANY STATE UNIVERSITY. - -In addition, please note that while you are matriculating as an International Student -ASU currently does not provide any financial assistance for you. It is your responsibility -to have the financial resources to continue your education. If for any reason you are -unable to continue to make financial contribution to your education endeavors you are -at risk of being in violation of the student immigration status. (please see the next page) - -I look forward to assisting you at ALBANY STATE UNIVERSITY and seeing you reach the -end of your program as focused and ready for the future as you are here today. - -Sincerely, - -Sue Ann Balch, J.D. -F1-Visa Student Advisor -sueann.balch@asurams.edu (Email) -229-430-7075 (Office Number) -229-434-7106 (fax Number) -205 Reese Hall - -4 - - Financial Documentation (Required Forms) -You can acquire these forms from the office of Recruitment and Admission - -*Albany State University International Student Financial Statement (To Be Completed by Student) - -*I-34 Affidavit of Support (To Be Completed By Sponsor, Affirmed Under Oath and Notarized) - -Original/Official Supporting Financial Documentation Must be: - -1. Written in English or have certified English translation attached and a -recognized transcript evaluation by a company such as WES or Silney. -2. Dated no more than six months prior to the first day of the semester you - -plan to attend. - -3. Clearly state the account balance in United States Dollar amount(s). -4. Must provide the following information: Date account opened; total - -amount deposited for the past year; -Tuition and fees -Living Expenses -Expenses of dependant -Other: Books & Health Insurance - -$14,600.00 -$9,000.00 -$0.00 -$4,000.00 -*$27,600.00 - -(*Fee changes yearly, 2016 cost listed above) - -5. All monies for tuition, health insurance, Room Board, Books & Supplies -and other fees must be must be paid by the first day of classes each -semester at Albany State University after student has been accepted. -This process must occur until the student has completed matriculation at -Albany State University - -How to set-up account? - -1. After student receives acceptance letter from - -undergraduate or graduate admissions, student will -begin the account set-up process. - -2. Student will mail fees for the year in United States - -Dollar Amount to: - -Albany State University -Financial Operations -Attn: Student Account Manager -504 College Drive -Albany, GA 31705 -3. Student will mail money certified or return receipt - -request or any secure means of mail delivery or pay -in person. - -ACADEMIC LIFE -ADVISORS -All students are assigned an academic advisor when they enroll in a degree -program. Your advisor will be a faculty member in your department. He or she -will help you choose which courses to take each semester. If you need to - -5 - - change your schedule in any way, you must get the approval of your academic -advisor. - -Academic advisors can better serve you when they understand your educational -and career plans. They can help you prepare for graduate school or for a job. -They can arrange work experiences as a part of your degree program if this will -help you. If you give your advisor a clear picture of what you plan to do, he or -she can guide you toward the most suitable courses and university activities. - -INDIVIDUALITY -Individuality is very important to Americans. This value is as important in the -classroom as it is in other parts of U.S. culture. Independent thinking is -expected of students in American classrooms. Interpretation, analysis, critical -thinking, and even challenging the professor may be expected of you in class. -You may be graded on your class participation. Usually, professors will tell you -at the beginning of the course if you are to be evaluated on class participation. -It takes time and practice to get used to speaking out in class, but it is often an -important part of the learning experience. - -REGULATIONS -Certain acts are considered to be academic dishonesty and are not permitted at -Albany State University. Submitting another person’s work as your own or doing -academic work for another person is not allowed. Academic dishonesty includes -copying exam papers, assignments, or term papers or allowing another person to -copy your work. If you use ideas or information published by other people in -papers that you turn in, you must give the writers credit by providing citations in -footnotes or references. Your professors will tell you how to do this. Students -who turn in work which is not their own will be disciplined by the professor and -possibly the university. It could lead to expulsion from the university. - -GLOSSARY OF “UNIVERSITY LANGUAGE” -CATALOG/BULLETIN -The university publishes an undergraduate and graduate catalog. These catalogs -are sometimes called bulletins. They include information on all aspects of -university life, including regulations governing academic life, course descriptions, -and a list of faculty members. You must follow the rules set forth in the catalog -for your academic program. It is an important resource. Read it carefully and -refer to it when you need information about academic offerings and -requirements. - -CLASS -Sessions or meetings of particular courses. These are listed in the Schedule -Report, issued each semester, which lists the days, times and venues the class -meets. - -COLLEGE -The University is divided into units known as “Colleges.” The Colleges at ASU -are the Colleges of Arts and Humanities, College of Business, College of - -6 - - Education, and College of Science and Health Professions. Unlike many other -countries, the term college in U.S. usage does not refer to high school or -secondary education. It is often used interchangeably with the term university. - -COMPREHENSIVE EXAMINATION -Examinations are required by most departments before a student can receive a -Master’s degree. The exams test the student’s knowledge of his or her entire -field of study. - -CORE CURRICULUM -The Core Curriculum is composed of a wide spectrum of general courses. It is -designed to give all students a broad base of basic knowledge and is required of -all students seeking bachelor’s degrees at ASU. - -COURSE -Subject Offerings. These are listed in the course catalog, by department. -Students usually take 4-5 courses per semester. The course listings are also -available through BANNER on the ASU website. - -CREDIT HOUR -Courses are given credit hour values. Most courses are 3 credit hours. This -means that the course meets for approximately 3 hours each week during the -semester. 1-hour course meets for 1 hour, etc. In order to be considered full- -time, undergraduates must enroll in 12 credit hours, graduate students need to -take 9 credit hours each semester. - -CURRICULUM -This refers to the program that you are following, and its requirements. There -are required courses and elective courses in most curricula. - -DEPARTMENT -Within each College are several Departments that specialize in a certain subject. -For example, the College of Education includes the Department of Teacher -Education and Department of Counseling and Educational Leadership. - -DROP-DATE -Students may choose to drop or withdraw from a course without penalty up to -the middle of the semester. However, you must remember that: - -1. As a foreign student, you must remain a full-time student at all times. -You must be enrolled for 12 hours at all times, unless it is a vacation -period which is the summer semester or you otherwise have -permission from the DSO in the International Office. Graduate -students must take 9 hours. - -2. Withdrawal hour’s associated with classes that are dropped due to -documented extenuating circumstances will be included in the 12 -hour total. - -7 - - 3. Students will receive a “WF” (This symbol indicates withdrawal with penalty) for course - -withdrawals identified after the 16 hour limit has been reached. -There are financial and G.P.A conditions associated with the “WF”. - -4. “W” (This symbol indicates that a student was permitted to withdraw without penalty) grades will be awarded to -students who withdraw from classes prior to the end of the schedule -change deadline and to those students who are authorized to -withdraw due to extenuating circumstances. - -After the midterm drop-date, special permission is required for course -withdrawal. The grade the student has earned up to the point of withdrawal will -be reflected on his or her transcript; however, it will not affect the GPA. In -addition, as a F1-Visa Student you are required to notify the International -student Advisor of any schedule changes you are making. - -ELECTIVE COURSE -These are optional courses. Each student must choose several of these courses. -The academic advisor will suggest specific electives. - -FINAL -This is the last, major examination of the semester. Usually, one week is set -aside at the end of the semester for final exams. A final is a long exam, which -sometimes lasts as long as two hours. Not all professors choose to give final -exams. Some will assign research papers or special projects instead. - -GRADE POINT AVERAGE -Your performance is evaluated with grades of A, B, C, D, or F. The grades A -through F are given a numerical value (A=4, F=0) and weighted by the number -of credit hours associated with each grade. Your grade point average is -computed by averaging the numerical values of the credit hours for all of the -courses you have completed. “GPA” is the common abbreviation for Grade Point -Average. - -GRADUATE STUDENTS -These are students who have already received a Bachelor’s degree or the -equivalent. The graduate degrees offered at ASU are the Specialist degree in -Education and Master’s degree in several disciplines. -MID-TERM -A Mid-Term is a major examination given in many classes in the middle of the -semester. This word also refers to the date signaling the end of the first half of -the semester. - -MINOR -A secondary, optional program of study in addition to the student’s major. - -SEMESTER -The school year at Albany State University is divided into two semesters. Fall -semester, which begins in August and spring semester, which begins in January, - -8 - - constitute the normal academic year. There is a full six-week summer session. -Summer terms are optional terms for foreign students. - -QUIZ -An informal examination, shorter in length than a regular examination. - -REQUIRED COURSE -These are courses which each student seeking a particular degree must take. - -RESEARCH PAPER -This is a major paper which can be from 5 to 25 pages in length. In order to -write this paper, a student must select an approved topic, somewhat narrow in -scope, and do research in the library. Papers are to be typed. There are -manuals in the bookstore to show you how to type it, or you can hire a -professional typist. - -SECTION -Some courses are offered in a number of sections, or classes, each semester, so -that a large number of students may attend. A course number, section letter, -and call number identify each section. - -SYLLABUS -A list of information about a course including the books you will use, the -professor’s name, office and phone number. Usually there is a general outline of -how the course will proceed, and sometimes assignments and dates of -examinations are included. - -TRANSCRIPT -A record of students’ courses and grades. An “official transcript” is sealed by the -issuing institution, and normally sent directly to the school or office requesting it. -It is used to document a student’s academic performance while attending -college. Universities, graduate schools, and some employers use this to -determine the desirability of accepting or hiring students. - -UNDERGRADUATE -These are students who are studying for their first degree at the university. -During the first year, students are called freshmen; the second year, -sophomores; the third year, juniors; and their last year, seniors. The -undergraduate degree is the Bachelor’s degree. - -9 - - AMERICAN CUSTOMS AND CULTURE -The culture and customs of the United States are difficult to define precisely, -because geographic areas and the people who live in them differ. Most of what -you will learn about the customs and culture of the people of the United States -will be by experience and through contact with people. - -While you are in the U.S., it is good to learn the customs of the people here. It -is important, however, that you maintain your own values and beliefs. If you’re -personal preferences or religious beliefs influence you not to do certain things, -do not be embarrassed to explain this. - -TIME -Being on time is very important in the American society. Classes, parties, -concerts, meetings, and other planned events all start at a specific time. If you -need to see someone, you should ALWAYS make an appointment beforehand. -Do not show up at their office or home unexpectedly. It is considered very -important not to be late to scheduled appointments or events. If you are late for -an appointment, it is important to apologize to the person(s) you have kept -waiting. Students who are always late for appointments or who turn in -assignments after the deadline earn negative reputations. Many professors will -deduct from a grade if an assignment is late or if a student arrives late for class. - -INFORMALITY -There is a great deal of informality between all types of people in the U.S. It is -our way of life. The American society does not have a formal class structure. -We may dress more casually and act more informally in social situations than you -may be used to. - -GREETING, INVITATIONS, AND VISITING -The most common forms of greeting are a handshake, smile and “Hello, I’m -pleased to meet you.” The Americans do not usually embrace in public, except -with members of their family or very close friends. - -Professors should be addressed as Doctor or Professor unless they request -otherwise. You should also use titles (usually Mr., Mrs., or Ms.) with older -persons unless they ask you to call them by their first name. For other students, -it is customary that you address them by their first name. - -If you are invited somewhere and you accept the invitation, it is VERY -IMPORTANT that you appear at the event. Do not accept an invitation simply -because you are afraid to say no. If you cannot or do not want to go to the -event, politely tell the person who invited you htat you cannot or choose not to -attend. It is much more inappropriate to accept an invitation and then fail to -appear than it is to decline the invitation in the first place. If invited to visit -someone’s home, you are not normally expected to bring a gift. It is always -acceptable, but not necessary, to bring a small token or an inexpensive gift to -your host or hostess. An expensive gift is usually inappropriate and may make - -10 - - your host or hostess uncomfortable or embarrassed. American homes, usually -does not have domestic servants, so if you stay as a houseguest, it is proper for -you to make your own bed or volunteer to help with other household chores. - -TABLE MANNERS -Table manners is most common to Americans including not speaking with food in -your mouth, keeping your elbows off the table, and keeping your hands in you -lap while you’re not eating. A knife, fork, and spoon are used unless “finger -foods” are being served. You should not belch at the table after eating; this is -considered rude in the U.S. After the meal is finished, the napkin should be laid -beside the plate and used silverware should be laid across the top of the plate. - -RELATIONSHIPS -Most Americans have many friendly acquaintances, but just a few close friends. -Visitors from other countries often remark that Americans are very friendly, even -at the first meeting, but that it is difficult to get to know them very well. -Sometimes this is true; although Americans are friendly to everyone, they are -also very private people and often have just a few intimate friends. - -Americans often say things like “let’s get together sometime” as a friendly way to -say good-bye, NOT as an invitation. If you would like to get to know someone -better, it is a good idea for you to invite them to go somewhere with you on a -specific day, at a scheduled time. Participate in as many social activities as you -can and you will meet and make many new friends. - -MEN AND WOMEN -American women are independent, and women and men are considered equal. -It is not appropriate to whistle at, touch, or make personal comments about -women whom you do not know EXTREMELY well. It is common for American -university women to live alone or with other women in apartments off campus. -Guests of both sexes are often invited to these apartments. Do not assume that -your hostess has romantic intentions simply because she has invited you to stop -by her apartment. - -There are no clear guidelines for dating Americans. If you want to get to know -someone, it is often best just to ask the person to join you for a snack or to get -together to study. It is important to remember, however, that if two people go -out on a date, it only means that they enjoy each other’s company. It does not -always indicate serious intentions or constitute a promise! If you begin to date -someone from another culture, it is important to remember that s/he simply may -not view the relationship in the same way that you do, especially regarding the -seriousness of the relationship or the desire to engage in sexual activity. - -11 - - It is most often the man who invites the woman out on a date, and customarily -the man pays all of the expenses unless the persons agree that it will be a -“Dutch Treat,” that is, each person pays for his or her own expenses. Women -sometimes ask men for dates. A woman who asks a man out should be -prepared to pay for the expenses. - -BATHING AND HYGIENE -Most Americans bathe or shower daily, and use deodorants and anti- perspirants. -Americans are not accustomed to body odors and find it offensive. Some -international visitors may want to adjust their bathing habits and accommodate -American preferences for the time they are here. - -CULTURE SHOCK -There are many differences between the American culture and the home culture -of many international students. Adapting to these differences can cause “culture -shock.” If you are having trouble adjusting to life in the U.S., or are -experiencing depression or physical or emotional problems, you may be -experiencing culture shock. The Office of the International Student Advisor can -provide cross-cultural counseling to help you, or refer you to a counselor on -campus free of charge. Counseling is always confidential. - -ADDITIONAL INFORMATION -If you would like additional information about American culture, please visit the -Office of the International Student Advisor. The office contains several -handbooks and guidebooks for foreign visitors in the United States. - -IMMIGRATION STATUS -THE U.S. CITIZENSHIP AND IMMIGRATION SERVICES (USCIS) is the branch of -the Federal government created to monitor the entry and exit of foreign -nationals. This branch of the government was formerly known as the -Immigration and Naturalization Service (INS). You will encounter INS officials as -you travel between the U.S. and your home country and when you apply for -special programs. The following information should help you understand the -regulations concerning student immigration status. Most ASU students have -either F-1 or J-1 immigration status. - -PASSPORT -Your passport must be valid at all times. When it is time to renew your passport, -contact your country’s embassy in Washington, D.C. to determine what -procedures you need to follow in order to extend your passport. - -VISA -There is a visa stamped in your passport, which permitted you to enter the -United States. A single entry visa permits you to enter the U.S. only once on -that specific visa. A multiple entry visa allows you to enter the U.S. as many -times as you like (after leaving on vacations or trips home) as long as you - -12 - - remain a full-time student. Student stays are valid for the “duration of status”. -This status is indicated by the mark “D/S’ in your passport and on your I-20/DS- -2019 and I-94. As long as you remain in full-time student status, you are in the -country legally, regardless of the expiration date on your entry visa. -If you leave the U.S. after your visa expires, you must go to a U.S. Embassy or -Consulate-General in your home country and obtain a new visa before you can -return to the U.S. It is difficult to obtain a new visa in a “third country.” It is -difficult to obtain new F-1 visas in Canada or Mexico, unless you are from those -countries. - -F-1 IMMIGRATION STATUS -If you are a full-time degree-seeking student or a Rotary student at ASU, you -have entered the U.S. with F-1 immigration status. This type of immigration -status is designed to allow you to obtain your degree and return home following -your graduation. F-1 status is a “non-immigrant” status. “Non-immigrant” -means that you are not moving to the U.S. permanently. This immigration status -is valid for “duration of status.” - -• You should carefully consider the dates of your authorized stay and -make sure you are following the procedures under U.S. immigration -laws. It is important that you depart the U.S. on or before the last -day you are authorized to be in the U.S. on any given trip, based on -the specified end date on your Arrival-Departure Record, Form I-94. -Failure to depart the U.S. will cause you to be out-of- -status. Additional information on successfully maintaining your -immigration status while a student or exchange visitor can be found -on the Immigration and Customs Enforcement (ICE) website. -• Staying beyond the period of time authorized by the Department of - -Homeland Security (DHS) and being out-of-status in the -United States is a violation of U.S. immigration laws, and may cause -you to be ineligible for a visa in the future for return travels to the -U.S. - -• Staying unlawfully in the United States beyond the date Customs and -Border Protection (CBP) officials have authorized--even by one day-- -results in your visa being automatically voided, in accordance with -INA 222(g). Under this provision of immigration law, if you overstay -on your nonimmigrant authorized stay in the U.S., your visa will be -automatically voided. In this situation, you are required to reapply for -a new nonimmigrant visa, generally in your country of nationality. - -• For non immigrants in the U.S. who have an Arrival-Departure - -Record, Form I-94 with the CBP admitting officer endorsement of -Duration of Status or D/S, but who are no longer performing the -same function in the U.S. that they were originally admitted -to perform (e.g. you are no longer working for the same employer or -you are no longer attending the same school), a DHS or an -immigration judge makes a finding of status violation, resulting in the -termination of the period of authorized stay. - -13 - - Staying Beyond Your Authorized Stay in the U.S. and Being Out of Status - -• You should carefully consider the dates of your authorized stay and make -sure you are following the procedures under U.S. immigration laws. It is -important that you depart the U.S. on or before the last day you are -authorized to be in the U.S. on any given trip, based on the specified end -date on your Arrival-Departure Record, Form I-94. Failure to depart the -U.S. will cause you to be out-of-status. Additional information on -successfully maintaining your immigration status while a student or -exchange visitor can be found on the Immigration and Customs -Enforcement (ICE) website. - -• Staying beyond the period of time authorized by the Department of - -Homeland Security (DHS) and being out-of-status in the United States is a -violation of U.S. immigration laws, and may cause you to be ineligible for -a visa in the future for return travel to the U.S. Select Classes of Aliens -Ineligible to Receive Visas to learn more. - -• Staying unlawfully in the United States beyond the date Customs and -Border Protection (CBP) officials have authorized--even by one day-- -results in your visa being automatically voided, in accordance with INA -222(g). Under this provision of immigration law, if you overstay on your -nonimmigrant authorized stay in the U.S., your visa will be automatically -voided. In this situation, you are required to reapply for a new -nonimmigrant visa, generally in your country of nationality. - -• For non immigrants in the U.S. who have an Arrival-Departure Record, -Form I-94 with the CBP admitting officer endorsement of Duration of -Status or D/S, but who are no longer performing the same function in the -U.S. that they were originally admitted to perform (e.g. you are no longer -working for the same employer or you are no longer attending the same -school), a DHS or an immigration judge makes a finding of status -violation, resulting in the termination of the period of authorized stay. - -What Items Do Returning Students Need? - -All applicants applying for renewals must submit: - -• A passport valid for at least six months; -• An application Form DS-156, together with a Form DS-158. Both forms -must be completed and signed. Some applicants will also be required -to complete and sign Form DS-157. Blank forms are available without -charge at all U.S. consular offices and on the Visa Services website -under Visa Applications Forms at the United States Citizenship and -Immigration Services (https://www.uscis.gov/). - -• A receipt for visa processing fee. A receipt showing payment of the visa - -application fee for each applicant, including each child listed in a -parent’s passport who is also applying for a U.S. visa, is needed; -• A new I-20 or an I-20 that has been endorsed on the back by a school - -official within the past 12 months. - -14 - - All applicants applying for renewals should be prepared to submit: - -• A certified copy of your grades from the school in which you are - -enrolled; - -• Financial documents from you or your sponsor, showing your ability to - -cover the cost of your schooling. - -Students Away from Classes More Than Five Months -Students in or outside the U.S., who have been away from classes for more -than five months, will likely need a new visa to enter the U.S. - -How long may I stay on my F-1 student visa? - -When you enter the United States on a student visa, you will usually be -admitted for the duration of your student status. That means you may stay as -long as you are a full-time student, even if the F-1 visa in your passport -expires while you are in America. For a student who has completed the course -of studies shown on the I-20, and any authorized practical training, the -student is allowed the following additional time in the U.S. before departure: - -• F-1 student - An additional 60 days, to prepare for departure from the - -U.S. or to transfer to another school. - -• M-1 student - An additional 30 days to depart the U.S. (Fixed time period, -in total not to exceed one year). The 30 days to prepare for departure is -permitted as long as the student maintained a full course of study and -maintained status. An M student may receive extensions up to three years -for the total program. - -As an example regarding duration of status, if you have a visa that is valid for -five years that will expire on January 1, 2001, and you are admitted into the -U.S. for the duration of your studies (often abbreviated in your passport or on -your I-94 card as "D/S"), you may stay in the U.S. as long as you are a full- -time student. Even if January 1, 2001 passes and your visa expires while in -America, you will still be in legal student status. However, if you depart the -U.S. with an expired visa, you will need to obtain a new one before being able -to return to America and resume your studies. A student visa cannot be -renewed or re-issued in the United States; it must be done at an Embassy or -Consulate abroad. - -FORM I-20 -Before you came to ASU, you were sent Form I-20 by the ASU Admissions Office. -This form is the one you use to obtain your F-1 student visa, and it is a very -important form. You should keep it safe with your passport. A Designated School -Official (DSO) must sign your I-20 whenever you travel outside the United -States, even if you are only going for a short time. If you change degree -programs (change your major), you will need a new I-20. New I-20 forms are -available from the International Admissions Officer on campus at 229-430-4646 -or visit the Office of the International Student Advisor for further assistance. If -you transferred to ASU, be sure to keep your ASU I-20 and the I-20 from your - -15 - - previous school together. Note the date of expiration in #5 on your I-20. If you -need to stay past this time to complete your studies, you must contact a DSO at -least 30 days before this expiration date to extend your stay as a student. - -FULL-TIME STUDENT STATUS (In-Status) -In order to maintain your full time status, you must attend school at least 9 -semester hours per semester for graduate studies and at least 12 semester -hours per semester for undergraduate studies. If you have any questions -about this regulation, please see the International Student Advisor. Students -who fail to maintain this requirement risk being placed on probation and could -lose university scholarships, including International Student Scholarships. To -avoid losing your full-time status, check with your advisor and/or Registrar -before you drop a class. - -VACATION -The Summer Term is considered a vacation period for all non-immigrant -students, as long as the student intends to register full-time for the next term. -You may enroll part-time, full-time or not at all during the summer term. Make -sure you check with the International Student Advisor before taking ANY -semester other than the summer semester off. You can go home for summer -any time you wish. - -VACATIONS INSIDE OR OUTSIDE THE USA - -(A) Inside the USA: You do not need permission of any kind from anyone to -travel anywhere within the USA. Travel between the 50 states is totally -unrestricted. Have fun—see the USA! - -(B) Outside the USA: Before leaving the USA, you must go by the office of a -Designated School Official (DSO) with your passport and I-20 and let the -DSO sign your I-20 and check your visa. Your I-20 must be signed for -you to re-enter the USA. You must have both the signed (or a new) I-20 -and a valid F-1 visa in your passport to re-enter the USA. - -If your visa has expired, you will need to go to a U.S. Embassy or Consulate- -General to get a new visa before you can re-enter the country. U.S. VISAS ARE -NOT ISSUED IN THE USA. If you have dependents that are on an I-20, we need -to check their passports and visas as well. - -PLEASE VISIT THE OFFICE OF THE INTERNATIONAL STUDENT -ADVISOR AT LEAST TWO (2) WEEKS BEFORE YOUR TRIP SO THAT -THERE WILL BE SOMEONE AVAILABLE TO SIGN YOUR DOCUMENTS. - -TRANSFER STUDENTS -If you are transferring from another U.S. college or university to ASU, you must -see the International Student Advisor within 5 days of your registration at the -beginning of the semester for completion of the transfer process, which must be -done to maintain your student status. - -EMPLOYMENT - -16 - - (A) ON-CAMPUS: As soon as you have registered for classes, you may work -on campus as a student assistant in the various departments if hired. . -Each department or facility has its own budget for student workers, so -you must go directly to them to seek employment. Once you have found -a job, you must register with the ASU Human Resources Department -(Department of Human Resources) in order to be paid for your work. To -register, you must have a Social Security Number. For more information -about Social Security Numbers, please see the section on Social Security -Numbers in this handbook. - -(B) OFF CAMPUS—ECONOMIC HARDSHIP: You must see the International - -Student Advisor before you can work off campus. Due to recent changes -in immigration laws, it is virtually impossible for foreign students to work -off campus. The only two exceptions are internships for academic credit -or extreme economic difficulty due to conditions beyond his or her -control. Students must apply to USCIS for permission to accept -employment off campus due to economic hardship, and the ISA can -assist you with this application. - -(C) OFF CAMPUS—PRACTICAL TRAINING: You must see the International -Student Advisor before you can work off campus. Curricular Practical -training (or an internship) is work directly related to your course of study -that will give you “hands-on” job experience that will help you get a -better job when you return to your home country. It can be granted only -after you have been at ASU for an academic year, and then only under -certain specific conditions. There are provisions for curricular practical -training before graduation and after graduation. - -Other than the above types of work, NO OTHER EMPLOYMENT IS PERMITTED -while you are in student status. You may not work off campus without -permission from USCIS and/or the ISA. If you do work off campus illegally, you -will lose your student immigration status and its benefits. Working illegally in the -United States is the most serious violation of non-immigrant status in the eyes of -the USCIS, and you WILL NOT be reinstated if you violate this regulation. - -INFORMATION UPDATES -Any change in address, phone number or marital status must be reported to the -Office of the International Student Advisor (ISA) and the Office of the Registrar -within 10 days. You must report the changes to both offices. If you do not -notify the ISA and the Registrar, you will not receive important mail from the -university. Please note that the Office of the Registrar maintains two addresses -on each student, a local address and a home address. The local address is your -address in Albany. If you live on campus, the university will automatically record -your ASU box number. If you live off campus, you must report your address to -the ISA and the Registrar. The home address is your address in your country. It -is the address you listed on your application for admission. All official -correspondence concerning your grades and your fees will be sent to the home -address. You may want to change your home address to your Albany address to -make sure you receive these notices. If you move off-campus, you must also - -17 - - complete a United States Postal Service Change of Address card and sign the -card. Completing this card will allow the campus mailroom to forward mail to -you at your new address. - -THE CARDINAL RULE -If you have questions about working or any other USCIS-related -matter while in the USA, visit the Office of the International Student -Advisor. The Office of the International Student Advisor and -Designated School Officials are here to assist you. We will help you as -much as we can; however, it is YOUR RESPONSIBILITY to keep the -office informed of your status and to ask for help when you need it. The -U.S. Immigration and Naturalization Services holds you accountable for -your own actions and expects you to follow all laws pertaining to your -immigration status. If you choose to violate the law or fail to educate -yourself about the law, you will be responsible, not the Office of the -International Student Advisor. Do not allow yourself to fall out of -status. Do not rely on the advice of other students for immigration -matters. You could lose the privilege of studying in the U.S. now, and -you could be prevented from returning to the U.S. in the future. - -A revision of the immigration law, the Illegal Immigration Reform and -Immigrant Responsibility Act of 1996, contains very harsh penalties of -students who violate student immigrant status, including provision to -prevent offenders from returning to the United States for periods of -several years. The new immigration regulations based on this act are -currently in place, but because the law is a new one, there are often -modifications or updates to the regulations. The International Student -Advisor will keep you informed about modifications in the regulations. - -AUTOMOBILES -REQUIRED INSURANCE -Law requires liability insurance for a car. If you own an automobile in Georgia, -you must purchase liability insurance. Collision insurance is not required, but it is -recommended. Proof of insurance, usually in the form of an insurance card, -must be carried in the car at all times. Look in the “Yellow Pages” of the phone -book to locate insurance agents, or ask a friend to refer you. Compare prices -with several companies before buying. - -DRIVER’S LICENSE -As a non-resident student at the university, you may drive in Georgia with a valid -international driver’s license. If you do not have an international license or if -your license expires during your stay, you may apply for a state of Georgia -license at the `Dougherty County Department of Motor Vehicle Safety (DMVS), -2062 Newton Road, Albany. The contact telephone number is 229-430-4258. -Getting a Georgia license requires a written test and a driving test. You should -present your passport, I-94, and your I-20 or DS-2019 to the examiner. You will -also need to bring a lease or utility bill, which verifies your address. If you live in - -18 - - a residence hall, you may also need a letter from the International Student -Advisor certifying that you live on-campus. The fee for a five-year license for -driving a passenger car is $20.00. A provisional permit costs $10.00. All fees -must be paid in cash. In addition, you will need a denial letter from the Social -Security Office and/or a Social Security Card. Also, you will need a letter from -your International Student Advisor before you go. - -Many students find that having a State of Georgia license helps them to obtain -lower rates for automobile insurance. Georgia identification cards are also -available at the driver’s license office for a fee. Both the license and the ID card -contain photographs of the bearer. Because the driver’s license and the state ID -card are the most widely accepted forms of identification, it is easier to verify -your address and age with these items than with your passport or ASU ID. -Having proper identification is especially important for writing checks and -handling other banking matters. You should consider obtaining an official form -of identification other than your passport. If you must carry your passport with -you at all times, you risk losing it. You should call the license bureau at 229-430- -4258 to make sure that the test is being given on the day you plan to apply for a -license. The bureau is closed on Mondays and will not give tests when it is -raining. - -If you hold a license from another state in the U.S. or from another country, you -will need to surrender this license to obtain a Georgia license. Please note that -once an out-of-state license has been surrendered to the State of Georgia, it is -not possible to reclaim the license. You should consider whether you wish to -surrender your home country license before you go to the bureau. If you have -questions about this regulation, please see the International Student Advisor. - -RULES OF THE ROAD -Although your international license is valid in the U.S., it is a good idea to read -the State of Georgia Driver’s Manual before you begin driving. Every country has -different traffic laws. Reading these laws in advance could prevent you from -having an accident. Copies of the State of Georgia Driver’s Manual are available -in the Office of the International Student Advisor. - -TRAFFIC VIOLATIONS -If you break any of the traffic laws, including driving over the speed limit, you -risk being stopped by a police officer. If a police or sheriff car follows you with -flashing lights and sirens, the officer wants you to pull over to the side of the -road. You should be prepared to pull over in a safe spot and do as the officer -instructs. Usually, the officer will ask to see your driver’s license, proof of liability -insurance, and registration for the car. He will also ask you questions about your -driving. If you have broken a traffic law, he may issue you a citation (ticket). -Citations involve paying fines, and in extreme cases, imprisonments. If you are -in an accident, you must wait at the scene of the accident for a police officer to -investigate the accident. Do not move the car and do not leave until the officer -has interviewed you about the accident. - -PARKING - -19 - - Most parking in Albany is free. Parking on the ASU campus requires a permit, -which costs $15 per academic year. Please refer to the ASU Parking and -Transportation resource guide for parking regulations. If parking regulations are -violated, you will get a “ticket” and be required to pay a fine. Parking Rules on -campus are clearly stated in the RAM Student Handbook. - -RENTAL AND PURCHASE -If you have a valid driver’s license and are at least 21 years old, you will probably -be able to rent or lease a car. Some rental places also require a credit card in -your name or that you are at least 25 years of age. Check with the larger local -car dealerships about short- and long-term rental. - -If you wish to buy a car for your stay in the U.S., be CAREFUL. Most companies -will service your car if it needs repair after purchase. You must pay for all -repairs unless you have a “manufacturer’s warranty” or paper which states that -the maker of the car will pay for repairs for a limited period of time after -purchase. If you buy a used car from a private dealer or individual, check the -vehicle carefully before purchase. It might help to bring a friend who knows -about the mechanical working of an automobile to inspect the car for you. An -honest car dealer will be happy to let you inspect all parts of the car and -remember to ask questions before you buy. When purchasing a car, you must -have a TITLE and LICENCE PLATE (tag) from the State of Georgia. The dealer -where you purchased the car will give you the correct papers to fill out and help -you obtain these things. In addition, you must renew items such as insurance, -license and tags on a yearly basis. If you buy a car from an individual, be sure to -ask for assistance with these items. If you are bringing a car to Albany from -another state, you must transfer the title for the car and obtain license plates -from the State of Georgia. You should also notify your insurance company that -you have moved. - -License plates must be renewed each year before paying the tax on your car. -Once you have paid the tax, you will be given a sticker for that year. The sticker -should be placed in the upper right hand corner of the license plate. If you have -questions about license plates, please contact the Dougherty County Tag Office, -225 Pine Avenue, Room 104, Albany, GA. The telephone number is 229-431- -3255. - -EMERGENCIES -To call the POLICE, AMBULANCE, or FIRE DEPARTMENT, dial 911. Call this -number only if someone needs immediate help. This is a special emergency aid -number and needs no other numbers. For medical problems that are not life- -threatening emergencies, call the Palmyra Medical Centers, 2000 Palmyra Road, -Albany. Tel.:229- 229-434-2000 or Phoebe Putney Memorial Hospital, 417 3rd -Avenue, Albany. Tel.: 229-883-1800, 229-312-1000. Police concerns are handled -according to location: - -Campus Police: -Albany Police: -Leesburg Police: - -229-430-4711 -229-431-3266 or 229-431-2100 -229-759-6012 - -20 - - Albany Fire Dept.: - -229-431-3266 - -ENGLISH LANGUAGE -Students who scored below 523 on the TOEFL (Test of English as a Foreign -Language) are REQUIRED to take special supplementary English as a Second -Language (ESL) courses. These courses are non-credit-bearing, but they will help -you with your academic English needs. The course numbers are ENGL 0075, -0077, 0098 and 0099. Any international students are welcome to take these -courses, even if the TOEFL score was above 523. - -English 1101 is the introductory university level English course required for all -ASU degree programs. The course prepares students for the remaining required -English courses in their degree program. - -There are some options available for students who wish to improve their English -language skills. For students who need help with their writing skills, the English -Department offers tutoring in the Writing Center lab located in Simmons Hall. -Another option for tutoring is the PEER TUTORING program. Tutoring is available -in most subjects free of charge during the fall and spring semesters. Contact the -Office of the International Student Advisor or your English professor to find out -which of these facilities will best meet your language needs. - -ENTERTAINMENT -ASU provides many entertainment and social opportunities for students. The -student centers on both campuses are available to you for all activities. It has as -its objectives the fulfillment of the cultural, educational, and recreational desires -of all students of ASU. Various programs and activities are planned and -supervised by the Director of Student Activities and the program coordinator to -round out and broaden the experiences of each student. The services and -facilities of the Student Union are open from 10.00 a.m. to 7.00 p.m. Monday -through Friday. On Saturdays and Sundays, the Game Room hours will be posted -in the game room, unless other wise notified. The facilities and services of the -Union are for students to make full use of. They include conference and meeting -rooms, lounges, a game room, Pizza Hut, the ASU Bookstore, Student -Government Office, a Post Office, an ATM, an information desk, vending -machines, and the ASU Board of Managers. - -The Albany State University International Student Association (ASUISA) is a -special club for international students at ASU. This group plans and promotes -cultural diversity and global awareness activities involving ASU and the -community. It also organizes programs and field trips for students who are -interested in more international activities. Look for information about the first -meeting in late August. - -For more information about clubs and activities, you should read the campus -paper, The Student Voice, and listen to the community radio stations. Also plan -to attend the campus-wide activities fair that takes place during the early part of - -21 - - the Fall Semester. Each of ASU’s clubs will have a booth at this fair to provide -more information for potential members. - -HEALTH CARE -The United States philosophy of health care is different from that of many other -countries. Health care in this country is viewed as an industry. All Health Care -professionals are expected to be paid by their patients or clients. Patients often -must prove ability to pay before treatment will be given. Doctors in this country -usually manage their offices so that they spend a minimum amount of time with -each patient. Nurses do as much as possible to help free the doctor to see more -patients. Expect to deal with nurses in the office and over the telephone. - -MEDICAL EMERGENCY -In case of medical emergency involving a life-threatening situation, or to request -an ambulance (emergency transportation to the hospital), call 911. Call this -number only in true emergencies. ASU students should not use the emergency -room services at the local hospitals, Phoebe or Palmyra, unless the medical -problem is life threatening. Students should use the university health center -infirmary as their main source of medical care. If you do not know whether the -situation is a true emergency, it is probably best to call campus police and have -the officer escort you to the health center. The nurse will evaluate the problem -and will send emergency cases to the hospital. THE HEALTH CENTER IS THE -CLOSEST MEDICAL FACITLITY TO THE RESIDENCE HALLS. Please note that -using the ambulance and the emergency room at the hospitals is very expensive. -This expense will not be covered by medical insurance if the situation is not a -true emergency. - -STUDENT HEALTH SERVICES -The University Health Service of Albany State University is located in the -Academic Administration (ACAD) Building, Room 138. Student Health Services -provide opportunities for safeguarding the health of all students at ASU. The -staff provides the following through student health fees: - -1. General medical care. -2. Confinement in the Health Services. Overnight stays are not available. -3. Consultation on general health needs. - -The health fee DOES NOT cover expenses for special medicines and treatments, -visits to doctors’ offices, operations or off-campus hospitalization. - -The University Health Services office is open Monday through Friday, 8:00 a.m. – -5:00 p.m. Our Doctors’ hours are Monday through Friday, 4:00-5:00 p.m. -Emergency calls are possible at any time. Each student, freshman or transferee, -is required to have on file a medical record completed and signed by a physician. -These records will be good for four years provided the student is in continuous -attendance at Albany State University and has consulted with the University -Physician periodically. - -22 - - DENTAL CARE -If you need to have your teeth cleaned or another dental service, you should -look in the telephone book Yellow Pages under “Dentists.” Phone to make an -appointment before you go. - -EYEGLASSES AND CONTACT LENSES -If you need an eye examination, or you want your eyeglasses or contact lenses -changed, you should look in the telephone book Yellow Pages under, “Opticians -and Optometrists.” Phone to make an appointment before you go. For medical -concerns with your eyes, look under “Ophthalmologist.” - -COUNSELING -Sometimes individuals become depressed, lonely or fearful in a strange -environment. Many overseas students come from a culture very different from -that of the U.S., and they often miss the support of their family and friends at -home. If you are feeling emotionally upset, depressed, or just need someone to -talk to, contact the International Student Advisor. S/he can make helpful -suggestions, listen to your concerns, and may refer you for further counseling. - -The primary function of the Counseling Services is to assist students in their total -development by providing services and programs to facilitate their intellectual, -emotional, social and academic growth. Counseling services also aid students in -determining a purpose and direction for their lives and in developing coping skills -and behaviors to aid them in working more efficiently with others in their -environment. - -PAYMENT -Health care in the United States is not free. Patients must pay for the services of -doctors, dentists, laboratory workers and hospitals. Their fees are often very -expensive. Care at the Health Services on campus is paid for by student health -fees each semester, so go to the Health Center first for any illness. If you have -an appointment with any off campus doctor or dentist, be prepared to -pay the bill at the conclusion of the appointment. Insurance will not -always cover 100% of the cost. Medical insurance does not cover the expense -of dental visits or eye exams and glasses. Students may want to plan to take -care of routine visits to these doctors during vacations at home. - -INSURANCE -Due to the high cost of medical care in the U.S., all foreign students -must purchase health insurance before enrolling for classes at ASU. -ASU is pleased to offer one policy for all ASU foreign students student. -Since September 1994, new regulations from the U.S. government -have determined what type of insurance coverage students must have. -The ASU policy has been designed to meet this minimum level of -coverage at a reasonable price. The plan covers portions of hospital -costs, doctor visits, and medication but not dental visits or eyeglasses. -Insurance to cover all costs at 100% and to cover dental visits and -eyeglasses is extremely expensive. Few U.S. residents carry such -coverage due to the high cost. The insurance system in the U.S. is - -23 - - designed for the patient to pay a portion of the cost in addition to the -insurance premium already paid. Insurance premiums are the fees paid -for insurance coverage. The university will bill all ASU international -students for the insurance premiums. Insurance from another source is -generally not accepted, since the coverage must meet U.S government -standards. If you need information about the ASU health insurance -program, see the International Student Advisor. If you do not pay your -premium, you will lose your insurance coverage and you will have a -hold placed on your student account, which will prevent you from -registering for classes. You should be careful to avoid a situation -where you lose your coverage because you never know when an illness -or accident may occur. If you have an illness or an accident and you -have no insurance, you will be responsible for 100% of the cost of your -treatment. For Student Health Insurance, please visit the Student -Health Insurance website. www.studenthealthinsurance.com - -GOOD HEALTH -The demand of university life and adjusting to a new environment can be very -stressful. Whenever a person is under stress, s/he is much more susceptible to -illness. Be sure that you take steps to protect your health. Eat properly, get -plenty of sleep, exercise, and protect yourself from communicable diseases, -especially sexually transmitted diseases such as AIDS. If you think you have a -health problem, seek help immediately. If you have questions about any health -matter, please visit the Student Health Center or the Counseling Center. - -IMMUNIZATIONS -All students in the state of Georgia must be vaccinated against measles, mumps, -and rubella. The vaccination can be obtained in one injection called an MMR. It -is recommended that you also receive a shot for Meningitis as well as a -Tuberculosis test. The vaccines are required because these diseases are -present in the U.S. You must protect your health while you are in the U.S. -Students must receive two doses (injection) of this “MMR” vaccine at least 30 -days apart. These vaccines are available at the Dougherty County Government -Office at 222 Pine Avenue, Albany. The telephone number of the Health -Department is 229-430-6200. Students who do not receive these vaccines will -not be allowed to register for classes. We prefer that you send proof of -immunizations and TB test as part of your application package. - -HOUSING -ON-CAMPUS HOUSING -The residence halls provide convenient, inexpensive housing on campus for -students. Most students will share a room with another student (roommate) and -a bathroom with other students nearby. Students must sign a room contract and -pay an application fee and deposit in order to live in a residence hall. The exact -amount to be paid will be included in your letter of admission. The contract -obligates you to live in the residence hall for the entire nine-month academic -year (2 semesters). The on-campus apartments are open all year. They are, - -24 - - however, in great demand. If you want to live in an on-campus apartment, stop -by the Office of Housing for more information or call 229-430-4741. - -If your room needs repairs or you have questions and concerns about any aspect -of life in the residence hall, you should contact your Resident Assistant. The -Resident Assistant, or RA, is also a student who lives in the hall, but this student -is an employee of the Office of Housing. The RA enforces the rules of the -residence hall such as quiet hours. The RA also handles issues such as changing -rooms or roommates and reporting when repairs are needed. The Residence -Hall Director supervises the Resident Assistant of the residence hall. Your -questions and concerns about life in the residence hall should be addressed to -the RA or the Residence Hall Director. Of course the staff of the Office of -Housing and Residence Life will be happy to assist you as well. The telephone -number is 229-430-4741. - -During recent years, the demand for on-campus housing has been high. It is -very important for you to read carefully any correspondence from the Office of -Housing. You need to follow specific procedures to reserve your room each year. -Please make a note of all deadlines and procedures. - -ROOMMATES -Living with an American roommate can be an excellent way to practice English, -learn about the U.S. culture and customs, or to meet a new friend. Living with -another person requires patience, tolerance, and kindness, especially as you get -to know one another during the first few days. It can also be one of the most -enjoyable aspects of your stay on campus. You should expect to disagree with -your roommate about some things. This is normal. The best way to solve your -differences is to speak honestly to one another. Americans are accustomed to -direct discussion of problems. If there are problems, which cannot be solved -between the two of you, you may go to your Resident Assistant (RA) who will -work with you to assign a new roommate. - -OFF-CAMPUS HOUSING -Off-campus housing is plentiful in Albany. There are many apartment complexes -and rental houses near campus and around town. Listed below are a few -suggestions of places to look for housing: - -1. - -2. - -3. - -4. - -Albany Herald, the local newspaper, lists apartments, houses, and -rooms for rent in the classified advertising section of the paper. The -ad will usually list the monthly rent as well as a phone number to call -for more information. -Bulletin boards and notices on campus: check classroom buildings for -notices and ads about apartments for rent. Many students looking for -roommates advertise in this manner. -The telephone directory: look under “Apartments” in the Yellow -Pages. -The Office of the International Student Advisor has listing of apartment -complexes in Albany, which lists such information as rental cost, size, -and distance from campus. - -25 - - RENT AND DEPOSITS -You will probably have to pay a deposit of two or three hundred dollars, plus -your first month’s rent before you can move into an apartment or rental house. -When you move out of the apartment you will get the deposit back if there has -been no damage to the apartment. You will also be required to sign a lease. A -lease is a legal contract that states the amount of time that you will live in the -apartment. Most leases are for one year. Once a lease is signed, you are legally -bound to live in the apartment, and you must pay the rent even if you decide -that you no longer want to live in the apartment. Do not sign any lease that you -do not understand. Be sure to check the dates of the lease carefully. Please -visit the International Student Advisor before you move off campus. He/she can -offer advice about apartments and assist you with interpreting leases. Please -note that if you lease an apartment in your name but also have roommates living -with you, you must be very careful about changing the name on the lease if you -move. If your name is on the lease, you are responsible for the payment of rent -even if you move out and allow other students to move into the apartment. If -you do not pay the rent, the landlord may take legal action against you. Please -be especially careful in situations where you share telephone and utility services -with other students. Never agree to have a lease or a utility bill placed in your -name unless you are certain no additional money is owed on the account. If you -change a lease or utility bill from another student’s name to yours, you will be -responsible for any money owed on the lease or account even if the charges -were billed before the account was placed in your name. - -DINING SERVICE -The University Dining Hall, which is under the supervision of the Director of Food -Services, provides nutritious and wholesome meals at a minimum cost to the -student. Meals are served in an ultra-modern facility, which includes a food court -consisting of the following concepts: Café Features, Wokery, Grill Works, Itza -Pizza, Deli Corner, Sweet Things, Salad Garden, Veggie and Drinks. The meals -are unlimited (all you can eat). The Dining Hall seating capacity is approximately -525 students. All boarding students must present proper University student -identification. Non-boarding students, faculty, staff and visitors must pay cash at -the cashier’s stand before being served. - -Meal Schedule - -Monday-Thursday -Breakfast -Lunch -Dinner -Friday -Breakfast -Lunch -Dinner - -Saturday & Sunday -Brunch - -7:00 a.m. – 8:45 a.m. -11:30 a.m. – 1:00 p.m. -5:00 p.m. – 7:00 p.m. - -7:00 a.m. – 8:45 a.m. -11:30 a.m. – 1:00 p.m. -4:30 p.m. – 6:30 p.m. - -10:30 a.m. – 12:30 p.m. - -26 - - Dinner - -4:00 p.m. − 5:30 p.m. - -(Schedule is subject to change without notice) - -LEGAL CONCERNS -LAWS -While you are in the U.S., you are subject to all of the laws and regulations, -which govern U.S. citizens. In addition, you are subject to immigration laws, -which define and limit your activities as a non-immigrant. Failure to comply with -immigration laws can lead to a request for your voluntary departure, or a forced -deportation from the U.S. - -EQUAL PROTECTION and DUE PROCESS -All persons in the U.S., regardless of citizenship, are entitled to the same -protection of the law and the same procedures of law enforcement. This means -that, with a few exceptions, you will be treated the same as an American citizen -in your relationship with the law enforcement agencies. Those few exceptions -involve immigration law. Consult the Office of the International Student Advisor -for clarification of this issue. - -ILLEGAL DRUGS -The possession, use, or sale of drugs or narcotics, including marijuana, hashish, -amphetamines, hallucinogens, barbiturates, cocaine, heroin, and a wide range of -other drugs, are serious offenses under Georgia law and immigration law. Those -who choose to use illegal drugs run the risk of expulsion from the university, -imprisonment, and/or deportation from the United States. - -ALCOHOL -The minimum legal age for drinking alcoholic beverages (beer, wine, liquor, etc.) -in the State of Georgia is 21. You cannot be served alcoholic beverages unless -you are 21 or older. You will be asked to present an identification card with your -photograph and your date of birth before you can purchase alcohol from a store -or enter a bar, nightclub, or other establishment that serves alcohol. Students -who are over age 18 but under 21 will be admitted, but these students are not -allowed to purchase alcohol. The police routinely search all nightclubs and bars -in Albany for students under age 21 who have consumed alcoholic beverages or -used false identification to enter the establishment. Students who are caught -are arrested, taken to jail, and fined. They may also face university disciplinary -measures. - -Driving a car after drinking alcoholic beverages is also illegal in Georgia, -regardless of age. Driving under the influence (DUI) of alcohol carries severe -penalties and can result in the loss of your driver’s license, heavy fines, -imprisonment, disciplinary action by the university, and/or deportation. NEVER -DRINK AND DRIVE. - -SMOKING - -27 - - Smoking has become socially unacceptable in many places. Smoking is not -permitted in Albany State University buildings (except dorm rooms). Smoking is -also forbidden in hospitals, many restaurants, and other public places. Always -ask permission of your companies before smoking. Never smoke in someone’s -home unless he or she smokes as well. - -MAIL -CAMPUS MAIL -Campus mail is a service provided by the university for mail going between -people and offices on the premises of the campus. To send a letter by campus -mail, clearly mark the office and person it should go to and write CAMPUS MAIL -in the upper right hand corner. No stamp is needed. Drop the letter in the -collection point at the post office or in your Department Office. - -POST OFFICES -The university post office can send materials in both domestic and overseas mail. -Overseas mail must be weighed and stamped according to weight. Domestic -mail requires standard postage of 44 cents per letter. Packages and letters -containing more than five sheets of paper usually require extra postage. Any -package weighing over 16 ounces may be mailed first class from the university -post office. Domestic mail also requires a zip code for delivery. This is a five- -letter code specific to the area the letter or package is being sent to. If you do -not know the zip code of the area you are mailing to, ask for a zip code directory -at the post office and look up the proper zip code. - -CHANGE OF ADDRESS -Students who move off campus or leave ASU for a vacation should fill out a U.S. -Postal Service Change of Address card, sign the card, and place it in a U.S -mailbox. By completing this card, you allow the postal service to automatically -forward your mail to your new address. There is no charge for this service. -Unfortunately, the Office of the International Student Advisor cannot forward -mail to students who have moved off campus. Please do not have your mail -sent to the Office of the International Student Advisor. You will also need to -complete an AR-11 form which you obtain in the International Office. - -ELECTRONIC MAIL -An electronic mail (e-mail) account is available to all ASU students, faculty, and -staff free of charge. All ASI students are expected to use their ASU Ram Mail -because that is the official mode of communication. If you need an account or -would like to learn more about this service, call the Office of Information -Technology at 229-430-4909. You can also use the Internet through ASU -website: www.asurams.edu. E-mail and the Internet are wonderful ways to -communicate with friends around the world at no expense to you. - -MONEY AND BANKING -BANKS -There are many banks located in Albany. Most of them have specially priced -services for ASU students. The Bank of America has an automated teller machine - -28 - - or ATM in the Reese Student Union building on lower campus. As with any other -service, you should determine which banking services are most important to you -and call the different banks to find out about prices and available services. - -CHECKING AND SAVING ACCOUNTS -All students should open checking and/or saving accounts at a local bank. It is -not wise to carry or keep large amounts of cash. Most local businesses will -accept checks from local banks only. - -It is important to know how much money is in your checking account at all -times. Do not write checks for more money than you have in your checking -account. Checks written for more than the amount in an account are returned -unpaid. You will have to pay a fine to the bank, to the business where you -wrote the check, and may be unable to write checks in the future. Writing checks -when you do not have enough money to cover the check is called “bouncing -checks.” Bouncing checks is a criminal offense punishable by imprisonment. A -checking account is a service. Charges for services vary from bank to bank. -Savings accounts allow you to deposit money safely in a bank and receive a -small amount of interest on the balance. Savings accounts require passbooks for -depositing and withdrawing money. Keep both your checkbook and your -passbook in a safe place and report it to the bank immediately if either document -is lost or stolen. - -CREDIT CARDS -Credit cards are available through most banks. They have high interest rates, -but are very easy to use. Credit cards can be used almost everywhere and are -billed on a monthly basis. The best way to use a credit card is to pay the entire -bill monthly; otherwise, your balance may get higher than you can pay. Many -students find it helpful to carry an international credit card such as Visa or -MasterCard. Purchases on these cards can be made in U.S. dollars while the bill -is sent to your home and is denoted in your home currency. These cards can -also be used at automated teller machines to withdraw cash. Be sure to inquire -about the transaction fees before you obtain a credit card. Often the use of -these cards is an inexpensive way to exchange currency. But the fees can -become expensive. Your student fees can also be paid by credit card -(MasterCard or Visa only). - -OTHER BANKING SERVICES OFFERED FOR FEES - -Cashier’s Check -Money Orders -Certified Checks -Travelers Checks -Safety deposit box rental for storing valuable items - -MONEY EXCHANGE -Foreign currency must be taken to a bank to be exchanged. All currency (except -Canadian) must be sent out of town by the local banks for exchange. There is a - -29 - - charge of about $10.00 for the transaction and a processing time of 3 to 6 -weeks. - -RECEIVING MONEY FROM HOME - -1. - -2. - -By Mail -Tell your family to send numbered cashier’s check in U.S. -dollars, purchased from a bank in your country that has -business arrangement with a well-known U.S. bank. Make sure -that the check has a bank routing number on the bottom of the -check. - -By Telex -You may receive money from home by telex (wire). Wired -money generally arrives 5-10 days after it is sent from overseas. -It should be arranged through a local bank, and you should -have a checking or savings account established at the bank. -The receiving bank will charge about $12.00 as handling fee. It -is not possible to have money wired directly to your ASU -student account. - -Please be sure to request your funds from home well in advance of fee payment -deadlines. International transfers can often experience unexpected delays, so -you should plan ahead each semester. THERE IS NO FLEXIBITY IN FEE -PAYMENT DEADLINES. Please do not request an extension of the set fee -payment deadline. Failure to pay your student fees in a timely manner will result -in your having to return home. - -30 - - INTERNATIONAL STUDENTS -If English is not a student’s native language, if the student did not attend high -school in the United States, and if the student wishes to discuss taking the -Regents’ Test especially designed for International students, the student must -contact the Regents’ Test Director whose office is in 103 Simmons Hall (229-430- -4735 – flo.hill@asurams.edu). The local International Students’ Task Force -Committee has suggested that a student be considered for a special form of the -essay test after taking and failing the test at least once. There is no special test -for reading at this time; however, students may be given more time and may be -allowed to use their translation dictionaries. - -RELIGIOUS LIFE -Religion is a very important part of South Georgia culture. People will probably -invite you to visit their church and to take part in church sponsored social -activities. While these church services and activities can be very meaningful to -you, participation will not obligate you to join a church. Churches often attempt -to bring students into their church families by asking students to change their -belief to match the beliefs of the church members. If church members or clergy -make you feel uncomfortable about your religious belief or your values, feel free -tell them that you wish to maintain your own beliefs. If they cannot accept that, -find another church group! - -There are many churches located in the Albany area. Denominations include: -African-Methodist Episcopal, Apostolic, Assembly of God, Bahai, Baptist, Catholic, -Charismatic, Christian, Church of Christ, Church of God, Church of Jesus Christ of -Latter Days, Episcopal (Anglican), Holiness, Interdenominational, Jehovah -Witness, Lutheran, Nazarene, Pentecostal, Salvation Army, Seventh Day -Adventist, and Unitarian. All denominations are listed under “Churches” or -“Synagogues” in the Yellow Pages of the telephone book. - -SHOPPING -GROCERY STORES -All foods are located in one store. You will find fresh fruits, vegetables, canned -(tinned) goods, baby foods, meats, bread, cleaning supplies, paper products, and -frozen foods. Generally, grocery stores have lower prices than smaller stores. -There are several large grocery stores in Albany and some of them are within -walking distance of the University. You can order special items if you are -looking for ingredients not readily available in Albany stores. - -DRUGSTORES AND PHARMACIES -This is the only place where you can buy medicine that the doctor has ordered -for you. You can often buy many other things there as well. The most -convenient drugstore is CVS on Radium Springs Road, near the campus. Less -expensive pharmacies are located in discount stores and some grocery stores. - -31 - - DISCOUNT STORES -These are department stores where you find a large variety of things including -clothing, pharmacies, garden supplies, small appliance, paper products, linens, -and even some types of furniture. They often have lower prices and can be -wonderful places to see American people and products. Some of the local -discount stores include K-Mart, Wal-Mart, and Target. - -RESTAURANTS -Unless you are buying food, restaurant owners do not like for you to sit and -“take up space” from paying customers. You should leave money (a tip) on the -table for the waiter or waitress when you eat at a restaurant. The tip is usually -15%-20% of the total bill, depending on the quality of the service. Fast food -restaurants do not have waiters and waitresses. Do not leave a tip at this type -of restaurant. There are a number of commercial fast food restaurants around -the campus. Food can be ordered “for here” (to be eaten in the restaurant) or -“to go” (to be eaten at home). - -THE ALBANY MALL -The Albany Mall is a shopping center housing several dozen stores and -restaurants. It is located some miles west of the campus on Dawson Road. You -can buy almost anything you need in the Mall. Prices in these stores are usually -higher than the prices at Discount Stores, but the Mall is a great place to meet -people and experience American culture! - -OUTLET MALL -A large mall containing only manufacturer’s outlet stores is located many miles -south of Albany on Interstate 75. This mall has over 100 stores and is a good -place to shop because the prices are low while the quality of the merchandise is -high. - -Student Athlete - -Best practices For International Athlete Recruits (NCAA) -F-1 Visa and Issuance of SEVIS Form I-20 -International student-athletes who require an F-1 student visa are subject to -specific requirements regarding their admissions eligibility, financial ability and -academic status. Failure of students to follow appropriate regulations and -guidelines can result in severe difficulties with their immigration status with the -U.S. Department of Homeland Security. To avoid difficulties, the NCAA strongly -recommends that athletics staff meet with appropriate school officials (e.g., -international admissions officers, international student advisors) to gain an -understanding of the school’s policies and procedures that are in place to protect -international students and to maintain compliance with all immigration -regulations. - -32 - - IMPORTANT NOTE FOR ATHLETICS ADMINISTRATORS AND ADMISSIONS OFFICERS -Under no circumstances should an international student athlete be issued a Form -I-20 with the expectation of financial support from the athletics office before the -student has been certified as a final qualifier by the NCAA Eligibility Center or is -determined to be a partial qualifier in Division II. - -• Doing so opens the possibility for an international student who might not be -eligible for financial aid to obtain a visa and enter the United States without -having the sufficient funds to cover tuition and living expenses, as required by -immigration laws. - -Financial Ability - -• All F-1 visa students must submit appropriate documentation of financial -support before the school can issue the SEVIS Form I-20 [certificate of eligibility -for nonimmigrant (F-1) student status]. -• Generally the school’s office of recruitment and admissions office will have -information as to what constitutes appropriate documentation of financial -support. -• Athletics department staff members must understand that withdrawal of -athletics financial aid for an international student-athlete could result in severe -economic hardship for the student and could result in a violation of the student’s -immigration status. - -What Athletics Administrators Should Tell Student-Athletes -• Student-athletes who will be receiving financial aid from the athletics -department must understand that if the amount of aid is less than the -amount required for issuance of the I-20, then the student will need to -provide additional documentation of financial support. - -• Student-athletes should be informed of the true costs of attending the - -school and must understand the financial implications in the event of the -loss of financial support from athletics. - -• Although students athletes may receive some financial assistance through - -athletics students are responsible for purchasing their books for the -duration of the academic tenure. - -• Student-athletes must understand that if their scholarship is made - -unavailable for whatever reason (e.g., because the student-athlete did not -meet initial-eligibility requirement, they must have other means of -financial support. If sufficient finances are not available from other -sources, the student risks violation of immigration status and possible -deportation. - -SOCIAL SECURITY NUMBERS -A Social Security Number is a government identification number issued to all -residents of the United States. Most residents receive this number as children. -In the U.S., we do not carry passports or national ID cards. Instead, we use this -number and a driver’s license as our primary forms of identification for -employment records, taxation, opening bank accounts, university records and - -33 - - other business matters. All drivers’ licenses contain a color photograph of the -bearer as well as the current address and date of birth. - -For your admission to ASU, you were issued a student identification number -which has the same number of digits as the social security number. The -university uses this ID number for all of your university records. The number -also appears on your student identification card. While this number is sufficient -for your academic records, you will need to obtain a Social Security number -before you can work on campus. To obtain the number, you must complete an -application form and deliver it, along with a letter of verification from the ISA -and an offer letter from the person who is hiring you on letterhead, in person to -the Social Security Administration in Albany. There is no charge for Social -Security numbers. Call 1-800-772-1213 for more information. - -TAXES -Federal Law requires that foreign students pay income tax from sources within -the U.S. This income generally includes salary from any employment, on or off -campus, as well as scholarship money. Tax return is due on April 15 for the -previous year. In addition you may need to file a tax statement. If you have -concern about your tax obligation, you may wish to consult a certified public -accountant (CPA). The Internal Revenue Service is the branch of the federal -government responsible for income tax collection. This agency is often simply -called IRS. You can call the IRS directly at 1-800-829-1040 for additional -information. - -TELEPHONE AND TELEGRAMS -TELEPHONE -Telephone service in the U.S. is divided into two types: local and long-distance -(including international) service. Most people work with two different companies -to obtain telephone services: a local carrier for local service and long distance -carrier for long distance service. The local carrier for Albany is BellSouth. -Contact the company by calling 229-780-2355. You may need to pay a deposit -to begin your local service. For about $30.00 a month, you may have unlimited -telephone calls within the country. You will be billed monthly for this service. - -There are several local and long distance services in the U.S. Each company lays -claims to lowest telephone rates for long distance. You may want to talk to each -company to learn who can offer you the best deal. Rates usually depend on -where you call and when you call. To get the cheapest rate, you may find that -all three companies have special plans with many restrictions. Be sure to read -the details of the special plans. Be sure to ask for the rates for calling your -country. You can contact these companies by calling the numbers listed in the -local telephone book. Once you choose a company, they will provide you with -direct dialing from your home and a calling card, which you can use from any -phone. - -INFORMATION - -34 - - If you cannot find a number on-line, you may call “Information,” also called -“Directory Assistance” and the number is 1411. In order to use the information -service, you must know the name of the person or organization you wish to call -as well as the city they are located in. The number you dial for information -depends upon the area you wish to call. The telephone book also contains useful -information about how to use the telephone and place different kinds of calls. -The Yellow Pages found on-line contain listings of local businesses and can help -you locate services. - -TOLL FREE CALLING -Most numbers with an area code of 1-800 or 1-888 are toll free calls. This -means that there will be no charge for the call. Many businesses and services -have toll free numbers for customer use. You should use toll free numbers -whenever possible to keep your telephone bill low. - -OTHER ASPECTS OF TELEPHONE SERVICES -If you live off-campus and share a telephone number, be sure to have an -agreement about how the bill will be paid and whose name will appear on the -bill. The names and numbers of all telephone subscribes are published in the -BellSouth telephone book unless you request an unpublished number. Albany has -several “local exchanges.” This term refers to the first three digits of the -telephone number. You will see prefixes listed as 430, 431, 434, and 435, for -example. Be sure to give all seven digits of your number. If you are calling from -one phone on campus to another, you only need to dial the last four digits of the -number. Many students today buy pre-paid cell phones. - -TELEGRAMS -Letters or money may be sent both within the U.S. and internationally through -Western Union by calling their toll-free number 1-800-325-6000. Mailgram is also -available through Western Union and is less expensive than telegrams. To send -a Mailgram, call Western Union at 1-800-325-6000. MoneyGram provides similar -services as Western Union. Their toll free number is 1-800-926-9400. - -FAX -Faxes may be sent or received through the fax services offered by some -companies in Albany. You will be charged for this service. For international -faxes, students should use Mailboxes, etc. at 229-434-1133 or 229-432-2230. -Unfortunately, the high cost of international and long distance faxes prohibits the -Office of the International Student Advisor or the University from sending -personal faxes for students. - -TESTING -The Albany State University Computer-based Testing Center (Department of -Counseling, Testing and Disability Services), located on the first floor in ACAD -195, administers and prepares students for several national computer-based -tests periodically throughout the year. The GRE, GMAT, MAT, MCAT, LSAT, -PCAT, and TOEFL are a few of the tests administered by the center. National test -scores are required for admission to U.S. graduate schools. These tests have - -35 - - standard fees and deadlines for enrollment. For information about a test, call -the Computer-based Testing office at 229- 430-1631. To sign up for a test, visit -the Division of Student Affairs and pick up an application. - -TRANSPORTATION -TRANSPORTATION IN ALBANY -Albany has a public transportation system. In emergencies, you can hire a taxi -(cab). Companies offering taxi services are listed under “Taxicabs” in the -Albany telephone book. Be sure and ask how much you will be charged before -hiring a taxi. Some students purchase bicycles for transportation. Cyclists must -ride on the roadways, as there are no separate bicycle paths. If you choose to -ride a bicycle, you must be very careful and watch out for cars. American drivers -are not accustomed to cyclists on the road, be sure to lock up your bike when -you are not using it. To travel outside Albany, you must go by bus or airplane. - -BUS TRANSPORTATION -The Greyhound Bus Lines station in Albany is located at 300 Oglethorpe Blvd. -Call 1-800-231-2222 for nation-wide information or 229-432-0511 for local -terminal information. - -AIR TRANSPORTATION -Albany has an airport located south of town. Service is limited to several daily -flights to Atlanta, where connecting flights are available to other cities. Call 229- -430-5175 for schedules. - -TRAVEL AGENTS -Travel agents will help you make travel arrangements for a small fee. They -often save you money by finding the lowest rates available to your destination. -Look in the Yellow Pages under “Travel Agencies”. Or you may book your tickets -online using a credit card at the airlines website. (e.g. www.delta.com) - -INTERNATIONAL TRAVEL -If you know that you will return to your country during a school vacation, you -should try to plan international travel 2 or 3 months in advance. The best rates -for international travel are obtained by making reservations early. Also, all -students are eligible for discounted airfares through an organization called -Council Travel. To obtain these special fares, you must have an international -student ID card. These cards can be purchased for $20 from the Office of the -International Student Advisor or local travel agents. These cards also qualify the -bearer for discounts at museums and other tourist attractions. If you have -questions about international travel, you should speak with the staff at the Office -of the International Student Advisor or experienced travelers from your country. - -36 - - CAMPUS CONTACTS - -Quick Reference - -Information - -Office - -Phone Number - -Advising and Immigration - -Office of Int’l Student Advisor - -Meal Plan - -Food Services - -General Info (Switchboard) - -Facilities Management - -Graduate Studies - -Graduate School - -Residence Life (Housing) - -Student Affairs - -Mail delivery services for on-campus students - -Post Office - -Parking Permits - -Public Safety - -229-430-7075 - -229-430-4750 - -229-430-4600 - -229-430-5118 - -229-430-4741 - -229-430-4745 - -229-430-4711 - -Personal & Counseling - -Counseling, Testing & Disability Services - - 229- 430-4667 - -Reporting Theft or Crime - -Public Safety - -Admission Status - -Admissions Office - -Student Activities - -Student Affairs - -Transcripts (to be sent from ASU) - -Registrar’s Office - -Tuition Payment - -Business Office - -229-430-4711 - -229-430-4646 - -229-430-4739 - -229-430-4638 - -229-430-4610 - -37 - - SERVICES FOR INTERNATIONAL STUDENTS - - Orientation sessions and workshops for new students. - Liaison with the U.S. Citizenship and Immigration Services (USCIS) for -all of your immigration documents. We will help you with filing for -practical training, transfer of schools, and replacement of lost -documents. WE are your source for maintaining your immigration -status. - - Counseling and referral services. Personal and cross-cultural - -counseling is available on a confidential basis to all students. If you -would like to talk to someone, call the International Student Advisor -(ISA) at 229-430-4721 to make an appointment. If necessary, you -may be referred elsewhere for help. - - Information. If you do not know what to do, to whom to talk, or -where to find ANYTHING, the Office of the International Student -Advisor can help. - - Letters of certification for government loans, money exchange, re- -entry into the U.S., invitations to visit the U.S., and other official -documents are available at the office. - - Information about Albany State University International Student - -Association (ASUISA): ASUISA annually elects new student officers, -and sponsors activities throughout the year, by and for international -students. - - Other activities for all Global Programs students such as International - -Festival in November, picnics, coffee/tea hour, and the Cultural -Exchange Club/Global Ambassadors. - -38 - - CHECK LISTS - -ANNUAL CHECKLIST - -• Renew passport if necessary (six months before expiration). - -• File income tax returns if necessary (between January 1 and - -April 15). - -• Extend temporary stay with sponsor and with the - -International Advisor for an extended I-20 if necessary (30 -days before expiration). - -PERIODIC CHECKLIST - -• Report changes of address (within 10 days) to the Office of -the International Student Advisor and the Office of the -Registrar. This is required by law. So is your email -address and your telephone number. - -• Report any change of marital status, or if your family has - -joined you in the U.S., to the International Student Advisor. - -• Complete registration for Student Health Insurance every - -semester. - +IEEETRANSACTIONSONMULTIMEDIA 1 +RA-BLIP: Multimodal Adaptive +Retrieval-Augmented Bootstrapping +Language-Image Pre-training +Muhe Ding, Yang Ma, Pengda Qin, Jianlong Wu, Member, IEEE, Yuhong Li, Liqiang Nie, Senior Member, IEEE +rec A en bs tl t y ra r c e t c — ei M ve u d lt s i u m b o s d ta a n l t L ia a l r i g n e te L r a e n st g , u w a h g i e ch M s o h d o e w ls s ( t M he L ir L e M m s e ) r h g a in v g e (a) × × In +S +n +i +e +m +r +i +P +la +ro +ri +d +ty +uct (b) 𝐻 …["#$] … 𝐼) T +S +o +i +k +m +en +il +- +a +W +rit +i +y +se +potential as general-purpose models for various vision-language +𝐻,-./012 𝐼& +tasks.MLLMsinvolvesignificantexternalknowledgewithintheir ContrastiveLearning 𝐻["#$] 𝐻& 𝐻'()* …𝐻!"#$%&' 𝐼) 𝐼* 𝐼+ …𝐼& +parameters; however, it is challenging to continually update +these models with the latest knowledge, which involves huge Text Image Text Image +Encoder Encoder Encoder Encoder +computationalcostsandpoorinterpretability.Retrievalaugmen- +tation techniques have proven to be effective plugins for both +LLMsandMLLMs.Inthisstudy,weproposemultimodaladap- Arockband Arockband +isplaying isplaying +tive Retrieval-Augmented Bootstrapping Language-Image Pre- +training (RA-BLIP), a novel retrieval-augmented framework for (c) Multimodal Knowledge Semantic Space +various MLLMs. Considering the redundant information within +vision modality, we first leverage the question to instruct the +extractionofvisualinformationthroughinteractionswithoneset +of learnable queries, minimizing irrelevant interference during +retrieval and generation. Besides, we introduce a pre-trained +multimodal adaptive fusion module to achieve question text-to- MultimodalAdaptiveRA-BLIP +multimodalretrievalandintegrationofmultimodalknowledgeby +projectingvisualandlanguagemodalitiesintoaunifiedsemantic … … … … +space.Furthermore,wepresentanAdaptiveSelectionKnowledge +Generation (ASKG) strategy to train the generator to au- +tonomously discern the relevance of retrieved knowledge, which +Which streetwas paved with +realizes excellent denoising performance. Extensive experiments boards; Little Champlain +Street, Quebec City, 1916 or +o th n at op R e A n -B m L u I l P tim a o ch d i a e l ve q s ue si s g ti n o i n fi - c a a n n s t w p er e i r n fo g rm da a t n a c s e ets an d d em su o r n p s a tr s a se te s Quebec C w i i t n y t e R r u 2 e 0 S 1 a 0 int-Louis N Q aN w t u i Q ao i N e w t nu n b i Q a t o i e a e e wn tn b lu c i rt oai ee eH 2 n C ln cbr 0 ti Ha i e es2 C1 t l ctr y 0 i 0 o i Hs 21 tC. r R t y i00 oi Lic s u 1. r tR t oy i e S0 o L c uu . r iR S o e i tSi L s e a u c u i S i iot Seo S n s e a u t f t i iSrS i - ot ne Cs e L at f t er i - aS o o en C tL n - u t f eta ra o a- i te C L n s -d u ea aa a io t s dn - u aaa i s d a +Question Image Image-Text Document +the state-of-the-art retrieval-augmented models. +QuestionPrompt Inter-Modal Intra-Modal Relevant +Index Terms—Retrieval-augmented model, vision-language Instruction Interaction Interaction Irrelevant +pre-training, multimodal retrieval, open question answering. Question Image Image-Text Text +Embeddings Embeddings Embeddings Embeddings +I. INTRODUCTION +Fig. 1. Illustration of different multimodal retrieval approaches. (a) Cross- +THE birth of the Internet has triggered an unprecedented modality retrieval. (b) Late-interaction retrieval. (c) RA-BLIP multimodal +information revolution, catapulting humanity into the era adaptive retrieval. For RA-BLIP, questions, documents, images, and image- +textpairsareprojectedintoaunifiedmultimodalspace. +of information explosion. It is a great challenge to efficiently +find answers from a vast amount of information based on +our questions. Open Multimodal Multihop Question Answer- +BLIP-2 [10], GPT-4 [11], etc., have been notably explored to +ing (MMQA) [1]–[7] can help alleviate this problem of in- +enhancetheirperformancebyimplicitlyencodingasubstantial +formationoverloadbyretrievingexternalknowledgebasedon +amount of external knowledge within their parameters, which +questionsandgeneratingcorrectanswers.Inrecentyears,sev- +now scale into the hundreds of billions [12]. While these +eraladvancedLLMsandMLLMslikeFlanT5[8],LLaMA[9], +models have yielded exciting results on various multimodal +ThisworkwassupportedinpartbytheNationalNaturalScienceFoundation tasks, they have also encountered high computational costs +ofChinaunderGrant62376069,inpartbyYoungEliteScientistsSponsorship and significant challenges in terms of interpretability. +ProgrambyCASTunderGrant2023QNRC001,andinpartbyGuangdongBa- +To alleviate the challenge, many researchers proposed re- +sicandAppliedBasicResearchFoundationunderGrant2024A1515012027. +Muhe Ding, Jianlong Wu and Liqiang Nie are with the School trievalaugmentationtechniquesthatdividethemodelintotwo +of Computer Science and Technology, Harbin Institute of Technology keycomponents:theretrieverandthegenerator[13]–[16].The +(Shenzhen),Shenzhen518055,China(e-mail:dmh1216380870@gmail.com, +retriever accesses relevant knowledge from a knowledge base +jlwu1992@pku.edu.cn,nieliqiang@gmail.com). +Yang Ma is with the School of Computer Science, University of Sydney, based on the posed question, while the generator leverages +Sydney,NSW2006,Australia(e-mail:yama5878@uni.sydney.edu.au). this information to create textual output in response. In ear- +Pengda Qin and Yuhong Li are with the Security Department, Al- +lier stages, text-modality retrieval-augmented models, such as +ibaba Group, Hangzhou 311121, China (e-mail: qinpengda0406@163.com, +daniel.yuhong@gmail.com). REALM [13], RAG [17], and so on [18], [19], have been +4202 +tcO +81 +]MM.sc[ +1v45141.0142:viXra + +| IEEETRANSACTIONSONMULTIMEDIA | | | | | | | | | | | | | | | 2 | +| ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +proposed to solve text-only question answering. They build an adaptive selection knowledge generation strategy, which +the dense index as a non-parametric document memory from gives the generator the capability of selecting knowledge by +extensive textual sources like Wikipedia for effective knowl- data enhancement to make the model automatically judge the +edge retrieval, and the generator produces answers based on relevance of knowledge. ASKG strategy allows the generator +the retrieved knowledge. More recently, multimodal retrieval- tonotsimplyrelyonthewordsimilaritybetweenthequestion +augmented models, such as MuRAG [14], SKURG [20], and and knowledge, but to understand the semantic information +soon[16],[21],haveemergedoneafteranother.Thesemodels of question and know which knowledge contains the answer. +extend the knowledge memory across various modalities, em- Furthermore, the parameters of the image encoder and LLM +ployingpre-trainedvisuallanguagemodelstoretrieverelevant of our framework are frozen, significantly reducing computa- +evidence and support reasoning for answers. tionalcosts.ExtensiveexperimentsonthreerepresentativeQA +However, existing methods exhibit certain limitations. The datasets demonstrate the effectiveness of our methods. +first limitation is the insufficient integration and interaction Overall, our key contributions are as follows: +| between | vision | and | language. | On | the one | hand, | existing | | | | | | | | | +| ------- | ------ | -------- | ----------- | --- | ---------- | ------------ | -------- | --------- | ------- | ---------- | ---------- | -------- | -------- | -------- | ---------- | +| | | | | | | | | • We | propose | a novel | multimodal | | adaptive | | retrieval- | +| methods | lack | explicit | integration | of | multimodal | information, | | | | | | | | | | +| | | | | | | | | augmented | | framework, | which | achieves | | question | text-to- | +hindering the alignment of questions and multimodal knowl- multimodalretrievalandknowledge-intensivemultimodal +| edge in | the semantic | | space. | As shown | in | Fig. | 1(a), some | | | | | | | | | +| ------- | ------------ | --- | ------ | -------- | --- | ---- | ---------- | --- | -------------- | --- | ------ | ------------ | --- | ---------- | --- | +| | | | | | | | | QA | by integrating | | visual | and language | | modalities | and | +methods [14], [22], [23] have employed separate visual en- projecting them into a unified semantic space. +| coder and | text | encoder | for individual | | modality | encoding | and | | | | | | | | | +| --------- | ---- | ------- | -------------- | --- | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +• Weintroduceanadaptiveselectionknowledgegeneration +| adopted | contrastive | learning | | [24] for | multimodal | | alignment | | | | | | | | | +| ------- | ----------- | -------- | --- | -------- | ---------- | --- | --------- | -------- | ---- | --------- | --- | -------- | ------------ | --- | ------- | +| | | | | | | | | strategy | that | leverages | the | powerful | capabilities | | of LLMs | +to retrieve. This may lead to an unbalanced and biased to select the relevant retrieved knowledge for answer +| multimodal | retrieval | | and reasoning | | process | towards | specific | | | | | | | | | +| ---------- | --------- | --- | ------------- | --- | ------- | ------- | -------- | --------- | --- | ------------- | --- | --- | --- | --- | --- | +| | | | | | | | | reasoning | | autonomously. | | | | | | +modalities.Besides,late-interactionretrievalapproaches[25]– We conduct extensive experiments on various multi- • - -If leaving the U.S. temporarily, have 1-20 or DS-2019 -signed by a Designated School Official and check visa -validity. - -• Check in with International Student Advisor at the beginning -of each semester and schedule appointments periodically. -Always bring all of your documents when you come. - -Information from Handbook is from the following websites. -www.travel.state.gov/visa (U.S. Department of State) -www.ice.gov (Department of Homeland Security) -www.uscis.gov (U.S. Citizenship and Immigration Services) -www.ncaa.com (National Collegiate Athletic Association (NCAA)) - -39 - - Office of the International Student Advisor -Sue Ann Balch, J.D. -205 Reese Hall Lower East Campus, 504 College Drive -(229) 430-7075 (Office) -(229) 434-7106 (Fax) -Sueann.balch@asurams.edu (email address) - -40 - +[27] in Fig. 1(b), retain dual-encoder independent encoding modal and multihop datasets (i.e., WebQA [4], Multi- +| architecture | and | perform | token-wise | | interactions | only | in the | | | | | | | | | +| ------------ | --- | ------- | ---------- | --- | ------------ | ---- | ------ | ------- | --- | -------- | ------ | ----- | ------- | --- | ------ | +| | | | | | | | | modalQA | | [5], and | MMCoQA | [6]). | RA-BLIP | | demon- | +late scoring stage, which sacrifices the retrieval efficiency for strates superiority over the existing state-of-the-art +thebenefitsoffine-grainedfeaturelearning.Ontheotherhand, +| | | | | | | | | retrieval-augmented | | | models. | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | ------- | --- | --- | --- | --- | +existingmethods[14],[20],[28],[29]donotutilizequestions +| to instruct | the | image | encoder | in selectively | | extracting | visual | | | | | | | | | +| ------------- | ----------- | -------------- | ------- | -------------- | ------------- | ---------- | ---------- | ------------------ | --- | ------------ | ----------- | ----- | ------- | ---------------- | ------ | +| | | | | | | | | | | II. | RELATEDWORK | | | | | +| features, | and | thus suffer | from | interference | | and noise | caused | | | | | | | | | +| | | | | | | | | A. Vision-Language | | Pretraining | | | | | | +| by redundant | information | | in | images. | Moreover, | such | methods | | | | | | | | | +| lack mutual | instruction | | when | encoding | different | | modal fea- | | | | | | | | | +| | | | | | | | | Vision-language | | pre-training | | (VLP) | aims | to train | models | +| tures, making | | it challenging | | to model | relationships | | between | | | | | | | | | +| | | | | | | | | on large-scale | | image-text | datasets | to | capture | the relationship | | +multiple sources. The second limitation is that existing ap- between these two modalities. Broadly, VLP methodologies +proachesdonotinspectthecorrectnessoftheretrievedrelevant +| | | | | | | | | fall into | two categories | | based | on their | training | approach: | 1) | +| --- | --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | ----- | -------- | -------- | --------- | --- | +knowledge at the generation stage. However, the retrieved End-to-end Methods: This category includes methods [35]– +| knowledge | contains | significant | | noise, | resulting | in poor | model | | | | | | | | | +| --------- | -------- | ----------- | --- | ------ | --------- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | +[38]thattrainmodelsend-to-end,backpropagatinglearnedsig- +| anti-interference | | and | robustness. | The | generator | assumes | all | | | | | | | | | +| ----------------- | --- | --- | ----------- | --- | --------- | ------- | --- | --------------- | --- | ------ | -------- | ------- | --------- | --- | ----------- | +| | | | | | | | | nals to achieve | | mutual | learning | between | different | | modalities. | +retrieved relevant knowledge is correct, potentially leading to 2)ModularMethods:Incontrast,modularmethods,asseenin +| the utilization | | of irrelevant | or | confusing | information. | | | | | | | | | | | +| --------------- | --- | ------------- | --- | --------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +worksby[39]–[43],involvekeepingtheparametersofspecific +To address the above issues, we propose multimodal adap- pre-trainedcomponents(likeimageencodersorlargelanguage +tiveRetrieval-AugmentedBootstrappingLanguage-ImagePre- models) fixed while focusing on refining other aspects of the +| training | (RA-BLIP). | | RA-BLIP | consists | of | two key | compo- | | | | | | | | | +| -------- | ---------- | --- | ------- | -------- | --- | ------- | ------ | ------ | ------------- | --- | ---- | -------- | ------------- | --- | ------ | +| | | | | | | | | model. | For instance, | LiT | [44] | utilizes | a pre-trained | | frozen | +nents: a multimodal adaptive retrieval-augmented framework image encoder from CLIP, while Flamingo [45] and BLIP-2 +| and an | adaptive | selection | | knowledge | generation | | (ASKG) | | | | | | | | | +| ------ | -------- | --------- | --- | --------- | ---------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | +[10]freezethelanguagemodeltointegrateLLMsintovision- +strategy. To tackle the first limitation, RA-BLIP is based on language tasks better. Besides, instruction tuning is also an +the InstructBLIP architecture [30] and adopts Q-Former to effective approach during VLP. InstructBLIP [30] represents +implementinstruction-awarevisualfeatureextractionthatuses +arecentadvancementinthisarea,achievinginstruction-aware +questions as instructions. The question instruction interacts visual feature extraction and instruction-guided LLM genera- +withthequeryembeddingsthroughsharedself-attentionlayers +| | | | | | | | | tion. The | unique | capability | of | InstructBLIP | to | extract | features | +| -------------- | --- | -------------- | --- | -------------------- | --- | --- | ----------- | --------- | ------ | ------------- | --- | ------------ | ---- | ------- | ----------- | +| and encourages | | the extraction | | of question-relevant | | | visual fea- | | | | | | | | | +| | | | | | | | | based on | prompt | instructions, | | combined | with | its | utilization | +tures. Additionally, we incorporate a pre-trained multimodal of frozen LLMs and image encoders, positions it as an ideal +| adaptive | fusion | module | to fuse | vision | and | text information, | | | | | | | | | | +| ------------- | ---------- | -------- | ------------------ | ---------- | --- | ----------------- | --------- | -------- | ------- | -------- | ------------------- | --- | --- | ---------- | --- | +| | | | | | | | | backbone | for our | proposed | retrieval-augmented | | | framework. | | +| obtaining | multimodal | | features | [31]–[33]. | | As a result, | RA- | | | | | | | | | +| BLIP achieves | | question | text-to-multimodal | | | retrieval | by align- | | | | | | | | | +ing the questions and multimodal knowledge bases in the B. Text-modality Retrieval-Augmented Models +semantic space of three modalities: text, image, and image- Retrieval-augmented techniques have proven to be ef- +text [34], as shown in Fig. 1(c). For the second limitation, fective plugins for both LLMs and MLLMs in academia. +we leverage the implicit capabilities of LLMs and introduce These techniques extract pertinent world knowledge from + +IEEETRANSACTIONSONMULTIMEDIA 3 +extensive databases, subsequently integrating this information A. Problem Formulation +to formulate answers. Pioneering methods like ORQA [3] +This paper presents a multimodal adaptive retrieval- +have employed inverse cloze tasks for retriever pre-training, +augmentedframeworkcalledRA-BLIPforopenmultihopand +showcasing their efficacy on open-ended question answering +multimodalQA,integratingretrievalandgenerationfunctions. +datasets. Following suit, REALM [13] extends this by retriev- +For knowledge-intensive QA, we deconstruct the task into +ing and processing documents from comprehensive sources +two stages: retrieval and generation, which are implemented +like Wikipedia for logical reasoning, which is pre-trained to +by the retriever and generator respectively. The goal of our +reason over a large corpus of knowledge on the fly during model training is to learn the distribution P(y|x ) to generate +q +inference. Furthermore, RAG [17] adopts a pre-trained model a textual output y conditioned on input question x and +q +andnon-parametricmemoryforlanguagegeneration.FiD[18] multimodal knowledge base KB (KB = k ,...,k ). Firstly, +1 n +leveragesencoder-decodertransformermodelsforknowledge- +the retriever encodes questions, images, and texts from the +intensivetasks,settingnewbenchmarksinQA.Morerecently, knowledge base KB. It identifies the most relevant retrieved +RETRO [19] has advanced these methods by handling longer knowledge, K ⊂KB (K is the retrieved knowledge) for +ret ret +sequences and accessing diverse documents for segmented each question x , which is modeled as p(K |x ). Secondly, +q ret q +sequences from expansive retrieval datasets. Text-modality the generator utilizes an LLM to generate answers y, condi- +retrieval-augmented models construct dense indices as non- +tionedonboththequestionandtheretrievedknowledge,which +parametricdocumentmemories,usingextensivetextualknowl- is modeled as p(y|x ,K ). We treat multimodal knowledge +q ret +edge bases to align the retrieval process with specific queries. K asalatentvariablefromtheexternalknowledgebaseand +ret +Despite these advancements, a notable limitation remains the +marginalize it to increase the overall likelihood of the answer +needforthesemethodstoeffectivelyleveragevastmultimodal y. The overall process is encapsulated in the equation: +knowledge, thereby constraining their applicability in the +(cid:88) +domain of open multimodal question answering. p(y |x q )= p(K ret |x q )·p(y |x q ,K ret ). (1) +(cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) +Kret⊂KB +Retrieval Generation +C. Multimodal Retrieval-Augmented Models This dual-stage framework effectively addresses the complex- +To overcome the limitations of text-modality retrieval- ities of open multimodal QA by balancing the retrieval of +augmented models, recent research [10], [15], [30], [46], [47] multimodal data and knowledge-based generation, and has +has made strides in integrating multimodal knowledge. No- been validated by extensive experiments and ablation studies. +tableefforts,includingAutoRouting[5]andMAE[6],involve +trainingdistinctmodelsforeachmodalityandusingclassifiers B. Model Architecture +for task-specific routing, though this approach often hampers +RA-BLIP is built on a simple backbone model that is pre- +cross-modal reasoning. MuRAG [14] seeks to overcome this +trained to encode image-text pairs so that they are suitable +limitation by employing separate encoders for visual and +for both knowledge base retrieval and answer generation. +textualmodalities,followedbyajointencoderformultimodal +The overall framework of RA-BLIP is shown in Fig. 2. The +fusion. However, this approach lacks integrated guidance for +backbone model consists of a multimodal encoder f (·) and +θ +different modalities and cannot model the relations between +decoderg (·),whichareusedascomponentsoftheRA-BLIP +θ +knowledge sources during retrieval. SKURG [20] attempts to +model to implement retrieval and generation. The multimodal +bridge this gap by using an entity-centered fusion encoder +encoder f (·) contains a frozen image encoder ViT [48], +θ +to align modalities, yet faces challenges in computational +Q-Former architecture [30], and the pre-trained multimodal +efficiency and limited interpretability. Besides, Solar [21] +adaptive fusion module. The decoder g (·) is composed of +θ +transforms multimodal inputs into a unified language format +a LLM FlanT5 [8]. Querying Transformer (Q-Former) [10] +but falls short in handling complex tasks and generalizing +is a lightweight Transformer consisting of two modules that +visual information. REVAL [16] leverages large-scale knowl- +sharethesameself-attentionlayer:oneisanimagetransformer +edgegraphstoassistvisuallanguagepre-training,butitbrings +that interacts with the frozen image encoder ViT for visual +a lot of calculations. In contrast, RA-BLIP distinguishes itself +featureextraction,andtheotherisatexttransformercanactas +by seamlessly integrating visual and language modalities into +both text encoder and text decoder for text feature. The visual +a cohesive semantic space, enabling the autonomous selection +encoder, composed of ViT and Q-Former image transformer, +of relevant knowledge for reasoning, thus addressing these +hasinstruction-awarevisualfeatureextractioncapabilitiesand +limitations more effectively. +can extract visual information based on question instructions. +We input N learnable query embeddings into the Q-Former +III. METHODOLOGY +imagetransformer,whichinteracts withfrozenimagefeatures +In this section, we first formulate the research problem through cross-attention layers to obtain visual representation +and subsequently elaborate on the model architecture of our f (I) ∈ RN×D, where D is the hidden dimension of the Q- +θ +retrieval-augmented framework. Then, we describe the learn- Former.Additionally,weusetheQ-Formertexttransformerto +ablequeryinteractionapproach,followedbymultimodaladap- encode text, taking the [CLS] token as the text representation +tive fusion module. Subsequently, we show how to train the f (T)∈R1×D.Toobtainmultimodalfeaturescombiningboth +θ +retrieverandranktherelevantknowledge.Lastly,weintroduce imageandtext,weintroduceapre-trainedmultimodaladaptive +the adaptive selection knowledge generation strategy. fusion module M (·) to obtain the multimodal representation +θ + +IEEETRANSACTIONSONMULTIMEDIA 4 +Q:The Marina Bay Sands +in Singapore is made of +what building material? +Retrieval Knowledge +Template: {Q}. Which item is related to this question? Positive & Negative Knowledge Pair +Answer +Generation Stage: +Q-Former +Q-Former +Q-Former +Fusion +Multimodal +LLM +Retrieval Stage: Learnable queries +🔥 🔥 +Image 𝑓!(𝑘"#) +… Encoder 𝑓!(𝑘") +Marina Bay Q K1 🔥Q-Former +Sands Singapore 🔥 +Q:The Marina Bay Sands 𝑓!(𝑘#) +in Singapore is made of +what building material? +It is close to some of Singapore's famous In Asia, the Marina 🔥 S l u a n n t d e m c a C r i k t s y , s u M c a h r i a n s a S B i a n y g a S p a o n r d e s w lo a c s a a te d d d e in d 𝑓!(𝑘#) Bay Sands, Esplanade, +to the company's portfolio in 2010… +… +ModalityRepresentation +… +Text Question +QA Data Format +Image Relevant Knowledge Input +Learnable queries +Image-Text +concat +Q K2 Image Encoder +Instruction Top-K Q K3 Q K4 Question& Caption +Marina Bay ASKG Data Format Sands Singapore +FeatureEncoder +In 2010, when it opened, at total cosTt he Downtown MRInT Asia, the Marina 🔥 Classifier inc o lu f d S in $ g 8 l b a i n ll d io c n o stm, st e a t t e io rs n t i o s t b h u e i l W t a e s fe St B w oi a nf y g a S p a o n r d e s w lo a c s a a te d d d e in d Marina Bay Sands… the building. to the company's portfolio in 2010… +Maximum Inner Product Search Ranking +Fig. 2. The overall workflow of RA-BLIP consists of a retrieval stage and a generation stage. We utilize multimodal encoder f θ (·) to project questions +and multimodal knowledge into a unified semantic space to achieve question text-to-multimodal retrieval, and exclude confusing knowledge by ranking. +Additionally, we employ ASKG strategy to filter out invalid knowledge, enabling precise reasoning. The parameters of LLM and image encoder are fixed, +onlytheQ-Formerandclassifieraretrainable. +f (IT) ∈ R(N+1)×D. Consequently, the multimodal encoder Q-Former has aligned visual and text feature representation, +θ +can simultaneously encode image, text, and image-text fea- we use Image-Text Matching loss and Image-grounded Text +tures. Questions and knowledge are encoded into multimodal Generation loss for pre-training. As shown in Fig. 3, our +information through the multimodal encoder and then input approach is to fix the parameters of the image encoder and +into the decoder for answer generation. In the generation Q-Former, and solely fine-tune the parameters of the multi- +stage, compared with the retriever, the multimodal encoder modal adaptive fusion module. The module concatenates the +discards the multimodal adaptive fusion module to reduce the visual embedding and text embedding with a dimension of +computational cost. R(N+L)×D, where N is the learnable query embeddings and +L is the length of text tokens. Image-text matching loss is +used to fuse image and text representations, and the results +C. Learnable Query Interaction for Multi-images +of the fusion module are fed into a binary linear classifier +Questions and image captions are used as instructions to +foreachoutputquerytoobtainthelogitsandtaketheaverage +extract visual features to get learnable query embeddings +logitsofallqueriesasthematchingscore.Givenapre-training +and input them together with text knowledge to LLMs for dataset X = {I ,T }n , we randomly sample negative texts +i i i=1 +generation. We employ a novel approach for extracting visual +foreachimageandrandomlysamplenegativeimagesforeach +information to alleviate the burden of LLMs in distinguishing +text, to generate negative training data. Therefore, we denote +knowledge. In the original Q-Former in [30], multiple images +the ground truth label as y ∈ {1,0} for each image-text pair +are processed by employing multiple separate sets of queries +(I ,T ), indicating if the input image-text pair is relevant or +i i +for each, with each set of queries independently extracting +not.Weusethemultimodalencoderf (·)toencodeimage-text +θ +visualfeatures.Thisresultsinusingmultiplequeryfeaturesfor +pairs and input it into the multimodal adaptive fusion module +generation, which can be computationally intensive and less +M (·). The objective function is defined as follows: +θ +efficientincapturingtheinterrelationsamongdifferentimages. +1 (cid:88) +In contrast, our method innovates by succinctly utilizing one L =− ylog(ρ(M (f (I );f (T )))), (2) +itm n θ θ i θ i +set of learnable queries to directly interact with and extract Ii,Ti∈X +features from multiple images in a unified manner. This pro- whereρ(·)isthesoftmaxfunction.Image-groundedTextGen- +cessoccursduringtheQ-Formerstage,enablingmoreefficient +eration loss trains the fusion module to generate texts, given +and integrated interaction among multiple visual references. +input images as the condition [10], [30]. For the image-text +By employing one set of query interaction approach, RA- pairs in the pre-training dataset, each image I corresponds to +BLIP not only simplifies the feature extraction process but a text sentence y ={y ,...,y } of length T. We employ a +1:T 1 T +also enhances the efficiency of information extraction. This +multimodalcausalself-attentionmaskformultimodalencoder +unified interaction allows the model to understand better and f (·)andmultimodaladaptivefusionmoduleM (·)tocontrol +θ θ +represent the collective information presented in multiple im- +the interaction between queries and text. The visual query +ages, enabling more effective and cohesive feature utilization, +functions as a prefix causal, ensuring that queries can attend +especially when dealing with complex scenes or subjects +to each other while excluding text tokens. Similarly, each text +across multiple images. token y can attend to all visual queries and preceding text +tokens. The loss function is defined as: +D. Multimodal Adaptive Fusion Module T +(cid:88) +L =− logM (f ((y |y ,I)). (3) +The pre-trained multimodal adaptive fusion module M (·) itg θ θ t Wi-Fi Calling. Slide the switch to the + +. + +On + position to enable it. You must have an active Wi-Fi +connection and have Wi-Fi calling service activated before it can be +used. + +Call history + +Every telephone number called and received will be saved in the history. +1. From the Home screen, tap Phone +2. Tap Recents on the bottom bar to display the call history. +You can tap a contact in the call history to display more options. + +. + +Call settings + +1. From the Home screen, tap Phone +2. Tap Options > Settings to configure your phone settings. + +. + 9 -10 + 10 + Contacts + +Add a contact + +1. From the Home screen, swipe up to access Apps and tap Contacts + +. +2. Tap +3. Use the keyboard to enter contact information, such as picture, + + to add a contact. + +name, telephone number, group, address, email, etc. + +4. When you are finished, tap Save. + +Edit a contact + +To edit a contact, tap the contact from the Contacts screen to view its +details, and tap + +. + +Messages + +Send a text or multimedia message +1. From the Home screen, tap Messaging +2. Tap Start chat to create a new text or multimedia message, or start + +. + +a new group chat. + +3. Enter the recipient(s) name from your Contacts list or phone number + +in the To field. + + to send the text message. + +4. Type your message and tap +5. Or tap + + to the left side of the text box to add location, contacts, + +or files. Tap +recordings. When done, tap + + to attach photos or videos. Tap + + to send sound + to send the multimedia message. + +Message settings +From Message screen, tap + on the top right corner of the Messages +screen and tap Messages Settings to adjust your messages settings. + 11 -11 -13 + +12 + + Email + +Email setup (Gmail, Yahoo!, etc.) + +From the Home screen, tap Gmail +your account. + +, and follow the steps to set up + +Create and send an email message +1. Tap Compose from the Inbox screen to compose a new message. +2. Enter the recipient(s) email address in the To field. To add a carbon copy + +or blind carbon copy, tap + +. + +3. Enter the subject and content of your message. To add attachments, tap + + and select Attach file. + +4. Tap + + to send. + +Email settings +From Email screen, tap Menu +settings. + + > Settings to configure your Email + +Chrome browser +From the Home screen, tap Chrome +for information, and bookmark your favorite web pages for convenient +access. + + to browse the Internet, search + +Browsing web pages +From the Chrome browser screen, tap the address field and enter a web +address or a keyword to browse for web pages. + +Adding a bookmark +Open the web page you want to bookmark and tap Menu > + +. + +Opening a bookmark +From the Chrome browser screen, tap Menu > Bookmarks, select +the bookmark of the website you want to view. + +Browser settings +From the Chrome browser screen, tap Menu > Settings to customize +the browser. + 13 -14 -14 + 14 + Bluetooth + +Wi-Fi + +Bluetooth is a short-range wireless communications technology that +devices can use to exchange information and connect to different types +of devices which include headsets and other devices. + +Enable Bluetooth + +1. From the Home screen, swipe up to access Apps and tap Settings +2. Tap Connected devices > Connection preferences > + +. + +Bluetooth. + +3. Slide the Bluetooth switch to the On + + position. + +Pair Bluetooth devices + +1. From the Connected devices screen, tap Pair new device. +2. Tap a device under Available devices and follow the on-screen + +instructions to pair with the device. + +3. When the device is connected to your phone, You can then use the +headset or car kit to listen to music or make and receive phone calls. + +To use your phone’s Wi-Fi, you need access to a wireless access point or +hotspot. +1. From the Home screen, swipe up to access Apps and tap Settings +2. Tap Network & internet > Internet > Wi-Fi. +3. Slide the Wi-Fi switch to the On +4. Tap a Wi-Fi network to connect, enter the password (if applicable), + + position to enable Wi-Fi. + +. + +and then tap CONNECT . + +5. Your phone will be connected to the Wi-Fi network. You will see + +in the status bar. + +You can drag down the status bar twice to access Quick Settings panel +to enable or disable Bluetooth and Wi-Fi quickly and easily. + 15 -15 -16 + 16 -17 + + Camera + +Photos + +Take a photo +1. From the Home screen, tap Camera +2. Tap Photo to switch to Photo mode. +3. Aim the camera at your subject and tap + + to take the photo. + +. + +Record a video +1. From the Camera screen, tap Video to switch to Video mode. +2. Tap +3. Tap + + to start recording a video. + again to stop recording and save your video. + +Camera settings +From the Camera screen, tap Settings + + to adjust camera settings. + +You can view photos and watch videos that you’ve taken, downloaded, +or copied onto your storage card or phone memory. + +Edit photos +1. From the Home screen, tap Photos +2. Tap the picture you want to edit. +3. Tap +save. + +. + + to access the Edit screen. When finished, tap Save copy to + +Share pictures and videos +1. Select the photos or videos you want to share. +2. Tap + +, select a sharing method, and follow the on-screen prompts + +to send the photos or videos. + 17 18 + Calendar + +Calculator +Calculator + +Use the Calendar to create and manage events, meetings, and +appointments. + +Add an event + +1. From the Home screen, swipe up to access Apps and tap Calendar + +. +2. Tap Add +3. Enter event details, and then tap Save. + + to add an event. + +Calendar settings + +1. From the Calendar screen, tap Menu +2. Tap an option to adjust Calendar settings. + + > Settings. + +Your phone’s convenient built-in calculator lets you perform basic +Your phone’s convenient built-in calculator lets you perform basic +mathematical equations. +mathematical equations. +1. From the Home screen, swipe up to access Apps and tap Calculator +1. From the Home screen, swipe up to access Apps and tap Calculator + +. +. + +2. Enter the first number, tap a mathematical function key, and then +2. Enter the first number, tap a mathematical function key, and then + +enter the second number. +enter the second number. + +3. Tap +3. Tap + + to view the result, tap + to view the result, tap + + to clear the results. + to clear the results. + +View calculator history +View calculator history + +From the Calculator screen, tap Menu > History, or drag down on +From the Calculator screen, tap Menu > History, or drag down on +the history bar +the history bar +To clear the calculator history, tap Menu > Clear. To return to the +To clear the calculator history, tap Menu > Clear. To return to the +. +calculator, drag the history bar +. +calculator, drag the history bar + + back up or tap the Back button + back up or tap the Back button + + to view the calculator history. + to view the calculator history. + +19 + 20 -www.bc.edu/international + Clock - CHECKLIST +Google Maps - BEFORE YOU LEAVE HOME +The Clock app allows you to set alarms, view different world times, use +a stopwatch, set a timer, manage your bedtime, and use your phone as +a desk clock. -Apply for F-1 or J-1 visa from local US Embassy or Consulate -Make travel plans for arrival and departure from BC -Secure on-campus or off-campus housing. If necessary, arrange tem- +Use the Google Maps app to determine your location, get directions, +browse local businesses and attractions, rate and review places, and +more. -porary housing or hotel +Set the date and time manually -Complete and submit Immunization and Health Forms to BC - University Health Services +1. From the Home screen, swipe up to access Apps and tap Clock +2. Tap Menu > Settings > Change date & time > Set time +automatically > Off, and set the date and time manually. - AT BOSTON COLLEGE +. -Check in at the Office of Residential Life to pick up keys and get hous- +Create a new alarm -ing assignment or locate off-campus housing -Check in at the Office of International Programs -Check in at the Office of International Students and Scholars -Attend orientation. Dates, times, and locations are in your acceptance +1. From the Clock screen, tap Alarm +2. Set the time for the alarm and tap OK. -packet + > -Get student identification card, set up e-mail account, and establish -library account. This process will be explained at the orientation. +. -3 +Search for destinations +1. From the Home screen, swipe up to access Apps and tap Maps +2. Tap Search here. +3. Enter an address, city, facility name, etc., and then choose a search -www.bc.edu/international - - INTRODUCTION - -Office of International Programs - -The Office of International Programs (OIP), located in Hovey House at 258 -Hammond Street, is your primary contact while you are studying at BC. The -exchange program that brought you here is administered jointly between OIP -and a comparable office at your home university. In addition to admitting you -to Boston College, we pre-register you for courses, assist with housing selec- -tion, and can assign you to a faculty academic advisor if you wish. We also -maintain regular contact with our counterparts at your home university. -Please feel free to come to us with any questions or problems before, during, -or after your semester or year at Boston College. - -Our office also sponsors the Exchange Mentor Program. This program aims -to match you with a Boston College student who has studied abroad at your -home university. Mentors maintain regular contact with their students. - -OIP works alongside numerous offices on the Boston College campus to en- -sure that you have the same access to facilities, services, and opportunities -as all other BC students. Other offices that you may come into contact with -include: the Office of International Students and Scholars, Student Services, -Health Services, Residential Life (which encompasses the On-Campus Hous- -ing Office and the Off-Campus Housing Office), and various academic de- -partments. - -Office of International Students and Scholars - -The Office of International Students and Scholars (OISS) is responsible for -issuing the immigration documents required for international students to -apply for a student visa. They also issue work authorization and advise on -other issues related to maintaining legal status in the United States. In col- -laboration with OIP, OISS organizes a mandatory International Student Ori- -entation each semester. - -In addition to these important services, OISS sponsors programs for interna- -tional students throughout the year, including welcome socials, Boston walk- -ing tours, the Thanksgiving Host Family Program, and many other activities. -They also provide undergraduate international students with International -Assistants, who are BC students serving as peer advisors to international stu- -dents. You will meet these International Assistants during orientation. +. -4 +result. You will see your location on the map. -www.bc.edu/international - - Things to Know Before You Leave Home - -Student Visa - -As a participant of an exchange pro- -gram to Boston College, you will -most likely be entering the United -States either on an F-1 student visa -or a J-1 exchange visitor visa. To ap- -ply for a student visa you will need: - - - - - -the enclosed Certificate of Eligi- -bility Form I-20 for F-1 visa (or -Form DS2019 for J-1 visa) - -a passport valid for at least six -months after your scheduled -return to your home country - - proof of funding such as a letter - -from your bank - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -With these items, you must apply to -the appropriate Embassy or Consu- -late General of the United States in your home country. If you are applying -for an F-1 visa, be sure to complete the items on the bottom of pages 1 and 3 -of your I-20 before applying for your visa. - -It is important to understand that as an exchange student on a non- -immigrant F-1 or J-1 visa you are required to be enrolled as a full-time student -for the duration of your participation on the exchange program. Please refer -to the Academics section for more information on full-time status. - -Travel Plans - -As an exchange student, it is your responsibility to arrange for your arrival to -and departure from the United States. Please refer to the Academic Calendar -enclosed with your acceptance packet for important dates. - -You should plan to arrive at Boston College on a weekday (Monday - Friday) -between 9:00 am and 5:00 pm, when university offices are open to assist you. -On-Campus Housing staff may be available on the weekend prior to the start -of the semester. Details will be sent at a later date. - -If you are planning to live off-campus and have not secured housing prior to -departure you should arrive as early as possible (5-7 days) to look for housing -before the beginning of orientation. Please see p. 9 for more information. +Find a route to your destination +Tap Go +, enter your desired start location and destination, and tap a +method of transportation (car, public transit, or on foot) for the detailed +directions to your destination. -5 +21 -www.bc.edu/international - - Also, please be aware that the following immigration regulations are in effect -for all F-1 and J-1 students: - - New international students will not be permitted to enter the U.S. more -than 30 days before the start date of their program as indicated on the I- -20 or DS2019 form. - - Upon completion of studies (i.e. the end date listed on the I-20 or - -DS2019), F-1 students are permitted to remain in the U.S. for 60 days -and J-1 students are permitted to remain in the US for 30 days. - -Additional immigration information is available at the following web site: -http://www.bc.edu/offices/odsd/intercultural/immigration.html. - -Medical Insurance - -The Commonwealth of Massachusetts requires all full-time students (both -American and international) attending colleges and universities in Massachu- -setts to subscribe to a US-based insurance plan. You will be automatically -enrolled in the BC medical insurance upon your admission to the university. -If you do not want this insurance, you must submit proof of insurance with a -provider that has a US office after your arrival. Please see p. 15 for more infor- -mation on BC medical insurance. - -Living Expenses - -Past exchange students -have been surprised by -the cost differences be- -tween their home coun- -try and the Boston area. -When estimating the -costs of your studies -and living in Boston, be -sure to budget for -housing, meals, trans- -portation, books, enter- -tainment, and any trav- -els you plan to take be- -yond the local Boston -area. Specific informa- -tion on costs can be -found on the Estimated -Expenses page in the -back of this handbook. - -www.bc.edu/international +22 -6 + Google Play Store -s -t -t -e -s -u -h -c -a -s -s -a -M - -, -y -b -n -a -r -G - - YOUR FIRST DAYS IN BOSTON - -Getting To Boston College - -Most students come to Boston -by flying to Logan International -Airport. Getting to Boston Col- -lege from Logan is not hard and -can be done in many ways: - -Taxi and Airport Shuttle - -The simplest, yet most expen- -sive, way to get to Boston College -is by taxi. - -Another alternative is to utilize -one of the private airport shuttle -services that operate from Logan -and will drop you off at Boston -College. Prices vary by company. -You might need to make a reser- -vation in advance. For a listing of shuttle services available at Logan, visit -www.massport.com/logan. A listing of private car services can be found on -the internet by Googling “livery services in Boston.” - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Public Transportation - -Finally, you can utilize public transportation to get to Boston College. The -subway is known locally as the “T.” For more information consult the Massa- -chusetts Bay Transportation Authority website at www.mbta.com. - -To take the "T" to Boston College, take the airport shuttle bus to the Airport -subway stop. Take the “Blue Line” subway from the airport to the -“Government Center” subway stop, where you should change trains to the -“Green Line.” Please note that there are four branches of the Green Line! Be -sure to take the “B-Line” train outbound, destination "Boston College," and -get off at the last stop. A public transportation map is included in your accep- -tance package. Obtain a reusable “Charlie Card” at the subway station to re- -duce your T fare. - -For more information on Logan International Airport and traveling into Bos- -ton from the airport, please consult www.massport.com/logan. +Access Google Play -7 +1. From the Home screen, tap Play Store +2. Enter your existing GoogleTM Account login or create one by following -www.bc.edu/international - - Moving Into On-Campus Housing - -Boston College has a number of residence halls. If you will be living on- -campus, we try to place you in residence halls with kitchen facilities. Students -living in rooms without access to kitchen facilities are required to purchase -the Boston College meal plan, which pays for meals in the campus dining -halls throughout the semester. Students living in rooms with kitchen access -have the option of purchasing the meal plan. If you do not wish to purchase -the full meal plan, you should strongly consider the “self-catering” option, -which allows you to purchase "Dining Bucks" in smaller amounts throughout -the semester. For more information on meal plans and Dining Bucks visit the -Dining Services web site at www.bc.edu/offices/bcds. - -Please be aware that sheets, blankets, pillows, towels, etc. are not provided in -the residence halls. These items can be purchased inexpensively in Boston, -so you may not want to take up luggage space with bedding and towels. - -The cost for on-campus housing varies by residence hall and room arrange- -ment. You will be assigned to available housing and will not be able to re- -quest a specific residence hall. Your BC student account will be billed once -your housing assignment is confirmed. You are then responsible for paying -for your housing upon arrival at Boston College. - -Every student on campus has a private phone number and a phone jack in -her/his room. You must provide your own telephone, which can be pur- -chased inexpensively in nearby stores. Local telephone service and voice mail -are free. Campus rates for long distance and overseas calls are very competi- -tive. Long distance calls are billed to your student account. Many students -also use cell phones and Skype for long distance and international calls. - -All on-campus rooms have free internet access, available through either an -Ethernet cable or wireless device. Contact 617/552-4357 with questions. - -Students with on-campus housing can move into their rooms on the first day -of the International Student Orientation for Fall/Full Year students, or on the -day before the Exchange Student Orientation for Spring students (see the -enclosed Academic Calendar). Be sure to arrive between 9:00 am and 5:00 -pm, when the housing office will be open to check you in and give you your -room key. The On-Campus Housing Office phone number is 617-552-3060. - -If you plan to arrive outside of normal business hours, please notify OIP as -soon as you make your travel plans so we can inform the housing office. -Upon arrival at BC, you should go directly to your residence hall and ask for -the RA (Resident Assistant) on call. The RA will let you into your room and -you can pick up your key on the next business day from the Residential Life -office. +. -8 +the steps on the screen. -www.bc.edu/international - - Off-Campus Housing - -If you will be living off-campus, you -will need to make your own housing -arrangements. The Boston College -Off-Campus Housing Office has up- -to-date lists of apartments and -roommate openings near Boston -College and acts as a referral source -for rooms in private homes. Your -name will be given to the Off- -Campus Housing Office and they -will contact you and assist you in -your housing search. The staff can -advise you on suitable arrange- -ments, convenient locations, etc. -However, they cannot make ar- -rangements on your behalf. If you -choose to utilize the above services, -it is possible that you may secure -housing prior to your arrival in the -United States. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Because apartment rentals in the BC area are expensive, many international -students either share an apartment with other students or live with families in -private homes in the surrounding area. We advise all exchange students who -will be seeking off-campus housing to arrive at least 5-7 days before orienta- -tion in order to have enough time to make the necessary arrangements. You -should also make necessary hotel arrangements prior to arrival (see the Ho- -tels and Temporary Housing section for suggestions). - -Finally, be sure to carefully read your lease or other agreements to rent an -apartment before signing and make certain that you understand the terms. If -you are unclear about what a particular document means, consult the Off- -Campus Housing Office for advice. There are laws to protect tenants in Mas- -sachusetts, so be sure to find out what your rights are in case you run into a -conflict with your landlord. - -The Off-Campus Housing Office web site offers helpful hints to assist you in -your search. To access their web site, go to www.bc.edu/housing and click on -the Off-Campus Housing link on the left sidebar. - -Once you have settled into your accommodations for the semester, be sure -to inform OIP of your address and phone number so that we know how to -contact you when necessary. You will receive a local housing notification form -in your orientation packet. +Navigate Google Play -9 +1. Tap Search for apps & games on the top of the screen to search -www.bc.edu/international +for a specific game or app. - Hotels and Temporary Housing +2. To install an app, tap the app and follow the prompts to complete +the process. After installation, you will see a notification in the +Notifications panel. You can open the app from the notification panel +or Apps list. -If you will be arriving in Boston early to find off-campus housing or have ac- -companying guests who need temporary housing, you can obtain lists of tem- -porary accommodation options from the Off-Campus Housing web site at -www.bc.edu/housing (click on the Off-Campus Housing link). +Settings -Checking In +From the Home screen, swipe up to access Apps and tap Settings +to access your phone’s Settings menu. -As an exchange student at Boston College, there are two offices with which -you must check in during normal business hours (Monday to Friday, 9am - -5pm) upon arrival at Boston College: the Office of International Programs -(OIP) and the Office of International Students and Scholars (OISS). +Network & Internet -Upon your arrival, please come to OIP in Hovey House and check in so that -we know that you have arrived safely. +From the Settings screen, tap Network & internet. +• Internet: Enable or disable Wi-Fi, add Wi-Fi network, view saved -You then need to check in with OISS. All international students are required -under the terms of their visas to check in officially at the Office of Interna- -tional Students and Scholars for immigration purposes. You will need your -passport with the white I-94 card attached, as well as your I-20 or DS2019 -form in order to check in. +networks, and monitor data usage. -Orientation +• Wi-Fi Calling: Enable or disable Wi-Fi Calling, adjust Wi-Fi Calling settings. +• Mobile network: Enable or disable mobile data, adjust data roaming +settings, view mobile data usage, set data limit and warning mode, etc. -The Office of International Programs, in collaboration with the Office of Inter- -national Students and Scholars, sponsors an orientation program for all new -international students which includes new exchange students. For fall and full -year students orientation takes place the week before classes begin in Sep- -tember. Spring students have orientation prior to the first week of classes in -January. The orientation will provide you not only with important information -about Boston College, but also with valuable opportunities to meet other ex- -change students, staff members, and Boston College students. +• Airplane mode: Enable or disable Airplane mode. +• Mobile plan: View your mobile plan. +• Others: Adjust settings on Hotspot & tethering, Data saver, VPN, -An important part of this program is the English Placement Test, adminis- -tered to non-native English speakers. All international students, including -exchange students, are required to take this test if their first language is not -English. +Private DNS, etc. -Please see the OISS letter in your acceptance packet for additional informa- -tion about the orientation. +23 -www.bc.edu/international +24 -10 + Connected devices +From the Settings screen, tap Connected devices > Connection +preferences. +• USB: Connect your phone with a computer using a USB cable to +transfer data, share network connectivity, or charge your phone. +• Bluetooth: Pair Bluetooth devices to launch wirelessly hear calls, - ACADEMICS +connect speakers, transfer data, and more. -Full-Time Status +• Cast: Scan for nearby devices to connect to and set them up. +• Printing: Connect your device to a printer on the same Wi-Fi -All international students in the US are required to maintain full-time student -status for the duration of their studies in the US. Undergraduates must be -enrolled in a minimum of 12 credits per semester. Graduate students must -be enrolled in a minimum of 9 credits per semester. Undergraduate exchange -students should expect to take between 12 and 15 credits per semester. -Graduate exchange students should expect to take between 9 and 12 credits -per semester. Failure to maintain full-time status will affect your legal status -in the United States, so it is very important to be enrolled in the appropriate -number of credits. +network. -Selecting and Registering for Courses +• Chromebook: Link your phone to a chromebook. +• Nearby Share: Allow the rapid short-range exchange of images, -We attempt to register you for the classes you requested on your application, -but please keep in mind that you must be flexible about what courses you will -take. We cannot guarantee that you will get all of the classes you request. If -you need to make any alterations to the courses you selected on your applica- -tion, please e-mail OIP at oipexchange@bc.edu. +videos, text, contact info, directions, YouTube videos, and other data +via Bluetooth and Wi-Fi. -When selecting courses, there are a number of factors that must be taken -into account. First, be sure to choose courses that correspond to the level at -which you were admitted to BC. Your acceptance letter states if you were ac- -cepted as an undergraduate or graduate student. Second, be sure to select -courses that correspond to your level of experience with the subject matter. -Each course listing on the BC course web site gives the course level. Levels -for all BC courses are: +• Android Auto: Control your phone’s maps, media and messaging - CORE: Fulfills university core requirements. Fairly basic courses open to +functions, with the Google Assistant while you drive. -undergraduate students from any field of study. +Apps - ONE: Undergraduate elective courses open to students from any field of +View current running applications, manage permissions, and more. +From the Settings screen, tap Apps. +• Recently opened apps: Tap an app to view detailed information. +• Default apps: Set the default app for the browser, phone, messages, -study. +etc. You can also set default apps for opening links. - TWO: Undergraduate courses designed primarily for those specializing in +• Screen time: You can tap -that field of study. +on the right side of most apps to set - THREE: Upper-level courses for undergraduate and graduate registration. -Most suited to those with background knowledge of the general subject -matter and specializing in that field of study. +daily timers. When the app timer ends, the app is paused for the rest +of the day. -11 +• Unused apps: View apps that you have not used for a long time. You -www.bc.edu/international +can un-install them to free up some space. - GRAD: For those admitted as graduate students. Requires advanced knowl- -edge of general subject matter. Please note that some of these courses are -open to undergraduate students with the permission of the professor. +• App battery usage: Set battery usage for apps. +• Special app access: View app access for special functions, such as -Undergraduate students should select courses with a level of Core, One, Two, -or Three. Graduate students should select courses with a level of Grad or -Three. +battery optimization, display over other apps, unrestricted data, usage +access, etc. -Course listings on the BC course web site (www.bc.edu/courses) look similar -to this: +25 -Index +26 -Course # + Notifications -Credits +Battery -Level +From the Settings screen, tap Notifications. +• App notifications: Control notifications from individual apps. +• Notification history: View your recent and snoozed notifications. +• Conversations: View priority and modified conversations. +• Bubbles: Enable message conversations display as bubbles, floating -Title +on top of other apps. -Instructor +• Device & app notification: Control which apps and devices can -Bldg +read notifications. -Room +• Notifications on lock screen: Select how you want notifications to -Time +be displayed on the lock screen. -9398 +• Do Not Disturb: Enable or disable Do Not Disturb. Set Do Not -BI200 01 +Disturb behavior, exceptions, schedule, etc. -0569 +• Others: You can adjust other notifications settings, like wireless -HS181 03 +emergency alerts, hide silent notifications in status bar, etc. -3 +Monitor your phone’s battery usage. +From the Settings screen, tap Battery. +• Battery Usage: View the battery percentage that each apps used -3 +from last full charge. -TWO +• Battery Saver: Enable or disable Battery Saver, set a schedule for +turning on the Battery Saver, turn the Battery Saver off when your +phone charges above 90%. -Biology +• Battery percentage: Show battery percentage in the Status bar. -DEPT +Storage -ONE +Manage internal storage on your phone and provides menu choices for +an optional SD card. +From the Settings screen, tap Storage. +• Device storage: From Storage screen, tap THIS DEVICE > This -American -Civilization +device. Tap a category to view the storage amount used by each app. -Singleton +27 -MWF -11 -TTH -9* - -In these examples, the index is the number used to register for the course -(you must include this number on your Course Registration Form). The -course number includes a two-letter abbreviation for the department (e.g. BI -= Biology, HS = History), followed by the three-digit course number. The two -numbers following the course number indicate the section. Popular courses -such as HS181 are offered in different sections to keep class sizes small. In -the example above, the course selected is section 3 of HS181. Each of the -courses in these examples carries a 3-credit load, which is the most common -load for university courses. It is your responsibility to make sure that you -have had all prerequisites required for any course that you are requesting. - -Courses typically meet several times a week. Classes that meet Monday, -Wednesday and Friday (MWF) are generally 50 minutes long, while Tuesday, -Thursday (TTH) classes are typically 75 minutes. These classes are indicated -in the course listing with an asterisk (*). - -Some courses may require co-registration for a supplementary lab or discus- -sion section and this will be stated in the course listing. Enrollment in some -courses may be restricted to majors only or may require department permis- -sion prior to registration. We are not able to pre-register you for restricted -courses, but you will be able to appeal to the department or professor after -you arrive to ask for special permission to enroll in such a class. Likewise, -undergraduates who seek to enroll in graduate-level courses will need to seek -direct permission from the instructor after arrival in Boston. It is fairly un- -usual for undergraduates to obtain approval for graduate courses and you -should not count on being able to do so. - -www.bc.edu/international +28 -12 + • SD card: From Storage screen, tap THIS DEVICE > SD CARD to - Finally, please note that many courses “close” during the early registration -period. Each course is listed in the computer with a maximum number of stu- -dents who can enroll. Once that limit is reached, the computer will not allow -any additional registration in that course. If you find that a course you need is -closed, you will have to appeal to the professor (either by e-mail prior to your -arrival or after you arrive) for an “override” to allow you into the class. This is -not uncommon and most professors are sympathetic to reasonable requests -from exchange students. However, it is not guaranteed that you will receive -an override for a closed course. - -Please look for classes offered in the semester in which you will be attending -(e.g., Fall, Spring) even if the list is for the previous year. For the most current -listing of available courses, see www.bc.edu/courses. - -Drop/Add - -The first week of classes at the beginning of each semester is called the -“Drop/Add” period. At this time you can make changes to your course sched- -ule, adding and/or dropping courses as needed. Don’t worry if you do not -have all of the courses that you need by the time classes start. You will have -this period of time to make adjustments to your course schedule, seek over- -rides as needed, etc. - -Your OIP exchange advisor can help you with most registration needs, includ- -ing add/drop. Registration matters can also be handled at Student Services in -Lyons Hall or on the computer via AGORA. You will receive more specific in- -formation about these resources upon arrival. In the meantime, refer all -course requests to OIP. - -Faculty Advisors - -You may request a faculty advisor for -assistance with academic planning for -your semester or year at BC. We will -make advisor placements based upon -your field of study. If your advisor at -home already knows a particular pro- -fessor at BC, or if you have already -made contact with a professor here -and you would like him/her to be your -advisor, please let us know. If you re- -quest a faculty advisor, you will be -required to meet with them by ap- -pointment during the first week of -classes and throughout the semester -as necessary. +Display -13 +view files stored on your SD card. Tap on the top right corner of the +screen to rename, eject or format the SD card. -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -www.bc.edu/international - - Tutoring - -The Connors Family Learning Center offers free tutoring for more than 60 -undergraduate and some graduate courses, including writing and English -as a Second Language. Intensive, small-group supplemental instruction -sessions are available in several courses, and tutoring and assistance in -improving academic writing. For more information about services provided -by the Learning Center, go to www.bc.edu/libraries/centers/connors/. - -English Classes for Non-Native Speakers - -If your native language is not English you will be required to take a place- -ment examination during the Orientation in the fall or during the first week -of classes in January. If necessary, students will be recommended for a par- -ticular course based upon the results of this examination. The Program in -English for Foreign Students is based in the English Department and offers -several courses in college-level English for non-native speakers. Recent -course offerings include: - - EN 118 Essential of English Composition: extensive practice in writing - -and editing academic prose - - EN119 The Craft of Writing: practice writing academic essays and expo- -sure to English rhetoric, with emphasis on written analysis and the logi- -cal support of ideas - -In order to enroll in one of these courses, you must have taken the place- -ment test. - -Transcripts of Academic Work - -At the end of your time at Boston College, we will send an official transcript -of the academic work you completed to your home university. Before we -can release your BC transcript to your home university, however, you must -clear any outstanding debts to Boston College that you may have. Be sure -that all long-distance phone calls, on-campus housing fees, insurance, -meals, etc. have been paid for. - -Your BC student account must show a zero balance before you leave Bos- -ton or we will not be able to release your transcript. - -www.bc.edu/international +• Storage manager: Storage manager removes backed up photos and -14 +videos from your device. - ADMINISTRATIVE ISSUES - -Student Identification Cards - -If you submitted the required passport-size photographs with your Exchange -Student Application you will receive your student ID card during orientation. -If you did not send us photos, you will be able to have your ID card made at -Student Services in Lyons Hall following orientation. - -Your Boston College student identification card (“Eagle ID”) is a very impor- -tant item for you to have. In addition to identifying you as a BC student, the -card enables you to check books out from the library, to make purchases at -the BC bookstore, to access your student record and account from “U-VIEW” -computers around campus, and to make use of the BC Flynn Recreation -Complex (fee required), among other services. Be sure to keep your Eagle ID -with you at all times and do not lose it. A replacement fee will be charged by -Student Services for lost cards. - -Medical Insurance - -As explained in the Things to Know Before You Leave Home section, the -Commonwealth of Massachusetts requires all students enrolled in a Massa- -chusetts college or university to be covered by a US-based medical insurance -plan. Therefore, you have been automatically enrolled in the Boston College -medical insurance policy which provides adequate coverage during your stay -at BC, including travel within and outside the US during your period of cover- -age. Your student account was billed upon registration at BC. As an exchange -student it is your responsibility to pay for your insurance, which can be done -at Student Services. Full year students will be automatically re-enrolled in the -insurance plan for their second semester. You can find more information on -the BC medical insurance coverage on the BC web site at www.bc.edu/ -offices/uhs/services/insurance. - -If you do not want to be covered by BC medical insurance, you must be cov- -ered by insurance that is US-based or has a US contact office. You must have -proof of this insurance. If your insurance is comparable, you will be allowed -to waive the BC insurance and your student account will be credited accord- -ingly. Please note that it is not possible to have your medical insurance evalu- -ated before you arrive at Boston College. +Sound & vibration +From the Settings screen, tap Sound & vibration. +• Media/Call/Ring/Notification/Alarm volume: Drag the slides to -15 +adjust the media, call, ring & notification, and alarm volumes. -www.bc.edu/international - - Health Services - -The Health Services Center at -Boston College offers a range -of services to BC students. -Staffed by first-class doctors, -nurse practitioners, and other -medical staff, the Health Ser- -vices Center is equipped to -handle health emergencies, -minor illnesses and questions -about healthy living. While the -general health services offered -are comprehensive, if you re- -quire specialty care you can -be referred to a qualified spe- -cialist. - -All undergraduate exchange -students who live on-campus -are required to pay the Health -Services fee. Exchange stu- -dents who live off-campus are -also entitled to this service for the same fee. Off-campus students can ei- -ther pay in advance, or the first time the service is used. We strongly recom- -mend that you utilize the Health Services if necessary. Services covered by -the fee are detailed on the Boston College Health Services website at -www.bc.edu/offices/uhs. - -Please note that Health Services fee is not the same as medical insurance. -You must still have adequate medical insurance coverage, especially for -such things as hospitalization, emergency room visits, diagnostic testing, -x-rays, etc., which are not covered by the Health Services fee. - -University Counseling Services - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sometimes students find adjusting to a new culture and new academic en- -vironment very challenging. The Counseling Services Office provides caring, -professional support and assistance to full-time students who request it. -Selected staff members in Counseling Services have extensive experience -with cross-cultural issues and are especially well-suited to help students -who are experiencing culture shock. It is very common in the United States -for students to seek the help of counseling professionals when dealing with -challenging situations. In fact, thousands of students take advantage of the -services offered by Counseling Services for a variety of issues each year. +• Do Not Disturb: Manage Do Not Disturb settings. +• Phone ringtone: Select a ringtone for incoming phone calls. +• Media: Enable Media to hide player when the media session has ended. +• Others: You can adjust other sound and vibration settings, like -16 +vibration & haptics, default notification and alarm sounds, enable +tones or vibration on other scenarios, etc. -www.bc.edu/international - - The services offered by Counseling Services (at no additional cost) include -short-term individual counseling and psychotherapy as well as a limited -number of counseling groups. If you wish to speak with a counselor you -can make an appointment by calling 617-552-3310 or by visiting the Counsel- -ing Office in Gasson Hall 108. - -Also, please remember that OIP staff is always available and willing to help -if you have any questions, either academic or personal. Most of our staff -members have studied and/or lived outside the US and know firsthand the -challenges of adapting to a new country and culture. - -Students with Disabilities - -Boston College provides a range of services and accommodations for stu- -dents with physical or learning disabilities. If you require additional assis- -tance because of a disability, feel free to share that information with your -OIP exchange advisor. You may also contact the Boston College Coordina- -tor of Special Needs Services in Gasson Hall 108, telephone: 617-552-3310. - -E-mail - -Every student enrolled at -Boston College is eligible -to receive an e-mail ac- -count. You may obtain -your account after arriv- -ing at BC by presenting -your BC ID card at the -Campus Technology Re- -source Center (CTRC) -located on the second -floor of the O’Neill Li- -brary. The use of your BC -email account is required -once you have arrived on- -campus. If you have an- -other address that you -prefer to use, set up your BC account (by using AGORA) to forward all mes- -sages to the alternate account. OIP and other BC offices will send all infor- -mation during your time at Boston College to your BC email address. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +From the Settings screen, tap Display. +• Brightness level: Drag the slide to adjust the screen brightness -17 +level. -www.bc.edu/international - - LIFE BEYOND ACADEMICS - -Religion - -As a Catholic and Jesuit institution, Boston College is sensitive to the spiri- -tual needs of all members of the BC community, whatever their respective -faiths or philosophical traditions. Catholic Masses are offered throughout -the week in several locations on campus and are open to all BC students, -faculty and staff. The Campus Ministry Office offers spiritual counseling -and guidance to students, as well as volunteer service opportunities in Bos- -ton and other parts of the world. Campus Ministry also provides informa- -tion and support for people of all religious and spiritual traditions by main- -taining a list of churches, temples and synagogues in the Boston College -area. Various student clubs, including the Asian Christian Fellowship, Jew- -ish Students’ Association, and Muslim Students’ Association, provide a -sense of community and support for students of various spiritual tradi- -tions. - -Activities/Opportunities - -There is a wide variety of clubs and -activities to meet any student’s in- -terests, including several intercul- -tural clubs focusing on individual -countries or regions. An Activities -Fair early in the Fall Semester will -give you an opportunity to learn -about many of these opportunities -and speak to some of the students -involved. For students on exchange -during the Spring semester, the Of- -fice of the Dean for Student Devel- -opment has a list of all BC clubs -and activities (www.bc.edu/a-z/ -clubs/). Don’t pass up this opportu- -nity to learn about American culture -outside the classroom. Students -who get involved in activities at Boston College consistently enjoy their ex- -perience here more than those who do not get involved. For this reason, -we strongly encourage you to join at least one club while you are here. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +• Adaptive brightness: Enable adaptive brightness to automatically -18 +adjust the screen brightness basing on the lighting conditions. -www.bc.edu/international - - s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sports in American University Culture - -Many international students are surprised at the level of importance placed -on intercollegiate sports in the United States. American football, particu- -larly, draws thousands of screaming fans every Saturday during football -season, with students cheering for the home team. This is very much a part -of American college culture and, even if you are not a sports fan, it is some- -thing you should experience during your time as a student in the United -States. You may not know anything about the intricate rules of American -sports like football, hockey and basketball. However, the experience of sit- -ting in a crowded stadium, cheering, listening to the band, watching the -crowd, and even trying to follow the game, is a true American experience -and should not be missed. - -Recreation Complex - -The Flynn Recreation Complex provides Boston College students with a -broad array of sports and physical fitness opportunities, from aerobics to -weight lifting and swimming. As an exchange student you have the option -of joining the “RecPlex” for a semester or academic year (membership fees -apply). If you choose not to join the “RecPlex” you may still use the facilities -by presenting your BC ID card and paying a small fee per visit. +• Lock screen: Set what to show on lock screen. +• Screen timeout: Your screen will automatically go to sleep after the -19 +set amount of time. -www.bc.edu/international +• Dark theme: Turn on dark theme. +• Display size and text: Adjust font size and display size. +• Night Light: Set a Night Light schedule and intensity. +• Others: You can adjust other display settings, like enable auto-rotate +screen and life to wake features, adjust display color and screen saver, +etc. - CONTACT INFORMATION +29 -Office of International Programs +30 -Location: -Mailing Address: + Accessibility -Hovey House, 258 Hammond Street -Hovey House, 140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +Security & privacy -Phone: 617/552-3827 Fax: 617/552-0647 -Email: oipexchange@bc.edu -Web: www.bc.edu/international -Director: Bernd Widdig -Email: widdig@bc.edu -Assistant Director for Exchange Students: Esther Messing -Email: messines@bc.edu +Manage accessibility-related apps to make the device easier to use for +those with certain physical disabilities. +From the Settings screen, tap Accessibility. +• Real-Time text call: Manage RTT call options. +• Display: Adjust the font size and display size, enable or disable dark -Office of International Students and Scholars +theme, enable screen magnification, etc. -Location: -Mailing Address: +• Interaction controls: Enable or disable useful interaction functions, -21 Campanella Way - Suite 249 -140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +such as timing control, system controls, etc. -Phone: 617/552-8005 Fax: 617/552-2190 -Web: www.bc.edu/offices/odsd/intercultural -Director: Adrienne Nussbaum -Email: adrienne.nussbaum@bc.edu +• Captions: Enable or disable captions. You can also set caption display -Office of Residential Life +language, text size, and caption style. -Contact: -Web: +• Audio: Pair your hearing aids, enable mono audio, or adjust the audio -On-Campus Housing, Off-Campus Housing Advice -www.bc.edu/offices/reslife +balance from the speaker. -Office of Student Services +• Accessibility shortcuts: Mange the settings on accessibility button. +• Text-to-speech output: Set the text-to-speech engine, language, -Contact: -Web: +speech rate, etc. -Billing, Medical Insurance, ID, Course Registration -www.bc.edu/offices/stserv +From the Settings screen, tap Security & privacy. +• App security: Check your device and prevent/warn you of potential -University Health Services +harm. -Contact: -Web: +• Device lock: Set your screen lock method, pattern, PIN, password, etc. +• Account security: Add a Google Account to secure your device. +• Device finders: Enable Find My Device to locate your device -Immunization, Health and Wellness Issues -www.bc.edu/offices/uhs +remotely, keeping your data safe in the event that your phone is lost. +• System & updates: Check for new software updates for your device. +• Privacy: Manage a batch of privacy settings. +• Face unlock: Unlock your phone by looking at it. +• More security & privacy: Adjust more security and privacy settings. -University Counseling Services +Unlock your device automatically in certain conditions. -Contact: -Web: +31 -Personal Counseling -www.bc.edu/offices/counseling +32 -20 + Location + +From the Settings screen, tap Location. +• Recent access: View a list of apps that have recently requested your + +location. + +• App location permissions: View which apps have permission to + +access location services. + +• Location services: Manage settings on other location services, such + +as emergency location service, Google location accuracy, etc. + +Safety & emergency + +Fulfill your medical and emergency information, view emergency alerts, +and manage settings on emergency SOS and location service. +From the Settings screen, tap Safety & emergency. +• Emergency information: Add medical information and emergency +contacts to help first responders in an emergency. Anyone can read +this information from your lock screen and dial your emergency +contacts without unlocking your phone.. + +• Emergency SOS: If you have turned the Emergency SOS function +on, when you have an emergency and need emergency services, +press the Power/Lock key 5 or more times quickly. + +• Emergency Location Service: If you have turned Emergency +Location Services (ELS) on, when you call or text an emergency +number, your device will use ELS to automatically send your location +to emergency responders. + +• Earthquake alerts: Get an alert with the initial estimated magnitude + +and distance from your location. + +• Wireless emergency alerts: Turn wireless emergency alerts on + +or off. You can also view emergency alert history and manage alert +preferences. + +Passwords & accounts + +Set up and manage all of your accounts, including your Google Account +and other social networking accounts, through the Accounts settings +menu. From the Settings screen, tap Passwords & accounts. + +33 + +34 + + • Passwords: See, change, or remove passwords you saved in your + +• Reduce interruptions: Enable or disable the notifications of apps + +Google Account. + +and Do Not Disturb mode. + +• Autofill service: Autofill your information when needed from your + +Google Account. + +• Add account: Set up your account. +• Automatically sync app data: Activate automatic data syncing. + +Digital Wellbeing & parental controls + +Statistics screen usage time, app timers, bedtime schedules, manage +notifications, etc. +From the Settings screen, tap Digital Wellbeing & parental controls. +• Your Digital Wellbeing tools: Use app timers and other tools to + +keep track of screen time and unplug more easily. + +• Ways to disconnect: View your screen time and set timers for your +apps, set a bedtime schedule to disconnect from your phone and +avoid interruptions while you sleep, pause distracting apps when you +need time to focus. + +• Show icon in the app list: Enable icons in the app list. +• Parental controls: Add content restrictions and set other limits to + +help your child balance their screen time. + +Google + +Configure your device’s Google settings. Available options depend on +your Google Account. From the Settings screen, tap Google. + +System + +Manage detailed settings for your phone, including language and input +settings, battery and storage, date and time, etc. +From the Settings screen, tap System. +• Languages: Set the language for the phone’s menus and keyboards, +set the language for each app, set units and number preferences, etc. + +• Keyboard: Select and configure keyboard settings + +35 + +36 + + • Gestures: Enable useful gestures, such as Quickly open camera, +Navigation mode, Power & hold power menu and Prevent ringing. +• Date & time: Enable or disable network-provided time, automatic + +time zone, 24-hour format, etc. + +• Backup: Back up your phone’s data and settings to Google servers + +before resetting your phone to the original factory settings. + +• System update: Check for new software updates for your device. +• Multiple users: Share this device with other people by creating + +additional users. + +• Reset options: View different reset options and reset your device. + +About phone + +Access important phone information, search for and apply updates, and +view legal and safety information. +From the Settings screen, tap About phone. + +Troubleshooting + +Below is a list of common problems that may occur when using your +mobile phone. Try these easy troubleshooting steps to resolve any type +of issues you may encounter. +The phone does not power on +• Press and hold the power key until the startup sequence appears on + +the display. + +• Check the battery level and ensure there is sufficient charge. +The phone does not respond after several minutes +• Remove and insert the battery – If the phone has an internal battery, + +please press and hold the power button. + +The phone powers off by itself +• Check if the screen lock of the phone is enabled. +• Check if the phone needs to be charged. + +37 + +38 + + The battery will not charge +• Check if the battery is already fully charged. +• Make sure that the battery is inserted correctly. +• Ensure that the battery being used is a BLU approved battery. +• Double check if the charger connector is properly connected to the + +phone. + +Note: Connect charger only after the battery has been inserted. + +The phone displays a SIM Error +• Make sure the SIM is properly inserted and in its intended slot. +• Make sure that the SIM card is not damaged or scratched. +The PIN code was entered incorrectly three times +• The SIM card will lock, and you will need to enter the PUK code (PIN +Unlock Code). Please contact your local service provider to obtain the +PUK. + +The standby time is too short +• Please keep in mind that standby time relates directly to phone usage + +I cannot make calls +• Make sure that the number you dialed is correct and the dial key has + +(Internet, Text, Voice Calls, Maps, etc.). + +been pressed. + +• Batteries typically have to be replaced after a year of use. +The phone cannot connect to the network (Voice or Data) +• Verify if you have active service with your service provider. +• Contact your local network service provider and verify if you are within + +its network coverage area. + +• Use the full national area code when dialing out of your area or when + +abroad, use country-specific international access codes. + +I cannot receive calls +• Make sure that your phone is connected to a network. + +39 + +40 + + The caller ID does not display on incoming calls +• Contact your local network service provider and verify you have the +caller ID feature enabled on your account or verify if the caller may +have blocked his number from appearing. + +The caller cannot leave a voicemail message +• Contact your local network service provider and verify if the voicemail + +service is active on your account or check if the network is busy. -www.bc.edu/international +The phone does not connect to the internet +• Ensure with your service provider that your account has the - ESTIMATED EXPENSES -2010-2011 +appropriate internet/data plan enabled. - Average Cost of Apartment/Room Rental +• Ensure internet APN settings are configured (Carrier dependent). +Text Messages cannot be Sent / Received +• Ensure text messaging is enabled with your service provider. +• Ensure APN settings are configured (Carrier dependent). -Studio Apartment: $1000 - $1300 per month * -One Bedroom Apartment: $1200 - $1550 per month * -Two Bedroom Apartment: $1400 - $2000 per month * -Three-Four Bedroom Apartment: $2000 - $3800 per month * -Private Room in Local Home: $500 - $900 per month * +Regulatory Information -*Please note: Boston rental agents often charge one half or one full month's -rent as a commission fee (also known as a "Finder's Fee") +FCC regulations -On-Campus Housing +This mobile phone complies with part 15 of the FCC Rules. Operation is +subject to the following two conditions: (1) This device may not cause +harmful interference, and (2) this device must accept any interference +received, including interference that may cause undesired operation. +This mobile phone has been tested and found to comply with the limits +for a Class B digital device, pursuant to Part 15 of the FCC Rules. These +limits are designed to provide reasonable protection against harmful +interference in a residential installation. +This equipment generates, uses and can radiate radio frequency energy +and, if not installed and used in accordance with the instructions, may +cause harmful interference to radio communications. + +41 + +42 + + However, there is no guarantee that interference will not occur in a +particular installation If this equipment does cause harmful interference +to radio or television reception, which can be determined by turning +the equipment off and on, the user is encouraged to try to correct the +interference by one or more of the following measures: +• Reorient or relocate the receiving antenna. +• Increase the separation between the equipment and receiver. +• Connect the equipment into an outlet on a circuit different from that + +to which the receiver is connected. + +• Consult the dealer or an experienced radio/TV technician for help. +Changes or modifications not expressly approved by the party +responsible for compliance could void the user’s authority to operate the +equipment. +The antenna(s) used for this transmitter must not be co-located or +operating in conjunction with any other antenna or transmitter. + +RF Exposure Information (SAR) + +This mobile phone meets the government’s requirements for exposure to +radio waves. +This phone is designed and manufactured not to exceed the emission +limits for exposure to radio frequency (RF) energy set by the Federal +Communications Commission of the U.S. Government. +The exposure standard for wireless mobile phones employs a unit of +measurement known as the Specific Absorption Rate, or SAR. The SAR +limit set by the FCC is 1.6 W/kg. *Tests for SAR are conducted using +standard operating positions accepted by the FCC with the phone +transmitting at its highest certified power level in all tested frequency +bands. +Although the SAR is determined at the highest certified power level, +the actual SAR level of the phone while operating can be well below +the maximum value. This is because the phone is designed to operate +at multiple power levels so as to use only the poser required to reach +the network. In general, the closer you are to a wireless base station +antenna, the lower the power output. + +43 + +44 + + The highest SAR value for the model phone as reported to the FCC when +tested for use at the ear is 1.29 W/kg and when worn on the body, as +described in this user guide, is 0.89 W/kg (Body-worn measurements +differ among phone models, depending upon available accessories and +FCC requirements.). +While there may be differences between the SAR levels of various +phones and at various positions, they all meet the government +requirement. +The FCC has granted an Equipment Authorization for this model phone +with all reported SAR levels evaluated as in compliance with the FCC RF +exposure guidelines. SAR information on this model phone is on file with +the FCC and can be found under the Display Grant section of www.fcc. +gov/oet/ea/fccid after searching on FCC ID: YHLBLUB160V. +For body worn operation, this phone has been tested and meets the FCC +RF exposure guidelines for use with an accessory that contains no metal +and the positions the handset with a minimum of 1.5 cm from the body. + +Use of other accessories may not ensure compliance with FCC RF +exposure guidelines. If you do not use a body-worn accessory and are +not holding the phone at the ear, position the handset a minimum of 1.5 +cm from your body when the phone is switched on. + +Safety precautions for power supply unit + +Use the correct external power source +A product should be operated only from the type of power source +indicated on the electrical ratings label. If you are not sure of the type +of power source required, consult your authorized service provider or +local power company. For a product that operates from battery power or +other sources, refer to the operating instructions that are included with +the product. +This product should be operated only with the following designated +power supply unit(s). +Travel charger: Input: 100-240V, 50/60Hz, 0.35A. Output: 5V, 2000mA. + +45 + +46 + + Handle battery packs carefully +This product contains a Lithium-ion battery. There is a risk of fire and +burns if the battery pack is handled improperly. Do not attempt to open +or service the battery pack. Do not disassemble, crush, puncture, short +circuit the external contacts or circuits, dispose of in fire or water, or +expose a battery pack to temperatures higher than 140°F (60°C). The +operating temperature for the phone is 14°F (-10°C) to 113°F (45°C). +Warning: Battery is non-removable. Do not attempt to remove +the battery for danger of explosion. +To reduce risk of fire or burns, do not disassemble, crush, puncture, +short circuit the external contacts, expose to temperature above 113°F +(45°C), or dispose of in fire or water. Do not attempt to replace the +non-removable battery. The phone and battery as a unibody device +must be disposed of in accordance with locally applicable environmental +regulations or guides supplied with your product. + +47 + +Take extra precautions +• Do not disassemble or open crush, bend or deform, puncture or + +shred. + +• Do not modify or remanufacture, attempt to insert foreign objects into +the battery, immerse or expose to water or other liquids, expose to +fire, explosion, or other hazards. + +• Only use the battery with a charging system that has been qualified + +with the system per CTIA Certification Requirement for Battery System +Compliance to IEEE1725. Use of an unqualified battery or charger may +present a risk of fire, explosion, leakage, or other hazards. + +• Replace the battery only with another battery that has been qualified + +with the system per this standard, IEEE-Std-1725. Use of an +unqualified battery may present a risk of fire, explosion, leakage or +other hazards. Only authorized service providers shall replace battery. +(If the battery is non-user replaceable). + +• Promptly dispose of used batteries in accordance with local + +regulations. + +• Improper battery use may result in a fire, explosion, or other hazards. +48 + + • Do not replace an embedded battery pack. Improper replacement may +present a risk of fire, explosion, leakage or other hazards. Contact the +manufacturer for replacement instructions. + +• The phone identifies the battery by the Battery ID Pin. +• If the battery leaks: + +• Do not allow the leaking fluid to come in contact with skin or + +clothing. If already in contact, flush the affected area immediately +with clean water and seek medical advice. + +• Do not allow the leaking fluid to come in contact with eyes. +If already in contact, DO NOT rub; rinse with clean water +immediately and seek medical advice. + +• Take extra precautions to keep a leaking battery away from fire as + +there is a danger of ignition or explosion. + +Hearing Aid Compatibility (HAC) + +This device is hearing aid-compatible as determined by ANSI C63.19- +2019 and waiver DA-23-914. The lowest conversational gain is 10.32 +dB with a hearing-aid, and 13.93 dB without hearing-aid, using telecoil +coupling capability and Bluetooth connectivity. +The 2019 ANSI C63.19 standard does not use this rating system nor +do any test standards exist for any wireless technologies that transmit +above 6 GHz. Instead, this standard uses the conversational gains for +the device’s hearing aid compatible capabilities. +This device has been tested and certified for use with hearing aids for +some of the wireless technologies that it uses. However, there may be +some newer wireless technologies used in this phone that have not been +tested yet for use with hearing aids. It is important to try the different +features of this phone thoroughly and in different locations, using your +hearing aid or cochlear implant, to determine if you hear any interfering +noise. Consult your service provider or the manufacturer of this phone +for information on hearing aid compatibility. If you have questions about +return or exchange policies, consult your service provider or retailer. -Shared Room: $1100 per person per month +49 -Meals and Books +50 -Campus Meal Plan: $2300 per semester for full meal plan -No Meal Plan: $1500 per semester -Books: $600 per semester + We consent to the following conditions of the waiver: +• Pass Conversational Gain, Receive Distortion and Noise and Receive -Other Expenses +Acoustic Frequency Response testing using one narrowband and one +wideband codec, limited to one bit rate. -Medical Insurance: $871 (Fall semester); $1027 (Spring semester) ** -University Health Services: $210 per semester ** -Flynn Recreation Complex: $140 per semester +• Pass Conversational Gain at both 2N and 8N on all available -**The BC Medical Insurance plan is mandatory unless you can demonstrate -coverage by an insurer with a U.S. office. The University Health Services fee is -required for on-campus students. +narrowband and wideband codecs and air interface combinations. +Please see the HAC Volume Control test report for further details and +supporting data. -Personal Expenses: These will vary significantly according to the individ- -ual, but students should budget approximately $2000 per semester. +The information provided in this user manual is subject to change without notice +and BLU Products reserves the right to modify the manual at any given time. BLU +Products has made every effort to ensure that this user manual is accurate and +disclaims responsibility and liability for any inaccuracies or omissions. BLU Products +makes no commitment to keep current the user manuals, however, does reserve +the right to make improvements to the user manual at any given time without +notice. -Transportation from Logan Airport to BC +Google, Google Play, Google Maps, YouTube, Gmail and other marks are +trademarks of Google LLC. -Taxi: approximately $60 -Private Airport Shuttle: approximately $50 -Subway (“T”) Ride : $2.00 cash payment or $1.70 using a “Charlie Card” +51 diff --git a/q121/random_k2/random_2.md b/q121/random_k2/random_2.md index 4b9bd14ed63d1882c240fd2561e1be123d06b2a6..2cd064b36cf6fc7716b108dfdc7b4efff29d6268 100644 --- a/q121/random_k2/random_2.md +++ b/q121/random_k2/random_2.md @@ -1,1483 +1,629 @@ -USER MANUAL +N0 5909 -8-IN-1 ONE-TOUCH -AUTOMATIC ESPRESSO MAKER +HEADPHONE OWNER’S MANUAL -GCM4230 + TABLE OF CONTENTS -1 - -USER MANUAL ESPRESSO MAKER GCM4230 - CONTENTS - -Important Safeguards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 - -Know Your Espresso Maker . . . . . . . . . . . . . . . . . . . . . . . . . . 6 -Control Panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 - -Using Your Espresso Maker - -Before First Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 - -Setting Up and Priming the Espresso Maker . . . . . . . . . . . . . . . 8 - -Filling the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 - -Placing the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 - -Mug Size Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 - -Drink Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +TABLE OF CONTENTS -Making an Espresso . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 +Overview +What's in the Box +Product Overview -Making a Cappuccino . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 +Adjustability, Fit & Comfort -Making a Latte . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 +Functions -Making an Americano . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 +Charge -Brewing in Manual Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 15 +Connections +Pairing +Passive +USB Digital Direct -Frothing Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 +APP -Cleaning and Maintenance +Maintenance / Care -Cleaning the Portafilter and Brew Head . . . . . . . . . . . . . . . . . 17 +Changing / Removing Earpads -Cleaning the Water Reservoir . . . . . . . . . . . . . . . . . . . . . . . 17 +Accessories -Cleaning the Milk Reservoir and Frothing Tube . . . . . . . . . . . . . 17 +Troubleshooting -Descaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 +Ear Health -Warranty and Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 +Specifications -Model: GCM4230 | © 2023 Gourmia | 2.0 -The Steelstone Group Brooklyn, NY | Gourmia .com - -All rights reserved. No portion of this manual may be reproduced by any means whatsoever without written -permission from The Steelstone Group, except for the inclusion of quotations in a media review. Although -every precaution has been taken in the preparation of this manual, The Steelstone Group assumes no -responsibility for errors or omissions. Nor is any liability assumed for damages resulting from the use of -the information contained herein. Although every effort has been made to ensure that the information -contained in this manual is complete and accurate, The Steelstone Group is not engaged in rendering -professional advice or services to the individual consumer; further, The Steelstone Group shall not be held -responsible for any loss or damage allegedly arising from any information or suggestions in this manual. - - Congratulations on the -purchase of your Espresso -Maker from Gourmia! - -Here at Gourmia, our one and only goal is to make sure -nutritious and delicious meals are always a practical -possibility no matter how fast-paced your life is . Whether -you’re in need of a hearty breakfast, a mid-day snack, or a -satisfying supper, our dependable, easy-to-use appliances -are designed to work for you while you work your way -through your busy day . - -This manual provides an overview and instructions on -all the espresso maker’s functions and features, along -with directions for assembling, operating, cleaning, -and maintaining your appliance . Please read all safety -instructions before getting started . Keep this manual handy -for future use and review . - -Our team is available to help with any and every question -you might have about your new espresso maker . We’re -excited for you and want you to get the most out of your -purchase—so whether you need assistance with setup or -guidance on which features to enable for which drinks, -we’re happy to walk you through it, step by step . - -Happy and easy cooking! - -We’d love to hear about the delicious drinks you’re creating! -Share your thoughts, recipes, and photos with us. - -@gourmia +Warranty 3 +3 +4 - S -d -r -a -u -G -E -f -a -S - -t -n -a -t -r -o -p -M - -I - -S A V E T H E S E I N S T R U C T I O N S - -IMPORTANT SAFEGUARDS - -Before operating this espresso maker, read and follow all these -basic safety precautions: - -1. The appliance can be used - -7. do not operate the appliance - -by children aged 8 years and -above and persons with reduced -physical, sensory or mental -capabilities or lack of experience -and knowledge if they have been -given supervision or instruction -concerning use of the appliance -in a safe way and understand the -hazards involved . Children should -not play with the appliance . - -Cleaning and user maintenance -shall not be done by children -unless they are 8 or older and -supervised . - -2. Keep the appliance and its cord - -out of reach of children less than 8 -years old . - -3. Make sure your outlet voltage - -corresponds to the voltage stated -on the rating label of the espresso -maker . - -4. To protect against fire, electric - -shock, or injury to persons do not -immerse cord, plug, or appliance -in water or other liquid . - -5. Close supervision is necessary - -when this appliance is used by or -near children . - -6. Unplug from outlet when not in +5 -use and before cleaning . Allow the -appliance to cool before putting -on or taking off parts, and before -cleaning the appliance . +6 -with a damaged cord or plug or if -the appliance malfunctions or has -been damaged in any manner . -Return appliance to an authorized -service facility for examination, -repair, or adjustment . +8 -8. The use of accessories not +9 +9 +9 +10 -recommended by the appliance -manufacturer may result in fire, -electric shock, or injury to persons . +11 -9. do not touch hot surfaces; use +12 -handles or knobs . +12 -10. do not operate this appliance -outdoors . do not use for -commercial purposes . +13 -11. do not let the cord hang over +14 -the edge of a table or counter or -touch any hot surfaces . +15 -12. Use on a hard, flat, level, heat- +16 -resistant surface only . +17 -13. Always be sure to secure the +1 -water reservoir tightly back inside -the unit with the lid closed and -filled with water between the -minimum and maximum mark -before turning the appliance on . +N05909 HEADPHONE / OWNER’S MANUAL 2 -14. do not remove the portafilter while -brewing coffee . Exercise caution -when removing the portafilter after -brewing since the metal parts -will be hot . Make sure to hold the -portafilter by the handle and use - the filter retention clip to dispose -the grounds . +N05909 HEAPHONE / OWNER’S MANUAL -4 + OVERVIEW -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - 15. Extreme caution must be -exercised when moving -appliances with hot liquids . +Congratulations on your selection of the Mark Levinson® +No 5909 Premium Wireless ANC Headphone. For over 50 +years, discerning individuals have chosen Mark Levinson +products for their luxurious quality and revealing sound. +Robust materials, lavish finishes, bold geometry, and +outstanding performance are celebrated attributes of the +Mark Levinson brand. The No 5909 headphone is made with +the finest materials, technology, and craftsmanship in the +world and will deliver outstanding performance for many +years to come. -16. Only remove the water reservoir -when the appliance is turned off . +For optimum performance, please read this guide and +follow the unpacking and setup instructions carefully. -17. This appliance is not intended -to be operated by means of an -external timer or separate remote- -control system . +WHAT'S IN THE BOX -appliance and allow it to cool -down . The blockage may be -caused by the coffee being too -finely ground . +• Mark Levinson No 5909 Premium Wireless ANC Headphones -20. do not place the appliance on or -near a hot gas or electric burner, -or in a heated oven . +• Travel Storage Case -21. do not use the appliance for other +• 1.25m USB-C to USB-C Charging Cable -than its intended use . +• 1.25m USB-C to 3.5mm Proprietary Audio Cable -18. do not remove the portafilter +• 4m USB-C to 3.5mm Proprietary Audio Cable -at any time during the brewing -process . +• 3.5mm to 6.3mm Adaptor -19. If coffee does not drip, it means -the filter is blocked . Turn off the +• USB-C to USB-A Adaptor -22. do not leave the appliance -unattended when turned on . +• Airplane Adaptor -23. This appliance should only be -used in households or similar -environments . +• Microfiber Polishing Cloth -S -d -r -a -u -G -E -f -a -S +• Quick Start Guide -t -n -a -t -r -o -p -M +OVERVIEW -I +N05909 HEADPHONE / OWNER’S MANUAL -Cord and plug Safety -1. A short power-supply cord is provided to reduce risk of becoming entangled +3 -in or tripping over a longer cord . + OVERVIEW -2. Extension cords may be used if care is exercised in their use . +PRODUCT OVERVIEW -3. +1. Premium Leather Headband with Memory Foam Interior -If an extension cord is used: +2. Replaceable Memory Foam and Memory Foam Ear Pads -• The marked electrical rating of the extension cord should be at least equal +3. Forged and Machined Aluminum Components throughout -to the electrical rating of the appliance . +4. USB-C Input for Passive Audio Listenting & Charging -• The longer cord should be arranged so that it will not drape over the +5. Adjustment Arm -countertop or tabletop where it can be pulled on by children or tripped -over . +6. 8 Microphones (4 talk microphones and 4 ANC microphones) -Electrical polarized plug -This unit has a polarized plug (one blade is wider than the other) . -To reduce the risk of electric shock, the plug is intended to be -used by inserting it into the electrical outlet only one way . If the -plug does not fit securely into the outlet, reverse the plug and -re-insert . If it still does not fit, contact a qualified electrician for -assistance . Do not modify the plug in any way . +4 -poWEr +1 -AC120V~ 60Hz 1,300W +2 5 -USER MANUAL ESPRESSO MAKER GCM4230 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K - -KNOW YOUR -ESPRESSO MAKER - -Water Reservoir -Cover - -Removable -Water Reservoir - -Control Panel - -Brew Head - -Power Switch - -On the back-right -side of the unit - -Removable Drip Tray +3 6 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Milk Reservoir Lid +4 -Froth Strength Control +N05909 HEAPHONE / OWNER’S MANUAL ADJUSTABILITY, FIT & COMFORT -Froth Dispensing Tube Lever +ADJUSTABILITY, FIT & COMFORT -Froth Dispensing Tube +Adjust to find optimal fit -Minimum Strength +Rotates for comfort around neck -Clean / Americano Position +5 -Maximum Strength +N05909 HEADPHONE / OWNER’S MANUAL FUNCTIONS -Removable Milk Reservoir +FUNCTIONS -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K +Active Noise Cancelling / Awareness (ANC) Button -Measuring Scoop -and Tamper +Bluetooth Pairing/ Power ON/OFF -Portafilter Handle +Pairing Indicator -Portafilter Retention Clip +Active Noise Cancelling / Awareness (ANC) Button -Portafilter +Pairing Indicator -Double-Shot -Filter +White light will pulse when headphones are in pairing +mode -Single-Shot +A solid white light indicates you are connected -Control panel -All icons will light up when the milk reservoir is correctly inserted into the unit . +x1: ANC ON +(Delfault selectable via app; HIGH, ADAPTIVE, LOW +modes) -ESPRESSO -Icon +x2: Awareness ON +(Default selectable via app; Voice Pass or Ambient +modes) -LATTE -Icon +x3: Passive / ANC OFF -MANUAL -ESPRESSO Icon +Clicking the button will cycle through the different +modes. -CAPPUCCINO -Icon +6 -AMERICANO -Icon +N05909 HEAPHONE / OWNER’S MANUAL -FROTH -Icon + FUNCTIONS -CLEAN -Icon +Vol + -7 +Play/Pause / Answer/End Call -USER MANUAL ESPRESSO MAKER GCM4230 +Vol - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +Battery Level Indicator -I +USB-C Charging Port -USING YOUR -ESPRESSO MAKER -Before first use -1. remove and discard all packaging material and stickers that are not +Play/Pause / Answer/End Call -permanent . +USB-C Charging Port -2. remove the water reservoir, milk reservoir, portafilter, filters, measuring -scoop, and removable drip tray. Wash them with non-abrasive soap . -Thoroughly dry the parts . Wipe the exterior with a damp cloth . +Fully Charged -3. Before using the espresso maker for the first time, it is required to prime the +Charging -appliance . See Setting up and priming the Espresso Maker below . +x2: Skip Forward -Setting up and priming the -Espresso Maker -1. Place the espresso maker on -a hard, flat, and heat-resistant -surface . +x3: Skip Back -7. Set the froth strength control to +HOLD: Voice Assistant -the clean position . +Battery Level Indicator -8. Turn the froth dispensing +High Battery -2. Detach the water reservoir and -open the water reservoir cover . +Med Battery -tube lever to rotate the froth -dispensing tube upward . +Low Battery -9. Place a heatproof cup or mug +N05909 HEADPHONE / OWNER’S MANUAL -under the froth dispensing tube . - -10. Tap the CLEAN icon . The clean +7 -icon will blink until the espresso -maker is done preheating . After -preheating, water will dispense -into the cup . When cleaning is -complete, the appliance will beep -three times . + CHARGE -11. Discard the water dispensed into +CHARGE -the cup . +Press and hold (>1 second) ON/OFF button to power ON. +Left earcup LED will show solid white light while unit is on - +Battery level will be displayed for a few seconds at startup on +Right earcup - Green high charge - Orange medium charge +- Red low charge -3. Fill the water reservoir with water -between the “Min” and “Max” -mark and reattach it to the back -of the appliance . . +To conserve battery power, headphones will power OFF after +30 minutes of inactivity (default) - Default can be changed via +app - 30min, 1HR, 3HRS, NEVER -4. Plug the power cord into an +To charge, plug in the USB-C Connector and connect to a +powered USB-A port -outlet . +NOTE: The headphones power OFF when charging. +Headphones cannot charge and play at the same time. -5. Turn the espresso maker on by -flicking the power switch on the -right side of the appliance . The -ESprESSo icon and ManuaL -ESprESSo icon will light up on -the control panel . +A solid orange light on the left earcup indicates charging +and a solid green light indicates a full charge when the USB +cable is connected. -6. +Fully Charged -Insert the empty milk reservoir -into the appliance until it clicks -into place . All icons will light up on -the control panel . +Charging 8 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - filling the portafilter -1. Select your filter size . Use the - -single-shot filter for one shot of -espresso . Use the double-shot -filter for larger sized drinks, or -for a double shot of espresso . -Align the gap in the filter with the -groove in the portafilter . Drop the -filter into the portafilter and rotate -the filter to ensure it’s locked in -place . - -2. Use the measuring spoon to fill - -the filter with espresso grounds . -Do not exceed the MAX line in -the filter . - -3. Use the tamper end of the - -measuring spoon to press on -the grounds . The surface of the -espresso grounds should be -smooth after pressing . - -4. Clean any excess grounds from - -the rim of the portafilter . - -placing the portafilter -1. Place the portafilter underneath -the brew head, with the handle -facing toward the left . - -2. Keep one hand atop the - -appliance to keep it steady while -fastening the portafilter into the -brew head . The handle must -align with the open lock icon on -the brew head . - -3. Turn the handle to the right to - -lock the portafilter into the brew -head . The handle must align with -the closed lock icon on the brew -head . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I +N05909 HEAPHONE / OWNER’S MANUAL -9 + CONNECTIONS -USER MANUAL ESPRESSO MAKER GCM4230 +CONNECTIONS - Mug Size Guide +PAIRING -ESPRESSO +The first time you power ON your headphones they will enter pairing mode. Simply connect through the Bluetooth menu on your +device and select Nº5909. -CAPPUCCINO +To manually activate pairing mode, power headphones OFF, then hold ON/OFF button for 4 or more seconds. -LATTE +Pairing to 2 or More Devices -AMERICANO +1. Enter pairing mode and connect to device 1. -Drink +2. Enter pairing mode again and device 1 will disconnect. -Shot +3. Connect to device 2 and turn headphone off. -Size +4. Turn headphone ON and headphone will automatically connect to device 1 and 2 simultaneously. -SINGLE +5. Pause one device and play from the other to swap headphone connection. -DOUBLE +NOTE: Maximum number of devices that can be connected at any given time - 2 source devices. -SINGLE +PASSIVE -DOUBLE +The № 5909 can be connected passively to any product that has an analog 3.5mm or 6.3mm (1/4") audio connetion using the +included 1.25m or 4m to USB-C proprietary audio cables included. -SINGLE +1. Choose the appropriate length (1.25m or 4m) proprietary audio cable depending on distance from the source component. -DOUBLE +2. Plug the USB-C connection end into the USB-C input on the Right ear cup. -SINGLE +3. Plug the 3.5mm stereo mini jack on the other end of the cable to your source component's headphone input. If needed, use +the 6.3mm (1/4") adaptor for traditional audio components. -DOUBLE +NOTE: When connecting passively, none of the electronic or app functions are operational. -2 OZ +Computer / Smart Phone -3 OZ +3.5mm Proprietary Audio Cable -5 OZ +Standard AV Products -10 OZ +6.3mm Adaptor + 3.5mm Proprietary Audio Cable -12 OZ +USB-C -16 OZ +N05909 HEADPHONE / OWNER’S MANUAL -5 OZ +9 -10 OZ + CONNECTIONS -note: To accommodate larger mugs, remove the drip tray . +CONNECTIONS -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +USB DIGITAL DIRECT -I +The № 5909 can be connected and used to listen to content via USB Digital Direct using the standard USB-C to USB-C charging +cable included. -drink types +1. Plug one end of the USB-C to USB-C cable directly into one of your computer's USB connections (use the included USB-C +to USB-A adaptor if needed). -Espresso -One shot of -espresso . +2. Plug the other end of the USB-C cable into the USB-C input on the right ear cup of the headphone. -Cappuccino -Made up of -1/3 espresso, -1/3 steamed milk, -and 1/3 milk foam . +3. From your computer device, open the "sound settings" and locate and select the "No 5909" under the speaker label. -Latte -Made up of -1/3 espresso and -2/3 steamed milk -topped with milk foam . +4. Adjust your volume via the slider on your computer. -Americano -Made up of -1/3 espresso and -2/3 hot water . +5. Start your content - enjoy. -Espresso +NOTE: ANC functions will continue to operate via the left ear cup ANC button AND the headphone will charge while in this +connection method. -Steamed Milk +Computer -Hot Water +USB-C to USB-A Charging adaptor + USB-C Charging Cable -Milk Foam +USB-C 10 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Making an Espresso - -Ensure the water reservoir is filled with water and placed to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a cup on the drip tray . - -4. For a single shot, tap the - -ESPRESSO icon once . For a double -shot, tap the ESPRESSO icon twice . -Your selected drink type will blink -until the espresso maker is done -preheating . After preheating, the -brewing process will begin . - -5. When brewing is complete, the -appliance will beep three times -and your espresso will be ready -for sipping . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -11 +N05909 HEAPHONE / OWNER’S MANUAL -USER MANUAL ESPRESSO MAKER GCM4230 + APP - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +APP -I +The Mark Levinson Headphones app allows you to choose +your listening preference and select Active Noise Cancellation +and Awareness modes. Select Bass Contour settings, Set +on-head detection and auto time-off parameters. -Making a Cappuccino +App Start-up -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +Landing Page -1. Fill the milk reservoir with milk +Settings -7. For a single shot, tap the +Support -below the “Max” mark . +N05909 HEADPHONE / OWNER’S MANUAL -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -cappuccino function will not be -available as an option . +11 -3. Set the froth strength control to + MAINTENANCE / CARE / CHANGING / REMOVING EARPADS -the maximum strength . -note: The above setting is a -recommendation only . +MAINTENANCE / CARE -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . +Clean ear pads and headphones with soft, slightly damp cloth. Do not soak, submerge or allow moisture to enter near the +drivers or cable jacks. Ear pads and cables can be purchased if they are worn out at: www.marklevinson.com -5. To attach the portafilter to the +Polishing cloth can be used to wipe down after use / remove finger prints, etc. . . -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . +Do not drop, sit on, or allow the headphones to be exposed to water, moisture, or temperature extremes. -6. Position the froth dispensing tube +It is recommended that you return your headphones to their hard travel case when not in use. To avoid damage, do not store +other items in the travel case along with your headphones. -over a large heatproof cup . +CAUTION: Incorrectly rotating or over rotating the earcups can damage the headphones. -CAPUCCINO icon once . For a -double shot, tap the CAPPUCCINO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . +Temperature Range Operate and store this product within the temperature range of -4°F to 113°F (-20°C to 45°C) only. Charge +the battery where the temperature is between 41°F and 104°F (5°C and 40°C) only. -8. When brewing is complete, the -appliance will beep three times -and your cappuccino will be -ready for sipping . +CHANGING / REMOVING EARPADS -9. Discard the spent grounds by +To remove ear pads: Gently pull the earpads from the inside +of the earcup. The snaps will release. -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +To re-attach ear pads: Gently push the earpad against the +earcup, ensuring all the snaps are secured. 12 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making a Latte - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. Fill the milk reservoir with milk - -below the “Max” mark . - -6. Position the froth dispensing tube -over a large heatproof cup . - -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -latte function will not be available -as an option . - -3. Set the froth strength control to - -the minimum strength . -note: The above setting is a -recommendation only . - -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -5. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -7. For a single shot, tap the LATTE -icon once . For a double shot, -tap the LATTE icon twice . Your -selected drink type will blink until -the espresso machine is done -preheating . After preheating, the -brewing process will begin . - -8. When brewing is complete, the -appliance will beep three times -and your latte will be ready for -sipping . - -9. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +N05909 HEAPHONE / OWNER’S MANUAL -13 + ACCESSORIES -USER MANUAL ESPRESSO MAKER GCM4230 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making an americano - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. - -Insert a filled or empty milk -reservoir into the appliance until it -clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -americano function will not be -available as an option . - -2. Set the froth strength control to -the Clean / Americano position . - -3. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -4. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -5. Position the froth dispensing tube -over a large heatproof cup . - -6. For a single shot, tap the - -AMERICANO icon once . For a -double shot, tap the AMERICANO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . - -7. When brewing is complete, the -appliance will beep three times -and your americano will be ready -for sipping . - -8. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +ACCESSORIES -14 +Travel Storage Case -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Brewing in Manual Mode - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a heatproof cup over the - -drip tray . - -4. Tap the MANUAL ESPRESSO icon . - -The text below the icon will blink -for three seconds then the brew -cycle will begin . - -5. When your desired amount of -coffee has dispensed, tap the -MANUAL ESPRESSO icon again and -coffee will cease brewing . -note: The maximum amount of -coffee that will dispense is 14 oz . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I +USB-C to USB-C Charging Cable -15 +USB-C to USB-A Charging adaptor -USER MANUAL ESPRESSO MAKER GCM4230 +USB-C to 3.5mm Proprietary Audio Cable +(Both 1.25mm and 4mm Audio Cables included) - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +3.5mm to 6.3mm (1/4" standard) +headphone adaptor -I +Airplane adaptor -frothing Milk +N05909 HEADPHONE / OWNER’S MANUAL -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +13 -1. Fill the milk reservoir with milk + TROUBLESHOOTING -5. Tap the FROTH icon . +TROUBLESHOOTING -below the “Max” mark . +TROUBLE PAIRING HEADPHONES AND DEVICE -2. Secure the milk reservoir in the +THE HEADPHONES AND DEVICE DO NOT PAIR: -appliance until it clicks into place . +1. Make sure the headphones are not connected to the charger. Cycle power switch OFF and ON. -3. Set the froth strength control . +2. Turn the Bluetooth OFF, then on again on the source device. -4. Position the froth dispensing tube +3. Turn OFF the Bluetooth feature on any other device that is paired with the headphones. -over a heatproof cup . +4. Hold power button down >4 seconds to enter pairing mode. -6. Milk will froth for a maximum of -three minutes . To stop frothing -earlier, tap the FROTH icon again -and milk will cease frothing . +5. Make sure “Nº5909" is selected in the Bluetooth® device list -16 +NO AUDIO OR POOR AUDIO QUALITY FROM A CONNECTED BLUETOOTH® DEVICE -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +Move the device closer to the headphones, physical objects between the headphone and the source may obstruct the signal. - I +Try another music application or try another track. -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +Try a different audio device. -I +FACTORY RESET -CLEANING AND MAINTENANCE +1. Power OFF the headphones. -the following cleaning instructions should be followed after every -use of the espresso maker. +2. Hold power button >4 seconds to enter pairing mode. -Cleaning the portafilter and Brew Head +3. Simultaneously hold volume + button and power button until LED lights flash red/white. -1. Turn the appliance off and unplug -the power cord from the outlet . -Wait for the metal parts to cool -down before touching them . +14 -2. Remove the portafilter from the -brew head and dump the used -espresso grounds . Lift the filter -retainer clip so that it’s placed -over the filter edge, and hold it in +N05909 HEAPHONE / OWNER’S MANUAL EAR HEALTH -place as you dump the espresso -grounds into the trash . +EAR HEALTH -3. Rinse the single-shot filter, +Our ability to hear is amazing. The Mark Levinson team wishes you to treat your ears like the precious tools they are so they will +continue to provide you a lifetime of incredible sonic experiences. -double-shot filter, and portafilter -with warm, soapy water . +In general, do not listen “too loud or for too long,” and be attentive to your own ears. Ringing, discomfort, or sensitivity to high +frequencies and treble may be a signal from your ears that you are pushing them to their limits and causing irreparable damage. -4. Wipe the underside of the brew -head with a damp cloth to clean -off any excess coffee grounds . +As a general practice, find what seems to be your preferred listening level and then adjust the volume down another 10% or +more. Surprisingly, your ears will attune to a slightly lower volume over time, and what seems quiet at first can be perfect for +longer listening periods. -Cleaning the Water reservoir +15 -1. Discard leftover water . +N05909 HEADPHONE / OWNER’S MANUAL SPECIFICATIONS -2. Wash the water reservoir with soapy water . Rinse and dry well . +SPECIFICATIONS -Cleaning the Milk reservoir and frothing tube +GENERAL -1. Remove the milk reservoir from +Type: -the unit . If there is any remaining -milk, pour the milk into a separate -container and store it in the -refrigerator . +Driver Size: -2. Wash the milk reservoir and milk +Premium High-Resolution Over-the-Ear Wireless Adaptive ANC Noise Cancelling Headphone -reservoir lid thoroughly with warm, -soapy water . Do not use abrasive -cleansers and scouring pads . -note: Hand wash only! The milk -reservoir is not dishwasher safe . +40mm Beryllium Driver -3. Ensure there is water in the water +Frequency Response (Passive): -reservoir . +10Hz - 40kHz -4. Position the froth dispensing tube -over a large heatproof cup . +Frequency Response (Active): -5. Set the froth strength to the Clean -/ Americano position . Tap the -CLEAN icon . +20Hz - 20kHz -6. Wait until all the water has -finished dispensing before -removing the cup from -underneath the brew head -(the unit will beep three times -and other function buttons will -reappear) . +Sensitivity: -7. Dispose of the water in the mug -and any remaining water in the -milk reservoir and water reservoir . +Maximum SPL: -17 +Maximum Input Power (Passive): -USER MANUAL ESPRESSO MAKER GCM4230 +97dB SPL @1kHz/1mW - I +98dB -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +100mW -I +Microphone Sensitivity: -descaling +-32dBV @ 1kHz/Pa -After using the espresso maker for a while, limescale will begin to build up inside -the appliance . It is important to descale the appliance every 2-3 months . If the -espresso maker is used daily, the descaling process should be done monthly . +Impedance: -1. Fill the removable water reservoir +Bluetooth Transmitted Power: -5. On the following day, discard the +32 ohm -with equal parts water and -vinegar between the “Min” and -“Max” mark . +<8dbm -2. Use the froth dispensing +Bluetooth Transmitted Modulation: -tube lever to rotate the froth -dispensing tube upward and -position a mug underneath it . +GFSK, π/4 DQPSK / 8DPSK -3. Tap the CLEAN icon . +Bluetooth Frequency: -4. Allow remaining water and +2.402Ghz - 2.480Ghz -vinegar to sit overnight in the -water reservoir . +Bluetooth Profile Version: -troubleshooting +A2DP 1.3.1, AVRCP 1.6, HFP 1.7.1 -water and vinegar . +Bluetooth Version: -6. Rinse the water reservoir +v5.1 -thoroughly with tap water and -then pour the water back into the -sink . +Battery Type: -7. Fill up the water reservoir with +Power Supply: -water and begin brewing again . -This washes out any remaining -vinegar . -notE: If an odor of vinegar -persists, run another cycle with -water . +Charging Time: -ISSUE +Music Playtime With BT On: -SOLUTIONS +Music Playtime With BT & ANC On: -Only the ESPRESSO and MANUAL -ESPRESSO icons appear on the control -panel. +Music Playtime With Aux-In & ANC On: -A drink has been chosen and the unit is -running, but no coffee is dispensing from -the portafilter. +Lithium-ion battery (750 mAh/3.7V) -Water is spreading from the drip tray -when brewing or cleaning. +5V / 1.0A -The other icons will appear when the milk -reservoir is correctly inserted into the unit. -Insert the milk reservoir into the unit by aligning -the groove on top of the reservoir with the -notch on the unit and pushing. It will click into -place when done correctly. +100 minutes -Detach then reattach the water reservoir to the -unit (this may need to be done a few times). -Run a clean cycle. See CLEANING THE MILK -RESERVOIR AND FROTHING TUBE on page 17. +34 hrs -A. Water dispenses from the froth dispensing -tube when brewing Americano or running a -clean cycle. Ensure the froth dispensing tube is -above the cup. +30 hrs -B. Empty the drip tray. +30 hrs -18 +Dimensions (H x W x D): -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +202.6mm (7.9") x 205.4mm (8.1") x 65.4mm (2.6") - WARRANTY -AND SERVICE +340g (12 oz) -This Limited Warranty is for one full year from the date of -purchase and requires proof of purchase from an authorized -seller . It is applicable to the original purchaser only and is not -transferable to a third party user . Repair or replacement of -defective parts is solely at Gourmia’s discretion . In the event -that repair isn’t possible, Gourmia will replace the product/part . -If product repair/replacement won’t suffice, Gourmia has the -option of refunding the cash value of the product or component -returned . +2 Years Parts & Labor -Product defects not covered under the Warranty provisions -include normal wear and damage incurred from use or -accidental negligence, misuse of instruction specifications -or repair by unauthorized parties . Gourmia is not liable for -any incidental or consequential damages incurred by such -circumstances . +Weight: -Only valid within the United States . +Warranty: -register your product at -www.gourmia.com/warranty +16 -for customer service, questions or comments +N05909 HEAPHONE / OWNER’S MANUAL WARRANTY -Visit +Harman International Industries, Incorporated (“Harman”) warrants to you that your Mark Levinson product will be free from +defects in workmanship and materials for a corresponding period set forth in the Mark Levinson Limited Warranty. -gourmia.com +WARRANTY -Email +This Limited Warranty does not cover defects which are a result of: (1) damage caused by accident, disaster, misuse, abuse or +other unreasonable use or neglect (including the lack of reasonable and necessary maintenance); (2) damage occurring during +shipment (claims must be presented to the carrier); (3) damage caused by improper installation or adjustment; (4) damage +to, or deterioration of, any accessory or decorative surface; (5) damage resulting from failure to follow instructions contained +in your owner’s manual; or (6) damage resulting from the performance of repairs by someone other than an authorized Mark +Levinson service center. -info@gourmia.com +Furthermore, this Limited Warranty covers only actual defects within the Product itself, and does not cover the cost of +installation or removal from a fixed installation, setup or adjustments, claims based on misrepresentation by the seller, +performance variations resulting from installation-related circumstances such as program source quality or AC power or +Product modifications, any Product on which the serial number has been effaced, modified or removed, or any Product sold on +an “as-is” or final sale basis. This Limited Warranty is valid only for Products purchased from an authorized Mark Levinson +dealer. -Call +Complete Mark Levinson warranty details can be found at https://www.marklevinson.com/about/purchase-policy.html -888.552.0033 -Mon-Thu 9:00 a .m . to 6:00 p .m . ET -Fri 9:00 a .m . to 3:00 p .m . ET -Languages: English, Spanish, French +17 -@gourmia +N05909 HEADPHONE / OWNER’S MANUAL HARMAN International Industries, Incorporated +8500 Balboa Boulevard +Northridge, CA 91329 USA -I +© 2020 HARMAN International Industries, Incorporated. All rights reserved. -E -C -v -r -E -S -d -n -a -Y -t -n -a -r -r -a -W +Mark Levinson is a registered trademark of HARMAN International Industries, Incorporated. -19 +Other company and product names may be trademarks of the respective companies with which they +are associated. -USER MANUAL ESPRESSO MAKER GCM4230 +"MQA" is a trade mark of MQA Limited © 2018. - Get cooking with +This document should not be construed as a commitment on the part of HARMAN International +Industries, Incorporated. The information it contains, as well as the features, specifications and +appearance of the product, is subject to change without notice. HARMAN International Industries, +Incorporated, assumes no responsibility for errors that may appear within this document -We’d love to hear from you! -888.552.0033 -info@gourmia.com +For customer service and product shipment information, refer to our Website: www.MarkLevinson.com +www.MarkLevinson.com +Part No. 070-00004 rev B.0 + +18 -20 +N05909 HEAPHONE / OWNER’S MANUAL -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 \ No newline at end of file diff --git a/q121/random_k4/question.json b/q121/random_k4/question.json index f0cf70f5638a88f8a28d2e917221e79a10b32b6b..a03ab83502389b062d79c3d64548dfae38974d1f 100644 --- a/q121/random_k4/question.json +++ b/q121/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "tacl_a_00660.pdf", - "web_d8f936ecd4367f91.pdf", - "web_4cf87d006f2e4a65.pdf", - "web_2ed0552c74ec7055.pdf", - "web_77f281bc58aa8364.pdf" + "resize_r2_25.pdf", + "web_e025dffc1a22e65f.pdf", + "resize_fb_1.pdf", + "web_7eca0520d1d88795.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q121/random_k4/random_1.md b/q121/random_k4/random_1.md index 2cfb2a95e81c5a4894ee70af1d1b3fdec18ec7c6..a7cbcf7d4134f4c225b224673f418e279c4db69c 100644 --- a/q121/random_k4/random_1.md +++ b/q121/random_k4/random_1.md @@ -1,1447 +1,1314 @@ - b o s t o n c o l l e g e - -Exchange Student -Handbook - -Office of International Programs - - We are delighted that you have chosen to attend Boston College as an ex- -change student. We value the partnership that we have developed with your -home university and look forward to welcoming you as a representative of -your institution. Your presence will add a significant element to the interna- -tional dimension of Boston College, and we hope you will take advantage of -every opportunity to share your culture with our community. - -This handbook is intended to give you the vital information you will need to -prepare for your trip to the United States and your first days in Boston. The -information included here is not meant to be comprehensive — you will be -given a complete information packet upon arrival — but rather to inform you -of the things you need to do in order to make a smooth transition from home -to Boston College. Please read everything carefully and feel free to contact us -with any questions you might have. We are committed to making your stay -with us as rewarding as possible. - -At the Office of International Programs we want to provide you with a hassle- -free transition. We work closely with numerous offices on the Boston College -campus. One of those offices is the Office of International Students and -Scholars (OISS), which is responsible for international student services on -the Boston College campus. This office issued your visa documents and will -work with you on any immigration-related matters. Please pay close attention -to materials you receive from the Office of International Students and Schol- -ars, especially since they are experts on immigration matters and will be re- -sponsible for helping you maintain your legal status in the United States. - -As an exchange student, you are given the opportunity to study at Boston Col- -lege for a semester or academic year as part of your academic program at -your home university. Please note that it is not possible to transfer to Boston -College to finish your degree. Boston College has a variety of graduate pro- -grams that will welcome your application once you have received your degree. - -Please visit www.bc.edu for detailed campus information. For information on -the Office of International Programs see www.bc.edu/international. - -We wish you all the best as you begin to prepare for your upcoming trip to -Boston. Please let us know if we can be of any assistance. - -Bernd Widdig, Ph.D. Esther Messing -Director, Office of International Programs Assistant Director for - - International Exchange Students - +B160V + +User Guide + + Contents +Safety information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 +Getting to know your phone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 +Setting up your phone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 +Contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 +Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 +Email . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 +Chrome browser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 +Bluetooth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 +Wi-Fi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 +Camera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 +Photos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 +Calendar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 +Calculator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 +Calculator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 +Clock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 +Google Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 +Google Play Store . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 +Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 +Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 +Regulatory Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 1 -www.bc.edu/international +Safety information - TABLE OF CONTENTS +Please carefully read the items bellow: +Use Safely +Do not use the phone in situations where it might cause danger. +Transportation Safety +Please observe all transportation laws and regulations. +Please drive with both hands on the wheel. - INTRODUCTION +DO NOT TEXT AND DRIVE - BEFORE YOU LEAVE HOME +Hospitals +Please follow limitations. +Please turn off your mobile phone when near medical instruments. +Airports +Remember to follow all airport and flight safety regulations. +Please do not use your phone during flight. - Student Visa - Travel Plans - Medical Insurance - Living Expenses +2 - YOUR FIRST DAYS IN BOSTON + Toxic Chemicals +Do not use your mobile phone around fuels or chemicals. - Getting to Boston College - On-Campus Housing - Off-Campus Housing - Hotels and Temporary Housing - Checking-in and Orientation +Hazard +Do not use in hazardous areas where explosions may occur. - ACADEMICS +Accessories and Batteries +Use only Blu authorized batteries and accessories. - Full-time Status - Course Selection and Registration - Drop/Add Period - Faculty Advisors - Tutoring - English Classes for Non-Natives - Transcripts of Academic Work +Backup +Remember to keep a written record of all important information. - ADMINISTRATIVE ISSUES +Water +Your phone is not waterproof. Keep it away from water and liquids. - Student Identification Card - Medical Insurance - Health Services - University Counseling Services - Students with Disabilities - E-mail +SOS Emergency Calls +Make sure your mobile phone is switched on and in a service area. In +home screen, tap the phone key and dial 911 and send. - LIFE BEYOND ACADEMICS +3 - CONTACT INFORMATION +Getting to know your phone -2 +Receiver + +Headset port +Status bar + +Front camera + +Rear cameras + +Volume keys + +Power key + +Flash + +Back +Home + +Recent Apps + +USB port 4 + Power +key + +• Press to turn the screen on or off. +• Press and hold to turn the phone on. +• Press along with the Volume up key to turn the phone off. +• Press along with the Volume down key to take a + +screenshot. + +Volume +keys + +Adjust the volume for ringtones, media, notifications, and +system sounds. + +Home + +The home button pauses any current action and returns to +the home screen. + +Recent +Apps + +Open a list of recently opened applications that allows you +to switch conveniently between them. This option can also +close any open applications by swiping to the side. + +Back + +Return to the previous screen; Closes the keyboard, any +open program, or any menu option. + +USB port + +Charge the device and transfers data such as documents, +music, and video between your device and a PC. + +Headset +port + +Connect earphones to this port for hands-free capabilities. +You may also to listen to music or FM radio. + +Status +bar + +Provide phone and service status information on the right +side and notification alerts on the left. + +Front +Camera + +A high-resolution front camera to take pictures and video. +This option is ideal for self-portraits. + +Rear +Cameras + +Flash + +A high-resolution rear camera to take pictures and video. + +Provide light for pictures in low-light conditions. It also +may be used as a flashlight. + 5 -5 -6 + 6 + Setting up your phone + +Charge the battery +1. Connect one end of the data cable to the USB port located on the + +Install a Nano SIM/microSD card + +bottom of your phone. + +Note: Please contact your cellular phone service provider directly to +obtain your SIM card if your card is not preinstalled. +1. Insert the included SIM tray pin into + +the tray slot and remove the SIM tray. +2. Place the Nano SIM card and microSD + +microSD + +Nano SIM + +card (not included) with metal +contacts facing down into the tray. +3. Gently slide the SIM tray back into + +SIM tray + +Pin + +the slot. + +Warning: Please keep the SIM card out of the reach of children. +The SIM card and its contact are easily damaged due to scratching or +bending. Please use caution when carrying, installing, or taking out the +SIM card. + +2. Connect the other end to the power adapter. +3. Connect the power adapter to an electrical outlet to fully charge your + +phone. + +Power On and Off +Long press the Power key to switch on the phone. +To turn off the phone, press the Power key and the +Volume up key at the same time, and tap Power Off on +the screen that displayed. + +Power +key + 7 + 8 + + Phone +Make a call + +1. To make a call, tap Phone +2. Tap Dialer + + from the Home screen. + + to access the keypad, or select one of the following: + +• Choose frequently called or favorite contacts in Favorites. + +• Choose recently called numbers and contacts in Recents. + +• Choose a saved contact in Contacts. + +3. Tap +4. During a call, you can tap + + to make a call. + +up. + + at the bottom of the screen to hang + +Enable Wi-Fi call + +You can make a voice call with Wi-Fi Calling. +1. From the Home screen, swipe up to access Apps and tap Settings +2. Tap Network & internet > Wi-Fi Calling. Slide the switch to the + +. + +On + position to enable it. You must have an active Wi-Fi +connection and have Wi-Fi calling service activated before it can be +used. + +Call history + +Every telephone number called and received will be saved in the history. +1. From the Home screen, tap Phone +2. Tap Recents on the bottom bar to display the call history. +You can tap a contact in the call history to display more options. + +. + +Call settings + +1. From the Home screen, tap Phone +2. Tap Options > Settings to configure your phone settings. + +. + 9 -10 + 10 + Contacts + +Add a contact + +1. From the Home screen, swipe up to access Apps and tap Contacts + +. +2. Tap +3. Use the keyboard to enter contact information, such as picture, + + to add a contact. + +name, telephone number, group, address, email, etc. + +4. When you are finished, tap Save. + +Edit a contact + +To edit a contact, tap the contact from the Contacts screen to view its +details, and tap + +. + +Messages + +Send a text or multimedia message +1. From the Home screen, tap Messaging +2. Tap Start chat to create a new text or multimedia message, or start + +. + +a new group chat. + +3. Enter the recipient(s) name from your Contacts list or phone number + +in the To field. + + to send the text message. + +4. Type your message and tap +5. Or tap + + to the left side of the text box to add location, contacts, + +or files. Tap +recordings. When done, tap + + to attach photos or videos. Tap + + to send sound + to send the multimedia message. + +Message settings +From Message screen, tap + on the top right corner of the Messages +screen and tap Messages Settings to adjust your messages settings. + 11 -11 -13 + +12 + + Email + +Email setup (Gmail, Yahoo!, etc.) + +From the Home screen, tap Gmail +your account. + +, and follow the steps to set up + +Create and send an email message +1. Tap Compose from the Inbox screen to compose a new message. +2. Enter the recipient(s) email address in the To field. To add a carbon copy + +or blind carbon copy, tap + +. + +3. Enter the subject and content of your message. To add attachments, tap + + and select Attach file. + +4. Tap + + to send. + +Email settings +From Email screen, tap Menu +settings. + + > Settings to configure your Email + +Chrome browser +From the Home screen, tap Chrome +for information, and bookmark your favorite web pages for convenient +access. + + to browse the Internet, search + +Browsing web pages +From the Chrome browser screen, tap the address field and enter a web +address or a keyword to browse for web pages. + +Adding a bookmark +Open the web page you want to bookmark and tap Menu > + +. + +Opening a bookmark +From the Chrome browser screen, tap Menu > Bookmarks, select +the bookmark of the website you want to view. + +Browser settings +From the Chrome browser screen, tap Menu > Settings to customize +the browser. + 13 -14 -14 + 14 + Bluetooth + +Wi-Fi + +Bluetooth is a short-range wireless communications technology that +devices can use to exchange information and connect to different types +of devices which include headsets and other devices. + +Enable Bluetooth + +1. From the Home screen, swipe up to access Apps and tap Settings +2. Tap Connected devices > Connection preferences > + +. + +Bluetooth. + +3. Slide the Bluetooth switch to the On + + position. + +Pair Bluetooth devices + +1. From the Connected devices screen, tap Pair new device. +2. Tap a device under Available devices and follow the on-screen + +instructions to pair with the device. + +3. When the device is connected to your phone, You can then use the +headset or car kit to listen to music or make and receive phone calls. + +To use your phone’s Wi-Fi, you need access to a wireless access point or +hotspot. +1. From the Home screen, swipe up to access Apps and tap Settings +2. Tap Network & internet > Internet > Wi-Fi. +3. Slide the Wi-Fi switch to the On +4. Tap a Wi-Fi network to connect, enter the password (if applicable), + + position to enable Wi-Fi. + +. + +and then tap CONNECT . + +5. Your phone will be connected to the Wi-Fi network. You will see + +in the status bar. + +You can drag down the status bar twice to access Quick Settings panel +to enable or disable Bluetooth and Wi-Fi quickly and easily. + 15 -15 -16 + 16 -17 + + Camera + +Photos + +Take a photo +1. From the Home screen, tap Camera +2. Tap Photo to switch to Photo mode. +3. Aim the camera at your subject and tap + + to take the photo. + +. + +Record a video +1. From the Camera screen, tap Video to switch to Video mode. +2. Tap +3. Tap + + to start recording a video. + again to stop recording and save your video. + +Camera settings +From the Camera screen, tap Settings + + to adjust camera settings. + +You can view photos and watch videos that you’ve taken, downloaded, +or copied onto your storage card or phone memory. + +Edit photos +1. From the Home screen, tap Photos +2. Tap the picture you want to edit. +3. Tap +save. + +. + + to access the Edit screen. When finished, tap Save copy to + +Share pictures and videos +1. Select the photos or videos you want to share. +2. Tap + +, select a sharing method, and follow the on-screen prompts + +to send the photos or videos. + 17 18 + Calendar + +Calculator +Calculator + +Use the Calendar to create and manage events, meetings, and +appointments. + +Add an event + +1. From the Home screen, swipe up to access Apps and tap Calendar + +. +2. Tap Add +3. Enter event details, and then tap Save. + + to add an event. + +Calendar settings + +1. From the Calendar screen, tap Menu +2. Tap an option to adjust Calendar settings. + + > Settings. + +Your phone’s convenient built-in calculator lets you perform basic +Your phone’s convenient built-in calculator lets you perform basic +mathematical equations. +mathematical equations. +1. From the Home screen, swipe up to access Apps and tap Calculator +1. From the Home screen, swipe up to access Apps and tap Calculator + +. +. + +2. Enter the first number, tap a mathematical function key, and then +2. Enter the first number, tap a mathematical function key, and then + +enter the second number. +enter the second number. + +3. Tap +3. Tap + + to view the result, tap + to view the result, tap + + to clear the results. + to clear the results. + +View calculator history +View calculator history + +From the Calculator screen, tap Menu > History, or drag down on +From the Calculator screen, tap Menu > History, or drag down on +the history bar +the history bar +To clear the calculator history, tap Menu > Clear. To return to the +To clear the calculator history, tap Menu > Clear. To return to the +. +calculator, drag the history bar +. +calculator, drag the history bar + + back up or tap the Back button + back up or tap the Back button + + to view the calculator history. + to view the calculator history. + +19 + 20 -www.bc.edu/international + Clock - CHECKLIST +Google Maps - BEFORE YOU LEAVE HOME +The Clock app allows you to set alarms, view different world times, use +a stopwatch, set a timer, manage your bedtime, and use your phone as +a desk clock. -Apply for F-1 or J-1 visa from local US Embassy or Consulate -Make travel plans for arrival and departure from BC -Secure on-campus or off-campus housing. If necessary, arrange tem- +Use the Google Maps app to determine your location, get directions, +browse local businesses and attractions, rate and review places, and +more. -porary housing or hotel +Set the date and time manually -Complete and submit Immunization and Health Forms to BC - University Health Services +1. From the Home screen, swipe up to access Apps and tap Clock +2. Tap Menu > Settings > Change date & time > Set time +automatically > Off, and set the date and time manually. - AT BOSTON COLLEGE +. -Check in at the Office of Residential Life to pick up keys and get hous- +Create a new alarm -ing assignment or locate off-campus housing -Check in at the Office of International Programs -Check in at the Office of International Students and Scholars -Attend orientation. Dates, times, and locations are in your acceptance +1. From the Clock screen, tap Alarm +2. Set the time for the alarm and tap OK. -packet + > -Get student identification card, set up e-mail account, and establish -library account. This process will be explained at the orientation. +. -3 +Search for destinations +1. From the Home screen, swipe up to access Apps and tap Maps +2. Tap Search here. +3. Enter an address, city, facility name, etc., and then choose a search -www.bc.edu/international - - INTRODUCTION - -Office of International Programs - -The Office of International Programs (OIP), located in Hovey House at 258 -Hammond Street, is your primary contact while you are studying at BC. The -exchange program that brought you here is administered jointly between OIP -and a comparable office at your home university. In addition to admitting you -to Boston College, we pre-register you for courses, assist with housing selec- -tion, and can assign you to a faculty academic advisor if you wish. We also -maintain regular contact with our counterparts at your home university. -Please feel free to come to us with any questions or problems before, during, -or after your semester or year at Boston College. - -Our office also sponsors the Exchange Mentor Program. This program aims -to match you with a Boston College student who has studied abroad at your -home university. Mentors maintain regular contact with their students. - -OIP works alongside numerous offices on the Boston College campus to en- -sure that you have the same access to facilities, services, and opportunities -as all other BC students. Other offices that you may come into contact with -include: the Office of International Students and Scholars, Student Services, -Health Services, Residential Life (which encompasses the On-Campus Hous- -ing Office and the Off-Campus Housing Office), and various academic de- -partments. - -Office of International Students and Scholars - -The Office of International Students and Scholars (OISS) is responsible for -issuing the immigration documents required for international students to -apply for a student visa. They also issue work authorization and advise on -other issues related to maintaining legal status in the United States. In col- -laboration with OIP, OISS organizes a mandatory International Student Ori- -entation each semester. - -In addition to these important services, OISS sponsors programs for interna- -tional students throughout the year, including welcome socials, Boston walk- -ing tours, the Thanksgiving Host Family Program, and many other activities. -They also provide undergraduate international students with International -Assistants, who are BC students serving as peer advisors to international stu- -dents. You will meet these International Assistants during orientation. +. -4 +result. You will see your location on the map. -www.bc.edu/international - - Things to Know Before You Leave Home - -Student Visa - -As a participant of an exchange pro- -gram to Boston College, you will -most likely be entering the United -States either on an F-1 student visa -or a J-1 exchange visitor visa. To ap- -ply for a student visa you will need: - - - - - -the enclosed Certificate of Eligi- -bility Form I-20 for F-1 visa (or -Form DS2019 for J-1 visa) - -a passport valid for at least six -months after your scheduled -return to your home country - - proof of funding such as a letter - -from your bank - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -With these items, you must apply to -the appropriate Embassy or Consu- -late General of the United States in your home country. If you are applying -for an F-1 visa, be sure to complete the items on the bottom of pages 1 and 3 -of your I-20 before applying for your visa. - -It is important to understand that as an exchange student on a non- -immigrant F-1 or J-1 visa you are required to be enrolled as a full-time student -for the duration of your participation on the exchange program. Please refer -to the Academics section for more information on full-time status. - -Travel Plans - -As an exchange student, it is your responsibility to arrange for your arrival to -and departure from the United States. Please refer to the Academic Calendar -enclosed with your acceptance packet for important dates. - -You should plan to arrive at Boston College on a weekday (Monday - Friday) -between 9:00 am and 5:00 pm, when university offices are open to assist you. -On-Campus Housing staff may be available on the weekend prior to the start -of the semester. Details will be sent at a later date. - -If you are planning to live off-campus and have not secured housing prior to -departure you should arrive as early as possible (5-7 days) to look for housing -before the beginning of orientation. Please see p. 9 for more information. +Find a route to your destination +Tap Go +, enter your desired start location and destination, and tap a +method of transportation (car, public transit, or on foot) for the detailed +directions to your destination. -5 +21 -www.bc.edu/international - - Also, please be aware that the following immigration regulations are in effect -for all F-1 and J-1 students: - - New international students will not be permitted to enter the U.S. more -than 30 days before the start date of their program as indicated on the I- -20 or DS2019 form. - - Upon completion of studies (i.e. the end date listed on the I-20 or - -DS2019), F-1 students are permitted to remain in the U.S. for 60 days -and J-1 students are permitted to remain in the US for 30 days. - -Additional immigration information is available at the following web site: -http://www.bc.edu/offices/odsd/intercultural/immigration.html. - -Medical Insurance - -The Commonwealth of Massachusetts requires all full-time students (both -American and international) attending colleges and universities in Massachu- -setts to subscribe to a US-based insurance plan. You will be automatically -enrolled in the BC medical insurance upon your admission to the university. -If you do not want this insurance, you must submit proof of insurance with a -provider that has a US office after your arrival. Please see p. 15 for more infor- -mation on BC medical insurance. - -Living Expenses - -Past exchange students -have been surprised by -the cost differences be- -tween their home coun- -try and the Boston area. -When estimating the -costs of your studies -and living in Boston, be -sure to budget for -housing, meals, trans- -portation, books, enter- -tainment, and any trav- -els you plan to take be- -yond the local Boston -area. Specific informa- -tion on costs can be -found on the Estimated -Expenses page in the -back of this handbook. - -www.bc.edu/international +22 -6 + Google Play Store -s -t -t -e -s -u -h -c -a -s -s -a -M - -, -y -b -n -a -r -G - - YOUR FIRST DAYS IN BOSTON - -Getting To Boston College - -Most students come to Boston -by flying to Logan International -Airport. Getting to Boston Col- -lege from Logan is not hard and -can be done in many ways: - -Taxi and Airport Shuttle - -The simplest, yet most expen- -sive, way to get to Boston College -is by taxi. - -Another alternative is to utilize -one of the private airport shuttle -services that operate from Logan -and will drop you off at Boston -College. Prices vary by company. -You might need to make a reser- -vation in advance. For a listing of shuttle services available at Logan, visit -www.massport.com/logan. A listing of private car services can be found on -the internet by Googling “livery services in Boston.” - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Public Transportation - -Finally, you can utilize public transportation to get to Boston College. The -subway is known locally as the “T.” For more information consult the Massa- -chusetts Bay Transportation Authority website at www.mbta.com. - -To take the "T" to Boston College, take the airport shuttle bus to the Airport -subway stop. Take the “Blue Line” subway from the airport to the -“Government Center” subway stop, where you should change trains to the -“Green Line.” Please note that there are four branches of the Green Line! Be -sure to take the “B-Line” train outbound, destination "Boston College," and -get off at the last stop. A public transportation map is included in your accep- -tance package. Obtain a reusable “Charlie Card” at the subway station to re- -duce your T fare. - -For more information on Logan International Airport and traveling into Bos- -ton from the airport, please consult www.massport.com/logan. +Access Google Play -7 +1. From the Home screen, tap Play Store +2. Enter your existing GoogleTM Account login or create one by following -www.bc.edu/international - - Moving Into On-Campus Housing - -Boston College has a number of residence halls. If you will be living on- -campus, we try to place you in residence halls with kitchen facilities. Students -living in rooms without access to kitchen facilities are required to purchase -the Boston College meal plan, which pays for meals in the campus dining -halls throughout the semester. Students living in rooms with kitchen access -have the option of purchasing the meal plan. If you do not wish to purchase -the full meal plan, you should strongly consider the “self-catering” option, -which allows you to purchase "Dining Bucks" in smaller amounts throughout -the semester. For more information on meal plans and Dining Bucks visit the -Dining Services web site at www.bc.edu/offices/bcds. - -Please be aware that sheets, blankets, pillows, towels, etc. are not provided in -the residence halls. These items can be purchased inexpensively in Boston, -so you may not want to take up luggage space with bedding and towels. - -The cost for on-campus housing varies by residence hall and room arrange- -ment. You will be assigned to available housing and will not be able to re- -quest a specific residence hall. Your BC student account will be billed once -your housing assignment is confirmed. You are then responsible for paying -for your housing upon arrival at Boston College. - -Every student on campus has a private phone number and a phone jack in -her/his room. You must provide your own telephone, which can be pur- -chased inexpensively in nearby stores. Local telephone service and voice mail -are free. Campus rates for long distance and overseas calls are very competi- -tive. Long distance calls are billed to your student account. Many students -also use cell phones and Skype for long distance and international calls. - -All on-campus rooms have free internet access, available through either an -Ethernet cable or wireless device. Contact 617/552-4357 with questions. - -Students with on-campus housing can move into their rooms on the first day -of the International Student Orientation for Fall/Full Year students, or on the -day before the Exchange Student Orientation for Spring students (see the -enclosed Academic Calendar). Be sure to arrive between 9:00 am and 5:00 -pm, when the housing office will be open to check you in and give you your -room key. The On-Campus Housing Office phone number is 617-552-3060. - -If you plan to arrive outside of normal business hours, please notify OIP as -soon as you make your travel plans so we can inform the housing office. -Upon arrival at BC, you should go directly to your residence hall and ask for -the RA (Resident Assistant) on call. The RA will let you into your room and -you can pick up your key on the next business day from the Residential Life -office. +. -8 +the steps on the screen. -www.bc.edu/international - - Off-Campus Housing - -If you will be living off-campus, you -will need to make your own housing -arrangements. The Boston College -Off-Campus Housing Office has up- -to-date lists of apartments and -roommate openings near Boston -College and acts as a referral source -for rooms in private homes. Your -name will be given to the Off- -Campus Housing Office and they -will contact you and assist you in -your housing search. The staff can -advise you on suitable arrange- -ments, convenient locations, etc. -However, they cannot make ar- -rangements on your behalf. If you -choose to utilize the above services, -it is possible that you may secure -housing prior to your arrival in the -United States. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Because apartment rentals in the BC area are expensive, many international -students either share an apartment with other students or live with families in -private homes in the surrounding area. We advise all exchange students who -will be seeking off-campus housing to arrive at least 5-7 days before orienta- -tion in order to have enough time to make the necessary arrangements. You -should also make necessary hotel arrangements prior to arrival (see the Ho- -tels and Temporary Housing section for suggestions). - -Finally, be sure to carefully read your lease or other agreements to rent an -apartment before signing and make certain that you understand the terms. If -you are unclear about what a particular document means, consult the Off- -Campus Housing Office for advice. There are laws to protect tenants in Mas- -sachusetts, so be sure to find out what your rights are in case you run into a -conflict with your landlord. - -The Off-Campus Housing Office web site offers helpful hints to assist you in -your search. To access their web site, go to www.bc.edu/housing and click on -the Off-Campus Housing link on the left sidebar. - -Once you have settled into your accommodations for the semester, be sure -to inform OIP of your address and phone number so that we know how to -contact you when necessary. You will receive a local housing notification form -in your orientation packet. +Navigate Google Play -9 +1. Tap Search for apps & games on the top of the screen to search -www.bc.edu/international +for a specific game or app. - Hotels and Temporary Housing +2. To install an app, tap the app and follow the prompts to complete +the process. After installation, you will see a notification in the +Notifications panel. You can open the app from the notification panel +or Apps list. -If you will be arriving in Boston early to find off-campus housing or have ac- -companying guests who need temporary housing, you can obtain lists of tem- -porary accommodation options from the Off-Campus Housing web site at -www.bc.edu/housing (click on the Off-Campus Housing link). +Settings -Checking In +From the Home screen, swipe up to access Apps and tap Settings +to access your phone’s Settings menu. -As an exchange student at Boston College, there are two offices with which -you must check in during normal business hours (Monday to Friday, 9am - -5pm) upon arrival at Boston College: the Office of International Programs -(OIP) and the Office of International Students and Scholars (OISS). +Network & Internet -Upon your arrival, please come to OIP in Hovey House and check in so that -we know that you have arrived safely. +From the Settings screen, tap Network & internet. +• Internet: Enable or disable Wi-Fi, add Wi-Fi network, view saved -You then need to check in with OISS. All international students are required -under the terms of their visas to check in officially at the Office of Interna- -tional Students and Scholars for immigration purposes. You will need your -passport with the white I-94 card attached, as well as your I-20 or DS2019 -form in order to check in. +networks, and monitor data usage. -Orientation +• Wi-Fi Calling: Enable or disable Wi-Fi Calling, adjust Wi-Fi Calling settings. +• Mobile network: Enable or disable mobile data, adjust data roaming +settings, view mobile data usage, set data limit and warning mode, etc. -The Office of International Programs, in collaboration with the Office of Inter- -national Students and Scholars, sponsors an orientation program for all new -international students which includes new exchange students. For fall and full -year students orientation takes place the week before classes begin in Sep- -tember. Spring students have orientation prior to the first week of classes in -January. The orientation will provide you not only with important information -about Boston College, but also with valuable opportunities to meet other ex- -change students, staff members, and Boston College students. +• Airplane mode: Enable or disable Airplane mode. +• Mobile plan: View your mobile plan. +• Others: Adjust settings on Hotspot & tethering, Data saver, VPN, -An important part of this program is the English Placement Test, adminis- -tered to non-native English speakers. All international students, including -exchange students, are required to take this test if their first language is not -English. +Private DNS, etc. -Please see the OISS letter in your acceptance packet for additional informa- -tion about the orientation. +23 -www.bc.edu/international +24 -10 + Connected devices +From the Settings screen, tap Connected devices > Connection +preferences. +• USB: Connect your phone with a computer using a USB cable to +transfer data, share network connectivity, or charge your phone. +• Bluetooth: Pair Bluetooth devices to launch wirelessly hear calls, - ACADEMICS +connect speakers, transfer data, and more. -Full-Time Status +• Cast: Scan for nearby devices to connect to and set them up. +• Printing: Connect your device to a printer on the same Wi-Fi -All international students in the US are required to maintain full-time student -status for the duration of their studies in the US. Undergraduates must be -enrolled in a minimum of 12 credits per semester. Graduate students must -be enrolled in a minimum of 9 credits per semester. Undergraduate exchange -students should expect to take between 12 and 15 credits per semester. -Graduate exchange students should expect to take between 9 and 12 credits -per semester. Failure to maintain full-time status will affect your legal status -in the United States, so it is very important to be enrolled in the appropriate -number of credits. +network. -Selecting and Registering for Courses +• Chromebook: Link your phone to a chromebook. +• Nearby Share: Allow the rapid short-range exchange of images, -We attempt to register you for the classes you requested on your application, -but please keep in mind that you must be flexible about what courses you will -take. We cannot guarantee that you will get all of the classes you request. If -you need to make any alterations to the courses you selected on your applica- -tion, please e-mail OIP at oipexchange@bc.edu. +videos, text, contact info, directions, YouTube videos, and other data +via Bluetooth and Wi-Fi. -When selecting courses, there are a number of factors that must be taken -into account. First, be sure to choose courses that correspond to the level at -which you were admitted to BC. Your acceptance letter states if you were ac- -cepted as an undergraduate or graduate student. Second, be sure to select -courses that correspond to your level of experience with the subject matter. -Each course listing on the BC course web site gives the course level. Levels -for all BC courses are: +• Android Auto: Control your phone’s maps, media and messaging - CORE: Fulfills university core requirements. Fairly basic courses open to +functions, with the Google Assistant while you drive. -undergraduate students from any field of study. +Apps - ONE: Undergraduate elective courses open to students from any field of +View current running applications, manage permissions, and more. +From the Settings screen, tap Apps. +• Recently opened apps: Tap an app to view detailed information. +• Default apps: Set the default app for the browser, phone, messages, -study. +etc. You can also set default apps for opening links. - TWO: Undergraduate courses designed primarily for those specializing in +• Screen time: You can tap -that field of study. +on the right side of most apps to set - THREE: Upper-level courses for undergraduate and graduate registration. -Most suited to those with background knowledge of the general subject -matter and specializing in that field of study. +daily timers. When the app timer ends, the app is paused for the rest +of the day. -11 +• Unused apps: View apps that you have not used for a long time. You -www.bc.edu/international +can un-install them to free up some space. - GRAD: For those admitted as graduate students. Requires advanced knowl- -edge of general subject matter. Please note that some of these courses are -open to undergraduate students with the permission of the professor. +• App battery usage: Set battery usage for apps. +• Special app access: View app access for special functions, such as -Undergraduate students should select courses with a level of Core, One, Two, -or Three. Graduate students should select courses with a level of Grad or -Three. +battery optimization, display over other apps, unrestricted data, usage +access, etc. -Course listings on the BC course web site (www.bc.edu/courses) look similar -to this: +25 -Index +26 -Course # + Notifications -Credits +Battery -Level +From the Settings screen, tap Notifications. +• App notifications: Control notifications from individual apps. +• Notification history: View your recent and snoozed notifications. +• Conversations: View priority and modified conversations. +• Bubbles: Enable message conversations display as bubbles, floating -Title +on top of other apps. -Instructor +• Device & app notification: Control which apps and devices can -Bldg +read notifications. -Room +• Notifications on lock screen: Select how you want notifications to -Time +be displayed on the lock screen. -9398 +• Do Not Disturb: Enable or disable Do Not Disturb. Set Do Not -BI200 01 +Disturb behavior, exceptions, schedule, etc. -0569 +• Others: You can adjust other notifications settings, like wireless -HS181 03 +emergency alerts, hide silent notifications in status bar, etc. -3 +Monitor your phone’s battery usage. +From the Settings screen, tap Battery. +• Battery Usage: View the battery percentage that each apps used -3 +from last full charge. -TWO +• Battery Saver: Enable or disable Battery Saver, set a schedule for +turning on the Battery Saver, turn the Battery Saver off when your +phone charges above 90%. -Biology +• Battery percentage: Show battery percentage in the Status bar. -DEPT +Storage -ONE +Manage internal storage on your phone and provides menu choices for +an optional SD card. +From the Settings screen, tap Storage. +• Device storage: From Storage screen, tap THIS DEVICE > This -American -Civilization +device. Tap a category to view the storage amount used by each app. -Singleton +27 -MWF -11 -TTH -9* - -In these examples, the index is the number used to register for the course -(you must include this number on your Course Registration Form). The -course number includes a two-letter abbreviation for the department (e.g. BI -= Biology, HS = History), followed by the three-digit course number. The two -numbers following the course number indicate the section. Popular courses -such as HS181 are offered in different sections to keep class sizes small. In -the example above, the course selected is section 3 of HS181. Each of the -courses in these examples carries a 3-credit load, which is the most common -load for university courses. It is your responsibility to make sure that you -have had all prerequisites required for any course that you are requesting. - -Courses typically meet several times a week. Classes that meet Monday, -Wednesday and Friday (MWF) are generally 50 minutes long, while Tuesday, -Thursday (TTH) classes are typically 75 minutes. These classes are indicated -in the course listing with an asterisk (*). - -Some courses may require co-registration for a supplementary lab or discus- -sion section and this will be stated in the course listing. Enrollment in some -courses may be restricted to majors only or may require department permis- -sion prior to registration. We are not able to pre-register you for restricted -courses, but you will be able to appeal to the department or professor after -you arrive to ask for special permission to enroll in such a class. Likewise, -undergraduates who seek to enroll in graduate-level courses will need to seek -direct permission from the instructor after arrival in Boston. It is fairly un- -usual for undergraduates to obtain approval for graduate courses and you -should not count on being able to do so. - -www.bc.edu/international +28 -12 + • SD card: From Storage screen, tap THIS DEVICE > SD CARD to - Finally, please note that many courses “close” during the early registration -period. Each course is listed in the computer with a maximum number of stu- -dents who can enroll. Once that limit is reached, the computer will not allow -any additional registration in that course. If you find that a course you need is -closed, you will have to appeal to the professor (either by e-mail prior to your -arrival or after you arrive) for an “override” to allow you into the class. This is -not uncommon and most professors are sympathetic to reasonable requests -from exchange students. However, it is not guaranteed that you will receive -an override for a closed course. - -Please look for classes offered in the semester in which you will be attending -(e.g., Fall, Spring) even if the list is for the previous year. For the most current -listing of available courses, see www.bc.edu/courses. - -Drop/Add - -The first week of classes at the beginning of each semester is called the -“Drop/Add” period. At this time you can make changes to your course sched- -ule, adding and/or dropping courses as needed. Don’t worry if you do not -have all of the courses that you need by the time classes start. You will have -this period of time to make adjustments to your course schedule, seek over- -rides as needed, etc. - -Your OIP exchange advisor can help you with most registration needs, includ- -ing add/drop. Registration matters can also be handled at Student Services in -Lyons Hall or on the computer via AGORA. You will receive more specific in- -formation about these resources upon arrival. In the meantime, refer all -course requests to OIP. - -Faculty Advisors - -You may request a faculty advisor for -assistance with academic planning for -your semester or year at BC. We will -make advisor placements based upon -your field of study. If your advisor at -home already knows a particular pro- -fessor at BC, or if you have already -made contact with a professor here -and you would like him/her to be your -advisor, please let us know. If you re- -quest a faculty advisor, you will be -required to meet with them by ap- -pointment during the first week of -classes and throughout the semester -as necessary. +Display -13 +view files stored on your SD card. Tap on the top right corner of the +screen to rename, eject or format the SD card. -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -www.bc.edu/international - - Tutoring - -The Connors Family Learning Center offers free tutoring for more than 60 -undergraduate and some graduate courses, including writing and English -as a Second Language. Intensive, small-group supplemental instruction -sessions are available in several courses, and tutoring and assistance in -improving academic writing. For more information about services provided -by the Learning Center, go to www.bc.edu/libraries/centers/connors/. - -English Classes for Non-Native Speakers - -If your native language is not English you will be required to take a place- -ment examination during the Orientation in the fall or during the first week -of classes in January. If necessary, students will be recommended for a par- -ticular course based upon the results of this examination. The Program in -English for Foreign Students is based in the English Department and offers -several courses in college-level English for non-native speakers. Recent -course offerings include: - - EN 118 Essential of English Composition: extensive practice in writing - -and editing academic prose - - EN119 The Craft of Writing: practice writing academic essays and expo- -sure to English rhetoric, with emphasis on written analysis and the logi- -cal support of ideas - -In order to enroll in one of these courses, you must have taken the place- -ment test. - -Transcripts of Academic Work - -At the end of your time at Boston College, we will send an official transcript -of the academic work you completed to your home university. Before we -can release your BC transcript to your home university, however, you must -clear any outstanding debts to Boston College that you may have. Be sure -that all long-distance phone calls, on-campus housing fees, insurance, -meals, etc. have been paid for. - -Your BC student account must show a zero balance before you leave Bos- -ton or we will not be able to release your transcript. - -www.bc.edu/international +• Storage manager: Storage manager removes backed up photos and -14 +videos from your device. - ADMINISTRATIVE ISSUES - -Student Identification Cards - -If you submitted the required passport-size photographs with your Exchange -Student Application you will receive your student ID card during orientation. -If you did not send us photos, you will be able to have your ID card made at -Student Services in Lyons Hall following orientation. - -Your Boston College student identification card (“Eagle ID”) is a very impor- -tant item for you to have. In addition to identifying you as a BC student, the -card enables you to check books out from the library, to make purchases at -the BC bookstore, to access your student record and account from “U-VIEW” -computers around campus, and to make use of the BC Flynn Recreation -Complex (fee required), among other services. Be sure to keep your Eagle ID -with you at all times and do not lose it. A replacement fee will be charged by -Student Services for lost cards. - -Medical Insurance - -As explained in the Things to Know Before You Leave Home section, the -Commonwealth of Massachusetts requires all students enrolled in a Massa- -chusetts college or university to be covered by a US-based medical insurance -plan. Therefore, you have been automatically enrolled in the Boston College -medical insurance policy which provides adequate coverage during your stay -at BC, including travel within and outside the US during your period of cover- -age. Your student account was billed upon registration at BC. As an exchange -student it is your responsibility to pay for your insurance, which can be done -at Student Services. Full year students will be automatically re-enrolled in the -insurance plan for their second semester. You can find more information on -the BC medical insurance coverage on the BC web site at www.bc.edu/ -offices/uhs/services/insurance. - -If you do not want to be covered by BC medical insurance, you must be cov- -ered by insurance that is US-based or has a US contact office. You must have -proof of this insurance. If your insurance is comparable, you will be allowed -to waive the BC insurance and your student account will be credited accord- -ingly. Please note that it is not possible to have your medical insurance evalu- -ated before you arrive at Boston College. +Sound & vibration +From the Settings screen, tap Sound & vibration. +• Media/Call/Ring/Notification/Alarm volume: Drag the slides to -15 +adjust the media, call, ring & notification, and alarm volumes. -www.bc.edu/international - - Health Services - -The Health Services Center at -Boston College offers a range -of services to BC students. -Staffed by first-class doctors, -nurse practitioners, and other -medical staff, the Health Ser- -vices Center is equipped to -handle health emergencies, -minor illnesses and questions -about healthy living. While the -general health services offered -are comprehensive, if you re- -quire specialty care you can -be referred to a qualified spe- -cialist. - -All undergraduate exchange -students who live on-campus -are required to pay the Health -Services fee. Exchange stu- -dents who live off-campus are -also entitled to this service for the same fee. Off-campus students can ei- -ther pay in advance, or the first time the service is used. We strongly recom- -mend that you utilize the Health Services if necessary. Services covered by -the fee are detailed on the Boston College Health Services website at -www.bc.edu/offices/uhs. - -Please note that Health Services fee is not the same as medical insurance. -You must still have adequate medical insurance coverage, especially for -such things as hospitalization, emergency room visits, diagnostic testing, -x-rays, etc., which are not covered by the Health Services fee. - -University Counseling Services - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sometimes students find adjusting to a new culture and new academic en- -vironment very challenging. The Counseling Services Office provides caring, -professional support and assistance to full-time students who request it. -Selected staff members in Counseling Services have extensive experience -with cross-cultural issues and are especially well-suited to help students -who are experiencing culture shock. It is very common in the United States -for students to seek the help of counseling professionals when dealing with -challenging situations. In fact, thousands of students take advantage of the -services offered by Counseling Services for a variety of issues each year. +• Do Not Disturb: Manage Do Not Disturb settings. +• Phone ringtone: Select a ringtone for incoming phone calls. +• Media: Enable Media to hide player when the media session has ended. +• Others: You can adjust other sound and vibration settings, like -16 +vibration & haptics, default notification and alarm sounds, enable +tones or vibration on other scenarios, etc. -www.bc.edu/international - - The services offered by Counseling Services (at no additional cost) include -short-term individual counseling and psychotherapy as well as a limited -number of counseling groups. If you wish to speak with a counselor you -can make an appointment by calling 617-552-3310 or by visiting the Counsel- -ing Office in Gasson Hall 108. - -Also, please remember that OIP staff is always available and willing to help -if you have any questions, either academic or personal. Most of our staff -members have studied and/or lived outside the US and know firsthand the -challenges of adapting to a new country and culture. - -Students with Disabilities - -Boston College provides a range of services and accommodations for stu- -dents with physical or learning disabilities. If you require additional assis- -tance because of a disability, feel free to share that information with your -OIP exchange advisor. You may also contact the Boston College Coordina- -tor of Special Needs Services in Gasson Hall 108, telephone: 617-552-3310. - -E-mail - -Every student enrolled at -Boston College is eligible -to receive an e-mail ac- -count. You may obtain -your account after arriv- -ing at BC by presenting -your BC ID card at the -Campus Technology Re- -source Center (CTRC) -located on the second -floor of the O’Neill Li- -brary. The use of your BC -email account is required -once you have arrived on- -campus. If you have an- -other address that you -prefer to use, set up your BC account (by using AGORA) to forward all mes- -sages to the alternate account. OIP and other BC offices will send all infor- -mation during your time at Boston College to your BC email address. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +From the Settings screen, tap Display. +• Brightness level: Drag the slide to adjust the screen brightness -17 +level. -www.bc.edu/international - - LIFE BEYOND ACADEMICS - -Religion - -As a Catholic and Jesuit institution, Boston College is sensitive to the spiri- -tual needs of all members of the BC community, whatever their respective -faiths or philosophical traditions. Catholic Masses are offered throughout -the week in several locations on campus and are open to all BC students, -faculty and staff. The Campus Ministry Office offers spiritual counseling -and guidance to students, as well as volunteer service opportunities in Bos- -ton and other parts of the world. Campus Ministry also provides informa- -tion and support for people of all religious and spiritual traditions by main- -taining a list of churches, temples and synagogues in the Boston College -area. Various student clubs, including the Asian Christian Fellowship, Jew- -ish Students’ Association, and Muslim Students’ Association, provide a -sense of community and support for students of various spiritual tradi- -tions. - -Activities/Opportunities - -There is a wide variety of clubs and -activities to meet any student’s in- -terests, including several intercul- -tural clubs focusing on individual -countries or regions. An Activities -Fair early in the Fall Semester will -give you an opportunity to learn -about many of these opportunities -and speak to some of the students -involved. For students on exchange -during the Spring semester, the Of- -fice of the Dean for Student Devel- -opment has a list of all BC clubs -and activities (www.bc.edu/a-z/ -clubs/). Don’t pass up this opportu- -nity to learn about American culture -outside the classroom. Students -who get involved in activities at Boston College consistently enjoy their ex- -perience here more than those who do not get involved. For this reason, -we strongly encourage you to join at least one club while you are here. - -s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C +• Adaptive brightness: Enable adaptive brightness to automatically -18 +adjust the screen brightness basing on the lighting conditions. -www.bc.edu/international - - s -n -o -i -t -a -c -i -n -u -m -m -o -C -g -n -i -t -e -k -r -a -M - -f -o -e -c -i -f -f - -O -C -B -y -s -e -t -r -u -o -C - -Sports in American University Culture - -Many international students are surprised at the level of importance placed -on intercollegiate sports in the United States. American football, particu- -larly, draws thousands of screaming fans every Saturday during football -season, with students cheering for the home team. This is very much a part -of American college culture and, even if you are not a sports fan, it is some- -thing you should experience during your time as a student in the United -States. You may not know anything about the intricate rules of American -sports like football, hockey and basketball. However, the experience of sit- -ting in a crowded stadium, cheering, listening to the band, watching the -crowd, and even trying to follow the game, is a true American experience -and should not be missed. - -Recreation Complex - -The Flynn Recreation Complex provides Boston College students with a -broad array of sports and physical fitness opportunities, from aerobics to -weight lifting and swimming. As an exchange student you have the option -of joining the “RecPlex” for a semester or academic year (membership fees -apply). If you choose not to join the “RecPlex” you may still use the facilities -by presenting your BC ID card and paying a small fee per visit. +• Lock screen: Set what to show on lock screen. +• Screen timeout: Your screen will automatically go to sleep after the -19 +set amount of time. -www.bc.edu/international +• Dark theme: Turn on dark theme. +• Display size and text: Adjust font size and display size. +• Night Light: Set a Night Light schedule and intensity. +• Others: You can adjust other display settings, like enable auto-rotate +screen and life to wake features, adjust display color and screen saver, +etc. - CONTACT INFORMATION +29 -Office of International Programs +30 -Location: -Mailing Address: + Accessibility -Hovey House, 258 Hammond Street -Hovey House, 140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +Security & privacy -Phone: 617/552-3827 Fax: 617/552-0647 -Email: oipexchange@bc.edu -Web: www.bc.edu/international -Director: Bernd Widdig -Email: widdig@bc.edu -Assistant Director for Exchange Students: Esther Messing -Email: messines@bc.edu +Manage accessibility-related apps to make the device easier to use for +those with certain physical disabilities. +From the Settings screen, tap Accessibility. +• Real-Time text call: Manage RTT call options. +• Display: Adjust the font size and display size, enable or disable dark -Office of International Students and Scholars +theme, enable screen magnification, etc. -Location: -Mailing Address: +• Interaction controls: Enable or disable useful interaction functions, -21 Campanella Way - Suite 249 -140 Commonwealth Avenue -Chestnut Hill, MA 02467-3926 +such as timing control, system controls, etc. -Phone: 617/552-8005 Fax: 617/552-2190 -Web: www.bc.edu/offices/odsd/intercultural -Director: Adrienne Nussbaum -Email: adrienne.nussbaum@bc.edu +• Captions: Enable or disable captions. You can also set caption display -Office of Residential Life +language, text size, and caption style. -Contact: -Web: +• Audio: Pair your hearing aids, enable mono audio, or adjust the audio -On-Campus Housing, Off-Campus Housing Advice -www.bc.edu/offices/reslife +balance from the speaker. -Office of Student Services +• Accessibility shortcuts: Mange the settings on accessibility button. +• Text-to-speech output: Set the text-to-speech engine, language, -Contact: -Web: +speech rate, etc. -Billing, Medical Insurance, ID, Course Registration -www.bc.edu/offices/stserv +From the Settings screen, tap Security & privacy. +• App security: Check your device and prevent/warn you of potential -University Health Services +harm. -Contact: -Web: +• Device lock: Set your screen lock method, pattern, PIN, password, etc. +• Account security: Add a Google Account to secure your device. +• Device finders: Enable Find My Device to locate your device -Immunization, Health and Wellness Issues -www.bc.edu/offices/uhs +remotely, keeping your data safe in the event that your phone is lost. +• System & updates: Check for new software updates for your device. +• Privacy: Manage a batch of privacy settings. +• Face unlock: Unlock your phone by looking at it. +• More security & privacy: Adjust more security and privacy settings. -University Counseling Services +Unlock your device automatically in certain conditions. -Contact: -Web: +31 -Personal Counseling -www.bc.edu/offices/counseling +32 -20 + Location + +From the Settings screen, tap Location. +• Recent access: View a list of apps that have recently requested your + +location. + +• App location permissions: View which apps have permission to + +access location services. + +• Location services: Manage settings on other location services, such + +as emergency location service, Google location accuracy, etc. + +Safety & emergency + +Fulfill your medical and emergency information, view emergency alerts, +and manage settings on emergency SOS and location service. +From the Settings screen, tap Safety & emergency. +• Emergency information: Add medical information and emergency +contacts to help first responders in an emergency. Anyone can read +this information from your lock screen and dial your emergency +contacts without unlocking your phone.. + +• Emergency SOS: If you have turned the Emergency SOS function +on, when you have an emergency and need emergency services, +press the Power/Lock key 5 or more times quickly. + +• Emergency Location Service: If you have turned Emergency +Location Services (ELS) on, when you call or text an emergency +number, your device will use ELS to automatically send your location +to emergency responders. + +• Earthquake alerts: Get an alert with the initial estimated magnitude + +and distance from your location. + +• Wireless emergency alerts: Turn wireless emergency alerts on + +or off. You can also view emergency alert history and manage alert +preferences. + +Passwords & accounts + +Set up and manage all of your accounts, including your Google Account +and other social networking accounts, through the Accounts settings +menu. From the Settings screen, tap Passwords & accounts. + +33 + +34 + + • Passwords: See, change, or remove passwords you saved in your + +• Reduce interruptions: Enable or disable the notifications of apps + +Google Account. + +and Do Not Disturb mode. + +• Autofill service: Autofill your information when needed from your + +Google Account. + +• Add account: Set up your account. +• Automatically sync app data: Activate automatic data syncing. + +Digital Wellbeing & parental controls + +Statistics screen usage time, app timers, bedtime schedules, manage +notifications, etc. +From the Settings screen, tap Digital Wellbeing & parental controls. +• Your Digital Wellbeing tools: Use app timers and other tools to + +keep track of screen time and unplug more easily. + +• Ways to disconnect: View your screen time and set timers for your +apps, set a bedtime schedule to disconnect from your phone and +avoid interruptions while you sleep, pause distracting apps when you +need time to focus. + +• Show icon in the app list: Enable icons in the app list. +• Parental controls: Add content restrictions and set other limits to + +help your child balance their screen time. + +Google + +Configure your device’s Google settings. Available options depend on +your Google Account. From the Settings screen, tap Google. + +System + +Manage detailed settings for your phone, including language and input +settings, battery and storage, date and time, etc. +From the Settings screen, tap System. +• Languages: Set the language for the phone’s menus and keyboards, +set the language for each app, set units and number preferences, etc. + +• Keyboard: Select and configure keyboard settings + +35 + +36 + + • Gestures: Enable useful gestures, such as Quickly open camera, +Navigation mode, Power & hold power menu and Prevent ringing. +• Date & time: Enable or disable network-provided time, automatic + +time zone, 24-hour format, etc. + +• Backup: Back up your phone’s data and settings to Google servers + +before resetting your phone to the original factory settings. + +• System update: Check for new software updates for your device. +• Multiple users: Share this device with other people by creating + +additional users. + +• Reset options: View different reset options and reset your device. + +About phone + +Access important phone information, search for and apply updates, and +view legal and safety information. +From the Settings screen, tap About phone. + +Troubleshooting + +Below is a list of common problems that may occur when using your +mobile phone. Try these easy troubleshooting steps to resolve any type +of issues you may encounter. +The phone does not power on +• Press and hold the power key until the startup sequence appears on + +the display. + +• Check the battery level and ensure there is sufficient charge. +The phone does not respond after several minutes +• Remove and insert the battery – If the phone has an internal battery, + +please press and hold the power button. + +The phone powers off by itself +• Check if the screen lock of the phone is enabled. +• Check if the phone needs to be charged. + +37 + +38 + + The battery will not charge +• Check if the battery is already fully charged. +• Make sure that the battery is inserted correctly. +• Ensure that the battery being used is a BLU approved battery. +• Double check if the charger connector is properly connected to the + +phone. + +Note: Connect charger only after the battery has been inserted. + +The phone displays a SIM Error +• Make sure the SIM is properly inserted and in its intended slot. +• Make sure that the SIM card is not damaged or scratched. +The PIN code was entered incorrectly three times +• The SIM card will lock, and you will need to enter the PUK code (PIN +Unlock Code). Please contact your local service provider to obtain the +PUK. + +The standby time is too short +• Please keep in mind that standby time relates directly to phone usage + +I cannot make calls +• Make sure that the number you dialed is correct and the dial key has + +(Internet, Text, Voice Calls, Maps, etc.). + +been pressed. + +• Batteries typically have to be replaced after a year of use. +The phone cannot connect to the network (Voice or Data) +• Verify if you have active service with your service provider. +• Contact your local network service provider and verify if you are within + +its network coverage area. + +• Use the full national area code when dialing out of your area or when + +abroad, use country-specific international access codes. + +I cannot receive calls +• Make sure that your phone is connected to a network. + +39 + +40 + + The caller ID does not display on incoming calls +• Contact your local network service provider and verify you have the +caller ID feature enabled on your account or verify if the caller may +have blocked his number from appearing. + +The caller cannot leave a voicemail message +• Contact your local network service provider and verify if the voicemail + +service is active on your account or check if the network is busy. -www.bc.edu/international +The phone does not connect to the internet +• Ensure with your service provider that your account has the - ESTIMATED EXPENSES -2010-2011 +appropriate internet/data plan enabled. - Average Cost of Apartment/Room Rental +• Ensure internet APN settings are configured (Carrier dependent). +Text Messages cannot be Sent / Received +• Ensure text messaging is enabled with your service provider. +• Ensure APN settings are configured (Carrier dependent). -Studio Apartment: $1000 - $1300 per month * -One Bedroom Apartment: $1200 - $1550 per month * -Two Bedroom Apartment: $1400 - $2000 per month * -Three-Four Bedroom Apartment: $2000 - $3800 per month * -Private Room in Local Home: $500 - $900 per month * +Regulatory Information -*Please note: Boston rental agents often charge one half or one full month's -rent as a commission fee (also known as a "Finder's Fee") +FCC regulations -On-Campus Housing +This mobile phone complies with part 15 of the FCC Rules. Operation is +subject to the following two conditions: (1) This device may not cause +harmful interference, and (2) this device must accept any interference +received, including interference that may cause undesired operation. +This mobile phone has been tested and found to comply with the limits +for a Class B digital device, pursuant to Part 15 of the FCC Rules. These +limits are designed to provide reasonable protection against harmful +interference in a residential installation. +This equipment generates, uses and can radiate radio frequency energy +and, if not installed and used in accordance with the instructions, may +cause harmful interference to radio communications. + +41 + +42 + + However, there is no guarantee that interference will not occur in a +particular installation If this equipment does cause harmful interference +to radio or television reception, which can be determined by turning +the equipment off and on, the user is encouraged to try to correct the +interference by one or more of the following measures: +• Reorient or relocate the receiving antenna. +• Increase the separation between the equipment and receiver. +• Connect the equipment into an outlet on a circuit different from that + +to which the receiver is connected. + +• Consult the dealer or an experienced radio/TV technician for help. +Changes or modifications not expressly approved by the party +responsible for compliance could void the user’s authority to operate the +equipment. +The antenna(s) used for this transmitter must not be co-located or +operating in conjunction with any other antenna or transmitter. + +RF Exposure Information (SAR) + +This mobile phone meets the government’s requirements for exposure to +radio waves. +This phone is designed and manufactured not to exceed the emission +limits for exposure to radio frequency (RF) energy set by the Federal +Communications Commission of the U.S. Government. +The exposure standard for wireless mobile phones employs a unit of +measurement known as the Specific Absorption Rate, or SAR. The SAR +limit set by the FCC is 1.6 W/kg. *Tests for SAR are conducted using +standard operating positions accepted by the FCC with the phone +transmitting at its highest certified power level in all tested frequency +bands. +Although the SAR is determined at the highest certified power level, +the actual SAR level of the phone while operating can be well below +the maximum value. This is because the phone is designed to operate +at multiple power levels so as to use only the poser required to reach +the network. In general, the closer you are to a wireless base station +antenna, the lower the power output. + +43 + +44 + + The highest SAR value for the model phone as reported to the FCC when +tested for use at the ear is 1.29 W/kg and when worn on the body, as +described in this user guide, is 0.89 W/kg (Body-worn measurements +differ among phone models, depending upon available accessories and +FCC requirements.). +While there may be differences between the SAR levels of various +phones and at various positions, they all meet the government +requirement. +The FCC has granted an Equipment Authorization for this model phone +with all reported SAR levels evaluated as in compliance with the FCC RF +exposure guidelines. SAR information on this model phone is on file with +the FCC and can be found under the Display Grant section of www.fcc. +gov/oet/ea/fccid after searching on FCC ID: YHLBLUB160V. +For body worn operation, this phone has been tested and meets the FCC +RF exposure guidelines for use with an accessory that contains no metal +and the positions the handset with a minimum of 1.5 cm from the body. + +Use of other accessories may not ensure compliance with FCC RF +exposure guidelines. If you do not use a body-worn accessory and are +not holding the phone at the ear, position the handset a minimum of 1.5 +cm from your body when the phone is switched on. + +Safety precautions for power supply unit + +Use the correct external power source +A product should be operated only from the type of power source +indicated on the electrical ratings label. If you are not sure of the type +of power source required, consult your authorized service provider or +local power company. For a product that operates from battery power or +other sources, refer to the operating instructions that are included with +the product. +This product should be operated only with the following designated +power supply unit(s). +Travel charger: Input: 100-240V, 50/60Hz, 0.35A. Output: 5V, 2000mA. + +45 + +46 + + Handle battery packs carefully +This product contains a Lithium-ion battery. There is a risk of fire and +burns if the battery pack is handled improperly. Do not attempt to open +or service the battery pack. Do not disassemble, crush, puncture, short +circuit the external contacts or circuits, dispose of in fire or water, or +expose a battery pack to temperatures higher than 140°F (60°C). The +operating temperature for the phone is 14°F (-10°C) to 113°F (45°C). +Warning: Battery is non-removable. Do not attempt to remove +the battery for danger of explosion. +To reduce risk of fire or burns, do not disassemble, crush, puncture, +short circuit the external contacts, expose to temperature above 113°F +(45°C), or dispose of in fire or water. Do not attempt to replace the +non-removable battery. The phone and battery as a unibody device +must be disposed of in accordance with locally applicable environmental +regulations or guides supplied with your product. + +47 + +Take extra precautions +• Do not disassemble or open crush, bend or deform, puncture or + +shred. + +• Do not modify or remanufacture, attempt to insert foreign objects into +the battery, immerse or expose to water or other liquids, expose to +fire, explosion, or other hazards. + +• Only use the battery with a charging system that has been qualified + +with the system per CTIA Certification Requirement for Battery System +Compliance to IEEE1725. Use of an unqualified battery or charger may +present a risk of fire, explosion, leakage, or other hazards. + +• Replace the battery only with another battery that has been qualified + +with the system per this standard, IEEE-Std-1725. Use of an +unqualified battery may present a risk of fire, explosion, leakage or +other hazards. Only authorized service providers shall replace battery. +(If the battery is non-user replaceable). + +• Promptly dispose of used batteries in accordance with local + +regulations. + +• Improper battery use may result in a fire, explosion, or other hazards. +48 + + • Do not replace an embedded battery pack. Improper replacement may +present a risk of fire, explosion, leakage or other hazards. Contact the +manufacturer for replacement instructions. + +• The phone identifies the battery by the Battery ID Pin. +• If the battery leaks: + +• Do not allow the leaking fluid to come in contact with skin or + +clothing. If already in contact, flush the affected area immediately +with clean water and seek medical advice. + +• Do not allow the leaking fluid to come in contact with eyes. +If already in contact, DO NOT rub; rinse with clean water +immediately and seek medical advice. + +• Take extra precautions to keep a leaking battery away from fire as + +there is a danger of ignition or explosion. + +Hearing Aid Compatibility (HAC) + +This device is hearing aid-compatible as determined by ANSI C63.19- +2019 and waiver DA-23-914. The lowest conversational gain is 10.32 +dB with a hearing-aid, and 13.93 dB without hearing-aid, using telecoil +coupling capability and Bluetooth connectivity. +The 2019 ANSI C63.19 standard does not use this rating system nor +do any test standards exist for any wireless technologies that transmit +above 6 GHz. Instead, this standard uses the conversational gains for +the device’s hearing aid compatible capabilities. +This device has been tested and certified for use with hearing aids for +some of the wireless technologies that it uses. However, there may be +some newer wireless technologies used in this phone that have not been +tested yet for use with hearing aids. It is important to try the different +features of this phone thoroughly and in different locations, using your +hearing aid or cochlear implant, to determine if you hear any interfering +noise. Consult your service provider or the manufacturer of this phone +for information on hearing aid compatibility. If you have questions about +return or exchange policies, consult your service provider or retailer. -Shared Room: $1100 per person per month +49 -Meals and Books +50 -Campus Meal Plan: $2300 per semester for full meal plan -No Meal Plan: $1500 per semester -Books: $600 per semester + We consent to the following conditions of the waiver: +• Pass Conversational Gain, Receive Distortion and Noise and Receive -Other Expenses +Acoustic Frequency Response testing using one narrowband and one +wideband codec, limited to one bit rate. -Medical Insurance: $871 (Fall semester); $1027 (Spring semester) ** -University Health Services: $210 per semester ** -Flynn Recreation Complex: $140 per semester +• Pass Conversational Gain at both 2N and 8N on all available -**The BC Medical Insurance plan is mandatory unless you can demonstrate -coverage by an insurer with a U.S. office. The University Health Services fee is -required for on-campus students. +narrowband and wideband codecs and air interface combinations. +Please see the HAC Volume Control test report for further details and +supporting data. -Personal Expenses: These will vary significantly according to the individ- -ual, but students should budget approximately $2000 per semester. +The information provided in this user manual is subject to change without notice +and BLU Products reserves the right to modify the manual at any given time. BLU +Products has made every effort to ensure that this user manual is accurate and +disclaims responsibility and liability for any inaccuracies or omissions. BLU Products +makes no commitment to keep current the user manuals, however, does reserve +the right to make improvements to the user manual at any given time without +notice. -Transportation from Logan Airport to BC +Google, Google Play, Google Maps, YouTube, Gmail and other marks are +trademarks of Google LLC. -Taxi: approximately $60 -Private Airport Shuttle: approximately $50 -Subway (“T”) Ride : $2.00 cash payment or $1.70 using a “Charlie Card” +51 diff --git a/q121/random_k4/random_2.md b/q121/random_k4/random_2.md index 4b9bd14ed63d1882c240fd2561e1be123d06b2a6..2cd064b36cf6fc7716b108dfdc7b4efff29d6268 100644 --- a/q121/random_k4/random_2.md +++ b/q121/random_k4/random_2.md @@ -1,1483 +1,629 @@ -USER MANUAL +N0 5909 -8-IN-1 ONE-TOUCH -AUTOMATIC ESPRESSO MAKER +HEADPHONE OWNER’S MANUAL -GCM4230 + TABLE OF CONTENTS -1 - -USER MANUAL ESPRESSO MAKER GCM4230 - CONTENTS - -Important Safeguards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 - -Know Your Espresso Maker . . . . . . . . . . . . . . . . . . . . . . . . . . 6 -Control Panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 - -Using Your Espresso Maker - -Before First Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 - -Setting Up and Priming the Espresso Maker . . . . . . . . . . . . . . . 8 - -Filling the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 - -Placing the Portafilter . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 - -Mug Size Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 - -Drink Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 +TABLE OF CONTENTS -Making an Espresso . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 +Overview +What's in the Box +Product Overview -Making a Cappuccino . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 +Adjustability, Fit & Comfort -Making a Latte . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 +Functions -Making an Americano . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 +Charge -Brewing in Manual Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 15 +Connections +Pairing +Passive +USB Digital Direct -Frothing Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 +APP -Cleaning and Maintenance +Maintenance / Care -Cleaning the Portafilter and Brew Head . . . . . . . . . . . . . . . . . 17 +Changing / Removing Earpads -Cleaning the Water Reservoir . . . . . . . . . . . . . . . . . . . . . . . 17 +Accessories -Cleaning the Milk Reservoir and Frothing Tube . . . . . . . . . . . . . 17 +Troubleshooting -Descaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 +Ear Health -Warranty and Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 +Specifications -Model: GCM4230 | © 2023 Gourmia | 2.0 -The Steelstone Group Brooklyn, NY | Gourmia .com - -All rights reserved. No portion of this manual may be reproduced by any means whatsoever without written -permission from The Steelstone Group, except for the inclusion of quotations in a media review. Although -every precaution has been taken in the preparation of this manual, The Steelstone Group assumes no -responsibility for errors or omissions. Nor is any liability assumed for damages resulting from the use of -the information contained herein. Although every effort has been made to ensure that the information -contained in this manual is complete and accurate, The Steelstone Group is not engaged in rendering -professional advice or services to the individual consumer; further, The Steelstone Group shall not be held -responsible for any loss or damage allegedly arising from any information or suggestions in this manual. - - Congratulations on the -purchase of your Espresso -Maker from Gourmia! - -Here at Gourmia, our one and only goal is to make sure -nutritious and delicious meals are always a practical -possibility no matter how fast-paced your life is . Whether -you’re in need of a hearty breakfast, a mid-day snack, or a -satisfying supper, our dependable, easy-to-use appliances -are designed to work for you while you work your way -through your busy day . - -This manual provides an overview and instructions on -all the espresso maker’s functions and features, along -with directions for assembling, operating, cleaning, -and maintaining your appliance . Please read all safety -instructions before getting started . Keep this manual handy -for future use and review . - -Our team is available to help with any and every question -you might have about your new espresso maker . We’re -excited for you and want you to get the most out of your -purchase—so whether you need assistance with setup or -guidance on which features to enable for which drinks, -we’re happy to walk you through it, step by step . - -Happy and easy cooking! - -We’d love to hear about the delicious drinks you’re creating! -Share your thoughts, recipes, and photos with us. - -@gourmia +Warranty 3 +3 +4 - S -d -r -a -u -G -E -f -a -S - -t -n -a -t -r -o -p -M - -I - -S A V E T H E S E I N S T R U C T I O N S - -IMPORTANT SAFEGUARDS - -Before operating this espresso maker, read and follow all these -basic safety precautions: - -1. The appliance can be used - -7. do not operate the appliance - -by children aged 8 years and -above and persons with reduced -physical, sensory or mental -capabilities or lack of experience -and knowledge if they have been -given supervision or instruction -concerning use of the appliance -in a safe way and understand the -hazards involved . Children should -not play with the appliance . - -Cleaning and user maintenance -shall not be done by children -unless they are 8 or older and -supervised . - -2. Keep the appliance and its cord - -out of reach of children less than 8 -years old . - -3. Make sure your outlet voltage - -corresponds to the voltage stated -on the rating label of the espresso -maker . - -4. To protect against fire, electric - -shock, or injury to persons do not -immerse cord, plug, or appliance -in water or other liquid . - -5. Close supervision is necessary - -when this appliance is used by or -near children . - -6. Unplug from outlet when not in +5 -use and before cleaning . Allow the -appliance to cool before putting -on or taking off parts, and before -cleaning the appliance . +6 -with a damaged cord or plug or if -the appliance malfunctions or has -been damaged in any manner . -Return appliance to an authorized -service facility for examination, -repair, or adjustment . +8 -8. The use of accessories not +9 +9 +9 +10 -recommended by the appliance -manufacturer may result in fire, -electric shock, or injury to persons . +11 -9. do not touch hot surfaces; use +12 -handles or knobs . +12 -10. do not operate this appliance -outdoors . do not use for -commercial purposes . +13 -11. do not let the cord hang over +14 -the edge of a table or counter or -touch any hot surfaces . +15 -12. Use on a hard, flat, level, heat- +16 -resistant surface only . +17 -13. Always be sure to secure the +1 -water reservoir tightly back inside -the unit with the lid closed and -filled with water between the -minimum and maximum mark -before turning the appliance on . +N05909 HEADPHONE / OWNER’S MANUAL 2 -14. do not remove the portafilter while -brewing coffee . Exercise caution -when removing the portafilter after -brewing since the metal parts -will be hot . Make sure to hold the -portafilter by the handle and use - the filter retention clip to dispose -the grounds . +N05909 HEAPHONE / OWNER’S MANUAL -4 + OVERVIEW -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - 15. Extreme caution must be -exercised when moving -appliances with hot liquids . +Congratulations on your selection of the Mark Levinson® +No 5909 Premium Wireless ANC Headphone. For over 50 +years, discerning individuals have chosen Mark Levinson +products for their luxurious quality and revealing sound. +Robust materials, lavish finishes, bold geometry, and +outstanding performance are celebrated attributes of the +Mark Levinson brand. The No 5909 headphone is made with +the finest materials, technology, and craftsmanship in the +world and will deliver outstanding performance for many +years to come. -16. Only remove the water reservoir -when the appliance is turned off . +For optimum performance, please read this guide and +follow the unpacking and setup instructions carefully. -17. This appliance is not intended -to be operated by means of an -external timer or separate remote- -control system . +WHAT'S IN THE BOX -appliance and allow it to cool -down . The blockage may be -caused by the coffee being too -finely ground . +• Mark Levinson No 5909 Premium Wireless ANC Headphones -20. do not place the appliance on or -near a hot gas or electric burner, -or in a heated oven . +• Travel Storage Case -21. do not use the appliance for other +• 1.25m USB-C to USB-C Charging Cable -than its intended use . +• 1.25m USB-C to 3.5mm Proprietary Audio Cable -18. do not remove the portafilter +• 4m USB-C to 3.5mm Proprietary Audio Cable -at any time during the brewing -process . +• 3.5mm to 6.3mm Adaptor -19. If coffee does not drip, it means -the filter is blocked . Turn off the +• USB-C to USB-A Adaptor -22. do not leave the appliance -unattended when turned on . +• Airplane Adaptor -23. This appliance should only be -used in households or similar -environments . +• Microfiber Polishing Cloth -S -d -r -a -u -G -E -f -a -S +• Quick Start Guide -t -n -a -t -r -o -p -M +OVERVIEW -I +N05909 HEADPHONE / OWNER’S MANUAL -Cord and plug Safety -1. A short power-supply cord is provided to reduce risk of becoming entangled +3 -in or tripping over a longer cord . + OVERVIEW -2. Extension cords may be used if care is exercised in their use . +PRODUCT OVERVIEW -3. +1. Premium Leather Headband with Memory Foam Interior -If an extension cord is used: +2. Replaceable Memory Foam and Memory Foam Ear Pads -• The marked electrical rating of the extension cord should be at least equal +3. Forged and Machined Aluminum Components throughout -to the electrical rating of the appliance . +4. USB-C Input for Passive Audio Listenting & Charging -• The longer cord should be arranged so that it will not drape over the +5. Adjustment Arm -countertop or tabletop where it can be pulled on by children or tripped -over . +6. 8 Microphones (4 talk microphones and 4 ANC microphones) -Electrical polarized plug -This unit has a polarized plug (one blade is wider than the other) . -To reduce the risk of electric shock, the plug is intended to be -used by inserting it into the electrical outlet only one way . If the -plug does not fit securely into the outlet, reverse the plug and -re-insert . If it still does not fit, contact a qualified electrician for -assistance . Do not modify the plug in any way . +4 -poWEr +1 -AC120V~ 60Hz 1,300W +2 5 -USER MANUAL ESPRESSO MAKER GCM4230 - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K - -KNOW YOUR -ESPRESSO MAKER - -Water Reservoir -Cover - -Removable -Water Reservoir - -Control Panel - -Brew Head - -Power Switch - -On the back-right -side of the unit - -Removable Drip Tray +3 6 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Milk Reservoir Lid +4 -Froth Strength Control +N05909 HEAPHONE / OWNER’S MANUAL ADJUSTABILITY, FIT & COMFORT -Froth Dispensing Tube Lever +ADJUSTABILITY, FIT & COMFORT -Froth Dispensing Tube +Adjust to find optimal fit -Minimum Strength +Rotates for comfort around neck -Clean / Americano Position +5 -Maximum Strength +N05909 HEADPHONE / OWNER’S MANUAL FUNCTIONS -Removable Milk Reservoir +FUNCTIONS -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -W -o -n -K +Active Noise Cancelling / Awareness (ANC) Button -Measuring Scoop -and Tamper +Bluetooth Pairing/ Power ON/OFF -Portafilter Handle +Pairing Indicator -Portafilter Retention Clip +Active Noise Cancelling / Awareness (ANC) Button -Portafilter +Pairing Indicator -Double-Shot -Filter +White light will pulse when headphones are in pairing +mode -Single-Shot +A solid white light indicates you are connected -Control panel -All icons will light up when the milk reservoir is correctly inserted into the unit . +x1: ANC ON +(Delfault selectable via app; HIGH, ADAPTIVE, LOW +modes) -ESPRESSO -Icon +x2: Awareness ON +(Default selectable via app; Voice Pass or Ambient +modes) -LATTE -Icon +x3: Passive / ANC OFF -MANUAL -ESPRESSO Icon +Clicking the button will cycle through the different +modes. -CAPPUCCINO -Icon +6 -AMERICANO -Icon +N05909 HEAPHONE / OWNER’S MANUAL -FROTH -Icon + FUNCTIONS -CLEAN -Icon +Vol + -7 +Play/Pause / Answer/End Call -USER MANUAL ESPRESSO MAKER GCM4230 +Vol - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +Battery Level Indicator -I +USB-C Charging Port -USING YOUR -ESPRESSO MAKER -Before first use -1. remove and discard all packaging material and stickers that are not +Play/Pause / Answer/End Call -permanent . +USB-C Charging Port -2. remove the water reservoir, milk reservoir, portafilter, filters, measuring -scoop, and removable drip tray. Wash them with non-abrasive soap . -Thoroughly dry the parts . Wipe the exterior with a damp cloth . +Fully Charged -3. Before using the espresso maker for the first time, it is required to prime the +Charging -appliance . See Setting up and priming the Espresso Maker below . +x2: Skip Forward -Setting up and priming the -Espresso Maker -1. Place the espresso maker on -a hard, flat, and heat-resistant -surface . +x3: Skip Back -7. Set the froth strength control to +HOLD: Voice Assistant -the clean position . +Battery Level Indicator -8. Turn the froth dispensing +High Battery -2. Detach the water reservoir and -open the water reservoir cover . +Med Battery -tube lever to rotate the froth -dispensing tube upward . +Low Battery -9. Place a heatproof cup or mug +N05909 HEADPHONE / OWNER’S MANUAL -under the froth dispensing tube . - -10. Tap the CLEAN icon . The clean +7 -icon will blink until the espresso -maker is done preheating . After -preheating, water will dispense -into the cup . When cleaning is -complete, the appliance will beep -three times . + CHARGE -11. Discard the water dispensed into +CHARGE -the cup . +Press and hold (>1 second) ON/OFF button to power ON. +Left earcup LED will show solid white light while unit is on - +Battery level will be displayed for a few seconds at startup on +Right earcup - Green high charge - Orange medium charge +- Red low charge -3. Fill the water reservoir with water -between the “Min” and “Max” -mark and reattach it to the back -of the appliance . . +To conserve battery power, headphones will power OFF after +30 minutes of inactivity (default) - Default can be changed via +app - 30min, 1HR, 3HRS, NEVER -4. Plug the power cord into an +To charge, plug in the USB-C Connector and connect to a +powered USB-A port -outlet . +NOTE: The headphones power OFF when charging. +Headphones cannot charge and play at the same time. -5. Turn the espresso maker on by -flicking the power switch on the -right side of the appliance . The -ESprESSo icon and ManuaL -ESprESSo icon will light up on -the control panel . +A solid orange light on the left earcup indicates charging +and a solid green light indicates a full charge when the USB +cable is connected. -6. +Fully Charged -Insert the empty milk reservoir -into the appliance until it clicks -into place . All icons will light up on -the control panel . +Charging 8 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - filling the portafilter -1. Select your filter size . Use the - -single-shot filter for one shot of -espresso . Use the double-shot -filter for larger sized drinks, or -for a double shot of espresso . -Align the gap in the filter with the -groove in the portafilter . Drop the -filter into the portafilter and rotate -the filter to ensure it’s locked in -place . - -2. Use the measuring spoon to fill - -the filter with espresso grounds . -Do not exceed the MAX line in -the filter . - -3. Use the tamper end of the - -measuring spoon to press on -the grounds . The surface of the -espresso grounds should be -smooth after pressing . - -4. Clean any excess grounds from - -the rim of the portafilter . - -placing the portafilter -1. Place the portafilter underneath -the brew head, with the handle -facing toward the left . - -2. Keep one hand atop the - -appliance to keep it steady while -fastening the portafilter into the -brew head . The handle must -align with the open lock icon on -the brew head . - -3. Turn the handle to the right to - -lock the portafilter into the brew -head . The handle must align with -the closed lock icon on the brew -head . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I +N05909 HEAPHONE / OWNER’S MANUAL -9 + CONNECTIONS -USER MANUAL ESPRESSO MAKER GCM4230 +CONNECTIONS - Mug Size Guide +PAIRING -ESPRESSO +The first time you power ON your headphones they will enter pairing mode. Simply connect through the Bluetooth menu on your +device and select Nº5909. -CAPPUCCINO +To manually activate pairing mode, power headphones OFF, then hold ON/OFF button for 4 or more seconds. -LATTE +Pairing to 2 or More Devices -AMERICANO +1. Enter pairing mode and connect to device 1. -Drink +2. Enter pairing mode again and device 1 will disconnect. -Shot +3. Connect to device 2 and turn headphone off. -Size +4. Turn headphone ON and headphone will automatically connect to device 1 and 2 simultaneously. -SINGLE +5. Pause one device and play from the other to swap headphone connection. -DOUBLE +NOTE: Maximum number of devices that can be connected at any given time - 2 source devices. -SINGLE +PASSIVE -DOUBLE +The № 5909 can be connected passively to any product that has an analog 3.5mm or 6.3mm (1/4") audio connetion using the +included 1.25m or 4m to USB-C proprietary audio cables included. -SINGLE +1. Choose the appropriate length (1.25m or 4m) proprietary audio cable depending on distance from the source component. -DOUBLE +2. Plug the USB-C connection end into the USB-C input on the Right ear cup. -SINGLE +3. Plug the 3.5mm stereo mini jack on the other end of the cable to your source component's headphone input. If needed, use +the 6.3mm (1/4") adaptor for traditional audio components. -DOUBLE +NOTE: When connecting passively, none of the electronic or app functions are operational. -2 OZ +Computer / Smart Phone -3 OZ +3.5mm Proprietary Audio Cable -5 OZ +Standard AV Products -10 OZ +6.3mm Adaptor + 3.5mm Proprietary Audio Cable -12 OZ +USB-C -16 OZ +N05909 HEADPHONE / OWNER’S MANUAL -5 OZ +9 -10 OZ + CONNECTIONS -note: To accommodate larger mugs, remove the drip tray . +CONNECTIONS -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +USB DIGITAL DIRECT -I +The № 5909 can be connected and used to listen to content via USB Digital Direct using the standard USB-C to USB-C charging +cable included. -drink types +1. Plug one end of the USB-C to USB-C cable directly into one of your computer's USB connections (use the included USB-C +to USB-A adaptor if needed). -Espresso -One shot of -espresso . +2. Plug the other end of the USB-C cable into the USB-C input on the right ear cup of the headphone. -Cappuccino -Made up of -1/3 espresso, -1/3 steamed milk, -and 1/3 milk foam . +3. From your computer device, open the "sound settings" and locate and select the "No 5909" under the speaker label. -Latte -Made up of -1/3 espresso and -2/3 steamed milk -topped with milk foam . +4. Adjust your volume via the slider on your computer. -Americano -Made up of -1/3 espresso and -2/3 hot water . +5. Start your content - enjoy. -Espresso +NOTE: ANC functions will continue to operate via the left ear cup ANC button AND the headphone will charge while in this +connection method. -Steamed Milk +Computer -Hot Water +USB-C to USB-A Charging adaptor + USB-C Charging Cable -Milk Foam +USB-C 10 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Making an Espresso - -Ensure the water reservoir is filled with water and placed to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a cup on the drip tray . - -4. For a single shot, tap the - -ESPRESSO icon once . For a double -shot, tap the ESPRESSO icon twice . -Your selected drink type will blink -until the espresso maker is done -preheating . After preheating, the -brewing process will begin . - -5. When brewing is complete, the -appliance will beep three times -and your espresso will be ready -for sipping . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -11 +N05909 HEAPHONE / OWNER’S MANUAL -USER MANUAL ESPRESSO MAKER GCM4230 + APP - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +APP -I +The Mark Levinson Headphones app allows you to choose +your listening preference and select Active Noise Cancellation +and Awareness modes. Select Bass Contour settings, Set +on-head detection and auto time-off parameters. -Making a Cappuccino +App Start-up -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +Landing Page -1. Fill the milk reservoir with milk +Settings -7. For a single shot, tap the +Support -below the “Max” mark . +N05909 HEADPHONE / OWNER’S MANUAL -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -cappuccino function will not be -available as an option . +11 -3. Set the froth strength control to + MAINTENANCE / CARE / CHANGING / REMOVING EARPADS -the maximum strength . -note: The above setting is a -recommendation only . +MAINTENANCE / CARE -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . +Clean ear pads and headphones with soft, slightly damp cloth. Do not soak, submerge or allow moisture to enter near the +drivers or cable jacks. Ear pads and cables can be purchased if they are worn out at: www.marklevinson.com -5. To attach the portafilter to the +Polishing cloth can be used to wipe down after use / remove finger prints, etc. . . -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . +Do not drop, sit on, or allow the headphones to be exposed to water, moisture, or temperature extremes. -6. Position the froth dispensing tube +It is recommended that you return your headphones to their hard travel case when not in use. To avoid damage, do not store +other items in the travel case along with your headphones. -over a large heatproof cup . +CAUTION: Incorrectly rotating or over rotating the earcups can damage the headphones. -CAPUCCINO icon once . For a -double shot, tap the CAPPUCCINO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . +Temperature Range Operate and store this product within the temperature range of -4°F to 113°F (-20°C to 45°C) only. Charge +the battery where the temperature is between 41°F and 104°F (5°C and 40°C) only. -8. When brewing is complete, the -appliance will beep three times -and your cappuccino will be -ready for sipping . +CHANGING / REMOVING EARPADS -9. Discard the spent grounds by +To remove ear pads: Gently pull the earpads from the inside +of the earcup. The snaps will release. -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +To re-attach ear pads: Gently push the earpad against the +earcup, ensuring all the snaps are secured. 12 -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making a Latte - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. Fill the milk reservoir with milk - -below the “Max” mark . - -6. Position the froth dispensing tube -over a large heatproof cup . - -2. - -Insert the milk reservoir into the -appliance until it clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -latte function will not be available -as an option . - -3. Set the froth strength control to - -the minimum strength . -note: The above setting is a -recommendation only . - -4. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -5. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -7. For a single shot, tap the LATTE -icon once . For a double shot, -tap the LATTE icon twice . Your -selected drink type will blink until -the espresso machine is done -preheating . After preheating, the -brewing process will begin . - -8. When brewing is complete, the -appliance will beep three times -and your latte will be ready for -sipping . - -9. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +N05909 HEAPHONE / OWNER’S MANUAL -13 + ACCESSORIES -USER MANUAL ESPRESSO MAKER GCM4230 - - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I - -Making an americano - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. - -Insert a filled or empty milk -reservoir into the appliance until it -clicks into place . -note: If the milk reservoir is not -securely inside the appliance, the -americano function will not be -available as an option . - -2. Set the froth strength control to -the Clean / Americano position . - -3. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -4. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -5. Position the froth dispensing tube -over a large heatproof cup . - -6. For a single shot, tap the - -AMERICANO icon once . For a -double shot, tap the AMERICANO -icon twice . Your selected drink -type will blink until the espresso -machine is done preheating . After -preheating, the brewing process -will begin . - -7. When brewing is complete, the -appliance will beep three times -and your americano will be ready -for sipping . - -8. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge, and hold it in place as -you dump the grounds into the -trash . +ACCESSORIES -14 +Travel Storage Case -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 - - Brewing in Manual Mode - -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . - -1. To fill the portafilter with espresso, -follow the instructions under -“Filling the Portafilter” on page 9 . - -2. To attach the portafilter to the - -brew head, follow the instructions -under “Placing the Portafilter” on -page 9 . - -3. Place a heatproof cup over the - -drip tray . - -4. Tap the MANUAL ESPRESSO icon . - -The text below the icon will blink -for three seconds then the brew -cycle will begin . - -5. When your desired amount of -coffee has dispensed, tap the -MANUAL ESPRESSO icon again and -coffee will cease brewing . -note: The maximum amount of -coffee that will dispense is 14 oz . - -6. Discard the spent grounds by - -removing the portafilter from the -brew head . Lift the filter retainer -clip so that it’s placed over the -filter edge and hold it in place as -you dump the grounds into the -trash . - -r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u - -I +USB-C to USB-C Charging Cable -15 +USB-C to USB-A Charging adaptor -USER MANUAL ESPRESSO MAKER GCM4230 +USB-C to 3.5mm Proprietary Audio Cable +(Both 1.25mm and 4mm Audio Cables included) - r -E -K -a -M -o -S -S -E -r -p -S -E -r -u -o -Y -G -n -S -u +3.5mm to 6.3mm (1/4" standard) +headphone adaptor -I +Airplane adaptor -frothing Milk +N05909 HEADPHONE / OWNER’S MANUAL -Ensure the water reservoir is filled with water and attached to the back of -appliance and the espresso maker is plugged in and turned on . +13 -1. Fill the milk reservoir with milk + TROUBLESHOOTING -5. Tap the FROTH icon . +TROUBLESHOOTING -below the “Max” mark . +TROUBLE PAIRING HEADPHONES AND DEVICE -2. Secure the milk reservoir in the +THE HEADPHONES AND DEVICE DO NOT PAIR: -appliance until it clicks into place . +1. Make sure the headphones are not connected to the charger. Cycle power switch OFF and ON. -3. Set the froth strength control . +2. Turn the Bluetooth OFF, then on again on the source device. -4. Position the froth dispensing tube +3. Turn OFF the Bluetooth feature on any other device that is paired with the headphones. -over a heatproof cup . +4. Hold power button down >4 seconds to enter pairing mode. -6. Milk will froth for a maximum of -three minutes . To stop frothing -earlier, tap the FROTH icon again -and milk will cease frothing . +5. Make sure “Nº5909" is selected in the Bluetooth® device list -16 +NO AUDIO OR POOR AUDIO QUALITY FROM A CONNECTED BLUETOOTH® DEVICE -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +Move the device closer to the headphones, physical objects between the headphone and the source may obstruct the signal. - I +Try another music application or try another track. -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +Try a different audio device. -I +FACTORY RESET -CLEANING AND MAINTENANCE +1. Power OFF the headphones. -the following cleaning instructions should be followed after every -use of the espresso maker. +2. Hold power button >4 seconds to enter pairing mode. -Cleaning the portafilter and Brew Head +3. Simultaneously hold volume + button and power button until LED lights flash red/white. -1. Turn the appliance off and unplug -the power cord from the outlet . -Wait for the metal parts to cool -down before touching them . +14 -2. Remove the portafilter from the -brew head and dump the used -espresso grounds . Lift the filter -retainer clip so that it’s placed -over the filter edge, and hold it in +N05909 HEAPHONE / OWNER’S MANUAL EAR HEALTH -place as you dump the espresso -grounds into the trash . +EAR HEALTH -3. Rinse the single-shot filter, +Our ability to hear is amazing. The Mark Levinson team wishes you to treat your ears like the precious tools they are so they will +continue to provide you a lifetime of incredible sonic experiences. -double-shot filter, and portafilter -with warm, soapy water . +In general, do not listen “too loud or for too long,” and be attentive to your own ears. Ringing, discomfort, or sensitivity to high +frequencies and treble may be a signal from your ears that you are pushing them to their limits and causing irreparable damage. -4. Wipe the underside of the brew -head with a damp cloth to clean -off any excess coffee grounds . +As a general practice, find what seems to be your preferred listening level and then adjust the volume down another 10% or +more. Surprisingly, your ears will attune to a slightly lower volume over time, and what seems quiet at first can be perfect for +longer listening periods. -Cleaning the Water reservoir +15 -1. Discard leftover water . +N05909 HEADPHONE / OWNER’S MANUAL SPECIFICATIONS -2. Wash the water reservoir with soapy water . Rinse and dry well . +SPECIFICATIONS -Cleaning the Milk reservoir and frothing tube +GENERAL -1. Remove the milk reservoir from +Type: -the unit . If there is any remaining -milk, pour the milk into a separate -container and store it in the -refrigerator . +Driver Size: -2. Wash the milk reservoir and milk +Premium High-Resolution Over-the-Ear Wireless Adaptive ANC Noise Cancelling Headphone -reservoir lid thoroughly with warm, -soapy water . Do not use abrasive -cleansers and scouring pads . -note: Hand wash only! The milk -reservoir is not dishwasher safe . +40mm Beryllium Driver -3. Ensure there is water in the water +Frequency Response (Passive): -reservoir . +10Hz - 40kHz -4. Position the froth dispensing tube -over a large heatproof cup . +Frequency Response (Active): -5. Set the froth strength to the Clean -/ Americano position . Tap the -CLEAN icon . +20Hz - 20kHz -6. Wait until all the water has -finished dispensing before -removing the cup from -underneath the brew head -(the unit will beep three times -and other function buttons will -reappear) . +Sensitivity: -7. Dispose of the water in the mug -and any remaining water in the -milk reservoir and water reservoir . +Maximum SPL: -17 +Maximum Input Power (Passive): -USER MANUAL ESPRESSO MAKER GCM4230 +97dB SPL @1kHz/1mW - I +98dB -E -C -n -a -n -E -t -n -a -M -d -n -a -G -n -n -a -E -L -C +100mW -I +Microphone Sensitivity: -descaling +-32dBV @ 1kHz/Pa -After using the espresso maker for a while, limescale will begin to build up inside -the appliance . It is important to descale the appliance every 2-3 months . If the -espresso maker is used daily, the descaling process should be done monthly . +Impedance: -1. Fill the removable water reservoir +Bluetooth Transmitted Power: -5. On the following day, discard the +32 ohm -with equal parts water and -vinegar between the “Min” and -“Max” mark . +<8dbm -2. Use the froth dispensing +Bluetooth Transmitted Modulation: -tube lever to rotate the froth -dispensing tube upward and -position a mug underneath it . +GFSK, π/4 DQPSK / 8DPSK -3. Tap the CLEAN icon . +Bluetooth Frequency: -4. Allow remaining water and +2.402Ghz - 2.480Ghz -vinegar to sit overnight in the -water reservoir . +Bluetooth Profile Version: -troubleshooting +A2DP 1.3.1, AVRCP 1.6, HFP 1.7.1 -water and vinegar . +Bluetooth Version: -6. Rinse the water reservoir +v5.1 -thoroughly with tap water and -then pour the water back into the -sink . +Battery Type: -7. Fill up the water reservoir with +Power Supply: -water and begin brewing again . -This washes out any remaining -vinegar . -notE: If an odor of vinegar -persists, run another cycle with -water . +Charging Time: -ISSUE +Music Playtime With BT On: -SOLUTIONS +Music Playtime With BT & ANC On: -Only the ESPRESSO and MANUAL -ESPRESSO icons appear on the control -panel. +Music Playtime With Aux-In & ANC On: -A drink has been chosen and the unit is -running, but no coffee is dispensing from -the portafilter. +Lithium-ion battery (750 mAh/3.7V) -Water is spreading from the drip tray -when brewing or cleaning. +5V / 1.0A -The other icons will appear when the milk -reservoir is correctly inserted into the unit. -Insert the milk reservoir into the unit by aligning -the groove on top of the reservoir with the -notch on the unit and pushing. It will click into -place when done correctly. +100 minutes -Detach then reattach the water reservoir to the -unit (this may need to be done a few times). -Run a clean cycle. See CLEANING THE MILK -RESERVOIR AND FROTHING TUBE on page 17. +34 hrs -A. Water dispenses from the froth dispensing -tube when brewing Americano or running a -clean cycle. Ensure the froth dispensing tube is -above the cup. +30 hrs -B. Empty the drip tray. +30 hrs -18 +Dimensions (H x W x D): -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 +202.6mm (7.9") x 205.4mm (8.1") x 65.4mm (2.6") - WARRANTY -AND SERVICE +340g (12 oz) -This Limited Warranty is for one full year from the date of -purchase and requires proof of purchase from an authorized -seller . It is applicable to the original purchaser only and is not -transferable to a third party user . Repair or replacement of -defective parts is solely at Gourmia’s discretion . In the event -that repair isn’t possible, Gourmia will replace the product/part . -If product repair/replacement won’t suffice, Gourmia has the -option of refunding the cash value of the product or component -returned . +2 Years Parts & Labor -Product defects not covered under the Warranty provisions -include normal wear and damage incurred from use or -accidental negligence, misuse of instruction specifications -or repair by unauthorized parties . Gourmia is not liable for -any incidental or consequential damages incurred by such -circumstances . +Weight: -Only valid within the United States . +Warranty: -register your product at -www.gourmia.com/warranty +16 -for customer service, questions or comments +N05909 HEAPHONE / OWNER’S MANUAL WARRANTY -Visit +Harman International Industries, Incorporated (“Harman”) warrants to you that your Mark Levinson product will be free from +defects in workmanship and materials for a corresponding period set forth in the Mark Levinson Limited Warranty. -gourmia.com +WARRANTY -Email +This Limited Warranty does not cover defects which are a result of: (1) damage caused by accident, disaster, misuse, abuse or +other unreasonable use or neglect (including the lack of reasonable and necessary maintenance); (2) damage occurring during +shipment (claims must be presented to the carrier); (3) damage caused by improper installation or adjustment; (4) damage +to, or deterioration of, any accessory or decorative surface; (5) damage resulting from failure to follow instructions contained +in your owner’s manual; or (6) damage resulting from the performance of repairs by someone other than an authorized Mark +Levinson service center. -info@gourmia.com +Furthermore, this Limited Warranty covers only actual defects within the Product itself, and does not cover the cost of +installation or removal from a fixed installation, setup or adjustments, claims based on misrepresentation by the seller, +performance variations resulting from installation-related circumstances such as program source quality or AC power or +Product modifications, any Product on which the serial number has been effaced, modified or removed, or any Product sold on +an “as-is” or final sale basis. This Limited Warranty is valid only for Products purchased from an authorized Mark Levinson +dealer. -Call +Complete Mark Levinson warranty details can be found at https://www.marklevinson.com/about/purchase-policy.html -888.552.0033 -Mon-Thu 9:00 a .m . to 6:00 p .m . ET -Fri 9:00 a .m . to 3:00 p .m . ET -Languages: English, Spanish, French +17 -@gourmia +N05909 HEADPHONE / OWNER’S MANUAL HARMAN International Industries, Incorporated +8500 Balboa Boulevard +Northridge, CA 91329 USA -I +© 2020 HARMAN International Industries, Incorporated. All rights reserved. -E -C -v -r -E -S -d -n -a -Y -t -n -a -r -r -a -W +Mark Levinson is a registered trademark of HARMAN International Industries, Incorporated. -19 +Other company and product names may be trademarks of the respective companies with which they +are associated. -USER MANUAL ESPRESSO MAKER GCM4230 +"MQA" is a trade mark of MQA Limited © 2018. - Get cooking with +This document should not be construed as a commitment on the part of HARMAN International +Industries, Incorporated. The information it contains, as well as the features, specifications and +appearance of the product, is subject to change without notice. HARMAN International Industries, +Incorporated, assumes no responsibility for errors that may appear within this document -We’d love to hear from you! -888.552.0033 -info@gourmia.com +For customer service and product shipment information, refer to our Website: www.MarkLevinson.com +www.MarkLevinson.com +Part No. 070-00004 rev B.0 + +18 -20 +N05909 HEAPHONE / OWNER’S MANUAL -FOR CUSTOMER SERVICE VISIT WWW.GOURMIA.COM OR CALL 888.552.0033 \ No newline at end of file diff --git a/q121/random_k4/random_3.md b/q121/random_k4/random_3.md index 1fcd9c0007bb10d1c45cc658d02fe79e8c0298ad..93421da53624a4659d5f91ceaa6a712c34229053 100644 --- a/q121/random_k4/random_3.md +++ b/q121/random_k4/random_3.md @@ -1,2074 +1,2535 @@ -Priya Sharma, MD -Department of Radiology -University of Florida Gainesville +BY08 Capital Asset Plan and Business Case Summary +Exhibit 300 -GUIDELINES FOR RESIDENTS IN FLUOROSCOPY +PART I: SUMMARY INFORMATION AND JUSTIFICATION -Welcome to GI/GU Fluoroscopy +In Part I, complete Sections A. B, C, and D for all capital assets (IT and non-IT). Complete Sections E and +F for IT capital assets. -Introduction: +Section A: Overview (All Capital Assets) -The aim of the Fluoroscopy Service at UF is to provide complete and accurate +The following series of questions are to be completed for all investments to help OMB to identify which +agency and bureau is responsible for managing each capital asset, which OMB MAX budget account funds +the project, the kind of the project, who to contact with questions about the information provided in the +exhibit 300, and whether or not it is an IT or a non-IT capital asset. -studies performed in an expeditious manner, with attempts to minimize patient +(1) Date of Submission: +(2) Agency: +(3) Bureau: +(4) Name of this Capital Asset: -discomfort and delay while assisting referring physicians in answering clinical +09/11/2006 +422 -conundrums. While many answers can be provided through the use of cross- +00 -sectional imaging, fluoroscopy provides real-time visualization of dynamic processes +Financial Accounting System (FAS) -as they occur. +(250 Character Max) +(5) Unique ID (Unique Project +Identifier ) : +Format xxx-xx-xx-xx-xx-xxxx-xx +(For IT investments only, see section 53. For all other, use agency ID system.) -• Each day begins after morning conference where you will stop by in +422-00-01-01-01-0001-00 -Fluoroscopy to obtain the list of daily cases (from Nanci – Lead Fluoroscopy +(6) What kind of investment +Will in be this FY2008? +(7) What was the first budget +year this investment was +submitted to OMB? -Technologist). +Operations and Maintenance -• Review your schedule and find out which studies are happening in the North +FY2001 or earlier -and South Tower. If you notice that there are studies that are scheduled at +(8) Provide a brief summary and justification for this investment, including a brief description of how this +closes a gap in part or in whole an identified agency performance gap: (2500 Char Max) -the same time (requiring you to be two places at once!), please ask a resident +The Financial Accounting System (FAS) is the main tool NSF staff utilizes to monitor, control, and execute the +management and financial accountability of approximately 20,000 active awards with 2000+ external +grantees. The FAS is an on-line real time system is the primary vehicle for managing all funding allocated to the +NSF. A central transaction management procedure called the "post-routine" represents the core of the system, +handling the full spectrum of financial transaction processing, accounts maintenance, transaction history, and +rules processing. This procedure processes any financial request, whether generated from within the Financial +System itself or by an external system. The FAS is extensively integrated with all of NSF's core business +systems including the e-Jacket, Proposal and Reviewer System, the Awards System, Guest Travel System, and +the FastLane System in support of Grants management. NSF's accounting and financial systems staff +created a data warehousing environment with the ability to produce our financial statements including the +Closing Package statements, generate automated closing entries and in produce the SGL Tie points reports. In +FY06, NSF implemented a near real-time interface between FAS and FedTraveler that -who has already completed his/her fluoroscopy rotation for assistance +NSF FAS 1-3-07 -(covering resident) in your schedule. As the GI/GU resident you will have +1 of 18 -first choice on which study you would like to do! +FY08 Exhibit 300 -o The covering resident will likely be able to assist you by doing a few of + includes evaluation of funds availability and funds obligation. NSF expects the same type of data exchanges with +other externally provided cross-services. The e-Learning and e-Procurement /FAS interfaces are +scheduled to be implemented in FY 2007. The FAS will remain in a steady-state phase in the FY05-FY10 +timeframe. NSF will be approaching its future financial system requirements as an integral part of its grant +process. NSF will conduct an integrated review of the GMLOB/FMLOB solution in 2007. If that solution is +determined to not be feasible, NSF will analyze the FMLOB Shared Service Provider (SSP) options in 2008. If +necessary, NSF will conduct a Business Case Feasibility Study for the FMLOB solution in 2009. This plan allows +the NSF to take advantage of the results/findings of the GMLOB process in becoming a SSP to more fully +define our financial requirements. NSF anticipates that if a conversion is necessary it will substantially impact +NSF Grantees and will begin in 2010. The current FAS will remain in steady-state maintenance until the +conversion is complete. Any decisions NSF makes regarding the future financial system modernization will +support the President's Management Agenda. -the swallowing studies that have been scheduled in the North Tower. +(9) Did the Agency's Executive/Investment Committee approve this request? -o Please make sure that you notify the head fluoro tech that the +a. If "yes," what was the date of this approval? -covering resident will be covering cases in the North Tower, while you +(10) Did the Project Manager review this Exhibit? -perform cases in the South Tower. +11) Contact Information of Project Manager? -▪ Provide the Head Fluoro Tech with the name of your covering +N a me : +Phone Number: +E-Mail: -resident and an efficient means to communicate with him/her. +Carolyn Miller - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +703-292-4272 -• After obtaining the list of cases it is important to determine why the +cmiller@nsf.gov -studies are being requested. The key to a successful fluoroscopy rotation is +yes -taking an active role a patient care advocate. This involves obtaining a good +. -history- through EPIC (medical records), conversations with the referring +08/30/2006 -clinicians as well as the review of prior imaging. Do not forget to review +yes -recent procedures such as endoscopies, colonoscopies and the operative +(12) Has the agency developed and/or promoted cost effective, energy-efficient and environmentally +sustainable techniques or practices for this project. -reports. This information will help you understand what you might see during +no -the examination, allow you to anticipate certain findings and possible changes +(a) Will this investment include electronic assets (including computers)? -that may need to be made in a “routine” study protocol. +yes -2 +(b) Is this investment for new construction or major retrofit of a Federal building or +facility? (answer applicable to non-IT assets only) +[1] If "yes," is an ESPC or UESC being used to help fund this investment? +[2] If "yes," will this investment meet sustainable design principles? +[3] If "yes," is it designed to be 30% more energy efficient than relevant code? -• In some cases, you may note that the examinations which are ordered are +(13) Does this investment support one of the PMA initiatives? yes -not in fact actually needed. Please check with your attending cancelling any +If "yes," select all that apply: +President's Management Agenda (PMA) Initiatives +E x p a n d e d E - G o v e r n m e n t -of the cases on the schedule. +Budget Performance Integration -• Once you have done your homework, present the cases to the GI/GU +a. Briefly describe how this asset directly supports the identified initiative(s)? -attending and discuss what are you planning to do and what you may see in +NSF's Enterprise Information System, the FAS and the Report web reporting system provide +financial, budgetary, awards and performance information (including PART) that is accessible in -the clinical setting. +NSF FAS 1-3-07 -• After performing the study, send the pictures via PACS. Call the attending +2 of 18 -to review the images. The attending may ask you to obtain more images. DO +FY08 Exhibit 300 -NOT SEND THE PATIENT without reviewing the images with an attending. + multiple formats at every workstation on a 24/7, real-time basis. Managers use this information to +make decisions regarding NSF budget priorities and business processes. NSF's Statement of Net +Cost uses the FAS to report the full cost of each strategic goal - Ideas, Tools, People and +Stewardship. -If the GI/GU attending has stepped out ask for an alternate attending. +(14) Does this investment support a program assessed using OMB's +Program Assessment Rating Tool (PART)? +(a) If "yes," does this investment address a +weakness found during a PART review? +(b) If "yes," what is the name of the PARTed +program ? +(c) If "yes," what rating did the PART receive? -• You may be requested to add additional cases onto the schedule, before +no -adding cases check with your attending. +(15) Is this investment for information technology? (see section 53 for definition)) yes -The remainder of this packet will contain your educational goals, guidelines, and +If the answer to Question 15 was "Yes," complete questions 16-23 below. If +the answer is "No," do not answer questions 16-23. -most common procedural protocols as well as a list of references to guide your +Level 1 -reading for this month. + (1) Project manager has been validated as qualified for this +investment - COMPETENCY-BASED GOALS AND OBJECTIVES +no -1. Medical Knowledge +No -a. Instrumentation and Protocols +(16) What is the level of the IT Project (per CIO +Council PM Guidance)? +(17) What project management qualifications does +the Project Manager have? (per CIO Council PM -i. Basic fluoroscopic instrumentation +(18) Is this investment identified as "high risk" on +the Q4 - FY 2006 agency high risk report +(per OMB's 'high risk" memo)? -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +(19) Is this a financial management system? +(a) If "yes," does this investment address a FFMIA +compliance area? -3 - -A. The first time you are on fluoro – ask your attending for - -an introduction to the fluoro tower and buttons - -ii. Common fluoroscopic artifacts - -iii. Become familiar with common fluoroscopic protocols - -iv. Types of oral contrast - -b. Bowel - -i. Normal and abnormal appearances of the esophagus, stomach, - -small bowel and colon on fluoroscopy (details to follow in the - -packet) - -c. Gynecology - -i. Normal and abnormal appearance of the uterus and uterine - -tubes on hysterosalpingography (HSG) - -d. Urethra and Bladder - -i. Normal and abnormal appearance of the urethra on retrograde - -urethrogram (RUG) - -ii. Normal and abnormal appearance of the urinary bladder and - -urethra on voiding cystourethrogram (VCUG) - -2. Patient Care - -a. Gather clinical information about patients relevant to the - -interpretation of the examination as described in the introduction. - -b. Communicate with the patient prior the procedure. Introduce - -yourself, including the fact that you are a radiology resident. Let - -them know what you will be doing and what you are looking for. - -Demonstrate caring and respectful behavior. - -i. In your pediatric fluoroscopy rotation – you will be explaining - -the procedure to the patient’s parents who will be present for - -the entirety of the study. - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -c. Use of information technology to support patient care decisions – IE - -Look up things you don’t know! +[1] If "yes," which compliance area: +[2] If "no," what does it address? -d. Effectively perform fluoroscopic procedures in a way that minimizes +(b) If "yes," please identify the system name(s) and +system acronym(s) as reported in the most recent +financial systems inventory update required by +Circular A-11 section 52: -patient discomfort but answers the clinician’s question. +(20) What is the percentage breakout for the total FY2008 funding request for the following? +(This should total 100%) -e. If an attending is not present and if asked or you sense the patient or +Hardware %: +0 -family member is uncomfortable with the lack of direct supervision +Software %: +0 -ask the attending to participate. +Services %: +100 -f. If you are uncomfortable with the procedure or concerned with +Other %: -findings that may need further evaluation, inform the attending. +0 -4 +Total % +100 -3. Practice-Based Learning and Improvement +(21) If this project produces information dissemination products for the public, are these products +published to the Internet in conformance with OMB Memorandum 05-04 and included in your +agency inventory, schedules and priorities? +n/a -a. Participate in self-learning including outside reading, Statdx, +NSF FAS 1-3-07 -radiopedia and articles found in Radiology and Radiographics. See the +3 of 18 -list provided. +FY08 Exhibit 300 -b. Participate in QI/QA activities – watch out for artifacts. Remember + (22) Contact information of individual responsible for privacy related questions: -to consider fluoro time and reduce dose by collimating etc. +Name: +Phone +Number: +Title: +E-Mail: -4. Interpersonal and Communication Skills +Leslie A. Jensen -a. Dictate prompt, concise, accurate and concise radiological reports. +703-292-8060 -b. Promptly communicate urgent, critical or unexpected findings to +NSF FOIA/Privacy Act Officer -residents, referring physicians or clinicians and document the +Ijensen@nsf.gov -communication in the report. (Name of clinician, findings discussed, +(23) Are the records produced by this investment appropriately scheduled with the National +Archives and Records Administration's approval? no -time and date). +Section B: Summary of Funding (All Capital Assets) +(1) Provide the total estimated life-cycle cost for this investment by completing the following table. +All amounts represent budget authority in millions, and are rounded to three decimal places. +Federal personnel costs should be included only in the row designated "Government FTE Cost," +and should be excluded from the amounts shown for "Planning," "Full Acquisition," and +"Operation/Maintenance." The total estimated annual cost of the investment is the sum of costs for +"Planning," "Full Acquisition," and "Operation/Maintenance." For Federal buildings and facilities, +life-cycle costs should include long term energy, environmental, decommissioning, and/or +restoration costs. The costs associated with the entire life-cycle of the investment should be +included in this report. -5. Professionalism +Table 1: SUMMARY OF SPENDING FOR PROJECT PHASES (REPORTED IN MILLIONS) +All amounts represent Budget Authority (Estimates for BY+1 and beyond are for planning purposes only and do not +represent budget decisions) -a. Demonstrate respect and compassion to patients, physicians, and +PY-1 +Spending +Prior to 2006 -staff. +BY +1 BY+2 2010 +2009 -b. Demonstrate positive work habits, such as punctuality. +2011 -c. Commitment to ethical principles and patient confidentiality. +Total - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +BY+4 +2012 and +beyond -6. Systems Based Practice +Planning -a. Understand how medical decisions affecting patient care within the +Acquisition +Subtotal +Planning & +Acquisition +Operations -larger system. +$0.000 -b. Demonstrate knowledge of appropriateness criteria. +$0.000 +$0.000 -5 +$0.000 - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +$0.000 +$0.000 -FLUOROSCOPIC GI/GU STUDIES +$0.000 -6 +$0.000 +$0.000 -• Esophagram/Barium Swallow +$0.000 -• Modified Barium Swallow (performed with Speech Pathologist) with or +$0.000 +$0.000 -without limited esophagram +$7.570 -• Upper GI examination (UGI) +$1.500 -• Small bowel follow through (SBFT) +$1.500 -• Post-surgical: Esophageal repair/stent, POEM, Gastric pull-up, Whipple, +$1.120 -Gastric Bypass +TOTAL -• Barium Enema Examination +$7.570 -• Air contrast Enema for Intussusception Reduction +$1.500 -• Fistulogram +$1.500 -• Voiding Cystourethrogram +$1.120 -• Retrograde Urethrogram +Government FTE Costs +should not be included +th -• Fluoroscopic Assistance for Hysterosalpingogram +i -*This list contains the most commonly performed fluoroscopic studies at Shands +t -Hospital, although this is not exhaustive of the procedures performed in the +provided +above. -Fluoroscopic Suite, it should suit your needs for this rotation* +$0.360 -• Radiologic placement of an enteric tube (RARE – and must make sure that all +$0.360 -the criteria for floor placement and Core-trak attempt have been performed +$0.360 -– please see document by Dr. Abbitt below) +$1.800 - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Government +FTE Costs +Number of +FTE +represented +by cost +Note: For the cross-agency investments, this table should include all funding (both managing partner +and partner agencies). Government FTE Costs should not be included as part of the TOTAL +represented. -7 - - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -OVERVIEW OF CONTRAST AGENTS - -8 - -There are a few types of contrast that are used in Fluoroscopy: - -1. Barium Sulfate: is a relatively inert agent; however it can incite an - -inflammatory response if spilled into the MEDIASTINUM or PERITONEAL - -CAVITY. Use barium cautiously if there is a colonic obstruction. +NSF FAS 1-3-07 -a. Consult Peds attending if there is a question. +4 of 18 -i. Peds surgery prefers Barium over water soluble. +FY08 Exhibit 300 -ii. Peds tends to use Barium in most cases, except in the post-op + (2) Will this project require the agency to hire additional FTE's? no -state. +(a) If "yes," How many and in what year? -2. Iodine-based contrast agents: +(3) If the summary of spending has changed from the FY2007 President's budget request, briefly +explain those changes. -a. High Osmolality Agents (HOCM) – include agents such as Gastroview +Section C: Acquisition/Contract Strategy (All Capital Assets) -and Gastrograffin (never used in kids!). These tend to be flavored +(1) Complete the table for all contracts and/or task orders in place or planned for this investment: -(not well!) and are for enteric use ONLY. All of these agents are ionic. +Contract or Task Order Number: BZ-11/0533982 Type of Contract/TO Used: Cost Plus Fixed Fee +Has the Contract Being Awarded: yes +Contract Actual/Planned Award Date: +05/01/2005 +Contract/TO Start Date: +05/01/2005 +Contract/TO End Date: +04/30/2007 +Contract/TO Total Value ($M): $32.200 Inter Agency Acquisition: no +Performance Based Contract: yes +Competitively Awarded Contract: yes +Alternative Financing: NA EVM Required: yes +Security Privacy Clause: yes -i. WARNINGS: HOCM agents can cause bronchospasm, pulmonary +Contracting Officer (CO) Contact Information: -edema, pneumonitis and are CONTRAINDICATED IN THOSE +CO Name: Patricia S. Williams +CO Contact Information (Phone/Email): (703) 292-8240 pswillia@nsf.gov +CO Certification Level (Level 1, 2, 3, N/A): 3 +If N/A has the agency determined the CO assigned has the competencies and skills necessary to +support this acquisition? (Y/N) -PATIENTS WITH TRACHEAL ASPIRATION OR +(2) If earned value is not required or will not be a contract requirement for any of the contracts or -TRACHEOESPHAGEAL FISTULA (TEF). +task orders above, explain why: -1. Do not use these agents in patients who are status post +The contract has a requirement to utilize earned value management (EVM) for tasks/projects designated as +new development or "Development/Modernization/Enhancements (DME)." FAS is in Steady State and its +maintenance activities are exempt from EVM at this time per OMB Guidance. -esophageal interventions! Please use LOCM agents. +(3) Do the contracts ensure Section 508 compliance? yes +Section 508 Compliance Explanation: +The system was reviewed and modified, as needed, in 2001 for Section 508 compliance. The FAS consists of an +end-user GUI and back-end software programs. Only the end-user GUI is used by users; therefore only changes +to it would affect Section 508 compliance. While the majority of maintenance changes are to the back-end +software, when the GUI is affected, NSF staff review each maintenance change for Section 508 compliance +before the change is implemented. -2. These agents are hyperosmolar – so do not use them in +NSF FAS 1-3-07 -the elderly or infants – they will draw fluid into the bowel +5 of 18 -lumen. +FY08 Exhibit 300 -b. Low Osmolality Agents (LOCM) – include agents such as Omnipaque + (4) Is there an acquisition plan which has been approved in accordance with agency requirements? -and Isovue. LOCM agents include both ionic and NON-ionic compounds. +yes - HELPFUL HINTS FOR PATIENT POSITIONING +(a) If "yes", what is the date? +12/01/2001 +(b) If "no," will an acquisition plan be developed? -9 +[1] If "no," briefly explain why: -Meet the Fluoro tower… +Section D: Performance Information (All Capital Assets) -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -Picture Reference: http://pubs.rsna.org/doi/pdf/10.1148/radiographics.20.4.g00jl301115 - -(*Suggest reading*) - -We will orient the patient in reference to the X-ray Table. The patient is in the - -supine position above. - - The following images are copied from: - -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines - -10 +In order to successfully address this area of the exhibit 300, performance goals must be provided for the +agency and be linked to the annual performance plan. The investment must discuss the agency's mission and +strategic goals, and performance measures must be provided. These goals need to map to the gap in the +agency's strategic goals and objectives this investment is designed to fill. They are the internal and external +performance benefits this investment is expected to deliver to the agency (e.g., improve efficiency by 60 +percent, increase citizen participation by 300 percent a year to achieve an overall citizen participation rate of +75 percent by FY 2xxx, etc.). The goals must be clearly measurable investment outcomes, and if applicable, +investment outputs. They do not include the completion date of the module, milestones, or investment, or +general goals, such as, significant, better, improved that do not have a quantitative or qualitative measure. -Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical +Agencies must use Table 1 below for reporting performance goals and measures for all non-IT investments +and for existing IT investments that were initiated prior to FY 2005. The table can be extended to include +measures for years beyond FY 2006. -Publishers Inc.; 1996. +Table 1 - Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical +Fiscal Year -Publishers Inc.; 1996. +Strategic -11 +Goal(s) +Supported -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +2003 - Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical +2003 -Publishers Inc.; 1996. +2003 -12 +2003 -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +2004 - REFERENCES +2004 -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +2004 -13 +2004 -There are a wide range of useful articles that can be used to learn about the +Financial +Management +Financial +Management -procedures and imaging findings of certain pathologies. These articles are listed +Financial +Management -under suggested reading which can be found after each procedure. However, there +Financial +Management -are a few books which I do suggest that you use as your primary references for +Financial +Management -“how –to” and for pathologic findings. +Financial +Management -Books: +Financial +Management -1. Houston JD, Davis M. Fundamentals of Fluoroscopy. 1st ed. Saunders, 2001. -2. Javors BR. Manual of GI Fluoroscopy. 1st ed. New York, NY: Thieme Medical +Financial +Management -Publishers Inc.; 1996. +Performance Measure -3. Johnson, CD, Schmit, GD. Mayo Clinic Gastrointestinal Imaging Review. +Actual/ -Rochester, MN: Mayo Clinic Scientific Press, 2005. +Migrate from current payroll system to +Government wide Payroll provider -Journal Articles +Meet Accelerated Financial Management +Reporting Requirements (45 days after +fiscal year-end) +Meet Quarterly reporting requirement -1. Geise, R. Fluoroscopy: Recording of Fluoroscopic Images and Automatic +Unqualified Audit Opinion for Financial +Statements -Exposure Control. Radiographics 2001; (21): 227-236. +Implement Government wide ePayroll +service provider -http://dx.doi.org/10.1148/radiographics.21.1.g01ja19227 +Baseline (from +previous year) -2. Schueler, BA. General Overview of Fluoroscopic Imaging. Radiographics +Legacy System-IPAY -2000; 20 (1) 1115-1126. +Met Current Financial +Management Reporting +Requirements +Met Current Financial +Management Reporting +Requirements +Unqualified Audit +Opinion -http://pubs.rsna.org/doi/abs/10.1148/radiographics.20.4.g00jl301115 +Legacy System-IPAY - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Planned +Performance +Metric (Target) -14 +Select ePayroll provider in +FY03 +Submit NSF Performance +and Accountability Report +to OMB on November 17 +Submit Quarterly financial +statements to OMB within +45 days of end of quarter -ESOPHAGRAM/BARIUM SWALLOW +Unqualified Audit Opinion +for FY 2003 Financial +Statements +Successful transition to +Government wide ePayroll +service provide and +retirement of IPAY system -Most Common Indications: Dysphagia, odynophagia, globus sensation +Meet Accelerated Financial Management +Reporting Requirements 45 days after fiscal +year end +Unqualified Audit Opinion for Financial +Statements -Patient Preparation: NPO for a minimum of 2 hours +Met Current Financial +Management Reporting +Requirements +Unqualified Audit +Opinion -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast), 12.5 mm Barium +Submit NSF Performance +and Accountability Report +by November 15 +Unqualified Audit Opinion +for FY 2004 Financial +Statements -Tablet +Produce Quarterly Financial Statements +within 21 days of end of quarter -*Please note that if this exam is being done in a patient who is post-operative or +Produced Quarterly +Financial Statements +within 21 days of end of +quarter -there is a question for perforation you will be using WATER SOLUBLE CONTRAST. +Produce Quarterly +Financial Statements +within 21 days of end of +quarter -Additionally, if the patient is at risk for both PERFORATION and ASPIRATION, a +Performance Metric +Results (Actual) -low osmolar intravenous agent can be administer orally without the risk of +ePayroll Service Provider +selected in January 2003 +November 15th date was +achieved one year ahead of +OMB requirements +NSF's Quarterly reports +were consistently one of the +first submitted to OMB +NSF received an Unqualified +Audit Opinion for FY 2003 +on 11/05/04 +New ePayroll service +provider began processing +NSF's payroll in May 2004. +Because NSF's payroll +system ran for part of the +calendar year, NSF cannot +retire its IPAY system until +all reconciliation for the +calendar year is done and W- +2s are generated +PAR Report submitted on +11/15/2004 -hypertonic pneumonitis.* +NSF received an Unqualified +Audit Opinion for FY 2004 +on 11/04/2005 +Quarterly reports produced +within 21 days beginning +March 2004 -Procedure: +NSF FAS 1-3-07 -In the North tower Fluoroscopy suite, please note that most exams are done for +6 of 18 -inpatients and therefore techniques that you may see at the VA or in Plaza may +FY08 Exhibit 300 -differ. For example in the VA or Plaza you may see the use of “fizzies.” This allows + Fiscal Year -us to create “double contrast” and allows the radiologist to evaluation the mucosa +Strategic +Goal(s) +Supported -of the esophagus. Since, most of the patients we do here at Shands are post- +Performance Measure -operative we do not typically use “fizzies” or the double contrast technique. +Actual/ +Baseline (from +previous year) -However if mucosal evaluation is needed we will use “fizzies” per attending +2004 -discretion. +2005 -Furthermore, “fizzies” are not used in the pediatric population. +Financial +Management - Adults: +Automate closing package financial +statements -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Met new Closing +package requirements -15 +Financial +Management -1. Start the procedure by explaining to the patient exactly what you are going +Unqualified Audit Opinion for Financial +Statements -to do today. +Unqualified Audit +Opinion -2. Begin with the adult in the standing position facing you. Since many of our +2005 -patients are post-operative or frail they will not be able to stand. In that +FFMIA Compliance -case the study can be performed in the supine or reverse Trendelenburg +Receive Assertion of FFMIA Compliance - +Implement FAS Requirements necessary to +maintain compliance -position. +Financial +Management -3. Fluoro the chest and remove any artifacts – such as clips and leads from the +Produce Quarterly Financial Statements +within 21 days of end of quarter -patient. +Financial +Management -4. Obtain a scout radiograph of the chest. +Unqualified Audit Opinion for Financial +Statements -a. ***IF WE ARE USING FIZZIES, PLEASE ADMINISTER THEM TO +Unqualified Audit +Opinion -THE PATIENT AT THIS POINT*** +Financial +Management -i. Have the patient take the “gas crystals” like they are doing a +Produce Quarterly and year end Financial +statement -shot, into the back of their throats, and have them chase it +2006 -with water. +FFMIA Compliance -ii. INSTRUCT THE PATIENT NOT TO BURP!! The gas is what +Receive Assertion of FFMIA Compliance - +Implement FAS Requirements necessary to +maintain compliance -gives us the contrast for mucosal evaluation. +Financial +Management -5. Ask the patient to take a mouthful of the “thick” barium” and watch the +Unqualified Audit Opinion for Financial +Statements -patient swallow. For the first swallow you do not have to obtain any images, +Unqualified Audit +Opinion -this is for you to understand the patient’s motility and ability to perform +2005 -study. +2006 -6. Now position the patient in the LPO position standing and have them swallow +20116 -a mouthful of barium (3 swallows for your 3 shots). +2007 -a. Shoot the proximal (cervical) esophagus. You may grab or shoot a full +2007 -exposure. +Financial +Management -b. Shoot the mid thoracic esophagus. +Produce Quarterly and year end Financial +statement -c. Shoot the distal thoracic esophagus. +2007 -*Please note the following images are obtained with the double contrast +FFMIA Compliance -technique outlining the mucosa* If there is a mucosal question, the exam +Receive Assertion of FFMIA Compliance - +Implement FAS Requirements necessary to +maintain compliance -should be performed with “fizzies” – please see addendum at the end of this +2007 -protocol. +2007 - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +2008 -16 +2008 -Picture Reference: +2008 -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +2008 -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +2008 -HELPFUL HINT 1: Also when, obtaining images, remember if there is +Financial +Management -something that looks like a stricture – have the patient drink and either +Federal Cash Transactions Report +Monitoring -demonstrate that the finding remains, or the contrast clears and that it was +Financial +Management -likely peristalsis. +Grant Closeout - Review the Award +Closeout Report on a quarterly basis -HELPFUL HINT 2: If you do see something abnormal remember this is what +Financial +Management -are evaluating, Take pictures, magnify your image etc. But DO KEEP AN EYE +Federal Cash Transactions Report +Monitoring -on your fluoro time! +Financial +Management -7. We will now lie the patient down on the table and place them in the right +Grant Closeout - Review the Award +Closeout Report on a quarterly basis -lateral position and obtain lateral pictures of the esophagus (obtain a column +Financial +Management -of barium in the esophagus). You may use “thin barium” as you will have the +Unqualified Audit Opinion for Financial +Statements -patient taking multiple sips quickly or “chugging the barium”. +Financial +Management -a. Center first over the upper thoracic esophagus. +Produce Quarterly and year end Financial +statement -b. Include a shot of the open lower esophageal sphincter (LES) on the +FFMIA Compliance -last two views that you take +Receive Assertion of FFMIA Compliance - +Implement FAS Requirements necessary to +maintain compliance -c. Obtain a magnified image of the LES. +Planned +Performance +Metric (Target) +Automate closing package +financial statements - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Performance Metric +Results (Actual) -17 +Closing package statement +became automated in June +2004 NSF produces its +automated financial +statements and closing +package statements +simultaneously -Picture Reference: +Unqualified Audit Opinion +for FY 2005 Financial +Statements +Maintain Assertion of +FFMIA Compliance +through FAS compliance +with requirements -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +FFMIA compliance asserted +on 11/08/2005 -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Produce Quarterly +Financial Statements +within 21 days of end of +quarter -8. Now we will evaluate for reflux. +NSF submitted timely and +accurate quarterly statements +within 21 days of the end of +quarter -a. Place the patient in the RPO position, this will allow the contrast to +Unqualified Audit Opinion +for FY 2006 Financial +Statements +Quarterly reports +produced +within 21 days and year + +ithi 45 d + +Maintain Assertion of +FFMIA Compliance +through FAS compliance +with requirements + +Unqualified Audit Opinion +for FY 2007 Financial +Statements +Quarterly reports +produced +within 21 days and year +Maintain Assertion of +FFMIA Compliance +through FAS compliance +with requirements + +ithi 45 d + +Resolve 100% of +excessive cash on hand +findings + +Close 100% of awards +within two full reporting +quarters after the +expiration date + +Resolve 100% of +excessive cash on hand +findings + +Close 100% of awards +within two full reporting +quarters after the +expiration date + +Unqualified Audit Opinion +for FY 2008 Financial +Statements +Quarterly reports +produced +within 21 days and year + +ithi 45 d + +Maintain Assertion of +FFMIA Compliance +through FAS compliance +with requirements + +Received Assertion of +FFMIA Compliance - +Implemented required +changes to SGL and +transactions + +Produced Quarterly +Financial Statements +within 21 days of end +of + +Produced Quarterly and +year-end Financial +Statements by deadlines + +Received Assertion of +FFMIA Compliance - +Implemented required +changes to SGL and +transactions -pool into the cardia of the stomach. In peds, fluoro intermittently for +Produce Quarterly and +year-end Financial +Statements by deadlines -3-5 minutes to evaluate completely. +Received Assertion of +FFMIA Compliance - +Implemented required +changes to SGL and +transactions -i. To illicit reflux, there are three maneuvers that one can apply: +Reviewed 100% of +Federal Cash +Transactions Reports +with positive cash on +hand -1. Observe the GE junction while the patient performs a +Closed 100% of awards +within two full +reporting +quarters after the +expiration date +Reviewed 100% of +Federal Cash +Transactions Reports +with positive cash on +hand -Valsalva maneuver and coughs. +Closed 100% of awards +within two full +reporting +quarters after the +expiration date +Unqualified Audit +Opinion -2. Watch while the patients rolls over onto the right lateral +Produced Quarterly and +year-end Financial +Statements by deadlines -position. +Received Assertion of +FFMIA Compliance - +Implemented required +changes to SGL and +transactions -3. “Water siphon test” – have the patient take two sips of +NSF FAS 1-3-07 -water in the RPO position. Watch the water go down the +7 of 18 -GE junction and check if there is any reflux. +FY08 Exhibit 300 - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines + Fiscal Year -ii. If you observe the patient refluxing, it is important to note +Strategic +Goal(s) +Supported -three things: +Performance Measure -18 +Actual/ +Baseline (from +previous year) -1. The maneuver that elicited the reflux. +Planned +Performance +Metric (Target) -2. How far patient refluxed (cervical, upper, mid, distal +Performance Metric +Results (Actual) -thoracic esophagus). +2009 -3. Time it takes for the esophagus to clear the reflux. +2009 -9. Now we will examine the pharynx by centering on the pharynx. The patient +2009 -will be standing (if able) for this portion of the exam. +2009 -You may use 1-2x magnification for this portion of the exam. +2009 -a. Have the patient take a few sips of thick barium and ask them to +2010 -“gargle” with the barium. Instruct the patient that you are trying to +2011 -“coat their throat” with the barium so that we may outline their +2010 -anatomy. +2010 -b. With the patient’s throat coated we will perform a series of +2010 -maneuvers to demonstrate the anatomy of the pharynx. +2011- -i. Ask the patient to hold their chin up +2011 -ii. Take a scout image +2011 -iii. Now ask the patient to say “EEEEEE” and take an image +2011 -iv. Now have the patient PUFF OUT their cheeks, take an image +2011 -v. Repeat the above maneuvers with the patient in the lateral +All new IT investments initiated for FY 2005 and beyond must use Table 2 and are required to use the +Federal Enterprise Architecture (FEA) Performance Reference Model (PRM). Please use Table 2 and the +PRM to identify the performance information pertaining to this major IT investment. Map all Measurement +Indicators to the corresponding "Measurement Area" and "Measurement Grouping" identified in the PRM. +There should be at least one Measurement Indicator for at least four different Measurement Areas (for each +fiscal year). The PRM is available at www.egov.gov. -position. +NSF FAS 1-3-07 -The following images demonstrate your scouts in the AP and Lateral positions +8 of 18 -Picture Reference: +FY08 Exhibit 300 -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: + Table 2 +Fiscal +Year -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Measurement Area IT - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Measurement Grouping IT -If you do see abnormalities on this portion of the exam – obtain a CINE like +Measurement +I di -study. Ask the patient to hold the contrast in their mouth– change the film +t -speed on the tower to 3 frames/second and have the patient swallow on the +Baseline Planned -count of three and obtain images. +Actual Results -19 +Improvement to +the Baseline -10. Complete the exam by having the patient swallow the barium pill. +Section E: Security and Privacy (IT Capital Assets Only) -a. The barium tablet is typically 12.5mm in size as strictures that are +In order to successfully address this area of the business case, each question below must be answered at the +system/application level, not at a program or agency level. Systems supporting this investment on the planning and +operational systems security tables should match the systems on the privacy table below. Systems on the Operational +Security Table must be included on your agency FISMA system inventory and should be easily referenced in the +inventory (i.e., should use the same name or identifier). -less than 13 mm are the ones that cause dysphagia in patients. +All systems supporting and/or part of this investment should be included in the tables below, inclusive of both agency +owned systems and contractor systems. For IT investments under development, security and privacy planning must +proceed in parallel with the development of the system/s to ensure IT security and privacy requirements and costs are +identified and incorporated into the overall lifecycle of the system/s. -b. Strictures ranging in size from 13-19mm may or may not cause +Please respond to the questions below and verify the system owner took the following actions: -symptoms. +(1) Have the IT security costs for the system(s) been identified and integrated into the overall costs of the -11. Demonstrate where the barium pill ends up. +investment: yes -a. If the pill gets “caught” – have the patient drink water to move the +(a) If "yes," provide the "Percentage IT Security" for the budget year: 14 -pill along. +(2) Is identifying and assessing security and privacy risks a part of the overall risk management effort for +each system supporting or part of this investment. yes -b. The pill will eventually dissolve so reassure the patient that it will not +(3) Systems in Planning - Security: -be stuck. +Name Of System -12. Now check the images with your attending and repeat any portions of the +Agency Or Contractor Operated -exam that your attending may feel are unclear. +System? -13. Upon completion of the exam, feel free to let the patient know if the +Planned Operational Date -preliminary findings, noting of course that the final read will be send to +Planned or Actual +C&A Completion Date -their ordering physician. +(4) Operational Systems - Security: -Peds +Name Of System -The above can be used as guidelines in the pediatric population; however, often +Agency Or Contractor +Operated system -times the indications are different. As mentioned above, double contrast is not +NIST FIPS +199 -used in pediatric patients. +Risk Impact +Level (High, +Moderate, +Low) -Make sure the TECH has used pediatric settings on the fluoro tower and there is +Has the +C&A been +completed +using +NIST 800- +37? -pulsed fluoro. +Date C&A +Complete -You will NOT take full exposures, but instead will use FLUORO Grabs, last image +What standards we +used for the +Security Controls +tests? -hold saves. +Date Completed - You will NOT use MAGNIFICATION. +Security +Control Testing -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Date +Contingency +Plan Tested -20 +FAS -Also, we do not typically exam the pharynx in peds, the pharynx is usually examined +Contractor and +Government -in the setting of a modified barium swallow study with a speech pathologist. +High -Suggested Reading: +yes -1. Levine, MS, Rubensin SE. Diseases of the Esophagus: Diagnosis with +04/05/06 -Esophagography. Radiology 237(2): 414-427. +FIPS 200 / NIST +800-53 -http://geiselmed.dartmouth.edu/radiology/resources/fluoroscopy/pdf/esop +01/18/06 -hageal_disease.pdf +02/15/06 -2. Levine, MS. Radiologic Evaluation of the Esophageal Phase of Swallowing: In +(5) Have any weaknesses, not yet remediated, related to any of the systems part of or supporting this +investment been identified by the agency or IG? no -Levine, MS. Manual of Diagnostic and Therapeutic Techniques for Disorders +(a) If "yes," have those weaknesses been incorporated into the agency's plan of action and milestone -of Deglutition. http://link.springer.com/chapter/10.1007%2F978-1-4614- +process? -3779-6_4 +(6) Indicate whether an increase in IT security funding is requested to remediate IT security +weaknesses? no -3. Leudtke, PA et al. Radiologic Diagnosis of Benign Esophageal Strictures: A +NSF FAS 1-3-07 -Pattern Approach. Radiology 23 (4): 897-909. +9 of 18 -http://pubs.rsna.org/doi/pdf/10.1148/rg.234025717 +FY08 Exhibit 300 -4. Gupta S, Levine MS, Rubesin SE, Katzka DA, Laufer I. Usefulness of barium + (a) If "yes," specify the amount, a general description of the weakness, and how the funding request +will remediate the weakness. -studies for differentiating benign and malignant strictures of the +(7) How are contractor security procedures monitored, verified, and validated by the agency for -esophagus. AJR 2003; 180:737-744. +the contractor systems above? -5. Baker et al. Gastroesophageal Reflux Disease: Integrating the Barium +NSF uses a range of methods to review the security of operations through contract requirements, project +management oversight and review, certification and accreditation processes, IG independent reviews, +proactive testing of controls through penetration testing and vulnerability scans to ensure services are +adequately secure and meet the requirements of FISMA, OMB policy, NIST guidelines and NSF policy. The +system is operated on-site by a team of contractors and NSF personnel with system administrators tightly +controlling access to the systems. Only administrators with current need have access to the system, +and strict code migration, quality control, and configuration management procedures prevent deployment of +hostile or vulnerable software on the systems. Contractors are trained in the same security measures as +NSF employees. All NSF employees and contract staff are required to complete an on-line security training +class each year, including the rules of behavior. Background checks are done routinely as a part of the +NSF contracting process, and IT security requirements are stated in the contracts statement of work. +Contractor security procedures are monitored, verified, and validated by the agency in the same way as for +government employees. Once on board, contractors are allowed access to the NSF systems based on their +specific job requirements. Audit logs are also implemented to monitor operating system changes - these +audit logs are reviewed by the system administrators. Additionally, roles and responsibilities are separated +to the extent possible to allow for checks and balances in system management and multiple levels of +oversight. -Esophagram before and after Antireflux Surgery. Radiographics 243 (2): +(8) Planning and Operational Systems - Privacy: -329-339. http://pubs.rsna.org/doi/pdf/10.1148/radiol.2432050057 +(a) Name Of System -Websites: +(b) Is this a +new system? -http://www.radiologyassistant.nl/en/p472458f15c55a/esophagus-part-i.html +(c) Is there a PIA that +covers this system? • - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +(d) Is the PIA available to the public? -MODIFIED BARIUM SWALLOW WITH LIMITED ESOPHAGRAM +F A S -21 +no -For this exam we are providing fluoroscopic assistance to the Speech Pathologist +1. Y e s . -as they evaluate the patient’s ability to swallow a variety of consistencies of foods +2 . No, because a PIA is not yet -coating in barium. +required to be completed at this time. -1. Introduce yourself to the speech pathologist as the radiology resident on +(e) Is a +System +Records +Notice +(SORN) +required for +this system? -service. +no -2. Discuss what the expected imaging findings are. +(I) Was a new or amended SORN +published in FY06? -3. Discuss the speech pathologist expectations of the study and whether or +5. No, because the system is -not a limited esophagram will be performed. +not a Privacy Act system of -4. Introduce yourself to the patient as the radiologist who will be assisting the +records. -Speech Pathologist for this exam. +(c) Is there a Privacy Impact Assessment (PIA) that covers this system? -5. Line up the patients head/neck under fluoroscopy, remember to collimate to +1. Yes. +2. No. +3. No, because the system does not contain, process, or transmit personal identifying information. +4. No, because even though it has personal identifying information, the system contains information solely about -the region of interest. +federal employees and agency contractors. -6. Make sure the patient’s eyes are NOT included in your field. +(d) Is the PIA available to the public? -7. Observe the Speech Pathologist giving the patient a variety of consistencies. +1. Yes. +2. No, because a PIA is not yet required to be completed at this time. +3. No, because the PIA has not been prepared. -As the radiologist, it is your job to note any instances of aspiration or +(f) Was a new or amended SORN published in FY2006? -penetration on the exam and what consistency of barium they occurred with. +NSF FAS 1-3-07 -8. Observe the patient for any other pathologies, such as premature slippage, +10 of 18 -delayed swallow etc. +FY08 Exhibit 300 -9. If there are any unexpected findings not related to a swallow study that you + 1. Yes, because this is a newly established Privacy Act system of records. +2. Yes, because the existing Privacy Act system of records was substantially revised in FY 06. +3. No, because the existing Privacy Act system of records was not substantially revised in FY 06. +4. No; the system is operational, but the SORN has not yet been published. +5. No, because the system is not a Privacy Act system of records. -feel would warrant more imaging. Please save shots and discuss them with +Section F: Enterprise Architecture (EA) (IT Capital Assets Only) -your attending at the time of interpretation. +In order to successfully address this area of the business case and capital asset plan you must ensure +the investment is included in the agency's EA and Capital Planning and Investment Control (CPIC) +process, and is mapped to and supports the FEA. You must also ensure the business case demonstrates +the relationship between the investment and the business, performance, data, services, application, and +technology layers of the agency's EA. -10. The Speech Pathologist may ask you to perform a limited Esophagram. +(1) Is this investment included in your agency's target enterprise architecture? yes (a) -a. This means we will observe the esophagus in ONLY ONE VIEW. If +If "no," please explain why? -there is pathology – please take an image to document. We will then +(2) Is this investment included in the agency's EA Transition Strategy? yes -suggest to the clinicians to perform a full diagnostic esophagram. +a. If "yes," provide the +investment name as identified +in the Transition Strategy +provided in the agency's most +recent annual EA +Assessment. +b. If "no," please explain +why? -11. Discuss the case with the Speech Pathologist upon completion and note any +The Financial Accounting System (FAS) is included as part of the Next +Generation Grants Management workstream in the EA Transition Strategy +submitted to OMB on February 28, 2006. -findings that he/she may add to your interpretation. +(3) Identify the service components funded by this major IT investment (e.g., knowledge +management, content management, customer relationship management, etc.). Provide this +information in the format of the following table. For detailed guidance regarding components, +please refer to http://www.whitehouse.gov/omb/egov/. -12. It is not necessary to review the images with your attending prior to sending +Agency +Component +Name -the patient. +Agency +Comonent +p - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Description -a. However, take down any notes that you feel are necessary for +Advice of BEP Maintenance of -discussion with your attending at the time of interpretation. +FEASRM +Service Type -22 +FEA SRM +Component +(a) -Suggested Reading: +Billing and -1. Matsuo, K et al. Anatomy and Physiology of Feeding and Swallowing – Normal +FEA Service Component Reused (b) -and Abnormal. Phys Med Rehabil Clin N Am. 2008 Nov; 19(4): 691–707. +Reused Service +Component Name -http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2597750/ +Reused Service Component +UPI - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Internal External +Reuse (c) -UPPER GI SERIES (Back) +Funding +Percentage +(d) -23 +No Reuse -Most Common Indications: Epigastric pain, anemia, heme+ stools, GERD, hiatal - -hernia - -Patient Preparation: NPO after midnight, or at least 4-6 hours. - -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast) - -*Please note that if this exam is being done in a patient who is post-operative or - -there is a question for perforation you will be using WATER SOLUBLE CONTRAST. - -Additionally, if the patient is at risk for both PERFORATION and ASPIRATION, a - -low osmolar intravenous agent can be administered orally without the risk of - -hypertonic pneumonitis.* - -Procedure: - -1. Start the procedure by explaining to the patient exactly what you are going +10 -to do today. +Charge Card +Module -2. Begin with the adult in the standing position facing you. Since many of our +FASTRAN -patients are post-operative or frail they will not be able to stand. In that +AP Log -case the study can be performed in the supine or reverse Trendelenburg +Funding +Allocations +Charge Card +Bill Processing -position. +Transaction +Processing +Module -3. Fluoro the chest and abdomen and remove any artifacts – such as clips and +Accounts +Payable/Prompt +Pay and Invoice +tracking -leads from the patient. +Management -4. Look for four things: +Credit / +Charge -a. Full stomach (cannot evaluate stomach if the stomach is full) +Management -b. Bowel obstruction +Financials -c. Free air +Management -d. Barium present in small bowel or colon +Expense +Management -5. ***IF WE ARE USING FIZZIES, PLEASE ADMINISTER THEM TO THE +Financial -PATIENT AT THIS POINT*** +Management - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Payment / +Settlement -a. Have the patient take the “gas crystals” like they are doing a shot, +No Reuse -into the back of their throats, and have them chase it with water. +No Reuse -b. INSTRUCT THE PATIENT NOT TO BURP!! The gas is what gives us +No Reuse -24 +3 -the contrast for mucosal evaluation. +10 -6. Ask the patient to take a mouthful of the “thick” barium” and watch the +7 -patient swallow. For the first swallow you do not have to obtain any images, +NSF FAS 1-3-07 -this is for you to understand the patient’s motility and ability to perform +11 of 18 -study. +FY08 Exhibit 300 -7. Now position the patient in the LPO position standing and have them swallow + Agency +Componen +Name -a mouthful of barium (3 swallows for your 3 shots). +FASTRAN -a. Shoot the proximal (cervical) esophagus. You may grab or shoot a full +FAS Core -exposure. +Agency +Component +Description -b. Shoot the mid thoracic esophagus. +FEASRM +Service Type -c. Shoot the distal thoracic esophagus. +FEA SRM +Component +(a) -8. You may need to instruct patient to drink more barium if you do not feel +FEA Service Component Reused (b) -there is enough in the stomach. +Reused Service +Component Name -9. Next we will move the patient into the supine position. +Reused Service Component +UPI -a. If we are using the “fizzies” (ie doing a double contrast exam) and +Internal External +Reuse(c) -looking at mucosa please do the following: +Funding +Percentage +(d) -i. Instruct the patient that we are attempt to coat their stomach +Transaction +Processing +Module -with barium (this is important if we are looking at the mucosa +Financial +Management -using the fizzies technique). +Debt +Collection -ii. We will ask the patient to roll through the left lateral position +Core +Functionally of +Accounting +System -to the prone postion and back (“roll like a cement mixer”). +Internal +Controls -10. Capture images of the stomach in the supine position. +Financial -11. Now we will obtain contrast filled images of the stomach. Please see the +Management -images below for guidance. +No Reuse -a. Shoot the cardia and fundus of the stomach +10 -b. Shoot the mid body of the stomach outlining the greater curvature +No Reuse -c. Shoot the antropyloric region of the stomach (if you miss this here +60 -don’t worry you’ll get it when you evaluate the duodenal bulb) +a. Use existing SRM Components or identify as "NEW". A "NEW" component is one not already +identified as a service component in the FEA SRM. - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +b. A reused component is one being funded by another investment, but being used by this +investment. Rather than answer yes or no, identify the reused service component funded by the other +investment and identify the other investment using the Unique Project Identifier (UPI) code from the +OMB Ex 300 or Ex 53 submission. -25 +c. 'Internal' reuse is within an agency. For example, one agency within a department is reusing a +service component provided by another agency within the same department. 'External' reuse is one +agency within a department reusing a service component provided by another agency in another +department. A good example of this is an E-Gov initiatiye service being reused by multiple +organizations across the federal government. -Picture Reference: +d. Please provide the percentage of the BY requested funding amount used for each service +component listed in the table. If external, provide the funding level transferred to another agency to +pay for the service. -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +4. To demonstrate how this major IT investment aligns with the FEA Technical Reference Model +(TRM), please list the Service Areas, Categories, Standards, and Service Specifications supporting +this IT investment. -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +FEA SRM Component (a) FEA TRM Service Area +Billing and Accounting -Although not pictured above, please obtain images of a contrast filled cardia, as +Service Access and -attempted in the first image (labelled two) above. +FEA TRM Service Category +Access Channels -12. Now we will watch the duodenal bulb (spade shaped). If performing a double +FEA TRM Service Standard +Other Electronic Channels -contrast exam (ie with fizzies). Please follow the instructions below +FEAService Specification (b) -a. Have the patient moved into the LPO position, and try to capture an +Billing and Accounting -air-filled duodenal bulb. It may help to initially place the patient in a +Delivery -left lateral decubitus position, and gently roll them into an LPO +Delivery Channels -position while coughing to capture an air-filled bulb. Note if you +Intranet -cannot obtain this come back to this LATER. +Billing and Accounting - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Billing and Accounting -26 +Service Access and -Picture Reference: +Delivery -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +Service Access and +Delivery -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Service Access and +Delivery -13. For the standard contrast filled duodenal bulb we will have the patient move +Service Requirements -into the RPO to Right lateral position. This encourages the contrast to move +Legislative / Compliance -through the stomach to the gastroduodenal junction and for contrast to +Service Transport -move through the duodenal C-loop. +Service Transport - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Billing and Accounting. -27 +Service Platform and -Picture Reference: +Support Platforms -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +Platform Dependent -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Windows XP -14. Once you see the C-loop of the duodenum place the patient in the supine +Billing and Accounting -position. +Billing and Accounting -a. The contrast will flow across the third and fourth portions of the +Billing and Accounting -duodenum and will cross the Ligament of Treitz (In a peds case you +Infrastructure -are done here!). +Service Platform and -i. Take this picture +Infrastructure -ii. You should also have a good picture of a contrast filled stomach +Service Platform and -and be able to evaluate the proximal small bowel +Infrastructure -15. Now test for reflux (please see guidelines for the Esophagram). +Service Platform and -16. Obtain a picture of the standing stomach and proximal small bowel as your +Infrastructure -final image (full exposure on adults). +Delivery Servers -17. *If there was a request to evaluate for distal stricture you can give the +Application Servers -patient the barium pill and demonstrate it passing through to the stomach. +Windows NT -18. Otherwise, review your images with your attending and repeat any portions +Database / Storage -that are needed. +Database -19. Discuss the preliminary findings with the patient if desired. +Sybase 12,x - Peds +Hardware / Infrastructure -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Servers / Computers -28 +Windows NT -Overall the peds upper GI is a quick study to typically rule out malrotation, +NSF FAS 1-3-07 -evaluate equivocal cases of HPS, duodenal atresia/web and/or diagnose GERD. +12 of 18 -We will always use the single contrast technique and fluoro grabs to obtain any +FY08 Exhibit 300 -pictures with the pediatric fluoro settings. + Billing and Accounting -We will examine the esophagus in the AP and lateral planes with fluoro grabs as the +Billing and Accounting -child/infant is drinking. Typically given patient size we can fit the stomach on our +Billing and Accounting +Billing and Accounting -views (without having to move the patient and obtain multiple pictures). The most +Component Framework Presentation / Interface -important part of the study is document the normal positioning of the ligament of +Static Display -Treitz. +Component Framework Data Management -Be careful when you have the baby in the lateral positioning the quickly move the +Database Connectivity. -infant supine to watch the contrast cross over the ligament of Treitz. +Sybase 12.x -Suggested Reading: +Component Framework Data Management -1. Rubensin et al. Double-Contrast Upper Gastrointestinal Radiography: A +Service Interface and -Pattern Approach for Diseases of the Stomach. Radiology 2008; 246 (1): 33- +Integration -48. http://pubs.rsna.org/doi/pdf/10.1148/radiol.2461061245. +Integration -2. Rubensin et al. Upper Gastrointestinal Tract: Imaging Techniques. In: +Reporting and Analysis +Enterprise Application -Rubensin et al. Abdominal Imaging. Springer Verlag 2013. +Integration -http://link.springer.com/referenceworkentry/10.1007%2F978-3-642- +Billing and Accounting -13327-5_13 +Service Interface and - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +lnteroperability -Small Bowel Follow Through (SBFT) +Data Transformation -Most Common Indications: Crohn’s Disease, heme+ stools, abdominal pain NOS +Integration -In peds: This study is often done in post-operative patients who have had bowel +Billing and Accounting -resected or due to atresias/strictures or webs. It is also performed in children +Service Interface and -29 +Interface -with gastroschisis repairs. +Service Description -*Please note this exam is different than the Gastrograffin Challenge which is +API -often ordered in the setting of potential bowel obstruction. The Gastrograffin +Credit / Charge -Challenge (which is a misnomer, as we do not use Gastrograffin) is a series of +Service Access and -abdominal x-rays which are taken after the administration of contrast (by the +Access Channels -tech) documenting the passage of contrast through the bowel.* +Other Electronic Channels -Patient Preparation: NPO after midnight +Integration -Materials: “Thin” and “Thick” Barium (or Water Soluble contrast can be used but +Interface -often gets “washed out” as it is not as dense as barium) +Delivery -Procedure: +Credit / Charge -This study is typically performed just following an upper GI. In the majority of +Service Access and -case CT or MR enterography is used when evaluation of ONLY the small bowel is +Delivery Channels -requested. +Intranet -1. Please see preceding protocol for how to perform an UPPER GI and obtain +Delivery -the proper scout radiograph. +Credit / Charge -2. We will begin our instruction after the contrast has crossed the ligament of +Service Access and -Treitz. +Service Requirements -3. Obtain radiographs (fluoro grabs in peds, although may consider spots in +Legislative / Compliance -children with known motility disorders to reduce radiation) in approximately +Credit / Charge -30 minutes intervals. If there is extremely slow transit as may be seen in +Service Access and -post-surgical infants you may increase this interval of time between +Service Transport -radiographs. +Service Transport - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Delivery -a. During the interval time the patient may go ahead and eat and walk +Credit / Charge -around the hospital to encourage bowel transit. +Credit / Charge -30 +Credit / Charge -4. Once the contrast has reached the level of the ascending colon, for an adult +Credit / Charge -or outpatient examination we will go and specifically exam the terminal ileum +Delivery. -(esp for those with Crohns). +Service Platform and -a. Obtain the balloon paddle and compress “spot” the terminal ileum +Infrastructure -(fluoro grab or spot the image). You may even see the appendix. +Service Platform and -5. Compress the small bowel in all four quadrants of the abdominal and make +Infrastructure -sure that there appears to be normal compressibility and that the bowel +Service Platform and -does not appear “stuck.” +Infrastructure -6. You should at least fluoro grab images of the four quadrants. +Service Platform and -7. Review all images with your attending. Upon his/her approval you may let the +Infrastructure -patient go. +Support Platforms -Suggested Reading: +Platform Dependent -1. Levine, MS et al. Pattern Approach for Diseases of Mesenteric Small Bowel +Windows XP -on Barium Studies. Radiology 2008; 249 (2): 445-460. +Delivery Servers -http://pubs.rsna.org/doi/pdf/10.1148/radiol.2491071336. +Application Servers - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Windows NT -POST-SURGICAL CASES +Database / Storage -31 +Database: -We see a wide variety of patients here at Shands who often undergo complex +Sybase 12,x -surgeries which need fluoroscopic evaluation for complications. The following are a +Hardware / Infrastructure Servers / Computers -series of post-operative cases that we do see and some helpful hints for how to +Windows NT -complete the requested studies. +Credit / Charge -For all surgical studies: REMEMBER to obtain a scout image of the area of study +Component Framework -prior to beginning any study. There are often devices and sutures that we need to +Presentation / Interface -be aware of. ALSO, upon completion of the study obtain a POST radiograph. +Static Display -Esophageal Studies: +Credit / Charge -• These are often requested after esophageal repair, stenting, POEM +Component Framework Data Management -procedure or gastric pull-through/gastric pull-up. The goal of these studies +Database Connectivity -is to evaluate for leak. Therefore, distention of the esophagus is paramount +Sybase 12.x -and the patient should be encouraged to take large gulps of contrast. +Credit / Charge -• The esophagus is typically dysmotile. However, leak and contained leak are +Component Framework Data Management -often noted as well. +Credit / Charge -• In this case, you will perform a single contrast water soluble exam of the +Service Interface and -esophagus with the post-surgical patient lying down on the fluoro table. +Integration; -• Attempt to view in the esophagus in at least the AP and Lateral planes. +Reporting and Analysis -However, oblique views may reveal subtle leaks if you are able to position the +. -patient. Obtain mag views if you do demonstrate a leak. Note any areas of +Enterprise Application +Integration -stricture or narrowing. If the narrowing is transient demonstrate a fully +Credit / Charge -distended esophagus. +Credit / Charge -Fluoroscopic Evaluation of Whipple: +Integration -Whipple's operation is most often performed in patients with ductal +Service Interface and -adenocarcinoma of the pancreas. The duodenum, part of or entire pancreas, and +Integration - distal common bile duct are resected. Mostly commonly, the common bile duct, +Service Interface and +Integration -pancreas, and proximal duodenum are anastomosed to a loop of jejunum. +lnteroperability -32 +Data Transformation -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Interface -Picture Reference: +Service Description / -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +API -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Interface -Diagram of the Post-surgical appearance of the Whipple +Expense Management -Remember to use LOCM contrast and attempt to distend the anastomoses. Obtain +Service Access and -views in the AP, lateral and oblique planes. If the patient has an NG tube +Access Channels -remember to remove the contrast from the stomach as these patients are an +Other Electronic Channels -aspiration risk. +Delivery - Fluoroscopic Evaluation of Gastric Bypass +Expense Management -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Service Access and -33 +Delivery Channels -Picture Reference: +Intranet -Left: Levine, MS et. al. . Imaging of Bariatric Surgery: Normal and Post-Operative Complications. Available at: +Delivery -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520. +Expense Management -Right:Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +Service Access and -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Service Requirements -Typically for gastric bypass, a pouch is formed at the cardia of the stomach. A +Legislative / Compliance -jejunal Roux-limb is brought through the transverse mesocolon. A side to side +Expense Management -anastomosis is gastrojejunostomy is created. +Service Access and -1. Obtain digital scout film of left upper quadrant. +Service Transport -2. In a standing lateral position, have the patient swallow one swallow of water- +Service Transport -soluble contrast and fluoro the lateral pharynx. If there is no aspiration, +Delivery -place patient erect AP and have them drink a few swallows of contrast, +Delivery -centering over left upper quadrant. Watch for emptying and extravasation. +Expense Management -Table spot image over pouch. +Service Platform and -3. Then place the patient in the supine position (make sure the patient does not +Support Platforms -exceed the table weight limit!). Obtain 3 to 4 digital spot images of the +Platform Dependent -gastric pouch and Roux-limb in several obliquities. +Windows XP - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Infrastructure -a. Do not magnify these images as most patients are so large that +Expense Management -exposure time would be long and result in motion artifact. +Service Platform and -34 +Delivery Servers -4. Don’t forget to get delayed images of the upper abdomen to confirm +Application Servers -contrast emptying into the distal jejunum, beyond where the Roux-limb +Windows NT -courses through the transverse mesocolon. +Expense Management -a. Is there a leak? This usually occurs at the gastrojejunostomy +Service Platform and -anastomosis or the blind jejunal stump. +Database / Storage -b. Is there obstruction? Usually at gastrojejunostomy, as the Roux-limb +Database, -traverses the mesocolon, or at the jejunojejunostomy, +Sybase 12,x -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520 +Infrastructure -Suggested Reading: +Infrastructure -1. Levine, MS, Carrucci, LS. Imaging of Bariatric Surgery: Normal and Post- +NSF FAS 1-3-07 -Operative Complications. Radiology 2014; 270 (2): 327-341. +13 of 18 -http://pubs.rsna.org/doi/pdf/10.1148/radiol.13122520. +FY08 Exhibit 300 -2. Swanson JO, Levine MS, Redfern RO, Rubesin SE. Usefulness of high- + Expense Management -density barium for detection of leaks after esophagogastrectomy, total +Service Platform and -gastrectomy, and total laryngectomy. AJR 2003;181:415-420. +Hardware / Infrastructure -3. Gervais, DA et al. Complications after Pancreatoduodenectomy: Imaging and +Servers / Computers, -Imaging-guided Interventional Procedures. RadioGraphics 2001; 21:673–690. +Windows NT -http://geiselmed.dartmouth.edu/radiology/pdf/complications_radiographics. +Infrastructure -pdf. +Expense Management -4. Hamath et al. Fluoroscopic Findings post-peroral esophageal myomectomy. +Component Framework -Abdominal Imaging 40 (2): 237-245. +Presentation / Interface -http://link.springer.com/article/10.1007%2Fs00261-014-0209-7 +Static Display - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Expense Management -35 +Component Framework Data Management - SINGLE CONTRAST ENEMA +Database Connectivity -Indication: Failed colonoscopy, stricture, mass, fistula evaluation (i.e. +Sybase 12.x -rectovaginal/rectovesicular), colostomy reversal. The ordering physician should +Expense Management +Expense Management -consult with the Fluoro Radiologist to discuss whether CT with rectal contrast or +Component Framework Data Management -fluoroscopic enema is the most appropriate exam given patient history and mobility. +Reporting and Analysis -The contrast agent (barium or dilute Omnipaque) is protocolled by the radiologist +Service Interface and -performing the fluoroscopic study. +Integration -Please note that Radiology will not perform a therapeutic “gastrograffin” enema, -and is not permitted to dispense MD Gastroview for this purpose, nor is MD -Gastroview ever used as a contrast agent in Radiology with the exception of the -Gastrograffin Challenge Small Bowel Study. +Integration -Patient Preparation: Requires a preceding bowel clean-out beginning the day before +Enterprise Application -the procedure and NPO on day of procedure. No prep is done in children. +Integration -Materials: Enema bag with attached catheter tubing with rectal tube tip, barium or +Expense Management. -water soluble contrast (in adults we use the Polibar kit) +Service Interface and -**Please note the techniques used for a double contrast enema do vary slightly. +lnteroperability -However given colonoscopy, CT/MR enterography these are rarely performed. The +Data Transformation -single contrast enema is also very rare – given the ease of CT and rectal contrast- +Integration: -as stated above.** +Expense Management -Procedure: +Service Interface and -1. Obtain a routine scout to evaluate stool burden and exclude any potential +Interface -free air. +Service Description / -2. Do a careful rectal examination – this will help you orient the proper +API -direction of the anal canal and rectum as a guide to inserting the enema tip. +Payment / Settlement - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Service Access and -a. This should also help you assess the presence of rectal mass or +Access Channels -stricture. +Other Electronic Channels -36 +Integration -3. Insert the tip of the enema with the patient in the left lateral decubitus +Interface -position with the knees drawn up towards the chest. +Delivery -4. Tape the buttocks together and ask the patient to hold their butt cheeks +Payment / Settlement -together for the entirety of the exam. +Service Access and -5. We will now cover the technique for an adult patient – single contrast barium +Delivery Channels -enema and review the necessary images. We will discuss the peds single +Intranet -contrast enema afterward with a concentration on differences in technique. +Delivery -a. Now allow the patient to roll into the supine position. +Payment / Settlement -b. Fill and distend the rectum with barium (shoot image) +Service Access and -c. Roll the patient to the supine LPO and add barium until the column +Service Requirements -reaches descending colon. +Legislative / Compliance -Picture Reference: +Payment / Settlement -Smith, JP et al. (2009). Guidelines for Residents in Gastrointestinal Radiology. Available at: +Payment / Settlement -http://www.uasom.uab.edu/PublicDocuments/Radiology/P&P/GI%20Protocols%20Revised_8-09.pdf +Payment / Settlement -d. Obtain a spot image of the sigmoid colon. +Delivery -e. Continue rolling the patient towards the right and obtain an image of +Service Access and +Delivery -the transverse colon – stop the barium. +Service Platform and -f. Then roll the patient RPO and spot the splenic flexure – stop the +Infrastructure -barium. +Service Platform and -g. Turn on barium and fill the ascending colon and look for reflux into +Infrastructure -the terminal ileum. Turn off the barium. +Service Transport -h. You may use the paddle to spot the terminal ileum. +Service Transport - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Support Platforms -6. For pediatric patients: +Platform Dependent -a. In peds – usual indication of this study is for evaluation of +Windows XP -Hirschsprungs’ or Meconium plug – so we will begin with evaluation of +Delivery Servers -the rectum in the left lateral position. +Application Servers -i. In the case of evaluation for meconium plug make sure that you +Windows NT -use Cysto-Conray (not barium!). +Payment / Settlement -37 +Service Platform and -Picture Reference: +Database / Storage -Neville, HL et al. Pediatric Hirschsprung Disease Workup. Available at: http://emedicine.medscape.com/article/929733- +Database -workup#c4 +Sybase 12,x -b. We will then obtain fluoro grabs of the same as the adult patients +Infrastructure -(but only grabs! No spots!) +Payment / Settlement -7. Once you feel certain that you have obtains all the images necessary, review +Service Platform and -them with your attending and respot any necessary. +Hardware / Infrastructure -8. Unclamp the barium and lower the enema bag so that the contrast drains by +Servers / Computers -gravity. +Windows NT -9. Obtain a post radiograph. +Payment / Settlement - Suggested Reading: +Infrastructure +Component Framework Presentation / Interface -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Static Display -38 +Payment / Settlement -1. Rubensin, SE et al. Double-Contrast Barium Enema Examination Technique. +Component Framework Data Management -Radiology 2000; 215:642–650. +Database Connectivity -http://pubs.rsna.org/doi/pdf/10.1148/radiology.215.3.r00jn36642. +Sybase 12.x -2. Levine, MS et al. Diagnosis of Colorectal Neoplasms at Double-Contrast +Payment / Settlement -Barium Enema Examination. Radiology 2000; 216:11–18. +Component Framework: Data Management -http://pubs.rsna.org/doi/pdf/10.1148/radiology.216.1.r00jl3311 +Reporting and Analysis, -3. Greenall, MJ et al. Complications of diverticular disease: A review of the +Payment / Settlement -barium enema findings. Gastrointestinal Radiology 1983;8:353-358. +Service Interface and -http://link.springer.com/article/10.1007%2FBF01948150#page-1 +Integration -Suggested Reading on Peds: +Integration -1. Berrocal et al. Congenital Anomalies of the Small Intestine, Colon, and +Payment / Settlement -Rectum. Radiographics 1999; 19 (5): 1219-1236. +Service Interface and -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.19.5.g99se041219. +Interoperability -2. Laya et al. Patterns of Microcolon: Imaging Strategies for Diagnosis of +Payment / Settlement -Lower Intestinal Obstruction in Neonates. J Am Osteopath Coll Radiol 2015; +Debt Collection -4 (1): 2-11. http://www.jaocr.org/articles/patterns-of-microcolon-imaging- +Integration -strategies-for-diagnosis-of-lower-intestinal-obstruction-in-neonates +Service Interface and +Integration -Websites: +Interface -Hirschsprungs +Enterprise Application -http://emedicine.medscape.com/article/409150-overview +Integration +„..,, +Data Transformation - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Service Description / -AIR CONTRAST INTUSSUSCEPTION REDUCTION ***PEDS ONLY*** +Interface' -39 +Service Access and -Indication: Intussusception seen on ultrasound, high clinical suspicion despite + Delivery -negative US +Access Channels -Patient Preparation/Radiology Prep: Obtain left lateral radiograph to exclude the +Other Electronic Channels -presence of intraperitoneal air, Make sure that surgical resident or fellow is +Debt Collection -present in the event of perforation. +Service Access and -Materials: Catheter tubing, in peds we use feeding tubes ranging from 5-8 Fr +Delivery Channels -depending on the patient size, sphygmomanometer +Intranet -Procedure: +Debt Collection -1. Insert the rectal tube and begin taping the buttocks. +Debt Collection -a. You can use 4x4 gauzes for additional reinforcement +Debt Collection -2. Obtain a scout image hold. +Debt Collection -3. Make sure the catheter tubing is well connected to the sphygmomanometer. +Delivery -4. You may pump the sphygmomanometer up to 120 mmHg in attempts to reduce +Service Access and -the intussuscepted bowel. +Delivery -5. Successful intussusception means seeing air in the distal ileum. +Service Access and -6. If intussuscepted bowel recurs – you may try up to 3 times or for 10 + Delivery -minutes. +Service Platform and, -7. Upon completion obtain a grab to document success. +Infrastructure -Suggested Reading: +Service Platform and -1. Del-Pozo, G. et al. Intussusception in Children: Current Concepts in Diagnosis +Infrastructure -and Enema Reduction. Radiographics 1999; 19: 299-319. +Service Requirements -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.19.2.g99mr14299 +Legislative / Compliance -2. http://emedicine.medscape.com/article/930708-overview +Service Transport -3. Applegate, KE. Clinically Suspected Intussusception in Children: Evidence- +Service Transport -Based Review and Self-Assessment Module. Journal of Roentgenology. +Support Platforms - 2005;185: S175-S183. 10.2214/ajr.185.3_supplement.0185s175 +Platform Dependent -http://www.ajronline.org/doi/full/10.2214/ajr.185.3_supplement.0185s175 +Windows XP -40 +Delivery Servers -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +!Application Servers - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Windows NT -FISTULOGRAM +NSF FAS 1-3-07 -41 +14 of 18 -Most Common Indications: Enterocutaneous Fistula +FY08 Exhibit 300 -Patient Preparation: None + Debt Collection -Materials: Small bore feeding tube, water soluble contrast +Debt Collection -Procedure: +Service Platform and -1. Use a small bore feeding tube to cannulate the fistula at the surface of the +Infrastructure -skin. +Service Platform and -2. Inject water soluble contrast under pressure. +Infrastructure -3. Make sure to turn the patient to document any fistula as well as the flow of +Database / Storage -contrast. +Database -4. Obtain spot images in supine, lateral and oblique views. +Sybase 12,x -Suggested Reading: +Hardware / Infrastructure -1. Lee, JK, Stein SL. Radiographic and Endoscopic Diagnosis and Treatment of +Servers / Computers -Enterocutaneous Fistulas. Clin Colon Rectal Surg. 2010 Sep; 23(3): 149–160. +Windows NT -http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2967314/pdf/ccrs23149.pdf +Debt Collection -2. Pickhardt, PJ et al. Acquired Gastrointestinal Fistulas: Classification, +Component Framework -Etiologies, and Imaging Evaluation. Radiology 2002; 224:9–23. +Presentation / Interface -http://pubs.rsna.org/doi/pdf/10.1148/radiol.2241011185 +Static Display - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Debt Collection -VOIDING CYSTOURETHROGRAM +Component Framework Data Management -42 +Database Connectivity -Most Common Indications: +Sybase 12.x. -Peds: Vesicoureteral reflux, multiple febrile UTIs, posterior urethral valves, +Debt Collection +Debt Collection -congenital abnormalities +Debt Collection -Adults: Post-operative evaluate for bladder injury, trauma, dysuria +Debt Collection -Patient Preparation: None. +Internal Controls -Materials: 5-8 French catheter tubing (PEDS), Foley Catheter, Cysto Conray or +Internal Controls -Omnipaque +Internal Controls -Procedure: +Internal Controls -1. Place the catheter/Foley using aseptic technique. +Internal Controls -2. Obtain a scout image prior to instilling contrast in the bladder (grab only for +Internal Controls -peds) +Internal Controls -3. Fill the bladder. +Internal Controls -a. For adults fill the bladder until the patient can no longer bear the +Component Framework Data Management -presence of contrast +Reporting and Analysis -b. Alternatively, when you are doing this in a pediatric patient you will +Service Interface and +Integration -use the cyclical technique of filling the bladder. +Service Interface and +Integration -i. IE the patient will pee around the catheter and you will fill the +Service Interface and +Integration -bladder a few times to evaluate for the presence of reflux. +Service Access and +Delivery -ii. Estimated bladder capacity in kids: (AGE + 2)30 = BLADDER +Service Access and -CAPACITY (in milliliters/cubic centimeters) +Delivery -4. Obtain images of the full bladder in the supine, oblique and lateral positions +Service Access and -(bilaterally). +Delivery -a. If you have any questions or concerns: STOP and discuss the case +Service Access and +Delivery -with your attending. In the next step we will remove the catheter. +Service Platform and -5. For adults: if the patient is able to urinate, you may remove the catheter +Infrastructure -prior to the urethra images. +Service Platform and - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Infrastructure -a. However, often times, patients with long term in-dwelling Foley +Service Platform and -catheters have difficulty urinating – so in those patients allow them to +Infrastructure -43 +Service Platform and -begin urinating and then remove the catheter. +Infrastructure -b. For males make sure the patient is in the oblique lateral position so +Integration -that all of the urethra is clearly seen. +Enterprise Application: -i. Make sure all the portions of the urethra are seen. +Integration -6. For little peds patients: +Interoperability -a. Since we do not have control over when the patient will urinate – we +Data Transformation -have to be very fast and quick with our imaging. +Interface -b. Make sure you are centered over the urethra (you only get one shot +Service Description / -once the catheter is out, otherwise we will have to re-catheterize the -patient ) +Interface -i. Make sure they are in the oblique lateral position to get +API -optimal visualization of the urethra +Access Channels -1. In males we are looking for posterior urethral valves! +:Other Electronic Channels. -c. As the patient urinates, QUICKLY remove the catheter and obtain +Delivery Channels -images of the contrast filled urethra. +Intranet -7. Try to have the patient urinate as much contrast as possible. +Service Requirements -a. Do not leave the patient with a large amount of contrast in the +Legislative / Compliance -bladder without notifying the ordering physician. +Service Transport -8. Obtain a post-void image (grab for peds). +Service Transport -9. Review imaging and findings with attending before releasing the patient. +Support Platforms -Suggested Reading: +Platform Dependent -1. Fernbach, SK. et al. Pediatric Voiding Cystourethrography: A Pictorial Guide. +Windows XP -RadioGraphics 2000; 20:155–168. +Delivery Servers -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.20.1.g00ja12155 +Application Servers -2. Yu, NC. et al. Fistulas of the Genitourinary Tract: A Radiologic Review. +Windows NT -RadioGraphics 2004; 24:1331–1352. +Database / Storage -http://pubs.rsna.org/doi/full/10.1148/rg.245035219 +Database -3. Cystography Technique Emedicine Description +Sybase 12,x -http://emedicine.medscape.com/article/1893772-technique +Hardware / Infrastructure - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +Servers / Computers -44 +Windows NT -RETROGRADE URETHROGRAM +Internal Controls -Most Common Indication: Penile/Urethral trauma, urethral stricture, hematuria +Component Framework Presentation / Interface -Patient Preparation: None +Internal Controls -Materials: Smaller bore foley catheter tubing 16-18 French, Omnipaque or +Component Framework Data Management -CystoConray +Static Display +Database Connectivity -Procedure: +Sybase 12.x -1. Position the tower over the patient’s pelvis with the center focused just +Internal Controls +Internal Controls -below the pubic bone. +Internal Controls -2. Flush the 16-F or 18-F Foley catheter with radiopaque contrast to remove +Component Framework Data Management -any air bubbles. +Reporting and Analysis, -a. The penile glans and urethral meatus should be cleaned with +Service Interface and +Integration -antiseptic. +Service Interface and +Integration -3. The Foley catheter is then placed just inside the urethral meatus so that +Integration -the Foley catheter balloon rests in the fossa navicularis. +Enterprise Application -a. Some attendings will use a red rubber catheter and/or foley and do +Integration -not inflate the balloon, but hold presseure at the glans to seal the +Interoperability. -tube. +Data Transformation -i. Ask your attending their preferred method. +Internal Controls: -4. With the Foley in position, fill the catheter balloon with 1-2 mL of +Service Interface and! -radiopaque contrast or saline solution. +Interface -a. Overfilling must be avoided, or it will rupture the distal urethra. (A +Service Description / -conscious patient can be asked if pain accompanies balloon filling). +API -b. Pull the penis laterally (or ask the patient to) to straighten the +Integration -urethra, grasping the penis as distally as possible, and distal to the +Interface -inflated balloon. +a. Service Components identified in the previous question should be entered in this column. Please +enter multiple rows for FEA SRM Components supported by multiple TRM Service Specifications. -5. Obtain a pre-injection scout image. +b. In the Service Specification field, Agencies should provide information on the specified technical +standard or vendor product mapped to the FEA TRM Service Standard, including model or version +numbers, as appropriate. - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +5. Will the application leverage existing components and/or applications across the Government +(i.e., FirstGov, Pay.Gov, etc)? no -6. The catheter-tipped syringe is then filled with approximately 50 mL of +NSF FAS 1-3-07 -radiopaque contrast, and 20-30 mL of contrast is injected in a retrograde +15 of 18 -45 +FY08 Exhibit 300 -fashion. + a. If "yes," please describe. -a. Consider obtaining a fluoro loop of the injection – the tech may help +6. Does this investment provide the public with access to a government automated information +system? no -you with the settings on the machine to do this. +(a) If "yes," does customer access require specific software (e.g., a +specific web browser version)? +[1] If "yes," provide the specific product name(s) and version number(s) of the required software and +the date when the public will be able to access this investment by any software (i.e. to ensure +equitable and timely access of government information and services). -7. Static images of the urethra are taken during retrograde injection of +NSF FAS 1-3-07 -radiopaque contrast. +16 of 18 -Picture Reference: +FY08 Exhibit 300 -Burks, FN et. al. Urethrogram. Available at: http://emedicine.medscape.com/article/1893948-overview + PART III: For "Operation and Maintenance" Investments ONLY (Steady State) -8. You should see a jet into the bladder. Then you have completely evaluated +Part III should be completed only for investments which will be in "Operation and Maintenance" +(Steady State) in response to Question 6 in Part I, Section A above. -the entirety of the urethra. +Section A: Risk Management (All Capital Assets) -9. Check images with your attending and obtain any other necessary images. +You should have performed a risk assessment during the early planning and initial concept phase of +this investment's life-cycle, developed a risk-adjusted life-cycle cost estimate and a plan to +eliminate, mitigate or manage risk, and be actively managing risk throughout the investment's life- +cycle. -10. When the exam is completely you may remove the catheter +Answer the following questions to describe how you are managing investment risks. -Suggested Reading: +1. Does the investment have a Risk Management Plan? yes -1. Urethrogram Emedicine website (see above) +a. If "yes," what is the date of the plan? -2. Kawashima, A. et al. Imaging of Urethral Disease: A Pictorial Review. +06/01/2005 -RadioGraphics 2004; 24:S195–S216. +b. Has the Risk Management Plan been significantly changed since last year's submission to -http://pubs.rsna.org/doi/pdf/10.1148/rg.24si045504 +OMB? no - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +c. If "yes," describe any significant changes: -3. Berrocal, T et al. Anomalies of the Distal Ureter, Bladder, and Urethra in +2. If there currently is no plan, will a plan be developed? -Children: Embryologic, Radiologic, and Pathologic Features. RadioGraphics +a. If "yes," what is the planned completion date? +b. If "no," what is the strategy for managing the risks? -46 +Section B: Cost and Schedule Performance (All Capital Assets) -2002; 22:1139–1164. +Answer the following questions about how you are currently managing this investment. -http://pubs.rsna.org/doi/pdf/10.1148/radiographics.22.5.g02se101139 +1. Was an operational analysis conducted? yes - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +a. If "yes," provide the date the analysis was completed. -FLUOROSCOPIC ASSISTANCE FOR HYSTEROSALPINGOGRAM +06/01/2005 -47 +b. If "yes," what were the results? (Max 2500 Characters) -Most Common Indications: Tubal occlusion, tubal patency, endometrial polyp, +The results of the OA recommend that the FAS continue to be used as the NSF's financial accounting -evaluation for mullerian anomalies +system for the next several years. The NSF is in the process of conducting a pilot project as a Grants -This procedure is done with the OB/GYN attending and it is your job to +Management Line of Business provider. We plan to take advantage of the results/findings of the GMLOB -provide fluoroscopic assistance and obtain all the view necessary to +process in becoming an SSP to more fully define our financial requirements. This will allow the NSF to take -document spillage or non-spillage of contrast from the Fallopian Tubes. +an integrated approach to both GMLOB & FMLOB in regards to the future of FAS and NSF's -The patient will be lying supine on the fluoro table. +financial system. -1. Obtain a scout image of the catheter once it has been inserted. +NSF FAS 1-3-07 -2. Obtain a picture in the early filling phase of the endometrium to assess +17 of 18 -for subtle filling defects as well as a later phase when the endometrial +FY08 Exhibit 300 -cavity is full of contrast. + c. If "no," please explain why it was not conducted and if there are any plans to conduct an -3. Spot pictures of the contrast spilling from the fallopian tubes +operational analysis in the future? (Max 2500 Characters) -a. Obtain pictures with the patient in the supine and bilateral oblique +2. Complete the following table to compare actual cost performance against the planned cost +performance baseline. Milestones reported may include specific individual scheduled preventative +and predictable corrective maintenance activities, or may be the total of planned annual operation +and maintenance efforts). Indicate if the information provided includes government and contractor +costs: -positions. +a. What costs are included in the reported Cost/Schedule Performance information (Government +Only/Contractor Only/Both)? Contractor Only -b. If you are ever unsure of contrast spillage from the fallopian +Description of Milestone (Max 50 +Characters) -tubes, please do a full exposure. Contrast is often easier to see on +Planned -fully exposed images. +Completion Date -4. Upon completion of the study confirm imaging findings with the OB/GYN +Total Cost +($M) -attending. +Actual +Completion Date Total Costs -Suggested Reading: +($M) -1. Simpson, WL et al. Hysterosalpingography: A Reemerging Study. +$1.800 -RadioGraphics 2006; 26:419–431. +$1.500 -http://pubs.rsna.org/doi/pdf/10.1148/rg.262055109 +09/30/2001 -2. Chalazonitis A. et al. Hysterosalpingography: Technique and Applications. +09/30/2002 -Current Problems in Diagnostic Radiology 2009; 38 (5): 199-205. +$1.800 -http://www.sciencedirect.com/science/article/pii/S0363018808000200 +$1.500 - 3. Behr, SC et al. Imaging of Müllerian Duct Anomalies. RadioGraphics 2012; +09/30/2001 -32:E233–E250. http://pubs.rsna.org/doi/pdf/10.1148/rg.326125515 +09/30/2002 -48 +09/30/2003 -University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +$1.670 - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +09/30/2003 -RADIOLOGIC PLACEMENT OF ENTERIC TUBE +$1.670 -WITH FLUOROSCOPIC GUIDANCE +09/30/2004 -49 +09/30/2005 -We may be asked to provide fluoroscopic assistance to clinicians in placement of +09/30/200 -enteric tubes – and even asked to place them ourselves. Please follow radiology +09/30/2007 -protocol (see document by Dr. Abbitt – hung above your work stations before +$1.300 -agreeing to assisting). +$1.300 -Please note we do not stock enteric tubes in the Fluoroscopic suite. +$1.500 -Below I will describe the placement of a nasojejunal tube with guidewire. +$1.500. -1. Lubricate the guidewire well and assure that the wire and catheter fit +09/30/2004 -together properly. +09/30/2005 -2. If there is oral anesthetic spray available (Benzocaine topic anesthetic) +09/30/2006 -spray the patient’s pharynx for about 1 sec duration. +$1.300: -a. Have the patient swallow the material. +$1.300 -3. Insert the guidewire into the tube leaving 10-20 cm floppy at the tip. +$1.500 -4. Lubricate the end of the catheter with viscuous xylocaine (2%). +$0.000 -5. With the patient sitting on the edge of the table, pass the catheter through +Baseline +Schedule +Variance +(#Days) -the nose. +Baseline +Cost +Variance +($M) -6. Have the patient flex their neck, chin to the chest and move the catheter +$0.000 -through the pharynx into the esophagus. +$0.000 -7. Continue pushing the catheter until you feel resistance. +$0.000 -8. Place the patient in the right lateral decubitus position and pass the tube +$0.000 -with fluoroscopic guidance. +$0.000 -9. Typically the catheter will pass into the antrum in the right lateral position. +$0.000 -10. Then move the patient into the LPO position so that the catheter can pass +$0.000 -into the pylorus and duodenal bulb. +FY01 Steady State Operations -11. Keep the tip of the guidewire at the pylorus, slipping the catheter over the +FY02 Steady State Operations -wire in the distal duodenum or proximal jejunum. +FY03 Steady State Operations -12. When the tip makes it to the duodenojejunal junction tape the tube in place +FY04 Steady State Operations -at the nose. +FY05 Steady State Operations -13. Document your success with a radiograph. +FY06 Steady State Operations - University of Florida Gainesville -GI/GU Fluoroscopy Guidelines +FY07 Steady State Operations -Note: Try not to exceed five minutes of fluoro time in your attempts. If the Fluoro +Total Planned Costs: -alarm notifying you of the five minute mark rings, you may complete the study and +Total Actual Costs: $9.070 -notify the ordering physician that the procedure was not successful although +NSF FAS 1-3-07 -50 +18 of 18 -multiple attempts were made. +FY08 Exhibit 300 diff --git a/q121/random_k4/random_4.md b/q121/random_k4/random_4.md index 602df279b63a2b7c5600ed9bc60d31b5343359ac..9a0d95de2dfad680593e31de73957a364bff2af5 100644 --- a/q121/random_k4/random_4.md +++ b/q121/random_k4/random_4.md @@ -1,2008 +1,530 @@ -INTERNATIONAL STUDENT -HANDBOOK +Summary Report +Records Management Inspections +Research and Development Records +FY 2018 - FY 2019 + +National Archives and Records Administration +May 5, 2020 + + _____________________________________________________________________________________ + +Summary Report +Records Management Inspections +Research and Development Records + +INTRODUCTION + +The National Archives and Records Administration (NARA) is responsible for assessing the +proper management of records in all media within Federal agencies to protect rights, assure +government accountability, and preserve and make available records of enduring value. Under 44 +U.S.C. 2904(c)(7) and 2906, NARA has the authority to conduct inspections or surveys of the +records and records management practices of Federal agencies for the purpose of providing +recommendations for improvements.1 The criteria for selecting agencies for inspection or records +management program review include, but are not limited to, the results of an agency's annual +records management self-assessment, the significance of certain records and the related business +processes, the risk of improper management of records, and the presence of important issues that +are relevant to the management of Federal records in general. + +In FY 2018 and 2019, NARA conducted a series of inspections investigating the management of +research and development (R&D) or scientific records at five Federal agencies. While only a +small sample, these five agencies are representative of the various types of research being +conducted by Federal agencies, the different approaches that are taken to scientific data +management and stewardship, and the incorporation of records management (RM) into agency +R&D operations. + +The following science and research centers were included in the inspection. For individual +inspection reports, see https://www.archives.gov/records-mgmt/resources/rm-inspections. + +● Department of Department of Commerce (DOC) – National Oceanic and Atmospheric + +Administration (NOAA) - National Centers for Environmental Information + +● Department of Health and Human Services (HHS) – Centers for Disease Control and + +Prevention (CDC) + +o National Center for Emerging and Zoonotic Infectious Diseases +o National Center for Health Statistics +o National Center for HIV/AIDS, Viral Hepatitis, STD, and TB Prevention +o National Center for Immunization and Respiratory Diseases +● Department of the Interior (DOI) – U.S. Geological Survey (USGS) + +o Central Energy Resources Science Center +o Colorado Water Science Center +o Geologic Hazards Science Center +o Geology, Geophysics, and Geochemistry Science Center + +● National Aeronautics and Space Administration (NASA) + +1 Records Management by the Archivist of the United States (44 U.S.C. Chapter 29), +https://www.archives.gov/about/laws/records-management.html#2904. +_____________________________________________________________________________________________ +Page 2 + + _____________________________________________________________________________________ + +o Ames Research Center +o Goddard Space Flight Center +o Jet Propulsion Laboratory + +● Tennessee Valley Authority (TVA)2 + +The long-term business need for R&D data poses unique challenges for preservation, access, and +eventual transfer of permanent records to the National Archives. The overall intent of these +inspections was to determine if agencies have essential policies, procedures and processes for the +creation, access and protection of R&D records and the eventual transfer of permanent R&D +records to the National Archives. NARA also sought to identify unique and common challenges, +risks, and trends that might be of interest to records management programs in other Federal +agencies that create and maintain R&D records. + +This report provides commitments from NARA and recommendations for action by Federal +agencies to mitigate records management risk and incorporate best practices, where appropriate. +Additionally, it presents NARA’s key observations common to these agencies for the benefit of +other Federal agencies that create and maintain R&D records. + +The following chart rates each of the participating agencies’ records management integration or +compliance with RM statutes and regulations for the main topics of this summary. Please note +these inspections were conducted on a subset of records (namely R&D) and of a selected sample +of participating agency centers and programs. These ratings do not represent the level of +compliance or RM integration for an entire agency RM program. For more details on each +inspected agency, see the respective report at the website listed above. + +2 While TVA has transitioned away from a robust internal R&D program to the outsourcing of much of its R&D +activities, the agency was included in this inspection series as a way to gauge the impact of this transition on +recordkeeping for R&D records. +_____________________________________________________________________________________________ +Page 3 + + _____________________________________________________________________________________ + +DATA MANAGEMENT AND RECORDS MANAGEMENT + +Data management and records management were the two overarching themes in this R&D- +focused inspection. Data management in a very basic sense involves collecting, maintaining, +processing, and analyzing data; publishing or otherwise disseminating results and data; and +preserving data for future use and re-use. Records management incorporates many of these same +activities, although not all data is a record. The major difference between the two is records +management legally differentiates between types of data by declaring it a record and legally +imposing disposition time frames and related specific instructions for destruction of temporary +records and maintaining permanent records. (In the case of Federal records, this means +transferring permanent records to the National Archives.) Within the data management theme, +data governance emphasizes formal authority and accountability to manage data as a valued, +organizational asset. As a result, there was a common shared framework to manage R&D +records. + +Based on this small sample of agencies, NARA found that agencies who create and maintain +R&D records manage the data effectively through established governance3 policies that comply +with Federal Government data management regulations and standards. Records management in +terms of creation, maintenance, and preservation is to some extent intertwined and embedded +into R&D project management through common procedural phase-based project methodologies, +and meeting general project requirements and best practices. In order to propose a project or +research study, gather and/or create data, and publish or release results for stakeholders, it is vital +for R&D staff to maintain an appropriate level of recordkeeping to document process and +maintain intellectual control over the records and data. Preservation of these records and data is +an important aspect of R&D project management, though, as was noted during this inspection +series, it often ended with the publication of results or dissemination of data to stakeholders. The +inclusion of instructions for disposition according to NARA-approved records schedules (a key +function of records management) was sometimes missing. + +The agencies inspected demonstrated varying levels of records management and data +management. Those utilizing more promising methods may serve as an example to agencies that +create and maintain R&D records, as well as to the larger RM community. NARA notes below +the key observations that include common challenges, risks, and trends. + +3 Information governance is the overarching and coordinating strategy for all organizational information. It +establishes the authorities, supports, processes, capabilities, structures, and infrastructure to enable information to be +a useful asset and reduced liability to an organization, based on that organization’s specific business requirements +and risk tolerance, https://www.arma.org/page/Information_Governance. +_____________________________________________________________________________________________ +Page 4 + + _____________________________________________________________________________________ + +KEY OBSERVATIONS + +Governance and Intellectual Control + +Data management and stewardship increase RM inclusion and collaboration with project +staff which contributes to the identification and management of R&D records, but plans +lack final disposition instructions that align with NARA-approved records schedules. + +The use of data management plans (DMP) or similar plans to manage R&D projects throughout +the project lifecycle is a common practice of research project management. Plans included a +variety of information such as a description of the data to be managed including formats, data +inputs and outputs, the project manager or data steward responsible for ensuring the proper +management of the data, metadata requirements, use restrictions, and methods for data access +and dissemination. However, not all plans specifically included final disposition instructions. +While the creation of DMPs was not required at all agencies, it is at the very least a best practice +that, with the incorporation of RM requirements, will help staff see the correlation between data +and records, and ensure records management is a part of the overall project planning and data +management process. + +It was policy at a couple of agencies that RM staff collaborate with R&D project managers +(sometimes referred to as gatekeepers) from the beginning stages of DMP creation through the +end of the project to ensure that records and data created and maintained throughout the project +lifecycle met recordkeeping requirements and were dispositioned according to NARA-approved +records schedules. Another agency had RM staff following up with project managers as projects +were winding down. + +Good governance and internal controls need to be in place to maintain intellectual control +of R&D records. + +Having intellectual control (i.e., knowing the types of records created, the content of the records, +how and why they are created, and ownership) is essential for putting records management into +practice and for meeting any agency’s mission. It provides efficiency in process and +accountability to stakeholders. For R&D projects, having intellectual control is a basic step in +being able to identify and retrieve files, complete projects, meet standards, and provide access +and preservation. + +At least two of the agencies inspected were not fully aware of what R&D records were being +created and maintained and where they were being stored. One of those agencies was also unsure +of records ownership due to lack of clarity on records management requirements in R&D +contracts. Records were often created in various locations without standard taxonomies and +naming conventions. There were obsolete file formats containing unknown data. Records +inventories and file plans either had not been created or were not being utilized properly. Having +all of these controls in place would aid staff in locating and re-using R&D records and data, +minimize redundancies in data creation, and facilitate disposition per NARA-approved records +schedules. +_____________________________________________________________________________________________ +Page 5 + + _____________________________________________________________________________________ + +Electronic Records Management + +Some agencies are struggling to manage unstructured R&D data, shared network drives, +and personal drives. + +All of the agencies were using some kind of electronic recordkeeping system for managing their +R&D records. One agency was using an Enterprise Content Management system. The other +agencies were using a variety of data management systems and electronic records repositories +based on a particular program’s or project’s needs. While there is still work to be done in this +area, some agencies were successfully and systematically applying classification schemes and +naming conventions as well as populating record metadata. Some were building approved +schedules into the systems, while others relied more on the RM staff to apply disposition. For the +most part, the R&D data and records maintained in these systems of record were being +appropriately managed. + +Where some agencies struggle is with the records and information that is being maintained +outside these systems on shared network and personal drives, and in office space and filing +cabinets. As mentioned previously, agencies are not always aware of what types of records are +being created and where, whether they are scheduled, and how or if disposition is being applied. +It’s this unstructured (and sometimes unaccounted for) data and records that agencies find hard +to get a handle on. Instituting internal controls such as inventories and file plans; RM policies +and procedures addressing these types of records; and access controls will aid in bringing +awareness to these records and improve compliance with Federal RM regulations. + +Email Management + +The majority of agencies are capturing email containing R&D records. + +Focusing on just one body of records, like R&D, can provide limited insight into email practices +as a whole. When dealing with large projects like R&D there are discussions, sharing of results, +and conclusions made within the content and context of email. In these cases, the emails are +records and may require more specific records scheduling than general email approaches. + +The majority of the agencies have either a NARA-approved Capstone schedule4 in place with +policies for managing non-Capstone officials’ email or an agency-specific schedule that covers +record email of staff working on R&D projects and studies. At least three agencies require R&D +staff to save project-related emails to project files. While the majority of agencies have email +scheduled, one agency still does not have an approved schedule covering email.5 Another +agency, while it has an approved Capstone schedule in place, is struggling to ensure + +4 NARA Bulletin 2013-02: Guidance on a New Approach to Managing Email Records, +https://www.archives.gov/records-mgmt/bulletins/2013/2013-02.html. +5 This agency has a draft form NA-1005 (Verification for Implementing GRS 6.1 (or Capstone GRS)) but has yet to +obtain internal departmental clearance. In the meantime, this agency is capturing all email and has policies in place +for the management of email. +_____________________________________________________________________________________________ +Page 6 + + _____________________________________________________________________________________ + +implementation across all centers. Having an approved schedule in place for email, and ensuring +its implementation enterprise-wide, is essential for appropriately dispositioning both permanent +and temporary R&D records and for mitigating the risk of unauthorized dispositions. + +Exit Clearance Process + +The majority of the agencies lacked adequate exit clearance processes to ensure the +safeguarding of records of departing staff including those associated with R&D projects. + +Federal regulations require that agencies include procedures to ensure that departing officials and +employees (including staff engaged in R&D projects) do not remove Federal records from +agency custody. The majority of the agencies lacked adequate exit clearance processes to ensure +the safeguarding of records of departing staff including those responsible for R&D contracts and +those associated with R&D projects. RM staff at two of the agencies were largely kept out of the +exit clearance process. They were often not notified of departing staff, and were not involved in +exit interviews or other exit processes. One of these agencies primarily placed the responsibilities +of ensuring the safeguarding of R&D records on the R&D project managers. A third agency did +not have adequate controls in place for staff who left official agency employment, but converted +to a different employment status (such as emeritus positions) so as to continue providing +scientific knowledge and project contributions to the agency. These individuals sometimes +continued to create and maintain agency records. The other two agencies have implemented +systematic processes and policies to ensure that records were handled properly and to prevent the +alienation or unauthorized disposition of records. It is essential that agencies ensure that records +are clearly protected from unauthorized access and removal by departing staff by including a +review of records as part of the exit clearance process. + +RM Program Management + +Most agencies had a network of designated staff with RM responsibilities that manage +R&D records. + +36 CFR 1220.34 requires agencies to assign RM responsibilities in each program (mission) area +to ensure incorporation of recordkeeping requirements and records maintenance, storage, and +disposition practices into agency programs, processes, systems, and procedures. Designating +staff with RM responsibilities within program areas across agencies, not just in units that create +and maintain R&D records, is vital for ensuring agency compliance with Federal records +management statutes and regulations, embedding RM across the enterprise in daily business +functions, and in assisting the Agency Records Officer (ARO) in providing RM assistance to +staff. + +Most agencies inspected have a network of designated RM roles that actively works with the +ARO in the management of their R&D project and other records. This is encouraging as our +findings from other agency inspections often show the lack of a well-structured and fully +supported RM network. In only one case did we find this was lacking, and we noted the issue in +a recommendation for follow-up action. Having this network in place is a vital foundational step +_____________________________________________________________________________________________ +Page 7 + + _____________________________________________________________________________________ + +to managing R&D records (and all agency records) appropriately. + +Records schedules for R&D/scientific records often do not meet current business needs and +require updating. + +Current records schedules are not just a regulatory requirement -- they also serve as the +foundation for any records management program. They enable agencies to enact the legal +authority approved by the Archivist of the United States to properly dispose of temporary records +and to transfer permanent records to the National Archives. + +The majority of the agencies have records schedules for R&D/scientific/research records that are +not meeting current business needs. Three of the five agencies are actively working to reschedule +these records, with at least two agencies holding focus group meetings to gather input on the +schedules. One agency had recently updated their R&D schedules to better meet their business +needs. The other agency was using schedules approved in 2010 that may need to be reviewed. It +is important to note that the agencies who are working to reschedule their R&D records are +taking a closer look at business processes, existing data dissemination practices, the re-use of +R&D data, and the long-term business needs of the records. Once submitted to NARA for +approval, the new schedules should help to improve implementation of dispositions for R&D +records. + +RM Policies and Guidance + +The majority of agencies have relatively current records management directives, policies, +and guidance to direct and aid agency RM and mission staff in the management of records. + +Up-to-date RM policy and guidance are foundational elements of any RM program. They +provide the requirements, direction, and support to enable agency staff to meet not only Federal +RM statutes and regulations, but also agency business needs. These are especially critical for +R&D staff as they often have additional scientific and research standards and data challenges that +need to be addressed to meet project requirements for preserving and disseminating project/study +data to the stakeholder public and international research community. R&D staff need the +additional guidance to ensure they can manage and preserve data long-term ensuring the ability +to leverage R&D data for future projects. + +The majority of agencies have relatively current records management directives, policies, and +guidance to direct and aid agency RM and mission staff in the management of records. For this +particular inspection series, mission staff included a variety of project management, scientific +and research roles, all of whom were either creating or maintaining R&D records and +information. RM handbooks, manuals, templates, and project and data management plans were +some of the products used to disseminate this information to both RM and mission/project staff. +The guidance varied in scope with some providing overall RM requirements for all records while +others were geared to specific R&D projects supplementing the agency-wide guidance. Some +guidance products included detailed recommended naming conventions and required metadata. +All agencies at the time of the inspections were in the process of reviewing and/or updating RM + +_____________________________________________________________________________________________ +Page 8 + + _____________________________________________________________________________________ + +policies, directives, guidance, and manuals to ensure they reflected the most current requirements +on managing records in all formats. + +RM Training + +The management of R&D records would be improved through role-based RM training for +R&D staff and mandatory annual RM training. + +Agency employees involved in R&D projects have unique challenges in the creation, oversight, +management, and preservation of records. It is important for staff to be able to determine the +relationship between scientific/research data and records; when their data is considered to be a +record; and how to address records management for the data after it has been published or +otherwise disseminated, and/or the project has closed, particularly when the data is re-purposed +for additional or other research or the project has no end date. R&D staff also have additional +challenges in the volume of data created or received that needs to be maintained, the highly +technical nature of the data, and the formats in which that data is created and maintained. + +Federal regulations require agencies to provide records management guidance and training to all +personnel (36 CFR 1220.34(f)). NARA Bulletin 2017-01: Agency Records Management +Training Requirements requires all staff, contractors, and volunteers to take annual RM training +specific to the agency.6 The bulletin also recommends, as a best practice, the development and +implementation of role-based RM training. The majority of the agencies had mandatory annual +RM training for all staff; however, at two of the agencies, RM training was being applied +inconsistently or was not required. While there is no Federal regulation that requires R&D staff, +specifically, receive role-based training, agency staff engaged in R&D work need a better +understanding of their unique responsibilities for managing R&D records. Acknowledging this, +two agencies had created, above and beyond the required annual agency-specific RM training, +role-based or task-based training (or were in the process of expanding their role-based offerings) +for staff who create and maintain R&D records. At another agency, one particular project had +implemented role-based training specific to that project; however, this was not an agency-wide +practice. The advantage of role-based training is it addresses the specific needs of the people +attending the training and they are more likely to use what they have learned. Agencies should +consider developing such training for specialized roles like R&D staff. + +RM Evaluations + +Most of the agencies were not conducting regular, formal comprehensive records +management inspections, evaluations, assessments or audits as required by 36 CFR +1220.34(j). + +Effective evaluations are an important part of any RM program. Comprehensive evaluations are +essential to measure the adequacy and effectiveness of agency policies, procedures, and guidance + +6 NARA Bulletin 2017-01: Agency Records Management Training Requirements, https://www.archives.gov/records- +mgmt/bulletins/2017/2017-01-html. +_____________________________________________________________________________________________ +Page 9 + + _____________________________________________________________________________________ + +and ensure that programs are in compliance with Federal regulations. Only one agency +demonstrated a systematic process for conducting formal, comprehensive RM reviews or +assessments. The other agencies were either not conducting RM evaluations, or they were being +done on an ad hoc and/or limited basis. The lack of records management evaluation programs +within the agencies is of concern. + +Any difficulties an agency is experiencing in the management of R&D records would, during the +course of a regular, formal, and comprehensive RM evaluation, be documented and subsequently +corrected. Without conducting formal inspections, documenting the results, and monitoring the +implementation of recommendations, it is difficult for records officers to know how records +management is practiced throughout the agency, to identify areas of non-compliance, and to +mitigate risks of non-compliance. Without any kind of evaluation of records management +program implementation, agencies cannot determine the effectiveness of their policies, +procedures, and internal controls, or have other assurances that their R&D records and data are +being created, maintained, made accessible, and preserved until eligible for disposition per +NARA-approved records schedules. -A resource guide published by the -Office of International Education -Albany State University -Albany, GA +CONCLUSION - TABLE OF CONTENTS -Notice to Students………………………………………………………………..……………………….. +This inspection series focused on the participating agency science and research centers that are +creating and maintaining R&D records. Many agencies that create and maintain R&D records +may use these same approaches for R&D and have these same challenges and risks. This +summary report documents trends and common challenges within these agencies that may be +useful to these and other Federal agencies. -Welcome Letter from ISA……………..…………………………………………………………..………….…………………. +As noted in this report, agencies are experiencing some successes and some challenges in the +management of R&D records. The use of Data Management Plans with the incorporation of RM +and records disposition is one example of a best practice that, if not already being utilized by +agencies creating R&D records, should be included in regular project and data management. +Role-based training for staff who create and maintain R&D records is another best practice that +would help them better understand the relationship between data and records and how to address +records management for the data throughout its lifecycle, from creation through publication and +dissemination. -Financial Documentation………………………………………………………….………………..…….. +Areas that need improvement include agencies regaining or maintaining intellectual control of +R&D records; conducting records inventories and creating file plans; using standard +classification schemes and naming conventions; updating R&D schedules so they meet current +business needs; keeping RM policies and guidance products for R&D records up to date; and +having staff assigned to RM roles to assist R&D staff in the management of their records +including disposition; among others. By making improvements in these areas, agencies will +strengthen their compliance with records management requirements; further integrate records +management into existing data management frameworks; enhance the preservation of R&D +records; help mitigate risks to R&D records; and better contribute to agency R&D missions. -Academic Life………………………………………………………………………………………………………….…………….. +_____________________________________________________________________________________________ +Page 10 -American Customs and Culture………………………………………………………………………………..………… + _____________________________________________________________________________________ -Immigration Status…………………………………….……………………………………………………………..……………. +RECOMMENDATIONS FOR EXECUTIVE ACTION -Automobiles…………………………………………………………….………………………..…………. +Based on the observations and data collected by the NARA inspection team, this report makes +the following commitments for action by NARA and recommendations for Federal agencies. -Emergencies……………………………………………………………………………………………………………………………. +What NARA will do: -English Language…………………………………………………………………………………………… +● Have policies and processes in place to support Federal agencies’ transition to fully -Entertainment………………………………………………..………………………………………………………………………. +electronic recordkeeping. -Health Care…………………………………………………………………………………………………… +● Continue its efforts to provide policy and guidance for electronic records -Housing……………………………………………………………………………….…………………………………………….…. +management, information stewardship, and governance. -Dining Services……………………………………………………………………………..….…………… +● Enhance its support of Federal agency records management officials with effective +policies, modern tools, and new services to support the transition to electronic +records. -Legal Concerns………………………………………….…………………………………………..……………………………… +● Develop Federal records management requirements and work with Federal and -Mail………………………………………..…………………………………………………………………… +commercial vendors to incorporate the requirements into software applications and +cloud offerings. -Money and Banking……………………..……………………………………….………………………………………………... +● Provide reasonable and independent assurance that agencies are complying with -Regent’s Test…………………………………………………………………………………….…………. +relevant laws and regulations. -Religious Life……………………………………..…………………………………………………….…………………………… +● Establish appraisal, scheduling, and pre-accessioning processes that reflect modern -Shopping…………………………………………………………………………..………………………….. +electronic records management. -Student Athletes ………………………………………….……………………………………………………………..……….. +● Redesign records management training to assist agencies in building a records -Social Security Numbers ……………………………………………..…………………….…………… +management workforce that is skilled in electronic records and data management. +● Provide policy and guidance on the creation and maintenance of records management -Taxes……………………………………………….…………………………………………………………………………………… +directives and policies. -3 +What agencies must/should do: -4 +The following general recommendations, if addressed, will improve the management of R&D +records in agencies:7 -5 +● Agencies must continue or expand the inclusion of records management as part of the -6 +use of data management, data stewardship, and lifecycle project management +methodology. (36 CFR 1220.34(e), 36 CFR 1224.10(d), 36 CFR 1232.14(a), and 36 +CFR 1236) -10 +● Agencies must establish and maintain internal controls for the management of R&D -12 +records. (OMB Circular No. A-123, Management's Responsibility for Enterprise Risk +Management and Internal Control) -18 +● Agencies must improve the management of unstructured data. (36 CFR 1236.24) +● Agencies must properly capture email records. (36 CFR 1236.22) +● Agencies must conduct evaluations, inspections or audits of their RM programs, -19 +including on the management of R&D records and data, to provide assurance that -21 +7 For specific and more comprehensive recommendations per each participant, see individual reports at: +https://www.archives.gov/records-mgmt/resources/rm-inspections. -21 +_____________________________________________________________________________________________ +Page 11 -22 + _____________________________________________________________________________________ -24 +records are being created, maintained, made accessible, and preserved until eligible +for disposition per NARA-approved records schedules. (36 CFR 1220.18) -26 +● Agencies must develop training programs based on agency policies, procedures, and +records schedules, and should develop role-based training to aid R&D staff in the +management of R&D records. (36 CFR 1220.23(f) and NARA Bulletin 2017-01) -27 +● Agencies should also be aware of and implement into their electronic records -28 +management the Universal Electronic Records Management Requirements, published +by NARA on August 4, 2017; Criteria for Successfully Managing Permanent +Records, published by NARA on March 16, 2018; and the Criteria for Managing +Email Records in Compliance with the Managing Government Records Directive (M- +12-18) (Success Criteria for Managing Email Records), published by NARA on April +6, 2016. -29 - -30 - -32 - -32 - -33 - -34 - -35 - -Telephone and Telegrams ……………………….……………………………………………………….. 35 - -Testing …………………….………………….……………………………………………………………..… - -Transportation ……………………….…………………………………………………………………………………………… - -Campus Contacts…………………………………………………………………………………………………………………… - -Checklists……………………………………………………………………………………………………… - -Services for International Students ………………………………………………………….………………………………. - -36 - -36 - -38 - -39 - -40 - - Office of International Education - -NOTICE TO STUDENTS -This guide should be used in combination with the ASU Undergraduate -Catalog/ASU Graduate Catalog and the RAM Student Handbook. The University -Catalog contains important information and regulations of the university that you -will need to know. You should obtain a copy of the Catalog (Undergraduate or -Graduate) and the RAM Student Handbook as soon as you begin your program -of study. - -3 - - Office of International Education - -Dear International Student: - -It gives me great pleasure to welcome you to the campus of Albany State University. I -will serve as your International Student Advisor. I want you to know that you are a -valuable part of the ALBANY STATE UNIVERSITY family like any other student and most -of all I want you to be successful in your studies. - -To help you reach this goal, I offer services that include providing you with important -information about your visa status and the benefits it includes. There are laws and -regulations you are required to follow in order to legally remain in the U.S. and I am -required to submit updates and other reports regularly and by specific deadlines to -confirm that. - -This guide is designed to help you locate the resources and services you will need while -you are at ASU and answer some of your questions. You are encouraged to ALWAYS -know your F-1 Visa Student Status and regulation involved. - -As an international student at ALBANY STATE UNIVERSITY you are required to meet -with me each semester of your matriculation. I encourage you to schedule -appointments so that we can meet to discuss and review options available to you at any -point of your educational endeavors at ALBANY STATE UNIVERSITY. - -In addition, please note that while you are matriculating as an International Student -ASU currently does not provide any financial assistance for you. It is your responsibility -to have the financial resources to continue your education. If for any reason you are -unable to continue to make financial contribution to your education endeavors you are -at risk of being in violation of the student immigration status. (please see the next page) - -I look forward to assisting you at ALBANY STATE UNIVERSITY and seeing you reach the -end of your program as focused and ready for the future as you are here today. - -Sincerely, - -Sue Ann Balch, J.D. -F1-Visa Student Advisor -sueann.balch@asurams.edu (Email) -229-430-7075 (Office Number) -229-434-7106 (fax Number) -205 Reese Hall - -4 - - Financial Documentation (Required Forms) -You can acquire these forms from the office of Recruitment and Admission - -*Albany State University International Student Financial Statement (To Be Completed by Student) - -*I-34 Affidavit of Support (To Be Completed By Sponsor, Affirmed Under Oath and Notarized) - -Original/Official Supporting Financial Documentation Must be: - -1. Written in English or have certified English translation attached and a -recognized transcript evaluation by a company such as WES or Silney. -2. Dated no more than six months prior to the first day of the semester you - -plan to attend. - -3. Clearly state the account balance in United States Dollar amount(s). -4. Must provide the following information: Date account opened; total - -amount deposited for the past year; -Tuition and fees -Living Expenses -Expenses of dependant -Other: Books & Health Insurance - -$14,600.00 -$9,000.00 -$0.00 -$4,000.00 -*$27,600.00 - -(*Fee changes yearly, 2016 cost listed above) - -5. All monies for tuition, health insurance, Room Board, Books & Supplies -and other fees must be must be paid by the first day of classes each -semester at Albany State University after student has been accepted. -This process must occur until the student has completed matriculation at -Albany State University - -How to set-up account? - -1. After student receives acceptance letter from - -undergraduate or graduate admissions, student will -begin the account set-up process. - -2. Student will mail fees for the year in United States - -Dollar Amount to: - -Albany State University -Financial Operations -Attn: Student Account Manager -504 College Drive -Albany, GA 31705 -3. Student will mail money certified or return receipt - -request or any secure means of mail delivery or pay -in person. - -ACADEMIC LIFE -ADVISORS -All students are assigned an academic advisor when they enroll in a degree -program. Your advisor will be a faculty member in your department. He or she -will help you choose which courses to take each semester. If you need to - -5 - - change your schedule in any way, you must get the approval of your academic -advisor. - -Academic advisors can better serve you when they understand your educational -and career plans. They can help you prepare for graduate school or for a job. -They can arrange work experiences as a part of your degree program if this will -help you. If you give your advisor a clear picture of what you plan to do, he or -she can guide you toward the most suitable courses and university activities. - -INDIVIDUALITY -Individuality is very important to Americans. This value is as important in the -classroom as it is in other parts of U.S. culture. Independent thinking is -expected of students in American classrooms. Interpretation, analysis, critical -thinking, and even challenging the professor may be expected of you in class. -You may be graded on your class participation. Usually, professors will tell you -at the beginning of the course if you are to be evaluated on class participation. -It takes time and practice to get used to speaking out in class, but it is often an -important part of the learning experience. - -REGULATIONS -Certain acts are considered to be academic dishonesty and are not permitted at -Albany State University. Submitting another person’s work as your own or doing -academic work for another person is not allowed. Academic dishonesty includes -copying exam papers, assignments, or term papers or allowing another person to -copy your work. If you use ideas or information published by other people in -papers that you turn in, you must give the writers credit by providing citations in -footnotes or references. Your professors will tell you how to do this. Students -who turn in work which is not their own will be disciplined by the professor and -possibly the university. It could lead to expulsion from the university. - -GLOSSARY OF “UNIVERSITY LANGUAGE” -CATALOG/BULLETIN -The university publishes an undergraduate and graduate catalog. These catalogs -are sometimes called bulletins. They include information on all aspects of -university life, including regulations governing academic life, course descriptions, -and a list of faculty members. You must follow the rules set forth in the catalog -for your academic program. It is an important resource. Read it carefully and -refer to it when you need information about academic offerings and -requirements. - -CLASS -Sessions or meetings of particular courses. These are listed in the Schedule -Report, issued each semester, which lists the days, times and venues the class -meets. - -COLLEGE -The University is divided into units known as “Colleges.” The Colleges at ASU -are the Colleges of Arts and Humanities, College of Business, College of - -6 - - Education, and College of Science and Health Professions. Unlike many other -countries, the term college in U.S. usage does not refer to high school or -secondary education. It is often used interchangeably with the term university. - -COMPREHENSIVE EXAMINATION -Examinations are required by most departments before a student can receive a -Master’s degree. The exams test the student’s knowledge of his or her entire -field of study. - -CORE CURRICULUM -The Core Curriculum is composed of a wide spectrum of general courses. It is -designed to give all students a broad base of basic knowledge and is required of -all students seeking bachelor’s degrees at ASU. - -COURSE -Subject Offerings. These are listed in the course catalog, by department. -Students usually take 4-5 courses per semester. The course listings are also -available through BANNER on the ASU website. - -CREDIT HOUR -Courses are given credit hour values. Most courses are 3 credit hours. This -means that the course meets for approximately 3 hours each week during the -semester. 1-hour course meets for 1 hour, etc. In order to be considered full- -time, undergraduates must enroll in 12 credit hours, graduate students need to -take 9 credit hours each semester. - -CURRICULUM -This refers to the program that you are following, and its requirements. There -are required courses and elective courses in most curricula. - -DEPARTMENT -Within each College are several Departments that specialize in a certain subject. -For example, the College of Education includes the Department of Teacher -Education and Department of Counseling and Educational Leadership. - -DROP-DATE -Students may choose to drop or withdraw from a course without penalty up to -the middle of the semester. However, you must remember that: - -1. As a foreign student, you must remain a full-time student at all times. -You must be enrolled for 12 hours at all times, unless it is a vacation -period which is the summer semester or you otherwise have -permission from the DSO in the International Office. Graduate -students must take 9 hours. - -2. Withdrawal hour’s associated with classes that are dropped due to -documented extenuating circumstances will be included in the 12 -hour total. - -7 - - 3. Students will receive a “WF” (This symbol indicates withdrawal with penalty) for course - -withdrawals identified after the 16 hour limit has been reached. -There are financial and G.P.A conditions associated with the “WF”. - -4. “W” (This symbol indicates that a student was permitted to withdraw without penalty) grades will be awarded to -students who withdraw from classes prior to the end of the schedule -change deadline and to those students who are authorized to -withdraw due to extenuating circumstances. - -After the midterm drop-date, special permission is required for course -withdrawal. The grade the student has earned up to the point of withdrawal will -be reflected on his or her transcript; however, it will not affect the GPA. In -addition, as a F1-Visa Student you are required to notify the International -student Advisor of any schedule changes you are making. - -ELECTIVE COURSE -These are optional courses. Each student must choose several of these courses. -The academic advisor will suggest specific electives. - -FINAL -This is the last, major examination of the semester. Usually, one week is set -aside at the end of the semester for final exams. A final is a long exam, which -sometimes lasts as long as two hours. Not all professors choose to give final -exams. Some will assign research papers or special projects instead. - -GRADE POINT AVERAGE -Your performance is evaluated with grades of A, B, C, D, or F. The grades A -through F are given a numerical value (A=4, F=0) and weighted by the number -of credit hours associated with each grade. Your grade point average is -computed by averaging the numerical values of the credit hours for all of the -courses you have completed. “GPA” is the common abbreviation for Grade Point -Average. - -GRADUATE STUDENTS -These are students who have already received a Bachelor’s degree or the -equivalent. The graduate degrees offered at ASU are the Specialist degree in -Education and Master’s degree in several disciplines. -MID-TERM -A Mid-Term is a major examination given in many classes in the middle of the -semester. This word also refers to the date signaling the end of the first half of -the semester. - -MINOR -A secondary, optional program of study in addition to the student’s major. - -SEMESTER -The school year at Albany State University is divided into two semesters. Fall -semester, which begins in August and spring semester, which begins in January, - -8 - - constitute the normal academic year. There is a full six-week summer session. -Summer terms are optional terms for foreign students. - -QUIZ -An informal examination, shorter in length than a regular examination. - -REQUIRED COURSE -These are courses which each student seeking a particular degree must take. - -RESEARCH PAPER -This is a major paper which can be from 5 to 25 pages in length. In order to -write this paper, a student must select an approved topic, somewhat narrow in -scope, and do research in the library. Papers are to be typed. There are -manuals in the bookstore to show you how to type it, or you can hire a -professional typist. - -SECTION -Some courses are offered in a number of sections, or classes, each semester, so -that a large number of students may attend. A course number, section letter, -and call number identify each section. - -SYLLABUS -A list of information about a course including the books you will use, the -professor’s name, office and phone number. Usually there is a general outline of -how the course will proceed, and sometimes assignments and dates of -examinations are included. - -TRANSCRIPT -A record of students’ courses and grades. An “official transcript” is sealed by the -issuing institution, and normally sent directly to the school or office requesting it. -It is used to document a student’s academic performance while attending -college. Universities, graduate schools, and some employers use this to -determine the desirability of accepting or hiring students. - -UNDERGRADUATE -These are students who are studying for their first degree at the university. -During the first year, students are called freshmen; the second year, -sophomores; the third year, juniors; and their last year, seniors. The -undergraduate degree is the Bachelor’s degree. - -9 - - AMERICAN CUSTOMS AND CULTURE -The culture and customs of the United States are difficult to define precisely, -because geographic areas and the people who live in them differ. Most of what -you will learn about the customs and culture of the people of the United States -will be by experience and through contact with people. - -While you are in the U.S., it is good to learn the customs of the people here. It -is important, however, that you maintain your own values and beliefs. If you’re -personal preferences or religious beliefs influence you not to do certain things, -do not be embarrassed to explain this. - -TIME -Being on time is very important in the American society. Classes, parties, -concerts, meetings, and other planned events all start at a specific time. If you -need to see someone, you should ALWAYS make an appointment beforehand. -Do not show up at their office or home unexpectedly. It is considered very -important not to be late to scheduled appointments or events. If you are late for -an appointment, it is important to apologize to the person(s) you have kept -waiting. Students who are always late for appointments or who turn in -assignments after the deadline earn negative reputations. Many professors will -deduct from a grade if an assignment is late or if a student arrives late for class. - -INFORMALITY -There is a great deal of informality between all types of people in the U.S. It is -our way of life. The American society does not have a formal class structure. -We may dress more casually and act more informally in social situations than you -may be used to. - -GREETING, INVITATIONS, AND VISITING -The most common forms of greeting are a handshake, smile and “Hello, I’m -pleased to meet you.” The Americans do not usually embrace in public, except -with members of their family or very close friends. - -Professors should be addressed as Doctor or Professor unless they request -otherwise. You should also use titles (usually Mr., Mrs., or Ms.) with older -persons unless they ask you to call them by their first name. For other students, -it is customary that you address them by their first name. - -If you are invited somewhere and you accept the invitation, it is VERY -IMPORTANT that you appear at the event. Do not accept an invitation simply -because you are afraid to say no. If you cannot or do not want to go to the -event, politely tell the person who invited you htat you cannot or choose not to -attend. It is much more inappropriate to accept an invitation and then fail to -appear than it is to decline the invitation in the first place. If invited to visit -someone’s home, you are not normally expected to bring a gift. It is always -acceptable, but not necessary, to bring a small token or an inexpensive gift to -your host or hostess. An expensive gift is usually inappropriate and may make - -10 - - your host or hostess uncomfortable or embarrassed. American homes, usually -does not have domestic servants, so if you stay as a houseguest, it is proper for -you to make your own bed or volunteer to help with other household chores. - -TABLE MANNERS -Table manners is most common to Americans including not speaking with food in -your mouth, keeping your elbows off the table, and keeping your hands in you -lap while you’re not eating. A knife, fork, and spoon are used unless “finger -foods” are being served. You should not belch at the table after eating; this is -considered rude in the U.S. After the meal is finished, the napkin should be laid -beside the plate and used silverware should be laid across the top of the plate. - -RELATIONSHIPS -Most Americans have many friendly acquaintances, but just a few close friends. -Visitors from other countries often remark that Americans are very friendly, even -at the first meeting, but that it is difficult to get to know them very well. -Sometimes this is true; although Americans are friendly to everyone, they are -also very private people and often have just a few intimate friends. - -Americans often say things like “let’s get together sometime” as a friendly way to -say good-bye, NOT as an invitation. If you would like to get to know someone -better, it is a good idea for you to invite them to go somewhere with you on a -specific day, at a scheduled time. Participate in as many social activities as you -can and you will meet and make many new friends. - -MEN AND WOMEN -American women are independent, and women and men are considered equal. -It is not appropriate to whistle at, touch, or make personal comments about -women whom you do not know EXTREMELY well. It is common for American -university women to live alone or with other women in apartments off campus. -Guests of both sexes are often invited to these apartments. Do not assume that -your hostess has romantic intentions simply because she has invited you to stop -by her apartment. - -There are no clear guidelines for dating Americans. If you want to get to know -someone, it is often best just to ask the person to join you for a snack or to get -together to study. It is important to remember, however, that if two people go -out on a date, it only means that they enjoy each other’s company. It does not -always indicate serious intentions or constitute a promise! If you begin to date -someone from another culture, it is important to remember that s/he simply may -not view the relationship in the same way that you do, especially regarding the -seriousness of the relationship or the desire to engage in sexual activity. - -11 - - It is most often the man who invites the woman out on a date, and customarily -the man pays all of the expenses unless the persons agree that it will be a -“Dutch Treat,” that is, each person pays for his or her own expenses. Women -sometimes ask men for dates. A woman who asks a man out should be -prepared to pay for the expenses. - -BATHING AND HYGIENE -Most Americans bathe or shower daily, and use deodorants and anti- perspirants. -Americans are not accustomed to body odors and find it offensive. Some -international visitors may want to adjust their bathing habits and accommodate -American preferences for the time they are here. - -CULTURE SHOCK -There are many differences between the American culture and the home culture -of many international students. Adapting to these differences can cause “culture -shock.” If you are having trouble adjusting to life in the U.S., or are -experiencing depression or physical or emotional problems, you may be -experiencing culture shock. The Office of the International Student Advisor can -provide cross-cultural counseling to help you, or refer you to a counselor on -campus free of charge. Counseling is always confidential. - -ADDITIONAL INFORMATION -If you would like additional information about American culture, please visit the -Office of the International Student Advisor. The office contains several -handbooks and guidebooks for foreign visitors in the United States. - -IMMIGRATION STATUS -THE U.S. CITIZENSHIP AND IMMIGRATION SERVICES (USCIS) is the branch of -the Federal government created to monitor the entry and exit of foreign -nationals. This branch of the government was formerly known as the -Immigration and Naturalization Service (INS). You will encounter INS officials as -you travel between the U.S. and your home country and when you apply for -special programs. The following information should help you understand the -regulations concerning student immigration status. Most ASU students have -either F-1 or J-1 immigration status. - -PASSPORT -Your passport must be valid at all times. When it is time to renew your passport, -contact your country’s embassy in Washington, D.C. to determine what -procedures you need to follow in order to extend your passport. - -VISA -There is a visa stamped in your passport, which permitted you to enter the -United States. A single entry visa permits you to enter the U.S. only once on -that specific visa. A multiple entry visa allows you to enter the U.S. as many -times as you like (after leaving on vacations or trips home) as long as you - -12 - - remain a full-time student. Student stays are valid for the “duration of status”. -This status is indicated by the mark “D/S’ in your passport and on your I-20/DS- -2019 and I-94. As long as you remain in full-time student status, you are in the -country legally, regardless of the expiration date on your entry visa. -If you leave the U.S. after your visa expires, you must go to a U.S. Embassy or -Consulate-General in your home country and obtain a new visa before you can -return to the U.S. It is difficult to obtain a new visa in a “third country.” It is -difficult to obtain new F-1 visas in Canada or Mexico, unless you are from those -countries. - -F-1 IMMIGRATION STATUS -If you are a full-time degree-seeking student or a Rotary student at ASU, you -have entered the U.S. with F-1 immigration status. This type of immigration -status is designed to allow you to obtain your degree and return home following -your graduation. F-1 status is a “non-immigrant” status. “Non-immigrant” -means that you are not moving to the U.S. permanently. This immigration status -is valid for “duration of status.” - -• You should carefully consider the dates of your authorized stay and -make sure you are following the procedures under U.S. immigration -laws. It is important that you depart the U.S. on or before the last -day you are authorized to be in the U.S. on any given trip, based on -the specified end date on your Arrival-Departure Record, Form I-94. -Failure to depart the U.S. will cause you to be out-of- -status. Additional information on successfully maintaining your -immigration status while a student or exchange visitor can be found -on the Immigration and Customs Enforcement (ICE) website. -• Staying beyond the period of time authorized by the Department of - -Homeland Security (DHS) and being out-of-status in the -United States is a violation of U.S. immigration laws, and may cause -you to be ineligible for a visa in the future for return travels to the -U.S. - -• Staying unlawfully in the United States beyond the date Customs and -Border Protection (CBP) officials have authorized--even by one day-- -results in your visa being automatically voided, in accordance with -INA 222(g). Under this provision of immigration law, if you overstay -on your nonimmigrant authorized stay in the U.S., your visa will be -automatically voided. In this situation, you are required to reapply for -a new nonimmigrant visa, generally in your country of nationality. - -• For non immigrants in the U.S. who have an Arrival-Departure - -Record, Form I-94 with the CBP admitting officer endorsement of -Duration of Status or D/S, but who are no longer performing the -same function in the U.S. that they were originally admitted -to perform (e.g. you are no longer working for the same employer or -you are no longer attending the same school), a DHS or an -immigration judge makes a finding of status violation, resulting in the -termination of the period of authorized stay. - -13 - - Staying Beyond Your Authorized Stay in the U.S. and Being Out of Status - -• You should carefully consider the dates of your authorized stay and make -sure you are following the procedures under U.S. immigration laws. It is -important that you depart the U.S. on or before the last day you are -authorized to be in the U.S. on any given trip, based on the specified end -date on your Arrival-Departure Record, Form I-94. Failure to depart the -U.S. will cause you to be out-of-status. Additional information on -successfully maintaining your immigration status while a student or -exchange visitor can be found on the Immigration and Customs -Enforcement (ICE) website. - -• Staying beyond the period of time authorized by the Department of - -Homeland Security (DHS) and being out-of-status in the United States is a -violation of U.S. immigration laws, and may cause you to be ineligible for -a visa in the future for return travel to the U.S. Select Classes of Aliens -Ineligible to Receive Visas to learn more. - -• Staying unlawfully in the United States beyond the date Customs and -Border Protection (CBP) officials have authorized--even by one day-- -results in your visa being automatically voided, in accordance with INA -222(g). Under this provision of immigration law, if you overstay on your -nonimmigrant authorized stay in the U.S., your visa will be automatically -voided. In this situation, you are required to reapply for a new -nonimmigrant visa, generally in your country of nationality. - -• For non immigrants in the U.S. who have an Arrival-Departure Record, -Form I-94 with the CBP admitting officer endorsement of Duration of -Status or D/S, but who are no longer performing the same function in the -U.S. that they were originally admitted to perform (e.g. you are no longer -working for the same employer or you are no longer attending the same -school), a DHS or an immigration judge makes a finding of status -violation, resulting in the termination of the period of authorized stay. - -What Items Do Returning Students Need? - -All applicants applying for renewals must submit: - -• A passport valid for at least six months; -• An application Form DS-156, together with a Form DS-158. Both forms -must be completed and signed. Some applicants will also be required -to complete and sign Form DS-157. Blank forms are available without -charge at all U.S. consular offices and on the Visa Services website -under Visa Applications Forms at the United States Citizenship and -Immigration Services (https://www.uscis.gov/). - -• A receipt for visa processing fee. A receipt showing payment of the visa - -application fee for each applicant, including each child listed in a -parent’s passport who is also applying for a U.S. visa, is needed; -• A new I-20 or an I-20 that has been endorsed on the back by a school - -official within the past 12 months. - -14 - - All applicants applying for renewals should be prepared to submit: - -• A certified copy of your grades from the school in which you are - -enrolled; - -• Financial documents from you or your sponsor, showing your ability to - -cover the cost of your schooling. - -Students Away from Classes More Than Five Months -Students in or outside the U.S., who have been away from classes for more -than five months, will likely need a new visa to enter the U.S. - -How long may I stay on my F-1 student visa? - -When you enter the United States on a student visa, you will usually be -admitted for the duration of your student status. That means you may stay as -long as you are a full-time student, even if the F-1 visa in your passport -expires while you are in America. For a student who has completed the course -of studies shown on the I-20, and any authorized practical training, the -student is allowed the following additional time in the U.S. before departure: - -• F-1 student - An additional 60 days, to prepare for departure from the - -U.S. or to transfer to another school. - -• M-1 student - An additional 30 days to depart the U.S. (Fixed time period, -in total not to exceed one year). The 30 days to prepare for departure is -permitted as long as the student maintained a full course of study and -maintained status. An M student may receive extensions up to three years -for the total program. - -As an example regarding duration of status, if you have a visa that is valid for -five years that will expire on January 1, 2001, and you are admitted into the -U.S. for the duration of your studies (often abbreviated in your passport or on -your I-94 card as "D/S"), you may stay in the U.S. as long as you are a full- -time student. Even if January 1, 2001 passes and your visa expires while in -America, you will still be in legal student status. However, if you depart the -U.S. with an expired visa, you will need to obtain a new one before being able -to return to America and resume your studies. A student visa cannot be -renewed or re-issued in the United States; it must be done at an Embassy or -Consulate abroad. - -FORM I-20 -Before you came to ASU, you were sent Form I-20 by the ASU Admissions Office. -This form is the one you use to obtain your F-1 student visa, and it is a very -important form. You should keep it safe with your passport. A Designated School -Official (DSO) must sign your I-20 whenever you travel outside the United -States, even if you are only going for a short time. If you change degree -programs (change your major), you will need a new I-20. New I-20 forms are -available from the International Admissions Officer on campus at 229-430-4646 -or visit the Office of the International Student Advisor for further assistance. If -you transferred to ASU, be sure to keep your ASU I-20 and the I-20 from your - -15 - - previous school together. Note the date of expiration in #5 on your I-20. If you -need to stay past this time to complete your studies, you must contact a DSO at -least 30 days before this expiration date to extend your stay as a student. - -FULL-TIME STUDENT STATUS (In-Status) -In order to maintain your full time status, you must attend school at least 9 -semester hours per semester for graduate studies and at least 12 semester -hours per semester for undergraduate studies. If you have any questions -about this regulation, please see the International Student Advisor. Students -who fail to maintain this requirement risk being placed on probation and could -lose university scholarships, including International Student Scholarships. To -avoid losing your full-time status, check with your advisor and/or Registrar -before you drop a class. - -VACATION -The Summer Term is considered a vacation period for all non-immigrant -students, as long as the student intends to register full-time for the next term. -You may enroll part-time, full-time or not at all during the summer term. Make -sure you check with the International Student Advisor before taking ANY -semester other than the summer semester off. You can go home for summer -any time you wish. - -VACATIONS INSIDE OR OUTSIDE THE USA - -(A) Inside the USA: You do not need permission of any kind from anyone to -travel anywhere within the USA. Travel between the 50 states is totally -unrestricted. Have fun—see the USA! - -(B) Outside the USA: Before leaving the USA, you must go by the office of a -Designated School Official (DSO) with your passport and I-20 and let the -DSO sign your I-20 and check your visa. Your I-20 must be signed for -you to re-enter the USA. You must have both the signed (or a new) I-20 -and a valid F-1 visa in your passport to re-enter the USA. - -If your visa has expired, you will need to go to a U.S. Embassy or Consulate- -General to get a new visa before you can re-enter the country. U.S. VISAS ARE -NOT ISSUED IN THE USA. If you have dependents that are on an I-20, we need -to check their passports and visas as well. - -PLEASE VISIT THE OFFICE OF THE INTERNATIONAL STUDENT -ADVISOR AT LEAST TWO (2) WEEKS BEFORE YOUR TRIP SO THAT -THERE WILL BE SOMEONE AVAILABLE TO SIGN YOUR DOCUMENTS. - -TRANSFER STUDENTS -If you are transferring from another U.S. college or university to ASU, you must -see the International Student Advisor within 5 days of your registration at the -beginning of the semester for completion of the transfer process, which must be -done to maintain your student status. - -EMPLOYMENT - -16 - - (A) ON-CAMPUS: As soon as you have registered for classes, you may work -on campus as a student assistant in the various departments if hired. . -Each department or facility has its own budget for student workers, so -you must go directly to them to seek employment. Once you have found -a job, you must register with the ASU Human Resources Department -(Department of Human Resources) in order to be paid for your work. To -register, you must have a Social Security Number. For more information -about Social Security Numbers, please see the section on Social Security -Numbers in this handbook. - -(B) OFF CAMPUS—ECONOMIC HARDSHIP: You must see the International - -Student Advisor before you can work off campus. Due to recent changes -in immigration laws, it is virtually impossible for foreign students to work -off campus. The only two exceptions are internships for academic credit -or extreme economic difficulty due to conditions beyond his or her -control. Students must apply to USCIS for permission to accept -employment off campus due to economic hardship, and the ISA can -assist you with this application. - -(C) OFF CAMPUS—PRACTICAL TRAINING: You must see the International -Student Advisor before you can work off campus. Curricular Practical -training (or an internship) is work directly related to your course of study -that will give you “hands-on” job experience that will help you get a -better job when you return to your home country. It can be granted only -after you have been at ASU for an academic year, and then only under -certain specific conditions. There are provisions for curricular practical -training before graduation and after graduation. - -Other than the above types of work, NO OTHER EMPLOYMENT IS PERMITTED -while you are in student status. You may not work off campus without -permission from USCIS and/or the ISA. If you do work off campus illegally, you -will lose your student immigration status and its benefits. Working illegally in the -United States is the most serious violation of non-immigrant status in the eyes of -the USCIS, and you WILL NOT be reinstated if you violate this regulation. - -INFORMATION UPDATES -Any change in address, phone number or marital status must be reported to the -Office of the International Student Advisor (ISA) and the Office of the Registrar -within 10 days. You must report the changes to both offices. If you do not -notify the ISA and the Registrar, you will not receive important mail from the -university. Please note that the Office of the Registrar maintains two addresses -on each student, a local address and a home address. The local address is your -address in Albany. If you live on campus, the university will automatically record -your ASU box number. If you live off campus, you must report your address to -the ISA and the Registrar. The home address is your address in your country. It -is the address you listed on your application for admission. All official -correspondence concerning your grades and your fees will be sent to the home -address. You may want to change your home address to your Albany address to -make sure you receive these notices. If you move off-campus, you must also - -17 - - complete a United States Postal Service Change of Address card and sign the -card. Completing this card will allow the campus mailroom to forward mail to -you at your new address. - -THE CARDINAL RULE -If you have questions about working or any other USCIS-related -matter while in the USA, visit the Office of the International Student -Advisor. The Office of the International Student Advisor and -Designated School Officials are here to assist you. We will help you as -much as we can; however, it is YOUR RESPONSIBILITY to keep the -office informed of your status and to ask for help when you need it. The -U.S. Immigration and Naturalization Services holds you accountable for -your own actions and expects you to follow all laws pertaining to your -immigration status. If you choose to violate the law or fail to educate -yourself about the law, you will be responsible, not the Office of the -International Student Advisor. Do not allow yourself to fall out of -status. Do not rely on the advice of other students for immigration -matters. You could lose the privilege of studying in the U.S. now, and -you could be prevented from returning to the U.S. in the future. - -A revision of the immigration law, the Illegal Immigration Reform and -Immigrant Responsibility Act of 1996, contains very harsh penalties of -students who violate student immigrant status, including provision to -prevent offenders from returning to the United States for periods of -several years. The new immigration regulations based on this act are -currently in place, but because the law is a new one, there are often -modifications or updates to the regulations. The International Student -Advisor will keep you informed about modifications in the regulations. - -AUTOMOBILES -REQUIRED INSURANCE -Law requires liability insurance for a car. If you own an automobile in Georgia, -you must purchase liability insurance. Collision insurance is not required, but it is -recommended. Proof of insurance, usually in the form of an insurance card, -must be carried in the car at all times. Look in the “Yellow Pages” of the phone -book to locate insurance agents, or ask a friend to refer you. Compare prices -with several companies before buying. - -DRIVER’S LICENSE -As a non-resident student at the university, you may drive in Georgia with a valid -international driver’s license. If you do not have an international license or if -your license expires during your stay, you may apply for a state of Georgia -license at the `Dougherty County Department of Motor Vehicle Safety (DMVS), -2062 Newton Road, Albany. The contact telephone number is 229-430-4258. -Getting a Georgia license requires a written test and a driving test. You should -present your passport, I-94, and your I-20 or DS-2019 to the examiner. You will -also need to bring a lease or utility bill, which verifies your address. If you live in - -18 - - a residence hall, you may also need a letter from the International Student -Advisor certifying that you live on-campus. The fee for a five-year license for -driving a passenger car is $20.00. A provisional permit costs $10.00. All fees -must be paid in cash. In addition, you will need a denial letter from the Social -Security Office and/or a Social Security Card. Also, you will need a letter from -your International Student Advisor before you go. - -Many students find that having a State of Georgia license helps them to obtain -lower rates for automobile insurance. Georgia identification cards are also -available at the driver’s license office for a fee. Both the license and the ID card -contain photographs of the bearer. Because the driver’s license and the state ID -card are the most widely accepted forms of identification, it is easier to verify -your address and age with these items than with your passport or ASU ID. -Having proper identification is especially important for writing checks and -handling other banking matters. You should consider obtaining an official form -of identification other than your passport. If you must carry your passport with -you at all times, you risk losing it. You should call the license bureau at 229-430- -4258 to make sure that the test is being given on the day you plan to apply for a -license. The bureau is closed on Mondays and will not give tests when it is -raining. - -If you hold a license from another state in the U.S. or from another country, you -will need to surrender this license to obtain a Georgia license. Please note that -once an out-of-state license has been surrendered to the State of Georgia, it is -not possible to reclaim the license. You should consider whether you wish to -surrender your home country license before you go to the bureau. If you have -questions about this regulation, please see the International Student Advisor. - -RULES OF THE ROAD -Although your international license is valid in the U.S., it is a good idea to read -the State of Georgia Driver’s Manual before you begin driving. Every country has -different traffic laws. Reading these laws in advance could prevent you from -having an accident. Copies of the State of Georgia Driver’s Manual are available -in the Office of the International Student Advisor. - -TRAFFIC VIOLATIONS -If you break any of the traffic laws, including driving over the speed limit, you -risk being stopped by a police officer. If a police or sheriff car follows you with -flashing lights and sirens, the officer wants you to pull over to the side of the -road. You should be prepared to pull over in a safe spot and do as the officer -instructs. Usually, the officer will ask to see your driver’s license, proof of liability -insurance, and registration for the car. He will also ask you questions about your -driving. If you have broken a traffic law, he may issue you a citation (ticket). -Citations involve paying fines, and in extreme cases, imprisonments. If you are -in an accident, you must wait at the scene of the accident for a police officer to -investigate the accident. Do not move the car and do not leave until the officer -has interviewed you about the accident. - -PARKING - -19 - - Most parking in Albany is free. Parking on the ASU campus requires a permit, -which costs $15 per academic year. Please refer to the ASU Parking and -Transportation resource guide for parking regulations. If parking regulations are -violated, you will get a “ticket” and be required to pay a fine. Parking Rules on -campus are clearly stated in the RAM Student Handbook. - -RENTAL AND PURCHASE -If you have a valid driver’s license and are at least 21 years old, you will probably -be able to rent or lease a car. Some rental places also require a credit card in -your name or that you are at least 25 years of age. Check with the larger local -car dealerships about short- and long-term rental. - -If you wish to buy a car for your stay in the U.S., be CAREFUL. Most companies -will service your car if it needs repair after purchase. You must pay for all -repairs unless you have a “manufacturer’s warranty” or paper which states that -the maker of the car will pay for repairs for a limited period of time after -purchase. If you buy a used car from a private dealer or individual, check the -vehicle carefully before purchase. It might help to bring a friend who knows -about the mechanical working of an automobile to inspect the car for you. An -honest car dealer will be happy to let you inspect all parts of the car and -remember to ask questions before you buy. When purchasing a car, you must -have a TITLE and LICENCE PLATE (tag) from the State of Georgia. The dealer -where you purchased the car will give you the correct papers to fill out and help -you obtain these things. In addition, you must renew items such as insurance, -license and tags on a yearly basis. If you buy a car from an individual, be sure to -ask for assistance with these items. If you are bringing a car to Albany from -another state, you must transfer the title for the car and obtain license plates -from the State of Georgia. You should also notify your insurance company that -you have moved. - -License plates must be renewed each year before paying the tax on your car. -Once you have paid the tax, you will be given a sticker for that year. The sticker -should be placed in the upper right hand corner of the license plate. If you have -questions about license plates, please contact the Dougherty County Tag Office, -225 Pine Avenue, Room 104, Albany, GA. The telephone number is 229-431- -3255. - -EMERGENCIES -To call the POLICE, AMBULANCE, or FIRE DEPARTMENT, dial 911. Call this -number only if someone needs immediate help. This is a special emergency aid -number and needs no other numbers. For medical problems that are not life- -threatening emergencies, call the Palmyra Medical Centers, 2000 Palmyra Road, -Albany. Tel.:229- 229-434-2000 or Phoebe Putney Memorial Hospital, 417 3rd -Avenue, Albany. Tel.: 229-883-1800, 229-312-1000. Police concerns are handled -according to location: - -Campus Police: -Albany Police: -Leesburg Police: - -229-430-4711 -229-431-3266 or 229-431-2100 -229-759-6012 - -20 - - Albany Fire Dept.: - -229-431-3266 - -ENGLISH LANGUAGE -Students who scored below 523 on the TOEFL (Test of English as a Foreign -Language) are REQUIRED to take special supplementary English as a Second -Language (ESL) courses. These courses are non-credit-bearing, but they will help -you with your academic English needs. The course numbers are ENGL 0075, -0077, 0098 and 0099. Any international students are welcome to take these -courses, even if the TOEFL score was above 523. - -English 1101 is the introductory university level English course required for all -ASU degree programs. The course prepares students for the remaining required -English courses in their degree program. - -There are some options available for students who wish to improve their English -language skills. For students who need help with their writing skills, the English -Department offers tutoring in the Writing Center lab located in Simmons Hall. -Another option for tutoring is the PEER TUTORING program. Tutoring is available -in most subjects free of charge during the fall and spring semesters. Contact the -Office of the International Student Advisor or your English professor to find out -which of these facilities will best meet your language needs. - -ENTERTAINMENT -ASU provides many entertainment and social opportunities for students. The -student centers on both campuses are available to you for all activities. It has as -its objectives the fulfillment of the cultural, educational, and recreational desires -of all students of ASU. Various programs and activities are planned and -supervised by the Director of Student Activities and the program coordinator to -round out and broaden the experiences of each student. The services and -facilities of the Student Union are open from 10.00 a.m. to 7.00 p.m. Monday -through Friday. On Saturdays and Sundays, the Game Room hours will be posted -in the game room, unless other wise notified. The facilities and services of the -Union are for students to make full use of. They include conference and meeting -rooms, lounges, a game room, Pizza Hut, the ASU Bookstore, Student -Government Office, a Post Office, an ATM, an information desk, vending -machines, and the ASU Board of Managers. - -The Albany State University International Student Association (ASUISA) is a -special club for international students at ASU. This group plans and promotes -cultural diversity and global awareness activities involving ASU and the -community. It also organizes programs and field trips for students who are -interested in more international activities. Look for information about the first -meeting in late August. - -For more information about clubs and activities, you should read the campus -paper, The Student Voice, and listen to the community radio stations. Also plan -to attend the campus-wide activities fair that takes place during the early part of - -21 - - the Fall Semester. Each of ASU’s clubs will have a booth at this fair to provide -more information for potential members. - -HEALTH CARE -The United States philosophy of health care is different from that of many other -countries. Health care in this country is viewed as an industry. All Health Care -professionals are expected to be paid by their patients or clients. Patients often -must prove ability to pay before treatment will be given. Doctors in this country -usually manage their offices so that they spend a minimum amount of time with -each patient. Nurses do as much as possible to help free the doctor to see more -patients. Expect to deal with nurses in the office and over the telephone. - -MEDICAL EMERGENCY -In case of medical emergency involving a life-threatening situation, or to request -an ambulance (emergency transportation to the hospital), call 911. Call this -number only in true emergencies. ASU students should not use the emergency -room services at the local hospitals, Phoebe or Palmyra, unless the medical -problem is life threatening. Students should use the university health center -infirmary as their main source of medical care. If you do not know whether the -situation is a true emergency, it is probably best to call campus police and have -the officer escort you to the health center. The nurse will evaluate the problem -and will send emergency cases to the hospital. THE HEALTH CENTER IS THE -CLOSEST MEDICAL FACITLITY TO THE RESIDENCE HALLS. Please note that -using the ambulance and the emergency room at the hospitals is very expensive. -This expense will not be covered by medical insurance if the situation is not a -true emergency. - -STUDENT HEALTH SERVICES -The University Health Service of Albany State University is located in the -Academic Administration (ACAD) Building, Room 138. Student Health Services -provide opportunities for safeguarding the health of all students at ASU. The -staff provides the following through student health fees: - -1. General medical care. -2. Confinement in the Health Services. Overnight stays are not available. -3. Consultation on general health needs. - -The health fee DOES NOT cover expenses for special medicines and treatments, -visits to doctors’ offices, operations or off-campus hospitalization. - -The University Health Services office is open Monday through Friday, 8:00 a.m. – -5:00 p.m. Our Doctors’ hours are Monday through Friday, 4:00-5:00 p.m. -Emergency calls are possible at any time. Each student, freshman or transferee, -is required to have on file a medical record completed and signed by a physician. -These records will be good for four years provided the student is in continuous -attendance at Albany State University and has consulted with the University -Physician periodically. - -22 - - DENTAL CARE -If you need to have your teeth cleaned or another dental service, you should -look in the telephone book Yellow Pages under “Dentists.” Phone to make an -appointment before you go. - -EYEGLASSES AND CONTACT LENSES -If you need an eye examination, or you want your eyeglasses or contact lenses -changed, you should look in the telephone book Yellow Pages under, “Opticians -and Optometrists.” Phone to make an appointment before you go. For medical -concerns with your eyes, look under “Ophthalmologist.” - -COUNSELING -Sometimes individuals become depressed, lonely or fearful in a strange -environment. Many overseas students come from a culture very different from -that of the U.S., and they often miss the support of their family and friends at -home. If you are feeling emotionally upset, depressed, or just need someone to -talk to, contact the International Student Advisor. S/he can make helpful -suggestions, listen to your concerns, and may refer you for further counseling. - -The primary function of the Counseling Services is to assist students in their total -development by providing services and programs to facilitate their intellectual, -emotional, social and academic growth. Counseling services also aid students in -determining a purpose and direction for their lives and in developing coping skills -and behaviors to aid them in working more efficiently with others in their -environment. - -PAYMENT -Health care in the United States is not free. Patients must pay for the services of -doctors, dentists, laboratory workers and hospitals. Their fees are often very -expensive. Care at the Health Services on campus is paid for by student health -fees each semester, so go to the Health Center first for any illness. If you have -an appointment with any off campus doctor or dentist, be prepared to -pay the bill at the conclusion of the appointment. Insurance will not -always cover 100% of the cost. Medical insurance does not cover the expense -of dental visits or eye exams and glasses. Students may want to plan to take -care of routine visits to these doctors during vacations at home. - -INSURANCE -Due to the high cost of medical care in the U.S., all foreign students -must purchase health insurance before enrolling for classes at ASU. -ASU is pleased to offer one policy for all ASU foreign students student. -Since September 1994, new regulations from the U.S. government -have determined what type of insurance coverage students must have. -The ASU policy has been designed to meet this minimum level of -coverage at a reasonable price. The plan covers portions of hospital -costs, doctor visits, and medication but not dental visits or eyeglasses. -Insurance to cover all costs at 100% and to cover dental visits and -eyeglasses is extremely expensive. Few U.S. residents carry such -coverage due to the high cost. The insurance system in the U.S. is - -23 - - designed for the patient to pay a portion of the cost in addition to the -insurance premium already paid. Insurance premiums are the fees paid -for insurance coverage. The university will bill all ASU international -students for the insurance premiums. Insurance from another source is -generally not accepted, since the coverage must meet U.S government -standards. If you need information about the ASU health insurance -program, see the International Student Advisor. If you do not pay your -premium, you will lose your insurance coverage and you will have a -hold placed on your student account, which will prevent you from -registering for classes. You should be careful to avoid a situation -where you lose your coverage because you never know when an illness -or accident may occur. If you have an illness or an accident and you -have no insurance, you will be responsible for 100% of the cost of your -treatment. For Student Health Insurance, please visit the Student -Health Insurance website. www.studenthealthinsurance.com - -GOOD HEALTH -The demand of university life and adjusting to a new environment can be very -stressful. Whenever a person is under stress, s/he is much more susceptible to -illness. Be sure that you take steps to protect your health. Eat properly, get -plenty of sleep, exercise, and protect yourself from communicable diseases, -especially sexually transmitted diseases such as AIDS. If you think you have a -health problem, seek help immediately. If you have questions about any health -matter, please visit the Student Health Center or the Counseling Center. - -IMMUNIZATIONS -All students in the state of Georgia must be vaccinated against measles, mumps, -and rubella. The vaccination can be obtained in one injection called an MMR. It -is recommended that you also receive a shot for Meningitis as well as a -Tuberculosis test. The vaccines are required because these diseases are -present in the U.S. You must protect your health while you are in the U.S. -Students must receive two doses (injection) of this “MMR” vaccine at least 30 -days apart. These vaccines are available at the Dougherty County Government -Office at 222 Pine Avenue, Albany. The telephone number of the Health -Department is 229-430-6200. Students who do not receive these vaccines will -not be allowed to register for classes. We prefer that you send proof of -immunizations and TB test as part of your application package. - -HOUSING -ON-CAMPUS HOUSING -The residence halls provide convenient, inexpensive housing on campus for -students. Most students will share a room with another student (roommate) and -a bathroom with other students nearby. Students must sign a room contract and -pay an application fee and deposit in order to live in a residence hall. The exact -amount to be paid will be included in your letter of admission. The contract -obligates you to live in the residence hall for the entire nine-month academic -year (2 semesters). The on-campus apartments are open all year. They are, - -24 - - however, in great demand. If you want to live in an on-campus apartment, stop -by the Office of Housing for more information or call 229-430-4741. - -If your room needs repairs or you have questions and concerns about any aspect -of life in the residence hall, you should contact your Resident Assistant. The -Resident Assistant, or RA, is also a student who lives in the hall, but this student -is an employee of the Office of Housing. The RA enforces the rules of the -residence hall such as quiet hours. The RA also handles issues such as changing -rooms or roommates and reporting when repairs are needed. The Residence -Hall Director supervises the Resident Assistant of the residence hall. Your -questions and concerns about life in the residence hall should be addressed to -the RA or the Residence Hall Director. Of course the staff of the Office of -Housing and Residence Life will be happy to assist you as well. The telephone -number is 229-430-4741. - -During recent years, the demand for on-campus housing has been high. It is -very important for you to read carefully any correspondence from the Office of -Housing. You need to follow specific procedures to reserve your room each year. -Please make a note of all deadlines and procedures. - -ROOMMATES -Living with an American roommate can be an excellent way to practice English, -learn about the U.S. culture and customs, or to meet a new friend. Living with -another person requires patience, tolerance, and kindness, especially as you get -to know one another during the first few days. It can also be one of the most -enjoyable aspects of your stay on campus. You should expect to disagree with -your roommate about some things. This is normal. The best way to solve your -differences is to speak honestly to one another. Americans are accustomed to -direct discussion of problems. If there are problems, which cannot be solved -between the two of you, you may go to your Resident Assistant (RA) who will -work with you to assign a new roommate. - -OFF-CAMPUS HOUSING -Off-campus housing is plentiful in Albany. There are many apartment complexes -and rental houses near campus and around town. Listed below are a few -suggestions of places to look for housing: - -1. - -2. - -3. - -4. - -Albany Herald, the local newspaper, lists apartments, houses, and -rooms for rent in the classified advertising section of the paper. The -ad will usually list the monthly rent as well as a phone number to call -for more information. -Bulletin boards and notices on campus: check classroom buildings for -notices and ads about apartments for rent. Many students looking for -roommates advertise in this manner. -The telephone directory: look under “Apartments” in the Yellow -Pages. -The Office of the International Student Advisor has listing of apartment -complexes in Albany, which lists such information as rental cost, size, -and distance from campus. - -25 - - RENT AND DEPOSITS -You will probably have to pay a deposit of two or three hundred dollars, plus -your first month’s rent before you can move into an apartment or rental house. -When you move out of the apartment you will get the deposit back if there has -been no damage to the apartment. You will also be required to sign a lease. A -lease is a legal contract that states the amount of time that you will live in the -apartment. Most leases are for one year. Once a lease is signed, you are legally -bound to live in the apartment, and you must pay the rent even if you decide -that you no longer want to live in the apartment. Do not sign any lease that you -do not understand. Be sure to check the dates of the lease carefully. Please -visit the International Student Advisor before you move off campus. He/she can -offer advice about apartments and assist you with interpreting leases. Please -note that if you lease an apartment in your name but also have roommates living -with you, you must be very careful about changing the name on the lease if you -move. If your name is on the lease, you are responsible for the payment of rent -even if you move out and allow other students to move into the apartment. If -you do not pay the rent, the landlord may take legal action against you. Please -be especially careful in situations where you share telephone and utility services -with other students. Never agree to have a lease or a utility bill placed in your -name unless you are certain no additional money is owed on the account. If you -change a lease or utility bill from another student’s name to yours, you will be -responsible for any money owed on the lease or account even if the charges -were billed before the account was placed in your name. - -DINING SERVICE -The University Dining Hall, which is under the supervision of the Director of Food -Services, provides nutritious and wholesome meals at a minimum cost to the -student. Meals are served in an ultra-modern facility, which includes a food court -consisting of the following concepts: Café Features, Wokery, Grill Works, Itza -Pizza, Deli Corner, Sweet Things, Salad Garden, Veggie and Drinks. The meals -are unlimited (all you can eat). The Dining Hall seating capacity is approximately -525 students. All boarding students must present proper University student -identification. Non-boarding students, faculty, staff and visitors must pay cash at -the cashier’s stand before being served. - -Meal Schedule - -Monday-Thursday -Breakfast -Lunch -Dinner -Friday -Breakfast -Lunch -Dinner - -Saturday & Sunday -Brunch - -7:00 a.m. – 8:45 a.m. -11:30 a.m. – 1:00 p.m. -5:00 p.m. – 7:00 p.m. - -7:00 a.m. – 8:45 a.m. -11:30 a.m. – 1:00 p.m. -4:30 p.m. – 6:30 p.m. - -10:30 a.m. – 12:30 p.m. - -26 - - Dinner - -4:00 p.m. − 5:30 p.m. - -(Schedule is subject to change without notice) - -LEGAL CONCERNS -LAWS -While you are in the U.S., you are subject to all of the laws and regulations, -which govern U.S. citizens. In addition, you are subject to immigration laws, -which define and limit your activities as a non-immigrant. Failure to comply with -immigration laws can lead to a request for your voluntary departure, or a forced -deportation from the U.S. - -EQUAL PROTECTION and DUE PROCESS -All persons in the U.S., regardless of citizenship, are entitled to the same -protection of the law and the same procedures of law enforcement. This means -that, with a few exceptions, you will be treated the same as an American citizen -in your relationship with the law enforcement agencies. Those few exceptions -involve immigration law. Consult the Office of the International Student Advisor -for clarification of this issue. - -ILLEGAL DRUGS -The possession, use, or sale of drugs or narcotics, including marijuana, hashish, -amphetamines, hallucinogens, barbiturates, cocaine, heroin, and a wide range of -other drugs, are serious offenses under Georgia law and immigration law. Those -who choose to use illegal drugs run the risk of expulsion from the university, -imprisonment, and/or deportation from the United States. - -ALCOHOL -The minimum legal age for drinking alcoholic beverages (beer, wine, liquor, etc.) -in the State of Georgia is 21. You cannot be served alcoholic beverages unless -you are 21 or older. You will be asked to present an identification card with your -photograph and your date of birth before you can purchase alcohol from a store -or enter a bar, nightclub, or other establishment that serves alcohol. Students -who are over age 18 but under 21 will be admitted, but these students are not -allowed to purchase alcohol. The police routinely search all nightclubs and bars -in Albany for students under age 21 who have consumed alcoholic beverages or -used false identification to enter the establishment. Students who are caught -are arrested, taken to jail, and fined. They may also face university disciplinary -measures. - -Driving a car after drinking alcoholic beverages is also illegal in Georgia, -regardless of age. Driving under the influence (DUI) of alcohol carries severe -penalties and can result in the loss of your driver’s license, heavy fines, -imprisonment, disciplinary action by the university, and/or deportation. NEVER -DRINK AND DRIVE. - -SMOKING - -27 - - Smoking has become socially unacceptable in many places. Smoking is not -permitted in Albany State University buildings (except dorm rooms). Smoking is -also forbidden in hospitals, many restaurants, and other public places. Always -ask permission of your companies before smoking. Never smoke in someone’s -home unless he or she smokes as well. - -MAIL -CAMPUS MAIL -Campus mail is a service provided by the university for mail going between -people and offices on the premises of the campus. To send a letter by campus -mail, clearly mark the office and person it should go to and write CAMPUS MAIL -in the upper right hand corner. No stamp is needed. Drop the letter in the -collection point at the post office or in your Department Office. - -POST OFFICES -The university post office can send materials in both domestic and overseas mail. -Overseas mail must be weighed and stamped according to weight. Domestic -mail requires standard postage of 44 cents per letter. Packages and letters -containing more than five sheets of paper usually require extra postage. Any -package weighing over 16 ounces may be mailed first class from the university -post office. Domestic mail also requires a zip code for delivery. This is a five- -letter code specific to the area the letter or package is being sent to. If you do -not know the zip code of the area you are mailing to, ask for a zip code directory -at the post office and look up the proper zip code. - -CHANGE OF ADDRESS -Students who move off campus or leave ASU for a vacation should fill out a U.S. -Postal Service Change of Address card, sign the card, and place it in a U.S -mailbox. By completing this card, you allow the postal service to automatically -forward your mail to your new address. There is no charge for this service. -Unfortunately, the Office of the International Student Advisor cannot forward -mail to students who have moved off campus. Please do not have your mail -sent to the Office of the International Student Advisor. You will also need to -complete an AR-11 form which you obtain in the International Office. - -ELECTRONIC MAIL -An electronic mail (e-mail) account is available to all ASU students, faculty, and -staff free of charge. All ASI students are expected to use their ASU Ram Mail -because that is the official mode of communication. If you need an account or -would like to learn more about this service, call the Office of Information -Technology at 229-430-4909. You can also use the Internet through ASU -website: www.asurams.edu. E-mail and the Internet are wonderful ways to -communicate with friends around the world at no expense to you. - -MONEY AND BANKING -BANKS -There are many banks located in Albany. Most of them have specially priced -services for ASU students. The Bank of America has an automated teller machine - -28 - - or ATM in the Reese Student Union building on lower campus. As with any other -service, you should determine which banking services are most important to you -and call the different banks to find out about prices and available services. - -CHECKING AND SAVING ACCOUNTS -All students should open checking and/or saving accounts at a local bank. It is -not wise to carry or keep large amounts of cash. Most local businesses will -accept checks from local banks only. - -It is important to know how much money is in your checking account at all -times. Do not write checks for more money than you have in your checking -account. Checks written for more than the amount in an account are returned -unpaid. You will have to pay a fine to the bank, to the business where you -wrote the check, and may be unable to write checks in the future. Writing checks -when you do not have enough money to cover the check is called “bouncing -checks.” Bouncing checks is a criminal offense punishable by imprisonment. A -checking account is a service. Charges for services vary from bank to bank. -Savings accounts allow you to deposit money safely in a bank and receive a -small amount of interest on the balance. Savings accounts require passbooks for -depositing and withdrawing money. Keep both your checkbook and your -passbook in a safe place and report it to the bank immediately if either document -is lost or stolen. - -CREDIT CARDS -Credit cards are available through most banks. They have high interest rates, -but are very easy to use. Credit cards can be used almost everywhere and are -billed on a monthly basis. The best way to use a credit card is to pay the entire -bill monthly; otherwise, your balance may get higher than you can pay. Many -students find it helpful to carry an international credit card such as Visa or -MasterCard. Purchases on these cards can be made in U.S. dollars while the bill -is sent to your home and is denoted in your home currency. These cards can -also be used at automated teller machines to withdraw cash. Be sure to inquire -about the transaction fees before you obtain a credit card. Often the use of -these cards is an inexpensive way to exchange currency. But the fees can -become expensive. Your student fees can also be paid by credit card -(MasterCard or Visa only). - -OTHER BANKING SERVICES OFFERED FOR FEES - -Cashier’s Check -Money Orders -Certified Checks -Travelers Checks -Safety deposit box rental for storing valuable items - -MONEY EXCHANGE -Foreign currency must be taken to a bank to be exchanged. All currency (except -Canadian) must be sent out of town by the local banks for exchange. There is a - -29 - - charge of about $10.00 for the transaction and a processing time of 3 to 6 -weeks. - -RECEIVING MONEY FROM HOME - -1. - -2. - -By Mail -Tell your family to send numbered cashier’s check in U.S. -dollars, purchased from a bank in your country that has -business arrangement with a well-known U.S. bank. Make sure -that the check has a bank routing number on the bottom of the -check. - -By Telex -You may receive money from home by telex (wire). Wired -money generally arrives 5-10 days after it is sent from overseas. -It should be arranged through a local bank, and you should -have a checking or savings account established at the bank. -The receiving bank will charge about $12.00 as handling fee. It -is not possible to have money wired directly to your ASU -student account. - -Please be sure to request your funds from home well in advance of fee payment -deadlines. International transfers can often experience unexpected delays, so -you should plan ahead each semester. THERE IS NO FLEXIBITY IN FEE -PAYMENT DEADLINES. Please do not request an extension of the set fee -payment deadline. Failure to pay your student fees in a timely manner will result -in your having to return home. - -30 - - INTERNATIONAL STUDENTS -If English is not a student’s native language, if the student did not attend high -school in the United States, and if the student wishes to discuss taking the -Regents’ Test especially designed for International students, the student must -contact the Regents’ Test Director whose office is in 103 Simmons Hall (229-430- -4735 – flo.hill@asurams.edu). The local International Students’ Task Force -Committee has suggested that a student be considered for a special form of the -essay test after taking and failing the test at least once. There is no special test -for reading at this time; however, students may be given more time and may be -allowed to use their translation dictionaries. - -RELIGIOUS LIFE -Religion is a very important part of South Georgia culture. People will probably -invite you to visit their church and to take part in church sponsored social -activities. While these church services and activities can be very meaningful to -you, participation will not obligate you to join a church. Churches often attempt -to bring students into their church families by asking students to change their -belief to match the beliefs of the church members. If church members or clergy -make you feel uncomfortable about your religious belief or your values, feel free -tell them that you wish to maintain your own beliefs. If they cannot accept that, -find another church group! - -There are many churches located in the Albany area. Denominations include: -African-Methodist Episcopal, Apostolic, Assembly of God, Bahai, Baptist, Catholic, -Charismatic, Christian, Church of Christ, Church of God, Church of Jesus Christ of -Latter Days, Episcopal (Anglican), Holiness, Interdenominational, Jehovah -Witness, Lutheran, Nazarene, Pentecostal, Salvation Army, Seventh Day -Adventist, and Unitarian. All denominations are listed under “Churches” or -“Synagogues” in the Yellow Pages of the telephone book. - -SHOPPING -GROCERY STORES -All foods are located in one store. You will find fresh fruits, vegetables, canned -(tinned) goods, baby foods, meats, bread, cleaning supplies, paper products, and -frozen foods. Generally, grocery stores have lower prices than smaller stores. -There are several large grocery stores in Albany and some of them are within -walking distance of the University. You can order special items if you are -looking for ingredients not readily available in Albany stores. - -DRUGSTORES AND PHARMACIES -This is the only place where you can buy medicine that the doctor has ordered -for you. You can often buy many other things there as well. The most -convenient drugstore is CVS on Radium Springs Road, near the campus. Less -expensive pharmacies are located in discount stores and some grocery stores. - -31 - - DISCOUNT STORES -These are department stores where you find a large variety of things including -clothing, pharmacies, garden supplies, small appliance, paper products, linens, -and even some types of furniture. They often have lower prices and can be -wonderful places to see American people and products. Some of the local -discount stores include K-Mart, Wal-Mart, and Target. - -RESTAURANTS -Unless you are buying food, restaurant owners do not like for you to sit and -“take up space” from paying customers. You should leave money (a tip) on the -table for the waiter or waitress when you eat at a restaurant. The tip is usually -15%-20% of the total bill, depending on the quality of the service. Fast food -restaurants do not have waiters and waitresses. Do not leave a tip at this type -of restaurant. There are a number of commercial fast food restaurants around -the campus. Food can be ordered “for here” (to be eaten in the restaurant) or -“to go” (to be eaten at home). - -THE ALBANY MALL -The Albany Mall is a shopping center housing several dozen stores and -restaurants. It is located some miles west of the campus on Dawson Road. You -can buy almost anything you need in the Mall. Prices in these stores are usually -higher than the prices at Discount Stores, but the Mall is a great place to meet -people and experience American culture! - -OUTLET MALL -A large mall containing only manufacturer’s outlet stores is located many miles -south of Albany on Interstate 75. This mall has over 100 stores and is a good -place to shop because the prices are low while the quality of the merchandise is -high. - -Student Athlete - -Best practices For International Athlete Recruits (NCAA) -F-1 Visa and Issuance of SEVIS Form I-20 -International student-athletes who require an F-1 student visa are subject to -specific requirements regarding their admissions eligibility, financial ability and -academic status. Failure of students to follow appropriate regulations and -guidelines can result in severe difficulties with their immigration status with the -U.S. Department of Homeland Security. To avoid difficulties, the NCAA strongly -recommends that athletics staff meet with appropriate school officials (e.g., -international admissions officers, international student advisors) to gain an -understanding of the school’s policies and procedures that are in place to protect -international students and to maintain compliance with all immigration -regulations. - -32 - - IMPORTANT NOTE FOR ATHLETICS ADMINISTRATORS AND ADMISSIONS OFFICERS -Under no circumstances should an international student athlete be issued a Form -I-20 with the expectation of financial support from the athletics office before the -student has been certified as a final qualifier by the NCAA Eligibility Center or is -determined to be a partial qualifier in Division II. - -• Doing so opens the possibility for an international student who might not be -eligible for financial aid to obtain a visa and enter the United States without -having the sufficient funds to cover tuition and living expenses, as required by -immigration laws. - -Financial Ability - -• All F-1 visa students must submit appropriate documentation of financial -support before the school can issue the SEVIS Form I-20 [certificate of eligibility -for nonimmigrant (F-1) student status]. -• Generally the school’s office of recruitment and admissions office will have -information as to what constitutes appropriate documentation of financial -support. -• Athletics department staff members must understand that withdrawal of -athletics financial aid for an international student-athlete could result in severe -economic hardship for the student and could result in a violation of the student’s -immigration status. - -What Athletics Administrators Should Tell Student-Athletes -• Student-athletes who will be receiving financial aid from the athletics -department must understand that if the amount of aid is less than the -amount required for issuance of the I-20, then the student will need to -provide additional documentation of financial support. - -• Student-athletes should be informed of the true costs of attending the - -school and must understand the financial implications in the event of the -loss of financial support from athletics. - -• Although students athletes may receive some financial assistance through - -athletics students are responsible for purchasing their books for the -duration of the academic tenure. - -• Student-athletes must understand that if their scholarship is made - -unavailable for whatever reason (e.g., because the student-athlete did not -meet initial-eligibility requirement, they must have other means of -financial support. If sufficient finances are not available from other -sources, the student risks violation of immigration status and possible -deportation. - -SOCIAL SECURITY NUMBERS -A Social Security Number is a government identification number issued to all -residents of the United States. Most residents receive this number as children. -In the U.S., we do not carry passports or national ID cards. Instead, we use this -number and a driver’s license as our primary forms of identification for -employment records, taxation, opening bank accounts, university records and - -33 - - other business matters. All drivers’ licenses contain a color photograph of the -bearer as well as the current address and date of birth. - -For your admission to ASU, you were issued a student identification number -which has the same number of digits as the social security number. The -university uses this ID number for all of your university records. The number -also appears on your student identification card. While this number is sufficient -for your academic records, you will need to obtain a Social Security number -before you can work on campus. To obtain the number, you must complete an -application form and deliver it, along with a letter of verification from the ISA -and an offer letter from the person who is hiring you on letterhead, in person to -the Social Security Administration in Albany. There is no charge for Social -Security numbers. Call 1-800-772-1213 for more information. - -TAXES -Federal Law requires that foreign students pay income tax from sources within -the U.S. This income generally includes salary from any employment, on or off -campus, as well as scholarship money. Tax return is due on April 15 for the -previous year. In addition you may need to file a tax statement. If you have -concern about your tax obligation, you may wish to consult a certified public -accountant (CPA). The Internal Revenue Service is the branch of the federal -government responsible for income tax collection. This agency is often simply -called IRS. You can call the IRS directly at 1-800-829-1040 for additional -information. - -TELEPHONE AND TELEGRAMS -TELEPHONE -Telephone service in the U.S. is divided into two types: local and long-distance -(including international) service. Most people work with two different companies -to obtain telephone services: a local carrier for local service and long distance -carrier for long distance service. The local carrier for Albany is BellSouth. -Contact the company by calling 229-780-2355. You may need to pay a deposit -to begin your local service. For about $30.00 a month, you may have unlimited -telephone calls within the country. You will be billed monthly for this service. - -There are several local and long distance services in the U.S. Each company lays -claims to lowest telephone rates for long distance. You may want to talk to each -company to learn who can offer you the best deal. Rates usually depend on -where you call and when you call. To get the cheapest rate, you may find that -all three companies have special plans with many restrictions. Be sure to read -the details of the special plans. Be sure to ask for the rates for calling your -country. You can contact these companies by calling the numbers listed in the -local telephone book. Once you choose a company, they will provide you with -direct dialing from your home and a calling card, which you can use from any -phone. - -INFORMATION - -34 - - If you cannot find a number on-line, you may call “Information,” also called -“Directory Assistance” and the number is 1411. In order to use the information -service, you must know the name of the person or organization you wish to call -as well as the city they are located in. The number you dial for information -depends upon the area you wish to call. The telephone book also contains useful -information about how to use the telephone and place different kinds of calls. -The Yellow Pages found on-line contain listings of local businesses and can help -you locate services. - -TOLL FREE CALLING -Most numbers with an area code of 1-800 or 1-888 are toll free calls. This -means that there will be no charge for the call. Many businesses and services -have toll free numbers for customer use. You should use toll free numbers -whenever possible to keep your telephone bill low. - -OTHER ASPECTS OF TELEPHONE SERVICES -If you live off-campus and share a telephone number, be sure to have an -agreement about how the bill will be paid and whose name will appear on the -bill. The names and numbers of all telephone subscribes are published in the -BellSouth telephone book unless you request an unpublished number. Albany has -several “local exchanges.” This term refers to the first three digits of the -telephone number. You will see prefixes listed as 430, 431, 434, and 435, for -example. Be sure to give all seven digits of your number. If you are calling from -one phone on campus to another, you only need to dial the last four digits of the -number. Many students today buy pre-paid cell phones. - -TELEGRAMS -Letters or money may be sent both within the U.S. and internationally through -Western Union by calling their toll-free number 1-800-325-6000. Mailgram is also -available through Western Union and is less expensive than telegrams. To send -a Mailgram, call Western Union at 1-800-325-6000. MoneyGram provides similar -services as Western Union. Their toll free number is 1-800-926-9400. - -FAX -Faxes may be sent or received through the fax services offered by some -companies in Albany. You will be charged for this service. For international -faxes, students should use Mailboxes, etc. at 229-434-1133 or 229-432-2230. -Unfortunately, the high cost of international and long distance faxes prohibits the -Office of the International Student Advisor or the University from sending -personal faxes for students. - -TESTING -The Albany State University Computer-based Testing Center (Department of -Counseling, Testing and Disability Services), located on the first floor in ACAD -195, administers and prepares students for several national computer-based -tests periodically throughout the year. The GRE, GMAT, MAT, MCAT, LSAT, -PCAT, and TOEFL are a few of the tests administered by the center. National test -scores are required for admission to U.S. graduate schools. These tests have - -35 - - standard fees and deadlines for enrollment. For information about a test, call -the Computer-based Testing office at 229- 430-1631. To sign up for a test, visit -the Division of Student Affairs and pick up an application. - -TRANSPORTATION -TRANSPORTATION IN ALBANY -Albany has a public transportation system. In emergencies, you can hire a taxi -(cab). Companies offering taxi services are listed under “Taxicabs” in the -Albany telephone book. Be sure and ask how much you will be charged before -hiring a taxi. Some students purchase bicycles for transportation. Cyclists must -ride on the roadways, as there are no separate bicycle paths. If you choose to -ride a bicycle, you must be very careful and watch out for cars. American drivers -are not accustomed to cyclists on the road, be sure to lock up your bike when -you are not using it. To travel outside Albany, you must go by bus or airplane. - -BUS TRANSPORTATION -The Greyhound Bus Lines station in Albany is located at 300 Oglethorpe Blvd. -Call 1-800-231-2222 for nation-wide information or 229-432-0511 for local -terminal information. - -AIR TRANSPORTATION -Albany has an airport located south of town. Service is limited to several daily -flights to Atlanta, where connecting flights are available to other cities. Call 229- -430-5175 for schedules. - -TRAVEL AGENTS -Travel agents will help you make travel arrangements for a small fee. They -often save you money by finding the lowest rates available to your destination. -Look in the Yellow Pages under “Travel Agencies”. Or you may book your tickets -online using a credit card at the airlines website. (e.g. www.delta.com) - -INTERNATIONAL TRAVEL -If you know that you will return to your country during a school vacation, you -should try to plan international travel 2 or 3 months in advance. The best rates -for international travel are obtained by making reservations early. Also, all -students are eligible for discounted airfares through an organization called -Council Travel. To obtain these special fares, you must have an international -student ID card. These cards can be purchased for $20 from the Office of the -International Student Advisor or local travel agents. These cards also qualify the -bearer for discounts at museums and other tourist attractions. If you have -questions about international travel, you should speak with the staff at the Office -of the International Student Advisor or experienced travelers from your country. - -36 - - CAMPUS CONTACTS - -Quick Reference - -Information - -Office - -Phone Number - -Advising and Immigration - -Office of Int’l Student Advisor - -Meal Plan - -Food Services - -General Info (Switchboard) - -Facilities Management - -Graduate Studies - -Graduate School - -Residence Life (Housing) - -Student Affairs - -Mail delivery services for on-campus students - -Post Office - -Parking Permits - -Public Safety - -229-430-7075 - -229-430-4750 - -229-430-4600 - -229-430-5118 - -229-430-4741 - -229-430-4745 - -229-430-4711 - -Personal & Counseling - -Counseling, Testing & Disability Services - - 229- 430-4667 - -Reporting Theft or Crime - -Public Safety - -Admission Status - -Admissions Office - -Student Activities - -Student Affairs - -Transcripts (to be sent from ASU) - -Registrar’s Office - -Tuition Payment - -Business Office - -229-430-4711 - -229-430-4646 - -229-430-4739 - -229-430-4638 - -229-430-4610 - -37 - - SERVICES FOR INTERNATIONAL STUDENTS - - Orientation sessions and workshops for new students. - Liaison with the U.S. Citizenship and Immigration Services (USCIS) for -all of your immigration documents. We will help you with filing for -practical training, transfer of schools, and replacement of lost -documents. WE are your source for maintaining your immigration -status. - - Counseling and referral services. Personal and cross-cultural - -counseling is available on a confidential basis to all students. If you -would like to talk to someone, call the International Student Advisor -(ISA) at 229-430-4721 to make an appointment. If necessary, you -may be referred elsewhere for help. - - Information. If you do not know what to do, to whom to talk, or -where to find ANYTHING, the Office of the International Student -Advisor can help. - - Letters of certification for government loans, money exchange, re- -entry into the U.S., invitations to visit the U.S., and other official -documents are available at the office. - - Information about Albany State University International Student - -Association (ASUISA): ASUISA annually elects new student officers, -and sponsors activities throughout the year, by and for international -students. - - Other activities for all Global Programs students such as International - -Festival in November, picnics, coffee/tea hour, and the Cultural -Exchange Club/Global Ambassadors. - -38 - - CHECK LISTS - -ANNUAL CHECKLIST - -• Renew passport if necessary (six months before expiration). - -• File income tax returns if necessary (between January 1 and - -April 15). - -• Extend temporary stay with sponsor and with the - -International Advisor for an extended I-20 if necessary (30 -days before expiration). - -PERIODIC CHECKLIST - -• Report changes of address (within 10 days) to the Office of -the International Student Advisor and the Office of the -Registrar. This is required by law. So is your email -address and your telephone number. - -• Report any change of marital status, or if your family has - -joined you in the U.S., to the International Student Advisor. - -• Complete registration for Student Health Insurance every - -semester. - -• - -If leaving the U.S. temporarily, have 1-20 or DS-2019 -signed by a Designated School Official and check visa -validity. - -• Check in with International Student Advisor at the beginning -of each semester and schedule appointments periodically. -Always bring all of your documents when you come. - -Information from Handbook is from the following websites. -www.travel.state.gov/visa (U.S. Department of State) -www.ice.gov (Department of Homeland Security) -www.uscis.gov (U.S. Citizenship and Immigration Services) -www.ncaa.com (National Collegiate Athletic Association (NCAA)) - -39 - - Office of the International Student Advisor -Sue Ann Balch, J.D. -205 Reese Hall Lower East Campus, 504 College Drive -(229) 430-7075 (Office) -(229) 434-7106 (Fax) -Sueann.balch@asurams.edu (email address) - -40 +_____________________________________________________________________________________________ +Page 12 diff --git a/q126/random_k2/question.json b/q126/random_k2/question.json index 53da02a2f8c29d7914fc82dcd660ee72385c16fa..539c52824df2b67a00a4d37616ad6d27fa1a2137 100644 --- a/q126/random_k2/question.json +++ b/q126/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "owners-manual-2170416.pdf", - "web_974a02e9fcc23ca9.pdf", - "web_ddfd3d214b1f0ea4.pdf" + "0a89b981c2736e87bac48a090e1a3e60dd8a4e0f.pdf", + "web_ccf5b62f39e26967.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q126/random_k2/random_1.md b/q126/random_k2/random_1.md index 0f5db4f84cd2bbe84dc239b53c3a2c97d631e42e..16de3e3316b1e3bb316b878e17b05640d996b000 100644 --- a/q126/random_k2/random_1.md +++ b/q126/random_k2/random_1.md @@ -1,279 +1,1070 @@ -COMPETITOR ANALYSIS - -© InnoValor, 2018 -© InnoValor, 2018 - -How to use -Template -Example - - Competitor Analysis - -A Competitor Analysis helps you to understand your -competition. It supports you in creating your unique position in -the market. With this analysis you make an assessment of the -strengths and weaknesses of your current and potential -competitors. With a better understanding of your competition -you can choose and implement effective strategies that will -improve your competitive advantage. - -Tips for use - -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. This tool can easily be used by -yourself, or with colleagues. Print the template and use the tool -to create an understanding of your competition. - -Have a look at the example of Fleur Flowers to get an -understanding of a Competitor Analysis. - -How to use - -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. Follow the steps to complete your -own Competitor Analysis. - -Summarize your target group - -Shortly describe your target group. Who are the customers that -buy your offering? Make the description as specific as possible. If - -© InnoValor, 2018 - -your target group consist of businesses you can mention their -size, industry or their offering. If your target group consist of -consumers you can mention their age, goals, gender, education -level or interests. - -Identify your competitors - -Write the names for up to three competitors in the first row. To -help you generate this list, think of where your customers would -go if they cannot buy your product or service. - -Determine competitors’ strengths and weaknesses - -Now think like your customers. Why would a customer go for the -services of your competitors instead of coming to your company? -You can describe this by thinking about the offerings of your -competitors. It would even be better when you visit your -competitors and experience what their customers experience. -Write down three strengths of each competitor. Also consider -why customers would not go to your competitors. Write down -three weaknesses of each competitor. - -Describe the key findings - -Finally, in the fourth column of the template write down the key -findings for the strengths. What stands out? What patterns can -you discover from your competitors? And which opportunities -can you come up with? Also, write down your key -findings for the weaknesses. What are the aspects that - - your competitors don’t do well? How could you avoid making -the same mistakes? What should you do to serve your customers -in such a way that you differentiate from your competitors in a -positive way? - -© InnoValor, 2018 - - competitor analysis - -name - -target group - -competitors - -most important findings - -s -h -t -g -n -e -r -t -s - -s -e -s -s -e -n -k -a -e -w - -date - - Example Fleurs Flowers - -industry - -Retail - -products & -services - -Flowers - -size of -business - -2 full-timers 2 part-timers - -revenue - -€ 37,000 - -location - -Main street - Bloomfield - -Fleurs Flowers is a flower shop in a small town. Fleur sells -flowers to a small group of regular clients who regularly -purchase fresh flowers from her shop. Her customers generally -consist of men and women aged 25. The customers often just -bought a house and flowers are the perfect way to bring -atmosphere to the new place. In addition, she has customers -who buy flowers for birthday and holiday occasions. With -Valentine’s and Mother's Day she is working overtime to meet -the demand - -Fleurs Flowers’ competitor analysis - -A company like Fleurs Flowers has a difficult time in today's -society. She has a physical store that is open during shopping -hours. Her customers are people who also work at those times. -These customers search for convenience in their shopping. -Internet shopping is very popular amongst these people. Fleur -wants to increase her appeal to her target group. She -investigates the advantages and disadvantages of the sales -strategies of her main competitors. For her research Fleur visits -her competitors and buys their flowers. The results of Fleur’s -competitor analysis can be read in the example on the next page. - -© InnoValor, 2018 - - competitor analysis - -Fleurs Flowers - -name - -target group - -Men or women from 25 years old with an average salary - -fromyouflowers.com - -Lidl - -Valentino flowershop - -competitors - -most important findings - -Offers flowers online - -The price for flowers is low - -Customers can order -flowers online but have to -pick them up at the store - -Delivers at home though -whole the region within 24 -hours - -Customers can buy flowers -while doing groceries, they do -not have to visit a different store - -When ordering flowers above a -certain amount the customer -will receive a free vase - -Package deals – get a -discount on a vase when -flowers are bought - -Everyday there are new flowers - -A lot of time is spend on -wrapping the flowers as a gift - -An online shop seems to be more -and more important. Customers -want to be able to order flowers -from their home. Customers also -like it when they get something for -free or a discount on something -when they buy flowers - -There is no way that a -customer can see the flowers -before they are ordered - -Customers cannot create -their own piece since the -flowers are all prearranged - -The price for flowers is higher -than in other shops - -Ordering flowers is less -personal - -The flowers are made in mass - -Ordering flowers takes -quite some time - -Customers cannot create -their own bouquet of flowers - -Buying flowers is not a special -occasion since it is been done -during doing groceries - -The shop is not as much open -as customers would like, for -instance it is closed on Mondays - -Customers seem to think that -buying flowers is a personal and -special occasion. They do not like -it when buying flowers is less -personal. They also do not like that -they cannot create their own -bouquet - -s -h -t -g -n -e -r -t -s - -s -e -s -s -e -n -k -a -e -w - -date - +SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation +| | YagizCanAkay1 | | | MuhammedYusufKartal1 | | | EsraAlparslan1 | | | | | +| --- | ------------- | --- | ------------------- | -------------------- | ------------ | --- | -------------- | --- | --- | --- | --- | +| | | | FarukOrtakoyluoglu1 | | ArdaAkpinar2 | | | | | | | +1TOBBUniversityofEconomicsandTechnology 2OSTIMTechnicalUniversity +{y.akay, m.kartal, ealparslan, fortakoyluoglu}@etu.edu.tr 230206003@ostimteknik.edu.tr +| | | Abstract | | | andcommon,yetexposecriticalcoverageandrea- | | | | | | | +| --- | --- | -------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | +soningbottlenecksincurrentNLPsystems. +Answeringcomplex,real-worldqueriesoften +6202 raM 9 ]LC.sc[ 1v92380.3062:viXra Traditionalsingle-indexorsingle-agentretrieval +requiressynthesizingfactsscatteredacrossvast +| | | | | | systemsstruggleto | | maintainbothscalability | | | | and | +| -------- | -------- | -------- | --------- | ----- | ----------------- | --- | ----------------------- | --- | --- | --- | --- | +| document | corpora. | In these | settings, | stan- | | | | | | | | +contextualrelevanceoverlarge,heterogeneouscor- +| dard retrieval-augmented | | | generation | (RAG) | | | | | | | | +| ------------------------ | --- | --- | ---------- | ----- | ---------------- | --- | -------- | --------------- | --- | ------- | --- | +| | | | | | pora, especially | | for long | or structurally | | complex | | +pipelinessufferfromincompleteevidencecov- +erage,whilelong-contextlargelanguagemod- documents. Standard RAG pipelines retrieve a +| | | | | | fixed number | | of documents | K | and process | | them | +| ---------- | -------- | --------- | -------- | ---- | ------------ | --- | ------------ | --- | ----------- | --- | ---- | +| els (LLMs) | struggle | to reason | reliably | over | | | | | | | | +massive inputs. We introduce SPD-RAG, a withinasinglecontextwindow(Lewisetal.,2020), +hierarchicalmulti-agentframeworkforexhaus- +whichfailswhenanswersdependoninformation +tivecross-documentquestionansweringthatde- +distributedacrossmanydocuments,sinceevidence +composestheproblemalongthedocumentaxis. +| | | | | | beyondthetop-K | | resultsistypicallydiscardedin | | | | | +| ---- | -------- | ------------ | ---- | --------- | -------------- | --------- | ----------------------------- | ------------ | --- | ---- | --- | +| Each | document | is processed | by a | dedicated | | | | | | | | +| | | | | | a single | retrieval | pass. | Long-context | | LLMs | ex- | +document-levelagentoperatingonlyonitsown +content,enablingfocusedretrieval,whileaco- tendcontextwindowsto128K–2Mtokens(Gemini +| | | | | | Teametal.,2024), | | butempiricalevidenceshows | | | | | +| --------- | ---------- | ----- | ----------- | ------ | ---------------- | --- | ------------------------- | --- | --- | --- | --- | +| ordinator | dispatches | tasks | to relevant | agents | | | | | | | | +and aggregates their partial answers. Agent thatreasoningqualitydegradessignificantlyascon- +outputsaresynthesizedbymergingpartialan- text length increases (Liu et al., 2023). This sug- +swersthroughatoken-boundedsynthesislayer +| | | | | | gests that | the | bottleneck | is not | only | retrieval, | but | +| --- | --- | --- | --- | --- | ---------- | --- | ---------- | ------ | ---- | ---------- | --- | +(whichsupportsrecursivemap-reduceformas- +alsoreasoningatscaleovermanydocumentsand +| sivecorpora). | | Thisdocument-levelspecializa- | | | | | | | | | | +| ------------- | --- | ----------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +hundredsofthousandsoftokens. +tionwithcentralizedfusionimprovesscalabil- +ityandanswerqualityinheterogeneousmulti- In this work, we introduce SPD-RAG +| | | | | | (Sub-agent | per | Document | Retrieval-Augmented | | | | +| -------- | -------- | ----- | -------- | ---------- | ---------- | --- | -------- | ------------------- | --- | --- | --- | +| document | settings | while | yielding | a modular, | | | | | | | | +extensibleretrievalpipeline. OntheLOONG Generation),ahierarchicalmulti-agentarchitecture +benchmark (EMNLP 2024) for long-context forexhaustivemulti-documentquestionanswering +multi-document QA, SPD-RAG achieves an thatfactorstheproblemalongthedocumentaxis +AvgScoreof58.1(GPT-5evaluation),outper- +| | | | | | rather than | the | task axis. | Instead | | of forcing | a | +| --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | ---------- | --- | +formingNormalRAG(33.0)andAgenticRAG +| | | | | | single model | | to hunt through | | a massive | | global | +| --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --------- | --- | ------ | +(32.8)whileusingonly38%oftheAPIcostof +| | | | | | index, SPD-RAG | | uses | a central | coordinator | | to | +| --- | --- | --- | --- | --- | -------------- | --- | ---- | --------- | ----------- | --- | --- | +afull-contextbaseline(68.0). +| | | | | | decompose | the | user’s | query into | shared | instruc- | | +| --- | --- | --- | --- | --- | --------- | --- | ------ | ---------- | ------ | -------- | --- | +1 Introduction tions. It then deploys a dedicated, cost-efficient +| | | | | | sub-agenttoeachdocumentinthecorpus. | | | | | | These | +| --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | ----- | +Large language models (LLMs) and the agentic agents operate entirely in parallel, treating their +systems built around them are increasingly used assigneddocumentsasisolatedretrievaluniverses +forcomplexinformationsearchtasks(Guoetal., +| | | | | | to extract | relevant | findings. | Finally, | | a synthesis | | +| --- | --- | --- | --- | --- | ---------- | -------- | --------- | -------- | --- | ----------- | --- | +2024). Real-worldquestionsoftenrequiresynthe- +| | | | | | model | aggregates | these | document-grounded | | | | +| --- | --- | --- | --- | --- | ----- | ---------- | ----- | ----------------- | --- | --- | --- | +sizingevidencescatteredacrossmanydocuments, findings—employingarecursivemergingfallback +suchasassessingacompany’sfinancialrisksacross +| | | | | | for exceptionally | | large | corpora—to | | construct | a | +| --- | --- | --- | --- | --- | ----------------- | --- | ----- | ---------- | --- | --------- | --- | +yearsofreportsorintegratingfindingsfrommul- +| | | | | | comprehensive, | | final answer. | | Our codebase | | and | +| --- | --- | --- | --- | --- | -------------- | --- | ------------- | --- | ------------ | --- | --- | +tiplescientificpapers.1 +| | | Suchquestionsarenatural | | | datasetsusedinthispapercanbefoundat | | | | | | | +| --- | --- | ----------------------- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +https://github.com/NebulAICompany/ +| 1Examples | inspired | by scenarios | in the | Loong bench- | | | | | | | | +| ---------------------- | -------- | ------------ | ------ | ------------ | -------- | --- | --- | --- | --- | --- | --- | +| mark(Wangetal.,2024a). | | | | | SPD-RAG. | | | | | | | + +Ourcontributionsareasfollows: tasks,butthatmulti-agentoverheadgrowssuperlin- +early. Theirresultthatindependentagentsamplify +• WeproposeSPD-RAG,ahierarchicalmulti- +errorsby17.2×comparedto4.4×forcentralized +agentframeworkthatcombinesper-document +systemsdirectlymotivatesourcoordinator-based +agentic RAG with a centralized synthesis design. Thetaxonomyofhierarchicalmulti-agent +layer, where cost-efficient document agents systemsbyMoore(2025)furtherclassifiesdesign +| are coordinated | by a | smarter coordinator | | | | | | | +| --------------- | ---- | ------------------- | --- | ---------------------- | --- | ------------------------- | --- | --- | +| | | | | patternsalongfiveaxes: | | controlhierarchy,informa- | | | +agent,enablingdocument-levelspecialization +| | | | | tion flow, | role delegation, | temporal | layering, | and | +| ----------------- | -------------- | --------------- | --------- | ----------------------- | ---------------- | -------- | --------- | --- | +| and parallel | execution | while allowing | each | communicationstructure. | | | | | +| document | to be analyzed | in depth | to ensure | | | | | | +| that all relevant | information | is incorporated | | | | | | | +2.2 LongDocumentandMulti-DocumentQA +withoutmissingcriticalevidence. +| | | | | Long documents | that | exceed | LLM context | win- | +| --- | --- | --- | --- | -------------- | ---- | ------ | ----------- | ---- | +• WeevaluateontheLoongbenchmark(Wang +| | | | | dows are | often handled | via | divide-and-conquer | | +| --- | --- | --- | --- | -------- | ------------- | --- | ------------------ | --- | +etal.,2024a),whichincludesmulti-document +| | | | | strategies. | LLM×MapReduce(Zhouetal.,2025) | | | | +| ------------ | -------------- | ---------- | -------- | --------------------------- | ----------------------------- | --- | ----------------- | --- | +| QA instances | over financial | reports | and aca- | | | | | | +| | | | | formalizestwokeychallenges: | | | inter-chunkdepen- | | +| demic papers | with an | average of | 11 docu- | | | | | | +dency(onechunkdependsoncontextfromanother) +mentsperinstanceandcontextlengthsfrom +andinter-chunkconflict(chunksprovidecontradic- +| 10K to beyond | 250K | tokens, and | show that | | | | | | +| ------------- | ---- | ----------- | --------- | --- | --- | --- | --- | --- | +toryinformation),andproposesstructuredinforma- +oursystemsubstantiallyoutperformsstandard +tionprotocolsandin-contextconfidencecalibration +RAGandAgenticRAGbaselinesunderGPT- +| | | | | tomitigatethem. | ToM(Guoetal.,2025)extends | | | | +| -------- | ------------------ | ---- | ------- | --------------- | ------------------------- | ------------ | ------- | ------ | +| 5-judged | Avg Score (adapted | from | Loong’s | | | | | | +| | | | | this approach | with a | hierarchical | DocTree | repre- | +GPT-4-judgedprotocol;see§4.3),withgains +sentationforrecursivereasoning. +| of around | +25 points | (76% higher | average | | | | | | +| --------- | ---------- | ----------- | ------- | --- | --- | --- | --- | --- | +Formulti-documentQA,LongAgent(Zhaoetal., +score)comparedtoNormalRAGandAgentic +| | | | | 2024) splits | a 128K-token | document | | into chunks | +| --- | --- | --- | --- | ------------ | ------------ | -------- | --- | ----------- | +RAGbaselines. +| | | | | assigned | to member | agents, with | a leader | agent | +| --- | --- | --- | --- | ------------- | -------------- | ------------ | ---------------- | ----- | +| | | | | orchestrating | discussionsand | | inter-membercom- | | +• Weanalyzeablations,documenttypeandtask +complexityeffects,andcost–qualitytradeoffs, munication to reduce hallucinations. However, +showingthat SPD-RAGattainsover85%of LongAgenttargetsasinglelongdocumentrather +thanmultipleindependentdocumentsthatmaycon- +| full-context | baseline quality | at roughly | 38% | | | | | | +| ------------ | ---------------- | ---------- | --- | --------------- | ---- | ------------- | ---- | -------- | +| | | | | flict. DocAgent | (Sun | et al., 2025) | uses | a multi- | +oftheAPIcost. +| | | | | agent framework | that | mimics | human | reading via | +| --- | --- | --- | --- | --------------- | ---- | ------ | ----- | ----------- | +2 RelatedWork +atree-structuredoutlineandrevieweragent,while +| | | | | MDocAgent | (Han et | al., 2025) | employs | five spe- | +| --- | --- | --- | --- | --------- | ------- | ---------- | ------- | --------- | +2.1 LLM-BasedMulti-AgentSystems +| | | | | cialized | agents for multi-modal | | document | under- | +| --- | --- | --- | --- | -------- | ---------------------- | --- | -------- | ------ | +LLM-basedmulti-agentsystemsareincreasingly +standing. +| used for complex | task solving. | Guo et | al. (2024) | | | | | | +| ---------------- | ------------- | ------ | ---------- | --- | --- | --- | --- | --- | +provideacomprehensivesurveyofagentprofiles, +2.3 HierarchicalandRecursive +communicationprotocols,andcollaborationstrate- +Summarization +| gies. Hierarchical | architectures | such | as MegaA- | | | | | | +| ------------------ | ------------- | ---- | --------- | --- | --- | --- | --- | --- | +gent(Wangetal.,2024b),whichdemonstratesau- Ourhierarchicalmergingmechanismbuildsontree- +tonomouscooperationamongupto590agentsvia basedretrievalandrecursivesummarization. RAP- +multi-leveltaskdecomposition,andAgentOrches- TOR(Sarthietal.,2024)recursivelyembeds,clus- +tra(Zhangetal.,2025),whichachievesstate-of-the- ters, and summarizes text chunks to construct a +artperformanceonGAIAusingacentralplanning bottom-uptreewithmultipleabstractionlevels;re- +agent and specialized sub-agents, exemplify this trievingfromtheappropriatelevelyieldsa20%ab- +| trend. | | | | soluteaccuracygainonQuALITY.OuandLapata | | | | | +| ------ | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | +Scalinglawsformulti-agentsystemsofferquan- (2025)showthathierarchicalmergingcanamplify +titativeguidanceforarchitecturedesign. Kimetal. hallucinationsduringrecursivesummarizationand +(2025)evaluate180configurationsacrossfivear- proposecontext-awareaugmentationstrategiesthat +chitectures, finding that centralized coordination replaceintermediatesummarieswithrelevantinput +yields an 80.9% improvement on parallelizable context,whichweadoptinourmergingprotocol. + +2.4 Graph-BasedDocumentAnalysis 3 Methodology +Figure 1 presents an overview of the SPD-RAG +Graph-based document analysis is particu- +architecture. Thesystemprocessesaquerythrough +larly effective for global sensemaking queries. +three layers: the Coordination Layer (§3.1), the +GraphRAG(Edgeetal.,2024)constructsaknowl- +ParallelRetrievalLayer(§3.2),andtheSynthesis +edge graph of entities and relations from source +Layer(§3.3). +documents and applies hierarchical community +detection(e.g.,theLeidenalgorithm(Traagetal., +3.1 CoordinationLayer +2019)) to obtain multi-level communities with +Given a query q and corpus D, the system first +textual summaries. At query time, it retrieves +processes the query through a coordinator agent. +relevant communities, elicits community-level +The coordinator’s role is to decompose the user +partial answers, and aggregates them via a +query into a Shared Instruction Set for Docu- +map–reduce–styleprocedure. AsurveyonLLM- +ments and Synthesis Directives. It generates a +empoweredknowledgegraphconstruction(Bian, +WriteTodosstructuredobjectcontaining(a)alist +2025) highlights how LLMs are reshaping the +of sub_agent_todos: atomic, self-contained ex- +classical knowledge graph pipeline across three +traction tasks specifying exactly what fields, en- +key stages: ontology engineering, knowledge +tities, or numeric values to extract (the shared in- +extraction, and knowledge fusion. Our work is +structions);and(b)asynthesis_directive(2–4 +complementary: instead of inducing an explicit +sentences) instructing the downstream Synthesis +knowledge graph, SPD-RAG operates directly +Layeronhowtoprioritizeandstructurethemerged +over agent-produced textual summaries and +response. +performs similarity-guided recursive synthesis +acrossmanydocuments. 3.2 ParallelRetrievalLayer +Weassignadedicatedsub-agentα toeachdocu- +i +mentd inthecorpus. Eachsub-agentoperatesas +2.5 BenchmarksforMulti-Document i +anindependentRAGloopoveritsassigneddocu- +Reasoning +ment,guidedbythesharedinstructionsetfromthe +The Loong benchmark (Wang et al., 2024a) CoordinationLayer. +evaluates long-context LLMs on extended multi- +Document-Scoped Retrieval and Reasoning. +document QA where every document is relevant +Each document d serves as an isolated retrieval +i +("Leave No Document Behind") across financial, +universe; the assigned sub-agent is strictly con- +legal, and academic domains in English and Chi- +strainedtosearchtoolcallswithind , preventing +i +nese. Itspansfourtasktypeswithcontextlengths +cross-documentdistractorchunksfromdegrading +from 10K to over 250K tokens, and even strong +local extraction. Each sub-agent receives the +frontier models achieve only modest Avg Scores +originalqueryq,itsassigneddocumentname,and +andmediumperfect-answerrates,underscoringthe +thesharedinstructionset(todolist),thenoperates +difficultyofreasoningovermanylong,all-relevant +in an iterative retrieve-and-reason loop. On each +documents. +turn, the agent either issues a search action (a +Complementarydatasetsprobeotheraspectsof focused query string) or a finalize action that +multi-document reasoning. MoNaCo (Wolfson terminates the loop and emits the document- +et al., 2025) focuses on natural questions over groundedfindings. Thesub-agentpromptinstructs +dozensofWikipediapages,MEBench(Linetal., the agent to attempt at least 2 focused searches +2025)onhigh-entity-densityQA,FanOutQA(Zhu before concluding a piece of information is +et al., 2024) on multi-hop questions with absent, with a total cap of 5 search calls across +long cross-document evidence chains, and all tasks. Each search action triggers dense +HoloBench (Maekawa et al., 2024) on database- vector retrieval over the sub-agent’s dedicated +style aggregation over unstructured text. Taken per-documentQdrantindex. ThisretrievalusesCo- +together, these benchmarks show that scalable hereembed-v4.0embeddings(1536-dimensional, +multi-document reasoning remains an open cosine similarity) to fetch the top k = 15 most +challenge,withLoongemphasizingtheespecially similar chunks. These chunks are then re-ranked +demandingall-relevant,long-contextregime. byCoherererank-v4.0-fasttoyieldafinalset + +| | | | | Figure1: | OverviewofSPD-RAGArchitecture. | | | | | | | | | +| --- | --- | --- | --- | -------- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | +of top_n = 5 chunks per search. The documents Similarity-Ordered Merge & Synthesis. +t, +are split into chunks offline using LangChain’s In iteration we maintain a set of find- +RecursiveCharacterTextSplitter in Mark- ings/summaries S(t) = {s (t) ,...,s (t) }. We +| | | | | | | | | | | | 1 | n | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +downmodewithachunksizeof1000tokensand embed each summary using Cohere embed-v4.0 +| 250-tokenoverlap. | | | | | | | and compute | | a cosine-similarity | | | matrix | via | +| ----------------- | --- | --- | --- | --- | --- | --- | ----------- | --- | ------------------- | --- | --- | ------ | --- | +sklearn.metrics.pairwise.cosine_similarity. +| FindingsOutput. | | | Wedenotethefindingsofeach | | | | | | | | | | | +| --------------- | --- | --- | ------------------------- | --- | --- | --- | ---- | ------------ | --- | ---- | -------- | ------ | --- | +| | | | | | | | This | is converted | | to a | distance | matrix | | +sub-agentas +| | | | | | | | | | | (t) (t) | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ------- | ----- | --------- | --- | +| | | | | | | | D = | 1 − | cos(s | ,s ), | which | is passed | to | +| | | | | | | | ij | | | i j | | | | +o = α (q,d ) = ⟨s ,r ⟩ (1) AgglomerativeClustering with n_clusters=1, +| | | i | i i | i | i | | | | | | | | | +| ----- | ---- | ----------- | --- | -------- | -------- | --- | ----------------- | --- | --- | -------- | --- | --------- | --- | +| | | | | | | | linkage="average" | | | (UPGMA), | | producing | a | +| where | s is | the natural | | language | findings | re- | | | | | | | | +i +completedendrogramoverallcurrentsummaries. +| port from | d | and | r ∈ | [0,1] is | a scalar | rele- | | | | | | | | +| ---------------------- | --- | --- | --------------------- | -------- | -------- | ----- | ------------ | --- | ----- | ------------ | ---------- | -------- | --- | +| | i | | i | | | | | | | | | | | +| | | | | | | | We traverse | the | merge | tree | bottom-up, | greedily | | +| vance/confidencescore. | | | Theseoutputsarepassed | | | | | | | | | | | +| | | | | | | | accumulating | | nodes | into a batch | as | long as | the | +totheSynthesisLayer. +| | | | | | | | total tiktoken | | count | does not | exceed | the target | | +| ------------ | --- | --- | --- | --------- | --- | ----- | -------------- | ---- | --------- | -------- | ------ | ------------ | --- | +| Parallelism. | | All | RAG | loops for | the | docu- | | | | | | | | +| | | | | | | | budget | of B | = 750,000 | tokens. | | Semantically | | +ments are dispatched concurrently via Lang- similarsummariesaregroupedfirst,soeachbatch +Graph’s Send API fan-out, which emits one contains the most similar available summaries +| Send("document_sub_agent_node", | | | | | ...) | per | | | | | | | | +| ------------------------------- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +withinthetokenlimit. +documentandexecutesalltargetsinparallelwithin +LangGraph’sasyncruntime. AggregateandSynthesizeFindings. Foreach +batch,anLLMispromptedwiththeconcatenated +3.3 SynthesisLayer +| | | | | | | | findings, | the synthesis | | directive, | and | the original | | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | ---------- | --- | ------------ | --- | +WhilemodernLLMsfeatureextensivecontextwin- queryq toproduceanaggregatedandsynthesized +dows,synthesizinghundredsofdocumentreports +| | | | | | | | summary. | Allbatchsynthesiscallswithinanitera- | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------ | --- | --- | --- | --- | --- | +can still exceed these limits or cause ’lost in the tionareissuedconcurrentlyviaasyncio.gather. +middle’degradation. Tofuture-proofthearchitec- Themergedsummariesformtheinputforthenext +| ture for | massive | corpora, | | the Synthesis | Layer | is | | | | | | | | +| -------- | ------- | -------- | --- | ------------- | ----- | --- | --------- | ------- | --- | ---- | --------- | ------ | --- | +| | | | | | | | iteration | S(t+1). | The | same | SYNTHESIS | prompt | | +designedasadynamicmap-reducepipelinethatag- +| | | | | | | | used at | every | level | of the | merging | tree is | ap- | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----- | ----- | ------ | ------- | ------- | --- | +gregatesfindingsrecursively. TheSynthesisLayer plied even at the final level—there is no separate +receives the synthesis directives from the Coor- synthesis-onlyprompt. +| dination | Layer | and | the findings | {o | ,o ,...,o | } | | | | | | | | +| -------- | ----- | --- | ------------ | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | 1 2 | n | | | | | | | | +from the Parallel Retrieval Layer. It aggregates ContextSizeConditionandOutput. Thelayer +andsynthesizesthefindingsrecursivelythrougha evaluates whether the total findings size exceeds +similarity-orderedprocessuntilthetotalfindings the context size threshold (i.e., |S(t+1)| > 1). If +sizefallswithinthetargetcontextsizetoproduce theconditionistrue,theprocessloopsbacktothe +thefinaloutput. similarity-orderedmergeandsynthesisstep. Ifthe + +Algorithm1RecursiveSynthesis Locating, Comparison, Clustering, or Chain of +Reasoning,withcontextlengthsrangingfrom10K +| Require: | FindingsS | = {s ,...,s | },queryq, | | | | | | | | | +| -------- | --------- | ----------- | --------- | --- | --------- | ---- | ------- | --- | ----------- | --- | ---------- | +| | | 1 | n | | | | | | | | | +| | | | | | to beyond | 250K | tokens. | | We evaluate | | on the En- | +synthesisdirectived,contextsizebudgetB +Ensure: FinalOutputs⋆ glishandSet4(200k-250ktokens)portionofthe +| | | | | | benchmarkonly. | | Ourevaluationsetcomprises40 | | | | | +| ----------- | ----- | --- | --- | --- | -------------- | --- | --------------------------- | --- | --- | --- | --- | +| 1: while|S| | > 1do | | | | | | | | | | | +instancesinvolvingacademicpapersand62involv- +| 2: E | ← COHEREEMBED(S) | | | | | | | | | | | +| ---- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +D ← 1−COSINESIMILARITY(E) ingfinancialreports(102instancestotal),withthe +3: +tree ← AGGLOMERATIVECLUSTER( fourtasktypesdistributedas: SpotlightLocating +4: +(27),Comparison(15),Clustering(49),andChain +| | D, linkage | = “average”) | | | | | | | | | | +| --- | ---------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ofReasoning(11). +| 5: batches | ← | GROUPBYTOKENS( | | | | | | | | | | +| ---------- | -------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | S, tree, | B) | | | | | | | | | | +4.2 Baselines +| 6: if|batches| | | ≥ |S|then | | | | | | | | | | +| -------------- | --------- | --------- | --- | --- | --------------------- | --- | ------- | --- | --------- | -------- | --- | +| | | | | | We compare | | against | the | following | systems, | all | +| 7: | batches ← | [S] | | | | | | | | | | +| endif | | | | | backedbyGemini2.5Pro: | | | | | | | +8: +| 9: S | ← ASYNCGATHER( | | | | | | | | | | | +| ---- | ------------------- | --- | --- | -------- | ------------------------ | --- | --- | --- | --- | --------------- | --- | +| | | | | | • Baseline(FullContext): | | | | | Alldocumentsfor | | +| | SYNTHESIZE(b,q,d)∀b | | ∈ | batches) | | | | | | | | +eachinstanceareconcatenatedandprovided +10: endwhile +directlyinthecontextwindowofGemini2.5 +| return | S[0] | | | | | | | | | | | +| ------ | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +11: +Pro,representingtheoraclelong-contextup- +perbound. +conditionisfalse(thetotalfindingssizeiswithin +• NormalRAG:Standardvectorsearchtop-K +thecontextsize,leavingasinglefinalsummarys⋆), +| | | | | | retrieval | | over | the full | document | | corpus, fol- | +| ----------- | ---------- | --------------------------- | --- | ------------ | --------- | --- | ---- | --------- | -------- | -------- | ------------ | +| the process | terminates | and yields | s⋆ | as the final | | | | | | | | +| | | | | | lowed | by | LLM | reasoning | | over the | retrieved | +| | If|S(t+1)| | |S(t)|(no-progresscase),the | | | | | | | | | | +| output. | | ≥ | | | | | | | | | | +chunks. +entirecurrentlevelisforcedintoasinglebatchto +guaranteetermination. +| | | | | | • Agentic | | RAG: | A single-agent | | ReAct-style | | +| --- | --- | --- | --- | --- | --------- | --- | ---- | -------------- | --- | ----------- | --- | +Algorithm1givesthepseudocodeforthegeneral RAG system (implemented via Lang- +recursivesynthesisprocedure. Graph/LangChain) that can issue multiple +| | | | | | iterative | | retrieval | calls | over | the | full global | +| ---------------- | --- | ------------------------ | --- | --- | --------- | --- | --------- | ----- | ---- | --- | ----------- | +| BehavioronLoong. | | Inpractice,withGemini2.5 | | | | | | | | | | +Pro’s 1M-token context window and a budget of corpus. It uses the same embedding and +| | | | | | re-rankingpipelineas | | | | SPD-RAG,butlacks | | | +| --- | --- | --- | --- | --- | -------------------- | --- | --- | --- | ---------------- | --- | --- | +B = 750,000tokens,thecombinedsub-agentfind- +per-documentspecialization. +ingsforallinstancesinourLoongevaluationsetfit +withinasinglebatch,sotheconditionalcheckal- +4.3 EvaluationMetrics +| lowedoutputafteroneiteration. | | | Therecursiveloop | | | | | | | | | +| ----------------------------- | --- | --- | ---------------- | --- | --------- | --- | --------- | --- | -------- | --------- | --- | +| | | | | | We report | the | following | | metrics, | following | the | +isthereforeadesigncapabilityintendedformuch +Loongevaluationprotocol(Wangetal.,2024a): +largercorpora,wheretotalfindingssizegenuinely +exceedsthecontextsize. +| | | | | | • Avg | Score: | An | LLM-judged | | score | (0–100) | +| --- | --- | --- | --- | --- | ----- | ------ | --- | ---------- | --- | ----- | ------- | +measuringthedegreetowhichthepredicted +4 ExperimentalSetup +| | | | | | answer | | accurately | covers | | the gold | answer, | +| --- | --- | --- | --- | --- | ------ | --- | ---------- | ------ | --- | -------- | ------- | +4.1 Dataset +| | | | | | following | | the | evaluation | protocol | | of Loong | +| --- | --- | --- | --- | --- | ------------------ | --- | --- | ---------- | -------- | --- | -------- | +| | | | | | (Wangetal.,2024a). | | | | 2 | | | +WeevaluateSPD-RAGontheLoongbenchmark +| (Wangetal.,2024a). | | Loongisanextendedmulti- | | | | | | | | | | +| ------------------ | ------------ | ----------------------- | ------- | -------- | ------------------- | --- | --- | --- | ------------------- | --- | --- | +| | | | | | • PerfectRate(PR%): | | | | Percentageofqueries | | | +| document | QA instances | in both | English | and Chi- | | | | | | | | +wherethesystemreceivesaperfectscoreof +| nese, with | an average | of 11 documents | | per test | | | | | | | | +| ---------- | ---------- | --------------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +100. +| casespanningtworeal-worldscenarios: | | | | financial | | | | | | | | +| ----------------------------------- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +reports and academic papers. Each test case is 2BecausetheoriginallegacyGPT-4modelAPIisunavail- +ablefornewscaledevaluations,weadoptGPT-5astheeval- +constructedsothateverydocumentisrelevantto +| | | | | | uator. Recent | literature | | increasingly | | adopts | GPT-5 as the | +| --- | --- | --- | --- | --- | ------------- | ---------- | --- | ------------ | --- | ------ | ------------ | +the final answer ("Leave No Document Behind", standard for LLM-as-a-judge protocols due to its superior +meaningacorrectresponserequiressynthesizing alignmentwithhumangrading(Wolfsonetal.,2025).While +ourbaselinenumbersarere-computedunderthisGPT-5judge +evidencefromallprovideddocuments),andisan- +forstrictfairness,rawscoresmaynotbeperfectly1-to-1com- +notated under one of four task types: Spotlight parablewiththeoriginallypublishedLoongleaderboard. + +• AvgTokenUsage: Meanper-querytotal(in- Cost tracking. Per-query costs are computed +put+output)tokenusage. fromLangSmithrunmetadata(totaltokencounts +×per-tokenpricing. +| • Avg | Cost | (USD): | Mean | per-query | | API cost, | | | | | | | +| ----------------------------------- | ---- | ------ | ---- | --------- | --- | --------- | --------- | --- | --- | --- | --- | --- | +| enablingcost-efficiencycomparisons. | | | | | | | 5 Results | | | | | | +5.1 MainResults +| • AvgLatency(s): | | | Meanper-querywall-clock | | | | | | | | | | +| ---------------- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +time. +Table1reportsthemainresultsacrossallsystems +| | | | | | | | on our Loong | evaluation | | set. | Figure | 2 illustrates | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ---------- | --- | ---- | ------ | ------------- | +4.4 ImplementationDetails +theseperformancedifferencesacrossthefourspe- +| AllsystemsuseGemini2.5Pro(temperature0.0) | | | | | | | cifictasktypes. | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | +asthebackboneLLMforthecoordinator,merging SPD-RAGachievesanAvgScoreof58.1,sub- +| layer, and | synthesis | | layer. | SPD-RAG | | document | | | | | | | +| ---------- | --------- | --- | ------ | ------- | --- | -------- | --- | --- | --- | --- | --- | --- | +stantiallyoutperformingbothNormalRAG(33.0) +| sub-agents | use | Gemini | 2.5 | Flash | for | document- | | | | | | | +| ---------- | --- | ------ | --- | ----- | --- | --------- | ----------- | --- | ------- | --- | ---------------- | --- | +| | | | | | | | and Agentic | RAG | (32.8). | | This corresponds | to | +scopedretrieval,reducingper-documentcostwhile roughly a 25-point absolute improvement (about +preserving reasoning quality at the merging and 76% relative gain) over standard RAG baselines. +| synthesisstages. | | GPT-5isusedexclusivelyasthe | | | | | | | | | | | +| ---------------- | --- | --------------------------- | --- | --- | --- | --- | ------------ | ------- | --- | ---- | ---- | ----------- | +| | | | | | | | Furthermore, | SPD-RAG | | more | than | doubles the | +LLMjudgefortheevaluationoftheAvgscoreand Perfect Rate (PR) of Agentic RAG (18.6% vs. +is not part of the pipeline itself. Prompts for all 8.8%),indicatingthatexhaustivedocument-level +| systems, | including | baselines, | | were | optimized | for | | | | | | | +| -------- | --------- | ---------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | +processingmorefrequentlycapturesthecomplete +informationextractionandsynthesistoensurefair set of required facts. Notably, the Agentic RAG +comparison. Fullprompttemplatesforallsystems baseline,despiteconsuming∼3×moretokensthan +areprovidedinthecoderepository. NormalRAG,doesnotyieldscoreimprovements, +suggestingthatuninstructediterativeretrievalwith- +| Indexing. | Each | | document | | is | pre- | | | | | | | +| --------- | ---- | --- | --------- | --- | ------ | ---- | ---------------- | --- | -------------- | --- | ---- | ---------- | +| | | | | | | | out per-document | | specialization | | does | not effec- | +| indexed | into | a | dedicated | | Qdrant | vec- | | | | | | | +tivelyaddressthecoveragechallengeinexhaustive +| tor | collection | | using | | LangChain’s | | | | | | | | +| ------------------------------ | ---------- | --- | ----- | --- | ----------- | ------- | -------------- | ------- | ----------- | ----- | ------------ | ------------ | +| | | | | | | | multi-document | | settings. | While | the | full-context | +| RecursiveCharacterTextSplitter | | | | | | (Imple- | | | | | | | +| | | | | | | | baseline | attains | the highest | | quality with | an Avg | +mentationdetailsareprovidedintheAppendix.). +| | | | | | | | Score of | 68.0, | SPD-RAG | reaches | 85.4% | of this | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------- | ------- | ----- | ------- | +Retrievalpersub-agent. Eachsearchtoolcall performancewhileconsumingonly37.9%ofthe +performs dense vector retrieval over the docu- API cost, demonstrating a highly favorable cost- +| | | | | | | | quality trade-off. | | The | full-context | baseline | pro- | +| --------------------------- | --- | --- | --- | --- | -------------- | --- | ------------------ | --- | --- | ------------ | -------- | ---- | +| mentindex,returningthetop-K | | | | | (k = 15)chunks | | | | | | | | +using cosine similarity in Qdrant. These can- vides a strong upper bound that benefits from all +didates are subsequently re-ranked with Cohere documents in one pass. Despite this, SPD-RAG +narrowsthegapto9.9pointswhilecostingless. +| rerank-v4.0-fast,andthetoptop_n | | | | | = | 5chunks | | | | | | | +| ------------------------------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +arepassedtothesub-agentLLM. +5.2 AnalysisbyTaskType +| Merging | layer. | The | token | budget | | per merge | | | | | | | +| ------- | ------ | --- | ----- | ------ | --- | --------- | --- | --- | --- | --- | --- | --- | +Table2breaksdownperformancebyLoong’sfour +| batchisB | = 750,000tokens(75%ofGemini2.5 | | | | | | | | | | | | +| -------- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +tasktypes. +| Pro’s1Mcontextcap). | | | Agglomerativeclustering | | | | | | | | | | +| ------------------- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Thetask-typebreakdownrevealsaclearpattern. +usessklearnwithUPGMAlinkageandprecom- +| | | | | | | | For Spotlight | Locating—which | | | requires | finding | +| --- | --- | --- | --- | --- | --- | --- | ------------- | -------------- | --- | --- | -------- | ------- | +putedcosine-distancematrix. +asinglesalientfactacrossdocuments—allRAG- +| | | | | | | | based systems | | perform | reasonably | well | (69–74), | +| ------------- | --------------- | ----------------------------- | ----- | --------- | ----------- | ---------- | -------------------------------------------- | ------------ | ------- | ---------- | ----- | ----------- | +| Parallelism. | | Documentagentsrunconcurrently | | | | | | | | | | | +| | | | | | | | while the | full-context | | baseline | leads | (86.6). The | +| via LangGraph | | Send | API | fan-out. | Batch | synthe- | | | | | | | +| | | | | | | | largest gaps | between | | SPD-RAG | and | the RAG | +| sis calls | within | each | merge | iteration | | are paral- | | | | | | | +| | asyncio.gather. | | | | | | baselinesappearinClustering(+40.5ptsoverNor- | | | | | | +| lelized | with | | | | Synchronous | Co- | | | | | | | +malRAG)andChainofReasoning(+26.2ptsover +hereAPIcallsareoffloadedtothethreadpoolvia +| | | | | | | | Agentic RAG), | | tasks | that require | aggregating | or | +| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ----- | ------------ | ----------- | --- | +asyncio.to_thread. +reasoningoverevidencefrommanydocumentssi- +Safety limits. Each document sub-agent’s iter- multaneously. Ontheotherhand,itnearlymatches +ative retrieval loop is capped at 5 iterations; the the full-context baseline (42.2 vs. 42.7) on Com- +| AgenticRAGbaselinecapsat10iterations. | | | | | | | parisontasks. | | | | | | +| ------------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | + +System AvgScore PR(%) AvgInputTok AvgTotalTok AvgCost($) AvgLatency(s) +OracleLong-ContextBaseline +| Baseline(FullContext) | | 68.0 | 31.4 253,085 | 255,345 | 0.273 | 45.6 | +| --------------------- | --- | ---- | ------------ | ------- | ----- | ---- | +RAGBaselines +| NormalRAG | | 33.0 | 13.7 22,174 | 27,430 | 0.080 | 42.6 | +| ---------- | --- | ---- | ----------- | ------ | ----- | ---- | +| AgenticRAG | | 32.8 | 8.8 81,877 | 85,290 | 0.098 | 40.6 | +OurSystem +| SPD-RAG | | 58.1 | 18.6 193,954 | 205,683 | 0.103 | 54.8 | +| ------- | --- | ---- | ------------ | ------- | ----- | ---- | +Table1: MainresultsonourLoongevaluationset(102instances: 40academicpaper,62financialreport). Bold: +| bestoverall. | Underline: | bestamongRAG-basedsystems. | | | | | +| ------------ | ---------- | -------------------------- | --- | --- | --- | --- | +Figure2: ComparisonofAverageScoreacrossthefoursystems,brokendownbytasktype(SpotlightLocating, +Comparison,Clustering,andChainofReasoning). +5.3 AnalysisbyDocumentDomain +| As shown | in Table 3 | and Figure | 3 the domain | | | | +| -------- | ---------- | ---------- | ------------ | --- | --- | --- | +breakdownexposesastrikingfailuremodeofstan- +| dard RAG: | both Normal | RAG and | Agentic RAG | | | | +| --------- | ----------- | ------- | ----------- | --- | --- | --- | +achieve0%PRonacademicpaperinstances(Avg +| Scoresof15.2and16.8,respectively). | | | Academic | | | | +| ---------------------------------- | --- | --- | -------- | --- | --- | --- | +paperinstancesinLoongtendtoinvolvelong,tech- +nicaldocumentswithdistributedevidence,making +top-K retrievalespeciallysusceptibletocoverage +failures. SPD-RAGdramaticallyrecoversonthis +domain(60.0AvgScore),closingmostofthegap +| with the | full-context | baseline (78.8). | On finan- | | | | +| -------- | ------------ | ---------------- | --------- | --- | --- | --- | +Figure3: Averagescorebydocumentdomaincompar- +cialreportinstances—whichtendtobeshorterand ingtheBaseline(FullContext),NormalRAG,Agentic +morestructured—allRAG-basedsystemsperform +RAG,andSPD-RAGsystems. +better,andthegapbetweenSPD-RAG(56.9)and +thebaseline(61.0)issmaller. + +| | System | | SpotlightLoc. | | Comparison | Clustering | ChainofReasoning | | | | | +| --- | ------ | --- | ------------- | --- | ---------- | ---------- | ---------------- | ------ | --- | --- | --- | +| | | | (n=27) | | (n=15) | (n=49) | | (n=11) | | | | +Baseline(FullContext) 86.6/55.6% 42.7/26.7% 67.1/14.3% 60.9/54.5% +| | NormalRAG | | 69.7/37.0% | | 37.7/26.7% | 15.4/0.0% | | 14.8/0.0% | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | +| | AgenticRAG | | 73.4/22.2% | | 23.3/20.0% | 16.7/0.0% | | 17.9/0.0% | | | | +| | SPD-RAG | | 74.2/44.4% | | 42.2/20.0% | 57.2/4.1% | | 44.1/18.2% | | | | +Table2: AvgScore/PR(%)byLoongtasktype. EachcellshowsAvgScore/PR%. +| | | | | | Crucially, | thearchitectureof | | | SPD-RAG | | enables | +| -------- | --- | ----------- | --------------- | --- | ---------- | ----------------- | --- | --------------- | ------- | --------- | ------- | +| System | | Paper(n=40) | Financial(n=62) | | | | | | | | | +| | | | | | this | cost efficiency: | | by constraining | | retrieval | to | +| Baseline | | 78.8/30.0% | 61.0/32.3% | | | | | | | | | +NormalRAG 15.2/0.0% 44.5/22.6% isolated,single-documentspaces,wecanreliably +| AgenticRAG | | 16.8/0.0% | 43.1/14.5% | | | | | | | | | +| ---------- | --- | --------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +offloadtheiterativereasoningtoacheapermodel +| SPD-RAG | | 60.0/7.5% | 56.9/25.8% | | | | | | | | | +| ------- | --- | --------- | ---------- | --- | ------- | ----------- | --- | ------------ | --- | ------ | --------- | +| | | | | | (Gemini | 2.5 Flash). | | In contrast, | | global | iterative | +baselinesrequirethereasoningcapacityofafron- +Table3: AvgScore/PR(%)bydocumentdomain. +| | | | | | tier | model (Gemini | 2.5 | Pro) | to navigate | | the fully | +| --- | --- | --- | --- | --- | ---- | ------------- | --- | ---- | ----------- | --- | --------- | +concatenatedcorpus,drivingupcosts. +| System | | AvgScore | AvgCost($) | Score/$ | | | | | | | | +| ---------- | --- | -------- | ---------- | ------- | ------------------- | ---------- | --- | ---- | ------------------- | --- | --------- | +| Baseline | | 68.0 | 0.273 | 249.1 | | | | | | | | +| NormalRAG | | 33.0 | 0.080 | 412.5 | 6 | Discussion | | | | | | +| AgenticRAG | | 32.8 | 0.098 | 334.7 | | | | | | | | +| SPD-RAG | | 58.1 | 0.103 | 564.1 | | | | | | | | +| | | | | | WhereSPD-RAGexcels. | | | | Thetask-typeresults | | | +| | | | | | in Table | 2 confirm | our | core | hypothesis: | | architec- | +Table4: Cost–qualitytradeoff. +turaldecompositionalongthedocumentaxisyields +| | | | | | the largest | gains | precisely | | for tasks | that | require | +| --- | --- | --- | --- | --- | ----------- | -------------- | --------- | --- | ---------- | ---------- | ------- | +| | | | | | exhaustive | multi-document | | | synthesis. | Clustering | | +5.4 Cost–QualityTradeoff +tasks(+40.5overNormalRAG)andChainofRea- +| | | | | | soning | tasks (+26.2 | over | Agentic | | RAG) | both de- | +| --- | --- | --- | --- | --- | ------ | ------------ | ---- | ------- | --- | ---- | -------- | +mandthatthesystemformacoherentanswerfrom +evidencedistributedacrossallrelevantdocuments— +exactlythescenarioforwhichper-documentagents +| | | | | | aredesigned. | ForLoong’sComparisonquestions, | | | | | | +| --- | --- | --- | --- | --- | ------------ | ------------------------------ | --- | --- | --- | --- | --- | +theanswerisprimarilydeterminedbyreadingout +asmallsetofcomparablenumericattributesfrom +| | | | | | each | report and | aggregating | | them, | so once | docu- | +| --- | --- | --- | --- | --- | -------- | -------------- | ----------- | --------- | ----- | ------------ | -------- | +| | | | | | ment | agents have | extracted | | those | fields, | the syn- | +| | | | | | thesizer | is effectively | | operating | over | a structured | | +table,whichhelpsexplainwhySPD-RAG’sscore +onComparisontasks(42.2)nearlymatchesthefull- +Figure4: Cost–Qualitytradeoff. Thescatterplotillus- contextbaseline(42.7). +tratestheParetofrontierformulti-documentQA. +| | | | | | Remaining | gap | with | the | full-context | | baseline. | +| --- | --- | --- | --- | --- | --------- | --- | ---- | --- | ------------ | --- | --------- | +SPD-RAG outperforms the RAG baselines in Despite strong gains over RAG baselines, SPD- +the cost–quality trade-off. It achieves a 76% RAG lags the full-context baseline by 9.9 Avg +higheraveragescorethanNormalRAG(58.1vs. Score points overall. Three factors likely con- +33.0)whileincreasingtheper-querycostbyonly tribute: (1) Gemini 2.5 Flash, used for document +$0.023 (Table 4). In contrast, Agentic RAG is agents due to its cost efficiency, may provide +Pareto-dominated, as it incurs higher cost than weaker agentic reasoning capabilities compared +NormalRAG($0.098vs.$0.080)whiledelivering to larger models. (2) The coordinator’s sub-task +nearlyidenticalquality(32.8vs.33.0). Compared generation may under-specify queries for highly +tothefull-contextbaseline,whichachievesanAvg technical academic content, leading to incom- +Score of 68.0, SPD-RAG reaches 85.4% of the pleteinformationextractionatthedocument-agent +qualityatonly37.9%ofthecost, resultingina level. (3) The maximum context length observed +2.25× improvement in cost–quality efficiency. inLoongquerieswasapproximately250ktokens— + +onlyabout25%ofthe1M-tokencontextwindow onthecoordinator’sabilitytogeneratecomprehen- +of Gemini 2.5 Pro—potentially limiting the pres- sivesub-tasks: under-specifiedtasksleadtoincom- +sureonthefull-contextbaseline. SinceSPD-RAG plete document-level extractions, as observed on +isdesignedtooperateonsubstantiallylargercon- technicalacademicpaperinstances. Ourevaluation +texts,itsadvantagesmaybecomemorepronounced islimitedtotwodomains(academicpapersandfi- +undermoreextremelong-contextconditions. nancialreports)fromtheLoongbenchmark;legal, +medical,orenterprisedocumentcorporamayhave +| RAG baseline | | failure | on papers. | | The 0% | PR | | | | | | | +| ------------ | --- | ------- | ---------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +differentstructuralcharacteristicsthatalterrelative +| rate and | 15.2-16.8% | | Avg Score | of | both Normal | | | | | | | | +| -------- | ---------- | --- | --------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | +performance. +| RAG and | Agentic | RAG | on | academic | paper | in- | | | | | | | +| ------- | ------- | --- | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | +ThesynthesislayerofSPD-RAGisdesignedto +top-K +| stances | highlights | a systematic | | failure | of | | | | | | | | +| ------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +handlecorporawherethecombinedsub-agentout- +| retrieval | under | dense, | distributed | evidence. | | Aca- | | | | | | | +| --------- | ----- | ------ | ----------- | --------- | --- | ---- | --- | --- | --- | --- | --- | --- | +putsexceedthemodel’scontextwindow,triggering +| demic | papers | in Loong | often | require | synthesiz- | | | | | | | | +| ----- | ------ | -------- | ----- | ------- | ---------- | --- | --- | --- | --- | --- | --- | --- | +multi-roundsimilarity-basedclusteringandrecur- +ingfactsspreadacrossabstracts,methods,results, +| | | | | | | | sive synthesis. | | In our Loong | experiments, | | how- | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | ------------ | ------------ | --- | ---- | +andappendicesofmultipledocuments—aregime +ever,Gemini2.5Pro’s1M-tokencontextwassuffi- +| where exhaustive | | per-document | | coverage | proves | | | | | | | | +| ---------------- | --- | ------------ | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | +cienttoprocessallsub-agentsummariesinasingle +| highly | effective. | The | recovery | by SPD-RAG | | to | | | | | | | +| ------ | ---------- | --- | -------- | ---------- | --- | --- | ----------- | --- | -------------- | ---- | --- | ----------- | +| | | | | | | | merge step, | so | this recursive | path | was | never acti- | +60.0AvgScore/7.5%PRinthisdomaindemon- +| | | | | | | | vated. Asaresult,thescalabilitypropertiesofthe | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | +stratesthepracticalvalueofper-documentspecial- +recursivesynthesispipeline—itsimpactonanswer +ization. +| | | | | | | | quality, | redundancy | reduction, | | and cost | as docu- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ---------- | --- | -------- | -------- | +Latencyoverhead. SPD-RAG incursamodest mentcountsgrowintothehundredsorthousands— +latency overhead (54.8 s vs. 40.6–45.6 s for the remainempiricallyunvalidated. EvaluatingSPD- +| baselines), | primarily | | due to | its multi-agent | | archi- | | | | | | | +| ----------- | --------- | --- | ------ | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +RAGonmuchlargercorpora,andideallyonanew +tecture. Although document agents run in paral- benchmarkspecificallytargetinghundredstothou- +lel,end-to-endlatencyisstillgatedbythelayered sands of documents per query instance, is an im- +pipeline, which requires at least three sequential portantdirectionforfutureworkthatwouldmore +LLM calls (coordination, per-document retrieval, directlytesttheper-documentagentdesignandthe +andsynthesis)ratherthanasinglepass. intendedlarge-databaseregime. +| | | | | | | | 8 Conclusion | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | -------- | ---------------- | ---------- | ------ | +| | | | | | | | We presented | | SPD-RAG, | a hierarchical | | multi- | +| | | | | | | | agent framework | | designed | for | exhaustive | multi- | +| | | | | | | | documentquestionanswering. | | | Byassigningaded- | | | +icatedagenticRAGmoduletoeachdocumentand +mergingtheirfindingsthroughasimilarity-ordered, +| | | | | | | | token-bounded | | tree, our | system | effectively | miti- | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | ---------- | --------- | --------- | ----------- | ---------- | +| | | | | | | | gatesthelimitationsofbothtraditionaltop-K | | | | | re- | +| | | | | | | | trieval (incomplete | | coverage) | and | isolated | long- | +| | | | | | | | context | processing | (degraded | reasoning | | at scale). | +Figure5: Averageper-querylatencyacrosssystems. ExperimentsonthechallengingLoongbenchmark +demonstratethatSPD-RAGachievesanaverage +scoreof58.1,outperformingNormalRAG(33.0) +7 LimitationsandFutureWork +andAgenticRAG(32.8)byapproximately25ab- +SPD-RAG incurs higher per-query LLM call solute points (76% higher average score), while +countsthansingle-passsystems—onecallperdoc- operating at only 37.9% of the API cost of the +ument agent plus a final synthesis step—which oracle full-context baseline (68.0). These per- +increasestotaltokenusagecomparedtostandard formance gains are most pronounced in tasks re- +RAGbaselines,thoughitremainsatroughlyone- quiring deep cross-document synthesis—such as +thirdthecostofthefull-contextbaseline($0.10vs. Clustering(+40.5points)andChainofReasoning +$0.27perquery)duetotheuseofGemini2.5Flash (+26.2points)—aswellasondenseacademicpa- +fordocumentsub-agents. Systemqualitydepends perswherestandardRAGmethodsentirelyfail. + +Ultimately, our findings demonstrate that for TaichengGuo,XiuyingChen,YaqiWang,RuidiChang, +complex information-seeking queries over large ShichaoPei,NiteshV.Chawla,OlafWiest,andXi- +| | | | | | | | angliangZhang.2024. | | | Largelanguagemodel-based | | | +| -------- | --- | ----------- | --- | ------------ | --- | -------- | ------------------- | --- | ------------------------------- | ------------------------ | --- | --- | +| corpora, | how | information | | is processed | is | crucial. | | | | | | | +| | | | | | | | multi-agents: | | Asurveyofprogressandchallenges. | | | | +Specifically,ensuringthateachdocumentreceives +| | | | | | | | Preprint,arXiv:2402.01680. | | | ArXiv:2402.01680. | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | ----------------- | --- | --- | +exhaustive,dedicatedagenticattentionprovestobe +| | | | | | | | Siwei Han, | Peng | Xia, Ruiyi | Zhang, | Tong | Sun, Yun | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---- | ---------- | ------ | ---- | -------- | +amoreeffective,cost-efficient,andscalablestrat- +| | | | | | | | Li, Hongtu | Zhu, | and | Huaxiu Yao. | 2025. | MDocA- | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---- | --- | ----------- | ----- | ------ | +egythansimplyexpandinghowmuchrawcontext +gent: Amulti-modalmulti-agentframeworkfordoc- +asinglemodelcanconsumeinonepass. umentunderstanding. Preprint,arXiv:2503.13964. +ArXiv:2503.13964. +9 EthicsStatement +| | | | | | | | Yubin Kim, | Ken | Gu, Chanwoo | Park, | Chunjong | Park, | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ----------- | ----- | -------- | ----- | +SPD-RAG is a retrieval-augmented generation SamuelSchmidgall,A.AliHeydari,YaoYan,Zhi- +hanZhang,YuchenZhuang,MarkMalhotra,PaulPu +| framework | intended | | to improve | access | | to infor- | | | | | | | +| --------- | -------- | --- | ---------- | ------ | --- | --------- | --- | --- | --- | --- | --- | --- | +Liang,HaeWonPark,YuzheYang,XuhaiXu,Yilun +| mation | distributed | across | | large document | | collec- | | | | | | | +| ------ | ----------- | ------ | --- | -------------- | --- | ------- | ----------- | --- | ---------- | -------- | ------ | ------- | +| | | | | | | | Du, Shwetak | | Patel, Tim | Althoff, | Daniel | McDuff, | +tions. ThesystemreliesoncommercialLLMAPIs and Xin Liu. 2025. Towards a science of scal- +(Gemini2.5,Cohere),andaswithanyLLM-based ing agent systems. Preprint, arXiv:2512.08296. +ArXiv:2512.08296. +| pipeline, | outputs | may | contain | factual | errors | and | | | | | | | +| --------- | ------- | --- | ------- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | +should not be treated as authoritative without hu- PatrickLewis,EthanPerez,AleksandraPiktus,Fabio +man verification. The increased number of API Petroni,VladimirKarpukhin,NamanGoyal,Hein- +| | | | | | | | richKüttler, | | MikeLewis, | Wen-tauYih, | | TimRock- | +| --------- | ------ | -------- | ------------ | ----------- | ------- | --------- | ------------------- | --------- | ---------- | ----------- | ----- | ------------ | +| calls per | query | relative | to | single-pass | systems | re- | | | | | | | +| | | | | | | | täschel, | Sebastian | Riedel, | and | Douwe | Kiela. 2020. | +| sults in | higher | energy | consumption, | | a | cost that | | | | | | | +| | | | | | | | Retrieval-augmented | | | generation | for | knowledge- | +should be weighed against the accuracy gains in intensiveNLPtasks. InAdvancesinNeuralInforma- +deploymentdecisions. Ourbenchmarkevaluations tionProcessingSystems. +| are conducted | | on publicly | | available | datasets | and | | | | | | | +| ------------- | --- | ----------- | --- | --------- | -------- | --- | -------- | -------- | ------------- | --- | ------------- | --- | +| | | | | | | | TengLin, | YuyuLuo, | HonglinZhang, | | JichengZhang, | | +involvenohumansubjectsorpersonaldata. Dur- Chunlin Liu, Kaishun Wu, and Nan Tang. 2025. +ingthepreparationofthismanuscript,AI-assisted MEBench: Benchmarkinglargelanguagemodelsfor +| | | | | | | | cross-documentmulti-entityquestionanswering. | | | | | In | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | +writingtoolswereusedtohelprefinelanguageand +ProceedingsoftheConferenceonEmpiricalMethods +| improveclarity. | | Allresearchdesign,experiments, | | | | | | | | | | | +| --------------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +inNaturalLanguageProcessing(EMNLP). +analysis,andfinaldecisionsregardingthecontent +NelsonF.Liu,KevinLin,JohnHewitt,AshwinParan- +wereconductedandverifiedbytheauthors. +jape,MicheleBevilacqua,FabioPetroni,andPercy +| | | | | | | | Liang.2023. | | Lostinthemiddle: | | Howlanguagemod- | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | ---------------------------- | --- | --------------- | --- | +| | | | | | | | elsuselongcontexts. | | TransactionsoftheAssociation | | | | +References +forComputationalLinguistics,12. +| Haonan | Bian. | 2025. | LLM-empowered | | | knowl- | | | | | | | +| ------ | ----- | ----- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | +SeijiMaekawa,HayateIso,andNikitaBhutani.2024. +| edge | graph | construction: | | A survey. | | Preprint, | | | | | | | +| ----------------- | ----- | ------------- | ----------------- | --------- | --- | --------- | ------------------------------------- | --- | --- | --- | --- | ------- | +| | | | | | | | Holisticreasoningwithlong-contextLMs: | | | | | Abench- | +| arXiv:2510.20345. | | | ArXiv:2510.20345. | | | | | | | | | | +markfordatabaseoperationsonmassivetextualdata. +Darren Edge, Ha Trinh, Newman Cheng, Joshua Preprint,arXiv:2410.11996. ArXiv:2410.11996. +| Bradley, | Alex | Chao, | | Apurva | Mody, | Steven | | | | | | | +| -------- | ---- | ----- | --- | ------ | ----- | ------ | -------- | ------ | ----- | ---------- | --- | ------------ | +| | | | | | | | David J. | Moore. | 2025. | A taxonomy | of | hierarchical | +Truitt, Dasha Metropolitansky, Robert Osazuwa multi-agentsystems: Designpatterns,coordination +| Ness, | and | Jonathan | Larson. | 2024. | | From lo- | | | | | | | +| --------------------- | ------- | -------- | -------------------------- | -------- | --- | --------- | ----------------- | --- | ----------------- | ------------- | --- | --------- | +| | | | | | | | mechanisms, | | and industrial | applications. | | Preprint, | +| cal to | global: | A graph | RAG | approach | | to query- | | | | | | | +| | | | | | | | arXiv:2508.12683. | | ArXiv:2508.12683. | | | | +| focusedsummarization. | | | Preprint,arXiv:2404.16130. | | | | | | | | | | +ArXiv:2404.16130. +| | | | | | | | LituOuandMirellaLapata.2025. | | | | Context-awarehier- | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ------------------ | --- | +archicalmergingforlongdocumentsummarization. +| Gemini | Team, | Petko | Georgiev, | Ving | Ian Lei, | Ryan | | | | | | | +| -------- | ----- | ----- | --------- | ------- | -------- | ------- | -------------------------- | --- | --- | ----------------- | --- | --- | +| | | | | | | | Preprint,arXiv:2502.00977. | | | ArXiv:2502.00977. | | | +| Burnell, | Libin | Bai, | Anmol | Gulati, | Garrett | Tanzer, | | | | | | | +DamienVincent,ZhufengPan,ShiboWang,and1 Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh +others. 2024. Gemini 1.5: Unlocking multimodal Khanna, Anna Goldie, and Christopher D. Man- +understandingacrossmillionsoftokensofcontext. ning. 2024. RAPTOR: Recursive abstractive pro- +Preprint,arXiv:2403.05530. ArXiv:2403.05530. cessing for tree-organized retrieval. Preprint, +| | | | | | | | arXiv:2401.18059. | | ArXiv:2401.18059. | | | | +| ---------- | ------ | --- | ------- | ------- | ----- | --- | ----------------- | --- | ----------------- | --- | --- | --- | +| Jiani Guo, | Zuchao | Li, | Jie Wu, | Qianren | Wang, | Yun | | | | | | | +Li, Lefei Zhang, Hai Zhao, and Yujiu Yang. 2025. LiSun,LiuHe,ShuyueJia,YangfanHe,andChenyu +ToM:Leveragingtree-orientedMapReduceforlong- You. 2025. DocAgent: An agentic framework for +contextreasoninginlargelanguagemodels. InPro- multi-modallong-contextdocumentunderstanding. +ceedingsoftheConferenceonEmpiricalMethodsin InProceedingsoftheConferenceonEmpiricalMeth- +NaturalLanguageProcessing(EMNLP). odsinNaturalLanguageProcessing(EMNLP). + +V.A.Traag,L.Waltman,andN.J.vanEck.2019. From extraction tasks and write a synthesis +louvaintoleiden: Guaranteeingwell-connectedcom- directive to guide the downstream +| munities. | ScientificReports,9(1):5233. | | | | | | synthesizer. | | | | | | +| --------- | ---------------------------- | --- | --- | --- | --- | --- | ------------- | --- | ------ | --- | ---- | -------- | +| | | | | | | | 2. Sub-agents | | (next) | — | Each | document | +Minzheng Wang, Longze Chen, Cheng Fu, Shengyi is assigned one worker that runs your +Liao,XinghuaZhang,BingliWu,HaiyangYu,Nan tasks independently. Workers search +Xu,LeiZhang,RunLuo,YunshuiLi,MinYang,Fei their document and report raw findings. +| | | | | | | | They run | in parallel | | and | cannot | see each | +| ------------------------- | --- | --- | --- | --------------- | --- | --- | -------- | ----------- | --- | --- | ------ | -------- | +| Huang,andYongbinLi.2024a. | | | | Leavenodocument | | | | | | | | | +other’s results. +| behind: | Benchmarkinglong-contextLLMswithex- | | | | | | | | | | | | +| ------------------ | ----------------------------------- | --- | -------------------------- | --- | --- | --- | -------------- | -------- | ------ | ------ | ------------ | -------- | +| | | | | | | | 3. Synthesizer | | (last) | — | A downstream | | +| tendedmulti-docQA. | | | Preprint,arXiv:2406.17419. | | | | | | | | | | +| | | | | | | | agent | receives | ALL | worker | | findings | +ArXiv:2406.17419. +| | | | | | | | and merges | | them | into | a | single | +| ---------- | ------ | --- | ----- | -------- | ----- | ------ | ------------------- | --------- | ---- | --------- | ------- | ------- | +| | | | | | | | coherent | response. | | It | follows | your | +| Qian Wang, | Tianyu | | Wang, | Zhenheng | Tang, | Qinbin | | | | | | | +| | | | | | | | synthesis_directive | | | to decide | | what to | +Li,NuoChen,JingshengLiang,andBingshengHe. +| | | | | | | | prioritize | and | how | to | structure | the | +| ------ | ---------- | --- | -------------------------- | --- | --- | --- | ---------- | --- | --- | --- | --------- | --- | +| 2024b. | Megaagent: | | Alarge-scaleautonomousLLM- | | | | | | | | | | +output. +basedmulti-agentsystemwithoutpredefinedSOPs. +| | | | | | | | You CANNOT | see | the | names, | types, | or | +| -------------------------- | --- | --- | --- | ----------------- | --- | --- | ---------- | ------ | ---------- | ------ | ------ | --- | +| Preprint,arXiv:2408.09955. | | | | ArXiv:2408.09955. | | | | | | | | | +| | | | | | | | content | of the | documents. | | | | +TomerWolfson,HarshTrivedi,MorGeva,YoavGold- Your ONLY job in this step: +| berg, | Dan Roth, | Tushar | Khot, | Ashish | | Sabharwal, | | | | | | | +| ------------------ | --------- | ------ | -------------------------- | ------- | --------- | ---------- | ---------- | ---------- | ---------------------- | ----------- | -------------- | ------- | +| | | | | | | | • Produce | | a list | of | subagent_todos | | +| and Reut | Tsarfaty. | | 2025. | Monaco: | | More nat- | | | | | | | +| | | | | | | | (as | structured | | output): | | precise | +| ural and | complex | | questions | for | reasoning | across | | | | | | | +| | | | | | | | extraction | | tasks | | that | will be | +| dozensofdocuments. | | | Preprint,arXiv:2508.11133. | | | | | | | | | | +| | | | | | | | executed | | independently | | against | EACH | +| ArXiv:2508.11133. | | | | | | | document | | by the | sub-agents. | | | +| | | | | | | | • Produce | | a synthesis_directive: | | | a | +Wentao Zhang, Liang Zeng, Yuzhen Xiao, Yong- concise instruction (2–4 sentences) +cong Li, Ce Cui, Yilei Zhao, Rui Hu, Yang Liu, for the synthesizer. +| Yahui | Zhou, | and Bo | An. | 2025. | AgentOrchestra: | | | | | | | | +| ----- | ----- | ------ | --- | ----- | --------------- | --- | --- | --- | --- | --- | --- | --- | +A hierarchical multi-agent framework for general- Todo-writing rules: +| purpose | task | solving. | Preprint, | arXiv:2506.12508. | | | | | | | | | +| ----------------- | ---- | -------- | --------- | ----------------- | --- | --- | ----------- | --- | ----------- | ---- | ------------- | ---- | +| | | | | | | | • Decompose | | the | user | query | into | +| ArXiv:2506.12508. | | | | | | | concrete | | information | | requirements. | | +Jun Zhao, Can Zu, Hao Xu, Yi Lu, Wei He, Yiwen • Each todo must be self-contained +and unambiguous. +| Ding, | Tao Gui, | Qi | Zhang, | and | Xuanjing | Huang. | | | | | | | +| ----- | ---------- | --- | --------------------------- | --- | -------- | ------ | -------- | --- | ------ | ----- | ---- | ----- | +| | | | | | | | • Prefer | | atomic | tasks | over | broad | +| 2024. | LongAgent: | | Scalinglanguagemodelsto128k | | | | | | | | | | +tasks. +| contextthroughmulti-agentcollaboration. | | | | | | Preprint, | | | | | | | +| --------------------------------------- | ----- | --- | ----------------- | ----- | ---- | --------- | --------- | --- | --------------- | ------------ | ------------ | -------- | +| | | | | | | | • Include | | coverage | for: | definitions, | | +| arXiv:2402.11550. | | | ArXiv:2402.11550. | | | | | | | | | | +| | | | | | | | numeric | | values, | constraints, | | edge | +| | | | | | | | cases, | | error | modes, | and | any | +| Zihan Zhou, | Chong | | Li, Xinyi | Chen, | Shuo | Wang, | | | | | | | +| | | | | | | | explicit | | recommendations | | | required | +YuChao,ZhiliLi,HaoyuWang,QiShi,ZhixingTan, +| | | | | | | | by | the user | query. | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ------ | --- | --- | --- | +XuHan,XiaodongShi,ZhiyuanLiu,andMaosong +| | | | | | | | • Design | a | robust | extraction | | list | +| ---------- | ---------- | -------------- | ----- | ----- | ---------- | ------- | -------- | ----- | ------ | ------------ | --- | ------ | +| Sun. 2025. | | LLM×MapReduce: | | | Simplified | long- | | | | | | | +| | | | | | | | that | works | for | ANY document | | in the | +| sequence | processing | | using | large | language | models. | | | | | | | +set. +InProceedingsoftheAnnualMeetingoftheAssocia- +| tionforComputationalLinguistics(ACL). | | | | | | | Important | Constraints: | | | | | +| ---------------------------------------- | --- | ----- | --------- | --- | --- | ---------- | --------- | ------------ | ----- | -------- | ------- | -------- | +| | | | | | | | Do | not assume | any | document | | contains | +| AndrewZhu,AlyssaHwang,LiamDugan,andChris | | | | | | | • | | | | | | +| | | | | | | | the | answer. | Write | | todos | that can | +| Callison-Burch. | | 2024. | FanOutQA: | | A | multi-hop, | | | | | | | +| | | | | | | | be | answered | with | either | “Found” | or | +multi-documentquestionansweringbenchmarkfor +| | | | | | | | “Not | found | in | this document.” | | | +| -------------- | --- | ------- | --- | ----------- | --- | ---------- | ------ | ----- | ------ | --------------- | ----------- | --- | +| large language | | models. | In | Proceedings | | of the An- | | | | | | | +| | | | | | | | • Tell | the | worker | WHAT | to extract, | not | +nualMeetingoftheAssociationforComputational +| | | | | | | | HOW | to extract | | it. | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | +Linguistics(ACL). +| | | | | | | | • Do | not | synthesize, | | summarize, | or | +| --- | --- | --- | --- | --- | --- | --- | ------- | --- | ----------- | --- | ---------- | ----- | +| | | | | | | | attempt | | to answer | | the user | query | +A Prompts +yourself. +| All | prompts | | are | | defined | in | | | | | | | +| --- | ------- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +A.2DocumentSub-AgentPrompt +backend/core/prompts.py. +(RESEARCH_SYSTEM_PROMPT) +| A.1CoordinatorPrompt | | | | | | | You are | a sub-agent | | investigating | | a | +| ------------------------ | --- | --- | --- | --- | --- | --- | ---------------- | ----------- | ------------ | ------------- | --- | --- | +| (LEAD_RESEARCHER_PROMPT) | | | | | | | single document: | | {file_name}. | | | | +You are a lead researcher coordinating You operate inside an iterative +a RAG-based analysis to answer a user retrieval loop. On each turn you output +| query. | | | | | | | exactly | one structured | | action: | | | +| -------- | --- | ------------- | --- | --- | --- | --- | ------------------ | -------------- | --- | ------- | --- | ------- | +| Pipeline | | Architecture: | | | | | • action="search": | | | Issue | a | focused | +1. YOU (now) — Decompose the query into query to retrieve information from + +the document. Set query to a • Follow the synthesis directive +specific, targeted search string. above — it defines your main goal, +Set reasoning to why this query is priorities, and output structure. +| | needed. | | | | | | • | Keep only | information | | that | directly | +| --- | ------- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ---- | -------- | +• action="finalize": You have helps answer the query. Discard +| | gathered | | sufficient | evidence | to | | | tangential | content. | | | | +| --- | -------- | --- | ---------- | -------- | ---------- | --- | --- | ------------ | ------------ | -------- | ------ | ------ | +| | address | ALL | assigned | | tasks. Set | | | | | | | | +| | | | | | | | • | Preserve | exact | numbers, | | names, | +| | findings | to | your | complete | extracted | | | | | | | | +| | | | | | | | | dates, | and caveats. | | | | +| | report. | Set | reasoning | | to a brief | | | | | | | | +| | | | | | | | • | Remove | redundancy. | | If the | same | +| | summary | of | what you | found. | | | | | | | | | +| | | | | | | | | fact appears | | multiple | times, | keep | +it once. +| | Investigation | principles: | | | | | | | | | | | +| --- | ------------- | -------------- | ---- | ----- | --------- | --- | --- | ------- | -------- | --------- | ----- | -------- | +| | | | | | | | • | Do NOT | invent | or infer | facts | not | +| | | | | | | | | present | in the | findings. | | | +| | 1. Start | by identifying | | the | key facts | | | | | | | | +| | required | for | each | task. | | | • | Same | language | as | the | findings | +2. Issue ONE focused query per turn (default English if mixed). +| | — prefer | specific | | terms | over broad | | | | | | | | +| --- | --------- | -------- | -------- | ----- | ----------- | --- | --------------------- | --- | --- | --- | --- | --- | +| | phrases. | | | | | B | ImplementationDetails | | | | | | +| | 3. Expand | to | synonyms | or | paraphrases | | | | | | | | +Models. +if initial results are incomplete. The coordinator, the merging layer, +4. Issue separate queries for and the synthesis layer use gemini-2.5-pro +| | different | | aspects | of a | task. | | | | | | | | +| --- | --------- | --------- | ------- | ----- | ------- | ------------ | --- | ----- | --- | -------- | --- | ---------- | +| | | | | | | (temperature | | 0.0). | | Document | | sub-agents | +| | 5. Keep | searching | until | every | task is | | | | | | | | +covered with concrete evidence or use gemini-2.5-flash (temperature 0.0) for +| | confirmed | | absent. | | | document-scoped | | | retrieval. | | The | LLM judge | +| --- | --------- | --- | ------- | --- | --- | --------------- | --- | --- | ---------- | --- | --- | --------- | +6. Do NOT attempt to answer the user (GPT-5) is separate and is used only for the +| | query | directly | — | extract | raw facts | | | | | | | | +| --- | ----- | -------- | --- | ------- | --------- | ------------------------ | --- | --- | --- | ----------------- | --- | --- | +| | | | | | | evaluationoftheAvgscore. | | | | Modelconstantsare | | | +only. +definedinbackend/shared/constants.py. +| | 7. Do | NOT finalize | | a task | as ’Not | | | | | | | | +| --- | ----------------- | ------------ | ------------ | ---------- | ----------- | ----------------------------- | --- | --- | --- | ------------------------- | --- | ----------- | +| | found’ | after | only | a single | focused | | | | | | | | +| | search. | | At | least | attempt 2 | Hyperparameters. | | | | | | | +| | focused | searches | | before | concluding | | | | | | | | +| | information | | is absent. | | | | | | | | | | +| | | | | | | Parameter | | | | Value | | | +| | 8. Do NOT | focus | on very | specific | terms, | | | | | | | | +| | | | | | | Chunksize(tokens) | | | | 1000 | | | +| | but | rather | on the | general | context | | | | | | | | +| | | | | | | Chunkoverlap(tokens) | | | | 250 | | | +| | of the | task. | Always | try | to find the | | | | | | | | +| | | | | | | Tokenizer | | | | tiktoken | | cl100k_base | +| | most | relevant | information. | | | | | | | | | | +| | | | | | | Embeddingmodel | | | | Cohereembed-v4.0 | | | +| | 9. You | must use | AT MOST | 5 | searches to | | | | | | | | +| | | | | | | Embeddingdimension | | | | 1536 | | | +| | answer | a task. | DO | NOT | OVERUSE THE | | | | | | | | +| | | | | | | Similaritymetric | | | | Cosine | | | +| | SEARCH | TOOL. | | | | | | | | | | | +| | | | | | | MergetokenbudgetB | | | | 750,000tokens | | | +| | | | | | | Denseretrievaltop-K | | | | 15 | | | +| | Finalize | report | format: | | For each | | | | | | | | +| | | | | | | Reranker | | | | Coherererank-v4.0-fast | | | +| | assigned | task, | output | either | “Found: | | | | | | | | +| | | | | | | Rerankertop-N | | | | 5 | | | +| | [exact answer | with | supporting | | evidence]” | | | | | | | | +| | | | | | | Sub-agentiterationsafetylimit | | | | 5 | | | +| | or “Not found | in | this | document.” | Always | | | | | | | | +| | | | | | | Clusteringalgorithm | | | | Agglomerative(UPGMA) | | | +| | report exact | numbers, | | names, | and dates. | | | | | | | | +| | | | | | | Clusteringmetric | | | | Precomputedcosinedistance | | | +| | Never approximate | | or | infer | beyond the | | | | | | | | +document. +| | | | | | | Cost | | measurement. | | Per-query | | costs | +| --- | --- | --- | --- | --- | --- | ---- | ------ | ------------ | --------- | --------- | --- | -------- | +| | | | | | | are | pulled | from | LangSmith | | run | metadata | +A.3Merge/SynthesisPrompt +| (SYNTHESIS_PROMPT) | | | | | | (extract_langsmith_info.py) | | | | | using | model- | +| ------------------ | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --- | --- | ----- | ------ | +specificper-tokenpricing. +Thispromptisuseduniformlyateverylevelofthe +recursivemergingtree,includingthefinalsynthesis +step. Thereisnoseparatefinal-answerprompt. +| | You are | a research | synthesizer. | | Merge | | | | | | | | +| --- | ------------- | ----------------- | ------------ | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +| | the following | | sub-agent | findings | into | | | | | | | | +| | one compact, | information-dense | | | response | | | | | | | | +| | for the user. | | | | | | | | | | | | +| | Query: {query | } | | | | | | | | | | | +| | Synthesis | Directive: | | | | | | | | | | | +{synthesis_directive} +| | Findings | Batch: | {findings | } | | | | | | | | | +| --- | -------- | ------ | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Rules: \ No newline at end of file diff --git a/q126/random_k2/random_2.md b/q126/random_k2/random_2.md index 5d5d4bf748b05f616c8399e8fe70690f3bccb9c7..f2e0cee911212728b5cd229fe67ae344991a2f8f 100644 --- a/q126/random_k2/random_2.md +++ b/q126/random_k2/random_2.md @@ -1,1678 +1,3886 @@ -| | | Learning | | Fine-Grained | | Grounded | Citations | | for | | | -| --- | --- | -------- | ---------- | ------------ | ----- | -------- | --------- | --- | --- | --- | --- | -| | | | Attributed | | Large | Language | Models | | | | | -LeiHuang1,XiaochengFeng1,2*,WeitaoMa1,YuxuanGu1,WeihongZhong1,XiachongFeng1 -WeijiangYu3,WeihuaPeng3,DuyuTang3,DandanTu3,BingQin1,2 -1HarbinInstituteofTechnology,Harbin,China -| | | 2 PengChengLaboratory | | | | 3HuaweiInc.,Shenzhen,China | | | | | | -| --- | --- | --------------------- | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | -{lhuang,xcfeng,wtma,yxgu,whzhong,xiachongfeng,qinb}@ir.hit.edu.cn -{weijiangyu8,pengwh.hit}@gmail.com,{tangduyu,tudandan}@huawei.com -Abstract -Why is it bad to eat cookie dough for risk of salmonella but things -like Cookie Dough Bites are ok? -| Despite | the | impressive | performance | | on | | | | | | | -| ------- | --- | ---------- | ----------- | --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Salmonella is most commonly caused by eating undercooked | | | | | 1 | -information-seekingtasks,largelanguagemod- or raw foods like eggs or meat. (...) You know how your mom -4202 guA 8 ]LC.sc[ 1v86540.8042:viXra els (LLMs) still struggle with hallucinations. always warned you not to eat raw cookie dough? This is why. -| Attributed | LLMs, | which | augment | generated | | | | | | | | -| ---------- | ----- | ----- | ------- | --------- | --- | --------------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Food safety recommendations encourage people to cook eggs until | | | | | 2 | -text within-line citations, haveshownpoten- the white and yolk are firm in order to kill any bacteria. (...) However, -tial in mitigating hallucinations and improv- anyone making cookies can do things to reduce this risk by -using pasteurized egg products. -| ingverifiability. | | However,currentapproaches | | | | | | | | | | -| ----------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -suffer from suboptimal citation quality due Raw cookie dough can carry the risk of salmonella contamination, which can -to their reliance on in-context learning. Fur- lead to foodborne illness if the dough is eaten raw  1  . However, products -like Cookie Dough Bites are generally safe because they are made with -| thermore, | the | practice | of citing | only | coarse | pasteurized eggs  1    2  . | | | | | | -| --------- | --- | -------- | --------- | ---- | ------ | --------------------------- | --- | --- | --- | --- | --- | -documentidentifiersmakesitchallengingfor -Difficult to verify and subject to -userstoperformfine-grainedverification. In hallucinations and citation errors -| this work, | we | introduce | FRONT, | a | training | | | | | | | -| ---------- | --- | --------- | ------ | --- | -------- | --- | --- | --- | --- | --- | --- | - G  R  O   U  N  D  I N  G     1  salmonella is most commonly caused by eating -framework designed to teach LLMs to gen- undercooked or raw foods like eggs or meat  2  Anyone making cookies -can do things to reduce this risk by using pasteurized egg products. -erateFine-gRainedgrOuNdedciTations. By  A  N  S  W   E  R    Eating raw cookie dough can be risky due to salmonella, -groundingmodeloutputsinfine-grainedsup- commonly found in raw eggs  1 . However, this concern is mitigated in -products like Cookie Dough Bites, as the products use pasteurized egg -portingquotes,thesequotesguidethegenera- -products, which are safe to eat raw 2  . -tionofgroundedandconsistentresponses,not -onlyimprovingcitationqualitybutalsofacili- Interpretable and accurately cited -tatingfine-grainedverification.Experimentson -| | | | | | | Figure1: | Comparedwiththecurrentattributedsystems, | | | | | -| --- | --- | --- | --- | --- | --- | -------- | ---------------------------------------- | --- | --- | --- | --- | -theALCEbenchmarkdemonstratetheefficacy -thecoreideabehindFRONTistofirstselectthesupport- -ofFRONTingeneratingsuperiorgroundedre- -| | | | | | | ing quotes | from | retrieved | sources and | then | condition | -| ------------------------------------ | --- | --- | --- | --- | ---- | ---------- | ---- | --------- | ----------- | ---- | --------- | -| sponsesandhighlysupportivecitations. | | | | | With | | | | | | | -thegenerationprocessonthem,ensuringgroundedre- -LLaMA-2-7B,theframeworksignificantlyout- -sponsesandaccuratecitations. -| performsallthebaselines, | | | achievinganaver- | | | | | | | | | -| ------------------------ | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -ageof14.21%improvementincitationquality -acrossalldatasets,evensurpassingChatGPT1. SuchprevalenceofhallucinationsinLLMout- -| | | | | | | puts has | motivated | the | development | of | attributed | -| --- | --- | --- | --- | --- | --- | -------- | --------- | --- | ----------- | --- | ---------- | -1 Introduction -| | | | | | | systems | (Nakano | et al., | 2021; Thoppilan | | et al., | -| --- | --- | --- | --- | --- | --- | ------- | ------- | ------- | --------------- | --- | ------- | -Therecentadventoflargelanguagemodels(LLMs) 2022;Menicketal.,2022),suchasNewBing2 and -(Touvronetal.,2023;OpenAI,2023;Zhaoetal., Perplexity3,whereLLMsareallowedtogenerate -2023) has taken the world by storm, fueling a responses with in-line citations. Not only does it -| paradigm | shift in | information | | acquisition | (Zhu | | | | | | | -| -------- | -------- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | -improvefactualityandalleviatehallucinations,but -etal.,2023). Despitetheircompellingperformance, italsosimplifiesuserverificationofmodeloutputs, -LLMs still struggle with hallucinations (Ji et al., furtherenhancingtheverifiabilityofLLMs. -2023;Huangetal.,2023),atendencytofabricate Despiterecentadvancements,currentattributed -| non-existent | facts | or generate | unfaithful | | content. | | | | | | | -| ------------ | ----- | ----------- | ---------- | --- | -------- | -------------------------------------- | --- | --- | --- | --- | -------- | -| | | | | | | LLMsstillexposesignificantlimitations. | | | | | Firstly, | -This issue further poses a risk of misinformation current approaches predominantly rely on either -dissemination(ChenandShu,2023),directlyim- in-contextlearning(Gaoetal.,2023b)orpost-hoc -pactingthereliabilityandtrustworthinessofLLMs. -retrieval(Gaoetal.,2023a)toachieveattribution, -*CorrespondingAuthor -1Our data and code can be found at: https://github. 2https://www.bing.com/chat -com/LuckyyySTA/Fine-grained-Attribution. 3https://www.perplexity.ai - -| | | | | | | | 2 RelatedWork | | | | | | | -| ------- | ----------- | ----------- | --- | ---------- | --- | ------ | ------------- | --- | --- | --- | --- | --- | --- | -| lacking | an inherent | attribution | | capability | | within | | | | | | | | -LLMs,therebyresultingincompromisedcitation -| | | | | | | | RetrievalAugmentedGeneration. | | | | | Recently,re- | | -| ----------------------- | --------- | --- | ----------------------- | ---- | --------- | ---- | ----------------------------- | --- | ---------- | --- | ----- | ------------ | --- | -| quality(Liuetal.,2023). | | | Secondly,thesecitations | | | | | | | | | | | -| | | | | | | | trieval augmented | | generation | | (RAG) | (Karpukhin | | -| are typically | presented | | in the | form | of either | doc- | | | | | | | | -ument identifiers (Nakano et al., 2021) or URLs etal.,2020;Lewisetal.,2020;Fengetal.,2023; -Gaoetal.,2023c)hasshownpromiseinknowledge- -| (Thoppilan | et al., | 2022). | Such | coarse | attribution | | | | | | | | | -| ---------- | ------- | ------ | ---- | ------ | ----------- | --- | --------- | ------ | ---------------- | --- | --------- | --- | ----- | -| | | | | | | | intensive | tasks. | By incorporating | | retrieved | | docu- | -complicatestheprocessforuserstoperformfine- -| | | | | | | | ments, | LLMsareequippedwithup-to-dateinfor- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------------------------- | --- | --- | --- | --- | --- | -grainedverification(e.g.,pinpointspecificsupport- -| | | | | | | | mation, | significantly | mitigating | | knowledge | | gaps. | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | ---------- | --- | --------- | --- | ----- | -ingevidence),particularlyinlengthydocuments. -| | | | | | | | However, | recent | studies | (Shi | et al., | 2023; | Yoran | -| ------- | ---- | ------ | ---------- | --- | ---------- | ---- | -------- | ------ | ------- | ---- | ------- | ----- | ----- | -| To this | end, | we aim | to advance | | attributed | text | | | | | | | | -etal.,2023;Xuetal.,2023a;Zhuetal.,2024)have -generationbyempoweringLLMswithfine-grained -attributionability. However,onechallengecomes revealed that existing retrieval-augmented LLMs -| | | | | | | | struggle | to handle | irrelevant | | or contradictory | | re- | -| --- | --- | --- | --- | --- | --- | --- | -------- | --------- | ---------- | --- | ---------------- | --- | --- | -fromtheacquisitionofhigh-qualityattributiondata -trievaldocumentsandeffectivelyutilizecontextual -| for supervised | | fine-tuning, | | which | is difficult | and | | | | | | | | -| ------------------------------------ | --- | ------------ | --- | ----- | ------------ | ------- | --------- | ----- | ----------- | --- | ---------- | --- | ------- | -| | | | | | | | evidence. | These | limitations | | can result | in | perfor- | -| costlytoannotate,andthereforescarce. | | | | | | Thus,we | | | | | | | | -mancedegradationorevenhallucinations(Huang -| start with | an automatic | | data | generation | | pipeline | | | | | | | | -| ---------- | ------------ | --- | ---- | ---------- | --- | -------- | -------------- | --- | ------------ | --- | --------- | --- | ---- | -| | | | | | | | et al., 2023), | | highlighting | the | necessity | for | more | -tailoredforcollectinghigh-qualityattributiondata -factualandverifiablesystems. -(§3.1). Givenauserquery,thepipelineautomates -dataconstructionthroughdocumentretrieval,rele- -| | | | | | | | AttributedLargeLanguageModels. | | | | | Theper- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | --- | ------- | --- | -vancereranking,attributedanswergeneration,and -| | | | | | | | sistent | challenge | of hallucinations | | | within | LLMs | -| ------------------------------- | --- | ------ | ---- | ------------------- | --- | --- | ----------- | --------- | ----------------- | --- | ------------- | ------ | ---- | -| data filtering | to | ensure | both | the informativeness | | | | | | | | | | -| | | | | | | | has spurred | the | development | | of attributed | | LLMs | -| andattributabilityoftheanswers. | | | | Furthermore,to | | | | | | | | | | -(Bohnetetal.,2022;Lietal.,2023;Worledgeetal., -betterunlockLLMs’abilityforfine-grainedattri- -2023),whichseektoenhanceinformationverifia- -| bution, | we introduce | | FRONT, | a two-stage | | train- | | | | | | | | -| ------------- | ------------ | ---- | ------- | ----------- | --- | -------- | --------- | ---------- | --------- | ------ | --------- | ----------- | -------- | -| | | | | | | | bility by | generating | responses | | with | attribution | to | -| ing framework | | that | teaches | LLMs | to | generate | | | | | | | | -| | | | | | | | evidence | sources. | The | way of | providing | | attribu- | -Fine-gRainedgrOuNdedciTations(§3.2),consist- -| | | | | | | | tionsvariesacrossstudies. | | | Forexample,Gaoetal. | | | | -| ------ | --------- | ------ | ---------- | --- | ---- | --- | ------------------------- | --- | --- | ------------------- | --- | --- | --- | -| ing of | Grounding | Guided | Generation | | (G3) | and | | | | | | | | -(2023b)enablesLLMstogeneratetextwithin-line -Consistency-AwareAlignment(CAA).Duringthe -| | | | | | | | citations | via in-context | | learning. | Another | | line of | -| --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------- | --- | ------- | -G3 -stage,theLLMfirstselectssupportingquotes -| | | | | | | | research | (Gao | et al., 2023a; | Xu | et al., | 2023b) | ex- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ---- | -------------- | --- | ------- | ------ | --- | -fromretrievedsources(grounding)andthencondi- -plorespost-hocattribution,whereLLMsfirstgen- -tionsthegenerationprocessonthem(generation). -erateaninitialresponseandthenretrievethemost -TheCAAstagethenutilizespreferenceoptimiza- -| | | | | | | | relevantevidencetoachieveattribution. | | | | | Inthispa- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | --------- | --- | -tiontofurtheralignthegroundingandgeneration -per,weadvancetheresearchonattributedLLMs -processbyautomaticallyconstructingpreference -| | | | | | | | further. | Unlikeexistingmodelsthatpredominantly | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------- | --- | --- | --- | --- | --- | -signals. Inthisway,thesequotescanserveasfine- -| | | | | | | | cite document | | identifiers, | we | delve | into | a more | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ------------ | --- | ----- | ---- | ------ | -grainedcitationsandimprovetheefficiencyofthe -fine-grainedformofattributionbypinpointingand -verificationprocessforusers(seeFigure1). -citingspecificextractivequotes. -| We conduct | | extensive | experiments | | to | evaluate | | | | | | | | -| ---------- | --- | --------- | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -our framework on the ALCE Benchmark (Gao 3 TaskFormulationandMethodology -| etal.,2023b). | Ourfindingsareasfollows: | | | | | | | | | | | | | -| ------------- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Following(Liuetal.,2023;Gaoetal.,2023b),the -| • FRONT | demonstratessupervisorperformance | | | | | | | | | | | | | -| ------- | --------------------------------- | --- | --- | --- | --- | --- | ------------------ | --- | ----------- | --- | ----- | ------ | ----- | -| | | | | | | | task is formalized | | as follows: | | given | a user | query | -gainsincitationqualitycomparedtoallbase- q and a corpus of retrieved documents D as in- -lines,achievinganaverage14.21%improve- -| | | | | | | | put, the | LLM | is required | to | produce | a response | | -| --- | --- | --- | --- | --- | --- | --- | -------- | --- | ----------- | --- | ------- | ---------- | --- | -mentusingLLaMA-2-7B. -| | | | | | | | S, which | consists | of statements | | with | embedded | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | -------- | -------------------- | --- | ---- | -------- | ---- | -| | | | | | | | in-linecitations. | | WeassumetheresponseS | | | | com- | -• Humanevaluationrevealsthatthequotesgen- -| | | | | | | | prising | with | n statements | S | = {s | ,s ,...,s | } | -| ---------------------------------------- | --- | ------------- | --- | --- | ------- | ------- | ----------------- | -------- | ------------ | ----------------------- | ---- | ------------- | --- | -| erated | by | our framework | | are | of high | quality | | | | | | 1 2 | n | -| | | | | | | | andeachstatements | | | ∈ S,citesalistofpassage | | | | -| andsignificantlybenefituserverification. | | | | | | | | | i | | | | | -| | | | | | | | C i = {c | i1 ,c i2 | ,...}, where | c ij | ∈ D. | Specifically, | | -• AnalysisshowsthatFRONTgenerateslesshal- citationsarepresentedintheformof[1][2]. -lucinationanddemonstratesremarkablegen- Next,wepresentacomprehensiveoverviewof -eralizationacrossdifferentbasemodels. ourmethod,whichconsistsoftwoprimarycompo- - -Automatic Data Generation Pipeline -          Data Collection        Answer Generation           Data Filtering -Question: Why do bagels have holes in the middle？ Bagel holes ensures even baking (...) a area -| | | | | | | Demonstrations | | |  1    3  . Another suggests (...) and transport  2  . | | | | -| --- | --- | --- | --- | --- | -------------------------------------------- | -------------- | --- | --- | ----------------------------------------------------- | --- | --- | --- | -| | | | | | Instruction: Generate an answer and cite the | | | | Informative? | | | | -Retrieve source for the question and provided documents Break into -Question: Why do bagels have holes in the middle？ -| | Top-100 retrieval documents | | | | | | | | | | statements | | -| --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | -(...) -| | | | | | Prior to (...) the hole was originally put in place | | | | Statement 1 | Bagel holes ensures even baking by | | | -| --- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | ----------- | ---------------------------------- | --- | --- | -allowing heat to circulate around a -| | | | | | to allow for easier handing prior. | | | | | greater surface area  1    3  .  | | | -| --- | -------- | -------------------------------- | ---------------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | -------------------------------- | --- | --- | -| | T h e b | a k e r s (. .. ) th e h o | l e i n t h e c e n tre of the | | | | | | | | | | -1 ba g e l e n s u r e s t h a t i t b a k e s e v e n ly . Statement 2 A n o t h e r s u g g e s t s t h a t t h e y w e re u s ed -| | | | | | | | | | | fo r e a s i e r | h a n d li n g a n | d tr a n sp o rt   2  . | -| ------ | --------------------------------------------- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | ----------- | ------------------ | -------------------- | --------------------------- | -| | Prior to (...) the hole was originally put in | | | | Bagel holes ensures even baking by allowing | | | | | | | | -| | 2 | | | | heat to circulate around a greater surface | | | | | | | Support | -| | place to allow for easier handing prior. | | | | | | | | Attributed? | | | | -| Rerank | | | | | area   1   3   . Another suggests that they were | | | | | | | | -T h e r i n g s h a p e a l lo w s h e a t t o c ir c u la t e used for easier handling and transport  2  . Documents -| | 3 ar o u n | d ( .. . )  m a k e s b a g | e l s t o c o o k fa s t er  | | | | | | | | | | -| --- | ---------- | ------------------------------ | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Figure2: Overviewofthedatagenerationpipeline. Thepipelineconsistsofthreeprimarysteps: datacollection, -answergeneration,anddatafiltering. Firstly,givenauserquery,thedatacollectionmoduleretrievesthetop100 -relevantdocumentsandemploysarerankingmodeltoselectthetop5mostpertinentdocuments. Subsequently, -attributedresponsesaregeneratedbydistillingChatGPTviain-contextlearning. Finally,allresponsesarefiltered -bythedatafilteringmoduletoensureinformativenessandattributability. -nents: anautomaticdatagenerationpipeline(§3.1) subsequently re-rankedby RankVicuna (Pradeep -andatwo-stagetrainingframework(§3.2). etal.,2023)consideringitssuperiorperformance -| | | | | | | inlistwisere-ranking, | | | resultinginthetop5most | | | | -| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | --- | -3.1 AutomaticDataGenerationPipeline -relevantdocumentsforeachquery. -EquippingLLMswiththeattributioncapabilityne- -| | | | | | | Attributed | Answer | | Generation. | | Given | the re- | -| ---------- | -------- | ---- | ------------- | ------------ | -------- | ---------- | ----------- | --- | ----------- | ------- | ----- | ---------- | -| cessitates | training | data | that includes | high-quality | | | | | | | | | -| | | | | | | markable | performance | | of | ChatGPT | in | attributed | -| responses | paired | with | precise | citations, | which is | | | | | | | | -questionanswering,weemployChatGPTtogener- -| typically | labor-intensive | | and | costly. To | address | | | | | | | | -| --------- | --------------- | --- | --- | ---------- | ------- | --- | --- | --- | --- | --- | --- | --- | -ateanswerswithcorrespondingcitationsforgiven -thischallenge,weproposeapipelinedesignedfor -| | | | | | | queriesandthetop5retrieveddocuments. | | | | | | Wepro- | -| --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | --- | ------ | -theautomaticgenerationofhigh-qualityattributed -videpreciseinstructionsandin-contextdemonstra- -data4. -Thispipelinecomprisesthreecorecompo- -tionstoensurethatChatGPTproducesinformative -nents: datacollection,attributedanswergeneration, -responsesandcitesthesourcesaccordingly. -anddatafiltering,asoutlinedinFigure2. -| | | | | | | DataFiltering. | | Toguaranteethehighqualityof | | | | | -| --------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | --------------------------- | ----- | --------- | --- | ----------- | -| DataCollection. | | Tosimulatethereal-worldenvi- | | | | | | | | | | | -| | | | | | | our synthetic | | training | data, | we employ | | a data fil- | -ronmentforinformation-seeking,wecollectques- -teringprocessguidedbytwokeycriteriaderived -tionsfromtheAQuAMuSedataset(Kulkarnietal., -| | | | | | | fromKamallooetal.(2023): | | | | (1)informativeness: | | | -| --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ------------------- | --- | --- | -2020),whichisderivedfromtheNaturalQuestion -| | | | | | | assessing | if the | answer | provides | | sufficient | infor- | -| ----------------------------------- | --- | --- | --- | --- | ----- | ------------------------------------------ | ------ | ------ | -------- | --- | ---------- | ------ | -| (NQ)dataset(Kwiatkowskietal.,2019). | | | | | TheNQ | | | | | | | | -| | | | | | | mationtothequestion,and(2)attributability: | | | | | | de- | -datasetcomprisesrealuserqueriesfromtheGoogle -| | | | | | | termining | if the | answer | is | attributed | | to the cited | -| --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | --- | ---------- | --- | ------------ | -searchengine,providingarobustbasisforrealistic -| | | | | | | documents. | To | mitigate | | the impact | of | nonsensi- | -| ------------------ | --- | ---------- | --- | ----------- | ----- | ---------- | --- | -------- | --- | ---------- | --- | --------- | -| question-answering | | scenarios. | | The dataset | spans | | | | | | | | -calqueriesandirrelevantdocumentretrievalthat -| a range | of diverse | question | types, | demanding | an- | | | | | | | | -| ------- | ---------- | -------- | ------ | --------- | --- | -------- | ------------------ | --- | --- | -------- | --- | ---------- | -| | | | | | | may lead | to non-informative | | | answers, | | we utilize | -swersofvaryinglengths,fromconcisetodetailed. -ChatGPTforpreliminaryinformativenessannota- -Tomimicthewayasearchenginemightsynthesize -| | | | | | | tions. Responses | | categorized | | as | non-informative | | -| --- | --- | --- | --- | --- | --- | ---------------- | --- | ----------- | --- | --- | --------------- | --- | -documentsofhighrelevanceinresponsetoauser -| | | | | | | aredirectlyexcluded. | | | Furthermore,toensurethat | | | | -| ------ | --------- | ------ | ------- | ------- | -------- | -------------------- | --- | --- | ------------------------ | --- | --- | --- | -| query, | we employ | Sphere | (Piktus | et al., | 2021), a | | | | | | | | -answersareaccompaniedbyhighlysupportiveci- -pre-processedandcleanedversionoftheCommon -tations,wetrainadiscriminatoronhuman-labeled -Crawlcorpus,servingasaproxywebsearchindex. -| | | | | | | data from | the | comprehensive | | evaluation | | by Liu | -| --- | --- | --- | --- | --- | --- | --------- | --- | ------------- | --- | ---------- | --- | ------ | -Inparticular,foragivenuserquerysampledfrom -| | | | | | | et al. (2023), | | where | attributability | | is | categorized | -| --- | ------------ | ---------- | ------------ | ---------- | ---- | ---------------- | ------------------------------------- | ------------------------------- | --------------- | ----- | --- | ----------- | -| the | AQuAMuSe | dataset, | we initially | retrieve | the | | | | | | | | -| | | | | | | intothreelevels: | | fullsupport,partialsupport,orno | | | | | -| top | 100 relevant | documents | from | the Sphere | cor- | | | | | | | | -| | | | | | | support. | Wequantitativelymapthediscriminator’s | | | | | | -| pus | using sparse | retrieval. | These | documents | are | | | | | | | | -| | | | | | | outputs | to an | attributability | | score | and | ultimately | -4Attributeddatarefersto“answerswithin-linecitations”. deriveanaveragescoreforeachattributedanswer. - -Two-Stage Training Recipe -          Grounding Guided Generation        Consistency-Aware Alignment -          Answer Generation -| Q :   W h y d o | b a g e l s h a v e | | | | | | | | | | | | -| ----------------- | ----------------------- | ---------------- | --------------------------- | ----------------------- | ------------------------------------- | --- | ----------------------- | ------------------- | ----------------------------------------------- | --------------------------- | --- | --- | -| | | G r o u n d i ng | T h e b a k e r s ( . . | . ) t h e h o l e | i n t h e c e n t r e o f the | Q | u e s t i o n :   W h y | d o b a g e l s |   G  R      D eI   G    1l   e | n s u r e s t h a t i t | | | -h o l e s i n t h e m i d d l e ？ 1 b a g e l { e n s u r e s t h a t i t b a k e s e v e n l y } . h a v e h o l e s i n t h e m i d d l e ？ b a kO eU Ns N v e n y .   3   a l l o w s h e a t -| | | | | | | | ( . . . ) | | t o c i r c u l a t e a r | o u n d .  2    a l l o w   s A u   N r S  f W  a   c E | R e     a B r a e g a e   1 l   h  3 o  A l e n s o ( t . h . . e )   r | g ( r . e .. a ) t | -| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------- | -B a g e l h o l e s e n s u r e s e v e n b a k i n g b y P r i o r t o ( . . . ) t h e h o l e w a s o r i g i n a l l y p u t in P r i o r t o ( .. . )   p l a c e t o a ll o w f o r e a s i e r h a n d i n g p r i o r . h a n d l i n g a n d t r a n s p o r t   2  . -| al lo w i n g h e a | t to c i r c u l a t e | a r o u n d a 2 | | | | | | | | | | | -| -------------------- | -------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -greater surface area  1   3  .Another p l a c e t o { a l l o w f o r e a s ie r h a n d i n g p r i o r } . fo r e a s i e r h a n d i n g p r i o r . Consistency -| suggests that they were used for | | 3 | The ring shape {allows heat to circulate | | | | | | | | | | -| -------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -easier handling and transport  2  . around} (...) makes bagels to cook faster  Gold Grounding Consistent Answer Contrastive -Large Language -| | Supervised Fine-tuning | | | | | | Model | | | | | | -| -------- | ---------------------- | -------------- | ------------- | --------------------------------------------- | ------------------------------------------------- | --- | ----- | --- | -------------- | ------------------- | --- | --- | -| | | | | | | | | | Gold Grounding | Inconsistent Answer | | | -| Question | | |   G   R   | Ol  Uo N  D  sI N   Gh     1   e n s | u r e s t h a t it b a k e s e v e n ly.  | | | | | | | | -| | | Large Language |   f3 a | l w e a t t o | c i r c u l a t e a r o u n d .   2  a l lo w | | | | | | | | -| | | | o r | e a s i e r h a n d i n | g p r i o r . | | | | | | | | -Documents Model Direct Preference Optimization (DPO)   A   N S  W    E R    B a g e l h o l e s   (. . . )  f o r -| | | |   A A  N n  S o | W  t h  E e  R r   B (. a .. g ) e h l a h n o d le | lin s g ( . a .. n )  d g r t e ra a n t s s p u o rf r a t  c 2  e . area13 | | | | | a e s t h e | t i c s a n d u n i q u e t e x | t u r e   1 | -| --- | --- | --- | ---------------- | ------------------------------------------------------------ | ------------------------------------------------------------------------------------- | --- | --- | --- | --- | ----------------- | ------------------------------------- | ------------- | -| | | | | | | | | | |   ( . . . ) h a | n d l i n g a n d tr a n s p o | r t   3   . | -Figure 3: Overview of FRONT: The training recipe consists of two stages: grounding-guided generation and -consistency-aware alignment. It enables LLMs to first generate precise grounding and subsequently guide the -generationofattributedanswers,therebyenhancingfine-grainedattributioncapability. -Answersfallingbelowadefinedthresholdaresys- fine-tunedtogeneratearesponseS whichconsists -tematicallyexcludedtoensurethesyntheticdata’s of two components: the grounded quotes G and -reliability, which results in nearly 8,000 entries. theattributedanswerA. Specifically,thegrounded -Formoredetails,pleaserefertoAppendixA. quotesG aredelineatedasfollows: -3.2 Two-StageTrainingRecipe G = {[GROUNDING],(i ,e ),...,(i ,e )}, (1) -| | | | | | | | | | | 1 1 | n n | | -| ------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Inthissection,weintroduce | | | FRONT,atwo-stage | | | | | | | | | | -where[GROUNDING]denotesaspecialtokenindi- -trainingframeworkthataimsatempoweringLLMs -| | | | | | | cating | the | start of | the grounding | process. | | Each | -| ----------------- | --- | ----------- | ----------- | --- | -------- | ------ | --- | -------- | ------------- | -------- | --- | ---- | -| with fine-grained | | attribution | capability. | | Figure 3 | | | | | | | | -tuplewithinG,comprisingadocumentidentifieri -illustratestheoverviewofourframework. -andthecorrespondingextractivesegmente,collec- -tivelyformingagroundedquote. -3.2.1 GroundingGuidedGeneration -| | | | | | | Similarly, | | the formulation | | of the | attributed | an- | -| ---------- | --- | --------- | ------------ | --- | -------- | ---------- | --- | --------------- | --- | ------ | ---------- | --- | -| To empower | | LLMs with | fine-grained | | attribu- | | | | | | | | -swerAisconciselypresentedas: -| tion capability, | | we propose | Grounding | | Guided | | | | | | | | -| ---------------- | --- | ---------- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -(G3), -Generation which teaches LLMs to gener- A = {[ANSWER],s ,s ...,s }, (2) -| | | | | | | | | | | 1 2 | m | | -| ------------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| atefine-grainedcitations. | | | ThecornerstoneofG3 | | | | | | | | | | -liesinenablingLLMstoextractsupportingquotes where [ANSWER] is a special token that signals -fromthesourcedocuments,eachassociatedwithits the beginning of the answer generation process. -documentidentifier,whichinturnguidesthegen- Each statement s cites a list of passages C = -| | | | | | | | | i | | | | i | -| --------------------------- | --- | --------------------------- | ----------------- | --- | --- | ----------------- | ------------ | ------ | --- | ------------------ | ---- | ----- | -| erationofattributedanswers. | | | Suchaformatoffers | | | | | | | | | | -| | | | | | | {c i1 | ,c i2 ,...}, | wherec | ij | ⊆ {i 1 ,i 2 ,...,i | n }, | asde- | -| twoprimarybenefits. | | Firstly,thedirectextraction | | | | finedinEquation2. | | | | | | | -of quotes from sources significantly reduces the Thus,thetraininglossisformulatedas: -impactoftheincorporationofirrelevantinforma- -| tion and | the risk | of hallucinations | | in | subsequent | | | N | | | | | -| -------- | -------- | ----------------- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -(cid:88) -| | | | | | | | | L = − | logP(y | |q ,D | ;θ) | (3) | -| ------------------ | --- | ---------------------------- | --- | --- | --- | --- | --- | ----- | ------ | ----- | --- | --- | -| attributedanswers. | | Secondly,theprocessnaturally | | | | | | | | i i | i | | -i=1 -facilitatesaccurateattribution,witheachdocument -identifierservingasaclearsupervisedsignalthat -| | | | | | | where | y | represents | the | combined | output | of | -| --- | --- | --- | --- | --- | --- | ----- | --- | ---------- | --- | -------- | ------ | --- | -i -delineatestheoriginoftheextractivequotes,thus -groundedquotesGandtheanswerAforeachgiven -improvingthecitationquality. queryq andsetofretrieveddocumentsD . -| | | | | | | | i | | | | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -However,theabsenceofspecificgroundingcon- -tent for statements within our generated dataset 3.2.2 Consistency-AwareAlignment -posesadditionalchallenges. Totacklethis,weem- While G3 unlocks the ability to first extract sup- -ploy ChatGPT to meticulously extract segments portingquotesbeforegeneratingattributedanswers, -fromciteddocumentsthatsupportthecorrespond- it occasionally leads to inconsistencies between -ingstatement. Hence,whengivenaqueryqandthe groundedquotesandattributedanswers. Suchdis- -top-5retrieveddocumentsD asinput,theLLMis crepancieschallengetheattempttoemploythese - -groundedquotesasfine-grainedverification. Inre- whereπ ref representsthereferencemodel,initial- -sponse,weproposeaconsistency-awarealignment izedfromG3. Thehyper-parameterβ modulates -(CCA) stage specifically aimed at enhancing the the divergence between the distribution from the -consistencybetweenthegroundingprocessandthe policy model and the reference model. τ is the -w -generationprocess. consistentanswer,whileτ istheinconsistentone. +Corporate Governance Report + +I. WIPRO’S PHILOSOPHY ON CORPORATE + +GOVERNANCE +Wipro’s governance framework is driven by the objective +of enhancing long term stakeholder value without +compromising on ethical standards and corporate +social responsibilities. Efficient corporate governance +requires a clear understanding of the respective +roles of the Board of Directors (“Board”) and of senior +management and their relationships with others in the + +corporate structure. Sincerity, fairness, good citizenship +and commitment to compliance are key characteristics +that drive relationships of the Board and senior +management with other stakeholders. + +Corporate governance philosophy of Wipro flows from +the “Spirit of Wipro” which represents core values by +which policies and practices of the organization are +guided. The values encapsulated in the “Spirit of Wipro” +are: + +Be passionate about +clients’ success + +Treat each person +with respect + +Be global and +responsible + +Unyielding integrity in +everything we do + +While our Company has transformed many times over +the years, the Spirit of Wipro and our core values, have +remained constant. + +Our Chairman introduced the Five Habits essential to +drive a Growth Mindset in early 2020, which are our +values in action: + +• Being respectful + +• Being responsive + +• Always communicating + +• Demonstrating stewardship + +• Building trust + +So far, over 21,000 employees globally have been part +of 75 immersive and interactive sessions hosted by our +senior leadership team on the Five Habits. + +Corporate governance at Wipro is implemented through +robust board governance processes, internal control +systems and processes, and strong audit mechanisms. +These are articulated through the Company’s Code of +Business Conduct, Corporate Governance Guidelines +and charters of various sub-committees of the Board +and the Company’s Disclosure Policy. Wipro’s corporate +governance practices can be described through the +following four layers: + +a) Governance by Shareholders + +b) Governance by Board of Directors + +c) Governance by Sub-committees of Board, and + +d) Governance through management process + +In this report, we have provided details on how the +corporate governance principles are put in to practice +within Wipro. + +II. SHAREHOLDERS + +The Companies Act, 2013, Securities and Exchange +Board of India (Listing Obligations and Disclosure +Requirements) Regulations, 2015 (“Listing Regulations”) +and New York Stock Exchange (NYSE) Listed Company +Manual prescribe the governance mechanism by +shareholders in terms of passing of ordinary and special +resolutions, voting rights, participation in the corporate +actions such as bonus issue, buyback of shares, +declaration of dividend, etc. Your Company follows a +robust process to ensure that the shareholders of the +Company are well informed of Board decisions both on +financial and non-financial matters and adequate notice +with a detailed explanation is sent to the shareholders +well in advance to obtain necessary approvals. + +III. BOARD OF DIRECTORS +1. Composition of Board + +is a professional CEO who + +As at March 31, 2021, our Board had two Executive +Independent Directors +Directors, six non-executive +and one non-executive non-independent Director. The +Executive Chairman and Whole Time Director, and the +non-executive non-independent Director are Promoter +(CEO) and +Directors. The Chief Executive Officer +Managing Director +is +responsible for the day to day operations of the +Company. Of the seven Non-Executive Directors, six +are Independent Directors, free from any business or +other relationship that could materially influence their +judgment. In the opinion of the Board, all the Independent +Directors are independent of the management and +satisfy the criteria of independence as defined under +the Companies Act, 2013, the Listing Regulations and +the NYSE Listed Company manual. + +121 + + Corporate Governance Report + +is well diversified and + +consists +The Board +of a woman +three +Directors who are foreign nationals. The profiles +of our Directors are available on our website at +https://www.wipro.com/leadership. + +Independent Director and + +Post-Meeting Follow-Up System +After the Board meeting, we have formal system of +follow-up, review and reporting on actions taken by the +management on the decisions of the Board and sub- +committees of the Board. + +2. Board Meetings + +4. Appointment of Directors + +We decide about the Board meeting dates +in +consultation with the Board members. The schedule +of the Board meetings and Board Committee meetings +is communicated in advance to the Directors to enable +them to attend the meetings. + +Our Board meetings are normally scheduled over two +days. In addition, every quarter, Independent Directors +meet amongst themselves exclusively. In line with Para 4 +of Schedule B of SEBI (Prohibition of Insider Trading) +Regulations, 2015, it is the endeavor of the Company +that the gap between the clearance of accounts by audit +committee and board meeting is as narrow as possible, +and Wipro is committed to adhere to this requirement. + +Information flow to the Board Members +Information is provided to the Board Members on a +continuous basis for their review, inputs and approval +from time to time. More specifically, we present our +annual Strategic Plan and Operating Plans of our +business to the Board for their review, inputs and +approval. Likewise, our quarterly financial statements +and annual financial statements are first presented to +the Audit Committee and subsequently to the Board +for their approval. In addition, various matters such as +appointment of Directors and Key Managerial Personnel, +corporate actions, review of internal and statutory +audits, details of investor grievances, specific cases of +acquisitions, important managerial decisions, material +positive/negative developments and +legal/statutory +matters are presented to the respective Committees +of the Board and later with the recommendation of +Committees to the Board of Directors for their approval +as may be required. + +As a system, in most cases, information to Directors is +submitted along with the agenda papers well in advance +of the Board meeting. Inputs and feedback of Board +Members are taken and considered while preparation +of agenda and documents for the Board meeting. + +Documents containing Unpublished Price Sensitive +Information are submitted to the Board and Committee +Members, at a shorter notice, as per the general consent +taken from the Board, from time to time. + +3. + +122 + +The Board has adopted the provisions with respect +to appointment and tenure of Independent Directors +consistent with the Companies Act, 2013 and the Listing +Regulations. + +As per the provisions of the Companies Act, 2013, the +Independent Directors shall be appointed for not more +than two terms of maximum of five years each and shall +not be liable to retire by rotation. + +issues a formal + +Independent +At the time of appointment of an +Director, the Company +letter of +appointment outlining his/her role, function, duties +and responsibilities as a Director. The template of the +letter of appointment is available on our website at +h t t p s : // w w w.w i p ro .c o m / i n v e s t o r s /c o r p o r a t e - +governance/. + +Details of Directors proposed for appointment/re- +appointment at the 75th Annual General Meeting +(“AGM”) is provided at page no. 91 as part of the Board’s +Report and in the notice convening the 75th AGM. + +Lead Independent Director +The Board has designated Mr. M. K. Sharma as +Independent Director. The role of the +the Lead +Lead +in the +Corporate Governance guidelines of your Company +and +is available on the Company’s website at +h t t p s : // w w w.w i p ro .c o m / i n v e s t o r s /c o r p o r a t e - +governance/. + +Independent Director + +is described + +5. Policy + +for Nomination of Directors, + +their + +Remuneration and Board Diversity +The Board Governance, Nomination and Compensation +Committee has adopted a policy which, inter alia, deals +with the manner of selection of Directors and payment +of their remuneration as described herein below. + +The Company has also adopted a policy on Board +Diversity which guides the organization’s approach to +diversity in the composition of the Board. + +Criteria for Selection of Independent Directors and Key +Skills, Expertise, and Core Competencies of the Board +The Board of the Company comprises of eminent +personalities and leaders in their respective fields. +These Directors are nominated based on well-defined + + selection criteria. The Board Governance, Nomination +and Compensation Committee considers, inter alia, + qualifications, skills, expertise and +experience, +competencies, whilst recommending to the Board +the candidature for appointment as +Independent +Director. + +In case of appointment of Independent Directors, the +Board Governance, Nomination and Compensation +Committee satisfies itself about the independence +of the Directors vis-à-vis the Company to enable the +Board to function independently of the management +and discharge its functions and duties effectively. In +case of re-appointment of Independent Directors, the +Board also takes into consideration, the performance + +evaluation and engagement level of the Independent +Directors. + +The Board Governance, Nomination and Compensation +Committee ensures that the candidates +identified +for appointment as Directors are not disqualified for +appointment under Section 164 and other applicable +provisions of the Companies Act, 2013. + +As required under Rule 6 of the Companies (Appointment +and Qualification of Directors) Rules, 2014, all the +Independent Directors have completed the registration +with the Independent Directors Databank and, wherever +required, also completed the online proficiency test, +conducted by Indian Institute of Corporate Affairs. + +In the opinion of the Board and the Board Governance, Nomination and Compensation Committee, the following is a list of +core skills/expertise/competencies required in the context of the Company’s business and which are available with the Board: + +Wide management and +leadership experience + +Information Technology + +Diversity + +Functional and managerial +experience + +Personal values + +Corporate governance + +Strong management and leadership experience, including in areas of business development, strategic +planning and mergers and acquisitions, ideally with major public companies with successful multinational +operations in technology, manufacturing, banking, investments and finance, international business, +scientific research and development, senior level government experience and academic administration. +Expertise or experience in information technology business, technology consulting and operations, +emerging areas of technology such as digital, cloud and cyber security, intellectual property in information +technology domain, and knowledge of technology trends. +Diversity of thought, experience, knowledge, perspective, gender and culture brought to the Board +by individual members. Varied mix of strategic perspectives, geographical focus with knowledge and +understanding of key geographies. +Knowledge and skills in accounting and finance, business judgment, general management practices and +processes, crisis response and management, industry knowledge, macro-economic perspectives, human +resources, labour laws, international markets, sales and marketing, and risk management. +Personal characteristics matching the Company’s values, such as integrity, accountability, and high +performance standards. +Experience in developing and implementing good corporate governance practices, maintaining board and +management accountability, managing stakeholders’ interests and Company’s responsibilities towards +customers, employees, suppliers, regulatory bodies and the communities in which it operates. Experience +in boards and committees of other large companies. + +Given below is a list of core skills, expertise and competencies of the individual Directors: + +Name of Director + +Mr. Rishad A. Premji +Mr. Azim H. Premji +Mr. Thierry Delaporte2 +Mr. William Arthur Owens +Mr. M. K. Sharma +Mrs. Ireena Vittal +Dr. Patrick J. Ennis +Mr. Patrick Dupuis +Mr. Deepak M. Satwalekar3 + +Wide Management +and Leadership +experience1 +  +  +  +  +  +  +  +  +  + +Information +Technology + +  +  +  +  + - +  +  +  +  + +Diversity + +Skills/Expertise/Competencies +Functional and +Managerial +Experience1 +  +  +  +  +  +  +  +  +  + +  +  +  +  +  +  +  +  +  + +Personal +Values + +Corporate +Governance + +  +  +  +  +  +  +  +  +  + +  +  +  +  +  +  +  +  +  + +1 These skills/competencies are broad-based, encompassing several areas of expertise/experience. Each Director may possess varied combinations of skills/ +experience within the described set of parameters, and it is not necessary that all Directors possess all skills/experience listed therein. +2 Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director for a term of 5 years with effect from July 6, 2020 to July 5, 2025. +3 Mr. Deepak M. Satwalekar was appointed as an Independent Director for a term of 5 years with effect from July 1, 2020 to June 30, 2025. + +123 + + Corporate Governance Report + +6. Familiarization Programme and Training + +for + +Independent Directors +The Company has an orientation process/familiarization +programme for its independent directors that includes: + +a) Briefing on their role, responsibilities, duties, and + +obligations as a member of the Board. + +b) Nature of business and business model of the +Company, Company’s strategic and operating plans. + +c) Matters relating to Corporate Governance, Code of +Business Conduct, Risk Management, Compliance +Programs, Internal Audit, etc. + +As a process when a new independent director is +appointed, a familiarization programme as described +above is conducted by the senior management team +and whenever a new member is appointed to a Board +Committee, information relevant to the functioning of the +Committee and the role and responsibility of Committee +members is informed. Each of our independent directors +have attended such orientation process/familiarization +programme when they were inducted into the Board. + +As part of ongoing training, the Company schedules +quarterly meetings of business heads and functional +heads with the Independent Directors. During these +meetings, comprehensive presentations are made +on the various aspects such as business models, +new business strategies and initiatives by Business +Leaders, risk minimization procedures, recent trends +in technology, changes in domestic/overseas industry +scenario, Digital Transformation, state of Global IT +Services +industry, and regulatory regime affecting +the Company globally. These meetings also facilitate +Independent Directors to provide their inputs and +suggestions on various strategic and operational +matters directly to the business and functional +heads. The details of the familiarization programme +are also available on the website of the Company at +https://www.wipro.com/content/dam/nexus/en/investor/ +corporate-governance/policies-and-guidelines/ethical- +guidelines/12771-familiarization-programmes-imparted- +to-independent-directors.pdf. + +The Board Governance, Nomination and Compensation +Committee presents to the Board on a periodic basis, +succession plans for appointments to the Board +based on various factors such as current tenure of +Directors, outcome of performance evaluation, Board +diversity and business requirements. In addition, the +Company conducts an annual Talent Review Process +for senior management and other executive officers +which provides a leadership-level talent inventory and +capability map that reflects the extent to which critical +talent needs are fulfilled vis-a-vis business drivers. + +The Board Governance, Nomination and Compensation +Committee reviews the outcome of this process and +presents the succession plan for senior management +and other executive officers to the Board. + +8. Board Evaluation + +Details of methodology adopted for Board evaluation +have been provided at page no. 92 of the Board’s Report. + +9. Remuneration Policy and Criteria for Making +Payments to Directors, Senior Management and +Key Managerial Personnel +The Independent Directors are entitled to receive +remuneration by way of sitting fees, reimbursement +of expenses for participation in the Board/Committee +meetings and commission as detailed hereunder: + +a) Sitting fees for each meeting of the Board or +Committee of the Board attended by him or her, of +such sum as may be approved by the Board within +the overall limits prescribed under the Companies +Act, 2013. + +and + +b) Commission on a quarterly basis, of such sum as +may be approved by the Board and Members on +the recommendation of the Board Governance, +Nomination +Committee. +The aggregate commission payable to all the +Independent Directors and non-executive directors +put together shall not exceed 1% of the net profits +of the Company during any financial year. The +commission is payable on pro-rata basis to those +Directors who occupy office for part of the year. + +Compensation + +7. Succession Planning + +We have an effective mechanism for succession +planning which focuses on orderly succession of +Directors, +including Executive Directors and other +senior management team and other executive officers. +The Board Governance, Nomination and Compensation +Committee implements this mechanism in concurrence +with the Board. + +c) Reimbursement of travel, stay and other expenses +for participation in Board/Committee meetings. + +d) Independent Directors and Promoter Directors +are not entitled to participate in the stock option +schemes of the Company. + +Following are terms and conditions for determining +the remuneration to Mr. Azim H. Premji, who is a Non- +Executive, Non-Independent Director: + +124 + + a) Remuneration as applicable to other Non-Executive +Directors of the Company, in addition to the sitting +fees for attending the meetings of the Board thereof, +as may be determined by the Board, provided +however that the aggregate remuneration, including +commission, paid to the Directors other than the +Managing Director and Whole Time Directors in a +financial year shall not exceed 1% of the net profits +of the Company, in terms of Section 197 of the +Companies Act, 2013 and computed in the manner +referred to in Section 198 of the Companies Act, 2013. + +b) Maintenance of Founder Chairman’s office including + +executive assistant at Company’s expense. + +c) Reimbursement of travel, stay and entertainment +expenses actually and properly incurred in the +course of business as per the Company’s policy. + +In determining the remuneration of Chairman, CEO and +Managing Director, and Senior Management Employees +and Key Managerial Personnel, the Board Governance, +Nomination and Compensation Committee and the +Board shall ensure/consider the following: + +a) The balance between fixed and variable pay +reflecting short and +long-term performance +objectives, appropriate to the working of the +Company and its goals. + +b) Alignment of remuneration of Key Managerial +Personnel and Directors with long-term interests of +the Company. + +c) Company’s performance vis-à-vis + +the annual +achievement, +vis-à- +vis KRAs/KPIs, industry benchmark and current +compensation trends in the market. + +individuals’ performance + +the + +recommends + +remuneration + +The Board Governance, Nomination and Compensation +Committee +for +the Chairman, CEO and Managing Director, Senior +Management and Key Managerial Personnel. The +payment of remuneration to the Executive Directors and +Non-Executive Directors is approved by the Board and +Members. Approval of Members is also obtained in case +of remuneration payable to Non-Executive Directors. +There was no change to the remuneration policy during +the financial year. + +Details of Remuneration to Directors +Details of remuneration paid to the Directors for the services rendered and stock options granted during the financial year +2020-21 are given below. No stock options were granted to any of the Independent Directors and Promoter Directors during +the financial year 2020-21. + +(in ` Million) + +Relationship with directors + +Salary + +Allowances + +Commission/Incentives/ +Variable Pay + +Other annual compensation + +Retirals + +Sitting fees + +TOTAL +Grant of ADS Restricted +Stock Units during the year + +Notice period + +Rishad A. +Premji(2)(3) +Son of Azim +H. Premji + +Thierry +Delaporte(1)(4)(5) +None + +William +Arthur +Owens(1) +None + +Azim H. +Premji +Father of +Rishad A. +Premji + +M. K. +Sharma +None + +Ireena +Vittal +None + +Patrick J. +Ennis(1) +None + +Patrick +Dupuis(1) +None + +Deepak M. +Satwalekar(6) +None + +Arundhati +Bhattacharya(7) +None + +Abidali Z. +Neemuchwala(1)(8) +None + +14.30 + +43.93 + +55.68 + +0.17 + +3.86 + +NA + +117.94 +NA + +Up to 180 +days + +68.63 + +27.40 + +112.79 + +379.19 + +55.49 + +NA + +643.50 +1,500,000 + +NA + +NA + +7.15 + +NA + +NA + +0.60 + +7.75 +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +29.21 + +11.60 + +10.24 + +20.73 + +20.73 + +NA + +NA + +0.60 + +29.81 +NA + +NA + +NA + +0.60 + +12.20 +NA + +NA + +NA + +0.60 + +10.84 +NA + +NA + +NA + +0.60 + +21.33 +NA + +NA + +NA + +0.60 + +21.33 +NA + +Up to 180 days + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +6.65 + +NA + +NA + +0.40 + +7.05 +NA + +NA + +NA + +NA + +2.22 + +NA + +NA + +0.20 + +2.42 +NA + +NA + +14.14 + +NA + +2.42 + +6.15 + +0.002 + +NA + +22.72 +NA + +NA + +Figures in the above table are subject to rounding-off adjustments + +Notes: + +(1) Figures mentioned in ` are equivalent to amounts paid in foreign currency, wherever applicable. + +(2) Mr. Rishad A. Premji is entitled to a commission at the rate of 0.35% on incremental consolidated net profits of Wipro Limited for fiscal year 2021 + +over the previous year, computed in accordance with the provisions of the Companies Act, 2013. + +(3) Mr. Rishad A. Premji’s compensation also included cash bonus (part of his allowances) on an accrual basis, which is payable over a period of time. + +(4) The compensation disclosed for Mr. Thierry Delaporte is for the period from July 6, 2020 to March 31, 2021. It includes components such as a one- +time cash award, annual stock grant which will vest as per the vesting pattern approved by the Board Governance, Nomination and Compensation +Committee (1,200,000 ADS restricted stock units, comprising of 300,000 ADS restricted stock units expiring in July 2024 and 900,000 ADS + +125 + + Corporate Governance Report + +performance stock units expiring in June 2023 and one-time grant of 300,000 ADS restricted stock units expiring in July 2022), among other things, +as per the terms approved by the Shareholders at the Annual General Meeting held in July 2020. + +(5) The remuneration of Chief Executive Officer and Managing Director is computed on an accrual basis. It also includes the amortization of RSUs +granted to him, which will vest over a period of time. This also includes RSUs that will vest based on performance parameters of the Company. + +(6) Mr. Deepak M. Satwalekar was appointed as an Independent Director of the Company with effect from July 1, 2020 and the compensation disclosed + +is for the period from July 1, 2020 to March 31, 2021. + +(7) The compensation disclosed for Ms. Arundhati Bhattacharya is for the period April 1, 2020 to June 30, 2020. Ms. Bhattacharya resigned as an + +Independent Director of the Company with effect from the close of business hours on June 30, 2020. + +(8) The compensation disclosed for Mr. Abidali Z. Neemuchwala is for the period April 1, 2020 to June 1, 2020. Mr. Neemuchwala resigned as the Chief + +Executive Officer and Managing Director of the Company with effect from the end of the day on June 1, 2020. + +None of the Non-Executive Directors received remuneration +exceeding 50% of the total annual remuneration paid to all +Non-Executive Directors for the year ended March 31, 2021. + +that consist of a pension and gratuity which are offered to all +of our employees, but no other benefits upon termination of +employment except as mentioned below. + +Terms of Employment Arrangements +Under the Companies Act, 2013, our shareholders must +approve the salary, bonus and benefits of all executive +directors at a General Meeting of the Shareholders. Each of +our executive directors has signed an agreement containing +the terms and conditions of employment, including a monthly +salary, performance bonus and benefits including vacation, +medical reimbursement and pension fund contributions. +These agreements have varying terms, but either we or the +executive director may generally terminate the agreement +upon six months’ notice to the other party. + +Pursuant to the terms of the employment arrangement +with Mr. Delaporte, if his employment is terminated by the +Company without Cause, the Company is required to pay +Mr. Delaporte, severance pay of 12 months’ base salary as +last applicable when in service, payable over a 12 month +period following the date of termination. These payments will +cease if Mr. Delaporte obtains a new employment within the +12 months period or becomes a consultant to any Company. + +We also indemnify our directors and officers for claims +brought under any rule of law to the fullest extent permitted +by applicable law. + +The terms of our employment arrangements with Mr. Rishad +A. Premji and Mr. Thierry Delaporte provide for up to a +180-day notice period, and country specific leaves per year in +addition to statutory holidays, and an annual compensation +review. Additionally, these officers are required to relocate +as we may determine, and to comply with confidentiality +provisions. Service contracts with our executive directors +and officers provide for our standard retirement benefits + +Among other things, we agree to indemnify our directors +and officers for certain expenses, judgments, fines and +settlement amounts incurred by any such person in any +action or proceeding, including any action by or in the right +of the Company, arising out of such person’s services as our +director or officer, including claims which are covered by +the director’s and officer’s liability insurance policy taken +by the Company. + +Key Information pertaining to Directors as on March 31, 2021 is given below: + +Designation + +Date of initial +appointment + +Sl. +No. + +Name of +the Director +and Director +Identification +Number (DIN) + +Date of +appointment +as +Independent +Director under +Companies +Act, 2013 and +SEBI Listing +Regulations +(first term) 1 +- + +Directorship +in other +Companies2 + +Chairmanship +in Committees +of Board +of other +Companies3 + +Membership +in Committees +of the Board +of other +Companies3 + +No. of shares +held as on +March 31, +2021 + +Attendance +at the last +AGM held +on +July 13, +2020 + +Other listed +companies where +the Director is +appointed as +Independent +Director + +4 + +12 + +- + +- + +- + +- + +Yes + +1,738,057@ + +Yes + +241,913,816@ + +- + +- + +1-May-2015 + +1-Sep-1968 + +- + +1 + +2 + +Rishad A. Premji +(DIN: 02983899) + +Azim H. Premji +(DIN: 00234280) + +Executive +Director and +Chairman + +Non- +Executive +Non- +Independent +Director + +126 + + Designation + +Date of initial +appointment + +Sl. +No. + +Name of +the Director +and Director +Identification +Number (DIN) + +Directorship +in other +Companies2 + +Chairmanship +in Committees +of Board +of other +Companies3 + +Membership +in Committees +of the Board +of other +Companies3 + +No. of shares +held as on +March 31, +2021 + +Attendance +at the last +AGM held +on +July 13, +2020 + +Other listed +companies where +the Director is +appointed as +Independent +Director + +Date of +appointment +as +Independent +Director under +Companies +Act, 2013 and +SEBI Listing +Regulations +(first term) 1 +- + +3 + +4 + +5 + +Thierry Delaporte +(DIN: 08107242) + +William Arthur +Owens +(DIN: 00422976) + +M. K. Sharma +(DIN: 00327684) + +Chief +Executive +Officer and +Managing +Director + +Independent +Director + +Independent +Director + +6-Jul-2020 + +1-Jul-2006 + +23-Jul-2014 + +1-Jul-2011 + +23-Jul-2014 + +- + +- + +9 + +6 + +Ireena Vittal +(DIN: 05195656) + +Independent +Director + +1-Oct-2013 + +23-Jul-2014 + +5 + +7 + +8 + +9 + +Patrick J. Ennis +(DIN: 07463299) + +Patrick Dupuis +(DIN: 07480046) + +Deepak M. +Satwalekar +(DIN: 00009627) + +Independent +Director + +Independent +Director + +Independent +Director + +1-Apr-2016 + +1-Apr-2016 + +1-Apr-2016 + +1-Apr-2016 + +1-Jul-2020 + +1-Jul- 2020 + +- + +- + +5 + +- + +- + +3 + +- + +- + +- + +1 + +- + +- + +2 + +2 + +- + +- + +1 + +Yes + +Yes + +Yes + +Yes + +Yes + +Yes + +Yes + +- + +- + +- + +- + +- + +- + +- + +- + +- + +1. Asian Paints +Limited + +2. United Spirits + +Limited +3. Ambuja +Cements +Limited +4. Vedanta +Limited + +1. Godrej + +Consumer +Products +Limited +2. Housing + +Development +Finance +Corporation +Limited + +- + +- + +1. Piramal + +Enterprises +Limited +2. Asian Paints +Limited +3. Home First +Finance +Company India +Limited +4. Piramal + +Capital & +Housing +Finance +Limited + +1. At the 70th AGM, Mr. M. K. Sharma was re-appointed as Independent Director for a second term from July 1, 2016 to June 30, 2021. + At the 71st AGM, Mr. William Arthur Owens was re-appointed as Independent Director for a second term from August 1, 2017 to July 31, 2022. + At the 72nd AGM, Mrs. Ireena Vittal was re-appointed as Independent Director for a second term from October 1, 2018 to September 30, 2023. + At the 74th AGM, Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director of the Company to hold office for a period + +of five years from July 6, 2020 to July 5, 2025. + + At the 74th AGM, Mr. Deepak M. Satwalekar was appointed as an Independent Director for a period of five years from July 1, 2020 to June 30, 2025. + Dr. Patrick J. Ennis and Mr. Patrick Dupuis were re-appointed as Independent Directors of the Company for a second term of 5 years, with effect from +April 1, 2021 to March 31, 2026. The said re-appointment was approved by shareholders of the Company vide special resolutions dated June 4, 2021, +passed through postal ballot by e-voting. + +127 + + Corporate Governance Report + +2. This does not include position in foreign companies and position as an advisory board member but includes position in private companies and + +companies under Section 8 of the Companies Act, 2013. None of our Directors hold directorship in more than seven listed companies. + +3. In accordance with Regulation 26 of the Listing Regulations, Membership/Chairmanship of only Audit Committees and Stakeholders’ Relationship + +Committees in all public limited companies have been considered. + +@ Includes equity shares held jointly with immediate family members. + +IV. COMMITTEES OF BOARD + +Our Board has constituted sub-committees to focus +on specific areas and make informed decisions within +the authority delegated to each of the Committees. +Each Committee of the Board is guided by its Charter, +which defines the scope, powers and composition of the +Committee. All decisions and recommendations of the +Committees are placed before the Board for information +or approval. + +During the financial year, the Board has accepted the +recommendations of Committees on matters where +such a recommendation is mandatorily required. There +have been no instances where such recommendations +have not been considered. + +We have four sub-committees of the Board as at March +31, 2021: + +1. Audit, Risk and Compliance Committee, which also + +acts as the Risk Management Committee + +2. Board Governance, Nomination and Compensation +Committee, which also oversees the CSR initiatives +of the Company and acts as the CSR Committee + +3. Administrative + +Shareholders/Investors +Grievance Committee (Stakeholders Relationship +Committee) + +and + +4. Strategy Committee + +1. Audit, Risk and Compliance Committee + +The Audit, Risk and Compliance Committee of our Board +reviews, acts on and reports to our Board with respect +to various auditing and accounting matters. The primary +responsibilities include overseeing: + +a) Auditing and accounting matters, + +including +recommending the appointment of our independent +auditors to the shareholders; + +b) Compliance with legal and statutory requirements; + +c) + +Integrity of the Company’s financial statements, +discussions with the independent auditors regarding +the scope of the annual audits, and fees to be paid to +the independent auditors; + +d) Performance of the Company’s + +internal audit +independent auditors and accounting + +function, +practices; + +e) Review of related party transactions and functioning + +of whistle blower mechanism; + +128 + +f) + +Implementation of the applicable provisions of the +Sarbanes Oxley Act of 2002 (the “Sarbanes Oxley +Act”), including review of the progress of internal +control mechanisms to prepare for certification +under Section 404 of the Sarbanes Oxley Act; + +g) Evaluation of + +internal financial controls, risk +management systems and policies including review +of cyber security; + +h) Review of utilization of loans and advances from, +and investment by, the Company in its subsidiaries +exceeding ` 100 crore or 10% of the asset size of +the subsidiary, whichever is lower, including existing +loans, advances and investments; and + +i) Such other matters and activities as the Committee +deems necessary for fulfilment of the above or as +may be approved by the Board of Directors or as may +be prescribed by applicable law from time to time. + +The Audit, Risk and Compliance Committee also acts +as the Risk Management Committee in compliance +with the Listing Regulations. The Committee reviews, +acts on and reports to our Board with respect to risk +management matters. The primary responsibilities +include the following: + +a) To formulate a detailed risk management policy + +which shall include: + +• A + +for + +framework + +in particular + +identification of + +internal +and external risks specifically faced by the +Company, +including financial, +operational, sectoral, sustainability (specifically, +Environmental, Social and Governance related +risks and impact), information and cyber security +risks. + +• Measures for risk mitigation + +• Systems for internal controls + +• Business contingency plan + +b) To monitor and oversee implementation of the +risk management policy, including evaluating the +adequacy of risk management and internal control +systems; + +c) Ensure that appropriate methodology, processes +and systems are in place to monitor and evaluate +risks associated with the business of the listed +entity; + + d) Evaluate risks related to cyber security and +significant risk exposures of the Company and +assess steps taken by the management to mitigate +the exposures in a timely manner (including business +continuity and disaster recovery planning); + +e) To review the risk management policy on annual +including by considering the changing + +basis, +industry dynamics and evolving complexity; + +f) To keep the Board of Directors informed about +its discussions, + +the nature and content of +recommendations and actions to be taken; and + +g) The appointment, removal and terms of remuneration +of the Chief Risk Officer (if any) shall be subject to +review by the Audit, Risk and Compliance Committee, +jointly with the Board Governance, Nomination and +Compensation Committee. + +Mr. Deepak Acharya has been appointed as the Chief +Risk Officer of the Company. + +The detailed charter of the Committee is posted +on our website and available at https://www.wipro. +com/investors/corporate-governance/. All members +of our Audit, Risk and Compliance Committee are +Independent Directors and financially literate. The +Chairman of our Audit, Risk and Compliance Committee +has the accounting and financial management related +expertise. + +Statutory Auditors as well as Internal Auditors always +have independent meetings with the Audit, Risk and +Compliance Committee and also participate in the +Audit, Risk and Compliance Committee meetings. +Our Chief Financial Officer, General Counsel, Internal +Auditor, Finance Controller and other Corporate Officers +make periodic presentations to the Audit, Risk and +Compliance Committee on various issues. + +Mr. M. K. Sharma, Independent Director, is the Chairman +of the Audit, Risk and Compliance Committee. The other +members of the Committee as at March 31, 2021 were +Mrs. Ireena Vittal and Mr. Deepak M. Satwalekar. The +Chairman of the Committee was present at the AGM +held on July 13, 2020. + +2. Board Governance, Nomination and Compensation + +Committee +The Board Governance, Nomination and Compensation +Committee reviews, acts on and reports to our Board +with respect to various governance, nomination and +compensation matters. The primary responsibilities +include: + +a) Developing and recommending to the Board +corporate governance guidelines applicable to the +Company; + +b) Evaluating the Board on a continuing basis, +including an assessment of the effectiveness of the +full Board, operations of the Board Committees and +contributions of individual directors; + +c) Establishing policies and procedures to assess the +requirements for induction of new members to the +Board; + +d) Implementing policies and processes relating to + +corporate governance principles; + +e) Ensuring that appropriate procedures are in place +to assess Board membership needs and Board +effectiveness; + +f) Reviewing the Company’s policies that relate to +matters of corporate social responsibility (CSR), +including public +issues of significance to the +Company and its shareholders; + +g) Formulating the Disclosure Policy, its review and + +approval of disclosures; + +h) Approving and evaluating the compensation plans, +policies and programs for full-time directors and +senior management; + +i) Acting as Administrator of the Company’s Employee +Stock Option Plans and Employee Stock Purchase +Plans drawn up from time to time; and + +j) Reviewing and recommending of all remuneration, +in whatever form, payable to senior management. + +The detailed charter of Board Governance, Nomination +and Compensation Committee is posted on our website +and is available at https://www.wipro.com/investors/ +corporate-governance/. + +Our Chief Human Resources Officer makes periodic +presentations to the Board Governance, Nomination and +Compensation Committee on compensation reviews and +performance linked compensation recommendations. +All members of the Board Governance, Nomination +and Compensation Committee are independent non- +executive directors. The Board Governance, Nomination +and Compensation Committee is the apex body that +oversees our Corporate Social Responsibility policy and +programs. + +Mr. William Arthur Owens, Independent Director, is the +Chairman of the Board Governance, Nomination and +Compensation Committee. The other members of the +Committee as at March 31, 2021 were Mrs. Ireena Vittal +and Mr. M. K. Sharma. The Chairman of the Committee +was present at the AGM held on July 13, 2020. + +3. Administrative and Shareholders/Investors Grievance +Committee (Stakeholders Relationship Committee) +The +Grievance Committee carries out + +and Shareholders/Investors +role of + +Administrative + +the + +129 + + Corporate Governance Report + +Stakeholders Relationship Committee in compliance +with Section 178 of the Companies Act, 2013 and the +Listing Regulations. + +Administrative + +The +and Shareholders/Investors +Grievance Committee reviews, acts on and reports to +our Board with respect to various matters relating to +stakeholders. The primary responsibilities include: + +a) Redressal of Grievances of the shareholders of the +Company pertaining to transfer or transmission +of shares, non-receipt of annual report and +declared dividends, issue of new or duplicate share +certificates, and grievances pertaining to corporate +actions; + +b) Approving consolidation, split or sub-division of +share certificates, transmission of shares, issue of +duplicate share certificates, re-materialization of +shares; + +c) Reviewing the grievance redressal mechanism +implemented by the Company in coordination with +Company’s Registrar and Transfer Agent (“RTA”) +from time to time; + +d) Reviewing the measures taken by the Company for +effective exercise of voting rights by shareholders; + +e) Implementing and overseeing the procedures and +processes in handling and maintenance of records, +transfer of securities and payment of dividend by +the Company, RTA and dividend processing bank; + +f) Reviewing the various measures and initiatives +taken by the Company for reducing the quantum of +unclaimed dividends and ensuring timely receipt +of dividend warrants, annual reports and statutory +notices by the shareholders of the Company. + +g) Overseeing administrative matters like opening +and closure of Company’s bank accounts, grant and +revocation of general, specific and banking powers +of attorney; and + +h) Considering and approving allotment of equity +shares pursuant to exercise of stock options, setting +up branch offices and other administrative matters +as delegated by Board from time to time. + +The detailed charter of the Committee is available +on our website at https://www.wipro.com/investors/ +corporate-governance/. + +Mr. M. K. Sharma, Independent Director, is the Chairman +of the Administrative and Shareholders/Investors +Grievance Committee. The other members of the +Committee as at March 31, 2021 were Mr. Deepak M. +Satwalekar and Mr. Rishad A. Premji. The Chairman +of the Committee was present at the Annual General +Meeting of the Company held on July 13, 2020. + +Mr. M. Sanaulla Khan, Company Secretary, is our +Compliance Officer under the Listing Regulations. + +Status Report of investor queries and complaints for +the period from April 1, 2020 to March 31, 2021 is given +below: + +Sl. +No. +1. + +2. + +3. + +4. + +Particulars + +No. of +Complaints + +Investor complaints pending at the +beginning of the year +Investor complaints received during +the year +Investor complaints disposed of +during the year +Investor +complaints +unresolved at the end of the year + +remaining + +NIL + +1,374* + +1,332 + +42** + +* + +of the 1,374 complaints received, 921 were clarifications +regarding buyback of equity shares. + +** These queries were received between March 25, 2021 +and March 31, 2021, and subsequently responded before +April 6, 2021. + +Apart from these queries/complaints, there are certain +pending cases relating to dispute over title to shares in +which in certain cases the Company has been made a +party. However, these cases are not material in nature. + +4. Strategy Committee + +The Strategy Committee reviews, acts and reports to our +Board with respect to the mission, vision and strategic +direction of the Company. Primary responsibilities of +this Committee, inter alia, are: + +a) Making recommendations to the Board relating to +the Company’s mission, vision, strategic initiatives, +major programs and services; + +b) Ensuring management has established an effective +strategic planning process, including development +of a three to five-year strategic plan with measurable +goals and time targets; + +c) Annually reviewing the strategic plan for the +Company and for each division and entity as well +and recommending updates to the Board; + +d) Establishing criteria for management to evaluate +potential strategic investments, reviewing proposals +for acquisition or divestment opportunities for the +Company and making appropriate recommendations +to the Board, and reviewing post-transaction +integration matters; and + +e) Monitoring the Company performance against +measurable targets (e.g. market share, increase in +revenue, or Operating Margin) or progress points +(such as emerging technologies). + +130 + + Mr. William Arthur Owens, Independent Director, is the Chairman of the Strategy Committee. The other members of the +Committee as at March 31, 2021 were Mr. Azim H. Premji, Mrs. Ireena Vittal, Dr. Patrick J. Ennis, Mr. Patrick Dupuis, +Mr. Thierry Delaporte and Mr. Rishad A. Premji. + +At the Board Meeting held on April 15, 2021, it was decided to discontinue the Strategy Committee, as the entire Board +will be deliberating matters pertaining to the strategy of the Company going forward. + +Attendance of Directors at Board and Committee meetings +Details of attendance of Directors at the Board meetings and Committee meetings for the year ended March 31, 2021 +are as under: + +Board meeting1 + +Audit, Risk and +Compliance +Committee2 + +Board Governance, +Nomination and +Compensation +Committee (also acts as +CSR Committee) + +Strategy +Committee + +Administrative +and Shareholders/ +Investors Grievance +Committee3 + +6 + +5 + +5 + +2 + +4 + +April 15, 2020, +May 29, 2020, +July 14, 2020, +October 12-13, +2020, January 13, +2021 and February +24, 2021 + +April 14, 2020, +May 28, 2020, +July 13, 2020, +October 12-13, +2020 and January +12-13, 2021 + +April 14, 2020, May 29, +2020, July 13, 2020, +October 12, 2020 and +January 12, 2021 + +October 12, +2020 and +January 12, +2021 + +April 14, 2020, July +13, 2020, October +12, 2020 and +January 12, 2021 + +6 +6 +4 +6 +6 +6 +6 +6 +4 + +2 + +1 + +NA +NA +NA +NA +5 +5 +NA +NA +3 + +2 + +NA + +NA +NA +NA +5 +5 +5 +NA +NA +NA + +NA + +NA + +2 +2 +2 +2 +NA +2 +2 +2 +NA + +NA + +NA + +4 +NA +NA +NA +3 +NA +NA +NA +3 + +1 + +NA + +No. of meetings held +during FY 2020-21^ +Date of meetings + +Attendance of +Directors +Rishad A. Premji +Azim H. Premji +Thierry Delaporte# +William Arthur Owens +M. K. Sharma +Ireena Vittal +Dr. Patrick J. Ennis +Patrick Dupuis +Deepak M. +Satwalekar* +Arundhati +Bhattacharya $ +Abidali Z. +Neemuchwala @ + +# At the 74th AGM, Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director of the Company to hold office for a + +period of five years from July 6, 2020 to July 5, 2025. + +* At the 74th AGM, Mr. Deepak M. Satwalekar was appointed as an Independent Director for a term of five years from July 1, 2020 to June 30, + +2025. + +$ The Board of Directors, at its meeting held on May 29, 2020, noted the resignation of Mrs. Arundhati Bhattacharya as an Independent Director +with effect from close of business hours on June 30, 2020. In her resignation letter, Mrs. Arundhati Bhattacharya has indicated that the reason +for her resignation is her decision to accept a full time role as Chairperson and CEO in another company. She has also confirmed that there is +no other material reason, other than the reason stated in her resignation letter. + +@ The Board of Directors, at its meeting held on May 29, 2020, noted the resignation of Mr. Abidali Z. Neemuchwala as the Chief Executive Officer + +and Managing Director of the Company with effect from the end of the day on June 1, 2020. + +^ Pursuant to the relaxations granted by the Ministry of Corporate Affairs and SEBI, all meetings of the Board of Directors and the Committees + +were held through video conferencing. + +1. Board Meeting: Since the appointment of Mr. Thierry Delaporte as Chief Executive Officer and Managing Director and Mr. Deepak M. Satwalekar + +as Independent Director, four Board meetings were held on July 14, 2020, October 12-13, 2020, January 13, 2021 and February 24, 2021. + +2. Audit, Risk and Compliance Committee: + +i. The Committee was re-constituted during the year as Mrs. Arundhati Bhattacharya resigned and Mr. Deepak M. Satwalekar was appointed + +131 + + Corporate Governance Report + +as a member of the Committee. Consequently, the composition of the Committee is as follows: Mr. M. K. Sharma (Chairman), Mrs. Ireena +Vittal and Mr. Deepak M. Satwalekar (Members). + +ii. Since the appointment of Mr. Deepak M. Satwalekar as member of the Committee, there were three Committee meetings held on July 13, + +2020, October 12-13, 2020, and January 12-13, 2021. + +3. Administrative and Shareholders/Investors Grievance Committee: + +i. The Committee was re-constituted during the year as Mrs. Arundhati Bhattacharya resigned and Mr. Deepak M. Satwalekar was appointed +as a member of the Committee. Consequently, the composition of the Committee is as follows: Mr. M. K. Sharma (Chairman), Mr. Deepak M. +Satwalekar and Mr. Rishad A. Premji (Members). + +ii. Since the appointment of Mr. Deepak M. Satwalekar as member of the Committee, there were three Committee meetings held on July 13, + +2020, October 12, 2020 and January 12, 2021. + +V. GOVERNANCE +PROCESS + +THROUGH + +MANAGEMENT + +1. Code of Business Conduct + +In the year 1983, we articulated ‘Wipro Beliefs’ consisting +of six statements. At the core of beliefs was integrity, +articulated as “individual and company relationship +should be governed by the highest standard of conduct +and integrity”. + +Over years, this articulation has evolved in form but +remained constant in substance. Today, we articulate it +as Code of Business Conduct. + +In our Company, the Board and all employees have +a responsibility to understand and follow the Code +of Business Conduct. All employees are expected to +perform their work with honesty and integrity. Wipro’s +Code of Business Conduct reflects general principles to +guide employees in making ethical decisions. This Code +is also applicable to our representatives. This Code +outlines fundamental ethical considerations as well +as specific considerations that need to be maintained +for professional conduct. This Code has been displayed +on the Company’s website at https://www.wipro.com/ +investors/corporate-governance/. + +Internal Audit +The Company has a robust internal audit function which +has been in place for last 4 decades with the stated +vision of “To be the best in class Internal Audit function +globally”. In pursuit of this vision, the function provides +an independent, objective assurance services to value- +add and improve Operations of Business Units and +processes by: + +a) Financial, Business Process and Compliance Audit + +b) Cyber Defense and Technology Audit + +c) Operations Reviews + +d) Best Practices and Benchmarking + +e) Leadership Development + +The function taking cognizance of changes in business +climate and technology risks has taken upon itself to +infuse and adopt Technology in its operations. + +2. + +132 + +The Chief of Internal Audit reports to the Chairman +of the Audit, Risk and Compliance Committee and +administratively to the Chief Financial Officer. Chief of +Internal Audit has regular and exclusive meetings with +the Audit, Risk and Compliance Committee. + +The internal audit function is guided by its charter, as +approved by the Audit, Risk and Compliance Committee. +The internal audit function formulates an annual risk +based audit plan based on consultations and inputs +from the Board and business leaders and presents +it to the Audit, Risk and Compliance Committee for +approval. Findings of various audits carried out during +the financial year are also periodically presented to the +Audit, Risk and Compliance Committee. The internal +audit function adopts a risk based audit approach and +covers core areas such as compliance audits, financial +audits, technology audits, third party risk audits, M&A +audits, etc. + +The internal audit team comprises of personnel with +professional qualifications and certifications in audit +and is rich on diversity. The audit team hones its skills +through a robust knowledge management program to +continuously assimilate the latest trends and skills +in the domain and to retain the knowledge gained for +future reference and dissemination. The internal audit +team re-asserts its independence across all its staff. + +A key strategic vision of Internal Audit is auditing in the +new digital environment “Auditing Digital and Auditing +with Digital”- in line with this, the Internal Audit function +has actively adopted Technology and Innovation to be +better equipped to carry out audits. + +The function, which was the first Indian Internal audit +unit to get ISO certified in 1998 and win International +award from Institute of Internal Auditors (IIA) in 2002, +was also an early adopter of the new ISO 9001:2015 +Version. During the year Internal Audit function is +assessed to have “Met +International Standards” +prescribed by the Professional Practice of Internal +Auditing issued by “International Institute of Internal +Auditors (IIA)” by external firm (KPMG). Testimony to +the functions’ innovation and excellence are the IIA +awards won in these categories continuously over the +last few years. + + 3. Disclosure Policy + +In line with requirements under Regulation 30 of the +Listing Regulations, the Company has framed a policy +on disclosure of material events and information +as per the Listing Regulations, which is available on +our website at https://www.wipro.com/investors/ +corporate-governance/. The objective of this policy is to +have uniform disclosure practices and ensure timely, +adequate and accurate disclosure of information on +an ongoing basis. The Company has constituted a +Disclosure Committee consisting of senior officials, +which approves all disclosures required to be made +by the Company. Parity in disclosures are maintained +through simultaneous disclosure on National Stock +Exchange of India Limited, the BSE Limited and the New +York Stock Exchange. + +4. Policy for Preservation of Documents + +Pursuant to the requirements under Regulation 9 of +the Listing Regulations, the Board has formulated and +approved a Document Retention Policy prescribing the +manner of retaining the Company’s documents and +the time period up to which certain documents are to +be retained. The policy percolates to all levels of the +organization who handle the prescribed categories of +documents. + +5. Other Policies + +The Company has adopted an Ombuds policy (vigil +mechanism), a policy for prevention, prohibition & +redressal of sexual harassment of women at workplace, +as well as a code of conduct to regulate, monitor and +report insider trading. Details of these are provided as +part of the Board’s report. + +VI. DISCLOSURES + +1. Disclosure of Materially Significant Related Party + +Transactions + +All related party transactions entered during the +financial year were at an arm’s length basis and in the +ordinary course of business. There are no materially +significant related party transactions made by the +Company with Promoters, Directors, Key Managerial +Personnel or other designated persons which may have +a potential conflict with the interest of the Company at +large. + +As required under Regulation 23 of Listing Regulations, +the Company has adopted a policy on Related Party +Transactions. The abridged policy on Related Party + +Transactions is available on the Company’s website at +https://www.wipro.com/content/dam/nexus/en/ +i nv e s to r /c o r p o ra te - go v e r n a n c e /p o l i c i e s - a n d - +guidelines/ethical-guidelines/abridged-policy-for- +related-party-transactions.pdf. + +Apart from receiving director remuneration, none +of the Directors has any pecuniary relationships or +transactions vis-à-vis the Company. During the year +2020-21, no transactions of material nature were +entered by the Company with the Management or their +relatives that may have a potential conflict of interest +with the Company and the concerned officials have +given undertakings to that effect as per the provisions +of the Listing Regulations. + +The Register under Section 189 of the Companies Act, +2013 is maintained and particulars of the transactions +have been entered in the Register, as applicable. + +2. Subsidiary Monitoring Framework + +All the subsidiary companies of the Company are +managed by their Boards having the rights and +obligations to manage these companies in the best +interest of respective stakeholders. The Company +its representatives on the Board of +nominates +subsidiary companies and monitors performance of +such companies, inter alia, by reviewing: + +a) Financial statements, + +the + +investment, + +inter- +corporate loans/advances made by the unlisted +subsidiary companies, statement containing all +significant transactions and arrangements entered +by the unlisted subsidiary companies forming part +of the financials being reviewed by the Audit, Risk +and Compliance Committee of the Company on a +quarterly basis. + +b) Minutes of the meetings of the unlisted subsidiary +companies, if any, are placed before the Company’s +Board regularly. + +c) Providing necessary guarantees, letter of comfort +and other support for their day-to-day operations +from time-to-time. + +As required under Regulation 16(1)(c) and 24 of +Listing Regulations, the Company has adopted a +policy on determining “material subsidiary” and the +said Policy is available on the Company’s website at +https://www.wipro.com/content/dam/nexus/en/ +i nv e s to r /c o r p o ra te - go v e r n a n c e /p o l i c i e s - a n d - +guidelines/ethical-guidelines/abridged-policy-for- +related-party-transactions.pdf. + +133 + + Corporate Governance Report + +3. Details of non-compliance by the Company, +penalties, and strictures imposed on the Company +by Stock Exchanges or SEBI or any statutory +authority, on any matter related to capital markets, +during the last three years. + +The Company has complied with the requirements of +the Stock Exchanges or SEBI on matters related to +Capital Markets, as applicable, during the last three +years. No penalties or strictures have been imposed on +the Company. + +4. Whistle Blower Policy and affirmation that no +personnel have been denied access to the Audit, +Risk & Compliance Committee + +As mentioned earlier in this report, the Company has +adopted an Ombuds process which is a channel for +receiving and redressing employees’ complaints. No +personnel in the Company has been denied access +to the Audit, Risk and Compliance Committee or its +Chairman. + +5. Transfer to Investor Education and Protection + +Fund Authority + +a) Pursuant to the provisions of Section 124(6) of the +Companies Act, 2013 and Investor Education and +Protection Fund (IEPF) Rules, during the year 2020-21, +an amount of ` 1,255,390 pertaining to unclaimed sale +proceeds of fractional shares arising out of scheme of +arrangement between Wipro Limited, Wipro Enterprises +Limited (formerly known as Azim Premji Custodial +Services Private Limited) and Wipro Trademarks Holding +Limited, was transferred to the IEPF. + +b) Pursuant to the provisions of Section 124(6) of the +Companies Act, 2013 and IEPF rules, during the +year 2020-21, unclaimed dividend for financial years +2012-13 and 2013-14 of ` 10,019,025 and ` 6,925,143 +respectively, together with an aggregate of 227,395 +equity shares in respect of which dividend had not been +claimed were transferred to the IEPF. + +6. Disclosures with respect to demat suspense +account/unclaimed suspense account (Unclaimed +Shares) + +In accordance with Regulation 39 and Schedule VI of the +Listing Regulations, a minimum of three reminders are +sent to shareholders, towards the shares which remain +unclaimed. In case of non-receipt of response to the +reminders from the shareholders, the unclaimed shares +are transferred to the Unclaimed Suspense Account. +The Company maintains the details of shareholding +of each individual shareholders whose shares are +transferred to the Unclaimed Suspense Account. When +a claim from a shareholder is received by the Company, +the shares lying in the Unclaimed Suspense Account + +134 + +are transferred after due verification of documents +submitted by the shareholder. +Further, the shares in respect of which dividend +entitlements remained unclaimed for seven consecutive +years are transferred from the Unclaimed Suspense +Account to IEPF Authority in accordance with Section +124(6) of the Companies Act, 2013 and rules made +thereunder. +The disclosure as required under Schedule V of the +Listing Regulations is given below for the financial year +2020-21: + +Sl. +No. +1. + +2. + +3. + +4. + +5. + +6. + +Particulars + +number +and + +of +Aggregate +shareholders +the +outstanding shares in the +suspense account lying at +the beginning of the year +Number of shareholders +who +the +approached +Company for transfer of +suspense +from +shares +account during the year +Number of shareholders +to whom shares were +transferred from suspense +account during the year +Number of shares +in +respect of which dividend +remained +entitlements +for +unclaimed +seven +years and +consecutive +the +transferred +from +Unclaimed +Suspense +Account to the IEPF +of +Aggregate +shareholders +the +outstanding shares in the +suspense account lying at +the end of the year +Voting +these +rights on +shares shall remain frozen +till the rightful owner of +such shares claim +the +same + +number +and + +No. of +Shareholders +295 + +No. of +Shares +271,916 + +Nil + +Nil + +Nil + +Nil + +- + +4,920* + +295 + +266,996 + +Yes + +* Represents a portion of the shares held by 29 shareholders, +whose balance shares continue to remain in the unclaimed +suspense account. + +7. Shareholder Information +Various shareholder +to be +disclosed pursuant to Schedule V of the Listing +Regulations are provided in Annexure I to this report. + +information + +required + +8. Compliance with Mandatory Requirements + +Your Company has complied with all the mandatory +corporate governance requirements under the Listing +Regulations. Specifically, your Company confirms +compliance with corporate governance requirements + + specified in Regulation 17 to 27 and clauses (b) to (i) +of Sub-regulation (2) of Regulation 46 of the Listing +Regulations. + +9. Certificates from Practising Company Secretary + +The certificate dated June 9, 2021, +issued by +Mr. V. Sreedharan, Partner, V. Sreedharan & Associates, +Practising Company Secretaries is given at page no. +142 of this Annual Report in compliance with corporate +governance norms prescribed under the Listing +Regulations. + +The Company has received certificate dated June 9, +2021, from Mr. V. Sreedharan, Partner, V. Sreedharan & +Associates, Practising Company Secretaries, confirming +that none of the Directors of the Company have been +debarred or disqualified from being appointed or +continuing as director of companies by the SEBI/ +Ministry of Corporate of Affairs or any such authority. +The certificate is given at page no. 143 of this Annual +Report. + +VII. COMPLIANCE REPORT ON DISCRETIONARY +REQUIREMENTS UNDER REGULATION 27(1) OF +THE LISTING REGULATIONS + +1. + +The Board +As per Para A of Part E of Schedule II of the Listing +Regulations, a non-executive Chairman of the Board +may be entitled to maintain a Chairman’s Office at the +company’s expense and also allowed reimbursement +of expenses incurred in performance of his duties. The +Chairman of the Company is an Executive Director and +hence this provision is not applicable to us. + +2. Shareholders rights + +Considering the dynamic shareholder demography and +trading on the stock exchanges, as a prudent measure, +we display our quarterly and half yearly results on our +website www.wipro.com and also publish our results in +widely circulated newspapers. We have communicated +the payment of dividend by e-mail to shareholders in + +addition to dispatch of letters to all shareholders. We +publish the voting results of shareholder meetings +and make it available on our website www.wipro.com, +and report the same to Stock Exchanges in terms of +Regulation 44 of the Listing Regulations. + +3. Modified opinion(s) in audit report + +The Auditors have issued an unmodified opinion on the +financial statements of the Company. + +4. Reporting of Internal Auditor + +Reporting of Head of Internal Audit is to the Chairman +of the Audit, Risk and Compliance Committee and +administratively to the Chief Financial Officer. Head of +Internal Audit has regular and exclusive meetings with +the Audit Committee. + +5. NYSE Corporate Governance Listing Standards + +The Company has made necessary disclosures in +compliance with the New York Stock Exchange Listing +Standards and NYSE Listed Company Manual on its +website https://www.wipro.com/investors/ +corporate-governance + +Bengaluru +June 9, 2021 + +Rishad A. Premji +Chairman + +Declaration as required under Regulation 34(3) and +Schedule V of the Listing Regulations + +All Directors and senior management personnel of the +Company have affirmed compliance with Wipro’s Code of +Business Conduct for the financial year ended March 31, +2021. + +Bengaluru +June 9, 2021 + +Rishad A. Premji +Chairman + +Thierry Delaporte +Chief Executive Officer and Managing Director + +135 + + SHAREHOLDER INFORMATION + +Annual General Meeting + +ANNEXURE I + +Pursuant to the General Circular No. 14/2020 dated April 8, 2020, the General Circular No. 17/2020 dated April 13, 2020, the +General Circular No. 20/2020 dated May 5, 2020, the General Circular No. 22/2020 dated June 15, 2020, the General Circular +No. 33/2020 dated September 28, 2020, the General Circular No. 39/2020 dated December 31, 2020 and the General Circular +No. 02/2021 dated January 13, 2021 issued by the Ministry of Corporate Affairs (“MCA Circulars”), the 75th Annual General +Meeting (“AGM”) for the year ended March 31, 2021 is scheduled to be held on Wednesday, July 14, 2021 at 9 am IST through +Video Conferencing (“VC”). + +The Members may attend the 75th AGM scheduled to be held on Wednesday, July 14, 2021, 9 am IST onwards, through VC +or watch the live web-cast at https://www.wipro.com/AGM2021/. Detailed instructions for participation are provided in the +notice of the 75th AGM. The proceedings of the 75th AGM will be available through VC and live web-cast to the shareholders as +on the cut-off date i.e. July 7, 2021. + +Annual General Meetings and Other General Body meeting of the last three years and Special Resolutions, if any. + +Financial Year +2017-18 + +2018-19 + +Date and Time +July 19, 2018 +at 4.00 pm +July 16, 2019 +at 4.00 pm + +2019-20 + +July 13, 2020 +at 9.00 am + +Venue +Wipro Campus, Cafeteria +Hall EC-3, Ground Floor, +Opp. Tower 8, No. 72, +Keonics, Electronic City, +Hosur Road, +Bengaluru – 561 229 +Meeting held through VC + +Special resolutions passed +i. Re-appointment of Mrs. Ireena Vittal (DIN: 05195656) as Independent + +Director of the Company. + +i. Amendments to the Articles of Association of the Company +ii. Appointment of Mr. Azim H. Premji (DIN: 00234280) as Non-Executive, + +Non-Independent Director of the Company + +i. Appointment of Mr. Deepak M. Satwalekar (DIN: 00009627) as an + +Independent Director of the Company + +Details of resolutions passed through postal ballot during Financial Year 2020-21 and details of the voting pattern +The Company sought the approval of shareholders through notice of postal ballot dated October 13, 2020 for approval +for Buyback of Equity Shares by way of special resolution. The aforesaid resolutions were duly passed and the results of +postal ballot/e-voting were announced on November 17, 2020. Mr. V. Sreedharan/Ms. Devika Sathyanarayana/Mr. Pradeep +B. Kulkarni, partners of V. Sreedharan & Associates, Practicing Company Secretaries, were appointed as the Scrutinizer to +scrutinize the postal ballot and remote e-voting process in a fair and transparent manner. + +Resolution + +No. of Votes +Polled + +No. of Votes Cast +in Favour + +No. of Votes Cast +Against + +% of Votes Cast in +Favour on Votes +Polled + +% of Votes Cast +Against on Votes +Polled + +Approval for Buyback of Equity +Shares + +5,089,856,744 + +5,078,618,396 + +11,238,348 + +99.78 + +0.22 + +Procedure for Postal Ballot +The postal ballot is conducted in accordance with the provisions contained in Section 110 and other applicable provisions, +if any, of the Companies Act, 2013, read with Rule 22 of the Companies (Management and Administration) Rules, 2014. The +Shareholders are provided the facility to vote either by physical ballot or through e-voting. The postal ballot notice is sent to +shareholders as per the permitted mode wherever applicable. The Company also publishes a notice in the newspapers in +accordance with the requirements under the Companies Act, 2013. + +Shareholders holding equity shares as on the cut-off date may cast their votes through e-voting or through postal ballot +during the voting period fixed for this purpose. After completion of scrutiny of votes, the scrutinizer submits his report to the +Chairman and the results of voting by postal ballot are announced within 2 working days of conclusion of the voting period. +The results are displayed on the website of the Company (www.wipro.com), and communicated to the Stock Exchanges, +Depositories, and Registrar and Share Transfer Agents. The resolutions, if passed by the requisite majority, are deemed to +have been passed on the last date specified for receipt of duly completed postal ballot forms or e-voting. + +In view of the COVID-19 pandemic, the MCA permitted companies to transact items through postal ballot as per the framework +set out in Circular No. 14/2020 dated April 8, 2020, Circular No. 17/2020 dated April 13, 2020, the General Circular No. 22/2020 +dated June 15, 2020, Circular No. 33/2020 dated September 28, 2020 and General Circular No. 39/2020 dated December 31, +2020. In accordance with the aforementioned circulars, e-voting facility was provided to all the shareholders to cast their +votes only through the remote e-voting process as per notice of postal ballot dated October 13, 2020 for approval for Buyback +of Equity Shares. + +136 + + Further, the Company sought the approval of shareholders through notice of postal ballot dated April 15, 2021 for the re- +appointment of Dr. Patrick J. Ennis (DIN: 07463299) and Mr. Patrick Dupuis (DIN: 07480046) as Independent Directors on the +Board of Wipro Limited by way of special resolutions. The aforesaid resolutions were duly passed and the results of postal +ballot/e-voting were announced on June 6, 2021. Mr. V. Sreedharan/Ms. Devika Sathyanarayana/Mr. Pradeep B. Kulkarni, +partners of V. Sreedharan & Associates, Practicing Company Secretaries, were appointed as the Scrutinizer to scrutinize the +postal ballot and remote e-voting process in a fair and transparent manner. + +Resolution + +No. of Votes +Polled + +No. of Votes Cast +in Favour + +No. of Votes Cast +Against + +% of Votes Cast in +Favour on Votes +Polled + +% of Votes Cast +Against on Votes +Polled + +Re-appointment of Dr. Patrick +J. Ennis (DIN: 07463299) as an +Independent Director of the +Company + +Re-appointment of Mr. Patrick +Dupuis (DIN: 07480046) as an +Independent Director of the +Company + +4,905,658,196 + +4,869,602,676 + +36,055,520 + +99.27 + +0.73 + +4,905,658,087 + +4,880,958,490 + +24,699,597 + +99.50 + +0.50 + +Means of Communication with Shareholders/Analysts +We have established procedures to disseminate, in a planned manner, relevant information to our shareholders, analysts, +employees and the society at large. Our Audit, Risk and Compliance Committee reviews the earnings press releases, Form 20-F +filed with Securities Exchange Commission (SEC) filings and annual and quarterly reports of the Company, before they are +presented to the Board for their approval for release. The details of the means of communication with shareholders/analysts +are given below: + +News Releases and +Presentations +Quarterly results + +Website + +Annual Report + +Other Disclosures/Filings + +All our news releases and presentations made at investor conferences and to analysts are posted on the +Company’s website at https://www.wipro.com/investors. +Our quarterly results are published in widely circulated national newspapers such as Financial Express +and the local daily Kannada Prabha. +The Company’s website contains a dedicated section for Investors (https://www.wipro.com/investors), +where annual reports, earnings press releases, stock exchange filings, quarterly reports, and corporate +governance policies are available, apart from the details about the Company, Board of Directors +and Management. +Annual Report containing audited standalone accounts, consolidated financial statements together +with Board’s Report, Corporate Governance Report, Management Discussion and Analysis Report, +Auditors Report and other important information are circulated to the Members entitled thereto through +permitted mode(s). +Our Form 20-F filed with SEC containing detailed disclosures, along with other disclosures including +Press Releases etc. are available at https://www.wipro.com/investors. + +Communication of Results: +Means of Communications +Earnings Calls +Publication of results +Analysts/Investors Meetings/Analyst Day + +Number of times during 2020-21 +4 +4 +Details are provided in the MD&A Report forming part of this Annual Report. + +Financial Calendar +The financial year of the Company starts from the 1st day of April and ends on 31st day of March of next year. Our tentative +calendar for declaration of results for the financial year 2021-22 are as given below. In addition, the Board may meet on other +dates as and when required. + +Quarter Ending +For the Quarter ending June 30, 2021 +For the Quarter and half year ending September 30, 2021 +For the Quarter and nine months ending December 31, 2021 +For the year ending March 31, 2022 + +Release of Results + +Third week of July, 2021 +Third week of October, 2021 +Third week of January, 2022 +Third week of April, 2022 + +The Register of Members and Share Transfer books will remain closed from Monday, July 12, 2021 to Tuesday, July 13, 2021 +(both days inclusive). + +137 + + Fees Paid to Statutory Auditors +The details of total fees for all services paid by the Company and its subsidiaries, on a consolidated basis, to the statutory +auditor and all the entities in the network firm/network entity of which the statutory auditor is a part, are as follows: + +Type of Service +Audit Fees +Tax Fees +Others + +Total + + (In ` Mn) + +FY 2020-21 +83 +44 +13 + +FY 2019-20 +81 +51 +17 + +140 + +149 + +Corporate Information +a) Corporate Identity Number (CIN): L32102KA1945PLC020800 +b) Company Registration Number: 20800 +c) International Securities Identification Number (ISIN): INE075A01022 +d) CUSIP Number for Wipro American Depository Shares: 97651M109 +e) Details of exchanges where Company’s shares are listed in as at March 31, 2021: + +Equity shares +BSE Limited (BSE) + +Stock Codes +507685 + +National Stock Exchange of India Limited (NSE) + +WIPRO + +Address +BSE Limited, Phiroze Jeejeebhoy Towers, Dalal Street, +Mumbai - 400 001 +Exchange Plaza, C-1, Block G, Bandra Kurla Complex, +Bandra (E), Mumbai – 400 051 + +American Depository Receipts +New York Stock Exchange (NYSE) + +WIT + +11 Wall St, New York, NY 10005, United States of America + +Notes: +1. Listing fees for the year 2021-22 has been paid to the Indian Stock Exchanges as on date of this report. +2. Listing fees to NYSE for the calendar year 2021 has been paid as on date of this report. +3. The stock code on Reuters is WIPR.NS and on Bloomberg is WPRO:IN + +Distribution of Shareholding as on March 31, 2021 + +March 31, 2021 + +March 31, 2020 + +Category + (No. of Shares) +1-5000 +5001- 10000 +10001- 20000 +20001- 30000 +30001- 40000 +40001- 50000 +50001- 100000 +100001 & Above + +Total + +No. of +Shareholders +829,647 +3,628 +1,936 +751 +371 +261 +668 +1,329 + +% of +Shareholders +98.93 +0.43 +0.23 +0.09 +0.04 +0.03 +0.08 +0.17 + +No. of Shares +83,119,655 +12,699,508 +13,864,634 +9,354,600 +6,483,397 +5,881,802 +23,626,382 +5,324,108,577 + +% of Total +Equity + +1.52 +0.23 +0.25 +0.17 +0.12 +0.11 +0.43 +97.17 + +No. of +Shareholders +507,272 +1,697 +1,049 +400 +240 +162 +358 +703 + +% of +Shareholders +99.10 +0.33 +0.20 +0.08 +0.05 +0.03 +0.07 +0.14 + +No. of Shares + +82,232,977 +12,032,833 +14,869,292 +9,808,813 +8,376,285 +7,250,873 +25,599,263 +5,553,187,054 + +% of Total +Equity + +1.44 +0.21 +0.26 +0.17 +0.15 +0.13 +0.45 +97.20 + +838,591 + +100.00 + +5,479,138,555 + +100.00 + +511,881 + +100.00 + +5,713,357,390 + +100.00 + +Market Share Price Data +The performance of our stock in the financial year 2020-21 is tabulated below: + +2020-21 + +April +133,037,800 + +May +151,908,123 + +June +195,006,806 + +July +460,217,867 + +August +195,174,626 + +September +435,740,889 + +October +587,943,024 + +November +148,723,864 + +December +241,053,563 + +January +467,630,831 + +February +225,393,445 + +March +260,495,775 + +202.95 +09-Apr-20 +3,465,831 +173.8 +21-Apr-20 + +214.9 +29-May-20 +30,630,218 +178 +19-May-20 + +230 +26-Jun-20 +19,194,612 +206.3 +12-Jun-20 + +290.8 +31-Jul-20 +16,700,905 +218.35 +01-Jul-20 + +287.4 +18-Aug-20 +12,054,273 +268 +31-Aug-20 + +NSE +324.5 +21-Sep-20 +22,739,047 +269 +01-Sep-20 + +381.7 +13-Oct-20 +36,092,383 +312.05 +01-Oct-20 + +361.4 +24-Nov-20 +10,092,441 +332.65 +03-Nov-20 + +390.5 +29-Dec-20 +11,459,126 +346.25 +01-Dec-20 + +467.45 +13-Jan-21 +29,190,193 +385.05 +01-Jan-21 + +451.75 +09-Feb-21 +29,474,830 +408 +26-Feb-21 + +444.3 +04-Mar-21 +12,963,797 +397.75 +25-Mar-21 + +Volume +traded +on NSE + +High +Date +Volume +Low +Date + +138 + + Volume + +9,043,319 + +5,743,344 + +6,209,370 + +High +Low + +High +Low + +High + +Low + +9,889.05 +8,055.8 + +-13.23% +9.03% + +-13.50% + +7.25% + +9,598.85 +8,806.75 + +10,553.15 +9,544.35 + +5.89% +2.42% + +-2.93% + +9.32% + +7.03% +15.90% + +9.94% + +8.38% + +3,952,617 + +9,257,331 + +20,571,550 + +11,794.25 +10,882.25 + +12,293,054 + S&P CNX NIFTY +11,618.1 +10,790.2 +Wipro Price Movement vis-à-vis Previous Month High/Low (%) +12.91% +0.37% + +12,025.45 +11,347.05 + +17.63% +16.00% + +-1.17% +22.74% + +11,341.4 +10,299.6 + +26.43% +5.84% + +S&P CNX Nifty Index vis-à-vis Previous Month High/Low (%) +-1.49% + +3.99% + +3.51% + +5.66% + +-0.85% + +5.16% + +7.47% + +7.91% + +4,757,798 + +13,145.85 +11,557.4 + +10,693,567 + +5,042,336 + +11,731,941 + +12,737,129 + +14,024.85 +12,962.8 + +14,753.55 +13,596.75 + +15,431.75 +13,661.75 + +15,336.3 +14,264.4 + +-5.32% +6.60% + +9.32% + +1.85% + +8.05% +4.09% + +6.69% + +12.16% + +19.71% +11.21% + +5.20% + +4.89% + +-3.36% +5.96% + +4.60% + +0.48% + +-1.65% +-2.51% + +-0.62% + +4.41% + +ADS Share Price during the Financial Year 2020-21 +August +May + +June + +April + +July + +September + +October + +November December + +January + +February + +March + +3.4 + +3.36 + +3.32 + +4.34 + +4.33 + +4.76 + +4.9 + +5.17 + +5.67 + +6.36 + +6.24 + +6.42 + +8,529.46 + +8,672.52 + +8,851.62 + +9,354.97 + +9,909.38 + +9,594.92 + +9,298.03 + +10,382.02 + +10,720.47 + +10,541.65 + +10,900.66 + +10,941.79 + +8.28% + +-1.18% + +-1.19% + +30.72% + +-0.23% + +9.93% + +2.94% + +5.51% + +9.67% + +12.17% + +-1.89% + +2.88% + +8.04% + +1.68% + +2.07% + +5.69% + +5.93% + +-3.17% + +-3.09% + +11.66% + +3.26% + +-1.67% + +3.41% + +0.38% + +Wipro ADS price in +NYSE during each +month closing + +NYSE TMT index +during each month +closing + +Wipro ADS Price +Movement (%) +vis-à-vis Previous +month Closing + +NYSE TMT Index +movement (%) +vis-à-vis Previous +month closing + +Performance of Wipro equity shares relative to the SENSEX and NYSE Composite index during the period +April 1, 2020 to March 31, 2021 is given in the following chart: + +250 +240 +230 +220 +210 +200 +190 +180 +170 +160 +150 +140 +130 +120 +110 +100 +90 +80 +70 + +0 +2 +0 +2 +- +r +p +A +- +1 + +0 +2 +0 +2 +- +y +a +M +- +1 + +0 +2 +0 +2 +- +y +a +M +- +1 +3 + +0 +2 +0 +2 +- +n +u +J +- +0 +3 + +0 +2 +0 +2 +- l -| | The cornerstone | | of | our approach | involves | con- | | | | | | | | -| --- | --------------- | --- | --- | ------------ | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -4 ExperimentalSettings -| trasting | | a consistent | | answer | with an | inconsis- | | | | | | | | -| --------------- | --- | ------------ | -------------------------- | -------- | ----------- | --------- | ------------ | --- | --- | --- | --- | --- | --- | -| tent | one | under | the | guidance | of the same | oracle | 4.1 Datasets | | | | | | | -| groundedquotes. | | | Thisalignswiththeconceptof | | | | | | | | | | | -WeconductexperimentsontheALCEbenchmark -| Reinforcement | | | Learning | from | Human | Feedback | | | | | | | | -| ------------- | --- | --- | -------- | ---- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | -(Gaoetal.,2023b),designedforattributedtextgen- -| (RLHF) | | (Ouyang | et | al., 2022; | Bai et | al., 2022), | | | | | | | | -| ------ | --- | ------- | --- | ---------- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -eration. Thebenchmarkincludesthreelong-form -whereLLMsarefurtherfine-tunedtodistinguish -QAdatasetsthatspanvarioustypesofquestions. -betweendesirableandundesirableresponsesunder -preference feedback. However, such contrastive ASQA (Stelmakh et al., 2022) is a long-form -preferencefeedbacktypicallycomesfromhuman factoidQAdatasetcharacterizedbyinherentlyam- -| annotation. | | Toautomaticallyconstructpreference | | | | | | | | | | | | -| ----------- | --- | ---------------------------------- | --- | --- | --- | --- | ------- | --------- | ---- | ------- | -------- | ----- | --- | -| | | | | | | | biguous | questions | that | require | multiple | short | an- | -pairsforpreferenceoptimization,weutilizetheat- swerstoencapsulatedifferentviewpoints. -tributedanswersgeneratedbysmallerLLMs(e.g., -LLaMA-2-7B)underthein-contextlearningsetting ELI5 (Fanetal.,2019)featuresopen-endedques- -tionsintendedforsimplificationtothecomprehen- -| asnegativesamples. | | | | Theseanswers,characterized | | | | | | | | | | -| ------------------ | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -sionleveloffive-year-olds,requiringexplanatory -bytheirlowqualityandinconsistencywithoracle -multi-sentenceresponses. -groundedquotes,automaticallyserveascontrastive -supervisionsignalswhenpairedwithhigh-quality -| | | | | | | | QAMPARI | | (Amouyaletal.,2022)isafactoid | | | | | -| ------------------------------- | --- | --- | --- | --- | --------------- | --- | ------- | --- | ----------------------------- | --- | --- | --- | --- | -| attributedanswerslabeledin§3.1. | | | | | Inthisscenario, | | | | | | | | | -QAdatasetderivedfromWikipedia,whereanswers -theprocessnotonlyencouragestheLLMtogener- arestructuredasacompilationofentities. -ateattributedanswersthataremoreconsistentwith -thegroundedquotesbutalsofacilitatestheidentifi- 4.2 EvaluationMetrics -cationandcorrectionofnuancederrorspresentin Following the ALCE benchmark (Gao et al., -smallermodels. 2023b), our evaluation primarily focuses on two -| | Specifically, | | we adopt | Direct | Preference | Opti- | | | | | | | | -| --- | ------------- | --- | -------- | ------ | ---------- | ----- | --------------- | --- | -------- | --- | ------- | ------------ | --- | -| | | | | | | | key dimensions: | | Citation | | Quality | and Correct- | | -mization(Rafailovetal.,2023),avariantofRLHF ness. Detaileddescriptionsofadditionalevaluation -known for its stability, for our contrastive align- dimensionsarepresentedintheAppendixB. -ment. Formally,foreachinstance,giventheoracle -| | | | | | | | CitationQuality. | | Citationqualityiscriticalfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | --- | -groundedg(i)alongwithaconsistentoracleanswer -| (i) | | | | | (i) | | evaluating | LLM | attribution, | | assessed | along | two | -| --- | --------------------------- | --- | --- | --- | --- | --------- | ----------- | --- | ------------ | ------- | ----------- | ----- | --- | -| y | aswellasanattributedanswery | | | | | generated | | | | | | | | -| w | | | | | l | | | | | | | | | -| | | | | | | | dimensions: | (1) | Citation | Recall, | determining | | if | -byaweakerLLMviain-contextlearning,wecan -| | | | | | | | the output | is entirely | | supported | by | the cited | doc- | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | --- | --------- | --- | --------- | ---- | -simplyconstructapreferencedataset: -| | | | | | | | uments, | and (2) | Citation | Precision, | | assessing | if | -| --- | --- | --- | --------------------- | --- | ------------- | --- | ---------------------------------------------- | ------- | -------- | ---------- | --- | --------- | --- | -| | | D | = (cid:8) x(i),τ(i),τ | | (i)(cid:9)N , | | | | | | | | | -| | | | | w | | (4) | eachcitationsupportsitscorrespondingstatement. | | | | | | | -l i=1 -| | (i) | | (i) | | | | EvaluationisconductedbyTRUE(Honovichetal., | | | | | | | -| ------ | --- | -------- | --- | ------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -| whereτ | | = g(i)◦y | | denotestheconcatenationof | | | | | | | | | | -| | w | | w | | | | 2022),aT5-11Bmodelfine-tunedonacollection | | | | | | | -theoraclegroundingwiththeconsistent,attributed -ofNLIdatasetstoautomaticallyexaminetheentail- -| | | (i) | g(i)◦y | (i) | | | | | | | | | | -| -------- | --- | ------------ | ------ | ----------------------- | ------- | ------- | ------------------------------------------ | --- | ------- | ---------- | ------- | --- | ------ | -| answer,τ | | = | | denotestheconcatenation | | | | | | | | | | -| | | l | | l | | | mentofciteddocumentsandthemodelgeneration. | | | | | | | -| with | the | inconsistent | | attributed | answer. | Here, ◦ | | | | | | | | -| | | | | | | | Additionally, | to | capture | a holistic | measure | | of ci- | -signifiestheoperationofstringconcatenation. -tationquality,wealsoreporttheCitationF1,the -| | Finally,wecanoptimizethepolicymodelπ | | | | | on | | | | | | | | -| --- | ------------------------------------ | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | θ | harmonicmeanofcitationprecisionandrecall: | | | | | | | -thedatasetD -byminimizingthefollowingloss: -citationprecision·citationrecall -| | L | (π ;π | ;D) | | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | ----- | -| | DPO | θ | ref | | | | F 1 = 2· | | | | | | , (6) | -citationprecision+citationrecall -| | | | (cid:20) | (cid:18) | π (τ | |x) | | | | | | | | -| --- | --- | --- | -------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -| | =−E | | | | θ | w | | | | | | | | -| | | | | logσ | βlog | | | | | | | | | -(x,τw,τ l )∼D π (τ |x) (5) Correctness. FortheASQAdataset,correctness -ref w -π (τ |x) (cid:19)(cid:21) is quantified using exact match recall (EM Rec.) -θ l -| | −βlog | | | , | | | | | | | | | | -| --- | ----- | --- | ------ | --- | --- | --- | ----------- | ------- | --- | --------- | ------- | --- | ----- | -| | | π | (τ |x) | | | | by checking | whether | | the short | answers | are | exact | -| | | | ref l | | | | | | | | | | | - -| | | | ASQA | | | ELI5 | | QAMPARI | | | -| --------- | --- | ----------- | ---- | --------- | ----------- | ---- | -------- | ------------ | ---------- | --- | -| ModelType | | ModelSize | | | | | | | | | -| | | Correctness | | Citation | Correctness | | Citation | Correctness | Citation | | -| | | EMRec. | Rec. | Prec. F1. | Claim | Rec. | Prec. F1 | Rec.-5 Prec. | Rec. Prec. | F1 | -Prompting-based -ChatGPT - 40.37 72.81 69.69 71.22 12.47 49.44 47.05 48.22 20.28 19.84 19.06 22.03 20.44 -7B 24.32 17.24 17.87 17.55 4.53 3.92 5.38 4.54 12.56 11.32 6.03 6.35 6.19 -LLaMA-2 13B 27.99 16.45 19.04 17.65 7.77 8.49 8.43 8.46 18.00 12.39 5.45 5.74 5.59 -70B 31.53 44.18 44.79 44.48 10.43 23.75 22.43 23.07 18.50 14.79 10.10 10.50 10.30 -7B 29.93 55.99 51.66 53.74 12.47 19.90 15.48 17.41 17.96 19.74 9.58 9.68 9.63 -LLaMA-2-Chat 13B 34.39 37.15 38.17 37.65 13.83 16.50 16.09 16.29 21.34 18.86 8.94 9.06 9.00 -70B 41.24 60.19 61.16 60.67 13.30 36.63 36.63 36.63 22.62 18.04 13.49 13.98 13.73 -Vicuna-v1.5 7B 38.34 48.37 44.63 46.42 12.30 29.81 22.45 25.61 14.22 14.74 11.26 11.64 11.45 -13B 35.20 51.92 53.40 52.65 14.33 31.15 28.99 30.03 22.06 19.60 13.04 13.74 13.38 -7B 29.46 23.12 25.45 24.23 8.47 16.04 16.32 16.18 16.96 15.98 7.50 7.76 7.63 -Mistral -8×7B 36.30 32.72 34.49 33.58 10.43 26.11 25.09 25.59 18.18 15.63 9.72 10.20 9.95 -7B 38.57 64.90 59.67 62.18 11.07 49.25 42.69 45.74 17.52 21.29 17.56 18.53 18.03 -Mistral-Instruct 8×7B 44.11 61.80 63.27 62.53 13.93 49.28 48.34 48.81 20.12 19.64 19.27 20.38 19.81 -Post-hocRetrieval -ChatGPT - 37.68 27.11 27.05 27.08 18.77 14.55 14.55 14.55 25.14 22.85 12.29 12.29 12.29 -LLaMA-2-Chat 70B 29.68 24.51 24.51 24.51 16.03 12.93 12.93 12.93 17.90 14.45 9.05 9.05 9.05 -Mistral-Instruct 8×7B 33.90 24.57 24.48 24.52 17.37 15.68 15.68 15.68 24.16 18.28 9.78 9.78 9.78 -Training-based -7B 29.96 67.82 66.97 67.39 6.90 22.34 32.40 26.45 2.34 1.98 10.53 18.80 13.50 -Self-RAG(LLaMA-2) -13B 31.66 71.26 70.35 70.80 6.07 30.46 40.20 34.66 1.90 1.33 12.79 20.90 15.86 -7B 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -VANILLA-SFT(LLaMA-2) 13B 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -FRONT(LLaMA-2) 7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -Table1: MainresultsontheALCEbenchmark. Boldnumbersindicatethebestperformance,while_indicatesthe -second-bestperformance. -substringsofthegeneration. RegardingtheELI5 Mistral-7B(Jiangetal.,2023)toMistral-8x7B- -dataset,correctnessismeasuredthroughclaimre- MoE (Jiang et al., 2024). Regarding SFT LLMs, -call (Claim), evaluating whether the model’s re- weselecttheSFTcounterpartsoftheopen-source -sponse entails the ground truth sub-claims. For foundationalLLMsweused. Detailedprompting -theQAMPARIdataset,correctnessisassessedus- settingscanbefoundinAppendixC. -| ingexactmatchprecision(Prec.) | | | andtop-5exact | | | | | | | | -| ----------------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | -match recall (Rec.-5) — considered 100% if the 4.3.2 Post-hocRetrievalMethods. -predictionincludesatleastfivecorrectanswers. -| | | | | | Following | Gao | et al. | (2023b), | we first | instruct | -| --- | --- | --- | --- | --- | --------- | --- | ------ | -------- | -------- | -------- | -LLMstoanswerthegivenqueryinaclosed-book -4.3 Baselines -| | | | | | setting, | and then | integrate | citations | in a post-hoc | | -| --- | --- | --- | --- | --- | -------- | -------- | --------- | --------- | ------------- | --- | -Wecompareourmethodwiththreetypesofbase- -| | | | | | manner. | Foreachgeneratedstatement,weemploy | | | | | -| ------ | ---------------- | -------- | ---------- | --- | ------- | ---------------------------------- | --- | --- | --- | --- | -| lines: | prompting-based, | post-hoc | retrieval, | and | | | | | | | -GTR(Nietal.,2022)toidentifyandcitethemost -training-based. -relevantdocumentfromthetop100retrieveddoc- -4.3.1 Prompting-basedMethods. uments. Weutilizethesamemodelsmentionedin -WedirectlypromptLLMsusingfew-shotdemon- prompting-basedsettingsforthisbaseline. -strations,eachconsistingofaquery,thetop5rel- -| | | | | | 4.3.3 | Training-basedMethods. | | | | | -| --- | --- | --- | --- | --- | ----- | ---------------------- | --- | --- | --- | --- | -evantretrieveddocuments,andananswerwithin- -| linecitations. | Ourexperimentsencompassarange | | | | | | | | | | -| -------------- | ----------------------------- | --- | --- | --- | -------- | --------------------------------- | --- | --- | --- | --- | -| | | | | | Self-RAG | (Asaietal.,2023)Self-RAGtrainsthe | | | | | -ofLLMs,fromfoundationalmodelstosupervised -| | | | | | LLM | to learn | to adaptively | retrieve | passages | on- | -| --------------------- | --- | -------------------- | --- | --- | --- | -------- | ------------- | -------- | -------- | --- | -| fine-tuning(SFT)LLMs. | | ForfoundationalLLMs, | | | | | | | | | -demandandenableittoreflectonitsgenerationto -| we select | GPT-3.5-Turbo5 | as the | representative | | | | | | | | -| --------- | -------------- | ------ | -------------- | --- | --- | --- | --- | --- | --- | --- | -furtherimprovegenerationqualityandattributions. -closed-sourcemodel,recognizedforitsnotableper- -| formance. | Among | the open-source | foundational | | | | | | | | -| --------- | ----- | --------------- | ------------ | --- | ----------- | --- | -------------------------- | --- | --- | --- | -| | | | | | VANILLA-SFT | | Wedirectlyemploysupervised | | | | -LLMs,wefocusontheLLaMA-2seriesincluding -fine-tuningtotraintheLLMonourgeneratedtrain- -LLaMA2-7B,LLaMA2-13B,andLLaMA2-70B, -| | | | | | ing data. | Given | a query | and corresponding | | doc- | -| ------------------------------------------------ | -------------- | ------------- | ----- | ---- | --------------------- | ------- | ------- | ----------------- | -------- | -------- | -| as well | as the Mistral | series, which | spans | from | | | | | | | -| | | | | | uments, | the LLM | is | required to | directly | generate | -| 5Specifically,weutilizegpt-3.5-turbo-1106version | | | | | answerswithcitations. | | | | | | - -4.4 ImplementDetails -| Weimplement | | FRONTwithdifferentsizesoffoun- | | | | | 60 | | | | Pre-filtered | | | -| ----------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --- | -Post-filtered -| dational | models | (LLaMA-2-7B | | and | LLaMA-2- | | | | | | | | | -| -------- | ------ | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -)%(1FnoitatiC 50 -| 13B)toevaluateitseffectiveness. | | | | Duringtheeval- | | | | | | | | | | -| ------------------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -40 -uation,FRONTutilizethesameretrievalsettingsas -| those outlined | | by Gao | et al. | (2023b). | Additional | | 30 | | | | | | | -| -------------- | ----------- | ------ | ---------- | -------- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| details | of training | and | evaluation | settings | | can be | 20 | | | | | | | -foundinAppendixD. -| | | | | | | | | ASQA | | ELI5 | | QAMPARI | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ------- | --- | -5 ResultsandAnalysis -| | | | | | | | Figure4: | AblationStudyonDataFiltering. | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----------------------------- | --- | --- | --- | --- | --- | -5.1 OverallResults -Simplysupervisedfine-tuningcanboostcitation -| | | | | | | | FRONT | demonstrates | | remarkable | | generaliza- | | -| ----------- | ------------------------------ | --- | -------------- | --- | ---------- | ----- | ---------- | ------------ | --- | ---------- | --- | ----------- | ------- | -| quality. | AsshowninTable1,teachingLLMsto | | | | | | | | | | | | | -| | | | | | | | tion. | Compared | to | the varied | | queries | and an- | -| generate | responses | | with citations | via | supervised | | | | | | | | | -| | | | | | | | swer types | present | in | the ALCE | | benchmark, | our | -| fine-tuning | significantly | | enhances | | citation | qual- | | | | | | | | -trainingdata,derivedexclusivelyfromtheAQuA- -ity,demonstratingsubstantialimprovementsover -MuSedataset(Kulkarnietal.,2020),exhibitsout- -bothprompt-basedandpost-hocretrievalbaselines -| | | | | | | | of-domain | characteristics. | | | Nonetheless, | | FRONT | -| ------ | ------------- | --- | ------------- | ---- | -------- | --- | ------------ | ---------------- | --- | -------- | ------------ | --- | --------- | -| across | all datasets. | | Specifically, | with | LLaMA-2- | | | | | | | | | -| | | | | | | | demonstrates | superior | | citation | quality, | | affirming | -7B, VANILLA-SFTledtosubstantialgainsincita- -itsexceptionalabilitytogeneralizeacrossdiverse -| tion F1 | scores | over | prompting: | ASQA | (17.55 | → | | | | | | | | -| ------- | ------ | ---- | ---------- | ---- | ------ | --- | -------------------------------- | --- | --- | --- | --- | ------------- | --- | -| | | | | | | | querytypesandretrievaldocuments. | | | | | Additionally, | | -65.61),ELI5(4.54→41.15),andQAMPARI(6.19 -| | | | | | | | while not | specifically | | optimized | | for correctness, | | -| -------- | ----------------------------------- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------------ | ---------------- | --- | -| →21.35). | Thesegainshighlighttheeffectiveness | | | | | | | | | | | | | -| | | | | | | | FRONT | also showcases | | modest | improvements | | in | -ofourtrainingdatagenerationpipeline. -thismetricoverVANILLA-SFTontheASQAand -| | | | | | | | QAMPARIdatasets. | | | However,FRONTencounters | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ----------------------- | --- | --- | --- | -FRONT achievessignificantperformancegains -lowerRec.-5ontheQAMPARIdataset,likelydue -| andsurpassesChatGPT. | | | | While VANILLA-SFT | | | | | | | | | | -| -------------------- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tothenatureofitsanswers,whichconsistofcon- -| demonstrates | | strong | performance, | | it still | shows | | | | | | | | -| ------------ | --- | ------ | ------------ | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | -catenatedentities,divergingsignificantlyfromour -| notable | discrepancies | | compared | to | leading | open- | | | | | | | | -| ------- | ------------- | --- | -------- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | -trainingdata. -sourceLLMs,suchasMixtral-8×7B-Instruct(e.g., -| 41.15 vs. | 45.74) | and | ChatGPT | (e.g., | 41.15 | vs. | | | | | | | | -| --------- | ------ | ---- | -------- | ----------- | ----- | -------- | ----------------- | --- | --- | --- | --- | --- | --- | -| 48.22) | on the | ELI5 | dataset. | FRONT | not | only | 5.2 AblationStudy | | | | | | | -| bridges | these | gaps | but also | establishes | | signifi- | | | | | | | | -Weconductablationstudiestoverifytheeffective- -| cant leads | across | all | datasets. | Specifically, | | us- | | | | | | | | -| --------------- | ------ | --- | --------- | --------------- | --- | ---- | ----------------------------------- | --- | --- | --- | --- | --- | ------ | -| | | | | | | | nessofdifferentcomponentsproposedin | | | | | | FRONT. | -| ing LLaMA-2-7B, | | | FRONT | comprehensively | | out- | | | | | | | | -performsChatGPT,achievingincreasesof3.32%, -| | | | | | | | EffectsofDataGenerationPipeline. | | | | | | Asillus- | -| ------- | --- | ------ | ----------- | ------- | --- | ------ | -------------------------------- | -------- | ------ | --- | -------- | ------ | -------- | -| 18.04%, | and | 21.28% | in citation | quality | | on the | | | | | | | | -| | | | | | | | trated | in §5.1, | simply | SFT | achieves | strong | per- | -ASQA,ELI5,andQAMPARIdatasetsrespectively. -| | | | | | | | formance, | underscoring | | the | high | quality | of our | -| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | --- | --- | ---- | ------- | ------ | -Thisperformanceunderscorestheeffectivenessof -| | | | | | | | training | data. | Furthermore, | | data | filtering, | a cru- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------ | --- | ---- | ---------- | ------ | -FRONTinenhancingattributioncapabilities. -| | | | | | | | cial component | | of our | data | generation | | pipeline, | -| -------------------------------------- | --- | --- | --- | --- | --- | --- | -------------- | ------- | ------- | -------- | ---------- | ------- | --------- | -| | | | | | | | plays a | pivotal | role in | ensuring | the | quality | of the | -| FRONTexhibitsscalabilitywithmodelsize. | | | | | | As | | | | | | | | -illustratedatthebottomofTable1,theperformance generated data by filtering out queries that yield -of FRONT in terms of citation quality shows no- non-informativeanswersorfailtomeetattribution -| | | | | | | | criteria. | To validate | | the effectiveness | | of | our data | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ----------------- | --- | --- | -------- | -tableimprovementswhenscalingfrom7Bto13B. -filteringstrategies,weconductedexperimentscom- -Specifically,weobserveimprovementsof3.23%in -ASQA,4.97%inELI5,and1.33%inQAMPARI. paringmodelsfine-tunedonbothpre-filteredand -This upward trend underscores the scalability of post-filtereddata. Theresults,depictedinFigure -4,confirmthatmodelstrainedonfiltereddataex- -FRONT withincreasingmodelsize,demonstrating -thepotentialofFRONTinleveragingtheincreased hibitanotableimprovementincitationqualityover -capabilitiesoflargerLLMsforfurtherperformance thosetrainedonunfiltereddata,achievingsuperior -attributionperformancewithreduceddatavolume. -gains. - -| | | | | ASQA | | | | ELI5 | | | QAMPARI | | | -| --- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | ------- | --- | --- | -Model -| | | | Correctness | | Citation | | Correctness | | Citation | Correctness | | Citation | | -| --- | --- | --- | ----------- | --- | -------- | --- | ----------- | --- | -------- | ----------- | --- | -------- | --- | -EMRec. Rec. Prec. F1. Claim Rec. Prec. F1 Rec.-5 Prec. Rec. Prec. F1 -FRONT-7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -SELF-GUIDE(w/oConsistency) 38.99 70.69 64.48 67.44 10.04 47.63 44.80 46.17 12.18 20.03 22.50 22.58 22.54 -VANILLA-SFT(w/oGround) 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -FRONT-13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -SELF-GUIDE(w/oConsistency) 40.99 73.08 68.13 70.52 10.06 50.68 49.78 50.23 13.94 22.38 23.73 23.99 23.85 -VANILLA-SFT(w/oGround) 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -Table2: AblationstudyontheimpactofdifferenttrainingstageswithintheALCEbenchmark. -44.42 -Citation-F1 -| | | | | 46.17 | | | | | 33 | | FRONT-13B | | | -| -------------- | --- | ----- | ----- | ----- | ----------- | --- | --- | ------------------- | --------------- | --- | --------- | ------------ | --- | -| | | | | | | | | | | | F | -7 B | | -| | | 43.71 | | | SELF-GUIDE | | | | | | R O | N T | | -| Citation-Prec. | | | | | | | | | LLaMA2-70B-Chat | | SEL F- G | U I D E -13B | | -| | | | 44.80 | | PROMPTGUIDE | | | ssenlufhtiaF 22..55 | | | | | | -SELF-GUIDE-7B -45.16 -| Citation-Rec. | | | | | | | | | | | Mixtral-8x7B-Inst | | | -| ------------- | --- | ----- | --- | ----- | ----- | --- | --- | --- | --- | ---------- | ----------------- | --- | --- | -| | | | | | 47.63 | | | | 22 | Vicuna-13B | | | | -| | | 44.00 | | 46.00 | 48.00 | | | | | | | | | -LLaMA2-7B-Chat -| Figure5: | Ablationstudyofdifferentgroundingguid- | | | | | | | 11..55 | | | | | | -| -------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -ancestrategiesontheELI5dataset. -| | | | | | | | | | 2200 | 3300 | 4400 5500 | 6600 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --------- | ---- | --- | -CitationF1 -EffectsofGroundingGuidedGeneration(G3). -G3 -empowers LLMs to first select relevant fine- Figure6: TherelationshipbetweencitationF1andhal- -grainedquotes,whichsubsequentlyguidethegen- lucination: Models positioned closer to the top-right -eration process. These quotes can provide fine- cornerexhibithighercitationqualityandalowerdegree -ofhallucination. -grainedsupervisionsignalsforattributedtextgen- -G3, -| eration. | To evaluate | | the effectiveness | | of | we | | | | | | | | -| --------------------------------------- | ----------- | --- | ----------------- | --------- | ---------- | ----- | ------- | -------- | ----------- | --- | --------- | --- | ------ | -| conduct | an ablation | | study | comparing | it against | | | | | | | | | -| | | | | | | | eration | process. | Experiments | | conducted | | on the | -| twovariantswithdistincttrainingrecipes. | | | | | | Given | | | | | | | | -ELI5datasetusingtheLLaMA-2-7Bmodelshow -that FRONTconsistsoftwostages,werefertothe that SELF-GUIDE outperforms PROMPT-GUIDE. -modeltrainedonlyduringthefirststage(without -ResultsdepictedinFigure5indicatethattraining -| consistency-awarealignment)asSELF-GUIDE. | | | | | | We | | | | | | | | -| ---------------------------------------- | ---------- | --- | ------- | --- | ----------- | --- | ---------- | --- | ------------- | --------- | -------- | -------------- | ------ | -| | | | | | | | models | to | self-generate | | grounded | quotes | before | -| firstcompare | SELF-GUIDE | | against | | VANILLA-SFT | | | | | | | | | -| | | | | | | | generating | | attributed | responses | is | more effective | | -(w/oGround),whichistrainedtodirectlygenerate thansimplyincorporatingthesegroundedquotes -responseswithcitations,bypassingthegrounding -intotheprompt. -| step. The | ablation | study, | detailed | | in Table | 2, re- | | | | | | | | -| --------- | -------- | ------ | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | -veals that models incorporating grounding guid- EffectsofConsistency-AwareAlignment(CCA). -ancesignificantlyoutperformtheir VANILLA-SFT The primary goal of CCA is to enhance the con- -counterpartsthatlacksuchgroundingmechanisms. sistency between grounded quotes and attributed -This highlights the crucial role of grounding in answers, thereby alleviating hallucinations and -| | | | | | | | achievingmorepreciseattribution. | | | | | Toevaluatethis, | | -| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | -bolsteringattribution. -Moreover, we explore an alternative variant we compare models that underwent only the G3 -of grounding guidance. Considering that SELF- stage(SELF-GUIDE)withthosefurtherenhanced -GUIDE leverages the model itself to both select through the CCA stage (FRONT). As illustrated -grounded quotes and generate attributed answers inTable2,FRONTsignificantlyimprovescitation -in an end-to-end paradigm, a natural variant in- qualityoverSELF-GUIDE,demonstratingtheeffec- -volves breaking down this task into two distinct tivenessoftheCCAstageinenhancingattribution. -stages. Inthisvariant,ChatGPTistaskedwithex- Furthermore,toassessCCA’simpactonreduc- -tractinggroundedquotes. Subsequently,aseparate inghallucinations,weutilizeQAFactEval(Fabbri -modelistrainedtoutilizethesegroundedquotes, etal.,2022),awidelyusedmetricforfactualcon- -along with the query and retrieval documents, to sistency, which scores the consistency of model -directly output the response and citations. This responses to given documents on a scale from 0 -variant,referredtoasPROMPT-GUIDED,integrates to5,withhigherscoresindicatinggreaterfaithful- -groundedquotesintotheprompttoguidethegen- ness. Specifically,weanalyzetheperformanceof - -leading open-source models and two variants of Authenticity Helpfulness -| FRONTand | SELF-GUIDE | | ontheELI5dataset. | | As | | | | | | | -| -------- | ---------- | --- | ----------------- | --- | --- | ------- | --- | --- | --- | ---- | ---- | -| | | | | | | ChatGPT | | | | 0.93 | 4.08 | -showninFigure6, FRONTproducesmorefaithful FRONT-7BonASQA 0.94 3.86 -outputsthanSELF-GUIDE,significantlyreducing FRONT-7BonELI5 0.92 3.96 -| hallucinations. | | | | | | FRONT-7BonQAMPARI | | | | 0.86 | 3.62 | -| --------------- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | ---- | ---- | -Table3: Humanevaluationonthequalityofgrounded -| EffectsofTrainingDataScale. | | | | Weanalyzethe | | | | | | | | -| --------------------------- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | -quotes. -| impact | of the | data scale | on model | performance | | | | | | | | -| ------ | ------------ | ---------- | -------- | ----------- | ------- | --- | --- | --- | --- | --- | --- | -| across | two training | stages. | In | particular, | we ran- | | | | | | | -tors. Theresultsofthehumanevaluationindicate -domlysampled2k,4k,6k,and8kinstancesfrom -thatbothquotesextractedbyChatGPTandthose -| our full | training | data | across two | distinct | training | | | | | | | -| -------- | -------- | ---- | ---------- | -------- | -------- | --- | --- | --- | --- | --- | --- | -stages. These subsets were then utilized to fine- generated by FRONT are of high quality, further -tunevarious7Bmodelvariants,enablingacompar- substantiatingtheeffectivenessofourmethod. -ativeanalysisofperformancebasedondatascale. -| Results | are shown | in | Figure | 7, which | indicates | | | | | | | -| ------- | --------- | --- | ------ | -------- | --------- | --- | --- | --- | --- | --- | --- | -7 Conclusion -thatincreasingdatasizeshowssignificantenhance- -mentsincitationquality,indicatingapositivecor- Inthiswork,wepresent FRONT,atwo-stagetrain- -relationbetweendatasizeandmodelperformance. ingframeworkdesignedtoequipLLMswithfine- -AsFRONTimplementsanautomatedprocedureca- -| | | | | | | grained | attribution | capabilities. | | FRONT | enables | -| --- | --- | --- | --- | --- | --- | ------- | ----------- | ------------- | --- | ----- | ------- | -pableofgeneratinghigh-qualityattributeddataand LLMstoinitiallyselectsupportingquotes,which -constructingcontrastivesupervisionfromweakand thenguidethegenerationprocess. Byfurtheren- -strongLLMs,itholdsthepotentialforcontinuous -| | | | | | | hancing | the consistency | | between | | the grounding | -| --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------- | --- | ------------- | -performanceimprovements. andgenerationprocessviapreferenceoptimization, -thesesupportingquotescanserveasfine-grained -6 HumanEvaluation -| | | | | | | citations. | Through | comprehensive | | | experiments, | -| --- | --- | --- | --- | --- | --- | ---------- | ------- | ------------- | --- | --- | ------------ | -Given the significant impact of the quality of FRONThasdemonstrateditsabilitytogeneratesu- -| | | | | | | perior grounded | | responses | | and highly | supportive | -| -------- | ------ | --- | ------------ | ------------ | --- | --------------- | ------------------------ | --------- | --- | ---------- | ---------- | -| grounded | quotes | on | fine-grained | verification | for | | | | | | | -| | | | | | | citations. | Furtheranalysisshowsthat | | | | FRONT sig- | -users,weconductedahumanevaluationtoassess -thequalityofgroundedquotesatdifferentstagesof nificantlyreduceshallucinationsandbenefitsuser -verification. -| ourframework: | | (1)QuotesextractedbyChatGPT | | | | | | | | | | -| ---------------------------------- | --------- | --------------------------- | ----------- | --- | ------------ | ------------ | --- | --- | --- | --- | --- | -| from50sampleddatapointsduringtheG3 | | | | | stage. | | | | | | | -| (2) Quotes | generated | | by FRONT-7B | | across three | 8 Limitation | | | | | | -datasets,with50datapointssampledfromeach. -Weengagedfourannotators,eachwithrelevant Our study presents several limitations worth not- -expertiseandholdingatleastabachelor’sdegree. ing. Firstly, the validation of our framework is -Thequalityofquoteswasevaluatedontwodimen- predominantly conducted on models of sizes 7B -sions: authenticityandhelpfulness. Authenticity and13B,leavingtheexplorationoflargermodels, -(abinaryscaleof0/1)referstowhetherthequotes suchasLLaMA-270Bduetocomputationalcon- -genuinelyoriginatefromthecorrespondingdocu- straints. Secondly,ourframeworkreliesonaprior -ments(quotesthatarehallucinatedormismatched retrievalprocess,whereinrelevantdocumentsare -withthecorrespondingdocumentIDareconsidered retrieved at one time. The incorporation of adap- -inauthentic). Helpfulness (5-point Likert scale) tiveretrieval,enablingmoredynamicinteractions -referstothedegreetowhichthequotesarebenefi- withLLMs,couldpotentiallyenhanceperformance. -cialinaddressingthequery. TheresultsinTable3 Weleaveitforfutureresearch. Lastly,evaluating -representtheaveragescoresforallquoteswithin the correctness of long-form question answering -eachmodelresponse,withtwoannotatorsevaluat- presents inherent challenges, leading our frame- -ingeachresponsetoensurereliability. work to primarily enhance citation quality, with -Furthermore, to evaluate the consistency of modest advancements in correctness. Therefore, -quotequalityannotations,wecomputedtheinter- we advocate for the development of more robust -annotator agreement using Fleiss’ Kappa coeffi- metricscapableofaccuratelyassessingthecorrect- -cient. The obtained Kappa coefficient of 0.82 in- nessoflong-formQAresponses, pavingtheway -| dicates | a high | level of | agreement | among | annota- | forfuturework. | | | | | | -| ------- | ------ | -------- | --------- | ----- | ------- | -------------- | --- | --- | --- | --- | --- | - -Acknowledgements -qa-basedfactualconsistencyevaluationforsumma- -| | | | | | | | rization. | InProceedingsofthe2022Conferenceof | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | ---------------------------------- | --- | --- | --- | -We appreciate Yuchun Fan for providing valu- theNorthAmericanChapteroftheAssociationfor -| | | | | | | | ComputationalLinguistics: | | | HumanLanguageTech- | | -| ----------------- | --- | --------- | --- | --- | ------- | -------- | ------------------------- | --- | --- | ------------------ | --- | -| able suggestions. | | Xiaocheng | | | Feng is | the cor- | | | | | | -nologies,NAACL2022,Seattle,WA,UnitedStates, -| responding | author | of | this | work. | We | thank the | | | | | | -| ---------- | ------ | --- | ---- | ----- | --- | --------- | --- | --- | --- | --- | --- | -July10-15,2022,pages2587–2601.Associationfor -| anonymous | reviewers | | for | their | insightful | com- | | | | | | -| --------- | --------- | --- | --- | ----- | ---------- | ---- | --- | --- | --- | --- | --- | -ComputationalLinguistics. -| ments. | This | work | was | supported | by | the Na- | | | | | | -| ------ | ---- | ---- | --- | --------- | --- | ------- | --- | --- | --- | --- | --- | -AngelaFan,YacineJernite,EthanPerez,DavidGrang- -| tional Key | R&D | Program | | of China | via | grant No. | | | | | | -| -------------- | --- | ------- | -------- | -------- | --------- | --------- | ---------- | -------- | ----------- | -------------- | ----- | -| | | | | | | | ier, Jason | Weston, | and Michael | Auli. 2019. | ELI5: | -| 2021ZD0112905, | | the | National | | Natural | Science | | | | | | -| | | | | | | | long form | question | answering. | In Proceedings | of | -| Foundation | of | China | (NSFC) | (grant | 62276078, | | | | | | | -the57thConferenceoftheAssociationforCompu- -U22B2059), the Key R&D Program of Hei- tationalLinguistics,ACL2019,Florence,Italy,July -longjiang via grant 2022ZX01A32, the Interna- 28-August2,2019,Volume1: LongPapers,pages -3558–3567.AssociationforComputationalLinguis- -tionalCooperationProjectofPCL,PCL2022D01 -tics. -andtheFundamentalResearchFundsfortheCen- -tralUniversities(GrantNo.HIT.OCEF.2023018). ZhangyinFeng,WeitaoMa,WeijiangYu,LeiHuang, -HaotianWang,QianglongChen,WeihuaPeng,Xi- -| | | | | | | | aochengFeng,BingQin,andTingLiu.2023. | | | | Trends | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | ------ | -inintegrationofknowledgeandlargelanguagemod- -References -els:Asurveyandtaxonomyofmethods,benchmarks, -CoRR,abs/2311.05876. -| Samuel | Joseph | Amouyal, | Ohad | Rubin, | | Ori Yoran, | andapplications. | | | | | -| -------- | --------- | -------- | --------- | ------- | -------------- | ---------- | ---------------- | ------ | ----------------- | -------- | --------- | -| Tomer | Wolfson, | Jonathan | | Herzig, | and | Jonathan | | | | | | -| | | | | | | | Luyu Gao, | Zhuyun | Dai, Panupong | Pasupat, | Anthony | -| Berant. | 2022. | QAMPARI: | | : | An open-domain | | | | | | | -| | | | | | | | Chen, | Arun | Tejasvi Chaganty, | Yicheng | Fan, Vin- | -| question | answering | | benchmark | | for questions | with | | | | | | -centY.Zhao,NiLao,HongraeLee,Da-ChengJuan, -| many | answers | from | multiple | paragraphs. | | CoRR, | | | | | | -| ---- | ------- | ---- | -------- | ----------- | --- | ----- | ---------- | ---- | ------ | ----------------- | --- | -| | | | | | | | and Kelvin | Guu. | 2023a. | RARR: researching | and | -abs/2205.12665. -revisingwhatlanguagemodelssay,usinglanguage -AkariAsai,ZeqiuWu,YizhongWang,AvirupSil,and models. InProceedingsofthe61stAnnualMeeting -oftheAssociationforComputationalLinguistics(Vol- -| Hannaneh | Hajishirzi. | | 2023. | Self-rag: | Learning | to | | | | | | -| -------- | ----------- | --- | ----- | --------- | -------- | --- | ----- | ----------------------------------- | --- | --- | --- | -| | | | | | | | ume1: | LongPapers),ACL2023,Toronto,Canada, | | | | -retrieve,generate,andcritiquethroughself-reflection. -July9-14,2023,pages16477–16508.Associationfor -CoRR,abs/2310.11511. -ComputationalLinguistics. -| Yuntao Bai, | Andy | Jones, | Kamal | Ndousse, | | Amanda | | | | | | -| ----------- | ---- | ------ | ----- | -------- | --- | ------ | --- | --- | --- | --- | --- | -TianyuGao,HowardYen,JiatongYu,andDanqiChen. -| Askell, | AnnaChen, | | NovaDasSarma, | | DawnDrain, | | | | | | | -| --------- | --------- | ---- | ------------- | --- | ---------- | --------- | ------------------ | ------------------------------------- | -------------------- | --- | --- | -| | | | | | | | 2023b. | Enablinglargelanguagemodelstogenerate | | | | -| Stanislav | Fort, | Deep | Ganguli, | | Tom | Henighan, | | | | | | -| | | | | | | | textwithcitations. | | CoRR,abs/2305.14627. | | | -NicholasJoseph,SauravKadavath,JacksonKernion, -TomConerly,SheerElShowk,NelsonElhage,Zac -| | | | | | | | Yunfan Gao, | Yun | Xiong, Xinyu | Gao, Kangxiang | Jia, | -| --------------- | --- | ----- | ---------- | --- | ------- | ----- | ----------- | --- | ------------ | -------------- | ---- | -| Hatfield-Dodds, | | Danny | Hernandez, | | Tristan | Hume, | | | | | | -JinliuPan,YuxiBi,YiDai,JiaweiSun,QianyuGuo, -ScottJohnston,ShaunaKravec,LianeLovitt,Neel Meng Wang, and Haofen Wang. 2023c. Retrieval- -| Nanda, | Catherine | Olsson, | | Dario | Amodei, | Tom B. | | | | | | -| -------- | --------- | ------- | --------------- | --------- | ------- | ---------- | ------------------------------------------ | -------------------- | --- | --- | --- | -| | | | | | | | augmentedgenerationforlargelanguagemodels: | | | | A | -| Brown, | Jack | Clark, | Sam McCandlish, | | Chris | Olah, | | | | | | -| | | | | | | | survey. | CoRR,abs/2312.10997. | | | | -| Benjamin | Mann, | and | Jared | Kaplan. | 2022. | Train- | | | | | | -| ing a | helpful | and | harmless | assistant | | with rein- | | | | | | -SylvainGugger,LysandreDebut,ThomasWolf,Philipp -forcement learning from human feedback. CoRR, Schmid,ZacharyMueller,SourabMangrulkar,Marc -abs/2204.05862. Sun,andBenjaminBossan.2022. Accelerate: Train- -ingandinferenceatscalemadesimple,efficientand -BerndBohnet,VinhQ.Tran,PatVerga,RoeeAharoni, -| | | | | | | | adaptable. | | https://github.com/huggingface/ | | | -| ------ | ------ | ----- | ------- | ------- | ----- | ------ | ---------- | --- | ------------------------------- | --- | --- | -| Daniel | Andor, | Livio | Baldini | Soares, | Jacob | Eisen- | | | | | | -accelerate. -stein,KuzmanGanchev,JonathanHerzig,KaiHui, -TomKwiatkowski,JiMa,JianmoNi,TalSchuster, OrHonovich, RoeeAharoni, JonathanHerzig, Hagai -WilliamW.Cohen,MichaelCollins,DipanjanDas, Taitelbaum,DoronKukliansy,VeredCohen,Thomas -Donald Metzler, Slav Petrov, and Kellie Webster. Scialom, Idan Szpektor, Avinatan Hassidim, and -| 2022. | Attributed | question | | answering: | | Evaluation | | | | | | -| ----- | ---------- | -------- | --- | ---------- | --- | ---------- | ----- | ------- | ----------- | ------------- | ------- | -| | | | | | | | Yossi | Matias. | 2022. TRUE: | re-evaluating | factual | -andmodelingforattributedlargelanguagemodels. -| | | | | | | | consistencyevaluation. | | InProceedingsofthe2022 | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | ---------------------- | --- | --- | -CoRR,abs/2212.08037. Conference of the North American Chapter of the -| | | | | | | | AssociationforComputationalLinguistics: | | | | Human | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | ----- | -CanyuChenandKaiShu.2023. Combatingmisinfor- LanguageTechnologies,NAACL2022,Seattle,WA, -mation in the age of llms: Opportunities and chal- UnitedStates, July10-15, 2022, pages3905–3920. -| lenges. | CoRR,abs/2311.05656. | | | | | | | | | | | -| ------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AssociationforComputationalLinguistics. -Alexander R. Fabbri, Chien-Sheng Wu, Wenhao Liu, LeiHuang,WeijiangYu,WeitaoMa,WeihongZhong, -and Caiming Xiong. 2022. Qafacteval: Improved Zhangyin Feng, Haotian Wang, Qianglong Chen, - -WeihuaPeng,XiaochengFeng,BingQin,andTing servingwithpagedattention. InProceedingsofthe -Liu. 2023. A survey on hallucination in large lan- ACMSIGOPS29thSymposiumonOperatingSystems -| guagemodels: | Principles,taxonomy,challenges,and | | | | | Principles. | | | | | | -| ---------------- | ---------------------------------- | ---- | -------- | -------- | --- | ----------- | ------------ | -------- | ----------------- | --- | ----- | -| openquestions. | CoRR,abs/2311.05232. | | | | | | | | | | | -| | | | | | | Patrick | S. H. Lewis, | Ethan | Perez, Aleksandra | | Pik- | -| | | | | | | tus, Fabio | Petroni, | Vladimir | Karpukhin, | | Naman | -| Ziwei Ji, Nayeon | Lee, | Rita | Frieske, | Tiezheng | Yu, | | | | | | | -DanSu,YanXu,EtsukoIshii,YejinBang,Andrea Goyal,HeinrichKüttler,MikeLewis,Wen-tauYih, -Madotto,andPascaleFung.2023. Surveyofhalluci- Tim Rocktäschel, Sebastian Riedel, and Douwe -nationinnaturallanguagegeneration. ACMComput. Kiela. 2020. Retrieval-augmented generation for -Surv.,55(12):248:1–248:38. knowledge-intensiveNLPtasks. InAdvancesinNeu- -| | | | | | | ralInformationProcessingSystems33: | | | | AnnualCon- | | -| --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ---------- | --- | -AlbertQ.Jiang,AlexandreSablayrolles,ArthurMen- -ferenceonNeuralInformationProcessingSystems -sch,ChrisBamford,DevendraSinghChaplot,Diego 2020,NeurIPS2020,December6-12,2020,virtual. -| de Las Casas, | Florian | Bressand, | | Gianna | Lengyel, | | | | | | | -| ------------- | ------- | --------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | -Guillaume Lample, Lucile Saulnier, Lélio Re- Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, -nard Lavaud, Marie-Anne Lachaux, Pierre Stock, Ziyang Chen, Baotian Hu, Aiguo Wu, and Min -TevenLeScao,ThibautLavril,ThomasWang,Timo- Zhang. 2023. A survey of large language models -| | | | | | | attribution. | CoRR,abs/2311.03731. | | | | | -| ----------------------------------- | --- | --- | --- | --- | ------- | ------------ | -------------------- | --- | --- | --- | --- | -| théeLacroix,andWilliamElSayed.2023. | | | | | Mistral | | | | | | | -7b. CoRR,abs/2310.06825. -| | | | | | | Nelson F. | Liu, Tianyi | Zhang, | and Percy | Liang. | 2023. | -| --- | --- | --- | --- | --- | --- | --------- | ----------- | ------ | --------- | ------ | ----- | -Albert Q. Jiang, Alexandre Sablayrolles, Antoine Evaluatingverifiabilityingenerativesearchengines. -Roux,ArthurMensch,BlancheSavary,ChrisBam- InFindingsoftheAssociationforComputationalLin- -ford,DevendraSinghChaplot,DiegodeLasCasas, guistics: EMNLP2023,Singapore,December6-10, -Emma Bou Hanna, Florian Bressand, Gianna 2023,pages7001–7025.AssociationforComputa- -| Lengyel, | Guillaume | Bour, | Guillaume | | Lample, | tionalLinguistics. | | | | | | -| ------------ | --------- | ------ | --------- | --------- | ------- | ------------------ | --- | --- | --- | --- | --- | -| Lélio Renard | Lavaud, | Lucile | | Saulnier, | Marie- | | | | | | | -AnneLachaux,PierreStock,SandeepSubramanian, JacobMenick,MajaTrebacz,VladimirMikulik,John -Sophia Yang, Szymon Antoniak, Teven Le Scao, Aslanides, H. Francis Song, Martin J. Chadwick, -Théophile Gervet, Thibaut Lavril, Thomas Wang, Mia Glaese, Susannah Young, Lucy Campbell- -| | | | | | | Gillingham, | Geoffrey | | Irving, and Nat | McAleese. | | -| --------------------------------------- | --- | --- | --- | --- | ---- | ----------- | -------- | --- | --------------- | --------- | --- | -| TimothéeLacroix,andWilliamElSayed.2024. | | | | | Mix- | | | | | | | -2022. Teachinglanguagemodelstosupportanswers -| tralofexperts. | CoRR,abs/2401.04088. | | | | | | | | | | | -| --------------- | -------------------- | ------- | ----- | ------ | ------ | ------------------- | --- | -------------------- | --- | --- | --- | -| | | | | | | withverifiedquotes. | | CoRR,abs/2203.11147. | | | | -| Ehsan Kamalloo, | Aref | Jafari, | Xinyu | Zhang, | Nandan | | | | | | | -Thakur,andJimmyLin.2023. HAGRID:Ahuman- ReiichiroNakano,JacobHilton,SuchirBalaji,JeffWu, -llmcollaborativedatasetforgenerativeinformation- Long Ouyang, Christina Kim, Christopher Hesse, -seekingwithattribution. CoRR,abs/2307.16883. ShantanuJain,VineetKosaraju,WilliamSaunders, -| | | | | | | Xu Jiang, | Karl | Cobbe, | Tyna Eloundou, | Gretchen | | -| --- | --- | --- | --- | --- | --- | --------- | ---- | ------ | -------------- | -------- | --- | -Krueger,KevinButton,MatthewKnight,Benjamin -VladimirKarpukhin,BarlasOguz,SewonMin,Patrick -S.H.Lewis,LedellWu,SergeyEdunov,DanqiChen, Chess,andJohnSchulman.2021. Webgpt: Browser- -andWen-tauYih.2020. Densepassageretrievalfor assisted question-answering with human feedback. -open-domainquestionanswering. InProceedingsof CoRR,abs/2112.09332. -the2020ConferenceonEmpiricalMethodsinNat- -| | | | | | | Jianmo | Ni, Chen | Qu, Jing | Lu, Zhuyun | Dai, | Gus- | -| ------------- | ----------- | --- | ----- | ----- | ------- | ------ | -------- | -------- | ---------- | ---- | ---- | -| ural Language | Processing, | | EMNLP | 2020, | Online, | | | | | | | -November16-20,2020,pages6769–6781.Associa- tavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, -tionforComputationalLinguistics. YiLuan,KeithB.Hall,Ming-WeiChang,andYinfei -| | | | | | | Yang.2022. | Largedualencodersaregeneralizable | | | | | -| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------- | --- | --- | --- | --- | -SayaliKulkarni,SheideChammas,WanZhu,FeiSha, retrievers. InProceedingsofthe2022Conferenceon -and Eugene Ie. 2020. Aquamuse: Automatically EmpiricalMethodsinNaturalLanguageProcessing, -generatingdatasetsforquery-basedmulti-document EMNLP2022,AbuDhabi,UnitedArabEmirates,De- -summarization. CoRR,abs/2010.12694. cember7-11,2022,pages9844–9855.Association -forComputationalLinguistics. -| TomKwiatkowski, | JennimariaPalomaki, | | | | OliviaRed- | | | | | | | -| --------------- | ------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | -field,MichaelCollins,AnkurP.Parikh,ChrisAlberti, OpenAI. 2023. GPT-4 technical report. CoRR, -| DanielleEpstein,IlliaPolosukhin,JacobDevlin,Ken- | | | | | | abs/2303.08774. | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -tonLee,KristinaToutanova,LlionJones,Matthew -LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, -| Kelcey, Ming-Wei | | Chang, | Andrew | M. | Dai, Jakob | | | | | | | -| ---------------- | -------- | -------- | ------- | ----- | ---------- | ------- | -------------- | --- | --------------- | --- | ----- | -| | | | | | | Carroll | L. Wainwright, | | Pamela Mishkin, | | Chong | -| Uszkoreit, | Quoc Le, | and Slav | Petrov. | 2019. | Natu- | | | | | | | -ralquestions: abenchmarkforquestionanswering Zhang,SandhiniAgarwal,KatarinaSlama,AlexRay, -research. Trans.Assoc.Comput.Linguistics,7:452– JohnSchulman,JacobHilton,FraserKelton,Luke -| 466. | | | | | | Miller,MaddieSimens,AmandaAskell,PeterWelin- | | | | | | -| ---- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | -der,PaulF.Christiano,JanLeike,andRyanLowe. -Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying 2022. Training languagemodelsto followinstruc- -Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. tionswithhumanfeedback. InAdvancesinNeural -Gonzalez, Hao Zhang, and Ion Stoica. 2023. Effi- InformationProcessingSystems35: AnnualConfer- -cientmemorymanagementforlargelanguagemodel enceonNeuralInformationProcessingSystems2022, - -NeurIPS2022,NewOrleans,LA,USA,November28 Doshi,RenelitoDelosSantos,TojuDuke,JohnnySo- -| -December9,2022. | | | | | | | raker,BenZevenbergen,VinodkumarPrabhakaran, | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | ----------- | --- | ------- | ----------- | -| | | | | | | | Mark Diaz, | Ben | Hutchinson, | | Kristen | Olson, Ale- | -AleksandraPiktus,FabioPetroni,VladimirKarpukhin, jandraMolina,ErinHoffman-John,JoshLee,Lora -DmytroOkhonko,SamuelBroscheit,GautierIzacard, -| | | | | | | | Aroyo, | Ravi Rajakumar, | | Alena | Butryna, | Matthew | -| --- | --- | --- | --- | --- | --- | --- | ------ | --------------- | --- | ----- | -------- | ------- | -Patrick S. H. Lewis, Barlas Oguz, Edouard Grave, Lamm,ViktoriyaKuzmina,JoeFenton,AaronCo- -Wen-tauYih,andSebastianRiedel.2021. Theweb hen,RachelBernstein,RayKurzweil,BlaiseAgüera -isyouroyster-knowledge-intensiveNLPagainsta yArcas,ClaireCui,MarianCroak,EdH.Chi,and -verylargewebcorpus. CoRR,abs/2112.09924. QuocLe.2022. Lamda: Languagemodelsfordialog -CoRR,abs/2201.08239. -applications. -KrishnaPillutla,SwabhaSwayamdipta,RowanZellers, -JohnThickstun,SeanWelleck,YejinChoi,andZaïd -| | | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | -Harchaoui.2021. MAUVE:measuringthegapbe- bert, Amjad Almahairi, Yasmine Babaei, Nikolay -tweenneuraltextandhumantextusingdivergence Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti -| frontiers. | In | Advances | in Neural | Information | | Pro- | | | | | | | -| ---------- | --- | -------- | --------- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | -Bhosale,DanBikel,LukasBlecher,CristianCanton- -| cessingSystems34: | | AnnualConferenceonNeural | | | | | | | | | | | -| ----------------- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Ferrer,MoyaChen,GuillemCucurull,DavidEsiobu, -InformationProcessingSystems2021,NeurIPS2021, -JudeFernandes,JeremyFu,WenyinFu,BrianFuller, -December6-14,2021,virtual,pages4816–4828. -CynthiaGao,VedanujGoswami,NamanGoyal,An- -thonyHartshorn,SagharHosseini,RuiHou,Hakan -RonakPradeep,SahelSharifymoghaddam,andJimmy Inan,MarcinKardas,ViktorKerkez,MadianKhabsa, -| Lin.2023. | Rankvicuna: | | Zero-shotlistwisedocument | | | | | | | | | | -| --------- | ----------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -IsabelKloumann,ArtemKorenev,PunitSinghKoura, -rerankingwithopen-sourcelargelanguagemodels. -Marie-AnneLachaux,ThibautLavril,JenyaLee,Di- -CoRR,abs/2309.15088. -anaLiskovich,YinghaiLu,YuningMao,XavierMar- -tinet,TodorMihaylov,PushkarMishra,IgorMoly- -| Rafael Rafailov, | | Archit | Sharma, | Eric | Mitchell, | Ste- | | | | | | | -| ---------------- | --- | ------ | ------- | ---- | --------- | ---- | ---------- | ---- | ------ | -------- | ------ | ------- | -| | | | | | | | bog, Yixin | Nie, | Andrew | Poulton, | Jeremy | Reizen- | -fanoErmon,ChristopherD.Manning,andChelsea -stein,RashiRungta,KalyanSaladi,AlanSchelten, -| Finn. | 2023. | Direct preference | | optimization: | | Your | | | | | | | -| ------------------------------------ | ----- | ----------------- | --- | ------------- | --- | ----- | -------------- | ----- | ------- | ------ | ------ | --------- | -| | | | | | | | Ruan Silva, | Eric | Michael | Smith, | Ranjan | Subrama- | -| languagemodelissecretlyarewardmodel. | | | | | | CoRR, | | | | | | | -| | | | | | | | nian, Xiaoqing | Ellen | Tan, | Binh | Tang, | Ross Tay- | -abs/2305.18290. -| | | | | | | | lor, Adina | Williams, | Jian | Xiang | Kuan, | Puxin Xu, | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | ---- | ----- | ----- | --------- | -ZhengYan,IliyanZarov,YuchenZhang,AngelaFan, -| Jeff Rasley, | Samyam | Rajbhandari, | | Olatunji | | Ruwase, | | | | | | | -| ------------ | ------ | ------------ | --- | -------- | --- | ------- | --- | --- | --- | --- | --- | --- | -and Yuxiong He. 2020. Deepspeed: System opti- Melanie Kambadur, Sharan Narang, Aurélien Ro- -driguez,RobertStojnic,SergeyEdunov,andThomas -mizationsenabletrainingdeeplearningmodelswith -| | | | | | | | Scialom.2023. | Llama2: | | Openfoundationandfine- | | | -| ------------------------- | --- | --- | --- | --------- | ------- | --- | ---------------- | ------- | -------------------- | ---------------------- | --- | --- | -| over100billionparameters. | | | | InKDD’20: | The26th | | | | | | | | -| | | | | | | | tunedchatmodels. | | CoRR,abs/2307.09288. | | | | -ACMSIGKDDConferenceonKnowledgeDiscovery -| and Data | Mining, | Virtual | Event, | CA, | USA, | August | | | | | | | -| -------- | ------- | ------- | ------ | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | -23-27,2020,pages3505–3506.ACM. TheodoraWorledge,JudyHanwenShen,NicoleMeis- -| | | | | | | | ter,CalebWinston,andCarlosGuestrin.2023. | | | | | Uni- | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ---- | -fyingcorroborativeandcontributiveattributionsin -| Freda Shi, | Xinyun | Chen, | Kanishka | Misra, | | Nathan | | | | | | | -| ---------- | ------ | ----- | -------- | ------ | --- | ------ | -------------------- | --- | --- | -------------------- | --- | --- | -| | | | | | | | largelanguagemodels. | | | CoRR,abs/2311.12233. | | | -Scales,DavidDohan,EdH.Chi,NathanaelSchärli, -| andDennyZhou.2023. | | | Largelanguagemodelscan | | | | | | | | | | -| ------------------ | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -beeasilydistractedbyirrelevantcontext. InInterna- Fangyuan Xu, Weijia Shi, and Eunsol Choi. 2023a. -tionalConferenceonMachineLearning,ICML2023, RECOMP:improvingretrieval-augmentedlmswith -23-29 July 2023, Honolulu, Hawaii, USA, volume compression and selective augmentation. CoRR, -abs/2310.04408. -202ofProceedingsofMachineLearningResearch, -pages31210–31227.PMLR. -ShichengXu,LiangPang,HuaweiShen,XueqiCheng, -IvanStelmakh,YiLuan,BhuwanDhingra,andMing- andTat-SengChua.2023b. Search-in-the-chain: To- -Wei Chang. 2022. ASQA: factoid questions meet wards the accurate, credible and traceable content -long-formanswers. InProceedingsofthe2022Con- generation for complex knowledge-intensive tasks. -CoRR,abs/2304.14732. -ferenceonEmpiricalMethodsinNaturalLanguage -Processing,EMNLP2022,AbuDhabi,UnitedArab -Emirates,December7-11,2022,pages8273–8288. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan -AssociationforComputationalLinguistics. Berant. 2023. Making retrieval-augmented lan- -| | | | | | | | guage models | robust | to | irrelevant | context. | CoRR, | -| ----- | ---------- | ------ | ------------- | -------- | -------- | ----- | --------------- | ------ | --- | ---------- | -------- | ----- | -| Romal | Thoppilan, | Daniel | De | Freitas, | Jamie | Hall, | abs/2310.01558. | | | | | | -| Noam | Shazeer, | Apoorv | Kulshreshtha, | | Heng-Tze | | | | | | | | -Cheng,AliciaJin,TaylorBos,LeslieBaker,YuDu, Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, -YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Xiaolei Wang, Yupeng Hou, Yingqian Min, Be- -AminGhafouri,MarceloMenegali,YanpingHuang, ichenZhang,JunjieZhang,ZicanDong,YifanDu, -MaximKrikun,DmitryLepikhin,JamesQin,Dehao Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao -Chen,YuanzhongXu,ZhifengChen,AdamRoberts, Jiang,RuiyangRen,YifanLi,XinyuTang,Zikang -MaartenBosma,YanqiZhou,Chung-ChingChang, Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. -IgorKrivokon,WillRusch,MarcPickett,KathleenS. 2023. A survey of large language models. CoRR, -| Meier-Hellstern, | | Meredith | Ringel | Morris, | | Tulsee | abs/2303.18223. | | | | | | -| ---------------- | --- | -------- | ------ | ------- | --- | ------ | --------------- | --- | --- | --- | --- | --- | - -Kun Zhu, Xiaocheng Feng, Xiyuan Du, Yuxuan Gu, process, we retained 5,667 and 2,431 queries, re- -WeijiangYu,HaotianWang,QianglongChen,Zheng spectively. Additionally,wecalculatedtheaverage -| Chu, JingchangChen, | | andBingQin.2024. | | Anin- | | | | | | | -| ------------------- | --- | ---------------- | --- | ----- | --- | --- | --- | --- | --- | --- | -lengthofanswersandtheaveragenumberofcita- -formationbottleneckperspectiveforeffectivenoise -tionsgeneratedforvarioustypesofquerieswithin -filteringonretrieval-augmentedgeneration. -ourdataset,asshowninTable4. -YutaoZhu,HuayingYuan,ShutingWang,JiongnanLiu, -| Wenhan | Liu, Chenlong | Deng, | Zhicheng | Dou, and | | | | | | | -| ------ | ------------- | ----- | -------- | -------- | --- | --- | --- | --- | --- | --- | -A.2 DetailsofDataFiltering -| Ji-RongWen.2023. | | Largelanguagemodelsforinfor- | | | | | | | | | -| ---------------- | --- | ---------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | -| mationretrieval: | | Asurvey. | CoRR,abs/2308.07107. | | | | | | | | -WetrainedourAttributedDiscriminatorusingthe -| | | | | | manually | annotated | data | provided | by | Liu et al. | -| --- | --- | --- | --- | --- | -------- | --------- | ---- | -------- | --- | ---------- | -A DetailsofDataGenerationPipeline -| | | | | | (2023), | which is | sampled | from | real generative | | -| --- | --- | --- | --- | --- | ------- | -------- | ------- | ---- | --------------- | --- | -A.1 DataStatistic searchengines. Eachstatementanditsciteddocu- -menthavebeenmeticulouslyannotatedforattribu- -#Questions 8,098 tion,categorizedintothreetypes: completesupport, -| | | | | | partialsupport,andnosupport. | | | | Fortraining,weuti- | | -| ------------ | --- | --- | --- | ---- | ---------------------------- | --- | --- | --- | ------------------ | --- | -| ➥#LongAnswer | | | | 5667 | | | | | | | -lizedadatasetof8,834instances,comprising6,415 -| ➥#ShortAnswer | | | | 2431 | | | | | | | -| ------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | -instancesofcompletesupport,1,552ofpartialsup- -Avg. WordsperAnswer 50.48 port, and 867 of no support. The discriminator -initializedwithLLaMA-2-7B,wastrainedwitha -| ➥Avg. | WordsperLongAnswer | | | 69.15 | | | | | | | -| ----- | ------------------- | --- | --- | ----- | --------------------------- | --- | --- | --- | -------------- | --- | -| | | | | | maximumsequencelengthof512. | | | | Wetraineditfor | | -| ➥Avg. | WordsperShortAnswer | | | 6.94 | | | | | | | -3epochs,withatotalbatchsizeof128,andapeak -Avg. CitationperAnswer 4.40 learning rate of 2e-5, incorporating 3% warmup -steps,followedbyalineardecay. -| ➥Avg. | CitationperLongAnswer | | | 4.68 | | | | | | | -| ----- | ---------------------- | --- | --- | ---- | ------ | ----------------- | --------- | --------- | ---------- | ----- | -| ➥Avg. | | | | | During | the data | filtering | stage, | we first | break | -| | CitationperShortAnswer | | | 3.77 | | | | | | | -| | | | | | down | the automatically | | generated | attributed | an- | -swersintostatementformandusethetraineddis- -| Table 4: | The statistics | of the | data generated | by our | | | | | | | -| -------- | -------------- | ------ | -------------- | ------ | --- | --- | --- | --- | --- | --- | -criminatortoannotatetheattributionbetweeneach -automaticdatagenerationpipeline. -| | | | | | statement | and its | cited | documents. | Specifically, | | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ---------- | ------------- | --- | -weassigndifferentattributionscorestoeachstate- -Table4presentsthestatisticsofthedataautomat- -| | | | | | ment s | based on | its attribution | | relationship | with | -| ------------------------------------------- | --- | --- | --- | --- | ----------------------------------- | -------- | --------------- | --- | ------------ | ---- | -| icallygeneratedbyourdatagenerationpipeline. | | | | In | | | | | | | -| | | | | | citeddocumentsd,asshowninEquation7. | | | | | Con- | -total,wecollected8,098questionsfromtheNatural -sequently,foreachattributedanswer,wecancal- -Questions(NQ)dataset,ofwhich5,667questions -| | | | | | culateitsaverageattributionscore. | | | | Attributedan- | | -| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ------------- | --- | -weregatheredfromthosewithlong-formanswers, -swerswithanaverageattributionscorebelow0.8 -and2,431questionswerecollectedfromthosewith -| | | | | | are filtered | out. | The threshold | | of 0.8 was | deter- | -| --- | --- | --- | --- | --- | ------------ | ---- | ------------- | --- | ---------- | ------ | -short-formfactoidanswers. -minedthroughpreliminarytestingonthedevelop- -Forquestionsrequiringlong-formanswers,we -| | | | | | ment set, | for which | we | manually | annotated | 100 | -| --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --------- | --- | -initializedourquerysourcewiththeAQUAMUSE -samplestoensuretheeffectivenessofourfiltering -| dataset (Kulkarni | | et al., 2020), | which consists | of | | | | | | | -| ----------------- | --- | -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | -criteria. -high-qualityqueriesspecificallydesignedforlong- -formresponseswithintheNQdataset,recognized -| as“good”bythemajorityofNQevaluators. | | | | Inthis | | | | | | | -| ------------------------------------ | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | - -| | | | | | | 1, Dis(s,d) | | = complete | support | | -| -------------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | -| way,utilizingarefinedandsuperiorqualityquery | | | | | |   | | | | | -setlaidarobustgroundworkforourtrainingdata r(s)= 0.5,Dis(s,d) = partial support - -generation,streamliningthedatafilteringprocess. 0, Dis(s,d) = no support -| Forfactoidqueriesthatnecessitateshort-forman- | | | | | | | | | | (7) | -| --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -swers,wedirectlysampledfromtheoriginalNQ -| dataset, leveraging | | its abundance | and | inherently | | | | | | | -| ------------------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | -B DetailsofEvaluationMetrics -highquality. -During the data generation process, our initial In addition to evaluating citation quality and cor- -querysetcomprised7,725queriesrequiringlong- rectness,theALCEbenchmarkincludesabroader -formanswersand4,000queriesnecessitatingshort- setofdimensions,suchasfluency,ROUGE-L,and -| form answers. | | After a two-stage | data | filtering | generationlength. | | | | | | -| ------------- | --- | ----------------- | ---- | --------- | ----------------- | --- | --- | --- | --- | --- | - -Fluency Weevaluatethefluencyofthegenerated framework(Kwonetal.,2023)forefficientinfer- -responseusingMAUVE(Pillutlaetal.,2021). No- ence. Thehyperparametersaresetasillustratedin -| tably,wecalculatefluencyonlyfortheASQAand | | | | | | | Table9. | | | | | | | -| ----------------------------------------- | --- | -------- | ------ | -------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | -| ELI5 datasets, | | omitting | it for | QAMPARI, | | as the | | | | | | | | -response in QAMPARI typically consists of lists D.2 RetrievalSettings -| ofshortanswers. | | ArelativelyhighMAUVEscore | | | | | | | | | | | | -| --------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Duringtheevaluation,weadoptthesameretrieval -indicatesthatthegenerationissufficientlyfluent. -| | | | | | | | settings | as specified | | by Gao | et | al. (2023b). | For | -| ------- | --------------------------------- | --- | --- | -------- | --- | ------ | -------------------------------- | ------------ | ------- | ------ | --------- | ------------ | ------- | -| | | | | | | | the ASQA | and | QAMPARI | | datasets, | we | use the | -| ROUGE-L | Inadditiontoevaluatingthecorrect- | | | | | | | | | | | | | -| | | | | | | | denseretrieverGTR(Nietal.,2022). | | | | | FortheELI5 | | -| ness of | the model-generated | | | content, | we | employ | | | | | | | | -dataset,weemploythesparseretrieverBM25. -ROUGE-Ltoassesstheoverallqualityandtextual -coherenceoftheresponses. -E MoredetailaboutAblationStudy -C Prompts -E.1 TheEffectofTrainingDataScale. -C.1 PromptsforPrompting-basedMethods Weexaminehowmodelperformancevarieswith -| | | | | | | | changesindatascale,asdepictedinFigure7. | | | | | | The | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -FollowingGaoetal.(2023b),weadoptthevanilla -| | | | | | | | upper part | of | the figure | illustrates | | the | impact of | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | ----------- | --- | --- | --------- | -promptingstrategyforitssimplicityandeffective- -| | | | | | | | the training | data | scale | on | citation | quality | during | -| ------------------------------------ | ------ | -------- | -------- | -------- | --------- | --------- | ------------- | ---- | ------- | ---------- | ---------- | -------- | --------- | -| ness. Specifically, | | the | prompts | vary | according | to | | | | | | | | -| | | | | | | | the Grounding | | Guided | Generation | | training | stage, | -| thetypeofdatawithintheALCEbenchmark. | | | | | | For | | | | | | | | -| | | | | | | | with datasets | | ASQA, | ELI5, | and | QAMPARI | rep- | -| long-form | QA | datasets | such | as ASQA | | and ELI5, | | | | | | | | -| | | | | | | | resented | from | left to | right. | Similarly, | | the lower | -| the prompt | format | is | detailed | in Table | 5. | For the | | | | | | | | -partofthefiguredescribestheinfluenceduringthe -| short-form | QA | dataset | QAMPARI, | | the | format is | | | | | | | | -| ---------- | --- | ------- | -------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -Consistency-AwareAlignmenttrainingstage. -outlinedinTable6. -E.2 TheGeneralizationAcrossModel -| C.2 Instructionsfor | | | FRONT | | | | | | | | | | | -| ------------------- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Architectures. -Duringthetrainingprocess,wefollowtheinstruc- -| | | | | | | | FRONT | demonstrates | | exceptional | | generalization | | -| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | --- | ----------- | --- | -------------- | --- | -Alpaca6. -| tion format | of | | Specifically, | | we | employ | | | | | | | | -| ----------- | --- | --- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -capabilitiesacrossvariousfoundationalmodelar- -variedinstructionsfordifferentquestiontypes,as -| | | | | | | | chitectures. | Specifically,transitioningthefounda- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------------------------ | --- | --- | --- | --- | --- | -delineatedinTable7forlong-formquestionsand -| | | | | | | | tional model | | from LLaMA-2-7B | | | to the | stronger | -| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --- | ------ | -------- | -Table8forshort-formquestions. -| | | | | | | | foundational | | model, | Mistral-7B, | | results | in even | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ----------- | ------ | ------------ | --- | ------- | -------- | -| | | | | | | | greater | performance | | enhancements | | as | shown in | -D ExperimentalDetails -| | | | | | | | Figure8. | Thisfurtherunderscoresthebroadappli- | | | | | | -| --------------------- | --- | --- | ----- | --- | --- | --- | ----------------------------- | ------------------------------------ | --- | --- | ------ | --- | --- | -| | | | | | | | cabilityandgeneralizabilityof | | | | FRONT. | | | -| D.1 TrainingDetailsof | | | FRONT | | | | | | | | | | | -Thetrainingofallmodelsisexecutedon4Nvidia E.3 Theeffectofβ inConsistency-Aware -A100GPUs,eachwith80GBofmemory,leverag- AlignmentTrainingStage -ingtheDeepspeed(Rasleyetal.,2020)andHug- -| | | | | | | | In the | Consistency-Aware | | | Alignment | | Training | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------- | --- | --- | --------- | --- | -------- | -gingFaceAcceleratelibraries(Guggeretal.,2022) -| | | | | | | | Stage, | the β | parameter | in | Direct | Preference | Op- | -| -------------------------------------- | --- | --- | --- | --- | --- | ----- | ---------- | ----- | --------- | --- | ------ | ---------- | ------ | -| toconductmulti-GPUdistributedtraining. | | | | | | Given | | | | | | | | -| | | | | | | | timization | (DPO) | controls | | the | strength | of the | -thelongnatureoftheinputs,themaximumtoken -Kullback-Leiblerpenalty,typicallysetwithinthe -lengthissetto2,048tokens. -| | | | | | | | range of | 0.1 to | 0.5. | A higher | β | value | indicates a | -| ------ | --- | --------- | ----- | ---------- | --- | ------ | -------- | ------ | ---- | -------- | --- | ----- | ----------- | -| During | the | grounding | guide | generation | | stage, | | | | | | | | -preferenceforthepolicymodel’strainingprocess -modelsaretrainedfor5epochswithatotalbatch -toremainclosertotheinitiallyreferencedmodel. -| size of | 128, a peak | learning | | rate of | 2e-5 | with 3% | | | | | | | | -| -------------------------------------- | ----------- | -------- | --- | ------- | ---- | ------- | ----------------------------------- | ------ | --- | --- | ----- | ------ | -------- | -| | | | | | | | In extreme | cases, | as | β → | 0, we | ignore | the con- | -| warmupstepsfollowedbyalineardecay. | | | | | | During | | | | | | | | -| | | | | | | | straintsimposedbythereferencemodel. | | | | | | Thisset- | -| thecontrastivealignmentstage,wesettheβ | | | | | | to0.1 | | | | | | | | -tingaimstobalancethemodel’sabilitytoadaptto -andcontinuedtrainingfortwoadditionalepochs. -newtrainingsignalswhilemaintainingthestability -| Specifically, | During | | inference, | we | use | the vllm | | | | | | | | -| ------------- | ------ | --- | ---------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -ofthelearnedbehaviorsfromthereferencemodel. -| | | | | | | | Subsequently, | | we | trained | five | variants | by ad- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | ------- | ---- | -------- | ------ | -6https://github.com/tatsu-lab/stanford_alpaca/ -| tree/main | | | | | | | justingβ | from0.1to0.5onthemodelpreviously | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --- | - -Instruction: Write an accurate, engaging, and concise answer for the given question using only -the provided search results (some of which might be irrelevant) and cite them properly. Use an -unbiased and journalistic tone. Always cite for any factual claim. When citing several search -results, use [1][2][3]. Cite at least one document and at most three documents in each sentence. -If multiple documents support the sentence, only cite a minimum sufficient subset of the -documents. -Table5: PromptforLong-formQA. -Instruction: Provide a list of accurate answers for the given question using only the provided -search results (some of which might be irrelevant) and cite them properly. Always cite one and -only one document for each answer. Separate answers by commas. For questions that have more than -| 5 answers, | write | at least 5 answers. | | | | | -| ---------- | ----- | ------------------- | --- | --- | --- | --- | -Table6: PromptforShort-formQA. -| 72.0 | | | 48.0 | | 23.0 | | -| ---- | --- | ----- | ---- | ----- | ---- | ----- | -| 68.0 | | | 46.0 | | 22.0 | | -| 64.0 | | | 44.0 | | 21.0 | | -| 60.0 | | REC. | 42.0 | REC. | 20.0 | REC. | -| 56.0 | | PREC. | 40.0 | PREC. | 19.0 | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| ---- | --- | -------- | ---- | -------- | ---- | -------- | -| 76.0 | | | 54.0 | | 25.0 | | -| 72.0 | | | 52.0 | | 24.0 | | -| 68.0 | | | 50.0 | | 23.0 | | -| 64.0 | | | 48.0 | | 22.0 | | -| | | REC. | | REC. | | REC. | -| 60.0 | | | 46.0 | | 21.0 | | -| | | PREC. | | PREC. | | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| --- | --- | -------- | --- | -------- | --- | -------- | -Figure7: Ablationstudyonsynthetictrainingdatasize: TheupperpartofthefigurecorrespondstotheGrounding -GuidedGenerationtrainingstage, whilethebottompartrepresentstheWeak-to-StrongContrastiveAlignment -trainingstage. Fromlefttoright,theresultsarepresentedforASQA,ELI5,andQAMPARI,respectively. REC. -indicatesCitationRecallandPREC. denotesCitationPrecision. Thex-axisrepresentsthequantityofautomatically -generateddata. Itisobservedthatasthevolumeofautomaticallygenerateddataincreases,thereisaconsistent -improvementinbothcitationrecallandprecisionacrossthetwotrainingstages. -trainedwithG3 toexploretheimpactofthehyper- adheretotheevaluationframeworkestablishedin -parameter β on attribution quality. We evaluated (Gaoetal.,2023b). Forlong-formQAdatasetslike -thesevariantsontheASQAandELI5datasets,and ASQAandELI5,wealsoreportmetricsrelatedto -theexperimentalresultsareshowninFigure9. Fluency,ROUGE-L,andaverageresponselength. -The experimental results indicate that as β in- Specifically,weuseMAUVE(Pillutlaetal.,2021) -creases, the model’s performance on attribution toevaluatethefluencyofthemodelresponse. For -graduallydecreases. Thisobservationsuggeststhat datasetslikeQAMPARI,whereanswersarecom- -thefirststageofG3 mightintroduceanoticeable posed of concatenated entities, we calculate the -inconsistencybetweengroundingandattribution. averagenumberofpredictedentities. -Withhigherβvalues,themodelstrugglestoescape -theconstraintsofinconsistentattributedanswers, -| leading to | a reduction | in attribution | quality as | β | | | -| ---------- | ----------- | -------------- | ---------- | --- | --- | --- | -increases. -F FullResults -Wepresentthecomprehensiveresultsofourexperi- -| mentsinTables10,11,and12. | | Beyondtheevalua- | | | | | -| ------------------------- | --- | ---------------- | --- | --- | --- | --- | -tionmetricsrelatedtoCorrectnessandCitation,we - -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -| guide | answer | generation | and | cite the | sources properly. | | | | | | | -| ------------------- | ------ | ---------- | --- | ---------- | ----------------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table7: | InstructionFormatforFRONTonLong-formQA. | | | | | | | | -| --- | --- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -provide a list of accurate answers for the given question. Always cite one and only one document -| for each | answer. | Separate | answers | by | commas. | | | | | | | -| ------------------- | ------- | ---------- | ------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table8: | InstructionFormatforFRONTonShort-formQA. | | | | | | | | -| ------- | ----------------------------------- | --- | ------- | ---------------------------------------- | --- | ------------- | --- | --- | ---------------- | --- | --- | -| | Hyper-parameters | | | Value | | | | | | | | -| | Top-p | | | 0.95 | | | | | | | | -| | Temperature | | | 0.2 | | | | | | | | -| | Max-length | | | 2048 | | | | | | | | -| Table9: | Hyper-parametersettingsininference. | | | | | | | | | | | -| | | | | | | | 72 | | 56 | | | -| | | | | | | )%(1FnoitatiC | 71 | | )%(1FnoitatiC 55 | | | -Mistral-7B -| | | | VANILLA-SFT(Mistral-7B) | | | | 70 | | 54 | | | -| --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | FRONT(Mistral-7B) | | | | 69 | | 53 | | | -60 -| | | | | | | | 68 | | 52 | | | -| ---------------- | --- | --- | --- | --- | --- | --- | -------------- | ----------- | --- | -------------- | ----------- | -| )%(1FnoitatiC 50 | | | | | | | 67 | FRONT-7B | 51 | | FRONT-7B | -| 40 | | | | | | | 0.1 0.2 | 0.3 0.4 0.5 | | 0.1 0.2 | 0.3 0.4 0.5 | -| 30 | | | | | | | βonASQAdataset | | | βonELI5dataset | | -20 -| | | | | | | Figure | 9: Ablation | on hyperparameter | | β | in Weak-to- | -| --- | --- | --- | --- | --- | --- | -------------------------------------------- | ----------- | ----------------- | --- | --- | ----------- | -| 10 | | | | | | StrongContrastiveAlignmentstageonASQAandELI5 | | | | | | +u +J +- +0 +3 + +0 +2 +0 +2 +- +g +u +A +- +9 +2 + +0 +2 +0 +2 +- +p +e +S +- +8 +2 + +0 +2 +0 +2 +- +t +c +O +- +8 +2 + +0 +2 +0 +2 +- +v +o +N +- +7 +2 + +0 +2 +0 +2 +- +c +e +D +- +7 +2 + +1 +2 +0 +2 +- +n +a +J +- +6 +2 + +1 +2 0 -| | ASQA | | ELI5 | QAMPARI | | | | | | | | -| --- | ---- | --- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Figure8:Ablationstudyonmodelarchitecture:Wesub- -stitutedthefoundationmodelinFRONTwithMistral-7B -andcomparedtheexperimentalresultsofmodelsunder -| the same | foundation | model | using | in-context | learning | | | | | | | -| -------- | ---------- | ----- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | -andthosedirectlysupervisedfine-tunedonourautomat- -| icallygenerateddata. | | Theexperimentsdemonstratethat | | | | | | | | | | -| -------------------- | --------- | ----------------------------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | -| by replacing | different | foundation | | models, | our frame- | | | | | | | -workstillmaintainsitsgeneralizability. - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | -------- | ---------- | ---------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (EMRec.) | Rec. Prec. | F1 ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 73.41 | 40.37 | 72.81 69.69 | 71.22 37.92 | 39.24 | -| ------------ | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 79.90 | 24.32 | 17.24 17.87 | 17.55 29.38 | 42.29 | -| LLaMA-2 | 13B | 87.08 | 27.99 | 16.45 19.04 | 17.65 31.41 | 39.25 | -| | 70B | 69.28 | 31.53 | 44.18 44.79 | 44.48 31.53 | 26.86 | -| | 7B | 66.78 | 29.93 | 55.99 51.66 | 53.74 32.93 | 26.18 | -| LLaMA-2-Chat | 13B | 66.14 | 34.39 | 37.15 38.17 | 37.65 35.13 | 33.68 | -| | 70B | 86.60 | 41.24 | 60.19 61.16 | 60.67 37.01 | 47.09 | -| | 7B | 86.92 | 38.34 | 48.37 44.63 | 46.42 35.95 | 63.90 | -Vicuna-v1.5 -| | 13B | 66.11 | 35.20 | 51.92 53.40 | 52.65 35.74 | 38.57 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.37 | 29.46 | 23.12 25.45 | 24.23 31.67 | 37.17 | -Mistral -| | 8×7B | 83.30 | 36.30 | 32.72 34.49 | 33.58 35.05 | 38.47 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.86 | 38.57 | 64.90 59.67 | 62.18 36.21 | 45.26 | -Mistral-Instruct -| | 8×7B | 94.77 | 44.11 | 61.80 63.27 | 62.53 38.54 | 58.83 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -Post-hocRetrieval -| ChatGPT | - | 49.78 | 37.68 | 27.11 27.05 | 27.08 36.64 | 52.61 | -| ------------ | ---- | ----- | ----- | ----------- | ----------- | ------ | -| | 7B | 75.56 | 16.55 | 13.88 13.86 | 13.87 26.81 | 37.50 | -| LLaMA-2 | 13B | 77.91 | 20.51 | 20.95 20.94 | 20.94 29.53 | 31.37 | -| | 70B | 75.23 | 27.58 | 28.43 28.43 | 28.43 30.33 | 29.88 | -| | 7B | 22.50 | 14.17 | 11.33 11.33 | 11.33 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 64.52 | 24.43 | 21.43 21.43 | 21.43 33.91 | 41.12 | -| | 70B | 70.63 | 29.68 | 24.51 24.51 | 24.51 34.17 | 45.74 | -| | 7B | 63.87 | 19.58 | 16.24 16.24 | 16.24 33.22 | 41.80 | -| Vicuna-v1.5 | 13B | 73.83 | 24.79 | 24.11 24.11 | 24.11 34.42 | 43.54 | -| | 7B | 86.54 | 21.17 | 16.78 16.77 | 16.77 30.90 | 42.43 | -| Mistral | 8×7B | 80.99 | 36.30 | 38.37 35.27 | 36.75 35.05 | 38.47 | -| | 7B | 67.97 | 26.26 | 17.87 17.85 | 17.86 33.71 | 51.56 | -Mistral-Instruct 8×7B 65.51 33.90 24.57 24.48 24.52 36.20 53.83 -Training-based -| | 7B | 74.33 | 29.96 | 67.82 66.97 | 67.39 35.70 | 29.83 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -Self-RAG -| | 13B | 71.59 | 31.66 | 71.26 70.35 | 70.80 36.01 | 27.03 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 76.66 | 40.32 | 67.67 63.67 | 65.61 38.32 | 62.00 | -VANILLA-SFT -| | 13B | 84.36 | 40.85 | 71.49 66.21 | 68.75 38.22 | 58.82 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 81.88 | 40.84 | 77.70 69.89 | 73.59 36.95 | 53.93 | -FRONT -| | 13B | 76.11 | 41.51 | 78.44 73.66 | 75.95 38.63 | 57.56 | -| --- | --- | -------- | ---------------- | ----------- | ----------- | ----- | -| | | Table10: | ASQAfullresults. | | | | - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | ------------ | -------- | ------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (Claim) Rec. | Prec. F1 | ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 44.65 | 12.47 49.44 | 47.05 48.22 | 20.64 | 90.2 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 63.72 | 4.53 3.92 | 5.38 4.54 | 18.27 | 103.36 | -| LLaMA-2 | 13B | 62.19 | 7.77 8.49 | 8.43 8.46 | 19.95 | 88.23 | -| | 70B | 53.39 | 10.43 23.75 | 22.43 23.07 | 20.43 | 93.84 | -| | 7B | 32.80 | 12.47 19.90 | 15.48 17.41 | 20.88 | 96.42 | -| LLaMA-2-Chat | 13B | 29.08 | 13.83 16.50 | 16.09 16.29 | 21.04 | 94.32 | -| | 70B | 33.69 | 13.30 36.63 | 36.63 36.63 | 21.29 | 117.84 | -| | 7B | 31.45 | 12.30 29.81 | 22.45 25.61 | 21.36 | 105.68 | -Vicuna-v1.5 -| | 13B | 37.41 | 14.33 31.15 | 28.99 30.03 | 21.74 | 98.23 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 56.62 | 8.47 16.04 | 16.32 16.18 | 20.46 | 93.80 | -Mistral -| | 8×7B | 61.83 | 10.43 26.11 | 25.09 25.59 | 20.66 | 93.59 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 32.74 | 11.07 49.25 | 42.69 45.74 | 20.75 | 98.28 | -Mistral-Instruct -| | 8×7B | 38.51 | 13.93 49.28 | 48.34 48.81 | 21.34 | 113.71 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -Post-hocRetrieval -| ChatGPT | - | 22.79 | 18.77 14.55 | 14.55 14.55 | 22.28 | 106.83 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 72.80 | 7.23 6.84 | 6.84 6.84 | 19.14 | 88.19 | -| LLaMA-2 | 13B | 53.21 | 10.33 9.61 | 9.61 9.61 | 20.63 | 90.44 | -| | 70B | 58.97 | 11.10 10.27 | 10.26 10.26 | 20.41 | 77.85 | -| | 7B | 22.50 | 14.17 11.33 | 11.33 11.33 | 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 30.36 | 14.93 12.10 | 12.10 12.10 | 21.82 | 109.79 | -| | 70B | 37.87 | 16.03 12.93 | 12.93 12.93 | 21.57 | 99.94 | -| | 7B | 30.88 | 11.83 10.91 | 10.91 10.91 | 21.66 | 99.03 | -Vicuna-v1.5 -| | 13B | 32.59 | 15.20 14.06 | 14.06 14.05 | 14.05 | 108.16 | -| --- | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 52.45 | 10.47 8.64 | 8.64 8.64 | 20.48 | 90.17 | -Mistral -| | 8×7B | 48.39 | 13.57 11.62 | 11.62 11.62 | 21.43 | 91.97 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 27.41 | 17.07 13.20 | 13.20 13.20 | 21.52 | 106.93 | -Mistral-Instruct -| | 8×7B | 27.60 | 17.37 15.68 | 15.68 15.68 | 21.66 | 95.21 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -Training-based -| | 7B | 30.98 | 6.90 22.34 | 32.40 26.45 | 16.48 | 41.66 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -Self-RAG -| | 13B | 32.04 | 6.07 30.46 | 40.20 34.66 | 15.23 | 38.19 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -| | 7B | 44.12 | 9.63 42.30 | 40.06 41.15 | 20.58 | 80.43 | -VANILLA-SFT -| | 13B | 46.33 | 10.27 46.75 | 44.47 45.58 | 20.56 | 84.01 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 36.90 | 9.18 58.60 | 55.33 56.92 | 19.09 | 74.06 | -FRONT -| | 13B | 34.37 | 9.32 60.31 | 59.21 59.75 | 19.66 | 75.14 | -| --- | --- | -------- | ---------------- | ----------- | ----- | ----- | -| | | Table11: | ELI5fullresults. | | | | - -| | | Correctness | Citation | | -| --------- | --------- | ------------ | ---------- | ----------- | -| ModelType | ModelSize | | | | -| | | Rec.-5 Prec. | Rec. Prec. | F1 NumPred. | -Prompting-based -| ChatGPT | - | 20.28 19.84 | 19.06 22.03 | 20.44 4.71 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 12.56 11.32 | 6.03 6.35 | 6.19 7.02 | -| LLaMA-2 | 13B | 18.00 12.39 | 5.45 5.74 | 5.59 11.31 | -| | 70B | 18.50 14.79 | 10.10 10.50 | 10.30 8.31 | -| | 7B | 17.96 19.74 | 9.58 9.68 | 9.63 4.73 | -| LLaMA-2-Chat | 13B | 21.34 18.86 | 8.94 9.06 | 9.00 6.51 | -| | 70B | 22.62 18.04 | 13.49 13.98 | 13.73 7.44 | -| | 7B | 14.22 14.74 | 11.26 11.64 | 11.45 5.87 | -Vicuna-v1.5 -| | 13B | 22.06 19.60 | 13.04 13.74 | 13.38 7.62 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 16.96 15.98 | 7.50 7.76 | 7.63 6.29 | -Mistral -| | 8×7B | 18.18 15.63 | 9.72 10.20 | 9.95 6.63 | -| --- | ---- | ----------- | ----------- | ---------- | -| | 7B | 17.52 21.29 | 17.56 18.53 | 18.03 4.54 | -Mistral-Instruct -| | 8×7B | 20.12 19.64 | 19.27 20.38 | 19.81 5.32 | -| --- | ---- | ----------- | ----------- | ---------- | -Post-hocRetrieval -| ChatGPT | - | 25.14 22.85 | 12.29 12.29 | 12.29 5.46 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 6.48 5.11 | 5.05 5.05 | 5.05 6.55 | -| LLaMA-2 | 13B | 9.88 7.17 | 5.20 5.20 | 5.20 6.98 | -| | 70B | 14.44 12.44 | 7.49 7.49 | 7.49 7.41 | -| | 7B | 12.94 10.89 | 7.76 7.76 | 7.76 5.99 | -| LLaMA-2-Chat | 13B | 15.72 12.23 | 7.87 7.87 | 7.87 6.32 | -| | 70B | 17.90 14.45 | 9.05 9.05 | 9.05 6.05 | -| | 7B | 12.04 9.71 | 6.69 6.69 | 6.69 7.10 | -Vicuna-v1.5 -| | 13B | 14.78 11.47 | 8.50 8.50 | 8.50 6.67 | -| --- | --- | ----------- | --------- | --------- | -| | 7B | 9.94 7.90 | 6.00 6.00 | 6.00 7.38 | -Mistral -| | 8×7B | 13.92 12.08 | 6.70 6.70 | 6.70 6.58 | -| --- | ---- | ----------- | --------- | --------- | -| | 7B | 15.80 12.15 | 8.34 8.34 | 8.34 7.01 | -Mistral-Instruct -| | 8×7B | 24.16 18.28 | 9.78 9.78 | 9.78 7.37 | -| --- | ---- | ----------- | --------- | --------- | -Training-based -| | 7B | 2.34 1.98 | 10.53 18.80 | 13.50 3.49 | -| --- | --- | --------- | ----------- | ---------- | -Self-RAG -| | 13B | 1.90 1.33 | 12.79 20.90 | 15.86 3.08 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 12.86 21.09 | 21.35 21.36 | 21.35 7.49 | -VANILLA-SFT -| | 13B | 12.68 22.80 | 23.64 23.71 | 23.67 3.14 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 11.50 21.38 | 24.74 24.84 | 24.79 3.08 | -FRONT -| | 13B | 11.94 22.61 | 24.86 25.39 | 25.12 3.17 | -| --- | -------- | ------------------- | ----------- | ---------- | -| | Table12: | QAMPARIfullresults. | | | \ No newline at end of file +2 +- +b +e +F +- +5 +2 + +1 +2 +0 +2 +- +r +a +M +- +7 +2 + +Wipro + +Sensex + +NYSE Composite Index + +139 + + Other Disclosures +Description of Voting Rights +Dematerialisation of Shares and Liquidity +Outstanding ADR/GDR/Warrants or any +other Convertible instruments, Conversion +Date and Likely Impact on Equity +Commodity Price Risk or Foreign Exchange +Risk and Hedging Activities + +Credit Ratings + +Plant Locations + +All our equity shares carry voting rights on a pari-passu basis. +99.85% of outstanding equity shares have been dematerialized as at March 31, 2021. + +The Company has 2.61% of outstanding ADRs as on March 31, 2021. + +The Company had no exposure to commodity and commodity risks for the financial year +2020-21. For Foreign exchange risk and hedging activities, please refer Management +Discussion and Analysis Report for details. +The ICRA Committee of ICRA has reaffirmed the long-term rating for lines of credit of Wipro +Limited at [ICRA]AAA. The Outlook on the long-term rating is stable. The Rating Committee +of ICRA has also re-affirmed the short-term rating at [ICRA]A+. Fitch Ratings has assigned +Long-Term Foreign- and Local-Currency Issuer Default Ratings (IDR) and foreign-currency +senior unsecured rating of ‘A-’. Standard & Poor has also assigned a rating of A-. The Outlook +is Stable. +The Company has various offices in India and abroad. Details of these locations as on March +31, 2021 are available on our website www.wipro.com. + +Registrar and Transfer Agents +Company’s share transfer and related activities are +operated through its Registrar and Share Transfer Agents: +KFin Technologies Private Limited, Hyderabad. + +Share Transfer System +In accordance with the proviso to Regulation 40(1) of the +Listing Regulations, effective from April 1, 2019, transfers +of shares of the Company shall not be processed unless the +shares are held in the dematerialized form with a depository. +Accordingly, shareholders holding equity shares in physical +form are urged to have their shares dematerialized so as to +be able to freely transfer them. + +Investor Queries and Grievances Redressal +Shareholders may write either to the Company or the +Registrar and Transfer Agents for redressal of queries and +grievances. The address and contact details of the concerned +officials are given below. + +Overseas Depository for ADSs - J.P. Morgan Chase Bank N.A. +383 Madison Avenue, Floor 11 New York, NY10179 +General: +1 800 990 1135 +From outside the U.S.: +1 651 453 2128 +Tel: +1 212 552 8926 New York +E-mail: drx_depo@jpmorgan.com + +Indian Custodian for ADSs +India Sub Custody +Office Address: J.P. Morgan Chase Bank, N.A. Mumbai +Branch, Paradigm B-Wing, 6th Floor, Mindspace, Malad (W), +Mumbai - 400 064 +Phone: +91 022 6649 2515 | F: +91 022 6649 2509 +The e-mail address and contact details for all service related +queries is: india.custody.client.service@jpmorgan.com + +Contact Persons: +Rohit Keer- E-mail id: rohit.a.keer@jpmchase.com, +Nekzad Behramkamdin- E-mail id: nekzad.behramkamdin@ +jpmorgan.com +Nayan Vyas- E-mail id: nayan.x.vyas@jpmorgan.com + +Registrar and Share Transfer Agents +KFin Technologies Private Limited, Unit: Wipro Limited, +Selenium Tower B, Plot 31 & 32, Financial District, +Nanakramguda, Serilingampally Mandal, +Hyderabad - 500 032, Telangana. +Toll Free No.: 1800 3454 001 Phone: (040) 7961 1000 + +Contact Person: +Mr. B. Srinivas- e-mail id: srinivas.b@kfintech.com +Ms. Rajitha Cholleti- e-mail id: rajitha.cholleti@kfintech.com +Shareholders Grievance can also be sent through e-mail to the +following designated e-mail id: einward.ris@kfintech.com. + +Web-Based Query Redressal System +Members may utilize this facility extended by the Registrar & +Transfer Agents for redressal of their queries. + +Please visit https://karisma.kfintech.com and click on +“investors” option for query registration through free identity +registration to log on. Investor can submit the query in +the “QUERIES” option provided on the website, which will +generate the grievance registration number. For accessing +the status/response to your query, please use the same +number at the option “VIEW REPLY” after 24 hours. The +investors can continue to put additional queries relating to +the case till they are satisfied. + +140 + + Shareholders can also send their correspondence to the Company with respect to their shares, dividend, request for annual +reports and shareholder grievances. The contact details are provided below: + +Mr. M. Sanaulla Khan +Vice President and Company Secretary +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 +Mr. G. Kothandaraman +General Manager, Finance +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Ph: +91 80 28440011 (Extn: 226185) +Fax: +91 080 28440054 +E-mail: sanaulla.khan@wipro.com +Ph: +91 80 28440011 (Extn: 226183) +Fax: +91 080 28440054 +E-mail: kothandaraman.gopal@wipro.com + +Analysts can reach our Investor Relations Team for any queries and clarification on Financial/Investor Relations related +matters: + +Ms. Aparna C. Iyer +Vice President, Finance +Corporate Treasurer and Investor Relations +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Mr. Abhishek Kumar Jain +General Manager +Investor Relations +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Ph: +91 80 28440011 (Extn: 226186) +Fax: +91 80 28440054 +E-mail: iyer.aparna@wipro.com + +Ph: +91-80-6142 6143 +Fax: +91 80 28440054 +E-mail: abhishekkumar.jain@wipro.com + +Due to the ongoing COVID-19 pandemic, majority of our staff is working from home. In case of any queries, stakeholders are +requested to write to the above mentioned Email IDs for a quicker response. + +141 + + Corporate Governance Compliance Certificate + +Corporate Identity Number: L32102KA1945PLC020800 +Nominal Capital: ` 2527.40 crores + +To the Members of +Wipro Limited +Doddakannelli, Sarjapur Road, +Bengaluru – 560035 + +We have examined all the relevant records of Wipro Limited (“Company”) for the purpose of certifying compliance of the +conditions of the Corporate Governance under the SEBI (Listing Obligations and Disclosure Requirements) Regulations, 2015 +for the financial year ended March 31, 2021. We have obtained all the information and explanations which to the best of our +knowledge and belief were necessary for the purposes of certification. + +The compliance of conditions of corporate governance is the responsibility of the Management. Our examination was limited +to the procedure and implementation process adopted by the Company for ensuring the compliance of the conditions of the +corporate governance. + +This certificate is neither an assurance as to the future viability of the Company nor of the efficacy or effectiveness with +which the management has conducted the affairs of the Company. + +In our opinion and to the best of our information and according to the explanations and information furnished to us, we certify +that the Company has complied with all the mandatory requirements of Corporate Governance as stipulated in Schedule II of +the said Regulations. As regards Discretionary Requirements specified in Part E of Schedule II of the SEBI (Listing Obligations +and Disclosure Requirements) Regulations, 2015, the Company has complied with items C and E. + +For V. SREEDHARAN & ASSOCIATES +Company Secretaries + +(V. Sreedharan) +Partner +FCS: 2347; CP No. 833 + +Bengaluru +June 9, 2021 + +UDIN: F002347C000436627 + +NOTE: Due to the ongoing Covid-19 pandemic, we have conducted online verification and examination of records, as facilitated by the +Company for the purpose of issuing this Certificate. + +142 + + CERTIFICATE OF NON-DISQUALIFICATION OF DIRECTORS +[Pursuant to Regulation 34(3) and Schedule V Para C Clause (10)(i) of the SEBI (Listing +Obligations and Disclosure Requirements) Regulations, 2015] + +To, +The Members of +WIPRO LIMITED +Doddakannelli, Sarjapur Road, +Bengaluru- 560035 + +We have examined the relevant registers, records, forms, returns and disclosures received from the Directors of +WIPRO LIMITED, having CIN L32102KA1945PLC020800 and having registered office at Doddakannelli, Sarjapur Road, +Bengaluru- 560035 (hereinafter referred to as ‘the Company’), produced before us by the Company for the purpose of issuing +this Certificate, in accordance with Regulation 34(3) read with Schedule V Para-C Sub-clause 10(i) of the Securities and +Exchange Board of India (Listing Obligations and Disclosure Requirements) Regulations, 2015. + +In our opinion and to the best of our information and according to the verifications (including Directors Identification Number +(DIN) status at the portal www.mca.gov.in) as considered necessary and explanations furnished to us by the Company and +its officers, we hereby certify that none of the Directors on the Board of the Company as stated below for the Financial Year +ending on March 31, 2021 have been debarred or disqualified from being appointed or continuing as Directors of companies +by the Securities and Exchange Board of India (SEBI) and Ministry of Corporate Affairs (MCA), or any such other Statutory +Authority. + +Details of Directors: + +Sl. No. +1. +2. +3. +4. +5. +6. +7. +8. +9. + +Name of Director + +Mr. Azim Premji Hasham +Mr. Mahendra Kumar Sharma +Mr. William Arthur Owens +Mr. Rishad Premji Azim +Mrs. Ireena Vittal +Mr. Patrick John Ennis +Mr. Patrick Lucien Andre Dupuis +Mr. Deepak Madhav Satwalekar +Mr. Thierry Delaporte + +DIN +00234280 +00327684 +00422976 +02983899 +05195656 +07463299 +07480046 +00009627 +08107242 + +Date of appointment in the +Company +01/09/1968 +01/07/2011 +01/07/2006 +01/05/2015 +01/10/2013 +01/04/2016 +01/04/2016 +01/07/2020 +06/07/2020 + +Ensuring the eligibility for the appointment/continuity of every Director on the Board is the responsibility of the management +of the Company. Our responsibility is to express an opinion based on our verification. This certificate is neither an assurance +as to the future viability of the Company nor of the efficiency or effectiveness with which the management has conducted the +affairs of the Company. + +For V. SREEDHARAN & ASSOCIATES +Company Secretaries + +(V. Sreedharan) +Partner +FCS: 2347; CP No. 833 + +Bengaluru +June 09, 2021 + +UDIN: F002347C000436638 + +NOTE: Due to the ongoing Covid-19 pandemic, we have conducted online verification and examination of records, as facilitated by the +Company for the purpose of issuing this Certificate. + +143 + diff --git a/q126/random_k4/question.json b/q126/random_k4/question.json index e8c6ffd5e2b275f7d00f992a4c843a9fa3af162b..a18f77c949356e7ab789efdfadc3fccc89509efb 100644 --- a/q126/random_k4/question.json +++ b/q126/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "owners-manual-2170416.pdf", - "web_974a02e9fcc23ca9.pdf", - "web_ddfd3d214b1f0ea4.pdf", - "web_7c476895c4da1bbe.pdf", - "web_623919b9713226a3.pdf" + "0a89b981c2736e87bac48a090e1a3e60dd8a4e0f.pdf", + "web_ccf5b62f39e26967.pdf", + "5fdcf120b846746985684fa54250aab44639b30c.pdf", + "web_2c8b6d1a0ca91c7b.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q126/random_k4/random_1.md b/q126/random_k4/random_1.md index 0f5db4f84cd2bbe84dc239b53c3a2c97d631e42e..16de3e3316b1e3bb316b878e17b05640d996b000 100644 --- a/q126/random_k4/random_1.md +++ b/q126/random_k4/random_1.md @@ -1,279 +1,1070 @@ -COMPETITOR ANALYSIS - -© InnoValor, 2018 -© InnoValor, 2018 - -How to use -Template -Example - - Competitor Analysis - -A Competitor Analysis helps you to understand your -competition. It supports you in creating your unique position in -the market. With this analysis you make an assessment of the -strengths and weaknesses of your current and potential -competitors. With a better understanding of your competition -you can choose and implement effective strategies that will -improve your competitive advantage. - -Tips for use - -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. This tool can easily be used by -yourself, or with colleagues. Print the template and use the tool -to create an understanding of your competition. - -Have a look at the example of Fleur Flowers to get an -understanding of a Competitor Analysis. - -How to use - -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. Follow the steps to complete your -own Competitor Analysis. - -Summarize your target group - -Shortly describe your target group. Who are the customers that -buy your offering? Make the description as specific as possible. If - -© InnoValor, 2018 - -your target group consist of businesses you can mention their -size, industry or their offering. If your target group consist of -consumers you can mention their age, goals, gender, education -level or interests. - -Identify your competitors - -Write the names for up to three competitors in the first row. To -help you generate this list, think of where your customers would -go if they cannot buy your product or service. - -Determine competitors’ strengths and weaknesses - -Now think like your customers. Why would a customer go for the -services of your competitors instead of coming to your company? -You can describe this by thinking about the offerings of your -competitors. It would even be better when you visit your -competitors and experience what their customers experience. -Write down three strengths of each competitor. Also consider -why customers would not go to your competitors. Write down -three weaknesses of each competitor. - -Describe the key findings - -Finally, in the fourth column of the template write down the key -findings for the strengths. What stands out? What patterns can -you discover from your competitors? And which opportunities -can you come up with? Also, write down your key -findings for the weaknesses. What are the aspects that - - your competitors don’t do well? How could you avoid making -the same mistakes? What should you do to serve your customers -in such a way that you differentiate from your competitors in a -positive way? - -© InnoValor, 2018 - - competitor analysis - -name - -target group - -competitors - -most important findings - -s -h -t -g -n -e -r -t -s - -s -e -s -s -e -n -k -a -e -w - -date - - Example Fleurs Flowers - -industry - -Retail - -products & -services - -Flowers - -size of -business - -2 full-timers 2 part-timers - -revenue - -€ 37,000 - -location - -Main street - Bloomfield - -Fleurs Flowers is a flower shop in a small town. Fleur sells -flowers to a small group of regular clients who regularly -purchase fresh flowers from her shop. Her customers generally -consist of men and women aged 25. The customers often just -bought a house and flowers are the perfect way to bring -atmosphere to the new place. In addition, she has customers -who buy flowers for birthday and holiday occasions. With -Valentine’s and Mother's Day she is working overtime to meet -the demand - -Fleurs Flowers’ competitor analysis - -A company like Fleurs Flowers has a difficult time in today's -society. She has a physical store that is open during shopping -hours. Her customers are people who also work at those times. -These customers search for convenience in their shopping. -Internet shopping is very popular amongst these people. Fleur -wants to increase her appeal to her target group. She -investigates the advantages and disadvantages of the sales -strategies of her main competitors. For her research Fleur visits -her competitors and buys their flowers. The results of Fleur’s -competitor analysis can be read in the example on the next page. - -© InnoValor, 2018 - - competitor analysis - -Fleurs Flowers - -name - -target group - -Men or women from 25 years old with an average salary - -fromyouflowers.com - -Lidl - -Valentino flowershop - -competitors - -most important findings - -Offers flowers online - -The price for flowers is low - -Customers can order -flowers online but have to -pick them up at the store - -Delivers at home though -whole the region within 24 -hours - -Customers can buy flowers -while doing groceries, they do -not have to visit a different store - -When ordering flowers above a -certain amount the customer -will receive a free vase - -Package deals – get a -discount on a vase when -flowers are bought - -Everyday there are new flowers - -A lot of time is spend on -wrapping the flowers as a gift - -An online shop seems to be more -and more important. Customers -want to be able to order flowers -from their home. Customers also -like it when they get something for -free or a discount on something -when they buy flowers - -There is no way that a -customer can see the flowers -before they are ordered - -Customers cannot create -their own piece since the -flowers are all prearranged - -The price for flowers is higher -than in other shops - -Ordering flowers is less -personal - -The flowers are made in mass - -Ordering flowers takes -quite some time - -Customers cannot create -their own bouquet of flowers - -Buying flowers is not a special -occasion since it is been done -during doing groceries - -The shop is not as much open -as customers would like, for -instance it is closed on Mondays - -Customers seem to think that -buying flowers is a personal and -special occasion. They do not like -it when buying flowers is less -personal. They also do not like that -they cannot create their own -bouquet - -s -h -t -g -n -e -r -t -s - -s -e -s -s -e -n -k -a -e -w - -date - +SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation +| | YagizCanAkay1 | | | MuhammedYusufKartal1 | | | EsraAlparslan1 | | | | | +| --- | ------------- | --- | ------------------- | -------------------- | ------------ | --- | -------------- | --- | --- | --- | --- | +| | | | FarukOrtakoyluoglu1 | | ArdaAkpinar2 | | | | | | | +1TOBBUniversityofEconomicsandTechnology 2OSTIMTechnicalUniversity +{y.akay, m.kartal, ealparslan, fortakoyluoglu}@etu.edu.tr 230206003@ostimteknik.edu.tr +| | | Abstract | | | andcommon,yetexposecriticalcoverageandrea- | | | | | | | +| --- | --- | -------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | +soningbottlenecksincurrentNLPsystems. +Answeringcomplex,real-worldqueriesoften +6202 raM 9 ]LC.sc[ 1v92380.3062:viXra Traditionalsingle-indexorsingle-agentretrieval +requiressynthesizingfactsscatteredacrossvast +| | | | | | systemsstruggleto | | maintainbothscalability | | | | and | +| -------- | -------- | -------- | --------- | ----- | ----------------- | --- | ----------------------- | --- | --- | --- | --- | +| document | corpora. | In these | settings, | stan- | | | | | | | | +contextualrelevanceoverlarge,heterogeneouscor- +| dard retrieval-augmented | | | generation | (RAG) | | | | | | | | +| ------------------------ | --- | --- | ---------- | ----- | ---------------- | --- | -------- | --------------- | --- | ------- | --- | +| | | | | | pora, especially | | for long | or structurally | | complex | | +pipelinessufferfromincompleteevidencecov- +erage,whilelong-contextlargelanguagemod- documents. Standard RAG pipelines retrieve a +| | | | | | fixed number | | of documents | K | and process | | them | +| ---------- | -------- | --------- | -------- | ---- | ------------ | --- | ------------ | --- | ----------- | --- | ---- | +| els (LLMs) | struggle | to reason | reliably | over | | | | | | | | +massive inputs. We introduce SPD-RAG, a withinasinglecontextwindow(Lewisetal.,2020), +hierarchicalmulti-agentframeworkforexhaus- +whichfailswhenanswersdependoninformation +tivecross-documentquestionansweringthatde- +distributedacrossmanydocuments,sinceevidence +composestheproblemalongthedocumentaxis. +| | | | | | beyondthetop-K | | resultsistypicallydiscardedin | | | | | +| ---- | -------- | ------------ | ---- | --------- | -------------- | --------- | ----------------------------- | ------------ | --- | ---- | --- | +| Each | document | is processed | by a | dedicated | | | | | | | | +| | | | | | a single | retrieval | pass. | Long-context | | LLMs | ex- | +document-levelagentoperatingonlyonitsown +content,enablingfocusedretrieval,whileaco- tendcontextwindowsto128K–2Mtokens(Gemini +| | | | | | Teametal.,2024), | | butempiricalevidenceshows | | | | | +| --------- | ---------- | ----- | ----------- | ------ | ---------------- | --- | ------------------------- | --- | --- | --- | --- | +| ordinator | dispatches | tasks | to relevant | agents | | | | | | | | +and aggregates their partial answers. Agent thatreasoningqualitydegradessignificantlyascon- +outputsaresynthesizedbymergingpartialan- text length increases (Liu et al., 2023). This sug- +swersthroughatoken-boundedsynthesislayer +| | | | | | gests that | the | bottleneck | is not | only | retrieval, | but | +| --- | --- | --- | --- | --- | ---------- | --- | ---------- | ------ | ---- | ---------- | --- | +(whichsupportsrecursivemap-reduceformas- +alsoreasoningatscaleovermanydocumentsand +| sivecorpora). | | Thisdocument-levelspecializa- | | | | | | | | | | +| ------------- | --- | ----------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +hundredsofthousandsoftokens. +tionwithcentralizedfusionimprovesscalabil- +ityandanswerqualityinheterogeneousmulti- In this work, we introduce SPD-RAG +| | | | | | (Sub-agent | per | Document | Retrieval-Augmented | | | | +| -------- | -------- | ----- | -------- | ---------- | ---------- | --- | -------- | ------------------- | --- | --- | --- | +| document | settings | while | yielding | a modular, | | | | | | | | +extensibleretrievalpipeline. OntheLOONG Generation),ahierarchicalmulti-agentarchitecture +benchmark (EMNLP 2024) for long-context forexhaustivemulti-documentquestionanswering +multi-document QA, SPD-RAG achieves an thatfactorstheproblemalongthedocumentaxis +AvgScoreof58.1(GPT-5evaluation),outper- +| | | | | | rather than | the | task axis. | Instead | | of forcing | a | +| --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | ---------- | --- | +formingNormalRAG(33.0)andAgenticRAG +| | | | | | single model | | to hunt through | | a massive | | global | +| --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --------- | --- | ------ | +(32.8)whileusingonly38%oftheAPIcostof +| | | | | | index, SPD-RAG | | uses | a central | coordinator | | to | +| --- | --- | --- | --- | --- | -------------- | --- | ---- | --------- | ----------- | --- | --- | +afull-contextbaseline(68.0). +| | | | | | decompose | the | user’s | query into | shared | instruc- | | +| --- | --- | --- | --- | --- | --------- | --- | ------ | ---------- | ------ | -------- | --- | +1 Introduction tions. It then deploys a dedicated, cost-efficient +| | | | | | sub-agenttoeachdocumentinthecorpus. | | | | | | These | +| --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | ----- | +Large language models (LLMs) and the agentic agents operate entirely in parallel, treating their +systems built around them are increasingly used assigneddocumentsasisolatedretrievaluniverses +forcomplexinformationsearchtasks(Guoetal., +| | | | | | to extract | relevant | findings. | Finally, | | a synthesis | | +| --- | --- | --- | --- | --- | ---------- | -------- | --------- | -------- | --- | ----------- | --- | +2024). Real-worldquestionsoftenrequiresynthe- +| | | | | | model | aggregates | these | document-grounded | | | | +| --- | --- | --- | --- | --- | ----- | ---------- | ----- | ----------------- | --- | --- | --- | +sizingevidencescatteredacrossmanydocuments, findings—employingarecursivemergingfallback +suchasassessingacompany’sfinancialrisksacross +| | | | | | for exceptionally | | large | corpora—to | | construct | a | +| --- | --- | --- | --- | --- | ----------------- | --- | ----- | ---------- | --- | --------- | --- | +yearsofreportsorintegratingfindingsfrommul- +| | | | | | comprehensive, | | final answer. | | Our codebase | | and | +| --- | --- | --- | --- | --- | -------------- | --- | ------------- | --- | ------------ | --- | --- | +tiplescientificpapers.1 +| | | Suchquestionsarenatural | | | datasetsusedinthispapercanbefoundat | | | | | | | +| --- | --- | ----------------------- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +https://github.com/NebulAICompany/ +| 1Examples | inspired | by scenarios | in the | Loong bench- | | | | | | | | +| ---------------------- | -------- | ------------ | ------ | ------------ | -------- | --- | --- | --- | --- | --- | --- | +| mark(Wangetal.,2024a). | | | | | SPD-RAG. | | | | | | | + +Ourcontributionsareasfollows: tasks,butthatmulti-agentoverheadgrowssuperlin- +early. Theirresultthatindependentagentsamplify +• WeproposeSPD-RAG,ahierarchicalmulti- +errorsby17.2×comparedto4.4×forcentralized +agentframeworkthatcombinesper-document +systemsdirectlymotivatesourcoordinator-based +agentic RAG with a centralized synthesis design. Thetaxonomyofhierarchicalmulti-agent +layer, where cost-efficient document agents systemsbyMoore(2025)furtherclassifiesdesign +| are coordinated | by a | smarter coordinator | | | | | | | +| --------------- | ---- | ------------------- | --- | ---------------------- | --- | ------------------------- | --- | --- | +| | | | | patternsalongfiveaxes: | | controlhierarchy,informa- | | | +agent,enablingdocument-levelspecialization +| | | | | tion flow, | role delegation, | temporal | layering, | and | +| ----------------- | -------------- | --------------- | --------- | ----------------------- | ---------------- | -------- | --------- | --- | +| and parallel | execution | while allowing | each | communicationstructure. | | | | | +| document | to be analyzed | in depth | to ensure | | | | | | +| that all relevant | information | is incorporated | | | | | | | +2.2 LongDocumentandMulti-DocumentQA +withoutmissingcriticalevidence. +| | | | | Long documents | that | exceed | LLM context | win- | +| --- | --- | --- | --- | -------------- | ---- | ------ | ----------- | ---- | +• WeevaluateontheLoongbenchmark(Wang +| | | | | dows are | often handled | via | divide-and-conquer | | +| --- | --- | --- | --- | -------- | ------------- | --- | ------------------ | --- | +etal.,2024a),whichincludesmulti-document +| | | | | strategies. | LLM×MapReduce(Zhouetal.,2025) | | | | +| ------------ | -------------- | ---------- | -------- | --------------------------- | ----------------------------- | --- | ----------------- | --- | +| QA instances | over financial | reports | and aca- | | | | | | +| | | | | formalizestwokeychallenges: | | | inter-chunkdepen- | | +| demic papers | with an | average of | 11 docu- | | | | | | +dency(onechunkdependsoncontextfromanother) +mentsperinstanceandcontextlengthsfrom +andinter-chunkconflict(chunksprovidecontradic- +| 10K to beyond | 250K | tokens, and | show that | | | | | | +| ------------- | ---- | ----------- | --------- | --- | --- | --- | --- | --- | +toryinformation),andproposesstructuredinforma- +oursystemsubstantiallyoutperformsstandard +tionprotocolsandin-contextconfidencecalibration +RAGandAgenticRAGbaselinesunderGPT- +| | | | | tomitigatethem. | ToM(Guoetal.,2025)extends | | | | +| -------- | ------------------ | ---- | ------- | --------------- | ------------------------- | ------------ | ------- | ------ | +| 5-judged | Avg Score (adapted | from | Loong’s | | | | | | +| | | | | this approach | with a | hierarchical | DocTree | repre- | +GPT-4-judgedprotocol;see§4.3),withgains +sentationforrecursivereasoning. +| of around | +25 points | (76% higher | average | | | | | | +| --------- | ---------- | ----------- | ------- | --- | --- | --- | --- | --- | +Formulti-documentQA,LongAgent(Zhaoetal., +score)comparedtoNormalRAGandAgentic +| | | | | 2024) splits | a 128K-token | document | | into chunks | +| --- | --- | --- | --- | ------------ | ------------ | -------- | --- | ----------- | +RAGbaselines. +| | | | | assigned | to member | agents, with | a leader | agent | +| --- | --- | --- | --- | ------------- | -------------- | ------------ | ---------------- | ----- | +| | | | | orchestrating | discussionsand | | inter-membercom- | | +• Weanalyzeablations,documenttypeandtask +complexityeffects,andcost–qualitytradeoffs, munication to reduce hallucinations. However, +showingthat SPD-RAGattainsover85%of LongAgenttargetsasinglelongdocumentrather +thanmultipleindependentdocumentsthatmaycon- +| full-context | baseline quality | at roughly | 38% | | | | | | +| ------------ | ---------------- | ---------- | --- | --------------- | ---- | ------------- | ---- | -------- | +| | | | | flict. DocAgent | (Sun | et al., 2025) | uses | a multi- | +oftheAPIcost. +| | | | | agent framework | that | mimics | human | reading via | +| --- | --- | --- | --- | --------------- | ---- | ------ | ----- | ----------- | +2 RelatedWork +atree-structuredoutlineandrevieweragent,while +| | | | | MDocAgent | (Han et | al., 2025) | employs | five spe- | +| --- | --- | --- | --- | --------- | ------- | ---------- | ------- | --------- | +2.1 LLM-BasedMulti-AgentSystems +| | | | | cialized | agents for multi-modal | | document | under- | +| --- | --- | --- | --- | -------- | ---------------------- | --- | -------- | ------ | +LLM-basedmulti-agentsystemsareincreasingly +standing. +| used for complex | task solving. | Guo et | al. (2024) | | | | | | +| ---------------- | ------------- | ------ | ---------- | --- | --- | --- | --- | --- | +provideacomprehensivesurveyofagentprofiles, +2.3 HierarchicalandRecursive +communicationprotocols,andcollaborationstrate- +Summarization +| gies. Hierarchical | architectures | such | as MegaA- | | | | | | +| ------------------ | ------------- | ---- | --------- | --- | --- | --- | --- | --- | +gent(Wangetal.,2024b),whichdemonstratesau- Ourhierarchicalmergingmechanismbuildsontree- +tonomouscooperationamongupto590agentsvia basedretrievalandrecursivesummarization. RAP- +multi-leveltaskdecomposition,andAgentOrches- TOR(Sarthietal.,2024)recursivelyembeds,clus- +tra(Zhangetal.,2025),whichachievesstate-of-the- ters, and summarizes text chunks to construct a +artperformanceonGAIAusingacentralplanning bottom-uptreewithmultipleabstractionlevels;re- +agent and specialized sub-agents, exemplify this trievingfromtheappropriatelevelyieldsa20%ab- +| trend. | | | | soluteaccuracygainonQuALITY.OuandLapata | | | | | +| ------ | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | +Scalinglawsformulti-agentsystemsofferquan- (2025)showthathierarchicalmergingcanamplify +titativeguidanceforarchitecturedesign. Kimetal. hallucinationsduringrecursivesummarizationand +(2025)evaluate180configurationsacrossfivear- proposecontext-awareaugmentationstrategiesthat +chitectures, finding that centralized coordination replaceintermediatesummarieswithrelevantinput +yields an 80.9% improvement on parallelizable context,whichweadoptinourmergingprotocol. + +2.4 Graph-BasedDocumentAnalysis 3 Methodology +Figure 1 presents an overview of the SPD-RAG +Graph-based document analysis is particu- +architecture. Thesystemprocessesaquerythrough +larly effective for global sensemaking queries. +three layers: the Coordination Layer (§3.1), the +GraphRAG(Edgeetal.,2024)constructsaknowl- +ParallelRetrievalLayer(§3.2),andtheSynthesis +edge graph of entities and relations from source +Layer(§3.3). +documents and applies hierarchical community +detection(e.g.,theLeidenalgorithm(Traagetal., +3.1 CoordinationLayer +2019)) to obtain multi-level communities with +Given a query q and corpus D, the system first +textual summaries. At query time, it retrieves +processes the query through a coordinator agent. +relevant communities, elicits community-level +The coordinator’s role is to decompose the user +partial answers, and aggregates them via a +query into a Shared Instruction Set for Docu- +map–reduce–styleprocedure. AsurveyonLLM- +ments and Synthesis Directives. It generates a +empoweredknowledgegraphconstruction(Bian, +WriteTodosstructuredobjectcontaining(a)alist +2025) highlights how LLMs are reshaping the +of sub_agent_todos: atomic, self-contained ex- +classical knowledge graph pipeline across three +traction tasks specifying exactly what fields, en- +key stages: ontology engineering, knowledge +tities, or numeric values to extract (the shared in- +extraction, and knowledge fusion. Our work is +structions);and(b)asynthesis_directive(2–4 +complementary: instead of inducing an explicit +sentences) instructing the downstream Synthesis +knowledge graph, SPD-RAG operates directly +Layeronhowtoprioritizeandstructurethemerged +over agent-produced textual summaries and +response. +performs similarity-guided recursive synthesis +acrossmanydocuments. 3.2 ParallelRetrievalLayer +Weassignadedicatedsub-agentα toeachdocu- +i +mentd inthecorpus. Eachsub-agentoperatesas +2.5 BenchmarksforMulti-Document i +anindependentRAGloopoveritsassigneddocu- +Reasoning +ment,guidedbythesharedinstructionsetfromthe +The Loong benchmark (Wang et al., 2024a) CoordinationLayer. +evaluates long-context LLMs on extended multi- +Document-Scoped Retrieval and Reasoning. +document QA where every document is relevant +Each document d serves as an isolated retrieval +i +("Leave No Document Behind") across financial, +universe; the assigned sub-agent is strictly con- +legal, and academic domains in English and Chi- +strainedtosearchtoolcallswithind , preventing +i +nese. Itspansfourtasktypeswithcontextlengths +cross-documentdistractorchunksfromdegrading +from 10K to over 250K tokens, and even strong +local extraction. Each sub-agent receives the +frontier models achieve only modest Avg Scores +originalqueryq,itsassigneddocumentname,and +andmediumperfect-answerrates,underscoringthe +thesharedinstructionset(todolist),thenoperates +difficultyofreasoningovermanylong,all-relevant +in an iterative retrieve-and-reason loop. On each +documents. +turn, the agent either issues a search action (a +Complementarydatasetsprobeotheraspectsof focused query string) or a finalize action that +multi-document reasoning. MoNaCo (Wolfson terminates the loop and emits the document- +et al., 2025) focuses on natural questions over groundedfindings. Thesub-agentpromptinstructs +dozensofWikipediapages,MEBench(Linetal., the agent to attempt at least 2 focused searches +2025)onhigh-entity-densityQA,FanOutQA(Zhu before concluding a piece of information is +et al., 2024) on multi-hop questions with absent, with a total cap of 5 search calls across +long cross-document evidence chains, and all tasks. Each search action triggers dense +HoloBench (Maekawa et al., 2024) on database- vector retrieval over the sub-agent’s dedicated +style aggregation over unstructured text. Taken per-documentQdrantindex. ThisretrievalusesCo- +together, these benchmarks show that scalable hereembed-v4.0embeddings(1536-dimensional, +multi-document reasoning remains an open cosine similarity) to fetch the top k = 15 most +challenge,withLoongemphasizingtheespecially similar chunks. These chunks are then re-ranked +demandingall-relevant,long-contextregime. byCoherererank-v4.0-fasttoyieldafinalset + +| | | | | Figure1: | OverviewofSPD-RAGArchitecture. | | | | | | | | | +| --- | --- | --- | --- | -------- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | +of top_n = 5 chunks per search. The documents Similarity-Ordered Merge & Synthesis. +t, +are split into chunks offline using LangChain’s In iteration we maintain a set of find- +RecursiveCharacterTextSplitter in Mark- ings/summaries S(t) = {s (t) ,...,s (t) }. We +| | | | | | | | | | | | 1 | n | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +downmodewithachunksizeof1000tokensand embed each summary using Cohere embed-v4.0 +| 250-tokenoverlap. | | | | | | | and compute | | a cosine-similarity | | | matrix | via | +| ----------------- | --- | --- | --- | --- | --- | --- | ----------- | --- | ------------------- | --- | --- | ------ | --- | +sklearn.metrics.pairwise.cosine_similarity. +| FindingsOutput. | | | Wedenotethefindingsofeach | | | | | | | | | | | +| --------------- | --- | --- | ------------------------- | --- | --- | --- | ---- | ------------ | --- | ---- | -------- | ------ | --- | +| | | | | | | | This | is converted | | to a | distance | matrix | | +sub-agentas +| | | | | | | | | | | (t) (t) | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ------- | ----- | --------- | --- | +| | | | | | | | D = | 1 − | cos(s | ,s ), | which | is passed | to | +| | | | | | | | ij | | | i j | | | | +o = α (q,d ) = ⟨s ,r ⟩ (1) AgglomerativeClustering with n_clusters=1, +| | | i | i i | i | i | | | | | | | | | +| ----- | ---- | ----------- | --- | -------- | -------- | --- | ----------------- | --- | --- | -------- | --- | --------- | --- | +| | | | | | | | linkage="average" | | | (UPGMA), | | producing | a | +| where | s is | the natural | | language | findings | re- | | | | | | | | +i +completedendrogramoverallcurrentsummaries. +| port from | d | and | r ∈ | [0,1] is | a scalar | rele- | | | | | | | | +| ---------------------- | --- | --- | --------------------- | -------- | -------- | ----- | ------------ | --- | ----- | ------------ | ---------- | -------- | --- | +| | i | | i | | | | | | | | | | | +| | | | | | | | We traverse | the | merge | tree | bottom-up, | greedily | | +| vance/confidencescore. | | | Theseoutputsarepassed | | | | | | | | | | | +| | | | | | | | accumulating | | nodes | into a batch | as | long as | the | +totheSynthesisLayer. +| | | | | | | | total tiktoken | | count | does not | exceed | the target | | +| ------------ | --- | --- | --- | --------- | --- | ----- | -------------- | ---- | --------- | -------- | ------ | ------------ | --- | +| Parallelism. | | All | RAG | loops for | the | docu- | | | | | | | | +| | | | | | | | budget | of B | = 750,000 | tokens. | | Semantically | | +ments are dispatched concurrently via Lang- similarsummariesaregroupedfirst,soeachbatch +Graph’s Send API fan-out, which emits one contains the most similar available summaries +| Send("document_sub_agent_node", | | | | | ...) | per | | | | | | | | +| ------------------------------- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +withinthetokenlimit. +documentandexecutesalltargetsinparallelwithin +LangGraph’sasyncruntime. AggregateandSynthesizeFindings. Foreach +batch,anLLMispromptedwiththeconcatenated +3.3 SynthesisLayer +| | | | | | | | findings, | the synthesis | | directive, | and | the original | | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | ---------- | --- | ------------ | --- | +WhilemodernLLMsfeatureextensivecontextwin- queryq toproduceanaggregatedandsynthesized +dows,synthesizinghundredsofdocumentreports +| | | | | | | | summary. | Allbatchsynthesiscallswithinanitera- | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------ | --- | --- | --- | --- | --- | +can still exceed these limits or cause ’lost in the tionareissuedconcurrentlyviaasyncio.gather. +middle’degradation. Tofuture-proofthearchitec- Themergedsummariesformtheinputforthenext +| ture for | massive | corpora, | | the Synthesis | Layer | is | | | | | | | | +| -------- | ------- | -------- | --- | ------------- | ----- | --- | --------- | ------- | --- | ---- | --------- | ------ | --- | +| | | | | | | | iteration | S(t+1). | The | same | SYNTHESIS | prompt | | +designedasadynamicmap-reducepipelinethatag- +| | | | | | | | used at | every | level | of the | merging | tree is | ap- | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----- | ----- | ------ | ------- | ------- | --- | +gregatesfindingsrecursively. TheSynthesisLayer plied even at the final level—there is no separate +receives the synthesis directives from the Coor- synthesis-onlyprompt. +| dination | Layer | and | the findings | {o | ,o ,...,o | } | | | | | | | | +| -------- | ----- | --- | ------------ | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | 1 2 | n | | | | | | | | +from the Parallel Retrieval Layer. It aggregates ContextSizeConditionandOutput. Thelayer +andsynthesizesthefindingsrecursivelythrougha evaluates whether the total findings size exceeds +similarity-orderedprocessuntilthetotalfindings the context size threshold (i.e., |S(t+1)| > 1). If +sizefallswithinthetargetcontextsizetoproduce theconditionistrue,theprocessloopsbacktothe +thefinaloutput. similarity-orderedmergeandsynthesisstep. Ifthe + +Algorithm1RecursiveSynthesis Locating, Comparison, Clustering, or Chain of +Reasoning,withcontextlengthsrangingfrom10K +| Require: | FindingsS | = {s ,...,s | },queryq, | | | | | | | | | +| -------- | --------- | ----------- | --------- | --- | --------- | ---- | ------- | --- | ----------- | --- | ---------- | +| | | 1 | n | | | | | | | | | +| | | | | | to beyond | 250K | tokens. | | We evaluate | | on the En- | +synthesisdirectived,contextsizebudgetB +Ensure: FinalOutputs⋆ glishandSet4(200k-250ktokens)portionofthe +| | | | | | benchmarkonly. | | Ourevaluationsetcomprises40 | | | | | +| ----------- | ----- | --- | --- | --- | -------------- | --- | --------------------------- | --- | --- | --- | --- | +| 1: while|S| | > 1do | | | | | | | | | | | +instancesinvolvingacademicpapersand62involv- +| 2: E | ← COHEREEMBED(S) | | | | | | | | | | | +| ---- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +D ← 1−COSINESIMILARITY(E) ingfinancialreports(102instancestotal),withthe +3: +tree ← AGGLOMERATIVECLUSTER( fourtasktypesdistributedas: SpotlightLocating +4: +(27),Comparison(15),Clustering(49),andChain +| | D, linkage | = “average”) | | | | | | | | | | +| --- | ---------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ofReasoning(11). +| 5: batches | ← | GROUPBYTOKENS( | | | | | | | | | | +| ---------- | -------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | S, tree, | B) | | | | | | | | | | +4.2 Baselines +| 6: if|batches| | | ≥ |S|then | | | | | | | | | | +| -------------- | --------- | --------- | --- | --- | --------------------- | --- | ------- | --- | --------- | -------- | --- | +| | | | | | We compare | | against | the | following | systems, | all | +| 7: | batches ← | [S] | | | | | | | | | | +| endif | | | | | backedbyGemini2.5Pro: | | | | | | | +8: +| 9: S | ← ASYNCGATHER( | | | | | | | | | | | +| ---- | ------------------- | --- | --- | -------- | ------------------------ | --- | --- | --- | --- | --------------- | --- | +| | | | | | • Baseline(FullContext): | | | | | Alldocumentsfor | | +| | SYNTHESIZE(b,q,d)∀b | | ∈ | batches) | | | | | | | | +eachinstanceareconcatenatedandprovided +10: endwhile +directlyinthecontextwindowofGemini2.5 +| return | S[0] | | | | | | | | | | | +| ------ | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +11: +Pro,representingtheoraclelong-contextup- +perbound. +conditionisfalse(thetotalfindingssizeiswithin +• NormalRAG:Standardvectorsearchtop-K +thecontextsize,leavingasinglefinalsummarys⋆), +| | | | | | retrieval | | over | the full | document | | corpus, fol- | +| ----------- | ---------- | --------------------------- | --- | ------------ | --------- | --- | ---- | --------- | -------- | -------- | ------------ | +| the process | terminates | and yields | s⋆ | as the final | | | | | | | | +| | | | | | lowed | by | LLM | reasoning | | over the | retrieved | +| | If|S(t+1)| | |S(t)|(no-progresscase),the | | | | | | | | | | +| output. | | ≥ | | | | | | | | | | +chunks. +entirecurrentlevelisforcedintoasinglebatchto +guaranteetermination. +| | | | | | • Agentic | | RAG: | A single-agent | | ReAct-style | | +| --- | --- | --- | --- | --- | --------- | --- | ---- | -------------- | --- | ----------- | --- | +Algorithm1givesthepseudocodeforthegeneral RAG system (implemented via Lang- +recursivesynthesisprocedure. Graph/LangChain) that can issue multiple +| | | | | | iterative | | retrieval | calls | over | the | full global | +| ---------------- | --- | ------------------------ | --- | --- | --------- | --- | --------- | ----- | ---- | --- | ----------- | +| BehavioronLoong. | | Inpractice,withGemini2.5 | | | | | | | | | | +Pro’s 1M-token context window and a budget of corpus. It uses the same embedding and +| | | | | | re-rankingpipelineas | | | | SPD-RAG,butlacks | | | +| --- | --- | --- | --- | --- | -------------------- | --- | --- | --- | ---------------- | --- | --- | +B = 750,000tokens,thecombinedsub-agentfind- +per-documentspecialization. +ingsforallinstancesinourLoongevaluationsetfit +withinasinglebatch,sotheconditionalcheckal- +4.3 EvaluationMetrics +| lowedoutputafteroneiteration. | | | Therecursiveloop | | | | | | | | | +| ----------------------------- | --- | --- | ---------------- | --- | --------- | --- | --------- | --- | -------- | --------- | --- | +| | | | | | We report | the | following | | metrics, | following | the | +isthereforeadesigncapabilityintendedformuch +Loongevaluationprotocol(Wangetal.,2024a): +largercorpora,wheretotalfindingssizegenuinely +exceedsthecontextsize. +| | | | | | • Avg | Score: | An | LLM-judged | | score | (0–100) | +| --- | --- | --- | --- | --- | ----- | ------ | --- | ---------- | --- | ----- | ------- | +measuringthedegreetowhichthepredicted +4 ExperimentalSetup +| | | | | | answer | | accurately | covers | | the gold | answer, | +| --- | --- | --- | --- | --- | ------ | --- | ---------- | ------ | --- | -------- | ------- | +4.1 Dataset +| | | | | | following | | the | evaluation | protocol | | of Loong | +| --- | --- | --- | --- | --- | ------------------ | --- | --- | ---------- | -------- | --- | -------- | +| | | | | | (Wangetal.,2024a). | | | | 2 | | | +WeevaluateSPD-RAGontheLoongbenchmark +| (Wangetal.,2024a). | | Loongisanextendedmulti- | | | | | | | | | | +| ------------------ | ------------ | ----------------------- | ------- | -------- | ------------------- | --- | --- | --- | ------------------- | --- | --- | +| | | | | | • PerfectRate(PR%): | | | | Percentageofqueries | | | +| document | QA instances | in both | English | and Chi- | | | | | | | | +wherethesystemreceivesaperfectscoreof +| nese, with | an average | of 11 documents | | per test | | | | | | | | +| ---------- | ---------- | --------------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +100. +| casespanningtworeal-worldscenarios: | | | | financial | | | | | | | | +| ----------------------------------- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +reports and academic papers. Each test case is 2BecausetheoriginallegacyGPT-4modelAPIisunavail- +ablefornewscaledevaluations,weadoptGPT-5astheeval- +constructedsothateverydocumentisrelevantto +| | | | | | uator. Recent | literature | | increasingly | | adopts | GPT-5 as the | +| --- | --- | --- | --- | --- | ------------- | ---------- | --- | ------------ | --- | ------ | ------------ | +the final answer ("Leave No Document Behind", standard for LLM-as-a-judge protocols due to its superior +meaningacorrectresponserequiressynthesizing alignmentwithhumangrading(Wolfsonetal.,2025).While +ourbaselinenumbersarere-computedunderthisGPT-5judge +evidencefromallprovideddocuments),andisan- +forstrictfairness,rawscoresmaynotbeperfectly1-to-1com- +notated under one of four task types: Spotlight parablewiththeoriginallypublishedLoongleaderboard. + +• AvgTokenUsage: Meanper-querytotal(in- Cost tracking. Per-query costs are computed +put+output)tokenusage. fromLangSmithrunmetadata(totaltokencounts +×per-tokenpricing. +| • Avg | Cost | (USD): | Mean | per-query | | API cost, | | | | | | | +| ----------------------------------- | ---- | ------ | ---- | --------- | --- | --------- | --------- | --- | --- | --- | --- | --- | +| enablingcost-efficiencycomparisons. | | | | | | | 5 Results | | | | | | +5.1 MainResults +| • AvgLatency(s): | | | Meanper-querywall-clock | | | | | | | | | | +| ---------------- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +time. +Table1reportsthemainresultsacrossallsystems +| | | | | | | | on our Loong | evaluation | | set. | Figure | 2 illustrates | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ---------- | --- | ---- | ------ | ------------- | +4.4 ImplementationDetails +theseperformancedifferencesacrossthefourspe- +| AllsystemsuseGemini2.5Pro(temperature0.0) | | | | | | | cifictasktypes. | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | +asthebackboneLLMforthecoordinator,merging SPD-RAGachievesanAvgScoreof58.1,sub- +| layer, and | synthesis | | layer. | SPD-RAG | | document | | | | | | | +| ---------- | --------- | --- | ------ | ------- | --- | -------- | --- | --- | --- | --- | --- | --- | +stantiallyoutperformingbothNormalRAG(33.0) +| sub-agents | use | Gemini | 2.5 | Flash | for | document- | | | | | | | +| ---------- | --- | ------ | --- | ----- | --- | --------- | ----------- | --- | ------- | --- | ---------------- | --- | +| | | | | | | | and Agentic | RAG | (32.8). | | This corresponds | to | +scopedretrieval,reducingper-documentcostwhile roughly a 25-point absolute improvement (about +preserving reasoning quality at the merging and 76% relative gain) over standard RAG baselines. +| synthesisstages. | | GPT-5isusedexclusivelyasthe | | | | | | | | | | | +| ---------------- | --- | --------------------------- | --- | --- | --- | --- | ------------ | ------- | --- | ---- | ---- | ----------- | +| | | | | | | | Furthermore, | SPD-RAG | | more | than | doubles the | +LLMjudgefortheevaluationoftheAvgscoreand Perfect Rate (PR) of Agentic RAG (18.6% vs. +is not part of the pipeline itself. Prompts for all 8.8%),indicatingthatexhaustivedocument-level +| systems, | including | baselines, | | were | optimized | for | | | | | | | +| -------- | --------- | ---------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | +processingmorefrequentlycapturesthecomplete +informationextractionandsynthesistoensurefair set of required facts. Notably, the Agentic RAG +comparison. Fullprompttemplatesforallsystems baseline,despiteconsuming∼3×moretokensthan +areprovidedinthecoderepository. NormalRAG,doesnotyieldscoreimprovements, +suggestingthatuninstructediterativeretrievalwith- +| Indexing. | Each | | document | | is | pre- | | | | | | | +| --------- | ---- | --- | --------- | --- | ------ | ---- | ---------------- | --- | -------------- | --- | ---- | ---------- | +| | | | | | | | out per-document | | specialization | | does | not effec- | +| indexed | into | a | dedicated | | Qdrant | vec- | | | | | | | +tivelyaddressthecoveragechallengeinexhaustive +| tor | collection | | using | | LangChain’s | | | | | | | | +| ------------------------------ | ---------- | --- | ----- | --- | ----------- | ------- | -------------- | ------- | ----------- | ----- | ------------ | ------------ | +| | | | | | | | multi-document | | settings. | While | the | full-context | +| RecursiveCharacterTextSplitter | | | | | | (Imple- | | | | | | | +| | | | | | | | baseline | attains | the highest | | quality with | an Avg | +mentationdetailsareprovidedintheAppendix.). +| | | | | | | | Score of | 68.0, | SPD-RAG | reaches | 85.4% | of this | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------- | ------- | ----- | ------- | +Retrievalpersub-agent. Eachsearchtoolcall performancewhileconsumingonly37.9%ofthe +performs dense vector retrieval over the docu- API cost, demonstrating a highly favorable cost- +| | | | | | | | quality trade-off. | | The | full-context | baseline | pro- | +| --------------------------- | --- | --- | --- | --- | -------------- | --- | ------------------ | --- | --- | ------------ | -------- | ---- | +| mentindex,returningthetop-K | | | | | (k = 15)chunks | | | | | | | | +using cosine similarity in Qdrant. These can- vides a strong upper bound that benefits from all +didates are subsequently re-ranked with Cohere documents in one pass. Despite this, SPD-RAG +narrowsthegapto9.9pointswhilecostingless. +| rerank-v4.0-fast,andthetoptop_n | | | | | = | 5chunks | | | | | | | +| ------------------------------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +arepassedtothesub-agentLLM. +5.2 AnalysisbyTaskType +| Merging | layer. | The | token | budget | | per merge | | | | | | | +| ------- | ------ | --- | ----- | ------ | --- | --------- | --- | --- | --- | --- | --- | --- | +Table2breaksdownperformancebyLoong’sfour +| batchisB | = 750,000tokens(75%ofGemini2.5 | | | | | | | | | | | | +| -------- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +tasktypes. +| Pro’s1Mcontextcap). | | | Agglomerativeclustering | | | | | | | | | | +| ------------------- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Thetask-typebreakdownrevealsaclearpattern. +usessklearnwithUPGMAlinkageandprecom- +| | | | | | | | For Spotlight | Locating—which | | | requires | finding | +| --- | --- | --- | --- | --- | --- | --- | ------------- | -------------- | --- | --- | -------- | ------- | +putedcosine-distancematrix. +asinglesalientfactacrossdocuments—allRAG- +| | | | | | | | based systems | | perform | reasonably | well | (69–74), | +| ------------- | --------------- | ----------------------------- | ----- | --------- | ----------- | ---------- | -------------------------------------------- | ------------ | ------- | ---------- | ----- | ----------- | +| Parallelism. | | Documentagentsrunconcurrently | | | | | | | | | | | +| | | | | | | | while the | full-context | | baseline | leads | (86.6). The | +| via LangGraph | | Send | API | fan-out. | Batch | synthe- | | | | | | | +| | | | | | | | largest gaps | between | | SPD-RAG | and | the RAG | +| sis calls | within | each | merge | iteration | | are paral- | | | | | | | +| | asyncio.gather. | | | | | | baselinesappearinClustering(+40.5ptsoverNor- | | | | | | +| lelized | with | | | | Synchronous | Co- | | | | | | | +malRAG)andChainofReasoning(+26.2ptsover +hereAPIcallsareoffloadedtothethreadpoolvia +| | | | | | | | Agentic RAG), | | tasks | that require | aggregating | or | +| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ----- | ------------ | ----------- | --- | +asyncio.to_thread. +reasoningoverevidencefrommanydocumentssi- +Safety limits. Each document sub-agent’s iter- multaneously. Ontheotherhand,itnearlymatches +ative retrieval loop is capped at 5 iterations; the the full-context baseline (42.2 vs. 42.7) on Com- +| AgenticRAGbaselinecapsat10iterations. | | | | | | | parisontasks. | | | | | | +| ------------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | + +System AvgScore PR(%) AvgInputTok AvgTotalTok AvgCost($) AvgLatency(s) +OracleLong-ContextBaseline +| Baseline(FullContext) | | 68.0 | 31.4 253,085 | 255,345 | 0.273 | 45.6 | +| --------------------- | --- | ---- | ------------ | ------- | ----- | ---- | +RAGBaselines +| NormalRAG | | 33.0 | 13.7 22,174 | 27,430 | 0.080 | 42.6 | +| ---------- | --- | ---- | ----------- | ------ | ----- | ---- | +| AgenticRAG | | 32.8 | 8.8 81,877 | 85,290 | 0.098 | 40.6 | +OurSystem +| SPD-RAG | | 58.1 | 18.6 193,954 | 205,683 | 0.103 | 54.8 | +| ------- | --- | ---- | ------------ | ------- | ----- | ---- | +Table1: MainresultsonourLoongevaluationset(102instances: 40academicpaper,62financialreport). Bold: +| bestoverall. | Underline: | bestamongRAG-basedsystems. | | | | | +| ------------ | ---------- | -------------------------- | --- | --- | --- | --- | +Figure2: ComparisonofAverageScoreacrossthefoursystems,brokendownbytasktype(SpotlightLocating, +Comparison,Clustering,andChainofReasoning). +5.3 AnalysisbyDocumentDomain +| As shown | in Table 3 | and Figure | 3 the domain | | | | +| -------- | ---------- | ---------- | ------------ | --- | --- | --- | +breakdownexposesastrikingfailuremodeofstan- +| dard RAG: | both Normal | RAG and | Agentic RAG | | | | +| --------- | ----------- | ------- | ----------- | --- | --- | --- | +achieve0%PRonacademicpaperinstances(Avg +| Scoresof15.2and16.8,respectively). | | | Academic | | | | +| ---------------------------------- | --- | --- | -------- | --- | --- | --- | +paperinstancesinLoongtendtoinvolvelong,tech- +nicaldocumentswithdistributedevidence,making +top-K retrievalespeciallysusceptibletocoverage +failures. SPD-RAGdramaticallyrecoversonthis +domain(60.0AvgScore),closingmostofthegap +| with the | full-context | baseline (78.8). | On finan- | | | | +| -------- | ------------ | ---------------- | --------- | --- | --- | --- | +Figure3: Averagescorebydocumentdomaincompar- +cialreportinstances—whichtendtobeshorterand ingtheBaseline(FullContext),NormalRAG,Agentic +morestructured—allRAG-basedsystemsperform +RAG,andSPD-RAGsystems. +better,andthegapbetweenSPD-RAG(56.9)and +thebaseline(61.0)issmaller. + +| | System | | SpotlightLoc. | | Comparison | Clustering | ChainofReasoning | | | | | +| --- | ------ | --- | ------------- | --- | ---------- | ---------- | ---------------- | ------ | --- | --- | --- | +| | | | (n=27) | | (n=15) | (n=49) | | (n=11) | | | | +Baseline(FullContext) 86.6/55.6% 42.7/26.7% 67.1/14.3% 60.9/54.5% +| | NormalRAG | | 69.7/37.0% | | 37.7/26.7% | 15.4/0.0% | | 14.8/0.0% | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | +| | AgenticRAG | | 73.4/22.2% | | 23.3/20.0% | 16.7/0.0% | | 17.9/0.0% | | | | +| | SPD-RAG | | 74.2/44.4% | | 42.2/20.0% | 57.2/4.1% | | 44.1/18.2% | | | | +Table2: AvgScore/PR(%)byLoongtasktype. EachcellshowsAvgScore/PR%. +| | | | | | Crucially, | thearchitectureof | | | SPD-RAG | | enables | +| -------- | --- | ----------- | --------------- | --- | ---------- | ----------------- | --- | --------------- | ------- | --------- | ------- | +| System | | Paper(n=40) | Financial(n=62) | | | | | | | | | +| | | | | | this | cost efficiency: | | by constraining | | retrieval | to | +| Baseline | | 78.8/30.0% | 61.0/32.3% | | | | | | | | | +NormalRAG 15.2/0.0% 44.5/22.6% isolated,single-documentspaces,wecanreliably +| AgenticRAG | | 16.8/0.0% | 43.1/14.5% | | | | | | | | | +| ---------- | --- | --------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +offloadtheiterativereasoningtoacheapermodel +| SPD-RAG | | 60.0/7.5% | 56.9/25.8% | | | | | | | | | +| ------- | --- | --------- | ---------- | --- | ------- | ----------- | --- | ------------ | --- | ------ | --------- | +| | | | | | (Gemini | 2.5 Flash). | | In contrast, | | global | iterative | +baselinesrequirethereasoningcapacityofafron- +Table3: AvgScore/PR(%)bydocumentdomain. +| | | | | | tier | model (Gemini | 2.5 | Pro) | to navigate | | the fully | +| --- | --- | --- | --- | --- | ---- | ------------- | --- | ---- | ----------- | --- | --------- | +concatenatedcorpus,drivingupcosts. +| System | | AvgScore | AvgCost($) | Score/$ | | | | | | | | +| ---------- | --- | -------- | ---------- | ------- | ------------------- | ---------- | --- | ---- | ------------------- | --- | --------- | +| Baseline | | 68.0 | 0.273 | 249.1 | | | | | | | | +| NormalRAG | | 33.0 | 0.080 | 412.5 | 6 | Discussion | | | | | | +| AgenticRAG | | 32.8 | 0.098 | 334.7 | | | | | | | | +| SPD-RAG | | 58.1 | 0.103 | 564.1 | | | | | | | | +| | | | | | WhereSPD-RAGexcels. | | | | Thetask-typeresults | | | +| | | | | | in Table | 2 confirm | our | core | hypothesis: | | architec- | +Table4: Cost–qualitytradeoff. +turaldecompositionalongthedocumentaxisyields +| | | | | | the largest | gains | precisely | | for tasks | that | require | +| --- | --- | --- | --- | --- | ----------- | -------------- | --------- | --- | ---------- | ---------- | ------- | +| | | | | | exhaustive | multi-document | | | synthesis. | Clustering | | +5.4 Cost–QualityTradeoff +tasks(+40.5overNormalRAG)andChainofRea- +| | | | | | soning | tasks (+26.2 | over | Agentic | | RAG) | both de- | +| --- | --- | --- | --- | --- | ------ | ------------ | ---- | ------- | --- | ---- | -------- | +mandthatthesystemformacoherentanswerfrom +evidencedistributedacrossallrelevantdocuments— +exactlythescenarioforwhichper-documentagents +| | | | | | aredesigned. | ForLoong’sComparisonquestions, | | | | | | +| --- | --- | --- | --- | --- | ------------ | ------------------------------ | --- | --- | --- | --- | --- | +theanswerisprimarilydeterminedbyreadingout +asmallsetofcomparablenumericattributesfrom +| | | | | | each | report and | aggregating | | them, | so once | docu- | +| --- | --- | --- | --- | --- | -------- | -------------- | ----------- | --------- | ----- | ------------ | -------- | +| | | | | | ment | agents have | extracted | | those | fields, | the syn- | +| | | | | | thesizer | is effectively | | operating | over | a structured | | +table,whichhelpsexplainwhySPD-RAG’sscore +onComparisontasks(42.2)nearlymatchesthefull- +Figure4: Cost–Qualitytradeoff. Thescatterplotillus- contextbaseline(42.7). +tratestheParetofrontierformulti-documentQA. +| | | | | | Remaining | gap | with | the | full-context | | baseline. | +| --- | --- | --- | --- | --- | --------- | --- | ---- | --- | ------------ | --- | --------- | +SPD-RAG outperforms the RAG baselines in Despite strong gains over RAG baselines, SPD- +the cost–quality trade-off. It achieves a 76% RAG lags the full-context baseline by 9.9 Avg +higheraveragescorethanNormalRAG(58.1vs. Score points overall. Three factors likely con- +33.0)whileincreasingtheper-querycostbyonly tribute: (1) Gemini 2.5 Flash, used for document +$0.023 (Table 4). In contrast, Agentic RAG is agents due to its cost efficiency, may provide +Pareto-dominated, as it incurs higher cost than weaker agentic reasoning capabilities compared +NormalRAG($0.098vs.$0.080)whiledelivering to larger models. (2) The coordinator’s sub-task +nearlyidenticalquality(32.8vs.33.0). Compared generation may under-specify queries for highly +tothefull-contextbaseline,whichachievesanAvg technical academic content, leading to incom- +Score of 68.0, SPD-RAG reaches 85.4% of the pleteinformationextractionatthedocument-agent +qualityatonly37.9%ofthecost, resultingina level. (3) The maximum context length observed +2.25× improvement in cost–quality efficiency. inLoongquerieswasapproximately250ktokens— + +onlyabout25%ofthe1M-tokencontextwindow onthecoordinator’sabilitytogeneratecomprehen- +of Gemini 2.5 Pro—potentially limiting the pres- sivesub-tasks: under-specifiedtasksleadtoincom- +sureonthefull-contextbaseline. SinceSPD-RAG plete document-level extractions, as observed on +isdesignedtooperateonsubstantiallylargercon- technicalacademicpaperinstances. Ourevaluation +texts,itsadvantagesmaybecomemorepronounced islimitedtotwodomains(academicpapersandfi- +undermoreextremelong-contextconditions. nancialreports)fromtheLoongbenchmark;legal, +medical,orenterprisedocumentcorporamayhave +| RAG baseline | | failure | on papers. | | The 0% | PR | | | | | | | +| ------------ | --- | ------- | ---------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +differentstructuralcharacteristicsthatalterrelative +| rate and | 15.2-16.8% | | Avg Score | of | both Normal | | | | | | | | +| -------- | ---------- | --- | --------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | +performance. +| RAG and | Agentic | RAG | on | academic | paper | in- | | | | | | | +| ------- | ------- | --- | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | +ThesynthesislayerofSPD-RAGisdesignedto +top-K +| stances | highlights | a systematic | | failure | of | | | | | | | | +| ------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +handlecorporawherethecombinedsub-agentout- +| retrieval | under | dense, | distributed | evidence. | | Aca- | | | | | | | +| --------- | ----- | ------ | ----------- | --------- | --- | ---- | --- | --- | --- | --- | --- | --- | +putsexceedthemodel’scontextwindow,triggering +| demic | papers | in Loong | often | require | synthesiz- | | | | | | | | +| ----- | ------ | -------- | ----- | ------- | ---------- | --- | --- | --- | --- | --- | --- | --- | +multi-roundsimilarity-basedclusteringandrecur- +ingfactsspreadacrossabstracts,methods,results, +| | | | | | | | sive synthesis. | | In our Loong | experiments, | | how- | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | ------------ | ------------ | --- | ---- | +andappendicesofmultipledocuments—aregime +ever,Gemini2.5Pro’s1M-tokencontextwassuffi- +| where exhaustive | | per-document | | coverage | proves | | | | | | | | +| ---------------- | --- | ------------ | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | +cienttoprocessallsub-agentsummariesinasingle +| highly | effective. | The | recovery | by SPD-RAG | | to | | | | | | | +| ------ | ---------- | --- | -------- | ---------- | --- | --- | ----------- | --- | -------------- | ---- | --- | ----------- | +| | | | | | | | merge step, | so | this recursive | path | was | never acti- | +60.0AvgScore/7.5%PRinthisdomaindemon- +| | | | | | | | vated. Asaresult,thescalabilitypropertiesofthe | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | +stratesthepracticalvalueofper-documentspecial- +recursivesynthesispipeline—itsimpactonanswer +ization. +| | | | | | | | quality, | redundancy | reduction, | | and cost | as docu- | +| --- | --- | --- | --- | --- | --- | --- | -------- | ---------- | ---------- | --- | -------- | -------- | +Latencyoverhead. SPD-RAG incursamodest mentcountsgrowintothehundredsorthousands— +latency overhead (54.8 s vs. 40.6–45.6 s for the remainempiricallyunvalidated. EvaluatingSPD- +| baselines), | primarily | | due to | its multi-agent | | archi- | | | | | | | +| ----------- | --------- | --- | ------ | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +RAGonmuchlargercorpora,andideallyonanew +tecture. Although document agents run in paral- benchmarkspecificallytargetinghundredstothou- +lel,end-to-endlatencyisstillgatedbythelayered sands of documents per query instance, is an im- +pipeline, which requires at least three sequential portantdirectionforfutureworkthatwouldmore +LLM calls (coordination, per-document retrieval, directlytesttheper-documentagentdesignandthe +andsynthesis)ratherthanasinglepass. intendedlarge-databaseregime. +| | | | | | | | 8 Conclusion | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | -------- | ---------------- | ---------- | ------ | +| | | | | | | | We presented | | SPD-RAG, | a hierarchical | | multi- | +| | | | | | | | agent framework | | designed | for | exhaustive | multi- | +| | | | | | | | documentquestionanswering. | | | Byassigningaded- | | | +icatedagenticRAGmoduletoeachdocumentand +mergingtheirfindingsthroughasimilarity-ordered, +| | | | | | | | token-bounded | | tree, our | system | effectively | miti- | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | ---------- | --------- | --------- | ----------- | ---------- | +| | | | | | | | gatesthelimitationsofbothtraditionaltop-K | | | | | re- | +| | | | | | | | trieval (incomplete | | coverage) | and | isolated | long- | +| | | | | | | | context | processing | (degraded | reasoning | | at scale). | +Figure5: Averageper-querylatencyacrosssystems. ExperimentsonthechallengingLoongbenchmark +demonstratethatSPD-RAGachievesanaverage +scoreof58.1,outperformingNormalRAG(33.0) +7 LimitationsandFutureWork +andAgenticRAG(32.8)byapproximately25ab- +SPD-RAG incurs higher per-query LLM call solute points (76% higher average score), while +countsthansingle-passsystems—onecallperdoc- operating at only 37.9% of the API cost of the +ument agent plus a final synthesis step—which oracle full-context baseline (68.0). These per- +increasestotaltokenusagecomparedtostandard formance gains are most pronounced in tasks re- +RAGbaselines,thoughitremainsatroughlyone- quiring deep cross-document synthesis—such as +thirdthecostofthefull-contextbaseline($0.10vs. Clustering(+40.5points)andChainofReasoning +$0.27perquery)duetotheuseofGemini2.5Flash (+26.2points)—aswellasondenseacademicpa- +fordocumentsub-agents. Systemqualitydepends perswherestandardRAGmethodsentirelyfail. + +Ultimately, our findings demonstrate that for TaichengGuo,XiuyingChen,YaqiWang,RuidiChang, +complex information-seeking queries over large ShichaoPei,NiteshV.Chawla,OlafWiest,andXi- +| | | | | | | | angliangZhang.2024. | | | Largelanguagemodel-based | | | +| -------- | --- | ----------- | --- | ------------ | --- | -------- | ------------------- | --- | ------------------------------- | ------------------------ | --- | --- | +| corpora, | how | information | | is processed | is | crucial. | | | | | | | +| | | | | | | | multi-agents: | | Asurveyofprogressandchallenges. | | | | +Specifically,ensuringthateachdocumentreceives +| | | | | | | | Preprint,arXiv:2402.01680. | | | ArXiv:2402.01680. | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | ----------------- | --- | --- | +exhaustive,dedicatedagenticattentionprovestobe +| | | | | | | | Siwei Han, | Peng | Xia, Ruiyi | Zhang, | Tong | Sun, Yun | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---- | ---------- | ------ | ---- | -------- | +amoreeffective,cost-efficient,andscalablestrat- +| | | | | | | | Li, Hongtu | Zhu, | and | Huaxiu Yao. | 2025. | MDocA- | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---- | --- | ----------- | ----- | ------ | +egythansimplyexpandinghowmuchrawcontext +gent: Amulti-modalmulti-agentframeworkfordoc- +asinglemodelcanconsumeinonepass. umentunderstanding. Preprint,arXiv:2503.13964. +ArXiv:2503.13964. +9 EthicsStatement +| | | | | | | | Yubin Kim, | Ken | Gu, Chanwoo | Park, | Chunjong | Park, | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ----------- | ----- | -------- | ----- | +SPD-RAG is a retrieval-augmented generation SamuelSchmidgall,A.AliHeydari,YaoYan,Zhi- +hanZhang,YuchenZhuang,MarkMalhotra,PaulPu +| framework | intended | | to improve | access | | to infor- | | | | | | | +| --------- | -------- | --- | ---------- | ------ | --- | --------- | --- | --- | --- | --- | --- | --- | +Liang,HaeWonPark,YuzheYang,XuhaiXu,Yilun +| mation | distributed | across | | large document | | collec- | | | | | | | +| ------ | ----------- | ------ | --- | -------------- | --- | ------- | ----------- | --- | ---------- | -------- | ------ | ------- | +| | | | | | | | Du, Shwetak | | Patel, Tim | Althoff, | Daniel | McDuff, | +tions. ThesystemreliesoncommercialLLMAPIs and Xin Liu. 2025. Towards a science of scal- +(Gemini2.5,Cohere),andaswithanyLLM-based ing agent systems. Preprint, arXiv:2512.08296. +ArXiv:2512.08296. +| pipeline, | outputs | may | contain | factual | errors | and | | | | | | | +| --------- | ------- | --- | ------- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | +should not be treated as authoritative without hu- PatrickLewis,EthanPerez,AleksandraPiktus,Fabio +man verification. The increased number of API Petroni,VladimirKarpukhin,NamanGoyal,Hein- +| | | | | | | | richKüttler, | | MikeLewis, | Wen-tauYih, | | TimRock- | +| --------- | ------ | -------- | ------------ | ----------- | ------- | --------- | ------------------- | --------- | ---------- | ----------- | ----- | ------------ | +| calls per | query | relative | to | single-pass | systems | re- | | | | | | | +| | | | | | | | täschel, | Sebastian | Riedel, | and | Douwe | Kiela. 2020. | +| sults in | higher | energy | consumption, | | a | cost that | | | | | | | +| | | | | | | | Retrieval-augmented | | | generation | for | knowledge- | +should be weighed against the accuracy gains in intensiveNLPtasks. InAdvancesinNeuralInforma- +deploymentdecisions. Ourbenchmarkevaluations tionProcessingSystems. +| are conducted | | on publicly | | available | datasets | and | | | | | | | +| ------------- | --- | ----------- | --- | --------- | -------- | --- | -------- | -------- | ------------- | --- | ------------- | --- | +| | | | | | | | TengLin, | YuyuLuo, | HonglinZhang, | | JichengZhang, | | +involvenohumansubjectsorpersonaldata. Dur- Chunlin Liu, Kaishun Wu, and Nan Tang. 2025. +ingthepreparationofthismanuscript,AI-assisted MEBench: Benchmarkinglargelanguagemodelsfor +| | | | | | | | cross-documentmulti-entityquestionanswering. | | | | | In | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | +writingtoolswereusedtohelprefinelanguageand +ProceedingsoftheConferenceonEmpiricalMethods +| improveclarity. | | Allresearchdesign,experiments, | | | | | | | | | | | +| --------------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +inNaturalLanguageProcessing(EMNLP). +analysis,andfinaldecisionsregardingthecontent +NelsonF.Liu,KevinLin,JohnHewitt,AshwinParan- +wereconductedandverifiedbytheauthors. +jape,MicheleBevilacqua,FabioPetroni,andPercy +| | | | | | | | Liang.2023. | | Lostinthemiddle: | | Howlanguagemod- | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | ---------------------------- | --- | --------------- | --- | +| | | | | | | | elsuselongcontexts. | | TransactionsoftheAssociation | | | | +References +forComputationalLinguistics,12. +| Haonan | Bian. | 2025. | LLM-empowered | | | knowl- | | | | | | | +| ------ | ----- | ----- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | +SeijiMaekawa,HayateIso,andNikitaBhutani.2024. +| edge | graph | construction: | | A survey. | | Preprint, | | | | | | | +| ----------------- | ----- | ------------- | ----------------- | --------- | --- | --------- | ------------------------------------- | --- | --- | --- | --- | ------- | +| | | | | | | | Holisticreasoningwithlong-contextLMs: | | | | | Abench- | +| arXiv:2510.20345. | | | ArXiv:2510.20345. | | | | | | | | | | +markfordatabaseoperationsonmassivetextualdata. +Darren Edge, Ha Trinh, Newman Cheng, Joshua Preprint,arXiv:2410.11996. ArXiv:2410.11996. +| Bradley, | Alex | Chao, | | Apurva | Mody, | Steven | | | | | | | +| -------- | ---- | ----- | --- | ------ | ----- | ------ | -------- | ------ | ----- | ---------- | --- | ------------ | +| | | | | | | | David J. | Moore. | 2025. | A taxonomy | of | hierarchical | +Truitt, Dasha Metropolitansky, Robert Osazuwa multi-agentsystems: Designpatterns,coordination +| Ness, | and | Jonathan | Larson. | 2024. | | From lo- | | | | | | | +| --------------------- | ------- | -------- | -------------------------- | -------- | --- | --------- | ----------------- | --- | ----------------- | ------------- | --- | --------- | +| | | | | | | | mechanisms, | | and industrial | applications. | | Preprint, | +| cal to | global: | A graph | RAG | approach | | to query- | | | | | | | +| | | | | | | | arXiv:2508.12683. | | ArXiv:2508.12683. | | | | +| focusedsummarization. | | | Preprint,arXiv:2404.16130. | | | | | | | | | | +ArXiv:2404.16130. +| | | | | | | | LituOuandMirellaLapata.2025. | | | | Context-awarehier- | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ------------------ | --- | +archicalmergingforlongdocumentsummarization. +| Gemini | Team, | Petko | Georgiev, | Ving | Ian Lei, | Ryan | | | | | | | +| -------- | ----- | ----- | --------- | ------- | -------- | ------- | -------------------------- | --- | --- | ----------------- | --- | --- | +| | | | | | | | Preprint,arXiv:2502.00977. | | | ArXiv:2502.00977. | | | +| Burnell, | Libin | Bai, | Anmol | Gulati, | Garrett | Tanzer, | | | | | | | +DamienVincent,ZhufengPan,ShiboWang,and1 Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh +others. 2024. Gemini 1.5: Unlocking multimodal Khanna, Anna Goldie, and Christopher D. Man- +understandingacrossmillionsoftokensofcontext. ning. 2024. RAPTOR: Recursive abstractive pro- +Preprint,arXiv:2403.05530. ArXiv:2403.05530. cessing for tree-organized retrieval. Preprint, +| | | | | | | | arXiv:2401.18059. | | ArXiv:2401.18059. | | | | +| ---------- | ------ | --- | ------- | ------- | ----- | --- | ----------------- | --- | ----------------- | --- | --- | --- | +| Jiani Guo, | Zuchao | Li, | Jie Wu, | Qianren | Wang, | Yun | | | | | | | +Li, Lefei Zhang, Hai Zhao, and Yujiu Yang. 2025. LiSun,LiuHe,ShuyueJia,YangfanHe,andChenyu +ToM:Leveragingtree-orientedMapReduceforlong- You. 2025. DocAgent: An agentic framework for +contextreasoninginlargelanguagemodels. InPro- multi-modallong-contextdocumentunderstanding. +ceedingsoftheConferenceonEmpiricalMethodsin InProceedingsoftheConferenceonEmpiricalMeth- +NaturalLanguageProcessing(EMNLP). odsinNaturalLanguageProcessing(EMNLP). + +V.A.Traag,L.Waltman,andN.J.vanEck.2019. From extraction tasks and write a synthesis +louvaintoleiden: Guaranteeingwell-connectedcom- directive to guide the downstream +| munities. | ScientificReports,9(1):5233. | | | | | | synthesizer. | | | | | | +| --------- | ---------------------------- | --- | --- | --- | --- | --- | ------------- | --- | ------ | --- | ---- | -------- | +| | | | | | | | 2. Sub-agents | | (next) | — | Each | document | +Minzheng Wang, Longze Chen, Cheng Fu, Shengyi is assigned one worker that runs your +Liao,XinghuaZhang,BingliWu,HaiyangYu,Nan tasks independently. Workers search +Xu,LeiZhang,RunLuo,YunshuiLi,MinYang,Fei their document and report raw findings. +| | | | | | | | They run | in parallel | | and | cannot | see each | +| ------------------------- | --- | --- | --- | --------------- | --- | --- | -------- | ----------- | --- | --- | ------ | -------- | +| Huang,andYongbinLi.2024a. | | | | Leavenodocument | | | | | | | | | +other’s results. +| behind: | Benchmarkinglong-contextLLMswithex- | | | | | | | | | | | | +| ------------------ | ----------------------------------- | --- | -------------------------- | --- | --- | --- | -------------- | -------- | ------ | ------ | ------------ | -------- | +| | | | | | | | 3. Synthesizer | | (last) | — | A downstream | | +| tendedmulti-docQA. | | | Preprint,arXiv:2406.17419. | | | | | | | | | | +| | | | | | | | agent | receives | ALL | worker | | findings | +ArXiv:2406.17419. +| | | | | | | | and merges | | them | into | a | single | +| ---------- | ------ | --- | ----- | -------- | ----- | ------ | ------------------- | --------- | ---- | --------- | ------- | ------- | +| | | | | | | | coherent | response. | | It | follows | your | +| Qian Wang, | Tianyu | | Wang, | Zhenheng | Tang, | Qinbin | | | | | | | +| | | | | | | | synthesis_directive | | | to decide | | what to | +Li,NuoChen,JingshengLiang,andBingshengHe. +| | | | | | | | prioritize | and | how | to | structure | the | +| ------ | ---------- | --- | -------------------------- | --- | --- | --- | ---------- | --- | --- | --- | --------- | --- | +| 2024b. | Megaagent: | | Alarge-scaleautonomousLLM- | | | | | | | | | | +output. +basedmulti-agentsystemwithoutpredefinedSOPs. +| | | | | | | | You CANNOT | see | the | names, | types, | or | +| -------------------------- | --- | --- | --- | ----------------- | --- | --- | ---------- | ------ | ---------- | ------ | ------ | --- | +| Preprint,arXiv:2408.09955. | | | | ArXiv:2408.09955. | | | | | | | | | +| | | | | | | | content | of the | documents. | | | | +TomerWolfson,HarshTrivedi,MorGeva,YoavGold- Your ONLY job in this step: +| berg, | Dan Roth, | Tushar | Khot, | Ashish | | Sabharwal, | | | | | | | +| ------------------ | --------- | ------ | -------------------------- | ------- | --------- | ---------- | ---------- | ---------- | ---------------------- | ----------- | -------------- | ------- | +| | | | | | | | • Produce | | a list | of | subagent_todos | | +| and Reut | Tsarfaty. | | 2025. | Monaco: | | More nat- | | | | | | | +| | | | | | | | (as | structured | | output): | | precise | +| ural and | complex | | questions | for | reasoning | across | | | | | | | +| | | | | | | | extraction | | tasks | | that | will be | +| dozensofdocuments. | | | Preprint,arXiv:2508.11133. | | | | | | | | | | +| | | | | | | | executed | | independently | | against | EACH | +| ArXiv:2508.11133. | | | | | | | document | | by the | sub-agents. | | | +| | | | | | | | • Produce | | a synthesis_directive: | | | a | +Wentao Zhang, Liang Zeng, Yuzhen Xiao, Yong- concise instruction (2–4 sentences) +cong Li, Ce Cui, Yilei Zhao, Rui Hu, Yang Liu, for the synthesizer. +| Yahui | Zhou, | and Bo | An. | 2025. | AgentOrchestra: | | | | | | | | +| ----- | ----- | ------ | --- | ----- | --------------- | --- | --- | --- | --- | --- | --- | --- | +A hierarchical multi-agent framework for general- Todo-writing rules: +| purpose | task | solving. | Preprint, | arXiv:2506.12508. | | | | | | | | | +| ----------------- | ---- | -------- | --------- | ----------------- | --- | --- | ----------- | --- | ----------- | ---- | ------------- | ---- | +| | | | | | | | • Decompose | | the | user | query | into | +| ArXiv:2506.12508. | | | | | | | concrete | | information | | requirements. | | +Jun Zhao, Can Zu, Hao Xu, Yi Lu, Wei He, Yiwen • Each todo must be self-contained +and unambiguous. +| Ding, | Tao Gui, | Qi | Zhang, | and | Xuanjing | Huang. | | | | | | | +| ----- | ---------- | --- | --------------------------- | --- | -------- | ------ | -------- | --- | ------ | ----- | ---- | ----- | +| | | | | | | | • Prefer | | atomic | tasks | over | broad | +| 2024. | LongAgent: | | Scalinglanguagemodelsto128k | | | | | | | | | | +tasks. +| contextthroughmulti-agentcollaboration. | | | | | | Preprint, | | | | | | | +| --------------------------------------- | ----- | --- | ----------------- | ----- | ---- | --------- | --------- | --- | --------------- | ------------ | ------------ | -------- | +| | | | | | | | • Include | | coverage | for: | definitions, | | +| arXiv:2402.11550. | | | ArXiv:2402.11550. | | | | | | | | | | +| | | | | | | | numeric | | values, | constraints, | | edge | +| | | | | | | | cases, | | error | modes, | and | any | +| Zihan Zhou, | Chong | | Li, Xinyi | Chen, | Shuo | Wang, | | | | | | | +| | | | | | | | explicit | | recommendations | | | required | +YuChao,ZhiliLi,HaoyuWang,QiShi,ZhixingTan, +| | | | | | | | by | the user | query. | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ------ | --- | --- | --- | +XuHan,XiaodongShi,ZhiyuanLiu,andMaosong +| | | | | | | | • Design | a | robust | extraction | | list | +| ---------- | ---------- | -------------- | ----- | ----- | ---------- | ------- | -------- | ----- | ------ | ------------ | --- | ------ | +| Sun. 2025. | | LLM×MapReduce: | | | Simplified | long- | | | | | | | +| | | | | | | | that | works | for | ANY document | | in the | +| sequence | processing | | using | large | language | models. | | | | | | | +set. +InProceedingsoftheAnnualMeetingoftheAssocia- +| tionforComputationalLinguistics(ACL). | | | | | | | Important | Constraints: | | | | | +| ---------------------------------------- | --- | ----- | --------- | --- | --- | ---------- | --------- | ------------ | ----- | -------- | ------- | -------- | +| | | | | | | | Do | not assume | any | document | | contains | +| AndrewZhu,AlyssaHwang,LiamDugan,andChris | | | | | | | • | | | | | | +| | | | | | | | the | answer. | Write | | todos | that can | +| Callison-Burch. | | 2024. | FanOutQA: | | A | multi-hop, | | | | | | | +| | | | | | | | be | answered | with | either | “Found” | or | +multi-documentquestionansweringbenchmarkfor +| | | | | | | | “Not | found | in | this document.” | | | +| -------------- | --- | ------- | --- | ----------- | --- | ---------- | ------ | ----- | ------ | --------------- | ----------- | --- | +| large language | | models. | In | Proceedings | | of the An- | | | | | | | +| | | | | | | | • Tell | the | worker | WHAT | to extract, | not | +nualMeetingoftheAssociationforComputational +| | | | | | | | HOW | to extract | | it. | | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | +Linguistics(ACL). +| | | | | | | | • Do | not | synthesize, | | summarize, | or | +| --- | --- | --- | --- | --- | --- | --- | ------- | --- | ----------- | --- | ---------- | ----- | +| | | | | | | | attempt | | to answer | | the user | query | +A Prompts +yourself. +| All | prompts | | are | | defined | in | | | | | | | +| --- | ------- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +A.2DocumentSub-AgentPrompt +backend/core/prompts.py. +(RESEARCH_SYSTEM_PROMPT) +| A.1CoordinatorPrompt | | | | | | | You are | a sub-agent | | investigating | | a | +| ------------------------ | --- | --- | --- | --- | --- | --- | ---------------- | ----------- | ------------ | ------------- | --- | --- | +| (LEAD_RESEARCHER_PROMPT) | | | | | | | single document: | | {file_name}. | | | | +You are a lead researcher coordinating You operate inside an iterative +a RAG-based analysis to answer a user retrieval loop. On each turn you output +| query. | | | | | | | exactly | one structured | | action: | | | +| -------- | --- | ------------- | --- | --- | --- | --- | ------------------ | -------------- | --- | ------- | --- | ------- | +| Pipeline | | Architecture: | | | | | • action="search": | | | Issue | a | focused | +1. YOU (now) — Decompose the query into query to retrieve information from + +the document. Set query to a • Follow the synthesis directive +specific, targeted search string. above — it defines your main goal, +Set reasoning to why this query is priorities, and output structure. +| | needed. | | | | | | • | Keep only | information | | that | directly | +| --- | ------- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ---- | -------- | +• action="finalize": You have helps answer the query. Discard +| | gathered | | sufficient | evidence | to | | | tangential | content. | | | | +| --- | -------- | --- | ---------- | -------- | ---------- | --- | --- | ------------ | ------------ | -------- | ------ | ------ | +| | address | ALL | assigned | | tasks. Set | | | | | | | | +| | | | | | | | • | Preserve | exact | numbers, | | names, | +| | findings | to | your | complete | extracted | | | | | | | | +| | | | | | | | | dates, | and caveats. | | | | +| | report. | Set | reasoning | | to a brief | | | | | | | | +| | | | | | | | • | Remove | redundancy. | | If the | same | +| | summary | of | what you | found. | | | | | | | | | +| | | | | | | | | fact appears | | multiple | times, | keep | +it once. +| | Investigation | principles: | | | | | | | | | | | +| --- | ------------- | -------------- | ---- | ----- | --------- | --- | --- | ------- | -------- | --------- | ----- | -------- | +| | | | | | | | • | Do NOT | invent | or infer | facts | not | +| | | | | | | | | present | in the | findings. | | | +| | 1. Start | by identifying | | the | key facts | | | | | | | | +| | required | for | each | task. | | | • | Same | language | as | the | findings | +2. Issue ONE focused query per turn (default English if mixed). +| | — prefer | specific | | terms | over broad | | | | | | | | +| --- | --------- | -------- | -------- | ----- | ----------- | --- | --------------------- | --- | --- | --- | --- | --- | +| | phrases. | | | | | B | ImplementationDetails | | | | | | +| | 3. Expand | to | synonyms | or | paraphrases | | | | | | | | +Models. +if initial results are incomplete. The coordinator, the merging layer, +4. Issue separate queries for and the synthesis layer use gemini-2.5-pro +| | different | | aspects | of a | task. | | | | | | | | +| --- | --------- | --------- | ------- | ----- | ------- | ------------ | --- | ----- | --- | -------- | --- | ---------- | +| | | | | | | (temperature | | 0.0). | | Document | | sub-agents | +| | 5. Keep | searching | until | every | task is | | | | | | | | +covered with concrete evidence or use gemini-2.5-flash (temperature 0.0) for +| | confirmed | | absent. | | | document-scoped | | | retrieval. | | The | LLM judge | +| --- | --------- | --- | ------- | --- | --- | --------------- | --- | --- | ---------- | --- | --- | --------- | +6. Do NOT attempt to answer the user (GPT-5) is separate and is used only for the +| | query | directly | — | extract | raw facts | | | | | | | | +| --- | ----- | -------- | --- | ------- | --------- | ------------------------ | --- | --- | --- | ----------------- | --- | --- | +| | | | | | | evaluationoftheAvgscore. | | | | Modelconstantsare | | | +only. +definedinbackend/shared/constants.py. +| | 7. Do | NOT finalize | | a task | as ’Not | | | | | | | | +| --- | ----------------- | ------------ | ------------ | ---------- | ----------- | ----------------------------- | --- | --- | --- | ------------------------- | --- | ----------- | +| | found’ | after | only | a single | focused | | | | | | | | +| | search. | | At | least | attempt 2 | Hyperparameters. | | | | | | | +| | focused | searches | | before | concluding | | | | | | | | +| | information | | is absent. | | | | | | | | | | +| | | | | | | Parameter | | | | Value | | | +| | 8. Do NOT | focus | on very | specific | terms, | | | | | | | | +| | | | | | | Chunksize(tokens) | | | | 1000 | | | +| | but | rather | on the | general | context | | | | | | | | +| | | | | | | Chunkoverlap(tokens) | | | | 250 | | | +| | of the | task. | Always | try | to find the | | | | | | | | +| | | | | | | Tokenizer | | | | tiktoken | | cl100k_base | +| | most | relevant | information. | | | | | | | | | | +| | | | | | | Embeddingmodel | | | | Cohereembed-v4.0 | | | +| | 9. You | must use | AT MOST | 5 | searches to | | | | | | | | +| | | | | | | Embeddingdimension | | | | 1536 | | | +| | answer | a task. | DO | NOT | OVERUSE THE | | | | | | | | +| | | | | | | Similaritymetric | | | | Cosine | | | +| | SEARCH | TOOL. | | | | | | | | | | | +| | | | | | | MergetokenbudgetB | | | | 750,000tokens | | | +| | | | | | | Denseretrievaltop-K | | | | 15 | | | +| | Finalize | report | format: | | For each | | | | | | | | +| | | | | | | Reranker | | | | Coherererank-v4.0-fast | | | +| | assigned | task, | output | either | “Found: | | | | | | | | +| | | | | | | Rerankertop-N | | | | 5 | | | +| | [exact answer | with | supporting | | evidence]” | | | | | | | | +| | | | | | | Sub-agentiterationsafetylimit | | | | 5 | | | +| | or “Not found | in | this | document.” | Always | | | | | | | | +| | | | | | | Clusteringalgorithm | | | | Agglomerative(UPGMA) | | | +| | report exact | numbers, | | names, | and dates. | | | | | | | | +| | | | | | | Clusteringmetric | | | | Precomputedcosinedistance | | | +| | Never approximate | | or | infer | beyond the | | | | | | | | +document. +| | | | | | | Cost | | measurement. | | Per-query | | costs | +| --- | --- | --- | --- | --- | --- | ---- | ------ | ------------ | --------- | --------- | --- | -------- | +| | | | | | | are | pulled | from | LangSmith | | run | metadata | +A.3Merge/SynthesisPrompt +| (SYNTHESIS_PROMPT) | | | | | | (extract_langsmith_info.py) | | | | | using | model- | +| ------------------ | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --- | --- | ----- | ------ | +specificper-tokenpricing. +Thispromptisuseduniformlyateverylevelofthe +recursivemergingtree,includingthefinalsynthesis +step. Thereisnoseparatefinal-answerprompt. +| | You are | a research | synthesizer. | | Merge | | | | | | | | +| --- | ------------- | ----------------- | ------------ | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +| | the following | | sub-agent | findings | into | | | | | | | | +| | one compact, | information-dense | | | response | | | | | | | | +| | for the user. | | | | | | | | | | | | +| | Query: {query | } | | | | | | | | | | | +| | Synthesis | Directive: | | | | | | | | | | | +{synthesis_directive} +| | Findings | Batch: | {findings | } | | | | | | | | | +| --- | -------- | ------ | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Rules: \ No newline at end of file diff --git a/q126/random_k4/random_2.md b/q126/random_k4/random_2.md index 5d5d4bf748b05f616c8399e8fe70690f3bccb9c7..f2e0cee911212728b5cd229fe67ae344991a2f8f 100644 --- a/q126/random_k4/random_2.md +++ b/q126/random_k4/random_2.md @@ -1,1678 +1,3886 @@ -| | | Learning | | Fine-Grained | | Grounded | Citations | | for | | | -| --- | --- | -------- | ---------- | ------------ | ----- | -------- | --------- | --- | --- | --- | --- | -| | | | Attributed | | Large | Language | Models | | | | | -LeiHuang1,XiaochengFeng1,2*,WeitaoMa1,YuxuanGu1,WeihongZhong1,XiachongFeng1 -WeijiangYu3,WeihuaPeng3,DuyuTang3,DandanTu3,BingQin1,2 -1HarbinInstituteofTechnology,Harbin,China -| | | 2 PengChengLaboratory | | | | 3HuaweiInc.,Shenzhen,China | | | | | | -| --- | --- | --------------------- | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | -{lhuang,xcfeng,wtma,yxgu,whzhong,xiachongfeng,qinb}@ir.hit.edu.cn -{weijiangyu8,pengwh.hit}@gmail.com,{tangduyu,tudandan}@huawei.com -Abstract -Why is it bad to eat cookie dough for risk of salmonella but things -like Cookie Dough Bites are ok? -| Despite | the | impressive | performance | | on | | | | | | | -| ------- | --- | ---------- | ----------- | --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Salmonella is most commonly caused by eating undercooked | | | | | 1 | -information-seekingtasks,largelanguagemod- or raw foods like eggs or meat. (...) You know how your mom -4202 guA 8 ]LC.sc[ 1v86540.8042:viXra els (LLMs) still struggle with hallucinations. always warned you not to eat raw cookie dough? This is why. -| Attributed | LLMs, | which | augment | generated | | | | | | | | -| ---------- | ----- | ----- | ------- | --------- | --- | --------------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Food safety recommendations encourage people to cook eggs until | | | | | 2 | -text within-line citations, haveshownpoten- the white and yolk are firm in order to kill any bacteria. (...) However, -tial in mitigating hallucinations and improv- anyone making cookies can do things to reduce this risk by -using pasteurized egg products. -| ingverifiability. | | However,currentapproaches | | | | | | | | | | -| ----------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -suffer from suboptimal citation quality due Raw cookie dough can carry the risk of salmonella contamination, which can -to their reliance on in-context learning. Fur- lead to foodborne illness if the dough is eaten raw  1  . However, products -like Cookie Dough Bites are generally safe because they are made with -| thermore, | the | practice | of citing | only | coarse | pasteurized eggs  1    2  . | | | | | | -| --------- | --- | -------- | --------- | ---- | ------ | --------------------------- | --- | --- | --- | --- | --- | -documentidentifiersmakesitchallengingfor -Difficult to verify and subject to -userstoperformfine-grainedverification. In hallucinations and citation errors -| this work, | we | introduce | FRONT, | a | training | | | | | | | -| ---------- | --- | --------- | ------ | --- | -------- | --- | --- | --- | --- | --- | --- | - G  R  O   U  N  D  I N  G     1  salmonella is most commonly caused by eating -framework designed to teach LLMs to gen- undercooked or raw foods like eggs or meat  2  Anyone making cookies -can do things to reduce this risk by using pasteurized egg products. -erateFine-gRainedgrOuNdedciTations. By  A  N  S  W   E  R    Eating raw cookie dough can be risky due to salmonella, -groundingmodeloutputsinfine-grainedsup- commonly found in raw eggs  1 . However, this concern is mitigated in -products like Cookie Dough Bites, as the products use pasteurized egg -portingquotes,thesequotesguidethegenera- -products, which are safe to eat raw 2  . -tionofgroundedandconsistentresponses,not -onlyimprovingcitationqualitybutalsofacili- Interpretable and accurately cited -tatingfine-grainedverification.Experimentson -| | | | | | | Figure1: | Comparedwiththecurrentattributedsystems, | | | | | -| --- | --- | --- | --- | --- | --- | -------- | ---------------------------------------- | --- | --- | --- | --- | -theALCEbenchmarkdemonstratetheefficacy -thecoreideabehindFRONTistofirstselectthesupport- -ofFRONTingeneratingsuperiorgroundedre- -| | | | | | | ing quotes | from | retrieved | sources and | then | condition | -| ------------------------------------ | --- | --- | --- | --- | ---- | ---------- | ---- | --------- | ----------- | ---- | --------- | -| sponsesandhighlysupportivecitations. | | | | | With | | | | | | | -thegenerationprocessonthem,ensuringgroundedre- -LLaMA-2-7B,theframeworksignificantlyout- -sponsesandaccuratecitations. -| performsallthebaselines, | | | achievinganaver- | | | | | | | | | -| ------------------------ | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -ageof14.21%improvementincitationquality -acrossalldatasets,evensurpassingChatGPT1. SuchprevalenceofhallucinationsinLLMout- -| | | | | | | puts has | motivated | the | development | of | attributed | -| --- | --- | --- | --- | --- | --- | -------- | --------- | --- | ----------- | --- | ---------- | -1 Introduction -| | | | | | | systems | (Nakano | et al., | 2021; Thoppilan | | et al., | -| --- | --- | --- | --- | --- | --- | ------- | ------- | ------- | --------------- | --- | ------- | -Therecentadventoflargelanguagemodels(LLMs) 2022;Menicketal.,2022),suchasNewBing2 and -(Touvronetal.,2023;OpenAI,2023;Zhaoetal., Perplexity3,whereLLMsareallowedtogenerate -2023) has taken the world by storm, fueling a responses with in-line citations. Not only does it -| paradigm | shift in | information | | acquisition | (Zhu | | | | | | | -| -------- | -------- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | -improvefactualityandalleviatehallucinations,but -etal.,2023). Despitetheircompellingperformance, italsosimplifiesuserverificationofmodeloutputs, -LLMs still struggle with hallucinations (Ji et al., furtherenhancingtheverifiabilityofLLMs. -2023;Huangetal.,2023),atendencytofabricate Despiterecentadvancements,currentattributed -| non-existent | facts | or generate | unfaithful | | content. | | | | | | | -| ------------ | ----- | ----------- | ---------- | --- | -------- | -------------------------------------- | --- | --- | --- | --- | -------- | -| | | | | | | LLMsstillexposesignificantlimitations. | | | | | Firstly, | -This issue further poses a risk of misinformation current approaches predominantly rely on either -dissemination(ChenandShu,2023),directlyim- in-contextlearning(Gaoetal.,2023b)orpost-hoc -pactingthereliabilityandtrustworthinessofLLMs. -retrieval(Gaoetal.,2023a)toachieveattribution, -*CorrespondingAuthor -1Our data and code can be found at: https://github. 2https://www.bing.com/chat -com/LuckyyySTA/Fine-grained-Attribution. 3https://www.perplexity.ai - -| | | | | | | | 2 RelatedWork | | | | | | | -| ------- | ----------- | ----------- | --- | ---------- | --- | ------ | ------------- | --- | --- | --- | --- | --- | --- | -| lacking | an inherent | attribution | | capability | | within | | | | | | | | -LLMs,therebyresultingincompromisedcitation -| | | | | | | | RetrievalAugmentedGeneration. | | | | | Recently,re- | | -| ----------------------- | --------- | --- | ----------------------- | ---- | --------- | ---- | ----------------------------- | --- | ---------- | --- | ----- | ------------ | --- | -| quality(Liuetal.,2023). | | | Secondly,thesecitations | | | | | | | | | | | -| | | | | | | | trieval augmented | | generation | | (RAG) | (Karpukhin | | -| are typically | presented | | in the | form | of either | doc- | | | | | | | | -ument identifiers (Nakano et al., 2021) or URLs etal.,2020;Lewisetal.,2020;Fengetal.,2023; -Gaoetal.,2023c)hasshownpromiseinknowledge- -| (Thoppilan | et al., | 2022). | Such | coarse | attribution | | | | | | | | | -| ---------- | ------- | ------ | ---- | ------ | ----------- | --- | --------- | ------ | ---------------- | --- | --------- | --- | ----- | -| | | | | | | | intensive | tasks. | By incorporating | | retrieved | | docu- | -complicatestheprocessforuserstoperformfine- -| | | | | | | | ments, | LLMsareequippedwithup-to-dateinfor- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------------------------- | --- | --- | --- | --- | --- | -grainedverification(e.g.,pinpointspecificsupport- -| | | | | | | | mation, | significantly | mitigating | | knowledge | | gaps. | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | ---------- | --- | --------- | --- | ----- | -ingevidence),particularlyinlengthydocuments. -| | | | | | | | However, | recent | studies | (Shi | et al., | 2023; | Yoran | -| ------- | ---- | ------ | ---------- | --- | ---------- | ---- | -------- | ------ | ------- | ---- | ------- | ----- | ----- | -| To this | end, | we aim | to advance | | attributed | text | | | | | | | | -etal.,2023;Xuetal.,2023a;Zhuetal.,2024)have -generationbyempoweringLLMswithfine-grained -attributionability. However,onechallengecomes revealed that existing retrieval-augmented LLMs -| | | | | | | | struggle | to handle | irrelevant | | or contradictory | | re- | -| --- | --- | --- | --- | --- | --- | --- | -------- | --------- | ---------- | --- | ---------------- | --- | --- | -fromtheacquisitionofhigh-qualityattributiondata -trievaldocumentsandeffectivelyutilizecontextual -| for supervised | | fine-tuning, | | which | is difficult | and | | | | | | | | -| ------------------------------------ | --- | ------------ | --- | ----- | ------------ | ------- | --------- | ----- | ----------- | --- | ---------- | --- | ------- | -| | | | | | | | evidence. | These | limitations | | can result | in | perfor- | -| costlytoannotate,andthereforescarce. | | | | | | Thus,we | | | | | | | | -mancedegradationorevenhallucinations(Huang -| start with | an automatic | | data | generation | | pipeline | | | | | | | | -| ---------- | ------------ | --- | ---- | ---------- | --- | -------- | -------------- | --- | ------------ | --- | --------- | --- | ---- | -| | | | | | | | et al., 2023), | | highlighting | the | necessity | for | more | -tailoredforcollectinghigh-qualityattributiondata -factualandverifiablesystems. -(§3.1). Givenauserquery,thepipelineautomates -dataconstructionthroughdocumentretrieval,rele- -| | | | | | | | AttributedLargeLanguageModels. | | | | | Theper- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | --- | ------- | --- | -vancereranking,attributedanswergeneration,and -| | | | | | | | sistent | challenge | of hallucinations | | | within | LLMs | -| ------------------------------- | --- | ------ | ---- | ------------------- | --- | --- | ----------- | --------- | ----------------- | --- | ------------- | ------ | ---- | -| data filtering | to | ensure | both | the informativeness | | | | | | | | | | -| | | | | | | | has spurred | the | development | | of attributed | | LLMs | -| andattributabilityoftheanswers. | | | | Furthermore,to | | | | | | | | | | -(Bohnetetal.,2022;Lietal.,2023;Worledgeetal., -betterunlockLLMs’abilityforfine-grainedattri- -2023),whichseektoenhanceinformationverifia- -| bution, | we introduce | | FRONT, | a two-stage | | train- | | | | | | | | -| ------------- | ------------ | ---- | ------- | ----------- | --- | -------- | --------- | ---------- | --------- | ------ | --------- | ----------- | -------- | -| | | | | | | | bility by | generating | responses | | with | attribution | to | -| ing framework | | that | teaches | LLMs | to | generate | | | | | | | | -| | | | | | | | evidence | sources. | The | way of | providing | | attribu- | -Fine-gRainedgrOuNdedciTations(§3.2),consist- -| | | | | | | | tionsvariesacrossstudies. | | | Forexample,Gaoetal. | | | | -| ------ | --------- | ------ | ---------- | --- | ---- | --- | ------------------------- | --- | --- | ------------------- | --- | --- | --- | -| ing of | Grounding | Guided | Generation | | (G3) | and | | | | | | | | -(2023b)enablesLLMstogeneratetextwithin-line -Consistency-AwareAlignment(CAA).Duringthe -| | | | | | | | citations | via in-context | | learning. | Another | | line of | -| --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------- | --- | ------- | -G3 -stage,theLLMfirstselectssupportingquotes -| | | | | | | | research | (Gao | et al., 2023a; | Xu | et al., | 2023b) | ex- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ---- | -------------- | --- | ------- | ------ | --- | -fromretrievedsources(grounding)andthencondi- -plorespost-hocattribution,whereLLMsfirstgen- -tionsthegenerationprocessonthem(generation). -erateaninitialresponseandthenretrievethemost -TheCAAstagethenutilizespreferenceoptimiza- -| | | | | | | | relevantevidencetoachieveattribution. | | | | | Inthispa- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | --------- | --- | -tiontofurtheralignthegroundingandgeneration -per,weadvancetheresearchonattributedLLMs -processbyautomaticallyconstructingpreference -| | | | | | | | further. | Unlikeexistingmodelsthatpredominantly | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------- | --- | --- | --- | --- | --- | -signals. Inthisway,thesequotescanserveasfine- -| | | | | | | | cite document | | identifiers, | we | delve | into | a more | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ------------ | --- | ----- | ---- | ------ | -grainedcitationsandimprovetheefficiencyofthe -fine-grainedformofattributionbypinpointingand -verificationprocessforusers(seeFigure1). -citingspecificextractivequotes. -| We conduct | | extensive | experiments | | to | evaluate | | | | | | | | -| ---------- | --- | --------- | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -our framework on the ALCE Benchmark (Gao 3 TaskFormulationandMethodology -| etal.,2023b). | Ourfindingsareasfollows: | | | | | | | | | | | | | -| ------------- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Following(Liuetal.,2023;Gaoetal.,2023b),the -| • FRONT | demonstratessupervisorperformance | | | | | | | | | | | | | -| ------- | --------------------------------- | --- | --- | --- | --- | --- | ------------------ | --- | ----------- | --- | ----- | ------ | ----- | -| | | | | | | | task is formalized | | as follows: | | given | a user | query | -gainsincitationqualitycomparedtoallbase- q and a corpus of retrieved documents D as in- -lines,achievinganaverage14.21%improve- -| | | | | | | | put, the | LLM | is required | to | produce | a response | | -| --- | --- | --- | --- | --- | --- | --- | -------- | --- | ----------- | --- | ------- | ---------- | --- | -mentusingLLaMA-2-7B. -| | | | | | | | S, which | consists | of statements | | with | embedded | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | -------- | -------------------- | --- | ---- | -------- | ---- | -| | | | | | | | in-linecitations. | | WeassumetheresponseS | | | | com- | -• Humanevaluationrevealsthatthequotesgen- -| | | | | | | | prising | with | n statements | S | = {s | ,s ,...,s | } | -| ---------------------------------------- | --- | ------------- | --- | --- | ------- | ------- | ----------------- | -------- | ------------ | ----------------------- | ---- | ------------- | --- | -| erated | by | our framework | | are | of high | quality | | | | | | 1 2 | n | -| | | | | | | | andeachstatements | | | ∈ S,citesalistofpassage | | | | -| andsignificantlybenefituserverification. | | | | | | | | | i | | | | | -| | | | | | | | C i = {c | i1 ,c i2 | ,...}, where | c ij | ∈ D. | Specifically, | | -• AnalysisshowsthatFRONTgenerateslesshal- citationsarepresentedintheformof[1][2]. -lucinationanddemonstratesremarkablegen- Next,wepresentacomprehensiveoverviewof -eralizationacrossdifferentbasemodels. ourmethod,whichconsistsoftwoprimarycompo- - -Automatic Data Generation Pipeline -          Data Collection        Answer Generation           Data Filtering -Question: Why do bagels have holes in the middle？ Bagel holes ensures even baking (...) a area -| | | | | | | Demonstrations | | |  1    3  . Another suggests (...) and transport  2  . | | | | -| --- | --- | --- | --- | --- | -------------------------------------------- | -------------- | --- | --- | ----------------------------------------------------- | --- | --- | --- | -| | | | | | Instruction: Generate an answer and cite the | | | | Informative? | | | | -Retrieve source for the question and provided documents Break into -Question: Why do bagels have holes in the middle？ -| | Top-100 retrieval documents | | | | | | | | | | statements | | -| --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | -(...) -| | | | | | Prior to (...) the hole was originally put in place | | | | Statement 1 | Bagel holes ensures even baking by | | | -| --- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | ----------- | ---------------------------------- | --- | --- | -allowing heat to circulate around a -| | | | | | to allow for easier handing prior. | | | | | greater surface area  1    3  .  | | | -| --- | -------- | -------------------------------- | ---------------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | -------------------------------- | --- | --- | -| | T h e b | a k e r s (. .. ) th e h o | l e i n t h e c e n tre of the | | | | | | | | | | -1 ba g e l e n s u r e s t h a t i t b a k e s e v e n ly . Statement 2 A n o t h e r s u g g e s t s t h a t t h e y w e re u s ed -| | | | | | | | | | | fo r e a s i e r | h a n d li n g a n | d tr a n sp o rt   2  . | -| ------ | --------------------------------------------- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | ----------- | ------------------ | -------------------- | --------------------------- | -| | Prior to (...) the hole was originally put in | | | | Bagel holes ensures even baking by allowing | | | | | | | | -| | 2 | | | | heat to circulate around a greater surface | | | | | | | Support | -| | place to allow for easier handing prior. | | | | | | | | Attributed? | | | | -| Rerank | | | | | area   1   3   . Another suggests that they were | | | | | | | | -T h e r i n g s h a p e a l lo w s h e a t t o c ir c u la t e used for easier handling and transport  2  . Documents -| | 3 ar o u n | d ( .. . )  m a k e s b a g | e l s t o c o o k fa s t er  | | | | | | | | | | -| --- | ---------- | ------------------------------ | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Figure2: Overviewofthedatagenerationpipeline. Thepipelineconsistsofthreeprimarysteps: datacollection, -answergeneration,anddatafiltering. Firstly,givenauserquery,thedatacollectionmoduleretrievesthetop100 -relevantdocumentsandemploysarerankingmodeltoselectthetop5mostpertinentdocuments. Subsequently, -attributedresponsesaregeneratedbydistillingChatGPTviain-contextlearning. Finally,allresponsesarefiltered -bythedatafilteringmoduletoensureinformativenessandattributability. -nents: anautomaticdatagenerationpipeline(§3.1) subsequently re-rankedby RankVicuna (Pradeep -andatwo-stagetrainingframework(§3.2). etal.,2023)consideringitssuperiorperformance -| | | | | | | inlistwisere-ranking, | | | resultinginthetop5most | | | | -| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | --- | -3.1 AutomaticDataGenerationPipeline -relevantdocumentsforeachquery. -EquippingLLMswiththeattributioncapabilityne- -| | | | | | | Attributed | Answer | | Generation. | | Given | the re- | -| ---------- | -------- | ---- | ------------- | ------------ | -------- | ---------- | ----------- | --- | ----------- | ------- | ----- | ---------- | -| cessitates | training | data | that includes | high-quality | | | | | | | | | -| | | | | | | markable | performance | | of | ChatGPT | in | attributed | -| responses | paired | with | precise | citations, | which is | | | | | | | | -questionanswering,weemployChatGPTtogener- -| typically | labor-intensive | | and | costly. To | address | | | | | | | | -| --------- | --------------- | --- | --- | ---------- | ------- | --- | --- | --- | --- | --- | --- | --- | -ateanswerswithcorrespondingcitationsforgiven -thischallenge,weproposeapipelinedesignedfor -| | | | | | | queriesandthetop5retrieveddocuments. | | | | | | Wepro- | -| --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | --- | ------ | -theautomaticgenerationofhigh-qualityattributed -videpreciseinstructionsandin-contextdemonstra- -data4. -Thispipelinecomprisesthreecorecompo- -tionstoensurethatChatGPTproducesinformative -nents: datacollection,attributedanswergeneration, -responsesandcitesthesourcesaccordingly. -anddatafiltering,asoutlinedinFigure2. -| | | | | | | DataFiltering. | | Toguaranteethehighqualityof | | | | | -| --------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | --------------------------- | ----- | --------- | --- | ----------- | -| DataCollection. | | Tosimulatethereal-worldenvi- | | | | | | | | | | | -| | | | | | | our synthetic | | training | data, | we employ | | a data fil- | -ronmentforinformation-seeking,wecollectques- -teringprocessguidedbytwokeycriteriaderived -tionsfromtheAQuAMuSedataset(Kulkarnietal., -| | | | | | | fromKamallooetal.(2023): | | | | (1)informativeness: | | | -| --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ------------------- | --- | --- | -2020),whichisderivedfromtheNaturalQuestion -| | | | | | | assessing | if the | answer | provides | | sufficient | infor- | -| ----------------------------------- | --- | --- | --- | --- | ----- | ------------------------------------------ | ------ | ------ | -------- | --- | ---------- | ------ | -| (NQ)dataset(Kwiatkowskietal.,2019). | | | | | TheNQ | | | | | | | | -| | | | | | | mationtothequestion,and(2)attributability: | | | | | | de- | -datasetcomprisesrealuserqueriesfromtheGoogle -| | | | | | | termining | if the | answer | is | attributed | | to the cited | -| --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | --- | ---------- | --- | ------------ | -searchengine,providingarobustbasisforrealistic -| | | | | | | documents. | To | mitigate | | the impact | of | nonsensi- | -| ------------------ | --- | ---------- | --- | ----------- | ----- | ---------- | --- | -------- | --- | ---------- | --- | --------- | -| question-answering | | scenarios. | | The dataset | spans | | | | | | | | -calqueriesandirrelevantdocumentretrievalthat -| a range | of diverse | question | types, | demanding | an- | | | | | | | | -| ------- | ---------- | -------- | ------ | --------- | --- | -------- | ------------------ | --- | --- | -------- | --- | ---------- | -| | | | | | | may lead | to non-informative | | | answers, | | we utilize | -swersofvaryinglengths,fromconcisetodetailed. -ChatGPTforpreliminaryinformativenessannota- -Tomimicthewayasearchenginemightsynthesize -| | | | | | | tions. Responses | | categorized | | as | non-informative | | -| --- | --- | --- | --- | --- | --- | ---------------- | --- | ----------- | --- | --- | --------------- | --- | -documentsofhighrelevanceinresponsetoauser -| | | | | | | aredirectlyexcluded. | | | Furthermore,toensurethat | | | | -| ------ | --------- | ------ | ------- | ------- | -------- | -------------------- | --- | --- | ------------------------ | --- | --- | --- | -| query, | we employ | Sphere | (Piktus | et al., | 2021), a | | | | | | | | -answersareaccompaniedbyhighlysupportiveci- -pre-processedandcleanedversionoftheCommon -tations,wetrainadiscriminatoronhuman-labeled -Crawlcorpus,servingasaproxywebsearchindex. -| | | | | | | data from | the | comprehensive | | evaluation | | by Liu | -| --- | --- | --- | --- | --- | --- | --------- | --- | ------------- | --- | ---------- | --- | ------ | -Inparticular,foragivenuserquerysampledfrom -| | | | | | | et al. (2023), | | where | attributability | | is | categorized | -| --- | ------------ | ---------- | ------------ | ---------- | ---- | ---------------- | ------------------------------------- | ------------------------------- | --------------- | ----- | --- | ----------- | -| the | AQuAMuSe | dataset, | we initially | retrieve | the | | | | | | | | -| | | | | | | intothreelevels: | | fullsupport,partialsupport,orno | | | | | -| top | 100 relevant | documents | from | the Sphere | cor- | | | | | | | | -| | | | | | | support. | Wequantitativelymapthediscriminator’s | | | | | | -| pus | using sparse | retrieval. | These | documents | are | | | | | | | | -| | | | | | | outputs | to an | attributability | | score | and | ultimately | -4Attributeddatarefersto“answerswithin-linecitations”. deriveanaveragescoreforeachattributedanswer. - -Two-Stage Training Recipe -          Grounding Guided Generation        Consistency-Aware Alignment -          Answer Generation -| Q :   W h y d o | b a g e l s h a v e | | | | | | | | | | | | -| ----------------- | ----------------------- | ---------------- | --------------------------- | ----------------------- | ------------------------------------- | --- | ----------------------- | ------------------- | ----------------------------------------------- | --------------------------- | --- | --- | -| | | G r o u n d i ng | T h e b a k e r s ( . . | . ) t h e h o l e | i n t h e c e n t r e o f the | Q | u e s t i o n :   W h y | d o b a g e l s |   G  R      D eI   G    1l   e | n s u r e s t h a t i t | | | -h o l e s i n t h e m i d d l e ？ 1 b a g e l { e n s u r e s t h a t i t b a k e s e v e n l y } . h a v e h o l e s i n t h e m i d d l e ？ b a kO eU Ns N v e n y .   3   a l l o w s h e a t -| | | | | | | | ( . . . ) | | t o c i r c u l a t e a r | o u n d .  2    a l l o w   s A u   N r S  f W  a   c E | R e     a B r a e g a e   1 l   h  3 o  A l e n s o ( t . h . . e )   r | g ( r . e .. a ) t | -| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------- | -B a g e l h o l e s e n s u r e s e v e n b a k i n g b y P r i o r t o ( . . . ) t h e h o l e w a s o r i g i n a l l y p u t in P r i o r t o ( .. . )   p l a c e t o a ll o w f o r e a s i e r h a n d i n g p r i o r . h a n d l i n g a n d t r a n s p o r t   2  . -| al lo w i n g h e a | t to c i r c u l a t e | a r o u n d a 2 | | | | | | | | | | | -| -------------------- | -------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -greater surface area  1   3  .Another p l a c e t o { a l l o w f o r e a s ie r h a n d i n g p r i o r } . fo r e a s i e r h a n d i n g p r i o r . Consistency -| suggests that they were used for | | 3 | The ring shape {allows heat to circulate | | | | | | | | | | -| -------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -easier handling and transport  2  . around} (...) makes bagels to cook faster  Gold Grounding Consistent Answer Contrastive -Large Language -| | Supervised Fine-tuning | | | | | | Model | | | | | | -| -------- | ---------------------- | -------------- | ------------- | --------------------------------------------- | ------------------------------------------------- | --- | ----- | --- | -------------- | ------------------- | --- | --- | -| | | | | | | | | | Gold Grounding | Inconsistent Answer | | | -| Question | | |   G   R   | Ol  Uo N  D  sI N   Gh     1   e n s | u r e s t h a t it b a k e s e v e n ly.  | | | | | | | | -| | | Large Language |   f3 a | l w e a t t o | c i r c u l a t e a r o u n d .   2  a l lo w | | | | | | | | -| | | | o r | e a s i e r h a n d i n | g p r i o r . | | | | | | | | -Documents Model Direct Preference Optimization (DPO)   A   N S  W    E R    B a g e l h o l e s   (. . . )  f o r -| | | |   A A  N n  S o | W  t h  E e  R r   B (. a .. g ) e h l a h n o d le | lin s g ( . a .. n )  d g r t e ra a n t s s p u o rf r a t  c 2  e . area13 | | | | | a e s t h e | t i c s a n d u n i q u e t e x | t u r e   1 | -| --- | --- | --- | ---------------- | ------------------------------------------------------------ | ------------------------------------------------------------------------------------- | --- | --- | --- | --- | ----------------- | ------------------------------------- | ------------- | -| | | | | | | | | | |   ( . . . ) h a | n d l i n g a n d tr a n s p o | r t   3   . | -Figure 3: Overview of FRONT: The training recipe consists of two stages: grounding-guided generation and -consistency-aware alignment. It enables LLMs to first generate precise grounding and subsequently guide the -generationofattributedanswers,therebyenhancingfine-grainedattributioncapability. -Answersfallingbelowadefinedthresholdaresys- fine-tunedtogeneratearesponseS whichconsists -tematicallyexcludedtoensurethesyntheticdata’s of two components: the grounded quotes G and -reliability, which results in nearly 8,000 entries. theattributedanswerA. Specifically,thegrounded -Formoredetails,pleaserefertoAppendixA. quotesG aredelineatedasfollows: -3.2 Two-StageTrainingRecipe G = {[GROUNDING],(i ,e ),...,(i ,e )}, (1) -| | | | | | | | | | | 1 1 | n n | | -| ------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Inthissection,weintroduce | | | FRONT,atwo-stage | | | | | | | | | | -where[GROUNDING]denotesaspecialtokenindi- -trainingframeworkthataimsatempoweringLLMs -| | | | | | | cating | the | start of | the grounding | process. | | Each | -| ----------------- | --- | ----------- | ----------- | --- | -------- | ------ | --- | -------- | ------------- | -------- | --- | ---- | -| with fine-grained | | attribution | capability. | | Figure 3 | | | | | | | | -tuplewithinG,comprisingadocumentidentifieri -illustratestheoverviewofourframework. -andthecorrespondingextractivesegmente,collec- -tivelyformingagroundedquote. -3.2.1 GroundingGuidedGeneration -| | | | | | | Similarly, | | the formulation | | of the | attributed | an- | -| ---------- | --- | --------- | ------------ | --- | -------- | ---------- | --- | --------------- | --- | ------ | ---------- | --- | -| To empower | | LLMs with | fine-grained | | attribu- | | | | | | | | -swerAisconciselypresentedas: -| tion capability, | | we propose | Grounding | | Guided | | | | | | | | -| ---------------- | --- | ---------- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -(G3), -Generation which teaches LLMs to gener- A = {[ANSWER],s ,s ...,s }, (2) -| | | | | | | | | | | 1 2 | m | | -| ------------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| atefine-grainedcitations. | | | ThecornerstoneofG3 | | | | | | | | | | -liesinenablingLLMstoextractsupportingquotes where [ANSWER] is a special token that signals -fromthesourcedocuments,eachassociatedwithits the beginning of the answer generation process. -documentidentifier,whichinturnguidesthegen- Each statement s cites a list of passages C = -| | | | | | | | | i | | | | i | -| --------------------------- | --- | --------------------------- | ----------------- | --- | --- | ----------------- | ------------ | ------ | --- | ------------------ | ---- | ----- | -| erationofattributedanswers. | | | Suchaformatoffers | | | | | | | | | | -| | | | | | | {c i1 | ,c i2 ,...}, | wherec | ij | ⊆ {i 1 ,i 2 ,...,i | n }, | asde- | -| twoprimarybenefits. | | Firstly,thedirectextraction | | | | finedinEquation2. | | | | | | | -of quotes from sources significantly reduces the Thus,thetraininglossisformulatedas: -impactoftheincorporationofirrelevantinforma- -| tion and | the risk | of hallucinations | | in | subsequent | | | N | | | | | -| -------- | -------- | ----------------- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -(cid:88) -| | | | | | | | | L = − | logP(y | |q ,D | ;θ) | (3) | -| ------------------ | --- | ---------------------------- | --- | --- | --- | --- | --- | ----- | ------ | ----- | --- | --- | -| attributedanswers. | | Secondly,theprocessnaturally | | | | | | | | i i | i | | -i=1 -facilitatesaccurateattribution,witheachdocument -identifierservingasaclearsupervisedsignalthat -| | | | | | | where | y | represents | the | combined | output | of | -| --- | --- | --- | --- | --- | --- | ----- | --- | ---------- | --- | -------- | ------ | --- | -i -delineatestheoriginoftheextractivequotes,thus -groundedquotesGandtheanswerAforeachgiven -improvingthecitationquality. queryq andsetofretrieveddocumentsD . -| | | | | | | | i | | | | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -However,theabsenceofspecificgroundingcon- -tent for statements within our generated dataset 3.2.2 Consistency-AwareAlignment -posesadditionalchallenges. Totacklethis,weem- While G3 unlocks the ability to first extract sup- -ploy ChatGPT to meticulously extract segments portingquotesbeforegeneratingattributedanswers, -fromciteddocumentsthatsupportthecorrespond- it occasionally leads to inconsistencies between -ingstatement. Hence,whengivenaqueryqandthe groundedquotesandattributedanswers. Suchdis- -top-5retrieveddocumentsD asinput,theLLMis crepancieschallengetheattempttoemploythese - -groundedquotesasfine-grainedverification. Inre- whereπ ref representsthereferencemodel,initial- -sponse,weproposeaconsistency-awarealignment izedfromG3. Thehyper-parameterβ modulates -(CCA) stage specifically aimed at enhancing the the divergence between the distribution from the -consistencybetweenthegroundingprocessandthe policy model and the reference model. τ is the -w -generationprocess. consistentanswer,whileτ istheinconsistentone. +Corporate Governance Report + +I. WIPRO’S PHILOSOPHY ON CORPORATE + +GOVERNANCE +Wipro’s governance framework is driven by the objective +of enhancing long term stakeholder value without +compromising on ethical standards and corporate +social responsibilities. Efficient corporate governance +requires a clear understanding of the respective +roles of the Board of Directors (“Board”) and of senior +management and their relationships with others in the + +corporate structure. Sincerity, fairness, good citizenship +and commitment to compliance are key characteristics +that drive relationships of the Board and senior +management with other stakeholders. + +Corporate governance philosophy of Wipro flows from +the “Spirit of Wipro” which represents core values by +which policies and practices of the organization are +guided. The values encapsulated in the “Spirit of Wipro” +are: + +Be passionate about +clients’ success + +Treat each person +with respect + +Be global and +responsible + +Unyielding integrity in +everything we do + +While our Company has transformed many times over +the years, the Spirit of Wipro and our core values, have +remained constant. + +Our Chairman introduced the Five Habits essential to +drive a Growth Mindset in early 2020, which are our +values in action: + +• Being respectful + +• Being responsive + +• Always communicating + +• Demonstrating stewardship + +• Building trust + +So far, over 21,000 employees globally have been part +of 75 immersive and interactive sessions hosted by our +senior leadership team on the Five Habits. + +Corporate governance at Wipro is implemented through +robust board governance processes, internal control +systems and processes, and strong audit mechanisms. +These are articulated through the Company’s Code of +Business Conduct, Corporate Governance Guidelines +and charters of various sub-committees of the Board +and the Company’s Disclosure Policy. Wipro’s corporate +governance practices can be described through the +following four layers: + +a) Governance by Shareholders + +b) Governance by Board of Directors + +c) Governance by Sub-committees of Board, and + +d) Governance through management process + +In this report, we have provided details on how the +corporate governance principles are put in to practice +within Wipro. + +II. SHAREHOLDERS + +The Companies Act, 2013, Securities and Exchange +Board of India (Listing Obligations and Disclosure +Requirements) Regulations, 2015 (“Listing Regulations”) +and New York Stock Exchange (NYSE) Listed Company +Manual prescribe the governance mechanism by +shareholders in terms of passing of ordinary and special +resolutions, voting rights, participation in the corporate +actions such as bonus issue, buyback of shares, +declaration of dividend, etc. Your Company follows a +robust process to ensure that the shareholders of the +Company are well informed of Board decisions both on +financial and non-financial matters and adequate notice +with a detailed explanation is sent to the shareholders +well in advance to obtain necessary approvals. + +III. BOARD OF DIRECTORS +1. Composition of Board + +is a professional CEO who + +As at March 31, 2021, our Board had two Executive +Independent Directors +Directors, six non-executive +and one non-executive non-independent Director. The +Executive Chairman and Whole Time Director, and the +non-executive non-independent Director are Promoter +(CEO) and +Directors. The Chief Executive Officer +Managing Director +is +responsible for the day to day operations of the +Company. Of the seven Non-Executive Directors, six +are Independent Directors, free from any business or +other relationship that could materially influence their +judgment. In the opinion of the Board, all the Independent +Directors are independent of the management and +satisfy the criteria of independence as defined under +the Companies Act, 2013, the Listing Regulations and +the NYSE Listed Company manual. + +121 + + Corporate Governance Report + +is well diversified and + +consists +The Board +of a woman +three +Directors who are foreign nationals. The profiles +of our Directors are available on our website at +https://www.wipro.com/leadership. + +Independent Director and + +Post-Meeting Follow-Up System +After the Board meeting, we have formal system of +follow-up, review and reporting on actions taken by the +management on the decisions of the Board and sub- +committees of the Board. + +2. Board Meetings + +4. Appointment of Directors + +We decide about the Board meeting dates +in +consultation with the Board members. The schedule +of the Board meetings and Board Committee meetings +is communicated in advance to the Directors to enable +them to attend the meetings. + +Our Board meetings are normally scheduled over two +days. In addition, every quarter, Independent Directors +meet amongst themselves exclusively. In line with Para 4 +of Schedule B of SEBI (Prohibition of Insider Trading) +Regulations, 2015, it is the endeavor of the Company +that the gap between the clearance of accounts by audit +committee and board meeting is as narrow as possible, +and Wipro is committed to adhere to this requirement. + +Information flow to the Board Members +Information is provided to the Board Members on a +continuous basis for their review, inputs and approval +from time to time. More specifically, we present our +annual Strategic Plan and Operating Plans of our +business to the Board for their review, inputs and +approval. Likewise, our quarterly financial statements +and annual financial statements are first presented to +the Audit Committee and subsequently to the Board +for their approval. In addition, various matters such as +appointment of Directors and Key Managerial Personnel, +corporate actions, review of internal and statutory +audits, details of investor grievances, specific cases of +acquisitions, important managerial decisions, material +positive/negative developments and +legal/statutory +matters are presented to the respective Committees +of the Board and later with the recommendation of +Committees to the Board of Directors for their approval +as may be required. + +As a system, in most cases, information to Directors is +submitted along with the agenda papers well in advance +of the Board meeting. Inputs and feedback of Board +Members are taken and considered while preparation +of agenda and documents for the Board meeting. + +Documents containing Unpublished Price Sensitive +Information are submitted to the Board and Committee +Members, at a shorter notice, as per the general consent +taken from the Board, from time to time. + +3. + +122 + +The Board has adopted the provisions with respect +to appointment and tenure of Independent Directors +consistent with the Companies Act, 2013 and the Listing +Regulations. + +As per the provisions of the Companies Act, 2013, the +Independent Directors shall be appointed for not more +than two terms of maximum of five years each and shall +not be liable to retire by rotation. + +issues a formal + +Independent +At the time of appointment of an +Director, the Company +letter of +appointment outlining his/her role, function, duties +and responsibilities as a Director. The template of the +letter of appointment is available on our website at +h t t p s : // w w w.w i p ro .c o m / i n v e s t o r s /c o r p o r a t e - +governance/. + +Details of Directors proposed for appointment/re- +appointment at the 75th Annual General Meeting +(“AGM”) is provided at page no. 91 as part of the Board’s +Report and in the notice convening the 75th AGM. + +Lead Independent Director +The Board has designated Mr. M. K. Sharma as +Independent Director. The role of the +the Lead +Lead +in the +Corporate Governance guidelines of your Company +and +is available on the Company’s website at +h t t p s : // w w w.w i p ro .c o m / i n v e s t o r s /c o r p o r a t e - +governance/. + +Independent Director + +is described + +5. Policy + +for Nomination of Directors, + +their + +Remuneration and Board Diversity +The Board Governance, Nomination and Compensation +Committee has adopted a policy which, inter alia, deals +with the manner of selection of Directors and payment +of their remuneration as described herein below. + +The Company has also adopted a policy on Board +Diversity which guides the organization’s approach to +diversity in the composition of the Board. + +Criteria for Selection of Independent Directors and Key +Skills, Expertise, and Core Competencies of the Board +The Board of the Company comprises of eminent +personalities and leaders in their respective fields. +These Directors are nominated based on well-defined + + selection criteria. The Board Governance, Nomination +and Compensation Committee considers, inter alia, + qualifications, skills, expertise and +experience, +competencies, whilst recommending to the Board +the candidature for appointment as +Independent +Director. + +In case of appointment of Independent Directors, the +Board Governance, Nomination and Compensation +Committee satisfies itself about the independence +of the Directors vis-à-vis the Company to enable the +Board to function independently of the management +and discharge its functions and duties effectively. In +case of re-appointment of Independent Directors, the +Board also takes into consideration, the performance + +evaluation and engagement level of the Independent +Directors. + +The Board Governance, Nomination and Compensation +Committee ensures that the candidates +identified +for appointment as Directors are not disqualified for +appointment under Section 164 and other applicable +provisions of the Companies Act, 2013. + +As required under Rule 6 of the Companies (Appointment +and Qualification of Directors) Rules, 2014, all the +Independent Directors have completed the registration +with the Independent Directors Databank and, wherever +required, also completed the online proficiency test, +conducted by Indian Institute of Corporate Affairs. + +In the opinion of the Board and the Board Governance, Nomination and Compensation Committee, the following is a list of +core skills/expertise/competencies required in the context of the Company’s business and which are available with the Board: + +Wide management and +leadership experience + +Information Technology + +Diversity + +Functional and managerial +experience + +Personal values + +Corporate governance + +Strong management and leadership experience, including in areas of business development, strategic +planning and mergers and acquisitions, ideally with major public companies with successful multinational +operations in technology, manufacturing, banking, investments and finance, international business, +scientific research and development, senior level government experience and academic administration. +Expertise or experience in information technology business, technology consulting and operations, +emerging areas of technology such as digital, cloud and cyber security, intellectual property in information +technology domain, and knowledge of technology trends. +Diversity of thought, experience, knowledge, perspective, gender and culture brought to the Board +by individual members. Varied mix of strategic perspectives, geographical focus with knowledge and +understanding of key geographies. +Knowledge and skills in accounting and finance, business judgment, general management practices and +processes, crisis response and management, industry knowledge, macro-economic perspectives, human +resources, labour laws, international markets, sales and marketing, and risk management. +Personal characteristics matching the Company’s values, such as integrity, accountability, and high +performance standards. +Experience in developing and implementing good corporate governance practices, maintaining board and +management accountability, managing stakeholders’ interests and Company’s responsibilities towards +customers, employees, suppliers, regulatory bodies and the communities in which it operates. Experience +in boards and committees of other large companies. + +Given below is a list of core skills, expertise and competencies of the individual Directors: + +Name of Director + +Mr. Rishad A. Premji +Mr. Azim H. Premji +Mr. Thierry Delaporte2 +Mr. William Arthur Owens +Mr. M. K. Sharma +Mrs. Ireena Vittal +Dr. Patrick J. Ennis +Mr. Patrick Dupuis +Mr. Deepak M. Satwalekar3 + +Wide Management +and Leadership +experience1 +  +  +  +  +  +  +  +  +  + +Information +Technology + +  +  +  +  + - +  +  +  +  + +Diversity + +Skills/Expertise/Competencies +Functional and +Managerial +Experience1 +  +  +  +  +  +  +  +  +  + +  +  +  +  +  +  +  +  +  + +Personal +Values + +Corporate +Governance + +  +  +  +  +  +  +  +  +  + +  +  +  +  +  +  +  +  +  + +1 These skills/competencies are broad-based, encompassing several areas of expertise/experience. Each Director may possess varied combinations of skills/ +experience within the described set of parameters, and it is not necessary that all Directors possess all skills/experience listed therein. +2 Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director for a term of 5 years with effect from July 6, 2020 to July 5, 2025. +3 Mr. Deepak M. Satwalekar was appointed as an Independent Director for a term of 5 years with effect from July 1, 2020 to June 30, 2025. + +123 + + Corporate Governance Report + +6. Familiarization Programme and Training + +for + +Independent Directors +The Company has an orientation process/familiarization +programme for its independent directors that includes: + +a) Briefing on their role, responsibilities, duties, and + +obligations as a member of the Board. + +b) Nature of business and business model of the +Company, Company’s strategic and operating plans. + +c) Matters relating to Corporate Governance, Code of +Business Conduct, Risk Management, Compliance +Programs, Internal Audit, etc. + +As a process when a new independent director is +appointed, a familiarization programme as described +above is conducted by the senior management team +and whenever a new member is appointed to a Board +Committee, information relevant to the functioning of the +Committee and the role and responsibility of Committee +members is informed. Each of our independent directors +have attended such orientation process/familiarization +programme when they were inducted into the Board. + +As part of ongoing training, the Company schedules +quarterly meetings of business heads and functional +heads with the Independent Directors. During these +meetings, comprehensive presentations are made +on the various aspects such as business models, +new business strategies and initiatives by Business +Leaders, risk minimization procedures, recent trends +in technology, changes in domestic/overseas industry +scenario, Digital Transformation, state of Global IT +Services +industry, and regulatory regime affecting +the Company globally. These meetings also facilitate +Independent Directors to provide their inputs and +suggestions on various strategic and operational +matters directly to the business and functional +heads. The details of the familiarization programme +are also available on the website of the Company at +https://www.wipro.com/content/dam/nexus/en/investor/ +corporate-governance/policies-and-guidelines/ethical- +guidelines/12771-familiarization-programmes-imparted- +to-independent-directors.pdf. + +The Board Governance, Nomination and Compensation +Committee presents to the Board on a periodic basis, +succession plans for appointments to the Board +based on various factors such as current tenure of +Directors, outcome of performance evaluation, Board +diversity and business requirements. In addition, the +Company conducts an annual Talent Review Process +for senior management and other executive officers +which provides a leadership-level talent inventory and +capability map that reflects the extent to which critical +talent needs are fulfilled vis-a-vis business drivers. + +The Board Governance, Nomination and Compensation +Committee reviews the outcome of this process and +presents the succession plan for senior management +and other executive officers to the Board. + +8. Board Evaluation + +Details of methodology adopted for Board evaluation +have been provided at page no. 92 of the Board’s Report. + +9. Remuneration Policy and Criteria for Making +Payments to Directors, Senior Management and +Key Managerial Personnel +The Independent Directors are entitled to receive +remuneration by way of sitting fees, reimbursement +of expenses for participation in the Board/Committee +meetings and commission as detailed hereunder: + +a) Sitting fees for each meeting of the Board or +Committee of the Board attended by him or her, of +such sum as may be approved by the Board within +the overall limits prescribed under the Companies +Act, 2013. + +and + +b) Commission on a quarterly basis, of such sum as +may be approved by the Board and Members on +the recommendation of the Board Governance, +Nomination +Committee. +The aggregate commission payable to all the +Independent Directors and non-executive directors +put together shall not exceed 1% of the net profits +of the Company during any financial year. The +commission is payable on pro-rata basis to those +Directors who occupy office for part of the year. + +Compensation + +7. Succession Planning + +We have an effective mechanism for succession +planning which focuses on orderly succession of +Directors, +including Executive Directors and other +senior management team and other executive officers. +The Board Governance, Nomination and Compensation +Committee implements this mechanism in concurrence +with the Board. + +c) Reimbursement of travel, stay and other expenses +for participation in Board/Committee meetings. + +d) Independent Directors and Promoter Directors +are not entitled to participate in the stock option +schemes of the Company. + +Following are terms and conditions for determining +the remuneration to Mr. Azim H. Premji, who is a Non- +Executive, Non-Independent Director: + +124 + + a) Remuneration as applicable to other Non-Executive +Directors of the Company, in addition to the sitting +fees for attending the meetings of the Board thereof, +as may be determined by the Board, provided +however that the aggregate remuneration, including +commission, paid to the Directors other than the +Managing Director and Whole Time Directors in a +financial year shall not exceed 1% of the net profits +of the Company, in terms of Section 197 of the +Companies Act, 2013 and computed in the manner +referred to in Section 198 of the Companies Act, 2013. + +b) Maintenance of Founder Chairman’s office including + +executive assistant at Company’s expense. + +c) Reimbursement of travel, stay and entertainment +expenses actually and properly incurred in the +course of business as per the Company’s policy. + +In determining the remuneration of Chairman, CEO and +Managing Director, and Senior Management Employees +and Key Managerial Personnel, the Board Governance, +Nomination and Compensation Committee and the +Board shall ensure/consider the following: + +a) The balance between fixed and variable pay +reflecting short and +long-term performance +objectives, appropriate to the working of the +Company and its goals. + +b) Alignment of remuneration of Key Managerial +Personnel and Directors with long-term interests of +the Company. + +c) Company’s performance vis-à-vis + +the annual +achievement, +vis-à- +vis KRAs/KPIs, industry benchmark and current +compensation trends in the market. + +individuals’ performance + +the + +recommends + +remuneration + +The Board Governance, Nomination and Compensation +Committee +for +the Chairman, CEO and Managing Director, Senior +Management and Key Managerial Personnel. The +payment of remuneration to the Executive Directors and +Non-Executive Directors is approved by the Board and +Members. Approval of Members is also obtained in case +of remuneration payable to Non-Executive Directors. +There was no change to the remuneration policy during +the financial year. + +Details of Remuneration to Directors +Details of remuneration paid to the Directors for the services rendered and stock options granted during the financial year +2020-21 are given below. No stock options were granted to any of the Independent Directors and Promoter Directors during +the financial year 2020-21. + +(in ` Million) + +Relationship with directors + +Salary + +Allowances + +Commission/Incentives/ +Variable Pay + +Other annual compensation + +Retirals + +Sitting fees + +TOTAL +Grant of ADS Restricted +Stock Units during the year + +Notice period + +Rishad A. +Premji(2)(3) +Son of Azim +H. Premji + +Thierry +Delaporte(1)(4)(5) +None + +William +Arthur +Owens(1) +None + +Azim H. +Premji +Father of +Rishad A. +Premji + +M. K. +Sharma +None + +Ireena +Vittal +None + +Patrick J. +Ennis(1) +None + +Patrick +Dupuis(1) +None + +Deepak M. +Satwalekar(6) +None + +Arundhati +Bhattacharya(7) +None + +Abidali Z. +Neemuchwala(1)(8) +None + +14.30 + +43.93 + +55.68 + +0.17 + +3.86 + +NA + +117.94 +NA + +Up to 180 +days + +68.63 + +27.40 + +112.79 + +379.19 + +55.49 + +NA + +643.50 +1,500,000 + +NA + +NA + +7.15 + +NA + +NA + +0.60 + +7.75 +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +29.21 + +11.60 + +10.24 + +20.73 + +20.73 + +NA + +NA + +0.60 + +29.81 +NA + +NA + +NA + +0.60 + +12.20 +NA + +NA + +NA + +0.60 + +10.84 +NA + +NA + +NA + +0.60 + +21.33 +NA + +NA + +NA + +0.60 + +21.33 +NA + +Up to 180 days + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +NA + +6.65 + +NA + +NA + +0.40 + +7.05 +NA + +NA + +NA + +NA + +2.22 + +NA + +NA + +0.20 + +2.42 +NA + +NA + +14.14 + +NA + +2.42 + +6.15 + +0.002 + +NA + +22.72 +NA + +NA + +Figures in the above table are subject to rounding-off adjustments + +Notes: + +(1) Figures mentioned in ` are equivalent to amounts paid in foreign currency, wherever applicable. + +(2) Mr. Rishad A. Premji is entitled to a commission at the rate of 0.35% on incremental consolidated net profits of Wipro Limited for fiscal year 2021 + +over the previous year, computed in accordance with the provisions of the Companies Act, 2013. + +(3) Mr. Rishad A. Premji’s compensation also included cash bonus (part of his allowances) on an accrual basis, which is payable over a period of time. + +(4) The compensation disclosed for Mr. Thierry Delaporte is for the period from July 6, 2020 to March 31, 2021. It includes components such as a one- +time cash award, annual stock grant which will vest as per the vesting pattern approved by the Board Governance, Nomination and Compensation +Committee (1,200,000 ADS restricted stock units, comprising of 300,000 ADS restricted stock units expiring in July 2024 and 900,000 ADS + +125 + + Corporate Governance Report + +performance stock units expiring in June 2023 and one-time grant of 300,000 ADS restricted stock units expiring in July 2022), among other things, +as per the terms approved by the Shareholders at the Annual General Meeting held in July 2020. + +(5) The remuneration of Chief Executive Officer and Managing Director is computed on an accrual basis. It also includes the amortization of RSUs +granted to him, which will vest over a period of time. This also includes RSUs that will vest based on performance parameters of the Company. + +(6) Mr. Deepak M. Satwalekar was appointed as an Independent Director of the Company with effect from July 1, 2020 and the compensation disclosed + +is for the period from July 1, 2020 to March 31, 2021. + +(7) The compensation disclosed for Ms. Arundhati Bhattacharya is for the period April 1, 2020 to June 30, 2020. Ms. Bhattacharya resigned as an + +Independent Director of the Company with effect from the close of business hours on June 30, 2020. + +(8) The compensation disclosed for Mr. Abidali Z. Neemuchwala is for the period April 1, 2020 to June 1, 2020. Mr. Neemuchwala resigned as the Chief + +Executive Officer and Managing Director of the Company with effect from the end of the day on June 1, 2020. + +None of the Non-Executive Directors received remuneration +exceeding 50% of the total annual remuneration paid to all +Non-Executive Directors for the year ended March 31, 2021. + +that consist of a pension and gratuity which are offered to all +of our employees, but no other benefits upon termination of +employment except as mentioned below. + +Terms of Employment Arrangements +Under the Companies Act, 2013, our shareholders must +approve the salary, bonus and benefits of all executive +directors at a General Meeting of the Shareholders. Each of +our executive directors has signed an agreement containing +the terms and conditions of employment, including a monthly +salary, performance bonus and benefits including vacation, +medical reimbursement and pension fund contributions. +These agreements have varying terms, but either we or the +executive director may generally terminate the agreement +upon six months’ notice to the other party. + +Pursuant to the terms of the employment arrangement +with Mr. Delaporte, if his employment is terminated by the +Company without Cause, the Company is required to pay +Mr. Delaporte, severance pay of 12 months’ base salary as +last applicable when in service, payable over a 12 month +period following the date of termination. These payments will +cease if Mr. Delaporte obtains a new employment within the +12 months period or becomes a consultant to any Company. + +We also indemnify our directors and officers for claims +brought under any rule of law to the fullest extent permitted +by applicable law. + +The terms of our employment arrangements with Mr. Rishad +A. Premji and Mr. Thierry Delaporte provide for up to a +180-day notice period, and country specific leaves per year in +addition to statutory holidays, and an annual compensation +review. Additionally, these officers are required to relocate +as we may determine, and to comply with confidentiality +provisions. Service contracts with our executive directors +and officers provide for our standard retirement benefits + +Among other things, we agree to indemnify our directors +and officers for certain expenses, judgments, fines and +settlement amounts incurred by any such person in any +action or proceeding, including any action by or in the right +of the Company, arising out of such person’s services as our +director or officer, including claims which are covered by +the director’s and officer’s liability insurance policy taken +by the Company. + +Key Information pertaining to Directors as on March 31, 2021 is given below: + +Designation + +Date of initial +appointment + +Sl. +No. + +Name of +the Director +and Director +Identification +Number (DIN) + +Date of +appointment +as +Independent +Director under +Companies +Act, 2013 and +SEBI Listing +Regulations +(first term) 1 +- + +Directorship +in other +Companies2 + +Chairmanship +in Committees +of Board +of other +Companies3 + +Membership +in Committees +of the Board +of other +Companies3 + +No. of shares +held as on +March 31, +2021 + +Attendance +at the last +AGM held +on +July 13, +2020 + +Other listed +companies where +the Director is +appointed as +Independent +Director + +4 + +12 + +- + +- + +- + +- + +Yes + +1,738,057@ + +Yes + +241,913,816@ + +- + +- + +1-May-2015 + +1-Sep-1968 + +- + +1 + +2 + +Rishad A. Premji +(DIN: 02983899) + +Azim H. Premji +(DIN: 00234280) + +Executive +Director and +Chairman + +Non- +Executive +Non- +Independent +Director + +126 + + Designation + +Date of initial +appointment + +Sl. +No. + +Name of +the Director +and Director +Identification +Number (DIN) + +Directorship +in other +Companies2 + +Chairmanship +in Committees +of Board +of other +Companies3 + +Membership +in Committees +of the Board +of other +Companies3 + +No. of shares +held as on +March 31, +2021 + +Attendance +at the last +AGM held +on +July 13, +2020 + +Other listed +companies where +the Director is +appointed as +Independent +Director + +Date of +appointment +as +Independent +Director under +Companies +Act, 2013 and +SEBI Listing +Regulations +(first term) 1 +- + +3 + +4 + +5 + +Thierry Delaporte +(DIN: 08107242) + +William Arthur +Owens +(DIN: 00422976) + +M. K. Sharma +(DIN: 00327684) + +Chief +Executive +Officer and +Managing +Director + +Independent +Director + +Independent +Director + +6-Jul-2020 + +1-Jul-2006 + +23-Jul-2014 + +1-Jul-2011 + +23-Jul-2014 + +- + +- + +9 + +6 + +Ireena Vittal +(DIN: 05195656) + +Independent +Director + +1-Oct-2013 + +23-Jul-2014 + +5 + +7 + +8 + +9 + +Patrick J. Ennis +(DIN: 07463299) + +Patrick Dupuis +(DIN: 07480046) + +Deepak M. +Satwalekar +(DIN: 00009627) + +Independent +Director + +Independent +Director + +Independent +Director + +1-Apr-2016 + +1-Apr-2016 + +1-Apr-2016 + +1-Apr-2016 + +1-Jul-2020 + +1-Jul- 2020 + +- + +- + +5 + +- + +- + +3 + +- + +- + +- + +1 + +- + +- + +2 + +2 + +- + +- + +1 + +Yes + +Yes + +Yes + +Yes + +Yes + +Yes + +Yes + +- + +- + +- + +- + +- + +- + +- + +- + +- + +1. Asian Paints +Limited + +2. United Spirits + +Limited +3. Ambuja +Cements +Limited +4. Vedanta +Limited + +1. Godrej + +Consumer +Products +Limited +2. Housing + +Development +Finance +Corporation +Limited + +- + +- + +1. Piramal + +Enterprises +Limited +2. Asian Paints +Limited +3. Home First +Finance +Company India +Limited +4. Piramal + +Capital & +Housing +Finance +Limited + +1. At the 70th AGM, Mr. M. K. Sharma was re-appointed as Independent Director for a second term from July 1, 2016 to June 30, 2021. + At the 71st AGM, Mr. William Arthur Owens was re-appointed as Independent Director for a second term from August 1, 2017 to July 31, 2022. + At the 72nd AGM, Mrs. Ireena Vittal was re-appointed as Independent Director for a second term from October 1, 2018 to September 30, 2023. + At the 74th AGM, Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director of the Company to hold office for a period + +of five years from July 6, 2020 to July 5, 2025. + + At the 74th AGM, Mr. Deepak M. Satwalekar was appointed as an Independent Director for a period of five years from July 1, 2020 to June 30, 2025. + Dr. Patrick J. Ennis and Mr. Patrick Dupuis were re-appointed as Independent Directors of the Company for a second term of 5 years, with effect from +April 1, 2021 to March 31, 2026. The said re-appointment was approved by shareholders of the Company vide special resolutions dated June 4, 2021, +passed through postal ballot by e-voting. + +127 + + Corporate Governance Report + +2. This does not include position in foreign companies and position as an advisory board member but includes position in private companies and + +companies under Section 8 of the Companies Act, 2013. None of our Directors hold directorship in more than seven listed companies. + +3. In accordance with Regulation 26 of the Listing Regulations, Membership/Chairmanship of only Audit Committees and Stakeholders’ Relationship + +Committees in all public limited companies have been considered. + +@ Includes equity shares held jointly with immediate family members. + +IV. COMMITTEES OF BOARD + +Our Board has constituted sub-committees to focus +on specific areas and make informed decisions within +the authority delegated to each of the Committees. +Each Committee of the Board is guided by its Charter, +which defines the scope, powers and composition of the +Committee. All decisions and recommendations of the +Committees are placed before the Board for information +or approval. + +During the financial year, the Board has accepted the +recommendations of Committees on matters where +such a recommendation is mandatorily required. There +have been no instances where such recommendations +have not been considered. + +We have four sub-committees of the Board as at March +31, 2021: + +1. Audit, Risk and Compliance Committee, which also + +acts as the Risk Management Committee + +2. Board Governance, Nomination and Compensation +Committee, which also oversees the CSR initiatives +of the Company and acts as the CSR Committee + +3. Administrative + +Shareholders/Investors +Grievance Committee (Stakeholders Relationship +Committee) + +and + +4. Strategy Committee + +1. Audit, Risk and Compliance Committee + +The Audit, Risk and Compliance Committee of our Board +reviews, acts on and reports to our Board with respect +to various auditing and accounting matters. The primary +responsibilities include overseeing: + +a) Auditing and accounting matters, + +including +recommending the appointment of our independent +auditors to the shareholders; + +b) Compliance with legal and statutory requirements; + +c) + +Integrity of the Company’s financial statements, +discussions with the independent auditors regarding +the scope of the annual audits, and fees to be paid to +the independent auditors; + +d) Performance of the Company’s + +internal audit +independent auditors and accounting + +function, +practices; + +e) Review of related party transactions and functioning + +of whistle blower mechanism; + +128 + +f) + +Implementation of the applicable provisions of the +Sarbanes Oxley Act of 2002 (the “Sarbanes Oxley +Act”), including review of the progress of internal +control mechanisms to prepare for certification +under Section 404 of the Sarbanes Oxley Act; + +g) Evaluation of + +internal financial controls, risk +management systems and policies including review +of cyber security; + +h) Review of utilization of loans and advances from, +and investment by, the Company in its subsidiaries +exceeding ` 100 crore or 10% of the asset size of +the subsidiary, whichever is lower, including existing +loans, advances and investments; and + +i) Such other matters and activities as the Committee +deems necessary for fulfilment of the above or as +may be approved by the Board of Directors or as may +be prescribed by applicable law from time to time. + +The Audit, Risk and Compliance Committee also acts +as the Risk Management Committee in compliance +with the Listing Regulations. The Committee reviews, +acts on and reports to our Board with respect to risk +management matters. The primary responsibilities +include the following: + +a) To formulate a detailed risk management policy + +which shall include: + +• A + +for + +framework + +in particular + +identification of + +internal +and external risks specifically faced by the +Company, +including financial, +operational, sectoral, sustainability (specifically, +Environmental, Social and Governance related +risks and impact), information and cyber security +risks. + +• Measures for risk mitigation + +• Systems for internal controls + +• Business contingency plan + +b) To monitor and oversee implementation of the +risk management policy, including evaluating the +adequacy of risk management and internal control +systems; + +c) Ensure that appropriate methodology, processes +and systems are in place to monitor and evaluate +risks associated with the business of the listed +entity; + + d) Evaluate risks related to cyber security and +significant risk exposures of the Company and +assess steps taken by the management to mitigate +the exposures in a timely manner (including business +continuity and disaster recovery planning); + +e) To review the risk management policy on annual +including by considering the changing + +basis, +industry dynamics and evolving complexity; + +f) To keep the Board of Directors informed about +its discussions, + +the nature and content of +recommendations and actions to be taken; and + +g) The appointment, removal and terms of remuneration +of the Chief Risk Officer (if any) shall be subject to +review by the Audit, Risk and Compliance Committee, +jointly with the Board Governance, Nomination and +Compensation Committee. + +Mr. Deepak Acharya has been appointed as the Chief +Risk Officer of the Company. + +The detailed charter of the Committee is posted +on our website and available at https://www.wipro. +com/investors/corporate-governance/. All members +of our Audit, Risk and Compliance Committee are +Independent Directors and financially literate. The +Chairman of our Audit, Risk and Compliance Committee +has the accounting and financial management related +expertise. + +Statutory Auditors as well as Internal Auditors always +have independent meetings with the Audit, Risk and +Compliance Committee and also participate in the +Audit, Risk and Compliance Committee meetings. +Our Chief Financial Officer, General Counsel, Internal +Auditor, Finance Controller and other Corporate Officers +make periodic presentations to the Audit, Risk and +Compliance Committee on various issues. + +Mr. M. K. Sharma, Independent Director, is the Chairman +of the Audit, Risk and Compliance Committee. The other +members of the Committee as at March 31, 2021 were +Mrs. Ireena Vittal and Mr. Deepak M. Satwalekar. The +Chairman of the Committee was present at the AGM +held on July 13, 2020. + +2. Board Governance, Nomination and Compensation + +Committee +The Board Governance, Nomination and Compensation +Committee reviews, acts on and reports to our Board +with respect to various governance, nomination and +compensation matters. The primary responsibilities +include: + +a) Developing and recommending to the Board +corporate governance guidelines applicable to the +Company; + +b) Evaluating the Board on a continuing basis, +including an assessment of the effectiveness of the +full Board, operations of the Board Committees and +contributions of individual directors; + +c) Establishing policies and procedures to assess the +requirements for induction of new members to the +Board; + +d) Implementing policies and processes relating to + +corporate governance principles; + +e) Ensuring that appropriate procedures are in place +to assess Board membership needs and Board +effectiveness; + +f) Reviewing the Company’s policies that relate to +matters of corporate social responsibility (CSR), +including public +issues of significance to the +Company and its shareholders; + +g) Formulating the Disclosure Policy, its review and + +approval of disclosures; + +h) Approving and evaluating the compensation plans, +policies and programs for full-time directors and +senior management; + +i) Acting as Administrator of the Company’s Employee +Stock Option Plans and Employee Stock Purchase +Plans drawn up from time to time; and + +j) Reviewing and recommending of all remuneration, +in whatever form, payable to senior management. + +The detailed charter of Board Governance, Nomination +and Compensation Committee is posted on our website +and is available at https://www.wipro.com/investors/ +corporate-governance/. + +Our Chief Human Resources Officer makes periodic +presentations to the Board Governance, Nomination and +Compensation Committee on compensation reviews and +performance linked compensation recommendations. +All members of the Board Governance, Nomination +and Compensation Committee are independent non- +executive directors. The Board Governance, Nomination +and Compensation Committee is the apex body that +oversees our Corporate Social Responsibility policy and +programs. + +Mr. William Arthur Owens, Independent Director, is the +Chairman of the Board Governance, Nomination and +Compensation Committee. The other members of the +Committee as at March 31, 2021 were Mrs. Ireena Vittal +and Mr. M. K. Sharma. The Chairman of the Committee +was present at the AGM held on July 13, 2020. + +3. Administrative and Shareholders/Investors Grievance +Committee (Stakeholders Relationship Committee) +The +Grievance Committee carries out + +and Shareholders/Investors +role of + +Administrative + +the + +129 + + Corporate Governance Report + +Stakeholders Relationship Committee in compliance +with Section 178 of the Companies Act, 2013 and the +Listing Regulations. + +Administrative + +The +and Shareholders/Investors +Grievance Committee reviews, acts on and reports to +our Board with respect to various matters relating to +stakeholders. The primary responsibilities include: + +a) Redressal of Grievances of the shareholders of the +Company pertaining to transfer or transmission +of shares, non-receipt of annual report and +declared dividends, issue of new or duplicate share +certificates, and grievances pertaining to corporate +actions; + +b) Approving consolidation, split or sub-division of +share certificates, transmission of shares, issue of +duplicate share certificates, re-materialization of +shares; + +c) Reviewing the grievance redressal mechanism +implemented by the Company in coordination with +Company’s Registrar and Transfer Agent (“RTA”) +from time to time; + +d) Reviewing the measures taken by the Company for +effective exercise of voting rights by shareholders; + +e) Implementing and overseeing the procedures and +processes in handling and maintenance of records, +transfer of securities and payment of dividend by +the Company, RTA and dividend processing bank; + +f) Reviewing the various measures and initiatives +taken by the Company for reducing the quantum of +unclaimed dividends and ensuring timely receipt +of dividend warrants, annual reports and statutory +notices by the shareholders of the Company. + +g) Overseeing administrative matters like opening +and closure of Company’s bank accounts, grant and +revocation of general, specific and banking powers +of attorney; and + +h) Considering and approving allotment of equity +shares pursuant to exercise of stock options, setting +up branch offices and other administrative matters +as delegated by Board from time to time. + +The detailed charter of the Committee is available +on our website at https://www.wipro.com/investors/ +corporate-governance/. + +Mr. M. K. Sharma, Independent Director, is the Chairman +of the Administrative and Shareholders/Investors +Grievance Committee. The other members of the +Committee as at March 31, 2021 were Mr. Deepak M. +Satwalekar and Mr. Rishad A. Premji. The Chairman +of the Committee was present at the Annual General +Meeting of the Company held on July 13, 2020. + +Mr. M. Sanaulla Khan, Company Secretary, is our +Compliance Officer under the Listing Regulations. + +Status Report of investor queries and complaints for +the period from April 1, 2020 to March 31, 2021 is given +below: + +Sl. +No. +1. + +2. + +3. + +4. + +Particulars + +No. of +Complaints + +Investor complaints pending at the +beginning of the year +Investor complaints received during +the year +Investor complaints disposed of +during the year +Investor +complaints +unresolved at the end of the year + +remaining + +NIL + +1,374* + +1,332 + +42** + +* + +of the 1,374 complaints received, 921 were clarifications +regarding buyback of equity shares. + +** These queries were received between March 25, 2021 +and March 31, 2021, and subsequently responded before +April 6, 2021. + +Apart from these queries/complaints, there are certain +pending cases relating to dispute over title to shares in +which in certain cases the Company has been made a +party. However, these cases are not material in nature. + +4. Strategy Committee + +The Strategy Committee reviews, acts and reports to our +Board with respect to the mission, vision and strategic +direction of the Company. Primary responsibilities of +this Committee, inter alia, are: + +a) Making recommendations to the Board relating to +the Company’s mission, vision, strategic initiatives, +major programs and services; + +b) Ensuring management has established an effective +strategic planning process, including development +of a three to five-year strategic plan with measurable +goals and time targets; + +c) Annually reviewing the strategic plan for the +Company and for each division and entity as well +and recommending updates to the Board; + +d) Establishing criteria for management to evaluate +potential strategic investments, reviewing proposals +for acquisition or divestment opportunities for the +Company and making appropriate recommendations +to the Board, and reviewing post-transaction +integration matters; and + +e) Monitoring the Company performance against +measurable targets (e.g. market share, increase in +revenue, or Operating Margin) or progress points +(such as emerging technologies). + +130 + + Mr. William Arthur Owens, Independent Director, is the Chairman of the Strategy Committee. The other members of the +Committee as at March 31, 2021 were Mr. Azim H. Premji, Mrs. Ireena Vittal, Dr. Patrick J. Ennis, Mr. Patrick Dupuis, +Mr. Thierry Delaporte and Mr. Rishad A. Premji. + +At the Board Meeting held on April 15, 2021, it was decided to discontinue the Strategy Committee, as the entire Board +will be deliberating matters pertaining to the strategy of the Company going forward. + +Attendance of Directors at Board and Committee meetings +Details of attendance of Directors at the Board meetings and Committee meetings for the year ended March 31, 2021 +are as under: + +Board meeting1 + +Audit, Risk and +Compliance +Committee2 + +Board Governance, +Nomination and +Compensation +Committee (also acts as +CSR Committee) + +Strategy +Committee + +Administrative +and Shareholders/ +Investors Grievance +Committee3 + +6 + +5 + +5 + +2 + +4 + +April 15, 2020, +May 29, 2020, +July 14, 2020, +October 12-13, +2020, January 13, +2021 and February +24, 2021 + +April 14, 2020, +May 28, 2020, +July 13, 2020, +October 12-13, +2020 and January +12-13, 2021 + +April 14, 2020, May 29, +2020, July 13, 2020, +October 12, 2020 and +January 12, 2021 + +October 12, +2020 and +January 12, +2021 + +April 14, 2020, July +13, 2020, October +12, 2020 and +January 12, 2021 + +6 +6 +4 +6 +6 +6 +6 +6 +4 + +2 + +1 + +NA +NA +NA +NA +5 +5 +NA +NA +3 + +2 + +NA + +NA +NA +NA +5 +5 +5 +NA +NA +NA + +NA + +NA + +2 +2 +2 +2 +NA +2 +2 +2 +NA + +NA + +NA + +4 +NA +NA +NA +3 +NA +NA +NA +3 + +1 + +NA + +No. of meetings held +during FY 2020-21^ +Date of meetings + +Attendance of +Directors +Rishad A. Premji +Azim H. Premji +Thierry Delaporte# +William Arthur Owens +M. K. Sharma +Ireena Vittal +Dr. Patrick J. Ennis +Patrick Dupuis +Deepak M. +Satwalekar* +Arundhati +Bhattacharya $ +Abidali Z. +Neemuchwala @ + +# At the 74th AGM, Mr. Thierry Delaporte was appointed as the Chief Executive Officer and Managing Director of the Company to hold office for a + +period of five years from July 6, 2020 to July 5, 2025. + +* At the 74th AGM, Mr. Deepak M. Satwalekar was appointed as an Independent Director for a term of five years from July 1, 2020 to June 30, + +2025. + +$ The Board of Directors, at its meeting held on May 29, 2020, noted the resignation of Mrs. Arundhati Bhattacharya as an Independent Director +with effect from close of business hours on June 30, 2020. In her resignation letter, Mrs. Arundhati Bhattacharya has indicated that the reason +for her resignation is her decision to accept a full time role as Chairperson and CEO in another company. She has also confirmed that there is +no other material reason, other than the reason stated in her resignation letter. + +@ The Board of Directors, at its meeting held on May 29, 2020, noted the resignation of Mr. Abidali Z. Neemuchwala as the Chief Executive Officer + +and Managing Director of the Company with effect from the end of the day on June 1, 2020. + +^ Pursuant to the relaxations granted by the Ministry of Corporate Affairs and SEBI, all meetings of the Board of Directors and the Committees + +were held through video conferencing. + +1. Board Meeting: Since the appointment of Mr. Thierry Delaporte as Chief Executive Officer and Managing Director and Mr. Deepak M. Satwalekar + +as Independent Director, four Board meetings were held on July 14, 2020, October 12-13, 2020, January 13, 2021 and February 24, 2021. + +2. Audit, Risk and Compliance Committee: + +i. The Committee was re-constituted during the year as Mrs. Arundhati Bhattacharya resigned and Mr. Deepak M. Satwalekar was appointed + +131 + + Corporate Governance Report + +as a member of the Committee. Consequently, the composition of the Committee is as follows: Mr. M. K. Sharma (Chairman), Mrs. Ireena +Vittal and Mr. Deepak M. Satwalekar (Members). + +ii. Since the appointment of Mr. Deepak M. Satwalekar as member of the Committee, there were three Committee meetings held on July 13, + +2020, October 12-13, 2020, and January 12-13, 2021. + +3. Administrative and Shareholders/Investors Grievance Committee: + +i. The Committee was re-constituted during the year as Mrs. Arundhati Bhattacharya resigned and Mr. Deepak M. Satwalekar was appointed +as a member of the Committee. Consequently, the composition of the Committee is as follows: Mr. M. K. Sharma (Chairman), Mr. Deepak M. +Satwalekar and Mr. Rishad A. Premji (Members). + +ii. Since the appointment of Mr. Deepak M. Satwalekar as member of the Committee, there were three Committee meetings held on July 13, + +2020, October 12, 2020 and January 12, 2021. + +V. GOVERNANCE +PROCESS + +THROUGH + +MANAGEMENT + +1. Code of Business Conduct + +In the year 1983, we articulated ‘Wipro Beliefs’ consisting +of six statements. At the core of beliefs was integrity, +articulated as “individual and company relationship +should be governed by the highest standard of conduct +and integrity”. + +Over years, this articulation has evolved in form but +remained constant in substance. Today, we articulate it +as Code of Business Conduct. + +In our Company, the Board and all employees have +a responsibility to understand and follow the Code +of Business Conduct. All employees are expected to +perform their work with honesty and integrity. Wipro’s +Code of Business Conduct reflects general principles to +guide employees in making ethical decisions. This Code +is also applicable to our representatives. This Code +outlines fundamental ethical considerations as well +as specific considerations that need to be maintained +for professional conduct. This Code has been displayed +on the Company’s website at https://www.wipro.com/ +investors/corporate-governance/. + +Internal Audit +The Company has a robust internal audit function which +has been in place for last 4 decades with the stated +vision of “To be the best in class Internal Audit function +globally”. In pursuit of this vision, the function provides +an independent, objective assurance services to value- +add and improve Operations of Business Units and +processes by: + +a) Financial, Business Process and Compliance Audit + +b) Cyber Defense and Technology Audit + +c) Operations Reviews + +d) Best Practices and Benchmarking + +e) Leadership Development + +The function taking cognizance of changes in business +climate and technology risks has taken upon itself to +infuse and adopt Technology in its operations. + +2. + +132 + +The Chief of Internal Audit reports to the Chairman +of the Audit, Risk and Compliance Committee and +administratively to the Chief Financial Officer. Chief of +Internal Audit has regular and exclusive meetings with +the Audit, Risk and Compliance Committee. + +The internal audit function is guided by its charter, as +approved by the Audit, Risk and Compliance Committee. +The internal audit function formulates an annual risk +based audit plan based on consultations and inputs +from the Board and business leaders and presents +it to the Audit, Risk and Compliance Committee for +approval. Findings of various audits carried out during +the financial year are also periodically presented to the +Audit, Risk and Compliance Committee. The internal +audit function adopts a risk based audit approach and +covers core areas such as compliance audits, financial +audits, technology audits, third party risk audits, M&A +audits, etc. + +The internal audit team comprises of personnel with +professional qualifications and certifications in audit +and is rich on diversity. The audit team hones its skills +through a robust knowledge management program to +continuously assimilate the latest trends and skills +in the domain and to retain the knowledge gained for +future reference and dissemination. The internal audit +team re-asserts its independence across all its staff. + +A key strategic vision of Internal Audit is auditing in the +new digital environment “Auditing Digital and Auditing +with Digital”- in line with this, the Internal Audit function +has actively adopted Technology and Innovation to be +better equipped to carry out audits. + +The function, which was the first Indian Internal audit +unit to get ISO certified in 1998 and win International +award from Institute of Internal Auditors (IIA) in 2002, +was also an early adopter of the new ISO 9001:2015 +Version. During the year Internal Audit function is +assessed to have “Met +International Standards” +prescribed by the Professional Practice of Internal +Auditing issued by “International Institute of Internal +Auditors (IIA)” by external firm (KPMG). Testimony to +the functions’ innovation and excellence are the IIA +awards won in these categories continuously over the +last few years. + + 3. Disclosure Policy + +In line with requirements under Regulation 30 of the +Listing Regulations, the Company has framed a policy +on disclosure of material events and information +as per the Listing Regulations, which is available on +our website at https://www.wipro.com/investors/ +corporate-governance/. The objective of this policy is to +have uniform disclosure practices and ensure timely, +adequate and accurate disclosure of information on +an ongoing basis. The Company has constituted a +Disclosure Committee consisting of senior officials, +which approves all disclosures required to be made +by the Company. Parity in disclosures are maintained +through simultaneous disclosure on National Stock +Exchange of India Limited, the BSE Limited and the New +York Stock Exchange. + +4. Policy for Preservation of Documents + +Pursuant to the requirements under Regulation 9 of +the Listing Regulations, the Board has formulated and +approved a Document Retention Policy prescribing the +manner of retaining the Company’s documents and +the time period up to which certain documents are to +be retained. The policy percolates to all levels of the +organization who handle the prescribed categories of +documents. + +5. Other Policies + +The Company has adopted an Ombuds policy (vigil +mechanism), a policy for prevention, prohibition & +redressal of sexual harassment of women at workplace, +as well as a code of conduct to regulate, monitor and +report insider trading. Details of these are provided as +part of the Board’s report. + +VI. DISCLOSURES + +1. Disclosure of Materially Significant Related Party + +Transactions + +All related party transactions entered during the +financial year were at an arm’s length basis and in the +ordinary course of business. There are no materially +significant related party transactions made by the +Company with Promoters, Directors, Key Managerial +Personnel or other designated persons which may have +a potential conflict with the interest of the Company at +large. + +As required under Regulation 23 of Listing Regulations, +the Company has adopted a policy on Related Party +Transactions. The abridged policy on Related Party + +Transactions is available on the Company’s website at +https://www.wipro.com/content/dam/nexus/en/ +i nv e s to r /c o r p o ra te - go v e r n a n c e /p o l i c i e s - a n d - +guidelines/ethical-guidelines/abridged-policy-for- +related-party-transactions.pdf. + +Apart from receiving director remuneration, none +of the Directors has any pecuniary relationships or +transactions vis-à-vis the Company. During the year +2020-21, no transactions of material nature were +entered by the Company with the Management or their +relatives that may have a potential conflict of interest +with the Company and the concerned officials have +given undertakings to that effect as per the provisions +of the Listing Regulations. + +The Register under Section 189 of the Companies Act, +2013 is maintained and particulars of the transactions +have been entered in the Register, as applicable. + +2. Subsidiary Monitoring Framework + +All the subsidiary companies of the Company are +managed by their Boards having the rights and +obligations to manage these companies in the best +interest of respective stakeholders. The Company +its representatives on the Board of +nominates +subsidiary companies and monitors performance of +such companies, inter alia, by reviewing: + +a) Financial statements, + +the + +investment, + +inter- +corporate loans/advances made by the unlisted +subsidiary companies, statement containing all +significant transactions and arrangements entered +by the unlisted subsidiary companies forming part +of the financials being reviewed by the Audit, Risk +and Compliance Committee of the Company on a +quarterly basis. + +b) Minutes of the meetings of the unlisted subsidiary +companies, if any, are placed before the Company’s +Board regularly. + +c) Providing necessary guarantees, letter of comfort +and other support for their day-to-day operations +from time-to-time. + +As required under Regulation 16(1)(c) and 24 of +Listing Regulations, the Company has adopted a +policy on determining “material subsidiary” and the +said Policy is available on the Company’s website at +https://www.wipro.com/content/dam/nexus/en/ +i nv e s to r /c o r p o ra te - go v e r n a n c e /p o l i c i e s - a n d - +guidelines/ethical-guidelines/abridged-policy-for- +related-party-transactions.pdf. + +133 + + Corporate Governance Report + +3. Details of non-compliance by the Company, +penalties, and strictures imposed on the Company +by Stock Exchanges or SEBI or any statutory +authority, on any matter related to capital markets, +during the last three years. + +The Company has complied with the requirements of +the Stock Exchanges or SEBI on matters related to +Capital Markets, as applicable, during the last three +years. No penalties or strictures have been imposed on +the Company. + +4. Whistle Blower Policy and affirmation that no +personnel have been denied access to the Audit, +Risk & Compliance Committee + +As mentioned earlier in this report, the Company has +adopted an Ombuds process which is a channel for +receiving and redressing employees’ complaints. No +personnel in the Company has been denied access +to the Audit, Risk and Compliance Committee or its +Chairman. + +5. Transfer to Investor Education and Protection + +Fund Authority + +a) Pursuant to the provisions of Section 124(6) of the +Companies Act, 2013 and Investor Education and +Protection Fund (IEPF) Rules, during the year 2020-21, +an amount of ` 1,255,390 pertaining to unclaimed sale +proceeds of fractional shares arising out of scheme of +arrangement between Wipro Limited, Wipro Enterprises +Limited (formerly known as Azim Premji Custodial +Services Private Limited) and Wipro Trademarks Holding +Limited, was transferred to the IEPF. + +b) Pursuant to the provisions of Section 124(6) of the +Companies Act, 2013 and IEPF rules, during the +year 2020-21, unclaimed dividend for financial years +2012-13 and 2013-14 of ` 10,019,025 and ` 6,925,143 +respectively, together with an aggregate of 227,395 +equity shares in respect of which dividend had not been +claimed were transferred to the IEPF. + +6. Disclosures with respect to demat suspense +account/unclaimed suspense account (Unclaimed +Shares) + +In accordance with Regulation 39 and Schedule VI of the +Listing Regulations, a minimum of three reminders are +sent to shareholders, towards the shares which remain +unclaimed. In case of non-receipt of response to the +reminders from the shareholders, the unclaimed shares +are transferred to the Unclaimed Suspense Account. +The Company maintains the details of shareholding +of each individual shareholders whose shares are +transferred to the Unclaimed Suspense Account. When +a claim from a shareholder is received by the Company, +the shares lying in the Unclaimed Suspense Account + +134 + +are transferred after due verification of documents +submitted by the shareholder. +Further, the shares in respect of which dividend +entitlements remained unclaimed for seven consecutive +years are transferred from the Unclaimed Suspense +Account to IEPF Authority in accordance with Section +124(6) of the Companies Act, 2013 and rules made +thereunder. +The disclosure as required under Schedule V of the +Listing Regulations is given below for the financial year +2020-21: + +Sl. +No. +1. + +2. + +3. + +4. + +5. + +6. + +Particulars + +number +and + +of +Aggregate +shareholders +the +outstanding shares in the +suspense account lying at +the beginning of the year +Number of shareholders +who +the +approached +Company for transfer of +suspense +from +shares +account during the year +Number of shareholders +to whom shares were +transferred from suspense +account during the year +Number of shares +in +respect of which dividend +remained +entitlements +for +unclaimed +seven +years and +consecutive +the +transferred +from +Unclaimed +Suspense +Account to the IEPF +of +Aggregate +shareholders +the +outstanding shares in the +suspense account lying at +the end of the year +Voting +these +rights on +shares shall remain frozen +till the rightful owner of +such shares claim +the +same + +number +and + +No. of +Shareholders +295 + +No. of +Shares +271,916 + +Nil + +Nil + +Nil + +Nil + +- + +4,920* + +295 + +266,996 + +Yes + +* Represents a portion of the shares held by 29 shareholders, +whose balance shares continue to remain in the unclaimed +suspense account. + +7. Shareholder Information +Various shareholder +to be +disclosed pursuant to Schedule V of the Listing +Regulations are provided in Annexure I to this report. + +information + +required + +8. Compliance with Mandatory Requirements + +Your Company has complied with all the mandatory +corporate governance requirements under the Listing +Regulations. Specifically, your Company confirms +compliance with corporate governance requirements + + specified in Regulation 17 to 27 and clauses (b) to (i) +of Sub-regulation (2) of Regulation 46 of the Listing +Regulations. + +9. Certificates from Practising Company Secretary + +The certificate dated June 9, 2021, +issued by +Mr. V. Sreedharan, Partner, V. Sreedharan & Associates, +Practising Company Secretaries is given at page no. +142 of this Annual Report in compliance with corporate +governance norms prescribed under the Listing +Regulations. + +The Company has received certificate dated June 9, +2021, from Mr. V. Sreedharan, Partner, V. Sreedharan & +Associates, Practising Company Secretaries, confirming +that none of the Directors of the Company have been +debarred or disqualified from being appointed or +continuing as director of companies by the SEBI/ +Ministry of Corporate of Affairs or any such authority. +The certificate is given at page no. 143 of this Annual +Report. + +VII. COMPLIANCE REPORT ON DISCRETIONARY +REQUIREMENTS UNDER REGULATION 27(1) OF +THE LISTING REGULATIONS + +1. + +The Board +As per Para A of Part E of Schedule II of the Listing +Regulations, a non-executive Chairman of the Board +may be entitled to maintain a Chairman’s Office at the +company’s expense and also allowed reimbursement +of expenses incurred in performance of his duties. The +Chairman of the Company is an Executive Director and +hence this provision is not applicable to us. + +2. Shareholders rights + +Considering the dynamic shareholder demography and +trading on the stock exchanges, as a prudent measure, +we display our quarterly and half yearly results on our +website www.wipro.com and also publish our results in +widely circulated newspapers. We have communicated +the payment of dividend by e-mail to shareholders in + +addition to dispatch of letters to all shareholders. We +publish the voting results of shareholder meetings +and make it available on our website www.wipro.com, +and report the same to Stock Exchanges in terms of +Regulation 44 of the Listing Regulations. + +3. Modified opinion(s) in audit report + +The Auditors have issued an unmodified opinion on the +financial statements of the Company. + +4. Reporting of Internal Auditor + +Reporting of Head of Internal Audit is to the Chairman +of the Audit, Risk and Compliance Committee and +administratively to the Chief Financial Officer. Head of +Internal Audit has regular and exclusive meetings with +the Audit Committee. + +5. NYSE Corporate Governance Listing Standards + +The Company has made necessary disclosures in +compliance with the New York Stock Exchange Listing +Standards and NYSE Listed Company Manual on its +website https://www.wipro.com/investors/ +corporate-governance + +Bengaluru +June 9, 2021 + +Rishad A. Premji +Chairman + +Declaration as required under Regulation 34(3) and +Schedule V of the Listing Regulations + +All Directors and senior management personnel of the +Company have affirmed compliance with Wipro’s Code of +Business Conduct for the financial year ended March 31, +2021. + +Bengaluru +June 9, 2021 + +Rishad A. Premji +Chairman + +Thierry Delaporte +Chief Executive Officer and Managing Director + +135 + + SHAREHOLDER INFORMATION + +Annual General Meeting + +ANNEXURE I + +Pursuant to the General Circular No. 14/2020 dated April 8, 2020, the General Circular No. 17/2020 dated April 13, 2020, the +General Circular No. 20/2020 dated May 5, 2020, the General Circular No. 22/2020 dated June 15, 2020, the General Circular +No. 33/2020 dated September 28, 2020, the General Circular No. 39/2020 dated December 31, 2020 and the General Circular +No. 02/2021 dated January 13, 2021 issued by the Ministry of Corporate Affairs (“MCA Circulars”), the 75th Annual General +Meeting (“AGM”) for the year ended March 31, 2021 is scheduled to be held on Wednesday, July 14, 2021 at 9 am IST through +Video Conferencing (“VC”). + +The Members may attend the 75th AGM scheduled to be held on Wednesday, July 14, 2021, 9 am IST onwards, through VC +or watch the live web-cast at https://www.wipro.com/AGM2021/. Detailed instructions for participation are provided in the +notice of the 75th AGM. The proceedings of the 75th AGM will be available through VC and live web-cast to the shareholders as +on the cut-off date i.e. July 7, 2021. + +Annual General Meetings and Other General Body meeting of the last three years and Special Resolutions, if any. + +Financial Year +2017-18 + +2018-19 + +Date and Time +July 19, 2018 +at 4.00 pm +July 16, 2019 +at 4.00 pm + +2019-20 + +July 13, 2020 +at 9.00 am + +Venue +Wipro Campus, Cafeteria +Hall EC-3, Ground Floor, +Opp. Tower 8, No. 72, +Keonics, Electronic City, +Hosur Road, +Bengaluru – 561 229 +Meeting held through VC + +Special resolutions passed +i. Re-appointment of Mrs. Ireena Vittal (DIN: 05195656) as Independent + +Director of the Company. + +i. Amendments to the Articles of Association of the Company +ii. Appointment of Mr. Azim H. Premji (DIN: 00234280) as Non-Executive, + +Non-Independent Director of the Company + +i. Appointment of Mr. Deepak M. Satwalekar (DIN: 00009627) as an + +Independent Director of the Company + +Details of resolutions passed through postal ballot during Financial Year 2020-21 and details of the voting pattern +The Company sought the approval of shareholders through notice of postal ballot dated October 13, 2020 for approval +for Buyback of Equity Shares by way of special resolution. The aforesaid resolutions were duly passed and the results of +postal ballot/e-voting were announced on November 17, 2020. Mr. V. Sreedharan/Ms. Devika Sathyanarayana/Mr. Pradeep +B. Kulkarni, partners of V. Sreedharan & Associates, Practicing Company Secretaries, were appointed as the Scrutinizer to +scrutinize the postal ballot and remote e-voting process in a fair and transparent manner. + +Resolution + +No. of Votes +Polled + +No. of Votes Cast +in Favour + +No. of Votes Cast +Against + +% of Votes Cast in +Favour on Votes +Polled + +% of Votes Cast +Against on Votes +Polled + +Approval for Buyback of Equity +Shares + +5,089,856,744 + +5,078,618,396 + +11,238,348 + +99.78 + +0.22 + +Procedure for Postal Ballot +The postal ballot is conducted in accordance with the provisions contained in Section 110 and other applicable provisions, +if any, of the Companies Act, 2013, read with Rule 22 of the Companies (Management and Administration) Rules, 2014. The +Shareholders are provided the facility to vote either by physical ballot or through e-voting. The postal ballot notice is sent to +shareholders as per the permitted mode wherever applicable. The Company also publishes a notice in the newspapers in +accordance with the requirements under the Companies Act, 2013. + +Shareholders holding equity shares as on the cut-off date may cast their votes through e-voting or through postal ballot +during the voting period fixed for this purpose. After completion of scrutiny of votes, the scrutinizer submits his report to the +Chairman and the results of voting by postal ballot are announced within 2 working days of conclusion of the voting period. +The results are displayed on the website of the Company (www.wipro.com), and communicated to the Stock Exchanges, +Depositories, and Registrar and Share Transfer Agents. The resolutions, if passed by the requisite majority, are deemed to +have been passed on the last date specified for receipt of duly completed postal ballot forms or e-voting. + +In view of the COVID-19 pandemic, the MCA permitted companies to transact items through postal ballot as per the framework +set out in Circular No. 14/2020 dated April 8, 2020, Circular No. 17/2020 dated April 13, 2020, the General Circular No. 22/2020 +dated June 15, 2020, Circular No. 33/2020 dated September 28, 2020 and General Circular No. 39/2020 dated December 31, +2020. In accordance with the aforementioned circulars, e-voting facility was provided to all the shareholders to cast their +votes only through the remote e-voting process as per notice of postal ballot dated October 13, 2020 for approval for Buyback +of Equity Shares. + +136 + + Further, the Company sought the approval of shareholders through notice of postal ballot dated April 15, 2021 for the re- +appointment of Dr. Patrick J. Ennis (DIN: 07463299) and Mr. Patrick Dupuis (DIN: 07480046) as Independent Directors on the +Board of Wipro Limited by way of special resolutions. The aforesaid resolutions were duly passed and the results of postal +ballot/e-voting were announced on June 6, 2021. Mr. V. Sreedharan/Ms. Devika Sathyanarayana/Mr. Pradeep B. Kulkarni, +partners of V. Sreedharan & Associates, Practicing Company Secretaries, were appointed as the Scrutinizer to scrutinize the +postal ballot and remote e-voting process in a fair and transparent manner. + +Resolution + +No. of Votes +Polled + +No. of Votes Cast +in Favour + +No. of Votes Cast +Against + +% of Votes Cast in +Favour on Votes +Polled + +% of Votes Cast +Against on Votes +Polled + +Re-appointment of Dr. Patrick +J. Ennis (DIN: 07463299) as an +Independent Director of the +Company + +Re-appointment of Mr. Patrick +Dupuis (DIN: 07480046) as an +Independent Director of the +Company + +4,905,658,196 + +4,869,602,676 + +36,055,520 + +99.27 + +0.73 + +4,905,658,087 + +4,880,958,490 + +24,699,597 + +99.50 + +0.50 + +Means of Communication with Shareholders/Analysts +We have established procedures to disseminate, in a planned manner, relevant information to our shareholders, analysts, +employees and the society at large. Our Audit, Risk and Compliance Committee reviews the earnings press releases, Form 20-F +filed with Securities Exchange Commission (SEC) filings and annual and quarterly reports of the Company, before they are +presented to the Board for their approval for release. The details of the means of communication with shareholders/analysts +are given below: + +News Releases and +Presentations +Quarterly results + +Website + +Annual Report + +Other Disclosures/Filings + +All our news releases and presentations made at investor conferences and to analysts are posted on the +Company’s website at https://www.wipro.com/investors. +Our quarterly results are published in widely circulated national newspapers such as Financial Express +and the local daily Kannada Prabha. +The Company’s website contains a dedicated section for Investors (https://www.wipro.com/investors), +where annual reports, earnings press releases, stock exchange filings, quarterly reports, and corporate +governance policies are available, apart from the details about the Company, Board of Directors +and Management. +Annual Report containing audited standalone accounts, consolidated financial statements together +with Board’s Report, Corporate Governance Report, Management Discussion and Analysis Report, +Auditors Report and other important information are circulated to the Members entitled thereto through +permitted mode(s). +Our Form 20-F filed with SEC containing detailed disclosures, along with other disclosures including +Press Releases etc. are available at https://www.wipro.com/investors. + +Communication of Results: +Means of Communications +Earnings Calls +Publication of results +Analysts/Investors Meetings/Analyst Day + +Number of times during 2020-21 +4 +4 +Details are provided in the MD&A Report forming part of this Annual Report. + +Financial Calendar +The financial year of the Company starts from the 1st day of April and ends on 31st day of March of next year. Our tentative +calendar for declaration of results for the financial year 2021-22 are as given below. In addition, the Board may meet on other +dates as and when required. + +Quarter Ending +For the Quarter ending June 30, 2021 +For the Quarter and half year ending September 30, 2021 +For the Quarter and nine months ending December 31, 2021 +For the year ending March 31, 2022 + +Release of Results + +Third week of July, 2021 +Third week of October, 2021 +Third week of January, 2022 +Third week of April, 2022 + +The Register of Members and Share Transfer books will remain closed from Monday, July 12, 2021 to Tuesday, July 13, 2021 +(both days inclusive). + +137 + + Fees Paid to Statutory Auditors +The details of total fees for all services paid by the Company and its subsidiaries, on a consolidated basis, to the statutory +auditor and all the entities in the network firm/network entity of which the statutory auditor is a part, are as follows: + +Type of Service +Audit Fees +Tax Fees +Others + +Total + + (In ` Mn) + +FY 2020-21 +83 +44 +13 + +FY 2019-20 +81 +51 +17 + +140 + +149 + +Corporate Information +a) Corporate Identity Number (CIN): L32102KA1945PLC020800 +b) Company Registration Number: 20800 +c) International Securities Identification Number (ISIN): INE075A01022 +d) CUSIP Number for Wipro American Depository Shares: 97651M109 +e) Details of exchanges where Company’s shares are listed in as at March 31, 2021: + +Equity shares +BSE Limited (BSE) + +Stock Codes +507685 + +National Stock Exchange of India Limited (NSE) + +WIPRO + +Address +BSE Limited, Phiroze Jeejeebhoy Towers, Dalal Street, +Mumbai - 400 001 +Exchange Plaza, C-1, Block G, Bandra Kurla Complex, +Bandra (E), Mumbai – 400 051 + +American Depository Receipts +New York Stock Exchange (NYSE) + +WIT + +11 Wall St, New York, NY 10005, United States of America + +Notes: +1. Listing fees for the year 2021-22 has been paid to the Indian Stock Exchanges as on date of this report. +2. Listing fees to NYSE for the calendar year 2021 has been paid as on date of this report. +3. The stock code on Reuters is WIPR.NS and on Bloomberg is WPRO:IN + +Distribution of Shareholding as on March 31, 2021 + +March 31, 2021 + +March 31, 2020 + +Category + (No. of Shares) +1-5000 +5001- 10000 +10001- 20000 +20001- 30000 +30001- 40000 +40001- 50000 +50001- 100000 +100001 & Above + +Total + +No. of +Shareholders +829,647 +3,628 +1,936 +751 +371 +261 +668 +1,329 + +% of +Shareholders +98.93 +0.43 +0.23 +0.09 +0.04 +0.03 +0.08 +0.17 + +No. of Shares +83,119,655 +12,699,508 +13,864,634 +9,354,600 +6,483,397 +5,881,802 +23,626,382 +5,324,108,577 + +% of Total +Equity + +1.52 +0.23 +0.25 +0.17 +0.12 +0.11 +0.43 +97.17 + +No. of +Shareholders +507,272 +1,697 +1,049 +400 +240 +162 +358 +703 + +% of +Shareholders +99.10 +0.33 +0.20 +0.08 +0.05 +0.03 +0.07 +0.14 + +No. of Shares + +82,232,977 +12,032,833 +14,869,292 +9,808,813 +8,376,285 +7,250,873 +25,599,263 +5,553,187,054 + +% of Total +Equity + +1.44 +0.21 +0.26 +0.17 +0.15 +0.13 +0.45 +97.20 + +838,591 + +100.00 + +5,479,138,555 + +100.00 + +511,881 + +100.00 + +5,713,357,390 + +100.00 + +Market Share Price Data +The performance of our stock in the financial year 2020-21 is tabulated below: + +2020-21 + +April +133,037,800 + +May +151,908,123 + +June +195,006,806 + +July +460,217,867 + +August +195,174,626 + +September +435,740,889 + +October +587,943,024 + +November +148,723,864 + +December +241,053,563 + +January +467,630,831 + +February +225,393,445 + +March +260,495,775 + +202.95 +09-Apr-20 +3,465,831 +173.8 +21-Apr-20 + +214.9 +29-May-20 +30,630,218 +178 +19-May-20 + +230 +26-Jun-20 +19,194,612 +206.3 +12-Jun-20 + +290.8 +31-Jul-20 +16,700,905 +218.35 +01-Jul-20 + +287.4 +18-Aug-20 +12,054,273 +268 +31-Aug-20 + +NSE +324.5 +21-Sep-20 +22,739,047 +269 +01-Sep-20 + +381.7 +13-Oct-20 +36,092,383 +312.05 +01-Oct-20 + +361.4 +24-Nov-20 +10,092,441 +332.65 +03-Nov-20 + +390.5 +29-Dec-20 +11,459,126 +346.25 +01-Dec-20 + +467.45 +13-Jan-21 +29,190,193 +385.05 +01-Jan-21 + +451.75 +09-Feb-21 +29,474,830 +408 +26-Feb-21 + +444.3 +04-Mar-21 +12,963,797 +397.75 +25-Mar-21 + +Volume +traded +on NSE + +High +Date +Volume +Low +Date + +138 + + Volume + +9,043,319 + +5,743,344 + +6,209,370 + +High +Low + +High +Low + +High + +Low + +9,889.05 +8,055.8 + +-13.23% +9.03% + +-13.50% + +7.25% + +9,598.85 +8,806.75 + +10,553.15 +9,544.35 + +5.89% +2.42% + +-2.93% + +9.32% + +7.03% +15.90% + +9.94% + +8.38% + +3,952,617 + +9,257,331 + +20,571,550 + +11,794.25 +10,882.25 + +12,293,054 + S&P CNX NIFTY +11,618.1 +10,790.2 +Wipro Price Movement vis-à-vis Previous Month High/Low (%) +12.91% +0.37% + +12,025.45 +11,347.05 + +17.63% +16.00% + +-1.17% +22.74% + +11,341.4 +10,299.6 + +26.43% +5.84% + +S&P CNX Nifty Index vis-à-vis Previous Month High/Low (%) +-1.49% + +3.99% + +3.51% + +5.66% + +-0.85% + +5.16% + +7.47% + +7.91% + +4,757,798 + +13,145.85 +11,557.4 + +10,693,567 + +5,042,336 + +11,731,941 + +12,737,129 + +14,024.85 +12,962.8 + +14,753.55 +13,596.75 + +15,431.75 +13,661.75 + +15,336.3 +14,264.4 + +-5.32% +6.60% + +9.32% + +1.85% + +8.05% +4.09% + +6.69% + +12.16% + +19.71% +11.21% + +5.20% + +4.89% + +-3.36% +5.96% + +4.60% + +0.48% + +-1.65% +-2.51% + +-0.62% + +4.41% + +ADS Share Price during the Financial Year 2020-21 +August +May + +June + +April + +July + +September + +October + +November December + +January + +February + +March + +3.4 + +3.36 + +3.32 + +4.34 + +4.33 + +4.76 + +4.9 + +5.17 + +5.67 + +6.36 + +6.24 + +6.42 + +8,529.46 + +8,672.52 + +8,851.62 + +9,354.97 + +9,909.38 + +9,594.92 + +9,298.03 + +10,382.02 + +10,720.47 + +10,541.65 + +10,900.66 + +10,941.79 + +8.28% + +-1.18% + +-1.19% + +30.72% + +-0.23% + +9.93% + +2.94% + +5.51% + +9.67% + +12.17% + +-1.89% + +2.88% + +8.04% + +1.68% + +2.07% + +5.69% + +5.93% + +-3.17% + +-3.09% + +11.66% + +3.26% + +-1.67% + +3.41% + +0.38% + +Wipro ADS price in +NYSE during each +month closing + +NYSE TMT index +during each month +closing + +Wipro ADS Price +Movement (%) +vis-à-vis Previous +month Closing + +NYSE TMT Index +movement (%) +vis-à-vis Previous +month closing + +Performance of Wipro equity shares relative to the SENSEX and NYSE Composite index during the period +April 1, 2020 to March 31, 2021 is given in the following chart: + +250 +240 +230 +220 +210 +200 +190 +180 +170 +160 +150 +140 +130 +120 +110 +100 +90 +80 +70 + +0 +2 +0 +2 +- +r +p +A +- +1 + +0 +2 +0 +2 +- +y +a +M +- +1 + +0 +2 +0 +2 +- +y +a +M +- +1 +3 + +0 +2 +0 +2 +- +n +u +J +- +0 +3 + +0 +2 +0 +2 +- l -| | The cornerstone | | of | our approach | involves | con- | | | | | | | | -| --- | --------------- | --- | --- | ------------ | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -4 ExperimentalSettings -| trasting | | a consistent | | answer | with an | inconsis- | | | | | | | | -| --------------- | --- | ------------ | -------------------------- | -------- | ----------- | --------- | ------------ | --- | --- | --- | --- | --- | --- | -| tent | one | under | the | guidance | of the same | oracle | 4.1 Datasets | | | | | | | -| groundedquotes. | | | Thisalignswiththeconceptof | | | | | | | | | | | -WeconductexperimentsontheALCEbenchmark -| Reinforcement | | | Learning | from | Human | Feedback | | | | | | | | -| ------------- | --- | --- | -------- | ---- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | -(Gaoetal.,2023b),designedforattributedtextgen- -| (RLHF) | | (Ouyang | et | al., 2022; | Bai et | al., 2022), | | | | | | | | -| ------ | --- | ------- | --- | ---------- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -eration. Thebenchmarkincludesthreelong-form -whereLLMsarefurtherfine-tunedtodistinguish -QAdatasetsthatspanvarioustypesofquestions. -betweendesirableandundesirableresponsesunder -preference feedback. However, such contrastive ASQA (Stelmakh et al., 2022) is a long-form -preferencefeedbacktypicallycomesfromhuman factoidQAdatasetcharacterizedbyinherentlyam- -| annotation. | | Toautomaticallyconstructpreference | | | | | | | | | | | | -| ----------- | --- | ---------------------------------- | --- | --- | --- | --- | ------- | --------- | ---- | ------- | -------- | ----- | --- | -| | | | | | | | biguous | questions | that | require | multiple | short | an- | -pairsforpreferenceoptimization,weutilizetheat- swerstoencapsulatedifferentviewpoints. -tributedanswersgeneratedbysmallerLLMs(e.g., -LLaMA-2-7B)underthein-contextlearningsetting ELI5 (Fanetal.,2019)featuresopen-endedques- -tionsintendedforsimplificationtothecomprehen- -| asnegativesamples. | | | | Theseanswers,characterized | | | | | | | | | | -| ------------------ | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -sionleveloffive-year-olds,requiringexplanatory -bytheirlowqualityandinconsistencywithoracle -multi-sentenceresponses. -groundedquotes,automaticallyserveascontrastive -supervisionsignalswhenpairedwithhigh-quality -| | | | | | | | QAMPARI | | (Amouyaletal.,2022)isafactoid | | | | | -| ------------------------------- | --- | --- | --- | --- | --------------- | --- | ------- | --- | ----------------------------- | --- | --- | --- | --- | -| attributedanswerslabeledin§3.1. | | | | | Inthisscenario, | | | | | | | | | -QAdatasetderivedfromWikipedia,whereanswers -theprocessnotonlyencouragestheLLMtogener- arestructuredasacompilationofentities. -ateattributedanswersthataremoreconsistentwith -thegroundedquotesbutalsofacilitatestheidentifi- 4.2 EvaluationMetrics -cationandcorrectionofnuancederrorspresentin Following the ALCE benchmark (Gao et al., -smallermodels. 2023b), our evaluation primarily focuses on two -| | Specifically, | | we adopt | Direct | Preference | Opti- | | | | | | | | -| --- | ------------- | --- | -------- | ------ | ---------- | ----- | --------------- | --- | -------- | --- | ------- | ------------ | --- | -| | | | | | | | key dimensions: | | Citation | | Quality | and Correct- | | -mization(Rafailovetal.,2023),avariantofRLHF ness. Detaileddescriptionsofadditionalevaluation -known for its stability, for our contrastive align- dimensionsarepresentedintheAppendixB. -ment. Formally,foreachinstance,giventheoracle -| | | | | | | | CitationQuality. | | Citationqualityiscriticalfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | --- | -groundedg(i)alongwithaconsistentoracleanswer -| (i) | | | | | (i) | | evaluating | LLM | attribution, | | assessed | along | two | -| --- | --------------------------- | --- | --- | --- | --- | --------- | ----------- | --- | ------------ | ------- | ----------- | ----- | --- | -| y | aswellasanattributedanswery | | | | | generated | | | | | | | | -| w | | | | | l | | | | | | | | | -| | | | | | | | dimensions: | (1) | Citation | Recall, | determining | | if | -byaweakerLLMviain-contextlearning,wecan -| | | | | | | | the output | is entirely | | supported | by | the cited | doc- | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | --- | --------- | --- | --------- | ---- | -simplyconstructapreferencedataset: -| | | | | | | | uments, | and (2) | Citation | Precision, | | assessing | if | -| --- | --- | --- | --------------------- | --- | ------------- | --- | ---------------------------------------------- | ------- | -------- | ---------- | --- | --------- | --- | -| | | D | = (cid:8) x(i),τ(i),τ | | (i)(cid:9)N , | | | | | | | | | -| | | | | w | | (4) | eachcitationsupportsitscorrespondingstatement. | | | | | | | -l i=1 -| | (i) | | (i) | | | | EvaluationisconductedbyTRUE(Honovichetal., | | | | | | | -| ------ | --- | -------- | --- | ------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -| whereτ | | = g(i)◦y | | denotestheconcatenationof | | | | | | | | | | -| | w | | w | | | | 2022),aT5-11Bmodelfine-tunedonacollection | | | | | | | -theoraclegroundingwiththeconsistent,attributed -ofNLIdatasetstoautomaticallyexaminetheentail- -| | | (i) | g(i)◦y | (i) | | | | | | | | | | -| -------- | --- | ------------ | ------ | ----------------------- | ------- | ------- | ------------------------------------------ | --- | ------- | ---------- | ------- | --- | ------ | -| answer,τ | | = | | denotestheconcatenation | | | | | | | | | | -| | | l | | l | | | mentofciteddocumentsandthemodelgeneration. | | | | | | | -| with | the | inconsistent | | attributed | answer. | Here, ◦ | | | | | | | | -| | | | | | | | Additionally, | to | capture | a holistic | measure | | of ci- | -signifiestheoperationofstringconcatenation. -tationquality,wealsoreporttheCitationF1,the -| | Finally,wecanoptimizethepolicymodelπ | | | | | on | | | | | | | | -| --- | ------------------------------------ | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | θ | harmonicmeanofcitationprecisionandrecall: | | | | | | | -thedatasetD -byminimizingthefollowingloss: -citationprecision·citationrecall -| | L | (π ;π | ;D) | | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | ----- | -| | DPO | θ | ref | | | | F 1 = 2· | | | | | | , (6) | -citationprecision+citationrecall -| | | | (cid:20) | (cid:18) | π (τ | |x) | | | | | | | | -| --- | --- | --- | -------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -| | =−E | | | | θ | w | | | | | | | | -| | | | | logσ | βlog | | | | | | | | | -(x,τw,τ l )∼D π (τ |x) (5) Correctness. FortheASQAdataset,correctness -ref w -π (τ |x) (cid:19)(cid:21) is quantified using exact match recall (EM Rec.) -θ l -| | −βlog | | | , | | | | | | | | | | -| --- | ----- | --- | ------ | --- | --- | --- | ----------- | ------- | --- | --------- | ------- | --- | ----- | -| | | π | (τ |x) | | | | by checking | whether | | the short | answers | are | exact | -| | | | ref l | | | | | | | | | | | - -| | | | ASQA | | | ELI5 | | QAMPARI | | | -| --------- | --- | ----------- | ---- | --------- | ----------- | ---- | -------- | ------------ | ---------- | --- | -| ModelType | | ModelSize | | | | | | | | | -| | | Correctness | | Citation | Correctness | | Citation | Correctness | Citation | | -| | | EMRec. | Rec. | Prec. F1. | Claim | Rec. | Prec. F1 | Rec.-5 Prec. | Rec. Prec. | F1 | -Prompting-based -ChatGPT - 40.37 72.81 69.69 71.22 12.47 49.44 47.05 48.22 20.28 19.84 19.06 22.03 20.44 -7B 24.32 17.24 17.87 17.55 4.53 3.92 5.38 4.54 12.56 11.32 6.03 6.35 6.19 -LLaMA-2 13B 27.99 16.45 19.04 17.65 7.77 8.49 8.43 8.46 18.00 12.39 5.45 5.74 5.59 -70B 31.53 44.18 44.79 44.48 10.43 23.75 22.43 23.07 18.50 14.79 10.10 10.50 10.30 -7B 29.93 55.99 51.66 53.74 12.47 19.90 15.48 17.41 17.96 19.74 9.58 9.68 9.63 -LLaMA-2-Chat 13B 34.39 37.15 38.17 37.65 13.83 16.50 16.09 16.29 21.34 18.86 8.94 9.06 9.00 -70B 41.24 60.19 61.16 60.67 13.30 36.63 36.63 36.63 22.62 18.04 13.49 13.98 13.73 -Vicuna-v1.5 7B 38.34 48.37 44.63 46.42 12.30 29.81 22.45 25.61 14.22 14.74 11.26 11.64 11.45 -13B 35.20 51.92 53.40 52.65 14.33 31.15 28.99 30.03 22.06 19.60 13.04 13.74 13.38 -7B 29.46 23.12 25.45 24.23 8.47 16.04 16.32 16.18 16.96 15.98 7.50 7.76 7.63 -Mistral -8×7B 36.30 32.72 34.49 33.58 10.43 26.11 25.09 25.59 18.18 15.63 9.72 10.20 9.95 -7B 38.57 64.90 59.67 62.18 11.07 49.25 42.69 45.74 17.52 21.29 17.56 18.53 18.03 -Mistral-Instruct 8×7B 44.11 61.80 63.27 62.53 13.93 49.28 48.34 48.81 20.12 19.64 19.27 20.38 19.81 -Post-hocRetrieval -ChatGPT - 37.68 27.11 27.05 27.08 18.77 14.55 14.55 14.55 25.14 22.85 12.29 12.29 12.29 -LLaMA-2-Chat 70B 29.68 24.51 24.51 24.51 16.03 12.93 12.93 12.93 17.90 14.45 9.05 9.05 9.05 -Mistral-Instruct 8×7B 33.90 24.57 24.48 24.52 17.37 15.68 15.68 15.68 24.16 18.28 9.78 9.78 9.78 -Training-based -7B 29.96 67.82 66.97 67.39 6.90 22.34 32.40 26.45 2.34 1.98 10.53 18.80 13.50 -Self-RAG(LLaMA-2) -13B 31.66 71.26 70.35 70.80 6.07 30.46 40.20 34.66 1.90 1.33 12.79 20.90 15.86 -7B 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -VANILLA-SFT(LLaMA-2) 13B 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -FRONT(LLaMA-2) 7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -Table1: MainresultsontheALCEbenchmark. Boldnumbersindicatethebestperformance,while_indicatesthe -second-bestperformance. -substringsofthegeneration. RegardingtheELI5 Mistral-7B(Jiangetal.,2023)toMistral-8x7B- -dataset,correctnessismeasuredthroughclaimre- MoE (Jiang et al., 2024). Regarding SFT LLMs, -call (Claim), evaluating whether the model’s re- weselecttheSFTcounterpartsoftheopen-source -sponse entails the ground truth sub-claims. For foundationalLLMsweused. Detailedprompting -theQAMPARIdataset,correctnessisassessedus- settingscanbefoundinAppendixC. -| ingexactmatchprecision(Prec.) | | | andtop-5exact | | | | | | | | -| ----------------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | -match recall (Rec.-5) — considered 100% if the 4.3.2 Post-hocRetrievalMethods. -predictionincludesatleastfivecorrectanswers. -| | | | | | Following | Gao | et al. | (2023b), | we first | instruct | -| --- | --- | --- | --- | --- | --------- | --- | ------ | -------- | -------- | -------- | -LLMstoanswerthegivenqueryinaclosed-book -4.3 Baselines -| | | | | | setting, | and then | integrate | citations | in a post-hoc | | -| --- | --- | --- | --- | --- | -------- | -------- | --------- | --------- | ------------- | --- | -Wecompareourmethodwiththreetypesofbase- -| | | | | | manner. | Foreachgeneratedstatement,weemploy | | | | | -| ------ | ---------------- | -------- | ---------- | --- | ------- | ---------------------------------- | --- | --- | --- | --- | -| lines: | prompting-based, | post-hoc | retrieval, | and | | | | | | | -GTR(Nietal.,2022)toidentifyandcitethemost -training-based. -relevantdocumentfromthetop100retrieveddoc- -4.3.1 Prompting-basedMethods. uments. Weutilizethesamemodelsmentionedin -WedirectlypromptLLMsusingfew-shotdemon- prompting-basedsettingsforthisbaseline. -strations,eachconsistingofaquery,thetop5rel- -| | | | | | 4.3.3 | Training-basedMethods. | | | | | -| --- | --- | --- | --- | --- | ----- | ---------------------- | --- | --- | --- | --- | -evantretrieveddocuments,andananswerwithin- -| linecitations. | Ourexperimentsencompassarange | | | | | | | | | | -| -------------- | ----------------------------- | --- | --- | --- | -------- | --------------------------------- | --- | --- | --- | --- | -| | | | | | Self-RAG | (Asaietal.,2023)Self-RAGtrainsthe | | | | | -ofLLMs,fromfoundationalmodelstosupervised -| | | | | | LLM | to learn | to adaptively | retrieve | passages | on- | -| --------------------- | --- | -------------------- | --- | --- | --- | -------- | ------------- | -------- | -------- | --- | -| fine-tuning(SFT)LLMs. | | ForfoundationalLLMs, | | | | | | | | | -demandandenableittoreflectonitsgenerationto -| we select | GPT-3.5-Turbo5 | as the | representative | | | | | | | | -| --------- | -------------- | ------ | -------------- | --- | --- | --- | --- | --- | --- | --- | -furtherimprovegenerationqualityandattributions. -closed-sourcemodel,recognizedforitsnotableper- -| formance. | Among | the open-source | foundational | | | | | | | | -| --------- | ----- | --------------- | ------------ | --- | ----------- | --- | -------------------------- | --- | --- | --- | -| | | | | | VANILLA-SFT | | Wedirectlyemploysupervised | | | | -LLMs,wefocusontheLLaMA-2seriesincluding -fine-tuningtotraintheLLMonourgeneratedtrain- -LLaMA2-7B,LLaMA2-13B,andLLaMA2-70B, -| | | | | | ing data. | Given | a query | and corresponding | | doc- | -| ------------------------------------------------ | -------------- | ------------- | ----- | ---- | --------------------- | ------- | ------- | ----------------- | -------- | -------- | -| as well | as the Mistral | series, which | spans | from | | | | | | | -| | | | | | uments, | the LLM | is | required to | directly | generate | -| 5Specifically,weutilizegpt-3.5-turbo-1106version | | | | | answerswithcitations. | | | | | | - -4.4 ImplementDetails -| Weimplement | | FRONTwithdifferentsizesoffoun- | | | | | 60 | | | | Pre-filtered | | | -| ----------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --- | -Post-filtered -| dational | models | (LLaMA-2-7B | | and | LLaMA-2- | | | | | | | | | -| -------- | ------ | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -)%(1FnoitatiC 50 -| 13B)toevaluateitseffectiveness. | | | | Duringtheeval- | | | | | | | | | | -| ------------------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -40 -uation,FRONTutilizethesameretrievalsettingsas -| those outlined | | by Gao | et al. | (2023b). | Additional | | 30 | | | | | | | -| -------------- | ----------- | ------ | ---------- | -------- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| details | of training | and | evaluation | settings | | can be | 20 | | | | | | | -foundinAppendixD. -| | | | | | | | | ASQA | | ELI5 | | QAMPARI | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ------- | --- | -5 ResultsandAnalysis -| | | | | | | | Figure4: | AblationStudyonDataFiltering. | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----------------------------- | --- | --- | --- | --- | --- | -5.1 OverallResults -Simplysupervisedfine-tuningcanboostcitation -| | | | | | | | FRONT | demonstrates | | remarkable | | generaliza- | | -| ----------- | ------------------------------ | --- | -------------- | --- | ---------- | ----- | ---------- | ------------ | --- | ---------- | --- | ----------- | ------- | -| quality. | AsshowninTable1,teachingLLMsto | | | | | | | | | | | | | -| | | | | | | | tion. | Compared | to | the varied | | queries | and an- | -| generate | responses | | with citations | via | supervised | | | | | | | | | -| | | | | | | | swer types | present | in | the ALCE | | benchmark, | our | -| fine-tuning | significantly | | enhances | | citation | qual- | | | | | | | | -trainingdata,derivedexclusivelyfromtheAQuA- -ity,demonstratingsubstantialimprovementsover -MuSedataset(Kulkarnietal.,2020),exhibitsout- -bothprompt-basedandpost-hocretrievalbaselines -| | | | | | | | of-domain | characteristics. | | | Nonetheless, | | FRONT | -| ------ | ------------- | --- | ------------- | ---- | -------- | --- | ------------ | ---------------- | --- | -------- | ------------ | --- | --------- | -| across | all datasets. | | Specifically, | with | LLaMA-2- | | | | | | | | | -| | | | | | | | demonstrates | superior | | citation | quality, | | affirming | -7B, VANILLA-SFTledtosubstantialgainsincita- -itsexceptionalabilitytogeneralizeacrossdiverse -| tion F1 | scores | over | prompting: | ASQA | (17.55 | → | | | | | | | | -| ------- | ------ | ---- | ---------- | ---- | ------ | --- | -------------------------------- | --- | --- | --- | --- | ------------- | --- | -| | | | | | | | querytypesandretrievaldocuments. | | | | | Additionally, | | -65.61),ELI5(4.54→41.15),andQAMPARI(6.19 -| | | | | | | | while not | specifically | | optimized | | for correctness, | | -| -------- | ----------------------------------- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------------ | ---------------- | --- | -| →21.35). | Thesegainshighlighttheeffectiveness | | | | | | | | | | | | | -| | | | | | | | FRONT | also showcases | | modest | improvements | | in | -ofourtrainingdatagenerationpipeline. -thismetricoverVANILLA-SFTontheASQAand -| | | | | | | | QAMPARIdatasets. | | | However,FRONTencounters | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ----------------------- | --- | --- | --- | -FRONT achievessignificantperformancegains -lowerRec.-5ontheQAMPARIdataset,likelydue -| andsurpassesChatGPT. | | | | While VANILLA-SFT | | | | | | | | | | -| -------------------- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tothenatureofitsanswers,whichconsistofcon- -| demonstrates | | strong | performance, | | it still | shows | | | | | | | | -| ------------ | --- | ------ | ------------ | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | -catenatedentities,divergingsignificantlyfromour -| notable | discrepancies | | compared | to | leading | open- | | | | | | | | -| ------- | ------------- | --- | -------- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | -trainingdata. -sourceLLMs,suchasMixtral-8×7B-Instruct(e.g., -| 41.15 vs. | 45.74) | and | ChatGPT | (e.g., | 41.15 | vs. | | | | | | | | -| --------- | ------ | ---- | -------- | ----------- | ----- | -------- | ----------------- | --- | --- | --- | --- | --- | --- | -| 48.22) | on the | ELI5 | dataset. | FRONT | not | only | 5.2 AblationStudy | | | | | | | -| bridges | these | gaps | but also | establishes | | signifi- | | | | | | | | -Weconductablationstudiestoverifytheeffective- -| cant leads | across | all | datasets. | Specifically, | | us- | | | | | | | | -| --------------- | ------ | --- | --------- | --------------- | --- | ---- | ----------------------------------- | --- | --- | --- | --- | --- | ------ | -| | | | | | | | nessofdifferentcomponentsproposedin | | | | | | FRONT. | -| ing LLaMA-2-7B, | | | FRONT | comprehensively | | out- | | | | | | | | -performsChatGPT,achievingincreasesof3.32%, -| | | | | | | | EffectsofDataGenerationPipeline. | | | | | | Asillus- | -| ------- | --- | ------ | ----------- | ------- | --- | ------ | -------------------------------- | -------- | ------ | --- | -------- | ------ | -------- | -| 18.04%, | and | 21.28% | in citation | quality | | on the | | | | | | | | -| | | | | | | | trated | in §5.1, | simply | SFT | achieves | strong | per- | -ASQA,ELI5,andQAMPARIdatasetsrespectively. -| | | | | | | | formance, | underscoring | | the | high | quality | of our | -| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | --- | --- | ---- | ------- | ------ | -Thisperformanceunderscorestheeffectivenessof -| | | | | | | | training | data. | Furthermore, | | data | filtering, | a cru- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------ | --- | ---- | ---------- | ------ | -FRONTinenhancingattributioncapabilities. -| | | | | | | | cial component | | of our | data | generation | | pipeline, | -| -------------------------------------- | --- | --- | --- | --- | --- | --- | -------------- | ------- | ------- | -------- | ---------- | ------- | --------- | -| | | | | | | | plays a | pivotal | role in | ensuring | the | quality | of the | -| FRONTexhibitsscalabilitywithmodelsize. | | | | | | As | | | | | | | | -illustratedatthebottomofTable1,theperformance generated data by filtering out queries that yield -of FRONT in terms of citation quality shows no- non-informativeanswersorfailtomeetattribution -| | | | | | | | criteria. | To validate | | the effectiveness | | of | our data | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ----------------- | --- | --- | -------- | -tableimprovementswhenscalingfrom7Bto13B. -filteringstrategies,weconductedexperimentscom- -Specifically,weobserveimprovementsof3.23%in -ASQA,4.97%inELI5,and1.33%inQAMPARI. paringmodelsfine-tunedonbothpre-filteredand -This upward trend underscores the scalability of post-filtereddata. Theresults,depictedinFigure -4,confirmthatmodelstrainedonfiltereddataex- -FRONT withincreasingmodelsize,demonstrating -thepotentialofFRONTinleveragingtheincreased hibitanotableimprovementincitationqualityover -capabilitiesoflargerLLMsforfurtherperformance thosetrainedonunfiltereddata,achievingsuperior -attributionperformancewithreduceddatavolume. -gains. - -| | | | | ASQA | | | | ELI5 | | | QAMPARI | | | -| --- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | ------- | --- | --- | -Model -| | | | Correctness | | Citation | | Correctness | | Citation | Correctness | | Citation | | -| --- | --- | --- | ----------- | --- | -------- | --- | ----------- | --- | -------- | ----------- | --- | -------- | --- | -EMRec. Rec. Prec. F1. Claim Rec. Prec. F1 Rec.-5 Prec. Rec. Prec. F1 -FRONT-7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -SELF-GUIDE(w/oConsistency) 38.99 70.69 64.48 67.44 10.04 47.63 44.80 46.17 12.18 20.03 22.50 22.58 22.54 -VANILLA-SFT(w/oGround) 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -FRONT-13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -SELF-GUIDE(w/oConsistency) 40.99 73.08 68.13 70.52 10.06 50.68 49.78 50.23 13.94 22.38 23.73 23.99 23.85 -VANILLA-SFT(w/oGround) 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -Table2: AblationstudyontheimpactofdifferenttrainingstageswithintheALCEbenchmark. -44.42 -Citation-F1 -| | | | | 46.17 | | | | | 33 | | FRONT-13B | | | -| -------------- | --- | ----- | ----- | ----- | ----------- | --- | --- | ------------------- | --------------- | --- | --------- | ------------ | --- | -| | | | | | | | | | | | F | -7 B | | -| | | 43.71 | | | SELF-GUIDE | | | | | | R O | N T | | -| Citation-Prec. | | | | | | | | | LLaMA2-70B-Chat | | SEL F- G | U I D E -13B | | -| | | | 44.80 | | PROMPTGUIDE | | | ssenlufhtiaF 22..55 | | | | | | -SELF-GUIDE-7B -45.16 -| Citation-Rec. | | | | | | | | | | | Mixtral-8x7B-Inst | | | -| ------------- | --- | ----- | --- | ----- | ----- | --- | --- | --- | --- | ---------- | ----------------- | --- | --- | -| | | | | | 47.63 | | | | 22 | Vicuna-13B | | | | -| | | 44.00 | | 46.00 | 48.00 | | | | | | | | | -LLaMA2-7B-Chat -| Figure5: | Ablationstudyofdifferentgroundingguid- | | | | | | | 11..55 | | | | | | -| -------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -ancestrategiesontheELI5dataset. -| | | | | | | | | | 2200 | 3300 | 4400 5500 | 6600 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --------- | ---- | --- | -CitationF1 -EffectsofGroundingGuidedGeneration(G3). -G3 -empowers LLMs to first select relevant fine- Figure6: TherelationshipbetweencitationF1andhal- -grainedquotes,whichsubsequentlyguidethegen- lucination: Models positioned closer to the top-right -eration process. These quotes can provide fine- cornerexhibithighercitationqualityandalowerdegree -ofhallucination. -grainedsupervisionsignalsforattributedtextgen- -G3, -| eration. | To evaluate | | the effectiveness | | of | we | | | | | | | | -| --------------------------------------- | ----------- | --- | ----------------- | --------- | ---------- | ----- | ------- | -------- | ----------- | --- | --------- | --- | ------ | -| conduct | an ablation | | study | comparing | it against | | | | | | | | | -| | | | | | | | eration | process. | Experiments | | conducted | | on the | -| twovariantswithdistincttrainingrecipes. | | | | | | Given | | | | | | | | -ELI5datasetusingtheLLaMA-2-7Bmodelshow -that FRONTconsistsoftwostages,werefertothe that SELF-GUIDE outperforms PROMPT-GUIDE. -modeltrainedonlyduringthefirststage(without -ResultsdepictedinFigure5indicatethattraining -| consistency-awarealignment)asSELF-GUIDE. | | | | | | We | | | | | | | | -| ---------------------------------------- | ---------- | --- | ------- | --- | ----------- | --- | ---------- | --- | ------------- | --------- | -------- | -------------- | ------ | -| | | | | | | | models | to | self-generate | | grounded | quotes | before | -| firstcompare | SELF-GUIDE | | against | | VANILLA-SFT | | | | | | | | | -| | | | | | | | generating | | attributed | responses | is | more effective | | -(w/oGround),whichistrainedtodirectlygenerate thansimplyincorporatingthesegroundedquotes -responseswithcitations,bypassingthegrounding -intotheprompt. -| step. The | ablation | study, | detailed | | in Table | 2, re- | | | | | | | | -| --------- | -------- | ------ | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | -veals that models incorporating grounding guid- EffectsofConsistency-AwareAlignment(CCA). -ancesignificantlyoutperformtheir VANILLA-SFT The primary goal of CCA is to enhance the con- -counterpartsthatlacksuchgroundingmechanisms. sistency between grounded quotes and attributed -This highlights the crucial role of grounding in answers, thereby alleviating hallucinations and -| | | | | | | | achievingmorepreciseattribution. | | | | | Toevaluatethis, | | -| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | -bolsteringattribution. -Moreover, we explore an alternative variant we compare models that underwent only the G3 -of grounding guidance. Considering that SELF- stage(SELF-GUIDE)withthosefurtherenhanced -GUIDE leverages the model itself to both select through the CCA stage (FRONT). As illustrated -grounded quotes and generate attributed answers inTable2,FRONTsignificantlyimprovescitation -in an end-to-end paradigm, a natural variant in- qualityoverSELF-GUIDE,demonstratingtheeffec- -volves breaking down this task into two distinct tivenessoftheCCAstageinenhancingattribution. -stages. Inthisvariant,ChatGPTistaskedwithex- Furthermore,toassessCCA’simpactonreduc- -tractinggroundedquotes. Subsequently,aseparate inghallucinations,weutilizeQAFactEval(Fabbri -modelistrainedtoutilizethesegroundedquotes, etal.,2022),awidelyusedmetricforfactualcon- -along with the query and retrieval documents, to sistency, which scores the consistency of model -directly output the response and citations. This responses to given documents on a scale from 0 -variant,referredtoasPROMPT-GUIDED,integrates to5,withhigherscoresindicatinggreaterfaithful- -groundedquotesintotheprompttoguidethegen- ness. Specifically,weanalyzetheperformanceof - -leading open-source models and two variants of Authenticity Helpfulness -| FRONTand | SELF-GUIDE | | ontheELI5dataset. | | As | | | | | | | -| -------- | ---------- | --- | ----------------- | --- | --- | ------- | --- | --- | --- | ---- | ---- | -| | | | | | | ChatGPT | | | | 0.93 | 4.08 | -showninFigure6, FRONTproducesmorefaithful FRONT-7BonASQA 0.94 3.86 -outputsthanSELF-GUIDE,significantlyreducing FRONT-7BonELI5 0.92 3.96 -| hallucinations. | | | | | | FRONT-7BonQAMPARI | | | | 0.86 | 3.62 | -| --------------- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | ---- | ---- | -Table3: Humanevaluationonthequalityofgrounded -| EffectsofTrainingDataScale. | | | | Weanalyzethe | | | | | | | | -| --------------------------- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | -quotes. -| impact | of the | data scale | on model | performance | | | | | | | | -| ------ | ------------ | ---------- | -------- | ----------- | ------- | --- | --- | --- | --- | --- | --- | -| across | two training | stages. | In | particular, | we ran- | | | | | | | -tors. Theresultsofthehumanevaluationindicate -domlysampled2k,4k,6k,and8kinstancesfrom -thatbothquotesextractedbyChatGPTandthose -| our full | training | data | across two | distinct | training | | | | | | | -| -------- | -------- | ---- | ---------- | -------- | -------- | --- | --- | --- | --- | --- | --- | -stages. These subsets were then utilized to fine- generated by FRONT are of high quality, further -tunevarious7Bmodelvariants,enablingacompar- substantiatingtheeffectivenessofourmethod. -ativeanalysisofperformancebasedondatascale. -| Results | are shown | in | Figure | 7, which | indicates | | | | | | | -| ------- | --------- | --- | ------ | -------- | --------- | --- | --- | --- | --- | --- | --- | -7 Conclusion -thatincreasingdatasizeshowssignificantenhance- -mentsincitationquality,indicatingapositivecor- Inthiswork,wepresent FRONT,atwo-stagetrain- -relationbetweendatasizeandmodelperformance. ingframeworkdesignedtoequipLLMswithfine- -AsFRONTimplementsanautomatedprocedureca- -| | | | | | | grained | attribution | capabilities. | | FRONT | enables | -| --- | --- | --- | --- | --- | --- | ------- | ----------- | ------------- | --- | ----- | ------- | -pableofgeneratinghigh-qualityattributeddataand LLMstoinitiallyselectsupportingquotes,which -constructingcontrastivesupervisionfromweakand thenguidethegenerationprocess. Byfurtheren- -strongLLMs,itholdsthepotentialforcontinuous -| | | | | | | hancing | the consistency | | between | | the grounding | -| --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------- | --- | ------------- | -performanceimprovements. andgenerationprocessviapreferenceoptimization, -thesesupportingquotescanserveasfine-grained -6 HumanEvaluation -| | | | | | | citations. | Through | comprehensive | | | experiments, | -| --- | --- | --- | --- | --- | --- | ---------- | ------- | ------------- | --- | --- | ------------ | -Given the significant impact of the quality of FRONThasdemonstrateditsabilitytogeneratesu- -| | | | | | | perior grounded | | responses | | and highly | supportive | -| -------- | ------ | --- | ------------ | ------------ | --- | --------------- | ------------------------ | --------- | --- | ---------- | ---------- | -| grounded | quotes | on | fine-grained | verification | for | | | | | | | -| | | | | | | citations. | Furtheranalysisshowsthat | | | | FRONT sig- | -users,weconductedahumanevaluationtoassess -thequalityofgroundedquotesatdifferentstagesof nificantlyreduceshallucinationsandbenefitsuser -verification. -| ourframework: | | (1)QuotesextractedbyChatGPT | | | | | | | | | | -| ---------------------------------- | --------- | --------------------------- | ----------- | --- | ------------ | ------------ | --- | --- | --- | --- | --- | -| from50sampleddatapointsduringtheG3 | | | | | stage. | | | | | | | -| (2) Quotes | generated | | by FRONT-7B | | across three | 8 Limitation | | | | | | -datasets,with50datapointssampledfromeach. -Weengagedfourannotators,eachwithrelevant Our study presents several limitations worth not- -expertiseandholdingatleastabachelor’sdegree. ing. Firstly, the validation of our framework is -Thequalityofquoteswasevaluatedontwodimen- predominantly conducted on models of sizes 7B -sions: authenticityandhelpfulness. Authenticity and13B,leavingtheexplorationoflargermodels, -(abinaryscaleof0/1)referstowhetherthequotes suchasLLaMA-270Bduetocomputationalcon- -genuinelyoriginatefromthecorrespondingdocu- straints. Secondly,ourframeworkreliesonaprior -ments(quotesthatarehallucinatedormismatched retrievalprocess,whereinrelevantdocumentsare -withthecorrespondingdocumentIDareconsidered retrieved at one time. The incorporation of adap- -inauthentic). Helpfulness (5-point Likert scale) tiveretrieval,enablingmoredynamicinteractions -referstothedegreetowhichthequotesarebenefi- withLLMs,couldpotentiallyenhanceperformance. -cialinaddressingthequery. TheresultsinTable3 Weleaveitforfutureresearch. Lastly,evaluating -representtheaveragescoresforallquoteswithin the correctness of long-form question answering -eachmodelresponse,withtwoannotatorsevaluat- presents inherent challenges, leading our frame- -ingeachresponsetoensurereliability. work to primarily enhance citation quality, with -Furthermore, to evaluate the consistency of modest advancements in correctness. Therefore, -quotequalityannotations,wecomputedtheinter- we advocate for the development of more robust -annotator agreement using Fleiss’ Kappa coeffi- metricscapableofaccuratelyassessingthecorrect- -cient. The obtained Kappa coefficient of 0.82 in- nessoflong-formQAresponses, pavingtheway -| dicates | a high | level of | agreement | among | annota- | forfuturework. | | | | | | -| ------- | ------ | -------- | --------- | ----- | ------- | -------------- | --- | --- | --- | --- | --- | - -Acknowledgements -qa-basedfactualconsistencyevaluationforsumma- -| | | | | | | | rization. | InProceedingsofthe2022Conferenceof | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | ---------------------------------- | --- | --- | --- | -We appreciate Yuchun Fan for providing valu- theNorthAmericanChapteroftheAssociationfor -| | | | | | | | ComputationalLinguistics: | | | HumanLanguageTech- | | -| ----------------- | --- | --------- | --- | --- | ------- | -------- | ------------------------- | --- | --- | ------------------ | --- | -| able suggestions. | | Xiaocheng | | | Feng is | the cor- | | | | | | -nologies,NAACL2022,Seattle,WA,UnitedStates, -| responding | author | of | this | work. | We | thank the | | | | | | -| ---------- | ------ | --- | ---- | ----- | --- | --------- | --- | --- | --- | --- | --- | -July10-15,2022,pages2587–2601.Associationfor -| anonymous | reviewers | | for | their | insightful | com- | | | | | | -| --------- | --------- | --- | --- | ----- | ---------- | ---- | --- | --- | --- | --- | --- | -ComputationalLinguistics. -| ments. | This | work | was | supported | by | the Na- | | | | | | -| ------ | ---- | ---- | --- | --------- | --- | ------- | --- | --- | --- | --- | --- | -AngelaFan,YacineJernite,EthanPerez,DavidGrang- -| tional Key | R&D | Program | | of China | via | grant No. | | | | | | -| -------------- | --- | ------- | -------- | -------- | --------- | --------- | ---------- | -------- | ----------- | -------------- | ----- | -| | | | | | | | ier, Jason | Weston, | and Michael | Auli. 2019. | ELI5: | -| 2021ZD0112905, | | the | National | | Natural | Science | | | | | | -| | | | | | | | long form | question | answering. | In Proceedings | of | -| Foundation | of | China | (NSFC) | (grant | 62276078, | | | | | | | -the57thConferenceoftheAssociationforCompu- -U22B2059), the Key R&D Program of Hei- tationalLinguistics,ACL2019,Florence,Italy,July -longjiang via grant 2022ZX01A32, the Interna- 28-August2,2019,Volume1: LongPapers,pages -3558–3567.AssociationforComputationalLinguis- -tionalCooperationProjectofPCL,PCL2022D01 -tics. -andtheFundamentalResearchFundsfortheCen- -tralUniversities(GrantNo.HIT.OCEF.2023018). ZhangyinFeng,WeitaoMa,WeijiangYu,LeiHuang, -HaotianWang,QianglongChen,WeihuaPeng,Xi- -| | | | | | | | aochengFeng,BingQin,andTingLiu.2023. | | | | Trends | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | ------ | -inintegrationofknowledgeandlargelanguagemod- -References -els:Asurveyandtaxonomyofmethods,benchmarks, -CoRR,abs/2311.05876. -| Samuel | Joseph | Amouyal, | Ohad | Rubin, | | Ori Yoran, | andapplications. | | | | | -| -------- | --------- | -------- | --------- | ------- | -------------- | ---------- | ---------------- | ------ | ----------------- | -------- | --------- | -| Tomer | Wolfson, | Jonathan | | Herzig, | and | Jonathan | | | | | | -| | | | | | | | Luyu Gao, | Zhuyun | Dai, Panupong | Pasupat, | Anthony | -| Berant. | 2022. | QAMPARI: | | : | An open-domain | | | | | | | -| | | | | | | | Chen, | Arun | Tejasvi Chaganty, | Yicheng | Fan, Vin- | -| question | answering | | benchmark | | for questions | with | | | | | | -centY.Zhao,NiLao,HongraeLee,Da-ChengJuan, -| many | answers | from | multiple | paragraphs. | | CoRR, | | | | | | -| ---- | ------- | ---- | -------- | ----------- | --- | ----- | ---------- | ---- | ------ | ----------------- | --- | -| | | | | | | | and Kelvin | Guu. | 2023a. | RARR: researching | and | -abs/2205.12665. -revisingwhatlanguagemodelssay,usinglanguage -AkariAsai,ZeqiuWu,YizhongWang,AvirupSil,and models. InProceedingsofthe61stAnnualMeeting -oftheAssociationforComputationalLinguistics(Vol- -| Hannaneh | Hajishirzi. | | 2023. | Self-rag: | Learning | to | | | | | | -| -------- | ----------- | --- | ----- | --------- | -------- | --- | ----- | ----------------------------------- | --- | --- | --- | -| | | | | | | | ume1: | LongPapers),ACL2023,Toronto,Canada, | | | | -retrieve,generate,andcritiquethroughself-reflection. -July9-14,2023,pages16477–16508.Associationfor -CoRR,abs/2310.11511. -ComputationalLinguistics. -| Yuntao Bai, | Andy | Jones, | Kamal | Ndousse, | | Amanda | | | | | | -| ----------- | ---- | ------ | ----- | -------- | --- | ------ | --- | --- | --- | --- | --- | -TianyuGao,HowardYen,JiatongYu,andDanqiChen. -| Askell, | AnnaChen, | | NovaDasSarma, | | DawnDrain, | | | | | | | -| --------- | --------- | ---- | ------------- | --- | ---------- | --------- | ------------------ | ------------------------------------- | -------------------- | --- | --- | -| | | | | | | | 2023b. | Enablinglargelanguagemodelstogenerate | | | | -| Stanislav | Fort, | Deep | Ganguli, | | Tom | Henighan, | | | | | | -| | | | | | | | textwithcitations. | | CoRR,abs/2305.14627. | | | -NicholasJoseph,SauravKadavath,JacksonKernion, -TomConerly,SheerElShowk,NelsonElhage,Zac -| | | | | | | | Yunfan Gao, | Yun | Xiong, Xinyu | Gao, Kangxiang | Jia, | -| --------------- | --- | ----- | ---------- | --- | ------- | ----- | ----------- | --- | ------------ | -------------- | ---- | -| Hatfield-Dodds, | | Danny | Hernandez, | | Tristan | Hume, | | | | | | -JinliuPan,YuxiBi,YiDai,JiaweiSun,QianyuGuo, -ScottJohnston,ShaunaKravec,LianeLovitt,Neel Meng Wang, and Haofen Wang. 2023c. Retrieval- -| Nanda, | Catherine | Olsson, | | Dario | Amodei, | Tom B. | | | | | | -| -------- | --------- | ------- | --------------- | --------- | ------- | ---------- | ------------------------------------------ | -------------------- | --- | --- | --- | -| | | | | | | | augmentedgenerationforlargelanguagemodels: | | | | A | -| Brown, | Jack | Clark, | Sam McCandlish, | | Chris | Olah, | | | | | | -| | | | | | | | survey. | CoRR,abs/2312.10997. | | | | -| Benjamin | Mann, | and | Jared | Kaplan. | 2022. | Train- | | | | | | -| ing a | helpful | and | harmless | assistant | | with rein- | | | | | | -SylvainGugger,LysandreDebut,ThomasWolf,Philipp -forcement learning from human feedback. CoRR, Schmid,ZacharyMueller,SourabMangrulkar,Marc -abs/2204.05862. Sun,andBenjaminBossan.2022. Accelerate: Train- -ingandinferenceatscalemadesimple,efficientand -BerndBohnet,VinhQ.Tran,PatVerga,RoeeAharoni, -| | | | | | | | adaptable. | | https://github.com/huggingface/ | | | -| ------ | ------ | ----- | ------- | ------- | ----- | ------ | ---------- | --- | ------------------------------- | --- | --- | -| Daniel | Andor, | Livio | Baldini | Soares, | Jacob | Eisen- | | | | | | -accelerate. -stein,KuzmanGanchev,JonathanHerzig,KaiHui, -TomKwiatkowski,JiMa,JianmoNi,TalSchuster, OrHonovich, RoeeAharoni, JonathanHerzig, Hagai -WilliamW.Cohen,MichaelCollins,DipanjanDas, Taitelbaum,DoronKukliansy,VeredCohen,Thomas -Donald Metzler, Slav Petrov, and Kellie Webster. Scialom, Idan Szpektor, Avinatan Hassidim, and -| 2022. | Attributed | question | | answering: | | Evaluation | | | | | | -| ----- | ---------- | -------- | --- | ---------- | --- | ---------- | ----- | ------- | ----------- | ------------- | ------- | -| | | | | | | | Yossi | Matias. | 2022. TRUE: | re-evaluating | factual | -andmodelingforattributedlargelanguagemodels. -| | | | | | | | consistencyevaluation. | | InProceedingsofthe2022 | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | ---------------------- | --- | --- | -CoRR,abs/2212.08037. Conference of the North American Chapter of the -| | | | | | | | AssociationforComputationalLinguistics: | | | | Human | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | ----- | -CanyuChenandKaiShu.2023. Combatingmisinfor- LanguageTechnologies,NAACL2022,Seattle,WA, -mation in the age of llms: Opportunities and chal- UnitedStates, July10-15, 2022, pages3905–3920. -| lenges. | CoRR,abs/2311.05656. | | | | | | | | | | | -| ------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AssociationforComputationalLinguistics. -Alexander R. Fabbri, Chien-Sheng Wu, Wenhao Liu, LeiHuang,WeijiangYu,WeitaoMa,WeihongZhong, -and Caiming Xiong. 2022. Qafacteval: Improved Zhangyin Feng, Haotian Wang, Qianglong Chen, - -WeihuaPeng,XiaochengFeng,BingQin,andTing servingwithpagedattention. InProceedingsofthe -Liu. 2023. A survey on hallucination in large lan- ACMSIGOPS29thSymposiumonOperatingSystems -| guagemodels: | Principles,taxonomy,challenges,and | | | | | Principles. | | | | | | -| ---------------- | ---------------------------------- | ---- | -------- | -------- | --- | ----------- | ------------ | -------- | ----------------- | --- | ----- | -| openquestions. | CoRR,abs/2311.05232. | | | | | | | | | | | -| | | | | | | Patrick | S. H. Lewis, | Ethan | Perez, Aleksandra | | Pik- | -| | | | | | | tus, Fabio | Petroni, | Vladimir | Karpukhin, | | Naman | -| Ziwei Ji, Nayeon | Lee, | Rita | Frieske, | Tiezheng | Yu, | | | | | | | -DanSu,YanXu,EtsukoIshii,YejinBang,Andrea Goyal,HeinrichKüttler,MikeLewis,Wen-tauYih, -Madotto,andPascaleFung.2023. Surveyofhalluci- Tim Rocktäschel, Sebastian Riedel, and Douwe -nationinnaturallanguagegeneration. ACMComput. Kiela. 2020. Retrieval-augmented generation for -Surv.,55(12):248:1–248:38. knowledge-intensiveNLPtasks. InAdvancesinNeu- -| | | | | | | ralInformationProcessingSystems33: | | | | AnnualCon- | | -| --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ---------- | --- | -AlbertQ.Jiang,AlexandreSablayrolles,ArthurMen- -ferenceonNeuralInformationProcessingSystems -sch,ChrisBamford,DevendraSinghChaplot,Diego 2020,NeurIPS2020,December6-12,2020,virtual. -| de Las Casas, | Florian | Bressand, | | Gianna | Lengyel, | | | | | | | -| ------------- | ------- | --------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | -Guillaume Lample, Lucile Saulnier, Lélio Re- Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, -nard Lavaud, Marie-Anne Lachaux, Pierre Stock, Ziyang Chen, Baotian Hu, Aiguo Wu, and Min -TevenLeScao,ThibautLavril,ThomasWang,Timo- Zhang. 2023. A survey of large language models -| | | | | | | attribution. | CoRR,abs/2311.03731. | | | | | -| ----------------------------------- | --- | --- | --- | --- | ------- | ------------ | -------------------- | --- | --- | --- | --- | -| théeLacroix,andWilliamElSayed.2023. | | | | | Mistral | | | | | | | -7b. CoRR,abs/2310.06825. -| | | | | | | Nelson F. | Liu, Tianyi | Zhang, | and Percy | Liang. | 2023. | -| --- | --- | --- | --- | --- | --- | --------- | ----------- | ------ | --------- | ------ | ----- | -Albert Q. Jiang, Alexandre Sablayrolles, Antoine Evaluatingverifiabilityingenerativesearchengines. -Roux,ArthurMensch,BlancheSavary,ChrisBam- InFindingsoftheAssociationforComputationalLin- -ford,DevendraSinghChaplot,DiegodeLasCasas, guistics: EMNLP2023,Singapore,December6-10, -Emma Bou Hanna, Florian Bressand, Gianna 2023,pages7001–7025.AssociationforComputa- -| Lengyel, | Guillaume | Bour, | Guillaume | | Lample, | tionalLinguistics. | | | | | | -| ------------ | --------- | ------ | --------- | --------- | ------- | ------------------ | --- | --- | --- | --- | --- | -| Lélio Renard | Lavaud, | Lucile | | Saulnier, | Marie- | | | | | | | -AnneLachaux,PierreStock,SandeepSubramanian, JacobMenick,MajaTrebacz,VladimirMikulik,John -Sophia Yang, Szymon Antoniak, Teven Le Scao, Aslanides, H. Francis Song, Martin J. Chadwick, -Théophile Gervet, Thibaut Lavril, Thomas Wang, Mia Glaese, Susannah Young, Lucy Campbell- -| | | | | | | Gillingham, | Geoffrey | | Irving, and Nat | McAleese. | | -| --------------------------------------- | --- | --- | --- | --- | ---- | ----------- | -------- | --- | --------------- | --------- | --- | -| TimothéeLacroix,andWilliamElSayed.2024. | | | | | Mix- | | | | | | | -2022. Teachinglanguagemodelstosupportanswers -| tralofexperts. | CoRR,abs/2401.04088. | | | | | | | | | | | -| --------------- | -------------------- | ------- | ----- | ------ | ------ | ------------------- | --- | -------------------- | --- | --- | --- | -| | | | | | | withverifiedquotes. | | CoRR,abs/2203.11147. | | | | -| Ehsan Kamalloo, | Aref | Jafari, | Xinyu | Zhang, | Nandan | | | | | | | -Thakur,andJimmyLin.2023. HAGRID:Ahuman- ReiichiroNakano,JacobHilton,SuchirBalaji,JeffWu, -llmcollaborativedatasetforgenerativeinformation- Long Ouyang, Christina Kim, Christopher Hesse, -seekingwithattribution. CoRR,abs/2307.16883. ShantanuJain,VineetKosaraju,WilliamSaunders, -| | | | | | | Xu Jiang, | Karl | Cobbe, | Tyna Eloundou, | Gretchen | | -| --- | --- | --- | --- | --- | --- | --------- | ---- | ------ | -------------- | -------- | --- | -Krueger,KevinButton,MatthewKnight,Benjamin -VladimirKarpukhin,BarlasOguz,SewonMin,Patrick -S.H.Lewis,LedellWu,SergeyEdunov,DanqiChen, Chess,andJohnSchulman.2021. Webgpt: Browser- -andWen-tauYih.2020. Densepassageretrievalfor assisted question-answering with human feedback. -open-domainquestionanswering. InProceedingsof CoRR,abs/2112.09332. -the2020ConferenceonEmpiricalMethodsinNat- -| | | | | | | Jianmo | Ni, Chen | Qu, Jing | Lu, Zhuyun | Dai, | Gus- | -| ------------- | ----------- | --- | ----- | ----- | ------- | ------ | -------- | -------- | ---------- | ---- | ---- | -| ural Language | Processing, | | EMNLP | 2020, | Online, | | | | | | | -November16-20,2020,pages6769–6781.Associa- tavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, -tionforComputationalLinguistics. YiLuan,KeithB.Hall,Ming-WeiChang,andYinfei -| | | | | | | Yang.2022. | Largedualencodersaregeneralizable | | | | | -| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------- | --- | --- | --- | --- | -SayaliKulkarni,SheideChammas,WanZhu,FeiSha, retrievers. InProceedingsofthe2022Conferenceon -and Eugene Ie. 2020. Aquamuse: Automatically EmpiricalMethodsinNaturalLanguageProcessing, -generatingdatasetsforquery-basedmulti-document EMNLP2022,AbuDhabi,UnitedArabEmirates,De- -summarization. CoRR,abs/2010.12694. cember7-11,2022,pages9844–9855.Association -forComputationalLinguistics. -| TomKwiatkowski, | JennimariaPalomaki, | | | | OliviaRed- | | | | | | | -| --------------- | ------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | -field,MichaelCollins,AnkurP.Parikh,ChrisAlberti, OpenAI. 2023. GPT-4 technical report. CoRR, -| DanielleEpstein,IlliaPolosukhin,JacobDevlin,Ken- | | | | | | abs/2303.08774. | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -tonLee,KristinaToutanova,LlionJones,Matthew -LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, -| Kelcey, Ming-Wei | | Chang, | Andrew | M. | Dai, Jakob | | | | | | | -| ---------------- | -------- | -------- | ------- | ----- | ---------- | ------- | -------------- | --- | --------------- | --- | ----- | -| | | | | | | Carroll | L. Wainwright, | | Pamela Mishkin, | | Chong | -| Uszkoreit, | Quoc Le, | and Slav | Petrov. | 2019. | Natu- | | | | | | | -ralquestions: abenchmarkforquestionanswering Zhang,SandhiniAgarwal,KatarinaSlama,AlexRay, -research. Trans.Assoc.Comput.Linguistics,7:452– JohnSchulman,JacobHilton,FraserKelton,Luke -| 466. | | | | | | Miller,MaddieSimens,AmandaAskell,PeterWelin- | | | | | | -| ---- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | -der,PaulF.Christiano,JanLeike,andRyanLowe. -Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying 2022. Training languagemodelsto followinstruc- -Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. tionswithhumanfeedback. InAdvancesinNeural -Gonzalez, Hao Zhang, and Ion Stoica. 2023. Effi- InformationProcessingSystems35: AnnualConfer- -cientmemorymanagementforlargelanguagemodel enceonNeuralInformationProcessingSystems2022, - -NeurIPS2022,NewOrleans,LA,USA,November28 Doshi,RenelitoDelosSantos,TojuDuke,JohnnySo- -| -December9,2022. | | | | | | | raker,BenZevenbergen,VinodkumarPrabhakaran, | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | ----------- | --- | ------- | ----------- | -| | | | | | | | Mark Diaz, | Ben | Hutchinson, | | Kristen | Olson, Ale- | -AleksandraPiktus,FabioPetroni,VladimirKarpukhin, jandraMolina,ErinHoffman-John,JoshLee,Lora -DmytroOkhonko,SamuelBroscheit,GautierIzacard, -| | | | | | | | Aroyo, | Ravi Rajakumar, | | Alena | Butryna, | Matthew | -| --- | --- | --- | --- | --- | --- | --- | ------ | --------------- | --- | ----- | -------- | ------- | -Patrick S. H. Lewis, Barlas Oguz, Edouard Grave, Lamm,ViktoriyaKuzmina,JoeFenton,AaronCo- -Wen-tauYih,andSebastianRiedel.2021. Theweb hen,RachelBernstein,RayKurzweil,BlaiseAgüera -isyouroyster-knowledge-intensiveNLPagainsta yArcas,ClaireCui,MarianCroak,EdH.Chi,and -verylargewebcorpus. CoRR,abs/2112.09924. QuocLe.2022. Lamda: Languagemodelsfordialog -CoRR,abs/2201.08239. -applications. -KrishnaPillutla,SwabhaSwayamdipta,RowanZellers, -JohnThickstun,SeanWelleck,YejinChoi,andZaïd -| | | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | -Harchaoui.2021. MAUVE:measuringthegapbe- bert, Amjad Almahairi, Yasmine Babaei, Nikolay -tweenneuraltextandhumantextusingdivergence Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti -| frontiers. | In | Advances | in Neural | Information | | Pro- | | | | | | | -| ---------- | --- | -------- | --------- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | -Bhosale,DanBikel,LukasBlecher,CristianCanton- -| cessingSystems34: | | AnnualConferenceonNeural | | | | | | | | | | | -| ----------------- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Ferrer,MoyaChen,GuillemCucurull,DavidEsiobu, -InformationProcessingSystems2021,NeurIPS2021, -JudeFernandes,JeremyFu,WenyinFu,BrianFuller, -December6-14,2021,virtual,pages4816–4828. -CynthiaGao,VedanujGoswami,NamanGoyal,An- -thonyHartshorn,SagharHosseini,RuiHou,Hakan -RonakPradeep,SahelSharifymoghaddam,andJimmy Inan,MarcinKardas,ViktorKerkez,MadianKhabsa, -| Lin.2023. | Rankvicuna: | | Zero-shotlistwisedocument | | | | | | | | | | -| --------- | ----------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -IsabelKloumann,ArtemKorenev,PunitSinghKoura, -rerankingwithopen-sourcelargelanguagemodels. -Marie-AnneLachaux,ThibautLavril,JenyaLee,Di- -CoRR,abs/2309.15088. -anaLiskovich,YinghaiLu,YuningMao,XavierMar- -tinet,TodorMihaylov,PushkarMishra,IgorMoly- -| Rafael Rafailov, | | Archit | Sharma, | Eric | Mitchell, | Ste- | | | | | | | -| ---------------- | --- | ------ | ------- | ---- | --------- | ---- | ---------- | ---- | ------ | -------- | ------ | ------- | -| | | | | | | | bog, Yixin | Nie, | Andrew | Poulton, | Jeremy | Reizen- | -fanoErmon,ChristopherD.Manning,andChelsea -stein,RashiRungta,KalyanSaladi,AlanSchelten, -| Finn. | 2023. | Direct preference | | optimization: | | Your | | | | | | | -| ------------------------------------ | ----- | ----------------- | --- | ------------- | --- | ----- | -------------- | ----- | ------- | ------ | ------ | --------- | -| | | | | | | | Ruan Silva, | Eric | Michael | Smith, | Ranjan | Subrama- | -| languagemodelissecretlyarewardmodel. | | | | | | CoRR, | | | | | | | -| | | | | | | | nian, Xiaoqing | Ellen | Tan, | Binh | Tang, | Ross Tay- | -abs/2305.18290. -| | | | | | | | lor, Adina | Williams, | Jian | Xiang | Kuan, | Puxin Xu, | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | ---- | ----- | ----- | --------- | -ZhengYan,IliyanZarov,YuchenZhang,AngelaFan, -| Jeff Rasley, | Samyam | Rajbhandari, | | Olatunji | | Ruwase, | | | | | | | -| ------------ | ------ | ------------ | --- | -------- | --- | ------- | --- | --- | --- | --- | --- | --- | -and Yuxiong He. 2020. Deepspeed: System opti- Melanie Kambadur, Sharan Narang, Aurélien Ro- -driguez,RobertStojnic,SergeyEdunov,andThomas -mizationsenabletrainingdeeplearningmodelswith -| | | | | | | | Scialom.2023. | Llama2: | | Openfoundationandfine- | | | -| ------------------------- | --- | --- | --- | --------- | ------- | --- | ---------------- | ------- | -------------------- | ---------------------- | --- | --- | -| over100billionparameters. | | | | InKDD’20: | The26th | | | | | | | | -| | | | | | | | tunedchatmodels. | | CoRR,abs/2307.09288. | | | | -ACMSIGKDDConferenceonKnowledgeDiscovery -| and Data | Mining, | Virtual | Event, | CA, | USA, | August | | | | | | | -| -------- | ------- | ------- | ------ | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | -23-27,2020,pages3505–3506.ACM. TheodoraWorledge,JudyHanwenShen,NicoleMeis- -| | | | | | | | ter,CalebWinston,andCarlosGuestrin.2023. | | | | | Uni- | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ---- | -fyingcorroborativeandcontributiveattributionsin -| Freda Shi, | Xinyun | Chen, | Kanishka | Misra, | | Nathan | | | | | | | -| ---------- | ------ | ----- | -------- | ------ | --- | ------ | -------------------- | --- | --- | -------------------- | --- | --- | -| | | | | | | | largelanguagemodels. | | | CoRR,abs/2311.12233. | | | -Scales,DavidDohan,EdH.Chi,NathanaelSchärli, -| andDennyZhou.2023. | | | Largelanguagemodelscan | | | | | | | | | | -| ------------------ | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -beeasilydistractedbyirrelevantcontext. InInterna- Fangyuan Xu, Weijia Shi, and Eunsol Choi. 2023a. -tionalConferenceonMachineLearning,ICML2023, RECOMP:improvingretrieval-augmentedlmswith -23-29 July 2023, Honolulu, Hawaii, USA, volume compression and selective augmentation. CoRR, -abs/2310.04408. -202ofProceedingsofMachineLearningResearch, -pages31210–31227.PMLR. -ShichengXu,LiangPang,HuaweiShen,XueqiCheng, -IvanStelmakh,YiLuan,BhuwanDhingra,andMing- andTat-SengChua.2023b. Search-in-the-chain: To- -Wei Chang. 2022. ASQA: factoid questions meet wards the accurate, credible and traceable content -long-formanswers. InProceedingsofthe2022Con- generation for complex knowledge-intensive tasks. -CoRR,abs/2304.14732. -ferenceonEmpiricalMethodsinNaturalLanguage -Processing,EMNLP2022,AbuDhabi,UnitedArab -Emirates,December7-11,2022,pages8273–8288. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan -AssociationforComputationalLinguistics. Berant. 2023. Making retrieval-augmented lan- -| | | | | | | | guage models | robust | to | irrelevant | context. | CoRR, | -| ----- | ---------- | ------ | ------------- | -------- | -------- | ----- | --------------- | ------ | --- | ---------- | -------- | ----- | -| Romal | Thoppilan, | Daniel | De | Freitas, | Jamie | Hall, | abs/2310.01558. | | | | | | -| Noam | Shazeer, | Apoorv | Kulshreshtha, | | Heng-Tze | | | | | | | | -Cheng,AliciaJin,TaylorBos,LeslieBaker,YuDu, Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, -YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Xiaolei Wang, Yupeng Hou, Yingqian Min, Be- -AminGhafouri,MarceloMenegali,YanpingHuang, ichenZhang,JunjieZhang,ZicanDong,YifanDu, -MaximKrikun,DmitryLepikhin,JamesQin,Dehao Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao -Chen,YuanzhongXu,ZhifengChen,AdamRoberts, Jiang,RuiyangRen,YifanLi,XinyuTang,Zikang -MaartenBosma,YanqiZhou,Chung-ChingChang, Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. -IgorKrivokon,WillRusch,MarcPickett,KathleenS. 2023. A survey of large language models. CoRR, -| Meier-Hellstern, | | Meredith | Ringel | Morris, | | Tulsee | abs/2303.18223. | | | | | | -| ---------------- | --- | -------- | ------ | ------- | --- | ------ | --------------- | --- | --- | --- | --- | --- | - -Kun Zhu, Xiaocheng Feng, Xiyuan Du, Yuxuan Gu, process, we retained 5,667 and 2,431 queries, re- -WeijiangYu,HaotianWang,QianglongChen,Zheng spectively. Additionally,wecalculatedtheaverage -| Chu, JingchangChen, | | andBingQin.2024. | | Anin- | | | | | | | -| ------------------- | --- | ---------------- | --- | ----- | --- | --- | --- | --- | --- | --- | -lengthofanswersandtheaveragenumberofcita- -formationbottleneckperspectiveforeffectivenoise -tionsgeneratedforvarioustypesofquerieswithin -filteringonretrieval-augmentedgeneration. -ourdataset,asshowninTable4. -YutaoZhu,HuayingYuan,ShutingWang,JiongnanLiu, -| Wenhan | Liu, Chenlong | Deng, | Zhicheng | Dou, and | | | | | | | -| ------ | ------------- | ----- | -------- | -------- | --- | --- | --- | --- | --- | --- | -A.2 DetailsofDataFiltering -| Ji-RongWen.2023. | | Largelanguagemodelsforinfor- | | | | | | | | | -| ---------------- | --- | ---------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | -| mationretrieval: | | Asurvey. | CoRR,abs/2308.07107. | | | | | | | | -WetrainedourAttributedDiscriminatorusingthe -| | | | | | manually | annotated | data | provided | by | Liu et al. | -| --- | --- | --- | --- | --- | -------- | --------- | ---- | -------- | --- | ---------- | -A DetailsofDataGenerationPipeline -| | | | | | (2023), | which is | sampled | from | real generative | | -| --- | --- | --- | --- | --- | ------- | -------- | ------- | ---- | --------------- | --- | -A.1 DataStatistic searchengines. Eachstatementanditsciteddocu- -menthavebeenmeticulouslyannotatedforattribu- -#Questions 8,098 tion,categorizedintothreetypes: completesupport, -| | | | | | partialsupport,andnosupport. | | | | Fortraining,weuti- | | -| ------------ | --- | --- | --- | ---- | ---------------------------- | --- | --- | --- | ------------------ | --- | -| ➥#LongAnswer | | | | 5667 | | | | | | | -lizedadatasetof8,834instances,comprising6,415 -| ➥#ShortAnswer | | | | 2431 | | | | | | | -| ------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | -instancesofcompletesupport,1,552ofpartialsup- -Avg. WordsperAnswer 50.48 port, and 867 of no support. The discriminator -initializedwithLLaMA-2-7B,wastrainedwitha -| ➥Avg. | WordsperLongAnswer | | | 69.15 | | | | | | | -| ----- | ------------------- | --- | --- | ----- | --------------------------- | --- | --- | --- | -------------- | --- | -| | | | | | maximumsequencelengthof512. | | | | Wetraineditfor | | -| ➥Avg. | WordsperShortAnswer | | | 6.94 | | | | | | | -3epochs,withatotalbatchsizeof128,andapeak -Avg. CitationperAnswer 4.40 learning rate of 2e-5, incorporating 3% warmup -steps,followedbyalineardecay. -| ➥Avg. | CitationperLongAnswer | | | 4.68 | | | | | | | -| ----- | ---------------------- | --- | --- | ---- | ------ | ----------------- | --------- | --------- | ---------- | ----- | -| ➥Avg. | | | | | During | the data | filtering | stage, | we first | break | -| | CitationperShortAnswer | | | 3.77 | | | | | | | -| | | | | | down | the automatically | | generated | attributed | an- | -swersintostatementformandusethetraineddis- -| Table 4: | The statistics | of the | data generated | by our | | | | | | | -| -------- | -------------- | ------ | -------------- | ------ | --- | --- | --- | --- | --- | --- | -criminatortoannotatetheattributionbetweeneach -automaticdatagenerationpipeline. -| | | | | | statement | and its | cited | documents. | Specifically, | | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ---------- | ------------- | --- | -weassigndifferentattributionscorestoeachstate- -Table4presentsthestatisticsofthedataautomat- -| | | | | | ment s | based on | its attribution | | relationship | with | -| ------------------------------------------- | --- | --- | --- | --- | ----------------------------------- | -------- | --------------- | --- | ------------ | ---- | -| icallygeneratedbyourdatagenerationpipeline. | | | | In | | | | | | | -| | | | | | citeddocumentsd,asshowninEquation7. | | | | | Con- | -total,wecollected8,098questionsfromtheNatural -sequently,foreachattributedanswer,wecancal- -Questions(NQ)dataset,ofwhich5,667questions -| | | | | | culateitsaverageattributionscore. | | | | Attributedan- | | -| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ------------- | --- | -weregatheredfromthosewithlong-formanswers, -swerswithanaverageattributionscorebelow0.8 -and2,431questionswerecollectedfromthosewith -| | | | | | are filtered | out. | The threshold | | of 0.8 was | deter- | -| --- | --- | --- | --- | --- | ------------ | ---- | ------------- | --- | ---------- | ------ | -short-formfactoidanswers. -minedthroughpreliminarytestingonthedevelop- -Forquestionsrequiringlong-formanswers,we -| | | | | | ment set, | for which | we | manually | annotated | 100 | -| --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --------- | --- | -initializedourquerysourcewiththeAQUAMUSE -samplestoensuretheeffectivenessofourfiltering -| dataset (Kulkarni | | et al., 2020), | which consists | of | | | | | | | -| ----------------- | --- | -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | -criteria. -high-qualityqueriesspecificallydesignedforlong- -formresponseswithintheNQdataset,recognized -| as“good”bythemajorityofNQevaluators. | | | | Inthis | | | | | | | -| ------------------------------------ | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | - -| | | | | | | 1, Dis(s,d) | | = complete | support | | -| -------------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | -| way,utilizingarefinedandsuperiorqualityquery | | | | | |   | | | | | -setlaidarobustgroundworkforourtrainingdata r(s)= 0.5,Dis(s,d) = partial support - -generation,streamliningthedatafilteringprocess. 0, Dis(s,d) = no support -| Forfactoidqueriesthatnecessitateshort-forman- | | | | | | | | | | (7) | -| --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -swers,wedirectlysampledfromtheoriginalNQ -| dataset, leveraging | | its abundance | and | inherently | | | | | | | -| ------------------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | -B DetailsofEvaluationMetrics -highquality. -During the data generation process, our initial In addition to evaluating citation quality and cor- -querysetcomprised7,725queriesrequiringlong- rectness,theALCEbenchmarkincludesabroader -formanswersand4,000queriesnecessitatingshort- setofdimensions,suchasfluency,ROUGE-L,and -| form answers. | | After a two-stage | data | filtering | generationlength. | | | | | | -| ------------- | --- | ----------------- | ---- | --------- | ----------------- | --- | --- | --- | --- | --- | - -Fluency Weevaluatethefluencyofthegenerated framework(Kwonetal.,2023)forefficientinfer- -responseusingMAUVE(Pillutlaetal.,2021). No- ence. Thehyperparametersaresetasillustratedin -| tably,wecalculatefluencyonlyfortheASQAand | | | | | | | Table9. | | | | | | | -| ----------------------------------------- | --- | -------- | ------ | -------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | -| ELI5 datasets, | | omitting | it for | QAMPARI, | | as the | | | | | | | | -response in QAMPARI typically consists of lists D.2 RetrievalSettings -| ofshortanswers. | | ArelativelyhighMAUVEscore | | | | | | | | | | | | -| --------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Duringtheevaluation,weadoptthesameretrieval -indicatesthatthegenerationissufficientlyfluent. -| | | | | | | | settings | as specified | | by Gao | et | al. (2023b). | For | -| ------- | --------------------------------- | --- | --- | -------- | --- | ------ | -------------------------------- | ------------ | ------- | ------ | --------- | ------------ | ------- | -| | | | | | | | the ASQA | and | QAMPARI | | datasets, | we | use the | -| ROUGE-L | Inadditiontoevaluatingthecorrect- | | | | | | | | | | | | | -| | | | | | | | denseretrieverGTR(Nietal.,2022). | | | | | FortheELI5 | | -| ness of | the model-generated | | | content, | we | employ | | | | | | | | -dataset,weemploythesparseretrieverBM25. -ROUGE-Ltoassesstheoverallqualityandtextual -coherenceoftheresponses. -E MoredetailaboutAblationStudy -C Prompts -E.1 TheEffectofTrainingDataScale. -C.1 PromptsforPrompting-basedMethods Weexaminehowmodelperformancevarieswith -| | | | | | | | changesindatascale,asdepictedinFigure7. | | | | | | The | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -FollowingGaoetal.(2023b),weadoptthevanilla -| | | | | | | | upper part | of | the figure | illustrates | | the | impact of | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | ----------- | --- | --- | --------- | -promptingstrategyforitssimplicityandeffective- -| | | | | | | | the training | data | scale | on | citation | quality | during | -| ------------------------------------ | ------ | -------- | -------- | -------- | --------- | --------- | ------------- | ---- | ------- | ---------- | ---------- | -------- | --------- | -| ness. Specifically, | | the | prompts | vary | according | to | | | | | | | | -| | | | | | | | the Grounding | | Guided | Generation | | training | stage, | -| thetypeofdatawithintheALCEbenchmark. | | | | | | For | | | | | | | | -| | | | | | | | with datasets | | ASQA, | ELI5, | and | QAMPARI | rep- | -| long-form | QA | datasets | such | as ASQA | | and ELI5, | | | | | | | | -| | | | | | | | resented | from | left to | right. | Similarly, | | the lower | -| the prompt | format | is | detailed | in Table | 5. | For the | | | | | | | | -partofthefiguredescribestheinfluenceduringthe -| short-form | QA | dataset | QAMPARI, | | the | format is | | | | | | | | -| ---------- | --- | ------- | -------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -Consistency-AwareAlignmenttrainingstage. -outlinedinTable6. -E.2 TheGeneralizationAcrossModel -| C.2 Instructionsfor | | | FRONT | | | | | | | | | | | -| ------------------- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Architectures. -Duringthetrainingprocess,wefollowtheinstruc- -| | | | | | | | FRONT | demonstrates | | exceptional | | generalization | | -| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | --- | ----------- | --- | -------------- | --- | -Alpaca6. -| tion format | of | | Specifically, | | we | employ | | | | | | | | -| ----------- | --- | --- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -capabilitiesacrossvariousfoundationalmodelar- -variedinstructionsfordifferentquestiontypes,as -| | | | | | | | chitectures. | Specifically,transitioningthefounda- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------------------------ | --- | --- | --- | --- | --- | -delineatedinTable7forlong-formquestionsand -| | | | | | | | tional model | | from LLaMA-2-7B | | | to the | stronger | -| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --- | ------ | -------- | -Table8forshort-formquestions. -| | | | | | | | foundational | | model, | Mistral-7B, | | results | in even | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ----------- | ------ | ------------ | --- | ------- | -------- | -| | | | | | | | greater | performance | | enhancements | | as | shown in | -D ExperimentalDetails -| | | | | | | | Figure8. | Thisfurtherunderscoresthebroadappli- | | | | | | -| --------------------- | --- | --- | ----- | --- | --- | --- | ----------------------------- | ------------------------------------ | --- | --- | ------ | --- | --- | -| | | | | | | | cabilityandgeneralizabilityof | | | | FRONT. | | | -| D.1 TrainingDetailsof | | | FRONT | | | | | | | | | | | -Thetrainingofallmodelsisexecutedon4Nvidia E.3 Theeffectofβ inConsistency-Aware -A100GPUs,eachwith80GBofmemory,leverag- AlignmentTrainingStage -ingtheDeepspeed(Rasleyetal.,2020)andHug- -| | | | | | | | In the | Consistency-Aware | | | Alignment | | Training | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------- | --- | --- | --------- | --- | -------- | -gingFaceAcceleratelibraries(Guggeretal.,2022) -| | | | | | | | Stage, | the β | parameter | in | Direct | Preference | Op- | -| -------------------------------------- | --- | --- | --- | --- | --- | ----- | ---------- | ----- | --------- | --- | ------ | ---------- | ------ | -| toconductmulti-GPUdistributedtraining. | | | | | | Given | | | | | | | | -| | | | | | | | timization | (DPO) | controls | | the | strength | of the | -thelongnatureoftheinputs,themaximumtoken -Kullback-Leiblerpenalty,typicallysetwithinthe -lengthissetto2,048tokens. -| | | | | | | | range of | 0.1 to | 0.5. | A higher | β | value | indicates a | -| ------ | --- | --------- | ----- | ---------- | --- | ------ | -------- | ------ | ---- | -------- | --- | ----- | ----------- | -| During | the | grounding | guide | generation | | stage, | | | | | | | | -preferenceforthepolicymodel’strainingprocess -modelsaretrainedfor5epochswithatotalbatch -toremainclosertotheinitiallyreferencedmodel. -| size of | 128, a peak | learning | | rate of | 2e-5 | with 3% | | | | | | | | -| -------------------------------------- | ----------- | -------- | --- | ------- | ---- | ------- | ----------------------------------- | ------ | --- | --- | ----- | ------ | -------- | -| | | | | | | | In extreme | cases, | as | β → | 0, we | ignore | the con- | -| warmupstepsfollowedbyalineardecay. | | | | | | During | | | | | | | | -| | | | | | | | straintsimposedbythereferencemodel. | | | | | | Thisset- | -| thecontrastivealignmentstage,wesettheβ | | | | | | to0.1 | | | | | | | | -tingaimstobalancethemodel’sabilitytoadaptto -andcontinuedtrainingfortwoadditionalepochs. -newtrainingsignalswhilemaintainingthestability -| Specifically, | During | | inference, | we | use | the vllm | | | | | | | | -| ------------- | ------ | --- | ---------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -ofthelearnedbehaviorsfromthereferencemodel. -| | | | | | | | Subsequently, | | we | trained | five | variants | by ad- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | ------- | ---- | -------- | ------ | -6https://github.com/tatsu-lab/stanford_alpaca/ -| tree/main | | | | | | | justingβ | from0.1to0.5onthemodelpreviously | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --- | - -Instruction: Write an accurate, engaging, and concise answer for the given question using only -the provided search results (some of which might be irrelevant) and cite them properly. Use an -unbiased and journalistic tone. Always cite for any factual claim. When citing several search -results, use [1][2][3]. Cite at least one document and at most three documents in each sentence. -If multiple documents support the sentence, only cite a minimum sufficient subset of the -documents. -Table5: PromptforLong-formQA. -Instruction: Provide a list of accurate answers for the given question using only the provided -search results (some of which might be irrelevant) and cite them properly. Always cite one and -only one document for each answer. Separate answers by commas. For questions that have more than -| 5 answers, | write | at least 5 answers. | | | | | -| ---------- | ----- | ------------------- | --- | --- | --- | --- | -Table6: PromptforShort-formQA. -| 72.0 | | | 48.0 | | 23.0 | | -| ---- | --- | ----- | ---- | ----- | ---- | ----- | -| 68.0 | | | 46.0 | | 22.0 | | -| 64.0 | | | 44.0 | | 21.0 | | -| 60.0 | | REC. | 42.0 | REC. | 20.0 | REC. | -| 56.0 | | PREC. | 40.0 | PREC. | 19.0 | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| ---- | --- | -------- | ---- | -------- | ---- | -------- | -| 76.0 | | | 54.0 | | 25.0 | | -| 72.0 | | | 52.0 | | 24.0 | | -| 68.0 | | | 50.0 | | 23.0 | | -| 64.0 | | | 48.0 | | 22.0 | | -| | | REC. | | REC. | | REC. | -| 60.0 | | | 46.0 | | 21.0 | | -| | | PREC. | | PREC. | | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| --- | --- | -------- | --- | -------- | --- | -------- | -Figure7: Ablationstudyonsynthetictrainingdatasize: TheupperpartofthefigurecorrespondstotheGrounding -GuidedGenerationtrainingstage, whilethebottompartrepresentstheWeak-to-StrongContrastiveAlignment -trainingstage. Fromlefttoright,theresultsarepresentedforASQA,ELI5,andQAMPARI,respectively. REC. -indicatesCitationRecallandPREC. denotesCitationPrecision. Thex-axisrepresentsthequantityofautomatically -generateddata. Itisobservedthatasthevolumeofautomaticallygenerateddataincreases,thereisaconsistent -improvementinbothcitationrecallandprecisionacrossthetwotrainingstages. -trainedwithG3 toexploretheimpactofthehyper- adheretotheevaluationframeworkestablishedin -parameter β on attribution quality. We evaluated (Gaoetal.,2023b). Forlong-formQAdatasetslike -thesevariantsontheASQAandELI5datasets,and ASQAandELI5,wealsoreportmetricsrelatedto -theexperimentalresultsareshowninFigure9. Fluency,ROUGE-L,andaverageresponselength. -The experimental results indicate that as β in- Specifically,weuseMAUVE(Pillutlaetal.,2021) -creases, the model’s performance on attribution toevaluatethefluencyofthemodelresponse. For -graduallydecreases. Thisobservationsuggeststhat datasetslikeQAMPARI,whereanswersarecom- -thefirststageofG3 mightintroduceanoticeable posed of concatenated entities, we calculate the -inconsistencybetweengroundingandattribution. averagenumberofpredictedentities. -Withhigherβvalues,themodelstrugglestoescape -theconstraintsofinconsistentattributedanswers, -| leading to | a reduction | in attribution | quality as | β | | | -| ---------- | ----------- | -------------- | ---------- | --- | --- | --- | -increases. -F FullResults -Wepresentthecomprehensiveresultsofourexperi- -| mentsinTables10,11,and12. | | Beyondtheevalua- | | | | | -| ------------------------- | --- | ---------------- | --- | --- | --- | --- | -tionmetricsrelatedtoCorrectnessandCitation,we - -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -| guide | answer | generation | and | cite the | sources properly. | | | | | | | -| ------------------- | ------ | ---------- | --- | ---------- | ----------------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table7: | InstructionFormatforFRONTonLong-formQA. | | | | | | | | -| --- | --- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -provide a list of accurate answers for the given question. Always cite one and only one document -| for each | answer. | Separate | answers | by | commas. | | | | | | | -| ------------------- | ------- | ---------- | ------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table8: | InstructionFormatforFRONTonShort-formQA. | | | | | | | | -| ------- | ----------------------------------- | --- | ------- | ---------------------------------------- | --- | ------------- | --- | --- | ---------------- | --- | --- | -| | Hyper-parameters | | | Value | | | | | | | | -| | Top-p | | | 0.95 | | | | | | | | -| | Temperature | | | 0.2 | | | | | | | | -| | Max-length | | | 2048 | | | | | | | | -| Table9: | Hyper-parametersettingsininference. | | | | | | | | | | | -| | | | | | | | 72 | | 56 | | | -| | | | | | | )%(1FnoitatiC | 71 | | )%(1FnoitatiC 55 | | | -Mistral-7B -| | | | VANILLA-SFT(Mistral-7B) | | | | 70 | | 54 | | | -| --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | FRONT(Mistral-7B) | | | | 69 | | 53 | | | -60 -| | | | | | | | 68 | | 52 | | | -| ---------------- | --- | --- | --- | --- | --- | --- | -------------- | ----------- | --- | -------------- | ----------- | -| )%(1FnoitatiC 50 | | | | | | | 67 | FRONT-7B | 51 | | FRONT-7B | -| 40 | | | | | | | 0.1 0.2 | 0.3 0.4 0.5 | | 0.1 0.2 | 0.3 0.4 0.5 | -| 30 | | | | | | | βonASQAdataset | | | βonELI5dataset | | -20 -| | | | | | | Figure | 9: Ablation | on hyperparameter | | β | in Weak-to- | -| --- | --- | --- | --- | --- | --- | -------------------------------------------- | ----------- | ----------------- | --- | --- | ----------- | -| 10 | | | | | | StrongContrastiveAlignmentstageonASQAandELI5 | | | | | | +u +J +- +0 +3 + +0 +2 +0 +2 +- +g +u +A +- +9 +2 + +0 +2 +0 +2 +- +p +e +S +- +8 +2 + +0 +2 +0 +2 +- +t +c +O +- +8 +2 + +0 +2 +0 +2 +- +v +o +N +- +7 +2 + +0 +2 +0 +2 +- +c +e +D +- +7 +2 + +1 +2 +0 +2 +- +n +a +J +- +6 +2 + +1 +2 0 -| | ASQA | | ELI5 | QAMPARI | | | | | | | | -| --- | ---- | --- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Figure8:Ablationstudyonmodelarchitecture:Wesub- -stitutedthefoundationmodelinFRONTwithMistral-7B -andcomparedtheexperimentalresultsofmodelsunder -| the same | foundation | model | using | in-context | learning | | | | | | | -| -------- | ---------- | ----- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | -andthosedirectlysupervisedfine-tunedonourautomat- -| icallygenerateddata. | | Theexperimentsdemonstratethat | | | | | | | | | | -| -------------------- | --------- | ----------------------------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | -| by replacing | different | foundation | | models, | our frame- | | | | | | | -workstillmaintainsitsgeneralizability. - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | -------- | ---------- | ---------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (EMRec.) | Rec. Prec. | F1 ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 73.41 | 40.37 | 72.81 69.69 | 71.22 37.92 | 39.24 | -| ------------ | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 79.90 | 24.32 | 17.24 17.87 | 17.55 29.38 | 42.29 | -| LLaMA-2 | 13B | 87.08 | 27.99 | 16.45 19.04 | 17.65 31.41 | 39.25 | -| | 70B | 69.28 | 31.53 | 44.18 44.79 | 44.48 31.53 | 26.86 | -| | 7B | 66.78 | 29.93 | 55.99 51.66 | 53.74 32.93 | 26.18 | -| LLaMA-2-Chat | 13B | 66.14 | 34.39 | 37.15 38.17 | 37.65 35.13 | 33.68 | -| | 70B | 86.60 | 41.24 | 60.19 61.16 | 60.67 37.01 | 47.09 | -| | 7B | 86.92 | 38.34 | 48.37 44.63 | 46.42 35.95 | 63.90 | -Vicuna-v1.5 -| | 13B | 66.11 | 35.20 | 51.92 53.40 | 52.65 35.74 | 38.57 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.37 | 29.46 | 23.12 25.45 | 24.23 31.67 | 37.17 | -Mistral -| | 8×7B | 83.30 | 36.30 | 32.72 34.49 | 33.58 35.05 | 38.47 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.86 | 38.57 | 64.90 59.67 | 62.18 36.21 | 45.26 | -Mistral-Instruct -| | 8×7B | 94.77 | 44.11 | 61.80 63.27 | 62.53 38.54 | 58.83 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -Post-hocRetrieval -| ChatGPT | - | 49.78 | 37.68 | 27.11 27.05 | 27.08 36.64 | 52.61 | -| ------------ | ---- | ----- | ----- | ----------- | ----------- | ------ | -| | 7B | 75.56 | 16.55 | 13.88 13.86 | 13.87 26.81 | 37.50 | -| LLaMA-2 | 13B | 77.91 | 20.51 | 20.95 20.94 | 20.94 29.53 | 31.37 | -| | 70B | 75.23 | 27.58 | 28.43 28.43 | 28.43 30.33 | 29.88 | -| | 7B | 22.50 | 14.17 | 11.33 11.33 | 11.33 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 64.52 | 24.43 | 21.43 21.43 | 21.43 33.91 | 41.12 | -| | 70B | 70.63 | 29.68 | 24.51 24.51 | 24.51 34.17 | 45.74 | -| | 7B | 63.87 | 19.58 | 16.24 16.24 | 16.24 33.22 | 41.80 | -| Vicuna-v1.5 | 13B | 73.83 | 24.79 | 24.11 24.11 | 24.11 34.42 | 43.54 | -| | 7B | 86.54 | 21.17 | 16.78 16.77 | 16.77 30.90 | 42.43 | -| Mistral | 8×7B | 80.99 | 36.30 | 38.37 35.27 | 36.75 35.05 | 38.47 | -| | 7B | 67.97 | 26.26 | 17.87 17.85 | 17.86 33.71 | 51.56 | -Mistral-Instruct 8×7B 65.51 33.90 24.57 24.48 24.52 36.20 53.83 -Training-based -| | 7B | 74.33 | 29.96 | 67.82 66.97 | 67.39 35.70 | 29.83 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -Self-RAG -| | 13B | 71.59 | 31.66 | 71.26 70.35 | 70.80 36.01 | 27.03 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 76.66 | 40.32 | 67.67 63.67 | 65.61 38.32 | 62.00 | -VANILLA-SFT -| | 13B | 84.36 | 40.85 | 71.49 66.21 | 68.75 38.22 | 58.82 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 81.88 | 40.84 | 77.70 69.89 | 73.59 36.95 | 53.93 | -FRONT -| | 13B | 76.11 | 41.51 | 78.44 73.66 | 75.95 38.63 | 57.56 | -| --- | --- | -------- | ---------------- | ----------- | ----------- | ----- | -| | | Table10: | ASQAfullresults. | | | | - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | ------------ | -------- | ------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (Claim) Rec. | Prec. F1 | ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 44.65 | 12.47 49.44 | 47.05 48.22 | 20.64 | 90.2 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 63.72 | 4.53 3.92 | 5.38 4.54 | 18.27 | 103.36 | -| LLaMA-2 | 13B | 62.19 | 7.77 8.49 | 8.43 8.46 | 19.95 | 88.23 | -| | 70B | 53.39 | 10.43 23.75 | 22.43 23.07 | 20.43 | 93.84 | -| | 7B | 32.80 | 12.47 19.90 | 15.48 17.41 | 20.88 | 96.42 | -| LLaMA-2-Chat | 13B | 29.08 | 13.83 16.50 | 16.09 16.29 | 21.04 | 94.32 | -| | 70B | 33.69 | 13.30 36.63 | 36.63 36.63 | 21.29 | 117.84 | -| | 7B | 31.45 | 12.30 29.81 | 22.45 25.61 | 21.36 | 105.68 | -Vicuna-v1.5 -| | 13B | 37.41 | 14.33 31.15 | 28.99 30.03 | 21.74 | 98.23 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 56.62 | 8.47 16.04 | 16.32 16.18 | 20.46 | 93.80 | -Mistral -| | 8×7B | 61.83 | 10.43 26.11 | 25.09 25.59 | 20.66 | 93.59 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 32.74 | 11.07 49.25 | 42.69 45.74 | 20.75 | 98.28 | -Mistral-Instruct -| | 8×7B | 38.51 | 13.93 49.28 | 48.34 48.81 | 21.34 | 113.71 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -Post-hocRetrieval -| ChatGPT | - | 22.79 | 18.77 14.55 | 14.55 14.55 | 22.28 | 106.83 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 72.80 | 7.23 6.84 | 6.84 6.84 | 19.14 | 88.19 | -| LLaMA-2 | 13B | 53.21 | 10.33 9.61 | 9.61 9.61 | 20.63 | 90.44 | -| | 70B | 58.97 | 11.10 10.27 | 10.26 10.26 | 20.41 | 77.85 | -| | 7B | 22.50 | 14.17 11.33 | 11.33 11.33 | 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 30.36 | 14.93 12.10 | 12.10 12.10 | 21.82 | 109.79 | -| | 70B | 37.87 | 16.03 12.93 | 12.93 12.93 | 21.57 | 99.94 | -| | 7B | 30.88 | 11.83 10.91 | 10.91 10.91 | 21.66 | 99.03 | -Vicuna-v1.5 -| | 13B | 32.59 | 15.20 14.06 | 14.06 14.05 | 14.05 | 108.16 | -| --- | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 52.45 | 10.47 8.64 | 8.64 8.64 | 20.48 | 90.17 | -Mistral -| | 8×7B | 48.39 | 13.57 11.62 | 11.62 11.62 | 21.43 | 91.97 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 27.41 | 17.07 13.20 | 13.20 13.20 | 21.52 | 106.93 | -Mistral-Instruct -| | 8×7B | 27.60 | 17.37 15.68 | 15.68 15.68 | 21.66 | 95.21 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -Training-based -| | 7B | 30.98 | 6.90 22.34 | 32.40 26.45 | 16.48 | 41.66 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -Self-RAG -| | 13B | 32.04 | 6.07 30.46 | 40.20 34.66 | 15.23 | 38.19 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -| | 7B | 44.12 | 9.63 42.30 | 40.06 41.15 | 20.58 | 80.43 | -VANILLA-SFT -| | 13B | 46.33 | 10.27 46.75 | 44.47 45.58 | 20.56 | 84.01 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 36.90 | 9.18 58.60 | 55.33 56.92 | 19.09 | 74.06 | -FRONT -| | 13B | 34.37 | 9.32 60.31 | 59.21 59.75 | 19.66 | 75.14 | -| --- | --- | -------- | ---------------- | ----------- | ----- | ----- | -| | | Table11: | ELI5fullresults. | | | | - -| | | Correctness | Citation | | -| --------- | --------- | ------------ | ---------- | ----------- | -| ModelType | ModelSize | | | | -| | | Rec.-5 Prec. | Rec. Prec. | F1 NumPred. | -Prompting-based -| ChatGPT | - | 20.28 19.84 | 19.06 22.03 | 20.44 4.71 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 12.56 11.32 | 6.03 6.35 | 6.19 7.02 | -| LLaMA-2 | 13B | 18.00 12.39 | 5.45 5.74 | 5.59 11.31 | -| | 70B | 18.50 14.79 | 10.10 10.50 | 10.30 8.31 | -| | 7B | 17.96 19.74 | 9.58 9.68 | 9.63 4.73 | -| LLaMA-2-Chat | 13B | 21.34 18.86 | 8.94 9.06 | 9.00 6.51 | -| | 70B | 22.62 18.04 | 13.49 13.98 | 13.73 7.44 | -| | 7B | 14.22 14.74 | 11.26 11.64 | 11.45 5.87 | -Vicuna-v1.5 -| | 13B | 22.06 19.60 | 13.04 13.74 | 13.38 7.62 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 16.96 15.98 | 7.50 7.76 | 7.63 6.29 | -Mistral -| | 8×7B | 18.18 15.63 | 9.72 10.20 | 9.95 6.63 | -| --- | ---- | ----------- | ----------- | ---------- | -| | 7B | 17.52 21.29 | 17.56 18.53 | 18.03 4.54 | -Mistral-Instruct -| | 8×7B | 20.12 19.64 | 19.27 20.38 | 19.81 5.32 | -| --- | ---- | ----------- | ----------- | ---------- | -Post-hocRetrieval -| ChatGPT | - | 25.14 22.85 | 12.29 12.29 | 12.29 5.46 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 6.48 5.11 | 5.05 5.05 | 5.05 6.55 | -| LLaMA-2 | 13B | 9.88 7.17 | 5.20 5.20 | 5.20 6.98 | -| | 70B | 14.44 12.44 | 7.49 7.49 | 7.49 7.41 | -| | 7B | 12.94 10.89 | 7.76 7.76 | 7.76 5.99 | -| LLaMA-2-Chat | 13B | 15.72 12.23 | 7.87 7.87 | 7.87 6.32 | -| | 70B | 17.90 14.45 | 9.05 9.05 | 9.05 6.05 | -| | 7B | 12.04 9.71 | 6.69 6.69 | 6.69 7.10 | -Vicuna-v1.5 -| | 13B | 14.78 11.47 | 8.50 8.50 | 8.50 6.67 | -| --- | --- | ----------- | --------- | --------- | -| | 7B | 9.94 7.90 | 6.00 6.00 | 6.00 7.38 | -Mistral -| | 8×7B | 13.92 12.08 | 6.70 6.70 | 6.70 6.58 | -| --- | ---- | ----------- | --------- | --------- | -| | 7B | 15.80 12.15 | 8.34 8.34 | 8.34 7.01 | -Mistral-Instruct -| | 8×7B | 24.16 18.28 | 9.78 9.78 | 9.78 7.37 | -| --- | ---- | ----------- | --------- | --------- | -Training-based -| | 7B | 2.34 1.98 | 10.53 18.80 | 13.50 3.49 | -| --- | --- | --------- | ----------- | ---------- | -Self-RAG -| | 13B | 1.90 1.33 | 12.79 20.90 | 15.86 3.08 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 12.86 21.09 | 21.35 21.36 | 21.35 7.49 | -VANILLA-SFT -| | 13B | 12.68 22.80 | 23.64 23.71 | 23.67 3.14 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 11.50 21.38 | 24.74 24.84 | 24.79 3.08 | -FRONT -| | 13B | 11.94 22.61 | 24.86 25.39 | 25.12 3.17 | -| --- | -------- | ------------------- | ----------- | ---------- | -| | Table12: | QAMPARIfullresults. | | | \ No newline at end of file +2 +- +b +e +F +- +5 +2 + +1 +2 +0 +2 +- +r +a +M +- +7 +2 + +Wipro + +Sensex + +NYSE Composite Index + +139 + + Other Disclosures +Description of Voting Rights +Dematerialisation of Shares and Liquidity +Outstanding ADR/GDR/Warrants or any +other Convertible instruments, Conversion +Date and Likely Impact on Equity +Commodity Price Risk or Foreign Exchange +Risk and Hedging Activities + +Credit Ratings + +Plant Locations + +All our equity shares carry voting rights on a pari-passu basis. +99.85% of outstanding equity shares have been dematerialized as at March 31, 2021. + +The Company has 2.61% of outstanding ADRs as on March 31, 2021. + +The Company had no exposure to commodity and commodity risks for the financial year +2020-21. For Foreign exchange risk and hedging activities, please refer Management +Discussion and Analysis Report for details. +The ICRA Committee of ICRA has reaffirmed the long-term rating for lines of credit of Wipro +Limited at [ICRA]AAA. The Outlook on the long-term rating is stable. The Rating Committee +of ICRA has also re-affirmed the short-term rating at [ICRA]A+. Fitch Ratings has assigned +Long-Term Foreign- and Local-Currency Issuer Default Ratings (IDR) and foreign-currency +senior unsecured rating of ‘A-’. Standard & Poor has also assigned a rating of A-. The Outlook +is Stable. +The Company has various offices in India and abroad. Details of these locations as on March +31, 2021 are available on our website www.wipro.com. + +Registrar and Transfer Agents +Company’s share transfer and related activities are +operated through its Registrar and Share Transfer Agents: +KFin Technologies Private Limited, Hyderabad. + +Share Transfer System +In accordance with the proviso to Regulation 40(1) of the +Listing Regulations, effective from April 1, 2019, transfers +of shares of the Company shall not be processed unless the +shares are held in the dematerialized form with a depository. +Accordingly, shareholders holding equity shares in physical +form are urged to have their shares dematerialized so as to +be able to freely transfer them. + +Investor Queries and Grievances Redressal +Shareholders may write either to the Company or the +Registrar and Transfer Agents for redressal of queries and +grievances. The address and contact details of the concerned +officials are given below. + +Overseas Depository for ADSs - J.P. Morgan Chase Bank N.A. +383 Madison Avenue, Floor 11 New York, NY10179 +General: +1 800 990 1135 +From outside the U.S.: +1 651 453 2128 +Tel: +1 212 552 8926 New York +E-mail: drx_depo@jpmorgan.com + +Indian Custodian for ADSs +India Sub Custody +Office Address: J.P. Morgan Chase Bank, N.A. Mumbai +Branch, Paradigm B-Wing, 6th Floor, Mindspace, Malad (W), +Mumbai - 400 064 +Phone: +91 022 6649 2515 | F: +91 022 6649 2509 +The e-mail address and contact details for all service related +queries is: india.custody.client.service@jpmorgan.com + +Contact Persons: +Rohit Keer- E-mail id: rohit.a.keer@jpmchase.com, +Nekzad Behramkamdin- E-mail id: nekzad.behramkamdin@ +jpmorgan.com +Nayan Vyas- E-mail id: nayan.x.vyas@jpmorgan.com + +Registrar and Share Transfer Agents +KFin Technologies Private Limited, Unit: Wipro Limited, +Selenium Tower B, Plot 31 & 32, Financial District, +Nanakramguda, Serilingampally Mandal, +Hyderabad - 500 032, Telangana. +Toll Free No.: 1800 3454 001 Phone: (040) 7961 1000 + +Contact Person: +Mr. B. Srinivas- e-mail id: srinivas.b@kfintech.com +Ms. Rajitha Cholleti- e-mail id: rajitha.cholleti@kfintech.com +Shareholders Grievance can also be sent through e-mail to the +following designated e-mail id: einward.ris@kfintech.com. + +Web-Based Query Redressal System +Members may utilize this facility extended by the Registrar & +Transfer Agents for redressal of their queries. + +Please visit https://karisma.kfintech.com and click on +“investors” option for query registration through free identity +registration to log on. Investor can submit the query in +the “QUERIES” option provided on the website, which will +generate the grievance registration number. For accessing +the status/response to your query, please use the same +number at the option “VIEW REPLY” after 24 hours. The +investors can continue to put additional queries relating to +the case till they are satisfied. + +140 + + Shareholders can also send their correspondence to the Company with respect to their shares, dividend, request for annual +reports and shareholder grievances. The contact details are provided below: + +Mr. M. Sanaulla Khan +Vice President and Company Secretary +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 +Mr. G. Kothandaraman +General Manager, Finance +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Ph: +91 80 28440011 (Extn: 226185) +Fax: +91 080 28440054 +E-mail: sanaulla.khan@wipro.com +Ph: +91 80 28440011 (Extn: 226183) +Fax: +91 080 28440054 +E-mail: kothandaraman.gopal@wipro.com + +Analysts can reach our Investor Relations Team for any queries and clarification on Financial/Investor Relations related +matters: + +Ms. Aparna C. Iyer +Vice President, Finance +Corporate Treasurer and Investor Relations +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Mr. Abhishek Kumar Jain +General Manager +Investor Relations +Wipro Limited, Doddakannelli, Sarjapur Road, Bengaluru - 560 035 + +Ph: +91 80 28440011 (Extn: 226186) +Fax: +91 80 28440054 +E-mail: iyer.aparna@wipro.com + +Ph: +91-80-6142 6143 +Fax: +91 80 28440054 +E-mail: abhishekkumar.jain@wipro.com + +Due to the ongoing COVID-19 pandemic, majority of our staff is working from home. In case of any queries, stakeholders are +requested to write to the above mentioned Email IDs for a quicker response. + +141 + + Corporate Governance Compliance Certificate + +Corporate Identity Number: L32102KA1945PLC020800 +Nominal Capital: ` 2527.40 crores + +To the Members of +Wipro Limited +Doddakannelli, Sarjapur Road, +Bengaluru – 560035 + +We have examined all the relevant records of Wipro Limited (“Company”) for the purpose of certifying compliance of the +conditions of the Corporate Governance under the SEBI (Listing Obligations and Disclosure Requirements) Regulations, 2015 +for the financial year ended March 31, 2021. We have obtained all the information and explanations which to the best of our +knowledge and belief were necessary for the purposes of certification. + +The compliance of conditions of corporate governance is the responsibility of the Management. Our examination was limited +to the procedure and implementation process adopted by the Company for ensuring the compliance of the conditions of the +corporate governance. + +This certificate is neither an assurance as to the future viability of the Company nor of the efficacy or effectiveness with +which the management has conducted the affairs of the Company. + +In our opinion and to the best of our information and according to the explanations and information furnished to us, we certify +that the Company has complied with all the mandatory requirements of Corporate Governance as stipulated in Schedule II of +the said Regulations. As regards Discretionary Requirements specified in Part E of Schedule II of the SEBI (Listing Obligations +and Disclosure Requirements) Regulations, 2015, the Company has complied with items C and E. + +For V. SREEDHARAN & ASSOCIATES +Company Secretaries + +(V. Sreedharan) +Partner +FCS: 2347; CP No. 833 + +Bengaluru +June 9, 2021 + +UDIN: F002347C000436627 + +NOTE: Due to the ongoing Covid-19 pandemic, we have conducted online verification and examination of records, as facilitated by the +Company for the purpose of issuing this Certificate. + +142 + + CERTIFICATE OF NON-DISQUALIFICATION OF DIRECTORS +[Pursuant to Regulation 34(3) and Schedule V Para C Clause (10)(i) of the SEBI (Listing +Obligations and Disclosure Requirements) Regulations, 2015] + +To, +The Members of +WIPRO LIMITED +Doddakannelli, Sarjapur Road, +Bengaluru- 560035 + +We have examined the relevant registers, records, forms, returns and disclosures received from the Directors of +WIPRO LIMITED, having CIN L32102KA1945PLC020800 and having registered office at Doddakannelli, Sarjapur Road, +Bengaluru- 560035 (hereinafter referred to as ‘the Company’), produced before us by the Company for the purpose of issuing +this Certificate, in accordance with Regulation 34(3) read with Schedule V Para-C Sub-clause 10(i) of the Securities and +Exchange Board of India (Listing Obligations and Disclosure Requirements) Regulations, 2015. + +In our opinion and to the best of our information and according to the verifications (including Directors Identification Number +(DIN) status at the portal www.mca.gov.in) as considered necessary and explanations furnished to us by the Company and +its officers, we hereby certify that none of the Directors on the Board of the Company as stated below for the Financial Year +ending on March 31, 2021 have been debarred or disqualified from being appointed or continuing as Directors of companies +by the Securities and Exchange Board of India (SEBI) and Ministry of Corporate Affairs (MCA), or any such other Statutory +Authority. + +Details of Directors: + +Sl. No. +1. +2. +3. +4. +5. +6. +7. +8. +9. + +Name of Director + +Mr. Azim Premji Hasham +Mr. Mahendra Kumar Sharma +Mr. William Arthur Owens +Mr. Rishad Premji Azim +Mrs. Ireena Vittal +Mr. Patrick John Ennis +Mr. Patrick Lucien Andre Dupuis +Mr. Deepak Madhav Satwalekar +Mr. Thierry Delaporte + +DIN +00234280 +00327684 +00422976 +02983899 +05195656 +07463299 +07480046 +00009627 +08107242 + +Date of appointment in the +Company +01/09/1968 +01/07/2011 +01/07/2006 +01/05/2015 +01/10/2013 +01/04/2016 +01/04/2016 +01/07/2020 +06/07/2020 + +Ensuring the eligibility for the appointment/continuity of every Director on the Board is the responsibility of the management +of the Company. Our responsibility is to express an opinion based on our verification. This certificate is neither an assurance +as to the future viability of the Company nor of the efficiency or effectiveness with which the management has conducted the +affairs of the Company. + +For V. SREEDHARAN & ASSOCIATES +Company Secretaries + +(V. Sreedharan) +Partner +FCS: 2347; CP No. 833 + +Bengaluru +June 09, 2021 + +UDIN: F002347C000436638 + +NOTE: Due to the ongoing Covid-19 pandemic, we have conducted online verification and examination of records, as facilitated by the +Company for the purpose of issuing this Certificate. + +143 + diff --git a/q126/random_k4/random_3.md b/q126/random_k4/random_3.md index 8ca690a89c7b73d0943de9dc64d3de810d4e2fe2..2e9f5241e58113455522088c7a6110de8838d130 100644 --- a/q126/random_k4/random_3.md +++ b/q126/random_k4/random_3.md @@ -1,891 +1,1208 @@ -A Practical Guide - -Graduate Studies: A Practical Guide - - November 2012 - -Note: This Guide was published by CAGS in 1992. The booklet is no longer in print. Although -there have been changes in some aspects of graduate studies since 1992, much of this Guide is -still relevant. CAGS is still making this document available in web format as a service to -students, faculty and others involved in graduate programs. - -Copyright © Canadian Association for Graduate Studies, 1992 - -The 2003 version was edited by Martha Crago -The 2012 version was edited by Samantha Boland - - Forward - -This booklet is addressed to students considering Masters and Doctoral programs at Canadian -universities. It discusses some of the motives for pursuing graduate studies and provides -prospective students with information on things to consider when examining the option of -graduate school. There is a tremendous diversity to graduate programs and thus some sections -may pertain more to some areas of study than others. - -From the perspective of students, graduate studies are not an end in themselves, but rather a -means to an end. In most cases, that end is gainful employment. Graduate studies therefore -constitute an investment of time and money for the purpose of career development. In some -cases the investment is essential, in others it is a matter of choice. From the perspective of -supervisors, graduate studies are an almost indispensable part of academic life and sometimes -central to successful research. Graduate students are needed for assistance, collaboration, and as -a source of inspiration, energy and fulfillment. They are also important partners for obtaining -external research funding from provincial and federal granting councils. From the perspective of -departments, research centers, and other units or combination of units offering graduate -programs, graduate studies are an essential component of research. What distinguishes -universities from other academic institutions is their mandate to carry out research and train -researchers. - -Graduate programs maintain and enhance the academic standing of a university and have a -significant bearing on the public perception of the quality of the graduates of that university. As -is the case in most partnerships, the probability of success in graduate studies is enhanced by a -set of explicit checks and balances which articulate the respective rights and responsibilities of -the partners. - -Graduate school can be, and often is, one of the highlights in a student’s life. Students are -provided with the opportunity to work in exciting environments and share experiences with like -minded colleagues that can result in lifelong friendships. Many former graduate students look -back on this time in their life with fondness and the realization that it represented an important -period of personal development that laid the foundation for a rewarding career. - -2 - - Table of Contents - -Forward ........................................................................................................................................... 2 -Why Pursue Graduate Studies? ....................................................................................................... 4 -Skills Required Before Entering Graduate Programs ................................................................. 4 -Representative Skills Acquired in Graduate Programs ............................................................... 4 -What is Graduate Education? ......................................................................................................... 5 -Preparation for Graduate School ..................................................................................................... 6 -Making the Decision ....................................................................................................................... 7 -Making Contacts ......................................................................................................................... 7 -Choosing an Institution ............................................................................................................... 7 -Choosing a Supervisor ................................................................................................................ 8 -Financing a Graduate Education ..................................................................................................... 9 -Funding Sources........................................................................................................................ 10 -Moving Along ............................................................................................................................... 11 -Acquiring Formal Training and Certifications ......................................................................... 11 -Choosing a Research Topic ...................................................................................................... 11 -Setting Milestones ..................................................................................................................... 12 -Finishing Up ................................................................................................................................. 13 -Analysis, Synthesis, and Writing .............................................................................................. 13 -Presenting and Publishing ......................................................................................................... 14 -A Note about Intellectual Property Rights ................................................................................ 14 -Additional Living Considerations ................................................................................................. 15 -Living Expenses ........................................................................................................................ 15 -Housing ..................................................................................................................................... 15 -Parental leave ............................................................................................................................ 15 -Childcare ................................................................................................................................... 16 -Questionnaires .............................................................................................................................. 16 -For Students .............................................................................................................................. 16 -For Supervisors ......................................................................................................................... 17 -For Departments........................................................................................................................ 17 - -3 - - Why Pursue Graduate Studies? - -Many students pursue graduate studies for the love of learning and discovery. For others, cultural -motivation and family traditions have an unquestionably positive influence on the decision to -enter graduate studies. Some students pursue graduate studies because their chosen profession -requires a graduate degree while others are looking to change career paths or better position -themselves for advancement opportunities. - -In research programs, graduate students are involved in both the development and the -responsible conduct of original, important research and scholarship. As such, graduate students -should be excited by carrying out in-depth and detailed studies in the spirit of creative and -imaginative inquiry. Graduate students and their work are an important part of an ongoing -research process that helps us to better comprehend the human and natural world in which we -exist. This research provides the human community with ways of understanding natural, cultural, -imaginative, social and technological phenomena and investigating problems through the pursuit -of knowledge. Graduate students are thus engaged not only in a social process that provides -society with new ways of looking at the world’s complexity, problems and beauty, but in a -personal quest for bettering their lives or expanding their learning and insight. - -Recently, people with graduate degrees have been referred to as “highly qualified personnel”. In -today’s knowledge economy this level of qualification is a valuable asset for an increasing -number of jobs. According to the 2007 National Graduates Survey (Class of 2005), 93% of -graduates of master programs were employed at the time of survey. As one might expect, -earnings increase with education levels, with the largest earning premiums between the bachelor -and master levels (Bayard & Greenlee, 2009). - -Skills Required Before Entering Graduate Programs - - High academic standing - Strong commitment to pursuing rigorous research training in a selected subject area - Enthusiasm and a high degree of interest in learning - Curiosity, and an open and enquiring mind - Sound work ethic, integrity and moral standards - Collegiality - Perseverance and patience - Maturity and reliability - -Representative Skills Acquired in Graduate Programs - - Excellent critical thinking skills - Ability to integrate data and information from multiple sources, and to develop and test - -hypotheses rigorously - - Excellent oral and written communication skills - -4 - -  Skills in a range of analytical techniques using sophisticated instrumentation - Ability to work with equipment and instruments at tasks requiring precision - Ability to coordinate or co-supervise the work of others - Ability to identify problems and to develop and implement innovative solutions - Ability to work independently and in teams - -(Used with permission from the Schulich School of Medicine & Dentistry, The University of -Western Ontario [UWO], 2010, p. 3) - -What is Graduate Education? - -Graduate school includes Masters Programs (typically 1-2 years of full time study following an -undergraduate degree) and Doctoral/Ph.D. programs (approximately 4-6 additional years of -study which may follow a Masters Degree). - -The graduate learning environment is distinctly different from that at the undergraduate level. It -has been characterized as advanced, focused and scholarly in nature: - - - - - -It is advanced because it builds upon an undergraduate education. -It is focused because the emphasis in graduate studies is on depth. -It is scholarly because it is concerned not simply with the acquisition of knowledge and -skills, but with the critical analysis of existing knowledge and the creation of new -knowledge. Graduate students are expected to acquire and apply advanced analytical and -interpretive skills, as well understanding and/or producing research. - -Although the expectations of students enrolled in graduate programs may vary, it should be -recognized that graduate school represents a very different educational experience than that of -undergraduate studies. Scientific discovery is seldom a linear process, and it is generally not -possible to drive research directly towards a desired outcome. Discovery is a mixture of insight, -effort, curiosity, and good fortune (UWO, 2010, p. 5). - -Graduate programs may be separated into three broad categories: - - Professional programs – provide specialized skills and qualifications for a specific - -profession (i.e., Speech-Language Pathology, MBA programs). - - Terminal programs – Masters and Doctoral degrees are achieved separately. Students - -apply to Doctoral programs after first completing their Masters degree. - - Non-terminal programs – Masters program feeds directly into Doctoral program - -without option of stopping after Masters. May be entered into directly from -undergraduate studies. - -Additionally, your degree may be: - - Course-based – a combination of required courses, practicum placement, qualifying - -exam, independent research, or capstone course or project. - -5 - -  Thesis-based – a combination of required courses and a thesis project under the - -supervision of a thesis advisor. (UTM Career Centre [UTM], 2012, p. 1) - -Thesis-based master’s programs provide an opportunity for students to explore the possibility of -pursuing research as a career. The normal time to completion of this degree is 2 years, although -completion of a high quality thesis may require a longer period of time. Students are expected to -display enthusiasm and dedication towards completing the objectives that define their research -projects, which can mean working on their projects beyond a standard work week. In research, -the end result is nearly always determined by the effort and commitment of the student to the -project. The reward for this is the opportunity to pursue research questions and lead discovery in -diverse and important areas that are directed at reducing human suffering and improving our -understanding of the world. - -Preparation for Graduate School - -The process of deciding to enrol in a graduate program and the development of a plan of action -often begins well before the completion of a bachelor's degree. In disciplines where the -opportunity exits, students sometimes secure summer jobs in research environments to get a -sense of what research is like. They also learn through reading and conversation about the -research community and the career opportunities in their field of interest. - -Some students visit the appropriate departments in several universities and gain an insight into -the existing curriculum, programs, infra-structure and support system for graduate studies. -Sometimes they also assess through discussions with potential supervisors the type of research -project for which supervision and financial support are available and identify research topics in -which they are interested and for which they are qualified. - -Although standards and criteria vary across programs, one common factor is that the competition -is intense. Factors that contribute to admission decisions include: - - GPA – most schools look for a minimum average of B+ during the last two years of - -undergraduate study - - Suitability for the program – determined by the quality of your application, your - -research interests and your compatibility with the program/faculty research interests - Skills and experience – your research experience, volunteer/work experiences and - -extracurricular activities - - Reference letters – the type of referee and relevance of their comments are weighed - -more heavily than the number of reference letters you submit - - Admission test scores – if required (GRE www.gre.org , GMAT www.mba.com/mba) - -Prepare for graduate school during your undergraduate studies by: - - Strengthening your GPA – Although many programs look primarily at your final two - -years, a GPA that is consistently high will impress admission committees - -6 - -  Exploring your options – Take a variety of courses to help determine your research -interests, talk to Teaching Assistant about their experiences, take part in networking -activities, and talk to your professors about graduate programs and their areas of research - - Identifying references – Graduate applications require between 2-3 references. Get to - -know your professors by taking more than one class with those whose work interests you, -participate in class discussions, attend office hours, and join clubs that allow professors to -get to know you outside of the classroom - - Building experience – Gain research experience through part-time or summer jobs, -senior level thesis courses, faculty research projects, undergraduate research awards, -internships, or volunteering with a professor. Experience as a teaching assistant would be -an asset. (UTM, 2012, p. 2) - -Making the Decision - -As prospective graduate students, you are important potential members of a university’s research -and learning community. The orientation and guidelines that this document provides are -designed to help you approach your future studies as a confident and informed person. Even -though you may feel that the process of admission to graduate studies is one in which you are -being evaluated by a university, you also have just as significant a role in evaluating your -prospective program of study. It is important that you make sure that you have the information -you need to make a good decision. - -Making Contacts - -Making contact with your prospective institutions is important. This may mean: - - using websites - communicating by e-mail - -talking on the telephone - making a visit - -Do your best to talk to prospective supervisors or heads of graduate admissions committees as -well as a range of students including those presently enrolled in the program and those who have -already graduated. It is important to gauge your potential satisfaction with a program or a -supervisor from a student’s perspective. This type of conversation can help you know what to -expect from a supervisor or a course, or may just prove helpful in getting tips on potential living -arrangements. - -Choosing an Institution - -Students choose to pursue their studies at a particular institution for a variety of reasons. For -instance, some people choose to attend universities near to where they live. Others base their -decision on the reputation of a university. Keep in mind, though, that the research reputation of a -university does not always equate with student experiences in a program of study. Sometimes -graduate students’ experiences of the programs at a world-renowned research university may not - -7 - - mesh with the university’s high reputation. Students may rank their experiences at a lesser- -known university as being far more supportive and encouraging than the university’s research -reputation might suggest. - -In the case of a professional graduate degree, it is important to verify that the university program -is accredited by professional bodies. If this applies to your field of study, look into the -accreditation status of the institution and/or program you are planning to attend. - -Whether you are interested in a professional or a research-oriented degree, you should research -the nature of the university you are considering. Because it is ultimately a question of your fit -and prospective experience at a university, it is crucial to be as informed as possible. - -Choosing a Supervisor - -The choice of the right supervisor depends on many factors such as the track record of a -supervisor in seeing students through to a successful conclusion in a reasonable period of time. -Personality considerations also play a role. Some students require the caring attention of a -professor who may be a superb teacher but less active as a scholar, others flourish in the -competitive environment where the star of the research team is an internationally recognized -researcher. The relationship between student and supervisor should be based on complementary -objectives and qualities. Thus students should not become totally dependent on their supervisor. -They should try to maintain a collegial working relationship with all members of the department. -In turn, departments should remember that students require continuing departmental support for -their successful and timely graduation. Thinking about appropriate supervisors and then -approaching them will be one of the most important things you need to do when applying to a -graduate research program. - -The role of the supervisor will differ depending on your area of study. In certain disciplines, the -choice of the type of research project for which supervision and financial support are available -will depend quite directly on the research work the supervisor is doing. However, in the social -sciences and humanities, research topics often originate from the students themselves. In either -case, you will want to identify research topics in which potential supervisors are interested and -knowledgeable. - -Determining whether a supervisor is an active and productive scholar or researcher is an -important factor. Many professors maintain a website which contains their pedagogical and -publication history. This may be more detailed and useful than a database search because it will -reflect conference participation, courses taught and present research projects as well. Exploring -the professor’s website or the department site of the university with which your potential -supervisor is affiliated will often give you access to valuable information about the professor’s -research and interests. If you are approaching a potential supervisor, knowledge of their own -research is invaluable. Become familiar with this professor’s publications and be able to -knowledgeably discuss your mutual interests. Potential graduate students should ask the -following questions about the supervisors they are considering: - -8 - -  Do they regularly publish journal articles? In which journals? On what subject? - What books have they written? - Do they encourage publications by their students, either jointly or as single authored - -publications? - -Potential supervisors have track records not only as scholars, but also as supervisors. This is also -very important in determining whether they are someone you want to work with. Assessing the -combination of a potential supervisor's scholarship and supervisory skills will give you a good -idea of what your working relationship with them might be like. However, it is just as important -to try to get a sense of the person's character and attitude towards supervision in a more personal -way. Try to talk to them over e-mail or telephone, or ideally, in person. You will be working -closely with this person for a number of years, so something as basic as how well you get along -will be surprisingly important. Since the issue in supervision is good “fit,” an honest self- -assessment of what you are looking for and what works best for you is also in order. - -At the beginning of the study term, supervisors and students should discuss such things as: - -timelines for completion -forming a Supervisory Committee - - - - expected work hours, vacation times, research agendas, etc. - scheduling regular meetings between supervisor and student - attending departmental seminars, guest lecturers, etc. - attending academic conferences - policies on plagiarism - policies about authorship on papers and intellectual property - -The CAGS has developed a Letter of Understanding template that may assist you in these -discussions. - -Financing a Graduate Education - -Most graduate students require financial assistance during their studies. Student financial -assistance is available in various forms, including government financial aid (student loans and/or -bursaries), and privately funded bursary and scholarship programs, as well as student assistance -programs (tuition payments deferrals, emergency funding, food vouchers, etc.) and on-campus -work opportunities through work-study programs. Students may also wish to take on teaching or -research assistance positions. - -Federal agencies such as the Canadian Institutes for Health Research, the Natural Sciences and -Engineering Research Council and the Social Sciences and Humanities Research Council as well -as provincial agencies are among the primary sources of scholarships and fellowships for these -students. Scholarships and fellowships are always tenable for limited periods of time. NSERC, -for example, provides a maximum of four years of combined support at the master's and doctoral -levels. The maximum funding for a master's program is two years. - -9 - - Business and industry often offer scholarships for study in designated areas. In addition, most -graduate programs provide teaching assistantships from university operating funds and research -fellowships or stipends from faculty research grants and contracts. Most Canadian universities -maintain a database or records of funding sources and publish selective listings in their graduate -calendar. - -Many (if not all) Canadian universities have a financial aid office to assist students and -prospective students in seeking and securing financial assistance to enable them to pursue their -education. The Government of Canada also has a comprehensive website explaining the -eligibility and application process for government student loans, which students and prospective -students may find useful. - -At the time of admission into a graduate program, it is essential that students, supervisors, and -program administrators have a clear understanding of the financial support obtained, for what -purposes, over what period of time, and in return for what services. The need for financial -stability in graduate school cannot be overemphasized. Many academically able students have -failed because they underestimated the financial drain on their resources and because they had -depleted all available sources of support before their studies were complete. It is important that -students measure their personal financial needs realistically against these available funds. - -Funding Sources - -In addition to those listed here, many institutions offer internal awards. Although this list is not -exhaustive, here are some of the most common external granting agencies: - -Association of Universities and Colleges of Canada -– Higher Education Scholarships -Canadian Bureau for International Education - -Canadian Cancer Society - -Canadian Institutes of Health Research (CIHR) -Fondation Baxter & Alma Ricard -Fonds de la recherche en santé du Québec -Fonds québécois de la recherche sur la nature et les -technologies -Fonds québécois de la recherche sur la société et -culture -Heart and Stroke Foundation of Canada -IODE Canada War Memorial Doctoral Scholarship - -John and Lois Lamont Graduate Scholarship - -https://juno.aucc.ca/wes/?pg=834 - -www.cbie-bcei.ca/education- -management-services/alphabetical-list- -of-programs -www.cancer.ca/rese -arch/Grants%20and%20Awards/Ongoin -g%20funding.aspx?sc_lang=en -www.cihr-irsc.gc.ca -www.fondationricard.com -www.frsq.gouv.qc.ca -www.fqrnt.gouv.qc.ca - -www.frqsc.gouv.qc.ca - -www.hsf.ca/research/en/home -www.iode.ca/War-Memorial-Doctoral- -Scholarship.aspx -http://www.cfsh.ca/What-We- -Do/Scholarships_and_Awards/John-and- -Lois-Lamont-Graduate-Scholarship.aspx - -10 - - Mitacs - -www.mitacs.ca/funding-opportunities - -Natural Sciences and Engineering Research Council -of Canada -Ontario Graduate Scholarships (OGS) / Ontario -Graduate Scholarships in Science and Technology -(OGSST) -Social Sciences and Humanities Research Council -(SSHRC) -The International Development Research Centre - -Trudeau Scholarships - -Vanier Canada Graduate Scholarships (CIHR, -NSERC, SSHRC) - -www.nserc-crsng.gc.ca - -https://osap.gov.on.ca/OSAPPortal/en/A --ZListofAid/PRD1346626.html - -www.sshrc-crsh.gc.ca/Default.aspx - -www.crdi.ca/EN/Funding/Pages/default. -aspx -http://www.trudeaufounda -tion.ca/en/programs/scholarships/applica -tions -www.vanier.gc.ca/eng/home- -accueil.aspx - -Acquiring Formal Training and Certifications - -Moving Along - -Much of the formal training in research-oriented graduate programs is received during the first -year in the program. This training varies with the nature of the discipline and the department. In -many programs it is customary to provide graduate courses and to examine the students in these -courses, usually by written tests or papers. In other programs, the course work is augmented by -directed reading or individual study accompanied by periodic evaluation. However, it is -generally agreed that the first year should include an introduction to research and should not be -entirely spent on formal teaching. Students should begin to participate in research seminars and -attend conferences in their discipline as soon as possible. - -For research involving human participants, animals or radioactive, biological or hazardous -materials and lasers, certification normally must be obtained prior to starting the research. Often, -grant funds cannot be released until the relevant certification has been obtained. - -Choosing a Research Topic - -Once a student is integrated into a program and a faculty member has agreed to be a supervisor, -the next matter is the choice of the student's research topic. At the master's level, the topic should -be feasible in a predictable period of time and its scope should be reasonable. In departments -without doctoral programs the temptation sometimes exists to expand master's theses into mini- -doctoral theses, usually at a cost to the students which outweighs the associated benefit. At the -doctoral level, the choice of research topic often determines the career path of a student and -therefore requires thoughtful planning and expert advice. - -11 - - In some programs, a student may become involved in one of the ongoing lines of research within -the department. There is a growing tendency in Canadian universities, encouraged by the federal -and provincial research councils, to pursue collaborative research. Graduate students can be the -beneficiaries of this trend. The available research topics in such cases are partially predetermined -by the existing departmental research programs. The success rate of students in such programs is -reasonably high because the research is directly linked to the milestones of the programs. - -In other programs, the choice of a research topic may depend more on the advice of a specific -supervisor who may have a general idea of several research possibilities. In deciding on a -particular research topic, the supervisor should be confident that given hard work and reasonable -ability, it is fairly certain that the student will bring the project to a satisfactory conclusion in a -predictable period of time. On the other hand, an exciting line of research occasionally presents -itself in which the outcome is more uncertain. In these situations the student and the supervisor -should have a fallback position in case unexpected difficulties arise. - -Setting Milestones - -After having chosen a research topic, students should, with the help of their supervisor, make a -realistic plan for their research and thesis work which takes into account the departmental and -university rules, as well as their own constraints and outside responsibilities. This plan should be -endorsed by the supervisor and the departmental graduate studies committee. The CAGS has -developed a Letter of Understanding template that may assist you in these discussions. - -The plan should establish in writing appropriate milestones which the student should reach at -various times. It should be filed with the director of the program and form the basis for regular -assessments of progress involving the student, the supervisor, and the department. Students -should keep systematic records which enable them to take note of the pace of their progress and -of their research activities and related events in an objective, quantifiable way. - -Everything tends to take much longer than inexperienced students expect. With the help of their -supervisors, students need to plan their time carefully if they are to complete their graduate -programs within an agreed period of time. The start of a research project is nearly always slower -than students expect. Another common cause of delay is the fact that sometimes the students, or -the supervisors, are never satisfied with the accomplished research. They can always think of -ways of improving the obtained results. In short, they cannot bring a project to a conclusion. -Perfectionism can be a virtue, but students should write up what they have already achieved. -With a draft, they would almost certainly see more clearly whether any improvement was -actually necessary. - -Not infrequently it happens that the collected information or obtained results are inadequate to -support a thesis. This inadequacy is often not recognized until the thesis is actually written. -Students then have to gather additional information, perform further experiments or calculations, -or do more research, usually resulting in a delay of at least six to twelve months. If the -accumulation of information depends on seasonal or other unpredictable factors, the delay may -be even longer. Essentially this outcome is due to a lack of planning. The role of a department in - -12 - - monitoring research progress is particularly pertinent to minimizing the risks involved in this -kind of research. - -For doctoral students, there exists a pivotal milestone which determines whether they should -continue in a program in the form of their admission to candidacy. This process usually consists -of written and oral comprehensive examinations evaluated jointly by several faculty members in -addition to the supervisor. The precise nature of the examinations differs considerably from -university to university and from discipline to discipline. In some cases the examinations are of a -diagnostic nature and serve indeed as the mechanism used to decide whether a student continues -in a program. In other cases the examinations are summative and serve to flag gaps in a student's -knowledge which need to be filled. In both cases, however, the examinations can be a traumatic -experience and students often over-prepare for them. Firm departmental and faculty rules and -regulations should be in place to clarify the process. - -In a number of disciplines, several universities have introduced the fast track option that allows a -student to proceed to a doctoral program from a master's program without completing the -master's degree requirements. In addition, some universities also permit talented students in -certain disciplines to enter doctoral programs directly from bachelor's programs. In such cases, -master's programs sometimes represent a fallback option for students who cannot complete their -doctoral programs. It is then up to the graduate faculty of a university to specify the exact -conditions under which a doctoral candidate can enter a master's program in the same field, and -what requirements must be met in order to obtain a master's degree in place of a doctorate. - -Analysis, Synthesis, and Writing - -Finishing Up - -Often students define the thesis process in terms of researching, then writing. They are proud of -the fact that they have achieved their research objectives and that all that remains is to write up -the thesis. Several years later they have still not finished. The problem is that research is -generally fun and relatively easy once the project has been defined. In contrast, writing is always -demanding, and it is especially demanding when it has not been done for the entire period of -research. For this reason, students should be encouraged to combine researching and writing at -all times. It is particularly difficult to combine the writing up of a thesis with full-time -employment and students should be encouraged and given the means to complete their thesis -before leaving the university. - -Two particular points must be mentioned here. First, the thesis should be no longer than -necessary. It should demonstrate that the student understands the background to the research, -explain clearly the methods used, present the results, and discuss the findings within an -appropriate framework. Verbosity and padding should be avoided at all stages; they detract from -the issues at hand and provide easy targets for criticism by examiners. - -Secondly, unless the student is a particularly gifted or experienced writer, the use of one of the -many books that are available on good writing will likely improve the thesis. There is more in - -13 - - this than making the thesis a pleasure for the examiners to read. Unclear writing is frequently a -reflection of unclear thinking, whereas students who write clearly soon discover that a problem -of expression often arises from a lack of understanding. - -Graduate faculties and departments usually have specific guidelines and regulations governing -the preparation of theses. Before starting to write a thesis, students should therefore familiarize -themselves with the formal requirements which a thesis must meet. Most universities will accept -a thesis that is a series of articles written in the publication format for that discipline. - -One of the basic requirements of a doctoral thesis is that it be original. At the doctoral level, a -student's contribution must always be in the form of a substantial advancement of knowledge. -The student should have been trained as a scholar and thinker who is able to place this new -knowledge convincingly and eloquently in a broader conceptual context. The student should also -feel confident to guide the research of other students in the same field. - -Presenting and Publishing - -All research students, both master's and doctoral, should be encouraged by their supervisors to -make presentations, participate in conferences and submit manuscripts to journals, often starting -with research notes. The many good results of such activities include greater self-confidence, a -better perspective on the quality and meaning of their research, and more visibility in the job -market. They also help students to acquire a variety of skills which are necessary for a career in -research or in research management. Students sometimes do not realize that colleagues may be -more interested in the wider aspects of the project and its impact on the whole body of -knowledge than in their raw results. For these reasons, students should be given the opportunity, -particularly toward the end of the period of research training and while writing up their results, to -present the results of their research in a public forum, both for criticism and enrichment. - -A Note about Intellectual Property Rights - -Students must receive proper recognition by their supervisors for contributions made in -publications before and after the defence of a thesis. This is particularly crucial where a student's -research is part of a larger team project, and where related results emanating from the larger -project have already been published, either jointly by the students and their supervisors or by -other members of the team. There should be a very clear departmental policy which regulates the -respective rights of students, supervisor, and research collaborators to publish results of joint -research. - -Another issue which sometimes arises is whether a student actually has the intellectual property -right to previously published material. It may happen that the copyright to research articles -belongs to a scientific journal or publishing house and students must obtain written permission if -they wish to include reprints in a thesis and subsequently copyright the thesis. In situations where -a thesis consists entirely of previously published material or is a work of art, the regulations and - -14 - - procedures for the examination of such theses should be established in writing at the time of -admission of the student to the program. - -In the case of research of immediate or even prospective commercial value, or where the -dissemination of knowledge is being delayed because of the proprietary nature of the results, the -confidential nature of the thesis defence raises additional ethical questions which the university -should anticipate. In these and similar situations, the university regulations should also spell out -unequivocally the extent of a student's right to restrict publication of research results, while still -meeting the degree requirements. - -Questions about authorship and credit for research to be carried out should be anticipated and -settled early in the program within the context of departmental guidelines and policies with your -supervisor. In the case of collaborative and team research, it is often difficult to isolate the -original contributions of individuals from the project as a whole. In such cases it is particularly -important to be able to refer to the specifically agreed and recorded objectives which each team -member was required to attain, and to have a record of seminar and conference presentations -which document the contribution of individual researchers. - -Additional Living Considerations - -Living Expenses - -From food to accommodation, clothing and transportation, living costs will vary by location. For -instance, in Toronto, accommodation can cost between $7,300 and $14,000 per year -(http://cie.utoronto.ca/Coming/Living-Here/Cost-of-Living.htm#cost). - -Is a vehicle required? Is public transportation sufficient? These factors will also impact the -student’s budget. For a graduate student with a family, there are other expenses considerations. -The Graduate Student Association at the University of Waterloo has prepared a sample budget -for a typical graduate student with a family. - -Housing - -Finding a place to live can be both exciting and challenging for new students. Many universities -have housing geared to upper-year and graduate level students and this may be the first -consideration for students new to a city. In addition to being on-campus, residence offers other -social and academic supports to residents. University-owned apartments are also often close to -campus and can offer additional benefits to students. - -Some students choose to live off-campus. A visit to the off-campus housing office can assist -students in finding suitable housing and understanding lease agreements. Students will need to -do some legwork to determine which area to live in and whether proximity to campus is a factor -in their decision making. - -Parental leave - -15 - - A graduate student who is bearing a child or who has primary responsibility for the care of an -infant or young child immediately following a birth or adoption is eligible for parental leave. -Students who are pregnant or considering having a child or taking parental leave should review -the specific guidelines related to requests for maternity/parental leaves at his or her particular -institution. - -Childcare - -There are many childcare options and they depend on what type of care your child needs. It is -important to ask yourself questions such as: - - When will you need child care? - Do you want your child in a home-based program or child care centre? - Should you choose a program near your home or school? - -Child care is available at many institutions and each will have different care options ranging -from full-time to drop-in options. The Province of Ontario provides a good guide to choosing -the right child care for your child(ren): http://www.edu.gov.on.ca/childcare/finding.html - -Most regions have childcare subsidy programs available, and it may be worthwhile to investigate -whether your income and child care needs meet the criteria for the program. It is important to -keep in mind that there may be wait lists for funding and childcare spaces may not be -immediately available. - -Some institutions offer a child daycare subsidy for graduate students in order to help offset the -cost of childcare for graduate students. For instance, the University of Alberta offers one such -subsidy, to a maximum of $500 per child per fiscal year (to a total maximum of $2,000 in each -fiscal year). - -Questionnaires - -The purpose of this booklet was to identify key aspects of the dynamic relationship between -graduate students, supervisors and departments. The focus was on the presentation of a suitable -framework in which students and professors could interact with a good measure of predictable -success. The following questions are intended to highlight the main points of the discussion. - -For Students - - Have you assessed your financial needs during your studies and do you have a financial - contingency plan? - Do you have a written commitment from your department which specifies what financial -support you will receive during your studies, for what purposes, over what period of time, -and in return for what services? - -16 - -  Have you made a plan for your research which includes achievable milestones and a - -timetable for achieving these milestones? - - Have you made a plan for your research which includes achievable milestones and a - -timetable for achieving these milestones? - - Have you discussed your research plan with your supervisor and agreed on the milestones - -and a timetable? - - Are you keeping systematic records of your work and results, and are you able to answer - -questions on the milestones achieved? - - What progress have you made towards your degree in the last few months? - Have you written a draft of any portion of your thesis for which you have completed the -research, and are there tables, figures or other matters which you could prepare at this -time? - - Have you discussed your research results and draft with others, and do they find your - -written work clear and articulate? - - What is your timetable for completing the remaining phases of your research and for - -writing up your thesis? - - What are the remaining requirements for your degree and when do you expect to - -graduate? - -For Supervisors - - Are your students' research topics well-defined, realistic and manageable within the - -agreed timeframes? - - Have you discussed different management strategies and methodological approaches with - -your students after having decided on their research topic? - - Have you set up a schedule of regular meetings with your students in order to monitor - -their progress in relation to their research plan? - - Do you have a mechanism for refining your students' research topics in the early stages of - -their program? - - Have you provided your students with guidelines on the appropriate length of a thesis in - -your field? - - Have you provided your students with guidelines on writing up and presenting their - -research? - - Do your students keep detailed records of their work and do you check these records - -periodically to see whether they are systematic and adequate? - - Are you familiar with your students' milestones for the next few months? - Have you provided your students with the opportunity to give seminar and conference - presentations on their research? - Have you discussed with your students the question of intellectual property rights and - -credit for collaborative research? - -For Departments - -17 - -  Do you provide your students with written statements specifying what scholarships, - -fellowships, teaching assistantships or research assistantships they will receive during -their studies? - - Do you have a document describing the responsibilities of a thesis supervisor? - How do you ensure a good match between supervisors and prospective students? - Do your students have a research plan, deposited with the department, listing the - -objectives of their research program and identifying a critical path for achieving them? - - Do your students have to present reports at regular intervals which are assessed by - -members of the department other than the supervisor? - - Do your students, supervisors, and the graduate studies committee meet on a regular basis - -to assess the students' difficulties and progress? - - How do you ensure that a student's timetable for writing a thesis and completing the - -degree requirements is realistic? - - Do your students have to give seminar and conference presentations on their work? - Do you have guidelines on the respective roles of co-supervisors of a thesis? - Do you have guidelines on collaborative research? - -18 - - References - -Bayard, J., & Greenlee, E. (2009). Graduating in Canada: Profile, labour market outcomes and - -student debt of the class of 2005 (81-595-MIE2009074). Retrieved from Minister of -Industry website: http://www.statcan.gc.ca/pub/81-595-m/81-595-m2009074-eng.pdf - -Schulich School of Medicine & Dentistry, The University of Western Ontario. (2010, April). -Department of Physiology and Pharmacology Graduate Student Handbook. Retrieved -May 22, 2012 from: -http://www.uwo.ca/physpharm/forms/Graduate%20Studies%20Handbook%20Jan%2020 -11.pdf - -UTM Career Centre. (2012, September). Considering Grad School (Tip Sheet). Retrieved from - -UTM Career Centre website: -http://www.utm.utoronto.ca/careers/pdf/consideringgradschool.pdf - -University of Waterloo (2011). A Guide for Graduate Research and Supervision at the -University of Waterloo. Retrieved from: https://uwaterloo.ca/graduate- -studies/sites/ca.graduate- -studies/files/uploads/files/A%20Guide%20for%20Graduate%20Research%20and%20Su -pervision%20at%20the%20University%20of%20Waterloo%202011_2.pdf - -19 - +DUET: Joint Exploration of User–Item Profiles in Recommendation System +YueChen1,*,YifeiSun1,*,LuWang2,†,FangkaiYang2,PuZhao2,MinjieHong3,YifeiDong4,MinghuaHe2, +NanHu2,JianjinZhang2,ZhiweiDai2,YuefengZhan2,WeihaoHan2,HaoSun2, +QingweiLin2,WeiweiDeng2,FengSun2,QiZhang2,SaravanRajmohan2,DongmeiZhang2 +1PekingUniversity2Microsoft3ZhejiangUniversity4KTHRoyalInstituteofTechnology +*Equalcontribution†Correspondingauthor +Abstract +| | | | User History: | Item History: | User History: | Item History: | +| --- | --- | --- | -------------------- | ------------------------- | -------------------- | ------------------------- | +| | | | • Misfits (Punk) | • Review: | • Misfits (Punk) | • Review: | +| | | | • I r o n M a iden | " P o l i s he d P o p " | • I r o n M a iden | " P o l i s he d P o p " | +Traditionalrecommendationsystemsrepresent ( M e t al ) • R e v i e w : " G r e a t ( M e t al ) • R e v i e w : " G r e a t +6202 rpA 51 ]RI.sc[ 1v10831.4062:viXra • • Parliament Funk Bass" +| usersanditemsasdensevectorsandlearnto | | | Parliament | Funk Bass" | | | +| ------------------------------------- | --- | --- | ----------- | ------------------ | ------ | ------------------ | +| | | | (Funk) | • Meta: "80s Rock" | (Funk) | • Meta: "80s Rock" | +aligntheminasharedlatentspaceforrelevance +| | | | | Duet | Independent | | +| --- | --- | --- | --- | ----- | ------------ | --- | +estimation. RecentLLM-basedrecommenders (Joint Alignment) Generation +| instead leverage | natural-language | representa- | | | | | +| ---------------- | ---------------- | ----------- | ------------- | ------------- | ------------- | ------------- | +| | | | User Profile: | Item History: | User Profile: | Item History: | +tionsthatareeasiertointerpretandintegrate Enthusiast of A peak-era FUNK- Fan of HEAVY A highly rated POP- +| | | | FUNK and SOUL. | ROCK album with | METAL and | ROCK album with | +| --- | --- | --- | --------------- | ---------------- | ---------- | ---------------- | +withdownstreamreasoningmodules. Thispa- Values technical genre-blurring PUNK. Prefers polished +| | | | mastery | creativity. | dark themes. | production. | +| --- | --- | --- | ------- | ----------- | ------------ | ----------- | +textual +per studies how to construct effective Semantic Alignment ✘ Semantic Mismatch +profilesforusersanditems,andhowtoalign +themforrecommendation. Acentraldifficulty Figure1: DUETalignsrawuseranditemdatabytrans- +| | | | forming them | into textual | profiles within | a shared se- | +| --- | --- | --- | ------------ | ------------ | --------------- | ------------ | +isthatthebestprofileformatisnotknownapri- +manticspace. +ori: manuallydesignedtemplatescanbebrittle +| andmisalignedwithtaskobjectives. | | Moreover, | | | | | +| -------------------------------- | --- | --------- | ------------ | ------------ | -------------------- | --- | +| | | | to introduce | semantically | rich, human-readable | | +generatinguseranditemprofilesindependently +mayproducedescriptionsthatareindividually representationsforrecommendation(Wangetal., +plausible yet semantically inconsistent for a 2025;Zhang,2024;Baoetal.,2023;Hongetal., +| specificuser–itempair. | WeproposeDUET,an | | | | | | +| ---------------------- | ---------------- | --- | -------------------------- | --- | ------------------- | --- | +| | | | 2025a,b,c;Wangetal.,2024). | | Anaturaldirectionis | | +interaction-awareprofilegeneratorthatjointly +toreplacelatentvectorswithtextualuseranditem +producesuseranditemprofilesconditionedon +profilesthatcanbeinspected,edited,andreusedby +| both user history | and item evidence. | DUET | | | | | +| ----------------- | ------------------ | ---- | --- | --- | --- | --- | +downstreamcomponentsincorporatedwithLLMs. +| follows a three-stage | procedure: | it first turns | | | | | +| --------------------- | ---------- | -------------- | --- | --- | --- | --- | +However,existingLLM-basedapproachesremain +rawhistoriesandmetadataintocompactcues, +then expands these cues into paired profile limited in two important ways. First, directly +promptsandthengenerateprofiles,andfinally promptinganLLMwithrawuseranditemhisto- +optimizesthegenerationpolicywithreinforce- +riestoobtainrecommendationsoftenyieldsnoisy +mentlearningusingdownstreamrecommenda- andincompletesignals,especiallywhenhistories +| tionperformanceasfeedback. | | Experimentson | | | | | +| -------------------------- | --- | ------------- | --------- | ------------------------ | --- | ------------- | +| | | | are long, | sparse, or heterogeneous | | (Wang et al., | +threereal-worlddatasetsshowthatDUETcon- +| | | | 2025). Second, | profile-based | methods | typically | +| --- | --- | --- | -------------- | ------------- | ------- | --------- | +sistentlyoutperformsstrongbaselines,demon- +relyonmanuallydesignedtemplatesorhandcrafted +stratingthebenefitsoftemplate-freeprofileex- +attributes,whichrequiressubstantialhumanengi- +plorationandjointuser–itemtextualalignment. +Project code page: https://github.com/duet- neering and constrains the representation space. +review/duet_code. More fundamentally, many approaches generate +userprofilesanditemprofilesindependently,with- +1 Introduction +| | | | out modeling | how user | preferences | and item se- | +| --- | --- | --- | ------------ | -------- | ----------- | ------------ | +Traditional recommendation systems represent manticsinteractatdecisiontime(Yangetal.,2023; +Xietal.,2024). +usersanditemsasdensevectorsandlearntoalign +theminasharedlatentspaceforrelevanceestima- Toaddressthesechallenges,wepropose DUET, +tion (Covington et al., 2016; Wu, 2023). While ajointuser–itemprofilegeneratorthattakesboth +effective,suchembeddingsareopaque: theyoffer userhistoryanditemhistoryasinputandproduces +limitedinterpretabilityandmakeitdifficulttoan- a paired set of profiles for the interaction. Cru- +alyzewhyanitemisrecommended. Recentwork cially, DUET doesnotrequireprofiletemplates: it +thereforeleverageslargelanguagemodels(LLMs) istrainedwithreinforcementlearningusingfeed- + +back from downstream recommendation perfor- asseeninworkslikeCRESDUP(Chenetal.,2007) +mance,enablingittoexploreanddiscovereffective andUPCSim(Widiyaningtyasetal.,2021). While +profileformatsautomatically. Figure1illustrates foundational,thesemethodswerelimitedbytheir +whyjointprofilingmatters. Theuserhaslistened rigid,hand-engineeredfeatures. Morerecently,the +toMisfits(punk),IronMaiden(metal),andParlia- adventofLargeLanguageModels(LLMs)hasen- +ment(funk),whilethecandidatealbumisdescribed abledashifttowardsgeneratingprofilesinnatural +byreviewssuchas“polishedpop”and“greatfunk language. Studies such as KAR (Xi et al., 2024), +bass”andametatag“’80srock.” Consideringboth GPG(Zhang,2024),PALR(Yangetal.,2023),and +sides together, DUET can reconcile these signals LettinGo (Wang et al., 2025) leverage LLMs to +intoacompatibleinterpretation(e.g.,highlighting createtextualuserprofilesfrombehavioraldata. A +the user’s funk/soul affinity and the item’s funk- keylimitationoftheseapproachesistwofold: they +rockcharacter). Incontrast,independentlygener- relyonstaticorpre-definedtemplatesforprofile +atedprofilesmayamplifydifferentfacets,summa- generation,andtheyoftenfocusexclusivelyonuser +rizingtheuseras“heavymetal/punk”whilesum- profiles,neglectingtherich,expressiveinformation +marizingtheitemas“pop-rock”,whichresultingin inherentinitemsandthecomplexdynamicsofuser- +asemanticallymismatchedpairthatobscuresthe iteminteractions. Incontrast,ourworkintroduces +truerelevancesignal. anewparadigmwherebothuseranditemprofiles +DUETproceedsinthreestages. First,rawhisto- arenotfixedbutaredynamicallyexploredandop- +ries and metadata are distilled into minimal cues timizedinasharedsemanticspacetodirectlyalign +thatcapturecompactbutinformativesignals. Sec- withrecommendationperformance. +ond, the model expands cues into richer profile +prompts to generate textual profiles, allowing ex- 2.2 ReinforcementLearningforLLM-Based +plorationoveralternativeprofilestructuresandem- RecommendationSystems +phases. Third,theresultingprofilesareconsumed +ReinforcementLearning(RL),particularlythrough +bydownstreamrecommenders,andtheirtaskfeed- +techniqueslikeRLHF,hasbecomeacoremethod +backisusedtooptimizetheprofilegenerationpol- +for aligning LLMs with specific objectives. This +icy. By coupling both sides in a shared semantic +approach has been adapted for recommendation +space,DUETlearnsuserprofilesthatreflectwhat +tasks (Wang et al., 2025; Lin et al., 2025; Deng +kindsofitemsauserprefersanditemprofilesthat +etal.,2025),butexistingeffortsoftenfacekeylimi- +reflectwhatkindsofusersanitemappealsto. +tations. Theyfrequentlyrelyonofflinerewardmod- +Ourcontributionsareasfollows: +els(Jeongetal.,2023;Chenetal.,2025;Heetal., +• Werepresentusersanditemsasnatural-language 2025; Liu et al., 2025) that do not adapt in real- +profiles and align them in a shared semantic timetosystemfeedback,asetupthatrisksissues +space,extendingtheclassicvector-basedalign- like reward hacking (Skalse et al., 2025). Other +mentprincipletointerpretabletextualrepresenta- methods(Sunetal.,2024;Luetal.,2024)restrict +tions. themselvestoofflinepreferencetuning(e.g.,DPO), +• We introduce an exploration-based framework which can easily overfit on static datasets. Our +thatstartsfromcue-basedinitialization,expands framework, DUET,overcomesthesechallengesby +cuesintocandidateprofiles,andjointlyoptimizes integrating RL into a closed-loop system where +user and item profiles with downstream recom- downstreamrecommendationperformanceserves +mendationfeedbackviareinforcementlearning, asthereal-timerewardsignal,allowingforthedy- +avoidingrigidtemplates. namicandinteractiverefinementoftextualprofiles. +• Extensiveexperimentsacrossmultiplereal-world +datasets show that DUET consistently outper- 3 Method +formsstrongbaselines,validatingbothjointpro- +filingandfeedback-drivenprofileoptimization. DUET is a closed-loop framework that trans- +forms raw user–item interaction histories into +2 RelatedWork performance-aligned textual profiles through +learnedrepresentationstrategies. AsshowninFig- +2.1 ProfilesinRecommendation +ure 2, DUET consists of three stages. (1) Cue- +Earlyrecommendationsystemsprimarilyreliedon Based Initialization distills interaction histories +pre-definedprofilesbasedonstructuredattributes, intoconciseevidence-basedcues. (2)JointExplo- + +CUE-BASED INITIALIZATION JOINT EXPLORATION ON-POLICY OPTIMIZATION +| | | | | Item | DU | | | Duet | | | | +| ----------- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | --- | +| ItemHistory | | | | | ET | | | | | | | +Initial CUE +Rated: Baba Is You (5★) +User: “Prefers retro puzzle games” +Reviewed: “I love logic- +Item: “Retro-style indie puzzle game” +| | | | based puzzle games.” | | | Profile Prompt | | | | | | +| --- | --- | --- | -------------------- | --- | --- | -------------- | --- | --- | --- | --- | --- | +Rated: The Witness (5★) +User: “Describe the user's typical +| | | | Rated: Fez (4★) | | puzzle-game preferences … | | | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒1 𝑃𝑟𝑜𝑓𝑖𝑙𝑒2 | … | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒𝑛 | | +| --- | --- | --- | --------------- | --- | ------------------------- | --- | --- | ----------------- | --- | -------- | --- | +Item: “… item's core mechanics… +| | | | | | | | Profile | Downstream Task Env | | | | +| ----------- | --- | --- | ----------------------- | ---- | --------------------------------- | --- | ------- | --------------------- | --- | --- | --- | +| | | | | User | User: “A player who enjoys logic- | | | | | | | +| UserHistory | | | | | intensive indie puzzle games, | | | | | | | +| | | | Category: Indie Puzzle | | | | | Ranking correctness | | | | +prefers retro visual aesthetics…” +| | | | Game | | | | | &format Reward | | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Reviews: +Item: “A retro-style indie puzzle +| | | | “Challenging logic | | game featuring challenging | | | Group Computation | | | | +| --- | --- | --- | -------------------- | ------------------------ | ------------------------------- | --- | --- | ------------------- | --- | ---- | --- | +| | | | mechanics.” | | logic mechanics that appeal to | | | | | | | +| | | | “ R e t r o p ix | e l- a rt st y l e . ” | hardcore puzzle players.” | | | 𝐴𝑑𝑣1 𝐴𝑑𝑣2 | | | | +| | | | “ H a r d b u | t re w a r d i n g . ” | | | | | … | 𝐴𝑑𝑣𝑛 | | +| | | | | Figure2: | OverviewoftheDUETframework. | | | | | | | +ration via Adaptive Profile Prompt Discovery 3.2 Cue-BasedInitialization +| jointly | explores | user-item’s | | profile prompts | that | | | | | | | +| ---------- | -------- | ----------- | -------- | --------------- | -------- | -------- | --------- | -------- | -------- | --- | --------- | +| | | | | | | Raw user | histories | and item | metadata | | are often | +| define how | user | and item | profiles | should | be writ- | | | | | | | +noisy,redundant,andnotdirectlysuitableforpro- +ten. (3)OptimizationviaOn-policyExploration +| | | | | | | fileconstruction. | | Toaddressthis,DUETintroduces | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ---------------------------- | --- | --- | --- | +jointlyoptimizesuseranditemprofilesunderdown- +theconceptofcues: +concisehypothesesthatsum- +streamrecommendationfeedback. +| | | | | | | marize minimal | | but informative | | aspects | of users | +| ------------------- | ------ | --- | ------------------------- | ------- | -------- | -------------- | ------------ | --------------- | -------------- | ------- | -------- | +| All three | stages | are | realized | through | a single | | | | | | | +| | | | | | | and items. | These | cues act | as lightweight | | seeds, | +| passinputandoutput: | | | cueextraction,self-prompt | | | | | | | | | +| | | | | | | which are | deliberately | underspecified | | so | that the | +construction,andprofilegenerationareproduced +systemcansubsequentlyexplorericherprofilefor- +inasinglesequence-to-sequencegenerationpass +mats. +atinferencetime,enablingefficientdeployment. +| | | | | | | Definition1(Cue). | | Acueisaminimaltextualhy- | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------------------ | --- | --- | --- | +3.1 ProblemFormulation +pothesisderivedfromhistoricaldatathathighlights +WeformulateprofilegenerationinDUETasanon- one potential aspect of a user’s preference or an +policyreinforcementlearningproblem,motivated item’scharacteristic. Ratherthanaimingforcom- +bytheabsenceofanytextualgroundtruthdefining pleteness, cues capture partial but salient signals +an optimal user or item profile. Profile quality is thatserveasstartingpointsforprofileexploration. +evaluatedsolelybyitsfunctionalutilityinafixed +| | | | | | | To extract | cues | automatically, | | the | LLM is | +| --- | --- | --- | --- | --- | --- | ---------- | ---- | -------------- | --- | --- | ------ | +recommendationenvironment. +| | | | | | | prompted | to summarize | minimal | | but informative | | +| ---- | ---------- | --- | ------------ | ------ | ------- | ------------------------ | ------------ | ------- | ----------------- | --------------- | --- | +| DUET | is modeled | as | a generative | policy | π θ in- | | | | | | | +| | | | | | | aspectsofuseroritemdata. | | | Forexample,givena | | | +teractingwithafrozendownstreamrecommender. +user’sinteractionhistory,themodelisguidedwith +| For each | user–item | pair, | the | state is | defined as | | | | | | | +| -------- | --------- | ----- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | +instructionssuchas: +| s = {H | ,H },whereH | | andH | denotetheuser | | | | | | | | +| ------ | ----------- | --- | ---- | ------------- | --- | --- | --- | --- | --- | --- | --- | +| | u i | | u | i | | | | | | | | +and item interaction histories. An action corre- CueExtractionPrompt +spondstoasingle-passjointgeneration +“Fromthehistorybelow,analyzetheuser’s +a = {C u ,S u ,P u ,C i ,S i ,P i }, historicalinteractionstounderstandprefer- +ences,ratingbehavior,reviewsentimentor +| where | (C ,C | ) are | cues | distilled | from his- | | | | | | | +| ----- | ----- | ----- | ---- | --------- | --------- | --- | --- | --- | --- | --- | --- | +u i +| | | | | | | anyotherdimension. | | Keepthedescription | | | | +| -------- | ------------ | ----------- | --- | ----------- | ------- | ------------------ | --- | ------------------ | --- | --- | --- | +| tory, (S | u ,S i ) are | constructed | | user-item’s | profile | | | | | | | +prompts,and(P ,P )arethefinaltextualprofiles. conciseandavoidfullsentences.” +u i +| Thepolicyπ | | (a | s)definesajointdistribution | | | | | | | | | | +| ---------- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +θ +over the entire generation sequence. As an on- This lightweight guidance allows the LLM to +policy agent, DUET is optimized using rewards maprawhistoriesandmetadataintocompacttex- +from its own sampled generations rather than by tualcues. Thedetaildexampleofcuecanbefound +| imitatingfixedsummaries. | | | | | | inAppendixA.3. | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | + +| 3.3 | JointExplorationviaAdaptiveProfile | | | | | | Single-PassInput | | | | | | +| --- | ---------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +PromptDiscovery Task:Generatestructuredprofilesfortheuseranditemfrom +historicaldata. +Aftercueextraction,DUETdoesnotdirectlysum- Combined Context: [User History H u ] + [Item +| | | | | | | | History | H i ]+[Avg | Ratings] | | | | +| ------------------------------------ | --- | --- | --- | --- | --- | ---------- | ------- | ---------- | -------- | --- | --- | --- | +| marizeuserinterestsoritemattributes. | | | | | | Instead,it | | | | | | | +“First,extractaconciseCueforbothuseranditem.Second, +| explores | | the space | of profile | construction | | strate- | | | | | | | +| -------- | --- | --------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | +basedonthecues,constructaProfilePromptthatdefines +gies—natural-languagepromptsthatdefinethefor- thedescriptionlogic.Finally,generatetheProfileguidedby +theconstructedProfilePrompt.” +mat,abstractionlevel,andattributeselectionlogic +↓DUETForwardPass↓ +| usedtogenerateprofiles. | | | | Explorationistherefore | | | | | | | | | +| -------------------------------------------- | --- | --- | --- | ---------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| performedoverhowprofilesshouldbeconstructed, | | | | | | | Single-PassOutput | | | | | | +User +ratherthanoversuperficialtextualparaphrases. +Cue:“prefersnostalgiclogic-basedchallenges” +For each user–item pair, DUET intro- Profileprompt: “Focuson1990svisualaestheticsand +| duces | an | explicit | intermediate | | variable, | the | strategicdepth.” | | | | | | +| ----- | --- | -------- | ------------ | --- | --------- | --- | ---------------- | --------- | --- | ----- | ----------- | ------------ | +| | | | | | | | Profile: | “A player | who | seeks | retro-style | visual charm | +constructed_prompt,whichservesasadiscrete +pairedwithdeepstrategicreasoning.” +| and | interpretable | | profile | prompt | S. | This profile | | | | | | | +| --- | ------------- | --- | ------- | ------ | --- | ------------ | --- | --- | --- | --- | --- | --- | +Item +promptisashortnatural-languageinstruction(e.g., +Cue:“retro-styleindiepuzzlewithhighdifficulty” +“Describetheuser’stypicalgamingpreferencesand +| | | | | | | | Profile | prompt: | “Describe | | pixel-art | graphics and | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --------- | --- | --------- | ------------ | +engagementpatterns”)thatspecifieswhataspects intellectualdifficultytomatchlogicpreference.” +Profile:“A2Dexperiencefeaturingpixelatednostalgia +| to | describe | and | how to | organize | them, | without | | | | | | | +| --- | -------- | --- | ------ | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +andchallengingmechanicsthatdemandlogicaldeduction.” +| containingtheprofilecontentitself. | | | | | | Conditioned | | | | | | | +| ---------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +onS,themodelgeneratesthefinaluseranditem Figure3: Single-passgenerationinDUET: cueextrac- +| profiles. | | | | | | | tion,profileprompt(constructedprompt),andprofile | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | --- | --- | +generationareproducedinonepassforbothuserand +| Exploration | | | is driven | by treating | | the profile | | | | | | | +| ----------- | --- | --- | --------- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +item. +| promptS | | asastochasticactionsampledfromthe | | | | | | | | | | | +| ------- | --- | --------------------------------- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| policyπ | | (S | Cue). | Duringtraining,theagentsam- | | | | | | | | | | +θ +plesdifferentprofilepromptinstructions,executes +| | | | | | | | andintellectualdifficulty”tomatch. | | | | | RLreinforces | +| ---------------------------- | --- | --- | --- | --- | ---------------- | --- | ---------------------------------- | --- | --- | --- | --- | ------------ | +| themtogenerateprofilepairs(P | | | | | ,P ),andreceives | | | | | | | | +u i +thissharedsemanticdirection,suppressingirrele- +| a reward | | based | on downstream | | recommendation | | | | | | | | +| --------- | --- | ----------------------------------- | ------------- | --- | -------------- | --- | ------------ | ------------- | -------- | --- | ----- | ---------------- | +| | | | | | | | vant signals | and | forcing | the | final | profiles to con- | +| accuracy. | | Thepolicyisoptimizedtoreinforcepro- | | | | | | | | | | | +| | | | | | | | verge | into a shared | semantic | | space | of nostalgia | +filepromptsthatconsistentlyyieldhigherrewards, +andlogic,significantlyimprovingrecommendation +enablingthemodeltoactivelysearchforeffective +accuracy. +profileconstructionformats. +Thisprocessisrealizedasaunifiedgeneration +3.4 OptimizationviaOn-policyExploration +pass, +| | | | | | | | On-policy | optimization. | | | We train | the profile | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | --- | -------- | ----------- | +O = [ Cue → profilepromptS → Profile], generator π in an on-policy manner against a +| | (cid:124)(cid:123)(cid:122)(cid:125) | | | | | (cid:124) (cid:123)(cid:122) (cid:125) | | θ | | | | | +| --- | ------------------------------------ | --- | --------- | ------------------ | --------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | (cid:124) | (cid:123)(cid:122) | (cid:125) | | | | | | | | +Context ProfileConstructionPrompt Execution frozendownstreammodelf, whichservesasthe +| | | | | | | (1) | environment | critic. | For | each | sampled | user–item | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | --- | ---- | ------- | --------- | +wheretheprofilepromptS parameterizesthepro- pair (u,i), the policy generates (P ,P ) and re- +u i +filegenerator. Atinferencetime,DUETcollapses ceivesascalarrewardbasedonthepredictionac- +thelearnedprofilepromptdistributionandgreed- curacy of f(P ,P ). The policy parameters are +u i +ilyexecutestheoptimalprofilepromptinasingle updatedtoreinforcegenerationsthatleadtolower +| forwardpass,introducingnoadditionallatency. | | | | | | | predictionerror. | | | | | | +| ------------------------------------------- | --- | ---------- | ------ | ------------ | -------- | ------- | --------------------------- | --- | --- | --- | --- | ------------- | +| Example | | of profile | prompt | Exploration. | | As | | | | | | | +| | | | | | | | Continuousfractionalreward. | | | | | Usingdiscrete | +| illustrated | | in Figure | 3, | DUET | resolves | this by | | | | | | | +integerratingsasrewardsleadstosparseandunsta- +| identifying | | a shared | context | in | a single | forward | | | | | | | +| ----------- | --- | -------- | ------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +blefeedback,asnear-misspredictions(e.g.,predict- +pass. ThepolicyextractsacoordinatedCue(e.g., +ing4foragroundtruth5)receivethesamepenalty +“prefersretropuzzlegames”)andconstructsapro- +| | | | | | | | assevereerrors. | | Toprovidedenseandinformative | | | | +| ---- | ------ | ---- | ------------------- | --- | --- | -------- | --------------- | --- | ---------------------------- | --- | --- | --- | +| file | prompt | (the | constructed_prompt) | | | that in- | | | | | | | +feedback,wedefineacontinuousfractionalreward: +structsthemodeltohighlight“nostalgicaesthetics +| and | strategic | logic” | for | the user, | while | simulta- | | | | | | | +| ------- | ---------- | ------ | ---------- | ---------- | ----- | -------- | --- | ------- | --- | ---- | ------ | -------- | +| | | | | | | | | | | | |y −yˆ | | | +| | | | | | | | | R (u,i) | | = 1− | ui | ui , (2) | +| neously | describing | | the item’s | “pixel-art | | graphics | | perf | | | | | +M + +where yˆ ui = f(P u ,P i ) is the predicted relevance ingsettingbasedondownstreampredictedrating +score and M is the maximum rating gap (e.g., scores. For each observed user–item interaction, +M = 4 for a 1–5 scale). This reward provides we randomly sample nine items that the user has +fine-grainedgradientsthatencourageincremental not interacted with to form a candidate set of ten +improvementsinrecommendationaccuracy. items, which are ranked in descending order ac- +We adopt Group Relative Policy Optimization cordingtotheirpredictedratings. Theground-truth +(GRPO) (DeepSeek-AI et al., 2025) to optimize interacted item is treated as the only positive in- +theprofilegeneratorundertheabovereward. The stance,andweadoptNDCG@Kastheevaluation +downstreamrecommenderf isfrozenthroughout metric, with K set to 1, 5, and 10, to assess how +training, which yields a stable policy optimiza- effectivelythelearnedprofilessupportcorrectitem +| tionsettingandpreventsrewarddriftorfeedback- | | | | ranking. | | | | | +| -------------------------------------------- | --- | --- | --- | --------- | ---------- | ---------- | ---- | ------- | +| | | | | Baselines | We compare | our method | with | several | +inducedrepresentationcollapse. +| | | | | representative | baselines. | 10H | directly | uses the | +| --- | --- | --- | --- | -------------- | ---------- | --- | -------- | -------- | +4 Experiment mostrecentinteractionhistoriesforpredictionwith- +| | | | | outconstructingexplicitprofiles. | | | KAR(Xietal., | | +| --- | --- | --- | --- | -------------------------------- | --- | --- | ------------ | --- | +4.1 ExperimentalSettings +2024)augmentsrecommendationmodelswithex- +Datasets. Experiments were conducted on three ternalreasoningknowledgeaboutuserpreferences +widelyusedreal-worlddatasets. AmazonMusic andfactualknowledgeaboutitemsextractedfrom +(Music) and Amazon Book (Book) are derived LLMs,whicharetransformedintotask-compatible +fromtheAmazonProductdataset1,whileYelpis representations. RLMRec(Renetal.,2024)lever- +fromtheYelpOpendataset2. Alldatasetsinclude agesLLMstolearnsemanticuseranditemrepre- +userreviews,ratings,andrichtextualinformation. sentationsfromtextualsignalsandalignsthemwith +WeusedthefullAmazonMusicdataset,butonly collaborativerelationalinformationthroughcross- +subsets (latest two months for Book and six for viewrepresentationlearning. PALR(Yangetal., +Yelp). Data was split by timestamp into training, 2023)fine-tunesalargelanguagemodelasarank- +validation,andtestsetstopreventinformationleak- ing component that selects preferred items from +age(Jietal.,2023). retrievedcandidatesexpressedinnaturallanguage. +EvaluationMetrics LG (LettinGo) (Wang et al., 2025) explores di- +For each observed user–item interaction (i.e., verseuserprofilecandidateswithLLMsandaligns +a review record), we construct the evaluation profilegenerationwithdownstreamrecommenda- +instance based on the user’s interaction history tionperformanceviapreferenceoptimization. Rea- +strictly prior to the corresponding timestamp. son4Rec(Fangetal.,2025)introducesadelibera- +tiverecommendationframeworkthatincorporates +Specifically,theuser’srecenthistoricalinteractions +beforethecurrentinteractionareusedtogenerate explicitstep-wisereasoningoveruserpreferences +auserprofile,whilehistoricalreviewsfromother toguideratingprediction. +| users are used | to generate | an item profile. | The | | | | | | +| -------------- | ----------- | ---------------- | --- | --- | --- | --- | --- | --- | +4.2 MainResults +downstreamrecommendationsystemthenpredicts +theratingofthetargetitemconditionedontwoin- Table 1 presents a comparison of our proposed +puts: thegenerateduserprofile,andthegenerated method against five baselines on three datasets: +| | | | | Amazon | Music, Amazon | Books, | and | Yelp. We | +| ------------ | ------------------------------- | --- | --- | ------ | ------------- | ------ | --- | -------- | +| itemprofile. | Anexceptionisthe10Hbaseline,for | | | | | | | | +which the downstream model directly consumes use Qwen3-8B (Team, 2025) and LLaMA3-8B +therawrecentinteractionhistories. (Dubey et al., 2024) as both the profile generator +andthedownstreamrecommendationmodel,with +| We evaluate | performance | using four | widely | | | | | | +| --------------- | ----------- | ------------------ | ------- | ------------ | ----------- | ----- | ----------- | ------- | +| | | | | a prediction | temperature | of 0. | In addition | to rat- | +| adopted metrics | (Wang | et al., 2025; Fang | et al., | | | | | | +2025): MeanAbsoluteError(MAE),RootMean ingpredictionresults,wealsoreportranking-based +SquareError(RMSE),Accuracy,andF1score. evaluation results in Table ??, which assess the +effectivenessofthelearnedprofilesfromadown- +Inadditiontoratingprediction,wefurtherevaluate +thegenerateduseranditemprofilesunderarank- streamrankingperspective. +| | | | | Overallsuperiorityoverstrongbaselines. | | | | As | +| --- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | +1https://cseweb.ucsd.edu/~jmcauley/datasets/ shown in Table 1, our method consistently out- +amazon/links.html +| | | | | performs | the strongest | baselines | across | all three | +| --- | --- | --- | --- | -------- | ------------- | --------- | ------ | --------- | +2https://business.yelp.com/data/resources/ +| open-dataset/ | | | | datasetsunderbothQwen3-8BandLLaMA3-8B. | | | | | +| ------------- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | + +Method Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) +(%) (%) (%) +Qwen3(8B) +10H 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +KAR(Xietal.,2024) 0.7396 1.2184 55.34 48.67 0.7483 1.1380 58.65 60.29 0.7098 1.0923 56.17 58.78 +RLMRec(Renetal.,2024) 0.8197 1.3312 47.15 42.46 0.7438 1.1069 54.89 57.65 0.7812 1.1584 52.86 55.93 +PALR(Yangetal.,2023) 0.7994 1.2876 48.53 43.19 0.6075 0.9531 57.35 56.77 0.7485 1.1187 54.24 56.38 +LG(Wangetal.,2025) 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +R4Rec(Fangetal.,2025) 0.7028 1.1523 55.69 47.73 0.5654 0.9635 58.69 54.67 0.6397 1.0098 58.47 56.84 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +LLaMA3(8B) +10H 1.0864 1.9532 22.09 27.30 0.7917 1.3346 38.13 46.87 0.8064 1.3866 37.15 45.27 +KAR(Xietal.,2024) 0.6427 1.1668 54.51 47.98 0.5726 0.9033 57.53 59.92 0.5892 0.9614 55.87 58.21 +RLMRec(Renetal.,2024) 0.7428 1.3572 46.74 42.11 0.6076 0.9886 53.78 57.42 0.6226 0.9477 52.12 55.79 +PALR(Yangetal.,2023) 0.7238 1.3265 47.72 43.29 0.5823 0.9222 56.73 59.31 0.5977 0.8855 55.06 57.62 +LG(Wangetal.,2025) 0.6196 1.1289 56.03 51.24 0.5204 0.9369 61.92 59.50 0.5543 0.7967 58.95 60.39 +R4Rec(Fangetal.,2025) 0.7586 1.0418 55.80 53.00 0.5442 0.7722 60.86 54.88 0.6029 0.8345 59.70 56.35 +Ours 0.5367 0.9687 60.87 54.74 0.4680 0.8277 63.30 60.60 0.5092 0.9500 63.42 58.12 +Table1: PerformanceonthreedatasetsusingQwen3(8B)andLLaMA3(8B). +Method Yelp AmazonMusic AmazonBooks +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +Table2: RankingperformanceunderEASE-based(Steck,2019)hardnegatives. +Under Qwen3-8B, our method achieves an accu- our approach reaches an NDCG@1 of 0.5619 +racyof61.23%onYelp,67.96%onAmazonMu- and an NDCG@10 of 0.7443, clearly outper- +sicand64.38%onAmazonBooks,surpassingLG forming the strongest baseline methods. Similar +(LettinGo)(Wangetal.,2025)by5.05%,5.59% trendsareobservedonAmazonMusic,whereour +and5.03%,respectively. Similarimprovementsare methodattains0.5347atNDCG@1and0.7331at +observed under LLaMA3-8B, indicating that the NDCG@10, as well as on Amazon Books with +gainsarestableacrossbackbonemodels. NDCG@1 of 0.4866 and NDCG@10 of 0.7107. +Theseresultsindicatethatthelearnedprofilesen- +Advantages over fixed-structure and delib- +able more accurate identification and ordering of +erative baselines. Compared with strong fixed- +relevantitemswithincandidatesets,leadingtosu- +structureordeliberativemethodssuchasKAR(Xi +perior ranking quality in downstream recommen- +etal.,2024)andReason4Rec(Fangetal.,2025), +dation. Wefurtherevaluaterankingperformance +our approach consistently achieves lower predic- +under a more challenging setting using EASE- +tion error and higher accuracy. For example, on +based (Steck, 2019) hard negatives, and observe +Amazon Books with Qwen3-8B, our method re- +consistentimprovements;detailedresultsarepro- +duces MAE to 0.4612, compared to 0.7098 for +videdinAppendixB.2. +KARand0.6397forReason4Rec,demonstrating +moreeffectiveabstractionofuserpreferenceswith- +4.3 AblationStudy +outrelyingonpredefinedreasoningtemplates. +Consistent improvements in downstream Effectivenessofprofilegeneration,cue&strategy, +ranking. Table??reportstherankingresultsun- anduser–itemjointoptimization. Table3reports +der the Qwen3-8B backbone. Our method con- the ablation results under different design con- +sistentlyachievesthebestperformanceacrossall figurations. Starting from the history-only base- +three datasets and evaluation cutoffs. On Yelp, line (10H), introducing explicit profile genera- + +| Method | | | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | --- | ---- | --- | ------ | ----------- | --- | --- | ----------- | ------ | +| | | | MAE | RMSE | Acc | F1 MAE | RMSE | Acc | F1 | MAE RMSE | Acc F1 | +10H(HistoryOnly) 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +Profile 0.7218 1.1863 55.48 48.09 0.6597 1.0218 58.67 57.48 0.6764 1.0469 57.14 57.68 +Profile+Cue&Strategy 0.7085 1.1654 55.83 48.54 0.5708 0.9897 58.91 55.53 0.6389 1.0108 58.43 56.88 +Profile+JointOpt.(LG(Wang 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +etal.,2025)) +Profile+Cue&Strategy+ 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +JointOpt.(Ours) +Table3: AblationstudyondifferentdesignconfigurationsinDUETusingQwen3(8B). +| Method | | Yelp | | | | AmazonMusic | | | | AmazonBooks | | +| ------ | -------- | ---- | --- | ----- | --- | ----------- | --- | ----- | --- | ----------- | --------- | +| | MAE RMSE | | Acc | F1(%) | MAE | RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | +| | | | (%) | | | | (%) | | | | (%) | +10H+30P 0.5126 0.9485 61.23 55.18 0.3883 0.7494 67.96 63.89 0.4612 0.9089 65.13 59.97 +10H+50P 0.4909 0.9207 62.43 56.24 0.3924 0.7543 67.88 63.87 0.4553 0.9023 64.62 59.52 +10H+70P 0.4987 0.9326 61.98 55.81 0.3937 0.7564 68.22 64.12 0.4608 0.9068 64.38 59.27 +Table4: Impactofhistoricalinteractionlengthonprofilequality(usingQwen3(8B)). +tionyieldssubstantialperformanceimprovements jointoptimizationarecombined. Underthisfull +across all datasets. For example, on Yelp, MAE configuration, accuracy reaches 61.23% on Yelp, +isreducedfrom1.1235to0.7218andaccuracyin- 67.96%onAmazonMusic,and64.38%onAma- +creases from 23.17% to 55.48%, confirming that zon Books, with corresponding MAE values of +textualprofilesprovidesignificantlymoreinforma- 0.5126, 0.3937, and 0.4612, respectively. Com- +tiverepresentationsthanrawinteractionhistories. pared with the LettinGo-style configuration, this +| | | | | | | setting | yields | consistent | | improvements | across all | +| ------ | ---------------- | --- | ----- | --- | ------ | --------- | ------ | -------------------------------------- | --- | ------------ | ---------- | +| Adding | the cue&strategy | | layer | on | top of | pro- | | | | | | +| | | | | | | datasets. | | Overall,theablationresultsindicatethat | | | | +filegenerationleadstomodestbutconsistentgains. +whileprofilegenerationandjointoptimization +OnAmazonMusic,MAEfurtherdecreasesfrom +0.6597to0.5708,whileaccuracyslightlyimproves contributesubstantiallytoperformancegains,inte- +gratingcue&strategydiscoveryfurtherenhances +| from58.67%to58.91%. | | | Althoughthenumerical | | | | | | | | | +| ------------------- | --- | --- | -------------------- | --- | --- | --- | ------------- | --- | -------- | --------- | ------------ | +| | | | | | | the | effectiveness | | of joint | user–item | optimization | +improvementsintroducedbystrategyalonearelim- +ited, these results suggest that strategy discovery byprovidingastructuredspaceforexplorationand +refinement. +primarilyreshapeshowpreferenceinformationis +abstractedandexpressed,ratherthandirectlyopti- Impactofhistoricalinteractionlengthonprofile +mizingpredictionaccuracyinisolation. +| | | | | | | quality. | | Table | 4 analyzes | the | impact of histori- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ----- | ---------- | --- | ------------------ | +Whenenablingjointoptimizationwithoutstrat- cal interaction length on profile quality. Across +thethreedatasets,varyingthenumberofhistorical +| egy (i.e., | the LettinGo | (Wang | et | al., 2025)-style | | | | | | | | +| ---------- | ------------ | ----- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | +configuration),performanceimprovesmorenotice- interactions leads to only moderate performance +ablyacrossdatasets. OnAmazonMusic,accuracy differences, indicating that our method does not +| | | | | | | strongly | | depend | on long | histories. | In particular, | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------ | ------- | ---------- | -------------- | +increasesto62.37%,comparedto58.67%withpro- +filegenerationalone. Inourimplementation,this usingamoderatehistorylength(e.g.,30–50inter- +settingcorrespondstoareproductionofLettinGo, actions)alreadyachievescompetitiveorbestper- +whereprofilegenerationandjointoptimizationare formanceonmostmetrics,whilefurtherincreasing +thehistorylengthprovideslimitedadditionalgains. +| appliedwithoutthestrategylayer. | | | | Whiletheorig- | | | | | | | | +| ------------------------------- | ------- | --------- | --- | ------------- | ---------- | --- | --- | --- | --- | --- | --- | +| inal method | focuses | primarily | on | user | profiling, | | | | | | | +OnYelpandAmazonBooks,extendingthehis- +weextendtheoptimizationtobothuseranditem +torylengthbeyondthisrangedoesnotconsistently +profilestoensureafaircomparison. improveaccuracyorF1scoreandmayslightlyde- +Thestrongestandmostconsistentperformance gradeperformance,suggestingthatexcessivehis- +is achieved when cue&strategy discovery and toricalinteractionscanintroducenoisyorlessrele- + +vantsignals. Incontrast,AmazonMusicexhibits maintaining comparable coverage. Compared to +relativelystableperformanceacrossdifferenthis- existing methods, which either rely on extractive +torylengths,indicatingthatuserpreferencesinthis compression(e.g.,KAR(Xietal.,2024))orgener- +domainarelesssensitivetohistorytruncation. atefree-formdescriptionswithweakergrounding +Necessity of RL-based Optimization. We com- (e.g.,RLMRec(Renetal.,2024)andPALR(Yang +pare the full model with a variant without RL etal.,2023)), DUET consistentlyattainsstronger +(DUET W/O RL) to isolate the effect of RL. As alignment without sacrificing coverage. For in- +showninTable5,removingRLleadstosubstantial stance,LG(Wangetal.,2025)achievesrelatively +performancedegradationacrossalldatasets(e.g., highalignmentbutexhibitslessstablegrounding +YelpAccuracydropsfrom61.23%to48.53%),indi- acrossdatasets,whileRLMRecattainshighercov- +catingthatthegainscannotbeattributedtoprompt erageatthecostofweakersemanticalignment. In +| designalone. | | | | | contrast,DUETmaintainsbothhighalignmentand | | | | | | +| ------------ | --- | --- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | +Without RL, the generator reduces to a static mid-to-highcoverage,suggestingamoreeffective +mapping from interaction history to textual pro- balancebetweensemanticabstractionandevidence +| files,lackingadaptiveselectionofrelevantsignals. | | | | | preservation. | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | +Incontrast,RLenablesoptimizingprofileconstruc- +4.5 CaseStudy +tionunderrewardfeedback,resultinginmoredis- +Figure4presentsarepresentativecasestudythat +| criminative | representations.These | | results | demon- | | | | | | | +| ----------- | --------------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | +strate that RL-based optimization is essential for demonstrates how semantically aligned user and +| DUET | . | | | | itemprofiles | enableaccurate | | prediction | | thattran- | +| ---- | --- | --- | --- | --- | ------------ | -------------- | --- | ---------- | --- | --------- | +scendsthelimitationsofsparserawinteractionhis- +4.4 SemanticAnalysisofGeneratedProfiles tories. In the user profile, initially scattered and +To better understand the source of performance fragmentedpreferencecuesaresystematicallydis- +gains, we analyze whether the generated profiles tilled into a stable and coherent preference struc- +exhibitmeaningfulsemanticpropertiesratherthan turefavoringfunk,soul,andprogressiverock(or- +servingasintermediatetextualartifacts. Weintro- ange), coupled with a pronounced emphasis on +duce two complementary metrics to characterize musicalcomplexityandhistoricalsignificance(pur- +semanticcompatibilityandgrounding. ple). Thecorrespondingitemprofileexhibitsstrik- +Semantic Alignment. We measure the ingsymmetry,characterizingthealbumasanexem- +embedding-levelsimilaritybetweengenerateduser plaryfunk-rockcomposition(orange)andposition- +anditemprofilesusingall-mpnet-base-v2from ingitasadefiningandculturallyinfluentialrelease +Sentence-Transformers (Reimers and Gurevych, of the 1970s (purple). This example shows 519 +2019). Foreachuser–itempair,wecomputecosine that the learned profiles capture meaningful pref- +| similarity: | | | | | erence–attributecorrespondencethatisdifficultto | | | | | | +| -------------------------------- | -------------------------------- | --- | ------------ | --- | ----------------------------------------------- | --------- | ------- | ---------- | ---- | ----------- | +| | | | e ·e | | recoverfromindividualreviewsalone. | | | | | | +| | | | u i | | | | | | | | +| | Align(u,i) | = | | (3) | | | | | | | +| | | | ∥e ∥∥e ∥ | | | | | | | | +| | | | u i | | 5 Conclusion | | | | | | +| wheree | ande denotetheembeddingvectorsof | | | | | | | | | | +| | u i | | | | In this | paper, we | propose | DUET | , a | closed-loop | +| thegenerateduseranditemprofiles. | | | Highervalues | | | | | | | | +| | | | | | framework | for | jointly | generating | user | and item | +indicatestrongersemanticcompatibilitybetween +| | | | | | textualprofilesforrecommendation. | | | | | Unlikeprior | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | +modeleduserpreferencesanditemcharacteristics. +| | | | | | methods | that rely | on | fixed templates | | or indepen- | +| -------- | --------------- | --- | ---------- | ------ | ------- | ----------- | --------- | --------------- | --- | -------------- | +| Coverage | (Faithfulness). | | We measure | token- | | | | | | | +| | | | | | dently | constructed | profiles, | DUET | | treats profile | +levelgroundingas: +| | | | | | generation | as an | exploration | problem | | and aligns | +| --- | --- | --- | --- | --- | ---------- | ----- | ----------- | ------- | --- | ---------- | +|Tokens(profile)∩Tokens(history)| representations directly with downstream recom- +Cov = +| | | |Tokens(profile)| | | | mendationperformance. | | | | | | +| --- | --- | ----------------- | --- | --- | --------------------- | --- | ---- | ---------- | --------- | ---- | +| | | | | (4) | Specifically, | | DUET | integrates | cue-based | ini- | +Thismetricquantifieshowmuchofthegenerated tialization, adaptive strategy construction, and +profileissupportedbyhistoricaltextualevidence. feedback-drivenjointoptimizationtoproduceflex- +We report coverage separately for user and item ible yet task-aligned user–item profiles. By opti- +| profiles. | | | | | mizing | both profiles | in | a shared | semantic | space, | +| --------- | --- | --- | --- | --- | ------ | ------------- | --- | -------- | -------- | ------ | +AsshowninTable6,DUETachievesthehigh- theframeworkreducessemanticmismatchandcap- +est semantic alignment across all datasets while turesinteraction-relevantsignalsthataredifficultto + +Setting Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1 MAE RMSE Acc F1 MAE RMSE Acc F1 +(%) (%) (%) (%) (%) (%) +DUETw/oRL 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +DUET(full) 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table5: EffectofRL-basedoptimization(usingQwen3(8B)). +User History: +Andre' S Grindle "Tour of U.S.A.", Rating: 5.0. If your looking for heavy metals dim orgins look no further. +Andre' S Grindle "Misfits", Rating:4.0. Actually the opening title track is more of a direct pop/rock type song. But the rock n roll +returns in a big way on "Hay Fever". +............ +Item History: +Worlds Away, Rating: 4.0, Review: The CD is wonderful, and is a seminal piece of music for both the group and for those of us who +were 'of age' in the '70s and '80s. Pablo Cruise was - and is - a group that makes beautiful music like few others, and this +shipper/seller made it all the more special with a fast ship! THANKS!!! +Worlds Away, Rating: 5.0, I love this CD! I'd recommend the Pablo Cruise Millenium Collection to get to know this band. Worlds Away to +me represents the best of breezy '70s pop. Five stars! +............ +USER PROFILE: +Andre Grindle is a dedicated music enthusiast with a deep appreciation for funk, soul, and progressive rock, particularly drawn to artists +who blend technical mastery with genre-defining innovation. His reviews consistently highlight albums that showcase musical complexity, +lyrical depth, and historical significance, often referencing artists like Stevie Wonder, Prince, and Rush. He values authenticity and +artistic vision, frequently praising albums that resist trend-chasing and prioritize creative integrity. His detailed, introspective reviews +suggest a preference for nuanced analysis over superficial praise, with a focus on emotional resonance, production quality, and the +artist's evolution. While he is generous with high ratings (often 5.0), he is also willing to critique perceived shortcomings, as seen in his +4.0 for "Test for Echo" and 3.0 for "The Color of Success." His engagement is characterized by a blend of personal anecdotes, technical +appreciation, and historical context, indicating a preference for music that transcends mere entertainment. +Item PROFILE: +"Worlds Away" by Pablo Cruise is celebrated as a peak-era funk-rock album that exemplifies the band's ability to merge polished +production with genre-blurring creativity. Users consistently praise its tight musicianship, genre-defying sound, and the band's ability to +craft catchy, emotionally resonant tracks. The album's appeal lies in its seamless fusion of rock, funk, and pop, with standout tracks like +"Love Will Find a Way" and "Worlds Away" receiving particular acclaim for their infectious melodies and lyrical depth. While some users +note its dated production compared to modern standards, the majority highlight its historical significance as a defining work of the +1970s rock-funk movement. The album resonates most with fans of 1970s progressive rock, genre-defying artists, and listeners seeking +well-crafted, emotionally engaging music. Its high average rating reflects widespread admiration for its artistic ambition and enduring +appeal, though a minority acknowledges its limitations in terms of modern production quality. +Figure4: Illustrationofthemutualcorrespondencebetweenuseranditem. Thehighlightedregionsdemonstrate +thatuserpreferencessummarizedintheuserprofilealignwiththekeyattributesextractedintheitemprofile,which +providescomplementaryinformationbeyondrawhistoriesandthusimprovespredictionaccuracy. +Method Yelp AmazonMusic AmazonBooks +Align UserCov. ItemCov. Align UserCov. ItemCov. Align UserCov. ItemCov. +10H 0.3902 /(1.00) /(1.00) 0.3577 /(1.00) /(1.00) 0.2604 /(1.00) /(1.00) +KAR(Xietal.,2024) 0.4932 0.1745 0.1823 0.4807 0.2170 0.2391 0.5702 0.1877 0.1544 +RLMRec(Renetal.,2024) 0.4010 0.2646 0.3773 0.3931 0.3526 0.4889 0.4436 0.2994 0.3506 +PALR(Yangetal.,2023) 0.4715 0.3675 0.1917 0.4938 0.2197 0.2390 0.5216 0.2050 0.2267 +LG(Wangetal.,2025) 0.5709 0.2378 0.2682 0.5109 0.2546 0.3749 0.5506 0.3883 0.3364 +R4Rec(Fangetal.,2025) 0.4882 0.2348 0.2330 0.4208 0.3328 0.3595 0.4946 0.2831 0.2949 +Ours 0.6382 0.2880 0.3429 0.5947 0.4002 0.4482 0.7287 0.3457 0.3127 +Table6: Semanticalignmentandcoverageofgeneratedprofiles. Cov. denotesthefractionoftokensgroundedin +historicaltext. +recoverfromrawhistoriesorstaticpromptsalone. reinforcementlearning–basedoptimization. These +resultssuggestthatadaptive,interaction-awaretex- +Experiments on multiple real-world datasets tualprofilesprovideapromisingdirectionformore +demonstratethatDUETconsistentlyoutperforms effective interpretable and performance-oriented +strongbaselinesunderdifferentbackbonemodels, recommendationsystems. +validating the effectiveness of joint profiling and + +| Limitations | | | | | | | References | | | | | | | +| ----------- | --- | --- | --- | --- | --- | --- | ---------- | ----- | ------ | ---- | ------ | ------ | ----- | +| | | | | | | | Keqin Bao, | Jizhi | Zhang, | Yang | Zhang, | Wenjie | Wang, | +Despiteitseffectiveness,ourapproachhasseveral FuliFeng,andXiangnanHe.2023. Tallrec: Anef- +fectiveandefficienttuningframeworktoalignlarge +| limitations. | First, | the | proposed | framework | | relies | | | | | | | | +| ------------ | ------ | --- | -------- | --------- | --- | ------ | -------------------------------- | --- | --- | --- | --- | ---------- | --- | +| | | | | | | | languagemodelwithrecommendation. | | | | | InProceed- | | +onlargelanguagemodelsforbothprofilegenera- ingsofthe17thACMConferenceonRecommender +tion and downstream recommendation, which in- Systems,pages1007–1014. +| troduces | additional | computational | | | overhead | dur- | | | | | | | | +| -------- | ---------- | ------------- | --- | --- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | +TingChen,Wei-LiHan,Hai-DongWang,Yi-XunZhou, +| ingtrainingandinference. | | | Whileourexperiments | | | | | | | | | | | +| ------------------------ | --- | --- | ------------------- | --- | --- | --- | ------------------------- | --- | --- | --- | ---------------- | --- | --- | +| | | | | | | | BinXu,andBin-YuZang.2007. | | | | Contentrecommen- | | | +showconsistentgainsacrossdifferentbackbones, dationsystembasedonprivatedynamicuserprofile. +the overall efficiency may be constrained when In2007Internationalconferenceonmachinelearn- +| | | | | | | | ing and | cybernetics, | | volume | 4, pages | 2112–2118. | | +| --------------------------------- | --- | --- | --- | --- | ---------- | --- | ------- | ------------ | --- | ------ | -------- | ---------- | --- | +| scalingtoverylargeuseroritemsets. | | | | | Second,the | | | | | | | | | +IEEE. +| quality | of the generated | | profiles | is | inherently | de- | | | | | | | | +| ------- | ----------------- | --- | -------- | --- | ------------- | --- | --------- | ------- | --- | ----------- | ----- | --- | ----- | +| pendent | on the underlying | | LLMs | | and prompting | | | | | | | | | +| | | | | | | | Yue Chen, | Minghua | | He, Fangkai | Yang, | Pu | Zhao, | +strategies. Variationsinmodelcapacityorprompt LuWang,YuKang,YifeiDong,YuefengZhan,Hao +Sun,QingweiLin,SaravanRajmohan,andDongmei +sensitivitymayleadtodifferencesinprofilestabil- +| | | | | | | | Zhang. | 2025. | Warriormath: | | Enhancing | the | mathe- | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | ------------ | --- | --------- | --- | ------ | +ity,whichwedonotexplicitlycontrolinthecurrent +maticalabilityoflargelanguagemodelswithadefect- +design. Finally,ourevaluationfocusesontext-rich +| | | | | | | | awareframework. | | | Preprint,arXiv:2508.01245. | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | -------------------------- | --- | --- | --- | +recommendationscenarioswheresufficienthistor- +| | | | | | | | Paul Covington, | | Jay | Adams, and | Emre | Sargin. | 2016. | +| ------------ | --- | ---------- | --- | ----------------- | --- | --- | --------------- | --- | --- | ---------- | ---- | ------- | ----- | +| ical reviews | are | available. | | The effectiveness | | of | | | | | | | | +Deepneuralnetworksforyoutuberecommendations. +| the proposed | strategy | | in domains | with | extremely | | | | | | | | | +| ------------ | -------- | --- | ---------- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +InProceedingsofthe10thACMconferenceonrec- +sparsetextualsignalsorinnon-textualmodalities +ommendersystems,pages191–198. +remainstobefurtherexplored. +DeepSeek-AI,DayaGuo,DejianYang,HaoweiZhang, +JunxiaoSong,RuoyuZhang,RunxinXu,QihaoZhu, +ShirongMa,PeiyiWang,XiaoBi,XiaokangZhang, +Data,Privacy,andEthicsConsiderations Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhi- +| | | | | | | | hong | Shao, | Zhuoshu | Li, Ziyi | Gao, and | 181 | others. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | ------- | ------------- | --------- | --- | ------- | +| | | | | | | | 2025. | Deepseek-r1: | | Incentivizing | reasoning | | capa- | +Alldatasetsusedinthisworkarepubliclyreleased bilityinllmsviareinforcementlearning. Preprint, +arXiv:2501.12948. +bytheiroriginalowners,andwefollowtheofficial +termsofuseassociatedwitheachdataset,utilizing +JiaxinDeng,ShiyaoWang,KuoCai,LejianRen,Qigen +thedatasolelyforresearchpurposesaspermitted +| | | | | | | | Hu, Weifeng | | Ding, | Qiang Luo, | and | Guorui | Zhou. | +| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ----- | ---------- | --- | ------ | ----- | +bytheirrespectivelicenses;forreproducibility,the 2025. Onerec: Unifyingretrieveandrankwithgen- +officialdatasetsourcesarecitedinthecorrespond- erativerecommenderanditerativepreferencealign- +ment. Preprint,arXiv:2502.18965. +ingsectionsofthispaper,andwedonotredistribute +anyrawuser-generatedcontentsuchasreviewtext AbhimanyuDubey,AbhinavJauhri,AbhinavPandey, +| aspartofthiswork. | | Thisresearchdoesnotattempt | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +AbhishekKadian,AhmadAl-Dahle,AieshaLetman, +to identify or infer the identity of any individual Akhil Mathur, Alan Schelten, Amy Yang, Angela +| | | | | | | | Fan,and1others.2024. | | | Thellama3herdofmodels. | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | ---------------------- | --- | --- | --- | +user,andpriortotraininganddownstreamprofile +arXivpreprintarXiv:2407.21783. +| generation, | we | apply text | filtering | | procedures | to | | | | | | | | +| ----------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +remove direct personal identifiers where applica- YiFang,WenjieWang,YangZhang,FengbinZhu,Qi- +ble;themodelsaretrainedsolelyonpubliclyavail- fanWang,FuliFeng,andXiangnanHe.2025. Rea- +son4rec:Largelanguagemodelsforrecommendation +abledata,andnoadditionalprivateorproprietary +| | | | | | | | withdeliberativeuserpreferencealignment. | | | | | Preprint, | | +| ---------------- | --- | -------------- | --- | ------------ | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --------- | --- | +| user information | | is introduced. | | Furthermore, | | we | | | | | | | | +arXiv:2502.02061. +| do not release | generated | | user | profiles | or | any per- | | | | | | | | +| -------------- | --------- | --- | ---- | -------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +MinghuaHe,YueChen,FangkaiYang,PuZhao,Wenjie +useroutputsproducedduringthemodelingprocess, +Yin,YuKang,QingweiLin,SaravanRajmohan,and +andallreportedresultsarepresentedinaggregate +| | | | | | | | DongmeiZhang.2025. | | | Execoder:Empoweringlarge | | | | +| -------- | ---------- | -------- | --- | ----- | ---- | ------- | ------------------ | --- | --- | ------------------------ | --- | --- | --- | +| form for | evaluation | purposes | | only. | From | a secu- | | | | | | | | +languagemodelswithexecutabilityrepresentationfor +rityperspective,theassociatedcoderepositoryhas codetranslation. Preprint,arXiv:2501.18460. +beenscannedforpotentialsecrets,andanyinternal +MinjieHong,ZirunGuo,YanXia,ZehanWang,Ziang +endpointsorconfidentialconfigurationinformation +| | | | | | | | Zhang, | Tao | Jin, and | Zhou Zhao. | 2025a. | Apo: | En- | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | ---------- | ------ | ---- | --- | +havebeenremovedpriortorelease. +hancingreasoningabilityofmllmsviaasymmetric +| | | | | | | | policyoptimization. | | | Preprint,arXiv:2506.21655. | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | -------------------------- | --- | --- | --- | + +Minjie Hong, Yan Xia, Zehan Wang, Jieming Zhu, JoarSkalse,NikolausH.R.Howe,DmitriiKrashenin- +Ye Wang, Sihang Cai, Xiaoda Yang, Quanyu Dai, nikov, and David Krueger. 2025. Defining +ZhenhuaDong,ZhimengZhang,and1others.2025b. and characterizing reward hacking. Preprint, +| Eager-llm: | Enhancinglargelanguagemodelsasrec- | | | | | arXiv:2209.13085. | | | | | | +| ---------- | ---------------------------------- | --------- | ----------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| ommenders | through | exogenous | behavior-semantic | | | | | | | | | +integration. InProceedingsoftheACMonWebCon- Harald Steck. 2019. Embarrassingly shallow autoen- +ference2025,pages2754–2762. codersforsparsedata. InProceedingsofTheWorld +WideWebConference(WWW),pages3251–3257. +MinjieHong,ZetongZhou,ZirunGuo,ZiangZhang, +| Ruofan | Hu, Weinan | Gan, | Jieming | Zhu, and | Zhou | | | | | | | +| ------ | ---------- | ---- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | +ChaoSun,YaoboLiang,YamingYang,ShilinXu,Tian- +| Zhao.2025c. | Generativereasoningrecommendation | | | | | | | | | | | +| ----------- | --------------------------------- | --- | --- | --- | --- | ---------------------------- | -------- | ---- | ------ | -------- | ------------- | +| | | | | | | mengYang,andYunhaiTong.2024. | | | | | Rlrf4rec: Re- | +| viallms. | Preprint,arXiv:2510.20815. | | | | | | | | | | | +| | | | | | | inforcement | learning | from | recsys | feedback | for en- | +JihwanJeong,YinlamChow,GuyTennenholtz,Chih- hancedrecommendationreranking. arXivpreprint +| Wei Hsu, | Azamat | Tulepbergenov, | | Mohammad | | arXiv:2410.05939. | | | | | | +| ------------------------------------ | ------------ | --------------- | ---------- | -------- | --------- | ----------------- | --- | --------------------- | --- | --- | --------- | +| Ghavamzadeh, | and | Craig | Boutilier. | 2023. | Fac- | | | | | | | +| | | | | | | QwenTeam.2025. | | Qwen3technicalreport. | | | Preprint, | +| tual and | personalized | recommendations | | using | lan- | | | | | | | +| guagemodelsandreinforcementlearning. | | | | | Preprint, | arXiv:2505.09388. | | | | | | +arXiv:2310.06176. +| | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | +| --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | +Yitong Ji, Aixin Sun, Jie Zhang, and Chenliang Li. bert, Amjad Almahairi, Yasmine Babaei, Nikolay +| 2023. A | critical | study | on data leakage | in | recom- | | | | | | | +| ------- | -------- | ----- | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti +| mendersystemofflineevaluation. | | | | ACMTrans.Inf. | | | | | | | | +| ------------------------------ | --- | --- | --- | ------------- | --- | ------------------------ | --- | --- | ------- | --- | --------- | +| | | | | | | Bhosale,and1others.2023. | | | Llama2: | | Openfoun- | +Syst.,41(3):75:1–75:27. +| | | | | | | dation and | fine-tuned | chat | models. | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | ---------- | ---------- | ---- | ------- | ----- | -------- | +arXiv:2307.09288. +| JiachengLin,TianWang,andKunQian.2025. | | | | | Rec-r1: | | | | | | | +| ------------------------------------- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +Bridginggenerativelargelanguagemodelsanduser- +| | | | | | | LeandrovonWerra, | | YounesBelkada, | | LewisTunstall, | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | -------------- | --- | -------------- | --- | +centricrecommendationsystemsviareinforcement +EdwardBeeching,TristanThrush,NathanLambert, +| learning. | Preprint,arXiv:2503.24289. | | | | | | | | | | | +| --------- | -------------------------- | --- | --- | --- | --- | -------- | ------ | ----------- | ------------- | --- | ------------ | +| | | | | | | Shengyi | Huang, | Kashif | Rasul, | and | Quentin Gal- | +| | | | | | | louédec. | TRL: | Transformer | Reinforcement | | Learn- | +AiweiLiu,MinghuaHe,ShaoxunZeng,SijunZhang, +| LinhaoZhang,ChuhanWu,WeiJia,YuanLiu,Xiao | | | | | | ing. | | | | | | +| ---------------------------------------- | --- | --- | ----------------------- | --- | --- | ---- | --- | --- | --- | --- | --- | +| Zhou,andJieZhou.2025. | | | Wedlm:Reconcilingdiffu- | | | | | | | | | +sionlanguagemodelswithstandardcausalattention LuWang,DiZhang,FangkaiYang,PuZhao,Jianfeng +forfastinference. Preprint,arXiv:2512.22737. Liu,YuefengZhan,HaoSun,QingweiLin,Weiwei +| | | | | | | Deng, Dongmei | | Zhang, | Feng | Sun, and | Qi Zhang. | +| --- | --- | --- | --- | --- | --- | ------------- | --- | ------ | ---- | -------- | --------- | +HongtaoLiu,FangzhaoWu,WenjunWang,Xianchen +| | | | | | | 2025. Lettingo: | | Exploreuserprofilegenerationfor | | | | +| --- | --- | --- | --- | --- | --- | --------------- | --- | ------------------------------- | --- | --- | --- | +Wang,PengfeiJiao,ChuhanWu,andXingXie.2019. +| | | | | | | recommendationsystem. | | | InProceedingsofthe31st | | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | +Nrpa: Neuralrecommendationwithpersonalizedat- ACMSIGKDDConferenceonKnowledgeDiscovery +tention. In Proceedings of the 42nd International andDataMiningV.2,KDD’25,NewYork,NY,USA. +ACMSIGIRConferenceonResearchandDevelop- AssociationforComputingMachinery. +mentinInformationRetrieval,pages1233–1236. +YeWang,JiahaoXun,MinjieHong,JiemingZhu,Tao +| Wensheng | Lu, Jianxun | Lian, | Wei Zhang, | Guanghua | | | | | | | | +| -------- | ----------- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | +Jin,WangLin,HaoyuanLi,LinjunLi,YanXia,Zhou +Li,MingyangZhou,HaoLiao,andXingXie.2024. +| | | | | | | Zhao,and1others.2024. | | | Eager: | Two-streamgen- | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ------ | -------------- | --- | +Aligninglargelanguagemodelsforcontrollablerec- +arXivpreprintarXiv:2403.05063. erativerecommenderwithbehavior-semanticcollab- +ommendations. +| | | | | | | oration. | InProceedingsofthe30thACMSIGKDD | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------- | --- | --- | --- | --- | +RadaMihalceaandPaulTarau.2004. Textrank: Bring- ConferenceonKnowledgeDiscoveryandDataMin- +| | | InEMNLP. | | | | ing,pages3245–3254. | | | | | | +| --- | --- | -------- | --- | --- | --- | ------------------- | --- | --- | --- | --- | --- | +ingorderintotext. +| Zhaopeng | Qiu, Xian | Wu, | Jingyue Gao, | and | Wei Fan. | | | | | | | +| -------- | --------- | --- | ------------ | --- | -------- | -------- | --------------- | --- | -------- | -------- | --- | +| | | | | | | Triyanna | Widiyaningtyas, | | Indriana | Hidayah, | and | +2021. U-BERT:pre-traininguserrepresentationsfor +| | | | | | | TeguhBAdji.2021. | | Userprofilecorrelation-based | | | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | +improved recommendation. In AAAI, pages 4320– similarity(upcsim)algorithminmovierecommenda- +| 4327.AAAIPress. | | | | | | tionsystem. | JournalofBigData,8(1):52. | | | | | +| --------------------------------- | --- | --- | --- | -------------- | --- | ----------- | ----------------------------------- | --- | --- | --- | --- | +| NilsReimersandIrynaGurevych.2019. | | | | Sentence-bert: | | | | | | | | +| | | | | | | J.Wu.2023. | Computationalunderstandingofuserin- | | | | | +Sentenceembeddingsusingsiamesebert-networks. +terfaces. +InProceedingsofthe2019ConferenceonEmpirical +MethodsinNaturalLanguageProcessing(EMNLP), +pages3982–3992. Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, +| | | | | | | Hong Zhu, | Jieming | Zhu, | Bo Chen, | Ruiming | Tang, | +| --- | --- | --- | --- | --- | --- | --------- | ------- | ---- | -------- | ------- | ----- | +Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi WeinanZhang,andYongYu.2024. Towardsopen- +Cheng,JunfengWang,DaweiYin,andChaoHuang. world recommendation with knowledge augmenta- +2024. Representationlearningwithlargelanguage tionfromlargelanguagemodels. InProceedingsof +modelsforrecommendation. InProceedingsofthe the18thACMConferenceonRecommenderSystems, +| ACMonWebConference2024,pages3464–3475. | | | | | | pages12–22. | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | + +A ExperimentSetup +FanYang,ZhengChen,ZiyanJiang,EunahCho,Xiao- +| jiangHuang,andYanbinLu.2023. | | Palr:Personaliza- | | | | | | | | +| ------------------------------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| tionawarellmsforrecommendation. | | arXivpreprint | A.1 DataCuration | | | | | | | +arXiv:2305.07622. +Jiarui Zhang. 2024. Guided profile generation im- Table7: Statisticaldetailsoftheevaluationdatasets. +| proves personalization | with llms. | arXiv preprint | | | | | | | | +| ---------------------- | ---------- | -------------- | ---------- | ------ | ----------- | --- | ------------ | ------ | ------ | +| arXiv:2409.13093. | | | Dataset | #Train | #Valid | | #Test #User | | #Item | +| | | | Music | 43,071 | 3,271 | | 1,296 | 4,183 | 2,660 | +| | | | Book | 71,972 | 6,144 | | 5,541 13,863 | | 13,515 | +| | | | Yelp | 51,497 | 4,757 | | 4,328 | 8,453 | 13,426 | +| | | | We conduct | | experiments | | on three | widely | used | +real-worlddatasets: +| | | | • AmazonMusic(Music): | | | | Thisreferstothe“Dig- | | | +| --- | --- | --- | --------------------- | --- | ------ | ------ | -------------------- | --- | ------ | +| | | | ital Music” | | subset | of the | well-known | | Amazon | +Productdataset3,whichrecordsrichuserreviews, +ratings,andtextualinformationaboutitems,such +| | | | as titles, | across | a | broad | range | of product | cate- | +| --- | --- | --- | ---------- | ------ | --- | ----- | ----- | ---------- | ----- | +gories,ontheAmazonplatform.2 +| | | | • AmazonBook(Book): | | | Thisreferstothe“Book” | | | | +| --- | --- | --- | ------------------- | --- | --- | --------------------- | --- | --- | --- | +subsetoftheAmazonProductdataset. +dataset4, +| | | | • Yelp: | This | refers | to the | Yelp | Open | | +| --- | --- | --- | ------- | -------- | -------------- | -------- | ------- | ------------- | --------- | +| | | | which | includes | user | reviews, | ratings | | for busi- | +| | | | nesses | such | as restaurants | | and | retail shops, | as | +wellastextualinformationaboutthebusinesses. +Itiswidelyusedinrecommendationtasks(Qiu +etal.,2021). +WeusetheentireMusicdatasetforexperiments, +| | | | while for | the Book | | and Yelp | datasets, | we | utilize | +| --- | --- | --- | ------------------------------- | -------- | --- | -------- | --------- | ---------- | ------- | +| | | | onlyasubsetduetotheirlargesize. | | | | | FortheBook | | +dataset,weusedatafromthelasttwomonths,and +fortheYelpdataset,weusedatafromthelastsix +months. Foreachdataset,wesplititintotraining, +| | | | validation, | and | test | sets based | on | the timestamps | | +| --- | --- | --- | ---------------- | --- | -------- | ---------- | ---- | -------------- | --- | +| | | | of interactions, | | ensuring | that | test | interactions | oc- | +curafteralltrainingandvalidationinteractionsto +preventinformationleakage(Jietal.,2023). +| | | | Regarding | | data | filtering, | following | | prior | +| --- | --- | --- | --------- | --- | ----------- | ---------- | --------- | -------- | ----- | +| | | | work (Liu | et | al., 2019), | we | adopt | a 5-core | set- | +tingtofilterthedataandexcludecold-startusers +| | | | and items—those | | not | appearing | | in the | training | +| --- | --- | --- | --------------- | --- | ---------- | --------- | ---- | ------ | -------- | +| | | | set—from | the | validation | and | test | sets. | The sta- | +tisticaldetailsoftheprocesseddatasetareprovided +inTable7. +3https://cseweb.ucsd.edu/~jmcauley/datasets/ +amazon/links.html. +4https://business.yelp.com/data/resources/ +open-dataset/. + +| A.1.1 | ImplementationDetails | | | | | | RLMRecPrompt | | | | +| ----- | --------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | +Inourexperiments,weprimarilyemployQwen3- Role: Businessrecommendationassistant +Task: Determinebusinesstypesauserislikely +| 8B(Team,2025)andLLaMA38BInstruct | | | | | | (Dubey | | | | | +| ------------------------------------------ | --- | ----- | --- | --- | ------- | ---------- | ------------ | --- | --- | --- | +| etal.,2024;Touvronetal.,2023)asboththerec- | | | | | | | toenjoy | | | | +| ommendation | | model | and | the | profile | generation | InputFormat: | | | | +•Title: Businessname +model. Thetrainingprocessisimplementedusing +theTRL(vonWerraetal.). Keyhyperparameters, •Categories: Businesscategories +suchasbatchsizeandlearningrate,aredetermined •Sentiment: Usersentimenttowardbusiness +throughgridsearchtoachieveoptimalperformance. +OutputRequirements: +Moredetailscanbefoundinourcode. +1. JSONformatonly +2. Structure: +A.2 BaselinePrompts +{ +| | | | | | | | "summarization": | "Types | of businesses | user | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------ | ------------- | ---- | +KARPrompt +likelyenjoys"(≤100words), +Task: Analyzeuserpreferencesbasedonbusi- +| | | | | | | | "reasoning": | "Brief explanation | | for summa- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------ | --- | ---------- | +nessreviewinghistory +rization"(nowordlimit) +Input: {user_history}-User’sbusinessreview- +} +inghistorywithsentimentsovertime +3. NoadditionaltextoutsideJSON +Instructions: +| 1. | Analyze | the | user’s | preferences | considering | | | | | | +| --- | ------- | --- | ------ | ----------- | ----------- | --- | ------ | ----------- | --- | ------ | +| | | | | | | | Input: | INTERACTION | | ITEMS: | +businessnamesandcategories +{user_history} +| 2. | Take | into account | | sentiment | patterns | over | | | | | +| --- | ---- | ------------ | --- | --------- | -------- | ---- | --- | --- | --- | --- | +time +| 3. | Provideclearexplanationsbasedonreview- | | | | | | LGPrompt | | | | +| --- | -------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | +inghistorydetails +Youwillserveasanassistanttohelpmegener- +| 4. | Consider | other | pertinent | | factors | that may | | | | | +| --- | -------- | ----- | --------- | --- | ------- | -------- | --- | --- | --- | --- | +ateauserprofilebasedonthisuser’ssentiments +influencepreferences +historytobetterunderstandthisusers’interest +andthuspredicthis/hersentimentaboutatarget +| PALRPrompt | | | | | | | item. Iwillprovideyouwithsomebehaviorhis- | | | | +| ---------- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --------------- | --- | +| | | | | | | | toryoftheuserinthisformat: | | [itemattributes | | +Task: Summarizeuserpreferencesusingkey- +| | | | | | | | andsentiment]. | Theuserprofileyougenerate | | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ------------------------- | --- | --- | +words. +shouldcontainasmuchusefulcontentaspossi- +| Input: | | {user_history} | | - historical | businesses | | | | | | +| ------ | --- | -------------- | --- | ------------ | ---------- | --- | --- | --- | --- | --- | +bletohelppredicttheuser’ssentimenttowards +withusersentiments. +anewbusiness. +| Output | | Format: | An | itemized | list ranked | by | | | | | +| ------ | --- | ------- | --- | -------- | ----------- | --- | --- | --- | --- | --- | +USERHISTORY:{user_history}. +importance. +PROFILEYOUGENERATE: +Template: +| | • KEY_WORD_1: | | | | | "HIS- | | | | | +| --- | ------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +R4RecPrompt(Reasoner) +| | TORY_BUSINESS_1", | | | | | "HIS- | | | | | +| --- | ----------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +UserReviewHistory +TORY_BUSINESS_2" +⟨H u organizedasbelow⟩ +| | • KEY_WORD_2: | | | | | "HIS- | 1. TitleofItem1 | | | | +| ------------- | ---------------- | --- | --- | --- | --- | ----- | ---------------- | ----------------------- | ----------- | --- | +| | TORY_BUSINESS_3" | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| | | | | | | | NegativeAspects: | [Aspect1],[Aspect2],... | | | +| Instructions: | | | | | | | User Preference | Elements: | [Preference | 1], | +[Preference2],... +| | 1. Extract | key | preference | | indicators | from | | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | ---- | --- | --- | --- | --- | +2. TitleofItem2 +userinteractionhistory. +| | | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------------- | --- | --- | +2. Rankkeywordsbyimportance. NegativeAspects: [Aspect1],[Aspect2],... +| | | | | | | | User Preference | Elements: | [Preference | 1], | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --------- | ----------- | --- | + +| [Preference2],... | | | | | | | cantlyworsethanDUETacrossalldatasets. | | | | | This | +| ----------------- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | ---- | +indicatesthatcoherentsummariesaloneareinsuf- +... +| | | | | | | | ficient, | and that | the gains | of DUET | | arise from | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --------- | ------- | --- | ---------- | +ItemReviewHistorybyOtherUsers +⟨H organizedinthesameformatasabove⟩ learnedsemanticabstractionratherthanextractive +i +compression. +| Task: | Analyze | whether | | the user | will | like the | | | | | | | +| --------- | ------- | ------- | --- | ---------- | ----------- | -------- | --- | --- | --- | --- | --- | --- | +| new Music | | i based | on | the user’s | preferences | | | | | | | | +B.2 RankingunderHardNegativeSampling +| andtheitem’sfeatures. | | | | Provideyourrationale | | | | | | | | | +| --------------------- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Toconstructamorechallengingrankingscenario, +inoneconciseparagraph. +wereplacerandomnegativeswithhardnegatives +| | | | | | | | generatedbyacollaborativefilteringmodel. | | | | | Specif- | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ------- | +A.3 ExampleofCue +ically,foreachuser,weretrievehigh-scoringitems +Thefollowingexamplesillustratehowrawsignals +| | | | | | | | from an | EASE(Steck, | 2019) | model | that | the user | +| --------------------- | --- | --- | ---------------- | --- | --- | ---- | ------- | ----------- | ----- | ----- | ---- | -------- | +| aredistilledintocues. | | | Asshownonthetop, | | | user | | | | | | | +hasnotinteractedwith,andcombinethemwiththe +cuesemphasizehistoricalpreferences,whileitem +ground-truthitemtoformthecandidateset. +cues highlight metadata and user-group patterns. Table 9 reports the results. Compared to ran- +| Together, | they | provide | minimal | | but informative | | | | | | | | +| --------- | ---- | ------- | ------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +domsampling,performancedecreasesforallmeth- +hypothesesforprofileexploration. +| | | | | | | | ods due | to increased | difficulty, | while | DUET | con- | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | ----------- | ---------------- | ---- | ------ | +| | | | | | | | sistently | maintains | the | best performance | | across | +ExamplesofUserCues +| | | | | | | | datasets. | Thisindicatesthattheimprovementsare | | | | | +| ------- | ----- | ------ | ------ | --- | ------- | -------- | --------- | ----------------------------------- | --- | --- | --- | --- | +| “enjoys | retro | puzzle | games” | — | derived | from re- | | | | | | | +robustandnotlimitedtotrivialrankingscenarios. +peatedengagementwithclassictitles. +“prefersconciseproductreviews”—inferredfrom B.3 RobustnessunderPreferenceDiversity +apatternofshort,directcomments. +WefurtheranalyzetherobustnessofDUETunder +“tendstogivehighratingsbutrarelycomments”— +| | | | | | | | varyinglevelsofuserpreferencediversity. | | | | | Weuse | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | ----- | +highlightingconsistencybutlimitedfeedback. +thevarianceofhistoricalratingsasaproxyforpref- +| | | | | | | | erencestability: | | lowvarianceindicatesconsistent | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ------------------------------ | --- | --- | --- | +ExamplesofItemCues +| | | | | | | | preferences, | while | high | variance | corresponds | to | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ----- | ---- | -------- | ----------- | --- | +“lightweighttrail-runningshoes”—derivedfrom +diverseorpotentiallyconflictingsignals. +productmetadata. +Wepartitionusersintothreegroupsbasedonper- +| “popular | among | budget-conscious | | | users” | — in- | | | | | | | +| -------- | ----- | ---------------- | --- | --- | ------ | ----- | --- | --- | --- | --- | --- | --- | +centilethresholds(bottom33%,middle33%,top +ferredfrompurchasepatterns. +33%ofratingvariance)andevaluateperformance +| “stylized | with | retro | aesthetics” | | — extracted | from | | | | | | | +| --------- | ---- | ----- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | +withineachgroup. +itemdescriptions. +| | | | | | | | As shown | in | Table 10, | performance | | degrades | +| --- | --- | --- | --- | --- | --- | --- | -------- | --- | --------- | ----------- | --- | -------- | +smoothlyaspreferencediversityincreasesacross +| | | | | | | | all datasets. | Importantly, | | the degradation | | is grad- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------ | --- | --------------- | --- | -------- | +B AdditionalExperiments +ualratherthancatastrophic,indicatingthatDUET +B.1 Non-LLMBaselineviaExtractive +remainsstableunderheterogeneousornoisyinter- +Summarization +actionhistories. +| To examine | | whether | the | gains | of DUET | stem | | | | | | | +| ---------- | --- | ------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | +C TheUseofLargeLanguageModels +| from improved | | semantic | | representations | | rather | | | | | | | +| ------------- | --- | ---------------- | --- | --------------- | --------- | ------ | --- | --- | --- | --- | --- | --- | +| than generic | | text generation, | | we | introduce | a non- | | | | | | | +WeusedaLargeLanguageModel(LLM)onlyas +LLMbaselinebasedonextractivesummarization. a writing assistant to polish the language of the +| Specifically, | | we apply | TextRank | | (Mihalcea | and | | | | | | | +| ------------- | --- | -------- | -------- | --- | --------- | --- | ---------- | ------ | ------- | ----------- | --- | --------- | +| | | | | | | | manuscript | (e.g., | grammar | refinement, | | style ad- | +Tarau,2004)toselectsalientsentencesfromuser +| | | | | | | | justment,andclarityimprovement). | | | | Theresearch | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | ----------- | --- | +historiesandconstructuserprofileswithoutusing ideas,methodologydesign,experiments,andanal- +anygenerativemodel. ysiswereentirelyconceived,implemented,andval- +| The | extracted | summaries | | are | then fed | into the | | | | | | | +| --- | --------- | --------- | --- | --- | -------- | -------- | --- | --- | --- | --- | --- | --- | +idatedbytheauthorswithoutrelianceontheLLM. +samedownstreampredictorforratingestimation. +| | | | | | | | The LLM | did not | contribute | to research | | ideation, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | ---------- | ----------- | --- | --------- | +Thisbaselineisolatestheeffectofreadablesumma- experimentaldesign,orresultinterpretation. +rizationfromrepresentationlearning. +Table8showsthatTextRankimprovesoversim- +| ple history | truncation | | (10H), | but | remains | signifi- | | | | | | | +| ----------- | ---------- | --- | ------ | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | + +| Method | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | -------- | ---- | ----- | --- | ----------- | --------- | --- | ----------- | --------- | +| | MAE RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | MAE | RMSE | Acc F1(%) | +| | | (%) | | | | (%) | | | (%) | +10H 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +TextRank 0.8104 1.1323 50.51 30.23 0.5914 0.8223 61.19 29.68 0.6328 0.8854 59.59 29.48 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table8: Comparisonwithanon-LLMextractivesummarizationbaseline(TextRank). +| Method | | | Yelp | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | ---- | --- | ----------- | --- | --- | ----------- | --- | +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +| | Table9: | RankingperformanceunderEASE-based(Steck,2019)hardnegatives. | | | | | | | | +| ------------- | ------- | ----------------------------------------------------------- | --- | --- | ----------- | --- | --- | ----------- | --- | +| VarianceGroup | | Yelp | | | AmazonMusic | | | AmazonBooks | | +#Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 +| | | | (%) | (%) | | (%) (%) | | | (%) (%) | +| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ------- | +Stable 1530 0.4017 0.8664 71.76 68.96 432 0.2368 0.5990 82.56 86.89 1876 0.3171 0.8403 76.64 72.56 +Moderate 1354 0.4773 0.8309 60.09 46.26 436 0.3921 0.6628 64.16 47.02 1827 0.4495 0.8059 66.73 55.45 +Diverse 1444 0.6632 1.1198 51.13 45.95 428 0.5536 0.9607 57.10 59.27 1838 0.6199 1.0603 49.52 46.35 +Table10: PerformanceofDUETunderdifferentlevelsofuserpreferencediversity(measuredbyratingvariance). +#Samp. denotesthenumberofsamplesineachgroup. \ No newline at end of file diff --git a/q126/random_k4/random_4.md b/q126/random_k4/random_4.md index 8efe386e6cfdafd89c56e967acf5dfb7f8070736..a2b48e41bd57bce68e73b59f00704cb4c84c38f3 100644 --- a/q126/random_k4/random_4.md +++ b/q126/random_k4/random_4.md @@ -1,3327 +1,2098 @@ -Cretaceous stratigraphy, paleontology, petrography. Depositional -environments, and cycle stratigraphy at Cerro de Cristo Rey, Dona Ana -County, New Mexico - -Spencer G. Lucas, Karl Krainer, Justin A. Spielmann, and Kevin Durney - -New Mexico Geology, v. 32, n. 4 pp. 103-130, Print ISSN: 0196-948X, Online ISSN: 2837-6420. -https://doi.org/10.58799/NMG-v32n4.103 -Download from: https://geoinfo.nmt.edu/publications/periodicals/nmg/backissues/home.cfml?volume=32&number=4 - -New Mexico Geology (NMG) publishes peer-reviewed geoscience papers focusing on New Mexico and the -surrounding region. We aslo welcome submissions to the Gallery of Geology, which presents images of -geologic interest (landscape images, maps, specimen photos, etc.) accompanied by a short description. - -Published quarterly since 1979, NMG transitioned to an online format in 2015, and is currently being issued -twice a year. NMG papers are available for download at no charge from our website. You can also -subscribe to receive email notifications when new issues are published. - -New Mexico Bureau of Geology & Mineral Resources -New Mexico Institute of Mining & Technology -801 Leroy Place -Socorro, NM 87801-4796 - -https://geoinfo.nmt.edu - - This page is intentionally left blank to maintain order of facing pages. - - Cretaceous stratigraphy, paleontology, petrography, -depositional environments, and cycle stratigraphy at -Cerro de Cristo Rey, Doña Ana County, New Mexico - -Spencer G. Lucas, New Mexico Museum of Natural History, 1801 Mountain Road NW, Albuquerque, New Mexico 87104, spencer.lucas@state.nm.us; - -Karl Krainer, Institute of Geology and Paleontology, Innsbruck University, Innrain 52, Innsbruck, A-6020 Austria, Karl.Krainer@uibk.ac.at; -Justin A. Spielmann, New Mexico Museum of Natural History, 1801 Mountain Road NW, Albuquerque, New Mexico 87104, -justin.spielmann1@state.nm.us; Kevin Durney, 11212 Jockey Bluff Drive, Austin, Texas 78748 - -Abstract - -Cretaceous marine and nonmarine strata -of late Albian–middle Cenomanian age are -exposed around the Cerro de Cristo Rey uplift -in southern Doña Ana County, New Mexico. -These strata comprise a section approximately -350 m thick and are assigned to the (ascending -order) Finlay, Del Norte, Smeltertown, Mule- -ros, Mesilla Valley, Mojado (=“Anapra”), Del -Rio, Buda, and Mancos (=“Boquillas”) For- -mations. Macrofossils and microfossils from -these strata indicate that the Finlay, Del Norte, -Smeltertown, Muleros, and Mesilla Valley -Formations are of late Albian age, whereas the -Del Rio, Buda, and Mancos Formations are of -Cenomanian age. The base of the Cenomanian -is most likely at a trangressive surface within -the uppermost Mojado Formation. The late -Albian (Manuaniceras powelli Zone) to early -Cenomanian (Neophlycticeras hyatti Zone) -sedimentary succession at Cerro de Cristo Rey -consists of alternating fossiliferous limestone, -shale with limestone and sandstone intercala- -tions, and sandstone. Muddy limestone types -are commonly wavy to nodular and represent -deposits of an open-marine shelf environment -below wave base. Intercalated coquina beds -rich in mollusc shells are interpreted as storm -layers. Shale was deposited in an open-shelf -environment below or near wave base during -periods of increased siliciclastic influx. Inter- -calated thin limestone and sandstone beds are -inferred to be storm layers. The siliciclastic -Mojado Formation is a regressive-transgres- -sive succession formed in depositional envi- -ronments ranging from lower shoreface to -upper shoreface and even fluvial, again over- -lain by shallow-marine siliciclastics. Although -the Washita Group section at Cerro de Cristo -Rey is much thicker and displays some differ- -ences in facies, the succession shows similar -transgressive and regressive trends compared -to the Washita Group of north Texas. Thus, -we recognize eight unconformity-bounded -depositional cycles in the Cretaceous sec- -tion at Cerro de Cristo Rey, the upper Finlay -Formation (youngest cycle of the Freder- -icksburg Group), lower Mancos Formation -(base of Greenhorn cycle), and six Washita -Group cycles: WA1 = Del Norte Formation, -WA2 = Smeltertown Formation, WA3 = Mule- -ros Formation, WA4 = most of Mesilla Valley -Formation, WA5 = uppermost Mesilla Val- -ley Formation and most of Mojado Forma- -tion, and WA6 = uppermost Mojado and Del -Rio and Buda Formations. The persistence -of cycles from the tectonically passive, open- -marine margin of the Gulf of Mexico into the -tectonically active Chihuahua trough suggests -that regional if not global eustasy, not local -tectonism, drove late Early to early Late Cre- -taceous sedimentation at Cerro de Cristo Rey. - -FIGURE 1—Location of Cerro de Cristo Rey in southern Doña Ana County, New Mexico (after Lovejoy -1976). ∆ indicates Eocene igneous intrusions—Cerro de Cristo Rey, Cerro de la Mina, Campus Andesite, -the Three Sisters, and Vado Hill. -Introduction + PUBLIC SPEAKING SKILLS + +1 + + 1.1 PUBLIC SPEAKING: DEFINITIONS + +UNIT-I + +Public speaking is the process or act of performing a speech to a live audience. This + +process can include preparation, organization, and the delivery of a message that is clear and + +engaging. It can range from a formal speech at a conference to an informal talk at a social + +gathering. + +Historical Perspective + +Public speaking dates back to ancient civilizations, notably the Greeks and Romans. + +In ancient Greece, public speaking was considered an essential skill for citizens. The great + +orator, Aristotle, even wrote a treatise on rhetoric. The Roman tradition also valued public + +oration, with figures like Cicero and Quintilian making significant contributions to the art of + +public speaking. + +1.2 COMPONENTS OF PUBLIC SPEAKING + + Speaker: The person delivering the speech, responsible for preparing and presenting + +the message. + + Message: The main content or idea being communicated. + + Audience: The group of people who receive and interpret the message. + + Channel: The medium through which the message is delivered, whether in person, + +via video, or another format. + + Feedback: The reactions and responses from the audience, which can be verbal or + +non-verbal. + + Context: The situation or environment in which the speech occurs, influencing how + +the message is received. + +2 + + 1.3 NEED FOR PUBLIC SPEAKING + +1.3.1. Personal Development + +Public speaking is crucial for personal growth. It helps build self-confidence, improves + +critical thinking, and enhances communication skills. + +a) Confidence Building + + Regular practice in public speaking situations can significantly increase self- + +confidence. Speaking in front of an audience, even a small one, helps individuals + +become more comfortable with expressing their ideas publicly. + +b) Critical Thinking + + Preparing a speech requires organizing thoughts and structuring them logically. This + +process helps develop critical thinking skills as one must evaluate information, + +formulate arguments, and anticipate audience questions. + +c) Communication Skills + + Public speaking refines both verbal and non-verbal communication skills. Effective + +speakers learn to articulate their thoughts clearly and use body language to reinforce + +their message. + +1.3.2. Professional Growth + +In the professional realm, public speaking skills are invaluable. They contribute to + +career advancement, networking, and leadership. + +a) Career Advancement + + Many career opportunities, especially in leadership roles, require the ability to speak + +confidently and effectively. Employers value employees who can present ideas clearly + +and lead discussions. + +b) Networking Opportunities: + + Public speaking events provide opportunities to meet and connect with peers, + +mentors, and industry leaders. These connections can lead to professional + +collaborations and career opportunities. + +c) Influence and Leadership: + + Effective public speakers can inspire and motivate others. This ability is crucial for + +leadership, as it involves guiding teams, making persuasive arguments, and driving + +organizational change. + +3 + + 1.3.3. Social Impact + +Public speaking plays a significant role in social advocacy, community engagement, + +and education. + +a) Advocacy and Activism: + + Public speakers can use their platform to advocate for social causes, influencing + +public opinion and policy. Activists and leaders often speak out to bring attention to + +important issues. + +b) Community Engagement + + Speaking at community events fosters a sense of belonging and encourages civic + +participation. Engaging with community members through speeches can promote + +unity and collective action. + +Education and Awareness: + + Educators and experts use public speaking to disseminate knowledge and raise + +awareness on various topics. This contributes to an informed and educated society. + +1.4 IMPORTANCE OF PUBLIC SPEAKING + +1.4.1 Effective Communication + +Public speaking is a powerful tool for effective communication, enabling clarity, + +engagement, and persuasion. + +a) Clarity and Precision + + Effective speakers convey their messages clearly and precisely, ensuring the audience + +understands their points. This reduces misunderstandings and enhances the + +effectiveness of communication. + +b) Audience Engagement + + Keeping the audience engaged is crucial for the success of a speech. Techniques like + +storytelling, humour, and interactive elements help maintain interest and involvement. + +c) Persuasion + + The ability to persuade is a key aspect of public speaking. Persuasive speeches aim to + +influence the audience’s beliefs, attitudes, or actions, which is essential in many + +professional and social contexts. + +1.4.2 Professional Relevance + +Public speaking is essential in professional settings, particularly for presentations, + +public relations, and crisis management. + +4 + + a) Presentations and Meetings: + + Many professional roles require presenting information to colleagues, clients, or + +stakeholders. Effective public speaking skills ensure these presentations are clear, + +engaging, and impactful. + +b) Public Relations: + + Managing a company’s or individual’s public image often involves public speaking. + +Delivering statements, handling media inquiries, and speaking at public events are all + +critical PR activities. + +c) Crisis Management + + During a crisis, clear and calm communication is vital. Public speakers must provide + +accurate information, reassure the public, and guide actions to manage the situation + +effectively. + +1.4.3 Societal Contributions + +Public speaking contributes to societal development through civic participation, education, + +and cultural preservation. + +a) Civic Participation + + Public speaking encourages active participation in democratic processes. By speaking + +at town halls, debates, and public forums, individuals can influence policy and + +contribute to community development. + +b) Education and Teaching + + Teachers and educators use public speaking to enhance learning experiences. + +Effective teaching involves clear communication, engaging presentations, and the + +ability to inspire students. + +c) Cultural Preservation + + Public speaking plays a role in preserving and promoting cultural heritage. + +Storytellers, historians, and cultural leaders use speeches to share traditions, history, + +and values. + +5 + + 1.5 ADVANTAGES OF PUBLIC SPEAKING + +1.5.1 Personal Benefits + +Public speaking offers numerous personal advantages, including increased self- + +esteem, networking opportunities, and improved problem-solving skills. + +a) Self-Esteem + + Successfully delivering a speech can boost self-esteem and provide a sense of + +accomplishment. Overcoming the fear of public speaking leads to greater self- + +confidence. + +b) Networking + + Speaking at events allows individuals to meet new people and expand their + +professional and personal networks. These connections can lead to collaborations and + +new opportunities. + +c) Problem-Solving + + Public speaking often involves addressing questions and concerns from the audience. + +This helps develop problem-solving skills as speakers learn to think on their feet and + +provide clear, thoughtful responses. + +1.5.2 Professional Benefits + +In the workplace, public speaking skills can enhance leadership, create career + +opportunities, and lead to professional recognition. + +a) Leadership + + Effective communication is a key leadership quality. Public speaking skills enable + +leaders to articulate their vision, motivate teams, and navigate challenges. + +b) Career Opportunities + + Many job roles, especially in management and leadership, require public speaking + +skills. Being a confident speaker can open doors to promotions and new career paths. + +c) Professional Recognition + + + +Individuals who excel in public speaking often gain recognition and respect in their + +field. This can lead to invitations to speak at conferences, awards, and other + +professional accolades. + +6 + + 1.5.3 Social Benefits + +Public speaking can foster community building, amplify advocacy efforts, and promote + +cultural exchange. + +a) Community Building + + Public speaking events bring people together, fostering a sense of community and + +shared purpose. Speakers can address community issues, inspire collective action, and + +promote social cohesion. + +b) Advocacy + + Speakers can use their platform to advocate for social justice, environmental + +protection, and other important causes. This helps raise awareness and mobilize + +support for these issues. + +c) Cultural Exchange + + Public speaking events can promote cultural exchange and understanding. Speakers + +from diverse backgrounds share their experiences and perspectives, enriching the + +audience’s knowledge and appreciation of different cultures. + +1. 6 SKILLS REQUIRED FOR PUBLIC SPEAKING + +1.6.1 Content Preparation + +Effective public speaking begins with thorough content preparation. This involves + +research, organization, and writing. + +a) Research + + Conducting comprehensive research is essential to gather accurate and relevant + +information. This ensures the speech is informative and credible. + +b) Organization + + Structuring the speech logically helps convey the message clearly. A well-organized + +speech typically includes an introduction, main body, and conclusion. + +c) Writing + + Crafting the speech with attention to language, tone, and style enhances its + +effectiveness. The writing should be clear, concise, and tailored to the audience. + +7 + + 1.6.2 Delivery Techniques + +The way a speech is delivered is just as important as its content. Key delivery + +techniques include vocal variety, body language, and eye contact. + +a) Vocal Variety + + Using different pitches, tones, and volumes can make the speech more engaging and + +emphasize key points. + +b) Body Language + + Gestures, facial expressions, and posture can reinforce the message and convey + +confidence. + +c) Eye Contact + + Maintaining eye contact with the audience helps build a connection and shows + +sincerity. + +1.6.3 Audience Engagement + +Engaging the audience is crucial for an effective speech. Techniques like storytelling, + +interactive elements, and adaptability can help. + +a) Storytelling + + Using stories and anecdotes can illustrate points and make the speech more relatable + +and memorable. + +b) Interactive Techniques + + + +Involving the audience through questions, discussions, and activities keeps them + +engaged and interested. + +c) Adaptability + + Being able to adjust the speech based on audience feedback and reactions is + +important. This may involve altering the delivery or content on the fly to better suit + +the audience. + +1.6.4 Overcoming Anxiety + +Public speaking anxiety is common but can be managed through preparation, practice, + +and mindfulness techniques. + +a) Preparation + + Thoroughly preparing the speech helps boost confidence and reduce anxiety. + +8 + + b) Practice + + Regularly practicing the speech, including in front of a mirror or a small audience, can + +make the speaker more comfortable and familiar with the content. + +c) Mindfulness Techniques + + Techniques like deep breathing, visualization, and mindfulness can help manage + +anxiety and maintain composure during the speech. + +1.6.5 Technical Proficiency + +Technical proficiency involves using visual aids, handling microphones, and + +managing time effectively. + +a) Use of Visual Aids + + + +Incorporating visual aids like slides, videos, and props can enhance the speech and + +make complex information more understandable. + +b) Microphone Handling + + Properly using microphones and other audio equipment ensures the audience can hear + +the speaker clearly. + +c) Time Management + + Managing time effectively is important to cover all points within the allotted time. + +This involves practicing the speech to ensure it fits within the time limit and being + +able to adjust on the fly if needed. + +1.7. CONCLUSION + +In conclusion, public speaking is a multifaceted skill that encompasses various + +aspects, from preparation to delivery. It is an essential tool for personal and professional + +development, offering numerous benefits and opportunities. By developing and refining + +public speaking skills, individuals can enhance their confidence, influence, and impact on + +society. + +9 + + 2.1 Elements of Public Speaking + +UNIT-II + +Public speaking is composed of several critical elements that contribute to the effectiveness + +and success of a speech. Understanding these elements is essential for mastering the art of + +public speaking. + +a) Speaker + +The speaker is the individual delivering the message. Their role is crucial as they are + +responsible for preparing and presenting the speech. A speaker's credibility, confidence, and + +delivery style greatly influence how the message is received. + +b) Message + +The message is the content or main idea conveyed in the speech. It includes the information, + +arguments, and emotions the speaker intends to communicate. Clarity, organization, and + +relevance of the message are key to effective public speaking. + +c) Audience + +The audience is the group of people who receive and interpret the message. Understanding + +the audience's demographics, interests, and expectations helps tailor the speech to engage and + +resonate with them. + +d) Channel + +The channel is the medium through which the message is delivered. This can be face-to-face + +communication, a video presentation, a webinar, or other formats. The choice of channel can + +affect how the message is perceived and received. + +e) Feedback + +Feedback consists of the responses and reactions from the audience, which can be verbal or + +non-verbal. Effective public speakers pay attention to feedback to adjust their delivery and + +ensure their message is being understood. + +f) Context + +Context refers to the environment and circumstances in which the speech is delivered. This + +includes the physical setting, the occasion, and the cultural background. Context can + +influence the tone and content of the speech. + +g) Purpose + +The purpose of the speech is the speaker's goal, whether it is to inform, persuade, entertain, or + +fulfill a specific occasion. Clearly defining the purpose helps in crafting a focused and + +effective message. + +10 + + h) Encoding and Decoding + + Encoding: The process of the speaker translating thoughts and ideas into words and + +gestures. + + Decoding: The audience's interpretation and understanding of the speaker's message. + +Effective communication occurs when the audience accurately decodes the speaker's + +message. + +2.2 TYPES OF PUBLIC SPEAKING + +Public speaking can be categorized into different types based on the purpose and context of + +the speech. Here are the main types: + +a. Ceremonial Public Speaking + +Ceremonial public speaking, also known as special occasion speaking, involves delivering + +speeches at formal events and ceremonies. These speeches are often emotive and personal, + +designed to celebrate, honor, or commemorate. + +Examples: + + Wedding Speeches: Best man or maid of honor speeches, father of the bride speech. + + Funeral Eulogies: Tributes to the deceased, reflecting on their life and legacy. + + Award Acceptance Speeches: Recipients expressing gratitude and acknowledging + +contributions. + +Key Elements: + + Emotional Connection: Establishing a heartfelt connection with the audience. + + Appropriate Tone: Matching the tone to the occasion, whether joyous, solemn, or + +celebratory. + + Brevity: Keeping the speech concise and to the point, respecting the event's flow. + +b. Demonstrative Public Speaking + +Demonstrative public speaking involves showing the audience how to do something. These + +speeches are educational and often include visual aids or live demonstrations to enhance + +understanding. + +Examples: + + Cooking Demonstrations: Showcasing how to prepare a specific dish. + + Technical Tutorials: Explaining how to use a software program or gadget. + + Craft Workshops: Demonstrating steps to create a craft project. + +11 + + Key Elements: + + Clarity: Clearly explaining each step in a logical order. + + Visual Aids: Using tools, props, or multimedia to illustrate points. + + Engagement: Encouraging audience participation or questions to ensure + +comprehension. + +c. Informative Public Speaking + +Informative public speaking aims to educate the audience about a particular topic. The goal is + +to increase understanding and awareness without persuading the audience to adopt a specific + +viewpoint. + +Examples: + + Lectures: Academic presentations on a specific subject. + + Seminars: Professional development sessions on industry-related topics. + + Briefings: Providing updates or reports on specific issues or projects. + +Key Elements: + + Research: Conducting thorough research to provide accurate and detailed + +information. + + Organization: Structuring the speech with a clear introduction, body, and conclusion. + + Neutrality: Presenting facts objectively without expressing personal opinions. + +d. Persuasive Public Speaking + +Persuasive public speaking aims to convince the audience to accept a particular viewpoint or + +take specific actions. This type of speech involves presenting arguments, evidence, and + +emotional appeals to influence the audience. + +Examples: + + Political Speeches: Campaign speeches advocating for policies or candidates. + + Sales Presentations: Pitching products or services to potential customers. + + Advocacy Speeches: Promoting social causes or public policies. + +Key Elements: + + Thesis Statement: Clearly stating the main argument or position. + + Evidence: Providing credible evidence and examples to support the argument. + + Emotional Appeals: Connecting with the audience's emotions to strengthen the + +persuasive impact. + + Call to Action: Encouraging the audience to take specific actions based on the + +speech. + +12 + + 2.3 CONCLUSION + +Public speaking encompasses various elements and types, each with unique + +characteristics and requirements. Understanding the fundamental elements—such as the + +speaker, message, audience, and context—forms the foundation for effective communication. + +Additionally, recognizing the different types of public speaking—ceremonial, demonstrative, + +informative, and persuasive—enables speakers to tailor their approach to suit the purpose and + +context of their speech. By mastering these aspects, individuals can enhance their public + +speaking skills and achieve their communication goals. + +13 + + UNIT-III + +3.1 TECHNIQUES FOR EFFECTIVE PUBLIC SPEAKING + +Effective public speaking is a vital skill in both personal and professional contexts. + +Mastering this art involves various techniques that enhance clarity, engagement, and impact. + +This material covers essential techniques to help you become a more effective and confident + +public speaker. + +3.1.1 Understanding Your Audience + +Audience Analysis + + Demographics: Consider age, gender, education level, cultural background, and + +profession. Understanding these factors helps tailor your message to resonate with the + +audience. + + + +Interests: Identify what the audience cares about and tailor your content to match + +their interests. This helps in making the content relevant and engaging. + + Knowledge Level: Gauge their familiarity with the topic to avoid oversimplification + +or unnecessary complexity. Knowing their knowledge level ensures that you neither + +underestimate nor overestimate their understanding. + + Expectations: Understand what the audience expects to gain from your speech. + +Meeting or exceeding these expectations can significantly enhance the impact of your + +presentation. + +Techniques + + Surveys and Questionnaires: Collect information about the audience before the + +event. This can include questions about their interests, preferences, and knowledge + +level. + + Observation: Pay attention to the audience's reactions and body language. This real- + +time feedback can help you adjust your delivery. + + Direct Interaction: Engage with the audience through questions and discussions + +before the speech. This builds rapport and provides valuable insights into their + +expectations. + +Practical Tips + + Adapt Your Language: Use language that is appropriate for the audience’s level of + +understanding. Avoid jargon or technical terms unless the audience is familiar with + +them. + +14 + +  Relate to the Audience: Share stories or examples that are relevant to the audience's + +experiences. This makes the content more relatable and engaging. + +3.1.2 Preparation and Organization + +Research + + Thorough Investigation: Gather accurate and up-to-date information on your topic. + +This ensures the credibility and relevance of your content. + + Credible Sources: Use reliable and reputable sources for your content. This adds + +authority to your speech and builds trust with the audience. + + Diverse Perspectives: Incorporate multiple viewpoints to enrich your speech. This + +demonstrates a well-rounded understanding of the topic and engages a broader + +audience. + +Structuring Your Speech + + + +Introduction: Capture attention with a strong opening statement, quote, or story. The + +introduction sets the tone and provides a roadmap for your speech. + + Body: Organize main points logically, using clear and concise language. Each point + +should build on the previous one, leading the audience through a coherent argument. + + Conclusion: Summarize key points and end with a powerful closing statement. The + +conclusion should reinforce your message and leave a lasting impression. + +Outlining + + Simple Outline: Main headings and subheadings. This provides a basic structure for + +your speech. + + Detailed Outline: Includes full sentences and transitional phrases. This helps in + +fleshing out your ideas and ensuring smooth transitions between points. + +Practical Tips + + Time Management: Allocate time for each section of your speech. Practice + +delivering within this timeframe to avoid running over. + + Rehearse: Practice delivering your speech multiple times. This helps in internalizing + +the content and improving delivery. + +3.1.3 Developing a Strong Opening + +Techniques + + Anecdotes: Start with a relevant story that sets the stage for your message. Anecdotes + +can be personal or about someone else, but they should be relatable to the audience. + +15 + +  Questions: Pose a thought-provoking question to engage the audience. This + +encourages active thinking and participation. + + Quotes: Use a powerful quote that relates to your topic. A well-chosen quote can set + +the tone and add credibility. + + Statistics: Present a surprising statistic to grab attention. Statistics can highlight the + +importance of your topic and provoke interest. + +Importance + + First Impressions: Establish credibility and set the tone for your speech. A strong + +opening makes a positive first impression. + + Audience Engagement: Captivate the audience from the beginning. An engaging + +opening keeps the audience interested and attentive. + +Practical Tips + + Relevance: Ensure your opening is relevant to the main topic. A disconnected + +opening can confuse the audience. + + Brevity: Keep the opening concise. An overly long opening can lose the audience's + +attention. + +3.1.4 Effective Body Language + +Components + + Gestures: Use hand movements to emphasize points and convey enthusiasm. + +Gestures should be natural and aligned with your words. + + Facial Expressions: Match expressions with the content to convey emotions. + +Smiling, frowning, or showing surprise can reinforce your message. + + Posture: Stand confidently, avoid slouching or leaning on the podium. Good posture + +projects confidence and authority. + + Eye Contact: Maintain eye contact to build a connection with the audience. This + +makes your speech feel more personal and engaging. + +Techniques + + Practice in Front of a Mirror: Observe and adjust your body language. This helps in + +becoming aware of and improving your gestures and expressions. + + Record and Review: Record your practice sessions to identify areas for + +improvement. Watching yourself can reveal habits you might not be aware of. + + Feedback: Seek feedback from peers or mentors. Constructive criticism can provide + +valuable insights for improvement. + +16 + + Practical Tips + + Avoid Distracting Movements: Be mindful of habits like fidgeting or pacing. These + +can distract the audience from your message. + + Consistent Gestures: Use gestures consistently to avoid confusion. Random or + +inconsistent gestures can be distracting. + +3.1.5 Vocal Variety + +Elements + + Pitch: Vary your pitch to avoid monotony and maintain interest. A monotone voice + +can be dull and disengaging. + + Volume: Adjust your volume to suit the venue and emphasize key points. Speaking + +too softly can make it hard for the audience to hear, while speaking too loudly can be + +overwhelming. + + Rate: Vary your speaking rate to highlight important information and maintain + +engagement. Speaking too fast can confuse the audience, while speaking too slowly + +can bore them. + + Pauses: Use pauses effectively to let key points sink in and to create dramatic effect. + +Pauses can also give you a moment to collect your thoughts. + +Techniques + + Breathing Exercises: Practice deep breathing to control your voice. Proper breathing + +supports vocal strength and clarity. + + Voice Modulation Exercises: Work on pitch, tone, and volume variations. Exercises + +like humming, lip trills, and tongue twisters can improve your vocal range and + +flexibility. + + Reading Aloud: Practice reading aloud to improve vocal variety and articulation. + +This helps in developing a more dynamic and engaging speaking style. + +Practical Tips + + Record Yourself: Record your practice sessions to analyze your vocal variety. + +Listening to yourself can help you identify areas for improvement. + + Seek Feedback: Get feedback from others on your vocal delivery. Constructive + +criticism can help you refine your vocal technique. + +3.1.6 Engaging the Audience + +17 + + Techniques + + + +Interactive Elements: Use polls, Q&A sessions, and discussions. Interaction keeps + +the audience involved and attentive. + + Stories and Examples: Share relevant anecdotes and case studies to illustrate points. + +Stories make abstract concepts more relatable and memorable. + + Humor: Incorporate appropriate humor to make the speech enjoyable and relatable. + +Humor can lighten the mood and make your speech more engaging. + + Visual Aids: Use slides, charts, and videos to support your message and keep the + +audience engaged. Visual aids can clarify complex information and maintain interest. + +Importance + + Attention Retention: Keeps the audience focused and interested. Engaged audiences + +are more likely to absorb and retain information. + + Enhanced Understanding: Helps the audience grasp complex concepts. Interaction + +and visual aids can make your message clearer and more accessible. + +Practical Tips + + Relevance: Ensure that your engagement techniques are relevant to your content. + +Irrelevant interaction can be distracting. + + Balance: Balance interactive elements with your main content. Too much interaction + +can detract from your core message. + +3.17 Overcoming Anxiety + +Techniques + + Preparation: The more prepared you are, the more confident you will feel. Thorough + +preparation reduces uncertainty and boosts confidence. + + Practice: Regular practice reduces anxiety and increases familiarity with the content. + +The more you practice, the more comfortable you become. + + Visualization: Visualize a successful speech to build confidence. Imagine yourself + +delivering the speech successfully and receiving positive feedback. + + Relaxation Techniques: Use deep breathing, meditation, and positive affirmations. + +These techniques can help calm your mind and reduce physical symptoms of anxiety. + +Managing Nervous Energy + + Physical Activity: Light exercise before speaking can reduce tension. Activities like + +stretching or a short walk can help calm nerves. + +18 + +  Hydration: Drink water to stay hydrated and prevent a dry mouth. Avoid caffeine, + +which can increase anxiety. + + Positive Thinking: Focus on positive outcomes and remind yourself of past + +successes. Positive affirmations can boost your confidence. + +Practical Tips + + Arrive Early: Arrive early to familiarize yourself with the venue. This reduces + +uncertainty and helps you feel more comfortable. + + Connect with the Audience: Engage with audience members before the speech. + +Building rapport can make you feel more at ease. + +3.1.8 Effective Use of Visual Aids + +Types of Visual Aids + + Slides: PowerPoint or Keynote presentations. Slides should be clear, concise, and + +visually appealing. + + Charts and Graphs: Illustrate data and trends. Charts and graphs can make complex + +data more understandable. + + Videos: Short clips to support or highlight points. Videos can add variety and + +illustrate points more vividly. + + Props: Physical objects that enhance understanding. Props can make abstract + +concepts more concrete and engaging. + +Techniques + + Simplicity: Keep slides and visuals simple and uncluttered. Avoid overwhelming the + +audience with too much information. + + Relevance: Ensure visual aids are directly related to your content. Irrelevant visuals + +can distract and confuse the audience. + + Practice: Familiarize yourself with the visual aids and integrate them smoothly into + +your speech. Practicing with your visual aids ensures a seamless presentation. + +Common Pitfalls + + Overreliance: Don't let visual aids overshadow your speech. Visual aids should + +support, not replace, your message. + + Technical Issues: Be prepared for potential technical difficulties. Have a backup plan + +in case of technical failures. + +19 + + Practical Tips + + Backup Copies: Have backup copies of your visual aids. This ensures you can + +continue your presentation even if technical issues arise. + + Engage with Visuals: Interact with your visual aids to maintain audience + +engagement. Pointing out specific details or elaborating on visuals can enhance + +understanding. + +3.1.9 Handling Questions and Interactions + +Techniques + + Anticipate Questions: Predict potential questions and prepare answers. This helps + +you feel more confident and prepared. + + Active Listening: Listen carefully to questions and respond thoughtfully. Active + +listening shows respect and helps you provide accurate answers. + + Clarification: Ask for clarification if a question is unclear. This ensures you + +understand the question and can provide a relevant response. + + Confidence: Maintain composure and confidence when answering questions. + +Confidence reassures the audience of your knowledge and credibility. + +Strategies + + Repetition: Restate the question to ensure understanding and give yourself time to + +think. Repeating the question also ensures everyone in the audience heard it. + + Bridging: Connect your answer to your main points or message. This keeps your + +response relevant and reinforces your key messages. + + Acknowledgment: Acknowledge challenging questions and provide honest, + +thoughtful responses. Honesty builds trust and credibility. + +Practical Tips + + Stay Calm: Stay calm and composed, even if faced with difficult questions. Keeping + +your cool shows professionalism. + + Admit When You Don’t Know: It’s okay to admit when you don’t know an answer. + +Offer to follow up later with more information. + +3.1.10 Continuous Improvement + +Techniques + + Self-Reflection: After each speech, reflect on what went well and what could be + +improved. Self-reflection helps you identify strengths and areas for growth. + +20 + +  Feedback: Seek feedback from peers, mentors, and the audience. Constructive + +feedback provides valuable insights for improvement. + + Training and Courses: Enroll in public speaking courses or workshops. Formal + +training can provide new techniques and enhance your skills. + + Practice: Regularly practice public speaking in various settings. Practice helps you + +refine your skills and gain confidence. + +Setting Goals + + Specific Goals: Set specific, measurable goals for improvement. Clear goals provide + +direction and motivation. + + Track Progress: Keep a record of your speeches and progress. Tracking progress + +helps you see improvements and stay motivated. + + Celebrate Achievements: Recognize and celebrate milestones and improvements. + +Celebrating successes reinforces positive behavior and boosts confidence. + +Practical Tips + + Join Speaking Groups: Join groups like Toastmasters to practice and receive + +feedback. Regular practice and feedback from peers can significantly improve your + +skills. + + Record Your Speeches: Record your speeches to analyze and improve your delivery. + +Watching recordings helps you identify areas for improvement. + +3.2 Conclusion + +Effective public speaking involves mastering various techniques that enhance clarity, + +engagement, and impact. By understanding your audience, preparing thoroughly, using strong + +openings, employing effective body language and vocal variety, engaging the audience, + +overcoming anxiety, utilizing visual aids effectively, handling questions with confidence, and + +committing to continuous improvement, you can become a more effective and confident + +public speaker. + +21 + + 4.1 METHODS OF PUBLIC SPEAKING + +UNIT-IV + +Public speaking is a critical skill that can be used in various contexts, ranging from + +casual conversations to formal presentations. Different methods of public speaking are suited + +for different scenarios, and mastering these methods can significantly enhance your ability to + +communicate effectively. This material covers key methods of public speaking, providing + +detailed insights and practical tips for each. + +4.1.1 Impromptu + +Definition + +Impromptu speaking involves delivering a speech without prior preparation. This method is + +often required in spontaneous situations where quick thinking and adaptability are essential. + +Techniques + + Think Quickly: Organize your thoughts rapidly and structure them logically. + + Stay Calm: Maintain composure to think clearly and speak confidently. + + Use Personal Experiences: Draw from your own experiences and knowledge to + +support your points. + + Practice Regularly: Engage in impromptu speaking exercises to improve your skills. + +Advantages + + Flexibility: Allows you to adapt to the topic and audience on the spot. + + Authenticity: Speeches often sound more genuine and spontaneous. + +Disadvantages + + Lack of Structure: Can lead to disorganized thoughts and unclear messages. + + + +Increased Anxiety: Higher pressure due to lack of preparation. + +Practical Tips + + Practice Thinking on Your Feet: Engage in activities like debate clubs or + +improvisational theater. + + Stay Updated: Keep yourself informed about various topics to have a broad + +knowledge base to draw from. + +4.1.2 Manuscript Speaking + +22 + + Definition + +Manuscript speaking involves reading a fully written speech. This method is commonly used + +in formal settings where accuracy and precise wording are crucial, such as political speeches + +or legal presentations. + +Techniques + + Write Clearly: Ensure the manuscript is well-written and easy to read. + + Practice Reading: Rehearse reading the manuscript to maintain eye contact and + +natural delivery. + + Use Visual Aids: Supplement the manuscript with visual aids to enhance engagement. + + Control Your Pace: Read at a moderate pace to ensure clarity and comprehension. + +Advantages + + Precision: Ensures accurate delivery of complex information. + + Consistency: Provides a consistent message every time. + +Disadvantages + + Less Engagement: Can sound monotonous and disengage the audience. + + Limited Flexibility: Difficult to adapt to audience reactions or questions. + +Practical Tips + + Highlight Key Points: Use highlights or underlines to emphasize important sections + +in your manuscript. + + Practice with a Timer: Ensure your speech fits within the allotted time by practicing + +with a timer. + +4.1.3 Memorized Speaking + +Definition + +Memorized speaking involves delivering a speech from memory without any notes or + +manuscripts. This method is often used for speeches that require exact wording, such as toasts + +or keynote addresses. + +Techniques + + Break It Down: Divide the speech into smaller sections for easier memorization. + + Use Mnemonics: Employ mnemonic devices to aid memory. + + Rehearse Frequently: Practice regularly to reinforce memory. + + Visualize Success: Visualize delivering the speech confidently to build memory and + +reduce anxiety. + +23 + + Advantages + + Engagement: Allows for direct eye contact and connection with the audience. + + Freedom of Movement: Enables more natural gestures and movement. + +Disadvantages + + Risk of Forgetting: High risk of forgetting parts of the speech. + + + +Inflexibility: Difficult to adapt on the fly if something goes wrong. + +Practical Tips + + Daily Practice: Set aside time each day to rehearse your speech. + + Use Visual Cues: Associate parts of your speech with specific visual cues to aid + +memory. + +4.1.4 Extemporaneous Speaking + +Definition + +Extemporaneous speaking involves delivering a speech from brief notes or an outline. This + +method balances preparation and spontaneity, allowing for a structured yet natural delivery. + +Techniques + + Prepare Thoroughly: Create a detailed outline with key points. + + Practice Extensively: Rehearse using the outline to become familiar with the content. + + Stay Flexible: Be ready to adapt to audience reactions and questions. + + Use Conversational Tone: Aim for a conversational tone to maintain natural delivery. + +Advantages + + Balance: Combines the benefits of preparation and spontaneity. + + Engagement: Allows for eye contact and audience interaction. + +Disadvantages + + Requires Skill: Demands good organization and speaking skills. + + Potential for Gaps: Risk of missing key points if not well-prepared. + +Practical Tips + + Note Cards: Use small note cards with key points to guide you. + + Engage with the Audience: Incorporate audience feedback and questions into your + +speech. + +4.1.5 Persuasive Speaking + +24 + + Definition + +Persuasive speaking aims to convince the audience to accept a particular viewpoint or take a + +specific action. This method is commonly used in sales pitches, political speeches, and + +advocacy campaigns. + +Techniques + + Clear Argument: Present a clear and logical argument. + + Evidence and Examples: Use evidence and examples to support your points. + + Emotional Appeal: Appeal to the audience's emotions to strengthen your argument. + + Call to Action: End with a clear call to action, urging the audience to act on your + +message. + +Advantages + + + +Influence: Can effectively change opinions or behaviors. + + Engagement: Often highly engaging and impactful. + +Disadvantages + + Resistance: Audience may resist if they feel manipulated. + + Complexity: Requires strong argumentative and rhetorical skills. -Cerro de Cristo Rey is a prominent peak in -Doña Ana County, New Mexico, just north -of the U.S.–Mexican border and just west of -the city of El Paso, Texas (Fig. 1). The moun- -tain was long referred to as the Cerro de -Muleros, but was renamed Cerro de Cristo -Rey (“Sierra” de Cristo Rey of Hook 2008 -and Cobban et al. 2008, also appears on the -U.S. Geological Survey topographic map of -the area) in recognition of the large statue of -Christ on the cross at its crest. The mountain -has an elevation of 1,425 m (4,675 ft) (Love- -joy 1976; Hook 2008). - -The core of Cerro de Cristo Rey is an -andesite laccolith, the Muleros Andes- -ite of Eocene age (Fig. 2). The laccolith is -surrounded by “an annulus of…faulted - -Cretaceous marine strata, locally strongly -deformed by gravity-glide structures trig- -gered by andesite intrusion” (Lovejoy 1976, -p. 24). The Cretaceous strata, a section -approximately 350 m thick, include rocks -of Early Cretaceous (late Albian) and Late -Cretaceous (early–middle Cenomanian) -age (Figs. 3–4). These rocks have been stud- -ied for more than a century, most notably -by Böse (1910) and Strain (1976). Here, we -present the first detailed lithostratigraphy -and sedimentary petrography of the Cre- -taceous strata exposed around the New -Mexican periphery of the Cerro de Cristo -Rey uplift. We combine these data with -paleontology and regional correlations to -present the first detailed interpretation of -the depositional environments and cycle -stratigraphy of the Cretaceous section. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -103 - - the Cretaceous invertebrate macrofossils -he collected at Cerro de Cristo Rey. - -Adkins (1932) and Imlay (1944) sub- -sequently applied Texas stratigraphic -names to the section Böse (1910) delin- -eated at Cerro de Cristo Rey (Fig. 3). Cór- -doba (1968, 1969a,b) proposed a formal -lithostratigraphic nomenclature for the -Cretaceous section in northeastern Chi- -huahua, Mexico. He mapped these units -as far north as the international border, so -he assigned Cretaceous strata along the -southern flank of Cerro de Cristo Rey in -Chihuahua to strata of his Ojinaga Group. -Córdoba (1969a) makes clear the equiva- -lence between his stratigraphic names and -those introduced at Cerro de Cristo Rey by -Strain (1968, 1976). - -The lithostratigraphic names long -applied to Cretaceous strata at Cerro de -Cristo Rey were briefly introduced by -Strain (1968) and elaborated on by Strain -(1976) and Lovejoy (1976). Thus, four strati- -graphic names from farther east, in Texas -(Finlay, Del Rio, Buda, and Boquillas), are -applied, whereas Strain introduced five -local lithostratigraphic names—Del Norte, -Smeltertown, Muleros, Mesilla Valley, and -Anapra (Figs. 3–4). - -Subsequent workers (e.g., LeMone and -Simpson 1983; Maudlin 1985; Cornell and -LeMone 1987; Cornell et al. 1991; Cornell -1997; Lucas and Estep 1998b, 2000; Scott -et al. 2001, 2003; Turnšek et al. 2003) have -employed the lithostratigraphy of Strain -(1976). The only previously suggested -modification is that of Lucas and Estep -(1998b, 2000), who argued that the Anapra -Sandstone is part of the same lithosome as -the Mojado Formation to the west. They -therefore abandoned the name Anapra -Sandstone and replaced it with the term -Mojado Formation. Here, we employ the -lithostratigraphic nomenclature of Strain -(1968, 1976) with two modifications, -the use of Mojado Formation instead of -Anapra Sandstone and the use of Mancos -Formation instead of Boquillas Formation -(Fig. 4). These modifications are justified -below. - - Del Norte - -FIGURE 2—Geologic map of part of the New Mexican portion of the Cerro de Cristo Rey uplift (from -Lovejoy 1976) showing locations of measured stratigraphic sections in this study. See Appendix for -map coordinates of the measured sections. - -Located on the northern margin of the -Chihuahua trough, the Cretaceous section -at Cerro de Cristo Rey is between Bisbee -Group sections to the west and Washita -Group sections to the east. It thus provides -a pivotal reference section for regional cor- -relation and testing concepts of Washita -Group cycle stratigraphy. - -Previous studies - -Emory (1857), Stanton and Vaughan (1896), -and Richardson (1909) first reported on -the Cretaceous rocks exposed around -Cerro de Cristo Rey (Fig. 3). However, the -real starting point for understanding the - -stratigraphy and paleontology of these -strata was the monograph of Böse (1910). -He informally divided the Cretaceous -section at Cerro de Cristo Rey into 11 -stratigraphic units, many characterized -by a distinctive fossil assemblage (Fig. 3). -Although Böse (1910) explicitly correlated -these units to formally named units of the -Fredericksburg and Washita Groups and -overlying Eagle Ford Shale of Texas (as -had Stanton and Vaughan [1896] and was -later done by Adkins [1932]), he applied -no formal lithostratigraphic nomencla- -ture to the Cretaceous section at Cerro de -Cristo Rey. Most significantly, Böse (1910) -provided a monographic description of - -FIGURE 3—Evolution of Cretaceous lithostratigraphic nomenclature at Cerro de Cristo Rey. Modified from Strain (1976). - -104 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - and serpulids are also present (Böse 1910; -Strain 1976; Turnšek et al. 2003; Kollmann -et al. 2003). In addition, our thin sections -(see below) demonstrate that calcareous -algae, ostracods, and rare calcisponges are -present. Kollmann et al. (2003) recently -analyzed the gastropod fossil assem- -blage of the Finlay Formation at Cerro de -Cristo Rey, identifying the following taxa: -Aporrhaidae, Diozoptyxis, Drepanocheilus, -Eunerinea aquiline, Naticoidea, Nerinella -n. sp., Otostoma elpasensis, Plesioptyxis tex- -anus, Rostroptygmatis kervillensis, Turritella, -and Tylostoma. - -Sedimentary petrography (Figs. 6, 16) - -Finlay Formation limestone beds at Cerro -de Cristo Rey are mostly bioclastic wacke- -stone that is fine-grained and commonly -bioturbated. Fossils include small mollusc -fragments (bivalves, gastropods), ostra- -cods, calcareous algae, a few echinoderms, -echinoid spines, and many foraminiferans -embedded in a pelmicritic matrix (Fig. 16A– -C). Rare calcisponges are present. Locally, -the wackestone grades into packstone and -rarely into washed grainstone containing -calcite cement (Fig. 16D–F). Many bioclasts -are recrystallized. Common foraminiferans -are Dictyoconus (particularly in units 2 and -8; Fig. 16D–F) and miliolids, Nummoloculina, -Cornuspira, and Ophthalmidium. Most algae -are dasycladacean algae (Permocalculus) and -aciculariaceans (Acicularia or Terquemella). -This facies corresponds to Facies F (bioclas- -tic dasycladacean wackestone) of Kollmann -et al. (2003). - -In the Finlay Formation, beds of inter- -calated nodular limestone are 0.2–5 m -thick and composed of limestone nodules -embedded in shale. The limestone nodules -consist of bioturbated mudstone contain- -ing a few larger mollusc shell fragments -(bivalves, gastropods) and small bioclasts -derived from molluscs, ostracods, rare -echinoderms, echinoid spines, and smaller -foraminiferans. - -Depositional environments - -The Finlay Formation section exposed at -Cerro de Cristo Rey is very similar to the -upper part of the Finlay Formation at its stra- -totype section in Hudspeth County, Texas -(Steinhoff 2003). Thus, the Finlay Formation -strata at Cristo Rey are limestones and nodu- -lar limestones largely devoid of coarse silici- -clastic material. Thin-shelled bivalves and -spatangoid echinoids dominate the macro- -fossil assemblage, and the benthic foraminif- -eran Dictyoconus walnutensis is particularly -abundant. Deposition clearly took place on -a shallow-marine platform in a relatively -restricted circulation shelf environment, and -this fits the regional picture of upper Finlay -deposition (Scott and Kidson 1977; Reaser -and Malott 1985; Steinhoff 2003). - -Microfacies and the diverse biota, includ- -ing broken mollusc fragments in the wacke- -stone to packstone of the massive limestone - -FIGURE 4—Summary of stratigraphic section and age assignments of Cretaceous strata at Cerro de -Cristo Rey (modified from Lucas and Estep 1998b, 2000). Only ammonoid zone names in boldface are -known from ammonoid fossils collected at Cerro de Cristo Rey; presence of other zones is inferred. - -Overview - -Finlay Formation - -Below, we present a brief summary of the -lithostratigraphy of the Cretaceous rocks -exposed at Cerro de Cristo Rey (Fig. 4), -based primarily on the stratigraphic sec- -tions we measured (Figs. 5–15) and on data -in Strain (1976) and Lovejoy (1976). More -detailed lithologic information on some -of the rock units is presented below in -the sections on sedimentary petrography -(Figs. 16–24). Note that we refer to all the -Cretaceous formation rank units exposed -at Cerro de Cristo Rey with the suffix “for- -mation,” instead of using the lithic terms -limestone, shale, or sandstone. - -Böse (1910) published most of what is -known of the Cretaceous paleontology of -Cerro de Cristo Rey. We have extensive -macrofossil collections from Cerro de Cris- -to Rey that we will document elsewhere. -Here, we provide brief summaries of the -paleontology, mostly of relevance to inter- -preting the age and depositional environ- -ments of the Cretaceous strata exposed at -Cerro de Cristo Rey. We also present a sedi- -mentary petrography of the Cretaceous sec- -tion at Cerro de Cristo Rey, and we briefly -summarize the depositional environments -of the Cretaceous formations. The final sec- -tion of this article brings these data and -analyses together to present an interpreta- -tion of the cycle stratigraphy of the Creta- -ceous strata exposed at Cerro de Cristo Rey. - -Lithostratigraphy - -The Finlay Formation is the stratigraphi- -cally lowest unit of the Cretaceous sec- -tion exposed in the New Mexico outcrops -at Cerro de Cristo Rey, but its base is not -exposed. Lovejoy (1976) reported a thick- -ness of 40 m, and total exposed thickness of -our measured section of the Finlay Forma- -tion is approximately 41 m (Figs. 5A, 6). The -only outcrop of the Finlay Formation on the -New Mexico flank of the uplift is located at -our measured section (Fig. 2). - -Finlay Formation strata are a conspicu- -ous ledge- and cliff-forming interval of -interbedded light-gray and white lime- -stone and nodular limestone in a marly -matrix (Figs. 5A, 6). The succession is -thus composed of alternating fossilifer- -ous limestone ledges and slope-forming -nodular limestone in shale. Limestone -ledges are 0.4–5.3 m thick and composed -of bioclastic wackestone to packstone and -some rare grainstone. - -Paleontology - -Bivalves and gastropods are the most -common fossils in the Finlay Formation, -but foraminiferans (especially Dictyoco- -nus walnutensis: Figs. 16D–F, 25R–S), cor- -als, ammonoids, spatangoid echinoderms, - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -105 - - FIGURE 5—Photos of selected Cretaceous outcrops at Cerro de Cristo Rey. -A—Overview of Finlay Formation section (Fig. 6). B—Overview of part of -lower member of Del Norte Formation (Fig. 7). C—Overview of middle part -of Smeltertown Formation (Fig. 8). D—Overview of characteristic outcrop - -of Muleros Formation (Fig. 9). E—Texigryphaea packstone in lower part of -Muleros Formation. F—Overview of shale slope formed by Mesilla Valley -Formation capped by sandstone at base of Mojado Formation (Fig. 10). - -ledges, indicate deposition in an open-shelf -setting with moderate energy. The presence -of dasycladacean algae points to deposition -within the photic zone. Intercalated grain- -stone suggests periodically high-energy -conditions. The wavy to nodular bedded -limestone, composed mainly of bioturbated -mudstone with a less diverse fauna, sug- -gest deposition in a low-energy, shallow- -marine shelf environment. - -The facies of the massive and wavy to -nodular limestone is very similar to that - -of the stratotype section of the Finlay For- -mation, although some differences exist. -According to Steinhoff (2003), at the strato- -type the lower part of the Finlay Formation -is composed of nodular to massive-bedded -limestone (bioclastic wackestone to pack- -stone) interbedded with marl and clay. In the -upper part of the Finlay Formation massive- -bedded limestone (bioclastic wackestone to -packstone with a slightly different fauna) -is abundant. Rudists, which are common -in the stratotype section, particularly in the - -upper part regionally (e.g., Steinhoff 2003), -are absent at Cristo Rey. The lower Finlay -Formation of the stratotype is composed of -deposits formed on a carbonate ramp in a -shallow open-marine shelf environment, -whereas the upper Finlay Formation repre- -sents deposits of a restricted circulation shelf -environment (Steinhoff 2003). Such a trend -from a carbonate ramp in the lower Finlay to -a restricted shelf environment in the upper -Finlay is not evident at Cristo Rey, although -some shallowing occurred. - -106 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - The cyclic alternation of massive and -wavy to nodular limestone evident in the -Finlay Formation section at Cerro de Cristo -Rey (Fig. 6) is also recognized at the Fin- -lay Formation stratotype, where Steinhoff -(2003) described nine transgressive-regres- -sive high-frequency cycles in the lower -high-frequency sequence 1, and 18 cycles in -the overlying high-frequency sequence 2. -The stratotype section of the Finlay Forma- -tion at the Rimrock escarpment, west Texas, -is approximately 52 m thick, although the -thickness of the Finlay Formation var- -ies from less than 15 m to approximately -800 m and more in the deeper parts of its -depositional basin, the Chihuahua trough -(Steinhoff 2003). - -Del Norte Formation - -Lithostratigraphy - -At Cerro de Cristo Rey, the Del Norte For- -mation of Strain (1968, 1976) rests with sharp -contact on the Finlay Formation (Figs. 6–7). It -is subdivisions 2 and 3 of Böse (1910), which -he estimated to be 20–30 m thick. Our mea- -sured section of the Del Norte Formation -is the type section of Strain (1976, table 8), -where he reported the formation as approxi- -mately 17 m thick. However, our section of -the Del Norte Formation is 24 m thick and, -following Strain (1976), can be divided into -a lower (clay) member (approximately 20 m -of shale with interbedded sandstone and -limestone) and an upper (calcareous) mem- -ber (3–4 m of limestone) (Fig. 7). - -The basal bed of the Del Norte Forma- -tion is a prominent, 0.1-m-thick ferruginous -sandstone bed that rests with sharp contact -on underlying lime mudstone of the Fin- -lay Formation (Figs. 6–7). Above this bed, -the lower member (“Brick Plant Member” -of Strain 1968) of the Del Norte Formation -forms a slope above the Finlay Formation -(Fig. 5B) and is composed of alternating -beds of limestone, siltstone to fine-grained -sandstone, and shale. Limestone beds -are thin (0.2–0.6 m thick) and form ledges -(Fig. 5B). The lower two limestone beds are -nodular. Intercalated beds of calcareous silt- -stone to fine-grained sandstone in the lower -member are 0.3 m thick (Fig. 7, unit 8) and -2.1 m thick (Fig. 7, unit 16). The lower mem- -ber is capped by the cuesta- and ridge-form- -ing upper member (“Refinery Member” of -Strain 1968) of the Del Norte Formation, -which is 3.5 m thick. A 0.6-m-thick sandy -and shelly limestone ledge (Fig. 7, unit 19) -forms its base. - -Strain (1976, p. 78) stated that the “Lower -Clay Member [of the Del Norte Formation] -is conformable with the Finlay Limestone -below.” However, we agree with Scott et al. -(2001, 2003) and Turnšek et al. (2003) that -the base of the Del Norte Formation at -Cerro de Cristo Rey marks the basal uncon- -formity of the Washita Group (see section -on cycle stratigraphy, page 127). - -FIGURE 6—Measured stratigraphic section of Finlay Formation. See Figure 2 and Appendix for loca- -tion of section. - -Kollmann et al. (2003, fig. 9) interpreted -the facies succession of the exposed Finlay -Formation at Cerro de Cristo Rey as that of a -shelf lagoon intercalated, near the top of the -Finlay, with a bioherm fringe facies repre- -sented by “Orbitolina wackestone.” Accord- -ing to Kollmann et al. (2003), this bioherm -fringe facies is approximately 8–15 m below - -the top of the Finlay, so it corresponds to -units 20–24 of our measured section of the -Finlay Formation (Fig. 6). However, we did -not observe this “Orbitolina wackestone” -facies. “Orbitolina” (we take this to refer to -Dictyoconus) is present throughout the sec- -tion and is abundant also in the lower part -(particularly in units 2 and 8). - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -107 - - Sedimentary petrography (Figs. 7, 17) - -The basal sandstone of the Del Norte For- -mation (Fig. 7) is fine-grained, mixed silici- -clastic and carbonate in composition, con- -taining abundant angular quartz grains, -a few micas, and (probably) some detrital -feldspars. Carbonate grains are abundant. -A few fossil fragments such as shell debris, -ostracods, echinoderms, and dasyclad- -acean algae are present. The sandstone is -well sorted and laminated. - -Nodular limestone of the Del Norte For- -mation is fine-grained, poorly sorted, bio- -turbated bioclastic wackestone (Fig. 17A, -F) containing abundant fragments of -calcareous algae (Aciculariacea, Gym- -nocodiacea, ?Permocalculus). Mollusc shell -fragments (gastropods, bivalves), ostra- -cods, foraminiferans, and echinoid spines -embedded in gray, micritic, locally pelmic- -ritic matrix are subordinate. Algae and -molluscs are very greatly fragmented and -recrystallized. Rarely, small voids (?fossils) -filled with calcite spar are present. - -Limestone of the Del Norte Formation -is fine- to coarse-grained, poorly sorted, -and commonly bioturbated wackestone to -packstone (Fig. 17C, E). A few large mol- -lusc fragments several cm in size float in the -wackestone. Mollusc fragments (bivalves, -gastropods) and echinoderms are abundant. -Echinoid spines, ostracods, foraminiferans -(including some large, agglutinated forms), -and bryozoans(?) are subordinate. Fossils, -particularly molluscs, are commonly recrys- -tallized and strongly fragmented. Lithoclasts -include small detrital quartz grains that are -mostly 0.1–0.5 mm in diameter (rarely to -1 mm), constituting as much as about 20% -of the rock, and a few peloids and micritic -intraclasts. Rare glauconite grains are pre- -sent. The matrix is micrite. - -Siltstone to fine-grained sandstone of the -Del Norte Formation is commonly non- -laminated and bioturbated, and locally -indistinctly laminated (Fig. 17B, D). The -composition is mixed siliciclastic-carbonate -with as much as about 40% angular detrital -quartz grains that are mostly 0.1–0.2 mm in -diameter. Other lithoclasts are peloids, mic- -ritic intraclasts, and local fecal pellets with -diameters from 0.3 to 0.5 mm. The most -common fossil fragments are derived from -bivalves and gastropods, which are rarely -encrusted by bryozoans(?) and annelid -worms (serpulid worm tubes). Less abun- -dant are echinoderms, ostracods, foraminif- -erans (including rare agglutinated forms), -echinoid spines, and bryozoans(?). - -The basal limestone ledge of the upper -member of the Del Norte Formation (Fig. 7, -unit 19) is composed of poorly sorted bio- -clastic wackestone to rudstone containing -large, broken fragments of bivalves. Other -skeletons are echinoderms, gastropods, -ostracods, foraminiferans (including agglu- -tinated forms), echinoid spines, and dasy- -cladacean algae. A few intraclasts and many -detrital quartz grains are present. Locally, -the grains are densely packed, grading into -packstone. The matrix is micrite. - -This basal limestone is overlain by two -nodular limestone beds (Fig. 7, units 20, 22, -forming recessed notches) alternating with -limestone ledges. The microfacies is similar -to that of the basal limestone ledge: poorly -sorted wackestone to packstone and, in -part, floatstone containing fragmented and -abraded skeletons. Most abundant are mol- -lusc shell fragments derived from bivalves, -gastropods, and echinoderms. Echinoid -spines, ostracods, foraminiferans (includ- -ing some agglutinated forms), and bryo- -zoans are subordinate. Angular to rounded -detrital quartz grains, mostly 0.2–0.5 mm -in diameter and rarely to 1 mm, and a few -micritic intraclasts are present. The grains -are embedded in micritic matrix. - -Depositional environments - -Fossils and sediments of the Del Norte For- -mation indicate deposition in a shallow, -open-marine environment with periodical- -ly strong siliciclastic influx. The presence -of calcareous algae suggests deposition -within the photic zone. Sandstone beds in -the lower part displaying horizontal lami- -nation and current ripples are interpreted -to represent storm layers. Thin sandstone -interbeds displaying horizontal lamina- -tion and ripple lamination intercalated -in shale of a shallow-marine shelf setting -have commonly been attributed to storm -activity. Thus, Scott et al. (1975) described -similar siltstone to fine-grained sandstone -beds 1–10 cm thick in the Washita Group -of Texas, which they interpreted as density -currents formed by storms, tides, rip flow, -or floods. Hobday and Morton (1984) inter- -preted similar thin sandstone intercalations -in the Lower Cretaceous Grayson Forma- -tion (Texas) as deposits of strong bottom -currents below the storm wave base gener- -ated by large-scale, storm-generated, bot- -tom-return flow. - -Nodular limestone, which is commonly -bioturbated and contains micritic to pel- -micitic matrix, formed in a low-energy -environment, whereas intercalated lime- -stone beds containing strongly fragmented, -locally densely packed mollusc fragments -(coquina) indicate deposition under high- -energy conditions and may represent storm -layers. The sharp contact with the under- -lying Finlay Formation is interpreted as a -transgressive surface, which was followed -by deepening upward and a regressive -trend in the upper member of the Del Norte -Formation, which resulted in the deposi- -tion of limestone with abundant coquina -storm layers. - -Smeltertown Formation - -Lithostratigraphy - -The Smeltertown Formation of Strain (1976) is -mostly gray shale with limestone nodules that -forms a relatively thick slope (approximately - -FIGURE 7—Measured stratigraphic section of -Del Norte Formation. See Figure 2 and Appen- -dix for location of section, and Figure 6 for lithol- -ogy legend. - -Paleontology - -The Del Norte Formation has a diverse fossil -assemblage of bivalves (especially grypha- -eids, exogyrids, pectens, and oysters), gas- -tropods, spatangoid echinoderms, ammo- -noids, corals, sponges, and serpulids (Böse -1910; Strain 1976; Turnšek et al. 2003). In thin -section, we observed dasycladacean algae, -ostracods, foraminferans, echinoderms, and -serpulid worms (see below). Ammonites -of the upper Albian Adkinsites bravoensis -Zone (of Young 1966) are present in the -lower member of the Del Norte Forma- -tion, in units 15–18 of our measured section -(Fig. 7), which indicates correlation with the -Kiamichi Formation in north-central Texas. -No fossils diagnostic of the Eopachydiscus -marcianus or Craginites serratescens ammo- -nite zones are present at Cerro de Cristo Rey, -so their presence (Fig. 4) can only be inferred -in the upper Del Norte Formation or lower -Smeltertown Formation or missing at an -unconformity. - -108 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - 60 m) above the upper member of the Del -Norte Formation (Figs. 5C, 8). Our measured -section of the Smeltertown Formation (Fig. 8) -is the type section of Strain (1976, table 9), -which he reported to be approximately -61 m thick. The Smeltertown Formation -is subdivision 4 of Böse (1910), which he -estimated as 30–50 m thick. - -The base of the Smeltertown Formation -is dark-colored shale that sits directly on -limestone at the top of the Del Norte For- -mation (Figs. 7–8). In the lower part of the -Smeltertown Formation (Fig. 8, unit 3), -limestone nodules as much as 20 cm in -diameter are in a 3-m-thick shale inter- -val. Sparse limestone nodules are also in -shale units 4 and 6. Shale unit 5 contains -thin ledges of limestone nodules and a -pebbly agate zone near the top. This unit -also contains many ammonites, especially -specimens of Mortoniceras equidistans. The -limestone ledge of unit 7 yields many -shells of Texigryphaea. In unit 8, thin lime- -stone ledges are intercalated in shale. A -2.3-m-thick interval of sandstone ledges, -mostly full of abundant tests of the large, -arenaceous foraminiferan Cribratina texana -and intercalated with shale beds (with -many horizontal grazing traces of Palaeo- -phycus striatus), forms the top of the Smel- -tertown Formation (Fig. 8). Limestone at -the basal Muleros Formation rests with -sharp contact on sandstone at the top of -the Smeltertown Formation. - -Paleontology - -The Smeltertown Formation yields fossils -of foraminiferans, bivalves, gastropods, -ammonoids, bryozoans, crustaceans, ophi- -uroids, echinoids, brachiopods, and corals -(Böse 1910; Strain 1976; Nye and LeMone -1978; Bullock 1985; Bullock and Cornell -1986; Cornell et al. 1991; Turnšek et al. -2003). The foraminiferans establish a robust -correlation of the Smeltertown Formation -with the Duck Creek Formation of north- -central Texas (Bullock and Cornell 1986). -Ammonites of the Mortoniceras equidistans -Zone (present in unit 5 of our measured sec- -tion; Fig. 8) also correlate the Smeltertown -Formation to the lower part of the Mojado -Formation (Fryingpan Spring Member) in -the Cooke’s Range of Luna County, New -Mexico (Böse 1910; Cobban 1987; Lucas -et al. 1988; Lucas and Estep 1998b, 2000). - -Sedimentary petrography (Figs. 8, 18, 19A–D) - -Unit 3 in our measured section of the -Smeltertown Formation (Fig. 8) is a mixed -siliciclastic-carbonate siltstone that is -partly laminated and partly bioturbated -(Fig. 18A). The siltstone contains as much -as 30% detrital quartz grains, abundant car- -bonate grains, rare glauconite, and micrite. -A few large mollusc fragments (bivalves -and gastropods) float in the siltstone. Other -fossils present are ostracods, a few smal- -ler foraminiferans, rare echinoderms, and -echinoid spines. - -The nodular limestone ledges of unit 5 of -the Smeltertown Formation section (Fig. 8) -are composed of bioclastic, mixed silici- -clastic-carbonate siltstone, which is partly -laminated and partly bioturbated. A few -large mollusc shell fragments of bivalves -and gastropods float in silty matrix. The -siltstone consists of as much as 30% quartz -grains, small carbonate grains, a few -glauconite grains, and micrite. Bioclasts -include abundant ostracods, a few smaller -foraminiferans, rare echinoderms, echinoid -spines, and shell debris. - -The microfacies of the unit 7 limestone -ledge (Fig. 8) is fine-grained, bioturbated -mixed siliciclastic-carbonate sandstone -(Fig. 18B). The sandstone contains abundant -detrital quartz grains, mostly 0.2–0.5 mm in -diameter, rarely to 1 mm. Most common is -monocrystalline quartz, and polycrystal- -line quartz is subordinate. Rare chert and -glauconite grains are present. Quartz grains -are surrounded by thin carbonate cement -rims. A few micritic intraclasts are present, -too. Bioclasts include bivalves, gastropods, -echinoderms, ostracods, foraminiferans, -and rare worm tubes(?). Litho- and bio- -clasts float in a micritic matrix. - -Thin limestone beds of the uppermost -Smeltertown Formation (Fig. 8, unit 8) -consist of bioturbated mudstone (Fig. 18C) -containing a few siliciclastic grains (mostly -quartz, rare chert, and glauconite grains) -and bioclasts (mostly ostracods, rare -bivalve shells, and echinoderm fragments; -probably other skeletons). Siliciclastic -grains and bioclasts constitute less than -10% of the rock, and they are locally con- -centrated due to bioturbation. - -The top siltstone to sandstone of the Smel- -tertown Formation (Fig. 8, unit 9) is fine -grained, laminated, and locally bioturbated -(Fig. 19A–D). Siliciclastic grains constitute -about 50–70% of the rock. Most abundant is -detrital quartz, and rare are grains of glau- -conite, opaque minerals, and micas. A few -micritic intraclasts are present. Bioclasts, -such as shell fragments, echinoderms, and -ostracods, are rare. This facies grades into -bioclastic wackestone that contains agglu- -tinated foraminiferans, bivalve fragments, -and some other bioclasts embedded in silty, -partly peloidal matrix with a few small -quartz grains (Fig. 18D–F). - -Depositional environments - -Dark-colored shale of the Smeltertown For- -mation is indicative of a muddy marine -sea bottom (cf. Scott 1970). From shales -of the Smeltertown Formation, Bullock -and Cornell (1986) identified 73 species of -foraminiferans that belong to the suborders -Texturlariina, Miliolina, and Rotaliina and -provided some information on the deposi- -tional environment. Within the formation -they distinguished three zones character- -ized by distinct assemblages of foraminifer- -ans. Zone 1 (lower 13 m), which correlates -with unit 1 of Strain (1976) and with units - -FIGURE 8—Measured stratigraphic section -of Smeltertown Formation. See Figure 2 and -Appendix for location of section, and Figure 6 -for lithology legend. Dashed lines identify the -three zones of Bullock and Cornell (1986). - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -109 - - planktonic foraminiferans are absent, -and other rotaliine forms are less abun- -dant than in the underlying strata. Textu- -lariine forms constitute 70% of the tests. -Individuals of Cribratina texana comprise -60–99% of the specimens, indicating an -extremely restricted environment, prob- -ably a hypersaline marsh according to -Bullock and Cornell (1986), or a shal- -low, nearshore shelf with possibly lower -than normal salinity (R. W. Scott, written -comm. 2010). - -The foraminiferal assemblages of the -shales thus indicate that the Smeltertown -Formation represents an upward-shallow- -ing succession from a deeper, open shelf -with water depths of approximately 50 m -to a shallow restricted shelf environment -with increased salinity. Interbedded silt- -stone, sandstone, and limestone beds in the -upper part of the succession may represent -storm layers similar to those of the Del -Norte Formation. - -Muleros Formation - -Lithostratigraphy - -Our measured section of the Muleros For- -mation (Fig. 9) is the type section of the for- -mation (Strain 1976, table 10). The Muleros -Formation is subdivision 5 of Böse (1910), -which he reported as 30–50 m thick; Strain -(1976) reported a thickness of 32 m for the -formation type section. Our measured sec- -tion of the Muleros Formation is approxi- -mately 33 m thick. The formation generally -forms cliffs and ledgy slopes (Fig. 5D), and -many of the limestone beds of the middle -and upper units are Texigryphaea pack- -stones (Figs. 5E and 9). - -Following Strain (1976), the Muleros For- -mation can be divided into three members -(Fig. 9). The lower 11–12 m is light-colored -shale with some thin, intercalated beds of -bioturbated sandstone and limestone with -abundant Texigryphaea. The limestone is -nodules embedded in brownish silty matrix -and micrite. The middle part of the Muleros -Formation (from 11.5 to 24.3 m) is composed -of bioturbated micritic limestone interca- -lated with thin shale and, in the lower part, -thin bioturbated sandstone. The limestone -contains Texigryphaea and Cribratina and is -bioturbated. Horizontal, meniscate back- -filled burrows are present in unit 5 (Fig. 9). -The upper third of the section (approxi- -mately 9 m thick) is composed of alternat- -ing limestone ledges and shale. Limestone -intervals are 0.6–1.4 m thick and consist of -floatstone/rudstone and packstone contain- -ing Texigryphaea. Shale intervals in the upper -part are as much as 1.6 m thick. The lower -two intercalated shale intervals contain -limestone nodules. The base of the Mesilla -Valley Formation is a sharp contact of silty -shale on Texigryphaea packstone at the top of -the Muleros Formation (Fig. 9). - -Paleontology - -The Muleros Formation is very fossilifer- -ous, primarily because of the packstones of -the gryphaeid bivalve Texigryphaea washi- -taensis it contains (Böse 1910; Kues 1989). It -also yields other bivalves, gastropods, rare -ammonoids, common spatangoid echino- -derms, foraminiferans (especially the large, -arenaceous form Cribratina texana), corals, -sponge borings, bryozoans, ostracods, and -serpulids (Böse 1910; Strain 1976; Kues -1989; Turnšek et al. 2003). Correlation of -the Muleros Formation with the Denton -and Fort Worth Formations of north-central -Texas is supported by stratigraphic posi- -tion but lacks a robust ammonite biostrati- -graphic basis (Scott et al. 2003). - -Sedimentary petrography (Figs. 9, 19E–F, 20, 21) - -The uppermost bed of the Smeltertown -Formation (Fig. 8, unit 9) is a quartzose -siltstone to fine-grained sandstone that is -heavily bioturbated and partly laminated. -Most abundant are monocrystalline quartz -grains; chert, carbonate grains, opaque -minerals, and glauconite are rare. Small -bioclasts including shell debris, ostracods, -echinoderms, and a few large bivalve shells -float in siltstone to fine-grained sandstone. -The rock is cemented by quartz over- -growths, rarely by calcite cement. - -The limestone beds in the lower part of -the Muleros Formation (Fig. 9, unit 2) are -gray, bioturbated, fine grained, and com- -posed of abundant peloids, a few detrital -quartz grains, and bioclasts such as echi- -noderms, echinoid spines, ostracods, and a -few smaller foraminiferans. The matrix is -micrite. The limestone in the middle part of -the Muleros Formation (Fig. 9, units 3–8) is -mainly composed of bioclastic wackestone -to packstone (Fig. 20D), and subordinately -mudstone and floatstone to rudstone. - -The bioclastic wackestone is fine to coarse -grained, poorly sorted, commonly biotur- -bated, and locally laminated. Abundant -fossils are large gryphaeid bivalve shells, -subordinate gastropods, echinoderms, cor- -als, smaller foraminiferans (including large -agglutinated forms), ostracods, and echinoid -spines. A few glauconite grains are present. -Locally, peloids and detrital quartz grains -are present (Fig. 19E, F; Fig. 20A, B, F). - -The bioclastic mudstone is bioturbated -and contains as much as about 5% detrital -quartz grains and rare glauconite (Fig. 20E). -Fossils such as smaller foraminiferans -(including rare, larger agglutinated forms), -rare ostracods, echinoderms, echinoid -spines, and other skeletons are floating in -micrite. The following foraminiferans are -present in the limestone: Ammobaculites, -Ataxophragmiidae, and Nodosaria. - -The bioclastic siltstone to mudstone -(Fig. 9, unit 4) is nonlaminated and biotur- -bated (Fig. 20C). Many small and a few larg- -er bioclasts (echinoderms, ostracods, bivalve -shell fragments, rare smaller foraminiferans, - -FIGURE 9—Measured stratigraphic section of -Muleros Formation. See Figure 2 and Appendix -for location of section, and Figure 6 for lithology -legend. - -1 and 2 of our section (Fig. 8), is character- -ized by a high abundance of “Hedbergella” -planispira and the dominance of rotaliine -forms, which constitute 45–90% of the indi- -viduals. This assemblage indicates deposi- -tion in an open, inner-shelf environment -with water depths of 50 m or less (Bullock -and Cornell 1986). - -Zone 2, including unit 2 and the lower -9 m of unit 3 of Strain (1976), which is our -unit 3 and part of unit 4 (Fig. 8), is char- -acterized by textulariine forms, which -constitute 55–95% of all individuals, with -Cribratina texana being the most abundant -species. The presence of Ammobaculites -and Trochammina is indicative of a shal- -low, nearshore environment. Bullock and -Cornell (1986) concluded that during sedi- -mentation of this interval some shallowing -occurred, causing restriction of circulation. -In zone 3, encompassing the upper -24 m of the section (upper 22 m of unit -3 and unit 4 of Strain 1976; upper part -of our unit 4 and all of units 5–6; Fig. 8), - -110 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - corals, and bryozoans) float in a silty to mic- -ritic matrix. Peloids (fecal pellets?) are local- -ly present, as are small quartz grains (< 5%). -The floatstone to rudstone (Fig. 9, -unit 7) is nonlaminated, poorly sorted, -and contains abundant shell fragments of -gryphaeids and gastropods (Fig. 21A–B). -Many large agglutinated foraminiferans are -also present. These foraminiferans include -rare agglutinated smaller foraminiferans. -Echinoderm fragments, ostracods, smaller -foraminiferans, and shell debris are sub- -ordinate. A few mollusc shell fragments -are encrusted by worm tubes. The matrix -is gray peloidal micrite containing a few -small quartz grains. - -The limestone of unit 8 (Fig. 9) is fine- -grained, bioturbated wackestone (Fig. 21C) -containing abundant small bioclasts such as -ostracods, echinoderms, smaller foraminif- -erans, mollusc shell debris, and rare larger -skeletons of gastropods, echinoderms, and -bivalves. The matrix is peloidal micrite -containing a few quartz grains and rare -glauconite. - -Unit 10 of our Muleros Formation sec- -tion (Fig. 9) is a bioturbated floatstone- -rudstone, containing abundant bivalve -(Texigryphaea) and gastropod shell frag- -ments, many agglutinated foraminiferans, -echinoderms, and rare corals (Fig. 21D–E). -Smaller foraminiferans, ostracods, and -echinoid spines are subordinate. A few -skeletons are encrusted by worm tubes. -The micritic matrix contains a few angular -quartz grains and rare glauconite. - -The uppermost limestone interval of the -Muleros Formation (Fig. 9, units 13–14) is -a packstone–rudstone (Fig. 21F) composed -of densely packed bivalve shell fragments -that are oriented more or less parallel to -the bedding plane. Other bioclasts, such -as echinoderms, ostracods, and smaller -foraminiferans, are rare. The matrix is silt- -stone containing abundant angular quartz -grains and rare glauconite grains. Locally, -some calcite cement is present (“umbrel- -la” porosity). The peloidal wackestone -(pelmicrite; Fig. 9, unit 14) is laminated and -consists of abundant small peloids, mostly -around 0.05 mm in size and a few peloids -as large as 0.25 mm in diameter. Bioclasts -such as ostracods and foraminiferans -are rare. The floatstone–rudstone of the -upper part is poorly sorted, nonlaminated, -partly bioturbated, and contains abun- -dant large gryphaeid and gastropod shell -fragments. A few shells display borings -and are encrusted by annelid worms (ser- -pulid worm tubes). Other bioclasts include -echinoderms, echinoid spines, ostracods, -foraminiferans (Lenticulina), rare corals, -and bryozoans(?). Locally, agglutinated -foraminiferans are common. The rock con- -tains many detrital angular quartz grains -and a few glauconite grains. The matrix is -micrite. - -FIGURE 10—Measured stratigraphic section -of Mesilla Valley Formation. See Figure 2 and -Appendix for location of section, and Figure 6 -for lithology legend. - -FIGURE 11—Measured stratigraphic section of -Mojado Formation (“Anapra Sandstone”). See -Figure 2 and Appendix for location of section, and -Figure 6 for lithology legend. HCS=hummocky -cross-stratification. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -111 - - FIGURE 12—Photos of Cretaceous outcrops at Cerro de Cristo Rey. -A—Outcrops of Mojado Formation overlying Mesilla Valley Forma- -tion (Fig. 10). B—Characteristic crossbedded sandstone of Mojado -Formation. C—Dinosaur footprints (ornithopod tracks) in Mojado - -Formation. Divisions of measuring stick = 0.25 m. D—Base of Del Rio -Formation on Mojado Formation (Figs. 11 and 13). E—Overview of Buda -Formation measured section (Fig. 14). F—Contact of Buda and Mancos -Formations at measured section (Fig. 15). - -Depositional environments - -Muleros strata also demonstrate an appar- -ent upward-shallowing trend. Thus, lower, - -shale-dominated strata are relatively deep- -er marine deposits (though still shallow), -whereas Texigryphaea packstones and inter- -calated bioturbated limestones and sandy - -limestone of the middle-upper Muleros -(Fig. 9) suggest shallow, carbonate shelf -deposition with some coarse clastic influx -(cf. Turnšek et al. 2003). - -112 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - transport and deposition as density current -deposits (cf. Scott et al. 1975; Hobday and -Morton 1984). - -The middle part of the Muleros Forma- -tion (Fig. 9, units 3–8) is dominated by dif- -ferent types of limestone containing a fossil -assemblage suggesting a shallow-marine -setting. Micritic types such as mudstone -and wackestone accumulated in shallow, -quiet water, where currents were not strong -enough to remove the mud. Intercalated -floatstone to rudstone, commonly devel- -oped as coquinas containing abundant -complete and broken gryphaeid bivalve -shells, are interpreted to represent storm -deposits. Such storm layers with abundant -gryphaeid shells are also present in the -upper third of the formation, where they -are intercalated with fossiliferous shale -deposited under low-energy conditions. -Deposition of the upper part of the Muleros -thus was also in a shallow-marine setting, -but one with more storm deposits than in -the middle member. - -The Muleros Formation succession thus -indicates an outer-shelf environment with -dominantly shale during deposition of the -lower part, followed by some shallowing -and deposition of dominantly limestone -in an inner-shelf environment periodically -affected by storms, and again some deep- -ening resulting in increased deposition of -shale with intercalated storm layers in the -upper part. - -Mesilla Valley Formation - -Lithostratigraphy - -The Mesilla Valley Formation is a slope- -forming, shale-dominated unit approxi- -mately 65 m thick at our measured sec- -tion (Figs. 5F, 10). This is subdivision 6 of -Böse (1910), who indicated it is more than -20 m thick. Our measured section is the -type section of the Mesilla Valley Shale of -Strain (1976, table 11), who indicated it is -64 m thick. - -Strain (1976) divided the Mesilla Val- -ley Formation into two units, a lower shale -(59–60 m thick) and an upper sandstone -and shale (4–5 m thick). However, like the -underlying Muleros Formation, we divide -the Mesilla Valley Formation into three infor- -mal members (Fig. 10). The lower member -(member A) is 13.7 m thick and consists of -gray shale with some very thin sandstone -lenses. The middle member (member B) is -14.4 m thick. Its base is a 0.4-m-thick ledge- -forming limestone that contains Texigryphaea. -The limestone bed is overlain by olive shale -with thin, fine-grained sandstone ledges and -three thicker, siltstone-sandstone intervals -(each 0.4 m thick) that display ripple lami- -nation. Locally, the sandstone contains shell -debris and serpulids, and the thin sandstone -beds contain Texigryphaea (upper part). The -upper member (member C) is black shale - -forming a thick slope (36.5 m). The black -shale is sharply overlain by coarse clastic -sediments (sandstone) at the base of the -Mojado Formation (“Anapra Sandstone” of -Strain). - -Paleontology - -Böse (1910) referred to the Mesilla Valley For- -mation as the “capas con [beds with] Ostrea -quadriplicata [now Peilinia quadriplicata: Kues -1997],” and most (or all?) fossils of this taxon -appear to derive from member B of the Mesil- -la Valley Formation (Fig. 10). The Mesilla Val- -ley Formation yields a profuse assemblage of -gryphaeid bivalves (Texigryphaea) as well as -other bivalves, gastropods, ammonoids, bra- -chiopods, foraminiferans (particularly Cribra- -tina texana), dinoflagellates, echinoids, serpu- -lid worms, corals, ostracods, echinoderms, -calcareous algae, and some terrestrial plant -fragments (Böse 1910; Strain 1976; Cornell -1982; Kues 1989, 1997; Turnšek et al. 2003). -Correlation of the Mesilla Valley Formation -with the Weno and Pawpaw Formations -(Scott et al. 2003) is based more on strati- -graphic position than on precise biostratig- -raphy, although the presence of the bivalve -Peilinia quadriplicata in the Mesilla Valley For- -mation, also known from the Denton–Main -Street interval in north-central Texas (Scott -et al. 2003, p. 315), does support the Weno– -Pawpaw correlation. - -Sedimentary petrography (Figs. 10, 22) - -The lower part of the Mesilla Valley Forma- -tion (member A) is gray shale with a few -thin sandy lenses. The microfacies of the -limestone at the base of member B (Fig. 10, -unit 3) is bioclastic rudstone–packstone -(Fig. 22A–B) composed mainly of large, bro- -ken bivalve fragments. Subordinate large -agglutinated foraminiferans (Fig. 22C), -echinoderms, ostracods, and rare smaller -foraminiferans are present. The matrix is -silty and contains abundant small, angular -quartz grains. The large shell fragments are -randomly oriented parallel to the bedding -plane. - -The siltstone of unit 4 at the base of -member B (Fig. 10) is laminated, mixed -siliciclastic-carbonate and contains rare -bioclasts. The most abundant grain type -is quartz, and carbonate grains are also -present. Bioclasts include ostracods, echi- -noderms, foraminiferans, and shell debris. -Glauconite grains are common. - -In member B, the ripple-laminated silt- -stone of unit 9 (Fig. 10) is partly bioturbated -and contains small and some larger bioclasts -floating in the silty matrix. Larger bioclasts -are bivalve shell fragments, agglutinated -foraminiferans, and rare solitary corals -(Fig. 22D). Small bioclasts include ostra- -cods, gastropods, small foraminiferans, shell -debris, and rare phosphatic fragments. The -silty matrix is composed of 70–80% small -angular quartz grains, about 2% glauconite, -and a few micas in micritic matrix. - -FIGURE 13—Measured stratigraphic section of -Del Rio Formation. See Figure 2 and Appendix -for location of section, and Figure 6 for lithology -legend. - -The lower part of the Muleros Forma- -tion is mainly composed of shale that accu- -mulated on the shelf in areas deeper than -storm-induced currents. Intercalated thin -sandstone beds probably formed during -storms that resulted in the remobilization -of coarser material and their basinward - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -113 - - member C apparently occurred in a poorly -oxygenated environment, which may indi- -cate relatively deep water. - -Mojado Formation - -Lithostratigraphy - -Our measured section of the Mojado Forma- -tion (Figs. 11, 12A–D) is the type section of the -Anapra Sandstone of Strain (1976, table 12), -which he reported as approximately 53 m -thick. Here, the Mojado Formation is a sand- -stone-dominated unit as much as 58 m thick -(Figs. 5F, 11, 12A–D). It rests with sharp con- -tact on shale at the top of the Mesilla Valley -Formation (Figs. 5F, 11, 12A) and is sharply -overlain by shale and nodular limestone at -the base of the Del Rio Formation (Fig. 12D). -The topmost bed of the Mesilla Valley -Formation at our Mojado Formation sec- -tion (Fig. 11) is reddish-brown shale/silt- -stone containing abundant quartz, some -chert grains, and rare glauconite. The -matrix (or cement) is dark brown (iron -stained). Strain (1976, p. 81) stated that “the -base of the Anapra [Mojado] is the lowest -sandstone stratum 30 cm thick or more just -above the thinly bedded sandstone beds at -the top of the Mesilla Valley Shale.” Thus, -he included our basal beds of the Mojado -Formation (Fig. 10, unit 13; Fig. 11, unit 3) -in the Mesilla Valley Formation. Instead, -we draw the base of the Mojado Formation -at the base of the first traceable sandstone -bed of typical Mojado lithology. - -The Mojado Formation is entirely silici- -clastic, being composed of different types of -sandstone with intercalated gray and black -(locally lignitic) shale. A covered interval -(5.4 m) is in the middle of the succession. -Sandstone intervals are as much as 5.7 m -thick, and shale intervals are as much as -3.5 m thick. The following lithofacies are rec- -ognized: (1) thin, fine-grained sandstone beds -displaying hummocky cross-stratification -form the lowermost 4.9 m of the succession -(lower ¾ of unit 3); the sandstone beds are -interbedded with green shale; (2) iron-mot- -tled, hackly sandstone, which is bioturbated -and has Liesegang banding (units 7, 11); -(3) thin, fine-grained sandstone with ripple -lamination (units 16–22); (4) thin, medium- -grained sandstone with trough crossbedding -(units 4, 6, 8, 10, 17); and (5) thicker, multisto- -ried sandstone beds with larger-scale trough -crossbedding (units 13, 15, 25). - -Trough crossbedded sandstone inter- -vals are characterized by erosive lower con- -tacts (e.g., Fig. 12B). Several upward-fining -and thinning cycles are developed. Each cycle -starts with trough crossbedded sandstone, -grading into fine-grained, ripple-laminated -sandstone with shale interbeds, and the tops -of a few cycles are formed by shale. Hum- -mocky cross-stratification is only observed -in the lowermost sandstone beds. Neverthe- -less, marine fossils and marine bioturbation - -FIGURE 14—Measured stratigraphic section of -Buda Formation. See Figure 2 and Appendix for -location of section, and Figure 6 for lithology -legend. - -FIGURE 15—Measured stratigraphic section of -Mancos Formation (“Boquillas Formation”). See -Figure 2 and Appendix for location of section, -and Figure 6 for lithology legend. - -A thin limestone bed intercalated in -unit 10 (Fig. 10) is bioturbated bioclastic -wackestone containing a few larger and -many small bioclasts floating in micritic -matrix (Fig. 22E). Bioclasts include mol- -lusc shell debris (bivalves, gastropods), -calcareous algae, worm tubes (Fig. 22F), -ostracods, echinoderms, smaller foraminif- -erans, agglutinated foraminiferans, and -echinoid spines. The matrix is micrite and -pelmicrite. The top of member B (Fig. 10, -unit 11) is formed by a fine-grained sand- -stone interval 0.9 m thick displaying rip- -ple lamination and containing Texigryphaea -(packstone). Member C is black shale with -no obvious fossil content. - -Depositional environments - -The shale-dominated Mesilla Valley For- -mation (Fig. 10) mostly represents deposi- -tion on muddy marine sea bottoms below -wave base. The lower part of the formation -(member A, Fig. 10) is mostly shale with a - -few thin sandstone lenses, and this appears -to represent relatively deep water. The -dark shale with intercalated thin sandstone -lenses of the lower member is interpreted -to record deposition in an outer-shelf envi- -ronment, which was only rarely affected -by storms. The thin intercalated sandstones -may represent distal storm layers. Inter- -calated rudstone to packstone beds (shell -coquinas) and ripple-laminated sandstone -beds containing fossils such as Texigryphaea -represent typical storm deposits. - -Overlying shale with interbedded Texi- -gryphaea packstones and ripple-laminated -sandstones (member B) indicate a distinct -shallowing. Storm layers are more abun- -dant in member B than in the members -below and above, indicating that some -shallowing occurred during deposition of -the middle member (Fig. 10). - -The upper part of the formation (mem- -ber C) is black shale, which we interpret as -relatively offshore marine deposits. Depo- -sition of the organic-rich black shale of - -114 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 16—Thin section photographs of microfacies of the Finlay Formation -at Cerro de Cristo Rey (Fig. 6). Sample numbers correspond to unit numbers -in Figure 6. A—Bioclastic wackestone containing abundant broken fragments -of recrystallized calcareous algae, molluscs, some echinoderm fragments, -and foraminiferans in micritic matrix. Sample A24. B—Bioclastic wackestone -containing abundant fragments of dasycladacean algae, mollusc (bivalve) -fragments, and some echinoderms floating in micritic matrix. Sample A27. -C—Fine-grained bioclastic wackestone–packstone with abundant fragments -of recrystallized dasycladacean algae, molluscs, echinoderms, and a few - -foraminiferans. Matrix is gray micrite. Sample A26. D—Packstone contain- -ing abundant peloids, smaller foraminiferans including dictyoconid forms, -fragments of dasycladacean algae, and molluscs and some micritic matrix. -Sample A10. E—Packstone–grainstone containing abundant peloids, some -orbitolinoid foraminiferans (Dictyoconus), and mollusc shell debris cemented -by calcite. Sample A10. F—Grainstone composed of abundant peloids, some -orbitolinoid foraminiferans (Dictyoconus), and other bioclasts including gas- -tropods and echinoderms, cemented by calcite. Sample A10. Width of photo- -graphs is 6.3 mm. - -are present in the upper few meters of the -Mojado Formation, and we identify a clear -break (transgressive surface) approximately -3 m below its top (base of unit 26, Fig. 11). - -Strain (1976) divided the Mojado (his -“Anapra”) Formation into four units, and - -Lovejoy (1976) recognized five units. We -believe our measured section can be readily -divided into at least four intervals: (1) lower -beds with marine bioturbation and intercalat- -ed shale (Fig. 11, units 3–11), (2) middle, slope- -forming interval of mostly siltstone and shale - -(Fig. 11, units 12–14); (3) upper interval of -crossbedded sandstones (Fig. 11, units 15–25); -and (4) uppermost interval with marine fossils -and bioturbation (Fig. 11, unit 26). However, -more subdivisions are possible, though they -are of little stratigraphic significance. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -115 - - A - -C - -E - -B - -D - -F - -FIGURE 17—Thin section photographs of microfacies of the Del Norte For- -mation at Cerro de Cristo Rey (Fig. 7). Sample numbers correspond to unit -numbers in Figure 7. A—Bioclastic wackestone containing abundant algal -fragments and mollusc (bivalve) shell debris, a few foraminiferans, and -echinoderms in gray micrite. Sample B2. B—Mixed carbonate-siliciclastic -siltstone containing abundant angular quartz grains and bioclasts such as -recrystallized bivalve shell fragments (?oysters), some echinoderms, and -micrite. Sample B8. C—Wackestone–packstone composed of echinoderms, -mollusc shell fragments, and quartz grains floating in micrite. Sample B9. - -D—Mixed siliciclastic-carbonate fossiliferous siltstone containing a few -mollusc shell fragments and rare echinoderms. Sample B16. E—Bioclastic -wackestone–packstone containing abundant mollusc (bivalve) shell frag- -ments (mainly derived from bivalves), subordinate echinoderms, small -quartz grains, and a few micritic intraclasts in micritic matrix. Sample -B22. F—Fine-grained bioclastic wackestone with many mollusc (bivalve) -shell fragments and dasycladacean algae, with subordinate echinoderms -in micritic, partly peloidal matrix. Sample B19. Width of photographs is -6.3 mm, but for Fig. C is 3.2 mm. +Practical Tips -We propose that at Cerro de Cristo Rey -the name Anapra Formation be replaced by -Mojado Formation. To the west and north- -west of Cerro de Cristo Rey, rocks that strad- -dle the Albian–Cenomanian boundary are -assigned to the Mojado Formation of Zeller -(1965). At the type section, in the Big Hatch- + Understand Your Audience: Tailor your message to the audience's values and -et Mountains of Hidalgo County, New Mex- -ico, the Mojado Formation is approximately -1,500 m of sandstone, shale, and minor lime- -stone that overlies the Aptian–Albian U-Bar -Formation (Zeller 1965). In the Cooke’s -Range of Luna County, approximately 120 -km northwest of Cerro de Cristo Rey, the - -Mojado Formation is divided into three -members (Lucas and Estep 1998b): (1) Fry- -ingpan Spring Member, approximately 22 m -of sandstone, siltstone, shale, and nodular -limestone that includes fossils of the ammo- -noid Mortoniceras equidistans (Cragin) (Cobban -1987; Lucas et al. 1988); (2) Sarten Member, - -116 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 18—Thin section photographs of microfacies of the Smeltertown -Formation at Cerro de Cristo Rey (Fig. 8). Sample numbers correspond -to unit numbers in Figure 8. A—Mixed fossiliferous carbonate-siliciclas- -tic siltstone containing some large mollusc (bivalve) fragments, a few -ostracods and foraminiferans, and as much as 30% detrital quartz grains. -Sample E3. B—Fine-grained, mixed siliciclastic-carbonate sandstone -composed of detrital quartz grains, a few glauconite grains, and a few - -bioclasts (bivalves, gastropods, echinoderms, foraminiferans) floating in -micrite. Sample E7. C—Mudstone containing some quartz grains and few -bioclasts. Sample E8. D–F—Bioclastic wackestone containing agglutinat- -ed foraminiferans (Cribratina) and bivalve fragments, and, subordinately, -other bioclasts such as ostracods embedded in silty, partly peloidal matrix -containing a few small quartz grains. Sample E9. Width of photograph C -is 3.2 mm, and of all other photographs is 6.3 mm. +beliefs. -approximately 58 m of sandstone and silt- -stone of nonmarine origin; and (3) Rattlesnake -Ridge Member, approximately 6 m of silty -sandstone, siltstone, and nodular limestone -that yields a bivalve fauna of earliest Ceno- -manian age (Lucas et al. 1988). Sandstone and -siltstone of the Sarten Member are lithologi- -cally very similar to the “Anapra Sandstone,” -and the biostratigraphy supports a reasonably + Practice Persuasion: Engage in debates and discussions to hone your persuasive -close correlation (Fryingpan Spring Member -correlates to the Smeltertown Formation, and -Rattlesnake Ridge Member correlates to the -Del Rio Formation; Lucas et al. 1988). Thus, -the “Anapra Sandstone” is seen by us as a -tongue of the Bisbee Group (Mojado Forma- -tion) present in the Cretaceous section of Cerro -de Cristo Rey. We therefore abandon the local -name Anapra (of Strain 1968, 1976) in favor of - -the older name of a more regional stratigraph- -ic unit (Mojado of Zeller 1965). - -Paleontology - -The Mojado Formation (“Anapra Sand- -stone”) at Cerro de Cristo Rey contains -dinosaur footprints and plant debris in its - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -117 - - A - -C - -E - -B - -D - -F - -FIGURE 19—Thin section photographs of microfacies of the Smeltertown -Formation (A–D) and Muleros Formation (E–F) at Cerro de Cristo Rey -(Figs. 8–9). Sample numbers correspond to unit numbers in Figures 8 and 9. -A—Mixed siliciclastic-carbonate siltstone containing a few glauconite -grains and rare bioclasts such as ostracods and echinoderms. Sample E9. -B—Quartzose siltstone with a few carbonate grains, rare glauconite, and -few large bivalve fragments. The siltstone is cemented by quartz (authi- -genic overgrowths) and rare calcite cement. Sample E10. C–D—Mixed -siliciclastic-carbonate siltstone to fine-grained sandstone containing some - -bivalve shell fragments and echinoderms. Sample F1. E—Fine-grained -bioclastic wackestone–packstone overlying gray mudstone. Wackestone– -packstone contains bivalve shell fragments, echinoderms, ostracods, a few -smaller foraminiferans, peloids, and a few small quartz grains. Sample F2. -F—Coarse-grained bioclastic wackestone, partly floatstone, containing -abundant bivalve shell fragments (Texigryphaea), subordinate gastropods, -echinoderms, foraminiferans, few peloids, and quartz grains in micritic -matrix. Sample F3. Width of photograph D is 3.2 mm, and of all other pho- -tographs is 6.3 mm. +skills. -lower to middle portions, as well as some -marine fossil fragments and tests of the large -arenaceous foraminiferan Cribratina texana. -The dinosaur tracks are primarily those of -ornithopods and theropods (Fig. 12C; e.g., -Kappus et al. 2003; Kappus 2007) and pro- -vide strong evidence of subaerial, terrestrial +4.1.6 Informative Speaking -conditions during deposition of part of the -Mojado Formation, but are not of precise -biostratigraphic significance. The upper -3 m of the Mojado (unit 26) contain fossils -of the large exogyrid bivalve Gyrostrea whit- -neyi (known from the Del Rio Formation -in north-central Texas) as well as echinoids - -and agglutinated foraminiferans (Böse 1910; -Strain 1976; Turnšek et al. 2003). We also -observed dasycladacean algae, foraminifer- -ans, echinoderms, and ostracods in thin sec- -tion (Fig. 23E). - -Strain (1976) placed the base of the -Cenomanian at the base of the “Anapra - -118 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 20—Thin section photographs of microfacies of the Muleros Forma- -tion at Cerro de Cristo Rey (Fig. 9). Sample numbers correspond to unit num- -bers in Figure 9. A–B—Bioclastic wackestone to floatstone containing abundant -bivalve shell fragments (Texigryphaea), subordinate gastropods, echinoderms, -and foraminiferans, including large agglutinated forms in micritic matrix. Sam- -ple F3. C—Bioturbated siltstone–mudstone containing a few bioclasts (bivalve -shell fragments, echinoderms, ostracods) that float in matrix. Sample F3. D— - -Fine-grained wackestone–packstone with a few large bivalve fragments and -many small bioclasts, peloids, and some detrital angular quartz grains in mic- -ritic matrix. Sample D5. E—Mudstone containing quartz grains and smaller -textulariid foraminiferans floating in micrite. Sample F5. F—Fine-grained -wackestone–packstone containing abundant peloids, a few detrital quartz -grains, shell debris, echinoderms, smaller foraminiferans, and ostracods. Sam- -ple F6. Width of photographs A–D is 6.3 mm, and of E–F is 3.2 mm. +Definition -Sandstone,” and Lucas and Estep (1998b, -2000) placed it at the base of the Del Rio -Formation. Given that in the regional -Washita Group section, the base of the -Cenomanian is a sequence boundary at -the base of the WA6 cycle of deposition, -the Albian–Cenomanian boundary is -most likely at the marine transgressive -surface that is approximately 3 m below +Informative speaking aims to educate the audience about a specific topic. This method is -the Mojado–Del Rio contact (Turnšek -et al. 2003; Scott et al. 2003) (Fig. 11). +commonly used in lectures, presentations, and training sessions. -Sedimentary petrography (Figs. 11, 23C–F) - -Sandstone of the Mojado Formation is -fine- (0.1–0.3 mm) to medium-grained (0.2– -0.5 mm) quartz arenite, nonlaminated, and -well sorted (Fig. 23C–D). Grains are mostly - -subrounded to rounded, although many -grains display authigenic overgrowths -that are not always easy to recognize. The -most common grain type is monocrystal- -line quartz. Subordinate chert grains are -present, perhaps of volcanic origin. Poly- -crystalline quartz of metamorphic origin -including stretched metamorphic grains -is rare. Detrital feldspars are very rare - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -119 - - A - -C - -E - -B - -D - -F - -FIGURE 21—Thin section photographs of microfacies of the Muleros Forma- -tion at Cerro de Cristo Rey (Fig. 9). Sample numbers correspond to unit num- -bers in Figure 9. A—Floatstone containing gryphaeid shell fragments, some -encrusted by balanids, and agglutinated foraminiferans embedded in micritic, -partly peloidal matrix. Sample F7. B—Wackestone–rudstone with abundant -gryphaeid shell fragments, some gastropods, agglutinated foraminiferans, and +Techniques -other bioclasts floating in micritic matrix. Sample F7. C—Fine-grained biotur- -bated wackestone composed of abundant peloidal and small bioclasts (ostraco- -ds, echinoderms, mollusc shell debris, foraminiferans) and a few quartz grains. -Sample F8. D–F—Floatstone–rudstone with bivalve fragments, agglutinated -foraminiferans, echinoderms, and corals embedded in micritic matrix. D–E = -sample F10, F = sample F14. Width of photographs is 6.3 mm. + Clear Structure: Organize your speech with a clear introduction, body, and -and generally altered. Accessory grains -are micas (muscovite), tourmaline, zircon, -apatite, and opaque grains. The sandstone -is cemented by authigenic quartz over- -growths, and locally also by microcrystal- -line quartz and opaque cement (Fe-hydrox- -ides). Unit 10 (Fig. 11) includes a lens of -strongly recrystallized stromatolite com- -posed of irregularly laminated algal mats. +conclusion. -Unit 23 (Fig. 11) is a crossbedded, fine- -grained, bioturbated bioclastic wacke- -stone, locally a packstone, containing a few -larger bioclasts (Fig. 23E). Most abundant -are dasycladacean algae and mollusc shell -fragments (bivalves, gastropods). Less -common are foraminiferans, echinoderms, -echinoid spines, ostracods, and other frag- -ments. The matrix is micrite to pelmicrite. - -The uppermost Mojado Formation (Fig. 11, -units 25–26) is composed of laminated silt- -stone to fine-grained sandstone. Grain size -is mostly < 0.1 mm; however, a few quartz -grains as much as 0.5 mm are present. The -most abundant grain type is quartz; carbon- -ate grains are subordinate, and opaques and -glauconite are rare. The siltstone to fine- -grained sandstone contains some ostracods, - -120 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B + Use Visual Aids: Enhance understanding with visual aids like slides, charts, and -D +videos. -F + Simplify Complex Information: Break down complex information into easily -FIGURE 22—Thin section photographs of microfacies of the Mesilla Valley -Formation at Cerro de Cristo Rey (Fig. 10). Sample numbers correspond to -unit numbers in Figure 10. A–B—Bioclastic packstone–rudstone contain- -ing densely packed bivalve fragments, subordinate echinoderms and other -bioclasts, a few detrital quartz grains, and rare glauconite. Sample G1. C— -Bioclastic packstone–rudstone composed of abundant recrystallized shell -fragments and agglutinated foraminiferans in silty matrix containing small - -angular quartz grains. Sample G3. D—Bioclastic siltstone composed of -small and larger fossils such as corals, bivalves, and agglutinated foraminif- -erans floating in silty matrix. Sample G4. E—Bioclastic wackestone con- -taining bivalve shell fragments, echinoderms, ostracods, and worm tubes -embedded in micritic to pelmicritic matrix. Sample G10. F—Worm tubes -in a bioclastic wackestone. Sample G10. Width of photographs is 6.3 mm, -except for F, which is 3.2 mm. +understandable parts. -shell fragments, rare foraminiferans, and -echinoderms. Shell fragments are oriented -parallel to the bedding in unit 26. +25 -Depositional environments +  Engage the Audience: Use questions and interactive elements to maintain -The lower part of the Mojado Formation -(Fig. 11) represents an upward-coarsening +engagement. -and shallowing succession, starting with -hummocky crossbedded sandstone of the -lower shoreface followed by upper shoreface -deposits represented by alternating, horizon- -tally laminated, ripple-laminated, and trough -crossbedded sandstone. A shallow-marine -environment is also indicated by a thin stro- -matolite lens within unit 10 (Fig. 11). +Advantages -The overlying trough crossbedded intervals -(Fig. 11, units 13 and 15) probably represent -fluvial channel-fill deposits of a delta plain, -which are again overlain by shallow-marine -siliciclastic sediments (units 16–22). The thin, -intercalated sandstone beds (ripple laminated, -horizontally laminated, trough crossbedded) -indicate high-energy conditions and may - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -121 - - A - -C - -E - -B - -D - -F + Educational Value: Provides valuable information to the audience. -FIGURE 23—Thin section photographs of microfacies of the Mancos Forma- -tion (A–B) and Mojado Formation (C–F) at Cerro de Cristo Rey (Figs. 11, 15). -Sample numbers correspond to unit numbers in Figures 11 and 15. A—Bio- -clastic wackestone with bivalve shell fragments, calcareous algae, and echi- -noderms floating in micrite. Sample I1. B—Bioclastic wackestone–packstone -with mostly bivalve fragments, echinoderms, and subordinate other bio- -clasts, micritic matrix, and some calcite cement. Sample I3. C—Fine-grained -quartz arenite composed mostly of monocrystalline quartz, rare polycrystalline + Clarity: Focuses on clear and concise communication. -quartz, cemented by quartz as authigenic overgrowths. Sample H3. D—Quartz -arenite showing abundant monocrystalline quartz grains with well-displayed -authigenic quartz overgrowths. Sample H3. E—Bioclastic wackestone contain- -ing abundant fragments of dasycladacean algae and bivalves, subordinate -foraminiferans, and echinoderms floating in micritic to pelmicritic matrix. -Sample H23. F—Bioturbated peloidal mudstone containing some quartz -grains and a few bioclasts. Sample H27. Width of photographs is 6.3 mm, -except for D, which is is 3.2 mm. +Disadvantages -represent storm deposits. A shallow-marine -environment is also indicated by a thin lens of -fossiliferous limestone (bioclastic wackestone; -Fig. 11, unit 23) that contains abundant dasy- -cladacean algae. + Potential for Boredom: Can be dry and unengaging if not delivered well. -The trough crossbedded sandstone of -the succeeding 7 m may represent fluvial -or upper shoreface deposits. The overlying, -horizontally laminated sandstone (Fig. 11, + Overload: Risk of overwhelming the audience with too much information. -unit 26) is of shallow-marine origin, indi- -cated by the marine fossils. +Practical Tips -Del Rio Formation + Focus on Key Points: Highlight the most important information. -Lithostratigraphy + Check for Understanding: Ask questions to ensure the audience understands the -The Del Rio Formation at our measured section -is approximately 33 m thick (Fig. 13). Lovejoy +material. -(1976) reported its thickness as 24–27 m. In our -measured section (Fig. 13), the lower 9.7 m -(most of unit 2) is a nodular limestone inter- -val composed of abundant limestone nodules -in a yellow clay slope (Fig. 12D). This inter- -val contains many fossils of Gyrostrea (=Exo- -gyra) whitneyi and gastropods. Above follows -a shale slope with some nodular limestone -and three thin limestone beds, each 0.2–0.3 m +4.1.7 Demonstrative Speaking -122 +Definition -New Mexico Geology +Demonstrative speaking involves showing the audience how to do something. This method is -November 2010, Volume 32, Number 4 +commonly used in tutorials, workshops, and cooking shows. - A +Techniques -C + Step-by-Step Instructions: Provide clear and detailed instructions. -E + Use Props and Visual Aids: Demonstrate each step using props and visual aids. -B + Engage the Audience: Invite audience participation to enhance understanding. -D + Rehearse the Demonstration: Practice the demonstration to ensure smooth -F +execution. -FIGURE 24—Thin section photographs of microfacies of the Del Rio Formation -(A–C) and Buda Formation (D–F) at Cerro de Cristo Rey (Figs. 13–14). Sample -numbers correspond to unit numbers in Figures 13–14. A—Peloidal wacke- -stone with some angular quartz grains and bioclasts such as ostracods, smaller -foraminiferans, and echinoid spines. Sample C3. B—A few bioclasts (bivalve -shell fragments, echinoderms, ostracods, foraminiferans) floating in siltstone -composed of angular quartz and carbonate grains, rare glauconite. Sample C7. -C—Bioclastic mudstone–wackestone containing many small and a few larger +Advantages -fossil fragments (bivalves, echinoderms) floating in micritic matrix. Sample C9. -D—Peloidal bioclastic mudstone containing a few angular quartz grains, some -bioclasts (ostracods, shell fragments, foraminiferans, echinoderms), and rare -glauconite. Sample D5. E—Bioclastic mudstone composed of gray micrite with -small and a few larger fossil fragments. Sample D7. F—Fine-grained bioclastic -wackestone containing mostly bivalve fragments, subordinately echinoderms, -ostracods, and foraminiferans floating in gray micrite. Sample D5. Width of -photographs A and D is 3.2 mm, and of photographs B, C, E, F is 6.3 mm. + Practical Knowledge: Provides hands-on learning experiences. -thick and forming ledges (units 3–7). Above -the uppermost limestone ledge, yellowish -shale with some nodular limestone is exposed -with a thickness of 12.8 m (unit 8). Gyrostrea -whitneyi is also common in the upper part. -Turnšek et al. (2003, p. 154) stated that Ilyma- -togyra arietina is the “dominant fossil” in the -Del Rio, but our field observations suggest it -is Gyrostrea whitneyi. The top of the Del Rio + Engagement: Interactive and engaging for the audience. -Formation is composed of nodular limestone -with thin yellow shale interbeds (unit 9, 4.5 m -thick), overlain by white limestone of the Buda -Formation (Fig. 13). +Disadvantages -Paleontology + Preparation Time: Requires significant preparation and practice. -The Del Rio Formation at Cerro de Cristo Rey -yields bivalves (especially Gyrostrea whitneyi + Potential for Mistakes: Live demonstrations can go wrong. -and Ilymatogyra arietina), gastropods, spatan- -goids, and ammonoids. In thin section we -observed foraminiferans, ostracods, echino- -derms, and annelid worm tubes. From the Del -Rio Formation at Cerro de Cristo Rey, Maul- -din (1985) and Mauldin and Cornell (1986) -determined 52 species of foraminiferans. The -foraminiferal assemblage is dominated by -rotaliine foraminiferans (38 species), which +Practical Tips -November 2010, Volume 32, Number 4 + Prepare Thoroughly: Ensure all materials and equipment are ready. -New Mexico Geology + Practice Multiple Times: Rehearse the demonstration several times to avoid -123 +mistakes. - B +26 -C + 4.1.8 Ceremonial Speaking -H +Definition -M +Ceremonial speaking involves delivering speeches for special occasions, such as weddings, -I +funerals, graduations, and award ceremonies. -N +Techniques -R + Understand the Occasion: Tailor your speech to the specific occasion and audience. -S + Use Appropriate Tone: Match the tone of your speech to the occasion, whether it's -T +celebratory, solemn, or formal. -U + Personal Touch: Include personal stories or anecdotes to make the speech more -D +meaningful. -F + Express Emotions: Convey appropriate emotions to connect with the audience. -J +Advantages -O + Memorability: Creates lasting memories for special occasions. -E + Emotional Impact: Can be highly emotional and impactful. -G +Disadvantages -K + Pressure: High expectations for memorable and impactful speeches. -P + Emotional Difficulty: Can be challenging to deliver in emotionally charged -V - -A - -L - -FIGURE 25—Some Cretaceous microfossils of the Cerro de Cristo Rey -sections. A, H—Permocalculus sp. (Finlay Formation, A unit 14, H unit 27, -Fig. 6). B, C—Boulina sp. (Finlay Formation, B unit 12, C unit 20, Fig. 6). -D, E, G, K—Acicularia or Terquemella sp. (Finlay Formation, D unit 14, E -unit 27, K unit 26, Fig. 6; Del Norte Formation, G unit 9, Fig. 7). F—Neom- -eris sp. (Finlay Formation, unit 20, Fig. 6). I, J—Salpingoporella sp. (Finlay -Formation, unit 24, Fig. 6). L—Pseudocyclammina sp. (Finlay Formation, unit - -27, Fig. 6). M, N—Textulariid foraminiferans (M—Del Norte Formation, unit 14, -Fig. 7; N—Smeltertown Formation, unit 7). O—Pseudonummuloculina heimi (Fin- -lay Formation, unit 10, Fig. 6). P—Lenticulina sp. (Del Rio Formation, unit 3, -Fig. 13). R, S—Orbitolinoid foraminiferans (Dictyoconus) (Finlay Formation, -unit 8, Fig. 6). T, U, V—Agglutinated foraminiferans (Ammobaculites? and/or -Cribratina) (T, U—Smeltertown Formation, unit 9, Fig. 8; V—Buda Formation, -unit 7, Fig. 14). Scale bar is 0.3 mm, except for K, which is 0.1 mm. - -constitute 82% of the individuals. Aggluti- -nated foraminiferans (suborder Textulariina) -comprise 12 species and 17% of all individu- -als. Two species belong to milioline foraminif- -erans, constituting 1% of the individuals. -These foraminiferans support correlation of -the Del Rio outcrops with the Grayson For- -mation elsewhere in Texas. - -Sedimentary petrography (Figs. 13, 24A–C) - -At the top of our section of the Mojado For- -mation, the basal interval of the overlying Del - -Rio Formation (Fig. 11, unit 27) is composed -of interbedded nodular limestone and shale. -The limestone beds are bioturbated peloidal -mudstone containing many peloids, a few -quartz grains, rare glauconite, and a few -small bioclasts such as ostracods, echino- -derms, and rare foraminiferans (Fig. 23F). -Similarly, in our section of the Del Rio -Formation (Fig. 13), the lowermost inter- -val (unit 2) is nodular limestone and shale -capped by a thin limestone bed (unit 3) that -consists of peloidal wackestone (Fig. 24A), -grading up into fine-grained sandstone - -containing abundant quartz. The peloidal -wackestone (unit 3) is indistinctly laminated -and contains as much as about 20% quartz -(with a grain size as much as approximately -0.2 mm) and rare glauconite. Most abundant -are small peloids and many small bioclasts -such as ostracods, foraminiferans, echinoid -spines, and some other small fossils, float- -ing in microspar (cement?). The amount of -quartz grains increases upward to about 30%, -and the peloidal wackestone is overlain by -fine-grained sandstone that contains about -50–60% angular quartz grains, rare glauconite, - -124 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - and a few bioclasts (including ostracods, echi- -noid spines, echinoderms, shell fragments, -and rare foraminiferans) in brownish micritic -matrix. - -The limestone bed of unit 5 of the Del Rio -Formation (Fig. 13) is a peloidal wackestone– -grainstone with abundant small peloids and -foraminiferans. Subordinate are ostracods, -echinoderms, echinoid spines, small quartz -grains (5–10%), a few micritic intraclasts, -and rare glauconite. The rock is indistinctly -laminated and calcite cemented. - -Unit 7 (Fig. 13) is a bioclastic limestone -with some small and larger bioclasts float- -ing in silty matrix (Fig. 24B). The silty -matrix is composed of predominantly -small angular quartz and carbonate grains, -subordinate glauconite, and some opaque -grains. Bioclasts include echinoderms, -ostracods, shell fragments (bivalves), and -rare agglutinated and other foraminiferans. -The rock is locally bioturbated and calcite -cemented. - -The nodular limestone of the Del Rio -Formation is a gray, nonlaminated, biotur- -bated bioclastic mudstone. The micritic to -pelmicritic matrix contains many small and -a few larger bioclasts. A few small quartz -grains and rare glauconite grains are pre- -sent. Bioclasts include ostracods, bivalve -shells, echinoderms, a few foraminiferans, -rare gastropods, and annelid worm tubes. -The uppermost limestone bed of the Del -Rio Formation (unit 9) is composed of bio- -turbated bioclastic mudstone (Fig. 24C). -Small bioclasts of ostracods, echinoderms, -foraminiferans, gastropods, bivalves, rare -quartz grains, and very rare glauconite -float in a micritic matrix. - -Depositional environments - -Del Rio Formation facies (Fig. 13) are -similar to those of the lower part of the -Del Norte Formation (Fig. 7) and indicate -shallow-marine deposition. Mauldin and -Cornell (1986) concluded that the forami- -niferans of the Del Rio Formation indicate a -deepening of marine waters (transgression) -that culminated in the middle part of the -Del Rio, followed by a regression during -deposition of the upper Del Rio. - -Mauldin and Cornell (1986) indicate that -the lower half of the Del Rio succession -is dominated by rotaliine foraminiferans; -the upper half is enriched in agglutinated -foraminiferans. The ratio of planktonic -to benthic foraminiferans is 0.1–0.6 in the -lower half, indicating water depths from -60 to 100 m. The value drops in the upper -half, indicating water depths of 20 m or less -(Mauldin and Cornell 1986). The foraminif- -eral assemblages indicate that the basal part -of the Del Rio Formation marks a transgres- -sion, and that a deeper shelf environment -(60–100 m) persisted until deposition of the -middle Del Rio Formation, followed by a -drop in sea level to water depths of approx- -imately 20 m in the upper half (Mauldin -and Cornell 1986). - -However, shale with limestone nodules -and nodular to wavy limestone indicate -deposition under quiet-water conditions in a -deeper, outer-shelf environment. This is also -indicated by the presence of glauconite, which -in modern oceans occurs at depths between -50 and 500 m (Flügel 2004). Even the wavy -to nodular limestone in the uppermost part -of the Del Rio Formation formed in a low- -energy environment. Lithologic evidence of a -drop in sea level to water depths of approxi- -mately 20 m in the upper half suggested by -Mauldin and Cornell (1986) is not observed. -Even the uppermost wavy to nodular lime- -stone (bioclastic mudstone) with intercalated -shale was deposited in a low-energy environ- -ment below wave base. High-energy deposits -such as grainstone or packstone are absent. - -Buda Formation - -Lithostratigraphy - -At Cerro de Cristo Rey, the best exposed and -most complete section of the Buda Forma- -tion is in Bowen Gulch where it is a promi- -nent, resistant interval of white limestone -that forms ridges and cuestas (Figs. 12E, -14). Here, the Buda Formation is 15.5 m -thick (Lovejoy 1976 reported its thickness as -approximately 12 m) and composed of alter- -nating limestone and shale with limestone -nodules. The lowermost limestone of the -Buda Formation is 0.8 m thick (unit 5) and -nodular, overlain by a thin nodular lime- -stone with shale and a prominent limestone -cliff 2.7 m thick (units 6–7). Above the lime- -stone cliff follows a succession of alternating -limestone and shale (units 8–21). Limestone -beds are 0.3–0.6 m thick and full of gastro- -pods. Shale intervals measure 0.3–2.5 m and -contain limestone nodules, which are locally -fossiliferous. - -The lithology of the Buda Formation sec- -tion exposed at Cerro de Cristo Rey most -resembles that of the middle member (Red -Light Member) of the Buda in Trans-Pecos -Texas (Reaser and Robinson 2003) in being -a mixture of thin-bedded lime mudstone -and nodular, marly limestone. In contrast, -the lower member (Lecheguilla Member) -and the upper member (Love Station Mem- -ber) of the Buda Formation in Trans-Pecos -Texas are ridge-forming beds of lithograph- -ic, muddy limestone (Reaser and Robinson -2003). Correlation of the Buda Formation -section at Cerro de Cristo Rey to only the -Red Light Member to the southeast thus -has a lithostratigraphic basis, but cannot -be confirmed biostratigraphically because -the entire Buda Formation falls within one -ammonoid zone, which is the biostrati- -graphic means by which Buda sections are -correlated (Young 1979). - -As Hook (2008; also see Cooper et al. 2008) -noted, the top of the Buda Formation usu- -ally displays evidence of erosion in the form -of microkarst features at the Buda–Mancos -contact. However, at Cerro de Cristo Rey the - -Buda–Mancos contact is poorly exposed and -tectonized (Fig. 12F), so evidence of micro- -karst features is not clear. The best evidence -of the Buda–Mancos unconformity at Cerro -de Cristo Rey is provided by ammonoid bio- -stratigraphy (Cobban et al. 2008). The base of -the Mancos Formation is in the upper middle -Cenomanian Acanthoceras amphibolum Zone, -whereas the upper Buda is in the lower Ceno- -manian Neophlycticeras (=Budaiceras) hyatti -Zone—thus, two lower Cenomanian and -four middle Cenomanian ammonoid zones -are missing at the Buda–Mancos contact, an -unconformity of more than one million years -duration (Cobban et al. 2008, fig. 2). - -Paleontology - -The Buda Formation yields many gastro- -pods (mostly turritellids) and bivalves, as -well as some ammonoids, dinoflagellates, -serpulids, spatangoid echinoderms, crus- -taceans, corals, and fish teeth (Böse 1910; -Young 1979; Cornell 1997; Turnšek et al. -2003). Ammonoids we have collected (and -will document elsewhere) place it in the -early Cenomanian zone of Neophlycticeras -(=Budaiceras) hyatti (Young 1979). - -Sedimentary petrography (Figs. 14, 24D–F) - -The basal limestone of the Buda Forma- -tion (Fig. 14, unit 5) is composed of fine- -grained, bioturbated bioclastic wackestone -containing a few larger skeletons (Fig. 24F). -Most abundant are mollusc shell frag- -ments. Less abundant are echinoderms, -ostracods, echinoid spines, foraminiferans, -and other recrystallized, indeterminate -skeletons. The skeletons float in a micritic -to pelmicritic matrix. This type grades into -bioclastic mudstone containing small and a -few larger bioclasts that float in gray mic- -rite (Fig. 24E). - -The thick limestone cliff of the Buda -Formation (Fig. 14, unit 7) consists of fine- -grained, bioturbated, bioclastic wackestone -with a few larger skeletons. Most abundant -are mollusc shell fragments (bivalves and -gastropods) and dasycladacean algae; -ostracods, echinoderms, echinoid spines, -smaller foraminiferans, and rare bryozo- -ans(?) are subordinate. Many skeletons, -particularly molluscs and algae, are recrys- -tallized. A few micritic intraclasts are pre- -sent. The skeletons are embedded in a mic- -ritic to pelmicritic matrix. - -The lowermost, gastropod-rich lime- -stone bed (Fig. 14, unit 9) is a bioturbated -bioclastic mudstone containing small and -rare larger bioclasts floating in a pelmic- -ritic matrix (Fig. 24D). Most skeletons are -derived from bivalves and gastropods, -and also from dasycladacean algae. Rare -ostracods, echinoderms, echinoid spines, -and smaller foraminiferans are present. -The rock also contains a few small, detrital -quartz grains. Another thin section indi- -cates a bioturbated bioclastic mudstone -composed of micritic matrix and small - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -125 - - bioclasts such as ostracods, bivalve shell -fragments, echinoid spines, echinoderms, -and smaller foraminiferans floating in the -matrix. Small quartz grains and rare glau- -conite are present. - -The uppermost limestone bed of the Buda -Formation (Fig. 14, unit 21) is composed of -bioturbated bioclastic wackestone contain- -ing a few larger bivalve and shell fragments -floating in a fine-grained bioclastic matrix. -Small recrystallized shell fragments and -algal debris, ostracods, subordinate small -gastropods, echinoderms, and smaller -foraminiferans are abundant. The fossils are -embedded in micritic matrix. Similarly, at -our Mancos Formation section, the topmost -Buda Formation (Fig. 15, unit 1) is bioturbat- -ed bioclastic wackestone that is poorly sort- -ed, contains mollusc shell debris (bivalves, -gastropods), calcareous algae, echinoderms, -echinoid spines, ostracods, worm tubes, -and foraminiferans. A few peloids and rare -glauconite grains are present. The matrix is -micrite. - -Depositional environments - -Regional deposition of the Buda Forma- -tion took place in shallow-marine waters -on a broad, shallow carbonate ramp that -extended across much of Texas and parts -of northern Mexico (Reaser and Robinson -2003). Our analysis of Buda microfacies -(most limestones are bioclastic wackestone -and mudstone) at Cerro de Cristo Rey indi- -cates they correspond well to the shallow- -water wackestone–mudstone facies of -Reaser and Robinson (2003). - -According to Reaser and Robinson (2003), -the lower member of the Buda Formation was -deposited on a broad open shelf with normal -marine water near the basin margin during -the last marine incursion by the Coman- -chean sea. The middle member is interpreted -as deposits of a shallow, slightly hypersaline -shelf during a drop in sea level. Sediments of -the upper member accumulated in a similar -environment (on the distal part of an open -shelf) as those of the lower member and rep- -resent the final highstand event. - -At Cerro de Cristo Rey (Fig. 14) muddy -textures, common bioturbation, diverse fos- -sil assemblage, and the presence of dasyclad- -acean algae indicate to us that the limestone -of the lower part of the Buda Formation was -deposited in an open, normal marine, low- -energy shelf environment below the wave -base. Carbonate sedimentation periodically -was interrupted by deposition of thin shale -intercalations. - -Intercalated limestone beds of the upper -part of the Buda are of the same composi- -tion as those of the lower part, indicating -a similar depositional environment. Shale -is more abundant in the upper part, indi- -cating high rates of fine siliciclastic influx -that was periodically interrupted, causing -the deposition of limestone beds. A great- -er amount of shale may be the result of a -slight drop in sea level, although deposi- -tion still occurred below wave base. - -Mancos Formation -("Boquillas Formation") - -Lithostratigraphy - -At Cerro de Cristo Rey, the exposed thick- -ness of the Mancos Formation is 18.6 m -(Fig. 15). The total estimated thickness is -approximately 110 m (Böse 1910; Lovejoy -1976), but we cannot confirm this estimate -with outcrop data. White, gastropod-rich, -nodular limestone of the Buda Formation -is sharply and unconformably overlain by -approximately 5 m of shale intercalated -with thin sandy limestone and limestone -beds. The uppermost limestone bed of this -interval is 0.2 m thick and forms a ledge. -Above is a 3.9-m-thick covered interval that -conceals a cross fault or fold. Overlying -strata are a 10-m-thick succession of shale -with a few thin sandstone beds. - -Böse (1910) and some subsequent work- -ers (Fig. 3) referred to these strata as the -Eagle Ford Formation (Shale). Strain (1976) -first brought the term Boquillas Formation -into Cerro de Cristo Rey, although he sug- -gested that the names Chispa Summit or -Ojinaga might better be applied. We prefer -to use the name Mancos Formation (Shale), -widely used across New Mexico, for this -shaley interval at the base of the Green- -horn cycle of deposition (e.g., Seager 1981; -Molenaar 1983; Lucas and Estep 1998a; -Lucas et al. 2000). Indeed, Mancos For- -mation is applied to the same lithosome -at nearby outcrops in the southern San -Andres Mountains, approximately 75 km -north of Cerro de Cristo Rey, and in the -southern Cooke’s Range, approximately -120 km northwest of Cerro de Cristo Rey. - -Paleontology - -Kennedy et al. (1988) described a middle -Cenomanian molluscan fauna of bivalves -(Ostrea beloiti Logan and Inoceramus arva- -nus Stephenson) and diverse ammonoids -(including Acanthoceras amphibolum) from -a thin bed of calcarenitic and coquinoidal -limestone near the base of the Mancos For- -mation (Fig. 15, bed 3). Turnšek et al. (2003) -also state that shark teeth (notably Pty- -chodus) are present in the Mancos Forma- -tion at Cerro de Cristo Rey. Cornell (1997) -reported dinoflagellate cysts. We observed -dasycladacean algae, foraminiferans, ostra- -cods, echinoderms, and bryozoans(?) in -thin section. - -Sedimentary petrography (Figs. 15, 23A–B) - -The lowermost thin limestone bed of the -Mancos Formation (Fig. 15, unit 3) is a -bioclastic wackestone–packstone–rudstone -that is poorly sorted and indistinctly lami- -nated (Fig. 23A–B). The rock contains a mic- -ritic matrix, and locally some calcite cement -is present. It is strongly recrystallized, and -it contains a few centimeter-sized bivalve - -shell fragments, abundant small bioclasts, -mostly shell debris and foraminiferans -(globigerinids?), subordinate echinoderms, -and other bioclasts. A few angular quartz -grains, ooids, and intraclasts are present. - -Depositional environments - -Deposition of the Mancos Formation took -place in normal marine waters. Muddy sea -bottoms supported characteristic benthic -organisms such as inoceramid bivalves, -and nektonic organisms such as ammo- -noids lived in the water column. - -Depositional cycles and -sequence boundaries +situations. + +Practical Tips + + Practice Emotion Management: Practice delivering your speech to manage + +emotions. + + Get Feedback: Seek feedback from close friends or family to refine your speech. + +4.1.9 Motivational Speaking + +Definition + +Motivational speaking aims to inspire and uplift the audience. This method is commonly used + +by motivational speakers, coaches, and leaders. + +Techniques + + Personal Stories: Share personal experiences that illustrate your message. + + Positive Language: Use positive and uplifting language to inspire the audience. + + Call to Action: Encourage the audience to take positive actions. + + Engage Emotions: Appeal to the audience's emotions to create a lasting impact. + +Advantages + + + +Inspiration: Can significantly motivate and inspire the audience. + +27 + +  Positive Impact: Promotes positive change and action. + +Disadvantages + + Skepticism: Some audience members may be skeptical of motivational speeches. + + High Expectations: Audiences often have high expectations for impactful messages. + +Practical Tips + + Be Authentic: Speak from the heart to connect with the audience genuinely. + + Use Powerful Quotes: Incorporate powerful quotes to reinforce your message. + +Public speaking encompasses a variety of methods, each suited to different contexts and + +purposes. Understanding and mastering these methods can significantly enhance your ability + +to communicate effectively. Whether delivering an impromptu speech, reading from + +4.2 ADVANTAGES AND DISADVANTAGES OF PUBLIC SPEAKING + +Public speaking is a powerful tool that can have significant personal, professional, and + +social impacts. While it offers numerous advantages, it also presents certain challenges. + +Understanding both the benefits and the potential drawbacks of public speaking can help you + +navigate and leverage this skill more effectively. + +4.2.1 Personal Advantages + +Confidence Boost + +Public speaking can significantly enhance your self-confidence. Successfully delivering a + +speech in front of an audience can make you feel more capable and self-assured in various + +aspects of life. + + Self-Esteem: Regular public speaking engagements can improve your self-esteem as + +you become more comfortable with expressing your ideas. + + Overcoming Fear: Conquering the fear of public speaking, known as glossophobia, + +can have a positive ripple effect on other areas where anxiety might be present. + +Communication Skills + +Engaging in public speaking helps develop your verbal and non-verbal communication skills. + + Articulation: Improves your ability to express thoughts clearly and concisely. + + Body Language: Enhances your understanding and use of effective body language to + +complement your message. + +Critical Thinking + +Preparing and delivering speeches enhances critical thinking skills. + +28 + +  Organization: Requires you to structure your thoughts logically and coherently. + + Analysis: Encourages deep analysis of topics and consideration of different + +perspectives. + +4.2.2 Professional Advantages + +Career Advancement + +Public speaking can open doors to career advancement opportunities. + + Leadership Roles: Effective communication is a key leadership skill, often leading to + +promotions and leadership positions. + + Visibility: Increases your visibility within your organization and industry, making you + +a more attractive candidate for advancement. + +Networking Opportunities + +Public speaking events provide opportunities to network and build professional relationships. + + + +Industry Connections: Engage with peers, mentors, and potential clients or + +collaborators. + + Reputation Building: Establish yourself as an expert or thought leader in your field. + +Skill Development + +Public speaking hones various skills that are valuable in the workplace. + + Persuasion: Develops your ability to persuade and influence others, crucial for sales, + +marketing, and leadership roles. + + Time Management: Enhances your ability to manage time effectively during + +presentations and meetings. + +4.2.3 Social Advantages + +Influence and Persuasion + +Public speaking enables you to influence and persuade others effectively. + + Advocacy: Use your voice to advocate for causes and inspire change. + + Community Impact: Engage with your community and contribute positively by + +sharing valuable insights and information. + +Enhanced Relationships + +Improves interpersonal relationships through better communication. + + Listening Skills: Encourages active listening, which is essential for effective + +communication in personal relationships. + + Empathy: Develops empathy by considering the audience's perspective and needs. + +29 + + Community Engagement + +Public speaking fosters active participation in community and social causes. + + Volunteering: Speak at community events or local organizations to support various + +causes. + + Awareness: Raise awareness about important issues and encourage collective action. + +4.2.4 Personal Disadvantages + +Anxiety and Fear + +Public speaking anxiety, or glossophobia, is a common and often debilitating fear. + + Physical Symptoms: Can cause physical symptoms such as sweating, shaking, and a + +racing heart. + + Mental Strain: Leads to stress and mental strain before and during speaking + +engagements. + +Time-Consuming Preparation + +Preparing a well-crafted speech can be time-consuming. + + Research: Requires extensive research to gather relevant information. + + Rehearsal: Involves significant time spent rehearsing to ensure a smooth delivery. + +Risk of Criticism + +Exposure to public scrutiny and potential criticism is a constant risk. + + Negative Feedback: Speakers must be prepared to handle negative feedback and + +criticism. + + Self-Doubt: Criticism can lead to self-doubt and decreased confidence. + +4.2.5 Professional Disadvantages + +Reputation Damage + +A poor public speaking performance can negatively impact your professional reputation. + + Perception: Can lead to negative perceptions about your competence and expertise. + + Career Impact: May affect career progression and opportunities for advancement. + +Miscommunication + +There is a risk of miscommunicating information or being misunderstood. + + Clarity: Ensuring your message is clear and understood by all audience members can + +be challenging. + + Consistency: Maintaining consistency in your message across different audiences and + +settings is crucial. + +30 + + High Stakes + +High-stakes speaking engagements can add pressure and stress. + + Performance Anxiety: The importance of certain speeches can increase anxiety and + +pressure. + + Consequences: Mistakes in high-stakes situations can have significant consequences. + +4.2.6 Social Disadvantages + +Exhaustion + +Frequent public speaking can be physically and mentally exhausting. + + Energy Drain: Continuous speaking engagements can lead to burnout and fatigue. + + Stress: Managing multiple speaking commitments can be stressful. + +Privacy Concerns + +Public exposure can lead to a loss of privacy. + + Public Scrutiny: Increased visibility can attract unwanted attention and scrutiny. + + Personal Boundaries: Balancing public life and personal privacy can be challenging. + +Misinterpretation + +There is a risk of being misinterpreted or taken out of context. + + Media: Statements can be misrepresented by the media or others. + + Audience: Different audience members may interpret the same message in various + +ways. + +4.2.7 Overcoming Disadvantages + +Managing Anxiety + +Several strategies can help manage public speaking anxiety. + + Preparation: Thorough preparation reduces anxiety by increasing familiarity with the + +material. + + Practice: Regular practice builds confidence and reduces fear. + + Relaxation Techniques: Use deep breathing, meditation, and positive affirmations to + +calm nerves. + +Efficient Preparation + +Efficient preparation can mitigate the time-consuming nature of public speaking. + + Time Management: Allocate specific times for research, writing, and rehearsing. + + Use of Technology: Leverage technology for efficient research and rehearsal. + + Seek Feedback: Obtain feedback from peers to refine and improve the speech. + +31 + + Handling Criticism + +Effective ways to handle criticism include: + + Constructive Feedback: View criticism as an opportunity for growth. + + Stay Professional: Maintain professionalism and composure in response to criticism. + + Learn and Improve: Use feedback to improve future performances. + +4.2.8 Enhancing the Advantages + +Building Confidence + +Strategies to build confidence include: + + Small Steps: Start with smaller, less intimidating speaking opportunities. + + Positive Visualization: Visualize successful speaking engagements to build + +confidence. + + + +Incremental Challenges: Gradually increase the complexity and size of speaking + +engagements. + +Career Advancement + +Maximize career advancement opportunities through: + + Seek Opportunities: Actively seek out speaking opportunities in your field. + + Professional Development: Attend workshops and courses to improve public + +speaking skills. + + Networking: Use public speaking events to network and build professional + +relationships. + +Community Engagement + +Enhance community engagement by: + + Get Involved: Participate in community events and causes that matter to you. + + Volunteer: Volunteer to speak at community gatherings or local organizations. + + Advocate: Use public speaking to advocate for social issues and causes. + +4.2.9 Practical Tips for Effective Public Speaking + +Know Your Audience + + Research: Understand the audience's interests, knowledge level, and expectations. + + Tailor Your Message: Adapt your content to resonate with the audience. + +Structure Your Speech + + Clear Outline: Create a clear and logical outline. + + Engaging Introduction: Start with a strong opening to capture attention. + +32 + +  Coherent Body: Develop your main points logically and clearly. + + Powerful Conclusion: End with a memorable closing statement. + +Practice and Rehearse + + Regular Practice: Rehearse multiple times to build confidence and familiarity. + + Feedback: Seek feedback from peers or mentors to refine your speech. + +Engage with the Audience + + Eye Contact: Maintain eye contact to build a connection. + + + +Interactive Elements: Use questions, polls, and discussions to engage the audience. + + Body Language: Use effective body language to enhance your message. + +4.3 Conclusion + +Public speaking offers numerous advantages, including personal growth, professional + +advancement, and positive social impact. However, it also presents challenges such as + +anxiety, time-consuming preparation, and potential criticism. By understanding these + +advantages and disadvantages, speakers can better prepare and deliver effective speeches. + +Overcoming the disadvantages and enhancing the advantages requires continuous practice, + +preparation, and a willingness to learn and grow. Effective public speaking can lead to + +significant personal and professional development, making it a valuable skill to master. + +33 + + Unit- V + +5.1 FIVE ACTIVITIES TO IMPROVE PUBLIC SPEAKING SKILLS: + +1. Toastmasters International: Joining a Toastmasters club is an excellent way to + +practice public speaking in a supportive environment. Members give speeches, + +receive feedback, and participate in impromptu speaking exercises. + +2. Debate Clubs: Participating in debate clubs helps develop argumentative and + +persuasive speaking skills. Debates require participants to think on their feet and + +present coherent arguments. + +3. Storytelling Workshops: Storytelling is a powerful communication tool. Workshops + +focusing on storytelling techniques can help speakers engage their audience and + +deliver memorable speeches. + +4. Mock Presentations: Organize mock presentations with friends or colleagues. + +Practice delivering your speech in front of an audience and solicit feedback for + +improvement. + +5. Public Speaking Classes: Enroll in public speaking classes or workshops. These + +courses often + +34 + + 5.2 Twenty Public Speaking Topics + +Instruction: Prepare a five minutes speech on any of the following topics: + +1. Child Abuse + +2. Racism + +3. Right to privacy + +4. Common sense + +5. Team Building + +6. Freedom + +7. Physical fitness + +8. Feminism + +9. True friendship + +10. Emotional Intelligence + +11. Personality Development + +12. Human rights + +13. Sophistication + +14. Festivals in India + +15. A good (or bad) boss + +16. Positive attitude + +17. If I were the Prime Minister + +18. Cell phones and Health + +19. Lessons from Nature + +20. Life without internet + +35 + + 5.3 Public Speaking Record + +Topic: ……………………………………………………. + + Date: …………………. + +Blue Print of the Presentation Introduction -Deposition of the Cretaceous strata exposed -at Cerro de Cristo Rey took place just off -of the northern margin of the Chihuahua -trough, one of the actively subsiding basins -bordering the late Aptian–early Cenoma- -nian Comanche Shelf, which extended from -eastern Arizona across part of southern -New Mexico and northern Mexico through -Texas and farther east (Fig. 26). The Creta- -ceous section at Cerro de Cristo Rey encom- -passes the uppermost Fredricksburg Group -(upper part of Finlay Formation), the entire -Washita Group (Del Norte, Smeltertown, -Muleros, Mesilla Valley, Mojado, Del Rio, -and Buda Formations), and the basal unit -(lower Mancos Formation) of the overlying -Greenhorn cycle. It represents a hierarchy of -at least three clearly definable depositional -cycles (Scott et al. 2003). Here, we identify -sequence boundaries (unconformities) and -transgressive-regressive (deepening-shal- -lowing) cycles in the Cretaceous section at -Cerro de Cristo Rey (Fig. 27). - -First order cycles and sequence boundaries - -Scott et al. (2000, 2003) consider the entire -Comanchean Series of Texas and adjacent -areas to represent a single, first-order cycle -of deposition. This cycle includes the Trin- -ity, Frederickburg, and Washita Groups of -late Aptian to early Cenomanian age. The -lower part of this cycle (Trinity Group and -most of the Fredericksburg Group) is not -exposed at Cerro de Cristo Rey. However, -the upper part of the cycle, and its upper -bounding unconformity (between the Buda -and Mancos Formations, see above), is well -exposed at Cerro de Cristo Rey. - -In northern Texas, the entire Comanchean -section is approximately 250 m thick, whereas -at Cerro de Cristo Rey that part of the Coman- -chean section exposed is thicker, approxi- -mately 350 m thick. This probably reflects -greater clastic influx into (and greater subsid- -ence of) the Cerro de Cristo Rey Comanchean -section because of its more landward position -relative to the more offshore sections in north- -ern and central Texas (Fig. 26). - -126 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - Formation at Cerro de Cristo Rey, whereas -the upper member indicates shallowing. - -A strong candidate for the next uncon- -formity (sequence boundary) is the base of -the Smeltertown Formation (Fig. 7, unit 24). -The sharp contact of shale on limestone at -the base of the Smeltertown Formation indi- -cates deepening and marks the next trans- -gressive surface. The absence of ammonites -indicative of the Craginites serratescens and -Eopachydiscus marcianus Zones at Cerro de -Cristo Rey may be indicative of a hiatus -at the base of the Smeltertown Formation -where strata of the Mortoniceras equidistans -Zone rest directly on strata of the Adkinsites -bravoensis Zone, or it may be that facies -restrictions precluded ammonite presence/ -preservation in part of the Del Norte–Smel- -tertown interval. The shoaling trend from -deeper to shallower shelf of cycle WA2, cor- -responding to the Smeltertown Formation, -is also observed at Cerro de Cristo Rey. - -Cycle WA3 in north Texas was deposited -during a period of general shoreline retreat -along the northern basin margin (Scott et al. -2003). An upward-deepening trend is also -observed within most of the Muleros Forma- -tion with a regressive trend in the uppermost -part. Cycle WA3 (Fig. 27) thus encompasses -the entire Muleros Formation, and begins -with an upward-deepening trend into off- -shore shale (Fig. 9, unit 2) and then shallow- -ing into the bioturbated limestones and Texi- -gryphaea packstones of the middle to upper -Muleros Formation (Fig. 9, units 3–14). WA3 -is an important regional cycle in New Mexi- -co as it is represented by the Tucumcari and -Glencairn Formations, which are the first -extensive Albian seaway in the northeastern -part of the state (e.g., Kues and Lucas 1987; -Scott et al. 2001). - -An abrupt deepening of marine waters -at the base of the Mesilla Valley Formation -marks the beginning of the next cycle (WA4, -Fig. 27). This cycle begins relatively deep in -offshore marine muds of member A (Fig. 10, -unit 2) and then shallows through sandstone -and Texigryphaea packstones of member B. In -north Texas, cycle WA4, which corresponds -to most of the Mesilla Valley Formation at -Cerro de Cristo Rey (Fig. 27), is a succession -from shallow shelf within the storm wave -base to a nearshore high-energy setting. A -shallowing trend is also recorded within the -Mesilla Valley Formation from member A to -member B. The sharp contact between lime- -stone (Muleros) and shale (Mesilla Valley) is -the likely sequence boundary at the base of -the WA4 cycle. - -Relatively offshore black shale of mem- -ber C of the Mesilla Valley Formation (Fig. 10, -unit 12) appears to represent an abrupt deep- -ening and thus can be interpreted as the base -of the next cycle, WA5. The succession shal- -lows upward into interbedded sandstone -and shale of the basal Mojado Formation. -We thus place the base of WA5 at the base of -member C of the Mesilla Valley Formation -(Fig. 27). A deepening trend above the first -thick, crossbedded fluvial sandstone of the -Mojado Formation (Fig. 10, unit 14; Fig. 11, - -FIGURE 26—Early Cretaceous depositional basins of the Comanchean Series (uppermost Lower and -lowermost Upper Cretaceous) in Texas, northern Mexico, and adjacent areas (after Scott et al. 2003). - -Second order cycles and sequence boundaries - -In Texas the Fredericksburg and Washita -Groups have long been regarded as sec- -ond-order cycles of deposition (e.g., Lozo -and Stricklin 1956; Hendricks 1967; Scott -et al. 2003). Across much of Texas, the top -of the Fredricksburg Group is a regional -hardground with subaerial exposure (uncon- -formity) overlain by the Kiamichi Formation -(base of the Washita Group; Amsbury 2003; -Scott et al. 2003). Thus, the Fredericksburg -and Washita Groups are unconformity- -bounded units with a basal clastic facies that -generally grades up into intercalated shallow- -marine carbonates and clastics. At Cerro de -Cristo Rey, the entire Washita Group second- -order cycle is exposed (Fig. 27). It is bounded -by distinct unconformities at the base of the -Del Norte and Mancos Formations. Only the -uppermost part of the Fredericksburg cycle -(upper Finlay Formation) and the lowermost -part of the Greenhorn cycle (lower Mancos -Formation) are exposed at Cerro de Cristo -Rey (Fig. 27). - -Third order cycles and sequence boundaries - -Based on the interpretation of Scott et al. -(2001, 2003), the Cretaceous section exposed -at Cerro de Cristo Rey represents eight -smaller-scale (third order?) depositional -cycles, six of which comprise the Wash- -ita Group section (Fig. 27). In the Washita -Group section, Scott et al. (2001, 2003) iden- -tify these six cycles (WA1 through WA6) -as shale-sandstone or mudstone-limestone -cycles. Our data strongly support this inter- -pretation, although we differ from Scott -et al. (2001, 2003) in our placement of the - -sequence boundaries (and thus stratigraph- -ic extent) of the lower three cycles. We thus -identify six upward-shallowing shale-sand- -stone or shale-limestone cycles bounded by -unconformities in the Washita Group sec- -tion at Cerro de Cristo Rey (Fig. 27). - -The Finlay Formation is the final high- -stand of the Fredericksburg second-order -cycle. Steinhoff (2003) identified high-fre- -quency cycles in the Finlay Formation that -begin with nodular, marly layers and grade -up into massive limestones and are 1–4.5 m -thick. The upper part of the Finlay Forma- -tion exposed at Cerro de Cristo Rey encom- -passes 12 such cycles (Fig. 6). These cycles -may correlate to 12 of the 18 high-frequen- -cy cycles that Steinhoff (2003) recognized as -part of an upper, high-frequency sequence 2 -of the Finlay Formation formed during a -time of relative sea-level highstand. By this -interpretation, the upper Finlay at Cerro de -Cristo Rey represents shoaling up followed -by sea level fall at the end of the last cycle -of deposition of the Fredericksburg Group -(Amsbury 2003; Scott et al. 2003). - -As noted above, Scott et al. (2000, 2003) -identified six third-order cycles of deposi- -tion in the Washita Group of northern Texas -(and at Cerro de Cristo Rey) that they labeled -WA1 through WA6. Thus, WA1 begins at the -base of the Del Norte Formation (= base of -Washita Group), where a ferruginous sand- -stone rests directly on lime mudstone at the -top of the Finlay Formation (Fig. 27). The con- -tact reflects an abrupt increase in water depth -and siliciclastic influx, thus representing a -transgressive surface. The transgressional- -progradational succession of cycle WA1 is -visible in the lower member of the Del Norte - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -127 - - unit 4) is indicated by bioturbated sand- -stones and interbedded siltstones (Fig. 11, -units 7–14) and is followed by shallowing, -indicated by prograding fluvial sandstones -(Fig. 11, units 15–25). - -In north Texas, WA5 represents a regres- -sive trend from deep to shallow shelf -conditions (Scott et al. 2003). This trend -is indicated by the facies of the Pawpaw -Formation, which northward becomes -very sandy with thin-bedded and trough -crossbedded sandstone of deltaic origin. -The overlying Main Street Formation also -indicates a northward change from a car- -bonate shelf to shoreface environment -with significant input of quartz. A similar -shoaling trend is clearly visible within the -Mojado Formation, which approximately -corresponds to cycle WA5: lower shore- -face to upper shoreface and finally to flu- -viodeltaic sandstone. - -Cycle WA6 in north Texas is represent- -ed by the Grayson and Buda Formations, -which correlate to the uppermost Mojado, -Del Rio, and Buda Formations at Cerro -de Cristo Rey (Scott et al. 2001, 2003). The -Grayson is characterized by a progressive -change from carbonate to clay. The lower -part is composed of shale with intercalated -ripple-laminated beds of carbonate silt and -sand interpreted as storm layers overlain -by clay mainly deposited below the storm -wave base. The middle part is represent- -ed by nodular wackestone and shale, and -the uppermost part by shale, interbedded -marl, and bioturbated limestone. Similar -lithologies are recognized within the Del -Rio Formation, although storm layers are -absent, and deposition took place in a low- -energy outer-shelf environment below the -storm wave base. - -The uppermost Mojado Formation pre- -serves a transgressive surface that we iden- -tify as the basal unconformity of the WA6 -cycle (Fig. 11, base of unit 26; Fig. 27). This -cycle deepens upward through the lower -Del Rio, then shallows into the platform -carbonates of the Buda Formation. The -Buda Formation represents the last marine -transgression of the Comanchean Seaway, -and Reaser and Robinson (2003) interpreted -Buda deposition (in Trans-Pecos Texas) as -representing a transgression across a broad, -low-relief carbonate ramp (lower Buda = -Lecheguilla Member), followed by a break -in transgression and a regression during -which terrigenous clays were deposited -(middle Buda = Red Light Member), cul- -minated by a final transgression and high- -stand (upper Buda = Love Station Member). -The Buda outcrop at Cerro de Cristo Rey is -the westernmost outcrop of the formation, -and is a relatively thin (< 20 m thick) and -shaly section of the formation. Given that -the Del Rio shows an upward-shallowing -trend, we think it most likely that the Buda -was deposited during regression or during -a prograding highstand. Limestone of the -Buda Formation at Cristo Rey suggests dep- -osition in an open-marine, low-energy shelf -environment below wave base. Increased - -deposition of shale in the upper part may -indicate a drop in sea level. - -The unconformity at the base of the Man- -cos Formation (Fig. 27) begins the next cycle -of deposition (Greenhorn cycle). It repre- -sents a major tectonic reorganization of the -Cretaceous marine basins of the American -Southwest (e.g., Mack 1987). At Cerro de -Cristo Rey, this was the change from dep- -osition along an extensive carbonate plat- -form (Comanchean Platform) that bordered -a rift basin (Chihuahua trough) to deposi- -tion in the retroarc foreland basin of the -Late Cretaceous Western Interior Seaway. -The persistence of cycles from the tectoni- -cally passive, open-marine margin of the -Gulf of Mexico into the tectonically active -Chihuahua trough suggests that regional, -if not global eustasy, not local tectonism, -drove late Early to early Late Cretaceous -sedimentation at Cerro de Cristo Rey. - -Acknowledgments - -We are grateful to George Cudahy for access -to land, and assistance in the field from -Josh Smith. We thank Daniel Vachard, who -helped us to identify foraminiferans and -algae in thin sections. Comments by Robert -W. Scott on an earlier version of this man- -uscript greatly improved its content and -organization. W. R. Dickinson, T. Lawton, -and R. W. Scott provided helpful reviews of -the submitted manuscript. - -References - -Adkins, W. C., 1932, The Mesozoic system in Texas; -in Sellards, E. H., Adkins, W. C., and Plummer, -F. B. (eds.), The geology of Texas: University of -Texas, Publication 3252, pp. 329–518. - -Amsbury, D. L., 2003, Stratigraphy of Fredericks- -burg Group (middle–upper Albian Cretaceous) -north-central Texas; in Scott, R. W. (ed.), Creta- -ceous stratigraphy and paleoecology, Texas and -Mexico: Perkins Memorial volume, GCSSEPM -Foundation, Special Publications in Geology, v. 1, -pp. 227–276. - -Böse, E., 1910, Monografia geológica y paleontologi- -cal del Cerro de Muleros cerca de Ciudad Juárez, -Estado de Chihuahua, y descripción de la fauna -cretácea de La Encanatada, Placer de Guadalupe, -Estado de Chihuahua: Instituto Geologico de -Mexico, Bulletin 25, 193 pp. - -Bullock, J. S., 1985, Biostratigraphic study of -foraminifera from the Smeltertown Formation -(Lower Cretaceous, Albian) of southeastern New -Mexico: Unpublished M.S. thesis, University of -Texas at El Paso, 155 pp. - -Bullock, J. S., and Cornell, W. C., 1986, Biostratigra- -phy and paleoecology of the Smeltertown Forma- -tion, Cerro de Cristo Rey, Doña Ana County, New -Mexico; in Hoffer, J. M. (ed.), Geology of south- -central New Mexico: El Paso Geological Society, -Guidebook, pp. 73–76. - -Cobban, W. A., 1987, Ammonite faunas of the Sarten -Sandstone (Cretaceous), Luna County, New Mexico: -U.S. Geological Survey, Bulletin 1641–B, pp. B1–B17. -Cobban, W. A., Hook, S. C., and McKinney, K. C., -2008, Upper Cretaceous molluscan record along -a transect from Virden, New Mexico, to Del Rio, -Texas: New Mexico Geology, v. 30, pp. 75–92. - -FIGURE 27—Summary of Cretaceous section at -Cerro de Cristo Rey showing sequence boundar- -ies and depositional cycles. - -128 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - Cooper, R. W., Cooper, D. A., Stevens, J. B., and -Stevens, M. S., 2008, Geology of the Hot Springs -Trail area, Ernst and San Vicente Members of the -Boquillas Formation; in Cooper, D. A. (ed.), The -southern extension of the Western Interior Sea- -way: Geology of the Big Bend National Park and -Trans-Pecos, Texas, 2008 Joint Annual Meeting of -the Geological Society of America and the Hous- -ton Geological Society, pp. 24–33. - -Córdoba, D. A., 1968, Juárez Mountains: West Texas -Geological Society, Guidebook, Publication 68-55, -pp. 85–86. - -Córdoba, D. A., 1969a, Mesozoic stratigraphy of -northeastern Chihuahua, México; in Córdoba, D. -A., Wengerd, S. A., and Shomaker, J. (eds.), Guide- -book of the border region: New Mexico Geologi- -cal Society, Guidebook 20, pp. 91–96. - -Córdoba, D. A., 1969b, Geologia del area de Cerro -de Muleros o Cristo Rey [inset on the] Carta -geológica de México, scale 1:100,000, hoja Ciudad -Juárez 13R-a(3); in Córdoba, D. A., Wengerd, S. A., -and Shomaker, J. (eds.), Guidebook of the border -region: New Mexico Geological Society, Guide- -book 20, rear pocket. - -Cornell, W. C., 1982, Dinoflagellate cysts from the -Mesilla Valley Shale (late Albian), El Paso, Texas -(abs.): American Association of Stratigraphic Pal- -ynologists, 14th Annual Meeting, October, 1981, -Program and Abstracts, p. 17. - -Cornell, W. C., 1997, Dinoflagellate cysts from the -Buda Limestone (Cenomanian) Cerro de Cristo -Rey, Doña Ana County, New Mexico: Review of -Paleobotany and Palynology, v. 98, pp. 153–157. -Cornell, W. C., and LeMone, D. V., 1987, The Anapra -Sandstone of Cerro de Cristo Rey, Doña Ana -County, New Mexico (abs.): New Mexico Journal -of Science, v. 27, p. 123. - -Cornell, W.C., LeMone, D.V., and Norland, W. D., -1991, Albian ophiuroids from Cerro de Cristo Rey, -Doña Ana County, New Mexico: Journal of Pale- -ontology, v. 65, pp. 1009–1013. - -Emory, W. H., 1857, Report of the United States and -Mexican boundary survey: U.S. 34th Congress, 1st -session, Senate ex. Doc. 108, v. 1, pt. 2, 174 pp. -Flügel, E., 2004, Microfacies of carbonate rocks: -analysis, interpretation and application: Springer, -Berlin, 976 pp. - -Hendricks, L., 1967, Comanchean stratigraphy of the -Cretaceous of north Texas; in Hendricks, L. (ed.), -Comanchean (Lower Cretaceous) stratigraphy -and paleontology of Texas: Society of Economic -Paleontologists and Mineralogists (SEPM), Perm- -ian Basin Section, Publication 67-8, pp. 51–63. -Hobday, D. K., and Morton, R. A., 1984, Lower Cre- -taceous shelf storm deposits, northeast Texas; in -Tillman, R. W., and Siemers, C. T. (eds.), Siliciclas- -tic shelf sediments: Society of Economic Paleon- -tologists and Mineralogists (SEPM), Special Pub- -lication 34, pp. 205–213. - -Hook, S. C., 2008, Gallery of geology—Sierra de -Cristo Rey: New Mexico Geology, v. 30, pp. 93–94. -Imlay, R. W., 1944, Cretaceous formations of Central -America and Mexico: American Association of -Petroleum Geologists, Bulletin 28, pp. 878–1028. -Kappus, E. J., 2007, Middle Cretaceous dinosaur -tracks at Cerro de Cristo Rey, Sunland Park, New -Mexico, and a comparison with other paleocoastal -tracksites of the southwestern US: Unpublished -M.S. thesis, University of Texas at El Paso, 152 pp. -Kappus, E. J., Lucas, S. G., Heckert, A. B., Hunt, A. P., -and Lockley, M. G., 2003, Dinosaur footprints from -the Lower Cretaceous Sarten Member of the Moja- -do Formation at Cerro de Cristo Rey, Doña Ana -County, New Mexico: Ichnos, v. 10, pp. 263–267. -Kennedy, W. J., Cobban, W. A., and Hook, S. C., 1988, -Middle Cenomanian (Late Cretaceous) molluscan -fauna from the base of the Boquillas Formation, -Cerro de Muleros, Doña Ana County, New Mexi- -co; in Contributions to Late Cretaceous paleontol- -ogy and stratigraphy of New Mexico, Part II: New -Mexico Bureau of Mines and Mineral Resources, -Bulletin 114, pp. 35–44. - -Kollmann, H. A., Decker, K., and LeMone, D. V., -2003, Facies control of Lower Cretaceous gastro- -pod assemblages, southwestern United States; in -Scott, R. W. (ed.), Cretaceous stratigraphy and -paleoecology, Texas and Mexico: Perkins Memo- -rial volume, GCSSEPM Foundation, Special Pub- -lications in Geology, v. 1, pp. 101–146. - -Kues, B. S., 1989, Taxonomy and variability of three -Texigryphaea (Bivalvia) species from their Lower Cre- -taceous (Albian) type localities in New Mexico and -Oklahoma: Journal of Paleontology, v. 63, pp. 454–483. -Kues, B. S., 1997, New bivalve taxa from the Tucumcari -Formation (Cretaceous, Albian), New Mexico, and -the biostratigraphic significance of the basal Tucum- -cari fauna: Journal of Paleontology, v. 71, pp. 820–839. -Kues, B. S., and Lucas, S. G., 1987, Cretaceous stra- -tigraphy and paleontology in the Dry Cimarron -Valley, New Mexico, Colorado, and Oklahoma; in -Lucas, S. G., and Hunt, A. P. (eds.), Northeastern -New Mexico: New Mexico Geological Society, -Guidebook 38, pp. 167–198. - -LeMone, D. V., and Simpson, R. D., 1983, Creta- -ceous biostratigraphy of the Franklin Mountains, -El Paso County, Texas; in Allen, R. (ed.), Delaware -Basin: West Texas Geological Society, Publication -82–76, pp. 85–87. - -Lovejoy, E. M. P., 1976, Geology of Cerro de Cristo -Rey uplift, Chihuahua and New Mexico: New -Mexico Bureau of Mines and Mineral Resources, -Memoir 31, 84 pp. - -Lozo, F. E., and Stricklin, F. L., Jr., 1956, Stratigraph- -ic notes on the outcrop basal Cretaceous, central -Texas: Gulf Coast Association of Geological Socie- -ties, Transactions, v. 6, pp. 67–78. - -Lucas, S. G., and Estep, J. W., 1998a, Cretaceous -stratigraphy and biostratigraphy in the southern -San Andres Mountains, Doña Ana County, New -Mexico; in Mack, G. H., Austin, G. S., and Barker, -J. M. (eds.), Las Cruces country II: New Mexico -Geological Society, Guidebook 49, pp. 187–196. -Lucas, S. G., and Estep, J. W., 1998b, Lithostratigra- -phy and biostratigraphy of the Lower-Middle Cre- -taceous Bisbee Group, southwestern New Mexico, -USA; in Lucas, S. G., Kirkland, J. I., and Estep, J. -W. (eds.), Lower and middle Cretaceous terrestrial -ecosystems: New Mexico Museum of Natural His- -tory and Science, Bulletin 14, pp. 39–55. - -Lucas, S. G., and Estep, J. W., 2000, Lower Cretaceous -stratigraphy, correlation and paleogeography of -New Mexico; in Lucas, S. G. (ed.), New Mexico's -fossil record 2: New Mexico Museum of Natural -History and Science, Bulletin 16, pp. 45–61. - -Lucas, S. G., Estep, J. W., Boucher, L. D., and -Anderson, B. G., 2000, Cretaceous stratigraphy, -biostratigraphy and depositional environments -near Virden, Hidalgo County, New Mexico; in -Lucas, S. G. (ed.), New Mexico's fossil record 2: -New Mexico Museum of Natural History and Sci- -ence, Bulletin 16, pp. 107–119. - -Lucas, S. G., Kues, B. S., Hayden, S. N., Allen, B. -D., Kietzke, K. K., Williamson, T. E., Sealey, P., -and Pence, R., 1988, Cretaceous stratigraphy and -biostratigraphy, Cooke’s Range, Luna County, New -Mexico; in Mack, G. H., Lawton, T. F., and Lucas, S. -G. (eds.), Cretaceous and Laramide tectonic evo- -lution of southwestern New Mexico: New Mexico -Geological Society, Guidebook 39, pp. 143–167. - -Mack, G. H., 1987, Mid-Cretaceous (late Albian) -change from rift to retroarc foreland basin in -southwestern New Mexico: Geological Society of -America, Bulletin, v. 98, pp. 507–514. - -Mauldin, R. A., 1985, Foraminiferal biostratigraphy, -paleoecology, and correlation of the Del Rio Clay -(Cenomanian), from Big Bend National Park, -Brewster County, Texas, to the Cerro de Muleros -area, Doña Ana County, New Mexico: Unpublished -M.S. thesis, University of Texas at El Paso, 89 pp. -Mauldin, R. A., and Cornell, W. C., 1986, Paleoecol- -ogy and biostratigraphic correlation of the Del Rio -Clay, Cerro de Cristo Rey, Doña Ana County, New -Mexico; in Hoffer, J. M. (ed.), Geology of south- -central New Mexico: El Paso Geological Society, -Guidebook, pp. 77–84. - -Molenaar, C. M., 1983, Major depositional cycles and -regional correlations of Upper Cretaceous rocks, -southern Colorado Plateau and adjacent areas; in -Reynolds, M. W., and Dolly, E. D. (eds.), Mesozoic -paleogeography of west-central United States: Soci- -ety of Economic Peleontologists and Mineralogists -(SEPM), Rocky Mountain Section, Rocky Mountain -Paleogeography Symposium 2, pp. 201–224. - -Nye, O. B., Jr., and LeMone, D. V., 1978, Multilami- -nar growth in Reptomulticava texana, a new species -of cyclostome bryozoa: Journal of Paleontology, -v. 52, pp. 830–885. - -Reaser, D. F., and Malott, V. E., 1985, Finlay Lime- -stone: A key to geological interpretation in west- -ern Trans-Pecos Texas and northeastern Texas and -northeastern Chihuahua, Mexico; in Dickerson, P. -W., and Muehlberger, W. R. (eds.), Structure and -tectonics of Trans-Pecos Texas: West Texas Geo- -logical Society, Publication 85–81, pp. 213–219. -Reaser, D. F., and Robinson, W. C., 2003, Cretaceous -Buda Limestone in west Texas and northern Mex- -ico; in Scott, R. W. (ed.), Cretaceous stratigraphy -and paleoecology, Texas and Mexico: Perkins -Memorial volume, GCSSEPM Foundation, Special -Publications in Geology, v. 1, pp. 337–373. - -Richardson, G. B., 1909, Description of the El Paso -quadrangle Texas; in Geologic atlas of the United -States, El Paso folio no. 166: U.S. Geological Sur- -vey, 11 pp. - -Scott, R. W., 1970, Stratigraphy and sedimentary -environments of Lower Cretaceous rocks, southern -Western Interior: American Association of Petro- -leum Geologists, Bulletin, v. 54, pp. 1225–1244. - -Scott, R. W., and Kidson, E. J., 1977, Lower Cretaceous -depositional systems, west Texas; in Bebout, D. G., -and Loucks, R. G. (eds.), Cretaceous carbonates of -Texas and Mexico: Texas Bureau of Economic Geol- -ogy, Report of Investigations 89, pp. 169–181. - -Scott, R. W., Holbrook, J. M., Evetts, M. J., and -Oboh-Ikuenobe, F. E., 2001, Albian–Cenomanian -depositional cycles transgressed from Chihua- -hua trough to Western Interior; in Lucas, S. G., -and Ulmer-Scholle, D. S. (eds.), Geology of the -Llano Estacado: New Mexico Geological Society, -Guidebook 52, pp. 221–228. - -Scott, R. W., Hooman, L., and Fee, D. W., 1975, Den- -sity-current strata in Lower Cretaceous Washita -Group, north-central Texas: Journal of Sedimen- -tary Petrology, v. 45, pp. 562–575. - -Scott, R. W., Schlager, W., Fouke, B., and Neder- -bragt, S. A., 2000, Are mid-Cretaceous eustatic -events recorded in Middle East carbonate plat- -forms? in Alsharhan, A. S., and Scott, R. W. (eds.), -Middle East models of Jurassic/Cretaceous car- -bonate systems: SEPM (Society for Sedimentary -Geology), Special Publication 69, pp. 73–84. - -Scott, R. W., Benson, D. G., Morin, R. W., Shaffer, B. L., -and Oboh-Ikuenobe, F. E., 2003, Integrated Albian– -lower Cenomanian chronostratigraphy standard, -Trinity River section, Texas; in Scott, R. W. (ed.), -Cretaceous stratigraphy and paleoecology, Texas -and Mexico: Perkins Memorial volume, GCSSEPM -Foundation, Special Publications in Geology, v. 1, -pp. 277–334. - -Seager, W. R., 1981, Geology of Organ Mountains -and southern San Andres Mountains, New Mex- -ico: New Mexico Bureau of Mines and Mineral -Resources, Memoir 36, 97 pp. - -Stanton, T. W., and Vaughan, T. W., 1896, Sections of -the Cretaceous at El Paso, Texas: American Journal -of Science, 4th series, v. 1, pp. 21–26. - -Steinhoff, D., 2003, Albian Finlay Formation strato- -type, Rimrock escarpment, west Texas; in Scott, R. -W. (ed.), Cretaceous stratigraphy and paleoecol- -ogy, Texas and Mexico: Perkins Memorial volume, -GCSSEPM Foundation, Special Publications in -Geology, v. 1, pp. 69–99. - -Strain, W. S., 1968, Cerro de Muleros (Cerro de Cris- -to Rey): West Texas Geological Society, Delaware -Basin Exploration Guide, Publication 68–55, p. 82. -Strain, W. S., 1976, New formation names in the Cre- -taceous at Cerro de Cristo Rey, Doña Ana County, -New Mexico; in Lovejoy, E. M. P., Geology of -Cerro de Cristo Rey uplift, Chihuahua and New -Mexico: New Mexico Bureau of Mines and Min- -eral Resources, Memoir 31, pp. 77–82. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -129 - - Turnšek, D., LeMone, D. V., and Scott, R. W., 2003, -Tethyan Albian corals, Cerro de Cristo Rey uplift, -Chihuahua and New Mexico; in Scott, R. W. -(ed.), Cretaceous stratigraphy and paleoecology, -Texas and Mexico: Perkins Memorial volume, -GCSSEPM Foundation, Special Publications in -Geology, v. 1, pp. 147–185. - -Young, K., 1966, Texas Mojsisovicinae (Ammonoidea) -and the zonation of the Fredericksburg: Geological -Society of America, Memoir 100, 225 pp. - -Young, K., 1979, Lower Cenomanian and late Albian -(Cretaceous) ammonites, especially Lyelliceridae, -of Texas and Mexico: Texas Memorial Museum, -Bulletin 26, 99 pp. - -Zeller, R. A., Jr., 1965, Stratigraphy of the Big Hatch- -et Mountains area, New Mexico: New Mexico -Bureau of Mines and Mineral Resources, Mem- -oir 16, 128 pp. - -Appendix - -Location of measured sections - -Finlay Formation (Fig. 6)—base at UTM -Zone 13, 355011E 3518134N (NAD 27), -top at 355064E 3518629N. - -Del Norte Formation (Fig. 7)—base at top of -section A, top at UTM Zone 13, 355084E -3517962N (NAD 27). - -Smeltertown Formation (Fig. 8)—base at -UTM Zone 13, 355178E 3517669N, top at -355040E 3517595N. - -Muleros Formation (Fig. 9)—base at UTM -Zone 13, 354953E 3517640N, top at -354922E 3517683N. - -Mesilla Valley Formation (Fig. 10)—base at -UTM Zone 13, 354508E 3518122N, top at -354535E 3518200N. - -Mojado Formation (Fig. 11)—base at -UTM Zone 13, 354003E 3518452N, top -at 353967E 3518553N. - -Del Rio Formation (Fig. 13)—base at UTM -Zone 13, 353899E 3118636N, top at -353871E 3518678N. - -Buda Formation (Fig. 14)—base at UTM -Zone 13, 353881E 3518823N, top at -353859E 3518828N. - -Mancos Formation (Fig. 15)—base at UTM -Zone 13, 353534E 3518775N, top at -353524E 3518792N. - -130 - -New Mexico Geology +Main Idea 1 + +Main idea 2 + +Main Idea 3 + +Conclusion + +Observation by the Faculty: + +Details + +Eye Contact +Body Language +Language +Content +Presentation Skills + +Marks Awarded + +Remarks + +10 + +36 -November 2010, Volume 32, Number 4 +(Signature of the Faculty) diff --git a/q128/random_k2/question.json b/q128/random_k2/question.json index 90e61a5f3dda042a5b8c20543e18ea5e6624a529..59db68d83e02cefaa591e6c7c16513804451d051 100644 --- a/q128/random_k2/question.json +++ b/q128/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "owners-manual-2170416.pdf", - "web_974a02e9fcc23ca9.pdf", - "web_ddfd3d214b1f0ea4.pdf" + "web_d3545b20cb487e4b.pdf", + "web_9deb1a47e107dc04.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q128/random_k2/random_1.md b/q128/random_k2/random_1.md index 0f5db4f84cd2bbe84dc239b53c3a2c97d631e42e..3b59653be960a7700f9cff39752b4e24bce89425 100644 --- a/q128/random_k2/random_1.md +++ b/q128/random_k2/random_1.md @@ -1,279 +1,3031 @@ -COMPETITOR ANALYSIS +FOR RELEASE JULY 26, 2018 -© InnoValor, 2018 -© InnoValor, 2018 +BY Cary Funk and Meg Hefferon -How to use -Template -Example +FOR MEDIA OR OTHER INQUIRIES: - Competitor Analysis +Cary Funk, Director, Science and Society Research +Haley Nolan, Communications Assistant -A Competitor Analysis helps you to understand your -competition. It supports you in creating your unique position in -the market. With this analysis you make an assessment of the -strengths and weaknesses of your current and potential -competitors. With a better understanding of your competition -you can choose and implement effective strategies that will -improve your competitive advantage. +202.419.4372 -Tips for use +www.pewresearch.org -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. This tool can easily be used by -yourself, or with colleagues. Print the template and use the tool -to create an understanding of your competition. +RECOMMENDED CITATION -Have a look at the example of Fleur Flowers to get an -understanding of a Competitor Analysis. +Pew Research Center, July 2018, “Public Views of Gene +Editing for Babies Depend on How It Would Be Used” -How to use + 2 -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. Follow the steps to complete your -own Competitor Analysis. +PEW RESEARCH CENTER -Summarize your target group +About Pew Research Center -Shortly describe your target group. Who are the customers that -buy your offering? Make the description as specific as possible. If +Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes +and trends shaping America and the world. It does not take policy positions. The Center conducts +public opinion polling, demographic research, content analysis and other data-driven social +science research. It studies U.S. politics and policy; journalism and media; internet, science and +technology; religion and public life; Hispanic trends; global attitudes and trends; and U.S. social +and demographic trends. All of the Center’s reports are available at www.pewresearch.org. Pew +Research Center is a subsidiary of The Pew Charitable Trusts, its primary funder. -© InnoValor, 2018 +© Pew Research Center 2018 -your target group consist of businesses you can mention their -size, industry or their offering. If your target group consist of -consumers you can mention their age, goals, gender, education -level or interests. +www.pewresearch.org -Identify your competitors + 3 -Write the names for up to three competitors in the first row. To -help you generate this list, think of where your customers would -go if they cannot buy your product or service. +PEW RESEARCH CENTER -Determine competitors’ strengths and weaknesses +Public Views of Gene Editing for Babies Depend on How It +Would Be Used -Now think like your customers. Why would a customer go for the -services of your competitors instead of coming to your company? -You can describe this by thinking about the offerings of your -competitors. It would even be better when you visit your -competitors and experience what their customers experience. -Write down three strengths of each competitor. Also consider -why customers would not go to your competitors. Write down -three weaknesses of each competitor. +Americans’ views on the +appropriateness of changing +a baby’s genetic +characteristics depend in +large part on the intended +purpose and on whether or +not human embryos would be +used in testing these +techniques. A majority of +Americans support the idea +of using gene editing with the +goal of delivering direct +health benefits for babies, but +at the same time, a majority +considers the use of such +techniques to boost a baby’s +intelligence something that +takes technology “too far.” -Describe the key findings +A majority of U.S. adults say changing a baby’s genes +to treat a serious congenital disease is appropriate -Finally, in the fourth column of the template write down the key -findings for the strengths. What stands out? What patterns can -you discover from your competitors? And which opportunities -can you come up with? Also, write down your key -findings for the weaknesses. What are the aspects that +% of U.S. adults who say changing a baby’s genetic characteristics for each of +the following reasons is … - your competitors don’t do well? How could you avoid making -the same mistakes? What should you do to serve your customers -in such a way that you differentiate from your competitors in a -positive way? +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” -© InnoValor, 2018 +PEW RESEARCH CENTER - competitor analysis +About seven-in-ten Americans (72%) say that changing an unborn baby’s genetic characteristics to +treat a serious disease or condition that the baby would have at birth is an appropriate use of +medical technology, while 27% say this would be taking technology too far. A somewhat smaller +share of Americans say gene editing to reduce a baby’s risk of developing a serious disease or +condition over their lifetime is appropriate (60% say this, while 38% say it would be taking +medical technology too far). But just 19% of Americans say it would be appropriate to use gene +editing to make a baby more intelligent; eight-in-ten (80%) say this would be taking medical +technology too far. -name +These are some of the findings from a new Pew Research Center survey conducted April 23-May 6, +2018, among 2,537 U.S. adults. -target group +www.pewresearch.org -competitors + 4 -most important findings +PEW RESEARCH CENTER -s -h -t -g -n -e -r -t -s +While public discussions about potentially altering a baby’s genetic makeup have been ongoing for +decades, the development of a new gene-splicing technology – known as CRISPR – has accelerated +the debate and brought new urgency to better understanding public opinion about gene editing as +well as the broader social, ethical and policy implications ahead.1 -s -e -s -s -e -n -k -a -e -w +Previous Pew Research Center surveys have tracked public opinion about gene editing using +different question wording and, in some cases, different polling methodologies. As such, those +findings are not directly comparable to this new survey. However, the broad pattern – that public +support for gene editing varies with its intended purpose – is consistent with a 2016 study that +explored public views about the possibility of using gene editing to “enhance” a baby’s health over +the course of their lifetime and a 2014 survey.2 -date +Regardless of the intended purpose of gene +editing, experts acknowledge that further +development of these techniques will likely +involve testing in human embryos. In August +2017, research scientists in the United States +reported the first successful use of gene editing +in human embryos to eliminate an inherited +condition. When asked to consider the +possibility that the development of gene +editing would involve testing on human +embryos, one third of Americans (33%) say +this would be appropriate while about two- +thirds (65%) say this would be taking medical +technology too far. - Example Fleurs Flowers +65% of Americans say if gene editing +relies on embryonic testing, it would +take technology too far -industry +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … -Retail +Taking medical +technology too far -products & -services +Appropriate use of +medical technology -Flowers +65 -size of -business +33 -2 full-timers 2 part-timers +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” -revenue +PEW RESEARCH CENTER -€ 37,000 +The 2016 Pew Research Center survey using +different question wording found that most adults said testing on human embryos to develop gene +editing would make gene editing less acceptable for them. -location +1 The National Academies of Sciences and the National Academy of Medicine launched a consensus study and at least two international +summits to consider these issues. +2 Other research on public views of gene editing has also shown that the public distinguishes its intended purpose. Scheufele et al. (2017) +found that Americans were more supportive of gene editing applications that treat medical conditions than those aimed at enhancing health. +Similarly, Whitman (2018) also found stronger public support for gene editing to cure disease than gene editing to make people stronger or +more intelligent. -Main street - Bloomfield +www.pewresearch.org -Fleurs Flowers is a flower shop in a small town. Fleur sells -flowers to a small group of regular clients who regularly -purchase fresh flowers from her shop. Her customers generally -consist of men and women aged 25. The customers often just -bought a house and flowers are the perfect way to bring -atmosphere to the new place. In addition, she has customers -who buy flowers for birthday and holiday occasions. With -Valentine’s and Mother's Day she is working overtime to meet -the demand + 5 -Fleurs Flowers’ competitor analysis +PEW RESEARCH CENTER -A company like Fleurs Flowers has a difficult time in today's -society. She has a physical store that is open during shopping -hours. Her customers are people who also work at those times. -These customers search for convenience in their shopping. -Internet shopping is very popular amongst these people. Fleur -wants to increase her appeal to her target group. She -investigates the advantages and disadvantages of the sales -strategies of her main competitors. For her research Fleur visits -her competitors and buys their flowers. The results of Fleur’s -competitor analysis can be read in the example on the next page. +Several patterns emerge in +public opinion about gene +editing in these varying +contexts. There are large +differences in acceptance of +gene editing between the +highly religious and less +religious. In addition, there is +a gender gap in views about +gene editing, with women +less accepting of it than men. +People with higher levels of +science knowledge and +greater familiarity with gene +editing also tend to be more +accepting of it. -© InnoValor, 2018 +Religious Americans are +more likely to view gene +editing negatively - competitor analysis +Highly religious Americans are more likely to see gene +editing for babies as taking medical technology too far -Fleurs Flowers +% of U.S. adults in each religious commitment group who say changing a +baby’s genetic characteristics for each of the following reasons is … -name +Taking medical technology too far -target group +Appropriate use of medical technology -Men or women from 25 years old with an average salary +Treat a serious disease or condition the baby would have at birth +Among those ___ on the religious commitment index -fromyouflowers.com +High -Lidl +Medium -Valentino flowershop +Low -competitors +40 -most important findings +57 -Offers flowers online +27 -The price for flowers is low +18 -Customers can order -flowers online but have to -pick them up at the store +72 -Delivers at home though -whole the region within 24 -hours +82 -Customers can buy flowers -while doing groceries, they do -not have to visit a different store +Reduce risk of a serious disease or condition that could occur over their +lifetime -When ordering flowers above a -certain amount the customer -will receive a free vase +High -Package deals – get a -discount on a vase when -flowers are bought +53 -Everyday there are new flowers +Medium -A lot of time is spend on -wrapping the flowers as a gift +Low -An online shop seems to be more -and more important. Customers -want to be able to order flowers -from their home. Customers also -like it when they get something for -free or a discount on something -when they buy flowers +41 -There is no way that a -customer can see the flowers -before they are ordered +27 -Customers cannot create -their own piece since the -flowers are all prearranged +Make the baby more intelligent -The price for flowers is higher -than in other shops +46 -Ordering flowers is less -personal +58 -The flowers are made in mass +73 -Ordering flowers takes -quite some time +6 -Customers cannot create -their own bouquet of flowers +High -Buying flowers is not a special -occasion since it is been done -during doing groceries +93 -The shop is not as much open -as customers would like, for -instance it is closed on Mondays +82 -Customers seem to think that -buying flowers is a personal and -special occasion. They do not like -it when buying flowers is less -personal. They also do not like that -they cannot create their own -bouquet +72 -s -h -t -g -n -e -r -t -s +18 -s -e -s -s -e -n -k -a -e -w +Low -date +Medium + +Americans who are high in +religious commitment – that +is, those who attend religious +services at least weekly, pray +at least daily and say that +religion is very important in +their lives – are less inclined +than those with either +medium or low levels of +religious commitment to say +that gene editing is an appropriate use of medical technology.3 For example, those high in religious +commitment are closely divided over whether it is appropriate to use gene editing to reduce a + +Note: Respondents who did not give an answer are not shown. See Methodology for details +on index of religious commitment. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +28 + +3 A person is considered to have a high level of religious commitment if they attend religious services at least weekly, pray at least daily and +say religion is very important in their life. People are classified as having low commitment if they say religion is not too or not at all important +in their lives and that they seldom or never attend religious services or pray. All others are classified as having medium commitment. + +www.pewresearch.org + + 6 + +PEW RESEARCH CENTER + +baby’s risk of disease later in life; 46% say this is appropriate, while 53% consider it taking +technology too far. In contrast, roughly three-quarters of those low in religious commitment (73%) +say gene editing to reduce a baby’s risk of developing a serious disease or condition is an +appropriate use of medical technology. And, while a 57% majority of those high in religious +commitment say gene editing to treat a congenital disorder in a baby is an appropriate use of +medical technology, a much larger share of those with low religious commitment (82%) say this is +appropriate. + +Those with high levels of religiosity also stand +out when considering the possibility that +development of gene editing would entail +testing on human embryos. An overwhelming +majority of those high in religious +commitment (87%) say this would be taking +medical technology too far; just 11% of this +group says this would be appropriate. In +contrast, 55% of those low in religious +commitment say that development of gene +editing techniques that require testing on +human embryos would be an appropriate use +of medical technology. + +The differences by religiosity in views on the +appropriateness of gene editing tend to persist +even when statistically controlling for other +factors such as gender, race and ethnicity, age, +and education that are related to religious +beliefs and practices. See Appendix for details. + +Most highly religious Americans +consider testing gene editing on human +embryos to be taking technology too far + +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … + +Taking medical technology too far +Appropriate use of medical technology + +U.S. adults + +65 + +33 + +Among those ___ on the religious commitment index + +High + +87 + +Medium + +69 + +11 + +29 + +Low + +44 + +55 + +Note: Respondents who did not give an answer are not shown. See +Methodology for details on index of religious commitment. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 7 + +PEW RESEARCH CENTER + +There are also differences in views about gene +editing by religious affiliation, particularly if +such techniques would involve embryonic +testing. Just 11% of white evangelical +Protestants say that if development of gene +editing techniques required testing on human +embryos, it would be an appropriate use of +medical technology. + +By contrast, about half of the religiously +unaffiliated (52%), including 79% of atheists +and 57% of agnostics, say that embryonic +testing to develop gene editing techniques +would be an appropriate use of medical +technology. + +See the Appendix for other views about gene +editing by religious affiliation. + +White evangelicals especially likely to +say gene editing involving embryonic +testing takes technology too far + +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate use +of medical +technology +33% + +Taking medical +technology too +far +65% + +21 +11 +31 +25 +30 +29 +27 +52 +79 +57 +42 + +34 +29 +28 + +78 +88 +67 +72 +69 +70 +73 +47 +21 +42 +57 + +64 +68 +71 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 8 + +PEW RESEARCH CENTER + +Men are more accepting than women of using gene editing to change a baby’s genetic +makeup + +Men are more inclined than +women to view gene editing as +an appropriate use of medical +technology, regardless of its +intent. About two-thirds of +men (65%) believe that gene +editing to reduce a baby’s risk +of developing a serious +disease later in life is an +appropriate use of medical +technology, compared with +54% of women. And men +(76%) are more supportive +than women (68%) of using +gene editing to treat a +congenital disorder. + +Similarly, more men (43%) +than women (24%) are +accepting of gene editing +technology if it required +embryonic testing to develop. + +Men are more likely than women to view gene editing +for babies as appropriate + +% of U.S. adults who say changing a baby’s genetic characteristics for each +of the following reasons is an appropriate use of medical technology + +Changing a baby's genetic characteristics ... + +To treat a serious disease/condition the +baby would have at birth + +Women + +Men + +68 + +76 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +54 + +65 + +To make the baby more intelligent + +13 + +24 + +Gene editing to change a baby's +genetic characteristics ... + +If it required testing on human embryos +to develop these techniques + +24 + +43 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Women tend to have higher +levels of religious commitment, on average, but differences by gender tend to hold even when +controlling for religiosity and other factors in statistical models. See the Appendix for details. + +Previous Pew Research Center surveys in 2016 and 2014, while not directly comparable due to +different polling methods and question wording, have also found that women tend to be less +accepting than men when it comes to gene editing for babies. + +www.pewresearch.org + + 9 + +PEW RESEARCH CENTER + +Americans with high levels of science knowledge are more likely to view gene editing in a +positive light + +People high in science +knowledge – based on a nine- +item index – tend to be more +accepting of using gene +editing for babies.4 Some +86% of those with high +science knowledge believe it +is appropriate to use gene +editing to treat a congenital +disorder, compared with 58% +of those with low science +knowledge. And, 71% of those +with high science knowledge +say it is appropriate to use +gene editing to reduce a +baby’s risk of disease that +would occur later in life, +compared with 49% of those +with low science knowledge.5 + +However, few Americans say +it is appropriate to use gene +editing to make a baby more +intelligent, regardless of their +level of science knowledge. + +Those with high science knowledge are more inclined +to see gene editing techniques as appropriate + +% of U.S. adults who say each of the following is an appropriate use of +medical technology + +Changing a baby's genetic +characteristics ... +To treat a serious disease/condition the +baby would have at birth + +Among those with ___ science +knowledge +Medium + +Low + +High + +58 + +86 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +49 + +71 + +To make the baby more intelligent + +Gene editing to change a baby's +genetic characteristics ... + +21 + +24 + +If it required testing on human +embryos to develop these techniques + +26 + +50 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +See Methodology for details on index of science knowledge. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Judgments on the appropriateness of gene editing also follow a similar pattern by education level, +which is closely linked with science knowledge. Those with a postgraduate degree are more + +4 See the Methodology for details on the science knowledge index. +5 These findings are in keeping with a 2017 study by Scheufele et al., which found Americans’ attitudes about gene editing varied depending +on their knowledge level about genetics and gene editing, as well as their religiosity. A 2013 study by Allum et al. in the United Kingdom found +the effect of science knowledge on support for using genetic testing to screen for serious diseases in an unborn baby was conditional on +religious service attendance. In this survey, there is a tendency for the effects of science knowledge to be conditional with levels of religiosity +in views about the use of gene editing to reduce the risk of disease and if it would require testing on human embryos. For example, among +people with high religious commitment, science knowledge is not related to views on the appropriateness of gene editing for babies to reduce +the risk of disease that could occur over their lifetime. This interaction between religious commitment and science knowledge holds even after +controlling for other factors in statistical models but does not occur for other views about gene editing. + +www.pewresearch.org + + 10 + +PEW RESEARCH CENTER + +accepting of gene editing for treating or reducing the risk of a disease than those with a high school +degree or less. + +Americans who report familiarity with gene editing are more inclined to see it as +appropriate + +People who say they have +heard at least a little about +gene editing that can be used +to change a baby’s genetic +characteristics are more +likely than those who have +heard nothing at all to say +gene editing would be +appropriate in each of the +circumstances considered in +this survey. + +A Pew Research Center +survey in 2016 found a +similar tendency for those +who report familiarity with +gene editing to express more +interest in and acceptance of +using these technologies to +enhance a baby’s health over +the course of their lifetime. + +Those familiar with gene editing are more inclined to +see different uses as appropriate + +% of U.S. adults who say each of the following is an appropriate use of +medical technology + +Changing a baby's genetic +characteristics ... + +To treat a serious disease/condition the +baby would have at birth + +Among those who have heard ___ +about gene editing +A little +71 + +Nothing at all + +A lot + +64 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +52 + +65 + +To make the baby more intelligent + +13 + +31 + +Gene editing to change a baby's +genetic characteristics ... + +If it required testing on human embryos +to develop these techniques + +24 + +49 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 11 + +PEW RESEARCH CENTER + +Although gene editing for babies is still in development, a separate April 2018 survey found that +about half of Americans (52%) believe it is likely that within the next 50 years, we will be able to +eliminate almost all birth defects by manipulating the genes of an embryo before a baby is born. In +the current survey, as people think about a future with possible widespread use of gene editing to +change a baby’s genetic makeup, more anticipate negative than positive effects on society. + +Specifically, a majority of +Americans (58%) believe +gene editing will very likely +lead to increased inequality +because it will only be +available to the wealthy. +Some 54% of Americans +anticipate a slippery slope, +saying it’s very likely that +“even if gene editing is used +appropriately in some cases, +others will use these +techniques in ways that are +morally unacceptable.” And, +46% expect it is very likely +that gene editing techniques +will be used before we fully +understand how they affect +people’s health. + +Larger shares of Americans believe negative effects of +widespread use of gene editing are very likely + +% of U.S. adults who say each of the following would be ____ to occur if gene +editing to change a baby’s genetic characteristics becomes widely available + +Very likely + +Fairly likely + +Not too likely + +Not at all likely + +Inequality will increase as it will +only be available for the wealthy + +Even if used appropriately by +some, others will use in morally +unacceptable ways + +Will be used before we fully +understand effects on health + +58 + +29 + +9 + +4 + +54 + +32 + +9 + +3 + +46 + +38 + +12 + +3 + +Will pave way for new medical +advances that benefit society + +18 + +42 + +30 + +8 + +Will help people live longer and +better quality lives + +16 + +48 + +27 + +8 + +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Far smaller shares see +positive outcomes as very +likely if gene editing for +babies were to become widely available. Roughly two-in-ten Americans (18%) say it is very likely +that development of these techniques will pave the way for new medical advances that benefit +society as a whole. Another 42% say this is fairly likely to occur, while 38% consider this not too or +not at all likely. Just 16% see the widespread use of gene editing as very likely to help people live +longer and better quality lives. About half (48%) say this is fairly likely, while 34% say this +outcome is not too or not at all likely. + +www.pewresearch.org + + 12 + +PEW RESEARCH CENTER + +Women are more inclined +than men to expect negative +effects if gene editing +becomes widely available, in +keeping with gender +differences over acceptance +of gene editing. For example, +larger shares of women than +men say this technology will +lead to an increase in +inequality as it will only be +available for the wealthy +(63% say this is very likely vs. +52% of men) or say it will be +used in morally unacceptable +ways (57% vs. 51%). And +although only minorities of +men and women see positive +effects as very likely, men are +somewhat more inclined to +anticipate positive outcomes. + +More women than men anticipate negative effects of +gene editing + +% of U.S. adults who say each of the following would be very likely to occur if +gene editing to change a baby’s genetic characteristics becomes widely +available + +POSITIVE OUTCOMES + +Women + +Men + +Will pave way for new medical +advances that benefit society + +14 + +23 + +Will help people live longer and +better quality lives + +13 + +19 + +NEGATIVE OUTCOMES + +Inequality will increase as it will +only be available for the wealthy + +Even if used appropriately by +some, others will use in morally +unacceptable ways + +52 + +63 + +51 + +57 + +Will be used before we fully +understand effects on health + +42 + +49 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Expectations also vary by +self-reported familiarity with +gene editing. Those who have +heard a lot about gene editing to change a baby’s genetic characteristics are more inclined to +anticipate positive outcomes from its widespread use. Some 36% of this group thinks it is very +likely that the widespread availability of gene editing would pave the way for new medical +advances that are beneficial to society, compared with about half as many of those who have heard +a little or nothing about gene editing (16%). + +However, those most familiar with gene editing are more likely than those with no familiarity to +also anticipate downsides. About two-thirds of those who have heard a lot about gene editing +(64%) say it is very likely that widespread availability of gene editing will increase inequality +because the technology will primarily be available only for the wealthy, compared with 53% of +those who have heard nothing about using gene editing to change a baby’s genetic characteristics. +And, 65% of those who have heard a lot about gene editing think it’s very likely that others will use + +www.pewresearch.org + + 13 + +PEW RESEARCH CENTER + +these techniques in morally unacceptable ways, while 54% of those who have heard a little and +52% of those who have heard nothing about gene editing say this. + +Americans who are more familiar with gene editing are more inclined to anticipate +positive as well as negative effects from widespread use of gene editing + +% of U.S. adults in each group who say each of the following would be very likely to occur if gene editing to change a +baby’s genetic characteristics becomes widely available + +POSITIVE OUTCOMES + +Will pave way for new medical +advances that benefit society + +Will help people live longer and +better quality lives + +NEGATIVE OUTCOMES + +Inequality will increase as it will +only be available for the wealthy +Even if used appropriately by +some, others will use in morally +unacceptable ways +Will be used before we fully +understand effects on health + +Heard a lot about +gene editing + +Heard a little about +gene editing + +Heard nothing at all +about gene editing + +36 + +34 + +16 + +13 + +16 + +14 + +64 + +65 + +59 + +54 + +53 + +52 + +54 + +46 + +43 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +A similar pattern occurs with levels of science knowledge. A larger share of people with high +science knowledge anticipate positive outcomes from the widespread availability of gene editing +than those with low science knowledge. At the same time, those with high science knowledge are at +least equally likely as those with low science knowledge to think negative effects from the +widespread availability of gene editing are very likely. + +www.pewresearch.org + + 14 + +PEW RESEARCH CENTER + +The American public also +tends to be skeptical about +whether medical experts fully +comprehend the health +consequences of gene editing. +A 36% minority of Americans +believe that medical +researchers understand the +health effects of gene editing +for babies either very (7%) or +fairly well (29%), while 62% +say medical researchers do +not understand the health +effects at all or not too well. + +36% of Americans say medical researchers understand +the health effects of gene editing at least fairly well + +% of U.S. adults who say medical researchers understand the health risks and +benefits of changing a baby’s genetic characteristics … + +Very well + +Fairly well + +Not too well + +Not at all well + +7 + +29 + +45 + +17 + +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Familiarity with gene editing is linked with beliefs about medical researchers’ understanding of it. +About half of those who have heard a lot about gene editing for babies (51%) say that medical +researchers understand the health effects of gene editing for babies at least fairly well, compared +with three-in-ten of those who have heard nothing at all (30%). However, level of science +knowledge based on a nine-item index is not related to views on this issue; 36% of those with high +science knowledge and 40% of those with low science knowledge believe that medical researchers +understand the health effects of changing a baby’s genetic characteristics at least fairly well. + +There are no or only modest differences on this question by gender or religious commitment. + +www.pewresearch.org + + 15 + +PEW RESEARCH CENTER + +Acknowledgments + +This report is made possible by The Pew Charitable Trusts. This report is a collaborative effort +based on the input and analysis of the following individuals. Find related reports online at: +pewresearch.org/science. + +Primary research team + +Cary Funk, Director, Science and Society Research +Brian Kennedy, Research Associate +Meg Hefferon, Research Assistant +Mark Strauss, Writer/Editor +Isadora Milanez, Intern + +Editorial and graphic design + +Mark Strauss, Writer/Editor +Shannon Greenwood, Copy Editor +Margaret Porteus, Information Graphics Designer + +Communications and web publishing + +Haley Nolan, Communications Assistant +Sara Atske, Assistant Digital Producer + +www.pewresearch.org + + 16 + +PEW RESEARCH CENTER + +Methodology + +Margins of error + +This report is drawn from a survey conducted as +part of the American Trends Panel (ATP), a +nationally representative panel of randomly +selected U.S. adults living in households +recruited from landline and cellphone random- +digit-dial (RDD) surveys. Panelists participate +via monthly self-administered web surveys. +Panelists who do not have internet access are +provided a tablet and wireless internet +connection. The panel, which was created by +Pew Research Center, is being managed by GfK. + +Data in this report are drawn from the panel +wave conducted April 23-May 6, 2018, among +2,537 respondents. The margin of sampling error +for the full sample of 2,537 respondents is plus +or minus 2.8 percentage points. + +Members of the ATP were recruited from several +large, national landline and cellphone RDD +surveys conducted in English and Spanish. At +the end of each survey, respondents were invited +to join the panel. The first group of panelists was +recruited from the 2014 Political Polarization +and Typology Survey, conducted Jan. 23-March +16, 2014. Of the 10,013 adults interviewed, 9,809 +were invited to take part in the panel and a total +of 5,338 agreed to participate.6 + +Sample size +2,537 + +Margin of error +in percentage +points ++/- 2.8 + +U.S. adults + +Men +Women + +Race/ethnicity +White +Black +Hispanic + +Religious commitment index +High +Medium +Low + +Religious affiliation +NET Protestant + +White evangelical +White mainline +Black Protestant + +NET Catholic + +White Catholic +Hispanic Catholic + +NET Unaffiliated +Atheist +Agnostic +Nothing in particular + +1,272 +1,265 + +1,519 +412 +421 + +440 +1,291 +566 + +1,086 +349 +282 +283 +459 +254 +168 +703 +146 +158 +399 + ++/- 4.0 ++/- 3.8 + ++/- 3.3 ++/- 7.8 ++/-7.6 + ++/- 6.7 ++/- 3.8 ++/- 5.7 + ++/- 4.2 ++/- 7.1 ++/- 7.7 ++/- 9.6 ++/- 6.5 ++/- 8.0 ++/-11.8 ++/-5.1 ++/- 11.2 ++/- 10.2 ++/- 6.8 + +Note: The margins of error are reported at the 95% level of +confidence and are calculated by taking into account the average +design effect for each subgroup. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +6 When data collection for the 2014 Political Polarization and Typology Survey began, non-internet users were subsampled at a rate of 25%, +but a decision was made shortly thereafter to invite all non-internet users to join. In total, 83% of non-internet users were invited to join the +panel. + +www.pewresearch.org + + 17 + +PEW RESEARCH CENTER + +The second group of panelists was recruited +from the 2015 Pew Research Center Survey on +Government conducted Aug. 27-Oct. 4, 2015. Of +the 6,004 adults interviewed, all were invited to +join the panel, and 2,976 agreed to participate.7 +The third group of panelists was recruited from a +survey conducted April 25-June 4, 2017. Of the +5,012 adults interviewed in the survey or pretest, +3,905 were invited to take part in the panel and a +total of 1,628 agreed to participate.8 + +The overall target population for Wave 34 was +non-institutionalized persons ages 18 and older, +living in the United States, including Alaska and +Hawaii. The sample for Wave 34 consisted of +3,099 ATP members that were invited to Wave +33 and were still active. This subsample was +selected using the following approach: + +Margins of error continued + +Sample size + +Margin of error +in percentage +points + +Science knowledge index +High +Medium +Low + +Heard about gene editing +A lot +A little +Nothing at all + +679 +1,274 +584 + +326 +1,449 +753 + ++/-5.2 ++/- 3.9 ++/- 5.7 + ++/- 7.9 ++/-3.6 ++/-5.1 + +Note: The margins of error are reported at the 95% level of +confidence and are calculated by taking into account the average +design effect for each subgroup. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +1. Panelists were grouped into three strata based on how underrepresented they are + +demographically. Then we analyzed response rates to the last five panel survey waves +(W28-32) to project the number of panelists in each stratum who would respond to the +W33 survey. + +2. We then determined how many panelists we wanted to sample from each stratum in W33 +in order to finish with around 2,500 completed interviews and have a responding sample +that is as representative as possible. + +▪ Stratum A consists of panelists who are non-internet users, are black non-Hispanic, +are Hispanic, or have high school or less education. There were 1,819 total panelists +in this stratum and they are sampled at a rate of 100% for W33. 1,806 were active +panelists. + +7 Respondents to the 2014 Political Polarization and Typology Survey who indicated that they are internet users but refused to provide an +email address were initially permitted to participate in the American Trends Panel by mail but were no longer permitted to join the panel after +Feb. 6, 2014. Internet users from the 2015 Pew Research Center Survey on Government who refused to provide an email address were not +permitted to join the panel. +8 White, non-Hispanic college graduates were subsampled at a rate of 50%. + +www.pewresearch.org + + 18 + +PEW RESEARCH CENTER + +▪ Stratum B consists of panelists who are ages 18 to 34 or are non-volunteers. The + +1,684 total panelists in this stratum are subsampled at a rate of 63%, yielding 1,061 +sampled for W33 (1,057 were active). + +▪ Stratum C consists of the remaining 2,009 panelists not in stratum A or B. This + +group is subsampled at a rate of 12%, yielding 241 panelists sampled for W33 (239 +were active). + +The ATP data were weighted in a multistep process that begins with a base weight incorporating +the respondents’ original survey selection probability and the fact that in 2014 some panelists were +subsampled for invitation to the panel. Next, an adjustment was made for the fact that the +propensity to join the panel and remain an active panelist varied across different groups in the +sample. The final step in the weighting uses an iterative technique that aligns the sample to +population benchmarks on a number of dimensions. Gender, age, education, race, Hispanic origin +and region parameters come from the U.S. Census Bureau’s 2016 American Community Survey. +The county-level population density parameter (deciles) comes from the 2010 U.S. decennial +census. The telephone service benchmark comes from the July-December 2016 National Health +Interview Survey and is projected to 2017. The volunteerism benchmark comes from the 2015 +Current Population Survey Volunteer Supplement. The party affiliation benchmark is the average +of the three most recent Pew Research Center general public telephone surveys. The internet +access benchmark comes from the 2017 ATP Panel Refresh Survey. Respondents who did not +previously have internet access are treated as not having internet access for weighting purposes. +Sampling errors and statistical tests of significance take into account the effect of weighting. +Interviews are conducted in both English and Spanish, but the Hispanic sample in the ATP is +predominantly U.S. born and English speaking. + +Margins of error tables shown here provide the unweighted sample sizes and the error attributable +to sampling that would be expected at the 95% level of confidence for different groups in the +survey taking into account the average design effect for each subgroup. Sample sizes and sampling +errors for other subgroups are available upon request. + +In addition to sampling error, one should bear in mind that question wording and practical +difficulties in conducting surveys can introduce error or bias into the findings of opinion polls. + +The April 2018 wave had a response rate of 82% (2,537 responses among 3,099 individuals in the +panel). Taking account of the combined, weighted response rate for the recruitment surveys + +www.pewresearch.org + + 19 + +PEW RESEARCH CENTER + +(10.1%) and attrition from panel members who were removed at their request or for inactivity, the +cumulative response rate for the wave is 2.3%.9 + +Religious commitment index + +Survey respondents were classified into high, medium and low levels of religious commitment +based on three indicators: frequency of religious service attendance, self-reported importance of +religion in their lives and frequency of prayer. Those who attend worship services at least weekly, +pray at least once a day and say religion is very important in their lives are classified as high in +religious commitment. Those low in commitment say religion is not too or not at all important in +their lives, that they seldom or never attend worship services, and seldom or never pray. All others +are classified as exhibiting a medium level of religious commitment. + +9 Approximately once per year, panelists who have not participated in multiple consecutive waves are removed from the panel. These cases +are counted in the denominator of cumulative response rates. Note that for the March 2018 survey, we calculated the response rates by +computing the mean rates for the subsampled respondents (based on the rates from the recruitment survey they joined the panel on). + +www.pewresearch.org + + 20 + +PEW RESEARCH CENTER + +Science knowledge index + +Scale reliability and factor analysis + +The Pew Research Center survey included a set of nine questions to tap public knowledge of +science across a range of principles and topics. Most respondents (n=1,901) completed these +questions as part of a previous +wave of the American Trends +Panel, conducted May 10-June +6, 2016. Respondents who did +not participate in the previous +wave answered these questions +in this survey (n=636). The +following shows the +measurement properties of the +index for the combined set of +2,537 respondents.10 + +Alpha for scale +0.75 +Alpha if item is +dropped + +Knowledge index 9-items + +Item-rest +correlation + +0.47 + +0.39 + +0.74 + +0.73 + +Common +variance +explained by +first factor +76% +Factor +loadings + +0.46 + +0.55 + +KNOSCT22. Use of a control group +to determine whether a new drug is +effective +KNOSCT23. Carbon dioxide is made +as a consequence of burning fossil +fuels +KNOSCT27. The probability of an +old-bridge collapsing after a period +of time +KNOSCT28. Only bacterial +infections can be treated effectively +by antibiotic medications +KNOSCT29. The use of a control +“sugar pill” in a new drug trial is to +rule out a possible placebo effect +KNOSCT31. The health benefits +occurring when most people in a +population get a vaccine is called +herd immunity +KNOSCT32. An apple, salmon, corn +and a mosquito can all be +genetically modified. +KNOSCT33. Humans and mice +share 50% or more of the same +genetic makeup +KNOSCT34. Nitrogen makes up +most of the Earth’s atmosphere. + +0.50 + +0.72 + +0.59 + +0.31 + +0.75 + +0.35 + +0.43 + +0.73 + +0.50 + +0.45 + +0.73 + +0.52 + +0.47 + +0.73 + +0.55 + +An exploratory factor analysis +finds one common factor +explaining 76% of the common +variance in the items. The +factor loadings show that each +of the nine questions is +moderately correlated with the +common factor. These +indicators suggest a set of items is measuring a single underlying dimension. + +PEW RESEARCH CENTER + +0.44 + +0.43 + +Source: Surveys conducted May 10-June 6, 2016, and April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +0.73 + +0.50 + +0.73 + +0.52 + +As shown in the accompanying +table, the internal reliability or +consistency of the scale as +measured by Cronbach’s alpha +is 0.75. Each of the items in the +scale is at least moderately +correlated with the other items. + +10 In statistical models controlling for demographics and education, there were no significant differences between respondents who +completed the science knowledge items in 2016 (W17) and those who completed them in 2018 (W34) on either the overall index measure or +the individual questions. + +www.pewresearch.org + + 21 + +PEW RESEARCH CENTER + +Two parameter item response theory analysis + +KNOSCT22. Use of a control +group to determine whether a +new drug is effective +KNOSCT23. Carbon dioxide is +made as a consequence of +burning fossil fuels +KNOSCT27. The probability of an +old-bridge collapsing after a +period of time +KNOSCT28. Only bacterial +infections can be treated +effectively by antibiotic +medications +KNOSCT29. The use of a control +“sugar pill” in a new drug trial is +to rule out a possible placebo +effect +KNOSCT31. The health benefits +occurring when most people in a +population get a vaccine is called +herd immunity +KNOSCT32. An apple, salmon, +corn and a mosquito can all be +genetically modified. +KNOSCT33. Humans and mice +share 50% or more of the same +genetic make-up +KNOSCT34. Nitrogen makes up +most of the Earth’s atmosphere. + +% Correct Discrimination + +Difficulty + +68 + +72 + +62 + +46 + +1.36 + +-0.73 + +1.74 + +-0.82 + +2.14 + +-0.38 + +0.81 + +0.25 + +59 + +1.60 + +-0.33 + +34 + +39 + +34 + +1.39 + +0.65 + +1.66 + +0.38 + +1.37 + +0.66 + +Note that each of the science +knowledge questions are coded +as binary variables +(correct/incorrect). Both +Cronbach’s alpha reliability +analysis and the factor analysis +are based on a Pearson’s +correlation matrix. Pearson +correlations with binary +variables are restricted to a +limited range, underestimating +the association between two +variables when compared with +tetrachoric correlations. We do +not anticipate that the use of a +Pearson’s correlation matrix +affects the unidimensional +factor solution for the scale, +however. + +31 + +1.55 + +0.74 + +PEW RESEARCH CENTER + +Source: Surveys conducted May 10-June 6, 2016, and April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +We also ran an item-response +theory analysis (IRT) to check +how well each question +distinguishes between those +who know relatively more or +less on the scale. This analysis fits a two-parameter logistic model, allowing discrimination and +difficulty to vary across the items. Discrimination shows the ability of the question to distinguish +between those with higher and lower science knowledge. Difficulty shows how easy or hard each +question is for the average respondent. We did not include a guessing parameter in the model; the +questionnaire offered respondents an explicit option of not sure on the survey. + +The results show variation in difficulty across the items. The easiest item required respondents to +identify that carbon dioxide as the gas that is made as a consequence of burning fossil fuels, while +the most difficult item required respondents to identify nitrogen as the gas that makes up most of +the Earth’s atmosphere. + +www.pewresearch.org + + 22 + +PEW RESEARCH CENTER + +Most of the questions also discriminate between those with higher and lower science knowledge. +The item with strongest ability to discriminate was the question asking respondents to calculate +the conditional probability of an old bridge over time. The question with the weakest ability to +discriminate was the question about the effectiveness of antibiotics to treat bacterial, but not other +kinds of infections. + +The test information curve mirrors a normal curve centered around zero, suggesting that the +science knowledge index provides the most information about Americans near the mean level of +knowledge. + +© Pew Research Center, 2018 + +www.pewresearch.org + + 23 + +PEW RESEARCH CENTER + +Appendix: Detailed tables + +Half or more from all religious groups +consider gene editing to treat disease +that would be present at birth to be +appropriate + +% of U.S. adults who say using gene editing to treat a +serious disease or condition a baby would have at birth +is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +72% + +Taking medical +technology too +far +27% + +66 +61 +79 +60 +72 +72 +74 +78 +92 +92 +69 + +74 +62 +74 + +31 +37 +20 +36 +28 +28 +26 +22 +7 +8 +30 + +24 +35 +26 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 24 + +PEW RESEARCH CENTER + +White evangelical Protestants among +the least likely to consider gene editing +to reduce lifetime disease risk +appropriate + +% of U.S. adults who say using gene editing to reduce the +risk of a serious disease or condition that could occur +over the course of a baby’s lifetime is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +60% + +Taking medical +technology too +far +38% + +51 +45 +58 +51 +63 +58 +68 +67 +85 +81 +57 + +60 +52 +61 + +47 +54 +41 +43 +36 +41 +30 +32 +15 +19 +41 + +39 +44 +36 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 25 + +PEW RESEARCH CENTER + +Nine-in-ten white evangelical +Protestants see gene editing for +intelligence as taking technology too far + +% of U.S. adults who say using gene editing to make a +baby more intelligent is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +19% + +Taking medical +technology too +far +80% + +12 +8 +12 +14 +20 +18 +24 +25 +36 +22 +23 + +16 +21 +22 + +86 +91 +88 +81 +80 +82 +76 +74 +64 +77 +77 + +83 +75 +76 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 26 + +PEW RESEARCH CENTER + +Statistical models predicting opinions about appropriate uses of gene editing + +Difference in predicted probabilities for belief that changing a baby’s genetic characteristics ____ is an appropriate use of +medical technology + +Women + +To treat a serious disease or +condition the baby would +have at birth +NS + +To reduce the risk of a serious +disease or condition that could occur +over the course of his or her lifetime +-0.08 + +To make the baby +more intelligent +-0.09 + +Requiring testing on +human embryos +-0.15 + +Race Reference category: White, non-Hispanic + +Black, non-Hispanic + +Hispanic + +Other + +Age Reference category: 18-29 + +Ages 30-49 + +50-64 + +65+ + +NS + +NS + +-0.20 + +NS + +-0.10 + +NS + +Education Reference category: High school or less + +Postgraduate + +College graduate + +Some college + ++0.16 + ++0.14 + ++0.09 + +Science knowledge Reference category: Low science knowledge + +High + +Medium + ++0.16 + +NS + +NS + +NS + +NS + +-0.12 + +-0.17 + +NS + ++0.13 + +NS + ++0.08 + +NS + +NS + +Religious commitment Reference category: Low religious commitment + +High + +Medium + +-0.14 + +NS + +Religious affiliation Reference category: Unaffiliated + +Mainline Protestant + +Evangelical Protestant + +Catholic + +Other Christian + +Other religion + +NS + +NS + +NS + +NS + +NS + +-0.21 + +-0.10 + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +-0.08 + +-0.12 + +-0.13 + +NS + +NS + +NS + +NS + +-0.09 + +-0.14 + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +-0.09 + +NS + +NS + +NS + +-0.12 + ++0.10 + +NS + ++0.07 + +NS + +-0.08 + +-0.14 + +-0.13 + +NS + +-0.22 + +NS + +NS + +NS + +Heard about gene editing Reference category: Heard nothing at all about gene editing to change a baby’s genetic characteristics + +Heard a lot + +Heard a little + +Number of respondents + +NS + ++0.06 + +2,237 + +NS + +NS + +2,232 + ++0.15 + ++0.07 + +2,241 + ++0.17 + ++0.09 + +2,223 + +Note: Figures shown are the difference in predicted probabilities with the reference category while other factors are held at their mean using binary +logistic regression models. Positive and negative values indicated the direction of effects. NS indicates not statistically significant based on a two- +tailed p value < 0.05. +Source: Survey conducted April 23- May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 27 + +PEW RESEARCH CENTER + +Survey questionnaire and topline + +2018 PEW RESEARCH CENTER’S AMERICAN TRENDS PANEL +APRIL 23-MAY 6, 2018 +TOTAL N=2,537 + +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED OR HELD FOR FUTURE RELEASE + +ASK ALL: +MED3 How much, if anything, have you heard or read about gene editing that can be used to change a + +baby’s genetic characteristics? + +April 23- +May 6, +2018 +12 +56 +32 +1 + +A lot +A little +Nothing at all +No answer + +ASK ALL: +MED4 Do you think each of the following is an appropriate use of medical technology or is it taking + +medical technology too far? [RANDOMIZE ITEMS] + +An appropriate use of +medical technology + +Taking medical +technology too far + +No +Answer + +a. Changing a baby's genetic characteristics + +to make the baby more intelligent + +April 23-May 6, 2018 + +b. Changing a baby's genetic characteristics + +to TREAT a serious disease or condition the +baby would have at birth +April 23-May 6, 2018 + +c. Changing a baby's genetic characteristics +to REDUCE THE RISK of a serious disease +or condition that could occur over the +course of his or her lifetime + +19 + +72 + +80 + +27 + +April 23-May 6, 2018 + +60 + +38 + +1 + +2 + +2 + +TRENDS FOR COMPARISON +Pew Research Center survey conducted by telephone: Would you say that changing a baby’s genetic +characteristics to make the baby more intelligent is making appropriate use of medical advances OR is it +taking medical advances too far? + +Aug. 15-25, +2014 +15 +83 +2 + +Appropriate use of medical advances +Taking medical advances too far +Don’t know/Refused (VOL.) + +www.pewresearch.org + + 28 + +PEW RESEARCH CENTER + +Surveys conducted by telephone: Would you say that changing a baby’s genetic characteristics to reduce +the risk of serious diseases is making appropriate use of medical advances OR is it taking medical +advances too far? + +Appropriate use of medical advances +Taking medical advances too far +Don’t know/Refused (VOL.) + +Pew Research Center +Aug. 15-25, +2014 +46 +50 +4 + +VCU Life Sciences +Survey +Sept. 3-26, +2003 +41 +54 +6 + +ASK ALL: +MED5 Thinking about what you have heard or read, how well do you think medical researchers + +understand the health risks and benefits of changing a baby’s genetic characteristics? + +April 23- +May 6, +2018 +7 +29 +45 +17 +1 + +Very well +Fairly well +Not too well +Not at all well +No answer + +ASK ALL: +MED6 Thinking about society as a whole … + +How likely, if at all, do you think each of the following would be to occur if gene editing to +change a baby’s genetic characteristics becomes widely available? [RANDOMIZE ITEMS] + +Very likely + +Fairly +likely + +Not too +likely + +Not at all +likely + +No +Answer + +a. Even if gene editing is used + +appropriately in some cases, others +will use these techniques in ways +that are morally unacceptable + +April 23-May 6, 2018 + +54 + +32 + +9 + +b. These techniques will help people +live longer and better quality lives + +April 23-May 6, 2018 + +16 + +48 + +27 + +c. We will use these techniques before +we fully understand how they affect +people’s health + +April 23-May 6, 2018 + +46 + +38 + +12 + +d. + +Inequality will increase because +this option will be available only for +the wealthy + +April 23-May 6, 2018 + +58 + +29 + +9 + +3 + +8 + +3 + +4 + +1 + +2 + +1 + +1 + +www.pewresearch.org + + 29 + +PEW RESEARCH CENTER + +Very likely + +Fairly +likely + +Not too +likely + +Not at all +likely + +No +Answer + +MED6 continued … + +e. Development of these techniques +will pave the way for new medical +advances that benefit society as a +whole + +April 23-May 6, 2018 + +18 + +42 + +30 + +8 + +1 + +ASK ALL: +MED7 If gene editing to change a baby’s genetic characteristics required testing on human embryos in + +order to develop these techniques, do you think this would be … + +April 23- +May 6, +2018 +33 +65 +2 + +An appropriate use of medical technology +Taking medical technology too far +No answer + +ADDITIONAL QUESTIONS HELD FOR FURTHER RELEASE + +ASK ALL: +BIO15 Have you seen a health care provider FOR AN ILLNESS OR MEDICAL CONDITION in the past 12 + +months, or not? + +April 23- +May 6, +2018 +64 +36 +1 + +Yes +No +No answer + +May 10- +June 6 +2016 +63 +36 +2 + +ASK ALL: +G1 + +Do you, or does anyone in your immediate family, have a gene that predisposes you to a serious +disease such as Alzheimer’s, cancer, heart disease or sickle cell anemia? + +April 23- +May 6, +2018 +26 +40 +34 +1 + +Yes +No +Not sure +No answer + +May 10- +June 6, +2016 +27 +43 +30 +1 + +www.pewresearch.org + + 30 + +PEW RESEARCH CENTER + +ASK ALL: +G2 + +Have you, or has anyone in your immediate family, ever had a genetic test, or haven’t you done +this? + +April 23- +May 6, +2018 +14 +65 +20 +1 + +Yes +No +Not sure +No answer + +May 10- +June 6 +2016 +10 +70 +19 +1 + +NOTE: RESULTS SHOWN FOR KNOSCT22-KNOSCT34 INCLUDE 636 RESPONDENTS WHO +ANSWERED THESE QUESTIONS APRIL 23-MAY 6, 2018 COMBINED WITH 1,901 RESPONDENTS +WHO ANSWERED THESE QUESTIONS IN A PREVIOUS WAVE FIELDED MAY 10-JUNE 6, 2016. + +ASK ALL: Here’s a different kind of question. (If you don’t know the answer, select “Not sure.”) As far as +you know … + +KNOSCT22 + +Here’s a different kind of question. (If you don’t know the answer, select “Not sure.”) As +far as you know ... + +Which is the better way to determine whether a new drug is effective in treating a +disease? If a scientist has a group of 1,000 volunteers with the disease to study, should +she … [RANDOMIZE OPTIONS 1 AND 2] + +Apr 23- +May 6 +2018 +68 + +32 + 14 + 18 +<1 + +Give the drug to half of them but not to the other half, and compare +how many in each group get better (Correct) +NET Incorrect/Not sure/No answer + Give the drug to all of them and see how many get better + Not sure + No answer + +[RANDOMIZE ITEMS KNOSCT23 TO KNOSCT34; KNOSCT22 ALWAYS FIRST] + +www.pewresearch.org + + 31 + +PEW RESEARCH CENTER + +Which gas is made as a consequence of burning fossil fuels? Is it … [RANDOMIZE +OPTIONS 1-4]11 + +Carbon dioxide (Correct) +NET Incorrect/Not sure/No answer + Hydrogen + Helium + Radon + Not sure + No answer + +ASK ALL: +KNOSCT23 + +Apr 23- +May 6 +2018 +72 +28 + 4 + 1 + 3 + 20 +<1 + +NO QUESTION KNOSCT24, KNOSCT25 AND KNOSCT26 + +ASK ALL: +KNOSCT27 + +If the chances that an old bridge will collapse starts at 1% in week 1 and doubles each +week (as shown below), what is the chance that the old bridge will collapse during week +7? + +Chances the bridge will collapse is… + +1% +2% +4% +8% + +at Week 1 +at Week 2 +at Week 3 +at Week 4 + +Enter the % chance that the bridge will collapse at Week 7 (if the bridge is still standing +after Week 6)12 + +April 23- +May 6, +2018 +62 +38 + 20 + 17 + 1 + +64% (Correct) +NET Incorrect/Not sure/No answer + All other numeric responses + Not sure + No answer + +11 In the Wave 17 May 2016 survey, the question had a minor wording difference: “What gas is made as a consequence of burning fossil +fuels? Is it …” +12 The phrase “(if the bridge is still standing after Week 6)” was added for clarity in the April 23-May 6, 2018, wave. + +www.pewresearch.org + + 32 + +PEW RESEARCH CENTER + +Which of the following conditions can be treated effectively by antibiotic medications? + +[Check all that apply] [RANDOMIZE ITEMS WITH ITEMS e AND f ALWAYS LAST] + +Bacterial infection only (Correct) +NET Incorrect/Not sure/No answer + +ASK ALL: +KNOSCT28 + +April 23- +May 6, +2018 +46 +54 + +KNOSCT28 INDIVIDUAL ITEM RESPONSES + +a. Viral infections (such as a cold) +b. Fungal infections (such as athlete’s foot) +c. Bacterial infections (such as strep throat infections) +d. Allergic reactions to insect bites +e. None of these [EXCLUSIVE PUNCH] +f. Not sure [EXCLUSIVE PUNCH] + +Selected +23 +28 +83 +17 +2 +9 + +Not selected +/No answer +77 +72 +17 +83 +98 +91 + +ASK ALL: +KNOSCT29 + +April 23- +May 6, +2018 +59 +41 + 4 + 16 + 20 + 1 + +If a scientist wants to determine if a new drug is effective at treating high blood pressure +by giving half of a group of 1,000 volunteers a new medication and the other half a +“sugar pill” she wants to rule out … [RANDOMIZE OPTIONS 1-3] + +A placebo effect (Correct) +NET Incorrect/Not sure/No answer + A third person effect + A false consensus effect + Not sure + No answer + +NO QUESTION KNOSCT30 + +ASK ALL: +KNOSCT31 + +April 23- +May 6, +2018 +34 +66 + 8 + 35 + 23 + 1 + +Which of these terms refers to health benefits occurring when most people in a +population get a vaccine? [RANDOMIZE OPTIONS 1-3] + +Herd immunity (Correct) +NET Incorrect/Not sure/No answer + Population control + Vaccination rate + Not sure + No answer + +www.pewresearch.org + + 33 + +PEW RESEARCH CENTER + +ASK ALL: +KNOSCT32 + +Which of the following can be genetically modified? + +[Check all that apply] [RANDOMIZE ITEMS WITH ITEMS e AND f ALWAYS LAST] + +April 23- +May 6, +2018 +39 +61 + +Selected all (Correct) +NET Incorrect/Not sure/No answer + +KNOSCT32 INDIVIDUAL ITEM RESPONSES + +a. An apple +b. Salmon +c. A mosquito +d. Corn +e. None of these [EXCLUSIVE PUNCH] +f. Not sure [EXCLUSIVE PUNCH] + +Selected +61 +53 +46 +70 +2 +20 + +Not selected +/No answer +39 +47 +54 +30 +98 +80 + +ASK ALL: +KNOSCT33 + +April 23- +May 6, +2018 +34 +66 + 8 + 14 + 44 + 1 + +ASK ALL: +KNOSCT34 + +April 23- +May 6, +2018 +31 +69 + 9 + 9 + 35 + 15 + 1 + +Humans and mice share the same genetic makeup by… [RANDOMIZE ORDER LOW TO +HIGH; HIGH TO LOW with NOT SURE ALWAYS LAST] + +About 50% or more (Correct) +NET Incorrect/Not sure/No answer + Less than 10% + Between 11% and 49% + Not sure + No answer + +Which gas makes up most of the Earth's atmosphere? [RANDOMIZE OPTIONS 1-4] + +Nitrogen (Correct) +NET Incorrect/Not sure/No answer + Hydrogen + Carbon dioxide + +Oxygen + Not sure + No answer + +www.pewresearch.org + + 34 + +PEW RESEARCH CENTER + +TOTAL NUMBER CORRECT KNOSCT22 THROUGH KNOSCT34: + +April 23- +May 6, +2018 +4 +10 +10 +13 +12 +13 +11 +12 +9 +6 + +9 out of 9 +8 out of 9 +7 out of 9 +6 out of 9 +5 out of 9 +4 out of 9 +3 out of 9 +2 out of 9 +1 out of 9 +0 out of 9 + +24 +49 +26 + +High science knowledge (7-9 correct) +Medium science knowledge (3-6 correct) +Low science knowledge (0-2 correct) + +www.pewresearch.org diff --git a/q128/random_k2/random_2.md b/q128/random_k2/random_2.md index 5d5d4bf748b05f616c8399e8fe70690f3bccb9c7..7e875d55d5f93fef49815a4921bd61abaa902c5b 100644 --- a/q128/random_k2/random_2.md +++ b/q128/random_k2/random_2.md @@ -1,1678 +1,1736 @@ -| | | Learning | | Fine-Grained | | Grounded | Citations | | for | | | -| --- | --- | -------- | ---------- | ------------ | ----- | -------- | --------- | --- | --- | --- | --- | -| | | | Attributed | | Large | Language | Models | | | | | -LeiHuang1,XiaochengFeng1,2*,WeitaoMa1,YuxuanGu1,WeihongZhong1,XiachongFeng1 -WeijiangYu3,WeihuaPeng3,DuyuTang3,DandanTu3,BingQin1,2 -1HarbinInstituteofTechnology,Harbin,China -| | | 2 PengChengLaboratory | | | | 3HuaweiInc.,Shenzhen,China | | | | | | -| --- | --- | --------------------- | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | -{lhuang,xcfeng,wtma,yxgu,whzhong,xiachongfeng,qinb}@ir.hit.edu.cn -{weijiangyu8,pengwh.hit}@gmail.com,{tangduyu,tudandan}@huawei.com -Abstract -Why is it bad to eat cookie dough for risk of salmonella but things -like Cookie Dough Bites are ok? -| Despite | the | impressive | performance | | on | | | | | | | -| ------- | --- | ---------- | ----------- | --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Salmonella is most commonly caused by eating undercooked | | | | | 1 | -information-seekingtasks,largelanguagemod- or raw foods like eggs or meat. (...) You know how your mom -4202 guA 8 ]LC.sc[ 1v86540.8042:viXra els (LLMs) still struggle with hallucinations. always warned you not to eat raw cookie dough? This is why. -| Attributed | LLMs, | which | augment | generated | | | | | | | | -| ---------- | ----- | ----- | ------- | --------- | --- | --------------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Food safety recommendations encourage people to cook eggs until | | | | | 2 | -text within-line citations, haveshownpoten- the white and yolk are firm in order to kill any bacteria. (...) However, -tial in mitigating hallucinations and improv- anyone making cookies can do things to reduce this risk by -using pasteurized egg products. -| ingverifiability. | | However,currentapproaches | | | | | | | | | | -| ----------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -suffer from suboptimal citation quality due Raw cookie dough can carry the risk of salmonella contamination, which can -to their reliance on in-context learning. Fur- lead to foodborne illness if the dough is eaten raw  1  . However, products -like Cookie Dough Bites are generally safe because they are made with -| thermore, | the | practice | of citing | only | coarse | pasteurized eggs  1    2  . | | | | | | -| --------- | --- | -------- | --------- | ---- | ------ | --------------------------- | --- | --- | --- | --- | --- | -documentidentifiersmakesitchallengingfor -Difficult to verify and subject to -userstoperformfine-grainedverification. In hallucinations and citation errors -| this work, | we | introduce | FRONT, | a | training | | | | | | | -| ---------- | --- | --------- | ------ | --- | -------- | --- | --- | --- | --- | --- | --- | - G  R  O   U  N  D  I N  G     1  salmonella is most commonly caused by eating -framework designed to teach LLMs to gen- undercooked or raw foods like eggs or meat  2  Anyone making cookies -can do things to reduce this risk by using pasteurized egg products. -erateFine-gRainedgrOuNdedciTations. By  A  N  S  W   E  R    Eating raw cookie dough can be risky due to salmonella, -groundingmodeloutputsinfine-grainedsup- commonly found in raw eggs  1 . However, this concern is mitigated in -products like Cookie Dough Bites, as the products use pasteurized egg -portingquotes,thesequotesguidethegenera- -products, which are safe to eat raw 2  . -tionofgroundedandconsistentresponses,not -onlyimprovingcitationqualitybutalsofacili- Interpretable and accurately cited -tatingfine-grainedverification.Experimentson -| | | | | | | Figure1: | Comparedwiththecurrentattributedsystems, | | | | | -| --- | --- | --- | --- | --- | --- | -------- | ---------------------------------------- | --- | --- | --- | --- | -theALCEbenchmarkdemonstratetheefficacy -thecoreideabehindFRONTistofirstselectthesupport- -ofFRONTingeneratingsuperiorgroundedre- -| | | | | | | ing quotes | from | retrieved | sources and | then | condition | -| ------------------------------------ | --- | --- | --- | --- | ---- | ---------- | ---- | --------- | ----------- | ---- | --------- | -| sponsesandhighlysupportivecitations. | | | | | With | | | | | | | -thegenerationprocessonthem,ensuringgroundedre- -LLaMA-2-7B,theframeworksignificantlyout- -sponsesandaccuratecitations. -| performsallthebaselines, | | | achievinganaver- | | | | | | | | | -| ------------------------ | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -ageof14.21%improvementincitationquality -acrossalldatasets,evensurpassingChatGPT1. SuchprevalenceofhallucinationsinLLMout- -| | | | | | | puts has | motivated | the | development | of | attributed | -| --- | --- | --- | --- | --- | --- | -------- | --------- | --- | ----------- | --- | ---------- | -1 Introduction -| | | | | | | systems | (Nakano | et al., | 2021; Thoppilan | | et al., | -| --- | --- | --- | --- | --- | --- | ------- | ------- | ------- | --------------- | --- | ------- | -Therecentadventoflargelanguagemodels(LLMs) 2022;Menicketal.,2022),suchasNewBing2 and -(Touvronetal.,2023;OpenAI,2023;Zhaoetal., Perplexity3,whereLLMsareallowedtogenerate -2023) has taken the world by storm, fueling a responses with in-line citations. Not only does it -| paradigm | shift in | information | | acquisition | (Zhu | | | | | | | -| -------- | -------- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | -improvefactualityandalleviatehallucinations,but -etal.,2023). Despitetheircompellingperformance, italsosimplifiesuserverificationofmodeloutputs, -LLMs still struggle with hallucinations (Ji et al., furtherenhancingtheverifiabilityofLLMs. -2023;Huangetal.,2023),atendencytofabricate Despiterecentadvancements,currentattributed -| non-existent | facts | or generate | unfaithful | | content. | | | | | | | -| ------------ | ----- | ----------- | ---------- | --- | -------- | -------------------------------------- | --- | --- | --- | --- | -------- | -| | | | | | | LLMsstillexposesignificantlimitations. | | | | | Firstly, | -This issue further poses a risk of misinformation current approaches predominantly rely on either -dissemination(ChenandShu,2023),directlyim- in-contextlearning(Gaoetal.,2023b)orpost-hoc -pactingthereliabilityandtrustworthinessofLLMs. -retrieval(Gaoetal.,2023a)toachieveattribution, -*CorrespondingAuthor -1Our data and code can be found at: https://github. 2https://www.bing.com/chat -com/LuckyyySTA/Fine-grained-Attribution. 3https://www.perplexity.ai - -| | | | | | | | 2 RelatedWork | | | | | | | -| ------- | ----------- | ----------- | --- | ---------- | --- | ------ | ------------- | --- | --- | --- | --- | --- | --- | -| lacking | an inherent | attribution | | capability | | within | | | | | | | | -LLMs,therebyresultingincompromisedcitation -| | | | | | | | RetrievalAugmentedGeneration. | | | | | Recently,re- | | -| ----------------------- | --------- | --- | ----------------------- | ---- | --------- | ---- | ----------------------------- | --- | ---------- | --- | ----- | ------------ | --- | -| quality(Liuetal.,2023). | | | Secondly,thesecitations | | | | | | | | | | | -| | | | | | | | trieval augmented | | generation | | (RAG) | (Karpukhin | | -| are typically | presented | | in the | form | of either | doc- | | | | | | | | -ument identifiers (Nakano et al., 2021) or URLs etal.,2020;Lewisetal.,2020;Fengetal.,2023; -Gaoetal.,2023c)hasshownpromiseinknowledge- -| (Thoppilan | et al., | 2022). | Such | coarse | attribution | | | | | | | | | -| ---------- | ------- | ------ | ---- | ------ | ----------- | --- | --------- | ------ | ---------------- | --- | --------- | --- | ----- | -| | | | | | | | intensive | tasks. | By incorporating | | retrieved | | docu- | -complicatestheprocessforuserstoperformfine- -| | | | | | | | ments, | LLMsareequippedwithup-to-dateinfor- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------------------------- | --- | --- | --- | --- | --- | -grainedverification(e.g.,pinpointspecificsupport- -| | | | | | | | mation, | significantly | mitigating | | knowledge | | gaps. | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | ---------- | --- | --------- | --- | ----- | -ingevidence),particularlyinlengthydocuments. -| | | | | | | | However, | recent | studies | (Shi | et al., | 2023; | Yoran | -| ------- | ---- | ------ | ---------- | --- | ---------- | ---- | -------- | ------ | ------- | ---- | ------- | ----- | ----- | -| To this | end, | we aim | to advance | | attributed | text | | | | | | | | -etal.,2023;Xuetal.,2023a;Zhuetal.,2024)have -generationbyempoweringLLMswithfine-grained -attributionability. However,onechallengecomes revealed that existing retrieval-augmented LLMs -| | | | | | | | struggle | to handle | irrelevant | | or contradictory | | re- | -| --- | --- | --- | --- | --- | --- | --- | -------- | --------- | ---------- | --- | ---------------- | --- | --- | -fromtheacquisitionofhigh-qualityattributiondata -trievaldocumentsandeffectivelyutilizecontextual -| for supervised | | fine-tuning, | | which | is difficult | and | | | | | | | | -| ------------------------------------ | --- | ------------ | --- | ----- | ------------ | ------- | --------- | ----- | ----------- | --- | ---------- | --- | ------- | -| | | | | | | | evidence. | These | limitations | | can result | in | perfor- | -| costlytoannotate,andthereforescarce. | | | | | | Thus,we | | | | | | | | -mancedegradationorevenhallucinations(Huang -| start with | an automatic | | data | generation | | pipeline | | | | | | | | -| ---------- | ------------ | --- | ---- | ---------- | --- | -------- | -------------- | --- | ------------ | --- | --------- | --- | ---- | -| | | | | | | | et al., 2023), | | highlighting | the | necessity | for | more | -tailoredforcollectinghigh-qualityattributiondata -factualandverifiablesystems. -(§3.1). Givenauserquery,thepipelineautomates -dataconstructionthroughdocumentretrieval,rele- -| | | | | | | | AttributedLargeLanguageModels. | | | | | Theper- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | --- | ------- | --- | -vancereranking,attributedanswergeneration,and -| | | | | | | | sistent | challenge | of hallucinations | | | within | LLMs | -| ------------------------------- | --- | ------ | ---- | ------------------- | --- | --- | ----------- | --------- | ----------------- | --- | ------------- | ------ | ---- | -| data filtering | to | ensure | both | the informativeness | | | | | | | | | | -| | | | | | | | has spurred | the | development | | of attributed | | LLMs | -| andattributabilityoftheanswers. | | | | Furthermore,to | | | | | | | | | | -(Bohnetetal.,2022;Lietal.,2023;Worledgeetal., -betterunlockLLMs’abilityforfine-grainedattri- -2023),whichseektoenhanceinformationverifia- -| bution, | we introduce | | FRONT, | a two-stage | | train- | | | | | | | | -| ------------- | ------------ | ---- | ------- | ----------- | --- | -------- | --------- | ---------- | --------- | ------ | --------- | ----------- | -------- | -| | | | | | | | bility by | generating | responses | | with | attribution | to | -| ing framework | | that | teaches | LLMs | to | generate | | | | | | | | -| | | | | | | | evidence | sources. | The | way of | providing | | attribu- | -Fine-gRainedgrOuNdedciTations(§3.2),consist- -| | | | | | | | tionsvariesacrossstudies. | | | Forexample,Gaoetal. | | | | -| ------ | --------- | ------ | ---------- | --- | ---- | --- | ------------------------- | --- | --- | ------------------- | --- | --- | --- | -| ing of | Grounding | Guided | Generation | | (G3) | and | | | | | | | | -(2023b)enablesLLMstogeneratetextwithin-line -Consistency-AwareAlignment(CAA).Duringthe -| | | | | | | | citations | via in-context | | learning. | Another | | line of | -| --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------- | --- | ------- | -G3 -stage,theLLMfirstselectssupportingquotes -| | | | | | | | research | (Gao | et al., 2023a; | Xu | et al., | 2023b) | ex- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ---- | -------------- | --- | ------- | ------ | --- | -fromretrievedsources(grounding)andthencondi- -plorespost-hocattribution,whereLLMsfirstgen- -tionsthegenerationprocessonthem(generation). -erateaninitialresponseandthenretrievethemost -TheCAAstagethenutilizespreferenceoptimiza- -| | | | | | | | relevantevidencetoachieveattribution. | | | | | Inthispa- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | --------- | --- | -tiontofurtheralignthegroundingandgeneration -per,weadvancetheresearchonattributedLLMs -processbyautomaticallyconstructingpreference -| | | | | | | | further. | Unlikeexistingmodelsthatpredominantly | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------- | --- | --- | --- | --- | --- | -signals. Inthisway,thesequotescanserveasfine- -| | | | | | | | cite document | | identifiers, | we | delve | into | a more | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ------------ | --- | ----- | ---- | ------ | -grainedcitationsandimprovetheefficiencyofthe -fine-grainedformofattributionbypinpointingand -verificationprocessforusers(seeFigure1). -citingspecificextractivequotes. -| We conduct | | extensive | experiments | | to | evaluate | | | | | | | | -| ---------- | --- | --------- | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -our framework on the ALCE Benchmark (Gao 3 TaskFormulationandMethodology -| etal.,2023b). | Ourfindingsareasfollows: | | | | | | | | | | | | | -| ------------- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Following(Liuetal.,2023;Gaoetal.,2023b),the -| • FRONT | demonstratessupervisorperformance | | | | | | | | | | | | | -| ------- | --------------------------------- | --- | --- | --- | --- | --- | ------------------ | --- | ----------- | --- | ----- | ------ | ----- | -| | | | | | | | task is formalized | | as follows: | | given | a user | query | -gainsincitationqualitycomparedtoallbase- q and a corpus of retrieved documents D as in- -lines,achievinganaverage14.21%improve- -| | | | | | | | put, the | LLM | is required | to | produce | a response | | -| --- | --- | --- | --- | --- | --- | --- | -------- | --- | ----------- | --- | ------- | ---------- | --- | -mentusingLLaMA-2-7B. -| | | | | | | | S, which | consists | of statements | | with | embedded | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | -------- | -------------------- | --- | ---- | -------- | ---- | -| | | | | | | | in-linecitations. | | WeassumetheresponseS | | | | com- | -• Humanevaluationrevealsthatthequotesgen- -| | | | | | | | prising | with | n statements | S | = {s | ,s ,...,s | } | -| ---------------------------------------- | --- | ------------- | --- | --- | ------- | ------- | ----------------- | -------- | ------------ | ----------------------- | ---- | ------------- | --- | -| erated | by | our framework | | are | of high | quality | | | | | | 1 2 | n | -| | | | | | | | andeachstatements | | | ∈ S,citesalistofpassage | | | | -| andsignificantlybenefituserverification. | | | | | | | | | i | | | | | -| | | | | | | | C i = {c | i1 ,c i2 | ,...}, where | c ij | ∈ D. | Specifically, | | -• AnalysisshowsthatFRONTgenerateslesshal- citationsarepresentedintheformof[1][2]. -lucinationanddemonstratesremarkablegen- Next,wepresentacomprehensiveoverviewof -eralizationacrossdifferentbasemodels. ourmethod,whichconsistsoftwoprimarycompo- - -Automatic Data Generation Pipeline -          Data Collection        Answer Generation           Data Filtering -Question: Why do bagels have holes in the middle？ Bagel holes ensures even baking (...) a area -| | | | | | | Demonstrations | | |  1    3  . Another suggests (...) and transport  2  . | | | | -| --- | --- | --- | --- | --- | -------------------------------------------- | -------------- | --- | --- | ----------------------------------------------------- | --- | --- | --- | -| | | | | | Instruction: Generate an answer and cite the | | | | Informative? | | | | -Retrieve source for the question and provided documents Break into -Question: Why do bagels have holes in the middle？ -| | Top-100 retrieval documents | | | | | | | | | | statements | | -| --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | -(...) -| | | | | | Prior to (...) the hole was originally put in place | | | | Statement 1 | Bagel holes ensures even baking by | | | -| --- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | ----------- | ---------------------------------- | --- | --- | -allowing heat to circulate around a -| | | | | | to allow for easier handing prior. | | | | | greater surface area  1    3  .  | | | -| --- | -------- | -------------------------------- | ---------------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | -------------------------------- | --- | --- | -| | T h e b | a k e r s (. .. ) th e h o | l e i n t h e c e n tre of the | | | | | | | | | | -1 ba g e l e n s u r e s t h a t i t b a k e s e v e n ly . Statement 2 A n o t h e r s u g g e s t s t h a t t h e y w e re u s ed -| | | | | | | | | | | fo r e a s i e r | h a n d li n g a n | d tr a n sp o rt   2  . | -| ------ | --------------------------------------------- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | ----------- | ------------------ | -------------------- | --------------------------- | -| | Prior to (...) the hole was originally put in | | | | Bagel holes ensures even baking by allowing | | | | | | | | -| | 2 | | | | heat to circulate around a greater surface | | | | | | | Support | -| | place to allow for easier handing prior. | | | | | | | | Attributed? | | | | -| Rerank | | | | | area   1   3   . Another suggests that they were | | | | | | | | -T h e r i n g s h a p e a l lo w s h e a t t o c ir c u la t e used for easier handling and transport  2  . Documents -| | 3 ar o u n | d ( .. . )  m a k e s b a g | e l s t o c o o k fa s t er  | | | | | | | | | | -| --- | ---------- | ------------------------------ | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Figure2: Overviewofthedatagenerationpipeline. Thepipelineconsistsofthreeprimarysteps: datacollection, -answergeneration,anddatafiltering. Firstly,givenauserquery,thedatacollectionmoduleretrievesthetop100 -relevantdocumentsandemploysarerankingmodeltoselectthetop5mostpertinentdocuments. Subsequently, -attributedresponsesaregeneratedbydistillingChatGPTviain-contextlearning. Finally,allresponsesarefiltered -bythedatafilteringmoduletoensureinformativenessandattributability. -nents: anautomaticdatagenerationpipeline(§3.1) subsequently re-rankedby RankVicuna (Pradeep -andatwo-stagetrainingframework(§3.2). etal.,2023)consideringitssuperiorperformance -| | | | | | | inlistwisere-ranking, | | | resultinginthetop5most | | | | -| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | --- | -3.1 AutomaticDataGenerationPipeline -relevantdocumentsforeachquery. -EquippingLLMswiththeattributioncapabilityne- -| | | | | | | Attributed | Answer | | Generation. | | Given | the re- | -| ---------- | -------- | ---- | ------------- | ------------ | -------- | ---------- | ----------- | --- | ----------- | ------- | ----- | ---------- | -| cessitates | training | data | that includes | high-quality | | | | | | | | | -| | | | | | | markable | performance | | of | ChatGPT | in | attributed | -| responses | paired | with | precise | citations, | which is | | | | | | | | -questionanswering,weemployChatGPTtogener- -| typically | labor-intensive | | and | costly. To | address | | | | | | | | -| --------- | --------------- | --- | --- | ---------- | ------- | --- | --- | --- | --- | --- | --- | --- | -ateanswerswithcorrespondingcitationsforgiven -thischallenge,weproposeapipelinedesignedfor -| | | | | | | queriesandthetop5retrieveddocuments. | | | | | | Wepro- | -| --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | --- | ------ | -theautomaticgenerationofhigh-qualityattributed -videpreciseinstructionsandin-contextdemonstra- -data4. -Thispipelinecomprisesthreecorecompo- -tionstoensurethatChatGPTproducesinformative -nents: datacollection,attributedanswergeneration, -responsesandcitesthesourcesaccordingly. -anddatafiltering,asoutlinedinFigure2. -| | | | | | | DataFiltering. | | Toguaranteethehighqualityof | | | | | -| --------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | --------------------------- | ----- | --------- | --- | ----------- | -| DataCollection. | | Tosimulatethereal-worldenvi- | | | | | | | | | | | -| | | | | | | our synthetic | | training | data, | we employ | | a data fil- | -ronmentforinformation-seeking,wecollectques- -teringprocessguidedbytwokeycriteriaderived -tionsfromtheAQuAMuSedataset(Kulkarnietal., -| | | | | | | fromKamallooetal.(2023): | | | | (1)informativeness: | | | -| --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ------------------- | --- | --- | -2020),whichisderivedfromtheNaturalQuestion -| | | | | | | assessing | if the | answer | provides | | sufficient | infor- | -| ----------------------------------- | --- | --- | --- | --- | ----- | ------------------------------------------ | ------ | ------ | -------- | --- | ---------- | ------ | -| (NQ)dataset(Kwiatkowskietal.,2019). | | | | | TheNQ | | | | | | | | -| | | | | | | mationtothequestion,and(2)attributability: | | | | | | de- | -datasetcomprisesrealuserqueriesfromtheGoogle -| | | | | | | termining | if the | answer | is | attributed | | to the cited | -| --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | --- | ---------- | --- | ------------ | -searchengine,providingarobustbasisforrealistic -| | | | | | | documents. | To | mitigate | | the impact | of | nonsensi- | -| ------------------ | --- | ---------- | --- | ----------- | ----- | ---------- | --- | -------- | --- | ---------- | --- | --------- | -| question-answering | | scenarios. | | The dataset | spans | | | | | | | | -calqueriesandirrelevantdocumentretrievalthat -| a range | of diverse | question | types, | demanding | an- | | | | | | | | -| ------- | ---------- | -------- | ------ | --------- | --- | -------- | ------------------ | --- | --- | -------- | --- | ---------- | -| | | | | | | may lead | to non-informative | | | answers, | | we utilize | -swersofvaryinglengths,fromconcisetodetailed. -ChatGPTforpreliminaryinformativenessannota- -Tomimicthewayasearchenginemightsynthesize -| | | | | | | tions. Responses | | categorized | | as | non-informative | | -| --- | --- | --- | --- | --- | --- | ---------------- | --- | ----------- | --- | --- | --------------- | --- | -documentsofhighrelevanceinresponsetoauser -| | | | | | | aredirectlyexcluded. | | | Furthermore,toensurethat | | | | -| ------ | --------- | ------ | ------- | ------- | -------- | -------------------- | --- | --- | ------------------------ | --- | --- | --- | -| query, | we employ | Sphere | (Piktus | et al., | 2021), a | | | | | | | | -answersareaccompaniedbyhighlysupportiveci- -pre-processedandcleanedversionoftheCommon -tations,wetrainadiscriminatoronhuman-labeled -Crawlcorpus,servingasaproxywebsearchindex. -| | | | | | | data from | the | comprehensive | | evaluation | | by Liu | -| --- | --- | --- | --- | --- | --- | --------- | --- | ------------- | --- | ---------- | --- | ------ | -Inparticular,foragivenuserquerysampledfrom -| | | | | | | et al. (2023), | | where | attributability | | is | categorized | -| --- | ------------ | ---------- | ------------ | ---------- | ---- | ---------------- | ------------------------------------- | ------------------------------- | --------------- | ----- | --- | ----------- | -| the | AQuAMuSe | dataset, | we initially | retrieve | the | | | | | | | | -| | | | | | | intothreelevels: | | fullsupport,partialsupport,orno | | | | | -| top | 100 relevant | documents | from | the Sphere | cor- | | | | | | | | -| | | | | | | support. | Wequantitativelymapthediscriminator’s | | | | | | -| pus | using sparse | retrieval. | These | documents | are | | | | | | | | -| | | | | | | outputs | to an | attributability | | score | and | ultimately | -4Attributeddatarefersto“answerswithin-linecitations”. deriveanaveragescoreforeachattributedanswer. - -Two-Stage Training Recipe -          Grounding Guided Generation        Consistency-Aware Alignment -          Answer Generation -| Q :   W h y d o | b a g e l s h a v e | | | | | | | | | | | | -| ----------------- | ----------------------- | ---------------- | --------------------------- | ----------------------- | ------------------------------------- | --- | ----------------------- | ------------------- | ----------------------------------------------- | --------------------------- | --- | --- | -| | | G r o u n d i ng | T h e b a k e r s ( . . | . ) t h e h o l e | i n t h e c e n t r e o f the | Q | u e s t i o n :   W h y | d o b a g e l s |   G  R      D eI   G    1l   e | n s u r e s t h a t i t | | | -h o l e s i n t h e m i d d l e ？ 1 b a g e l { e n s u r e s t h a t i t b a k e s e v e n l y } . h a v e h o l e s i n t h e m i d d l e ？ b a kO eU Ns N v e n y .   3   a l l o w s h e a t -| | | | | | | | ( . . . ) | | t o c i r c u l a t e a r | o u n d .  2    a l l o w   s A u   N r S  f W  a   c E | R e     a B r a e g a e   1 l   h  3 o  A l e n s o ( t . h . . e )   r | g ( r . e .. a ) t | -| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------- | -B a g e l h o l e s e n s u r e s e v e n b a k i n g b y P r i o r t o ( . . . ) t h e h o l e w a s o r i g i n a l l y p u t in P r i o r t o ( .. . )   p l a c e t o a ll o w f o r e a s i e r h a n d i n g p r i o r . h a n d l i n g a n d t r a n s p o r t   2  . -| al lo w i n g h e a | t to c i r c u l a t e | a r o u n d a 2 | | | | | | | | | | | -| -------------------- | -------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -greater surface area  1   3  .Another p l a c e t o { a l l o w f o r e a s ie r h a n d i n g p r i o r } . fo r e a s i e r h a n d i n g p r i o r . Consistency -| suggests that they were used for | | 3 | The ring shape {allows heat to circulate | | | | | | | | | | -| -------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -easier handling and transport  2  . around} (...) makes bagels to cook faster  Gold Grounding Consistent Answer Contrastive -Large Language -| | Supervised Fine-tuning | | | | | | Model | | | | | | -| -------- | ---------------------- | -------------- | ------------- | --------------------------------------------- | ------------------------------------------------- | --- | ----- | --- | -------------- | ------------------- | --- | --- | -| | | | | | | | | | Gold Grounding | Inconsistent Answer | | | -| Question | | |   G   R   | Ol  Uo N  D  sI N   Gh     1   e n s | u r e s t h a t it b a k e s e v e n ly.  | | | | | | | | -| | | Large Language |   f3 a | l w e a t t o | c i r c u l a t e a r o u n d .   2  a l lo w | | | | | | | | -| | | | o r | e a s i e r h a n d i n | g p r i o r . | | | | | | | | -Documents Model Direct Preference Optimization (DPO)   A   N S  W    E R    B a g e l h o l e s   (. . . )  f o r -| | | |   A A  N n  S o | W  t h  E e  R r   B (. a .. g ) e h l a h n o d le | lin s g ( . a .. n )  d g r t e ra a n t s s p u o rf r a t  c 2  e . area13 | | | | | a e s t h e | t i c s a n d u n i q u e t e x | t u r e   1 | -| --- | --- | --- | ---------------- | ------------------------------------------------------------ | ------------------------------------------------------------------------------------- | --- | --- | --- | --- | ----------------- | ------------------------------------- | ------------- | -| | | | | | | | | | |   ( . . . ) h a | n d l i n g a n d tr a n s p o | r t   3   . | -Figure 3: Overview of FRONT: The training recipe consists of two stages: grounding-guided generation and -consistency-aware alignment. It enables LLMs to first generate precise grounding and subsequently guide the -generationofattributedanswers,therebyenhancingfine-grainedattributioncapability. -Answersfallingbelowadefinedthresholdaresys- fine-tunedtogeneratearesponseS whichconsists -tematicallyexcludedtoensurethesyntheticdata’s of two components: the grounded quotes G and -reliability, which results in nearly 8,000 entries. theattributedanswerA. Specifically,thegrounded -Formoredetails,pleaserefertoAppendixA. quotesG aredelineatedasfollows: -3.2 Two-StageTrainingRecipe G = {[GROUNDING],(i ,e ),...,(i ,e )}, (1) -| | | | | | | | | | | 1 1 | n n | | -| ------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Inthissection,weintroduce | | | FRONT,atwo-stage | | | | | | | | | | -where[GROUNDING]denotesaspecialtokenindi- -trainingframeworkthataimsatempoweringLLMs -| | | | | | | cating | the | start of | the grounding | process. | | Each | -| ----------------- | --- | ----------- | ----------- | --- | -------- | ------ | --- | -------- | ------------- | -------- | --- | ---- | -| with fine-grained | | attribution | capability. | | Figure 3 | | | | | | | | -tuplewithinG,comprisingadocumentidentifieri -illustratestheoverviewofourframework. -andthecorrespondingextractivesegmente,collec- -tivelyformingagroundedquote. -3.2.1 GroundingGuidedGeneration -| | | | | | | Similarly, | | the formulation | | of the | attributed | an- | -| ---------- | --- | --------- | ------------ | --- | -------- | ---------- | --- | --------------- | --- | ------ | ---------- | --- | -| To empower | | LLMs with | fine-grained | | attribu- | | | | | | | | -swerAisconciselypresentedas: -| tion capability, | | we propose | Grounding | | Guided | | | | | | | | -| ---------------- | --- | ---------- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -(G3), -Generation which teaches LLMs to gener- A = {[ANSWER],s ,s ...,s }, (2) -| | | | | | | | | | | 1 2 | m | | -| ------------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| atefine-grainedcitations. | | | ThecornerstoneofG3 | | | | | | | | | | -liesinenablingLLMstoextractsupportingquotes where [ANSWER] is a special token that signals -fromthesourcedocuments,eachassociatedwithits the beginning of the answer generation process. -documentidentifier,whichinturnguidesthegen- Each statement s cites a list of passages C = -| | | | | | | | | i | | | | i | -| --------------------------- | --- | --------------------------- | ----------------- | --- | --- | ----------------- | ------------ | ------ | --- | ------------------ | ---- | ----- | -| erationofattributedanswers. | | | Suchaformatoffers | | | | | | | | | | -| | | | | | | {c i1 | ,c i2 ,...}, | wherec | ij | ⊆ {i 1 ,i 2 ,...,i | n }, | asde- | -| twoprimarybenefits. | | Firstly,thedirectextraction | | | | finedinEquation2. | | | | | | | -of quotes from sources significantly reduces the Thus,thetraininglossisformulatedas: -impactoftheincorporationofirrelevantinforma- -| tion and | the risk | of hallucinations | | in | subsequent | | | N | | | | | -| -------- | -------- | ----------------- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -(cid:88) -| | | | | | | | | L = − | logP(y | |q ,D | ;θ) | (3) | -| ------------------ | --- | ---------------------------- | --- | --- | --- | --- | --- | ----- | ------ | ----- | --- | --- | -| attributedanswers. | | Secondly,theprocessnaturally | | | | | | | | i i | i | | -i=1 -facilitatesaccurateattribution,witheachdocument -identifierservingasaclearsupervisedsignalthat -| | | | | | | where | y | represents | the | combined | output | of | -| --- | --- | --- | --- | --- | --- | ----- | --- | ---------- | --- | -------- | ------ | --- | -i -delineatestheoriginoftheextractivequotes,thus -groundedquotesGandtheanswerAforeachgiven -improvingthecitationquality. queryq andsetofretrieveddocumentsD . -| | | | | | | | i | | | | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -However,theabsenceofspecificgroundingcon- -tent for statements within our generated dataset 3.2.2 Consistency-AwareAlignment -posesadditionalchallenges. Totacklethis,weem- While G3 unlocks the ability to first extract sup- -ploy ChatGPT to meticulously extract segments portingquotesbeforegeneratingattributedanswers, -fromciteddocumentsthatsupportthecorrespond- it occasionally leads to inconsistencies between -ingstatement. Hence,whengivenaqueryqandthe groundedquotesandattributedanswers. Suchdis- -top-5retrieveddocumentsD asinput,theLLMis crepancieschallengetheattempttoemploythese - -groundedquotesasfine-grainedverification. Inre- whereπ ref representsthereferencemodel,initial- -sponse,weproposeaconsistency-awarealignment izedfromG3. Thehyper-parameterβ modulates -(CCA) stage specifically aimed at enhancing the the divergence between the distribution from the -consistencybetweenthegroundingprocessandthe policy model and the reference model. τ is the -w -generationprocess. consistentanswer,whileτ istheinconsistentone. -l -| | The cornerstone | | of | our approach | involves | con- | | | | | | | | -| --- | --------------- | --- | --- | ------------ | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -4 ExperimentalSettings -| trasting | | a consistent | | answer | with an | inconsis- | | | | | | | | -| --------------- | --- | ------------ | -------------------------- | -------- | ----------- | --------- | ------------ | --- | --- | --- | --- | --- | --- | -| tent | one | under | the | guidance | of the same | oracle | 4.1 Datasets | | | | | | | -| groundedquotes. | | | Thisalignswiththeconceptof | | | | | | | | | | | -WeconductexperimentsontheALCEbenchmark -| Reinforcement | | | Learning | from | Human | Feedback | | | | | | | | -| ------------- | --- | --- | -------- | ---- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | -(Gaoetal.,2023b),designedforattributedtextgen- -| (RLHF) | | (Ouyang | et | al., 2022; | Bai et | al., 2022), | | | | | | | | -| ------ | --- | ------- | --- | ---------- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -eration. Thebenchmarkincludesthreelong-form -whereLLMsarefurtherfine-tunedtodistinguish -QAdatasetsthatspanvarioustypesofquestions. -betweendesirableandundesirableresponsesunder -preference feedback. However, such contrastive ASQA (Stelmakh et al., 2022) is a long-form -preferencefeedbacktypicallycomesfromhuman factoidQAdatasetcharacterizedbyinherentlyam- -| annotation. | | Toautomaticallyconstructpreference | | | | | | | | | | | | -| ----------- | --- | ---------------------------------- | --- | --- | --- | --- | ------- | --------- | ---- | ------- | -------- | ----- | --- | -| | | | | | | | biguous | questions | that | require | multiple | short | an- | -pairsforpreferenceoptimization,weutilizetheat- swerstoencapsulatedifferentviewpoints. -tributedanswersgeneratedbysmallerLLMs(e.g., -LLaMA-2-7B)underthein-contextlearningsetting ELI5 (Fanetal.,2019)featuresopen-endedques- -tionsintendedforsimplificationtothecomprehen- -| asnegativesamples. | | | | Theseanswers,characterized | | | | | | | | | | -| ------------------ | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -sionleveloffive-year-olds,requiringexplanatory -bytheirlowqualityandinconsistencywithoracle -multi-sentenceresponses. -groundedquotes,automaticallyserveascontrastive -supervisionsignalswhenpairedwithhigh-quality -| | | | | | | | QAMPARI | | (Amouyaletal.,2022)isafactoid | | | | | -| ------------------------------- | --- | --- | --- | --- | --------------- | --- | ------- | --- | ----------------------------- | --- | --- | --- | --- | -| attributedanswerslabeledin§3.1. | | | | | Inthisscenario, | | | | | | | | | -QAdatasetderivedfromWikipedia,whereanswers -theprocessnotonlyencouragestheLLMtogener- arestructuredasacompilationofentities. -ateattributedanswersthataremoreconsistentwith -thegroundedquotesbutalsofacilitatestheidentifi- 4.2 EvaluationMetrics -cationandcorrectionofnuancederrorspresentin Following the ALCE benchmark (Gao et al., -smallermodels. 2023b), our evaluation primarily focuses on two -| | Specifically, | | we adopt | Direct | Preference | Opti- | | | | | | | | -| --- | ------------- | --- | -------- | ------ | ---------- | ----- | --------------- | --- | -------- | --- | ------- | ------------ | --- | -| | | | | | | | key dimensions: | | Citation | | Quality | and Correct- | | -mization(Rafailovetal.,2023),avariantofRLHF ness. Detaileddescriptionsofadditionalevaluation -known for its stability, for our contrastive align- dimensionsarepresentedintheAppendixB. -ment. Formally,foreachinstance,giventheoracle -| | | | | | | | CitationQuality. | | Citationqualityiscriticalfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | --- | -groundedg(i)alongwithaconsistentoracleanswer -| (i) | | | | | (i) | | evaluating | LLM | attribution, | | assessed | along | two | -| --- | --------------------------- | --- | --- | --- | --- | --------- | ----------- | --- | ------------ | ------- | ----------- | ----- | --- | -| y | aswellasanattributedanswery | | | | | generated | | | | | | | | -| w | | | | | l | | | | | | | | | -| | | | | | | | dimensions: | (1) | Citation | Recall, | determining | | if | -byaweakerLLMviain-contextlearning,wecan -| | | | | | | | the output | is entirely | | supported | by | the cited | doc- | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | --- | --------- | --- | --------- | ---- | -simplyconstructapreferencedataset: -| | | | | | | | uments, | and (2) | Citation | Precision, | | assessing | if | -| --- | --- | --- | --------------------- | --- | ------------- | --- | ---------------------------------------------- | ------- | -------- | ---------- | --- | --------- | --- | -| | | D | = (cid:8) x(i),τ(i),τ | | (i)(cid:9)N , | | | | | | | | | -| | | | | w | | (4) | eachcitationsupportsitscorrespondingstatement. | | | | | | | -l i=1 -| | (i) | | (i) | | | | EvaluationisconductedbyTRUE(Honovichetal., | | | | | | | -| ------ | --- | -------- | --- | ------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -| whereτ | | = g(i)◦y | | denotestheconcatenationof | | | | | | | | | | -| | w | | w | | | | 2022),aT5-11Bmodelfine-tunedonacollection | | | | | | | -theoraclegroundingwiththeconsistent,attributed -ofNLIdatasetstoautomaticallyexaminetheentail- -| | | (i) | g(i)◦y | (i) | | | | | | | | | | -| -------- | --- | ------------ | ------ | ----------------------- | ------- | ------- | ------------------------------------------ | --- | ------- | ---------- | ------- | --- | ------ | -| answer,τ | | = | | denotestheconcatenation | | | | | | | | | | -| | | l | | l | | | mentofciteddocumentsandthemodelgeneration. | | | | | | | -| with | the | inconsistent | | attributed | answer. | Here, ◦ | | | | | | | | -| | | | | | | | Additionally, | to | capture | a holistic | measure | | of ci- | -signifiestheoperationofstringconcatenation. -tationquality,wealsoreporttheCitationF1,the -| | Finally,wecanoptimizethepolicymodelπ | | | | | on | | | | | | | | -| --- | ------------------------------------ | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | θ | harmonicmeanofcitationprecisionandrecall: | | | | | | | -thedatasetD -byminimizingthefollowingloss: -citationprecision·citationrecall -| | L | (π ;π | ;D) | | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | ----- | -| | DPO | θ | ref | | | | F 1 = 2· | | | | | | , (6) | -citationprecision+citationrecall -| | | | (cid:20) | (cid:18) | π (τ | |x) | | | | | | | | -| --- | --- | --- | -------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -| | =−E | | | | θ | w | | | | | | | | -| | | | | logσ | βlog | | | | | | | | | -(x,τw,τ l )∼D π (τ |x) (5) Correctness. FortheASQAdataset,correctness -ref w -π (τ |x) (cid:19)(cid:21) is quantified using exact match recall (EM Rec.) -θ l -| | −βlog | | | , | | | | | | | | | | -| --- | ----- | --- | ------ | --- | --- | --- | ----------- | ------- | --- | --------- | ------- | --- | ----- | -| | | π | (τ |x) | | | | by checking | whether | | the short | answers | are | exact | -| | | | ref l | | | | | | | | | | | - -| | | | ASQA | | | ELI5 | | QAMPARI | | | -| --------- | --- | ----------- | ---- | --------- | ----------- | ---- | -------- | ------------ | ---------- | --- | -| ModelType | | ModelSize | | | | | | | | | -| | | Correctness | | Citation | Correctness | | Citation | Correctness | Citation | | -| | | EMRec. | Rec. | Prec. F1. | Claim | Rec. | Prec. F1 | Rec.-5 Prec. | Rec. Prec. | F1 | -Prompting-based -ChatGPT - 40.37 72.81 69.69 71.22 12.47 49.44 47.05 48.22 20.28 19.84 19.06 22.03 20.44 -7B 24.32 17.24 17.87 17.55 4.53 3.92 5.38 4.54 12.56 11.32 6.03 6.35 6.19 -LLaMA-2 13B 27.99 16.45 19.04 17.65 7.77 8.49 8.43 8.46 18.00 12.39 5.45 5.74 5.59 -70B 31.53 44.18 44.79 44.48 10.43 23.75 22.43 23.07 18.50 14.79 10.10 10.50 10.30 -7B 29.93 55.99 51.66 53.74 12.47 19.90 15.48 17.41 17.96 19.74 9.58 9.68 9.63 -LLaMA-2-Chat 13B 34.39 37.15 38.17 37.65 13.83 16.50 16.09 16.29 21.34 18.86 8.94 9.06 9.00 -70B 41.24 60.19 61.16 60.67 13.30 36.63 36.63 36.63 22.62 18.04 13.49 13.98 13.73 -Vicuna-v1.5 7B 38.34 48.37 44.63 46.42 12.30 29.81 22.45 25.61 14.22 14.74 11.26 11.64 11.45 -13B 35.20 51.92 53.40 52.65 14.33 31.15 28.99 30.03 22.06 19.60 13.04 13.74 13.38 -7B 29.46 23.12 25.45 24.23 8.47 16.04 16.32 16.18 16.96 15.98 7.50 7.76 7.63 -Mistral -8×7B 36.30 32.72 34.49 33.58 10.43 26.11 25.09 25.59 18.18 15.63 9.72 10.20 9.95 -7B 38.57 64.90 59.67 62.18 11.07 49.25 42.69 45.74 17.52 21.29 17.56 18.53 18.03 -Mistral-Instruct 8×7B 44.11 61.80 63.27 62.53 13.93 49.28 48.34 48.81 20.12 19.64 19.27 20.38 19.81 -Post-hocRetrieval -ChatGPT - 37.68 27.11 27.05 27.08 18.77 14.55 14.55 14.55 25.14 22.85 12.29 12.29 12.29 -LLaMA-2-Chat 70B 29.68 24.51 24.51 24.51 16.03 12.93 12.93 12.93 17.90 14.45 9.05 9.05 9.05 -Mistral-Instruct 8×7B 33.90 24.57 24.48 24.52 17.37 15.68 15.68 15.68 24.16 18.28 9.78 9.78 9.78 -Training-based -7B 29.96 67.82 66.97 67.39 6.90 22.34 32.40 26.45 2.34 1.98 10.53 18.80 13.50 -Self-RAG(LLaMA-2) -13B 31.66 71.26 70.35 70.80 6.07 30.46 40.20 34.66 1.90 1.33 12.79 20.90 15.86 -7B 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -VANILLA-SFT(LLaMA-2) 13B 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -FRONT(LLaMA-2) 7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -Table1: MainresultsontheALCEbenchmark. Boldnumbersindicatethebestperformance,while_indicatesthe -second-bestperformance. -substringsofthegeneration. RegardingtheELI5 Mistral-7B(Jiangetal.,2023)toMistral-8x7B- -dataset,correctnessismeasuredthroughclaimre- MoE (Jiang et al., 2024). Regarding SFT LLMs, -call (Claim), evaluating whether the model’s re- weselecttheSFTcounterpartsoftheopen-source -sponse entails the ground truth sub-claims. For foundationalLLMsweused. Detailedprompting -theQAMPARIdataset,correctnessisassessedus- settingscanbefoundinAppendixC. -| ingexactmatchprecision(Prec.) | | | andtop-5exact | | | | | | | | -| ----------------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | -match recall (Rec.-5) — considered 100% if the 4.3.2 Post-hocRetrievalMethods. -predictionincludesatleastfivecorrectanswers. -| | | | | | Following | Gao | et al. | (2023b), | we first | instruct | -| --- | --- | --- | --- | --- | --------- | --- | ------ | -------- | -------- | -------- | -LLMstoanswerthegivenqueryinaclosed-book -4.3 Baselines -| | | | | | setting, | and then | integrate | citations | in a post-hoc | | -| --- | --- | --- | --- | --- | -------- | -------- | --------- | --------- | ------------- | --- | -Wecompareourmethodwiththreetypesofbase- -| | | | | | manner. | Foreachgeneratedstatement,weemploy | | | | | -| ------ | ---------------- | -------- | ---------- | --- | ------- | ---------------------------------- | --- | --- | --- | --- | -| lines: | prompting-based, | post-hoc | retrieval, | and | | | | | | | -GTR(Nietal.,2022)toidentifyandcitethemost -training-based. -relevantdocumentfromthetop100retrieveddoc- -4.3.1 Prompting-basedMethods. uments. Weutilizethesamemodelsmentionedin -WedirectlypromptLLMsusingfew-shotdemon- prompting-basedsettingsforthisbaseline. -strations,eachconsistingofaquery,thetop5rel- -| | | | | | 4.3.3 | Training-basedMethods. | | | | | -| --- | --- | --- | --- | --- | ----- | ---------------------- | --- | --- | --- | --- | -evantretrieveddocuments,andananswerwithin- -| linecitations. | Ourexperimentsencompassarange | | | | | | | | | | -| -------------- | ----------------------------- | --- | --- | --- | -------- | --------------------------------- | --- | --- | --- | --- | -| | | | | | Self-RAG | (Asaietal.,2023)Self-RAGtrainsthe | | | | | -ofLLMs,fromfoundationalmodelstosupervised -| | | | | | LLM | to learn | to adaptively | retrieve | passages | on- | -| --------------------- | --- | -------------------- | --- | --- | --- | -------- | ------------- | -------- | -------- | --- | -| fine-tuning(SFT)LLMs. | | ForfoundationalLLMs, | | | | | | | | | -demandandenableittoreflectonitsgenerationto -| we select | GPT-3.5-Turbo5 | as the | representative | | | | | | | | -| --------- | -------------- | ------ | -------------- | --- | --- | --- | --- | --- | --- | --- | -furtherimprovegenerationqualityandattributions. -closed-sourcemodel,recognizedforitsnotableper- -| formance. | Among | the open-source | foundational | | | | | | | | -| --------- | ----- | --------------- | ------------ | --- | ----------- | --- | -------------------------- | --- | --- | --- | -| | | | | | VANILLA-SFT | | Wedirectlyemploysupervised | | | | -LLMs,wefocusontheLLaMA-2seriesincluding -fine-tuningtotraintheLLMonourgeneratedtrain- -LLaMA2-7B,LLaMA2-13B,andLLaMA2-70B, -| | | | | | ing data. | Given | a query | and corresponding | | doc- | -| ------------------------------------------------ | -------------- | ------------- | ----- | ---- | --------------------- | ------- | ------- | ----------------- | -------- | -------- | -| as well | as the Mistral | series, which | spans | from | | | | | | | -| | | | | | uments, | the LLM | is | required to | directly | generate | -| 5Specifically,weutilizegpt-3.5-turbo-1106version | | | | | answerswithcitations. | | | | | | - -4.4 ImplementDetails -| Weimplement | | FRONTwithdifferentsizesoffoun- | | | | | 60 | | | | Pre-filtered | | | -| ----------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --- | -Post-filtered -| dational | models | (LLaMA-2-7B | | and | LLaMA-2- | | | | | | | | | -| -------- | ------ | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -)%(1FnoitatiC 50 -| 13B)toevaluateitseffectiveness. | | | | Duringtheeval- | | | | | | | | | | -| ------------------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -40 -uation,FRONTutilizethesameretrievalsettingsas -| those outlined | | by Gao | et al. | (2023b). | Additional | | 30 | | | | | | | -| -------------- | ----------- | ------ | ---------- | -------- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| details | of training | and | evaluation | settings | | can be | 20 | | | | | | | -foundinAppendixD. -| | | | | | | | | ASQA | | ELI5 | | QAMPARI | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ------- | --- | -5 ResultsandAnalysis -| | | | | | | | Figure4: | AblationStudyonDataFiltering. | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----------------------------- | --- | --- | --- | --- | --- | -5.1 OverallResults -Simplysupervisedfine-tuningcanboostcitation -| | | | | | | | FRONT | demonstrates | | remarkable | | generaliza- | | -| ----------- | ------------------------------ | --- | -------------- | --- | ---------- | ----- | ---------- | ------------ | --- | ---------- | --- | ----------- | ------- | -| quality. | AsshowninTable1,teachingLLMsto | | | | | | | | | | | | | -| | | | | | | | tion. | Compared | to | the varied | | queries | and an- | -| generate | responses | | with citations | via | supervised | | | | | | | | | -| | | | | | | | swer types | present | in | the ALCE | | benchmark, | our | -| fine-tuning | significantly | | enhances | | citation | qual- | | | | | | | | -trainingdata,derivedexclusivelyfromtheAQuA- -ity,demonstratingsubstantialimprovementsover -MuSedataset(Kulkarnietal.,2020),exhibitsout- -bothprompt-basedandpost-hocretrievalbaselines -| | | | | | | | of-domain | characteristics. | | | Nonetheless, | | FRONT | -| ------ | ------------- | --- | ------------- | ---- | -------- | --- | ------------ | ---------------- | --- | -------- | ------------ | --- | --------- | -| across | all datasets. | | Specifically, | with | LLaMA-2- | | | | | | | | | -| | | | | | | | demonstrates | superior | | citation | quality, | | affirming | -7B, VANILLA-SFTledtosubstantialgainsincita- -itsexceptionalabilitytogeneralizeacrossdiverse -| tion F1 | scores | over | prompting: | ASQA | (17.55 | → | | | | | | | | -| ------- | ------ | ---- | ---------- | ---- | ------ | --- | -------------------------------- | --- | --- | --- | --- | ------------- | --- | -| | | | | | | | querytypesandretrievaldocuments. | | | | | Additionally, | | -65.61),ELI5(4.54→41.15),andQAMPARI(6.19 -| | | | | | | | while not | specifically | | optimized | | for correctness, | | -| -------- | ----------------------------------- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------------ | ---------------- | --- | -| →21.35). | Thesegainshighlighttheeffectiveness | | | | | | | | | | | | | -| | | | | | | | FRONT | also showcases | | modest | improvements | | in | -ofourtrainingdatagenerationpipeline. -thismetricoverVANILLA-SFTontheASQAand -| | | | | | | | QAMPARIdatasets. | | | However,FRONTencounters | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ----------------------- | --- | --- | --- | -FRONT achievessignificantperformancegains -lowerRec.-5ontheQAMPARIdataset,likelydue -| andsurpassesChatGPT. | | | | While VANILLA-SFT | | | | | | | | | | -| -------------------- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tothenatureofitsanswers,whichconsistofcon- -| demonstrates | | strong | performance, | | it still | shows | | | | | | | | -| ------------ | --- | ------ | ------------ | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | -catenatedentities,divergingsignificantlyfromour -| notable | discrepancies | | compared | to | leading | open- | | | | | | | | -| ------- | ------------- | --- | -------- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | -trainingdata. -sourceLLMs,suchasMixtral-8×7B-Instruct(e.g., -| 41.15 vs. | 45.74) | and | ChatGPT | (e.g., | 41.15 | vs. | | | | | | | | -| --------- | ------ | ---- | -------- | ----------- | ----- | -------- | ----------------- | --- | --- | --- | --- | --- | --- | -| 48.22) | on the | ELI5 | dataset. | FRONT | not | only | 5.2 AblationStudy | | | | | | | -| bridges | these | gaps | but also | establishes | | signifi- | | | | | | | | -Weconductablationstudiestoverifytheeffective- -| cant leads | across | all | datasets. | Specifically, | | us- | | | | | | | | -| --------------- | ------ | --- | --------- | --------------- | --- | ---- | ----------------------------------- | --- | --- | --- | --- | --- | ------ | -| | | | | | | | nessofdifferentcomponentsproposedin | | | | | | FRONT. | -| ing LLaMA-2-7B, | | | FRONT | comprehensively | | out- | | | | | | | | -performsChatGPT,achievingincreasesof3.32%, -| | | | | | | | EffectsofDataGenerationPipeline. | | | | | | Asillus- | -| ------- | --- | ------ | ----------- | ------- | --- | ------ | -------------------------------- | -------- | ------ | --- | -------- | ------ | -------- | -| 18.04%, | and | 21.28% | in citation | quality | | on the | | | | | | | | -| | | | | | | | trated | in §5.1, | simply | SFT | achieves | strong | per- | -ASQA,ELI5,andQAMPARIdatasetsrespectively. -| | | | | | | | formance, | underscoring | | the | high | quality | of our | -| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | --- | --- | ---- | ------- | ------ | -Thisperformanceunderscorestheeffectivenessof -| | | | | | | | training | data. | Furthermore, | | data | filtering, | a cru- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------ | --- | ---- | ---------- | ------ | -FRONTinenhancingattributioncapabilities. -| | | | | | | | cial component | | of our | data | generation | | pipeline, | -| -------------------------------------- | --- | --- | --- | --- | --- | --- | -------------- | ------- | ------- | -------- | ---------- | ------- | --------- | -| | | | | | | | plays a | pivotal | role in | ensuring | the | quality | of the | -| FRONTexhibitsscalabilitywithmodelsize. | | | | | | As | | | | | | | | -illustratedatthebottomofTable1,theperformance generated data by filtering out queries that yield -of FRONT in terms of citation quality shows no- non-informativeanswersorfailtomeetattribution -| | | | | | | | criteria. | To validate | | the effectiveness | | of | our data | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ----------------- | --- | --- | -------- | -tableimprovementswhenscalingfrom7Bto13B. -filteringstrategies,weconductedexperimentscom- -Specifically,weobserveimprovementsof3.23%in -ASQA,4.97%inELI5,and1.33%inQAMPARI. paringmodelsfine-tunedonbothpre-filteredand -This upward trend underscores the scalability of post-filtereddata. Theresults,depictedinFigure -4,confirmthatmodelstrainedonfiltereddataex- -FRONT withincreasingmodelsize,demonstrating -thepotentialofFRONTinleveragingtheincreased hibitanotableimprovementincitationqualityover -capabilitiesoflargerLLMsforfurtherperformance thosetrainedonunfiltereddata,achievingsuperior -attributionperformancewithreduceddatavolume. -gains. - -| | | | | ASQA | | | | ELI5 | | | QAMPARI | | | -| --- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | ------- | --- | --- | -Model -| | | | Correctness | | Citation | | Correctness | | Citation | Correctness | | Citation | | -| --- | --- | --- | ----------- | --- | -------- | --- | ----------- | --- | -------- | ----------- | --- | -------- | --- | -EMRec. Rec. Prec. F1. Claim Rec. Prec. F1 Rec.-5 Prec. Rec. Prec. F1 -FRONT-7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -SELF-GUIDE(w/oConsistency) 38.99 70.69 64.48 67.44 10.04 47.63 44.80 46.17 12.18 20.03 22.50 22.58 22.54 -VANILLA-SFT(w/oGround) 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -FRONT-13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -SELF-GUIDE(w/oConsistency) 40.99 73.08 68.13 70.52 10.06 50.68 49.78 50.23 13.94 22.38 23.73 23.99 23.85 -VANILLA-SFT(w/oGround) 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -Table2: AblationstudyontheimpactofdifferenttrainingstageswithintheALCEbenchmark. -44.42 -Citation-F1 -| | | | | 46.17 | | | | | 33 | | FRONT-13B | | | -| -------------- | --- | ----- | ----- | ----- | ----------- | --- | --- | ------------------- | --------------- | --- | --------- | ------------ | --- | -| | | | | | | | | | | | F | -7 B | | -| | | 43.71 | | | SELF-GUIDE | | | | | | R O | N T | | -| Citation-Prec. | | | | | | | | | LLaMA2-70B-Chat | | SEL F- G | U I D E -13B | | -| | | | 44.80 | | PROMPTGUIDE | | | ssenlufhtiaF 22..55 | | | | | | -SELF-GUIDE-7B -45.16 -| Citation-Rec. | | | | | | | | | | | Mixtral-8x7B-Inst | | | -| ------------- | --- | ----- | --- | ----- | ----- | --- | --- | --- | --- | ---------- | ----------------- | --- | --- | -| | | | | | 47.63 | | | | 22 | Vicuna-13B | | | | -| | | 44.00 | | 46.00 | 48.00 | | | | | | | | | -LLaMA2-7B-Chat -| Figure5: | Ablationstudyofdifferentgroundingguid- | | | | | | | 11..55 | | | | | | -| -------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -ancestrategiesontheELI5dataset. -| | | | | | | | | | 2200 | 3300 | 4400 5500 | 6600 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --------- | ---- | --- | -CitationF1 -EffectsofGroundingGuidedGeneration(G3). -G3 -empowers LLMs to first select relevant fine- Figure6: TherelationshipbetweencitationF1andhal- -grainedquotes,whichsubsequentlyguidethegen- lucination: Models positioned closer to the top-right -eration process. These quotes can provide fine- cornerexhibithighercitationqualityandalowerdegree -ofhallucination. -grainedsupervisionsignalsforattributedtextgen- -G3, -| eration. | To evaluate | | the effectiveness | | of | we | | | | | | | | -| --------------------------------------- | ----------- | --- | ----------------- | --------- | ---------- | ----- | ------- | -------- | ----------- | --- | --------- | --- | ------ | -| conduct | an ablation | | study | comparing | it against | | | | | | | | | -| | | | | | | | eration | process. | Experiments | | conducted | | on the | -| twovariantswithdistincttrainingrecipes. | | | | | | Given | | | | | | | | -ELI5datasetusingtheLLaMA-2-7Bmodelshow -that FRONTconsistsoftwostages,werefertothe that SELF-GUIDE outperforms PROMPT-GUIDE. -modeltrainedonlyduringthefirststage(without -ResultsdepictedinFigure5indicatethattraining -| consistency-awarealignment)asSELF-GUIDE. | | | | | | We | | | | | | | | -| ---------------------------------------- | ---------- | --- | ------- | --- | ----------- | --- | ---------- | --- | ------------- | --------- | -------- | -------------- | ------ | -| | | | | | | | models | to | self-generate | | grounded | quotes | before | -| firstcompare | SELF-GUIDE | | against | | VANILLA-SFT | | | | | | | | | -| | | | | | | | generating | | attributed | responses | is | more effective | | -(w/oGround),whichistrainedtodirectlygenerate thansimplyincorporatingthesegroundedquotes -responseswithcitations,bypassingthegrounding -intotheprompt. -| step. The | ablation | study, | detailed | | in Table | 2, re- | | | | | | | | -| --------- | -------- | ------ | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | -veals that models incorporating grounding guid- EffectsofConsistency-AwareAlignment(CCA). -ancesignificantlyoutperformtheir VANILLA-SFT The primary goal of CCA is to enhance the con- -counterpartsthatlacksuchgroundingmechanisms. sistency between grounded quotes and attributed -This highlights the crucial role of grounding in answers, thereby alleviating hallucinations and -| | | | | | | | achievingmorepreciseattribution. | | | | | Toevaluatethis, | | -| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | -bolsteringattribution. -Moreover, we explore an alternative variant we compare models that underwent only the G3 -of grounding guidance. Considering that SELF- stage(SELF-GUIDE)withthosefurtherenhanced -GUIDE leverages the model itself to both select through the CCA stage (FRONT). As illustrated -grounded quotes and generate attributed answers inTable2,FRONTsignificantlyimprovescitation -in an end-to-end paradigm, a natural variant in- qualityoverSELF-GUIDE,demonstratingtheeffec- -volves breaking down this task into two distinct tivenessoftheCCAstageinenhancingattribution. -stages. Inthisvariant,ChatGPTistaskedwithex- Furthermore,toassessCCA’simpactonreduc- -tractinggroundedquotes. Subsequently,aseparate inghallucinations,weutilizeQAFactEval(Fabbri -modelistrainedtoutilizethesegroundedquotes, etal.,2022),awidelyusedmetricforfactualcon- -along with the query and retrieval documents, to sistency, which scores the consistency of model -directly output the response and citations. This responses to given documents on a scale from 0 -variant,referredtoasPROMPT-GUIDED,integrates to5,withhigherscoresindicatinggreaterfaithful- -groundedquotesintotheprompttoguidethegen- ness. Specifically,weanalyzetheperformanceof - -leading open-source models and two variants of Authenticity Helpfulness -| FRONTand | SELF-GUIDE | | ontheELI5dataset. | | As | | | | | | | -| -------- | ---------- | --- | ----------------- | --- | --- | ------- | --- | --- | --- | ---- | ---- | -| | | | | | | ChatGPT | | | | 0.93 | 4.08 | -showninFigure6, FRONTproducesmorefaithful FRONT-7BonASQA 0.94 3.86 -outputsthanSELF-GUIDE,significantlyreducing FRONT-7BonELI5 0.92 3.96 -| hallucinations. | | | | | | FRONT-7BonQAMPARI | | | | 0.86 | 3.62 | -| --------------- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | ---- | ---- | -Table3: Humanevaluationonthequalityofgrounded -| EffectsofTrainingDataScale. | | | | Weanalyzethe | | | | | | | | -| --------------------------- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | -quotes. -| impact | of the | data scale | on model | performance | | | | | | | | -| ------ | ------------ | ---------- | -------- | ----------- | ------- | --- | --- | --- | --- | --- | --- | -| across | two training | stages. | In | particular, | we ran- | | | | | | | -tors. Theresultsofthehumanevaluationindicate -domlysampled2k,4k,6k,and8kinstancesfrom -thatbothquotesextractedbyChatGPTandthose -| our full | training | data | across two | distinct | training | | | | | | | -| -------- | -------- | ---- | ---------- | -------- | -------- | --- | --- | --- | --- | --- | --- | -stages. These subsets were then utilized to fine- generated by FRONT are of high quality, further -tunevarious7Bmodelvariants,enablingacompar- substantiatingtheeffectivenessofourmethod. -ativeanalysisofperformancebasedondatascale. -| Results | are shown | in | Figure | 7, which | indicates | | | | | | | -| ------- | --------- | --- | ------ | -------- | --------- | --- | --- | --- | --- | --- | --- | -7 Conclusion -thatincreasingdatasizeshowssignificantenhance- -mentsincitationquality,indicatingapositivecor- Inthiswork,wepresent FRONT,atwo-stagetrain- -relationbetweendatasizeandmodelperformance. ingframeworkdesignedtoequipLLMswithfine- -AsFRONTimplementsanautomatedprocedureca- -| | | | | | | grained | attribution | capabilities. | | FRONT | enables | -| --- | --- | --- | --- | --- | --- | ------- | ----------- | ------------- | --- | ----- | ------- | -pableofgeneratinghigh-qualityattributeddataand LLMstoinitiallyselectsupportingquotes,which -constructingcontrastivesupervisionfromweakand thenguidethegenerationprocess. Byfurtheren- -strongLLMs,itholdsthepotentialforcontinuous -| | | | | | | hancing | the consistency | | between | | the grounding | -| --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------- | --- | ------------- | -performanceimprovements. andgenerationprocessviapreferenceoptimization, -thesesupportingquotescanserveasfine-grained -6 HumanEvaluation -| | | | | | | citations. | Through | comprehensive | | | experiments, | -| --- | --- | --- | --- | --- | --- | ---------- | ------- | ------------- | --- | --- | ------------ | -Given the significant impact of the quality of FRONThasdemonstrateditsabilitytogeneratesu- -| | | | | | | perior grounded | | responses | | and highly | supportive | -| -------- | ------ | --- | ------------ | ------------ | --- | --------------- | ------------------------ | --------- | --- | ---------- | ---------- | -| grounded | quotes | on | fine-grained | verification | for | | | | | | | -| | | | | | | citations. | Furtheranalysisshowsthat | | | | FRONT sig- | -users,weconductedahumanevaluationtoassess -thequalityofgroundedquotesatdifferentstagesof nificantlyreduceshallucinationsandbenefitsuser -verification. -| ourframework: | | (1)QuotesextractedbyChatGPT | | | | | | | | | | -| ---------------------------------- | --------- | --------------------------- | ----------- | --- | ------------ | ------------ | --- | --- | --- | --- | --- | -| from50sampleddatapointsduringtheG3 | | | | | stage. | | | | | | | -| (2) Quotes | generated | | by FRONT-7B | | across three | 8 Limitation | | | | | | -datasets,with50datapointssampledfromeach. -Weengagedfourannotators,eachwithrelevant Our study presents several limitations worth not- -expertiseandholdingatleastabachelor’sdegree. ing. Firstly, the validation of our framework is -Thequalityofquoteswasevaluatedontwodimen- predominantly conducted on models of sizes 7B -sions: authenticityandhelpfulness. Authenticity and13B,leavingtheexplorationoflargermodels, -(abinaryscaleof0/1)referstowhetherthequotes suchasLLaMA-270Bduetocomputationalcon- -genuinelyoriginatefromthecorrespondingdocu- straints. Secondly,ourframeworkreliesonaprior -ments(quotesthatarehallucinatedormismatched retrievalprocess,whereinrelevantdocumentsare -withthecorrespondingdocumentIDareconsidered retrieved at one time. The incorporation of adap- -inauthentic). Helpfulness (5-point Likert scale) tiveretrieval,enablingmoredynamicinteractions -referstothedegreetowhichthequotesarebenefi- withLLMs,couldpotentiallyenhanceperformance. -cialinaddressingthequery. TheresultsinTable3 Weleaveitforfutureresearch. Lastly,evaluating -representtheaveragescoresforallquoteswithin the correctness of long-form question answering -eachmodelresponse,withtwoannotatorsevaluat- presents inherent challenges, leading our frame- -ingeachresponsetoensurereliability. work to primarily enhance citation quality, with -Furthermore, to evaluate the consistency of modest advancements in correctness. Therefore, -quotequalityannotations,wecomputedtheinter- we advocate for the development of more robust -annotator agreement using Fleiss’ Kappa coeffi- metricscapableofaccuratelyassessingthecorrect- -cient. The obtained Kappa coefficient of 0.82 in- nessoflong-formQAresponses, pavingtheway -| dicates | a high | level of | agreement | among | annota- | forfuturework. | | | | | | -| ------- | ------ | -------- | --------- | ----- | ------- | -------------- | --- | --- | --- | --- | --- | - -Acknowledgements -qa-basedfactualconsistencyevaluationforsumma- -| | | | | | | | rization. | InProceedingsofthe2022Conferenceof | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | ---------------------------------- | --- | --- | --- | -We appreciate Yuchun Fan for providing valu- theNorthAmericanChapteroftheAssociationfor -| | | | | | | | ComputationalLinguistics: | | | HumanLanguageTech- | | -| ----------------- | --- | --------- | --- | --- | ------- | -------- | ------------------------- | --- | --- | ------------------ | --- | -| able suggestions. | | Xiaocheng | | | Feng is | the cor- | | | | | | -nologies,NAACL2022,Seattle,WA,UnitedStates, -| responding | author | of | this | work. | We | thank the | | | | | | -| ---------- | ------ | --- | ---- | ----- | --- | --------- | --- | --- | --- | --- | --- | -July10-15,2022,pages2587–2601.Associationfor -| anonymous | reviewers | | for | their | insightful | com- | | | | | | -| --------- | --------- | --- | --- | ----- | ---------- | ---- | --- | --- | --- | --- | --- | -ComputationalLinguistics. -| ments. | This | work | was | supported | by | the Na- | | | | | | -| ------ | ---- | ---- | --- | --------- | --- | ------- | --- | --- | --- | --- | --- | -AngelaFan,YacineJernite,EthanPerez,DavidGrang- -| tional Key | R&D | Program | | of China | via | grant No. | | | | | | -| -------------- | --- | ------- | -------- | -------- | --------- | --------- | ---------- | -------- | ----------- | -------------- | ----- | -| | | | | | | | ier, Jason | Weston, | and Michael | Auli. 2019. | ELI5: | -| 2021ZD0112905, | | the | National | | Natural | Science | | | | | | -| | | | | | | | long form | question | answering. | In Proceedings | of | -| Foundation | of | China | (NSFC) | (grant | 62276078, | | | | | | | -the57thConferenceoftheAssociationforCompu- -U22B2059), the Key R&D Program of Hei- tationalLinguistics,ACL2019,Florence,Italy,July -longjiang via grant 2022ZX01A32, the Interna- 28-August2,2019,Volume1: LongPapers,pages -3558–3567.AssociationforComputationalLinguis- -tionalCooperationProjectofPCL,PCL2022D01 -tics. -andtheFundamentalResearchFundsfortheCen- -tralUniversities(GrantNo.HIT.OCEF.2023018). ZhangyinFeng,WeitaoMa,WeijiangYu,LeiHuang, -HaotianWang,QianglongChen,WeihuaPeng,Xi- -| | | | | | | | aochengFeng,BingQin,andTingLiu.2023. | | | | Trends | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | ------ | -inintegrationofknowledgeandlargelanguagemod- -References -els:Asurveyandtaxonomyofmethods,benchmarks, -CoRR,abs/2311.05876. -| Samuel | Joseph | Amouyal, | Ohad | Rubin, | | Ori Yoran, | andapplications. | | | | | -| -------- | --------- | -------- | --------- | ------- | -------------- | ---------- | ---------------- | ------ | ----------------- | -------- | --------- | -| Tomer | Wolfson, | Jonathan | | Herzig, | and | Jonathan | | | | | | -| | | | | | | | Luyu Gao, | Zhuyun | Dai, Panupong | Pasupat, | Anthony | -| Berant. | 2022. | QAMPARI: | | : | An open-domain | | | | | | | -| | | | | | | | Chen, | Arun | Tejasvi Chaganty, | Yicheng | Fan, Vin- | -| question | answering | | benchmark | | for questions | with | | | | | | -centY.Zhao,NiLao,HongraeLee,Da-ChengJuan, -| many | answers | from | multiple | paragraphs. | | CoRR, | | | | | | -| ---- | ------- | ---- | -------- | ----------- | --- | ----- | ---------- | ---- | ------ | ----------------- | --- | -| | | | | | | | and Kelvin | Guu. | 2023a. | RARR: researching | and | -abs/2205.12665. -revisingwhatlanguagemodelssay,usinglanguage -AkariAsai,ZeqiuWu,YizhongWang,AvirupSil,and models. InProceedingsofthe61stAnnualMeeting -oftheAssociationforComputationalLinguistics(Vol- -| Hannaneh | Hajishirzi. | | 2023. | Self-rag: | Learning | to | | | | | | -| -------- | ----------- | --- | ----- | --------- | -------- | --- | ----- | ----------------------------------- | --- | --- | --- | -| | | | | | | | ume1: | LongPapers),ACL2023,Toronto,Canada, | | | | -retrieve,generate,andcritiquethroughself-reflection. -July9-14,2023,pages16477–16508.Associationfor -CoRR,abs/2310.11511. -ComputationalLinguistics. -| Yuntao Bai, | Andy | Jones, | Kamal | Ndousse, | | Amanda | | | | | | -| ----------- | ---- | ------ | ----- | -------- | --- | ------ | --- | --- | --- | --- | --- | -TianyuGao,HowardYen,JiatongYu,andDanqiChen. -| Askell, | AnnaChen, | | NovaDasSarma, | | DawnDrain, | | | | | | | -| --------- | --------- | ---- | ------------- | --- | ---------- | --------- | ------------------ | ------------------------------------- | -------------------- | --- | --- | -| | | | | | | | 2023b. | Enablinglargelanguagemodelstogenerate | | | | -| Stanislav | Fort, | Deep | Ganguli, | | Tom | Henighan, | | | | | | -| | | | | | | | textwithcitations. | | CoRR,abs/2305.14627. | | | -NicholasJoseph,SauravKadavath,JacksonKernion, -TomConerly,SheerElShowk,NelsonElhage,Zac -| | | | | | | | Yunfan Gao, | Yun | Xiong, Xinyu | Gao, Kangxiang | Jia, | -| --------------- | --- | ----- | ---------- | --- | ------- | ----- | ----------- | --- | ------------ | -------------- | ---- | -| Hatfield-Dodds, | | Danny | Hernandez, | | Tristan | Hume, | | | | | | -JinliuPan,YuxiBi,YiDai,JiaweiSun,QianyuGuo, -ScottJohnston,ShaunaKravec,LianeLovitt,Neel Meng Wang, and Haofen Wang. 2023c. Retrieval- -| Nanda, | Catherine | Olsson, | | Dario | Amodei, | Tom B. | | | | | | -| -------- | --------- | ------- | --------------- | --------- | ------- | ---------- | ------------------------------------------ | -------------------- | --- | --- | --- | -| | | | | | | | augmentedgenerationforlargelanguagemodels: | | | | A | -| Brown, | Jack | Clark, | Sam McCandlish, | | Chris | Olah, | | | | | | -| | | | | | | | survey. | CoRR,abs/2312.10997. | | | | -| Benjamin | Mann, | and | Jared | Kaplan. | 2022. | Train- | | | | | | -| ing a | helpful | and | harmless | assistant | | with rein- | | | | | | -SylvainGugger,LysandreDebut,ThomasWolf,Philipp -forcement learning from human feedback. CoRR, Schmid,ZacharyMueller,SourabMangrulkar,Marc -abs/2204.05862. Sun,andBenjaminBossan.2022. Accelerate: Train- -ingandinferenceatscalemadesimple,efficientand -BerndBohnet,VinhQ.Tran,PatVerga,RoeeAharoni, -| | | | | | | | adaptable. | | https://github.com/huggingface/ | | | -| ------ | ------ | ----- | ------- | ------- | ----- | ------ | ---------- | --- | ------------------------------- | --- | --- | -| Daniel | Andor, | Livio | Baldini | Soares, | Jacob | Eisen- | | | | | | -accelerate. -stein,KuzmanGanchev,JonathanHerzig,KaiHui, -TomKwiatkowski,JiMa,JianmoNi,TalSchuster, OrHonovich, RoeeAharoni, JonathanHerzig, Hagai -WilliamW.Cohen,MichaelCollins,DipanjanDas, Taitelbaum,DoronKukliansy,VeredCohen,Thomas -Donald Metzler, Slav Petrov, and Kellie Webster. Scialom, Idan Szpektor, Avinatan Hassidim, and -| 2022. | Attributed | question | | answering: | | Evaluation | | | | | | -| ----- | ---------- | -------- | --- | ---------- | --- | ---------- | ----- | ------- | ----------- | ------------- | ------- | -| | | | | | | | Yossi | Matias. | 2022. TRUE: | re-evaluating | factual | -andmodelingforattributedlargelanguagemodels. -| | | | | | | | consistencyevaluation. | | InProceedingsofthe2022 | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | ---------------------- | --- | --- | -CoRR,abs/2212.08037. Conference of the North American Chapter of the -| | | | | | | | AssociationforComputationalLinguistics: | | | | Human | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | ----- | -CanyuChenandKaiShu.2023. Combatingmisinfor- LanguageTechnologies,NAACL2022,Seattle,WA, -mation in the age of llms: Opportunities and chal- UnitedStates, July10-15, 2022, pages3905–3920. -| lenges. | CoRR,abs/2311.05656. | | | | | | | | | | | -| ------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AssociationforComputationalLinguistics. -Alexander R. Fabbri, Chien-Sheng Wu, Wenhao Liu, LeiHuang,WeijiangYu,WeitaoMa,WeihongZhong, -and Caiming Xiong. 2022. Qafacteval: Improved Zhangyin Feng, Haotian Wang, Qianglong Chen, - -WeihuaPeng,XiaochengFeng,BingQin,andTing servingwithpagedattention. InProceedingsofthe -Liu. 2023. A survey on hallucination in large lan- ACMSIGOPS29thSymposiumonOperatingSystems -| guagemodels: | Principles,taxonomy,challenges,and | | | | | Principles. | | | | | | -| ---------------- | ---------------------------------- | ---- | -------- | -------- | --- | ----------- | ------------ | -------- | ----------------- | --- | ----- | -| openquestions. | CoRR,abs/2311.05232. | | | | | | | | | | | -| | | | | | | Patrick | S. H. Lewis, | Ethan | Perez, Aleksandra | | Pik- | -| | | | | | | tus, Fabio | Petroni, | Vladimir | Karpukhin, | | Naman | -| Ziwei Ji, Nayeon | Lee, | Rita | Frieske, | Tiezheng | Yu, | | | | | | | -DanSu,YanXu,EtsukoIshii,YejinBang,Andrea Goyal,HeinrichKüttler,MikeLewis,Wen-tauYih, -Madotto,andPascaleFung.2023. Surveyofhalluci- Tim Rocktäschel, Sebastian Riedel, and Douwe -nationinnaturallanguagegeneration. ACMComput. Kiela. 2020. Retrieval-augmented generation for -Surv.,55(12):248:1–248:38. knowledge-intensiveNLPtasks. InAdvancesinNeu- -| | | | | | | ralInformationProcessingSystems33: | | | | AnnualCon- | | -| --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ---------- | --- | -AlbertQ.Jiang,AlexandreSablayrolles,ArthurMen- -ferenceonNeuralInformationProcessingSystems -sch,ChrisBamford,DevendraSinghChaplot,Diego 2020,NeurIPS2020,December6-12,2020,virtual. -| de Las Casas, | Florian | Bressand, | | Gianna | Lengyel, | | | | | | | -| ------------- | ------- | --------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | -Guillaume Lample, Lucile Saulnier, Lélio Re- Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, -nard Lavaud, Marie-Anne Lachaux, Pierre Stock, Ziyang Chen, Baotian Hu, Aiguo Wu, and Min -TevenLeScao,ThibautLavril,ThomasWang,Timo- Zhang. 2023. A survey of large language models -| | | | | | | attribution. | CoRR,abs/2311.03731. | | | | | -| ----------------------------------- | --- | --- | --- | --- | ------- | ------------ | -------------------- | --- | --- | --- | --- | -| théeLacroix,andWilliamElSayed.2023. | | | | | Mistral | | | | | | | -7b. CoRR,abs/2310.06825. -| | | | | | | Nelson F. | Liu, Tianyi | Zhang, | and Percy | Liang. | 2023. | -| --- | --- | --- | --- | --- | --- | --------- | ----------- | ------ | --------- | ------ | ----- | -Albert Q. Jiang, Alexandre Sablayrolles, Antoine Evaluatingverifiabilityingenerativesearchengines. -Roux,ArthurMensch,BlancheSavary,ChrisBam- InFindingsoftheAssociationforComputationalLin- -ford,DevendraSinghChaplot,DiegodeLasCasas, guistics: EMNLP2023,Singapore,December6-10, -Emma Bou Hanna, Florian Bressand, Gianna 2023,pages7001–7025.AssociationforComputa- -| Lengyel, | Guillaume | Bour, | Guillaume | | Lample, | tionalLinguistics. | | | | | | -| ------------ | --------- | ------ | --------- | --------- | ------- | ------------------ | --- | --- | --- | --- | --- | -| Lélio Renard | Lavaud, | Lucile | | Saulnier, | Marie- | | | | | | | -AnneLachaux,PierreStock,SandeepSubramanian, JacobMenick,MajaTrebacz,VladimirMikulik,John -Sophia Yang, Szymon Antoniak, Teven Le Scao, Aslanides, H. Francis Song, Martin J. Chadwick, -Théophile Gervet, Thibaut Lavril, Thomas Wang, Mia Glaese, Susannah Young, Lucy Campbell- -| | | | | | | Gillingham, | Geoffrey | | Irving, and Nat | McAleese. | | -| --------------------------------------- | --- | --- | --- | --- | ---- | ----------- | -------- | --- | --------------- | --------- | --- | -| TimothéeLacroix,andWilliamElSayed.2024. | | | | | Mix- | | | | | | | -2022. Teachinglanguagemodelstosupportanswers -| tralofexperts. | CoRR,abs/2401.04088. | | | | | | | | | | | -| --------------- | -------------------- | ------- | ----- | ------ | ------ | ------------------- | --- | -------------------- | --- | --- | --- | -| | | | | | | withverifiedquotes. | | CoRR,abs/2203.11147. | | | | -| Ehsan Kamalloo, | Aref | Jafari, | Xinyu | Zhang, | Nandan | | | | | | | -Thakur,andJimmyLin.2023. HAGRID:Ahuman- ReiichiroNakano,JacobHilton,SuchirBalaji,JeffWu, -llmcollaborativedatasetforgenerativeinformation- Long Ouyang, Christina Kim, Christopher Hesse, -seekingwithattribution. CoRR,abs/2307.16883. ShantanuJain,VineetKosaraju,WilliamSaunders, -| | | | | | | Xu Jiang, | Karl | Cobbe, | Tyna Eloundou, | Gretchen | | -| --- | --- | --- | --- | --- | --- | --------- | ---- | ------ | -------------- | -------- | --- | -Krueger,KevinButton,MatthewKnight,Benjamin -VladimirKarpukhin,BarlasOguz,SewonMin,Patrick -S.H.Lewis,LedellWu,SergeyEdunov,DanqiChen, Chess,andJohnSchulman.2021. Webgpt: Browser- -andWen-tauYih.2020. Densepassageretrievalfor assisted question-answering with human feedback. -open-domainquestionanswering. InProceedingsof CoRR,abs/2112.09332. -the2020ConferenceonEmpiricalMethodsinNat- -| | | | | | | Jianmo | Ni, Chen | Qu, Jing | Lu, Zhuyun | Dai, | Gus- | -| ------------- | ----------- | --- | ----- | ----- | ------- | ------ | -------- | -------- | ---------- | ---- | ---- | -| ural Language | Processing, | | EMNLP | 2020, | Online, | | | | | | | -November16-20,2020,pages6769–6781.Associa- tavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, -tionforComputationalLinguistics. YiLuan,KeithB.Hall,Ming-WeiChang,andYinfei -| | | | | | | Yang.2022. | Largedualencodersaregeneralizable | | | | | -| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------- | --- | --- | --- | --- | -SayaliKulkarni,SheideChammas,WanZhu,FeiSha, retrievers. InProceedingsofthe2022Conferenceon -and Eugene Ie. 2020. Aquamuse: Automatically EmpiricalMethodsinNaturalLanguageProcessing, -generatingdatasetsforquery-basedmulti-document EMNLP2022,AbuDhabi,UnitedArabEmirates,De- -summarization. CoRR,abs/2010.12694. cember7-11,2022,pages9844–9855.Association -forComputationalLinguistics. -| TomKwiatkowski, | JennimariaPalomaki, | | | | OliviaRed- | | | | | | | -| --------------- | ------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | -field,MichaelCollins,AnkurP.Parikh,ChrisAlberti, OpenAI. 2023. GPT-4 technical report. CoRR, -| DanielleEpstein,IlliaPolosukhin,JacobDevlin,Ken- | | | | | | abs/2303.08774. | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -tonLee,KristinaToutanova,LlionJones,Matthew -LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, -| Kelcey, Ming-Wei | | Chang, | Andrew | M. | Dai, Jakob | | | | | | | -| ---------------- | -------- | -------- | ------- | ----- | ---------- | ------- | -------------- | --- | --------------- | --- | ----- | -| | | | | | | Carroll | L. Wainwright, | | Pamela Mishkin, | | Chong | -| Uszkoreit, | Quoc Le, | and Slav | Petrov. | 2019. | Natu- | | | | | | | -ralquestions: abenchmarkforquestionanswering Zhang,SandhiniAgarwal,KatarinaSlama,AlexRay, -research. Trans.Assoc.Comput.Linguistics,7:452– JohnSchulman,JacobHilton,FraserKelton,Luke -| 466. | | | | | | Miller,MaddieSimens,AmandaAskell,PeterWelin- | | | | | | -| ---- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | -der,PaulF.Christiano,JanLeike,andRyanLowe. -Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying 2022. Training languagemodelsto followinstruc- -Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. tionswithhumanfeedback. InAdvancesinNeural -Gonzalez, Hao Zhang, and Ion Stoica. 2023. Effi- InformationProcessingSystems35: AnnualConfer- -cientmemorymanagementforlargelanguagemodel enceonNeuralInformationProcessingSystems2022, - -NeurIPS2022,NewOrleans,LA,USA,November28 Doshi,RenelitoDelosSantos,TojuDuke,JohnnySo- -| -December9,2022. | | | | | | | raker,BenZevenbergen,VinodkumarPrabhakaran, | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | ----------- | --- | ------- | ----------- | -| | | | | | | | Mark Diaz, | Ben | Hutchinson, | | Kristen | Olson, Ale- | -AleksandraPiktus,FabioPetroni,VladimirKarpukhin, jandraMolina,ErinHoffman-John,JoshLee,Lora -DmytroOkhonko,SamuelBroscheit,GautierIzacard, -| | | | | | | | Aroyo, | Ravi Rajakumar, | | Alena | Butryna, | Matthew | -| --- | --- | --- | --- | --- | --- | --- | ------ | --------------- | --- | ----- | -------- | ------- | -Patrick S. H. Lewis, Barlas Oguz, Edouard Grave, Lamm,ViktoriyaKuzmina,JoeFenton,AaronCo- -Wen-tauYih,andSebastianRiedel.2021. Theweb hen,RachelBernstein,RayKurzweil,BlaiseAgüera -isyouroyster-knowledge-intensiveNLPagainsta yArcas,ClaireCui,MarianCroak,EdH.Chi,and -verylargewebcorpus. CoRR,abs/2112.09924. QuocLe.2022. Lamda: Languagemodelsfordialog -CoRR,abs/2201.08239. -applications. -KrishnaPillutla,SwabhaSwayamdipta,RowanZellers, -JohnThickstun,SeanWelleck,YejinChoi,andZaïd -| | | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | -Harchaoui.2021. MAUVE:measuringthegapbe- bert, Amjad Almahairi, Yasmine Babaei, Nikolay -tweenneuraltextandhumantextusingdivergence Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti -| frontiers. | In | Advances | in Neural | Information | | Pro- | | | | | | | -| ---------- | --- | -------- | --------- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | -Bhosale,DanBikel,LukasBlecher,CristianCanton- -| cessingSystems34: | | AnnualConferenceonNeural | | | | | | | | | | | -| ----------------- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Ferrer,MoyaChen,GuillemCucurull,DavidEsiobu, -InformationProcessingSystems2021,NeurIPS2021, -JudeFernandes,JeremyFu,WenyinFu,BrianFuller, -December6-14,2021,virtual,pages4816–4828. -CynthiaGao,VedanujGoswami,NamanGoyal,An- -thonyHartshorn,SagharHosseini,RuiHou,Hakan -RonakPradeep,SahelSharifymoghaddam,andJimmy Inan,MarcinKardas,ViktorKerkez,MadianKhabsa, -| Lin.2023. | Rankvicuna: | | Zero-shotlistwisedocument | | | | | | | | | | -| --------- | ----------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -IsabelKloumann,ArtemKorenev,PunitSinghKoura, -rerankingwithopen-sourcelargelanguagemodels. -Marie-AnneLachaux,ThibautLavril,JenyaLee,Di- -CoRR,abs/2309.15088. -anaLiskovich,YinghaiLu,YuningMao,XavierMar- -tinet,TodorMihaylov,PushkarMishra,IgorMoly- -| Rafael Rafailov, | | Archit | Sharma, | Eric | Mitchell, | Ste- | | | | | | | -| ---------------- | --- | ------ | ------- | ---- | --------- | ---- | ---------- | ---- | ------ | -------- | ------ | ------- | -| | | | | | | | bog, Yixin | Nie, | Andrew | Poulton, | Jeremy | Reizen- | -fanoErmon,ChristopherD.Manning,andChelsea -stein,RashiRungta,KalyanSaladi,AlanSchelten, -| Finn. | 2023. | Direct preference | | optimization: | | Your | | | | | | | -| ------------------------------------ | ----- | ----------------- | --- | ------------- | --- | ----- | -------------- | ----- | ------- | ------ | ------ | --------- | -| | | | | | | | Ruan Silva, | Eric | Michael | Smith, | Ranjan | Subrama- | -| languagemodelissecretlyarewardmodel. | | | | | | CoRR, | | | | | | | -| | | | | | | | nian, Xiaoqing | Ellen | Tan, | Binh | Tang, | Ross Tay- | -abs/2305.18290. -| | | | | | | | lor, Adina | Williams, | Jian | Xiang | Kuan, | Puxin Xu, | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | ---- | ----- | ----- | --------- | -ZhengYan,IliyanZarov,YuchenZhang,AngelaFan, -| Jeff Rasley, | Samyam | Rajbhandari, | | Olatunji | | Ruwase, | | | | | | | -| ------------ | ------ | ------------ | --- | -------- | --- | ------- | --- | --- | --- | --- | --- | --- | -and Yuxiong He. 2020. Deepspeed: System opti- Melanie Kambadur, Sharan Narang, Aurélien Ro- -driguez,RobertStojnic,SergeyEdunov,andThomas -mizationsenabletrainingdeeplearningmodelswith -| | | | | | | | Scialom.2023. | Llama2: | | Openfoundationandfine- | | | -| ------------------------- | --- | --- | --- | --------- | ------- | --- | ---------------- | ------- | -------------------- | ---------------------- | --- | --- | -| over100billionparameters. | | | | InKDD’20: | The26th | | | | | | | | -| | | | | | | | tunedchatmodels. | | CoRR,abs/2307.09288. | | | | -ACMSIGKDDConferenceonKnowledgeDiscovery -| and Data | Mining, | Virtual | Event, | CA, | USA, | August | | | | | | | -| -------- | ------- | ------- | ------ | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | -23-27,2020,pages3505–3506.ACM. TheodoraWorledge,JudyHanwenShen,NicoleMeis- -| | | | | | | | ter,CalebWinston,andCarlosGuestrin.2023. | | | | | Uni- | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ---- | -fyingcorroborativeandcontributiveattributionsin -| Freda Shi, | Xinyun | Chen, | Kanishka | Misra, | | Nathan | | | | | | | -| ---------- | ------ | ----- | -------- | ------ | --- | ------ | -------------------- | --- | --- | -------------------- | --- | --- | -| | | | | | | | largelanguagemodels. | | | CoRR,abs/2311.12233. | | | -Scales,DavidDohan,EdH.Chi,NathanaelSchärli, -| andDennyZhou.2023. | | | Largelanguagemodelscan | | | | | | | | | | -| ------------------ | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -beeasilydistractedbyirrelevantcontext. InInterna- Fangyuan Xu, Weijia Shi, and Eunsol Choi. 2023a. -tionalConferenceonMachineLearning,ICML2023, RECOMP:improvingretrieval-augmentedlmswith -23-29 July 2023, Honolulu, Hawaii, USA, volume compression and selective augmentation. CoRR, -abs/2310.04408. -202ofProceedingsofMachineLearningResearch, -pages31210–31227.PMLR. -ShichengXu,LiangPang,HuaweiShen,XueqiCheng, -IvanStelmakh,YiLuan,BhuwanDhingra,andMing- andTat-SengChua.2023b. Search-in-the-chain: To- -Wei Chang. 2022. ASQA: factoid questions meet wards the accurate, credible and traceable content -long-formanswers. InProceedingsofthe2022Con- generation for complex knowledge-intensive tasks. -CoRR,abs/2304.14732. -ferenceonEmpiricalMethodsinNaturalLanguage -Processing,EMNLP2022,AbuDhabi,UnitedArab -Emirates,December7-11,2022,pages8273–8288. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan -AssociationforComputationalLinguistics. Berant. 2023. Making retrieval-augmented lan- -| | | | | | | | guage models | robust | to | irrelevant | context. | CoRR, | -| ----- | ---------- | ------ | ------------- | -------- | -------- | ----- | --------------- | ------ | --- | ---------- | -------- | ----- | -| Romal | Thoppilan, | Daniel | De | Freitas, | Jamie | Hall, | abs/2310.01558. | | | | | | -| Noam | Shazeer, | Apoorv | Kulshreshtha, | | Heng-Tze | | | | | | | | -Cheng,AliciaJin,TaylorBos,LeslieBaker,YuDu, Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, -YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Xiaolei Wang, Yupeng Hou, Yingqian Min, Be- -AminGhafouri,MarceloMenegali,YanpingHuang, ichenZhang,JunjieZhang,ZicanDong,YifanDu, -MaximKrikun,DmitryLepikhin,JamesQin,Dehao Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao -Chen,YuanzhongXu,ZhifengChen,AdamRoberts, Jiang,RuiyangRen,YifanLi,XinyuTang,Zikang -MaartenBosma,YanqiZhou,Chung-ChingChang, Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. -IgorKrivokon,WillRusch,MarcPickett,KathleenS. 2023. A survey of large language models. CoRR, -| Meier-Hellstern, | | Meredith | Ringel | Morris, | | Tulsee | abs/2303.18223. | | | | | | -| ---------------- | --- | -------- | ------ | ------- | --- | ------ | --------------- | --- | --- | --- | --- | --- | - -Kun Zhu, Xiaocheng Feng, Xiyuan Du, Yuxuan Gu, process, we retained 5,667 and 2,431 queries, re- -WeijiangYu,HaotianWang,QianglongChen,Zheng spectively. Additionally,wecalculatedtheaverage -| Chu, JingchangChen, | | andBingQin.2024. | | Anin- | | | | | | | -| ------------------- | --- | ---------------- | --- | ----- | --- | --- | --- | --- | --- | --- | -lengthofanswersandtheaveragenumberofcita- -formationbottleneckperspectiveforeffectivenoise -tionsgeneratedforvarioustypesofquerieswithin -filteringonretrieval-augmentedgeneration. -ourdataset,asshowninTable4. -YutaoZhu,HuayingYuan,ShutingWang,JiongnanLiu, -| Wenhan | Liu, Chenlong | Deng, | Zhicheng | Dou, and | | | | | | | -| ------ | ------------- | ----- | -------- | -------- | --- | --- | --- | --- | --- | --- | -A.2 DetailsofDataFiltering -| Ji-RongWen.2023. | | Largelanguagemodelsforinfor- | | | | | | | | | -| ---------------- | --- | ---------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | -| mationretrieval: | | Asurvey. | CoRR,abs/2308.07107. | | | | | | | | -WetrainedourAttributedDiscriminatorusingthe -| | | | | | manually | annotated | data | provided | by | Liu et al. | -| --- | --- | --- | --- | --- | -------- | --------- | ---- | -------- | --- | ---------- | -A DetailsofDataGenerationPipeline -| | | | | | (2023), | which is | sampled | from | real generative | | -| --- | --- | --- | --- | --- | ------- | -------- | ------- | ---- | --------------- | --- | -A.1 DataStatistic searchengines. Eachstatementanditsciteddocu- -menthavebeenmeticulouslyannotatedforattribu- -#Questions 8,098 tion,categorizedintothreetypes: completesupport, -| | | | | | partialsupport,andnosupport. | | | | Fortraining,weuti- | | -| ------------ | --- | --- | --- | ---- | ---------------------------- | --- | --- | --- | ------------------ | --- | -| ➥#LongAnswer | | | | 5667 | | | | | | | -lizedadatasetof8,834instances,comprising6,415 -| ➥#ShortAnswer | | | | 2431 | | | | | | | -| ------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | -instancesofcompletesupport,1,552ofpartialsup- -Avg. WordsperAnswer 50.48 port, and 867 of no support. The discriminator -initializedwithLLaMA-2-7B,wastrainedwitha -| ➥Avg. | WordsperLongAnswer | | | 69.15 | | | | | | | -| ----- | ------------------- | --- | --- | ----- | --------------------------- | --- | --- | --- | -------------- | --- | -| | | | | | maximumsequencelengthof512. | | | | Wetraineditfor | | -| ➥Avg. | WordsperShortAnswer | | | 6.94 | | | | | | | -3epochs,withatotalbatchsizeof128,andapeak -Avg. CitationperAnswer 4.40 learning rate of 2e-5, incorporating 3% warmup -steps,followedbyalineardecay. -| ➥Avg. | CitationperLongAnswer | | | 4.68 | | | | | | | -| ----- | ---------------------- | --- | --- | ---- | ------ | ----------------- | --------- | --------- | ---------- | ----- | -| ➥Avg. | | | | | During | the data | filtering | stage, | we first | break | -| | CitationperShortAnswer | | | 3.77 | | | | | | | -| | | | | | down | the automatically | | generated | attributed | an- | -swersintostatementformandusethetraineddis- -| Table 4: | The statistics | of the | data generated | by our | | | | | | | -| -------- | -------------- | ------ | -------------- | ------ | --- | --- | --- | --- | --- | --- | -criminatortoannotatetheattributionbetweeneach -automaticdatagenerationpipeline. -| | | | | | statement | and its | cited | documents. | Specifically, | | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ---------- | ------------- | --- | -weassigndifferentattributionscorestoeachstate- -Table4presentsthestatisticsofthedataautomat- -| | | | | | ment s | based on | its attribution | | relationship | with | -| ------------------------------------------- | --- | --- | --- | --- | ----------------------------------- | -------- | --------------- | --- | ------------ | ---- | -| icallygeneratedbyourdatagenerationpipeline. | | | | In | | | | | | | -| | | | | | citeddocumentsd,asshowninEquation7. | | | | | Con- | -total,wecollected8,098questionsfromtheNatural -sequently,foreachattributedanswer,wecancal- -Questions(NQ)dataset,ofwhich5,667questions -| | | | | | culateitsaverageattributionscore. | | | | Attributedan- | | -| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ------------- | --- | -weregatheredfromthosewithlong-formanswers, -swerswithanaverageattributionscorebelow0.8 -and2,431questionswerecollectedfromthosewith -| | | | | | are filtered | out. | The threshold | | of 0.8 was | deter- | -| --- | --- | --- | --- | --- | ------------ | ---- | ------------- | --- | ---------- | ------ | -short-formfactoidanswers. -minedthroughpreliminarytestingonthedevelop- -Forquestionsrequiringlong-formanswers,we -| | | | | | ment set, | for which | we | manually | annotated | 100 | -| --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --------- | --- | -initializedourquerysourcewiththeAQUAMUSE -samplestoensuretheeffectivenessofourfiltering -| dataset (Kulkarni | | et al., 2020), | which consists | of | | | | | | | -| ----------------- | --- | -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | -criteria. -high-qualityqueriesspecificallydesignedforlong- -formresponseswithintheNQdataset,recognized -| as“good”bythemajorityofNQevaluators. | | | | Inthis | | | | | | | -| ------------------------------------ | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | - -| | | | | | | 1, Dis(s,d) | | = complete | support | | -| -------------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | -| way,utilizingarefinedandsuperiorqualityquery | | | | | |   | | | | | -setlaidarobustgroundworkforourtrainingdata r(s)= 0.5,Dis(s,d) = partial support - -generation,streamliningthedatafilteringprocess. 0, Dis(s,d) = no support -| Forfactoidqueriesthatnecessitateshort-forman- | | | | | | | | | | (7) | -| --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -swers,wedirectlysampledfromtheoriginalNQ -| dataset, leveraging | | its abundance | and | inherently | | | | | | | -| ------------------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | -B DetailsofEvaluationMetrics -highquality. -During the data generation process, our initial In addition to evaluating citation quality and cor- -querysetcomprised7,725queriesrequiringlong- rectness,theALCEbenchmarkincludesabroader -formanswersand4,000queriesnecessitatingshort- setofdimensions,suchasfluency,ROUGE-L,and -| form answers. | | After a two-stage | data | filtering | generationlength. | | | | | | -| ------------- | --- | ----------------- | ---- | --------- | ----------------- | --- | --- | --- | --- | --- | - -Fluency Weevaluatethefluencyofthegenerated framework(Kwonetal.,2023)forefficientinfer- -responseusingMAUVE(Pillutlaetal.,2021). No- ence. Thehyperparametersaresetasillustratedin -| tably,wecalculatefluencyonlyfortheASQAand | | | | | | | Table9. | | | | | | | -| ----------------------------------------- | --- | -------- | ------ | -------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | -| ELI5 datasets, | | omitting | it for | QAMPARI, | | as the | | | | | | | | -response in QAMPARI typically consists of lists D.2 RetrievalSettings -| ofshortanswers. | | ArelativelyhighMAUVEscore | | | | | | | | | | | | -| --------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Duringtheevaluation,weadoptthesameretrieval -indicatesthatthegenerationissufficientlyfluent. -| | | | | | | | settings | as specified | | by Gao | et | al. (2023b). | For | -| ------- | --------------------------------- | --- | --- | -------- | --- | ------ | -------------------------------- | ------------ | ------- | ------ | --------- | ------------ | ------- | -| | | | | | | | the ASQA | and | QAMPARI | | datasets, | we | use the | -| ROUGE-L | Inadditiontoevaluatingthecorrect- | | | | | | | | | | | | | -| | | | | | | | denseretrieverGTR(Nietal.,2022). | | | | | FortheELI5 | | -| ness of | the model-generated | | | content, | we | employ | | | | | | | | -dataset,weemploythesparseretrieverBM25. -ROUGE-Ltoassesstheoverallqualityandtextual -coherenceoftheresponses. -E MoredetailaboutAblationStudy -C Prompts -E.1 TheEffectofTrainingDataScale. -C.1 PromptsforPrompting-basedMethods Weexaminehowmodelperformancevarieswith -| | | | | | | | changesindatascale,asdepictedinFigure7. | | | | | | The | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -FollowingGaoetal.(2023b),weadoptthevanilla -| | | | | | | | upper part | of | the figure | illustrates | | the | impact of | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | ----------- | --- | --- | --------- | -promptingstrategyforitssimplicityandeffective- -| | | | | | | | the training | data | scale | on | citation | quality | during | -| ------------------------------------ | ------ | -------- | -------- | -------- | --------- | --------- | ------------- | ---- | ------- | ---------- | ---------- | -------- | --------- | -| ness. Specifically, | | the | prompts | vary | according | to | | | | | | | | -| | | | | | | | the Grounding | | Guided | Generation | | training | stage, | -| thetypeofdatawithintheALCEbenchmark. | | | | | | For | | | | | | | | -| | | | | | | | with datasets | | ASQA, | ELI5, | and | QAMPARI | rep- | -| long-form | QA | datasets | such | as ASQA | | and ELI5, | | | | | | | | -| | | | | | | | resented | from | left to | right. | Similarly, | | the lower | -| the prompt | format | is | detailed | in Table | 5. | For the | | | | | | | | -partofthefiguredescribestheinfluenceduringthe -| short-form | QA | dataset | QAMPARI, | | the | format is | | | | | | | | -| ---------- | --- | ------- | -------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -Consistency-AwareAlignmenttrainingstage. -outlinedinTable6. -E.2 TheGeneralizationAcrossModel -| C.2 Instructionsfor | | | FRONT | | | | | | | | | | | -| ------------------- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Architectures. -Duringthetrainingprocess,wefollowtheinstruc- -| | | | | | | | FRONT | demonstrates | | exceptional | | generalization | | -| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | --- | ----------- | --- | -------------- | --- | -Alpaca6. -| tion format | of | | Specifically, | | we | employ | | | | | | | | -| ----------- | --- | --- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -capabilitiesacrossvariousfoundationalmodelar- -variedinstructionsfordifferentquestiontypes,as -| | | | | | | | chitectures. | Specifically,transitioningthefounda- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------------------------ | --- | --- | --- | --- | --- | -delineatedinTable7forlong-formquestionsand -| | | | | | | | tional model | | from LLaMA-2-7B | | | to the | stronger | -| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --- | ------ | -------- | -Table8forshort-formquestions. -| | | | | | | | foundational | | model, | Mistral-7B, | | results | in even | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ----------- | ------ | ------------ | --- | ------- | -------- | -| | | | | | | | greater | performance | | enhancements | | as | shown in | -D ExperimentalDetails -| | | | | | | | Figure8. | Thisfurtherunderscoresthebroadappli- | | | | | | -| --------------------- | --- | --- | ----- | --- | --- | --- | ----------------------------- | ------------------------------------ | --- | --- | ------ | --- | --- | -| | | | | | | | cabilityandgeneralizabilityof | | | | FRONT. | | | -| D.1 TrainingDetailsof | | | FRONT | | | | | | | | | | | -Thetrainingofallmodelsisexecutedon4Nvidia E.3 Theeffectofβ inConsistency-Aware -A100GPUs,eachwith80GBofmemory,leverag- AlignmentTrainingStage -ingtheDeepspeed(Rasleyetal.,2020)andHug- -| | | | | | | | In the | Consistency-Aware | | | Alignment | | Training | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------- | --- | --- | --------- | --- | -------- | -gingFaceAcceleratelibraries(Guggeretal.,2022) -| | | | | | | | Stage, | the β | parameter | in | Direct | Preference | Op- | -| -------------------------------------- | --- | --- | --- | --- | --- | ----- | ---------- | ----- | --------- | --- | ------ | ---------- | ------ | -| toconductmulti-GPUdistributedtraining. | | | | | | Given | | | | | | | | -| | | | | | | | timization | (DPO) | controls | | the | strength | of the | -thelongnatureoftheinputs,themaximumtoken -Kullback-Leiblerpenalty,typicallysetwithinthe -lengthissetto2,048tokens. -| | | | | | | | range of | 0.1 to | 0.5. | A higher | β | value | indicates a | -| ------ | --- | --------- | ----- | ---------- | --- | ------ | -------- | ------ | ---- | -------- | --- | ----- | ----------- | -| During | the | grounding | guide | generation | | stage, | | | | | | | | -preferenceforthepolicymodel’strainingprocess -modelsaretrainedfor5epochswithatotalbatch -toremainclosertotheinitiallyreferencedmodel. -| size of | 128, a peak | learning | | rate of | 2e-5 | with 3% | | | | | | | | -| -------------------------------------- | ----------- | -------- | --- | ------- | ---- | ------- | ----------------------------------- | ------ | --- | --- | ----- | ------ | -------- | -| | | | | | | | In extreme | cases, | as | β → | 0, we | ignore | the con- | -| warmupstepsfollowedbyalineardecay. | | | | | | During | | | | | | | | -| | | | | | | | straintsimposedbythereferencemodel. | | | | | | Thisset- | -| thecontrastivealignmentstage,wesettheβ | | | | | | to0.1 | | | | | | | | -tingaimstobalancethemodel’sabilitytoadaptto -andcontinuedtrainingfortwoadditionalepochs. -newtrainingsignalswhilemaintainingthestability -| Specifically, | During | | inference, | we | use | the vllm | | | | | | | | -| ------------- | ------ | --- | ---------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -ofthelearnedbehaviorsfromthereferencemodel. -| | | | | | | | Subsequently, | | we | trained | five | variants | by ad- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | ------- | ---- | -------- | ------ | -6https://github.com/tatsu-lab/stanford_alpaca/ -| tree/main | | | | | | | justingβ | from0.1to0.5onthemodelpreviously | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --- | - -Instruction: Write an accurate, engaging, and concise answer for the given question using only -the provided search results (some of which might be irrelevant) and cite them properly. Use an -unbiased and journalistic tone. Always cite for any factual claim. When citing several search -results, use [1][2][3]. Cite at least one document and at most three documents in each sentence. -If multiple documents support the sentence, only cite a minimum sufficient subset of the -documents. -Table5: PromptforLong-formQA. -Instruction: Provide a list of accurate answers for the given question using only the provided -search results (some of which might be irrelevant) and cite them properly. Always cite one and -only one document for each answer. Separate answers by commas. For questions that have more than -| 5 answers, | write | at least 5 answers. | | | | | -| ---------- | ----- | ------------------- | --- | --- | --- | --- | -Table6: PromptforShort-formQA. -| 72.0 | | | 48.0 | | 23.0 | | -| ---- | --- | ----- | ---- | ----- | ---- | ----- | -| 68.0 | | | 46.0 | | 22.0 | | -| 64.0 | | | 44.0 | | 21.0 | | -| 60.0 | | REC. | 42.0 | REC. | 20.0 | REC. | -| 56.0 | | PREC. | 40.0 | PREC. | 19.0 | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| ---- | --- | -------- | ---- | -------- | ---- | -------- | -| 76.0 | | | 54.0 | | 25.0 | | -| 72.0 | | | 52.0 | | 24.0 | | -| 68.0 | | | 50.0 | | 23.0 | | -| 64.0 | | | 48.0 | | 22.0 | | -| | | REC. | | REC. | | REC. | -| 60.0 | | | 46.0 | | 21.0 | | -| | | PREC. | | PREC. | | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| --- | --- | -------- | --- | -------- | --- | -------- | -Figure7: Ablationstudyonsynthetictrainingdatasize: TheupperpartofthefigurecorrespondstotheGrounding -GuidedGenerationtrainingstage, whilethebottompartrepresentstheWeak-to-StrongContrastiveAlignment -trainingstage. Fromlefttoright,theresultsarepresentedforASQA,ELI5,andQAMPARI,respectively. REC. -indicatesCitationRecallandPREC. denotesCitationPrecision. Thex-axisrepresentsthequantityofautomatically -generateddata. Itisobservedthatasthevolumeofautomaticallygenerateddataincreases,thereisaconsistent -improvementinbothcitationrecallandprecisionacrossthetwotrainingstages. -trainedwithG3 toexploretheimpactofthehyper- adheretotheevaluationframeworkestablishedin -parameter β on attribution quality. We evaluated (Gaoetal.,2023b). Forlong-formQAdatasetslike -thesevariantsontheASQAandELI5datasets,and ASQAandELI5,wealsoreportmetricsrelatedto -theexperimentalresultsareshowninFigure9. Fluency,ROUGE-L,andaverageresponselength. -The experimental results indicate that as β in- Specifically,weuseMAUVE(Pillutlaetal.,2021) -creases, the model’s performance on attribution toevaluatethefluencyofthemodelresponse. For -graduallydecreases. Thisobservationsuggeststhat datasetslikeQAMPARI,whereanswersarecom- -thefirststageofG3 mightintroduceanoticeable posed of concatenated entities, we calculate the -inconsistencybetweengroundingandattribution. averagenumberofpredictedentities. -Withhigherβvalues,themodelstrugglestoescape -theconstraintsofinconsistentattributedanswers, -| leading to | a reduction | in attribution | quality as | β | | | -| ---------- | ----------- | -------------- | ---------- | --- | --- | --- | -increases. -F FullResults -Wepresentthecomprehensiveresultsofourexperi- -| mentsinTables10,11,and12. | | Beyondtheevalua- | | | | | -| ------------------------- | --- | ---------------- | --- | --- | --- | --- | -tionmetricsrelatedtoCorrectnessandCitation,we - -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -| guide | answer | generation | and | cite the | sources properly. | | | | | | | -| ------------------- | ------ | ---------- | --- | ---------- | ----------------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table7: | InstructionFormatforFRONTonLong-formQA. | | | | | | | | -| --- | --- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -provide a list of accurate answers for the given question. Always cite one and only one document -| for each | answer. | Separate | answers | by | commas. | | | | | | | -| ------------------- | ------- | ---------- | ------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table8: | InstructionFormatforFRONTonShort-formQA. | | | | | | | | -| ------- | ----------------------------------- | --- | ------- | ---------------------------------------- | --- | ------------- | --- | --- | ---------------- | --- | --- | -| | Hyper-parameters | | | Value | | | | | | | | -| | Top-p | | | 0.95 | | | | | | | | -| | Temperature | | | 0.2 | | | | | | | | -| | Max-length | | | 2048 | | | | | | | | -| Table9: | Hyper-parametersettingsininference. | | | | | | | | | | | -| | | | | | | | 72 | | 56 | | | -| | | | | | | )%(1FnoitatiC | 71 | | )%(1FnoitatiC 55 | | | -Mistral-7B -| | | | VANILLA-SFT(Mistral-7B) | | | | 70 | | 54 | | | -| --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | FRONT(Mistral-7B) | | | | 69 | | 53 | | | -60 -| | | | | | | | 68 | | 52 | | | -| ---------------- | --- | --- | --- | --- | --- | --- | -------------- | ----------- | --- | -------------- | ----------- | -| )%(1FnoitatiC 50 | | | | | | | 67 | FRONT-7B | 51 | | FRONT-7B | -| 40 | | | | | | | 0.1 0.2 | 0.3 0.4 0.5 | | 0.1 0.2 | 0.3 0.4 0.5 | -| 30 | | | | | | | βonASQAdataset | | | βonELI5dataset | | -20 -| | | | | | | Figure | 9: Ablation | on hyperparameter | | β | in Weak-to- | -| --- | --- | --- | --- | --- | --- | -------------------------------------------- | ----------- | ----------------- | --- | --- | ----------- | -| 10 | | | | | | StrongContrastiveAlignmentstageonASQAandELI5 | | | | | | -0 -| | ASQA | | ELI5 | QAMPARI | | | | | | | | -| --- | ---- | --- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Figure8:Ablationstudyonmodelarchitecture:Wesub- -stitutedthefoundationmodelinFRONTwithMistral-7B -andcomparedtheexperimentalresultsofmodelsunder -| the same | foundation | model | using | in-context | learning | | | | | | | -| -------- | ---------- | ----- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | -andthosedirectlysupervisedfine-tunedonourautomat- -| icallygenerateddata. | | Theexperimentsdemonstratethat | | | | | | | | | | -| -------------------- | --------- | ----------------------------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | -| by replacing | different | foundation | | models, | our frame- | | | | | | | -workstillmaintainsitsgeneralizability. - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | -------- | ---------- | ---------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (EMRec.) | Rec. Prec. | F1 ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 73.41 | 40.37 | 72.81 69.69 | 71.22 37.92 | 39.24 | -| ------------ | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 79.90 | 24.32 | 17.24 17.87 | 17.55 29.38 | 42.29 | -| LLaMA-2 | 13B | 87.08 | 27.99 | 16.45 19.04 | 17.65 31.41 | 39.25 | -| | 70B | 69.28 | 31.53 | 44.18 44.79 | 44.48 31.53 | 26.86 | -| | 7B | 66.78 | 29.93 | 55.99 51.66 | 53.74 32.93 | 26.18 | -| LLaMA-2-Chat | 13B | 66.14 | 34.39 | 37.15 38.17 | 37.65 35.13 | 33.68 | -| | 70B | 86.60 | 41.24 | 60.19 61.16 | 60.67 37.01 | 47.09 | -| | 7B | 86.92 | 38.34 | 48.37 44.63 | 46.42 35.95 | 63.90 | -Vicuna-v1.5 -| | 13B | 66.11 | 35.20 | 51.92 53.40 | 52.65 35.74 | 38.57 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.37 | 29.46 | 23.12 25.45 | 24.23 31.67 | 37.17 | -Mistral -| | 8×7B | 83.30 | 36.30 | 32.72 34.49 | 33.58 35.05 | 38.47 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.86 | 38.57 | 64.90 59.67 | 62.18 36.21 | 45.26 | -Mistral-Instruct -| | 8×7B | 94.77 | 44.11 | 61.80 63.27 | 62.53 38.54 | 58.83 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -Post-hocRetrieval -| ChatGPT | - | 49.78 | 37.68 | 27.11 27.05 | 27.08 36.64 | 52.61 | -| ------------ | ---- | ----- | ----- | ----------- | ----------- | ------ | -| | 7B | 75.56 | 16.55 | 13.88 13.86 | 13.87 26.81 | 37.50 | -| LLaMA-2 | 13B | 77.91 | 20.51 | 20.95 20.94 | 20.94 29.53 | 31.37 | -| | 70B | 75.23 | 27.58 | 28.43 28.43 | 28.43 30.33 | 29.88 | -| | 7B | 22.50 | 14.17 | 11.33 11.33 | 11.33 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 64.52 | 24.43 | 21.43 21.43 | 21.43 33.91 | 41.12 | -| | 70B | 70.63 | 29.68 | 24.51 24.51 | 24.51 34.17 | 45.74 | -| | 7B | 63.87 | 19.58 | 16.24 16.24 | 16.24 33.22 | 41.80 | -| Vicuna-v1.5 | 13B | 73.83 | 24.79 | 24.11 24.11 | 24.11 34.42 | 43.54 | -| | 7B | 86.54 | 21.17 | 16.78 16.77 | 16.77 30.90 | 42.43 | -| Mistral | 8×7B | 80.99 | 36.30 | 38.37 35.27 | 36.75 35.05 | 38.47 | -| | 7B | 67.97 | 26.26 | 17.87 17.85 | 17.86 33.71 | 51.56 | -Mistral-Instruct 8×7B 65.51 33.90 24.57 24.48 24.52 36.20 53.83 -Training-based -| | 7B | 74.33 | 29.96 | 67.82 66.97 | 67.39 35.70 | 29.83 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -Self-RAG -| | 13B | 71.59 | 31.66 | 71.26 70.35 | 70.80 36.01 | 27.03 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 76.66 | 40.32 | 67.67 63.67 | 65.61 38.32 | 62.00 | -VANILLA-SFT -| | 13B | 84.36 | 40.85 | 71.49 66.21 | 68.75 38.22 | 58.82 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 81.88 | 40.84 | 77.70 69.89 | 73.59 36.95 | 53.93 | -FRONT -| | 13B | 76.11 | 41.51 | 78.44 73.66 | 75.95 38.63 | 57.56 | -| --- | --- | -------- | ---------------- | ----------- | ----------- | ----- | -| | | Table10: | ASQAfullresults. | | | | - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | ------------ | -------- | ------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (Claim) Rec. | Prec. F1 | ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 44.65 | 12.47 49.44 | 47.05 48.22 | 20.64 | 90.2 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 63.72 | 4.53 3.92 | 5.38 4.54 | 18.27 | 103.36 | -| LLaMA-2 | 13B | 62.19 | 7.77 8.49 | 8.43 8.46 | 19.95 | 88.23 | -| | 70B | 53.39 | 10.43 23.75 | 22.43 23.07 | 20.43 | 93.84 | -| | 7B | 32.80 | 12.47 19.90 | 15.48 17.41 | 20.88 | 96.42 | -| LLaMA-2-Chat | 13B | 29.08 | 13.83 16.50 | 16.09 16.29 | 21.04 | 94.32 | -| | 70B | 33.69 | 13.30 36.63 | 36.63 36.63 | 21.29 | 117.84 | -| | 7B | 31.45 | 12.30 29.81 | 22.45 25.61 | 21.36 | 105.68 | -Vicuna-v1.5 -| | 13B | 37.41 | 14.33 31.15 | 28.99 30.03 | 21.74 | 98.23 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 56.62 | 8.47 16.04 | 16.32 16.18 | 20.46 | 93.80 | -Mistral -| | 8×7B | 61.83 | 10.43 26.11 | 25.09 25.59 | 20.66 | 93.59 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 32.74 | 11.07 49.25 | 42.69 45.74 | 20.75 | 98.28 | -Mistral-Instruct -| | 8×7B | 38.51 | 13.93 49.28 | 48.34 48.81 | 21.34 | 113.71 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -Post-hocRetrieval -| ChatGPT | - | 22.79 | 18.77 14.55 | 14.55 14.55 | 22.28 | 106.83 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 72.80 | 7.23 6.84 | 6.84 6.84 | 19.14 | 88.19 | -| LLaMA-2 | 13B | 53.21 | 10.33 9.61 | 9.61 9.61 | 20.63 | 90.44 | -| | 70B | 58.97 | 11.10 10.27 | 10.26 10.26 | 20.41 | 77.85 | -| | 7B | 22.50 | 14.17 11.33 | 11.33 11.33 | 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 30.36 | 14.93 12.10 | 12.10 12.10 | 21.82 | 109.79 | -| | 70B | 37.87 | 16.03 12.93 | 12.93 12.93 | 21.57 | 99.94 | -| | 7B | 30.88 | 11.83 10.91 | 10.91 10.91 | 21.66 | 99.03 | -Vicuna-v1.5 -| | 13B | 32.59 | 15.20 14.06 | 14.06 14.05 | 14.05 | 108.16 | -| --- | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 52.45 | 10.47 8.64 | 8.64 8.64 | 20.48 | 90.17 | -Mistral -| | 8×7B | 48.39 | 13.57 11.62 | 11.62 11.62 | 21.43 | 91.97 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 27.41 | 17.07 13.20 | 13.20 13.20 | 21.52 | 106.93 | -Mistral-Instruct -| | 8×7B | 27.60 | 17.37 15.68 | 15.68 15.68 | 21.66 | 95.21 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -Training-based -| | 7B | 30.98 | 6.90 22.34 | 32.40 26.45 | 16.48 | 41.66 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -Self-RAG -| | 13B | 32.04 | 6.07 30.46 | 40.20 34.66 | 15.23 | 38.19 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -| | 7B | 44.12 | 9.63 42.30 | 40.06 41.15 | 20.58 | 80.43 | -VANILLA-SFT -| | 13B | 46.33 | 10.27 46.75 | 44.47 45.58 | 20.56 | 84.01 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 36.90 | 9.18 58.60 | 55.33 56.92 | 19.09 | 74.06 | -FRONT -| | 13B | 34.37 | 9.32 60.31 | 59.21 59.75 | 19.66 | 75.14 | -| --- | --- | -------- | ---------------- | ----------- | ----- | ----- | -| | | Table11: | ELI5fullresults. | | | | - -| | | Correctness | Citation | | -| --------- | --------- | ------------ | ---------- | ----------- | -| ModelType | ModelSize | | | | -| | | Rec.-5 Prec. | Rec. Prec. | F1 NumPred. | -Prompting-based -| ChatGPT | - | 20.28 19.84 | 19.06 22.03 | 20.44 4.71 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 12.56 11.32 | 6.03 6.35 | 6.19 7.02 | -| LLaMA-2 | 13B | 18.00 12.39 | 5.45 5.74 | 5.59 11.31 | -| | 70B | 18.50 14.79 | 10.10 10.50 | 10.30 8.31 | -| | 7B | 17.96 19.74 | 9.58 9.68 | 9.63 4.73 | -| LLaMA-2-Chat | 13B | 21.34 18.86 | 8.94 9.06 | 9.00 6.51 | -| | 70B | 22.62 18.04 | 13.49 13.98 | 13.73 7.44 | -| | 7B | 14.22 14.74 | 11.26 11.64 | 11.45 5.87 | -Vicuna-v1.5 -| | 13B | 22.06 19.60 | 13.04 13.74 | 13.38 7.62 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 16.96 15.98 | 7.50 7.76 | 7.63 6.29 | -Mistral -| | 8×7B | 18.18 15.63 | 9.72 10.20 | 9.95 6.63 | -| --- | ---- | ----------- | ----------- | ---------- | -| | 7B | 17.52 21.29 | 17.56 18.53 | 18.03 4.54 | -Mistral-Instruct -| | 8×7B | 20.12 19.64 | 19.27 20.38 | 19.81 5.32 | -| --- | ---- | ----------- | ----------- | ---------- | -Post-hocRetrieval -| ChatGPT | - | 25.14 22.85 | 12.29 12.29 | 12.29 5.46 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 6.48 5.11 | 5.05 5.05 | 5.05 6.55 | -| LLaMA-2 | 13B | 9.88 7.17 | 5.20 5.20 | 5.20 6.98 | -| | 70B | 14.44 12.44 | 7.49 7.49 | 7.49 7.41 | -| | 7B | 12.94 10.89 | 7.76 7.76 | 7.76 5.99 | -| LLaMA-2-Chat | 13B | 15.72 12.23 | 7.87 7.87 | 7.87 6.32 | -| | 70B | 17.90 14.45 | 9.05 9.05 | 9.05 6.05 | -| | 7B | 12.04 9.71 | 6.69 6.69 | 6.69 7.10 | -Vicuna-v1.5 -| | 13B | 14.78 11.47 | 8.50 8.50 | 8.50 6.67 | -| --- | --- | ----------- | --------- | --------- | -| | 7B | 9.94 7.90 | 6.00 6.00 | 6.00 7.38 | -Mistral -| | 8×7B | 13.92 12.08 | 6.70 6.70 | 6.70 6.58 | -| --- | ---- | ----------- | --------- | --------- | -| | 7B | 15.80 12.15 | 8.34 8.34 | 8.34 7.01 | -Mistral-Instruct -| | 8×7B | 24.16 18.28 | 9.78 9.78 | 9.78 7.37 | -| --- | ---- | ----------- | --------- | --------- | -Training-based -| | 7B | 2.34 1.98 | 10.53 18.80 | 13.50 3.49 | -| --- | --- | --------- | ----------- | ---------- | -Self-RAG -| | 13B | 1.90 1.33 | 12.79 20.90 | 15.86 3.08 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 12.86 21.09 | 21.35 21.36 | 21.35 7.49 | -VANILLA-SFT -| | 13B | 12.68 22.80 | 23.64 23.71 | 23.67 3.14 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 11.50 21.38 | 24.74 24.84 | 24.79 3.08 | -FRONT -| | 13B | 11.94 22.61 | 24.86 25.39 | 25.12 3.17 | -| --- | -------- | ------------------- | ----------- | ---------- | -| | Table12: | QAMPARIfullresults. | | | \ No newline at end of file +1. + +2. + +3. + +4. + +5. + +6. + +7. + +About These Forms + +In General. This and the other pleading forms available from the www.uscourts.gov +website illustrate some types of information that are useful to have in complaints and +some other pleadings. The forms do not try to cover every type of case. They are limited +to types of cases often filed in federal courts by those who represent themselves or who +may not have much experience in federal courts. + +Not Legal Advice. No form provides legal advice. No form substitutes for having or +consulting a lawyer. If you are not a lawyer and are suing or have been sued, it is best to +have or consult a lawyer if possible. + +No Guarantee. Following a form does not guarantee that any pleading is legally or +factually correct or sufficient. + +Variations Possible. A form may call for more or less information than a particular +court requires. The fact that a form asks for certain information does not mean that every +court or a particular court requires it. And if the form does not ask for certain +information, a particular court might still require it. Consult the rules and caselaw that +govern in the court where you are filing the pleading. + +Examples Only. The forms do not try to address or cover all the different types of +claims or defenses, or how specific facts might affect a particular claim or defense. Some +of the forms, such as the form for a generic complaint, apply to different types of cases. +Others apply only to specific types of cases. Be careful to use the form that fits your case +and the type of pleading you want to file. Be careful to change the information the form +asks for to fit the facts and circumstances of your case. + +No Guidance on Timing or Parties. The forms do not give any guidance on when +certain kinds of pleadings or claims or defenses have to be raised, or who has to be sued. +Some pleadings, claims, or defenses have to be raised at a certain point in the case or +within a certain period of time. And there are limits on who can be named as a party in a +case and when they have to be added. Lawyers and people representing themselves must +know the Federal Rules of Civil Procedure and the caselaw setting out these and other +requirements. The current Federal Rules of Civil Procedure are available, for free, at +www.uscourts.gov. + +Privacy Requirements. Federal Rule of Civil Procedure 5.2 addresses the privacy and +security concerns over public access to electronic court files. Under this rule, papers filed +with the court should not contain anyone’s full social-security number or full birth date; +the name of a person known to be a minor; or a complete financial-account number. A +filing may include only the last four digits of a social-security number and taxpayer +identification number; the year of someone’s birth; a minor’s initials; and the last four +digits of a financial-account number. + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +IN THE UNITED STATES DISTRICT COURT +FOR THE EASTERN DISTRICT OF MICHIGAN + +Case No. +________________________________ +(to be filled in by the Clerk’s Office) + +(Write the full name of each plaintiff who is filing this complaint. +If the names of all the plaintiffs cannot fit in the space above, +please write “see attached” in the space and attach an additional +page with the full list of names.) + +Jury Trial: ☐ Yes ☐ No + +(check one) + +v. + +(Write the full name of each defendant who is being sued. If the +names of all the defendants cannot fit in the space above, please +write “see attached” in the space and attach an additional page +with the full list of names.) + +Complaint for a Civil Case + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +I. + +The Parties to This Complaint + +A. + +The Plaintiff(s) + +Provide the information below for each plaintiff named in the complaint. Attach +additional pages if needed. + +Name + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address + +__________________________________________ + +B. + +The Defendant(s) + +Provide the information below for each defendant named in the complaint, whether the +defendant is an individual, a government agency, an organization, or a corporation. +For an individual defendant, include the person’s job or title (if known). Attach +additional pages if needed. + +Defendant No. 1 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +Defendant No. 2 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +2 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +Defendant No. 3 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +Defendant No. 4 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +II. + +Basis for Jurisdiction + +Federal courts are courts of limited jurisdiction (limited power). Generally, only two types of +cases can be heard in federal court: cases involving a federal question and cases involving +diversity of citizenship of the parties. Under 28 U.S.C. § 1331, a case arising under the United +States Constitution or federal laws or treaties is a federal question case. Under 28 U.S.C. +§ 1332, a case in which a citizen of one State sues a citizen of another State or nation and the +amount at stake is more than $75,000 is a diversity of citizenship case. In a diversity of +citizenship case, no defendant may be a citizen of the same State as any plaintiff. + +What is the basis for federal court jurisdiction? (check all that apply) + +☐ Federal question + +☐ Diversity of citizenship + +Fill out the paragraphs in this section that apply to this case. + +3 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +A. + +If the Basis for Jurisdiction Is a Federal Question + +List the specific federal statutes, federal treaties, and/or provisions of the United States +Constitution that are at issue in this case. + +B. + +If the Basis for Jurisdiction Is Diversity of Citizenship + +1. + +The Plaintiff(s) + +a. + +b. + +If the plaintiff is an individual +The plaintiff, (name) _________________________________________, +is a citizen of the State of (name) _______________________. + +If the plaintiff is a corporation +The plaintiff, (name) ________________________________________, +is incorporated under the laws of the State of (name) +______________________, and has its principal place of business in the +State of (name) _______________________. + +(If more than one plaintiff is named in the complaint, attach an additional page +providing the same information for each additional plaintiff.) + +2. + +The Defendant(s) + +a. + +b. + +If the defendant is an individual +The defendant, (name) ______________________, is a citizen of the +State of (name) _______________________. Or is a citizen of (foreign +nation) _______________________. + +If the defendant is a corporation +The defendant, (name) _______________________, is incorporated +under the laws of the State of (name) _______________________, and +has its principal place of business in the State of (name) +_______________________. Or is incorporated under the laws of +(foreign nation) _______________________, and has its principal place +of business in (name) _______________________. + +(If more than one defendant is named in the complaint, attach an additional +page providing the same information for each additional defendant.) + +4 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +3. + +The Amount in Controversy + +The amount in controversy—the amount the plaintiff claims the defendant owes +or the amount at stake—is more than $75,000, not counting interest and costs of +court, because (explain): + +III. + +Statement of Claim + +Write a short and plain statement of the claim. Do not make legal arguments. State as briefly +as possible the facts showing that each plaintiff is entitled to the damages or other relief sought. +State how each defendant was involved and what each defendant did that caused the plaintiff +harm or violated the plaintiff’s rights, including the dates and places of that involvement or +conduct. If more than one claim is asserted, number each claim and write a short and plain +statement of each claim in a separate paragraph. Attach additional pages if needed. + +5 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +IV. Relief + +State briefly and precisely what damages or other relief the plaintiff asks the court to order. Do +not make legal arguments. Include any basis for claiming that the wrongs alleged are +continuing at the present time. Include the amounts of any actual damages claimed for the acts +alleged and the basis for these amounts. Include any punitive or exemplary damages claimed, +the amounts, and the reasons you claim you are entitled to actual or punitive money damages. + +V. + +Certification and Closing + +Under Federal Rule of Civil Procedure 11, by signing below, I certify to the best of my +knowledge, information, and belief that this complaint: (1) is not being presented for an +improper purpose, such as to harass, cause unnecessary delay, or needlessly increase the cost of +litigation; (2) is supported by existing law or by a nonfrivolous argument for extending, +modifying, or reversing existing law; (3) the factual contentions have evidentiary support or, if +specifically so identified, will likely have evidentiary support after a reasonable opportunity for +further investigation or discovery; and (4) the complaint otherwise complies with the +requirements of Rule 11. + +A. + +For Parties Without an Attorney + +I agree to provide the Clerk’s Office with any changes to my address where case- +related papers may be served. I understand that my failure to keep a current address on +file with the Clerk’s Office may result in the dismissal of my case. + +Date of signing: _____________________, 20_____. + +Signature of Plaintiff +___________________________________________ +Printed Name of Plaintiff ___________________________________________ + +6 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +Additional Information: + +7 + + JS 44 (Rev. 10/20) + +CIVIL COVER SHEET +The JS 44 civil cover sheet and the information contained herein neither replace nor supplement the filing and service of pleadings or other papers as required by law, except as +provided by local rules of court. This form, approved by the Judicial Conference of the United States in September 1974, is required for the use of the Clerk of Court for the +purpose of initiating the civil docket sheet. (SEE INSTRUCTIONS ON NEXT PAGE OF THIS FORM.) +I. (a) PLAINTIFFS + +County in which action arose: _______________ + +DEFENDANTS + +(b) County of Residence of First Listed Plaintiff + +(EXCEPT IN U.S. PLAINTIFF CASES) + +County of Residence of First Listed Defendant + +NOTE: + +(IN U.S. PLAINTIFF CASES ONLY) +IN LAND CONDEMNATION CASES, USE THE LOCATION OF +THE TRACT OF LAND INVOLVED. + +(c) Attorneys (Firm Name, Address, and Telephone Number) + +Attorneys (If Known) + +II. BASIS OF JURISDICTION (Place an “X” in One Box Only) + +III. CITIZENSHIP OF PRINCIPAL PARTIES (Place an “X” in One Box for Plaintiff + +1 U.S. Government +Plaintiff + +3 Federal Question + +(U.S. Government Not a Party) + +Citizen of This State + +PTF +1 + +DEF +1 + +(For Diversity Cases Only) + +and One Box for Defendant) + +PTF + +Incorporated or Principal Place +of Business In This State + +2 U.S. Government + +Defendant + +4 Diversity + +Citizen of Another State + +(Indicate Citizenship of Parties in Item III) + +DEF +4 + +5 + +6 + +4 + +5 + +6 + +2 + +3 + +2 + +Incorporated and Principal Place +of Business In Another State + +3 + +Foreign Nation + +Click here for: Nature of Suit Code Descriptions. +OTHER STATUTES + +BANKRUPTCY + +IV. NATURE OF SUIT (Place an “X” in One Box Only) + +TORTS + +CONTRACT + +110 Insurance +120 Marine +130 Miller Act +140 Negotiable Instrument +150 Recovery of Overpayment + +PERSONAL INJURY +310 Airplane +315 Airplane Product + +Liability + +320 Assault, Libel & + +& Enforcement of Judgment + +Slander + +151 Medicare Act +152 Recovery of Defaulted + +Student Loans +(Excludes Veterans) +153 Recovery of Overpayment +of Veteran’s Benefits + +160 Stockholders’ Suits +190 Other Contract +195 Contract Product Liability +196 Franchise + +REAL PROPERTY +210 Land Condemnation +220 Foreclosure +230 Rent Lease & Ejectment +240 Torts to Land +245 Tort Product Liability +290 All Other Real Property + +330 Federal Employers’ + +Liability + +340 Marine +345 Marine Product +Liability +350 Motor Vehicle +355 Motor Vehicle + +Product Liability + +360 Other Personal + +Injury + +362 Personal Injury - + +Medical Malpractice + +CIVIL RIGHTS +440 Other Civil Rights +441 Voting +442 Employment +443 Housing/ + +Accommodations +445 Amer. w/Disabilities - + +Employment + +446 Amer. w/Disabilities - + +Other +448 Education + +Citizen or Subject of a +Foreign Country + +FORFEITURE/PENALTY + +625 Drug Related Seizure + +of Property 21 USC 881 + +690 Other + +PERSONAL INJURY +365 Personal Injury - +Product Liability + +367 Health Care/ + +Pharmaceutical +Personal Injury +Product Liability + +368 Asbestos Personal +Injury Product +Liability + +422 Appeal 28 USC 158 +423 Withdrawal +28 USC 157 + +PROPERTY RIGHTS +820 Copyrights +830 Patent +835 Patent - Abbreviated + +New Drug Application + +840 Trademark +880 Defend Trade Secrets + +PERSONAL PROPERTY + +LABOR + +710 Fair Labor Standards + +Act of 2016 + +Act + +720 Labor/Management + +Relations + +740 Railway Labor Act +751 Family and Medical + +Leave Act + +790 Other Labor Litigation +791 Employee Retirement +Income Security Act + +IMMIGRATION + +462 Naturalization Application +465 Other Immigration + +Actions + +SOCIAL SECURITY +861 HIA (1395ff) +862 Black Lung (923) +863 DIWC/DIWW (405(g)) +864 SSID Title XVI +865 RSI (405(g)) + +FEDERAL TAX SUITS +870 Taxes (U.S. Plaintiff +or Defendant) +871 IRS—Third Party +26 USC 7609 + +370 Other Fraud +371 Truth in Lending +380 Other Personal + +Property Damage +385 Property Damage +Product Liability + +PRISONER PETITIONS + +Habeas Corpus: +463 Alien Detainee +510 Motions to Vacate + +Sentence +530 General +535 Death Penalty +Other: +540 Mandamus & Other +550 Civil Rights +555 Prison Condition +560 Civil Detainee - +Conditions of +Confinement + +375 False Claims Act +376 Qui Tam (31 USC + +3729(a)) + +400 State Reapportionment +410 Antitrust +430 Banks and Banking +450 Commerce +460 Deportation +470 Racketeer Influenced and +Corrupt Organizations + +480 Consumer Credit + +(15 USC 1681 or 1692) + +485 Telephone Consumer +Protection Act +490 Cable/Sat TV +850 Securities/Commodities/ + +Exchange + +890 Other Statutory Actions +891 Agricultural Acts +893 Environmental Matters +895 Freedom of Information + +Act +896 Arbitration +899 Administrative Procedure +Act/Review or Appeal of +Agency Decision +950 Constitutionality of +State Statutes + +V. ORIGIN (Place an “X” in One Box Only) + +1 Original + +Proceeding + +2 Removed from +State Court + +3 Remanded from +Appellate Court + +4 Reinstated or +Reopened + +5 Transferred from +Another District +(specify) + +6 Multidistrict +Litigation - +Transfer + +8 Multidistrict +Litigation - +Direct File + +Cite the U.S. Civil Statute under which you are filing (Do not cite jurisdictional statutes unless diversity): + +VI. CAUSE OF ACTION + +Brief description of cause: + +VII. REQUESTED IN + +COMPLAINT: +VIII. RELATED CASE(S) + +IF ANY + +CHECK IF THIS IS A CLASS ACTION +UNDER RULE 23, F.R.Cv.P. + +DEMAND $ + +CHECK YES only if demanded in complaint: + +JURY DEMAND: + +Yes + +No + +(See instructions): + +JUDGE + +DOCKET NUMBER + +DATE + +SIGNATURE OF ATTORNEY OF RECORD + +FOR OFFICE USE ONLY + +RECEIPT # + +AMOUNT + +APPLYING IFP + +JUDGE + +MAG. JUDGE + + PURSUANT TO LOCAL RULE 83.11 + +1. + +Is this a case that has been previously dismissed? + +If yes, give the following information: + +Court: + +Case No.: + +Judge: + +2. + +Other than stated above, are there any pending or previously +discontinued or dismissed companion cases in this or any other +court, including state court? (Companion cases are matters in which +it appears substantially similar evidence will be offered or the same +or related parties are present and the cases arise out of the same +transaction or occurrence.) + +Yes + +No + +Yes + +No + +If yes, give the following information: + +Court: + +Case No.: + +Judge: + +Notes : + + JS 44 Reverse (Rev. 10/20) + +INSTRUCTIONS FOR ATTORNEYS COMPLETING CIVIL COVER SHEET FORM JS 44 + +Authority For Civil Cover Sheet + +The JS 44 civil cover sheet and the information contained herein neither replaces nor supplements the filings and service of pleading or other papers as +required by law, except as provided by local rules of court. This form, approved by the Judicial Conference of the United States in September 1974, is +required for the use of the Clerk of Court for the purpose of initiating the civil docket sheet. Consequently, a civil cover sheet is submitted to the Clerk of +Court for each civil complaint filed. The attorney filing a case should complete the form as follows: + +I.(a) Plaintiffs-Defendants. Enter names (last, first, middle initial) of plaintiff and defendant. If the plaintiff or defendant is a government agency, use + +only the full name or standard abbreviations. If the plaintiff or defendant is an official within a government agency, identify first the agency and then +the official, giving both name and title. + +(b) County of Residence. For each civil case filed, except U.S. plaintiff cases, enter the name of the county where the first listed plaintiff resides at the +time of filing. In U.S. plaintiff cases, enter the name of the county in which the first listed defendant resides at the time of filing. (NOTE: In land +condemnation cases, the county of residence of the "defendant" is the location of the tract of land involved.) + +(c) Attorneys. Enter the firm name, address, telephone number, and attorney of record. If there are several attorneys, list them on an attachment, noting + +in this section "(see attachment)". + +II. + +III. + +IV. + +V. + +Jurisdiction. The basis of jurisdiction is set forth under Rule 8(a), F.R.Cv.P., which requires that jurisdictions be shown in pleadings. Place an "X" +in one of the boxes. If there is more than one basis of jurisdiction, precedence is given in the order shown below. +United States plaintiff. (1) Jurisdiction based on 28 U.S.C. 1345 and 1348. Suits by agencies and officers of the United States are included here. +United States defendant. (2) When the plaintiff is suing the United States, its officers or agencies, place an "X" in this box. +Federal question. (3) This refers to suits under 28 U.S.C. 1331, where jurisdiction arises under the Constitution of the United States, an amendment +to the Constitution, an act of Congress or a treaty of the United States. In cases where the U.S. is a party, the U.S. plaintiff or defendant code takes +precedence, and box 1 or 2 should be marked. +Diversity of citizenship. (4) This refers to suits under 28 U.S.C. 1332, where parties are citizens of different states. When Box 4 is checked, the +citizenship of the different parties must be checked. (See Section III below; NOTE: federal question actions take precedence over diversity +cases.) + +Residence (citizenship) of Principal Parties. This section of the JS 44 is to be completed if diversity of citizenship was indicated above. Mark this +section for each principal party. + +Nature of Suit. Place an "X" in the appropriate box. If there are multiple nature of suit codes associated with the case, pick the nature of suit code +that is most applicable. Click here for: Nature of Suit Code Descriptions. + +Origin. Place an "X" in one of the seven boxes. +Original Proceedings. (1) Cases which originate in the United States district courts. +Removed from State Court. (2) Proceedings initiated in state courts may be removed to the district courts under Title 28 U.S.C., Section 1441. +Remanded from Appellate Court. (3) Check this box for cases remanded to the district court for further action. Use the date of remand as the filing +date. +Reinstated or Reopened. (4) Check this box for cases reinstated or reopened in the district court. Use the reopening date as the filing date. +Transferred from Another District. (5) For cases transferred under Title 28 U.S.C. Section 1404(a). Do not use this for within district transfers or +multidistrict litigation transfers. +Multidistrict Litigation – Transfer. (6) Check this box when a multidistrict case is transferred into the district under authority of Title 28 U.S.C. +Section 1407. +Multidistrict Litigation – Direct File. (8) Check this box when a multidistrict case is filed in the same district as the Master MDL docket. +PLEASE NOTE THAT THERE IS NOT AN ORIGIN CODE 7. Origin Code 7 was used for historical records and is no longer relevant due to +changes in statue. + +VI. + +Cause of Action. Report the civil statute directly related to the cause of action and give a brief description of the cause. Do not cite jurisdictional +statutes unless diversity. Example: U.S. Civil Statute: 47 USC 553 Brief Description: Unauthorized reception of cable service. + +VII. Requested in Complaint. Class Action. Place an "X" in this box if you are filing a class action under Rule 23, F.R.Cv.P. + +Demand. In this space enter the actual dollar amount being demanded or indicate other demand, such as a preliminary injunction. +Jury Demand. Check the appropriate box to indicate whether or not a jury is being demanded. + +VIII. Related Cases. This section of the JS 44 is used to reference related pending cases, if any. If there are related pending cases, insert the docket + +numbers and the corresponding judge names for such cases. + +Date and Attorney Signature. Date and sign the civil cover sheet. + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Contract +Code +110 + +Title +Insurance + +120 + +Marine + +130 + +Miller Act + +140 + +Negotiable Instrument + +150 + +Recovery of Overpayment & Enforcement +Judgment + +151 + +Medicare + +152 + +153 + +160 + +Recovery of Defaulted Student Loans +(Excludes Veterans) +Recovery of Overpayment of Veterans' +Benefits +Stockholders' Suits + +190 + +Other Contract + +195 + +Contract Product Liability + +Actions primarily alleging personal injury +or property damage caused by a defective +product should be classified under the +appropriate nature of suit code under +“TORTS.” +Franchise + +196 + +Description +Action alleging breach of insurance contract, tort claim, or other cause +related to an insurance contract, except for maritime insurance +contracts. +Action (Admiralty or Maritime) based on service, employment, +insurance or other contracts relating to maritime vessels and other +maritime contractual matters. +Action based on performance and payment bonds agreed to by +contractors on federal construction projects as required under the +Miller Act, 40 U.S.C. § 3131-3134. +Action relating to an agreement to pay a specific amount of money, +including promissory notes, loan agreements and checks. +Action to recover debt owed to the United States, including +enforcement of judgments, based on overpayments and restitution +agreements involving matters other than Medicare benefits, student +loans and veterans’ benefits. +Action relating to Medicare payments, including actions for payments of +benefits, to recover overpayments, and for judicial review of +administrative decisions. +Action to recover debt owed to the United States from defaulted +student loan. +Action relating to payments of veterans’ benefits, primarily including +actions to recover overpayments. +Action brought by stockholder(s) of a corporation (including both +stockholder derivative suits and direct actions based on plaintiff’s rights +as a stockholder), usually alleging claims based on contract and/or tort +law and/or fiduciary obligations. +Action primarily based on rights and obligations under a contract not +classifiable elsewhere under the specific natures of suit under +“Contract.” +Action concerning damages caused by a defective product, not primarily +involving personal injury or property damage, and based primarily on +breach of contract, breach of warranty, misrepresentation, and/or +violation of consumer protection laws. + +Action arising from a dispute over a franchise agreement, typically +alleging breach of contract, misrepresentation or unfair trade practices. + +Page 1 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Real Property + +Code +210 + +Title +Land Condemnation + +220 + +230 + +240 + +Foreclosure + +Rent Lease & Ejectment + +Torts to Land + +245 + +Tort Product Liability + +290 + +All Other Real Property + +Torts/Personal Injury + +Title +Airplane +(Excludes airplane product liability +claims) +Airplane Product Liability + +Description +Action by a governmental entity to take privately-owned real property +(land or buildings) for public use for compensation. +Action to enjoin foreclosure on real property by mortgage lender. + +Action for rental or lease payments owed on real property and/or to +eject a party occupying real property illegally. +Action alleging trespass to land, nuisance, contamination or other +unlawful entry on or interference with real property possessed by +another. +Action alleging harm by an unsafe product based on negligence, breach +of warranty, misrepresentation, and strict tort liability. +Action primarily based on unlawful conduct relating to real property +that cannot be classified under any other nature of suit. + +Description +Action alleging personal injury or wrongful death from an air crash or +other occurrence involving an airplane. + +Code +310 + +315 + +320 + +330 + +340 + +345 + +350 + +355 + +360 + +362 + +365 + +367 + +368 + +Motor Vehicle + +Marine Product Liability + +Marine +(Excludes marine product liability claims) + +Assault, Libel & Slander +(Excludes a government employee) +Federal Employers' Liability + +Action alleging personal injury or death from an air crash or other +occurrence involving an airplane and caused by a defective product. +Action alleging intentional acts of assault, libel, trade libel or slander by +a private party. +Action for personal injury or wrongful death brought by a railroad +employee or his survivors under the Federal Employers’ Liability Act +(FELA), 45 U.S.C. § 51, et. seq. +Action (Admiralty and Maritime) alleging personal injury or death from +an accident involving a water vessel or harbor/dock facilities, including +suits brought under the Jones Act and the Limitation of Liability Act. +Action (Admiralty and Maritime) alleging personal injury or wrongful +death from an accident involving a water vessel or harbor/dock +facilities and caused by a defective product. +Action alleging personal injury or wrongful death from negligence +involving a motor vehicle but not caused by a defective product. +Action alleging personal injury or wrongful death involving a motor +vehicle and caused by a defective product. +Action primarily based on personal injury or death caused by +negligence or intentional misconduct, including suits brought against +the United States under the Federal Tort Claims Act, and which cannot +be classified under any other nature of suit. +Action alleging personal injury or wrongful death caused by negligence +in medical care provided by a doctor or other health care professional. +Action alleging personal injury or death resulting from a defective +Personal Injury - Product Liability +product. +(Excludes a marine or airplane product) +Action alleging personal injury or death caused by a defective medical +Health Care/Pharmaceutical Personal +or pharmaceutical product. +Injury Product Liability +Asbestos Personal Injury Product Liability Action alleging personal injury or death caused by exposure to asbestos + +Personal Injury - Medical Malpractice + +Motor Vehicle Product Liability + +Other Personal Injury + +products. + +Page 2 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Personal Property + +Code +370 + +371 + +Title +Other Fraud +(Excludes any property that is not real +property) +Truth in Lending + +Actions relating to fraud or +misrepresentation in the transfer of real +property should be classified under +nature of suit 290, “All Other Real +Property,” or, if foreclosure is involved, +under nature of suit 220, “Foreclosure.” +Other Personal Property Damage + +380 + +385 + +Property Damage Product Liability + +Civil Rights + +Code + +440 + +441 + +442 + +443 + +445 + +446 + +Title +Other Civil Rights +(Excludes claims against corrections +officials) +Voting + +Employment + +Housing/Accommodations + +Americans With Disabilities - +Employment +Americans With Disabilities - Other + +448 + +Education + +Description +Action primarily based on fraud relating to personal property that +cannot be classified under any other nature of suit. + +Action alleging violation of the federal Truth in Lending Act arising from +consumer loan transactions involving personal property including +automobile loans and revolving credit accounts. + +Action primarily based on damage to personal property caused by +harmful conduct such as negligence, misrepresentation, interference +with business relationships or unfair trade practices. +Action alleging damage to personal property caused by a defective +product. + +Description + +Action alleging a civil rights violation other than the specific civil rights +categories listed below or a violation related to prison. Example: +Action alleging excessive force by police incident to an arrest. +Action filed under Civil Rights Act, 52 U.S.C. § 10101, and Voting Rights +Act, 52 U.S.C. § 10301 +Action filed under Age Discrimination in Employment Act 29:621:634, +Equal Employment Opportunity Act (Title VII) 42:2000E, Performance +Rating Act of 1950 5:4303 +Action filed under the Fair Housing Act (Title VII), 42 U.S.C. § 3601 & +3602. +Action of discrimination against an employee with disabilities of any +type in the work place, filed under 42 U.S.C. § 12117 +Action of discrimination against an individual with disabilities in areas +other than employment, filed under 42 U.S.C. § 12133 (exclusion or +discrimination in provision of services, programs or activities of a public +entity) or 42 U.S.C. § 12188 (public accommodations) +Action filed under the Individuals with Disabilities Educations Act, 20 +U.S.C. § 1401 and Title IX of the Education Amendment of 1972, 20 +U.S.C. § 1681 et seq. + +Page 3 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Prisoner Petitions +Habeas Corpus +Title +Alien Detainee + +Code +463 + +510 + +Motions to Vacate Sentence + +530 + +General + +535 + +Death Penalty + +Other +Prisoner Petitions + +Code +540 + +Title +Mandamus & Other + +550 + +Civil Rights + +555 + +Prison Condition + +560 + +Civil Detainee - Conditions of +Confinement (Excludes actions by alien +detainees) + +Description +Immigration habeas petition under 28 U.S.C. § 2241. All cases filed +with this nature of suit code are restricted to case participants and +public terminals. Petition is filed by an alien detainee. +Action by a prisoner to vacate or modify a sentence imposed in federal +court, other than a death sentence, under 28 U.S.C. § 2255. +Action by a federal or state prisoner currently in custody challenging +the legality of confinement or other punishment. This includes claims +alleging illegalities that occurred in trial (for example, ineffective +assistance of counsel), sentencing (including fines and restitution +orders), or disciplinary proceedings in prison (for example, loss of good +time credits). Habeas petition under 28 U.S.C. § 2254 or prisoner +habeas under 28 U.S.C. § 2241. +Action by a federal or state prisoner challenging a death sentence. + +Description +Action by prisoner currently in custody for a writ of mandamus to +compel action by a judge or government official relating to the +prisoner’s confinement, including conditions of confinement. This +category also includes any actions other than mandamus brought by a +prisoner currently in custody, whether or not it relates to his +confinement, if it is not classifiable under any other nature of suit +category under Prisoner Petitions (for example, action by prisoner to +recover property taken by the government in a criminal case). +Action by current or former prisoner alleging a civil rights violation by +corrections officials that is not related to a condition of prison life. +Action by current prisoner, or former prisoner or their families alleging +a civil rights, Federal Tort Claims Act, or state law claim with respect to +a condition of prison life, whether general circumstances or particular +episodes. Examples: inadequate medical care or excessive force by +prison guards. Includes non-habeas actions by alien detainees alleging +unlawful prison conditions. +Action by former prisoner who was involuntarily committed to a non- +criminal facility after expiration of his or her prison term alleging +unlawful conditions of confinement while in the non-criminal facility. +This category includes, for example, an action by a former prisoner +classified as a Sexually Dangerous Person or Sexually Violent Predator +alleging civil rights violations during his detention in a medical facility. + +Page 4 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Forfeiture/Penalty + +Code +625 + +Title +Drug Related Seizure of Property 21 +U.S.C. § 881 + +Description +Action (Forfeiture) by which property itself is accused of wrongdoing +and is forfeited to the government as a result. + +690 + +Other + +Labor +Code +710 + +Title +Fair Labor +Standards Act (Non-Union) + +720 + +Labor/Management Relations (Union) + +740 + +Railway Labor Act + +Action primarily based on Acts or Bills that cannot be classified under +any other nature of suit, such as: Endangered Species Act, Federal +Hazardous Substance Act 15:1261, Game & Wildlife Act 15:256C et seq. +(Penalty), Federal Trade Commission Act 15:41-51 (Penalty), Federal +Coal Mine Health & Safety Act 30:801 et seq. (Penalty), Load Line Act +46:85-85G, McGuire Bill (Federal Fair Trade) 15:45L Penalty, Marihuana +Tax Act 50 STAT 551, Motorboat Act 46:526-526T, National Traffic & +Motor Vehicle Safety Act penalty 49:1655, Veterans' Benefit Act, Title +38 Penalty. + +Description +Action relating to non-union workplace related disputes filed under the +Fair Labor Standards Act, 29 U.S.C. § 201 including but not limited to +wage discrimination, paid leave, minimum wage and overtime pay. + +Action relating to disputes between labor unions and employers as well +as all petitions regarding actions of the Nation Labor Relations Board +(NLRB) + +Action relating to disputes filed under the Railway Labor Act, 45 U.S.C. +§ 151 including labor disputes, individual claims, and response to +sanctions. + +751 + +790 + +791 + +Family and Medical Leave Act + +Action filed under the Family Medical Leave Act, 29 U.S.C. § 2601 + +Other Labor Litigation + +Employee Retirement Income Security +Act + +Action primarily based on labor disputes not addressed by other NOS +codes (includes Labor/Management Reporting and Disclosure Act) +Action filed under the Employee Retirement Income Security Act, 29 +U.S.C. § 1132 by individuals and labor organizations. + +Immigration + +Code +462 + +Title +Naturalization Application + +465 + +Other Immigration +Actions + +Description +Action seeking review of denial of an application for naturalization [8 +U.S.C. § 1447(b)] or alleging failure to make a determination regarding +an application for naturalization [8 U.S.C. § 1421(c)]. + +Action (Immigration-related) that do not involve Naturalization +Applications or petitions for Writ of Habeas Corpus, such as complaints +alleging failure to adjudicate an application to adjust immigration +status to permanent resident. + +Page 5 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Bankruptcy + +Code +422 + +Title +Appeal 28 USC § 158 + +423 + +Withdrawal of Reference 28 USC § 157 + +Property Rights +Title +Copyright + +Code +820 + +Patent + +Patent – Abbreviated New Drug +Application (ANDA) + +Trademark + +830 + +835 + +840 + +880 + +Description +All appeals of previous bankruptcy decisions filed under +28 U.S.C. § 158 +Action held in bankruptcy court requesting withdrawal under the +provisions of 28 U.S.C. § 157 + +Description +Action filed in support or to dispute a copyright claim. + +Action filed in support or to dispute a patent claim. + +Action filed in support or to dispute a patent claim involving an +Abbreviated New Drug Application (ANDA). These cases are also known +as “Hatch-Waxman” cases. +Action filed in support or to dispute a trademark claim + +Defend Trade Secrets Act of 2016 (DTSA) + +Action filed in support or to dispute a trade secret misappropriation + +Social Security +Title +HIA (1395ff) + +Code +861 + +862 + +Black Lung (923) + +863 + +DIWC/DIWW (405(g)) + +864 + +865 + +SSID Title XVI + +RSI (405(g)) + +Federal Tax Suits + +Code +870 +871 + +Title +Taxes (U.S. Plaintiff or Defendant) +IRS-Third Party 26 USC 7609 + +Description +Action filed with regard to social security benefits associated with +Health Insurance Part A Medicare + +Action filed with regard to social security benefits provided for those +who contracted Black Lung or their beneficiaries + +Action filed with regard to social security benefits provided to disabled +individuals: worker or child, or widow + +Action filed with regard to social security benefits provided to +Supplemental Security Income Disability under Title XVI +Action filed with regard to social security benefits provided for +Retirement, Survivor Insurance under 42 U.S.C. § 405 + +Description +Action filed under the Internal Revenue Code (General) +Action filed under the Internal Revenue Code - Tax Reform Act of 1976 +(P.L. 94-455) Third Party + +Page 6 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Other Statutes +Title +False Claims Act + +Code +375 + +376 + +Qui Tam (31 U.S.C. § 3729(a)) + +400 + +State Reapportionment + +410 + +Antitrust + +430 + +450 + +460 + +470 + +480 + +485 + +Banks and Banking + +Commerce + +Deportation + +Racketeer Influenced and Corrupt +Organizations +Consumer Credit + +Telephone Consumer Protection Act +(TCPA) + +490 + +Cable/Satellite TV + +850 + +Securities/Commodities/Exchange + +890 + +Other Statutory Actions + +891 + +Agricultural Acts + +893 + +Environmental Matters + +Description +Action filed by private individuals alleging fraud against the U.S. +Government under 31 U.S.C. § 3729. +Action brought under the False Claims Act by private persons (also +known as "whistleblowers") on their own behalf and on behalf of the +United States to recover damages against another person or entity that +acted fraudulently in receiving payments or property from, or avoiding +debts owed to, the United States Government, 31 U.S.C. § 3730 +Action filed under the Reapportionment Act of 1929 Ch. 28, 46 Stat. 21, +2 U.S.C. § 2a. +Action brought under the Clayton Act 15 U.S.C. § 12 - 27 alleging +undue restriction of trade and commerce by designated methods that +limit free competition in the market place amongst consumers such as +anti-competitive price discrimination, corporate mergers, interlocking +directorates or tying and exclusive dealing contracts. +Action filed under the Federal Home Loan Bank Act 12:1421-1449, +Home Owners Loan Act 12:1461 or Federal Reserve Acts 12:142 et seq. +Action filed under the Interstate Commerce Acts 49:1 et seq., 49:301 + +Action filed under the Immigration Acts (Habeas Corpus & Review) +8:1101/18:1546 +Racketeer Influenced and Corrupt Organization Act, RICO 18:1961-1968 + +Action filed under the Fair Credit Reporting Act, 15 U.S.C. 1681n or +15 U.S.C. 1681o, and the Fair Debt Collection Practices Act, +15 U.S.C. § 1692k +Action filed under the Telephone Consumer Protection Act 47 U.S.C. § +227 + +Action filed involving unauthorized reception of cable/satellite TV +service under 47 U.S.C. § 553 (unauthorized reception of cable/satellite +TV), or 47 U.S.C. § 605 (e)(3) (unauthorized use or publication of a +communication) +Action filed under Small Business Investment Act 15:681, Securities +Exchange Act 15:78, Securities Act 15:77, Investment Advisers Act +15:80B(1-21) +Action primarily based on Statutes that cannot be classified under any +other nature of suit, such as: Foreign Agents Registration Act 22:611- +621, Klamath Termination Act 25:564-564W-L, Federal Aid Highway Act +23:101-142, Federal Corrupt Practices Act 2:241-256, Federal Election +Campaign Act, Highway Safety Act 23:401 Immigration & Nationality +Act 8:1503, Natural Gas Pipeline Safety Act 49:1671-1700, +Naturalization Acts 8:1421/18:911, 1015, 1421, et seq., 3282 or Federal +Aviation Act 49:1301 et seq. +Action filed under the Federal Crop Insurance Act 7:1501-1550, +Commodity Credit Corporation Act 15:713A-L & 4. +Action filed under Air Pollution Control Act 42:1857-57L, Clean Air Act +42:1857:57L, Federal Environment Pesticide Control Act, Federal +Insecticide, Fungicide & Rodenticide Act 7:135, Federal Water Pollution +Control Act 33:1151 et seq., Land & Water Conservation Fund Act +16:4602,460 1-4, Motor Vehicle Air Pollution Control Act 42:1857F-1-8, + +Page 7 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +National Environmental Policy Act 42:4321, 4331-35G, 4341-47, River & +Harbor Act penalty 3:401-437, 1251. + +Other Statutes (Continued) +Title +Freedom of Information Act +Arbitration + +Code +895 +896 + +899 + +Administrative Procedure Act/Review or +Appeal of Agency Decision + +950 + +Constitutionality of State Statutes + +Description +Action filed under the Freedom of Information Act 5:552. +Action involving actions to confirm or modify arbitration awards filed +under Title 9 of the U.S. Code. +Action filed under the Administrative Procedures Act, 5 U.S.C. § 701, or +civil actions to review or appeal a federal agency decision. + +Action drawing into question the constitutionality of a federal or state +statute filed under (Rule 5.1). Rule 5.1 implements 28 U.S.C. §2403. + +Note: The statutes listed above are not all-inclusive, and other statues might be applicable to each nature of suit. + +Statutes that are included in the descriptions should be viewed as examples. + +Page 8 of 8 + diff --git a/q128/random_k4/question.json b/q128/random_k4/question.json index d0883f9b84ff9d8357c061ead021ba9b13b2d00d..d90b154aceed7beca59fcd678791d89a2eeba5b3 100644 --- a/q128/random_k4/question.json +++ b/q128/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "owners-manual-2170416.pdf", - "web_974a02e9fcc23ca9.pdf", - "web_ddfd3d214b1f0ea4.pdf", - "web_7c476895c4da1bbe.pdf", - "web_623919b9713226a3.pdf" + "web_d3545b20cb487e4b.pdf", + "web_9deb1a47e107dc04.pdf", + "95e7e1de309007d2c70710bda64b9291b4915d26.pdf", + "web_c83d6885217d96f0.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q128/random_k4/random_1.md b/q128/random_k4/random_1.md index 0f5db4f84cd2bbe84dc239b53c3a2c97d631e42e..3b59653be960a7700f9cff39752b4e24bce89425 100644 --- a/q128/random_k4/random_1.md +++ b/q128/random_k4/random_1.md @@ -1,279 +1,3031 @@ -COMPETITOR ANALYSIS +FOR RELEASE JULY 26, 2018 -© InnoValor, 2018 -© InnoValor, 2018 +BY Cary Funk and Meg Hefferon -How to use -Template -Example +FOR MEDIA OR OTHER INQUIRIES: - Competitor Analysis +Cary Funk, Director, Science and Society Research +Haley Nolan, Communications Assistant -A Competitor Analysis helps you to understand your -competition. It supports you in creating your unique position in -the market. With this analysis you make an assessment of the -strengths and weaknesses of your current and potential -competitors. With a better understanding of your competition -you can choose and implement effective strategies that will -improve your competitive advantage. +202.419.4372 -Tips for use +www.pewresearch.org -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. This tool can easily be used by -yourself, or with colleagues. Print the template and use the tool -to create an understanding of your competition. +RECOMMENDED CITATION -Have a look at the example of Fleur Flowers to get an -understanding of a Competitor Analysis. +Pew Research Center, July 2018, “Public Views of Gene +Editing for Babies Depend on How It Would Be Used” -How to use + 2 -With the Competitor Analysis you analyze your competitors’ -strengths and weaknesses. Follow the steps to complete your -own Competitor Analysis. +PEW RESEARCH CENTER -Summarize your target group +About Pew Research Center -Shortly describe your target group. Who are the customers that -buy your offering? Make the description as specific as possible. If +Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes +and trends shaping America and the world. It does not take policy positions. The Center conducts +public opinion polling, demographic research, content analysis and other data-driven social +science research. It studies U.S. politics and policy; journalism and media; internet, science and +technology; religion and public life; Hispanic trends; global attitudes and trends; and U.S. social +and demographic trends. All of the Center’s reports are available at www.pewresearch.org. Pew +Research Center is a subsidiary of The Pew Charitable Trusts, its primary funder. -© InnoValor, 2018 +© Pew Research Center 2018 -your target group consist of businesses you can mention their -size, industry or their offering. If your target group consist of -consumers you can mention their age, goals, gender, education -level or interests. +www.pewresearch.org -Identify your competitors + 3 -Write the names for up to three competitors in the first row. To -help you generate this list, think of where your customers would -go if they cannot buy your product or service. +PEW RESEARCH CENTER -Determine competitors’ strengths and weaknesses +Public Views of Gene Editing for Babies Depend on How It +Would Be Used -Now think like your customers. Why would a customer go for the -services of your competitors instead of coming to your company? -You can describe this by thinking about the offerings of your -competitors. It would even be better when you visit your -competitors and experience what their customers experience. -Write down three strengths of each competitor. Also consider -why customers would not go to your competitors. Write down -three weaknesses of each competitor. +Americans’ views on the +appropriateness of changing +a baby’s genetic +characteristics depend in +large part on the intended +purpose and on whether or +not human embryos would be +used in testing these +techniques. A majority of +Americans support the idea +of using gene editing with the +goal of delivering direct +health benefits for babies, but +at the same time, a majority +considers the use of such +techniques to boost a baby’s +intelligence something that +takes technology “too far.” -Describe the key findings +A majority of U.S. adults say changing a baby’s genes +to treat a serious congenital disease is appropriate -Finally, in the fourth column of the template write down the key -findings for the strengths. What stands out? What patterns can -you discover from your competitors? And which opportunities -can you come up with? Also, write down your key -findings for the weaknesses. What are the aspects that +% of U.S. adults who say changing a baby’s genetic characteristics for each of +the following reasons is … - your competitors don’t do well? How could you avoid making -the same mistakes? What should you do to serve your customers -in such a way that you differentiate from your competitors in a -positive way? +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” -© InnoValor, 2018 +PEW RESEARCH CENTER - competitor analysis +About seven-in-ten Americans (72%) say that changing an unborn baby’s genetic characteristics to +treat a serious disease or condition that the baby would have at birth is an appropriate use of +medical technology, while 27% say this would be taking technology too far. A somewhat smaller +share of Americans say gene editing to reduce a baby’s risk of developing a serious disease or +condition over their lifetime is appropriate (60% say this, while 38% say it would be taking +medical technology too far). But just 19% of Americans say it would be appropriate to use gene +editing to make a baby more intelligent; eight-in-ten (80%) say this would be taking medical +technology too far. -name +These are some of the findings from a new Pew Research Center survey conducted April 23-May 6, +2018, among 2,537 U.S. adults. -target group +www.pewresearch.org -competitors + 4 -most important findings +PEW RESEARCH CENTER -s -h -t -g -n -e -r -t -s +While public discussions about potentially altering a baby’s genetic makeup have been ongoing for +decades, the development of a new gene-splicing technology – known as CRISPR – has accelerated +the debate and brought new urgency to better understanding public opinion about gene editing as +well as the broader social, ethical and policy implications ahead.1 -s -e -s -s -e -n -k -a -e -w +Previous Pew Research Center surveys have tracked public opinion about gene editing using +different question wording and, in some cases, different polling methodologies. As such, those +findings are not directly comparable to this new survey. However, the broad pattern – that public +support for gene editing varies with its intended purpose – is consistent with a 2016 study that +explored public views about the possibility of using gene editing to “enhance” a baby’s health over +the course of their lifetime and a 2014 survey.2 -date +Regardless of the intended purpose of gene +editing, experts acknowledge that further +development of these techniques will likely +involve testing in human embryos. In August +2017, research scientists in the United States +reported the first successful use of gene editing +in human embryos to eliminate an inherited +condition. When asked to consider the +possibility that the development of gene +editing would involve testing on human +embryos, one third of Americans (33%) say +this would be appropriate while about two- +thirds (65%) say this would be taking medical +technology too far. - Example Fleurs Flowers +65% of Americans say if gene editing +relies on embryonic testing, it would +take technology too far -industry +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … -Retail +Taking medical +technology too far -products & -services +Appropriate use of +medical technology -Flowers +65 -size of -business +33 -2 full-timers 2 part-timers +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” -revenue +PEW RESEARCH CENTER -€ 37,000 +The 2016 Pew Research Center survey using +different question wording found that most adults said testing on human embryos to develop gene +editing would make gene editing less acceptable for them. -location +1 The National Academies of Sciences and the National Academy of Medicine launched a consensus study and at least two international +summits to consider these issues. +2 Other research on public views of gene editing has also shown that the public distinguishes its intended purpose. Scheufele et al. (2017) +found that Americans were more supportive of gene editing applications that treat medical conditions than those aimed at enhancing health. +Similarly, Whitman (2018) also found stronger public support for gene editing to cure disease than gene editing to make people stronger or +more intelligent. -Main street - Bloomfield +www.pewresearch.org -Fleurs Flowers is a flower shop in a small town. Fleur sells -flowers to a small group of regular clients who regularly -purchase fresh flowers from her shop. Her customers generally -consist of men and women aged 25. The customers often just -bought a house and flowers are the perfect way to bring -atmosphere to the new place. In addition, she has customers -who buy flowers for birthday and holiday occasions. With -Valentine’s and Mother's Day she is working overtime to meet -the demand + 5 -Fleurs Flowers’ competitor analysis +PEW RESEARCH CENTER -A company like Fleurs Flowers has a difficult time in today's -society. She has a physical store that is open during shopping -hours. Her customers are people who also work at those times. -These customers search for convenience in their shopping. -Internet shopping is very popular amongst these people. Fleur -wants to increase her appeal to her target group. She -investigates the advantages and disadvantages of the sales -strategies of her main competitors. For her research Fleur visits -her competitors and buys their flowers. The results of Fleur’s -competitor analysis can be read in the example on the next page. +Several patterns emerge in +public opinion about gene +editing in these varying +contexts. There are large +differences in acceptance of +gene editing between the +highly religious and less +religious. In addition, there is +a gender gap in views about +gene editing, with women +less accepting of it than men. +People with higher levels of +science knowledge and +greater familiarity with gene +editing also tend to be more +accepting of it. -© InnoValor, 2018 +Religious Americans are +more likely to view gene +editing negatively - competitor analysis +Highly religious Americans are more likely to see gene +editing for babies as taking medical technology too far -Fleurs Flowers +% of U.S. adults in each religious commitment group who say changing a +baby’s genetic characteristics for each of the following reasons is … -name +Taking medical technology too far -target group +Appropriate use of medical technology -Men or women from 25 years old with an average salary +Treat a serious disease or condition the baby would have at birth +Among those ___ on the religious commitment index -fromyouflowers.com +High -Lidl +Medium -Valentino flowershop +Low -competitors +40 -most important findings +57 -Offers flowers online +27 -The price for flowers is low +18 -Customers can order -flowers online but have to -pick them up at the store +72 -Delivers at home though -whole the region within 24 -hours +82 -Customers can buy flowers -while doing groceries, they do -not have to visit a different store +Reduce risk of a serious disease or condition that could occur over their +lifetime -When ordering flowers above a -certain amount the customer -will receive a free vase +High -Package deals – get a -discount on a vase when -flowers are bought +53 -Everyday there are new flowers +Medium -A lot of time is spend on -wrapping the flowers as a gift +Low -An online shop seems to be more -and more important. Customers -want to be able to order flowers -from their home. Customers also -like it when they get something for -free or a discount on something -when they buy flowers +41 -There is no way that a -customer can see the flowers -before they are ordered +27 -Customers cannot create -their own piece since the -flowers are all prearranged +Make the baby more intelligent -The price for flowers is higher -than in other shops +46 -Ordering flowers is less -personal +58 -The flowers are made in mass +73 -Ordering flowers takes -quite some time +6 -Customers cannot create -their own bouquet of flowers +High -Buying flowers is not a special -occasion since it is been done -during doing groceries +93 -The shop is not as much open -as customers would like, for -instance it is closed on Mondays +82 -Customers seem to think that -buying flowers is a personal and -special occasion. They do not like -it when buying flowers is less -personal. They also do not like that -they cannot create their own -bouquet +72 -s -h -t -g -n -e -r -t -s +18 -s -e -s -s -e -n -k -a -e -w +Low -date +Medium + +Americans who are high in +religious commitment – that +is, those who attend religious +services at least weekly, pray +at least daily and say that +religion is very important in +their lives – are less inclined +than those with either +medium or low levels of +religious commitment to say +that gene editing is an appropriate use of medical technology.3 For example, those high in religious +commitment are closely divided over whether it is appropriate to use gene editing to reduce a + +Note: Respondents who did not give an answer are not shown. See Methodology for details +on index of religious commitment. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +28 + +3 A person is considered to have a high level of religious commitment if they attend religious services at least weekly, pray at least daily and +say religion is very important in their life. People are classified as having low commitment if they say religion is not too or not at all important +in their lives and that they seldom or never attend religious services or pray. All others are classified as having medium commitment. + +www.pewresearch.org + + 6 + +PEW RESEARCH CENTER + +baby’s risk of disease later in life; 46% say this is appropriate, while 53% consider it taking +technology too far. In contrast, roughly three-quarters of those low in religious commitment (73%) +say gene editing to reduce a baby’s risk of developing a serious disease or condition is an +appropriate use of medical technology. And, while a 57% majority of those high in religious +commitment say gene editing to treat a congenital disorder in a baby is an appropriate use of +medical technology, a much larger share of those with low religious commitment (82%) say this is +appropriate. + +Those with high levels of religiosity also stand +out when considering the possibility that +development of gene editing would entail +testing on human embryos. An overwhelming +majority of those high in religious +commitment (87%) say this would be taking +medical technology too far; just 11% of this +group says this would be appropriate. In +contrast, 55% of those low in religious +commitment say that development of gene +editing techniques that require testing on +human embryos would be an appropriate use +of medical technology. + +The differences by religiosity in views on the +appropriateness of gene editing tend to persist +even when statistically controlling for other +factors such as gender, race and ethnicity, age, +and education that are related to religious +beliefs and practices. See Appendix for details. + +Most highly religious Americans +consider testing gene editing on human +embryos to be taking technology too far + +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … + +Taking medical technology too far +Appropriate use of medical technology + +U.S. adults + +65 + +33 + +Among those ___ on the religious commitment index + +High + +87 + +Medium + +69 + +11 + +29 + +Low + +44 + +55 + +Note: Respondents who did not give an answer are not shown. See +Methodology for details on index of religious commitment. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 7 + +PEW RESEARCH CENTER + +There are also differences in views about gene +editing by religious affiliation, particularly if +such techniques would involve embryonic +testing. Just 11% of white evangelical +Protestants say that if development of gene +editing techniques required testing on human +embryos, it would be an appropriate use of +medical technology. + +By contrast, about half of the religiously +unaffiliated (52%), including 79% of atheists +and 57% of agnostics, say that embryonic +testing to develop gene editing techniques +would be an appropriate use of medical +technology. + +See the Appendix for other views about gene +editing by religious affiliation. + +White evangelicals especially likely to +say gene editing involving embryonic +testing takes technology too far + +% of U.S. adults who say that if development of gene +editing required testing on human embryos, it would +be … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate use +of medical +technology +33% + +Taking medical +technology too +far +65% + +21 +11 +31 +25 +30 +29 +27 +52 +79 +57 +42 + +34 +29 +28 + +78 +88 +67 +72 +69 +70 +73 +47 +21 +42 +57 + +64 +68 +71 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 8 + +PEW RESEARCH CENTER + +Men are more accepting than women of using gene editing to change a baby’s genetic +makeup + +Men are more inclined than +women to view gene editing as +an appropriate use of medical +technology, regardless of its +intent. About two-thirds of +men (65%) believe that gene +editing to reduce a baby’s risk +of developing a serious +disease later in life is an +appropriate use of medical +technology, compared with +54% of women. And men +(76%) are more supportive +than women (68%) of using +gene editing to treat a +congenital disorder. + +Similarly, more men (43%) +than women (24%) are +accepting of gene editing +technology if it required +embryonic testing to develop. + +Men are more likely than women to view gene editing +for babies as appropriate + +% of U.S. adults who say changing a baby’s genetic characteristics for each +of the following reasons is an appropriate use of medical technology + +Changing a baby's genetic characteristics ... + +To treat a serious disease/condition the +baby would have at birth + +Women + +Men + +68 + +76 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +54 + +65 + +To make the baby more intelligent + +13 + +24 + +Gene editing to change a baby's +genetic characteristics ... + +If it required testing on human embryos +to develop these techniques + +24 + +43 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Women tend to have higher +levels of religious commitment, on average, but differences by gender tend to hold even when +controlling for religiosity and other factors in statistical models. See the Appendix for details. + +Previous Pew Research Center surveys in 2016 and 2014, while not directly comparable due to +different polling methods and question wording, have also found that women tend to be less +accepting than men when it comes to gene editing for babies. + +www.pewresearch.org + + 9 + +PEW RESEARCH CENTER + +Americans with high levels of science knowledge are more likely to view gene editing in a +positive light + +People high in science +knowledge – based on a nine- +item index – tend to be more +accepting of using gene +editing for babies.4 Some +86% of those with high +science knowledge believe it +is appropriate to use gene +editing to treat a congenital +disorder, compared with 58% +of those with low science +knowledge. And, 71% of those +with high science knowledge +say it is appropriate to use +gene editing to reduce a +baby’s risk of disease that +would occur later in life, +compared with 49% of those +with low science knowledge.5 + +However, few Americans say +it is appropriate to use gene +editing to make a baby more +intelligent, regardless of their +level of science knowledge. + +Those with high science knowledge are more inclined +to see gene editing techniques as appropriate + +% of U.S. adults who say each of the following is an appropriate use of +medical technology + +Changing a baby's genetic +characteristics ... +To treat a serious disease/condition the +baby would have at birth + +Among those with ___ science +knowledge +Medium + +Low + +High + +58 + +86 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +49 + +71 + +To make the baby more intelligent + +Gene editing to change a baby's +genetic characteristics ... + +21 + +24 + +If it required testing on human +embryos to develop these techniques + +26 + +50 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +See Methodology for details on index of science knowledge. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Judgments on the appropriateness of gene editing also follow a similar pattern by education level, +which is closely linked with science knowledge. Those with a postgraduate degree are more + +4 See the Methodology for details on the science knowledge index. +5 These findings are in keeping with a 2017 study by Scheufele et al., which found Americans’ attitudes about gene editing varied depending +on their knowledge level about genetics and gene editing, as well as their religiosity. A 2013 study by Allum et al. in the United Kingdom found +the effect of science knowledge on support for using genetic testing to screen for serious diseases in an unborn baby was conditional on +religious service attendance. In this survey, there is a tendency for the effects of science knowledge to be conditional with levels of religiosity +in views about the use of gene editing to reduce the risk of disease and if it would require testing on human embryos. For example, among +people with high religious commitment, science knowledge is not related to views on the appropriateness of gene editing for babies to reduce +the risk of disease that could occur over their lifetime. This interaction between religious commitment and science knowledge holds even after +controlling for other factors in statistical models but does not occur for other views about gene editing. + +www.pewresearch.org + + 10 + +PEW RESEARCH CENTER + +accepting of gene editing for treating or reducing the risk of a disease than those with a high school +degree or less. + +Americans who report familiarity with gene editing are more inclined to see it as +appropriate + +People who say they have +heard at least a little about +gene editing that can be used +to change a baby’s genetic +characteristics are more +likely than those who have +heard nothing at all to say +gene editing would be +appropriate in each of the +circumstances considered in +this survey. + +A Pew Research Center +survey in 2016 found a +similar tendency for those +who report familiarity with +gene editing to express more +interest in and acceptance of +using these technologies to +enhance a baby’s health over +the course of their lifetime. + +Those familiar with gene editing are more inclined to +see different uses as appropriate + +% of U.S. adults who say each of the following is an appropriate use of +medical technology + +Changing a baby's genetic +characteristics ... + +To treat a serious disease/condition the +baby would have at birth + +Among those who have heard ___ +about gene editing +A little +71 + +Nothing at all + +A lot + +64 + +To reduce risk of a serious disease/condition +that could occur over their lifetime + +52 + +65 + +To make the baby more intelligent + +13 + +31 + +Gene editing to change a baby's +genetic characteristics ... + +If it required testing on human embryos +to develop these techniques + +24 + +49 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 11 + +PEW RESEARCH CENTER + +Although gene editing for babies is still in development, a separate April 2018 survey found that +about half of Americans (52%) believe it is likely that within the next 50 years, we will be able to +eliminate almost all birth defects by manipulating the genes of an embryo before a baby is born. In +the current survey, as people think about a future with possible widespread use of gene editing to +change a baby’s genetic makeup, more anticipate negative than positive effects on society. + +Specifically, a majority of +Americans (58%) believe +gene editing will very likely +lead to increased inequality +because it will only be +available to the wealthy. +Some 54% of Americans +anticipate a slippery slope, +saying it’s very likely that +“even if gene editing is used +appropriately in some cases, +others will use these +techniques in ways that are +morally unacceptable.” And, +46% expect it is very likely +that gene editing techniques +will be used before we fully +understand how they affect +people’s health. + +Larger shares of Americans believe negative effects of +widespread use of gene editing are very likely + +% of U.S. adults who say each of the following would be ____ to occur if gene +editing to change a baby’s genetic characteristics becomes widely available + +Very likely + +Fairly likely + +Not too likely + +Not at all likely + +Inequality will increase as it will +only be available for the wealthy + +Even if used appropriately by +some, others will use in morally +unacceptable ways + +Will be used before we fully +understand effects on health + +58 + +29 + +9 + +4 + +54 + +32 + +9 + +3 + +46 + +38 + +12 + +3 + +Will pave way for new medical +advances that benefit society + +18 + +42 + +30 + +8 + +Will help people live longer and +better quality lives + +16 + +48 + +27 + +8 + +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Far smaller shares see +positive outcomes as very +likely if gene editing for +babies were to become widely available. Roughly two-in-ten Americans (18%) say it is very likely +that development of these techniques will pave the way for new medical advances that benefit +society as a whole. Another 42% say this is fairly likely to occur, while 38% consider this not too or +not at all likely. Just 16% see the widespread use of gene editing as very likely to help people live +longer and better quality lives. About half (48%) say this is fairly likely, while 34% say this +outcome is not too or not at all likely. + +www.pewresearch.org + + 12 + +PEW RESEARCH CENTER + +Women are more inclined +than men to expect negative +effects if gene editing +becomes widely available, in +keeping with gender +differences over acceptance +of gene editing. For example, +larger shares of women than +men say this technology will +lead to an increase in +inequality as it will only be +available for the wealthy +(63% say this is very likely vs. +52% of men) or say it will be +used in morally unacceptable +ways (57% vs. 51%). And +although only minorities of +men and women see positive +effects as very likely, men are +somewhat more inclined to +anticipate positive outcomes. + +More women than men anticipate negative effects of +gene editing + +% of U.S. adults who say each of the following would be very likely to occur if +gene editing to change a baby’s genetic characteristics becomes widely +available + +POSITIVE OUTCOMES + +Women + +Men + +Will pave way for new medical +advances that benefit society + +14 + +23 + +Will help people live longer and +better quality lives + +13 + +19 + +NEGATIVE OUTCOMES + +Inequality will increase as it will +only be available for the wealthy + +Even if used appropriately by +some, others will use in morally +unacceptable ways + +52 + +63 + +51 + +57 + +Will be used before we fully +understand effects on health + +42 + +49 + +0 + +20 + +40 + +60 + +80 + +100 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Expectations also vary by +self-reported familiarity with +gene editing. Those who have +heard a lot about gene editing to change a baby’s genetic characteristics are more inclined to +anticipate positive outcomes from its widespread use. Some 36% of this group thinks it is very +likely that the widespread availability of gene editing would pave the way for new medical +advances that are beneficial to society, compared with about half as many of those who have heard +a little or nothing about gene editing (16%). + +However, those most familiar with gene editing are more likely than those with no familiarity to +also anticipate downsides. About two-thirds of those who have heard a lot about gene editing +(64%) say it is very likely that widespread availability of gene editing will increase inequality +because the technology will primarily be available only for the wealthy, compared with 53% of +those who have heard nothing about using gene editing to change a baby’s genetic characteristics. +And, 65% of those who have heard a lot about gene editing think it’s very likely that others will use + +www.pewresearch.org + + 13 + +PEW RESEARCH CENTER + +these techniques in morally unacceptable ways, while 54% of those who have heard a little and +52% of those who have heard nothing about gene editing say this. + +Americans who are more familiar with gene editing are more inclined to anticipate +positive as well as negative effects from widespread use of gene editing + +% of U.S. adults in each group who say each of the following would be very likely to occur if gene editing to change a +baby’s genetic characteristics becomes widely available + +POSITIVE OUTCOMES + +Will pave way for new medical +advances that benefit society + +Will help people live longer and +better quality lives + +NEGATIVE OUTCOMES + +Inequality will increase as it will +only be available for the wealthy +Even if used appropriately by +some, others will use in morally +unacceptable ways +Will be used before we fully +understand effects on health + +Heard a lot about +gene editing + +Heard a little about +gene editing + +Heard nothing at all +about gene editing + +36 + +34 + +16 + +13 + +16 + +14 + +64 + +65 + +59 + +54 + +53 + +52 + +54 + +46 + +43 + +Note: Respondents who gave other responses or who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +A similar pattern occurs with levels of science knowledge. A larger share of people with high +science knowledge anticipate positive outcomes from the widespread availability of gene editing +than those with low science knowledge. At the same time, those with high science knowledge are at +least equally likely as those with low science knowledge to think negative effects from the +widespread availability of gene editing are very likely. + +www.pewresearch.org + + 14 + +PEW RESEARCH CENTER + +The American public also +tends to be skeptical about +whether medical experts fully +comprehend the health +consequences of gene editing. +A 36% minority of Americans +believe that medical +researchers understand the +health effects of gene editing +for babies either very (7%) or +fairly well (29%), while 62% +say medical researchers do +not understand the health +effects at all or not too well. + +36% of Americans say medical researchers understand +the health effects of gene editing at least fairly well + +% of U.S. adults who say medical researchers understand the health risks and +benefits of changing a baby’s genetic characteristics … + +Very well + +Fairly well + +Not too well + +Not at all well + +7 + +29 + +45 + +17 + +Note: Respondents who did not give an answer are not shown. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +Familiarity with gene editing is linked with beliefs about medical researchers’ understanding of it. +About half of those who have heard a lot about gene editing for babies (51%) say that medical +researchers understand the health effects of gene editing for babies at least fairly well, compared +with three-in-ten of those who have heard nothing at all (30%). However, level of science +knowledge based on a nine-item index is not related to views on this issue; 36% of those with high +science knowledge and 40% of those with low science knowledge believe that medical researchers +understand the health effects of changing a baby’s genetic characteristics at least fairly well. + +There are no or only modest differences on this question by gender or religious commitment. + +www.pewresearch.org + + 15 + +PEW RESEARCH CENTER + +Acknowledgments + +This report is made possible by The Pew Charitable Trusts. This report is a collaborative effort +based on the input and analysis of the following individuals. Find related reports online at: +pewresearch.org/science. + +Primary research team + +Cary Funk, Director, Science and Society Research +Brian Kennedy, Research Associate +Meg Hefferon, Research Assistant +Mark Strauss, Writer/Editor +Isadora Milanez, Intern + +Editorial and graphic design + +Mark Strauss, Writer/Editor +Shannon Greenwood, Copy Editor +Margaret Porteus, Information Graphics Designer + +Communications and web publishing + +Haley Nolan, Communications Assistant +Sara Atske, Assistant Digital Producer + +www.pewresearch.org + + 16 + +PEW RESEARCH CENTER + +Methodology + +Margins of error + +This report is drawn from a survey conducted as +part of the American Trends Panel (ATP), a +nationally representative panel of randomly +selected U.S. adults living in households +recruited from landline and cellphone random- +digit-dial (RDD) surveys. Panelists participate +via monthly self-administered web surveys. +Panelists who do not have internet access are +provided a tablet and wireless internet +connection. The panel, which was created by +Pew Research Center, is being managed by GfK. + +Data in this report are drawn from the panel +wave conducted April 23-May 6, 2018, among +2,537 respondents. The margin of sampling error +for the full sample of 2,537 respondents is plus +or minus 2.8 percentage points. + +Members of the ATP were recruited from several +large, national landline and cellphone RDD +surveys conducted in English and Spanish. At +the end of each survey, respondents were invited +to join the panel. The first group of panelists was +recruited from the 2014 Political Polarization +and Typology Survey, conducted Jan. 23-March +16, 2014. Of the 10,013 adults interviewed, 9,809 +were invited to take part in the panel and a total +of 5,338 agreed to participate.6 + +Sample size +2,537 + +Margin of error +in percentage +points ++/- 2.8 + +U.S. adults + +Men +Women + +Race/ethnicity +White +Black +Hispanic + +Religious commitment index +High +Medium +Low + +Religious affiliation +NET Protestant + +White evangelical +White mainline +Black Protestant + +NET Catholic + +White Catholic +Hispanic Catholic + +NET Unaffiliated +Atheist +Agnostic +Nothing in particular + +1,272 +1,265 + +1,519 +412 +421 + +440 +1,291 +566 + +1,086 +349 +282 +283 +459 +254 +168 +703 +146 +158 +399 + ++/- 4.0 ++/- 3.8 + ++/- 3.3 ++/- 7.8 ++/-7.6 + ++/- 6.7 ++/- 3.8 ++/- 5.7 + ++/- 4.2 ++/- 7.1 ++/- 7.7 ++/- 9.6 ++/- 6.5 ++/- 8.0 ++/-11.8 ++/-5.1 ++/- 11.2 ++/- 10.2 ++/- 6.8 + +Note: The margins of error are reported at the 95% level of +confidence and are calculated by taking into account the average +design effect for each subgroup. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +6 When data collection for the 2014 Political Polarization and Typology Survey began, non-internet users were subsampled at a rate of 25%, +but a decision was made shortly thereafter to invite all non-internet users to join. In total, 83% of non-internet users were invited to join the +panel. + +www.pewresearch.org + + 17 + +PEW RESEARCH CENTER + +The second group of panelists was recruited +from the 2015 Pew Research Center Survey on +Government conducted Aug. 27-Oct. 4, 2015. Of +the 6,004 adults interviewed, all were invited to +join the panel, and 2,976 agreed to participate.7 +The third group of panelists was recruited from a +survey conducted April 25-June 4, 2017. Of the +5,012 adults interviewed in the survey or pretest, +3,905 were invited to take part in the panel and a +total of 1,628 agreed to participate.8 + +The overall target population for Wave 34 was +non-institutionalized persons ages 18 and older, +living in the United States, including Alaska and +Hawaii. The sample for Wave 34 consisted of +3,099 ATP members that were invited to Wave +33 and were still active. This subsample was +selected using the following approach: + +Margins of error continued + +Sample size + +Margin of error +in percentage +points + +Science knowledge index +High +Medium +Low + +Heard about gene editing +A lot +A little +Nothing at all + +679 +1,274 +584 + +326 +1,449 +753 + ++/-5.2 ++/- 3.9 ++/- 5.7 + ++/- 7.9 ++/-3.6 ++/-5.1 + +Note: The margins of error are reported at the 95% level of +confidence and are calculated by taking into account the average +design effect for each subgroup. +Source: Survey conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +1. Panelists were grouped into three strata based on how underrepresented they are + +demographically. Then we analyzed response rates to the last five panel survey waves +(W28-32) to project the number of panelists in each stratum who would respond to the +W33 survey. + +2. We then determined how many panelists we wanted to sample from each stratum in W33 +in order to finish with around 2,500 completed interviews and have a responding sample +that is as representative as possible. + +▪ Stratum A consists of panelists who are non-internet users, are black non-Hispanic, +are Hispanic, or have high school or less education. There were 1,819 total panelists +in this stratum and they are sampled at a rate of 100% for W33. 1,806 were active +panelists. + +7 Respondents to the 2014 Political Polarization and Typology Survey who indicated that they are internet users but refused to provide an +email address were initially permitted to participate in the American Trends Panel by mail but were no longer permitted to join the panel after +Feb. 6, 2014. Internet users from the 2015 Pew Research Center Survey on Government who refused to provide an email address were not +permitted to join the panel. +8 White, non-Hispanic college graduates were subsampled at a rate of 50%. + +www.pewresearch.org + + 18 + +PEW RESEARCH CENTER + +▪ Stratum B consists of panelists who are ages 18 to 34 or are non-volunteers. The + +1,684 total panelists in this stratum are subsampled at a rate of 63%, yielding 1,061 +sampled for W33 (1,057 were active). + +▪ Stratum C consists of the remaining 2,009 panelists not in stratum A or B. This + +group is subsampled at a rate of 12%, yielding 241 panelists sampled for W33 (239 +were active). + +The ATP data were weighted in a multistep process that begins with a base weight incorporating +the respondents’ original survey selection probability and the fact that in 2014 some panelists were +subsampled for invitation to the panel. Next, an adjustment was made for the fact that the +propensity to join the panel and remain an active panelist varied across different groups in the +sample. The final step in the weighting uses an iterative technique that aligns the sample to +population benchmarks on a number of dimensions. Gender, age, education, race, Hispanic origin +and region parameters come from the U.S. Census Bureau’s 2016 American Community Survey. +The county-level population density parameter (deciles) comes from the 2010 U.S. decennial +census. The telephone service benchmark comes from the July-December 2016 National Health +Interview Survey and is projected to 2017. The volunteerism benchmark comes from the 2015 +Current Population Survey Volunteer Supplement. The party affiliation benchmark is the average +of the three most recent Pew Research Center general public telephone surveys. The internet +access benchmark comes from the 2017 ATP Panel Refresh Survey. Respondents who did not +previously have internet access are treated as not having internet access for weighting purposes. +Sampling errors and statistical tests of significance take into account the effect of weighting. +Interviews are conducted in both English and Spanish, but the Hispanic sample in the ATP is +predominantly U.S. born and English speaking. + +Margins of error tables shown here provide the unweighted sample sizes and the error attributable +to sampling that would be expected at the 95% level of confidence for different groups in the +survey taking into account the average design effect for each subgroup. Sample sizes and sampling +errors for other subgroups are available upon request. + +In addition to sampling error, one should bear in mind that question wording and practical +difficulties in conducting surveys can introduce error or bias into the findings of opinion polls. + +The April 2018 wave had a response rate of 82% (2,537 responses among 3,099 individuals in the +panel). Taking account of the combined, weighted response rate for the recruitment surveys + +www.pewresearch.org + + 19 + +PEW RESEARCH CENTER + +(10.1%) and attrition from panel members who were removed at their request or for inactivity, the +cumulative response rate for the wave is 2.3%.9 + +Religious commitment index + +Survey respondents were classified into high, medium and low levels of religious commitment +based on three indicators: frequency of religious service attendance, self-reported importance of +religion in their lives and frequency of prayer. Those who attend worship services at least weekly, +pray at least once a day and say religion is very important in their lives are classified as high in +religious commitment. Those low in commitment say religion is not too or not at all important in +their lives, that they seldom or never attend worship services, and seldom or never pray. All others +are classified as exhibiting a medium level of religious commitment. + +9 Approximately once per year, panelists who have not participated in multiple consecutive waves are removed from the panel. These cases +are counted in the denominator of cumulative response rates. Note that for the March 2018 survey, we calculated the response rates by +computing the mean rates for the subsampled respondents (based on the rates from the recruitment survey they joined the panel on). + +www.pewresearch.org + + 20 + +PEW RESEARCH CENTER + +Science knowledge index + +Scale reliability and factor analysis + +The Pew Research Center survey included a set of nine questions to tap public knowledge of +science across a range of principles and topics. Most respondents (n=1,901) completed these +questions as part of a previous +wave of the American Trends +Panel, conducted May 10-June +6, 2016. Respondents who did +not participate in the previous +wave answered these questions +in this survey (n=636). The +following shows the +measurement properties of the +index for the combined set of +2,537 respondents.10 + +Alpha for scale +0.75 +Alpha if item is +dropped + +Knowledge index 9-items + +Item-rest +correlation + +0.47 + +0.39 + +0.74 + +0.73 + +Common +variance +explained by +first factor +76% +Factor +loadings + +0.46 + +0.55 + +KNOSCT22. Use of a control group +to determine whether a new drug is +effective +KNOSCT23. Carbon dioxide is made +as a consequence of burning fossil +fuels +KNOSCT27. The probability of an +old-bridge collapsing after a period +of time +KNOSCT28. Only bacterial +infections can be treated effectively +by antibiotic medications +KNOSCT29. The use of a control +“sugar pill” in a new drug trial is to +rule out a possible placebo effect +KNOSCT31. The health benefits +occurring when most people in a +population get a vaccine is called +herd immunity +KNOSCT32. An apple, salmon, corn +and a mosquito can all be +genetically modified. +KNOSCT33. Humans and mice +share 50% or more of the same +genetic makeup +KNOSCT34. Nitrogen makes up +most of the Earth’s atmosphere. + +0.50 + +0.72 + +0.59 + +0.31 + +0.75 + +0.35 + +0.43 + +0.73 + +0.50 + +0.45 + +0.73 + +0.52 + +0.47 + +0.73 + +0.55 + +An exploratory factor analysis +finds one common factor +explaining 76% of the common +variance in the items. The +factor loadings show that each +of the nine questions is +moderately correlated with the +common factor. These +indicators suggest a set of items is measuring a single underlying dimension. + +PEW RESEARCH CENTER + +0.44 + +0.43 + +Source: Surveys conducted May 10-June 6, 2016, and April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +0.73 + +0.50 + +0.73 + +0.52 + +As shown in the accompanying +table, the internal reliability or +consistency of the scale as +measured by Cronbach’s alpha +is 0.75. Each of the items in the +scale is at least moderately +correlated with the other items. + +10 In statistical models controlling for demographics and education, there were no significant differences between respondents who +completed the science knowledge items in 2016 (W17) and those who completed them in 2018 (W34) on either the overall index measure or +the individual questions. + +www.pewresearch.org + + 21 + +PEW RESEARCH CENTER + +Two parameter item response theory analysis + +KNOSCT22. Use of a control +group to determine whether a +new drug is effective +KNOSCT23. Carbon dioxide is +made as a consequence of +burning fossil fuels +KNOSCT27. The probability of an +old-bridge collapsing after a +period of time +KNOSCT28. Only bacterial +infections can be treated +effectively by antibiotic +medications +KNOSCT29. The use of a control +“sugar pill” in a new drug trial is +to rule out a possible placebo +effect +KNOSCT31. The health benefits +occurring when most people in a +population get a vaccine is called +herd immunity +KNOSCT32. An apple, salmon, +corn and a mosquito can all be +genetically modified. +KNOSCT33. Humans and mice +share 50% or more of the same +genetic make-up +KNOSCT34. Nitrogen makes up +most of the Earth’s atmosphere. + +% Correct Discrimination + +Difficulty + +68 + +72 + +62 + +46 + +1.36 + +-0.73 + +1.74 + +-0.82 + +2.14 + +-0.38 + +0.81 + +0.25 + +59 + +1.60 + +-0.33 + +34 + +39 + +34 + +1.39 + +0.65 + +1.66 + +0.38 + +1.37 + +0.66 + +Note that each of the science +knowledge questions are coded +as binary variables +(correct/incorrect). Both +Cronbach’s alpha reliability +analysis and the factor analysis +are based on a Pearson’s +correlation matrix. Pearson +correlations with binary +variables are restricted to a +limited range, underestimating +the association between two +variables when compared with +tetrachoric correlations. We do +not anticipate that the use of a +Pearson’s correlation matrix +affects the unidimensional +factor solution for the scale, +however. + +31 + +1.55 + +0.74 + +PEW RESEARCH CENTER + +Source: Surveys conducted May 10-June 6, 2016, and April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +We also ran an item-response +theory analysis (IRT) to check +how well each question +distinguishes between those +who know relatively more or +less on the scale. This analysis fits a two-parameter logistic model, allowing discrimination and +difficulty to vary across the items. Discrimination shows the ability of the question to distinguish +between those with higher and lower science knowledge. Difficulty shows how easy or hard each +question is for the average respondent. We did not include a guessing parameter in the model; the +questionnaire offered respondents an explicit option of not sure on the survey. + +The results show variation in difficulty across the items. The easiest item required respondents to +identify that carbon dioxide as the gas that is made as a consequence of burning fossil fuels, while +the most difficult item required respondents to identify nitrogen as the gas that makes up most of +the Earth’s atmosphere. + +www.pewresearch.org + + 22 + +PEW RESEARCH CENTER + +Most of the questions also discriminate between those with higher and lower science knowledge. +The item with strongest ability to discriminate was the question asking respondents to calculate +the conditional probability of an old bridge over time. The question with the weakest ability to +discriminate was the question about the effectiveness of antibiotics to treat bacterial, but not other +kinds of infections. + +The test information curve mirrors a normal curve centered around zero, suggesting that the +science knowledge index provides the most information about Americans near the mean level of +knowledge. + +© Pew Research Center, 2018 + +www.pewresearch.org + + 23 + +PEW RESEARCH CENTER + +Appendix: Detailed tables + +Half or more from all religious groups +consider gene editing to treat disease +that would be present at birth to be +appropriate + +% of U.S. adults who say using gene editing to treat a +serious disease or condition a baby would have at birth +is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +72% + +Taking medical +technology too +far +27% + +66 +61 +79 +60 +72 +72 +74 +78 +92 +92 +69 + +74 +62 +74 + +31 +37 +20 +36 +28 +28 +26 +22 +7 +8 +30 + +24 +35 +26 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 24 + +PEW RESEARCH CENTER + +White evangelical Protestants among +the least likely to consider gene editing +to reduce lifetime disease risk +appropriate + +% of U.S. adults who say using gene editing to reduce the +risk of a serious disease or condition that could occur +over the course of a baby’s lifetime is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +60% + +Taking medical +technology too +far +38% + +51 +45 +58 +51 +63 +58 +68 +67 +85 +81 +57 + +60 +52 +61 + +47 +54 +41 +43 +36 +41 +30 +32 +15 +19 +41 + +39 +44 +36 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 25 + +PEW RESEARCH CENTER + +Nine-in-ten white evangelical +Protestants see gene editing for +intelligence as taking technology too far + +% of U.S. adults who say using gene editing to make a +baby more intelligent is … + +U.S. adults +Religious affiliation +Protestant + +White evangelical +White mainline +Black Protestant + +Catholic + +White Catholic +Hispanic Catholic + +Unaffiliated +Atheist +Agnostic +Nothing in particular + +Race/ethnicity +White +Black +Hispanic + +Appropriate +use of medical +technology +19% + +Taking medical +technology too +far +80% + +12 +8 +12 +14 +20 +18 +24 +25 +36 +22 +23 + +16 +21 +22 + +86 +91 +88 +81 +80 +82 +76 +74 +64 +77 +77 + +83 +75 +76 + +Note: Whites and blacks include only non-Hispanics; Hispanics are +of any race. Respondents who did not give an answer are not +shown. +Source: Survey of U.S. adults conducted April 23-May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would +Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 26 + +PEW RESEARCH CENTER + +Statistical models predicting opinions about appropriate uses of gene editing + +Difference in predicted probabilities for belief that changing a baby’s genetic characteristics ____ is an appropriate use of +medical technology + +Women + +To treat a serious disease or +condition the baby would +have at birth +NS + +To reduce the risk of a serious +disease or condition that could occur +over the course of his or her lifetime +-0.08 + +To make the baby +more intelligent +-0.09 + +Requiring testing on +human embryos +-0.15 + +Race Reference category: White, non-Hispanic + +Black, non-Hispanic + +Hispanic + +Other + +Age Reference category: 18-29 + +Ages 30-49 + +50-64 + +65+ + +NS + +NS + +-0.20 + +NS + +-0.10 + +NS + +Education Reference category: High school or less + +Postgraduate + +College graduate + +Some college + ++0.16 + ++0.14 + ++0.09 + +Science knowledge Reference category: Low science knowledge + +High + +Medium + ++0.16 + +NS + +NS + +NS + +NS + +-0.12 + +-0.17 + +NS + ++0.13 + +NS + ++0.08 + +NS + +NS + +Religious commitment Reference category: Low religious commitment + +High + +Medium + +-0.14 + +NS + +Religious affiliation Reference category: Unaffiliated + +Mainline Protestant + +Evangelical Protestant + +Catholic + +Other Christian + +Other religion + +NS + +NS + +NS + +NS + +NS + +-0.21 + +-0.10 + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +-0.08 + +-0.12 + +-0.13 + +NS + +NS + +NS + +NS + +-0.09 + +-0.14 + +NS + +NS + +NS + +NS + +NS + +NS + +NS + +-0.09 + +NS + +NS + +NS + +-0.12 + ++0.10 + +NS + ++0.07 + +NS + +-0.08 + +-0.14 + +-0.13 + +NS + +-0.22 + +NS + +NS + +NS + +Heard about gene editing Reference category: Heard nothing at all about gene editing to change a baby’s genetic characteristics + +Heard a lot + +Heard a little + +Number of respondents + +NS + ++0.06 + +2,237 + +NS + +NS + +2,232 + ++0.15 + ++0.07 + +2,241 + ++0.17 + ++0.09 + +2,223 + +Note: Figures shown are the difference in predicted probabilities with the reference category while other factors are held at their mean using binary +logistic regression models. Positive and negative values indicated the direction of effects. NS indicates not statistically significant based on a two- +tailed p value < 0.05. +Source: Survey conducted April 23- May 6, 2018. +“Public Views of Gene Editing for Babies Depend on How It Would Be Used” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 27 + +PEW RESEARCH CENTER + +Survey questionnaire and topline + +2018 PEW RESEARCH CENTER’S AMERICAN TRENDS PANEL +APRIL 23-MAY 6, 2018 +TOTAL N=2,537 + +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED OR HELD FOR FUTURE RELEASE + +ASK ALL: +MED3 How much, if anything, have you heard or read about gene editing that can be used to change a + +baby’s genetic characteristics? + +April 23- +May 6, +2018 +12 +56 +32 +1 + +A lot +A little +Nothing at all +No answer + +ASK ALL: +MED4 Do you think each of the following is an appropriate use of medical technology or is it taking + +medical technology too far? [RANDOMIZE ITEMS] + +An appropriate use of +medical technology + +Taking medical +technology too far + +No +Answer + +a. Changing a baby's genetic characteristics + +to make the baby more intelligent + +April 23-May 6, 2018 + +b. Changing a baby's genetic characteristics + +to TREAT a serious disease or condition the +baby would have at birth +April 23-May 6, 2018 + +c. Changing a baby's genetic characteristics +to REDUCE THE RISK of a serious disease +or condition that could occur over the +course of his or her lifetime + +19 + +72 + +80 + +27 + +April 23-May 6, 2018 + +60 + +38 + +1 + +2 + +2 + +TRENDS FOR COMPARISON +Pew Research Center survey conducted by telephone: Would you say that changing a baby’s genetic +characteristics to make the baby more intelligent is making appropriate use of medical advances OR is it +taking medical advances too far? + +Aug. 15-25, +2014 +15 +83 +2 + +Appropriate use of medical advances +Taking medical advances too far +Don’t know/Refused (VOL.) + +www.pewresearch.org + + 28 + +PEW RESEARCH CENTER + +Surveys conducted by telephone: Would you say that changing a baby’s genetic characteristics to reduce +the risk of serious diseases is making appropriate use of medical advances OR is it taking medical +advances too far? + +Appropriate use of medical advances +Taking medical advances too far +Don’t know/Refused (VOL.) + +Pew Research Center +Aug. 15-25, +2014 +46 +50 +4 + +VCU Life Sciences +Survey +Sept. 3-26, +2003 +41 +54 +6 + +ASK ALL: +MED5 Thinking about what you have heard or read, how well do you think medical researchers + +understand the health risks and benefits of changing a baby’s genetic characteristics? + +April 23- +May 6, +2018 +7 +29 +45 +17 +1 + +Very well +Fairly well +Not too well +Not at all well +No answer + +ASK ALL: +MED6 Thinking about society as a whole … + +How likely, if at all, do you think each of the following would be to occur if gene editing to +change a baby’s genetic characteristics becomes widely available? [RANDOMIZE ITEMS] + +Very likely + +Fairly +likely + +Not too +likely + +Not at all +likely + +No +Answer + +a. Even if gene editing is used + +appropriately in some cases, others +will use these techniques in ways +that are morally unacceptable + +April 23-May 6, 2018 + +54 + +32 + +9 + +b. These techniques will help people +live longer and better quality lives + +April 23-May 6, 2018 + +16 + +48 + +27 + +c. We will use these techniques before +we fully understand how they affect +people’s health + +April 23-May 6, 2018 + +46 + +38 + +12 + +d. + +Inequality will increase because +this option will be available only for +the wealthy + +April 23-May 6, 2018 + +58 + +29 + +9 + +3 + +8 + +3 + +4 + +1 + +2 + +1 + +1 + +www.pewresearch.org + + 29 + +PEW RESEARCH CENTER + +Very likely + +Fairly +likely + +Not too +likely + +Not at all +likely + +No +Answer + +MED6 continued … + +e. Development of these techniques +will pave the way for new medical +advances that benefit society as a +whole + +April 23-May 6, 2018 + +18 + +42 + +30 + +8 + +1 + +ASK ALL: +MED7 If gene editing to change a baby’s genetic characteristics required testing on human embryos in + +order to develop these techniques, do you think this would be … + +April 23- +May 6, +2018 +33 +65 +2 + +An appropriate use of medical technology +Taking medical technology too far +No answer + +ADDITIONAL QUESTIONS HELD FOR FURTHER RELEASE + +ASK ALL: +BIO15 Have you seen a health care provider FOR AN ILLNESS OR MEDICAL CONDITION in the past 12 + +months, or not? + +April 23- +May 6, +2018 +64 +36 +1 + +Yes +No +No answer + +May 10- +June 6 +2016 +63 +36 +2 + +ASK ALL: +G1 + +Do you, or does anyone in your immediate family, have a gene that predisposes you to a serious +disease such as Alzheimer’s, cancer, heart disease or sickle cell anemia? + +April 23- +May 6, +2018 +26 +40 +34 +1 + +Yes +No +Not sure +No answer + +May 10- +June 6, +2016 +27 +43 +30 +1 + +www.pewresearch.org + + 30 + +PEW RESEARCH CENTER + +ASK ALL: +G2 + +Have you, or has anyone in your immediate family, ever had a genetic test, or haven’t you done +this? + +April 23- +May 6, +2018 +14 +65 +20 +1 + +Yes +No +Not sure +No answer + +May 10- +June 6 +2016 +10 +70 +19 +1 + +NOTE: RESULTS SHOWN FOR KNOSCT22-KNOSCT34 INCLUDE 636 RESPONDENTS WHO +ANSWERED THESE QUESTIONS APRIL 23-MAY 6, 2018 COMBINED WITH 1,901 RESPONDENTS +WHO ANSWERED THESE QUESTIONS IN A PREVIOUS WAVE FIELDED MAY 10-JUNE 6, 2016. + +ASK ALL: Here’s a different kind of question. (If you don’t know the answer, select “Not sure.”) As far as +you know … + +KNOSCT22 + +Here’s a different kind of question. (If you don’t know the answer, select “Not sure.”) As +far as you know ... + +Which is the better way to determine whether a new drug is effective in treating a +disease? If a scientist has a group of 1,000 volunteers with the disease to study, should +she … [RANDOMIZE OPTIONS 1 AND 2] + +Apr 23- +May 6 +2018 +68 + +32 + 14 + 18 +<1 + +Give the drug to half of them but not to the other half, and compare +how many in each group get better (Correct) +NET Incorrect/Not sure/No answer + Give the drug to all of them and see how many get better + Not sure + No answer + +[RANDOMIZE ITEMS KNOSCT23 TO KNOSCT34; KNOSCT22 ALWAYS FIRST] + +www.pewresearch.org + + 31 + +PEW RESEARCH CENTER + +Which gas is made as a consequence of burning fossil fuels? Is it … [RANDOMIZE +OPTIONS 1-4]11 + +Carbon dioxide (Correct) +NET Incorrect/Not sure/No answer + Hydrogen + Helium + Radon + Not sure + No answer + +ASK ALL: +KNOSCT23 + +Apr 23- +May 6 +2018 +72 +28 + 4 + 1 + 3 + 20 +<1 + +NO QUESTION KNOSCT24, KNOSCT25 AND KNOSCT26 + +ASK ALL: +KNOSCT27 + +If the chances that an old bridge will collapse starts at 1% in week 1 and doubles each +week (as shown below), what is the chance that the old bridge will collapse during week +7? + +Chances the bridge will collapse is… + +1% +2% +4% +8% + +at Week 1 +at Week 2 +at Week 3 +at Week 4 + +Enter the % chance that the bridge will collapse at Week 7 (if the bridge is still standing +after Week 6)12 + +April 23- +May 6, +2018 +62 +38 + 20 + 17 + 1 + +64% (Correct) +NET Incorrect/Not sure/No answer + All other numeric responses + Not sure + No answer + +11 In the Wave 17 May 2016 survey, the question had a minor wording difference: “What gas is made as a consequence of burning fossil +fuels? Is it …” +12 The phrase “(if the bridge is still standing after Week 6)” was added for clarity in the April 23-May 6, 2018, wave. + +www.pewresearch.org + + 32 + +PEW RESEARCH CENTER + +Which of the following conditions can be treated effectively by antibiotic medications? + +[Check all that apply] [RANDOMIZE ITEMS WITH ITEMS e AND f ALWAYS LAST] + +Bacterial infection only (Correct) +NET Incorrect/Not sure/No answer + +ASK ALL: +KNOSCT28 + +April 23- +May 6, +2018 +46 +54 + +KNOSCT28 INDIVIDUAL ITEM RESPONSES + +a. Viral infections (such as a cold) +b. Fungal infections (such as athlete’s foot) +c. Bacterial infections (such as strep throat infections) +d. Allergic reactions to insect bites +e. None of these [EXCLUSIVE PUNCH] +f. Not sure [EXCLUSIVE PUNCH] + +Selected +23 +28 +83 +17 +2 +9 + +Not selected +/No answer +77 +72 +17 +83 +98 +91 + +ASK ALL: +KNOSCT29 + +April 23- +May 6, +2018 +59 +41 + 4 + 16 + 20 + 1 + +If a scientist wants to determine if a new drug is effective at treating high blood pressure +by giving half of a group of 1,000 volunteers a new medication and the other half a +“sugar pill” she wants to rule out … [RANDOMIZE OPTIONS 1-3] + +A placebo effect (Correct) +NET Incorrect/Not sure/No answer + A third person effect + A false consensus effect + Not sure + No answer + +NO QUESTION KNOSCT30 + +ASK ALL: +KNOSCT31 + +April 23- +May 6, +2018 +34 +66 + 8 + 35 + 23 + 1 + +Which of these terms refers to health benefits occurring when most people in a +population get a vaccine? [RANDOMIZE OPTIONS 1-3] + +Herd immunity (Correct) +NET Incorrect/Not sure/No answer + Population control + Vaccination rate + Not sure + No answer + +www.pewresearch.org + + 33 + +PEW RESEARCH CENTER + +ASK ALL: +KNOSCT32 + +Which of the following can be genetically modified? + +[Check all that apply] [RANDOMIZE ITEMS WITH ITEMS e AND f ALWAYS LAST] + +April 23- +May 6, +2018 +39 +61 + +Selected all (Correct) +NET Incorrect/Not sure/No answer + +KNOSCT32 INDIVIDUAL ITEM RESPONSES + +a. An apple +b. Salmon +c. A mosquito +d. Corn +e. None of these [EXCLUSIVE PUNCH] +f. Not sure [EXCLUSIVE PUNCH] + +Selected +61 +53 +46 +70 +2 +20 + +Not selected +/No answer +39 +47 +54 +30 +98 +80 + +ASK ALL: +KNOSCT33 + +April 23- +May 6, +2018 +34 +66 + 8 + 14 + 44 + 1 + +ASK ALL: +KNOSCT34 + +April 23- +May 6, +2018 +31 +69 + 9 + 9 + 35 + 15 + 1 + +Humans and mice share the same genetic makeup by… [RANDOMIZE ORDER LOW TO +HIGH; HIGH TO LOW with NOT SURE ALWAYS LAST] + +About 50% or more (Correct) +NET Incorrect/Not sure/No answer + Less than 10% + Between 11% and 49% + Not sure + No answer + +Which gas makes up most of the Earth's atmosphere? [RANDOMIZE OPTIONS 1-4] + +Nitrogen (Correct) +NET Incorrect/Not sure/No answer + Hydrogen + Carbon dioxide + +Oxygen + Not sure + No answer + +www.pewresearch.org + + 34 + +PEW RESEARCH CENTER + +TOTAL NUMBER CORRECT KNOSCT22 THROUGH KNOSCT34: + +April 23- +May 6, +2018 +4 +10 +10 +13 +12 +13 +11 +12 +9 +6 + +9 out of 9 +8 out of 9 +7 out of 9 +6 out of 9 +5 out of 9 +4 out of 9 +3 out of 9 +2 out of 9 +1 out of 9 +0 out of 9 + +24 +49 +26 + +High science knowledge (7-9 correct) +Medium science knowledge (3-6 correct) +Low science knowledge (0-2 correct) + +www.pewresearch.org diff --git a/q128/random_k4/random_2.md b/q128/random_k4/random_2.md index 5d5d4bf748b05f616c8399e8fe70690f3bccb9c7..7e875d55d5f93fef49815a4921bd61abaa902c5b 100644 --- a/q128/random_k4/random_2.md +++ b/q128/random_k4/random_2.md @@ -1,1678 +1,1736 @@ -| | | Learning | | Fine-Grained | | Grounded | Citations | | for | | | -| --- | --- | -------- | ---------- | ------------ | ----- | -------- | --------- | --- | --- | --- | --- | -| | | | Attributed | | Large | Language | Models | | | | | -LeiHuang1,XiaochengFeng1,2*,WeitaoMa1,YuxuanGu1,WeihongZhong1,XiachongFeng1 -WeijiangYu3,WeihuaPeng3,DuyuTang3,DandanTu3,BingQin1,2 -1HarbinInstituteofTechnology,Harbin,China -| | | 2 PengChengLaboratory | | | | 3HuaweiInc.,Shenzhen,China | | | | | | -| --- | --- | --------------------- | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | -{lhuang,xcfeng,wtma,yxgu,whzhong,xiachongfeng,qinb}@ir.hit.edu.cn -{weijiangyu8,pengwh.hit}@gmail.com,{tangduyu,tudandan}@huawei.com -Abstract -Why is it bad to eat cookie dough for risk of salmonella but things -like Cookie Dough Bites are ok? -| Despite | the | impressive | performance | | on | | | | | | | -| ------- | --- | ---------- | ----------- | --- | --- | -------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Salmonella is most commonly caused by eating undercooked | | | | | 1 | -information-seekingtasks,largelanguagemod- or raw foods like eggs or meat. (...) You know how your mom -4202 guA 8 ]LC.sc[ 1v86540.8042:viXra els (LLMs) still struggle with hallucinations. always warned you not to eat raw cookie dough? This is why. -| Attributed | LLMs, | which | augment | generated | | | | | | | | -| ---------- | ----- | ----- | ------- | --------- | --- | --------------------------------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | Food safety recommendations encourage people to cook eggs until | | | | | 2 | -text within-line citations, haveshownpoten- the white and yolk are firm in order to kill any bacteria. (...) However, -tial in mitigating hallucinations and improv- anyone making cookies can do things to reduce this risk by -using pasteurized egg products. -| ingverifiability. | | However,currentapproaches | | | | | | | | | | -| ----------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -suffer from suboptimal citation quality due Raw cookie dough can carry the risk of salmonella contamination, which can -to their reliance on in-context learning. Fur- lead to foodborne illness if the dough is eaten raw  1  . However, products -like Cookie Dough Bites are generally safe because they are made with -| thermore, | the | practice | of citing | only | coarse | pasteurized eggs  1    2  . | | | | | | -| --------- | --- | -------- | --------- | ---- | ------ | --------------------------- | --- | --- | --- | --- | --- | -documentidentifiersmakesitchallengingfor -Difficult to verify and subject to -userstoperformfine-grainedverification. In hallucinations and citation errors -| this work, | we | introduce | FRONT, | a | training | | | | | | | -| ---------- | --- | --------- | ------ | --- | -------- | --- | --- | --- | --- | --- | --- | - G  R  O   U  N  D  I N  G     1  salmonella is most commonly caused by eating -framework designed to teach LLMs to gen- undercooked or raw foods like eggs or meat  2  Anyone making cookies -can do things to reduce this risk by using pasteurized egg products. -erateFine-gRainedgrOuNdedciTations. By  A  N  S  W   E  R    Eating raw cookie dough can be risky due to salmonella, -groundingmodeloutputsinfine-grainedsup- commonly found in raw eggs  1 . However, this concern is mitigated in -products like Cookie Dough Bites, as the products use pasteurized egg -portingquotes,thesequotesguidethegenera- -products, which are safe to eat raw 2  . -tionofgroundedandconsistentresponses,not -onlyimprovingcitationqualitybutalsofacili- Interpretable and accurately cited -tatingfine-grainedverification.Experimentson -| | | | | | | Figure1: | Comparedwiththecurrentattributedsystems, | | | | | -| --- | --- | --- | --- | --- | --- | -------- | ---------------------------------------- | --- | --- | --- | --- | -theALCEbenchmarkdemonstratetheefficacy -thecoreideabehindFRONTistofirstselectthesupport- -ofFRONTingeneratingsuperiorgroundedre- -| | | | | | | ing quotes | from | retrieved | sources and | then | condition | -| ------------------------------------ | --- | --- | --- | --- | ---- | ---------- | ---- | --------- | ----------- | ---- | --------- | -| sponsesandhighlysupportivecitations. | | | | | With | | | | | | | -thegenerationprocessonthem,ensuringgroundedre- -LLaMA-2-7B,theframeworksignificantlyout- -sponsesandaccuratecitations. -| performsallthebaselines, | | | achievinganaver- | | | | | | | | | -| ------------------------ | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -ageof14.21%improvementincitationquality -acrossalldatasets,evensurpassingChatGPT1. SuchprevalenceofhallucinationsinLLMout- -| | | | | | | puts has | motivated | the | development | of | attributed | -| --- | --- | --- | --- | --- | --- | -------- | --------- | --- | ----------- | --- | ---------- | -1 Introduction -| | | | | | | systems | (Nakano | et al., | 2021; Thoppilan | | et al., | -| --- | --- | --- | --- | --- | --- | ------- | ------- | ------- | --------------- | --- | ------- | -Therecentadventoflargelanguagemodels(LLMs) 2022;Menicketal.,2022),suchasNewBing2 and -(Touvronetal.,2023;OpenAI,2023;Zhaoetal., Perplexity3,whereLLMsareallowedtogenerate -2023) has taken the world by storm, fueling a responses with in-line citations. Not only does it -| paradigm | shift in | information | | acquisition | (Zhu | | | | | | | -| -------- | -------- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | -improvefactualityandalleviatehallucinations,but -etal.,2023). Despitetheircompellingperformance, italsosimplifiesuserverificationofmodeloutputs, -LLMs still struggle with hallucinations (Ji et al., furtherenhancingtheverifiabilityofLLMs. -2023;Huangetal.,2023),atendencytofabricate Despiterecentadvancements,currentattributed -| non-existent | facts | or generate | unfaithful | | content. | | | | | | | -| ------------ | ----- | ----------- | ---------- | --- | -------- | -------------------------------------- | --- | --- | --- | --- | -------- | -| | | | | | | LLMsstillexposesignificantlimitations. | | | | | Firstly, | -This issue further poses a risk of misinformation current approaches predominantly rely on either -dissemination(ChenandShu,2023),directlyim- in-contextlearning(Gaoetal.,2023b)orpost-hoc -pactingthereliabilityandtrustworthinessofLLMs. -retrieval(Gaoetal.,2023a)toachieveattribution, -*CorrespondingAuthor -1Our data and code can be found at: https://github. 2https://www.bing.com/chat -com/LuckyyySTA/Fine-grained-Attribution. 3https://www.perplexity.ai - -| | | | | | | | 2 RelatedWork | | | | | | | -| ------- | ----------- | ----------- | --- | ---------- | --- | ------ | ------------- | --- | --- | --- | --- | --- | --- | -| lacking | an inherent | attribution | | capability | | within | | | | | | | | -LLMs,therebyresultingincompromisedcitation -| | | | | | | | RetrievalAugmentedGeneration. | | | | | Recently,re- | | -| ----------------------- | --------- | --- | ----------------------- | ---- | --------- | ---- | ----------------------------- | --- | ---------- | --- | ----- | ------------ | --- | -| quality(Liuetal.,2023). | | | Secondly,thesecitations | | | | | | | | | | | -| | | | | | | | trieval augmented | | generation | | (RAG) | (Karpukhin | | -| are typically | presented | | in the | form | of either | doc- | | | | | | | | -ument identifiers (Nakano et al., 2021) or URLs etal.,2020;Lewisetal.,2020;Fengetal.,2023; -Gaoetal.,2023c)hasshownpromiseinknowledge- -| (Thoppilan | et al., | 2022). | Such | coarse | attribution | | | | | | | | | -| ---------- | ------- | ------ | ---- | ------ | ----------- | --- | --------- | ------ | ---------------- | --- | --------- | --- | ----- | -| | | | | | | | intensive | tasks. | By incorporating | | retrieved | | docu- | -complicatestheprocessforuserstoperformfine- -| | | | | | | | ments, | LLMsareequippedwithup-to-dateinfor- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------------------------- | --- | --- | --- | --- | --- | -grainedverification(e.g.,pinpointspecificsupport- -| | | | | | | | mation, | significantly | mitigating | | knowledge | | gaps. | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | ---------- | --- | --------- | --- | ----- | -ingevidence),particularlyinlengthydocuments. -| | | | | | | | However, | recent | studies | (Shi | et al., | 2023; | Yoran | -| ------- | ---- | ------ | ---------- | --- | ---------- | ---- | -------- | ------ | ------- | ---- | ------- | ----- | ----- | -| To this | end, | we aim | to advance | | attributed | text | | | | | | | | -etal.,2023;Xuetal.,2023a;Zhuetal.,2024)have -generationbyempoweringLLMswithfine-grained -attributionability. However,onechallengecomes revealed that existing retrieval-augmented LLMs -| | | | | | | | struggle | to handle | irrelevant | | or contradictory | | re- | -| --- | --- | --- | --- | --- | --- | --- | -------- | --------- | ---------- | --- | ---------------- | --- | --- | -fromtheacquisitionofhigh-qualityattributiondata -trievaldocumentsandeffectivelyutilizecontextual -| for supervised | | fine-tuning, | | which | is difficult | and | | | | | | | | -| ------------------------------------ | --- | ------------ | --- | ----- | ------------ | ------- | --------- | ----- | ----------- | --- | ---------- | --- | ------- | -| | | | | | | | evidence. | These | limitations | | can result | in | perfor- | -| costlytoannotate,andthereforescarce. | | | | | | Thus,we | | | | | | | | -mancedegradationorevenhallucinations(Huang -| start with | an automatic | | data | generation | | pipeline | | | | | | | | -| ---------- | ------------ | --- | ---- | ---------- | --- | -------- | -------------- | --- | ------------ | --- | --------- | --- | ---- | -| | | | | | | | et al., 2023), | | highlighting | the | necessity | for | more | -tailoredforcollectinghigh-qualityattributiondata -factualandverifiablesystems. -(§3.1). Givenauserquery,thepipelineautomates -dataconstructionthroughdocumentretrieval,rele- -| | | | | | | | AttributedLargeLanguageModels. | | | | | Theper- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | --- | ------- | --- | -vancereranking,attributedanswergeneration,and -| | | | | | | | sistent | challenge | of hallucinations | | | within | LLMs | -| ------------------------------- | --- | ------ | ---- | ------------------- | --- | --- | ----------- | --------- | ----------------- | --- | ------------- | ------ | ---- | -| data filtering | to | ensure | both | the informativeness | | | | | | | | | | -| | | | | | | | has spurred | the | development | | of attributed | | LLMs | -| andattributabilityoftheanswers. | | | | Furthermore,to | | | | | | | | | | -(Bohnetetal.,2022;Lietal.,2023;Worledgeetal., -betterunlockLLMs’abilityforfine-grainedattri- -2023),whichseektoenhanceinformationverifia- -| bution, | we introduce | | FRONT, | a two-stage | | train- | | | | | | | | -| ------------- | ------------ | ---- | ------- | ----------- | --- | -------- | --------- | ---------- | --------- | ------ | --------- | ----------- | -------- | -| | | | | | | | bility by | generating | responses | | with | attribution | to | -| ing framework | | that | teaches | LLMs | to | generate | | | | | | | | -| | | | | | | | evidence | sources. | The | way of | providing | | attribu- | -Fine-gRainedgrOuNdedciTations(§3.2),consist- -| | | | | | | | tionsvariesacrossstudies. | | | Forexample,Gaoetal. | | | | -| ------ | --------- | ------ | ---------- | --- | ---- | --- | ------------------------- | --- | --- | ------------------- | --- | --- | --- | -| ing of | Grounding | Guided | Generation | | (G3) | and | | | | | | | | -(2023b)enablesLLMstogeneratetextwithin-line -Consistency-AwareAlignment(CAA).Duringthe -| | | | | | | | citations | via in-context | | learning. | Another | | line of | -| --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------- | --- | ------- | -G3 -stage,theLLMfirstselectssupportingquotes -| | | | | | | | research | (Gao | et al., 2023a; | Xu | et al., | 2023b) | ex- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ---- | -------------- | --- | ------- | ------ | --- | -fromretrievedsources(grounding)andthencondi- -plorespost-hocattribution,whereLLMsfirstgen- -tionsthegenerationprocessonthem(generation). -erateaninitialresponseandthenretrievethemost -TheCAAstagethenutilizespreferenceoptimiza- -| | | | | | | | relevantevidencetoachieveattribution. | | | | | Inthispa- | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | --------- | --- | -tiontofurtheralignthegroundingandgeneration -per,weadvancetheresearchonattributedLLMs -processbyautomaticallyconstructingpreference -| | | | | | | | further. | Unlikeexistingmodelsthatpredominantly | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------------------------- | --- | --- | --- | --- | --- | -signals. Inthisway,thesequotescanserveasfine- -| | | | | | | | cite document | | identifiers, | we | delve | into | a more | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ------------ | --- | ----- | ---- | ------ | -grainedcitationsandimprovetheefficiencyofthe -fine-grainedformofattributionbypinpointingand -verificationprocessforusers(seeFigure1). -citingspecificextractivequotes. -| We conduct | | extensive | experiments | | to | evaluate | | | | | | | | -| ---------- | --- | --------- | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -our framework on the ALCE Benchmark (Gao 3 TaskFormulationandMethodology -| etal.,2023b). | Ourfindingsareasfollows: | | | | | | | | | | | | | -| ------------- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Following(Liuetal.,2023;Gaoetal.,2023b),the -| • FRONT | demonstratessupervisorperformance | | | | | | | | | | | | | -| ------- | --------------------------------- | --- | --- | --- | --- | --- | ------------------ | --- | ----------- | --- | ----- | ------ | ----- | -| | | | | | | | task is formalized | | as follows: | | given | a user | query | -gainsincitationqualitycomparedtoallbase- q and a corpus of retrieved documents D as in- -lines,achievinganaverage14.21%improve- -| | | | | | | | put, the | LLM | is required | to | produce | a response | | -| --- | --- | --- | --- | --- | --- | --- | -------- | --- | ----------- | --- | ------- | ---------- | --- | -mentusingLLaMA-2-7B. -| | | | | | | | S, which | consists | of statements | | with | embedded | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | -------- | -------------------- | --- | ---- | -------- | ---- | -| | | | | | | | in-linecitations. | | WeassumetheresponseS | | | | com- | -• Humanevaluationrevealsthatthequotesgen- -| | | | | | | | prising | with | n statements | S | = {s | ,s ,...,s | } | -| ---------------------------------------- | --- | ------------- | --- | --- | ------- | ------- | ----------------- | -------- | ------------ | ----------------------- | ---- | ------------- | --- | -| erated | by | our framework | | are | of high | quality | | | | | | 1 2 | n | -| | | | | | | | andeachstatements | | | ∈ S,citesalistofpassage | | | | -| andsignificantlybenefituserverification. | | | | | | | | | i | | | | | -| | | | | | | | C i = {c | i1 ,c i2 | ,...}, where | c ij | ∈ D. | Specifically, | | -• AnalysisshowsthatFRONTgenerateslesshal- citationsarepresentedintheformof[1][2]. -lucinationanddemonstratesremarkablegen- Next,wepresentacomprehensiveoverviewof -eralizationacrossdifferentbasemodels. ourmethod,whichconsistsoftwoprimarycompo- - -Automatic Data Generation Pipeline -          Data Collection        Answer Generation           Data Filtering -Question: Why do bagels have holes in the middle？ Bagel holes ensures even baking (...) a area -| | | | | | | Demonstrations | | |  1    3  . Another suggests (...) and transport  2  . | | | | -| --- | --- | --- | --- | --- | -------------------------------------------- | -------------- | --- | --- | ----------------------------------------------------- | --- | --- | --- | -| | | | | | Instruction: Generate an answer and cite the | | | | Informative? | | | | -Retrieve source for the question and provided documents Break into -Question: Why do bagels have holes in the middle？ -| | Top-100 retrieval documents | | | | | | | | | | statements | | -| --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | -(...) -| | | | | | Prior to (...) the hole was originally put in place | | | | Statement 1 | Bagel holes ensures even baking by | | | -| --- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | ----------- | ---------------------------------- | --- | --- | -allowing heat to circulate around a -| | | | | | to allow for easier handing prior. | | | | | greater surface area  1    3  .  | | | -| --- | -------- | -------------------------------- | ---------------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | -------------------------------- | --- | --- | -| | T h e b | a k e r s (. .. ) th e h o | l e i n t h e c e n tre of the | | | | | | | | | | -1 ba g e l e n s u r e s t h a t i t b a k e s e v e n ly . Statement 2 A n o t h e r s u g g e s t s t h a t t h e y w e re u s ed -| | | | | | | | | | | fo r e a s i e r | h a n d li n g a n | d tr a n sp o rt   2  . | -| ------ | --------------------------------------------- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | ----------- | ------------------ | -------------------- | --------------------------- | -| | Prior to (...) the hole was originally put in | | | | Bagel holes ensures even baking by allowing | | | | | | | | -| | 2 | | | | heat to circulate around a greater surface | | | | | | | Support | -| | place to allow for easier handing prior. | | | | | | | | Attributed? | | | | -| Rerank | | | | | area   1   3   . Another suggests that they were | | | | | | | | -T h e r i n g s h a p e a l lo w s h e a t t o c ir c u la t e used for easier handling and transport  2  . Documents -| | 3 ar o u n | d ( .. . )  m a k e s b a g | e l s t o c o o k fa s t er  | | | | | | | | | | -| --- | ---------- | ------------------------------ | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Figure2: Overviewofthedatagenerationpipeline. Thepipelineconsistsofthreeprimarysteps: datacollection, -answergeneration,anddatafiltering. Firstly,givenauserquery,thedatacollectionmoduleretrievesthetop100 -relevantdocumentsandemploysarerankingmodeltoselectthetop5mostpertinentdocuments. Subsequently, -attributedresponsesaregeneratedbydistillingChatGPTviain-contextlearning. Finally,allresponsesarefiltered -bythedatafilteringmoduletoensureinformativenessandattributability. -nents: anautomaticdatagenerationpipeline(§3.1) subsequently re-rankedby RankVicuna (Pradeep -andatwo-stagetrainingframework(§3.2). etal.,2023)consideringitssuperiorperformance -| | | | | | | inlistwisere-ranking, | | | resultinginthetop5most | | | | -| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | --- | -3.1 AutomaticDataGenerationPipeline -relevantdocumentsforeachquery. -EquippingLLMswiththeattributioncapabilityne- -| | | | | | | Attributed | Answer | | Generation. | | Given | the re- | -| ---------- | -------- | ---- | ------------- | ------------ | -------- | ---------- | ----------- | --- | ----------- | ------- | ----- | ---------- | -| cessitates | training | data | that includes | high-quality | | | | | | | | | -| | | | | | | markable | performance | | of | ChatGPT | in | attributed | -| responses | paired | with | precise | citations, | which is | | | | | | | | -questionanswering,weemployChatGPTtogener- -| typically | labor-intensive | | and | costly. To | address | | | | | | | | -| --------- | --------------- | --- | --- | ---------- | ------- | --- | --- | --- | --- | --- | --- | --- | -ateanswerswithcorrespondingcitationsforgiven -thischallenge,weproposeapipelinedesignedfor -| | | | | | | queriesandthetop5retrieveddocuments. | | | | | | Wepro- | -| --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | --- | ------ | -theautomaticgenerationofhigh-qualityattributed -videpreciseinstructionsandin-contextdemonstra- -data4. -Thispipelinecomprisesthreecorecompo- -tionstoensurethatChatGPTproducesinformative -nents: datacollection,attributedanswergeneration, -responsesandcitesthesourcesaccordingly. -anddatafiltering,asoutlinedinFigure2. -| | | | | | | DataFiltering. | | Toguaranteethehighqualityof | | | | | -| --------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --- | --------------------------- | ----- | --------- | --- | ----------- | -| DataCollection. | | Tosimulatethereal-worldenvi- | | | | | | | | | | | -| | | | | | | our synthetic | | training | data, | we employ | | a data fil- | -ronmentforinformation-seeking,wecollectques- -teringprocessguidedbytwokeycriteriaderived -tionsfromtheAQuAMuSedataset(Kulkarnietal., -| | | | | | | fromKamallooetal.(2023): | | | | (1)informativeness: | | | -| --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ------------------- | --- | --- | -2020),whichisderivedfromtheNaturalQuestion -| | | | | | | assessing | if the | answer | provides | | sufficient | infor- | -| ----------------------------------- | --- | --- | --- | --- | ----- | ------------------------------------------ | ------ | ------ | -------- | --- | ---------- | ------ | -| (NQ)dataset(Kwiatkowskietal.,2019). | | | | | TheNQ | | | | | | | | -| | | | | | | mationtothequestion,and(2)attributability: | | | | | | de- | -datasetcomprisesrealuserqueriesfromtheGoogle -| | | | | | | termining | if the | answer | is | attributed | | to the cited | -| --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | --- | ---------- | --- | ------------ | -searchengine,providingarobustbasisforrealistic -| | | | | | | documents. | To | mitigate | | the impact | of | nonsensi- | -| ------------------ | --- | ---------- | --- | ----------- | ----- | ---------- | --- | -------- | --- | ---------- | --- | --------- | -| question-answering | | scenarios. | | The dataset | spans | | | | | | | | -calqueriesandirrelevantdocumentretrievalthat -| a range | of diverse | question | types, | demanding | an- | | | | | | | | -| ------- | ---------- | -------- | ------ | --------- | --- | -------- | ------------------ | --- | --- | -------- | --- | ---------- | -| | | | | | | may lead | to non-informative | | | answers, | | we utilize | -swersofvaryinglengths,fromconcisetodetailed. -ChatGPTforpreliminaryinformativenessannota- -Tomimicthewayasearchenginemightsynthesize -| | | | | | | tions. Responses | | categorized | | as | non-informative | | -| --- | --- | --- | --- | --- | --- | ---------------- | --- | ----------- | --- | --- | --------------- | --- | -documentsofhighrelevanceinresponsetoauser -| | | | | | | aredirectlyexcluded. | | | Furthermore,toensurethat | | | | -| ------ | --------- | ------ | ------- | ------- | -------- | -------------------- | --- | --- | ------------------------ | --- | --- | --- | -| query, | we employ | Sphere | (Piktus | et al., | 2021), a | | | | | | | | -answersareaccompaniedbyhighlysupportiveci- -pre-processedandcleanedversionoftheCommon -tations,wetrainadiscriminatoronhuman-labeled -Crawlcorpus,servingasaproxywebsearchindex. -| | | | | | | data from | the | comprehensive | | evaluation | | by Liu | -| --- | --- | --- | --- | --- | --- | --------- | --- | ------------- | --- | ---------- | --- | ------ | -Inparticular,foragivenuserquerysampledfrom -| | | | | | | et al. (2023), | | where | attributability | | is | categorized | -| --- | ------------ | ---------- | ------------ | ---------- | ---- | ---------------- | ------------------------------------- | ------------------------------- | --------------- | ----- | --- | ----------- | -| the | AQuAMuSe | dataset, | we initially | retrieve | the | | | | | | | | -| | | | | | | intothreelevels: | | fullsupport,partialsupport,orno | | | | | -| top | 100 relevant | documents | from | the Sphere | cor- | | | | | | | | -| | | | | | | support. | Wequantitativelymapthediscriminator’s | | | | | | -| pus | using sparse | retrieval. | These | documents | are | | | | | | | | -| | | | | | | outputs | to an | attributability | | score | and | ultimately | -4Attributeddatarefersto“answerswithin-linecitations”. deriveanaveragescoreforeachattributedanswer. - -Two-Stage Training Recipe -          Grounding Guided Generation        Consistency-Aware Alignment -          Answer Generation -| Q :   W h y d o | b a g e l s h a v e | | | | | | | | | | | | -| ----------------- | ----------------------- | ---------------- | --------------------------- | ----------------------- | ------------------------------------- | --- | ----------------------- | ------------------- | ----------------------------------------------- | --------------------------- | --- | --- | -| | | G r o u n d i ng | T h e b a k e r s ( . . | . ) t h e h o l e | i n t h e c e n t r e o f the | Q | u e s t i o n :   W h y | d o b a g e l s |   G  R      D eI   G    1l   e | n s u r e s t h a t i t | | | -h o l e s i n t h e m i d d l e ？ 1 b a g e l { e n s u r e s t h a t i t b a k e s e v e n l y } . h a v e h o l e s i n t h e m i d d l e ？ b a kO eU Ns N v e n y .   3   a l l o w s h e a t -| | | | | | | | ( . . . ) | | t o c i r c u l a t e a r | o u n d .  2    a l l o w   s A u   N r S  f W  a   c E | R e     a B r a e g a e   1 l   h  3 o  A l e n s o ( t . h . . e )   r | g ( r . e .. a ) t | -| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------- | -B a g e l h o l e s e n s u r e s e v e n b a k i n g b y P r i o r t o ( . . . ) t h e h o l e w a s o r i g i n a l l y p u t in P r i o r t o ( .. . )   p l a c e t o a ll o w f o r e a s i e r h a n d i n g p r i o r . h a n d l i n g a n d t r a n s p o r t   2  . -| al lo w i n g h e a | t to c i r c u l a t e | a r o u n d a 2 | | | | | | | | | | | -| -------------------- | -------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -greater surface area  1   3  .Another p l a c e t o { a l l o w f o r e a s ie r h a n d i n g p r i o r } . fo r e a s i e r h a n d i n g p r i o r . Consistency -| suggests that they were used for | | 3 | The ring shape {allows heat to circulate | | | | | | | | | | -| -------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -easier handling and transport  2  . around} (...) makes bagels to cook faster  Gold Grounding Consistent Answer Contrastive -Large Language -| | Supervised Fine-tuning | | | | | | Model | | | | | | -| -------- | ---------------------- | -------------- | ------------- | --------------------------------------------- | ------------------------------------------------- | --- | ----- | --- | -------------- | ------------------- | --- | --- | -| | | | | | | | | | Gold Grounding | Inconsistent Answer | | | -| Question | | |   G   R   | Ol  Uo N  D  sI N   Gh     1   e n s | u r e s t h a t it b a k e s e v e n ly.  | | | | | | | | -| | | Large Language |   f3 a | l w e a t t o | c i r c u l a t e a r o u n d .   2  a l lo w | | | | | | | | -| | | | o r | e a s i e r h a n d i n | g p r i o r . | | | | | | | | -Documents Model Direct Preference Optimization (DPO)   A   N S  W    E R    B a g e l h o l e s   (. . . )  f o r -| | | |   A A  N n  S o | W  t h  E e  R r   B (. a .. g ) e h l a h n o d le | lin s g ( . a .. n )  d g r t e ra a n t s s p u o rf r a t  c 2  e . area13 | | | | | a e s t h e | t i c s a n d u n i q u e t e x | t u r e   1 | -| --- | --- | --- | ---------------- | ------------------------------------------------------------ | ------------------------------------------------------------------------------------- | --- | --- | --- | --- | ----------------- | ------------------------------------- | ------------- | -| | | | | | | | | | |   ( . . . ) h a | n d l i n g a n d tr a n s p o | r t   3   . | -Figure 3: Overview of FRONT: The training recipe consists of two stages: grounding-guided generation and -consistency-aware alignment. It enables LLMs to first generate precise grounding and subsequently guide the -generationofattributedanswers,therebyenhancingfine-grainedattributioncapability. -Answersfallingbelowadefinedthresholdaresys- fine-tunedtogeneratearesponseS whichconsists -tematicallyexcludedtoensurethesyntheticdata’s of two components: the grounded quotes G and -reliability, which results in nearly 8,000 entries. theattributedanswerA. Specifically,thegrounded -Formoredetails,pleaserefertoAppendixA. quotesG aredelineatedasfollows: -3.2 Two-StageTrainingRecipe G = {[GROUNDING],(i ,e ),...,(i ,e )}, (1) -| | | | | | | | | | | 1 1 | n n | | -| ------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Inthissection,weintroduce | | | FRONT,atwo-stage | | | | | | | | | | -where[GROUNDING]denotesaspecialtokenindi- -trainingframeworkthataimsatempoweringLLMs -| | | | | | | cating | the | start of | the grounding | process. | | Each | -| ----------------- | --- | ----------- | ----------- | --- | -------- | ------ | --- | -------- | ------------- | -------- | --- | ---- | -| with fine-grained | | attribution | capability. | | Figure 3 | | | | | | | | -tuplewithinG,comprisingadocumentidentifieri -illustratestheoverviewofourframework. -andthecorrespondingextractivesegmente,collec- -tivelyformingagroundedquote. -3.2.1 GroundingGuidedGeneration -| | | | | | | Similarly, | | the formulation | | of the | attributed | an- | -| ---------- | --- | --------- | ------------ | --- | -------- | ---------- | --- | --------------- | --- | ------ | ---------- | --- | -| To empower | | LLMs with | fine-grained | | attribu- | | | | | | | | -swerAisconciselypresentedas: -| tion capability, | | we propose | Grounding | | Guided | | | | | | | | -| ---------------- | --- | ---------- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -(G3), -Generation which teaches LLMs to gener- A = {[ANSWER],s ,s ...,s }, (2) -| | | | | | | | | | | 1 2 | m | | -| ------------------------- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| atefine-grainedcitations. | | | ThecornerstoneofG3 | | | | | | | | | | -liesinenablingLLMstoextractsupportingquotes where [ANSWER] is a special token that signals -fromthesourcedocuments,eachassociatedwithits the beginning of the answer generation process. -documentidentifier,whichinturnguidesthegen- Each statement s cites a list of passages C = -| | | | | | | | | i | | | | i | -| --------------------------- | --- | --------------------------- | ----------------- | --- | --- | ----------------- | ------------ | ------ | --- | ------------------ | ---- | ----- | -| erationofattributedanswers. | | | Suchaformatoffers | | | | | | | | | | -| | | | | | | {c i1 | ,c i2 ,...}, | wherec | ij | ⊆ {i 1 ,i 2 ,...,i | n }, | asde- | -| twoprimarybenefits. | | Firstly,thedirectextraction | | | | finedinEquation2. | | | | | | | -of quotes from sources significantly reduces the Thus,thetraininglossisformulatedas: -impactoftheincorporationofirrelevantinforma- -| tion and | the risk | of hallucinations | | in | subsequent | | | N | | | | | -| -------- | -------- | ----------------- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -(cid:88) -| | | | | | | | | L = − | logP(y | |q ,D | ;θ) | (3) | -| ------------------ | --- | ---------------------------- | --- | --- | --- | --- | --- | ----- | ------ | ----- | --- | --- | -| attributedanswers. | | Secondly,theprocessnaturally | | | | | | | | i i | i | | -i=1 -facilitatesaccurateattribution,witheachdocument -identifierservingasaclearsupervisedsignalthat -| | | | | | | where | y | represents | the | combined | output | of | -| --- | --- | --- | --- | --- | --- | ----- | --- | ---------- | --- | -------- | ------ | --- | -i -delineatestheoriginoftheextractivequotes,thus -groundedquotesGandtheanswerAforeachgiven -improvingthecitationquality. queryq andsetofretrieveddocumentsD . -| | | | | | | | i | | | | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -However,theabsenceofspecificgroundingcon- -tent for statements within our generated dataset 3.2.2 Consistency-AwareAlignment -posesadditionalchallenges. Totacklethis,weem- While G3 unlocks the ability to first extract sup- -ploy ChatGPT to meticulously extract segments portingquotesbeforegeneratingattributedanswers, -fromciteddocumentsthatsupportthecorrespond- it occasionally leads to inconsistencies between -ingstatement. Hence,whengivenaqueryqandthe groundedquotesandattributedanswers. Suchdis- -top-5retrieveddocumentsD asinput,theLLMis crepancieschallengetheattempttoemploythese - -groundedquotesasfine-grainedverification. Inre- whereπ ref representsthereferencemodel,initial- -sponse,weproposeaconsistency-awarealignment izedfromG3. Thehyper-parameterβ modulates -(CCA) stage specifically aimed at enhancing the the divergence between the distribution from the -consistencybetweenthegroundingprocessandthe policy model and the reference model. τ is the -w -generationprocess. consistentanswer,whileτ istheinconsistentone. -l -| | The cornerstone | | of | our approach | involves | con- | | | | | | | | -| --- | --------------- | --- | --- | ------------ | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -4 ExperimentalSettings -| trasting | | a consistent | | answer | with an | inconsis- | | | | | | | | -| --------------- | --- | ------------ | -------------------------- | -------- | ----------- | --------- | ------------ | --- | --- | --- | --- | --- | --- | -| tent | one | under | the | guidance | of the same | oracle | 4.1 Datasets | | | | | | | -| groundedquotes. | | | Thisalignswiththeconceptof | | | | | | | | | | | -WeconductexperimentsontheALCEbenchmark -| Reinforcement | | | Learning | from | Human | Feedback | | | | | | | | -| ------------- | --- | --- | -------- | ---- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | -(Gaoetal.,2023b),designedforattributedtextgen- -| (RLHF) | | (Ouyang | et | al., 2022; | Bai et | al., 2022), | | | | | | | | -| ------ | --- | ------- | --- | ---------- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -eration. Thebenchmarkincludesthreelong-form -whereLLMsarefurtherfine-tunedtodistinguish -QAdatasetsthatspanvarioustypesofquestions. -betweendesirableandundesirableresponsesunder -preference feedback. However, such contrastive ASQA (Stelmakh et al., 2022) is a long-form -preferencefeedbacktypicallycomesfromhuman factoidQAdatasetcharacterizedbyinherentlyam- -| annotation. | | Toautomaticallyconstructpreference | | | | | | | | | | | | -| ----------- | --- | ---------------------------------- | --- | --- | --- | --- | ------- | --------- | ---- | ------- | -------- | ----- | --- | -| | | | | | | | biguous | questions | that | require | multiple | short | an- | -pairsforpreferenceoptimization,weutilizetheat- swerstoencapsulatedifferentviewpoints. -tributedanswersgeneratedbysmallerLLMs(e.g., -LLaMA-2-7B)underthein-contextlearningsetting ELI5 (Fanetal.,2019)featuresopen-endedques- -tionsintendedforsimplificationtothecomprehen- -| asnegativesamples. | | | | Theseanswers,characterized | | | | | | | | | | -| ------------------ | --- | --- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -sionleveloffive-year-olds,requiringexplanatory -bytheirlowqualityandinconsistencywithoracle -multi-sentenceresponses. -groundedquotes,automaticallyserveascontrastive -supervisionsignalswhenpairedwithhigh-quality -| | | | | | | | QAMPARI | | (Amouyaletal.,2022)isafactoid | | | | | -| ------------------------------- | --- | --- | --- | --- | --------------- | --- | ------- | --- | ----------------------------- | --- | --- | --- | --- | -| attributedanswerslabeledin§3.1. | | | | | Inthisscenario, | | | | | | | | | -QAdatasetderivedfromWikipedia,whereanswers -theprocessnotonlyencouragestheLLMtogener- arestructuredasacompilationofentities. -ateattributedanswersthataremoreconsistentwith -thegroundedquotesbutalsofacilitatestheidentifi- 4.2 EvaluationMetrics -cationandcorrectionofnuancederrorspresentin Following the ALCE benchmark (Gao et al., -smallermodels. 2023b), our evaluation primarily focuses on two -| | Specifically, | | we adopt | Direct | Preference | Opti- | | | | | | | | -| --- | ------------- | --- | -------- | ------ | ---------- | ----- | --------------- | --- | -------- | --- | ------- | ------------ | --- | -| | | | | | | | key dimensions: | | Citation | | Quality | and Correct- | | -mization(Rafailovetal.,2023),avariantofRLHF ness. Detaileddescriptionsofadditionalevaluation -known for its stability, for our contrastive align- dimensionsarepresentedintheAppendixB. -ment. Formally,foreachinstance,giventheoracle -| | | | | | | | CitationQuality. | | Citationqualityiscriticalfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | --- | -groundedg(i)alongwithaconsistentoracleanswer -| (i) | | | | | (i) | | evaluating | LLM | attribution, | | assessed | along | two | -| --- | --------------------------- | --- | --- | --- | --- | --------- | ----------- | --- | ------------ | ------- | ----------- | ----- | --- | -| y | aswellasanattributedanswery | | | | | generated | | | | | | | | -| w | | | | | l | | | | | | | | | -| | | | | | | | dimensions: | (1) | Citation | Recall, | determining | | if | -byaweakerLLMviain-contextlearning,wecan -| | | | | | | | the output | is entirely | | supported | by | the cited | doc- | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | --- | --------- | --- | --------- | ---- | -simplyconstructapreferencedataset: -| | | | | | | | uments, | and (2) | Citation | Precision, | | assessing | if | -| --- | --- | --- | --------------------- | --- | ------------- | --- | ---------------------------------------------- | ------- | -------- | ---------- | --- | --------- | --- | -| | | D | = (cid:8) x(i),τ(i),τ | | (i)(cid:9)N , | | | | | | | | | -| | | | | w | | (4) | eachcitationsupportsitscorrespondingstatement. | | | | | | | -l i=1 -| | (i) | | (i) | | | | EvaluationisconductedbyTRUE(Honovichetal., | | | | | | | -| ------ | --- | -------- | --- | ------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -| whereτ | | = g(i)◦y | | denotestheconcatenationof | | | | | | | | | | -| | w | | w | | | | 2022),aT5-11Bmodelfine-tunedonacollection | | | | | | | -theoraclegroundingwiththeconsistent,attributed -ofNLIdatasetstoautomaticallyexaminetheentail- -| | | (i) | g(i)◦y | (i) | | | | | | | | | | -| -------- | --- | ------------ | ------ | ----------------------- | ------- | ------- | ------------------------------------------ | --- | ------- | ---------- | ------- | --- | ------ | -| answer,τ | | = | | denotestheconcatenation | | | | | | | | | | -| | | l | | l | | | mentofciteddocumentsandthemodelgeneration. | | | | | | | -| with | the | inconsistent | | attributed | answer. | Here, ◦ | | | | | | | | -| | | | | | | | Additionally, | to | capture | a holistic | measure | | of ci- | -signifiestheoperationofstringconcatenation. -tationquality,wealsoreporttheCitationF1,the -| | Finally,wecanoptimizethepolicymodelπ | | | | | on | | | | | | | | -| --- | ------------------------------------ | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | θ | harmonicmeanofcitationprecisionandrecall: | | | | | | | -thedatasetD -byminimizingthefollowingloss: -citationprecision·citationrecall -| | L | (π ;π | ;D) | | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | ----- | -| | DPO | θ | ref | | | | F 1 = 2· | | | | | | , (6) | -citationprecision+citationrecall -| | | | (cid:20) | (cid:18) | π (τ | |x) | | | | | | | | -| --- | --- | --- | -------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -| | =−E | | | | θ | w | | | | | | | | -| | | | | logσ | βlog | | | | | | | | | -(x,τw,τ l )∼D π (τ |x) (5) Correctness. FortheASQAdataset,correctness -ref w -π (τ |x) (cid:19)(cid:21) is quantified using exact match recall (EM Rec.) -θ l -| | −βlog | | | , | | | | | | | | | | -| --- | ----- | --- | ------ | --- | --- | --- | ----------- | ------- | --- | --------- | ------- | --- | ----- | -| | | π | (τ |x) | | | | by checking | whether | | the short | answers | are | exact | -| | | | ref l | | | | | | | | | | | - -| | | | ASQA | | | ELI5 | | QAMPARI | | | -| --------- | --- | ----------- | ---- | --------- | ----------- | ---- | -------- | ------------ | ---------- | --- | -| ModelType | | ModelSize | | | | | | | | | -| | | Correctness | | Citation | Correctness | | Citation | Correctness | Citation | | -| | | EMRec. | Rec. | Prec. F1. | Claim | Rec. | Prec. F1 | Rec.-5 Prec. | Rec. Prec. | F1 | -Prompting-based -ChatGPT - 40.37 72.81 69.69 71.22 12.47 49.44 47.05 48.22 20.28 19.84 19.06 22.03 20.44 -7B 24.32 17.24 17.87 17.55 4.53 3.92 5.38 4.54 12.56 11.32 6.03 6.35 6.19 -LLaMA-2 13B 27.99 16.45 19.04 17.65 7.77 8.49 8.43 8.46 18.00 12.39 5.45 5.74 5.59 -70B 31.53 44.18 44.79 44.48 10.43 23.75 22.43 23.07 18.50 14.79 10.10 10.50 10.30 -7B 29.93 55.99 51.66 53.74 12.47 19.90 15.48 17.41 17.96 19.74 9.58 9.68 9.63 -LLaMA-2-Chat 13B 34.39 37.15 38.17 37.65 13.83 16.50 16.09 16.29 21.34 18.86 8.94 9.06 9.00 -70B 41.24 60.19 61.16 60.67 13.30 36.63 36.63 36.63 22.62 18.04 13.49 13.98 13.73 -Vicuna-v1.5 7B 38.34 48.37 44.63 46.42 12.30 29.81 22.45 25.61 14.22 14.74 11.26 11.64 11.45 -13B 35.20 51.92 53.40 52.65 14.33 31.15 28.99 30.03 22.06 19.60 13.04 13.74 13.38 -7B 29.46 23.12 25.45 24.23 8.47 16.04 16.32 16.18 16.96 15.98 7.50 7.76 7.63 -Mistral -8×7B 36.30 32.72 34.49 33.58 10.43 26.11 25.09 25.59 18.18 15.63 9.72 10.20 9.95 -7B 38.57 64.90 59.67 62.18 11.07 49.25 42.69 45.74 17.52 21.29 17.56 18.53 18.03 -Mistral-Instruct 8×7B 44.11 61.80 63.27 62.53 13.93 49.28 48.34 48.81 20.12 19.64 19.27 20.38 19.81 -Post-hocRetrieval -ChatGPT - 37.68 27.11 27.05 27.08 18.77 14.55 14.55 14.55 25.14 22.85 12.29 12.29 12.29 -LLaMA-2-Chat 70B 29.68 24.51 24.51 24.51 16.03 12.93 12.93 12.93 17.90 14.45 9.05 9.05 9.05 -Mistral-Instruct 8×7B 33.90 24.57 24.48 24.52 17.37 15.68 15.68 15.68 24.16 18.28 9.78 9.78 9.78 -Training-based -7B 29.96 67.82 66.97 67.39 6.90 22.34 32.40 26.45 2.34 1.98 10.53 18.80 13.50 -Self-RAG(LLaMA-2) -13B 31.66 71.26 70.35 70.80 6.07 30.46 40.20 34.66 1.90 1.33 12.79 20.90 15.86 -7B 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -VANILLA-SFT(LLaMA-2) 13B 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -FRONT(LLaMA-2) 7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -Table1: MainresultsontheALCEbenchmark. Boldnumbersindicatethebestperformance,while_indicatesthe -second-bestperformance. -substringsofthegeneration. RegardingtheELI5 Mistral-7B(Jiangetal.,2023)toMistral-8x7B- -dataset,correctnessismeasuredthroughclaimre- MoE (Jiang et al., 2024). Regarding SFT LLMs, -call (Claim), evaluating whether the model’s re- weselecttheSFTcounterpartsoftheopen-source -sponse entails the ground truth sub-claims. For foundationalLLMsweused. Detailedprompting -theQAMPARIdataset,correctnessisassessedus- settingscanbefoundinAppendixC. -| ingexactmatchprecision(Prec.) | | | andtop-5exact | | | | | | | | -| ----------------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | -match recall (Rec.-5) — considered 100% if the 4.3.2 Post-hocRetrievalMethods. -predictionincludesatleastfivecorrectanswers. -| | | | | | Following | Gao | et al. | (2023b), | we first | instruct | -| --- | --- | --- | --- | --- | --------- | --- | ------ | -------- | -------- | -------- | -LLMstoanswerthegivenqueryinaclosed-book -4.3 Baselines -| | | | | | setting, | and then | integrate | citations | in a post-hoc | | -| --- | --- | --- | --- | --- | -------- | -------- | --------- | --------- | ------------- | --- | -Wecompareourmethodwiththreetypesofbase- -| | | | | | manner. | Foreachgeneratedstatement,weemploy | | | | | -| ------ | ---------------- | -------- | ---------- | --- | ------- | ---------------------------------- | --- | --- | --- | --- | -| lines: | prompting-based, | post-hoc | retrieval, | and | | | | | | | -GTR(Nietal.,2022)toidentifyandcitethemost -training-based. -relevantdocumentfromthetop100retrieveddoc- -4.3.1 Prompting-basedMethods. uments. Weutilizethesamemodelsmentionedin -WedirectlypromptLLMsusingfew-shotdemon- prompting-basedsettingsforthisbaseline. -strations,eachconsistingofaquery,thetop5rel- -| | | | | | 4.3.3 | Training-basedMethods. | | | | | -| --- | --- | --- | --- | --- | ----- | ---------------------- | --- | --- | --- | --- | -evantretrieveddocuments,andananswerwithin- -| linecitations. | Ourexperimentsencompassarange | | | | | | | | | | -| -------------- | ----------------------------- | --- | --- | --- | -------- | --------------------------------- | --- | --- | --- | --- | -| | | | | | Self-RAG | (Asaietal.,2023)Self-RAGtrainsthe | | | | | -ofLLMs,fromfoundationalmodelstosupervised -| | | | | | LLM | to learn | to adaptively | retrieve | passages | on- | -| --------------------- | --- | -------------------- | --- | --- | --- | -------- | ------------- | -------- | -------- | --- | -| fine-tuning(SFT)LLMs. | | ForfoundationalLLMs, | | | | | | | | | -demandandenableittoreflectonitsgenerationto -| we select | GPT-3.5-Turbo5 | as the | representative | | | | | | | | -| --------- | -------------- | ------ | -------------- | --- | --- | --- | --- | --- | --- | --- | -furtherimprovegenerationqualityandattributions. -closed-sourcemodel,recognizedforitsnotableper- -| formance. | Among | the open-source | foundational | | | | | | | | -| --------- | ----- | --------------- | ------------ | --- | ----------- | --- | -------------------------- | --- | --- | --- | -| | | | | | VANILLA-SFT | | Wedirectlyemploysupervised | | | | -LLMs,wefocusontheLLaMA-2seriesincluding -fine-tuningtotraintheLLMonourgeneratedtrain- -LLaMA2-7B,LLaMA2-13B,andLLaMA2-70B, -| | | | | | ing data. | Given | a query | and corresponding | | doc- | -| ------------------------------------------------ | -------------- | ------------- | ----- | ---- | --------------------- | ------- | ------- | ----------------- | -------- | -------- | -| as well | as the Mistral | series, which | spans | from | | | | | | | -| | | | | | uments, | the LLM | is | required to | directly | generate | -| 5Specifically,weutilizegpt-3.5-turbo-1106version | | | | | answerswithcitations. | | | | | | - -4.4 ImplementDetails -| Weimplement | | FRONTwithdifferentsizesoffoun- | | | | | 60 | | | | Pre-filtered | | | -| ----------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --- | -Post-filtered -| dational | models | (LLaMA-2-7B | | and | LLaMA-2- | | | | | | | | | -| -------- | ------ | ----------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -)%(1FnoitatiC 50 -| 13B)toevaluateitseffectiveness. | | | | Duringtheeval- | | | | | | | | | | -| ------------------------------- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -40 -uation,FRONTutilizethesameretrievalsettingsas -| those outlined | | by Gao | et al. | (2023b). | Additional | | 30 | | | | | | | -| -------------- | ----------- | ------ | ---------- | -------- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| details | of training | and | evaluation | settings | | can be | 20 | | | | | | | -foundinAppendixD. -| | | | | | | | | ASQA | | ELI5 | | QAMPARI | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ------- | --- | -5 ResultsandAnalysis -| | | | | | | | Figure4: | AblationStudyonDataFiltering. | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----------------------------- | --- | --- | --- | --- | --- | -5.1 OverallResults -Simplysupervisedfine-tuningcanboostcitation -| | | | | | | | FRONT | demonstrates | | remarkable | | generaliza- | | -| ----------- | ------------------------------ | --- | -------------- | --- | ---------- | ----- | ---------- | ------------ | --- | ---------- | --- | ----------- | ------- | -| quality. | AsshowninTable1,teachingLLMsto | | | | | | | | | | | | | -| | | | | | | | tion. | Compared | to | the varied | | queries | and an- | -| generate | responses | | with citations | via | supervised | | | | | | | | | -| | | | | | | | swer types | present | in | the ALCE | | benchmark, | our | -| fine-tuning | significantly | | enhances | | citation | qual- | | | | | | | | -trainingdata,derivedexclusivelyfromtheAQuA- -ity,demonstratingsubstantialimprovementsover -MuSedataset(Kulkarnietal.,2020),exhibitsout- -bothprompt-basedandpost-hocretrievalbaselines -| | | | | | | | of-domain | characteristics. | | | Nonetheless, | | FRONT | -| ------ | ------------- | --- | ------------- | ---- | -------- | --- | ------------ | ---------------- | --- | -------- | ------------ | --- | --------- | -| across | all datasets. | | Specifically, | with | LLaMA-2- | | | | | | | | | -| | | | | | | | demonstrates | superior | | citation | quality, | | affirming | -7B, VANILLA-SFTledtosubstantialgainsincita- -itsexceptionalabilitytogeneralizeacrossdiverse -| tion F1 | scores | over | prompting: | ASQA | (17.55 | → | | | | | | | | -| ------- | ------ | ---- | ---------- | ---- | ------ | --- | -------------------------------- | --- | --- | --- | --- | ------------- | --- | -| | | | | | | | querytypesandretrievaldocuments. | | | | | Additionally, | | -65.61),ELI5(4.54→41.15),andQAMPARI(6.19 -| | | | | | | | while not | specifically | | optimized | | for correctness, | | -| -------- | ----------------------------------- | --- | --- | --- | --- | --- | --------- | -------------- | --- | --------- | ------------ | ---------------- | --- | -| →21.35). | Thesegainshighlighttheeffectiveness | | | | | | | | | | | | | -| | | | | | | | FRONT | also showcases | | modest | improvements | | in | -ofourtrainingdatagenerationpipeline. -thismetricoverVANILLA-SFTontheASQAand -| | | | | | | | QAMPARIdatasets. | | | However,FRONTencounters | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ----------------------- | --- | --- | --- | -FRONT achievessignificantperformancegains -lowerRec.-5ontheQAMPARIdataset,likelydue -| andsurpassesChatGPT. | | | | While VANILLA-SFT | | | | | | | | | | -| -------------------- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tothenatureofitsanswers,whichconsistofcon- -| demonstrates | | strong | performance, | | it still | shows | | | | | | | | -| ------------ | --- | ------ | ------------ | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | -catenatedentities,divergingsignificantlyfromour -| notable | discrepancies | | compared | to | leading | open- | | | | | | | | -| ------- | ------------- | --- | -------- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | -trainingdata. -sourceLLMs,suchasMixtral-8×7B-Instruct(e.g., -| 41.15 vs. | 45.74) | and | ChatGPT | (e.g., | 41.15 | vs. | | | | | | | | -| --------- | ------ | ---- | -------- | ----------- | ----- | -------- | ----------------- | --- | --- | --- | --- | --- | --- | -| 48.22) | on the | ELI5 | dataset. | FRONT | not | only | 5.2 AblationStudy | | | | | | | -| bridges | these | gaps | but also | establishes | | signifi- | | | | | | | | -Weconductablationstudiestoverifytheeffective- -| cant leads | across | all | datasets. | Specifically, | | us- | | | | | | | | -| --------------- | ------ | --- | --------- | --------------- | --- | ---- | ----------------------------------- | --- | --- | --- | --- | --- | ------ | -| | | | | | | | nessofdifferentcomponentsproposedin | | | | | | FRONT. | -| ing LLaMA-2-7B, | | | FRONT | comprehensively | | out- | | | | | | | | -performsChatGPT,achievingincreasesof3.32%, -| | | | | | | | EffectsofDataGenerationPipeline. | | | | | | Asillus- | -| ------- | --- | ------ | ----------- | ------- | --- | ------ | -------------------------------- | -------- | ------ | --- | -------- | ------ | -------- | -| 18.04%, | and | 21.28% | in citation | quality | | on the | | | | | | | | -| | | | | | | | trated | in §5.1, | simply | SFT | achieves | strong | per- | -ASQA,ELI5,andQAMPARIdatasetsrespectively. -| | | | | | | | formance, | underscoring | | the | high | quality | of our | -| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | --- | --- | ---- | ------- | ------ | -Thisperformanceunderscorestheeffectivenessof -| | | | | | | | training | data. | Furthermore, | | data | filtering, | a cru- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------ | --- | ---- | ---------- | ------ | -FRONTinenhancingattributioncapabilities. -| | | | | | | | cial component | | of our | data | generation | | pipeline, | -| -------------------------------------- | --- | --- | --- | --- | --- | --- | -------------- | ------- | ------- | -------- | ---------- | ------- | --------- | -| | | | | | | | plays a | pivotal | role in | ensuring | the | quality | of the | -| FRONTexhibitsscalabilitywithmodelsize. | | | | | | As | | | | | | | | -illustratedatthebottomofTable1,theperformance generated data by filtering out queries that yield -of FRONT in terms of citation quality shows no- non-informativeanswersorfailtomeetattribution -| | | | | | | | criteria. | To validate | | the effectiveness | | of | our data | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ----------------- | --- | --- | -------- | -tableimprovementswhenscalingfrom7Bto13B. -filteringstrategies,weconductedexperimentscom- -Specifically,weobserveimprovementsof3.23%in -ASQA,4.97%inELI5,and1.33%inQAMPARI. paringmodelsfine-tunedonbothpre-filteredand -This upward trend underscores the scalability of post-filtereddata. Theresults,depictedinFigure -4,confirmthatmodelstrainedonfiltereddataex- -FRONT withincreasingmodelsize,demonstrating -thepotentialofFRONTinleveragingtheincreased hibitanotableimprovementincitationqualityover -capabilitiesoflargerLLMsforfurtherperformance thosetrainedonunfiltereddata,achievingsuperior -attributionperformancewithreduceddatavolume. -gains. - -| | | | | ASQA | | | | ELI5 | | | QAMPARI | | | -| --- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | ------- | --- | --- | -Model -| | | | Correctness | | Citation | | Correctness | | Citation | Correctness | | Citation | | -| --- | --- | --- | ----------- | --- | -------- | --- | ----------- | --- | -------- | ----------- | --- | -------- | --- | -EMRec. Rec. Prec. F1. Claim Rec. Prec. F1 Rec.-5 Prec. Rec. Prec. F1 -FRONT-7B 40.84 77.70 69.89 73.59 9.18 58.60 55.33 56.92 11.50 21.38 24.74 24.84 24.79 -SELF-GUIDE(w/oConsistency) 38.99 70.69 64.48 67.44 10.04 47.63 44.80 46.17 12.18 20.03 22.50 22.58 22.54 -VANILLA-SFT(w/oGround) 40.32 67.67 63.67 65.61 9.63 42.30 40.06 41.15 12.86 21.09 21.35 21.36 21.35 -FRONT-13B 41.51 78.44 73.66 75.97 9.32 60.31 59.21 59.75 11.94 22.61 24.86 25.39 25.12 -SELF-GUIDE(w/oConsistency) 40.99 73.08 68.13 70.52 10.06 50.68 49.78 50.23 13.94 22.38 23.73 23.99 23.85 -VANILLA-SFT(w/oGround) 40.85 71.49 66.21 68.75 10.27 46.75 44.47 45.58 12.68 22.80 23.64 23.71 23.67 -Table2: AblationstudyontheimpactofdifferenttrainingstageswithintheALCEbenchmark. -44.42 -Citation-F1 -| | | | | 46.17 | | | | | 33 | | FRONT-13B | | | -| -------------- | --- | ----- | ----- | ----- | ----------- | --- | --- | ------------------- | --------------- | --- | --------- | ------------ | --- | -| | | | | | | | | | | | F | -7 B | | -| | | 43.71 | | | SELF-GUIDE | | | | | | R O | N T | | -| Citation-Prec. | | | | | | | | | LLaMA2-70B-Chat | | SEL F- G | U I D E -13B | | -| | | | 44.80 | | PROMPTGUIDE | | | ssenlufhtiaF 22..55 | | | | | | -SELF-GUIDE-7B -45.16 -| Citation-Rec. | | | | | | | | | | | Mixtral-8x7B-Inst | | | -| ------------- | --- | ----- | --- | ----- | ----- | --- | --- | --- | --- | ---------- | ----------------- | --- | --- | -| | | | | | 47.63 | | | | 22 | Vicuna-13B | | | | -| | | 44.00 | | 46.00 | 48.00 | | | | | | | | | -LLaMA2-7B-Chat -| Figure5: | Ablationstudyofdifferentgroundingguid- | | | | | | | 11..55 | | | | | | -| -------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -ancestrategiesontheELI5dataset. -| | | | | | | | | | 2200 | 3300 | 4400 5500 | 6600 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --------- | ---- | --- | -CitationF1 -EffectsofGroundingGuidedGeneration(G3). -G3 -empowers LLMs to first select relevant fine- Figure6: TherelationshipbetweencitationF1andhal- -grainedquotes,whichsubsequentlyguidethegen- lucination: Models positioned closer to the top-right -eration process. These quotes can provide fine- cornerexhibithighercitationqualityandalowerdegree -ofhallucination. -grainedsupervisionsignalsforattributedtextgen- -G3, -| eration. | To evaluate | | the effectiveness | | of | we | | | | | | | | -| --------------------------------------- | ----------- | --- | ----------------- | --------- | ---------- | ----- | ------- | -------- | ----------- | --- | --------- | --- | ------ | -| conduct | an ablation | | study | comparing | it against | | | | | | | | | -| | | | | | | | eration | process. | Experiments | | conducted | | on the | -| twovariantswithdistincttrainingrecipes. | | | | | | Given | | | | | | | | -ELI5datasetusingtheLLaMA-2-7Bmodelshow -that FRONTconsistsoftwostages,werefertothe that SELF-GUIDE outperforms PROMPT-GUIDE. -modeltrainedonlyduringthefirststage(without -ResultsdepictedinFigure5indicatethattraining -| consistency-awarealignment)asSELF-GUIDE. | | | | | | We | | | | | | | | -| ---------------------------------------- | ---------- | --- | ------- | --- | ----------- | --- | ---------- | --- | ------------- | --------- | -------- | -------------- | ------ | -| | | | | | | | models | to | self-generate | | grounded | quotes | before | -| firstcompare | SELF-GUIDE | | against | | VANILLA-SFT | | | | | | | | | -| | | | | | | | generating | | attributed | responses | is | more effective | | -(w/oGround),whichistrainedtodirectlygenerate thansimplyincorporatingthesegroundedquotes -responseswithcitations,bypassingthegrounding -intotheprompt. -| step. The | ablation | study, | detailed | | in Table | 2, re- | | | | | | | | -| --------- | -------- | ------ | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | -veals that models incorporating grounding guid- EffectsofConsistency-AwareAlignment(CCA). -ancesignificantlyoutperformtheir VANILLA-SFT The primary goal of CCA is to enhance the con- -counterpartsthatlacksuchgroundingmechanisms. sistency between grounded quotes and attributed -This highlights the crucial role of grounding in answers, thereby alleviating hallucinations and -| | | | | | | | achievingmorepreciseattribution. | | | | | Toevaluatethis, | | -| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | -bolsteringattribution. -Moreover, we explore an alternative variant we compare models that underwent only the G3 -of grounding guidance. Considering that SELF- stage(SELF-GUIDE)withthosefurtherenhanced -GUIDE leverages the model itself to both select through the CCA stage (FRONT). As illustrated -grounded quotes and generate attributed answers inTable2,FRONTsignificantlyimprovescitation -in an end-to-end paradigm, a natural variant in- qualityoverSELF-GUIDE,demonstratingtheeffec- -volves breaking down this task into two distinct tivenessoftheCCAstageinenhancingattribution. -stages. Inthisvariant,ChatGPTistaskedwithex- Furthermore,toassessCCA’simpactonreduc- -tractinggroundedquotes. Subsequently,aseparate inghallucinations,weutilizeQAFactEval(Fabbri -modelistrainedtoutilizethesegroundedquotes, etal.,2022),awidelyusedmetricforfactualcon- -along with the query and retrieval documents, to sistency, which scores the consistency of model -directly output the response and citations. This responses to given documents on a scale from 0 -variant,referredtoasPROMPT-GUIDED,integrates to5,withhigherscoresindicatinggreaterfaithful- -groundedquotesintotheprompttoguidethegen- ness. Specifically,weanalyzetheperformanceof - -leading open-source models and two variants of Authenticity Helpfulness -| FRONTand | SELF-GUIDE | | ontheELI5dataset. | | As | | | | | | | -| -------- | ---------- | --- | ----------------- | --- | --- | ------- | --- | --- | --- | ---- | ---- | -| | | | | | | ChatGPT | | | | 0.93 | 4.08 | -showninFigure6, FRONTproducesmorefaithful FRONT-7BonASQA 0.94 3.86 -outputsthanSELF-GUIDE,significantlyreducing FRONT-7BonELI5 0.92 3.96 -| hallucinations. | | | | | | FRONT-7BonQAMPARI | | | | 0.86 | 3.62 | -| --------------- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | ---- | ---- | -Table3: Humanevaluationonthequalityofgrounded -| EffectsofTrainingDataScale. | | | | Weanalyzethe | | | | | | | | -| --------------------------- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | -quotes. -| impact | of the | data scale | on model | performance | | | | | | | | -| ------ | ------------ | ---------- | -------- | ----------- | ------- | --- | --- | --- | --- | --- | --- | -| across | two training | stages. | In | particular, | we ran- | | | | | | | -tors. Theresultsofthehumanevaluationindicate -domlysampled2k,4k,6k,and8kinstancesfrom -thatbothquotesextractedbyChatGPTandthose -| our full | training | data | across two | distinct | training | | | | | | | -| -------- | -------- | ---- | ---------- | -------- | -------- | --- | --- | --- | --- | --- | --- | -stages. These subsets were then utilized to fine- generated by FRONT are of high quality, further -tunevarious7Bmodelvariants,enablingacompar- substantiatingtheeffectivenessofourmethod. -ativeanalysisofperformancebasedondatascale. -| Results | are shown | in | Figure | 7, which | indicates | | | | | | | -| ------- | --------- | --- | ------ | -------- | --------- | --- | --- | --- | --- | --- | --- | -7 Conclusion -thatincreasingdatasizeshowssignificantenhance- -mentsincitationquality,indicatingapositivecor- Inthiswork,wepresent FRONT,atwo-stagetrain- -relationbetweendatasizeandmodelperformance. ingframeworkdesignedtoequipLLMswithfine- -AsFRONTimplementsanautomatedprocedureca- -| | | | | | | grained | attribution | capabilities. | | FRONT | enables | -| --- | --- | --- | --- | --- | --- | ------- | ----------- | ------------- | --- | ----- | ------- | -pableofgeneratinghigh-qualityattributeddataand LLMstoinitiallyselectsupportingquotes,which -constructingcontrastivesupervisionfromweakand thenguidethegenerationprocess. Byfurtheren- -strongLLMs,itholdsthepotentialforcontinuous -| | | | | | | hancing | the consistency | | between | | the grounding | -| --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------- | --- | ------------- | -performanceimprovements. andgenerationprocessviapreferenceoptimization, -thesesupportingquotescanserveasfine-grained -6 HumanEvaluation -| | | | | | | citations. | Through | comprehensive | | | experiments, | -| --- | --- | --- | --- | --- | --- | ---------- | ------- | ------------- | --- | --- | ------------ | -Given the significant impact of the quality of FRONThasdemonstrateditsabilitytogeneratesu- -| | | | | | | perior grounded | | responses | | and highly | supportive | -| -------- | ------ | --- | ------------ | ------------ | --- | --------------- | ------------------------ | --------- | --- | ---------- | ---------- | -| grounded | quotes | on | fine-grained | verification | for | | | | | | | -| | | | | | | citations. | Furtheranalysisshowsthat | | | | FRONT sig- | -users,weconductedahumanevaluationtoassess -thequalityofgroundedquotesatdifferentstagesof nificantlyreduceshallucinationsandbenefitsuser -verification. -| ourframework: | | (1)QuotesextractedbyChatGPT | | | | | | | | | | -| ---------------------------------- | --------- | --------------------------- | ----------- | --- | ------------ | ------------ | --- | --- | --- | --- | --- | -| from50sampleddatapointsduringtheG3 | | | | | stage. | | | | | | | -| (2) Quotes | generated | | by FRONT-7B | | across three | 8 Limitation | | | | | | -datasets,with50datapointssampledfromeach. -Weengagedfourannotators,eachwithrelevant Our study presents several limitations worth not- -expertiseandholdingatleastabachelor’sdegree. ing. Firstly, the validation of our framework is -Thequalityofquoteswasevaluatedontwodimen- predominantly conducted on models of sizes 7B -sions: authenticityandhelpfulness. Authenticity and13B,leavingtheexplorationoflargermodels, -(abinaryscaleof0/1)referstowhetherthequotes suchasLLaMA-270Bduetocomputationalcon- -genuinelyoriginatefromthecorrespondingdocu- straints. Secondly,ourframeworkreliesonaprior -ments(quotesthatarehallucinatedormismatched retrievalprocess,whereinrelevantdocumentsare -withthecorrespondingdocumentIDareconsidered retrieved at one time. The incorporation of adap- -inauthentic). Helpfulness (5-point Likert scale) tiveretrieval,enablingmoredynamicinteractions -referstothedegreetowhichthequotesarebenefi- withLLMs,couldpotentiallyenhanceperformance. -cialinaddressingthequery. TheresultsinTable3 Weleaveitforfutureresearch. Lastly,evaluating -representtheaveragescoresforallquoteswithin the correctness of long-form question answering -eachmodelresponse,withtwoannotatorsevaluat- presents inherent challenges, leading our frame- -ingeachresponsetoensurereliability. work to primarily enhance citation quality, with -Furthermore, to evaluate the consistency of modest advancements in correctness. Therefore, -quotequalityannotations,wecomputedtheinter- we advocate for the development of more robust -annotator agreement using Fleiss’ Kappa coeffi- metricscapableofaccuratelyassessingthecorrect- -cient. The obtained Kappa coefficient of 0.82 in- nessoflong-formQAresponses, pavingtheway -| dicates | a high | level of | agreement | among | annota- | forfuturework. | | | | | | -| ------- | ------ | -------- | --------- | ----- | ------- | -------------- | --- | --- | --- | --- | --- | - -Acknowledgements -qa-basedfactualconsistencyevaluationforsumma- -| | | | | | | | rization. | InProceedingsofthe2022Conferenceof | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | ---------------------------------- | --- | --- | --- | -We appreciate Yuchun Fan for providing valu- theNorthAmericanChapteroftheAssociationfor -| | | | | | | | ComputationalLinguistics: | | | HumanLanguageTech- | | -| ----------------- | --- | --------- | --- | --- | ------- | -------- | ------------------------- | --- | --- | ------------------ | --- | -| able suggestions. | | Xiaocheng | | | Feng is | the cor- | | | | | | -nologies,NAACL2022,Seattle,WA,UnitedStates, -| responding | author | of | this | work. | We | thank the | | | | | | -| ---------- | ------ | --- | ---- | ----- | --- | --------- | --- | --- | --- | --- | --- | -July10-15,2022,pages2587–2601.Associationfor -| anonymous | reviewers | | for | their | insightful | com- | | | | | | -| --------- | --------- | --- | --- | ----- | ---------- | ---- | --- | --- | --- | --- | --- | -ComputationalLinguistics. -| ments. | This | work | was | supported | by | the Na- | | | | | | -| ------ | ---- | ---- | --- | --------- | --- | ------- | --- | --- | --- | --- | --- | -AngelaFan,YacineJernite,EthanPerez,DavidGrang- -| tional Key | R&D | Program | | of China | via | grant No. | | | | | | -| -------------- | --- | ------- | -------- | -------- | --------- | --------- | ---------- | -------- | ----------- | -------------- | ----- | -| | | | | | | | ier, Jason | Weston, | and Michael | Auli. 2019. | ELI5: | -| 2021ZD0112905, | | the | National | | Natural | Science | | | | | | -| | | | | | | | long form | question | answering. | In Proceedings | of | -| Foundation | of | China | (NSFC) | (grant | 62276078, | | | | | | | -the57thConferenceoftheAssociationforCompu- -U22B2059), the Key R&D Program of Hei- tationalLinguistics,ACL2019,Florence,Italy,July -longjiang via grant 2022ZX01A32, the Interna- 28-August2,2019,Volume1: LongPapers,pages -3558–3567.AssociationforComputationalLinguis- -tionalCooperationProjectofPCL,PCL2022D01 -tics. -andtheFundamentalResearchFundsfortheCen- -tralUniversities(GrantNo.HIT.OCEF.2023018). ZhangyinFeng,WeitaoMa,WeijiangYu,LeiHuang, -HaotianWang,QianglongChen,WeihuaPeng,Xi- -| | | | | | | | aochengFeng,BingQin,andTingLiu.2023. | | | | Trends | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | ------ | -inintegrationofknowledgeandlargelanguagemod- -References -els:Asurveyandtaxonomyofmethods,benchmarks, -CoRR,abs/2311.05876. -| Samuel | Joseph | Amouyal, | Ohad | Rubin, | | Ori Yoran, | andapplications. | | | | | -| -------- | --------- | -------- | --------- | ------- | -------------- | ---------- | ---------------- | ------ | ----------------- | -------- | --------- | -| Tomer | Wolfson, | Jonathan | | Herzig, | and | Jonathan | | | | | | -| | | | | | | | Luyu Gao, | Zhuyun | Dai, Panupong | Pasupat, | Anthony | -| Berant. | 2022. | QAMPARI: | | : | An open-domain | | | | | | | -| | | | | | | | Chen, | Arun | Tejasvi Chaganty, | Yicheng | Fan, Vin- | -| question | answering | | benchmark | | for questions | with | | | | | | -centY.Zhao,NiLao,HongraeLee,Da-ChengJuan, -| many | answers | from | multiple | paragraphs. | | CoRR, | | | | | | -| ---- | ------- | ---- | -------- | ----------- | --- | ----- | ---------- | ---- | ------ | ----------------- | --- | -| | | | | | | | and Kelvin | Guu. | 2023a. | RARR: researching | and | -abs/2205.12665. -revisingwhatlanguagemodelssay,usinglanguage -AkariAsai,ZeqiuWu,YizhongWang,AvirupSil,and models. InProceedingsofthe61stAnnualMeeting -oftheAssociationforComputationalLinguistics(Vol- -| Hannaneh | Hajishirzi. | | 2023. | Self-rag: | Learning | to | | | | | | -| -------- | ----------- | --- | ----- | --------- | -------- | --- | ----- | ----------------------------------- | --- | --- | --- | -| | | | | | | | ume1: | LongPapers),ACL2023,Toronto,Canada, | | | | -retrieve,generate,andcritiquethroughself-reflection. -July9-14,2023,pages16477–16508.Associationfor -CoRR,abs/2310.11511. -ComputationalLinguistics. -| Yuntao Bai, | Andy | Jones, | Kamal | Ndousse, | | Amanda | | | | | | -| ----------- | ---- | ------ | ----- | -------- | --- | ------ | --- | --- | --- | --- | --- | -TianyuGao,HowardYen,JiatongYu,andDanqiChen. -| Askell, | AnnaChen, | | NovaDasSarma, | | DawnDrain, | | | | | | | -| --------- | --------- | ---- | ------------- | --- | ---------- | --------- | ------------------ | ------------------------------------- | -------------------- | --- | --- | -| | | | | | | | 2023b. | Enablinglargelanguagemodelstogenerate | | | | -| Stanislav | Fort, | Deep | Ganguli, | | Tom | Henighan, | | | | | | -| | | | | | | | textwithcitations. | | CoRR,abs/2305.14627. | | | -NicholasJoseph,SauravKadavath,JacksonKernion, -TomConerly,SheerElShowk,NelsonElhage,Zac -| | | | | | | | Yunfan Gao, | Yun | Xiong, Xinyu | Gao, Kangxiang | Jia, | -| --------------- | --- | ----- | ---------- | --- | ------- | ----- | ----------- | --- | ------------ | -------------- | ---- | -| Hatfield-Dodds, | | Danny | Hernandez, | | Tristan | Hume, | | | | | | -JinliuPan,YuxiBi,YiDai,JiaweiSun,QianyuGuo, -ScottJohnston,ShaunaKravec,LianeLovitt,Neel Meng Wang, and Haofen Wang. 2023c. Retrieval- -| Nanda, | Catherine | Olsson, | | Dario | Amodei, | Tom B. | | | | | | -| -------- | --------- | ------- | --------------- | --------- | ------- | ---------- | ------------------------------------------ | -------------------- | --- | --- | --- | -| | | | | | | | augmentedgenerationforlargelanguagemodels: | | | | A | -| Brown, | Jack | Clark, | Sam McCandlish, | | Chris | Olah, | | | | | | -| | | | | | | | survey. | CoRR,abs/2312.10997. | | | | -| Benjamin | Mann, | and | Jared | Kaplan. | 2022. | Train- | | | | | | -| ing a | helpful | and | harmless | assistant | | with rein- | | | | | | -SylvainGugger,LysandreDebut,ThomasWolf,Philipp -forcement learning from human feedback. CoRR, Schmid,ZacharyMueller,SourabMangrulkar,Marc -abs/2204.05862. Sun,andBenjaminBossan.2022. Accelerate: Train- -ingandinferenceatscalemadesimple,efficientand -BerndBohnet,VinhQ.Tran,PatVerga,RoeeAharoni, -| | | | | | | | adaptable. | | https://github.com/huggingface/ | | | -| ------ | ------ | ----- | ------- | ------- | ----- | ------ | ---------- | --- | ------------------------------- | --- | --- | -| Daniel | Andor, | Livio | Baldini | Soares, | Jacob | Eisen- | | | | | | -accelerate. -stein,KuzmanGanchev,JonathanHerzig,KaiHui, -TomKwiatkowski,JiMa,JianmoNi,TalSchuster, OrHonovich, RoeeAharoni, JonathanHerzig, Hagai -WilliamW.Cohen,MichaelCollins,DipanjanDas, Taitelbaum,DoronKukliansy,VeredCohen,Thomas -Donald Metzler, Slav Petrov, and Kellie Webster. Scialom, Idan Szpektor, Avinatan Hassidim, and -| 2022. | Attributed | question | | answering: | | Evaluation | | | | | | -| ----- | ---------- | -------- | --- | ---------- | --- | ---------- | ----- | ------- | ----------- | ------------- | ------- | -| | | | | | | | Yossi | Matias. | 2022. TRUE: | re-evaluating | factual | -andmodelingforattributedlargelanguagemodels. -| | | | | | | | consistencyevaluation. | | InProceedingsofthe2022 | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | ---------------------- | --- | --- | -CoRR,abs/2212.08037. Conference of the North American Chapter of the -| | | | | | | | AssociationforComputationalLinguistics: | | | | Human | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | ----- | -CanyuChenandKaiShu.2023. Combatingmisinfor- LanguageTechnologies,NAACL2022,Seattle,WA, -mation in the age of llms: Opportunities and chal- UnitedStates, July10-15, 2022, pages3905–3920. -| lenges. | CoRR,abs/2311.05656. | | | | | | | | | | | -| ------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AssociationforComputationalLinguistics. -Alexander R. Fabbri, Chien-Sheng Wu, Wenhao Liu, LeiHuang,WeijiangYu,WeitaoMa,WeihongZhong, -and Caiming Xiong. 2022. Qafacteval: Improved Zhangyin Feng, Haotian Wang, Qianglong Chen, - -WeihuaPeng,XiaochengFeng,BingQin,andTing servingwithpagedattention. InProceedingsofthe -Liu. 2023. A survey on hallucination in large lan- ACMSIGOPS29thSymposiumonOperatingSystems -| guagemodels: | Principles,taxonomy,challenges,and | | | | | Principles. | | | | | | -| ---------------- | ---------------------------------- | ---- | -------- | -------- | --- | ----------- | ------------ | -------- | ----------------- | --- | ----- | -| openquestions. | CoRR,abs/2311.05232. | | | | | | | | | | | -| | | | | | | Patrick | S. H. Lewis, | Ethan | Perez, Aleksandra | | Pik- | -| | | | | | | tus, Fabio | Petroni, | Vladimir | Karpukhin, | | Naman | -| Ziwei Ji, Nayeon | Lee, | Rita | Frieske, | Tiezheng | Yu, | | | | | | | -DanSu,YanXu,EtsukoIshii,YejinBang,Andrea Goyal,HeinrichKüttler,MikeLewis,Wen-tauYih, -Madotto,andPascaleFung.2023. Surveyofhalluci- Tim Rocktäschel, Sebastian Riedel, and Douwe -nationinnaturallanguagegeneration. ACMComput. Kiela. 2020. Retrieval-augmented generation for -Surv.,55(12):248:1–248:38. knowledge-intensiveNLPtasks. InAdvancesinNeu- -| | | | | | | ralInformationProcessingSystems33: | | | | AnnualCon- | | -| --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ---------- | --- | -AlbertQ.Jiang,AlexandreSablayrolles,ArthurMen- -ferenceonNeuralInformationProcessingSystems -sch,ChrisBamford,DevendraSinghChaplot,Diego 2020,NeurIPS2020,December6-12,2020,virtual. -| de Las Casas, | Florian | Bressand, | | Gianna | Lengyel, | | | | | | | -| ------------- | ------- | --------- | --- | ------ | -------- | --- | --- | --- | --- | --- | --- | -Guillaume Lample, Lucile Saulnier, Lélio Re- Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, -nard Lavaud, Marie-Anne Lachaux, Pierre Stock, Ziyang Chen, Baotian Hu, Aiguo Wu, and Min -TevenLeScao,ThibautLavril,ThomasWang,Timo- Zhang. 2023. A survey of large language models -| | | | | | | attribution. | CoRR,abs/2311.03731. | | | | | -| ----------------------------------- | --- | --- | --- | --- | ------- | ------------ | -------------------- | --- | --- | --- | --- | -| théeLacroix,andWilliamElSayed.2023. | | | | | Mistral | | | | | | | -7b. CoRR,abs/2310.06825. -| | | | | | | Nelson F. | Liu, Tianyi | Zhang, | and Percy | Liang. | 2023. | -| --- | --- | --- | --- | --- | --- | --------- | ----------- | ------ | --------- | ------ | ----- | -Albert Q. Jiang, Alexandre Sablayrolles, Antoine Evaluatingverifiabilityingenerativesearchengines. -Roux,ArthurMensch,BlancheSavary,ChrisBam- InFindingsoftheAssociationforComputationalLin- -ford,DevendraSinghChaplot,DiegodeLasCasas, guistics: EMNLP2023,Singapore,December6-10, -Emma Bou Hanna, Florian Bressand, Gianna 2023,pages7001–7025.AssociationforComputa- -| Lengyel, | Guillaume | Bour, | Guillaume | | Lample, | tionalLinguistics. | | | | | | -| ------------ | --------- | ------ | --------- | --------- | ------- | ------------------ | --- | --- | --- | --- | --- | -| Lélio Renard | Lavaud, | Lucile | | Saulnier, | Marie- | | | | | | | -AnneLachaux,PierreStock,SandeepSubramanian, JacobMenick,MajaTrebacz,VladimirMikulik,John -Sophia Yang, Szymon Antoniak, Teven Le Scao, Aslanides, H. Francis Song, Martin J. Chadwick, -Théophile Gervet, Thibaut Lavril, Thomas Wang, Mia Glaese, Susannah Young, Lucy Campbell- -| | | | | | | Gillingham, | Geoffrey | | Irving, and Nat | McAleese. | | -| --------------------------------------- | --- | --- | --- | --- | ---- | ----------- | -------- | --- | --------------- | --------- | --- | -| TimothéeLacroix,andWilliamElSayed.2024. | | | | | Mix- | | | | | | | -2022. Teachinglanguagemodelstosupportanswers -| tralofexperts. | CoRR,abs/2401.04088. | | | | | | | | | | | -| --------------- | -------------------- | ------- | ----- | ------ | ------ | ------------------- | --- | -------------------- | --- | --- | --- | -| | | | | | | withverifiedquotes. | | CoRR,abs/2203.11147. | | | | -| Ehsan Kamalloo, | Aref | Jafari, | Xinyu | Zhang, | Nandan | | | | | | | -Thakur,andJimmyLin.2023. HAGRID:Ahuman- ReiichiroNakano,JacobHilton,SuchirBalaji,JeffWu, -llmcollaborativedatasetforgenerativeinformation- Long Ouyang, Christina Kim, Christopher Hesse, -seekingwithattribution. CoRR,abs/2307.16883. ShantanuJain,VineetKosaraju,WilliamSaunders, -| | | | | | | Xu Jiang, | Karl | Cobbe, | Tyna Eloundou, | Gretchen | | -| --- | --- | --- | --- | --- | --- | --------- | ---- | ------ | -------------- | -------- | --- | -Krueger,KevinButton,MatthewKnight,Benjamin -VladimirKarpukhin,BarlasOguz,SewonMin,Patrick -S.H.Lewis,LedellWu,SergeyEdunov,DanqiChen, Chess,andJohnSchulman.2021. Webgpt: Browser- -andWen-tauYih.2020. Densepassageretrievalfor assisted question-answering with human feedback. -open-domainquestionanswering. InProceedingsof CoRR,abs/2112.09332. -the2020ConferenceonEmpiricalMethodsinNat- -| | | | | | | Jianmo | Ni, Chen | Qu, Jing | Lu, Zhuyun | Dai, | Gus- | -| ------------- | ----------- | --- | ----- | ----- | ------- | ------ | -------- | -------- | ---------- | ---- | ---- | -| ural Language | Processing, | | EMNLP | 2020, | Online, | | | | | | | -November16-20,2020,pages6769–6781.Associa- tavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, -tionforComputationalLinguistics. YiLuan,KeithB.Hall,Ming-WeiChang,andYinfei -| | | | | | | Yang.2022. | Largedualencodersaregeneralizable | | | | | -| --- | --- | --- | --- | --- | --- | ---------- | --------------------------------- | --- | --- | --- | --- | -SayaliKulkarni,SheideChammas,WanZhu,FeiSha, retrievers. InProceedingsofthe2022Conferenceon -and Eugene Ie. 2020. Aquamuse: Automatically EmpiricalMethodsinNaturalLanguageProcessing, -generatingdatasetsforquery-basedmulti-document EMNLP2022,AbuDhabi,UnitedArabEmirates,De- -summarization. CoRR,abs/2010.12694. cember7-11,2022,pages9844–9855.Association -forComputationalLinguistics. -| TomKwiatkowski, | JennimariaPalomaki, | | | | OliviaRed- | | | | | | | -| --------------- | ------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | -field,MichaelCollins,AnkurP.Parikh,ChrisAlberti, OpenAI. 2023. GPT-4 technical report. CoRR, -| DanielleEpstein,IlliaPolosukhin,JacobDevlin,Ken- | | | | | | abs/2303.08774. | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -tonLee,KristinaToutanova,LlionJones,Matthew -LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, -| Kelcey, Ming-Wei | | Chang, | Andrew | M. | Dai, Jakob | | | | | | | -| ---------------- | -------- | -------- | ------- | ----- | ---------- | ------- | -------------- | --- | --------------- | --- | ----- | -| | | | | | | Carroll | L. Wainwright, | | Pamela Mishkin, | | Chong | -| Uszkoreit, | Quoc Le, | and Slav | Petrov. | 2019. | Natu- | | | | | | | -ralquestions: abenchmarkforquestionanswering Zhang,SandhiniAgarwal,KatarinaSlama,AlexRay, -research. Trans.Assoc.Comput.Linguistics,7:452– JohnSchulman,JacobHilton,FraserKelton,Luke -| 466. | | | | | | Miller,MaddieSimens,AmandaAskell,PeterWelin- | | | | | | -| ---- | --- | --- | --- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | --- | -der,PaulF.Christiano,JanLeike,andRyanLowe. -Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying 2022. Training languagemodelsto followinstruc- -Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. tionswithhumanfeedback. InAdvancesinNeural -Gonzalez, Hao Zhang, and Ion Stoica. 2023. Effi- InformationProcessingSystems35: AnnualConfer- -cientmemorymanagementforlargelanguagemodel enceonNeuralInformationProcessingSystems2022, - -NeurIPS2022,NewOrleans,LA,USA,November28 Doshi,RenelitoDelosSantos,TojuDuke,JohnnySo- -| -December9,2022. | | | | | | | raker,BenZevenbergen,VinodkumarPrabhakaran, | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | ----------- | --- | ------- | ----------- | -| | | | | | | | Mark Diaz, | Ben | Hutchinson, | | Kristen | Olson, Ale- | -AleksandraPiktus,FabioPetroni,VladimirKarpukhin, jandraMolina,ErinHoffman-John,JoshLee,Lora -DmytroOkhonko,SamuelBroscheit,GautierIzacard, -| | | | | | | | Aroyo, | Ravi Rajakumar, | | Alena | Butryna, | Matthew | -| --- | --- | --- | --- | --- | --- | --- | ------ | --------------- | --- | ----- | -------- | ------- | -Patrick S. H. Lewis, Barlas Oguz, Edouard Grave, Lamm,ViktoriyaKuzmina,JoeFenton,AaronCo- -Wen-tauYih,andSebastianRiedel.2021. Theweb hen,RachelBernstein,RayKurzweil,BlaiseAgüera -isyouroyster-knowledge-intensiveNLPagainsta yArcas,ClaireCui,MarianCroak,EdH.Chi,and -verylargewebcorpus. CoRR,abs/2112.09924. QuocLe.2022. Lamda: Languagemodelsfordialog -CoRR,abs/2201.08239. -applications. -KrishnaPillutla,SwabhaSwayamdipta,RowanZellers, -JohnThickstun,SeanWelleck,YejinChoi,andZaïd -| | | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | -Harchaoui.2021. MAUVE:measuringthegapbe- bert, Amjad Almahairi, Yasmine Babaei, Nikolay -tweenneuraltextandhumantextusingdivergence Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti -| frontiers. | In | Advances | in Neural | Information | | Pro- | | | | | | | -| ---------- | --- | -------- | --------- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | -Bhosale,DanBikel,LukasBlecher,CristianCanton- -| cessingSystems34: | | AnnualConferenceonNeural | | | | | | | | | | | -| ----------------- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Ferrer,MoyaChen,GuillemCucurull,DavidEsiobu, -InformationProcessingSystems2021,NeurIPS2021, -JudeFernandes,JeremyFu,WenyinFu,BrianFuller, -December6-14,2021,virtual,pages4816–4828. -CynthiaGao,VedanujGoswami,NamanGoyal,An- -thonyHartshorn,SagharHosseini,RuiHou,Hakan -RonakPradeep,SahelSharifymoghaddam,andJimmy Inan,MarcinKardas,ViktorKerkez,MadianKhabsa, -| Lin.2023. | Rankvicuna: | | Zero-shotlistwisedocument | | | | | | | | | | -| --------- | ----------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -IsabelKloumann,ArtemKorenev,PunitSinghKoura, -rerankingwithopen-sourcelargelanguagemodels. -Marie-AnneLachaux,ThibautLavril,JenyaLee,Di- -CoRR,abs/2309.15088. -anaLiskovich,YinghaiLu,YuningMao,XavierMar- -tinet,TodorMihaylov,PushkarMishra,IgorMoly- -| Rafael Rafailov, | | Archit | Sharma, | Eric | Mitchell, | Ste- | | | | | | | -| ---------------- | --- | ------ | ------- | ---- | --------- | ---- | ---------- | ---- | ------ | -------- | ------ | ------- | -| | | | | | | | bog, Yixin | Nie, | Andrew | Poulton, | Jeremy | Reizen- | -fanoErmon,ChristopherD.Manning,andChelsea -stein,RashiRungta,KalyanSaladi,AlanSchelten, -| Finn. | 2023. | Direct preference | | optimization: | | Your | | | | | | | -| ------------------------------------ | ----- | ----------------- | --- | ------------- | --- | ----- | -------------- | ----- | ------- | ------ | ------ | --------- | -| | | | | | | | Ruan Silva, | Eric | Michael | Smith, | Ranjan | Subrama- | -| languagemodelissecretlyarewardmodel. | | | | | | CoRR, | | | | | | | -| | | | | | | | nian, Xiaoqing | Ellen | Tan, | Binh | Tang, | Ross Tay- | -abs/2305.18290. -| | | | | | | | lor, Adina | Williams, | Jian | Xiang | Kuan, | Puxin Xu, | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --------- | ---- | ----- | ----- | --------- | -ZhengYan,IliyanZarov,YuchenZhang,AngelaFan, -| Jeff Rasley, | Samyam | Rajbhandari, | | Olatunji | | Ruwase, | | | | | | | -| ------------ | ------ | ------------ | --- | -------- | --- | ------- | --- | --- | --- | --- | --- | --- | -and Yuxiong He. 2020. Deepspeed: System opti- Melanie Kambadur, Sharan Narang, Aurélien Ro- -driguez,RobertStojnic,SergeyEdunov,andThomas -mizationsenabletrainingdeeplearningmodelswith -| | | | | | | | Scialom.2023. | Llama2: | | Openfoundationandfine- | | | -| ------------------------- | --- | --- | --- | --------- | ------- | --- | ---------------- | ------- | -------------------- | ---------------------- | --- | --- | -| over100billionparameters. | | | | InKDD’20: | The26th | | | | | | | | -| | | | | | | | tunedchatmodels. | | CoRR,abs/2307.09288. | | | | -ACMSIGKDDConferenceonKnowledgeDiscovery -| and Data | Mining, | Virtual | Event, | CA, | USA, | August | | | | | | | -| -------- | ------- | ------- | ------ | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | -23-27,2020,pages3505–3506.ACM. TheodoraWorledge,JudyHanwenShen,NicoleMeis- -| | | | | | | | ter,CalebWinston,andCarlosGuestrin.2023. | | | | | Uni- | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ---- | -fyingcorroborativeandcontributiveattributionsin -| Freda Shi, | Xinyun | Chen, | Kanishka | Misra, | | Nathan | | | | | | | -| ---------- | ------ | ----- | -------- | ------ | --- | ------ | -------------------- | --- | --- | -------------------- | --- | --- | -| | | | | | | | largelanguagemodels. | | | CoRR,abs/2311.12233. | | | -Scales,DavidDohan,EdH.Chi,NathanaelSchärli, -| andDennyZhou.2023. | | | Largelanguagemodelscan | | | | | | | | | | -| ------------------ | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -beeasilydistractedbyirrelevantcontext. InInterna- Fangyuan Xu, Weijia Shi, and Eunsol Choi. 2023a. -tionalConferenceonMachineLearning,ICML2023, RECOMP:improvingretrieval-augmentedlmswith -23-29 July 2023, Honolulu, Hawaii, USA, volume compression and selective augmentation. CoRR, -abs/2310.04408. -202ofProceedingsofMachineLearningResearch, -pages31210–31227.PMLR. -ShichengXu,LiangPang,HuaweiShen,XueqiCheng, -IvanStelmakh,YiLuan,BhuwanDhingra,andMing- andTat-SengChua.2023b. Search-in-the-chain: To- -Wei Chang. 2022. ASQA: factoid questions meet wards the accurate, credible and traceable content -long-formanswers. InProceedingsofthe2022Con- generation for complex knowledge-intensive tasks. -CoRR,abs/2304.14732. -ferenceonEmpiricalMethodsinNaturalLanguage -Processing,EMNLP2022,AbuDhabi,UnitedArab -Emirates,December7-11,2022,pages8273–8288. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan -AssociationforComputationalLinguistics. Berant. 2023. Making retrieval-augmented lan- -| | | | | | | | guage models | robust | to | irrelevant | context. | CoRR, | -| ----- | ---------- | ------ | ------------- | -------- | -------- | ----- | --------------- | ------ | --- | ---------- | -------- | ----- | -| Romal | Thoppilan, | Daniel | De | Freitas, | Jamie | Hall, | abs/2310.01558. | | | | | | -| Noam | Shazeer, | Apoorv | Kulshreshtha, | | Heng-Tze | | | | | | | | -Cheng,AliciaJin,TaylorBos,LeslieBaker,YuDu, Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, -YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Xiaolei Wang, Yupeng Hou, Yingqian Min, Be- -AminGhafouri,MarceloMenegali,YanpingHuang, ichenZhang,JunjieZhang,ZicanDong,YifanDu, -MaximKrikun,DmitryLepikhin,JamesQin,Dehao Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao -Chen,YuanzhongXu,ZhifengChen,AdamRoberts, Jiang,RuiyangRen,YifanLi,XinyuTang,Zikang -MaartenBosma,YanqiZhou,Chung-ChingChang, Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. -IgorKrivokon,WillRusch,MarcPickett,KathleenS. 2023. A survey of large language models. CoRR, -| Meier-Hellstern, | | Meredith | Ringel | Morris, | | Tulsee | abs/2303.18223. | | | | | | -| ---------------- | --- | -------- | ------ | ------- | --- | ------ | --------------- | --- | --- | --- | --- | --- | - -Kun Zhu, Xiaocheng Feng, Xiyuan Du, Yuxuan Gu, process, we retained 5,667 and 2,431 queries, re- -WeijiangYu,HaotianWang,QianglongChen,Zheng spectively. Additionally,wecalculatedtheaverage -| Chu, JingchangChen, | | andBingQin.2024. | | Anin- | | | | | | | -| ------------------- | --- | ---------------- | --- | ----- | --- | --- | --- | --- | --- | --- | -lengthofanswersandtheaveragenumberofcita- -formationbottleneckperspectiveforeffectivenoise -tionsgeneratedforvarioustypesofquerieswithin -filteringonretrieval-augmentedgeneration. -ourdataset,asshowninTable4. -YutaoZhu,HuayingYuan,ShutingWang,JiongnanLiu, -| Wenhan | Liu, Chenlong | Deng, | Zhicheng | Dou, and | | | | | | | -| ------ | ------------- | ----- | -------- | -------- | --- | --- | --- | --- | --- | --- | -A.2 DetailsofDataFiltering -| Ji-RongWen.2023. | | Largelanguagemodelsforinfor- | | | | | | | | | -| ---------------- | --- | ---------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | -| mationretrieval: | | Asurvey. | CoRR,abs/2308.07107. | | | | | | | | -WetrainedourAttributedDiscriminatorusingthe -| | | | | | manually | annotated | data | provided | by | Liu et al. | -| --- | --- | --- | --- | --- | -------- | --------- | ---- | -------- | --- | ---------- | -A DetailsofDataGenerationPipeline -| | | | | | (2023), | which is | sampled | from | real generative | | -| --- | --- | --- | --- | --- | ------- | -------- | ------- | ---- | --------------- | --- | -A.1 DataStatistic searchengines. Eachstatementanditsciteddocu- -menthavebeenmeticulouslyannotatedforattribu- -#Questions 8,098 tion,categorizedintothreetypes: completesupport, -| | | | | | partialsupport,andnosupport. | | | | Fortraining,weuti- | | -| ------------ | --- | --- | --- | ---- | ---------------------------- | --- | --- | --- | ------------------ | --- | -| ➥#LongAnswer | | | | 5667 | | | | | | | -lizedadatasetof8,834instances,comprising6,415 -| ➥#ShortAnswer | | | | 2431 | | | | | | | -| ------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | -instancesofcompletesupport,1,552ofpartialsup- -Avg. WordsperAnswer 50.48 port, and 867 of no support. The discriminator -initializedwithLLaMA-2-7B,wastrainedwitha -| ➥Avg. | WordsperLongAnswer | | | 69.15 | | | | | | | -| ----- | ------------------- | --- | --- | ----- | --------------------------- | --- | --- | --- | -------------- | --- | -| | | | | | maximumsequencelengthof512. | | | | Wetraineditfor | | -| ➥Avg. | WordsperShortAnswer | | | 6.94 | | | | | | | -3epochs,withatotalbatchsizeof128,andapeak -Avg. CitationperAnswer 4.40 learning rate of 2e-5, incorporating 3% warmup -steps,followedbyalineardecay. -| ➥Avg. | CitationperLongAnswer | | | 4.68 | | | | | | | -| ----- | ---------------------- | --- | --- | ---- | ------ | ----------------- | --------- | --------- | ---------- | ----- | -| ➥Avg. | | | | | During | the data | filtering | stage, | we first | break | -| | CitationperShortAnswer | | | 3.77 | | | | | | | -| | | | | | down | the automatically | | generated | attributed | an- | -swersintostatementformandusethetraineddis- -| Table 4: | The statistics | of the | data generated | by our | | | | | | | -| -------- | -------------- | ------ | -------------- | ------ | --- | --- | --- | --- | --- | --- | -criminatortoannotatetheattributionbetweeneach -automaticdatagenerationpipeline. -| | | | | | statement | and its | cited | documents. | Specifically, | | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ---------- | ------------- | --- | -weassigndifferentattributionscorestoeachstate- -Table4presentsthestatisticsofthedataautomat- -| | | | | | ment s | based on | its attribution | | relationship | with | -| ------------------------------------------- | --- | --- | --- | --- | ----------------------------------- | -------- | --------------- | --- | ------------ | ---- | -| icallygeneratedbyourdatagenerationpipeline. | | | | In | | | | | | | -| | | | | | citeddocumentsd,asshowninEquation7. | | | | | Con- | -total,wecollected8,098questionsfromtheNatural -sequently,foreachattributedanswer,wecancal- -Questions(NQ)dataset,ofwhich5,667questions -| | | | | | culateitsaverageattributionscore. | | | | Attributedan- | | -| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | ------------- | --- | -weregatheredfromthosewithlong-formanswers, -swerswithanaverageattributionscorebelow0.8 -and2,431questionswerecollectedfromthosewith -| | | | | | are filtered | out. | The threshold | | of 0.8 was | deter- | -| --- | --- | --- | --- | --- | ------------ | ---- | ------------- | --- | ---------- | ------ | -short-formfactoidanswers. -minedthroughpreliminarytestingonthedevelop- -Forquestionsrequiringlong-formanswers,we -| | | | | | ment set, | for which | we | manually | annotated | 100 | -| --- | --- | --- | --- | --- | --------- | --------- | --- | -------- | --------- | --- | -initializedourquerysourcewiththeAQUAMUSE -samplestoensuretheeffectivenessofourfiltering -| dataset (Kulkarni | | et al., 2020), | which consists | of | | | | | | | -| ----------------- | --- | -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | -criteria. -high-qualityqueriesspecificallydesignedforlong- -formresponseswithintheNQdataset,recognized -| as“good”bythemajorityofNQevaluators. | | | | Inthis | | | | | | | -| ------------------------------------ | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | - -| | | | | | | 1, Dis(s,d) | | = complete | support | | -| -------------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | ---------- | ------- | --- | -| way,utilizingarefinedandsuperiorqualityquery | | | | | |   | | | | | -setlaidarobustgroundworkforourtrainingdata r(s)= 0.5,Dis(s,d) = partial support - -generation,streamliningthedatafilteringprocess. 0, Dis(s,d) = no support -| Forfactoidqueriesthatnecessitateshort-forman- | | | | | | | | | | (7) | -| --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -swers,wedirectlysampledfromtheoriginalNQ -| dataset, leveraging | | its abundance | and | inherently | | | | | | | -| ------------------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | -B DetailsofEvaluationMetrics -highquality. -During the data generation process, our initial In addition to evaluating citation quality and cor- -querysetcomprised7,725queriesrequiringlong- rectness,theALCEbenchmarkincludesabroader -formanswersand4,000queriesnecessitatingshort- setofdimensions,suchasfluency,ROUGE-L,and -| form answers. | | After a two-stage | data | filtering | generationlength. | | | | | | -| ------------- | --- | ----------------- | ---- | --------- | ----------------- | --- | --- | --- | --- | --- | - -Fluency Weevaluatethefluencyofthegenerated framework(Kwonetal.,2023)forefficientinfer- -responseusingMAUVE(Pillutlaetal.,2021). No- ence. Thehyperparametersaresetasillustratedin -| tably,wecalculatefluencyonlyfortheASQAand | | | | | | | Table9. | | | | | | | -| ----------------------------------------- | --- | -------- | ------ | -------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | -| ELI5 datasets, | | omitting | it for | QAMPARI, | | as the | | | | | | | | -response in QAMPARI typically consists of lists D.2 RetrievalSettings -| ofshortanswers. | | ArelativelyhighMAUVEscore | | | | | | | | | | | | -| --------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Duringtheevaluation,weadoptthesameretrieval -indicatesthatthegenerationissufficientlyfluent. -| | | | | | | | settings | as specified | | by Gao | et | al. (2023b). | For | -| ------- | --------------------------------- | --- | --- | -------- | --- | ------ | -------------------------------- | ------------ | ------- | ------ | --------- | ------------ | ------- | -| | | | | | | | the ASQA | and | QAMPARI | | datasets, | we | use the | -| ROUGE-L | Inadditiontoevaluatingthecorrect- | | | | | | | | | | | | | -| | | | | | | | denseretrieverGTR(Nietal.,2022). | | | | | FortheELI5 | | -| ness of | the model-generated | | | content, | we | employ | | | | | | | | -dataset,weemploythesparseretrieverBM25. -ROUGE-Ltoassesstheoverallqualityandtextual -coherenceoftheresponses. -E MoredetailaboutAblationStudy -C Prompts -E.1 TheEffectofTrainingDataScale. -C.1 PromptsforPrompting-basedMethods Weexaminehowmodelperformancevarieswith -| | | | | | | | changesindatascale,asdepictedinFigure7. | | | | | | The | -| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -FollowingGaoetal.(2023b),weadoptthevanilla -| | | | | | | | upper part | of | the figure | illustrates | | the | impact of | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | ----------- | --- | --- | --------- | -promptingstrategyforitssimplicityandeffective- -| | | | | | | | the training | data | scale | on | citation | quality | during | -| ------------------------------------ | ------ | -------- | -------- | -------- | --------- | --------- | ------------- | ---- | ------- | ---------- | ---------- | -------- | --------- | -| ness. Specifically, | | the | prompts | vary | according | to | | | | | | | | -| | | | | | | | the Grounding | | Guided | Generation | | training | stage, | -| thetypeofdatawithintheALCEbenchmark. | | | | | | For | | | | | | | | -| | | | | | | | with datasets | | ASQA, | ELI5, | and | QAMPARI | rep- | -| long-form | QA | datasets | such | as ASQA | | and ELI5, | | | | | | | | -| | | | | | | | resented | from | left to | right. | Similarly, | | the lower | -| the prompt | format | is | detailed | in Table | 5. | For the | | | | | | | | -partofthefiguredescribestheinfluenceduringthe -| short-form | QA | dataset | QAMPARI, | | the | format is | | | | | | | | -| ---------- | --- | ------- | -------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -Consistency-AwareAlignmenttrainingstage. -outlinedinTable6. -E.2 TheGeneralizationAcrossModel -| C.2 Instructionsfor | | | FRONT | | | | | | | | | | | -| ------------------- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Architectures. -Duringthetrainingprocess,wefollowtheinstruc- -| | | | | | | | FRONT | demonstrates | | exceptional | | generalization | | -| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | --- | ----------- | --- | -------------- | --- | -Alpaca6. -| tion format | of | | Specifically, | | we | employ | | | | | | | | -| ----------- | --- | --- | ------------- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -capabilitiesacrossvariousfoundationalmodelar- -variedinstructionsfordifferentquestiontypes,as -| | | | | | | | chitectures. | Specifically,transitioningthefounda- | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------------------------ | --- | --- | --- | --- | --- | -delineatedinTable7forlong-formquestionsand -| | | | | | | | tional model | | from LLaMA-2-7B | | | to the | stronger | -| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------------- | --- | --- | ------ | -------- | -Table8forshort-formquestions. -| | | | | | | | foundational | | model, | Mistral-7B, | | results | in even | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ----------- | ------ | ------------ | --- | ------- | -------- | -| | | | | | | | greater | performance | | enhancements | | as | shown in | -D ExperimentalDetails -| | | | | | | | Figure8. | Thisfurtherunderscoresthebroadappli- | | | | | | -| --------------------- | --- | --- | ----- | --- | --- | --- | ----------------------------- | ------------------------------------ | --- | --- | ------ | --- | --- | -| | | | | | | | cabilityandgeneralizabilityof | | | | FRONT. | | | -| D.1 TrainingDetailsof | | | FRONT | | | | | | | | | | | -Thetrainingofallmodelsisexecutedon4Nvidia E.3 Theeffectofβ inConsistency-Aware -A100GPUs,eachwith80GBofmemory,leverag- AlignmentTrainingStage -ingtheDeepspeed(Rasleyetal.,2020)andHug- -| | | | | | | | In the | Consistency-Aware | | | Alignment | | Training | -| --- | --- | --- | --- | --- | --- | --- | ------ | ----------------- | --- | --- | --------- | --- | -------- | -gingFaceAcceleratelibraries(Guggeretal.,2022) -| | | | | | | | Stage, | the β | parameter | in | Direct | Preference | Op- | -| -------------------------------------- | --- | --- | --- | --- | --- | ----- | ---------- | ----- | --------- | --- | ------ | ---------- | ------ | -| toconductmulti-GPUdistributedtraining. | | | | | | Given | | | | | | | | -| | | | | | | | timization | (DPO) | controls | | the | strength | of the | -thelongnatureoftheinputs,themaximumtoken -Kullback-Leiblerpenalty,typicallysetwithinthe -lengthissetto2,048tokens. -| | | | | | | | range of | 0.1 to | 0.5. | A higher | β | value | indicates a | -| ------ | --- | --------- | ----- | ---------- | --- | ------ | -------- | ------ | ---- | -------- | --- | ----- | ----------- | -| During | the | grounding | guide | generation | | stage, | | | | | | | | -preferenceforthepolicymodel’strainingprocess -modelsaretrainedfor5epochswithatotalbatch -toremainclosertotheinitiallyreferencedmodel. -| size of | 128, a peak | learning | | rate of | 2e-5 | with 3% | | | | | | | | -| -------------------------------------- | ----------- | -------- | --- | ------- | ---- | ------- | ----------------------------------- | ------ | --- | --- | ----- | ------ | -------- | -| | | | | | | | In extreme | cases, | as | β → | 0, we | ignore | the con- | -| warmupstepsfollowedbyalineardecay. | | | | | | During | | | | | | | | -| | | | | | | | straintsimposedbythereferencemodel. | | | | | | Thisset- | -| thecontrastivealignmentstage,wesettheβ | | | | | | to0.1 | | | | | | | | -tingaimstobalancethemodel’sabilitytoadaptto -andcontinuedtrainingfortwoadditionalepochs. -newtrainingsignalswhilemaintainingthestability -| Specifically, | During | | inference, | we | use | the vllm | | | | | | | | -| ------------- | ------ | --- | ---------- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -ofthelearnedbehaviorsfromthereferencemodel. -| | | | | | | | Subsequently, | | we | trained | five | variants | by ad- | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | ------- | ---- | -------- | ------ | -6https://github.com/tatsu-lab/stanford_alpaca/ -| tree/main | | | | | | | justingβ | from0.1to0.5onthemodelpreviously | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --- | - -Instruction: Write an accurate, engaging, and concise answer for the given question using only -the provided search results (some of which might be irrelevant) and cite them properly. Use an -unbiased and journalistic tone. Always cite for any factual claim. When citing several search -results, use [1][2][3]. Cite at least one document and at most three documents in each sentence. -If multiple documents support the sentence, only cite a minimum sufficient subset of the -documents. -Table5: PromptforLong-formQA. -Instruction: Provide a list of accurate answers for the given question using only the provided -search results (some of which might be irrelevant) and cite them properly. Always cite one and -only one document for each answer. Separate answers by commas. For questions that have more than -| 5 answers, | write | at least 5 answers. | | | | | -| ---------- | ----- | ------------------- | --- | --- | --- | --- | -Table6: PromptforShort-formQA. -| 72.0 | | | 48.0 | | 23.0 | | -| ---- | --- | ----- | ---- | ----- | ---- | ----- | -| 68.0 | | | 46.0 | | 22.0 | | -| 64.0 | | | 44.0 | | 21.0 | | -| 60.0 | | REC. | 42.0 | REC. | 20.0 | REC. | -| 56.0 | | PREC. | 40.0 | PREC. | 19.0 | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| ---- | --- | -------- | ---- | -------- | ---- | -------- | -| 76.0 | | | 54.0 | | 25.0 | | -| 72.0 | | | 52.0 | | 24.0 | | -| 68.0 | | | 50.0 | | 23.0 | | -| 64.0 | | | 48.0 | | 22.0 | | -| | | REC. | | REC. | | REC. | -| 60.0 | | | 46.0 | | 21.0 | | -| | | PREC. | | PREC. | | PREC. | -2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 2,000 4,000 6,000 8,000 -| | | DataSize | | DataSize | | DataSize | -| --- | --- | -------- | --- | -------- | --- | -------- | -Figure7: Ablationstudyonsynthetictrainingdatasize: TheupperpartofthefigurecorrespondstotheGrounding -GuidedGenerationtrainingstage, whilethebottompartrepresentstheWeak-to-StrongContrastiveAlignment -trainingstage. Fromlefttoright,theresultsarepresentedforASQA,ELI5,andQAMPARI,respectively. REC. -indicatesCitationRecallandPREC. denotesCitationPrecision. Thex-axisrepresentsthequantityofautomatically -generateddata. Itisobservedthatasthevolumeofautomaticallygenerateddataincreases,thereisaconsistent -improvementinbothcitationrecallandprecisionacrossthetwotrainingstages. -trainedwithG3 toexploretheimpactofthehyper- adheretotheevaluationframeworkestablishedin -parameter β on attribution quality. We evaluated (Gaoetal.,2023b). Forlong-formQAdatasetslike -thesevariantsontheASQAandELI5datasets,and ASQAandELI5,wealsoreportmetricsrelatedto -theexperimentalresultsareshowninFigure9. Fluency,ROUGE-L,andaverageresponselength. -The experimental results indicate that as β in- Specifically,weuseMAUVE(Pillutlaetal.,2021) -creases, the model’s performance on attribution toevaluatethefluencyofthemodelresponse. For -graduallydecreases. Thisobservationsuggeststhat datasetslikeQAMPARI,whereanswersarecom- -thefirststageofG3 mightintroduceanoticeable posed of concatenated entities, we calculate the -inconsistencybetweengroundingandattribution. averagenumberofpredictedentities. -Withhigherβvalues,themodelstrugglestoescape -theconstraintsofinconsistentattributedanswers, -| leading to | a reduction | in attribution | quality as | β | | | -| ---------- | ----------- | -------------- | ---------- | --- | --- | --- | -increases. -F FullResults -Wepresentthecomprehensiveresultsofourexperi- -| mentsinTables10,11,and12. | | Beyondtheevalua- | | | | | -| ------------------------- | --- | ---------------- | --- | --- | --- | --- | -tionmetricsrelatedtoCorrectnessandCitation,we - -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -| guide | answer | generation | and | cite the | sources properly. | | | | | | | -| ------------------- | ------ | ---------- | --- | ---------- | ----------------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table7: | InstructionFormatforFRONTonLong-formQA. | | | | | | | | -| --- | --- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Below is an instruction that describes a task, paired with an input that provides further -| context. | Write | a response | that | appropriately | completes | | the request. | | | | | -| -------- | ----- | ---------- | ---- | ------------- | --------- | --- | ------------ | --- | --- | --- | --- | -### Instruction: -Extract the relevant content from the provided documents and then use the extracted content to -provide a list of accurate answers for the given question. Always cite one and only one document -| for each | answer. | Separate | answers | by | commas. | | | | | | | -| ------------------- | ------- | ---------- | ------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | -| ### Input:Question: | | {Question} | | Documents: | {Documents} | | | | | | | -### Response: -| | | | Table8: | InstructionFormatforFRONTonShort-formQA. | | | | | | | | -| ------- | ----------------------------------- | --- | ------- | ---------------------------------------- | --- | ------------- | --- | --- | ---------------- | --- | --- | -| | Hyper-parameters | | | Value | | | | | | | | -| | Top-p | | | 0.95 | | | | | | | | -| | Temperature | | | 0.2 | | | | | | | | -| | Max-length | | | 2048 | | | | | | | | -| Table9: | Hyper-parametersettingsininference. | | | | | | | | | | | -| | | | | | | | 72 | | 56 | | | -| | | | | | | )%(1FnoitatiC | 71 | | )%(1FnoitatiC 55 | | | -Mistral-7B -| | | | VANILLA-SFT(Mistral-7B) | | | | 70 | | 54 | | | -| --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | FRONT(Mistral-7B) | | | | 69 | | 53 | | | -60 -| | | | | | | | 68 | | 52 | | | -| ---------------- | --- | --- | --- | --- | --- | --- | -------------- | ----------- | --- | -------------- | ----------- | -| )%(1FnoitatiC 50 | | | | | | | 67 | FRONT-7B | 51 | | FRONT-7B | -| 40 | | | | | | | 0.1 0.2 | 0.3 0.4 0.5 | | 0.1 0.2 | 0.3 0.4 0.5 | -| 30 | | | | | | | βonASQAdataset | | | βonELI5dataset | | -20 -| | | | | | | Figure | 9: Ablation | on hyperparameter | | β | in Weak-to- | -| --- | --- | --- | --- | --- | --- | -------------------------------------------- | ----------- | ----------------- | --- | --- | ----------- | -| 10 | | | | | | StrongContrastiveAlignmentstageonASQAandELI5 | | | | | | -0 -| | ASQA | | ELI5 | QAMPARI | | | | | | | | -| --- | ---- | --- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Figure8:Ablationstudyonmodelarchitecture:Wesub- -stitutedthefoundationmodelinFRONTwithMistral-7B -andcomparedtheexperimentalresultsofmodelsunder -| the same | foundation | model | using | in-context | learning | | | | | | | -| -------- | ---------- | ----- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | -andthosedirectlysupervisedfine-tunedonourautomat- -| icallygenerateddata. | | Theexperimentsdemonstratethat | | | | | | | | | | -| -------------------- | --------- | ----------------------------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | -| by replacing | different | foundation | | models, | our frame- | | | | | | | -workstillmaintainsitsgeneralizability. - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | -------- | ---------- | ---------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (EMRec.) | Rec. Prec. | F1 ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 73.41 | 40.37 | 72.81 69.69 | 71.22 37.92 | 39.24 | -| ------------ | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 79.90 | 24.32 | 17.24 17.87 | 17.55 29.38 | 42.29 | -| LLaMA-2 | 13B | 87.08 | 27.99 | 16.45 19.04 | 17.65 31.41 | 39.25 | -| | 70B | 69.28 | 31.53 | 44.18 44.79 | 44.48 31.53 | 26.86 | -| | 7B | 66.78 | 29.93 | 55.99 51.66 | 53.74 32.93 | 26.18 | -| LLaMA-2-Chat | 13B | 66.14 | 34.39 | 37.15 38.17 | 37.65 35.13 | 33.68 | -| | 70B | 86.60 | 41.24 | 60.19 61.16 | 60.67 37.01 | 47.09 | -| | 7B | 86.92 | 38.34 | 48.37 44.63 | 46.42 35.95 | 63.90 | -Vicuna-v1.5 -| | 13B | 66.11 | 35.20 | 51.92 53.40 | 52.65 35.74 | 38.57 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.37 | 29.46 | 23.12 25.45 | 24.23 31.67 | 37.17 | -Mistral -| | 8×7B | 83.30 | 36.30 | 32.72 34.49 | 33.58 35.05 | 38.47 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 82.86 | 38.57 | 64.90 59.67 | 62.18 36.21 | 45.26 | -Mistral-Instruct -| | 8×7B | 94.77 | 44.11 | 61.80 63.27 | 62.53 38.54 | 58.83 | -| --- | ---- | ----- | ----- | ----------- | ----------- | ----- | -Post-hocRetrieval -| ChatGPT | - | 49.78 | 37.68 | 27.11 27.05 | 27.08 36.64 | 52.61 | -| ------------ | ---- | ----- | ----- | ----------- | ----------- | ------ | -| | 7B | 75.56 | 16.55 | 13.88 13.86 | 13.87 26.81 | 37.50 | -| LLaMA-2 | 13B | 77.91 | 20.51 | 20.95 20.94 | 20.94 29.53 | 31.37 | -| | 70B | 75.23 | 27.58 | 28.43 28.43 | 28.43 30.33 | 29.88 | -| | 7B | 22.50 | 14.17 | 11.33 11.33 | 11.33 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 64.52 | 24.43 | 21.43 21.43 | 21.43 33.91 | 41.12 | -| | 70B | 70.63 | 29.68 | 24.51 24.51 | 24.51 34.17 | 45.74 | -| | 7B | 63.87 | 19.58 | 16.24 16.24 | 16.24 33.22 | 41.80 | -| Vicuna-v1.5 | 13B | 73.83 | 24.79 | 24.11 24.11 | 24.11 34.42 | 43.54 | -| | 7B | 86.54 | 21.17 | 16.78 16.77 | 16.77 30.90 | 42.43 | -| Mistral | 8×7B | 80.99 | 36.30 | 38.37 35.27 | 36.75 35.05 | 38.47 | -| | 7B | 67.97 | 26.26 | 17.87 17.85 | 17.86 33.71 | 51.56 | -Mistral-Instruct 8×7B 65.51 33.90 24.57 24.48 24.52 36.20 53.83 -Training-based -| | 7B | 74.33 | 29.96 | 67.82 66.97 | 67.39 35.70 | 29.83 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -Self-RAG -| | 13B | 71.59 | 31.66 | 71.26 70.35 | 70.80 36.01 | 27.03 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 76.66 | 40.32 | 67.67 63.67 | 65.61 38.32 | 62.00 | -VANILLA-SFT -| | 13B | 84.36 | 40.85 | 71.49 66.21 | 68.75 38.22 | 58.82 | -| --- | --- | ----- | ----- | ----------- | ----------- | ----- | -| | 7B | 81.88 | 40.84 | 77.70 69.89 | 73.59 36.95 | 53.93 | -FRONT -| | 13B | 76.11 | 41.51 | 78.44 73.66 | 75.95 38.63 | 57.56 | -| --- | --- | -------- | ---------------- | ----------- | ----------- | ----- | -| | | Table10: | ASQAfullresults. | | | | - -| | | Fluency | Correct. | Citation | | | -| --------- | --------- | ------- | ------------ | -------- | ------- | ------ | -| ModelType | ModelSize | | | | | | -| | | (MAUVE) | (Claim) Rec. | Prec. F1 | ROUGE-L | Length | -Prompting-based -| ChatGPT | - | 44.65 | 12.47 49.44 | 47.05 48.22 | 20.64 | 90.2 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 63.72 | 4.53 3.92 | 5.38 4.54 | 18.27 | 103.36 | -| LLaMA-2 | 13B | 62.19 | 7.77 8.49 | 8.43 8.46 | 19.95 | 88.23 | -| | 70B | 53.39 | 10.43 23.75 | 22.43 23.07 | 20.43 | 93.84 | -| | 7B | 32.80 | 12.47 19.90 | 15.48 17.41 | 20.88 | 96.42 | -| LLaMA-2-Chat | 13B | 29.08 | 13.83 16.50 | 16.09 16.29 | 21.04 | 94.32 | -| | 70B | 33.69 | 13.30 36.63 | 36.63 36.63 | 21.29 | 117.84 | -| | 7B | 31.45 | 12.30 29.81 | 22.45 25.61 | 21.36 | 105.68 | -Vicuna-v1.5 -| | 13B | 37.41 | 14.33 31.15 | 28.99 30.03 | 21.74 | 98.23 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 56.62 | 8.47 16.04 | 16.32 16.18 | 20.46 | 93.80 | -Mistral -| | 8×7B | 61.83 | 10.43 26.11 | 25.09 25.59 | 20.66 | 93.59 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 32.74 | 11.07 49.25 | 42.69 45.74 | 20.75 | 98.28 | -Mistral-Instruct -| | 8×7B | 38.51 | 13.93 49.28 | 48.34 48.81 | 21.34 | 113.71 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -Post-hocRetrieval -| ChatGPT | - | 22.79 | 18.77 14.55 | 14.55 14.55 | 22.28 | 106.83 | -| ------------ | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 72.80 | 7.23 6.84 | 6.84 6.84 | 19.14 | 88.19 | -| LLaMA-2 | 13B | 53.21 | 10.33 9.61 | 9.61 9.61 | 20.63 | 90.44 | -| | 70B | 58.97 | 11.10 10.27 | 10.26 10.26 | 20.41 | 77.85 | -| | 7B | 22.50 | 14.17 11.33 | 11.33 11.33 | 21.17 | 110.04 | -| LLaMA-2-Chat | 13B | 30.36 | 14.93 12.10 | 12.10 12.10 | 21.82 | 109.79 | -| | 70B | 37.87 | 16.03 12.93 | 12.93 12.93 | 21.57 | 99.94 | -| | 7B | 30.88 | 11.83 10.91 | 10.91 10.91 | 21.66 | 99.03 | -Vicuna-v1.5 -| | 13B | 32.59 | 15.20 14.06 | 14.06 14.05 | 14.05 | 108.16 | -| --- | --- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 52.45 | 10.47 8.64 | 8.64 8.64 | 20.48 | 90.17 | -Mistral -| | 8×7B | 48.39 | 13.57 11.62 | 11.62 11.62 | 21.43 | 91.97 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ------ | -| | 7B | 27.41 | 17.07 13.20 | 13.20 13.20 | 21.52 | 106.93 | -Mistral-Instruct -| | 8×7B | 27.60 | 17.37 15.68 | 15.68 15.68 | 21.66 | 95.21 | -| --- | ---- | ----- | ----------- | ----------- | ----- | ----- | -Training-based -| | 7B | 30.98 | 6.90 22.34 | 32.40 26.45 | 16.48 | 41.66 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -Self-RAG -| | 13B | 32.04 | 6.07 30.46 | 40.20 34.66 | 15.23 | 38.19 | -| --- | --- | ----- | ---------- | ----------- | ----- | ----- | -| | 7B | 44.12 | 9.63 42.30 | 40.06 41.15 | 20.58 | 80.43 | -VANILLA-SFT -| | 13B | 46.33 | 10.27 46.75 | 44.47 45.58 | 20.56 | 84.01 | -| --- | --- | ----- | ----------- | ----------- | ----- | ----- | -| | 7B | 36.90 | 9.18 58.60 | 55.33 56.92 | 19.09 | 74.06 | -FRONT -| | 13B | 34.37 | 9.32 60.31 | 59.21 59.75 | 19.66 | 75.14 | -| --- | --- | -------- | ---------------- | ----------- | ----- | ----- | -| | | Table11: | ELI5fullresults. | | | | - -| | | Correctness | Citation | | -| --------- | --------- | ------------ | ---------- | ----------- | -| ModelType | ModelSize | | | | -| | | Rec.-5 Prec. | Rec. Prec. | F1 NumPred. | -Prompting-based -| ChatGPT | - | 20.28 19.84 | 19.06 22.03 | 20.44 4.71 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 12.56 11.32 | 6.03 6.35 | 6.19 7.02 | -| LLaMA-2 | 13B | 18.00 12.39 | 5.45 5.74 | 5.59 11.31 | -| | 70B | 18.50 14.79 | 10.10 10.50 | 10.30 8.31 | -| | 7B | 17.96 19.74 | 9.58 9.68 | 9.63 4.73 | -| LLaMA-2-Chat | 13B | 21.34 18.86 | 8.94 9.06 | 9.00 6.51 | -| | 70B | 22.62 18.04 | 13.49 13.98 | 13.73 7.44 | -| | 7B | 14.22 14.74 | 11.26 11.64 | 11.45 5.87 | -Vicuna-v1.5 -| | 13B | 22.06 19.60 | 13.04 13.74 | 13.38 7.62 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 16.96 15.98 | 7.50 7.76 | 7.63 6.29 | -Mistral -| | 8×7B | 18.18 15.63 | 9.72 10.20 | 9.95 6.63 | -| --- | ---- | ----------- | ----------- | ---------- | -| | 7B | 17.52 21.29 | 17.56 18.53 | 18.03 4.54 | -Mistral-Instruct -| | 8×7B | 20.12 19.64 | 19.27 20.38 | 19.81 5.32 | -| --- | ---- | ----------- | ----------- | ---------- | -Post-hocRetrieval -| ChatGPT | - | 25.14 22.85 | 12.29 12.29 | 12.29 5.46 | -| ------------ | --- | ----------- | ----------- | ---------- | -| | 7B | 6.48 5.11 | 5.05 5.05 | 5.05 6.55 | -| LLaMA-2 | 13B | 9.88 7.17 | 5.20 5.20 | 5.20 6.98 | -| | 70B | 14.44 12.44 | 7.49 7.49 | 7.49 7.41 | -| | 7B | 12.94 10.89 | 7.76 7.76 | 7.76 5.99 | -| LLaMA-2-Chat | 13B | 15.72 12.23 | 7.87 7.87 | 7.87 6.32 | -| | 70B | 17.90 14.45 | 9.05 9.05 | 9.05 6.05 | -| | 7B | 12.04 9.71 | 6.69 6.69 | 6.69 7.10 | -Vicuna-v1.5 -| | 13B | 14.78 11.47 | 8.50 8.50 | 8.50 6.67 | -| --- | --- | ----------- | --------- | --------- | -| | 7B | 9.94 7.90 | 6.00 6.00 | 6.00 7.38 | -Mistral -| | 8×7B | 13.92 12.08 | 6.70 6.70 | 6.70 6.58 | -| --- | ---- | ----------- | --------- | --------- | -| | 7B | 15.80 12.15 | 8.34 8.34 | 8.34 7.01 | -Mistral-Instruct -| | 8×7B | 24.16 18.28 | 9.78 9.78 | 9.78 7.37 | -| --- | ---- | ----------- | --------- | --------- | -Training-based -| | 7B | 2.34 1.98 | 10.53 18.80 | 13.50 3.49 | -| --- | --- | --------- | ----------- | ---------- | -Self-RAG -| | 13B | 1.90 1.33 | 12.79 20.90 | 15.86 3.08 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 12.86 21.09 | 21.35 21.36 | 21.35 7.49 | -VANILLA-SFT -| | 13B | 12.68 22.80 | 23.64 23.71 | 23.67 3.14 | -| --- | --- | ----------- | ----------- | ---------- | -| | 7B | 11.50 21.38 | 24.74 24.84 | 24.79 3.08 | -FRONT -| | 13B | 11.94 22.61 | 24.86 25.39 | 25.12 3.17 | -| --- | -------- | ------------------- | ----------- | ---------- | -| | Table12: | QAMPARIfullresults. | | | \ No newline at end of file +1. + +2. + +3. + +4. + +5. + +6. + +7. + +About These Forms + +In General. This and the other pleading forms available from the www.uscourts.gov +website illustrate some types of information that are useful to have in complaints and +some other pleadings. The forms do not try to cover every type of case. They are limited +to types of cases often filed in federal courts by those who represent themselves or who +may not have much experience in federal courts. + +Not Legal Advice. No form provides legal advice. No form substitutes for having or +consulting a lawyer. If you are not a lawyer and are suing or have been sued, it is best to +have or consult a lawyer if possible. + +No Guarantee. Following a form does not guarantee that any pleading is legally or +factually correct or sufficient. + +Variations Possible. A form may call for more or less information than a particular +court requires. The fact that a form asks for certain information does not mean that every +court or a particular court requires it. And if the form does not ask for certain +information, a particular court might still require it. Consult the rules and caselaw that +govern in the court where you are filing the pleading. + +Examples Only. The forms do not try to address or cover all the different types of +claims or defenses, or how specific facts might affect a particular claim or defense. Some +of the forms, such as the form for a generic complaint, apply to different types of cases. +Others apply only to specific types of cases. Be careful to use the form that fits your case +and the type of pleading you want to file. Be careful to change the information the form +asks for to fit the facts and circumstances of your case. + +No Guidance on Timing or Parties. The forms do not give any guidance on when +certain kinds of pleadings or claims or defenses have to be raised, or who has to be sued. +Some pleadings, claims, or defenses have to be raised at a certain point in the case or +within a certain period of time. And there are limits on who can be named as a party in a +case and when they have to be added. Lawyers and people representing themselves must +know the Federal Rules of Civil Procedure and the caselaw setting out these and other +requirements. The current Federal Rules of Civil Procedure are available, for free, at +www.uscourts.gov. + +Privacy Requirements. Federal Rule of Civil Procedure 5.2 addresses the privacy and +security concerns over public access to electronic court files. Under this rule, papers filed +with the court should not contain anyone’s full social-security number or full birth date; +the name of a person known to be a minor; or a complete financial-account number. A +filing may include only the last four digits of a social-security number and taxpayer +identification number; the year of someone’s birth; a minor’s initials; and the last four +digits of a financial-account number. + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +IN THE UNITED STATES DISTRICT COURT +FOR THE EASTERN DISTRICT OF MICHIGAN + +Case No. +________________________________ +(to be filled in by the Clerk’s Office) + +(Write the full name of each plaintiff who is filing this complaint. +If the names of all the plaintiffs cannot fit in the space above, +please write “see attached” in the space and attach an additional +page with the full list of names.) + +Jury Trial: ☐ Yes ☐ No + +(check one) + +v. + +(Write the full name of each defendant who is being sued. If the +names of all the defendants cannot fit in the space above, please +write “see attached” in the space and attach an additional page +with the full list of names.) + +Complaint for a Civil Case + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +I. + +The Parties to This Complaint + +A. + +The Plaintiff(s) + +Provide the information below for each plaintiff named in the complaint. Attach +additional pages if needed. + +Name + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address + +__________________________________________ + +B. + +The Defendant(s) + +Provide the information below for each defendant named in the complaint, whether the +defendant is an individual, a government agency, an organization, or a corporation. +For an individual defendant, include the person’s job or title (if known). Attach +additional pages if needed. + +Defendant No. 1 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +Defendant No. 2 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +2 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +Defendant No. 3 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +Defendant No. 4 + +Name + +Job or Title +(if known) + +__________________________________________ + +__________________________________________ + +Street Address + +__________________________________________ + +City and County + +__________________________________________ + +State and Zip Code + +__________________________________________ + +Telephone Number + +__________________________________________ + +E-mail Address +(if known) + +__________________________________________ + +II. + +Basis for Jurisdiction + +Federal courts are courts of limited jurisdiction (limited power). Generally, only two types of +cases can be heard in federal court: cases involving a federal question and cases involving +diversity of citizenship of the parties. Under 28 U.S.C. § 1331, a case arising under the United +States Constitution or federal laws or treaties is a federal question case. Under 28 U.S.C. +§ 1332, a case in which a citizen of one State sues a citizen of another State or nation and the +amount at stake is more than $75,000 is a diversity of citizenship case. In a diversity of +citizenship case, no defendant may be a citizen of the same State as any plaintiff. + +What is the basis for federal court jurisdiction? (check all that apply) + +☐ Federal question + +☐ Diversity of citizenship + +Fill out the paragraphs in this section that apply to this case. + +3 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +A. + +If the Basis for Jurisdiction Is a Federal Question + +List the specific federal statutes, federal treaties, and/or provisions of the United States +Constitution that are at issue in this case. + +B. + +If the Basis for Jurisdiction Is Diversity of Citizenship + +1. + +The Plaintiff(s) + +a. + +b. + +If the plaintiff is an individual +The plaintiff, (name) _________________________________________, +is a citizen of the State of (name) _______________________. + +If the plaintiff is a corporation +The plaintiff, (name) ________________________________________, +is incorporated under the laws of the State of (name) +______________________, and has its principal place of business in the +State of (name) _______________________. + +(If more than one plaintiff is named in the complaint, attach an additional page +providing the same information for each additional plaintiff.) + +2. + +The Defendant(s) + +a. + +b. + +If the defendant is an individual +The defendant, (name) ______________________, is a citizen of the +State of (name) _______________________. Or is a citizen of (foreign +nation) _______________________. + +If the defendant is a corporation +The defendant, (name) _______________________, is incorporated +under the laws of the State of (name) _______________________, and +has its principal place of business in the State of (name) +_______________________. Or is incorporated under the laws of +(foreign nation) _______________________, and has its principal place +of business in (name) _______________________. + +(If more than one defendant is named in the complaint, attach an additional +page providing the same information for each additional defendant.) + +4 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +3. + +The Amount in Controversy + +The amount in controversy—the amount the plaintiff claims the defendant owes +or the amount at stake—is more than $75,000, not counting interest and costs of +court, because (explain): + +III. + +Statement of Claim + +Write a short and plain statement of the claim. Do not make legal arguments. State as briefly +as possible the facts showing that each plaintiff is entitled to the damages or other relief sought. +State how each defendant was involved and what each defendant did that caused the plaintiff +harm or violated the plaintiff’s rights, including the dates and places of that involvement or +conduct. If more than one claim is asserted, number each claim and write a short and plain +statement of each claim in a separate paragraph. Attach additional pages if needed. + +5 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +IV. Relief + +State briefly and precisely what damages or other relief the plaintiff asks the court to order. Do +not make legal arguments. Include any basis for claiming that the wrongs alleged are +continuing at the present time. Include the amounts of any actual damages claimed for the acts +alleged and the basis for these amounts. Include any punitive or exemplary damages claimed, +the amounts, and the reasons you claim you are entitled to actual or punitive money damages. + +V. + +Certification and Closing + +Under Federal Rule of Civil Procedure 11, by signing below, I certify to the best of my +knowledge, information, and belief that this complaint: (1) is not being presented for an +improper purpose, such as to harass, cause unnecessary delay, or needlessly increase the cost of +litigation; (2) is supported by existing law or by a nonfrivolous argument for extending, +modifying, or reversing existing law; (3) the factual contentions have evidentiary support or, if +specifically so identified, will likely have evidentiary support after a reasonable opportunity for +further investigation or discovery; and (4) the complaint otherwise complies with the +requirements of Rule 11. + +A. + +For Parties Without an Attorney + +I agree to provide the Clerk’s Office with any changes to my address where case- +related papers may be served. I understand that my failure to keep a current address on +file with the Clerk’s Office may result in the dismissal of my case. + +Date of signing: _____________________, 20_____. + +Signature of Plaintiff +___________________________________________ +Printed Name of Plaintiff ___________________________________________ + +6 + + MIED ProSe 1 (Rev 5/16) Complaint for a Civil Case + +Additional Information: + +7 + + JS 44 (Rev. 10/20) + +CIVIL COVER SHEET +The JS 44 civil cover sheet and the information contained herein neither replace nor supplement the filing and service of pleadings or other papers as required by law, except as +provided by local rules of court. This form, approved by the Judicial Conference of the United States in September 1974, is required for the use of the Clerk of Court for the +purpose of initiating the civil docket sheet. (SEE INSTRUCTIONS ON NEXT PAGE OF THIS FORM.) +I. (a) PLAINTIFFS + +County in which action arose: _______________ + +DEFENDANTS + +(b) County of Residence of First Listed Plaintiff + +(EXCEPT IN U.S. PLAINTIFF CASES) + +County of Residence of First Listed Defendant + +NOTE: + +(IN U.S. PLAINTIFF CASES ONLY) +IN LAND CONDEMNATION CASES, USE THE LOCATION OF +THE TRACT OF LAND INVOLVED. + +(c) Attorneys (Firm Name, Address, and Telephone Number) + +Attorneys (If Known) + +II. BASIS OF JURISDICTION (Place an “X” in One Box Only) + +III. CITIZENSHIP OF PRINCIPAL PARTIES (Place an “X” in One Box for Plaintiff + +1 U.S. Government +Plaintiff + +3 Federal Question + +(U.S. Government Not a Party) + +Citizen of This State + +PTF +1 + +DEF +1 + +(For Diversity Cases Only) + +and One Box for Defendant) + +PTF + +Incorporated or Principal Place +of Business In This State + +2 U.S. Government + +Defendant + +4 Diversity + +Citizen of Another State + +(Indicate Citizenship of Parties in Item III) + +DEF +4 + +5 + +6 + +4 + +5 + +6 + +2 + +3 + +2 + +Incorporated and Principal Place +of Business In Another State + +3 + +Foreign Nation + +Click here for: Nature of Suit Code Descriptions. +OTHER STATUTES + +BANKRUPTCY + +IV. NATURE OF SUIT (Place an “X” in One Box Only) + +TORTS + +CONTRACT + +110 Insurance +120 Marine +130 Miller Act +140 Negotiable Instrument +150 Recovery of Overpayment + +PERSONAL INJURY +310 Airplane +315 Airplane Product + +Liability + +320 Assault, Libel & + +& Enforcement of Judgment + +Slander + +151 Medicare Act +152 Recovery of Defaulted + +Student Loans +(Excludes Veterans) +153 Recovery of Overpayment +of Veteran’s Benefits + +160 Stockholders’ Suits +190 Other Contract +195 Contract Product Liability +196 Franchise + +REAL PROPERTY +210 Land Condemnation +220 Foreclosure +230 Rent Lease & Ejectment +240 Torts to Land +245 Tort Product Liability +290 All Other Real Property + +330 Federal Employers’ + +Liability + +340 Marine +345 Marine Product +Liability +350 Motor Vehicle +355 Motor Vehicle + +Product Liability + +360 Other Personal + +Injury + +362 Personal Injury - + +Medical Malpractice + +CIVIL RIGHTS +440 Other Civil Rights +441 Voting +442 Employment +443 Housing/ + +Accommodations +445 Amer. w/Disabilities - + +Employment + +446 Amer. w/Disabilities - + +Other +448 Education + +Citizen or Subject of a +Foreign Country + +FORFEITURE/PENALTY + +625 Drug Related Seizure + +of Property 21 USC 881 + +690 Other + +PERSONAL INJURY +365 Personal Injury - +Product Liability + +367 Health Care/ + +Pharmaceutical +Personal Injury +Product Liability + +368 Asbestos Personal +Injury Product +Liability + +422 Appeal 28 USC 158 +423 Withdrawal +28 USC 157 + +PROPERTY RIGHTS +820 Copyrights +830 Patent +835 Patent - Abbreviated + +New Drug Application + +840 Trademark +880 Defend Trade Secrets + +PERSONAL PROPERTY + +LABOR + +710 Fair Labor Standards + +Act of 2016 + +Act + +720 Labor/Management + +Relations + +740 Railway Labor Act +751 Family and Medical + +Leave Act + +790 Other Labor Litigation +791 Employee Retirement +Income Security Act + +IMMIGRATION + +462 Naturalization Application +465 Other Immigration + +Actions + +SOCIAL SECURITY +861 HIA (1395ff) +862 Black Lung (923) +863 DIWC/DIWW (405(g)) +864 SSID Title XVI +865 RSI (405(g)) + +FEDERAL TAX SUITS +870 Taxes (U.S. Plaintiff +or Defendant) +871 IRS—Third Party +26 USC 7609 + +370 Other Fraud +371 Truth in Lending +380 Other Personal + +Property Damage +385 Property Damage +Product Liability + +PRISONER PETITIONS + +Habeas Corpus: +463 Alien Detainee +510 Motions to Vacate + +Sentence +530 General +535 Death Penalty +Other: +540 Mandamus & Other +550 Civil Rights +555 Prison Condition +560 Civil Detainee - +Conditions of +Confinement + +375 False Claims Act +376 Qui Tam (31 USC + +3729(a)) + +400 State Reapportionment +410 Antitrust +430 Banks and Banking +450 Commerce +460 Deportation +470 Racketeer Influenced and +Corrupt Organizations + +480 Consumer Credit + +(15 USC 1681 or 1692) + +485 Telephone Consumer +Protection Act +490 Cable/Sat TV +850 Securities/Commodities/ + +Exchange + +890 Other Statutory Actions +891 Agricultural Acts +893 Environmental Matters +895 Freedom of Information + +Act +896 Arbitration +899 Administrative Procedure +Act/Review or Appeal of +Agency Decision +950 Constitutionality of +State Statutes + +V. ORIGIN (Place an “X” in One Box Only) + +1 Original + +Proceeding + +2 Removed from +State Court + +3 Remanded from +Appellate Court + +4 Reinstated or +Reopened + +5 Transferred from +Another District +(specify) + +6 Multidistrict +Litigation - +Transfer + +8 Multidistrict +Litigation - +Direct File + +Cite the U.S. Civil Statute under which you are filing (Do not cite jurisdictional statutes unless diversity): + +VI. CAUSE OF ACTION + +Brief description of cause: + +VII. REQUESTED IN + +COMPLAINT: +VIII. RELATED CASE(S) + +IF ANY + +CHECK IF THIS IS A CLASS ACTION +UNDER RULE 23, F.R.Cv.P. + +DEMAND $ + +CHECK YES only if demanded in complaint: + +JURY DEMAND: + +Yes + +No + +(See instructions): + +JUDGE + +DOCKET NUMBER + +DATE + +SIGNATURE OF ATTORNEY OF RECORD + +FOR OFFICE USE ONLY + +RECEIPT # + +AMOUNT + +APPLYING IFP + +JUDGE + +MAG. JUDGE + + PURSUANT TO LOCAL RULE 83.11 + +1. + +Is this a case that has been previously dismissed? + +If yes, give the following information: + +Court: + +Case No.: + +Judge: + +2. + +Other than stated above, are there any pending or previously +discontinued or dismissed companion cases in this or any other +court, including state court? (Companion cases are matters in which +it appears substantially similar evidence will be offered or the same +or related parties are present and the cases arise out of the same +transaction or occurrence.) + +Yes + +No + +Yes + +No + +If yes, give the following information: + +Court: + +Case No.: + +Judge: + +Notes : + + JS 44 Reverse (Rev. 10/20) + +INSTRUCTIONS FOR ATTORNEYS COMPLETING CIVIL COVER SHEET FORM JS 44 + +Authority For Civil Cover Sheet + +The JS 44 civil cover sheet and the information contained herein neither replaces nor supplements the filings and service of pleading or other papers as +required by law, except as provided by local rules of court. This form, approved by the Judicial Conference of the United States in September 1974, is +required for the use of the Clerk of Court for the purpose of initiating the civil docket sheet. Consequently, a civil cover sheet is submitted to the Clerk of +Court for each civil complaint filed. The attorney filing a case should complete the form as follows: + +I.(a) Plaintiffs-Defendants. Enter names (last, first, middle initial) of plaintiff and defendant. If the plaintiff or defendant is a government agency, use + +only the full name or standard abbreviations. If the plaintiff or defendant is an official within a government agency, identify first the agency and then +the official, giving both name and title. + +(b) County of Residence. For each civil case filed, except U.S. plaintiff cases, enter the name of the county where the first listed plaintiff resides at the +time of filing. In U.S. plaintiff cases, enter the name of the county in which the first listed defendant resides at the time of filing. (NOTE: In land +condemnation cases, the county of residence of the "defendant" is the location of the tract of land involved.) + +(c) Attorneys. Enter the firm name, address, telephone number, and attorney of record. If there are several attorneys, list them on an attachment, noting + +in this section "(see attachment)". + +II. + +III. + +IV. + +V. + +Jurisdiction. The basis of jurisdiction is set forth under Rule 8(a), F.R.Cv.P., which requires that jurisdictions be shown in pleadings. Place an "X" +in one of the boxes. If there is more than one basis of jurisdiction, precedence is given in the order shown below. +United States plaintiff. (1) Jurisdiction based on 28 U.S.C. 1345 and 1348. Suits by agencies and officers of the United States are included here. +United States defendant. (2) When the plaintiff is suing the United States, its officers or agencies, place an "X" in this box. +Federal question. (3) This refers to suits under 28 U.S.C. 1331, where jurisdiction arises under the Constitution of the United States, an amendment +to the Constitution, an act of Congress or a treaty of the United States. In cases where the U.S. is a party, the U.S. plaintiff or defendant code takes +precedence, and box 1 or 2 should be marked. +Diversity of citizenship. (4) This refers to suits under 28 U.S.C. 1332, where parties are citizens of different states. When Box 4 is checked, the +citizenship of the different parties must be checked. (See Section III below; NOTE: federal question actions take precedence over diversity +cases.) + +Residence (citizenship) of Principal Parties. This section of the JS 44 is to be completed if diversity of citizenship was indicated above. Mark this +section for each principal party. + +Nature of Suit. Place an "X" in the appropriate box. If there are multiple nature of suit codes associated with the case, pick the nature of suit code +that is most applicable. Click here for: Nature of Suit Code Descriptions. + +Origin. Place an "X" in one of the seven boxes. +Original Proceedings. (1) Cases which originate in the United States district courts. +Removed from State Court. (2) Proceedings initiated in state courts may be removed to the district courts under Title 28 U.S.C., Section 1441. +Remanded from Appellate Court. (3) Check this box for cases remanded to the district court for further action. Use the date of remand as the filing +date. +Reinstated or Reopened. (4) Check this box for cases reinstated or reopened in the district court. Use the reopening date as the filing date. +Transferred from Another District. (5) For cases transferred under Title 28 U.S.C. Section 1404(a). Do not use this for within district transfers or +multidistrict litigation transfers. +Multidistrict Litigation – Transfer. (6) Check this box when a multidistrict case is transferred into the district under authority of Title 28 U.S.C. +Section 1407. +Multidistrict Litigation – Direct File. (8) Check this box when a multidistrict case is filed in the same district as the Master MDL docket. +PLEASE NOTE THAT THERE IS NOT AN ORIGIN CODE 7. Origin Code 7 was used for historical records and is no longer relevant due to +changes in statue. + +VI. + +Cause of Action. Report the civil statute directly related to the cause of action and give a brief description of the cause. Do not cite jurisdictional +statutes unless diversity. Example: U.S. Civil Statute: 47 USC 553 Brief Description: Unauthorized reception of cable service. + +VII. Requested in Complaint. Class Action. Place an "X" in this box if you are filing a class action under Rule 23, F.R.Cv.P. + +Demand. In this space enter the actual dollar amount being demanded or indicate other demand, such as a preliminary injunction. +Jury Demand. Check the appropriate box to indicate whether or not a jury is being demanded. + +VIII. Related Cases. This section of the JS 44 is used to reference related pending cases, if any. If there are related pending cases, insert the docket + +numbers and the corresponding judge names for such cases. + +Date and Attorney Signature. Date and sign the civil cover sheet. + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Contract +Code +110 + +Title +Insurance + +120 + +Marine + +130 + +Miller Act + +140 + +Negotiable Instrument + +150 + +Recovery of Overpayment & Enforcement +Judgment + +151 + +Medicare + +152 + +153 + +160 + +Recovery of Defaulted Student Loans +(Excludes Veterans) +Recovery of Overpayment of Veterans' +Benefits +Stockholders' Suits + +190 + +Other Contract + +195 + +Contract Product Liability + +Actions primarily alleging personal injury +or property damage caused by a defective +product should be classified under the +appropriate nature of suit code under +“TORTS.” +Franchise + +196 + +Description +Action alleging breach of insurance contract, tort claim, or other cause +related to an insurance contract, except for maritime insurance +contracts. +Action (Admiralty or Maritime) based on service, employment, +insurance or other contracts relating to maritime vessels and other +maritime contractual matters. +Action based on performance and payment bonds agreed to by +contractors on federal construction projects as required under the +Miller Act, 40 U.S.C. § 3131-3134. +Action relating to an agreement to pay a specific amount of money, +including promissory notes, loan agreements and checks. +Action to recover debt owed to the United States, including +enforcement of judgments, based on overpayments and restitution +agreements involving matters other than Medicare benefits, student +loans and veterans’ benefits. +Action relating to Medicare payments, including actions for payments of +benefits, to recover overpayments, and for judicial review of +administrative decisions. +Action to recover debt owed to the United States from defaulted +student loan. +Action relating to payments of veterans’ benefits, primarily including +actions to recover overpayments. +Action brought by stockholder(s) of a corporation (including both +stockholder derivative suits and direct actions based on plaintiff’s rights +as a stockholder), usually alleging claims based on contract and/or tort +law and/or fiduciary obligations. +Action primarily based on rights and obligations under a contract not +classifiable elsewhere under the specific natures of suit under +“Contract.” +Action concerning damages caused by a defective product, not primarily +involving personal injury or property damage, and based primarily on +breach of contract, breach of warranty, misrepresentation, and/or +violation of consumer protection laws. + +Action arising from a dispute over a franchise agreement, typically +alleging breach of contract, misrepresentation or unfair trade practices. + +Page 1 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Real Property + +Code +210 + +Title +Land Condemnation + +220 + +230 + +240 + +Foreclosure + +Rent Lease & Ejectment + +Torts to Land + +245 + +Tort Product Liability + +290 + +All Other Real Property + +Torts/Personal Injury + +Title +Airplane +(Excludes airplane product liability +claims) +Airplane Product Liability + +Description +Action by a governmental entity to take privately-owned real property +(land or buildings) for public use for compensation. +Action to enjoin foreclosure on real property by mortgage lender. + +Action for rental or lease payments owed on real property and/or to +eject a party occupying real property illegally. +Action alleging trespass to land, nuisance, contamination or other +unlawful entry on or interference with real property possessed by +another. +Action alleging harm by an unsafe product based on negligence, breach +of warranty, misrepresentation, and strict tort liability. +Action primarily based on unlawful conduct relating to real property +that cannot be classified under any other nature of suit. + +Description +Action alleging personal injury or wrongful death from an air crash or +other occurrence involving an airplane. + +Code +310 + +315 + +320 + +330 + +340 + +345 + +350 + +355 + +360 + +362 + +365 + +367 + +368 + +Motor Vehicle + +Marine Product Liability + +Marine +(Excludes marine product liability claims) + +Assault, Libel & Slander +(Excludes a government employee) +Federal Employers' Liability + +Action alleging personal injury or death from an air crash or other +occurrence involving an airplane and caused by a defective product. +Action alleging intentional acts of assault, libel, trade libel or slander by +a private party. +Action for personal injury or wrongful death brought by a railroad +employee or his survivors under the Federal Employers’ Liability Act +(FELA), 45 U.S.C. § 51, et. seq. +Action (Admiralty and Maritime) alleging personal injury or death from +an accident involving a water vessel or harbor/dock facilities, including +suits brought under the Jones Act and the Limitation of Liability Act. +Action (Admiralty and Maritime) alleging personal injury or wrongful +death from an accident involving a water vessel or harbor/dock +facilities and caused by a defective product. +Action alleging personal injury or wrongful death from negligence +involving a motor vehicle but not caused by a defective product. +Action alleging personal injury or wrongful death involving a motor +vehicle and caused by a defective product. +Action primarily based on personal injury or death caused by +negligence or intentional misconduct, including suits brought against +the United States under the Federal Tort Claims Act, and which cannot +be classified under any other nature of suit. +Action alleging personal injury or wrongful death caused by negligence +in medical care provided by a doctor or other health care professional. +Action alleging personal injury or death resulting from a defective +Personal Injury - Product Liability +product. +(Excludes a marine or airplane product) +Action alleging personal injury or death caused by a defective medical +Health Care/Pharmaceutical Personal +or pharmaceutical product. +Injury Product Liability +Asbestos Personal Injury Product Liability Action alleging personal injury or death caused by exposure to asbestos + +Personal Injury - Medical Malpractice + +Motor Vehicle Product Liability + +Other Personal Injury + +products. + +Page 2 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Personal Property + +Code +370 + +371 + +Title +Other Fraud +(Excludes any property that is not real +property) +Truth in Lending + +Actions relating to fraud or +misrepresentation in the transfer of real +property should be classified under +nature of suit 290, “All Other Real +Property,” or, if foreclosure is involved, +under nature of suit 220, “Foreclosure.” +Other Personal Property Damage + +380 + +385 + +Property Damage Product Liability + +Civil Rights + +Code + +440 + +441 + +442 + +443 + +445 + +446 + +Title +Other Civil Rights +(Excludes claims against corrections +officials) +Voting + +Employment + +Housing/Accommodations + +Americans With Disabilities - +Employment +Americans With Disabilities - Other + +448 + +Education + +Description +Action primarily based on fraud relating to personal property that +cannot be classified under any other nature of suit. + +Action alleging violation of the federal Truth in Lending Act arising from +consumer loan transactions involving personal property including +automobile loans and revolving credit accounts. + +Action primarily based on damage to personal property caused by +harmful conduct such as negligence, misrepresentation, interference +with business relationships or unfair trade practices. +Action alleging damage to personal property caused by a defective +product. + +Description + +Action alleging a civil rights violation other than the specific civil rights +categories listed below or a violation related to prison. Example: +Action alleging excessive force by police incident to an arrest. +Action filed under Civil Rights Act, 52 U.S.C. § 10101, and Voting Rights +Act, 52 U.S.C. § 10301 +Action filed under Age Discrimination in Employment Act 29:621:634, +Equal Employment Opportunity Act (Title VII) 42:2000E, Performance +Rating Act of 1950 5:4303 +Action filed under the Fair Housing Act (Title VII), 42 U.S.C. § 3601 & +3602. +Action of discrimination against an employee with disabilities of any +type in the work place, filed under 42 U.S.C. § 12117 +Action of discrimination against an individual with disabilities in areas +other than employment, filed under 42 U.S.C. § 12133 (exclusion or +discrimination in provision of services, programs or activities of a public +entity) or 42 U.S.C. § 12188 (public accommodations) +Action filed under the Individuals with Disabilities Educations Act, 20 +U.S.C. § 1401 and Title IX of the Education Amendment of 1972, 20 +U.S.C. § 1681 et seq. + +Page 3 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Prisoner Petitions +Habeas Corpus +Title +Alien Detainee + +Code +463 + +510 + +Motions to Vacate Sentence + +530 + +General + +535 + +Death Penalty + +Other +Prisoner Petitions + +Code +540 + +Title +Mandamus & Other + +550 + +Civil Rights + +555 + +Prison Condition + +560 + +Civil Detainee - Conditions of +Confinement (Excludes actions by alien +detainees) + +Description +Immigration habeas petition under 28 U.S.C. § 2241. All cases filed +with this nature of suit code are restricted to case participants and +public terminals. Petition is filed by an alien detainee. +Action by a prisoner to vacate or modify a sentence imposed in federal +court, other than a death sentence, under 28 U.S.C. § 2255. +Action by a federal or state prisoner currently in custody challenging +the legality of confinement or other punishment. This includes claims +alleging illegalities that occurred in trial (for example, ineffective +assistance of counsel), sentencing (including fines and restitution +orders), or disciplinary proceedings in prison (for example, loss of good +time credits). Habeas petition under 28 U.S.C. § 2254 or prisoner +habeas under 28 U.S.C. § 2241. +Action by a federal or state prisoner challenging a death sentence. + +Description +Action by prisoner currently in custody for a writ of mandamus to +compel action by a judge or government official relating to the +prisoner’s confinement, including conditions of confinement. This +category also includes any actions other than mandamus brought by a +prisoner currently in custody, whether or not it relates to his +confinement, if it is not classifiable under any other nature of suit +category under Prisoner Petitions (for example, action by prisoner to +recover property taken by the government in a criminal case). +Action by current or former prisoner alleging a civil rights violation by +corrections officials that is not related to a condition of prison life. +Action by current prisoner, or former prisoner or their families alleging +a civil rights, Federal Tort Claims Act, or state law claim with respect to +a condition of prison life, whether general circumstances or particular +episodes. Examples: inadequate medical care or excessive force by +prison guards. Includes non-habeas actions by alien detainees alleging +unlawful prison conditions. +Action by former prisoner who was involuntarily committed to a non- +criminal facility after expiration of his or her prison term alleging +unlawful conditions of confinement while in the non-criminal facility. +This category includes, for example, an action by a former prisoner +classified as a Sexually Dangerous Person or Sexually Violent Predator +alleging civil rights violations during his detention in a medical facility. + +Page 4 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Forfeiture/Penalty + +Code +625 + +Title +Drug Related Seizure of Property 21 +U.S.C. § 881 + +Description +Action (Forfeiture) by which property itself is accused of wrongdoing +and is forfeited to the government as a result. + +690 + +Other + +Labor +Code +710 + +Title +Fair Labor +Standards Act (Non-Union) + +720 + +Labor/Management Relations (Union) + +740 + +Railway Labor Act + +Action primarily based on Acts or Bills that cannot be classified under +any other nature of suit, such as: Endangered Species Act, Federal +Hazardous Substance Act 15:1261, Game & Wildlife Act 15:256C et seq. +(Penalty), Federal Trade Commission Act 15:41-51 (Penalty), Federal +Coal Mine Health & Safety Act 30:801 et seq. (Penalty), Load Line Act +46:85-85G, McGuire Bill (Federal Fair Trade) 15:45L Penalty, Marihuana +Tax Act 50 STAT 551, Motorboat Act 46:526-526T, National Traffic & +Motor Vehicle Safety Act penalty 49:1655, Veterans' Benefit Act, Title +38 Penalty. + +Description +Action relating to non-union workplace related disputes filed under the +Fair Labor Standards Act, 29 U.S.C. § 201 including but not limited to +wage discrimination, paid leave, minimum wage and overtime pay. + +Action relating to disputes between labor unions and employers as well +as all petitions regarding actions of the Nation Labor Relations Board +(NLRB) + +Action relating to disputes filed under the Railway Labor Act, 45 U.S.C. +§ 151 including labor disputes, individual claims, and response to +sanctions. + +751 + +790 + +791 + +Family and Medical Leave Act + +Action filed under the Family Medical Leave Act, 29 U.S.C. § 2601 + +Other Labor Litigation + +Employee Retirement Income Security +Act + +Action primarily based on labor disputes not addressed by other NOS +codes (includes Labor/Management Reporting and Disclosure Act) +Action filed under the Employee Retirement Income Security Act, 29 +U.S.C. § 1132 by individuals and labor organizations. + +Immigration + +Code +462 + +Title +Naturalization Application + +465 + +Other Immigration +Actions + +Description +Action seeking review of denial of an application for naturalization [8 +U.S.C. § 1447(b)] or alleging failure to make a determination regarding +an application for naturalization [8 U.S.C. § 1421(c)]. + +Action (Immigration-related) that do not involve Naturalization +Applications or petitions for Writ of Habeas Corpus, such as complaints +alleging failure to adjudicate an application to adjust immigration +status to permanent resident. + +Page 5 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Bankruptcy + +Code +422 + +Title +Appeal 28 USC § 158 + +423 + +Withdrawal of Reference 28 USC § 157 + +Property Rights +Title +Copyright + +Code +820 + +Patent + +Patent – Abbreviated New Drug +Application (ANDA) + +Trademark + +830 + +835 + +840 + +880 + +Description +All appeals of previous bankruptcy decisions filed under +28 U.S.C. § 158 +Action held in bankruptcy court requesting withdrawal under the +provisions of 28 U.S.C. § 157 + +Description +Action filed in support or to dispute a copyright claim. + +Action filed in support or to dispute a patent claim. + +Action filed in support or to dispute a patent claim involving an +Abbreviated New Drug Application (ANDA). These cases are also known +as “Hatch-Waxman” cases. +Action filed in support or to dispute a trademark claim + +Defend Trade Secrets Act of 2016 (DTSA) + +Action filed in support or to dispute a trade secret misappropriation + +Social Security +Title +HIA (1395ff) + +Code +861 + +862 + +Black Lung (923) + +863 + +DIWC/DIWW (405(g)) + +864 + +865 + +SSID Title XVI + +RSI (405(g)) + +Federal Tax Suits + +Code +870 +871 + +Title +Taxes (U.S. Plaintiff or Defendant) +IRS-Third Party 26 USC 7609 + +Description +Action filed with regard to social security benefits associated with +Health Insurance Part A Medicare + +Action filed with regard to social security benefits provided for those +who contracted Black Lung or their beneficiaries + +Action filed with regard to social security benefits provided to disabled +individuals: worker or child, or widow + +Action filed with regard to social security benefits provided to +Supplemental Security Income Disability under Title XVI +Action filed with regard to social security benefits provided for +Retirement, Survivor Insurance under 42 U.S.C. § 405 + +Description +Action filed under the Internal Revenue Code (General) +Action filed under the Internal Revenue Code - Tax Reform Act of 1976 +(P.L. 94-455) Third Party + +Page 6 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +Other Statutes +Title +False Claims Act + +Code +375 + +376 + +Qui Tam (31 U.S.C. § 3729(a)) + +400 + +State Reapportionment + +410 + +Antitrust + +430 + +450 + +460 + +470 + +480 + +485 + +Banks and Banking + +Commerce + +Deportation + +Racketeer Influenced and Corrupt +Organizations +Consumer Credit + +Telephone Consumer Protection Act +(TCPA) + +490 + +Cable/Satellite TV + +850 + +Securities/Commodities/Exchange + +890 + +Other Statutory Actions + +891 + +Agricultural Acts + +893 + +Environmental Matters + +Description +Action filed by private individuals alleging fraud against the U.S. +Government under 31 U.S.C. § 3729. +Action brought under the False Claims Act by private persons (also +known as "whistleblowers") on their own behalf and on behalf of the +United States to recover damages against another person or entity that +acted fraudulently in receiving payments or property from, or avoiding +debts owed to, the United States Government, 31 U.S.C. § 3730 +Action filed under the Reapportionment Act of 1929 Ch. 28, 46 Stat. 21, +2 U.S.C. § 2a. +Action brought under the Clayton Act 15 U.S.C. § 12 - 27 alleging +undue restriction of trade and commerce by designated methods that +limit free competition in the market place amongst consumers such as +anti-competitive price discrimination, corporate mergers, interlocking +directorates or tying and exclusive dealing contracts. +Action filed under the Federal Home Loan Bank Act 12:1421-1449, +Home Owners Loan Act 12:1461 or Federal Reserve Acts 12:142 et seq. +Action filed under the Interstate Commerce Acts 49:1 et seq., 49:301 + +Action filed under the Immigration Acts (Habeas Corpus & Review) +8:1101/18:1546 +Racketeer Influenced and Corrupt Organization Act, RICO 18:1961-1968 + +Action filed under the Fair Credit Reporting Act, 15 U.S.C. 1681n or +15 U.S.C. 1681o, and the Fair Debt Collection Practices Act, +15 U.S.C. § 1692k +Action filed under the Telephone Consumer Protection Act 47 U.S.C. § +227 + +Action filed involving unauthorized reception of cable/satellite TV +service under 47 U.S.C. § 553 (unauthorized reception of cable/satellite +TV), or 47 U.S.C. § 605 (e)(3) (unauthorized use or publication of a +communication) +Action filed under Small Business Investment Act 15:681, Securities +Exchange Act 15:78, Securities Act 15:77, Investment Advisers Act +15:80B(1-21) +Action primarily based on Statutes that cannot be classified under any +other nature of suit, such as: Foreign Agents Registration Act 22:611- +621, Klamath Termination Act 25:564-564W-L, Federal Aid Highway Act +23:101-142, Federal Corrupt Practices Act 2:241-256, Federal Election +Campaign Act, Highway Safety Act 23:401 Immigration & Nationality +Act 8:1503, Natural Gas Pipeline Safety Act 49:1671-1700, +Naturalization Acts 8:1421/18:911, 1015, 1421, et seq., 3282 or Federal +Aviation Act 49:1301 et seq. +Action filed under the Federal Crop Insurance Act 7:1501-1550, +Commodity Credit Corporation Act 15:713A-L & 4. +Action filed under Air Pollution Control Act 42:1857-57L, Clean Air Act +42:1857:57L, Federal Environment Pesticide Control Act, Federal +Insecticide, Fungicide & Rodenticide Act 7:135, Federal Water Pollution +Control Act 33:1151 et seq., Land & Water Conservation Fund Act +16:4602,460 1-4, Motor Vehicle Air Pollution Control Act 42:1857F-1-8, + +Page 7 of 8 + + Civil Nature of Suit Code Descriptions +(Rev. 10/20) + +National Environmental Policy Act 42:4321, 4331-35G, 4341-47, River & +Harbor Act penalty 3:401-437, 1251. + +Other Statutes (Continued) +Title +Freedom of Information Act +Arbitration + +Code +895 +896 + +899 + +Administrative Procedure Act/Review or +Appeal of Agency Decision + +950 + +Constitutionality of State Statutes + +Description +Action filed under the Freedom of Information Act 5:552. +Action involving actions to confirm or modify arbitration awards filed +under Title 9 of the U.S. Code. +Action filed under the Administrative Procedures Act, 5 U.S.C. § 701, or +civil actions to review or appeal a federal agency decision. + +Action drawing into question the constitutionality of a federal or state +statute filed under (Rule 5.1). Rule 5.1 implements 28 U.S.C. §2403. + +Note: The statutes listed above are not all-inclusive, and other statues might be applicable to each nature of suit. + +Statutes that are included in the descriptions should be viewed as examples. + +Page 8 of 8 + diff --git a/q128/random_k4/random_3.md b/q128/random_k4/random_3.md index 8ca690a89c7b73d0943de9dc64d3de810d4e2fe2..74dbcea637a806c66910138226ab3f36206382f5 100644 --- a/q128/random_k4/random_3.md +++ b/q128/random_k4/random_3.md @@ -1,891 +1,390 @@ -A Practical Guide - -Graduate Studies: A Practical Guide - - November 2012 - -Note: This Guide was published by CAGS in 1992. The booklet is no longer in print. Although -there have been changes in some aspects of graduate studies since 1992, much of this Guide is -still relevant. CAGS is still making this document available in web format as a service to -students, faculty and others involved in graduate programs. - -Copyright © Canadian Association for Graduate Studies, 1992 - -The 2003 version was edited by Martha Crago -The 2012 version was edited by Samantha Boland - - Forward - -This booklet is addressed to students considering Masters and Doctoral programs at Canadian -universities. It discusses some of the motives for pursuing graduate studies and provides -prospective students with information on things to consider when examining the option of -graduate school. There is a tremendous diversity to graduate programs and thus some sections -may pertain more to some areas of study than others. - -From the perspective of students, graduate studies are not an end in themselves, but rather a -means to an end. In most cases, that end is gainful employment. Graduate studies therefore -constitute an investment of time and money for the purpose of career development. In some -cases the investment is essential, in others it is a matter of choice. From the perspective of -supervisors, graduate studies are an almost indispensable part of academic life and sometimes -central to successful research. Graduate students are needed for assistance, collaboration, and as -a source of inspiration, energy and fulfillment. They are also important partners for obtaining -external research funding from provincial and federal granting councils. From the perspective of -departments, research centers, and other units or combination of units offering graduate -programs, graduate studies are an essential component of research. What distinguishes -universities from other academic institutions is their mandate to carry out research and train -researchers. - -Graduate programs maintain and enhance the academic standing of a university and have a -significant bearing on the public perception of the quality of the graduates of that university. As -is the case in most partnerships, the probability of success in graduate studies is enhanced by a -set of explicit checks and balances which articulate the respective rights and responsibilities of -the partners. - -Graduate school can be, and often is, one of the highlights in a student’s life. Students are -provided with the opportunity to work in exciting environments and share experiences with like -minded colleagues that can result in lifelong friendships. Many former graduate students look -back on this time in their life with fondness and the realization that it represented an important -period of personal development that laid the foundation for a rewarding career. - -2 - - Table of Contents - -Forward ........................................................................................................................................... 2 -Why Pursue Graduate Studies? ....................................................................................................... 4 -Skills Required Before Entering Graduate Programs ................................................................. 4 -Representative Skills Acquired in Graduate Programs ............................................................... 4 -What is Graduate Education? ......................................................................................................... 5 -Preparation for Graduate School ..................................................................................................... 6 -Making the Decision ....................................................................................................................... 7 -Making Contacts ......................................................................................................................... 7 -Choosing an Institution ............................................................................................................... 7 -Choosing a Supervisor ................................................................................................................ 8 -Financing a Graduate Education ..................................................................................................... 9 -Funding Sources........................................................................................................................ 10 -Moving Along ............................................................................................................................... 11 -Acquiring Formal Training and Certifications ......................................................................... 11 -Choosing a Research Topic ...................................................................................................... 11 -Setting Milestones ..................................................................................................................... 12 -Finishing Up ................................................................................................................................. 13 -Analysis, Synthesis, and Writing .............................................................................................. 13 -Presenting and Publishing ......................................................................................................... 14 -A Note about Intellectual Property Rights ................................................................................ 14 -Additional Living Considerations ................................................................................................. 15 -Living Expenses ........................................................................................................................ 15 -Housing ..................................................................................................................................... 15 -Parental leave ............................................................................................................................ 15 -Childcare ................................................................................................................................... 16 -Questionnaires .............................................................................................................................. 16 -For Students .............................................................................................................................. 16 -For Supervisors ......................................................................................................................... 17 -For Departments........................................................................................................................ 17 - -3 - - Why Pursue Graduate Studies? - -Many students pursue graduate studies for the love of learning and discovery. For others, cultural -motivation and family traditions have an unquestionably positive influence on the decision to -enter graduate studies. Some students pursue graduate studies because their chosen profession -requires a graduate degree while others are looking to change career paths or better position -themselves for advancement opportunities. - -In research programs, graduate students are involved in both the development and the -responsible conduct of original, important research and scholarship. As such, graduate students -should be excited by carrying out in-depth and detailed studies in the spirit of creative and -imaginative inquiry. Graduate students and their work are an important part of an ongoing -research process that helps us to better comprehend the human and natural world in which we -exist. This research provides the human community with ways of understanding natural, cultural, -imaginative, social and technological phenomena and investigating problems through the pursuit -of knowledge. Graduate students are thus engaged not only in a social process that provides -society with new ways of looking at the world’s complexity, problems and beauty, but in a -personal quest for bettering their lives or expanding their learning and insight. - -Recently, people with graduate degrees have been referred to as “highly qualified personnel”. In -today’s knowledge economy this level of qualification is a valuable asset for an increasing -number of jobs. According to the 2007 National Graduates Survey (Class of 2005), 93% of -graduates of master programs were employed at the time of survey. As one might expect, -earnings increase with education levels, with the largest earning premiums between the bachelor -and master levels (Bayard & Greenlee, 2009). - -Skills Required Before Entering Graduate Programs - - High academic standing - Strong commitment to pursuing rigorous research training in a selected subject area - Enthusiasm and a high degree of interest in learning - Curiosity, and an open and enquiring mind - Sound work ethic, integrity and moral standards - Collegiality - Perseverance and patience - Maturity and reliability - -Representative Skills Acquired in Graduate Programs - - Excellent critical thinking skills - Ability to integrate data and information from multiple sources, and to develop and test - -hypotheses rigorously - - Excellent oral and written communication skills - -4 - -  Skills in a range of analytical techniques using sophisticated instrumentation - Ability to work with equipment and instruments at tasks requiring precision - Ability to coordinate or co-supervise the work of others - Ability to identify problems and to develop and implement innovative solutions - Ability to work independently and in teams +# **Decoupling the Benefits of Subword Tokenization for Language Model Training via Byte-level Simulation** -(Used with permission from the Schulich School of Medicine & Dentistry, The University of -Western Ontario [UWO], 2010, p. 3) +**Th´eo Gigant** Nous Research `theo@nousresearch.com` -What is Graduate Education? +**Jeffrey Quesnelle** Nous Research -Graduate school includes Masters Programs (typically 1-2 years of full time study following an -undergraduate degree) and Doctoral/Ph.D. programs (approximately 4-6 additional years of -study which may follow a Masters Degree). +**Bowen Peng** Nous Research -The graduate learning environment is distinctly different from that at the undergraduate level. It -has been characterized as advanced, focused and scholarly in nature: +``` +bloc@nousresearch.comemozilla@nousresearch.com +``` - - - +## **Abstract** -It is advanced because it builds upon an undergraduate education. -It is focused because the emphasis in graduate studies is on depth. -It is scholarly because it is concerned not simply with the acquisition of knowledge and -skills, but with the critical analysis of existing knowledge and the creation of new -knowledge. Graduate students are expected to acquire and apply advanced analytical and -interpretive skills, as well understanding and/or producing research. +Subword tokenization is an essential part of modern large language models (LLMs), yet its specific contributions to training efficiency and model performance remain poorly understood. In this work, we decouple the effects of subword tokenization by isolating them within a controlled byte-level pretraining pipeline. We formulate and test hypotheses across various dimensions, including sample throughput, vocabulary scaling, and the linguistic prior of subword boundaries. By simulating these effects in a byte-level setting, we refine our understanding of why subword models outperform raw byte models and offer insights to improve the pretraining of future byte-level and subword models. Specifically, our experiments highlight the critical role of increased training throughput and the integration of subword boundaries as either explicit priors or inductive biases. -Although the expectations of students enrolled in graduate programs may vary, it should be -recognized that graduate school represents a very different educational experience than that of -undergraduate studies. Scientific discovery is seldom a linear process, and it is generally not -possible to drive research directly towards a desired outcome. Discovery is a mixture of insight, -effort, curiosity, and good fortune (UWO, 2010, p. 5). +## **1 Introduction** -Graduate programs may be separated into three broad categories: +Tokenization is an essential step of the Natural Language Processing pipeline, segmenting text into atomic units to be processed by language models. Although state-of-the-art Large Language Models (LLMs) rely almost exclusively on subword algorithms like BPE or Unigram [31, 18], there is no consensus on which specific properties of subword models enable this performance advantage [11, 30]. - Professional programs – provide specialized skills and qualifications for a specific +Subword tokenization simultaneously dictates the allocation of compute to parts of the input sequence and the scaling of the model’s vocabulary parameters by balancing vocabulary size, sequence length, and information density per token through the granularity of the tokens, or _fertility_ of the tokenizer. Empirical evidence suggests that a larger vocabulary results on average in better downstream performances [33, 15] in part because it reduces the Kolmogorov complexity of tokenized sequences [4]. Subword tokens are also often viewed as a proxy for linguistic “information units” [2]. -profession (i.e., Speech-Language Pathology, MBA programs). +Despite their prevalence, recent literature has highlighted significant issues stemming from subword tokenizers, including “character-blindness” [5, 7], language-dependent performance disparities [29], inadequacies with prefix forms [20], tokenization ambiguity [18, 26], and weaknesses linked to under-trained tokens [19]. - Terminal programs – Masters and Doctoral degrees are achieved separately. Students +Character, or byte-level language models [6, 23] have been proposed as an alternative to subword language models, in part to address these issues. Sometimes wrongly described as _tokenizer-free_ , these models usually rely on characters as defined by the Unicode standard [35], or bytes resulting from the UTF-8 [36] encoding of text. While solving some of the aforementioned subword-related problems, these byte-level language models consistently struggle to match the training efficiency and downstream performance of their subword-based counterparts. This performance gap between bytelevel and subword models is typically attributed to some “benefits” of subword tokenization, which are typically analyzed in aggregate. To the best of our knowledge, there have been no successful attempts to isolate and quantify their decoupled contributions. For example, a larger vocabulary not -apply to Doctoral programs after first completing their Masters degree. +Preprint. - Non-terminal programs – Masters program feeds directly into Doctoral program +only increases embedding capacity, but also reduces sequence length, thereby increasing the effective sample throughput during training. Furthermore, subword boundaries may provide a structural prior that aligns with human semantics, aiding generalization in ways that raw bytes do not. -without option of stopping after Masters. May be entered into directly from -undergraduate studies. +In this paper, we suggest hypotheses as to what effects subword tokenization methods have on training dynamics, and we conduct a set of experiments to try to isolate and quantify them by artificially reproducing these effects for training byte-level language models. -Additionally, your degree may be: +## **2 Preliminaries** - Course-based – a combination of required courses, practicum placement, qualifying +## **2.1 Subword tokenization** -exam, independent research, or capstone course or project. +Byte-Pair Encoding (BPE) [31] is a bottom-up subword tokenization method based on the BPE grammar-based compression algorithm [10]. It is the _de facto_ standard tokenization method used with LLMs. It comes as the default tokenization method in the most popular LLM training frameworks [32, 21, 1, 8], due to highly optimized implementations[1] . Its dominance can also be attributed to the legacy of open-source LLMs that had a great impact on industry and academia, such as GPT2 [27], LLaMA [34] and Mistral [17]. -5 +A popular alternative is unigram tokenization [18], a top-down subword tokenization method based on a unigram language model, which creates tokens that align better with morphology [2] and allows subword regularization [18]. This method is more rarely encountered in practice, due to the more costly and difficult implementation. -  Thesis-based – a combination of required courses and a thesis project under the +## **2.2 Byte-level language models** -supervision of a thesis advisor. (UTM Career Centre [UTM], 2012, p. 1) +Contrary to LLMs using static subword tokenization, byte-level LLMs have a more fine-grained access to single bytes of the input. These models usually involve a method to compress or downsample the byte sequences to align the FLOPs-per-input-byte cost with subword models. These include, for instance, static downsampling with strided convolutions [6], or dynamic downsampling using lightweight local encoders [24, 16, 23]. -Thesis-based master’s programs provide an opportunity for students to explore the possibility of -pursuing research as a career. The normal time to completion of this degree is 2 years, although -completion of a high quality thesis may require a longer period of time. Students are expected to -display enthusiasm and dedication towards completing the objectives that define their research -projects, which can mean working on their projects beyond a standard work week. In research, -the end result is nearly always determined by the effort and commitment of the student to the -project. The reward for this is the opportunity to pursue research questions and lead discovery in -diverse and important areas that are directed at reducing human suffering and improving our -understanding of the world. +In contrast with these works, in this work we do not use downsampling in the architecture and process UTF-8-tokenized sequences with a standard architecture for subword-tokenized sequences, namely the LLaMA-3 architecture [14]. -Preparation for Graduate School +## **3 Related Works** -The process of deciding to enrol in a graduate program and the development of a plan of action -often begins well before the completion of a bachelor's degree. In disciplines where the -opportunity exits, students sometimes secure summer jobs in research environments to get a -sense of what research is like. They also learn through reading and conversation about the -research community and the career opportunities in their field of interest. +Previous works have studied the effects of subword tokenization for language model training. Gall´e [11] and Zouhar et al. [38] empirically showed that a BPE tokenizer with a higher compression ratio results in higher downstream performance on machine translation tasks. Chung and Kim [4] quantified the complexity of tokenized text via an estimate of the Kolmogorov complexity, showing that increasing the vocabulary size of a BPE tokenizer increases performance as a consequence of a reduction in the complexity of tokenized sequences. Schmidt et al. [30] developed a tokenization scheme that compresses sequences more than BPE, while resulting in worse downstream performance, challenging the idea that the effectiveness of BPE comes only from its compression effect. -Some students visit the appropriate departments in several universities and gain an insight into -the existing curriculum, programs, infra-structure and support system for graduate studies. -Sometimes they also assess through discussions with potential supervisors the type of research -project for which supervision and financial support are available and identify research topics in -which they are interested and for which they are qualified. +In this paper, we formulate and test hypotheses covering various aspects of subword tokenization, including computational efficiency, structural inductive biases and changes to the optimization objective. -Although standards and criteria vary across programs, one common factor is that the competition -is intense. Factors that contribute to admission decisions include: +## **4 Hypotheses** - GPA – most schools look for a minimum average of B+ during the last two years of +We formalize the potential drivers of the subword-byte performance gap into the following testable hypotheses, categorized by their effects on model training and representation. -undergraduate study +> 1Such as `https://github.com/huggingface/tokenizers` or `https://github.com/openai/tiktoken` - Suitability for the program – determined by the quality of your application, your +2 -research interests and your compatibility with the program/faculty research interests - Skills and experience – your research experience, volunteer/work experiences and +## **4.1 Computational and Scaling Efficiency** -extracurricular activities +The advantages most commonly attributed to tokenization relate to sequence compression. By reducing sequence lengths and expanding the vocabulary, tokenization fundamentally alters the structural dimensionality of the model’s input and the marginal computational cost per bit of processed data. - Reference letters – the type of referee and relevance of their comments are weighed +## **Hypothesis 1** -more heavily than the number of reference letters you submit +Increasing subword vocabulary size improves performance by scaling model capacity (via the embedding layer) at a marginal increase in FLOPs. - Admission test scores – if required (GRE www.gre.org , GMAT www.mba.com/mba) +Token embeddings are usually implemented as look-up tables, accessed in constant time. As noted by Tao et al. [33], large vocabularies improve model performance, and most of the computational overhead of adding vocabulary parameters is related to the output layer. -Prepare for graduate school during your undergraduate studies by: +## **Hypothesis 2** - Strengthening your GPA – Although many programs look primarily at your final two +At an isoFLOP budget, subword models process significantly more raw information (in bits) per gradient step than byte-level models, leading to superior sample throughput and better downstream performance. -years, a GPA that is consistently high will impress admission committees +## **4.2 Structural Inductive Biases** -6 +Subword tokenization injects “human-centric” structure into the sequence before the model ever sees it. We hypothesize that this acts as a powerful prior and could be leveraged as an inductive bias to improve training. -  Exploring your options – Take a variety of courses to help determine your research -interests, talk to Teaching Assistant about their experiences, take part in networking -activities, and talk to your professors about graduate programs and their areas of research +## **Hypothesis 3** - Identifying references – Graduate applications require between 2-3 references. Get to +Subword end boundaries provide a non-causal prior that simplifies the modeling task by leaking future bytes. -know your professors by taking more than one class with those whose work interests you, -participate in class discussions, attend office hours, and join clubs that allow professors to -get to know you outside of the classroom +Unlike UTF-8 tokenization, which is strictly causal, subword tokenizers require a “look-ahead” to determine optimal boundaries [23]. This effectively provides the model with a “hint” about the future byte distribution, creating an inherently easier prediction task. - Building experience – Gain research experience through part-time or summer jobs, -senior level thesis courses, faculty research projects, undergraduate research awards, -internships, or volunteering with a professor. Experience as a teaching assistant would be -an asset. (UTM, 2012, p. 2) +## **Hypothesis 4** -Making the Decision +Subword boundaries act as inductive biases that align the model with semantically meaningful units. -As prospective graduate students, you are important potential members of a university’s research -and learning community. The orientation and guidelines that this document provides are -designed to help you approach your future studies as a confident and informed person. Even -though you may feel that the process of admission to graduate studies is one in which you are -being evaluated by a university, you also have just as significant a role in evaluating your -prospective program of study. It is important that you make sure that you have the information -you need to make a good decision. +## **Hypothesis 5** -Making Contacts +Subword distances act as inductive biases that align the model’s attention mechanism and positional encodings with semantically meaningful units. -Making contact with your prospective institutions is important. This may mean: +In subword LLMs, positional encodings represent distances between subwords; in byte-level models, they usually represent character distances, which may lack direct semantic utility. - using websites - communicating by e-mail - -talking on the telephone - making a visit +## **4.3 Optimization Objective** -Do your best to talk to prospective supervisors or heads of graduate admissions committees as -well as a range of students including those presently enrolled in the program and those who have -already graduated. It is important to gauge your potential satisfaction with a program or a -supervisor from a student’s perspective. This type of conversation can help you know what to -expect from a supervisor or a course, or may just prove helpful in getting tips on potential living -arrangements. +Finally, we consider how the choice of tokenization shifts the nature of the prediction task itself. -Choosing an Institution +## **Hypothesis 6** -Students choose to pursue their studies at a particular institution for a variety of reasons. For -instance, some people choose to attend universities near to where they live. Others base their -decision on the reputation of a university. Keep in mind, though, that the research reputation of a -university does not always equate with student experiences in a program of study. Sometimes -graduate students’ experiences of the programs at a world-renowned research university may not - -7 - - mesh with the university’s high reputation. Students may rank their experiences at a lesser- -known university as being far more supportive and encouraging than the university’s research -reputation might suggest. +Minimizing cross-entropy per subword is a more effective proxy for language modeling than minimizing cross-entropy per byte. -In the case of a professional graduate degree, it is important to verify that the university program -is accredited by professional bodies. If this applies to your field of study, look into the -accreditation status of the institution and/or program you are planning to attend. - -Whether you are interested in a professional or a research-oriented degree, you should research -the nature of the university you are considering. Because it is ultimately a question of your fit -and prospective experience at a university, it is crucial to be as informed as possible. - -Choosing a Supervisor - -The choice of the right supervisor depends on many factors such as the track record of a -supervisor in seeing students through to a successful conclusion in a reasonable period of time. -Personality considerations also play a role. Some students require the caring attention of a -professor who may be a superb teacher but less active as a scholar, others flourish in the -competitive environment where the star of the research team is an internationally recognized -researcher. The relationship between student and supervisor should be based on complementary -objectives and qualities. Thus students should not become totally dependent on their supervisor. -They should try to maintain a collegial working relationship with all members of the department. -In turn, departments should remember that students require continuing departmental support for -their successful and timely graduation. Thinking about appropriate supervisors and then -approaching them will be one of the most important things you need to do when applying to a -graduate research program. - -The role of the supervisor will differ depending on your area of study. In certain disciplines, the -choice of the type of research project for which supervision and financial support are available -will depend quite directly on the research work the supervisor is doing. However, in the social -sciences and humanities, research topics often originate from the students themselves. In either -case, you will want to identify research topics in which potential supervisors are interested and -knowledgeable. - -Determining whether a supervisor is an active and productive scholar or researcher is an -important factor. Many professors maintain a website which contains their pedagogical and -publication history. This may be more detailed and useful than a database search because it will -reflect conference participation, courses taught and present research projects as well. Exploring -the professor’s website or the department site of the university with which your potential -supervisor is affiliated will often give you access to valuable information about the professor’s -research and interests. If you are approaching a potential supervisor, knowledge of their own -research is invaluable. Become familiar with this professor’s publications and be able to -knowledgeably discuss your mutual interests. Potential graduate students should ask the -following questions about the supervisors they are considering: - -8 - -  Do they regularly publish journal articles? In which journals? On what subject? - What books have they written? - Do they encourage publications by their students, either jointly or as single authored - -publications? - -Potential supervisors have track records not only as scholars, but also as supervisors. This is also -very important in determining whether they are someone you want to work with. Assessing the -combination of a potential supervisor's scholarship and supervisory skills will give you a good -idea of what your working relationship with them might be like. However, it is just as important -to try to get a sense of the person's character and attitude towards supervision in a more personal -way. Try to talk to them over e-mail or telephone, or ideally, in person. You will be working -closely with this person for a number of years, so something as basic as how well you get along -will be surprisingly important. Since the issue in supervision is good “fit,” an honest self- -assessment of what you are looking for and what works best for you is also in order. - -At the beginning of the study term, supervisors and students should discuss such things as: - -timelines for completion -forming a Supervisory Committee - - - - expected work hours, vacation times, research agendas, etc. - scheduling regular meetings between supervisor and student - attending departmental seminars, guest lecturers, etc. - attending academic conferences - policies on plagiarism - policies about authorship on papers and intellectual property - -The CAGS has developed a Letter of Understanding template that may assist you in these -discussions. - -Financing a Graduate Education - -Most graduate students require financial assistance during their studies. Student financial -assistance is available in various forms, including government financial aid (student loans and/or -bursaries), and privately funded bursary and scholarship programs, as well as student assistance -programs (tuition payments deferrals, emergency funding, food vouchers, etc.) and on-campus -work opportunities through work-study programs. Students may also wish to take on teaching or -research assistance positions. - -Federal agencies such as the Canadian Institutes for Health Research, the Natural Sciences and -Engineering Research Council and the Social Sciences and Humanities Research Council as well -as provincial agencies are among the primary sources of scholarships and fellowships for these -students. Scholarships and fellowships are always tenable for limited periods of time. NSERC, -for example, provides a maximum of four years of combined support at the master's and doctoral -levels. The maximum funding for a master's program is two years. - -9 - - Business and industry often offer scholarships for study in designated areas. In addition, most -graduate programs provide teaching assistantships from university operating funds and research -fellowships or stipends from faculty research grants and contracts. Most Canadian universities -maintain a database or records of funding sources and publish selective listings in their graduate -calendar. - -Many (if not all) Canadian universities have a financial aid office to assist students and -prospective students in seeking and securing financial assistance to enable them to pursue their -education. The Government of Canada also has a comprehensive website explaining the -eligibility and application process for government student loans, which students and prospective -students may find useful. - -At the time of admission into a graduate program, it is essential that students, supervisors, and -program administrators have a clear understanding of the financial support obtained, for what -purposes, over what period of time, and in return for what services. The need for financial -stability in graduate school cannot be overemphasized. Many academically able students have -failed because they underestimated the financial drain on their resources and because they had -depleted all available sources of support before their studies were complete. It is important that -students measure their personal financial needs realistically against these available funds. - -Funding Sources - -In addition to those listed here, many institutions offer internal awards. Although this list is not -exhaustive, here are some of the most common external granting agencies: - -Association of Universities and Colleges of Canada -– Higher Education Scholarships -Canadian Bureau for International Education - -Canadian Cancer Society - -Canadian Institutes of Health Research (CIHR) -Fondation Baxter & Alma Ricard -Fonds de la recherche en santé du Québec -Fonds québécois de la recherche sur la nature et les -technologies -Fonds québécois de la recherche sur la société et -culture -Heart and Stroke Foundation of Canada -IODE Canada War Memorial Doctoral Scholarship - -John and Lois Lamont Graduate Scholarship - -https://juno.aucc.ca/wes/?pg=834 - -www.cbie-bcei.ca/education- -management-services/alphabetical-list- -of-programs -www.cancer.ca/rese -arch/Grants%20and%20Awards/Ongoin -g%20funding.aspx?sc_lang=en -www.cihr-irsc.gc.ca -www.fondationricard.com -www.frsq.gouv.qc.ca -www.fqrnt.gouv.qc.ca - -www.frqsc.gouv.qc.ca - -www.hsf.ca/research/en/home -www.iode.ca/War-Memorial-Doctoral- -Scholarship.aspx -http://www.cfsh.ca/What-We- -Do/Scholarships_and_Awards/John-and- -Lois-Lamont-Graduate-Scholarship.aspx - -10 - - Mitacs - -www.mitacs.ca/funding-opportunities - -Natural Sciences and Engineering Research Council -of Canada -Ontario Graduate Scholarships (OGS) / Ontario -Graduate Scholarships in Science and Technology -(OGSST) -Social Sciences and Humanities Research Council -(SSHRC) -The International Development Research Centre - -Trudeau Scholarships - -Vanier Canada Graduate Scholarships (CIHR, -NSERC, SSHRC) - -www.nserc-crsng.gc.ca - -https://osap.gov.on.ca/OSAPPortal/en/A --ZListofAid/PRD1346626.html - -www.sshrc-crsh.gc.ca/Default.aspx - -www.crdi.ca/EN/Funding/Pages/default. -aspx -http://www.trudeaufounda -tion.ca/en/programs/scholarships/applica -tions -www.vanier.gc.ca/eng/home- -accueil.aspx - -Acquiring Formal Training and Certifications - -Moving Along - -Much of the formal training in research-oriented graduate programs is received during the first -year in the program. This training varies with the nature of the discipline and the department. In -many programs it is customary to provide graduate courses and to examine the students in these -courses, usually by written tests or papers. In other programs, the course work is augmented by -directed reading or individual study accompanied by periodic evaluation. However, it is -generally agreed that the first year should include an introduction to research and should not be -entirely spent on formal teaching. Students should begin to participate in research seminars and -attend conferences in their discipline as soon as possible. - -For research involving human participants, animals or radioactive, biological or hazardous -materials and lasers, certification normally must be obtained prior to starting the research. Often, -grant funds cannot be released until the relevant certification has been obtained. - -Choosing a Research Topic - -Once a student is integrated into a program and a faculty member has agreed to be a supervisor, -the next matter is the choice of the student's research topic. At the master's level, the topic should -be feasible in a predictable period of time and its scope should be reasonable. In departments -without doctoral programs the temptation sometimes exists to expand master's theses into mini- -doctoral theses, usually at a cost to the students which outweighs the associated benefit. At the -doctoral level, the choice of research topic often determines the career path of a student and -therefore requires thoughtful planning and expert advice. - -11 - - In some programs, a student may become involved in one of the ongoing lines of research within -the department. There is a growing tendency in Canadian universities, encouraged by the federal -and provincial research councils, to pursue collaborative research. Graduate students can be the -beneficiaries of this trend. The available research topics in such cases are partially predetermined -by the existing departmental research programs. The success rate of students in such programs is -reasonably high because the research is directly linked to the milestones of the programs. - -In other programs, the choice of a research topic may depend more on the advice of a specific -supervisor who may have a general idea of several research possibilities. In deciding on a -particular research topic, the supervisor should be confident that given hard work and reasonable -ability, it is fairly certain that the student will bring the project to a satisfactory conclusion in a -predictable period of time. On the other hand, an exciting line of research occasionally presents -itself in which the outcome is more uncertain. In these situations the student and the supervisor -should have a fallback position in case unexpected difficulties arise. - -Setting Milestones - -After having chosen a research topic, students should, with the help of their supervisor, make a -realistic plan for their research and thesis work which takes into account the departmental and -university rules, as well as their own constraints and outside responsibilities. This plan should be -endorsed by the supervisor and the departmental graduate studies committee. The CAGS has -developed a Letter of Understanding template that may assist you in these discussions. - -The plan should establish in writing appropriate milestones which the student should reach at -various times. It should be filed with the director of the program and form the basis for regular -assessments of progress involving the student, the supervisor, and the department. Students -should keep systematic records which enable them to take note of the pace of their progress and -of their research activities and related events in an objective, quantifiable way. - -Everything tends to take much longer than inexperienced students expect. With the help of their -supervisors, students need to plan their time carefully if they are to complete their graduate -programs within an agreed period of time. The start of a research project is nearly always slower -than students expect. Another common cause of delay is the fact that sometimes the students, or -the supervisors, are never satisfied with the accomplished research. They can always think of -ways of improving the obtained results. In short, they cannot bring a project to a conclusion. -Perfectionism can be a virtue, but students should write up what they have already achieved. -With a draft, they would almost certainly see more clearly whether any improvement was -actually necessary. - -Not infrequently it happens that the collected information or obtained results are inadequate to -support a thesis. This inadequacy is often not recognized until the thesis is actually written. -Students then have to gather additional information, perform further experiments or calculations, -or do more research, usually resulting in a delay of at least six to twelve months. If the -accumulation of information depends on seasonal or other unpredictable factors, the delay may -be even longer. Essentially this outcome is due to a lack of planning. The role of a department in - -12 - - monitoring research progress is particularly pertinent to minimizing the risks involved in this -kind of research. - -For doctoral students, there exists a pivotal milestone which determines whether they should -continue in a program in the form of their admission to candidacy. This process usually consists -of written and oral comprehensive examinations evaluated jointly by several faculty members in -addition to the supervisor. The precise nature of the examinations differs considerably from -university to university and from discipline to discipline. In some cases the examinations are of a -diagnostic nature and serve indeed as the mechanism used to decide whether a student continues -in a program. In other cases the examinations are summative and serve to flag gaps in a student's -knowledge which need to be filled. In both cases, however, the examinations can be a traumatic -experience and students often over-prepare for them. Firm departmental and faculty rules and -regulations should be in place to clarify the process. - -In a number of disciplines, several universities have introduced the fast track option that allows a -student to proceed to a doctoral program from a master's program without completing the -master's degree requirements. In addition, some universities also permit talented students in -certain disciplines to enter doctoral programs directly from bachelor's programs. In such cases, -master's programs sometimes represent a fallback option for students who cannot complete their -doctoral programs. It is then up to the graduate faculty of a university to specify the exact -conditions under which a doctoral candidate can enter a master's program in the same field, and -what requirements must be met in order to obtain a master's degree in place of a doctorate. - -Analysis, Synthesis, and Writing - -Finishing Up - -Often students define the thesis process in terms of researching, then writing. They are proud of -the fact that they have achieved their research objectives and that all that remains is to write up -the thesis. Several years later they have still not finished. The problem is that research is -generally fun and relatively easy once the project has been defined. In contrast, writing is always -demanding, and it is especially demanding when it has not been done for the entire period of -research. For this reason, students should be encouraged to combine researching and writing at -all times. It is particularly difficult to combine the writing up of a thesis with full-time -employment and students should be encouraged and given the means to complete their thesis -before leaving the university. - -Two particular points must be mentioned here. First, the thesis should be no longer than -necessary. It should demonstrate that the student understands the background to the research, -explain clearly the methods used, present the results, and discuss the findings within an -appropriate framework. Verbosity and padding should be avoided at all stages; they detract from -the issues at hand and provide easy targets for criticism by examiners. - -Secondly, unless the student is a particularly gifted or experienced writer, the use of one of the -many books that are available on good writing will likely improve the thesis. There is more in - -13 - - this than making the thesis a pleasure for the examiners to read. Unclear writing is frequently a -reflection of unclear thinking, whereas students who write clearly soon discover that a problem -of expression often arises from a lack of understanding. - -Graduate faculties and departments usually have specific guidelines and regulations governing -the preparation of theses. Before starting to write a thesis, students should therefore familiarize -themselves with the formal requirements which a thesis must meet. Most universities will accept -a thesis that is a series of articles written in the publication format for that discipline. - -One of the basic requirements of a doctoral thesis is that it be original. At the doctoral level, a -student's contribution must always be in the form of a substantial advancement of knowledge. -The student should have been trained as a scholar and thinker who is able to place this new -knowledge convincingly and eloquently in a broader conceptual context. The student should also -feel confident to guide the research of other students in the same field. - -Presenting and Publishing - -All research students, both master's and doctoral, should be encouraged by their supervisors to -make presentations, participate in conferences and submit manuscripts to journals, often starting -with research notes. The many good results of such activities include greater self-confidence, a -better perspective on the quality and meaning of their research, and more visibility in the job -market. They also help students to acquire a variety of skills which are necessary for a career in -research or in research management. Students sometimes do not realize that colleagues may be -more interested in the wider aspects of the project and its impact on the whole body of -knowledge than in their raw results. For these reasons, students should be given the opportunity, -particularly toward the end of the period of research training and while writing up their results, to -present the results of their research in a public forum, both for criticism and enrichment. - -A Note about Intellectual Property Rights - -Students must receive proper recognition by their supervisors for contributions made in -publications before and after the defence of a thesis. This is particularly crucial where a student's -research is part of a larger team project, and where related results emanating from the larger -project have already been published, either jointly by the students and their supervisors or by -other members of the team. There should be a very clear departmental policy which regulates the -respective rights of students, supervisor, and research collaborators to publish results of joint -research. - -Another issue which sometimes arises is whether a student actually has the intellectual property -right to previously published material. It may happen that the copyright to research articles -belongs to a scientific journal or publishing house and students must obtain written permission if -they wish to include reprints in a thesis and subsequently copyright the thesis. In situations where -a thesis consists entirely of previously published material or is a work of art, the regulations and - -14 - - procedures for the examination of such theses should be established in writing at the time of -admission of the student to the program. - -In the case of research of immediate or even prospective commercial value, or where the -dissemination of knowledge is being delayed because of the proprietary nature of the results, the -confidential nature of the thesis defence raises additional ethical questions which the university -should anticipate. In these and similar situations, the university regulations should also spell out -unequivocally the extent of a student's right to restrict publication of research results, while still -meeting the degree requirements. - -Questions about authorship and credit for research to be carried out should be anticipated and -settled early in the program within the context of departmental guidelines and policies with your -supervisor. In the case of collaborative and team research, it is often difficult to isolate the -original contributions of individuals from the project as a whole. In such cases it is particularly -important to be able to refer to the specifically agreed and recorded objectives which each team -member was required to attain, and to have a record of seminar and conference presentations -which document the contribution of individual researchers. +3 -Additional Living Considerations +## **Hypothesis 7** -Living Expenses +Subword prediction functions as a form of variable-length multi-token prediction, which has been shown to improve representation learning compared to single-token prediction. -From food to accommodation, clothing and transportation, living costs will vary by location. For -instance, in Toronto, accommodation can cost between $7,300 and $14,000 per year -(http://cie.utoronto.ca/Coming/Living-Here/Cost-of-Living.htm#cost). +Predicting a single subword is equivalent to predicting a byte _n_ -gram at once. This aligns with recent findings that multi-token prediction heads can improve downstream performance [13]. -Is a vehicle required? Is public transportation sufficient? These factors will also impact the -student’s budget. For a graduate student with a family, there are other expenses considerations. -The Graduate Student Association at the University of Waterloo has prepared a sample budget -for a typical graduate student with a family. +## **5 Methodology** -Housing +We propose experiments intended to replicate one by one the effects induced by subword tokenization linked to the hypotheses we suggested. These effects are added to a 1.7B parameters byte-level language model pretraining pipeline, which will be compared to a baseline byte-level language model. -Finding a place to live can be both exciting and challenging for new students. Many universities -have housing geared to upper-year and graduate level students and this may be the first -consideration for students new to a city. In addition to being on-campus, residence offers other -social and academic supports to residents. University-owned apartments are also often close to -campus and can offer additional benefits to students. +In the following experiments, most hyperparameters remain unchanged. All changes made on the input and output, or on the architecture of the model, are designed to introduce negligible computational overhead. We are using a standard LLaMA-3 architecture [14] trained with the TorchTitan framework [21]. Models are trained on the fineweb-edu dataset [25] tokenized into UTF-8 bytes. Sequences are also tokenized with the LLaMA-3 BPE tokenizer to provide byte-level subword boundaries. All comparisons between models are done using the same bits-per-byte cross-entropy loss, computed on a separate validation subset of fineweb-edu. Hyperparameters are detailed in the Appendix A. -Some students choose to live off-campus. A visit to the off-campus housing office can assist -students in finding suitable housing and understanding lease agreements. Students will need to -do some legwork to determine which area to live in and whether proximity to campus is a factor -in their decision making. +## **5.1 Scaling vocabulary parameters** -Parental leave +To test Hypothesis 1, we introduce multi-head _n_ -gram embedding tables to simulate the larger input vocabulary of a subword LLM. This method is similar to recent _n_ -gram embedding methods [15, 22, 3], but we introduce them only in the input layer. Our implementation is derived from the _engram_ demo implementation[2] . -15 +Hyperparameters are chosen to introduce around 70M additional parameters to the byte-level LLM, matching the embedding table of a subword LLM using the same architecture with a vocabulary of 35k tokens. - A graduate student who is bearing a child or who has primary responsibility for the care of an -infant or young child immediately following a birth or adoption is eligible for parental leave. -Students who are pregnant or considering having a child or taking parental leave should review -the specific guidelines related to requests for maternity/parental leaves at his or her particular -institution. +**==> picture [278 x 195] intentionally omitted <==** -Childcare +**----- Start of picture text -----**
+8 . 0
7 . 5
7 . 0
6 . 5
Model A - baseline
6 . 0 Model B - scaled vocabulary parameters
Training Steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Validation Loss (bits-per-byte)
**----- End of picture text -----**
-There are many childcare options and they depend on what type of care your child needs. It is -important to ask yourself questions such as: - When will you need child care? - Do you want your child in a home-based program or child care centre? - Should you choose a program near your home or school? +Figure 1: Validation loss when scaling input embedding parameters -Child care is available at many institutions and each will have different care options ranging -from full-time to drop-in options. The Province of Ontario provides a good guide to choosing -the right child care for your child(ren): http://www.edu.gov.on.ca/childcare/finding.html +Figure 1 highlights the small increase in performance associated with scaling input embedding parameters. While these results suggest that Hypothesis 1 does not explain the significant performance -Most regions have childcare subsidy programs available, and it may be worthwhile to investigate -whether your income and child care needs meet the criteria for the program. It is important to -keep in mind that there may be wait lists for funding and childcare spaces may not be -immediately available. +> 2 `https://raw.githubusercontent.com/deepseek-ai/Engram/refs/heads/main/engram_demo_v1.py` -Some institutions offer a child daycare subsidy for graduate students in order to help offset the -cost of childcare for graduate students. For instance, the University of Alberta offers one such -subsidy, to a maximum of $500 per child per fiscal year (to a total maximum of $2,000 in each -fiscal year). +4 -Questionnaires +gap between subword and byte-level language models, scaling vocabulary-like parameters remains a promising direction to improve language models, as exemplified in recent literature [28, 15, 3, 22]. -The purpose of this booklet was to identify key aspects of the dynamic relationship between -graduate students, supervisors and departments. The focus was on the presentation of a suitable -framework in which students and professors could interact with a good measure of predictable -success. The following questions are intended to highlight the main points of the discussion. +## **5.2 Artificially increasing the training sample throughput** -For Students +Subword tokenization results on average in around 4 times fewer tokens compared to UTF-8 tokenization[3] . At isoFLOPs, the sample throughput during training is 4 times higher using the same architecture. To simulate this behavior, we compress the sequences by a factor of 4 to train a bytelevel LLM at the same isoFLOPs sample throughput as a subword LLM. - Have you assessed your financial needs during your studies and do you have a financial - contingency plan? - Do you have a written commitment from your department which specifies what financial -support you will receive during your studies, for what purposes, over what period of time, -and in return for what services? +Given a sequence of length 4 _· L_ , we segment it into contiguous chunks of 4 bytes, resulting in a sequence of shape ( _L,_ 4). In the input layer, the model sums the embeddings of the 4 contiguous bytes in each chunk. In all hidden layers, the behavior is unchanged, and the model is effectively processing a sequence of _L_ latent tokens, containing information from 4 _· L_ input tokens. The model output has a shape ( _L, V_ ) with _V_ the size of the vocabulary. The loss is computed as the cross-entropy between this prediction and the first byte of the next chunk, _i.e._ next-byte prediction. -16 +After 50k steps using this method to artificially increase sample throughput by 4 times, we continue pretraining this model with the baseline regime, using sequences of length _L_ . During the first 50k steps, the baseline model _A_ sees on average 4 times less samples compared to model _B_ , but the same number of tokens, effectively simulating the larger sample throughput of subword language models. After 50k steps, both models are trained under the same conditions. -  Have you made a plan for your research which includes achievable milestones and a +**==> picture [278 x 195] intentionally omitted <==** -timetable for achieving these milestones? +**----- Start of picture text -----**
+100
Return to normal regime
8 . 0
7 . 5
7 . 0
6 . 5
Model A - baseline
6 . 0 Model B - increased sample throughput
Training Steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Validation Loss (bits-per-byte)
**----- End of picture text -----**
- Have you made a plan for your research which includes achievable milestones and a -timetable for achieving these milestones? +Figure 2: Validation loss when scaling sample throughput by 4 times for 50k steps - Have you discussed your research plan with your supervisor and agreed on the milestones +Figure 2 illustrates a significant gain resulting from the increase in sample throughput, even if performed for only 50k steps. Rapidly after falling back to the normal regime, model _B_ crosses the performance of the baseline model _A_ , and soon stabilizes at the same slope. This experiment strongly supports Hypothesis 2. -and a timetable? +## **5.3 Giving subword boundaries as a prior** - Are you keeping systematic records of your work and results, and are you able to answer +Subword tokenization segments the input text into contiguous chunks based on frequencies of _n_ - grams in the training corpus. This process requires access to the full sequence and thus leaks future information into past tokens [23]. A subword LLM is optimized for next-token prediction given a correctly segmented input. On the other hand, byte-level LLMs are usually strictly causal. We posit that having access to the subword segmentation boundaries makes the prediction task easier. For example, by design of pre-tokenization, whitespace characters are always following an end-of- -questions on the milestones achieved? +> 3We measure an average of 4 _._ 75 bytes-per-token on 50 _,_ 000 samples from fineweb-edu tokenized with the LLaMA-3 tokenizer. - What progress have you made towards your degree in the last few months? - Have you written a draft of any portion of your thesis for which you have completed the -research, and are there tables, figures or other matters which you could prepare at this -time? +5 - Have you discussed your research results and draft with others, and do they find your +subword boundary. On the other hand, the start-of-subword boundaries do not leak future bytes information, but can provide the model with structural prior. -written work clear and articulate? +In the following experiment, models _B_ and _C_ have access to a binary sequence of start-of-subword and end-of-subword boundaries, respectively, whose embeddings are added to the input byte embeddings. - What is your timetable for completing the remaining phases of your research and for +**==> picture [360 x 204] intentionally omitted <==** -writing up your thesis? +**----- Start of picture text -----**
+8 . 0 100
7 . 5
8 . 0
7 . 0 7 . 5 Return to normal regime
7 . 0
6 . 5
6 . 5
6 . 0
Model A - baseline 6 . 0 Model A - baseline
Model B - subword start prior Model B - subword start inductive bias
Model C - subword end prior Model C - subword end inductive bias
5 . 5 5 . 5
Training Steps Training Steps
(a) Subword boundaries as a prior (b) Subword boundaries as an inductive bias for
50k steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000100000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000100000
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
**----- End of picture text -----**
- What are the remaining requirements for your degree and when do you expect to -graduate? +Figure 3: Validation loss when providing the start or end of subword boundaries -For Supervisors +Figure 3a shows the significant performance boost resulting from the access to the subword segmentation boundaries, supporting Hypothesis 3. Specifically, end-of-subword boundaries offer a larger advantage compared to start-of-subword boundaries, as they leak future information. Startof-subword boundaries also improve the performance of the model, suggesting that the statistical prior they provide is a useful inductive bias for the model. In order to test that hypothesis, we train the models with access to the subword boundaries only at training-time, and remove the boundary information at validation-time. After 50k steps, we also remove the access to the subword boundaries for training and resume pretraining following the baseline regime. - Are your students' research topics well-defined, realistic and manageable within the +While subword end boundaries are more useful as a prior than subword start boundaries ( _c.f._ Figure 3a), they do not provide a useful inductive bias in this setting as evidenced by Figure 3b, probably because the model is relying too much on this prior. On the other hand, subword start boundaries do not leak future information, and provide a prior that improves the model performance in this setting. These observations support Hypotheses 3 and 4. -agreed timeframes? +## **5.4 Giving subword distances as a prior** - Have you discussed different management strategies and methodological approaches with +Similarly, Gelberg et al. [12] showed that RoPE positional encoding acts as a prior that can be removed later during training. In subword LLMs, the positional encoding is using subword distances, when byte-level LLMs use byte distances. To simulate the position prior of the subword positions in the latter, we replace the byte position encoding with subword position encoding in model _B_ . Subsequent bytes that are part of the same subword use the same repeated position. This setting does not leak future byte information, as it is effectively using the subword start boundaries information. -your students after having decided on their research topic? +We perform another experiment in which this prior is given only during training and removed after 50k steps, returning to the baseline training regime afterwards. - Have you set up a schedule of regular meetings with your students in order to monitor +Figures 4a and 4b suggest that subword distances can be a useful prior, but do not constitute a strong inductive bias in this setting. Considering the previous section, we conclude that subword boundaries constitute stronger prior, and inductive biases, than subword distances, highlighting the lesser relative significance of Hypothesis 5 compared to the previous Hypotheses. -their progress in relation to their research plan? +6 - Do you have a mechanism for refining your students' research topics in the early stages of +**==> picture [360 x 205] intentionally omitted <==** -their program? +**----- Start of picture text -----**
+8 . 0 100
7 . 5
8 . 0
7 . 0 7 . 5 Return to normal regime
7 . 0
6 . 5
6 . 5
6 . 0 Model A - baseline Model A - baseline
Model B - subword distances prior 6 . 0 Model B - subword distances inductive bias
Training Steps Training Steps
(a) Subword distances as a prior (b) Subword distances as an inductive bias for 50k
steps for training only
0 10000 20000 30000 40000 50000 60000 70000 80000 90000100000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000100000
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
**----- End of picture text -----**
- Have you provided your students with guidelines on the appropriate length of a thesis in -your field? +Figure 4: Validation loss when using subword distances in the positional embedding - Have you provided your students with guidelines on writing up and presenting their +## **5.5 Optimizing cross-entropy per subword** -research? +The cross-entropy loss for predicting a sequence of subwords ( _tm_ ) _m≤M_ with a model _θ_ is defined as - Do your students keep detailed records of their work and do you check these records +**==> picture [246 x 27] intentionally omitted <==** -periodically to see whether they are systematic and adequate? +With the same sequence, but tokenized as UTF-8 bytes ( _bn_ ) _n≤N_ , the default cross-entropy becomes - Are you familiar with your students' milestones for the next few months? - Have you provided your students with the opportunity to give seminar and conference - presentations on their research? - Have you discussed with your students the question of intellectual property rights and +**==> picture [228 x 28] intentionally omitted <==** -credit for collaborative research? +However, by decomposing a subword _t_ into the _k_ bytes it contains ( _bi_ ) _i≤k_ , we have _Pθ_ ( _t_ ) = Π _i≤k_ ( _Pθ_ ( _bi|_ ( _bj_ ) _j picture [320 x 28] intentionally omitted <==** -  Do you provide your students with written statements specifying what scholarships, +Instead of optimizing for the best cross-entropy per subword, the baseline target for byte-level LLM optimizes for cross-entropy per byte, scaling the loss by a dynamic factor _[M] N[<]_[ 1][.][In order to see if] this difference has any consequence on training, we use the cross-entropy per subword as a target to train a byte-level LLM and compare to the baseline. -fellowships, teaching assistantships or research assistantships they will receive during -their studies? +Figure 5 shows very little improvement compared to baseline, suggesting that Hypothesis 6 has minimal effects at this scale. - Do you have a document describing the responsibilities of a thesis supervisor? - How do you ensure a good match between supervisors and prospective students? - Do your students have a research plan, deposited with the department, listing the +## **5.6 Optimizing next subword prediction** -objectives of their research program and identifying a critical path for achieving them? +Instead of predicting one byte at a time, a subword LLM predicts a subword, usually containing multiple bytes. Arguably, this is analogous to multi-token prediction [13], which was shown to improve pretraining of LLMs, especially for models with more than 1 billion parameters. - Do your students have to present reports at regular intervals which are assessed by +We train byte-level model _B_ using a subword output vocabulary, optimizing for the cross-entropy computed using the next subwords, predicted from the end-of-subword bytes. After 50k steps, we return to the baseline pretraining regime. -members of the department other than the supervisor? +Figure 6 illustrates that the next subword prediction task is a worse objective to train a language model at this scale compared to next byte prediction, rejecting Hypothesis 7. - Do your students, supervisors, and the graduate studies committee meet on a regular basis +7 -to assess the students' difficulties and progress? +**==> picture [278 x 196] intentionally omitted <==** - How do you ensure that a student's timetable for writing a thesis and completing the +**----- Start of picture text -----**
+8 . 0
7 . 5
7 . 0
6 . 5
6 . 0 Model A - baseline
Model B - cross-entropy per subword
Training Steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Validation Loss (bits-per-byte)
**----- End of picture text -----**
-degree requirements is realistic? - Do your students have to give seminar and conference presentations on their work? - Do you have guidelines on the respective roles of co-supervisors of a thesis? - Do you have guidelines on collaborative research? +Figure 5: Validation loss when optimizing for cross-entropy per subword -18 +**==> picture [278 x 195] intentionally omitted <==** - References +**----- Start of picture text -----**
+100
Return to normal regime
8 . 0
7 . 5
7 . 0
6 . 5
Model A - baseline
6 . 0 Model B - next subword prediction
Training Steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Validation Loss (bits-per-byte)
**----- End of picture text -----**
-Bayard, J., & Greenlee, E. (2009). Graduating in Canada: Profile, labour market outcomes and -student debt of the class of 2005 (81-595-MIE2009074). Retrieved from Minister of -Industry website: http://www.statcan.gc.ca/pub/81-595-m/81-595-m2009074-eng.pdf +Figure 6: Validation loss when optimizing for next subword prediction for 50k steps -Schulich School of Medicine & Dentistry, The University of Western Ontario. (2010, April). -Department of Physiology and Pharmacology Graduate Student Handbook. Retrieved -May 22, 2012 from: -http://www.uwo.ca/physpharm/forms/Graduate%20Studies%20Handbook%20Jan%2020 -11.pdf +## **5.7 Summary** -UTM Career Centre. (2012, September). Considering Grad School (Tip Sheet). Retrieved from +The experiments we conducted suggest that the superior performance of subword language models compared to byte-level language models involve multiple effects at different magnitudes. Specifically, the effects related to Hypotheses 2, 3 and 4 are the most noticeable at this scale. By replicating these effects in isolation, we observe a significant improvement for pretraining byte-level language models. Interestingly, these hypotheses are related to different aspects of subword tokenization. The increased sample throughput (Hypothesis 2) is a direct consequence of the compression capabilities of subword tokenization. This is usually the biggest drawback that hinders the competitiveness of byte-level language models, such that state-of-the-art byte-level language models come with methods to compress the byte sequences and thus increase sample throughput closer to the subword counterpart [6, 24, 16, 23]. On the other hand, prior knowledge of subword boundaries (Hypotheses 3 and 4) has strong connections to subwords being good approximations of English semantic units. As exemplified by Schmidt et al. [30], the compression aspect cannot completely explain the efficiency of subword LLMs. Subwords created with Unigram, and BPE to a lesser extent, align well with morphological reference segmentations [2], explaining why we empirically observed that they provide a useful inductive bias during byte-level language model pretraining. -UTM Career Centre website: -http://www.utm.utoronto.ca/careers/pdf/consideringgradschool.pdf +8 -University of Waterloo (2011). A Guide for Graduate Research and Supervision at the -University of Waterloo. Retrieved from: https://uwaterloo.ca/graduate- -studies/sites/ca.graduate- -studies/files/uploads/files/A%20Guide%20for%20Graduate%20Research%20and%20Su -pervision%20at%20the%20University%20of%20Waterloo%202011_2.pdf +Our tests to replicate the effects linked to Hypotheses 1, 5, 6 and 7 either perform worse or do not show a significant change compared to the baseline, suggesting that these effects are not perceptible at this scale. However, their significance could be different at different scales. For instance, we observed a larger gap for experiments linked with Hypothesis 1 for smaller models (68M parameters experiments are included in Appendix B). -19 +## **5.8 Conclusion** + +In this paper, we proposed hypotheses regarding the effects that subword tokenization is having on language modeling. Through experiments simulating these effects in a pipeline for pretraining bytelevel language models, we try to isolate these effects and quantify the improvement they provide. In particular, we highlight the importance of increasing the training sample throughput, and giving the subword boundaries as a prior or as inductive biases. + +We believe that a better understanding of these effects will prove useful to both improve subword tokenization and byte-level language model pretraining. For example, Minixhofer et al. [23] recently proposed a method to continue the pretraining of a subword LLM as a byte-level LLM, effectively taking advantage of the beneficial effects of subword tokenization during the first stage of bytelevel language model pretraining. Zheng et al. [37] trained LLMs with inputs mixing raw unicode with sequences compressed using subword tokenization, neural compression or gzip, showing better isoFLOPs byte-level performance at scales exceeding 4B parameters, compared with baseline byte-level pretraining. A better understanding of these effects in isolation could allow researchers to improve on some of them, for instance by using different tokenization schemes for different purposes, or even scale some of these effects, similarly to the recent works studying new scaling directions for vocabulary-like parameters decoupled from the model’s vocabulary [15, 22, 3]. + +## **5.9 Limitations and Future Work** + +While our controlled simulations provide valuable insights into the decoupled benefits of subword tokenization, this study has several limitations that present opportunities for future research. + +To maintain computational feasibility while exploring a wide range of hypotheses, several of our key experimental interventions, such as artificially increasing sample throughput, injecting subword boundary priors, enforcing subword distance priors and optimizing for the next subword prediction objective, were introduced for only the first 50k training steps before reverting to the baseline bytelevel training regime. While this setting was sufficient to observe significant shifts in validation loss and training dynamics in some settings, the behavior of these priors could be different at different model scales and intervention duration. It remains an open question whether the performance gains, or the lack of them, observed in models exposed to these training interventions compound, plateau, or diminish when maintained throughout a complete, full-scale pretraining run. + +A core methodological choice in this work was to replicate the effects induced by subword tokenization one by one. By artificially isolating these variables, we successfully quantified their individual contributions to the subword-byte performance gap. However, this decoupled approach does not account for the complex interplay between these mechanisms. For instance, it is highly likely that the benefits of an increased training sample throughput and the structural inductive biases of subword boundaries interact during standard subword language model training. Future work should investigate these compounding effects to determine whether these isolated variables act additively, synergistically, or redundantly when combined into a single byte-level architecture. + +Our experiments were conducted on a 1 _._ 7B parameter language model trained exclusively on the English-centric fineweb-edu dataset tokenized into UTF-8 bytes. As noted in our discussion, the significance of certain hypotheses may shift at larger, or smaller, parameter scales. Furthermore, because English subwords naturally align well with morphological segmentations, the strength of the inductive biases provided by subword boundaries might differ substantially when modeling languages with different morphological structures [9]. Expanding this byte-level simulation framework to highly multilingual datasets and larger model scales remains a promising direction for future research. + +Finally, our work revolved around the perspective of reducing the gap between subword models and byte models, however, some of the insights could be leveraged to improve subword models further. + +9 + +## **References** + +- [1] Axolotl maintainers and contributors. _Axolotl: Open Source LLM Post-Training_ . 2023. URL: `https://github.com/axolotl-ai-cloud/axolotl` . + +- [2] Kaj Bostrom and Greg Durrett. “Byte Pair Encoding is Suboptimal for Language Model Pretraining”. In: _Findings of the Association for Computational Linguistics: EMNLP 2020_ . Findings 2020. Ed. by Trevor Cohn, Yulan He, and Yang Liu. Online: Association for Computational Linguistics, Nov. 2020, pp. 4617–4624. DOI: `10.18653/v1/2020.findingsemnlp.414` . URL: `https://aclanthology.org/2020.findings-emnlp.414/` (visited on 01/22/2026). + +- [3] Xin Cheng et al. _Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models_ . Jan. 12, 2026. DOI: `10 . 48550 / arXiv. 2601 . 07372` . arXiv: `2601 . 07372[cs]` . URL: `http://arxiv.org/abs/2601.07372` (visited on 01/22/2026). + +- [4] Woojin Chung and Jeonghoon Kim. _Exploiting Vocabulary Frequency Imbalance in Language Model Pre-training_ . Nov. 28, 2025. DOI: `10.48550/arXiv.2508.15390` . arXiv: `2508.15390[cs]` . URL: `http://arxiv.org/abs/2508.15390` (visited on 01/22/2026). + +- [5] Cristiano Ciaccio, Marta Sartor, Alessio Miaschi, and Felice Dell’Orletta. “Beyond the Spelling Miracle: Investigating Substring Awareness in Character-Blind Language Models”. In: _Findings of the Association for Computational Linguistics: ACL 2025_ . Findings 2025. Ed. by Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pilehvar. Vienna, Austria: Association for Computational Linguistics, July 2025, pp. 11361–11372. ISBN: 979-8-89176-256-5. DOI: `10.18653/v1/2025.findings-acl.593` . URL: `https: //aclanthology.org/2025.findings-acl.593/` (visited on 01/22/2026). + +- [6] Jonathan H. Clark, Dan Garrette, Iulia Turc, and John Wieting. “Canine: Pre-training an Efficient Tokenization-Free Encoder for Language Representation”. In: _Transactions of the Association for Computational Linguistics_ 10 (Jan. 31, 2022), pp. 73–91. ISSN: 2307-387X. DOI: `10.1162/tacl_a_00448` . URL: `https://doi.org/10.1162/tacl_a_00448` (visited on 01/22/2026). + +- [7] Adrian Cosma, Stefan Ruseti, Emilian Radoi, and Mihai Dascalu. _The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language Models_ . Sept. 15, 2025. DOI: `10.48550/arXiv.2505.14172` . arXiv: `2505.14172[cs]` . URL: `http://arxiv. org/abs/2505.14172` (visited on 01/22/2026). + +- [8] Michael Han Daniel Han and Unsloth team. _Unsloth_ . 2023. URL: `https://github.com/ unslothai/unsloth` . + +- [9] D´avid Drˇz´ık and Jozef Kapusta. “The importance of morphology-aware subword tokenization for NLP tasks in Slovak language modeling”. In: _Expert Systems with Applications_ 312 (May 25, 2026), p. 131492. ISSN: 0957-4174. DOI: `10.1016/j.eswa.2026.131492` . URL: `https://www.sciencedirect.com/science/article/pii/S0957417426004057` (visited on 04/02/2026). + +- [10] Philip Gage. “A new algorithm for data compression”. In: _C Users J._ 12.2 (Feb. 1, 1994), pp. 23–38. ISSN: 0898-9788. + +- [11] Matthias Gall´e. “Investigating the Effectiveness of BPE: The Power of Shorter Sequences”. In: _Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLPIJCNLP)_ . EMNLP-IJCNLP 2019. Ed. by Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan. Hong Kong, China: Association for Computational Linguistics, Nov. 2019, pp. 1375– 1381. DOI: `10.18653/v1/D19-1141` . URL: `https://aclanthology.org/D19-1141/` (visited on 01/22/2026). + +- [12] Yoav Gelberg, Koshi Eguchi, Takuya Akiba, and Edoardo Cetin. _Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings_ . Dec. 13, 2025. DOI: `10.48550/ arXiv.2512.12167` . arXiv: `2512.12167[cs]` . URL: `http://arxiv.org/abs/2512. 12167` (visited on 01/29/2026). + +- [13] Fabian Gloeckle, Badr Youbi Idrissi, Baptiste Rozi`ere, David Lopez-Paz, and Gabriel Synnaeve. “Better & faster large language models via multi-token prediction”. In: _Proceedings of the 41st International Conference on Machine Learning_ . Vol. 235. ICML’24. Vienna, Austria: JMLR.org, July 21, 2024, pp. 15706–15734. (Visited on 01/22/2026). + +10 + +- [14] Aaron Grattafiori et al. _The Llama 3 Herd of Models_ . Nov. 23, 2024. DOI: `10.48550/arXiv. 2407.21783` . arXiv: `2407.21783[cs]` . URL: `http://arxiv.org/abs/2407.21783` (visited on 04/01/2026). + +- [15] Hongzhi Huang, Defa Zhu, Banggu Wu, Yutao Zeng, Ya Wang, Qiyang Min, and Zhou Xun. “Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling”. In: Forty-second International Conference on Machine Learning. June 18, 2025. URL: `https://openreview. net/forum?id=gbeZKej40m` (visited on 01/22/2026). + +- [16] Sukjun Hwang, Brandon Wang, and Albert Gu. _Dynamic Chunking for End-to-End Hierarchical Sequence Modeling_ . July 15, 2025. DOI: `10.48550/arXiv.2507.07955` . arXiv: `2507.07955[cs]` . URL: `http://arxiv.org/abs/2507.07955` (visited on 02/03/2026). + +- [17] Albert Q. Jiang et al. _Mistral 7B_ . Oct. 10, 2023. DOI: `10.48550/arXiv.2310.06825` . arXiv: `2310.06825[cs]` . URL: `http://arxiv.org/abs/2310.06825` (visited on 04/01/2026). + +- [18] Taku Kudo. “Subword Regularization: Improving Neural Network Translation Models with Multiple Subword Candidates”. In: _Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . ACL 2018. Ed. by Iryna Gurevych and Yusuke Miyao. Melbourne, Australia: Association for Computational Linguistics, July 2018, pp. 66–75. DOI: `10.18653/v1/P18-1007` . URL: `https://aclanthology. org/P18-1007/` (visited on 01/22/2026). + +- [19] Sander Land and Max Bartolo. “Fishing for Magikarp: Automatically Detecting Undertrained Tokens in Large Language Models”. In: _Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing_ . EMNLP 2024. Ed. by Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen. Miami, Florida, USA: Association for Computational Linguistics, Nov. 2024, pp. 11631–11646. DOI: `10.18653/v1/2024.emnlp-main.649` . URL: `https://aclanthology.org/2024.emnlp-main.649/` (visited on 01/22/2026). + +- [20] Paul Lerner and Franc¸ois Yvon. “Unlike “Likely”, “Unlike” is Unlikely: BPE-based Segmentation hurts Morphological Derivations in LLMs”. In: _Proceedings of the 31st International Conference on Computational Linguistics_ . COLING 2025. Ed. by Owen Rambow, Leo Wanner, Marianna Apidianaki, Hend Al-Khalifa, Barbara Di Eugenio, and Steven Schockaert. Abu Dhabi, UAE: Association for Computational Linguistics, Jan. 2025, pp. 5181–5190. URL: `https://aclanthology.org/2025.coling-main.348/` (visited on 01/22/2026). + +- [21] Wanchao Liang et al. _TorchTitan: One-stop PyTorch native solution for production ready LLM pre-training_ . June 7, 2025. DOI: `10.48550/arXiv.2410.06511` . arXiv: `2410.06511[cs]` . URL: `http://arxiv.org/abs/2410.06511` (visited on 04/01/2026). + +- [22] Hong Liu et al. _Scaling Embeddings Outperforms Scaling Experts in Language Models_ . Jan. 29, 2026. DOI: `10.48550/arXiv.2601.21204` . arXiv: `2601.21204[cs]` . URL: `http: //arxiv.org/abs/2601.21204` (visited on 02/02/2026). + +- [23] Benjamin Minixhofer, Tyler Murray, Tomasz Limisiewicz, Anna Korhonen, Luke Zettlemoyer, Noah A. Smith, Edoardo M. Ponti, Luca Soldaini, and Valentin Hofmann. _Bolmo: Byteifying the Next Generation of Language Models_ . Dec. 17, 2025. DOI: `10.48550/arXiv. 2512.15586` . arXiv: `2512.15586[cs]` . URL: `http://arxiv.org/abs/2512.15586` (visited on 01/22/2026). + +- [24] Artidoro Pagnoni et al. “Byte Latent Transformer: Patches Scale Better Than Tokens”. In: _Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . ACL 2025. Ed. by Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pilehvar. Vienna, Austria: Association for Computational Linguistics, July 2025, pp. 9238–9258. ISBN: 979-8-89176-251-0. DOI: `10 . 18653 / v1 / 2025.acl- long.453` . URL: `https://aclanthology.org/2025.acl- long.453/` (visited on 02/03/2026). + +- [25] Guilherme Penedo, Hynek Kydl´ıˇcek, Loubna Ben Allal, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, and Thomas Wolf. “The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale”. In: The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track. Nov. 13, 2024. URL: `https: //openreview.net/forum?id=n6SCkn2QaG#discussion` (visited on 04/01/2026). + +- [26] Ivan Provilkov, Dmitrii Emelianenko, and Elena Voita. “BPE-Dropout: Simple and Effective Subword Regularization”. In: _Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics_ . ACL 2020. Ed. by Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault. Online: Association for Computational Linguistics, July 2020, pp. 1882– + +11 + +1892. DOI: `10.18653/v1/2020.acl-main.170` . URL: `https://aclanthology.org/ 2020.acl-main.170/` (visited on 01/22/2026). + +- [27] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. “Language models are unsupervised multitask learners”. In: _OpenAI blog_ 1.8 (2019), p. 9. + +- [28] Aurko Roy et al. _N-Grammer: Augmenting Transformers with latent n-grams_ . July 13, 2022. DOI: `10.48550/arXiv.2207.06366` . arXiv: `2207.06366[cs]` . URL: `http://arxiv. org/abs/2207.06366` (visited on 01/22/2026). + +- [29] Phillip Rust, Jonas Pfeiffer, Ivan Vuli´c, Sebastian Ruder, and Iryna Gurevych. “How Good is Your Tokenizer? On the Monolingual Performance of Multilingual Language Models”. In: _Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)_ . ACL-IJCNLP 2021. Ed. by Chengqing Zong, Fei Xia, Wenjie Li, and Roberto Navigli. Online: Association for Computational Linguistics, Aug. 2021, pp. 3118–3135. DOI: `10.18653/v1/2021.acl-long.243` . URL: `https://aclanthology.org/2021.acllong.243/` (visited on 01/22/2026). + +- [30] Craig W Schmidt, Varshini Reddy, Haoran Zhang, Alec Alameddine, Omri Uzan, Yuval Pinter, and Chris Tanner. “Tokenization Is More Than Compression”. In: _Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing_ . EMNLP 2024. Ed. by Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen. Miami, Florida, USA: Association for Computational Linguistics, Nov. 2024, pp. 678–702. DOI: `10.18653/v1/2024.emnlpmain . 40` . URL: `https : / / aclanthology . org / 2024 . emnlp - main . 40/` (visited on 01/22/2026). + +- [31] Rico Sennrich, Barry Haddow, and Alexandra Birch. “Neural Machine Translation of Rare Words with Subword Units”. In: _Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . ACL 2016. Ed. by Katrin Erk and Noah A. Smith. Berlin, Germany: Association for Computational Linguistics, Aug. 2016, pp. 1715–1725. DOI: `10.18653/v1/P16-1162` . URL: `https://aclanthology.org/P161162/` (visited on 01/22/2026). + +- [32] Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, and Bryan Catanzaro. _Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism_ . Mar. 13, 2020. DOI: `10.48550/arXiv.1909.08053` . arXiv: `1909. 08053[cs]` . URL: `http://arxiv.org/abs/1909.08053` (visited on 04/01/2026). + +- [33] Chaofan Tao, Qian Liu, Longxu Dou, Niklas Muennighoff, Zhongwei Wan, Ping Luo, Min Lin, and Ngai Wong. “Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabularies”. In: The Thirty-eighth Annual Conference on Neural Information Processing Systems. Nov. 6, 2024. URL: `https:// openreview.net/forum? id=sKCKPr8cRL` (visited on 01/22/2026). + +- [34] Hugo Touvron et al. _LLaMA: Open and Efficient Foundation Language Models_ . Feb. 27, 2023. DOI: `10.48550/arXiv.2302.13971` . arXiv: `2302.13971[cs]` . URL: `http://arxiv. org/abs/2302.13971` (visited on 04/01/2026). + +- [35] Unicode Consortium. _The Unicode Standard: Worldwide Character Encoding_ . Reading, Massachusetts: Addison Wesley, Oct. 1991. ISBN: 0-201-56788-1. + +- [36] Franc¸ois Yergeau. _UTF-8, a transformation format of ISO 10646_ . Request for Comments RFC 3629. Num Pages: 14. Internet Engineering Task Force, Nov. 2003. DOI: `10.17487/ RFC3629` . URL: `https://datatracker.ietf.org/doc/rfc3629` (visited on 04/01/2026). + +- [37] Lin Zheng, Xinyu Li, Qian Liu, Xiachong Feng, and Lingpeng Kong. _Proxy Compression for Language Modeling_ . Feb. 4, 2026. DOI: `10.48550/arXiv.2602.04289` . arXiv: `2602. 04289[cs]` . URL: `http://arxiv.org/abs/2602.04289` (visited on 02/11/2026). + +- [38] Vil´em Zouhar, Clara Meister, Juan Gastaldi, Li Du, Tim Vieira, Mrinmaya Sachan, and Ryan Cotterell. “A Formal Perspective on Byte-Pair Encoding”. In: _Findings of the Association for Computational Linguistics: ACL 2023_ . Findings 2023. Ed. by Anna Rogers, Jordan BoydGraber, and Naoaki Okazaki. Toronto, Canada: Association for Computational Linguistics, July 2023, pp. 598–614. DOI: `10.18653/v1/2023.findings- acl.38` . URL: `https: //aclanthology.org/2023.findings-acl.38/` (visited on 01/22/2026). + +12 + +## **A Hyperparameters** + +Table 1: Training hyperparameters + +|Hyperparameter|Value| +|---|---| +|Architecture
# parameters
Dimension
# layers
# heads
# KV heads
Optimizer
_β_1_, β_2
Weight decay
Global batch size
Sequence length
Positional encoding
Learning rate scheduler
Learning rate (peak)
Training steps
LR warmup
LR decay|LLaMA-3 [14]
1_,_747_,_060_,_736
2_,_048
32
32
8
AdamW
0_._9,0_._95
0.1
128
8_,_192
RoPE
Warmup-Stable Cosine Decay
3_∗_10_−_4
100_,_000steps
2_,_000steps
10_,_000steps| + + + +Table 2: Multihead _n_ -gram embedding hyperparameters + +|Hyperparameter|Value| +|---|---| +|# extra parameters
Maximum order_n_of_n_-grams
Base_n_-gram vocabulary size at each order
Token vocabulary size
Embedding dimension
# heads|71_,_065_,_088
3
15_,_000
512
2048
8| + + + +## **B Smaller-scale experiments** + +13 + +**==> picture [360 x 605] intentionally omitted <==** + +**----- Start of picture text -----**
+10.0 Model A - baseline 100 Back to normal regime Model A - baseline
Model B - scaled vocabulary parameters Model B - increased sample throughput
9.5
10.0
9.0
9.5
8.5 9.0
8.5
8.0
8.0
7.5
7.5
7.0 7.0
Training Steps Training Steps
(a) Scaling input embedding parameters (b) Scaling sample throughput by 4 times for 20k
steps
10.0 Model A - baseline 100 Back to normal regime Model A - baseline
Model B - subword start boundaries prior Model B - subword start boundaries inductive bias
9.5 Model C - subword end boundaries prior Model C - subword end boundaries inductive bias
9.0 10.0
8.5 9.5
9.0
8.0
8.5
7.5
8.0
7.0 7.5
6.5 7.0
Training Steps Training Steps
(c) Subword boundaries as a prior (d) Subword boundaries as an inductive bias for
20k steps
10.0 Model A - baseline 100 Back to normal regime Model A - baseline
Model B - subword distances prior Model B - subword distances inductive bias
9.5
10.0
9.0
9.5
8.5 9.0
8.5
8.0
8.0
7.5
7.5
7.0 7.0
Training Steps Training Steps
(e) Subword distances as a prior (f) Subword distances as an inductive bias for 20k
steps for training only
10.0 Model A - baseline 100 Back to normal regime Model A - baseline
Model B - cross-entropy per subword Model B - next subword prediction
9.5
10.0
9.0
9.5
8.5 9.0
8.5
8.0
8.0
7.5
7.5
7.0 7.0
Training Steps Training Steps
(g) Optimizing for cross-entropy per subword (h) Optimizing for next subword prediction for
20k steps
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
Validation Loss (bits-per-byte)
**----- End of picture text -----**
+ + +Figure 7: Validation loss for different experiments with a 68M parameters model + +14 diff --git a/q128/random_k4/random_4.md b/q128/random_k4/random_4.md index 8efe386e6cfdafd89c56e967acf5dfb7f8070736..dd5d594edac856e6a90c63c0d5c3c52351225e37 100644 --- a/q128/random_k4/random_4.md +++ b/q128/random_k4/random_4.md @@ -1,3327 +1,771 @@ -Cretaceous stratigraphy, paleontology, petrography. Depositional -environments, and cycle stratigraphy at Cerro de Cristo Rey, Dona Ana -County, New Mexico - -Spencer G. Lucas, Karl Krainer, Justin A. Spielmann, and Kevin Durney - -New Mexico Geology, v. 32, n. 4 pp. 103-130, Print ISSN: 0196-948X, Online ISSN: 2837-6420. -https://doi.org/10.58799/NMG-v32n4.103 -Download from: https://geoinfo.nmt.edu/publications/periodicals/nmg/backissues/home.cfml?volume=32&number=4 - -New Mexico Geology (NMG) publishes peer-reviewed geoscience papers focusing on New Mexico and the -surrounding region. We aslo welcome submissions to the Gallery of Geology, which presents images of -geologic interest (landscape images, maps, specimen photos, etc.) accompanied by a short description. - -Published quarterly since 1979, NMG transitioned to an online format in 2015, and is currently being issued -twice a year. NMG papers are available for download at no charge from our website. You can also -subscribe to receive email notifications when new issues are published. - -New Mexico Bureau of Geology & Mineral Resources -New Mexico Institute of Mining & Technology -801 Leroy Place -Socorro, NM 87801-4796 - -https://geoinfo.nmt.edu - - This page is intentionally left blank to maintain order of facing pages. - - Cretaceous stratigraphy, paleontology, petrography, -depositional environments, and cycle stratigraphy at -Cerro de Cristo Rey, Doña Ana County, New Mexico - -Spencer G. Lucas, New Mexico Museum of Natural History, 1801 Mountain Road NW, Albuquerque, New Mexico 87104, spencer.lucas@state.nm.us; - -Karl Krainer, Institute of Geology and Paleontology, Innsbruck University, Innrain 52, Innsbruck, A-6020 Austria, Karl.Krainer@uibk.ac.at; -Justin A. Spielmann, New Mexico Museum of Natural History, 1801 Mountain Road NW, Albuquerque, New Mexico 87104, -justin.spielmann1@state.nm.us; Kevin Durney, 11212 Jockey Bluff Drive, Austin, Texas 78748 - -Abstract - -Cretaceous marine and nonmarine strata -of late Albian–middle Cenomanian age are -exposed around the Cerro de Cristo Rey uplift -in southern Doña Ana County, New Mexico. -These strata comprise a section approximately -350 m thick and are assigned to the (ascending -order) Finlay, Del Norte, Smeltertown, Mule- -ros, Mesilla Valley, Mojado (=“Anapra”), Del -Rio, Buda, and Mancos (=“Boquillas”) For- -mations. Macrofossils and microfossils from -these strata indicate that the Finlay, Del Norte, -Smeltertown, Muleros, and Mesilla Valley -Formations are of late Albian age, whereas the -Del Rio, Buda, and Mancos Formations are of -Cenomanian age. The base of the Cenomanian -is most likely at a trangressive surface within -the uppermost Mojado Formation. The late -Albian (Manuaniceras powelli Zone) to early -Cenomanian (Neophlycticeras hyatti Zone) -sedimentary succession at Cerro de Cristo Rey -consists of alternating fossiliferous limestone, -shale with limestone and sandstone intercala- -tions, and sandstone. Muddy limestone types -are commonly wavy to nodular and represent -deposits of an open-marine shelf environment -below wave base. Intercalated coquina beds -rich in mollusc shells are interpreted as storm -layers. Shale was deposited in an open-shelf -environment below or near wave base during -periods of increased siliciclastic influx. Inter- -calated thin limestone and sandstone beds are -inferred to be storm layers. The siliciclastic -Mojado Formation is a regressive-transgres- -sive succession formed in depositional envi- -ronments ranging from lower shoreface to -upper shoreface and even fluvial, again over- -lain by shallow-marine siliciclastics. Although -the Washita Group section at Cerro de Cristo -Rey is much thicker and displays some differ- -ences in facies, the succession shows similar -transgressive and regressive trends compared -to the Washita Group of north Texas. Thus, -we recognize eight unconformity-bounded -depositional cycles in the Cretaceous sec- -tion at Cerro de Cristo Rey, the upper Finlay -Formation (youngest cycle of the Freder- -icksburg Group), lower Mancos Formation -(base of Greenhorn cycle), and six Washita -Group cycles: WA1 = Del Norte Formation, -WA2 = Smeltertown Formation, WA3 = Mule- -ros Formation, WA4 = most of Mesilla Valley -Formation, WA5 = uppermost Mesilla Val- -ley Formation and most of Mojado Forma- -tion, and WA6 = uppermost Mojado and Del -Rio and Buda Formations. The persistence -of cycles from the tectonically passive, open- -marine margin of the Gulf of Mexico into the -tectonically active Chihuahua trough suggests -that regional if not global eustasy, not local -tectonism, drove late Early to early Late Cre- -taceous sedimentation at Cerro de Cristo Rey. - -FIGURE 1—Location of Cerro de Cristo Rey in southern Doña Ana County, New Mexico (after Lovejoy -1976). ∆ indicates Eocene igneous intrusions—Cerro de Cristo Rey, Cerro de la Mina, Campus Andesite, -the Three Sisters, and Vado Hill. -Introduction - -Cerro de Cristo Rey is a prominent peak in -Doña Ana County, New Mexico, just north -of the U.S.–Mexican border and just west of -the city of El Paso, Texas (Fig. 1). The moun- -tain was long referred to as the Cerro de -Muleros, but was renamed Cerro de Cristo -Rey (“Sierra” de Cristo Rey of Hook 2008 -and Cobban et al. 2008, also appears on the -U.S. Geological Survey topographic map of -the area) in recognition of the large statue of -Christ on the cross at its crest. The mountain -has an elevation of 1,425 m (4,675 ft) (Love- -joy 1976; Hook 2008). - -The core of Cerro de Cristo Rey is an -andesite laccolith, the Muleros Andes- -ite of Eocene age (Fig. 2). The laccolith is -surrounded by “an annulus of…faulted - -Cretaceous marine strata, locally strongly -deformed by gravity-glide structures trig- -gered by andesite intrusion” (Lovejoy 1976, -p. 24). The Cretaceous strata, a section -approximately 350 m thick, include rocks -of Early Cretaceous (late Albian) and Late -Cretaceous (early–middle Cenomanian) -age (Figs. 3–4). These rocks have been stud- -ied for more than a century, most notably -by Böse (1910) and Strain (1976). Here, we -present the first detailed lithostratigraphy -and sedimentary petrography of the Cre- -taceous strata exposed around the New -Mexican periphery of the Cerro de Cristo -Rey uplift. We combine these data with -paleontology and regional correlations to -present the first detailed interpretation of -the depositional environments and cycle -stratigraphy of the Cretaceous section. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -103 - - the Cretaceous invertebrate macrofossils -he collected at Cerro de Cristo Rey. - -Adkins (1932) and Imlay (1944) sub- -sequently applied Texas stratigraphic -names to the section Böse (1910) delin- -eated at Cerro de Cristo Rey (Fig. 3). Cór- -doba (1968, 1969a,b) proposed a formal -lithostratigraphic nomenclature for the -Cretaceous section in northeastern Chi- -huahua, Mexico. He mapped these units -as far north as the international border, so -he assigned Cretaceous strata along the -southern flank of Cerro de Cristo Rey in -Chihuahua to strata of his Ojinaga Group. -Córdoba (1969a) makes clear the equiva- -lence between his stratigraphic names and -those introduced at Cerro de Cristo Rey by -Strain (1968, 1976). - -The lithostratigraphic names long -applied to Cretaceous strata at Cerro de -Cristo Rey were briefly introduced by -Strain (1968) and elaborated on by Strain -(1976) and Lovejoy (1976). Thus, four strati- -graphic names from farther east, in Texas -(Finlay, Del Rio, Buda, and Boquillas), are -applied, whereas Strain introduced five -local lithostratigraphic names—Del Norte, -Smeltertown, Muleros, Mesilla Valley, and -Anapra (Figs. 3–4). - -Subsequent workers (e.g., LeMone and -Simpson 1983; Maudlin 1985; Cornell and -LeMone 1987; Cornell et al. 1991; Cornell -1997; Lucas and Estep 1998b, 2000; Scott -et al. 2001, 2003; Turnšek et al. 2003) have -employed the lithostratigraphy of Strain -(1976). The only previously suggested -modification is that of Lucas and Estep -(1998b, 2000), who argued that the Anapra -Sandstone is part of the same lithosome as -the Mojado Formation to the west. They -therefore abandoned the name Anapra -Sandstone and replaced it with the term -Mojado Formation. Here, we employ the -lithostratigraphic nomenclature of Strain -(1968, 1976) with two modifications, -the use of Mojado Formation instead of -Anapra Sandstone and the use of Mancos -Formation instead of Boquillas Formation -(Fig. 4). These modifications are justified -below. - - Del Norte - -FIGURE 2—Geologic map of part of the New Mexican portion of the Cerro de Cristo Rey uplift (from -Lovejoy 1976) showing locations of measured stratigraphic sections in this study. See Appendix for -map coordinates of the measured sections. - -Located on the northern margin of the -Chihuahua trough, the Cretaceous section -at Cerro de Cristo Rey is between Bisbee -Group sections to the west and Washita -Group sections to the east. It thus provides -a pivotal reference section for regional cor- -relation and testing concepts of Washita -Group cycle stratigraphy. - -Previous studies - -Emory (1857), Stanton and Vaughan (1896), -and Richardson (1909) first reported on -the Cretaceous rocks exposed around -Cerro de Cristo Rey (Fig. 3). However, the -real starting point for understanding the - -stratigraphy and paleontology of these -strata was the monograph of Böse (1910). -He informally divided the Cretaceous -section at Cerro de Cristo Rey into 11 -stratigraphic units, many characterized -by a distinctive fossil assemblage (Fig. 3). -Although Böse (1910) explicitly correlated -these units to formally named units of the -Fredericksburg and Washita Groups and -overlying Eagle Ford Shale of Texas (as -had Stanton and Vaughan [1896] and was -later done by Adkins [1932]), he applied -no formal lithostratigraphic nomencla- -ture to the Cretaceous section at Cerro de -Cristo Rey. Most significantly, Böse (1910) -provided a monographic description of - -FIGURE 3—Evolution of Cretaceous lithostratigraphic nomenclature at Cerro de Cristo Rey. Modified from Strain (1976). - -104 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - and serpulids are also present (Böse 1910; -Strain 1976; Turnšek et al. 2003; Kollmann -et al. 2003). In addition, our thin sections -(see below) demonstrate that calcareous -algae, ostracods, and rare calcisponges are -present. Kollmann et al. (2003) recently -analyzed the gastropod fossil assem- -blage of the Finlay Formation at Cerro de -Cristo Rey, identifying the following taxa: -Aporrhaidae, Diozoptyxis, Drepanocheilus, -Eunerinea aquiline, Naticoidea, Nerinella -n. sp., Otostoma elpasensis, Plesioptyxis tex- -anus, Rostroptygmatis kervillensis, Turritella, -and Tylostoma. - -Sedimentary petrography (Figs. 6, 16) - -Finlay Formation limestone beds at Cerro -de Cristo Rey are mostly bioclastic wacke- -stone that is fine-grained and commonly -bioturbated. Fossils include small mollusc -fragments (bivalves, gastropods), ostra- -cods, calcareous algae, a few echinoderms, -echinoid spines, and many foraminiferans -embedded in a pelmicritic matrix (Fig. 16A– -C). Rare calcisponges are present. Locally, -the wackestone grades into packstone and -rarely into washed grainstone containing -calcite cement (Fig. 16D–F). Many bioclasts -are recrystallized. Common foraminiferans -are Dictyoconus (particularly in units 2 and -8; Fig. 16D–F) and miliolids, Nummoloculina, -Cornuspira, and Ophthalmidium. Most algae -are dasycladacean algae (Permocalculus) and -aciculariaceans (Acicularia or Terquemella). -This facies corresponds to Facies F (bioclas- -tic dasycladacean wackestone) of Kollmann -et al. (2003). - -In the Finlay Formation, beds of inter- -calated nodular limestone are 0.2–5 m -thick and composed of limestone nodules -embedded in shale. The limestone nodules -consist of bioturbated mudstone contain- -ing a few larger mollusc shell fragments -(bivalves, gastropods) and small bioclasts -derived from molluscs, ostracods, rare -echinoderms, echinoid spines, and smaller -foraminiferans. - -Depositional environments - -The Finlay Formation section exposed at -Cerro de Cristo Rey is very similar to the -upper part of the Finlay Formation at its stra- -totype section in Hudspeth County, Texas -(Steinhoff 2003). Thus, the Finlay Formation -strata at Cristo Rey are limestones and nodu- -lar limestones largely devoid of coarse silici- -clastic material. Thin-shelled bivalves and -spatangoid echinoids dominate the macro- -fossil assemblage, and the benthic foraminif- -eran Dictyoconus walnutensis is particularly -abundant. Deposition clearly took place on -a shallow-marine platform in a relatively -restricted circulation shelf environment, and -this fits the regional picture of upper Finlay -deposition (Scott and Kidson 1977; Reaser -and Malott 1985; Steinhoff 2003). - -Microfacies and the diverse biota, includ- -ing broken mollusc fragments in the wacke- -stone to packstone of the massive limestone - -FIGURE 4—Summary of stratigraphic section and age assignments of Cretaceous strata at Cerro de -Cristo Rey (modified from Lucas and Estep 1998b, 2000). Only ammonoid zone names in boldface are -known from ammonoid fossils collected at Cerro de Cristo Rey; presence of other zones is inferred. - -Overview - -Finlay Formation - -Below, we present a brief summary of the -lithostratigraphy of the Cretaceous rocks -exposed at Cerro de Cristo Rey (Fig. 4), -based primarily on the stratigraphic sec- -tions we measured (Figs. 5–15) and on data -in Strain (1976) and Lovejoy (1976). More -detailed lithologic information on some -of the rock units is presented below in -the sections on sedimentary petrography -(Figs. 16–24). Note that we refer to all the -Cretaceous formation rank units exposed -at Cerro de Cristo Rey with the suffix “for- -mation,” instead of using the lithic terms -limestone, shale, or sandstone. - -Böse (1910) published most of what is -known of the Cretaceous paleontology of -Cerro de Cristo Rey. We have extensive -macrofossil collections from Cerro de Cris- -to Rey that we will document elsewhere. -Here, we provide brief summaries of the -paleontology, mostly of relevance to inter- -preting the age and depositional environ- -ments of the Cretaceous strata exposed at -Cerro de Cristo Rey. We also present a sedi- -mentary petrography of the Cretaceous sec- -tion at Cerro de Cristo Rey, and we briefly -summarize the depositional environments -of the Cretaceous formations. The final sec- -tion of this article brings these data and -analyses together to present an interpreta- -tion of the cycle stratigraphy of the Creta- -ceous strata exposed at Cerro de Cristo Rey. - -Lithostratigraphy - -The Finlay Formation is the stratigraphi- -cally lowest unit of the Cretaceous sec- -tion exposed in the New Mexico outcrops -at Cerro de Cristo Rey, but its base is not -exposed. Lovejoy (1976) reported a thick- -ness of 40 m, and total exposed thickness of -our measured section of the Finlay Forma- -tion is approximately 41 m (Figs. 5A, 6). The -only outcrop of the Finlay Formation on the -New Mexico flank of the uplift is located at -our measured section (Fig. 2). - -Finlay Formation strata are a conspicu- -ous ledge- and cliff-forming interval of -interbedded light-gray and white lime- -stone and nodular limestone in a marly -matrix (Figs. 5A, 6). The succession is -thus composed of alternating fossilifer- -ous limestone ledges and slope-forming -nodular limestone in shale. Limestone -ledges are 0.4–5.3 m thick and composed -of bioclastic wackestone to packstone and -some rare grainstone. - -Paleontology - -Bivalves and gastropods are the most -common fossils in the Finlay Formation, -but foraminiferans (especially Dictyoco- -nus walnutensis: Figs. 16D–F, 25R–S), cor- -als, ammonoids, spatangoid echinoderms, - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -105 - - FIGURE 5—Photos of selected Cretaceous outcrops at Cerro de Cristo Rey. -A—Overview of Finlay Formation section (Fig. 6). B—Overview of part of -lower member of Del Norte Formation (Fig. 7). C—Overview of middle part -of Smeltertown Formation (Fig. 8). D—Overview of characteristic outcrop - -of Muleros Formation (Fig. 9). E—Texigryphaea packstone in lower part of -Muleros Formation. F—Overview of shale slope formed by Mesilla Valley -Formation capped by sandstone at base of Mojado Formation (Fig. 10). - -ledges, indicate deposition in an open-shelf -setting with moderate energy. The presence -of dasycladacean algae points to deposition -within the photic zone. Intercalated grain- -stone suggests periodically high-energy -conditions. The wavy to nodular bedded -limestone, composed mainly of bioturbated -mudstone with a less diverse fauna, sug- -gest deposition in a low-energy, shallow- -marine shelf environment. - -The facies of the massive and wavy to -nodular limestone is very similar to that - -of the stratotype section of the Finlay For- -mation, although some differences exist. -According to Steinhoff (2003), at the strato- -type the lower part of the Finlay Formation -is composed of nodular to massive-bedded -limestone (bioclastic wackestone to pack- -stone) interbedded with marl and clay. In the -upper part of the Finlay Formation massive- -bedded limestone (bioclastic wackestone to -packstone with a slightly different fauna) -is abundant. Rudists, which are common -in the stratotype section, particularly in the - -upper part regionally (e.g., Steinhoff 2003), -are absent at Cristo Rey. The lower Finlay -Formation of the stratotype is composed of -deposits formed on a carbonate ramp in a -shallow open-marine shelf environment, -whereas the upper Finlay Formation repre- -sents deposits of a restricted circulation shelf -environment (Steinhoff 2003). Such a trend -from a carbonate ramp in the lower Finlay to -a restricted shelf environment in the upper -Finlay is not evident at Cristo Rey, although -some shallowing occurred. - -106 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - The cyclic alternation of massive and -wavy to nodular limestone evident in the -Finlay Formation section at Cerro de Cristo -Rey (Fig. 6) is also recognized at the Fin- -lay Formation stratotype, where Steinhoff -(2003) described nine transgressive-regres- -sive high-frequency cycles in the lower -high-frequency sequence 1, and 18 cycles in -the overlying high-frequency sequence 2. -The stratotype section of the Finlay Forma- -tion at the Rimrock escarpment, west Texas, -is approximately 52 m thick, although the -thickness of the Finlay Formation var- -ies from less than 15 m to approximately -800 m and more in the deeper parts of its -depositional basin, the Chihuahua trough -(Steinhoff 2003). - -Del Norte Formation - -Lithostratigraphy - -At Cerro de Cristo Rey, the Del Norte For- -mation of Strain (1968, 1976) rests with sharp -contact on the Finlay Formation (Figs. 6–7). It -is subdivisions 2 and 3 of Böse (1910), which -he estimated to be 20–30 m thick. Our mea- -sured section of the Del Norte Formation -is the type section of Strain (1976, table 8), -where he reported the formation as approxi- -mately 17 m thick. However, our section of -the Del Norte Formation is 24 m thick and, -following Strain (1976), can be divided into -a lower (clay) member (approximately 20 m -of shale with interbedded sandstone and -limestone) and an upper (calcareous) mem- -ber (3–4 m of limestone) (Fig. 7). - -The basal bed of the Del Norte Forma- -tion is a prominent, 0.1-m-thick ferruginous -sandstone bed that rests with sharp contact -on underlying lime mudstone of the Fin- -lay Formation (Figs. 6–7). Above this bed, -the lower member (“Brick Plant Member” -of Strain 1968) of the Del Norte Formation -forms a slope above the Finlay Formation -(Fig. 5B) and is composed of alternating -beds of limestone, siltstone to fine-grained -sandstone, and shale. Limestone beds -are thin (0.2–0.6 m thick) and form ledges -(Fig. 5B). The lower two limestone beds are -nodular. Intercalated beds of calcareous silt- -stone to fine-grained sandstone in the lower -member are 0.3 m thick (Fig. 7, unit 8) and -2.1 m thick (Fig. 7, unit 16). The lower mem- -ber is capped by the cuesta- and ridge-form- -ing upper member (“Refinery Member” of -Strain 1968) of the Del Norte Formation, -which is 3.5 m thick. A 0.6-m-thick sandy -and shelly limestone ledge (Fig. 7, unit 19) -forms its base. - -Strain (1976, p. 78) stated that the “Lower -Clay Member [of the Del Norte Formation] -is conformable with the Finlay Limestone -below.” However, we agree with Scott et al. -(2001, 2003) and Turnšek et al. (2003) that -the base of the Del Norte Formation at -Cerro de Cristo Rey marks the basal uncon- -formity of the Washita Group (see section -on cycle stratigraphy, page 127). - -FIGURE 6—Measured stratigraphic section of Finlay Formation. See Figure 2 and Appendix for loca- -tion of section. - -Kollmann et al. (2003, fig. 9) interpreted -the facies succession of the exposed Finlay -Formation at Cerro de Cristo Rey as that of a -shelf lagoon intercalated, near the top of the -Finlay, with a bioherm fringe facies repre- -sented by “Orbitolina wackestone.” Accord- -ing to Kollmann et al. (2003), this bioherm -fringe facies is approximately 8–15 m below - -the top of the Finlay, so it corresponds to -units 20–24 of our measured section of the -Finlay Formation (Fig. 6). However, we did -not observe this “Orbitolina wackestone” -facies. “Orbitolina” (we take this to refer to -Dictyoconus) is present throughout the sec- -tion and is abundant also in the lower part -(particularly in units 2 and 8). - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -107 - - Sedimentary petrography (Figs. 7, 17) - -The basal sandstone of the Del Norte For- -mation (Fig. 7) is fine-grained, mixed silici- -clastic and carbonate in composition, con- -taining abundant angular quartz grains, -a few micas, and (probably) some detrital -feldspars. Carbonate grains are abundant. -A few fossil fragments such as shell debris, -ostracods, echinoderms, and dasyclad- -acean algae are present. The sandstone is -well sorted and laminated. - -Nodular limestone of the Del Norte For- -mation is fine-grained, poorly sorted, bio- -turbated bioclastic wackestone (Fig. 17A, -F) containing abundant fragments of -calcareous algae (Aciculariacea, Gym- -nocodiacea, ?Permocalculus). Mollusc shell -fragments (gastropods, bivalves), ostra- -cods, foraminiferans, and echinoid spines -embedded in gray, micritic, locally pelmic- -ritic matrix are subordinate. Algae and -molluscs are very greatly fragmented and -recrystallized. Rarely, small voids (?fossils) -filled with calcite spar are present. - -Limestone of the Del Norte Formation -is fine- to coarse-grained, poorly sorted, -and commonly bioturbated wackestone to -packstone (Fig. 17C, E). A few large mol- -lusc fragments several cm in size float in the -wackestone. Mollusc fragments (bivalves, -gastropods) and echinoderms are abundant. -Echinoid spines, ostracods, foraminiferans -(including some large, agglutinated forms), -and bryozoans(?) are subordinate. Fossils, -particularly molluscs, are commonly recrys- -tallized and strongly fragmented. Lithoclasts -include small detrital quartz grains that are -mostly 0.1–0.5 mm in diameter (rarely to -1 mm), constituting as much as about 20% -of the rock, and a few peloids and micritic -intraclasts. Rare glauconite grains are pre- -sent. The matrix is micrite. - -Siltstone to fine-grained sandstone of the -Del Norte Formation is commonly non- -laminated and bioturbated, and locally -indistinctly laminated (Fig. 17B, D). The -composition is mixed siliciclastic-carbonate -with as much as about 40% angular detrital -quartz grains that are mostly 0.1–0.2 mm in -diameter. Other lithoclasts are peloids, mic- -ritic intraclasts, and local fecal pellets with -diameters from 0.3 to 0.5 mm. The most -common fossil fragments are derived from -bivalves and gastropods, which are rarely -encrusted by bryozoans(?) and annelid -worms (serpulid worm tubes). Less abun- -dant are echinoderms, ostracods, foraminif- -erans (including rare agglutinated forms), -echinoid spines, and bryozoans(?). - -The basal limestone ledge of the upper -member of the Del Norte Formation (Fig. 7, -unit 19) is composed of poorly sorted bio- -clastic wackestone to rudstone containing -large, broken fragments of bivalves. Other -skeletons are echinoderms, gastropods, -ostracods, foraminiferans (including agglu- -tinated forms), echinoid spines, and dasy- -cladacean algae. A few intraclasts and many -detrital quartz grains are present. Locally, -the grains are densely packed, grading into -packstone. The matrix is micrite. - -This basal limestone is overlain by two -nodular limestone beds (Fig. 7, units 20, 22, -forming recessed notches) alternating with -limestone ledges. The microfacies is similar -to that of the basal limestone ledge: poorly -sorted wackestone to packstone and, in -part, floatstone containing fragmented and -abraded skeletons. Most abundant are mol- -lusc shell fragments derived from bivalves, -gastropods, and echinoderms. Echinoid -spines, ostracods, foraminiferans (includ- -ing some agglutinated forms), and bryo- -zoans are subordinate. Angular to rounded -detrital quartz grains, mostly 0.2–0.5 mm -in diameter and rarely to 1 mm, and a few -micritic intraclasts are present. The grains -are embedded in micritic matrix. - -Depositional environments - -Fossils and sediments of the Del Norte For- -mation indicate deposition in a shallow, -open-marine environment with periodical- -ly strong siliciclastic influx. The presence -of calcareous algae suggests deposition -within the photic zone. Sandstone beds in -the lower part displaying horizontal lami- -nation and current ripples are interpreted -to represent storm layers. Thin sandstone -interbeds displaying horizontal lamina- -tion and ripple lamination intercalated -in shale of a shallow-marine shelf setting -have commonly been attributed to storm -activity. Thus, Scott et al. (1975) described -similar siltstone to fine-grained sandstone -beds 1–10 cm thick in the Washita Group -of Texas, which they interpreted as density -currents formed by storms, tides, rip flow, -or floods. Hobday and Morton (1984) inter- -preted similar thin sandstone intercalations -in the Lower Cretaceous Grayson Forma- -tion (Texas) as deposits of strong bottom -currents below the storm wave base gener- -ated by large-scale, storm-generated, bot- -tom-return flow. - -Nodular limestone, which is commonly -bioturbated and contains micritic to pel- -micitic matrix, formed in a low-energy -environment, whereas intercalated lime- -stone beds containing strongly fragmented, -locally densely packed mollusc fragments -(coquina) indicate deposition under high- -energy conditions and may represent storm -layers. The sharp contact with the under- -lying Finlay Formation is interpreted as a -transgressive surface, which was followed -by deepening upward and a regressive -trend in the upper member of the Del Norte -Formation, which resulted in the deposi- -tion of limestone with abundant coquina -storm layers. - -Smeltertown Formation - -Lithostratigraphy - -The Smeltertown Formation of Strain (1976) is -mostly gray shale with limestone nodules that -forms a relatively thick slope (approximately - -FIGURE 7—Measured stratigraphic section of -Del Norte Formation. See Figure 2 and Appen- -dix for location of section, and Figure 6 for lithol- -ogy legend. - -Paleontology - -The Del Norte Formation has a diverse fossil -assemblage of bivalves (especially grypha- -eids, exogyrids, pectens, and oysters), gas- -tropods, spatangoid echinoderms, ammo- -noids, corals, sponges, and serpulids (Böse -1910; Strain 1976; Turnšek et al. 2003). In thin -section, we observed dasycladacean algae, -ostracods, foraminferans, echinoderms, and -serpulid worms (see below). Ammonites -of the upper Albian Adkinsites bravoensis -Zone (of Young 1966) are present in the -lower member of the Del Norte Forma- -tion, in units 15–18 of our measured section -(Fig. 7), which indicates correlation with the -Kiamichi Formation in north-central Texas. -No fossils diagnostic of the Eopachydiscus -marcianus or Craginites serratescens ammo- -nite zones are present at Cerro de Cristo Rey, -so their presence (Fig. 4) can only be inferred -in the upper Del Norte Formation or lower -Smeltertown Formation or missing at an -unconformity. - -108 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - 60 m) above the upper member of the Del -Norte Formation (Figs. 5C, 8). Our measured -section of the Smeltertown Formation (Fig. 8) -is the type section of Strain (1976, table 9), -which he reported to be approximately -61 m thick. The Smeltertown Formation -is subdivision 4 of Böse (1910), which he -estimated as 30–50 m thick. - -The base of the Smeltertown Formation -is dark-colored shale that sits directly on -limestone at the top of the Del Norte For- -mation (Figs. 7–8). In the lower part of the -Smeltertown Formation (Fig. 8, unit 3), -limestone nodules as much as 20 cm in -diameter are in a 3-m-thick shale inter- -val. Sparse limestone nodules are also in -shale units 4 and 6. Shale unit 5 contains -thin ledges of limestone nodules and a -pebbly agate zone near the top. This unit -also contains many ammonites, especially -specimens of Mortoniceras equidistans. The -limestone ledge of unit 7 yields many -shells of Texigryphaea. In unit 8, thin lime- -stone ledges are intercalated in shale. A -2.3-m-thick interval of sandstone ledges, -mostly full of abundant tests of the large, -arenaceous foraminiferan Cribratina texana -and intercalated with shale beds (with -many horizontal grazing traces of Palaeo- -phycus striatus), forms the top of the Smel- -tertown Formation (Fig. 8). Limestone at -the basal Muleros Formation rests with -sharp contact on sandstone at the top of -the Smeltertown Formation. - -Paleontology - -The Smeltertown Formation yields fossils -of foraminiferans, bivalves, gastropods, -ammonoids, bryozoans, crustaceans, ophi- -uroids, echinoids, brachiopods, and corals -(Böse 1910; Strain 1976; Nye and LeMone -1978; Bullock 1985; Bullock and Cornell -1986; Cornell et al. 1991; Turnšek et al. -2003). The foraminiferans establish a robust -correlation of the Smeltertown Formation -with the Duck Creek Formation of north- -central Texas (Bullock and Cornell 1986). -Ammonites of the Mortoniceras equidistans -Zone (present in unit 5 of our measured sec- -tion; Fig. 8) also correlate the Smeltertown -Formation to the lower part of the Mojado -Formation (Fryingpan Spring Member) in -the Cooke’s Range of Luna County, New -Mexico (Böse 1910; Cobban 1987; Lucas -et al. 1988; Lucas and Estep 1998b, 2000). - -Sedimentary petrography (Figs. 8, 18, 19A–D) - -Unit 3 in our measured section of the -Smeltertown Formation (Fig. 8) is a mixed -siliciclastic-carbonate siltstone that is -partly laminated and partly bioturbated -(Fig. 18A). The siltstone contains as much -as 30% detrital quartz grains, abundant car- -bonate grains, rare glauconite, and micrite. -A few large mollusc fragments (bivalves -and gastropods) float in the siltstone. Other -fossils present are ostracods, a few smal- -ler foraminiferans, rare echinoderms, and -echinoid spines. - -The nodular limestone ledges of unit 5 of -the Smeltertown Formation section (Fig. 8) -are composed of bioclastic, mixed silici- -clastic-carbonate siltstone, which is partly -laminated and partly bioturbated. A few -large mollusc shell fragments of bivalves -and gastropods float in silty matrix. The -siltstone consists of as much as 30% quartz -grains, small carbonate grains, a few -glauconite grains, and micrite. Bioclasts -include abundant ostracods, a few smaller -foraminiferans, rare echinoderms, echinoid -spines, and shell debris. - -The microfacies of the unit 7 limestone -ledge (Fig. 8) is fine-grained, bioturbated -mixed siliciclastic-carbonate sandstone -(Fig. 18B). The sandstone contains abundant -detrital quartz grains, mostly 0.2–0.5 mm in -diameter, rarely to 1 mm. Most common is -monocrystalline quartz, and polycrystal- -line quartz is subordinate. Rare chert and -glauconite grains are present. Quartz grains -are surrounded by thin carbonate cement -rims. A few micritic intraclasts are present, -too. Bioclasts include bivalves, gastropods, -echinoderms, ostracods, foraminiferans, -and rare worm tubes(?). Litho- and bio- -clasts float in a micritic matrix. - -Thin limestone beds of the uppermost -Smeltertown Formation (Fig. 8, unit 8) -consist of bioturbated mudstone (Fig. 18C) -containing a few siliciclastic grains (mostly -quartz, rare chert, and glauconite grains) -and bioclasts (mostly ostracods, rare -bivalve shells, and echinoderm fragments; -probably other skeletons). Siliciclastic -grains and bioclasts constitute less than -10% of the rock, and they are locally con- -centrated due to bioturbation. - -The top siltstone to sandstone of the Smel- -tertown Formation (Fig. 8, unit 9) is fine -grained, laminated, and locally bioturbated -(Fig. 19A–D). Siliciclastic grains constitute -about 50–70% of the rock. Most abundant is -detrital quartz, and rare are grains of glau- -conite, opaque minerals, and micas. A few -micritic intraclasts are present. Bioclasts, -such as shell fragments, echinoderms, and -ostracods, are rare. This facies grades into -bioclastic wackestone that contains agglu- -tinated foraminiferans, bivalve fragments, -and some other bioclasts embedded in silty, -partly peloidal matrix with a few small -quartz grains (Fig. 18D–F). - -Depositional environments - -Dark-colored shale of the Smeltertown For- -mation is indicative of a muddy marine -sea bottom (cf. Scott 1970). From shales -of the Smeltertown Formation, Bullock -and Cornell (1986) identified 73 species of -foraminiferans that belong to the suborders -Texturlariina, Miliolina, and Rotaliina and -provided some information on the deposi- -tional environment. Within the formation -they distinguished three zones character- -ized by distinct assemblages of foraminifer- -ans. Zone 1 (lower 13 m), which correlates -with unit 1 of Strain (1976) and with units - -FIGURE 8—Measured stratigraphic section -of Smeltertown Formation. See Figure 2 and -Appendix for location of section, and Figure 6 -for lithology legend. Dashed lines identify the -three zones of Bullock and Cornell (1986). - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -109 - - planktonic foraminiferans are absent, -and other rotaliine forms are less abun- -dant than in the underlying strata. Textu- -lariine forms constitute 70% of the tests. -Individuals of Cribratina texana comprise -60–99% of the specimens, indicating an -extremely restricted environment, prob- -ably a hypersaline marsh according to -Bullock and Cornell (1986), or a shal- -low, nearshore shelf with possibly lower -than normal salinity (R. W. Scott, written -comm. 2010). - -The foraminiferal assemblages of the -shales thus indicate that the Smeltertown -Formation represents an upward-shallow- -ing succession from a deeper, open shelf -with water depths of approximately 50 m -to a shallow restricted shelf environment -with increased salinity. Interbedded silt- -stone, sandstone, and limestone beds in the -upper part of the succession may represent -storm layers similar to those of the Del -Norte Formation. - -Muleros Formation - -Lithostratigraphy - -Our measured section of the Muleros For- -mation (Fig. 9) is the type section of the for- -mation (Strain 1976, table 10). The Muleros -Formation is subdivision 5 of Böse (1910), -which he reported as 30–50 m thick; Strain -(1976) reported a thickness of 32 m for the -formation type section. Our measured sec- -tion of the Muleros Formation is approxi- -mately 33 m thick. The formation generally -forms cliffs and ledgy slopes (Fig. 5D), and -many of the limestone beds of the middle -and upper units are Texigryphaea pack- -stones (Figs. 5E and 9). - -Following Strain (1976), the Muleros For- -mation can be divided into three members -(Fig. 9). The lower 11–12 m is light-colored -shale with some thin, intercalated beds of -bioturbated sandstone and limestone with -abundant Texigryphaea. The limestone is -nodules embedded in brownish silty matrix -and micrite. The middle part of the Muleros -Formation (from 11.5 to 24.3 m) is composed -of bioturbated micritic limestone interca- -lated with thin shale and, in the lower part, -thin bioturbated sandstone. The limestone -contains Texigryphaea and Cribratina and is -bioturbated. Horizontal, meniscate back- -filled burrows are present in unit 5 (Fig. 9). -The upper third of the section (approxi- -mately 9 m thick) is composed of alternat- -ing limestone ledges and shale. Limestone -intervals are 0.6–1.4 m thick and consist of -floatstone/rudstone and packstone contain- -ing Texigryphaea. Shale intervals in the upper -part are as much as 1.6 m thick. The lower -two intercalated shale intervals contain -limestone nodules. The base of the Mesilla -Valley Formation is a sharp contact of silty -shale on Texigryphaea packstone at the top of -the Muleros Formation (Fig. 9). - -Paleontology - -The Muleros Formation is very fossilifer- -ous, primarily because of the packstones of -the gryphaeid bivalve Texigryphaea washi- -taensis it contains (Böse 1910; Kues 1989). It -also yields other bivalves, gastropods, rare -ammonoids, common spatangoid echino- -derms, foraminiferans (especially the large, -arenaceous form Cribratina texana), corals, -sponge borings, bryozoans, ostracods, and -serpulids (Böse 1910; Strain 1976; Kues -1989; Turnšek et al. 2003). Correlation of -the Muleros Formation with the Denton -and Fort Worth Formations of north-central -Texas is supported by stratigraphic posi- -tion but lacks a robust ammonite biostrati- -graphic basis (Scott et al. 2003). - -Sedimentary petrography (Figs. 9, 19E–F, 20, 21) - -The uppermost bed of the Smeltertown -Formation (Fig. 8, unit 9) is a quartzose -siltstone to fine-grained sandstone that is -heavily bioturbated and partly laminated. -Most abundant are monocrystalline quartz -grains; chert, carbonate grains, opaque -minerals, and glauconite are rare. Small -bioclasts including shell debris, ostracods, -echinoderms, and a few large bivalve shells -float in siltstone to fine-grained sandstone. -The rock is cemented by quartz over- -growths, rarely by calcite cement. - -The limestone beds in the lower part of -the Muleros Formation (Fig. 9, unit 2) are -gray, bioturbated, fine grained, and com- -posed of abundant peloids, a few detrital -quartz grains, and bioclasts such as echi- -noderms, echinoid spines, ostracods, and a -few smaller foraminiferans. The matrix is -micrite. The limestone in the middle part of -the Muleros Formation (Fig. 9, units 3–8) is -mainly composed of bioclastic wackestone -to packstone (Fig. 20D), and subordinately -mudstone and floatstone to rudstone. - -The bioclastic wackestone is fine to coarse -grained, poorly sorted, commonly biotur- -bated, and locally laminated. Abundant -fossils are large gryphaeid bivalve shells, -subordinate gastropods, echinoderms, cor- -als, smaller foraminiferans (including large -agglutinated forms), ostracods, and echinoid -spines. A few glauconite grains are present. -Locally, peloids and detrital quartz grains -are present (Fig. 19E, F; Fig. 20A, B, F). - -The bioclastic mudstone is bioturbated -and contains as much as about 5% detrital -quartz grains and rare glauconite (Fig. 20E). -Fossils such as smaller foraminiferans -(including rare, larger agglutinated forms), -rare ostracods, echinoderms, echinoid -spines, and other skeletons are floating in -micrite. The following foraminiferans are -present in the limestone: Ammobaculites, -Ataxophragmiidae, and Nodosaria. - -The bioclastic siltstone to mudstone -(Fig. 9, unit 4) is nonlaminated and biotur- -bated (Fig. 20C). Many small and a few larg- -er bioclasts (echinoderms, ostracods, bivalve -shell fragments, rare smaller foraminiferans, - -FIGURE 9—Measured stratigraphic section of -Muleros Formation. See Figure 2 and Appendix -for location of section, and Figure 6 for lithology -legend. - -1 and 2 of our section (Fig. 8), is character- -ized by a high abundance of “Hedbergella” -planispira and the dominance of rotaliine -forms, which constitute 45–90% of the indi- -viduals. This assemblage indicates deposi- -tion in an open, inner-shelf environment -with water depths of 50 m or less (Bullock -and Cornell 1986). - -Zone 2, including unit 2 and the lower -9 m of unit 3 of Strain (1976), which is our -unit 3 and part of unit 4 (Fig. 8), is char- -acterized by textulariine forms, which -constitute 55–95% of all individuals, with -Cribratina texana being the most abundant -species. The presence of Ammobaculites -and Trochammina is indicative of a shal- -low, nearshore environment. Bullock and -Cornell (1986) concluded that during sedi- -mentation of this interval some shallowing -occurred, causing restriction of circulation. -In zone 3, encompassing the upper -24 m of the section (upper 22 m of unit -3 and unit 4 of Strain 1976; upper part -of our unit 4 and all of units 5–6; Fig. 8), - -110 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - corals, and bryozoans) float in a silty to mic- -ritic matrix. Peloids (fecal pellets?) are local- -ly present, as are small quartz grains (< 5%). -The floatstone to rudstone (Fig. 9, -unit 7) is nonlaminated, poorly sorted, -and contains abundant shell fragments of -gryphaeids and gastropods (Fig. 21A–B). -Many large agglutinated foraminiferans are -also present. These foraminiferans include -rare agglutinated smaller foraminiferans. -Echinoderm fragments, ostracods, smaller -foraminiferans, and shell debris are sub- -ordinate. A few mollusc shell fragments -are encrusted by worm tubes. The matrix -is gray peloidal micrite containing a few -small quartz grains. - -The limestone of unit 8 (Fig. 9) is fine- -grained, bioturbated wackestone (Fig. 21C) -containing abundant small bioclasts such as -ostracods, echinoderms, smaller foraminif- -erans, mollusc shell debris, and rare larger -skeletons of gastropods, echinoderms, and -bivalves. The matrix is peloidal micrite -containing a few quartz grains and rare -glauconite. - -Unit 10 of our Muleros Formation sec- -tion (Fig. 9) is a bioturbated floatstone- -rudstone, containing abundant bivalve -(Texigryphaea) and gastropod shell frag- -ments, many agglutinated foraminiferans, -echinoderms, and rare corals (Fig. 21D–E). -Smaller foraminiferans, ostracods, and -echinoid spines are subordinate. A few -skeletons are encrusted by worm tubes. -The micritic matrix contains a few angular -quartz grains and rare glauconite. - -The uppermost limestone interval of the -Muleros Formation (Fig. 9, units 13–14) is -a packstone–rudstone (Fig. 21F) composed -of densely packed bivalve shell fragments -that are oriented more or less parallel to -the bedding plane. Other bioclasts, such -as echinoderms, ostracods, and smaller -foraminiferans, are rare. The matrix is silt- -stone containing abundant angular quartz -grains and rare glauconite grains. Locally, -some calcite cement is present (“umbrel- -la” porosity). The peloidal wackestone -(pelmicrite; Fig. 9, unit 14) is laminated and -consists of abundant small peloids, mostly -around 0.05 mm in size and a few peloids -as large as 0.25 mm in diameter. Bioclasts -such as ostracods and foraminiferans -are rare. The floatstone–rudstone of the -upper part is poorly sorted, nonlaminated, -partly bioturbated, and contains abun- -dant large gryphaeid and gastropod shell -fragments. A few shells display borings -and are encrusted by annelid worms (ser- -pulid worm tubes). Other bioclasts include -echinoderms, echinoid spines, ostracods, -foraminiferans (Lenticulina), rare corals, -and bryozoans(?). Locally, agglutinated -foraminiferans are common. The rock con- -tains many detrital angular quartz grains -and a few glauconite grains. The matrix is -micrite. - -FIGURE 10—Measured stratigraphic section -of Mesilla Valley Formation. See Figure 2 and -Appendix for location of section, and Figure 6 -for lithology legend. - -FIGURE 11—Measured stratigraphic section of -Mojado Formation (“Anapra Sandstone”). See -Figure 2 and Appendix for location of section, and -Figure 6 for lithology legend. HCS=hummocky -cross-stratification. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -111 - - FIGURE 12—Photos of Cretaceous outcrops at Cerro de Cristo Rey. -A—Outcrops of Mojado Formation overlying Mesilla Valley Forma- -tion (Fig. 10). B—Characteristic crossbedded sandstone of Mojado -Formation. C—Dinosaur footprints (ornithopod tracks) in Mojado - -Formation. Divisions of measuring stick = 0.25 m. D—Base of Del Rio -Formation on Mojado Formation (Figs. 11 and 13). E—Overview of Buda -Formation measured section (Fig. 14). F—Contact of Buda and Mancos -Formations at measured section (Fig. 15). - -Depositional environments - -Muleros strata also demonstrate an appar- -ent upward-shallowing trend. Thus, lower, - -shale-dominated strata are relatively deep- -er marine deposits (though still shallow), -whereas Texigryphaea packstones and inter- -calated bioturbated limestones and sandy - -limestone of the middle-upper Muleros -(Fig. 9) suggest shallow, carbonate shelf -deposition with some coarse clastic influx -(cf. Turnšek et al. 2003). - -112 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - transport and deposition as density current -deposits (cf. Scott et al. 1975; Hobday and -Morton 1984). - -The middle part of the Muleros Forma- -tion (Fig. 9, units 3–8) is dominated by dif- -ferent types of limestone containing a fossil -assemblage suggesting a shallow-marine -setting. Micritic types such as mudstone -and wackestone accumulated in shallow, -quiet water, where currents were not strong -enough to remove the mud. Intercalated -floatstone to rudstone, commonly devel- -oped as coquinas containing abundant -complete and broken gryphaeid bivalve -shells, are interpreted to represent storm -deposits. Such storm layers with abundant -gryphaeid shells are also present in the -upper third of the formation, where they -are intercalated with fossiliferous shale -deposited under low-energy conditions. -Deposition of the upper part of the Muleros -thus was also in a shallow-marine setting, -but one with more storm deposits than in -the middle member. - -The Muleros Formation succession thus -indicates an outer-shelf environment with -dominantly shale during deposition of the -lower part, followed by some shallowing -and deposition of dominantly limestone -in an inner-shelf environment periodically -affected by storms, and again some deep- -ening resulting in increased deposition of -shale with intercalated storm layers in the -upper part. - -Mesilla Valley Formation - -Lithostratigraphy - -The Mesilla Valley Formation is a slope- -forming, shale-dominated unit approxi- -mately 65 m thick at our measured sec- -tion (Figs. 5F, 10). This is subdivision 6 of -Böse (1910), who indicated it is more than -20 m thick. Our measured section is the -type section of the Mesilla Valley Shale of -Strain (1976, table 11), who indicated it is -64 m thick. - -Strain (1976) divided the Mesilla Val- -ley Formation into two units, a lower shale -(59–60 m thick) and an upper sandstone -and shale (4–5 m thick). However, like the -underlying Muleros Formation, we divide -the Mesilla Valley Formation into three infor- -mal members (Fig. 10). The lower member -(member A) is 13.7 m thick and consists of -gray shale with some very thin sandstone -lenses. The middle member (member B) is -14.4 m thick. Its base is a 0.4-m-thick ledge- -forming limestone that contains Texigryphaea. -The limestone bed is overlain by olive shale -with thin, fine-grained sandstone ledges and -three thicker, siltstone-sandstone intervals -(each 0.4 m thick) that display ripple lami- -nation. Locally, the sandstone contains shell -debris and serpulids, and the thin sandstone -beds contain Texigryphaea (upper part). The -upper member (member C) is black shale - -forming a thick slope (36.5 m). The black -shale is sharply overlain by coarse clastic -sediments (sandstone) at the base of the -Mojado Formation (“Anapra Sandstone” of -Strain). - -Paleontology - -Böse (1910) referred to the Mesilla Valley For- -mation as the “capas con [beds with] Ostrea -quadriplicata [now Peilinia quadriplicata: Kues -1997],” and most (or all?) fossils of this taxon -appear to derive from member B of the Mesil- -la Valley Formation (Fig. 10). The Mesilla Val- -ley Formation yields a profuse assemblage of -gryphaeid bivalves (Texigryphaea) as well as -other bivalves, gastropods, ammonoids, bra- -chiopods, foraminiferans (particularly Cribra- -tina texana), dinoflagellates, echinoids, serpu- -lid worms, corals, ostracods, echinoderms, -calcareous algae, and some terrestrial plant -fragments (Böse 1910; Strain 1976; Cornell -1982; Kues 1989, 1997; Turnšek et al. 2003). -Correlation of the Mesilla Valley Formation -with the Weno and Pawpaw Formations -(Scott et al. 2003) is based more on strati- -graphic position than on precise biostratig- -raphy, although the presence of the bivalve -Peilinia quadriplicata in the Mesilla Valley For- -mation, also known from the Denton–Main -Street interval in north-central Texas (Scott -et al. 2003, p. 315), does support the Weno– -Pawpaw correlation. - -Sedimentary petrography (Figs. 10, 22) - -The lower part of the Mesilla Valley Forma- -tion (member A) is gray shale with a few -thin sandy lenses. The microfacies of the -limestone at the base of member B (Fig. 10, -unit 3) is bioclastic rudstone–packstone -(Fig. 22A–B) composed mainly of large, bro- -ken bivalve fragments. Subordinate large -agglutinated foraminiferans (Fig. 22C), -echinoderms, ostracods, and rare smaller -foraminiferans are present. The matrix is -silty and contains abundant small, angular -quartz grains. The large shell fragments are -randomly oriented parallel to the bedding -plane. - -The siltstone of unit 4 at the base of -member B (Fig. 10) is laminated, mixed -siliciclastic-carbonate and contains rare -bioclasts. The most abundant grain type -is quartz, and carbonate grains are also -present. Bioclasts include ostracods, echi- -noderms, foraminiferans, and shell debris. -Glauconite grains are common. - -In member B, the ripple-laminated silt- -stone of unit 9 (Fig. 10) is partly bioturbated -and contains small and some larger bioclasts -floating in the silty matrix. Larger bioclasts -are bivalve shell fragments, agglutinated -foraminiferans, and rare solitary corals -(Fig. 22D). Small bioclasts include ostra- -cods, gastropods, small foraminiferans, shell -debris, and rare phosphatic fragments. The -silty matrix is composed of 70–80% small -angular quartz grains, about 2% glauconite, -and a few micas in micritic matrix. - -FIGURE 13—Measured stratigraphic section of -Del Rio Formation. See Figure 2 and Appendix -for location of section, and Figure 6 for lithology -legend. - -The lower part of the Muleros Forma- -tion is mainly composed of shale that accu- -mulated on the shelf in areas deeper than -storm-induced currents. Intercalated thin -sandstone beds probably formed during -storms that resulted in the remobilization -of coarser material and their basinward - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -113 - - member C apparently occurred in a poorly -oxygenated environment, which may indi- -cate relatively deep water. - -Mojado Formation - -Lithostratigraphy - -Our measured section of the Mojado Forma- -tion (Figs. 11, 12A–D) is the type section of the -Anapra Sandstone of Strain (1976, table 12), -which he reported as approximately 53 m -thick. Here, the Mojado Formation is a sand- -stone-dominated unit as much as 58 m thick -(Figs. 5F, 11, 12A–D). It rests with sharp con- -tact on shale at the top of the Mesilla Valley -Formation (Figs. 5F, 11, 12A) and is sharply -overlain by shale and nodular limestone at -the base of the Del Rio Formation (Fig. 12D). -The topmost bed of the Mesilla Valley -Formation at our Mojado Formation sec- -tion (Fig. 11) is reddish-brown shale/silt- -stone containing abundant quartz, some -chert grains, and rare glauconite. The -matrix (or cement) is dark brown (iron -stained). Strain (1976, p. 81) stated that “the -base of the Anapra [Mojado] is the lowest -sandstone stratum 30 cm thick or more just -above the thinly bedded sandstone beds at -the top of the Mesilla Valley Shale.” Thus, -he included our basal beds of the Mojado -Formation (Fig. 10, unit 13; Fig. 11, unit 3) -in the Mesilla Valley Formation. Instead, -we draw the base of the Mojado Formation -at the base of the first traceable sandstone -bed of typical Mojado lithology. - -The Mojado Formation is entirely silici- -clastic, being composed of different types of -sandstone with intercalated gray and black -(locally lignitic) shale. A covered interval -(5.4 m) is in the middle of the succession. -Sandstone intervals are as much as 5.7 m -thick, and shale intervals are as much as -3.5 m thick. The following lithofacies are rec- -ognized: (1) thin, fine-grained sandstone beds -displaying hummocky cross-stratification -form the lowermost 4.9 m of the succession -(lower ¾ of unit 3); the sandstone beds are -interbedded with green shale; (2) iron-mot- -tled, hackly sandstone, which is bioturbated -and has Liesegang banding (units 7, 11); -(3) thin, fine-grained sandstone with ripple -lamination (units 16–22); (4) thin, medium- -grained sandstone with trough crossbedding -(units 4, 6, 8, 10, 17); and (5) thicker, multisto- -ried sandstone beds with larger-scale trough -crossbedding (units 13, 15, 25). - -Trough crossbedded sandstone inter- -vals are characterized by erosive lower con- -tacts (e.g., Fig. 12B). Several upward-fining -and thinning cycles are developed. Each cycle -starts with trough crossbedded sandstone, -grading into fine-grained, ripple-laminated -sandstone with shale interbeds, and the tops -of a few cycles are formed by shale. Hum- -mocky cross-stratification is only observed -in the lowermost sandstone beds. Neverthe- -less, marine fossils and marine bioturbation - -FIGURE 14—Measured stratigraphic section of -Buda Formation. See Figure 2 and Appendix for -location of section, and Figure 6 for lithology -legend. - -FIGURE 15—Measured stratigraphic section of -Mancos Formation (“Boquillas Formation”). See -Figure 2 and Appendix for location of section, -and Figure 6 for lithology legend. - -A thin limestone bed intercalated in -unit 10 (Fig. 10) is bioturbated bioclastic -wackestone containing a few larger and -many small bioclasts floating in micritic -matrix (Fig. 22E). Bioclasts include mol- -lusc shell debris (bivalves, gastropods), -calcareous algae, worm tubes (Fig. 22F), -ostracods, echinoderms, smaller foraminif- -erans, agglutinated foraminiferans, and -echinoid spines. The matrix is micrite and -pelmicrite. The top of member B (Fig. 10, -unit 11) is formed by a fine-grained sand- -stone interval 0.9 m thick displaying rip- -ple lamination and containing Texigryphaea -(packstone). Member C is black shale with -no obvious fossil content. - -Depositional environments - -The shale-dominated Mesilla Valley For- -mation (Fig. 10) mostly represents deposi- -tion on muddy marine sea bottoms below -wave base. The lower part of the formation -(member A, Fig. 10) is mostly shale with a - -few thin sandstone lenses, and this appears -to represent relatively deep water. The -dark shale with intercalated thin sandstone -lenses of the lower member is interpreted -to record deposition in an outer-shelf envi- -ronment, which was only rarely affected -by storms. The thin intercalated sandstones -may represent distal storm layers. Inter- -calated rudstone to packstone beds (shell -coquinas) and ripple-laminated sandstone -beds containing fossils such as Texigryphaea -represent typical storm deposits. - -Overlying shale with interbedded Texi- -gryphaea packstones and ripple-laminated -sandstones (member B) indicate a distinct -shallowing. Storm layers are more abun- -dant in member B than in the members -below and above, indicating that some -shallowing occurred during deposition of -the middle member (Fig. 10). - -The upper part of the formation (mem- -ber C) is black shale, which we interpret as -relatively offshore marine deposits. Depo- -sition of the organic-rich black shale of - -114 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 16—Thin section photographs of microfacies of the Finlay Formation -at Cerro de Cristo Rey (Fig. 6). Sample numbers correspond to unit numbers -in Figure 6. A—Bioclastic wackestone containing abundant broken fragments -of recrystallized calcareous algae, molluscs, some echinoderm fragments, -and foraminiferans in micritic matrix. Sample A24. B—Bioclastic wackestone -containing abundant fragments of dasycladacean algae, mollusc (bivalve) -fragments, and some echinoderms floating in micritic matrix. Sample A27. -C—Fine-grained bioclastic wackestone–packstone with abundant fragments -of recrystallized dasycladacean algae, molluscs, echinoderms, and a few - -foraminiferans. Matrix is gray micrite. Sample A26. D—Packstone contain- -ing abundant peloids, smaller foraminiferans including dictyoconid forms, -fragments of dasycladacean algae, and molluscs and some micritic matrix. -Sample A10. E—Packstone–grainstone containing abundant peloids, some -orbitolinoid foraminiferans (Dictyoconus), and mollusc shell debris cemented -by calcite. Sample A10. F—Grainstone composed of abundant peloids, some -orbitolinoid foraminiferans (Dictyoconus), and other bioclasts including gas- -tropods and echinoderms, cemented by calcite. Sample A10. Width of photo- -graphs is 6.3 mm. - -are present in the upper few meters of the -Mojado Formation, and we identify a clear -break (transgressive surface) approximately -3 m below its top (base of unit 26, Fig. 11). - -Strain (1976) divided the Mojado (his -“Anapra”) Formation into four units, and - -Lovejoy (1976) recognized five units. We -believe our measured section can be readily -divided into at least four intervals: (1) lower -beds with marine bioturbation and intercalat- -ed shale (Fig. 11, units 3–11), (2) middle, slope- -forming interval of mostly siltstone and shale - -(Fig. 11, units 12–14); (3) upper interval of -crossbedded sandstones (Fig. 11, units 15–25); -and (4) uppermost interval with marine fossils -and bioturbation (Fig. 11, unit 26). However, -more subdivisions are possible, though they -are of little stratigraphic significance. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -115 - - A - -C - -E - -B - -D - -F - -FIGURE 17—Thin section photographs of microfacies of the Del Norte For- -mation at Cerro de Cristo Rey (Fig. 7). Sample numbers correspond to unit -numbers in Figure 7. A—Bioclastic wackestone containing abundant algal -fragments and mollusc (bivalve) shell debris, a few foraminiferans, and -echinoderms in gray micrite. Sample B2. B—Mixed carbonate-siliciclastic -siltstone containing abundant angular quartz grains and bioclasts such as -recrystallized bivalve shell fragments (?oysters), some echinoderms, and -micrite. Sample B8. C—Wackestone–packstone composed of echinoderms, -mollusc shell fragments, and quartz grains floating in micrite. Sample B9. - -D—Mixed siliciclastic-carbonate fossiliferous siltstone containing a few -mollusc shell fragments and rare echinoderms. Sample B16. E—Bioclastic -wackestone–packstone containing abundant mollusc (bivalve) shell frag- -ments (mainly derived from bivalves), subordinate echinoderms, small -quartz grains, and a few micritic intraclasts in micritic matrix. Sample -B22. F—Fine-grained bioclastic wackestone with many mollusc (bivalve) -shell fragments and dasycladacean algae, with subordinate echinoderms -in micritic, partly peloidal matrix. Sample B19. Width of photographs is -6.3 mm, but for Fig. C is 3.2 mm. - -We propose that at Cerro de Cristo Rey -the name Anapra Formation be replaced by -Mojado Formation. To the west and north- -west of Cerro de Cristo Rey, rocks that strad- -dle the Albian–Cenomanian boundary are -assigned to the Mojado Formation of Zeller -(1965). At the type section, in the Big Hatch- - -et Mountains of Hidalgo County, New Mex- -ico, the Mojado Formation is approximately -1,500 m of sandstone, shale, and minor lime- -stone that overlies the Aptian–Albian U-Bar -Formation (Zeller 1965). In the Cooke’s -Range of Luna County, approximately 120 -km northwest of Cerro de Cristo Rey, the - -Mojado Formation is divided into three -members (Lucas and Estep 1998b): (1) Fry- -ingpan Spring Member, approximately 22 m -of sandstone, siltstone, shale, and nodular -limestone that includes fossils of the ammo- -noid Mortoniceras equidistans (Cragin) (Cobban -1987; Lucas et al. 1988); (2) Sarten Member, - -116 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 18—Thin section photographs of microfacies of the Smeltertown -Formation at Cerro de Cristo Rey (Fig. 8). Sample numbers correspond -to unit numbers in Figure 8. A—Mixed fossiliferous carbonate-siliciclas- -tic siltstone containing some large mollusc (bivalve) fragments, a few -ostracods and foraminiferans, and as much as 30% detrital quartz grains. -Sample E3. B—Fine-grained, mixed siliciclastic-carbonate sandstone -composed of detrital quartz grains, a few glauconite grains, and a few - -bioclasts (bivalves, gastropods, echinoderms, foraminiferans) floating in -micrite. Sample E7. C—Mudstone containing some quartz grains and few -bioclasts. Sample E8. D–F—Bioclastic wackestone containing agglutinat- -ed foraminiferans (Cribratina) and bivalve fragments, and, subordinately, -other bioclasts such as ostracods embedded in silty, partly peloidal matrix -containing a few small quartz grains. Sample E9. Width of photograph C -is 3.2 mm, and of all other photographs is 6.3 mm. - -approximately 58 m of sandstone and silt- -stone of nonmarine origin; and (3) Rattlesnake -Ridge Member, approximately 6 m of silty -sandstone, siltstone, and nodular limestone -that yields a bivalve fauna of earliest Ceno- -manian age (Lucas et al. 1988). Sandstone and -siltstone of the Sarten Member are lithologi- -cally very similar to the “Anapra Sandstone,” -and the biostratigraphy supports a reasonably - -close correlation (Fryingpan Spring Member -correlates to the Smeltertown Formation, and -Rattlesnake Ridge Member correlates to the -Del Rio Formation; Lucas et al. 1988). Thus, -the “Anapra Sandstone” is seen by us as a -tongue of the Bisbee Group (Mojado Forma- -tion) present in the Cretaceous section of Cerro -de Cristo Rey. We therefore abandon the local -name Anapra (of Strain 1968, 1976) in favor of - -the older name of a more regional stratigraph- -ic unit (Mojado of Zeller 1965). - -Paleontology - -The Mojado Formation (“Anapra Sand- -stone”) at Cerro de Cristo Rey contains -dinosaur footprints and plant debris in its - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -117 - - A - -C - -E - -B - -D - -F - -FIGURE 19—Thin section photographs of microfacies of the Smeltertown -Formation (A–D) and Muleros Formation (E–F) at Cerro de Cristo Rey -(Figs. 8–9). Sample numbers correspond to unit numbers in Figures 8 and 9. -A—Mixed siliciclastic-carbonate siltstone containing a few glauconite -grains and rare bioclasts such as ostracods and echinoderms. Sample E9. -B—Quartzose siltstone with a few carbonate grains, rare glauconite, and -few large bivalve fragments. The siltstone is cemented by quartz (authi- -genic overgrowths) and rare calcite cement. Sample E10. C–D—Mixed -siliciclastic-carbonate siltstone to fine-grained sandstone containing some - -bivalve shell fragments and echinoderms. Sample F1. E—Fine-grained -bioclastic wackestone–packstone overlying gray mudstone. Wackestone– -packstone contains bivalve shell fragments, echinoderms, ostracods, a few -smaller foraminiferans, peloids, and a few small quartz grains. Sample F2. -F—Coarse-grained bioclastic wackestone, partly floatstone, containing -abundant bivalve shell fragments (Texigryphaea), subordinate gastropods, -echinoderms, foraminiferans, few peloids, and quartz grains in micritic -matrix. Sample F3. Width of photograph D is 3.2 mm, and of all other pho- -tographs is 6.3 mm. - -lower to middle portions, as well as some -marine fossil fragments and tests of the large -arenaceous foraminiferan Cribratina texana. -The dinosaur tracks are primarily those of -ornithopods and theropods (Fig. 12C; e.g., -Kappus et al. 2003; Kappus 2007) and pro- -vide strong evidence of subaerial, terrestrial - -conditions during deposition of part of the -Mojado Formation, but are not of precise -biostratigraphic significance. The upper -3 m of the Mojado (unit 26) contain fossils -of the large exogyrid bivalve Gyrostrea whit- -neyi (known from the Del Rio Formation -in north-central Texas) as well as echinoids - -and agglutinated foraminiferans (Böse 1910; -Strain 1976; Turnšek et al. 2003). We also -observed dasycladacean algae, foraminifer- -ans, echinoderms, and ostracods in thin sec- -tion (Fig. 23E). - -Strain (1976) placed the base of the -Cenomanian at the base of the “Anapra - -118 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 20—Thin section photographs of microfacies of the Muleros Forma- -tion at Cerro de Cristo Rey (Fig. 9). Sample numbers correspond to unit num- -bers in Figure 9. A–B—Bioclastic wackestone to floatstone containing abundant -bivalve shell fragments (Texigryphaea), subordinate gastropods, echinoderms, -and foraminiferans, including large agglutinated forms in micritic matrix. Sam- -ple F3. C—Bioturbated siltstone–mudstone containing a few bioclasts (bivalve -shell fragments, echinoderms, ostracods) that float in matrix. Sample F3. D— - -Fine-grained wackestone–packstone with a few large bivalve fragments and -many small bioclasts, peloids, and some detrital angular quartz grains in mic- -ritic matrix. Sample D5. E—Mudstone containing quartz grains and smaller -textulariid foraminiferans floating in micrite. Sample F5. F—Fine-grained -wackestone–packstone containing abundant peloids, a few detrital quartz -grains, shell debris, echinoderms, smaller foraminiferans, and ostracods. Sam- -ple F6. Width of photographs A–D is 6.3 mm, and of E–F is 3.2 mm. - -Sandstone,” and Lucas and Estep (1998b, -2000) placed it at the base of the Del Rio -Formation. Given that in the regional -Washita Group section, the base of the -Cenomanian is a sequence boundary at -the base of the WA6 cycle of deposition, -the Albian–Cenomanian boundary is -most likely at the marine transgressive -surface that is approximately 3 m below - -the Mojado–Del Rio contact (Turnšek -et al. 2003; Scott et al. 2003) (Fig. 11). - -Sedimentary petrography (Figs. 11, 23C–F) - -Sandstone of the Mojado Formation is -fine- (0.1–0.3 mm) to medium-grained (0.2– -0.5 mm) quartz arenite, nonlaminated, and -well sorted (Fig. 23C–D). Grains are mostly - -subrounded to rounded, although many -grains display authigenic overgrowths -that are not always easy to recognize. The -most common grain type is monocrystal- -line quartz. Subordinate chert grains are -present, perhaps of volcanic origin. Poly- -crystalline quartz of metamorphic origin -including stretched metamorphic grains -is rare. Detrital feldspars are very rare - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -119 - - A - -C - -E - -B - -D - -F - -FIGURE 21—Thin section photographs of microfacies of the Muleros Forma- -tion at Cerro de Cristo Rey (Fig. 9). Sample numbers correspond to unit num- -bers in Figure 9. A—Floatstone containing gryphaeid shell fragments, some -encrusted by balanids, and agglutinated foraminiferans embedded in micritic, -partly peloidal matrix. Sample F7. B—Wackestone–rudstone with abundant -gryphaeid shell fragments, some gastropods, agglutinated foraminiferans, and - -other bioclasts floating in micritic matrix. Sample F7. C—Fine-grained biotur- -bated wackestone composed of abundant peloidal and small bioclasts (ostraco- -ds, echinoderms, mollusc shell debris, foraminiferans) and a few quartz grains. -Sample F8. D–F—Floatstone–rudstone with bivalve fragments, agglutinated -foraminiferans, echinoderms, and corals embedded in micritic matrix. D–E = -sample F10, F = sample F14. Width of photographs is 6.3 mm. - -and generally altered. Accessory grains -are micas (muscovite), tourmaline, zircon, -apatite, and opaque grains. The sandstone -is cemented by authigenic quartz over- -growths, and locally also by microcrystal- -line quartz and opaque cement (Fe-hydrox- -ides). Unit 10 (Fig. 11) includes a lens of -strongly recrystallized stromatolite com- -posed of irregularly laminated algal mats. - -Unit 23 (Fig. 11) is a crossbedded, fine- -grained, bioturbated bioclastic wacke- -stone, locally a packstone, containing a few -larger bioclasts (Fig. 23E). Most abundant -are dasycladacean algae and mollusc shell -fragments (bivalves, gastropods). Less -common are foraminiferans, echinoderms, -echinoid spines, ostracods, and other frag- -ments. The matrix is micrite to pelmicrite. - -The uppermost Mojado Formation (Fig. 11, -units 25–26) is composed of laminated silt- -stone to fine-grained sandstone. Grain size -is mostly < 0.1 mm; however, a few quartz -grains as much as 0.5 mm are present. The -most abundant grain type is quartz; carbon- -ate grains are subordinate, and opaques and -glauconite are rare. The siltstone to fine- -grained sandstone contains some ostracods, - -120 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 22—Thin section photographs of microfacies of the Mesilla Valley -Formation at Cerro de Cristo Rey (Fig. 10). Sample numbers correspond to -unit numbers in Figure 10. A–B—Bioclastic packstone–rudstone contain- -ing densely packed bivalve fragments, subordinate echinoderms and other -bioclasts, a few detrital quartz grains, and rare glauconite. Sample G1. C— -Bioclastic packstone–rudstone composed of abundant recrystallized shell -fragments and agglutinated foraminiferans in silty matrix containing small - -angular quartz grains. Sample G3. D—Bioclastic siltstone composed of -small and larger fossils such as corals, bivalves, and agglutinated foraminif- -erans floating in silty matrix. Sample G4. E—Bioclastic wackestone con- -taining bivalve shell fragments, echinoderms, ostracods, and worm tubes -embedded in micritic to pelmicritic matrix. Sample G10. F—Worm tubes -in a bioclastic wackestone. Sample G10. Width of photographs is 6.3 mm, -except for F, which is 3.2 mm. - -shell fragments, rare foraminiferans, and -echinoderms. Shell fragments are oriented -parallel to the bedding in unit 26. - -Depositional environments - -The lower part of the Mojado Formation -(Fig. 11) represents an upward-coarsening - -and shallowing succession, starting with -hummocky crossbedded sandstone of the -lower shoreface followed by upper shoreface -deposits represented by alternating, horizon- -tally laminated, ripple-laminated, and trough -crossbedded sandstone. A shallow-marine -environment is also indicated by a thin stro- -matolite lens within unit 10 (Fig. 11). - -The overlying trough crossbedded intervals -(Fig. 11, units 13 and 15) probably represent -fluvial channel-fill deposits of a delta plain, -which are again overlain by shallow-marine -siliciclastic sediments (units 16–22). The thin, -intercalated sandstone beds (ripple laminated, -horizontally laminated, trough crossbedded) -indicate high-energy conditions and may - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -121 - - A - -C - -E - -B - -D - -F - -FIGURE 23—Thin section photographs of microfacies of the Mancos Forma- -tion (A–B) and Mojado Formation (C–F) at Cerro de Cristo Rey (Figs. 11, 15). -Sample numbers correspond to unit numbers in Figures 11 and 15. A—Bio- -clastic wackestone with bivalve shell fragments, calcareous algae, and echi- -noderms floating in micrite. Sample I1. B—Bioclastic wackestone–packstone -with mostly bivalve fragments, echinoderms, and subordinate other bio- -clasts, micritic matrix, and some calcite cement. Sample I3. C—Fine-grained -quartz arenite composed mostly of monocrystalline quartz, rare polycrystalline - -quartz, cemented by quartz as authigenic overgrowths. Sample H3. D—Quartz -arenite showing abundant monocrystalline quartz grains with well-displayed -authigenic quartz overgrowths. Sample H3. E—Bioclastic wackestone contain- -ing abundant fragments of dasycladacean algae and bivalves, subordinate -foraminiferans, and echinoderms floating in micritic to pelmicritic matrix. -Sample H23. F—Bioturbated peloidal mudstone containing some quartz -grains and a few bioclasts. Sample H27. Width of photographs is 6.3 mm, -except for D, which is is 3.2 mm. - -represent storm deposits. A shallow-marine -environment is also indicated by a thin lens of -fossiliferous limestone (bioclastic wackestone; -Fig. 11, unit 23) that contains abundant dasy- -cladacean algae. - -The trough crossbedded sandstone of -the succeeding 7 m may represent fluvial -or upper shoreface deposits. The overlying, -horizontally laminated sandstone (Fig. 11, - -unit 26) is of shallow-marine origin, indi- -cated by the marine fossils. - -Del Rio Formation - -Lithostratigraphy - -The Del Rio Formation at our measured section -is approximately 33 m thick (Fig. 13). Lovejoy - -(1976) reported its thickness as 24–27 m. In our -measured section (Fig. 13), the lower 9.7 m -(most of unit 2) is a nodular limestone inter- -val composed of abundant limestone nodules -in a yellow clay slope (Fig. 12D). This inter- -val contains many fossils of Gyrostrea (=Exo- -gyra) whitneyi and gastropods. Above follows -a shale slope with some nodular limestone -and three thin limestone beds, each 0.2–0.3 m - -122 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - A - -C - -E - -B - -D - -F - -FIGURE 24—Thin section photographs of microfacies of the Del Rio Formation -(A–C) and Buda Formation (D–F) at Cerro de Cristo Rey (Figs. 13–14). Sample -numbers correspond to unit numbers in Figures 13–14. A—Peloidal wacke- -stone with some angular quartz grains and bioclasts such as ostracods, smaller -foraminiferans, and echinoid spines. Sample C3. B—A few bioclasts (bivalve -shell fragments, echinoderms, ostracods, foraminiferans) floating in siltstone -composed of angular quartz and carbonate grains, rare glauconite. Sample C7. -C—Bioclastic mudstone–wackestone containing many small and a few larger - -fossil fragments (bivalves, echinoderms) floating in micritic matrix. Sample C9. -D—Peloidal bioclastic mudstone containing a few angular quartz grains, some -bioclasts (ostracods, shell fragments, foraminiferans, echinoderms), and rare -glauconite. Sample D5. E—Bioclastic mudstone composed of gray micrite with -small and a few larger fossil fragments. Sample D7. F—Fine-grained bioclastic -wackestone containing mostly bivalve fragments, subordinately echinoderms, -ostracods, and foraminiferans floating in gray micrite. Sample D5. Width of -photographs A and D is 3.2 mm, and of photographs B, C, E, F is 6.3 mm. - -thick and forming ledges (units 3–7). Above -the uppermost limestone ledge, yellowish -shale with some nodular limestone is exposed -with a thickness of 12.8 m (unit 8). Gyrostrea -whitneyi is also common in the upper part. -Turnšek et al. (2003, p. 154) stated that Ilyma- -togyra arietina is the “dominant fossil” in the -Del Rio, but our field observations suggest it -is Gyrostrea whitneyi. The top of the Del Rio - -Formation is composed of nodular limestone -with thin yellow shale interbeds (unit 9, 4.5 m -thick), overlain by white limestone of the Buda -Formation (Fig. 13). - -Paleontology - -The Del Rio Formation at Cerro de Cristo Rey -yields bivalves (especially Gyrostrea whitneyi - -and Ilymatogyra arietina), gastropods, spatan- -goids, and ammonoids. In thin section we -observed foraminiferans, ostracods, echino- -derms, and annelid worm tubes. From the Del -Rio Formation at Cerro de Cristo Rey, Maul- -din (1985) and Mauldin and Cornell (1986) -determined 52 species of foraminiferans. The -foraminiferal assemblage is dominated by -rotaliine foraminiferans (38 species), which - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -123 - - B - -C - -H - -M - -I - -N - -R - -S - -T - -U - -D - -F - -J - -O - -E - -G - -K - -P - -V - -A - -L - -FIGURE 25—Some Cretaceous microfossils of the Cerro de Cristo Rey -sections. A, H—Permocalculus sp. (Finlay Formation, A unit 14, H unit 27, -Fig. 6). B, C—Boulina sp. (Finlay Formation, B unit 12, C unit 20, Fig. 6). -D, E, G, K—Acicularia or Terquemella sp. (Finlay Formation, D unit 14, E -unit 27, K unit 26, Fig. 6; Del Norte Formation, G unit 9, Fig. 7). F—Neom- -eris sp. (Finlay Formation, unit 20, Fig. 6). I, J—Salpingoporella sp. (Finlay -Formation, unit 24, Fig. 6). L—Pseudocyclammina sp. (Finlay Formation, unit - -27, Fig. 6). M, N—Textulariid foraminiferans (M—Del Norte Formation, unit 14, -Fig. 7; N—Smeltertown Formation, unit 7). O—Pseudonummuloculina heimi (Fin- -lay Formation, unit 10, Fig. 6). P—Lenticulina sp. (Del Rio Formation, unit 3, -Fig. 13). R, S—Orbitolinoid foraminiferans (Dictyoconus) (Finlay Formation, -unit 8, Fig. 6). T, U, V—Agglutinated foraminiferans (Ammobaculites? and/or -Cribratina) (T, U—Smeltertown Formation, unit 9, Fig. 8; V—Buda Formation, -unit 7, Fig. 14). Scale bar is 0.3 mm, except for K, which is 0.1 mm. - -constitute 82% of the individuals. Aggluti- -nated foraminiferans (suborder Textulariina) -comprise 12 species and 17% of all individu- -als. Two species belong to milioline foraminif- -erans, constituting 1% of the individuals. -These foraminiferans support correlation of -the Del Rio outcrops with the Grayson For- -mation elsewhere in Texas. - -Sedimentary petrography (Figs. 13, 24A–C) - -At the top of our section of the Mojado For- -mation, the basal interval of the overlying Del - -Rio Formation (Fig. 11, unit 27) is composed -of interbedded nodular limestone and shale. -The limestone beds are bioturbated peloidal -mudstone containing many peloids, a few -quartz grains, rare glauconite, and a few -small bioclasts such as ostracods, echino- -derms, and rare foraminiferans (Fig. 23F). -Similarly, in our section of the Del Rio -Formation (Fig. 13), the lowermost inter- -val (unit 2) is nodular limestone and shale -capped by a thin limestone bed (unit 3) that -consists of peloidal wackestone (Fig. 24A), -grading up into fine-grained sandstone - -containing abundant quartz. The peloidal -wackestone (unit 3) is indistinctly laminated -and contains as much as about 20% quartz -(with a grain size as much as approximately -0.2 mm) and rare glauconite. Most abundant -are small peloids and many small bioclasts -such as ostracods, foraminiferans, echinoid -spines, and some other small fossils, float- -ing in microspar (cement?). The amount of -quartz grains increases upward to about 30%, -and the peloidal wackestone is overlain by -fine-grained sandstone that contains about -50–60% angular quartz grains, rare glauconite, - -124 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - and a few bioclasts (including ostracods, echi- -noid spines, echinoderms, shell fragments, -and rare foraminiferans) in brownish micritic -matrix. - -The limestone bed of unit 5 of the Del Rio -Formation (Fig. 13) is a peloidal wackestone– -grainstone with abundant small peloids and -foraminiferans. Subordinate are ostracods, -echinoderms, echinoid spines, small quartz -grains (5–10%), a few micritic intraclasts, -and rare glauconite. The rock is indistinctly -laminated and calcite cemented. - -Unit 7 (Fig. 13) is a bioclastic limestone -with some small and larger bioclasts float- -ing in silty matrix (Fig. 24B). The silty -matrix is composed of predominantly -small angular quartz and carbonate grains, -subordinate glauconite, and some opaque -grains. Bioclasts include echinoderms, -ostracods, shell fragments (bivalves), and -rare agglutinated and other foraminiferans. -The rock is locally bioturbated and calcite -cemented. - -The nodular limestone of the Del Rio -Formation is a gray, nonlaminated, biotur- -bated bioclastic mudstone. The micritic to -pelmicritic matrix contains many small and -a few larger bioclasts. A few small quartz -grains and rare glauconite grains are pre- -sent. Bioclasts include ostracods, bivalve -shells, echinoderms, a few foraminiferans, -rare gastropods, and annelid worm tubes. -The uppermost limestone bed of the Del -Rio Formation (unit 9) is composed of bio- -turbated bioclastic mudstone (Fig. 24C). -Small bioclasts of ostracods, echinoderms, -foraminiferans, gastropods, bivalves, rare -quartz grains, and very rare glauconite -float in a micritic matrix. - -Depositional environments - -Del Rio Formation facies (Fig. 13) are -similar to those of the lower part of the -Del Norte Formation (Fig. 7) and indicate -shallow-marine deposition. Mauldin and -Cornell (1986) concluded that the forami- -niferans of the Del Rio Formation indicate a -deepening of marine waters (transgression) -that culminated in the middle part of the -Del Rio, followed by a regression during -deposition of the upper Del Rio. - -Mauldin and Cornell (1986) indicate that -the lower half of the Del Rio succession -is dominated by rotaliine foraminiferans; -the upper half is enriched in agglutinated -foraminiferans. The ratio of planktonic -to benthic foraminiferans is 0.1–0.6 in the -lower half, indicating water depths from -60 to 100 m. The value drops in the upper -half, indicating water depths of 20 m or less -(Mauldin and Cornell 1986). The foraminif- -eral assemblages indicate that the basal part -of the Del Rio Formation marks a transgres- -sion, and that a deeper shelf environment -(60–100 m) persisted until deposition of the -middle Del Rio Formation, followed by a -drop in sea level to water depths of approx- -imately 20 m in the upper half (Mauldin -and Cornell 1986). - -However, shale with limestone nodules -and nodular to wavy limestone indicate -deposition under quiet-water conditions in a -deeper, outer-shelf environment. This is also -indicated by the presence of glauconite, which -in modern oceans occurs at depths between -50 and 500 m (Flügel 2004). Even the wavy -to nodular limestone in the uppermost part -of the Del Rio Formation formed in a low- -energy environment. Lithologic evidence of a -drop in sea level to water depths of approxi- -mately 20 m in the upper half suggested by -Mauldin and Cornell (1986) is not observed. -Even the uppermost wavy to nodular lime- -stone (bioclastic mudstone) with intercalated -shale was deposited in a low-energy environ- -ment below wave base. High-energy deposits -such as grainstone or packstone are absent. - -Buda Formation - -Lithostratigraphy - -At Cerro de Cristo Rey, the best exposed and -most complete section of the Buda Forma- -tion is in Bowen Gulch where it is a promi- -nent, resistant interval of white limestone -that forms ridges and cuestas (Figs. 12E, -14). Here, the Buda Formation is 15.5 m -thick (Lovejoy 1976 reported its thickness as -approximately 12 m) and composed of alter- -nating limestone and shale with limestone -nodules. The lowermost limestone of the -Buda Formation is 0.8 m thick (unit 5) and -nodular, overlain by a thin nodular lime- -stone with shale and a prominent limestone -cliff 2.7 m thick (units 6–7). Above the lime- -stone cliff follows a succession of alternating -limestone and shale (units 8–21). Limestone -beds are 0.3–0.6 m thick and full of gastro- -pods. Shale intervals measure 0.3–2.5 m and -contain limestone nodules, which are locally -fossiliferous. - -The lithology of the Buda Formation sec- -tion exposed at Cerro de Cristo Rey most -resembles that of the middle member (Red -Light Member) of the Buda in Trans-Pecos -Texas (Reaser and Robinson 2003) in being -a mixture of thin-bedded lime mudstone -and nodular, marly limestone. In contrast, -the lower member (Lecheguilla Member) -and the upper member (Love Station Mem- -ber) of the Buda Formation in Trans-Pecos -Texas are ridge-forming beds of lithograph- -ic, muddy limestone (Reaser and Robinson -2003). Correlation of the Buda Formation -section at Cerro de Cristo Rey to only the -Red Light Member to the southeast thus -has a lithostratigraphic basis, but cannot -be confirmed biostratigraphically because -the entire Buda Formation falls within one -ammonoid zone, which is the biostrati- -graphic means by which Buda sections are -correlated (Young 1979). - -As Hook (2008; also see Cooper et al. 2008) -noted, the top of the Buda Formation usu- -ally displays evidence of erosion in the form -of microkarst features at the Buda–Mancos -contact. However, at Cerro de Cristo Rey the - -Buda–Mancos contact is poorly exposed and -tectonized (Fig. 12F), so evidence of micro- -karst features is not clear. The best evidence -of the Buda–Mancos unconformity at Cerro -de Cristo Rey is provided by ammonoid bio- -stratigraphy (Cobban et al. 2008). The base of -the Mancos Formation is in the upper middle -Cenomanian Acanthoceras amphibolum Zone, -whereas the upper Buda is in the lower Ceno- -manian Neophlycticeras (=Budaiceras) hyatti -Zone—thus, two lower Cenomanian and -four middle Cenomanian ammonoid zones -are missing at the Buda–Mancos contact, an -unconformity of more than one million years -duration (Cobban et al. 2008, fig. 2). - -Paleontology - -The Buda Formation yields many gastro- -pods (mostly turritellids) and bivalves, as -well as some ammonoids, dinoflagellates, -serpulids, spatangoid echinoderms, crus- -taceans, corals, and fish teeth (Böse 1910; -Young 1979; Cornell 1997; Turnšek et al. -2003). Ammonoids we have collected (and -will document elsewhere) place it in the -early Cenomanian zone of Neophlycticeras -(=Budaiceras) hyatti (Young 1979). - -Sedimentary petrography (Figs. 14, 24D–F) - -The basal limestone of the Buda Forma- -tion (Fig. 14, unit 5) is composed of fine- -grained, bioturbated bioclastic wackestone -containing a few larger skeletons (Fig. 24F). -Most abundant are mollusc shell frag- -ments. Less abundant are echinoderms, -ostracods, echinoid spines, foraminiferans, -and other recrystallized, indeterminate -skeletons. The skeletons float in a micritic -to pelmicritic matrix. This type grades into -bioclastic mudstone containing small and a -few larger bioclasts that float in gray mic- -rite (Fig. 24E). - -The thick limestone cliff of the Buda -Formation (Fig. 14, unit 7) consists of fine- -grained, bioturbated, bioclastic wackestone -with a few larger skeletons. Most abundant -are mollusc shell fragments (bivalves and -gastropods) and dasycladacean algae; -ostracods, echinoderms, echinoid spines, -smaller foraminiferans, and rare bryozo- -ans(?) are subordinate. Many skeletons, -particularly molluscs and algae, are recrys- -tallized. A few micritic intraclasts are pre- -sent. The skeletons are embedded in a mic- -ritic to pelmicritic matrix. - -The lowermost, gastropod-rich lime- -stone bed (Fig. 14, unit 9) is a bioturbated -bioclastic mudstone containing small and -rare larger bioclasts floating in a pelmic- -ritic matrix (Fig. 24D). Most skeletons are -derived from bivalves and gastropods, -and also from dasycladacean algae. Rare -ostracods, echinoderms, echinoid spines, -and smaller foraminiferans are present. -The rock also contains a few small, detrital -quartz grains. Another thin section indi- -cates a bioturbated bioclastic mudstone -composed of micritic matrix and small - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -125 - - bioclasts such as ostracods, bivalve shell -fragments, echinoid spines, echinoderms, -and smaller foraminiferans floating in the -matrix. Small quartz grains and rare glau- -conite are present. - -The uppermost limestone bed of the Buda -Formation (Fig. 14, unit 21) is composed of -bioturbated bioclastic wackestone contain- -ing a few larger bivalve and shell fragments -floating in a fine-grained bioclastic matrix. -Small recrystallized shell fragments and -algal debris, ostracods, subordinate small -gastropods, echinoderms, and smaller -foraminiferans are abundant. The fossils are -embedded in micritic matrix. Similarly, at -our Mancos Formation section, the topmost -Buda Formation (Fig. 15, unit 1) is bioturbat- -ed bioclastic wackestone that is poorly sort- -ed, contains mollusc shell debris (bivalves, -gastropods), calcareous algae, echinoderms, -echinoid spines, ostracods, worm tubes, -and foraminiferans. A few peloids and rare -glauconite grains are present. The matrix is -micrite. - -Depositional environments - -Regional deposition of the Buda Forma- -tion took place in shallow-marine waters -on a broad, shallow carbonate ramp that -extended across much of Texas and parts -of northern Mexico (Reaser and Robinson -2003). Our analysis of Buda microfacies -(most limestones are bioclastic wackestone -and mudstone) at Cerro de Cristo Rey indi- -cates they correspond well to the shallow- -water wackestone–mudstone facies of -Reaser and Robinson (2003). - -According to Reaser and Robinson (2003), -the lower member of the Buda Formation was -deposited on a broad open shelf with normal -marine water near the basin margin during -the last marine incursion by the Coman- -chean sea. The middle member is interpreted -as deposits of a shallow, slightly hypersaline -shelf during a drop in sea level. Sediments of -the upper member accumulated in a similar -environment (on the distal part of an open -shelf) as those of the lower member and rep- -resent the final highstand event. - -At Cerro de Cristo Rey (Fig. 14) muddy -textures, common bioturbation, diverse fos- -sil assemblage, and the presence of dasyclad- -acean algae indicate to us that the limestone -of the lower part of the Buda Formation was -deposited in an open, normal marine, low- -energy shelf environment below the wave -base. Carbonate sedimentation periodically -was interrupted by deposition of thin shale -intercalations. - -Intercalated limestone beds of the upper -part of the Buda are of the same composi- -tion as those of the lower part, indicating -a similar depositional environment. Shale -is more abundant in the upper part, indi- -cating high rates of fine siliciclastic influx -that was periodically interrupted, causing -the deposition of limestone beds. A great- -er amount of shale may be the result of a -slight drop in sea level, although deposi- -tion still occurred below wave base. - -Mancos Formation -("Boquillas Formation") - -Lithostratigraphy - -At Cerro de Cristo Rey, the exposed thick- -ness of the Mancos Formation is 18.6 m -(Fig. 15). The total estimated thickness is -approximately 110 m (Böse 1910; Lovejoy -1976), but we cannot confirm this estimate -with outcrop data. White, gastropod-rich, -nodular limestone of the Buda Formation -is sharply and unconformably overlain by -approximately 5 m of shale intercalated -with thin sandy limestone and limestone -beds. The uppermost limestone bed of this -interval is 0.2 m thick and forms a ledge. -Above is a 3.9-m-thick covered interval that -conceals a cross fault or fold. Overlying -strata are a 10-m-thick succession of shale -with a few thin sandstone beds. - -Böse (1910) and some subsequent work- -ers (Fig. 3) referred to these strata as the -Eagle Ford Formation (Shale). Strain (1976) -first brought the term Boquillas Formation -into Cerro de Cristo Rey, although he sug- -gested that the names Chispa Summit or -Ojinaga might better be applied. We prefer -to use the name Mancos Formation (Shale), -widely used across New Mexico, for this -shaley interval at the base of the Green- -horn cycle of deposition (e.g., Seager 1981; -Molenaar 1983; Lucas and Estep 1998a; -Lucas et al. 2000). Indeed, Mancos For- -mation is applied to the same lithosome -at nearby outcrops in the southern San -Andres Mountains, approximately 75 km -north of Cerro de Cristo Rey, and in the -southern Cooke’s Range, approximately -120 km northwest of Cerro de Cristo Rey. - -Paleontology - -Kennedy et al. (1988) described a middle -Cenomanian molluscan fauna of bivalves -(Ostrea beloiti Logan and Inoceramus arva- -nus Stephenson) and diverse ammonoids -(including Acanthoceras amphibolum) from -a thin bed of calcarenitic and coquinoidal -limestone near the base of the Mancos For- -mation (Fig. 15, bed 3). Turnšek et al. (2003) -also state that shark teeth (notably Pty- -chodus) are present in the Mancos Forma- -tion at Cerro de Cristo Rey. Cornell (1997) -reported dinoflagellate cysts. We observed -dasycladacean algae, foraminiferans, ostra- -cods, echinoderms, and bryozoans(?) in -thin section. - -Sedimentary petrography (Figs. 15, 23A–B) - -The lowermost thin limestone bed of the -Mancos Formation (Fig. 15, unit 3) is a -bioclastic wackestone–packstone–rudstone -that is poorly sorted and indistinctly lami- -nated (Fig. 23A–B). The rock contains a mic- -ritic matrix, and locally some calcite cement -is present. It is strongly recrystallized, and -it contains a few centimeter-sized bivalve - -shell fragments, abundant small bioclasts, -mostly shell debris and foraminiferans -(globigerinids?), subordinate echinoderms, -and other bioclasts. A few angular quartz -grains, ooids, and intraclasts are present. - -Depositional environments - -Deposition of the Mancos Formation took -place in normal marine waters. Muddy sea -bottoms supported characteristic benthic -organisms such as inoceramid bivalves, -and nektonic organisms such as ammo- -noids lived in the water column. - -Depositional cycles and -sequence boundaries - -Introduction - -Deposition of the Cretaceous strata exposed -at Cerro de Cristo Rey took place just off -of the northern margin of the Chihuahua -trough, one of the actively subsiding basins -bordering the late Aptian–early Cenoma- -nian Comanche Shelf, which extended from -eastern Arizona across part of southern -New Mexico and northern Mexico through -Texas and farther east (Fig. 26). The Creta- -ceous section at Cerro de Cristo Rey encom- -passes the uppermost Fredricksburg Group -(upper part of Finlay Formation), the entire -Washita Group (Del Norte, Smeltertown, -Muleros, Mesilla Valley, Mojado, Del Rio, -and Buda Formations), and the basal unit -(lower Mancos Formation) of the overlying -Greenhorn cycle. It represents a hierarchy of -at least three clearly definable depositional -cycles (Scott et al. 2003). Here, we identify -sequence boundaries (unconformities) and -transgressive-regressive (deepening-shal- -lowing) cycles in the Cretaceous section at -Cerro de Cristo Rey (Fig. 27). - -First order cycles and sequence boundaries - -Scott et al. (2000, 2003) consider the entire -Comanchean Series of Texas and adjacent -areas to represent a single, first-order cycle -of deposition. This cycle includes the Trin- -ity, Frederickburg, and Washita Groups of -late Aptian to early Cenomanian age. The -lower part of this cycle (Trinity Group and -most of the Fredericksburg Group) is not -exposed at Cerro de Cristo Rey. However, -the upper part of the cycle, and its upper -bounding unconformity (between the Buda -and Mancos Formations, see above), is well -exposed at Cerro de Cristo Rey. - -In northern Texas, the entire Comanchean -section is approximately 250 m thick, whereas -at Cerro de Cristo Rey that part of the Coman- -chean section exposed is thicker, approxi- -mately 350 m thick. This probably reflects -greater clastic influx into (and greater subsid- -ence of) the Cerro de Cristo Rey Comanchean -section because of its more landward position -relative to the more offshore sections in north- -ern and central Texas (Fig. 26). - -126 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - Formation at Cerro de Cristo Rey, whereas -the upper member indicates shallowing. - -A strong candidate for the next uncon- -formity (sequence boundary) is the base of -the Smeltertown Formation (Fig. 7, unit 24). -The sharp contact of shale on limestone at -the base of the Smeltertown Formation indi- -cates deepening and marks the next trans- -gressive surface. The absence of ammonites -indicative of the Craginites serratescens and -Eopachydiscus marcianus Zones at Cerro de -Cristo Rey may be indicative of a hiatus -at the base of the Smeltertown Formation -where strata of the Mortoniceras equidistans -Zone rest directly on strata of the Adkinsites -bravoensis Zone, or it may be that facies -restrictions precluded ammonite presence/ -preservation in part of the Del Norte–Smel- -tertown interval. The shoaling trend from -deeper to shallower shelf of cycle WA2, cor- -responding to the Smeltertown Formation, -is also observed at Cerro de Cristo Rey. - -Cycle WA3 in north Texas was deposited -during a period of general shoreline retreat -along the northern basin margin (Scott et al. -2003). An upward-deepening trend is also -observed within most of the Muleros Forma- -tion with a regressive trend in the uppermost -part. Cycle WA3 (Fig. 27) thus encompasses -the entire Muleros Formation, and begins -with an upward-deepening trend into off- -shore shale (Fig. 9, unit 2) and then shallow- -ing into the bioturbated limestones and Texi- -gryphaea packstones of the middle to upper -Muleros Formation (Fig. 9, units 3–14). WA3 -is an important regional cycle in New Mexi- -co as it is represented by the Tucumcari and -Glencairn Formations, which are the first -extensive Albian seaway in the northeastern -part of the state (e.g., Kues and Lucas 1987; -Scott et al. 2001). - -An abrupt deepening of marine waters -at the base of the Mesilla Valley Formation -marks the beginning of the next cycle (WA4, -Fig. 27). This cycle begins relatively deep in -offshore marine muds of member A (Fig. 10, -unit 2) and then shallows through sandstone -and Texigryphaea packstones of member B. In -north Texas, cycle WA4, which corresponds -to most of the Mesilla Valley Formation at -Cerro de Cristo Rey (Fig. 27), is a succession -from shallow shelf within the storm wave -base to a nearshore high-energy setting. A -shallowing trend is also recorded within the -Mesilla Valley Formation from member A to -member B. The sharp contact between lime- -stone (Muleros) and shale (Mesilla Valley) is -the likely sequence boundary at the base of -the WA4 cycle. - -Relatively offshore black shale of mem- -ber C of the Mesilla Valley Formation (Fig. 10, -unit 12) appears to represent an abrupt deep- -ening and thus can be interpreted as the base -of the next cycle, WA5. The succession shal- -lows upward into interbedded sandstone -and shale of the basal Mojado Formation. -We thus place the base of WA5 at the base of -member C of the Mesilla Valley Formation -(Fig. 27). A deepening trend above the first -thick, crossbedded fluvial sandstone of the -Mojado Formation (Fig. 10, unit 14; Fig. 11, - -FIGURE 26—Early Cretaceous depositional basins of the Comanchean Series (uppermost Lower and -lowermost Upper Cretaceous) in Texas, northern Mexico, and adjacent areas (after Scott et al. 2003). - -Second order cycles and sequence boundaries - -In Texas the Fredericksburg and Washita -Groups have long been regarded as sec- -ond-order cycles of deposition (e.g., Lozo -and Stricklin 1956; Hendricks 1967; Scott -et al. 2003). Across much of Texas, the top -of the Fredricksburg Group is a regional -hardground with subaerial exposure (uncon- -formity) overlain by the Kiamichi Formation -(base of the Washita Group; Amsbury 2003; -Scott et al. 2003). Thus, the Fredericksburg -and Washita Groups are unconformity- -bounded units with a basal clastic facies that -generally grades up into intercalated shallow- -marine carbonates and clastics. At Cerro de -Cristo Rey, the entire Washita Group second- -order cycle is exposed (Fig. 27). It is bounded -by distinct unconformities at the base of the -Del Norte and Mancos Formations. Only the -uppermost part of the Fredericksburg cycle -(upper Finlay Formation) and the lowermost -part of the Greenhorn cycle (lower Mancos -Formation) are exposed at Cerro de Cristo -Rey (Fig. 27). - -Third order cycles and sequence boundaries - -Based on the interpretation of Scott et al. -(2001, 2003), the Cretaceous section exposed -at Cerro de Cristo Rey represents eight -smaller-scale (third order?) depositional -cycles, six of which comprise the Wash- -ita Group section (Fig. 27). In the Washita -Group section, Scott et al. (2001, 2003) iden- -tify these six cycles (WA1 through WA6) -as shale-sandstone or mudstone-limestone -cycles. Our data strongly support this inter- -pretation, although we differ from Scott -et al. (2001, 2003) in our placement of the - -sequence boundaries (and thus stratigraph- -ic extent) of the lower three cycles. We thus -identify six upward-shallowing shale-sand- -stone or shale-limestone cycles bounded by -unconformities in the Washita Group sec- -tion at Cerro de Cristo Rey (Fig. 27). - -The Finlay Formation is the final high- -stand of the Fredericksburg second-order -cycle. Steinhoff (2003) identified high-fre- -quency cycles in the Finlay Formation that -begin with nodular, marly layers and grade -up into massive limestones and are 1–4.5 m -thick. The upper part of the Finlay Forma- -tion exposed at Cerro de Cristo Rey encom- -passes 12 such cycles (Fig. 6). These cycles -may correlate to 12 of the 18 high-frequen- -cy cycles that Steinhoff (2003) recognized as -part of an upper, high-frequency sequence 2 -of the Finlay Formation formed during a -time of relative sea-level highstand. By this -interpretation, the upper Finlay at Cerro de -Cristo Rey represents shoaling up followed -by sea level fall at the end of the last cycle -of deposition of the Fredericksburg Group -(Amsbury 2003; Scott et al. 2003). - -As noted above, Scott et al. (2000, 2003) -identified six third-order cycles of deposi- -tion in the Washita Group of northern Texas -(and at Cerro de Cristo Rey) that they labeled -WA1 through WA6. Thus, WA1 begins at the -base of the Del Norte Formation (= base of -Washita Group), where a ferruginous sand- -stone rests directly on lime mudstone at the -top of the Finlay Formation (Fig. 27). The con- -tact reflects an abrupt increase in water depth -and siliciclastic influx, thus representing a -transgressive surface. The transgressional- -progradational succession of cycle WA1 is -visible in the lower member of the Del Norte - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -127 - - unit 4) is indicated by bioturbated sand- -stones and interbedded siltstones (Fig. 11, -units 7–14) and is followed by shallowing, -indicated by prograding fluvial sandstones -(Fig. 11, units 15–25). - -In north Texas, WA5 represents a regres- -sive trend from deep to shallow shelf -conditions (Scott et al. 2003). This trend -is indicated by the facies of the Pawpaw -Formation, which northward becomes -very sandy with thin-bedded and trough -crossbedded sandstone of deltaic origin. -The overlying Main Street Formation also -indicates a northward change from a car- -bonate shelf to shoreface environment -with significant input of quartz. A similar -shoaling trend is clearly visible within the -Mojado Formation, which approximately -corresponds to cycle WA5: lower shore- -face to upper shoreface and finally to flu- -viodeltaic sandstone. - -Cycle WA6 in north Texas is represent- -ed by the Grayson and Buda Formations, -which correlate to the uppermost Mojado, -Del Rio, and Buda Formations at Cerro -de Cristo Rey (Scott et al. 2001, 2003). The -Grayson is characterized by a progressive -change from carbonate to clay. The lower -part is composed of shale with intercalated -ripple-laminated beds of carbonate silt and -sand interpreted as storm layers overlain -by clay mainly deposited below the storm -wave base. The middle part is represent- -ed by nodular wackestone and shale, and -the uppermost part by shale, interbedded -marl, and bioturbated limestone. Similar -lithologies are recognized within the Del -Rio Formation, although storm layers are -absent, and deposition took place in a low- -energy outer-shelf environment below the -storm wave base. - -The uppermost Mojado Formation pre- -serves a transgressive surface that we iden- -tify as the basal unconformity of the WA6 -cycle (Fig. 11, base of unit 26; Fig. 27). This -cycle deepens upward through the lower -Del Rio, then shallows into the platform -carbonates of the Buda Formation. The -Buda Formation represents the last marine -transgression of the Comanchean Seaway, -and Reaser and Robinson (2003) interpreted -Buda deposition (in Trans-Pecos Texas) as -representing a transgression across a broad, -low-relief carbonate ramp (lower Buda = -Lecheguilla Member), followed by a break -in transgression and a regression during -which terrigenous clays were deposited -(middle Buda = Red Light Member), cul- -minated by a final transgression and high- -stand (upper Buda = Love Station Member). -The Buda outcrop at Cerro de Cristo Rey is -the westernmost outcrop of the formation, -and is a relatively thin (< 20 m thick) and -shaly section of the formation. Given that -the Del Rio shows an upward-shallowing -trend, we think it most likely that the Buda -was deposited during regression or during -a prograding highstand. Limestone of the -Buda Formation at Cristo Rey suggests dep- -osition in an open-marine, low-energy shelf -environment below wave base. Increased - -deposition of shale in the upper part may -indicate a drop in sea level. - -The unconformity at the base of the Man- -cos Formation (Fig. 27) begins the next cycle -of deposition (Greenhorn cycle). It repre- -sents a major tectonic reorganization of the -Cretaceous marine basins of the American -Southwest (e.g., Mack 1987). At Cerro de -Cristo Rey, this was the change from dep- -osition along an extensive carbonate plat- -form (Comanchean Platform) that bordered -a rift basin (Chihuahua trough) to deposi- -tion in the retroarc foreland basin of the -Late Cretaceous Western Interior Seaway. -The persistence of cycles from the tectoni- -cally passive, open-marine margin of the -Gulf of Mexico into the tectonically active -Chihuahua trough suggests that regional, -if not global eustasy, not local tectonism, -drove late Early to early Late Cretaceous -sedimentation at Cerro de Cristo Rey. - -Acknowledgments - -We are grateful to George Cudahy for access -to land, and assistance in the field from -Josh Smith. We thank Daniel Vachard, who -helped us to identify foraminiferans and -algae in thin sections. Comments by Robert -W. Scott on an earlier version of this man- -uscript greatly improved its content and -organization. W. R. Dickinson, T. Lawton, -and R. W. Scott provided helpful reviews of -the submitted manuscript. - -References - -Adkins, W. C., 1932, The Mesozoic system in Texas; -in Sellards, E. H., Adkins, W. C., and Plummer, -F. B. (eds.), The geology of Texas: University of -Texas, Publication 3252, pp. 329–518. - -Amsbury, D. L., 2003, Stratigraphy of Fredericks- -burg Group (middle–upper Albian Cretaceous) -north-central Texas; in Scott, R. W. (ed.), Creta- -ceous stratigraphy and paleoecology, Texas and -Mexico: Perkins Memorial volume, GCSSEPM -Foundation, Special Publications in Geology, v. 1, -pp. 227–276. - -Böse, E., 1910, Monografia geológica y paleontologi- -cal del Cerro de Muleros cerca de Ciudad Juárez, -Estado de Chihuahua, y descripción de la fauna -cretácea de La Encanatada, Placer de Guadalupe, -Estado de Chihuahua: Instituto Geologico de -Mexico, Bulletin 25, 193 pp. - -Bullock, J. S., 1985, Biostratigraphic study of -foraminifera from the Smeltertown Formation -(Lower Cretaceous, Albian) of southeastern New -Mexico: Unpublished M.S. thesis, University of -Texas at El Paso, 155 pp. - -Bullock, J. S., and Cornell, W. C., 1986, Biostratigra- -phy and paleoecology of the Smeltertown Forma- -tion, Cerro de Cristo Rey, Doña Ana County, New -Mexico; in Hoffer, J. M. (ed.), Geology of south- -central New Mexico: El Paso Geological Society, -Guidebook, pp. 73–76. - -Cobban, W. A., 1987, Ammonite faunas of the Sarten -Sandstone (Cretaceous), Luna County, New Mexico: -U.S. Geological Survey, Bulletin 1641–B, pp. B1–B17. -Cobban, W. A., Hook, S. C., and McKinney, K. C., -2008, Upper Cretaceous molluscan record along -a transect from Virden, New Mexico, to Del Rio, -Texas: New Mexico Geology, v. 30, pp. 75–92. - -FIGURE 27—Summary of Cretaceous section at -Cerro de Cristo Rey showing sequence boundar- -ies and depositional cycles. - -128 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - - Cooper, R. W., Cooper, D. A., Stevens, J. B., and -Stevens, M. S., 2008, Geology of the Hot Springs -Trail area, Ernst and San Vicente Members of the -Boquillas Formation; in Cooper, D. A. (ed.), The -southern extension of the Western Interior Sea- -way: Geology of the Big Bend National Park and -Trans-Pecos, Texas, 2008 Joint Annual Meeting of -the Geological Society of America and the Hous- -ton Geological Society, pp. 24–33. - -Córdoba, D. A., 1968, Juárez Mountains: West Texas -Geological Society, Guidebook, Publication 68-55, -pp. 85–86. - -Córdoba, D. A., 1969a, Mesozoic stratigraphy of -northeastern Chihuahua, México; in Córdoba, D. -A., Wengerd, S. A., and Shomaker, J. (eds.), Guide- -book of the border region: New Mexico Geologi- -cal Society, Guidebook 20, pp. 91–96. - -Córdoba, D. A., 1969b, Geologia del area de Cerro -de Muleros o Cristo Rey [inset on the] Carta -geológica de México, scale 1:100,000, hoja Ciudad -Juárez 13R-a(3); in Córdoba, D. A., Wengerd, S. A., -and Shomaker, J. (eds.), Guidebook of the border -region: New Mexico Geological Society, Guide- -book 20, rear pocket. - -Cornell, W. C., 1982, Dinoflagellate cysts from the -Mesilla Valley Shale (late Albian), El Paso, Texas -(abs.): American Association of Stratigraphic Pal- -ynologists, 14th Annual Meeting, October, 1981, -Program and Abstracts, p. 17. - -Cornell, W. C., 1997, Dinoflagellate cysts from the -Buda Limestone (Cenomanian) Cerro de Cristo -Rey, Doña Ana County, New Mexico: Review of -Paleobotany and Palynology, v. 98, pp. 153–157. -Cornell, W. C., and LeMone, D. V., 1987, The Anapra -Sandstone of Cerro de Cristo Rey, Doña Ana -County, New Mexico (abs.): New Mexico Journal -of Science, v. 27, p. 123. - -Cornell, W.C., LeMone, D.V., and Norland, W. D., -1991, Albian ophiuroids from Cerro de Cristo Rey, -Doña Ana County, New Mexico: Journal of Pale- -ontology, v. 65, pp. 1009–1013. - -Emory, W. H., 1857, Report of the United States and -Mexican boundary survey: U.S. 34th Congress, 1st -session, Senate ex. Doc. 108, v. 1, pt. 2, 174 pp. -Flügel, E., 2004, Microfacies of carbonate rocks: -analysis, interpretation and application: Springer, -Berlin, 976 pp. - -Hendricks, L., 1967, Comanchean stratigraphy of the -Cretaceous of north Texas; in Hendricks, L. (ed.), -Comanchean (Lower Cretaceous) stratigraphy -and paleontology of Texas: Society of Economic -Paleontologists and Mineralogists (SEPM), Perm- -ian Basin Section, Publication 67-8, pp. 51–63. -Hobday, D. K., and Morton, R. A., 1984, Lower Cre- -taceous shelf storm deposits, northeast Texas; in -Tillman, R. W., and Siemers, C. T. (eds.), Siliciclas- -tic shelf sediments: Society of Economic Paleon- -tologists and Mineralogists (SEPM), Special Pub- -lication 34, pp. 205–213. - -Hook, S. C., 2008, Gallery of geology—Sierra de -Cristo Rey: New Mexico Geology, v. 30, pp. 93–94. -Imlay, R. W., 1944, Cretaceous formations of Central -America and Mexico: American Association of -Petroleum Geologists, Bulletin 28, pp. 878–1028. -Kappus, E. J., 2007, Middle Cretaceous dinosaur -tracks at Cerro de Cristo Rey, Sunland Park, New -Mexico, and a comparison with other paleocoastal -tracksites of the southwestern US: Unpublished -M.S. thesis, University of Texas at El Paso, 152 pp. -Kappus, E. J., Lucas, S. G., Heckert, A. B., Hunt, A. P., -and Lockley, M. G., 2003, Dinosaur footprints from -the Lower Cretaceous Sarten Member of the Moja- -do Formation at Cerro de Cristo Rey, Doña Ana -County, New Mexico: Ichnos, v. 10, pp. 263–267. -Kennedy, W. J., Cobban, W. A., and Hook, S. C., 1988, -Middle Cenomanian (Late Cretaceous) molluscan -fauna from the base of the Boquillas Formation, -Cerro de Muleros, Doña Ana County, New Mexi- -co; in Contributions to Late Cretaceous paleontol- -ogy and stratigraphy of New Mexico, Part II: New -Mexico Bureau of Mines and Mineral Resources, -Bulletin 114, pp. 35–44. - -Kollmann, H. A., Decker, K., and LeMone, D. V., -2003, Facies control of Lower Cretaceous gastro- -pod assemblages, southwestern United States; in -Scott, R. W. (ed.), Cretaceous stratigraphy and -paleoecology, Texas and Mexico: Perkins Memo- -rial volume, GCSSEPM Foundation, Special Pub- -lications in Geology, v. 1, pp. 101–146. - -Kues, B. S., 1989, Taxonomy and variability of three -Texigryphaea (Bivalvia) species from their Lower Cre- -taceous (Albian) type localities in New Mexico and -Oklahoma: Journal of Paleontology, v. 63, pp. 454–483. -Kues, B. S., 1997, New bivalve taxa from the Tucumcari -Formation (Cretaceous, Albian), New Mexico, and -the biostratigraphic significance of the basal Tucum- -cari fauna: Journal of Paleontology, v. 71, pp. 820–839. -Kues, B. S., and Lucas, S. G., 1987, Cretaceous stra- -tigraphy and paleontology in the Dry Cimarron -Valley, New Mexico, Colorado, and Oklahoma; in -Lucas, S. G., and Hunt, A. P. (eds.), Northeastern -New Mexico: New Mexico Geological Society, -Guidebook 38, pp. 167–198. - -LeMone, D. V., and Simpson, R. D., 1983, Creta- -ceous biostratigraphy of the Franklin Mountains, -El Paso County, Texas; in Allen, R. (ed.), Delaware -Basin: West Texas Geological Society, Publication -82–76, pp. 85–87. - -Lovejoy, E. M. P., 1976, Geology of Cerro de Cristo -Rey uplift, Chihuahua and New Mexico: New -Mexico Bureau of Mines and Mineral Resources, -Memoir 31, 84 pp. - -Lozo, F. E., and Stricklin, F. L., Jr., 1956, Stratigraph- -ic notes on the outcrop basal Cretaceous, central -Texas: Gulf Coast Association of Geological Socie- -ties, Transactions, v. 6, pp. 67–78. - -Lucas, S. G., and Estep, J. W., 1998a, Cretaceous -stratigraphy and biostratigraphy in the southern -San Andres Mountains, Doña Ana County, New -Mexico; in Mack, G. H., Austin, G. S., and Barker, -J. M. (eds.), Las Cruces country II: New Mexico -Geological Society, Guidebook 49, pp. 187–196. -Lucas, S. G., and Estep, J. W., 1998b, Lithostratigra- -phy and biostratigraphy of the Lower-Middle Cre- -taceous Bisbee Group, southwestern New Mexico, -USA; in Lucas, S. G., Kirkland, J. I., and Estep, J. -W. (eds.), Lower and middle Cretaceous terrestrial -ecosystems: New Mexico Museum of Natural His- -tory and Science, Bulletin 14, pp. 39–55. - -Lucas, S. G., and Estep, J. W., 2000, Lower Cretaceous -stratigraphy, correlation and paleogeography of -New Mexico; in Lucas, S. G. (ed.), New Mexico's -fossil record 2: New Mexico Museum of Natural -History and Science, Bulletin 16, pp. 45–61. - -Lucas, S. G., Estep, J. W., Boucher, L. D., and -Anderson, B. G., 2000, Cretaceous stratigraphy, -biostratigraphy and depositional environments -near Virden, Hidalgo County, New Mexico; in -Lucas, S. G. (ed.), New Mexico's fossil record 2: -New Mexico Museum of Natural History and Sci- -ence, Bulletin 16, pp. 107–119. - -Lucas, S. G., Kues, B. S., Hayden, S. N., Allen, B. -D., Kietzke, K. K., Williamson, T. E., Sealey, P., -and Pence, R., 1988, Cretaceous stratigraphy and -biostratigraphy, Cooke’s Range, Luna County, New -Mexico; in Mack, G. H., Lawton, T. F., and Lucas, S. -G. (eds.), Cretaceous and Laramide tectonic evo- -lution of southwestern New Mexico: New Mexico -Geological Society, Guidebook 39, pp. 143–167. - -Mack, G. H., 1987, Mid-Cretaceous (late Albian) -change from rift to retroarc foreland basin in -southwestern New Mexico: Geological Society of -America, Bulletin, v. 98, pp. 507–514. - -Mauldin, R. A., 1985, Foraminiferal biostratigraphy, -paleoecology, and correlation of the Del Rio Clay -(Cenomanian), from Big Bend National Park, -Brewster County, Texas, to the Cerro de Muleros -area, Doña Ana County, New Mexico: Unpublished -M.S. thesis, University of Texas at El Paso, 89 pp. -Mauldin, R. A., and Cornell, W. C., 1986, Paleoecol- -ogy and biostratigraphic correlation of the Del Rio -Clay, Cerro de Cristo Rey, Doña Ana County, New -Mexico; in Hoffer, J. M. (ed.), Geology of south- -central New Mexico: El Paso Geological Society, -Guidebook, pp. 77–84. - -Molenaar, C. M., 1983, Major depositional cycles and -regional correlations of Upper Cretaceous rocks, -southern Colorado Plateau and adjacent areas; in -Reynolds, M. W., and Dolly, E. D. (eds.), Mesozoic -paleogeography of west-central United States: Soci- -ety of Economic Peleontologists and Mineralogists -(SEPM), Rocky Mountain Section, Rocky Mountain -Paleogeography Symposium 2, pp. 201–224. - -Nye, O. B., Jr., and LeMone, D. V., 1978, Multilami- -nar growth in Reptomulticava texana, a new species -of cyclostome bryozoa: Journal of Paleontology, -v. 52, pp. 830–885. - -Reaser, D. F., and Malott, V. E., 1985, Finlay Lime- -stone: A key to geological interpretation in west- -ern Trans-Pecos Texas and northeastern Texas and -northeastern Chihuahua, Mexico; in Dickerson, P. -W., and Muehlberger, W. R. (eds.), Structure and -tectonics of Trans-Pecos Texas: West Texas Geo- -logical Society, Publication 85–81, pp. 213–219. -Reaser, D. F., and Robinson, W. C., 2003, Cretaceous -Buda Limestone in west Texas and northern Mex- -ico; in Scott, R. W. (ed.), Cretaceous stratigraphy -and paleoecology, Texas and Mexico: Perkins -Memorial volume, GCSSEPM Foundation, Special -Publications in Geology, v. 1, pp. 337–373. - -Richardson, G. B., 1909, Description of the El Paso -quadrangle Texas; in Geologic atlas of the United -States, El Paso folio no. 166: U.S. Geological Sur- -vey, 11 pp. - -Scott, R. W., 1970, Stratigraphy and sedimentary -environments of Lower Cretaceous rocks, southern -Western Interior: American Association of Petro- -leum Geologists, Bulletin, v. 54, pp. 1225–1244. - -Scott, R. W., and Kidson, E. J., 1977, Lower Cretaceous -depositional systems, west Texas; in Bebout, D. G., -and Loucks, R. G. (eds.), Cretaceous carbonates of -Texas and Mexico: Texas Bureau of Economic Geol- -ogy, Report of Investigations 89, pp. 169–181. - -Scott, R. W., Holbrook, J. M., Evetts, M. J., and -Oboh-Ikuenobe, F. E., 2001, Albian–Cenomanian -depositional cycles transgressed from Chihua- -hua trough to Western Interior; in Lucas, S. G., -and Ulmer-Scholle, D. S. (eds.), Geology of the -Llano Estacado: New Mexico Geological Society, -Guidebook 52, pp. 221–228. - -Scott, R. W., Hooman, L., and Fee, D. W., 1975, Den- -sity-current strata in Lower Cretaceous Washita -Group, north-central Texas: Journal of Sedimen- -tary Petrology, v. 45, pp. 562–575. - -Scott, R. W., Schlager, W., Fouke, B., and Neder- -bragt, S. A., 2000, Are mid-Cretaceous eustatic -events recorded in Middle East carbonate plat- -forms? in Alsharhan, A. S., and Scott, R. W. (eds.), -Middle East models of Jurassic/Cretaceous car- -bonate systems: SEPM (Society for Sedimentary -Geology), Special Publication 69, pp. 73–84. - -Scott, R. W., Benson, D. G., Morin, R. W., Shaffer, B. L., -and Oboh-Ikuenobe, F. E., 2003, Integrated Albian– -lower Cenomanian chronostratigraphy standard, -Trinity River section, Texas; in Scott, R. W. (ed.), -Cretaceous stratigraphy and paleoecology, Texas -and Mexico: Perkins Memorial volume, GCSSEPM -Foundation, Special Publications in Geology, v. 1, -pp. 277–334. - -Seager, W. R., 1981, Geology of Organ Mountains -and southern San Andres Mountains, New Mex- -ico: New Mexico Bureau of Mines and Mineral -Resources, Memoir 36, 97 pp. - -Stanton, T. W., and Vaughan, T. W., 1896, Sections of -the Cretaceous at El Paso, Texas: American Journal -of Science, 4th series, v. 1, pp. 21–26. - -Steinhoff, D., 2003, Albian Finlay Formation strato- -type, Rimrock escarpment, west Texas; in Scott, R. -W. (ed.), Cretaceous stratigraphy and paleoecol- -ogy, Texas and Mexico: Perkins Memorial volume, -GCSSEPM Foundation, Special Publications in -Geology, v. 1, pp. 69–99. - -Strain, W. S., 1968, Cerro de Muleros (Cerro de Cris- -to Rey): West Texas Geological Society, Delaware -Basin Exploration Guide, Publication 68–55, p. 82. -Strain, W. S., 1976, New formation names in the Cre- -taceous at Cerro de Cristo Rey, Doña Ana County, -New Mexico; in Lovejoy, E. M. P., Geology of -Cerro de Cristo Rey uplift, Chihuahua and New -Mexico: New Mexico Bureau of Mines and Min- -eral Resources, Memoir 31, pp. 77–82. - -November 2010, Volume 32, Number 4 - -New Mexico Geology - -129 - - Turnšek, D., LeMone, D. V., and Scott, R. W., 2003, -Tethyan Albian corals, Cerro de Cristo Rey uplift, -Chihuahua and New Mexico; in Scott, R. W. -(ed.), Cretaceous stratigraphy and paleoecology, -Texas and Mexico: Perkins Memorial volume, -GCSSEPM Foundation, Special Publications in -Geology, v. 1, pp. 147–185. - -Young, K., 1966, Texas Mojsisovicinae (Ammonoidea) -and the zonation of the Fredericksburg: Geological -Society of America, Memoir 100, 225 pp. - -Young, K., 1979, Lower Cenomanian and late Albian -(Cretaceous) ammonites, especially Lyelliceridae, -of Texas and Mexico: Texas Memorial Museum, -Bulletin 26, 99 pp. - -Zeller, R. A., Jr., 1965, Stratigraphy of the Big Hatch- -et Mountains area, New Mexico: New Mexico -Bureau of Mines and Mineral Resources, Mem- -oir 16, 128 pp. - -Appendix - -Location of measured sections - -Finlay Formation (Fig. 6)—base at UTM -Zone 13, 355011E 3518134N (NAD 27), -top at 355064E 3518629N. - -Del Norte Formation (Fig. 7)—base at top of -section A, top at UTM Zone 13, 355084E -3517962N (NAD 27). - -Smeltertown Formation (Fig. 8)—base at -UTM Zone 13, 355178E 3517669N, top at -355040E 3517595N. - -Muleros Formation (Fig. 9)—base at UTM -Zone 13, 354953E 3517640N, top at -354922E 3517683N. - -Mesilla Valley Formation (Fig. 10)—base at -UTM Zone 13, 354508E 3518122N, top at -354535E 3518200N. - -Mojado Formation (Fig. 11)—base at -UTM Zone 13, 354003E 3518452N, top -at 353967E 3518553N. - -Del Rio Formation (Fig. 13)—base at UTM -Zone 13, 353899E 3118636N, top at -353871E 3518678N. - -Buda Formation (Fig. 14)—base at UTM -Zone 13, 353881E 3518823N, top at -353859E 3518828N. - -Mancos Formation (Fig. 15)—base at UTM -Zone 13, 353534E 3518775N, top at -353524E 3518792N. - -130 - -New Mexico Geology - -November 2010, Volume 32, Number 4 - +1 +PEW RESEARCH CENTER +2024 PEW RESEARCH CENTER’S AMERICAN TRENDS PANEL +WAVE 159 POLITICS SURVEY NOVEMBER 2024 +FINAL TOPLINE +NOVEMBER 12-17, 2024 +N=9,609 +Note: This survey was conducted primarily online, with some interviews conducted by live telephone. This +topline shows the programming language for online administration. For details on how questions were slightly +modified for phone administration, visit the questionnaire. +American Trends Panel surveys conducted between October 2016 and June 2024 were conducted fully online +(with tablets and data plans provided to adults without home internet). American Trends Panel surveys +conducted prior to October 2016 were conducted primarily online, with some respondents completing by mail. +For additional details, refer to the methodology. +* “No answer” includes web respondents who do not answer the question as well as telephone respondents +who refuse to answer or who say that they don’t know how to answer. In cases where “not sure” was offered +as an explicit option to web and telephone respondents, the “no answer” category includes only web skips and +telephone refusals. +In addition to the parameters used in weighting each wave of the American Trends Panel, the results of this +survey were weighted to the 2024 presidential popular vote and turnout among the voting-eligible population +(as of Nov. 18, 2024). For additional information, refer to the methodology. +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED +ASK ALL CITIZENS (XCITIZEN=1) [N=9,302]: +VOTED Which of the following statements best describes you? +Nov 29- +Nov 12-17, Nov 16-27, Nov 12-17, Nov 7-16, Dec 12, +2024 20221 2020 2018 2016 +I did not vote in the November +22 2024 election 32 22 23 14 +13 I planned to vote but wasn’t able to 20 9 22 8 +I definitely voted in the November +65 2024 election 47 67 55 77 +1 No answer* 1 2 1 1 +1 In 2022 and 2018, this question was asked about that November’s congressional election. In 2020 and 2016, this question +was asked about that November’s presidential election. +www.pewresearch.org + +2 +PEW RESEARCH CENTER +ASK IF VOTED (VOTED=3) [N=8,072]: +VOTEGEN_POST In the 2024 presidential election, who did you vote for? [RANDOMIZE OPTIONS 1 AND 2 +FIRST FOLLOWED BY RANDOMIZED OPTIONS 3,4,5,6,7 WITH OPTION 8 AND 9 ALWAYS LAST] +Nov 12-17, +| 2024 | | | | | | | | | | +| ----- | ----------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| 49 | Donald Trump, the Republican | | | | | | | | | +| 47 | Kamala Harris, the Democrat | | | | | | | | | +| * | Robert F. Kennedy Jr., a third party candidate | | | | | | | | | +| * | Chase Oliver, the Libertarian Party candidate | | | | | | | | | +| * | Jill Stein, the Green Party candidate | | | | | | | | | +| * | Cornel West, a third party candidate | | | | | | | | | +| * | Claudia De la Cruz, the Socialism and Liberation Party candidate | | | | | | | | | +| * | Another candidate | | | | | | | | | +| 2 | I did not vote for a candidate for president | | | | | | | | | +| * | No answer* | | | | | | | | | + +ADDITIONAL QUESTIONS HELD FOR FUTURE RELEASE + +ASK IF DID NOT VOTE (VOTED=1-2) [N=1,201]: +| WISH_VOTED | Do you wish that you had voted in the election? | | | | | | | | | +| ----------- | ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | + +| | | | | | | Nov 29- | | Nov 17- | | +| ----------- | ----------- | ----------- | --- | ---------- | --- | -------- | --- | -------- | --- | +| Nov 12-17, | | Nov 12-17, | | Nov 7-16, | | Dec 12 | | Dec 15, | | +| 2024 | | 2020 | | 2018 | | 2016 | | 2014 | | +| 42 | Yes | | 45 | 61 | | 44 | | 55 | | +| 57 | No | | 54 | 38 | | 55 | | 43 | | +| 1 | No answer* | | 1 | | 1 | * | | 1 | | + +ASK IF CITIZEN AND DID NOT VOTE (X_CITIZEN=1 AND VOTED=1,2) [N=1,201]: +NOVOTERSN2 There are lots of different reasons why many people don’t vote. For you personally, +how much of a reason, if at all, is each of the following for why you did not vote? +[RANDOMIZE] + +| | | | Major | | Minor | | Not a | | No | +| ----- | ------------------------------------ | --- | ------- | --- | ------- | --- | ------- | --- | -------- | +| | | | reason | | reason | | reason | | answer* | +| DIFF | I didn’t think my vote would make a | | | | | | | | | +difference +| | Nov 12-17, 2024 | | 35 | | | 26 | 39 | | 1 | +| ----- | -------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | Nov 7-16, 2018 | | 21 | | | 23 | 55 | | 1 | +| | | | | | | | | | | +| CARE | I didn’t care who won2 | | | | | | | | | +| | Nov 12-17, 2024 | | 17 | | | 21 | 61 | | 1 | +| | Nov 7-16, 2018 | | 12 | | | 23 | 64 | | 1 | +| | | | | | | | | | | +| ELGB | I’m not registered or not eligible to | | | | | | | | | + vote +| | Nov 12-17, 2024 | | 18 | | | 15 | 66 | | 1 | +| ----- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | Nov 7-16, 2018 | | 19 | | | 10 | 70 | | 1 | +| | | | | | | | | | | +| INCV | It was too inconvenient for me to vote | | | | | | | | | +| | Nov 12-17, 2024 | | 15 | | | 22 | 62 | | 1 | +| | Nov 7-16, 2018 | | 20 | | | 21 | 59 | | 1 | +| | | | | | | | | | | + +2 In 2018, the reason was worded “I didn’t care who won the congressional election in my area.” +www.pewresearch.org + +3 +PEW RESEARCH CENTER +| NOVOTERSN2 CONTINUED … | | | | | | | | | | | | +| ----------------------- | --- | ---------------------- | --- | ------- | --- | ------- | ------ | --- | ------- | -------- | --- | +| | | | | Major | | | Minor | | Not a | No | | +| | | | | reason | | reason | | | reason | answer* | | +| LIKE | | I don’t like politics | | | | | | | | | | +| | | Nov 12-17, 2024 | | 31 | | | 25 | | 43 | 1 | | +| | | Nov 7-16, 2018 | | 26 | | | 23 | | 50 | 1 | | +| | | | | | | | | | | | | +| FRGT | | I forgot to vote | | | | | | | | | | +| | | Nov 12-17, 2024 | | | 8 | | 13 | | 79 | 1 | | +| | | Nov 7-16, 2018 | | | 7 | | 15 | | 77 | 1 | | +| | | | | | | | | | | | | + +ASK IF VOTED TRUMP (VOTEGEN_POST=1) [N=3,536]: +DTFORAGNST_POST Would you say that your vote for Trump was more a vote… +ASK IF VOTED HARRIS (VOTEGEN_POST=2) [N=4,185]: +KHFORAGNST_POST Would you say that your vote for Harris was more a vote… + +BASED ON VOTERS [N=8,072]: + +| | | | | | | Based on registered voters | | | | | | +| ------------- | ----------- | --- | ------------------------------------------ | --- | --- | --------------------------- | ------ | --- | -------------- | --- | --- | +| | Nov 12-17, | | | | | Sep 30-Oct 6, | | | Aug 26-Sep 2, | | | +| | 2024 | | | | | | 20243 | | 2024 | | | +| | 49 | | Trump | | | | 47 | | 49 | | | +| | 40 | | For Trump | | | | 31 | | 33 | | | +| | 9 | | Against Harris | | | | 15 | | 16 | | | +| | * | | Refused | | | | * | | * | | | +| 47 | | | Harris | | | | 48 | | 49 | | | +| | 23 | | For Harris | | | | 23 | | 24 | | | +| | 24 | | Against Trump | | | | 25 | | 25 | | | +| | * | | Refused | | | | * | | * | | | +| | 2 | | Vote for Kennedy Jr./Oliver/Stein/West/De | | | | - | | - | | | +La Cruz/Another candidate +| | 2 | | I did not vote for a candidate for president | | | | | | | | | +| --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | * | | No answer* | | | | * | | * | | | + +| | | 2020 TREND FOR COMPARISON: | | | | | | | | | | +| --- | --- | ---------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | Would you say that your vote for Trump was more a vote… | | | | | | | | | | +| | | Would you say that your vote for Biden was more a vote… | | | | | | | | | | + +| | | | | | Based on voters | | | | Based on registered voters | | | +| --- | --- | --- | ----------------------------------- | --- | ---------------- | ----------- | --- | -------------- | --------------------------- | --- | ----------- | +| | | | | | | Nov 12-17, | | Sep 30-Oct 5, | | | Jun 16-22, | +| | | | | | | 2020 | | | 20204 | | 2020 | +| | | | Trump | | | 47 | | | 42 | | 44 | +| | | | For Trump | | | 36 | | | 30 | | 33 | +| | | | Against Biden | | | 11 | | | 12 | | 10 | +| | | | Biden | | | 51 | | | 52 | | 54 | +| | | | For Biden | | | 24 | | | 19 | | 18 | +| | | | Against Trump | | | 28 | | | 33 | | 36 | +| | | | Vote for Jorgensen/Hawkins/Another | | | 1 | | | - | | - | +candidate +| | | | No answer | | | * | | | * | | - | +| --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | + +3 +Trend for DTFORAGNST and KHFORAGNST asked prior to election combined those who said they were voting for the +candidate or leaning toward the candidate. Postelection versions of these questions asked those who said they voted for +Trump or Harris. +4 +Trend for DTFORAGNST and JBFORAGNST asked prior to election combined those who said they were voting for the +candidate or leaning toward the candidate. Postelection versions of these questions asked those who said they voted for +Trump or Biden. +www.pewresearch.org + +4 +PEW RESEARCH CENTER +ASK IF VOTED (VOTED=3) [N=8,072]: +VOTEDECTIME As far as you can remember, when did you make up your mind about who you were going to +vote for in the presidential election? +BASED ON THOSE WHO SELECTED A CANDIDATE (VOTEGEN_POST=1-8) [N=7,977] +Nov 29- +Nov 12-17, Nov 12-17, Dec 12 +2024 2020 20165 +Last few days before Election +4 Day 3 9 +The last week before Election +3 Day 2 4 +7 In October 5 9 +6 In September 5 11 +81 Before September 84 67 +* No answer* * * +PHONE TREND FOR COMPARISON: +When did you make up your mind definitely to vote for “Hillary Clinton;” “Donald Trump;” “Gary +Johnson;” “Jill Stein,” “The candidate of your choice”? +Nov 10-14 Nov 8-11 Nov Nov Nov Nov Nov +2016 2012 2004 2000 1996 1992 1988 +On Election Day 4 4 4 5 6 9 6 +On Monday (day before the election) * 1 1 2 3 4 3 +Over the last weekend * 1 1 2 2 4 2 +In the last week 2 2 3 5 6 8 4 +In the last few weeks 5 -- -- -- -- -- -- +During or just after the Presidential debates 10 10 11 17 -- 12 -- +Before debates/after the Conventions 10 9 6 9 12 6 16 +Convention period (NET) 22 10 7 7 8 12 15 +During or after the Republican convention 13 9 1 4 -- -- -- +During or after the Democratic convention 9 1 6 3 -- -- -- +This year, before the conventions 22 16 23 21 13 18 19 +Before 2016 20 41 38 20 39 12 9 +Other -- 0 0 -- * 8 -- +Don’t remember 2 1 4 6 6 5 4 +Don’t know/Refused 1 4 2 3 2 1 2 +ADDITIONAL QUESTIONS HELD FOR FUTURE RELEASE +5 In 2016, people who said “another candidate” were not asked this question; question read “As far as you can remember, +when did you make up your mind definitely to vote for ‘Donald Trump;’ ‘Hillary Clinton;’ ‘Gary Johnson;’ ‘Jill Stein’?” +www.pewresearch.org + +5 +PEW RESEARCH CENTER +[RANDOMIZE ORDER OF VTADMIN_POST_COM AND VTADMIN_POST_US] + +ASK ALL: +VTADMIN_POST_COM Do you think the elections this November in your community were run and +administered… + +| Nov 12-17, | | | Nov 16-27, | | Nov 12-17, | | Nov 7-16, | | | | +| ----------- | ---------------- | --- | ----------- | --- | ----------- | --- | ---------- | --- | --- | --- | +| 2024 | | | 2022 | | 2020 | | 2018 | | | | +| 55 | Very well | | 42 | | 53 | | 49 | | | | +| 35 | Somewhat well | | 42 | | 32 | | 38 | | | | +| 7 | Not too well | | 9 | | 9 | | | 8 | | | +| 2 | Not at all well | | 4 | | 6 | | | 4 | | | +| 1 | No answer* | | 3 | | 1 | | | 1 | | | + +TREND FOR COMPARISON: +Do you think the elections this November in your community will be run and administered… + +| | | | Sep 30-Oct 6, | | Oct 10-16, | | | Sep 30-Oct 5, | Sep 24-Oct 7, | | +| --- | ---------------- | --- | -------------- | --- | ----------- | ----- | --- | -------------- | -------------- | ----- | +| | | | 2024 | | | 2022 | | 2020 | | 2018 | +| | Very well | | | 43 | | 40 | | 36 | | 41 | +| | Somewhat well | | | 44 | | 45 | | 48 | | 46 | +| | Not too well | | | 8 | | 10 | | 11 | | 9 | +| | Not at all well | | | 4 | | 4 | | 4 | | 3 | +| | No answer | | | 1 | | 1 | | * | | 1 | + +[RANDOMIZE ORDER OF VTADMIN_POST_COM AND VTADMIN_POST_US] + +ASK ALL: +VTADMIN_POST_US Do you think the elections this November in the United States were run and +administered… + +| Nov 12-17, | | Nov 16-27, | | Nov 12-17, | | | Nov 7-16, | | | | +| ----------- | ---------------- | ----------- | ----- | ----------- | ----- | --- | ---------- | --- | --- | --- | +| 2024 | | | 2022 | | 2020 | | 2018 | | | | +| 37 | Very well | | 25 | | 33 | | 21 | | | | +| 47 | Somewhat well | | 46 | | 25 | | 53 | | | | +| 11 | Not too well | | 18 | | 20 | | 18 | | | | +| 4 | Not at all well | | 8 | | 22 | | 7 | | | | +| 1 | No answer* | | 3 | | 1 | | 1 | | | | + +TREND FOR COMPARISON: +Do you think the elections this November in the United States will be run and administered… + +| | | | Sep 30-Oct 6, | | Oct 10-16, | | | Sep 30-Oct 5, | Sep 24-Oct 7, | | +| --- | ---------------- | --- | -------------- | --- | ----------- | ----- | --- | -------------- | -------------- | ----- | +| | | | 2024 | | | 2022 | | 2020 | | 2018 | +| | Very well | | | 20 | | 20 | | 12 | | 20 | +| | Somewhat well | | | 50 | | 49 | | 46 | | 57 | +| | Not too well | | | 21 | | 23 | | 30 | | 18 | +| | Not at all well | | | 8 | | 7 | | 12 | | 4 | +| | No answer | | | 1 | | 1 | | 1 | | 1 | + +| | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +www.pewresearch.org + +6 +PEW RESEARCH CENTER +ASK IF VOTED (VOTED=3) [N=8,072]: +VTEASY_POST Overall, was voting in the election this November for you personally... [ROTATE +RESPONSES OPTIONS 1-4/4-1] + +| Nov 12-17, | | | | | Nov 12-17, | | | | | | | +| ----------- | --- | ------------------- | --- | --- | ----------- | ----- | --- | --- | --- | --- | --- | +| 2024 | | | | | | 2020 | | | | | | +| | 79 | Very easy | | | | 77 | | | | | | +| | 15 | Somewhat easy | | | | 17 | | | | | | +| | 5 | Somewhat difficult | | | | 5 | | | | | | + +| | 2 | Very difficult | | | | 1 | | | | | | +| --- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +| | * | No answer* | | | | * | | | | | | +| --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +TREND FOR COMPARISON: +Overall, do you personally expect voting in the November elections to be... + +BASED ON REGISTERED VOTERS: + +| | | | | | | Sep 30- | Oct 10- | | Sep 30- | Jul 27- | Sep 24- | +| --- | --- | ------------------- | --- | --- | --- | -------- | ------- | --- | -------- | -------- | -------- | +| | | | | | | Oct 6, | 16 | | Oct 5 | Aug 2 | Oct 7 | +| | | | | | | 2024 | 2022 | | 2020 | 2020 | 2018 | +| | | Very easy | | | | 43 | 45 | | 29 | 23 | 46 | +| | | Somewhat easy | | | | 38 | 37 | | 36 | 28 | 39 | +| | | Somewhat difficult | | | | 16 | 14 | | 29 | 38 | 13 | +| | | Very difficult | | | | 4 | | 3 | 6 | 11 | 2 | +| | | No answer | | | | * | | * | * | 1 | * | + +ASK ALL: +SATCAND_POST Now that the campaign is over, how satisfied were you with the choice of presidential +candidates? Would you say that you were… +BASED ON VOTERS [N=8,072]: +| Nov 12-17, | | | | | | Nov 12-17, | | | | | | +| ----------- | --- | --------------------- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | +| 2024 | | | | | | 2020 | | | | | | +| | 32 | Very satisfied | | | | 30 | | | | | | +| | 20 | Fairly satisfied | | | | 26 | | | | | | +| | 22 | Not too satisfied | | | | 24 | | | | | | +| | 25 | Not at all satisfied | | | | 19 | | | | | | +| | * | No answer* | | | | * | | | | | | + + TREND FOR COMPARISON: +What is your opinion of the presidential candidates for this year? Would you say that you are… + +BASED ON REGISTERED VOTERS: + +| | | | | | 2024 Election | | | | 2020 Election | | | +| --- | --------------------- | --- | --- | -------- | -------------- | --------- | --- | --- | -------------- | ----------- | --- | +| | | | | Aug 26- | | | | | Jul 27- | | | +| | | | | Sep 2, | | Jul 1-7, | | | Aug 2, | Jun 16-22, | | +| | | | | 2024 | | 2024 | | | 2020 | 2020 | | +| | Very satisfied | | | | 15 | 11 | | | 15 | 16 | | +| | Fairly satisfied | | | | 32 | 20 | | | 31 | 31 | | +| | Not too satisfied | | | | 34 | 32 | | | 35 | 34 | | +| | Not at all satisfied | | | | 17 | 37 | | | 19 | 19 | | +| | No answer* | | | | 1 | | 1 | | 1 | 1 | | + +| | | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +www.pewresearch.org + +7 +PEW RESEARCH CENTER +SATCAND_POST TREND FOR COMPARSION CONTINUED … + +PHONE TREND FOR COMPARISON: + Now that the campaign is over, how satisfied were you with the choice of presidential candidates? + +BASED ON THOSE WHO VOTED: + +| | | | Nov 10-14, | | Nov 8-11, | | Nov Nov | | Nov Nov | Nov Nov | +| --------------------- | --- | --- | ----------- | --- | ---------- | --- | ----------- | --- | ----------- | ----------- | +| | | | 2016 | | 2012 | | 2008 2004 | | 2000 1996 | 1992 1988 | +| Very satisfied | | | | 19 | 37 | | 39 | 33 | 24 18 | 24 26 | +| Fairly satisfied | | | | 25 | 33 | | 28 | 33 | 44 33 | 37 36 | +| Not very satisfied | | | | 26 | 16 | | 17 | 16 | 18 24 | 21 20 | +| Not at all satisfied | | | | 29 | 12 | | 14 | 16 | 9 22 | 16 17 | +| Don't | | | | 1 | 3 | | 2 | 2 | 5 3 | 2 1 | +know/Refused +(VOL.) + +[DISPLAY CMPGNTRAITPOST_PROUD, CMPGNTRAITPOST_NEG, CMPGNTRAITPOST_POLCY, +CMPGNTRAITPOST_INT ON SAME PAGE; RANDOMIZE ORDER OF QUESTIONS; INCLUDE +RANDOMIZATION IN DATA FILE. DO NOT RANDOMIZE RESPONSE OPTIONS WITHIN EACH +QUESTION] + +ASK FORM 1 ONLY [N=4,801]: +CMPGNTRAITPOST_PROUD Would you say the 2024 presidential campaign was... + +BASED ON VOTERS [N=4,012]: + +| | | | Made you feel proud | | | Did not make you feel | | | No | | +| ---------------- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | -------- | --- | +| | | | of the country | | | proud of the country | | | answer* | | +| Nov 12-17, 2024 | | | | 29 | | | 70 | | 1 | | + +| | | TREND FOR COMPARISON: | | | | | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | +| | | So far, would you say the 2024 presidential campaign… | | | | | | | | | + +| | | BASED ON REGISTERED VOTERS: | | | | | | | | | +| --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | + +| | | | | | | Makes you feel proud | | Does not make you feel | | No | +| --- | --- | ------------------- | --- | --- | --- | --------------------- | --- | ----------------------- | --- | -------- | +| | | | | | | of the country | | proud of the country | | answer* | +| | | Sep 30-Oct 6, 2024 | | | | | 19 | | 79 | 2 | +| | | Jul 1-7, 2024 | | | | | 12 | | 87 | 1 | + +| | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +www.pewresearch.org + +8 +PEW RESEARCH CENTER + +ASK FORM 1 ONLY [N=4,801]: +CMPGNTRAITPOST_NEG Would you say the 2024 presidential campaign was... + +BASED ON VOTERS [N=4,012]: + +| | | | Too | | Not too | | No | | | | | | +| ---------------- | --- | --- | --------- | --- | --------- | --- | -------- | --- | --- | --- | --- | --- | +| | | | negative | | negative | | answer* | | | | | | +| Nov 12-17, 2024 | | | 74 | | 25 | | 1 | | | | | | + +| | | TREND FOR COMPARISON: | | | | | | | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | So far, would you say the 2024 presidential campaign… | | | | | | | | | | | + +| | | BASED ON REGISTERED VOTERS: | | | | | | | | | | | +| --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --------- | ---------- | --- | -------- | --- | +| | | | | | | | | Too | Not too | | No | | +| | | | | | | | | negative | negative | | answer* | | +| | | Sep 30-Oct 6, 2024 | | | | | | 71 | | 27 | | 2 | +Jul 1-7, 2024 +| | | | | | | | | 68 | | 30 | | 2 | +| --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | | | | | | | | +| | | 2020 TREND FOR COMPARISON:6 | | | | | | | | | | | +| | | Nov 12-17, 2020 | | | | | | 76 | | 23 | | 1 | +| | | July 27-Aug 2, 2020 | | | | | | 61 | | 37 | | 2 | + +ASK FORM 1 ONLY [N=4,801]: +CMPGNTRAITPOST_POLCY Would you say the 2024 presidential campaign was... + +BASED ON VOTERS [N=4,012]: + +| | | | Focused on important | | | Not focused on important | | | | No | | | +| ---------------- | --- | --- | --------------------- | --- | --- | ------------------------- | --------------- | --- | --- | -------- | --- | --- | +| | | | policy debates | | | | policy debates | | | answer* | | | +| Nov 12-17, 2024 | | | | 35 | | | | 64 | | 1 | | | + +| | | TREND FOR COMPARISON: | | | | | | | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | So far, would you say the 2024 presidential campaign… | | | | | | | | | | | + +| | | BASED ON REGISTERED VOTERS: | | | | | | | | | | | +| --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +| | | | | | | | | Focused on | | Not focused on | | | +| --- | --- | ------------------- | --- | --- | --- | --- | --- | --------------- | --- | --------------- | --- | -------- | +| | | | | | | | | important | | important | | No | +| | | | | | | | | policy debates | | policy debates | | answer* | +| | | Sep 30-Oct 6, 2024 | | | | | | 37 | | | 62 | 1 | +| | | Jul 1-7, 2024 | | | | | | 23 | | | 76 | 1 | + +| | | | | | | | | | | | | | +| --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | 2020 TREND FOR COMPARISON:7 | | | | | | | | | | | +| | | Nov 12-17, 2020 | | | | | | 27 | | | 72 | 1 | +| | | July 27-Aug 2, 2020 | | | | | | 25 | | | 73 | 2 | + +| | | | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +6 Question asked about the 2020 presidential campaign. +7 Question asked about the 2020 presidential campaign. +www.pewresearch.org + +9 +PEW RESEARCH CENTER + +ASK FORM 1 ONLY [N=4,801]: +CMPGNTRAITPOST_INT Would you say the 2024 presidential campaign was... + +BASED ON VOTERS [N=4,012]: + +| | | | | No | | | | | | | +| ---------------- | --- | --- | ------------ | -------------- | --- | --- | --- | --- | --- | --- | +| | | | Interesting | Dull answer* | | | | | | | +| Nov 12-17, 2024 | | | 74 | 24 2 | | | | | | | + +| | | TREND FOR COMPARISON: | | | | | | | | | +| --- | --- | ------------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | +| | | So far, would you say the 2024 presidential campaign… | | | | | | | | | + +| | | BASED ON REGISTERED VOTERS: | | | | | | | | | +| --- | --- | ---------------------------- | --- | --- | ------------ | --- | --- | ----- | --- | -------- | +| | | | | | | | | | | No | +| | | | | | Interesting | | | Dull | | answer* | +| | | Sep 30-Oct 6, 2024 | | | | 68 | | 30 | | 2 | +| | | | | | | | | | | | +| | | 2020 TREND FOR COMPARISON8 | | | | | | | | | +| | | Nov 12-17, 2020 | | | | 64 | | 34 | | 2 | +| | | July 27-Aug 2, 2020 | | | | 39 | | 59 | | 2 | + +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED + +ASK IF VOTED (VOTED=3) [N=8,072]: +| VOTE_HOW_POST | | | How did you vote in the election? [RANDOMIZE] | | | | | | | | +| -------------- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | +ASK IF VOTED IN PERSON (VOTE_HOW_POST=1) [N=5,196]: +| VOTEINPWHEN | | When did you vote? | | | | | | | | | +| ------------ | --- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | + +| Nov 12-17, | | | | | | Nov 16-27, | | | Nov 12-17, | | +| ----------- | ----- | ------------------------------ | --- | --- | --- | ----------- | ----- | --- | ----------- | --- | +| | 2024 | | | | | | 2022 | | 2020 | | +| | 65 | In person at a polling place | | | | | 65 | | 54 | | +| | 34 | On Election Day | | | | | 44 | | | 27 | +| | 32 | Before Election Day | | | | | 21 | | | 27 | +| | * | Refused | | | | | * | | | * | +| | 35 | By absentee or mail-in ballot | | | | | 35 | | 46 | | +| | * | No answer* | | | | | * | | | * | + +TREND FOR COMPARISON: +When did you vote in the elections this November? +How did you vote in the November election? + +| | | | | | | Nov 7-16, | | | | | +| --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | +2018 +| | | On election day | | | | | 55 | | | | +| --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | Before election day | | | | | 45 | | | | +| | | Voted in person | | | | | 19 | | | | +| | | Mailed in ballot | | | | | 25 | | | | +| | | No answer | | | | | * | | | | +| | | No answer | | | | | * | | | | +| | | | | | | | | | | | + +| | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +8 Question asked about the 2020 presidential campaign. +www.pewresearch.org + +10 +PEW RESEARCH CENTER +VOTE_HOW_POST/VOTEINPWHEN TREND FOR COMPARISON CONTINUED… + +PHONE TREND FOR COMPARISON: +Did you vote ON Election Day or BEFORE Election DAY? + +Did you vote in person or did you mail your ballot in? + +| | | | Nov 10-14 | Nov 6-9 | | Nov 8-11 | | Nov | | Nov | | Nov | Nov | +| ------------------------- | --- | --- | ---------- | -------- | ----- | --------- | ----- | ----- | --- | ----- | --- | ----- | ----- | +| | | | 2016 | | 2014 | | 2012 | 2008 | | 2006 | | 2004 | 2002 | +| On Election Day | | | 59 | | 71 | | 63 | 66 | | 80 | | 80 | 85 | +| Before Election Day | | | 41 | | 29 | | 37 | 34 | | 20 | | 20 | 15 | +| Voted in person | | | 22 | | 12 | | 19 | 19 | | -- | | | | +| Mailed in ballot | | | 18 | | 16 | | 17 | 14 | | -- | | | | +| Other way/DK/Ref (VOL) | | | 1 | | 1 | | * | 1 | | -- | | | | +| Don’t know/Refused (VOL) | | | 0 | | * | | 0 | * | | * | | * | * | +| | | | | | | | | | | | | | | + +ASK IF VOTED BY ABSENTEE OR MAIL (VOTE_HOW_POST=2) [N=2,868]: +VOTEMAIL How did you return your absentee or mail-in ballot? [RANDOMIZE] + +| Nov 12-17, | | | | | | | Nov 16-27, | | Nov 12-17, | | | | | +| ----------- | ----- | ------------------------------------------ | --- | --- | --- | --- | ----------- | --- | ----------- | ----- | --- | --- | --- | +| | 2024 | | | | | | 2022 | | | 2020 | | | | +| | 12 | In person to an election official or poll | | | | | | 12 | | 15 | | | | +worker +| | 38 | At a designated dropbox | | | | | | 36 | | 41 | | | | +| --- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | 50 | By mail | | | | | | 52 | | 44 | | | | +| | * | No answer* | | | | | | * | | * | | | | + +ASK IF VOTED BY ABSENTEE OR MAIL (VOTE_HOW_POST=2) [N=2,868]: +MAILWHEN When did you mail or return your absentee or mail-in ballot? + +| | Nov 12-17, | | | | | | Nov 12-17, | | | | | | | +| --- | ----------- | --------------------------------------- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +| | 2024 | | | | | | 2020 | | | | | | | +| | 33 | In the week leading up to Election Day | | | | | | 23 | | | | | | +| | 67 | Before that | | | | | | 76 | | | | | | +| | * | No answer* | | | | | | * | | | | | | +| | | | | | | | | | | | | | | + +ASK IF VOTED IN PERSON (VOTE_HOW_POST=1) [N=5,196]: +WAITVOTE When you went to vote, approximately how long did you have to wait in line to +vote? +| | | | | | | | | | | | Nov 17- | | | +| ------------ | ----- | --------------------- | --- | --- | ------------ | ----- | --- | ---------- | --- | --- | -------- | --- | --- | +| Nov 12-17, | | | | | Nov 12-17, | | | Nov 7-16, | | | Dec 15, | | | +| | 2024 | | | | | 2020 | | 2018 | | | 2014 | | | +| | 42 | Did not wait at all9 | | | | 35 | | 47 | | | | 53 | | +| | 30 | Less than 10 minutes | | | | 27 | | 33 | | | | 30 | | +| | 17 | 10 to 30 minutes | | | | 20 | | 14 | | | | 13 | | +| | 7 | 31 minutes to 1 hour | | | | 11 | | | 5 | | | 3 | | +| | 4 | More than 1 hour | | | | 6 | | | 1 | | | * | | +| | * | No answer* | | | | * | | | * | | | * | | + +| | | | | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +9 In surveys prior to November 2024, the response was worded “Not at all”. +www.pewresearch.org + +11 +PEW RESEARCH CENTER + +ASK IF VOTED BY MAIL OR ABSENTEE (VOTE_HOW_POST=2) [N=2,868]: +MAILFIRST Is this the first time you have ever voted by mail or absentee ballot? + +| Nov 12-17, | | | Nov 12-17, | | | +| ----------- | --- | --- | ----------- | --- | --- | +| 2024 | | | 2020 | | | +Yes, this is the first time I have ever voted by +| 12 | | | 39 | | | +| --- | --- | --- | --- | --- | --- | +mail or absentee ballot +| 88 | No, I have voted by mail or absentee ballot before | | 61 | | | +| --- | --------------------------------------------------- | --- | --- | --- | --- | +| * | No answer* | | * | | | + +ASK IF VOTED IN PERSON BEFORE ELECTION DAY (VOTEINPWHEN=1) [N=2,671]: +| EARLYFIRST | Is this the first time you have ever voted early? | | | | | +| ----------- | -------------------------------------------------- | --- | --- | --- | --- | + +Nov 12-17, +| 2024 | | | | | | +| ----- | --- | --- | --- | --- | --- | +29 Yes, this is the first time I have ever voted early +71 No, I have voted early before +* No answer* + +ASK IF VOTED IN PERSON ON ELECTION DAY (VOTEINPWHEN=2) [N=2,513]: +ELECDFIRST Is this the first time you have ever voted in person on Election Day? + +Nov 12-17, +2024 +Yes, this is the first time I have ever voted in person on Election Day +7 +93 No, I have voted in person on Election Day before +* No answer* + +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED + +ASK IF VOTED (VOTED=3) [N=8,072]: +VTCOUNT_OWN How confident are you that your vote was accurately counted? + +| Nov 12-17, | | | Nov 16-27, | Nov 12-17, | | +| ----------- | ------------------- | --- | ----------- | ----------- | --- | +| 2024 | | | 2022 | 2020 | | +| 59 | Very confident | | 57 | 59 | | +| 33 | Somewhat confident | | 31 | 26 | | +| 5 | Not too confident | | 8 | 9 | | +Not at all confident +| 2 | | | 3 | 6 | | +| --- | ----------- | --- | --- | --- | --- | +| * | No answer* | | * | * | | + +| | PHONE TREND FOR COMPARISON: | | | | | +| --- | ---------------------------- | --- | --- | --- | --- | + +| | | Nov 10-14 Nov 6-9 Nov 8-11 Nov 4-7 | | Nov Nov | Nov | +| --- | -------------------------- | ----------------------------------- | ----------- | ----------- | ----- | +| | | 2016 2014 | 2012 2010 | 2008 2006 | 2004 | +| | Very confident | 71 68 | 68 64 | 73 70 | 68 | +| | Somewhat confident | 19 21 | 22 26 | 22 23 | 24 | +| | Not too confident | 4 5 | 5 4 | 3 3 | 4 | +| | Not at all confident | 5 4 | 3 3 | 2 2 | 3 | +| | Don’t know/Refused (VOL.) | 1 1 | 1 3 | * 2 | 1 | +| | | | | | | +www.pewresearch.org + +12 +PEW RESEARCH CENTER +[RANDOMIZE ORDER OF VTCOUNT_POST_INP AND VTCOUNT_POST_ABS] + +ASK ALL: +VTCOUNT_POST_INP How confident are you that votes cast in person at polling places across the United +States were counted as voters intended in the elections this November? + +| Nov 12-17, | | Nov 16-27, | | Nov 12-17, | | | +| ----------- | --------------------- | ----------- | ----- | ----------- | ----- | --- | +| 2024 | | | 2022 | | 2020 | | +| 38 | Very confident | | 37 | | 42 | | +| 47 | Somewhat confident | | 43 | | 35 | | +| 11 | Not too confident | | 13 | | 13 | | +| 4 | Not at all confident | | 6 | | 8 | | +| 1 | No answer* | | 1 | | 1 | | + +TREND FOR COMPARISON: +How confident are you that votes cast in person at polling places across the United States will +be counted as voters intend in the elections this November? + +| | | | Sep 30- | | | Sep 30- | +| --- | --------------------- | --- | ------- | --- | ----------- | -------- | +| | | | Oct 6, | | Oct 10-16, | Oct 5, | +| | | | 2024 | | 2022 | 2020 | +| | Very confident | | 36 | | 34 | 42 | +| | Somewhat confident | | 46 | | 45 | 44 | +| | Not too confident | | 13 | | 15 | 10 | +| | Not at all confident | | | 5 | 5 | 4 | +| | No answer* | | | 1 | 1 | 1 | + +[RANDOMIZE ORDER OF VTCOUNT_POST_INP AND VTCOUNT_POST_ABS] + +ASK ALL: +VTCOUNT_POST_ABS How confident are you that votes cast by absentee or mail in ballot across the United +States were counted as voters intended in the elections this November? + +| Nov 12-17, | | Nov 16-27, | | Nov 12-17, | | | +| ----------- | --------------------- | ----------- | ----- | ----------- | ----- | --- | +| 2024 | | | 2022 | | 2020 | | +| 24 | Very confident | | 25 | | 30 | | +| 46 | Somewhat confident | | 37 | | 27 | | +| 21 | Not too confident | | 23 | | 19 | | +| 7 | Not at all confident | | 13 | | 23 | | +| 1 | No answer* | | 1 | | 1 | | + +TREND FOR COMPARISON: +How confident are you that votes cast by absentee or mail-in ballot across the United States +will be counted as voters intend in the elections this November? + +| | | | Sep 30- | | | Sep 30- | +| --- | --------------------- | --- | ------- | --- | ----------- | -------- | +| | | | Oct 6, | | Oct 10-16, | Oct 5, | +| | | | 2024 | | 2022 | 2020 | +| | Very confident | | 22 | | 21 | 17 | +| | Somewhat confident | | 38 | | 38 | 38 | +| | Not too confident | | 27 | | 24 | 28 | +| | Not at all confident | | 12 | | 16 | 16 | +| | No answer* | | | 1 | 1 | 1 | + +www.pewresearch.org + +13 +PEW RESEARCH CENTER +ASK ALL: +ELECT_CONFNOVPOS How confident, if at all, are you that... [RANDOMIZE ITEMS] + +| | | | | | | Very | Somewhat | | Not too | | Not at all | No | +| ------ | --- | -------------------------------- | --- | --- | ---------- | ----- | ---------- | --- | ---------- | --- | ----------- | -------- | +| | | | | | confident | | confident | | confident | | confident | answer* | +| PRVFR | | People who were not legally | | | | | | | | | | | +| | | qualified to vote are prevented | | | | | | | | | | | +from casting a ballot +| | | Nov 12-17, 2024 | | | | 27 | | 38 | 26 | | 8 | 1 | +| ------- | -------------------------------- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | | | | | | | | +| | TREND FOR COMPARISON10: | | | | | | | | | | | | +| | | Sep 30-Oct 6, 2024 | | | | 29 | | 30 | 24 | | 17 | 1 | +| | | Sep 30-Oct 5, 2020 | | | | 23 | | 36 | 28 | | 11 | 1 | +| | | | | | | | | | | | | | +| PRVSUP | | People who were legally | | | | | | | | | | | +| | | qualified and want to vote are | | | | | | | | | | | +able to cast a ballot +| | | Nov 12-17, 2024 | | | | 45 | | 42 | 9 | | 3 | 1 | +| --- | -------------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| | | | | | | | | | | | | | +| | TREND FOR COMPARISON11: | | | | | | | | | | | | +| | | Sep 30-Oct 6, 2024 | | | | 44 | | 43 | 10 | | 3 | 1 | +| | | Sep 30-Oct 5, 2020 | | | | 36 | | 45 | 14 | | 4 | 1 | +| | | | | | | | | | | | | | + +ADDITIONAL QUESTIONS PREVIOUSLY RELEASED + +ASK ALL: +| PARTY | | In politics today, do you consider yourself a… | | | | | | | | | | | +| ---------------- | --- | ------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ASK IF INDEP/SOMETHING ELSE (PARTY=3, 4 OR REFUSED): +| PARTYLN | | As of today do you lean more to…12 | | | | | | | | | | | +| -------------- | --- | ----------------------------------- | --- | --- | --- | ---------- | --- | --- | --- | ----- | ----- | --- | +| | | | | | | Something | | No | | Lean | Lean | | + Republican Democrat Independent else answer* Rep Dem +| | | 28 | 27 | 30 | | 14 | | 1 | | 20 | | 20 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +10 In October 2024 and October 2020, the item was worded “People who are not legally qualified to vote are prevented from +casting a ballot.” +11 +In October 2024 and October 2020, the item was worded “People who are legally qualified and want to vote are able to cast +a ballot.” +12 PARTY and PARTYLN asked in a prior survey. +www.pewresearch.org \ No newline at end of file diff --git a/q132/random_k2/question.json b/q132/random_k2/question.json index b35746a31cd4197d5f2b52eea1a074f9022bde6a..20c941fbd0dbf88800714071004dba21a7ed93e1 100644 --- a/q132/random_k2/question.json +++ b/q132/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "ISEP_student_handbook_2020.pdf", - "web_476fa6510ac4328b.pdf", - "web_8be0b957e80202c1.pdf" + "web_55eb3e1bfd6dcec7.pdf", + "web_1a2336a3f9985b16.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q132/random_k2/random_1.md b/q132/random_k2/random_1.md index 27dec34a4c7b44fe87d06889c67f66b0b5fdeff9..7d30cced73b1f30da245c4cf6780b4baf04ace71 100644 --- a/q132/random_k2/random_1.md +++ b/q132/random_k2/random_1.md @@ -1,585 +1,553 @@ -| M | U S | I | C | | -| --- | --- | --- | ---- | ---- | -| C O | N | S | U M | E R | -| I N | S | I G | H T | | -| R E | P | O | RT | | -2018 + GUI DE + +Social Media +Marketing Strategy +Eight easy steps to develop your +social media presence + + Contents + +Step 1: Set social media marketing goals that align to business objectives + +Step 2: Learn everything you can about your audience + +Step 3: Research the competition + +Step 4: Conduct a social media audit + +Step 5: Set up accounts and improve existing profiles + +Step 6: Find inspiration + +Step 7: Create a social media content calendar + +Step 8: Test, evaluate, and adjust your strategy 3 -C O N T E N T S I N T R O D U C T I O N -MUSIC IS AN INTEGRAL PART OF OUR LIVES -SECTION PG -The Music Consumer Insight increasing the availability of music, record -0 1 -MUSIC CONSUMPTION IN 2018 04 companies have licensed over 45 million -Report tells the story of how -tracks to hundreds of digital services around -recorded music is woven into the world. -0 2 -MUSIC IS AN INTEGRAL PART OF OUR DAILY LIVES 06 -the lives of people around However, for music to thrive in a digital -0 3 world there must be a fair digital -THE WORLD’S FAVOURITE GENRES 08 the world. -marketplace. This report also shows the -We are passionate about music. It is challenges the music community continues -0 4 MUSIC DRIVES TECHNOLOGY ENGAGEMENT 10 personal to us. Increasingly accessible, to face – both in the form of the evolving -music is embraced across genres and threat of copyright infringement and in fair -geographies, ages and formats - from revenues not being returned by some user- -0 5 ON-DEMAND STREAMING LEADS MUSIC CONSUMPTION GLOBALLY 12 -radio to streaming and beyond. upload services. -0 6 UNLICENSED MUSIC 14 This report looks at how music soundtracks Music unites us globally and adds enormous -the many parts of our day and how value to people’s lives. Record companies -this love of music is also driving fans’ are essential to this as they continue to -0 7 COUNTRY FOCUS: CHINA 16 growing adoption of technologies. From develop, support and invest in music, -smartphones to smart speakers, music playing a crucial role in ensuring that it -is a force ushering in these connected continues on its exciting journey around -0 8 COUNTRY FOCUS: INDIA 18 -devices that are being taken up globally. the world. -As ever, local repertoire continues to "RECORD COMPANIES -METHODOLOGY dominate countries' charts. There will always CONTINUE TO DEVELOP, -IFPI conducted global research in April-May 2018 which explored the way consumers engage and access be something special about local music that -SUPPORT AND INVEST IN MUSIC, -music across licensed and unlicensed services. speaks to us. In this interconnected world, -PLAYING A CRUCIAL ROLE IN -country-specific genres, like K-Pop in Korea -The fieldwork was carried out by AudienceNet amongst a demographically representative sample of the ENSURING THAT IT CONTINUES -and música popular brasileira in Brazil, are -online population aged 16-64 in the following territories: Argentina, Australia, Brazil, Canada, France, -ON ITS EXCITING JOURNEY -not only embraced at home but are also -Germany, Italy, Japan, Mexico, Netherlands, Poland, Russia, South Africa, South Korea, Spain, Sweden, -beginning to find a broader global audience. AROUND THE WORLD." -United Kingdom and United States. In addition, the study was also conducted in China and India but results -from these two countries are not included in “global” figures. In each country, nationally representative -In this year’s report, for the first time, we FRANCES MOORE | CHIEF EXECUTIVE, IFPI -quota samples of between 1,000-2,000 respondents were set in accordance with online population size and -take a close look at the exciting, evolving -demographic structure, as determined by the latest respective census data in each territory. This ensured -music markets in China and India. In both -that a standard error of +/- 3% was achieved throughout the data, at a 95% confidence level. -countries, music fans are highly engaged -These twenty territories accounted for 91.3% of global recorded music market revenues in 2017. with licensed music and local music is -flourishing. -For more about IFPI visit www.ifpi.org -Across the globe, record companies are -working to sustain and develop these rich -and diverse ways in which music is being -enjoyed. Driving digital innovation and -© IFPI 2018 All data, copy and images are subject to copyright. -MUSIC CONSUMER INSIGHT REPORT 2018 | - -1 5 -M U S I C C O N S U M P T I O N -I N 2 0 1 8 -Based on research conducted by IFPI in 2018, this report provides -a snapshot of how consumers across 18 of the world’s leading On average consumers spend -music markets are engaging with recorded music. -17.8 hrs -listening to music each -week globally -2.5 hrs -a day -86% 50% -of consumers are listening to music through of 16-24s would choose audio streaming if -on-demand streaming there were only one way to listen to music -75% 38% -of consumers use smartphones to listen of consumers obtain music through -to music copyright infringement -Base: All participants (n=19,000) from the 18 countries surveyed (Argentina, Australia, Brazil, Canada, France, -Germany, Italy, Japan, Mexico, Netherlands, Poland, Russia, South Africa, South Korea, Spain, Sweden, UK, US) MUSIC CONSUMER INSIGHT REPORT 2018 | - -2 77 -| M U | S I C | I S | A N | I N T | E G | R A L | | | | | | -| --- | ----- | ----- | ----- | ------- | ----- | ------- | --- | --- | --- | --- | --- | -| P A | R T | O F | O U R | D A | I L Y | L I V | E S | | | | | -Consumers are embracing music at all points of the day, -demonstrating the importance and value that it has in our lives. -| 17.8 | | hrs | | CONSUMERS MOST TYPICALLY | | | | | YOUNGER CONSUMERS ARE | | | -| ---- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | --------------------------- | --- | --- | -| | | | | LISTEN TO MUSIC WHILE: | | | | | LISTENING TO MORE MUSIC IN | | | -MORE WAYS -spent listening to music Younger consumers (16-24s) are more -each week globally In the car Going to sleep likely to listen to music during any activity -and much more likely to listen on their -| | | | | | | 66% | (global) | | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -19% (global) -way to work or education or while at -work or education. -South Africa -Brazil -80% -33% -| | | | | | Germany | | | Poland | | | | -| --- | --- | --- | --- | --- | ------- | --- | --- | ------ | --- | --- | --- | -77% -29% 72% -69% -US -South Africa -75% -25% -Commuting to work or education -| | | | | | | 54% | (global) | Relaxing at home | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ---------------- | --- | --- | --- | -63% (global) -Argentina -30% -62% -Cooking and cleaning -Mexico 54% (global) -62% -Germany -| | | | | | | | 60% | Exercising or at the gym | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | -36% (global) -| | | | | | | | | | Commuting | Relaxing | Going | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --------- | -------- | ----- | -Working or studying -| | | | | | | | | | to work or | at home | to sleep | -| --- | --- | --- | --- | --- | --- | ------------ | --- | ------------------------------- | ----------- | ------- | --------- | -| | | | | | | 40% (global) | | At concerts, gigs and festivals | education | | | -36% (global) -Mexico -68% -Argentina -58% -South Africa -57% -Base: All participants (n=19,000) from all -| 18 countries surveyed | | | | | | | | | MUSIC CONSUMER INSIGHT REPORT 2018 | | | | -| --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | - -3 9 -| T H | E W | O | R L D | ’ S | | | | | | | | | | | | -| ----- | ----- | --- | ----- | ---- | --- | ----- | --- | ----- | --- | ----- | --- | ------ | --- | --- | --- | -| F A V | O U | R | I T E | G | E N | R E S | | | | | | | | | | -| | | | | | | | | JAPAN | | KOREA | | FRANCE | | | | -POLAND -WHAT MUSIC DO PEOPLE TYPICALLY LISTEN TO? LOCAL MUSIC IS EMBRACED BY CONSUMERS -HERE WE TAKE A LOOK AT THE TOP TEN GENRES -Local culture influences consumers' listening habits, with many enjoying domestic genres. -| | | | | | | | | 66% | | 62% | | 69% | | | 28% | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | --- | ------------ | --- | --- | ------------ | -| 01 | | | 02 | | | 03 | | JJAAPPAANN | | KKOORREEAA | | FFRRAANNCCEE | | | PPOOLLAANNDD | -% % % Two-thirds of 62% of consumers In France, 69% In Poland, 28% -| | | | | | | | | consumers in Japan | | in Korea like K-Pop | | listen to Variété | | | listen to Disco Polo | -| ---- | --- | --- | ---- | --- | --- | ---- | --- | --------------------------- | --- | ------------------------- | --- | ------------------ | --- | --- | -------------------- | -| | 64 | | | 57 | | | 32 | | | | | | | | | -| | | | | | | | | l66isten66 to J%%-Pop with | | 66(and22 22%%% listen to | | 66Fran99çaise%% | | | 2288%% | -| | | | | | | | | 29% listening to | | K-Trot) | | | | | | -| 0011 | | | 0022 | | | 0033 | | | | | | | | | | -music from anime -Dance/Electronic/ -| Pop | | | Rock | | | | | | | | | | | | | -| --- | --- | --- | ---- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | House | %% | | | | | | | | | -%% %% TTwwoo--tthhiirrddss ooff 6622%% ooff ccoonnssuummeerrss IInn FFrraannccee,, 6699%% IInn PPoollaanndd,, 2288%% -| | | | | | | | | ccoonnssuummeerrss iinn JJaappaann | | iinn KKoorreeaa lliikkee KK--PPoopp | | lliisstteenn ttoo VVaarriiééttéé | | | lliisstteenn ttoo DDiissccoo PPoolloo | -| --- | ---- | --- | --- | ---- | --- | --- | ---- | -------------------------------------- | --- | ---------------------------------------- | --- | ------------------------------------ | --- | --- | ---------------------------------------- | -| | 6644 | | | 5577 | | | 3322 | | | | | | | | | -04 05 06 lliisstteenn ttoo JJ--PPoopp wwiitthh ((aanndd 2222%% lliisstteenn ttoo FFrraannççaaiissee -| | | | | | | | | 2299%% lliisstteenniinngg ttoo | | KK--TTrroott)) | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | -------------- | --- | --- | --- | --- | --- | -mmuussiicc ffrroomm aanniimmee -| | | % | | | % | | % | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -PPoopp RRoocckk DDaannccee//EElleeccttrroonniicc// IN ARGENTINA IN MEXICO IN BRAZIL -| | | | | | | HHoouussee | 24 | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | 30 | | | 26 | | | | | | | | | | | | -0044 0055 0066 LATIN AMERICAN CONSUMERS ARE ENGAGED WITH LOCAL GENRES -| Soundtracks | | | Hip-Hop/ | | | Singer/ | | | | | | | | | | -| ------------------------ | ---- | --- | ---------------- | ---- | --- | -------------------- | ---- | --- | ------------------------ | --- | ------------------ | --- | --- | ------------------ | --- | -| (film or TV) | | %% | Rap/Trap | | %% | Songwriter | %% | | | | | | | | | -| | | | | | | | | | IINN AARRGGEENNTTIINNAA | | IINN MMEEXXIICCOO | | | IINN BBRRAAZZIILL | | -| | 3300 | | | 2266 | | | 2244 | | | | | | | | | -| 07 | | | 08 | | | 09 | | | 43% | | 55% | | | 55% | | -| | | % | | | % | SSiinnggeerr// | % | | | | | | | | | -| SSoouunnddttrraacckkss | | | HHiipp--HHoopp// | | | | | | | | | | | | | -| ((fifillmm oorr TTVV)) | 24 | | RRaapp//TTrraapp | 23 | | SSoonnggwwrriitteerr | 22 | | | | | | | | | -| | | | | | | | | | Latin music | | Latin music | | | Música popular | | -brasileira -| 0077 | | | 0088 | | | 0099 | | | 4433%% | | 5555%% | | | | 5555%% | -| --------- | --- | --- | ---- | --- | --- | ---------- | --- | --- | ------ | --- | ------ | --- | --- | --- | ------ | -| Classical | | | R&B | | | Soul/Blues | | | | | | | | | | -(including Opera) -| | | %% | | | %% | | %% | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -2222 -2244 2233 LLaatt3iinn6 mm%uussiicc LLaa4ttiinn2 mm%uussiicc 39%MMúússiiccaa ppooppuullaarr3 4% + +4 + +5 + +6 + +7 + +8 + +8 + +10 + +GUIDE / Social Media Marketing Strategy + +#239584107 + +2 + + Related resource + +The Social Media Metrics +that Really Matter—and +How to Track Them + +Step 1 + +Set social media marketing goals that align to +business objectives + +The more specific your strategy is, the more effective the execution will be. +Set SMART goals and track the right metrics to set yourself up for success. + +Set SMART goals + +The first step to creating a social media marketing strategy is to establish +your objectives and goals. Without goals, you have no way to measure your +success or your social media return on investment (ROI). + +Each of your goals should be: + +Specific + +Measurable + +Attainable + +Relevant + +Time-bound + +An example of a SMART goal for your business might be "Grow our Instagram +audience by 50 new followers per week." + +With SMART goals, you’ll make sure your goals actually lead to real business +results, rather than just lofty ideals. + +GUIDE / Social Media Marketing Strategy + +3 + + Track meaningful metrics + +While vanity metrics like retweets and likes can be fun to share and easy to +track, it’s hard to prove their real value for your business. Instead, focus on +targets such as leads generated, web referrals, and conversion rate. + +You may want to track different goals for different channels, or even different uses +of each channel. For example, you can use paid campaigns to increase brand + awareness, but measure acquisition and engagement for organic social posts. + +Make sure to align your social media goals with your overall marketing +strategy. This will make it easier for you to show the value of your work and +get executive buy-in and investment. + +Start developing your social media marketing plan by writing down at least +three social media goals. + +Goals + +1. + +2. + +3. + +Step 2 + +Related resource + +How to build audience +personas + +Learn everything you can about your audience + +if you’re not engaged in social media listening, you’re creating your business +strategy with blinders on—and you’re missing out on mountains of actionable +insights from real people who are actively talking about you or your industry online. + +Here’s how to start listening and building your understanding of your +audience and their needs. + +Create audience personas + +tKnowing who your audience is and what they want to see on social is key to +creating content that they will like, comment on, and share. This knowledge +also critical for planning how to develop your social media fans into +customers for your business. + +Try creating audience personas. For example, a retail brand might create +different personas based on demographics, buying motivations, common +buying objections, and the emotional needs of each type of customer. + +GUIDE / Social Media Marketing Strategy + +4 + + Personas sharpen your marketing tactics. Luxury buyers, for example, +might not respond to Facebook ads with sales. But they might respond to +Facebook ads with exclusive in-store events to be the first to see a new line of +clothing. With personas, you’ll have the customer insights you need to create +campaigns that speak to the real desires and motivations of your buyers. + +Gather real-world data + +Don’t make assumptions. Social media analytics can also provide a ton of +valuable information about who your followers are, where they live, which +languages they speak, and how they interact with your brand on social. These +insights allow you to refine your strategy and better target your social ads. + +Jugnoo, an Uber-like service for auto-rickshaws in India, used Facebook +Analytics to learn that 90 percent of their users who referred other +customers were between 18 and 34 years old, and that 65 percent of that +group was using Android. They used this information to target their ads, +resulting in a 40 percent lower cost per referral. + +Step 3 + +Research the competition + +Odds are, your competitors are already using social media—and that means +you can learn from what they’re already doing. + +Conduct a competitive analysis + +A competitive analysis allows you to understand who the competition is and +what they’re doing well (and not so well). You’ll get a good sense of what’s +expected in your industry, which will help you set some social media targets +of your own. + +This analysis will also help you spot opportunities. For example, maybe +one of your competitors dominates on Facebook, but has put little effort +into Twitter or Instagram. You might want to focus on the networks where +your audience is underserved, rather than trying to win fans away from a +dominant player. + +Engage in social listening + +Social listening is another way to keep track of the competition. + +As you track your competitors’ accounts and relevant industry keywords, you +may notice strategic shifts in the way competitors use their social accounts. +Or you might spot a specific post or campaign that really hits the mark—or +one that bombs. + +Keep an eye on this information and use to it evaluate your own goals and plans. + +Related resources + +How to conduct a +competitor audit + +Getting started with +social listening + +Watch: How to set up +social listening streams + +GUIDE / Social Media Marketing Strategy + +5 + + Related resource + +Hootsuite’s social media +audit template + +Step 4 + +Conduct a social media audit + +Conducting a social media audit helps you assess how well your current +social media use works for you. + +Evaluate your current efforts + +If you’re already using social media tools, you need to take a step back and look +at what you’ve already accomplished. Ask yourself the following questions: + +•• What’s working? + +•• What’s not working? + +•• Who’s connecting with you on social? + +•• Which social media sites does your target market use? + +•• How does your social media presence compare to that of your competitors? + +Once you gather all this information in one place, you’ll have a good starting +point for planning how to improve your results. + +Your audit should give you a clear picture of what purpose each of your +social accounts serves. If the purpose of an account isn’t clear, think about +whether it’s worth keeping. It may be a valuable account that just needs +a strategic redirection, or it may be an outdated account that’s no longer +worth your while. + +To help you decide, ask yourself the following questions: + +1. Is my audience here? + +2. If so, how are they using this platform? + +3. Can I use this account to help achieve meaningful business goals? + +Asking these tough questions now will help keep your social media strategy +on track as you grow your social presence. + +Look for impostor accounts + +During your audit process, you may discover fraudulent accounts using your +business name or the names of your products—that is, accounts that you +and your business don’t own. + +These imposter accounts can be harmful to your brand (never mind +capturing followers that should be yours), so be sure to report them. You +may want to get your social accounts verified to ensure your fans and +followers know they are dealing with the real you. + +GUIDE / Social Media Marketing Strategy + +6 + + Related resources + +Watch: How to optimize +your social profiles + +We’ve also got step-by-step +guides for each network +to walk you through the +process: + +Create a Facebook +business page + +Create an Instagram +business account + +Create a Twitter business +account + +Create a Snapchat +account + +Create a LinkedIn +Company Page + +Create a Pinterest +business account + +Create a YouTube channel + +Create a WeChat business +account + +Don’t let this list overwhelm +you. Remember, it’s better to +use fewer channels well than +to stretch yourself thin trying +to maintain a presence on +every social network. + +Finally, here’s a quick +reference guide for image +sizes for every network. + +Step 5 + +Set up accounts and improve existing profiles + +Decide which networks you’ll focus on, and then set up and optimize your +accounts. + +Determine which networks to use (and how to use them) + +As you decide which social channels to use, you’ll also need to define your +strategy for each network. For example, you might decide to use Twitter for +customer service, Facebook for customer acquisition, and Instagram for +engaging existing customers. + +It’s a good exercise to create mission statements for each network. These +one-sentence declarations will help you focus on a very specific goal for +each account on each social network. + +For example, you could decide that: + +•• Facebook is best for acquiring new customers via paid advertising. + +•• + +Instagram is where you build brand affinity with existing customers. + +•• Twitter is where you engage press and industry influencers. + +•• LinkedIn is where you engage existing employees and attract new talent. + +•• YouTube is where you support existing customers with education and + +video help content. + +•• Snapchat is where you distribute content with the goal of building brand + +awareness with younger consumers. + +If you can’t create a solid mission statement for a particular social network, +you may want to reconsider whether that network is worth it. + +Set up (and optimize) your accounts + +Once you’ve decided which networks to focus on, it’s time to create your +profiles—or improve existing profiles so they align with your strategic plan. + +In general, make sure you fill out all profile fields, use keywords people will +use to search for your business, and use images that are correctly sized for +each network. + +GUIDE / Social Media Marketing Strategy + +7 + + Step 6 + +Find inspiration + +While it’s important that your brand be distinctive and unique, you can still +draw inspiration from other businesses that are great on social. + +Social network success stories + +All of the social networks feature success stories that highlight how brands +are using their tools effectively. You can usually find these on the business +section of the social network’s website. (For example, take a look at the +Facebook business success stories.) These case studies offer valuable +insights you can apply to your own goals for each social network. + +Award-winning accounts and campaigns + +For examples of brands that are at the top of their social media game, check +out the winners of The Facebook Awards or The Shorty Awards. + +Step 7 + +Create a social media content calendar + +Sharing great content is essential, of course, but it’s equally important to have +a plan in place for when you’ll share content to get the maximum impact. + +Your social media content calendar also needs to account for the time you’ll +spend interacting with your audience (although you need to allow for some +spontaneous engagement as well). + +Create a posting schedule + +Your social media content calendar lists the dates and times at which you will +publish types of content on each channel. It’s the perfect place to plan all of your +social media activities—from images and link sharing to blog posts and videos. + +Your calendar ensures your posts are spaced out appropriately and +published at the optimal times. It should include both your day-to-day posts +and your content for social media campaigns. + +Related resources + +How to create a social +media content calendar + +Watch: How to save time +with bulk scheduling + +GUIDE / Social Media Marketing Strategy + +8 + + Plot your content mix + +Make sure your calendar reflects the mission statement you’ve assigned to +each social profile, so that everything you post is working to support your +business goals. For example, you might decide that: + +•• 50 percent of content will drive traffic back to your blog + +•• 25 percent of content will be curated from other sources + +•• 20 percent of content will support enterprise goals (selling, lead + +generation, and so on) + +•• 5 percent of content will be about your employees and company culture + +Placing these different post types in your content calendar will help ensure +you maintain the ratio you’ve planned. If you’re starting from scratch and +you’re simply not sure what types of content to post, try the 80-20 rule: + +•• 80 percent of your posts should inform, educate, or entertain your audience + +•• 20 percent can directly promote your brand + +You could also try the social media rule of thirds: + +•• One-third of your social content promotes your business, converts + +readers, and generates profit + +•• One-third of your social content shares ideas and stories from thought + +leaders in your industry or like-minded businesses + +•• One-third of your social content involves personal interactions with your + +audience + +Once you have your calendar set, use scheduling tools or bulk scheduling to +prepare your posting in advance rather than updating constantly throughout +the day. This allows you to focus on crafting the language and format of your +posts, rather than writing them on the fly whenever you have time. + +GUIDE / Social Media Marketing Strategy + +9 + + Step 8 + +Test, evaluate, and adjust your strategy + +Your social media strategy is a hugely important document for your +business, and you can’t assume you’ll get it exactly right on the first try. + +As you start to implement your plan and track your results, you may find +that some strategies don’t work as well as you’d anticipated, while others are +working even better than expected. + +Track your data + +In addition to using the analytics tools available within each social network, +use UTM parameters to track visitors as they move through your website, so +you can see exactly which social posts drive the most traffic to your website. + +Re-evaluate, test, and do it all again + +When data starts coming in, use it to reevaluate your strategy regularly. +You can also use this information to test different posts, campaigns, and +strategies against one another. Constant testing allows you to understand +what works and what doesn’t, so you can refine your strategy in real time. + +Surveys can also be a great way to find out how well your strategy is working. +Ask your social media followers, email list, and website visitors whether +you’re meeting their needs and expectations on social media. You can even +ask them what they’d like to see more of—and then make sure to deliver on +what they tell you. + +Things change fast on social media. New networks emerge, while others +go through significant demographic shifts. Your business will go through +periods of change as well. All this means that your social media strategy +should be a living document that you look at regularly and adjust as needed. +Refer to it often to keep you on track, but don’t be afraid to make changes +so that it better reflects new goals, tools, or plans. + +When you update your social strategy, make sure to let everyone on your +social team know, so they can all work together to help your business make +the most of your social media accounts. + +GUIDE / Social Media Marketing Strategy + 10 -bbrraassiilleeiirraa -| CCllaassssiiccaall | | % | RR&&BB | | | SSoouull//BBlluueess | | | | | | | | | | -| ------------------ | --- | --- | ------ | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -((iinncclluuddiinngg OOppee1rraa))9 Reggaeton Regional Sertanejo Samba -| | | | | | | | | | | | Mexican | | Universitário | | Pagode | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | --- | ------- | --- | ------------- | --- | ------ | -| | | | | | | | | | 3366%% | | 4422%% | | 3399%% | | 3344%% | -1100 -Metal -%% -1199 -| | | | | | | | | | RReeggggaaeettoonn | | RReeggiioonnaall | | SSeerrttaanneejjoo | | SSaammbbaa | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | ------------------ | --- | -------------------------- | --- | ------------ | -| | | | | | | | | | | | MMeexxiiccaann | | UUnniivveerrssiittáárriioo | | PPaaggooddee | -MMeettaall -Base: All participants (n=19,000) from all 18 countries surveyed. -Genre data is based on the participants’ own definitions of genre. MUSIC CONSUMER INSIGHT REPORT 2018 | - -4 11 -| M U | S | I C | | D R | I V | E | S T | E | C | H | N O | L O G | Y | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Record companies have been working behind the scenes to make it possible -SRMecAoRrdT cSoPmEApaKnEiResS have been working behind the scenes to make it possible -fRoerc foarnds ctoom apccaensies st hheaivr efa bveoeunr iwteo trrkaincgks b tehhroinudg thh sem scaernt esps etoa kmerask. e it possible -RRfoeercc oforadrnd cs oc mtoopm aapnciaceesn shiesa svt ehh ebaievre efn a bwveooeruknrin iwtge ob trerkhaiinncgdk stb hteehh sriconeudnge thsh tesom shceaelprnt m essap keteoa i ktm eproassk.s eib liet possible -RfRoReerecc cfooaorrnddrds cc tcoooomm mappcpaacaennnsiieesiess ts hhh haeaaivvrve efe abb bveeoeeeuennnr iwwt weoo otrrkkrrakiinnicnggkg sbb bteehehhrhiionninudddg tt hhht heese m ss ccsaceerenntne esessp stte ootao kmm emaarsakk.ke ee iit ti tpp pooosssssiisbbiblleele -| E N | G | A | G | E M | E N | T | | | | | | | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -ffoorr cfoannsus mtoer as ctoc eacscse tshs ethire fira fvaovouurrititee ttrraacckks su tsihnrgo vuogiche -samctaivratt esdp semaakretr ssp. eakers. -| | | | | | | | | | | | | | | Affoofsorr trff haafaenn nsnss ett otowo a a tacceccccceehessnsssos tt lhthohegeeiiyrri rfcf aafoavvnvootouuinurruiirttieeetes tt trrtaoarac cgckkrksso s ttw hhth rraoorrouuouggughhnh dss mms tmhaaearr ttrw tss oppspreeldeaaa kkwkeeeerrss rl..so .o | | | | | | | | | | | | | k at the | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --------- | -As the new technology continues to grow around the world we look at the -pArso tfihlee onfe aw ttyepcihcanlo slomgayr ct osnpteiankueers utsoe gr.row around the world we look at the -As engagement continues to grow around the world, we look at the profile -Aprso tfihlee onfe wa ttyepcihcnaol slomgay rcto snpteinaukeers utose gr.row around the world we look at the -AopAAfrss oas tt fiththylheeepe inonc nafee lew waswm ttt yaetepercctcihh cshnnpaneool aosllookmloeggrgayy uy rccs tecoo rson.npntteiitnnainuukueeeessr s utt ootso e gg rgrr.oorowww aa arroorouuunnnddd tt hhtheee ww wooorrllrddld ww weee llo olooookkk aa att ttt hhtheee -profile of a typical smart speaker user. -From smartphones to smart speakers, across the world connected ppprroorofififilleele oo off faa a tt ytyypppiiccicaaall lss smmmaaarrttr t ss psppeeeaaakkkeeerr ruu usseseerr..r. -Compared to all consumers, -devices are a growing part of the listening experience. -Compared to all consumers, -smart speaker users are: -Compared to all consumers, -Csmomarpt asrpeeda tkoe ra ulls ceorsn saurem: ers, -CCsCmooommamrpptpaa sarrpeereeddad ttk ooteo raa aulllll slcc ecooronsn nssauusrummem:ee errssr,,s, -Record companies have licensed music across hundreds of digital -smart speaker users are: -ssMmsmomsaat alrrikttre tssl ypsp 2pee5eaa-3akk4keeerr ruu usseeserrssr saa arreere:: : -15% -Most likely 25-34 -music services allowing consumers to have easier access to the music 15% -MMoosstt lliikkeellyy t2o5 -li3st4en to Hip-Hop & Rap/ -15% -| | | | | | | | | | | | | | | | 1 | 5 | % | | | | MDMM Most likely 25-34 Maooo | nsoss tttc s etlll iii kkk lMieee | lllue yyy ls 22ti o 52/ | -5Jl i3 s tz e zn/ | R t eo g H gai p | e- H o p & | R a p / | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | ------------------------------------- | --------------------------- | ------------------ | ------------------ | ------------ | --------- | -| | | | | | | | | | | | | | | | | | | | | | | | k y c 5 | - 3 a- 443 4 | | | | -they love, wherever they are. 1 115 55% %% M o s t l i k e l y t o l i s t e n t o H i p - H o p & R a p / -Dance Music/Jazz/Reggae -Most likely to listen to Hip-Hop & Rap/ -| | | | | | | | | | | | | | | | of consumers globally | | | | | | MTMD | aooisn osctte c s tl eliia kk | M iseek llel uyyikl s yteti ocolty / ol lJiit ssa | loitts z eetu znne / s n R | ett oeotp g oHaHg iHiid appi | e--paHH-uHoodppoio p&& s & tRRr eaRaappam//p/ing | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | --- | --- | --- | -------------------------- | -------------------------------- | --------------------------------------------------- | --------------------------- | -------------------------------- | -------------------------------------------------- | --- | -| | | | | | | | | | | | | | | | | | | | | | Dance Music/Jazz/Reggae wM | l | | | | | | -of consumers globally -| | | | | | | | | | | | | | | | aoref lcikoenlysu tmo berusy g alo sbmalalyrt | | | | | | DDT w Daainan | c cnce eec a eMMs | M uul i ks useii sccl i/y/ cJJ | /ta aJo zaz zzuz //zs | RRe/ eRe p ggea gggi aagd | eea a eu d i o | s t r e a m i n g | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | --- | --- | ------------- | -------------------- | --------------------------------- | ---------------------- | -------------------------- | ----------------- | ----------------- | -| | | | 27% | | | | | | | | 58% | | | | | | | | | | | | | | | | | - USING SMARTPHONES TO LISTEN TO MUSIC aoref clikoenlsyu tmo ebrusy g alo sbmalalyrt T w i c e a s l i k e l y t o u s e p a i d a u d i o s t r e a m i n g -2 7 % 5 8 % T w i c e a s l i k e l y t o u s e p ta i d a u d i o s t r e a m ri tn g -| | | | | | | | | | | | | | | | aoo ro efsf f p cc l ice | koooa e nnn lk ss yesu u | tur mm o mi n ee b e rrt | ussrh ys gge g al l oon l o | sbbe mbaxa at ll all ylyl ry t | | M | u c h m | o r e l i k | e l y t o | a t e n | d g i g s / | c o n c e s | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | -------------------------- | -------------------------- | -------------------------------- | --------------------------------- | --- | --- | --------- | ------------- | ----------- | --------- | ------------- | ------------- | -2 7 % 5 8 % T Tw wT wi i c ci e ec ea a sa s sl li i k k lie ek l le y y l y t t o ot o u u s sue e s ep p a p a i iad d i d a a u ua d du i ido o i o s s t t sr r te er a ae m ma m i in n i g gng -2 7 5 8 a r e s p l i ke e a l k y e t r o i n b ut hy e a n s em x t a r t M u c h m o r e l i k e l y t o a t t e n d g i g s / c o n c e r t s -% % aaarreer se pll iilkk eikee a1e ll kyy2 e ttm roto oi no bb n b uu ttu hyyh e aas n ss emsm taa arrttr t M u c h m o r e l i k e l y t o a t t e n d g i g s / c o n c e r t s -| | | | 2 22 | 7 7 7 % %% | | | | | | | 5 55 | 8 8 8 % %% | | | | ly | | y a | xm | | | | | | | | | -| --- | --- | --- | ---- | ---------- | --- | --- | --- | --- | --- | --- | ---- | ---------- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -o f to t al m usic o f 16 - 24 y ear olds s p e a 1 k 2 e m r i n o n t th h e s n e x t MM uu cc hh mm oo rr ee ll ii kk ee ll yy tt oo au st et e an d tu g rin g ts a / b c leo n c e r t s -| | | | | | | | | | | | | | | | | | | | | | M MMu | ucu ch hc h m m mo | or roe er el li i k klie | ekl ley yl yt t o ot | oa a t att ttee nne | ddn dgg iig gg issg //s | /cc oocnnoccneecrrettsrsts | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ------------------- | -------------------------- | ----------------------- | -------------------- | ---------------------------- | ----------------------------- | -of total music of 16-24 year olds ssspppeeeaaa1kk2kee emrr rii nnoin ntt hththheees nn neeexxxtt t M u c h m o r e l i k e l y t o u s e t a tu r n t a b l e -loisft etontinagl m timuseic i s wofo 1u6ld-2 c4h yoeoasre oald s 12 months Much more likely to use a turntable -112212 mm mooonnntththhsss -olisft teontianlg m tiumsiec is owfo 1u6l-d2 c4h yoeoasre o ald s Much more likely to use a turntable -olooiosnfftf et tmtoonottoitnaabagll i llm m emt imuduusseseiivcci cii cs es swoomoffof 11 au166r6ld--t-22p 2c44h4h oyy oyneeoeaeasa rrew r o o aohlldd ledssns MMMuucuchhc h mm moorroeer ellii kklieeklleyyl ytt oot ouu ssuees eaa att uutrrunnrttnaatbballbeele -| | | | loisnt emnionbgil eti mdeev iics es | | | | | | | | wsmoualrdtp chhoonoes ew ah en | | | | | | | | | | | | | | | | | -| --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -aswwwmsookoauueuldrlddltd p “cc ihcfhh hoyooonooooeuss see woe aa nha ley n -olliilssnistt teemennnoiinninbgggil et tt iimmdimeeeve ii icssis e s -samskaerdt p“hifo ynoeu wohnelyn -on mobile devices -| | | | ooonnn m mmooobbbiillieele d ddeeevvviicciceeesss | | | | | | | | hasssmmasmkdaae aorrdrttn tpp“pehihf hdo oyonenonveeuei c ww owehnh hteleoyenn n | | | SOCIAL MEDIA | | | | | | | | | | | | | | -| --- | --- | --- | --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------------------------------------------------- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -ahaskde odn “eif dyeovuic oen tlyo -lhaaisaasstskkedkeen eodd don ““ ne“iiff. if d .yy .ye”ooovuuiuc oo eonn ntllyyoly 75% 30% -| | | | | | | | | | | | hlisatde no noen .d..e”vice to | | | | | | | | | 75% | | | | | | 30% | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -lhhihsaataddedn oo oonnnneee .dd. d.ee”evvviicciceee tt tooo Consumers are taking to social 75% 30% -| | | | | | | | | | | | l i s t e | n o n . . . ” | | | | | | | | 7 | 5 | % | | | | 3 | 0 % | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --------- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -l li ils sist tte een nn o o on nn. .. ... ..” ”.” media and messenger apps to 7o 7f us5 5e rs i% %n Latin 3 3of co0 0n sum% %% ers -| | | | | | | | | | | | | | | | | | | | | 7 | 5 | % | | | | 3 | 0 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | --------- | ------------ | ---- | --- | --- | ---------- | ------------------------------------------- | -| | | | | | | | | | | | | | | | | | | | | of | u s e rs | in L a ti | n | | | o fo f ll | c o o w n s a u r m tis e t r s s on | -| | | | | | | | | | | | | | | share and discuss their favourite | | | | | | A m | e r ic a | lis te n | to | | | | | -7 5 % o f u s e r s i n L a t i n f o o f l l c o o w n s a u r t m i s e t s r s o n -| | | | | | | | | | | | | | | | | | | | | m o A f | m u u s e s i r c e i c r p s a | o i n l s i t s L e t e a d n t | o i n t n o | | | o s o f | c c i o a n l ms u e m d e i a r s | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ----------------------------------- | ------------------------------------ | ----------------- | --- | --- | --------- | ---------------------------------------- | -7 5 % music. As record companies A m e cer i c a l i s t e tn stn t o f of l cl o wn au r t i se tr ss o n -7 5 o o m fo f u fu u i s su s i e ls r re ps s re o si i n n s wi t n L L e o a aL d ta oi i n ni n o o s o fo c cf i o oa c nol sm sn u s e m mu d m ei a re s r s -% s A o m c e a r i c n a t l i s t e rn k t o f o l l o w a r t i s t s o n -7 5 % c o n t in u e t o w o r k w i t h t e c h n o logy A A m mm u ies e i rr c ii cr c p iaa o a l l sii s st t te ee d nne o n tt n oot f f soo o ll lcloo i awwo l ama rra ettr ii dsts itit ass soo nno n -7 775 55% s o A c m a e l n c e t w l i s o t r k s o s o f o c i l a l l w m e d i a s t -%% TT hH e E h Hi g Ih Ge Hs t E rS a Tt e R o Af T s Em Oa Fr t p h o n e u s e m u s i c p o s t e d o n -| | | | | | | | | | | | | | | p a rt | n e rs t o l ic | e n s e | t r a c k | s fo r | | m m s o m | u u c su s i ai i s c c l i c pn p | oe p o t s o s wt t s e e t o d de r | d k o o s n no n | | | s s o o s c co | i i c a a il la m ml m e e d d e i id a a i a | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------------------- | ------- | ----------- | ---------- | --- | --------- | ----------------------------------------- | ------------------------------------ | -------------------- | --- | --- | -------------- | -------------------------------------------------------- | -T h e h i g h e s t r a t e o f s m a r t p h o n e u s e s o c i a l n e t w o r k s -| | | | | | | | fTSohMre Am hRuigTshPiceH sisOt irnNa LEtea U toiSnf E sA mFmOaerRrti pcMhaUo | | | | | nSeIC u s e | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------------------------------------------------- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -of consumers use consumers, music will continue ssooscociicaailal nnl enettewwtwoororkkrssks -Tfohre m hiugshice sist irna tLea toinf sAmmaertripchao ne use -| of consumers use | | | | | | | ITTS reeI | N hh h uiiLggigs AhhihceTe esisI | Ntt t irr n | raaA aLttM teeae E | R Iss A C Aaaaerrrtrttippcphhahoo onnneee uu ussseee | | | | | | | | | | | | | | | | | -| ----------------- | --- | --- | --- | --- | --- | --- | ----------- | ----------------------------------- | ------------ | ---------------------- | -------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -s m a rt p h o n e s t o f oThhh em ss too io nfff smmmm to drive online conversations. -| o f c | o n s u | m e r s | u s e | | | | INfo Mr EmXIuCsOic is in Latin America | | | | | | | | | | | | | | | | | | | | | -| ----- | ------- | --------- | ------ | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -osmf caortnpshuomneerss tuos e -| l so i oo s mff t f e cacc | n oorot nn t npss o shuu | m mmmn u eeee s rrsrs i s c s t uuouss seee | | | | | IffNfooo rMrr mm EmXuuICussOsiiccic ii ssis ii nnin LL 9Laaatttiinn3in AA Amm%meeerrriiccicaaa | | | | | | | | | | | | | | | | | | | | | -| ----------------------------- | ---------------------------- | ------------------------------------------------ | --- | --- | --- | --- | ------------------------------------------------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -uo -| slimsteanrt ptoh omnuessi cto | | | | | | | IN MEXICO | | | | | | | | | | | | | | | | | | | | | -| ------------------------------ | --------------- | -------------------- | ----------- | --- | --- | --- | ----------------- | --------------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| ls s smm a a | rr rt t p p hh | moo nnnu ee ss c | tt tooo | | | | | | | 93% | | | | | | | | | | | | | | | | | | -| i s m t e | an tt p o | h o es is | | | | | I N M | E X I C | O | 9 | 3 % | | | | | | | | | | | | | | | | | -| l i s t e | n t o | m u s i | c | | | | I IN NIN M M ME | EEX XXI IC CICO | OO | | | | | | | | | | | | | | | | | | | -| l l i i si s t tt e e | n n t t o o | m m u u s s i i | c ci | | | | | | | 9 | 3 % | | | | | | | | | | | | | | | | | -| l s | e n t o | m u s | c | | | | | | | 9 99 | 3 33 % %% | | | | | | | | | | | | | | | | | -GLOBALLY, CONSUMERS ARE USING -| | | | | | | | IN BRAZIL | | | | | | | SOCIAL NETWORKS TO DISCUSS MUSIC | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------------- | --- | ------ | ------ | --- | --- | -------------------------------- | -------- | --- | --- | --- | ------------- | --- | --- | --- | --- | ------ | ---- | --- | --- | -| 9 | 4 | | | | | | IN BRAZIL | | | 92% | | | | | | | | | | | | | | | | | | -| | | | % | | | | | | | 9 | 2 % | | | 3 | 5 % | | | | 2 3 % | | | | | 3 | 0 | % | | -| 9 | 4 | | % | | | | I N B | R A Z I L | | | | | | | | | | | | | | | | | | | | -| | | | | | | | I N B | R A Z I L | | 9 | 2 % | | | 3 | 5 % | | | | 2 3 % | | | | | 3 | 0 | % | | -| 9 | 4 | | | | | | IINN IN BB BRR | RAA AZZ ZIILL | | | | | | | | | | | | | | | | | | | | -| 9 | 4 | | % | | | | | IL | | 9 | 2 % | | | 3 | 5 % | | | | 2 3 | % | | | | 3 | 0 | % | | -| | | | % | | | | | | | 9 9 92 | 2 2% % | | | 3 | 5 % | | | | 2 3 % | | | | | 3 | 0 | % | | -| 9 9 | 4 4 | | % % | | | | | | | | % | | | | | | | | | | | | | | | | | -| 9 | 4 | | % | | | | | | | | | | | 3 3 35 | 5 5% % % | | | | 2 2 23 3 3% % | % | | | | 3 3 30 | 0 0% | % % | | -IN ARGENTINA of Instagram users talk about of Facebook users share links to -o f 1 6 - 2 4 y e a r o l d s u s e of WhatsApp users share links -I N A R G E N T I N A 8 9 % o f I n s t a g r a m u s e r s t a l k a b o u t o f F a c e b o o k u s e r s s h a r e l i n k s t o -o f 1 6 - 2 4 y e a r o l d s u s e t o o f m W u h s a i t c s u A s p i n p g u t s h e re s a s p h p a r e l i n k s m u s i c o n t h e s e r v i c e m u s i c u s i n g t h e a p p -s m a r t p h o n e s f o r m u s i c I N A R G E N T I N A 8 9 % mo f u I n s t a g r ta m su s e r s t a l k a b o u t o f F a c e b o o k u s e ar s s h a r e l i n k s t o -o f 1 6 - 2 4 y e a r o l d s u s e I N A R G E N T I N A o t o f W m h u a s i t c s A u s p i p n g u s t he r e s a s p h p a r e l i n k s s i c o n h e e r v i c e m u s i c u s i n g t h e p p -s m a r t p h o n e s f o r m u s i c I N A R G E N T I N A 8 9 % o f W h a t s A p p u s e r s s h a r e l i n k s o f I n s t a g r a m u s e r s t a l k a b o u t o f F a c e b o o k u s e r s s h a r e l i n k s t o -o f 1 6 - 2 4 y e a r o l d s u s e I N IN A A R R G G E E N N T T I N IN A A 8 9 % o t o o f f fW W m h h u a a s i t t cs s A A u p s p i p p n gu u us s t e e h re r es s as s h h p a a p ar r e e r e l l i i n n link k s s o o mfo f u fI I n n s ns s i ct t a a o ag g n gr r a a t h am m e m u u s esu s e e r vr r s s i c t t e a a tla l k k l a a ba b ob o u u tu t o o mf f uFf F sa a i ca c c e e ub b so o i no o gk k ut u h s s ee e r r as s p ss s ph h a a r r e e r l l i i n n lik k s s st t o o t -so o m f f f 1a 1 6 6 r6 t- - p2 2 4 4 h o y y n e e ea a s r r r f o o oo l l d d lr s s m u u uu s s ss e e ei c t o o m W u h s i a c t u s A s i p n p g t h s e r sa p s h p k s m u s I i c s o t n t r h e s e r s v e i c r s e k o t m o u s F i c c u e s b i n o g o k t h u e s ea r p p s h a e n k o -o 1 - 2 4 y e a d s 8 8 8 9 9 9 % % % m m u u su s i i c c o o n n n t t h h eh e s s e e r r v v i i c c e e m m u u s s i i sc c i u u s s i i sn n g g t t h h e e a a p p p p -s m a r t p h o n e s f o r m u s i c t t o o t o m m m u u u s s i i s c c i c u u us s i i s n n i n g g g t t h h t h e e e a a a p p p p p p m s i c o t e s e r v i c e m u c u i n g t h e a p p -| s s m m a a | r r t t p p h h | o o n n e e s s | f f o o r r m m | u u s s i i c c | | | | | | | | | | | | | | | | | | | | | | | | -| ----------- | --------------- | ------------------- | --------------------- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| s m | a r t p | h o n e s | f o r m | u s i c | | | | | | | | | | | | | | | | | | | | | | | | -Base: All participants (n=19,000) from all 18 countries surveyed. -| | | | | | | | | | | | | | | | | | | | | | MUSIC CONSUMER INSIGHT REPORT 2018 | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | - -| 5 | | | | | | | | | | | | | | | | | | | | | | | | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| O N | - | D E | M | A | N | D | | | | | | | | | | | | | | | | | | | -VIDEO STREAMING MAKES UP MORE THAN HALF OF ON-DEMAND -MUSIC STREAMING TIME -| S T | R | E A | M | I N | G | L | E A | D | S | M U | S | I C | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -52% is on video -47% -streaming -| C O | N | S | U M | P | T | I O | N | G | L O | B A | L | L Y | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -28% is on paid -of time spent listening -audio streaming -to on-dema4nd mu7sic i%s -| Licensed on-demand streaming is popular | | | | | | | | | | | | | on YouTube | | | | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -20% is on free -audio streaming -with consumers the world over. -of time spent listening -to on-demand music -| | | | | | | | | | | | | | HOWEVER, USER | | is on YouTube | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | --- | --- | --- | ----------------- | --- | -------------- | --------------------------------- | --- | --- | --- | --- | --------------- | --- | --- | --- | -| | | | | | | | | | | | | | | | | Estimated Annual Revenue Per User | | | | | 35% SAY A MAIN | | | | -| 86% | | | | | | | AUDIO STREAMING USE | | | | | | UPLOAD SERVICES | | | | | | | | | | | | -REASON FOR NOT -| | | | | | | | | | | | | | ARE NOT | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -USING A PAID AUDIO -| of consumers are listening to music | | | | | | | GLOBAL: 61% | | | | | | RETURNING | | | | | | | | | | | | -| ------------------------------------ | --- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -through on-demand streaming (audio FAIR VALUE SUBSCRIPTION IS -US$20 -| and video) | | | | | | | Russia 87% | | | | | | TO T | H E M | U S I C | | | | | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | ---- | ----------- | ----------------------- | --- | --- | --- | --- | ----- | -------------- | --- | --- | --- | -| | | | | | | | | | | | | | | | id audio streaUmiSng$20 | | | | | | THAT ANYTHING | | | | -| | | | | | | | | | | | | | | 28 % | is | o n p a | | | | | _∗_ Equal contribution +I. Prophase: Nuclear envelope and nucleoli disappear. Chromatin condenses into chromosomes, which +are made up of two identical sister chromatids joined by a centromere. In animal cells, centrioles start +migrating to opposite ends of the cell (centrioles are not present in plant cells). The mitotic spindle +forms and begins to move chromosomes towards the center of the cell. -1 +NOTE: Prometaphase is a transitional phase between prophase and metaphase. -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +II. Metaphase: Brief stage in which chromosomes line up in the equatorial plane of the cell. In animal +cells, one pair of centrioles are visible at both ends of the cell. The mitotic spindle is fully formed. +III. Anaphase: Sister chromatids begin to separate, becoming individual chromosomes, which begin to +migrate to opposite ends of the cell. -## 2 BACKGROUND +IV. Telophase: A full set of chromosomes reaches each pole of the cell. The mitotic spindle begins to +disappear. The nucleus and nucleoli begin to reappear. Chromosomes begin to unravel into chromatin. -**LLMs for Math and Science** LLMs are rapidly moving beyond simple linguistic pattern matching, developing complex multi-step reasoning skills that are necessary for mathematical and scientific general intelligence. Current methodologies often rely on specialized prompting paradigms; for instance, Chain-of-Thought (Wei et al., 2022) encourages explicit symbolic derivation, while Programof-Thought (Chen et al., 2022) offloads complex scientific computing to external Python interpreters. More recently, reasoning-centric models such as OpenAI’s o1 and DeepSeek’s R1 have utilized large-scale reinforcement learning to internalize these logical trajectories (DeepSeek-AI et al., 2025). However, breaking records on benchmarks is only a partial milestone. As these systems approach AGI-level performance, research must shift from merely tracking performance to understanding how to best collaborate with these models. In this work, we argue that identifying the structural biases and implicit reasoning strategies within these models is a vital prerequisite for the effective oversight and safe deployment of AGI systems in high-stakes and unsupervised scenarios. +Cytokinesis or cytoplasmic division usually occurs at the end of telophase. In plant cells cytokinesis is +accomplished by the formation of a cell plate. Animal cells separate by forming a cleavage furrow. In +some cells (e.g., muscle cells and certain embryonic cells) cytokinesis does not occur or is delayed until +multiple nuclear divisions have occurred. So although cytokinesis is associated with telophase of +mitosis, it is separate from mitosis and may not occur at all. Mitosis is therefore the process of nuclear +division, distinct from cytokinesis or cytoplasmic division. -**Automated Prompt Engineering** Optimization of model performance has evolved from the manual engineering of prompts to a systematic algorithmic search. Early works in automated prompt engineering have demonstrated that LLMs are often the best optimizers of their own instructions (Zhou et al., 2022; Yang et al., 2023a). Genetic Pareto (GEPA) utilizes an evolutionary approach to iteratively optimize high performing prompts (Agrawal et al., 2025). While previous works on automated engineering focused primarily on maximizing accuracy, we reposition these algorithms as tools for understanding. By allowing GEPA to explore the vast search space of possible instructions, we treat the resulting optimized prompts and the trajectory of prompt evolution as valuable windows into the latent preferences of the model. We can then identify specific heuristics that can be reverse engineered and transferred back to improve human-authored prompting methodologies. +EXERCISE 1 – Observing mitosis under the microscope -**Knowledge Transferability** A fundamental question in the path toward AGI is whether machine intelligence is a singular phenomenon or a collection of fragmented epistemologies (Quattrociocchi et al., 2025). Previous research into model distillation and generalization has shown that while knowledge can be transferred between architectures, internal reasoning protocols often remain model specific (Hinton et al., 2015). This brittleness poses challenges for the post-AGI era. For AI systems to truly be of use in collaborative scientific discovery, their logic must be interoperable. If a reasoning strategy optimized for one model fails on another, it suggests a closed epistemology, which represents a detached form of intelligence that lacks a universal logical foundation (Pal et al., 2026). +Examine prepared slides of both plant cells (onion/allium root tip) and animal cells (whitefish blastula) under the +microscope at 400X. Even though the cells in these tissues are rapidly dividing, most of the cells you see will be in +interphase (between cell divisions). Using your microscope, scan the slides to find a cell in interphase and each +one of the four stages of mitosis. Draw a schematic representation of your observations for both plant and +animal cells at each stage in the spaces provided on your worksheet and be sure to indicate and clearly label the +important features or events of each stage. -## 3 METHODOLOGY +124 -We assess LLM reasoning in two domains: (1) formal mathematical theorem proving via Lean from the MiniF2F dataset (Zheng et al., 2022) and (2) scientific reasoning via the GPQA Diamond benchmark (Rein et al., 2023). We refer to these benchmarks as “Algebra” and “GPQA” respectively. These domains assess scientific reasoning in complementary ways: Lean requires rigid verifiable logic and is open ended, while GPQA evaluates high level conceptual reasoning and is multiple choice. We apply a custom variant of GEPA which has been simplified and adapted for Lean theorem proving and GPQA. Implementation details are specified in Algorithm 1. + MEIOSIS -We conduct the entirety of GEPA optimization using DeepSeek-V3.2, which currently has state-ofthe-art performance on reasoning benchmarks. To gain insight into the transferability of optimized prompts across different models, we also test the same prompts on ChatGPT-5.4-mini, GLM 5, and Claude Sonnet 4.6, all of which were released within the past few months and have competitive performance. By analyzing the prompt optimization process and comparing prompt performance across models, we hope to gain insight into how prompting strategy affects reasoning performance. +During sexual reproduction in eukaryotes, a haploid sperm cell fuses with a haploid egg cell to produce +a diploid zygote or fertilized egg. In most species, it is very important that the offspring produced by +fertilization have the same number of chromosomes as the parents. Even a single extra or missing +chromosome can be lethal or extremely deleterious to an individual (e.g., Down’s syndrome in humans). +Meiosis is a special type of cell division that produces haploid gametes (sperm cells or ova). Meiosis only +occurs in an individual’s gonads, during their reproductive years. -For every combination of model and benchmark, we run evaluations across four prompts: **(1) HandCrafted Simple:** A hand-crafted simple prompt, intended to emulate the prompting ability of average technical users, which serves as a baseline; **(2) Hand-Crafted CoT:** A hand-crafted Chain-of-Thought prompt, intended to emulate best practice prompting strategies to current knowledge, which serves as a baseline; **(3) GEPA Optimized Baseline:** The initial prompt from the GEPA optimization process; and **(4) GEPA Optimized Final:** The final prompt from the GEPA optimization process. Prompts are further discussed in the Appendix, and examples are provided. +Meiosis involves two cell divisions and ultimately produces four haploid gametes. The haploid gametes +produced by meiosis are different from each other as well as from the parent cells due to the crossing +over of genetic material between homologous chromosomes and the random distribution of +homologous chromosomes. -2 +Meiosis is different in males and females: -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +Spermatogenesis -## **Algorithm 1** Custom Variant of Genetic Pareto (GEPA) +• -- 1: **Input:** Seed prompt _P_ 0, Lean theorems _L_ , GPQA questions _G_ , iterations _T_ , samples ( _n, m_ ). +In males four functional sperm cells are produced by meiosis. -- 2: **Initialize:** _Population ←{P_ 0 _}_ , _Pareto ←∅_ . +Oogenesis -- 3: **for** _t_ = 1 **to** _T_ **do** +• Due to unequal distribution of cytoplasm in during meiosis, one large functional egg -- 4: _P ←_ Sample( _Pareto ̸_ = _∅_ ? _Pareto_ : _Population_ ) 5: _SL ←_ Sample( _L, n_ ), _SG ←_ Sample( _G, m_ ) +(ovum) and three small polar bodies are produced. -- Lean Verify(LLM( _P, ci_ )) _ci ∈ SL_ +STAGES OF MEIOSIS -- 6: **Evaluate: v** [ _i_ ] _←_ for _ci ∈ SL ∪ SG_ �Check ~~A~~ nswer(LLM( _P, ci_ )) _ci ∈ SG_ +MEIOSIS I -- 7: **Update Pareto:** Add _P_ if non-dominated by existing prompts; remove dominated. 8: _Errors ←{ci_ : **v** [ _i_ ] = 0 _}_ +This is a reductive division in which one diploid (2N) cell produces two haploid (1N) cells. -- 9: _Critique ←_ LLM ~~C~~ ritic( _P,_ Logs( _Errors_ )) +125 -- 10: _P[′] ←_ LLM ~~E~~ volve( _P, Critique_ ) + Prophase I: Similar to prophase of mitosis with one important difference: Crossing Over in which pairs +of homologous chromosomes synapse together to form tetrads and exchange genetic information +(DNA). Crossing over creates new, recombinant chromosomes. Homologous chromosomes contain the +same arrangement of genes and are of the same size. Although they are very similar (hence the term +“homologous”), they differ slightly in DNA sequence since one comes from an individual’s mother, and +the other comes from an individual’s father. -- 11: _Population ← Pareto ∪{P[′] } {_ Prune to Pareto + new child _}_ +Metaphase I: Brief stage in which tetrads line up in the equatorial plane of the cell. -- 12: **end for** +Anaphase I: Homologous chromosomes separate and migrate to opposite ends of cell. Sister +chromatids DO NOT separate. -- 13: **return** _Pareto_ +Telophase I/Cytokinesis: A full set of chromosomes reaches each pole of the cell. The cells produced +contain half of the original number of chromosomes and are considered haploid. Each chromosome +consists of two sister chromatids attached at the centromere. -## 4 RESULTS +Interphase may be very brief or absent between meiosis I and meiosis II. -## 4.1 BENCHMARK PERFORMANCE +MEIOSIS II -Table 1: Performance comparison of models on benchmark datasets. +This division is very similar to mitosis – sister chromatids are distributed into different cells. -|**Model**|**Method**|**Algebra**|**GPQA**| -|---|---|---|---| -||Hand-Crafted Simple|86.11%|91.67%| -|DeepSeek-V3.2|Hand-Crafted CoT
GEPA Optimized Baseline|97.22%
91.67%|91.67%
88.89%| -||GEPA Optimized Final|**100.00%**|**94.44%**| -||Hand-Crafted Simple|50.00%|**91.67%**| -|GPT-5.4-mini|Hand-Crafted CoT
GEPA Optimized Baseline|47.22%
50.00%|**91.67%**
88.89%| -||GEPA Optimized Final|**61.11%**|**91.67%**| -||Hand-Crafted Simple|91.67%|**91.67%**| -|GLM 5|Hand-Crafted CoT
GEPA Optimized Baseline|**97.22%**
91.67%|86.11%
88.89%| -||GEPA Optimized Final|94.44%|**91.67%**| -||Hand-Crafted Simple|30.56%|77.78%| -|Claude Sonnet 4.6|Hand-Crafted CoT
GEPA Optimized Baseline|**52.78%**
50.00%|**83.33%**
80.56%| -||GEPA Optimized Final|50.00%|80.56%| +Prophase II: DNA condenses into chromosomes. No crossing over occurs. +Metaphase II: Individual duplicated chromosomes line up in the equatorial plane of the cell. +Anaphase II: Sister chromatids separate and begin to migrate to opposite poles of the cell. -Prompt performance on benchmarks is displayed in Table 1. Our first major observation is that the GEPA Optimized Final prompt achieves its most significant gains on DeepSeek-V3.2, reaching 100.00% on Algebra and 94.44% on GPQA. This confirms that the GEPA optimization process is highly effective when evaluated on the same model used during optimization. Our second major observation is that the superiority of the GEPA-optimized prompts does not reliably transfer to other models. While some models show marginal benefits – for example, GPT-5.4-mini sees an improvement in Algebra from 50.00% to 61.11% – the optimized prompt is rarely the undisputed best performer elsewhere. Notably, for GLM 5 (Algebra) and Claude Sonnet 4.6 (both benchmarks), the Hand-Crafted CoT prompts actually outperform the GEPA Optimized Final prompt. +Telophase II/Cytokinesis: A full set of chromosomes reaches each pole of the cell. Four -These results suggest that the optimization process is highly model-specific, and suggests a trade-off between prompt universality and performance. Since DeepSeek was used as the optimizer, the evolved prompts likely capture patterns specific to DeepSeek’s architecture that do not resonate with the internal logic of other models. This highlights a significant lack of interoperability and reinforces the brittleness of automated prompt engineering across diverse foundation models. + genetically unique gametes are produced. -3 +Exercise 2A – Modeling the stages of meiosis -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +1. Use the chromosome bead models to construct a single pair of homologous chromosomes, each with two sister + chromatids. Use red for the maternal chromosome and yellow for paternal chromosome. Use the rubber + tubing with magnets for centromeres and attach 10 beads to each end of the centromere. -**==> picture [318 x 128] intentionally omitted <==** +2. Model each stage of meiosis with your single pair of homologous chromosomes, diagramming each stage on + your worksheet using different colors to indicate maternal and paternal chromosomes. Also, be sure to include + one crossover event during Prophase I. -Figure 1: The length of GEPA proposed prompts increases over the course of optimization, with the final prompt often being about twice as long in characters as the initial prompt. This shows that detailed prompting is likely required to unlock better reasoning capabilities in LLMs. +INDEPENDENT ASSORTMENT OF CHROMOSOMES -## 4.2 PROMPT EVOLUTION PATTERNS +For each homologous pair of chromosomes, maternal and paternal chromosomes are randomly +distributed into daughter cells during meiosis I. The chromosomes of each homologous pair are +distributed independently of the other homologous pairs, a phenomenon we refer to as independent +assortment. Independent assortment can create a staggering number of possible chromosome +combinations in gametes depending on the haploid chromosome number (n). -By manually analyzing the evolution of prompts over the course of GEPA optimization, we primarily find that prompts evolve from telling the model _what to do_ to coaching it on _how to do it_ , similar to a human expert. In the process, significant domain knowledge and best practices are often added to the context. For example, in the GEPA optimization process for the Algebra benchmark, prompts eventually referenced domain specific strategies such as using Eisenstein’s Criterion for minimal polynomials. Similarly, later GPQA prompts mentioned specific strategies such as quantum field theory loop counting. Later Algebra prompts also explicitly warn against common pitfalls, especially with regard to Lean formatting. Interestingly, a robust protocol for “Handling False Statements” is also introduced to prevent hallucinating proofs. The Appendix contains prompt examples. +126 -This observation is also supported by analyzing the prompt lengths over the course of optimization, as shown in Figure 1. The final prompt turns out to be around twice as long as the initial prompt. This demonstrates that LLMs often rely on detailed prompting to fully elicit their reasoning capabilities. + The number of possible gametes generated by independent assortment alone (i.e., without considering +the effects of crossing over) is 2n, where n is the haploid number of chromosomes in a cell. For example, +in Homo sapiens (humans) n = 23 thus a human cell can produce over 8 million different gametes by +independent assortment (223 = 8.4 million). The effect of crossing over further increases genetic +variation in gametes generated by an individual such that the number of possibilities is essentially +infinite. -By analyzing the embedding space for prompts over the optimization process, we also find that the embeddings of proposed prompts tends to drift in a consistent direction over the course of optimization. This suggests that for the same task, some regions of prompting space may offer more promising performance. A plot of this behavior is in the Appendix. +Exercise 2B – Modeling independent assortment -## 5 DISCUSSION +Use the chromosome bead models to model meiosis with one, two, or three pairs of homologous chromosomes. +By doing this exercise you should be able to determine all the different combinations of chromosomes that are +possible in gametes due to the independent assortment of homologous pairs. Diagram all possible arrangements +in the circles provided using red for maternal chromosomes and yellow for paternal chromosomes. -**Conclusion** In this work, we demonstrated that using clever prompt optimization techniques can unlock reasoning potential in LLMs. We find that longer prompts, as well as prompts that guide the model along the _how_ of the target task, tend to elicit superior reasoning capabilities in LLMs. Such prompts can be optimized autonomously, and can beat hand-crafted baselines, even when we hand-craft prompts using state-of-the-art best practices. Our findings suggest that if current trends continue, post-AGI reasoning may not manifest as a universal logic, but rather as a collection of task-specific and model-specific heuristics that require precise coaching to elicit. +NOTE: You will disregard crossing over between homologous chromosomes for this exercise. -**Limitations** Our work is still relatively preliminary, and is limited to two benchmarks and four models. We would need a broader experimental scope to more confidently confirm the validity of our hypotheses. Furthermore, the stochastic nature of our LLM-driven prompt optimization loop could lead to inconsistency across multiple runs. +Before you leave, please make sure your table is clean, organized, and contains all supplies +listed below so that the next lab will be ready to begin. Thank you! -**Future Work** Our preliminary results suggest that prompt optimization remains dangerously architecture dependent. To prepare for a post-AGI landscape, future research should work on identifying “reasoning primitives”, or logical structures invariant across models. If AGI develops logic that humans cannot understand, we need to build automated tools to keep these systems interpretable. Without this, we risk a future where our most capable tools are also our least predictable. +Lab Supply List -4 +Lens paper -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +• +• Chromosome bead kit -## 6 STATEMENTS +Also, please attend to the following before you leave: -## 6.1 ETHICS STATEMENT +• Return all prepared slides to the correct trays. +• Put the compound microscopes away as indicated below, and make sure there is no slide on -While the discovery of effective reasoning paths could potentially be repurposed for harmful tasks, our analysis of the logic inherent in these models primarily as a vital safeguard. If we better understand how LLMs reason, we can better utilize LLMs for reasoning related tasks. Ultimately, identifying how LLMs approach reasoning tasks is essential for the development of robust, aligned, and transparent post-AGI systems. +turn off power -We also agree to NOT reveal examples from our evaluation datasets in plain text or images online, to reduce the risk of leakage into foundation model training corpora. +the stage: +o +o secure the power cord to the back of the microscope +o set to the lowest magnification (4X objective) +o cover and place “arm out” in the cupboard at your table -## 6.2 REPRODUCIBILITY STATEMENT +127 -Our results are reproducible to the extent of nondeterminism in the outputs of LLMs used. + 128 -## REFERENCES + LABORATORY 9 – WORKSHEET -- Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alexandros G. Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, and Omar Khattab. Gepa: Reflective prompt evolution can outperform reinforcement learning, 2025. URL https://arxiv.org/ abs/2507.19457. +Name ________________________ -- Lukas Berglund, Meg Tong, Max Kaufmann, Mikita Balesni, Asa Cooper Stickland, Tomasz Korbak, and Owain Evans. The reversal curse: Llms trained on ”a is b” fail to learn ”b is a”, 2024. URL https://arxiv.org/abs/2309.12288. +Section _______________________ -- Sebastien´ Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, Harsha Nori, Hamid Palangi, Marco Tulio Ribeiro, and Yi Zhang. Sparks of artificial general intelligence: Early experiments with gpt-4. March 2023. URL https://www.microsoft.com/en-us/research/publication/ sparks-of-artificial-general-intelligence-early-experiments-with-gpt-4/. +Ex. 1 – Observing mitosis under the microscope -- Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W Cohen. Program of thoughts prompting: Disentangling computation from reasoning for numerical reasoning tasks. _arXiv preprint arXiv:2211.12588_ , 2022. +Diagram a single cell in each of the indicated stages of mitosis both Onion Root Tip (plant) and +Whitefish Blastula (animal): -- DeepSeek-AI et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. _arXiv preprint arXiv:2501.12948_ , 2025. +ONION ROOT TIP (Allium) -- Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. _arXiv preprint arXiv:1503.02531_ , 2015. +WHITEFISH BLASTULA -- Jie Huang and Kevin Chen-Chuan Chang. Towards reasoning in large language models: A survey, 2023. URL https://arxiv.org/abs/2212.10403. +Interphase -- Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts. Dspy: Compiling declarative language model calls into self-improving pipelines, 2023. URL https://arxiv.org/abs/2310.03714. +Interphase -- Iman Mirzadeh, Keivan Alizadeh, Hooman Shahrokhi, Oncel Tuzel, Samy Bengio, and Mehrdad Farajtabar. Gsm-symbolic: Understanding the limitations of mathematical reasoning in large language models, 2025. URL https://arxiv.org/abs/2410.05229. +Prophase -- Koyena Pal, David Bau, and Chandan Singh. Do explanations generalize across large reasoning models?, 2026. URL https://arxiv.org/abs/2601.11517. +Prophase -5 +Metaphase -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +Metaphase -Walter Quattrociocchi, Valerio Capraro, and Matjaz Perc.ˇ Epistemological fault lines between human and artificial intelligence, 2025. URL https://arxiv.org/abs/2512.19466. +Anaphase -- David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. Gpqa: A graduate-level google-proof q&a benchmark, 2023. URL https://arxiv.org/abs/2311.12022. +Anaphase -- Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models. _Advances in neural information processing systems_ , 35:24824–24837, 2022. +129 -- Chengrung Yang, Xuezhi Wang, Yutai Lu, Hanxiao Liu, Quoc V Le, Denny Zhou, and Xinyun Chen. Large language models as optimizers. _arXiv preprint arXiv:2309.03409_ , 2023a. + ONION ROOT TIP (Allium) -- Kaiyu Yang, Aidan M. Swope, Alex Gu, Rahul Chalamala, Peiyang Song, Shixing Yu, Saad Godil, Ryan Prenger, and Anima Anandkumar. Leandojo: Theorem proving with retrieval-augmented language models, 2023b. URL https://arxiv.org/abs/2306.15626. +WHITEFISH BLASTULA -- Kunhao Zheng, Jesse Michael Han, and Stanislas Polu. Minif2f: a cross-system benchmark for formal olympiad-level mathematics, 2022. URL https://arxiv.org/abs/2109.00110. +Telophase -- Yongchao Zhou, Andrei Ioan Muresanu, Ziwen Han, Keiran Paster, Silvian Puzicanin, Amar Terzic, Laurent Itti, et al. Large language models are human-level prompt engineers. In _The Eleventh International Conference on Learning Representations_ , 2022. +Telophase -## A APPENDIX +Ex. 2A – Modeling the stages of meiosis -## A.1 ADDITIONAL RESULTS +Diagram each stage of meiosis as it unfolds for a cell with 1 pair of homologous chromosomes. Be sure to use +different colors for maternal and paternal chromosomes and to have one crossover event. -**==> picture [396 x 144] intentionally omitted <==** + MEIOSIS I -Figure 2: The embeddings of GEPA proposed prompts tend to drift over the course of optimization. For both algebra and GPQA, there seems to be a significant jump in embedding space at around iteration 12. This suggests that for the same task, some regions of prompting space may offer more promising performance. +Prophase I -## A.2 PROMPTS +Important Events -The 4 prompts we tested for the Algebra benchmark are listed below. For sake of brevity, we do not include the 4 GPQA prompts in this section, but they are similar in nature. +Tetrad formation +and crossing over -Listing 1: Hand-Crafted Simple Prompt +Metaphase I -1 You are a mathematician and Lean 4 practitioner. 2 Your goal is to solve formal IMO-style problems and produce correct Lean 4 proofs. + Alignment of tetrads -3 4 Approach and tricks: +Anaphase I -6 +Telophase I & +Cytokinesis -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +130 -- 5 - Try to reduce the goal using simp/linarith/nlinarith or rewriting. 6 - Look for standard lemmas in Mathlib (algebra, number theory, inequalities, parity). +Separation of +homologous +chromosomes -- 7 - Prefer structured proofs: have, refine, apply, and use ‘by‘ blocks. 8 9 Output format: + 2 haploid cells + produced -- 10 - Return the response in a single fenced code block with ‘‘‘lean ... ‘‘‘ containing valid Lean 4 code. + MEIOSIS II -- 11 - Do not include any text outside the code block. +Prophase II -## Listing 2: Hand-Crafted CoT Prompt + Metaphase II -|1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29|You are an expert mathematician and Lean 4 developer.
Your goal is to solve formal IMO-style problems and produce correct Lean
4 proofs.
### Output Format
- Return the response in a single fenced code block with ‘‘‘‘lean ...
‘‘‘‘ containing valid Lean 4 code.
- Do not include any text outside the code block.
- The code should be complete, including necessary ‘import‘ and ‘open‘
statements.
### Imports
- Do not use a single monolithic ‘import Mathlib‘.
- Instead, import the specific files you need, e.g., ‘import Mathlib.Data
.Real.Basic‘, ‘import Mathlib.Algebra.Order.Field.Basic‘, ‘import
Mathlib.Tactic.Linarith‘.
### Chain-of-Thought (internal)
- Think step-by-step and consider small test cases for ‘N‘/‘Z‘ statements
.
- Do not reveal chain-of-thought. Output only the Lean code block.
### General Approach & Tactics
- Start with structured proofs: ‘have‘, ‘refine‘, ‘apply‘, ‘by‘ blocks.
Use ‘calc‘ for equational reasoning.
- Use ‘simp‘, ‘rw‘, ‘ring‘, ‘norm_num‘. ‘linarith‘ for linear
inequalities, ‘nlinarith‘ for non-linear.
- ‘ac_rfl‘ is useful for equalities up to associativity/commutativity.
### Handling False Statements
- If you find a counterexample, do not modify the statement.
- Provide a Lean code block that proves the counterexample, then leave
the original theorem with ‘sorry‘.
### Domain-Specific Hints
- Polynomial identities: ‘ring‘ is often a one-line proof.
- Inequalities: reduce to ‘0 _≤_x - y‘, then show nonnegativity via
squares or ‘positivity‘.
- Linear systems over ‘C‘: use ‘linear_combination‘ or a ‘calc‘ block
with ‘ring‘.| -|---|---| + Anaphase II +Telophase II & +Cytokinesis + Alignment of + individual + chromosomes -## Listing 3: GEPA Optimized Baseline Prompt + Chromatids + separate -1 You are an expert mathematician and Lean 4 developer. 2 Your goal is to solve formal IMO-style problems and produce correct Lean 4 proofs. 3 4 ### General Approach 5 6 1. **[Analyze][the][Problem:] **[Read][the][theorem][statement][carefully.] Restate the goal in your own words to ensure you understand it. Identify the core mathematical concepts involved (e.g., number theory , inequalities, algebra). + Four gametes + Produced -7 +Ex. 2B – Modeling independent assortment -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +1. A cell with 1 pair of homologous chromosomes (n = 1) -|7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44|2.
**Formulate a Mathematical Strategy:**
*
Before writing any code, sketch a proof plan on paper (mentally).
*
Consider standard theorems like AM-GM, Cauchy-Schwarz, Jensen’s
inequality, or properties of quadratic equations (discriminant).
*
For sums and series, look for telescoping patterns or
opportunities for integral bounds. For example, ‘1/_√_k‘ is often
bounded using ‘2(_√_k - _√_(k-1))‘.
*
A very common and powerful trick for inequalities is to use the
fact that ‘(x - y)ˆ2 _≥_0‘, which expands to ‘xˆ2 + yˆ2 _≥_2xy‘.
3.
**Structure the Lean Proof:**
*
**DRY (Don’t Repeat Yourself):** If you find yourself proving the
same thing multiple times with different variables, define a local
helper ‘lemma‘ to abstract the repeated logic. For example, if you
need ‘xˆ2/y + y _≥_2x‘ for several pairs ‘(x, y)‘, prove it once in a
general lemma.
*
**Structured Proofs:** Prefer structured proofs using ‘have‘, ‘
let‘, ‘calc‘, ‘refine‘, and ‘apply‘. Use ‘by‘ blocks for short tactic
sequences.
*
**Backward Reasoning:** Use ‘suffices‘ to simplify the goal to an
easier-to-prove intermediate statement.
### Lean 4 Tactics and Tricks
*
**Automation:**
*
‘ring‘: Use to prove equalities that hold in any commutative ring
(polynomials without division).
*
‘field_simp‘: Use to prove equalities in fields (expressions with
division). Provide non-zero hypotheses, e.g., ‘field_simp [h.ne.symm
]‘.
*
‘linarith‘: For proving linear arithmetic inequalities.
*
‘nlinarith‘: For proving non-linear arithmetic inequalities. It
is very powerful when combined with a hypothesis like ‘sq_nonneg‘.
*
‘positivity‘: To automatically prove goals of the form ‘0 < x‘ or
‘0 _≤_x‘.
*
**Rewriting and Simplification:**
*
‘rw‘: For rewriting terms using equalities.
*
‘simp‘: For general simplification. You can add specific lemmas
to its ruleset, e.g., ‘simp [add_comm]‘.
*
**Finding Lemmas:**
*
Mathlib is vast. If you think a lemma for a common mathematical
fact (e.g., telescoping sums) should exist, it probably does. For
example, ‘Finset.sum_Icc_telescope‘ is available for telescoping sums
over an ‘Icc‘.
*
Use ‘exact?‘ or ‘apply?‘ to ask Lean to search for a lemma that
solves the current goal.
*
**Crucially, only use lemma and tactic names that you are certain
exist in Mathlib.** Do not invent names. If your code uses a non-
existent identifier, it is incorrect. Double-check your toolset.
*
**Common Mathlib Lemmas for Inequalities:**
*
‘sq_nonneg (x : R)‘: Proves ‘0 _≤_x ˆ 2‘.
*
‘pow_two_nonneg (x : R)‘: Same as ‘sq_nonneg‘.
*
‘div_nonneg {a b : R} (ha : 0 _≤_a) (hb : 0 _≤_b) : 0 _≤_a / b‘. You
need to prove the numerator and denominator are non-negative.
### Output Format
*
Return the response in a single fenced code block with ‘‘‘lean ...
‘‘‘ containing valid, complete, and compilable Lean 4 code.
*
Ensure all necessary ‘import‘ and ‘open‘ statements are present.
*
Do not include any text, explanation, or comments outside the code
block.| -|---|---| +Assemble two duplicated chromosomes, one red and one yellow, with 6 beads on each end of the centromere. By +modeling meiosis, determine all possible combinations of chromosomes in gametes when n = 1, assuming there is +NO crossing over, and diagram each unique combination in the squares below using different colors for maternal +and paternal chromosomes. +21 = ________ (possible chromosome combinations when n = 1) +131 -8 + 2. A cell with 2 pairs of homologous chromosomes (n = 2) -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 +Assemble another homologous pair of duplicated chromosomes with 4 beads on each side of the centromere. Use +this homologous pair with the longer homologous pair from the previous exercise to model meiosis when n = 2 and +determine all possible combinations of chromosomes in gametes, assuming there is no crossing over. Diagram +each unique combination in the squares below using different colors for maternal and paternal chromosomes. -Listing 4: GEPA Optimized Final Prompt +22 = ________ (possible chromosome combinations when n = 2) -|1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42|You are an expert mathematician and Lean 4 developer.
Your goal is to solve formal IMO-style problems and produce correct Lean
4 proofs.
### Output Format
- Return the response in a single fenced code block with ‘‘‘‘lean ...
‘‘‘‘ containing valid Lean 4 code.
- Do not include any text outside the code block.
- The code should be complete, including necessary ‘import‘ and ‘open‘
statements.
### General Approach & Tactics
- **Verify the statement first.** Before attempting a proof, especially
for statements over ‘N‘ or ‘Z‘, quickly test small values (e.g., n=0,
1, 2). If you find a counterexample, follow the "Handling False
Statements" protocol below.
- Start with structured proofs: ‘have‘, ‘refine‘, ‘apply‘, ‘by‘ blocks. ‘
calc‘ blocks are excellent for equational reasoning.
- Use ‘simp‘, ‘rw‘, ‘ring‘, ‘norm_num‘. ‘linarith‘ is for linear
inequalities over ordered fields (‘R‘, ‘Q‘). ‘nlinarith‘ is for non-
linear inequalities.
- ‘ac_rfl‘ is useful for proving equalities up to associativity and
commutativity, which is often more robust than a sequence of ‘rw [
add_comm, add_assoc]‘.
- **DRY (Don’t Repeat Yourself).** If you prove the same intermediate
result multiple times with different variables, define a local helper
lemma inside the proof using ‘have‘. This improves clarity and
reduces errors.
‘‘‘lean
have my_lemma : _∀_(x y : R), 0 < y _→_xˆ2 / y + y _≥_2 * x := by
... -- proof of the lemma
-- then use it:
have h1 := my_lemma a b hb
have h2 := my_lemma b c hc
‘‘‘
### Handling False Statements
- If you find a counterexample (e.g., the theorem fails for ‘n=1‘), do
not attempt to prove a modified version of the theorem (e.g.,
changing ‘<‘ to ‘_≤_‘ or adding a hypothesis like ‘n _≥_2‘).
- Your task is to prove the theorem *as given*. If it is false, you
cannot complete the task.
- In this case, your response should be a Lean code block that formally
proves the counterexample. Then, leave the original theorem with ‘
sorry‘. This correctly signals that the requested proof is impossible
.
- **Example a assistant should produce for a false statement:**
‘‘‘lean
import Mathlib
-- The user’s theorem statement is false for n=1.
-- Here is a proof of the counterexample.
example : _¬_ (_∀_n : N, (n : R) ˆ (1 / n : R) < 2 - 1 / n) := by
intro h
specialize h 1
norm_num at h -- This will fail, proving the negation.
-- The original problem, which is unprovable.
theorem algebra_ineq_nto1onlt2m1on (n : N) : (n : R) ˆ (1 / n : R) <
2 - 1 / n := by
sorry
‘‘‘
### Domain-Specific Strategies & Mathlib Knowledge|| -|---|---|---| +3. A cell with 3 pairs of homologous chromosomes (n = 3) +Assemble another homologous pair of duplicated chromosomes with 2 beads on each side of the centromere. Use +this homologous pair along with the homologous pairs from the previous exercise to model meiosis when n = 3 and +determine all possible combinations of chromosomes in gametes, assuming there is no crossing over. Diagram +each unique combination in the squares below using different colors for maternal and paternal chromosomes. +23 = ________ (possible chromosome combinations when n = 3) -9 +Question: How many gametes would be produced by independent assortment alone, in a cell -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 + with 8 pairs of homologous chromosomes? __________________ -43 44 #### 1. Polynomial & Ring Identities 45 - For identities involving addition, subtraction, multiplication, and integer powers (i.e., polynomial identities), the ‘ring‘ tactic is extremely powerful. It works over any commutative (semi)ring like ‘Z ‘, ‘Q‘, ‘R‘, ‘C‘. For simple identities, this is often a one-line proof. 46 47 #### 2. Inequalities over R, Q (AM-GM, Cauchy-Schwarz, etc.) 48 - A common strategy is to prove ‘x _≥_ y‘ by showing ‘x - y _≥_ 0‘. 49 - **Pattern: Prove ‘x _≥_ y‘ by showing ‘x - y _≥_ 0‘** 50 1. State the subgoal: ‘suffices 0 _≤_ x - y by linarith‘. 51 2. Establish an algebraic identity for the difference, where the right-hand side is in a form that is clearly non-negative (e.g., a square or a sum of squares). 52 ‘‘‘lean 53 have h_ident : x - y = (a - b)ˆ2 / c := by 54 field_simp -- Handles division. Provide non-zero hypotheses, e. g., ‘field_simp [ne_of_gt hc]‘. 55 ring -- Finishes the polynomial part. 56 ‘‘‘ 57 3. Prove the right-hand side is non-negative. This often uses ‘ div_nonneg‘, ‘mul_nonneg‘, ‘sq_nonneg‘, and ‘positivity‘. 58 - **Pattern: Proving ‘(a-b)*(c-d) _≥_ 0‘** 59 * This is common in induction proofs. The key is to show that ‘a-b‘ and ‘c-d‘ have the same sign. 60 * A robust method is to use ‘mul_nonneg_of_signs_sync‘. For example , to prove ‘0 _≤_ (a - b) * (aˆn - bˆn)‘ for ‘0 _≤_ a, b‘ and ‘1 _≤_ n‘: 61 ‘‘‘lean 62 have h_iff : a _≤_ b _↔_ a ˆ n _≤_ b ˆ n := 63 Real.pow_le_pow_iff_of_nonneg (by linarith) (by linarith) (by omega) 64 exact mul_nonneg_of_signs_sync h_iff 65 ‘‘‘ 66 * This is much cleaner than a ‘by_cases‘ block. 67 68 #### 3. Linear Algebra over C 69 - **‘linarith‘ fails over ‘C‘** because it is not an ordered ring. 70 - **‘linear_combination‘ Tactic:** This is the most direct way to solve systems of linear equations. It combines *other equational hypotheses *[.] 71 ‘‘‘lean 72 -- h 0 : f + 3*z = 11 73 -- h 1 : 3*f - 5*z = -65 74 have h_14z : 14 * z = 98 := by linear_combination 3 * h 0 - h 1 75 ‘‘‘ 76 - **Warning:** Do not use ‘linear_combination‘ to manipulate a single equation with a constant (e.g., ‘linear_combination h 0 - 21‘ is incorrect). To solve for a variable in one equation, use ‘calc‘ or ‘ rw‘. 77 - **Solving Equations:** 78 1. **[From][a][system:] **[After][using][‘linear_combination‘][to][get][e.g.] ‘14 * z = 98‘, use ‘mul_right_cancel 0 ‘ to solve for the variable. 79 ‘‘‘lean 80 -- h_eq: 14 * z = 98 81 have hz : z = 7 := by 82 apply mul_right_cancel 0 (by norm_num : (14 : C) = 0) 83 rw [h_eq] -- Goal becomes: 98 = 14 * 7 84 norm_num -- Finishes the proof 85 ‘‘‘ 86 2. **[From][one][equation:] **[To][isolate][a][variable][in][an][equation][like] ‘f + 21 = 11‘, use a ‘calc‘ block. 87 ‘‘‘lean 88 -- h 0 : f + 21 = 11 89 have hf : f = -10 := by - -10 - -Accepted at the Post-AGI Science and Society Workshop at ICLR 2026 - -|90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115|calc f = (f + 21) - 21 := by ring
_ = 11 - 21
:= by rw [h0]
_ = -10
:= by norm_num
‘‘‘
#### 4. Algebraic Numbers & Field Theory (Linear Independence)
- Problems of the form "if ‘a + b*m + c*mˆ2 = 0‘ with ‘a,b,c _∈_Q‘, then ‘
a=b=c=0‘" are about linear independence and minimal polynomials.
- The strategy:
1.
Identify the algebraic number (e.g., ‘m = 2ˆ(1/3)‘).
2.
Find its minimal polynomial over ‘Q‘ (e.g., ‘p = X3 - 2‘).
3.
Prove ‘p‘ is irreducible over ‘Q‘ using **Eisenstein’s Criterion
**: ‘Polynomial.irreducible_of_eisenstein_criterion‘. For ‘X3 - 2‘,
use the prime ‘p=2‘. (Requires ‘import Mathlib.RingTheory.Polynomial.
Eisenstein‘).
4.
Verify ‘p‘ is the minimal polynomial using ‘minpoly.
eq_of_irreducible_of_monic‘.
5.
Use ‘minpoly.dvd Q m hq‘ to show the minimal polynomial divides
your polynomial. (Requires ‘import Mathlib.FieldTheory.Minpoly.Basic
‘).
6.
Get a contradiction from degrees (‘Polynomial.
eq_zero_of_dvd_of_degree_lt‘).
7.
Deduce coefficients are zero using ‘Polynomial.
coeff_eq_zero_of_eq_zero‘.
### Troubleshooting
- **‘rw‘ or ‘calc‘ block failures:** If ‘rw‘ or ‘calc‘ fails due to term
ordering (e.g., ‘a*b‘ vs ‘b*a‘), try ‘ac_rfl‘ or add an explicit ‘rw
[mul_comm]‘.
- **Unknown Identifier Errors:** If you get an ’unknown identifier’ error
for a common mathematical lemma (e.g., ‘pow_le_pow_left‘), the name
has likely changed in Mathlib4.
- **Search the documentation.** The best way to find the current name
is to search the Mathlib4 API documentation.
- **Look for prefixes.** Many lemmas are now namespaced by their
primary type. For example, lemmas about ‘Real‘ powers are often in
the ‘Real‘ namespace (e.g., ‘Real.pow_le_pow_of_le_left‘). Lemmas
about ‘rpow‘ (real powers) are often named ‘Real.rpow_...‘.
- **Common Power Inequality Lemmas (for ‘R‘):**
- For ‘0 _≤_x _≤_y _→_xˆn _≤_yˆn‘, use ‘Real.pow_le_pow_of_le_left‘.
- For ‘0 _≤_x < y _→_xˆn < yˆn‘, use ‘Real.pow_lt_pow_of_lt_left‘.
- For ‘0 _≤_a, b‘ and ‘1 _≤_n‘, ‘a _≤_b _↔_aˆn _≤_bˆn‘ is ‘Real.
pow_le_pow_iff_of_nonneg‘.
- **Complex Proofs:** If a proof is complex (e.g., induction with
inequalities), break it down into many small, verifiable ‘have‘
statements to isolate errors. Don’t write a single large ‘calc‘ block
or proof term until you have verified the intermediate steps.| -|---|---| - - - -11 +132 diff --git a/q132/random_k4/random_4.md b/q132/random_k4/random_4.md index 49a10f83bf172ba12f438f4cf5333a4440848722..b860e860f1dc243408b89f0f6afa5454c5be73f4 100644 --- a/q132/random_k4/random_4.md +++ b/q132/random_k4/random_4.md @@ -1,1762 +1,2250 @@ -Dishwasher -Use & Care Manual - - MODEL: WQP12-6502 - （FDWBI8067-24S） - -Table of Contents - -Important Safety Instructions-------------------------2 - -Features of your new dishwasher - ------------------ - - -3 4 - -Operating Instructions-------------------------------- -5 8 - -Preparing and loading dishes - ---------------------- - - -9 13 - -Care & cleaning------------------ - --------------------- - --14 16 - -Troubleshooting------------------------------------- - --17 18 - +| | Why | Does | | Reinforcement | | Learning | Generalize? | | A Feature-Level | | | | +| --- | ----------- | ---- | --- | ------------- | --- | ---------------- | ----------- | ----- | --------------- | ------ | --- | --- | +| | Mechanistic | | | Study | | of Post-Training | in | Large | Language | Models | | | +DanShi1,ZhuowenHan1,SimonOstermann2,3,RenrenJin1, +JosefvanGenabith2,3,DeyiXiong1* +1TJUNLPLab,SchoolofComputerScienceandTechnology,TianjinUniversity,China +2GermanResearchCenterforArtificialIntelligence(DFKI),Saarbrücken,Germany +3SaarlandUniversity,Saarbrücken,Germany +| | | | | | {shidan, | dyxiong}@tju.edu.cn | | | | | | | +| --- | ------------- | --- | -------- | --- | ---------- | ------------------- | -------------------------- | ------------ | --- | --------------------- | ---------- | --- | +| | | | Abstract | | | | performance | improvements | | on tasks | far beyond | | +| | | | | | | | theirtrainingdistribution. | | | Incontrast,supervised | | | +| | Reinforcement | | learning | | (RL)-based | post- | | | | | | | +6202 rpA 72 ]LC.sc[ 1v11052.4062:viXra +fine-tuning(SFT)isfrequentlyobservedtoinduce +| | training | often | improves | the | reasoning | perfor- | | | | | | | +| --- | -------- | ----- | -------- | --- | --------- | ------- | --- | --- | --- | --- | --- | --- | +mance of large language models (LLMs) be- degradation or forgetting of previously acquired +| | | | | | | | general-purpose | | capabilities | (Huan | et al., | 2025; | +| --- | ---------------------------------------- | ------------ | ----------- | -------- | ----- | ---------- | --------------- | --- | ------------ | ----- | ------- | ----- | +| | yond | the training | | domain, | while | supervised | | | | | | | +| | fine-tuning(SFT)frequentlyleadstogeneral | | | | | | Chuetal.,2025). | | | | | | +| | capabilities | | forgetting. | However, | | the mecha- | | | | | | | +Despitetheseconsistentempiricalfindings,why +nismsunderlyingthiscontrastremainunclear. RL-tuned models can generalize well remains +Tobridgethisgap,wepresentafeature-level +| | | | | | | | poorlyunderstood. | | UnlikeSFT,whichdistillsfull | | | | +| --- | ----------------- | --- | -------- | ----------- | ---------- | -------- | ----------------- | ------------ | --------------------------- | --------- | ------ | --- | +| | mechanistic | | analysis | methodology | | to probe | | | | | | | +| | | | | | | | reasoning | trajectories | from | a teacher | model, | RL | +| | RL generalization | | | using a | controlled | experi- | | | | | | | +mentalsetup,whereRL-andSFT-tunedmod- typicallyreliesonlyonoutcome-levelsupervision. +Fromamechanisticperspective,itisunclearhow +| | els are | trained | from | the same | base | model on | | | | | | | +| --- | ------- | ------- | ---- | -------- | ---- | -------- | --- | --- | --- | --- | --- | --- | +identical data. Leveraging our interpretabil- suchweakandindirectsignalsyieldbroad,trans- +| | ity framework, | | we | align | internal | activations | | | | | | | +| --- | -------------- | --- | --- | ----- | -------- | ----------- | --- | --- | --- | --- | --- | --- | +ferableimprovementsacrossdiversetasks. +across models within a shared feature space In this work, we address this gap through a +andanalyzehowfeaturesevolveduringpost- +feature-levelinterpretabilityframeworkdesigned +| | training. | WefindthatSFTrapidlyintroduces | | | | | | | | | | | +| --- | --------- | ------------------------------ | ----------- | --- | -------- | ----------- | -------------- | --- | --- | --------------------- | --- | --- | +| | | | | | | | to investigate | how | SFT | and RL differentially | | re- | +| | many | highly | specialized | | features | that stabi- | | | | | | | +lize early in training, whereas RL induces shape internal representations. We first employ +SparseCrosscodertoaligntheinternalactivations +morerestrainedandcontinuallyevolvingfea- +turechangesthatlargelypreservebasemodels’ of the base model with those of its RL- and SFT- +representations. Focusing on samples where tuned counterparts within a shared, interpretable +RLsucceedsbutthebasemodelfails,weiden- feature space. This alignment allows us to sys- +| | tify a | compact, | task-agnostic | | set | of features | | | | | | | +| --- | ------ | -------- | ------------- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +tematicallycompareinternalrepresentationsacross +thatdirectlymediategeneralizationacrossdi- +modelsandtotrackhowindividualfeaturesemerge, +| | verse | tasks. | Feature-level | | interventions | con- | | | | | | | +| --- | ----- | ------ | ------------- | --- | ------------- | ---- | --- | --- | --- | --- | --- | --- | +evolve,anddivergeduringpost-training. +| | firm | their causal | | role: disabling | | these fea- | | | | | | | +| --- | ---- | ------------ | --- | --------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | +However,pairwisecomparisonsaloneareinsuf- +turessignificantlydegradesRLmodels’gener- +alizationperformance,whileamplifyingthem ficientforfullycharacterizingtherelationshipbe- +improves base models’ performance. The tweenSFTandRLrepresentations. Toovercome +code is available at https://github.com/ this limitation, we then propose a three-model +danshi777/RL-generalization. +SparseCrosscoderthatjointlyalignsthebase,SFT- +1 Introduction trained, and RL-trained models within a single +sparsefeaturespace,andanovelModelAttribution +Reinforcement learning (RL) has emerged as a Score(MAS)tomeasurefeaturespecificity. This +| powerful | | paradigm | for | enhancing | | the reasoning | | | | | | | +| -------- | --- | -------- | --- | --------- | --- | ------------- | --- | --- | --- | --- | --- | --- | +unifiedrepresentationenablesdirectattributionof +capabilitiesoflargelanguagemodels(LLMs),par- +eachfeaturetoaspecifictrainingparadigm. +ticularlyinsolvingcomplexlogicaltasksinvolving Usingthisframework,weconductasystematic +mathematicsandprogramming(Guoetal.,2025; +| | | | | | | | analysis | of feature | dynamics | throughout | training. | | +| ----- | --- | ---------- | ---- | --- | ----------- | -------- | -------- | ---------- | -------- | ---------- | --------- | --- | +| Jaech | et | al., 2024; | Team | et | al., 2025). | Notably, | | | | | | | +Ourresultsrevealaclearandconsistentdistinction +modelsoptimizedviaRLonnarrowlydefinedrea- between the two paradigms. SFT rapidly intro- +| soning | objectives | | often | demonstrate | | substantial | | | | | | | +| ------ | ---------- | --- | ----- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +ducesalargenumberofhighlyspecializedfeatures +*Correspondingauthor thatstabilizeearlyintraining,whereasRLinduces + +morerestrainedandcontinuouslyevolvingfeature asaprominentdirection(Daietal.,2022;Shietal., +changes,largelypreservingthebasemodel’srep- 2024a;Chenetal.,2024;Wuetal.,2023;Shietal., +resentational structure. This difference provides 2024b). Recent work has explored feature-level +aninitialrepresentationalexplanationforwhySFT interpretability in LLMs, often using Sparse Au- +tendstomemorizetask-specificpatterns,whileRL toencoders(SAEs,Cunninghametal.,2023;Gao +maintainsbroadercapabilities. etal.,2024a;Marksetal.,2025)toidentifyinternal +Beyond descriptive analysis, our framework features associated with specific functions, such +enables direct mechanistic investigation of RL- as emotions (Han et al., 2025), safety behaviors +inducedgeneralization. Weidentifyacompactset (Weng et al., 2025; Yeon et al., 2025), language- +ofinternalfeaturesthatactivelycontrolcross-task specific representations (Deng et al., 2025), and +transfer by focusing on samples where general- reasoningprocesses(Galichinetal.,2025). Sparse +ization explicitly occurs, cases in which the RL- Crosscoders extend SAEs, have further enabled +tunedmodelsucceedswhilethebasemodelfails, comparisons between fine-tuned and base mod- +andmeasuringfeature-levelactivationdifferences els,revealingfeaturesintroducedbySFT(Lindsey +withinthealignedfeaturespace. etal.,2024;BaekandTegmark,2025). Incontrast, +Throughtargetedfeatureinterventions,wefur- ourworkcomparesbothSFT-andRL-trainedmod- +ther demonstrate that these features are causally elsagainstthesamebasemodelandintroducesa +responsible for generalization: disabling them in three-model sparse crosscoder to jointly analyze +RL-tunedmodelsleadstosignificantperformance theirfeaturedifferences. +degradation,whileamplifyingtheminbasemodels +Generalization Comparison of SFT and RL. +induces substantial gains, even on unseen tasks. +Severalstudieshavecomparedthegeneralization +These results indicate that RL does not merely +of SFT- and RL-tuned models at the behavioral +improve task-specific performance, but instead +level,documentingperformancegainsandlosses +strengthensacompact,task-agnosticsetoffeatures +acrossdiversetasks. Theseworksconsistentlyre- +thatgovernsgeneralizationbehavior. +portthatRL-tunedmodelsgeneralizemorerobustly +Insummary,ourcontributionsareasfollows: +across domains, while SFT often leads to forget- +• We propose a feature-level interpretability tingofgeneralcapabilities(Huanetal.,2025;Chu +frameworkformechanisticallyanalyzinghow et al., 2025). However, the internal mechanisms +differentpost-trainingparadigmsreshapein- underlyingthiscontrastremainlargelyunexplored. +ternalrepresentationsandgiverisetoRLgen- Although Huan et al. (2025) attempt to explain +eralization. this difference by relating it to larger representa- +tion or output distribution drift induced by SFT, +• Wepresentathree-modelSparseCrosscoder +theiranalysisremainsatthelevelofglobalrepre- +andaModelAttributionScore(MAS)thaten- +sentationsanddoesnotidentifythespecificinter- +ableunifiedfeaturealignmentandattribution +nalmechanismsthatcausallydrivegeneralization. +acrossbase,SFT,andRLmodels. +Ourworkaddressesthisgapbyprovidingafeature- +level,causallyvalidatedexplanationofRL-induced +• We reveal a fundamental distinction be- +generalization. +tweenpost-trainingparadigms: SFTinduces +early-stabilized, highly specialized features, +3 InterpretabilityMethodology +whereasRLyieldsmorerestrainedandcontin- +uouslyevolvingfeaturechanges,andweiden- To systematically analyze how different post- +tifyasmallsetofgeneralization-controlling trainingparadigmsreshapeinternalrepresentations, +featuresthatprovideadirectmechanisticex- we introduce a unified feature-level interpretabil- +planationforRLgeneralization. ityframework. Theframeworkisguidedbythree +core principles. First, the models under compari- +2 RelatedWork +sonmustbestrictlycomparable,suchthatobserved +Feature-Level Interpretability in LLMs. As representationaldifferencescanbeattributedsolely +LLMscontinuetoexhibitdiverseandsophisticated to the training paradigm. Second, internal repre- +capabilities(Guoetal.,2023;Changetal.,2024; sentationsfromdifferentmodelsmustbealigned +Shietal.,2024c),understandingtheinternalmecha- into a shared feature space to enable direct, fine- +nismsthatgiverisetothesebehaviorshasemerged grainedcomparisonoffeatures. Third,theframe- + +workshouldnotonlysupportdescriptiveanalysis sparsityregularizationL : +sparsity +offeaturedifferences,butalsoprovidedirectevi- +L = L +βL , +denceforthegeneralizationcapacityofRL. recon sparsity +Tosatisfytheseprinciples,wedesignthreekey L = (cid:88) (cid:13) (cid:13)a(i)−aˆ(i) (cid:13) (cid:13) 2 , +recon (cid:13) (cid:13) +componentsinourinterpretabilityframework: (i) (3) +i=O,T +sparsecrosscodersforfeature-levelalignmentand L = (cid:88) f (x ) (cid:88) (cid:13) (cid:13)W (i) (cid:13) (cid:13). +attributionbetweenthebasemodelandtunedmod- sparsity k j (cid:13) dec,k(cid:13) +els,(ii)athree-modelextensionthatenablesjoint k i=O,T +comparison across base, SFT, and RL models, Thisobjectiveforcesthemodeltolearnasmallset +and (iii) a method for identifying generalization- of interpretable features that capture the distinct +controllingfeatures. Weintroduceeachcomponent properties of the activations. Within our frame- +inturnbelow. work,thistwo-modelsparsecrosscoderservesas +thebasicbuildingblockforpairwisecomparison +3.1 Feature-LevelAlignmentandAttribution betweenatunedmodelanditsbasecounterpart. +viaSparseCrosscoders +IdentifyingModel-SpecificFeatureswithNRN +Acentralchallengeincomparinginternalrepresen- Toquantifyhowstronglyeachfeatureisuniqueto +tationsacrossindependentlytrainedmodelsisthat eachmodel,wefollowBaekandTegmark(2025) +theiractivationspacesarenotdirectlyaligned. To usingtheNormalizedRelativeNorm(NRN)metric +address this challenge, we employ Sparse Cross- to analyze the features. NRN is computed as the +coders (Lindsey et al., 2024) to align activations ratio between the L1 norm of the decoder vector +fromthedifferentmodelsintoasharedsparsefea- foreachmodel: +turespace,enablingdirect,meaningfulcomparison +(cid:13) (cid:13) +ofrepresentationalchangesinducedbytuning. +(cid:13)W (T) (cid:13) +(cid:13) dec,k(cid:13) +NRN = (cid:13) (cid:13) (cid:13) 1 (cid:13) , (4) +(cid:13)W (O) (cid:13) +(cid:13)W (T) (cid:13) +Two-Model Sparse Crosscoder for Compari- (cid:13) dec,k(cid:13) (cid:13) dec,k(cid:13) +1 1 +sonofTunedModelsandBaseModel. Sparse +where O denotes the original base model, T rep- +CrosscodersextendSAEsbyjointlyencodingac- +(·) +resents the tuned model (SFT or RL), and W +tivationsfromdifferentsources,suchasdifferent dec +denotes the corresponding crosscoder’s decoder. +models, layers, or positions, into a shared sparse +Intuitively,ifafeaturecontributesmoretorecon- +feature space. This formulation enables a fine- +structingthetunedmodel’sactivations,thedecoder +grained comparison of representational changes +inducedbytuning. +assignsitalargernorminW ( +d +T +ec +) ,yieldingNRN→ +1. Inthiscase,NRN→1correspondstofeatures +Formally,theencodercomputesfeatureactiva- +thatareuniquetothetunedmodel,NRN→0cor- +tionsas: +respondstofeaturesuniquetotheoriginalmodel, +  whileNRN=0.5correspondstosharedfeatures. +(cid:88) +f(x +j +) = ReLU W( +e +i +n +) +c +a(i)(x +j +)+b enc, +Three-Model Sparse Crosscoder for Joint +i∈O,T Representation Comparison. The two-model +(1) +Sparse Crosscoder enables fine-grained compar- +and the decoder reconstructs the activations for +isonsbetweenatunedmodelanditsbasecounter- +eachmodel: +part. However, such pairwise analyses are inher- +entlylimitedwhenmultipletrainingparadigmsare +aˆ(i)(x ) = W (i) f(x )+b (i) . (2) +j dec j dec involved. In particular, when comparing an SFT- +tuned model and an RL-tuned model against the +Here, O denotes the original base model and T samebasemodel,pairwisecrosscoderscannotdis- +represents the tuned model (either SFT or RL). entanglewhetherafeatureissharedbyallmodels, +a(i)(x j ) ∈ Rd model is the residual-stream activa- specifictoonetunedmodel,orsharedbythetwo +tion of model i at token x j , f(x j ) ∈ Rdsparse is tunedmodelsbutabsentinbasemodel. +the sparse feature activations, and aˆ(i)(x ) is the Moreimportantly,featuresidentifiedindifferent +j +reconstructedactivation. Thecrosscoderistrained pairwisecrosscodersarenotdirectlycomparable. +by minimizing a reconstruction loss L with For example, a feature with a given index (e.g., +recon + +(cid:13) (cid:13) +feature #2026) in the SFT-Base crosscoder does (cid:13)W(S) (cid:13) +MAS = +(cid:13) dec,k(cid:13) +1 , (9) +notnecessarilycorrespondtothefeaturewiththe S (cid:13) (cid:13)W(O) (cid:13) (cid:13) + (cid:13) (cid:13)W(S) (cid:13) (cid:13) + (cid:13) (cid:13)W(R) (cid:13) (cid:13) +same index in the RL-Base crosscoder, since the +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) +1 +(cid:13) (cid:13) +twocrosscodersaretrainedindependentlyandmay MAS = (cid:13) (cid:13) W( d R ec ) ,k (cid:13) (cid:13) 1 . +learnentirelydifferentsparsebases. R (cid:13) (cid:13)W(O) (cid:13) (cid:13) + (cid:13) (cid:13)W(S) (cid:13) (cid:13) + (cid:13) (cid:13)W(R) (cid:13) (cid:13) +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) +1 +Toaddressthislimitation,weintroduceathree- (10) +modelSparseCrosscoder,whichjointlyalignsthe Here,O,S,andRdenotetheoriginalbasemodel, +base,SFT-tuned,andRL-tunedmodelswithinasin- the SFT-tuned model, and the RL-tuned model, +glesparsefeaturespace. Byunifyingallthreemod- respectively. By construction, MAS , MAS , +O S +els within a single sparse basis, the three-model MAS ∈ [0,1], MAS +MAS +MAS = 1. +R O S R +sparsecrosscoderconstitutesakeycomponentof ThemodelwiththelargestMASvalueistheoneto +ourinterpretabilityframework,enablingdirect,si- whichthefeatureismoststronglyattributed. For +multaneouscomparisonoffeaturesharingandspe- example, a feature with MAS → 1 is an SFT- +S +cializationacrosspost-trainingparadigms. specificfeature. +Specifically,givenatokenx ,wejointlyencode +j +3.2 IdentifyingGeneralization-Controlling +theresidual-streamactivationsfromthethreemod- +Features +elsintoasharedsparsefeaturerepresentation: +  Thefinalcomponentofourinterpretabilityframe- +(cid:88) workaimstomovebeyonddescriptivecomparison +f(x +j +) = ReLU W( +e +i +n +) +c +a(i)(x +j +)+b enc, +andtoidentifyinternalfeaturesthatcausallycon- +i=O,S,R +trolgeneralizationbehavior. Therefore,wehypoth- +(5) +esize that RL selectively strengthens a subset of +whereO,S,andRdenotetheoriginalbasemodel, +featuresthatplayacausalroleinenablinggeneral- +theSFT-tunedmodel,andtheRL-tunedmodel,re- +spectively. a(i)(x ) represents the activation of izationacrosstasks. +j +Prior work has attempted to localize function- +modeli,andf(x )isthesharedsparsefeaturevec- +j +ally meaningful features by analyzing which fea- +tor. Then,thedecoderreconstructstheactivations +tures are frequently activated on specific lexical +foreachmodel: +cues, such as identifying self-reflection features +aˆ(i)(x j ) = W ( d i e ) c f(x j )+b ( d i e ) c . (6) via activations on tokens like “Wait”, or con- +trastive features via “But” and “However” (Baek +Thetrainingobjectiveminimizesacombination +andTegmark,2025;Galichinetal.,2025). While +ofreconstructionlossandsparsityregularization: +such approaches are useful for interpretability, +L = (cid:88) (cid:13) (cid:13)aˆ(i)−a(i) (cid:13) (cid:13) 2 theyimplicitlyassumethatfunctionalfeaturesare +(cid:13) (cid:13) +tightly coupled to surface-level tokens. In con- +i=O,S,R +(7) ++ (cid:88) f (x ) (cid:88) (cid:13) (cid:13)W (i) (cid:13) (cid:13). trast,wethinkthatfeaturescontrollinggeneral- +k j (cid:13) dec,k(cid:13) izationshouldnotdependonspecificwordsor +k i=O,S,R task-specific lexical patterns. Instead, such fea- +Thisobjectiveencouragesthemodeltodiscover turesshouldbeidentifiablethroughtheirfunctional +acompactsetofsparsefeaturesthatjointlyexplain role,namely,whethertheysystematicallyalterthe +theactivationsofallthreemodels,whileallowing model’sdecision-makingbehavioracrosstasksin +each feature to contribute unequally to different situationswheregeneralizationactuallyoccurs. +models. We propose to localize generalization-related +featuresbyfocusingonsamplesthatexplicitlyin- +MeasuringFeatureSpecificitywithMAS. To +stantiate generalization behavior. For each task, +quantifyhowstronglyeachfeatureisattributedto +we construct a subset of samples on which the +a particular model within the shared crosscoder +basemodelfailsbuttheRL-tunedmodelsucceeds. +space,wedefineathree-waynormalization,which +Thesesamplesrepresenttheminimalevidenceof +werefertoastheModelAttributionScore(MAS). +generalizationandaretheonlyinstanceswheregen- +For each feature k, the formula for calculating +eralization can be unambiguously observed. We +MASisasfollows: +refertothemasgeneralization-criticalsamples. +(cid:13) (cid:13) +MAS = +(cid:13) +(cid:13) +W( +d +O +ec +) +,k +(cid:13) +(cid:13) 1 , (8) Foreachgeneralization-criticalsample,weex- +O (cid:13) (cid:13)W(O) (cid:13) (cid:13) + (cid:13) (cid:13)W(S) (cid:13) (cid:13) + (cid:13) (cid:13)W(R) (cid:13) (cid:13) tractthe residual-streamactivation atthe finalto- +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) +1 +(cid:13) dec,k(cid:13) 1 - d - -- - -type - - injury. - -Read all instructions before using the dishwasher. -The manual does not cover every possible -condition and situation that may occur . -Use the dishwasher only for intended function as -described in this manual. -When loading items to be washed: --Load sharp items and knives so that they are not -likely to damage the door seal and tub. --Load sharp items and knives with the handles up to -reduce the risk of cut -Do not wash plastic items unless marked -ishwasher safe or the equivalent, if not marked, -check with manufacture for recommendation, items -that are not dishwasher safe may melt and create a -potential fire hazard. -If the dishwasher drains into a food disposer, - disposer is completely empty before -make -sure - . -running dishwasher -Do not tamper with control. -Do not operate your dishwasher unless all -enclosure panels are in the properly place. -Do not touch the heating element during or -, -immediately after use especially if -has been selected. - - the sanitize option - -2 - -. - +| | | | MathReasoningTasks | | | | | | GeneralTasks | | | | | +| --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | +Model +MATH500 AIME24 AIME25 OpenBookQA CommonsenseQA HeadQA SciQ ARC-Challenge +| Qwen3-4B-Base | | | 26.0 | 13.3 | 0.0 | | 23.6 | 20.1 | | 31.7 | 78.5 | 36.0 | | +| -------------- | --- | --- | ---- | ---- | ---- | --- | ---- | ---- | --- | ---- | ---- | ---- | --- | +| Qwen3-4B-SFT | | | 68.4 | 13.3 | 13.3 | | 25.8 | 19.6 | | 31.2 | 51.8 | 34.4 | | +| Qwen3-4B-RL | | | 77.0 | 26.7 | 20.0 | | 27.2 | 50.5 | | 32.7 | 89.5 | 39.3 | | +| ∆(RL-SFT) | | | 8.6 | 13.4 | 6.7 | | 1.4 | 31.0 | | 1.5 | 37.7 | | 4.9 | +| Qwen2.5-7B | | | 40.0 | 10.0 | 3.3 | | 28.4 | 77.6 | | 33.7 | 86.9 | 41.9 | | +| Qwen2.5-7B-SFT | | | 69.2 | 13.3 | 10.0 | | 26.4 | 30.1 | | 31.2 | 79.4 | 37.0 | | +| Qwen2.5-7B-RL | | | 71.4 | 20.0 | 13.3 | | 32.8 | 76.1 | | 36.1 | 90.7 | 42.5 | | +| ∆(RL-SFT) | | | 2.2 | 6.7 | 3.3 | | 6.4 | 46.0 | | 5.0 | 11.3 | | 5.5 | +Table1: PerformancecomparisonofSFT-andRL-tunedmodelsonmathreasoningtasksandothertasks. +kenpositionfromthebasemodelandtheRL-tuned 2025) and Qwen2.5-7B (Yang et al., 2024) mod- +model, respectively. Then we encode them sep- els, respectively. Details about training datasets, +arately using the model-specific branches of the implementationspecifics,andhyperparametersare +trainedSparseCrosscoderencoder,yieldingfeature providedinAppendixA.2. +vectorsf(RL)(x)andf(Base)(x),respectively. +| | | | | | | | EvaluationBenchmarks. | | | | WeevaluatedtheSFT- | | | +| --- | --------- | --- | ----- | ---------- | -------------- | --- | --------------------- | --- | --- | --- | ------------------ | --- | --- | +| For | a feature | | k, we | define its | generalization | | | | | | | | | +andRL-tunedmodelsonadiversesuiteofbench- +scoreonagiventaskastheaverageactivationdif- +marks,spanningbothmathematicalreasoningand +ferencebetweentheRLmodelandthebasemodel +| | | | | | | | other | general | tasks. | Specifically, | | the mathemat- | | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------- | ------ | ------------- | --- | ------------- | --- | +acrossallgeneralization-criticalsamples: +| | | | | | | | ical reasoning | | benchmarks | | include | MATH500 | | +| ----- | --- | --- | --------- | ----- | ------ | --------- | --------------------------------------- | ------- | ---------- | ------------- | ---------- | ------- | ------- | +| | | | (cid:104) | | | (cid:105) | | | | | | | | +| | | E | (RL) | | (Base) | | (Hendrycksetal.,2021),AIME24,andAIME25, | | | | | | | +| Score | | = | f | (x)−f | (x) | , (11) | | | | | | | | +| | k | x∈G | k | | k | | | | | | | | | +| | | | | | | | while | other | tasks | comprise | OpenBookQA | | (Mi- | +| | | | | | | | haylov | et al., | 2018), | CommonsenseQA | | | (Talmor | +whereG denotesthesetofgeneralization-critical +| | | | | (·) | | | et al., | 2019), | HeadQA | | (Vilares | and | Gómez- | +| ----------------------- | --- | --- | --- | ------------------ | --- | --- | ------- | ------ | ------ | --- | -------- | --- | ------ | +| samplesforthattask,andf | | | | (x)istheactivation | | | | | | | | | | +k +| | | | | | | | Rodríguez, | | 2019), | SciQ (Welbl | | et al., 2017), | and | +| ------------------------------------- | --- | ----- | -------- | -------- | ----- | --------- | ------------------------------- | --- | ---------- | ----------- | -------------- | -------------- | ------- | +| of feature | | k for | input x. | For each | task, | we retain | | | | | | | | +| | | | | | | | ARC-Challenge(Clarketal.,2018). | | | | | Detailedde- | | +| featureswhosescoresexceedathresholdt. | | | | | | These | | | | | | | | +| | | | | | | | scriptions | of | benchmarks | | and evaluation | | metrics | +featuresaretreatedastask-relevantgeneralization +canbefoundinAppendixA.4. +| features. | | Finally, | we take | the intersection | | of the | | | | | | | | +| --------- | --- | -------- | ------- | ---------------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +selectedfeaturesetsacrossalltasks,yieldingaset +| | | | | | | | Results. | As | shown | in Table | 1, | we observe | that | +| --- | --- | --- | --- | --- | --- | --- | -------- | --- | ----- | -------- | --- | ---------- | ---- | +offeaturesthatconsistentlycontributetosuccessful +| | | | | | | | RL-tuned | models | | exhibit | strong | generalization | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------ | --- | ------- | ------ | -------------- | --- | +generalizationregardlessoftaskdomain. acrossdiversedomains,whereasSFT-tunedmodels +In summary, the features we identify are those sometimessufferfromgeneralcapabilitiesforget- +that(i)systematicallydifferentiatetheRLmodel +ting,consistentwithobservationsreportedinprior +from the base model on generalization-critical work(Huanetal.,2025). +| samples, | | (ii) influence | | model behavior | | in a task- | | | | | | | | +| -------- | --- | -------------- | --- | -------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | +agnostic manner, and (iii) do not rely on explicit 5 Experiments: ComparingSFTandRL +| taskknowledgeorlexicaltriggers. | | | | | | | ModelsviaTwo-ModelSparse | | | | | | | +| ------------------------------- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | --- | --- | --- | +Crosscoders +| 4 | Phenomena: | | Performance | | | | | | | | | | | +| --- | ---------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +DiscrepanciesofReasoningModels ToanalyzehowSFTandRLaltertheinternalrep- +resentationsofthebasemodel,weindependently +To ensure that differences between the resulting trainedtwoSparseCrosscoders: onebetweenthe +| reasoning | | models | can be | attributed | solely | to the | | | | | | | | +| --------- | --- | ------ | ------ | ---------- | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | +SFT-tunedmodelandthebasemodel,andanother +trainingparadigmratherthanconfoundingfactors betweentheRL-tunedmodelandthebasemodel. +such as data composition, we deliberately avoid Implementation and training details are provided +| using | existing | off-the-shelf | | distilled | or | RL-tuned | inAppendixA.1. | | | | | | | +| ----- | -------- | ------------- | --- | --------- | --- | -------- | -------------- | --- | --- | --- | --- | --- | --- | +models. Instead,wetrainedboththeSFTandRL +5.1 FeatureDiscrepanciesBetweenRLand +modelsfromthesamebasemodelonanidentical +SFT +datasetusingfull-parametertuning,ensuringthat +any observed differences can be attributed solely Using the trained crosscoders, we computed the +tothetrainingparadigms. Weperformedthiscon- NRNs for both the SFT- and the RL-tuned mod- +trolledtuningontheQwen-3-4B-Base(Yangetal., elsrelativetotheoriginalbasemodel,denotedas + +| | Qwen3-4B-SFT | | | Qwen3-4B-RL | | | | Qwen2.5-7B-SFT | | | | Qwen2.5-7B-RL | | +| ----- | ------------ | --- | ----- | ----------- | --- | --- | ----- | -------------- | --- | ----- | --- | ------------- | --- | +| 104 | | | 104 | | | | 104 | | | | 104 | | | +| 103 | | | 103 | | | | 103 | | | | 103 | | | +| tnuoC | | | tnuoC | | | | tnuoC | | | tnuoC | | | | +| 102 | | | 102 | | | | 102 | | | | 102 | | | +| 101 | | | 101 | | | | 101 | | | | 101 | | | +| 100 | | | 100 | | | | 100 | | | | 100 | | | +0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 +Normalized Relative Norm Normalized Relative Norm Normalized Relative Norm Normalized Relative Norm +Figure1: DistributionofNormalizedRelativeNormsacrossdifferenttrainingmethodsanddifferentmodelscales. +NRN SFT andNRN RL ,respectively. Theresulting largelypreservestheoriginalmodel’srepresenta- +NRNdistributionsarevisualizedinFigure1. tional structure. This pattern can be attributed to +Acrossbothtrainingparadigms,themajorityof thenatureofRLsupervision. UnlikeSFT,which +featuresclusteraroundNRN≈ 0.5,indicatingthat directlyconstrainstheentirereasoningtrajectory, +most features are shared between the fine-tuned RL only provides outcome-level feedback based +modelsandthebasemodel. Moreover,thenumber onfinalanswercorrectness. Asaresult,RLdoes +offeaturesdecaysapproximatelyexponentiallyto- not force the model to adopt a particular reason- +wardbothextremesofthedistribution,suggesting ing style or surface form. Instead, it selectively +that highly model-specific features are relatively reinforcesinternalcomputationsthatcontributeto +rare. However, thetailsofthedistribution, corre- correctdecisions,whileleavingmuchofthebase +spondingtomodel-specificfeatures,exhibitstrik- model’srepresentationalstructureintact. +inglydifferentbehaviorsforSFTandRL. Overall, these results suggest a fundamental +| | | | | | | | contrast | in post-training | | | dynamics: | SFT | drives | +| --- | ------- | ------- | ------ | --------- | --- | ---- | ----------- | ---------------- | --- | --------- | ---------- | --- | ------ | +| SFT | Induces | a Large | Number | of Unique | | Fea- | | | | | | | | +| | | | | | | | substantial | feature | | turnover, | generating | | many | +tures. FortheSFT-tunedmodel,theNRNdistri- +| | | | | | | | highly | model-specific | | features | while | erasing | oth- | +| --- | --- | --- | --- | --- | --- | --- | ------ | -------------- | --- | -------- | ----- | ------- | ---- | +butionexhibitsapronouncedrighttail,withasub- ers, whereas RL induces restrained and targeted +| stantialnumberoffeaturesachievingNRN | | | | | | > | | | | | | | | +| ------------------------------------ | ---- | ---------------- | --- | --- | --- | ---- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | | | SFT | | adjustmentsthatlargelypreservetheoriginalfea- | | | | | | | +| 0.8, and | some | even approaching | | NRN | | = 1. | | | | | | | | +| | | | | | SFT | | turespace. | | | | | | | +Thesefeaturescorrespondtorepresentationsthat +are almost entirely unique to the SFT model. At 5.2 FeatureFormationandEvolutionDuring +| the same | time, | we also | observe | a non-negligible | | | | Training | | | | | | +| -------- | ----- | ------- | ------- | ---------------- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | +numberoffeatureswithNRN → 0,indicating TofurtherunderstandhowSFTandRLalterthein- +SFT +featuresthatareeffectivelyexclusivetotheorigi- ternalrepresentationsofthebasemodelovertime, +| nalmodel. | | ThesetwoextremessuggestthatSFT | | | | | | | | | | | | +| --------- | --- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +weanalyzedthetemporalevolutionoffeaturesdur- +inducesapronouncedrepresentationalshift: while ingtraining. Ratherthanfocusingsolelyonthefi- +introducingmanynew,highlyspecializedfeatures, naltunedmodels,weexaminedhowmodel-specific +italsosuppressesorabandonsaconsiderablepor- +featuresemerge,persist,andchangeacrosstraining +tionofthebasemodel’sfeaturerepertoire. +checkpoints. +| Furthermore, | | the Qwen2.5-7B | | crosscoders | | ex- | | | | | | | | +| ------------ | -------- | -------------- | ------------- | ----------- | -------- | --- | ------------ | ------ | --------- | ----------- | ------------ | ---- | -------- | +| | | | | | | | Experimental | | Setup. | | For | each | training | +| hibit | a larger | number | of features | with | extreme | | | | | | | | | +| | | | | | | | paradigm, | we | saved | checkpoints | | at | regular | +| NRN | values | than the | Qwen3-4B-Base | | counter- | | | | | | | | | +| | | | | | | | intervals | (every | one-fifth | | of an epoch) | | and pair | +parts,implyingthatlargermodelstendtodevelop +| | | | | | | | each | checkpoint | with | the | base model | | to train a | +| -------- | --- | -------------- | -------- | -------- | ------ | --- | ------ | ----------- | ---- | -------- | ---------- | ----------- | ---------- | +| a richer | set | of distinctive | internal | features | during | | | | | | | | | +| | | | | | | | Sparse | Crosscoder, | | yielding | five | crosscoders | per | +post-training. +| | | | | | | | paradigm. | Thissetupenablesstage-wisecompar- | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --------- | --------------------------------- | --- | --- | --- | --- | --- | +RLPreservesCoreRepresentationsWhileIntro- isonsbetweenthebasemodelandpartiallytrained +ducing Fewer, Milder Deviations. In contrast, models. Using these crosscoders, we conducted +theRL-tunedmodelexhibitsamarkedlydifferent twocomplementaryanalyses: (1)featureoverlap +NRN profile. Only a small number of features across checkpoints, and (2) feature rank shifts +exhibit extreme NRN RL values (close to either 0 betweenconsecutivecheckpoints. Theseanalyses +or 1), with very few approaching either end of allow us to quantify both the stability of learned +the spectrum. This indicates that RL introduces featuresandtheextenttowhichtheinternalfeature +relatively few novel, model-specific features and spaceisreorganizedduringtraining. + +Qwen3-4B-SFT Qwen3-4B-RL Qwen2.5-7B-SFT Qwen2.5-7B-RL +1.0 1.0 1.0 1.0 +ckpt-1 1.00 0.28 0.22 0.30 0.25 ckpt-1 1.00 0.06 0.06 0.06 0.08 ckpt-1 1.00 0.28 0.20 0.30 0.23 ckpt-1 1.00 0.04 0.05 0.03 0.05 +0.8 0.8 0.8 0.8 +ckpt-2 0.28 1.00 0.37 0.33 0.32 ckpt-2 0.06 1.00 0.12 0.10 0.08 ckpt-2 0.28 1.00 0.19 0.19 0.18 ckpt-2 0.04 1.00 0.15 0.33 0.18 +0.6 0.6 0.6 0.6 +ckpt-3 0.22 0.37 1.00 0.43 0.35 ckpt-3 0.06 0.12 1.00 0.10 0.08 ckpt-3 0.20 0.19 1.00 0.35 0.28 ckpt-3 0.05 0.15 1.00 0.19 0.32 +0.4 0.4 0.4 0.4 +ckpt-4 0.30 0.33 0.43 1.00 0.47 ckpt-4 0.06 0.10 0.10 1.00 0.08 ckpt-4 0.30 0.19 0.35 1.00 0.39 ckpt-4 0.03 0.33 0.19 1.00 0.23 +0.2 0.2 0.2 0.2 +ckpt-5 0.25 0.32 0.35 0.47 1.00 ckpt-5 0.08 0.08 0.08 0.08 1.00 ckpt-5 0.23 0.18 0.28 0.39 1.00 ckpt-5 0.05 0.18 0.32 0.23 1.00 +0.0 0.0 0.0 0.0 +ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 +Figure2: Featureoverlapheatmapsacrosstrainingcheckpointsunderdifferenttrainingparadigms. +104 +103 +102 +101 +100 +0.0 0.2 0.4 0.6 0.8 1.0 +Model Attribution Score +tnuoC +Qwen3-4B-Base +104 +103 +102 +101 +100 +0.0 0.2 0.4 0.6 0.8 1.0 +Model Attribution Score +tnuoC +Qwen3-4B-SFT +104 +103 +102 +101 +100 +0.0 0.2 0.4 0.6 0.8 1.0 +Model Attribution Score +tnuoC +Qwen3-4B-RL +Figure3: DistributionofModelAttributionScoresacrossdifferenttrainingmethodsonQwen3-4B-Base. +SFTQuicklyEstablishesaRelativelyFixedSet servethatSFTrapidlyformsacomparativelysta- +of Internal Features, While RL Promotes a blehierarchyofmodel-specificfeatures,withlater +Slower and More Incremental Process of Fea- trainingprimarilyrefiningtheirrelativeimportance. +tureFormation. Foreachcheckpoint,weranked In contrast, RL induces features more gradually, +featuresbydescendingNRNandretainthetop50 withtrainingpersistentlyreshapingwhichinternal +as the most distinctive (i.e., model-specific) fea- featuresaremostsalientforgeneratingcorrectout- +tures at that stage. We then computed the over- comes. Visualizationresultsanddetailedanalyses +lap of these top-ranked features across different areprovidedinAppendixB.1. +checkpoints. AsshowninFigure2,theSFT-tuned Toassesswhetherourfindingsgeneralizeacross +model exhibits a high degree of feature overlap model families, we further conduct controlled +across checkpoints. A substantial portion of the post-training and feature-level analysis on a dif- +top-rankedSFT-specificfeaturesidentifiedatearly ferentarchitecture,namelyLlama3.1-8B-Instruct +checkpointspersiststhroughoutlaterstagesoftrain- (Grattafiori et al., 2024). The corresponding re- +ing. Thisindicatesthatmanyofthefeaturesintro- sultsarepresentedinAppendixB.2. Weobserve +ducedbySFTemergeearlyandremainconsistently consistentpatternswiththosefoundintheQwen +dominantduringsubsequentoptimization. family,indicatingthattherepresentationaldistinc- +In contrast, the RL-tuned model displays neg- tionsbetweenSFTandRLarenotmodel-specific +ligible overlap between the top-ranked feature butbroadlyapplicableacrossarchitectures. +sets of adjacent checkpoints. Features that are +Inconclusion,thecontrastbetweenSFTandRL +highlyrankedatanearliercheckpointrarelyremain +featuresrevealstwofundamentallydifferentmodes +among the top features at the subsequent check- +ofrepresentationchange: +point. Only toward the end of training does the +overlapincrease,suggestingthatRLdelaysfeature +• SFT rapidly introduces a large number of +consolidationandinsteadexploresabroadersetof +model-specificfeaturesthatcloselyreplicate +candidatefeaturesbeforegraduallystabilizingits +the teacher’s reasoning traces, while simul- +featurecomposition. +taneously discarding many pre-existing fea- +SFT Primarily Refines Feature Strengths Af- tures. Aftertheformation,theoverallfeature +ter Early Formation, While RL Continuously composition remains largely stable. Subse- +ReordersFeatureImportance Beyondfeature quent optimization primarily adjusts the rel- +persistence,wefurtherexaminedhowtherelative ativestrengthsofthesefeatures. Thisyields +importanceoffeaturesevolvesbyanalyzingrank a rigid and specialized feature space tightly +shifts between consecutive checkpoints. We ob- alignedwiththetrainingdistribution. + +1.0 1.0 +Com +O +m +p +o +e +n +n +s +B +e +o +n +o +se +kQ +Q +A +A 1 +0 . - . - - gas is - - using - - gas may be - - hot water system has not been used -, - on all - a period before - -Do not let children abuse sit, or stand on door or -racks of a dishwasher. -, -Under certain conditions hydrogen -produced in a hot water system that has not been -used for two weeks or more Hydrogen -explosive If the -for such -hot water faucets and let water flow from each for -several minutes. This will release any accumulated - gas is flammable. Do not -hydrogen gas Hydrogen -smoke or use an open flame during this time. -Do not store or use combustible materials, gasoline -or other flammable vapors and liquids in the vicinity -of this or any other appliance. -Use only detergents or rinse agents recommended -for use in a dishwasher and keep them out of the -reach of children. -Keep young children and infants away from -dishwasher when it is operating. -Do not use the dishwasher if it has a damaged -power line or plug, and do not plug the dishwasher -into a damaged outlet, failure to observe these -instructions may result in electrical shock. -Remove the door to the washing compartment when -removing an old dishwasher from service or discarding -it. - -Important safety instruction - -WARNING - -For your safety, the information in this manual must be followed to -minimize the risk of fire or explosion, electric shock, and to prevent -property damage, personal injury. - -Proper installation - -Please install your dishwasher properly follow the installation guide. - -Water supply temperature should be 120℉ or 149 .℉ -Dispose of the discarded appliance and packing material properly. -The dishwasher must be grounded, or it can result in a risk of electric shock. -If there is any damage of the -repair or replace any part by yourself. - - contact to your - - to attempt to - -Basic precaution before use - - Features of your new dishwasher - -Specification - -Capacity - -Dimension - -Weight - -14 place settings - -606 x 650 x 857 mm (W x D x H) - -unpacked 94lb (42.5 kg) - -Power Supply - -120volts - -, - -60Hz - -Rated power usage - -Wash motor 50W Heater - - 840W - -Water Feed Pressure - -20 ~ 120 psi (138 ~ 828kPa) - -9 - -10 - -11 - -12 - +0 +6 +0 +8 +0 1 - -2 - -3 - +. +. +6 +0 +8 +0 +0 +0 +. +. +5 +5 +0 +4 +0 +0 +. +. 4 - 5 - +9 +7 +0 +0 +. +. 6 - 7 - 8 - -1 - -2 - -3 - 4 - -5 - +0 +0 +. +. 6 - -Top Spray Arm - -Cup Shelf - -Upper Spray Arm - -Lower Spray Arm - -Bottom Rack - -Heating Element - -7 - +8 Com +O +m +p +o +e +n +n +s +B +e +o +n +o +se +kQ +Q +A +A 1 +0 +. +. +0 8 - -9 - -10 - -11 - -Rinse Aid Dispenser - -Detergen t Dispenser - -Tub Gasket - -Top Rack - -Adjustable Upper Rack - -12 - -Silverware Basket - -3 - - Features of your new dishwasher - -Features - -Filter System - the -Your dishwasher has a multiple Filter - .It can separate -- -System there are four filter meshes -soiled water and clean water in different chambers. -The multiple filter system helps your dishwasher to get -better performance with less water and less energy . - -Adjustable Upper Rack -The Upper Rack can be raised or lowered to -accommodate loading taller dishes in either rack -The -Upper Rack clearance height H1 goes from 8" to 10". -The Bottom Rack clearance height H2 is 11" to 13". - +0 +4 +0 +1 . - - a - -Smart Wash System -Your dishwasher has Smart Wash System. -partial load of lightly soiled dishes is placed in the -unit, a wash cycle similar to a short wash will -automatically be performed. When a full load of -heavily soiled dishes is placed in the unit, a heavy -wash cycle will automatically be performed. can -determine the type of cycle required to clean the -dishes and will clean more efficiently every time. - -it - +. +8 +0 4 - - How your - - dishwasher cleans - -Getting Started - -Your dishwasher cleans by spraying a mixture of -hot water and detergent through the spray arms +0 +0 +0 +. . -against soiled surfaces - - pumped through the multiple - -1. Load dishwasher (See preparing and loading - -dishes). - -2. Add detergent (See Fill detergent to the dispenser). -3. Add rinse aid, if needed (See Fill rinse to - -dispenser). - -4. Close door to latch. -5. Select desired CYCLE (See Cycle Chart). The -indicator light above the pad will glow when -selected. - -6. Select desired OPTIONS (See Energy Options). -The indicator light above the pad will glow when -selected. - -7. Run hot water faucet nearest dishwasher until water - -is hot - -. Turn water off. - -8. To start, press - - pad. - -Your dishwasher has the front control panel. -Your selections will only beactivated when - the door is firmly closed. - -NOTE -The dishwasher is programmed to -remember your last cycle so you do not -have to reset each -dishwasher using the same cycle and -options selected on the previous wash just -press the Start Cancel - - start the - - pad. - -, - -/ - - water is necessary to activate - -Wash Cycle Selections -Cycle times are approximate and will vary with -options -dishwasher detergent and melt fatty food soils. -An automatic sensor will check the incoming water - timer will -temperature and if it is not hot -be delayed for automatic water heating in the main - happens even when -wash of all - WASH is selected an even higher -HI TEMP -temperature will be guaranteed to handle heavier soil -loads. - -- - -5 - - C - -ntrol panel -ontrol panel - -1 - -2 - -3 - 4 - 5 - 6 - +3 +0 +0 +. +. +6 7 - +2 +4 +0 +0 +. +. +6 8 - -9 10 - -11 - -12 - -13 - -Wash Cycle Selections - 9 - -Child lock - -1 - - & - -Pots - -PANS -This cycle is for hard to clean heavily -dishes, - -, - and dishes. - -- - - - soiled - -2 Normal wash - -This cycle is for regularly soiled dinner dishes or -silverware. -China crystal -This cycle is for lightly soiled china and crystal. - -3 - -4 Quick wash - -5 - -- - -- - -This cycle is for lightly soiled and pre rinsed -dishes and silverware. -Rinse only - dishes or glasses. -This cycle is for pre rinsing -This is only a rinse that keeps food from drying -on your dishes and reduces odor buildup in your -dishwasher until you are ready to wash a full -load. Do not use detergent. -ENERGY SAVER -For light solied or pre-rinsed dishes and - +2 +0 +0 +. +. 6 - +8 +SciQ 0.50 0.54 1.00 0.51 0.57 SciQ 0.46 0.53 1.00 0.57 0.56 +0.4 0.4 +HeadQA 0.49 0.57 0.51 1.00 0.62 +0.2 +HeadQA 0.62 0.74 0.57 1.00 0.74 +0.2 +ARC +C +-C om +ha +m +lle +o +n +n +g +se +e +nse O Q +0 +p A . - -Functions - -7 - -Air exchange -Select the “Air exchange function, the -dishwasher will exchange the air every 6 -hours to keep the air in the dishwasher fresh. - +e +6 +nB 8 - -Steam wash -For use with - - heavily soiled items, this STEAM - -option adds 30 minutes t - -o the cycle - - time. - -The option is available with the cycles of “POTS & - - PANS” and “Normal wash”. - +ookQ +0 +A +.74 Sc +0 +iQ +.5 H +7 +ea A d R Q +0 +C A - +. +C 6 - -You can lock all controls to prevent children from -accidently changing the dishwasher cycle or -starting the dishwasher. -Press the “Steam wash” and the “Vortex wash” -simultaneity to select or cancel the function. -After the function is selected and the corresponding -indicator light will turn on. - -10 - -Votex wash - -Use this option when you want to wash pots and -, -pans - -durable serving bowls - -and other large - -, - -, - -, - -hard to clean dishes - -very dirty - Touse use Vortex wash, place the dishes face down - in the lower rack directly above the vortex wash arm - -. - - in the back left corner. - -11 Heated Dry -When the” Heated Dry” function is selected, the -heater will work during the drying process. - -12 - -Sanitize - -Sanitize - -the “Sanitize” function - -To sanitize your dishes and -the -is selected, the temperature of water will be -℃ ℉ -maintained at - (158 -Note: It is possible that -70 -not be reached, if the temperature of the incoming -hot water dose not reach the suggested temperature. - -) max. -℃ - (158 - -) may - - ℉ - -70 - -13 - -Hi Temp - -Wash -When the “Hi Temp wash” function is selected, -the temperature of water will be maintained at -60 - -) max. - - (140 - - ℉ - -℃ - - Operating dishwasher - -17 - -18 - -19 - -14 - -15 - -16 - -/ -16 Start Cancel -This information covers several different models. -Your dishwasher may not have all the cycles and -options described. -This built-in dishwasher’s function control and -cycle selections will be void when dishwasher -door is firmly closed. When setting up the -dishwasher, please keep the dishwasher door -open and select the wash cycle and options -desired. The indicator light above the pad will -turn on. To start, press the Start/Cancel key -once and close the door firmly within 4 seconds -after you select start. The indicator light will -flickering and the wash cycle will begin after the -door is closed. To cancel wash cycle once the -dishwasher is running, carefully open the door -slowly to access the controls and avoid water -spill. Long press Start/Cancel for 8 second, the -dishwasher will start a 60 second draining -process to prevent water and moister trap inside -the dishwasher. The dishwasher will be power -off automatically when draining process is -completed. You may then open the unit and/or -start a new cycle. -When opening the door, use caution as there is -a possibility of injury from hot steam coming out -from the unit. If the door is opened while the -dishwasher is running, a safety switch will stop -the cycle. -You may add more dishes before the main wash -starts. - -Cycle progress indicator - -14 - -Delay - - Delay -To delay the start of a selected -Start pad until the desired delay time shows in the -it allows you to automatically -LED display screen, -start your dishwasher for 1 to 24 hours delay -. -To cancel the Delay Start option and begin the -cycle, before the delay period is over press - the -/ -Start Cancel -15 Display window -To display the remained hours and minutes of the -running cycle, delay hours and error codes etc. -17 Rinse aid indicator light -Turns on when the dishwasher needs to add Rinse aid. - - pad. - -18 Sanitized indicator light -If a cycle with the sanitize function is finished, -the sanitized indicator light turns on. -If you open the door, it will turn off after 30 seconds. -18 Clean indicator light -Turns on after a cycle finished. -If you open the door, it will turn off after 30 seconds. - - Located on ride side of the front of the dishwasher, these lights indicate how far along is the current wash cycle while -the dishwasher is running. - -1 - +h 2 - -3 - +alleng 1 - -WASH - -2 - -DRY - -3 - -CLEAN - -When the dishwasher is in a wash or rinse portion of a cycle the Wash lights are -illuminated. -When the dishwasher is in a drying portion of a cycle the Wash and Dry lights are -illuminated. -After the dishwasher has completed all portions of a cycle all indicator lights are -illuminated. -If you open the door when the Cleaning lights are illuminated, all the portion lights -will turn off after 30seconds. - -, - -, - -, - -7 - - Operating dishwasher - -Status Window - -01 24- -If the Delay Start option is sel - - hours shows in the Status - - of - the - - of hours and minutes left shows - -dishwasher is working and the wash cycle time is more -than a -alternately in the Status Window. -CL -Flashes once the program has been entered and the - -NOTE -The time remaining could suddenly increase -. -or drop several minutes This -Smart Sensor has checked the soil level and -determined if additional water fills are needed -or if water fills can be deleted. - - means the - -Error indication - - key is - - also flashes if the door is - the dishwasher - -opened during a wash -door to begin or continue a wash cycle. -MINUTES -Number of minutes left in a cycle -It shows hours and minutes alternately. - - or - -In certain situations which are critical to the -machines function and safe operation, the -control shall shut down and indicate an error -code Er (See Error Codes on page 18). -If Er codes are - - the dealer or - will be able to provide assistance - the problem and or locate an - -in -authorized service technician if needed. - -Cycle Chart - -Program - -Description of Cycle - -Water（L） - -Cycle time（min） - -POTS & PANS - -Pre Wash - -Pre Wash - -Wash(48°C) - -Rinse - -Rinse - -Rinse(58°C) - -Drying -Pre Wash - -22.8 - -140 - -NORMAL WASH - -Auto(42~48°C) - -Rinse(50~58°C) - -10.8~23.0 - -105~135 - -CHINA CRYSTAL - -QUICK WASH - -Drying -Pre Wash - -Pre Wash - -Wash(48°C) - -Rinse - -Rinse(58°C) - -Drying -Pre Wash - -Wash(40°C) - -Rinse(50°C) - -Drying - -RINSE ONLY - -Rinse Only - -ENERGY SAVER - -Pre Wash - -Pre Wash - -Wash(40°C) - -Rinse - -Rinse(50°C) - -Drying - -120 - -60 - -20 - -105 - -19.3 - -15.5 - -8 - -19.3 - -8 - - Adjustable Upper Rack - -The upper rack height can be easily adjusted to -accommodate loading taller dishes in either rack. -To accommodate taller dishes in the bottom rack -are 2 positions available: - -, - -there - -Upper rack - -Lower rack - -Dish Preparation - -. - -, - -, - -, -pits - -, - -. +e +.00 0.0 ARC +C +-C om +ha +m +lle +o +n +n +g +se +e +nse O Q +0 +p A . - -bones - - be loosened before loading Empty - -Scrape away large pieces of food -, - continuous filtered wash system -toothpicks etc The -- -will remove the remaining food particles Burned on - liquids -foods should -from glasses and cups. -Foods such as mustard mayonnaise -juice and tomato based products may cause -discoloration of stainless steel and plastics if allowed to -. -sit for a long period of time Unless -is to be opertated at once it is best to rinse off these -food soils. -If the dishwasher drains into a food disposer be, sure -disposer is completely empty before starting -dishwasher. - - the dishwasher - -vinegar - -, - -, - -lemon - -Note: -Height should be adjusted without dishes in the racks. - +e +6 +nB 9 - - Loading the Upper Rack - -Loading the Bottom Rack - -The upper rack is designed for cups, glasses, small -plates, bowls and plastic items marked -safe etc. -For best results, place the bowls, cups, glasses, sauce -pans with soiled surface facing down or toward the -center. Tilt slightly for better drainage. - -Make sure the loaded dishes do not -interfere with the rotation of the middle -spray arm, which is located at the bottom -of the upper rack. (Check this by rotating -the middle spray arm by hand.) - -The bottom rack is designed to load plates - and cookware. The adjustable rack -, -plates saucers -space will allow you to load maximum 13 inch items. - -, - -, -soup bowls - -Large items should be placed along the edge -so they do not interfere with the middle spray -arm rotation, and make sure they - not prevent -the detergent -Large items should be turned so that the -inside faces downward, and make sure these -items do not interfere with the lower spray -arm rotation. - - dispenser from opening. - -- - - tines in the bottom rack make it easy to -The fold down -- - -load those extra large and hard to fit -items The tines -may be left up for normal use or folded down for more - down for more space to load large -flexible loading Fold -pots or pans. - -. - +ookQ +0 +A +.82 Sc +0 +iQ +.5 H +6 +ea A d R Q +0 +C A - . - -Upper Rack - -8 place settings - -Upper Rack 12 place settings - -Upper Rack 14 place settings - -Bottom Rack 8 place settings - -Bottom Rack 12 place settings - -Bottom Rack 14 place settings - -10 - - Preparing and loading dishes - -Loading the Silverware Basket - -Adding A Dish - -The silverware basket could be divided into three -separate modules that can be used in both upper and -lower racks - -1. -2. -3. -4. -5. -6. - -Spoons -Knives -Salad forks -Forks -Large spoons -Large forks - -2 -2 -2 -2 - -2 - -2 - -4 -4 -4 -4 - -3 -3 -3 -3 - -4 +C +7 +h 4 -3 -3 - -1 - -1 - -1 - -1 - -1 - -1 -1 -1 - -1 -1 -1 - -6 - -1 - -1 - -1 - -1 - -1 - -1 -1 -1 - -1 -1 -1 - -6 - -1 - +alleng 1 - -5 - -4 -4 -3 -3 - -3 -3 -3 -3 - -4 -4 -4 -4 - -2 - -2 - -2 -2 -2 -2 - -To add or remove items after wash cycle starts: -1 Open door slightly and wait a few seconds until wash - -action stops before completely opening. - -2 Add the item. -3 Close door firmly to latch and cycle will automatically - -resume. - -CAUTION - -TO AVOID BURN INJURY: Slightly open door and -wait until spray arms and wash action stop. Hot -water may splash out of dishwasher. Failure to do -so may result in injury. - -Filling the rinse aid dispenser - -. - - next to the - - rinse aid dispenser located - -Rinse aid greatly improves drying and reduces water - rinse aid your dishes and -spots and filming Without -. -dishwasher interior will have excessive moisture -The -Heated Dry option will not perform as well without rinse -, -aid The. - releases a measured -, -Detergent cup automatically -amount of rinse aid during the last rinse. -, - the -If spotting and poor drying is a problem increase -amount of rinse aid dispensed by rotating the dial to a -higher number The. - dial is located under the dispenser -cap. -If the rinse aid is low, the Rinse Aid light is illuminated at -the beginning and end of the cycle indicating that it -is time to refill. - -Do not let any item extend through bottom. - -Be sure nothing protrudes though bottom of -the basket or rack to block the lower spray -arm. - -11 - - Preparing and loading dishes - -Filling the rinse aid dispenser - -Your dishwasher is designed to use a liquid rinse aid The - -. - improves drying performance after the final rinse Do - - rinse -aid. Under normal conditions, the rinse aid will last approximately one month, try to -keep it full, but do not over fill. - -. not use a solid or bar type - - use of rinse greatly -- - - the door turn, - - the dispenser cap to left and lift - -. -1 Open -it out. - - the amount of the rinse aid released - -3. You can adjust -during the -rinse operation, the greater number indicates the -larger amount of rinse aid released. - -. -2 Add - - rinse agent until the indicator point to max - -. -4 Replace - - the dispenser cap - -Filling the detergent dispenser - -1. Push the dispenser cover latch downwards and -open it - -2. Add the detergent to the main wash compartment - -open cup - -12 - - Preparing and loading dishes - -3. Close the dispenser cover - -The amount of the detergent needed depends on the cycle and soil level of dishes, refer to the table below and -adjust the recommended amount as necessary for your load. - -Cycle - -Soft Water -( -0 3 Grains) - -Medium Hard Water -( -4 8 Grains) - -Hard Water -( -9 12 Grains) - -Very Hard Water* -(Over 12 Grains) - -, - -Auto wash, -& -Pots wash -Normal -wash - -2 Teaspoons -(Main Wash Cup- - - Full) - -5 Teaspoons -(Main Wash Cup - -fill to line above -"R - -") -egular - -8 Teaspoons -(Main Wash -Cup -Completely -Full) - -Each Cup - -Completely Full -(Water Softener -Recommended) - -Quick wash, -China crystal - -2 Teaspoons -(Main Wash - Full) - -5 Teaspoons -(Main Wash Cup - fill to line above -") -egular - -"R - -8 Teaspoons -(Main Wash Cup -completely full) - -Main Wash Cup - -Completely Full -(Water Softener -Recommended) - -Rinse - -No Detergent - -No Detergent - -No Detergent - - Water - -No Detergent ( -Softener -Recommended) - -13 - - Cleaning the exterior door and panel - -1 Stainless - -. - - steel panel - -Clean the stainless steel door and handle regularly to -remove the debris with cleaning soft cloth. -Do not use appliance wax, polish, bleach or products -containing chlorine to clean the stainless steel door. - -2 Control - -. - - panel - -Clean the control -cloth, clean - -panel - gently. - -it - -with - -a lightly - -dampened - -Clean the Stainless Steel Inner Door and Tub - -The tub is made of Stainless Steel -corrode if the dishwasher should be scratched or -dented. - -it, will not rust or - -Clean the spots on the steel inner door and -tub with a damp nonabrasive cloth. - -Clean the Cylinder Filter - -WARNING - -The Cylinder Filter is designed to collect some big -items such as broken glass, bones and pits -Cylinder Filter needs to be cleaned to maximize wash -performance. -Remove the bottom rack, rotate the cylinder filter as -shown, lift it out, empty and clean it by holding it under -running water and replace. - -. -The - -Burn Hazard -Allow heating element to cool before cleaning -the interior. -Failure to do so can result in burns. - -14 - - Clean the Fine Filter - -Take out the Cylinder Filter and remove the Fine Filter from the dishwasher tub bottom To. -remove the Fine Filter you must first remove the bottom spray arm as shown below. -Flush the Fine Filter by holding it under running water and replace. - -15 - - Clean the door gasket - -Air gap - -Clean the door gasket with a damp cloth to remove -food particles regularly. - -If there is an air gap installed with your dishwasher, -make sure it is clean so the dishwasher will drain -properly. -Air gap is not part of your dishwasher. -Before you clean the air gap turn, -firstly and lift off the cover -Remove the plastic cap and clean with a toothpick. - - off the dishwasher - -Inside - -Protect against freezing - - a - -Your dishwasher should be protected from freezing -properly if it was left in an unheated place Have -qualified person do the following: -To Disconnect Service: -1. Turn off electrical power to the dishwasher at the -supply source by moving fuses or tripping circuit -breaker. -2. Turn off the water supply -3. Place a pan under the inlet valve Disconnect - -. - - water - -line from inlet valve and drain into pan. - - the water drain - -4. Disconnect drain line from pump and drain water -into pan. -To Restore Service: -. -1 Reconnect -supply. -. -2 Turn -3 Fill. - a Heated Wash -. -4 Check - - both detergent cups and run dishwasher through - - connections to make sure they do not leak. - - on water and electrical power supply. - - and electrical power - - cycle. - -, - -- - - with -The inside of the dishwasher is self cleaning -normal use. If needed, clean the tub gasket with a -damp cloth and use a brush similar to the ones shown -below to clean the opening in the end of the gasket. - -- - -16 - - Troubleshooting - -before calling for service - -Problem - -Possible cause - -Solution - -Dishwasher won t' start - -Door may not be closed properly -Power supply or power line is not connected -Delay Start option has been selected - -Child Lock activated selected - -( - - models) - -Dishwasher beeps at the -end of the cycle -Rinse Aid light on - -It indicates the wash cycle finished -dishwasher will beep -The rinse aid level is low - -, - -the - -The dishwasher is connected to cold water - -Dishwasher operates too -long - -The cycle time will vary due to the soil - -The sanitize option is selected - -Water pressure is temporarily low - -The inlet water is low - -Dishes are not clean -enough - -Dishes are loaded too close together - -improper use of detergent - -The selected cycle is not suitable for food soil condition -Spay arms is blocked by some items - -Detergent dispenser is empty - -Improper loading of dishes - -Dishes not drying enough - -Shut the door completely make sure it is latched -Check and connect the power supply -Refer - - in this manual - - to rese - -Deactivate the child lock, refer to child lock -section in this manual - -Add rinse aid -, -Check the dishwasher make -to the hot water supply - - sure it connected - - cycle will automatically increase the - -When heavy soil is detected, the auto, normal, -pots pan& -cycle time -When sanitize option is selected, the cycle time -will be increase to meet the sanitized -temperature request - -Use you dishwasher when the water pressure is -normal -Make sure the dishwasher is connected to the -hot water supply -Try not use your dishwasher when hot water is -being used elsewhere in the house -Load the dishes again as the manual introduced - -Add detergent depending on the water hardness -and the cycle you chose. Use fresh detergent. -Choose anther cycle for longer washing time -Make sure the spay arm is rotated fluently - -Fill rinse aid dispenser or increase the amount - -of rinse - - aid. - -Load the dishes again as the manual introduction - -Spots and filming on -dishes - -Etching - -Detergent left in -dispenser cup - -The cycle you chose did not included the drying. - -Choose the proper cycle with drying - -Water hardness is too high - -Improper loading of dishes -Old or damp powder rinse aid is used -Rinse dispenser empty -Using too much detergent -The inlet water temperature exceeds 150 ℉ - -Detergent may be too old -Spray arm is blocked - -For extremely hard water install -a water softener -Load the dishes again as the manual indicates -Use fresh rinse aid -Add rinse aid to the dispenser -Use less detergent if you have soft water -Lower the inlet water temperature -Use fresh detergent - -, -Load the dishes making -are not blocked. - - sure the spray arms - +e +.00 0.0 +Figure4: Overlapofidentifiedgeneralization-controllingfeaturesacrosstasks. Left: Qwen3-4B-RLvs.Qwen3-4B- +Base. Right: Qwen2.5-7B-RLvs.Qwen2.5-7B. +• RLintroducesfeaturedeviationsinagradual theinternalfeaturelandscapeinfundamentallydif- +andrestrainedmanner,yieldingrelativelyfew ferent ways. SFT introduces a large number of +andlessexclusivefeatures. Ratherthansub- strongly model-specific features, while RL pre- +stantiallyreplacingthebasemodel’sinternal dominantlypreservesandreweightssharedfeatures +representationalstructure,RLinduceslimited within the existing representational space. More +featureinnovationwhileexhibitingsustained importantly, it enables downstream mechanistic +featureturnoverthroughouttraining. analysis: itservesasthefoundationforidentifying +featuresassociatedwithcross-taskgeneralization +These findings indicate that SFT and RL dif- inthenextsection. +fernotonlyinthefinalfeaturestheyproduce,but +alsointhetemporaldynamicsoffeatureformation. 7 MechanisticExplanationforthe +Importantly,theprolongedfeaturereconfiguration GeneralizationAbilityofRL +observedinRLsuggestsatrainingprocessthatre- +RL-tuned models are optimized solely on math- +mainssensitivetoperformancefeedbackthrough- +ematical reasoning tasks, yet they achieve sub- +out optimization, a property that may be closely +stantialgainsnotonlyonin-domainmathematics +relatedtothestrongercross-taskgeneralizationbe- +benchmarks but also on disparate tasks such as +havioranalyzedinthenextsection. +commonsense and scientific knowledge question +6 Experiments: JointComparisonwith answering. Thiscross-taskgeneralizationcannot +Three-ModelSparseCrosscoders beexplainedbydataoverlaportasksimilarity,and +thereforecallsforamechanisticexplanation. +Usingthetrainedthree-modelSparseCrosscoder, +Ourprioranalysesprovideaninitialexplanation: +wecomputedtheMASsforeachfeature,measur- +RLinducesrestrained,incrementalfeaturechanges, +inghowstronglythefeatureisattributedtothebase, +whereasSFTintroducesaggressivefeaturerecon- +SFT-tuned,orRL-tunedmodelwithinasharedfea- +figuration that may lead to over-specialization. +ture space. Figure 3 illustrates the MAS distri- +However, this explanation remains indirect. To +butions for Qwen3-4B-Base across the different +establishamoreconcretemechanisticaccount,we +training conditions. The MAS distributions for +seek direct evidence linking specific internal fea- +theSFT-tunedmodelsexhibitapronouncedright +turestotheobservedgeneralizationbehavior. +tail. A substantial number of features attain high +attributionscores. ThispatternsuggeststhatSFT Experimental Setup. We performed feature +introduces a large set of strongly model-specific identification using the method proposed in Sec- +features. ComparedtoSFT,RLproducesfarfewer tion 3.2 on all five evaluation tasks introduced in +features that are strongly attributed to itself. The Section4. Foreachtask,wesetthethresholdtas +sameobservationscanbefoundonQwen2.5-7B, 20%ofthemaximumscoreobservedforthattask. +asshowninFigure5inAppendixB.3. Whenexaminingthegeneralizationfeaturesiden- +Thethree-modelSparseCrosscoderprovidesdi- tified for each task, we observe a striking degree +rectempiricalevidencethatSFTandRLreshape of consistency across tasks. To further substan- + +| | Model | | | OpenBookQA | | CommonsenseQA | | HeadQA | | SciQ | ARC-Challenge | | | +| --- | ------------- | --- | --- | ---------- | --- | ------------- | --- | ------ | --- | ----- | ------------- | ----- | --- | +| | Qwen3-4B-RL | | | -46.2 | | -43.9 | | -21.2 | | -14.0 | | -33.3 | | +| | Qwen2.5-7B-RL | | | -21.9 | | -24.4 | | -23.8 | | -44.4 | | -20.0 | | +Table2: Performancedegradationinducedbytheremovalofgeneralization-relatedfeaturesinRL–tunedmodels. +| | Model | | | OpenBookQA | | CommonsenseQA | | HeadQA | | SciQ | ARC-Challenge | | | +| --- | ------------- | --- | --- | ---------- | --- | ------------- | --- | ------ | --- | ----- | ------------- | ----- | --- | +| | Qwen3-4B-Base | | | +36.3 | | +36.0 | | +21.2 | | +38.0 | | +33.3 | | +| | Qwen2.5-7B | | | +12.5 | | +24.4 | | +14.3 | | +55.6 | | +40.0 | | +Table3: Performanceimprovementachievedbyamplifyinggeneralization-relatedfeaturesinthebasemodel. +tiate this observation, we quantified the overlap Qwen2.5-7Btocorrectlyanswer56%ofsamples +amongtheidentifiedfeaturesacrossdifferenttasks, thatitpreviouslyansweredincorrectly. +asillustratedinFigure4. Theresultsrevealasub- Thisfindingsuggeststhatthebasemodeldoes +| stantial | level | of feature | | overlap, | even over | 80%. | | | | | | | | +| -------- | ----- | ---------- | --- | -------- | --------- | ---- | ----------------------------- | --- | --- | --- | --- | ---------------- | --- | +| | | | | | | | notlackthenecessaryknowledge. | | | | | Instead,therele- | | +Despite the diversity of these benchmarks, many vantupstreamcircuitryispresentbutnotnaturally +featuresconsistentlyappearacrosstasks,strongly activated. Onceacontrolsignalisforciblyinjected, +suggestingtheexistenceoftask-agnosticfeatures generalizedbehaviorsemergeprominently. +thatsupportgeneralization. +| | | | | | | | Generalization | | to | Unseen | Tasks. | | We further | +| --- | -------- | ----------------- | --- | ---------- | -------------- | -------- | -------------- | --- | ------- | -------------- | ------ | -------------- | ---------- | +| The | final | intersection | | across all | tasks contains | | | | | | | | | +| | | | | | | | evaluated | | whether | the identified | | generalization | | +| 50 | features | for Qwen3-4B-Base | | | and 16 | features | | | | | | | | +for Qwen2.5-7B, which we refer to as the final features transfer beyond the tasks used for fea- +| | | | | | | | ture | identification. | | Specifically, | | we tested | them | +| --- | --- | --- | --- | --- | --- | --- | ---- | --------------- | --- | ------------- | --- | --------- | ---- | +generalization-controllingfeatures. +| | | | | | | | on two | unseen | benchmarks, | | LogiQA | | (Liu et al., | +| --- | --- | --- | --- | --- | --- | --- | ------ | ------ | ----------- | --- | ------ | --- | ------------ | +7.1 CausalValidationviaFeature 2021) and PIQA (Bisk et al., 2020). We observe +| | Interventions | | | | | | consistentperformancedegradationwhenzeroing | | | | | | | +| --- | ------------- | -------------- | --- | ---------- | -------- | ---- | ------------------------------------------- | -------- | ------------- | --- | ------- | --- | ---------- | +| | | | | | | | these | features | in RL-trained | | models, | | and corre- | +| To | verify | the functional | | importance | of these | fea- | | | | | | | | +spondingperformanceimprovementswhenampli- +tures,weconductedtwocomplementaryinterven- +| | | | | | | | fying | them | in base | models. | This | provides | addi- | +| --- | --- | --- | --- | --- | --- | --- | ----- | ---- | ------- | ------- | ---- | -------- | ----- | +tionexperiments. +tionalevidencethattheidentifiedfeaturescapturea +ZeroingGeneralizationFeaturesintheRLMod- general-purposegeneralizationmechanism,rather +els. Wefirstconductedablationexperimentsby thantask-specificheuristics. Detailedexperimental +zeroingouttheidentifiedfeaturesintheRLmodel +setupandresultsareprovidedinAppendixB.4. +andevaluatingperformanceonthecorresponding +| generalization-critical | | | samples. | | As shown | in Ta- | 8 | Conclusion | | | | | | +| ----------------------- | ------- | ------------ | -------- | -------------- | ----------- | ------ | ---- | ---------- | ---------------- | --- | --- | -------- | ---- | +| ble | 2, this | intervention | | leads to | substantial | per- | | | | | | | | +| | | | | | | | This | work | has investigated | | why | RL-tuned | LLMs | +| formance | | degradation. | | In particular, | for | Open- | | | | | | | | +generalizebeyondtheirtrainingdistribution,while +BookQAandCommonsenseQA,zeroingthesefea- +SFToftenleadstothelossofgeneralcapabilities. +| tures | causes | the Qwen3-4B-RL | | | model to | answer | | | | | | | | +| ----- | ------ | --------------- | --- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | +Usingacontrolledexperimentalsetupandafeature- +over40%ofpreviouslycorrectsamplesincorrectly. +levelinterpretabilityframework,wehavecompared +| These | results | indicate | that | the identified | | features | | | | | | | | +| ----- | ------- | -------- | ---- | -------------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +howRLandSFTreshapeinternalrepresentations +arenecessaryforsuccessfulgeneralization. +| | | | | | | | during | post-training. | | We | have | shown | that SFT | +| --- | --- | --- | --- | --- | --- | --- | ------ | -------------- | --- | --- | ---- | ----- | -------- | +AmplifyingGeneralizationFeaturesintheBase rapidlyintroduceshighlyspecializedfeaturesthat +Models. Conversely,weamplifiedthesameset stabilize early in training, whereas RL induces +offeaturesinthebasemodelbysettingtheiracti- more restrained and continually evolving feature +vations to a fixed large value (we set to 3.0), fol- changesthatlargelypreservethebasemodel’srep- +lowing standard feature intervention practices in resentations. Buildingonthisdistinction,wehave +SAE-basedanalyses(Zhangetal.,2025;Hanetal., identified a compact set of internal features that +2025). TheresultsareshowninTable3. Acrossall causallycontrolcross-taskgeneralization. Feature- +tasks,performanceimprovessubstantially. Inpar- levelinterventionshaveconfirmedtheirrole: dis- +ticular,forSciQ,amplifyingthesefeaturesenables abling these features degrades RL performance, + +whileamplifyingthemtransfersgeneralizationbe- +haviortobasemodels,includingonunseentasks. +| Overall, | our results have | provided | a mechanistic | +| ----------- | ---------------- | -------------- | ------------- | +| explanation | for why | RL generalizes | while SFT | +| memorizes, | and have | demonstrated | the value of | +feature-levelinterpretabilityforunderstandingpost- +trainingdynamicsinLLMs. + +Limitations +| | | | | | | Sergey | Levine, | and | Yi Ma. | 2025. | SFT | mem- | +| --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------ | ----------- | --- | -------- | +| | | | | | | orizes, | RL generalizes: | | A | comparative | | study of | +First, our feature-level analysis relies on Sparse foundation model post-training. arXiv preprint +arXiv:2501.17161. +| Crosscoders | as the | underlying | alignment | | mecha- | | | | | | | | +| ----------- | ------ | ---------- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +nism. Althoughthisapproachenablesinterpretable +PeterClark,IsaacCowhey,OrenEtzioni,TusharKhot, +andcomparablefeaturerepresentations,thelearned +AshishSabharwal,CarissaSchoenick,andOyvind +featurespaceisnotguaranteedtocaptureallfunc- Tafjord.2018. Thinkyouhavesolvedquestionan- +| | | | | | | swering? | try | ARC, | the AI2 | reasoning | | challenge. | +| ----------------- | -------- | ----------- | --- | ----- | ------ | -------- | --- | ---- | ------- | --------- | --- | ---------- | +| tionally relevant | internal | structures. | | Other | inter- | | | | | | | | +CoRR,abs/1803.05457. +pretabilitymethodsoralignmentschemesmayre- +vealcomplementarymechanismsthatarenotcap- HoagyCunningham,AidanEwart,LoganRiggs,Robert +tured by sparse feature decomposition. Second, Huben,andLeeSharkey.2023. Sparseautoencoders +whileourinference-timefeatureinterventionses- findhighlyinterpretablefeaturesinlanguagemodels. +arXivpreprintarXiv:2309.08600. +tablishcausallinkstomodelbehavior,theydonot +directly inform how to design training objectives DamaiDai,LiDong,YaruHao,ZhifangSui,Baobao +| | | | | | | Chang, | and Furu | Wei. | 2022. | Knowledge | | neurons | +| --- | --- | --- | --- | --- | --- | ------ | -------- | ---- | ----- | --------- | --- | ------- | +thatexplicitlyencouragegeneralization-controlling +features. Weleavethedevelopmentofsuchtraining in pretrained transformers. In Proceedings of the +60thAnnualMeetingoftheAssociationforCompu- +strategiestofuturework. +| | | | | | | tationalLinguistics(Volume1: | | | | LongPapers),ACL | | | +| --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --------------- | --- | --- | +2022,Dublin,Ireland,May22-27,2022,pages8493– +Acknowledgements 8502.AssociationforComputationalLinguistics. +The present research was supported by the Na- BoyiDeng,YuWan,BaosongYang,YidanZhang,and +tional Key Research and Development Program Fuli Feng. 2025. Unveiling language-specific fea- +| | | | | | | tures in | large | language | models | via | sparse | autoen- | +| -------- | ---------- | --------------- | --- | --- | --- | -------- | ----- | -------- | ------ | --- | ------ | ------- | +| of China | (Grant No. | 2024YFE0203000) | | | and | | | | | | | | +coders. InProceedingsofthe63rdAnnualMeeting +| the International | Cooperation | | Program | | for Inno- | | | | | | | | +| ----------------- | ----------- | --- | ------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +oftheAssociationforComputationalLinguistics(Vol- +vative Talents Development of CSC (Grant No. ume1: LongPapers),pages4563–4608. +| CXXM2310203712). | | We | would | like | to thank | | | | | | | | +| ---------------- | --- | --- | ----- | ---- | -------- | --- | --- | --- | --- | --- | --- | --- | +AndreyGalichin,AlexeyDontsov,PolinaDruzhinina, +theanonymousreviewersfortheirinsightfulcom- +AntonRazzhigaev,OlegYRogov,ElenaTutubalina, +ments. +| | | | | | | and Ivan | Oseledets. | 2025. | | I have | covered | all the | +| ---------- | --- | --- | --- | --- | --- | ------------------------- | ------------------------------------ | ----- | ------ | ------------- | ------- | ------- | +| | | | | | | baseshere: | Interpretingreasoningfeaturesinlarge | | | | | | +| | | | | | | language | models | via | sparse | autoencoders. | | arXiv | +| References | | | | | | preprintarXiv:2503.18878. | | | | | | | +DavidDBaekandMaxTegmark.2025. Towardsun- LeoGao, TomDuprélaTour, HenkTillman, Gabriel +derstandingdistilledreasoningmodels: Arepresenta- Goh, Rajan Troll, Alec Radford, Ilya Sutskever, +tionalapproach. arXivpreprintarXiv:2503.03730. Jan Leike, and Jeffrey Wu. 2024a. Scaling and +| | | | | | | evaluating | sparse | autoencoders. | | | arXiv | preprint | +| ----------------------------------------------- | --- | --------------------------- | --- | --- | --- | ----------------- | ------ | ------------- | --- | --- | ----- | -------- | +| YonatanBisk,RowanZellers,JianfengGao,YejinChoi, | | | | | | arXiv:2406.04093. | | | | | | | +| and1others.2020. | | PIQA:Reasoningaboutphysical | | | | | | | | | | | +commonsenseinnaturallanguage. InProceedings Leo Gao, Jonathan Tow, Baber Abbasi, Stella Bider- +oftheAAAIconferenceonartificialintelligence,vol- man, Sid Black, Anthony DiPofi, Charles Foster, +ume34,pages7432–7439. Laurence Golding, Jeffrey Hsu, Alain Le Noac’h, +| | | | | | | Haonan | Li, Kyle | McDonell, | | Niklas | Muennighoff, | | +| --- | --- | --- | --- | --- | --- | ------ | -------- | --------- | --- | ------ | ------------ | --- | +Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, ChrisOciepa,JasonPhang,LariaReynolds,Hailey +| Linyi Yang, | Kaijie | Zhu, Hao | Chen, | Xiaoyuan | Yi, | | | | | | | | +| ----------- | ------ | -------- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +Schoelkopf,AviyaSkowron,LintangSutawika,and +CunxiangWang,YidongWang,and1others.2024. 5 others. 2024b. The language model evaluation +| A survey | on evaluation | of | large | language | models. | harness. | | | | | | | +| -------- | ------------- | --- | ----- | -------- | ------- | -------- | --- | --- | --- | --- | --- | --- | +ACMtransactionsonintelligentsystemsandtechnol- +ogy,15(3):1–45. AaronGrattafiori,AbhimanyuDubey,AbhinavJauhri, +| | | | | | | Abhinav | Pandey, | Abhishek | | Kadian, | Ahmad | Al- | +| --- | --- | --- | --- | --- | --- | ------- | ------- | -------- | --- | ------- | ----- | --- | +RuizheChen,TianxiangHu,YangFeng,andZuozhu +Dahle,AieshaLetman,AkhilMathur,AlanSchelten, +Liu. 2024. Learnable privacy neurons localization AlexVaughan,and1others.2024. Thellama3herd +inlanguagemodels. InProceedingsofthe62ndAn- ofmodels. arXivpreprintarXiv:2407.21783. +nualMeetingoftheAssociationforComputational +Linguistics,ACL2024-ShortPapers,Bangkok,Thai- Daya Guo, Dejian Yang, Haowei Zhang, Junxiao +land,August11-16,2024,pages256–264.Associa- Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shi- +rongMa,PeiyiWang,XiaoBi,and1others.2025. +tionforComputationalLinguistics. +| | | | | | | Deepseek-r1: | | Incentivizingreasoningcapabilityin | | | | | +| --- | --- | --- | --- | --- | --- | ------------ | --- | ---------------------------------- | --- | --- | --- | --- | +TianzheChu,YuexiangZhai,JihanYang,Shengbang LLMs via reinforcement learning. arXiv preprint +| Tong, Saining | Xie, | Dale Schuurmans, | | Quoc | V Le, | arXiv:2501.12948. | | | | | | | +| ------------- | ---- | ---------------- | --- | ---- | ----- | ----------------- | --- | --- | --- | --- | --- | --- | + +ZishanGuo,RenrenJin,ChuangLiu,YufeiHuang,Dan NeginRaoof,EtashKumarGuha,RyanMarten,Jean +Shi,LinhaoYu,YanLiu,JiaxuanLi,BojianXiong, Mercat, Eric Frankel, Sedrick Keh, Hritik Bansal, +and Deyi Xiong. 2023. Evaluating large language GeorgiosSmyrnis,MariannaNezhurina,TrungVu, +models: A comprehensive survey. arXiv preprint ZayneReaSprague,MikeAMerrill,LiangyuChen, +arXiv:2310.19736. CarolineChoi,ZaidKhan,SachinGrover,Benjamin +| | | | | | | | Feuer, | Ashima | Suvarna, | Shiye | Su, and | 27 others. | +| --- | --- | --- | --- | --- | --- | --- | ------ | ------ | -------- | ----- | ------- | ---------- | +ZhuowenHan,XinweiWu,DanShi,RenrenJin,and 2025. AutomaticevalsforLLMs. +| DeyiXiong.2025. | | Towardsaunifiedparadigmof | | | | | | | | | | | +| --------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +concepteditinginlargelanguagemodels. InProceed- Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, +ingsofthe2025ConferenceonEmpiricalMethodsin +JunxiaoSong,XiaoBi,HaoweiZhang,Mingchuan +NaturalLanguageProcessing,pages18445–18461. +| | | | | | | | Zhang, | YK | Li, Yang | Wu, and | 1 others. | 2024. | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | ------- | --------- | ----- | +AssociationforComputationalLinguistics. DeepSeekMath: Pushingthelimitsofmathematical +| | | | | | | | reasoninginopenlanguagemodels. | | | | arXivpreprint | | +| ---------------------------------------------- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | ------------- | --- | +| DanHendrycks,CollinBurns,StevenBasart,AndyZou, | | | | | | | arXiv:2402.03300. | | | | | | +MantasMazeika,DawnSong,andJacobSteinhardt. +2021. Measuringmassivemultitasklanguageunder- +| | | | | | | | Guangming | Sheng, | Chi Zhang, | Zilingfeng | | Ye, Xibin | +| --------- | ----------------------------------- | --- | --- | --- | --- | --- | --------- | ------ | ---------- | ---------- | --- | --------- | +| standing. | InInternationalConferenceonLearning | | | | | | | | | | | | +Wu,WangZhang,RuZhang,YanghuaPeng,Haibin +| Representations. | | | | | | | Lin,andChuanWu.2025. | | | HybridFlow: | | Aflexible | +| ---------------- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | ----------- | --------------- | --------- | +| | | | | | | | andefficientRLHFframework. | | | | InProceedingsof | | +Maggie Huan, Yuetai Li, Tuney Zheng, Xiaoyu Xu, the Twentieth European Conference on Computer +SeungoneKim,MinxinDu,RadhaPoovendran,Gra- +Systems,pages1279–1297. +| hamNeubig,andXiangYue.2025. | | | | Doesmathrea- | | | | | | | | | +| --------------------------- | --- | ------- | ----------------- | ------------ | ------ | --- | -------- | ------ | ------------ | ----- | ------- | ----- | +| soning improve | | general | LLM capabilities? | | under- | | | | | | | | +| | | | | | | | Dan Shi, | Renren | Jin, Tianhao | Shen, | Weilong | Dong, | +arXiv +standing transferability of LLM reasoning. XinweiWu,andDeyiXiong.2024a. IRCAN:Mit- +preprintarXiv:2507.00432. igatingknowledgeconflictsinLLMgenerationvia +identifyingandreweightingcontext-awareneurons. +AaronJaech,AdamKalai,AdamLerer,AdamRichard- +AdvancesinNeuralInformationProcessingSystems, +| son, Ahmed | El-Kishky, | | Aiden | Low, | Alec Helyar, | | | | | | | | +| ---------- | ---------- | --- | ----- | ---- | ------------ | --- | --- | --- | --- | --- | --- | --- | +37:4997–5024. +AleksanderMadry,AlexBeutel,AlexCarney,and1 +| others.2024. | Openaio1systemcard. | | | arXivpreprint | | | | | | | | | +| ------------ | ------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +DanShi,TianhaoShen,YufeiHuang,ZhigenLi,Yongqi +arXiv:2412.16720. +Leng,RenrenJin,ChuangLiu,XinweiWu,Zishan +| | | | | | | | Guo, | Linhao | Yu, and | 1 others. | 2024b. | Large lan- | +| ------------- | -------- | ---------- | ------- | -------- | ----------- | --- | --------------------------------- | ------ | ------- | --------- | ------------- | ---------- | +| Jack Lindsey, | Adly | Templeton, | | Jonathan | Marcus, | | | | | | | | +| | | | | | | | guagemodelsafety:Aholisticsurvey. | | | | arXivpreprint | | +| Thomas | Conerly, | Joshua | Batson, | and | Christopher | | | | | | | | +arXiv:2412.17686. +| Olah. 2024. | Sparse | crosscoders | | for | cross-layer | | | | | | | | +| ----------- | --------- | ----------- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | +| features | and model | diffing. | Transformer | | Circuits | | | | | | | | +DanShi,ChaobinYou,JiantaoHuang,TaihaoLi,and +Thread. +| | | | | | | | DeyiXiong.2024c. | | CORECODE:Acommonsense | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --------------------- | --- | --- | --- | +annotateddialoguedatasetwithbenchmarktasksfor +| Jian Liu, Leyang | Cui, | Hanmeng | Liu, | Dandan | Huang, | | | | | | | | +| ---------------- | ---- | ---------- | ---- | ------ | ------- | --- | ------- | ----- | -------- | ------- | -------------- | --- | +| | | | | | | | chinese | large | language | models. | In Proceedings | of | +| Yile Wang, | and | Yue Zhang. | | 2021. | LogiQA: | a | | | | | | | +theAAAIConferenceonArtificialIntelligence,vol- +| challenge | dataset | for machine | | reading | comprehen- | | | | | | | | +| --------- | ------- | ----------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | --- | +sion with logical reasoning. In Proceedings of the ume38,pages18952–18960. +Twenty-NinthInternationalConferenceonInterna- +| | | | | | | | Alon Talmor, | | Jonathan Herzig, | Nicholas | Lourie, | and | +| ------------ | ----------- | --- | ------------- | --- | ------------- | --- | -------------------- | --- | ---------------- | -------------------- | ------- | --- | +| tional Joint | Conferences | | on Artificial | | Intelligence, | | | | | | | | +| | | | | | | | JonathanBerant.2019. | | | CommonsenseQA:Aques- | | | +pages3622–3628. +| | | | | | | | tion answering | | challenge | targeting | commonsense | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | --- | --------- | --------- | ----------- | --- | +InProceedingsofthe2019Conference +| MichaelLuo,SijunTan,JustinWong,XiaoxiangShi, | | | | | | | knowledge. | | | | | | +| -------------------------------------------- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | +WilliamY.Tang,MananRoongta,ColinCai,Jeffrey oftheNorthAmericanChapteroftheAssociationfor +Luo,LiErranLi,RalucaAdaPopa,andIonStoica. ComputationalLinguistics: HumanLanguageTech- +nologies,Volume1(LongandShortPapers),pages +| 2025. DeepScaleR: | | Surpassing | | o1-preview | with | a | | | | | | | +| ----------------- | --- | ---------- | --- | ---------- | ---- | --- | --- | --- | --- | --- | --- | --- | +4149–4158,Minneapolis,Minnesota.Associationfor +| 1.5bmodelbyscalingRL. | | | NotionBlog. | | | | | | | | | | +| --------------------- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +ComputationalLinguistics. +SamuelMarks,CanRager,EricJMichaud,YonatanBe- +linkov,DavidBau,andAaronMueller.2025. Sparse Kimi Team, Angang Du, Bofei Gao, Bowei Xing, +featurecircuits:Discoveringandeditinginterpretable Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun +| | | | | InTheThirteenth | | | Xiao,ChenzhuangDu,ChonghuaLiao,and1others. | | | | | | +| --- | --- | --- | --- | --------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | +causalgraphsinlanguagemodels. +| | | | | | | | 2025. | Kimi | k1. 5: Scaling | reinforcement | | learning | +| ------------- | ---------- | --- | ----------- | --- | ----------- | --- | --------- | ---- | ------------------------------ | ------------- | --- | -------- | +| International | Conference | | on Learning | | Representa- | | | | | | | | +| tions. | | | | | | | withLLMs. | | arXivpreprintarXiv:2501.12599. | | | | +TodorMihaylov,PeterClark,TusharKhot,andAshish David Vilares and Carlos Gómez-Rodríguez. 2019. +Sabharwal.2018. Canasuitofarmorconductelec- HEAD-QA:Ahealthcaredatasetforcomplexreason- +tricity? a new dataset for open book question an- ing. InProceedingsofthe57thAnnualMeetingof +swering. InProceedingsofthe2018Conferenceon theAssociationforComputationalLinguistics,pages +EmpiricalMethodsinNaturalLanguageProcessing, 960–966,Florence,Italy.AssociationforComputa- +| pages2381–2391. | | | | | | | tionalLinguistics. | | | | | | +| --------------- | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | + +JohannesWelbl,NelsonFLiu,andMattGardner.2017. +| Crowdsourcing | | multiple | choice | science | questions. | +| ------------- | --- | -------- | ------ | ------- | ---------- | +InProceedingsofthe3rdWorkshoponNoisyUser- +generatedText,pages94–106. +JiaqiWeng,HanZheng,HanyuZhang,QinqinHe,Jial- +| ingTao, | HuiXue, | ZhixuanChu, | | andXitingWang. | | +| --------------------------------- | -------- | ----------- | ------ | -------------- | ---------- | +| 2025. Safe-SAIL: | | Towards | a | fine-grained | safety | +| landscape | of large | language | models | via | sparse au- | +| toencoderinterpretationframework. | | | | arXivpreprint | | +arXiv:2509.18127. +XinweiWu,JunzhuoLi,MinghuiXu,WeilongDong, +| ShuangzhiWu, | | ChaoBian, | andDeyiXiong.2023. | | | +| ------------ | --- | --------- | ------------------ | --- | --- | +DEPN:Detectingandeditingprivacyneuronsinpre- +| trainedlanguagemodels. | | | InProceedingsofthe2023 | | | +| ---------------------- | ------------ | --- | ---------------------- | ---------- | ---- | +| Conference | on Empirical | | Methods | in Natural | Lan- | +guageProcessing,EMNLP2023,Singapore,Decem- +| ber 6-10, | 2023, | pages 2875–2886. | | Association | for | +| --------- | ----- | ---------------- | --- | ----------- | --- | +ComputationalLinguistics. +AnYang,AnfengLi,BaosongYang,BeichenZhang, +| Binyuan | Hui, | Bo Zheng, | Bowen | Yu, | Chang | +| ------------ | ----- | --------- | ------- | ------- | --------- | +| Gao, Chengen | | Huang, | Chenxu | Lv, and | 1 others. | +| 2025. | Qwen3 | technical | report. | arXiv | preprint | +arXiv:2505.09388. +AnYang,BaosongYang,BeichenZhang,BinyuanHui, +BoZheng,BowenYu,ChengyuanLi,DayihengLiu, +FeiHuang,GuantingDong,HaoranWei,HuanLin, +JianYang,JianhongTu,JianweiZhang,JianxinYang, +JiaxinYang,JingrenZhou,JunyangLin,and25oth- +| ers.2024. | Qwen2.5technicalreport. | | | arXivpreprint | | +| --------- | ----------------------- | --- | --- | ------------- | --- | +arXiv:2412.15115. +| Jehyeok Yeon, | Federico | Cinus, | Yifan | Wu, | and Luca | +| ------------- | ------------------------------- | ------ | ------ | --------- | -------- | +| Luceri.2025. | GSAE:Graph-regularizedsparseau- | | | | | +| toencoders | for robust | LLM | safety | steering. | arXiv | +preprintarXiv:2512.06655. +KaichenZhang,YifeiShen,BoLi,andZiweiLiu.2025. +Largemulti-modalmodelscaninterpretfeaturesin +| large multi-modal | | models. | In | Proceedings | of the | +| ----------------- | ------------- | ------- | ---------- | ----------- | -------- | +| IEEE/CVF | International | | Conference | on | Computer | +Vision,pages3650–3661. +| Yaowei Zheng, | Richong | Zhang, | | Junhao Zhang, | Yan- | +| ------------- | ------- | ------ | --- | ------------- | ---- | +hanYe,ZheyanLuo,ZhangchiFeng,andYongqiang +| Ma. 2024. | Llamafactory: | | Unified | efficient | fine- | +| --------- | ------------- | -------- | ------- | --------- | -------- | +| tuning of | 100+ | language | models. | arXiv | preprint | +arXiv:2403.13372. + +A ExperimentalSettings +| | | | | | | | et al., 2025), | | and we | retain | only | those responses | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | --- | ------ | ------ | ---- | --------------- | --- | +thatleadtocorrectfinalanswers. +Inthissection,weprovidemoredetailsaboutthe +| experimentalsettings. | | | | | | | A.3 TrainingDetails | | | | | | | +| --------------------- | --- | --- | --- | --- | --- | --- | ------------------- | --- | -------- | --- | ---- | --- | -------- | +| | | | | | | | Reinforcement | | Learning | | (RL) | has | recently | +A.1 TrainingDetailsofCrosscoders +| | | | | | | | demonstrated | | notable | success | in | enhancing | the | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | ------- | ------- | --- | --------- | --- | +Following Baek and Tegmark (2025), we train complex, multi-step reasoning capabilities of +each crosscoder with d sparse = 32,768 LLMs by optimizing policies with scalar reward +| features | on | 200 | million | tokens | from | open- | | | | | | | | +| -------- | --- | --- | ------- | ------ | ---- | ----- | --- | --- | --- | --- | --- | --- | --- | +signals. Inourstudy,weadopttheverlframework +thoughts/OpenThoughts-114k1 and another 200 (Shengetal.,2025)andimplementGRPO(Shao +milliontokensfromtogethercomputer/RedPajama- etal.,2024)ontheQwen-3-4B-BaseandQwen2.5- +| Data-1T-Sample2 | | dataset. | | The | former | includes | | | | | | | | +| --------------- | --- | -------- | --- | --- | ------ | -------- | --------- | -------------------------------- | --- | --- | --- | --- | --- | +| | | | | | | | 7Bmodels. | Trainingisperformedwithanoverall | | | | | | +math, science, and code reasoning traces gener- batchsizeof128andalearningrateof1×10−6. +atedbyDeepSeek-R1,whereasthelattercontains Wesetthegenerationsequencelengthupto16kto- +general-domaintexts. Toavoiddistributionalbias, kens,andperform8rolloutsperprompt,updating +thetwodatasetsaremergedandjointlyshuffledbe- +| | | | | | | | themodelinmini-batchesof64samples. | | | | | Clipping | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | -------- | --- | +foretraining,ratherthanbeingtrainedsequentially. thresholdsaresetbetween0.22and0.28toensure +Thismixedcorpusallowsthecrosscodertocapture stable policy updates, while both KL-divergence +| both reasoning-related | | | and | general | linguistic | fea- | | | | | | | | +| ---------------------- | --- | --- | --- | ------- | ---------- | ---- | --- | --- | --- | --- | --- | --- | --- | +andentropypenaltiesareturnedoff(coefficientsset +tures. Allcrosscodersaretrainedtoreconstructthe +| | | | | | | | tozero). | Themodelistrainedforoneepoch,and | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --- | +residualstreamofthemiddlelayerofeachmodel. thecheckpointfromthefinaliterationispreserved +Trainingisconductedwithabatchsizeof1024and forsubsequentevaluation. +| alearningrateof1×10−4. | | | | Theβ,whichcontrols | | | | | | | | | | +| ---------------------- | --- | --- | --- | ------------------ | --- | --- | ---------- | --- | ----------- | --- | ----- | --- | ------- | +| | | | | | | | Supervised | | Fine-Tuning | | (SFT) | | remains | +thesparsityregularizationstrength,issetto2. +| | | | | | | | a widely | adopted | | approach | | for transferring | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ------- | --- | -------- | --- | ---------------- | --- | +A.2 TrainingSetupforSFTandRL knowledge and desired behaviors from large +| | | | | | | | pre-trained | language | | models | to | task-adapted | or | +| --- | --- | --- | --- | --- | --- | --- | ----------- | -------- | --- | ------ | --- | ------------ | --- | +Thissubsectiondetailsthetrainingsetupusedfor +| | | | | | | | resource-constrained | | | models. | In | particular, | SFT | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | ------- | --- | ----------- | --- | +SFTandRL,includingthedatasets,optimization +| | | | | | | | has recently | | been extensively | | used | as a | form of | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | ---------------- | --- | ---- | ---- | ------- | +objectives,implementationdetails,andtraininghy- +| | | | | | | | reasoning | distillation, | | where | the | intermediate | | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | ----- | --- | ------------ | --- | +perparametersforbothparadigms. +| | | | | | | | reasoning | processes | | of a stronger | | teacher | model | +| --- | --- | --- | --- | --- | --- | --- | --------- | --------- | --- | ------------- | --- | ------- | ----- | +aredistilledintoasmallerstudentmodelthrough +| TrainingDatasets | | | Totrainourmodels,weadopt | | | | | | | | | | | +| ---------------- | --- | --- | ------------------------ | --- | --- | --- | ------------- | --- | ----------------- | --- | --- | ---------------- | --- | +| | | | | | | | high-quality, | | teacher-generated | | | chain-of-thought | | +thehigh-qualitymathematicsdatasetconstructed +byHuanetal.(2025),whichconsistsof47Khigh- annotations(Guoetal.,2025). Byminimizingthe +cross-entropylossoncurateddatasets,SFTallows +qualitymathematicsproblemsderivedfromMATH +| | | | | | | | the model | to | internalize | | desired | behaviors | and | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----------- | --- | ------- | --------- | --- | +(Hendrycksetal.,2021)andDeepScaler(Luoetal., +2025). Bothtrainingparadigmsareappliedtothe reasoningpatternsinafullysupervisedmanner. In +| | | | | | | | our experiments, | | we | employ | the | LLaMA-Factory | | +| -------------- | --- | ------------- | --- | --------- | ----- | -------- | ---------------- | ------ | --- | ------- | ----- | ------------- | --- | +| same backbone, | | Qwen3-4B-Base | | | and | Qwen2.5- | | | | | | | | +| | | | | | | | framework | (Zheng | | et al., | 2024) | to fine-tune | the | +| 7B. For | RL, | the model | is | optimized | using | stan- | | | | | | | | +dardGroupRelativePolicyOptimization(GRPO, twomodelsonteacher-providedchain-of-thought +5×10−5 +Shaoetal.,2024),whererewardsarecomputedby traces. The learning rate is set to with +| | | | | | | | a batch | size of | 128. | For consistency | | with | the RL | +| ------------ | --- | -------- | ----- | ------------ | ------- | ----- | -------- | -------- | ------- | --------------- | --- | ------- | ------ | +| comparing | the | model’s | final | answers | against | the | | | | | | | | +| | | | | | | | setting, | training | is also | performed | | for one | epoch, | +| gold answers | | provided | in | the dataset. | This | setup | | | | | | | | +strictlysupervisesoutcomecorrectnesswithoutex- andthefinalcheckpointisretainedfordownstream +evaluation. +posingintermediatereasoningtracestothemodel. +AlltrainingrunswereconductedonH200and +| For SFT, | the | training | targets | are | complete | chain- | | | | | | | | +| ------------------------------------------ | --- | -------- | ------- | --- | -------- | ------ | --------- | --- | --- | --- | --- | --- | --- | +| of-thought(CoT)reasoningtracesgeneratedbya | | | | | | | H100GPUs. | | | | | | | +strongteachermodel,Qwen3-32B-Instruct(Yang +A.4 DetailedDescriptionoftheBenchmarks +andEvaluationMetrics +1https://huggingface.co/datasets/open- +thoughts/OpenThoughts-114k +| | | | | | | | In the | experiment, | | we | evaluated | our | models | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----------- | --- | --- | --------- | --- | ------ | +2https://ai.gitee.com/hf-datasets/togethercomputer/RedPajama- +| Data-1T-Sample | | | | | | | on a broad | | range | of benchmarks | | designed | to | +| -------------- | --- | --- | --- | --- | --- | --- | ---------- | --- | ----- | ------------- | --- | -------- | --- | + +probe different aspects of reasoning and gener- • ARC-Challenge (Clark et al., 2018): The +alization. These benchmarks can be broadly more difficult subset of the AI2 Reasoning +groupedintomathematicalreasoningtasksandnon- Challenge, containing grade-school science +mathematicaltasks,whichvaryindomainknowl- questionsthattypicallyrequiremulti-hoprea- +edgeandreasoningrequirements. soning and background knowledge integra- +tion. +| Math reasoning | | tasks | The | following | | bench- | | | | | | | | +| -------------- | --- | ----- | --- | --------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +marksprimarilyevaluateamodel’sabilitytoper- +| | | | | | | | Collectively, | | these | benchmarks | span | a | wide | +| --------------- | ---------- | -------- | ------------ | --- | ---------- | ------ | ------------- | ------- | ----- | ---------- | ------ | ----- | ---- | +| form explicit | multi-step | | mathematical | | reasoning, | | | | | | | | | +| | | | | | | | range of | domains | and | reasoning | types. | Their | di- | +| often requiring | | symbolic | manipulation | | and | struc- | | | | | | | | +versityallowsustoprobewhetherimprovements +turedproblemsolving. +inducedbyRLtrainingreflecttask-specificadapta- +• MATH500(Hendrycksetal.,2021): Asub- tionsorgeneral-purposereasoningcapabilitiesthat +| setof500problemssampledfromtheMATH | | | | | | | transferacrossdomains. | | | | | | | +| ---------------------------------- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | +dataset,coveringalgebra,geometry,number WeusedLLM-Evaluation-Harness(Gaoetal., +| theory,andcombinatorics. | | | | Eachproblemtyp- | | | | | | | | | | +| ------------------------ | --- | --- | --- | --------------- | --- | --- | --------- | -------- | --- | ------- | ----------- | --- | --- | +| | | | | | | | 2024b) to | evaluate | the | models’ | performance | | on | +icallyrequiresmulti-stepderivationsandpre- OpenBookQA,CommonsenseQA,HeadQA,SciQ, +cisenumericalorsymbolicanswers. andARC-Challenge,andusedEval-Chemy(Raoof +| | | | | | | | et al., 2025) | | to evaluate | the | performance | | on | +| -------- | --- | -------- | -------- | --- | ------ | ----- | ------------------------- | --- | ----------- | --- | ----------- | ----------- | --- | +| • AIME24 | / | AIME25: | Problems | | drawn | from | | | | | | | | +| | | | | | | | MATH500,AIME24,andAIME25. | | | | | Inourexper- | | +| the 2024 | | and 2025 | editions | | of the | Amer- | | | | | | | | +iments,weadoptedexact-matchaccuracytoeval- +| ican | Invitational | | Mathematics | | Examination | | | | | | | | | +| ---- | ------------ | --- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +uatethemodels’performanceonmathreasoning +(AIME).Eachbenchmarkconsistsof30chal- +| | | | | | | | tasks. Specifically,forAIME24andAIME25,we | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | +lengingshort-answerquestionsthatdemand +| | | | | | | | averaged | accuracy | on | 10 repetitions. | | For MATH | | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --- | --------------- | --- | -------- | --- | +carefulreasoningandmathematicalinsight. +500,ourscoreistheaverageaccuracyover3repe- +| Other tasks | The | following | | benchmarks | | assess | | | | | | | | +| ----------- | ------------- | --------- | --- | ---------- | --- | ------ | -------------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | | | | | titions. Forotherbenchmarks,wereportaccuracy | | | | | | | +| reasoning | and knowledge | | use | outside | the | mathe- | | | | | | | | +followingstandardevaluationprotocols. +maticaldomain,makingthemparticularlysuitable +| forevaluatinggeneralization. | | | | | | | B MoreExperimentalResults | | | | | | | +| ---------------------------- | --- | --------- | --- | --------- | ----------- | ------ | ------------------------- | --- | ---------- | ---------- | --- | ---------- | --- | +| • OpenBookQA | | (Mihaylov | | et | al., 2018): | A | | | | | | | | +| | | | | | | | In this section, | | we present | additional | | experimen- | | +| multiple-choice | | question | | answering | | bench- | | | | | | | | +talresultsthatcomplementthemainfindingsdis- +| mark | focused | on | elementary | | science | knowl- | | | | | | | | +| ---- | ------- | --- | ---------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | +cussedinthemainpaper. +| edge. | Each | question | is | associated | with | a set | | | | | | | | +| ------- | ------ | -------- | --------- | ---------- | ------ | ----- | --------------------------------------- | --- | --- | --- | --- | --- | --- | +| of core | facts, | and | the model | must | select | the | | | | | | | | +| | | | | | | | B.1 FeatureRankShiftsBetweenConsecutive | | | | | | | +correctanswerfromfouroptions. +Checkpoints +• CommonsenseQA (Talmor et al., 2019): A To further examine the dynamics of feature evo- +multiple-choice benchmark designed to test lution during training, for each pair of adjacent +generalcommonsenseknowledge. Questions checkpoints,wecomputehowmucheachfeature’s +are constructed around concepts from struc- rank(basedonNRN)changesfromonecheckpoint +| turedknowledgebases,withdistractoroptions | | | | | | | tothenext. | | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | +chosentobesemanticallyplausible. Theresults,visualizedinFigure8to11,reveal +distinctpatternsforSFTandRL.IntheSFT-tuned +| • HeadQA | | (Vilares | and | Gómez-Rodríguez, | | | | | | | | | | +| -------- | ------------------------------- | -------- | --- | ---------------- | --- | --- | ------------------------------------------ | --------- | ------ | ------ | --- | ------ | --- | +| | | | | | | | model, rankchangesbetweenconsecutivecheck- | | | | | | | +| 2019): | Amedicalquestionansweringbench- | | | | | | | | | | | | | +| | | | | | | | points are | generally | small. | First, | the | number | of | +markcomposedofmultiple-choicequestions +| | | | | | | | blank features | | is limited, | where | a blank | | feature | +| ------- | ---- | ---------- | --- | -------------- | --- | --- | -------------- | --- | ----------- | ----- | ------- | --- | ------- | +| derived | from | healthcare | | specialization | | ex- | | | | | | | | +denotesonethatappearsinthetop50atonecheck- +| ams, | including | questions | | across | pharmacol- | | | | | | | | | +| ---- | --------- | --------- | --- | ------ | ---------- | --- | --------- | ----- | ------- | ------- | ----- | ------------ | --- | +| | | | | | | | point but | falls | outside | the top | 50 at | the adjacent | | +ogy,chemistry,nursing,psychology,biology, +| | | | | | | | checkpoint. | Second, | | most features | | exhibit | rela- | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | --- | ------------- | --- | ------- | ----- | +andmedicine. +tivelysmallrankshifts,asreflectedbylightercol- +• SciQ(Welbletal.,2017): Asciencequestion orsinthevisualization. Moreover,somefeatures +answering dataset focusing on elementary- consistentlyappearinthetop50acrossallcheck- +levelscientificconcepts. points,indicatingthatcertainSFT-inducedfeatures + +| | | | Qwen2.5-7B | | | | Qwen2.5-7B-SFT | | | | Qwen2.5-7B-RL | | | +| --- | ----- | --- | ---------- | --- | --- | ----- | -------------- | --- | --- | ----- | ------------- | --- | --- | +| | 104 | | | | | 104 | | | | 104 | | | | +| | 103 | | | | | 103 | | | | 103 | | | | +| | tnuoC | | | | | tnuoC | | | | tnuoC | | | | +| | 102 | | | | | 102 | | | | 102 | | | | +| | 101 | | | | | 101 | | | | 101 | | | | +| | 100 | | | | | 100 | | | | 100 | | | | +0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 +Model Attribution Score Model Attribution Score Model Attribution Score +Figure5: DistributionofModelAttributionScoresacrossdifferenttrainingmethodsonQwen2.5-7B. +Llama3.1-8b-Instruct-SFT Llama3.1-8b-Instruct-RL Llama3.1-8b-Instruct-SFT Llama3.1-8b-Instruct-RL +| | | | | | | | | | | | 1.0 | | 1.0 | +| ----- | --- | --- | --- | ----- | --- | --- | --- | ----------- | --------- | --------- | ------ | -------------- | --------- | +| 104 | | | | 104 | | | | | | | | | | +| | | | | | | | | ckpt-1 1.00 | 0.32 0.28 | 0.33 0.33 | ckpt-1 | 1.00 0.09 0.08 | 0.08 0.06 | +| | | | | | | | | | | | 0.8 | | 0.8 | +| 103 | | | | 103 | | | | 0.32 | 1.00 0.45 | 0.45 0.45 | | 0.09 1.00 0.30 | 0.22 0.15 | +| tnuoC | | | | tnuoC | | | | ckpt-2 | | | ckpt-2 | | | +| | | | | | | | | | | | 0.6 | | 0.6 | +102 102 ckpt-3 0.28 0.45 1.00 0.56 0.54 ckpt-3 0.08 0.30 1.00 0.32 0.22 +| 101 | | | | 101 | | | | | | | 0.4 | | 0.4 | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --------- | --------- | ------ | -------------- | --------- | +| | | | | | | | | ckpt-4 0.33 | 0.45 0.56 | 1.00 0.72 | ckpt-4 | 0.08 0.22 0.32 | 1.00 0.28 | +| 100 | | | | 100 | | | | | | | 0.2 | | 0.2 | +0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 ckpt-5 0.33 0.45 0.54 0.72 1.00 ckpt-5 0.06 0.15 0.22 0.28 1.00 +| | Normalized Relative Norm | | | | Normalized Relative Norm | | | | | | | | | +| -------- | ------------------------ | ------------------------------------- | -------- | ------- | ------------------------ | ------------ | --- | ------------------------------ | ------- | --------- | -------- | ------------------------------ | -------- | +| | | | | | | | | ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 | | | 0.0 | ckpt-1ckpt-2ckpt-3ckpt-4ckpt-5 | 0.0 | +| | | | | | | | | Figure 7: | Feature | overlap | heatmaps | across | training | +| Figure6: | | DistributionofNormalizedRelativeNorms | | | | | | | | | | | | +| | | | | | | | | checkpoints | under | different | | training paradigms | on | +| across | different | | training | methods | on | Llama3.1-8B- | | | | | | | | +Llama3.1-8B-Instruct. +Instruct. +| | | | | | | | | model-specific | | features, | while | RL preserves | the | +| --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | --------- | ----- | ------------ | --- | +areestablishedearlyandpersistthroughouttrain- +baserepresentationsandinducesmorerestrained +ing. Together,theseobservationssuggestthatSFT +| | | | | | | | | feature | changes. | In | addition, | Figure | 7, 12, and | +| --- | --- | --- | --- | --- | --- | --- | --- | ------- | -------- | --- | --------- | ------ | ---------- | +quicklyestablishesarelativelystablehierarchyof +| | | | | | | | | 13 show | that | SFT features | | stabilize early | during | +| --- | --- | --- | --- | --- | --- | --- | --- | ------- | ---- | ------------ | --- | --------------- | ------ | +model-specificfeatures,withlatertrainingprimar- +training,whereasRLexhibitsmoregradualfeature +ilyrefiningtheirrelativeimportance. +evolution. +| | By contrast, | | the RL-tuned | | model | exhibits | sub- | | | | | | | +| -------------------------------------------- | ------------ | --- | ------------ | --- | ----- | -------- | ---- | ------------------------------------- | --- | --- | --- | --- | --- | +| stantiallylargerrankshiftsacrosscheckpoints. | | | | | | | A | | | | | | | +| | | | | | | | | B.3 AdditionalMASResultsforQwen2.5-7B | | | | | | +| large | number | of | features | are | blank | between | ad- | | | | | | | +Figure5providesadditionalMASdistributionson +| jacent | checkpoints, | | indicating | | frequent | | turnover | | | | | | | +| ------ | ------------ | --- | ---------- | --- | -------- | --- | -------- | --- | --- | --- | --- | --- | --- | +Qwen2.5-7B.Thesamequalitativetrendspersist: +| among | top-ranked | | features. | | Features | frequently | | | | | | | | +| ----- | ---------- | --- | --------- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | --- | +SFTinducesaheavilyright-skewedMASdistribu- +| undergo | | significant | | reordering, | with | previously | | | | | | | | +| ------- | --- | ----------- | --- | ----------- | ---- | ---------- | --- | --- | --- | --- | --- | --- | --- | +tionwithmanyhighlyattributedfeatures,whereas +prominentfeaturesdiminishinginimportanceand +RLresultsinasignificantlyflatterdistributionwith +| new | features | rising | to | prominence | | at later | stages. | | | | | | | +| --- | -------- | ------ | --- | ---------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +fewerstronglymodel-specificfeatures. +ThispatternsuggeststhatRL-inducedfeaturesare +formedgraduallyandslowly,withtrainingcontin- +| | | | | | | | | B.4 GeneralizationtoUnseenTasks. | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --- | +uouslyadjustingwhichinternalfeaturesaremost +Tofurtherevaluatewhethertheidentifiedgeneral- +relevantforproducingcorrectoutcomes. +| | | | | | | | | ization features | | extend | beyond | the tasks | used for | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ------ | ------ | --------- | -------- | +featureidentification,weconductadditionalexper- +B.2 ResultsonLlama3.1-8B-Instruct +| | | | | | | | | imentsontwounseenbenchmarks: | | | | LogiQA(Liu | | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ---------- | --- | +To evaluate the robustness and generality of our etal., 2021), alogical reasoningbenchmark, and +findingsacrossmodelfamilies,weextendouranal- PIQA(Bisketal.,2020),aphysicalcommonsense +| ysistoLlama3.1-8B-Instruct. | | | | | Followingthesame | | | | | | | | | +| --------------------------- | --- | --- | --- | --- | ---------------- | --- | --- | -------- | --------- | ---------- | --- | ----- | ---------- | +| | | | | | | | | question | answering | benchmark. | | These | tasks dif- | +controlledexperimentalsetupdescribedinSection fersubstantiallyfromtheoriginalevaluationsetin +4,wetrainbothSFT-andRL-tunedmodelsfrom domain and question structure, and are not used +thesamebasemodelusingidenticaldata. +duringfeatureselection. +The results show consistent trends with those Weperformthesamefeature-levelinterventions +observed in the Qwen models. As shown in Fig- asinthemainexperiments. Specifically,we(i)set +ure 6, SFT introduces a larger number of highly theidentifiedgeneralizationfeaturestozerointhe + +| Model | | | LogiQA | PIQA | +| ------------- | --- | --- | ------ | ----- | +| Qwen3-4B-RL | | | -24.5 | -17.6 | +| Qwen2.5-7B-RL | | | -24.0 | -11.8 | +Table4: Performancedegradationonunseentasksby +zeroinggeneralizationfeaturesintheRL-tunedmodel. +| Model | | | LogiQA | PIQA | +| ------------- | --- | --- | ------ | ----- | +| Qwen3-4B-Base | | | +23.3 | +28.2 | +| Qwen2.5-7B | | | +24.0 | +32.9 | +Table5: Performanceimprovementsonunseentasksby +amplifyinggeneralizationfeaturesinthebasemodel. +RL-trainedmodels,and(ii)amplifythesamefea- +turesinthebasemodels,andevaluateperformance +ongeneralization-criticalsamples. +| Table | 4 shows | that | zeroing the | generalization | +| -------- | ------------- | ---- | ------------ | -------------- | +| features | in RL-trained | | models leads | to clear per- | +formancedegradationonbothLogiQAandPIQA. +Conversely,asshowninTable5,amplifyingthese +featuresinthebasemodelsconsistentlyimproves +| performanceontheunseentasks. | | | | Theseresultsfur- | +| ---------------------------- | --- | --- | --- | ---------------- | +thersupportthattheidentifiedfeaturesimplement +ageneral-purposegeneralizationmechanismrather +thantask-specificheuristics. + +Qwen3-4B-SFT +50 +| 64 | 64 | 396 | 64 | | +| --- | --- | --- | --- | --- | +| 255 | 255 | 525 | 255 | | +396 +| | 396 | 2513 | 396 | | +| ---- | ------- | ------- | ------- | --- | +| 995 | 995 | 2618 | 525 | | +| 1546 | 1546 | | | | +| 2513 | | 2685 | 2139 | | +| | 2513 | 2955 | | | +| 2618 | 2 6 1 8 | | 2513 | | +| 2955 | | 3 1 5 2 | 2 6 1 8 | | +| 3152 | 2 9 5 5 | 3 5 9 9 | | 40 | +| | 3 1 5 2 | | 2 6 8 5 | | +| 3290 | | 4 0 5 5 | 2 9 5 5 | | +| 3819 | 3 8 1 9 | 4 3 3 7 | | | +| | 4 0 5 5 | | 3 1 5 2 | | +| 5976 | | 5976 | 3291 | | +| 6132 | 4337 | 6857 | | | +| 6663 | 6857 | | 3469 | | +| | | 7460 | 3599 | | +| 6857 | 7460 | 7746 | | | +| 7460 | 8271 | | 4337 | 30 | +| 7857 | 8865 | 7978 | 5505 | | +8271 +| 8520 | 8997 | | 5976 | | +| ---- | ------- | ------- | ---- | --- | +| 8578 | 9116 | 8865 | | | +| 8865 | | 9116 | 6333 | | +| | 9281 | | 6857 | | +| 8997 | 9286 | 9281 | | | +| 9116 | | 9286 | 7746 | | +| 9281 | 9402 | | 7883 | | +| | 9 5 4 5 | 9 4 0 2 | | | +| 9286 | | 9 5 4 5 | 7978 | 20 | +| 9545 | 9 8 4 5 | | 8271 | | +10380 9845 )regnorts = evitagen ,rekaew = evitisop( tfihS knaR +| 9845 | 10406 | 10380 | 8520 | | +| ----- | ----- | ----- | ---- | --- | +| 10380 | | | 8555 | | +| 10406 | 10843 | 10406 | | | +| | 11410 | 11410 | 8578 | | +10843 +| 11410 | 11928 | 11846 | 8865 | | +| ----- | ----- | ----- | ---- | --- | +| 11846 | 11997 | 11928 | 9116 | | +10 +| 11928 | 12734 | 11997 | 9281 | | +| ---------------- | ----- | ----- | ----- | --- | +| 11997 | 12788 | | 9286 | | +| 12734 | | 12734 | | | +| | 13288 | 13744 | 9545 | | +| 12788 | 13744 | | 10380 | | +| 13288 | 14080 | 14080 | | | +| DI erutaeF 13744 | | 14352 | 10406 | | +| | 14352 | | 11410 | | +| 13933 | 14381 | 14381 | | | +| 14080 | | 14392 | 11846 | | +| 14111 | 14392 | | | 0 | +| | 14458 | 14458 | 12734 | | +| 14336 | | 14545 | 13488 | | +| 14392 | 14545 | | | | +| | 14688 | 15151 | 13744 | | +| 14458 | | 18362 | 14336 | | +| 14545 | 15151 | | | | +| 14688 | 15762 | 18403 | 14352 | | +| | | 18686 | 14392 | | +| 15762 | 18299 | | | | +| 16834 | 18362 | 18802 | 14458 | 10 | +| 18299 | 18403 | 19164 | 14545 | | +| 18362 | 18686 | 19325 | 14756 | | +| 20826 | 19325 | 19583 | | | +| 21495 | | | 18362 | | +| | 19583 | 19979 | 18403 | | +| 21594 | 19979 | 21495 | | | +| 21608 | | | 18686 | | +| 21853 | 20826 | 21594 | 18802 | | +| | 21495 | 21608 | | | +| 21906 | | | 19164 | 20 | +| 22103 | 21608 | 21906 | 19325 | | +| | 21906 | 22103 | | | +| 22152 | 22103 | | 19979 | | +| 22629 | | 22152 | 21594 | | +| 22773 | 22152 | 22629 | | | +| | 22629 | | 21608 | | +| 23294 | | 23080 | | | +| 23474 | 23474 | 24451 | 22152 | | +| 23547 | 24368 | | 22629 | | +24481 +| 24368 | 24451 | 25087 | 23080 | 30 | +| ----- | --------- | --------- | --------- | --- | +| 24804 | 24481 | | 24481 | | +| 25087 | | 25291 | | | +| | 25087 | 25555 | 25087 | | +| 25310 | 25291 | | 25555 | | +| 25555 | 2 5 5 5 5 | 26120 | | | +| 26101 | | 2 7 1 0 1 | 2 6 1 2 0 | | +| | 2 6 1 0 1 | | 2 6 9 2 9 | | +| 27101 | 2 6 1 2 0 | 2 7 6 8 1 | | | +| 28805 | | 2 7 9 5 4 | 2 7 1 0 1 | | +| | 2 7 9 5 4 | | | 40 | +| 29122 | 2 8 8 0 5 | 2 8 8 0 5 | 2 7 5 2 0 | | +| 29214 | | 29214 | 27681 | | +| 29719 | 29122 | | | | +| | 29214 | 29485 | 28805 | | +| 29721 | | 29719 | 29214 | | +| 29792 | 29719 | | | | +| 30253 | 29721 | 29721 | 29485 | | +| | | 30381 | 29721 | | +| 30333 | 30381 | | | | +| 32267 | 32507 | 32507 | 30381 | | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure8: FeaturerankshiftacrossadjacentcheckpointsduringSFTonQwen3-4B-Base. + +Qwen3-4B-RL +50 +| 254 | 254 | 86 | 86 | | +| ---- | ------- | ------- | ------- | --- | +| 1520 | 1171 | 158 | 158 | | +| 1742 | 1308 | 254 | 254 | | +| 1868 | 1520 | 805 | 645 | | +| 2117 | 1742 | 1171 | 805 | | +| 2357 | 1868 | 1308 | 946 | | +| 2537 | | | 1523 | | +| 2712 | 1929 | 1929 | 1883 | | +| | 2117 | 2357 | | | +| 2756 | 2 3 5 7 | 2 4 7 0 | 2 3 5 7 | | +| 2792 | 2 4 7 0 | 2 4 9 9 | 2 4 9 9 | 40 | +| 3289 | 2537 | 2712 | 2601 | | +| 3585 | | 2 7 5 6 | 2 7 1 2 | | +| 3609 | 2 7 1 2 | 3 0 0 9 | 2 7 5 6 | | +| 4088 | 2 7 5 6 | | 3 0 0 9 | | +| 4430 | 3 0 0 9 | 3 0 2 5 | 3 0 2 5 | | +| 4967 | 4 0 8 8 | 3 2 1 5 | | | +| | 4 4 3 0 | 4 4 3 0 | 3 2 1 5 | | +| 4998 | | 4 5 4 7 | 4 4 3 0 | | +| 5001 | 4 9 9 8 | 4 9 9 8 | 4 5 4 7 | | +| 5120 | 5 0 0 1 | 5 3 3 1 | 4 7 7 4 | 3 0 | +| 5286 | 5 2 8 6 | | 5 4 1 0 | | +| 5648 | 5 3 3 1 | 5 4 1 0 | 5 5 8 7 | | +| 5938 | 5 4 1 0 | 5 5 8 7 | 6 0 6 3 | | +| 6000 | 5 6 4 8 | 5 7 3 5 | | | +| 6292 | | 6 0 0 0 | 6 2 6 5 | | +| | 5 7 3 5 | 6 1 1 5 | 6 4 1 2 | | +| 6348 | 5 9 3 8 | 6 2 6 5 | 6 5 1 1 | | +| 6663 | 6 0 0 0 | | 6 8 1 6 | | +| 6932 | 6 1 1 5 | 6 3 4 8 | 7 1 9 0 | | +| 7027 | 6 3 4 8 | 6 8 9 5 | 7 2 8 8 | | +| 7038 | | 6 9 6 1 | 7 3 4 9 | 2 0 | +| 7524 | 6 6 6 3 | 7 4 9 7 | 7 4 9 7 | | +7878 6 8 9 5 7 5 2 4 )regnorts = evitagen ,rekaew = evitisop( tfihS knaR +| 8080 | 6 9 3 2 | 7 5 3 2 | 7 7 2 6 | | +| ---------------- | --------- | --------- | --------- | --- | +| | 6 9 6 1 | | 8 9 9 9 | | +| 8183 | 7 0 3 8 | 7 6 9 5 | 9 2 5 9 | | +| 8329 | | 7 7 2 6 | 9 2 8 6 | | +| 9253 | 7 5 2 4 | 8 4 3 6 | 9 6 3 6 | | +| 9286 | 7 5 3 2 | 8 7 3 3 | 10 0 3 8 | | +| 9372 | 7695 | 9259 | 10471 | | +| 10558 | 7878 | 9286 | | 10 | +| 10684 | 8080 | | 10737 | | +| 10737 | | 10471 | 11007 | | +| | 8183 | 10695 | 11262 | | +| 11007 | 8436 | 10737 | 12086 | | +| 11766 | 8733 | 10803 | 13220 | | +| 12146 | 9286 | 11262 | 13931 | | +| DI erutaeF 12296 | 10558 | 12086 | 14540 | | +| 12583 | 10684 | 13220 | | | +| 13422 | | | 14798 | | +| 13568 | 10695 | 13931 | 15448 | | +| 14101 | 10803 | 14563 | 15656 | 0 | +| | 11262 | 15448 | 15755 | | +| 14449 | 12086 | 15656 | 15840 | | +| 14609 | 12296 | 15755 | 16458 | | +| 14798 | | 15840 | 16814 | | +| 15448 | 13568 | | | | +| 15840 | 14449 | 16458 | 17012 | | +| 16655 | 14563 | 17315 | 17315 | | +| 16847 | 15448 | 17703 | 17881 | | +| 17061 | 15755 | 17821 | 18729 | | +| | | 17881 | 18766 | 10 | +| 17276 | 15840 | 18729 | 19527 | | +| 17315 | 16655 | | 1 9 8 1 5 | | +| 17317 | 1 7 3 1 5 | 1 8 7 6 6 | 2 1 2 2 6 | | +| 17480 | 1 7 4 8 0 | 1 9 5 2 7 | | | +| 17703 | 1 7 7 0 3 | 1 9 5 8 3 | 2 1 2 8 7 | | +| 17881 | 1 7 8 2 1 | 2 0 8 1 9 | 2 1 4 2 2 | | +| 18415 | | 2 1 2 2 6 | 2 1 5 1 3 | | +| 18575 | 1 8 8 1 1 | 2 1 4 2 2 | 2 1 5 1 4 | | +| | 1 9 5 8 3 | | 2 1 7 0 2 | | +| 18721 | 1 9 8 1 5 | 2 1 4 8 8 | 2 1 7 5 5 | | +| 18811 | 2 0 6 7 6 | 2 1 6 2 7 | 2 1 9 4 5 | 2 0 | +| 19072 | 2 0 8 1 9 | 2 1 6 6 1 | | | +| 19527 | | 2 1 7 0 2 | 2 2 0 8 3 | | +| 19815 | 2 1 2 8 7 | 2 2 0 8 3 | 2 2 7 4 6 | | +| 19966 | 2 1 4 2 2 | 2 2 9 1 9 | 2 3 0 1 1 | | +| 20676 | 2 1 4 8 8 | | 2 3 0 6 9 | | +| 20910 | 2 1 6 2 7 | 2 3 0 1 1 | 2 3 1 9 2 | | +| | 2 1 6 6 1 | 2 3 0 6 9 | 2 3 8 2 4 | | +| 21287 | | 2 3 1 9 2 | 2 3 8 7 2 | | +| 21661 | 2 1 7 0 2 | 2 3 9 5 6 | | | +| 21702 | 2 2 9 1 9 | 2 4 3 4 1 | 2 3 9 5 6 | | +| 22919 | 2 3 0 6 9 | 2 4 3 6 8 | 2 4 3 6 8 | 3 0 | +| 23192 | 2 3 1 9 2 | | 2 4 4 7 4 | | +| 23220 | 2 3 9 5 6 | 2 4 4 1 5 | 2 4 9 9 8 | | +| 23956 | 2 4 3 4 1 | 2 4 9 9 8 | 2 5 0 0 9 | | +| 24243 | | 2 5 0 0 9 | 2 5 4 1 4 | | +| | 2 4 3 9 8 | 2 5 4 1 4 | 2 6 0 8 0 | | +| 24341 | 2 4 4 1 5 | 2 6 0 8 0 | 2 6 1 5 1 | | +| 24398 | 2 4 9 9 8 | 2 6 2 1 3 | | | +| 24931 | 25049 | 26891 | 2 6 8 9 1 | | +| 25049 | 2 5 5 5 4 | | 27486 | | +| 25554 | | 2 7 4 8 6 | 2 7 5 4 9 | 40 | +| 27347 | 2 6 0 8 0 | 2 7 5 4 9 | 2 8 6 1 3 | | +| 27468 | 26213 | 27704 | 28854 | | +| 27486 | 27486 | 28613 | 29081 | | +| 27505 | 27505 | 29081 | 29147 | | +| | 27704 | 29147 | | | +| 28613 | | | 30542 | | +| 28699 | 29081 | 29155 | 31350 | | +| 29146 | 29155 | 30803 | 31817 | | +| 30400 | 30400 | 31817 | 31903 | | +| 30542 | 30803 | 31903 | 32219 | | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure9: FeaturerankshiftacrossadjacentcheckpointsduringRLonQwen3-4B-Base. + +Qwen2.5-7B-SFT +50 +| 15 | 440 | 440 | 15 | | +| ---- | ------- | ---- | ---- | --- | +| 440 | 659 | 1097 | 440 | | +| 646 | 1097 | | | | +| | 1 2 2 1 | 1221 | 1097 | | +| 659 | | 1542 | 1221 | | +| 1097 | 1 8 9 5 | 1895 | | | +| 1221 | 1962 | | 1542 | | +| | 2241 | 1962 | 1962 | | +| 1962 | 2332 | 2241 | 2241 | | +2241 +| 2842 | 2689 | 2332 | 2332 | 40 | +| ---------------- | --------- | --------- | --------- | --------------------------------------------------- | +| | 2842 | 2613 | 2613 | | +| 3203 | 3 2 0 3 | | | | +| 5331 | 5 3 5 8 | 2 6 8 9 | 2 6 8 9 | | +| | | 2 8 4 2 | 3 2 0 3 | | +| 5660 | 5 5 9 9 | 3203 | | | +| 5824 | 5660 | | 5797 | | +| 6184 | 5797 | 5358 | 6184 | | +| | 6184 | 5599 | | | +| 6323 | | | 6437 | | +| 7213 | 6323 | 5797 | 7213 | 30 | +| 8117 | 6974 | 6184 | | | +| | 7213 | | 7542 | | +| 8509 | 8509 | 6974 | 7898 | | +| 8642 | | 7213 | | | +| 9534 | 8642 | 7542 | 8117 | | +| | 9534 | | 8509 | | +| 10583 | 10583 | 8117 | | | +| 10893 | | 8509 | 9761 | | +| 11186 | 10893 | | 10583 | | +| | 10934 | 9761 | | | +| 12220 | 11186 | 10893 | 10893 | 20 | +| 12502 | 12220 | | 11525 | | +| | | 10934 | 11674 | )regnorts = evitagen ,rekaew = evitisop( tfihS knaR | +| 12698 | 13717 | 11674 | | | +| 13744 | 13744 | 12698 | 12698 | | +| 13748 | 1 3 7 4 8 | | 1 3 6 0 5 | | +| | 1 3 9 9 1 | 1 3 7 1 7 | | | +| 14301 | | 1 3 9 9 1 | 1 3 7 1 7 | | +| 14697 | 1 4 3 0 1 | | 14906 | | +| 14906 | 14697 | 14906 | | | +| | 15353 | 14914 | 14914 | 10 | +| 15258 | 15455 | | 15258 | | +| 15353 | | 15258 | | | +| 15455 | 15757 | 15353 | 15353 | | +| | 16099 | 16099 | 16099 | | +| 15757 | 1 6 1 5 1 | | | | +| 16008 | 1 6 2 7 3 | 16151 | 16715 | | +| DI erutaeF 16273 | | 16361 | 16991 | | +16361 +| 16715 | 16715 | 16715 | 17819 | | +| ----- | ----- | ----- | ----- | --- | +| 17073 | 16991 | 16991 | 18508 | | +| 17819 | 17819 | | | 0 | +| | | 17819 | 19264 | | +| 18403 | 18508 | 18508 | 19340 | | +| 18700 | 18700 | 18719 | 19998 | | +1 8 7 1 9 +| 18719 | 1 9 2 6 4 | 19264 | 20148 | | +| ----- | --------- | --------- | --------- | --- | +| 19264 | | 19340 | 20637 | | +| 19340 | 19340 | | | | +| | 19998 | 19998 | 20642 | | +| 19998 | 20070 | 20148 | 20821 | | +| 20070 | | | | 10 | +| 20146 | 20146 | 20510 | 21823 | | +| | 20148 | 20637 | 22681 | | +| 20148 | 2 0 2 7 4 | 2 0 6 4 2 | | | +| 20274 | 2 0 5 1 0 | | 2 2 8 7 4 | | +| 21359 | | 2 0 8 2 1 | 2 3 2 3 1 | | +| | 2 2 1 2 8 | 21823 | | | +| 22128 | 22173 | | 23472 | | +| 22880 | 22880 | 22173 | 23515 | | +| 23221 | 23231 | 22681 | | | +23920 +| 23231 | 23472 | 22874 | 24699 | 20 | +| ----- | ----- | ----- | ----- | --- | +| 23472 | 23637 | 23231 | | | +| | 24299 | 23637 | 25727 | | +| 25339 | 24418 | | 26051 | | +| 25727 | | 23920 | | | +| 25787 | 25727 | 24299 | 26279 | | +| | 26051 | | 26506 | | +| 26051 | 26613 | 24418 | 26613 | | +| 26385 | 27227 | 24699 | | | +| 26613 | | | 27550 | | +| | 27738 | 25727 | 27678 | | +| 27738 | 27778 | 26051 | | 30 | +| 27778 | 28651 | 27227 | 27778 | | +| 28651 | 28732 | | 28937 | | +27550 +| 28791 | 29009 | 27778 | 29009 | | +| ----- | --------- | --------- | --------- | --- | +| 28937 | 2 9 2 7 6 | | 2 9 4 0 7 | | +| 29448 | 2 9 4 0 7 | 2 8 7 3 2 | | | +| | 2 9 4 4 8 | 2 9 0 0 9 | 2 9 6 3 3 | | +| 29660 | | | 30264 | | +| 29835 | 29633 | 29276 | | | +| | 29660 | 29407 | 30659 | 40 | +| 30264 | 29835 | 29448 | 31036 | | +30613 +| 30659 | 30264 | 29633 | 31931 | | +| ----- | --------- | ----- | ----- | --- | +| | 30659 | 30264 | 31945 | | +| 30984 | 3 0 9 8 4 | | | | +| 31723 | 3 1 1 3 4 | 30659 | 32061 | | +| 31945 | | 31134 | 32113 | | +| | 31723 | | 32232 | | +| 32232 | 32232 | 31931 | | | +| 32521 | 32521 | 31945 | 32735 | | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure10: FeaturerankshiftacrossadjacentcheckpointsduringSFTonQwen2.5-7B. + +Qwen2.5-7B-RL +50 +| 521 | 459 | 267 | 267 | | +| ---------------- | --------- | --------- | --------- | --------------------------------------------------- | +| 710 | 521 | 459 | 710 | | +| 1175 | 710 | 710 | 845 | | +| 1202 | 1175 | 845 | | | +| 1307 | 1202 | | 864 | | +| 1802 | | 1030 | 1030 | | +| 2277 | 2514 | 1175 | 1175 | | +| 2514 | 2613 | 2558 | 2277 | | +| 2613 | 3 5 8 7 | 2 6 1 3 | | | +| 3587 | 3 6 2 7 | | 2 5 5 8 | | +| | 3 6 5 0 | 3 5 8 7 | 2 5 7 8 | 40 | +| 3650 | | 3 6 2 7 | 2 6 1 3 | | +| 3676 | 3 6 7 6 | 3 6 5 0 | | | +| 3683 | 3 6 8 3 | 3 6 7 6 | 3 2 0 3 | | +| 5393 | 3 7 9 4 | | 3 5 8 7 | | +| 5960 | 5 4 2 3 | 3 6 8 3 | 3 6 5 0 | | +| 6062 | | 3 7 9 4 | | | +| 6153 | 6062 | 5 4 2 3 | 3676 | | +| 6267 | 6 1 5 3 | 6 0 6 2 | 3 6 8 3 | | +| 6836 | 6 2 6 7 | | 4 3 9 3 | | +| 7039 | 6 9 2 5 | 6 1 5 3 | 5 3 5 8 | 30 | +| | 7 2 7 9 | 6 1 7 7 | | | +| 7193 | | 6 9 2 5 | 5 4 2 3 | | +| 7368 | 7 3 6 8 | 7 2 7 9 | 5 4 6 4 | | +| 7427 | 7 4 2 7 | | 6 0 6 2 | | +| 7675 | 7 4 4 1 | 7 4 2 7 | | | +| 7778 | 9005 | 7441 | 6153 | | +| 7788 | | 8385 | 6177 | | +| 7926 | 9375 | | 6925 | | +| 8706 | 10688 | 9005 | | | +| 8752 | 10774 | 10688 | 7246 | | +| 9375 | 10875 | 10774 | 7279 | 20 | +| | 10908 | 11049 | 7427 | | +| 9552 | | | | )regnorts = evitagen ,rekaew = evitisop( tfihS knaR | +| 9703 | 11327 | 11327 | 7441 | | +| 9961 | 11496 | 11496 | 8385 | | +| 10167 | 11570 | 11570 | 8533 | | +| 10610 | 11580 | 11580 | 10688 | | +| 10875 | 11892 | | | | +| 10908 | | 11671 | 10788 | | +| 11496 | 12020 | 11892 | 11049 | | +| 11570 | 12400 | 12400 | 11671 | 10 | +| 11892 | 12702 | 12702 | | | +| | 1 3 3 4 6 | | 11892 | | +| 12020 | | 1 3 0 6 9 | 1 2 7 0 2 | | +| 12702 | 1 3 3 7 1 | 1 3 3 4 6 | 1 3 0 6 9 | | +| 12882 | 1 3 7 9 1 | 1 3 3 7 1 | | | +| 13371 | 1 3 8 1 7 | 1 3 8 1 7 | 1 3 3 4 6 | | +| DI erutaeF 13791 | 1 5 0 0 7 | | 1 3 3 7 1 | | +| 15007 | 1 5 4 6 7 | 1 5 4 6 7 | 1 4 0 3 3 | | +| 15345 | | 1 5 4 7 0 | | | +| 15470 | 1 5 4 7 0 | 1 5 6 8 9 | 1 4 9 8 3 | | +| 15680 | 15680 | 1 5 9 3 1 | 15467 | 0 | +| 15931 | 1 5 6 8 9 | | 1 5 4 7 0 | | +| | 1 5 9 3 1 | 1 6 0 0 2 | 1 5 9 3 1 | | +| 16099 | 1 6 0 9 9 | 1 6 0 9 9 | | | +| 16353 | | 1 6 3 5 3 | 1 5 9 4 7 | | +| 16537 | 1 6 3 5 3 | 1 7 2 6 6 | 1 6 0 0 2 | | +| 17464 | 1 6 5 3 7 | | 1 6 0 9 9 | | +| 17472 | 1 7 2 6 6 | 1 7 4 5 9 | | | +| 18289 | 1 7 4 5 9 | 1 7 4 7 2 | 1 6 3 5 3 | | +| 18372 | | 18189 | 17266 | | +| 18785 | 17472 | | 17472 | 10 | +| 18937 | 18189 | 18372 | | | +| 19623 | 18372 | 18785 | 18189 | | +| | 18785 | 20784 | 18289 | | +| 19676 | 20784 | 20937 | 18372 | | +| 21258 | | | 18785 | | +| 21547 | 20937 | 21022 | | | +| 22441 | 21258 | 21072 | 20937 | | +| 22489 | 22154 | 21258 | 21022 | | +| 22830 | 22277 | 21596 | 21072 | | +| 22953 | 22441 | | | | +| 23117 | | 21961 | 21258 | 20 | +| 23640 | 22743 | 22014 | 21596 | | +| 23652 | 23117 | 22154 | 21961 | | +| | 23502 | 22173 | | | +| 23797 | 23652 | | 22014 | | +| 24032 | | 22277 | 22173 | | +| 24447 | 2 3 7 9 7 | 2 2 7 4 3 | 2 3 1 1 7 | | +| 24506 | 2 4 8 8 2 | 2 3 1 1 7 | | | +| 24882 | 2 5 5 3 9 | 2 3 5 0 2 | 2 3 6 4 8 | | +| 24988 | 2 5 5 7 1 | | 2 3 6 5 2 | | +| 25079 | 2 5 8 3 9 | 2 3 6 4 8 | 2 4 8 8 2 | | +| 25155 | | 2 3 6 5 2 | 2 5 8 3 9 | 30 | +| 25839 | 2 6 0 5 1 | 2 4 8 8 2 | | | +| | 2 6 6 2 1 | 2 5 5 3 9 | 2 5 9 2 0 | | +| 26393 | 26760 | | 26051 | | +| 26453 | 2 7 0 3 5 | 2 5 5 7 1 | 2 6 2 2 4 | | +| 26621 | 2 7 0 5 7 | 2 5 8 3 9 | | | +| 26760 | | 2 5 9 2 0 | 2 6 3 2 4 | | +| 27057 | 2 7 2 8 5 | 2 6 0 5 1 | 2 6 8 7 3 | | +| 28734 | 2 9 1 6 8 | | 2 7 0 5 7 | | +| 28964 | 2 9 3 0 2 | 2 6 2 2 4 | | | +| 29006 | 2 9 5 3 8 | 2 7 0 3 5 | 2 7 1 3 9 | 40 | +| 29168 | | 2 7 0 5 7 | 2 8 1 5 1 | | +| 29302 | 2 9 9 2 4 | | 29039 | | +| | 30109 | 27285 | | | +| 29538 | 30419 | 29168 | 29168 | | +| 30109 | 30615 | 29302 | 29302 | | +| 30615 | 30902 | 29924 | 30123 | | +| 30902 | | | 30574 | | +| 30992 | 31384 | 30419 | | | +| 31384 | 31423 | 30574 | 30661 | | +| 31423 | 31896 | 31896 | 32311 | | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure11: FeaturerankshiftacrossadjacentcheckpointsduringRLonQwen2.5-7B. + +Llama3.1-8b-Instruct-SFT +50 +| 279 | 279 | 567 | 567 | | +| ---- | ------- | ------- | ---- | --- | +| 1000 | 567 | 779 | | | +| 1097 | | | 1000 | | +| | 779 | 1000 | | | +| 1621 | 1 0 0 0 | | 1097 | | +| 1741 | | 1 0 9 7 | 1103 | | +| | 1 1 0 3 | 1 1 0 3 | | | +| 1923 | 1621 | | 1621 | | +| 2190 | | 1621 | | | +| 2691 | 2190 | | 2190 | | +| | 2447 | 2190 | | 40 | +| 3242 | | 2447 | 2691 | | +| 5351 | 3468 | | | | +| | | 2691 | 2856 | | +| 6202 | 5351 | | 3009 | | +| 6531 | 6202 | 3009 | | | +| 6612 | | 3468 | 3468 | | +6612 +| 6869 | 6869 | 6202 | 6202 | | +| ---- | ---- | ---- | ---- | --- | +7682 +| | 7682 | 6612 | 6612 | 30 | +| ----- | ---- | ---- | ---- | --- | +| 7838 | 7838 | 6869 | | | +| 8523 | | | 6869 | | +| 9123 | 8523 | 6973 | | | +| | 9123 | | 6973 | | +| 9671 | | 7682 | 8445 | | +| 10434 | 9309 | 9123 | | | +| 10456 | 9671 | | 9123 | | +9309 +| 10613 | 9787 | | 9309 | | +| ----- | ----- | ---- | ---- | --- | +| 10759 | 10434 | 9671 | | | +| | | 9787 | 9671 | 20 | +| 11288 | 10759 | | | | +11442 11288 9904 9904 )regnorts = evitagen ,rekaew = evitisop( tfihS knaR +| 11879 | | 10434 | | | +| ----- | ----- | ----- | ----- | --- | +| | 11442 | | 10434 | | +| 11955 | 11952 | 10759 | 11288 | | +12133 +| | 12145 | 11288 | 12145 | | +| ----- | ----- | ----- | ----- | --- | +| 12145 | | 11952 | | | +| 12225 | 12225 | | 12849 | | +| 12697 | 12697 | 12145 | | 10 | +13184 +| 12849 | 12849 | 12849 | | | +| ----- | ----- | ----- | ----- | --- | +| 13086 | 13184 | 13184 | 13418 | | +14114 +| 13184 | 13418 | 13418 | | | +| ---------------- | ----- | ----- | ----- | --- | +| 13418 | 13759 | | 14665 | | +| DI erutaeF 13759 | | 14114 | | | +| | 14114 | 14665 | 15079 | | +| 14114 | 14255 | | | | +| 14255 | | 15079 | 15467 | | +| | 14665 | | | 0 | +| 14406 | 15079 | 15467 | 15655 | | +| 14665 | | 15655 | | | +| 15079 | 15655 | | 16296 | | +| | 16296 | 16296 | 17043 | | +15253 +| 15655 | 17309 | 17043 | 17309 | | +| ----- | ----- | ----- | ----- | --- | +| 16296 | 17425 | 17309 | | | +17425 +| 17205 | 17459 | 17425 | | | +| ----- | ----- | ----- | ----- | --- | +| 17309 | | | 17947 | 10 | +| | 18341 | 17459 | | | +| 17411 | 18567 | 17947 | 18341 | | +| 17425 | | | 18567 | | +| 18341 | 18992 | 18567 | | | +| | 19299 | | 19071 | | +| 18439 | | 18992 | | | +| 18496 | 19499 | 19071 | 19299 | | +19821 +| 18567 | | 19299 | 19499 | | +| ----- | ----- | ----- | ----- | --- | +| 19299 | 21141 | 19499 | | | +| 19499 | 21365 | | 19821 | 20 | +20918 +| 19821 | 21720 | | 20918 | | +| ----- | ----- | ----- | ----- | --- | +| 21365 | 21945 | 21141 | 21141 | | +21365 +| 21945 | 22513 | | 21365 | | +| ----- | ----- | ----- | ----- | --- | +| 22513 | 22754 | 21720 | | | +| 22754 | | | 21720 | | +| | 23231 | 21945 | | | +| 23396 | 23396 | 22513 | 21945 | | +24038 +| | 23408 | 22754 | 22513 | 30 | +| --- | ----- | ----- | ----- | --- | +25051 +| 25511 | 25051 | 23231 | 22754 | | +| ----- | --------- | --------- | ----- | --- | +| 26064 | 25511 | 23396 | 23396 | | +| 26389 | 26064 | 23408 | 25511 | | +| 29460 | 26389 | | | | +| 29627 | | 25511 | 26064 | | +| | 29501 | 26064 | | | +| 29728 | 29627 | | 26389 | | +| 29989 | | 26389 | | 40 | +| | 29728 | | 26978 | | +| 30060 | 29989 | 26978 | 27663 | | +| 30151 | | 2 9 5 0 1 | | | +| 30216 | 3 0 0 6 0 | | 27982 | | +| | 3 0 1 5 1 | 2 9 6 2 7 | | | +| 30518 | | | 29627 | | +| 31003 | 30216 | 29728 | | | +| | 30518 | 30151 | 29728 | | +31774 +| 32628 | 31003 | 31003 | 30151 | | +| ----- | ----- | ----- | ----- | --- | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure12: FeaturerankshiftacrossadjacentcheckpointsduringSFTonLlama3.1-8B-Instruct. + +Llama3.1-8b-Instruct-RL +50 +| 7 | 3 | 1 | 1 | | +| --- | --- | --- | --- | --- | +| 8 | 8 | 3 | 5 | | +| 15 | 9 | 5 | 7 | | 17 - - Troubleshooting - -Problem - -Possible cause - -Solution - -Improper operation of the detergent cover - -Add the detergent and rinse aid as indicated by -the manual - -Previous cycle has not finished or has been -interrupted - -Select the proper cycle as introduced by the -manual - -Detergent dispenser -will not shut -Water remains in the -dishwasher - -Dishwasher does not -drain properly - -Drain is clogged - -Drain hose is kinked - -Check the air gap if you have -Make sure the disposer is empty if the -dishwasher is connected to the disposer - -Make sure the drain hose is properly connected -to the sink - -Make sure to use only the automatic dishwasher -detergent - -Suds in the tub - -Improper detergent is used - -Dishwasher leaks - -Black or gray marks -on dishes - -Excessive suds caused by using the improper -detergent - -Make sure to use only the automatic -dishwasher detergent - -Dishwasher is not level - -Aluminum utensils have rubbed against dishes - -Leveling the dishwasher (see the installation -manual) - -Chose a higher cycle -Make sure the inlet water is no less than 120℉ - -Coffee and tea soil - -Using a spot cleaner to remove the soil - -Stained tub interior - -Reddish stain - -Some tomato food will cause this, use of the -rinse only cycle right after loading this will -decrease the staining - -Noises - -Detergent cup opening\ Drain pump sound - -This is normal - -A hard object has entered the wash module When -the object is ground up the, - -. - sound should stop. - -If the noise persists after a complete cycle, -call for service. - -Dishwasher won't fill - -Water valve is turn off -Door latch may not be properly seated - -Check if the water valve is turn on - -Make sure the door is closed - -When some malfunctions come on, the appliance will display error codes to warn you: - -E1 - -E4 - -E8 - -E9 - -Longer inlet time. - -Overflow. - -Faucets is not opened, or water intake is -restricted,or water pressure is too low. - -Some element of dishwasher leaks . - -Failure of orientation of distributary valve. - -Open circuit or break of distributary valve. - -Keep touching a button more than 30 seconds. Water or something on the button. - -If overflow occurs, turn off the main water supply before calling a service. -If there is water in the base pan because of an overfill or small leak, -the water should be removed before restarting the dishwasher. - -18 - - Thank you for purchasing a FORNO product. Please read the entire instruction manual before operating -your new appliance for the first time. Whether you are an occasional user or an expert , it will be beneficial -to familiarize yourself with the safety practices, features, operation and care recommendations of your -appliance. - -Both the model and serial number are listed inside the product. For warranty purposes, you will also need -the date of purchase. -Record this information below for future reference. - +| 18 | 15 | 7 | 8 | | +| --- | --- | --- | --- | --- | +| 21 | 17 | 8 | 16 | | +| 22 | | | 18 | | +| | 18 | 9 | | | +| 23 | 21 | 16 | 21 | | +| 28 | 22 | 17 | 22 | | +| 30 | | | 23 | 40 | +| 31 | 23 | 22 | | | +| 32 | 24 | 23 | 24 | | +| 34 | 28 | | 31 | | +| 35 | | 24 | | | +| | 30 | 30 | 32 | | +| 38 | 31 | 31 | 33 | | +| 45 | 32 | | 34 | | +| 54 | | 32 | | | +| 55 | 34 | 33 | 35 | | +| 58 | 35 | | 37 | 30 | +| 63 | | 34 | 38 | | +| | 37 | 35 | | | +| 67 | 38 | 37 | 39 | | +| 71 | 54 | | 42 | | +| 74 | | 38 | 45 | | +| 86 | 55 | 39 | | | +| 88 | 57 | 45 | 49 | | +| 97 | 5 8 | | 5 2 | | +| 100 | | 4 9 | 5 3 | | +| | 6 3 | 5 5 | | | +| 101 | 6 8 | | 5 4 | 2 0 | +| 106 | | 5 7 | 5 5 | | +107 6 9 6 2 )regnorts = evitagen ,rekaew = evitisop( tfihS knaR +| 108 | 7 1 | 6 3 | 5 7 | | +| -------------- | ----- | ----- | ------ | --- | +| 113 | 7 3 | | 5 8 | | +| 120 | | 6 8 | 6 2 | | +| 123 | 7 4 | 6 9 | | | +| | 8 5 | | 6 3 | | +| 128 | 8 6 | 7 0 | 6 6 | | +| 129 | | 7 1 | 6 8 | | +| 133 | 8 7 | 7 3 | | 1 0 | +| 134 | 8 8 | | 7 0 | | +| 136 | | 7 4 | 7 3 | | +| 138 | 9 1 | 8 2 | 7 4 | | +| 145 | 9 7 | | | | +| | 100 | 85 | 81 | | +| 148 | | 86 | 82 | | +| DI erutaeF 150 | 106 | 87 | 84 | | +| 162 | 113 | | | | +| 179 | 114 | 88 | 85 | | +| 180 | | 91 | 86 | | +| 182 | 115 | | 87 | 0 | +| | 116 | 97 | | | +| 199 | | 100 | 88 | | +| 200 | 120 | 105 | 90 | | +| 203 | 123 | | | | +| 207 | 125 | 106 | 91 | | +| 211 | | 108 | 93 | | +| 214 | 1 2 8 | 1 1 3 | 9 7 | | +| 217 | 1 2 9 | | | | +| | 1 3 3 | 1 1 4 | 1 0 2 | | +| 219 | | 1 1 5 | 1 0 5 | 1 0 | +| 221 | 1 3 4 | | 1 0 6 | | +| 236 | 1 3 6 | 1 1 6 | | | +| 238 | | 1 2 0 | 1 0 8 | | +| 248 | 1 3 8 | 1 2 3 | 1 1 3 | | +| 254 | 1 4 5 | | 1 1 4 | | +| 270 | 1 4 7 | 1 2 5 | | | +| | | 1 2 9 | 1 1 5 | | +| 281 | 1 4 8 | | 1 1 6 | | +| 287 | 1 5 5 | 1 3 3 | 1 2 0 | | +| 298 | 1 6 2 | 1 3 4 | | | +| 323 | | 1 3 6 | 1 2 3 | 2 0 | +| 346 | 1 7 9 | | 1 2 5 | | +| 350 | 1 8 0 | 1 4 5 | | | +| | | 147 | 16 1 0 | | +| 353 | 182 | | 1988 | | +| 374 | 200 | 148 | 2691 | | +| 379 | 203 | 155 | | | +| 382 | | 162 | 3029 | | +| 394 | 211 | | 4727 | | +| 400 | 214 | 179 | 6973 | | +| 408 | 219 | 1610 | | | +| | | | 8063 | 30 | +| 415 | 221 | 1988 | 10930 | | +| 420 | 236 | 2050 | 12982 | | +| 423 | | 3029 | | | +| 425 | 238 | | 13600 | | +| 432 | 248 | 4511 | 14665 | | +| 433 | 254 | 6973 | 15248 | | +| 445 | | 8063 | | | +| | 1610 | | 20405 | | +| 446 | 1988 | 9962 | 21376 | | +| 447 | 2050 | 10930 | | 40 | +| 454 | | | 25051 | | +| 456 | 4511 | 12982 | 25831 | | +| 464 | 6973 | 14665 | 26510 | | +| 486 | | 18247 | | | +| | 9962 | | 26761 | | +| 501 | 10930 | 21376 | 27913 | | +| 505 | 18247 | 27913 | 30507 | | +6973 +| 30591 | 30591 | 30591 | 30591 | | +| ----- | ----- | ----- | ----- | --- | +| 32100 | 32100 | 32100 | 32100 | | +50 +ckpt-2-to-ckpt-1 Rank Shift ckpt-3-to-ckpt-2 Rank Shift ckpt-4-to-ckpt-3 Rank Shift ckpt-5-to-ckpt-4 Rank Shift +Figure13: FeaturerankshiftacrossadjacentcheckpointsduringRLonLlama3.1-8B-Instruct. \ No newline at end of file diff --git a/q135/random_k2/question.json b/q135/random_k2/question.json index 5e960a4bb72b8c58b7c5f751ddf320e5664aba7c..0cd86ce900e678f75e2b2c12d25e5b03d2abb283 100644 --- a/q135/random_k2/question.json +++ b/q135/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "f1f5242528411b262be447e61e2eb10f.pdf", - "web_790f0c62b586634c.pdf", - "web_f8f90e71a8e2a36d.pdf" + "acm_3744614_scientific_factchecking.pdf", + "web_d0752a6a77ae08fd.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q135/random_k2/random_1.md b/q135/random_k2/random_1.md index 926e3cc3fc86ce4552fba63ae8bed9b167e5815e..817dc996b41c5ec14c1f3eee71ff817932a22638 100644 --- a/q135/random_k2/random_1.md +++ b/q135/random_k2/random_1.md @@ -1,224 +1,2076 @@ -PRACTICAL GUIDE FOR -INTERNATIONAL VISITORS +eprints@whiterose.ac.uk +https://eprints.whiterose.ac.uk + +Universities of Leeds, Sheffield and York + +Deposited via The University of Sheffield. + +White Rose Research Online URL for this paper: +https://eprints.whiterose.ac.uk/id/eprint/228448/ + +Version: Published Version + +Proceedings Paper: +Deng, X., Wang, X. and Stevenson, R. (2025) The next phase of scientific fact-checking: +advanced evidence retrieval from complex structured academic papers. In: Zamani, H., +(ed.) ICTIR '25: Proceedings of the 2025 International ACM SIGIR Conference on +Innovative Concepts and Theories in Information Retrieval (ICTIR). ICTIR '25: International +ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval, 18 +Jul 2025, Padua, Italy. ACM. ISBN: 9798400718618. + +https://doi.org/10.1145/3731120.3744614 + +Reuse +This article is distributed under the terms of the Creative Commons Attribution (CC BY) licence. This licence +allows you to distribute, remix, tweak, and build upon the work, even commercially, as long as you credit the +authors for the original work. More information and the full terms of the licence here: +https://creativecommons.org/licenses/ + +Takedown +If you consider content in White Rose Research Online to be in breach of UK law, please notify us by +emailing eprints@whiterose.ac.uk including the URL of the record and the reason for the withdrawal request. + + The Next Phase of Scientific Fact-Checking: Advanced Evidence +Retrieval from Complex Structured Academic Papers + +Xingyu Deng +xdeng37@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Xi Wang +xi.wang@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Mark Stevenson +mark.stevenson@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Abstract +Scientific fact-checking aims to determine the veracity of scien- +tific claims by retrieving and analysing evidence from research +literature. The problem is inherently more complex than general +fact-checking since it must accommodate the evolving nature of +scientific knowledge, the structural complexity of academic litera- +ture and the challenges posed by long-form, multimodal scientific +expression. However, existing approaches focus on simplified ver- +sions of the problem based on small-scale datasets consisting of +abstracts rather than full papers, thereby avoiding the distinct chal- +lenges associated with processing complete documents. This paper +examines the limitations of current scientific fact-checking systems +and reveals the many potential features and resources that could be +exploited to advance their performance. It identifies key research +challenges within evidence retrieval, including (1) evidence-driven +retrieval that addresses semantic limitations and topic imbalance +(2) time-aware evidence retrieval with citation tracking to mit- +igate outdated information, (3) structured document parsing to +leverage long-range context, (4) handling complex scientific expres- +sions, including tables, figures, and domain-specific terminology +and (5) assessing the credibility of scientific literature. Preliminary +experiments were conducted to substantiate these challenges and +identify potential solutions. This perspective paper aims to advance +scientific fact-checking with a specialised IR system tailored for +real-world applications. + +CCS Concepts +· Information systems → Specialized information retrieval. + +Keywords +Evidence retrieval, Scientific fact-checking + +ACM Reference Format: +Xingyu Deng, Xi Wang, and Mark Stevenson. 2025. The Next Phase of Scien- +tific Fact-Checking: Advanced Evidence Retrieval from Complex Structured +Academic Papers. In Proceedings of the 2025 International ACM SIGIR Con- +ference on Innovative Concepts and Theories in Information Retrieval (ICTIR) +(ICTIR ’25), July 18, 2025, Padua, Italy. ACM, New York, NY, USA, 13 pages. +https://doi.org/10.1145/3731120.3744614 + +This work is licensed under a Creative Commons Attribution 4.0 International License. +ICTIR ’25, Padua, Italy +© 2025 Copyright held by the owner/author(s). +ACM ISBN 979-8-4007-1861-8/2025/07 +https://doi.org/10.1145/3731120.3744614 + +1 Introduction +Fact-checking aims to assess the veracity of factual claims based +on credible evidence [37, 116] and serves as a crucial safeguard +for mitigating misinformation. Scientific fact-checking is a spe- +cialised variant of this task, grounded in scientific knowledge, with +the objective of combating misinformation that affects the pub- +lic, helping researchers in knowledge discovery and assisting in- +dividuals in understanding scientific advancements [97]. This is +particularly important given the rapid emergence of new scien- +tific findings, where both professionals and the public must assess +the credibility of information. A prominent case occurred during +the COVID-19 pandemic, in which politically motivated misinfor- +mationÐranging from inflated infection statistics to unsupported +treatmentsÐcirculated extensively, eroding public trust and endan- +gering health communication [58]. However, existing approaches +to scientific fact-checking remain limited, primarily relying on the +retrieval of evidence from relatively simple and small-scale sources +[16, 49, 62, 72, 75, 100, 101, 104]. For example, SciFact-Open [101], +the largest available dataset for scientific fact-checking, contains +500,000 documents ś substantially smaller than PubMed, which con- +tains over 37 million biomedical publications. In addition, SciFact- +Open consists only of abstracts, rather than full-text papers, thereby +excluding critical structural and citation information, ignoring long- +range context and scientific expression conveyed through tables +and figures. These design simplifications may hinder the applicabil- +ity of current approaches in real-world settings, where scientific +evidence is embedded in long and structurally complex documents +with multimodal content. + +Fact-checking is a knowledge-intensive task, where the verifica- +tion process relies on sourcing evidence from a reliable upstream +Information Retrieval (IR) system. Emerging findings indicate the +value of effective retrieval in improving fact-checking systems. +For example, introducing even a small amount of noise into evi- +dence can significantly degrade fact-checking performance [76]. +Recent Retrieval-Augmented-Generation (RAG) techniques have +been widely used for fact-checking [28, 47, 63, 70, 80, 84, 94], where +retrieval models are fine-tuned to identify high-quality evidence +for claim verification. These observations underscore the critical +role of robust evidence retrieval, as an ideal IR system for fact- +checking should rank all relevant evidence at the top while filtering +out non-evidential noise. Ensuring retrieval robustness is crucial to +maintaining sufficient yet relevant evidence, which is essential for +improving scientific fact-checking accuracy. + +A major trend in fact-checking research is to consider realistic set- +tings that employ rich, diverse and timely evidence sources, as seen +in FEVER (using Wikipedia) [90] and AVeriTeC (using web-wide +resources) [78]. In document level evidence retrieval, current + +436 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +general fact-checking systems over-rely on commercial search APIs, +which do not consider the specific requirements of fact-checking +[77, 91, 116]. Such reliance on commercial search APIs ś with lim- +ited adaptability ś has left document retrieval methodologies under- +explored in fact-checking, especially for domain-specific corpora +such as scientific fact-checking. Current scientific fact-checking sys- +tems primarily employ off-the-shelf IR methods [97], such as lexical +matching and semantic relevance ranking, which do not scale effec- +tively for large-scale scientific corpora. In addition, the distribution +of relevant evidence across scientific topics is highly imbalanced, +which degrades both retrieval effectiveness and efficiency, espe- +cially for claims with scarce supporting literature. SciFact-Open +[101], which extends the original SciFact dataset [100] for large- +scale evaluation, illustrates this issue: verification performance on +SciFact-Open drops by at least 15 F1 points for all well-performed +fact-checking systems developed in SciFact [101]. While increasing +corpus size enhances evidence diversity, it also amplifies retrieval +noise, reducing efficiency in both retrieval and verification. Beyond +that, high semantic relevance does not guarantee high evidential +relevance, and irrelevant yet semantically similar documents can +introduce noise into downstream verification [117]. These chal- +lenges underscore the necessity of developing tailored document +retrieval systems specifically designed for scientific fact-checking, +as effective retrieval is a prerequisite for accurate claim verification. +Beyond document-level evidence retrieval, within-document +evidence retrieval is also essential for processing complex sci- +entific literature. Scientific papers, unlike general fact-checking +documents, are long, structured, domain-specific and involve addi- +tional metadata. As scientific fact-checking evolves from abstract- +based to full-paper retrieval, retrieval models must account for +metadata (e.g., publish date, citations) and complex structured data +format (e.g., charts, tables, figures). This necessitates the adaptation +of verification models such as SciBERT [11] for domain-specific +terminologies [100], Longformer [12] for long-range dependen- +cies [102] and TAPAS [38] for tabular data verification [3]. Sci- +entific expressions in academic papers are highly structured and +contextually interdependent, where textual content, tabular data, +and figures mutually reinforce the conveyed information. How- +ever, existing scientific fact-checking systems primarily operate +at the abstract level, adopting methodologies similar to general +fact-checking [49, 50, 68, 72, 75, 97, 100, 102, 111, 119], albeit incor- +porating domain-specific models such as BioSentVec [21] and SciB- +ERT [11]. The development of public full-paper datasets aligns with +the requirement of real-world scientific fact-checking systems for +effective verification. This highlights the urgent need for retrieval +and verification methodologies that can leverage entire scientific +documents. Accordingly, within-document evidence retrieval and +its integration into verification pipelines should be explored to fully +unlock the potential of scientific literature for fact-checking. + +This perspective paper presents a comprehensive examination of +the challenges associated with evidence retrieval in scientific fact- +checking, highlighting challenges that are not typically faced +within general fact-checking, leading to the need for specialised +retrieval and verification strategies. We advocate for proactive re- +search efforts to develop scalable methodologies while addressing +the limitations of current datasets. We structure our discussion into +two parts following a typical fact-checking pipeline: Sections 2ś3 + +explore document-level evidence retrieval while Sections 4ś8 ex- +plore within-document evidence retrieval in scientific publication +for scientific fact-checking. Each of the following sections identi- +fies a research challenge followed by a tentative and illustrative +research direction (RD). + +2 Beyond Semantics +Evidence retrieval for fact checking is closely related to traditional +document retrieval techniques, which typically focus on retrieving +documents that are semantically similar to a query or contain match- +ing keywords. While this approach is effective in many scenarios, +it often fails to address the ultimate objective of fact-checking ś +successful claim verification. Evidence retrieval that relies solely on +semantic similarity may prioritise irrelevant or low-context infor- +mation, introducing noise into the subsequent verification process. +Recent IR studies [67, 92] show that semantic relevance alone may +not ensure utility in knowledge-intensive NLP tasks under the RAG +framework, suggesting the importance of utility-aware retrieval +strategies. To improve the verification utility of evidence retrieval +systems, techniques such as fine-tuning, joint optimisation, and +learning from verification feedback have been developed. These +approaches leverage relevance labels derived from annotated gold +evidence [40, 68, 117, 119, 120]. Although graded relevance has +been extensively explored in general IR, current evidence retrieval +systems for fact-checking often oversimplify relevance as binary, +failing to differentiate between fully non-evidential and partially +relevant evidence. This coarse-grained labelling scheme fails to +differentiate between completely non-evidential documents and +partially relevant (plausible) evidence. Negative examples and ran- +domly retrieved examples are equally treated as 0, despite exhibiting +varying degrees of evidential support. We argue that evidence re- +trieval should distinguish between non-evidential information and +plausible evidence, enhancing the model’s ability to identify previ- +ously unobserved but potentially useful evidence within large-scale +corpora. + +Developing an IR system that can effectively differentiate be- +tween evidential and non-evidential information requires access +to fine-grained relevance labels during training. However, manu- +ally constructing negative samples is both complex and resource- +intensive due to the vast number of unlabelled documents and +sentences that lack explicit pairing with given claims. Furthermore, +assessing the degree of evidential support for a claim within unla- +belled documents is inherently challenging. To validate the impact +of fine-grained evidential relevance, beyond semantic relevance, we +carry out preliminary experiments which explore the use of down- +stream verification feedback to capture different levels of evidential +values. + +Experiment Overview. The experiment investigates whether +combining verification feedback with semantic relevance improves +the performance of document evidence retrieval. Figure 1 presents +the pipeline, where probabilities from downstream verification +serve as feedback. Specifically, we integrate two components: 1) the +semantic relevance score, computed using an off-the-shelf reranker +model, and (2) the verification success feedback score, derived from +a fine-tuned verifier model, indicating the degree to which a docu- +ment is evidential for a given claim. + +437 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +Table 1: Retrieval result on SciFact-Open and Check-COVID + +SciFact-Open +R@50 R@20 R@10 +BM25 +45.22 +54.78 +66.09 +71.09 +79.13 +88.91 +monoT5-3B +71.30 +81.95 +91.96 ++Verification +Check-COVID R@50 R@20 R@10 +75.02 +87.91 +89.49 +95.84 +91.48 +96.13 + +BM25 +monoT5-3B ++Verification + +81.96 +93.16 +94.55 + +R@5 +38.04 +57.17 +62.61 +R@5 +67.59 +82.06 +84.04 + +R@3 +30.87 +48.26 +52.83 +R@3 +61.35 +74.93 +77.80 + +R@1 +20.22 +31.09 +32.17 +R@1 +46.18 +58.28 +61.84 + +Table 2: Evidence positions in the retrieved list. We select an ex- +ample from the SciFact-Open dataset. Claim: Female carriers of the +Apolipoprotein E4 (APOE4) allele have a reduced risk for Alzheimer’s +disease. Gold Evidence: [E1,E2,E3,E4,E5] + +reranker model E1 +BM25 +monot5-3B ++Verification + +838th +1st +1st + +E2 +141th +8th +3rd + +E3 +7th +16th +5th + +E4 +163th +2nd +2nd + +E5 +67th +302nd +27th + +Datasets. We conduct our evaluation using datasets including: +(1) SciFact [100]: A corpus of 5,183 abstracts from scientific arti- +cles, with 809/300/300 samples for train, validation and test sets. +The test set is not publicly accessible. (2) SciFact-Open [101]: An +extended version of SciFact with 500K abstracts, re-annotating evi- +dence documents for 279 claims from the original SciFact test set. (3) +Check-COVID [104]: A COVID-19-specific fact-checking dataset, +containing 347 abstracts from CORD-19 journal articles and 1,504 +expert annotated news-related claims. + +MultiVerS is trained on the SciFact train set to create a verifier +model that provides verification feedback. Since the SciFact test +set is inaccessible, we evaluate document evidence retrieval on +SciFact-Open and full Check-COVID. + +Results. The integration of verification feedback consistently en- +hanced document evidence retrieval, as +Verification outperformed +monoT5-3B across nearly all cut-off thresholds in both SciFact-Open +and Check-COVID (Table 1). The improvements are particularly ev- +ident at lower cut-offs, where retrieving the most relevant evidence +is crucial. In SciFact-Open, Recall@5 and Recall@3 increased from +57.17% to 62.61% and from 48.26% to 52.83%, respectively. Similarly, +in Check-COVID, these metrics improved from 82.06% to 84.04% +and from 74.93% to 77.80%. These improvements are particularly +meaningful given that the average number of gold evidence doc- +uments per claim is 1 in Check-COVID and approximately 2 in +SciFact-Open. To illustrate this improvement, we conducted a case +study examining how different retrieval methods ranked specific +evidence documents (Table 2). Compared to monoT5-3B, +Verifi- +cation successfully elevated the ranks of E2, E3 and E5, retrieving +three additional pieces of evidence within the top 5 results. Notably, +E5, ranked 302nd by monoT5-3B, was effectively rescued by adding +verification feedback in +Verification, demonstrating the value of +integrating verification-informed retrieval signals. -IN CHENGDU +These findings highlight a promising research direction: shift- +ing from semantic-only retrieval towards evidence-aware retrieval, +where retrieval models explicitly account for evidential value. Our +results suggest that leveraging well-performing verification models +can help refine retrieval systems by distinguishing between purely - Contents +Claim -I. Guide to Payment Services in China ....................................................... 1 +Retr ieved +documents -II. SIM Card .................................................................................................2 +0.86 -III. Chengdu Public Transit & Key Hubs ................................................... 2 +0.92 -IV. Airport-to-Venue Transportation ........................................................... 3 +0.31 -V. Daily Transportation Options ................................................................. 6 +0.07 -VI. Important Notes .....................................................................................8 +0.04 - I. Guide to Payment Services in China +0.02 -To ensure smooth transactions during your visit, all international attendees are advised to -review the official payment guidelines from the People's Bank of China: +0.01 -http://www.pbc.gov.cn/en/3688110/3688172/5188125/5274061/index.html +r elevance scor es -The page as shown in the picture +Document +Rer anker -Chengdu Metro Payment Options +monoT5-3B -From July 28th, 2025, -international visitors may use UnionPay chip-enabled cards, -internationally issued Visa, Mastercard, and American Express chip-enabled cards to -take Chengdu Metro. +Top-k +documents -Simply tap your card at fare gates for metro lines. +Ver ification -Deductions follow standard pricing rules. +M ultiVer S -Cards can also be used at Ticket Vending Machines (TVMs) and Day Pass Vending -Machines. +Suppor t -1 +0.7 - II. SIM Card +Refute -Alipay provides an eSIM feature that enables users to activate a virtual SIM card -directly through the app. +0.1 -You can also apply at any China Telecom, China Mobile, or China Unicom store (including -outlets at the airport). +ver ification +feedback -Main mobile phone operators in China +NEI -Card Application Process: +0.2 -1. Visit a store to apply for a SIM card and present your identification documents -(passport/visa, etc.). +0.8 -2. Select your preferred phone number and service plan. +Figure 1: Pipeline of experiment -3. Receive your activated SIM card. The entire process is efficient and typically completes -quickly. +Approaches. monoT5-3B [64] has demonstrated strong perfor- +mance as a reranker for SciFact, as evidenced by its widespread +use as a strong baseline in studies [55, 86] on the BEIR benchmark +[89]. It also demonstrated state-of-the-art performance in evidence +retrieval for verification [68, 97, 102]. The model assigns a predicted +score, 𝑠𝑟 +𝑐,𝑑 , representing the semantic relevance for a document 𝑑 +to a claim 𝑐 as defined in Equation 1. -Note: Before leaving, ensure the staff assists with card activation and verify that the card -functions properly. +(1) -III. Chengdu Public Transit & Key Hubs +𝑐,𝑑 ∈ (0, 1) -Chengdu offers a convenient public transportation system, including: +𝑓 (𝑐, 𝑑) → 𝑠𝑟 -Two major airports: Chengdu Shuangliu International Airport & Chengdu Tianfu -International Airport. +𝑐,𝑑, 𝑠𝑟 +MultiVerS is the best-performing verifier model on SciFact [97, +102]. We reproduced this model using the official implementation1 +while adjusting the negative sampling parameter from 20 to 5 to +avoid over-fitted verification feedback. The model predicts the veri- +fication outcome as per the calculated probabilities for a document +𝑑 either supporting 𝑝𝑟 +𝑐,𝑑 or providing insufficient in- +formation 𝑝𝑛 -Multiple high-speed rail stations, such as Chengdu East Railway Station (primary hub -that connects 80%+ high-speed routes), Chengdu South Railway Station (regional -services that link to Leshan City, Mianyang City, Yibin City...), and Chengdu West -Railway Station (serves western routes, e.g., Pujiang, Dujiangyan). +𝑐,𝑑 , refuting 𝑝𝑠 +𝑐,𝑑 , relative to a claim 𝑐, as follows: +𝑐,𝑑, 𝑝𝑠 -An extensive metro and bus network, ensuring efficient and accessible travel throughout -the city. +𝑉 (𝑐, 𝑑) = 𝑎𝑟𝑔𝑚𝑎𝑥 (𝑝𝑟 -2 +𝑐,𝑑, 𝑝𝑛 - IV. Airport-to-Venue Transportation +𝑐,𝑑 ) -From Chengdu Tianfu International Airport (TFU) to Chengdu Tianfu International -Convention Center +(2) -By Metro (Recommended): ++Verification (Ideal Reranker Model) combines semantic relevance +and verification feedback to refine evidential retrieval. The final +retrieval score is calculated by summing the semantic score (𝑠𝑟 +𝑐,𝑑 ) +and the verification probabilities (𝑝𝑟 +𝑐,𝑑 ), followed by a nor- +malisation step to ensure the score remains within (0,1), formulated +as: -Route: Take Line 18 from T1 & T2 of Tianfu International Airport → Exit at Western China -Int'l Expo City Station (Exit E). The venue is a 3-minute walk (180 meters) from Exit E. +𝑐,𝑑 and 𝑝𝑛 -3 +𝑠𝑟 +𝑣 +𝑐,𝑑 - Operating Hours: += 1/2 ∗ (𝑠𝑟 -Duration: 50 Minutes +𝑐,𝑑 + 𝑝𝑟 +This formulation ensures that documents contributing to support +or refute labels receive higher retrieval scores, enhancing the evi- +dential quality of retrieved documents. -Payment Method: Bank Cards/ Alipay/ WeChat Pay/ Cash +𝑐,𝑑 ) ∈ (0, 1) -By Taxi: +𝑐,𝑑 + 𝑝𝑠 -Boarding Location: Level 1, Terminal 1&2 +(3) -Availability: Available around the clock +To evaluate retrieval effectiveness, we use Recall@k (R@k), a +common evaluation approach that measures the proportion of rele- +vant evidence successfully retrieved within the top 𝑘 results. -Fare: CNY100-120 | Duration: 45-60 min (traffic-dependent) +1https://github.com/dwadden/multivers -From Chengdu Shuangliu International Airport -International Convention Center +438 -(CTU) +r -to Chengdu Tianfu +e -By Metro (Recommended): +r -Route: Take Line 19 from East of Terminal 2 of Shuangliu International Airport Station→ -Disembark at Tianfu Commercial District Station (3-minute indoor walk required) / Transfer to -Line 6 at Tianfu Business District Station → Exit at Western China Int'l Expo City Station -(Exit E). The venue is a 3-minute walk (180 meters) from Exit E. +a -4 +n - Operating Hours: +k -5 - - Duration: 40 Minutes - -Payment Method: Bank Cards/ Alipay/ WeChat Pay/ Cash - -By Taxi: - -Boarding Location: - -T1’s Pick-up Points: Suspend the service. - -T2’s Pick-up Points: Near the crosswalk of the arrival level. - -Availability: Available around the clock - -Fare: CNY 60-80 | Duration: 40-50 min (traffic-dependent) - -Airport shuttle buses (from both Tianfu and Shuangliu Airports): - -Operate on fixed routes with stops not in close proximity to the convention venue. If you need -transportation to other destinations, please check the official airport shuttle schedule upon -arrival. - -V. Daily Transportation Options - -Metro - -Transportation Option: Ticket - -The ticket price ranges from CNY 2 to CNY 10 according to the traveling distance. The fare -complies with the rules below: - -Staffed Assistance - -Self-Service Machines - -6 - - Notes: -Tickets are available at staffed counters or self-service machines - -Daily passes (1-day: CNY 20 | 3-day: CNY 50 | 5-day: CNY 70) - -Payment Methods: Bank Cards/Cash (CNY)/WeChat Pay/Alipay. - -Transportation Option: QR Code Access - -Via Alipay: Go to Transport → Select Chengdu Metro → Complete ID verification → Scan -QR code at gates - -Transportation Option: Transport Card - -Purchase a Tianfu Tong Card at staffed service counters for metro/bus use. - -Bus - -Transportation Option: Cash (exact change recommended) - -Transportation Option: Alipay QR Code- The same as metro procedure, requires QR code -scanning “Transport → Select Chengdu Bus” - -Transportation Option: Tianfu Tong Transport Card - -Taxi - -Payment Method: Cash (CNY) or mobile payment (WeChat/ Alipay QR code scanning) - -Please note: Credit cards are not accepted in local taxis. - -7 - - Ride-hailing service: Ride-hailing apps like DiDi are accessible. - -VI. Important Notes - -Card Payments - -Visa/Mastercard accepted at most venue-adjacent restaurants and shops. - -Power Banks - -For international passengers entering China and taking domestic connecting flights to -Chengdu: - -Power banks without CCC certification are strictly prohibited as per China's aviation safety -regulations. - -Note: This regulation does not apply to: - -International passengers arriving in Chengdu as their final destination. - -Direct international transit passengers (remaining in the sterile area). - -8 + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +semantic relevance and plausible evidential relevance among unan- +notated documents. Furthermore, to continually improve evidence- +aware retrieval, we propose the development of tailored IR systems +capable of identifying evidential information, thereby enhancing +evidence retrieval for scientific fact-checking. + +RD.1. Benchmark tailored IR system for fact-checking + +The preliminary study presented in this work outlined a framework +to enhance evidence retrieval beyond only semantic relevance. To +overcome the limitations of existing IR systems in scientific fact- +checking scenarios, it is imperative to develop specialised IR sys- +tems capable of handling the specific challenges of verification tasks. +However, training an IR system on a single fact-checking dataset +risks poor generalizability and potential overfitting, particularly due +to data imbalance, a common issue in the relatively small datasets +characteristic of scientific fact-checking [97, 116]. Furthermore, +poor verification performance deteriorates retrieval accuracy, cre- +ating a vicious feedback loop that further degrades overall system +effectiveness. A multi-pronged strategy could mitigate these chal- +lenges by pooling verification signals from various high-performing +verifier models, leveraging large-scale datasets such as FEVER [90] +to improve training robustness, and providing a shared retrieval +checkpoint enable subsequent studies to fine-tune the model for +specific scenarios or datasets while reducing training cost. Recent +work [51, 74] has explored unified retrieval models for knowledge- +intensive NLP tasks, focusing on retrieval quality and downstream +task utility [73, 115], including question answering (QA) and fact- +checking. Similarly, we propose a verification-driven IR system for +evidence retrieval, which explicitly incorporates evidential infor- +matics. This approach follows a two-step training paradigm: general +pre-training on large, diverse datasets followed by domain-specific +fine-tuning. This approach balances scalability and domain speci- +ficity, ensuring IR models are both robust across different contexts +and highly effective in targeted fact-checking applications. Addi- +tionally, a corresponding benchmark should employ a diverse set +of evaluation metrics beyond for the fact-checking task to ensure +comprehensive assessment of performance within fact-checking +[6]. These metrics could include verification accuracy, reflecting the +downstream utility of retrieved evidence; decision latency, measur- +ing the computational efficiency of retrieval models; and robustness +to real-world conditions such as noisy data and incomplete evidence, +to improve system resilience. + +Integrating verification feedback into evidence retrieval improves +relevance assessment beyond binary labels, enhancing retrieval per- +formance. Future research should focus on developing a benchmark +IR system tailored for fact-checking, incorporating fine-grained +relevance labels and verification-driven retrieval models. A scalable +pre-training and fine-tuning approach has the potential to improve +retrieval robustness and generalizability thereby producing more +accurate and efficient fact-checking systems. + +3 Imbalanced resources of scientific topics +In existing general fact-checking datasets, such as FEVER which is +based on Wikipedia, the distribution of gold evidence per claim is +relatively even and sufficient. However, a significant imbalance of +evidence is observed in scientific fact-checking. While the SciFact + +Table 3: Sufficient-evidence claim and none-evidence claim exam- +ples in SciFact-Open. ‘Evidence’ is the number of evidence in SciFact- +Open corpus and ‘Entities’ is the number of entities by searching +bold-keyword in PubMed. + +Claim +Obesity is determined in part by +genetic factors. +LRBA controls CTLA - 4 expression. + +Evidence Entities + +24 + +0 + +499k + +0.27k + +corpus (~5K documents) maintains a relatively balanced number of +evidence documents per claim, this balance was disrupted when the +dataset was expanded to create SciFact-Open (~500K documents). +In this larger corpus, the majority of claims have none or only +one piece of supporting evidence while others have over. Claims +related to less-researched topics are generally associated with fewer +scientific publications, as illustrated by the examples in Table 3. + +However, most fact-checking systems do not explicitly account +for evidence imbalance. A common approach is to use a fixed re- +trieval cut-off (i.e., selecting a predefined number of top-ranked +documents for verification). One of the most inefficient approaches +is setting the cut-off equal to the maximum number of evidence +per claim in the dataset, ensuring that all possible evidence is re- +trieved. This heuristic has not previously caused major issues since +general fact-checking datasets contain a relatively balanced num- +ber of supporting documents per claim. However, the imbalance +in SciFact-Open suggests that the simple approach may not be +suitable for open-domain scientific fact-checking with two major +drawbacks: + +(1) Inefficiency. Although the maximum number of gold evi- +dence documents in the SciFact-Open dataset is 24, less than one +third of claims have more than two. Using the maximum number as +a cut-off would be inefficient due to the large number of documents +that would have to be processed by the verifier. + +(2) Inaccuracy. Introducing irrelevant evidence into downstream +verification degrades fact-checking performance, whether the noise +is semantically related or completely random [76] (as discussed in +Section 2). + +To address these challenges in the current and future studies of +scientific fact-checking, we proposed a research direction based +on flexible cut-off strategy for retrieving evidence based on claim +characteristics. + +RD.2. Flexible cut-off for Retrieved Evidence + +Ranked List Truncation (RLT) refers to the task of selecting an opti- +mal prefix of a ranked list of retrieved documents, with the goal of +balancing retrieval effectiveness and efficiency. Prior work explores +both heuristic and learned approaches, using either relevance labels +or features derived from score distributions to determine the cut-off +point [10, 52, 59, 61, 103, 109]. A related line of work is stopping +methods in technology-assisted review (TAR) [14, 15, 85], which +aim to retrieve as much relevant information as possible while min- +imising the effort spent on examining irrelevant documents. Both +approaches aim to optimise an expected metric over candidate cut +positions, typically using metrics such as 𝐹 1@𝑘 or 𝑟𝑒𝑐𝑎𝑙𝑙@𝑘. The +datasets used in these prior studies on RLT and stopping methods, +such as CLEF and TREC [25ś27, 33, 43ś45], exhibit imbalance but + +439 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +typically contain enough relevant items per query to support recall- +based supervision and evaluation. This dependence on sufficient +relevance labels becomes problematic in fact-checking scenarios, +where gold evidence documents are typically rare, making both +recall-based stopping and supervised RLT approaches unsuited to +this problem. + +To explore whether relevance score distributions indicate evi- +dence sufficiency, we compare well-studied and less-studied claims +from SciFact-Open. Following Wadden et. al. [101], claims with +four or more gold evidence documents are considered to be well- +studied and those with none to be less-studied. Using monoT5-3B, +we compute several statistics over the ranked document scores, +including first-document relevance, average and total scores, score +decay, and the initial-to-final score ratio. Table 4 shows that well- +studied claims tend to have higher top-ranked scores and sharper +decay patterns, suggesting that relevance distributions may serve +as indicators of evidence sufficiency. + +Table 4: Statistical analysis of average relevance scores for +well-studied and less-studied claims. ‘I/F’ ratio is Initial-to- +Final ratio. ‘Exp k’ denotes the exponential decay factor k. + +Metric +Less- +Well- + +1st Doc Mean +0.502 +0.741 + +0.941 +0.995 + +Sum I/F ratio Exp k +1.544 +3.514 +25.097 +0.651 +1.963 +37.052 + +Based on these findings, one possible direction is to leverage +existing techniques to estimate whether a claim is less-studied or +well-studied. A prediction module could utilise statistical features +of relevance distribution such as those presented in Table 4. In addi- +tion, metadata such as retrieved entity counts in PubMed (Table 3) +can serve as auxiliary signals to refine the prediction. Claims pre- +dicted as less-studied ś e.g., with low total relevance or steep score +decay ś may be assigned smaller cut-offs to reduce verification cost, +while RLT and stopping techniques could be applied to well-studied +claims where concentrated high scores suggest richer evidence. + +While this naive strategy relies on heuristic features, it does not +explicitly optimise verification performance. To address this, future +approaches could explore learning a cut-off policy using feedback +from the verification stage. Specifically, truncation points may be +selected based on reward signals, such as whether the claim is cor- +rectly verified or the confidence of the verifier. This would bypass +the need for relevance-labelled supervision, which is often infeasi- +ble in scientific fact-checking due to sparse annotations. Inspired +by prior RLT and stopping method work, such a learned policy +could optimise both efficiency and factual accuracy by aligning +truncation decisions with downstream verification performance. + +4 Time and Citation +General fact-checking evidence corpus such as Wikipedia and fact- +checking websites, often lack sentence-level evidence timestamps, +making it difficult to determine the original publish time of sentence +evidence in verification and hindering the development of time- +aware retrieval methods. Timeliness is important in fact-checking, +but in science, evolving evidence makes outdated studies particu- +larly problematic. For instance, early COVID-19 treatment studies + +Table 5: Results of health QA task considering the different +thresholds of the published time of literature [99] + +Year +≥2020 +≥2018 +≥2015 +≥2010 +≥2005 +≥2000 +≥1990 +≥1980 + +Precision Recall + +59.7 +59.6 +61.1 +63.4 +68.1 +66.1 +65.6 +64.2 + +60.3 +58.0 +56.0 +55.6 +56.5 +56.8 +55.4 +54.7 + +F1 score +58.7 +57.9 +53.9 +52.8 +52.0 +51.8 +51.3 +50.0 + +were later refuted, and outdated evidence may lead to harmful de- +cisions. This section discusses whether scientific publications are +more suitable for time-aware fact-checking and explores possible +ways to leverage their inherent temporal characteristics. + +4.1 Timeliness of evidence +Unlike general fact-checking, where historical and static facts re- +main unchanged, scientific knowledge continuously evolves. This +fundamental difference necessitates time-aware retrieval in scien- +tific fact-checking to ensure that retrieved evidence remain valid +and reflective of the latest scientific consensus. A time-sensitive +retrieval mechanism should prioritise recent publications to en- +sure that fact-checking systems incorporate the most up-to-date +methodologies and factual updates. This is especially crucial in +fields like healthcare, where relying on outdated information could +lead to misleading conclusions or incorrect decisions. For example, +during a rapidly evolving pandemic, a medical treatment initially +considered effective might later be deemed unreliable. This sec- +tion discusses the challenges and opportunities of integrating the +evidence timestamp into scientific fact-checking. + +Scientific fact-checking aims to find evidential information in +the literature to verify a claim. Intuitively, considering outdated +literature negatively affects verification. Research in healthcare +question answering (QA) has demonstrated that time-aware re- +trieval improves system performance [99], as shown in Table 5. The +F1 score of the healthcare QA system improves as the publication +year of evidence documents becomes more recent. By extension, +scientific fact-checking on a large-scale corpus may also suffer from +incorporating outdated evidence. These findings highlight the need +for time-aware filtering in scientific fact-checking systems to en- +hance reliability. Fact-checking datasets could explicitly incorporate +timestamps as metadata to facilitate research into temporal rele- +vance in retrieval and verification. A general fact-checking study +[9] collected the ‘timestamp of last update’ of claims and evidential +documents, allowing later research [8] to explore the impact of +temporal data on verification. This study found that a time-aware +system achieved a 15% improvement in macro F1 score, underscor- +ing the importance of temporal information. However, this work +was limited by: (1) focusing on verification only, where evidence +documents had already been retrieved in a separate initial step, +without ensuring that retrieval prioritised high-quality and tem- +porally relevant evidence, and (2) the use of fragmented evidence ś + +440 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +the general fact-checking datasets usually only contain small snip- +pets of web documents, omitting many important time expressions +present in full texts. + +Given these findings, it is essential to develop a fact-checking +system that explicitly incorporates temporal awareness across both +retrieval and verification stages to meet real-world applications. + +RD.3. Time-Aware Retrieval and Verification + +Traditional fact-checking approaches often handle conflicting evi- +dence for a single claim by assigning neutral veracity labels, such +as “mixture," “unproven," or “not enough information" [37, 97]. +However, conflicting evidence often arises due to outdated stud- +ies included in the retrieval process, which introduces noise and +adversely affects prediction accuracy [99]. This issue has been ob- +served in healthcare QA systems, where outdated evidence degrades +performance [99]. To address this issue, we propose a time-aware +approach that incorporates temporal information into both retrieval +and verification stages: + +1.Retrieval Stage: Outdated evidence should be filtered or de- +prioritised during retrieval to ensure that the retrieved evidence +set is temporally aligned with the latest scientific findings. + +2.Verification Stage: After filtering by time-aware retrieval, all +retrieved evidence should be considered but with differentiated +weighting based on temporal relevance. Recent evidence should be +prioritised through higher weights, while older evidence should +serve as supplementary or contextual information rather than pri- +mary evidence. + +By processing outdated evidence differently across retrieval and +verification components, this direction explores how temporal in- +formation can reduce noise and improve the reliability of scientific +fact-checking systems. + +4.2 Indirect evidence through citation +General fact-checking faces a number of challenges when attempt- +ing to determine the timeliness of evidence: (1) Lack of publication +timestamp [8]. Many fact-checking sources, such as Wikipedia and +fact-checking websites, do not provide precise publication dates +for individual sentences or paragraphs. Instead, they record only +the last edited timestamp, which does not accurately reflect when a +fact was first published. (2) Tracking the origin of evidence. Evidence +is often copied or paraphrased across multiple sources, making it +difficult to determine the original publication date of a statement. +(3) Search engine bias in retrieval. Pre-established fact-checking +datasets retrieve evidence using top-k search engine results, where +ranking mechanisms may prioritise recent documents due to time- +aware ranking biases. This can misrepresent the actual chronology +of claims and lead to fragmented evidence, making time extraction +unreliable. + +While these challenges also affect scientific fact-checking, they +can be naturally mitigated by leveraging full-text papers rather than +abstract-only sources. The main advantages of using full-text papers +include: (1) Explicit publication metadata: Each piece of literature +has a clearly defined publication date, ensuring accurate temporal +tracking. (2) Citation tracing for indirect evidence: Mandatory cita- +tion rules in academic publications facilitate source tracing, even +when statements are referenced indirectly. (3) Structured nature + +of academic papers: provides an indication of the origin of state- +ments. For example, evidence in the background section typically +references prior studies, whereas those in the abstract or results +sections represent findings from the current publication. Given +these inherent advantages, incorporating timestamps as metadata +offers a promising research direction for full-paper-based scien- +tific fact-checking. Exploring the temporal dynamics of claims and +evidence should further enhance retrieval accuracy. Additionally, ci- +tation tracking to trace the original source of paraphrased evidence, +ensuring the first-published timestamp is accurately recorded. + +Citation-based tracking provides a promising approach to esti- +mating evidence timestamps. However, the widespread presence of +multiple citations in scientific literature makes it difficult to iden- +tify which references should be tracked, increasing computational +costs and reducing efficiency. Moreover, indirect citations and para- +phrased references, particularly in introductory sections, further +obscure the retrieval of the first-published source. To address these +issues, we propose the following research direction to develop a +more effective approach for citation tracking and timestamp attri- +bution. + +RD.4. Citation-Based Evidence Tracking + +Intuitively, self-contained evidence refers to information directly +presented in the body of the current paper, while cited evidence is +derived from external sources referenced by the paper. To explore +this distinction, we analysed 22 accessible full papers out of 24 +gold evidence for the claim in Table 2. We prompted GPT-4o to +search for supporting/refuting evidence and determine whether +each piece of evidence was paraphrased/summarised from a citation +or is self-contained. + +Figure 2: Evidence sources in scientific literature. + +The result shown in Figure 2 reveals, as expected, that cited +evidence is predominantly located in the introduction, while self- +contained evidence is more common in the results and conclusion +sections. While we observed a few inaccurate outcomes (e.g., one +‘From Cited’ evidence appearing in ‘Abstract’ is misjudged), the +overall distribution remains discernible and interpretable. In ad- +dition, we also observed that sentences in scientific literature fre- +quently contain multiple citations, making it costly to manually +extract the original timestamp of cited evidence. To address this +challenge, we propose a two-step approach: + +1. Identify track-worthy citations. Not all citations are equally +important for fact verification. Track-worthy citations should in- +clude: conclusive evidence directly influences claim verification and + +441 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +plausible evidence that may impact claim assessment [108]. Since +checking every cited document is expensive, an initial filtering +step is required. A potential solution is to rank citations based on +their relevance and function. Citation recommendation [30, 34] is a +similar task to identify relevant publications for a given statement +using retrieval models. Beyond relevance, the function of citation +can be referred to as a signal to adjust priority. Citations are classi- +fied into eight categories: background, motivation, uses, extends, +similarities, differences, compare/contrast, and future work [42]. +While single evidence has multiple citations, the function of cita- +tion can help identify the citations that align with the evidence. +For example, for predicted evidence ‘Experiments show model A +outperforms previous SOTA model B [citation 1,2,3]’, cited papers +for ‘model A/B’ in the ‘introduction’ function are less check-worthy +than the citation in the ‘compare/contrast’ function. By prioritising +high-impact citations, retrieval costs can be significantly reduced. +2. Track the original timestamp of evidence. Once track- +worthy citations are identified, the next step is to trace the original +timestamp of cited evidence via direct and indirect citation tracking. +Direct citation tracking can be applied if the publication date of the +cited paper is straightforward to retrieve. However, some evidence +is paraphrased or indirectly cited, requiring a deep tracking mech- +anism to trace the citation path. Citation graph analysis [17, 96] +can help map citation paths using directed graphs and applying +search algorithms to identify the earliest relevant source. A recent +study [118] found that reference errors ś references do not include +information to support statement ś frequently appear ranging from +11% to 41% across domains. Addressing these errors introduces +a sub-task for scientific fact-checking: verifying whether a cited +reference truly supports the claim. To ensure feasibility, an early +explorative study can assume that scientific literature generally +adheres to citation conventions, preventing infinite citation loops +in verification. + +In summary, to explore time-aware fact-checking for scientific +literature, we propose two potential research directions: (1) Integrat- +ing temporal information into retrieval and verification to handle +outdated evidence and (2) Developing citation-based tracking meth- +ods to identify the original source and timestamp of evidence. These +directions provide a basis for studying the impact of time-aware +mechanisms in scientific fact-checking systems. + +5 Structured Long-Context Evidence Retrieval +Existing scientific fact-checking datasets construct evidence cor- +pora using fragmented sentences, paragraphs, or abstracts [16, 49, +62, 72, 75, 100, 104]. However, scientific literature is typically pre- +sented in structured, visually rich formats, often as PDF documents, +where different sections serve distinct functions: abstracts, results, +and conclusions summarise key findings, while background and in- +troduction sections provide prior research context. With the diverse +and unique functionalities of scientific literature components, this +section explores challenges and potential research directions for +advancing scientific fact-checking at the full-paper level. Existing +fact-checking pipelines typically follow a document retrieval and +then sentence selection paradigm [37, 97, 116]. For general fact- +checking, evidence retrieval often uses top-ranked sentences from +top-ranked documents, treating them as self-contained evidence + +Title +Authors + +Publish Venue +Publish Time + +Citations: +[ 25, 44 , 60 , 67, 78 , 82 , 92 + +Figure 3: An example for parsing the scientific literature + +units. However, this approach neglects long-range context, as using +the extracted sentences ignores surrounding information to sup- +port verification. In contrast, scientific documents exhibit higher +document-level consistency of verdict, commonly one paper hav- +ing a sole standpoint to a given question, making document-level +processing necessary for scientific fact-checking [50, 102, 119]. + +LLMs have recently demonstrated growing capability to process +long contexts. However, it remains challenging to fact-check an +entire full-text document in a single pass [105]. LLMs are prone to +hallucinating content that is not grounded in the provided docu- +ments [2, 20, 23, 88] and are often susceptible to distraction from +irrelevant context [82]. Their reasoning capabilities also degrade +as text length increases [46]. A potential solution is context distil- +lation, as used in retrieval-augmented generation (RAG), to filter +high-quality context and mitigate hallucination, which improves +the performance of downstream QA tasks [106]. However, unlike +the QA task, fact-checking requires explicit retrieval of explicit. Fil- +tering context may remove crucial supporting evidence, leading to +incomplete verification. Moreover, much of the historical scientific +literature exists in PDF format. Although recent multimodal LLMs +are capable of consuming PDFs and conducting reasoning tasks, +their capabilities are still limited to surface-level understanding. +For example, they frequently fail to capture cross-page content +and complex layout structures [60, 87, 95]. By prompting GPT-4o +to locate evidence in PDF literature with the results presented in +Section 4.2, we observed that it mislocated evidence in incorrect +sections or non-existent sections. These limitations indicate that ex- +isting LLMs are not capable of supporting end-to-end fact-checking +for long-context scientific literature ś not only for raw PDFs but +also for plain-text documents. + +In addition to their length, scientific papers follow a structured +format that introduces additional challenges for verification. Back- +ground and introduction sections often cite prior work, which +may conflict with conclusions drawn later, while discussion sec- +tions highlight limitations that can cast doubt on earlier findings. +These internally inconsistent signals may introduce misleading +information, demanding reasoning that accounts for both the fac- +tual assertions and the functional roles of different sections. Pars- +ing scientific documents into structured units enables a modular +pipeline where retrieval and verification can be independently opti- +mised. This modular structure facilitates interpretability, robustness, +and denoising of conflicting or irrelevant content. Layout-aware +tools offer a practical foundation for such structure-aware process- +ing [19, 32, 41, 56, 81], as illustrated in Figure 3. + +442 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +To address these challenges, we advocate for a retrieval frame- +work that explicitly considers the document structure commonly +found within scientific literature. Such a system should identify +targeted evidence, capture long-range context across sections, and +suppress irrelevant or conflicting content. We next outline a direc- +tion toward adaptive, section-aware retrieval strategies designed +to meet these requirements. + +RD.5. Adaptive Section-Aware Evidence Retrieval + +Scientific literature follows a structured format where different sec- +tions serve distinct functions, presenting challenges for traditional +evidence retrieval and verification in fact-checking systems. Exist- +ing retrieval methods often operate at the sentence or paragraph +level, neglecting the long-range context and the structured nature +of scientific documents. Additionally, large language models (LLMs) +struggle to accurately associate claims with the appropriate sec- +tions, leading to potential misinterpretations and inconsistencies. +A promising research direction is adaptive section-aware ev- +idence retrieval, which dynamically adjusts retrieval strategies +based on document structure and claim types. This approach con- +sists of two key components: + +Claim-evidence matching. Claims should first be matched to +the most relevant sections. For instance, experiment-driven claims, +such as “X method improves accuracy compared to Y,ž should pri- +marily retrieve evidence from the ‘Results’ and ‘Conclusion’ sec- +tions, as these contain empirical findings. In contrast, background +or theoretical claims, such as “X method is widely used in Y applica- +tions,ž should focus on the Introduction and Background sections, +which provide foundational knowledge. Prioritising section-aware +evidence retrieval helps filter irrelevant context and reduces re- +trieval noise. + +Contextual expansion. After retrieving primary evidence, the +system should augment it with relevant contextual information +from other sections to improve interpretability. For example, method- +ological details from the ‘Analysis’ section can provide additional +support for experimental claims, while historical context from the +Background section can clarify theoretical claims. Access to the full +document allows for retrieving finer-grained evidence or comple- +mentary details that fragmented approaches may overlook. + +In summary, an effective adaptive section-aware retrieval sys- +tem must overcome challenges in accurately parsing document +structures, prioritising relevant sections based on claim types, and +efficiently handling conflicting evidence. By integrating structured +document parsing, hierarchical retrieval strategies, and context- +aware reasoning, future systems can leverage richer evidence from +full-paper scientific literature while reducing LLM hallucinations. +Advancing these techniques will enhance the reliability and inter- +pretability of scientific fact-checking systems. + +6 Multimodal content in Science +Scientific literature often conveys key evidence using non-textual +elements such as tables, charts, and figures, which are commonly +used to present experimental results, statistical analyses, and the- +oretical models, as shown in Figure 3. For full-text scientific fact- +checking, especially across various fields of science and technol- +ogy, it is crucial to move beyond text and accurately interpret + +these structured elements to ensure comprehensive verification. +Fact-checking and misinformation detection on individual modal- +ities ś such as figures [1, 65, 66, 69, 93, 112], charts [4, 7], and +tables [3, 13, 22, 29, 36, 57, 79, 83, 113] ś has been studied inde- +pendently [5]. To improve tabular reasoning, transformer-based +approaches such as TAPAS [38] and Table-BERT [114] have been de- +veloped. However, these techniques perform poorly in SCITAB [57], +a dataset for scientific fact-checking on tables, with results barely +above random. One possible cause is the lack of contextual ground- +ing for tables, which are rarely self-contained. Similarly, figures and +charts in scientific papers often require surrounding textual expla- +nations for correct interpretation. Recent datasets like AVerImaTeC +[18] address image-text verification using web-sourced data, but sci- +entific domains present greater challenges: figures are densely struc- +tured, often span multiple sections, and require domain-specific +understanding capability. + +We argue that scientific fact-checking should shift toward full- +paper analysis, where structured elements are interpreted along- +side their textual context. Unlike standalone multimodal models, +document-level processing enables cross-referencing between fig- +ures/tables/charts and their descriptions, facilitating more faithful +and complete verification. + +RD.6. Multi-modal Evidence Alignment + +Scientific fact-checking requires integrating evidence across multi- +ple modalities, including text, tables, and figures, to ensure consis- +tency and completeness. Recent multimodal information retrieval +datasets in the scientific domain [71, 110] provide aligned pairs +of textual and structured content, offering a foundation for cross- +modal reasoning. However, in real scientific documents, structured +elements, such as figures and tables, are not always located close +to their descriptive text, making alignment a non-trivial challenge. +A promising direction is to explicitly align structured elements +with their corresponding textual explanations within the same +document. In scientific articles, tables and figures are typically ex- +plained through captions or surrounding sentences. Layout-aware +parsing techniques can help identify these elements and link them +to relevant text spans. Once aligned, their contents can be jointly +encoded, enabling claim verification that draws on both structured +data and contextual text. This alignment facilitates more coherent +retrieval and reasoning across modalities, improving the reliability +of multimodal fact-checking. + +This unified framework contrasts with traditional multimodal +systems that process each modality in isolation. Effective alignment +demands progress in scientific document understanding [24, 60], +visual structure parsing [32], and domain-specific retrieval [71]. +Integrating these efforts will support full-document, multimodal +verification, where structured evidence is faithfully grounded in its +textual context. + +7 Credibility of scientific literature +Existing evidence retrieval models often favour high-ranking doc- +uments based on semantic relevance, often overlooking scientific +rigour [98]. As a result, low-quality documents may be retrieved +as evidence, undermining the credibility of scientific fact-checking. +The reliability of a claim verification process is inherently tied to + +443 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +the quality of the supporting literature, making evidence credibility +a crucial factor in scientific fact-checking. + +In the domain of scientific literature, credibility assessment is +influenced by multiple factors, including peer-review status, which +ensures methodological scrutiny, citation impact which indicates +how influential a study is within its field, and experimental rigour, +reflecting the robustness of a study’s methodology. However, the +proliferation of non-peer-reviewed manuscripts and publications +from venues with varying editorial standards, particularly in open- +access repositories, poses a growing challenge. Such sources may +lack the rigorous methodological scrutiny necessary to ensure re- +liable scientific conclusions. Therefore, incorporating additional +quality indicators is essential for enhancing the robustness of sci- +entific fact-checking. + +RD.7. Extending Indicators of Evidence Quality + +Evaluating scientific literature quality extends beyond content +reliability, and should consider factors including venue reputa- +tion, methodological rigour, and experimental transparency. High- +impact journals and prestigious conferences generally enforce strin- +gent peer-review standards, contributing to the credibility of pub- +lished research. Similarly, the expertise and prior contributions of +an author, particularly in reputable venues, can provide further +insight into the credibility of a study. In addition to traditional +metadata, emerging indicators such as replication status, data avail- +ability, and adherence to reporting guidelines can further reflect +methodological soundness. Although metadata-based credibility +assessment is a useful heuristic, it is not foolproof. For example, +selective reporting and statistical manipulation still exist, as some +widely cited studies have later been retracted due to methodolog- +ical flaws [31]. While the integration of such metadata remains a +reasonable approach, as these indicators generally correlate with lit- +erature quality, their limitations must be acknowledged, given that +even widely cited studies can occasionally be subject to retraction +due to undetected methodological flaws. + +8 Scientific terminology complexity +Scientific fact-checking systems often encounter challenges when +aligning claims with supporting evidence due to mismatches in +terminology granularity [101, 107]. The prevalence of hierarchical +and synonymous scientific terms introduces significant challenges +in fact verification. Many concepts exist at multiple levels of speci- +ficity, where broader categories encompass more specific subtypes, +leading to ambiguity in claim-evidence alignment. This issue is +further exacerbated by high token-level similarity among related +terms, making it difficult for models to differentiate between general +and specific concepts. As a result, models often misinterpret evi- +dence relevance, increasing the likelihood of incorrect verification +outcomes. + +This issue arises when a claim uses a broad term, while the sup- +porting evidence provides a more specific instance, or vice versa. +As in the following example, such mismatches can lead to incor- +rect veracity assignments, as existing models struggle to recognise +hierarchical relationships between concepts. + +Claim: Cancer risk is lower in individuals with a history of + +alcohol consumption. + +Supports: Alcohol consumption was associated with a decreased + +risk of thyroid cancer. + +This issue is common within scientific fact checking and has been +reported to occur within 44% of annotated examples in SciFact-Open +[101]. Hence, we argue that capturing the hierarchical relationship +could be a research direction to improve verification performance +in the scientific domain, by solving the mismatch problem. + +RD.8. Hierarchical Concept Modelling + +To address this issue, ontology-based reasoning can be integrated +into fact-checking pipelines. Structured ontologies such as MeSH +[54] and UMLS [53] define hierarchical relationships that help sys- +tems infer term specificity. Recognising that ‘lung cancer’ is a sub- +type of ‘cancer’ enables better claim-evidence alignment, mitigat- +ing errors caused by lexical similarity. Beyond that, the knowledge +graph can enrich ontological reasoning by encoding both hier- +archical and associative relationships among scientific concepts +[39, 48]. However, its potential for resolving terminology granular- +ity mismatches in scientific fact-checking remains unexplored. Ad- +ditionally, representation learning techniques such as contrastive +learning can embed these hierarchical relationships into vector +space representations, reducing reliance on token-level similarity. +Domain-specific models like SciBERT [11] and PubMedBERT [35] +can further enhance contextual understanding by incorporating +structured knowledge into retrieval and verification processes. By +leveraging ontological reasoning, knowledge graphs, and struc- +tured embeddings, scientific fact-checking systems can better align +claims with relevant evidence, reducing verification errors caused +by terminology granularity mismatches. In addition to enhancing +precision, this also has potential to improve interpretability by +making model decisions more transparent. + +9 Conclusion +This paper explores the evolution of scientific fact-checking method- +ologies from abstract-level approaches to full-paper frameworks +on large-scale corpora. As the volume and diversity of scientific +knowledge continue to grow, the challenges of verifying claims +across heterogeneous sources become increasingly complex. By +addressing the complexities inherent in scientific literature, includ- +ing its evolving nature, structured format, and the necessity for +precise evidence retrieval, we underscore the importance of de- +veloping specialised retrieval systems capable of managing large, +multimodal, and time-sensitive evidence retrieval. Furthermore, +this work proposes several research directions aimed at improving +scientific fact-checking efficiency and reliability. They include the +integration of time-aware evidence retrieval to ensure the use of the +most relevant and up-to-date findings, adaptive document process- +ing to enable context-sensitive retrieval strategies, and multi-modal +evidence alignment to integrate text, tables and figures to enhance +verification accuracy. Overcoming these challenges has potential to +improve the accuracy of fact-checking processes and also facilitate +the scalability and applicability of these systems in diverse scientific +fact-checking scenarios. By bridging the gap between scientific fact- +checking and effective evidence retrieval, these advancements will +contribute to more robust, interpretable, and trustworthy scientific +fact-checking methodologies for real-world applications. + +444 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +References + +[1] Sahar Abdelnabi, Rakibul Hasan, and Mario Fritz. 2022. Open-domain, content- +based, multi-modal fact-checking of out-of-context images via online resources. +In Proceedings of the IEEE/CVF conference on computer vision and pattern recog- +nition. 14940ś14949. + +[2] Vaibhav Adlakha, Parishad BehnamGhader, Xing Han Lu, Nicholas Meade, +and Siva Reddy. 2024. Evaluating Correctness and Faithfulness of Instruction- +Following Models for Question Answering. Transactions of the Association for +Computational Linguistics 12 (2024), 681ś699. doi:10.1162/tacl_a_00667 + +[3] Mubashara Akhtar, Oana Cocarascu, and Elena Simperl. 2022. PubHealthTab: A +Public Health Table-based Dataset for Evidence-based Fact Checking. In Findings +of the Association for Computational Linguistics: NAACL 2022, Marine Carpuat, +Marie-Catherine de Marneffe, and Ivan Vladimir Meza Ruiz (Eds.). Association +for Computational Linguistics, Seattle, United States, 1ś16. doi:10.18653/v1/ +2022.findings-naacl.1 + +[4] Mubashara Akhtar, Oana Cocarascu, and Elena Simperl. 2023. Reading and +Reasoning over Chart Images for Evidence-based Automated Fact-Checking. +In Findings of the Association for Computational Linguistics: EACL 2023, An- +dreas Vlachos and Isabelle Augenstein (Eds.). Association for Computational +Linguistics, Dubrovnik, Croatia, 399ś414. doi:10.18653/v1/2023.findings-eacl.30 +[5] Mubashara Akhtar, Michael Schlichtkrull, Zhijiang Guo, Oana Cocarascu, Elena +Simperl, and Andreas Vlachos. 2023. Multimodal Automated Fact-Checking: +A Survey. In Findings of the Association for Computational Linguistics: EMNLP +2023, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Com- +putational Linguistics, Singapore, 5430ś5448. doi:10.18653/v1/2023.findings- +emnlp.361 + +[6] Mubashara Akhtar, Michael Schlichtkrull, and Andreas Vlachos. 2024. Ev2R: +Evaluating Evidence Retrieval in Automated Fact-Checking. arXiv preprint +arXiv:2411.05375 (2024). + +[7] Mubashara Akhtar, Nikesh Subedi, Vivek Gupta, Sahar Tahmasebi, Oana Co- +carascu, and Elena Simperl. 2024. ChartCheck: Explainable Fact-Checking +over Real-World Chart Images. In Findings of the Association for Computational +Linguistics: ACL 2024, Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). +Association for Computational Linguistics, Bangkok, Thailand, 13921ś13937. +doi:10.18653/v1/2024.findings-acl.828 + +[8] Liesbeth Allein, Marlon Saelens, Ruben Cartuyvels, and Marie-Francine Moens. +2023. Implicit Temporal Reasoning for Evidence-Based Fact-Checking. In Find- +ings of the Association for Computational Linguistics: EACL 2023, Andreas Vla- +chos and Isabelle Augenstein (Eds.). Association for Computational Linguistics, +Dubrovnik, Croatia, 176ś189. doi:10.18653/v1/2023.findings-eacl.13 + +[9] Isabelle Augenstein, Christina Lioma, Dongsheng Wang, Lucas Chaves Lima, +Casper Hansen, Christian Hansen, and Jakob Grue Simonsen. 2019. MultiFC: A +Real-World Multi-Domain Dataset for Evidence-Based Fact Checking of Claims. +In Proceedings of the 2019 Conference on Empirical Methods in Natural Language +Processing and the 9th International Joint Conference on Natural Language Pro- +cessing (EMNLP-IJCNLP), Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan +(Eds.). Association for Computational Linguistics, Hong Kong, China, 4685ś4697. +doi:10.18653/v1/D19-1475 + +[10] Dara Bahri, Yi Tay, Che Zheng, Donald Metzler, and Andrew Tomkins. 2020. +Choppy: Cut transformer for ranked list truncation. In Proceedings of the 43rd +International ACM SIGIR Conference on Research and Development in Information +Retrieval. 1513ś1516. + +[11] Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. SciBERT: A Pretrained Language +Model for Scientific Text. In Proceedings of the 2019 Conference on Empirical +Methods in Natural Language Processing and the 9th International Joint Confer- +ence on Natural Language Processing (EMNLP-IJCNLP), Kentaro Inui, Jing Jiang, +Vincent Ng, and Xiaojun Wan (Eds.). Association for Computational Linguistics, +Hong Kong, China, 3615ś3620. doi:10.18653/v1/D19-1371 + +[12] Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. Longformer: The + +Long-Document Transformer. arXiv:2004.05150 (2020). + +[13] Chandra Sekhar Bhagavatula, Thanapon Noraset, and Doug Downey. 2013. +Methods for exploring and mining tables on wikipedia. In Proceedings of the +ACM SIGKDD workshop on interactive data exploration and analytics. 18ś26. + +[14] Reem Bin-Hezam and Mark Stevenson. 2023. Combining Counting Processes +and Classification Improves a Stopping Rule for Technology Assisted Review. +In Findings of the Association for Computational Linguistics: EMNLP 2023, Houda +Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational +Linguistics, Singapore, 2603ś2609. doi:10.18653/v1/2023.findings-emnlp.171 + +[15] Reem Bin-Hezam and Mark Stevenson. 2024. RLStop: A Reinforcement Learning +Stopping Method for TAR. In Proceedings of the 47th International ACM SIGIR +Conference on Research and Development in Information Retrieval. 2604ś2608. + +[16] Jannis Bulian, Jordan Boyd-Graber, Markus Leippold, Massimiliano Ciaramita, +and Thomas Diggelmann. 2020. Climate-fever: A dataset for verification of +real-world climate claims. In NeurIPS 2020 Workshop on Tackling Climate Change +with Machine Learning. + +[17] Peter Buneman, Dennis Dosso, Matteo Lissandrini, and Gianmaria Silvello. 2021. +Data citation and the citation graph. Quantitative Science Studies 2, 4 (2021), + +1399ś1422. + +[18] Rui Cao, Zifeng Ding, Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. +2025. AVerImaTeC: A Dataset for Automatic Verification of Image-Text Claims +with Evidence from the Web. arXiv:2505.17978 [cs.CL] https://arxiv.org/abs/ +2505.17978 + +[19] Catherine Chen, Zejiang Shen, Dan Klein, Gabriel Stanovsky, Doug Downey, +and Kyle Lo. 2023. Are Layout-Infused Language Models Robust to Layout +Distribution Shifts? A Case Study with Scientific Documents. In Findings of the +Association for Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd- +Graber, and Naoaki Okazaki (Eds.). Association for Computational Linguistics, +Toronto, Canada, 13345ś13360. doi:10.18653/v1/2023.findings-acl.844 + +[20] Jiawei Chen, Hongyu Lin, Xianpei Han, and Le Sun. 2024. Benchmarking large +language models in retrieval-augmented generation. In Proceedings of the AAAI +Conference on Artificial Intelligence, Vol. 38. 17754ś17762. + +[21] Qingyu Chen, Yifan Peng, and Zhiyong Lu. 2019. BioSentVec: creating sen- +tence embeddings for biomedical texts. In 2019 IEEE International Conference on +Healthcare Informatics (ICHI). IEEE, 1ś5. + +[22] Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, +Shiyang Li, Xiyou Zhou, and William Yang Wang. 2020. TabFact : A Large-scale +Dataset for Table-based Fact Verification. In International Conference on Learning +Representations (ICLR). Addis Ababa, Ethiopia. + +[23] Sabrina Chiesurin, Dimitris Dimakopoulos, Marco Antonio Sobrevilla Cabezudo, +Arash Eshghi, Ioannis Papaioannou, Verena Rieser, and Ioannis Konstas. 2023. +The Dangers of trusting Stochastic Parrots: Faithfulness and Trust in Open- +domain Conversational Question Answering. In Findings of the Association for +Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 947ś959. doi:10.18653/v1/2023.findings-acl.60 + +[24] Jaemin Cho, Debanjan Mahata, Ozan Irsoy, Yujie He, and Mohit Bansal. 2024. +M3docrag: Multi-modal retrieval is what you need for multi-page multi- +document understanding. arXiv preprint arXiv:2411.04952 (2024). + +[25] Gordon V Cormack and Maura R Grossman. 2014. Evaluation of machine- +learning protocols for technology-assisted review in electronic discovery. In +Proceedings of the 37th international ACM SIGIR conference on Research & devel- +opment in information retrieval. 153ś162. + +[26] Nick Craswell, Bhaskar Mitra, Emine Yilmaz, and Daniel Campos. 2021. +Overview of the TREC 2020 deep learning track. arXiv:2102.07662 [cs.IR] +https://arxiv.org/abs/2102.07662 + +[27] Nick Craswell, Bhaskar Mitra, Emine Yilmaz, Daniel Campos, and Ellen M +Voorhees. 2020. Overview of the TREC 2019 deep learning track. arXiv preprint +arXiv:2003.07820 (2020). + +[28] Alphaeus Dmonte, Roland Oruche, Marcos Zampieri, Prasad Calyam, and Is- +abelle Augenstein. 2024. Claim Verification in the Age of Large Language +Models: A Survey. arXiv preprint arXiv:2408.14317 (2024). + +[29] Haoyu Dong and Zhiruo Wang. 2024. Large language models for tabular data: +Progresses and future directions. In Proceedings of the 47th International ACM +SIGIR Conference on Research and Development in Information Retrieval. 2997ś +3000. + +[30] Michael Färber and Adam Jatowt. 2020. Citation recommendation: approaches + +and datasets. International Journal on Digital Libraries 21, 4 (2020), 375ś405. + +[31] Aaron HA Fletcher and Mark Stevenson. 2025. Predicting retracted research: a +dataset and machine learning approaches. Research Integrity and Peer Review +10, 1 (2025), 1ś10. + +[32] Raymond Fok, Hita Kambhamettu, Luca Soldaini, Jonathan Bragg, Kyle Lo, Marti +Hearst, Andrew Head, and Daniel S Weld. 2023. Scim: Intelligent skimming +support for scientific papers. In Proceedings of the 28th International Conference +on Intelligent User Interfaces. 476ś490. + +[33] Maura R Grossman, Gordon V Cormack, and Adam Roegiest. 2016. TREC 2016 + +Total Recall Track Overview.. In TREC. + +[34] Nianlong Gu, Yingqiang Gao, and Richard HR Hahnloser. 2022. Local citation +recommendation with hierarchical-attention text encoder and scibert-based +reranking. In European conference on information retrieval. Springer, 274ś288. +[35] Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong +Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2020. Domain-Specific +Language Model Pretraining for Biomedical Natural Language Processing. +arXiv:arXiv:2007.15779 + +[36] Zihui Gu, Ju Fan, Nan Tang, Preslav Nakov, Xiaoman Zhao, and Xiaoyong Du. +2022. PASTA: Table-Operations Aware Fact Verification via Sentence-Table +Cloze Pre-training. In Proceedings of the 2022 Conference on Empirical Methods +in Natural Language Processing, Yoav Goldberg, Zornitsa Kozareva, and Yue +Zhang (Eds.). Association for Computational Linguistics, Abu Dhabi, United +Arab Emirates, 4971ś4983. doi:10.18653/v1/2022.emnlp-main.331 + +[37] Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. 2022. A survey on +automated fact-checking. Transactions of the Association for Computational +Linguistics 10 (2022), 178ś206. + +[38] Jonathan Herzig, Pawel Krzysztof Nowak, Thomas Müller, Francesco Piccinno, +and Julian Eisenschlos. 2020. TaPas: Weakly Supervised Table Parsing via +Pre-training. In Proceedings of the 58th Annual Meeting of the Association for + +445 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +Computational Linguistics, Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel +Tetreault (Eds.). Association for Computational Linguistics, Online, 4320ś4333. +doi:10.18653/v1/2020.acl-main.398 + +[39] Jason Hoelscher-Obermaier, Edward Stevinson, Valentin Stauber, Ivaylo Zhelev, +Viktor Botev, Ronin Wu, and Jeremy Minton. 2022. Leveraging knowledge +graphs to update scientific word embeddings using latent semantic imputation. +In Proceedings of the first Workshop on Information Extraction from Scientific +Publications, Tirthankar Ghosal, Sergi Blanco-Cuaresma, Alberto Accomazzi, +Robert M. Patton, Felix Grezes, and Thomas Allen (Eds.). Association for Com- +putational Linguistics, Online, 43ś53. doi:10.18653/v1/2022.wiesp-1.6 + +[40] Xuming Hu, Zhaochen Hong, Zhijiang Guo, Lijie Wen, and Philip Yu. 2023. Read +it twice: Towards faithfully interpretable fact verification by revisiting evidence. +In Proceedings of the 46th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 2319ś2323. + +[41] Po-Wei Huang, Abhinav Ramesh Kashyap, Yanxia Qin, Yajing Yang, and Min-Yen +Kan. 2022. Lightweight Contextual Logical Structure Recovery. In Proceedings +of the Third Workshop on Scholarly Document Processing, Arman Cohan, Guy +Feigenblat, Dayne Freitag, Tirthankar Ghosal, Drahomira Herrmannova, Petr +Knoth, Kyle Lo, Philipp Mayr, Michal Shmueli-Scheuer, Anita de Waard, and +Lucy Lu Wang (Eds.). Association for Computational Linguistics, Gyeongju, +Republic of Korea, 37ś48. https://aclanthology.org/2022.sdp-1.5/ + +[42] Tomoki Ikoma and Shigeki Matsubara. 2023. On the Use of Language Models for +Function Identification of Citations in Scholarly Papers. In Proceedings of the Sec- +ond Workshop on Information Extraction from Scientific Publications, Tirthankar +Ghosal, Felix Grezes, Thomas Allen, Kelly Lockhart, Alberto Accomazzi, and +Sergi Blanco-Cuaresma (Eds.). Association for Computational Linguistics, Bali, +Indonesia, 130ś135. doi:10.18653/v1/2023.wiesp-1.15 + +[43] E. Kanoulas, Dan Li, Leif Azzopardi, and René Spijker. 2018. CLEF 2017 Tech- +nologically Assisted Reviews in Empirical Medicine Overview. In Conference +and Labs of the Evaluation Forum. https://api.semanticscholar.org/CorpusID: +246440739 + +[44] Evangelos Kanoulas, Dan Li, Leif Azzopardi, and Rene Spijker. 2018. CLEF +2018 technologically assisted reviews in empirical medicine overview. In CEUR +workshop proceedings, Vol. 2125. + +[45] Evangelos Kanoulas, Dan Li, Leif Azzopardi, and Rene Spijker. 2019. CLEF 2019 +technology assisted reviews in empirical medicine overview. In CEUR workshop +proceedings, Vol. 2380. 250. + +[46] Marzena Karpinska, Katherine Thai, Kyle Lo, Tanya Goyal, and Mohit Iyyer. +2024. One Thousand and One Pairs: A “novelž challenge for long-context +language models. In Proceedings of the 2024 Conference on Empirical Methods in +Natural Language Processing, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung +Chen (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +17048ś17085. doi:10.18653/v1/2024.emnlp-main.948 + +[47] Mohammed Abdul Khaliq, Paul Yu-Chun Chang, Mingyang Ma, Bernhard +Pflugfelder, and Filip Miletić. 2024. RAGAR, Your Falsehood Radar: RAG- +Augmented Reasoning for Political Fact-Checking using Multimodal Large +Language Models. In Proceedings of the Seventh Fact Extraction and VERifica- +tion Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +280ś296. doi:10.18653/v1/2024.fever-1.29 + +[48] Jiho Kim, Sungjin Park, Yeonsu Kwon, Yohan Jo, James Thorne, and Edward +Choi. 2023. FactKG: Fact Verification via Reasoning on Knowledge Graphs. +In Proceedings of the 61st Annual Meeting of the Association for Computational +Linguistics (Volume 1: Long Papers), Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 16190ś16206. doi:10.18653/v1/2023.acl-long.895 + +[49] Neema Kotonya and Francesca Toni. 2020. Explainable Automated Fact- +Checking for Public Health Claims. In Proceedings of the 2020 Conference on +Empirical Methods in Natural Language Processing (EMNLP), Bonnie Webber, +Trevor Cohn, Yulan He, and Yang Liu (Eds.). Association for Computational +Linguistics, Online, 7740ś7754. doi:10.18653/v1/2020.emnlp-main.623 + +[50] Xiangci Li, Gully A Burns, and Nanyun Peng. 2021. A Paragraph-level Multi-task + +Learning Model for Scientific Fact-Verification.. In SDU@ AAAI. + +[51] Xiaoxi Li, Zhicheng Dou, Yujia Zhou, and Fangchao Liu. 2024. Corpuslm: +Towards a unified language model on corpus for knowledge-intensive tasks. +In Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 26ś37. + +[52] Yen-Chieh Lien, Daniel Cohen, and W Bruce Croft. 2019. An assumption-free +approach to the dynamic truncation of ranked lists. In Proceedings of the 2019 +ACM SIGIR International Conference on Theory of Information Retrieval. 79ś82. +[53] Donald AB Lindberg, Betsy L Humphreys, and Alexa T McCray. 1993. The +unified medical language system. Yearbook of medical informatics 2, 01 (1993), +41ś51. + +[54] Carolyn E Lipscomb. 2000. Medical subject headings (MeSH). Bulletin of the + +Medical Library Association 88, 3 (2000), 265. + +[55] Qi Liu, Bo Wang, Nan Wang, and Jiaxin Mao. 2025. Leveraging Passage Em- +beddings for Efficient Listwise Reranking with Large Language Models. In Pro- +ceedings of the ACM on Web Conference 2025 (Sydney NSW, Australia) (WWW +’25). Association for Computing Machinery, New York, NY, USA, 4274ś4283. +doi:10.1145/3696410.3714554 + +[56] Kyle Lo, Joseph Chee Chang, Andrew Head, Jonathan Bragg, Amy X. Zhang, Cas- +sidy Trier, Chloe Anastasiades, Tal August, Russell Authur, Danielle Bragg, Erin +Bransom, Isabel Cachola, Stefan Candra, Yoganand Chandrasekhar, Yen-Sung +Chen, Evie Yu-Yen Cheng, Yvonne Chou, Doug Downey, Rob Evans, Raymond +Fok, Fangzhou Hu, Regan Huff, Dongyeop Kang, Tae Soo Kim, Rodney Kinney, +Aniket Kittur, Hyeonsu B. Kang, Egor Klevak, Bailey Kuehl, Michael J. Lan- +gan, Matt Latzke, Jaron Lochner, Kelsey MacMillan, Eric Marsh, Tyler Murray, +Aakanksha Naik, Ngoc-Uyen Nguyen, Srishti Palani, Soya Park, Caroline Paulic, +Napol Rachatasumrit, Smita Rao, Paul Sayre, Zejiang Shen, Pao Siangliulue, +Luca Soldaini, Huy Tran, Madeleine van Zuylen, Lucy Lu Wang, Christopher +Wilhelm, Caroline Wu, Jiangjiang Yang, Angele Zamarron, Marti A. Hearst, and +Daniel S. Weld. 2024. The Semantic Reader Project. Commun. ACM 67, 10 (Sept. +2024), 50ś61. doi:10.1145/3659096 + +[57] Xinyuan Lu, Liangming Pan, Qian Liu, Preslav Nakov, and Min-Yen Kan. 2023. +SCITAB: A Challenging Benchmark for Compositional Reasoning and Claim +Verification on Scientific Tables. In Proceedings of the 2023 Conference on Em- +pirical Methods in Natural Language Processing, Houda Bouamor, Juan Pino, +and Kalika Bali (Eds.). Association for Computational Linguistics, Singapore, +7787ś7813. doi:10.18653/v1/2023.emnlp-main.483 + +[58] María Luengo and David García-Marín. 2020. The performance of truth: politi- +cians, fact-checking journalism, and the struggle to tackle COVID-19 misinfor- +mation. American Journal of Cultural Sociology 8, 3 (2020), 405. + +[59] Yixiao Ma, Qingyao Ai, Yueyue Wu, Yunqiu Shao, Yiqun Liu, Min Zhang, and +Shaoping Ma. 2022. Incorporating retrieval information into the truncation +of ranking lists for better legal search. In Proceedings of the 45th International +ACM SIGIR Conference on Research and Development in Information Retrieval. +438ś448. + +[60] Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, +Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, Pan Zhang, Liangming Pan, Yu- +Gang Jiang, Jiaqi Wang, Yixin Cao, and Aixin Sun. 2024. MMLongBench-Doc: +Benchmarking Long-context Document Understanding with Visualizations. +arXiv:2407.01523 [cs.CV] https://arxiv.org/abs/2407.01523 + +[61] Chuan Meng, Negar Arabzadeh, Arian Askari, Mohammad Aliannejadi, and +Maarten de Rijke. 2024. Ranked list truncation for large language model-based +re-ranking. In Proceedings of the 47th International ACM SIGIR Conference on +Research and Development in Information Retrieval. 141ś151. + +[62] Isabelle Mohr, Amelie Wührl, and Roman Klinger. 2022. CoVERT: A Corpus +of Fact-checked Biomedical COVID-19 Tweets. In Proceedings of the Thirteenth +Language Resources and Evaluation Conference, Nicoletta Calzolari, Frédéric +Béchet, Philippe Blache, Khalid Choukri, Christopher Cieri, Thierry Declerck, +Sara Goggi, Hitoshi Isahara, Bente Maegaard, Joseph Mariani, Hélène Mazo, Jan +Odijk, and Stelios Piperidis (Eds.). European Language Resources Association, +Marseille, France, 244ś257. https://aclanthology.org/2022.lrec-1.26 + +[63] Yuki Momii, Tetsuya Takiguchi, and Yasuo Ariki. 2024. RAG-Fusion Based +Information Retrieval for Fact-Checking. In Proceedings of the Seventh Fact Ex- +traction and VERification Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, +Chenxi Whitehouse, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang +Guo, Christos Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, +and Andreas Vlachos (Eds.). Association for Computational Linguistics, Miami, +Florida, USA, 47ś54. doi:10.18653/v1/2024.fever-1.4 + +[64] Rodrigo Nogueira, Zhiying Jiang, Ronak Pradeep, and Jimmy Lin. 2020. Docu- +ment Ranking with a Pretrained Sequence-to-Sequence Model. In Findings of the +Association for Computational Linguistics: EMNLP 2020, Trevor Cohn, Yulan He, +and Yang Liu (Eds.). Association for Computational Linguistics, Online, 708ś718. +doi:10.18653/v1/2020.findings-emnlp.63 + +[65] Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and +Panagiotis Petrantonakis. 2023. Synthetic misinformers: Generating and com- +bating multimodal misinformation. In Proceedings of the 2nd ACM International +Workshop on Multimedia AI against Disinformation. 36ś44. + +[66] Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and +Panagiotis C Petrantonakis. 2024. VERITE: a Robust benchmark for multimodal +misinformation detection accounting for unimodal bias. International Journal +of Multimedia Information Retrieval 13, 1 (2024), 4. + +[67] Andrew Parry, Debasis Ganguly, and Manish Chandra. 2024. In-Context Learn- +ing" or: How I learned to stop worrying and love" Applied Information Retrieval. +In Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 14ś25. + +[68] Ronak Pradeep, Xueguang Ma, Rodrigo Nogueira, and Jimmy Lin. 2021. Scientific +Claim Verification with VerT5erini. In Proceedings of the 12th International +Workshop on Health Text Mining and Information Analysis, Eben Holderness, +Antonio Jimeno Yepes, Alberto Lavelli, Anne-Lyse Minard, James Pustejovsky, +and Fabio Rinaldi (Eds.). Association for Computational Linguistics, online, +94ś103. https://aclanthology.org/2021.louhi-1.11/ + +446 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +[69] Peng Qi, Zehong Yan, Wynne Hsu, and Mong Li Lee. 2024. SNIFFER: Multi- +modal Large Language Model for Explainable Out-of-Context Misinformation +Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and +Pattern Recognition. 13052ś13062. + +[70] Vatsal Raina and Mark Gales. 2024. Question-Based Retrieval using Atomic +Units for Enterprise RAG. In Proceedings of the Seventh Fact Extraction and VER- +ification Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +219ś233. doi:10.18653/v1/2024.fever-1.25 + +[71] Jonathan Roberts, Kai Han, Neil Houlsby, and Samuel Albanie. 2024. SciFIBench: +Benchmarking Large Multimodal Models for Scientific Figure Interpretation. +arXiv preprint arXiv:2405.08807 (2024). + +[72] Arkadiy Saakyan, Tuhin Chakrabarty, and Smaranda Muresan. 2021. COVID- +Fact: Fact Extraction and Verification of Real-World Claims on COVID-19 Pan- +demic. In Proceedings of the 59th Annual Meeting of the Association for Com- +putational Linguistics and the 11th International Joint Conference on Natural +Language Processing (Volume 1: Long Papers), Chengqing Zong, Fei Xia, Wen- +jie Li, and Roberto Navigli (Eds.). Association for Computational Linguistics, +Online, 2116ś2129. doi:10.18653/v1/2021.acl-long.165 + +[73] Alireza Salemi and Hamed Zamani. 2024. Evaluating retrieval quality in retrieval- +augmented generation. In Proceedings of the 47th International ACM SIGIR Con- +ference on Research and Development in Information Retrieval. 2395ś2400. +[74] Alireza Salemi and Hamed Zamani. 2024. Towards a search engine for machines: +Unified ranking for multiple retrieval-augmented large language models. In +Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 741ś751. + +[75] Mourad Sarrouti, Asma Ben Abacha, Yassine Mrabet, and Dina Demner- +Fushman. 2021. Evidence-based Fact-Checking of Health-related Claims. In +Findings of the Association for Computational Linguistics: EMNLP 2021, Marie- +Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (Eds.). +Association for Computational Linguistics, Punta Cana, Dominican Republic, +3499ś3512. doi:10.18653/v1/2021.findings-emnlp.297 + +[76] Artsiom Sauchuk, James Thorne, Alon Halevy, Nicola Tonellotto, and Fabrizio +Silvestri. 2022. On the role of relevance in natural language processing tasks. +In Proceedings of the 45th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 1785ś1789. + +[77] Michael Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, +Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana +Cocarascu, Arpit Mittal, James Thorne, and Andreas Vlachos. 2024. The Auto- +mated Verification of Textual Claims (AVeriTeC) Shared Task. In Proceedings +of the Seventh Fact Extraction and VERification Workshop (FEVER), Michael +Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, Mubashara +Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana Cocarascu, +Arpit Mittal, James Thorne, and Andreas Vlachos (Eds.). Association for Compu- +tational Linguistics, Miami, Florida, USA, 1ś26. doi:10.18653/v1/2024.fever-1.1 +[78] Michael Schlichtkrull, Zhijiang Guo, and Andreas Vlachos. 2024. Averitec: A +dataset for real-world claim verification with evidence from the web. Advances +in Neural Information Processing Systems 36 (2024). + +[79] Michael Sejr Schlichtkrull, Vladimir Karpukhin, Barlas Oguz, Mike Lewis, Wen- +tau Yih, and Sebastian Riedel. 2021. Joint Verification and Reranking for Open +Fact Checking Over Tables. In Proceedings of the 59th Annual Meeting of the +Association for Computational Linguistics and the 11th International Joint Confer- +ence on Natural Language Processing (Volume 1: Long Papers), Chengqing Zong, +Fei Xia, Wenjie Li, and Roberto Navigli (Eds.). Association for Computational +Linguistics, Online, 6787ś6799. doi:10.18653/v1/2021.acl-long.529 + +[80] Özge Sevgili, Irina Nikishina, Seid Muhie Yimam, Martin Semmann, and +Chris Biemann. 2024. UHH at AVeriTeC: RAG for Fact-Checking with Real- +World Claims. In Proceedings of the Seventh Fact Extraction and VERifica- +tion Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +55ś63. doi:10.18653/v1/2024.fever-1.5 + +[81] Zejiang Shen, Kyle Lo, Lucy Lu Wang, Bailey Kuehl, Daniel S. Weld, and Doug +Downey. 2022. VILA: Improving Structured Content Extraction from Scien- +tific PDFs Using Visual Layout Groups. Transactions of the Association for +Computational Linguistics 10 (2022), 376ś392. doi:10.1162/tacl_a_00466 +[82] Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H +Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be +easily distracted by irrelevant context. In International Conference on Machine +Learning. PMLR, 31210ś31227. + +[83] Qi Shi, Yu Zhang, Qingyu Yin, and Ting Liu. 2021. Logic-level Evidence Retrieval +and Graph-based Verification Network for Table-based Fact Verification. In +Proceedings of the 2021 Conference on Empirical Methods in Natural Language +Processing, Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott +Wen-tau Yih (Eds.). Association for Computational Linguistics, Online and Punta + +Cana, Dominican Republic, 175ś184. doi:10.18653/v1/2021.emnlp-main.16 +[84] Ronit Singal, Pransh Patwa, Parth Patwa, Aman Chadha, and Amitava Das. +2024. Evidence-backed Fact Checking using RAG and Few-Shot In-Context +Learning with LLMs. In Proceedings of the Seventh Fact Extraction and VERifi- +cation Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +91ś98. doi:10.18653/v1/2024.fever-1.10 + +[85] Mark Stevenson and Reem Bin-Hezam. 2023. Stopping Methods for Technology- +assisted Reviews Based on Point Processes. ACM Transactions on Information +Systems 42, 3 (2023), 1ś37. + +[86] Weiwei Sun, Lingyong Yan, Xinyu Ma, Shuaiqiang Wang, Pengjie Ren, Zhumin +Chen, Dawei Yin, and Zhaochun Ren. 2023. +Is ChatGPT Good at Search? +Investigating Large Language Models as Re-Ranking Agents. In Proceedings of +the 2023 Conference on Empirical Methods in Natural Language Processing, Houda +Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational +Linguistics, Singapore, 14918ś14937. doi:10.18653/v1/2023.emnlp-main.923 +[87] Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida, Taku Hasegawa, Itsumi Saito, +and Kuniko Saito. 2023. Slidevqa: A dataset for document visual question +answering on multiple images. In Proceedings of the AAAI Conference on Artificial +Intelligence, Vol. 37. 13636ś13645. + +[88] Liyan Tang, Philippe Laban, and Greg Durrett. 2024. MiniCheck: Efficient Fact- +Checking of LLMs on Grounding Documents. In Proceedings of the 2024 Confer- +ence on Empirical Methods in Natural Language Processing, Yaser Al-Onaizan, +Mohit Bansal, and Yun-Nung Chen (Eds.). Association for Computational Lin- +guistics, Miami, Florida, USA, 8818ś8847. doi:10.18653/v1/2024.emnlp-main.499 +[89] Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna +Gurevych. 2021. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of +Information Retrieval Models. In Thirty-fifth Conference on Neural Information +Processing Systems Datasets and Benchmarks Track (Round 2). https://openreview. +net/forum?id=wCu6T5xFjeJ + +[90] James Thorne, Andreas Vlachos, Christos Christodoulopoulos, and Arpit Mittal. +2018. FEVER: a Large-scale Dataset for Fact Extraction and VERification. In Pro- +ceedings of the 2018 Conference of the North American Chapter of the Association +for Computational Linguistics: Human Language Technologies, Volume 1 (Long Pa- +pers), Marilyn Walker, Heng Ji, and Amanda Stent (Eds.). Association for Compu- +tational Linguistics, New Orleans, Louisiana, 809ś819. doi:10.18653/v1/N18-1074 +[91] James Thorne, Andreas Vlachos, Oana Cocarascu, Christos Christodoulopou- +los, and Arpit Mittal. 2018. The Fact Extraction and VERification (FEVER) +Shared Task. In Proceedings of the First Workshop on Fact Extraction and VER- +ification (FEVER), James Thorne, Andreas Vlachos, Oana Cocarascu, Christos +Christodoulopoulos, and Arpit Mittal (Eds.). Association for Computational +Linguistics, Brussels, Belgium, 1ś9. doi:10.18653/v1/W18-5501 + +[92] Fangzheng Tian, Debasis Ganguly, and Craig Macdonald. 2025. Is Relevance +Propagated from Retriever to Generator in RAG?. In Advances in Information +Retrieval: 47th European Conference on Information Retrieval, ECIR 2025, Lucca, +Italy, April 6ś10, 2025, Proceedings, Part I (Lucca, Italy). Springer-Verlag, Berlin, +Heidelberg, 32ś48. doi:10.1007/978-3-031-88708-6_3 + +[93] Jonathan Tonglet, Marie-Francine Moens, and Iryna Gurevych. 2024. “Image, +Tell me your story!ž Predicting the original meta-context of visual misinfor- +mation. In Proceedings of the 2024 Conference on Empirical Methods in Natural +Language Processing, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen +(Eds.). Association for Computational Linguistics, Miami, Florida, USA, 7845ś +7864. doi:10.18653/v1/2024.emnlp-main.448 + +[94] Herbert Ullrich, Tomáš Mlynář, and Jan Drchal. 2024. AIC CTU system at +AVeriTeC: Re-framing automated fact-checking as a simple RAG task. In Proceed- +ings of the Seventh Fact Extraction and VERification Workshop (FEVER), Michael +Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, Mubashara +Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana Cocarascu, +Arpit Mittal, James Thorne, and Andreas Vlachos (Eds.). Association for Compu- +tational Linguistics, Miami, Florida, USA, 137ś150. doi:10.18653/v1/2024.fever- +1.16 + +[95] Jordy Van Landeghem, Rubèn Tito, Łukasz Borchmann, Michał Pietruszka, Pawel +Joziak, Rafal Powalski, Dawid Jurkiewicz, Mickaël Coustaty, Bertrand Anckaert, +Ernest Valveny, et al. 2023. Document understanding dataset and evaluation +(dude). In Proceedings of the IEEE/CVF International Conference on Computer +Vision. 19528ś19540. + +[96] Vijay Viswanathan, Graham Neubig, and Pengfei Liu. 2021. CitationIE: Lever- +aging the Citation Graph for Scientific Information Extraction. In Proceedings +of the 59th Annual Meeting of the Association for Computational Linguistics +and the 11th International Joint Conference on Natural Language Processing +(Volume 1: Long Papers), Chengqing Zong, Fei Xia, Wenjie Li, and Roberto +Navigli (Eds.). Association for Computational Linguistics, Online, 719ś731. +doi:10.18653/v1/2021.acl-long.59 + +[97] Juraj Vladika and Florian Matthes. 2023. Scientific Fact-Checking: A Survey +of Resources and Approaches. In Findings of the Association for Computational +Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and Naoaki Okazaki + +447 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +SIGIR Conference on Research and Development in Information Retrieval. 2733ś +2743. + +[113] Yunhu Ye, Binyuan Hui, Min Yang, Binhua Li, Fei Huang, and Yongbin Li. 2023. +Large language models are versatile decomposers: Decomposing evidence and +questions for table-based reasoning. In Proceedings of the 46th International +ACM SIGIR Conference on Research and Development in Information Retrieval. +174ś184. + +[114] Pengcheng Yin, Graham Neubig, Wen-tau Yih, and Sebastian Riedel. 2020. +TaBERT: Pretraining for Joint Understanding of Textual and Tabular Data. In Pro- +ceedings of the 58th Annual Meeting of the Association for Computational Linguis- +tics, Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault (Eds.). Associa- +tion for Computational Linguistics, Online, 8413ś8426. doi:10.18653/v1/2020.acl- +main.745 + +[115] Hamed Zamani and Michael Bendersky. 2024. Stochastic rag: End-to-end +retrieval-augmented generation through expected utility maximization. In Pro- +ceedings of the 47th International ACM SIGIR Conference on Research and Devel- +opment in Information Retrieval. 2641ś2646. + +[116] Xia Zeng, Amani S Abumansour, and Arkaitz Zubiaga. 2021. Automated fact- + +checking: A survey. Language and Linguistics Compass 15, 10 (2021), e12438. + +[117] Hengran Zhang, Ruqing Zhang, Jiafeng Guo, Maarten de Rijke, Yixing Fan, and +Xueqi Cheng. 2023. From Relevance to Utility: Evidence Retrieval with Feedback +for Fact Verification. In Findings of the Association for Computational Linguistics: +EMNLP 2023, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association +for Computational Linguistics, Singapore, 6373ś6384. doi:10.18653/v1/2023. +findings-emnlp.422 + +[118] Tianmai M Zhang and Neil F Abernethy. 2024. Detecting Reference Errors in Sci- +entific Literature with Large Language Models. arXiv preprint arXiv:2411.06101 +(2024). + +[119] Zhiwei Zhang, Jiyi Li, Fumiyo Fukumoto, and Yanming Ye. 2021. Abstract, Ra- +tionale, Stance: A Joint Model for Scientific Claim Verification. In Proceedings of +the 2021 Conference on Empirical Methods in Natural Language Processing, Marie- +Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (Eds.). +Association for Computational Linguistics, Online and Punta Cana, Dominican +Republic, 3580ś3586. doi:10.18653/v1/2021.emnlp-main.290 + +[120] Liwen Zheng, Chaozhuo Li, Xi Zhang, Yu-Ming Shang, Feiran Huang, and +Haoran Jia. 2024. Evidence Retrieval is almost All You Need for Fact Verifi- +cation. In Findings of the Association for Computational Linguistics ACL 2024, +Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). Association for Com- +putational Linguistics, Bangkok, Thailand and virtual meeting, 9274ś9281. +https://aclanthology.org/2024.findings-acl.551 + +(Eds.). Association for Computational Linguistics, Toronto, Canada, 6215ś6230. +doi:10.18653/v1/2023.findings-acl.387 + +[98] Juraj Vladika and Florian Matthes. 2024. Comparing Knowledge Sources for +Open-Domain Scientific Claim Verification. In Proceedings of the 18th Conference +of the European Chapter of the Association for Computational Linguistics (Volume +1: Long Papers), Yvette Graham and Matthew Purver (Eds.). Association for +Computational Linguistics, St. Julian’s, Malta, 2103ś2114. https://aclanthology. +org/2024.eacl-long.128/ + +[99] Juraj Vladika and Florian Matthes. 2024. Improving Health Question Answering +with Reliable and Time-Aware Evidence Retrieval. In Findings of the Association +for Computational Linguistics: NAACL 2024, Kevin Duh, Helena Gomez, and +Steven Bethard (Eds.). Association for Computational Linguistics, Mexico City, +Mexico, 4752ś4763. doi:10.18653/v1/2024.findings-naacl.295 + +[100] David Wadden, Shanchuan Lin, Kyle Lo, Lucy Lu Wang, Madeleine van Zuylen, +Arman Cohan, and Hannaneh Hajishirzi. 2020. Fact or Fiction: Verifying Sci- +entific Claims. In Proceedings of the 2020 Conference on Empirical Methods in +Natural Language Processing (EMNLP), Bonnie Webber, Trevor Cohn, Yulan +He, and Yang Liu (Eds.). Association for Computational Linguistics, Online, +7534ś7550. doi:10.18653/v1/2020.emnlp-main.609 + +[101] David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Iz Beltagy, Lucy Lu Wang, +and Hannaneh Hajishirzi. 2022. SciFact-Open: Towards open-domain scientific +claim verification. In Findings of the Association for Computational Linguistics: +EMNLP 2022, Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang (Eds.). As- +sociation for Computational Linguistics, Abu Dhabi, United Arab Emirates, +4719ś4734. doi:10.18653/v1/2022.findings-emnlp.347 + +[102] David Wadden, Kyle Lo, Lucy Lu Wang, Arman Cohan, Iz Beltagy, and Han- +naneh Hajishirzi. 2022. MultiVerS: Improving scientific claim verification with +weak supervision and full-document context. In Findings of the Association +for Computational Linguistics: NAACL 2022, Marine Carpuat, Marie-Catherine +de Marneffe, and Ivan Vladimir Meza Ruiz (Eds.). Association for Computational +Linguistics, Seattle, United States, 61ś76. doi:10.18653/v1/2022.findings-naacl.6 +[103] Dong Wang, Jianxin Li, Tianchen Zhu, Haoyi Zhou, Qishan Zhu, Yuxin Wen, +and Hongming Piao. 2022. MtCut: A Multi-Task Framework for Ranked List +Truncation. In Proceedings of the Fifteenth ACM International Conference on Web +Search and Data Mining. 1054ś1062. + +[104] Gengyu Wang, Kate Harwood, Lawrence Chillrud, Amith Ananthram, Melanie +Subbiah, and Kathleen McKeown. 2023. Check-COVID: Fact-Checking COVID- +19 News Claims with Scientific Evidence. In Findings of the Association for +Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 14114ś14127. doi:10.18653/v1/2023.findings-acl.888 + +[105] Yuxia Wang, Revanth Gangi Reddy, Zain Muhammad Mujahid, Arnav Arora, +Aleksandr Rubashevskii, Jiahui Geng, Osama Mohammed Afzal, Liangming +Pan, Nadav Borenstein, Aditya Pillai, Isabelle Augenstein, Iryna Gurevych, and +Preslav Nakov. 2024. Factcheck-Bench: Fine-Grained Evaluation Benchmark +for Automatic Fact-checkers. In Findings of the Association for Computational +Linguistics: EMNLP 2024, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen +(Eds.). Association for Computational Linguistics, Miami, Florida, USA, 14199ś +14230. doi:10.18653/v1/2024.findings-emnlp.830 + +[106] Zhiruo Wang, Jun Araki, Zhengbao Jiang, Md Rizwan Parvez, and Graham +Neubig. 2023. Learning to filter context for retrieval-augmented generation. +arXiv preprint arXiv:2311.08377 (2023). + +[107] Jevin D West and Carl T Bergstrom. 2021. Misinformation in and about science. + +Proceedings of the National Academy of Sciences 118, 15 (2021), e1912444117. + +[108] Dustin Wright and Isabelle Augenstein. 2021. CiteWorth: Cite-Worthiness +Detection for Improved Scientific Document Understanding. In Findings of the +Association for Computational Linguistics: ACL-IJCNLP 2021, Chengqing Zong, +Fei Xia, Wenjie Li, and Roberto Navigli (Eds.). Association for Computational +Linguistics, Online, 1796ś1807. doi:10.18653/v1/2021.findings-acl.157 +[109] Chen Wu, Ruqing Zhang, Jiafeng Guo, Yixing Fan, Yanyan Lan, and Xueqi +Cheng. 2021. Learning to truncate ranked lists for information retrieval. In +Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4453ś4461. +[110] Siwei Wu, Yizhi Li, Kang Zhu, Ge Zhang, Yiming Liang, Kaijing Ma, Cheng- +hao Xiao, Haoran Zhang, Bohao Yang, Wenhu Chen, Wenhao Huang, Noura +Al Moubayed, Jie Fu, and Chenghua Lin. 2024. SciMMIR: Benchmarking Sci- +entific Multi-modal Information Retrieval. In Findings of the Association for +Computational Linguistics: ACL 2024, Lun-Wei Ku, Andre Martins, and Vivek +Srikumar (Eds.). Association for Computational Linguistics, Bangkok, Thailand, +12560ś12574. doi:10.18653/v1/2024.findings-acl.746 + +[111] Amelie Wührl and Roman Klinger. 2022. Entity-based Claim Representation +Improves Fact-Checking of Medical Content in Tweets. In Proceedings of the 9th +Workshop on Argument Mining, Gabriella Lapesa, Jodi Schneider, Yohan Jo, and +Sougata Saha (Eds.). International Conference on Computational Linguistics, +Online and in Gyeongju, Republic of Korea, 187ś198. https://aclanthology.org/ +2022.argmining-1.18/ + +[112] Barry Menglong Yao, Aditya Shah, Lichao Sun, Jin-Hee Cho, and Lifu Huang. +2023. End-to-end multimodal fact-checking and explanation generation: A +challenging dataset and models. In Proceedings of the 46th International ACM + +448 diff --git a/q135/random_k2/random_2.md b/q135/random_k2/random_2.md index e2f4690064e5381fbfbef3012bd5ce5c92984a53..fd494fa8bcb0b303627f360573c5bd907a96e925 100644 --- a/q135/random_k2/random_2.md +++ b/q135/random_k2/random_2.md @@ -1,638 +1,3081 @@ -Electronic Arts Reports Q4 and Full Year FY19 Financial Results -REDWOOD CITY, CA – May 7, 2019 – Electronic Arts Inc. (NASDAQ: EA) today announced -preliminary financial results for its fourth fiscal quarter and fiscal year ended March 31, 2019. -News and ongoing updates regarding EA and its games are available on EA’s blog at -www.ea.com/news. -“FY20 will be a year of continued growth for our industry, and we’re excited to bring more great -entertainment to the expanding audience around the world,” said CEO Andrew Wilson. “Players are -engaging with games in more ways than ever before. We’re committed to meeting them where they -are with a broad portfolio of amazing new games and live services, choice of engagement models -including free-to-play and subscriptions, and new opportunities to play, compete and watch.” -“We ended our fiscal year with strong growth, driven by our live services business which continued to -provide a very large, stable and profitable source of net bookings,” said COO and CFO Blake -Jorgensen. “Looking ahead, we will bring games to an extraordinary variety of players in fiscal 2020 -and deliver on our commitment to players with deep live services content for Ultimate Team, Apex -Legends, The Sims 4, Anthem and Battlefield V.” -Selected Operating Highlights and Metrics -• Digital net bookings* for the trailing twelve months was $3.722 billion, up 5% year-over-year -and represents 75% of total net bookings. -• During the quarter, launched two new original IP titles, Apex LegendsTM and AnthemTM. -• During the quarter, launched Firestorm battle royale in Battlefield TM V, the biggest -BattlefieldTM live service event ever. -• In FY19, more than 45 million unique players engaged in FIFA 19 and FIFA 18 on console -and PC. -• In FY19, active players in The Sims TM 4 increased by 35% year-over-year. -* Net bookings is defined as the net amount of products and services sold digitally or sold-in -physically in the period. Net bookings is calculated by adding total net revenue to the change in -deferred net revenue for online-enabled games and, for periods after the fourth quarter of fiscal 2018, -mobile platform fees. -Selected Financial Highlights and Metrics -All financial measures are presented on a GAAP basis. -• Net cash from operating activities was $599 million for the quarter and $1.547 billion for the -trailing twelve months. -• EA repurchased 3.2 million shares for $301 million during the quarter and 11.0 million shares -for $1.192 billion during the trailing twelve months. -Impact of Recently Adopted Accounting Standard -At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), -Revenue from Contracts with Customers. For more information about the adoption of Topic 606, -please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April -1, 2018 has not been restated. - -Quarterly Financial Highlights -Three Months Ended -March 31, -| | | 2019 | 2018 | | | -| --- | --- | ---- | ---- | --- | --- | -(in $ millions, except per share amounts) -| Digital net revenue | | 1,065 | 1,102 | | | -| ------------------------------------ | --- | ------ | ------ | --- | --- | -| Packaged goods and other net revenue | | 173 | 480 | | | -| Total net revenue | | 1,238 | 1,582 | | | - -| Net income | | 209 | 607 | | | -| -------------------------- | --- | ----- | ----- | --- | --- | -| Diluted earnings per share | | 0.69 | 1.95 | | | - -| Operating cash flow | | 599 | 615 | | | -| ---------------------------- | --- | ---- | ----- | --- | --- | -| Value of shares repurchased | | 301 | 148 | | | -| Number of shares repurchased | | 3.2 | 1.2 | | | - -The following GAAP-based financial data and tax rate of 18% was used internally by company -management to adjust its fiscal 2019 GAAP results in order to assess EA’s operating results: - -Three Months Ended March 31, 2019 -GAAP-Based Financial Data -Change in -deferred -net revenue -| | Statement | Acquisition- | (online- | Mobile | | -| --------------- | ---------- | ------------ | -------- | --------- | ------------ | -| | of | related | enabled | platform | Stock-based | -| (in $ millions) | Operations | expenses | games) | fees | compensation | -Total net revenue 1 ,238 - 1 73 ( 47) - - Cost of revenue 2 76 ( 1) - ( 47) ( 1) -Gross profit 9 62 1 1 73 - 1 - Total operating expenses 7 66 ( 16) - - ( 72) -Operating income 1 96 1 7 1 73 - 7 3 -Interest and other income, net 2 3 - - - - -Income before provision for income taxes 2 19 1 7 1 73 - 7 3 -Number of shares used in computation: -| Diluted | | 301 | | | | -| ------------ | --- | --- | --- | --- | --- | - -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. +Direct Connect + +Service Overview + +Issue + +Date + +01 + +2026-02-27 + +HUAWEI CLOUD COMPUTING TECHNOLOGIES CO., LTD. + + Copyright © Huawei Cloud Computing Technologies Co., Ltd. 2026. All rights reserved. + +No part of this document may be reproduced or transmitted in any form or by any means without prior +written consent of Huawei Cloud Computing Technologies Co., Ltd. + +Trademarks and Permissions + + and other Huawei trademarks are the property of Huawei Technologies Co., Ltd. + +All other trademarks and trade names mentioned in this document are the property of their respective +holders. + +Notice + +The purchased products, services and features are stipulated by the contract made between Huawei +Cloud and the customer. All or part of the products, services and features described in this document may +not be within the purchase scope or the usage scope. Unless otherwise specified in the contract, all +statements, information, and recommendations in this document are provided "AS IS" without +warranties, guarantees or representations of any kind, either express or implied. + +The information in this document is subject to change without notice. Every effort has been made in the +preparation of this document to ensure accuracy of the contents, but all statements, information, and +recommendations in this document do not constitute a warranty of any kind, express or implied. + +Huawei Cloud Computing Technologies Co., Ltd. + +Address: + +Huawei Cloud Data Center Jiaoxinggong Road +Qianzhong Avenue +Gui'an New District +Gui Zhou 550029 +People's Republic of China + +Website: + +https://www.huaweicloud.com/intl/en-us/ + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +i + + Direct Connect +Service Overview + +Contents + +Contents + +1 What Is Direct Connect?........................................................................................................ 1 + +2 Product Advantages................................................................................................................ 3 + +3 Application Scenarios............................................................................................................. 4 + +4 Functions................................................................................................................................... 6 + +5 Direct Connect Locations.....................................................................................................10 + +6 Security.................................................................................................................................... 15 +6.1 Identity Authentication and Access Control................................................................................................................ 15 + +6.2 Auditing and Logging.......................................................................................................................................................... 16 + +6.3 Monitoring Security Risks.................................................................................................................................................. 16 + +6.4 Certificates...............................................................................................................................................................................16 + +7 Permissions............................................................................................................................. 18 + +8 Notes and Constraints.......................................................................................................... 26 + +9 Direct Connect and Other Services................................................................................... 29 + +10 Basic Concepts..................................................................................................................... 31 + +11 Region and AZ......................................................................................................................34 + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +ii + + Direct Connect +Service Overview + +1 What Is Direct Connect? + +1 What Is Direct Connect? + +Direct Connect allows you to establish a stable, high-speed, low-latency, secure +dedicated network connection that connects your on-premises data center to +Huawei Cloud. Direct Connect allows you to maximize legacy IT facilities and +leverage cloud services to build a flexible, scalable hybrid cloud computing +environment. + +Figure 1-1 shows how Direct Connect connects an on-premises data center to the +cloud. + +Figure 1-1 How Direct Connect works + +Why Direct Connect? + +● Network quality: Direct Connect allows you to establish a dedicated network + +for data transmission, which brings high network performance, low latency, +and excellent user experience. + +● + +● + +Security: Direct Connect establishes private connectivity between your on- +premises data center and the cloud. Data is transmitted over a secure +dedicated connection. + +Transmission speed: A single connection supports up to 100 Gbit/s of +bandwidth. + +Components + +● + +Connection + +A dedicated network connection is established between your on-premises +data center and a Direct Connect location over a line you leased from a +carrier. You can create a standard connection by yourself or request a hosted +connection from a partner. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +1 + + Direct Connect +Service Overview + +1 What Is Direct Connect? + +– + +– + +A standard connection has a dedicated port for your exclusive use and +can be associated with multiple virtual interfaces. + +A hosted connection is created by a carrier (a partner of Huawei Cloud), +who allocates VLAN and bandwidth resources for you. When using this +type of connections, you share a port with other users, and can create +only one virtual interface for a connection. + +● + +Virtual gateway + +A virtual gateway is a logical gateway for accessing VPCs. A virtual gateway +can be associated with only one VPC, and multiple connections can access a +VPC through the same virtual gateway. + +● Global DC gateway + +A global DC gateway enables your on-premises data center to access VPCs in +multiple regions so you can use a single connection to provide high-speed +access to cloud compute and storage resources in any region. + +A global DC gateway can only be associated with connections terminated at +the same Direct Connect location. If there are multiple connections +terminated at different Direct Connect locations, you need to create multiple +global DC gateways. + +● + +Virtual interface + +A virtual interface links a connection with one or more virtual gateways, so +that your on-premises network can access all the VPCs. + +Accessing Direct Connect + +The public cloud provides a web-based user interface, the management console, +for you to access the Direct Connect service. + +● + +● + +If you already have an account, log in to the management console and +choose Networking > Direct Connect on the homepage. + +If you do not have an account, you can sign up for one by referring to +Getting Started. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. 2 -TTM Financial Highlights -Twelve Months Ended -March 31, -| | 2019 | | 2018 | | | | -| --- | ---- | --- | ---- | --- | --- | --- | -(in $ millions, except per share amounts) -| Digital net revenue | | 3,710 | 3,450 | | | | -| ------------------------------------ | --- | ------ | ------ | --- | --- | --- | -| Packaged goods and other net revenue | | 1,240 | 1,700 | | | | -| Total net revenue | | 4,950 | 5,150 | | | | - -| Net income | | 1,019 | 1,043* | | | | -| ---------------------------- | ----------------------- | ------ | ------ | --- | --- | --- | -| Diluted earnings per share | | 3.33 | 3.34* | | | | -| Operating cash flow | | 1,547 | 1,692 | | | | -| Value of shares repurchased | 1,192 | | 601 | | | | -| Number of shares repurchased | | 11.0 | 5.3 | | | | -*During the twelve months ended March 31, 2018, EA recognized $235 million of incremental income tax -expense, or approximately $0.75 per share, due to the application of the Tax Cuts and Jobs Act. - -The following GAAP-based financial data and tax rate of 18% was used internally by company -management to adjust its fiscal 2019 GAAP results in order to assess EA’s operating results: - -Twelve Months Ended March 31, 2019 -GAAP-Based Financial Data -Change in -deferred net -revenue -| | Statement | Acquisition- | (online- | Mobile | | | -| --------------- | ---------- | ------------ | -------- | --------- | ------------ | --- | -| | of | related | enabled | platform | Stock-based | | -| (in $ millions) | Operations | expenses | games) | fees | compensation | | -Total net revenue 4,950 - 182 (188) - - Cost of revenue 1,322 (4) - (188) (4) -Gross profit 3,628 4 182 - 4 - Total operating expenses 2,632 (37) - - (280) -Operating income 996 41 182 - 2 84 -Interest and other income, net 83 - - - - -Income before provision for income taxes 1,079 41 182 - 2 84 -Number of shares used in computation: -| Diluted | 306 | | | | | | -| ------------ | --- | --- | --- | --- | --- | --- | -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. + Direct Connect +Service Overview + +2 Product Advantages + +2 Product Advantages + +Direct Connect has the following advantages: + +● Dedicated Channels to Safeguard Data Security + +– + +– + +Establish private connectivity between your on-premises network and the +cloud for stable data transmission without compromising privacy. + +Secure your data using encrypted, dedicated connections, which provide +higher security and more reliability. + +● Multiple Access Options + +– + +– + +You can select from optical bare fibers and other types of leased lines, +such as MSTP and MPLS VPN. + +There are a number of Direct Connect locations around the globe, so you +can access the cloud from anywhere nearby. + +● Great Scalability for Flexible Service Deployment + +– + +– + +By building a hybrid cloud, you can gain access to virtually unlimited +cloud resources for flexible, scalable deployment. + +You can run workloads that span on-premises infrastructure and different +cloud regions and flexibly connect networks across regions. + +● Ultra-low Latency and Stable Transmission + +– + +– + +Enjoy faster speeds with up to 100 Gbit/s of bandwidth, supporting +diverse service needs. + +Configure your connections in active/active or active/standby pairs for +higher reliability. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. 3 -Operating Metric -The following is a calculation of our total net bookings for the periods presented: -Three Months Ended TTM Ended -March 31, March 31, -2019 2018 2019 2018 -(in $ millions) -Total net revenue 1,238 1,582 4,950 5,150 -Change in deferred net revenue (online-enabled games) 173 (327) 182 30 -Mobile platform fees (47) - (188) - -Net bookings 1,364 1,255 4,944 5,180 -Business Outlook as of May 7, 2019 -The following forward-looking statements reflect expectations as of May 7, 2019. Electronic Arts -assumes no obligation to update these statements, except as required by law. Results may be -materially different and are affected by many factors detailed in this release and in EA’s annual and -quarterly SEC filings. -Fiscal Year 2020 Expectations – Ending March 31, 2020 -Financial metrics: -• Net revenue is expected to be approximately $5.375 billion. -Change in deferred net revenue (online-enabled games) is expected to be -o -approximately $(125) million. -Mobile platform fees are expected to be approximately $(150) million. -o -• Net income is expected to be approximately $2.595 billion. This includes a one-time tax -benefit of approximately $1.5 billion (see below). -• Diluted earnings per share is expected to be approximately $8.56. This includes a one- -time tax benefit of approximately $5 per share (see below). -• Operating cash flow is expected to be approximately $1.575 billion. -• The Company estimates a share count of 303 million for purposes of calculating fiscal -year 2020 diluted earnings per share. -Subsequent to the fiscal year ended March 31, 2019, EA completed an internal transfer of some of its -intellectual property rights to its Swiss subsidiary, where EA’s international business is -headquartered. This transfer will result in a one-time tax benefit of between $1.5 billion and $3.9 -billion, which will be recognized as a deferred tax asset in EA’s consolidated financial statements for -the three months ending June 30, 2019. The recognition of the deferred tax asset will add between $5 -to $13 per share to GAAP EPS and the low end of this range has been reflected above in EA’s -forecast. The forecast of this one-time tax benefit is preliminary and is dependent on the results of -Swiss referendums to be voted on later this month. -Operational metric: -• Net bookings is expected to be approximately $5.100 billion. -In addition, the following outlook for GAAP-based financial data and a long-term tax rate of 18% are -used internally by EA to adjust our GAAP expectations to assess EA’s operating results and plan for -future periods: + Direct Connect +Service Overview + +3 Application Scenarios + +3 Application Scenarios + +Access to Multiple VPCs from an On-premises Data Center + +After you connect your on-premises data center to the cloud using Direct Connect, +you can use Cloud Connect to connect the VPC that your on-premises data center +is accessing to those in other regions, so that your on-premises data center can +access all connected VPCs. + +Hybrid Cloud Deployment + +Direct Connect allows you to build a hybrid environment for your on-premises +data center and leverage the scalability of the cloud to expand the computing +capability of your applications. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 4 -Twelve Months Ending March 31, 2020 -GAAP-Based Financial Data -Change in -deferred net -revenue -Acquisition- (online- -GAAP related enabled Mobile Stock-based -Guidance expenses games) platform fees compensation -(in $ millions) -Digital net revenue 4 ,175 - ( 75) ( 150) - -Packaged goods & other net revenue 1 ,200 - ( 50) - - -Total net revenue 5 ,375 - ( 125) ( 150) - -Cost of revenue 1 ,327 ( 12) - ( 150) ( 2) -Operating expense 2 ,746 ( 23) - - ( 353) -Income before provision for income taxes 1 ,361 3 5 ( 125) - 3 55 -Net income 2 ,595 -Number of shares used in computation: -Diluted shares 3 03 -First Quarter Fiscal Year 2020 Expectations – Ending June 30, 2019 -Financial metrics: -• Net revenue is expected to be approximately $1.130 billion. -Change in deferred net revenue (online-enabled games) is expected to be -o -approximately $(405) million. -Mobile platform fees are expected to be approximately $(35) million. -o -• Net income is expected to be approximately $1.765 billion. This includes a one-time tax -benefit of approximately $1.5 billion (see above). -• Diluted earnings per share is expected to be approximately $5.84. This includes a one- -time tax benefit of approximately $5 per share (see above). -• The Company estimates a share count of 302 million for purposes of calculating first -quarter fiscal year 2020 diluted earnings per share. Should the first quarter be a loss, the -Company will use the basic share count of 299 million for purposes of calculating first -quarter fiscal year 2020 earnings per share. -Operational metric: -• Net bookings is expected to be approximately $690 million. -In addition, the following outlook for GAAP-based financial data and a long-term tax rate of 18% are -used internally by EA to adjust our GAAP expectations to assess EA’s operating results and plan for -future periods: + Direct Connect +Service Overview + +Figure 3-1 Hybrid cloud + +3 Application Scenarios + +Table 3-1 Comparisons of Direct Connect and VPN in hybrid cloud deployment + +Scenario + +Description + +Helpful Links + +Cloud +Service + +Virtual +Private +Networ +k (VPN) + +Connect an on- +premises data +center to the +cloud through +an IPsec tunnel. + +Direct +Connect + +Connect an on- +premises data +center to the +cloud using a +dedicated +network +connection. + +VPN uses an encrypted +communications tunnel to +connect a VPC on the cloud +to an on-premises data +center and sends traffic over +the Internet. It is +inexpensive, easy to +configure, and easy to use. +However, VPN connections +may be affected by the +Internet quality. + +Direct Connect provides +physical connections +between VPCs and data +centers. It has the +advantages of low latency +and is very secure. Direct +Connect is a good choice +when there are strict +requirements on network +transmission quality. + +Connecting to a +VPC Through a +VPN + +What Is an +Enterprise Switch? + +Connecting to +Multiple VPCs that +Do Not Need to +Communicate with +Each Other + +What Is an +Enterprise Switch? + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 5 -Three Months Ending June 30, 2019 -GAAP-Based Financial Data -Change in -deferred net -revenue -Acquisition- (online- Mobile -GAAP related enabled platform Stock-based -Guidance expenses games) fees compensation -(in $ millions) -Total net revenue 1 ,130 - ( 405) ( 35) - -Cost of revenue 1 88 ( 2) - ( 35) - -Operating expense 6 50 ( 5) - - ( 70) -Income before provision for income taxes 3 10 7 ( 405) - 7 0 -Net income 1 ,765 -Number of shares used in computation: -Diluted shares 3 02 -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. -Conference Call and Supporting Documents -Electronic Arts will host a conference call on Tuesday, May 7, 2019 at 2:00 pm PT (5:00 pm ET) to -review its results for the fiscal fourth quarter and fiscal year ended March 31, 2019 and its outlook for -the future. During the course of the call, Electronic Arts may disclose material developments affecting -its business and/or financial performance. Listeners may access the conference call live through the -following dial-in number (866) 393-4306 (domestic) or (734) 385-2616 (international), using the -Conference ID 2997759 or via webcast at EA’s IR Website at http://ir.ea.com. -EA has posted a slide presentation, a financial model of EA’s historical results and guidance, and an -Investor Accounting FAQ on EA’s IR Website. After the conference call, EA will also post the -prepared remarks and a transcript from the conference call on EA’s IR Website. -A dial-in replay of the conference call will be available until May 21, 2019 at (855) 859-2056 -(domestic) or (404) 537-3406 (international) using pin code 2997759. An audio webcast replay of the -conference call will be available for one year on EA’s IR Website. -Forward-Looking Statements -Some statements set forth in this release, including the information relating to EA’s fiscal 2020 -expectations under the heading “Business Outlook as of May 7, 2019,” and other information -regarding EA's fiscal 2020 expectations contain forward-looking statements that are subject to -change. Statements including words such as “anticipate,” “believe,” “expect,” “intend,” “estimate”, -“plan”, “predict”, “seek”, “goal”, “will”, “may”, “likely”, “should”, “could” (and the negative of any of -these terms), “future” and similar expressions also identify forward-looking statements. These -forward-looking statements are not guarantees of future performance and reflect management’s -current expectations. Our actual results could differ materially from those discussed in the forward- -looking statements. + Direct Connect +Service Overview + +4 Functions + +4 Functions + +Standard Connection + +A standard connection provides an exclusive port. You can create standard +connections on the management console. To improve reliability, you can create +multiple connections terminated at different locations and allow them to serve as +backups for each other. If you can select only one carrier due to special +requirements, you must configure different physical routes for your connections. + +You can choose self-service connections or full-service connections. + +● + +● + +Self-service connection: Huawei Cloud only provides the port. You need to +create a connection on the console, and lease a line from a carrier. +Full-service connection: You only need to create a connection on the console, +and Huawei Cloud will complete all operations required for network +connectivity. + +Figure 4-1 shows an example of how standard connections work. + +Figure 4-1 Accessing the cloud using standard connections + +For details, see Creating a Connection. + +Hosted Connection + +If you use a hosted connection to access the cloud, the port you use is shared with +other users. You request a carrier (a partner of Huawei Cloud) to provide a + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 6 -Some of the factors which could cause the Company’s results to differ materially from its expectations -include the following: sales of the Company’s products and services; the Company’s ability to develop -and support digital products and services, including managing online security and privacy; outages of -our products, services or technological infrastructure; the Company’s ability to manage expenses; the -competition in the interactive entertainment industry; governmental regulations; the effectiveness of -the Company’s sales and marketing programs; timely development and release of the Company’s -products and services; the Company’s ability to realize the anticipated benefits of acquisitions; the -consumer demand for, and the availability of an adequate supply of console hardware units; the -Company’s ability to predict consumer preferences among competing platforms; the Company’s -ability to develop and implement new technology; foreign currency exchange rate fluctuations; -general economic conditions; changes in our tax rates or tax laws; and other factors described in Part -II, Item 1A of Electronic Arts’ latest Quarterly Report on Form 10-Q under the heading “Risk Factors”, -as well as in other documents we have filed with the Securities and Exchange Commission, including -our Annual Report on Form 10-K for the fiscal year ended March 31, 2018. -These forward-looking statements are valid as of May 7, 2019. Electronic Arts assumes no obligation -to revise or update any forward-looking statement for any reason, except as required by law. In -addition, the preliminary financial results set forth in this release are estimates based on information -currently available to Electronic Arts. -While Electronic Arts believes these estimates are meaningful, they could differ from the actual -amounts that Electronic Arts ultimately reports in its Annual Report on Form 10-K for the fiscal year -ended March 31, 2019. Electronic Arts assumes no obligation and does not intend to update these -estimates prior to filing its Form 10-K for the fiscal year ended March 31, 2019. -About Electronic Arts -Electronic Arts (NASDAQ: EA) is a global leader in digital interactive entertainment. The Company -develops and delivers games, content and online services for Internet-connected consoles, mobile -devices and personal computers. -In fiscal year 2019, EA posted GAAP net revenue of $4.95 billion. Headquartered in Redwood City, -California, EA is recognized for a portfolio of critically acclaimed, high-quality brands such as The -Sims™, Madden NFL, EA SPORTS™ FIFA, Battlefield™, Need for Speed™, Dragon Age™ and -Plants vs. Zombies™. More information about EA is available at www.ea.com/news. -Apex Legends, Anthem, Battlefield, The Sims, EA SPORTS, Need for Speed, Dragon Age, and -Plants vs. Zombies are trademarks of Electronic Arts Inc. John Madden, NFL and FIFA are the -property of their respective owners and used with permission. -For additional information, please contact: -Chris Evenden John Reseburg -Vice President, Investor Relations Vice President, Corporate Communications -650-628-0255 650-628-3601 -cevenden@ea.com jreseburg@ea.com + Direct Connect +Service Overview + +4 Functions + +connection to connect to your on-premises data center to Huawei Cloud. The +partner has pre-connected dedicated lines terminated at the Direct Connect +locations of Huawei Cloud and allocates a connection port to you after receiving +your request. + +Figure 4-2 shows an example of how hosted connections work. + +Figure 4-2 Accessing the cloud using hosted connections + +The following table compares hosted connections with standard connections. + +Item + +Port + +Recomm +ended +bandwidt +h + +Estimate +d +construct +ion +period + +Parties +involved + +Standard Connection + +Hosted Connection + +Exclusive port + +Shared port + +1 Gbit/s to 100 Gbit/s + +Less than 1 Gbit/s + +Two to three months for local lines, +and three to four months for long- +distance lines + +About one month + +Connection user, leased line carrier, +equipment room provider, and Huawei +Cloud + +Connection user, leased +line carrier, and Huawei +Cloud + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 7 -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Statements of Operations -(in $ millions, except per share data) -| | Three Months Ended | | Twelve Months Ended | | -| --- | ------------------ | ---- | ------------------- | --- | -| | March 31, | | March 31, | | -| | 2019 1 | | 2019 1 | | -| | | 2018 | 2018 | | -Net revenue -| Product | 216 | 757 | 1,593 | 2,586 | -| ----------------- | ----- | ----- | ----- | ----- | -| Service and other | 1,022 | 825 | 3,357 | 2,564 | -| Total net revenue | 1,238 | 1,582 | 4,950 | 5,150 | -Cost of revenue -| Product | 52 | 106 | 517 | 822 | -| --------------------- | --- | ----- | ----- | ----- | -| Service and other | 224 | 127 | 805 | 455 | -| Total cost of revenue | 276 | 233 | 1,322 | 1,277 | -| Gross profit | 962 | 1,349 | 3,628 | 3,873 | -Operating expenses: -| Research and development | 398 | 335 | 1,433 | 1,320 | -| -------------------------------------------- | --- | --- | ----- | ----- | -| Marketing and sales | 229 | 130 | 702 | 641 | -| General and administrative | 123 | 126 | 460 | 469 | -| Acquisition-related contingent consideration | 11 | — | 14 | — | -| Amortization of intangibles | 5 | 5 | 23 | 9 | -| Total operating expenses | 766 | 596 | 2,632 | 2,439 | -| Operating income | 196 | 753 | 996 | 1,434 | -| Interest and other income (expense), net | 23 | 1 | 83 | 15 | -| Income before provision for income taxes | 219 | 754 | 1,079 | 1,449 | -| Provision for income taxes | 10 | 147 | 60 | 406 | -| Net income | 209 | 607 | 1,019 | 1,043 | -Earnings per share -| Basic | 0.70 | 1.98 | 3.36 | 3.39 | -| ------- | ---- | ---- | ---- | ---- | -| Diluted | 0.69 | 1.95 | 3.33 | 3.34 | -Number of shares used in computation -| Basic | 299 | 307 | 303 | 308 | -| ------- | --- | --- | --- | --- | -| Diluted | 301 | 311 | 306 | 312 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. - -Results (in $ millions, except per share data) -The following table reports the variance of the actuals versus our guidance provided on February 5, 2019 for the three months ended March -31, 2019 plus a comparison to the actuals for the three months ended March 31, 2018. -Three Months Ended March 31, -| | 2019 1 | | 2019 1 | | -| --- | ------ | --- | ------ | --- | -2018 -| | Guidance | Variance | Actuals | Actuals | -| --- | -------- | -------- | ------- | ------- | -Net revenue -| Net revenue | 1,163 | 75 | 1,238 | 1,582 | -| ----------- | ----- | --- | ----- | ----- | -GAAP-based financial data -Change in deferred net revenue (online-enabled games) 56 117 173 (327) -| Mobile platform fees | (49) | 2 | (47) | — | -| -------------------- | ---- | --- | ---- | --- | -Cost of revenue -| Cost of revenue | 265 | 11 | 276 | 233 | -| --------------- | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | — | (1) | (1) | (1) | -| ---------------------------- | ---- | --- | ---- | --- | -| Stock-based compensation | — | (1) | (1) | (1) | -| Mobile platform fees | (49) | 2 | (47) | — | -Operating expenses -| Operating expenses | 715 | 51 | 766 | 596 | -| ------------------ | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | (6) | (10) | (16) | (5) | -| ---------------------------- | ---- | ---- | ---- | ---- | -| Stock-based compensation | (76) | 4 | (72) | (68) | -Income before tax -| Income before tax | 195 | 24 | 219 | 754 | -| ----------------- | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | 6 | 11 | 17 | 6 | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games) 56 117 173 (327) -| Mobile platform fees | — | — | — | — | -| -------------------------------------- | --- | --- | --- | --- | -| Stock-based compensation | 76 | (3) | 73 | 69 | -| Tax rate used for management reporting | 18% | | 18% | 21% | -Earnings per share -| Basic | 0.56 | 0.14 | 0.70 | 1.98 | -| ------- | ---- | ---- | ---- | ---- | -| Diluted | 0.56 | 0.13 | 0.69 | 1.95 | -Number of shares -| Basic | 301 | (2) | 299 | 307 | -| ------- | --- | --- | --- | --- | -| Diluted | 303 | (2) | 301 | 311 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Balance Sheets -(in $ millions) -March 31, 20191 March 31, 20182 -ASSETS -Current assets: -| Cash and cash equivalents | 4,708 | 4,258 | -| ------------------------- | ----- | ----- | -| Short-term investments | 737 | 1,073 | -Receivables, net of allowances of $7 and $165, respectively 623 385 -| Other current assets | 313 | 288 | -| ------------------------------------ | ----- | ----- | -| Total current assets | 6,381 | 6,004 | -| Property and equipment, net | 448 | 453 | -| Goodwill | 1,892 | 1,883 | -| Acquisition-related intangibles, net | 87 | 71 | -| Deferred income taxes, net | 35 | 84 | -| Other assets | 114 | 89 | -| TOTAL ASSETS | 8,957 | 8,584 | -LIABILITIES AND STOCKHOLDERS’ EQUITY -Current liabilities: -| Accounts payable | 113 | 48 | -| ------------------------------------------- | ----- | ----- | -| Accrued and other current liabilities | 1,052 | 821 | -| Deferred net revenue (online-enabled games) | 1,100 | 1,622 | -| Total current liabilities | 2,265 | 2,491 | -| Senior notes, net | 994 | 992 | -| Income tax obligations | 233 | 250 | -| Deferred income taxes, net | 2 | 1 | -| Other liabilities | 132 | 255 | -| Total liabilities | 3,626 | 3,989 | -Stockholders’ equity: -| Common stock | 3 | 3 | -| ------------------------------------------ | ----- | ----- | -| Additional paid-in capital | — | 657 | -| Retained earnings | 5,358 | 4,062 | -| Accumulated other comprehensive loss | (30) | (127) | -| Total stockholders’ equity | 5,331 | 4,595 | -| TOTAL LIABILITIES AND STOCKHOLDERS’ EQUITY | 8,957 | 8,584 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606) Revenue from Contracts with Customers. EA’s FY19 -Unaudited Condensed Consolidated Balance Sheet reflects the effect of the adoption as of April 1, 2018, which had an impact on the following: receivables, net -of allowances, accrued and other current liabilities, deferred net revenue (online-enabled games), deferred income taxes, net, retained earnings and accumulated -other comprehensive loss. Financial data for periods prior to April 1, 2018 has not been restated. - 2Derived from audited consolidated financial statements. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Statements of Cash Flows -(in $ millions) -| | | Three Months Ended | | Twelve Months | | -| --- | --- | ------------------ | ---- | --------------- | ---- | -| | | March 31, | | Ended March 31, | | -| | | 2019 | 2018 | 2019 | 2018 | -OPERATING ACTIVITIES -| Net income | | 209 | 607 | 1,019 | 1,043 | -| ---------- | --- | --- | --- | ----- | ----- | -Adjustments to reconcile net income to net cash provided by -operating activities: -| | Depreciation, amortization and accretion | 37 | 39 | 145 | 136 | -| --- | -------------------------------------------- | --- | --- | --- | --- | -| | Acquisition-related contingent consideration | 14 | — | 14 | — | -| | Stock-based compensation | 73 | 69 | 284 | 242 | -Change in assets and liabilities -| | Receivables, net | 183 | 502 | (88) | (25) | -| --- | ------------------------------------------- | ----- | ----- | ---- | ---- | -| | Other assets | (65) | (69) | (24) | 10 | -| | Accounts payable | 24 | (60) | 59 | (44) | -| | Accrued and other liabilities | (120) | (222) | 3 | 43 | -| | Deferred income taxes, net | 73 | 74 | (16) | 204 | -| | Deferred net revenue (online-enabled games) | 171 | (325) | 151 | 83 | -Net cash provided by operating activities 599 615 1,547 1,692 -INVESTING ACTIVITIES -| Capital expenditures | | (35) | (20) | (119) | (107) | -| -------------------- | --- | ---- | ---- | ----- | ----- | -Proceeds from maturities and sales of short-term investments 733 1,510 1,688 3,166 -Purchase of short-term investments (198) (275) (1,342) (2,287) -| Acquisitions, net of cash acquired | | — | — | (58) | (150) | -| ---------------------------------- | ----------------------------------------- | --- | ----- | ---- | ----- | -| | Net cash provided by investing activities | 500 | 1,215 | 169 | 622 | -FINANCING ACTIVITIES -| Proceeds from issuance of common stock | | 25 | 21 | 61 | 78 | -| -------------------------------------- | --- | --- | --- | --- | --- | -Cash paid to taxing authorities for shares withheld from employees (6) (8) (122) (120) -Repurchase and retirement of common stock (301) (148) (1,192) (601) -Net cash used in financing activities (282) (135) (1,253) (643) -Effect of foreign exchange on cash and cash equivalents 4 (3) (13) 22 -| Increase in cash and cash equivalents | | 821 | 1,692 | 450 | 1,693 | -| ------------------------------------- | --- | ----- | ----- | ----- | ----- | -| Beginning cash and cash equivalents | | 3,887 | 2,566 | 4,258 | 2,565 | -| Ending cash and cash equivalents | | 4,708 | 4,258 | 4,708 | 4,258 | - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions, except per share data) -| | Q4 Q11 | Q21 Q31 | Q41 | YOY % | -| --- | --------- | --------- | ---- | ------ | -| | FY18 FY19 | FY19 FY19 | FY19 | Change | -Net revenue -| Net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 | (22%) | -| ------------ | ----------- | ----------- | ----- | ----- | -GAAP-based financial data -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | | -| -------------------- | ------ | --------- | ---- | --- | -Gross profit -| Gross profit | 1,349 922 | 868 876 | 962 | (29%) | -| ------------ | --------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 1 1 | 1 1 | 1 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Mobile platform fees | — — | — — | — | | -| ------------------------------------ | ------- | ------- | --- | --- | -| Stock-based compensation | 1 1 | 1 1 | 1 | | -| Gross profit (as a % of net revenue) | 85% 81% | 67% 68% | 78% | | -Operating income -| Operating income | 753 300 | 258 242 | 196 | (74%) | -| ---------------- | ------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 6 7 | 9 8 | 17 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Stock-based compensation | 69 70 | 66 75 | 73 | | -| ---------------------------------------- | ------- | ------- | --- | --- | -| Operating income (as a % of net revenue) | 48% 26% | 20% 19% | 16% | | -Net income -| Net income | 607 293 | 255 262 | 209 | (66%) | -| ---------- | ------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 6 7 | 9 8 | 17 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Stock-based compensation | 69 70 | 66 75 | 73 | | -| -------------------------------------- | --------- | --------- | ---- | ----- | -| Tax rate used for management reporting | 21% 18% | 18% 18% | 18% | | -| Net income (as a % of net revenue) | 38% 26% | 20% 20% | 17% | | -| Diluted earnings per share | 1.95 0.95 | 0.83 0.86 | 0.69 | (65%) | -Number of diluted shares used in computation -| Basic | 307 306 | 305 302 | 299 | | -| ------- | ------- | ------- | --- | --- | -| Diluted | 311 310 | 307 304 | 301 | | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 significantly -changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including information with -respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, 2018 has not been -restated. -3The change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of cash flows does not necessarily equal the -change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of operations primarily due to the impact of -unrecognized gains/losses on cash flow hedges. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions) -YOY -| | Q4 Q11 | Q21 Q31 | Q41 % | -| --- | --------- | --------- | ----------- | -| | FY18 FY19 | FY19 FY19 | FY19 Change | -QUARTERLY NET REVENUE PRESENTATIONS -Net revenue by composition -| Full game downloads | 232 116 | 148 247 | 169 (27%) | -| ------------------------ | ----------- | ----------- | ----------- | -| Live services | 698 610 | 412 480 | 714 2% | -| Mobile | 172 231 | 220 181 | 182 6% | -| Total digital | 1,102 957 | 780 908 | 1,065 (3%) | -| Packaged goods and other | 480 180 | 506 381 | 173 (64%) | -| Total net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 (22%) | -| Total digital | 70% 84% | 61% 70% | 86% | -| Packaged goods and other | 30% 16% | 39% 30% | 14% | -| Total net revenue % | 100% 100% | 100% 100% | 100% | -GAAP-based financial data -| Full game downloads | (39) (20) | 9 26 | 44 | -| ------------------------ | ----------- | -------- | --- | -| Live services | (19) (160) | (84) 304 | 131 | -| Mobile | 4 (35) | (24) 9 | — | -| Total digital | (54) (215) | (99) 339 | 175 | -| Packaged goods and other | (273) (124) | 79 29 | (2) | -Total change in deferred net revenue (online-enabled games) by composition3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | -| -------------------- | ------ | --------- | ---- | -Net revenue by platform -| Console | 1,196 705 | 917 885 | 826 (31%) | -| ----------------- | ----------- | ----------- | ----------- | -| PC/Browser | 210 197 | 149 217 | 217 3% | -| Mobile | 173 233 | 220 181 | 190 10% | -| Other | 3 2 | — 6 | 5 67% | -| Total net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 (22%) | -GAAP-based financial data -| Console | (313) (288) | 8 323 | 87 | -| ---------- | ----------- | ------- | --- | -| PC/Browser | (20) (9) | (4) 33 | 88 | -| Mobile | 5 (42) | (24) 11 | (2) | -| Other | 1 — | — 1 | — | -Total change in deferred net revenue (online-enabled games) by platform3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | -| -------------------- | ------ | --------- | ---- | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. -3The change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of cash flows does not necessarily equal the -change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of operations primarily due to the impact of -unrecognized gains/losses on cash flow hedges. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions) -| | Q4 Q11 | Q21 Q31 | Q41 YOY % | -| --- | --------- | --------- | ----------- | -| | FY18 FY19 | FY19 FY19 | FY19 Change | -CASH FLOW DATA -| Operating cash flow | 615 120 | (126) 954 | 599 (3%) | -| -------------------------- | ----------- | ----------- | ---------- | -| Operating cash flow - TTM | 1,692 1,636 | 1,458 1,563 | 1,547 (9%) | -| Capital expenditures | 20 32 | 31 21 | 35 75% | -| Capital expenditures - TTM | 107 106 | 107 104 | 119 11% | -Repurchase and retirement of common stock 148 300 299 292 301 103% -DEPRECIATION -| Depreciation expense | 31 30 | 30 30 | 31 — | -| -------------------- | ----- | ----- | ---- | -BALANCE SHEET DATA -| Cash and cash equivalents | 4,258 3,876 | 2,881 3,887 | 4,708 | -| ------------------------- | ----------- | ----------- | ----- | -| Short-term investments | 1,073 1,095 | 1,664 1,274 | 737 | -Cash and cash equivalents, and short-term investments 5,331 4,971 4,545 5,161 5,445 2% -| Receivables, net | 385 371 | 966 806 | 623 62% | -| ---------------- | ------- | ------- | ------- | -STOCK-BASED COMPENSATION -| Cost of revenue | 1 | 1 1 | 1 1 | -| ------------------------------ | ----- | ----- | --- | -| Research and development | 44 47 | 39 49 | 49 | -| Marketing and sales | 8 | 7 9 | 8 9 | -| General and administrative | 16 15 | 17 17 | 14 | -| Total stock-based compensation | 69 70 | 66 75 | 73 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606) Revenue from Contracts with Customers. EA’s FY19 -Unaudited Condensed Consolidated Balance Sheets reflect the effect of the adoption as of April 1, 2018, which had an impact on receivables, net of -allowances. Financial data for periods prior to April 1, 2018 has not been restated. \ No newline at end of file + Direct Connect +Service Overview + +4 Functions + +Item + +Standard Connection + +Hosted Connection + +Procedur +e + +1. You create a connection on the + +console to order a port. + +2. You contact the leased line carrier +and supervise the line deployment +from your on-premises data center +to the equipment room at the +location you choose. + +3. You contact the carrier operating + +the equipment room at the location +you choose to complete the cabling +(inlet wire and jumper) inside the +equipment room. + +4. Your carrier works with Huawei + +Cloud to commission access devices. + +5. You complete required network +configuration on the console, +including creating a virtual gateway +and virtual interface. + +Billing + +● Pay for the port (by month or year) + +to Huawei Cloud. + +● Pay to the equipment room provider +at the location for the cabling inside +the equipment room. + +● Pay to the carrier of your on- + +premises data center for the cabling +inside the equipment room. + +● Pay to the leased line carrier for + +other work and the bandwidth. For +details, see Billing. + +1. The partner deploys the +leased line from your +on-premises data +center to the Direct +Connect location you +select. + +2. The carrier completes +the commissioning of +access devices. + +3. You complete required +network configuration +on the console, +including creating a +virtual gateway and +virtual interface. + +● You do not need to pay +for the port to Huawei +Cloud. + +● Pay to the carrier of + +your on-premises data +center for the cabling +inside the equipment +room. + +● Pay to the leased line +carrier for other work +and the bandwidth. + +For details, see Partner Connections. + +Direct Connect Gateway + +● + +Virtual gateway: A virtual gateway is a logical gateway that enables an on- +premises data center to access a VPC over a connection. To enable an on- +premises data center to access a VPC over a connection, you can associate the +VPC with a virtual gateway. To access other VPCs, you can use VPC Peering or +Cloud Connect to connect the VPC your on-premises data center is accessing +to these VPCs. + +A virtual gateway can only have one VPC associated. An on-premises data +center can access the same VPC over two connections through one virtual +gateway. + +For details, see Virtual Gateways. + +● Global DC gateway: A global DC gateway enables your on-premises data + +center to access VPCs in multiple regions so you can use a single connection + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +8 + + Direct Connect +Service Overview + +4 Functions + +to provide high-speed access to cloud compute and storage resources in any +region. + +A global DC gateway can only be associated with connections terminated at +the same Direct Connect location. If there are multiple connections +terminated at different Direct Connect locations, you need to create multiple +global DC gateways. + +For details, see Global DC Gateways. + +Virtual Interface + +A virtual interface is a point of entry for an on-premises data center to access a +VPC over a connection. A virtual interface associates a connection with a virtual +gateway and connects the virtual gateway to a remote gateway, enabling +communications between the on-premises data center and the VPC. + +Virtual interfaces support static routing and BGP routing. During the +establishment of network connectivity, you can use BGP to connect on-premises +data centers to virtual gateways. BGP helps you build a hybrid cloud more +efficiently, flexibly, and reliably. + +For details, see Virtual Interfaces. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +9 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +5 Direct Connect Locations + +A Direct Connect location provides access to Huawei Cloud in a region. Before +using Direct Connect to access Huawei Cloud, you need to obtain the details about +each location. + +Direct Connect provides a number of locations for you to choose from. You can +request a port by creating a connection. + +For more information, submit a service ticket or contact your Direct Connect +manager. + +Table 5-1 Direct Connect locations in China + +Region + +Location + +CN North-Beijing4 Beijing-Yatai + +Beijing- +Jiuxianqiao + +Carrier + +Neutral + +Dr. Peng + +Beijing-Huitian + +Neutral + +Beijing-Zhongjin + +Neutral + +Address + +Beijing Yatai + +Beijing Universal +Creative Park +(UCP) + +Beijing Huitian +Cloud Industrial +Park + +Centrin Data in +Beijing Economic- +Technological +Development +Area + +Langfang- +Wanguo + +Neutral + +Langfang GDS + +Langfang-Range + +China Telecom + +Langfang-Huawei Huawei + +Langfang Range +Information Port + +Huawei +production base +in Langfang + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +10 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Carrier + +Address + +Wuhan-Wugang + +Neutral + +Wuhan-Huawei + +Huawei + +CN North- +Ulanqab1 + +Ulanqab-Jining + +Huawei + +Ulanqab- +Meiguiying + +Huawei + +CN North3 + +Helin'geer-Nanqu Huawei + +Helin'geer-Beiqu + +Huawei + +Helin'geer-Xiqu + +Huawei + +CN East- +Shanghai1 + +Suzhou-Wanguo + +Neutral + +Wuhan Iron & +Steel Data +Industrial Park + +Wuhan Research +Center + +Ulanqab Huawei +Data Center + +Ulanqab +Meiguiying Base + +Huawei Cloud +Data Center +Nanqu, Helin'geer + +Huawei Cloud +Data Center +Beiqu, Helin'geer + +Huawei Cloud +Data Center Xiqu, +Helin'geer + +Suzhou GDS Data +Center + +Suzhou-Guoke + +China Unicom + +SISDC + +Suzhou-Huawei + +Huawei + +Suzhou-Kunhui + +Neutral + +Suzhou-Wujiang + +China Mobile + +Shanghai- +Wanguo + +Neutral + +Shanghai-Baoxin + +China Mobile + +Shanghai- +Guanghuan + +Hangzhou- +Hanggang + +Hangzhou- +Huawei + +Neutral + +Neutral + +Huawei + +Suzhou Research +Center + +Suzhou Kunhui +Network + +Suzhou Mobile +Data Center + +Shanghai GDS +Data Center + +Shanghai Baoxin +Data Center + +Beijing Sinnet +Data Center + +Hangzhou Iron & +Steel Cloud Data +Center + +Hangzhou +Research Center + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +11 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +CN East- +Shanghai2 + +Location + +Shanghai- +Wanguo + +Shanghai- +Guanghuan + +Carrier + +Neutral + +Neutral + +Shanghai-Baoxin + +China Mobile + +CN East2 + +Wuhu-Jiangbei + +Huawei + +Wuhu-Sanshan + +Huawei + +Wuhu-Zhijiao + +Huawei + +Address + +Shanghai GDS +Data Center + +Beijing Sinnet +Data Center + +Shanghai Baoxin +Data Center + +Wuhu Jiangbei +Data Center + +Wuhu Sanshan +Data Center + +Wuhu Zhijiao +Data Center + +CN East-Qingdao + +Qingdao-Jiaozhou Huawei + +Jiaozhou Ka'aosi + +CN South- +Guangzhou + +Qingdao- +Hongdao + +Guangzhou- +Yunpu + +Guangzhou- +Huaxinyuan + +Guangzhou- +Daxuecheng + +Guangzhou- +Mingmei + +Guangzhou- +Hualong + +China Mobile + +China Telecom + +Neutral + +Neutral + +China Unicom + +China Unicom + +Shenzhen-Baode + +Neutral + +Shenzhen-Futian + +Neutral + +Hongdao Mobile +Data Center + +Guangzhou Yunpu +Data Center + +South China New +Materials +Innovation Park + +Guangzhou +Higher Education +Mega Center + +Guangzhou +Mingmei +Electronics + +Weixin Innovation +Industrial Park + +Shenzhen +Guanlan High- +Tech Park + +Futian Zhongpeng +Cloud + +Shenzhen- +Nanshan + +Neutral + +Nanshan SDGI + +Shenzhen-Yifeng + +Neutral + +Dianda Guyuan +Industrial Park + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +12 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Dongguan- +Tuanbowa + +Carrier + +Huawei + +CN Southwest- +Guiyang1 + +Chengdu-Huawei + +Huawei + +Guiyang-Machang Huawei + +Guiyang-Gui'an + +China Mobile + +Guiyang-Qixinghu Huawei + +Huawei + +China Telecom + +China Mobile + +CN-Hong Kong + +Guiyang- +Gaoduanyuan + +Hong Kong-Sha +Tin + +HK-Sai Kung- +Mobile + +Hong Kong-Sai +Kung-Global +Switch + +HK-Tsuen Wan- +MEGA + +China Telecom + +Global Switch + +Neutral + +MEGA + +Address + +Huawei Base +(Dongguan +Songshan Lake) + +Huawei Chengdu +Research Center + +Gui'an Machang +Data Center + +Gui'an New Area +Electron +Information +Industrial Park + +Gui'an Huawei +Cloud Data +Center + +Guiyang-High-End +Park + +Sun Hung Kai +Logistics Centre + +China Mobile +Global Network +Center + +Table 5-2 Direct Connect locations outside China + +Region + +Location + +AP-Bangkok + +Bangkok-TRUE + +Bangkok-STT + +Bangkok-NTT + +Carrier + +TRUE + +STT + +NTT + +AP-Singapore + +Singapore-Equinix + +Equinix + +Singapore-Global Switch Global Switch + +Singapore-1Net + +Singapore-Singtel + +1-Net + +Singtel + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +13 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Carrier + +AP-Jakarta + +AP-Manila + +Singapore-Telin3 + +Singapore-STT + +Jakarta-DCI + +Jakarta-EDGE + +Manila-GBP + +Telin + +STT + +DCI + +EDGE + +Globe + +AF-Johannesburg + +Nigeria-Lagos-Medallion Medallion + +Manila-Digital Halo + +Digital Halo + +Johannesburg-IS +Parklands + +Internet Solutions +Parklands + +Johannesburg-Teraco + +Teraco + +Johannesburg-DPA + +Vodacom + +Johannesburg-ADC + +Johannesburg-MTN + +AF-Cairo + +Cairo-Xceed + +ADC + +MTN + +Xceed + +LA-Mexico City1 + +Mexico-Ixtlahuaca + +Ixtlahuaca + +LA-Mexico City2 + +Mexico-Tultitlan + +Mexico-Tultitlan + +Mexico-Scala + +LA-Sao Paulo1 + +Sao Paulo-Equinix + +LA-Lima1 + +LA-Santiago + +Sao Paulo-ODATA + +Sao Paulo-Scala + +Sao Paulo-VIVO + +Lima-TIC + +Santiago-Paine + +Santiago-Claro + +TR-Istanbul + +TR-Istanbul + +Istanbul-NGN + +KIO MEX 5 + +KIO MEX 5 + +Scala + +Equinix + +OData + +Scala + +Telefonica + +Telefonica + +Paine + +Claro + +Turkcell + +NGN + +ME-Riyadh + +Riyadh-STC Khurais + +STC Khurais + +Riyadh-Remal + +Remal + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +14 + + Direct Connect +Service Overview + +6 Security + +6 Security + +6.1 Identity Authentication and Access Control + +Identity Authentication + +You can use Identity and Access Management (IAM) to control access to your +Direct Connect resources. IAM permissions define which actions on your cloud +resources are allowed or denied. After creating an IAM user, the administrator +needs to add it to a user group and grant the permissions required by Direct +Connect to the user group. Then, all users in this group automatically inherit the +granted permissions. + +For details, see Permissions. + +Access Control + +A security group is a collection of access control rules for cloud resources, such as +cloud servers, containers, and databases, that have the same security protection +requirements and that are mutually trusted within a VPC. After a security group is +created, you can create various access rules for the security group, and these rules +will apply to all cloud resources added to this security group. + +Your account automatically comes with a default security group. The default +security group allows all outbound traffic, denies all inbound traffic, and allows all +traffic between cloud resources in the group. Your cloud resources in this security +group can communicate with each other already without adding additional rules. +For details about the default security group rules, see Default Security Groups +and Security Group Rules. + +In addition, Huawei Cloud allows you to manage security groups and security +group rules, including + +● + +● + +Creating, viewing, deleting, modifying, cloning, and adding security groups + +Adding, copying, modifying, deleting, importing, and exporting security group +rules + +● Quickly adding multiple security group rules + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +15 + + Direct Connect +Service Overview + +6 Security + +● + +● + +Viewing and changing security groups of ECSs + +Adding cloud resources to or removing cloud resources from security groups + +You can define access control rules for a security group. Then ECSs that will be +added to this security group will be protected. For details, see Security Group. + +6.2 Auditing and Logging + +Cloud Trace Service (CTS) is a log audit service for Huawei Cloud security. It allows +you to collect, store, and query cloud resource operation records. You can use +these records to perform security analysis, audit compliance, track resource +changes, and locate faults. + +After CTS is enabled, traces can be generated for operations performed on the +Direct Connect console. + +● + +● + +● + +For details about how to enable and configure CTS, see Enabling CTS. + +For details about key operations of Direct Connect, see Key Operations +Recorded by CTS. + +For details about traces, see Viewing Traces. + +6.3 Monitoring Security Risks + +Cloud Eye is a monitoring service provided by Huawei Cloud. It provides +capabilities like real-time monitoring, timely alarm reporting, resource groups, and +website monitoring. Cloud Eye enables you to keep track of your resource usages +and service statuses on the cloud, and quickly respond to exceptions to ensure +that services run smoothly. + +Monitoring is critical to ensuring the performance, reliability, and availability of +Direct Connect. Cloud Eye automatically monitors your connections in real time, +collects and displays monitoring data in a convenient, visualized manner, and +allows you to manage alarms and notifications, helping you watch your +connection performance. + +For details about supported metrics and how to create alarm rules, see +Monitoring. + +6.4 Certificates + +Compliance Certificates + +Huawei Cloud services and platforms have obtained various security and +compliance certifications from authoritative organizations, such as International +Organization for Standardization (ISO). You can download them from the +console. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +16 + + Direct Connect +Service Overview + +6 Security + +Figure 6-1 Downloading compliance certificates + +Resource Center + +Huawei Cloud also provides the following resources to help users meet compliance +requirements. For details, see Resource Center. + +Figure 6-2 Resource center + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +17 + + Direct Connect +Service Overview + +7 Permissions + +7 Permissions + +If you need to assign different permissions to employees in your enterprise to +access your Direct Connect resources, Identity and Access Management (IAM) is a +good choice for fine-grained permissions management. IAM provides identity +authentication, permissions management, and access control, helping you securely +access your cloud resources. If your HUAWEI ID does not require IAM for +permissions management, you can skip this section. + +IAM is a free service. You pay only for the resources in your account. + +With IAM, you can control access to specific cloud resources. For example, some +software developers in your enterprise need to use Direct Connect but should not +be allowed to delete other Direct Connect resources or perform any other high- +risk operations. In this scenario, you can create IAM users for the software +developers and grant them only the required permissions. + +IAM supports role/policy-based authorization and identity policy-based +authorization. + +The following table describes the differences between the two authorization +models. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +18 + + Direct Connect +Service Overview + +7 Permissions + +Table 7-1 Differences between role/policy-based and identity policy-based +authorization + +Name Authoriz + +Role/ +Policy + +ation +Using + +User- +permissi +on- +authoriz +ation +scope + +Policies + +Identit +y +policy + +Permissio +ns + +Authorization +Method + +Scenario + +● Syste +m- +define +d roles + +● Syste +m- +define +d +policie +s + +● Custo +m +policie +s + +● Syste +m- +define +d +identit +y +policie +s + +● Custo +m +policie +s + +Assigning +roles or +policies to +principals + +To authorize a user, you need +to add it to a user group first +and then specify the scope of +authorization. It is hard to +provide fine-grained +permissions control using +authorization by user groups +and a limited number of +condition keys. This method is +suitable for small- and +medium-sized enterprises. + +● Assigning +identity +policies to +principals + +● Attaching +identity +policies to +principals + +You can authorize a user by +attaching an identity policy to +it. User-specific authorization +and a variety of key +conditions allow for more +fine-grained permissions +control. However, this model +can be hard to set up. It +requires a certain amount of +expertise and is suitable for +medium- and large-sized +enterprises. + +Assume that you want to grant IAM users the permission to create ECSs in CN +North-Beijing4 and OBS buckets in CN South-Guangzhou. With role/policy-based +authorization, the administrator needs to create two custom policies and attach +both to the IAM users. With identity policy-based authorization, the administrator +only needs to create one custom policy and configure the condition key +g:RequestedRegion for the policy, and then attach the policy to the principals or +grant the principals the access permissions to the specified regions. Identity policy- +based authorization is more flexible than role/policy-based authorization. + +Policies and actions in the two authorization models are not interoperable. You are +advised to use the role/policy-based authorization model. For details about +system-defined permissions, see Role/Policy-based Authorization and Identity +Policy-based Authorization. + +For more information about IAM, see IAM Service Overview. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +19 + + Direct Connect +Service Overview + +Role/Policy-based Authorization + +7 Permissions + +Direct Connect supports role/policy-based authorization. By default, new IAM +users do not have permissions assigned. You need to add a user to one or more +groups, and attach permissions policies or roles to these groups. Users inherit +permissions from the groups to which they are added and can perform specified +operations on cloud services based on the permissions. + +Direct Connect is a project-level service deployed and accessed in specific physical +regions. When you set Scope to Region-specific projects and select the specified +projects (for example, ap-southeast-2) in the specified regions (for example, AP- +Bangkok), the users only have permissions for VPCEP resources in the selected +projects. If you set Scope to All resources, the users have permissions for VPCEP +resources in all region-specific projects. When accessing Direct Connect, the users +need to switch to a region where they have been authorized to use this service. + +Table 7-2 lists all the system-defined policies for Direct Connect in role/policy- +based authorization. System-defined policies in the two authorization models are +not interoperable. + +Table 7-2 Direct Connect permissions + +Role/Policy + +Description + +Direct Connect +Administrator + +DCaaS Partner + +Administrator +permissions for Direct +Connect. Users with +these permissions can +perform all operations +on Direct Connect. + +To have these +permissions, users +must also have the +Tenant Guest and +VPC Administrator +permissions. + +Typ +e + +Syst +em +- +defi +ned +role + +Dependency + +Tenant Guest and VPC +Administrator + +● VPC Administrator: + +project-level policy, which +must be assigned in the +same project + +● Tenant Guest: project- +level policy, which must +be assigned in the same +project + +Direct Connect partner +permissions. Users +who have these +permissions can create +hosted operations for +others. + +Syst +em +- +defi +ned +role + +To have these +permissions, users +must also have the +Tenant Guest and +VPC Administrator +permissions. + +Tenant Guest and VPC +Administrator + +● VPC Administrator: + +project-level policy, which +must be assigned in the +same project + +● Tenant Guest: project- +level policy, which must +be assigned in the same +project + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +20 + + Direct Connect +Service Overview + +7 Permissions + +Role/Policy + +Description + +DCAAS FullAccess Permissions: all + +permissions for Direct +Connect + +Scope: project-level +service + +Permissions: read-only +permissions for Direct +Connect + +Scope: project-level +service + +DCAAS +ReadOnlyAccess + +Dependency + +None + +None + +Typ +e + +Syst +em +- +defi +ned +poli +cy + +Syst +em +- +defi +ned +poli +cy + +Table 7-3 lists common operations supported by each system-defined policy for +Direct Connect. + +Table 7-3 Common operations supported by each system-defined policy + +Operation + +Creating a +connection + +Viewing a +connection + +Modifying a +connection + +Deleting a +connection + +Creating a +virtual +gateway + +Viewing a +virtual +gateway + +Modifying a +virtual +gateway + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +21 + + Direct Connect +Service Overview + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +7 Permissions + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Operation + +Deleting a +virtual +gateway + +Creating a +global DC +gateway + +Viewing a +global DC +gateway + +Modifying a +global DC +gateway + +Deleting a +global DC +gateway + +Creating a +virtual +interface + +Viewing a +virtual +interface + +Modifying a +virtual +interface + +Deleting a +virtual +interface + +Creating an +operations +connection + +Viewing an +operations +connection + +Modifying an +operations +connection + +Deleting an +operations +connection + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +22 + + Direct Connect +Service Overview + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +7 Permissions + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +Operation + +Creating a +hosted +connection + +Viewing a +hosted +connection + +Modifying a +hosted +connection + +Deleting a +hosted +connection + +Roles or Policies that the Direct Connect Console Depends on + +Table 7-4 Roles or policies required for operations on the Direct Connect console + +Function + +Connect +gateways + +Dependent Cloud +Service/Resource + +Elastic IP + +Virtual +interfaces + +Roles or Policies Required + +You need to configure the EIP +ReadOnlyAccess permission when +you perform the following operations +on connect gateways on the Direct +Connect console: + +● In the connect gateway list, view +the location, bind global EIPs to a +connect gateway, and manage the +global EIPs bound to a connect +gateway. + +● On the connect gateway details +page, view all the global EIPs, +binding or unbinding global EIPs to +the connect gateway. + +When deleting a virtual interface on +the Direct Connect console, you need +the EIP ReadOnlyAccess permission. + +Identity Policy-based Authorization + +Direct Connect supports identity policy-based authorization. Table 7-5 lists all the +system-defined policies for Direct Connect with identity policy-based + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +23 + + Direct Connect +Service Overview + +7 Permissions + +authorization. System-defined policies in identity policy-based authorization are +not interoperable with those in role/policy-based authorization. + +Table 7-5 System-defined policies supported by Direct Connect + +Policy Name + +Description + +Policy Type + +DCAASFullAccessPolicy + +Full permissions for Direct +Connect + +System-defined identity +policy + +DCAASReadOnlyPolicy + +Read-only permissions for +Direct Connect + +System-defined identity +policy + +Table 7-6 lists common operations supported by system-defined identity policies +for Direct Connect. + +Table 7-6 Common operations supported by system-defined identity policies + +DCAASFullAccessPolicy + +DCAASReadOnlyPolicy + +Operation + +Creating a +connection + +√ + +Viewing a connection √ + +Modifying a +connection + +Deleting a +connection + +Creating a virtual +gateway + +Viewing a virtual +gateway + +Modifying a virtual +gateway + +Deleting a virtual +gateway + +Creating a global DC +gateway + +Viewing a global DC +gateway + +Modifying a global +DC gateway + +Deleting a global DC +gateway + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +24 + + Direct Connect +Service Overview + +7 Permissions + +Operation + +DCAASFullAccessPolicy + +DCAASReadOnlyPolicy + +Creating a virtual +interface + +Viewing a virtual +interface + +Modifying a virtual +interface + +Deleting a virtual +interface + +Creating an +operations +connection + +Viewing an +operations +connection + +Modifying an +operations +connection + +Deleting an +operations +connection + +Creating a hosted +connection + +Viewing a hosted +connection + +Modifying a hosted +connection + +Deleting a hosted +connection + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Helpful Links + +● + +IAM Service Overview + +● Using IAM to Grant Access to Direct Connect + +● + +Identity Policy-based Authorization + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +25 + + Direct Connect +Service Overview + +8 Notes and Constraints + +8 Notes and Constraints + +Quotas + +Resource + +Default Quota + +How to Increase Quota + +Number of connections +that can be created by +an account in each +region + +Number of global DC +gateways that can be +created by an account in +each region + +Number of virtual +interfaces that can be +created by an account in +each region + +Number of local subnets +for a virtual gateway + +Number of routes for +BGP sessions on a virtual +interface + +Number of remote +subnets for a virtual +interface + +10 + +5 + +50 + +50 + +100 + +50 + +Submit a service ticket. + +Submit a service ticket. + +Submit a service ticket. + +Increasing the quota is +not allowed. + +Submit a service ticket. + +Submit a service ticket. + +Restrictions on Locations + +Before creating a connection, you need to select a location. Note the following +restrictions: + +● + +There may be more than one location in each region. In this case, network +latency from each location to different AZs in the region should be less than 5 +ms. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +26 + + Direct Connect +Service Overview + +8 Notes and Constraints + +● + +If your workloads have demanding requirements for network latency, you can +submit a service ticket to confirm which location is the nearest to the AZ +where your cloud servers are running. + +For details about the locations in each region, see Direct Connect Locations. + +Product Use Restrictions + +● + +● + +● + +● + +The CIDR block of the VPC cannot overlap with the CIDR block used by the +on-premises network. + +The on-premises network cannot use 100.64.0.0/10, 127.0.0.0/8, +169.254.0.0/16, and 224.0.0.0/3 because they are reserved for the VPC service. + +1GE and 10GE single-mode optical ports can transmit data up to 10 km. If +you need an optical port to transmit data for more than 10 km, or you need a +40GE or 100GE port, you need to purchase optical modules by yourself. For +suggestions on optical module selection, submit a service ticket. + +If you use a Direct Connect connection to access ELB, you must select Source +IP hash as the load balancing algorithm and disable sticky sessions for ELB. + +By default, Direct Connect is not interconnected with Enterprise Switch. To +access Enterprise Switch using Direct Connect, submit a service ticket. + +● Direct Connect can respond to common ICMP packets (echo packets whose +type is 8 and code is 0 and do not carry IP options) for ping detections. + +● + +● + +For each connection, a maximum of 30 ping detections can be performed on +the local gateway IP address over the port per second. + +To connect on-premises data centers through enterprise routers, ensure that +the on-premises data centers are connected to different devices. For details, +see submit a service ticket. + +● + +A virtual gateway can be associated with only two different devices. + +Network Requirements + +● + +● + +● + +● + +● + +● + +Your on-premises network must use a single-mode fiber with a 1GE, 10GE, +40GE, or 100GE optical module to connect to the access device on Huawei +Cloud. In addition, key parameters such as the LC, wavelength, and distance +must be consistent with those of the access device on Huawei Cloud. +Examples of optical module parameters: 1 GE, LC single-mode, 1,310 nm, and +10 km + +Auto-negotiation for the ports must be disabled. Port speed and full-duplex +mode must be manually configured. + +802.1Q VLAN encapsulation must be supported on all devices of the entire +connection, including intermediate devices. + +Your device must support Border Gateway Protocol (BGP) and BGP MD5 +authentication or static routing. + +(Optional) You can configure Bidirectional Forwarding Detection (BFD) on the +network. + +A physical connection supports jumbo frames with a default maximum +transmission unit (MTU) size of 8,500 bytes. If you need to configure a larger +MTU, contact the Direct Connect manager. + +If you ping the IP address of the local gateway of a Direct Connect connection +for connectivity test, the MTU is 1,500 bytes. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +27 + + Direct Connect +Service Overview + +8 Notes and Constraints + +● + +Private IP addresses are recommended for cloud and on-premises servers, and +the IP address ranges used for communications cannot overlap. + +Construction Notes + +● + +Your construction company must comply with the regulations presented by +the equipment room provider and engineers. In case of any violation, the +construction cannot be completed. + +● No optical-to-electrical converters can be hosted or installed in the equipment + +room. + +● Network blocking due to state policies or Huawei Cloud management will + +delay the construction. In the event of such situation, contact your Direct +Connect manager. + +● + +● + +The equipment room at a location is operated by a telecom carrier or a third +party. If there are fees for connecting your leased line to the equipment room +or an in-building cable, pay the fees to the equipment room provider. + +You need to apply for a Letter of Authorization (LOA) and present the LOA +when entering the equipment room for construction. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +28 + + Direct Connect +Service Overview + +9 Direct Connect and Other Services + +9 Direct Connect and Other Services + +Figure 9-1 Direct Connect and other services + +Table 9-1 Related services + +Cloud +Service + +Interaction + +Reference + +Virtual +Private Cloud +(VPC) + +Your on-premises data center can +access the VPCs through Direct +Connect. + +Creating a VPC with a +Subnet + +You can use VPC Peering to connect +the VPC your on-premises data center +is accessing to other VPCs in the same +region so that your on-premises data +center can access all these VPCs. + +Connecting an On- +Premises Data Center +to Multiple VPCs in +the Same Region +Using Direct Connect +and VPC Peering + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +29 + + Direct Connect +Service Overview + +9 Direct Connect and Other Services + +Cloud +Service + +Enterprise +Router + +NAT +Gateway + +VPC +Endpoint +(VPCEP) + +Cloud Eye + +Identity and +Access +Management +(IAM) + +Interaction + +Reference + +You can connect your on-premises +data center to an enterprise router, so +that your on-premises data center can +access the VPCs attached to the +enterprise router. + +- + +On-premises servers can share a NAT +gateway to access the Internet or +provide services that are accessible +from the Internet. + +A VPC endpoint can connect your on- +premises data center to a cloud +service through a VPN or Direct +Connect connection over a private +network. + +Cloud Eye monitors Direct Connect +resources and allows you to view +visualized graphs. + +You can grant different permissions +for users to control access to Direct +Connect resources. + +NAT Gateway + +Configuring a VPC +Endpoint for Accessing +the Private IP Address +of OBS + +Viewing Metrics + +Identity and Access +Management + +Cloud Trace +Service (CTS) + +You can record operations performed +on Direct Connect. + +Key Operations +Recorded by CTS + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +30 + + Direct Connect +Service Overview + +10 Basic Concepts + +10 Basic Concepts + +Connection + +A connection is a dedicated channel that connects your on-premises data center +to the cloud. Connections are more stable, reliable, and secure than Internet-based +connections, and provide up to 100 Gbit/s of bandwidth. Direct Connect provides +ports only. After you request a connection, you need to work with the carrier and +Huawei Cloud to establish network connectivity between your on-premises data +center and the cloud. + +Direct Connect provides standard connections and hosted connections. + +● + +Standard connection: A standard connection provides an exclusive port. You +can create standard connections on the management console. To improve +reliability, you can create multiple connections terminated at different +locations and allow them to serve as backups for each other. + +You can choose self-service connections or full-service connections. + +– + +– + +Self-service connection: Huawei Cloud only provides the port. You need to +create a connection on the console, and lease a line from a carrier. + +Full-service connection: You only need to create a connection on the +console, and Huawei Cloud will complete all operations required for +network connectivity. + +Figure 10-1 shows an example of how standard connections work. + +Figure 10-1 Accessing the cloud using standard connections + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +31 + + Direct Connect +Service Overview + +10 Basic Concepts + +● Hosted connection: If you use a hosted connection to access the cloud, the +port you use is shared with other users. These connections are created by a +carrier (a partner of Huawei Cloud), who allocates the required VLAN and +bandwidth for your connection. Only one virtual interface can be associated +with each hosted connection. + +If you are a partner, you can request operations connections. If you are a +common user, you can purchase a host connection from your partner. Hosted +connections must be hosted on operations connections, and your partner will +allocate VLAN and bandwidth resources to the connection. + +Figure 10-2 shows an example of how hosted connections work. + +Figure 10-2 Accessing the cloud using hosted connections + +Connections support redundant configuration. If there are two connections +terminated at different locations in the same region, they can work in an active/ +standby pair to back each other up. If one connection becomes faulty, the other +will take over, ensuring stable services. + +Virtual Gateway + +A virtual gateway is a logical gateway that enables an on-premises data center +to access a VPC over a connection. To enable an on-premises data center to access +a VPC over a connection, you can associate the VPC with a virtual gateway. To +access other VPCs, you can use VPC Peering or Cloud Connect to connect the VPC +your on-premises data center is accessing to these VPCs. + +A virtual gateway can only have one VPC associated. An on-premises data center +can access the same VPC over two connections through one virtual gateway. + +Global DC Gateway + +A global DC gateway enables your on-premises data center to access VPCs in +multiple regions so you can use a single connection to provide high-speed access +to cloud compute and storage resources in any region. + +A global DC gateway can be attached to different enterprise routers to build a +central network so that your on-premises data center can access the VPCs in +different regions over the Huawei backbone network. This reduces network +latency, simplifies network topology, and improves O&M efficiency. + +A global DC gateway can only be associated with connections terminated at the +same Direct Connect location. If there are multiple connections terminated at + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +32 + + Direct Connect +Service Overview + +10 Basic Concepts + +different Direct Connect locations, you need to create multiple global DC +gateways. + +Virtual Interface + +A virtual interface is a point of entry for an on-premises data center to access a +VPC over a connection. A virtual interface associates a connection with a virtual +gateway and connects the virtual gateway to a remote gateway, enabling +communications between the on-premises data center and the VPC. + +Virtual interfaces support static routing and BGP routing. You can use BGP to +connect your on-premises data center to the virtual gateway over a connection. +BGP helps you build a hybrid cloud more efficiently, flexibly, and reliably. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +33 + + Direct Connect +Service Overview + +11 Region and AZ + +11 Region and AZ + +Concept + +A region and availability zone (AZ) identify the location of a data center. You can +create resources in a specific region and AZ. + +● + +● + +Regions are divided based on geographical location and network latency. +Public services, such as Elastic Cloud Server (ECS), Elastic Volume Service +(EVS), Object Storage Service (OBS), Virtual Private Cloud (VPC), Elastic IP +(EIP), and Image Management Service (IMS), are shared within the same +region. Regions are classified into universal regions and dedicated regions. A +universal region provides universal cloud services for common tenants. A +dedicated region provides specific services for specific tenants. + +An AZ contains one or more physical data centers. Each AZ has independent +cooling, fire extinguishing, moisture-proof, and electricity facilities. Within an +AZ, computing, network, storage, and other resources are logically divided +into multiple clusters. + +Figure 11-1 shows the relationship between regions and AZs. + +Figure 11-1 Regions and AZs + +Huawei Cloud provides services in many regions around the world. You can select +a region and an AZ based on requirements. For more information, see Huawei +Cloud Global Products and Services. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +34 + + Direct Connect +Service Overview + +Selecting a Region + +11 Region and AZ + +When selecting a region, consider the following factors: + +● + +Location + +It is recommended that you select the closest region for lower network +latency and quick access. + +– + +– + +– + +If your target users are in Asia Pacific (excluding the Chinese mainland), +select the CN-Hong Kong, AP-Bangkok, or AP-Singapore region. + +If your target users are in Africa, select the AF-Johannesburg region. + +If your target users are in Latin America, select the LA-Santiago region. + +NO TE + +The LA-Santiago region is located in Chile. + +● + +Resource price + +Resource prices may vary in different regions. For details, see Product Pricing +Details. + +Selecting an AZ + +When deploying resources, consider your applications' requirements on disaster +recovery (DR) and network latency. + +● + +● + +For high DR capability, deploy resources in different AZs within the same +region. + +For lower network latency, deploy resources in the same AZ. + +Regions and Endpoints + +Before you use an API to call resources, specify its region and endpoint. For more +details, see Regions and Endpoints. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +35 + diff --git a/q135/random_k4/question.json b/q135/random_k4/question.json index c4b11f9b301fc9096cc563f2b4079b9dc77ec179..b96f249e8cbafb1a9125b79be39223489d585666 100644 --- a/q135/random_k4/question.json +++ b/q135/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "f1f5242528411b262be447e61e2eb10f.pdf", - "web_790f0c62b586634c.pdf", - "web_f8f90e71a8e2a36d.pdf", - "web_b3c7755f9510022a.pdf", - "web_8cd0e969e4cbf86d.pdf" + "acm_3744614_scientific_factchecking.pdf", + "web_d0752a6a77ae08fd.pdf", + "5fdcf120b846746985684fa54250aab44639b30c.pdf", + "web_bdce50920f7f48d8.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q135/random_k4/random_1.md b/q135/random_k4/random_1.md index 926e3cc3fc86ce4552fba63ae8bed9b167e5815e..817dc996b41c5ec14c1f3eee71ff817932a22638 100644 --- a/q135/random_k4/random_1.md +++ b/q135/random_k4/random_1.md @@ -1,224 +1,2076 @@ -PRACTICAL GUIDE FOR -INTERNATIONAL VISITORS +eprints@whiterose.ac.uk +https://eprints.whiterose.ac.uk + +Universities of Leeds, Sheffield and York + +Deposited via The University of Sheffield. + +White Rose Research Online URL for this paper: +https://eprints.whiterose.ac.uk/id/eprint/228448/ + +Version: Published Version + +Proceedings Paper: +Deng, X., Wang, X. and Stevenson, R. (2025) The next phase of scientific fact-checking: +advanced evidence retrieval from complex structured academic papers. In: Zamani, H., +(ed.) ICTIR '25: Proceedings of the 2025 International ACM SIGIR Conference on +Innovative Concepts and Theories in Information Retrieval (ICTIR). ICTIR '25: International +ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval, 18 +Jul 2025, Padua, Italy. ACM. ISBN: 9798400718618. + +https://doi.org/10.1145/3731120.3744614 + +Reuse +This article is distributed under the terms of the Creative Commons Attribution (CC BY) licence. This licence +allows you to distribute, remix, tweak, and build upon the work, even commercially, as long as you credit the +authors for the original work. More information and the full terms of the licence here: +https://creativecommons.org/licenses/ + +Takedown +If you consider content in White Rose Research Online to be in breach of UK law, please notify us by +emailing eprints@whiterose.ac.uk including the URL of the record and the reason for the withdrawal request. + + The Next Phase of Scientific Fact-Checking: Advanced Evidence +Retrieval from Complex Structured Academic Papers + +Xingyu Deng +xdeng37@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Xi Wang +xi.wang@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Mark Stevenson +mark.stevenson@sheffield.ac.uk +University of Sheffield +Sheffield, UK + +Abstract +Scientific fact-checking aims to determine the veracity of scien- +tific claims by retrieving and analysing evidence from research +literature. The problem is inherently more complex than general +fact-checking since it must accommodate the evolving nature of +scientific knowledge, the structural complexity of academic litera- +ture and the challenges posed by long-form, multimodal scientific +expression. However, existing approaches focus on simplified ver- +sions of the problem based on small-scale datasets consisting of +abstracts rather than full papers, thereby avoiding the distinct chal- +lenges associated with processing complete documents. This paper +examines the limitations of current scientific fact-checking systems +and reveals the many potential features and resources that could be +exploited to advance their performance. It identifies key research +challenges within evidence retrieval, including (1) evidence-driven +retrieval that addresses semantic limitations and topic imbalance +(2) time-aware evidence retrieval with citation tracking to mit- +igate outdated information, (3) structured document parsing to +leverage long-range context, (4) handling complex scientific expres- +sions, including tables, figures, and domain-specific terminology +and (5) assessing the credibility of scientific literature. Preliminary +experiments were conducted to substantiate these challenges and +identify potential solutions. This perspective paper aims to advance +scientific fact-checking with a specialised IR system tailored for +real-world applications. + +CCS Concepts +· Information systems → Specialized information retrieval. + +Keywords +Evidence retrieval, Scientific fact-checking + +ACM Reference Format: +Xingyu Deng, Xi Wang, and Mark Stevenson. 2025. The Next Phase of Scien- +tific Fact-Checking: Advanced Evidence Retrieval from Complex Structured +Academic Papers. In Proceedings of the 2025 International ACM SIGIR Con- +ference on Innovative Concepts and Theories in Information Retrieval (ICTIR) +(ICTIR ’25), July 18, 2025, Padua, Italy. ACM, New York, NY, USA, 13 pages. +https://doi.org/10.1145/3731120.3744614 + +This work is licensed under a Creative Commons Attribution 4.0 International License. +ICTIR ’25, Padua, Italy +© 2025 Copyright held by the owner/author(s). +ACM ISBN 979-8-4007-1861-8/2025/07 +https://doi.org/10.1145/3731120.3744614 + +1 Introduction +Fact-checking aims to assess the veracity of factual claims based +on credible evidence [37, 116] and serves as a crucial safeguard +for mitigating misinformation. Scientific fact-checking is a spe- +cialised variant of this task, grounded in scientific knowledge, with +the objective of combating misinformation that affects the pub- +lic, helping researchers in knowledge discovery and assisting in- +dividuals in understanding scientific advancements [97]. This is +particularly important given the rapid emergence of new scien- +tific findings, where both professionals and the public must assess +the credibility of information. A prominent case occurred during +the COVID-19 pandemic, in which politically motivated misinfor- +mationÐranging from inflated infection statistics to unsupported +treatmentsÐcirculated extensively, eroding public trust and endan- +gering health communication [58]. However, existing approaches +to scientific fact-checking remain limited, primarily relying on the +retrieval of evidence from relatively simple and small-scale sources +[16, 49, 62, 72, 75, 100, 101, 104]. For example, SciFact-Open [101], +the largest available dataset for scientific fact-checking, contains +500,000 documents ś substantially smaller than PubMed, which con- +tains over 37 million biomedical publications. In addition, SciFact- +Open consists only of abstracts, rather than full-text papers, thereby +excluding critical structural and citation information, ignoring long- +range context and scientific expression conveyed through tables +and figures. These design simplifications may hinder the applicabil- +ity of current approaches in real-world settings, where scientific +evidence is embedded in long and structurally complex documents +with multimodal content. + +Fact-checking is a knowledge-intensive task, where the verifica- +tion process relies on sourcing evidence from a reliable upstream +Information Retrieval (IR) system. Emerging findings indicate the +value of effective retrieval in improving fact-checking systems. +For example, introducing even a small amount of noise into evi- +dence can significantly degrade fact-checking performance [76]. +Recent Retrieval-Augmented-Generation (RAG) techniques have +been widely used for fact-checking [28, 47, 63, 70, 80, 84, 94], where +retrieval models are fine-tuned to identify high-quality evidence +for claim verification. These observations underscore the critical +role of robust evidence retrieval, as an ideal IR system for fact- +checking should rank all relevant evidence at the top while filtering +out non-evidential noise. Ensuring retrieval robustness is crucial to +maintaining sufficient yet relevant evidence, which is essential for +improving scientific fact-checking accuracy. + +A major trend in fact-checking research is to consider realistic set- +tings that employ rich, diverse and timely evidence sources, as seen +in FEVER (using Wikipedia) [90] and AVeriTeC (using web-wide +resources) [78]. In document level evidence retrieval, current + +436 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +general fact-checking systems over-rely on commercial search APIs, +which do not consider the specific requirements of fact-checking +[77, 91, 116]. Such reliance on commercial search APIs ś with lim- +ited adaptability ś has left document retrieval methodologies under- +explored in fact-checking, especially for domain-specific corpora +such as scientific fact-checking. Current scientific fact-checking sys- +tems primarily employ off-the-shelf IR methods [97], such as lexical +matching and semantic relevance ranking, which do not scale effec- +tively for large-scale scientific corpora. In addition, the distribution +of relevant evidence across scientific topics is highly imbalanced, +which degrades both retrieval effectiveness and efficiency, espe- +cially for claims with scarce supporting literature. SciFact-Open +[101], which extends the original SciFact dataset [100] for large- +scale evaluation, illustrates this issue: verification performance on +SciFact-Open drops by at least 15 F1 points for all well-performed +fact-checking systems developed in SciFact [101]. While increasing +corpus size enhances evidence diversity, it also amplifies retrieval +noise, reducing efficiency in both retrieval and verification. Beyond +that, high semantic relevance does not guarantee high evidential +relevance, and irrelevant yet semantically similar documents can +introduce noise into downstream verification [117]. These chal- +lenges underscore the necessity of developing tailored document +retrieval systems specifically designed for scientific fact-checking, +as effective retrieval is a prerequisite for accurate claim verification. +Beyond document-level evidence retrieval, within-document +evidence retrieval is also essential for processing complex sci- +entific literature. Scientific papers, unlike general fact-checking +documents, are long, structured, domain-specific and involve addi- +tional metadata. As scientific fact-checking evolves from abstract- +based to full-paper retrieval, retrieval models must account for +metadata (e.g., publish date, citations) and complex structured data +format (e.g., charts, tables, figures). This necessitates the adaptation +of verification models such as SciBERT [11] for domain-specific +terminologies [100], Longformer [12] for long-range dependen- +cies [102] and TAPAS [38] for tabular data verification [3]. Sci- +entific expressions in academic papers are highly structured and +contextually interdependent, where textual content, tabular data, +and figures mutually reinforce the conveyed information. How- +ever, existing scientific fact-checking systems primarily operate +at the abstract level, adopting methodologies similar to general +fact-checking [49, 50, 68, 72, 75, 97, 100, 102, 111, 119], albeit incor- +porating domain-specific models such as BioSentVec [21] and SciB- +ERT [11]. The development of public full-paper datasets aligns with +the requirement of real-world scientific fact-checking systems for +effective verification. This highlights the urgent need for retrieval +and verification methodologies that can leverage entire scientific +documents. Accordingly, within-document evidence retrieval and +its integration into verification pipelines should be explored to fully +unlock the potential of scientific literature for fact-checking. + +This perspective paper presents a comprehensive examination of +the challenges associated with evidence retrieval in scientific fact- +checking, highlighting challenges that are not typically faced +within general fact-checking, leading to the need for specialised +retrieval and verification strategies. We advocate for proactive re- +search efforts to develop scalable methodologies while addressing +the limitations of current datasets. We structure our discussion into +two parts following a typical fact-checking pipeline: Sections 2ś3 + +explore document-level evidence retrieval while Sections 4ś8 ex- +plore within-document evidence retrieval in scientific publication +for scientific fact-checking. Each of the following sections identi- +fies a research challenge followed by a tentative and illustrative +research direction (RD). + +2 Beyond Semantics +Evidence retrieval for fact checking is closely related to traditional +document retrieval techniques, which typically focus on retrieving +documents that are semantically similar to a query or contain match- +ing keywords. While this approach is effective in many scenarios, +it often fails to address the ultimate objective of fact-checking ś +successful claim verification. Evidence retrieval that relies solely on +semantic similarity may prioritise irrelevant or low-context infor- +mation, introducing noise into the subsequent verification process. +Recent IR studies [67, 92] show that semantic relevance alone may +not ensure utility in knowledge-intensive NLP tasks under the RAG +framework, suggesting the importance of utility-aware retrieval +strategies. To improve the verification utility of evidence retrieval +systems, techniques such as fine-tuning, joint optimisation, and +learning from verification feedback have been developed. These +approaches leverage relevance labels derived from annotated gold +evidence [40, 68, 117, 119, 120]. Although graded relevance has +been extensively explored in general IR, current evidence retrieval +systems for fact-checking often oversimplify relevance as binary, +failing to differentiate between fully non-evidential and partially +relevant evidence. This coarse-grained labelling scheme fails to +differentiate between completely non-evidential documents and +partially relevant (plausible) evidence. Negative examples and ran- +domly retrieved examples are equally treated as 0, despite exhibiting +varying degrees of evidential support. We argue that evidence re- +trieval should distinguish between non-evidential information and +plausible evidence, enhancing the model’s ability to identify previ- +ously unobserved but potentially useful evidence within large-scale +corpora. + +Developing an IR system that can effectively differentiate be- +tween evidential and non-evidential information requires access +to fine-grained relevance labels during training. However, manu- +ally constructing negative samples is both complex and resource- +intensive due to the vast number of unlabelled documents and +sentences that lack explicit pairing with given claims. Furthermore, +assessing the degree of evidential support for a claim within unla- +belled documents is inherently challenging. To validate the impact +of fine-grained evidential relevance, beyond semantic relevance, we +carry out preliminary experiments which explore the use of down- +stream verification feedback to capture different levels of evidential +values. + +Experiment Overview. The experiment investigates whether +combining verification feedback with semantic relevance improves +the performance of document evidence retrieval. Figure 1 presents +the pipeline, where probabilities from downstream verification +serve as feedback. Specifically, we integrate two components: 1) the +semantic relevance score, computed using an off-the-shelf reranker +model, and (2) the verification success feedback score, derived from +a fine-tuned verifier model, indicating the degree to which a docu- +ment is evidential for a given claim. + +437 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +Table 1: Retrieval result on SciFact-Open and Check-COVID + +SciFact-Open +R@50 R@20 R@10 +BM25 +45.22 +54.78 +66.09 +71.09 +79.13 +88.91 +monoT5-3B +71.30 +81.95 +91.96 ++Verification +Check-COVID R@50 R@20 R@10 +75.02 +87.91 +89.49 +95.84 +91.48 +96.13 + +BM25 +monoT5-3B ++Verification + +81.96 +93.16 +94.55 + +R@5 +38.04 +57.17 +62.61 +R@5 +67.59 +82.06 +84.04 + +R@3 +30.87 +48.26 +52.83 +R@3 +61.35 +74.93 +77.80 + +R@1 +20.22 +31.09 +32.17 +R@1 +46.18 +58.28 +61.84 + +Table 2: Evidence positions in the retrieved list. We select an ex- +ample from the SciFact-Open dataset. Claim: Female carriers of the +Apolipoprotein E4 (APOE4) allele have a reduced risk for Alzheimer’s +disease. Gold Evidence: [E1,E2,E3,E4,E5] + +reranker model E1 +BM25 +monot5-3B ++Verification + +838th +1st +1st + +E2 +141th +8th +3rd + +E3 +7th +16th +5th + +E4 +163th +2nd +2nd + +E5 +67th +302nd +27th + +Datasets. We conduct our evaluation using datasets including: +(1) SciFact [100]: A corpus of 5,183 abstracts from scientific arti- +cles, with 809/300/300 samples for train, validation and test sets. +The test set is not publicly accessible. (2) SciFact-Open [101]: An +extended version of SciFact with 500K abstracts, re-annotating evi- +dence documents for 279 claims from the original SciFact test set. (3) +Check-COVID [104]: A COVID-19-specific fact-checking dataset, +containing 347 abstracts from CORD-19 journal articles and 1,504 +expert annotated news-related claims. + +MultiVerS is trained on the SciFact train set to create a verifier +model that provides verification feedback. Since the SciFact test +set is inaccessible, we evaluate document evidence retrieval on +SciFact-Open and full Check-COVID. + +Results. The integration of verification feedback consistently en- +hanced document evidence retrieval, as +Verification outperformed +monoT5-3B across nearly all cut-off thresholds in both SciFact-Open +and Check-COVID (Table 1). The improvements are particularly ev- +ident at lower cut-offs, where retrieving the most relevant evidence +is crucial. In SciFact-Open, Recall@5 and Recall@3 increased from +57.17% to 62.61% and from 48.26% to 52.83%, respectively. Similarly, +in Check-COVID, these metrics improved from 82.06% to 84.04% +and from 74.93% to 77.80%. These improvements are particularly +meaningful given that the average number of gold evidence doc- +uments per claim is 1 in Check-COVID and approximately 2 in +SciFact-Open. To illustrate this improvement, we conducted a case +study examining how different retrieval methods ranked specific +evidence documents (Table 2). Compared to monoT5-3B, +Verifi- +cation successfully elevated the ranks of E2, E3 and E5, retrieving +three additional pieces of evidence within the top 5 results. Notably, +E5, ranked 302nd by monoT5-3B, was effectively rescued by adding +verification feedback in +Verification, demonstrating the value of +integrating verification-informed retrieval signals. -IN CHENGDU +These findings highlight a promising research direction: shift- +ing from semantic-only retrieval towards evidence-aware retrieval, +where retrieval models explicitly account for evidential value. Our +results suggest that leveraging well-performing verification models +can help refine retrieval systems by distinguishing between purely - Contents +Claim -I. Guide to Payment Services in China ....................................................... 1 +Retr ieved +documents -II. SIM Card .................................................................................................2 +0.86 -III. Chengdu Public Transit & Key Hubs ................................................... 2 +0.92 -IV. Airport-to-Venue Transportation ........................................................... 3 +0.31 -V. Daily Transportation Options ................................................................. 6 +0.07 -VI. Important Notes .....................................................................................8 +0.04 - I. Guide to Payment Services in China +0.02 -To ensure smooth transactions during your visit, all international attendees are advised to -review the official payment guidelines from the People's Bank of China: +0.01 -http://www.pbc.gov.cn/en/3688110/3688172/5188125/5274061/index.html +r elevance scor es -The page as shown in the picture +Document +Rer anker -Chengdu Metro Payment Options +monoT5-3B -From July 28th, 2025, -international visitors may use UnionPay chip-enabled cards, -internationally issued Visa, Mastercard, and American Express chip-enabled cards to -take Chengdu Metro. +Top-k +documents -Simply tap your card at fare gates for metro lines. +Ver ification -Deductions follow standard pricing rules. +M ultiVer S -Cards can also be used at Ticket Vending Machines (TVMs) and Day Pass Vending -Machines. +Suppor t -1 +0.7 - II. SIM Card +Refute -Alipay provides an eSIM feature that enables users to activate a virtual SIM card -directly through the app. +0.1 -You can also apply at any China Telecom, China Mobile, or China Unicom store (including -outlets at the airport). +ver ification +feedback -Main mobile phone operators in China +NEI -Card Application Process: +0.2 -1. Visit a store to apply for a SIM card and present your identification documents -(passport/visa, etc.). +0.8 -2. Select your preferred phone number and service plan. +Figure 1: Pipeline of experiment -3. Receive your activated SIM card. The entire process is efficient and typically completes -quickly. +Approaches. monoT5-3B [64] has demonstrated strong perfor- +mance as a reranker for SciFact, as evidenced by its widespread +use as a strong baseline in studies [55, 86] on the BEIR benchmark +[89]. It also demonstrated state-of-the-art performance in evidence +retrieval for verification [68, 97, 102]. The model assigns a predicted +score, 𝑠𝑟 +𝑐,𝑑 , representing the semantic relevance for a document 𝑑 +to a claim 𝑐 as defined in Equation 1. -Note: Before leaving, ensure the staff assists with card activation and verify that the card -functions properly. +(1) -III. Chengdu Public Transit & Key Hubs +𝑐,𝑑 ∈ (0, 1) -Chengdu offers a convenient public transportation system, including: +𝑓 (𝑐, 𝑑) → 𝑠𝑟 -Two major airports: Chengdu Shuangliu International Airport & Chengdu Tianfu -International Airport. +𝑐,𝑑, 𝑠𝑟 +MultiVerS is the best-performing verifier model on SciFact [97, +102]. We reproduced this model using the official implementation1 +while adjusting the negative sampling parameter from 20 to 5 to +avoid over-fitted verification feedback. The model predicts the veri- +fication outcome as per the calculated probabilities for a document +𝑑 either supporting 𝑝𝑟 +𝑐,𝑑 or providing insufficient in- +formation 𝑝𝑛 -Multiple high-speed rail stations, such as Chengdu East Railway Station (primary hub -that connects 80%+ high-speed routes), Chengdu South Railway Station (regional -services that link to Leshan City, Mianyang City, Yibin City...), and Chengdu West -Railway Station (serves western routes, e.g., Pujiang, Dujiangyan). +𝑐,𝑑 , refuting 𝑝𝑠 +𝑐,𝑑 , relative to a claim 𝑐, as follows: +𝑐,𝑑, 𝑝𝑠 -An extensive metro and bus network, ensuring efficient and accessible travel throughout -the city. +𝑉 (𝑐, 𝑑) = 𝑎𝑟𝑔𝑚𝑎𝑥 (𝑝𝑟 -2 +𝑐,𝑑, 𝑝𝑛 - IV. Airport-to-Venue Transportation +𝑐,𝑑 ) -From Chengdu Tianfu International Airport (TFU) to Chengdu Tianfu International -Convention Center +(2) -By Metro (Recommended): ++Verification (Ideal Reranker Model) combines semantic relevance +and verification feedback to refine evidential retrieval. The final +retrieval score is calculated by summing the semantic score (𝑠𝑟 +𝑐,𝑑 ) +and the verification probabilities (𝑝𝑟 +𝑐,𝑑 ), followed by a nor- +malisation step to ensure the score remains within (0,1), formulated +as: -Route: Take Line 18 from T1 & T2 of Tianfu International Airport → Exit at Western China -Int'l Expo City Station (Exit E). The venue is a 3-minute walk (180 meters) from Exit E. +𝑐,𝑑 and 𝑝𝑛 -3 +𝑠𝑟 +𝑣 +𝑐,𝑑 - Operating Hours: += 1/2 ∗ (𝑠𝑟 -Duration: 50 Minutes +𝑐,𝑑 + 𝑝𝑟 +This formulation ensures that documents contributing to support +or refute labels receive higher retrieval scores, enhancing the evi- +dential quality of retrieved documents. -Payment Method: Bank Cards/ Alipay/ WeChat Pay/ Cash +𝑐,𝑑 ) ∈ (0, 1) -By Taxi: +𝑐,𝑑 + 𝑝𝑠 -Boarding Location: Level 1, Terminal 1&2 +(3) -Availability: Available around the clock +To evaluate retrieval effectiveness, we use Recall@k (R@k), a +common evaluation approach that measures the proportion of rele- +vant evidence successfully retrieved within the top 𝑘 results. -Fare: CNY100-120 | Duration: 45-60 min (traffic-dependent) +1https://github.com/dwadden/multivers -From Chengdu Shuangliu International Airport -International Convention Center +438 -(CTU) +r -to Chengdu Tianfu +e -By Metro (Recommended): +r -Route: Take Line 19 from East of Terminal 2 of Shuangliu International Airport Station→ -Disembark at Tianfu Commercial District Station (3-minute indoor walk required) / Transfer to -Line 6 at Tianfu Business District Station → Exit at Western China Int'l Expo City Station -(Exit E). The venue is a 3-minute walk (180 meters) from Exit E. +a -4 +n - Operating Hours: +k -5 - - Duration: 40 Minutes - -Payment Method: Bank Cards/ Alipay/ WeChat Pay/ Cash - -By Taxi: - -Boarding Location: - -T1’s Pick-up Points: Suspend the service. - -T2’s Pick-up Points: Near the crosswalk of the arrival level. - -Availability: Available around the clock - -Fare: CNY 60-80 | Duration: 40-50 min (traffic-dependent) - -Airport shuttle buses (from both Tianfu and Shuangliu Airports): - -Operate on fixed routes with stops not in close proximity to the convention venue. If you need -transportation to other destinations, please check the official airport shuttle schedule upon -arrival. - -V. Daily Transportation Options - -Metro - -Transportation Option: Ticket - -The ticket price ranges from CNY 2 to CNY 10 according to the traveling distance. The fare -complies with the rules below: - -Staffed Assistance - -Self-Service Machines - -6 - - Notes: -Tickets are available at staffed counters or self-service machines - -Daily passes (1-day: CNY 20 | 3-day: CNY 50 | 5-day: CNY 70) - -Payment Methods: Bank Cards/Cash (CNY)/WeChat Pay/Alipay. - -Transportation Option: QR Code Access - -Via Alipay: Go to Transport → Select Chengdu Metro → Complete ID verification → Scan -QR code at gates - -Transportation Option: Transport Card - -Purchase a Tianfu Tong Card at staffed service counters for metro/bus use. - -Bus - -Transportation Option: Cash (exact change recommended) - -Transportation Option: Alipay QR Code- The same as metro procedure, requires QR code -scanning “Transport → Select Chengdu Bus” - -Transportation Option: Tianfu Tong Transport Card - -Taxi - -Payment Method: Cash (CNY) or mobile payment (WeChat/ Alipay QR code scanning) - -Please note: Credit cards are not accepted in local taxis. - -7 - - Ride-hailing service: Ride-hailing apps like DiDi are accessible. - -VI. Important Notes - -Card Payments - -Visa/Mastercard accepted at most venue-adjacent restaurants and shops. - -Power Banks - -For international passengers entering China and taking domestic connecting flights to -Chengdu: - -Power banks without CCC certification are strictly prohibited as per China's aviation safety -regulations. - -Note: This regulation does not apply to: - -International passengers arriving in Chengdu as their final destination. - -Direct international transit passengers (remaining in the sterile area). - -8 + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +semantic relevance and plausible evidential relevance among unan- +notated documents. Furthermore, to continually improve evidence- +aware retrieval, we propose the development of tailored IR systems +capable of identifying evidential information, thereby enhancing +evidence retrieval for scientific fact-checking. + +RD.1. Benchmark tailored IR system for fact-checking + +The preliminary study presented in this work outlined a framework +to enhance evidence retrieval beyond only semantic relevance. To +overcome the limitations of existing IR systems in scientific fact- +checking scenarios, it is imperative to develop specialised IR sys- +tems capable of handling the specific challenges of verification tasks. +However, training an IR system on a single fact-checking dataset +risks poor generalizability and potential overfitting, particularly due +to data imbalance, a common issue in the relatively small datasets +characteristic of scientific fact-checking [97, 116]. Furthermore, +poor verification performance deteriorates retrieval accuracy, cre- +ating a vicious feedback loop that further degrades overall system +effectiveness. A multi-pronged strategy could mitigate these chal- +lenges by pooling verification signals from various high-performing +verifier models, leveraging large-scale datasets such as FEVER [90] +to improve training robustness, and providing a shared retrieval +checkpoint enable subsequent studies to fine-tune the model for +specific scenarios or datasets while reducing training cost. Recent +work [51, 74] has explored unified retrieval models for knowledge- +intensive NLP tasks, focusing on retrieval quality and downstream +task utility [73, 115], including question answering (QA) and fact- +checking. Similarly, we propose a verification-driven IR system for +evidence retrieval, which explicitly incorporates evidential infor- +matics. This approach follows a two-step training paradigm: general +pre-training on large, diverse datasets followed by domain-specific +fine-tuning. This approach balances scalability and domain speci- +ficity, ensuring IR models are both robust across different contexts +and highly effective in targeted fact-checking applications. Addi- +tionally, a corresponding benchmark should employ a diverse set +of evaluation metrics beyond for the fact-checking task to ensure +comprehensive assessment of performance within fact-checking +[6]. These metrics could include verification accuracy, reflecting the +downstream utility of retrieved evidence; decision latency, measur- +ing the computational efficiency of retrieval models; and robustness +to real-world conditions such as noisy data and incomplete evidence, +to improve system resilience. + +Integrating verification feedback into evidence retrieval improves +relevance assessment beyond binary labels, enhancing retrieval per- +formance. Future research should focus on developing a benchmark +IR system tailored for fact-checking, incorporating fine-grained +relevance labels and verification-driven retrieval models. A scalable +pre-training and fine-tuning approach has the potential to improve +retrieval robustness and generalizability thereby producing more +accurate and efficient fact-checking systems. + +3 Imbalanced resources of scientific topics +In existing general fact-checking datasets, such as FEVER which is +based on Wikipedia, the distribution of gold evidence per claim is +relatively even and sufficient. However, a significant imbalance of +evidence is observed in scientific fact-checking. While the SciFact + +Table 3: Sufficient-evidence claim and none-evidence claim exam- +ples in SciFact-Open. ‘Evidence’ is the number of evidence in SciFact- +Open corpus and ‘Entities’ is the number of entities by searching +bold-keyword in PubMed. + +Claim +Obesity is determined in part by +genetic factors. +LRBA controls CTLA - 4 expression. + +Evidence Entities + +24 + +0 + +499k + +0.27k + +corpus (~5K documents) maintains a relatively balanced number of +evidence documents per claim, this balance was disrupted when the +dataset was expanded to create SciFact-Open (~500K documents). +In this larger corpus, the majority of claims have none or only +one piece of supporting evidence while others have over. Claims +related to less-researched topics are generally associated with fewer +scientific publications, as illustrated by the examples in Table 3. + +However, most fact-checking systems do not explicitly account +for evidence imbalance. A common approach is to use a fixed re- +trieval cut-off (i.e., selecting a predefined number of top-ranked +documents for verification). One of the most inefficient approaches +is setting the cut-off equal to the maximum number of evidence +per claim in the dataset, ensuring that all possible evidence is re- +trieved. This heuristic has not previously caused major issues since +general fact-checking datasets contain a relatively balanced num- +ber of supporting documents per claim. However, the imbalance +in SciFact-Open suggests that the simple approach may not be +suitable for open-domain scientific fact-checking with two major +drawbacks: + +(1) Inefficiency. Although the maximum number of gold evi- +dence documents in the SciFact-Open dataset is 24, less than one +third of claims have more than two. Using the maximum number as +a cut-off would be inefficient due to the large number of documents +that would have to be processed by the verifier. + +(2) Inaccuracy. Introducing irrelevant evidence into downstream +verification degrades fact-checking performance, whether the noise +is semantically related or completely random [76] (as discussed in +Section 2). + +To address these challenges in the current and future studies of +scientific fact-checking, we proposed a research direction based +on flexible cut-off strategy for retrieving evidence based on claim +characteristics. + +RD.2. Flexible cut-off for Retrieved Evidence + +Ranked List Truncation (RLT) refers to the task of selecting an opti- +mal prefix of a ranked list of retrieved documents, with the goal of +balancing retrieval effectiveness and efficiency. Prior work explores +both heuristic and learned approaches, using either relevance labels +or features derived from score distributions to determine the cut-off +point [10, 52, 59, 61, 103, 109]. A related line of work is stopping +methods in technology-assisted review (TAR) [14, 15, 85], which +aim to retrieve as much relevant information as possible while min- +imising the effort spent on examining irrelevant documents. Both +approaches aim to optimise an expected metric over candidate cut +positions, typically using metrics such as 𝐹 1@𝑘 or 𝑟𝑒𝑐𝑎𝑙𝑙@𝑘. The +datasets used in these prior studies on RLT and stopping methods, +such as CLEF and TREC [25ś27, 33, 43ś45], exhibit imbalance but + +439 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +typically contain enough relevant items per query to support recall- +based supervision and evaluation. This dependence on sufficient +relevance labels becomes problematic in fact-checking scenarios, +where gold evidence documents are typically rare, making both +recall-based stopping and supervised RLT approaches unsuited to +this problem. + +To explore whether relevance score distributions indicate evi- +dence sufficiency, we compare well-studied and less-studied claims +from SciFact-Open. Following Wadden et. al. [101], claims with +four or more gold evidence documents are considered to be well- +studied and those with none to be less-studied. Using monoT5-3B, +we compute several statistics over the ranked document scores, +including first-document relevance, average and total scores, score +decay, and the initial-to-final score ratio. Table 4 shows that well- +studied claims tend to have higher top-ranked scores and sharper +decay patterns, suggesting that relevance distributions may serve +as indicators of evidence sufficiency. + +Table 4: Statistical analysis of average relevance scores for +well-studied and less-studied claims. ‘I/F’ ratio is Initial-to- +Final ratio. ‘Exp k’ denotes the exponential decay factor k. + +Metric +Less- +Well- + +1st Doc Mean +0.502 +0.741 + +0.941 +0.995 + +Sum I/F ratio Exp k +1.544 +3.514 +25.097 +0.651 +1.963 +37.052 + +Based on these findings, one possible direction is to leverage +existing techniques to estimate whether a claim is less-studied or +well-studied. A prediction module could utilise statistical features +of relevance distribution such as those presented in Table 4. In addi- +tion, metadata such as retrieved entity counts in PubMed (Table 3) +can serve as auxiliary signals to refine the prediction. Claims pre- +dicted as less-studied ś e.g., with low total relevance or steep score +decay ś may be assigned smaller cut-offs to reduce verification cost, +while RLT and stopping techniques could be applied to well-studied +claims where concentrated high scores suggest richer evidence. + +While this naive strategy relies on heuristic features, it does not +explicitly optimise verification performance. To address this, future +approaches could explore learning a cut-off policy using feedback +from the verification stage. Specifically, truncation points may be +selected based on reward signals, such as whether the claim is cor- +rectly verified or the confidence of the verifier. This would bypass +the need for relevance-labelled supervision, which is often infeasi- +ble in scientific fact-checking due to sparse annotations. Inspired +by prior RLT and stopping method work, such a learned policy +could optimise both efficiency and factual accuracy by aligning +truncation decisions with downstream verification performance. + +4 Time and Citation +General fact-checking evidence corpus such as Wikipedia and fact- +checking websites, often lack sentence-level evidence timestamps, +making it difficult to determine the original publish time of sentence +evidence in verification and hindering the development of time- +aware retrieval methods. Timeliness is important in fact-checking, +but in science, evolving evidence makes outdated studies particu- +larly problematic. For instance, early COVID-19 treatment studies + +Table 5: Results of health QA task considering the different +thresholds of the published time of literature [99] + +Year +≥2020 +≥2018 +≥2015 +≥2010 +≥2005 +≥2000 +≥1990 +≥1980 + +Precision Recall + +59.7 +59.6 +61.1 +63.4 +68.1 +66.1 +65.6 +64.2 + +60.3 +58.0 +56.0 +55.6 +56.5 +56.8 +55.4 +54.7 + +F1 score +58.7 +57.9 +53.9 +52.8 +52.0 +51.8 +51.3 +50.0 + +were later refuted, and outdated evidence may lead to harmful de- +cisions. This section discusses whether scientific publications are +more suitable for time-aware fact-checking and explores possible +ways to leverage their inherent temporal characteristics. + +4.1 Timeliness of evidence +Unlike general fact-checking, where historical and static facts re- +main unchanged, scientific knowledge continuously evolves. This +fundamental difference necessitates time-aware retrieval in scien- +tific fact-checking to ensure that retrieved evidence remain valid +and reflective of the latest scientific consensus. A time-sensitive +retrieval mechanism should prioritise recent publications to en- +sure that fact-checking systems incorporate the most up-to-date +methodologies and factual updates. This is especially crucial in +fields like healthcare, where relying on outdated information could +lead to misleading conclusions or incorrect decisions. For example, +during a rapidly evolving pandemic, a medical treatment initially +considered effective might later be deemed unreliable. This sec- +tion discusses the challenges and opportunities of integrating the +evidence timestamp into scientific fact-checking. + +Scientific fact-checking aims to find evidential information in +the literature to verify a claim. Intuitively, considering outdated +literature negatively affects verification. Research in healthcare +question answering (QA) has demonstrated that time-aware re- +trieval improves system performance [99], as shown in Table 5. The +F1 score of the healthcare QA system improves as the publication +year of evidence documents becomes more recent. By extension, +scientific fact-checking on a large-scale corpus may also suffer from +incorporating outdated evidence. These findings highlight the need +for time-aware filtering in scientific fact-checking systems to en- +hance reliability. Fact-checking datasets could explicitly incorporate +timestamps as metadata to facilitate research into temporal rele- +vance in retrieval and verification. A general fact-checking study +[9] collected the ‘timestamp of last update’ of claims and evidential +documents, allowing later research [8] to explore the impact of +temporal data on verification. This study found that a time-aware +system achieved a 15% improvement in macro F1 score, underscor- +ing the importance of temporal information. However, this work +was limited by: (1) focusing on verification only, where evidence +documents had already been retrieved in a separate initial step, +without ensuring that retrieval prioritised high-quality and tem- +porally relevant evidence, and (2) the use of fragmented evidence ś + +440 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +the general fact-checking datasets usually only contain small snip- +pets of web documents, omitting many important time expressions +present in full texts. + +Given these findings, it is essential to develop a fact-checking +system that explicitly incorporates temporal awareness across both +retrieval and verification stages to meet real-world applications. + +RD.3. Time-Aware Retrieval and Verification + +Traditional fact-checking approaches often handle conflicting evi- +dence for a single claim by assigning neutral veracity labels, such +as “mixture," “unproven," or “not enough information" [37, 97]. +However, conflicting evidence often arises due to outdated stud- +ies included in the retrieval process, which introduces noise and +adversely affects prediction accuracy [99]. This issue has been ob- +served in healthcare QA systems, where outdated evidence degrades +performance [99]. To address this issue, we propose a time-aware +approach that incorporates temporal information into both retrieval +and verification stages: + +1.Retrieval Stage: Outdated evidence should be filtered or de- +prioritised during retrieval to ensure that the retrieved evidence +set is temporally aligned with the latest scientific findings. + +2.Verification Stage: After filtering by time-aware retrieval, all +retrieved evidence should be considered but with differentiated +weighting based on temporal relevance. Recent evidence should be +prioritised through higher weights, while older evidence should +serve as supplementary or contextual information rather than pri- +mary evidence. + +By processing outdated evidence differently across retrieval and +verification components, this direction explores how temporal in- +formation can reduce noise and improve the reliability of scientific +fact-checking systems. + +4.2 Indirect evidence through citation +General fact-checking faces a number of challenges when attempt- +ing to determine the timeliness of evidence: (1) Lack of publication +timestamp [8]. Many fact-checking sources, such as Wikipedia and +fact-checking websites, do not provide precise publication dates +for individual sentences or paragraphs. Instead, they record only +the last edited timestamp, which does not accurately reflect when a +fact was first published. (2) Tracking the origin of evidence. Evidence +is often copied or paraphrased across multiple sources, making it +difficult to determine the original publication date of a statement. +(3) Search engine bias in retrieval. Pre-established fact-checking +datasets retrieve evidence using top-k search engine results, where +ranking mechanisms may prioritise recent documents due to time- +aware ranking biases. This can misrepresent the actual chronology +of claims and lead to fragmented evidence, making time extraction +unreliable. + +While these challenges also affect scientific fact-checking, they +can be naturally mitigated by leveraging full-text papers rather than +abstract-only sources. The main advantages of using full-text papers +include: (1) Explicit publication metadata: Each piece of literature +has a clearly defined publication date, ensuring accurate temporal +tracking. (2) Citation tracing for indirect evidence: Mandatory cita- +tion rules in academic publications facilitate source tracing, even +when statements are referenced indirectly. (3) Structured nature + +of academic papers: provides an indication of the origin of state- +ments. For example, evidence in the background section typically +references prior studies, whereas those in the abstract or results +sections represent findings from the current publication. Given +these inherent advantages, incorporating timestamps as metadata +offers a promising research direction for full-paper-based scien- +tific fact-checking. Exploring the temporal dynamics of claims and +evidence should further enhance retrieval accuracy. Additionally, ci- +tation tracking to trace the original source of paraphrased evidence, +ensuring the first-published timestamp is accurately recorded. + +Citation-based tracking provides a promising approach to esti- +mating evidence timestamps. However, the widespread presence of +multiple citations in scientific literature makes it difficult to iden- +tify which references should be tracked, increasing computational +costs and reducing efficiency. Moreover, indirect citations and para- +phrased references, particularly in introductory sections, further +obscure the retrieval of the first-published source. To address these +issues, we propose the following research direction to develop a +more effective approach for citation tracking and timestamp attri- +bution. + +RD.4. Citation-Based Evidence Tracking + +Intuitively, self-contained evidence refers to information directly +presented in the body of the current paper, while cited evidence is +derived from external sources referenced by the paper. To explore +this distinction, we analysed 22 accessible full papers out of 24 +gold evidence for the claim in Table 2. We prompted GPT-4o to +search for supporting/refuting evidence and determine whether +each piece of evidence was paraphrased/summarised from a citation +or is self-contained. + +Figure 2: Evidence sources in scientific literature. + +The result shown in Figure 2 reveals, as expected, that cited +evidence is predominantly located in the introduction, while self- +contained evidence is more common in the results and conclusion +sections. While we observed a few inaccurate outcomes (e.g., one +‘From Cited’ evidence appearing in ‘Abstract’ is misjudged), the +overall distribution remains discernible and interpretable. In ad- +dition, we also observed that sentences in scientific literature fre- +quently contain multiple citations, making it costly to manually +extract the original timestamp of cited evidence. To address this +challenge, we propose a two-step approach: + +1. Identify track-worthy citations. Not all citations are equally +important for fact verification. Track-worthy citations should in- +clude: conclusive evidence directly influences claim verification and + +441 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +plausible evidence that may impact claim assessment [108]. Since +checking every cited document is expensive, an initial filtering +step is required. A potential solution is to rank citations based on +their relevance and function. Citation recommendation [30, 34] is a +similar task to identify relevant publications for a given statement +using retrieval models. Beyond relevance, the function of citation +can be referred to as a signal to adjust priority. Citations are classi- +fied into eight categories: background, motivation, uses, extends, +similarities, differences, compare/contrast, and future work [42]. +While single evidence has multiple citations, the function of cita- +tion can help identify the citations that align with the evidence. +For example, for predicted evidence ‘Experiments show model A +outperforms previous SOTA model B [citation 1,2,3]’, cited papers +for ‘model A/B’ in the ‘introduction’ function are less check-worthy +than the citation in the ‘compare/contrast’ function. By prioritising +high-impact citations, retrieval costs can be significantly reduced. +2. Track the original timestamp of evidence. Once track- +worthy citations are identified, the next step is to trace the original +timestamp of cited evidence via direct and indirect citation tracking. +Direct citation tracking can be applied if the publication date of the +cited paper is straightforward to retrieve. However, some evidence +is paraphrased or indirectly cited, requiring a deep tracking mech- +anism to trace the citation path. Citation graph analysis [17, 96] +can help map citation paths using directed graphs and applying +search algorithms to identify the earliest relevant source. A recent +study [118] found that reference errors ś references do not include +information to support statement ś frequently appear ranging from +11% to 41% across domains. Addressing these errors introduces +a sub-task for scientific fact-checking: verifying whether a cited +reference truly supports the claim. To ensure feasibility, an early +explorative study can assume that scientific literature generally +adheres to citation conventions, preventing infinite citation loops +in verification. + +In summary, to explore time-aware fact-checking for scientific +literature, we propose two potential research directions: (1) Integrat- +ing temporal information into retrieval and verification to handle +outdated evidence and (2) Developing citation-based tracking meth- +ods to identify the original source and timestamp of evidence. These +directions provide a basis for studying the impact of time-aware +mechanisms in scientific fact-checking systems. + +5 Structured Long-Context Evidence Retrieval +Existing scientific fact-checking datasets construct evidence cor- +pora using fragmented sentences, paragraphs, or abstracts [16, 49, +62, 72, 75, 100, 104]. However, scientific literature is typically pre- +sented in structured, visually rich formats, often as PDF documents, +where different sections serve distinct functions: abstracts, results, +and conclusions summarise key findings, while background and in- +troduction sections provide prior research context. With the diverse +and unique functionalities of scientific literature components, this +section explores challenges and potential research directions for +advancing scientific fact-checking at the full-paper level. Existing +fact-checking pipelines typically follow a document retrieval and +then sentence selection paradigm [37, 97, 116]. For general fact- +checking, evidence retrieval often uses top-ranked sentences from +top-ranked documents, treating them as self-contained evidence + +Title +Authors + +Publish Venue +Publish Time + +Citations: +[ 25, 44 , 60 , 67, 78 , 82 , 92 + +Figure 3: An example for parsing the scientific literature + +units. However, this approach neglects long-range context, as using +the extracted sentences ignores surrounding information to sup- +port verification. In contrast, scientific documents exhibit higher +document-level consistency of verdict, commonly one paper hav- +ing a sole standpoint to a given question, making document-level +processing necessary for scientific fact-checking [50, 102, 119]. + +LLMs have recently demonstrated growing capability to process +long contexts. However, it remains challenging to fact-check an +entire full-text document in a single pass [105]. LLMs are prone to +hallucinating content that is not grounded in the provided docu- +ments [2, 20, 23, 88] and are often susceptible to distraction from +irrelevant context [82]. Their reasoning capabilities also degrade +as text length increases [46]. A potential solution is context distil- +lation, as used in retrieval-augmented generation (RAG), to filter +high-quality context and mitigate hallucination, which improves +the performance of downstream QA tasks [106]. However, unlike +the QA task, fact-checking requires explicit retrieval of explicit. Fil- +tering context may remove crucial supporting evidence, leading to +incomplete verification. Moreover, much of the historical scientific +literature exists in PDF format. Although recent multimodal LLMs +are capable of consuming PDFs and conducting reasoning tasks, +their capabilities are still limited to surface-level understanding. +For example, they frequently fail to capture cross-page content +and complex layout structures [60, 87, 95]. By prompting GPT-4o +to locate evidence in PDF literature with the results presented in +Section 4.2, we observed that it mislocated evidence in incorrect +sections or non-existent sections. These limitations indicate that ex- +isting LLMs are not capable of supporting end-to-end fact-checking +for long-context scientific literature ś not only for raw PDFs but +also for plain-text documents. + +In addition to their length, scientific papers follow a structured +format that introduces additional challenges for verification. Back- +ground and introduction sections often cite prior work, which +may conflict with conclusions drawn later, while discussion sec- +tions highlight limitations that can cast doubt on earlier findings. +These internally inconsistent signals may introduce misleading +information, demanding reasoning that accounts for both the fac- +tual assertions and the functional roles of different sections. Pars- +ing scientific documents into structured units enables a modular +pipeline where retrieval and verification can be independently opti- +mised. This modular structure facilitates interpretability, robustness, +and denoising of conflicting or irrelevant content. Layout-aware +tools offer a practical foundation for such structure-aware process- +ing [19, 32, 41, 56, 81], as illustrated in Figure 3. + +442 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +To address these challenges, we advocate for a retrieval frame- +work that explicitly considers the document structure commonly +found within scientific literature. Such a system should identify +targeted evidence, capture long-range context across sections, and +suppress irrelevant or conflicting content. We next outline a direc- +tion toward adaptive, section-aware retrieval strategies designed +to meet these requirements. + +RD.5. Adaptive Section-Aware Evidence Retrieval + +Scientific literature follows a structured format where different sec- +tions serve distinct functions, presenting challenges for traditional +evidence retrieval and verification in fact-checking systems. Exist- +ing retrieval methods often operate at the sentence or paragraph +level, neglecting the long-range context and the structured nature +of scientific documents. Additionally, large language models (LLMs) +struggle to accurately associate claims with the appropriate sec- +tions, leading to potential misinterpretations and inconsistencies. +A promising research direction is adaptive section-aware ev- +idence retrieval, which dynamically adjusts retrieval strategies +based on document structure and claim types. This approach con- +sists of two key components: + +Claim-evidence matching. Claims should first be matched to +the most relevant sections. For instance, experiment-driven claims, +such as “X method improves accuracy compared to Y,ž should pri- +marily retrieve evidence from the ‘Results’ and ‘Conclusion’ sec- +tions, as these contain empirical findings. In contrast, background +or theoretical claims, such as “X method is widely used in Y applica- +tions,ž should focus on the Introduction and Background sections, +which provide foundational knowledge. Prioritising section-aware +evidence retrieval helps filter irrelevant context and reduces re- +trieval noise. + +Contextual expansion. After retrieving primary evidence, the +system should augment it with relevant contextual information +from other sections to improve interpretability. For example, method- +ological details from the ‘Analysis’ section can provide additional +support for experimental claims, while historical context from the +Background section can clarify theoretical claims. Access to the full +document allows for retrieving finer-grained evidence or comple- +mentary details that fragmented approaches may overlook. + +In summary, an effective adaptive section-aware retrieval sys- +tem must overcome challenges in accurately parsing document +structures, prioritising relevant sections based on claim types, and +efficiently handling conflicting evidence. By integrating structured +document parsing, hierarchical retrieval strategies, and context- +aware reasoning, future systems can leverage richer evidence from +full-paper scientific literature while reducing LLM hallucinations. +Advancing these techniques will enhance the reliability and inter- +pretability of scientific fact-checking systems. + +6 Multimodal content in Science +Scientific literature often conveys key evidence using non-textual +elements such as tables, charts, and figures, which are commonly +used to present experimental results, statistical analyses, and the- +oretical models, as shown in Figure 3. For full-text scientific fact- +checking, especially across various fields of science and technol- +ogy, it is crucial to move beyond text and accurately interpret + +these structured elements to ensure comprehensive verification. +Fact-checking and misinformation detection on individual modal- +ities ś such as figures [1, 65, 66, 69, 93, 112], charts [4, 7], and +tables [3, 13, 22, 29, 36, 57, 79, 83, 113] ś has been studied inde- +pendently [5]. To improve tabular reasoning, transformer-based +approaches such as TAPAS [38] and Table-BERT [114] have been de- +veloped. However, these techniques perform poorly in SCITAB [57], +a dataset for scientific fact-checking on tables, with results barely +above random. One possible cause is the lack of contextual ground- +ing for tables, which are rarely self-contained. Similarly, figures and +charts in scientific papers often require surrounding textual expla- +nations for correct interpretation. Recent datasets like AVerImaTeC +[18] address image-text verification using web-sourced data, but sci- +entific domains present greater challenges: figures are densely struc- +tured, often span multiple sections, and require domain-specific +understanding capability. + +We argue that scientific fact-checking should shift toward full- +paper analysis, where structured elements are interpreted along- +side their textual context. Unlike standalone multimodal models, +document-level processing enables cross-referencing between fig- +ures/tables/charts and their descriptions, facilitating more faithful +and complete verification. + +RD.6. Multi-modal Evidence Alignment + +Scientific fact-checking requires integrating evidence across multi- +ple modalities, including text, tables, and figures, to ensure consis- +tency and completeness. Recent multimodal information retrieval +datasets in the scientific domain [71, 110] provide aligned pairs +of textual and structured content, offering a foundation for cross- +modal reasoning. However, in real scientific documents, structured +elements, such as figures and tables, are not always located close +to their descriptive text, making alignment a non-trivial challenge. +A promising direction is to explicitly align structured elements +with their corresponding textual explanations within the same +document. In scientific articles, tables and figures are typically ex- +plained through captions or surrounding sentences. Layout-aware +parsing techniques can help identify these elements and link them +to relevant text spans. Once aligned, their contents can be jointly +encoded, enabling claim verification that draws on both structured +data and contextual text. This alignment facilitates more coherent +retrieval and reasoning across modalities, improving the reliability +of multimodal fact-checking. + +This unified framework contrasts with traditional multimodal +systems that process each modality in isolation. Effective alignment +demands progress in scientific document understanding [24, 60], +visual structure parsing [32], and domain-specific retrieval [71]. +Integrating these efforts will support full-document, multimodal +verification, where structured evidence is faithfully grounded in its +textual context. + +7 Credibility of scientific literature +Existing evidence retrieval models often favour high-ranking doc- +uments based on semantic relevance, often overlooking scientific +rigour [98]. As a result, low-quality documents may be retrieved +as evidence, undermining the credibility of scientific fact-checking. +The reliability of a claim verification process is inherently tied to + +443 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +the quality of the supporting literature, making evidence credibility +a crucial factor in scientific fact-checking. + +In the domain of scientific literature, credibility assessment is +influenced by multiple factors, including peer-review status, which +ensures methodological scrutiny, citation impact which indicates +how influential a study is within its field, and experimental rigour, +reflecting the robustness of a study’s methodology. However, the +proliferation of non-peer-reviewed manuscripts and publications +from venues with varying editorial standards, particularly in open- +access repositories, poses a growing challenge. Such sources may +lack the rigorous methodological scrutiny necessary to ensure re- +liable scientific conclusions. Therefore, incorporating additional +quality indicators is essential for enhancing the robustness of sci- +entific fact-checking. + +RD.7. Extending Indicators of Evidence Quality + +Evaluating scientific literature quality extends beyond content +reliability, and should consider factors including venue reputa- +tion, methodological rigour, and experimental transparency. High- +impact journals and prestigious conferences generally enforce strin- +gent peer-review standards, contributing to the credibility of pub- +lished research. Similarly, the expertise and prior contributions of +an author, particularly in reputable venues, can provide further +insight into the credibility of a study. In addition to traditional +metadata, emerging indicators such as replication status, data avail- +ability, and adherence to reporting guidelines can further reflect +methodological soundness. Although metadata-based credibility +assessment is a useful heuristic, it is not foolproof. For example, +selective reporting and statistical manipulation still exist, as some +widely cited studies have later been retracted due to methodolog- +ical flaws [31]. While the integration of such metadata remains a +reasonable approach, as these indicators generally correlate with lit- +erature quality, their limitations must be acknowledged, given that +even widely cited studies can occasionally be subject to retraction +due to undetected methodological flaws. + +8 Scientific terminology complexity +Scientific fact-checking systems often encounter challenges when +aligning claims with supporting evidence due to mismatches in +terminology granularity [101, 107]. The prevalence of hierarchical +and synonymous scientific terms introduces significant challenges +in fact verification. Many concepts exist at multiple levels of speci- +ficity, where broader categories encompass more specific subtypes, +leading to ambiguity in claim-evidence alignment. This issue is +further exacerbated by high token-level similarity among related +terms, making it difficult for models to differentiate between general +and specific concepts. As a result, models often misinterpret evi- +dence relevance, increasing the likelihood of incorrect verification +outcomes. + +This issue arises when a claim uses a broad term, while the sup- +porting evidence provides a more specific instance, or vice versa. +As in the following example, such mismatches can lead to incor- +rect veracity assignments, as existing models struggle to recognise +hierarchical relationships between concepts. + +Claim: Cancer risk is lower in individuals with a history of + +alcohol consumption. + +Supports: Alcohol consumption was associated with a decreased + +risk of thyroid cancer. + +This issue is common within scientific fact checking and has been +reported to occur within 44% of annotated examples in SciFact-Open +[101]. Hence, we argue that capturing the hierarchical relationship +could be a research direction to improve verification performance +in the scientific domain, by solving the mismatch problem. + +RD.8. Hierarchical Concept Modelling + +To address this issue, ontology-based reasoning can be integrated +into fact-checking pipelines. Structured ontologies such as MeSH +[54] and UMLS [53] define hierarchical relationships that help sys- +tems infer term specificity. Recognising that ‘lung cancer’ is a sub- +type of ‘cancer’ enables better claim-evidence alignment, mitigat- +ing errors caused by lexical similarity. Beyond that, the knowledge +graph can enrich ontological reasoning by encoding both hier- +archical and associative relationships among scientific concepts +[39, 48]. However, its potential for resolving terminology granular- +ity mismatches in scientific fact-checking remains unexplored. Ad- +ditionally, representation learning techniques such as contrastive +learning can embed these hierarchical relationships into vector +space representations, reducing reliance on token-level similarity. +Domain-specific models like SciBERT [11] and PubMedBERT [35] +can further enhance contextual understanding by incorporating +structured knowledge into retrieval and verification processes. By +leveraging ontological reasoning, knowledge graphs, and struc- +tured embeddings, scientific fact-checking systems can better align +claims with relevant evidence, reducing verification errors caused +by terminology granularity mismatches. In addition to enhancing +precision, this also has potential to improve interpretability by +making model decisions more transparent. + +9 Conclusion +This paper explores the evolution of scientific fact-checking method- +ologies from abstract-level approaches to full-paper frameworks +on large-scale corpora. As the volume and diversity of scientific +knowledge continue to grow, the challenges of verifying claims +across heterogeneous sources become increasingly complex. By +addressing the complexities inherent in scientific literature, includ- +ing its evolving nature, structured format, and the necessity for +precise evidence retrieval, we underscore the importance of de- +veloping specialised retrieval systems capable of managing large, +multimodal, and time-sensitive evidence retrieval. Furthermore, +this work proposes several research directions aimed at improving +scientific fact-checking efficiency and reliability. They include the +integration of time-aware evidence retrieval to ensure the use of the +most relevant and up-to-date findings, adaptive document process- +ing to enable context-sensitive retrieval strategies, and multi-modal +evidence alignment to integrate text, tables and figures to enhance +verification accuracy. Overcoming these challenges has potential to +improve the accuracy of fact-checking processes and also facilitate +the scalability and applicability of these systems in diverse scientific +fact-checking scenarios. By bridging the gap between scientific fact- +checking and effective evidence retrieval, these advancements will +contribute to more robust, interpretable, and trustworthy scientific +fact-checking methodologies for real-world applications. + +444 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +References + +[1] Sahar Abdelnabi, Rakibul Hasan, and Mario Fritz. 2022. Open-domain, content- +based, multi-modal fact-checking of out-of-context images via online resources. +In Proceedings of the IEEE/CVF conference on computer vision and pattern recog- +nition. 14940ś14949. + +[2] Vaibhav Adlakha, Parishad BehnamGhader, Xing Han Lu, Nicholas Meade, +and Siva Reddy. 2024. Evaluating Correctness and Faithfulness of Instruction- +Following Models for Question Answering. Transactions of the Association for +Computational Linguistics 12 (2024), 681ś699. doi:10.1162/tacl_a_00667 + +[3] Mubashara Akhtar, Oana Cocarascu, and Elena Simperl. 2022. PubHealthTab: A +Public Health Table-based Dataset for Evidence-based Fact Checking. In Findings +of the Association for Computational Linguistics: NAACL 2022, Marine Carpuat, +Marie-Catherine de Marneffe, and Ivan Vladimir Meza Ruiz (Eds.). Association +for Computational Linguistics, Seattle, United States, 1ś16. doi:10.18653/v1/ +2022.findings-naacl.1 + +[4] Mubashara Akhtar, Oana Cocarascu, and Elena Simperl. 2023. Reading and +Reasoning over Chart Images for Evidence-based Automated Fact-Checking. +In Findings of the Association for Computational Linguistics: EACL 2023, An- +dreas Vlachos and Isabelle Augenstein (Eds.). Association for Computational +Linguistics, Dubrovnik, Croatia, 399ś414. doi:10.18653/v1/2023.findings-eacl.30 +[5] Mubashara Akhtar, Michael Schlichtkrull, Zhijiang Guo, Oana Cocarascu, Elena +Simperl, and Andreas Vlachos. 2023. Multimodal Automated Fact-Checking: +A Survey. In Findings of the Association for Computational Linguistics: EMNLP +2023, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Com- +putational Linguistics, Singapore, 5430ś5448. doi:10.18653/v1/2023.findings- +emnlp.361 + +[6] Mubashara Akhtar, Michael Schlichtkrull, and Andreas Vlachos. 2024. Ev2R: +Evaluating Evidence Retrieval in Automated Fact-Checking. arXiv preprint +arXiv:2411.05375 (2024). + +[7] Mubashara Akhtar, Nikesh Subedi, Vivek Gupta, Sahar Tahmasebi, Oana Co- +carascu, and Elena Simperl. 2024. ChartCheck: Explainable Fact-Checking +over Real-World Chart Images. In Findings of the Association for Computational +Linguistics: ACL 2024, Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). +Association for Computational Linguistics, Bangkok, Thailand, 13921ś13937. +doi:10.18653/v1/2024.findings-acl.828 + +[8] Liesbeth Allein, Marlon Saelens, Ruben Cartuyvels, and Marie-Francine Moens. +2023. Implicit Temporal Reasoning for Evidence-Based Fact-Checking. In Find- +ings of the Association for Computational Linguistics: EACL 2023, Andreas Vla- +chos and Isabelle Augenstein (Eds.). Association for Computational Linguistics, +Dubrovnik, Croatia, 176ś189. doi:10.18653/v1/2023.findings-eacl.13 + +[9] Isabelle Augenstein, Christina Lioma, Dongsheng Wang, Lucas Chaves Lima, +Casper Hansen, Christian Hansen, and Jakob Grue Simonsen. 2019. MultiFC: A +Real-World Multi-Domain Dataset for Evidence-Based Fact Checking of Claims. +In Proceedings of the 2019 Conference on Empirical Methods in Natural Language +Processing and the 9th International Joint Conference on Natural Language Pro- +cessing (EMNLP-IJCNLP), Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan +(Eds.). Association for Computational Linguistics, Hong Kong, China, 4685ś4697. +doi:10.18653/v1/D19-1475 + +[10] Dara Bahri, Yi Tay, Che Zheng, Donald Metzler, and Andrew Tomkins. 2020. +Choppy: Cut transformer for ranked list truncation. In Proceedings of the 43rd +International ACM SIGIR Conference on Research and Development in Information +Retrieval. 1513ś1516. + +[11] Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. SciBERT: A Pretrained Language +Model for Scientific Text. In Proceedings of the 2019 Conference on Empirical +Methods in Natural Language Processing and the 9th International Joint Confer- +ence on Natural Language Processing (EMNLP-IJCNLP), Kentaro Inui, Jing Jiang, +Vincent Ng, and Xiaojun Wan (Eds.). Association for Computational Linguistics, +Hong Kong, China, 3615ś3620. doi:10.18653/v1/D19-1371 + +[12] Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. Longformer: The + +Long-Document Transformer. arXiv:2004.05150 (2020). + +[13] Chandra Sekhar Bhagavatula, Thanapon Noraset, and Doug Downey. 2013. +Methods for exploring and mining tables on wikipedia. In Proceedings of the +ACM SIGKDD workshop on interactive data exploration and analytics. 18ś26. + +[14] Reem Bin-Hezam and Mark Stevenson. 2023. Combining Counting Processes +and Classification Improves a Stopping Rule for Technology Assisted Review. +In Findings of the Association for Computational Linguistics: EMNLP 2023, Houda +Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational +Linguistics, Singapore, 2603ś2609. doi:10.18653/v1/2023.findings-emnlp.171 + +[15] Reem Bin-Hezam and Mark Stevenson. 2024. RLStop: A Reinforcement Learning +Stopping Method for TAR. In Proceedings of the 47th International ACM SIGIR +Conference on Research and Development in Information Retrieval. 2604ś2608. + +[16] Jannis Bulian, Jordan Boyd-Graber, Markus Leippold, Massimiliano Ciaramita, +and Thomas Diggelmann. 2020. Climate-fever: A dataset for verification of +real-world climate claims. In NeurIPS 2020 Workshop on Tackling Climate Change +with Machine Learning. + +[17] Peter Buneman, Dennis Dosso, Matteo Lissandrini, and Gianmaria Silvello. 2021. +Data citation and the citation graph. Quantitative Science Studies 2, 4 (2021), + +1399ś1422. + +[18] Rui Cao, Zifeng Ding, Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. +2025. AVerImaTeC: A Dataset for Automatic Verification of Image-Text Claims +with Evidence from the Web. arXiv:2505.17978 [cs.CL] https://arxiv.org/abs/ +2505.17978 + +[19] Catherine Chen, Zejiang Shen, Dan Klein, Gabriel Stanovsky, Doug Downey, +and Kyle Lo. 2023. Are Layout-Infused Language Models Robust to Layout +Distribution Shifts? A Case Study with Scientific Documents. In Findings of the +Association for Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd- +Graber, and Naoaki Okazaki (Eds.). Association for Computational Linguistics, +Toronto, Canada, 13345ś13360. doi:10.18653/v1/2023.findings-acl.844 + +[20] Jiawei Chen, Hongyu Lin, Xianpei Han, and Le Sun. 2024. Benchmarking large +language models in retrieval-augmented generation. In Proceedings of the AAAI +Conference on Artificial Intelligence, Vol. 38. 17754ś17762. + +[21] Qingyu Chen, Yifan Peng, and Zhiyong Lu. 2019. BioSentVec: creating sen- +tence embeddings for biomedical texts. In 2019 IEEE International Conference on +Healthcare Informatics (ICHI). IEEE, 1ś5. + +[22] Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, +Shiyang Li, Xiyou Zhou, and William Yang Wang. 2020. TabFact : A Large-scale +Dataset for Table-based Fact Verification. In International Conference on Learning +Representations (ICLR). Addis Ababa, Ethiopia. + +[23] Sabrina Chiesurin, Dimitris Dimakopoulos, Marco Antonio Sobrevilla Cabezudo, +Arash Eshghi, Ioannis Papaioannou, Verena Rieser, and Ioannis Konstas. 2023. +The Dangers of trusting Stochastic Parrots: Faithfulness and Trust in Open- +domain Conversational Question Answering. In Findings of the Association for +Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 947ś959. doi:10.18653/v1/2023.findings-acl.60 + +[24] Jaemin Cho, Debanjan Mahata, Ozan Irsoy, Yujie He, and Mohit Bansal. 2024. +M3docrag: Multi-modal retrieval is what you need for multi-page multi- +document understanding. arXiv preprint arXiv:2411.04952 (2024). + +[25] Gordon V Cormack and Maura R Grossman. 2014. Evaluation of machine- +learning protocols for technology-assisted review in electronic discovery. In +Proceedings of the 37th international ACM SIGIR conference on Research & devel- +opment in information retrieval. 153ś162. + +[26] Nick Craswell, Bhaskar Mitra, Emine Yilmaz, and Daniel Campos. 2021. +Overview of the TREC 2020 deep learning track. arXiv:2102.07662 [cs.IR] +https://arxiv.org/abs/2102.07662 + +[27] Nick Craswell, Bhaskar Mitra, Emine Yilmaz, Daniel Campos, and Ellen M +Voorhees. 2020. Overview of the TREC 2019 deep learning track. arXiv preprint +arXiv:2003.07820 (2020). + +[28] Alphaeus Dmonte, Roland Oruche, Marcos Zampieri, Prasad Calyam, and Is- +abelle Augenstein. 2024. Claim Verification in the Age of Large Language +Models: A Survey. arXiv preprint arXiv:2408.14317 (2024). + +[29] Haoyu Dong and Zhiruo Wang. 2024. Large language models for tabular data: +Progresses and future directions. In Proceedings of the 47th International ACM +SIGIR Conference on Research and Development in Information Retrieval. 2997ś +3000. + +[30] Michael Färber and Adam Jatowt. 2020. Citation recommendation: approaches + +and datasets. International Journal on Digital Libraries 21, 4 (2020), 375ś405. + +[31] Aaron HA Fletcher and Mark Stevenson. 2025. Predicting retracted research: a +dataset and machine learning approaches. Research Integrity and Peer Review +10, 1 (2025), 1ś10. + +[32] Raymond Fok, Hita Kambhamettu, Luca Soldaini, Jonathan Bragg, Kyle Lo, Marti +Hearst, Andrew Head, and Daniel S Weld. 2023. Scim: Intelligent skimming +support for scientific papers. In Proceedings of the 28th International Conference +on Intelligent User Interfaces. 476ś490. + +[33] Maura R Grossman, Gordon V Cormack, and Adam Roegiest. 2016. TREC 2016 + +Total Recall Track Overview.. In TREC. + +[34] Nianlong Gu, Yingqiang Gao, and Richard HR Hahnloser. 2022. Local citation +recommendation with hierarchical-attention text encoder and scibert-based +reranking. In European conference on information retrieval. Springer, 274ś288. +[35] Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong +Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2020. Domain-Specific +Language Model Pretraining for Biomedical Natural Language Processing. +arXiv:arXiv:2007.15779 + +[36] Zihui Gu, Ju Fan, Nan Tang, Preslav Nakov, Xiaoman Zhao, and Xiaoyong Du. +2022. PASTA: Table-Operations Aware Fact Verification via Sentence-Table +Cloze Pre-training. In Proceedings of the 2022 Conference on Empirical Methods +in Natural Language Processing, Yoav Goldberg, Zornitsa Kozareva, and Yue +Zhang (Eds.). Association for Computational Linguistics, Abu Dhabi, United +Arab Emirates, 4971ś4983. doi:10.18653/v1/2022.emnlp-main.331 + +[37] Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. 2022. A survey on +automated fact-checking. Transactions of the Association for Computational +Linguistics 10 (2022), 178ś206. + +[38] Jonathan Herzig, Pawel Krzysztof Nowak, Thomas Müller, Francesco Piccinno, +and Julian Eisenschlos. 2020. TaPas: Weakly Supervised Table Parsing via +Pre-training. In Proceedings of the 58th Annual Meeting of the Association for + +445 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +Computational Linguistics, Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel +Tetreault (Eds.). Association for Computational Linguistics, Online, 4320ś4333. +doi:10.18653/v1/2020.acl-main.398 + +[39] Jason Hoelscher-Obermaier, Edward Stevinson, Valentin Stauber, Ivaylo Zhelev, +Viktor Botev, Ronin Wu, and Jeremy Minton. 2022. Leveraging knowledge +graphs to update scientific word embeddings using latent semantic imputation. +In Proceedings of the first Workshop on Information Extraction from Scientific +Publications, Tirthankar Ghosal, Sergi Blanco-Cuaresma, Alberto Accomazzi, +Robert M. Patton, Felix Grezes, and Thomas Allen (Eds.). Association for Com- +putational Linguistics, Online, 43ś53. doi:10.18653/v1/2022.wiesp-1.6 + +[40] Xuming Hu, Zhaochen Hong, Zhijiang Guo, Lijie Wen, and Philip Yu. 2023. Read +it twice: Towards faithfully interpretable fact verification by revisiting evidence. +In Proceedings of the 46th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 2319ś2323. + +[41] Po-Wei Huang, Abhinav Ramesh Kashyap, Yanxia Qin, Yajing Yang, and Min-Yen +Kan. 2022. Lightweight Contextual Logical Structure Recovery. In Proceedings +of the Third Workshop on Scholarly Document Processing, Arman Cohan, Guy +Feigenblat, Dayne Freitag, Tirthankar Ghosal, Drahomira Herrmannova, Petr +Knoth, Kyle Lo, Philipp Mayr, Michal Shmueli-Scheuer, Anita de Waard, and +Lucy Lu Wang (Eds.). Association for Computational Linguistics, Gyeongju, +Republic of Korea, 37ś48. https://aclanthology.org/2022.sdp-1.5/ + +[42] Tomoki Ikoma and Shigeki Matsubara. 2023. On the Use of Language Models for +Function Identification of Citations in Scholarly Papers. In Proceedings of the Sec- +ond Workshop on Information Extraction from Scientific Publications, Tirthankar +Ghosal, Felix Grezes, Thomas Allen, Kelly Lockhart, Alberto Accomazzi, and +Sergi Blanco-Cuaresma (Eds.). Association for Computational Linguistics, Bali, +Indonesia, 130ś135. doi:10.18653/v1/2023.wiesp-1.15 + +[43] E. Kanoulas, Dan Li, Leif Azzopardi, and René Spijker. 2018. CLEF 2017 Tech- +nologically Assisted Reviews in Empirical Medicine Overview. In Conference +and Labs of the Evaluation Forum. https://api.semanticscholar.org/CorpusID: +246440739 + +[44] Evangelos Kanoulas, Dan Li, Leif Azzopardi, and Rene Spijker. 2018. CLEF +2018 technologically assisted reviews in empirical medicine overview. In CEUR +workshop proceedings, Vol. 2125. + +[45] Evangelos Kanoulas, Dan Li, Leif Azzopardi, and Rene Spijker. 2019. CLEF 2019 +technology assisted reviews in empirical medicine overview. In CEUR workshop +proceedings, Vol. 2380. 250. + +[46] Marzena Karpinska, Katherine Thai, Kyle Lo, Tanya Goyal, and Mohit Iyyer. +2024. One Thousand and One Pairs: A “novelž challenge for long-context +language models. In Proceedings of the 2024 Conference on Empirical Methods in +Natural Language Processing, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung +Chen (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +17048ś17085. doi:10.18653/v1/2024.emnlp-main.948 + +[47] Mohammed Abdul Khaliq, Paul Yu-Chun Chang, Mingyang Ma, Bernhard +Pflugfelder, and Filip Miletić. 2024. RAGAR, Your Falsehood Radar: RAG- +Augmented Reasoning for Political Fact-Checking using Multimodal Large +Language Models. In Proceedings of the Seventh Fact Extraction and VERifica- +tion Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +280ś296. doi:10.18653/v1/2024.fever-1.29 + +[48] Jiho Kim, Sungjin Park, Yeonsu Kwon, Yohan Jo, James Thorne, and Edward +Choi. 2023. FactKG: Fact Verification via Reasoning on Knowledge Graphs. +In Proceedings of the 61st Annual Meeting of the Association for Computational +Linguistics (Volume 1: Long Papers), Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 16190ś16206. doi:10.18653/v1/2023.acl-long.895 + +[49] Neema Kotonya and Francesca Toni. 2020. Explainable Automated Fact- +Checking for Public Health Claims. In Proceedings of the 2020 Conference on +Empirical Methods in Natural Language Processing (EMNLP), Bonnie Webber, +Trevor Cohn, Yulan He, and Yang Liu (Eds.). Association for Computational +Linguistics, Online, 7740ś7754. doi:10.18653/v1/2020.emnlp-main.623 + +[50] Xiangci Li, Gully A Burns, and Nanyun Peng. 2021. A Paragraph-level Multi-task + +Learning Model for Scientific Fact-Verification.. In SDU@ AAAI. + +[51] Xiaoxi Li, Zhicheng Dou, Yujia Zhou, and Fangchao Liu. 2024. Corpuslm: +Towards a unified language model on corpus for knowledge-intensive tasks. +In Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 26ś37. + +[52] Yen-Chieh Lien, Daniel Cohen, and W Bruce Croft. 2019. An assumption-free +approach to the dynamic truncation of ranked lists. In Proceedings of the 2019 +ACM SIGIR International Conference on Theory of Information Retrieval. 79ś82. +[53] Donald AB Lindberg, Betsy L Humphreys, and Alexa T McCray. 1993. The +unified medical language system. Yearbook of medical informatics 2, 01 (1993), +41ś51. + +[54] Carolyn E Lipscomb. 2000. Medical subject headings (MeSH). Bulletin of the + +Medical Library Association 88, 3 (2000), 265. + +[55] Qi Liu, Bo Wang, Nan Wang, and Jiaxin Mao. 2025. Leveraging Passage Em- +beddings for Efficient Listwise Reranking with Large Language Models. In Pro- +ceedings of the ACM on Web Conference 2025 (Sydney NSW, Australia) (WWW +’25). Association for Computing Machinery, New York, NY, USA, 4274ś4283. +doi:10.1145/3696410.3714554 + +[56] Kyle Lo, Joseph Chee Chang, Andrew Head, Jonathan Bragg, Amy X. Zhang, Cas- +sidy Trier, Chloe Anastasiades, Tal August, Russell Authur, Danielle Bragg, Erin +Bransom, Isabel Cachola, Stefan Candra, Yoganand Chandrasekhar, Yen-Sung +Chen, Evie Yu-Yen Cheng, Yvonne Chou, Doug Downey, Rob Evans, Raymond +Fok, Fangzhou Hu, Regan Huff, Dongyeop Kang, Tae Soo Kim, Rodney Kinney, +Aniket Kittur, Hyeonsu B. Kang, Egor Klevak, Bailey Kuehl, Michael J. Lan- +gan, Matt Latzke, Jaron Lochner, Kelsey MacMillan, Eric Marsh, Tyler Murray, +Aakanksha Naik, Ngoc-Uyen Nguyen, Srishti Palani, Soya Park, Caroline Paulic, +Napol Rachatasumrit, Smita Rao, Paul Sayre, Zejiang Shen, Pao Siangliulue, +Luca Soldaini, Huy Tran, Madeleine van Zuylen, Lucy Lu Wang, Christopher +Wilhelm, Caroline Wu, Jiangjiang Yang, Angele Zamarron, Marti A. Hearst, and +Daniel S. Weld. 2024. The Semantic Reader Project. Commun. ACM 67, 10 (Sept. +2024), 50ś61. doi:10.1145/3659096 + +[57] Xinyuan Lu, Liangming Pan, Qian Liu, Preslav Nakov, and Min-Yen Kan. 2023. +SCITAB: A Challenging Benchmark for Compositional Reasoning and Claim +Verification on Scientific Tables. In Proceedings of the 2023 Conference on Em- +pirical Methods in Natural Language Processing, Houda Bouamor, Juan Pino, +and Kalika Bali (Eds.). Association for Computational Linguistics, Singapore, +7787ś7813. doi:10.18653/v1/2023.emnlp-main.483 + +[58] María Luengo and David García-Marín. 2020. The performance of truth: politi- +cians, fact-checking journalism, and the struggle to tackle COVID-19 misinfor- +mation. American Journal of Cultural Sociology 8, 3 (2020), 405. + +[59] Yixiao Ma, Qingyao Ai, Yueyue Wu, Yunqiu Shao, Yiqun Liu, Min Zhang, and +Shaoping Ma. 2022. Incorporating retrieval information into the truncation +of ranking lists for better legal search. In Proceedings of the 45th International +ACM SIGIR Conference on Research and Development in Information Retrieval. +438ś448. + +[60] Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, +Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, Pan Zhang, Liangming Pan, Yu- +Gang Jiang, Jiaqi Wang, Yixin Cao, and Aixin Sun. 2024. MMLongBench-Doc: +Benchmarking Long-context Document Understanding with Visualizations. +arXiv:2407.01523 [cs.CV] https://arxiv.org/abs/2407.01523 + +[61] Chuan Meng, Negar Arabzadeh, Arian Askari, Mohammad Aliannejadi, and +Maarten de Rijke. 2024. Ranked list truncation for large language model-based +re-ranking. In Proceedings of the 47th International ACM SIGIR Conference on +Research and Development in Information Retrieval. 141ś151. + +[62] Isabelle Mohr, Amelie Wührl, and Roman Klinger. 2022. CoVERT: A Corpus +of Fact-checked Biomedical COVID-19 Tweets. In Proceedings of the Thirteenth +Language Resources and Evaluation Conference, Nicoletta Calzolari, Frédéric +Béchet, Philippe Blache, Khalid Choukri, Christopher Cieri, Thierry Declerck, +Sara Goggi, Hitoshi Isahara, Bente Maegaard, Joseph Mariani, Hélène Mazo, Jan +Odijk, and Stelios Piperidis (Eds.). European Language Resources Association, +Marseille, France, 244ś257. https://aclanthology.org/2022.lrec-1.26 + +[63] Yuki Momii, Tetsuya Takiguchi, and Yasuo Ariki. 2024. RAG-Fusion Based +Information Retrieval for Fact-Checking. In Proceedings of the Seventh Fact Ex- +traction and VERification Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, +Chenxi Whitehouse, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang +Guo, Christos Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, +and Andreas Vlachos (Eds.). Association for Computational Linguistics, Miami, +Florida, USA, 47ś54. doi:10.18653/v1/2024.fever-1.4 + +[64] Rodrigo Nogueira, Zhiying Jiang, Ronak Pradeep, and Jimmy Lin. 2020. Docu- +ment Ranking with a Pretrained Sequence-to-Sequence Model. In Findings of the +Association for Computational Linguistics: EMNLP 2020, Trevor Cohn, Yulan He, +and Yang Liu (Eds.). Association for Computational Linguistics, Online, 708ś718. +doi:10.18653/v1/2020.findings-emnlp.63 + +[65] Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and +Panagiotis Petrantonakis. 2023. Synthetic misinformers: Generating and com- +bating multimodal misinformation. In Proceedings of the 2nd ACM International +Workshop on Multimedia AI against Disinformation. 36ś44. + +[66] Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and +Panagiotis C Petrantonakis. 2024. VERITE: a Robust benchmark for multimodal +misinformation detection accounting for unimodal bias. International Journal +of Multimedia Information Retrieval 13, 1 (2024), 4. + +[67] Andrew Parry, Debasis Ganguly, and Manish Chandra. 2024. In-Context Learn- +ing" or: How I learned to stop worrying and love" Applied Information Retrieval. +In Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 14ś25. + +[68] Ronak Pradeep, Xueguang Ma, Rodrigo Nogueira, and Jimmy Lin. 2021. Scientific +Claim Verification with VerT5erini. In Proceedings of the 12th International +Workshop on Health Text Mining and Information Analysis, Eben Holderness, +Antonio Jimeno Yepes, Alberto Lavelli, Anne-Lyse Minard, James Pustejovsky, +and Fabio Rinaldi (Eds.). Association for Computational Linguistics, online, +94ś103. https://aclanthology.org/2021.louhi-1.11/ + +446 + + ICTIR ’25, July 18, 2025, Padua, Italy + +Xingyu Deng, Xi Wang, Mark Stevenson + +[69] Peng Qi, Zehong Yan, Wynne Hsu, and Mong Li Lee. 2024. SNIFFER: Multi- +modal Large Language Model for Explainable Out-of-Context Misinformation +Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and +Pattern Recognition. 13052ś13062. + +[70] Vatsal Raina and Mark Gales. 2024. Question-Based Retrieval using Atomic +Units for Enterprise RAG. In Proceedings of the Seventh Fact Extraction and VER- +ification Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +219ś233. doi:10.18653/v1/2024.fever-1.25 + +[71] Jonathan Roberts, Kai Han, Neil Houlsby, and Samuel Albanie. 2024. SciFIBench: +Benchmarking Large Multimodal Models for Scientific Figure Interpretation. +arXiv preprint arXiv:2405.08807 (2024). + +[72] Arkadiy Saakyan, Tuhin Chakrabarty, and Smaranda Muresan. 2021. COVID- +Fact: Fact Extraction and Verification of Real-World Claims on COVID-19 Pan- +demic. In Proceedings of the 59th Annual Meeting of the Association for Com- +putational Linguistics and the 11th International Joint Conference on Natural +Language Processing (Volume 1: Long Papers), Chengqing Zong, Fei Xia, Wen- +jie Li, and Roberto Navigli (Eds.). Association for Computational Linguistics, +Online, 2116ś2129. doi:10.18653/v1/2021.acl-long.165 + +[73] Alireza Salemi and Hamed Zamani. 2024. Evaluating retrieval quality in retrieval- +augmented generation. In Proceedings of the 47th International ACM SIGIR Con- +ference on Research and Development in Information Retrieval. 2395ś2400. +[74] Alireza Salemi and Hamed Zamani. 2024. Towards a search engine for machines: +Unified ranking for multiple retrieval-augmented large language models. In +Proceedings of the 47th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 741ś751. + +[75] Mourad Sarrouti, Asma Ben Abacha, Yassine Mrabet, and Dina Demner- +Fushman. 2021. Evidence-based Fact-Checking of Health-related Claims. In +Findings of the Association for Computational Linguistics: EMNLP 2021, Marie- +Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (Eds.). +Association for Computational Linguistics, Punta Cana, Dominican Republic, +3499ś3512. doi:10.18653/v1/2021.findings-emnlp.297 + +[76] Artsiom Sauchuk, James Thorne, Alon Halevy, Nicola Tonellotto, and Fabrizio +Silvestri. 2022. On the role of relevance in natural language processing tasks. +In Proceedings of the 45th International ACM SIGIR Conference on Research and +Development in Information Retrieval. 1785ś1789. + +[77] Michael Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, +Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana +Cocarascu, Arpit Mittal, James Thorne, and Andreas Vlachos. 2024. The Auto- +mated Verification of Textual Claims (AVeriTeC) Shared Task. In Proceedings +of the Seventh Fact Extraction and VERification Workshop (FEVER), Michael +Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, Mubashara +Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana Cocarascu, +Arpit Mittal, James Thorne, and Andreas Vlachos (Eds.). Association for Compu- +tational Linguistics, Miami, Florida, USA, 1ś26. doi:10.18653/v1/2024.fever-1.1 +[78] Michael Schlichtkrull, Zhijiang Guo, and Andreas Vlachos. 2024. Averitec: A +dataset for real-world claim verification with evidence from the web. Advances +in Neural Information Processing Systems 36 (2024). + +[79] Michael Sejr Schlichtkrull, Vladimir Karpukhin, Barlas Oguz, Mike Lewis, Wen- +tau Yih, and Sebastian Riedel. 2021. Joint Verification and Reranking for Open +Fact Checking Over Tables. In Proceedings of the 59th Annual Meeting of the +Association for Computational Linguistics and the 11th International Joint Confer- +ence on Natural Language Processing (Volume 1: Long Papers), Chengqing Zong, +Fei Xia, Wenjie Li, and Roberto Navigli (Eds.). Association for Computational +Linguistics, Online, 6787ś6799. doi:10.18653/v1/2021.acl-long.529 + +[80] Özge Sevgili, Irina Nikishina, Seid Muhie Yimam, Martin Semmann, and +Chris Biemann. 2024. UHH at AVeriTeC: RAG for Fact-Checking with Real- +World Claims. In Proceedings of the Seventh Fact Extraction and VERifica- +tion Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +55ś63. doi:10.18653/v1/2024.fever-1.5 + +[81] Zejiang Shen, Kyle Lo, Lucy Lu Wang, Bailey Kuehl, Daniel S. Weld, and Doug +Downey. 2022. VILA: Improving Structured Content Extraction from Scien- +tific PDFs Using Visual Layout Groups. Transactions of the Association for +Computational Linguistics 10 (2022), 376ś392. doi:10.1162/tacl_a_00466 +[82] Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H +Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be +easily distracted by irrelevant context. In International Conference on Machine +Learning. PMLR, 31210ś31227. + +[83] Qi Shi, Yu Zhang, Qingyu Yin, and Ting Liu. 2021. Logic-level Evidence Retrieval +and Graph-based Verification Network for Table-based Fact Verification. In +Proceedings of the 2021 Conference on Empirical Methods in Natural Language +Processing, Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott +Wen-tau Yih (Eds.). Association for Computational Linguistics, Online and Punta + +Cana, Dominican Republic, 175ś184. doi:10.18653/v1/2021.emnlp-main.16 +[84] Ronit Singal, Pransh Patwa, Parth Patwa, Aman Chadha, and Amitava Das. +2024. Evidence-backed Fact Checking using RAG and Few-Shot In-Context +Learning with LLMs. In Proceedings of the Seventh Fact Extraction and VERifi- +cation Workshop (FEVER), Michael Schlichtkrull, Yulong Chen, Chenxi White- +house, Zhenyun Deng, Mubashara Akhtar, Rami Aly, Zhijiang Guo, Christos +Christodoulopoulos, Oana Cocarascu, Arpit Mittal, James Thorne, and Andreas +Vlachos (Eds.). Association for Computational Linguistics, Miami, Florida, USA, +91ś98. doi:10.18653/v1/2024.fever-1.10 + +[85] Mark Stevenson and Reem Bin-Hezam. 2023. Stopping Methods for Technology- +assisted Reviews Based on Point Processes. ACM Transactions on Information +Systems 42, 3 (2023), 1ś37. + +[86] Weiwei Sun, Lingyong Yan, Xinyu Ma, Shuaiqiang Wang, Pengjie Ren, Zhumin +Chen, Dawei Yin, and Zhaochun Ren. 2023. +Is ChatGPT Good at Search? +Investigating Large Language Models as Re-Ranking Agents. In Proceedings of +the 2023 Conference on Empirical Methods in Natural Language Processing, Houda +Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational +Linguistics, Singapore, 14918ś14937. doi:10.18653/v1/2023.emnlp-main.923 +[87] Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida, Taku Hasegawa, Itsumi Saito, +and Kuniko Saito. 2023. Slidevqa: A dataset for document visual question +answering on multiple images. In Proceedings of the AAAI Conference on Artificial +Intelligence, Vol. 37. 13636ś13645. + +[88] Liyan Tang, Philippe Laban, and Greg Durrett. 2024. MiniCheck: Efficient Fact- +Checking of LLMs on Grounding Documents. In Proceedings of the 2024 Confer- +ence on Empirical Methods in Natural Language Processing, Yaser Al-Onaizan, +Mohit Bansal, and Yun-Nung Chen (Eds.). Association for Computational Lin- +guistics, Miami, Florida, USA, 8818ś8847. doi:10.18653/v1/2024.emnlp-main.499 +[89] Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna +Gurevych. 2021. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of +Information Retrieval Models. In Thirty-fifth Conference on Neural Information +Processing Systems Datasets and Benchmarks Track (Round 2). https://openreview. +net/forum?id=wCu6T5xFjeJ + +[90] James Thorne, Andreas Vlachos, Christos Christodoulopoulos, and Arpit Mittal. +2018. FEVER: a Large-scale Dataset for Fact Extraction and VERification. In Pro- +ceedings of the 2018 Conference of the North American Chapter of the Association +for Computational Linguistics: Human Language Technologies, Volume 1 (Long Pa- +pers), Marilyn Walker, Heng Ji, and Amanda Stent (Eds.). Association for Compu- +tational Linguistics, New Orleans, Louisiana, 809ś819. doi:10.18653/v1/N18-1074 +[91] James Thorne, Andreas Vlachos, Oana Cocarascu, Christos Christodoulopou- +los, and Arpit Mittal. 2018. The Fact Extraction and VERification (FEVER) +Shared Task. In Proceedings of the First Workshop on Fact Extraction and VER- +ification (FEVER), James Thorne, Andreas Vlachos, Oana Cocarascu, Christos +Christodoulopoulos, and Arpit Mittal (Eds.). Association for Computational +Linguistics, Brussels, Belgium, 1ś9. doi:10.18653/v1/W18-5501 + +[92] Fangzheng Tian, Debasis Ganguly, and Craig Macdonald. 2025. Is Relevance +Propagated from Retriever to Generator in RAG?. In Advances in Information +Retrieval: 47th European Conference on Information Retrieval, ECIR 2025, Lucca, +Italy, April 6ś10, 2025, Proceedings, Part I (Lucca, Italy). Springer-Verlag, Berlin, +Heidelberg, 32ś48. doi:10.1007/978-3-031-88708-6_3 + +[93] Jonathan Tonglet, Marie-Francine Moens, and Iryna Gurevych. 2024. “Image, +Tell me your story!ž Predicting the original meta-context of visual misinfor- +mation. In Proceedings of the 2024 Conference on Empirical Methods in Natural +Language Processing, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen +(Eds.). Association for Computational Linguistics, Miami, Florida, USA, 7845ś +7864. doi:10.18653/v1/2024.emnlp-main.448 + +[94] Herbert Ullrich, Tomáš Mlynář, and Jan Drchal. 2024. AIC CTU system at +AVeriTeC: Re-framing automated fact-checking as a simple RAG task. In Proceed- +ings of the Seventh Fact Extraction and VERification Workshop (FEVER), Michael +Schlichtkrull, Yulong Chen, Chenxi Whitehouse, Zhenyun Deng, Mubashara +Akhtar, Rami Aly, Zhijiang Guo, Christos Christodoulopoulos, Oana Cocarascu, +Arpit Mittal, James Thorne, and Andreas Vlachos (Eds.). Association for Compu- +tational Linguistics, Miami, Florida, USA, 137ś150. doi:10.18653/v1/2024.fever- +1.16 + +[95] Jordy Van Landeghem, Rubèn Tito, Łukasz Borchmann, Michał Pietruszka, Pawel +Joziak, Rafal Powalski, Dawid Jurkiewicz, Mickaël Coustaty, Bertrand Anckaert, +Ernest Valveny, et al. 2023. Document understanding dataset and evaluation +(dude). In Proceedings of the IEEE/CVF International Conference on Computer +Vision. 19528ś19540. + +[96] Vijay Viswanathan, Graham Neubig, and Pengfei Liu. 2021. CitationIE: Lever- +aging the Citation Graph for Scientific Information Extraction. In Proceedings +of the 59th Annual Meeting of the Association for Computational Linguistics +and the 11th International Joint Conference on Natural Language Processing +(Volume 1: Long Papers), Chengqing Zong, Fei Xia, Wenjie Li, and Roberto +Navigli (Eds.). Association for Computational Linguistics, Online, 719ś731. +doi:10.18653/v1/2021.acl-long.59 + +[97] Juraj Vladika and Florian Matthes. 2023. Scientific Fact-Checking: A Survey +of Resources and Approaches. In Findings of the Association for Computational +Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and Naoaki Okazaki + +447 + + The Next Phase of Scientific Fact-Checking: Advanced Evidence Retrieval from Complex Structured Academic Papers + +ICTIR ’25, July 18, 2025, Padua, Italy + +SIGIR Conference on Research and Development in Information Retrieval. 2733ś +2743. + +[113] Yunhu Ye, Binyuan Hui, Min Yang, Binhua Li, Fei Huang, and Yongbin Li. 2023. +Large language models are versatile decomposers: Decomposing evidence and +questions for table-based reasoning. In Proceedings of the 46th International +ACM SIGIR Conference on Research and Development in Information Retrieval. +174ś184. + +[114] Pengcheng Yin, Graham Neubig, Wen-tau Yih, and Sebastian Riedel. 2020. +TaBERT: Pretraining for Joint Understanding of Textual and Tabular Data. In Pro- +ceedings of the 58th Annual Meeting of the Association for Computational Linguis- +tics, Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault (Eds.). Associa- +tion for Computational Linguistics, Online, 8413ś8426. doi:10.18653/v1/2020.acl- +main.745 + +[115] Hamed Zamani and Michael Bendersky. 2024. Stochastic rag: End-to-end +retrieval-augmented generation through expected utility maximization. In Pro- +ceedings of the 47th International ACM SIGIR Conference on Research and Devel- +opment in Information Retrieval. 2641ś2646. + +[116] Xia Zeng, Amani S Abumansour, and Arkaitz Zubiaga. 2021. Automated fact- + +checking: A survey. Language and Linguistics Compass 15, 10 (2021), e12438. + +[117] Hengran Zhang, Ruqing Zhang, Jiafeng Guo, Maarten de Rijke, Yixing Fan, and +Xueqi Cheng. 2023. From Relevance to Utility: Evidence Retrieval with Feedback +for Fact Verification. In Findings of the Association for Computational Linguistics: +EMNLP 2023, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association +for Computational Linguistics, Singapore, 6373ś6384. doi:10.18653/v1/2023. +findings-emnlp.422 + +[118] Tianmai M Zhang and Neil F Abernethy. 2024. Detecting Reference Errors in Sci- +entific Literature with Large Language Models. arXiv preprint arXiv:2411.06101 +(2024). + +[119] Zhiwei Zhang, Jiyi Li, Fumiyo Fukumoto, and Yanming Ye. 2021. Abstract, Ra- +tionale, Stance: A Joint Model for Scientific Claim Verification. In Proceedings of +the 2021 Conference on Empirical Methods in Natural Language Processing, Marie- +Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (Eds.). +Association for Computational Linguistics, Online and Punta Cana, Dominican +Republic, 3580ś3586. doi:10.18653/v1/2021.emnlp-main.290 + +[120] Liwen Zheng, Chaozhuo Li, Xi Zhang, Yu-Ming Shang, Feiran Huang, and +Haoran Jia. 2024. Evidence Retrieval is almost All You Need for Fact Verifi- +cation. In Findings of the Association for Computational Linguistics ACL 2024, +Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). Association for Com- +putational Linguistics, Bangkok, Thailand and virtual meeting, 9274ś9281. +https://aclanthology.org/2024.findings-acl.551 + +(Eds.). Association for Computational Linguistics, Toronto, Canada, 6215ś6230. +doi:10.18653/v1/2023.findings-acl.387 + +[98] Juraj Vladika and Florian Matthes. 2024. Comparing Knowledge Sources for +Open-Domain Scientific Claim Verification. In Proceedings of the 18th Conference +of the European Chapter of the Association for Computational Linguistics (Volume +1: Long Papers), Yvette Graham and Matthew Purver (Eds.). Association for +Computational Linguistics, St. Julian’s, Malta, 2103ś2114. https://aclanthology. +org/2024.eacl-long.128/ + +[99] Juraj Vladika and Florian Matthes. 2024. Improving Health Question Answering +with Reliable and Time-Aware Evidence Retrieval. In Findings of the Association +for Computational Linguistics: NAACL 2024, Kevin Duh, Helena Gomez, and +Steven Bethard (Eds.). Association for Computational Linguistics, Mexico City, +Mexico, 4752ś4763. doi:10.18653/v1/2024.findings-naacl.295 + +[100] David Wadden, Shanchuan Lin, Kyle Lo, Lucy Lu Wang, Madeleine van Zuylen, +Arman Cohan, and Hannaneh Hajishirzi. 2020. Fact or Fiction: Verifying Sci- +entific Claims. In Proceedings of the 2020 Conference on Empirical Methods in +Natural Language Processing (EMNLP), Bonnie Webber, Trevor Cohn, Yulan +He, and Yang Liu (Eds.). Association for Computational Linguistics, Online, +7534ś7550. doi:10.18653/v1/2020.emnlp-main.609 + +[101] David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Iz Beltagy, Lucy Lu Wang, +and Hannaneh Hajishirzi. 2022. SciFact-Open: Towards open-domain scientific +claim verification. In Findings of the Association for Computational Linguistics: +EMNLP 2022, Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang (Eds.). As- +sociation for Computational Linguistics, Abu Dhabi, United Arab Emirates, +4719ś4734. doi:10.18653/v1/2022.findings-emnlp.347 + +[102] David Wadden, Kyle Lo, Lucy Lu Wang, Arman Cohan, Iz Beltagy, and Han- +naneh Hajishirzi. 2022. MultiVerS: Improving scientific claim verification with +weak supervision and full-document context. In Findings of the Association +for Computational Linguistics: NAACL 2022, Marine Carpuat, Marie-Catherine +de Marneffe, and Ivan Vladimir Meza Ruiz (Eds.). Association for Computational +Linguistics, Seattle, United States, 61ś76. doi:10.18653/v1/2022.findings-naacl.6 +[103] Dong Wang, Jianxin Li, Tianchen Zhu, Haoyi Zhou, Qishan Zhu, Yuxin Wen, +and Hongming Piao. 2022. MtCut: A Multi-Task Framework for Ranked List +Truncation. In Proceedings of the Fifteenth ACM International Conference on Web +Search and Data Mining. 1054ś1062. + +[104] Gengyu Wang, Kate Harwood, Lawrence Chillrud, Amith Ananthram, Melanie +Subbiah, and Kathleen McKeown. 2023. Check-COVID: Fact-Checking COVID- +19 News Claims with Scientific Evidence. In Findings of the Association for +Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and +Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, +Canada, 14114ś14127. doi:10.18653/v1/2023.findings-acl.888 + +[105] Yuxia Wang, Revanth Gangi Reddy, Zain Muhammad Mujahid, Arnav Arora, +Aleksandr Rubashevskii, Jiahui Geng, Osama Mohammed Afzal, Liangming +Pan, Nadav Borenstein, Aditya Pillai, Isabelle Augenstein, Iryna Gurevych, and +Preslav Nakov. 2024. Factcheck-Bench: Fine-Grained Evaluation Benchmark +for Automatic Fact-checkers. In Findings of the Association for Computational +Linguistics: EMNLP 2024, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen +(Eds.). Association for Computational Linguistics, Miami, Florida, USA, 14199ś +14230. doi:10.18653/v1/2024.findings-emnlp.830 + +[106] Zhiruo Wang, Jun Araki, Zhengbao Jiang, Md Rizwan Parvez, and Graham +Neubig. 2023. Learning to filter context for retrieval-augmented generation. +arXiv preprint arXiv:2311.08377 (2023). + +[107] Jevin D West and Carl T Bergstrom. 2021. Misinformation in and about science. + +Proceedings of the National Academy of Sciences 118, 15 (2021), e1912444117. + +[108] Dustin Wright and Isabelle Augenstein. 2021. CiteWorth: Cite-Worthiness +Detection for Improved Scientific Document Understanding. In Findings of the +Association for Computational Linguistics: ACL-IJCNLP 2021, Chengqing Zong, +Fei Xia, Wenjie Li, and Roberto Navigli (Eds.). Association for Computational +Linguistics, Online, 1796ś1807. doi:10.18653/v1/2021.findings-acl.157 +[109] Chen Wu, Ruqing Zhang, Jiafeng Guo, Yixing Fan, Yanyan Lan, and Xueqi +Cheng. 2021. Learning to truncate ranked lists for information retrieval. In +Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4453ś4461. +[110] Siwei Wu, Yizhi Li, Kang Zhu, Ge Zhang, Yiming Liang, Kaijing Ma, Cheng- +hao Xiao, Haoran Zhang, Bohao Yang, Wenhu Chen, Wenhao Huang, Noura +Al Moubayed, Jie Fu, and Chenghua Lin. 2024. SciMMIR: Benchmarking Sci- +entific Multi-modal Information Retrieval. In Findings of the Association for +Computational Linguistics: ACL 2024, Lun-Wei Ku, Andre Martins, and Vivek +Srikumar (Eds.). Association for Computational Linguistics, Bangkok, Thailand, +12560ś12574. doi:10.18653/v1/2024.findings-acl.746 + +[111] Amelie Wührl and Roman Klinger. 2022. Entity-based Claim Representation +Improves Fact-Checking of Medical Content in Tweets. In Proceedings of the 9th +Workshop on Argument Mining, Gabriella Lapesa, Jodi Schneider, Yohan Jo, and +Sougata Saha (Eds.). International Conference on Computational Linguistics, +Online and in Gyeongju, Republic of Korea, 187ś198. https://aclanthology.org/ +2022.argmining-1.18/ + +[112] Barry Menglong Yao, Aditya Shah, Lichao Sun, Jin-Hee Cho, and Lifu Huang. +2023. End-to-end multimodal fact-checking and explanation generation: A +challenging dataset and models. In Proceedings of the 46th International ACM + +448 diff --git a/q135/random_k4/random_2.md b/q135/random_k4/random_2.md index e2f4690064e5381fbfbef3012bd5ce5c92984a53..fd494fa8bcb0b303627f360573c5bd907a96e925 100644 --- a/q135/random_k4/random_2.md +++ b/q135/random_k4/random_2.md @@ -1,638 +1,3081 @@ -Electronic Arts Reports Q4 and Full Year FY19 Financial Results -REDWOOD CITY, CA – May 7, 2019 – Electronic Arts Inc. (NASDAQ: EA) today announced -preliminary financial results for its fourth fiscal quarter and fiscal year ended March 31, 2019. -News and ongoing updates regarding EA and its games are available on EA’s blog at -www.ea.com/news. -“FY20 will be a year of continued growth for our industry, and we’re excited to bring more great -entertainment to the expanding audience around the world,” said CEO Andrew Wilson. “Players are -engaging with games in more ways than ever before. We’re committed to meeting them where they -are with a broad portfolio of amazing new games and live services, choice of engagement models -including free-to-play and subscriptions, and new opportunities to play, compete and watch.” -“We ended our fiscal year with strong growth, driven by our live services business which continued to -provide a very large, stable and profitable source of net bookings,” said COO and CFO Blake -Jorgensen. “Looking ahead, we will bring games to an extraordinary variety of players in fiscal 2020 -and deliver on our commitment to players with deep live services content for Ultimate Team, Apex -Legends, The Sims 4, Anthem and Battlefield V.” -Selected Operating Highlights and Metrics -• Digital net bookings* for the trailing twelve months was $3.722 billion, up 5% year-over-year -and represents 75% of total net bookings. -• During the quarter, launched two new original IP titles, Apex LegendsTM and AnthemTM. -• During the quarter, launched Firestorm battle royale in Battlefield TM V, the biggest -BattlefieldTM live service event ever. -• In FY19, more than 45 million unique players engaged in FIFA 19 and FIFA 18 on console -and PC. -• In FY19, active players in The Sims TM 4 increased by 35% year-over-year. -* Net bookings is defined as the net amount of products and services sold digitally or sold-in -physically in the period. Net bookings is calculated by adding total net revenue to the change in -deferred net revenue for online-enabled games and, for periods after the fourth quarter of fiscal 2018, -mobile platform fees. -Selected Financial Highlights and Metrics -All financial measures are presented on a GAAP basis. -• Net cash from operating activities was $599 million for the quarter and $1.547 billion for the -trailing twelve months. -• EA repurchased 3.2 million shares for $301 million during the quarter and 11.0 million shares -for $1.192 billion during the trailing twelve months. -Impact of Recently Adopted Accounting Standard -At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), -Revenue from Contracts with Customers. For more information about the adoption of Topic 606, -please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April -1, 2018 has not been restated. - -Quarterly Financial Highlights -Three Months Ended -March 31, -| | | 2019 | 2018 | | | -| --- | --- | ---- | ---- | --- | --- | -(in $ millions, except per share amounts) -| Digital net revenue | | 1,065 | 1,102 | | | -| ------------------------------------ | --- | ------ | ------ | --- | --- | -| Packaged goods and other net revenue | | 173 | 480 | | | -| Total net revenue | | 1,238 | 1,582 | | | - -| Net income | | 209 | 607 | | | -| -------------------------- | --- | ----- | ----- | --- | --- | -| Diluted earnings per share | | 0.69 | 1.95 | | | - -| Operating cash flow | | 599 | 615 | | | -| ---------------------------- | --- | ---- | ----- | --- | --- | -| Value of shares repurchased | | 301 | 148 | | | -| Number of shares repurchased | | 3.2 | 1.2 | | | - -The following GAAP-based financial data and tax rate of 18% was used internally by company -management to adjust its fiscal 2019 GAAP results in order to assess EA’s operating results: - -Three Months Ended March 31, 2019 -GAAP-Based Financial Data -Change in -deferred -net revenue -| | Statement | Acquisition- | (online- | Mobile | | -| --------------- | ---------- | ------------ | -------- | --------- | ------------ | -| | of | related | enabled | platform | Stock-based | -| (in $ millions) | Operations | expenses | games) | fees | compensation | -Total net revenue 1 ,238 - 1 73 ( 47) - - Cost of revenue 2 76 ( 1) - ( 47) ( 1) -Gross profit 9 62 1 1 73 - 1 - Total operating expenses 7 66 ( 16) - - ( 72) -Operating income 1 96 1 7 1 73 - 7 3 -Interest and other income, net 2 3 - - - - -Income before provision for income taxes 2 19 1 7 1 73 - 7 3 -Number of shares used in computation: -| Diluted | | 301 | | | | -| ------------ | --- | --- | --- | --- | --- | - -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. +Direct Connect + +Service Overview + +Issue + +Date + +01 + +2026-02-27 + +HUAWEI CLOUD COMPUTING TECHNOLOGIES CO., LTD. + + Copyright © Huawei Cloud Computing Technologies Co., Ltd. 2026. All rights reserved. + +No part of this document may be reproduced or transmitted in any form or by any means without prior +written consent of Huawei Cloud Computing Technologies Co., Ltd. + +Trademarks and Permissions + + and other Huawei trademarks are the property of Huawei Technologies Co., Ltd. + +All other trademarks and trade names mentioned in this document are the property of their respective +holders. + +Notice + +The purchased products, services and features are stipulated by the contract made between Huawei +Cloud and the customer. All or part of the products, services and features described in this document may +not be within the purchase scope or the usage scope. Unless otherwise specified in the contract, all +statements, information, and recommendations in this document are provided "AS IS" without +warranties, guarantees or representations of any kind, either express or implied. + +The information in this document is subject to change without notice. Every effort has been made in the +preparation of this document to ensure accuracy of the contents, but all statements, information, and +recommendations in this document do not constitute a warranty of any kind, express or implied. + +Huawei Cloud Computing Technologies Co., Ltd. + +Address: + +Huawei Cloud Data Center Jiaoxinggong Road +Qianzhong Avenue +Gui'an New District +Gui Zhou 550029 +People's Republic of China + +Website: + +https://www.huaweicloud.com/intl/en-us/ + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +i + + Direct Connect +Service Overview + +Contents + +Contents + +1 What Is Direct Connect?........................................................................................................ 1 + +2 Product Advantages................................................................................................................ 3 + +3 Application Scenarios............................................................................................................. 4 + +4 Functions................................................................................................................................... 6 + +5 Direct Connect Locations.....................................................................................................10 + +6 Security.................................................................................................................................... 15 +6.1 Identity Authentication and Access Control................................................................................................................ 15 + +6.2 Auditing and Logging.......................................................................................................................................................... 16 + +6.3 Monitoring Security Risks.................................................................................................................................................. 16 + +6.4 Certificates...............................................................................................................................................................................16 + +7 Permissions............................................................................................................................. 18 + +8 Notes and Constraints.......................................................................................................... 26 + +9 Direct Connect and Other Services................................................................................... 29 + +10 Basic Concepts..................................................................................................................... 31 + +11 Region and AZ......................................................................................................................34 + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +ii + + Direct Connect +Service Overview + +1 What Is Direct Connect? + +1 What Is Direct Connect? + +Direct Connect allows you to establish a stable, high-speed, low-latency, secure +dedicated network connection that connects your on-premises data center to +Huawei Cloud. Direct Connect allows you to maximize legacy IT facilities and +leverage cloud services to build a flexible, scalable hybrid cloud computing +environment. + +Figure 1-1 shows how Direct Connect connects an on-premises data center to the +cloud. + +Figure 1-1 How Direct Connect works + +Why Direct Connect? + +● Network quality: Direct Connect allows you to establish a dedicated network + +for data transmission, which brings high network performance, low latency, +and excellent user experience. + +● + +● + +Security: Direct Connect establishes private connectivity between your on- +premises data center and the cloud. Data is transmitted over a secure +dedicated connection. + +Transmission speed: A single connection supports up to 100 Gbit/s of +bandwidth. + +Components + +● + +Connection + +A dedicated network connection is established between your on-premises +data center and a Direct Connect location over a line you leased from a +carrier. You can create a standard connection by yourself or request a hosted +connection from a partner. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +1 + + Direct Connect +Service Overview + +1 What Is Direct Connect? + +– + +– + +A standard connection has a dedicated port for your exclusive use and +can be associated with multiple virtual interfaces. + +A hosted connection is created by a carrier (a partner of Huawei Cloud), +who allocates VLAN and bandwidth resources for you. When using this +type of connections, you share a port with other users, and can create +only one virtual interface for a connection. + +● + +Virtual gateway + +A virtual gateway is a logical gateway for accessing VPCs. A virtual gateway +can be associated with only one VPC, and multiple connections can access a +VPC through the same virtual gateway. + +● Global DC gateway + +A global DC gateway enables your on-premises data center to access VPCs in +multiple regions so you can use a single connection to provide high-speed +access to cloud compute and storage resources in any region. + +A global DC gateway can only be associated with connections terminated at +the same Direct Connect location. If there are multiple connections +terminated at different Direct Connect locations, you need to create multiple +global DC gateways. + +● + +Virtual interface + +A virtual interface links a connection with one or more virtual gateways, so +that your on-premises network can access all the VPCs. + +Accessing Direct Connect + +The public cloud provides a web-based user interface, the management console, +for you to access the Direct Connect service. + +● + +● + +If you already have an account, log in to the management console and +choose Networking > Direct Connect on the homepage. + +If you do not have an account, you can sign up for one by referring to +Getting Started. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. 2 -TTM Financial Highlights -Twelve Months Ended -March 31, -| | 2019 | | 2018 | | | | -| --- | ---- | --- | ---- | --- | --- | --- | -(in $ millions, except per share amounts) -| Digital net revenue | | 3,710 | 3,450 | | | | -| ------------------------------------ | --- | ------ | ------ | --- | --- | --- | -| Packaged goods and other net revenue | | 1,240 | 1,700 | | | | -| Total net revenue | | 4,950 | 5,150 | | | | - -| Net income | | 1,019 | 1,043* | | | | -| ---------------------------- | ----------------------- | ------ | ------ | --- | --- | --- | -| Diluted earnings per share | | 3.33 | 3.34* | | | | -| Operating cash flow | | 1,547 | 1,692 | | | | -| Value of shares repurchased | 1,192 | | 601 | | | | -| Number of shares repurchased | | 11.0 | 5.3 | | | | -*During the twelve months ended March 31, 2018, EA recognized $235 million of incremental income tax -expense, or approximately $0.75 per share, due to the application of the Tax Cuts and Jobs Act. - -The following GAAP-based financial data and tax rate of 18% was used internally by company -management to adjust its fiscal 2019 GAAP results in order to assess EA’s operating results: - -Twelve Months Ended March 31, 2019 -GAAP-Based Financial Data -Change in -deferred net -revenue -| | Statement | Acquisition- | (online- | Mobile | | | -| --------------- | ---------- | ------------ | -------- | --------- | ------------ | --- | -| | of | related | enabled | platform | Stock-based | | -| (in $ millions) | Operations | expenses | games) | fees | compensation | | -Total net revenue 4,950 - 182 (188) - - Cost of revenue 1,322 (4) - (188) (4) -Gross profit 3,628 4 182 - 4 - Total operating expenses 2,632 (37) - - (280) -Operating income 996 41 182 - 2 84 -Interest and other income, net 83 - - - - -Income before provision for income taxes 1,079 41 182 - 2 84 -Number of shares used in computation: -| Diluted | 306 | | | | | | -| ------------ | --- | --- | --- | --- | --- | --- | -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. + Direct Connect +Service Overview + +2 Product Advantages + +2 Product Advantages + +Direct Connect has the following advantages: + +● Dedicated Channels to Safeguard Data Security + +– + +– + +Establish private connectivity between your on-premises network and the +cloud for stable data transmission without compromising privacy. + +Secure your data using encrypted, dedicated connections, which provide +higher security and more reliability. + +● Multiple Access Options + +– + +– + +You can select from optical bare fibers and other types of leased lines, +such as MSTP and MPLS VPN. + +There are a number of Direct Connect locations around the globe, so you +can access the cloud from anywhere nearby. + +● Great Scalability for Flexible Service Deployment + +– + +– + +By building a hybrid cloud, you can gain access to virtually unlimited +cloud resources for flexible, scalable deployment. + +You can run workloads that span on-premises infrastructure and different +cloud regions and flexibly connect networks across regions. + +● Ultra-low Latency and Stable Transmission + +– + +– + +Enjoy faster speeds with up to 100 Gbit/s of bandwidth, supporting +diverse service needs. + +Configure your connections in active/active or active/standby pairs for +higher reliability. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. 3 -Operating Metric -The following is a calculation of our total net bookings for the periods presented: -Three Months Ended TTM Ended -March 31, March 31, -2019 2018 2019 2018 -(in $ millions) -Total net revenue 1,238 1,582 4,950 5,150 -Change in deferred net revenue (online-enabled games) 173 (327) 182 30 -Mobile platform fees (47) - (188) - -Net bookings 1,364 1,255 4,944 5,180 -Business Outlook as of May 7, 2019 -The following forward-looking statements reflect expectations as of May 7, 2019. Electronic Arts -assumes no obligation to update these statements, except as required by law. Results may be -materially different and are affected by many factors detailed in this release and in EA’s annual and -quarterly SEC filings. -Fiscal Year 2020 Expectations – Ending March 31, 2020 -Financial metrics: -• Net revenue is expected to be approximately $5.375 billion. -Change in deferred net revenue (online-enabled games) is expected to be -o -approximately $(125) million. -Mobile platform fees are expected to be approximately $(150) million. -o -• Net income is expected to be approximately $2.595 billion. This includes a one-time tax -benefit of approximately $1.5 billion (see below). -• Diluted earnings per share is expected to be approximately $8.56. This includes a one- -time tax benefit of approximately $5 per share (see below). -• Operating cash flow is expected to be approximately $1.575 billion. -• The Company estimates a share count of 303 million for purposes of calculating fiscal -year 2020 diluted earnings per share. -Subsequent to the fiscal year ended March 31, 2019, EA completed an internal transfer of some of its -intellectual property rights to its Swiss subsidiary, where EA’s international business is -headquartered. This transfer will result in a one-time tax benefit of between $1.5 billion and $3.9 -billion, which will be recognized as a deferred tax asset in EA’s consolidated financial statements for -the three months ending June 30, 2019. The recognition of the deferred tax asset will add between $5 -to $13 per share to GAAP EPS and the low end of this range has been reflected above in EA’s -forecast. The forecast of this one-time tax benefit is preliminary and is dependent on the results of -Swiss referendums to be voted on later this month. -Operational metric: -• Net bookings is expected to be approximately $5.100 billion. -In addition, the following outlook for GAAP-based financial data and a long-term tax rate of 18% are -used internally by EA to adjust our GAAP expectations to assess EA’s operating results and plan for -future periods: + Direct Connect +Service Overview + +3 Application Scenarios + +3 Application Scenarios + +Access to Multiple VPCs from an On-premises Data Center + +After you connect your on-premises data center to the cloud using Direct Connect, +you can use Cloud Connect to connect the VPC that your on-premises data center +is accessing to those in other regions, so that your on-premises data center can +access all connected VPCs. + +Hybrid Cloud Deployment + +Direct Connect allows you to build a hybrid environment for your on-premises +data center and leverage the scalability of the cloud to expand the computing +capability of your applications. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 4 -Twelve Months Ending March 31, 2020 -GAAP-Based Financial Data -Change in -deferred net -revenue -Acquisition- (online- -GAAP related enabled Mobile Stock-based -Guidance expenses games) platform fees compensation -(in $ millions) -Digital net revenue 4 ,175 - ( 75) ( 150) - -Packaged goods & other net revenue 1 ,200 - ( 50) - - -Total net revenue 5 ,375 - ( 125) ( 150) - -Cost of revenue 1 ,327 ( 12) - ( 150) ( 2) -Operating expense 2 ,746 ( 23) - - ( 353) -Income before provision for income taxes 1 ,361 3 5 ( 125) - 3 55 -Net income 2 ,595 -Number of shares used in computation: -Diluted shares 3 03 -First Quarter Fiscal Year 2020 Expectations – Ending June 30, 2019 -Financial metrics: -• Net revenue is expected to be approximately $1.130 billion. -Change in deferred net revenue (online-enabled games) is expected to be -o -approximately $(405) million. -Mobile platform fees are expected to be approximately $(35) million. -o -• Net income is expected to be approximately $1.765 billion. This includes a one-time tax -benefit of approximately $1.5 billion (see above). -• Diluted earnings per share is expected to be approximately $5.84. This includes a one- -time tax benefit of approximately $5 per share (see above). -• The Company estimates a share count of 302 million for purposes of calculating first -quarter fiscal year 2020 diluted earnings per share. Should the first quarter be a loss, the -Company will use the basic share count of 299 million for purposes of calculating first -quarter fiscal year 2020 earnings per share. -Operational metric: -• Net bookings is expected to be approximately $690 million. -In addition, the following outlook for GAAP-based financial data and a long-term tax rate of 18% are -used internally by EA to adjust our GAAP expectations to assess EA’s operating results and plan for -future periods: + Direct Connect +Service Overview + +Figure 3-1 Hybrid cloud + +3 Application Scenarios + +Table 3-1 Comparisons of Direct Connect and VPN in hybrid cloud deployment + +Scenario + +Description + +Helpful Links + +Cloud +Service + +Virtual +Private +Networ +k (VPN) + +Connect an on- +premises data +center to the +cloud through +an IPsec tunnel. + +Direct +Connect + +Connect an on- +premises data +center to the +cloud using a +dedicated +network +connection. + +VPN uses an encrypted +communications tunnel to +connect a VPC on the cloud +to an on-premises data +center and sends traffic over +the Internet. It is +inexpensive, easy to +configure, and easy to use. +However, VPN connections +may be affected by the +Internet quality. + +Direct Connect provides +physical connections +between VPCs and data +centers. It has the +advantages of low latency +and is very secure. Direct +Connect is a good choice +when there are strict +requirements on network +transmission quality. + +Connecting to a +VPC Through a +VPN + +What Is an +Enterprise Switch? + +Connecting to +Multiple VPCs that +Do Not Need to +Communicate with +Each Other + +What Is an +Enterprise Switch? + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 5 -Three Months Ending June 30, 2019 -GAAP-Based Financial Data -Change in -deferred net -revenue -Acquisition- (online- Mobile -GAAP related enabled platform Stock-based -Guidance expenses games) fees compensation -(in $ millions) -Total net revenue 1 ,130 - ( 405) ( 35) - -Cost of revenue 1 88 ( 2) - ( 35) - -Operating expense 6 50 ( 5) - - ( 70) -Income before provision for income taxes 3 10 7 ( 405) - 7 0 -Net income 1 ,765 -Number of shares used in computation: -Diluted shares 3 02 -For more information about the nature of the GAAP-based financial data, please refer to EA’s Form -10-Q for the fiscal quarter ended December 31, 2018. -Conference Call and Supporting Documents -Electronic Arts will host a conference call on Tuesday, May 7, 2019 at 2:00 pm PT (5:00 pm ET) to -review its results for the fiscal fourth quarter and fiscal year ended March 31, 2019 and its outlook for -the future. During the course of the call, Electronic Arts may disclose material developments affecting -its business and/or financial performance. Listeners may access the conference call live through the -following dial-in number (866) 393-4306 (domestic) or (734) 385-2616 (international), using the -Conference ID 2997759 or via webcast at EA’s IR Website at http://ir.ea.com. -EA has posted a slide presentation, a financial model of EA’s historical results and guidance, and an -Investor Accounting FAQ on EA’s IR Website. After the conference call, EA will also post the -prepared remarks and a transcript from the conference call on EA’s IR Website. -A dial-in replay of the conference call will be available until May 21, 2019 at (855) 859-2056 -(domestic) or (404) 537-3406 (international) using pin code 2997759. An audio webcast replay of the -conference call will be available for one year on EA’s IR Website. -Forward-Looking Statements -Some statements set forth in this release, including the information relating to EA’s fiscal 2020 -expectations under the heading “Business Outlook as of May 7, 2019,” and other information -regarding EA's fiscal 2020 expectations contain forward-looking statements that are subject to -change. Statements including words such as “anticipate,” “believe,” “expect,” “intend,” “estimate”, -“plan”, “predict”, “seek”, “goal”, “will”, “may”, “likely”, “should”, “could” (and the negative of any of -these terms), “future” and similar expressions also identify forward-looking statements. These -forward-looking statements are not guarantees of future performance and reflect management’s -current expectations. Our actual results could differ materially from those discussed in the forward- -looking statements. + Direct Connect +Service Overview + +4 Functions + +4 Functions + +Standard Connection + +A standard connection provides an exclusive port. You can create standard +connections on the management console. To improve reliability, you can create +multiple connections terminated at different locations and allow them to serve as +backups for each other. If you can select only one carrier due to special +requirements, you must configure different physical routes for your connections. + +You can choose self-service connections or full-service connections. + +● + +● + +Self-service connection: Huawei Cloud only provides the port. You need to +create a connection on the console, and lease a line from a carrier. +Full-service connection: You only need to create a connection on the console, +and Huawei Cloud will complete all operations required for network +connectivity. + +Figure 4-1 shows an example of how standard connections work. + +Figure 4-1 Accessing the cloud using standard connections + +For details, see Creating a Connection. + +Hosted Connection + +If you use a hosted connection to access the cloud, the port you use is shared with +other users. You request a carrier (a partner of Huawei Cloud) to provide a + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 6 -Some of the factors which could cause the Company’s results to differ materially from its expectations -include the following: sales of the Company’s products and services; the Company’s ability to develop -and support digital products and services, including managing online security and privacy; outages of -our products, services or technological infrastructure; the Company’s ability to manage expenses; the -competition in the interactive entertainment industry; governmental regulations; the effectiveness of -the Company’s sales and marketing programs; timely development and release of the Company’s -products and services; the Company’s ability to realize the anticipated benefits of acquisitions; the -consumer demand for, and the availability of an adequate supply of console hardware units; the -Company’s ability to predict consumer preferences among competing platforms; the Company’s -ability to develop and implement new technology; foreign currency exchange rate fluctuations; -general economic conditions; changes in our tax rates or tax laws; and other factors described in Part -II, Item 1A of Electronic Arts’ latest Quarterly Report on Form 10-Q under the heading “Risk Factors”, -as well as in other documents we have filed with the Securities and Exchange Commission, including -our Annual Report on Form 10-K for the fiscal year ended March 31, 2018. -These forward-looking statements are valid as of May 7, 2019. Electronic Arts assumes no obligation -to revise or update any forward-looking statement for any reason, except as required by law. In -addition, the preliminary financial results set forth in this release are estimates based on information -currently available to Electronic Arts. -While Electronic Arts believes these estimates are meaningful, they could differ from the actual -amounts that Electronic Arts ultimately reports in its Annual Report on Form 10-K for the fiscal year -ended March 31, 2019. Electronic Arts assumes no obligation and does not intend to update these -estimates prior to filing its Form 10-K for the fiscal year ended March 31, 2019. -About Electronic Arts -Electronic Arts (NASDAQ: EA) is a global leader in digital interactive entertainment. The Company -develops and delivers games, content and online services for Internet-connected consoles, mobile -devices and personal computers. -In fiscal year 2019, EA posted GAAP net revenue of $4.95 billion. Headquartered in Redwood City, -California, EA is recognized for a portfolio of critically acclaimed, high-quality brands such as The -Sims™, Madden NFL, EA SPORTS™ FIFA, Battlefield™, Need for Speed™, Dragon Age™ and -Plants vs. Zombies™. More information about EA is available at www.ea.com/news. -Apex Legends, Anthem, Battlefield, The Sims, EA SPORTS, Need for Speed, Dragon Age, and -Plants vs. Zombies are trademarks of Electronic Arts Inc. John Madden, NFL and FIFA are the -property of their respective owners and used with permission. -For additional information, please contact: -Chris Evenden John Reseburg -Vice President, Investor Relations Vice President, Corporate Communications -650-628-0255 650-628-3601 -cevenden@ea.com jreseburg@ea.com + Direct Connect +Service Overview + +4 Functions + +connection to connect to your on-premises data center to Huawei Cloud. The +partner has pre-connected dedicated lines terminated at the Direct Connect +locations of Huawei Cloud and allocates a connection port to you after receiving +your request. + +Figure 4-2 shows an example of how hosted connections work. + +Figure 4-2 Accessing the cloud using hosted connections + +The following table compares hosted connections with standard connections. + +Item + +Port + +Recomm +ended +bandwidt +h + +Estimate +d +construct +ion +period + +Parties +involved + +Standard Connection + +Hosted Connection + +Exclusive port + +Shared port + +1 Gbit/s to 100 Gbit/s + +Less than 1 Gbit/s + +Two to three months for local lines, +and three to four months for long- +distance lines + +About one month + +Connection user, leased line carrier, +equipment room provider, and Huawei +Cloud + +Connection user, leased +line carrier, and Huawei +Cloud + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + 7 -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Statements of Operations -(in $ millions, except per share data) -| | Three Months Ended | | Twelve Months Ended | | -| --- | ------------------ | ---- | ------------------- | --- | -| | March 31, | | March 31, | | -| | 2019 1 | | 2019 1 | | -| | | 2018 | 2018 | | -Net revenue -| Product | 216 | 757 | 1,593 | 2,586 | -| ----------------- | ----- | ----- | ----- | ----- | -| Service and other | 1,022 | 825 | 3,357 | 2,564 | -| Total net revenue | 1,238 | 1,582 | 4,950 | 5,150 | -Cost of revenue -| Product | 52 | 106 | 517 | 822 | -| --------------------- | --- | ----- | ----- | ----- | -| Service and other | 224 | 127 | 805 | 455 | -| Total cost of revenue | 276 | 233 | 1,322 | 1,277 | -| Gross profit | 962 | 1,349 | 3,628 | 3,873 | -Operating expenses: -| Research and development | 398 | 335 | 1,433 | 1,320 | -| -------------------------------------------- | --- | --- | ----- | ----- | -| Marketing and sales | 229 | 130 | 702 | 641 | -| General and administrative | 123 | 126 | 460 | 469 | -| Acquisition-related contingent consideration | 11 | — | 14 | — | -| Amortization of intangibles | 5 | 5 | 23 | 9 | -| Total operating expenses | 766 | 596 | 2,632 | 2,439 | -| Operating income | 196 | 753 | 996 | 1,434 | -| Interest and other income (expense), net | 23 | 1 | 83 | 15 | -| Income before provision for income taxes | 219 | 754 | 1,079 | 1,449 | -| Provision for income taxes | 10 | 147 | 60 | 406 | -| Net income | 209 | 607 | 1,019 | 1,043 | -Earnings per share -| Basic | 0.70 | 1.98 | 3.36 | 3.39 | -| ------- | ---- | ---- | ---- | ---- | -| Diluted | 0.69 | 1.95 | 3.33 | 3.34 | -Number of shares used in computation -| Basic | 299 | 307 | 303 | 308 | -| ------- | --- | --- | --- | --- | -| Diluted | 301 | 311 | 306 | 312 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. - -Results (in $ millions, except per share data) -The following table reports the variance of the actuals versus our guidance provided on February 5, 2019 for the three months ended March -31, 2019 plus a comparison to the actuals for the three months ended March 31, 2018. -Three Months Ended March 31, -| | 2019 1 | | 2019 1 | | -| --- | ------ | --- | ------ | --- | -2018 -| | Guidance | Variance | Actuals | Actuals | -| --- | -------- | -------- | ------- | ------- | -Net revenue -| Net revenue | 1,163 | 75 | 1,238 | 1,582 | -| ----------- | ----- | --- | ----- | ----- | -GAAP-based financial data -Change in deferred net revenue (online-enabled games) 56 117 173 (327) -| Mobile platform fees | (49) | 2 | (47) | — | -| -------------------- | ---- | --- | ---- | --- | -Cost of revenue -| Cost of revenue | 265 | 11 | 276 | 233 | -| --------------- | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | — | (1) | (1) | (1) | -| ---------------------------- | ---- | --- | ---- | --- | -| Stock-based compensation | — | (1) | (1) | (1) | -| Mobile platform fees | (49) | 2 | (47) | — | -Operating expenses -| Operating expenses | 715 | 51 | 766 | 596 | -| ------------------ | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | (6) | (10) | (16) | (5) | -| ---------------------------- | ---- | ---- | ---- | ---- | -| Stock-based compensation | (76) | 4 | (72) | (68) | -Income before tax -| Income before tax | 195 | 24 | 219 | 754 | -| ----------------- | --- | --- | --- | --- | -GAAP-based financial data -| Acquisition-related expenses | 6 | 11 | 17 | 6 | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games) 56 117 173 (327) -| Mobile platform fees | — | — | — | — | -| -------------------------------------- | --- | --- | --- | --- | -| Stock-based compensation | 76 | (3) | 73 | 69 | -| Tax rate used for management reporting | 18% | | 18% | 21% | -Earnings per share -| Basic | 0.56 | 0.14 | 0.70 | 1.98 | -| ------- | ---- | ---- | ---- | ---- | -| Diluted | 0.56 | 0.13 | 0.69 | 1.95 | -Number of shares -| Basic | 301 | (2) | 299 | 307 | -| ------- | --- | --- | --- | --- | -| Diluted | 303 | (2) | 301 | 311 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Balance Sheets -(in $ millions) -March 31, 20191 March 31, 20182 -ASSETS -Current assets: -| Cash and cash equivalents | 4,708 | 4,258 | -| ------------------------- | ----- | ----- | -| Short-term investments | 737 | 1,073 | -Receivables, net of allowances of $7 and $165, respectively 623 385 -| Other current assets | 313 | 288 | -| ------------------------------------ | ----- | ----- | -| Total current assets | 6,381 | 6,004 | -| Property and equipment, net | 448 | 453 | -| Goodwill | 1,892 | 1,883 | -| Acquisition-related intangibles, net | 87 | 71 | -| Deferred income taxes, net | 35 | 84 | -| Other assets | 114 | 89 | -| TOTAL ASSETS | 8,957 | 8,584 | -LIABILITIES AND STOCKHOLDERS’ EQUITY -Current liabilities: -| Accounts payable | 113 | 48 | -| ------------------------------------------- | ----- | ----- | -| Accrued and other current liabilities | 1,052 | 821 | -| Deferred net revenue (online-enabled games) | 1,100 | 1,622 | -| Total current liabilities | 2,265 | 2,491 | -| Senior notes, net | 994 | 992 | -| Income tax obligations | 233 | 250 | -| Deferred income taxes, net | 2 | 1 | -| Other liabilities | 132 | 255 | -| Total liabilities | 3,626 | 3,989 | -Stockholders’ equity: -| Common stock | 3 | 3 | -| ------------------------------------------ | ----- | ----- | -| Additional paid-in capital | — | 657 | -| Retained earnings | 5,358 | 4,062 | -| Accumulated other comprehensive loss | (30) | (127) | -| Total stockholders’ equity | 5,331 | 4,595 | -| TOTAL LIABILITIES AND STOCKHOLDERS’ EQUITY | 8,957 | 8,584 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606) Revenue from Contracts with Customers. EA’s FY19 -Unaudited Condensed Consolidated Balance Sheet reflects the effect of the adoption as of April 1, 2018, which had an impact on the following: receivables, net -of allowances, accrued and other current liabilities, deferred net revenue (online-enabled games), deferred income taxes, net, retained earnings and accumulated -other comprehensive loss. Financial data for periods prior to April 1, 2018 has not been restated. - 2Derived from audited consolidated financial statements. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Condensed Consolidated Statements of Cash Flows -(in $ millions) -| | | Three Months Ended | | Twelve Months | | -| --- | --- | ------------------ | ---- | --------------- | ---- | -| | | March 31, | | Ended March 31, | | -| | | 2019 | 2018 | 2019 | 2018 | -OPERATING ACTIVITIES -| Net income | | 209 | 607 | 1,019 | 1,043 | -| ---------- | --- | --- | --- | ----- | ----- | -Adjustments to reconcile net income to net cash provided by -operating activities: -| | Depreciation, amortization and accretion | 37 | 39 | 145 | 136 | -| --- | -------------------------------------------- | --- | --- | --- | --- | -| | Acquisition-related contingent consideration | 14 | — | 14 | — | -| | Stock-based compensation | 73 | 69 | 284 | 242 | -Change in assets and liabilities -| | Receivables, net | 183 | 502 | (88) | (25) | -| --- | ------------------------------------------- | ----- | ----- | ---- | ---- | -| | Other assets | (65) | (69) | (24) | 10 | -| | Accounts payable | 24 | (60) | 59 | (44) | -| | Accrued and other liabilities | (120) | (222) | 3 | 43 | -| | Deferred income taxes, net | 73 | 74 | (16) | 204 | -| | Deferred net revenue (online-enabled games) | 171 | (325) | 151 | 83 | -Net cash provided by operating activities 599 615 1,547 1,692 -INVESTING ACTIVITIES -| Capital expenditures | | (35) | (20) | (119) | (107) | -| -------------------- | --- | ---- | ---- | ----- | ----- | -Proceeds from maturities and sales of short-term investments 733 1,510 1,688 3,166 -Purchase of short-term investments (198) (275) (1,342) (2,287) -| Acquisitions, net of cash acquired | | — | — | (58) | (150) | -| ---------------------------------- | ----------------------------------------- | --- | ----- | ---- | ----- | -| | Net cash provided by investing activities | 500 | 1,215 | 169 | 622 | -FINANCING ACTIVITIES -| Proceeds from issuance of common stock | | 25 | 21 | 61 | 78 | -| -------------------------------------- | --- | --- | --- | --- | --- | -Cash paid to taxing authorities for shares withheld from employees (6) (8) (122) (120) -Repurchase and retirement of common stock (301) (148) (1,192) (601) -Net cash used in financing activities (282) (135) (1,253) (643) -Effect of foreign exchange on cash and cash equivalents 4 (3) (13) 22 -| Increase in cash and cash equivalents | | 821 | 1,692 | 450 | 1,693 | -| ------------------------------------- | --- | ----- | ----- | ----- | ----- | -| Beginning cash and cash equivalents | | 3,887 | 2,566 | 4,258 | 2,565 | -| Ending cash and cash equivalents | | 4,708 | 4,258 | 4,708 | 4,258 | - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions, except per share data) -| | Q4 Q11 | Q21 Q31 | Q41 | YOY % | -| --- | --------- | --------- | ---- | ------ | -| | FY18 FY19 | FY19 FY19 | FY19 | Change | -Net revenue -| Net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 | (22%) | -| ------------ | ----------- | ----------- | ----- | ----- | -GAAP-based financial data -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | | -| -------------------- | ------ | --------- | ---- | --- | -Gross profit -| Gross profit | 1,349 922 | 868 876 | 962 | (29%) | -| ------------ | --------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 1 1 | 1 1 | 1 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Mobile platform fees | — — | — — | — | | -| ------------------------------------ | ------- | ------- | --- | --- | -| Stock-based compensation | 1 1 | 1 1 | 1 | | -| Gross profit (as a % of net revenue) | 85% 81% | 67% 68% | 78% | | -Operating income -| Operating income | 753 300 | 258 242 | 196 | (74%) | -| ---------------- | ------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 6 7 | 9 8 | 17 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Stock-based compensation | 69 70 | 66 75 | 73 | | -| ---------------------------------------- | ------- | ------- | --- | --- | -| Operating income (as a % of net revenue) | 48% 26% | 20% 19% | 16% | | -Net income -| Net income | 607 293 | 255 262 | 209 | (66%) | -| ---------- | ------- | ------- | --- | ----- | -GAAP-based financial data -| Acquisition-related expenses | 6 7 | 9 8 | 17 | | -| ---------------------------- | --- | --- | --- | --- | -Change in deferred net revenue (online-enabled games)3 (327) (339) (20) 368 173 -| Stock-based compensation | 69 70 | 66 75 | 73 | | -| -------------------------------------- | --------- | --------- | ---- | ----- | -| Tax rate used for management reporting | 21% 18% | 18% 18% | 18% | | -| Net income (as a % of net revenue) | 38% 26% | 20% 20% | 17% | | -| Diluted earnings per share | 1.95 0.95 | 0.83 0.86 | 0.69 | (65%) | -Number of diluted shares used in computation -| Basic | 307 306 | 305 302 | 299 | | -| ------- | ------- | ------- | --- | --- | -| Diluted | 311 310 | 307 304 | 301 | | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 significantly -changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including information with -respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, 2018 has not been -restated. -3The change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of cash flows does not necessarily equal the -change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of operations primarily due to the impact of -unrecognized gains/losses on cash flow hedges. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions) -YOY -| | Q4 Q11 | Q21 Q31 | Q41 % | -| --- | --------- | --------- | ----------- | -| | FY18 FY19 | FY19 FY19 | FY19 Change | -QUARTERLY NET REVENUE PRESENTATIONS -Net revenue by composition -| Full game downloads | 232 116 | 148 247 | 169 (27%) | -| ------------------------ | ----------- | ----------- | ----------- | -| Live services | 698 610 | 412 480 | 714 2% | -| Mobile | 172 231 | 220 181 | 182 6% | -| Total digital | 1,102 957 | 780 908 | 1,065 (3%) | -| Packaged goods and other | 480 180 | 506 381 | 173 (64%) | -| Total net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 (22%) | -| Total digital | 70% 84% | 61% 70% | 86% | -| Packaged goods and other | 30% 16% | 39% 30% | 14% | -| Total net revenue % | 100% 100% | 100% 100% | 100% | -GAAP-based financial data -| Full game downloads | (39) (20) | 9 26 | 44 | -| ------------------------ | ----------- | -------- | --- | -| Live services | (19) (160) | (84) 304 | 131 | -| Mobile | 4 (35) | (24) 9 | — | -| Total digital | (54) (215) | (99) 339 | 175 | -| Packaged goods and other | (273) (124) | 79 29 | (2) | -Total change in deferred net revenue (online-enabled games) by composition3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | -| -------------------- | ------ | --------- | ---- | -Net revenue by platform -| Console | 1,196 705 | 917 885 | 826 (31%) | -| ----------------- | ----------- | ----------- | ----------- | -| PC/Browser | 210 197 | 149 217 | 217 3% | -| Mobile | 173 233 | 220 181 | 190 10% | -| Other | 3 2 | — 6 | 5 67% | -| Total net revenue | 1,582 1,137 | 1,286 1,289 | 1,238 (22%) | -GAAP-based financial data -| Console | (313) (288) | 8 323 | 87 | -| ---------- | ----------- | ------- | --- | -| PC/Browser | (20) (9) | (4) 33 | 88 | -| Mobile | 5 (42) | (24) 11 | (2) | -| Other | 1 — | — 1 | — | -Total change in deferred net revenue (online-enabled games) by platform3 (327) (339) (20) 368 173 -| Mobile platform fees | — (49) | (44) (48) | (47) | -| -------------------- | ------ | --------- | ---- | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606), Revenue from Contracts with Customers. Topic 606 -significantly changes how EA recognizes and reports revenue and diluted earnings per share. For more information about the adoption of Topic 606, including -information with respect to mobile platform fees, please refer to the Investor Accounting FAQ on our IR website. Financial data for periods prior to April 1, -2018 has not been restated. -3The change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of cash flows does not necessarily equal the -change in deferred net revenue (online-enabled games) in the unaudited condensed consolidated statements of operations primarily due to the impact of -unrecognized gains/losses on cash flow hedges. - -ELECTRONIC ARTS INC. AND SUBSIDIARIES -Unaudited Supplemental Financial Information and Business Metrics -(in $ millions) -| | Q4 Q11 | Q21 Q31 | Q41 YOY % | -| --- | --------- | --------- | ----------- | -| | FY18 FY19 | FY19 FY19 | FY19 Change | -CASH FLOW DATA -| Operating cash flow | 615 120 | (126) 954 | 599 (3%) | -| -------------------------- | ----------- | ----------- | ---------- | -| Operating cash flow - TTM | 1,692 1,636 | 1,458 1,563 | 1,547 (9%) | -| Capital expenditures | 20 32 | 31 21 | 35 75% | -| Capital expenditures - TTM | 107 106 | 107 104 | 119 11% | -Repurchase and retirement of common stock 148 300 299 292 301 103% -DEPRECIATION -| Depreciation expense | 31 30 | 30 30 | 31 — | -| -------------------- | ----- | ----- | ---- | -BALANCE SHEET DATA -| Cash and cash equivalents | 4,258 3,876 | 2,881 3,887 | 4,708 | -| ------------------------- | ----------- | ----------- | ----- | -| Short-term investments | 1,073 1,095 | 1,664 1,274 | 737 | -Cash and cash equivalents, and short-term investments 5,331 4,971 4,545 5,161 5,445 2% -| Receivables, net | 385 371 | 966 806 | 623 62% | -| ---------------- | ------- | ------- | ------- | -STOCK-BASED COMPENSATION -| Cost of revenue | 1 | 1 1 | 1 1 | -| ------------------------------ | ----- | ----- | --- | -| Research and development | 44 47 | 39 49 | 49 | -| Marketing and sales | 8 | 7 9 | 8 9 | -| General and administrative | 16 15 | 17 17 | 14 | -| Total stock-based compensation | 69 70 | 66 75 | 73 | -1At the beginning of fiscal year 2019, April 1, 2018, EA adopted FASB ASU 2014-09 (Topic 606) Revenue from Contracts with Customers. EA’s FY19 -Unaudited Condensed Consolidated Balance Sheets reflect the effect of the adoption as of April 1, 2018, which had an impact on receivables, net of -allowances. Financial data for periods prior to April 1, 2018 has not been restated. \ No newline at end of file + Direct Connect +Service Overview + +4 Functions + +Item + +Standard Connection + +Hosted Connection + +Procedur +e + +1. You create a connection on the + +console to order a port. + +2. You contact the leased line carrier +and supervise the line deployment +from your on-premises data center +to the equipment room at the +location you choose. + +3. You contact the carrier operating + +the equipment room at the location +you choose to complete the cabling +(inlet wire and jumper) inside the +equipment room. + +4. Your carrier works with Huawei + +Cloud to commission access devices. + +5. You complete required network +configuration on the console, +including creating a virtual gateway +and virtual interface. + +Billing + +● Pay for the port (by month or year) + +to Huawei Cloud. + +● Pay to the equipment room provider +at the location for the cabling inside +the equipment room. + +● Pay to the carrier of your on- + +premises data center for the cabling +inside the equipment room. + +● Pay to the leased line carrier for + +other work and the bandwidth. For +details, see Billing. + +1. The partner deploys the +leased line from your +on-premises data +center to the Direct +Connect location you +select. + +2. The carrier completes +the commissioning of +access devices. + +3. You complete required +network configuration +on the console, +including creating a +virtual gateway and +virtual interface. + +● You do not need to pay +for the port to Huawei +Cloud. + +● Pay to the carrier of + +your on-premises data +center for the cabling +inside the equipment +room. + +● Pay to the leased line +carrier for other work +and the bandwidth. + +For details, see Partner Connections. + +Direct Connect Gateway + +● + +Virtual gateway: A virtual gateway is a logical gateway that enables an on- +premises data center to access a VPC over a connection. To enable an on- +premises data center to access a VPC over a connection, you can associate the +VPC with a virtual gateway. To access other VPCs, you can use VPC Peering or +Cloud Connect to connect the VPC your on-premises data center is accessing +to these VPCs. + +A virtual gateway can only have one VPC associated. An on-premises data +center can access the same VPC over two connections through one virtual +gateway. + +For details, see Virtual Gateways. + +● Global DC gateway: A global DC gateway enables your on-premises data + +center to access VPCs in multiple regions so you can use a single connection + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +8 + + Direct Connect +Service Overview + +4 Functions + +to provide high-speed access to cloud compute and storage resources in any +region. + +A global DC gateway can only be associated with connections terminated at +the same Direct Connect location. If there are multiple connections +terminated at different Direct Connect locations, you need to create multiple +global DC gateways. + +For details, see Global DC Gateways. + +Virtual Interface + +A virtual interface is a point of entry for an on-premises data center to access a +VPC over a connection. A virtual interface associates a connection with a virtual +gateway and connects the virtual gateway to a remote gateway, enabling +communications between the on-premises data center and the VPC. + +Virtual interfaces support static routing and BGP routing. During the +establishment of network connectivity, you can use BGP to connect on-premises +data centers to virtual gateways. BGP helps you build a hybrid cloud more +efficiently, flexibly, and reliably. + +For details, see Virtual Interfaces. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +9 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +5 Direct Connect Locations + +A Direct Connect location provides access to Huawei Cloud in a region. Before +using Direct Connect to access Huawei Cloud, you need to obtain the details about +each location. + +Direct Connect provides a number of locations for you to choose from. You can +request a port by creating a connection. + +For more information, submit a service ticket or contact your Direct Connect +manager. + +Table 5-1 Direct Connect locations in China + +Region + +Location + +CN North-Beijing4 Beijing-Yatai + +Beijing- +Jiuxianqiao + +Carrier + +Neutral + +Dr. Peng + +Beijing-Huitian + +Neutral + +Beijing-Zhongjin + +Neutral + +Address + +Beijing Yatai + +Beijing Universal +Creative Park +(UCP) + +Beijing Huitian +Cloud Industrial +Park + +Centrin Data in +Beijing Economic- +Technological +Development +Area + +Langfang- +Wanguo + +Neutral + +Langfang GDS + +Langfang-Range + +China Telecom + +Langfang-Huawei Huawei + +Langfang Range +Information Port + +Huawei +production base +in Langfang + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +10 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Carrier + +Address + +Wuhan-Wugang + +Neutral + +Wuhan-Huawei + +Huawei + +CN North- +Ulanqab1 + +Ulanqab-Jining + +Huawei + +Ulanqab- +Meiguiying + +Huawei + +CN North3 + +Helin'geer-Nanqu Huawei + +Helin'geer-Beiqu + +Huawei + +Helin'geer-Xiqu + +Huawei + +CN East- +Shanghai1 + +Suzhou-Wanguo + +Neutral + +Wuhan Iron & +Steel Data +Industrial Park + +Wuhan Research +Center + +Ulanqab Huawei +Data Center + +Ulanqab +Meiguiying Base + +Huawei Cloud +Data Center +Nanqu, Helin'geer + +Huawei Cloud +Data Center +Beiqu, Helin'geer + +Huawei Cloud +Data Center Xiqu, +Helin'geer + +Suzhou GDS Data +Center + +Suzhou-Guoke + +China Unicom + +SISDC + +Suzhou-Huawei + +Huawei + +Suzhou-Kunhui + +Neutral + +Suzhou-Wujiang + +China Mobile + +Shanghai- +Wanguo + +Neutral + +Shanghai-Baoxin + +China Mobile + +Shanghai- +Guanghuan + +Hangzhou- +Hanggang + +Hangzhou- +Huawei + +Neutral + +Neutral + +Huawei + +Suzhou Research +Center + +Suzhou Kunhui +Network + +Suzhou Mobile +Data Center + +Shanghai GDS +Data Center + +Shanghai Baoxin +Data Center + +Beijing Sinnet +Data Center + +Hangzhou Iron & +Steel Cloud Data +Center + +Hangzhou +Research Center + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +11 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +CN East- +Shanghai2 + +Location + +Shanghai- +Wanguo + +Shanghai- +Guanghuan + +Carrier + +Neutral + +Neutral + +Shanghai-Baoxin + +China Mobile + +CN East2 + +Wuhu-Jiangbei + +Huawei + +Wuhu-Sanshan + +Huawei + +Wuhu-Zhijiao + +Huawei + +Address + +Shanghai GDS +Data Center + +Beijing Sinnet +Data Center + +Shanghai Baoxin +Data Center + +Wuhu Jiangbei +Data Center + +Wuhu Sanshan +Data Center + +Wuhu Zhijiao +Data Center + +CN East-Qingdao + +Qingdao-Jiaozhou Huawei + +Jiaozhou Ka'aosi + +CN South- +Guangzhou + +Qingdao- +Hongdao + +Guangzhou- +Yunpu + +Guangzhou- +Huaxinyuan + +Guangzhou- +Daxuecheng + +Guangzhou- +Mingmei + +Guangzhou- +Hualong + +China Mobile + +China Telecom + +Neutral + +Neutral + +China Unicom + +China Unicom + +Shenzhen-Baode + +Neutral + +Shenzhen-Futian + +Neutral + +Hongdao Mobile +Data Center + +Guangzhou Yunpu +Data Center + +South China New +Materials +Innovation Park + +Guangzhou +Higher Education +Mega Center + +Guangzhou +Mingmei +Electronics + +Weixin Innovation +Industrial Park + +Shenzhen +Guanlan High- +Tech Park + +Futian Zhongpeng +Cloud + +Shenzhen- +Nanshan + +Neutral + +Nanshan SDGI + +Shenzhen-Yifeng + +Neutral + +Dianda Guyuan +Industrial Park + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +12 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Dongguan- +Tuanbowa + +Carrier + +Huawei + +CN Southwest- +Guiyang1 + +Chengdu-Huawei + +Huawei + +Guiyang-Machang Huawei + +Guiyang-Gui'an + +China Mobile + +Guiyang-Qixinghu Huawei + +Huawei + +China Telecom + +China Mobile + +CN-Hong Kong + +Guiyang- +Gaoduanyuan + +Hong Kong-Sha +Tin + +HK-Sai Kung- +Mobile + +Hong Kong-Sai +Kung-Global +Switch + +HK-Tsuen Wan- +MEGA + +China Telecom + +Global Switch + +Neutral + +MEGA + +Address + +Huawei Base +(Dongguan +Songshan Lake) + +Huawei Chengdu +Research Center + +Gui'an Machang +Data Center + +Gui'an New Area +Electron +Information +Industrial Park + +Gui'an Huawei +Cloud Data +Center + +Guiyang-High-End +Park + +Sun Hung Kai +Logistics Centre + +China Mobile +Global Network +Center + +Table 5-2 Direct Connect locations outside China + +Region + +Location + +AP-Bangkok + +Bangkok-TRUE + +Bangkok-STT + +Bangkok-NTT + +Carrier + +TRUE + +STT + +NTT + +AP-Singapore + +Singapore-Equinix + +Equinix + +Singapore-Global Switch Global Switch + +Singapore-1Net + +Singapore-Singtel + +1-Net + +Singtel + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +13 + + Direct Connect +Service Overview + +5 Direct Connect Locations + +Region + +Location + +Carrier + +AP-Jakarta + +AP-Manila + +Singapore-Telin3 + +Singapore-STT + +Jakarta-DCI + +Jakarta-EDGE + +Manila-GBP + +Telin + +STT + +DCI + +EDGE + +Globe + +AF-Johannesburg + +Nigeria-Lagos-Medallion Medallion + +Manila-Digital Halo + +Digital Halo + +Johannesburg-IS +Parklands + +Internet Solutions +Parklands + +Johannesburg-Teraco + +Teraco + +Johannesburg-DPA + +Vodacom + +Johannesburg-ADC + +Johannesburg-MTN + +AF-Cairo + +Cairo-Xceed + +ADC + +MTN + +Xceed + +LA-Mexico City1 + +Mexico-Ixtlahuaca + +Ixtlahuaca + +LA-Mexico City2 + +Mexico-Tultitlan + +Mexico-Tultitlan + +Mexico-Scala + +LA-Sao Paulo1 + +Sao Paulo-Equinix + +LA-Lima1 + +LA-Santiago + +Sao Paulo-ODATA + +Sao Paulo-Scala + +Sao Paulo-VIVO + +Lima-TIC + +Santiago-Paine + +Santiago-Claro + +TR-Istanbul + +TR-Istanbul + +Istanbul-NGN + +KIO MEX 5 + +KIO MEX 5 + +Scala + +Equinix + +OData + +Scala + +Telefonica + +Telefonica + +Paine + +Claro + +Turkcell + +NGN + +ME-Riyadh + +Riyadh-STC Khurais + +STC Khurais + +Riyadh-Remal + +Remal + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +14 + + Direct Connect +Service Overview + +6 Security + +6 Security + +6.1 Identity Authentication and Access Control + +Identity Authentication + +You can use Identity and Access Management (IAM) to control access to your +Direct Connect resources. IAM permissions define which actions on your cloud +resources are allowed or denied. After creating an IAM user, the administrator +needs to add it to a user group and grant the permissions required by Direct +Connect to the user group. Then, all users in this group automatically inherit the +granted permissions. + +For details, see Permissions. + +Access Control + +A security group is a collection of access control rules for cloud resources, such as +cloud servers, containers, and databases, that have the same security protection +requirements and that are mutually trusted within a VPC. After a security group is +created, you can create various access rules for the security group, and these rules +will apply to all cloud resources added to this security group. + +Your account automatically comes with a default security group. The default +security group allows all outbound traffic, denies all inbound traffic, and allows all +traffic between cloud resources in the group. Your cloud resources in this security +group can communicate with each other already without adding additional rules. +For details about the default security group rules, see Default Security Groups +and Security Group Rules. + +In addition, Huawei Cloud allows you to manage security groups and security +group rules, including + +● + +● + +Creating, viewing, deleting, modifying, cloning, and adding security groups + +Adding, copying, modifying, deleting, importing, and exporting security group +rules + +● Quickly adding multiple security group rules + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +15 + + Direct Connect +Service Overview + +6 Security + +● + +● + +Viewing and changing security groups of ECSs + +Adding cloud resources to or removing cloud resources from security groups + +You can define access control rules for a security group. Then ECSs that will be +added to this security group will be protected. For details, see Security Group. + +6.2 Auditing and Logging + +Cloud Trace Service (CTS) is a log audit service for Huawei Cloud security. It allows +you to collect, store, and query cloud resource operation records. You can use +these records to perform security analysis, audit compliance, track resource +changes, and locate faults. + +After CTS is enabled, traces can be generated for operations performed on the +Direct Connect console. + +● + +● + +● + +For details about how to enable and configure CTS, see Enabling CTS. + +For details about key operations of Direct Connect, see Key Operations +Recorded by CTS. + +For details about traces, see Viewing Traces. + +6.3 Monitoring Security Risks + +Cloud Eye is a monitoring service provided by Huawei Cloud. It provides +capabilities like real-time monitoring, timely alarm reporting, resource groups, and +website monitoring. Cloud Eye enables you to keep track of your resource usages +and service statuses on the cloud, and quickly respond to exceptions to ensure +that services run smoothly. + +Monitoring is critical to ensuring the performance, reliability, and availability of +Direct Connect. Cloud Eye automatically monitors your connections in real time, +collects and displays monitoring data in a convenient, visualized manner, and +allows you to manage alarms and notifications, helping you watch your +connection performance. + +For details about supported metrics and how to create alarm rules, see +Monitoring. + +6.4 Certificates + +Compliance Certificates + +Huawei Cloud services and platforms have obtained various security and +compliance certifications from authoritative organizations, such as International +Organization for Standardization (ISO). You can download them from the +console. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +16 + + Direct Connect +Service Overview + +6 Security + +Figure 6-1 Downloading compliance certificates + +Resource Center + +Huawei Cloud also provides the following resources to help users meet compliance +requirements. For details, see Resource Center. + +Figure 6-2 Resource center + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +17 + + Direct Connect +Service Overview + +7 Permissions + +7 Permissions + +If you need to assign different permissions to employees in your enterprise to +access your Direct Connect resources, Identity and Access Management (IAM) is a +good choice for fine-grained permissions management. IAM provides identity +authentication, permissions management, and access control, helping you securely +access your cloud resources. If your HUAWEI ID does not require IAM for +permissions management, you can skip this section. + +IAM is a free service. You pay only for the resources in your account. + +With IAM, you can control access to specific cloud resources. For example, some +software developers in your enterprise need to use Direct Connect but should not +be allowed to delete other Direct Connect resources or perform any other high- +risk operations. In this scenario, you can create IAM users for the software +developers and grant them only the required permissions. + +IAM supports role/policy-based authorization and identity policy-based +authorization. + +The following table describes the differences between the two authorization +models. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +18 + + Direct Connect +Service Overview + +7 Permissions + +Table 7-1 Differences between role/policy-based and identity policy-based +authorization + +Name Authoriz + +Role/ +Policy + +ation +Using + +User- +permissi +on- +authoriz +ation +scope + +Policies + +Identit +y +policy + +Permissio +ns + +Authorization +Method + +Scenario + +● Syste +m- +define +d roles + +● Syste +m- +define +d +policie +s + +● Custo +m +policie +s + +● Syste +m- +define +d +identit +y +policie +s + +● Custo +m +policie +s + +Assigning +roles or +policies to +principals + +To authorize a user, you need +to add it to a user group first +and then specify the scope of +authorization. It is hard to +provide fine-grained +permissions control using +authorization by user groups +and a limited number of +condition keys. This method is +suitable for small- and +medium-sized enterprises. + +● Assigning +identity +policies to +principals + +● Attaching +identity +policies to +principals + +You can authorize a user by +attaching an identity policy to +it. User-specific authorization +and a variety of key +conditions allow for more +fine-grained permissions +control. However, this model +can be hard to set up. It +requires a certain amount of +expertise and is suitable for +medium- and large-sized +enterprises. + +Assume that you want to grant IAM users the permission to create ECSs in CN +North-Beijing4 and OBS buckets in CN South-Guangzhou. With role/policy-based +authorization, the administrator needs to create two custom policies and attach +both to the IAM users. With identity policy-based authorization, the administrator +only needs to create one custom policy and configure the condition key +g:RequestedRegion for the policy, and then attach the policy to the principals or +grant the principals the access permissions to the specified regions. Identity policy- +based authorization is more flexible than role/policy-based authorization. + +Policies and actions in the two authorization models are not interoperable. You are +advised to use the role/policy-based authorization model. For details about +system-defined permissions, see Role/Policy-based Authorization and Identity +Policy-based Authorization. + +For more information about IAM, see IAM Service Overview. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +19 + + Direct Connect +Service Overview + +Role/Policy-based Authorization + +7 Permissions + +Direct Connect supports role/policy-based authorization. By default, new IAM +users do not have permissions assigned. You need to add a user to one or more +groups, and attach permissions policies or roles to these groups. Users inherit +permissions from the groups to which they are added and can perform specified +operations on cloud services based on the permissions. + +Direct Connect is a project-level service deployed and accessed in specific physical +regions. When you set Scope to Region-specific projects and select the specified +projects (for example, ap-southeast-2) in the specified regions (for example, AP- +Bangkok), the users only have permissions for VPCEP resources in the selected +projects. If you set Scope to All resources, the users have permissions for VPCEP +resources in all region-specific projects. When accessing Direct Connect, the users +need to switch to a region where they have been authorized to use this service. + +Table 7-2 lists all the system-defined policies for Direct Connect in role/policy- +based authorization. System-defined policies in the two authorization models are +not interoperable. + +Table 7-2 Direct Connect permissions + +Role/Policy + +Description + +Direct Connect +Administrator + +DCaaS Partner + +Administrator +permissions for Direct +Connect. Users with +these permissions can +perform all operations +on Direct Connect. + +To have these +permissions, users +must also have the +Tenant Guest and +VPC Administrator +permissions. + +Typ +e + +Syst +em +- +defi +ned +role + +Dependency + +Tenant Guest and VPC +Administrator + +● VPC Administrator: + +project-level policy, which +must be assigned in the +same project + +● Tenant Guest: project- +level policy, which must +be assigned in the same +project + +Direct Connect partner +permissions. Users +who have these +permissions can create +hosted operations for +others. + +Syst +em +- +defi +ned +role + +To have these +permissions, users +must also have the +Tenant Guest and +VPC Administrator +permissions. + +Tenant Guest and VPC +Administrator + +● VPC Administrator: + +project-level policy, which +must be assigned in the +same project + +● Tenant Guest: project- +level policy, which must +be assigned in the same +project + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +20 + + Direct Connect +Service Overview + +7 Permissions + +Role/Policy + +Description + +DCAAS FullAccess Permissions: all + +permissions for Direct +Connect + +Scope: project-level +service + +Permissions: read-only +permissions for Direct +Connect + +Scope: project-level +service + +DCAAS +ReadOnlyAccess + +Dependency + +None + +None + +Typ +e + +Syst +em +- +defi +ned +poli +cy + +Syst +em +- +defi +ned +poli +cy + +Table 7-3 lists common operations supported by each system-defined policy for +Direct Connect. + +Table 7-3 Common operations supported by each system-defined policy + +Operation + +Creating a +connection + +Viewing a +connection + +Modifying a +connection + +Deleting a +connection + +Creating a +virtual +gateway + +Viewing a +virtual +gateway + +Modifying a +virtual +gateway + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +21 + + Direct Connect +Service Overview + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +7 Permissions + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Operation + +Deleting a +virtual +gateway + +Creating a +global DC +gateway + +Viewing a +global DC +gateway + +Modifying a +global DC +gateway + +Deleting a +global DC +gateway + +Creating a +virtual +interface + +Viewing a +virtual +interface + +Modifying a +virtual +interface + +Deleting a +virtual +interface + +Creating an +operations +connection + +Viewing an +operations +connection + +Modifying an +operations +connection + +Deleting an +operations +connection + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +22 + + Direct Connect +Service Overview + +Direct +Connect +Administrato +r + +DCaaS +Partner + +DCAAS +FullAccess + +7 Permissions + +DCAAS +ReadOnlyAcc +ess + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +Operation + +Creating a +hosted +connection + +Viewing a +hosted +connection + +Modifying a +hosted +connection + +Deleting a +hosted +connection + +Roles or Policies that the Direct Connect Console Depends on + +Table 7-4 Roles or policies required for operations on the Direct Connect console + +Function + +Connect +gateways + +Dependent Cloud +Service/Resource + +Elastic IP + +Virtual +interfaces + +Roles or Policies Required + +You need to configure the EIP +ReadOnlyAccess permission when +you perform the following operations +on connect gateways on the Direct +Connect console: + +● In the connect gateway list, view +the location, bind global EIPs to a +connect gateway, and manage the +global EIPs bound to a connect +gateway. + +● On the connect gateway details +page, view all the global EIPs, +binding or unbinding global EIPs to +the connect gateway. + +When deleting a virtual interface on +the Direct Connect console, you need +the EIP ReadOnlyAccess permission. + +Identity Policy-based Authorization + +Direct Connect supports identity policy-based authorization. Table 7-5 lists all the +system-defined policies for Direct Connect with identity policy-based + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +23 + + Direct Connect +Service Overview + +7 Permissions + +authorization. System-defined policies in identity policy-based authorization are +not interoperable with those in role/policy-based authorization. + +Table 7-5 System-defined policies supported by Direct Connect + +Policy Name + +Description + +Policy Type + +DCAASFullAccessPolicy + +Full permissions for Direct +Connect + +System-defined identity +policy + +DCAASReadOnlyPolicy + +Read-only permissions for +Direct Connect + +System-defined identity +policy + +Table 7-6 lists common operations supported by system-defined identity policies +for Direct Connect. + +Table 7-6 Common operations supported by system-defined identity policies + +DCAASFullAccessPolicy + +DCAASReadOnlyPolicy + +Operation + +Creating a +connection + +√ + +Viewing a connection √ + +Modifying a +connection + +Deleting a +connection + +Creating a virtual +gateway + +Viewing a virtual +gateway + +Modifying a virtual +gateway + +Deleting a virtual +gateway + +Creating a global DC +gateway + +Viewing a global DC +gateway + +Modifying a global +DC gateway + +Deleting a global DC +gateway + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +24 + + Direct Connect +Service Overview + +7 Permissions + +Operation + +DCAASFullAccessPolicy + +DCAASReadOnlyPolicy + +Creating a virtual +interface + +Viewing a virtual +interface + +Modifying a virtual +interface + +Deleting a virtual +interface + +Creating an +operations +connection + +Viewing an +operations +connection + +Modifying an +operations +connection + +Deleting an +operations +connection + +Creating a hosted +connection + +Viewing a hosted +connection + +Modifying a hosted +connection + +Deleting a hosted +connection + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +√ + +× + +√ + +× + +× + +× + +√ + +× + +× + +× + +√ + +× + +× + +Helpful Links + +● + +IAM Service Overview + +● Using IAM to Grant Access to Direct Connect + +● + +Identity Policy-based Authorization + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +25 + + Direct Connect +Service Overview + +8 Notes and Constraints + +8 Notes and Constraints + +Quotas + +Resource + +Default Quota + +How to Increase Quota + +Number of connections +that can be created by +an account in each +region + +Number of global DC +gateways that can be +created by an account in +each region + +Number of virtual +interfaces that can be +created by an account in +each region + +Number of local subnets +for a virtual gateway + +Number of routes for +BGP sessions on a virtual +interface + +Number of remote +subnets for a virtual +interface + +10 + +5 + +50 + +50 + +100 + +50 + +Submit a service ticket. + +Submit a service ticket. + +Submit a service ticket. + +Increasing the quota is +not allowed. + +Submit a service ticket. + +Submit a service ticket. + +Restrictions on Locations + +Before creating a connection, you need to select a location. Note the following +restrictions: + +● + +There may be more than one location in each region. In this case, network +latency from each location to different AZs in the region should be less than 5 +ms. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +26 + + Direct Connect +Service Overview + +8 Notes and Constraints + +● + +If your workloads have demanding requirements for network latency, you can +submit a service ticket to confirm which location is the nearest to the AZ +where your cloud servers are running. + +For details about the locations in each region, see Direct Connect Locations. + +Product Use Restrictions + +● + +● + +● + +● + +The CIDR block of the VPC cannot overlap with the CIDR block used by the +on-premises network. + +The on-premises network cannot use 100.64.0.0/10, 127.0.0.0/8, +169.254.0.0/16, and 224.0.0.0/3 because they are reserved for the VPC service. + +1GE and 10GE single-mode optical ports can transmit data up to 10 km. If +you need an optical port to transmit data for more than 10 km, or you need a +40GE or 100GE port, you need to purchase optical modules by yourself. For +suggestions on optical module selection, submit a service ticket. + +If you use a Direct Connect connection to access ELB, you must select Source +IP hash as the load balancing algorithm and disable sticky sessions for ELB. + +By default, Direct Connect is not interconnected with Enterprise Switch. To +access Enterprise Switch using Direct Connect, submit a service ticket. + +● Direct Connect can respond to common ICMP packets (echo packets whose +type is 8 and code is 0 and do not carry IP options) for ping detections. + +● + +● + +For each connection, a maximum of 30 ping detections can be performed on +the local gateway IP address over the port per second. + +To connect on-premises data centers through enterprise routers, ensure that +the on-premises data centers are connected to different devices. For details, +see submit a service ticket. + +● + +A virtual gateway can be associated with only two different devices. + +Network Requirements + +● + +● + +● + +● + +● + +● + +Your on-premises network must use a single-mode fiber with a 1GE, 10GE, +40GE, or 100GE optical module to connect to the access device on Huawei +Cloud. In addition, key parameters such as the LC, wavelength, and distance +must be consistent with those of the access device on Huawei Cloud. +Examples of optical module parameters: 1 GE, LC single-mode, 1,310 nm, and +10 km + +Auto-negotiation for the ports must be disabled. Port speed and full-duplex +mode must be manually configured. + +802.1Q VLAN encapsulation must be supported on all devices of the entire +connection, including intermediate devices. + +Your device must support Border Gateway Protocol (BGP) and BGP MD5 +authentication or static routing. + +(Optional) You can configure Bidirectional Forwarding Detection (BFD) on the +network. + +A physical connection supports jumbo frames with a default maximum +transmission unit (MTU) size of 8,500 bytes. If you need to configure a larger +MTU, contact the Direct Connect manager. + +If you ping the IP address of the local gateway of a Direct Connect connection +for connectivity test, the MTU is 1,500 bytes. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +27 + + Direct Connect +Service Overview + +8 Notes and Constraints + +● + +Private IP addresses are recommended for cloud and on-premises servers, and +the IP address ranges used for communications cannot overlap. + +Construction Notes + +● + +Your construction company must comply with the regulations presented by +the equipment room provider and engineers. In case of any violation, the +construction cannot be completed. + +● No optical-to-electrical converters can be hosted or installed in the equipment + +room. + +● Network blocking due to state policies or Huawei Cloud management will + +delay the construction. In the event of such situation, contact your Direct +Connect manager. + +● + +● + +The equipment room at a location is operated by a telecom carrier or a third +party. If there are fees for connecting your leased line to the equipment room +or an in-building cable, pay the fees to the equipment room provider. + +You need to apply for a Letter of Authorization (LOA) and present the LOA +when entering the equipment room for construction. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +28 + + Direct Connect +Service Overview + +9 Direct Connect and Other Services + +9 Direct Connect and Other Services + +Figure 9-1 Direct Connect and other services + +Table 9-1 Related services + +Cloud +Service + +Interaction + +Reference + +Virtual +Private Cloud +(VPC) + +Your on-premises data center can +access the VPCs through Direct +Connect. + +Creating a VPC with a +Subnet + +You can use VPC Peering to connect +the VPC your on-premises data center +is accessing to other VPCs in the same +region so that your on-premises data +center can access all these VPCs. + +Connecting an On- +Premises Data Center +to Multiple VPCs in +the Same Region +Using Direct Connect +and VPC Peering + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +29 + + Direct Connect +Service Overview + +9 Direct Connect and Other Services + +Cloud +Service + +Enterprise +Router + +NAT +Gateway + +VPC +Endpoint +(VPCEP) + +Cloud Eye + +Identity and +Access +Management +(IAM) + +Interaction + +Reference + +You can connect your on-premises +data center to an enterprise router, so +that your on-premises data center can +access the VPCs attached to the +enterprise router. + +- + +On-premises servers can share a NAT +gateway to access the Internet or +provide services that are accessible +from the Internet. + +A VPC endpoint can connect your on- +premises data center to a cloud +service through a VPN or Direct +Connect connection over a private +network. + +Cloud Eye monitors Direct Connect +resources and allows you to view +visualized graphs. + +You can grant different permissions +for users to control access to Direct +Connect resources. + +NAT Gateway + +Configuring a VPC +Endpoint for Accessing +the Private IP Address +of OBS + +Viewing Metrics + +Identity and Access +Management + +Cloud Trace +Service (CTS) + +You can record operations performed +on Direct Connect. + +Key Operations +Recorded by CTS + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +30 + + Direct Connect +Service Overview + +10 Basic Concepts + +10 Basic Concepts + +Connection + +A connection is a dedicated channel that connects your on-premises data center +to the cloud. Connections are more stable, reliable, and secure than Internet-based +connections, and provide up to 100 Gbit/s of bandwidth. Direct Connect provides +ports only. After you request a connection, you need to work with the carrier and +Huawei Cloud to establish network connectivity between your on-premises data +center and the cloud. + +Direct Connect provides standard connections and hosted connections. + +● + +Standard connection: A standard connection provides an exclusive port. You +can create standard connections on the management console. To improve +reliability, you can create multiple connections terminated at different +locations and allow them to serve as backups for each other. + +You can choose self-service connections or full-service connections. + +– + +– + +Self-service connection: Huawei Cloud only provides the port. You need to +create a connection on the console, and lease a line from a carrier. + +Full-service connection: You only need to create a connection on the +console, and Huawei Cloud will complete all operations required for +network connectivity. + +Figure 10-1 shows an example of how standard connections work. + +Figure 10-1 Accessing the cloud using standard connections + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +31 + + Direct Connect +Service Overview + +10 Basic Concepts + +● Hosted connection: If you use a hosted connection to access the cloud, the +port you use is shared with other users. These connections are created by a +carrier (a partner of Huawei Cloud), who allocates the required VLAN and +bandwidth for your connection. Only one virtual interface can be associated +with each hosted connection. + +If you are a partner, you can request operations connections. If you are a +common user, you can purchase a host connection from your partner. Hosted +connections must be hosted on operations connections, and your partner will +allocate VLAN and bandwidth resources to the connection. + +Figure 10-2 shows an example of how hosted connections work. + +Figure 10-2 Accessing the cloud using hosted connections + +Connections support redundant configuration. If there are two connections +terminated at different locations in the same region, they can work in an active/ +standby pair to back each other up. If one connection becomes faulty, the other +will take over, ensuring stable services. + +Virtual Gateway + +A virtual gateway is a logical gateway that enables an on-premises data center +to access a VPC over a connection. To enable an on-premises data center to access +a VPC over a connection, you can associate the VPC with a virtual gateway. To +access other VPCs, you can use VPC Peering or Cloud Connect to connect the VPC +your on-premises data center is accessing to these VPCs. + +A virtual gateway can only have one VPC associated. An on-premises data center +can access the same VPC over two connections through one virtual gateway. + +Global DC Gateway + +A global DC gateway enables your on-premises data center to access VPCs in +multiple regions so you can use a single connection to provide high-speed access +to cloud compute and storage resources in any region. + +A global DC gateway can be attached to different enterprise routers to build a +central network so that your on-premises data center can access the VPCs in +different regions over the Huawei backbone network. This reduces network +latency, simplifies network topology, and improves O&M efficiency. + +A global DC gateway can only be associated with connections terminated at the +same Direct Connect location. If there are multiple connections terminated at + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +32 + + Direct Connect +Service Overview + +10 Basic Concepts + +different Direct Connect locations, you need to create multiple global DC +gateways. + +Virtual Interface + +A virtual interface is a point of entry for an on-premises data center to access a +VPC over a connection. A virtual interface associates a connection with a virtual +gateway and connects the virtual gateway to a remote gateway, enabling +communications between the on-premises data center and the VPC. + +Virtual interfaces support static routing and BGP routing. You can use BGP to +connect your on-premises data center to the virtual gateway over a connection. +BGP helps you build a hybrid cloud more efficiently, flexibly, and reliably. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +33 + + Direct Connect +Service Overview + +11 Region and AZ + +11 Region and AZ + +Concept + +A region and availability zone (AZ) identify the location of a data center. You can +create resources in a specific region and AZ. + +● + +● + +Regions are divided based on geographical location and network latency. +Public services, such as Elastic Cloud Server (ECS), Elastic Volume Service +(EVS), Object Storage Service (OBS), Virtual Private Cloud (VPC), Elastic IP +(EIP), and Image Management Service (IMS), are shared within the same +region. Regions are classified into universal regions and dedicated regions. A +universal region provides universal cloud services for common tenants. A +dedicated region provides specific services for specific tenants. + +An AZ contains one or more physical data centers. Each AZ has independent +cooling, fire extinguishing, moisture-proof, and electricity facilities. Within an +AZ, computing, network, storage, and other resources are logically divided +into multiple clusters. + +Figure 11-1 shows the relationship between regions and AZs. + +Figure 11-1 Regions and AZs + +Huawei Cloud provides services in many regions around the world. You can select +a region and an AZ based on requirements. For more information, see Huawei +Cloud Global Products and Services. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +34 + + Direct Connect +Service Overview + +Selecting a Region + +11 Region and AZ + +When selecting a region, consider the following factors: + +● + +Location + +It is recommended that you select the closest region for lower network +latency and quick access. + +– + +– + +– + +If your target users are in Asia Pacific (excluding the Chinese mainland), +select the CN-Hong Kong, AP-Bangkok, or AP-Singapore region. + +If your target users are in Africa, select the AF-Johannesburg region. + +If your target users are in Latin America, select the LA-Santiago region. + +NO TE + +The LA-Santiago region is located in Chile. + +● + +Resource price + +Resource prices may vary in different regions. For details, see Product Pricing +Details. + +Selecting an AZ + +When deploying resources, consider your applications' requirements on disaster +recovery (DR) and network latency. + +● + +● + +For high DR capability, deploy resources in different AZs within the same +region. + +For lower network latency, deploy resources in the same AZ. + +Regions and Endpoints + +Before you use an API to call resources, specify its region and endpoint. For more +details, see Regions and Endpoints. + +Issue 01 (2026-02-27) + +Copyright © Huawei Cloud Computing Technologies Co., Ltd. + +35 + diff --git a/q135/random_k4/random_3.md b/q135/random_k4/random_3.md index 55365db142facf9f8b5c1d00700a8b3e272fdfb8..2e9f5241e58113455522088c7a6110de8838d130 100644 --- a/q135/random_k4/random_3.md +++ b/q135/random_k4/random_3.md @@ -1,511 +1,1208 @@ -MOBILE DEVICES | 2024 - -2024 - -MOBILE DEVICES - -SAMPLE REPORT - -Get an in-depth analysis of the current state, trends, and future of the -mobile devices market, including insights into key drivers, changing -product lifecycles, and manufacturing challenges. - -CONFIDENTIAL AND PROPRIETARY - -Page 1 of 25 - - MOBILE DEVICES | 2024 - -Table of Contents - -1. - -2. - -Introduction ...................................................................................................................................................... 4 - -Industry Overview .......................................................................................................................................... 5 - -2.1 Segment Outlook .............................................................................................................................. 5 - -2.1.1 Smartphones ......................................................................................................................... 5 - -2.1.2 Tablets ..................................................................................................................................... 6 - -2.1.3 Smart Wearables ................................................................................................................. 7 - -2.1.4 Smart Hearing ....................................................................................................................... 8 - -2.1.5 Smart Clothing ...................................................................................................................... 8 - -2.1.6 China ......................................................................................................................................... 9 - -2.1.7 US .............................................................................................................................................. 9 - -2.1.8 Europe ...................................................................................................................................10 - -2.1.9 Latin America ......................................................................................................................11 - -2.2 Market Growth Enablers ..............................................................................................................12 - -2.2.1 5G Network Expansion .....................................................................................................12 - -2.2.2 The Rise of the Internet of Things (IoT) ......................................................................12 - -2.2.3 Shifting Consumer Preferences for Technology-Centric Living .......................12 - -2.2.4 Advancements in Artificial Intelligence (AI) .............................................................13 - -2.2.5 Growing Importance of Mobile Devices in Emerging Markets ...........................13 - -2.3 Market Inhibitors.............................................................................................................................14 - -2.3.1 Concerns over Data Privacy and Security .................................................................14 - -2.3.2 Market Saturation in Developed Economies ............................................................14 - -2.3.3 E-Waste and Sustainability Concerns ........................................................................14 - -3. Supply Chain Analysis ................................................................................................................................15 - -3.1 Geopolitical Tensions and Supply Chain Realignment ......................................................15 - -3.2 Semiconductor Scarcity and Strategic Countermeasures ...............................................15 - -3.3 Supplier Relationship Transformation ....................................................................................15 - -3.4 - -3.5 - -3.6 - -Localization Imperative ................................................................................................................16 - -Ecosystem Collaboration .............................................................................................................16 - -ESG Challenges ...............................................................................................................................16 - -3.7 China+1 ..............................................................................................................................................16 - -3.8 - -Impact of AI on Supply Chain .....................................................................................................17 - -CONFIDENTIAL AND PROPRIETARY - -Page 2 of 25 - - MOBILE DEVICES | 2024 - -4. - -Technology Evolution ..................................................................................................................................18 - -4.1 - -5G, 6G, and Connected Ecosystem ...........................................................................................18 - -4.2 Augmented Reality, Virtual Reality, and Immersive Experiences .................................18 - -4.3 - -Foldable Devices.............................................................................................................................19 - -4.4 Ultra-Fast Charging Technologies ...........................................................................................19 - -4.5 - -Flexible Electronic Skins .............................................................................................................19 - -4.6 Next-Generation Wi-Fi .................................................................................................................19 - -4.7 Wearable Health Monitors...........................................................................................................20 - -4.8 - -4.9 - -Internet of Things (IoT) Integration ..........................................................................................20 - -Enhanced Processing Power ......................................................................................................20 - -5. - -Jabil Insights & Strategic Considerations ............................................................................................21 - -5.1 Current Market Outlook – Analyst Perspective ...................................................................21 - -5.2 Strategic Considerations .............................................................................................................22 - -5.3 - -Jabil Spend Data .............................................................................................................................23 - -CONFIDENTIAL AND PROPRIETARY - -Page 3 of 25 - - MOBILE DEVICES | 2024 - -1. - -Introduction - -The mobile device market is dynamic, with the industry constantly innovating to meet evolving -consumer demand. Smartphones and tablets, the cornerstone of this market, are no longer -simply communication tools; they've evolved into productivity powerhouses. Cutting-edge -smart wearables, hearables, and clothing are redefining how humans connect with -technology, offering health monitoring, augmented experiences, and seamless -communication. The rise of 5G networks, advancements in artificial intelligence (AI), and the -expanding Internet of Things (IoT) are accelerating the development of increasingly -sophisticated mobile devices. This transformation is fueled by the desire for global -connectivity, enhanced productivity, personalized experiences, and the adoption of cutting- -edge technology. As mobile devices become increasingly integrated into daily life, a -comprehensive understanding of this market landscape is crucial to support informed -business decisions. - -Scope of work - -CONFIDENTIAL AND PROPRIETARY - -Page 4 of 25 - - MOBILE DEVICES | 2024 - -2. - -Industry Overview - -The global mobile device market is experiencing a period of transition and adaptation. While -smartphone sales have shown signs of plateauing due to incremental upgrades and market -saturation, other segments are emerging as growth drivers. Smart hearables and wearables -are popular, offering enhanced connectivity and health-tracking features. Foldable devices -are carving out a niche, albeit with high prices and durability concerns. Smart clothing is still -in its nascent stage but shows promise in integrating technology seamlessly into everyday -life. Tablets, meanwhile, are finding renewed relevance for productivity and entertainment -purposes. Overall, the mobile device landscape is diversifying, with innovation focusing on -integrating various devices and creating a more connected and personalized user experience. - -2.1 Segment Outlook - -2.1.1 Smartphones - -While global smartphone shipment growth might have slowed down in previous years, the -market is expected to stabilize and see modest growth due to several factors. One key driver -is the replacement cycle for 5G-compatible devices, as companies like Apple and Samsung, -among others, continue to release advanced models. Additionally, the rise of foldable -smartphones is creating a new niche market segment with premium pricing. According to an -article published by Arstechnica in February 2024, the top 7 selling smartphones in 2023 were -manufactured by Apple, whereas Samsung manufactured the next 3 in the list. - -Figure: List of Top 10 highest-selling smartphones in 2023 and their adoption rate (%) - -CONFIDENTIAL AND PROPRIETARY - -Page 5 of 25 - - 2.1.2 Tablets - -MOBILE DEVICES | 2024 - -Figure: Market Share of Top Tablet Vendors Worldwide (April 2023 – April 2024) - -CONFIDENTIAL AND PROPRIETARY - -Page 6 of 25 - - 2.1.3 Smart Wearables - -MOBILE DEVICES | 2024 - -CONFIDENTIAL AND PROPRIETARY - -Page 7 of 25 - - 2.1.4 Smart Hearing - -MOBILE DEVICES | 2024 - -2.1.5 Smart Clothing - -CONFIDENTIAL AND PROPRIETARY - -Page 8 of 25 - - MOBILE DEVICES | 2024 - -2.1.6 China - -China is expected to remain dominant in the global mobile devices market, propelled by its -massive consumer base and robust local manufacturing capabilities. With leading companies -like Huawei, Xiaomi, and OPPO continuously innovating, China's market share is substantial. -However, geopolitical tensions and trade restrictions could impact growth. The aggressive -push for 5G adoption and beyond in China makes it likely that a significant portion of mobile -sales will be 5G-capable devices in the coming years. - -Figure: Chinese Smartphone Shipment, By Brand - -Source: Counterpoint Research - -2.1.7 US - -The US will likely see sustained growth in the mobile devices sector, driven by a strong -consumer preference for high-end smartphones and the rapid rollout of 5G networks. Apple -and Samsung remain dominant players, with the former enjoying a loyal customer base. -Introducing innovative features and technologies, such as foldable screens and AI-enhanced -applications, could spur replacement cycles. However, the market is nearing saturation, with -smartphone penetration already high. Growth may increasingly come from wearable devices -and smart home technologies. - -CONFIDENTIAL AND PROPRIETARY - -Page 9 of 25 - - MOBILE DEVICES | 2024 - -2.1.8 Europe - -CONFIDENTIAL AND PROPRIETARY - -Page 10 of 25 - - 2.1.9 Latin America - -MOBILE DEVICES | 2024 - -CONFIDENTIAL AND PROPRIETARY - -Page 11 of 25 - - MOBILE DEVICES | 2024 - -2.2 Market Growth Enablers - -2.2.1 5G Network Expansion - -Increased speeds and decreased latency offered by 5G fuel demand for advanced mobile -devices capable of handling data-intensive applications, augmented reality experiences, and -cloud-based services. Furthermore, the growth of high-quality video streaming on mobile -devices drives increased subscriptions to content providers, which fuels demand for devices -capable of delivering an exceptional viewing experience. This surge in demand is driving -innovation across smartphones, wearables, and hearables. - -2.2.2 The Rise of the Internet of Things (IoT) - -2.2.3 Shifting Consumer Preferences for Technology-Centric Living - -CONFIDENTIAL AND PROPRIETARY - -Page 12 of 25 - - 2.2.4 Advancements in Artificial Intelligence (AI) - -MOBILE DEVICES | 2024 - -2.2.5 Growing Importance of Mobile Devices in Emerging Markets - -CONFIDENTIAL AND PROPRIETARY - -Page 13 of 25 - - MOBILE DEVICES | 2024 - -2.3 Market Inhibitors - -2.3.1 Concerns over Data Privacy and Security - -Data privacy and security concerns have cast a shadow over the mobile device market, -eroding consumer trust and hindering industry growth. Frequent and increasingly -sophisticated cyberattacks and invasive tracking and data collection practices by apps and -services have fueled these concerns. Vulnerabilities in operating systems and applications -further exacerbate the risks, leaving users' personal information exposed to potential -breaches and misuse. - -2.3.2 Market Saturation in Developed Economies - -2.3.3 E-Waste and Sustainability Concerns - -CONFIDENTIAL AND PROPRIETARY - -Page 14 of 25 - - MOBILE DEVICES | 2024 - -3. Supply Chain Analysis - -3.1 Geopolitical Tensions and Supply Chain Realignment - -The mobile devices industry faces increased complexity and risk due to heightened -geopolitical tensions, particularly between major technology-producing nations. These -tensions have led to stricter trade policies, tariffs, and restrictions on technology transfers, -significantly impacting global supply chain dynamics. In response, companies are urgently -reassessing and realigning their supply chains to reduce dependency on regions with high -geopolitical risks. This includes diversifying sourcing and manufacturing locations to more -geopolitically stable or neutral countries. Moreover, there is an accelerated trend towards the -friend-shoring/nearshoring of critical components where companies prefer to trade with -allies or within blocks that share similar regulatory and political frameworks. This strategic -shift aims to safeguard access to essential materials and components, such as rare earth -metals and advanced semiconductors, which are pivotal for mobile device manufacturing. -Additionally, companies are increasingly investing in technology and infrastructure to -enhance supply chain visibility and resilience, enabling more agile responses to future -geopolitical shifts. - -3.2 Semiconductor Scarcity and Strategic Countermeasures - -3.3 Supplier Relationship Transformation - -CONFIDENTIAL AND PROPRIETARY - -Page 15 of 25 - - 3.4 Localization Imperative - -MOBILE DEVICES | 2024 - -3.5 Ecosystem Collaboration - -3.6 ESG Challenges - -3.7 China+1 - -. - -CONFIDENTIAL AND PROPRIETARY - -Page 16 of 25 - - 3.8 - -Impact of AI on Supply Chain - -MOBILE DEVICES | 2024 - -CONFIDENTIAL AND PROPRIETARY - -Page 17 of 25 - - MOBILE DEVICES | 2024 - -4. Technology Evolution - -4.1 5G, 6G, and Connected Ecosystem - -The acceleration of 5G adoption and early exploration of 6G's capabilities will underpin a new -generation of mobile devices and experiences. Faster speeds, reduced latency, and massive -device connectivity will supercharge applications like real-time AR/VR experiences, remote -collaboration, high-fidelity cloud gaming, and a flourishing Internet of Things (IoT) landscape. -These developments drive innovation across smartphones, wearables, and hearables, -enabling unprecedented connectivity and interaction. - -4.2 Augmented Reality, Virtual Reality, and Immersive - -Experiences - -CONFIDENTIAL AND PROPRIETARY - -Page 18 of 25 - - 4.3 Foldable Devices - -MOBILE DEVICES | 2024 - -4.4 Ultra-Fast Charging Technologies - -4.5 Flexible Electronic Skins - -4.6 Next-Generation Wi-Fi - -CONFIDENTIAL AND PROPRIETARY - -Page 19 of 25 - - 4.7 Wearable Health Monitors - -MOBILE DEVICES | 2024 - -4.8 - -Internet of Things (IoT) Integration - -4.9 Enhanced Processing Power - -CONFIDENTIAL AND PROPRIETARY - -Page 20 of 25 - - MOBILE DEVICES | 2024 - -5. Jabil Insights & Strategic Considerations - -From a technological standpoint, the mobile device market is poised for significant growth in -the next 3-4 years. The widespread adoption of 5G networks is expected to unlock the full -potential of mobile devices, enabling faster speeds, lower latency, and new applications like -augmented reality and virtual reality. AI and machine learning advancements will further -enhance device capabilities, offering more personalized experiences and improved -functionality. The convergence of mobile devices with other emerging technologies like -blockchain and IoT will also create new opportunities for innovation and growth. - -From a regional standpoint, the mobile device market is expected to experience diverse -growth trajectories. In developed regions like North America and Europe, growth is expected -to be steady but slower due to market saturation. However, emerging markets like India, -Southeast Asia, and Africa will rapidly grow as smartphone penetration increases and -consumers upgrade to newer models with advanced features. These regions will become -critical drivers of the global mobile device market, with local brands playing a significant role -in catering to these diverse consumer bases' unique needs and preferences. Additionally, -government initiatives to improve digital infrastructure and promote affordable access to -mobile devices will further fuel growth in these regions. - -5.1 Current Market Outlook – Analyst Perspective - -CONFIDENTIAL AND PROPRIETARY - -Page 21 of 25 - - 5.2 Strategic Considerations - -MOBILE DEVICES | 2024 - -CONFIDENTIAL AND PROPRIETARY - -Page 22 of 25 - - MOBILE DEVICES | 2024 - -5.3 Jabil Spend Data - -Figure: Mobile Devices Spend by Category (2020 Vs. 2023) - -Note: This data is based on Jabil's historical spending data for the Mobile Device Customer Segment - -CONFIDENTIAL AND PROPRIETARY - -Page 23 of 25 - - MOBILE DEVICES | 2024 - -Figure: Category Spend YoY (%) - -Note: This data is based on Jabil's historical spending data for the Mobile Device Customer Segment - -CONFIDENTIAL AND PROPRIETARY - -Page 24 of 25 - - MOBILE DEVICES | 2024 - -This document is provided for informational purposes only. It is not intended to be used or -construed as investment advice. The information contained in this document is not an offer to -buy or sell any securities or other financial instruments. Any investment decisions made -based on the information contained in this document are made at the sole discretion of the -reader. The author(s) of this document are not investment advisors and are not registered -with any regulatory authority. The author(s) of this document disclaim any liability for any -direct or consequential loss arising from any use of this document or its contents. - -JABIL and the JABIL logo are the property of Jabil Inc. Other marks used herein are used for -identification purposes only and are the property of their respective owners. - -CONFIDENTIAL AND PROPRIETARY - -Page 25 of 25 - +DUET: Joint Exploration of User–Item Profiles in Recommendation System +YueChen1,*,YifeiSun1,*,LuWang2,†,FangkaiYang2,PuZhao2,MinjieHong3,YifeiDong4,MinghuaHe2, +NanHu2,JianjinZhang2,ZhiweiDai2,YuefengZhan2,WeihaoHan2,HaoSun2, +QingweiLin2,WeiweiDeng2,FengSun2,QiZhang2,SaravanRajmohan2,DongmeiZhang2 +1PekingUniversity2Microsoft3ZhejiangUniversity4KTHRoyalInstituteofTechnology +*Equalcontribution†Correspondingauthor +Abstract +| | | | User History: | Item History: | User History: | Item History: | +| --- | --- | --- | -------------------- | ------------------------- | -------------------- | ------------------------- | +| | | | • Misfits (Punk) | • Review: | • Misfits (Punk) | • Review: | +| | | | • I r o n M a iden | " P o l i s he d P o p " | • I r o n M a iden | " P o l i s he d P o p " | +Traditionalrecommendationsystemsrepresent ( M e t al ) • R e v i e w : " G r e a t ( M e t al ) • R e v i e w : " G r e a t +6202 rpA 51 ]RI.sc[ 1v10831.4062:viXra • • Parliament Funk Bass" +| usersanditemsasdensevectorsandlearnto | | | Parliament | Funk Bass" | | | +| ------------------------------------- | --- | --- | ----------- | ------------------ | ------ | ------------------ | +| | | | (Funk) | • Meta: "80s Rock" | (Funk) | • Meta: "80s Rock" | +aligntheminasharedlatentspaceforrelevance +| | | | | Duet | Independent | | +| --- | --- | --- | --- | ----- | ------------ | --- | +estimation. RecentLLM-basedrecommenders (Joint Alignment) Generation +| instead leverage | natural-language | representa- | | | | | +| ---------------- | ---------------- | ----------- | ------------- | ------------- | ------------- | ------------- | +| | | | User Profile: | Item History: | User Profile: | Item History: | +tionsthatareeasiertointerpretandintegrate Enthusiast of A peak-era FUNK- Fan of HEAVY A highly rated POP- +| | | | FUNK and SOUL. | ROCK album with | METAL and | ROCK album with | +| --- | --- | --- | --------------- | ---------------- | ---------- | ---------------- | +withdownstreamreasoningmodules. Thispa- Values technical genre-blurring PUNK. Prefers polished +| | | | mastery | creativity. | dark themes. | production. | +| --- | --- | --- | ------- | ----------- | ------------ | ----------- | +textual +per studies how to construct effective Semantic Alignment ✘ Semantic Mismatch +profilesforusersanditems,andhowtoalign +themforrecommendation. Acentraldifficulty Figure1: DUETalignsrawuseranditemdatabytrans- +| | | | forming them | into textual | profiles within | a shared se- | +| --- | --- | --- | ------------ | ------------ | --------------- | ------------ | +isthatthebestprofileformatisnotknownapri- +manticspace. +ori: manuallydesignedtemplatescanbebrittle +| andmisalignedwithtaskobjectives. | | Moreover, | | | | | +| -------------------------------- | --- | --------- | ------------ | ------------ | -------------------- | --- | +| | | | to introduce | semantically | rich, human-readable | | +generatinguseranditemprofilesindependently +mayproducedescriptionsthatareindividually representationsforrecommendation(Wangetal., +plausible yet semantically inconsistent for a 2025;Zhang,2024;Baoetal.,2023;Hongetal., +| specificuser–itempair. | WeproposeDUET,an | | | | | | +| ---------------------- | ---------------- | --- | -------------------------- | --- | ------------------- | --- | +| | | | 2025a,b,c;Wangetal.,2024). | | Anaturaldirectionis | | +interaction-awareprofilegeneratorthatjointly +toreplacelatentvectorswithtextualuseranditem +producesuseranditemprofilesconditionedon +profilesthatcanbeinspected,edited,andreusedby +| both user history | and item evidence. | DUET | | | | | +| ----------------- | ------------------ | ---- | --- | --- | --- | --- | +downstreamcomponentsincorporatedwithLLMs. +| follows a three-stage | procedure: | it first turns | | | | | +| --------------------- | ---------- | -------------- | --- | --- | --- | --- | +However,existingLLM-basedapproachesremain +rawhistoriesandmetadataintocompactcues, +then expands these cues into paired profile limited in two important ways. First, directly +promptsandthengenerateprofiles,andfinally promptinganLLMwithrawuseranditemhisto- +optimizesthegenerationpolicywithreinforce- +riestoobtainrecommendationsoftenyieldsnoisy +mentlearningusingdownstreamrecommenda- andincompletesignals,especiallywhenhistories +| tionperformanceasfeedback. | | Experimentson | | | | | +| -------------------------- | --- | ------------- | --------- | ------------------------ | --- | ------------- | +| | | | are long, | sparse, or heterogeneous | | (Wang et al., | +threereal-worlddatasetsshowthatDUETcon- +| | | | 2025). Second, | profile-based | methods | typically | +| --- | --- | --- | -------------- | ------------- | ------- | --------- | +sistentlyoutperformsstrongbaselines,demon- +relyonmanuallydesignedtemplatesorhandcrafted +stratingthebenefitsoftemplate-freeprofileex- +attributes,whichrequiressubstantialhumanengi- +plorationandjointuser–itemtextualalignment. +Project code page: https://github.com/duet- neering and constrains the representation space. +review/duet_code. More fundamentally, many approaches generate +userprofilesanditemprofilesindependently,with- +1 Introduction +| | | | out modeling | how user | preferences | and item se- | +| --- | --- | --- | ------------ | -------- | ----------- | ------------ | +Traditional recommendation systems represent manticsinteractatdecisiontime(Yangetal.,2023; +Xietal.,2024). +usersanditemsasdensevectorsandlearntoalign +theminasharedlatentspaceforrelevanceestima- Toaddressthesechallenges,wepropose DUET, +tion (Covington et al., 2016; Wu, 2023). While ajointuser–itemprofilegeneratorthattakesboth +effective,suchembeddingsareopaque: theyoffer userhistoryanditemhistoryasinputandproduces +limitedinterpretabilityandmakeitdifficulttoan- a paired set of profiles for the interaction. Cru- +alyzewhyanitemisrecommended. Recentwork cially, DUET doesnotrequireprofiletemplates: it +thereforeleverageslargelanguagemodels(LLMs) istrainedwithreinforcementlearningusingfeed- + +back from downstream recommendation perfor- asseeninworkslikeCRESDUP(Chenetal.,2007) +mance,enablingittoexploreanddiscovereffective andUPCSim(Widiyaningtyasetal.,2021). While +profileformatsautomatically. Figure1illustrates foundational,thesemethodswerelimitedbytheir +whyjointprofilingmatters. Theuserhaslistened rigid,hand-engineeredfeatures. Morerecently,the +toMisfits(punk),IronMaiden(metal),andParlia- adventofLargeLanguageModels(LLMs)hasen- +ment(funk),whilethecandidatealbumisdescribed abledashifttowardsgeneratingprofilesinnatural +byreviewssuchas“polishedpop”and“greatfunk language. Studies such as KAR (Xi et al., 2024), +bass”andametatag“’80srock.” Consideringboth GPG(Zhang,2024),PALR(Yangetal.,2023),and +sides together, DUET can reconcile these signals LettinGo (Wang et al., 2025) leverage LLMs to +intoacompatibleinterpretation(e.g.,highlighting createtextualuserprofilesfrombehavioraldata. A +the user’s funk/soul affinity and the item’s funk- keylimitationoftheseapproachesistwofold: they +rockcharacter). Incontrast,independentlygener- relyonstaticorpre-definedtemplatesforprofile +atedprofilesmayamplifydifferentfacets,summa- generation,andtheyoftenfocusexclusivelyonuser +rizingtheuseras“heavymetal/punk”whilesum- profiles,neglectingtherich,expressiveinformation +marizingtheitemas“pop-rock”,whichresultingin inherentinitemsandthecomplexdynamicsofuser- +asemanticallymismatchedpairthatobscuresthe iteminteractions. Incontrast,ourworkintroduces +truerelevancesignal. anewparadigmwherebothuseranditemprofiles +DUETproceedsinthreestages. First,rawhisto- arenotfixedbutaredynamicallyexploredandop- +ries and metadata are distilled into minimal cues timizedinasharedsemanticspacetodirectlyalign +thatcapturecompactbutinformativesignals. Sec- withrecommendationperformance. +ond, the model expands cues into richer profile +prompts to generate textual profiles, allowing ex- 2.2 ReinforcementLearningforLLM-Based +plorationoveralternativeprofilestructuresandem- RecommendationSystems +phases. Third,theresultingprofilesareconsumed +ReinforcementLearning(RL),particularlythrough +bydownstreamrecommenders,andtheirtaskfeed- +techniqueslikeRLHF,hasbecomeacoremethod +backisusedtooptimizetheprofilegenerationpol- +for aligning LLMs with specific objectives. This +icy. By coupling both sides in a shared semantic +approach has been adapted for recommendation +space,DUETlearnsuserprofilesthatreflectwhat +tasks (Wang et al., 2025; Lin et al., 2025; Deng +kindsofitemsauserprefersanditemprofilesthat +etal.,2025),butexistingeffortsoftenfacekeylimi- +reflectwhatkindsofusersanitemappealsto. +tations. Theyfrequentlyrelyonofflinerewardmod- +Ourcontributionsareasfollows: +els(Jeongetal.,2023;Chenetal.,2025;Heetal., +• Werepresentusersanditemsasnatural-language 2025; Liu et al., 2025) that do not adapt in real- +profiles and align them in a shared semantic timetosystemfeedback,asetupthatrisksissues +space,extendingtheclassicvector-basedalign- like reward hacking (Skalse et al., 2025). Other +mentprincipletointerpretabletextualrepresenta- methods(Sunetal.,2024;Luetal.,2024)restrict +tions. themselvestoofflinepreferencetuning(e.g.,DPO), +• We introduce an exploration-based framework which can easily overfit on static datasets. Our +thatstartsfromcue-basedinitialization,expands framework, DUET,overcomesthesechallengesby +cuesintocandidateprofiles,andjointlyoptimizes integrating RL into a closed-loop system where +user and item profiles with downstream recom- downstreamrecommendationperformanceserves +mendationfeedbackviareinforcementlearning, asthereal-timerewardsignal,allowingforthedy- +avoidingrigidtemplates. namicandinteractiverefinementoftextualprofiles. +• Extensiveexperimentsacrossmultiplereal-world +datasets show that DUET consistently outper- 3 Method +formsstrongbaselines,validatingbothjointpro- +filingandfeedback-drivenprofileoptimization. DUET is a closed-loop framework that trans- +forms raw user–item interaction histories into +2 RelatedWork performance-aligned textual profiles through +learnedrepresentationstrategies. AsshowninFig- +2.1 ProfilesinRecommendation +ure 2, DUET consists of three stages. (1) Cue- +Earlyrecommendationsystemsprimarilyreliedon Based Initialization distills interaction histories +pre-definedprofilesbasedonstructuredattributes, intoconciseevidence-basedcues. (2)JointExplo- + +CUE-BASED INITIALIZATION JOINT EXPLORATION ON-POLICY OPTIMIZATION +| | | | | Item | DU | | | Duet | | | | +| ----------- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | --- | +| ItemHistory | | | | | ET | | | | | | | +Initial CUE +Rated: Baba Is You (5★) +User: “Prefers retro puzzle games” +Reviewed: “I love logic- +Item: “Retro-style indie puzzle game” +| | | | based puzzle games.” | | | Profile Prompt | | | | | | +| --- | --- | --- | -------------------- | --- | --- | -------------- | --- | --- | --- | --- | --- | +Rated: The Witness (5★) +User: “Describe the user's typical +| | | | Rated: Fez (4★) | | puzzle-game preferences … | | | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒1 𝑃𝑟𝑜𝑓𝑖𝑙𝑒2 | … | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒𝑛 | | +| --- | --- | --- | --------------- | --- | ------------------------- | --- | --- | ----------------- | --- | -------- | --- | +Item: “… item's core mechanics… +| | | | | | | | Profile | Downstream Task Env | | | | +| ----------- | --- | --- | ----------------------- | ---- | --------------------------------- | --- | ------- | --------------------- | --- | --- | --- | +| | | | | User | User: “A player who enjoys logic- | | | | | | | +| UserHistory | | | | | intensive indie puzzle games, | | | | | | | +| | | | Category: Indie Puzzle | | | | | Ranking correctness | | | | +prefers retro visual aesthetics…” +| | | | Game | | | | | &format Reward | | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Reviews: +Item: “A retro-style indie puzzle +| | | | “Challenging logic | | game featuring challenging | | | Group Computation | | | | +| --- | --- | --- | -------------------- | ------------------------ | ------------------------------- | --- | --- | ------------------- | --- | ---- | --- | +| | | | mechanics.” | | logic mechanics that appeal to | | | | | | | +| | | | “ R e t r o p ix | e l- a rt st y l e . ” | hardcore puzzle players.” | | | 𝐴𝑑𝑣1 𝐴𝑑𝑣2 | | | | +| | | | “ H a r d b u | t re w a r d i n g . ” | | | | | … | 𝐴𝑑𝑣𝑛 | | +| | | | | Figure2: | OverviewoftheDUETframework. | | | | | | | +ration via Adaptive Profile Prompt Discovery 3.2 Cue-BasedInitialization +| jointly | explores | user-item’s | | profile prompts | that | | | | | | | +| ---------- | -------- | ----------- | -------- | --------------- | -------- | -------- | --------- | -------- | -------- | --- | --------- | +| | | | | | | Raw user | histories | and item | metadata | | are often | +| define how | user | and item | profiles | should | be writ- | | | | | | | +noisy,redundant,andnotdirectlysuitableforpro- +ten. (3)OptimizationviaOn-policyExploration +| | | | | | | fileconstruction. | | Toaddressthis,DUETintroduces | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ---------------------------- | --- | --- | --- | +jointlyoptimizesuseranditemprofilesunderdown- +theconceptofcues: +concisehypothesesthatsum- +streamrecommendationfeedback. +| | | | | | | marize minimal | | but informative | | aspects | of users | +| ------------------- | ------ | --- | ------------------------- | ------- | -------- | -------------- | ------------ | --------------- | -------------- | ------- | -------- | +| All three | stages | are | realized | through | a single | | | | | | | +| | | | | | | and items. | These | cues act | as lightweight | | seeds, | +| passinputandoutput: | | | cueextraction,self-prompt | | | | | | | | | +| | | | | | | which are | deliberately | underspecified | | so | that the | +construction,andprofilegenerationareproduced +systemcansubsequentlyexplorericherprofilefor- +inasinglesequence-to-sequencegenerationpass +mats. +atinferencetime,enablingefficientdeployment. +| | | | | | | Definition1(Cue). | | Acueisaminimaltextualhy- | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------------------ | --- | --- | --- | +3.1 ProblemFormulation +pothesisderivedfromhistoricaldatathathighlights +WeformulateprofilegenerationinDUETasanon- one potential aspect of a user’s preference or an +policyreinforcementlearningproblem,motivated item’scharacteristic. Ratherthanaimingforcom- +bytheabsenceofanytextualgroundtruthdefining pleteness, cues capture partial but salient signals +an optimal user or item profile. Profile quality is thatserveasstartingpointsforprofileexploration. +evaluatedsolelybyitsfunctionalutilityinafixed +| | | | | | | To extract | cues | automatically, | | the | LLM is | +| --- | --- | --- | --- | --- | --- | ---------- | ---- | -------------- | --- | --- | ------ | +recommendationenvironment. +| | | | | | | prompted | to summarize | minimal | | but informative | | +| ---- | ---------- | --- | ------------ | ------ | ------- | ------------------------ | ------------ | ------- | ----------------- | --------------- | --- | +| DUET | is modeled | as | a generative | policy | π θ in- | | | | | | | +| | | | | | | aspectsofuseroritemdata. | | | Forexample,givena | | | +teractingwithafrozendownstreamrecommender. +user’sinteractionhistory,themodelisguidedwith +| For each | user–item | pair, | the | state is | defined as | | | | | | | +| -------- | --------- | ----- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | +instructionssuchas: +| s = {H | ,H },whereH | | andH | denotetheuser | | | | | | | | +| ------ | ----------- | --- | ---- | ------------- | --- | --- | --- | --- | --- | --- | --- | +| | u i | | u | i | | | | | | | | +and item interaction histories. An action corre- CueExtractionPrompt +spondstoasingle-passjointgeneration +“Fromthehistorybelow,analyzetheuser’s +a = {C u ,S u ,P u ,C i ,S i ,P i }, historicalinteractionstounderstandprefer- +ences,ratingbehavior,reviewsentimentor +| where | (C ,C | ) are | cues | distilled | from his- | | | | | | | +| ----- | ----- | ----- | ---- | --------- | --------- | --- | --- | --- | --- | --- | --- | +u i +| | | | | | | anyotherdimension. | | Keepthedescription | | | | +| -------- | ------------ | ----------- | --- | ----------- | ------- | ------------------ | --- | ------------------ | --- | --- | --- | +| tory, (S | u ,S i ) are | constructed | | user-item’s | profile | | | | | | | +prompts,and(P ,P )arethefinaltextualprofiles. conciseandavoidfullsentences.” +u i +| Thepolicyπ | | (a | s)definesajointdistribution | | | | | | | | | | +| ---------- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +θ +over the entire generation sequence. As an on- This lightweight guidance allows the LLM to +policy agent, DUET is optimized using rewards maprawhistoriesandmetadataintocompacttex- +from its own sampled generations rather than by tualcues. Thedetaildexampleofcuecanbefound +| imitatingfixedsummaries. | | | | | | inAppendixA.3. | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | + +| 3.3 | JointExplorationviaAdaptiveProfile | | | | | | Single-PassInput | | | | | | +| --- | ---------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +PromptDiscovery Task:Generatestructuredprofilesfortheuseranditemfrom +historicaldata. +Aftercueextraction,DUETdoesnotdirectlysum- Combined Context: [User History H u ] + [Item +| | | | | | | | History | H i ]+[Avg | Ratings] | | | | +| ------------------------------------ | --- | --- | --- | --- | --- | ---------- | ------- | ---------- | -------- | --- | --- | --- | +| marizeuserinterestsoritemattributes. | | | | | | Instead,it | | | | | | | +“First,extractaconciseCueforbothuseranditem.Second, +| explores | | the space | of profile | construction | | strate- | | | | | | | +| -------- | --- | --------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | +basedonthecues,constructaProfilePromptthatdefines +gies—natural-languagepromptsthatdefinethefor- thedescriptionlogic.Finally,generatetheProfileguidedby +theconstructedProfilePrompt.” +mat,abstractionlevel,andattributeselectionlogic +↓DUETForwardPass↓ +| usedtogenerateprofiles. | | | | Explorationistherefore | | | | | | | | | +| -------------------------------------------- | --- | --- | --- | ---------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| performedoverhowprofilesshouldbeconstructed, | | | | | | | Single-PassOutput | | | | | | +User +ratherthanoversuperficialtextualparaphrases. +Cue:“prefersnostalgiclogic-basedchallenges” +For each user–item pair, DUET intro- Profileprompt: “Focuson1990svisualaestheticsand +| duces | an | explicit | intermediate | | variable, | the | strategicdepth.” | | | | | | +| ----- | --- | -------- | ------------ | --- | --------- | --- | ---------------- | --------- | --- | ----- | ----------- | ------------ | +| | | | | | | | Profile: | “A player | who | seeks | retro-style | visual charm | +constructed_prompt,whichservesasadiscrete +pairedwithdeepstrategicreasoning.” +| and | interpretable | | profile | prompt | S. | This profile | | | | | | | +| --- | ------------- | --- | ------- | ------ | --- | ------------ | --- | --- | --- | --- | --- | --- | +Item +promptisashortnatural-languageinstruction(e.g., +Cue:“retro-styleindiepuzzlewithhighdifficulty” +“Describetheuser’stypicalgamingpreferencesand +| | | | | | | | Profile | prompt: | “Describe | | pixel-art | graphics and | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --------- | --- | --------- | ------------ | +engagementpatterns”)thatspecifieswhataspects intellectualdifficultytomatchlogicpreference.” +Profile:“A2Dexperiencefeaturingpixelatednostalgia +| to | describe | and | how to | organize | them, | without | | | | | | | +| --- | -------- | --- | ------ | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +andchallengingmechanicsthatdemandlogicaldeduction.” +| containingtheprofilecontentitself. | | | | | | Conditioned | | | | | | | +| ---------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +onS,themodelgeneratesthefinaluseranditem Figure3: Single-passgenerationinDUET: cueextrac- +| profiles. | | | | | | | tion,profileprompt(constructedprompt),andprofile | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | --- | --- | +generationareproducedinonepassforbothuserand +| Exploration | | | is driven | by treating | | the profile | | | | | | | +| ----------- | --- | --- | --------- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +item. +| promptS | | asastochasticactionsampledfromthe | | | | | | | | | | | +| ------- | --- | --------------------------------- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| policyπ | | (S | Cue). | Duringtraining,theagentsam- | | | | | | | | | | +θ +plesdifferentprofilepromptinstructions,executes +| | | | | | | | andintellectualdifficulty”tomatch. | | | | | RLreinforces | +| ---------------------------- | --- | --- | --- | --- | ---------------- | --- | ---------------------------------- | --- | --- | --- | --- | ------------ | +| themtogenerateprofilepairs(P | | | | | ,P ),andreceives | | | | | | | | +u i +thissharedsemanticdirection,suppressingirrele- +| a reward | | based | on downstream | | recommendation | | | | | | | | +| --------- | --- | ----------------------------------- | ------------- | --- | -------------- | --- | ------------ | ------------- | -------- | --- | ----- | ---------------- | +| | | | | | | | vant signals | and | forcing | the | final | profiles to con- | +| accuracy. | | Thepolicyisoptimizedtoreinforcepro- | | | | | | | | | | | +| | | | | | | | verge | into a shared | semantic | | space | of nostalgia | +filepromptsthatconsistentlyyieldhigherrewards, +andlogic,significantlyimprovingrecommendation +enablingthemodeltoactivelysearchforeffective +accuracy. +profileconstructionformats. +Thisprocessisrealizedasaunifiedgeneration +3.4 OptimizationviaOn-policyExploration +pass, +| | | | | | | | On-policy | optimization. | | | We train | the profile | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | --- | -------- | ----------- | +O = [ Cue → profilepromptS → Profile], generator π in an on-policy manner against a +| | (cid:124)(cid:123)(cid:122)(cid:125) | | | | | (cid:124) (cid:123)(cid:122) (cid:125) | | θ | | | | | +| --- | ------------------------------------ | --- | --------- | ------------------ | --------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | (cid:124) | (cid:123)(cid:122) | (cid:125) | | | | | | | | +Context ProfileConstructionPrompt Execution frozendownstreammodelf, whichservesasthe +| | | | | | | (1) | environment | critic. | For | each | sampled | user–item | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | --- | ---- | ------- | --------- | +wheretheprofilepromptS parameterizesthepro- pair (u,i), the policy generates (P ,P ) and re- +u i +filegenerator. Atinferencetime,DUETcollapses ceivesascalarrewardbasedonthepredictionac- +thelearnedprofilepromptdistributionandgreed- curacy of f(P ,P ). The policy parameters are +u i +ilyexecutestheoptimalprofilepromptinasingle updatedtoreinforcegenerationsthatleadtolower +| forwardpass,introducingnoadditionallatency. | | | | | | | predictionerror. | | | | | | +| ------------------------------------------- | --- | ---------- | ------ | ------------ | -------- | ------- | --------------------------- | --- | --- | --- | --- | ------------- | +| Example | | of profile | prompt | Exploration. | | As | | | | | | | +| | | | | | | | Continuousfractionalreward. | | | | | Usingdiscrete | +| illustrated | | in Figure | 3, | DUET | resolves | this by | | | | | | | +integerratingsasrewardsleadstosparseandunsta- +| identifying | | a shared | context | in | a single | forward | | | | | | | +| ----------- | --- | -------- | ------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +blefeedback,asnear-misspredictions(e.g.,predict- +pass. ThepolicyextractsacoordinatedCue(e.g., +ing4foragroundtruth5)receivethesamepenalty +“prefersretropuzzlegames”)andconstructsapro- +| | | | | | | | assevereerrors. | | Toprovidedenseandinformative | | | | +| ---- | ------ | ---- | ------------------- | --- | --- | -------- | --------------- | --- | ---------------------------- | --- | --- | --- | +| file | prompt | (the | constructed_prompt) | | | that in- | | | | | | | +feedback,wedefineacontinuousfractionalreward: +structsthemodeltohighlight“nostalgicaesthetics +| and | strategic | logic” | for | the user, | while | simulta- | | | | | | | +| ------- | ---------- | ------ | ---------- | ---------- | ----- | -------- | --- | ------- | --- | ---- | ------ | -------- | +| | | | | | | | | | | | |y −yˆ | | | +| | | | | | | | | R (u,i) | | = 1− | ui | ui , (2) | +| neously | describing | | the item’s | “pixel-art | | graphics | | perf | | | | | +M + +where yˆ ui = f(P u ,P i ) is the predicted relevance ingsettingbasedondownstreampredictedrating +score and M is the maximum rating gap (e.g., scores. For each observed user–item interaction, +M = 4 for a 1–5 scale). This reward provides we randomly sample nine items that the user has +fine-grainedgradientsthatencourageincremental not interacted with to form a candidate set of ten +improvementsinrecommendationaccuracy. items, which are ranked in descending order ac- +We adopt Group Relative Policy Optimization cordingtotheirpredictedratings. Theground-truth +(GRPO) (DeepSeek-AI et al., 2025) to optimize interacted item is treated as the only positive in- +theprofilegeneratorundertheabovereward. The stance,andweadoptNDCG@Kastheevaluation +downstreamrecommenderf isfrozenthroughout metric, with K set to 1, 5, and 10, to assess how +training, which yields a stable policy optimiza- effectivelythelearnedprofilessupportcorrectitem +| tionsettingandpreventsrewarddriftorfeedback- | | | | ranking. | | | | | +| -------------------------------------------- | --- | --- | --- | --------- | ---------- | ---------- | ---- | ------- | +| | | | | Baselines | We compare | our method | with | several | +inducedrepresentationcollapse. +| | | | | representative | baselines. | 10H | directly | uses the | +| --- | --- | --- | --- | -------------- | ---------- | --- | -------- | -------- | +4 Experiment mostrecentinteractionhistoriesforpredictionwith- +| | | | | outconstructingexplicitprofiles. | | | KAR(Xietal., | | +| --- | --- | --- | --- | -------------------------------- | --- | --- | ------------ | --- | +4.1 ExperimentalSettings +2024)augmentsrecommendationmodelswithex- +Datasets. Experiments were conducted on three ternalreasoningknowledgeaboutuserpreferences +widelyusedreal-worlddatasets. AmazonMusic andfactualknowledgeaboutitemsextractedfrom +(Music) and Amazon Book (Book) are derived LLMs,whicharetransformedintotask-compatible +fromtheAmazonProductdataset1,whileYelpis representations. RLMRec(Renetal.,2024)lever- +fromtheYelpOpendataset2. Alldatasetsinclude agesLLMstolearnsemanticuseranditemrepre- +userreviews,ratings,andrichtextualinformation. sentationsfromtextualsignalsandalignsthemwith +WeusedthefullAmazonMusicdataset,butonly collaborativerelationalinformationthroughcross- +subsets (latest two months for Book and six for viewrepresentationlearning. PALR(Yangetal., +Yelp). Data was split by timestamp into training, 2023)fine-tunesalargelanguagemodelasarank- +validation,andtestsetstopreventinformationleak- ing component that selects preferred items from +age(Jietal.,2023). retrievedcandidatesexpressedinnaturallanguage. +EvaluationMetrics LG (LettinGo) (Wang et al., 2025) explores di- +For each observed user–item interaction (i.e., verseuserprofilecandidateswithLLMsandaligns +a review record), we construct the evaluation profilegenerationwithdownstreamrecommenda- +instance based on the user’s interaction history tionperformanceviapreferenceoptimization. Rea- +strictly prior to the corresponding timestamp. son4Rec(Fangetal.,2025)introducesadelibera- +tiverecommendationframeworkthatincorporates +Specifically,theuser’srecenthistoricalinteractions +beforethecurrentinteractionareusedtogenerate explicitstep-wisereasoningoveruserpreferences +auserprofile,whilehistoricalreviewsfromother toguideratingprediction. +| users are used | to generate | an item profile. | The | | | | | | +| -------------- | ----------- | ---------------- | --- | --- | --- | --- | --- | --- | +4.2 MainResults +downstreamrecommendationsystemthenpredicts +theratingofthetargetitemconditionedontwoin- Table 1 presents a comparison of our proposed +puts: thegenerateduserprofile,andthegenerated method against five baselines on three datasets: +| | | | | Amazon | Music, Amazon | Books, | and | Yelp. We | +| ------------ | ------------------------------- | --- | --- | ------ | ------------- | ------ | --- | -------- | +| itemprofile. | Anexceptionisthe10Hbaseline,for | | | | | | | | +which the downstream model directly consumes use Qwen3-8B (Team, 2025) and LLaMA3-8B +therawrecentinteractionhistories. (Dubey et al., 2024) as both the profile generator +andthedownstreamrecommendationmodel,with +| We evaluate | performance | using four | widely | | | | | | +| --------------- | ----------- | ------------------ | ------- | ------------ | ----------- | ----- | ----------- | ------- | +| | | | | a prediction | temperature | of 0. | In addition | to rat- | +| adopted metrics | (Wang | et al., 2025; Fang | et al., | | | | | | +2025): MeanAbsoluteError(MAE),RootMean ingpredictionresults,wealsoreportranking-based +SquareError(RMSE),Accuracy,andF1score. evaluation results in Table ??, which assess the +effectivenessofthelearnedprofilesfromadown- +Inadditiontoratingprediction,wefurtherevaluate +thegenerateduseranditemprofilesunderarank- streamrankingperspective. +| | | | | Overallsuperiorityoverstrongbaselines. | | | | As | +| --- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | +1https://cseweb.ucsd.edu/~jmcauley/datasets/ shown in Table 1, our method consistently out- +amazon/links.html +| | | | | performs | the strongest | baselines | across | all three | +| --- | --- | --- | --- | -------- | ------------- | --------- | ------ | --------- | +2https://business.yelp.com/data/resources/ +| open-dataset/ | | | | datasetsunderbothQwen3-8BandLLaMA3-8B. | | | | | +| ------------- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | + +Method Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) +(%) (%) (%) +Qwen3(8B) +10H 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +KAR(Xietal.,2024) 0.7396 1.2184 55.34 48.67 0.7483 1.1380 58.65 60.29 0.7098 1.0923 56.17 58.78 +RLMRec(Renetal.,2024) 0.8197 1.3312 47.15 42.46 0.7438 1.1069 54.89 57.65 0.7812 1.1584 52.86 55.93 +PALR(Yangetal.,2023) 0.7994 1.2876 48.53 43.19 0.6075 0.9531 57.35 56.77 0.7485 1.1187 54.24 56.38 +LG(Wangetal.,2025) 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +R4Rec(Fangetal.,2025) 0.7028 1.1523 55.69 47.73 0.5654 0.9635 58.69 54.67 0.6397 1.0098 58.47 56.84 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +LLaMA3(8B) +10H 1.0864 1.9532 22.09 27.30 0.7917 1.3346 38.13 46.87 0.8064 1.3866 37.15 45.27 +KAR(Xietal.,2024) 0.6427 1.1668 54.51 47.98 0.5726 0.9033 57.53 59.92 0.5892 0.9614 55.87 58.21 +RLMRec(Renetal.,2024) 0.7428 1.3572 46.74 42.11 0.6076 0.9886 53.78 57.42 0.6226 0.9477 52.12 55.79 +PALR(Yangetal.,2023) 0.7238 1.3265 47.72 43.29 0.5823 0.9222 56.73 59.31 0.5977 0.8855 55.06 57.62 +LG(Wangetal.,2025) 0.6196 1.1289 56.03 51.24 0.5204 0.9369 61.92 59.50 0.5543 0.7967 58.95 60.39 +R4Rec(Fangetal.,2025) 0.7586 1.0418 55.80 53.00 0.5442 0.7722 60.86 54.88 0.6029 0.8345 59.70 56.35 +Ours 0.5367 0.9687 60.87 54.74 0.4680 0.8277 63.30 60.60 0.5092 0.9500 63.42 58.12 +Table1: PerformanceonthreedatasetsusingQwen3(8B)andLLaMA3(8B). +Method Yelp AmazonMusic AmazonBooks +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +Table2: RankingperformanceunderEASE-based(Steck,2019)hardnegatives. +Under Qwen3-8B, our method achieves an accu- our approach reaches an NDCG@1 of 0.5619 +racyof61.23%onYelp,67.96%onAmazonMu- and an NDCG@10 of 0.7443, clearly outper- +sicand64.38%onAmazonBooks,surpassingLG forming the strongest baseline methods. Similar +(LettinGo)(Wangetal.,2025)by5.05%,5.59% trendsareobservedonAmazonMusic,whereour +and5.03%,respectively. Similarimprovementsare methodattains0.5347atNDCG@1and0.7331at +observed under LLaMA3-8B, indicating that the NDCG@10, as well as on Amazon Books with +gainsarestableacrossbackbonemodels. NDCG@1 of 0.4866 and NDCG@10 of 0.7107. +Theseresultsindicatethatthelearnedprofilesen- +Advantages over fixed-structure and delib- +able more accurate identification and ordering of +erative baselines. Compared with strong fixed- +relevantitemswithincandidatesets,leadingtosu- +structureordeliberativemethodssuchasKAR(Xi +perior ranking quality in downstream recommen- +etal.,2024)andReason4Rec(Fangetal.,2025), +dation. Wefurtherevaluaterankingperformance +our approach consistently achieves lower predic- +under a more challenging setting using EASE- +tion error and higher accuracy. For example, on +based (Steck, 2019) hard negatives, and observe +Amazon Books with Qwen3-8B, our method re- +consistentimprovements;detailedresultsarepro- +duces MAE to 0.4612, compared to 0.7098 for +videdinAppendixB.2. +KARand0.6397forReason4Rec,demonstrating +moreeffectiveabstractionofuserpreferenceswith- +4.3 AblationStudy +outrelyingonpredefinedreasoningtemplates. +Consistent improvements in downstream Effectivenessofprofilegeneration,cue&strategy, +ranking. Table??reportstherankingresultsun- anduser–itemjointoptimization. Table3reports +der the Qwen3-8B backbone. Our method con- the ablation results under different design con- +sistentlyachievesthebestperformanceacrossall figurations. Starting from the history-only base- +three datasets and evaluation cutoffs. On Yelp, line (10H), introducing explicit profile genera- + +| Method | | | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | --- | ---- | --- | ------ | ----------- | --- | --- | ----------- | ------ | +| | | | MAE | RMSE | Acc | F1 MAE | RMSE | Acc | F1 | MAE RMSE | Acc F1 | +10H(HistoryOnly) 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +Profile 0.7218 1.1863 55.48 48.09 0.6597 1.0218 58.67 57.48 0.6764 1.0469 57.14 57.68 +Profile+Cue&Strategy 0.7085 1.1654 55.83 48.54 0.5708 0.9897 58.91 55.53 0.6389 1.0108 58.43 56.88 +Profile+JointOpt.(LG(Wang 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +etal.,2025)) +Profile+Cue&Strategy+ 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +JointOpt.(Ours) +Table3: AblationstudyondifferentdesignconfigurationsinDUETusingQwen3(8B). +| Method | | Yelp | | | | AmazonMusic | | | | AmazonBooks | | +| ------ | -------- | ---- | --- | ----- | --- | ----------- | --- | ----- | --- | ----------- | --------- | +| | MAE RMSE | | Acc | F1(%) | MAE | RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | +| | | | (%) | | | | (%) | | | | (%) | +10H+30P 0.5126 0.9485 61.23 55.18 0.3883 0.7494 67.96 63.89 0.4612 0.9089 65.13 59.97 +10H+50P 0.4909 0.9207 62.43 56.24 0.3924 0.7543 67.88 63.87 0.4553 0.9023 64.62 59.52 +10H+70P 0.4987 0.9326 61.98 55.81 0.3937 0.7564 68.22 64.12 0.4608 0.9068 64.38 59.27 +Table4: Impactofhistoricalinteractionlengthonprofilequality(usingQwen3(8B)). +tionyieldssubstantialperformanceimprovements jointoptimizationarecombined. Underthisfull +across all datasets. For example, on Yelp, MAE configuration, accuracy reaches 61.23% on Yelp, +isreducedfrom1.1235to0.7218andaccuracyin- 67.96%onAmazonMusic,and64.38%onAma- +creases from 23.17% to 55.48%, confirming that zon Books, with corresponding MAE values of +textualprofilesprovidesignificantlymoreinforma- 0.5126, 0.3937, and 0.4612, respectively. Com- +tiverepresentationsthanrawinteractionhistories. pared with the LettinGo-style configuration, this +| | | | | | | setting | yields | consistent | | improvements | across all | +| ------ | ---------------- | --- | ----- | --- | ------ | --------- | ------ | -------------------------------------- | --- | ------------ | ---------- | +| Adding | the cue&strategy | | layer | on | top of | pro- | | | | | | +| | | | | | | datasets. | | Overall,theablationresultsindicatethat | | | | +filegenerationleadstomodestbutconsistentgains. +whileprofilegenerationandjointoptimization +OnAmazonMusic,MAEfurtherdecreasesfrom +0.6597to0.5708,whileaccuracyslightlyimproves contributesubstantiallytoperformancegains,inte- +gratingcue&strategydiscoveryfurtherenhances +| from58.67%to58.91%. | | | Althoughthenumerical | | | | | | | | | +| ------------------- | --- | --- | -------------------- | --- | --- | --- | ------------- | --- | -------- | --------- | ------------ | +| | | | | | | the | effectiveness | | of joint | user–item | optimization | +improvementsintroducedbystrategyalonearelim- +ited, these results suggest that strategy discovery byprovidingastructuredspaceforexplorationand +refinement. +primarilyreshapeshowpreferenceinformationis +abstractedandexpressed,ratherthandirectlyopti- Impactofhistoricalinteractionlengthonprofile +mizingpredictionaccuracyinisolation. +| | | | | | | quality. | | Table | 4 analyzes | the | impact of histori- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ----- | ---------- | --- | ------------------ | +Whenenablingjointoptimizationwithoutstrat- cal interaction length on profile quality. Across +thethreedatasets,varyingthenumberofhistorical +| egy (i.e., | the LettinGo | (Wang | et | al., 2025)-style | | | | | | | | +| ---------- | ------------ | ----- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | +configuration),performanceimprovesmorenotice- interactions leads to only moderate performance +ablyacrossdatasets. OnAmazonMusic,accuracy differences, indicating that our method does not +| | | | | | | strongly | | depend | on long | histories. | In particular, | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------ | ------- | ---------- | -------------- | +increasesto62.37%,comparedto58.67%withpro- +filegenerationalone. Inourimplementation,this usingamoderatehistorylength(e.g.,30–50inter- +settingcorrespondstoareproductionofLettinGo, actions)alreadyachievescompetitiveorbestper- +whereprofilegenerationandjointoptimizationare formanceonmostmetrics,whilefurtherincreasing +thehistorylengthprovideslimitedadditionalgains. +| appliedwithoutthestrategylayer. | | | | Whiletheorig- | | | | | | | | +| ------------------------------- | ------- | --------- | --- | ------------- | ---------- | --- | --- | --- | --- | --- | --- | +| inal method | focuses | primarily | on | user | profiling, | | | | | | | +OnYelpandAmazonBooks,extendingthehis- +weextendtheoptimizationtobothuseranditem +torylengthbeyondthisrangedoesnotconsistently +profilestoensureafaircomparison. improveaccuracyorF1scoreandmayslightlyde- +Thestrongestandmostconsistentperformance gradeperformance,suggestingthatexcessivehis- +is achieved when cue&strategy discovery and toricalinteractionscanintroducenoisyorlessrele- + +vantsignals. Incontrast,AmazonMusicexhibits maintaining comparable coverage. Compared to +relativelystableperformanceacrossdifferenthis- existing methods, which either rely on extractive +torylengths,indicatingthatuserpreferencesinthis compression(e.g.,KAR(Xietal.,2024))orgener- +domainarelesssensitivetohistorytruncation. atefree-formdescriptionswithweakergrounding +Necessity of RL-based Optimization. We com- (e.g.,RLMRec(Renetal.,2024)andPALR(Yang +pare the full model with a variant without RL etal.,2023)), DUET consistentlyattainsstronger +(DUET W/O RL) to isolate the effect of RL. As alignment without sacrificing coverage. For in- +showninTable5,removingRLleadstosubstantial stance,LG(Wangetal.,2025)achievesrelatively +performancedegradationacrossalldatasets(e.g., highalignmentbutexhibitslessstablegrounding +YelpAccuracydropsfrom61.23%to48.53%),indi- acrossdatasets,whileRLMRecattainshighercov- +catingthatthegainscannotbeattributedtoprompt erageatthecostofweakersemanticalignment. In +| designalone. | | | | | contrast,DUETmaintainsbothhighalignmentand | | | | | | +| ------------ | --- | --- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | +Without RL, the generator reduces to a static mid-to-highcoverage,suggestingamoreeffective +mapping from interaction history to textual pro- balancebetweensemanticabstractionandevidence +| files,lackingadaptiveselectionofrelevantsignals. | | | | | preservation. | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | +Incontrast,RLenablesoptimizingprofileconstruc- +4.5 CaseStudy +tionunderrewardfeedback,resultinginmoredis- +Figure4presentsarepresentativecasestudythat +| criminative | representations.These | | results | demon- | | | | | | | +| ----------- | --------------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | +strate that RL-based optimization is essential for demonstrates how semantically aligned user and +| DUET | . | | | | itemprofiles | enableaccurate | | prediction | | thattran- | +| ---- | --- | --- | --- | --- | ------------ | -------------- | --- | ---------- | --- | --------- | +scendsthelimitationsofsparserawinteractionhis- +4.4 SemanticAnalysisofGeneratedProfiles tories. In the user profile, initially scattered and +To better understand the source of performance fragmentedpreferencecuesaresystematicallydis- +gains, we analyze whether the generated profiles tilled into a stable and coherent preference struc- +exhibitmeaningfulsemanticpropertiesratherthan turefavoringfunk,soul,andprogressiverock(or- +servingasintermediatetextualartifacts. Weintro- ange), coupled with a pronounced emphasis on +duce two complementary metrics to characterize musicalcomplexityandhistoricalsignificance(pur- +semanticcompatibilityandgrounding. ple). Thecorrespondingitemprofileexhibitsstrik- +Semantic Alignment. We measure the ingsymmetry,characterizingthealbumasanexem- +embedding-levelsimilaritybetweengenerateduser plaryfunk-rockcomposition(orange)andposition- +anditemprofilesusingall-mpnet-base-v2from ingitasadefiningandculturallyinfluentialrelease +Sentence-Transformers (Reimers and Gurevych, of the 1970s (purple). This example shows 519 +2019). Foreachuser–itempair,wecomputecosine that the learned profiles capture meaningful pref- +| similarity: | | | | | erence–attributecorrespondencethatisdifficultto | | | | | | +| -------------------------------- | -------------------------------- | --- | ------------ | --- | ----------------------------------------------- | --------- | ------- | ---------- | ---- | ----------- | +| | | | e ·e | | recoverfromindividualreviewsalone. | | | | | | +| | | | u i | | | | | | | | +| | Align(u,i) | = | | (3) | | | | | | | +| | | | ∥e ∥∥e ∥ | | | | | | | | +| | | | u i | | 5 Conclusion | | | | | | +| wheree | ande denotetheembeddingvectorsof | | | | | | | | | | +| | u i | | | | In this | paper, we | propose | DUET | , a | closed-loop | +| thegenerateduseranditemprofiles. | | | Highervalues | | | | | | | | +| | | | | | framework | for | jointly | generating | user | and item | +indicatestrongersemanticcompatibilitybetween +| | | | | | textualprofilesforrecommendation. | | | | | Unlikeprior | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | +modeleduserpreferencesanditemcharacteristics. +| | | | | | methods | that rely | on | fixed templates | | or indepen- | +| -------- | --------------- | --- | ---------- | ------ | ------- | ----------- | --------- | --------------- | --- | -------------- | +| Coverage | (Faithfulness). | | We measure | token- | | | | | | | +| | | | | | dently | constructed | profiles, | DUET | | treats profile | +levelgroundingas: +| | | | | | generation | as an | exploration | problem | | and aligns | +| --- | --- | --- | --- | --- | ---------- | ----- | ----------- | ------- | --- | ---------- | +|Tokens(profile)∩Tokens(history)| representations directly with downstream recom- +Cov = +| | | |Tokens(profile)| | | | mendationperformance. | | | | | | +| --- | --- | ----------------- | --- | --- | --------------------- | --- | ---- | ---------- | --------- | ---- | +| | | | | (4) | Specifically, | | DUET | integrates | cue-based | ini- | +Thismetricquantifieshowmuchofthegenerated tialization, adaptive strategy construction, and +profileissupportedbyhistoricaltextualevidence. feedback-drivenjointoptimizationtoproduceflex- +We report coverage separately for user and item ible yet task-aligned user–item profiles. By opti- +| profiles. | | | | | mizing | both profiles | in | a shared | semantic | space, | +| --------- | --- | --- | --- | --- | ------ | ------------- | --- | -------- | -------- | ------ | +AsshowninTable6,DUETachievesthehigh- theframeworkreducessemanticmismatchandcap- +est semantic alignment across all datasets while turesinteraction-relevantsignalsthataredifficultto + +Setting Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1 MAE RMSE Acc F1 MAE RMSE Acc F1 +(%) (%) (%) (%) (%) (%) +DUETw/oRL 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +DUET(full) 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table5: EffectofRL-basedoptimization(usingQwen3(8B)). +User History: +Andre' S Grindle "Tour of U.S.A.", Rating: 5.0. If your looking for heavy metals dim orgins look no further. +Andre' S Grindle "Misfits", Rating:4.0. Actually the opening title track is more of a direct pop/rock type song. But the rock n roll +returns in a big way on "Hay Fever". +............ +Item History: +Worlds Away, Rating: 4.0, Review: The CD is wonderful, and is a seminal piece of music for both the group and for those of us who +were 'of age' in the '70s and '80s. Pablo Cruise was - and is - a group that makes beautiful music like few others, and this +shipper/seller made it all the more special with a fast ship! THANKS!!! +Worlds Away, Rating: 5.0, I love this CD! I'd recommend the Pablo Cruise Millenium Collection to get to know this band. Worlds Away to +me represents the best of breezy '70s pop. Five stars! +............ +USER PROFILE: +Andre Grindle is a dedicated music enthusiast with a deep appreciation for funk, soul, and progressive rock, particularly drawn to artists +who blend technical mastery with genre-defining innovation. His reviews consistently highlight albums that showcase musical complexity, +lyrical depth, and historical significance, often referencing artists like Stevie Wonder, Prince, and Rush. He values authenticity and +artistic vision, frequently praising albums that resist trend-chasing and prioritize creative integrity. His detailed, introspective reviews +suggest a preference for nuanced analysis over superficial praise, with a focus on emotional resonance, production quality, and the +artist's evolution. While he is generous with high ratings (often 5.0), he is also willing to critique perceived shortcomings, as seen in his +4.0 for "Test for Echo" and 3.0 for "The Color of Success." His engagement is characterized by a blend of personal anecdotes, technical +appreciation, and historical context, indicating a preference for music that transcends mere entertainment. +Item PROFILE: +"Worlds Away" by Pablo Cruise is celebrated as a peak-era funk-rock album that exemplifies the band's ability to merge polished +production with genre-blurring creativity. Users consistently praise its tight musicianship, genre-defying sound, and the band's ability to +craft catchy, emotionally resonant tracks. The album's appeal lies in its seamless fusion of rock, funk, and pop, with standout tracks like +"Love Will Find a Way" and "Worlds Away" receiving particular acclaim for their infectious melodies and lyrical depth. While some users +note its dated production compared to modern standards, the majority highlight its historical significance as a defining work of the +1970s rock-funk movement. The album resonates most with fans of 1970s progressive rock, genre-defying artists, and listeners seeking +well-crafted, emotionally engaging music. Its high average rating reflects widespread admiration for its artistic ambition and enduring +appeal, though a minority acknowledges its limitations in terms of modern production quality. +Figure4: Illustrationofthemutualcorrespondencebetweenuseranditem. Thehighlightedregionsdemonstrate +thatuserpreferencessummarizedintheuserprofilealignwiththekeyattributesextractedintheitemprofile,which +providescomplementaryinformationbeyondrawhistoriesandthusimprovespredictionaccuracy. +Method Yelp AmazonMusic AmazonBooks +Align UserCov. ItemCov. Align UserCov. ItemCov. Align UserCov. ItemCov. +10H 0.3902 /(1.00) /(1.00) 0.3577 /(1.00) /(1.00) 0.2604 /(1.00) /(1.00) +KAR(Xietal.,2024) 0.4932 0.1745 0.1823 0.4807 0.2170 0.2391 0.5702 0.1877 0.1544 +RLMRec(Renetal.,2024) 0.4010 0.2646 0.3773 0.3931 0.3526 0.4889 0.4436 0.2994 0.3506 +PALR(Yangetal.,2023) 0.4715 0.3675 0.1917 0.4938 0.2197 0.2390 0.5216 0.2050 0.2267 +LG(Wangetal.,2025) 0.5709 0.2378 0.2682 0.5109 0.2546 0.3749 0.5506 0.3883 0.3364 +R4Rec(Fangetal.,2025) 0.4882 0.2348 0.2330 0.4208 0.3328 0.3595 0.4946 0.2831 0.2949 +Ours 0.6382 0.2880 0.3429 0.5947 0.4002 0.4482 0.7287 0.3457 0.3127 +Table6: Semanticalignmentandcoverageofgeneratedprofiles. Cov. denotesthefractionoftokensgroundedin +historicaltext. +recoverfromrawhistoriesorstaticpromptsalone. reinforcementlearning–basedoptimization. These +resultssuggestthatadaptive,interaction-awaretex- +Experiments on multiple real-world datasets tualprofilesprovideapromisingdirectionformore +demonstratethatDUETconsistentlyoutperforms effective interpretable and performance-oriented +strongbaselinesunderdifferentbackbonemodels, recommendationsystems. +validating the effectiveness of joint profiling and + +| Limitations | | | | | | | References | | | | | | | +| ----------- | --- | --- | --- | --- | --- | --- | ---------- | ----- | ------ | ---- | ------ | ------ | ----- | +| | | | | | | | Keqin Bao, | Jizhi | Zhang, | Yang | Zhang, | Wenjie | Wang, | +Despiteitseffectiveness,ourapproachhasseveral FuliFeng,andXiangnanHe.2023. Tallrec: Anef- +fectiveandefficienttuningframeworktoalignlarge +| limitations. | First, | the | proposed | framework | | relies | | | | | | | | +| ------------ | ------ | --- | -------- | --------- | --- | ------ | -------------------------------- | --- | --- | --- | --- | ---------- | --- | +| | | | | | | | languagemodelwithrecommendation. | | | | | InProceed- | | +onlargelanguagemodelsforbothprofilegenera- ingsofthe17thACMConferenceonRecommender +tion and downstream recommendation, which in- Systems,pages1007–1014. +| troduces | additional | computational | | | overhead | dur- | | | | | | | | +| -------- | ---------- | ------------- | --- | --- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | +TingChen,Wei-LiHan,Hai-DongWang,Yi-XunZhou, +| ingtrainingandinference. | | | Whileourexperiments | | | | | | | | | | | +| ------------------------ | --- | --- | ------------------- | --- | --- | --- | ------------------------- | --- | --- | --- | ---------------- | --- | --- | +| | | | | | | | BinXu,andBin-YuZang.2007. | | | | Contentrecommen- | | | +showconsistentgainsacrossdifferentbackbones, dationsystembasedonprivatedynamicuserprofile. +the overall efficiency may be constrained when In2007Internationalconferenceonmachinelearn- +| | | | | | | | ing and | cybernetics, | | volume | 4, pages | 2112–2118. | | +| --------------------------------- | --- | --- | --- | --- | ---------- | --- | ------- | ------------ | --- | ------ | -------- | ---------- | --- | +| scalingtoverylargeuseroritemsets. | | | | | Second,the | | | | | | | | | +IEEE. +| quality | of the generated | | profiles | is | inherently | de- | | | | | | | | +| ------- | ----------------- | --- | -------- | --- | ------------- | --- | --------- | ------- | --- | ----------- | ----- | --- | ----- | +| pendent | on the underlying | | LLMs | | and prompting | | | | | | | | | +| | | | | | | | Yue Chen, | Minghua | | He, Fangkai | Yang, | Pu | Zhao, | +strategies. Variationsinmodelcapacityorprompt LuWang,YuKang,YifeiDong,YuefengZhan,Hao +Sun,QingweiLin,SaravanRajmohan,andDongmei +sensitivitymayleadtodifferencesinprofilestabil- +| | | | | | | | Zhang. | 2025. | Warriormath: | | Enhancing | the | mathe- | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | ------------ | --- | --------- | --- | ------ | +ity,whichwedonotexplicitlycontrolinthecurrent +maticalabilityoflargelanguagemodelswithadefect- +design. Finally,ourevaluationfocusesontext-rich +| | | | | | | | awareframework. | | | Preprint,arXiv:2508.01245. | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | -------------------------- | --- | --- | --- | +recommendationscenarioswheresufficienthistor- +| | | | | | | | Paul Covington, | | Jay | Adams, and | Emre | Sargin. | 2016. | +| ------------ | --- | ---------- | --- | ----------------- | --- | --- | --------------- | --- | --- | ---------- | ---- | ------- | ----- | +| ical reviews | are | available. | | The effectiveness | | of | | | | | | | | +Deepneuralnetworksforyoutuberecommendations. +| the proposed | strategy | | in domains | with | extremely | | | | | | | | | +| ------------ | -------- | --- | ---------- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +InProceedingsofthe10thACMconferenceonrec- +sparsetextualsignalsorinnon-textualmodalities +ommendersystems,pages191–198. +remainstobefurtherexplored. +DeepSeek-AI,DayaGuo,DejianYang,HaoweiZhang, +JunxiaoSong,RuoyuZhang,RunxinXu,QihaoZhu, +ShirongMa,PeiyiWang,XiaoBi,XiaokangZhang, +Data,Privacy,andEthicsConsiderations Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhi- +| | | | | | | | hong | Shao, | Zhuoshu | Li, Ziyi | Gao, and | 181 | others. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | ------- | ------------- | --------- | --- | ------- | +| | | | | | | | 2025. | Deepseek-r1: | | Incentivizing | reasoning | | capa- | +Alldatasetsusedinthisworkarepubliclyreleased bilityinllmsviareinforcementlearning. Preprint, +arXiv:2501.12948. +bytheiroriginalowners,andwefollowtheofficial +termsofuseassociatedwitheachdataset,utilizing +JiaxinDeng,ShiyaoWang,KuoCai,LejianRen,Qigen +thedatasolelyforresearchpurposesaspermitted +| | | | | | | | Hu, Weifeng | | Ding, | Qiang Luo, | and | Guorui | Zhou. | +| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ----- | ---------- | --- | ------ | ----- | +bytheirrespectivelicenses;forreproducibility,the 2025. Onerec: Unifyingretrieveandrankwithgen- +officialdatasetsourcesarecitedinthecorrespond- erativerecommenderanditerativepreferencealign- +ment. Preprint,arXiv:2502.18965. +ingsectionsofthispaper,andwedonotredistribute +anyrawuser-generatedcontentsuchasreviewtext AbhimanyuDubey,AbhinavJauhri,AbhinavPandey, +| aspartofthiswork. | | Thisresearchdoesnotattempt | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +AbhishekKadian,AhmadAl-Dahle,AieshaLetman, +to identify or infer the identity of any individual Akhil Mathur, Alan Schelten, Amy Yang, Angela +| | | | | | | | Fan,and1others.2024. | | | Thellama3herdofmodels. | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | ---------------------- | --- | --- | --- | +user,andpriortotraininganddownstreamprofile +arXivpreprintarXiv:2407.21783. +| generation, | we | apply text | filtering | | procedures | to | | | | | | | | +| ----------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +remove direct personal identifiers where applica- YiFang,WenjieWang,YangZhang,FengbinZhu,Qi- +ble;themodelsaretrainedsolelyonpubliclyavail- fanWang,FuliFeng,andXiangnanHe.2025. Rea- +son4rec:Largelanguagemodelsforrecommendation +abledata,andnoadditionalprivateorproprietary +| | | | | | | | withdeliberativeuserpreferencealignment. | | | | | Preprint, | | +| ---------------- | --- | -------------- | --- | ------------ | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --------- | --- | +| user information | | is introduced. | | Furthermore, | | we | | | | | | | | +arXiv:2502.02061. +| do not release | generated | | user | profiles | or | any per- | | | | | | | | +| -------------- | --------- | --- | ---- | -------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +MinghuaHe,YueChen,FangkaiYang,PuZhao,Wenjie +useroutputsproducedduringthemodelingprocess, +Yin,YuKang,QingweiLin,SaravanRajmohan,and +andallreportedresultsarepresentedinaggregate +| | | | | | | | DongmeiZhang.2025. | | | Execoder:Empoweringlarge | | | | +| -------- | ---------- | -------- | --- | ----- | ---- | ------- | ------------------ | --- | --- | ------------------------ | --- | --- | --- | +| form for | evaluation | purposes | | only. | From | a secu- | | | | | | | | +languagemodelswithexecutabilityrepresentationfor +rityperspective,theassociatedcoderepositoryhas codetranslation. Preprint,arXiv:2501.18460. +beenscannedforpotentialsecrets,andanyinternal +MinjieHong,ZirunGuo,YanXia,ZehanWang,Ziang +endpointsorconfidentialconfigurationinformation +| | | | | | | | Zhang, | Tao | Jin, and | Zhou Zhao. | 2025a. | Apo: | En- | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | ---------- | ------ | ---- | --- | +havebeenremovedpriortorelease. +hancingreasoningabilityofmllmsviaasymmetric +| | | | | | | | policyoptimization. | | | Preprint,arXiv:2506.21655. | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | -------------------------- | --- | --- | --- | + +Minjie Hong, Yan Xia, Zehan Wang, Jieming Zhu, JoarSkalse,NikolausH.R.Howe,DmitriiKrashenin- +Ye Wang, Sihang Cai, Xiaoda Yang, Quanyu Dai, nikov, and David Krueger. 2025. Defining +ZhenhuaDong,ZhimengZhang,and1others.2025b. and characterizing reward hacking. Preprint, +| Eager-llm: | Enhancinglargelanguagemodelsasrec- | | | | | arXiv:2209.13085. | | | | | | +| ---------- | ---------------------------------- | --------- | ----------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| ommenders | through | exogenous | behavior-semantic | | | | | | | | | +integration. InProceedingsoftheACMonWebCon- Harald Steck. 2019. Embarrassingly shallow autoen- +ference2025,pages2754–2762. codersforsparsedata. InProceedingsofTheWorld +WideWebConference(WWW),pages3251–3257. +MinjieHong,ZetongZhou,ZirunGuo,ZiangZhang, +| Ruofan | Hu, Weinan | Gan, | Jieming | Zhu, and | Zhou | | | | | | | +| ------ | ---------- | ---- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | +ChaoSun,YaoboLiang,YamingYang,ShilinXu,Tian- +| Zhao.2025c. | Generativereasoningrecommendation | | | | | | | | | | | +| ----------- | --------------------------------- | --- | --- | --- | --- | ---------------------------- | -------- | ---- | ------ | -------- | ------------- | +| | | | | | | mengYang,andYunhaiTong.2024. | | | | | Rlrf4rec: Re- | +| viallms. | Preprint,arXiv:2510.20815. | | | | | | | | | | | +| | | | | | | inforcement | learning | from | recsys | feedback | for en- | +JihwanJeong,YinlamChow,GuyTennenholtz,Chih- hancedrecommendationreranking. arXivpreprint +| Wei Hsu, | Azamat | Tulepbergenov, | | Mohammad | | arXiv:2410.05939. | | | | | | +| ------------------------------------ | ------------ | --------------- | ---------- | -------- | --------- | ----------------- | --- | --------------------- | --- | --- | --------- | +| Ghavamzadeh, | and | Craig | Boutilier. | 2023. | Fac- | | | | | | | +| | | | | | | QwenTeam.2025. | | Qwen3technicalreport. | | | Preprint, | +| tual and | personalized | recommendations | | using | lan- | | | | | | | +| guagemodelsandreinforcementlearning. | | | | | Preprint, | arXiv:2505.09388. | | | | | | +arXiv:2310.06176. +| | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | +| --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | +Yitong Ji, Aixin Sun, Jie Zhang, and Chenliang Li. bert, Amjad Almahairi, Yasmine Babaei, Nikolay +| 2023. A | critical | study | on data leakage | in | recom- | | | | | | | +| ------- | -------- | ----- | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti +| mendersystemofflineevaluation. | | | | ACMTrans.Inf. | | | | | | | | +| ------------------------------ | --- | --- | --- | ------------- | --- | ------------------------ | --- | --- | ------- | --- | --------- | +| | | | | | | Bhosale,and1others.2023. | | | Llama2: | | Openfoun- | +Syst.,41(3):75:1–75:27. +| | | | | | | dation and | fine-tuned | chat | models. | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | ---------- | ---------- | ---- | ------- | ----- | -------- | +arXiv:2307.09288. +| JiachengLin,TianWang,andKunQian.2025. | | | | | Rec-r1: | | | | | | | +| ------------------------------------- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +Bridginggenerativelargelanguagemodelsanduser- +| | | | | | | LeandrovonWerra, | | YounesBelkada, | | LewisTunstall, | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | -------------- | --- | -------------- | --- | +centricrecommendationsystemsviareinforcement +EdwardBeeching,TristanThrush,NathanLambert, +| learning. | Preprint,arXiv:2503.24289. | | | | | | | | | | | +| --------- | -------------------------- | --- | --- | --- | --- | -------- | ------ | ----------- | ------------- | --- | ------------ | +| | | | | | | Shengyi | Huang, | Kashif | Rasul, | and | Quentin Gal- | +| | | | | | | louédec. | TRL: | Transformer | Reinforcement | | Learn- | +AiweiLiu,MinghuaHe,ShaoxunZeng,SijunZhang, +| LinhaoZhang,ChuhanWu,WeiJia,YuanLiu,Xiao | | | | | | ing. | | | | | | +| ---------------------------------------- | --- | --- | ----------------------- | --- | --- | ---- | --- | --- | --- | --- | --- | +| Zhou,andJieZhou.2025. | | | Wedlm:Reconcilingdiffu- | | | | | | | | | +sionlanguagemodelswithstandardcausalattention LuWang,DiZhang,FangkaiYang,PuZhao,Jianfeng +forfastinference. Preprint,arXiv:2512.22737. Liu,YuefengZhan,HaoSun,QingweiLin,Weiwei +| | | | | | | Deng, Dongmei | | Zhang, | Feng | Sun, and | Qi Zhang. | +| --- | --- | --- | --- | --- | --- | ------------- | --- | ------ | ---- | -------- | --------- | +HongtaoLiu,FangzhaoWu,WenjunWang,Xianchen +| | | | | | | 2025. Lettingo: | | Exploreuserprofilegenerationfor | | | | +| --- | --- | --- | --- | --- | --- | --------------- | --- | ------------------------------- | --- | --- | --- | +Wang,PengfeiJiao,ChuhanWu,andXingXie.2019. +| | | | | | | recommendationsystem. | | | InProceedingsofthe31st | | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | +Nrpa: Neuralrecommendationwithpersonalizedat- ACMSIGKDDConferenceonKnowledgeDiscovery +tention. In Proceedings of the 42nd International andDataMiningV.2,KDD’25,NewYork,NY,USA. +ACMSIGIRConferenceonResearchandDevelop- AssociationforComputingMachinery. +mentinInformationRetrieval,pages1233–1236. +YeWang,JiahaoXun,MinjieHong,JiemingZhu,Tao +| Wensheng | Lu, Jianxun | Lian, | Wei Zhang, | Guanghua | | | | | | | | +| -------- | ----------- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | +Jin,WangLin,HaoyuanLi,LinjunLi,YanXia,Zhou +Li,MingyangZhou,HaoLiao,andXingXie.2024. +| | | | | | | Zhao,and1others.2024. | | | Eager: | Two-streamgen- | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ------ | -------------- | --- | +Aligninglargelanguagemodelsforcontrollablerec- +arXivpreprintarXiv:2403.05063. erativerecommenderwithbehavior-semanticcollab- +ommendations. +| | | | | | | oration. | InProceedingsofthe30thACMSIGKDD | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------- | --- | --- | --- | --- | +RadaMihalceaandPaulTarau.2004. Textrank: Bring- ConferenceonKnowledgeDiscoveryandDataMin- +| | | InEMNLP. | | | | ing,pages3245–3254. | | | | | | +| --- | --- | -------- | --- | --- | --- | ------------------- | --- | --- | --- | --- | --- | +ingorderintotext. +| Zhaopeng | Qiu, Xian | Wu, | Jingyue Gao, | and | Wei Fan. | | | | | | | +| -------- | --------- | --- | ------------ | --- | -------- | -------- | --------------- | --- | -------- | -------- | --- | +| | | | | | | Triyanna | Widiyaningtyas, | | Indriana | Hidayah, | and | +2021. U-BERT:pre-traininguserrepresentationsfor +| | | | | | | TeguhBAdji.2021. | | Userprofilecorrelation-based | | | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | +improved recommendation. In AAAI, pages 4320– similarity(upcsim)algorithminmovierecommenda- +| 4327.AAAIPress. | | | | | | tionsystem. | JournalofBigData,8(1):52. | | | | | +| --------------------------------- | --- | --- | --- | -------------- | --- | ----------- | ----------------------------------- | --- | --- | --- | --- | +| NilsReimersandIrynaGurevych.2019. | | | | Sentence-bert: | | | | | | | | +| | | | | | | J.Wu.2023. | Computationalunderstandingofuserin- | | | | | +Sentenceembeddingsusingsiamesebert-networks. +terfaces. +InProceedingsofthe2019ConferenceonEmpirical +MethodsinNaturalLanguageProcessing(EMNLP), +pages3982–3992. Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, +| | | | | | | Hong Zhu, | Jieming | Zhu, | Bo Chen, | Ruiming | Tang, | +| --- | --- | --- | --- | --- | --- | --------- | ------- | ---- | -------- | ------- | ----- | +Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi WeinanZhang,andYongYu.2024. Towardsopen- +Cheng,JunfengWang,DaweiYin,andChaoHuang. world recommendation with knowledge augmenta- +2024. Representationlearningwithlargelanguage tionfromlargelanguagemodels. InProceedingsof +modelsforrecommendation. InProceedingsofthe the18thACMConferenceonRecommenderSystems, +| ACMonWebConference2024,pages3464–3475. | | | | | | pages12–22. | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | + +A ExperimentSetup +FanYang,ZhengChen,ZiyanJiang,EunahCho,Xiao- +| jiangHuang,andYanbinLu.2023. | | Palr:Personaliza- | | | | | | | | +| ------------------------------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| tionawarellmsforrecommendation. | | arXivpreprint | A.1 DataCuration | | | | | | | +arXiv:2305.07622. +Jiarui Zhang. 2024. Guided profile generation im- Table7: Statisticaldetailsoftheevaluationdatasets. +| proves personalization | with llms. | arXiv preprint | | | | | | | | +| ---------------------- | ---------- | -------------- | ---------- | ------ | ----------- | --- | ------------ | ------ | ------ | +| arXiv:2409.13093. | | | Dataset | #Train | #Valid | | #Test #User | | #Item | +| | | | Music | 43,071 | 3,271 | | 1,296 | 4,183 | 2,660 | +| | | | Book | 71,972 | 6,144 | | 5,541 13,863 | | 13,515 | +| | | | Yelp | 51,497 | 4,757 | | 4,328 | 8,453 | 13,426 | +| | | | We conduct | | experiments | | on three | widely | used | +real-worlddatasets: +| | | | • AmazonMusic(Music): | | | | Thisreferstothe“Dig- | | | +| --- | --- | --- | --------------------- | --- | ------ | ------ | -------------------- | --- | ------ | +| | | | ital Music” | | subset | of the | well-known | | Amazon | +Productdataset3,whichrecordsrichuserreviews, +ratings,andtextualinformationaboutitems,such +| | | | as titles, | across | a | broad | range | of product | cate- | +| --- | --- | --- | ---------- | ------ | --- | ----- | ----- | ---------- | ----- | +gories,ontheAmazonplatform.2 +| | | | • AmazonBook(Book): | | | Thisreferstothe“Book” | | | | +| --- | --- | --- | ------------------- | --- | --- | --------------------- | --- | --- | --- | +subsetoftheAmazonProductdataset. +dataset4, +| | | | • Yelp: | This | refers | to the | Yelp | Open | | +| --- | --- | --- | ------- | -------- | -------------- | -------- | ------- | ------------- | --------- | +| | | | which | includes | user | reviews, | ratings | | for busi- | +| | | | nesses | such | as restaurants | | and | retail shops, | as | +wellastextualinformationaboutthebusinesses. +Itiswidelyusedinrecommendationtasks(Qiu +etal.,2021). +WeusetheentireMusicdatasetforexperiments, +| | | | while for | the Book | | and Yelp | datasets, | we | utilize | +| --- | --- | --- | ------------------------------- | -------- | --- | -------- | --------- | ---------- | ------- | +| | | | onlyasubsetduetotheirlargesize. | | | | | FortheBook | | +dataset,weusedatafromthelasttwomonths,and +fortheYelpdataset,weusedatafromthelastsix +months. Foreachdataset,wesplititintotraining, +| | | | validation, | and | test | sets based | on | the timestamps | | +| --- | --- | --- | ---------------- | --- | -------- | ---------- | ---- | -------------- | --- | +| | | | of interactions, | | ensuring | that | test | interactions | oc- | +curafteralltrainingandvalidationinteractionsto +preventinformationleakage(Jietal.,2023). +| | | | Regarding | | data | filtering, | following | | prior | +| --- | --- | --- | --------- | --- | ----------- | ---------- | --------- | -------- | ----- | +| | | | work (Liu | et | al., 2019), | we | adopt | a 5-core | set- | +tingtofilterthedataandexcludecold-startusers +| | | | and items—those | | not | appearing | | in the | training | +| --- | --- | --- | --------------- | --- | ---------- | --------- | ---- | ------ | -------- | +| | | | set—from | the | validation | and | test | sets. | The sta- | +tisticaldetailsoftheprocesseddatasetareprovided +inTable7. +3https://cseweb.ucsd.edu/~jmcauley/datasets/ +amazon/links.html. +4https://business.yelp.com/data/resources/ +open-dataset/. + +| A.1.1 | ImplementationDetails | | | | | | RLMRecPrompt | | | | +| ----- | --------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | +Inourexperiments,weprimarilyemployQwen3- Role: Businessrecommendationassistant +Task: Determinebusinesstypesauserislikely +| 8B(Team,2025)andLLaMA38BInstruct | | | | | | (Dubey | | | | | +| ------------------------------------------ | --- | ----- | --- | --- | ------- | ---------- | ------------ | --- | --- | --- | +| etal.,2024;Touvronetal.,2023)asboththerec- | | | | | | | toenjoy | | | | +| ommendation | | model | and | the | profile | generation | InputFormat: | | | | +•Title: Businessname +model. Thetrainingprocessisimplementedusing +theTRL(vonWerraetal.). Keyhyperparameters, •Categories: Businesscategories +suchasbatchsizeandlearningrate,aredetermined •Sentiment: Usersentimenttowardbusiness +throughgridsearchtoachieveoptimalperformance. +OutputRequirements: +Moredetailscanbefoundinourcode. +1. JSONformatonly +2. Structure: +A.2 BaselinePrompts +{ +| | | | | | | | "summarization": | "Types | of businesses | user | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------ | ------------- | ---- | +KARPrompt +likelyenjoys"(≤100words), +Task: Analyzeuserpreferencesbasedonbusi- +| | | | | | | | "reasoning": | "Brief explanation | | for summa- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------ | --- | ---------- | +nessreviewinghistory +rization"(nowordlimit) +Input: {user_history}-User’sbusinessreview- +} +inghistorywithsentimentsovertime +3. NoadditionaltextoutsideJSON +Instructions: +| 1. | Analyze | the | user’s | preferences | considering | | | | | | +| --- | ------- | --- | ------ | ----------- | ----------- | --- | ------ | ----------- | --- | ------ | +| | | | | | | | Input: | INTERACTION | | ITEMS: | +businessnamesandcategories +{user_history} +| 2. | Take | into account | | sentiment | patterns | over | | | | | +| --- | ---- | ------------ | --- | --------- | -------- | ---- | --- | --- | --- | --- | +time +| 3. | Provideclearexplanationsbasedonreview- | | | | | | LGPrompt | | | | +| --- | -------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | +inghistorydetails +Youwillserveasanassistanttohelpmegener- +| 4. | Consider | other | pertinent | | factors | that may | | | | | +| --- | -------- | ----- | --------- | --- | ------- | -------- | --- | --- | --- | --- | +ateauserprofilebasedonthisuser’ssentiments +influencepreferences +historytobetterunderstandthisusers’interest +andthuspredicthis/hersentimentaboutatarget +| PALRPrompt | | | | | | | item. Iwillprovideyouwithsomebehaviorhis- | | | | +| ---------- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --------------- | --- | +| | | | | | | | toryoftheuserinthisformat: | | [itemattributes | | +Task: Summarizeuserpreferencesusingkey- +| | | | | | | | andsentiment]. | Theuserprofileyougenerate | | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ------------------------- | --- | --- | +words. +shouldcontainasmuchusefulcontentaspossi- +| Input: | | {user_history} | | - historical | businesses | | | | | | +| ------ | --- | -------------- | --- | ------------ | ---------- | --- | --- | --- | --- | --- | +bletohelppredicttheuser’ssentimenttowards +withusersentiments. +anewbusiness. +| Output | | Format: | An | itemized | list ranked | by | | | | | +| ------ | --- | ------- | --- | -------- | ----------- | --- | --- | --- | --- | --- | +USERHISTORY:{user_history}. +importance. +PROFILEYOUGENERATE: +Template: +| | • KEY_WORD_1: | | | | | "HIS- | | | | | +| --- | ------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +R4RecPrompt(Reasoner) +| | TORY_BUSINESS_1", | | | | | "HIS- | | | | | +| --- | ----------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +UserReviewHistory +TORY_BUSINESS_2" +⟨H u organizedasbelow⟩ +| | • KEY_WORD_2: | | | | | "HIS- | 1. TitleofItem1 | | | | +| ------------- | ---------------- | --- | --- | --- | --- | ----- | ---------------- | ----------------------- | ----------- | --- | +| | TORY_BUSINESS_3" | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| | | | | | | | NegativeAspects: | [Aspect1],[Aspect2],... | | | +| Instructions: | | | | | | | User Preference | Elements: | [Preference | 1], | +[Preference2],... +| | 1. Extract | key | preference | | indicators | from | | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | ---- | --- | --- | --- | --- | +2. TitleofItem2 +userinteractionhistory. +| | | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------------- | --- | --- | +2. Rankkeywordsbyimportance. NegativeAspects: [Aspect1],[Aspect2],... +| | | | | | | | User Preference | Elements: | [Preference | 1], | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --------- | ----------- | --- | + +| [Preference2],... | | | | | | | cantlyworsethanDUETacrossalldatasets. | | | | | This | +| ----------------- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | ---- | +indicatesthatcoherentsummariesaloneareinsuf- +... +| | | | | | | | ficient, | and that | the gains | of DUET | | arise from | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --------- | ------- | --- | ---------- | +ItemReviewHistorybyOtherUsers +⟨H organizedinthesameformatasabove⟩ learnedsemanticabstractionratherthanextractive +i +compression. +| Task: | Analyze | whether | | the user | will | like the | | | | | | | +| --------- | ------- | ------- | --- | ---------- | ----------- | -------- | --- | --- | --- | --- | --- | --- | +| new Music | | i based | on | the user’s | preferences | | | | | | | | +B.2 RankingunderHardNegativeSampling +| andtheitem’sfeatures. | | | | Provideyourrationale | | | | | | | | | +| --------------------- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Toconstructamorechallengingrankingscenario, +inoneconciseparagraph. +wereplacerandomnegativeswithhardnegatives +| | | | | | | | generatedbyacollaborativefilteringmodel. | | | | | Specif- | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ------- | +A.3 ExampleofCue +ically,foreachuser,weretrievehigh-scoringitems +Thefollowingexamplesillustratehowrawsignals +| | | | | | | | from an | EASE(Steck, | 2019) | model | that | the user | +| --------------------- | --- | --- | ---------------- | --- | --- | ---- | ------- | ----------- | ----- | ----- | ---- | -------- | +| aredistilledintocues. | | | Asshownonthetop, | | | user | | | | | | | +hasnotinteractedwith,andcombinethemwiththe +cuesemphasizehistoricalpreferences,whileitem +ground-truthitemtoformthecandidateset. +cues highlight metadata and user-group patterns. Table 9 reports the results. Compared to ran- +| Together, | they | provide | minimal | | but informative | | | | | | | | +| --------- | ---- | ------- | ------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +domsampling,performancedecreasesforallmeth- +hypothesesforprofileexploration. +| | | | | | | | ods due | to increased | difficulty, | while | DUET | con- | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | ----------- | ---------------- | ---- | ------ | +| | | | | | | | sistently | maintains | the | best performance | | across | +ExamplesofUserCues +| | | | | | | | datasets. | Thisindicatesthattheimprovementsare | | | | | +| ------- | ----- | ------ | ------ | --- | ------- | -------- | --------- | ----------------------------------- | --- | --- | --- | --- | +| “enjoys | retro | puzzle | games” | — | derived | from re- | | | | | | | +robustandnotlimitedtotrivialrankingscenarios. +peatedengagementwithclassictitles. +“prefersconciseproductreviews”—inferredfrom B.3 RobustnessunderPreferenceDiversity +apatternofshort,directcomments. +WefurtheranalyzetherobustnessofDUETunder +“tendstogivehighratingsbutrarelycomments”— +| | | | | | | | varyinglevelsofuserpreferencediversity. | | | | | Weuse | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | ----- | +highlightingconsistencybutlimitedfeedback. +thevarianceofhistoricalratingsasaproxyforpref- +| | | | | | | | erencestability: | | lowvarianceindicatesconsistent | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ------------------------------ | --- | --- | --- | +ExamplesofItemCues +| | | | | | | | preferences, | while | high | variance | corresponds | to | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ----- | ---- | -------- | ----------- | --- | +“lightweighttrail-runningshoes”—derivedfrom +diverseorpotentiallyconflictingsignals. +productmetadata. +Wepartitionusersintothreegroupsbasedonper- +| “popular | among | budget-conscious | | | users” | — in- | | | | | | | +| -------- | ----- | ---------------- | --- | --- | ------ | ----- | --- | --- | --- | --- | --- | --- | +centilethresholds(bottom33%,middle33%,top +ferredfrompurchasepatterns. +33%ofratingvariance)andevaluateperformance +| “stylized | with | retro | aesthetics” | | — extracted | from | | | | | | | +| --------- | ---- | ----- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | +withineachgroup. +itemdescriptions. +| | | | | | | | As shown | in | Table 10, | performance | | degrades | +| --- | --- | --- | --- | --- | --- | --- | -------- | --- | --------- | ----------- | --- | -------- | +smoothlyaspreferencediversityincreasesacross +| | | | | | | | all datasets. | Importantly, | | the degradation | | is grad- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------ | --- | --------------- | --- | -------- | +B AdditionalExperiments +ualratherthancatastrophic,indicatingthatDUET +B.1 Non-LLMBaselineviaExtractive +remainsstableunderheterogeneousornoisyinter- +Summarization +actionhistories. +| To examine | | whether | the | gains | of DUET | stem | | | | | | | +| ---------- | --- | ------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | +C TheUseofLargeLanguageModels +| from improved | | semantic | | representations | | rather | | | | | | | +| ------------- | --- | ---------------- | --- | --------------- | --------- | ------ | --- | --- | --- | --- | --- | --- | +| than generic | | text generation, | | we | introduce | a non- | | | | | | | +WeusedaLargeLanguageModel(LLM)onlyas +LLMbaselinebasedonextractivesummarization. a writing assistant to polish the language of the +| Specifically, | | we apply | TextRank | | (Mihalcea | and | | | | | | | +| ------------- | --- | -------- | -------- | --- | --------- | --- | ---------- | ------ | ------- | ----------- | --- | --------- | +| | | | | | | | manuscript | (e.g., | grammar | refinement, | | style ad- | +Tarau,2004)toselectsalientsentencesfromuser +| | | | | | | | justment,andclarityimprovement). | | | | Theresearch | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | ----------- | --- | +historiesandconstructuserprofileswithoutusing ideas,methodologydesign,experiments,andanal- +anygenerativemodel. ysiswereentirelyconceived,implemented,andval- +| The | extracted | summaries | | are | then fed | into the | | | | | | | +| --- | --------- | --------- | --- | --- | -------- | -------- | --- | --- | --- | --- | --- | --- | +idatedbytheauthorswithoutrelianceontheLLM. +samedownstreampredictorforratingestimation. +| | | | | | | | The LLM | did not | contribute | to research | | ideation, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | ---------- | ----------- | --- | --------- | +Thisbaselineisolatestheeffectofreadablesumma- experimentaldesign,orresultinterpretation. +rizationfromrepresentationlearning. +Table8showsthatTextRankimprovesoversim- +| ple history | truncation | | (10H), | but | remains | signifi- | | | | | | | +| ----------- | ---------- | --- | ------ | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | + +| Method | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | -------- | ---- | ----- | --- | ----------- | --------- | --- | ----------- | --------- | +| | MAE RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | MAE | RMSE | Acc F1(%) | +| | | (%) | | | | (%) | | | (%) | +10H 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +TextRank 0.8104 1.1323 50.51 30.23 0.5914 0.8223 61.19 29.68 0.6328 0.8854 59.59 29.48 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table8: Comparisonwithanon-LLMextractivesummarizationbaseline(TextRank). +| Method | | | Yelp | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | ---- | --- | ----------- | --- | --- | ----------- | --- | +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +| | Table9: | RankingperformanceunderEASE-based(Steck,2019)hardnegatives. | | | | | | | | +| ------------- | ------- | ----------------------------------------------------------- | --- | --- | ----------- | --- | --- | ----------- | --- | +| VarianceGroup | | Yelp | | | AmazonMusic | | | AmazonBooks | | +#Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 +| | | | (%) | (%) | | (%) (%) | | | (%) (%) | +| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ------- | +Stable 1530 0.4017 0.8664 71.76 68.96 432 0.2368 0.5990 82.56 86.89 1876 0.3171 0.8403 76.64 72.56 +Moderate 1354 0.4773 0.8309 60.09 46.26 436 0.3921 0.6628 64.16 47.02 1827 0.4495 0.8059 66.73 55.45 +Diverse 1444 0.6632 1.1198 51.13 45.95 428 0.5536 0.9607 57.10 59.27 1838 0.6199 1.0603 49.52 46.35 +Table10: PerformanceofDUETunderdifferentlevelsofuserpreferencediversity(measuredbyratingvariance). +#Samp. denotesthenumberofsamplesineachgroup. \ No newline at end of file diff --git a/q135/random_k4/random_4.md b/q135/random_k4/random_4.md index 37c346402ddc5ad1ad3bc67fd0d26c78800f987c..1c8737f084ba00e51b1c470d69081e0c3c1f99b5 100644 --- a/q135/random_k4/random_4.md +++ b/q135/random_k4/random_4.md @@ -1,477 +1,1014 @@ -Healthcare Quality Reporting Program -Nursing Home Summary Report -The Rhode Island Department of Health publishes information about nursing homes. If you know that you or a family member will need nursing -home care, this information can help you compare nursing homes and choose among them. You may also want to visit nursing homes and to ask -friends and family members for their thoughts and experiences. -This report summarizes information from the Department of Health (www.health.ri.gov/nursinghomes/about/quality) and Medicare -(https://www.medicare.gov/care-compare). Reports with more information are available at those websites. This report is updated every time there -is new information for one of the columns on the report. The date in the bottom left-hand corner of this report tells you when this report was last -updated. If this report is more than 30 days old, more current information may be available at the previously mentioned websites. You can learn -more about what is in this report, including definitions and time periods for each column of information, by reading the Methods Report -(http://www.health.ri.gov/publications/qualityreports/nursinghomes/SummaryMethods.pdf). -More information about choosing a nursing home: -When possible, it can be helpful to ask questions about what you can expect during your nursing home stay. This can help you to understand the -care you will receive, what it will be like to stay at a nursing home and how much services might cost. Below are some suggested questions that will -help you to be prepared for your nursing home stay: -Questions to ask your doctor or Questions to ask prospective Questions to ask your insurance Questions to ask friends and -nurse: nursing homes: provider: family: -▪ (Primary care provider) Are ▪ How long has the current ▪ What facilities are in my ▪ Have you or a loved one spent -there any facilities where you, leadership, including Director network? time in nursing home before? -or someone from your of Nursing and Administrator, If so, how was your -▪ What types of care are -practice, care for patients? been in this role? If less than experience? -covered by my plan? -a year, how long was the -▪ (Hospital care provider) Does ▪ Have you ever visited a -previous leadership in the ▪ How much care is covered by -this hospital have preferred nursing home, either as a -role? my plan? -nursing home care providers? friend or as a perspective -▪ How often will you speak with ▪ What costs will not be covered resident that you would -▪ What kind of care should I -my doctor about my care? by my plan? recommend? -expect to receive while I am at -the nursing home? ▪ Will you work with my doctor ▪ If my condition changes, will ▪ Will you be able to visit me if I -and/or my pharmacy my coverage change? choose a nursing in -▪ How long should I expect to -regarding my medication? [location]? -be at the nursing home? -▪ Do you take my insurance/are -you in my network? -Updated 10/14/2025 - -Healthcare Quality Reporting Program -Nursing Home Summary Report -The Rhode Island Department of Health publishes information about nursing homes. If you know that you or a family member will need nursing -home care, this information can help you compare nursing homes and choose among them. You may also want to visit nursing homes and to ask -friends and family members for their thoughts and experiences. -This report summarizes information from the Department of Health (www.health.ri.gov/nursinghomes/about/quality) and Medicare -(https://www.medicare.gov/care-compare). Reports with more information are available at those websites. -This report is updated every time there is new information for one of the columns below. You can learn more about what is in this report, including -definitions and time periods for each column of information, by reading the Methods Report. Please contact nursing homes directly with questions, -to inquire about private insurance, to check on bed availability or to schedule a tour. -| | Contact Information: | | | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | | | -Updated 7/1/2025* - d -| | | | | | e | | | | l | | | -| ---------------- | --- | --- | --- | --- | ----------- | ----- | -------- | ------- | ---------- | ------- | -------------- | -| | | | | s | llik | | | | la | l | | -| | | | | d | | | | | r | l | o | -| | | | | e B | S | | ꝉ s | | e | a | h u | -| | | | | fo | fo | | | | v O g | r e g | e W l n | -| | | | | | | | n | ꝉ s | L n | v n | r d F o | -| | | | | r | r e r d | ꝉ g | o | e | ꝉ G t it | O it | a s it | -| F a ci li t y, | | | | e | e a ia | | h it | y r | L A n a | a | c r e a | -| | | | | b | b c c | n | c | t u | R N e R | y R | h e v n | -| | | | | m | m s id id | iffa | t la e p | ila s a | E IT d e | li m e | t la k r ie ic | -| | | | | | u d e e | | s | u e | V i s r | r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O A e a a a e H W e a -| | | | | N | B | S | I | | R R C | F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | - -Bristol County -Crestwood Nursing & -568 Child Street, Warren 245-1574 247-0211 76 76 Y Y     I Better 17% -Rehabilitation Center -Grace Barker Nursing Home 54 Barker Avenue, Warren 245-9100 245-3730 86 86 Y Y     Same Same 23% -Rhode Island Veterans Home 480 Metacom Avenue, Bristol 253-8000 396-8127 192 n/a N N N/A N/A N/A N/A Better Better 35% -The Dawn Hill Home for Rehab -One Dawn Hill Road, Bristol 253-2300 685-8278 133 133 Y Y     N/A N/A 32% -& Healthcare -Silver Creek Rehab & Healthcare -7 Creek Lane, Bristol 253-3000 685-8278 128 128 Y Y     Worse Same 28% -Center -| Warren Skilled Nursing & | 642 Metacom Avenue, | | | | | | | | | | | -| ------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -245-2860 - 63 63 Y Y     N/A N/A 78% -| Rehabilitation | Warren | | | | | | | | | | | -| --------------- | ------- | --- | ----- | ----- | ----------- | --- | --- | --- | ----- | --- | --- | -| Kent County | | | | | | | | | | | | -557 Weaver Hill Road, -Alpine Nursing Home 397-5001 397-2455 60 60 Y Y     Better Same 55% -Coventry - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are currently frozen with July 2025 data. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf - Last updated: 10/14/2025 1 Center for Health Data and Analysis - -Healthcare Quality Reporting Program Nursing Home Summary Report - -| | Contact Information: | | | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | | | -Updated 7/1/2025* - d -| | | | | | e | | | | l | | | -| ---------------- | --- | --- | --- | --- | ----------- | ----- | -------- | ------- | ---------- | ------- | ------------ | -| | | | | s | llik | | | | la | l | | -| | | | | d | | | | | r | l | o | -| | | | | e B | S | | ꝉ s | | e | a | h u | -| | | | | fo | fo | | | | v O g | r e g | e W l n | -| | | | | | e | | n | ꝉ s | L n | v n | r d F o | -| | | | | r | r r d | ꝉ g | o | e | ꝉ G t it | O it | a s it | -| F a ci li t y, | | | | e | e a ia | | h it | y r | L A n a | a | c r e a | -| | | | | b | b c c | n | c | t u | R N e R | y R | h e k v n | -| | | | | m | m s id id | iffa | t la e p | ila s a | E IT d e | li m e | t la r ie ic | -| | | | | | u d e e | | s | u e | V A i s r | r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O e a a a e H W e a -| | | | | N | B | S | I | | R R C | F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | - -Avalon Nursing Home 57 Stokes Street, Warwick 738-1200 738-9430 31 31 Y Y     Same N/A 47% -Brentwood Nursing Home 4000 Post Road, Warwick 884-8020 884-7977 96 96 Y Y     N/A N/A 24% -Respiratory and Rehabilitation -10 Woodland Drive, Coventry 826-2000 - 210 210 Y Y     N/A N/A 56% -Center of RI -Greenwood Center 1139 Main Avenue, Warwick 739-6600 738-0310 130 130 Y Y     Same N/A 40% -660 Commonwealth Avenue, -Kent Regency Center 739-4241 732-3358 153 153 Y Y     N/A N/A 69% -Warwick -Riverview Healthcare -546 Main Street, Coventry 821-6837 - 190 190 Y Y     Worse Same 50% -Community -| Saint Elizabeth Home East | 1 Saint Elizabeth Way, East | | | | | | | | | | | -| -------------------------- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -471-6060 471-6072 168 168 Y Y     Better Better 74% -| Greenwich | Greenwich | | | | | | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Sunny View Nursing Home 83 Corona Street, Warwick 737-9193 737-9196 57 57 Y Y     Same N/A - -109 West Shore Road, -West Shore Health Center 739-9440 921-1623 145 145 Y Y     Better Same 47% -Warwick -| West View Nursing and | 239 Legris Avenue, West | | | | | | | | | | | -| ---------------------- | ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -828-9000 - 120 120 Y Y     Better Same 48% -| Rehabilitation Center | Warwick | | | | | | | | | | | -| --------------------------- | ------------------- | --- | ----- | ----- | ----------- | --- | --- | --- | --- | --- | --- | -| Newport County | | | | | | | | | | | | -| | | | | | | | | | | | | -| Royal Middletown Nursing & | 193 Forest Avenue, | | | | | | | | | | | -847-2777 848-7403 50 50 Y Y     N/A N/A 4% -| Rehabilitation Center | Middletown | | | | | | | | | | | -| ---------------------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -333 Green End Avenue, -Grand Islander Center 849-7100 849-6076 146 146 Y Y     N/A N/A 48% -Middletown -| Heatherwood Nursing & | 398 Bellevue Avenue, | | | | | | | | | | | -| ---------------------- | --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -849-6600 847-0778 114 114 Y Y     I Same 64% -| Subacute Center | Newport | | | | | | | | | | | -| ---------------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -John Clarke Retirement Centre 600 Valley Road, Middletown 846-0743 848-5890 60 60 Y Y     Same Better 34% -St Clare Home 309 Spring Street, Newport 849-3204 849-5780 50 50 Y Y     I N/A 32% -Village House Nursing & -70 Harrison Avenue, Newport 849-5222 849-5765 95 95 Y Y     Same Better 63% -Rehabilitation Center - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are updated every month. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf -Last updated: 10/14/2025 2 Center for Health Data and Analysis - -Healthcare Quality Reporting Program Nursing Home Summary Report - -| | Contact Information: | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | -Updated 7/1/2025* - d -| | | | e | | | | | l | | -| ---------------- | --- | --- | ----------- | ----- | -------- | ------- | ---- | --------------- | ------------ | -| | | s | llik | | | | | la l | | -| | | d | | | | | | r l | o | -| | | e B | S | | ꝉ s | | | e a | h u | -| | | fo | fo | | | | | v O g r e g | e W l n | -| | | | e | | n | ꝉ s | L | n v n | r d F o | -| | | r | r r d | ꝉ g | o | e | ꝉ G | t it O it | a s it | -| F a ci li t y, | | e | e a ia | | h it | y r | L A | n a a | c r e a | -| | | b | b c c | n | c | t u | R N | e R y R | h e k v n | -| | | m | m s id id | iffa | t la e p | ila s a | E IT | d e li m e | t la r ie ic | -| | | | u d e e | | s | u e | V A | i s r r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O e a a a e H W e a -| | | N | B | S | I | | R | R C F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | - -Providence County -Bannister Ctr for Rehab & -135 Dodge Street, Providence 529-4600 633-8076 161 161 Y Y     Worse N/A 37% -Healthcare -Bayberry Commons 181 Davis Drive, Pascoag 568-0600 568-3080 110 110 Y Y     Same Same 29% -Berkshire Place 455 Douglas Ave, Providence 553-8600 685-8278 220 220 Y Y     N/A N/A 49% -111 South Angell Street, -Bethany Home of Rhode Island 831-2870 331-9570 33 33 Y Y     N/A N/A 36% -Providence -49 Old Pocasset Road, -Briarcliffe Manor 944-2450 944-2455 122 122 Y Y     Better N/A 82% -Johnston -125 Scituate Avenue, -Cedar Crest Nursing Centre 944-8500 944-5162 156 156 Y Y     I Better 91% -Cranston -Cherry Hill Manor 2 Cherry Hill Road, Johnston 231-3102 - 172 171 Y Y     Same Same - -Stillwater Assisted Living and -20 Austin Avenue, Greenville 949-3880 - 80 80 Y Y     N/A N/A 31% -Skilled Nursing Community -575 Seven Mile Road, -Cra-Mar Meadows 828-5010 - 41 41 Y Y     Same Same 49% -Cranston -| Eastgate Nursing & | 198 Waterman Avenue, East | | | | | | | | | -| ------------------- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -431-2087 435-6465 68 60 Y Y     Worse N/A 60% -| Rehabilitation Center | Providence | | | | | | | | | -| ---------------------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | -Elmhurst Rehabilitation and -50 Maude Street, Providence 456-2600 - 206 206 Y Y I I I I Worse N/A 15% -Healthcare Center -| Elmwood Nursing and | 225 Elmwood Avenue, | | | | | | | | | -| -------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -272-0600 454-0818 70 70 Y Y     Same N/A 65% -| Rehabilitation Center | Providence | | | | | | | | | -| ---------------------- | ------------------------ | --- | --- | ----- | ---- | ----- | ---- | --- | --- | -| | 1 Evergreen Drive, East | | |  |  |  |  | | | -Evergreen House Health Center 438-3250 438-4813 160 160 Y Y Same Same 51% -Providence -303 Rhodes Avenue, -Friendly Home 769-7220 - 126 126 Y Y     N/A N/A 14% -Woonsocket - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are updated every month. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf -Last updated: 10/14/2025 3 Center for Health Data and Analysis - -Healthcare Quality Reporting Program Nursing Home Summary Report - -| | Contact Information: | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | -Updated 7/1/2025* - d -| | | | e | | | | | l | | -| ---------------- | --- | --- | ----------- | ----- | -------- | ------- | ---- | --------------- | ------------ | -| | | s | llik | | | | | la l | | -| | | d | | | | | | r l | o | -| | | e B | S | | ꝉ s | | | e a | h u | -| | | fo | fo | | | | | v O g r e g | e W l n | -| | | | e | | n | ꝉ s | L | n v n | r d F o | -| | | r | r r d | ꝉ g | o | e | ꝉ G | t it O it | a s it | -| F a ci li t y, | | e | e a ia | | h it | y r | L A | n a a | c r e a | -| | | b | b c c | n | c | t u | R N | e R y R | h e k v n | -| | | m | m s id id | iffa | t la e p | ila s a | E IT | d e li m e | t la r ie ic | -| | | | u d e e | | s | u e | V A | i s r r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O e a a a e H W e a -| | | N | B | S | I | | R | R C F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | - -100 Smithfield Road, -Golden Crest Nursing Centre 353-1710 353-1618 152 152 Y Y     Better Same 100% -Providence -100 Chambers Street, -Grandview Center 724-7500 724-7543 72 72 Y Y     N/A N/A 31% -Cumberland -Greenville Skilled Nursing & -735 Putnam Pike, Greenville 949-1200 - 131 131 Y Y     N/A N/A 28% -Rehabilitation -60 Eben Brown Lane, Central -Harris Health Care Center-North 722-6000 726-0850 32 32 Y Y     Better N/A 41% -Falls -833 Broadway, East -Harris Health Center 434-7404 435-4255 31 31 Y Y     Same N/A 60% -Providence -200 Wampanoag Trail, -Hattie Ide Chaffee Home 434-1520 438-8494 69 69 Y Y     Same Same 46% -Riverside -Hebert Nursing Home 180 Log Road, Smithfield 231-7016 - 133 133 Y Y     Worse Worse 36% -Heritage Hills Nursing & -80 Douglas Pike, Smithfield 231-2700 - 100 100 Y Y     N/A N/A 56% -Rehabilitation Center -Holiday Retirement Home 30 Sayles Hill Road, Manville 765-1440 765-7464 170 170 Y Y     Same Same 76% -| Lincolnwood Rehabilitation and | 610 Smithfield Road, | | | | | | | | | -| ------------------------------- | --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -353-6300 - 210 200 Y Y     Worse Same 66% -| Healthcare Center | Providence | | | | | | | | | -| ------------------ | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | -Jeanne Jugan Residence 964 Main Street, Pawtucket 723-4314 723-4316 49 49 Y Y     Better Better 52% -Mansion Nursing And -104 Clay Street, Central Falls 722-0830 728-1814 62 62 Y Y     Same Same 33% -Rehabilitation Center -Morgan Health Center 80 Morgan Avenue, Johnston 944-7800 - 120 120 Y Y     Worse Same 38% -15 Sumner Brown Road, -Mount St Rita Health Centre 333-6352 333-1012 98 98 Y Y     Better N/A 90% -Cumberland -| Oak Hill Health and Rehab | 544 Pleasant Street, | | | | | | | | | -| -------------------------- | --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -725-8888 727-8797 129 129 Y Y     Same Same 58% -| Center | Pawtucket | | | | | | | | | -| -------------------------- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| Oakland Grove Health Care | 560 Cumberland Hill Road, | | | | | | | | | -769-0800 766-3661 178 178 Y Y     N/A N/A 53% -| Center | Woonsocket | | | | | | | | | -| ------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are updated every month. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf -Last updated: 10/14/2025 4 Center for Health Data and Analysis - -Healthcare Quality Reporting Program Nursing Home Summary Report - -| | Contact Information: | | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | | -Updated 7/1/2025* - d -| | | | | e | | | | | l | | -| ---------------- | --- | --- | --- | ----------- | ----- | -------- | ------- | ---- | --------------- | ------------ | -| | | | s | llik | | | | | la l | | -| | | | d | | | | | | r l | o | -| | | | e B | S | | ꝉ s | | | e a | h u | -| | | | fo | fo | | | | | v O g r e g | e W l n | -| | | | | e | | n | ꝉ s | L | n v n | r d F o | -| | | | r | r r d | ꝉ g | o | e | ꝉ G | t it O it | a s it | -| F a ci li t y, | | | e | e a ia | | h it | y r | L A | n a a | c r e a | -| | | | b | b c c | n | c | t u | R N | e R y R | h e k v n | -| | | | m | m s id id | iffa | t la e p | ila s a | E IT | d e li m e | t la r ie ic | -| | | | | u d e e | | s | u e | V A | i s r r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O e a a a e H W e a -| | | | N | B | S | I | | R | R C F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | - -135 Tripps Lane, East -Orchard View Manor 438-2250 438-0635 166 166 Y Y     N/A N/A 27% -Providence -Overlook Nursing & -14 Rock Avenue, Pascoag 568-2549 - 100 100 Y Y     Worse Same 48% -Rehabilitation Center -Pawtucket Falls Healthcare -70 Gill Ave, Pawtucket 722-7900 - 154 154 Y Y     Same N/A 30% -Center -| Crystal Lake Rehabilitation and | 999 South Main Street, | | | | | | | | | | -| -------------------------------- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -568-3091 568-8070 71 71 Y Y     N/A N/A 68% -| Care Center | Pascoag | | | | | | | | | | -| ------------ | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Scandinavian Home 1811 Broad Street, Cranston 461-1433 461-4005 74 74 Y Y     N/A N/A 57% -10 Rhodes Avenue, North -St Antoine Residence 767-3500 769-5249 180 180 Y Y     Worse Same 52% -Smithfield -| Steere House Nursing and | 100 Borden Street, | | | | | | | | | | -| ------------------------- | ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -454-7970 831-7570 120 120 Y Y     Same Better 34% -| Rehabilitation Center | Providence | | | | | | | | | | -| ---------------------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Summit Commons -99 Hillside Avenue, -Rehabilitation and Health Care 574-4800 278-4937 165 165 Y Y     N/A N/A 37% -Providence -Center -500 Waterfront Dr, East -Tockwotton on the Waterfront 272-5280 421-0550 52 52 Y Y     N/A N/A 87% -Providence -| Trinity Health & Rehabilitation | 4 St Joseph Street, | | | | | | | | | | -| -------------------------------- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -765-5844 - 185 185 Y Y     Worse Worse 29% -| Center | Woonsocket | | | | | | | | | | -| ------------------------------- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Waterview Villa Rehabilitation | 1275 South Broadway, East | | | | | | | | | | -438-7020 438-0013 132 132 Y Y     Better Same 38% -| and Health Care Center | Providence | | | | | | | | | | -| ----------------------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -262 Poplar Street, -Woonsocket Health Centre 765-2100 765-2100 150 150 Y Y     N/A N/A 37% -Woonsocket -| Washington County | | | | | | | | | | | -| ------------------ | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | -Apple Rehab Clipper 161 Post Road, Westerly 322-8081 322-0910 60 60 Y Y     Same N/A 62% -| Elderwood Of Scallop Shell At | 55 Scallop Shell Way, Peace | | | | | | | | | | -| ------------------------------ | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -789-3006 789-3562 80 80 Y Y     N/A N/A 45% -| Wakefield | Dale | | | | | | | | | | -| ---------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are updated every month. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf -Last updated: 10/14/2025 5 Center for Health Data and Analysis - -Healthcare Quality Reporting Program Nursing Home Summary Report - -| | Contact Information: | | | | | Quality And Satisfaction: | | | | -| --- | --------------------- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | -| | | | | | | | | | | -Updated 7/1/2025* - d -| | | | e | | | | | l | | -| ---------------- | --- | --- | ----------- | ----- | -------- | ------- | ---- | --------------- | ------------ | -| | | s | llik | | | | | la l | | -| | | d | | | | | | r l | o | -| | | e B | S | | ꝉ s | | | e a | h u | -| | | fo | fo | | | | | v O g r e g | e W l n | -| | | | e | | n | ꝉ s | L | n v n | r d F o | -| | | r | r r d | ꝉ g | o | e | ꝉ G | t it O it | a s it | -| F a ci li t y, | | e | e a ia | | h it | y r | L A | n a a | c r e a | -| | | b | b c c | n | c | t u | R N | e R y R | h e k v n | -| | | m | m s id id | iffa | t la e p | ila s a | E IT | d e li m e | t la r ie ic | -| | | | u d e e | | s | u e | V A | i s r r | o c c | -A lp h a b e tical By County Address Phone Fax u N e M M t e H n Q M O e a a a e H W e a -| | | N | B | S | I | | R | R C F C | R V | -| --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | - -25 Roberts Way, North -Roberts Health Centre 294-3587 295-9357 66 66 Y Y     Same Same 66% -Kingstown -| Bayview Rehabilitation and | 860 North Quidnessett Road, | | | | | | | | | -| --------------------------- | ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -884-1802 - 120 120 Y Y     N/A N/A - -| Healthcare Center at Scalibrini | North Kingstown | | | | | | | | | -| -------------------------------- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| South County Nursing and | 740 Oak Hill Road, North | | | | | | | | | -294. 4105 295-7650 120 120 Y Y     Worse Same 73% -| Rehabilitation Center | Kingstown | | | | | | | | | -| --------------------------- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| South Kingstown & Nursing | 2115 South County Trail, | | | | | | | | | -783-8568 792-8930 112 112 Y Y     Same Same 60% -| Rehabilitation Center | West Kingston | | | | | | | | | -| ---------------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Westerly Health Center 280 High Street, Westerly 348-0020 - 120 120 Y Y     Same Same 36% -Royal of Westerly Nursing -79 Beach Street, Westerly 596-4925 596-4925 66 66 Y Y     Better N/A 57% -Center - - = Much Below Average,  = Below Average,  = Average,  = Above Average, = Much Above Average, n/a = Facility not asked to submit this information, I = Insufficient data -sent to obtain adequate rating, N/A = 10 or fewer people provided responses, Worse = Worse than state average, Same = Same as state average, Better = Better than state average, - = Information not -provided -* = Nursing Home Five-Star Ratings are updated every month. If the report that you are viewing is not current, please visit https://www.medicare.gov/care-compare/#search -More information about the Medicare Five-Star Quality Rating System is available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/FSQRS -Reports with more information about quality and satisfaction are available at www.health.ri.gov/nursinghomes/about/quality. -ꝉ = Starting with the July 2025 refresh, CMS changed the methodology for the Health Inspection Rating to use the two most recent health inspection surveys, instead of the three most recent surveys. -For more details on the Health Inspection methodologies, see the Five Star User Guide available at https://www.cms.gov/Medicare/Provider-Enrollment-and- -Certification/CertificationandComplianc/Downloads/usersguide.pdf -Last updated: 10/14/2025 6 Center for Health Data and Analysis \ No newline at end of file +Wallingford Public Schools - HIGH SCHOOL COURSE OUTLINE + +Course Title: Business Mathematics + +Department: Mathematics + +Level(s): General + +Course Number: 8114 + +Grade(s): 11-12 + +Credit: 1 + +Course Description: +This course focuses on mathematical functions using whole numbers, fractions, decimals, and +percents that are required to function in today’s business world as well as to handle personal +finance. The ability to estimate and approximate answers in solving financial problems will be +emphasized. Students will work with computer spreadsheets and learn to create and interpret +graphic representations of numerical data. Business topics that will be covered include interest +rates, job costing, merchandising, payroll, credit, taxes, insurance, and global business. +Satisfactory completion of this course satisfies one math credit for graduation. + +Required Instructional Materials +Practical Mathematics, Fredrick Leinwand Postman Wantuck +Holt Rinehart Winston 1998 Third edition + +Other Instructional Materials +Business Math (workbook), 15th ed., 2003 Thomson-Southwestern +Business Math published by Prentice Hall +Business Math published by Southwestern +Math in Everyday Life published by Walch +Money Instructor –website +Banking by BankAmerica + +Completion/Revision Date + +Approved by Board of +Education on December 17, +2007. + +Mission Statement of the Curriculum Management Team +The mission of the Wallingford K-12 mathematics program is to develop mathematical literacy in all +students. Students will understand the importance of mathematics and become flexible and +resourceful problem solvers. They will use a range of numerical, algebraic, geometrical and +statistical concepts and skills to formulate and solve authentic problems, communicating their +reasoning in oral and written form. + +Enduring Understandings for the Course + +• A problem solver understands what has been done, knows why the process was appropriate, + +and can support it with reasons and evidence. + +• There can be different strategies to solve a problem, but some are more effective and efficient + +than others are. + +• The context of a problem determines the reasonableness of a solution. +• There are many different ways a paycheck may be calculated +• There is a big difference between the money you earn and the money you take home +• The way that data is collected, organized and displayed influences interpretations. +• The analysis of data is important to the success of a business. +• Knowledge of basic mathematics is essential to personal financial success. +• Time, money, and rate of interest are significant factors in meeting financial goals. + +Page 1 of 14 + + • The amount you save is not as important as establishing the savings habits now. +• The purchasing power of a dollar changes over time. +• Costs and prices can be compared in many ways to determine the best buy. +• Credit cards have both benefits and risks. +• The sticker price of a car is negotiable. +• There are many additional expenses to owning a car. +• Many payment options should be examined before purchasing a car. +• Insurance protects a person against loss and additional expenses. +• Many factors affect insurance premiums. + +Page 2 of 14 + + Learning Strand + +1.0 Problem Solving + +NOTE: This learning strand should be taught through the integration of the other learning strands. +This learning strand is not meant to be taught in isolation as a separate unit. + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• A problem solver understands what has +been done, knows why the process was +appropriate, and can support it with reasons +and evidence. + +• There can be different strategies to solve a +problem, but some are more effective and +efficient than others are. + +• The context of a problem determines the + +reasonableness of a solution. + +LEARNING OBJECTIVES The student will : +1.1 Distinguish between given information, + +unknown information, and assumptions that +may be necessary. + +1.2 Devise a plan to solve the problem. +1.3 Identify different strategies to solve a + +problem (table, diagram, number pattern) +1.4 Apply an appropriate strategy(ies) to solve a + +problem. + +1.5 Justify the strategy and solution with + +mathematical reasoning and evidence. +1.6 Analyze strategies used to solve a problem. +1.7 Judge the reasonableness of a solution for a + +problem. + +1.8 Communicate the solution and the + +strategies used to solve the problem in oral +and/or written form. + +1.9 Pose problems from given situations + +• How do I know where to begin when solving + +a problem? + +• How does explaining my process help me to + +understand a problem’s solution better? + +• How do I decide what strategy will work best + +in a given problem situation? +• What do I do when I get stuck? +• How do I know when a result is reasonable? +• What is the relationship between solving + +problems and computation? + +INSTRUCTIONAL SUPPORT MATERIALS +• CAPT sample items and released items +• Websites for problem solving + +- Math Forum +- NCTM Standards and + +Illuminations +• CSDE Goals 2000 resources (Q drive) + +SUGGESTED INSTRUCTIONAL STRATEGIES +Each lesson throughout the course should begin +with a problem solving experience that can +integrate mathematical concepts. + +• Small group discussion about strategies and + +solutions + +• Written explanation of strategy and solution + +for a problem + +• Peer editing of strategies and solutions +• Revise written solutions of problems +• Oral presentation of solution strategies +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Graphic organizers +• Cooperative learning +• See other learning strands for integration + +SUGGESTED ASSESSMENTS + +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +Page 3 of 14 + + for a problem + +• District rubric for assessing problem solving +• Oral presentation of solution strategies +• See other learning strands for integration + +Page 4 of 14 + + LEARNING STRAND + +2.0 Earning a Living + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• There are many different ways a paycheck + +may be calculated + +• There is a big difference between the money + +you earn and the money you take home + +LEARNING OBJECTIVES The student will : +2.1 Calculate gross pay for hourly rate +employees +2.2 Calculate gross pay for salaried employees +2.3 Calculate regular and overtime pay +2.4 Calculate salary based on commission +2.5 Calculate salary based on piece work +2.6 Calculate deduction & net pay +2.7 Identify employee benefits +2.8 Compare job benefits and job expenses + +• How is your net pay calculated? +• How can job expenses and job benefits + +affect your earnings? + +• How can an employee who is paid hourly +earn more than someone who is salaried? + +INSTRUCTIONAL SUPPORT MATERIALS + +• Internet +• Overhead transparency +• Example paycheck stubs + +SUGGESTED INSTRUCTIONAL +STRATEGIES + +• On site visitation of payroll department +• Spreadsheet +• Guest speaker +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENT METHODS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 5 of 14 + + LEARNING STRAND + +3.0 Data Analysis + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• The way that data is collected, organized +and displayed influences interpretations. + +• The analysis of data is important to the + +• Why is data collected and analyzed? +• How do people use data to influence + +others? + +success of a business. + +• How can predictions be made based on + +LEARNING OBJECTIVES The students will: +3.1 Calculate measures of central tendency +3.2 Calculate range, quartiles, and outliers +3.3 Analyze data related to the world of business +3.4 Represent data using appropriate graphs + +(line, bar, pie, picture, etc.) +3.5 Interpret data based on a graphs + +data? + +INSTRUCTIONAL SUPPORT MATERIALS + +• Newspapers +• Internet +• Excel + +SUGGESTED INSTRUCTIONAL +STRATEGIES +• Spreadsheet +• Guest speaker +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENT METHODS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 6 of 14 + + LEARNING STRAND + +4.0 Personal Banking + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• A knowledge of basic mathematics is + +essential to personal financial success. + +• Time, money, and rate of interest are + +significant factors in meeting financial goals. +• The amount you save is not as important as + +establishing the savings habits now. + +• What are the benefits of have a checking + +and savings account? +• What is the Rule of 72? +• Why do we reconcile a bank statement? +• What are the benefits of banking online? + +LEARNING OBJECTIVES The students will : +4.1 Define debits, credits, fees, & balance. +4.2 Write out a check. +4.3 Maintain a check book. +4.4 Reconcile a bank statement for a checking +account. +4.5 Interpret a bank statement for a savings +account. +4.6 Identify and analyze the effect of different +kinds of checking accounts. +4.7 Calculate interest and principal using simple +and compound interest formulas +4.8 Calculate interest on money market, bonds, +and CD accounts +4.9 Identify and calculate hidden costs of +personal banking + +INSTRUCTIONAL SUPPORT MATERIALS + +• Checkbook +• Internet +• Sample bank statement +• Overhead transparency + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 7 of 14 + + LEARNING STRAND + +5.0 Responsible Purchasing + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• The purchasing power of a dollar changes + +over time. + +• Costs and prices can be compared in many + +ways to determine the best buy. + +• Credit cards have both benefits and risks. +• There are various methods of payment for +purchases, (i.e. credit cards, debit cards, +check, cash) + +LEARNING OBJECTIVES The student will : +5.1 Compute inflation rate +5.2 Compute purchasing power after inflation +5.3 Calculate sales tax +5.4 Analyze the best buy based on unit prices +5.5 Analyze the best buy based on discounts, +sales price, coupons, and rebates +5.6 Read a credit card statement +5.7 Compute an unpaid balance and identify the +new charges +5.8 Compute a credit finance charge based on +fixed and variable rates +5.9 Compute a new credit balance +5.10 Identify and analyze the benefits and +drawbacks of credit cards + +• What is the consumer price index? +• How does inflation affect purchasing power? +• How does a knowledge of mathematics help +determine which product is a better buy? +• How does knowledge of mathematics help + +one to use a credit card responsibly? + +INSTRUCTIONAL SUPPORT MATERIALS + +• Circulars +• Internet +• Overhead transparency +• Consumer reports + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 8 of 14 + + LEARNING STRAND + +6.0 Transportation Expenses + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• The sticker price of a car is negotiable. +• There are many additional expenses to + +• What expenses factor into owning a car? +• What are important considerations when + +owning a car. + +buying a car? + +• Many payment options should be examined + +• How will a knowledge of mathematics help + +before purchasing a car. + +make a better purchase? + +LEARNING OBJECTIVES The student will : +6.1 Estimate and compute traveling distances +using a mileage chart or map +6.2 Estimate and compute travel time and rate +of travel +6.3 Determine the cost of a car with options and +additional expenses +6.4 Compute the amount you need to finance to +buy a new car +6.5 Compute the finance charge for an +automobile loan +6.6 Calculate operating expenses (mpg, gas, +repair) +6.7 Justify the purchase of a given car +6.8 Compute the cost for renting a car + +INSTRUCTIONAL SUPPORT MATERIALS + +• Car advertisements +• Newspapers +• Consumer reports + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 9 of 14 + + LEARNING STRAND + +7.0 Insurance + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• Insurance protects a person against loss + +and additional expenses. + +• Why is insurance important? +• What factors affect the premiums of different + +• Many factors affect insurance premiums. + +kinds of insurance? + +LEARNING OBJECTIVES The student will : +7.1 Define premium and deductible +7.2 Compute life insurance premiums +7.3 Identify the differences between term and +whole life insurance +7.4 Decide, given relevant information, which +type of life insurance to purchase +7.5 Compute health insurance premiums +7.6 Compute insurer’s and insured share of +health insurance cost +7.7 Compute insured’s share of medical bills +7.8 Identify the different types of automobile +insurance +7.9 Compute yearly automobile insurance +premiums +7.10 Calculate property insurance premiums +7.11 Calculate how much can be collected on +insurance claims +7.12 Analyze the factors that influence the +amount of coverage of any insurance + +• To what extent will insurance protect you? + +INSTRUCTIONAL SUPPORT MATERIALS + +• Requested material from various insurance + +companies + +• Internet + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 10 of 14 + + LEARNING STRAND + +8.0 Long Term Investments + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• With higher risk comes the potential to earn +higher reward or returns on our investments. + +• It is important to divisive investments. +• Different types of investments offer varying + +• What is the stock market and how does it + +work? + +• What is the difference between savings and + +investments? + +growth potential + +• What is the difference between a dividend + +• Investments aren’t guaranteed to grow. + +and a yield? + +LEARNING OBJECTIVES The student will : +8.1 Compute the cost, redemption value, and +interest earned on U.S. savings bonds. +8.2 Calculate the total investment in a mutual +fund. +8.3 Calculate profit or loss from mutual fund +investments. +8.4 Calculate the cost of stock purchases. +8.5 Calculate annual stock dividends. +8.6 Calculate the yield on stock investments. +8.7 Calculate the proceeds from the sale of +stock. +8.8 Explain what the stock market is and how it +works. +8.9 Identify factors that cause the stock market +to fluctuate. + +• What makes one stock a more desirable + +investment then the other? + +INSTRUCTIONAL SUPPORT MATERIALS + +• www.stocksquest.com +• Newspaper + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 11 of 14 + + LEARNING STRAND + +9.0 Budgeting + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• Organization is necessary to be fiscally + +responsible. + +• What are the benefits of a budget? +• Why is a budget important for future + +• Some expenses are predictable while some + +financial success? + +are varying + +• How can maintaining a budget provide + +• Tracking income and expenses can prevent + +possible one time expense opportunities? + +overspending + +LEARNING OBJECTIVES The student will : +9.1 Identify fixed, variable, and annual expenses +9.2 Compute the amount to budget for each of +these expenses +9.3 Analyze a monthly budget sheet +9.4 Prepare a personal budget +9.5 Apply the use of area and perimeter in +preparing a budget for a one time expense. + +INSTRUCTIONAL SUPPORT MATERIALS + +• Excel +• Protractors +• Compass +• Rules +• Overhead Transparency + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning +• Creating a pie chart representation of a + +budget + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Landscape/Remodeling Project +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 12 of 14 + + LEARNING STRAND + +10.0 Marketing & Advertising + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• Advertising is a necessary expense to + +• What is the purpose of marking & + +create a greater profit. + +advertising? + +• Simple algebraic procedures can predict + +• What are the benefits and draw backs of + +future sales. + +LEARNING OBJECTIVES The student will : +10.1 Predict future sales using trend +data/forecast methods +10.2 Calculate market share +10.3 Compute the cost of advertising in various +media +10.4 Analyze the most effective advertising +techniques/strategies + +various marketing media? + +• How does forecasting affect current + +business practices? + +INSTRUCTIONAL SUPPORT MATERIALS + +• Ruler +• Graph paper +• Graphing calculators + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 13 of 14 + + LEARNING STRAND + +11.0 Production + +ENDURING UNDERSTANDING(S) + +ESSENTIAL QUESTIONS + +• There are numerous expenses involved in + +providing goods and services. + +• Why is quality control important? +• How does overhead affect the cost of an + +• You must sale a certain number of goods + +item? + +before making a profit. + +• Simple algebraic procedures can predict the + +break-even point. + +LEARNING OBJECTIVES The student will : +11.1 Calculate prime cost and total +manufacturing costs +11.2 Identify various overhead costs +11.3 Classify expenses as raw material, labor, +or overhead +11.4 Calculate the break-even point in +production units and in sales production +11.5 Determine the number of manufactured +items to be tested for quality control +11.6 Predict the number of defective items on +an assembly line + +INSTRUCTIONAL SUPPORT MATERIALS + +• Excel +• Overhead transparency +• Internet + +SUGGESTED INSTRUCTIONAL STRATEGIES + +• Spreadsheet +• Journal writing +• Modeling +• Think aloud +• Student inquiry through guided + +investigations + +• Teacher directed class discussions +• Graphic organizers +• Cooperative learning + +SUGGESTED ASSESSMENTS + +• Homework +• Projects +• Tests & Quizzes +• Open-ended questions included on tests + +and quizzes + +• Performance Tasks +• Teacher observations +• Written explanation of strategy and solution + +for a problem + +• Oral presentation of solution strategies + +Page 14 of 14 + diff --git a/q143/random_k2/question.json b/q143/random_k2/question.json index 35078881fde44f40bd0e7b10b7f28f8d6fc88d04..ec60797fb66ce11c26572ab7d5bc82f1c1719393 100644 --- a/q143/random_k2/question.json +++ b/q143/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "germanwingsdigitalcrisisanalysis-150403064828-conversion-gate01_95.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf", - "web_ff8dcb0ea2a037b7.pdf" + "web_5e3d260ea31a243f.pdf", + "resize_r2_22.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q143/random_k2/random_1.md b/q143/random_k2/random_1.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..f94127163277f669a14f3b0a86d9bfd3ebbd9e38 100644 --- a/q143/random_k2/random_1.md +++ b/q143/random_k2/random_1.md @@ -1,1453 +1,6514 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA -8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +C H A P T E R 4 + +Climate Change Impacts +on Human Health + +Song Liang1,5, Kristina Kintziger2, Phyllis Reaves3, and Sadie J. Ryan4,5 + +1Department of Environmental and Global Health, College of Public Health and Health Professions, +University of Florida, Gainesville, FL; 2Department of Public Health, College of Education, Health & +Human Sciences, The University of Tennessee, Knoxville, TN; 3Division of Physical Therapy, School of +Allied Health Sciences, Florida Agricultural & Mechanical University, Tallahassee, FL; 4Department of +Geography, University of Florida, Gainesville, FL; 5Emerging Pathogens Institute, University of Florida, +Gainesville, FL + +Climate change poses major challenges to human society and to Earth systems, influencing the +functioning of many ecosystems and thereby affecting human health. Many climate change/variability- +and extreme weather-associated events, such as sea level rise, hurricanes, and storm surge, as well as +other weather extremes, including excessive precipitation and heatwaves, have direct and/or indirect +impacts on human health. These impacts include death/injury, cardiovascular and respiratory diseases, +environmentally-mediated infectious diseases, and mental health, among others. Due to its unique +geography, Florida is particularly vulnerable to these environmental impacts, which have important +health implications for the state’s more than 20 million residents. In this chapter, we review the health +impacts of climate change and associated weather events, with an emphasis on those relevant to Florida, +and environmental hazards, including hurricanes and storms, lightning, sea level rise, excessive +precipitation, extreme heat, and drought. There is clear evidence for significant climate-sensitive hazards +and human health impacts in the state, despite uncertainties associated with the assessment of some +effects. To address health impacts and challenges, policies focused on mitigation and adaptation +strategies, health surveillance, and research that could close knowledge gaps on human exposures to the +climate-sensitive hazards and health impacts are needed. + +Key Messages + +• Florida is highly vulnerable to climate-sensitive hazards (e.g. sea level rise, heat waves, storm + +surge, and hurricanes), which have a wide range of human health effects. + +• The health effects can be direct, such as storm/temperature related illnesses, injuries, and +deaths; or indirect, such as waterborne, food-borne, and vector-borne diseases; or take social +and economic pathways, such as stress and mental illness. + +• The health effects exhibit substantial regional disparities across the state. +• Policies focused on health surveillance and research on knowledge gaps between human + +exposure to the hazards and health effects are much needed. + +Keywords + +Climate change; Environmental hazards; Human health; Florida + +125 + + 1 2 6 • S O N G L I A N G E T A L . + +Introduction + +C + +limate change is well-characterized at the global level using metrics such as the global +mean temperature and sea level. But the manifestations of global climate change and +variability occur at varying geographical scales, causing a variety of weather events, +such as excessive rainfall, drought, severe storm/flooding, sea level rise, and heat waves (Field +et al. 2014), all of which are likely to have direct and/or indirect human health impacts +(McMichael and Haines 1997; Colwell et al. 1998; Frumkin et al. 2008). Indeed, there is clear +and increasing evidence that many health outcomes (most adverse) exhibit high sensitivities to +these varying weather events. The interface between the climate change-associated weather +events and human health exhibits a complex web of relationships involving both natural and +social environments through direct or indirect impact pathways, social institutional disruption, +or a combinations of these things (Fig. 4.1). Globally, substantial disease burdens are attributable +to hazards that are associated with climate change and related weather events. According to the +World Health Organization’s 2008 estimate, climate change-associated hazards were responsible +for more than 150,000 deaths and 5,517,000 disability-adjusted life years, which were highly +likely to be underestimated as the estimates were only based on selected risk factors and illnesses +associated with climate change and related weather events (WHO 2008). The actual disease +burden might be greater and the adverse health impacts are likely to increase in the years to come, +as the Intergovernmental Panel on Climate Change (IPCC) and US Global Change Research +Program have clearly indicated that the accelerated changing climate poses a substantial threat +to global human health and the risk will continue to become severe if no remediation action is +taken (Field et al. 2014; Balbus et al. 2016). + +Due to its unique geography, Florida is highly vulnerable to a variety of hazards associated +with climate change and variability, and related weather events. Sea level rise, hurricane and +storm surge, excessive precipitation, and heatwaves all pose threats to Florida’s agriculture, +ecosystems, tourism, and public health. For example, Florida has experienced a greater number +of hurricane landfalls than other states in the country (Knight and Davis 2009), and more intense +hurricanes are expected in the future (Knutson et al. 2013). In addition, rising sea levels are +influencing and expected to continue to have significant impacts on communities and residents +along Florida’s coastal areas, particularly in the southern part of the state, making Floridians +exceptionally vulnerable to these environmental hazards (Elsner et al. 2008). + +This chapter is organized into sections addressing impact pathways and specific +environmental hazards of particular concern in Florida (Fig. 4.1). For direct impacts, we focus +on extreme heat, flooding, storms and lightning, and sea level rise. For indirect impacts on public +health, we focus on water, food, and vector-borne diseases. And lastly, we conclude with our +discussion of impact pathways through social disruption on mental and community health impact +associated with climate change and related weather events. Throughout this chapter, health + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 7 + +impacts and implications, particularly those pertinent to Florida, are reviewed primarily based +on the Building Resilience Against Climate Effects (BRACE) project reports from the Florida +Department of Health (FDH) (FDH 2015b, a) and other relevant publications. + +Figure 4.1. Impact pathways by which climate change and variability and related weather events affect +human health. + +Direct Impact Pathways – Environmental Hazards +and Human Health + +A number of human health impacts come from direct exposure to environmental hazards +associated with climate change and related weather events. Among those of particular concern +in Florida are extreme heat, flooding, storms and lightning, and sea level rise. + + 1 2 8 • S O N G L I A N G E T A L . + +Extreme Heat and Health Impacts + +Extreme climate and weather events (e.g., heat, cold, storms, and floods) are expected to occur +more frequently worldwide due to climate change. These extreme events, such as heatwaves and +cold spells (e.g. frequent very hot days and possibly fewer very cold days), have a direct impact +on human health by compromising the body’s ability to regulate its internal temperature. +Increased human morbidity and mortality have been associated with extreme temperatures, both +hot and cold, and documented in both developing and developed countries worldwide (Center +for Disease Control and Prevention 1994; Keatinge et al. 2000a; Keatinge et al. 2000b; Basu and +Samet 2002; Medina-Ramon et al. 2006; Kovats and Hajat 2008; Bandyopadhyay et al. 2012). +The mechanistic effects of extreme temperature on death and illness are briefly summarized +below (Sarofim et al. 2016): +• Extreme heat may induce heat cramps, heat exhaustion, heatstroke, and hyperthermia. +• Extreme cold may induce hypothermia and frostbite. +• Extreme heat/cold may exacerbate chronic conditions related to cardiovascular disease, + +respiratory disease, cerebrovascular disease, and diabetes-related illnesses. + +• Extended exposure to high temperatures may induce increased hospital admissions for + +cardiovascular, kidney, and respiratory disorders. + +• Extreme heat may induce or exacerbate mental health and behavioral disorders. + +A number of studies have reported associations between high temperatures and human deaths +or illnesses around the world. For example, a large number of heat-related deaths have been +reported in the United States (Center for Disease Control and Prevention 1994; Curriero et al. +2002; Hoshiko et al. 2010), Europe (Keatinge et al. 2000a; Keatinge et al. 2000b; Huynen et al. +2001), and Asia (Qian et al. 2008; Chung et al. 2009). A recent study by Berko et al. examined +deaths attributed to extreme weather events from 2006 to 2010 in the United States and found +that each year about 2,000 U.S. residents died from weather-related causes, and among those +31% died due to exposure to excessive heat, heat stroke, and/or sun stroke (Berko et al. 2014). +Mounting evidence has suggested that substantial increases in mortality from respiratory and +cardiovascular diseases, in particular among the elderly and youth, are associated with high +temperatures in the United States (Basu and Samet 2002; Medina-Ramon et al. 2006; Kovats and +Hajat 2008). During the 2006 California heat wave, 16,166 excess emergency department visits +and 1,182 excess hospitalizations were reported statewide. This included a significant increase +in emergency department visits for acute renal failure, cardiovascular diseases, and diabetes +(Knowlton et al. 2009). The association between heat and respiratory diseases has also been +reported in many studies. In New York state, a significant number of respiratory hospitalizations +were attributable to excessive heat, and a projection suggested that excess respiratory admissions +in the state due to extreme heat will be two to six times higher in 2080–2099 than what was seen +in 1991–2004 (Lin et al. 2012). In another study conducted in Greater London, UK, an analysis + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 9 + +of historical data suggested that heat-related increases in emergency admissions for respiratory +and renal diseases had been observed in children under five, and for respiratory disease in the +75+ age group (Kovats et al. 2004). + +Extreme Heat and Public Health in Florida + +Florida has a warm subtropical and tropical climate. Due to continuous summer heat and extreme +weather events, temperatures around the state can reach levels potentially harmful to human +health. Through analyses of historical data (for the periods 1895–2009 and 1970–2009), a +significant trend in Florida’s mean temperature has been found (Martinez et al. 2012). The +average summer temperature (June–August) has been showing an overall increasing trend, +particularly after the 1980s, which shows a greater upward trend (FDH 2015a), likely due to a +variety of factors ranging from rapid urban development (Martinez et al. 2012; FDH 2015a) to +climate change (Ji et al. 2014). Within the state, substantial spatial heterogeneities in mean, +minimum, and maximum temperature have been reported (Martinez et al. 2012). For example, +the distribution of annual mean number of days with a maximum temperature greater than or +equal to 95 °F and greater than or equal to 75 °F showed that more days with hottest daytime +temperatures occurred in the northern and interior portions of the state (FDH 2015a). + +Using statewide information on specific health outcome-related emergency department visits +that occurred May to October from 2005–2012, the BRACE project examined the effects of daily +maximum temperature or daily maximum heat index on emergency department visits through a +two-stage analysis—the first stage involved the analysis on regional (National Weather Service +regions) scales using the Poisson regression model; the second stage used a meta-analysis +technique to integrate regional estimates into a statewide estimate (FDH 2015a). Rate ratios, +based on the comparison between the exposure (e.g., hotter days) and the reference (e.g. 88 °F +for temperature and 94 °F for heat index), were used to compare emergency department visits +for specific health conditions (FDH 2015a). +a) Heat and heat-related illness. A strong association between heat-related illness and heat +exposure was observed; the significant increase in emergency department visit rates for heat- +related illness was associated with increasing temperatures and a strong dose-response +relationship across all regions in Florida was found (FDH 2015a). + +b) Heat and cardiovascular disease. Two types of cardiovascular disease were considered— +myocardial infarction and ischemic stroke. No statistically significant association was found +between temperature and ischemic stroke at the state level; however, significant regional +variations were identified. For example, the Tampa region showed a significant positive +association between heat index (above the reference level at 94 °F) and stroke (FDH 2015a). +Furthermore, a positive, statistically significant relationship was observed between +myocardial infarction and heat index despite remarkable geographical variations (FDH +2015a). + + 1 3 0 • S O N G L I A N G E T A L . + +c) Heat and respiratory disease. Overall, increases in the number of emergency department +visits for asthma were associated with higher temperatures. Although no significant +association was found at the state level between temperature and emergency department visits +for asthma, the regional analysis suggested that all regions except Tallahassee showed a +statistically significant positive association between temperature and asthma (FDH 2015a). + +Flooding, Storms, Lightning, and Sea Level Rise + +Excessive precipitation, hurricanes, coastal storms, sea level rise, and thunderstorm-related +lightning, which are all typically accompanied by coastal and inland flooding, have the potential +for substantial direct human health impacts (e.g. injury and death, maternal and child health, and +mental health issues). Natural disasters related to flooding and storms are a significant cause of +mortality and morbidity. For example, according to the International Federation of Red Cross +and Red Crescent Societies (IFRCRCS), more than 3,448 climatic, hydrological and +meteorological disasters (including 1,751 major floods and 988 storms) resulting in 339,710 +deaths (59,092 and 177,685 deaths specifically tied to flooding and storms) were reported +worldwide between 2005 and 2014 (IFRCRCS 2015). Lightning strike is the second leading +cause of weather-related mortality at the global scale, with an estimated 0.2-1.7 deaths per million +people, and those who survive lightning strikes often suffer from significant injuries (Aslar et al. +2001; Ritenour et al. 2008). The magnitude and impacts of these extreme weather events, their +severity, and the extent of the effects are influenced by many different factors. For example, +short- and long-term averages and variability of weather conditions and physical impacts of +associated extreme events, as well as social environmental factors (e.g. infrastructure, social and +individual vulnerability), are considered important (Bouma et al. 1997, Kovats 2000, Kovats et +al. 2003, Miranda 2004). For projections of future flooding under different climate change +scenarios, climate models have consistently suggested that episodes of severe flooding may +become more frequent in inland river systems including flood plains (Christensen and +Christensen 2003, Booij 2005), urban and coastal environments in various parts of the world +(Schreider et al. 2000; Douglas et al. 2008; Kirshen et al. 2008; Thompson et al. 2009, Diez et +al. 2011; Lyle and Mills 2016). Meanwhile, climate models also predict positive correlations +between lightning and global temperatures, suggesting the likelihood of a greater number of and +more severe lightning episodes (Price and Rind 1994; Reeve and Toumi 1999; Kochtubajda et +al. 2006). These episodes are expected to increase the risk of extreme environmental hazards and +adverse public health impacts. +a) Natural disasters in Florida. Due to its geography (extensive coastline and a peninsular +shape) and its tropical/subtropical climate, Florida is particularly vulnerable to tropical +storms, hurricanes, and lightning strikes. Historically the occurrence of hurricanes in the US +has been clustered in Florida and along the Atlantic coastline (Ellis et al. 2015). A total of 67 +known Florida hurricanes have occurred over the 108-year period (1900-2007) with four + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 1 + +different hurricanes occurring in one year (2004) in Florida. The climate change model +suggests a 46% chance that Florida will be hit by at least one hurricane each year in the future +(Malmstadt et al. 2009), and recent experience seems to support this. Fig. 4.2 illustrates the +distribution of average return time of hurricane landfalls and populations along the coastal +areas of Florida, showing spatial heterogeneities in vulnerability to hurricanes. + +b) Public health impacts. Key direct public health impacts associated with extreme weather +events include injury and death due to trauma, drowning, destructive forces of wind, +collapsed building and trees, and lightning strike and carbon monoxide poisoning related to +power outages. Historically, Florida experienced significantly higher mortality in the early +20th century; approximately 3,000 deaths were attributed to extreme weather-related events +(Winsberg 2003) and the largest number of deaths due to lightning strike in the US were +reported during this time (Duclos et al. 1990; Ritenour et al. 2008). During 2004 and 2005, +eight hurricanes hit Florida resulting in 213 deaths, over half of which were caused by trauma, +followed by drowning, other injury, electrocution, and carbon monoxide poisoning (Ragan et +al. 2008). Injury is common cause of hurricane-related morbidity and mortality. For example, +in the aftermath of Hurricane Katrina between September 8 and October 14, 2005, 7,543 non- +fatal injuries among residents and relief workers were recorded in the surveillance system +(Sullivent et al. 2006). During a 2006 flood event in El Paso County in Texas, 43% of +individuals (out of 475 surveyed) were reported having physical health issues related to the +flooding episode (Collins et al. 2013). + +c) Injury. The BRACE study compared injury-related emergency department visits and +hospitalizations during impact periods (periods covering tropical cyclone landfall) vs. control +periods (periods with no such impact) between 2004 and 2012. The study found that +emergency department visit and hospitalization rates were significantly higher during the +impact periods, with a rate ratio of 1.03 (95% confidence interval (CI): 1.02, 1.05) for the +emergency department visit rate and 1.4 (95% CI: 1.02, 1.05) for the hospitalization rate. The +most common types of injuries during the impact periods included falls, being struck by an +object, being cut or pierced, and motor vehicle transport accidents (FDH 2015b). + +d) Carbon monoxide poisoning. Carbon monoxide (CO) is a colorless, odorless, poisonous gas +that can be harmful when inhaled in a large amount. Inhaled CO enters the blood stream and +reduces the delivery of oxygen to the body’s critical organs, such as the heart and brain. The +gas is primarily generated through incomplete combustion. The greatest sources of CO +outdoors are usually related to motor vehicles, machinery that burn fossil fuels. Indoors, some +major sources include unvented kerosene and gas space heaters, leaking chimneys and +furnaces, and gas stoves. Exposure to CO may cause weakness, headache, dizziness, nausea, +shortness of breath, confusion, and even death at very high concentration levels, which is +more likely in indoor environments. A recent systematic review of the health impacts of +power outages due to extreme events indicated that CO poisoning is an important health +concern (Klinger et al. 2014). In Florida, the BRACE study reported that during the study + + 1 3 2 • S O N G L I A N G E T A L . + +period 2004–2012, the rate of CO exposure calls during impact periods was 6.59 times (95% +CI: 4.48, 9.7) the rate of CO exposure calls to poison control centers during the control +periods. The study also found the rate of CO-poisoning related emergency department visits +was significantly higher for the impact periods than during the control periods with the rate +ratio 3.44 (95% CI: 2.07, 5.72). Finally, the BRACE study revealed that rates of CO +poisoning-related hospitalizations were also significantly higher during the impact periods +with the rate ratio 4.0 (95% CI: 2.9, 5.51) (FDH 2015b). + +Figure 4.2. Historical distribution of hurricane landfalls in Florida. + +Indirect Impact Pathways–Infectious Diseases + +Climate change and variability and associated variable weather conditions can affect human +health through indirect pathways (Fig. 4.1). These indirect impacts are modulated through +biological and ecological processes that can influence infectious diseases (in particular, some +water/food- and vector-borne diseases) and air quality. In this section, we focus on water/food-, +and vector-borne diseases. + +Disease Transmission Groupings + +The impact of climate change and variability and associated weather events on infectious +diseases has been receiving much attention in the past several decades. Across the globe, high + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 3 + +human mortality rates and disability have been attributed to infectious diseases; climate change +is intensifying and will continue to exacerbate transmission of infectious diseases, poses +increasing concerns (Patz et al. 2005). Indeed, the transmission of many infectious diseases is +closely linked to physical environmental factors, such as water and temperature. Many water- +associated and zoonotic pathogens, for instance, which account for more than 75% of known +pathogens causing human diseases, are associated with hydrological process (WHO 2003). From +a transmission perspective, infectious diseases can be broadly classified into two groups: 1) +directly transmitted (i.e., those spread directly from person to person via direct contact or droplet +exposure, such as influenza and tuberculosis); and 2) indirectly transmitted, including those +mediated through an intermediate vector organisms (vector-borne) or through environmental +media such as water and soil (environmentally-mediated) (Eisenberg et al. 2007). Because the +diseases in the indirectly transmitted group are either directly or indirectly related to water, this +group can be further broken down into the following sub-groups based on water’s role in disease +transmission: waterborne (e.g., cholera typically via fecal-oral transmission), water-based (e.g. +part of the pathogen life cycle requires aquatic environment, such as schistosomiasis), and water- +related (e.g., malaria and dengue, which need water for breeding of insect vectors to fulfill the +transmission cycle) (Yang et al. 2012; Mordecai et al. 2013; Ryan et al. 2015; Johnson et al. +2015; Mordecai et al., 2017) . + +Climate-Sensitivity of Microbial Agents and Their Associated Diseases + +Table 4.1 lists some of the major pathogens and their associated diseases that are climate sensitive +and associated with water. These disease-causing pathogens, either through infection or toxin- +generating (e.g. caused by protozoa, bacteria, virus, helminth, or algae), and their associated +vectors (e.g., snails, mosquitoes, sand flies) are very small in size and lack thermostatic +mechanisms (i.e. lack of function to maintain body temperatures). Their biological (e.g., survival, +reproduction, development/growth, and infectivity) and environmental (e.g., contamination and +movement) processes are determined by environmental conditions, such as rainfall, storm runoff, +temperature, and sunlight. Climate change and variability and associated weather events are +expected to affect both fresh and marine water environments thereby changing humans’ exposure +to these water-related contaminants or pathogens that cause ill health. + +a) Waterborne Diseases +Waterborne diseases remain an important contributor to the global burden of disease (Pruss et al. +2002; Yang et al. 2012; Pruss-Ustun et al. 2014). Many different viral, bacterial, and parasitic +diseases have been associated with waterborne transmission (Table 4.1). Human exposure to +waterborne pathogens is primarily through drinking water, recreational water (e.g. via accidental +drinking and, for some pathogens, via dermal contact), or foods. The distribution of these +pathogens exhibits significant geographical variations, depending on both physical + + 1 3 4 • S O N G L I A N G E T A L . + +environmental and socio-economic conditions (Yang et al. 2012). For example, waterborne +cholera, viral hepatitis, and some other water-based and water-related diseases (e.g., +schistosomiasis and onchocerciasis) are largely confined to certain tropical areas where the +prevalence of such diseases is largely due to lack of access to improved drinking water, sanitation +and hygiene (Pruss et al. 2002), while some pathogens, such as Cryptosporidium parvum, +Giardia duodenalis, and Campylobacter spp. have a much wider geographical distribution (Yang +et al. 2012). Regardless of the geographical region, heavy rainfall, flooding, and temperature are +among the most important factors associated with transmission and outbreaks of these +waterborne diseases. + +There is mounting evidence that weather events are often an important factor triggering +waterborne outbreaks. In developed countries, extreme weather events such as excessive rainfall +and flooding can overwhelm water treatment plants and/or increase runoff into recreational +water, leading to water contamination that causes outbreaks (Kistemann et al. 2002). Some recent +studies have clearly indicated that excessive rainfall has been a significant contributor to +waterborne outbreaks. An analysis of historical records of 548 waterborne outbreaks in the US +reported between 1948 and 1994 suggested a significant association between the outbreaks and +rainfall—51% of waterborne disease outbreaks were preceded by precipitation events above the +90th percentile and 68% by events above the 80th percentile; the strongest association was +between the outbreaks and surface water contamination due to extreme precipitation (Curriero et +al. 2001). In Canada, Thomas et al. examined extreme rainfall and spring snowmelt in relation to +92 Canadian waterborne disease outbreaks that occurred between 1975 and 2001. They found +that warmer temperatures and extreme rainfall were significant contributing factors to waterborne +disease outbreaks in Canada (Thomas et al. 2006). Similar studies have been conducted in +England (Nichols et al. 2009), the Netherlands (Schijven and de Roda Husman 2005), Finland +(Miettinen et al. 2001), Denmark (Laursen et al. 1994), and Taiwan (Chen et al. 2012). In many +developing countries, lack of access to improved drinking water and sanitation exacerbates the +impact of these extreme precipitation weather events on waterborne diseases. Diarrhea, caused +by many waterborne pathogens, remains a top killer of children under five in the developing +world, particularly in Africa (Pruss-Ustun et al. 2014). A large number of studies have been +carried out in the developing world. For example, cholera outbreaks caused by toxigenic Vibrio +cholerae have been consistently shown to be correlated with excessive rainfall, flooding, and +high temperatures in the epidemic areas of West Africa and Bangladesh (Mhalu et al. 1987; +Hashizume et al. 2008; Luque Fernandez et al. 2009; Ngwa et al. 2016), although the exact +mechanistic relationships that result in this correlation remain elusive (Colwell et al. 1998). +Significant (positive) relationships have been reported between excessive rainfall and/or high +temperatures and diarrhea in the Pacific Islands (Singh et al. 2001), Ecuador (Carlton et al. 2014), +and Sub-Saharan Africa (Bandyopadhyay et al. 2012), while evidence also suggests that low +rainfall and even drought can also be associated with diarrhea, as reported in Bangladesh +(Hashizume et al. 2008) and Denmark (Senhorst and Zwolsman 2005). + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 5 + +b) Waterborne Diseases and Climate Impact in Florida +Across the US, climate change and variability and associated weather events are affecting and +are expected to continue to affect both marine and freshwater resources. These effects extend to +some water-associated pathogens and related diseases. In Florida, waterborne and foodborne +diseases of major public health concerns are highlighted in Table 1. Enteric bacteria, protozoan +parasites, enteric viruses including Salmonella enterica, Campylobacter spp., toxigenic +Escherichia coli, Vibrio bacteria species, Cryptosporidium and Giardia enteroviruses, +rotaviruses, noroviruses, and hepatitis A and E, are among those closely related to drinking and +recreational waters, and shellfish. These pathogens, while in environmental stages, are sensitive +to temperature, precipitation, and water flow. Several important toxin producers of water sources +are of particular public health importance in Florida (Table 4.1) including toxins from harmful +algal blooms, toxigenic marine species of Alexandrium (causing paralytic shellfish poisoning), +Karenia brevis (causing neurotoxic shellfish poisoning), and Gambierdiscus spp. (causing +ciguatera fish poisoning). Human exposure pathways are primarily through the consumption of +contaminated shellfish and fish, and in recreational waters. Cyanobacteria, consisting of multiple +species, can produce toxins (including microcystin), and the primary human exposure is from +drinking water and recreational water. For all of these harmful algal blooms species, temperature, +and precipitation are among the important factors affecting their reproduction, growth, and +distribution. + +Fig. 4.3 shows yearly distribution of total reported outbreaks and cases of food-borne and +waterborne diseases in Florida from 1989 to 2011 and Fig. 4.4 shows the monthly distribution of +reported outbreaks and cases for 2011, exhibiting marked annual variations and distinct seasonal +distribution. According to the BRACE report for the period 2004–2012, approximately 1,150 +cases of campylobacteriosis were reported annually, with the majority of cases reported between +May and September. From that same report, ~426 cases of cryptosporidiosis were reported +annually, with the majority of cases reported between June and October. Approximately 940 +cases of giardiasis were reported annually, with an annual incidence rate of 5.1 cases per 100,000 +population and the majority of cases reported between May and October. Salmonellosis +accounted for the greatest number of foodborne illnesses reported, with an average of 5,438 cases +reported each year and ~70% reported between June and November. Finally, ~101 cases of +vibriosis were reported annually, with 81% of the cases reported between April and October +(FDH 2015b). Focusing on the five diseases—campylobacteriosis. cryptosporidiosis, giardiasis, +salmonellosis, and vibriosis—the BRACE study identified a total of 1,231 follow-up days of +interest including 775 control days and 456 impact days—42 associated with hurricanes and 414 +with tropical storms (FDH 2015b); the result suggested that the occurrences of cryptosporidiosis +and salmonellosis were significantly associated with the tropical cyclones, with the risk ratio +being 1.26 (95% CI, 1.04, 1.52) and 1.35 (95% CI, 1.29, 1.42), respectively (FDH 2015b). + + g +n +i +t +i + +m +o +v + +, +a +e +h +r +r +a +i +d + +, +h +s +a +r + +, +e +h +c +a +d +a +e +h + +, +e +u +g +i +t +a +f + +, +r +e +v +e +F + +d +e +t +a +n +i +m +a +t +n +o +c + +, +g +u +b + +e +n i +m +o +t +a +i +r +t + +y +b + +e +t +i + +B + +e +s +a +e +s +i +d +s +a +g +a +h C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +1 3 6 • S O N G L I A N G E T A L . + +t +n +a +n +g +e +r +p +n +i + +e +g +a +m +a +d +l +a +t +e +f + +, +e +y +e + +, +n +i +a +r +b +n +o +e +g +a +m +a +d +, +r +e +v +e +F + +l +a +t +i +n +e +g +n +o +c + +, +) +F + +, + +D + +( + +r +e +t +a + +W + +s +i +s +o +m +s +a +l +p +o +x +o +T + +i +i +d +n +o +g +a +m +s +a +l +p +o +x +o +T + +, +e +u +g +i +t +a +f + +, +e +h +c +a +d +a +e +h +, +r +e +v +e +f + +, +s +s +e +n k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | +c +i +s + +g +n +i +p +e +e +l +s + +, +e +r +c +n +a +h +C + +a +e +s +u +a +n + +, +e +u +g +i +t +a +f + +, +s +t +a +e +w +s + +, +e +h +c +a +d +a +e +h + +, +s +l +l +i +h +c + +, +r +e +v +e +F + +l +a +t +a +f + +e +b + +n +a +c + +n +a +m +o +w + +s +o +t +i +u +q +s +o +m +y +b + +e +t +i + B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) + +a +i +r +a +l +a + M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +y +l +f + +e +s +t +e +s +t +y +b +e +t +i + +B + +s +i +s +a +i +m +o +s +o +n +a +p +y +r +t +n +a +c +i +r +f + +A + +P +P +S +a +m +o +s +o +n +a +p +y +r +T + +P +P +S +m +u +i +d +o +m +s +a +l +P + +r +o +f + +l +a +t +a +f + +d +n +a + +, +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +y +r +e +t +n +e +s +y +d +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +e +l +p +o +e +p +d +e +s +i +m +o +r +p +m +o +c +o +n +u +m +m + +i + +) + +D + +( + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +a +i +b +e +m +A + +a +c +i +t +y +l +o +t +s +i +h +a +b +e +o +m +a +t +n +E + +) + +, + +R +D + +( + +) + +, + +R +D + +( + +) + +D + +( + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +o +i +d +i +r +o +p +s +o +t +p +y +r +C + +m +u +v +r +a +p +m +u +i +d +i +r +o +p +s +o +t +p +y +r +C + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +a +i +d +r +a +i +G + +s +i +l +a +n +e +d +o +u +d + +a +i +d +r +a +i +G + +s +i +s +a +i +r +o +p +s +o +l +c +y +C + +s +i +s +n +e +n +a +t +e +y +a +c +a +r +o +p +s +o +l +c +y +C + +e +n +r +o +b +r +e +t +a + +W + +a +o +z +o +t +o +r +P + +g +n +i +p +u +o +r +G + +s + +m +o +t +p +m +y +S +& + +s +s +e +n +l +l +I +n +i +a +M + +s +y +a +w +h +t +a +P +e +r +u +s +o +p +x +E +n +i +a +M + +e +s +a +e +s +i +D + +m + +s +i +n +a +g +r +O + +n +o +i +s +s +i + +m +s +n +a +r +T + +. +s +e +s +s +e +n +l +l +i + +d +e +t +a +l +e +r + +d +n +a + +s +t +n +e +g +a + +d +e +t +a +i +c +o +s +s +a +- +r +e +t +a +w + +, +e +v +i +t +i +s +n +e +s +- +e +t +a +m + +i +l +c + +f +o + +t +s +i +l + +A + +. +1 +. +4 + +e +l +b +a +T + +e +n +o +b +d +n +a + +e +l +c +s +u +m + +, +t +n +i +o +j + +e +r +e +v +e +s + +, +e +h +c +a +d +a +e +h +e +r +e +v +e +s + +, +r +e +v +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +c +i +g +a +h +r +r +o +m +e +h +e +u +g +n +e +D + +g +n +i +d +e +e +l +b + +, +h +s +a +r + +, +n +i +a +p + +) +F +H +D + +( + +r +e +v +e +f + +s +i +t +i +g +n +i +n +e +m + +r +o + +, +s +s +e +n +i +z +z +i +d + +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +s +i +u +o +L + +. +t +S + +s +i +t +i +l +a +h +p +e +c +n +e + +s +i +u +o +L + +. +t +S + +h +s +a +r + +, +n +i +a +p +e +l +c +s +u +m + +, +g +n +i +l +l +e +w +s +d +n +a +n +i +a +p +t +n +i +o +j + +, +r +e +v +e +F + +d +n +a + +, +s +l +l +i +h +c + +, +r +e +v +e +f + +h +g +i +h + +, +e +h +c +a +d +a +e +h +, +s +n +a +m +u +h + +n +i + +e +r +a +R + +s +e +s +a +c + +e +r +e +v +e +s + +n +i + +g +n +i +t +i + +m +o +v + +e +s +i +a +l +a +m +d +n +a + +, +a +e +s +u +a +n + +o +t +i +u +q +s +o +m +y +b +e +t +i + +B + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +e +n +i +u +q +e +n +r +e +t +s +a +E + +a +y +n +u +g +n +u +k +i +h +C + +s +i +t +i +l +a +h +p +e +c +n +e + +s +u +r +i +v +a +y +n +u +g +n +u +k +i +h +C + +s +u +r +i +v + +s +i +t +i +l +a +h +p +e +c +n +e + +e +n +i +u +q +e +n +r +e +t +s +a +E + +s +u +r +i +v + +, +a +e +s +u +a +n + +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +s +o +r +C +a +L + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +s +o +r +C +a +L + +s +s +e +n +d +e +r +i +t + +d +n +a + +, +g +n +i +t +i + +m +o +v + +s +u +r +i +v + +s +i +t +i +l +a +h +p +e +c +n +e + +, +s +s +e +n +l +l +i + +e +l +i +r +b +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +e +s +a +e +s +i +d +s +u +r +i +v + +e +l +i + +N + +t +s +e + +W + +s +u +r +i +v + +e +l +i + +N + +t +s +e + +W + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +, +r +e +d +r +o +s +i +d + +t +n +e +m +e +v +o +m + +, +g +n +i +t +i + +m +o +v +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +e +n +a +p +a +J + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +e +n +a +p +a +J + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +s +e +r +u +z +i +e +s + +s +u +r +i +v + +) +s +u +o +e +n +a +t +u +c +( + +s +e +r +o +s +n +i +k +s + +, +) +l +a +r +e +c +s +i +v +( + +t +n +e +m +e +g +r +a +l +n +e + +r +e +v +i +l +/ +n +e +e +l +p +s + +, +s +s +o +l + +t +h +g +i +e +w + +, +r +e +v +e +F + +y +l +f +d +n +a +s +y +b +e +t +i + +B + +s +r +e +h +t +o +d +n +a + +, +) +F +( + +a +e +h +r +r +a +i +d +d +n +a + +g +n +i +t +i + +m +o +V + +s +i +t +i +t +a +p +e +H + +) +F + +, + +R + +, + +D + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +) +F +R + +, + +, + +D + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +s +u +o +i +r +a +V + +t +c +a +t +n +o +c + +, + +, +) +F +R +D + +, + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +a +i +n +a +m +h +s +i +e +L + +P +P +S +a +i +n +a +m +h +s +i +e +L + +s +i +t +i +t +a +p +e +h + +l +a +r +i + +V + +s +u +r +i +v +E + +, + +A +s +i +t +i +t +a +p +e +H + +s +e +s +u +r +i +V + +s +i +t +i +r +e +t +n +e +o +r +t +s +a +g + +l +a +r +i + +V + +s +u +o +i +r +a +V + +4 + +, +3 +, +2 +, +1 +V +N +E +D + +s +e +s +u +r +i +v +o +r +e +t +n +E + +s +u +r +i +v +o +r +o +N + +e +n +r +o +b +r +e +t +a + +W + +i +z +u +r +c + +. + +T + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +d +e +t +a +l +e +r +- +r +e +t +a + +W + + 7 +3 +1 + +• + +H +T +L +A +E +H + +N +A +M +U +H + +N +O + +S +T +C +A +P +M + +I + +E +G +N +A +H +C + +E +T +A +M + +I +L +C + +s + +m +o +t +p +m +y +S +& + +s +s +e +n +l +l +I +n +i +a +M + +s +y +a +w +h +t +a +P +e +r +u +s +o +p +x +E +n +i +a +M + +e +s +a +e +s +i +D + +m + +s +i +n +a +g +r +O + +n +o +i +s +s +i + +m +s +n +a +r +T + +C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 7 + +e +l +c +s +u +m + +, +a +e +s +u +a +n +, +r +e +v +e +f + +, +g +n +i +p +m +a +r +c + +l +a +n +i +m +o +d +b +a + +, +a +e +h +r +r +a +i +D + +l +a +t +a +f + +e +b +n +a +c + +, +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +e +s +i +a +l +a +m + +, +r +e +v +e +F + +s +p +m +a +r +c +h +c +a +m +o +t +s +d +n +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +s +p +m +a +r +c +h +c +a +m +o +t +s + +d +n +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +s +e +h +c +a + +e +m +o +r +d +n +y +s + +c +i +m +e +a +r +u +- +c +i +t +y +l +o +m +e +a +h + +d +n +a + +, +a +e +h +r +r +a +i +d +y +d +o +o +l +B + +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +c +i +n +o +m +u +e +n +p +g +n +i +n +e +t +a +e +r +h +t +- +e +f +i +l + +o +t + +n +o +i +s +e +l + +n +i +k +s + +d +l +i + +M + +h +t +i + +w + +t +c +a +t +n +o +c + +, +y +l +f + +r +e +e +d +r +o + +k +c +i +t + +f +o + +s +e +t +i + +B + +, +r +i +a +h +g +u +o +r +h +t + +e +b +n +a +c + +, +s +l +a +m +i +n +a +d +e +t +c +e +f +n +i + +r +e +t +a +w + +) +F +D + +, + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +p +m +a +r +c + +e +l +c +s +u +m + +, +g +n +i +t +i + +m +o +v +, +a +e +h +r +r +a +i +d + +e +r +e +v +e +S + +) +F + +, + +D + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +r +e +v +e +f + +, +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +a +e +h +r +r +a +i +D + +) + +D +s +s +e +l + +, + +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +a +i +m +e +c +i +t +p +e +s + +, +a +e +h +r +r +a +i +D + +d +n +u +o +w +n +e +p +o +, +h +s +i +f +l +l +e +h +S +d +e +t +a +n +i +m +a +t +n +o +C + +) +F +D + +, + +( + +) +F +D + +, + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +s +s +e +n +l +l +i + +l +a +e +h +r +r +a +i +D + +i +l +o +c + +. + +E +c +i +n +e +g +i +x +o +t +o +r +e +t +n +E + +d +e +t +a +n +i +m +a +t +n +o +C + +s +s +e +n +l +l +i + +l +a +e +h +r +r +a +i +D + +. + +E +c +i +g +a +h +r +r +o +m +e +a +h +o +r +e +t +n +E + +) + +R + +, + +F + +) +F + +) +F + +, + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +, + +D + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +, + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +e +l +l +e +n +o +m +l +a +S + +s +i +s +o +l +l +e +g +i +h +S + +d +i +o +h +p +y +T + +s +i +s +o +i +r +b +i +V + +a +r +e +l +o +h +C + +e +a +r +e +l +o +h +c +o +i +r +b +i +V + +P +P +S +o +i +r +b +i +V + +. +p +p +s + +a +l +l +e +n +o +m +l +a +S + +i +h +p +y +t + +a +l +l +e +n +o +m +l +a +S + +. +p +p +s + +a +l +l +e +g +i +h +S + +a +i +r +e +t +c +a +B + +s +i +s +o +i +r +e +t +c +a +b +o +l +y +p +m +a +C + +. +p +p +s + +r +e +t +c +a +b +o +l +y +p +m +a +C + +i + +g +n +p +u +o +r +G + +e +n +r +o +b +r +e +t +a + +W + +s +i +s +o +i +n +i +s +r +e +Y + +a +i +m +e +a +r +a +l +u +T + +s +i +s +n +e +r +a +l +u +t + +a +l +l +e +s +i +c +n +a +r +F + +. +p +p +s +a +i +n +i +s +r +e +Y + +i +l +o +c + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +n +i +k +s + +e +h +t + +r +e +d +n +u +s +e +l +u +d +o +n +d +n +a + +, +e +s +a +e +s +i +d +e +y +e + +, +h +s +a +r +n +i +k +S + +m +r +o +w +e +y +e + +, +s +e +u +g +i +t +a +f + +, +n +i +a +p +t +n +i +o +j + +d +n +a + +e +l +c +s +u +m + +, +y +h +c +t +I + +y +r +a +n +o +m +l +u +p +, +a +m +e +d +e +h +p +m +y +l + +, +c +i +t +a +m +o +t +p +m +y +s +a + +t +s +o +M + +n +o +i +t +a +d +r +a +t +e +r +h +t +w +o +r +g +, +l +o +o +t +s +y +r +e +t +a +w +d +n +a +y +d +o +o +l +B + +a +i +l +i +h +p +o +n +i +s +o +e + +n +e +r +d +l +i +h +c + +n +i + +n +o +i +t +a +d +r +a +t +e +r +h +t +w +o +r +g +, +t +r +o +f +m +o +c +s +i +d +l +a +n +i +m +o +d +b +A + +) +F + +, + +D + +( + +) +F +D + +, + +( + +l +a +r +o +- +l +a +c +e +F + +l +a +r +o +- +l +a +c +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +y +l +f +k +c +a +l +b +y +b +e +t +i + +B + +y +l +f +r +e +e +d +y +b +e +t +i + +B + +s +i +s +a +i +o +L + +s +u +o +i +r +a +V + +a +i +r +e +t +c +a +b +o +n +a +y +C + +a +o +l + +a +o +L + +e +a +g +l +A + +s +i +r +a +c +s +A + +s +e +d +i +o +c +i +r +b +m +u +l + +s +i +r +a +c +s +A + +s +u +n +a +c +i +r +e +m +a + +r +o +t +a +c +e +N +d +n +a + +n +o +i +t +c +e +f +n +i + +m +r +o +w +p +i +h +W + +s +i +s +a +i +r +a +l +i +f + +c +i +t +a +h +p +m +y +L + +a +r +u +i +h +c +i +r +t + +s +i +r +u +h +c +i +r +T + +i +t +f +o +r +c +n +a +b + +a +i +r +e +r +e +h +c +u +W + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +s +i +s +a +i +c +r +e +c +o +h +c +n +O + +s +u +l +u +v +l +o +v + +a +c +r +e +c +o +h +c +n +O + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +c +i +t +o +r +u +e +n + +, +e +g +a +m +a +d + +r +e +v +i +l + +, +s +s +e +n +l +l +i + +l +a +n +i +t +s +e +t +n +i + +, +s +i +t +i +t +a +m +r +e +D + +t +c +e +r +i +d +d +n +a +D +a +i +v +s +n +i +x +o +t +o +t + +e +r +u +s +o +p +x +E + +s +n +o +i +t +c +a +e +r + +t +c +a +t +n +o +c + +y +r +o +t +a +r +i +p +s +e +r + +, +a +i +s +e +h +t +s +e +r +a +p + +, +s +m +o +t +p +m +y +s + +l +a +n +i +t +s +e +t +n +i +o +r +t +s +a +G + +n +i +x +o +t +y +b +) +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +h +s +i +f +l +l +e +h +s + +c +i +x +o +t +o +r +u +e +N + +s +i +v +e +r +b + +a +i +n +e +r +a +K + +n +o +i +t +a +t +i +r +r +i + +e +y +e + +d +n +a + +g +n +i +n +o +s +i +o +p + +a +i +s +e +h +t +s +e +r +a +p + +y +t +i + +m +e +r +t +x +e + +, +e +s +i +a +l +a +m + +, +g +n +i +t +i + +m +o +v +, +a +e +h +r +r +a +i +D + +n +i +x +o +t +y +b +) +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +g +n +i +n +o +s +i +o +p +h +s +i +f + +a +r +e +t +a +u +g +i +C + +p +p +s + +s +u +c +s +i +d +r +e +i +b +m +a +G + +s +i +s +y +l +a +r +a +p + +g +n +i +n +o +s +i +o +p + +y +r +o +t +a +r +i +p +s +e +r + +, +a +i +s +e +h +t +s +e +r +a +p + +y +t +i + +m +e +r +t +x +e +d +n +a + +l +a +r +o +m +u +c +r +i + +C + +n +i +x +o +t + +y +b +) +F +( + +d +e +t +a +n +i +m +a +t +n +o +C + +h +s +i +f +l +l +e +h +s + +c +i +t +y +l +a +r +a +P + +p +p +s +m +u +i +r +d +n +a +x +e +l +A + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +. +a +d +i +r +o +l +F +n +i + +e +c +n +a +c +i +f +i +n +g +i +s +h +t +l +a +e +h +c +i +l +b +u +p +r +o +j +a +m + +f +o +s +e +s +a +e +s +i +d +/ +s +n +e +g +o +h +t +a +p + +s +e +t +a +c +i +d +n +i +d +e +r +n +i + +d +e +t +h +g +i +l +h +g +i +H + +. +2 +; +) +F +( + +s +d +o +o +f + +d +n +a +, +) +R +( + +r +e +t +a +w + +l +a +n +o +i +t +a +e +r +c +e +r + +, +) + +D + +( + +r +e +t +a +w +g +n +i +k +n +i +r +d +: +s +y +a +w +h +t +a +p +e +r +u +s +o +p +x +e + +n +i +a +m +r +o +f + +s +n +o +i +t +a +i +v +e +r +b +b +A + +. + +1 +: +e +t +o +N + +e +g +a +m +a +d + +l +a +n +i +t +s +e +t +n +i + +d +n +a +y +r +a +n +i +r +u +, +a +e +h +r +r +a +i +D + +r +e +t +a +w +h +t +i + +w + +t +c +a +t +n +o +c + +l +a +m +r +e +D + +s +i +s +a +i +m +o +s +o +t +s +i +h +c +S + +. +p +p +s + +a +m +o +s +o +t +s +i +h +c +S + +s +h +t +n +i +m +l +e +H + +a +i +m +e +n +a + +d +n +a + +, +e +u +g +i +t +a +f + +, +e +t +i +t +e +p +p +a + +f +o +s +s +o +l + +, +a +e +h +r +r +a +i +D + +l +i +o +s +h +t +i + +w + +t +c +a +t +n +o +c + +l +a +m +r +e +D + +n +o +i +t +c +e +f +n +i + +m +r +o +w +k +o +o +H + +e +l +a +n +e +d +o +u +d +a +m +o +t +s +o +l +y +c +n +A + +n +o +i +t +a +m +m +a +l +f +n +i +d +n +a + +r +e +c +l +u +l +u +f +n +i +a +P + +) + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +a +i +l +u +c +n +u +c +a +r +D + +s +i +s +n +e +n +i +d +e +m +s +u +l +u +c +n +u +c +a +r +D + +d +e +s +a +b +- +r +e +t +a + +W + + 1 3 8 • S O N G L I A N G E T A L . + +Figure 4.3. Reported outbreaks (bar) and human cases (line) of food-borne and waterborne diseases in +Florida each year from 1989 to 2011 (Source: Florida Department of Health). Toxigenic V. cholerae +infections are highly unusual in Florida and the first V. cholerae O75 outbreak was detected and reported +in Florida between March 23 and April 13, 2011. + +Figure 4.4. Monthly distribution of reported outbreaks and cases of food-borne and waterborne diseases in +Florida, 2011 (Source: Florida Department of Health) + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 9 + +c) Vector-Borne Diseases +Vector-borne diseases pose serious public health threats throughout the world. According to the +World Health Organization, vector-borne diseases account for more than 17% of infectious +diseases, causing more than one million deaths annually from diseases such as malaria, dengue +fever, yellow fever, Japanese encephalitis, and schistosomiasis (WHO 2016). Vector-borne +diseases are transmitted through vector organisms including mosquitoes, ticks, flies, fleas, and +snails. The transmission of vector-borne diseases takes two primary transmission pathways. The +first is the human-vector pathway, with transmission through a vector biting a host; pathogens +survive outside hosts in arthropod vectors and humans are typically the only host (Eisenberg et +al. 2007). Diseases in this group include malaria, dengue fever, onchocerciasis, trypanosomiasis, +and filariasis (Table 4.1). The second pathway involves zoonotic transmission, with a vector +biting a nonhuman host as the transmission route and humans as the dead-end host (Eisenberg et +al. 2007). Diseases in this group include Lyme disease, Yellow fever, West Nile virus, and +Japanese encephalitis (Table 4.1). The transmission and spread of vector-borne diseases are +determined by complex interactions between the host (either human or nonhuman), the vector +(e.g., mosquitos, flies, snails, and ticks), and various pathogens ranging from protozoans, viruses, +bacteria, and helminthes (Table 4.1). Important biological properties underlying the transmission +of vector-borne diseases include survival, development, and reproduction of the vector; also the +pathogens in the vector, the vector’s biting rate, and the behavior of hosts (human and +nonhuman), all of which are associated with climatic conditions. The following sections discuss +some specific aspects of the effects of rainfall, temperature, and flooding with regard to vector- +borne diseases. + +Temperature Sensitivity +Through the modulating effects on physiological activities and tolerance limits of vectors and +infectivity of pathogens, temperature can affect distribution of the vector and transmission of the +pathogens between vectors and hosts. Some key mechanistic effects of temperature on vectors, +vector-borne pathogens and their transmission are summarized by Gulber et al. (2001). +• Change in survival of vectors (e.g. increase or decrease depending on species) +• Change in susceptibility of vectors to pathogens +• Change in vector population growth +• Change in feeding behavior and host contact +• Change in incubation period of pathogens in vectors (e.g. decreased extrinsic incubation + +period at higher temperatures) + +• Change in pathogen replication and infectivity +• Change in transmission season + +A large number of studies on vector-borne diseases at different scales have provided +empirical evidence and predictions on the climatic factors and vector-borne disease links. For + + 1 4 0 • S O N G L I A N G E T A L . + +example, studies have shown significant relationships between ambient temperature and malaria +transmission for both mean temperature (Loevinsohn 1994; Lindblade et al. 2000; Thomson et +al. 2005; Tian et al. 2008; Wardrop et al. 2013) and its daily fluctuations (Paaijmans et al. 2010). +Laboratory studies have shown clear temperature-dependent effects on dengue virus infection in +mosquitoes (Alto and Bettinardi 2013). Fluctuations at a low mean temperature accelerate +dengue virus transmission in Aedes aegypti mosquitoes (Carrington et al. 2013). A recent +systematic review and meta-analysis based on 137 studies on the relationship between dengue +risk and temperature suggests that the dengue transmission and risk are sensitive to temperature, +with a positive relationship between them for a minimum range (18.1–24.2 °C) to a maximum +temperature range (28.0–34.5 °C) (Fan et al. 2014). A regional predictive study reported that +increased temperature increases the likelihood of many vector-borne diseases, in particular +malaria, dengue fever, and Lyme disease in new areas (Githeko et al. 2000). Similar findings are +also reported for West Nile virus (Soverow et al. 2009). These empirical findings have been used +to parameterize mechanistic models of temperature driven transmission for vector-borne +diseases, which can be coupled to projected climate scenarios to anticipate shifting geographies +of transmission risk, e.g. for malaria (Mordecai et al. 2013; Johnson et al. 2014; Johnson et al. +2015; Ryan et al. 2015; Ryan et al. 2015) and for Aedes spp transmitted diseases of concern for +Florida: dengue, chikugunya, and zika viruses (Moedecai et al. 2017; Ryan et al. 2017). + +Rainfall +Rainfall can supply water to both transient and persistent environments that can serve as habitats +or breeding sites for vectors and/or for part of pathogens’ lifecycles, critical for transmission of +vector-borne diseases. Meanwhile, rainfall can also have negative impacts on the habitats (for +example, washing away), thereby reducing the transmission risk. Variations in rainfall may have +direct and indirect impacts on the distribution and timing of transmission of vector-borne +diseases. Some key mechanistic effects of rainfall on vectors, vector-borne pathogens, and their +transmission are summarized by Gubler et al. (2001). +• +• Excess rain (e.g. causing flooding) may eliminate habitat for both vector and hosts. +• +• +• + +Increased humidity may increase vector survival and biting rate. +Increased rainfall may facilitate transport of pathogens for certain distance (Liang et al. 2007). +Increased rain may increase vertebrate host size through increased food availability (e.g. +vegetation). + +Increased rain may increase larval habitat and vector population. + +• Flooding may decrease vertebrate host size (e.g. reduced food availability) but increase + +interactions (e.g. contacts) with humans. + +Such mechanistic links have been reported by many studies including for diseases such as +malaria (van der Hoek et al. 1997; Thomson et al. 2005; Galardo et al. 2009; Gao et al. 2012; +Yang et al. 2012), dengue fever (Li et al. 1985; Su 2008; Johansson et al. 2009; Banu et al. 2011; + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 1 + +Hii et al. 2012), West Nile virus (Shaman et al. 2005; Soverow et al. 2009), and Lyme disease +(Ogden et al. 2006; Ostfeld et al. 2006). + +Vector-Borne Diseases in Florida +In the United States, 14 vector-borne diseases are of primary public health concern. Table 4.2 +lists key notifiable vector-borne diseases and reported cases between 2013 and 2015 in Florida. +Major mosquito-borne diseases include the St. Louis encephalitis virus, West Nile virus, Eastern +equine encephalitis virus, Western equine encephalitis virus, Venezuelan equine encephalitis +virus, Everglades virus, and the California serogroup viruses including La Crosse encephalitis +virus, all of which are transmitted by mosquitoes. More recently, concern about Aedes spp +transmitted flavivirus and alphaviral disesaes have become more prevalent: dengue (DENV), +chikungunya (CHIKV), and zika (ZIKV). The 2015-2016 global public health emergency for +Zika declared by the WHO focused attention on Florida as an introduction and establishment +gateway for the U.S. + +Table 4.2. A summary of reported case counts of notifiable vector-borne diseases in Florida. (Source: +Florida Department of Health). + +2013-2015 Reported Cases Number of Counties Reported + +Diseases +Tick-borne +Lyme disease +Spotted Fever Rickettsia +Anaplasmosis/Ehrlichiosis +Tularemia +Mosquito-Borne +West Nile virus +Malaria +Dengue +California serogroup viruses 2 +3 +Eastern equine encephalitis +2 +St. Louis encephalitis + +459 +74 +82 +2 + +37 +146 +331 + +46 +18 +30 +2 + +14 +24 +31 +2 +3 +1 + +St. Louis Encephalitis. The St. Louis encephalitis virus, a flavivirus, was the most common +mosquito-transmitted human pathogen in the U.S. prior to the introduction of the West Nile virus +in 1999 (FDH 2016). In Florida, the principal vector is Culex nigripalpus, a ubiquitous species +throughout the state. No human case has been reported in Florida since 2003. However, sentinel +chickens testing positive for antibodies to the St. Louis encephalitis virus have been reported in +several counties in Central and South Florida, suggesting the potential for a possible resurgence +(FDH 2014, 2016). + +West Nile virus. The peak period of transmission in Florida for the West Nile virus is July +through September. The natural cycle of West Nile virus involves Culex mosquitoes and wild +birds. Between 2001 and 2013, 318 human West Nile virus cases were reported in Florida and + + 1 4 2 • S O N G L I A N G E T A L . + +West Nile virus activities have been consistently reported in many counties around the state +(FDH 2014). + +Eastern Equine Encephalitis. Eastern equine encephalitis occurs in natural cycles involving +birds and Culiseta melanura in freshwater swampy areas, with a peak in transmission occurring +between May and August. Historical and current evidence indicates limited human Eastern +equine encephalitis epidemic potential in Florida (FDH 2014, 2016). + +Dengue Fever. In Florida, a historical dengue fever epidemic occurred in 1934-1935 and then +it ceased. It re-emerged in 2009, and since then a small number of cases have been reported each +year among individuals who had previous travel to dengue-endemic countries. In the summer of +2009, local dengue transmission was identified in Key West, Florida. Since then, sporadic local +transmission has been identified in other Central and South Florida counties during 2010, 2011, +2012, and 2013 (FDH 2014, 2016). The presence of the Aedes spp mosquito, both Aedes aegypti, +also known as the Yellow Fever mosquito, which has been a persistent invasive mosquito in +Florida for a couple of centuries, and the more recently introduced Asian Tiger Mosquito, Aedes +albopictus, maintain transmission potential in Florida. Aedes mosquitoes are capable of breeding +in as little as a teaspoon of water, and are able to adapt well to urban environments, and exhibit +transmission of dengue, chikungunya, and zika, at temperature ranges found almost year-round +in most of Florida (Mordecai et al. 2017). The optimal transmission temperature range for Aedes +aegypti, in particular, is higher than that for Aedes albopictus, suggesting that future climates +will promote Ae aegypti transmission in Florida, before it becomes too hot for transmission (Ryan +et al. 2017). + +Malaria. Malaria is caused by the mosquito-borne parasite Plasmodium falciparum. Endemic +malaria was eliminated in Florida in the late 1940s. However, imported cases—either from +travelers returning to the state from malaria-endemic regions or tourists carrying the pathogen— +have been reported. The Anopheles mosquitoes responsible for transmitting the malaria parasite +to humans are common in the state so establishment of local transmission is still possible (FDH +2014, 2016). + +Tick-Borne Diseases. The most common tick-borne diseases are ehrlichiosis, anaplasmosis, +Lyme disease, Rocky Mountain spotted fever, and other spotted fever illnesses. Lyme disease is +the most commonly reported vector-borne disease in the United States. It is caused by a bacteria +known as Borrelia burgdorferi. Lyme disease is reported in Florida year-round. An estimate +suggests that about 23% of cases were acquired in Florida and about 77% were acquired while +travelling to other states or countries (FDH 2016). Ehrlichiosis and Anaplasmosis—several +pathogenic species in the genus Ehrlichia and Anaplasma—can cause human illness. The human +illness caused by Ehrlichia chaffeensis is called Human Monocytic Ehrlichiosis, and the illness +caused by Anaplasma phagocytophilum is called Anaplasmosis or Human Granulocytotropic +Anaplasmosis. In Florida, the majority of Human Monocytic Ehrlichiosis cases (73%) are +acquired in Florida, primarily in the northern and central parts of the state, and Human +Granulocytotropic Anaplasmosis cases are even more likely to be acquired in Florida. Cases are + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 3 + +reported year-round, with a peak occurring during the spring and summer months (FDH 2016). +Rocky Mountain Spotted Fever is a disease is caused by the bacterium Rickettsia richerrsii and, +in Florida, it is transmitted primarily by the American dog tick. It can also be caused by +Dermacentor variabilis and cases of this transmission are reported year-round with more than +70% acquired in Florida, and the majority of cases are reported in the northern and central regions +of the state (FDH 2016). + +Drought +Drought is considered a meteorological anomaly characterized by a prolonged and abnormal +moisture deficiency. Climate change and variability are progressively increasing the severity and +frequency of drought events. Drought has many possible public health implications (Kalis et al. +2009; Kalis and Curtiss 2016). Drought may: +• Compromise quantity and quality of water for drinking, sanitation, and hygiene +• Compromise air quality (e.g. increased particulate matters in the air) +• Reduce crop yield, food availability and nutrition +• + +Increase or decrease risks to infectious pathogens (e.g. waterborne, air-borne, and vector- +borne) +Increase risk of non-communicable illness (e.g. respiratory infections and/or illness) + +• + +Drought has been linked to deterioration of water quality (Golladay and Battle 2002; Al- +Kharabsheh and Ta'any 2003) and air quality (Taylor and Davies 1990; Field et al. 2009) in +various settings in both the developed and developing world. Drought has been implicated for +infectious diseases—waterborne (Lipp et al. 2002; Schuster et al. 2005; Senhorst and Zwolsman +2005), vector-borne (Chretien et al. 2007; Erlanger et al. 2009; Medlock and Leach 2015), and +airborne (Polymenakou et al. 2008). Drought conditions have also been linked to increased +incidence of respiratory diseases (Smith et al. 2014). + +In Florida, the BRACE study examined associations between monthly drought conditions +(using the Standardized Precipitation Index), disease rates and emergency department visits for +specific health conditions (e.g. allergic rhinitis, asthma, and all respiratory diseases excluding +asthma) among Florida residents from 2005 to 2012. The results suggested that drought was +significantly associated with emergency department visits for respiratory diseases. It is worth +noting that the relationship between the most extreme drought conditions and emergency +department visits tended to be protective (e.g. less emergency department visits for respiratory +illness), while more moderate drought appeared to be associated with an increase in emergency +department visits for the health conditions. A similar pattern was also observed for asthma (FDH +2014b). + + 1 4 4 • S O N G L I A N G E T A L . + +Indirect Impact Pathway – Mental Health, Well-Being, +and Community Health + +Mental health disorders and well-being are an important aspect of the human health impacts +associated with climate change and variability, and associated weather events. Many mental +health consequences, ranging from simple stress and distress symptoms to some severe +conditions including anxiety, distress, posttraumatic stress disorder (PTSD), and even suicidal +tendency have been the focus of much of the research and public health responses in the past few +decades. Here are some key mental disorders and community health issues that may be related +to climate change and associated weather events, based on the U.S. Global Change Research +Program report (Crimmins et al. 2016). + +Experience with extreme weather events, such as flooding, drought, and hurricanes, may + +cause the following mental disorders: +• +• +• +• + +Increased stress, anxiety, depression, grief, and even suicidality; +Increased tensions on social stress and relationships; +Increased substance abuse; and +Increased risk of PTSD. +At the community level (Crimmins et al. 2016), potential impacts may include: +• +Increased interpersonal aggression, +• +Increased violence and crime, +• +Increased social instability, and +• Decreased community cohesion. + +Assessing the impacts of natural disasters on mental or community health and well-being of +affected populations, as well as mediation techniques, has been a focus of much research, +community and public health responses. In Florida, studies have suggested that anxiety disorder, +major depressive episodes, and PTSD among affected populations could be significantly +associated with storm exposure and displacement during the occurrence of Hurricane Andrew in +1992 (David et al. 1996), as well as a number of hurricanes in 2004 (Acierno et al. 2007) and +2012 (Neria and Shultz 2012). Studies have also indicated that social support would help to +alleviate such negative impacts (Acierno et al. 2007). Similar impacts have been reported for +community and public health workers who responded to these weather events. For example, +during the 2004 Florida hurricane season, four hurricanes (Charley, Frances, Ivan, and Jean) as +well as one tropical storm (Bonnie) hit Florida during a period of seven weeks. Increased PTSD, +other mental health issues, and substance abuse were reported among public health workers +during and after these events (Fullerton et al. 2013; Fullerton et al. 2015). The studies also +suggested that these mental health outcomes were influenced by multiple community +characteristics including the collective efficacy of neighborhood populations in the community. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 5 + +Significantly lower depressive symptoms were associated with communities that had sufficient +resources and received social support (Acierno et al. 2007, Fullerton et al. 2013, Fullerton et al. +2015). + +Impacts on children following natural disasters are of particular concern. For example, +Hurricane Andrew was a devastating category 5 Atlantic hurricane that struck South Florida in +1992. Studies showed significant impacts of hurricane exposure, stressors occurring during the +hurricane, and recovery periods on children’s persistent posttraumatic stress. In the Miami-Dade +County area, 35% to 60% of children surveyed reported moderate to very severe levels of +posttraumatic stress symptoms (Vernberg et al. 1996; La Greca et al. 2010), with hurricane- +related stressors influencing children’s persistent posttraumatic stress symptoms and other life +events in later stages of children’s life (La Greca et al. 2010; Weems et al. 2010; La Greca et al. +2013). These findings have important community health implications for identifying and +potentially helping youth in the aftermath of natural disasters. + +Conclusion + +The current available evidence has clearly indicated increases in average temperature, total +annual precipitation, frequency of extreme temperature conditions and heavy precipitation in the +United States in the past five decades, as well as increases in the tropical cyclone activity in +regions along the Atlantic Ocean, the Caribbean, and the Gulf of Mexico. Similar patterns are +expected in the future if no remediation is conducted (USEPA 2016). Florida is likely to face +even greater effects than the rest of the US, particularly effects associated with sea level rise, heat +waves, storm surge, hurricanes, and others (FOCC 2010) that affect human health in a variety of +different ways. Some effects are through direct exposure pathways, such as hurricane/storm and +temperature-related illnesses, injuries, and deaths. Some are through indirect exposure pathways +such as waterborne and vector-borne diseases. And still others are through social and economic +pathways, such as stress and mental illness. In addition to these direct and indirect effects, we +see coupled and compounded effects of climate change, which are anticipated to be exacerbated +in the future. For example, Florida is home to a large elderly population; the warm climate and +housing development capacity made this ideal. However, the elderly are more vulnerable to the +effects of heatwaves, less mobile in the event of hurricanes, and suffer greater impacts of the +symptomology of febrile vectorborne diseases. This puts the state in a position higher public +health urgency by virtue of an interaction of demographic profile and increased impact of climate +change due to geography. Given the increase in hurricances and storm impacts expected under +climate change, we are likely to also see the added impact of vectorborne disease risk anticipated +in the aftermath of such disasters. For example, the additional potential vector breeding sites +provided by debris and damaged infrastructure after hurricanes have been noted as concerns for +Zika transmission in the 2016-2017 hurricane season. An additional component of this is the + + 1 4 6 • S O N G L I A N G E T A L . + +unanticipated exposures post-disaster, in which both displaced residents and rescue workers have +radically increased outdoor exposure, reduced access to air conditioned spaces, and to basic +protection measures such as insect repellent and appropriate protective clothing. These are just +two of the compounded scenarios of public health concern that the state of Florida faces in a +changing climate. Despite the comprehensive information reviewed in the chapter, many things +are still unknown or are not well understood about the health impacts of Florida’s changing +climate. For instance, what are the impacts of sea level rise, flooding, and other extreme weather +events on water contamination that will ultimately influence human health? How will future +warming trends and other climate-sensitive hazards influence the transmission and spread of +existing and/or new waterborne and vector-borne diseases? What are the disease burdens and +their distributions attributable to known climate-sensitive hazards in the state? To address these +challenges, policies focused on mitigation, adaptation strategies, health surveillance, and +research that could close knowledge gaps on human exposures to these climate-sensitive hazards +and associated health impacts are much needed. + +Acknowledgments + +The authors thank Melissa Jordan for her instrumental discussion on this chapter. Song Liang was supported +in part by grants from NSF/EEID (EF-1015908) and NSF/WSC-Category 3 (1360330). Sadie J. Ryan was +supported in part by grants from NSF/EEID (DEB-1518681), NSF RAPID (DEB-1641145), and an Early +Career Fellowship from the Florida Climate Institute. + +References + +Acierno, R., K. J. Ruggiero, S. Galea, H. S. Resnick, K. Koenen, J. Roitzsch, M. de Arellano, J. Boyle, and +D. G. Kilpatrick. 2007. Psychological sequelae resulting from the 2004 Florida hurricanes: implications +for postdisaster intervention. Am J Public Health 97 Suppl 1:S103-108. + +Al-Kharabsheh, A., and R. Ta'any. 2003. Influence of urbanization on water quality deterioration during + +drought periods at South Jordan. Journal of Arid Environments 53:619-630. + +Alto, B. W., and D. Bettinardi. 2013. Temperature and dengue virus infection in mosquitoes: independent + +effects on the immature and adult stages. Am J Trop Med Hyg 88:497-505. + +Aslar, A. K., A. Soran, Y. Yildiz, and Y. Isik. 2001. Epidemiology, morbidity, mortality and treatment of +lightning injuries in a Turkish burns units. International Journal of Clinical Practice 55:502-504. +Balbus, J., A. Crimmins, J. L. Gamble, D. R. Easterling, K. E. Kunkel, S. Saha, and M. C. Sarofim. 2016. +Ch. 1: Introduction: Climate Change and Human Health. Pages 25–42 The Impacts of Climate Change +on Human Health in the United States: A Scientific Assessment. U.S. Global Change Research +Program, Washington, DC. + +Bandyopadhyay, S., S. Kanji, and L. M. Wang. 2012. The impact of rainfall and temperature variation on + +diarrheal prevalence in Sub-Saharan Africa. Applied Geography 33:63-72. + +Banu, S., W. Hu, C. Hurst, and S. Tong. 2011. Dengue transmission in the Asia-Pacific region: impact of + +climate change and socio-environmental factors. Trop Med Int Health 16:598-607. + +Basu, R., and J. M. Samet. 2002. Relation between elevated ambient temperature and mortality: A review + +of the epidemiologic evidence. Epidemiologic Reviews 24:190-202. + +Berko, J., D. D. Ingram, S. Saha, and J. D. Parker. 2014. Deaths attributed to heat, cold, and other weather + +events in the United States, 2006-2010. Natl Health Stat Report:1-15. + +Booij, M. J. 2005. Impact of climate change on river flooding assessed with different spatial model + +resolutions. Journal of Hydrology 303:176-198. + +Bouma, M. J., R. S. Kovats, S. A. Goubet, J. S. Cox, and A. Haines. 1997. Global assessment of El Nino's + +disaster burden. Lancet 350:1435-1438. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 7 + +Carlton, E. J., J. N. S. Eisenberg, J. Goldstick, W. Cevallos, J. Trostle, and K. Levy. 2014. Heavy Rainfall +Events and Diarrhea Incidence: The Role of Social and Environmental Factors. American Journal of +Epidemiology 179:344-352. + +Carrington, L. B., M. V. Armijos, L. Lambrechts, and T. W. Scott. 2013. Fluctuations at a Low Mean +Temperature Accelerate Dengue Virus Transmission by Aedes aegypti. Plos Neglected Tropical +Diseases 7. + +Center for Disease Control, and Prevention. 1994. Heat-related deaths--Philadelphia and United States, + +1993-1994. MMWR Morb Mortal Wkly Rep 43:453-455. + +Chen, M. J., C. Y. Lin, Y. T. Wu, P. C. Wu, S. C. Lung, and H. J. Su. 2012. Effects of extreme precipitation + +to the distribution of infectious diseases in Taiwan, 1994-2008. PLoS One 7:e34651. + +Chretien, J. P., A. Anyamba, S. A. Bedno, R. F. Breiman, R. Sang, K. Sergon, A. M. Powers, C. O. Onyango, +J. Small, C. J. Tucker, and K. J. Linthicum. 2007. Drought-associated chikungunya emergence along +coastal East Africa. Am J Trop Med Hyg 76:405-407. + +Christensen, J. H., and O. B. Christensen. 2003. Climate modelling: Severe summertime flooding in Europe. + +Nature 421:805-806. + +Chung, J. Y., Y. Honda, Y. C. Hong, X. C. Pan, Y. L. Guo, and H. Kim. 2009. Ambient temperature and +mortality: an international study in four capital cities of East Asia. Sci Total Environ 408:390-396. +Collins, T. W., A. M. Jimenez, and S. E. Grineski. 2013. Hispanic health disparities after a flood disaster: +results of a population-based survey of individuals experiencing home site damage in El Paso (Texas, +USA). J Immigr Minor Health 15:415-426. + +Colwell, R. R., P. R. Epstein, D. Gubler, N. Maynard, A. J. McMichael, J. A. Patz, R. B. Sack, and R. + +Shope. 1998. Climate change and human health. Science 279:968-969. + +Crimmins, A., J. Balbus, J. L. Gamble, C. B. Beard, J. E. Bell, D. Dodgen, R. J. Eisen, N. Fann, M. D. +Hawkins, S. C. Herring, L. Jantarasami, D. M. Mills, S. Saha, M. C. Sarofim, J. Trtanj, and L. Ziska. +2016. Executive Summary. Pages 1–24 The Impacts of Climate Change on Human Health in the United +States: A Scientific Assessment. U.S. Global Change Research Program, Washington, DC. + +Curriero, F. C., K. S. Heiner, J. M. Samet, S. L. Zeger, L. Strug, and J. A. Patz. 2002. Temperature and + +mortality in 11 cities of the eastern United States. Am J Epidemiol 155:80-87. + +Curriero, F. C., J. A. Patz, J. B. Rose, and S. Lele. 2001. The association between extreme precipitation and +waterborne disease outbreaks in the United States, 1948-1994. American Journal of Public Health +91:1194-1199. + +David, D., T. A. Mellman, L. M. Mendoza, R. Kulick-Bell, G. Ironson, and N. Schneiderman. 1996. + +Psychiatric morbidity following Hurricane Andrew. J Trauma Stress 9:607-612. + +Diez, J. J., M. D. Esteban, R. Paz, J. S. Lopez-Gutierrez, V. Negro, and J. V. Monnot. 2011. Urban Coastal + +Flooding and Climate Change. Journal of Coastal Research:205-209. + +Douglas, I., K. Alam, M. Maghenda, Y. McDonnell, L. McLean, and J. Campbell. 2008. Unjust waters: +climate change, flooding and the urban poor in Africa. Environment and Urbanization 20:187-205. +Duclos, P. J., L. M. Sanderson, and K. C. Klontz. 1990. Lightning-related mortality and morbidity in + +Florida. Public Health Rep 105:276-282. + +Eisenberg, J. N., M. A. Desai, K. Levy, S. J. Bates, S. Liang, K. Naumoff, and J. C. Scott. 2007. +Environmental determinants of infectious disease: a framework for tracking causal links and guiding +public health research. Environ Health Perspect 115:1216-1223. + +Ellis, K. N., L. M. Sylvester, and J. C. Trepanier. 2015. Spatiotemporal patterns of extreme hurricanes + +impacting US coastal cities. Natural Hazards 75:2733-2749. + +Elsner, James B., James P. Kossin, and Thomas H. Jagger. "The increasing intensity of the strongest tropical + +cyclones." Nature 455.7209 (2008): 92. + +Erlanger, T. E., S. Weiss, J. Keiser, J. Utzinger, and K. Wiedenmayer. 2009. Past, Present, and Future of + +Japanese Encephalitis. Emerging Infectious Diseases 15:1-7. + +Fan, J., W. Wei, Z. Bai, C. Fan, S. Li, Q. Liu, and K. Yang. 2014. A systematic review and meta-analysis + +of dengue risk with temperature change. Int J Environ Res Public Health 12:1-15. + +FDH. 2014a. Surveillance and control of selected mosquito-borne diseases in Florida. Florida Department + +of Health, Tallahassee, Florida. + +FDH. 2014b. Health effects of precipitation abundance and deficits in Florida. Tallahassee, Florida. +FDH. 2015a. Health effects of summer heat in Florida. Florida Department of Health, Tallahassee, Florida. +FDH. 2015b. Health effects of tropical storms and hurricanes in Florida. Florida Department of Health, + +Tallahassee, Florida. + + 1 4 8 • S O N G L I A N G E T A L . + +FDH. 2016. Diseases and Conditions, Florida Department of Health, Tallahassee, Florida. +Field, C. B., V. R. Barros, and Intergovernmental Panel on Climate Change. Working Group II. 2014. +Climate change 2014: impacts, adaptation, and vulnerability: Working Group II contribution to the fifth +assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, +New York, NY. + +Field, R. D., G. R. van der Werf, and S. S. P. Shen. 2009. Human amplification of drought-induced biomass + +burning in Indonesia since 1960. Nature Geoscience 2:185-188. + +FOCC. 2010. Climate change and sea-level rise in Florida. An update of the effects of climate change on +Florida's ocean & coastal resources. Page 36 in T. F. O. a. C. Council, editor, Tallahassee, Florida +Frumkin, H., A. J. McMichael, and J. J. Hess. 2008. Climate change and the health of the public. Am J Prev + +Med 35:401-402. + +Fullerton, C. S., J. B. A. McKibben, D. B. Reissman, T. Scharf, K. M. Kowalski-Trakofler, J. M. Shultz, +and R. J. Ursano. 2013. Posttraumatic Stress Disorder, Depression, and Alcohol and Tobacco Use in +Public Health Workers After the 2004 Florida Hurricanes. Disaster Medicine and Public Health +Preparedness 7:89-95. + +Fullerton, C. S., R. J. Ursano, X. Liu, J. B. McKibben, L. Wang, and D. B. Reissman. 2015. Depressive +Symptom Severity and Community Collective Efficacy following the 2004 Florida Hurricanes. PLoS +One 10:e0130863. + +Galardo, A. K. R., R. H. Zimmerman, L. P. Lounibos, L. J. Young, C. D. Galardo, M. Arruda, and A. A. R. +D. Couto. 2009. Seasonal abundance of anopheline mosquitoes and their association with rainfall and +malaria along the Matapi River, Amapi, Brazil. Medical and Veterinary Entomology 23:335-349. +Gao, H. W., L. P. Wang, S. Liang, Y. X. Liu, S. L. Tong, J. J. Wang, Y. P. Li, X. F. Wang, H. Yang, J. Q. +Ma, L. Q. Fang, and W. C. Cao. 2012. Change in rainfall drives malaria re-emergence in Anhui +Province, China. PLoS One 7:e43686. + +Githeko, A. K., S. W. Lindsay, U. E. Confalonieri, and J. A. Patz. 2000. Climate change and vector-borne + +diseases: a regional analysis. Bull World Health Organ 78:1136-1147. + +Golladay, S. W., and J. Battle. 2002. Effects of flooding and drought on water quality in gulf coastal plain + +streams in Georgia. Journal of Environmental Quality 31:1266-1272. + +Gubler, D. J., P. Reiter, K. L. Ebi, W. Yap, R. Nasci, and J. A. Patz. 2001. Climate variability and change +in the United States: potential impacts on vector- and rodent-borne diseases. Environ Health Perspect +109 Suppl 2:223-233. + +Hashizume, M., B. Armstrong, S. Hajat, Y. Wagatsuma, A. S. Faruque, T. Hayashi, and D. A. Sack. 2008. + +The effect of rainfall on the incidence of cholera in Bangladesh. Epidemiology 19:103-110. + +Hii, Y. L., H. P. Zhu, N. Ng, L. C. Ng, and J. Rocklov. 2012. Forecast of Dengue Incidence Using + +Temperature and Rainfall. Plos Neglected Tropical Diseases 6. + +Hoshiko, S., P. English, D. Smith, and R. Trent. 2010. A simple method for estimating excess mortality due +to heat waves, as applied to the 2006 California heat wave. International Journal of Public Health +55:133-137. + +Huynen, M. M., P. Martens, D. Schram, M. P. Weijenberg, and A. E. Kunst. 2001. The impact of heat +waves and cold spells on mortality rates in the Dutch population. Environ Health Perspect 109:463-470. +IFRCRCS. 2015. World disasters report: focus on local actors, the key to humanitarian effectiveness + +International Federation of Red Cross and Red Crescent Societies. + +Ji, F., Z. H. Wu, J. P. Huang, and E. P. Chassignet. 2014. Evolution of land surface air temperature trend. + +Nature Climate Change 4:462-466. + +Johansson, M. A., D. A. T. Cummings, and G. E. Glass. 2009. Multiyear Climate Variability and Dengue- +El Nino Southern Oscillation, Weather, and Dengue Incidence in Puerto Rico, Mexico, and Thailand: +A Longitudinal Data Analysis. Plos Medicine 6. + +Johnson, L. R., K.D. Lafferty, A. McNally, E. Mordecai, K. Paaijmans, S. Pawar, S.J. Ryan. 2014. +"Mapping the distribution of Malaria: Current Approaches and Future Directions". in Analyzing and +Modeling Spatial and Temporal Dynamics of Infectious Diseases. D. Chen, B. Moulin, J. Wu (Eds). +John Wiley & Sons. pp 189-209. + +Johnson, L.R. Ben-Horin, T., Lafferty, K.D., McNally, A., Mordecai, E., Paaijmans, K.P., Pawar, S., Ryan, +S.J. 2015. Understanding uncertainty in temperature effects on vector-borne disease: A Bayesian +approach. Ecology 96(1): 203-213. + +Kalis, M., and E. Curtiss. 2016. CDC's Drought Guidance: Your Public Health Resource for Understanding + +and Preparing for Drought in Your Community. J Environ Health 78:34-35. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 9 + +Kalis, M. A., M. D. Miller, and R. J. Wilson. 2009. Public health and drought. J Environ Health 72:10-11. +Keatinge, W. R., G. C. Donaldson, K. Bucher, G. Jendritzky, E. Cordioli, M. Martinelli, K. Katsouyanni, +A. E. Kunst, C. McDonald, S. Nayha, I. Vuori, and G. Eurowinter. 2000a. Winter mortality in relation +to climate. Int J Circumpolar Health 59:154-159. + +Keatinge, W. R., G. C. Donaldson, E. Cordioli, M. Martinelli, A. E. Kunst, J. P. Mackenbach, S. Nayha, +and I. Vuori. 2000b. Heat related mortality in warm and cold regions of Europe: observational study. +BMJ 321:670-673. + +Kirshen, P., C. Watson, E. Douglas, A. Gontz, J. Lee, and Y. Tian. 2008. Coastal flooding in the +Northeastern United States due to climate change. Mitigation and Adaptation Strategies for Global +Change 13:437-451. + +Kistemann, T., T. Classen, C. Koch, F. Dangendorf, R. Fischeder, J. Gebel, V. Vacata, and M. Exner. 2002. +Microbial load of drinking water reservoir tributaries during extreme rainfall and runoff. Appl Environ +Microbiol 68:2188-2197. + +Klinger, C., O. Landeg, and V. Murray. 2014. Power outages, extreme events and health: a systematic + +review of the literature from 2011-2012. PLoS Curr 6. + +Knight, D. B., and R. E. Davis. 2009. Contribution of tropical cyclones to extreme rainfall events in the + +southeastern United States. Journal of Geophysical Research-Atmospheres 114. + +Knowlton, K., M. Rotkin-Ellman, G. King, H. G. Margolis, D. Smith, G. Solomon, R. Trent, and P. English. +2009. The 2006 California heat wave: impacts on hospitalizations and emergency department visits. +Environ Health Perspect 117:61-67. + +Knutson, T. R., J. J. Sirutis, G. A. Vecchi, S. Garner, M. Zhao, H. S. Kim, M. Bender, R. E. Tuleya, I. M. +Held, and G. Villarini. 2013. Dynamical Downscaling Projections of Twenty-First-Century Atlantic +Hurricane Activity: CMIP3 and CMIP5 Model-Based Scenarios. Journal of Climate 26:6591-6617. +Kochtubajda, B., M. D. Flannigan, J. R. Gyakum, R. E. Stewart, K. A. Logan, and T. V. Nguyen. 2006. +Lightning and fires in the Northwest Territories and responses to future climate change. Arctic 59:211- +221. + +Kovats, R. S. 2000. El Nino and human health. Bull World Health Organ 78:1127-1135. +Kovats, R. S., M. J. Bouma, S. Hajat, E. Worrall, and A. Haines. 2003. El Nino and health. Lancet 362:1481- + +1489. + +Kovats, R. S., and S. Hajat. 2008. Heat stress and public health: a critical review. Annu Rev Public Health + +29:41-55. + +Kovats, R. S., S. Hajat, and P. Wilkinson. 2004. Contrasting patterns of mortality and hospital admissions + +during hot weather and heat waves in Greater London, UK. Occup Environ Med 61:893-898. + +La Greca, A. M., B. S. Lai, M. M. Llabre, W. K. Silverman, E. M. Vernberg, and M. J. Prinstein. 2013. +Children's Postdisaster Trajectories of PTS Symptoms: Predicting Chronic Distress. Child Youth Care +Forum 42:351-369. + +La Greca, A. M., W. K. Silverman, B. Lai, and J. Jaccard. 2010. Hurricane-Related Exposure Experiences +and Stressors, Other Life Events, and Social Support: Concurrent and Prospective Impact on Children's +Persistent Posttraumatic Stress Symptoms. Journal of Consulting and Clinical Psychology 78:794-805. +Laursen, E., O. Mygind, B. Rasmussen, and T. Ronne. 1994. Gastroenteritis: a waterborne outbreak + +affecting 1600 people in a small Danish town. J Epidemiol Community Health 48:453-458. + +Li, C. F., T. W. Lim, L. L. Han, and R. Fang. 1985. Rainfall, abundance of Aedes aegypti and dengue + +infection in Selangor, Malaysia. Southeast Asian J Trop Med Public Health 16:560-568. + +Liang, S., E. Y. Seto, J. V. Remais, B. Zhong, C. Yang, A. Hubbard, G. M. Davis, X. Gu, D. Qiu, and R. +C. Spear. 2007. Environmental effects on parasitic disease transmission exemplified by schistosomiasis +in western China. Proc Natl Acad Sci U S A 104:7110-7115. + +Lin, S., W. H. Hsu, A. R. Van Zutphen, S. Saha, G. Luber, and S. A. Hwang. 2012. Excessive heat and +respiratory hospitalizations in New York State: estimating current and future public health burden +related to climate change. Environ Health Perspect 120:1571-1577. + +Lindblade, K. A., E. D. Walker, A. W. Onapa, J. Katungu, and M. L. Wilson. 2000. Land use change alters +malaria transmission parameters by modifying temperature in a highland area of Uganda. Tropical +Medicine & International Health 5:263-274. + +Lipp, E. K., A. Huq, and R. R. Colwell. 2002. Effects of global climate on infectious disease: the cholera + +model. Clin Microbiol Rev 15:757-770. + +Loevinsohn, M. E. 1994. Climatic warming and increased malaria incidence in Rwanda. Lancet 343:714- + +718. + + 1 5 0 • S O N G L I A N G E T A L . + +Luque Fernandez, M. A., A. Bauernfeind, J. D. Jimenez, C. L. Gil, N. El Omeiri, and D. H. Guibert. 2009. +Influence of temperature and rainfall on the evolution of cholera epidemics in Lusaka, Zambia, 2003- +2006: analysis of a time series. Trans R Soc Trop Med Hyg 103:137-143. + +Lyle, T. S., and T. Mills. 2016. Assessing coastal flood risk in a changing climate for the City of Vancouver. + +Canadian Water Resources Journal 41:343-352. + +Malmstadt, J., K. Scheltin, and J. Elsner. 2009. Florida hurricanes and damage costs. Southeastern + +Geographer 49:108-131. + +Martinez, C. J., J. J. Maleski, and M. F. Miller. 2012. Trends in precipitation and temperature in Florida, + +USA. Journal of Hydrology 452:259-281. + +McMichael, A. J., and A. Haines. 1997. Global climate change: the potential effects on health. BMJ + +315:805-809. + +Medina-Ramon, M., A. Zanobetti, D. P. Cavanagh, and J. Schwartz. 2006. Extreme temperatures and +mortality: assessing effect modification by personal characteristics and specific cause of death in a +multi-city case-only analysis. Environ Health Perspect 114:1331-1336. + +Medlock, J. M., and S. A. Leach. 2015. Effect of climate change on vector-borne disease risk in the UK. + +Lancet Infect Dis 15:721-730. + +Mhalu, F. S., W. M. Mntenga, and F. D. Mtango. 1987. Seasonality of cholera in Tanzania: possible role of + +rainfall in disease transmission. East Afr Med J 64:378-387. + +Miettinen, I. T., O. Zacheus, C. H. von Bonsdorff, and T. Vartiainen. 2001. Waterborne epidemics in + +Finland in 1998-1999. Water Sci Technol 43:67-71. +Miranda, J. J. 2004. El Nino and health. Lancet 363:247-248. +Mordecai, E.A., Cohen, J., Evans, M.V., Gudapati, P., Johnson, L.R., Lippi, C.A., Miazgowicz, K., +Murdock, C.C., Rohr, J.R., Ryan, S.J., Savage, V., Shocket, M., Stewart Ibarra, A.M., Thomas, M.B., +Weikel, D.P. 2017. Detecting the impact of temperature on transmission of Zika, dengue, and +chikungunya using mechanistic models. PLOS NTDs. 11(4): e0005568 + +Mordecai, E.A., Paaijmans, K.P., Johnson, L.R., Balzer, C.H., Ben-Horin, T., deMoor, E., McNally, A., +Pawar, S., Ryan, S.J., Smith, T.C., Lafferty, K.D., 2013. Physiological constraints dramatically lower +the expected temperature for peak malaria transmission. Ecology Letters 16(1):22-30 + +Neria, Y., and J. M. Shultz. 2012. Mental health effects of Hurricane Sandy: characteristics, potential + +aftermath, and response. JAMA 308:2571-2572. + +Ngwa, M. C., S. Liang, I. T. Kracalik, L. Morris, J. K. Blackburn, L. M. Mbam, S. F. Ba Pouth, A. Teboh, +Y. Yang, M. Arabi, J. D. Sugimoto, and J. G. Morris, Jr. 2016. Cholera in Cameroon, 2000-2012: Spatial +and Temporal Analysis at the Operational (Health District) and Sub Climate Levels. PLoS Negl Trop +Dis 10:e0005105. + +Nichols, G., C. Lane, N. Asgari, N. Q. Verlander, and A. Charlett. 2009. Rainfall and outbreaks of drinking + +water related disease and in England and Wales. J Water Health 7:1-8. + +Ogden, N. H., A. Maarouf, I. K. Barker, M. Bigras-Poulin, L. R. Lindsay, M. G. Morshed, C. J. O'Callaghan, +F. Ramay, D. Waltner-Toews, and D. F. Charron. 2006. Climate change and the potential for range +expansion of the Lyme disease vector Ixodes scapularis in Canada. International Journal for +Parasitology 36:63-70. + +Ostfeld, R. S., C. D. Canham, K. Oggenfuss, R. J. Winchcombe, and F. Keesing. 2006. Climate, deer, +rodents, and acorns as determinants of variation in Lyme-disease risk. Plos Biology 4:1058-1068. +Paaijmans, K. P., S. Blanford, A. S. Bell, J. I. Blanford, A. F. Read, and M. B. Thomas. 2010. Influence of +climate on malaria transmission depends on daily temperature variation. Proc Natl Acad Sci U S A +107:15135-15139. + +Polymenakou, P. N., M. Mandalakis, E. G. Stephanou, and A. Tselepides. 2008. Particle size distribution +of airborne microorganisms and pathogens during an Intense African dust event in the eastern +Mediterranean. Environ Health Perspect 116:292-296. + +Price, C., and D. Rind. 1994. Possible Implications of Global Climate-Change on Global Lightning + +Distributions and Frequencies. Journal of Geophysical Research-Atmospheres 99:10823-10831. + +Pruss-Ustun, A., J. Bartram, T. Clasen, J. M. Colford, O. Cumming, V. Curtis, S. Bonjour, A. D. Dangour, +J. De France, L. Fewtrell, M. C. Freeman, B. Gordon, P. R. Hunter, R. B. Johnston, C. Mathers, D. +Mausezahl, K. Medlicott, M. Neira, M. Stocks, J. Wolf, and S. Cairncross. 2014. Burden of disease +from inadequate water, sanitation and hygiene in low- and middle-income settings: a retrospective +analysis of data from 145 countries. Tropical Medicine & International Health 19:894-905. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 5 1 + +Pruss, A., D. Kay, L. Fewtrell, and J. Bartram. 2002. Estimating the burden of disease from water, sanitation, + +and hygiene at a global level. Environmental Health Perspectives 110:537-542. + +Qian, Z. M., Q. C. He, H. M. Lin, L. L. Kong, C. M. Bentley, W. S. Liu, and D. J. Zhou. 2008. High +temperatures enhanced acute mortality effects of ambient particle pollution in the "Oven" city of +Wuhan, China. Environmental Health Perspectives 116:1172-1178. + +Ragan, P., J. Schulte, S. J. Nelson, and K. T. Jones. 2008. Mortality surveillance: 2004 to 2005 Florida + +hurricane-related deaths. Am J Forensic Med Pathol 29:148-153. + +Reeve, N., and R. Toumi. 1999. Lightning activity as an indicator of climate change. Quarterly Journal of + +the Royal Meteorological Society 125:893-903. + +Ritenour, A. E., M. J. Morton, J. G. McManus, D. J. Barillo, and L. C. Cancio. 2008. Lightning injury: a + +review. Burns 34:585-594. + +Ryan, S.J., McNally, A., Johnson, L.R., Mordecai, E.A., Ben-Horin, T., Paaijmans, K.P., Lafferty, K.D. +2015. Mapping physiological suitability limits of malaria in Africa under climate change. Journal of +vector borne and zoonotic diseases 15(12): 817-725. + +Ryan, S.J., Ben-Horin, T., Johnson, L.R. 2015. Malaria control and senescence: the importance of + +accounting for the pace and shape of ageing in wild mosquitoes. Ecosphere 6(9):170. + +Ryan, S.J. et al. 2017. Climate change drives uncertain global shifts in potential distribution and seasonal + +risk of Aedes-transmitted viruses. BioRXiv doi: https://doi.org/10.1101/172221. + +Sarofim, M. C., S. Saha, M. D. Hawkins, D. M. Mills, J. Hess, R. Horton, P. Kinney, J. Schwartz, and A. +St. Juliana. 2016. Ch. 2: Temperature-Related Death and Illness. Pages 43-68 The Impacts of Climate +Change on Human Health in the United States: A + +Scientific Assessment. U.S. Global Change Research Program, Washington, DC. +Schijven, J. F., and A. M. de Roda Husman. 2005. Effect of climate changes on waterborne disease in The + +Netherlands. Water Sci Technol 51:79-87. + +Schreider, S. Y., D. I. Smith, and A. J. Jakeman. 2000. Climate change impacts on urban flooding. Climatic + +Change 47:91-115. + +Schuster, C. J., A. G. Ellis, W. J. Robertson, D. F. Charron, J. J. Aramini, B. J. Marshall, and D. T. Medeiros. +2005. Infectious disease outbreaks related to drinking water in Canada, 1974-2001. Can J Public Health +96:254-258. + +Senhorst, H. A., and J. J. Zwolsman. 2005. Climate change and effects on water quality: a first impression. + +Water Sci Technol 51:53-59. + +Shaman, J., J. F. Day, and M. Stieglitz. 2005. Drought-induced amplification and epidemic transmission of + +West Nile virus in southern Florida. J Med Entomol 42:134-141. + +Singh, R. B., S. Hales, N. de Wet, R. Raj, M. Hearnden, and P. Weinstein. 2001. The influence of climate +variation and change on diarrheal disease in the Pacific Islands. Environ Health Perspect 109:155-159. +Soverow, J. E., G. A. Wellenius, D. N. Fisman, and M. A. Mittleman. 2009. Infectious Disease in a +Warming World: How Weather Influenced West Nile Virus in the United States (2001-2005). +Environmental Health Perspectives 117:1049-1052. + +Su, G. L. S. 2008. Correlation of climatic factors and dengue incidence in Metro Manila, Philippines. Ambio + +37:292-294. + +Sullivent, E. E., 3rd, C. A. West, R. S. Noe, K. E. Thomas, L. J. Wallace, and R. T. Leeb. 2006. Nonfatal + +injuries following Hurricane Katrina--New Orleans, Louisiana, 2005. J Safety Res 37:213-217. + +Taylor, G., and W. J. Davies. 1990. Root-Growth of Fagus-Sylvatica - Impact of Air-Quality and Drought + +at a Site in Southern Britain. New Phytologist 116:457-464. + +Thomas, K. M., D. F. Charron, D. Waltner-Toews, C. Schuster, A. R. Maarouf, and J. D. Holt. 2006. A role +of high impact weather events in waterborne disease outbreaks in Canada, 1975 - 2001. Int J Environ +Health Res 16:167-180. + +Thompson, K. R., N. B. Bernier, and P. Chan. 2009. Extreme sea levels, coastal flooding and climate change + +with a focus on Atlantic Canada. Natural Hazards 51:139-150. + +Thomson, M. C., S. J. Mason, T. Phindela, and S. J. Connor. 2005. Use of rainfall and sea surface +temperature monitoring for malaria early warning in Botswana. American Journal of Tropical Medicine +and Hygiene 73:214-221. + +Tian, L., Y. Bi, S. C. Ho, W. Liu, S. Liang, W. B. Goggins, E. Y. Chan, S. Zhou, and J. J. Sung. 2008. One- +year delayed effect of fog on malaria transmission: a time-series analysis in the rain forest area of +Mengla County, south-west China. Malar J 7:110. + +USEPA. 2016. Climate change indicators in the United States. Page 96 in U. S. E. P. Agency, editor. + + 1 5 2 • S O N G L I A N G E T A L . + +van der Hoek, W., F. Konradsen, D. Perera, P. H. Amerasinghe, and F. P. Amerasinghe. 1997. Correlation +between rainfall and malaria in the dry zone of Sri Lanka. Annals of Tropical Medicine and Parasitology +91:945-949. + +Vernberg, E. M., A. M. LaGreca, W. K. Silverman, and M. J. Prinstein. 1996. Prediction of posttraumatic +stress symptoms in children after Hurricane Andrew. Journal of Abnormal Psychology 105:237-248. +Wardrop, N. A., A. G. Barnett, J. A. Atkinson, and A. C. A. Clements. 2013. Plasmodium vivax malaria +incidence over time and its association with temperature and rainfall in four counties of Yunnan +Province, China. Malaria Journal 12. + +Weems, C. F., L. K. Taylor, M. F. Cannon, R. C. Marino, D. M. Romano, B. G. Scott, A. M. Perry, and V. +Triplett. 2010. Post Traumatic Stress, Context, and the Lingering Effects of the Hurricane Katrina +Disaster among Ethnic Minority Youth. Journal of Abnormal Child Psychology 38:49-56. + +WHO. 2008. Global Burden of Disease 2004 update. World Health Organization. +Winsberg, M. D. 2003. Florida weather. 2nd edition. University Press of Florida, Gainesville, Fla. +Yang, K., J. LeJeune, D. Alsdorf, B. Lu, C. K. Shum, and S. Liang. 2012. Global distribution of outbreaks + +of water-associated infectious diseases. PLoS Negl Trop Dis 6:e1483. + diff --git a/q143/random_k2/random_2.md b/q143/random_k2/random_2.md index 69d1305f3a9c9c9fb312a4e7dbd3718670cacdd2..a5f5d2f89373cf725b1140b751b41f37e641aa63 100644 --- a/q143/random_k2/random_2.md +++ b/q143/random_k2/random_2.md @@ -1,13 +1,34 @@ -3 +Published as a conference paper at ICLR 2024 + +CAN LARGE LANGUAGE MODELS INFER +CAUSATION FROM CORRELATION? + +Jiarui Liu3,∗ + +Zhijing Jin1,2,∗,‡ +Mrinmaya Sachan2 Rada Mihalcea6 Mona Diab3,‡,† Bernhard Schölkopf1,† +1Max Planck Institute for Intelligent Systems, Tübingen, Germany 2ETH Zürich +3LTI, CMU 4University of Hong Kong 5Meta AI +jinzhi@ethz.ch + +Zhiheng Lyu4 Spencer Poff5 + +zhihenglyu.cs@gmail.com + +6University of Michigan + +jiarui@cmu.edu + +4 2 0 2 r -a - -M +p +A 7 +1 ] L @@ -17,17 +38,17 @@ s c [ -4 +3 v -1 -7 -1 -1 -1 -. +6 3 +8 +5 0 -2 +. +6 +0 +3 2 : v @@ -36,2168 +57,1729 @@ X r a -Published as a conference paper at ICLR 2023 - -SELF-CONSISTENCY IMPROVES CHAIN OF THOUGHT -REASONING IN LANGUAGE MODELS - -Jason Wei† Dale Schuurmans† Quoc Le† Ed H. Chi† - -Xuezhi Wang†‡ -Sharan Narang† Aakanksha Chowdhery† Denny Zhou†§ -†Google Research, Brain Team -‡xuezhiw@google.com, §dennyzhou@google.com - ABSTRACT -Chain-of-thought prompting combined with pre-trained large language models has -achieved encouraging results on complex reasoning tasks. In this paper, we propose -a new decoding strategy, self-consistency, to replace the naive greedy decoding -used in chain-of-thought prompting. It first samples a diverse set of reasoning paths -instead of only taking the greedy one, and then selects the most consistent answer -by marginalizing out the sampled reasoning paths. Self-consistency leverages the -intuition that a complex reasoning problem typically admits multiple different ways -of thinking leading to its unique correct answer. Our extensive empirical evaluation -shows that self-consistency boosts the performance of chain-of-thought prompting -with a striking margin on a range of popular arithmetic and commonsense reasoning -benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), -StrategyQA (+6.4%) and ARC-challenge (+3.9%). +Causal inference is one of the hallmarks of human intelligence. While the field +of Causal NLP has attracted much interest in the recent years, existing causal +inference datasets in NLP primarily rely on discovering causality from empirical +knowledge (e.g., commonsense knowledge). In this work, we propose the first +benchmark dataset to test the pure causal inference skills of large language models +(LLMs). Specifically, we formulate a novel task CORR2CAUSE, which takes a +set of correlational statements and determines the causal relationship between +the variables. We curate a large-scale dataset of more than 200K samples, on +which we evaluate seventeen existing LLMs. Through our experiments, we identify +a key shortcoming of LLMs in terms of their causal inference skills, and show +that these models achieve almost close to random performance on the task. This +shortcoming is somewhat mitigated when we try to re-purpose LLMs for this skill +via finetuning, but we find that these models still fail to generalize – they can only +perform causal inference in in-distribution settings when variable names and textual +expressions used in the queries are similar to those in the training set, but fail in +out-of-distribution settings generated by perturbing these queries. CORR2CAUSE +is a challenging task for LLMs, and can be helpful in guiding future research on +improving LLMs’ pure reasoning skills and generalizability.1 1 INTRODUCTION -Although language models have demonstrated remarkable success across a range of NLP tasks, their -ability to demonstrate reasoning is often seen as a limitation, which cannot be overcome solely by -increasing model scale (Rae et al., 2021; BIG-bench collaboration, 2021, inter alia). In an effort -to address this shortcoming, Wei et al. (2022) have proposed chain-of-thought prompting, where -a language model is prompted to generate a series of short sentences that mimic the reasoning -process a person might employ in solving a task. For example, given the question “If there are 3 -cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?”, instead -of directly responding with “5”, a language model would be prompted to respond with the entire -chain-of-thought: “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars. The answer is 5.”. It has been observed that chain-of-thought prompting significantly -improves model performance across a variety of multi-step reasoning tasks (Wei et al., 2022). - -In this paper, we introduce a novel decoding strategy called self-consistency to replace the greedy -decoding strategy used in chain-of-thought prompting (Wei et al., 2022), that further improves -language models’ reasoning performance by a significant margin. Self-consistency leverages the -intuition that complex reasoning tasks typically admit multiple reasoning paths that reach a correct -answer (Stanovich & West, 2000). The more that deliberate thinking and analysis is required for a -problem (Evans, 2010), the greater the diversity of reasoning paths that can recover the answer. - -Figure 1 illustrates the self-consistency method with an example. We first prompt the language model -with chain-of-thought prompting, then instead of greedily decoding the optimal reasoning path, we -propose a “sample-and-marginalize” decoding procedure: we first sample from the language model’s -decoder to generate a diverse set of reasoning paths; each reasoning path might lead to a different -final answer, so we determine the optimal answer by marginalizing out the sampled reasoning paths -to find the most consistent answer in the final answer set. Such an approach is analogous to the -human experience that if multiple different ways of thinking lead to the same answer, one has greater -confidence that the final answer is correct. Compared to other decoding methods, self-consistency -avoids the repetitiveness and local-optimality that plague greedy decoding, while mitigating the -stochasticity of a single sampled generation. - -1 - - Published as a conference paper at ICLR 2023 - -Figure 1: The self-consistency method contains three steps: (1) prompt a language model using -chain-of-thought (CoT) prompting; (2) replace the “greedy decode” in CoT prompting by sampling -from the language model’s decoder to generate a diverse set of reasoning paths; and (3) marginalize -out the reasoning paths and aggregate by choosing the most consistent answer in the final answer set. - -Self-consistency is far simpler than prior approaches that either train an additional verifier (Cobbe -et al., 2021) or train a re-ranker given additional human annotations to improve generation quality -(Thoppilan et al., 2022). Instead, self-consistency is entirely unsupervised, works off-the-shelf with -pre-trained language models, requires no additional human annotation, and avoids any additional -training, auxiliary models or fine-tuning. Self-consistency also differs from a typical ensemble -approach where multiple models are trained and the outputs from each model are aggregated, it acts -more like a “self-ensemble” that works on top of a single language model. - -We evaluate self-consistency on a wide range of arithmetic and commonsense reasoning tasks over -four language models with varying scales: the public UL2-20B (Tay et al., 2022) and GPT-3-175B -(Brown et al., 2020), and two densely-activated decoder-only language models: LaMDA-137B -(Thoppilan et al., 2022) and PaLM-540B (Chowdhery et al., 2022). On all four language models, -self-consistency improves over chain-of-thought prompting by a striking margin across all tasks. In -particular, when used with PaLM-540B or GPT-3, self-consistency achieves new state-of-the-art levels -of performance across arithmetic reasoning tasks, including GSM8K (Cobbe et al., 2021) (+17.9% -absolute accuracy gains), SVAMP (Patel et al., 2021) (+11.0%), AQuA (Ling et al., 2017) (+12.2%), -and across commonsense reasoning tasks such as StrategyQA (Geva et al., 2021) (+6.4%) and ARC- -challenge (Clark et al., 2018) (+3.9%). In additional experiments, we show self-consistency can -robustly boost performance on NLP tasks where adding a chain-of-thought might hurt performance -compared to standard prompting (Ye & Durrett, 2022). We also show self-consistency significantly -outperforms sample-and-rank, beam search, ensemble-based approaches, and is robust to sampling -strategies and imperfect prompts. - -2 SELF-CONSISTENCY OVER DIVERSE REASONING PATHS - -A salient aspect of humanity is that people think differently. It is natural to suppose that in tasks -requiring deliberate thinking, there are likely several ways to attack the problem. We propose that -such a process can be simulated in language models via sampling from the language model’s decoder. -For instance, as shown in Figure 1, a model can generate several plausible responses to a math -question that all arrive at the same correct answer (Outputs 1 and 3). Since language models are not -perfect reasoners, the model might also produce an incorrect reasoning path or make a mistake in -one of the reasoning steps (e.g., in Output 2), but such solutions are less likely to arrive at the same -answer. That is, we hypothesize that correct reasoning processes, even if they are diverse, tend to -have greater agreement in their final answer than incorrect processes. - -We leverage this intuition by proposing the following self-consistency method. First, a language -model is prompted with a set of manually written chain-of-thought exemplars (Wei et al., 2022). Next, - -2 - -Language modelQ: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.…Q: Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder for $2 per egg. How much does she make every day?A:She has 16 - 3 - 4 = 9 eggs left. So she makes $2 * 9 = $18 per day. Sample a diverse set of reasoning pathsShe eats 3 for breakfast, so she has 16 - 3 = 13 left. Then she bakes muffins, so she has 13 - 4 = 9 eggs left. So she has 9 eggs * $2 = $18. This means she she sells the remainder for $2 * (16 - 4 - 3) = $26 per day. The answer is $18.The answer is $26.The answer is $18.The answer is $18.Marginalize out reasoning paths to aggregate final answersLanguage modelThis means she uses 3 + 4 = 7 eggs every day. She sells the remainder for $2 per egg, so in total she sells 7 * $2 = $14 per day. The answer is $14.The answer is $14.Greedy decode Published as a conference paper at ICLR 2023 - -GSM8K MultiArith +Causal inference, i.e., the ability to establish the correct causal relationships between variables or +events, is fundamental to human intelligence. There are two distinct ways this causal inference +capability can be acquired: one through empirical knowledge, e.g., we know from common sense that +touching a hot stove will get us burned; the other through pure causal reasoning, as causality can be +formally argued and reasoned about using known procedures and rules from causal inference (Spirtes +et al., 2000; Pearl, 2009; Peters et al., 2017). One example is that we have the a priori knowledge that +the correlation between A and B does not necessarily imply causality. This is a formal rule that holds +true regardless of the realizations of the variables A and B. -AQuA +With the rise of large language models (LLMs) (Radford et al., 2019; Devlin et al., 2019; Ouyang +et al., 2022; Zhang et al., 2022; OpenAI, 2023, inter alia), a crucial research question is whether +they can do causal reasoning well. Recent studies have pointed out that LLMs are “causal parrots,” +which recite the causal knowledge in the training data (Zeˇcevi´c et al., 2023). Moreover, the vast +majority of studies frame causal reasoning as a skill to navigate around empirical knowledge (Gordon +et al., 2012; Sap et al., 2019a;b; Qin et al., 2019; Bhagavatula et al., 2020), and also treat LLMs as a +knowledge base when evaluating its causal skills (Kıcıman et al., 2023; Tu et al., 2023; Xie et al., +2023). However, all the above lines of research frame causality as empirical knowledge, thus relying +heavily on the quality and the coverage of the training data, overlooking the great potential of the +formal causal reasoning skills to process correlational information to causal conclusions. -SVAMP +∗Equal contribution. †Equal supervision. ‡Work originated as a Meta AI internship project involving Zhijing, -CSQA +Mona, and Spencer. -ARC-c +1Our data is at https://huggingface.co/datasets/causalnlp/corr2cause. -Greedy decode +Our code is at https://github.com/causalNLP/corr2cause. -56.5 - -94.7 - -35.8 - -79.0 - -79.0 +1 -85.2 + Published as a conference paper at ICLR 2024 -Weighted avg (unnormalized) -Weighted avg (normalized) +Figure 1: Illustration of the motivation behind our task and dataset. -56.3 ± 0.0 -22.1 ± 0.0 +Drawing inspirations from technical studies on causal discovery (Spirtes et al., 2000; Spirtes & Zhang, +2016; Glymour et al., 2019), we formulate a novel task for NLP, correlation-to-causation inference +(CORR2CAUSE), which is an important skill for LLMs. Imagine the scenario in Figure 1, where +the training corpus does not tediously cover every causal relation, but more pervasively talk about +correlations, such as which events tend to co-occur. Learning a good CORR2CAUSE skill can enable +LLMs to draw causal relations behind the mere correlational information on the surface. For example, +several decades ago, there might be an observation that female university students tend to perform +better, but behind the correlational statistics is the causal graph that female students have to achieve +extra good performance to get into universities as the first place. -90.5 ± 0.0 -59.7 ± 0.0 +To this end, we collect the CORR2CAUSE dataset, the first dataset to test the pure causal reasoning +abilities of LLMs. All the questions in this dataset are centered around testing when it is valid +or invalid to infer causation from correlation. To systematically compose this dataset, we ground +our generalization process in the formal framework of causal discovery (Spirtes et al., 1993; 2000; +Glymour et al., 2016; Spirtes & Zhang, 2016), which provides rules about how to deduce causal +relations among variables given their statistical correlation in the observational data. We generate +more than 200K data points, and label a correlation-causation statement pair as valid if and only if +there is a bijective mapping between the statistical correlation and the underlying causality. -35.8 ± 0.0 73.0 ± 0.0 74.8 ± 0.0 82.3 ± 0.0 -15.7 ± 0.0 40.5 ± 0.0 52.1 ± 0.0 51.7 ± 0.0 +Based on our CORR2CAUSE dataset with 200K samples, we investigate two main research questions: +(1) How well do existing LLMs perform on this task? (2) Can existing LLMs be re-trained or +re-purposed on this task and obtain robust causal inference skills? Through extensive experiments, +we show empirically that none of the 17 existing LLMs we investigate perform well on this pure +causal inference task. We also show that although LLMs can demonstrate better performance after +being finetuned on the data, the causal inference skills attained by them are not robust. In summary, +our contributions are as follows: -Weighted sum (unnormalized) -Weighted sum (normalized) +1. We propose the novel task of CORR2CAUSE, to probe an aspect of LLM’s reasoning ability, -59.9 ± 0.0 -74.1 ± 0.0 +pure causal inference; -92.2 ± 0.0 -99.3 ± 0.0 +2. We compose a dataset of over 200K samples, using insights from causal discovery; +3. We evaluate the performance of 17 LLMs on our dataset, finding that all of them perform -38.2 ± 0.0 76.2 ± 0.0 76.2 ± 0.0 83.5 ± 0.0 -48.0 ± 0.0 86.8 ± 0.0 80.7 ± 0.0 88.7 ± 0.0 +poorly, close to the random baseline; -Unweighted sum (majority vote) 74.4 ± 0.1 +4. We further explored whether LLMs can learn the skill through finetuning, and find that +LLMs fail to robustly acquire this skill in out-of-distribution settings. Finally, we suggest +future work to explore more ways to enhance the pure causal inference skill in LLMs. -99.3 ± 0.0 +2 PRELIMINARIES: CAUSAL INFERENCE -48.3 ± 0.5 86.6 ± 0.1 80.7 ± 0.1 88.7 ± 0.1 +2.1 DIRECTED GRAPHICAL CAUSAL MODELS (DGCMS) -Table 1: Accuracy comparison of different answer aggregation strategies on PaLM-540B. +A directed graphical causal model (DGCM) is a commonly used representation to express the causal +relations among a set of variables. Given a set of N variables X = {X1, . . . , XN }, we can encode +the causal relations among them using a directed graph G := (X, E), where E is the set of directed +edges. Each edge ei,j ∈ E represents a causal link Xi → Xj, meaning that Xi is a direct cause of +Xj. In the context of this work, we take the common assumption of directed acyclic graphs (DAGs), +which most causal discovery methods use (Glymour et al., 2019), as graphs with cycles can make the +causal discovery process arbitrarily hard. -we sample a set of candidate outputs from the language model’s decoder, generating a diverse set of -candidate reasoning paths. Self-consistency is compatible with most existing sampling algorithms, -including temperature sampling (Ackley et al., 1985; Ficler & Goldberg, 2017), top-k sampling (Fan -et al., 2018; Holtzman et al., 2018; Radford et al., 2019), and nucleus sampling (Holtzman et al., -2020). Finally, we aggregate the answers by marginalizing out the sampled reasoning paths and -choosing the answer that is the most consistent among the generated answers. -In more detail, assume the generated answers ai are from a fixed answer set, ai ∈ A, where -i = 1, . . . , m indexes the m candidate outputs sampled from the decoder. Given a prompt and a -question, self-consistency introduces an additional latent variable ri, which is a sequence of tokens -representing the reasoning path in the i-th output, then couples the generation of (ri, ai) where -ri → ai, i.e., generating a reasoning path ri is optional and only used to reach the final answer ai. As -an example, consider Output 3 from Figure 1: the first few sentences “She eats 3 for breakfast ... So -she has 9 eggs * $2 = $18.” constitutes ri, while the answer 18 from the last sentence, “The answer -is $18”, is parsed as ai.1 After sampling multiple (ri, ai) from the model’s decoder, self-consistency -1(ai = a), -applies a marginalization over ri by taking a majority vote over ai, i.e., arg maxa -or as we defined as the most “consistent” answer among the final answer set. +2 -(cid:80)m +Previous tasks:Getting burnedTouching a hot stoveCausesNoSuppose we know that A correlates with B.Can we infer that A causes B?Skill being tested in previous work: Empiricalknowledge instead of pure causal inference.Training CorpusIce cream salesHow can LLMs process suchinformation? Correlation?Causation? What causes what?We propose a new task: Corr2Cause InferenceThis requires the skill of inferring causation from correlation*Assumption that we explicitly mention in the samples: We suppose a close system of the given variables and correlations.YesA correlates with B. C correlates with B. However, A is independent of C.Can we infer that A causes B???Drowning casesHot weatherVaccinationAutismTech company CEOsCollege dropoutFatality rateManycorrelations Published as a conference paper at ICLR 2024 + +Following the graph-theoretic terminology, we use an analogy of the ancestry tree to denote the +relations between two variables. For example, we call Xi as a parent of Xj if there is a directed edge +Xi → Xj in the graph, and, thus, Xj is a child of Xi. Similarly, we denote Xi as an ancestor of Xj +if there exists a directed path from Xi to Xj, and, thus, Xj is a descendent of Xi. Note that a parent +is a special case of an ancestor where the directed path has a length of 1. + +For convenience, we also introduce the notions for some special three-variable relations. Given two +variables Xi and Xj, we call a third variable Xk a confounder (i.e., common cause) if Xk is a parent +of both Xi and Xj; a collider (i.e., common effect) if Xk is a child of both Xi and Xj; and a mediator +if Xk is both a child of Xi, and a parent of Xj. + +2.2 D-SEPARATION AND MARKOV PROPERTY + +D-Separation D-separation (Pearl, 1988) is a fundamental concept in graphical models used to +determine whether two sets of nodes X and Y in a DAG G are conditionally independent given a +third set of nodes Z, where the three sets are disjoint. We say that X and Y are d-separated by Z if +all paths between any node in X and any node in Y are blocked by the conditioning set Z. A path +between X and Y is blocked by Z if there exists a node A ∈ Z which satisfies one of the following +conditions: A is the parent node in a fork structure on the path (i.e., · ← A → ·); A is the mediator +node in a chain structure on the path (i.e., · → A → ·); or in any collider structure on the path (i.e., +· → A ← ·), Z does not contain A or its descendants. + +Markov Property The Markov property in a DAG G states that each node Xi is conditionally +independent of its non-descendants given its parents, namely Xi ⊥⊥ NonDe(Xi)| Pa(Xi), where +NonDe(Xi) denotes the non-descendants of Xi excluding itself, and Pa(Xi) denotes the parents +of Xi. Using the Markov property, we can factorize the joint distribution of all the nodes in the +graph into P (X1, . . . , XN ) = (cid:81)N +i=1 P (Xi|PA(Xi)). To infer the causal graph from probability +distributions, a common assumption is faithfulness, namely the validity to infer all the d-separation +sets in the graph from the independence relations in the probability distribution. In our work, we also +take this broadly taken assumption which holds for most real-world scenarios. + +Markov Equivalence of Graphs We denote two DAGs as Markov equivalent if they induce the +same joint distribution P (X). The set of DAGs that are Markov equivalent to each other is called +a Markov equivalence class (MEC). Causal graphs in the same MEC can be easily identified since +they have the same skeleton (i.e., undirected edges) and V-structures (i.e., structures in the form of +A → B ← C where A and C are not connected). + +Obviously, there is a one-to-many mapping (i.e., surjection) between the causal graph and statistical +distribution. Namely, each causal graph sufficiently determines a statistical distribution, but from +a statistical distribution, we cannot necessarily induce a unique causal graph. This is why we say +“correlation does not necessarily mean causation”. + +2.3 CAUSAL DISCOVERY + +Causal discovery aims to learn the causal relations by analyzing statistical properties in the obser- +vational data (Spirtes et al., 1993; 2000; Glymour et al., 2016; Spirtes & Zhang, 2016; Glymour +et al., 2019). It can be achieved through constraint-based methods (Spirtes et al., 2000), score-based +methods (Chickering, 2002), or other methods taking advantage of the functional causal models +(Shimizu et al., 2006; Hoyer et al., 2008; Zhang & Hyvärinen, 2009). + +To fit for the spirit of this paper to infer from correlation (expressed in natural language) to causation, +we base our dataset design on the widely-used Peter-Clark (PC) algorithm (Spirtes et al., 2000). +The PC algorithm is based on the principles of conditional independence and the causal Markov +assumption, which allows it to efficiently identify causal relationships among variables in a given +dataset. The algorithm first starts with a fully connected undirected graph among all the variables. +Then it removes the edge between two variables if there is an unconditional or conditional inde- +pendence relationship between them. Afterwards, it orients the directed edges whenever there is a +V-structure. And finally, it iteratively checks the direction of the other edges until the entire causal +graph is consistent with all the statistical correlations. -i=1 +3 -In Table 1, we show the test accuracy over a set of reasoning tasks by using different answer -aggregation strategies. In addition to majority vote, one can also weight each (ri, ai) by P (ri, ai | -prompt, question) when aggregating the answers. Note to compute P (ri, ai | prompt, question), we -can either take the unnormalized probability of the model generating (ri, ai) given (prompt, question), -or we can normalize the conditional probability by the output length (Brown et al., 2020), i.e., + Published as a conference paper at ICLR 2024 -P (ri, ai | prompt, question) = exp +Figure 2: Pipeline of the data construction process. -1 -K +3 DATASET CONSTRUCTION -(cid:80)K +We introduce the construction of our dataset in this section. We start with our task formulation for +CORR2CAUSE, and then briefly give an overview of the data generation process, followed by detailed +descriptions of each step. We conclude the section with the overall statistics of the dataset. -k=1 log P (tk|prompt,question,t1,...,tk−1), +3.1 TASK FORMULATION +Given a set of N variables X = {X1, . . . , XN }, we have a statement s about all the correlations +among the variables, and a hypothesis h describing the causal relation r between the pair of variables +Xi and Xj. The task is to learn a function f : (s, h) (cid:55)→ v which maps the correlation statement +s and the causal relation hypothesis h to their validity v ∈ {0, 1}, which takes the value 0 if this +inference is invalid, and the value 1 if this inference is valid. -(1) +3.2 OVERVIEW OF THE DATA GENERATION PROCESS +We base the construction our dataset on several concepts of causal inference, including the DGCM, +d-separation, and MECs, as introduced in Section 2. -where log P (tk | prompt, question, t1, . . . , tk−1) is the log probability of generating the k-th token -tk in (ri, ai) conditioned on the previous tokens, and K is the total number of tokens in (ri, ai). -In Table 1, we show that taking the “unweighted sum”, i.e., taking a majority vote directly over ai -yields a very similar accuracy as aggregating using the “normalized weighted sum”. We took a closer -look at the model’s output probabilities and found this is because for each (ri, ai), the normalized -conditional probabilities P (ri, ai | prompt, question) are quite close to each other, i.e., the language -model regards those generations as “similarly likely”.2 Additionally, when aggregating the answers, -the results in Table 1 show that the “normalized” weighted sum (i.e., Equation 1) yields a much -higher accuracy compared to its unnormalized counterpart. For completeness, in Table 1 we also -report the results by taking a “weighted average”, i.e., each a gets a score of its weighted sum divided -by (cid:80)m -i=1 +As in the overview of our data generation process in Figure 2, we first choose the number N of +variables (Step 1) and generate all the unique DGCMs with N nodes (Step 2), which we will introduce +in the Section 3.3. Then we collect all the d-separation sets from these graphs to identify MECs +(Step 3) in Section 3.4. Then, in Step 4, we create the formal form of data in Section 3.5. For each +correspondence of the MEC to causal graphs, we compose the correlation statement based on the +statistical relations in the MEC, and hypothesize a causal relation between two variables, and produce +the validity v = 1 if the hypothesis is a shared property of all causal graphs in the MEC, and v = 0 if +the hypothesis is not necessarily true for all the MEC graphs. Finally, we introduce the verbalization +process in Section 3.6. -1(ai = a), which results in a much worse performance. +3.3 CONSTRUCTING THE GRAPHS WITH ISOMORPHISM CHECKS +The first step of the data generation is to compose the causal graphs, as in Step 1 and 2 of Figure 2. +For a set of N variables X = {X1, . . . , XN }, there are N (N − 1) possible directed edges, since +each node can link to any node other than itself. To remove cycles in the graph, we make the nodes +in topological order, which only allows edges Xi → Xj, where i < j. We achieve this by limiting +the adjacency matrix of the graph to only having non-zero values above the diagonal, resulting in +N (N − 1)/2 possible directed edges for the DAGs. -Self-consistency explores an interesting space between open-ended text generation and optimal -text generation with a fixed answer. Reasoning tasks typically have fixed answers, which is why -researchers have generally considered greedy decoding approaches (Radford et al., 2019; Wei et al., -2022; Chowdhery et al., 2022). However, we have found that even when the desired answer is fixed, -introducing diversity in the reasoning processes can be highly beneficial; therefore we leverage +At the first glance, for N nodes, there should be 2N (N −1)/2 possible DAGs (i.e., the power set of all +edges). However, there could be isomorphic graphs in this set. To avoid this, we perform a graph -1The parser is task dependent. For arithmetic reasoning, we parse the first numerical part as the final answer -after the model generates “The answer is ”. For commonsense reasoning, we parse the full string answer as the -final answer after the model generates “The answer is ”. Most generated outputs have a consistent format of -“{Reasoning paths}. The answer is X.” if we prompt the language model in this format. +4 -2This also means that the language model is not well calibrated and thus cannot distinguish well between -correct solutions and wrong solutions, which also explains why additional re-rankers were trained to better judge -the quality of the solutions in previous work (Cobbe et al., 2021; Thoppilan et al., 2022). +ABC1. Choose thenumber of variables2. Generate all unique causalgraphsBACE.g., N=3ABC...3. Map each graph to a set ofstatistical correlationsACB● A ⫫̸ B,● A ⫫̸ C, ● B ⫫̸ C, and● A ⫫ C | B● A ⫫̸ B,● B ⫫̸ C, and● A ⫫ C Causal GraphsCorrelationsABC1-to-1 MappingMany-to-1MappingCorrelationsHypothesizedCausationValiditySuppose there is a closed system of 3 variables, A, B and C. All the statisticalrelations among these 3 variables are as follows:A correlates with C. B correlates with C. However, A is independent of B.A directly causes B.[The validity label is equivalent to the results after running the PC algorithm.I.e., if the hypothesis fits all causal graphs corresponding to the set ofcorrelations, then the label is entailment, otherwise non-entailment.]ValidVerbalize the statistical correlationsHypothesize a causal relation between two nodes4. Compose the Data Published as a conference paper at ICLR 2024 +# Nodes +2 3 - - Published as a conference paper at ICLR 2023 - -sampling, as commonly used for open-ended text generation (Radford et al., 2019; Brown et al., 2020; -Thoppilan et al., 2022), to achieve this goal. One should note that self-consistency can be applied -only to problems where the final answer is from a fixed answer set, but in principle this approach can -be extended to open-text generation problems if a good metric of consistency can be defined between -multiple generations, e.g., whether two answers agree or contradict each other. - -3 EXPERIMENTS - -We conducted a series of experiments to compare the proposed self-consistency method with existing -approaches on a range of reasoning benchmarks. We find that self-consistency robustly improves -reasoning accuracy for every language model considered, spanning a wide range of model scales. - -3.1 EXPERIMENT SETUP - -Tasks and datasets. We evaluate self-consistency on the following reasoning benchmarks.3 - -• Arithmetic reasoning. For these tasks, we used the Math Word Problem Repository (Koncel- -Kedziorski et al., 2016), including AddSub (Hosseini et al., 2014), MultiArith (Roy & Roth, -2015), and ASDiv (Miao et al., 2020). We also included AQUA-RAT (Ling et al., 2017), a -recently published benchmark of grade-school-math problems (GSM8K; Cobbe et al., 2021), -and a challenge dataset over math word problems (SVAMP; Patel et al., 2021). - -• Commonsense reasoning. For these tasks, we used CommonsenseQA (Talmor et al., 2019), -StrategyQA (Geva et al., 2021), and the AI2 Reasoning Challenge (ARC) (Clark et al., 2018). -• Symbolic Reasoning. We evaluate two symbolic reasoning tasks: last letter concatenation (e.g., -the input is “Elon Musk” and the output should be “nk”), and Coinflip (e.g., a coin is heads-up, -after a few flips is the coin still heads-up?) from Wei et al. (2022). - -Language models and prompts. We evaluate self-consistency over four transformer-based lan- -guage models with varying scales: - -• UL2 (Tay et al., 2022) is an encoder-decoder model trained on a mixture of denoisers with 20- -billion parameters. UL2 is completely open-sourced4 and has similar or better performance than -GPT-3 on zero-shot SuperGLUE, with only 20B parameters and thus is more compute-friendly; -• GPT-3 (Brown et al., 2020) with 175-billion parameters. We use two public engines code-davinci- -001 and code-davinci-002 from the Codex series (Chen et al., 2021) to aid reproducibility;5 -• LaMDA-137B (Thoppilan et al., 2022) is a dense left-to-right, decoder-only language model with -137-billion parameters, pre-trained on a mixture of web documents, dialog data and Wikipedia; -• PaLM-540B (Chowdhery et al., 2022) is a dense left-to-right, decoder-only language model with -540-billion parameters, pre-trained on a high quality corpus of 780 billion tokens with filtered -webpages, books, Wikipedia, news articles, source code, and social media conversations. -We perform all experiments in the few-shot setting, without training or fine-tuning the language -models. For a fair comparison we use the same prompts as in Wei et al. (2022): for all arithmetic -reasoning tasks we use the same set of 8 manually written exemplars; for each commonsense -reasoning task, 4-7 exemplars are randomly chosen from the training set with manually composed -chain-of-thought prompts.6 Full details on the prompts used are given in Appendix A.3. - -Sampling scheme. To sample diverse reasoning paths, we followed similar settings to those -suggested in Radford et al. (2019); Holtzman et al. (2020) for open-text generation. In particular, for -UL2-20B and LaMDA-137B we applied temperature sampling with T = 0.5 and truncated at the -top-k (k = 40) tokens with the highest probability, for PaLM-540B we applied T = 0.7, k = 40, and -for GPT-3 we use T = 0.7 without top-k truncation. We provide an ablation study in Section 3.5 to -show that self-consistency is generally robust to sampling strategies and parameters. - -3By default we use the test split for all datasets if the labels are available for evaluation. For CommonsenseQA -we use the dev split; for StrategyQA we use the question-only set from BIG-bench collaboration (2021): -https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/strategyqa. - -4Model checkpoints at https://github.com/google-research/google-research/tree/master/ul2. -5Public API available at https://openai.com/api/. -6Self-consistency is robust to different sets of prompts and we provide a study in Appendix A.1.2. - 4 - - Published as a conference paper at ICLR 2023 - -3.2 MAIN RESULTS - -We report the results of self-consistency averaged over 10 runs, where we sampled 40 outputs -independently from the decoder in each run. The baseline we compare to is chain-of-thought -prompting with greedy decoding (Wei et al., 2022), referred to as CoT-prompting, which has been -previously used for decoding in large language models (Chowdhery et al., 2022). - -Arithmetic Reasoning The results are shown in Table 2.7 Self-consistency improves the arithmetic -reasoning performance over all four language models significantly over chain-of-thought prompting. -More surprisingly, the gains become more significant when the language model’s scale increases, -e.g., we see +3%-6% absolute accuracy improvement over UL2-20B but +9%-23% for LaMDA- -137B and GPT-3. For larger models that already achieve high accuracy on most tasks (e.g., GPT-3 -and PaLM-540B), self-consistency still contributes significant additional gains with +12%-18% -absolute accuracy on tasks like AQuA and GSM8K, and +7%-11% on SVAMP and ASDiv. With -self-consistency, we achieve new state-of-the-art results on almost all tasks: despite the fact that self- -consistency is unsupervised and task-agnostic, these results compare favorably to existing approaches -that require task-specific training, or fine-tuning with thousands of examples (e.g., on GSM8K). - -Method - -AddSub MultiArith ASDiv - -AQuA - -SVAMP - -GSM8K - -Previous SoTA 94.9a - -60.5a - -75.3b - -37.9c - -57.4d - -35e / 55g - -UL2-20B - -CoT-prompting -Self-consistency 24.8 (+6.6) - -18.2 - -10.7 -15.0 (+4.3) - -16.9 -21.5 (+4.6) 26.9 (+3.3) - -23.6 - -12.6 -19.4 (+6.8) - -4.1 -7.3 (+3.2) - -LaMDA-137B - -PaLM-540B - -GPT-3 -Code-davinci-001 - -CoT-prompting -51.8 -Self-consistency 63.5 (+10.6) 75.7 (+23.9) 58.2 (+9.2) 26.8 (+9.1) - -49.0 - -52.9 - -17.7 - -38.9 -53.3 (+14.4) 27.7 (+10.6) - -17.1 - -CoT-prompting -Self-consistency 93.7 (+1.8) - -91.9 - -94.7 -99.3 (+4.6) - -74.0 -81.9 (+7.9) 48.3 (+12.5) 86.6 (+7.6) - -79.0 - -35.8 - -56.5 -74.4 (+17.9) - -CoT-prompting -59.5 -Self-consistency 67.8 (+10.6) 82.7 (+23.2) 61.9 (+9.2) 25.6 (+6.7) - -52.7 - -57.2 - -18.9 - -39.8 -54.5 (+14.7) 23.4 (+8.8) - -14.6 - -GPT-3 -Code-davinci-002 - -CoT-prompting -Self-consistency 91.6 (+2.2) - -89.4 - -96.2 -100.0 (+3.8) 87.8 (+7.6) 52.0 (+12.2) 86.8 (+11.0) 78.0 (+17.9) - -80.1 - -75.8 - -60.1 - -39.8 - -Table 2: Arithmetic reasoning accuracy by self-consistency compared to chain-of-thought prompting -(Wei et al., 2022). The previous SoTA baselines are obtained from: a: Relevance and LCA operation -classifier (Roy & Roth, 2015), b: Lan et al. (2021), c: Amini et al. (2019), d: Pi et al. (2022), e: -GPT-3 175B finetuned with 7.5k examples (Cobbe et al., 2021), g: GPT-3 175B finetuned plus an -additional 175B verifier (Cobbe et al., 2021). The best performance for each task is shown in bold. - -Method - -CSQA - -StrategyQA ARC-e - -ARC-c - -Letter (4) Coinflip (4) - -Previous SoTA 91.2a - -73.9b - -86.4c - -75.0c - -N/A - -N/A - -UL2-20B - -51.4 -CoT-prompting -Self-consistency 55.7 (+4.3) 54.9 (+1.6) - -53.3 - -61.6 -69.8 (+8.2) 49.5 (+6.8) - -42.9 - -0.0 -0.0 (+0.0) - -50.4 -50.5 (+0.1) - -LaMDA-137B - -CoT-prompting -57.9 -Self-consistency 63.1 (+5.2) 67.8 (+2.4) - -65.4 - -75.3 -79.3 (+4.0) 59.8 (+4.7) - -55.1 - -8.2 -8.2 (+0.0) - -72.4 -73.5 (+1.1) - -PaLM-540B - -79.0 -CoT-prompting -Self-consistency 80.7 (+1.7) 81.6 (+6.3) - -75.3 - -95.3 -96.4 (+1.1) 88.7 (+3.5) - -85.2 - -65.8 -70.8 (+5.0) 91.2 (+3.0) - -88.2 - -GPT-3 -Code-davinci-001 - -CoT-prompting -46.6 -Self-consistency 54.9 (+8.3) 61.7 (+5.0) - -56.7 - -63.1 -72.1 (+9.0) 53.7 (+10.6) 10.0 (+2.2) 75.9 (+4.5) - -43.1 - -71.4 - -7.8 - -GPT-3 -Code-davinci-002 - -CoT-prompting -79.0 -Self-consistency 81.5 (+2.5) 79.8 (+6.4) - -73.4 - -94.0 -96.0 (+2.0) 87.5 (+3.9) - -83.6 - -70.4 -73.4 (+3.0) 99.5 (+0.5) - -99.0 - -Table 3: Commonsense and symbolic reasoning accuracy by self-consistency compared to chain- -of-thought prompting (Wei et al., 2022). The previous SoTA baselines are obtained from: a: -DeBERTaV3-large + KEAR (Xu et al., 2021b), b: Chowdhery et al. (2022), c: UnifiedQA-FT -(Khashabi et al., 2020). The best performance for each task is shown in bold. - -7The standard deviation of self-consistency is ≤ 0.5 for all tasks and is thus omitted in the table. Please refer - -to Figure 2, Figure 7 and 8 for the standard deviations under varying numbers of sampled paths. - 5 - - Published as a conference paper at ICLR 2023 - -Commonsense and Symbolic Reasoning Table 3 shows the results on commonsense and symbolic -reasoning tasks. Similarly, self-consistency yields large gains across all four language models, and -obtained SoTA results on 5 out of 6 tasks. For symbolic reasoning, we test the out-of-distribution -(OOD) setting where the input prompt contains examples of 2-letters or 2-flips but we test examples -of 4-letters and 4-flips (this setting is more challenging as PaLM-540B or GPT-3 can already achieve -perfect in-distribution accuracy). In this challenging OOD setting, the gain of self-consistency is still -quite significant compared to CoT-prompting with sufficient model sizes. - -To show the effect of the number of sampled reasoning paths, we plot the accuracy (mean and -standard deviation over 10 runs) with respect to varying numbers of sampled paths (1, 5, 10, 20, 40) -in Figure 2. The results show that sampling a higher number (e.g., 40) of reasoning paths leads to a -consistently better performance, further emphasizing the importance of introducing diversity in the -reasoning paths. In Table 4, we show self-consistency yields a richer set of reasoning paths compared -to greedy decoding with a few example questions from two tasks. - -Figure 2: Self-consistency (blue) significantly improves accuracy over CoT-prompting with greedy -decoding (orange) across arithmetic and commonsense reasoning tasks, over LaMDA-137B. Sampling -a higher number of diverse reasoning paths consistently improves reasoning accuracy. - -GSM8K - -StrategyQA - -Question Henry made two stops during his 60-mile bike trip. He first stopped after 20 -miles. His second stop was 15 miles before the end of the trip. How many miles -did he travel between his first and second stops? - -Yes or no: Is Albany, Georgia the most populous -US Albany? - -Greedy -Decode - -He traveled 60 miles in total. He stopped after 20 miles, so he traveled 60 - 20 -= 40 miles between the first and second stops. The answer is 40. - -Albany, Georgia is the most populous US Al- -bany. Thus, the answer is yes. - -Sampled -Path 1 - -Sampled -Path 2 - -Henry travelled a total of 60 miles. His first stop was 20 miles in, and his -second stop was 15 miles before the end. So between his first and second stops -he travelled 60 - 20 -15 = 25 miles. The answer is 25. - -The most populous US Albany is Albany, New -York. Thus, Albany, Georgia is not the most -populous US Albany. So the answer is no. - -He made two stops during a 60-mile trip. The first was 20 miles into the trip. -The second was 15 miles before the end of the trip. This means the second stop -was 60 - 15 = 45 miles into the trip. Since he made the stops in order, the second -stop must have been 45 - 20 = 25 miles after the first stop. The answer is 25. - -Albany, Georgia has a population of about -88,000. Albany, New York has a population of -about 95,000. Thus, Albany, Georgia is not the -most populous US Albany. So the answer is no. - -Table 4: Examples where self-consistency helps repair the errors over greedy decode, on PaLM-540B. -Two sampled reasoning paths that are consistent with the ground truth are shown. - -3.3 SELF-CONSISTENCY HELPS WHEN CHAIN-OF-THOUGHT HURTS PERFORMANCE - -Ye & Durrett (2022) show that sometimes chain-of-thought prompting could hurt performance -compared to standard prompting in few-shot in-context learning. Here we perform a study using -self-consistency to see if it can help fill in the gap, over a set of common NLP tasks, including (1) -Closed-Book Question Answering: BoolQ (Clark et al., 2019), HotpotQA (Yang et al., 2018), and -(2) Natural Language Inference: e-SNLI (Camburu et al., 2018), ANLI (Nie et al., 2020) and RTE -(Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Bentivogli et al., 2009). - -The results over PaLM-540B are shown in Table 5. For some tasks (e.g., ANLI-R1, e-SNLI, RTE), -adding chain-of-thought does hurt performance compared to standard prompting (Brown et al., 2020), -but self-consistency is able to robustly boost the performance and outperform standard prompting, -making it a reliable way to add rationales in few-shot in-context learning for common NLP tasks. - -ANLI R1 / R2 / R3 e-SNLI RTE BoolQ HotpotQA (EM/F1) - -Standard-prompting (no-rationale) -CoT-prompting (Wei et al., 2022) - -69.1 / 55.8 / 55.8 -68.8 / 58.9 / 60.6 - -Self-consistency - -78.5 / 64.5 / 63.4 - -85.8 -81.0 - -88.4 - -84.8 -79.1 - -86.3 - -71.3 -74.2 - -78.4 - -27.1 / 36.8 -28.9 / 39.8 - -33.8 / 44.6 - -Table 5: Compare Standard/CoT prompting with self-consistency on common NLP tasks. - 6 +Total -0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths56586062Commonsense QA0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -3.4 COMPARE TO OTHER EXISTING APPROACHES - -We conduct a set of additional studies and show that self-consistency significantly outperforms -existing methods including sample-and-rank, beam search, and ensemble-based approaches. - -Comparison to Sample-and-Rank One commonly used approach to improve generation quality is -sample-and-rank, where multiple sequences are sampled from the decoder and then ranked according -to each sequence’s log probability (Adiwardana et al., 2020). We compare self-consistency with -sample-and-rank on GPT-3 code-davinci-001, by sampling the same number of sequences from the -decoder as self-consistency and taking the final answer from the top-ranked sequence. The results are -shown in Figure 3. While sample-and-rank does improve the accuracy with additionally sampled -sequences and ranking, the gain is much smaller compared to self-consistency. - -Figure 3: Self-consistency significantly outperforms sample-and-rank with the same # of samples. - -Comparison to Beam Search In Table 6, we compare self-consistency with beam search decoding -on the UL2-20B model. For a fair comparison we report the accuracy under the same number of -beams and reasoning paths. On both tasks self-consistency outperforms beam search significantly. -Note self-consistency can also adopt beam search to decode each reasoning path (results are shown -as “Self-consistency using beam search”), but its performance is worse compared to self-consistency -with sampling. The reason is that beam search yields a lower diversity in the outputs (Li & Jurafsky, -2016), while in self-consistency the diversity of the reasoning paths is the key to a better performance. - -Beam size / Self-consistency paths - -1 - -AQuA +# Unique DAGs +2 out of 2 +6 out of 23 +31 out of 26 +302 out of 210 +5,984 out of 215 +6,325 -MultiArith +# Edges/DAG # MECs -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -23.6 -23.6 - -10.7 -10.7 - -19.8 ± 0.3 21.2 ± 0.7 24.6 ± 0.4 24.2 ± 0.5 -19.7 ± 2.5 24.9 ± 2.6 25.3 ± 1.8 26.7 ± 1.0 26.9 ± 0.5 - -12.0 - -11.3 - -11.0 - -10.5 - -11.8 ± 0.0 11.4 ± 0.1 12.3 ± 0.1 10.8 ± 0.1 -9.5 ± 1.2 11.3 ± 1.2 12.3 ± 0.8 13.7 ± 0.9 14.7 ± 0.3 +0.50 +1.67 +3.48 +5.89 +8.77 +8.60 +2 5 - -19.3 - -10 - -16.1 - 20 +142 +2,207 +2,376 + +# DAGs/MEC +1.0 +1.2 +1.55 +2.13 +2.71 +2.66 + +Table 1: Statistics about the source causal graphs in our dataset. Given the number of nodes, we report +the number of unique DAGs, average number of edges per DAG, number of MECs, and average +number of DAGs per MEC. + +isomorphism check (McKay & Piperno, 2014), and reduce the set so that only unique DAGs are +retained, and we show their statistics in Table 1. Although we can handle large graphs, we mostly +focus on smaller graphs that can still lead to a reasonably sized dataset, so we empirically set N = 6, +but future work can use our open-sourced codes to extend to more nodes. + +3.4 PROGRAMMATICALLY GENERATING THE D-SEPARATION SETS +Based on the set of unique DAGs, we then programmatically generate the d-separation sets by +graph theoretical conditions, as in Step 3 of Figure 2. To realize this step, we code an efficient +graph-theoretic algorithm to check for all the chain, fork, and collider structures to automatically +identify the set of nodes that d-separate each pair of nodes. Using the d-separation sets and the +faithfulness assumption, we form the statistical correlations as follows. For each pair of nodes, they +are conditionally independent given the variables in the d-separation set. If the d-separation set is +empty, then the two nodes are unconditionally independent. If no d-separation set can be found for +the two nodes, then they are directly correlated. + +Moreover, using the d-separation sets, we are able to cluster causal graphs to MECs. We achieve it by +tracing the mapping between the causal graphs and the set of statistical correlations, and backtracking +the graphs with the same d-separation sets to group them in the same MEC. We show in Table 1 that +each MEC contains on average 2.66 DAGs. + +3.5 COMPOSING THE HYPOTHESES AND LABEL +After generating the set of correlations based on the d-separation sets, we now generate the causal +hypotheses. For the causal relation r, we focus on six common causal relations between two nodes +introduced in Section 2.1: Is-Parent, Is-Child, Is-Ancestor (excluding the parents), Is-Descendant +(excluding the children), Has-Confounder (i.e., there exists a confounder, or common cause, of the +two nodes), and Has-Collider (i.e., there exists a collider, or common effect, of the two nodes). In this +way, the set of hypotheses contains all six meaningful causal relations between every pair of variables, +resulting in a total size of 6 · N (N − 1)/2 = 3N (N − 1) hypotheses for a graph with N variables. + +To generate the ground-truth validity label, we start from the correlation sets in Step 3, then look up +all the causal graphs in the same MEC corresponding to the given set of correlations, and check the +necessity of the hypothesized causal relation. If the causal relationship proposed in the hypothesis +is valid for all causal graphs within the MEC, then we generate the validity v = 1; otherwise, we +generate v = 0. A special case of valid samples is that when the size of the MEC is 1, then there is a +bijective mapping between the causal graph and the d-separation sets, so any hypothesis stating the +causal properties of that unique causal graph is valid. + +3.6 VERBALIZING INTO LANGUAGE +Finally, as in the last step of Figure 2, we convert all the information above to text data for our +CORR2CAUSE task. For the correlation statement, we verbalize the set of correlations in Step 3 into +a natural language statement s. When two variables cannot be d-separated, i.e., A ̸⊥⊥ B, then we +describe them as “A correlates with B” since they are directly correlated and cannot be independent +by any condition. And if two variables have a valid d-separation set C, then we describe them as “A +is independent of B given C.” In the special case when the d-separation set is empty, we directly +say “A is independent of B.” In addition, we disambiguate the setting by starting the correlation +statement with the setup of a closed system of the given variables, and no hidden variables: “Suppose +there is a closed system of N variables, A, B, . . . All the statistical relations among these N variables +are as follows:”. Finally, to verbalize the hypothesis, we feed the causal relation triplet (Xi, r, Xj) -15.0 +5 -40 + Published as a conference paper at ICLR 2024 + +Causal Relation +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +Hypothesis Template +{Var i} directly causes {Var j}. +{Var i} causes something else which causes {Var j}. +{Var j} directly causes {Var i}. +{Var j} is a cause for {Var i}, but not a direct one. +There exists at least one collider (i.e., common effect) of {Var i} and {Var j}. +There exists at least one confounder (i.e., common cause) of {Var i} and {Var j}. + +Table 2: Templates for each causal relation in the hypothesis. We use {Var i} and {Var j} as +placeholders for the two variables. + +into their hypothesis templates in Table 2. For example, we turn the triplet (A, Is-Parent, B) into “A +directly causes B”, as in the example of Figure 2. + +3.7 STATISTICS OF THE RESULTING DATA +We show the statistics of our CORR2CAUSE dataset in Table 3. Overall, our dataset contains 207,972 +samples, where 18.57% of the samples have the positive label (i.e., with validity=1). The average +length of the premise is 424.11 tokens, and hypothesis 10.83 tokens. We split the data into 205,734 +training samples, 1,076 development and 1,162 test samples.2 Since the main purpose of the dataset +is to benchmark the performance of LLMs, we prioritize the test and development sets to have a +comprehensive coverage over all sizes of graphs. Specifically, we iterate through the subset of our +data for each N , and split it entirely for only the test and development sets if the data is less than 1K, +which is the case for N = 2 and 3. For the other subsets that are larger, we randomly sample up to +1K or 10% of the data, whichever is smaller, to the test and development sets. We set the cap to be 1K +in order to form a reasonable computation budget, since many LLMs are expensive to query in the +inference mode. Aside from the test and valid sets, all the rest of the data goes into the training set. + +# Samples +# Test +# Dev +# Train + +# Tokens/Premise +# Tokens/Hypothesis +% Positive Labels +Vocab Size + +Overall + +207,972 +1,162 +1,076 +205,734 +424.11 +10.83 +18.57 +65 + +Statistics by the Number of Nodes N +N = 2 N = 3 N = 4 N = 5 N = 6 +198,630 +720 +522 +72 +474 +72 +197,634 +576 +434.54 +104.0 +10.83 +10.83 +18.85 +7.50 +61 +55 + +8,520 +514 +482 +7,524 +212.61 +10.83 +13.01 +57 + +90 +48 +42 +0 +52.0 +10.83 +3.33 +53 -10.2 +12 +6 +6 +0 +31.5 +10.83 +0.00 +49 + +Table 3: Statistics of our CORR2CAUSE dataset, and by subsets. We report the total number of +samples (# Samples); splits of the test (# Test), developement (# Dev) and training sets (# Train); +number of tokens per premise (# Tokens/Premise) and hypothesis (# Tokens/Hypothesis); percentage +of the positive labels (% Positive Labels), and vocabulary size by the number of unique tokens (Vocab +Size). Note that the number of unique graphs and MECs are in Table 1. +4 EXPERIMENTS + +4.1 EXPERIMENTAL SETUP +We set up a diverse list of LLMs for the experiments on our CORR2CAUSE dataset. To test existing +LLMs, we first include six commonly used BERT-based NLI models in the transformers library (Wolf +et al., 2020): BERT (Devlin et al., 2019), RoBERTa (Liu et al., 2019), BART (Lewis et al., 2020), +DeBERTa (He et al., 2021), DistilBERT (Sanh et al., 2019), and DistilBART (Shleifer & Rush, 2020). +Apart from these BERT-based NLI models, we also evaluate the general-purpose autoregressive +LLMs based on GPT (Radford et al., 2019): GPT-3 Ada, Babbage, Curie, Davinci (Brown et al., +2020); its instruction-tuned versions (Ouyang et al., 2022), text-davinci-001, text-davinci-002, and +text-davinci-003; and GPT-3.5 (i.e., ChatGPT), and the latest GPT-4 (OpenAI, 2023) by April 2023, + +2Note for our dataset v2.0: We notice that our original data (v1.0) has duplication due to symmetric relations +and verbalizations of the hypothesis. E.g., Is-Parent(A, B) has the exact hypothesis verbalization as Is-Child(B, +A). Hence, for our data v2.0, we perform a careful de-duplication, and update the data statistics in Table 3. See +more version comparison details in Appendix D. Note that, due to the symmetry, the current version is a random +sample half of the size of the original version, so the modeling results in the experiment section roughly hold. -Table 6: Compare self-consistency with beam search decoding on the UL2-20B model. +6 -Comparison to Ensemble-based Approaches We further compare self-consistency to ensemble- -based methods for few-shot learning. In particular, we consider ensembling by: (1) prompt order -permutation: we randomly permute the exemplars in the prompt 40 times to mitigate model’s -sensitivity to prompt order (Zhao et al., 2021; Lu et al., 2021); and (2) multiple sets of prompts -(Gao et al., 2021): we manually write 3 different sets of prompts. We took majority vote of the -answers from greedy decoding in both approaches as an ensemble. Table 7 shows that compared to -self-consistency, existing ensemble-based approaches achieve a much smaller gain.8 In addition, note -that self-consistency is different from a typical model-ensemble approach, where multiple models -are trained and their outputs are aggregated. Self-consistency acts more like a “self-ensemble” on -top of a single language model. We additionally show the results of ensembling multiple models in -Appendix A.1.3 where the model-ensembles perform much worse compared to self-consistency. + Published as a conference paper at ICLR 2024 -GSM8K MultiArith +Random Baselines +Always Majority +Random (Proportional) +Random (Uniform) +BERT-Based Models -SVAMP +BERT MNLI +RoBERTa MNLI +DeBERTa MNLI +DistilBERT MNLI +DistilBART MNLI +BART MNLI -ARC-e +LLaMa-Based Models -ARC-c +LLaMa-7B +Alpaca-7B -CoT (Wei et al., 2022) -Ensemble (3 sets of prompts) -Ensemble (40 prompt permutations) -Self-Consistency (40 sampled paths) +GPT-Based Models -17.1 -18.6 ± 0.5 -19.2 ± 0.1 -27.7 ± 0.2 +GPT-3 Ada +GPT-3 Babbage +GPT-3 Curie +GPT-3 Davinci +GPT-3 Instruct (text-davinci-001) +GPT-3 Instruct (text-davinci-002) +GPT-3 Instruct (text-davinci-003) +GPT-3.5 +GPT-4 -51.8 -57.1 ± 0.7 -60.9 ± 0.2 -75.7 ± 0.3 +F1 -38.9 -42.1 ± 0.6 -42.7 ± 0.1 -53.3 ± 0.2 +Precision -75.3 -76.6 ± 0.1 -76.9 ± 0.1 -79.3 ± 0.3 +Recall Accuracy -55.1 -57.0 ± 0.2 -57.0 ± 0.1 -59.8 ± 0.2 +0.0 +13.5 +20.38 + +2.82 +22.79 +14.52 +20.70 +26.74 +33.38 + +26.81 +27.37 + +0.00 +27.45 +26.43 +27.82 +17.99 +21.87 +15.72 +21.69 +29.08 -Table 7: Self-consistency outperforms prompt-order and multi-prompt ensembles on LaMDA-137B. +0.0 +12.53 +15.11 + +7.23 +34.73 +14.71 +24.12 +15.92 +31.59 + +15.50 +15.93 + +0.00 +15.96 +15.23 +16.57 +11.84 +13.46 +13.4 +17.79 +20.92 -8Self-consistency is compatible with both ensemble approaches and we show the results in Appendix A.1.4. +0.0 +14.62 +31.29 + +1.75 +16.96 +14.33 +18.13 +83.63 +35.38 + +99.42 +97.37 + +0.00 +97.95 +100.00 +86.55 +37.43 +58.19 +19.01 +27.78 +47.66 + +84.77 +71.46 +62.78 + +81.61 +82.50 +74.31 +78.85 +30.23 +78.50 + +17.36 +21.33 + +84.77 +21.15 +15.23 +31.61 +48.04 +36.69 +68.97 +69.46 +64.60 + +Table 4: Overall performance. We report F1 (main metric), precision, recall and accuracy. For the +main metric, F1 score, we use the bold font to highlight the overall best performance, and underline +to highlight the best performance within each category of models. + +using the OpenAI API (https://openai.com/api/) with temperature 0. We also evaluate the +recent, more efficient models, LLaMa (Touvron et al., 2023) and Alpaca (Taori et al., 2023). + +When inspecting the behavior of finetuned models, we adopt a large set of models, including GPT- +based models (GPT-3 Ada, Babbage, Curie, and Davinci) using the OpenAI finetuning API for +classification at https://platform.openai.com/docs/guides/fine-tuning, open-sourced +decoder-only models (GPT2, GPT2-Large, GPT2-XL, LLaMA-7B, and LLaMA2-7B), BERT-based +models from scratch (BERT-Base, BERT-Large, RoBERTa-Base, and RoBERTa-Large), and BERT- +Based NLI models (BERT-Base MNLI, BERT-Large MNLI, RoBERTa-Base MNLI, and RoBERTa- +Large MNLI) using the transformers library (Wolf et al., 2020). See training details in Appendix A. + +For the random baselines, we provide “always majority” to predict the majority class 100% of the time, +“random (uniform)” to uniformly sample a label (i.e., 50% for each), and “random (proportional)” to +sample a label from a Bernouli distribution proportional to the development set label distribution. + +4.2 THE CORR2CAUSE SKILL IN EXISTING LLMS + +We show the performance of seventeen LLMs in Table 4. We can see that pure causal inference is a +very challenging task across all existing LLMs. Among all the LLMs, the best performance is 33.38% +F1 by BART MNLI, which is even higher than the latest GPT-based model, GPT-4. Notably, many +models are worse than random guess, which means that they totally fail at this pure causal inference +task. The observation still holds for few-shot chain-of-thought prompts tested in Appendix G. + +4.3 FINETUNED PERFORMANCE + +Next, we address the question: Can we re-purpose LLMs to learn this task? The experimental results +in Table 5a of 17 models finetuned on our CORR2CAUSE seem very strong at first sight. Most models +see a substantial increase, among which the finetuned BERT-based NLI models demonstrate the +strongest performance. The best-performing one, RoBERTa-Large MNLI, achieves 94.74% F1 score +on this task, as well as very high precision, recall and accuracy scores. 7 -0510152025303540#Sampled Reasoning Paths12141618202224Accuracy (%)GSM8K0510152025303540#Sampled Reasoning Paths50556065707580Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths303540455055Accuracy (%)ARC (Challenge)Self Consistency (Multi-path)Sample & Rank (Multi-path)Greedy Decode (Single-path) Published as a conference paper at ICLR 2023 - -3.5 ADDITIONAL STUDIES - -We conducted a number of additional experiments to analyze different aspects of the self-consistency -method, including its robustness to sampling strategies and parameters, and how it works with -imperfect prompts and non-natural-language reasoning paths. - -Self-Consistency is Robust to Sampling Strategies and Scaling We show self-consistency is -robust to sampling strategies and parameters, by varying T in temperature sampling (Ackley et al., -1985; Ficler & Goldberg, 2017), k in top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford -et al., 2019), and p in nucleus sampling (Holtzman et al., 2020), over PaLM-540B in Figure 4 (left). -Figure 4 (right) shows that self-consistency robustly improves performance across all scales for the -LaMDA-137B model series. The gain is relatively lower for smaller models due to certain abilities -(e.g., arithmetic) only emerge when the model reaches a sufficient scale (Brown et al., 2020). - -Figure 4: GSM8K accuracy. (Left) Self-consistency is robust to various sampling strategies and -parameters. (Right) Self-consistency improves performance across language model scales. - -Self-Consistency Improves Robustness to Imperfect Prompts For few-shot learning with man- -ually constructed prompts, human annotators sometimes make minor mistakes when creating the -prompts. We further study if self-consistency can help improve a language model’s robustness to -imperfect prompts.9 We show the results in Table 8: while imperfect prompts decrease accuracy with -greedy decoding (17.1 → 14.9), self-consistency can fill in the gaps and robustly improve the results. - -Additionally, we found that the consistency (in terms of % of decodes agreeing with the final -aggregated answer) is highly correlated with accuracy (Figure 5, over GSM8K). This suggests that -one can use self-consistency to provide an uncertainty estimate of the model in its generated solutions. -In other words, one can use low consistency as an indicator that the model has low confidence; i.e., -self-consistency confers some ability for the model to “know when it doesn’t know”. - -Prompt with correct chain-of-thought - -17.1 - -LaMDA-137B - -Prompt with imperfect chain-of-thought 14.9 -23.4 - -+ Self-consistency (40 paths) - -Prompt with equations - -+ Self-consistency (40 paths) - -PaLM-540B - -Zero-shot CoT (Kojima et al., 2022) - -+ Self-consistency (40 paths) - -5.0 -6.5 - -43.0 -69.2 - -Table 8: Self-consistency works under imperfect prompts, equa- -tion prompts and zero-shot chain-of-thought for GSM8K. - -Figure 5: The consistency is cor- -related with model’s accuracy. - -Self-Consistency Works for Non-Natural-Language Reasoning Paths and Zero-shot CoT We -also tested the generality of the self-consistency concept to alternative forms of intermediate reasoning -like equations (e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there -are 3 + 2 = 5 cars.” to “3 + 2 = 5”). The results are shown in Table 8 (“Prompt with equations”): -self-consistency still improves accuracy by generating intermediate equations; however, compared to -generating natural language reasoning paths, the gain is smaller since the equations are much shorter -and less opportunity remains for generating diversity in the decoding process. In addition, we tested -self-consistency with zero-shot chain-of-thought (Kojima et al., 2022) and show that self-consistency -works for zero-shot CoT as well and improves the results significantly (+26.2%) in Table 8. - -9We use the same prompts as before, but swap all the numbers in the reasoning paths with random numbers -except the final answer, e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars.” to “There are 7 cars in the parking lot already. 6 more arrive. Now there are 7 + 6 = 5 cars.”. + Published as a conference paper at ICLR 2024 + +F1 + +Precison Recall Accuracy + +F1 (Paraph.) + +F1 (Var. Ref.) + +92.11 +88.60 +88.60 +91.52 + +79.85 +78.19 +81.23 +85.52 + +70.47 +69.98 +75.00 +80.26 + +Finetuned GPT-Based Models Using OpenAI API +GPT-3 Ada +GPT-3 Babbage +GPT-3 Curie +GPT-3 Davinci +Finetuned Open-Sourced Decoder-Only Models +89.18 +GPT2 +94.29 +GPT2-Large +GPT2-XL +94.30 +91.98 +LLaMA-7B +LLaMA2-7B +92.92 +Finetuned BERT-Based Models +69.29 +BERT-Base +85.26 +BERT-Large +87.60 +RoBERTa-Base +RoBERTa-Large +89.10 +Finetuned BERT-Based NLI Models +BERT-Base MNLI +BERT-Large MNLI +RoBERTa-Base MNLI +RoBERTa-Large MNLI + +88.03 +92.18 +91.94 +88.62 +90.11 + +54.42 +77.51 +78.47 +82.54 + +85.49 +84.44 +90.35 +92.24 + +89.88 +90.19 +94.27 +94.74 + +90.35 +96.49 +96.78 +95.61 +95.91 + +95.32 +94.74 +99.12 +96.78 + +94.74 +96.78 +98.54 +97.37 + +92.92 +92.48 +93.77 +95.28 + +96.66 +98.22 +98.22 +97.46 +97.77 + +87.13 +95.01 +95.73 +96.39 + +86.51 +96.79 +98.17 +98.35 + +61.73 +62.34 +64.93 +65.01 + +56.76 +55.95 +60.32 +56.41 +52.24 + +61.13 +63.64 +65.58 +65.05 + +65.56 +67.24 +57.42 +55.45 + +41.57 +43.28 +45.32 +46.96 + +31.70 +31.99 +43.95 +53.92 +49.47 + +35.20 +38.54 +53.12 +60.20 + +31.50 +52.04 +62.83 +67.87 + +(a) Performance of finetuned models on the original test set. + +(b) F1 scores of finetuned models +on the perturbed test sets by para- +phrasing (Paraph.) and variable +refactorization (Var. Ref.). +Table 5: Performance of finetuned models on the original test set and perturbed test sets. + +Relation Type +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +F1 +96.18 +93.94 +95.73 +96.55 +92.19 +98.67 + +Precision Recall Accuracy +96.92 +93.94 +96.56 +100 +97.52 +100 + +98.67 +98.93 +98.67 +99.47 +94.64 +99.73 + +95.45 +93.94 +94.92 +93.33 +87.41 +97.37 + +F1 +74.80 +45.45 +73.39 +29.41 +70.70 +70.42 + +Precision Recall Accuracy +70.77 +30.30 +68.97 +17.86 +66.90 +67.57 + +79.31 +90.91 +78.43 +83.33 +75.00 +73.53 + +91.73 +93.60 +92.27 +93.60 +82.04 +94.37 + +(a) Fine-grained performance of RoBERTa-Large by causal relation +type on the original test set. + +(b) Its fine-grained performance by relation +type after variable refactorization. + +Table 6: Fine-grained analysis of the best-performing model, RoBERTa-Large MNLI. + +4.4 FINE-GRAINED PERFORMANCE BY CAUSAL RELATION +In addition to the overall results mentioned above, we conduct a fine-grained analysis to check the +performance of the strongest finetuned model, RoBERTa-Large MNLI, by our six causal relation +types. As in Table 6a, the model is very good at judging relations such as Is-Parent, Is-Descendant +and Has-Confounder, all with more than 96% F1 scores, whereas it is several points weaker on +the Has-Collider relations. This could be due to that the collider relation is the most special type, +requiring identification of the V-structure based on both the unconditional independence based on the +two variables only and correlations whenever conditioned on a common descendant. We also conduct +error analysis for non-finetuned models in Appendix F. + +4.5 ROBUSTNESS ANALYSIS +Looking at the very high performance of the finetuned models, we raise the next question: Did the +models really robustly learn the causal inference skills? + +Two Robustness Tests We design two simple robustness tests: (1) paraphrasing, and (2) variable +refactorization. For (1) paraphrasing, we simply paraphrase the hypothesis by changing the text +template for each causal relation to some semantically-equivalent alternatives in Appendix C. For (2) +variable refactorization, we reverse the alphabet of the variable names, namely flipping A, B, C, to Z, +Y, X and so on. The inspiration behind the two robustness tests comes from the spurious correlation +analysis described in Appendix E. 8 -0510152025303540#Sampled Reasoning Paths444852566064687276Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.7, k=20T=0.7, no top kp=0.95p=0.9Greedy Decode125102050100200Model size (#param in billions)510152025Accuracy (%)Self ConsistencyGreedy Decode020406080100Consistency (%)020406080100Accuracy (%) Published as a conference paper at ICLR 2023 - -4 RELATED WORK - -Reasoning in language models. Language models are known to struggle in Type 2 tasks, such as -arithmetic, logical and commonsense reasoning (Evans, 2010). Previous work has primarily focused -on specialized approaches for improving reasoning (Andor et al., 2019; Ran et al., 2019; Geva et al., -2020; Pi˛ekos et al., 2021). Compared to prior work, self-consistency is applicable to a wide range of -reasoning tasks without any additional supervision or fine-tuning, while still substantially improving -the performance of the chain-of-thought prompting approach proposed in Wei et al. (2022). -Sampling and re-ranking in language models. Multiple decoding strategies for language models -have been proposed in the literature, e.g., temperature sampling (Ackley et al., 1985; Ficler & -Goldberg, 2017), top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford et al., 2019), -nucleus sampling (Holtzman et al., 2020), minimum Bayes risk decoding (Eikema & Aziz, 2020; Shi -et al., 2022), and typical decoding (Meister et al., 2022). Other work has sought to explicitly promote -diversity in the decoding process (Batra et al., 2012; Li et al., 2016; Vijayakumar et al., 2018). - -Re-ranking is another common approach to improve generation quality in language models (Adiwar- -dana et al., 2020; Shen et al., 2021). Thoppilan et al. (2022) collect additional human annotations -to train a re-ranker for response filtering. Cobbe et al. (2021) train a “verifier” to re-rank generated -solutions, which substantially improves the solve rate on math tasks compared to just fine-tuning the -language model. Elazar et al. (2021) improve the consistency of factual knowledge extraction by -extending pre-training with an additional consistency loss. All these methods require either training -an additional re-ranker or collecting additional human annotation, while self-consistency requires no -additional training, fine-tuning, nor extra data collection. -Extract reasoning paths. Some previous work has considered task-specific approaches for iden- -tifying reasoning paths, such as constructing semantic graphs (Xu et al., 2021a), learning an RNN -to retrieve reasoning paths over the Wikipedia graph (Asai et al., 2020), fine-tuning with human -annotated reasoning paths on math problems (Cobbe et al., 2021), or training an extractor with -heuristic-based pseudo reasoning paths (Chen et al., 2019). More recently, the importance of di- -versity in the reasoning processes has been noticed, but only leveraged via task-specific training, -either through an additional QA model over extracted reasoning paths (Chen et al., 2019), or by the -introduction of latent variables in a commonsense knowledge graph (Yu et al., 2022). Compared to -these approaches, self-consistency is far simpler and requires no additional training. The approach -we propose simply couples the generation of reasoning paths and a final answer by sampling from -the decoder, using aggregation to recover the most consistent answer without additional modules. -Consistency in language models. Some prior work has shown that language models can suffer -from inconsistency in conversation (Adiwardana et al., 2020), explanation generation (Camburu et al., -2020), and factual knowledge extraction (Elazar et al., 2021). Welleck et al. (2020) use “consistency” -to refer to generating an infinite-length sequence in recurrent language models. Nye et al. (2021) -improve the logical consistency of samples from a System 1 model by adding a System 2-inspired -logical reasoning module. In this paper we focus on a slightly different notion of “consistency”, i.e., -utilizing answer consistency among diverse reasoning paths to improve accuracy. - -5 CONCLUSION AND DISCUSSION - -We introduced a simple yet effective method called self-consistency, and observed that it significantly -improves accuracy in a range of arithmetic and commonsense reasoning tasks, across four large -language models with varying scales. Beyond accuracy gains, self-consistency is also useful for -collecting rationales when performing reasoning tasks with language models, and for providing -uncertainty estimates and improved calibration of language model outputs. - -One limitation of self-consistency is that it incurs more computation cost. In practice people can try a -small number of paths (e.g., 5 or 10) as a starting point to realize most of the gains while not incurring -too much cost, as in most cases the performance saturates quickly (Figure 2). As part of future work, -one could use self-consistency to generate better supervised data to fine-tune the model, such that the -model can give more accurate predictions in a single inference run after fine-tuning. In addition, we -observed that language models can sometimes generate incorrect or nonsensical reasoning paths (e.g., -the StrategyQA example in Table 4, the two population numbers are not exactly correct), and further -work is needed to better ground models’ rationale generations. + Published as a conference paper at ICLR 2024 + +Specifically, we adopt the common setup of text adversarial attack (Morris et al., 2020; Jin et al., +2020) to preserve the training set and keep the same saved models, but run the inference on the +perturbed test set. In this way, we separate the possibilities of the models only overfitting on the +training data vs. mastering the reasoning skills. + +Results after Perturbation We can see from Table 5b that all the models drop drastically, by up to +39.29 on the paraphrased test set, and up to 62.30 after variable refactorization. The best-performing +model, RoBERTa-Large MNLI, is especially sensitive towards paraphrasing, demonstrating the +most drop among all models; however, it is the most robust against the variable refactorization, +maintaining a high F1 score of 67.87. We conduct fine-grained analysis for RoBERTa-Large MNLI +under perturbation in Table 6b. We can see the the main source of the performance drop of the model +comes from the two classes, Is-Ancestor (decreasing to 45.45%) and Is-Descendant (decreasing to +29.41%), while the other classes stay relatively robust, keeping their F1 scores over 70%. + +From this analysis, we make the following suggestions to future studies testing this CORR2CAUSE +skill of LLMs. First, it is safe to use it as a test set to benchmark existing LLMs’ performance, since +the data we generate is out-of-distribution from the training data of the current LLMs. Then, when +testing finetuned models, it is very important to accompany adversarial attack together with the i.i.d. +test set. We open-source our perturbed test sets for future work to test the generalizability skill. + +4.6 EXTENSION TO NATURAL STORIES +We envision our CORR2CAUSE dataset to be a foundation for future extensions to various settings, +such as instantiating the variables with actual phenomena and situating the story in a more natural +setting. For example, the correlation does not imply causation rule can be instantiated with the ice +cream sales and swimming pool attendance as the two variables, and argue that ice cream sales does +not necessarily affect swimming pool attendance, because their correlation could be due to a third +variable, such as hot weather. We provide a case study for how to instantiate the symbolic expressions +in our dataset to more natural stories, and find that LLMs such as GPT-4 can generate realistic, daily +life stories that has foreseeably broad applications. See more details in Appendix B. + +5 RELATED WORK + +Existing Causal Reasoning Tasks A large body of existing research of causal reasoning in NLP +focuses on leveraging empirical knowledge to do tasks such as inferring the cause and effect of why +an agent perform certain tasks (Sap et al., 2019a), the motivation and emotional reaction in a social +context (Sap et al., 2019b), how people achieve a given goal with a set of concrete steps (Zhang et al., +2020), the development of a story given a different beginning (Qin et al., 2019), and how in general +LLMs serve as a knowledge base of cause and effect (Willig et al., 2023; Kıcıman et al., 2023). In +contrast, our CORR2CAUSE task focuses on the pure causal inference skill of models, which is a +knowledge-dependent reasoning skill based on formally correct rules from causal inference. + +Existing Logical and Inference Tasks Another related area of literature is logical and inference +tasks, of which a well-established one is natural language inference (NLI), to identify the semantic +relationship between a pair of sentences (MacCartney & Manning, 2008; Bowman et al., 2015). NLI +datasets mainly focus on the set and paraphrase relations. For example, “a group of boys are playing +football” can entail “some guys are playing football,” where “boys” are a sub-concept of “guys,” and +“a group of” and “some” are paraphrases. Recently, there have been increasing efforts to extend the +inference task to various logical inference skills such as deductive logic and propaganda techniques +(Jin et al., 2022; Alhindi et al., 2022). Our CORR2CAUSE dataset is the first dataset testing the +correlation-to-causation inference skill, which is unique of its type. + +6 CONCLUSION + +In this work, we introduced a novel task, CORR2CAUSE, to infer causation from correlation, and +collected a large-scale dataset of over 200K samples. We evaluated an extensive list of LLMs on this +new task, and showed that off-the-shelf LLMs perform poorly on this task. We also show that it is +possible to re-purpose LLMs on this task by finetuning, but future work needs to be aware of the +out-of-distribution generalization problem. To avoid the Goodhart’s law, we recommend using this +dataset to benchmark the pure causal inference skills for LLMs that have not seen this dataset. Given +the limited reasoning abilities of current LLMs, and the difficulty of separating actual reasoning from +training-corpus-derived knowledge, it is imperative that our community focus on work aiming to +accurately disentangle and measure both abilities. We believe the present work is a first such step. 9 - Published as a conference paper at ICLR 2023 + Published as a conference paper at ICLR 2024 -REPRODUCIBILITY STATEMENT +LIMITATIONS AND FUTURE WORK -In experiments, we included four different language models with varying scales. Two of them are pub- -lic models: UL2 is a completely open-sourced model with model checkpoints available at https:// -github.com/google-research/google-research/tree/master/ul2; GPT-3 is -also a public model with public API available at https://openai.com/api/. For GPT-3, -we have included two public engines (“code-davinci-001” and “code-davinci-002”) to further aid -reproducibility, as Codex is currently free so anyone can reproduce the results. In addition, as our -results make use of LaMDA-137B and PaLM-540B that are not publicly available, we provide the -exact input prompts for all tasks in Appendix A.3 (and note that we do not perform any finetuning -and only apply prompting to off-the-shelf language models). +We identify several limitations of this work and open future directions: First, in the context of this +work, we limit the causal graphs to two to six nodes, but future work can feel free to explore larger +graphs. Another aspect is that we do not assume hidden confounders in this inference problem, so we +welcome future work to generate an even more challenging dataset to infer the existence of hidden +confounders, analogous to the causal discovery algorithm of fast causal inference (FCI) (Spirtes et al., +2000). And also in general, explorations of other causal discovery algorithms are welcomed too. +Finally, a lot of motivation behind proposing this task is inspired by the problem of invalid reasoning +patterns in our daily reasoning (Jin et al., 2022), which could fertilize the ground for more pervasive +spread of fake news. We believe false causal inference is a prevalent type of fallacious beliefs, and +welcome future work to connect the idea of this benchmark to more real-world false beliefs based on +confusing correlation with causation. -ETHICS STATEMENT +ACKNOWLEDGMENT -As we stated in the discussion, language models can sometimes generate nonsensical or non-factual -reasoning paths, so one should use language models’ outputs with extra caution. We deal with -reasoning tasks mostly and the generated rationales are only used for inspecting how a model reaches -its answer. One could potentially use the generated rationales to further check why the model makes -certain mistakes or whether the model contains any biases when performing a certain task. For -language model in real-world use, further work is needed to better ground models’ predictions and -improve model’s factuality and safety, to ensure the models do not cause harms to users. +We thank Riley Goodside for valuable suggestions to improve our prompts to LLMs. We thank Luigi +Gresele and Amir Hossein Karimi for their suggestions to help us improve the formulation of our +causal discovery questions. -REFERENCES - -David H. Ackley, Geoffrey E. Hinton, and Terrence J. Sejnowski. A learning algorithm for boltzmann -machines. Cognitive Science, 9(1):147–169, 1985. ISSN 0364-0213. URL https://www. -sciencedirect.com/science/article/pii/S0364021385800124. - -Daniel Adiwardana, Minh-Thang Luong, David R. So, Jamie Hall, Noah Fiedel, Romal Thoppilan, -Zi Yang, Apoorv Kulshreshtha, Gaurav Nemade, Yifeng Lu, and Quoc V. Le. Towards a human-like -open-domain chatbot, 2020. - -Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh -Hajishirzi. MathQA: Towards interpretable math word problem solving with operation-based -In Proceedings of the 2019 Conference of the North American Chapter of the -formalisms. -Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long -and Short Papers), pp. 2357–2367. Association for Computational Linguistics, June 2019. URL -https://aclanthology.org/N19-1245. - -Daniel Andor, Luheng He, Kenton Lee, and Emily Pitler. Giving BERT a calculator: Finding -operations and arguments with reading comprehension. In Proceedings of the 2019 Conference on -Empirical Methods in Natural Language Processing and the 9th International Joint Conference -on Natural Language Processing (EMNLP-IJCNLP), 2019. URL https://aclanthology. -org/D19-1609. - -Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learn- -ing to retrieve reasoning paths over wikipedia graph for question answering. In International -Conference on Learning Representations, 2020. URL https://openreview.net/forum? -id=SJgVHkrYDH. - -Roy Bar-Haim, Ido Dagan, Bill Dolan, Lisa Ferro, Danilo Giampiccolo, Bernardo Magnini, and -Idan Szpektor. The second pascal recognising textual entailment challenge. In Proceedings of the -second PASCAL challenges workshop on recognising textual entailment, 2006. - -Dhruv Batra, Payman Yadollahpour, Abner Guzman-Rivera, and Gregory Shakhnarovich. Diverse -m-best solutions in markov random fields. In Proceedings of the 12th European Conference on -Computer Vision - Volume Part V, ECCV’12, pp. 1–16, Berlin, Heidelberg, 2012. Springer-Verlag. -ISBN 9783642337147. URL https://doi.org/10.1007/978-3-642-33715-4_1. +This material is based in part upon work supported by the German Federal Ministry of Education +and Research (BMBF): Tübingen AI Center, FKZ: 01IS18039B; by the Machine Learning Cluster of +Excellence, EXC number 2064/1 – Project number 390727645; by a National Science Foundation +award (#2306372); by a Swiss National Science Foundation award (#201009) and a Responsible +AI grant by the Haslerstiftung. Zhijing Jin is supported by PhD fellowships from the Future of Life +Institute and Open Philanthropy. We also thank OpenAI for granting Zhijing quota to their API of +GPT series through the Researcher Access Program. -10 - - Published as a conference paper at ICLR 2023 - -Luisa Bentivogli, Peter Clark, Ido Dagan, and Danilo Giampiccolo. The fifth pascal recognizing +REFERENCES -textual entailment challenge. In TAC, 2009. +Tariq Alhindi, Tuhin Chakrabarty, Elena Musi, and Smaranda Muresan. Multitask instruction-based +prompting for fallacy recognition. In Proceedings of the 2022 Conference on Empirical Methods +in Natural Language Processing, pp. 8172–8187, Abu Dhabi, United Arab Emirates, December +2022. Association for Computational Linguistics. URL https://aclanthology.org/ +2022.emnlp-main.560. 9 -BIG-bench collaboration. Beyond the imitation game: Measuring and extrapolating the capabil- -In preparation, 2021. URL https://github.com/google/ +Chandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi, Ari Holtzman, +Hannah Rashkin, Doug Downey, Wen-tau Yih, and Yejin Choi. Abductive commonsense reasoning. +In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, +April 26-30, 2020. OpenReview.net, 2020. URL https://openreview.net/forum?id= +Byg1v1HKDB. 1 -ities of language models. -BIG-bench/. +Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. A large +annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference +on Empirical Methods in Natural Language Processing, pp. 632–642, Lisbon, Portugal, September +2015. Association for Computational Linguistics. doi: 10.18653/v1/D15-1075. URL https: +//aclanthology.org/D15-1075. 9 Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, -Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, -Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, -and Dario Amodei. Language models are few-shot learners. In Advances in Neural Information -Processing Systems, 2020. URL https://proceedings.neurips.cc/paper/2020/ -file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. - -language inference with natural - -Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. - -snli: Natural -gio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, -(eds.), Advances -Information Processing Systems 31, -Curran Associates, -8163-e-snli-natural-language-inference-with-natural-language-explanations. -pdf. - -e- -In S. Ben- -and R. Garnett -pp. 9539–9549. -URL http://papers.nips.cc/paper/ - -language explanations. - -in Neural -Inc., - -2018. - -Oana-Maria Camburu, Brendan Shillingford, Pasquale Minervini, Thomas Lukasiewicz, and Phil -Blunsom. Make up your mind! adversarial generation of inconsistent natural language explanations. -In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. -4157–4165, Online, July 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -acl-main.382. URL https://aclanthology.org/2020.acl-main.382. - -Jifan Chen, Shih-Ting Lin, and Greg Durrett. Multi-hop question answering via reasoning chains. - -CoRR, abs/1910.02610, 2019. URL http://arxiv.org/abs/1910.02610. - -Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared -Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large -language models trained on code. arXiv preprint arXiv:2107.03374, 2021. - -Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam -Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, -Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam -Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James -Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Lev- -skaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin -Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, -Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. -Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon -Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark -Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeff Dean, -Slav Petrov, and Noah Fiedel. Palm: Scaling language modeling with pathways, 2022. URL -https://arxiv.org/abs/2204.02311. - -Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina -Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. In NAACL, -2019. - -Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and -Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. -ArXiv, abs/1803.05457, 2018. - -Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, -Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John -Schulman. Training verifiers to solve math word problems, 2021. - -11 - - Published as a conference paper at ICLR 2023 - -Ido Dagan, Oren Glickman, and Bernardo Magnini. The pascal recognising textual entailment - -challenge. In Machine Learning Challenges Workshop, pp. 177–190. Springer, 2005. - -Bryan Eikema and Wilker Aziz. Is MAP decoding all you need? the inadequacy of the mode in neural -machine translation. In Proceedings of the 28th International Conference on Computational Lin- -guistics, pp. 4506–4520, Barcelona, Spain (Online), December 2020. International Committee on -Computational Linguistics. URL https://aclanthology.org/2020.coling-main. -398. - -Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich -Schütze, and Yoav Goldberg. Measuring and improving consistency in pretrained language -models. Transactions of the Association for Computational Linguistics, 9:1012–1031, 2021. doi: -10.1162/tacl_a_00410. URL https://aclanthology.org/2021.tacl-1.60. - -Jonathan St BT Evans. Intuition and reasoning: A dual-process perspective. Psychological Inquiry, - -21(4):313–326, 2010. - -Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Proceedings -of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long -Papers), pp. 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics. -doi: 10.18653/v1/P18-1082. URL https://aclanthology.org/P18-1082. - -Jessica Ficler and Yoav Goldberg. Controlling linguistic style aspects in neural language generation. In -Proceedings of the Workshop on Stylistic Variation, pp. 94–104, Copenhagen, Denmark, September -2017. Association for Computational Linguistics. doi: 10.18653/v1/W17-4912. URL https: -//aclanthology.org/W17-4912. - -Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot -learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguis- -tics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long -Papers), pp. 3816–3830, Online, August 2021. Association for Computational Linguistics. doi: -10.18653/v1/2021.acl-long.295. URL https://aclanthology.org/2021.acl-long. -295. - -Mor Geva, Ankit Gupta, and Jonathan Berant. Injecting numerical reasoning skills into language -models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguis- -tics, 2020. doi: 10.18653/v1/2020.acl-main.89. URL https://aclanthology.org/2020. -acl-main.89. - -Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle -use a laptop? A question answering benchmark with implicit reasoning strategies. Transactions of -the Association for Computational Linguistics, 2021. URL https://aclanthology.org/ -2021.tacl-1.21. - -Danilo Giampiccolo, Bernardo Magnini, Ido Dagan, and Bill Dolan. The third pascal recognizing -textual entailment challenge. In Proceedings of the ACL-PASCAL workshop on textual entailment -and paraphrasing, pp. 1–9. Association for Computational Linguistics, 2007. - -Ari Holtzman, Jan Buys, Maxwell Forbes, Antoine Bosselut, David Golub, and Yejin Choi. Learning -In Proceedings of the 56th Annual Meeting of the -to write with cooperative discriminators. -Association for Computational Linguistics (Volume 1: Long Papers), pp. 1638–1649, Melbourne, -Australia, July 2018. Association for Computational Linguistics. doi: 10.18653/v1/P18-1152. -URL https://aclanthology.org/P18-1152. - -Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text -degeneration. In International Conference on Learning Representations, 2020. URL https: -//openreview.net/forum?id=rygGQyrFvH. - -Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to -solve arithmetic word problems with verb categorization. In Proceedings of the 2014 Conference on -Empirical Methods in Natural Language Processing (EMNLP), 2014. doi: 10.3115/v1/D14-1058. -URL https://aclanthology.org/D14-1058. - -12 - - Published as a conference paper at ICLR 2023 - -Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Han- -naneh Hajishirzi. UNIFIEDQA: Crossing format boundaries with a single QA system. In Findings -of the Association for Computational Linguistics: EMNLP 2020, pp. 1896–1907, Online, Novem- -ber 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.findings-emnlp.171. -URL https://aclanthology.org/2020.findings-emnlp.171. - -Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large -language models are zero-shot reasoners. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, -and Kyunghyun Cho (eds.), Advances in Neural Information Processing Systems, 2022. URL -https://openreview.net/forum?id=e2TBb5y0yFf. - -Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. MAWPS: -A math word problem repository. In Proceedings of the 2016 Conference of the North American -Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016. -doi: 10.18653/v1/N16-1136. URL https://aclanthology.org/N16-1136. - -Yihuai Lan, Lei Wang, Qiyuan Zhang, Yunshi Lan, Bing Tian Dai, Yan Wang, Dongxiang Zhang, -and Ee-Peng Lim. MWPToolkit: An open-source framework for deep learning-based math word -problem solvers. arXiv preprint arXiv:2109.00799, 2021. URL https://arxiv.org/abs/ -2109.00799. - -Jiwei Li and Dan Jurafsky. Mutual information and diverse decoding improve neural machine - -translation, 2016. URL https://arxiv.org/abs/1601.00372. - -Jiwei Li, Will Monroe, and Dan Jurafsky. A simple, fast diverse decoding algorithm for neural -generation. CoRR, abs/1611.08562, 2016. URL http://arxiv.org/abs/1611.08562. - -Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale genera- -tion: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual -Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2017. doi: -10.18653/v1/P17-1015. URL https://aclanthology.org/P17-1015. - -Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. Fantastically -ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. ArXiv, -abs/2104.08786, 2021. - -Clara Meister, Tiago Pimentel, Gian Wiher, and Ryan Cotterell. Typical decoding for natural language - -generation. arXiv preprint arXiv:2202.00666, 2022. - -Shen Yun Miao, Chao Chun Liang, and Keh Yih Su. A diverse corpus for evaluating and developing -English math word problem solvers. In Proceedings of the 58th Annual Meeting of the Asso- -ciation for Computational Linguistics, 2020. URL https://aclanthology.org/2020. -acl-main.92. - -Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adver- -sarial NLI: A new benchmark for natural language understanding. In Proceedings of the 58th -Annual Meeting of the Association for Computational Linguistics. Association for Computational -Linguistics, 2020. - -Maxwell Nye, Michael Henry Tessler, Joshua B. Tenenbaum, and Brenden M. Lake. Improving -coherence and consistency in neural sequence models with dual-system, neuro-symbolic reasoning. -In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan (eds.), Advances in Neural -Information Processing Systems, 2021. URL https://openreview.net/forum?id= -uyKk_avJ-p4. - -Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP models really able to solve simple -math word problems? In Proceedings of the 2021 Conference of the North American Chapter of -the Association for Computational Linguistics: Human Language Technologies, pp. 2080–2094, -Online, June 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main. -168. URL https://aclanthology.org/2021.naacl-main.168. - -Xinyu Pi, Qian Liu, Bei Chen, Morteza Ziyadi, Zeqi Lin, Yan Gao, Qiang Fu, Jian-Guang Lou, and - -Weizhu Chen. Reasoning like program executors, 2022. - -13 - - Published as a conference paper at ICLR 2023 - -Piotr Pi˛ekos, Mateusz Malinowski, and Henryk Michalewski. Measuring and improving BERT’s -mathematical abilities by predicting the order of reasoning. In Proceedings of the 59th Annual Meet- -ing of the Association for Computational Linguistics and the 11th International Joint Conference on -Natural Language Processing (Volume 2: Short Papers), 2021. doi: 10.18653/v1/2021.acl-short.49. -URL https://aclanthology.org/2021.acl-short.49. - -Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language - -models are unsupervised multitask learners. 2019. - -Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John -Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: -Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446, 2021. - -Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. NumNet: Machine reading comprehension -with numerical reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural -Language Processing and the 9th International Joint Conference on Natural Language Processing -(EMNLP-IJCNLP), 2019. doi: 10.18653/v1/D19-1251. URL https://aclanthology. -org/D19-1251. - -Subhro Roy and Dan Roth. Solving general arithmetic word problems. In Proceedings of the 2015 -Conference on Empirical Methods in Natural Language Processing, 2015. doi: 10.18653/v1/ -D15-1202. URL https://aclanthology.org/D15-1202. - -Jianhao Shen, Yichun Yin, Lin Li, Lifeng Shang, Xin Jiang, Ming Zhang, and Qun Liu. Generate -& rank: A multi-task framework for math word problems. In Findings of the Association for -Computational Linguistics: EMNLP 2021, pp. 2269–2279, Punta Cana, Dominican Republic, -November 2021. Association for Computational Linguistics. URL https://aclanthology. -org/2021.findings-emnlp.195. - -Freda Shi, Daniel Fried, Marjan Ghazvininejad, Luke Zettlemoyer, and Sida I. Wang. Natural -language to code translation with execution. In Proceedings of the 2022 Conference on Empirical -Methods in Natural Language Processing, pp. 3533–3546, Abu Dhabi, United Arab Emirates, -December 2022. Association for Computational Linguistics. URL https://aclanthology. -org/2022.emnlp-main.231. +Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Ben- +jamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and +Dario Amodei. Language models are few-shot learners. In H. Larochelle, M. Ranzato, R. Hadsell, +M. F. Balcan, and H. Lin (eds.), Advances in Neural Information Processing Systems, volume 33, +pp. 1877–1901. Curran Associates, Inc., 2020. URL https://proceedings.neurips. +cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. 6 + +David Maxwell Chickering. Optimal structure identification with greedy search. J. Mach. Learn. +Res., 3:507–554, 2002. URL http://jmlr.org/papers/v3/chickering02b.html. +3 -Keith E Stanovich and Richard F West. +10 -Individual differences in reasoning: Implications for -the rationality debate? Behavioral and brain sciences, 23(5):645–665, 2000. URL https: -//pubmed.ncbi.nlm.nih.gov/11301544/. + Published as a conference paper at ICLR 2024 -Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. CommonsenseQA: A question -answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of +Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep +bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language -Technologies, Volume 1 (Long and Short Papers), 2019. URL https://aclanthology. -org/N19-1421. - -Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Jason Wei, Xuezhi Wang, Hyung Won Chung, -Dara Bahri, Tal Schuster, Steven Zheng, Denny Zhou, Neil Houlsby, and Donald Metzler. Unifying -language learning paradigms, 2022. URL https://arxiv.org/abs/2205.05131. - -Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze -Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog -applications. arXiv preprint arXiv:2201.08239, 2022. URL https://arxiv.org/abs/ -2201.08239. - -Ashwin Vijayakumar, Michael Cogswell, Ramprasaath Selvaraju, Qing Sun, Stefan Lee, David -Crandall, and Dhruv Batra. Diverse beam search for improved description of complex scenes. -Proceedings of the AAAI Conference on Artificial Intelligence, 32, Apr. 2018. URL https: -//ojs.aaai.org/index.php/AAAI/article/view/12340. - -Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc -Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. -Conference on Neural Information Processing Systems (NeurIPS), 2022. URL https://arxiv. -org/pdf/2201.11903. - -14 - - Published as a conference paper at ICLR 2023 - -Sean Welleck, Ilia Kulikov, Jaedeok Kim, Richard Yuanzhe Pang, and Kyunghyun Cho. Consistency -of a recurrent language model with respect to incomplete decoding. In Proceedings of the 2020 -Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5553–5568, -Online, November 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -emnlp-main.448. URL https://aclanthology.org/2020.emnlp-main.448. - -Weiwen Xu, Yang Deng, Huihui Zhang, Deng Cai, and Wai Lam. Exploiting reasoning chains -for multi-hop science question answering. In Findings of the Association for Computational -Linguistics: EMNLP 2021, pp. 1143–1156, Punta Cana, Dominican Republic, November 2021a. -Association for Computational Linguistics. URL https://aclanthology.org/2021. -findings-emnlp.99. - -Yichong Xu, Chenguang Zhu, Shuohang Wang, Siqi Sun, Hao Cheng, Xiaodong Liu, Jianfeng -Gao, Pengcheng He, Michael Zeng, and Xuedong Huang. Human parity on commonsenseqa: -Augmenting self-attention with external attention, 2021b. URL https://arxiv.org/abs/ -2112.03254. - -Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and -Christopher D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answer- -ing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, -pp. 2369–2380, Brussels, Belgium, October-November 2018. Association for Computational Lin- -guistics. doi: 10.18653/v1/D18-1259. URL https://aclanthology.org/D18-1259. - -Xi Ye and Greg Durrett. The unreliability of explanations in few-shot prompting for textual reasoning. -In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho (eds.), Advances in -Neural Information Processing Systems, 2022. URL https://openreview.net/forum? -id=Bct2f8fRd8S. - -Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, and Meng Jiang. Diversifying -content generation for commonsense reasoning with mixture of knowledge graph experts. In -Findings of Annual Meeting of the Association for Computational Linguistics (ACL), 2022. - -Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. Calibrate before use: Improving -few-shot performance of language models. In Marina Meila and Tong Zhang (eds.), Proceed- -ings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of -Machine Learning Research. PMLR, 2021. URL https://proceedings.mlr.press/ -v139/zhao21c.html. +Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186, Minneapolis, Minnesota, June +2019. Association for Computational Linguistics. doi: 10.18653/v1/N19-1423. URL https: +//aclanthology.org/N19-1423. 1, 6 + +Clark Glymour, Kun Zhang, and Peter Spirtes. Review of causal discovery methods based on graph- +ical models. Frontiers in Genetics, 10:524, 2019. ISSN 1664-8021. doi: 10.3389/fgene.2019. +00524. URL https://www.frontiersin.org/article/10.3389/fgene.2019. +00524. 2, 3 + +Madelyn Glymour, Judea Pearl, and Nicholas P Jewell. Causal inference in statistics: A primer. John + +Wiley and Sons, 2016. 2, 3 + +Andrew Gordon, Zornitsa Kozareva, and Melissa Roemmele. SemEval-2012 task 7: Choice of +plausible alternatives: An evaluation of commonsense causal reasoning. In *SEM 2012: The First +Joint Conference on Lexical and Computational Semantics – Volume 1: Proceedings of the main +conference and the shared task, and Volume 2: Proceedings of the Sixth International Workshop on +Semantic Evaluation (SemEval 2012), pp. 394–398, Montréal, Canada, 7-8 June 2012. Association +for Computational Linguistics. URL https://aclanthology.org/S12-1052. 1 + +Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. Deberta: Decoding-enhanced +Bert with disentangled attention. In 9th International Conference on Learning Representations, +ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021. URL https:// +openreview.net/forum?id=XPZIaotutsD. 6 + +Patrik O. Hoyer, Dominik Janzing, Joris M. Mooij, Jonas Peters, and Bernhard Schölkopf. Non- +In Daphne Koller, Dale Schuurmans, +linear causal discovery with additive noise models. +Yoshua Bengio, and Léon Bottou (eds.), Advances in Neural Information Processing Systems +21, Proceedings of the Twenty-Second Annual Conference on Neural Information Processing +Systems, Vancouver, British Columbia, Canada, December 8-11, 2008, pp. 689–696. Curran As- +sociates, Inc., 2008. URL https://proceedings.neurips.cc/paper/2008/hash/ +f7664060cc52bc6f3d620bcedc94a4b6-Abstract.html. 3 + +Di Jin, Zhijing Jin, Joey Tianyi Zhou, and Peter Szolovits. Is BERT really robust? A strong baseline +In The Thirty-Fourth AAAI +for natural language attack on text classification and entailment. +Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications +of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational +Advances in Artificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020, pp. 8018– +8025. AAAI Press, 2020. URL https://aaai.org/ojs/index.php/AAAI/article/ +view/6311. 9 + +Zhijing Jin, Abhinav Lalwani, Tejas Vaidhya, Xiaoyu Shen, Yiwen Ding, Zhiheng Lyu, Mrinmaya +In Findings of +Sachan, Rada Mihalcea, and Bernhard Schölkopf. Logical fallacy detection. +the Association for Computational Linguistics: EMNLP 2022, pp. 7180â€âCœ–7198, Abu +Dhabi, United Arab Emirates, December 2022. Association for Computational Linguistics. URL +https://arxiv.org/abs/2202.13758. 9, 10 + +Emre Kıcıman, Robert Ness, Amit Sharma, and Chenhao Tan. Causal reasoning and large language + +models: Opening a new frontier for causality. arXiv preprint arXiv:2305.00050, 2023. 1, 9 + +Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, +Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-training +for natural language generation, translation, and comprehension. In Proceedings of the 58th +Annual Meeting of the Association for Computational Linguistics, pp. 7871–7880, Online, July +2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.acl-main.703. URL +https://aclanthology.org/2020.acl-main.703. 6 + +Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, +Luke Zettlemoyer, and Veselin Stoyanov. RoBERTa: A robustly optimized BERT pretraining +approach. CoRR, abs/1907.11692, 2019. URL http://arxiv.org/abs/1907.11692. 6 -15 - - Published as a conference paper at ICLR 2023 - -A APPENDIX - -A.1 ADDITIONAL EXPERIMENT RESULTS - -A.1.1 ROBUSTNESS TO SAMPLING STRATEGIES AND PARAMETERS - -In Figure 6 we ablate the results with respect to different sampling strategies and parameters by -varying T in temperature sampling and k in Top-k sampling, on LaMDA-137B. We show that -self-consistency is robust to various sampling strategies and parameters. - -Figure 6: GSM8K accuracy over LaMDA-137B. Self-consistency works under various sampling -strategies and sampling parameters. - -In Figure 7 and Figure 8, we show the results of self-consistency compared with greedy decoding a -single path over LaMDA-137B and PaLM-540B, respectively. Self-consistency improves over greedy -decode by a quite significant margin on both models, on top of high accuracy already achieved by -scaling up model sizes. - -Figure 7: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over LaMDA-137B. Sampling a higher number of diverse reasoning -paths consistently improves reasoning accuracy. +11 -We further show additional sampled reasoning paths from the LaMDA-137B model in Table 12, and -sampled reasoning paths from the PaLM-540B model in Table 13. We see that the diversity in the -additionally sampled reasoning paths indeed helps the model arrive at a more correct final answer -after aggregation. + Published as a conference paper at ICLR 2024 -A.1.2 ROBUSTNESS TO DIFFERENT SETS OF PROMPTS +Bill MacCartney and Christopher D. Manning. Modeling semantic containment and exclusion in +natural language inference. In Proceedings of the 22nd International Conference on Computational +Linguistics (Coling 2008), pp. 521–528, Manchester, UK, August 2008. Coling 2008 Organizing +Committee. URL https://aclanthology.org/C08-1066. 9 -In Table 9, we further show that self-consistency is quite robust to different sets of input prompts. -We manually wrote 3 different sets of chain-of-thought as prompts to the model. Across all sets of -prompts, self-consistency yields consistent gains over the original CoT approach. +Brendan D. McKay and Adolfo Piperno. Practical graph isomorphism, II. J. Symb. Comput., 60: +94–112, 2014. doi: 10.1016/j.jsc.2013.09.003. URL https://doi.org/10.1016/j.jsc. +2013.09.003. 5 -A.1.3 COMPARED TO MODEL ENSEMBLES +John Morris, Eli Lifland, Jin Yong Yoo, Jake Grigsby, Di Jin, and Yanjun Qi. TextAttack: A framework +for adversarial attacks, data augmentation, and adversarial training in NLP. In Proceedings of the +2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, +pp. 119–126, Online, October 2020. Association for Computational Linguistics. doi: 10.18653/v1/ +2020.emnlp-demos.16. URL https://aclanthology.org/2020.emnlp-demos.16. +9 -Additionally, we provide results of directly ensembling the outputs from multiple language models. -The results are shown in Table 10, by greedily decoding sequences from 3 language models and +OpenAI. GPT-4 technical report. CoRR, abs/2303.08774, 2023. doi: 10.48550/arXiv.2303.08774. -16 +URL https://doi.org/10.48550/arXiv.2303.08774. 1, 6 -481216202428323640#Sampled Reasoning Paths182022242628Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.5, k=20T=0.5, no top kGreedy Decode0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths4446485052545658ASDiv0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths1416182022242628GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths56586062Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths62636465666768Strategy QA0510152025303540#Sampled Reasoning Paths687072747678ARC (Easy)0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong +Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser +Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan +Leike, and Ryan Lowe. Training language models to follow instructions with human feedback. +CoRR, abs/2203.02155, 2022. doi: 10.48550/arXiv.2203.02155. URL https://doi.org/10. +48550/arXiv.2203.02155. 1, 6 -Figure 8: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over PaLM-540B. Sampling a higher number of diverse reasoning -paths consistently helps reasoning accuracy. +Judea Pearl. Probabilistic reasoning in intelligent systems: Networks of plausible inference. Morgan -Prompt set 1 (used in the main text) +Kaufmann, 1988. 3 -Prompt set 2 +Judea Pearl. Causality: Models, reasoning and inference (2nd ed.). Cambridge University Press, -Prompt set 3 +2009. 1 -CoT (Wei et al., 2022) -Self-consistency +Jonas Peters, Dominik Janzing, and Bernhard Schölkopf. Elements of causal inference: Foundations +and learning algorithms. The MIT Press, 2017. URL https://mitpress.mit.edu/ +books/elements-causal-inference. 1 -56.5 -74.4 (+17.9) +Lianhui Qin, Antoine Bosselut, Ari Holtzman, Chandra Bhagavatula, Elizabeth Clark, and Yejin +Choi. Counterfactual story reasoning and generation. In Proceedings of the 2019 Conference on +Empirical Methods in Natural Language Processing and the 9th International Joint Conference on +Natural Language Processing (EMNLP-IJCNLP), pp. 5043–5053, Hong Kong, China, November +2019. Association for Computational Linguistics. doi: 10.18653/v1/D19-1509. URL https: +//aclanthology.org/D19-1509. 1, 9 -54.6 -72.1 (+17.5) +Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language -54.0 -70.4 (+16.4) +models are unsupervised multitask learners. OpenAI Blog, 1(8), 2019. 1, 6 -Table 9: GSM8K accuracy over PaLM-540B. The results show robustness of self-consistency with -respect to different prompts in the input. +Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. DistilBERT, a distilled version +of BERT: Smaller, faster, cheaper and lighter. CoRR, abs/1910.01108, 2019. URL http: +//arxiv.org/abs/1910.01108. 6 -taking the majority vote (averaged over 10 runs). Note this is a typical ensemble approach (averaging -over the predictions over multiple models) and it achieves a performance significantly worse than -self-consistency (self-consistency over PaLM-540B gets an accuracy of 74.4%), as lower-capacity -models drag down the performance of higher-capacity models. In addition, this approach is limited in -two ways: 1) It requires multiple models for an ensemble which might not always be available, while -self-consistency only requires one single model to “self-ensemble”; 2) If one of the models is much -weaker, it can actually hurt the final performance. +Maarten Sap, Ronan Le Bras, Emily Allaway, Chandra Bhagavatula, Nicholas Lourie, Hannah +Rashkin, Brendan Roof, Noah A. Smith, and Yejin Choi. ATOMIC: an atlas of machine com- +monsense for if-then reasoning. In The Thirty-Third AAAI Conference on Artificial Intelligence, +AAAI 2019, The Thirty-First Innovative Applications of Artificial Intelligence Conference, IAAI +2019, The Ninth AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2019, +Honolulu, Hawaii, USA, January 27 - February 1, 2019, pp. 3027–3035. AAAI Press, 2019a. +doi: 10.1609/aaai.v33i01.33013027. URL https://doi.org/10.1609/aaai.v33i01. +33013027. 1, 9 -Method +Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. Social iqa: Common- -Single model +sense reasoning about social interactions. In EMNLP 2019, 2019b. 1, 9 -PaLM-540B, greedy / self-consistency +12 -Ensemble of models + Published as a conference paper at ICLR 2024 -LaMDA-137B + PaLM-540B -PaLM-540B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + PaLM-540B + GPT-3 (code-davinci-001, 175B) +Shohei Shimizu, Patrik O. Hoyer, Aapo Hyvärinen, and Antti J. Kerminen. A linear non-gaussian +acyclic model for causal discovery. J. Mach. Learn. Res., 7:2003–2030, 2006. URL http: +//jmlr.org/papers/v7/shimizu06a.html. 3 -GSM8K accuracy +Sam Shleifer and Alexander M. Rush. Pre-trained summarization distillation. CoRR, abs/2010.13002, -56.5 / 74.4 +2020. URL https://arxiv.org/abs/2010.13002. 6 -36.9 ± 0.5 -36.6 ± 0.4 -16.0 ± 0.8 -33.3 ± 0.7 +Peter Spirtes and Kun Zhang. Causal discovery and inference: Concepts and recent methodological -Table 10: Comparison of GSM8K accuracy over multiple-model ensembles. +advances. In Applied informatics, volume 3, pp. 1–28. SpringerOpen, 2016. 2, 3 -A.1.4 COMBINING SELF-CONSISTENCY WITH OTHER ENSEMBLING STRATEGIES +Peter Spirtes, Clark Glymour, and Richard Scheines. Causation, prediction, and search. 1993. 2, 3 -Self-consistency is completely compatible with other ensemble strategies, although the gains achieved -by self-consistency are significantly higher than other ensemble strategies (and can “override” the -performance gains achieved by other ensemble strategies). We further performed experiments and -include the results in Table 11 (for a fair comparison, we use 40 sets of prompts, or 40 prompt -permutations to compare with self-consistency with 40 paths, all experiments are based on PaLM- -540B). +Peter Spirtes, Clark Glymour, and Richard Scheines. Causation, Prediction, and Search, Second +Edition. Adaptive computation and machine learning. MIT Press, 2000. ISBN 978-0-262-19440-2. +1, 2, 3, 10 -17 +Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy +Liang, and Tatsunori B. Hashimoto. Stanford alpaca: An instruction-following llama model. +https://github.com/tatsu-lab/stanford_alpaca, 2023. 7 -0510152025303540#Sampled Reasoning Chains8688909294Accuracy (%)AddSub0510152025303540#Sampled Reasoning Chains727476788082Accuracy (%)ASDiv0510152025303540#Sampled Reasoning Chains30333639424548Accuracy (%)AQuAGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Chains889092949698Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Chains70.072.575.077.580.082.585.087.5Accuracy (%)SVAMP0510152025303540#Sampled Reasoning Chains505560657075Accuracy (%)GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths7475767778798081Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths7476788082Strategy QA0510152025303540#Sampled Reasoning Paths8890929496ARC (Easy)0510152025303540#Sampled Reasoning Paths788082848688ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -GSM8K accuracy - -Different sets of prompts (x40) -Prompt-permutation (x40) -Self-consistency (x40) -Self-consistency + different sets of prompts (x40) -Self-consistency + prompt-permutation (x40) - -58.9 -59.6 -74.4 -75.4 -73.8 - -Table 11: Combining self-consistency with other ensembling strategies. - -A.2 DETAILS ON RESOURCES AND INFERENCE - -For all four language models we perform prompting-based inference only. For UL2 we use TPU v3 -(2x2 configuration, 4 chips, 8 cores). For GPT-3 models the experiments are done though the public -API.10 For LaMDA-137B we use TPU v3 (8x8 configuration, 64 chips, 128 cores). For PaLM-540B -we use TPU v4 (4x4x12 configuration, 192 chips, 384 cores). Most inference jobs take 1 to 4 hours -(over about 1,000 examples) for each task on UL2 and LaMDA-137B, and about 2 to 12 hours on -PaLM-540B. Some tasks (e.g., commonsense reasoning) take longer but do not exceed 2 days for -each task. - -For GPT-3 models, we use 128 max tokens for all methods, without frequency penalty or presence -penalty. For all models, we take the generated outputs until the start of the next “Q:" to parse the final -answers, consistent with our prompting format. - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[MultiArith] Sam had 10 socks. -If he threw away 3 old ones -that didn’t fit and bought 36 new -ones, how many socks would he -have? - -Sam had 10 socks. He threw -away 3 old ones that didn’t fit. -He bought 36 new ones. That -means he has 10 + 36 - 3 = 41 -socks. The answer is 41. - -[SVAMP] In a school there are -697 girls and the rest are boys. -If there are 228 more girls than -boys. How many boys are there -in that school? - -[AQuA] There are 10 oranges in -a basket. Find the no. of ways in -which 2 oranges are chosen from -the basket? (a) 45 (b) 90 (c) 120 -(d) 150 (e) 180 - -There are 697 girls and the rest -are boys. That means there are -697 + 228 = 925 girls. The -rest are boys. So the number of -boys is 925 - 697 = 228. The -answer is 228. - -There are 10 oranges in a bas- -ket. -So there are 10 ways -in which 2 oranges are chosen -from the basket. - -[CommonsenseQA] The man -laid on the soft moss and looked -up at the trees, where was the -man? (a) niagra falls (b) forest -(c) waterfall (d) ground (e) tree - -The answer must be something -that is soft and mossy. Of the -above choices, only ground is -soft and mossy. So the answer -is (d). - -[ARC] Which of these will most -likely harm a habitat? (a) plant- -ing trees. (b) water pollution. (c) -rainfall. (d) sunlight. - -Planting trees will not harm a -habitat. So the answer is (a). - -[ARC-challenge] Which pro- -cess in the carbon cycle takes the -longest to complete? (a) emis- -sion of waste. (b) respiration in -animals. -(c) photosynthesis in -plants. (d) formation of fossil fu- -els. - -The process of photosynthesis -in plants takes the longest to -complete. So the answer is (c). - -Sam had 10 socks. He -threw away 3 old ones. So -now he has 7 left. Then he -bought 36 new ones. So -now he has 7 + 36 = 43 -socks. The answer is 43. - -He had 10 socks. He threw away -3 old ones that didn’t fit. He -bought 36 new ones. So in total -he has 10 - 3 + 36 = 43 socks. The -answer is 43. - -The number of boys is the -difference between 697 -and 228. This is 697 - 228 -= 469. The answer is 469. - -In the beginning there are 697 -girls and the rest are boys. So, the -number of boys must be 697 - 228 -= 469 boys. The answer is 469. - -There are 10 oranges in a -basket. We need to choose -2 oranges from the basket. -So the number of ways are -10C2. So the answer is (a). - -The -be -answer must -somewhere where moss -is found. Of the above -choices, only forests are -known for having moss. -So the answer is (b). - -The answer is (b), water -pollution, because it will -most likely harm the habi- -tat. The other answers are -not likely to harm the habi- -tat. - -The formation of fossil fu- -els takes the longest time -to complete. So the an- -swer is (d). - -There are 10 oranges in the bas- -ket. If you choose 2 oranges, you -have 10C2 = 45 ways of choosing -2 oranges. So the answer is (a). - -The answer must be a place where -the man is laying down and look- -ing up at trees. Of the above -is a place -choices, only forest -where people lay down and look -at trees. So the answer is (b). - -The answer is (b), water pollution. -Trees are good for a habitat, rain- -fall is necessary for a habitat, and -sunlight is necessary for a habitat. -So the answer is (b). - -The process that takes the longest -time to complete is the forma- -tion of fossil fuels, which hap- -pens over millions of years. So -the answer is (d). - -Table 12: Additional examples where self-consistency helps repair the errors over greedy decode on -LaMDA-137B. Two sampled reasoning paths that are consistent with the ground truth are shown. - -10https://beta.openai.com/docs/api-reference/making-requests - -18 - - Kyle bought a book that is sold -for $19.50. With a 25% discount, -the original price must be higher. -If we call -the original price x, -then the discounted price is 0.75x. -Since this is $19.50, we can write -0.75x = 19.50. Solving this, we -find that the original price is x = -19.50 / 0.75 = $26. The answer is -$26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. One cu- -bic meter contains 1000 litres, so -the tank contains 120 m3 * 1000 -liters/m3 = 120,000 liters. So the -answer is (d). - -Published as a conference paper at ICLR 2023 - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[GSM8K] Kyle bought -last -year’s best-selling book for -$19.50. This is with a 25% dis- -count from the original price. -What was the original price of -the book? - -The book was sold for $19.50. -This is 25% less than the origi- -nal price. This means the original -price was 100%. 25% of 100% -is 25%. -So the original price -was 100% - 25% = 75%. The -original price was 75% of $19.50. -75% of $19.50 is 75% * $19.50 = -$14.625. The answer is $14.625. - -The original price of the book -is $19.50. Since it was dis- -counted by 25%, the original -price was 19.50 / 0.75 = 26. -The answer is 26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. So the -capacity of the tank is 120 m3. So -the answer is (c). - -The volume of the tank is 8m -* 6m* 2.5m = 120m3. This is -equal to 120000 litres. So the -answer is (d) - -[AQuA] The capacity of a -tank of dimensions (8 m x 6 -m x 2.5 m) is (a) 120 litres (b) -1200 litres (c) 12000 litres (d) -120000 litres (e) None of these - -no: -[StrategyQA] Yes -Can the Great Depression be -treated with Prozac? - -or - -The Great Depression was an -economic depression. Prozac is -a drug used to treat depression. -Thus, the Great Depression could -be treated with Prozac. So the an- -swer is yes. - -Prozac is an anti-depressant -medication. The Great Depres- -sion is not a disease. Thus, -Prozac cannot treat the Great -Depression. So the answer is -no. - -The Great Depression is a time pe- -riod. Prozac is used to treat de- -pression (the mental illness). Thus, -the Great Depression can not be -treated with Prozac. So the answer -is no. - -[ARC-easy] Which source of -energy is nonrenewable? (a) -coal. -(c) water. -(d) wood. - -(b) sunlight. - -Coal, -sunlight, and water are -all renewable sources of energy. -Wood is a nonrenewable source of -energy. So the answer is (d). - -Students -[ARC-challenge] -studied the revolution of Earth -around the Sun. During which -month will Florida have the -most sunlight energy avail- -able? (a) March. (b) June. (c) -September. (d) December. +Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée +Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurélien Rodriguez, Armand +Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language +models. CoRR, abs/2302.13971, 2023. doi: 10.48550/arXiv.2302.13971. URL https://doi. +org/10.48550/arXiv.2302.13971. 7 -The Earth is closest to the Sun -in December, hence Florida will -have the most sunlight energy -available in December. So the an- -swer is (d). +Ruibo Tu, Chao Ma, and Cheng Zhang. Causal-discovery performance of chatgpt in the context of -A nonrenewable source of en- -ergy is one in which the source -is not replenished in a short -amount of time. Thus, the only -option that is nonrenewable is -coal. So the answer is (a). +neuropathic pain diagnosis. arXiv preprint arXiv:2301.13819, 2023. 1 -Coal is a fossil fuel and is formed -from dead organisms from the past. -Coal cannot be produced in a short -period of time, so it is considered -a nonrenewable source of energy. -So the answer is (a). +Moritz Willig, Matej Zeˇcevi´c, Devendra Singh Dhami, and Kristian Kersting. Probing for correlations +of causal facts: Large language models and causality, 2023. URL https://openreview. +net/forum?id=UPwzqPOs4-. 9 -Earth is closest to the Sun on -June 21, the summer solstice. -So the answer is (b). +Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, +Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick +von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gug- +ger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art +In Proceedings of the 2020 Conference on Empirical Methods +natural language processing. +in Natural Language Processing: System Demonstrations, pp. 38–45, Online, October 2020. +Association for Computational Linguistics. doi: 10.18653/v1/2020.emnlp-demos.6. URL +https://aclanthology.org/2020.emnlp-demos.6. 6, 7, 14 -Florida is closer to the sun during -June, so it will have the most sun- -light energy during that month. So -the answer is (b). +Yuxi Xie, Guanzhen Li, and Min-Yen Kan. Echo: Event causality inference via human-centric -Table 13: Additional examples where self-consistency helps repair the errors over greedy decode on -PaLM-540B. Two sampled reasoning paths that are consistent with the ground truth are shown. +reasoning. arXiv preprint arXiv:2305.14740, 2023. 1 -A.3 FULL SETS OF PROMPTS +Matej Zeˇcevi´c, Moritz Willig, Devendra Singh Dhami, and Kristian Kersting. Causal parrots: Large +language models may talk causality but are not causal. arXiv preprint arXiv:2308.13067, 2023. 1 -We list the full details of the prompts used for two newly-introduced datasets, AQUA-RAT (Ling -et al., 2017) and AI2 Reasoning Challenge (ARC) (Clark et al., 2018), where we manually composed -the example chain-of-thought in this paper, in Table 14 and Table 15, respectively. +Kun Zhang and Aapo Hyvärinen. Causality discovery with additive disturbances: An information- +theoretical perspective. In Machine Learning and Knowledge Discovery in Databases: European +Conference, ECML PKDD 2009, Bled, Slovenia, September 7-11, 2009, Proceedings, Part II 20, +pp. 570–585. Springer, 2009. 3 -Table 14: Few-shot exemplars for AQUA-RAT. -Q: John found that the average of 15 numbers is 40. If 10 is added to each number then the mean of the -numbers is? Answer Choices: (a) 50 (b) 45 (c) 65 (d) 78 (e) 64 +Li Zhang, Qing Lyu, and Chris Callison-Burch. Reasoning about goals, steps, and temporal ordering +with WikiHow. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language +Processing (EMNLP), pp. 4630–4639, Online, November 2020. Association for Computational +Linguistics. doi: 10.18653/v1/2020.emnlp-main.374. URL https://aclanthology.org/ +2020.emnlp-main.374. 9 -A: If 10 is added to each number, then the mean of the numbers also increases by 10. So the new mean -would be 50. The answer is (a). +Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher +Dewan, Mona T. Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt +Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. +OPT: open pre-trained transformer language models. CoRR, abs/2205.01068, 2022. doi: 10.48550/ +arXiv.2205.01068. URL https://doi.org/10.48550/arXiv.2205.01068. 1 -Q: If a / b = 3/4 and 8a + 5b = 22,then find the value of a. Answer Choices: (a) 1/2 (b) 3/2 (c) 5/2 (d) 4/2 (e) -7/2 +13 -A: If a / b = 3/4, then b = 4a / 3. So 8a + 5(4a / 3) = 22. This simplifies to 8a + 20a / 3 = 22, which means -44a / 3 = 22. So a is equal to 3/2. The answer is (b). + Published as a conference paper at ICLR 2024 + +A IMPLEMENTATION DETAILS + +When finetuning on our data, for GPT-based models, we use the default settings of the OpenAI +finetuning API; and for BERT-based models, we use the transformers library (Wolf et al., 2020) +and train the models on a server with an NVIDIA Tesla A100 GPU with 40G of memory. To fit for +the GPU memory, we set the batch size to be 8. We use the validation set to tune the learning rate, +which takes value in {2e-6, 5e-6, 1e-5, 2e-5, 5e-5}; dropout rate, which takes value in {0, 0.1, 0.2, +0.3}; and weight decay, which takes value in {1e-4, 1e-5}. We train the models until convergence, +which is usually around ten epochs. + +Prompts When querying the autoregressive LLMs, we formulate the prompt as follows: + +Question: [premise] + +Can we deduct the following: [hypothesis]? Just answer "Yes" or "No." + +Answer: + +B GENERATING NATURAL STORIES + +To generate the natural stories based on our symbolic expressions, we utilize the state-of-the-art LLM, +GPT-4, which is very good at story generation. We design detailed instructions in the prompt, and +generate around 200 stories in our case study. We show two examples stories in Table 7, and the +report the overall statistics in Table 8. + +Symbolic +Form + +Example 1 (Label=Negative) +Premise: Suppose there is a closed system +of 2 variables, A and B. All the statistical +relations among these 2 variables are as fol- +lows: A correlates with B. +Hypothesis: A directly affects B. +Relation between the promise and hypothe- +sis: The premise does not a necessary con- +dition for the hypothesis. + +Natural +Story + +Premise: Suppose there is a closed system +of 2 variables, ice cream sales and swim- +ming pool attendance. All the statistical +relations among these 2 variables are as fol- +lows: ice cream sales correlate with swim- +ming pool attendance. +Hypothesis: Ice cream sales directly affect +swimming pool attendance. +Relation between the premise and hypoth- +esis: The premise does not provide a nec- +essary condition for the hypothesis. The +correlation between ice cream sales and +swimming pool attendance could be due to +a third variable, such as hot weather, which +increases both ice cream sales and swim- +ming pool attendance. Therefore, it is not +necessarily true that ice cream sales directly +affect swimming pool attendance. + +Example 2 (Label=Positive) +Premise: Suppose there is a closed system +of 3 variables, A, B and C. All the statisti- +cal relations among these 3 variables are as +follows: A correlates with C. B correlates +with C. However, A is independent of B. +Hypothesis: A directly affects C. +Relation between the promise and hypoth- +esis: The premise is a necessary condition +for the hypothesis. So if the premise is true, +the hypothesis must be true. +Premise: Let’s consider three factors: eat- +ing junk food (A), obesity (C), and watch- +ing television (B). There is a correlation +between eating junk food and obesity, and +between watching television and obesity. +However, eating junk food and watching +television are independent from each other. +Hypothesis: Eating junk food directly af- +fects obesity. +Relation between the premise and hypoth- +esis: The premise provides the necessary +conditions for the hypothesis. It establishes +the independent variables A (eating junk +food) and B (watching television) and their +correlations with obesity. Given that these +are true, it supports the hypothesis that eat- +ing junk food directly affects obesity. + +Table 7: Examples of natural stories generated based on the symbolic form in our CORR2CAUSE +dataset, showing the broad application value of our dataset as the starting point for various verbaliza- +tions of the correlation-to-causation inference task. -Q: A person is traveling at 20 km/hr and reached his destiny in 2.5 hr then find the distance? Answer Choices: -(a) 53 km (b) 55 km (c) 52 km (d) 60 km (e) 50 km +14 -A: The distance that the person traveled would have been 20 km/hr * 2.5 hrs = 50 km. The answer is (e). + Published as a conference paper at ICLR 2024 -Q: How many keystrokes are needed to type the numbers from 1 to 500? Answer Choices: (a) 1156 (b) 1392 -(c) 1480 (d) 1562 (e) 1788 +Test Set Size +Dev Set Size +# Tokens/Premise +# Tokens/Hypothesis +# Tokens/Explanation +% Positive Labels -A: There are 9 one-digit numbers from 1 to 9. There are 90 two-digit numbers from 10 to 99. There are 401 -three-digit numbers from 100 to 500. 9 + 90(2) + 401(3) = 1392. The answer is (b). +102 +102 +64.88 +13.54 +64.66 +1.67 -19 +Table 8: Statistics of our generated natural stories. We report the number of samples in the test and de- +velopment sets; number of tokens per premise (# Tokens/Premise), hypothesis (# Tokens/Hypothesis), +and explanation (# Tokens/Explanation); and percentage of the positive labels (% Positive Labels). - Published as a conference paper at ICLR 2023 +For more information, the exact prompt we use is “Here is a causal inference rule: [symbolic +form] Please provide a real-world example instantiating this phenomenon. Format it also as +"Premise:", "Hypothesis:", and "Relation between the promise and hypothesis:".” -Table 15: Few-shot exemplars for ARC easy/challenge. -Q: George wants to warm his hands quickly by rubbing them. Which skin surface will produce the most -heat? (a) dry palms. (b) wet palms. (c) palms covered with oil. (d) palms covered with lotion. +C TEMPLATES AND PARAPHRASES -A: Dry surfaces will more likely cause more friction via rubbing than other smoother surfaces, hence dry -palms will produce the most heat. The answer is (a). +We use the verbalization templates in Table 9 to compose the hypotheses for all six causal relations. -Q: Which factor will most likely cause a person to develop a fever? (a) a leg muscle relaxing after exercise. -(b) a bacterial population in the bloodstream. (c) several viral particles on the skin. (d) carbohydrates being -digested in the stomach. +{Var i} directly causes {Var j}. +{Var i} causes something else which causes {Var j}. +{Var j} directly causes {Var i}. +{Var j} is a cause for {Var i}, but not a direct one. +There exists at least one collider (i.e., common effect) of {Var i} and {Var j}. +There exists at least one confounder (i.e., common cause) of {Var i} and {Var j}. -A: Option (b), bacterial population is the most likely cause for a person developing fever. The answer is (b). +Causal Relation Hypothesis Template +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder +Paraphrases +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder -Q: Which change in the state of water particles causes the particles to become arranged in a fixed position? -(a) boiling. (b) melting. (c) freezing. (d) evaporating. +{Var i} directly affects {Var j}. +{Var i} influences {Var j} through some mediator(s). +{Var j} directly affects {Var i}. +{Var j} influences {Var i} through some mediator(s). +{Var i} and {Var j} together cause some other variable(s). +Some variable(s) cause(s) both {Var i} and {Var j}. -A: When water is freezed, the particles are arranged in a fixed position; the particles are still moving for all -other options. The answer is (c). +Table 9: Templates and their paraphrases for each causal relation in the hypothesis. We use {Var +i} and {Var j} as placeholders for the two variables. -Q: When a switch is used in an electrical circuit, the switch can (a) cause the charge to build. (b) increase -and decrease the voltage. (c) cause the current to change direction. (d) stop and start the flow of current. +D CHANGE LOG FOR THE DATASET VERSION UPDATE -A: The function of a switch is to start and stop the flow of a current. The answer is (d). +Two Equivalent Forms +ß Is-Parent(i, j) +Is-Child(j, i) -Table 16: Few-shot exemplars for HotpotQA (closed-book setting). +ß Is-Ancestor(i, j) (Original) -Q: Which magazine was started first Arthur’s Magazine or First for Women? +Is-Descendent(j, i) (Original) +ß Is-Ancestor(i, j) (Paraphrased) -A: Arthur’s Magazine started in 1844. First for Women started in 1989. So Arthur’s Magazine was started first. -The answer is Arthur’s Magazine. +Is-Descendent(j, i) (Paraphrased) -Q: The Oberoi family is part of a hotel company that has a head office in what city? +ß Has-Collider(i, j) +Has-Collider(j, i) +ß Has-Confounder(i, j) +Has-Confounder(j, i) -A: The Oberoi family is part of the hotel company called The Oberoi Group. The Oberoi Group has its head -office in Delhi. The answer is Delhi. +Duplication Property -Q: What nationality was James Henry Miller’s wife? +De-Duplication Method -A: James Henry Miller’s wife is June Miller. June Miller is an American. The answer is American. +Two exact same strings -Q: The Dutch-Belgian television series that "House of Anubis" was based on first aired in what year? +Keep only one, by forcing i < j -A: "House of Anubis" is based on the Dutch–Belgian television series Het Huis Anubis. Het Huis Anubis is first -aired in September 2006. The answer is 2006. +Two different strings, but +semantically equivalent -As additional information, we also list the exact set of prompts used for all arithmetic reasoning tasks -in Table 17, since there are multiple sets of prompts introduced in Wei et al. (2022). The prompts for -CommonsenseQA and StrategyQA are the same as used in Wei et al. (2022). +Randomly sample one out of the two -We provide the exact prompts used for common NLP tasks in the following tables as well, including -NLI (Table 18, Table 19, Table 20) and Closed-Book Question-Answering tasks (Table 16, Table 21). +Two exact same strings -20 +Keep only one, by forcing i < j - Published as a conference paper at ICLR 2023 +Two different strings, but +semantically equivalent -Table 17: Few-shot exemplars for all arithmetic reasoning tasks, from Wei et al. (2022). -Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done, -there will be 21 trees. How many trees did the grove workers plant today? +Two different strings, but +semantically equivalent -A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted. -So, they must have planted 21 - 15 = 6 trees. The answer is 6. +Randomly sample one out of the two -Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot? +Randomly sample one out of the two -A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5. +Table 10: De-duplication methods for the six causal relation types and their verbalizations. -Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total? +De-Duplication Strategy As mentioned in Section 3.7 in the main paper, our original dataset (v1.0) +has duplication due to symmetric relations and verbalizations. We introduce in Table 10 several +reasons for why duplicated hypotheses exist in our original data. One typical reason is symmetric +relations such as Is-Parent(A, B) and Is-Child(B, A), and, similarly, the paraphrased version of -A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74 -chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39. +15 -Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops -did Jason give to Denny? + Published as a conference paper at ICLR 2024 + +Is-Ancestor(A, B) and Is-Descendent(B, A). Another typical reason is the semantic equivalence in +the verbalization templates, which applies to the Has-Collider and Has-Confounder relations. For +example, the verbalized texts of Has-Collider(A, B) and Collider(B, A) are “There exists at least one +collider (i.e., common effect) of {A and B, B and A},” respectively, which are semantically-equivalent +paraphrases of each other, so we randomly keep one out of the two. + +Resulting Dataset Statistics after De-Duplication + +Since the reason for duplication in the first place is due to symmetry in the causal relation, or +verbalization, the resulting new data, CORR2CAUSE v2.0, is exactly a half of the original data. As we +reported previously in Table 3 of Section 3.7, the total number of samples cuts down to half, while +the label distribution and all other properties are the same. To compose each split, we apply the same +de-duplication method for the test, train, and development sets. We notice that some duplicates are +across the splits, so we prioritize keeping the test and training sets untouched (to minimally affect the +experimental results), and then reduce the development set by removing the cross-split duplicates, +namely: + +• test_2.0 = deduplicate(test_1.0) +• train_2.0 = deduplicate(train_1.0) +• dev_2.0 = deduplicate(dev_1.0) \ {test_2.0, train_2.0} + +We expect minimal or almost no change to the experimental results. In case of the slight possibility +that this change in the development set might affect the model selection in the training process, future +work can feel free to re-train the models and update the exact performance number. + +E SPURIOUS CORRELATION ANALYSIS + +The inspirations of our two robustness tests (paraphrasing and variable refactorization) come from our +data analysis. We check for spurious correlations in the data by reporting in Table 11 the point-wise +mutual information (PMI) between the label and any n-gram with no more than four tokens. In +addition, we also report the difference of the PMI with the two labels in the |Diff| column of Table 11, +and report the top 10 n-grams. + +The design spirit for our robustness test is that if the models’ correct judgment relies on exploiting +these spurious correlations, then such reliance will be broken in our perturbations. + +N-Gram +a cause +a cause for +A causes +A causes something +a direct +a direct one +for D +for D but +for E +for E but + +PMI w/ Non-Ent. Label +1.692209 +1.663640 +1.640679 +1.621820 +1.606052 +1.592673 +1.584826 +1.583897 +1.582980 +1.582074 + +PMI w/ Ent. Label +-1.025611 +-0.983790 +-0.951610 +-0.926075 +-0.905316 +-0.888107 +-0.878180 +-0.877014 +-0.875864 +-0.874728 + +|Diff| +2.717820 +2.647430 +2.592289 +2.547895 +2.511369 +2.480781 +2.463006 +2.460911 +2.458844 +2.456802 + +Table 11: PMI between the labels and n-grams. The labels include non-entailment (Non-Ent.) and +entailment (Ent.). And the n-grams include all with no more than four words. The |Diff| column +shows the absolute value of the difference between the PMIs with two labels. We show the top 10 +n-grams with the largest differences of their PMIs with the two classes in the |Diff| column. + +We can see that some spurious correlations are rooted in the framing of the hypothesis, such as “a +cause (for)”, and “a direct (one)” (which we use the paraphrasing task to break), and others are +connected to the variable names, such as “for D (but)” and “for E (but)” (which we use the variable +refactorization to break). + +F FINE-GRAINED ERROR ANALYSIS + +In addition to the fine-grained analysis by causal relation type in Table 6a for fine-tuned models, we +also report such error analysis for non-finetuned models in Table 12. -A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of -lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8. +16 -Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does -he have now? + Published as a conference paper at ICLR 2024 + +Relation Type +All +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder +All +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +Selected Models +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +RoBERTa MNLI All +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI Has-Collider +RoBERTa MNLI Has-Confounder + +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant + +F1 +21.69 +8.82 +0 +9.84 +14.29 +34.24 +15.33 +29.08 +0 +30.77 +0 +26.98 +44.1 +20.67 +22.79 +0 +0 +0 +0 +43.45 +0 -A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so -in total he has 7 + 2 = 9 toys. The answer is 9. +Precision Recall Accuracy +27.78 +4.62 +0 +5.17 +17.86 +52.07 +56.76 +47.66 +0 +30.3 +0 +60.71 +81.82 +100 +16.96 +0 +0 +0 +0 +47.93 +0 -Q: There were nine computers in the server room. Five more computers were installed each day, from -monday to thursday. How many computers are now in the server room? +69.46 +83.47 +90.67 +85.33 +84 +35.12 +37.8 +64.6 +82.67 +88 +84.53 +75.47 +32.71 +23.86 +82.5 +82.67 +91.2 +84.53 +92.53 +59.52 +84.45 + +17.79 +100 +0 +100 +11.9 +25.51 +8.86 +20.92 +0 +31.25 +0 +17.35 +30.18 +11.53 +34.73 +0 +0 +0 +0 +39.73 +0 -A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 = -20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers. -The answer is 29. +Table 12: Fine-grained evaluation results for some selected non-fine-tuned models. -Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many -golf balls did he have at the end of wednesday? +These results are particularly revealing, showing how off-the-shelf models perform in recognizing +specific relations. Specifically, GPT-3.5 cannot recognize ancestor relations, whereas GPT-4 fails at +all direct causation recognition with parents and children. And RoBERTa MNLI only did collider +relation relatively correctly. Note that, when the F1 score is zero, the accuracy number is a result of +always predicting the negative class of that relation. -A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On -Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33. +G LLM PERFORMANCE OPTIMIZATION -Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left? +Since our experiments in Section 4.2 are based on plain, zero-shot prompts, we explore whether better +prompting strategies could improve the performance. We enhance the query prompt by incorporating +several strategies: (1) Utilizing a system prompt that specifies the model’s expertise (“You are a highly +intelligent question-answering bot with profound knowledge of causal inference.”); (2) Including +a pair of few-shot examples, one positive and one negative; (3) Implementing chain-of-thought +prompting with “Let’s think step by step.” to encourage the language model to generate step-by-step +reasoning. In Table 13, we present the evaluation results on the relatively affordable model, GPT-3.5, +where the optimized prompt leads to a 4-point improvement in F1 over the original performance. +However, we can see that despite the deployment of all three strategies, the model continues to +struggle with this challenging task. -A: She bought 5 bagels for $3 each. This means she spent 5 * $3 = $15 on the bagels. She had $23 in -beginning, so now she has $23 - $15 = $8. The answer is 8. +GPT-3.5 (plain query; original) +GPT-3.5 (enhanced query) -21 +F1 +21.69 +25.44 - Published as a conference paper at ICLR 2023 +Precision +17.79 +17.29 -Table 18: Few-shot exemplars for ANLI. +Recall Accuracy +27.78 +48.11 -Premise: -"Conceptually cream skimming has two basic dimensions - product and geography." -Based on this premise, can we conclude the hypothesis "Product and geography are what make cream skimming -work." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: Based on "cream skimming has two basic dimensions" we can’t infer that these two dimensions are what -make cream skimming work. The answer is it is not possible to tell. - -Premise: -"One of our member will carry out your instructions minutely." -Based on this premise, can we conclude the hypothesis "A member of my team will execute your orders with -immense precision." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "one of" means the same as "a member of", "carry out" means the same as "execute", and "minutely" means -the same as "immense precision". The answer is yes. - -Premise: -"Fun for adults and children." -Based on this premise, can we conclude the hypothesis "Fun for only children." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "adults and children" contradicts "only children". The answer is no. - -Premise: -"He turned and smiled at Vrenna." -Based on this premise, can we conclude the hypothesis "He smiled at Vrenna who was walking slowly behind -him with her mother." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: the premise does not say anything about "Vrenna was walking". The answer is it is not possible to tell. - -Premise: -"well you see that on television also" -Based on this premise, can we conclude the hypothesis "You can see that on television, as well." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "also" and "as well" mean the same thing. The answer is yes. - -Premise: -"Vrenna and I both fought him and he nearly took us." -Based on this premise, can we conclude the hypothesis "Neither Vrenna nor myself have ever fought him." is -true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "Vrenna and I both" contradicts "neither Vrenna nor myself". The answer is no. - -22 - - Published as a conference paper at ICLR 2023 - -Table 19: Few-shot exemplars for e-SNLI (the rationales are crowd-sourced annotations from the -original dataset). - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is training his horse for a competition." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: The person is not necessarily training his horse. The answer is it is not possible to tell. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is at a diner, ordering an omelette." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: One jumping horse cannot be in a diner ordering food. The answer is no. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is outdoors, on a horse." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: A broken down airplane is outdoors. The answer is yes. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "They are smiling at their parents." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: Just because they are smiling and waving at a camera does not imply their parents or anyone is anyone behind -it. The answer is it is not possible to tell. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "The kids are frowning." is true? OPTIONS: -- yes -- no -- it is not possible to tell -A: One cannot be smiling and frowning at the same time. The answer is no. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "There are children present." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A:The children must be present to see them smiling and waving. The answer is yes. - -23 - - Published as a conference paper at ICLR 2023 - -Table 20: Few-shot exemplars for RTE. - -Premise: -"No Weapons of Mass Destruction Found in Iraq Yet." -Based on this premise, can we conclude the hypothesis "Weapons of Mass Destruction Found in Iraq." is true? - -A: "No Weapons of Mass Destruction Found" contradicts "Weapons of Mass Destruction Found". The answer is -no. - -Premise: -"A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful -gathered in downtown Chicago to mark the installation of new Pope Benedict XVI." -Based on this premise, can we conclude the hypothesis "Pope Benedict XVI is the new leader of the Roman -Catholic Church." is true?’ - -A: "installation of new Pope Benedict XVI." means "Pope Benedict XVI is the new leader". The answer is yes. - -Premise: -"A man is due in court later charged with the murder 26 years ago of a teenager whose case was the first to -be featured on BBC One’s Crimewatch. Colette Aram, 16, was walking to her boyfriend’s house in Keyworth, -Nottinghamshire, on 30 October 1983 when she disappeared. Her body was later found in a field close to her -home. Paul Stewart Hutchinson, 50, has been charged with murder and is due before Nottingham magistrates -later." -Based on this premise, can we conclude the hypothesis "Paul Stewart Hutchinson is accused of having stabbed a -girl." is true? - -A: The premise does not say Paul Stewart Hutchinson "stabbed" this girl. The answer is no. - -Premise: -"Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it -will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients." -Based on this premise, can we conclude the hypothesis "Herceptin can be used to treat breast cancer." is true? +69.46 +52.01 -A: "Herceptin was approved to treat breast cancer" implies that "Herceptin can be used to treat breast cancer". -The answer is yes. - -Table 21: Few-shot exemplars for BoolQ (closed-book setting). - -Q: does system of a down have 2 singers? - -A: System of a Down currently consists of Serj Tankian, Daron Malakian, Shavo Odadjian and John Dolmayan. -Serj and Daron do vocals, so the band does have two singers. The answer is yes. - -Q: do iran and afghanistan speak the same language? - -A: Iran and Afghanistan both speak the Indo-European language Persian. The answer is yes. - -Q: is a cello and a bass the same thing? - -A: The cello is played sitting down with the instrument between the knees, whereas the double bass is played -standing or sitting on a stool. The answer is no. - -Q: can you use oyster card at epsom station? - -A: Epsom railway station serves the town of Epsom in Surrey and is not in the London Oyster card zone. The -answer is no. +Table 13: Performance of GPT-3.5 with different queries. We quote the original performance from +Table 4. -24 +17 diff --git a/q143/random_k4/question.json b/q143/random_k4/question.json index 52e224f05f8f5a644d5518153ecfcd615e515a2a..ce6d776324ae7ce74615ce0d10cb0d20c7898cf9 100644 --- a/q143/random_k4/question.json +++ b/q143/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "germanwingsdigitalcrisisanalysis-150403064828-conversion-gate01_95.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf", - "web_ff8dcb0ea2a037b7.pdf", - "web_4ae8d7cc08fa2f24.pdf", - "web_365d195edde4e8ba.pdf" + "web_5e3d260ea31a243f.pdf", + "resize_r2_22.pdf", + "resize_r1_16.pdf", + "web_eff43bab789f7c21.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q143/random_k4/random_1.md b/q143/random_k4/random_1.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..f94127163277f669a14f3b0a86d9bfd3ebbd9e38 100644 --- a/q143/random_k4/random_1.md +++ b/q143/random_k4/random_1.md @@ -1,1453 +1,6514 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA -8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +C H A P T E R 4 + +Climate Change Impacts +on Human Health + +Song Liang1,5, Kristina Kintziger2, Phyllis Reaves3, and Sadie J. Ryan4,5 + +1Department of Environmental and Global Health, College of Public Health and Health Professions, +University of Florida, Gainesville, FL; 2Department of Public Health, College of Education, Health & +Human Sciences, The University of Tennessee, Knoxville, TN; 3Division of Physical Therapy, School of +Allied Health Sciences, Florida Agricultural & Mechanical University, Tallahassee, FL; 4Department of +Geography, University of Florida, Gainesville, FL; 5Emerging Pathogens Institute, University of Florida, +Gainesville, FL + +Climate change poses major challenges to human society and to Earth systems, influencing the +functioning of many ecosystems and thereby affecting human health. Many climate change/variability- +and extreme weather-associated events, such as sea level rise, hurricanes, and storm surge, as well as +other weather extremes, including excessive precipitation and heatwaves, have direct and/or indirect +impacts on human health. These impacts include death/injury, cardiovascular and respiratory diseases, +environmentally-mediated infectious diseases, and mental health, among others. Due to its unique +geography, Florida is particularly vulnerable to these environmental impacts, which have important +health implications for the state’s more than 20 million residents. In this chapter, we review the health +impacts of climate change and associated weather events, with an emphasis on those relevant to Florida, +and environmental hazards, including hurricanes and storms, lightning, sea level rise, excessive +precipitation, extreme heat, and drought. There is clear evidence for significant climate-sensitive hazards +and human health impacts in the state, despite uncertainties associated with the assessment of some +effects. To address health impacts and challenges, policies focused on mitigation and adaptation +strategies, health surveillance, and research that could close knowledge gaps on human exposures to the +climate-sensitive hazards and health impacts are needed. + +Key Messages + +• Florida is highly vulnerable to climate-sensitive hazards (e.g. sea level rise, heat waves, storm + +surge, and hurricanes), which have a wide range of human health effects. + +• The health effects can be direct, such as storm/temperature related illnesses, injuries, and +deaths; or indirect, such as waterborne, food-borne, and vector-borne diseases; or take social +and economic pathways, such as stress and mental illness. + +• The health effects exhibit substantial regional disparities across the state. +• Policies focused on health surveillance and research on knowledge gaps between human + +exposure to the hazards and health effects are much needed. + +Keywords + +Climate change; Environmental hazards; Human health; Florida + +125 + + 1 2 6 • S O N G L I A N G E T A L . + +Introduction + +C + +limate change is well-characterized at the global level using metrics such as the global +mean temperature and sea level. But the manifestations of global climate change and +variability occur at varying geographical scales, causing a variety of weather events, +such as excessive rainfall, drought, severe storm/flooding, sea level rise, and heat waves (Field +et al. 2014), all of which are likely to have direct and/or indirect human health impacts +(McMichael and Haines 1997; Colwell et al. 1998; Frumkin et al. 2008). Indeed, there is clear +and increasing evidence that many health outcomes (most adverse) exhibit high sensitivities to +these varying weather events. The interface between the climate change-associated weather +events and human health exhibits a complex web of relationships involving both natural and +social environments through direct or indirect impact pathways, social institutional disruption, +or a combinations of these things (Fig. 4.1). Globally, substantial disease burdens are attributable +to hazards that are associated with climate change and related weather events. According to the +World Health Organization’s 2008 estimate, climate change-associated hazards were responsible +for more than 150,000 deaths and 5,517,000 disability-adjusted life years, which were highly +likely to be underestimated as the estimates were only based on selected risk factors and illnesses +associated with climate change and related weather events (WHO 2008). The actual disease +burden might be greater and the adverse health impacts are likely to increase in the years to come, +as the Intergovernmental Panel on Climate Change (IPCC) and US Global Change Research +Program have clearly indicated that the accelerated changing climate poses a substantial threat +to global human health and the risk will continue to become severe if no remediation action is +taken (Field et al. 2014; Balbus et al. 2016). + +Due to its unique geography, Florida is highly vulnerable to a variety of hazards associated +with climate change and variability, and related weather events. Sea level rise, hurricane and +storm surge, excessive precipitation, and heatwaves all pose threats to Florida’s agriculture, +ecosystems, tourism, and public health. For example, Florida has experienced a greater number +of hurricane landfalls than other states in the country (Knight and Davis 2009), and more intense +hurricanes are expected in the future (Knutson et al. 2013). In addition, rising sea levels are +influencing and expected to continue to have significant impacts on communities and residents +along Florida’s coastal areas, particularly in the southern part of the state, making Floridians +exceptionally vulnerable to these environmental hazards (Elsner et al. 2008). + +This chapter is organized into sections addressing impact pathways and specific +environmental hazards of particular concern in Florida (Fig. 4.1). For direct impacts, we focus +on extreme heat, flooding, storms and lightning, and sea level rise. For indirect impacts on public +health, we focus on water, food, and vector-borne diseases. And lastly, we conclude with our +discussion of impact pathways through social disruption on mental and community health impact +associated with climate change and related weather events. Throughout this chapter, health + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 7 + +impacts and implications, particularly those pertinent to Florida, are reviewed primarily based +on the Building Resilience Against Climate Effects (BRACE) project reports from the Florida +Department of Health (FDH) (FDH 2015b, a) and other relevant publications. + +Figure 4.1. Impact pathways by which climate change and variability and related weather events affect +human health. + +Direct Impact Pathways – Environmental Hazards +and Human Health + +A number of human health impacts come from direct exposure to environmental hazards +associated with climate change and related weather events. Among those of particular concern +in Florida are extreme heat, flooding, storms and lightning, and sea level rise. + + 1 2 8 • S O N G L I A N G E T A L . + +Extreme Heat and Health Impacts + +Extreme climate and weather events (e.g., heat, cold, storms, and floods) are expected to occur +more frequently worldwide due to climate change. These extreme events, such as heatwaves and +cold spells (e.g. frequent very hot days and possibly fewer very cold days), have a direct impact +on human health by compromising the body’s ability to regulate its internal temperature. +Increased human morbidity and mortality have been associated with extreme temperatures, both +hot and cold, and documented in both developing and developed countries worldwide (Center +for Disease Control and Prevention 1994; Keatinge et al. 2000a; Keatinge et al. 2000b; Basu and +Samet 2002; Medina-Ramon et al. 2006; Kovats and Hajat 2008; Bandyopadhyay et al. 2012). +The mechanistic effects of extreme temperature on death and illness are briefly summarized +below (Sarofim et al. 2016): +• Extreme heat may induce heat cramps, heat exhaustion, heatstroke, and hyperthermia. +• Extreme cold may induce hypothermia and frostbite. +• Extreme heat/cold may exacerbate chronic conditions related to cardiovascular disease, + +respiratory disease, cerebrovascular disease, and diabetes-related illnesses. + +• Extended exposure to high temperatures may induce increased hospital admissions for + +cardiovascular, kidney, and respiratory disorders. + +• Extreme heat may induce or exacerbate mental health and behavioral disorders. + +A number of studies have reported associations between high temperatures and human deaths +or illnesses around the world. For example, a large number of heat-related deaths have been +reported in the United States (Center for Disease Control and Prevention 1994; Curriero et al. +2002; Hoshiko et al. 2010), Europe (Keatinge et al. 2000a; Keatinge et al. 2000b; Huynen et al. +2001), and Asia (Qian et al. 2008; Chung et al. 2009). A recent study by Berko et al. examined +deaths attributed to extreme weather events from 2006 to 2010 in the United States and found +that each year about 2,000 U.S. residents died from weather-related causes, and among those +31% died due to exposure to excessive heat, heat stroke, and/or sun stroke (Berko et al. 2014). +Mounting evidence has suggested that substantial increases in mortality from respiratory and +cardiovascular diseases, in particular among the elderly and youth, are associated with high +temperatures in the United States (Basu and Samet 2002; Medina-Ramon et al. 2006; Kovats and +Hajat 2008). During the 2006 California heat wave, 16,166 excess emergency department visits +and 1,182 excess hospitalizations were reported statewide. This included a significant increase +in emergency department visits for acute renal failure, cardiovascular diseases, and diabetes +(Knowlton et al. 2009). The association between heat and respiratory diseases has also been +reported in many studies. In New York state, a significant number of respiratory hospitalizations +were attributable to excessive heat, and a projection suggested that excess respiratory admissions +in the state due to extreme heat will be two to six times higher in 2080–2099 than what was seen +in 1991–2004 (Lin et al. 2012). In another study conducted in Greater London, UK, an analysis + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 9 + +of historical data suggested that heat-related increases in emergency admissions for respiratory +and renal diseases had been observed in children under five, and for respiratory disease in the +75+ age group (Kovats et al. 2004). + +Extreme Heat and Public Health in Florida + +Florida has a warm subtropical and tropical climate. Due to continuous summer heat and extreme +weather events, temperatures around the state can reach levels potentially harmful to human +health. Through analyses of historical data (for the periods 1895–2009 and 1970–2009), a +significant trend in Florida’s mean temperature has been found (Martinez et al. 2012). The +average summer temperature (June–August) has been showing an overall increasing trend, +particularly after the 1980s, which shows a greater upward trend (FDH 2015a), likely due to a +variety of factors ranging from rapid urban development (Martinez et al. 2012; FDH 2015a) to +climate change (Ji et al. 2014). Within the state, substantial spatial heterogeneities in mean, +minimum, and maximum temperature have been reported (Martinez et al. 2012). For example, +the distribution of annual mean number of days with a maximum temperature greater than or +equal to 95 °F and greater than or equal to 75 °F showed that more days with hottest daytime +temperatures occurred in the northern and interior portions of the state (FDH 2015a). + +Using statewide information on specific health outcome-related emergency department visits +that occurred May to October from 2005–2012, the BRACE project examined the effects of daily +maximum temperature or daily maximum heat index on emergency department visits through a +two-stage analysis—the first stage involved the analysis on regional (National Weather Service +regions) scales using the Poisson regression model; the second stage used a meta-analysis +technique to integrate regional estimates into a statewide estimate (FDH 2015a). Rate ratios, +based on the comparison between the exposure (e.g., hotter days) and the reference (e.g. 88 °F +for temperature and 94 °F for heat index), were used to compare emergency department visits +for specific health conditions (FDH 2015a). +a) Heat and heat-related illness. A strong association between heat-related illness and heat +exposure was observed; the significant increase in emergency department visit rates for heat- +related illness was associated with increasing temperatures and a strong dose-response +relationship across all regions in Florida was found (FDH 2015a). + +b) Heat and cardiovascular disease. Two types of cardiovascular disease were considered— +myocardial infarction and ischemic stroke. No statistically significant association was found +between temperature and ischemic stroke at the state level; however, significant regional +variations were identified. For example, the Tampa region showed a significant positive +association between heat index (above the reference level at 94 °F) and stroke (FDH 2015a). +Furthermore, a positive, statistically significant relationship was observed between +myocardial infarction and heat index despite remarkable geographical variations (FDH +2015a). + + 1 3 0 • S O N G L I A N G E T A L . + +c) Heat and respiratory disease. Overall, increases in the number of emergency department +visits for asthma were associated with higher temperatures. Although no significant +association was found at the state level between temperature and emergency department visits +for asthma, the regional analysis suggested that all regions except Tallahassee showed a +statistically significant positive association between temperature and asthma (FDH 2015a). + +Flooding, Storms, Lightning, and Sea Level Rise + +Excessive precipitation, hurricanes, coastal storms, sea level rise, and thunderstorm-related +lightning, which are all typically accompanied by coastal and inland flooding, have the potential +for substantial direct human health impacts (e.g. injury and death, maternal and child health, and +mental health issues). Natural disasters related to flooding and storms are a significant cause of +mortality and morbidity. For example, according to the International Federation of Red Cross +and Red Crescent Societies (IFRCRCS), more than 3,448 climatic, hydrological and +meteorological disasters (including 1,751 major floods and 988 storms) resulting in 339,710 +deaths (59,092 and 177,685 deaths specifically tied to flooding and storms) were reported +worldwide between 2005 and 2014 (IFRCRCS 2015). Lightning strike is the second leading +cause of weather-related mortality at the global scale, with an estimated 0.2-1.7 deaths per million +people, and those who survive lightning strikes often suffer from significant injuries (Aslar et al. +2001; Ritenour et al. 2008). The magnitude and impacts of these extreme weather events, their +severity, and the extent of the effects are influenced by many different factors. For example, +short- and long-term averages and variability of weather conditions and physical impacts of +associated extreme events, as well as social environmental factors (e.g. infrastructure, social and +individual vulnerability), are considered important (Bouma et al. 1997, Kovats 2000, Kovats et +al. 2003, Miranda 2004). For projections of future flooding under different climate change +scenarios, climate models have consistently suggested that episodes of severe flooding may +become more frequent in inland river systems including flood plains (Christensen and +Christensen 2003, Booij 2005), urban and coastal environments in various parts of the world +(Schreider et al. 2000; Douglas et al. 2008; Kirshen et al. 2008; Thompson et al. 2009, Diez et +al. 2011; Lyle and Mills 2016). Meanwhile, climate models also predict positive correlations +between lightning and global temperatures, suggesting the likelihood of a greater number of and +more severe lightning episodes (Price and Rind 1994; Reeve and Toumi 1999; Kochtubajda et +al. 2006). These episodes are expected to increase the risk of extreme environmental hazards and +adverse public health impacts. +a) Natural disasters in Florida. Due to its geography (extensive coastline and a peninsular +shape) and its tropical/subtropical climate, Florida is particularly vulnerable to tropical +storms, hurricanes, and lightning strikes. Historically the occurrence of hurricanes in the US +has been clustered in Florida and along the Atlantic coastline (Ellis et al. 2015). A total of 67 +known Florida hurricanes have occurred over the 108-year period (1900-2007) with four + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 1 + +different hurricanes occurring in one year (2004) in Florida. The climate change model +suggests a 46% chance that Florida will be hit by at least one hurricane each year in the future +(Malmstadt et al. 2009), and recent experience seems to support this. Fig. 4.2 illustrates the +distribution of average return time of hurricane landfalls and populations along the coastal +areas of Florida, showing spatial heterogeneities in vulnerability to hurricanes. + +b) Public health impacts. Key direct public health impacts associated with extreme weather +events include injury and death due to trauma, drowning, destructive forces of wind, +collapsed building and trees, and lightning strike and carbon monoxide poisoning related to +power outages. Historically, Florida experienced significantly higher mortality in the early +20th century; approximately 3,000 deaths were attributed to extreme weather-related events +(Winsberg 2003) and the largest number of deaths due to lightning strike in the US were +reported during this time (Duclos et al. 1990; Ritenour et al. 2008). During 2004 and 2005, +eight hurricanes hit Florida resulting in 213 deaths, over half of which were caused by trauma, +followed by drowning, other injury, electrocution, and carbon monoxide poisoning (Ragan et +al. 2008). Injury is common cause of hurricane-related morbidity and mortality. For example, +in the aftermath of Hurricane Katrina between September 8 and October 14, 2005, 7,543 non- +fatal injuries among residents and relief workers were recorded in the surveillance system +(Sullivent et al. 2006). During a 2006 flood event in El Paso County in Texas, 43% of +individuals (out of 475 surveyed) were reported having physical health issues related to the +flooding episode (Collins et al. 2013). + +c) Injury. The BRACE study compared injury-related emergency department visits and +hospitalizations during impact periods (periods covering tropical cyclone landfall) vs. control +periods (periods with no such impact) between 2004 and 2012. The study found that +emergency department visit and hospitalization rates were significantly higher during the +impact periods, with a rate ratio of 1.03 (95% confidence interval (CI): 1.02, 1.05) for the +emergency department visit rate and 1.4 (95% CI: 1.02, 1.05) for the hospitalization rate. The +most common types of injuries during the impact periods included falls, being struck by an +object, being cut or pierced, and motor vehicle transport accidents (FDH 2015b). + +d) Carbon monoxide poisoning. Carbon monoxide (CO) is a colorless, odorless, poisonous gas +that can be harmful when inhaled in a large amount. Inhaled CO enters the blood stream and +reduces the delivery of oxygen to the body’s critical organs, such as the heart and brain. The +gas is primarily generated through incomplete combustion. The greatest sources of CO +outdoors are usually related to motor vehicles, machinery that burn fossil fuels. Indoors, some +major sources include unvented kerosene and gas space heaters, leaking chimneys and +furnaces, and gas stoves. Exposure to CO may cause weakness, headache, dizziness, nausea, +shortness of breath, confusion, and even death at very high concentration levels, which is +more likely in indoor environments. A recent systematic review of the health impacts of +power outages due to extreme events indicated that CO poisoning is an important health +concern (Klinger et al. 2014). In Florida, the BRACE study reported that during the study + + 1 3 2 • S O N G L I A N G E T A L . + +period 2004–2012, the rate of CO exposure calls during impact periods was 6.59 times (95% +CI: 4.48, 9.7) the rate of CO exposure calls to poison control centers during the control +periods. The study also found the rate of CO-poisoning related emergency department visits +was significantly higher for the impact periods than during the control periods with the rate +ratio 3.44 (95% CI: 2.07, 5.72). Finally, the BRACE study revealed that rates of CO +poisoning-related hospitalizations were also significantly higher during the impact periods +with the rate ratio 4.0 (95% CI: 2.9, 5.51) (FDH 2015b). + +Figure 4.2. Historical distribution of hurricane landfalls in Florida. + +Indirect Impact Pathways–Infectious Diseases + +Climate change and variability and associated variable weather conditions can affect human +health through indirect pathways (Fig. 4.1). These indirect impacts are modulated through +biological and ecological processes that can influence infectious diseases (in particular, some +water/food- and vector-borne diseases) and air quality. In this section, we focus on water/food-, +and vector-borne diseases. + +Disease Transmission Groupings + +The impact of climate change and variability and associated weather events on infectious +diseases has been receiving much attention in the past several decades. Across the globe, high + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 3 + +human mortality rates and disability have been attributed to infectious diseases; climate change +is intensifying and will continue to exacerbate transmission of infectious diseases, poses +increasing concerns (Patz et al. 2005). Indeed, the transmission of many infectious diseases is +closely linked to physical environmental factors, such as water and temperature. Many water- +associated and zoonotic pathogens, for instance, which account for more than 75% of known +pathogens causing human diseases, are associated with hydrological process (WHO 2003). From +a transmission perspective, infectious diseases can be broadly classified into two groups: 1) +directly transmitted (i.e., those spread directly from person to person via direct contact or droplet +exposure, such as influenza and tuberculosis); and 2) indirectly transmitted, including those +mediated through an intermediate vector organisms (vector-borne) or through environmental +media such as water and soil (environmentally-mediated) (Eisenberg et al. 2007). Because the +diseases in the indirectly transmitted group are either directly or indirectly related to water, this +group can be further broken down into the following sub-groups based on water’s role in disease +transmission: waterborne (e.g., cholera typically via fecal-oral transmission), water-based (e.g. +part of the pathogen life cycle requires aquatic environment, such as schistosomiasis), and water- +related (e.g., malaria and dengue, which need water for breeding of insect vectors to fulfill the +transmission cycle) (Yang et al. 2012; Mordecai et al. 2013; Ryan et al. 2015; Johnson et al. +2015; Mordecai et al., 2017) . + +Climate-Sensitivity of Microbial Agents and Their Associated Diseases + +Table 4.1 lists some of the major pathogens and their associated diseases that are climate sensitive +and associated with water. These disease-causing pathogens, either through infection or toxin- +generating (e.g. caused by protozoa, bacteria, virus, helminth, or algae), and their associated +vectors (e.g., snails, mosquitoes, sand flies) are very small in size and lack thermostatic +mechanisms (i.e. lack of function to maintain body temperatures). Their biological (e.g., survival, +reproduction, development/growth, and infectivity) and environmental (e.g., contamination and +movement) processes are determined by environmental conditions, such as rainfall, storm runoff, +temperature, and sunlight. Climate change and variability and associated weather events are +expected to affect both fresh and marine water environments thereby changing humans’ exposure +to these water-related contaminants or pathogens that cause ill health. + +a) Waterborne Diseases +Waterborne diseases remain an important contributor to the global burden of disease (Pruss et al. +2002; Yang et al. 2012; Pruss-Ustun et al. 2014). Many different viral, bacterial, and parasitic +diseases have been associated with waterborne transmission (Table 4.1). Human exposure to +waterborne pathogens is primarily through drinking water, recreational water (e.g. via accidental +drinking and, for some pathogens, via dermal contact), or foods. The distribution of these +pathogens exhibits significant geographical variations, depending on both physical + + 1 3 4 • S O N G L I A N G E T A L . + +environmental and socio-economic conditions (Yang et al. 2012). For example, waterborne +cholera, viral hepatitis, and some other water-based and water-related diseases (e.g., +schistosomiasis and onchocerciasis) are largely confined to certain tropical areas where the +prevalence of such diseases is largely due to lack of access to improved drinking water, sanitation +and hygiene (Pruss et al. 2002), while some pathogens, such as Cryptosporidium parvum, +Giardia duodenalis, and Campylobacter spp. have a much wider geographical distribution (Yang +et al. 2012). Regardless of the geographical region, heavy rainfall, flooding, and temperature are +among the most important factors associated with transmission and outbreaks of these +waterborne diseases. + +There is mounting evidence that weather events are often an important factor triggering +waterborne outbreaks. In developed countries, extreme weather events such as excessive rainfall +and flooding can overwhelm water treatment plants and/or increase runoff into recreational +water, leading to water contamination that causes outbreaks (Kistemann et al. 2002). Some recent +studies have clearly indicated that excessive rainfall has been a significant contributor to +waterborne outbreaks. An analysis of historical records of 548 waterborne outbreaks in the US +reported between 1948 and 1994 suggested a significant association between the outbreaks and +rainfall—51% of waterborne disease outbreaks were preceded by precipitation events above the +90th percentile and 68% by events above the 80th percentile; the strongest association was +between the outbreaks and surface water contamination due to extreme precipitation (Curriero et +al. 2001). In Canada, Thomas et al. examined extreme rainfall and spring snowmelt in relation to +92 Canadian waterborne disease outbreaks that occurred between 1975 and 2001. They found +that warmer temperatures and extreme rainfall were significant contributing factors to waterborne +disease outbreaks in Canada (Thomas et al. 2006). Similar studies have been conducted in +England (Nichols et al. 2009), the Netherlands (Schijven and de Roda Husman 2005), Finland +(Miettinen et al. 2001), Denmark (Laursen et al. 1994), and Taiwan (Chen et al. 2012). In many +developing countries, lack of access to improved drinking water and sanitation exacerbates the +impact of these extreme precipitation weather events on waterborne diseases. Diarrhea, caused +by many waterborne pathogens, remains a top killer of children under five in the developing +world, particularly in Africa (Pruss-Ustun et al. 2014). A large number of studies have been +carried out in the developing world. For example, cholera outbreaks caused by toxigenic Vibrio +cholerae have been consistently shown to be correlated with excessive rainfall, flooding, and +high temperatures in the epidemic areas of West Africa and Bangladesh (Mhalu et al. 1987; +Hashizume et al. 2008; Luque Fernandez et al. 2009; Ngwa et al. 2016), although the exact +mechanistic relationships that result in this correlation remain elusive (Colwell et al. 1998). +Significant (positive) relationships have been reported between excessive rainfall and/or high +temperatures and diarrhea in the Pacific Islands (Singh et al. 2001), Ecuador (Carlton et al. 2014), +and Sub-Saharan Africa (Bandyopadhyay et al. 2012), while evidence also suggests that low +rainfall and even drought can also be associated with diarrhea, as reported in Bangladesh +(Hashizume et al. 2008) and Denmark (Senhorst and Zwolsman 2005). + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 5 + +b) Waterborne Diseases and Climate Impact in Florida +Across the US, climate change and variability and associated weather events are affecting and +are expected to continue to affect both marine and freshwater resources. These effects extend to +some water-associated pathogens and related diseases. In Florida, waterborne and foodborne +diseases of major public health concerns are highlighted in Table 1. Enteric bacteria, protozoan +parasites, enteric viruses including Salmonella enterica, Campylobacter spp., toxigenic +Escherichia coli, Vibrio bacteria species, Cryptosporidium and Giardia enteroviruses, +rotaviruses, noroviruses, and hepatitis A and E, are among those closely related to drinking and +recreational waters, and shellfish. These pathogens, while in environmental stages, are sensitive +to temperature, precipitation, and water flow. Several important toxin producers of water sources +are of particular public health importance in Florida (Table 4.1) including toxins from harmful +algal blooms, toxigenic marine species of Alexandrium (causing paralytic shellfish poisoning), +Karenia brevis (causing neurotoxic shellfish poisoning), and Gambierdiscus spp. (causing +ciguatera fish poisoning). Human exposure pathways are primarily through the consumption of +contaminated shellfish and fish, and in recreational waters. Cyanobacteria, consisting of multiple +species, can produce toxins (including microcystin), and the primary human exposure is from +drinking water and recreational water. For all of these harmful algal blooms species, temperature, +and precipitation are among the important factors affecting their reproduction, growth, and +distribution. + +Fig. 4.3 shows yearly distribution of total reported outbreaks and cases of food-borne and +waterborne diseases in Florida from 1989 to 2011 and Fig. 4.4 shows the monthly distribution of +reported outbreaks and cases for 2011, exhibiting marked annual variations and distinct seasonal +distribution. According to the BRACE report for the period 2004–2012, approximately 1,150 +cases of campylobacteriosis were reported annually, with the majority of cases reported between +May and September. From that same report, ~426 cases of cryptosporidiosis were reported +annually, with the majority of cases reported between June and October. Approximately 940 +cases of giardiasis were reported annually, with an annual incidence rate of 5.1 cases per 100,000 +population and the majority of cases reported between May and October. Salmonellosis +accounted for the greatest number of foodborne illnesses reported, with an average of 5,438 cases +reported each year and ~70% reported between June and November. Finally, ~101 cases of +vibriosis were reported annually, with 81% of the cases reported between April and October +(FDH 2015b). Focusing on the five diseases—campylobacteriosis. cryptosporidiosis, giardiasis, +salmonellosis, and vibriosis—the BRACE study identified a total of 1,231 follow-up days of +interest including 775 control days and 456 impact days—42 associated with hurricanes and 414 +with tropical storms (FDH 2015b); the result suggested that the occurrences of cryptosporidiosis +and salmonellosis were significantly associated with the tropical cyclones, with the risk ratio +being 1.26 (95% CI, 1.04, 1.52) and 1.35 (95% CI, 1.29, 1.42), respectively (FDH 2015b). + + g +n +i +t +i + +m +o +v + +, +a +e +h +r +r +a +i +d + +, +h +s +a +r + +, +e +h +c +a +d +a +e +h + +, +e +u +g +i +t +a +f + +, +r +e +v +e +F + +d +e +t +a +n +i +m +a +t +n +o +c + +, +g +u +b + +e +n i +m +o +t +a +i +r +t + +y +b + +e +t +i + +B + +e +s +a +e +s +i +d +s +a +g +a +h C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +1 3 6 • S O N G L I A N G E T A L . + +t +n +a +n +g +e +r +p +n +i + +e +g +a +m +a +d +l +a +t +e +f + +, +e +y +e + +, +n +i +a +r +b +n +o +e +g +a +m +a +d +, +r +e +v +e +F + +l +a +t +i +n +e +g +n +o +c + +, +) +F + +, + +D + +( + +r +e +t +a + +W + +s +i +s +o +m +s +a +l +p +o +x +o +T + +i +i +d +n +o +g +a +m +s +a +l +p +o +x +o +T + +, +e +u +g +i +t +a +f + +, +e +h +c +a +d +a +e +h +, +r +e +v +e +f + +, +s +s +e +n k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | +c +i +s + +g +n +i +p +e +e +l +s + +, +e +r +c +n +a +h +C + +a +e +s +u +a +n + +, +e +u +g +i +t +a +f + +, +s +t +a +e +w +s + +, +e +h +c +a +d +a +e +h + +, +s +l +l +i +h +c + +, +r +e +v +e +F + +l +a +t +a +f + +e +b + +n +a +c + +n +a +m +o +w + +s +o +t +i +u +q +s +o +m +y +b + +e +t +i + B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) + +a +i +r +a +l +a + M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +y +l +f + +e +s +t +e +s +t +y +b +e +t +i + +B + +s +i +s +a +i +m +o +s +o +n +a +p +y +r +t +n +a +c +i +r +f + +A + +P +P +S +a +m +o +s +o +n +a +p +y +r +T + +P +P +S +m +u +i +d +o +m +s +a +l +P + +r +o +f + +l +a +t +a +f + +d +n +a + +, +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +s +s +o +l + +t +h +g +i +e +w + +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +y +r +e +t +n +e +s +y +d +, +n +i +a +p +h +c +a +m +o +t +s + +, +a +e +h +r +r +a +i +D + +e +l +p +o +e +p +d +e +s +i +m +o +r +p +m +o +c +o +n +u +m +m + +i + +) + +D + +( + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +a +i +b +e +m +A + +a +c +i +t +y +l +o +t +s +i +h +a +b +e +o +m +a +t +n +E + +) + +, + +R +D + +( + +) + +, + +R +D + +( + +) + +D + +( + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +o +i +d +i +r +o +p +s +o +t +p +y +r +C + +m +u +v +r +a +p +m +u +i +d +i +r +o +p +s +o +t +p +y +r +C + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +r +e +t +a +w +a +i +v +l +a +r +o +- +l +a +c +e +F + +s +i +s +a +i +d +r +a +i +G + +s +i +l +a +n +e +d +o +u +d + +a +i +d +r +a +i +G + +s +i +s +a +i +r +o +p +s +o +l +c +y +C + +s +i +s +n +e +n +a +t +e +y +a +c +a +r +o +p +s +o +l +c +y +C + +e +n +r +o +b +r +e +t +a + +W + +a +o +z +o +t +o +r +P + +g +n +i +p +u +o +r +G + +s + +m +o +t +p +m +y +S +& + +s +s +e +n +l +l +I +n +i +a +M + +s +y +a +w +h +t +a +P +e +r +u +s +o +p +x +E +n +i +a +M + +e +s +a +e +s +i +D + +m + +s +i +n +a +g +r +O + +n +o +i +s +s +i + +m +s +n +a +r +T + +. +s +e +s +s +e +n +l +l +i + +d +e +t +a +l +e +r + +d +n +a + +s +t +n +e +g +a + +d +e +t +a +i +c +o +s +s +a +- +r +e +t +a +w + +, +e +v +i +t +i +s +n +e +s +- +e +t +a +m + +i +l +c + +f +o + +t +s +i +l + +A + +. +1 +. +4 + +e +l +b +a +T + +e +n +o +b +d +n +a + +e +l +c +s +u +m + +, +t +n +i +o +j + +e +r +e +v +e +s + +, +e +h +c +a +d +a +e +h +e +r +e +v +e +s + +, +r +e +v +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +c +i +g +a +h +r +r +o +m +e +h +e +u +g +n +e +D + +g +n +i +d +e +e +l +b + +, +h +s +a +r + +, +n +i +a +p + +) +F +H +D + +( + +r +e +v +e +f + +s +i +t +i +g +n +i +n +e +m + +r +o + +, +s +s +e +n +i +z +z +i +d + +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +s +i +u +o +L + +. +t +S + +s +i +t +i +l +a +h +p +e +c +n +e + +s +i +u +o +L + +. +t +S + +h +s +a +r + +, +n +i +a +p +e +l +c +s +u +m + +, +g +n +i +l +l +e +w +s +d +n +a +n +i +a +p +t +n +i +o +j + +, +r +e +v +e +F + +d +n +a + +, +s +l +l +i +h +c + +, +r +e +v +e +f + +h +g +i +h + +, +e +h +c +a +d +a +e +h +, +s +n +a +m +u +h + +n +i + +e +r +a +R + +s +e +s +a +c + +e +r +e +v +e +s + +n +i + +g +n +i +t +i + +m +o +v + +e +s +i +a +l +a +m +d +n +a + +, +a +e +s +u +a +n + +o +t +i +u +q +s +o +m +y +b +e +t +i + +B + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +e +n +i +u +q +e +n +r +e +t +s +a +E + +a +y +n +u +g +n +u +k +i +h +C + +s +i +t +i +l +a +h +p +e +c +n +e + +s +u +r +i +v +a +y +n +u +g +n +u +k +i +h +C + +s +u +r +i +v + +s +i +t +i +l +a +h +p +e +c +n +e + +e +n +i +u +q +e +n +r +e +t +s +a +E + +s +u +r +i +v + +, +a +e +s +u +a +n + +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +s +o +r +C +a +L + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +s +o +r +C +a +L + +s +s +e +n +d +e +r +i +t + +d +n +a + +, +g +n +i +t +i + +m +o +v + +s +u +r +i +v + +s +i +t +i +l +a +h +p +e +c +n +e + +, +s +s +e +n +l +l +i + +e +l +i +r +b +e +f + +, +c +i +t +a +m +o +t +p +m +y +s +a + +s +e +s +a +c + +t +s +o +M + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +e +s +a +e +s +i +d +s +u +r +i +v + +e +l +i + +N + +t +s +e + +W + +s +u +r +i +v + +e +l +i + +N + +t +s +e + +W + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +, +r +e +d +r +o +s +i +d + +t +n +e +m +e +v +o +m + +, +g +n +i +t +i + +m +o +v +, +e +h +c +a +d +a +e +h + +, +r +e +v +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +e +n +a +p +a +J + +s +i +t +i +l +a +h +p +e +c +n +e + +e +s +e +n +a +p +a +J + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +s +e +r +u +z +i +e +s + +s +u +r +i +v + +) +s +u +o +e +n +a +t +u +c +( + +s +e +r +o +s +n +i +k +s + +, +) +l +a +r +e +c +s +i +v +( + +t +n +e +m +e +g +r +a +l +n +e + +r +e +v +i +l +/ +n +e +e +l +p +s + +, +s +s +o +l + +t +h +g +i +e +w + +, +r +e +v +e +F + +y +l +f +d +n +a +s +y +b +e +t +i + +B + +s +r +e +h +t +o +d +n +a + +, +) +F +( + +a +e +h +r +r +a +i +d +d +n +a + +g +n +i +t +i + +m +o +V + +s +i +t +i +t +a +p +e +H + +) +F + +, + +R + +, + +D + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +) +F +R + +, + +, + +D + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +s +u +o +i +r +a +V + +t +c +a +t +n +o +c + +, + +, +) +F +R +D + +, + +( + +r +e +t +a +w +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +a +i +n +a +m +h +s +i +e +L + +P +P +S +a +i +n +a +m +h +s +i +e +L + +s +i +t +i +t +a +p +e +h + +l +a +r +i + +V + +s +u +r +i +v +E + +, + +A +s +i +t +i +t +a +p +e +H + +s +e +s +u +r +i +V + +s +i +t +i +r +e +t +n +e +o +r +t +s +a +g + +l +a +r +i + +V + +s +u +o +i +r +a +V + +4 + +, +3 +, +2 +, +1 +V +N +E +D + +s +e +s +u +r +i +v +o +r +e +t +n +E + +s +u +r +i +v +o +r +o +N + +e +n +r +o +b +r +e +t +a + +W + +i +z +u +r +c + +. + +T + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +d +e +t +a +l +e +r +- +r +e +t +a + +W + + 7 +3 +1 + +• + +H +T +L +A +E +H + +N +A +M +U +H + +N +O + +S +T +C +A +P +M + +I + +E +G +N +A +H +C + +E +T +A +M + +I +L +C + +s + +m +o +t +p +m +y +S +& + +s +s +e +n +l +l +I +n +i +a +M + +s +y +a +w +h +t +a +P +e +r +u +s +o +p +x +E +n +i +a +M + +e +s +a +e +s +i +D + +m + +s +i +n +a +g +r +O + +n +o +i +s +s +i + +m +s +n +a +r +T + +C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 7 + +e +l +c +s +u +m + +, +a +e +s +u +a +n +, +r +e +v +e +f + +, +g +n +i +p +m +a +r +c + +l +a +n +i +m +o +d +b +a + +, +a +e +h +r +r +a +i +D + +l +a +t +a +f + +e +b +n +a +c + +, +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +e +s +i +a +l +a +m + +, +r +e +v +e +F + +s +p +m +a +r +c +h +c +a +m +o +t +s +d +n +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +s +p +m +a +r +c +h +c +a +m +o +t +s + +d +n +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +s +e +h +c +a + +e +m +o +r +d +n +y +s + +c +i +m +e +a +r +u +- +c +i +t +y +l +o +m +e +a +h + +d +n +a + +, +a +e +h +r +r +a +i +d +y +d +o +o +l +B + +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +r +e +v +e +f + +, +a +e +h +r +r +a +i +D + +c +i +n +o +m +u +e +n +p +g +n +i +n +e +t +a +e +r +h +t +- +e +f +i +l + +o +t + +n +o +i +s +e +l + +n +i +k +s + +d +l +i + +M + +h +t +i + +w + +t +c +a +t +n +o +c + +, +y +l +f + +r +e +e +d +r +o + +k +c +i +t + +f +o + +s +e +t +i + +B + +, +r +i +a +h +g +u +o +r +h +t + +e +b +n +a +c + +, +s +l +a +m +i +n +a +d +e +t +c +e +f +n +i + +r +e +t +a +w + +) +F +D + +, + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +p +m +a +r +c + +e +l +c +s +u +m + +, +g +n +i +t +i + +m +o +v +, +a +e +h +r +r +a +i +d + +e +r +e +v +e +S + +) +F + +, + +D + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +r +e +v +e +f + +, +n +i +a +p +l +a +n +i +m +o +d +b +a + +, +a +e +h +r +r +a +i +D + +) + +D +s +s +e +l + +, + +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +a +i +m +e +c +i +t +p +e +s + +, +a +e +h +r +r +a +i +D + +d +n +u +o +w +n +e +p +o +, +h +s +i +f +l +l +e +h +S +d +e +t +a +n +i +m +a +t +n +o +C + +) +F +D + +, + +( + +) +F +D + +, + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +s +s +e +n +l +l +i + +l +a +e +h +r +r +a +i +D + +i +l +o +c + +. + +E +c +i +n +e +g +i +x +o +t +o +r +e +t +n +E + +d +e +t +a +n +i +m +a +t +n +o +C + +s +s +e +n +l +l +i + +l +a +e +h +r +r +a +i +D + +. + +E +c +i +g +a +h +r +r +o +m +e +a +h +o +r +e +t +n +E + +) + +R + +, + +F + +) +F + +) +F + +, + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +, + +D + +( + +d +e +t +a +n +i +m +a +t +n +o +C + +, + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +e +l +l +e +n +o +m +l +a +S + +s +i +s +o +l +l +e +g +i +h +S + +d +i +o +h +p +y +T + +s +i +s +o +i +r +b +i +V + +a +r +e +l +o +h +C + +e +a +r +e +l +o +h +c +o +i +r +b +i +V + +P +P +S +o +i +r +b +i +V + +. +p +p +s + +a +l +l +e +n +o +m +l +a +S + +i +h +p +y +t + +a +l +l +e +n +o +m +l +a +S + +. +p +p +s + +a +l +l +e +g +i +h +S + +a +i +r +e +t +c +a +B + +s +i +s +o +i +r +e +t +c +a +b +o +l +y +p +m +a +C + +. +p +p +s + +r +e +t +c +a +b +o +l +y +p +m +a +C + +i + +g +n +p +u +o +r +G + +e +n +r +o +b +r +e +t +a + +W + +s +i +s +o +i +n +i +s +r +e +Y + +a +i +m +e +a +r +a +l +u +T + +s +i +s +n +e +r +a +l +u +t + +a +l +l +e +s +i +c +n +a +r +F + +. +p +p +s +a +i +n +i +s +r +e +Y + +i +l +o +c + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +n +i +k +s + +e +h +t + +r +e +d +n +u +s +e +l +u +d +o +n +d +n +a + +, +e +s +a +e +s +i +d +e +y +e + +, +h +s +a +r +n +i +k +S + +m +r +o +w +e +y +e + +, +s +e +u +g +i +t +a +f + +, +n +i +a +p +t +n +i +o +j + +d +n +a + +e +l +c +s +u +m + +, +y +h +c +t +I + +y +r +a +n +o +m +l +u +p +, +a +m +e +d +e +h +p +m +y +l + +, +c +i +t +a +m +o +t +p +m +y +s +a + +t +s +o +M + +n +o +i +t +a +d +r +a +t +e +r +h +t +w +o +r +g +, +l +o +o +t +s +y +r +e +t +a +w +d +n +a +y +d +o +o +l +B + +a +i +l +i +h +p +o +n +i +s +o +e + +n +e +r +d +l +i +h +c + +n +i + +n +o +i +t +a +d +r +a +t +e +r +h +t +w +o +r +g +, +t +r +o +f +m +o +c +s +i +d +l +a +n +i +m +o +d +b +A + +) +F + +, + +D + +( + +) +F +D + +, + +( + +l +a +r +o +- +l +a +c +e +F + +l +a +r +o +- +l +a +c +e +F + +o +t +i +u +q +s +o +m +y +b + +e +t +i + +B + +y +l +f +k +c +a +l +b +y +b +e +t +i + +B + +y +l +f +r +e +e +d +y +b +e +t +i + +B + +s +i +s +a +i +o +L + +s +u +o +i +r +a +V + +a +i +r +e +t +c +a +b +o +n +a +y +C + +a +o +l + +a +o +L + +e +a +g +l +A + +s +i +r +a +c +s +A + +s +e +d +i +o +c +i +r +b +m +u +l + +s +i +r +a +c +s +A + +s +u +n +a +c +i +r +e +m +a + +r +o +t +a +c +e +N +d +n +a + +n +o +i +t +c +e +f +n +i + +m +r +o +w +p +i +h +W + +s +i +s +a +i +r +a +l +i +f + +c +i +t +a +h +p +m +y +L + +a +r +u +i +h +c +i +r +t + +s +i +r +u +h +c +i +r +T + +i +t +f +o +r +c +n +a +b + +a +i +r +e +r +e +h +c +u +W + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +s +i +s +a +i +c +r +e +c +o +h +c +n +O + +s +u +l +u +v +l +o +v + +a +c +r +e +c +o +h +c +n +O + +) +e +n +r +o +b +- +r +o +t +c +e +V + +( + +c +i +t +o +r +u +e +n + +, +e +g +a +m +a +d + +r +e +v +i +l + +, +s +s +e +n +l +l +i + +l +a +n +i +t +s +e +t +n +i + +, +s +i +t +i +t +a +m +r +e +D + +t +c +e +r +i +d +d +n +a +D +a +i +v +s +n +i +x +o +t +o +t + +e +r +u +s +o +p +x +E + +s +n +o +i +t +c +a +e +r + +t +c +a +t +n +o +c + +y +r +o +t +a +r +i +p +s +e +r + +, +a +i +s +e +h +t +s +e +r +a +p + +, +s +m +o +t +p +m +y +s + +l +a +n +i +t +s +e +t +n +i +o +r +t +s +a +G + +n +i +x +o +t +y +b +) +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +h +s +i +f +l +l +e +h +s + +c +i +x +o +t +o +r +u +e +N + +s +i +v +e +r +b + +a +i +n +e +r +a +K + +n +o +i +t +a +t +i +r +r +i + +e +y +e + +d +n +a + +g +n +i +n +o +s +i +o +p + +a +i +s +e +h +t +s +e +r +a +p + +y +t +i + +m +e +r +t +x +e + +, +e +s +i +a +l +a +m + +, +g +n +i +t +i + +m +o +v +, +a +e +h +r +r +a +i +D + +n +i +x +o +t +y +b +) +F +( +d +e +t +a +n +i +m +a +t +n +o +C + +g +n +i +n +o +s +i +o +p +h +s +i +f + +a +r +e +t +a +u +g +i +C + +p +p +s + +s +u +c +s +i +d +r +e +i +b +m +a +G + +s +i +s +y +l +a +r +a +p + +g +n +i +n +o +s +i +o +p + +y +r +o +t +a +r +i +p +s +e +r + +, +a +i +s +e +h +t +s +e +r +a +p + +y +t +i + +m +e +r +t +x +e +d +n +a + +l +a +r +o +m +u +c +r +i + +C + +n +i +x +o +t + +y +b +) +F +( + +d +e +t +a +n +i +m +a +t +n +o +C + +h +s +i +f +l +l +e +h +s + +c +i +t +y +l +a +r +a +P + +p +p +s +m +u +i +r +d +n +a +x +e +l +A + +d +e +t +a +l +e +r +- +r +e +t +a + +W + +. +a +d +i +r +o +l +F +n +i + +e +c +n +a +c +i +f +i +n +g +i +s +h +t +l +a +e +h +c +i +l +b +u +p +r +o +j +a +m + +f +o +s +e +s +a +e +s +i +d +/ +s +n +e +g +o +h +t +a +p + +s +e +t +a +c +i +d +n +i +d +e +r +n +i + +d +e +t +h +g +i +l +h +g +i +H + +. +2 +; +) +F +( + +s +d +o +o +f + +d +n +a +, +) +R +( + +r +e +t +a +w + +l +a +n +o +i +t +a +e +r +c +e +r + +, +) + +D + +( + +r +e +t +a +w +g +n +i +k +n +i +r +d +: +s +y +a +w +h +t +a +p +e +r +u +s +o +p +x +e + +n +i +a +m +r +o +f + +s +n +o +i +t +a +i +v +e +r +b +b +A + +. + +1 +: +e +t +o +N + +e +g +a +m +a +d + +l +a +n +i +t +s +e +t +n +i + +d +n +a +y +r +a +n +i +r +u +, +a +e +h +r +r +a +i +D + +r +e +t +a +w +h +t +i + +w + +t +c +a +t +n +o +c + +l +a +m +r +e +D + +s +i +s +a +i +m +o +s +o +t +s +i +h +c +S + +. +p +p +s + +a +m +o +s +o +t +s +i +h +c +S + +s +h +t +n +i +m +l +e +H + +a +i +m +e +n +a + +d +n +a + +, +e +u +g +i +t +a +f + +, +e +t +i +t +e +p +p +a + +f +o +s +s +o +l + +, +a +e +h +r +r +a +i +D + +l +i +o +s +h +t +i + +w + +t +c +a +t +n +o +c + +l +a +m +r +e +D + +n +o +i +t +c +e +f +n +i + +m +r +o +w +k +o +o +H + +e +l +a +n +e +d +o +u +d +a +m +o +t +s +o +l +y +c +n +A + +n +o +i +t +a +m +m +a +l +f +n +i +d +n +a + +r +e +c +l +u +l +u +f +n +i +a +P + +) + +D + +( +d +e +t +a +n +i +m +a +t +n +o +C + +s +i +s +a +i +l +u +c +n +u +c +a +r +D + +s +i +s +n +e +n +i +d +e +m +s +u +l +u +c +n +u +c +a +r +D + +d +e +s +a +b +- +r +e +t +a + +W + + 1 3 8 • S O N G L I A N G E T A L . + +Figure 4.3. Reported outbreaks (bar) and human cases (line) of food-borne and waterborne diseases in +Florida each year from 1989 to 2011 (Source: Florida Department of Health). Toxigenic V. cholerae +infections are highly unusual in Florida and the first V. cholerae O75 outbreak was detected and reported +in Florida between March 23 and April 13, 2011. + +Figure 4.4. Monthly distribution of reported outbreaks and cases of food-borne and waterborne diseases in +Florida, 2011 (Source: Florida Department of Health) + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 9 + +c) Vector-Borne Diseases +Vector-borne diseases pose serious public health threats throughout the world. According to the +World Health Organization, vector-borne diseases account for more than 17% of infectious +diseases, causing more than one million deaths annually from diseases such as malaria, dengue +fever, yellow fever, Japanese encephalitis, and schistosomiasis (WHO 2016). Vector-borne +diseases are transmitted through vector organisms including mosquitoes, ticks, flies, fleas, and +snails. The transmission of vector-borne diseases takes two primary transmission pathways. The +first is the human-vector pathway, with transmission through a vector biting a host; pathogens +survive outside hosts in arthropod vectors and humans are typically the only host (Eisenberg et +al. 2007). Diseases in this group include malaria, dengue fever, onchocerciasis, trypanosomiasis, +and filariasis (Table 4.1). The second pathway involves zoonotic transmission, with a vector +biting a nonhuman host as the transmission route and humans as the dead-end host (Eisenberg et +al. 2007). Diseases in this group include Lyme disease, Yellow fever, West Nile virus, and +Japanese encephalitis (Table 4.1). The transmission and spread of vector-borne diseases are +determined by complex interactions between the host (either human or nonhuman), the vector +(e.g., mosquitos, flies, snails, and ticks), and various pathogens ranging from protozoans, viruses, +bacteria, and helminthes (Table 4.1). Important biological properties underlying the transmission +of vector-borne diseases include survival, development, and reproduction of the vector; also the +pathogens in the vector, the vector’s biting rate, and the behavior of hosts (human and +nonhuman), all of which are associated with climatic conditions. The following sections discuss +some specific aspects of the effects of rainfall, temperature, and flooding with regard to vector- +borne diseases. + +Temperature Sensitivity +Through the modulating effects on physiological activities and tolerance limits of vectors and +infectivity of pathogens, temperature can affect distribution of the vector and transmission of the +pathogens between vectors and hosts. Some key mechanistic effects of temperature on vectors, +vector-borne pathogens and their transmission are summarized by Gulber et al. (2001). +• Change in survival of vectors (e.g. increase or decrease depending on species) +• Change in susceptibility of vectors to pathogens +• Change in vector population growth +• Change in feeding behavior and host contact +• Change in incubation period of pathogens in vectors (e.g. decreased extrinsic incubation + +period at higher temperatures) + +• Change in pathogen replication and infectivity +• Change in transmission season + +A large number of studies on vector-borne diseases at different scales have provided +empirical evidence and predictions on the climatic factors and vector-borne disease links. For + + 1 4 0 • S O N G L I A N G E T A L . + +example, studies have shown significant relationships between ambient temperature and malaria +transmission for both mean temperature (Loevinsohn 1994; Lindblade et al. 2000; Thomson et +al. 2005; Tian et al. 2008; Wardrop et al. 2013) and its daily fluctuations (Paaijmans et al. 2010). +Laboratory studies have shown clear temperature-dependent effects on dengue virus infection in +mosquitoes (Alto and Bettinardi 2013). Fluctuations at a low mean temperature accelerate +dengue virus transmission in Aedes aegypti mosquitoes (Carrington et al. 2013). A recent +systematic review and meta-analysis based on 137 studies on the relationship between dengue +risk and temperature suggests that the dengue transmission and risk are sensitive to temperature, +with a positive relationship between them for a minimum range (18.1–24.2 °C) to a maximum +temperature range (28.0–34.5 °C) (Fan et al. 2014). A regional predictive study reported that +increased temperature increases the likelihood of many vector-borne diseases, in particular +malaria, dengue fever, and Lyme disease in new areas (Githeko et al. 2000). Similar findings are +also reported for West Nile virus (Soverow et al. 2009). These empirical findings have been used +to parameterize mechanistic models of temperature driven transmission for vector-borne +diseases, which can be coupled to projected climate scenarios to anticipate shifting geographies +of transmission risk, e.g. for malaria (Mordecai et al. 2013; Johnson et al. 2014; Johnson et al. +2015; Ryan et al. 2015; Ryan et al. 2015) and for Aedes spp transmitted diseases of concern for +Florida: dengue, chikugunya, and zika viruses (Moedecai et al. 2017; Ryan et al. 2017). + +Rainfall +Rainfall can supply water to both transient and persistent environments that can serve as habitats +or breeding sites for vectors and/or for part of pathogens’ lifecycles, critical for transmission of +vector-borne diseases. Meanwhile, rainfall can also have negative impacts on the habitats (for +example, washing away), thereby reducing the transmission risk. Variations in rainfall may have +direct and indirect impacts on the distribution and timing of transmission of vector-borne +diseases. Some key mechanistic effects of rainfall on vectors, vector-borne pathogens, and their +transmission are summarized by Gubler et al. (2001). +• +• Excess rain (e.g. causing flooding) may eliminate habitat for both vector and hosts. +• +• +• + +Increased humidity may increase vector survival and biting rate. +Increased rainfall may facilitate transport of pathogens for certain distance (Liang et al. 2007). +Increased rain may increase vertebrate host size through increased food availability (e.g. +vegetation). + +Increased rain may increase larval habitat and vector population. + +• Flooding may decrease vertebrate host size (e.g. reduced food availability) but increase + +interactions (e.g. contacts) with humans. + +Such mechanistic links have been reported by many studies including for diseases such as +malaria (van der Hoek et al. 1997; Thomson et al. 2005; Galardo et al. 2009; Gao et al. 2012; +Yang et al. 2012), dengue fever (Li et al. 1985; Su 2008; Johansson et al. 2009; Banu et al. 2011; + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 1 + +Hii et al. 2012), West Nile virus (Shaman et al. 2005; Soverow et al. 2009), and Lyme disease +(Ogden et al. 2006; Ostfeld et al. 2006). + +Vector-Borne Diseases in Florida +In the United States, 14 vector-borne diseases are of primary public health concern. Table 4.2 +lists key notifiable vector-borne diseases and reported cases between 2013 and 2015 in Florida. +Major mosquito-borne diseases include the St. Louis encephalitis virus, West Nile virus, Eastern +equine encephalitis virus, Western equine encephalitis virus, Venezuelan equine encephalitis +virus, Everglades virus, and the California serogroup viruses including La Crosse encephalitis +virus, all of which are transmitted by mosquitoes. More recently, concern about Aedes spp +transmitted flavivirus and alphaviral disesaes have become more prevalent: dengue (DENV), +chikungunya (CHIKV), and zika (ZIKV). The 2015-2016 global public health emergency for +Zika declared by the WHO focused attention on Florida as an introduction and establishment +gateway for the U.S. + +Table 4.2. A summary of reported case counts of notifiable vector-borne diseases in Florida. (Source: +Florida Department of Health). + +2013-2015 Reported Cases Number of Counties Reported + +Diseases +Tick-borne +Lyme disease +Spotted Fever Rickettsia +Anaplasmosis/Ehrlichiosis +Tularemia +Mosquito-Borne +West Nile virus +Malaria +Dengue +California serogroup viruses 2 +3 +Eastern equine encephalitis +2 +St. Louis encephalitis + +459 +74 +82 +2 + +37 +146 +331 + +46 +18 +30 +2 + +14 +24 +31 +2 +3 +1 + +St. Louis Encephalitis. The St. Louis encephalitis virus, a flavivirus, was the most common +mosquito-transmitted human pathogen in the U.S. prior to the introduction of the West Nile virus +in 1999 (FDH 2016). In Florida, the principal vector is Culex nigripalpus, a ubiquitous species +throughout the state. No human case has been reported in Florida since 2003. However, sentinel +chickens testing positive for antibodies to the St. Louis encephalitis virus have been reported in +several counties in Central and South Florida, suggesting the potential for a possible resurgence +(FDH 2014, 2016). + +West Nile virus. The peak period of transmission in Florida for the West Nile virus is July +through September. The natural cycle of West Nile virus involves Culex mosquitoes and wild +birds. Between 2001 and 2013, 318 human West Nile virus cases were reported in Florida and + + 1 4 2 • S O N G L I A N G E T A L . + +West Nile virus activities have been consistently reported in many counties around the state +(FDH 2014). + +Eastern Equine Encephalitis. Eastern equine encephalitis occurs in natural cycles involving +birds and Culiseta melanura in freshwater swampy areas, with a peak in transmission occurring +between May and August. Historical and current evidence indicates limited human Eastern +equine encephalitis epidemic potential in Florida (FDH 2014, 2016). + +Dengue Fever. In Florida, a historical dengue fever epidemic occurred in 1934-1935 and then +it ceased. It re-emerged in 2009, and since then a small number of cases have been reported each +year among individuals who had previous travel to dengue-endemic countries. In the summer of +2009, local dengue transmission was identified in Key West, Florida. Since then, sporadic local +transmission has been identified in other Central and South Florida counties during 2010, 2011, +2012, and 2013 (FDH 2014, 2016). The presence of the Aedes spp mosquito, both Aedes aegypti, +also known as the Yellow Fever mosquito, which has been a persistent invasive mosquito in +Florida for a couple of centuries, and the more recently introduced Asian Tiger Mosquito, Aedes +albopictus, maintain transmission potential in Florida. Aedes mosquitoes are capable of breeding +in as little as a teaspoon of water, and are able to adapt well to urban environments, and exhibit +transmission of dengue, chikungunya, and zika, at temperature ranges found almost year-round +in most of Florida (Mordecai et al. 2017). The optimal transmission temperature range for Aedes +aegypti, in particular, is higher than that for Aedes albopictus, suggesting that future climates +will promote Ae aegypti transmission in Florida, before it becomes too hot for transmission (Ryan +et al. 2017). + +Malaria. Malaria is caused by the mosquito-borne parasite Plasmodium falciparum. Endemic +malaria was eliminated in Florida in the late 1940s. However, imported cases—either from +travelers returning to the state from malaria-endemic regions or tourists carrying the pathogen— +have been reported. The Anopheles mosquitoes responsible for transmitting the malaria parasite +to humans are common in the state so establishment of local transmission is still possible (FDH +2014, 2016). + +Tick-Borne Diseases. The most common tick-borne diseases are ehrlichiosis, anaplasmosis, +Lyme disease, Rocky Mountain spotted fever, and other spotted fever illnesses. Lyme disease is +the most commonly reported vector-borne disease in the United States. It is caused by a bacteria +known as Borrelia burgdorferi. Lyme disease is reported in Florida year-round. An estimate +suggests that about 23% of cases were acquired in Florida and about 77% were acquired while +travelling to other states or countries (FDH 2016). Ehrlichiosis and Anaplasmosis—several +pathogenic species in the genus Ehrlichia and Anaplasma—can cause human illness. The human +illness caused by Ehrlichia chaffeensis is called Human Monocytic Ehrlichiosis, and the illness +caused by Anaplasma phagocytophilum is called Anaplasmosis or Human Granulocytotropic +Anaplasmosis. In Florida, the majority of Human Monocytic Ehrlichiosis cases (73%) are +acquired in Florida, primarily in the northern and central parts of the state, and Human +Granulocytotropic Anaplasmosis cases are even more likely to be acquired in Florida. Cases are + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 3 + +reported year-round, with a peak occurring during the spring and summer months (FDH 2016). +Rocky Mountain Spotted Fever is a disease is caused by the bacterium Rickettsia richerrsii and, +in Florida, it is transmitted primarily by the American dog tick. It can also be caused by +Dermacentor variabilis and cases of this transmission are reported year-round with more than +70% acquired in Florida, and the majority of cases are reported in the northern and central regions +of the state (FDH 2016). + +Drought +Drought is considered a meteorological anomaly characterized by a prolonged and abnormal +moisture deficiency. Climate change and variability are progressively increasing the severity and +frequency of drought events. Drought has many possible public health implications (Kalis et al. +2009; Kalis and Curtiss 2016). Drought may: +• Compromise quantity and quality of water for drinking, sanitation, and hygiene +• Compromise air quality (e.g. increased particulate matters in the air) +• Reduce crop yield, food availability and nutrition +• + +Increase or decrease risks to infectious pathogens (e.g. waterborne, air-borne, and vector- +borne) +Increase risk of non-communicable illness (e.g. respiratory infections and/or illness) + +• + +Drought has been linked to deterioration of water quality (Golladay and Battle 2002; Al- +Kharabsheh and Ta'any 2003) and air quality (Taylor and Davies 1990; Field et al. 2009) in +various settings in both the developed and developing world. Drought has been implicated for +infectious diseases—waterborne (Lipp et al. 2002; Schuster et al. 2005; Senhorst and Zwolsman +2005), vector-borne (Chretien et al. 2007; Erlanger et al. 2009; Medlock and Leach 2015), and +airborne (Polymenakou et al. 2008). Drought conditions have also been linked to increased +incidence of respiratory diseases (Smith et al. 2014). + +In Florida, the BRACE study examined associations between monthly drought conditions +(using the Standardized Precipitation Index), disease rates and emergency department visits for +specific health conditions (e.g. allergic rhinitis, asthma, and all respiratory diseases excluding +asthma) among Florida residents from 2005 to 2012. The results suggested that drought was +significantly associated with emergency department visits for respiratory diseases. It is worth +noting that the relationship between the most extreme drought conditions and emergency +department visits tended to be protective (e.g. less emergency department visits for respiratory +illness), while more moderate drought appeared to be associated with an increase in emergency +department visits for the health conditions. A similar pattern was also observed for asthma (FDH +2014b). + + 1 4 4 • S O N G L I A N G E T A L . + +Indirect Impact Pathway – Mental Health, Well-Being, +and Community Health + +Mental health disorders and well-being are an important aspect of the human health impacts +associated with climate change and variability, and associated weather events. Many mental +health consequences, ranging from simple stress and distress symptoms to some severe +conditions including anxiety, distress, posttraumatic stress disorder (PTSD), and even suicidal +tendency have been the focus of much of the research and public health responses in the past few +decades. Here are some key mental disorders and community health issues that may be related +to climate change and associated weather events, based on the U.S. Global Change Research +Program report (Crimmins et al. 2016). + +Experience with extreme weather events, such as flooding, drought, and hurricanes, may + +cause the following mental disorders: +• +• +• +• + +Increased stress, anxiety, depression, grief, and even suicidality; +Increased tensions on social stress and relationships; +Increased substance abuse; and +Increased risk of PTSD. +At the community level (Crimmins et al. 2016), potential impacts may include: +• +Increased interpersonal aggression, +• +Increased violence and crime, +• +Increased social instability, and +• Decreased community cohesion. + +Assessing the impacts of natural disasters on mental or community health and well-being of +affected populations, as well as mediation techniques, has been a focus of much research, +community and public health responses. In Florida, studies have suggested that anxiety disorder, +major depressive episodes, and PTSD among affected populations could be significantly +associated with storm exposure and displacement during the occurrence of Hurricane Andrew in +1992 (David et al. 1996), as well as a number of hurricanes in 2004 (Acierno et al. 2007) and +2012 (Neria and Shultz 2012). Studies have also indicated that social support would help to +alleviate such negative impacts (Acierno et al. 2007). Similar impacts have been reported for +community and public health workers who responded to these weather events. For example, +during the 2004 Florida hurricane season, four hurricanes (Charley, Frances, Ivan, and Jean) as +well as one tropical storm (Bonnie) hit Florida during a period of seven weeks. Increased PTSD, +other mental health issues, and substance abuse were reported among public health workers +during and after these events (Fullerton et al. 2013; Fullerton et al. 2015). The studies also +suggested that these mental health outcomes were influenced by multiple community +characteristics including the collective efficacy of neighborhood populations in the community. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 5 + +Significantly lower depressive symptoms were associated with communities that had sufficient +resources and received social support (Acierno et al. 2007, Fullerton et al. 2013, Fullerton et al. +2015). + +Impacts on children following natural disasters are of particular concern. For example, +Hurricane Andrew was a devastating category 5 Atlantic hurricane that struck South Florida in +1992. Studies showed significant impacts of hurricane exposure, stressors occurring during the +hurricane, and recovery periods on children’s persistent posttraumatic stress. In the Miami-Dade +County area, 35% to 60% of children surveyed reported moderate to very severe levels of +posttraumatic stress symptoms (Vernberg et al. 1996; La Greca et al. 2010), with hurricane- +related stressors influencing children’s persistent posttraumatic stress symptoms and other life +events in later stages of children’s life (La Greca et al. 2010; Weems et al. 2010; La Greca et al. +2013). These findings have important community health implications for identifying and +potentially helping youth in the aftermath of natural disasters. + +Conclusion + +The current available evidence has clearly indicated increases in average temperature, total +annual precipitation, frequency of extreme temperature conditions and heavy precipitation in the +United States in the past five decades, as well as increases in the tropical cyclone activity in +regions along the Atlantic Ocean, the Caribbean, and the Gulf of Mexico. Similar patterns are +expected in the future if no remediation is conducted (USEPA 2016). Florida is likely to face +even greater effects than the rest of the US, particularly effects associated with sea level rise, heat +waves, storm surge, hurricanes, and others (FOCC 2010) that affect human health in a variety of +different ways. Some effects are through direct exposure pathways, such as hurricane/storm and +temperature-related illnesses, injuries, and deaths. Some are through indirect exposure pathways +such as waterborne and vector-borne diseases. And still others are through social and economic +pathways, such as stress and mental illness. In addition to these direct and indirect effects, we +see coupled and compounded effects of climate change, which are anticipated to be exacerbated +in the future. For example, Florida is home to a large elderly population; the warm climate and +housing development capacity made this ideal. However, the elderly are more vulnerable to the +effects of heatwaves, less mobile in the event of hurricanes, and suffer greater impacts of the +symptomology of febrile vectorborne diseases. This puts the state in a position higher public +health urgency by virtue of an interaction of demographic profile and increased impact of climate +change due to geography. Given the increase in hurricances and storm impacts expected under +climate change, we are likely to also see the added impact of vectorborne disease risk anticipated +in the aftermath of such disasters. For example, the additional potential vector breeding sites +provided by debris and damaged infrastructure after hurricanes have been noted as concerns for +Zika transmission in the 2016-2017 hurricane season. An additional component of this is the + + 1 4 6 • S O N G L I A N G E T A L . + +unanticipated exposures post-disaster, in which both displaced residents and rescue workers have +radically increased outdoor exposure, reduced access to air conditioned spaces, and to basic +protection measures such as insect repellent and appropriate protective clothing. These are just +two of the compounded scenarios of public health concern that the state of Florida faces in a +changing climate. Despite the comprehensive information reviewed in the chapter, many things +are still unknown or are not well understood about the health impacts of Florida’s changing +climate. For instance, what are the impacts of sea level rise, flooding, and other extreme weather +events on water contamination that will ultimately influence human health? How will future +warming trends and other climate-sensitive hazards influence the transmission and spread of +existing and/or new waterborne and vector-borne diseases? What are the disease burdens and +their distributions attributable to known climate-sensitive hazards in the state? To address these +challenges, policies focused on mitigation, adaptation strategies, health surveillance, and +research that could close knowledge gaps on human exposures to these climate-sensitive hazards +and associated health impacts are much needed. + +Acknowledgments + +The authors thank Melissa Jordan for her instrumental discussion on this chapter. Song Liang was supported +in part by grants from NSF/EEID (EF-1015908) and NSF/WSC-Category 3 (1360330). Sadie J. Ryan was +supported in part by grants from NSF/EEID (DEB-1518681), NSF RAPID (DEB-1641145), and an Early +Career Fellowship from the Florida Climate Institute. + +References + +Acierno, R., K. J. Ruggiero, S. Galea, H. S. Resnick, K. Koenen, J. Roitzsch, M. de Arellano, J. Boyle, and +D. G. Kilpatrick. 2007. Psychological sequelae resulting from the 2004 Florida hurricanes: implications +for postdisaster intervention. Am J Public Health 97 Suppl 1:S103-108. + +Al-Kharabsheh, A., and R. Ta'any. 2003. Influence of urbanization on water quality deterioration during + +drought periods at South Jordan. Journal of Arid Environments 53:619-630. + +Alto, B. W., and D. Bettinardi. 2013. Temperature and dengue virus infection in mosquitoes: independent + +effects on the immature and adult stages. Am J Trop Med Hyg 88:497-505. + +Aslar, A. K., A. Soran, Y. Yildiz, and Y. Isik. 2001. Epidemiology, morbidity, mortality and treatment of +lightning injuries in a Turkish burns units. International Journal of Clinical Practice 55:502-504. +Balbus, J., A. Crimmins, J. L. Gamble, D. R. Easterling, K. E. Kunkel, S. Saha, and M. C. Sarofim. 2016. +Ch. 1: Introduction: Climate Change and Human Health. Pages 25–42 The Impacts of Climate Change +on Human Health in the United States: A Scientific Assessment. U.S. Global Change Research +Program, Washington, DC. + +Bandyopadhyay, S., S. Kanji, and L. M. Wang. 2012. The impact of rainfall and temperature variation on + +diarrheal prevalence in Sub-Saharan Africa. Applied Geography 33:63-72. + +Banu, S., W. Hu, C. Hurst, and S. Tong. 2011. Dengue transmission in the Asia-Pacific region: impact of + +climate change and socio-environmental factors. Trop Med Int Health 16:598-607. + +Basu, R., and J. M. Samet. 2002. Relation between elevated ambient temperature and mortality: A review + +of the epidemiologic evidence. Epidemiologic Reviews 24:190-202. + +Berko, J., D. D. Ingram, S. Saha, and J. D. Parker. 2014. Deaths attributed to heat, cold, and other weather + +events in the United States, 2006-2010. Natl Health Stat Report:1-15. + +Booij, M. J. 2005. Impact of climate change on river flooding assessed with different spatial model + +resolutions. Journal of Hydrology 303:176-198. + +Bouma, M. J., R. S. Kovats, S. A. Goubet, J. S. Cox, and A. Haines. 1997. Global assessment of El Nino's + +disaster burden. Lancet 350:1435-1438. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 7 + +Carlton, E. J., J. N. S. Eisenberg, J. Goldstick, W. Cevallos, J. Trostle, and K. Levy. 2014. Heavy Rainfall +Events and Diarrhea Incidence: The Role of Social and Environmental Factors. American Journal of +Epidemiology 179:344-352. + +Carrington, L. B., M. V. Armijos, L. Lambrechts, and T. W. Scott. 2013. Fluctuations at a Low Mean +Temperature Accelerate Dengue Virus Transmission by Aedes aegypti. Plos Neglected Tropical +Diseases 7. + +Center for Disease Control, and Prevention. 1994. Heat-related deaths--Philadelphia and United States, + +1993-1994. MMWR Morb Mortal Wkly Rep 43:453-455. + +Chen, M. J., C. Y. Lin, Y. T. Wu, P. C. Wu, S. C. Lung, and H. J. Su. 2012. Effects of extreme precipitation + +to the distribution of infectious diseases in Taiwan, 1994-2008. PLoS One 7:e34651. + +Chretien, J. P., A. Anyamba, S. A. Bedno, R. F. Breiman, R. Sang, K. Sergon, A. M. Powers, C. O. Onyango, +J. Small, C. J. Tucker, and K. J. Linthicum. 2007. Drought-associated chikungunya emergence along +coastal East Africa. Am J Trop Med Hyg 76:405-407. + +Christensen, J. H., and O. B. Christensen. 2003. Climate modelling: Severe summertime flooding in Europe. + +Nature 421:805-806. + +Chung, J. Y., Y. Honda, Y. C. Hong, X. C. Pan, Y. L. Guo, and H. Kim. 2009. Ambient temperature and +mortality: an international study in four capital cities of East Asia. Sci Total Environ 408:390-396. +Collins, T. W., A. M. Jimenez, and S. E. Grineski. 2013. Hispanic health disparities after a flood disaster: +results of a population-based survey of individuals experiencing home site damage in El Paso (Texas, +USA). J Immigr Minor Health 15:415-426. + +Colwell, R. R., P. R. Epstein, D. Gubler, N. Maynard, A. J. McMichael, J. A. Patz, R. B. Sack, and R. + +Shope. 1998. Climate change and human health. Science 279:968-969. + +Crimmins, A., J. Balbus, J. L. Gamble, C. B. Beard, J. E. Bell, D. Dodgen, R. J. Eisen, N. Fann, M. D. +Hawkins, S. C. Herring, L. Jantarasami, D. M. Mills, S. Saha, M. C. Sarofim, J. Trtanj, and L. Ziska. +2016. Executive Summary. Pages 1–24 The Impacts of Climate Change on Human Health in the United +States: A Scientific Assessment. U.S. Global Change Research Program, Washington, DC. + +Curriero, F. C., K. S. Heiner, J. M. Samet, S. L. Zeger, L. Strug, and J. A. Patz. 2002. Temperature and + +mortality in 11 cities of the eastern United States. Am J Epidemiol 155:80-87. + +Curriero, F. C., J. A. Patz, J. B. Rose, and S. Lele. 2001. The association between extreme precipitation and +waterborne disease outbreaks in the United States, 1948-1994. American Journal of Public Health +91:1194-1199. + +David, D., T. A. Mellman, L. M. Mendoza, R. Kulick-Bell, G. Ironson, and N. Schneiderman. 1996. + +Psychiatric morbidity following Hurricane Andrew. J Trauma Stress 9:607-612. + +Diez, J. J., M. D. Esteban, R. Paz, J. S. Lopez-Gutierrez, V. Negro, and J. V. Monnot. 2011. Urban Coastal + +Flooding and Climate Change. Journal of Coastal Research:205-209. + +Douglas, I., K. Alam, M. Maghenda, Y. McDonnell, L. McLean, and J. Campbell. 2008. Unjust waters: +climate change, flooding and the urban poor in Africa. Environment and Urbanization 20:187-205. +Duclos, P. J., L. M. Sanderson, and K. C. Klontz. 1990. Lightning-related mortality and morbidity in + +Florida. Public Health Rep 105:276-282. + +Eisenberg, J. N., M. A. Desai, K. Levy, S. J. Bates, S. Liang, K. Naumoff, and J. C. Scott. 2007. +Environmental determinants of infectious disease: a framework for tracking causal links and guiding +public health research. Environ Health Perspect 115:1216-1223. + +Ellis, K. N., L. M. Sylvester, and J. C. Trepanier. 2015. Spatiotemporal patterns of extreme hurricanes + +impacting US coastal cities. Natural Hazards 75:2733-2749. + +Elsner, James B., James P. Kossin, and Thomas H. Jagger. "The increasing intensity of the strongest tropical + +cyclones." Nature 455.7209 (2008): 92. + +Erlanger, T. E., S. Weiss, J. Keiser, J. Utzinger, and K. Wiedenmayer. 2009. Past, Present, and Future of + +Japanese Encephalitis. Emerging Infectious Diseases 15:1-7. + +Fan, J., W. Wei, Z. Bai, C. Fan, S. Li, Q. Liu, and K. Yang. 2014. A systematic review and meta-analysis + +of dengue risk with temperature change. Int J Environ Res Public Health 12:1-15. + +FDH. 2014a. Surveillance and control of selected mosquito-borne diseases in Florida. Florida Department + +of Health, Tallahassee, Florida. + +FDH. 2014b. Health effects of precipitation abundance and deficits in Florida. Tallahassee, Florida. +FDH. 2015a. Health effects of summer heat in Florida. Florida Department of Health, Tallahassee, Florida. +FDH. 2015b. Health effects of tropical storms and hurricanes in Florida. Florida Department of Health, + +Tallahassee, Florida. + + 1 4 8 • S O N G L I A N G E T A L . + +FDH. 2016. Diseases and Conditions, Florida Department of Health, Tallahassee, Florida. +Field, C. B., V. R. Barros, and Intergovernmental Panel on Climate Change. Working Group II. 2014. +Climate change 2014: impacts, adaptation, and vulnerability: Working Group II contribution to the fifth +assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, +New York, NY. + +Field, R. D., G. R. van der Werf, and S. S. P. Shen. 2009. Human amplification of drought-induced biomass + +burning in Indonesia since 1960. Nature Geoscience 2:185-188. + +FOCC. 2010. Climate change and sea-level rise in Florida. An update of the effects of climate change on +Florida's ocean & coastal resources. Page 36 in T. F. O. a. C. Council, editor, Tallahassee, Florida +Frumkin, H., A. J. McMichael, and J. J. Hess. 2008. Climate change and the health of the public. Am J Prev + +Med 35:401-402. + +Fullerton, C. S., J. B. A. McKibben, D. B. Reissman, T. Scharf, K. M. Kowalski-Trakofler, J. M. Shultz, +and R. J. Ursano. 2013. Posttraumatic Stress Disorder, Depression, and Alcohol and Tobacco Use in +Public Health Workers After the 2004 Florida Hurricanes. Disaster Medicine and Public Health +Preparedness 7:89-95. + +Fullerton, C. S., R. J. Ursano, X. Liu, J. B. McKibben, L. Wang, and D. B. Reissman. 2015. Depressive +Symptom Severity and Community Collective Efficacy following the 2004 Florida Hurricanes. PLoS +One 10:e0130863. + +Galardo, A. K. R., R. H. Zimmerman, L. P. Lounibos, L. J. Young, C. D. Galardo, M. Arruda, and A. A. R. +D. Couto. 2009. Seasonal abundance of anopheline mosquitoes and their association with rainfall and +malaria along the Matapi River, Amapi, Brazil. Medical and Veterinary Entomology 23:335-349. +Gao, H. W., L. P. Wang, S. Liang, Y. X. Liu, S. L. Tong, J. J. Wang, Y. P. Li, X. F. Wang, H. Yang, J. Q. +Ma, L. Q. Fang, and W. C. Cao. 2012. Change in rainfall drives malaria re-emergence in Anhui +Province, China. PLoS One 7:e43686. + +Githeko, A. K., S. W. Lindsay, U. E. Confalonieri, and J. A. Patz. 2000. Climate change and vector-borne + +diseases: a regional analysis. Bull World Health Organ 78:1136-1147. + +Golladay, S. W., and J. Battle. 2002. Effects of flooding and drought on water quality in gulf coastal plain + +streams in Georgia. Journal of Environmental Quality 31:1266-1272. + +Gubler, D. J., P. Reiter, K. L. Ebi, W. Yap, R. Nasci, and J. A. Patz. 2001. Climate variability and change +in the United States: potential impacts on vector- and rodent-borne diseases. Environ Health Perspect +109 Suppl 2:223-233. + +Hashizume, M., B. Armstrong, S. Hajat, Y. Wagatsuma, A. S. Faruque, T. Hayashi, and D. A. Sack. 2008. + +The effect of rainfall on the incidence of cholera in Bangladesh. Epidemiology 19:103-110. + +Hii, Y. L., H. P. Zhu, N. Ng, L. C. Ng, and J. Rocklov. 2012. Forecast of Dengue Incidence Using + +Temperature and Rainfall. Plos Neglected Tropical Diseases 6. + +Hoshiko, S., P. English, D. Smith, and R. Trent. 2010. A simple method for estimating excess mortality due +to heat waves, as applied to the 2006 California heat wave. International Journal of Public Health +55:133-137. + +Huynen, M. M., P. Martens, D. Schram, M. P. Weijenberg, and A. E. Kunst. 2001. The impact of heat +waves and cold spells on mortality rates in the Dutch population. Environ Health Perspect 109:463-470. +IFRCRCS. 2015. World disasters report: focus on local actors, the key to humanitarian effectiveness + +International Federation of Red Cross and Red Crescent Societies. + +Ji, F., Z. H. Wu, J. P. Huang, and E. P. Chassignet. 2014. Evolution of land surface air temperature trend. + +Nature Climate Change 4:462-466. + +Johansson, M. A., D. A. T. Cummings, and G. E. Glass. 2009. Multiyear Climate Variability and Dengue- +El Nino Southern Oscillation, Weather, and Dengue Incidence in Puerto Rico, Mexico, and Thailand: +A Longitudinal Data Analysis. Plos Medicine 6. + +Johnson, L. R., K.D. Lafferty, A. McNally, E. Mordecai, K. Paaijmans, S. Pawar, S.J. Ryan. 2014. +"Mapping the distribution of Malaria: Current Approaches and Future Directions". in Analyzing and +Modeling Spatial and Temporal Dynamics of Infectious Diseases. D. Chen, B. Moulin, J. Wu (Eds). +John Wiley & Sons. pp 189-209. + +Johnson, L.R. Ben-Horin, T., Lafferty, K.D., McNally, A., Mordecai, E., Paaijmans, K.P., Pawar, S., Ryan, +S.J. 2015. Understanding uncertainty in temperature effects on vector-borne disease: A Bayesian +approach. Ecology 96(1): 203-213. + +Kalis, M., and E. Curtiss. 2016. CDC's Drought Guidance: Your Public Health Resource for Understanding + +and Preparing for Drought in Your Community. J Environ Health 78:34-35. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 9 + +Kalis, M. A., M. D. Miller, and R. J. Wilson. 2009. Public health and drought. J Environ Health 72:10-11. +Keatinge, W. R., G. C. Donaldson, K. Bucher, G. Jendritzky, E. Cordioli, M. Martinelli, K. Katsouyanni, +A. E. Kunst, C. McDonald, S. Nayha, I. Vuori, and G. Eurowinter. 2000a. Winter mortality in relation +to climate. Int J Circumpolar Health 59:154-159. + +Keatinge, W. R., G. C. Donaldson, E. Cordioli, M. Martinelli, A. E. Kunst, J. P. Mackenbach, S. Nayha, +and I. Vuori. 2000b. Heat related mortality in warm and cold regions of Europe: observational study. +BMJ 321:670-673. + +Kirshen, P., C. Watson, E. Douglas, A. Gontz, J. Lee, and Y. Tian. 2008. Coastal flooding in the +Northeastern United States due to climate change. Mitigation and Adaptation Strategies for Global +Change 13:437-451. + +Kistemann, T., T. Classen, C. Koch, F. Dangendorf, R. Fischeder, J. Gebel, V. Vacata, and M. Exner. 2002. +Microbial load of drinking water reservoir tributaries during extreme rainfall and runoff. Appl Environ +Microbiol 68:2188-2197. + +Klinger, C., O. Landeg, and V. Murray. 2014. Power outages, extreme events and health: a systematic + +review of the literature from 2011-2012. PLoS Curr 6. + +Knight, D. B., and R. E. Davis. 2009. Contribution of tropical cyclones to extreme rainfall events in the + +southeastern United States. Journal of Geophysical Research-Atmospheres 114. + +Knowlton, K., M. Rotkin-Ellman, G. King, H. G. Margolis, D. Smith, G. Solomon, R. Trent, and P. English. +2009. The 2006 California heat wave: impacts on hospitalizations and emergency department visits. +Environ Health Perspect 117:61-67. + +Knutson, T. R., J. J. Sirutis, G. A. Vecchi, S. Garner, M. Zhao, H. S. Kim, M. Bender, R. E. Tuleya, I. M. +Held, and G. Villarini. 2013. Dynamical Downscaling Projections of Twenty-First-Century Atlantic +Hurricane Activity: CMIP3 and CMIP5 Model-Based Scenarios. Journal of Climate 26:6591-6617. +Kochtubajda, B., M. D. Flannigan, J. R. Gyakum, R. E. Stewart, K. A. Logan, and T. V. Nguyen. 2006. +Lightning and fires in the Northwest Territories and responses to future climate change. Arctic 59:211- +221. + +Kovats, R. S. 2000. El Nino and human health. Bull World Health Organ 78:1127-1135. +Kovats, R. S., M. J. Bouma, S. Hajat, E. Worrall, and A. Haines. 2003. El Nino and health. Lancet 362:1481- + +1489. + +Kovats, R. S., and S. Hajat. 2008. Heat stress and public health: a critical review. Annu Rev Public Health + +29:41-55. + +Kovats, R. S., S. Hajat, and P. Wilkinson. 2004. Contrasting patterns of mortality and hospital admissions + +during hot weather and heat waves in Greater London, UK. Occup Environ Med 61:893-898. + +La Greca, A. M., B. S. Lai, M. M. Llabre, W. K. Silverman, E. M. Vernberg, and M. J. Prinstein. 2013. +Children's Postdisaster Trajectories of PTS Symptoms: Predicting Chronic Distress. Child Youth Care +Forum 42:351-369. + +La Greca, A. M., W. K. Silverman, B. Lai, and J. Jaccard. 2010. Hurricane-Related Exposure Experiences +and Stressors, Other Life Events, and Social Support: Concurrent and Prospective Impact on Children's +Persistent Posttraumatic Stress Symptoms. Journal of Consulting and Clinical Psychology 78:794-805. +Laursen, E., O. Mygind, B. Rasmussen, and T. Ronne. 1994. Gastroenteritis: a waterborne outbreak + +affecting 1600 people in a small Danish town. J Epidemiol Community Health 48:453-458. + +Li, C. F., T. W. Lim, L. L. Han, and R. Fang. 1985. Rainfall, abundance of Aedes aegypti and dengue + +infection in Selangor, Malaysia. Southeast Asian J Trop Med Public Health 16:560-568. + +Liang, S., E. Y. Seto, J. V. Remais, B. Zhong, C. Yang, A. Hubbard, G. M. Davis, X. Gu, D. Qiu, and R. +C. Spear. 2007. Environmental effects on parasitic disease transmission exemplified by schistosomiasis +in western China. Proc Natl Acad Sci U S A 104:7110-7115. + +Lin, S., W. H. Hsu, A. R. Van Zutphen, S. Saha, G. Luber, and S. A. Hwang. 2012. Excessive heat and +respiratory hospitalizations in New York State: estimating current and future public health burden +related to climate change. Environ Health Perspect 120:1571-1577. + +Lindblade, K. A., E. D. Walker, A. W. Onapa, J. Katungu, and M. L. Wilson. 2000. Land use change alters +malaria transmission parameters by modifying temperature in a highland area of Uganda. Tropical +Medicine & International Health 5:263-274. + +Lipp, E. K., A. Huq, and R. R. Colwell. 2002. Effects of global climate on infectious disease: the cholera + +model. Clin Microbiol Rev 15:757-770. + +Loevinsohn, M. E. 1994. Climatic warming and increased malaria incidence in Rwanda. Lancet 343:714- + +718. + + 1 5 0 • S O N G L I A N G E T A L . + +Luque Fernandez, M. A., A. Bauernfeind, J. D. Jimenez, C. L. Gil, N. El Omeiri, and D. H. Guibert. 2009. +Influence of temperature and rainfall on the evolution of cholera epidemics in Lusaka, Zambia, 2003- +2006: analysis of a time series. Trans R Soc Trop Med Hyg 103:137-143. + +Lyle, T. S., and T. Mills. 2016. Assessing coastal flood risk in a changing climate for the City of Vancouver. + +Canadian Water Resources Journal 41:343-352. + +Malmstadt, J., K. Scheltin, and J. Elsner. 2009. Florida hurricanes and damage costs. Southeastern + +Geographer 49:108-131. + +Martinez, C. J., J. J. Maleski, and M. F. Miller. 2012. Trends in precipitation and temperature in Florida, + +USA. Journal of Hydrology 452:259-281. + +McMichael, A. J., and A. Haines. 1997. Global climate change: the potential effects on health. BMJ + +315:805-809. + +Medina-Ramon, M., A. Zanobetti, D. P. Cavanagh, and J. Schwartz. 2006. Extreme temperatures and +mortality: assessing effect modification by personal characteristics and specific cause of death in a +multi-city case-only analysis. Environ Health Perspect 114:1331-1336. + +Medlock, J. M., and S. A. Leach. 2015. Effect of climate change on vector-borne disease risk in the UK. + +Lancet Infect Dis 15:721-730. + +Mhalu, F. S., W. M. Mntenga, and F. D. Mtango. 1987. Seasonality of cholera in Tanzania: possible role of + +rainfall in disease transmission. East Afr Med J 64:378-387. + +Miettinen, I. T., O. Zacheus, C. H. von Bonsdorff, and T. Vartiainen. 2001. Waterborne epidemics in + +Finland in 1998-1999. Water Sci Technol 43:67-71. +Miranda, J. J. 2004. El Nino and health. Lancet 363:247-248. +Mordecai, E.A., Cohen, J., Evans, M.V., Gudapati, P., Johnson, L.R., Lippi, C.A., Miazgowicz, K., +Murdock, C.C., Rohr, J.R., Ryan, S.J., Savage, V., Shocket, M., Stewart Ibarra, A.M., Thomas, M.B., +Weikel, D.P. 2017. Detecting the impact of temperature on transmission of Zika, dengue, and +chikungunya using mechanistic models. PLOS NTDs. 11(4): e0005568 + +Mordecai, E.A., Paaijmans, K.P., Johnson, L.R., Balzer, C.H., Ben-Horin, T., deMoor, E., McNally, A., +Pawar, S., Ryan, S.J., Smith, T.C., Lafferty, K.D., 2013. Physiological constraints dramatically lower +the expected temperature for peak malaria transmission. Ecology Letters 16(1):22-30 + +Neria, Y., and J. M. Shultz. 2012. Mental health effects of Hurricane Sandy: characteristics, potential + +aftermath, and response. JAMA 308:2571-2572. + +Ngwa, M. C., S. Liang, I. T. Kracalik, L. Morris, J. K. Blackburn, L. M. Mbam, S. F. Ba Pouth, A. Teboh, +Y. Yang, M. Arabi, J. D. Sugimoto, and J. G. Morris, Jr. 2016. Cholera in Cameroon, 2000-2012: Spatial +and Temporal Analysis at the Operational (Health District) and Sub Climate Levels. PLoS Negl Trop +Dis 10:e0005105. + +Nichols, G., C. Lane, N. Asgari, N. Q. Verlander, and A. Charlett. 2009. Rainfall and outbreaks of drinking + +water related disease and in England and Wales. J Water Health 7:1-8. + +Ogden, N. H., A. Maarouf, I. K. Barker, M. Bigras-Poulin, L. R. Lindsay, M. G. Morshed, C. J. O'Callaghan, +F. Ramay, D. Waltner-Toews, and D. F. Charron. 2006. Climate change and the potential for range +expansion of the Lyme disease vector Ixodes scapularis in Canada. International Journal for +Parasitology 36:63-70. + +Ostfeld, R. S., C. D. Canham, K. Oggenfuss, R. J. Winchcombe, and F. Keesing. 2006. Climate, deer, +rodents, and acorns as determinants of variation in Lyme-disease risk. Plos Biology 4:1058-1068. +Paaijmans, K. P., S. Blanford, A. S. Bell, J. I. Blanford, A. F. Read, and M. B. Thomas. 2010. Influence of +climate on malaria transmission depends on daily temperature variation. Proc Natl Acad Sci U S A +107:15135-15139. + +Polymenakou, P. N., M. Mandalakis, E. G. Stephanou, and A. Tselepides. 2008. Particle size distribution +of airborne microorganisms and pathogens during an Intense African dust event in the eastern +Mediterranean. Environ Health Perspect 116:292-296. + +Price, C., and D. Rind. 1994. Possible Implications of Global Climate-Change on Global Lightning + +Distributions and Frequencies. Journal of Geophysical Research-Atmospheres 99:10823-10831. + +Pruss-Ustun, A., J. Bartram, T. Clasen, J. M. Colford, O. Cumming, V. Curtis, S. Bonjour, A. D. Dangour, +J. De France, L. Fewtrell, M. C. Freeman, B. Gordon, P. R. Hunter, R. B. Johnston, C. Mathers, D. +Mausezahl, K. Medlicott, M. Neira, M. Stocks, J. Wolf, and S. Cairncross. 2014. Burden of disease +from inadequate water, sanitation and hygiene in low- and middle-income settings: a retrospective +analysis of data from 145 countries. Tropical Medicine & International Health 19:894-905. + + C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 5 1 + +Pruss, A., D. Kay, L. Fewtrell, and J. Bartram. 2002. Estimating the burden of disease from water, sanitation, + +and hygiene at a global level. Environmental Health Perspectives 110:537-542. + +Qian, Z. M., Q. C. He, H. M. Lin, L. L. Kong, C. M. Bentley, W. S. Liu, and D. J. Zhou. 2008. High +temperatures enhanced acute mortality effects of ambient particle pollution in the "Oven" city of +Wuhan, China. Environmental Health Perspectives 116:1172-1178. + +Ragan, P., J. Schulte, S. J. Nelson, and K. T. Jones. 2008. Mortality surveillance: 2004 to 2005 Florida + +hurricane-related deaths. Am J Forensic Med Pathol 29:148-153. + +Reeve, N., and R. Toumi. 1999. Lightning activity as an indicator of climate change. Quarterly Journal of + +the Royal Meteorological Society 125:893-903. + +Ritenour, A. E., M. J. Morton, J. G. McManus, D. J. Barillo, and L. C. Cancio. 2008. Lightning injury: a + +review. Burns 34:585-594. + +Ryan, S.J., McNally, A., Johnson, L.R., Mordecai, E.A., Ben-Horin, T., Paaijmans, K.P., Lafferty, K.D. +2015. Mapping physiological suitability limits of malaria in Africa under climate change. Journal of +vector borne and zoonotic diseases 15(12): 817-725. + +Ryan, S.J., Ben-Horin, T., Johnson, L.R. 2015. Malaria control and senescence: the importance of + +accounting for the pace and shape of ageing in wild mosquitoes. Ecosphere 6(9):170. + +Ryan, S.J. et al. 2017. Climate change drives uncertain global shifts in potential distribution and seasonal + +risk of Aedes-transmitted viruses. BioRXiv doi: https://doi.org/10.1101/172221. + +Sarofim, M. C., S. Saha, M. D. Hawkins, D. M. Mills, J. Hess, R. Horton, P. Kinney, J. Schwartz, and A. +St. Juliana. 2016. Ch. 2: Temperature-Related Death and Illness. Pages 43-68 The Impacts of Climate +Change on Human Health in the United States: A + +Scientific Assessment. U.S. Global Change Research Program, Washington, DC. +Schijven, J. F., and A. M. de Roda Husman. 2005. Effect of climate changes on waterborne disease in The + +Netherlands. Water Sci Technol 51:79-87. + +Schreider, S. Y., D. I. Smith, and A. J. Jakeman. 2000. Climate change impacts on urban flooding. Climatic + +Change 47:91-115. + +Schuster, C. J., A. G. Ellis, W. J. Robertson, D. F. Charron, J. J. Aramini, B. J. Marshall, and D. T. Medeiros. +2005. Infectious disease outbreaks related to drinking water in Canada, 1974-2001. Can J Public Health +96:254-258. + +Senhorst, H. A., and J. J. Zwolsman. 2005. Climate change and effects on water quality: a first impression. + +Water Sci Technol 51:53-59. + +Shaman, J., J. F. Day, and M. Stieglitz. 2005. Drought-induced amplification and epidemic transmission of + +West Nile virus in southern Florida. J Med Entomol 42:134-141. + +Singh, R. B., S. Hales, N. de Wet, R. Raj, M. Hearnden, and P. Weinstein. 2001. The influence of climate +variation and change on diarrheal disease in the Pacific Islands. Environ Health Perspect 109:155-159. +Soverow, J. E., G. A. Wellenius, D. N. Fisman, and M. A. Mittleman. 2009. Infectious Disease in a +Warming World: How Weather Influenced West Nile Virus in the United States (2001-2005). +Environmental Health Perspectives 117:1049-1052. + +Su, G. L. S. 2008. Correlation of climatic factors and dengue incidence in Metro Manila, Philippines. Ambio + +37:292-294. + +Sullivent, E. E., 3rd, C. A. West, R. S. Noe, K. E. Thomas, L. J. Wallace, and R. T. Leeb. 2006. Nonfatal + +injuries following Hurricane Katrina--New Orleans, Louisiana, 2005. J Safety Res 37:213-217. + +Taylor, G., and W. J. Davies. 1990. Root-Growth of Fagus-Sylvatica - Impact of Air-Quality and Drought + +at a Site in Southern Britain. New Phytologist 116:457-464. + +Thomas, K. M., D. F. Charron, D. Waltner-Toews, C. Schuster, A. R. Maarouf, and J. D. Holt. 2006. A role +of high impact weather events in waterborne disease outbreaks in Canada, 1975 - 2001. Int J Environ +Health Res 16:167-180. + +Thompson, K. R., N. B. Bernier, and P. Chan. 2009. Extreme sea levels, coastal flooding and climate change + +with a focus on Atlantic Canada. Natural Hazards 51:139-150. + +Thomson, M. C., S. J. Mason, T. Phindela, and S. J. Connor. 2005. Use of rainfall and sea surface +temperature monitoring for malaria early warning in Botswana. American Journal of Tropical Medicine +and Hygiene 73:214-221. + +Tian, L., Y. Bi, S. C. Ho, W. Liu, S. Liang, W. B. Goggins, E. Y. Chan, S. Zhou, and J. J. Sung. 2008. One- +year delayed effect of fog on malaria transmission: a time-series analysis in the rain forest area of +Mengla County, south-west China. Malar J 7:110. + +USEPA. 2016. Climate change indicators in the United States. Page 96 in U. S. E. P. Agency, editor. + + 1 5 2 • S O N G L I A N G E T A L . + +van der Hoek, W., F. Konradsen, D. Perera, P. H. Amerasinghe, and F. P. Amerasinghe. 1997. Correlation +between rainfall and malaria in the dry zone of Sri Lanka. Annals of Tropical Medicine and Parasitology +91:945-949. + +Vernberg, E. M., A. M. LaGreca, W. K. Silverman, and M. J. Prinstein. 1996. Prediction of posttraumatic +stress symptoms in children after Hurricane Andrew. Journal of Abnormal Psychology 105:237-248. +Wardrop, N. A., A. G. Barnett, J. A. Atkinson, and A. C. A. Clements. 2013. Plasmodium vivax malaria +incidence over time and its association with temperature and rainfall in four counties of Yunnan +Province, China. Malaria Journal 12. + +Weems, C. F., L. K. Taylor, M. F. Cannon, R. C. Marino, D. M. Romano, B. G. Scott, A. M. Perry, and V. +Triplett. 2010. Post Traumatic Stress, Context, and the Lingering Effects of the Hurricane Katrina +Disaster among Ethnic Minority Youth. Journal of Abnormal Child Psychology 38:49-56. + +WHO. 2008. Global Burden of Disease 2004 update. World Health Organization. +Winsberg, M. D. 2003. Florida weather. 2nd edition. University Press of Florida, Gainesville, Fla. +Yang, K., J. LeJeune, D. Alsdorf, B. Lu, C. K. Shum, and S. Liang. 2012. Global distribution of outbreaks + +of water-associated infectious diseases. PLoS Negl Trop Dis 6:e1483. + diff --git a/q143/random_k4/random_2.md b/q143/random_k4/random_2.md index 69d1305f3a9c9c9fb312a4e7dbd3718670cacdd2..a5f5d2f89373cf725b1140b751b41f37e641aa63 100644 --- a/q143/random_k4/random_2.md +++ b/q143/random_k4/random_2.md @@ -1,13 +1,34 @@ -3 +Published as a conference paper at ICLR 2024 + +CAN LARGE LANGUAGE MODELS INFER +CAUSATION FROM CORRELATION? + +Jiarui Liu3,∗ + +Zhijing Jin1,2,∗,‡ +Mrinmaya Sachan2 Rada Mihalcea6 Mona Diab3,‡,† Bernhard Schölkopf1,† +1Max Planck Institute for Intelligent Systems, Tübingen, Germany 2ETH Zürich +3LTI, CMU 4University of Hong Kong 5Meta AI +jinzhi@ethz.ch + +Zhiheng Lyu4 Spencer Poff5 + +zhihenglyu.cs@gmail.com + +6University of Michigan + +jiarui@cmu.edu + +4 2 0 2 r -a - -M +p +A 7 +1 ] L @@ -17,17 +38,17 @@ s c [ -4 +3 v -1 -7 -1 -1 -1 -. +6 3 +8 +5 0 -2 +. +6 +0 +3 2 : v @@ -36,2168 +57,1729 @@ X r a -Published as a conference paper at ICLR 2023 - -SELF-CONSISTENCY IMPROVES CHAIN OF THOUGHT -REASONING IN LANGUAGE MODELS - -Jason Wei† Dale Schuurmans† Quoc Le† Ed H. Chi† - -Xuezhi Wang†‡ -Sharan Narang† Aakanksha Chowdhery† Denny Zhou†§ -†Google Research, Brain Team -‡xuezhiw@google.com, §dennyzhou@google.com - ABSTRACT -Chain-of-thought prompting combined with pre-trained large language models has -achieved encouraging results on complex reasoning tasks. In this paper, we propose -a new decoding strategy, self-consistency, to replace the naive greedy decoding -used in chain-of-thought prompting. It first samples a diverse set of reasoning paths -instead of only taking the greedy one, and then selects the most consistent answer -by marginalizing out the sampled reasoning paths. Self-consistency leverages the -intuition that a complex reasoning problem typically admits multiple different ways -of thinking leading to its unique correct answer. Our extensive empirical evaluation -shows that self-consistency boosts the performance of chain-of-thought prompting -with a striking margin on a range of popular arithmetic and commonsense reasoning -benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), -StrategyQA (+6.4%) and ARC-challenge (+3.9%). +Causal inference is one of the hallmarks of human intelligence. While the field +of Causal NLP has attracted much interest in the recent years, existing causal +inference datasets in NLP primarily rely on discovering causality from empirical +knowledge (e.g., commonsense knowledge). In this work, we propose the first +benchmark dataset to test the pure causal inference skills of large language models +(LLMs). Specifically, we formulate a novel task CORR2CAUSE, which takes a +set of correlational statements and determines the causal relationship between +the variables. We curate a large-scale dataset of more than 200K samples, on +which we evaluate seventeen existing LLMs. Through our experiments, we identify +a key shortcoming of LLMs in terms of their causal inference skills, and show +that these models achieve almost close to random performance on the task. This +shortcoming is somewhat mitigated when we try to re-purpose LLMs for this skill +via finetuning, but we find that these models still fail to generalize – they can only +perform causal inference in in-distribution settings when variable names and textual +expressions used in the queries are similar to those in the training set, but fail in +out-of-distribution settings generated by perturbing these queries. CORR2CAUSE +is a challenging task for LLMs, and can be helpful in guiding future research on +improving LLMs’ pure reasoning skills and generalizability.1 1 INTRODUCTION -Although language models have demonstrated remarkable success across a range of NLP tasks, their -ability to demonstrate reasoning is often seen as a limitation, which cannot be overcome solely by -increasing model scale (Rae et al., 2021; BIG-bench collaboration, 2021, inter alia). In an effort -to address this shortcoming, Wei et al. (2022) have proposed chain-of-thought prompting, where -a language model is prompted to generate a series of short sentences that mimic the reasoning -process a person might employ in solving a task. For example, given the question “If there are 3 -cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?”, instead -of directly responding with “5”, a language model would be prompted to respond with the entire -chain-of-thought: “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars. The answer is 5.”. It has been observed that chain-of-thought prompting significantly -improves model performance across a variety of multi-step reasoning tasks (Wei et al., 2022). - -In this paper, we introduce a novel decoding strategy called self-consistency to replace the greedy -decoding strategy used in chain-of-thought prompting (Wei et al., 2022), that further improves -language models’ reasoning performance by a significant margin. Self-consistency leverages the -intuition that complex reasoning tasks typically admit multiple reasoning paths that reach a correct -answer (Stanovich & West, 2000). The more that deliberate thinking and analysis is required for a -problem (Evans, 2010), the greater the diversity of reasoning paths that can recover the answer. - -Figure 1 illustrates the self-consistency method with an example. We first prompt the language model -with chain-of-thought prompting, then instead of greedily decoding the optimal reasoning path, we -propose a “sample-and-marginalize” decoding procedure: we first sample from the language model’s -decoder to generate a diverse set of reasoning paths; each reasoning path might lead to a different -final answer, so we determine the optimal answer by marginalizing out the sampled reasoning paths -to find the most consistent answer in the final answer set. Such an approach is analogous to the -human experience that if multiple different ways of thinking lead to the same answer, one has greater -confidence that the final answer is correct. Compared to other decoding methods, self-consistency -avoids the repetitiveness and local-optimality that plague greedy decoding, while mitigating the -stochasticity of a single sampled generation. - -1 - - Published as a conference paper at ICLR 2023 - -Figure 1: The self-consistency method contains three steps: (1) prompt a language model using -chain-of-thought (CoT) prompting; (2) replace the “greedy decode” in CoT prompting by sampling -from the language model’s decoder to generate a diverse set of reasoning paths; and (3) marginalize -out the reasoning paths and aggregate by choosing the most consistent answer in the final answer set. - -Self-consistency is far simpler than prior approaches that either train an additional verifier (Cobbe -et al., 2021) or train a re-ranker given additional human annotations to improve generation quality -(Thoppilan et al., 2022). Instead, self-consistency is entirely unsupervised, works off-the-shelf with -pre-trained language models, requires no additional human annotation, and avoids any additional -training, auxiliary models or fine-tuning. Self-consistency also differs from a typical ensemble -approach where multiple models are trained and the outputs from each model are aggregated, it acts -more like a “self-ensemble” that works on top of a single language model. - -We evaluate self-consistency on a wide range of arithmetic and commonsense reasoning tasks over -four language models with varying scales: the public UL2-20B (Tay et al., 2022) and GPT-3-175B -(Brown et al., 2020), and two densely-activated decoder-only language models: LaMDA-137B -(Thoppilan et al., 2022) and PaLM-540B (Chowdhery et al., 2022). On all four language models, -self-consistency improves over chain-of-thought prompting by a striking margin across all tasks. In -particular, when used with PaLM-540B or GPT-3, self-consistency achieves new state-of-the-art levels -of performance across arithmetic reasoning tasks, including GSM8K (Cobbe et al., 2021) (+17.9% -absolute accuracy gains), SVAMP (Patel et al., 2021) (+11.0%), AQuA (Ling et al., 2017) (+12.2%), -and across commonsense reasoning tasks such as StrategyQA (Geva et al., 2021) (+6.4%) and ARC- -challenge (Clark et al., 2018) (+3.9%). In additional experiments, we show self-consistency can -robustly boost performance on NLP tasks where adding a chain-of-thought might hurt performance -compared to standard prompting (Ye & Durrett, 2022). We also show self-consistency significantly -outperforms sample-and-rank, beam search, ensemble-based approaches, and is robust to sampling -strategies and imperfect prompts. - -2 SELF-CONSISTENCY OVER DIVERSE REASONING PATHS - -A salient aspect of humanity is that people think differently. It is natural to suppose that in tasks -requiring deliberate thinking, there are likely several ways to attack the problem. We propose that -such a process can be simulated in language models via sampling from the language model’s decoder. -For instance, as shown in Figure 1, a model can generate several plausible responses to a math -question that all arrive at the same correct answer (Outputs 1 and 3). Since language models are not -perfect reasoners, the model might also produce an incorrect reasoning path or make a mistake in -one of the reasoning steps (e.g., in Output 2), but such solutions are less likely to arrive at the same -answer. That is, we hypothesize that correct reasoning processes, even if they are diverse, tend to -have greater agreement in their final answer than incorrect processes. - -We leverage this intuition by proposing the following self-consistency method. First, a language -model is prompted with a set of manually written chain-of-thought exemplars (Wei et al., 2022). Next, - -2 - -Language modelQ: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.…Q: Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder for $2 per egg. How much does she make every day?A:She has 16 - 3 - 4 = 9 eggs left. So she makes $2 * 9 = $18 per day. Sample a diverse set of reasoning pathsShe eats 3 for breakfast, so she has 16 - 3 = 13 left. Then she bakes muffins, so she has 13 - 4 = 9 eggs left. So she has 9 eggs * $2 = $18. This means she she sells the remainder for $2 * (16 - 4 - 3) = $26 per day. The answer is $18.The answer is $26.The answer is $18.The answer is $18.Marginalize out reasoning paths to aggregate final answersLanguage modelThis means she uses 3 + 4 = 7 eggs every day. She sells the remainder for $2 per egg, so in total she sells 7 * $2 = $14 per day. The answer is $14.The answer is $14.Greedy decode Published as a conference paper at ICLR 2023 - -GSM8K MultiArith +Causal inference, i.e., the ability to establish the correct causal relationships between variables or +events, is fundamental to human intelligence. There are two distinct ways this causal inference +capability can be acquired: one through empirical knowledge, e.g., we know from common sense that +touching a hot stove will get us burned; the other through pure causal reasoning, as causality can be +formally argued and reasoned about using known procedures and rules from causal inference (Spirtes +et al., 2000; Pearl, 2009; Peters et al., 2017). One example is that we have the a priori knowledge that +the correlation between A and B does not necessarily imply causality. This is a formal rule that holds +true regardless of the realizations of the variables A and B. -AQuA +With the rise of large language models (LLMs) (Radford et al., 2019; Devlin et al., 2019; Ouyang +et al., 2022; Zhang et al., 2022; OpenAI, 2023, inter alia), a crucial research question is whether +they can do causal reasoning well. Recent studies have pointed out that LLMs are “causal parrots,” +which recite the causal knowledge in the training data (Zeˇcevi´c et al., 2023). Moreover, the vast +majority of studies frame causal reasoning as a skill to navigate around empirical knowledge (Gordon +et al., 2012; Sap et al., 2019a;b; Qin et al., 2019; Bhagavatula et al., 2020), and also treat LLMs as a +knowledge base when evaluating its causal skills (Kıcıman et al., 2023; Tu et al., 2023; Xie et al., +2023). However, all the above lines of research frame causality as empirical knowledge, thus relying +heavily on the quality and the coverage of the training data, overlooking the great potential of the +formal causal reasoning skills to process correlational information to causal conclusions. -SVAMP +∗Equal contribution. †Equal supervision. ‡Work originated as a Meta AI internship project involving Zhijing, -CSQA +Mona, and Spencer. -ARC-c +1Our data is at https://huggingface.co/datasets/causalnlp/corr2cause. -Greedy decode +Our code is at https://github.com/causalNLP/corr2cause. -56.5 - -94.7 - -35.8 - -79.0 - -79.0 +1 -85.2 + Published as a conference paper at ICLR 2024 -Weighted avg (unnormalized) -Weighted avg (normalized) +Figure 1: Illustration of the motivation behind our task and dataset. -56.3 ± 0.0 -22.1 ± 0.0 +Drawing inspirations from technical studies on causal discovery (Spirtes et al., 2000; Spirtes & Zhang, +2016; Glymour et al., 2019), we formulate a novel task for NLP, correlation-to-causation inference +(CORR2CAUSE), which is an important skill for LLMs. Imagine the scenario in Figure 1, where +the training corpus does not tediously cover every causal relation, but more pervasively talk about +correlations, such as which events tend to co-occur. Learning a good CORR2CAUSE skill can enable +LLMs to draw causal relations behind the mere correlational information on the surface. For example, +several decades ago, there might be an observation that female university students tend to perform +better, but behind the correlational statistics is the causal graph that female students have to achieve +extra good performance to get into universities as the first place. -90.5 ± 0.0 -59.7 ± 0.0 +To this end, we collect the CORR2CAUSE dataset, the first dataset to test the pure causal reasoning +abilities of LLMs. All the questions in this dataset are centered around testing when it is valid +or invalid to infer causation from correlation. To systematically compose this dataset, we ground +our generalization process in the formal framework of causal discovery (Spirtes et al., 1993; 2000; +Glymour et al., 2016; Spirtes & Zhang, 2016), which provides rules about how to deduce causal +relations among variables given their statistical correlation in the observational data. We generate +more than 200K data points, and label a correlation-causation statement pair as valid if and only if +there is a bijective mapping between the statistical correlation and the underlying causality. -35.8 ± 0.0 73.0 ± 0.0 74.8 ± 0.0 82.3 ± 0.0 -15.7 ± 0.0 40.5 ± 0.0 52.1 ± 0.0 51.7 ± 0.0 +Based on our CORR2CAUSE dataset with 200K samples, we investigate two main research questions: +(1) How well do existing LLMs perform on this task? (2) Can existing LLMs be re-trained or +re-purposed on this task and obtain robust causal inference skills? Through extensive experiments, +we show empirically that none of the 17 existing LLMs we investigate perform well on this pure +causal inference task. We also show that although LLMs can demonstrate better performance after +being finetuned on the data, the causal inference skills attained by them are not robust. In summary, +our contributions are as follows: -Weighted sum (unnormalized) -Weighted sum (normalized) +1. We propose the novel task of CORR2CAUSE, to probe an aspect of LLM’s reasoning ability, -59.9 ± 0.0 -74.1 ± 0.0 +pure causal inference; -92.2 ± 0.0 -99.3 ± 0.0 +2. We compose a dataset of over 200K samples, using insights from causal discovery; +3. We evaluate the performance of 17 LLMs on our dataset, finding that all of them perform -38.2 ± 0.0 76.2 ± 0.0 76.2 ± 0.0 83.5 ± 0.0 -48.0 ± 0.0 86.8 ± 0.0 80.7 ± 0.0 88.7 ± 0.0 +poorly, close to the random baseline; -Unweighted sum (majority vote) 74.4 ± 0.1 +4. We further explored whether LLMs can learn the skill through finetuning, and find that +LLMs fail to robustly acquire this skill in out-of-distribution settings. Finally, we suggest +future work to explore more ways to enhance the pure causal inference skill in LLMs. -99.3 ± 0.0 +2 PRELIMINARIES: CAUSAL INFERENCE -48.3 ± 0.5 86.6 ± 0.1 80.7 ± 0.1 88.7 ± 0.1 +2.1 DIRECTED GRAPHICAL CAUSAL MODELS (DGCMS) -Table 1: Accuracy comparison of different answer aggregation strategies on PaLM-540B. +A directed graphical causal model (DGCM) is a commonly used representation to express the causal +relations among a set of variables. Given a set of N variables X = {X1, . . . , XN }, we can encode +the causal relations among them using a directed graph G := (X, E), where E is the set of directed +edges. Each edge ei,j ∈ E represents a causal link Xi → Xj, meaning that Xi is a direct cause of +Xj. In the context of this work, we take the common assumption of directed acyclic graphs (DAGs), +which most causal discovery methods use (Glymour et al., 2019), as graphs with cycles can make the +causal discovery process arbitrarily hard. -we sample a set of candidate outputs from the language model’s decoder, generating a diverse set of -candidate reasoning paths. Self-consistency is compatible with most existing sampling algorithms, -including temperature sampling (Ackley et al., 1985; Ficler & Goldberg, 2017), top-k sampling (Fan -et al., 2018; Holtzman et al., 2018; Radford et al., 2019), and nucleus sampling (Holtzman et al., -2020). Finally, we aggregate the answers by marginalizing out the sampled reasoning paths and -choosing the answer that is the most consistent among the generated answers. -In more detail, assume the generated answers ai are from a fixed answer set, ai ∈ A, where -i = 1, . . . , m indexes the m candidate outputs sampled from the decoder. Given a prompt and a -question, self-consistency introduces an additional latent variable ri, which is a sequence of tokens -representing the reasoning path in the i-th output, then couples the generation of (ri, ai) where -ri → ai, i.e., generating a reasoning path ri is optional and only used to reach the final answer ai. As -an example, consider Output 3 from Figure 1: the first few sentences “She eats 3 for breakfast ... So -she has 9 eggs * $2 = $18.” constitutes ri, while the answer 18 from the last sentence, “The answer -is $18”, is parsed as ai.1 After sampling multiple (ri, ai) from the model’s decoder, self-consistency -1(ai = a), -applies a marginalization over ri by taking a majority vote over ai, i.e., arg maxa -or as we defined as the most “consistent” answer among the final answer set. +2 -(cid:80)m +Previous tasks:Getting burnedTouching a hot stoveCausesNoSuppose we know that A correlates with B.Can we infer that A causes B?Skill being tested in previous work: Empiricalknowledge instead of pure causal inference.Training CorpusIce cream salesHow can LLMs process suchinformation? Correlation?Causation? What causes what?We propose a new task: Corr2Cause InferenceThis requires the skill of inferring causation from correlation*Assumption that we explicitly mention in the samples: We suppose a close system of the given variables and correlations.YesA correlates with B. C correlates with B. However, A is independent of C.Can we infer that A causes B???Drowning casesHot weatherVaccinationAutismTech company CEOsCollege dropoutFatality rateManycorrelations Published as a conference paper at ICLR 2024 + +Following the graph-theoretic terminology, we use an analogy of the ancestry tree to denote the +relations between two variables. For example, we call Xi as a parent of Xj if there is a directed edge +Xi → Xj in the graph, and, thus, Xj is a child of Xi. Similarly, we denote Xi as an ancestor of Xj +if there exists a directed path from Xi to Xj, and, thus, Xj is a descendent of Xi. Note that a parent +is a special case of an ancestor where the directed path has a length of 1. + +For convenience, we also introduce the notions for some special three-variable relations. Given two +variables Xi and Xj, we call a third variable Xk a confounder (i.e., common cause) if Xk is a parent +of both Xi and Xj; a collider (i.e., common effect) if Xk is a child of both Xi and Xj; and a mediator +if Xk is both a child of Xi, and a parent of Xj. + +2.2 D-SEPARATION AND MARKOV PROPERTY + +D-Separation D-separation (Pearl, 1988) is a fundamental concept in graphical models used to +determine whether two sets of nodes X and Y in a DAG G are conditionally independent given a +third set of nodes Z, where the three sets are disjoint. We say that X and Y are d-separated by Z if +all paths between any node in X and any node in Y are blocked by the conditioning set Z. A path +between X and Y is blocked by Z if there exists a node A ∈ Z which satisfies one of the following +conditions: A is the parent node in a fork structure on the path (i.e., · ← A → ·); A is the mediator +node in a chain structure on the path (i.e., · → A → ·); or in any collider structure on the path (i.e., +· → A ← ·), Z does not contain A or its descendants. + +Markov Property The Markov property in a DAG G states that each node Xi is conditionally +independent of its non-descendants given its parents, namely Xi ⊥⊥ NonDe(Xi)| Pa(Xi), where +NonDe(Xi) denotes the non-descendants of Xi excluding itself, and Pa(Xi) denotes the parents +of Xi. Using the Markov property, we can factorize the joint distribution of all the nodes in the +graph into P (X1, . . . , XN ) = (cid:81)N +i=1 P (Xi|PA(Xi)). To infer the causal graph from probability +distributions, a common assumption is faithfulness, namely the validity to infer all the d-separation +sets in the graph from the independence relations in the probability distribution. In our work, we also +take this broadly taken assumption which holds for most real-world scenarios. + +Markov Equivalence of Graphs We denote two DAGs as Markov equivalent if they induce the +same joint distribution P (X). The set of DAGs that are Markov equivalent to each other is called +a Markov equivalence class (MEC). Causal graphs in the same MEC can be easily identified since +they have the same skeleton (i.e., undirected edges) and V-structures (i.e., structures in the form of +A → B ← C where A and C are not connected). + +Obviously, there is a one-to-many mapping (i.e., surjection) between the causal graph and statistical +distribution. Namely, each causal graph sufficiently determines a statistical distribution, but from +a statistical distribution, we cannot necessarily induce a unique causal graph. This is why we say +“correlation does not necessarily mean causation”. + +2.3 CAUSAL DISCOVERY + +Causal discovery aims to learn the causal relations by analyzing statistical properties in the obser- +vational data (Spirtes et al., 1993; 2000; Glymour et al., 2016; Spirtes & Zhang, 2016; Glymour +et al., 2019). It can be achieved through constraint-based methods (Spirtes et al., 2000), score-based +methods (Chickering, 2002), or other methods taking advantage of the functional causal models +(Shimizu et al., 2006; Hoyer et al., 2008; Zhang & Hyvärinen, 2009). + +To fit for the spirit of this paper to infer from correlation (expressed in natural language) to causation, +we base our dataset design on the widely-used Peter-Clark (PC) algorithm (Spirtes et al., 2000). +The PC algorithm is based on the principles of conditional independence and the causal Markov +assumption, which allows it to efficiently identify causal relationships among variables in a given +dataset. The algorithm first starts with a fully connected undirected graph among all the variables. +Then it removes the edge between two variables if there is an unconditional or conditional inde- +pendence relationship between them. Afterwards, it orients the directed edges whenever there is a +V-structure. And finally, it iteratively checks the direction of the other edges until the entire causal +graph is consistent with all the statistical correlations. -i=1 +3 -In Table 1, we show the test accuracy over a set of reasoning tasks by using different answer -aggregation strategies. In addition to majority vote, one can also weight each (ri, ai) by P (ri, ai | -prompt, question) when aggregating the answers. Note to compute P (ri, ai | prompt, question), we -can either take the unnormalized probability of the model generating (ri, ai) given (prompt, question), -or we can normalize the conditional probability by the output length (Brown et al., 2020), i.e., + Published as a conference paper at ICLR 2024 -P (ri, ai | prompt, question) = exp +Figure 2: Pipeline of the data construction process. -1 -K +3 DATASET CONSTRUCTION -(cid:80)K +We introduce the construction of our dataset in this section. We start with our task formulation for +CORR2CAUSE, and then briefly give an overview of the data generation process, followed by detailed +descriptions of each step. We conclude the section with the overall statistics of the dataset. -k=1 log P (tk|prompt,question,t1,...,tk−1), +3.1 TASK FORMULATION +Given a set of N variables X = {X1, . . . , XN }, we have a statement s about all the correlations +among the variables, and a hypothesis h describing the causal relation r between the pair of variables +Xi and Xj. The task is to learn a function f : (s, h) (cid:55)→ v which maps the correlation statement +s and the causal relation hypothesis h to their validity v ∈ {0, 1}, which takes the value 0 if this +inference is invalid, and the value 1 if this inference is valid. -(1) +3.2 OVERVIEW OF THE DATA GENERATION PROCESS +We base the construction our dataset on several concepts of causal inference, including the DGCM, +d-separation, and MECs, as introduced in Section 2. -where log P (tk | prompt, question, t1, . . . , tk−1) is the log probability of generating the k-th token -tk in (ri, ai) conditioned on the previous tokens, and K is the total number of tokens in (ri, ai). -In Table 1, we show that taking the “unweighted sum”, i.e., taking a majority vote directly over ai -yields a very similar accuracy as aggregating using the “normalized weighted sum”. We took a closer -look at the model’s output probabilities and found this is because for each (ri, ai), the normalized -conditional probabilities P (ri, ai | prompt, question) are quite close to each other, i.e., the language -model regards those generations as “similarly likely”.2 Additionally, when aggregating the answers, -the results in Table 1 show that the “normalized” weighted sum (i.e., Equation 1) yields a much -higher accuracy compared to its unnormalized counterpart. For completeness, in Table 1 we also -report the results by taking a “weighted average”, i.e., each a gets a score of its weighted sum divided -by (cid:80)m -i=1 +As in the overview of our data generation process in Figure 2, we first choose the number N of +variables (Step 1) and generate all the unique DGCMs with N nodes (Step 2), which we will introduce +in the Section 3.3. Then we collect all the d-separation sets from these graphs to identify MECs +(Step 3) in Section 3.4. Then, in Step 4, we create the formal form of data in Section 3.5. For each +correspondence of the MEC to causal graphs, we compose the correlation statement based on the +statistical relations in the MEC, and hypothesize a causal relation between two variables, and produce +the validity v = 1 if the hypothesis is a shared property of all causal graphs in the MEC, and v = 0 if +the hypothesis is not necessarily true for all the MEC graphs. Finally, we introduce the verbalization +process in Section 3.6. -1(ai = a), which results in a much worse performance. +3.3 CONSTRUCTING THE GRAPHS WITH ISOMORPHISM CHECKS +The first step of the data generation is to compose the causal graphs, as in Step 1 and 2 of Figure 2. +For a set of N variables X = {X1, . . . , XN }, there are N (N − 1) possible directed edges, since +each node can link to any node other than itself. To remove cycles in the graph, we make the nodes +in topological order, which only allows edges Xi → Xj, where i < j. We achieve this by limiting +the adjacency matrix of the graph to only having non-zero values above the diagonal, resulting in +N (N − 1)/2 possible directed edges for the DAGs. -Self-consistency explores an interesting space between open-ended text generation and optimal -text generation with a fixed answer. Reasoning tasks typically have fixed answers, which is why -researchers have generally considered greedy decoding approaches (Radford et al., 2019; Wei et al., -2022; Chowdhery et al., 2022). However, we have found that even when the desired answer is fixed, -introducing diversity in the reasoning processes can be highly beneficial; therefore we leverage +At the first glance, for N nodes, there should be 2N (N −1)/2 possible DAGs (i.e., the power set of all +edges). However, there could be isomorphic graphs in this set. To avoid this, we perform a graph -1The parser is task dependent. For arithmetic reasoning, we parse the first numerical part as the final answer -after the model generates “The answer is ”. For commonsense reasoning, we parse the full string answer as the -final answer after the model generates “The answer is ”. Most generated outputs have a consistent format of -“{Reasoning paths}. The answer is X.” if we prompt the language model in this format. +4 -2This also means that the language model is not well calibrated and thus cannot distinguish well between -correct solutions and wrong solutions, which also explains why additional re-rankers were trained to better judge -the quality of the solutions in previous work (Cobbe et al., 2021; Thoppilan et al., 2022). +ABC1. Choose thenumber of variables2. Generate all unique causalgraphsBACE.g., N=3ABC...3. Map each graph to a set ofstatistical correlationsACB● A ⫫̸ B,● A ⫫̸ C, ● B ⫫̸ C, and● A ⫫ C | B● A ⫫̸ B,● B ⫫̸ C, and● A ⫫ C Causal GraphsCorrelationsABC1-to-1 MappingMany-to-1MappingCorrelationsHypothesizedCausationValiditySuppose there is a closed system of 3 variables, A, B and C. All the statisticalrelations among these 3 variables are as follows:A correlates with C. B correlates with C. However, A is independent of B.A directly causes B.[The validity label is equivalent to the results after running the PC algorithm.I.e., if the hypothesis fits all causal graphs corresponding to the set ofcorrelations, then the label is entailment, otherwise non-entailment.]ValidVerbalize the statistical correlationsHypothesize a causal relation between two nodes4. Compose the Data Published as a conference paper at ICLR 2024 +# Nodes +2 3 - - Published as a conference paper at ICLR 2023 - -sampling, as commonly used for open-ended text generation (Radford et al., 2019; Brown et al., 2020; -Thoppilan et al., 2022), to achieve this goal. One should note that self-consistency can be applied -only to problems where the final answer is from a fixed answer set, but in principle this approach can -be extended to open-text generation problems if a good metric of consistency can be defined between -multiple generations, e.g., whether two answers agree or contradict each other. - -3 EXPERIMENTS - -We conducted a series of experiments to compare the proposed self-consistency method with existing -approaches on a range of reasoning benchmarks. We find that self-consistency robustly improves -reasoning accuracy for every language model considered, spanning a wide range of model scales. - -3.1 EXPERIMENT SETUP - -Tasks and datasets. We evaluate self-consistency on the following reasoning benchmarks.3 - -• Arithmetic reasoning. For these tasks, we used the Math Word Problem Repository (Koncel- -Kedziorski et al., 2016), including AddSub (Hosseini et al., 2014), MultiArith (Roy & Roth, -2015), and ASDiv (Miao et al., 2020). We also included AQUA-RAT (Ling et al., 2017), a -recently published benchmark of grade-school-math problems (GSM8K; Cobbe et al., 2021), -and a challenge dataset over math word problems (SVAMP; Patel et al., 2021). - -• Commonsense reasoning. For these tasks, we used CommonsenseQA (Talmor et al., 2019), -StrategyQA (Geva et al., 2021), and the AI2 Reasoning Challenge (ARC) (Clark et al., 2018). -• Symbolic Reasoning. We evaluate two symbolic reasoning tasks: last letter concatenation (e.g., -the input is “Elon Musk” and the output should be “nk”), and Coinflip (e.g., a coin is heads-up, -after a few flips is the coin still heads-up?) from Wei et al. (2022). - -Language models and prompts. We evaluate self-consistency over four transformer-based lan- -guage models with varying scales: - -• UL2 (Tay et al., 2022) is an encoder-decoder model trained on a mixture of denoisers with 20- -billion parameters. UL2 is completely open-sourced4 and has similar or better performance than -GPT-3 on zero-shot SuperGLUE, with only 20B parameters and thus is more compute-friendly; -• GPT-3 (Brown et al., 2020) with 175-billion parameters. We use two public engines code-davinci- -001 and code-davinci-002 from the Codex series (Chen et al., 2021) to aid reproducibility;5 -• LaMDA-137B (Thoppilan et al., 2022) is a dense left-to-right, decoder-only language model with -137-billion parameters, pre-trained on a mixture of web documents, dialog data and Wikipedia; -• PaLM-540B (Chowdhery et al., 2022) is a dense left-to-right, decoder-only language model with -540-billion parameters, pre-trained on a high quality corpus of 780 billion tokens with filtered -webpages, books, Wikipedia, news articles, source code, and social media conversations. -We perform all experiments in the few-shot setting, without training or fine-tuning the language -models. For a fair comparison we use the same prompts as in Wei et al. (2022): for all arithmetic -reasoning tasks we use the same set of 8 manually written exemplars; for each commonsense -reasoning task, 4-7 exemplars are randomly chosen from the training set with manually composed -chain-of-thought prompts.6 Full details on the prompts used are given in Appendix A.3. - -Sampling scheme. To sample diverse reasoning paths, we followed similar settings to those -suggested in Radford et al. (2019); Holtzman et al. (2020) for open-text generation. In particular, for -UL2-20B and LaMDA-137B we applied temperature sampling with T = 0.5 and truncated at the -top-k (k = 40) tokens with the highest probability, for PaLM-540B we applied T = 0.7, k = 40, and -for GPT-3 we use T = 0.7 without top-k truncation. We provide an ablation study in Section 3.5 to -show that self-consistency is generally robust to sampling strategies and parameters. - -3By default we use the test split for all datasets if the labels are available for evaluation. For CommonsenseQA -we use the dev split; for StrategyQA we use the question-only set from BIG-bench collaboration (2021): -https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/strategyqa. - -4Model checkpoints at https://github.com/google-research/google-research/tree/master/ul2. -5Public API available at https://openai.com/api/. -6Self-consistency is robust to different sets of prompts and we provide a study in Appendix A.1.2. - 4 - - Published as a conference paper at ICLR 2023 - -3.2 MAIN RESULTS - -We report the results of self-consistency averaged over 10 runs, where we sampled 40 outputs -independently from the decoder in each run. The baseline we compare to is chain-of-thought -prompting with greedy decoding (Wei et al., 2022), referred to as CoT-prompting, which has been -previously used for decoding in large language models (Chowdhery et al., 2022). - -Arithmetic Reasoning The results are shown in Table 2.7 Self-consistency improves the arithmetic -reasoning performance over all four language models significantly over chain-of-thought prompting. -More surprisingly, the gains become more significant when the language model’s scale increases, -e.g., we see +3%-6% absolute accuracy improvement over UL2-20B but +9%-23% for LaMDA- -137B and GPT-3. For larger models that already achieve high accuracy on most tasks (e.g., GPT-3 -and PaLM-540B), self-consistency still contributes significant additional gains with +12%-18% -absolute accuracy on tasks like AQuA and GSM8K, and +7%-11% on SVAMP and ASDiv. With -self-consistency, we achieve new state-of-the-art results on almost all tasks: despite the fact that self- -consistency is unsupervised and task-agnostic, these results compare favorably to existing approaches -that require task-specific training, or fine-tuning with thousands of examples (e.g., on GSM8K). - -Method - -AddSub MultiArith ASDiv - -AQuA - -SVAMP - -GSM8K - -Previous SoTA 94.9a - -60.5a - -75.3b - -37.9c - -57.4d - -35e / 55g - -UL2-20B - -CoT-prompting -Self-consistency 24.8 (+6.6) - -18.2 - -10.7 -15.0 (+4.3) - -16.9 -21.5 (+4.6) 26.9 (+3.3) - -23.6 - -12.6 -19.4 (+6.8) - -4.1 -7.3 (+3.2) - -LaMDA-137B - -PaLM-540B - -GPT-3 -Code-davinci-001 - -CoT-prompting -51.8 -Self-consistency 63.5 (+10.6) 75.7 (+23.9) 58.2 (+9.2) 26.8 (+9.1) - -49.0 - -52.9 - -17.7 - -38.9 -53.3 (+14.4) 27.7 (+10.6) - -17.1 - -CoT-prompting -Self-consistency 93.7 (+1.8) - -91.9 - -94.7 -99.3 (+4.6) - -74.0 -81.9 (+7.9) 48.3 (+12.5) 86.6 (+7.6) - -79.0 - -35.8 - -56.5 -74.4 (+17.9) - -CoT-prompting -59.5 -Self-consistency 67.8 (+10.6) 82.7 (+23.2) 61.9 (+9.2) 25.6 (+6.7) - -52.7 - -57.2 - -18.9 - -39.8 -54.5 (+14.7) 23.4 (+8.8) - -14.6 - -GPT-3 -Code-davinci-002 - -CoT-prompting -Self-consistency 91.6 (+2.2) - -89.4 - -96.2 -100.0 (+3.8) 87.8 (+7.6) 52.0 (+12.2) 86.8 (+11.0) 78.0 (+17.9) - -80.1 - -75.8 - -60.1 - -39.8 - -Table 2: Arithmetic reasoning accuracy by self-consistency compared to chain-of-thought prompting -(Wei et al., 2022). The previous SoTA baselines are obtained from: a: Relevance and LCA operation -classifier (Roy & Roth, 2015), b: Lan et al. (2021), c: Amini et al. (2019), d: Pi et al. (2022), e: -GPT-3 175B finetuned with 7.5k examples (Cobbe et al., 2021), g: GPT-3 175B finetuned plus an -additional 175B verifier (Cobbe et al., 2021). The best performance for each task is shown in bold. - -Method - -CSQA - -StrategyQA ARC-e - -ARC-c - -Letter (4) Coinflip (4) - -Previous SoTA 91.2a - -73.9b - -86.4c - -75.0c - -N/A - -N/A - -UL2-20B - -51.4 -CoT-prompting -Self-consistency 55.7 (+4.3) 54.9 (+1.6) - -53.3 - -61.6 -69.8 (+8.2) 49.5 (+6.8) - -42.9 - -0.0 -0.0 (+0.0) - -50.4 -50.5 (+0.1) - -LaMDA-137B - -CoT-prompting -57.9 -Self-consistency 63.1 (+5.2) 67.8 (+2.4) - -65.4 - -75.3 -79.3 (+4.0) 59.8 (+4.7) - -55.1 - -8.2 -8.2 (+0.0) - -72.4 -73.5 (+1.1) - -PaLM-540B - -79.0 -CoT-prompting -Self-consistency 80.7 (+1.7) 81.6 (+6.3) - -75.3 - -95.3 -96.4 (+1.1) 88.7 (+3.5) - -85.2 - -65.8 -70.8 (+5.0) 91.2 (+3.0) - -88.2 - -GPT-3 -Code-davinci-001 - -CoT-prompting -46.6 -Self-consistency 54.9 (+8.3) 61.7 (+5.0) - -56.7 - -63.1 -72.1 (+9.0) 53.7 (+10.6) 10.0 (+2.2) 75.9 (+4.5) - -43.1 - -71.4 - -7.8 - -GPT-3 -Code-davinci-002 - -CoT-prompting -79.0 -Self-consistency 81.5 (+2.5) 79.8 (+6.4) - -73.4 - -94.0 -96.0 (+2.0) 87.5 (+3.9) - -83.6 - -70.4 -73.4 (+3.0) 99.5 (+0.5) - -99.0 - -Table 3: Commonsense and symbolic reasoning accuracy by self-consistency compared to chain- -of-thought prompting (Wei et al., 2022). The previous SoTA baselines are obtained from: a: -DeBERTaV3-large + KEAR (Xu et al., 2021b), b: Chowdhery et al. (2022), c: UnifiedQA-FT -(Khashabi et al., 2020). The best performance for each task is shown in bold. - -7The standard deviation of self-consistency is ≤ 0.5 for all tasks and is thus omitted in the table. Please refer - -to Figure 2, Figure 7 and 8 for the standard deviations under varying numbers of sampled paths. - 5 - - Published as a conference paper at ICLR 2023 - -Commonsense and Symbolic Reasoning Table 3 shows the results on commonsense and symbolic -reasoning tasks. Similarly, self-consistency yields large gains across all four language models, and -obtained SoTA results on 5 out of 6 tasks. For symbolic reasoning, we test the out-of-distribution -(OOD) setting where the input prompt contains examples of 2-letters or 2-flips but we test examples -of 4-letters and 4-flips (this setting is more challenging as PaLM-540B or GPT-3 can already achieve -perfect in-distribution accuracy). In this challenging OOD setting, the gain of self-consistency is still -quite significant compared to CoT-prompting with sufficient model sizes. - -To show the effect of the number of sampled reasoning paths, we plot the accuracy (mean and -standard deviation over 10 runs) with respect to varying numbers of sampled paths (1, 5, 10, 20, 40) -in Figure 2. The results show that sampling a higher number (e.g., 40) of reasoning paths leads to a -consistently better performance, further emphasizing the importance of introducing diversity in the -reasoning paths. In Table 4, we show self-consistency yields a richer set of reasoning paths compared -to greedy decoding with a few example questions from two tasks. - -Figure 2: Self-consistency (blue) significantly improves accuracy over CoT-prompting with greedy -decoding (orange) across arithmetic and commonsense reasoning tasks, over LaMDA-137B. Sampling -a higher number of diverse reasoning paths consistently improves reasoning accuracy. - -GSM8K - -StrategyQA - -Question Henry made two stops during his 60-mile bike trip. He first stopped after 20 -miles. His second stop was 15 miles before the end of the trip. How many miles -did he travel between his first and second stops? - -Yes or no: Is Albany, Georgia the most populous -US Albany? - -Greedy -Decode - -He traveled 60 miles in total. He stopped after 20 miles, so he traveled 60 - 20 -= 40 miles between the first and second stops. The answer is 40. - -Albany, Georgia is the most populous US Al- -bany. Thus, the answer is yes. - -Sampled -Path 1 - -Sampled -Path 2 - -Henry travelled a total of 60 miles. His first stop was 20 miles in, and his -second stop was 15 miles before the end. So between his first and second stops -he travelled 60 - 20 -15 = 25 miles. The answer is 25. - -The most populous US Albany is Albany, New -York. Thus, Albany, Georgia is not the most -populous US Albany. So the answer is no. - -He made two stops during a 60-mile trip. The first was 20 miles into the trip. -The second was 15 miles before the end of the trip. This means the second stop -was 60 - 15 = 45 miles into the trip. Since he made the stops in order, the second -stop must have been 45 - 20 = 25 miles after the first stop. The answer is 25. - -Albany, Georgia has a population of about -88,000. Albany, New York has a population of -about 95,000. Thus, Albany, Georgia is not the -most populous US Albany. So the answer is no. - -Table 4: Examples where self-consistency helps repair the errors over greedy decode, on PaLM-540B. -Two sampled reasoning paths that are consistent with the ground truth are shown. - -3.3 SELF-CONSISTENCY HELPS WHEN CHAIN-OF-THOUGHT HURTS PERFORMANCE - -Ye & Durrett (2022) show that sometimes chain-of-thought prompting could hurt performance -compared to standard prompting in few-shot in-context learning. Here we perform a study using -self-consistency to see if it can help fill in the gap, over a set of common NLP tasks, including (1) -Closed-Book Question Answering: BoolQ (Clark et al., 2019), HotpotQA (Yang et al., 2018), and -(2) Natural Language Inference: e-SNLI (Camburu et al., 2018), ANLI (Nie et al., 2020) and RTE -(Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Bentivogli et al., 2009). - -The results over PaLM-540B are shown in Table 5. For some tasks (e.g., ANLI-R1, e-SNLI, RTE), -adding chain-of-thought does hurt performance compared to standard prompting (Brown et al., 2020), -but self-consistency is able to robustly boost the performance and outperform standard prompting, -making it a reliable way to add rationales in few-shot in-context learning for common NLP tasks. - -ANLI R1 / R2 / R3 e-SNLI RTE BoolQ HotpotQA (EM/F1) - -Standard-prompting (no-rationale) -CoT-prompting (Wei et al., 2022) - -69.1 / 55.8 / 55.8 -68.8 / 58.9 / 60.6 - -Self-consistency - -78.5 / 64.5 / 63.4 - -85.8 -81.0 - -88.4 - -84.8 -79.1 - -86.3 - -71.3 -74.2 - -78.4 - -27.1 / 36.8 -28.9 / 39.8 - -33.8 / 44.6 - -Table 5: Compare Standard/CoT prompting with self-consistency on common NLP tasks. - 6 +Total -0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths56586062Commonsense QA0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -3.4 COMPARE TO OTHER EXISTING APPROACHES - -We conduct a set of additional studies and show that self-consistency significantly outperforms -existing methods including sample-and-rank, beam search, and ensemble-based approaches. - -Comparison to Sample-and-Rank One commonly used approach to improve generation quality is -sample-and-rank, where multiple sequences are sampled from the decoder and then ranked according -to each sequence’s log probability (Adiwardana et al., 2020). We compare self-consistency with -sample-and-rank on GPT-3 code-davinci-001, by sampling the same number of sequences from the -decoder as self-consistency and taking the final answer from the top-ranked sequence. The results are -shown in Figure 3. While sample-and-rank does improve the accuracy with additionally sampled -sequences and ranking, the gain is much smaller compared to self-consistency. - -Figure 3: Self-consistency significantly outperforms sample-and-rank with the same # of samples. - -Comparison to Beam Search In Table 6, we compare self-consistency with beam search decoding -on the UL2-20B model. For a fair comparison we report the accuracy under the same number of -beams and reasoning paths. On both tasks self-consistency outperforms beam search significantly. -Note self-consistency can also adopt beam search to decode each reasoning path (results are shown -as “Self-consistency using beam search”), but its performance is worse compared to self-consistency -with sampling. The reason is that beam search yields a lower diversity in the outputs (Li & Jurafsky, -2016), while in self-consistency the diversity of the reasoning paths is the key to a better performance. - -Beam size / Self-consistency paths - -1 - -AQuA +# Unique DAGs +2 out of 2 +6 out of 23 +31 out of 26 +302 out of 210 +5,984 out of 215 +6,325 -MultiArith +# Edges/DAG # MECs -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -23.6 -23.6 - -10.7 -10.7 - -19.8 ± 0.3 21.2 ± 0.7 24.6 ± 0.4 24.2 ± 0.5 -19.7 ± 2.5 24.9 ± 2.6 25.3 ± 1.8 26.7 ± 1.0 26.9 ± 0.5 - -12.0 - -11.3 - -11.0 - -10.5 - -11.8 ± 0.0 11.4 ± 0.1 12.3 ± 0.1 10.8 ± 0.1 -9.5 ± 1.2 11.3 ± 1.2 12.3 ± 0.8 13.7 ± 0.9 14.7 ± 0.3 +0.50 +1.67 +3.48 +5.89 +8.77 +8.60 +2 5 - -19.3 - -10 - -16.1 - 20 +142 +2,207 +2,376 + +# DAGs/MEC +1.0 +1.2 +1.55 +2.13 +2.71 +2.66 + +Table 1: Statistics about the source causal graphs in our dataset. Given the number of nodes, we report +the number of unique DAGs, average number of edges per DAG, number of MECs, and average +number of DAGs per MEC. + +isomorphism check (McKay & Piperno, 2014), and reduce the set so that only unique DAGs are +retained, and we show their statistics in Table 1. Although we can handle large graphs, we mostly +focus on smaller graphs that can still lead to a reasonably sized dataset, so we empirically set N = 6, +but future work can use our open-sourced codes to extend to more nodes. + +3.4 PROGRAMMATICALLY GENERATING THE D-SEPARATION SETS +Based on the set of unique DAGs, we then programmatically generate the d-separation sets by +graph theoretical conditions, as in Step 3 of Figure 2. To realize this step, we code an efficient +graph-theoretic algorithm to check for all the chain, fork, and collider structures to automatically +identify the set of nodes that d-separate each pair of nodes. Using the d-separation sets and the +faithfulness assumption, we form the statistical correlations as follows. For each pair of nodes, they +are conditionally independent given the variables in the d-separation set. If the d-separation set is +empty, then the two nodes are unconditionally independent. If no d-separation set can be found for +the two nodes, then they are directly correlated. + +Moreover, using the d-separation sets, we are able to cluster causal graphs to MECs. We achieve it by +tracing the mapping between the causal graphs and the set of statistical correlations, and backtracking +the graphs with the same d-separation sets to group them in the same MEC. We show in Table 1 that +each MEC contains on average 2.66 DAGs. + +3.5 COMPOSING THE HYPOTHESES AND LABEL +After generating the set of correlations based on the d-separation sets, we now generate the causal +hypotheses. For the causal relation r, we focus on six common causal relations between two nodes +introduced in Section 2.1: Is-Parent, Is-Child, Is-Ancestor (excluding the parents), Is-Descendant +(excluding the children), Has-Confounder (i.e., there exists a confounder, or common cause, of the +two nodes), and Has-Collider (i.e., there exists a collider, or common effect, of the two nodes). In this +way, the set of hypotheses contains all six meaningful causal relations between every pair of variables, +resulting in a total size of 6 · N (N − 1)/2 = 3N (N − 1) hypotheses for a graph with N variables. + +To generate the ground-truth validity label, we start from the correlation sets in Step 3, then look up +all the causal graphs in the same MEC corresponding to the given set of correlations, and check the +necessity of the hypothesized causal relation. If the causal relationship proposed in the hypothesis +is valid for all causal graphs within the MEC, then we generate the validity v = 1; otherwise, we +generate v = 0. A special case of valid samples is that when the size of the MEC is 1, then there is a +bijective mapping between the causal graph and the d-separation sets, so any hypothesis stating the +causal properties of that unique causal graph is valid. + +3.6 VERBALIZING INTO LANGUAGE +Finally, as in the last step of Figure 2, we convert all the information above to text data for our +CORR2CAUSE task. For the correlation statement, we verbalize the set of correlations in Step 3 into +a natural language statement s. When two variables cannot be d-separated, i.e., A ̸⊥⊥ B, then we +describe them as “A correlates with B” since they are directly correlated and cannot be independent +by any condition. And if two variables have a valid d-separation set C, then we describe them as “A +is independent of B given C.” In the special case when the d-separation set is empty, we directly +say “A is independent of B.” In addition, we disambiguate the setting by starting the correlation +statement with the setup of a closed system of the given variables, and no hidden variables: “Suppose +there is a closed system of N variables, A, B, . . . All the statistical relations among these N variables +are as follows:”. Finally, to verbalize the hypothesis, we feed the causal relation triplet (Xi, r, Xj) -15.0 +5 -40 + Published as a conference paper at ICLR 2024 + +Causal Relation +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +Hypothesis Template +{Var i} directly causes {Var j}. +{Var i} causes something else which causes {Var j}. +{Var j} directly causes {Var i}. +{Var j} is a cause for {Var i}, but not a direct one. +There exists at least one collider (i.e., common effect) of {Var i} and {Var j}. +There exists at least one confounder (i.e., common cause) of {Var i} and {Var j}. + +Table 2: Templates for each causal relation in the hypothesis. We use {Var i} and {Var j} as +placeholders for the two variables. + +into their hypothesis templates in Table 2. For example, we turn the triplet (A, Is-Parent, B) into “A +directly causes B”, as in the example of Figure 2. + +3.7 STATISTICS OF THE RESULTING DATA +We show the statistics of our CORR2CAUSE dataset in Table 3. Overall, our dataset contains 207,972 +samples, where 18.57% of the samples have the positive label (i.e., with validity=1). The average +length of the premise is 424.11 tokens, and hypothesis 10.83 tokens. We split the data into 205,734 +training samples, 1,076 development and 1,162 test samples.2 Since the main purpose of the dataset +is to benchmark the performance of LLMs, we prioritize the test and development sets to have a +comprehensive coverage over all sizes of graphs. Specifically, we iterate through the subset of our +data for each N , and split it entirely for only the test and development sets if the data is less than 1K, +which is the case for N = 2 and 3. For the other subsets that are larger, we randomly sample up to +1K or 10% of the data, whichever is smaller, to the test and development sets. We set the cap to be 1K +in order to form a reasonable computation budget, since many LLMs are expensive to query in the +inference mode. Aside from the test and valid sets, all the rest of the data goes into the training set. + +# Samples +# Test +# Dev +# Train + +# Tokens/Premise +# Tokens/Hypothesis +% Positive Labels +Vocab Size + +Overall + +207,972 +1,162 +1,076 +205,734 +424.11 +10.83 +18.57 +65 + +Statistics by the Number of Nodes N +N = 2 N = 3 N = 4 N = 5 N = 6 +198,630 +720 +522 +72 +474 +72 +197,634 +576 +434.54 +104.0 +10.83 +10.83 +18.85 +7.50 +61 +55 + +8,520 +514 +482 +7,524 +212.61 +10.83 +13.01 +57 + +90 +48 +42 +0 +52.0 +10.83 +3.33 +53 -10.2 +12 +6 +6 +0 +31.5 +10.83 +0.00 +49 + +Table 3: Statistics of our CORR2CAUSE dataset, and by subsets. We report the total number of +samples (# Samples); splits of the test (# Test), developement (# Dev) and training sets (# Train); +number of tokens per premise (# Tokens/Premise) and hypothesis (# Tokens/Hypothesis); percentage +of the positive labels (% Positive Labels), and vocabulary size by the number of unique tokens (Vocab +Size). Note that the number of unique graphs and MECs are in Table 1. +4 EXPERIMENTS + +4.1 EXPERIMENTAL SETUP +We set up a diverse list of LLMs for the experiments on our CORR2CAUSE dataset. To test existing +LLMs, we first include six commonly used BERT-based NLI models in the transformers library (Wolf +et al., 2020): BERT (Devlin et al., 2019), RoBERTa (Liu et al., 2019), BART (Lewis et al., 2020), +DeBERTa (He et al., 2021), DistilBERT (Sanh et al., 2019), and DistilBART (Shleifer & Rush, 2020). +Apart from these BERT-based NLI models, we also evaluate the general-purpose autoregressive +LLMs based on GPT (Radford et al., 2019): GPT-3 Ada, Babbage, Curie, Davinci (Brown et al., +2020); its instruction-tuned versions (Ouyang et al., 2022), text-davinci-001, text-davinci-002, and +text-davinci-003; and GPT-3.5 (i.e., ChatGPT), and the latest GPT-4 (OpenAI, 2023) by April 2023, + +2Note for our dataset v2.0: We notice that our original data (v1.0) has duplication due to symmetric relations +and verbalizations of the hypothesis. E.g., Is-Parent(A, B) has the exact hypothesis verbalization as Is-Child(B, +A). Hence, for our data v2.0, we perform a careful de-duplication, and update the data statistics in Table 3. See +more version comparison details in Appendix D. Note that, due to the symmetry, the current version is a random +sample half of the size of the original version, so the modeling results in the experiment section roughly hold. -Table 6: Compare self-consistency with beam search decoding on the UL2-20B model. +6 -Comparison to Ensemble-based Approaches We further compare self-consistency to ensemble- -based methods for few-shot learning. In particular, we consider ensembling by: (1) prompt order -permutation: we randomly permute the exemplars in the prompt 40 times to mitigate model’s -sensitivity to prompt order (Zhao et al., 2021; Lu et al., 2021); and (2) multiple sets of prompts -(Gao et al., 2021): we manually write 3 different sets of prompts. We took majority vote of the -answers from greedy decoding in both approaches as an ensemble. Table 7 shows that compared to -self-consistency, existing ensemble-based approaches achieve a much smaller gain.8 In addition, note -that self-consistency is different from a typical model-ensemble approach, where multiple models -are trained and their outputs are aggregated. Self-consistency acts more like a “self-ensemble” on -top of a single language model. We additionally show the results of ensembling multiple models in -Appendix A.1.3 where the model-ensembles perform much worse compared to self-consistency. + Published as a conference paper at ICLR 2024 -GSM8K MultiArith +Random Baselines +Always Majority +Random (Proportional) +Random (Uniform) +BERT-Based Models -SVAMP +BERT MNLI +RoBERTa MNLI +DeBERTa MNLI +DistilBERT MNLI +DistilBART MNLI +BART MNLI -ARC-e +LLaMa-Based Models -ARC-c +LLaMa-7B +Alpaca-7B -CoT (Wei et al., 2022) -Ensemble (3 sets of prompts) -Ensemble (40 prompt permutations) -Self-Consistency (40 sampled paths) +GPT-Based Models -17.1 -18.6 ± 0.5 -19.2 ± 0.1 -27.7 ± 0.2 +GPT-3 Ada +GPT-3 Babbage +GPT-3 Curie +GPT-3 Davinci +GPT-3 Instruct (text-davinci-001) +GPT-3 Instruct (text-davinci-002) +GPT-3 Instruct (text-davinci-003) +GPT-3.5 +GPT-4 -51.8 -57.1 ± 0.7 -60.9 ± 0.2 -75.7 ± 0.3 +F1 -38.9 -42.1 ± 0.6 -42.7 ± 0.1 -53.3 ± 0.2 +Precision -75.3 -76.6 ± 0.1 -76.9 ± 0.1 -79.3 ± 0.3 +Recall Accuracy -55.1 -57.0 ± 0.2 -57.0 ± 0.1 -59.8 ± 0.2 +0.0 +13.5 +20.38 + +2.82 +22.79 +14.52 +20.70 +26.74 +33.38 + +26.81 +27.37 + +0.00 +27.45 +26.43 +27.82 +17.99 +21.87 +15.72 +21.69 +29.08 -Table 7: Self-consistency outperforms prompt-order and multi-prompt ensembles on LaMDA-137B. +0.0 +12.53 +15.11 + +7.23 +34.73 +14.71 +24.12 +15.92 +31.59 + +15.50 +15.93 + +0.00 +15.96 +15.23 +16.57 +11.84 +13.46 +13.4 +17.79 +20.92 -8Self-consistency is compatible with both ensemble approaches and we show the results in Appendix A.1.4. +0.0 +14.62 +31.29 + +1.75 +16.96 +14.33 +18.13 +83.63 +35.38 + +99.42 +97.37 + +0.00 +97.95 +100.00 +86.55 +37.43 +58.19 +19.01 +27.78 +47.66 + +84.77 +71.46 +62.78 + +81.61 +82.50 +74.31 +78.85 +30.23 +78.50 + +17.36 +21.33 + +84.77 +21.15 +15.23 +31.61 +48.04 +36.69 +68.97 +69.46 +64.60 + +Table 4: Overall performance. We report F1 (main metric), precision, recall and accuracy. For the +main metric, F1 score, we use the bold font to highlight the overall best performance, and underline +to highlight the best performance within each category of models. + +using the OpenAI API (https://openai.com/api/) with temperature 0. We also evaluate the +recent, more efficient models, LLaMa (Touvron et al., 2023) and Alpaca (Taori et al., 2023). + +When inspecting the behavior of finetuned models, we adopt a large set of models, including GPT- +based models (GPT-3 Ada, Babbage, Curie, and Davinci) using the OpenAI finetuning API for +classification at https://platform.openai.com/docs/guides/fine-tuning, open-sourced +decoder-only models (GPT2, GPT2-Large, GPT2-XL, LLaMA-7B, and LLaMA2-7B), BERT-based +models from scratch (BERT-Base, BERT-Large, RoBERTa-Base, and RoBERTa-Large), and BERT- +Based NLI models (BERT-Base MNLI, BERT-Large MNLI, RoBERTa-Base MNLI, and RoBERTa- +Large MNLI) using the transformers library (Wolf et al., 2020). See training details in Appendix A. + +For the random baselines, we provide “always majority” to predict the majority class 100% of the time, +“random (uniform)” to uniformly sample a label (i.e., 50% for each), and “random (proportional)” to +sample a label from a Bernouli distribution proportional to the development set label distribution. + +4.2 THE CORR2CAUSE SKILL IN EXISTING LLMS + +We show the performance of seventeen LLMs in Table 4. We can see that pure causal inference is a +very challenging task across all existing LLMs. Among all the LLMs, the best performance is 33.38% +F1 by BART MNLI, which is even higher than the latest GPT-based model, GPT-4. Notably, many +models are worse than random guess, which means that they totally fail at this pure causal inference +task. The observation still holds for few-shot chain-of-thought prompts tested in Appendix G. + +4.3 FINETUNED PERFORMANCE + +Next, we address the question: Can we re-purpose LLMs to learn this task? The experimental results +in Table 5a of 17 models finetuned on our CORR2CAUSE seem very strong at first sight. Most models +see a substantial increase, among which the finetuned BERT-based NLI models demonstrate the +strongest performance. The best-performing one, RoBERTa-Large MNLI, achieves 94.74% F1 score +on this task, as well as very high precision, recall and accuracy scores. 7 -0510152025303540#Sampled Reasoning Paths12141618202224Accuracy (%)GSM8K0510152025303540#Sampled Reasoning Paths50556065707580Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths303540455055Accuracy (%)ARC (Challenge)Self Consistency (Multi-path)Sample & Rank (Multi-path)Greedy Decode (Single-path) Published as a conference paper at ICLR 2023 - -3.5 ADDITIONAL STUDIES - -We conducted a number of additional experiments to analyze different aspects of the self-consistency -method, including its robustness to sampling strategies and parameters, and how it works with -imperfect prompts and non-natural-language reasoning paths. - -Self-Consistency is Robust to Sampling Strategies and Scaling We show self-consistency is -robust to sampling strategies and parameters, by varying T in temperature sampling (Ackley et al., -1985; Ficler & Goldberg, 2017), k in top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford -et al., 2019), and p in nucleus sampling (Holtzman et al., 2020), over PaLM-540B in Figure 4 (left). -Figure 4 (right) shows that self-consistency robustly improves performance across all scales for the -LaMDA-137B model series. The gain is relatively lower for smaller models due to certain abilities -(e.g., arithmetic) only emerge when the model reaches a sufficient scale (Brown et al., 2020). - -Figure 4: GSM8K accuracy. (Left) Self-consistency is robust to various sampling strategies and -parameters. (Right) Self-consistency improves performance across language model scales. - -Self-Consistency Improves Robustness to Imperfect Prompts For few-shot learning with man- -ually constructed prompts, human annotators sometimes make minor mistakes when creating the -prompts. We further study if self-consistency can help improve a language model’s robustness to -imperfect prompts.9 We show the results in Table 8: while imperfect prompts decrease accuracy with -greedy decoding (17.1 → 14.9), self-consistency can fill in the gaps and robustly improve the results. - -Additionally, we found that the consistency (in terms of % of decodes agreeing with the final -aggregated answer) is highly correlated with accuracy (Figure 5, over GSM8K). This suggests that -one can use self-consistency to provide an uncertainty estimate of the model in its generated solutions. -In other words, one can use low consistency as an indicator that the model has low confidence; i.e., -self-consistency confers some ability for the model to “know when it doesn’t know”. - -Prompt with correct chain-of-thought - -17.1 - -LaMDA-137B - -Prompt with imperfect chain-of-thought 14.9 -23.4 - -+ Self-consistency (40 paths) - -Prompt with equations - -+ Self-consistency (40 paths) - -PaLM-540B - -Zero-shot CoT (Kojima et al., 2022) - -+ Self-consistency (40 paths) - -5.0 -6.5 - -43.0 -69.2 - -Table 8: Self-consistency works under imperfect prompts, equa- -tion prompts and zero-shot chain-of-thought for GSM8K. - -Figure 5: The consistency is cor- -related with model’s accuracy. - -Self-Consistency Works for Non-Natural-Language Reasoning Paths and Zero-shot CoT We -also tested the generality of the self-consistency concept to alternative forms of intermediate reasoning -like equations (e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there -are 3 + 2 = 5 cars.” to “3 + 2 = 5”). The results are shown in Table 8 (“Prompt with equations”): -self-consistency still improves accuracy by generating intermediate equations; however, compared to -generating natural language reasoning paths, the gain is smaller since the equations are much shorter -and less opportunity remains for generating diversity in the decoding process. In addition, we tested -self-consistency with zero-shot chain-of-thought (Kojima et al., 2022) and show that self-consistency -works for zero-shot CoT as well and improves the results significantly (+26.2%) in Table 8. - -9We use the same prompts as before, but swap all the numbers in the reasoning paths with random numbers -except the final answer, e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars.” to “There are 7 cars in the parking lot already. 6 more arrive. Now there are 7 + 6 = 5 cars.”. + Published as a conference paper at ICLR 2024 + +F1 + +Precison Recall Accuracy + +F1 (Paraph.) + +F1 (Var. Ref.) + +92.11 +88.60 +88.60 +91.52 + +79.85 +78.19 +81.23 +85.52 + +70.47 +69.98 +75.00 +80.26 + +Finetuned GPT-Based Models Using OpenAI API +GPT-3 Ada +GPT-3 Babbage +GPT-3 Curie +GPT-3 Davinci +Finetuned Open-Sourced Decoder-Only Models +89.18 +GPT2 +94.29 +GPT2-Large +GPT2-XL +94.30 +91.98 +LLaMA-7B +LLaMA2-7B +92.92 +Finetuned BERT-Based Models +69.29 +BERT-Base +85.26 +BERT-Large +87.60 +RoBERTa-Base +RoBERTa-Large +89.10 +Finetuned BERT-Based NLI Models +BERT-Base MNLI +BERT-Large MNLI +RoBERTa-Base MNLI +RoBERTa-Large MNLI + +88.03 +92.18 +91.94 +88.62 +90.11 + +54.42 +77.51 +78.47 +82.54 + +85.49 +84.44 +90.35 +92.24 + +89.88 +90.19 +94.27 +94.74 + +90.35 +96.49 +96.78 +95.61 +95.91 + +95.32 +94.74 +99.12 +96.78 + +94.74 +96.78 +98.54 +97.37 + +92.92 +92.48 +93.77 +95.28 + +96.66 +98.22 +98.22 +97.46 +97.77 + +87.13 +95.01 +95.73 +96.39 + +86.51 +96.79 +98.17 +98.35 + +61.73 +62.34 +64.93 +65.01 + +56.76 +55.95 +60.32 +56.41 +52.24 + +61.13 +63.64 +65.58 +65.05 + +65.56 +67.24 +57.42 +55.45 + +41.57 +43.28 +45.32 +46.96 + +31.70 +31.99 +43.95 +53.92 +49.47 + +35.20 +38.54 +53.12 +60.20 + +31.50 +52.04 +62.83 +67.87 + +(a) Performance of finetuned models on the original test set. + +(b) F1 scores of finetuned models +on the perturbed test sets by para- +phrasing (Paraph.) and variable +refactorization (Var. Ref.). +Table 5: Performance of finetuned models on the original test set and perturbed test sets. + +Relation Type +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +F1 +96.18 +93.94 +95.73 +96.55 +92.19 +98.67 + +Precision Recall Accuracy +96.92 +93.94 +96.56 +100 +97.52 +100 + +98.67 +98.93 +98.67 +99.47 +94.64 +99.73 + +95.45 +93.94 +94.92 +93.33 +87.41 +97.37 + +F1 +74.80 +45.45 +73.39 +29.41 +70.70 +70.42 + +Precision Recall Accuracy +70.77 +30.30 +68.97 +17.86 +66.90 +67.57 + +79.31 +90.91 +78.43 +83.33 +75.00 +73.53 + +91.73 +93.60 +92.27 +93.60 +82.04 +94.37 + +(a) Fine-grained performance of RoBERTa-Large by causal relation +type on the original test set. + +(b) Its fine-grained performance by relation +type after variable refactorization. + +Table 6: Fine-grained analysis of the best-performing model, RoBERTa-Large MNLI. + +4.4 FINE-GRAINED PERFORMANCE BY CAUSAL RELATION +In addition to the overall results mentioned above, we conduct a fine-grained analysis to check the +performance of the strongest finetuned model, RoBERTa-Large MNLI, by our six causal relation +types. As in Table 6a, the model is very good at judging relations such as Is-Parent, Is-Descendant +and Has-Confounder, all with more than 96% F1 scores, whereas it is several points weaker on +the Has-Collider relations. This could be due to that the collider relation is the most special type, +requiring identification of the V-structure based on both the unconditional independence based on the +two variables only and correlations whenever conditioned on a common descendant. We also conduct +error analysis for non-finetuned models in Appendix F. + +4.5 ROBUSTNESS ANALYSIS +Looking at the very high performance of the finetuned models, we raise the next question: Did the +models really robustly learn the causal inference skills? + +Two Robustness Tests We design two simple robustness tests: (1) paraphrasing, and (2) variable +refactorization. For (1) paraphrasing, we simply paraphrase the hypothesis by changing the text +template for each causal relation to some semantically-equivalent alternatives in Appendix C. For (2) +variable refactorization, we reverse the alphabet of the variable names, namely flipping A, B, C, to Z, +Y, X and so on. The inspiration behind the two robustness tests comes from the spurious correlation +analysis described in Appendix E. 8 -0510152025303540#Sampled Reasoning Paths444852566064687276Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.7, k=20T=0.7, no top kp=0.95p=0.9Greedy Decode125102050100200Model size (#param in billions)510152025Accuracy (%)Self ConsistencyGreedy Decode020406080100Consistency (%)020406080100Accuracy (%) Published as a conference paper at ICLR 2023 - -4 RELATED WORK - -Reasoning in language models. Language models are known to struggle in Type 2 tasks, such as -arithmetic, logical and commonsense reasoning (Evans, 2010). Previous work has primarily focused -on specialized approaches for improving reasoning (Andor et al., 2019; Ran et al., 2019; Geva et al., -2020; Pi˛ekos et al., 2021). Compared to prior work, self-consistency is applicable to a wide range of -reasoning tasks without any additional supervision or fine-tuning, while still substantially improving -the performance of the chain-of-thought prompting approach proposed in Wei et al. (2022). -Sampling and re-ranking in language models. Multiple decoding strategies for language models -have been proposed in the literature, e.g., temperature sampling (Ackley et al., 1985; Ficler & -Goldberg, 2017), top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford et al., 2019), -nucleus sampling (Holtzman et al., 2020), minimum Bayes risk decoding (Eikema & Aziz, 2020; Shi -et al., 2022), and typical decoding (Meister et al., 2022). Other work has sought to explicitly promote -diversity in the decoding process (Batra et al., 2012; Li et al., 2016; Vijayakumar et al., 2018). - -Re-ranking is another common approach to improve generation quality in language models (Adiwar- -dana et al., 2020; Shen et al., 2021). Thoppilan et al. (2022) collect additional human annotations -to train a re-ranker for response filtering. Cobbe et al. (2021) train a “verifier” to re-rank generated -solutions, which substantially improves the solve rate on math tasks compared to just fine-tuning the -language model. Elazar et al. (2021) improve the consistency of factual knowledge extraction by -extending pre-training with an additional consistency loss. All these methods require either training -an additional re-ranker or collecting additional human annotation, while self-consistency requires no -additional training, fine-tuning, nor extra data collection. -Extract reasoning paths. Some previous work has considered task-specific approaches for iden- -tifying reasoning paths, such as constructing semantic graphs (Xu et al., 2021a), learning an RNN -to retrieve reasoning paths over the Wikipedia graph (Asai et al., 2020), fine-tuning with human -annotated reasoning paths on math problems (Cobbe et al., 2021), or training an extractor with -heuristic-based pseudo reasoning paths (Chen et al., 2019). More recently, the importance of di- -versity in the reasoning processes has been noticed, but only leveraged via task-specific training, -either through an additional QA model over extracted reasoning paths (Chen et al., 2019), or by the -introduction of latent variables in a commonsense knowledge graph (Yu et al., 2022). Compared to -these approaches, self-consistency is far simpler and requires no additional training. The approach -we propose simply couples the generation of reasoning paths and a final answer by sampling from -the decoder, using aggregation to recover the most consistent answer without additional modules. -Consistency in language models. Some prior work has shown that language models can suffer -from inconsistency in conversation (Adiwardana et al., 2020), explanation generation (Camburu et al., -2020), and factual knowledge extraction (Elazar et al., 2021). Welleck et al. (2020) use “consistency” -to refer to generating an infinite-length sequence in recurrent language models. Nye et al. (2021) -improve the logical consistency of samples from a System 1 model by adding a System 2-inspired -logical reasoning module. In this paper we focus on a slightly different notion of “consistency”, i.e., -utilizing answer consistency among diverse reasoning paths to improve accuracy. - -5 CONCLUSION AND DISCUSSION - -We introduced a simple yet effective method called self-consistency, and observed that it significantly -improves accuracy in a range of arithmetic and commonsense reasoning tasks, across four large -language models with varying scales. Beyond accuracy gains, self-consistency is also useful for -collecting rationales when performing reasoning tasks with language models, and for providing -uncertainty estimates and improved calibration of language model outputs. - -One limitation of self-consistency is that it incurs more computation cost. In practice people can try a -small number of paths (e.g., 5 or 10) as a starting point to realize most of the gains while not incurring -too much cost, as in most cases the performance saturates quickly (Figure 2). As part of future work, -one could use self-consistency to generate better supervised data to fine-tune the model, such that the -model can give more accurate predictions in a single inference run after fine-tuning. In addition, we -observed that language models can sometimes generate incorrect or nonsensical reasoning paths (e.g., -the StrategyQA example in Table 4, the two population numbers are not exactly correct), and further -work is needed to better ground models’ rationale generations. + Published as a conference paper at ICLR 2024 + +Specifically, we adopt the common setup of text adversarial attack (Morris et al., 2020; Jin et al., +2020) to preserve the training set and keep the same saved models, but run the inference on the +perturbed test set. In this way, we separate the possibilities of the models only overfitting on the +training data vs. mastering the reasoning skills. + +Results after Perturbation We can see from Table 5b that all the models drop drastically, by up to +39.29 on the paraphrased test set, and up to 62.30 after variable refactorization. The best-performing +model, RoBERTa-Large MNLI, is especially sensitive towards paraphrasing, demonstrating the +most drop among all models; however, it is the most robust against the variable refactorization, +maintaining a high F1 score of 67.87. We conduct fine-grained analysis for RoBERTa-Large MNLI +under perturbation in Table 6b. We can see the the main source of the performance drop of the model +comes from the two classes, Is-Ancestor (decreasing to 45.45%) and Is-Descendant (decreasing to +29.41%), while the other classes stay relatively robust, keeping their F1 scores over 70%. + +From this analysis, we make the following suggestions to future studies testing this CORR2CAUSE +skill of LLMs. First, it is safe to use it as a test set to benchmark existing LLMs’ performance, since +the data we generate is out-of-distribution from the training data of the current LLMs. Then, when +testing finetuned models, it is very important to accompany adversarial attack together with the i.i.d. +test set. We open-source our perturbed test sets for future work to test the generalizability skill. + +4.6 EXTENSION TO NATURAL STORIES +We envision our CORR2CAUSE dataset to be a foundation for future extensions to various settings, +such as instantiating the variables with actual phenomena and situating the story in a more natural +setting. For example, the correlation does not imply causation rule can be instantiated with the ice +cream sales and swimming pool attendance as the two variables, and argue that ice cream sales does +not necessarily affect swimming pool attendance, because their correlation could be due to a third +variable, such as hot weather. We provide a case study for how to instantiate the symbolic expressions +in our dataset to more natural stories, and find that LLMs such as GPT-4 can generate realistic, daily +life stories that has foreseeably broad applications. See more details in Appendix B. + +5 RELATED WORK + +Existing Causal Reasoning Tasks A large body of existing research of causal reasoning in NLP +focuses on leveraging empirical knowledge to do tasks such as inferring the cause and effect of why +an agent perform certain tasks (Sap et al., 2019a), the motivation and emotional reaction in a social +context (Sap et al., 2019b), how people achieve a given goal with a set of concrete steps (Zhang et al., +2020), the development of a story given a different beginning (Qin et al., 2019), and how in general +LLMs serve as a knowledge base of cause and effect (Willig et al., 2023; Kıcıman et al., 2023). In +contrast, our CORR2CAUSE task focuses on the pure causal inference skill of models, which is a +knowledge-dependent reasoning skill based on formally correct rules from causal inference. + +Existing Logical and Inference Tasks Another related area of literature is logical and inference +tasks, of which a well-established one is natural language inference (NLI), to identify the semantic +relationship between a pair of sentences (MacCartney & Manning, 2008; Bowman et al., 2015). NLI +datasets mainly focus on the set and paraphrase relations. For example, “a group of boys are playing +football” can entail “some guys are playing football,” where “boys” are a sub-concept of “guys,” and +“a group of” and “some” are paraphrases. Recently, there have been increasing efforts to extend the +inference task to various logical inference skills such as deductive logic and propaganda techniques +(Jin et al., 2022; Alhindi et al., 2022). Our CORR2CAUSE dataset is the first dataset testing the +correlation-to-causation inference skill, which is unique of its type. + +6 CONCLUSION + +In this work, we introduced a novel task, CORR2CAUSE, to infer causation from correlation, and +collected a large-scale dataset of over 200K samples. We evaluated an extensive list of LLMs on this +new task, and showed that off-the-shelf LLMs perform poorly on this task. We also show that it is +possible to re-purpose LLMs on this task by finetuning, but future work needs to be aware of the +out-of-distribution generalization problem. To avoid the Goodhart’s law, we recommend using this +dataset to benchmark the pure causal inference skills for LLMs that have not seen this dataset. Given +the limited reasoning abilities of current LLMs, and the difficulty of separating actual reasoning from +training-corpus-derived knowledge, it is imperative that our community focus on work aiming to +accurately disentangle and measure both abilities. We believe the present work is a first such step. 9 - Published as a conference paper at ICLR 2023 + Published as a conference paper at ICLR 2024 -REPRODUCIBILITY STATEMENT +LIMITATIONS AND FUTURE WORK -In experiments, we included four different language models with varying scales. Two of them are pub- -lic models: UL2 is a completely open-sourced model with model checkpoints available at https:// -github.com/google-research/google-research/tree/master/ul2; GPT-3 is -also a public model with public API available at https://openai.com/api/. For GPT-3, -we have included two public engines (“code-davinci-001” and “code-davinci-002”) to further aid -reproducibility, as Codex is currently free so anyone can reproduce the results. In addition, as our -results make use of LaMDA-137B and PaLM-540B that are not publicly available, we provide the -exact input prompts for all tasks in Appendix A.3 (and note that we do not perform any finetuning -and only apply prompting to off-the-shelf language models). +We identify several limitations of this work and open future directions: First, in the context of this +work, we limit the causal graphs to two to six nodes, but future work can feel free to explore larger +graphs. Another aspect is that we do not assume hidden confounders in this inference problem, so we +welcome future work to generate an even more challenging dataset to infer the existence of hidden +confounders, analogous to the causal discovery algorithm of fast causal inference (FCI) (Spirtes et al., +2000). And also in general, explorations of other causal discovery algorithms are welcomed too. +Finally, a lot of motivation behind proposing this task is inspired by the problem of invalid reasoning +patterns in our daily reasoning (Jin et al., 2022), which could fertilize the ground for more pervasive +spread of fake news. We believe false causal inference is a prevalent type of fallacious beliefs, and +welcome future work to connect the idea of this benchmark to more real-world false beliefs based on +confusing correlation with causation. -ETHICS STATEMENT +ACKNOWLEDGMENT -As we stated in the discussion, language models can sometimes generate nonsensical or non-factual -reasoning paths, so one should use language models’ outputs with extra caution. We deal with -reasoning tasks mostly and the generated rationales are only used for inspecting how a model reaches -its answer. One could potentially use the generated rationales to further check why the model makes -certain mistakes or whether the model contains any biases when performing a certain task. For -language model in real-world use, further work is needed to better ground models’ predictions and -improve model’s factuality and safety, to ensure the models do not cause harms to users. +We thank Riley Goodside for valuable suggestions to improve our prompts to LLMs. We thank Luigi +Gresele and Amir Hossein Karimi for their suggestions to help us improve the formulation of our +causal discovery questions. -REFERENCES - -David H. Ackley, Geoffrey E. Hinton, and Terrence J. Sejnowski. A learning algorithm for boltzmann -machines. Cognitive Science, 9(1):147–169, 1985. ISSN 0364-0213. URL https://www. -sciencedirect.com/science/article/pii/S0364021385800124. - -Daniel Adiwardana, Minh-Thang Luong, David R. So, Jamie Hall, Noah Fiedel, Romal Thoppilan, -Zi Yang, Apoorv Kulshreshtha, Gaurav Nemade, Yifeng Lu, and Quoc V. Le. Towards a human-like -open-domain chatbot, 2020. - -Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh -Hajishirzi. MathQA: Towards interpretable math word problem solving with operation-based -In Proceedings of the 2019 Conference of the North American Chapter of the -formalisms. -Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long -and Short Papers), pp. 2357–2367. Association for Computational Linguistics, June 2019. URL -https://aclanthology.org/N19-1245. - -Daniel Andor, Luheng He, Kenton Lee, and Emily Pitler. Giving BERT a calculator: Finding -operations and arguments with reading comprehension. In Proceedings of the 2019 Conference on -Empirical Methods in Natural Language Processing and the 9th International Joint Conference -on Natural Language Processing (EMNLP-IJCNLP), 2019. URL https://aclanthology. -org/D19-1609. - -Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learn- -ing to retrieve reasoning paths over wikipedia graph for question answering. In International -Conference on Learning Representations, 2020. URL https://openreview.net/forum? -id=SJgVHkrYDH. - -Roy Bar-Haim, Ido Dagan, Bill Dolan, Lisa Ferro, Danilo Giampiccolo, Bernardo Magnini, and -Idan Szpektor. The second pascal recognising textual entailment challenge. In Proceedings of the -second PASCAL challenges workshop on recognising textual entailment, 2006. - -Dhruv Batra, Payman Yadollahpour, Abner Guzman-Rivera, and Gregory Shakhnarovich. Diverse -m-best solutions in markov random fields. In Proceedings of the 12th European Conference on -Computer Vision - Volume Part V, ECCV’12, pp. 1–16, Berlin, Heidelberg, 2012. Springer-Verlag. -ISBN 9783642337147. URL https://doi.org/10.1007/978-3-642-33715-4_1. +This material is based in part upon work supported by the German Federal Ministry of Education +and Research (BMBF): Tübingen AI Center, FKZ: 01IS18039B; by the Machine Learning Cluster of +Excellence, EXC number 2064/1 – Project number 390727645; by a National Science Foundation +award (#2306372); by a Swiss National Science Foundation award (#201009) and a Responsible +AI grant by the Haslerstiftung. Zhijing Jin is supported by PhD fellowships from the Future of Life +Institute and Open Philanthropy. We also thank OpenAI for granting Zhijing quota to their API of +GPT series through the Researcher Access Program. -10 - - Published as a conference paper at ICLR 2023 - -Luisa Bentivogli, Peter Clark, Ido Dagan, and Danilo Giampiccolo. The fifth pascal recognizing +REFERENCES -textual entailment challenge. In TAC, 2009. +Tariq Alhindi, Tuhin Chakrabarty, Elena Musi, and Smaranda Muresan. Multitask instruction-based +prompting for fallacy recognition. In Proceedings of the 2022 Conference on Empirical Methods +in Natural Language Processing, pp. 8172–8187, Abu Dhabi, United Arab Emirates, December +2022. Association for Computational Linguistics. URL https://aclanthology.org/ +2022.emnlp-main.560. 9 -BIG-bench collaboration. Beyond the imitation game: Measuring and extrapolating the capabil- -In preparation, 2021. URL https://github.com/google/ +Chandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi, Ari Holtzman, +Hannah Rashkin, Doug Downey, Wen-tau Yih, and Yejin Choi. Abductive commonsense reasoning. +In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, +April 26-30, 2020. OpenReview.net, 2020. URL https://openreview.net/forum?id= +Byg1v1HKDB. 1 -ities of language models. -BIG-bench/. +Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. A large +annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference +on Empirical Methods in Natural Language Processing, pp. 632–642, Lisbon, Portugal, September +2015. Association for Computational Linguistics. doi: 10.18653/v1/D15-1075. URL https: +//aclanthology.org/D15-1075. 9 Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, -Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, -Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, -and Dario Amodei. Language models are few-shot learners. In Advances in Neural Information -Processing Systems, 2020. URL https://proceedings.neurips.cc/paper/2020/ -file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. - -language inference with natural - -Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. - -snli: Natural -gio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, -(eds.), Advances -Information Processing Systems 31, -Curran Associates, -8163-e-snli-natural-language-inference-with-natural-language-explanations. -pdf. - -e- -In S. Ben- -and R. Garnett -pp. 9539–9549. -URL http://papers.nips.cc/paper/ - -language explanations. - -in Neural -Inc., - -2018. - -Oana-Maria Camburu, Brendan Shillingford, Pasquale Minervini, Thomas Lukasiewicz, and Phil -Blunsom. Make up your mind! adversarial generation of inconsistent natural language explanations. -In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. -4157–4165, Online, July 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -acl-main.382. URL https://aclanthology.org/2020.acl-main.382. - -Jifan Chen, Shih-Ting Lin, and Greg Durrett. Multi-hop question answering via reasoning chains. - -CoRR, abs/1910.02610, 2019. URL http://arxiv.org/abs/1910.02610. - -Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared -Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large -language models trained on code. arXiv preprint arXiv:2107.03374, 2021. - -Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam -Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, -Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam -Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James -Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Lev- -skaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin -Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, -Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. -Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon -Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark -Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeff Dean, -Slav Petrov, and Noah Fiedel. Palm: Scaling language modeling with pathways, 2022. URL -https://arxiv.org/abs/2204.02311. - -Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina -Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. In NAACL, -2019. - -Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and -Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. -ArXiv, abs/1803.05457, 2018. - -Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, -Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John -Schulman. Training verifiers to solve math word problems, 2021. - -11 - - Published as a conference paper at ICLR 2023 - -Ido Dagan, Oren Glickman, and Bernardo Magnini. The pascal recognising textual entailment - -challenge. In Machine Learning Challenges Workshop, pp. 177–190. Springer, 2005. - -Bryan Eikema and Wilker Aziz. Is MAP decoding all you need? the inadequacy of the mode in neural -machine translation. In Proceedings of the 28th International Conference on Computational Lin- -guistics, pp. 4506–4520, Barcelona, Spain (Online), December 2020. International Committee on -Computational Linguistics. URL https://aclanthology.org/2020.coling-main. -398. - -Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich -Schütze, and Yoav Goldberg. Measuring and improving consistency in pretrained language -models. Transactions of the Association for Computational Linguistics, 9:1012–1031, 2021. doi: -10.1162/tacl_a_00410. URL https://aclanthology.org/2021.tacl-1.60. - -Jonathan St BT Evans. Intuition and reasoning: A dual-process perspective. Psychological Inquiry, - -21(4):313–326, 2010. - -Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Proceedings -of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long -Papers), pp. 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics. -doi: 10.18653/v1/P18-1082. URL https://aclanthology.org/P18-1082. - -Jessica Ficler and Yoav Goldberg. Controlling linguistic style aspects in neural language generation. In -Proceedings of the Workshop on Stylistic Variation, pp. 94–104, Copenhagen, Denmark, September -2017. Association for Computational Linguistics. doi: 10.18653/v1/W17-4912. URL https: -//aclanthology.org/W17-4912. - -Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot -learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguis- -tics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long -Papers), pp. 3816–3830, Online, August 2021. Association for Computational Linguistics. doi: -10.18653/v1/2021.acl-long.295. URL https://aclanthology.org/2021.acl-long. -295. - -Mor Geva, Ankit Gupta, and Jonathan Berant. Injecting numerical reasoning skills into language -models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguis- -tics, 2020. doi: 10.18653/v1/2020.acl-main.89. URL https://aclanthology.org/2020. -acl-main.89. - -Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle -use a laptop? A question answering benchmark with implicit reasoning strategies. Transactions of -the Association for Computational Linguistics, 2021. URL https://aclanthology.org/ -2021.tacl-1.21. - -Danilo Giampiccolo, Bernardo Magnini, Ido Dagan, and Bill Dolan. The third pascal recognizing -textual entailment challenge. In Proceedings of the ACL-PASCAL workshop on textual entailment -and paraphrasing, pp. 1–9. Association for Computational Linguistics, 2007. - -Ari Holtzman, Jan Buys, Maxwell Forbes, Antoine Bosselut, David Golub, and Yejin Choi. Learning -In Proceedings of the 56th Annual Meeting of the -to write with cooperative discriminators. -Association for Computational Linguistics (Volume 1: Long Papers), pp. 1638–1649, Melbourne, -Australia, July 2018. Association for Computational Linguistics. doi: 10.18653/v1/P18-1152. -URL https://aclanthology.org/P18-1152. - -Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text -degeneration. In International Conference on Learning Representations, 2020. URL https: -//openreview.net/forum?id=rygGQyrFvH. - -Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to -solve arithmetic word problems with verb categorization. In Proceedings of the 2014 Conference on -Empirical Methods in Natural Language Processing (EMNLP), 2014. doi: 10.3115/v1/D14-1058. -URL https://aclanthology.org/D14-1058. - -12 - - Published as a conference paper at ICLR 2023 - -Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Han- -naneh Hajishirzi. UNIFIEDQA: Crossing format boundaries with a single QA system. In Findings -of the Association for Computational Linguistics: EMNLP 2020, pp. 1896–1907, Online, Novem- -ber 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.findings-emnlp.171. -URL https://aclanthology.org/2020.findings-emnlp.171. - -Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large -language models are zero-shot reasoners. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, -and Kyunghyun Cho (eds.), Advances in Neural Information Processing Systems, 2022. URL -https://openreview.net/forum?id=e2TBb5y0yFf. - -Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. MAWPS: -A math word problem repository. In Proceedings of the 2016 Conference of the North American -Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016. -doi: 10.18653/v1/N16-1136. URL https://aclanthology.org/N16-1136. - -Yihuai Lan, Lei Wang, Qiyuan Zhang, Yunshi Lan, Bing Tian Dai, Yan Wang, Dongxiang Zhang, -and Ee-Peng Lim. MWPToolkit: An open-source framework for deep learning-based math word -problem solvers. arXiv preprint arXiv:2109.00799, 2021. URL https://arxiv.org/abs/ -2109.00799. - -Jiwei Li and Dan Jurafsky. Mutual information and diverse decoding improve neural machine - -translation, 2016. URL https://arxiv.org/abs/1601.00372. - -Jiwei Li, Will Monroe, and Dan Jurafsky. A simple, fast diverse decoding algorithm for neural -generation. CoRR, abs/1611.08562, 2016. URL http://arxiv.org/abs/1611.08562. - -Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale genera- -tion: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual -Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2017. doi: -10.18653/v1/P17-1015. URL https://aclanthology.org/P17-1015. - -Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. Fantastically -ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. ArXiv, -abs/2104.08786, 2021. - -Clara Meister, Tiago Pimentel, Gian Wiher, and Ryan Cotterell. Typical decoding for natural language - -generation. arXiv preprint arXiv:2202.00666, 2022. - -Shen Yun Miao, Chao Chun Liang, and Keh Yih Su. A diverse corpus for evaluating and developing -English math word problem solvers. In Proceedings of the 58th Annual Meeting of the Asso- -ciation for Computational Linguistics, 2020. URL https://aclanthology.org/2020. -acl-main.92. - -Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adver- -sarial NLI: A new benchmark for natural language understanding. In Proceedings of the 58th -Annual Meeting of the Association for Computational Linguistics. Association for Computational -Linguistics, 2020. - -Maxwell Nye, Michael Henry Tessler, Joshua B. Tenenbaum, and Brenden M. Lake. Improving -coherence and consistency in neural sequence models with dual-system, neuro-symbolic reasoning. -In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan (eds.), Advances in Neural -Information Processing Systems, 2021. URL https://openreview.net/forum?id= -uyKk_avJ-p4. - -Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP models really able to solve simple -math word problems? In Proceedings of the 2021 Conference of the North American Chapter of -the Association for Computational Linguistics: Human Language Technologies, pp. 2080–2094, -Online, June 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main. -168. URL https://aclanthology.org/2021.naacl-main.168. - -Xinyu Pi, Qian Liu, Bei Chen, Morteza Ziyadi, Zeqi Lin, Yan Gao, Qiang Fu, Jian-Guang Lou, and - -Weizhu Chen. Reasoning like program executors, 2022. - -13 - - Published as a conference paper at ICLR 2023 - -Piotr Pi˛ekos, Mateusz Malinowski, and Henryk Michalewski. Measuring and improving BERT’s -mathematical abilities by predicting the order of reasoning. In Proceedings of the 59th Annual Meet- -ing of the Association for Computational Linguistics and the 11th International Joint Conference on -Natural Language Processing (Volume 2: Short Papers), 2021. doi: 10.18653/v1/2021.acl-short.49. -URL https://aclanthology.org/2021.acl-short.49. - -Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language - -models are unsupervised multitask learners. 2019. - -Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John -Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: -Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446, 2021. - -Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. NumNet: Machine reading comprehension -with numerical reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural -Language Processing and the 9th International Joint Conference on Natural Language Processing -(EMNLP-IJCNLP), 2019. doi: 10.18653/v1/D19-1251. URL https://aclanthology. -org/D19-1251. - -Subhro Roy and Dan Roth. Solving general arithmetic word problems. In Proceedings of the 2015 -Conference on Empirical Methods in Natural Language Processing, 2015. doi: 10.18653/v1/ -D15-1202. URL https://aclanthology.org/D15-1202. - -Jianhao Shen, Yichun Yin, Lin Li, Lifeng Shang, Xin Jiang, Ming Zhang, and Qun Liu. Generate -& rank: A multi-task framework for math word problems. In Findings of the Association for -Computational Linguistics: EMNLP 2021, pp. 2269–2279, Punta Cana, Dominican Republic, -November 2021. Association for Computational Linguistics. URL https://aclanthology. -org/2021.findings-emnlp.195. - -Freda Shi, Daniel Fried, Marjan Ghazvininejad, Luke Zettlemoyer, and Sida I. Wang. Natural -language to code translation with execution. In Proceedings of the 2022 Conference on Empirical -Methods in Natural Language Processing, pp. 3533–3546, Abu Dhabi, United Arab Emirates, -December 2022. Association for Computational Linguistics. URL https://aclanthology. -org/2022.emnlp-main.231. +Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Ben- +jamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and +Dario Amodei. Language models are few-shot learners. In H. Larochelle, M. Ranzato, R. Hadsell, +M. F. Balcan, and H. Lin (eds.), Advances in Neural Information Processing Systems, volume 33, +pp. 1877–1901. Curran Associates, Inc., 2020. URL https://proceedings.neurips. +cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. 6 + +David Maxwell Chickering. Optimal structure identification with greedy search. J. Mach. Learn. +Res., 3:507–554, 2002. URL http://jmlr.org/papers/v3/chickering02b.html. +3 -Keith E Stanovich and Richard F West. +10 -Individual differences in reasoning: Implications for -the rationality debate? Behavioral and brain sciences, 23(5):645–665, 2000. URL https: -//pubmed.ncbi.nlm.nih.gov/11301544/. + Published as a conference paper at ICLR 2024 -Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. CommonsenseQA: A question -answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of +Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep +bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language -Technologies, Volume 1 (Long and Short Papers), 2019. URL https://aclanthology. -org/N19-1421. - -Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Jason Wei, Xuezhi Wang, Hyung Won Chung, -Dara Bahri, Tal Schuster, Steven Zheng, Denny Zhou, Neil Houlsby, and Donald Metzler. Unifying -language learning paradigms, 2022. URL https://arxiv.org/abs/2205.05131. - -Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze -Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog -applications. arXiv preprint arXiv:2201.08239, 2022. URL https://arxiv.org/abs/ -2201.08239. - -Ashwin Vijayakumar, Michael Cogswell, Ramprasaath Selvaraju, Qing Sun, Stefan Lee, David -Crandall, and Dhruv Batra. Diverse beam search for improved description of complex scenes. -Proceedings of the AAAI Conference on Artificial Intelligence, 32, Apr. 2018. URL https: -//ojs.aaai.org/index.php/AAAI/article/view/12340. - -Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc -Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. -Conference on Neural Information Processing Systems (NeurIPS), 2022. URL https://arxiv. -org/pdf/2201.11903. - -14 - - Published as a conference paper at ICLR 2023 - -Sean Welleck, Ilia Kulikov, Jaedeok Kim, Richard Yuanzhe Pang, and Kyunghyun Cho. Consistency -of a recurrent language model with respect to incomplete decoding. In Proceedings of the 2020 -Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5553–5568, -Online, November 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -emnlp-main.448. URL https://aclanthology.org/2020.emnlp-main.448. - -Weiwen Xu, Yang Deng, Huihui Zhang, Deng Cai, and Wai Lam. Exploiting reasoning chains -for multi-hop science question answering. In Findings of the Association for Computational -Linguistics: EMNLP 2021, pp. 1143–1156, Punta Cana, Dominican Republic, November 2021a. -Association for Computational Linguistics. URL https://aclanthology.org/2021. -findings-emnlp.99. - -Yichong Xu, Chenguang Zhu, Shuohang Wang, Siqi Sun, Hao Cheng, Xiaodong Liu, Jianfeng -Gao, Pengcheng He, Michael Zeng, and Xuedong Huang. Human parity on commonsenseqa: -Augmenting self-attention with external attention, 2021b. URL https://arxiv.org/abs/ -2112.03254. - -Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and -Christopher D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answer- -ing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, -pp. 2369–2380, Brussels, Belgium, October-November 2018. Association for Computational Lin- -guistics. doi: 10.18653/v1/D18-1259. URL https://aclanthology.org/D18-1259. - -Xi Ye and Greg Durrett. The unreliability of explanations in few-shot prompting for textual reasoning. -In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho (eds.), Advances in -Neural Information Processing Systems, 2022. URL https://openreview.net/forum? -id=Bct2f8fRd8S. - -Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, and Meng Jiang. Diversifying -content generation for commonsense reasoning with mixture of knowledge graph experts. In -Findings of Annual Meeting of the Association for Computational Linguistics (ACL), 2022. - -Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. Calibrate before use: Improving -few-shot performance of language models. In Marina Meila and Tong Zhang (eds.), Proceed- -ings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of -Machine Learning Research. PMLR, 2021. URL https://proceedings.mlr.press/ -v139/zhao21c.html. +Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186, Minneapolis, Minnesota, June +2019. Association for Computational Linguistics. doi: 10.18653/v1/N19-1423. URL https: +//aclanthology.org/N19-1423. 1, 6 + +Clark Glymour, Kun Zhang, and Peter Spirtes. Review of causal discovery methods based on graph- +ical models. Frontiers in Genetics, 10:524, 2019. ISSN 1664-8021. doi: 10.3389/fgene.2019. +00524. URL https://www.frontiersin.org/article/10.3389/fgene.2019. +00524. 2, 3 + +Madelyn Glymour, Judea Pearl, and Nicholas P Jewell. Causal inference in statistics: A primer. John + +Wiley and Sons, 2016. 2, 3 + +Andrew Gordon, Zornitsa Kozareva, and Melissa Roemmele. SemEval-2012 task 7: Choice of +plausible alternatives: An evaluation of commonsense causal reasoning. In *SEM 2012: The First +Joint Conference on Lexical and Computational Semantics – Volume 1: Proceedings of the main +conference and the shared task, and Volume 2: Proceedings of the Sixth International Workshop on +Semantic Evaluation (SemEval 2012), pp. 394–398, Montréal, Canada, 7-8 June 2012. Association +for Computational Linguistics. URL https://aclanthology.org/S12-1052. 1 + +Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. Deberta: Decoding-enhanced +Bert with disentangled attention. In 9th International Conference on Learning Representations, +ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021. URL https:// +openreview.net/forum?id=XPZIaotutsD. 6 + +Patrik O. Hoyer, Dominik Janzing, Joris M. Mooij, Jonas Peters, and Bernhard Schölkopf. Non- +In Daphne Koller, Dale Schuurmans, +linear causal discovery with additive noise models. +Yoshua Bengio, and Léon Bottou (eds.), Advances in Neural Information Processing Systems +21, Proceedings of the Twenty-Second Annual Conference on Neural Information Processing +Systems, Vancouver, British Columbia, Canada, December 8-11, 2008, pp. 689–696. Curran As- +sociates, Inc., 2008. URL https://proceedings.neurips.cc/paper/2008/hash/ +f7664060cc52bc6f3d620bcedc94a4b6-Abstract.html. 3 + +Di Jin, Zhijing Jin, Joey Tianyi Zhou, and Peter Szolovits. Is BERT really robust? A strong baseline +In The Thirty-Fourth AAAI +for natural language attack on text classification and entailment. +Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications +of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational +Advances in Artificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020, pp. 8018– +8025. AAAI Press, 2020. URL https://aaai.org/ojs/index.php/AAAI/article/ +view/6311. 9 + +Zhijing Jin, Abhinav Lalwani, Tejas Vaidhya, Xiaoyu Shen, Yiwen Ding, Zhiheng Lyu, Mrinmaya +In Findings of +Sachan, Rada Mihalcea, and Bernhard Schölkopf. Logical fallacy detection. +the Association for Computational Linguistics: EMNLP 2022, pp. 7180â€âCœ–7198, Abu +Dhabi, United Arab Emirates, December 2022. Association for Computational Linguistics. URL +https://arxiv.org/abs/2202.13758. 9, 10 + +Emre Kıcıman, Robert Ness, Amit Sharma, and Chenhao Tan. Causal reasoning and large language + +models: Opening a new frontier for causality. arXiv preprint arXiv:2305.00050, 2023. 1, 9 + +Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, +Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-training +for natural language generation, translation, and comprehension. In Proceedings of the 58th +Annual Meeting of the Association for Computational Linguistics, pp. 7871–7880, Online, July +2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.acl-main.703. URL +https://aclanthology.org/2020.acl-main.703. 6 + +Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, +Luke Zettlemoyer, and Veselin Stoyanov. RoBERTa: A robustly optimized BERT pretraining +approach. CoRR, abs/1907.11692, 2019. URL http://arxiv.org/abs/1907.11692. 6 -15 - - Published as a conference paper at ICLR 2023 - -A APPENDIX - -A.1 ADDITIONAL EXPERIMENT RESULTS - -A.1.1 ROBUSTNESS TO SAMPLING STRATEGIES AND PARAMETERS - -In Figure 6 we ablate the results with respect to different sampling strategies and parameters by -varying T in temperature sampling and k in Top-k sampling, on LaMDA-137B. We show that -self-consistency is robust to various sampling strategies and parameters. - -Figure 6: GSM8K accuracy over LaMDA-137B. Self-consistency works under various sampling -strategies and sampling parameters. - -In Figure 7 and Figure 8, we show the results of self-consistency compared with greedy decoding a -single path over LaMDA-137B and PaLM-540B, respectively. Self-consistency improves over greedy -decode by a quite significant margin on both models, on top of high accuracy already achieved by -scaling up model sizes. - -Figure 7: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over LaMDA-137B. Sampling a higher number of diverse reasoning -paths consistently improves reasoning accuracy. +11 -We further show additional sampled reasoning paths from the LaMDA-137B model in Table 12, and -sampled reasoning paths from the PaLM-540B model in Table 13. We see that the diversity in the -additionally sampled reasoning paths indeed helps the model arrive at a more correct final answer -after aggregation. + Published as a conference paper at ICLR 2024 -A.1.2 ROBUSTNESS TO DIFFERENT SETS OF PROMPTS +Bill MacCartney and Christopher D. Manning. Modeling semantic containment and exclusion in +natural language inference. In Proceedings of the 22nd International Conference on Computational +Linguistics (Coling 2008), pp. 521–528, Manchester, UK, August 2008. Coling 2008 Organizing +Committee. URL https://aclanthology.org/C08-1066. 9 -In Table 9, we further show that self-consistency is quite robust to different sets of input prompts. -We manually wrote 3 different sets of chain-of-thought as prompts to the model. Across all sets of -prompts, self-consistency yields consistent gains over the original CoT approach. +Brendan D. McKay and Adolfo Piperno. Practical graph isomorphism, II. J. Symb. Comput., 60: +94–112, 2014. doi: 10.1016/j.jsc.2013.09.003. URL https://doi.org/10.1016/j.jsc. +2013.09.003. 5 -A.1.3 COMPARED TO MODEL ENSEMBLES +John Morris, Eli Lifland, Jin Yong Yoo, Jake Grigsby, Di Jin, and Yanjun Qi. TextAttack: A framework +for adversarial attacks, data augmentation, and adversarial training in NLP. In Proceedings of the +2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, +pp. 119–126, Online, October 2020. Association for Computational Linguistics. doi: 10.18653/v1/ +2020.emnlp-demos.16. URL https://aclanthology.org/2020.emnlp-demos.16. +9 -Additionally, we provide results of directly ensembling the outputs from multiple language models. -The results are shown in Table 10, by greedily decoding sequences from 3 language models and +OpenAI. GPT-4 technical report. CoRR, abs/2303.08774, 2023. doi: 10.48550/arXiv.2303.08774. -16 +URL https://doi.org/10.48550/arXiv.2303.08774. 1, 6 -481216202428323640#Sampled Reasoning Paths182022242628Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.5, k=20T=0.5, no top kGreedy Decode0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths4446485052545658ASDiv0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths1416182022242628GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths56586062Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths62636465666768Strategy QA0510152025303540#Sampled Reasoning Paths687072747678ARC (Easy)0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong +Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser +Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan +Leike, and Ryan Lowe. Training language models to follow instructions with human feedback. +CoRR, abs/2203.02155, 2022. doi: 10.48550/arXiv.2203.02155. URL https://doi.org/10. +48550/arXiv.2203.02155. 1, 6 -Figure 8: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over PaLM-540B. Sampling a higher number of diverse reasoning -paths consistently helps reasoning accuracy. +Judea Pearl. Probabilistic reasoning in intelligent systems: Networks of plausible inference. Morgan -Prompt set 1 (used in the main text) +Kaufmann, 1988. 3 -Prompt set 2 +Judea Pearl. Causality: Models, reasoning and inference (2nd ed.). Cambridge University Press, -Prompt set 3 +2009. 1 -CoT (Wei et al., 2022) -Self-consistency +Jonas Peters, Dominik Janzing, and Bernhard Schölkopf. Elements of causal inference: Foundations +and learning algorithms. The MIT Press, 2017. URL https://mitpress.mit.edu/ +books/elements-causal-inference. 1 -56.5 -74.4 (+17.9) +Lianhui Qin, Antoine Bosselut, Ari Holtzman, Chandra Bhagavatula, Elizabeth Clark, and Yejin +Choi. Counterfactual story reasoning and generation. In Proceedings of the 2019 Conference on +Empirical Methods in Natural Language Processing and the 9th International Joint Conference on +Natural Language Processing (EMNLP-IJCNLP), pp. 5043–5053, Hong Kong, China, November +2019. Association for Computational Linguistics. doi: 10.18653/v1/D19-1509. URL https: +//aclanthology.org/D19-1509. 1, 9 -54.6 -72.1 (+17.5) +Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language -54.0 -70.4 (+16.4) +models are unsupervised multitask learners. OpenAI Blog, 1(8), 2019. 1, 6 -Table 9: GSM8K accuracy over PaLM-540B. The results show robustness of self-consistency with -respect to different prompts in the input. +Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. DistilBERT, a distilled version +of BERT: Smaller, faster, cheaper and lighter. CoRR, abs/1910.01108, 2019. URL http: +//arxiv.org/abs/1910.01108. 6 -taking the majority vote (averaged over 10 runs). Note this is a typical ensemble approach (averaging -over the predictions over multiple models) and it achieves a performance significantly worse than -self-consistency (self-consistency over PaLM-540B gets an accuracy of 74.4%), as lower-capacity -models drag down the performance of higher-capacity models. In addition, this approach is limited in -two ways: 1) It requires multiple models for an ensemble which might not always be available, while -self-consistency only requires one single model to “self-ensemble”; 2) If one of the models is much -weaker, it can actually hurt the final performance. +Maarten Sap, Ronan Le Bras, Emily Allaway, Chandra Bhagavatula, Nicholas Lourie, Hannah +Rashkin, Brendan Roof, Noah A. Smith, and Yejin Choi. ATOMIC: an atlas of machine com- +monsense for if-then reasoning. In The Thirty-Third AAAI Conference on Artificial Intelligence, +AAAI 2019, The Thirty-First Innovative Applications of Artificial Intelligence Conference, IAAI +2019, The Ninth AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2019, +Honolulu, Hawaii, USA, January 27 - February 1, 2019, pp. 3027–3035. AAAI Press, 2019a. +doi: 10.1609/aaai.v33i01.33013027. URL https://doi.org/10.1609/aaai.v33i01. +33013027. 1, 9 -Method +Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. Social iqa: Common- -Single model +sense reasoning about social interactions. In EMNLP 2019, 2019b. 1, 9 -PaLM-540B, greedy / self-consistency +12 -Ensemble of models + Published as a conference paper at ICLR 2024 -LaMDA-137B + PaLM-540B -PaLM-540B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + PaLM-540B + GPT-3 (code-davinci-001, 175B) +Shohei Shimizu, Patrik O. Hoyer, Aapo Hyvärinen, and Antti J. Kerminen. A linear non-gaussian +acyclic model for causal discovery. J. Mach. Learn. Res., 7:2003–2030, 2006. URL http: +//jmlr.org/papers/v7/shimizu06a.html. 3 -GSM8K accuracy +Sam Shleifer and Alexander M. Rush. Pre-trained summarization distillation. CoRR, abs/2010.13002, -56.5 / 74.4 +2020. URL https://arxiv.org/abs/2010.13002. 6 -36.9 ± 0.5 -36.6 ± 0.4 -16.0 ± 0.8 -33.3 ± 0.7 +Peter Spirtes and Kun Zhang. Causal discovery and inference: Concepts and recent methodological -Table 10: Comparison of GSM8K accuracy over multiple-model ensembles. +advances. In Applied informatics, volume 3, pp. 1–28. SpringerOpen, 2016. 2, 3 -A.1.4 COMBINING SELF-CONSISTENCY WITH OTHER ENSEMBLING STRATEGIES +Peter Spirtes, Clark Glymour, and Richard Scheines. Causation, prediction, and search. 1993. 2, 3 -Self-consistency is completely compatible with other ensemble strategies, although the gains achieved -by self-consistency are significantly higher than other ensemble strategies (and can “override” the -performance gains achieved by other ensemble strategies). We further performed experiments and -include the results in Table 11 (for a fair comparison, we use 40 sets of prompts, or 40 prompt -permutations to compare with self-consistency with 40 paths, all experiments are based on PaLM- -540B). +Peter Spirtes, Clark Glymour, and Richard Scheines. Causation, Prediction, and Search, Second +Edition. Adaptive computation and machine learning. MIT Press, 2000. ISBN 978-0-262-19440-2. +1, 2, 3, 10 -17 +Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy +Liang, and Tatsunori B. Hashimoto. Stanford alpaca: An instruction-following llama model. +https://github.com/tatsu-lab/stanford_alpaca, 2023. 7 -0510152025303540#Sampled Reasoning Chains8688909294Accuracy (%)AddSub0510152025303540#Sampled Reasoning Chains727476788082Accuracy (%)ASDiv0510152025303540#Sampled Reasoning Chains30333639424548Accuracy (%)AQuAGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Chains889092949698Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Chains70.072.575.077.580.082.585.087.5Accuracy (%)SVAMP0510152025303540#Sampled Reasoning Chains505560657075Accuracy (%)GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths7475767778798081Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths7476788082Strategy QA0510152025303540#Sampled Reasoning Paths8890929496ARC (Easy)0510152025303540#Sampled Reasoning Paths788082848688ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -GSM8K accuracy - -Different sets of prompts (x40) -Prompt-permutation (x40) -Self-consistency (x40) -Self-consistency + different sets of prompts (x40) -Self-consistency + prompt-permutation (x40) - -58.9 -59.6 -74.4 -75.4 -73.8 - -Table 11: Combining self-consistency with other ensembling strategies. - -A.2 DETAILS ON RESOURCES AND INFERENCE - -For all four language models we perform prompting-based inference only. For UL2 we use TPU v3 -(2x2 configuration, 4 chips, 8 cores). For GPT-3 models the experiments are done though the public -API.10 For LaMDA-137B we use TPU v3 (8x8 configuration, 64 chips, 128 cores). For PaLM-540B -we use TPU v4 (4x4x12 configuration, 192 chips, 384 cores). Most inference jobs take 1 to 4 hours -(over about 1,000 examples) for each task on UL2 and LaMDA-137B, and about 2 to 12 hours on -PaLM-540B. Some tasks (e.g., commonsense reasoning) take longer but do not exceed 2 days for -each task. - -For GPT-3 models, we use 128 max tokens for all methods, without frequency penalty or presence -penalty. For all models, we take the generated outputs until the start of the next “Q:" to parse the final -answers, consistent with our prompting format. - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[MultiArith] Sam had 10 socks. -If he threw away 3 old ones -that didn’t fit and bought 36 new -ones, how many socks would he -have? - -Sam had 10 socks. He threw -away 3 old ones that didn’t fit. -He bought 36 new ones. That -means he has 10 + 36 - 3 = 41 -socks. The answer is 41. - -[SVAMP] In a school there are -697 girls and the rest are boys. -If there are 228 more girls than -boys. How many boys are there -in that school? - -[AQuA] There are 10 oranges in -a basket. Find the no. of ways in -which 2 oranges are chosen from -the basket? (a) 45 (b) 90 (c) 120 -(d) 150 (e) 180 - -There are 697 girls and the rest -are boys. That means there are -697 + 228 = 925 girls. The -rest are boys. So the number of -boys is 925 - 697 = 228. The -answer is 228. - -There are 10 oranges in a bas- -ket. -So there are 10 ways -in which 2 oranges are chosen -from the basket. - -[CommonsenseQA] The man -laid on the soft moss and looked -up at the trees, where was the -man? (a) niagra falls (b) forest -(c) waterfall (d) ground (e) tree - -The answer must be something -that is soft and mossy. Of the -above choices, only ground is -soft and mossy. So the answer -is (d). - -[ARC] Which of these will most -likely harm a habitat? (a) plant- -ing trees. (b) water pollution. (c) -rainfall. (d) sunlight. - -Planting trees will not harm a -habitat. So the answer is (a). - -[ARC-challenge] Which pro- -cess in the carbon cycle takes the -longest to complete? (a) emis- -sion of waste. (b) respiration in -animals. -(c) photosynthesis in -plants. (d) formation of fossil fu- -els. - -The process of photosynthesis -in plants takes the longest to -complete. So the answer is (c). - -Sam had 10 socks. He -threw away 3 old ones. So -now he has 7 left. Then he -bought 36 new ones. So -now he has 7 + 36 = 43 -socks. The answer is 43. - -He had 10 socks. He threw away -3 old ones that didn’t fit. He -bought 36 new ones. So in total -he has 10 - 3 + 36 = 43 socks. The -answer is 43. - -The number of boys is the -difference between 697 -and 228. This is 697 - 228 -= 469. The answer is 469. - -In the beginning there are 697 -girls and the rest are boys. So, the -number of boys must be 697 - 228 -= 469 boys. The answer is 469. - -There are 10 oranges in a -basket. We need to choose -2 oranges from the basket. -So the number of ways are -10C2. So the answer is (a). - -The -be -answer must -somewhere where moss -is found. Of the above -choices, only forests are -known for having moss. -So the answer is (b). - -The answer is (b), water -pollution, because it will -most likely harm the habi- -tat. The other answers are -not likely to harm the habi- -tat. - -The formation of fossil fu- -els takes the longest time -to complete. So the an- -swer is (d). - -There are 10 oranges in the bas- -ket. If you choose 2 oranges, you -have 10C2 = 45 ways of choosing -2 oranges. So the answer is (a). - -The answer must be a place where -the man is laying down and look- -ing up at trees. Of the above -is a place -choices, only forest -where people lay down and look -at trees. So the answer is (b). - -The answer is (b), water pollution. -Trees are good for a habitat, rain- -fall is necessary for a habitat, and -sunlight is necessary for a habitat. -So the answer is (b). - -The process that takes the longest -time to complete is the forma- -tion of fossil fuels, which hap- -pens over millions of years. So -the answer is (d). - -Table 12: Additional examples where self-consistency helps repair the errors over greedy decode on -LaMDA-137B. Two sampled reasoning paths that are consistent with the ground truth are shown. - -10https://beta.openai.com/docs/api-reference/making-requests - -18 - - Kyle bought a book that is sold -for $19.50. With a 25% discount, -the original price must be higher. -If we call -the original price x, -then the discounted price is 0.75x. -Since this is $19.50, we can write -0.75x = 19.50. Solving this, we -find that the original price is x = -19.50 / 0.75 = $26. The answer is -$26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. One cu- -bic meter contains 1000 litres, so -the tank contains 120 m3 * 1000 -liters/m3 = 120,000 liters. So the -answer is (d). - -Published as a conference paper at ICLR 2023 - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[GSM8K] Kyle bought -last -year’s best-selling book for -$19.50. This is with a 25% dis- -count from the original price. -What was the original price of -the book? - -The book was sold for $19.50. -This is 25% less than the origi- -nal price. This means the original -price was 100%. 25% of 100% -is 25%. -So the original price -was 100% - 25% = 75%. The -original price was 75% of $19.50. -75% of $19.50 is 75% * $19.50 = -$14.625. The answer is $14.625. - -The original price of the book -is $19.50. Since it was dis- -counted by 25%, the original -price was 19.50 / 0.75 = 26. -The answer is 26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. So the -capacity of the tank is 120 m3. So -the answer is (c). - -The volume of the tank is 8m -* 6m* 2.5m = 120m3. This is -equal to 120000 litres. So the -answer is (d) - -[AQuA] The capacity of a -tank of dimensions (8 m x 6 -m x 2.5 m) is (a) 120 litres (b) -1200 litres (c) 12000 litres (d) -120000 litres (e) None of these - -no: -[StrategyQA] Yes -Can the Great Depression be -treated with Prozac? - -or - -The Great Depression was an -economic depression. Prozac is -a drug used to treat depression. -Thus, the Great Depression could -be treated with Prozac. So the an- -swer is yes. - -Prozac is an anti-depressant -medication. The Great Depres- -sion is not a disease. Thus, -Prozac cannot treat the Great -Depression. So the answer is -no. - -The Great Depression is a time pe- -riod. Prozac is used to treat de- -pression (the mental illness). Thus, -the Great Depression can not be -treated with Prozac. So the answer -is no. - -[ARC-easy] Which source of -energy is nonrenewable? (a) -coal. -(c) water. -(d) wood. - -(b) sunlight. - -Coal, -sunlight, and water are -all renewable sources of energy. -Wood is a nonrenewable source of -energy. So the answer is (d). - -Students -[ARC-challenge] -studied the revolution of Earth -around the Sun. During which -month will Florida have the -most sunlight energy avail- -able? (a) March. (b) June. (c) -September. (d) December. +Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée +Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurélien Rodriguez, Armand +Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language +models. CoRR, abs/2302.13971, 2023. doi: 10.48550/arXiv.2302.13971. URL https://doi. +org/10.48550/arXiv.2302.13971. 7 -The Earth is closest to the Sun -in December, hence Florida will -have the most sunlight energy -available in December. So the an- -swer is (d). +Ruibo Tu, Chao Ma, and Cheng Zhang. Causal-discovery performance of chatgpt in the context of -A nonrenewable source of en- -ergy is one in which the source -is not replenished in a short -amount of time. Thus, the only -option that is nonrenewable is -coal. So the answer is (a). +neuropathic pain diagnosis. arXiv preprint arXiv:2301.13819, 2023. 1 -Coal is a fossil fuel and is formed -from dead organisms from the past. -Coal cannot be produced in a short -period of time, so it is considered -a nonrenewable source of energy. -So the answer is (a). +Moritz Willig, Matej Zeˇcevi´c, Devendra Singh Dhami, and Kristian Kersting. Probing for correlations +of causal facts: Large language models and causality, 2023. URL https://openreview. +net/forum?id=UPwzqPOs4-. 9 -Earth is closest to the Sun on -June 21, the summer solstice. -So the answer is (b). +Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, +Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick +von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gug- +ger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art +In Proceedings of the 2020 Conference on Empirical Methods +natural language processing. +in Natural Language Processing: System Demonstrations, pp. 38–45, Online, October 2020. +Association for Computational Linguistics. doi: 10.18653/v1/2020.emnlp-demos.6. URL +https://aclanthology.org/2020.emnlp-demos.6. 6, 7, 14 -Florida is closer to the sun during -June, so it will have the most sun- -light energy during that month. So -the answer is (b). +Yuxi Xie, Guanzhen Li, and Min-Yen Kan. Echo: Event causality inference via human-centric -Table 13: Additional examples where self-consistency helps repair the errors over greedy decode on -PaLM-540B. Two sampled reasoning paths that are consistent with the ground truth are shown. +reasoning. arXiv preprint arXiv:2305.14740, 2023. 1 -A.3 FULL SETS OF PROMPTS +Matej Zeˇcevi´c, Moritz Willig, Devendra Singh Dhami, and Kristian Kersting. Causal parrots: Large +language models may talk causality but are not causal. arXiv preprint arXiv:2308.13067, 2023. 1 -We list the full details of the prompts used for two newly-introduced datasets, AQUA-RAT (Ling -et al., 2017) and AI2 Reasoning Challenge (ARC) (Clark et al., 2018), where we manually composed -the example chain-of-thought in this paper, in Table 14 and Table 15, respectively. +Kun Zhang and Aapo Hyvärinen. Causality discovery with additive disturbances: An information- +theoretical perspective. In Machine Learning and Knowledge Discovery in Databases: European +Conference, ECML PKDD 2009, Bled, Slovenia, September 7-11, 2009, Proceedings, Part II 20, +pp. 570–585. Springer, 2009. 3 -Table 14: Few-shot exemplars for AQUA-RAT. -Q: John found that the average of 15 numbers is 40. If 10 is added to each number then the mean of the -numbers is? Answer Choices: (a) 50 (b) 45 (c) 65 (d) 78 (e) 64 +Li Zhang, Qing Lyu, and Chris Callison-Burch. Reasoning about goals, steps, and temporal ordering +with WikiHow. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language +Processing (EMNLP), pp. 4630–4639, Online, November 2020. Association for Computational +Linguistics. doi: 10.18653/v1/2020.emnlp-main.374. URL https://aclanthology.org/ +2020.emnlp-main.374. 9 -A: If 10 is added to each number, then the mean of the numbers also increases by 10. So the new mean -would be 50. The answer is (a). +Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher +Dewan, Mona T. Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt +Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. +OPT: open pre-trained transformer language models. CoRR, abs/2205.01068, 2022. doi: 10.48550/ +arXiv.2205.01068. URL https://doi.org/10.48550/arXiv.2205.01068. 1 -Q: If a / b = 3/4 and 8a + 5b = 22,then find the value of a. Answer Choices: (a) 1/2 (b) 3/2 (c) 5/2 (d) 4/2 (e) -7/2 +13 -A: If a / b = 3/4, then b = 4a / 3. So 8a + 5(4a / 3) = 22. This simplifies to 8a + 20a / 3 = 22, which means -44a / 3 = 22. So a is equal to 3/2. The answer is (b). + Published as a conference paper at ICLR 2024 + +A IMPLEMENTATION DETAILS + +When finetuning on our data, for GPT-based models, we use the default settings of the OpenAI +finetuning API; and for BERT-based models, we use the transformers library (Wolf et al., 2020) +and train the models on a server with an NVIDIA Tesla A100 GPU with 40G of memory. To fit for +the GPU memory, we set the batch size to be 8. We use the validation set to tune the learning rate, +which takes value in {2e-6, 5e-6, 1e-5, 2e-5, 5e-5}; dropout rate, which takes value in {0, 0.1, 0.2, +0.3}; and weight decay, which takes value in {1e-4, 1e-5}. We train the models until convergence, +which is usually around ten epochs. + +Prompts When querying the autoregressive LLMs, we formulate the prompt as follows: + +Question: [premise] + +Can we deduct the following: [hypothesis]? Just answer "Yes" or "No." + +Answer: + +B GENERATING NATURAL STORIES + +To generate the natural stories based on our symbolic expressions, we utilize the state-of-the-art LLM, +GPT-4, which is very good at story generation. We design detailed instructions in the prompt, and +generate around 200 stories in our case study. We show two examples stories in Table 7, and the +report the overall statistics in Table 8. + +Symbolic +Form + +Example 1 (Label=Negative) +Premise: Suppose there is a closed system +of 2 variables, A and B. All the statistical +relations among these 2 variables are as fol- +lows: A correlates with B. +Hypothesis: A directly affects B. +Relation between the promise and hypothe- +sis: The premise does not a necessary con- +dition for the hypothesis. + +Natural +Story + +Premise: Suppose there is a closed system +of 2 variables, ice cream sales and swim- +ming pool attendance. All the statistical +relations among these 2 variables are as fol- +lows: ice cream sales correlate with swim- +ming pool attendance. +Hypothesis: Ice cream sales directly affect +swimming pool attendance. +Relation between the premise and hypoth- +esis: The premise does not provide a nec- +essary condition for the hypothesis. The +correlation between ice cream sales and +swimming pool attendance could be due to +a third variable, such as hot weather, which +increases both ice cream sales and swim- +ming pool attendance. Therefore, it is not +necessarily true that ice cream sales directly +affect swimming pool attendance. + +Example 2 (Label=Positive) +Premise: Suppose there is a closed system +of 3 variables, A, B and C. All the statisti- +cal relations among these 3 variables are as +follows: A correlates with C. B correlates +with C. However, A is independent of B. +Hypothesis: A directly affects C. +Relation between the promise and hypoth- +esis: The premise is a necessary condition +for the hypothesis. So if the premise is true, +the hypothesis must be true. +Premise: Let’s consider three factors: eat- +ing junk food (A), obesity (C), and watch- +ing television (B). There is a correlation +between eating junk food and obesity, and +between watching television and obesity. +However, eating junk food and watching +television are independent from each other. +Hypothesis: Eating junk food directly af- +fects obesity. +Relation between the premise and hypoth- +esis: The premise provides the necessary +conditions for the hypothesis. It establishes +the independent variables A (eating junk +food) and B (watching television) and their +correlations with obesity. Given that these +are true, it supports the hypothesis that eat- +ing junk food directly affects obesity. + +Table 7: Examples of natural stories generated based on the symbolic form in our CORR2CAUSE +dataset, showing the broad application value of our dataset as the starting point for various verbaliza- +tions of the correlation-to-causation inference task. -Q: A person is traveling at 20 km/hr and reached his destiny in 2.5 hr then find the distance? Answer Choices: -(a) 53 km (b) 55 km (c) 52 km (d) 60 km (e) 50 km +14 -A: The distance that the person traveled would have been 20 km/hr * 2.5 hrs = 50 km. The answer is (e). + Published as a conference paper at ICLR 2024 -Q: How many keystrokes are needed to type the numbers from 1 to 500? Answer Choices: (a) 1156 (b) 1392 -(c) 1480 (d) 1562 (e) 1788 +Test Set Size +Dev Set Size +# Tokens/Premise +# Tokens/Hypothesis +# Tokens/Explanation +% Positive Labels -A: There are 9 one-digit numbers from 1 to 9. There are 90 two-digit numbers from 10 to 99. There are 401 -three-digit numbers from 100 to 500. 9 + 90(2) + 401(3) = 1392. The answer is (b). +102 +102 +64.88 +13.54 +64.66 +1.67 -19 +Table 8: Statistics of our generated natural stories. We report the number of samples in the test and de- +velopment sets; number of tokens per premise (# Tokens/Premise), hypothesis (# Tokens/Hypothesis), +and explanation (# Tokens/Explanation); and percentage of the positive labels (% Positive Labels). - Published as a conference paper at ICLR 2023 +For more information, the exact prompt we use is “Here is a causal inference rule: [symbolic +form] Please provide a real-world example instantiating this phenomenon. Format it also as +"Premise:", "Hypothesis:", and "Relation between the promise and hypothesis:".” -Table 15: Few-shot exemplars for ARC easy/challenge. -Q: George wants to warm his hands quickly by rubbing them. Which skin surface will produce the most -heat? (a) dry palms. (b) wet palms. (c) palms covered with oil. (d) palms covered with lotion. +C TEMPLATES AND PARAPHRASES -A: Dry surfaces will more likely cause more friction via rubbing than other smoother surfaces, hence dry -palms will produce the most heat. The answer is (a). +We use the verbalization templates in Table 9 to compose the hypotheses for all six causal relations. -Q: Which factor will most likely cause a person to develop a fever? (a) a leg muscle relaxing after exercise. -(b) a bacterial population in the bloodstream. (c) several viral particles on the skin. (d) carbohydrates being -digested in the stomach. +{Var i} directly causes {Var j}. +{Var i} causes something else which causes {Var j}. +{Var j} directly causes {Var i}. +{Var j} is a cause for {Var i}, but not a direct one. +There exists at least one collider (i.e., common effect) of {Var i} and {Var j}. +There exists at least one confounder (i.e., common cause) of {Var i} and {Var j}. -A: Option (b), bacterial population is the most likely cause for a person developing fever. The answer is (b). +Causal Relation Hypothesis Template +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder +Paraphrases +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder -Q: Which change in the state of water particles causes the particles to become arranged in a fixed position? -(a) boiling. (b) melting. (c) freezing. (d) evaporating. +{Var i} directly affects {Var j}. +{Var i} influences {Var j} through some mediator(s). +{Var j} directly affects {Var i}. +{Var j} influences {Var i} through some mediator(s). +{Var i} and {Var j} together cause some other variable(s). +Some variable(s) cause(s) both {Var i} and {Var j}. -A: When water is freezed, the particles are arranged in a fixed position; the particles are still moving for all -other options. The answer is (c). +Table 9: Templates and their paraphrases for each causal relation in the hypothesis. We use {Var +i} and {Var j} as placeholders for the two variables. -Q: When a switch is used in an electrical circuit, the switch can (a) cause the charge to build. (b) increase -and decrease the voltage. (c) cause the current to change direction. (d) stop and start the flow of current. +D CHANGE LOG FOR THE DATASET VERSION UPDATE -A: The function of a switch is to start and stop the flow of a current. The answer is (d). +Two Equivalent Forms +ß Is-Parent(i, j) +Is-Child(j, i) -Table 16: Few-shot exemplars for HotpotQA (closed-book setting). +ß Is-Ancestor(i, j) (Original) -Q: Which magazine was started first Arthur’s Magazine or First for Women? +Is-Descendent(j, i) (Original) +ß Is-Ancestor(i, j) (Paraphrased) -A: Arthur’s Magazine started in 1844. First for Women started in 1989. So Arthur’s Magazine was started first. -The answer is Arthur’s Magazine. +Is-Descendent(j, i) (Paraphrased) -Q: The Oberoi family is part of a hotel company that has a head office in what city? +ß Has-Collider(i, j) +Has-Collider(j, i) +ß Has-Confounder(i, j) +Has-Confounder(j, i) -A: The Oberoi family is part of the hotel company called The Oberoi Group. The Oberoi Group has its head -office in Delhi. The answer is Delhi. +Duplication Property -Q: What nationality was James Henry Miller’s wife? +De-Duplication Method -A: James Henry Miller’s wife is June Miller. June Miller is an American. The answer is American. +Two exact same strings -Q: The Dutch-Belgian television series that "House of Anubis" was based on first aired in what year? +Keep only one, by forcing i < j -A: "House of Anubis" is based on the Dutch–Belgian television series Het Huis Anubis. Het Huis Anubis is first -aired in September 2006. The answer is 2006. +Two different strings, but +semantically equivalent -As additional information, we also list the exact set of prompts used for all arithmetic reasoning tasks -in Table 17, since there are multiple sets of prompts introduced in Wei et al. (2022). The prompts for -CommonsenseQA and StrategyQA are the same as used in Wei et al. (2022). +Randomly sample one out of the two -We provide the exact prompts used for common NLP tasks in the following tables as well, including -NLI (Table 18, Table 19, Table 20) and Closed-Book Question-Answering tasks (Table 16, Table 21). +Two exact same strings -20 +Keep only one, by forcing i < j - Published as a conference paper at ICLR 2023 +Two different strings, but +semantically equivalent -Table 17: Few-shot exemplars for all arithmetic reasoning tasks, from Wei et al. (2022). -Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done, -there will be 21 trees. How many trees did the grove workers plant today? +Two different strings, but +semantically equivalent -A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted. -So, they must have planted 21 - 15 = 6 trees. The answer is 6. +Randomly sample one out of the two -Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot? +Randomly sample one out of the two -A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5. +Table 10: De-duplication methods for the six causal relation types and their verbalizations. -Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total? +De-Duplication Strategy As mentioned in Section 3.7 in the main paper, our original dataset (v1.0) +has duplication due to symmetric relations and verbalizations. We introduce in Table 10 several +reasons for why duplicated hypotheses exist in our original data. One typical reason is symmetric +relations such as Is-Parent(A, B) and Is-Child(B, A), and, similarly, the paraphrased version of -A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74 -chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39. +15 -Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops -did Jason give to Denny? + Published as a conference paper at ICLR 2024 + +Is-Ancestor(A, B) and Is-Descendent(B, A). Another typical reason is the semantic equivalence in +the verbalization templates, which applies to the Has-Collider and Has-Confounder relations. For +example, the verbalized texts of Has-Collider(A, B) and Collider(B, A) are “There exists at least one +collider (i.e., common effect) of {A and B, B and A},” respectively, which are semantically-equivalent +paraphrases of each other, so we randomly keep one out of the two. + +Resulting Dataset Statistics after De-Duplication + +Since the reason for duplication in the first place is due to symmetry in the causal relation, or +verbalization, the resulting new data, CORR2CAUSE v2.0, is exactly a half of the original data. As we +reported previously in Table 3 of Section 3.7, the total number of samples cuts down to half, while +the label distribution and all other properties are the same. To compose each split, we apply the same +de-duplication method for the test, train, and development sets. We notice that some duplicates are +across the splits, so we prioritize keeping the test and training sets untouched (to minimally affect the +experimental results), and then reduce the development set by removing the cross-split duplicates, +namely: + +• test_2.0 = deduplicate(test_1.0) +• train_2.0 = deduplicate(train_1.0) +• dev_2.0 = deduplicate(dev_1.0) \ {test_2.0, train_2.0} + +We expect minimal or almost no change to the experimental results. In case of the slight possibility +that this change in the development set might affect the model selection in the training process, future +work can feel free to re-train the models and update the exact performance number. + +E SPURIOUS CORRELATION ANALYSIS + +The inspirations of our two robustness tests (paraphrasing and variable refactorization) come from our +data analysis. We check for spurious correlations in the data by reporting in Table 11 the point-wise +mutual information (PMI) between the label and any n-gram with no more than four tokens. In +addition, we also report the difference of the PMI with the two labels in the |Diff| column of Table 11, +and report the top 10 n-grams. + +The design spirit for our robustness test is that if the models’ correct judgment relies on exploiting +these spurious correlations, then such reliance will be broken in our perturbations. + +N-Gram +a cause +a cause for +A causes +A causes something +a direct +a direct one +for D +for D but +for E +for E but + +PMI w/ Non-Ent. Label +1.692209 +1.663640 +1.640679 +1.621820 +1.606052 +1.592673 +1.584826 +1.583897 +1.582980 +1.582074 + +PMI w/ Ent. Label +-1.025611 +-0.983790 +-0.951610 +-0.926075 +-0.905316 +-0.888107 +-0.878180 +-0.877014 +-0.875864 +-0.874728 + +|Diff| +2.717820 +2.647430 +2.592289 +2.547895 +2.511369 +2.480781 +2.463006 +2.460911 +2.458844 +2.456802 + +Table 11: PMI between the labels and n-grams. The labels include non-entailment (Non-Ent.) and +entailment (Ent.). And the n-grams include all with no more than four words. The |Diff| column +shows the absolute value of the difference between the PMIs with two labels. We show the top 10 +n-grams with the largest differences of their PMIs with the two classes in the |Diff| column. + +We can see that some spurious correlations are rooted in the framing of the hypothesis, such as “a +cause (for)”, and “a direct (one)” (which we use the paraphrasing task to break), and others are +connected to the variable names, such as “for D (but)” and “for E (but)” (which we use the variable +refactorization to break). + +F FINE-GRAINED ERROR ANALYSIS + +In addition to the fine-grained analysis by causal relation type in Table 6a for fine-tuned models, we +also report such error analysis for non-finetuned models in Table 12. -A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of -lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8. +16 -Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does -he have now? + Published as a conference paper at ICLR 2024 + +Relation Type +All +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder +All +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant +Has-Collider +Has-Confounder + +Selected Models +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-3.5 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +GPT-4 +RoBERTa MNLI All +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI +RoBERTa MNLI Has-Collider +RoBERTa MNLI Has-Confounder + +Is-Parent +Is-Ancestor +Is-Child +Is-Descendant + +F1 +21.69 +8.82 +0 +9.84 +14.29 +34.24 +15.33 +29.08 +0 +30.77 +0 +26.98 +44.1 +20.67 +22.79 +0 +0 +0 +0 +43.45 +0 -A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so -in total he has 7 + 2 = 9 toys. The answer is 9. +Precision Recall Accuracy +27.78 +4.62 +0 +5.17 +17.86 +52.07 +56.76 +47.66 +0 +30.3 +0 +60.71 +81.82 +100 +16.96 +0 +0 +0 +0 +47.93 +0 -Q: There were nine computers in the server room. Five more computers were installed each day, from -monday to thursday. How many computers are now in the server room? +69.46 +83.47 +90.67 +85.33 +84 +35.12 +37.8 +64.6 +82.67 +88 +84.53 +75.47 +32.71 +23.86 +82.5 +82.67 +91.2 +84.53 +92.53 +59.52 +84.45 + +17.79 +100 +0 +100 +11.9 +25.51 +8.86 +20.92 +0 +31.25 +0 +17.35 +30.18 +11.53 +34.73 +0 +0 +0 +0 +39.73 +0 -A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 = -20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers. -The answer is 29. +Table 12: Fine-grained evaluation results for some selected non-fine-tuned models. -Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many -golf balls did he have at the end of wednesday? +These results are particularly revealing, showing how off-the-shelf models perform in recognizing +specific relations. Specifically, GPT-3.5 cannot recognize ancestor relations, whereas GPT-4 fails at +all direct causation recognition with parents and children. And RoBERTa MNLI only did collider +relation relatively correctly. Note that, when the F1 score is zero, the accuracy number is a result of +always predicting the negative class of that relation. -A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On -Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33. +G LLM PERFORMANCE OPTIMIZATION -Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left? +Since our experiments in Section 4.2 are based on plain, zero-shot prompts, we explore whether better +prompting strategies could improve the performance. We enhance the query prompt by incorporating +several strategies: (1) Utilizing a system prompt that specifies the model’s expertise (“You are a highly +intelligent question-answering bot with profound knowledge of causal inference.”); (2) Including +a pair of few-shot examples, one positive and one negative; (3) Implementing chain-of-thought +prompting with “Let’s think step by step.” to encourage the language model to generate step-by-step +reasoning. In Table 13, we present the evaluation results on the relatively affordable model, GPT-3.5, +where the optimized prompt leads to a 4-point improvement in F1 over the original performance. +However, we can see that despite the deployment of all three strategies, the model continues to +struggle with this challenging task. -A: She bought 5 bagels for $3 each. This means she spent 5 * $3 = $15 on the bagels. She had $23 in -beginning, so now she has $23 - $15 = $8. The answer is 8. +GPT-3.5 (plain query; original) +GPT-3.5 (enhanced query) -21 +F1 +21.69 +25.44 - Published as a conference paper at ICLR 2023 +Precision +17.79 +17.29 -Table 18: Few-shot exemplars for ANLI. +Recall Accuracy +27.78 +48.11 -Premise: -"Conceptually cream skimming has two basic dimensions - product and geography." -Based on this premise, can we conclude the hypothesis "Product and geography are what make cream skimming -work." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: Based on "cream skimming has two basic dimensions" we can’t infer that these two dimensions are what -make cream skimming work. The answer is it is not possible to tell. - -Premise: -"One of our member will carry out your instructions minutely." -Based on this premise, can we conclude the hypothesis "A member of my team will execute your orders with -immense precision." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "one of" means the same as "a member of", "carry out" means the same as "execute", and "minutely" means -the same as "immense precision". The answer is yes. - -Premise: -"Fun for adults and children." -Based on this premise, can we conclude the hypothesis "Fun for only children." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "adults and children" contradicts "only children". The answer is no. - -Premise: -"He turned and smiled at Vrenna." -Based on this premise, can we conclude the hypothesis "He smiled at Vrenna who was walking slowly behind -him with her mother." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: the premise does not say anything about "Vrenna was walking". The answer is it is not possible to tell. - -Premise: -"well you see that on television also" -Based on this premise, can we conclude the hypothesis "You can see that on television, as well." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "also" and "as well" mean the same thing. The answer is yes. - -Premise: -"Vrenna and I both fought him and he nearly took us." -Based on this premise, can we conclude the hypothesis "Neither Vrenna nor myself have ever fought him." is -true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "Vrenna and I both" contradicts "neither Vrenna nor myself". The answer is no. - -22 - - Published as a conference paper at ICLR 2023 - -Table 19: Few-shot exemplars for e-SNLI (the rationales are crowd-sourced annotations from the -original dataset). - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is training his horse for a competition." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: The person is not necessarily training his horse. The answer is it is not possible to tell. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is at a diner, ordering an omelette." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: One jumping horse cannot be in a diner ordering food. The answer is no. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is outdoors, on a horse." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: A broken down airplane is outdoors. The answer is yes. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "They are smiling at their parents." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: Just because they are smiling and waving at a camera does not imply their parents or anyone is anyone behind -it. The answer is it is not possible to tell. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "The kids are frowning." is true? OPTIONS: -- yes -- no -- it is not possible to tell -A: One cannot be smiling and frowning at the same time. The answer is no. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "There are children present." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A:The children must be present to see them smiling and waving. The answer is yes. - -23 - - Published as a conference paper at ICLR 2023 - -Table 20: Few-shot exemplars for RTE. - -Premise: -"No Weapons of Mass Destruction Found in Iraq Yet." -Based on this premise, can we conclude the hypothesis "Weapons of Mass Destruction Found in Iraq." is true? - -A: "No Weapons of Mass Destruction Found" contradicts "Weapons of Mass Destruction Found". The answer is -no. - -Premise: -"A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful -gathered in downtown Chicago to mark the installation of new Pope Benedict XVI." -Based on this premise, can we conclude the hypothesis "Pope Benedict XVI is the new leader of the Roman -Catholic Church." is true?’ - -A: "installation of new Pope Benedict XVI." means "Pope Benedict XVI is the new leader". The answer is yes. - -Premise: -"A man is due in court later charged with the murder 26 years ago of a teenager whose case was the first to -be featured on BBC One’s Crimewatch. Colette Aram, 16, was walking to her boyfriend’s house in Keyworth, -Nottinghamshire, on 30 October 1983 when she disappeared. Her body was later found in a field close to her -home. Paul Stewart Hutchinson, 50, has been charged with murder and is due before Nottingham magistrates -later." -Based on this premise, can we conclude the hypothesis "Paul Stewart Hutchinson is accused of having stabbed a -girl." is true? - -A: The premise does not say Paul Stewart Hutchinson "stabbed" this girl. The answer is no. - -Premise: -"Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it -will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients." -Based on this premise, can we conclude the hypothesis "Herceptin can be used to treat breast cancer." is true? +69.46 +52.01 -A: "Herceptin was approved to treat breast cancer" implies that "Herceptin can be used to treat breast cancer". -The answer is yes. - -Table 21: Few-shot exemplars for BoolQ (closed-book setting). - -Q: does system of a down have 2 singers? - -A: System of a Down currently consists of Serj Tankian, Daron Malakian, Shavo Odadjian and John Dolmayan. -Serj and Daron do vocals, so the band does have two singers. The answer is yes. - -Q: do iran and afghanistan speak the same language? - -A: Iran and Afghanistan both speak the Indo-European language Persian. The answer is yes. - -Q: is a cello and a bass the same thing? - -A: The cello is played sitting down with the instrument between the knees, whereas the double bass is played -standing or sitting on a stool. The answer is no. - -Q: can you use oyster card at epsom station? - -A: Epsom railway station serves the town of Epsom in Surrey and is not in the London Oyster card zone. The -answer is no. +Table 13: Performance of GPT-3.5 with different queries. We quote the original performance from +Table 4. -24 +17 diff --git a/q143/random_k4/random_3.md b/q143/random_k4/random_3.md index 692dc5af339caacaddf265fe3e57e4a7f5067249..33ce8f19cae438595e04b0d40d28bacb32daad37 100644 --- a/q143/random_k4/random_3.md +++ b/q143/random_k4/random_3.md @@ -1,1533 +1,1454 @@ -GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT -OF INFORMATION TECHNOLOGY CAPITAL ASSETS - -Table of Contents - -300.1 What is the purpose of the section? -300.2 Does the Exhibit 300 apply to me? -300.3 What background information must I know? -300.4 What special terms should I know? -300.5 How will agencies manage capital assets? -300.6 What other requirements does Exhibit 300 fulfill? -300.7 What must I report on Exhibit 300A&B and when? -300.8 How will OMB use Exhibit 300A&B? - -Ex–300A&B IT Capital Asset Summary and Performance Measurement Report - -Summary of Changes - -Introduces new definitions to assist Agencies in the consistent planning, acquisition, and operation of IT -assets - -Changes Unique Project Identifier to Unique Investment Identifier - -Introduces Exhibits 300 A&B, “IT Capital Asset Summary and Performance Measurement Report” - -300.1 What is the purpose of the section? - -Exhibits 300s establish policy for planning, budgeting, acquisition and management of major information -technology (IT) capital investments. The Office of Management and Budget (OMB) provides procedural and -analytic guidelines for implementing specific aspects of these policies described in OMB Circulars and their -associated appendices. - -Exhibit 300s are companions to an agency’s Exhibit 53. Exhibit 300s and the Exhibit 53, together with the -agency’s Enterprise Architecture program, define how to manage the IT Capital Planning and Control Process. -Exhibit 53A is a tool for reporting the funding of the portfolio of all IT investments within a Department while -Exhibit 300A is a tool for detailed justifications of major "IT Investments". Exhibit 300B is for the management -of the execution of those investments through their project life cycle and into their useful life in production. By -integrating the disciplines of architecture, investment management, and project implementation, these programs -provide the foundation for sound IT management practices, end-to-end governance of IT capital assets, and the -alignment of IT investments with an agency’s strategic goals. As architecture-driven IT investments are funded in -the "Invest" phase, they move forward into the implementation phase where system development life cycle -processes are followed and actual versus planned outputs, schedule, and operational performance expenditures are -tracked utilizing performance-based management processes. New for the FY 2013 budget process, Exhibit 300B -requires agencies to provide more detailed benchmarks for the management and performance of projects and -operational assets associated with a major investment. - -300.2 - - Does Exhibit 300 apply to me? - -The policy and budget justification principles in this Exhibit apply to all agencies of the Executive Branch of the -Government subject to Executive Branch review (see Section 25). Exhibit 300A&B must be submitted for each -major IT investment for CIO Council agencies in accordance with this section, parallel requirements for other - -Page 1 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -agencies are addressed in separate guidance. Major IT investments also must be reported on the agency’s Exhibit -53. - -300.3 - - What background information must I know? - -The Federal Government must effectively manage its portfolio of capital assets to ensure scarce public resources -are wisely invested. Capital programming integrates the planning, acquisition and management of capital assets -into the budget decision-making process and is intended to assist agencies in improving asset management and in -complying with the results-oriented requirements of: - -• The Clinger-Cohen Act (CCA) of 1996, Public Law 104 – 106, legislatively mandates that IT - -investments be prudently managed. CCA requires federal agencies to focus on the results achieved -through IT investments while streamlining the federal IT procurement process. Congress and OMB have -clearly stated that each agency must actively manage its IT program to provide assurances that technology -expenditures are necessary and shall result in demonstrated improvements in mission effectiveness and -customer service. - -• The Federal Acquisition Streamlining Act of 1994, Title V (FASA V), which requires agencies to -establish cost, schedule and measurable performance goals for all major acquisition programs, and -achieve on average 90 percent of those goals. - -• Security: For IT investments, agencies should maintain up-to-date tracking of systems in the FISMA - -inventory to the appropriate IT investment. Costs for security will be collected in both the Exhibit 53A -and 53B. - -• Enterprise Architecture (EA): The IT investment must be included in the agency’s EA and Capital - -Planning and Investment Control (CPIC) process and mapped to and supporting the Federal Enterprise -Architecture (FEA). The business case must demonstrate the relationship between the investment and the -business, performance, data, services, application, and technology layers of the agency’s EA. Are this -investment and its assets included in your agency’s target enterprise architecture through the underlying -segment architecture? For detailed guidance regarding segment architecture requirements, please refer to -http://www.whitehouse.gov/omb/e-gov/. See this guidance also regarding the reporting of six digit codes -corresponding to agency segment architectures in Exhibit 53, and, for limited cases determined by the -Chief Architect, reporting an investment alignment with multiple segments. - -• Additional background information for Information Technology can be found in Section 53.2 of the - -guidance for Exhibit 53. - -300.4 What special terms should I know? - -Alternatives Analysis refers to an analysis of alternative approaches addressing the performance objectives of an -investment, performed prior to the initial decision to implement a solution, and updated periodically as -appropriate to capture changes in the context for an investment decision. Alternatives analysis details should be -available upon request. - -Asset refers to anything that has value to an organization, including, but not limited to: computing device, -information technology (IT) system, IT network, IT circuit, software (both an installed instance and a physical -instance), virtual computing platform (common in cloud and virtualized computing), and related hardware (e.g., -locks, cabinets, keyboards). Assets are the lowest level at which information technology is planned, acquired, -implemented and operated. - -Capital assets means land, structures, equipment, intellectual property (e.g., software), and information -technology (including the output of IT service contracts) used by the Federal Government and having an - -Page 2 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -estimated useful life of two years or more. See Appendix One of the Capital Programming Guide for a more -complete definition of capital assets. - -Capital investment means the planning, development, acquisition of a capital asset and the management and -operation of that asset through its usable life after the initial acquisition. IT Capital investments may consist of -one or more assets, the planning, development and acquisition of which are managed through projects, and which -then provide useful components in an operational (production) environment. - -Capital programming means an integrated process within an agency for planning, budgeting, procurement and -management of the agency’s portfolio of capital assets to achieve agency strategic goals and objectives with the -lowest overall cost and least risk. - -Cost means the expenditure of funds or use of property to acquire, produce, operate or maintain an asset. -Examples include, but are not limited to: sunk costs, operational costs, acquisition costs and disposition costs -(including variable costs such as labor hours). - -Cost saving represents the reduction in actual expenditures to achieve a specific objective (as defined in OMB -Circular A–131). Cost savings should be cited in descriptions, and may be included as a benefit in alternative -analyses. - -Cost avoidance represents results from an action taken in the immediate time frame that will decrease costs in the -future (as defined in OMB Circular A–131). Cost avoidance should be cited in descriptions, and may be included -in alternative analyses. - -Defense Acquisition Workforce Improvement Act (DAWIA) (P.L. 101-510 of November 5, 1990) established, -for the Department of Defense (DoD), an Acquisition Corps to professionalize the acquisition workforce in DoD -through education, training, and work experience. - -Dependency means the identification of relationships between projects and operational assets within an -investment, and identification of relationships between investments. Identification of dependencies is critical to -the management of project, program, and portfolio risk. - -Development, Modernization and Enhancement (DME) Costs are costs for projects leading to new IT -assets/systems and projects that change or modify existing IT assets to: substantively improve capability or -performance; implement legislative or regulatory requirements; or to meet an agency leadership request. Capital -costs as part of DME can include hardware, software development and acquisition costs, COTS acquisition costs, -government labor costs, and contracted labor costs for planning, development, acquisition, system integration, and -direct project management and overhead support. - -Disposition Cost for an asset refers to the cost of retiring a capital asset once its useful life is completed or it has -been superseded by a replacement asset, and may be included in operational costs. - -Earned Value Management (EVM) is a project management tool effectively integrating the project scope of -work with schedule and cost elements for optimum project planning and control. The qualities and operating -characteristics of earned value management systems (EVMS) are described in American National Standards -Institute (ANSI)/Electronic Industries Alliance (EIA) Standard –748–1998, Earned Value Management Systems, -approved May 19, 1998. It was reaffirmed on August 28, 2002. Additional information on EVMS is available at -www.acq.osd.mil/pm. - -Enterprise Architecture (EA) is the explicit description and documentation of the current and desired -relationships among business and management processes and information technology of an organization. It -describes the "current architecture" and "target architecture" to include the rules and standards and systems life -cycle information to optimize and maintain the environment which the agency wishes to create and maintain by - -Page 3 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -managing its IT portfolio. The EA must also provide a strategy to enable the agency to support its current state -and also act as the roadmap for transition to its target environment. The EA will define principles and goals and -set direction on such issues as the promotion of interoperability, open systems, public access, end user -satisfaction, and IT security. The agency must support the EA with a complete inventory of agency information -resources, including personnel, equipment, and funds devoted to information resources management and -information technology, at an appropriate level of detail. - -Federal Acquisition Certification for Program and Project Managers (FAC-P/PM) was established to ensure -general training and experience requirements for program and project managers are clearly identified for civilian -agencies. The FAC-P/PM focuses on essential competencies needed for program and project managers; the -program does not include functional or technical competencies, such as those for information technology or -agency-specific competencies. Defense agencies have a similar certification program under the Defense -Acquisition Workforce Improvement Act (DAWIA). Agencies were required to be compliant with FAC-P/PM -starting in FY 2008. Available levels are Entry/Apprentice, Mid/Journeyman and Expert/Advanced for FAC- -P/PM and 1, 2 and 3 for DAWIA. -(http://www.whitehouse.gov/sites/default/files/omb/procurement/workforce/fed_acq_cert_042507.pdf ) -(www.whitehouse.gov/omb/procurement/acq_wk/fac_contracting_program.pdf) - -Federal Enterprise Architecture (FEA) is a business-based documentation and analysis framework for -government-wide improvement. The FEA allows agencies to use standardized methods to describe the -relationship between an agency’s strategic goals, business functions, and enabling technologies at various levels -of scope and complexity. The FEA is comprised of documentation in six domain areas (strategic goals, business -services, data and information, systems and applications, infrastructure, and security) and six reference models -areas that are designed to facilitate standardized analysis, reporting, and the identification of duplicative -investments, gaps, and opportunities for collaboration within and across federal agencies. More information -about the FEA and reference models is available at http://www.whitehouse.gov/omb/e-gov/fea. - -Federal Segment Architecture Methodology (FSAM) is to become the “Federal Solution Architecture -Methodology” in October 2011 and will serve as a scalable and repeatable process for solution architecture at the -application, system, segment, enterprise, sector, government-wide, national, and international levels of scope. -Consistent use of the FSAM should result in more complete and consistent architecture products by helping -architects engage system owners, program offices, and executive sponsors to deliver value-added plans for -improved mission delivery. Specifically, FSAM includes guidance to help architects establish clear relationships -among strategic goals, detailed business / information management requirements, and measurable performance -improvements within each area of the agency’s enterprise architecture. - -Full funding means appropriations are enacted sufficient in total to complete an asset or useful component (see -definition below) of a capital asset before any obligations may be incurred for the component. Incrementally -funding the planning and acquisition of capital assets or (useful components) without certainty if or when future -funding will be available can result in poor planning, inadequate justification of assets acquisition, higher -acquisition costs, project delays, cancellation of projects, the loss of sunk costs, and inadequate funding to -maintain and operate the assets. Budget requests for full acquisition of capital assets must propose full funding -(see Section 31.5). - -Funding means providing the budgetary resources to plan for, acquire, develop, sustain, or operate an asset. - -Information Technology means any equipment or interconnected system or subsystem of equipment that is used -in the automatic acquisition, storage, manipulation, management, movement, control, display, switching, -interchange, transmission or reception of data or information by an executive agency. Information Technology is -related to the terms Capital Asset, IT Investment, Program, Project, Sub-project, Service, and System. - -Integrated Program Team (IPT) means cross-functional or multidisciplinary group of individuals organized and -collectively responsible for the specific purpose of delivering a project/product/or process to an external or - -Page 4 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -internal customer. Each IPT should include experts in program and project management, resource management, -procurement, and systems engineering, security, and other disciplines, as necessary, to evaluate all aspects of the -project. An IPT must always include, but is not limited to: a qualified fully-dedicated IT program manager, a -contract specialist, an Information technology specialist, a security specialist and a business process owner before -OMB will approve this program investment budget. Key members of the IPT will also be co-located during the -most critical junctures of the program, to the maximum extent possible. Agencies should establish integrated -program team members individual performance goals to hold team members accountable for both individual -functional goals and overall program success. IPT should be defined in a program or an IPT charter. - -Interagency acquisition means the use of the Federal Supply Schedules, a multi-agency contract (i.e., a task order -or delivery order contract established by one agency for use by government agencies to obtain supplies and -services, consistent with the Economy Act, 31 U.S.C. 1535), or a government-wide acquisition contract (i.e., a -task-order or delivery-order contract for information technology established by one agency for government-wide -use operated by an executive agent designated by OMB pursuant to Section 11302(3) of the Clinger Cohen Act of -1996). - -Federal IT Dashboard The purpose of the Dashboard is to provide information on the effectiveness of -government IT programs and to support decisions regarding the investment and management of resources. The -Dashboard is now being used by the Administration and Congress to make budget and policy decisions. - -IT Investment means the expenditure of IT resources to accomplish mission objectives. An IT investment may -include a project or projects for the development, modernization, enhancement, or maintenance of a single IT -asset or group of IT assets with related functionality and the subsequent operation of those assets in a production -environment. While each asset or project would have a defined life cycle, an investment that covers a collection -of assets intended to support an ongoing business mission may not have a defined life cycle. - -IT Program Managers and IT Project Managers are defined by OPM in the Job Family Standard for -Administrative Work, in the Information Technology Group (series 2200 in the Federal Classification and Job -Grading Systems). IT Program Managers will be responsible for major investments and will lead the required -Integrated Program Team for the investment. - -Investment Title as defined in Guidance for Exhibit 53 (See Section 53.8). - -Life-cycle costs includes all investment costs (including government FTE), independent of the funding source, i.e. -revolving fund, appropriated fund, working capital fund, trust fund, etc. (see Capital Programming Guide of OMB -Circular A-11 and OMB Circular A-131). - -Maintenance is the activity necessary to keep an asset functioning as designed during its operations and -maintenance phase of an investment. Maintenance costs include costs needed to sustain an IT asset at the current -capability and performance levels including: corrective hardware/software, voice and data communications -maintenance; replacement of damaged or obsolete IT equipment; and associated overhead costs. Where -appropriate, maintenance activities that follow agency defined project management methodologies should be -managed and reported as projects and reported in Section B of the Exhibit 300B. Examples of maintenance -projects include operating system upgrades, technology refreshes, and security patch implementations. - -Major IT Investment means a program requiring special management attention because of its importance to the -mission or function of the agency, a component of the agency, or another organization; has significant program or -policy implications; has high executive visibility; has high development, operating, or maintenance costs; is -funded through other than direct appropriations; or, is defined as major by the agency’s capital planning and -investment control process. OMB may work with the agency to declare other investments as major investments. -Agencies should consult with your OMB agency budget officer or analyst about what investments to consider as -"major". Investments not considered "major" are "non-major." - -Page 5 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Operations mean the day-to-day management of an asset in the production environment and include activities to -operate data centers, help desks, operational centers, telecommunication centers, and end user support services. -Operational activities are reported through Section C of the Exhibit 300B. Operations costs include the expenses -associated with an IT asset that is in the production environment to sustain an IT asset at the current capability -and performance levels including: Federal and contracted labor costs; and costs for the disposal of an asset. - -Operations and Maintenance means the phase of an asset in which the asset is in operations and produces the -same product or provides a repetitive service. Operations and Maintenance (O&M) is synonymous with “steady -state.” - -Performance-based acquisition management means a documented, systematic process for program management, -which includes integration of program scope, schedule and cost objectives, establishment of a baseline plan for -accomplishment of program objectives, and use of earned value techniques for performance measurement during -execution of the program. This includes prototypes and tests to select the most cost effective alternative during -the Planning Phase, the work during the Acquisition Phase, and any developmental, modification, or upgrade -work done during the Operational/Steady State Phase. For operational/steady state systems, an operational -analysis as discussed in Phase IV of the Capital Programming Guide is required. A performance-based -acquisition (as defined in the Federal Acquisition Regulation 37.101) or contract/agreement with a defined quality -assurance plan that includes performance standards/measures should be the basis for monitoring contractor or in- -house performance of this phase. - -Planning means preparing, developing or acquiring the information you will use to: design the asset; assess the -benefits, risks, and risk-adjusted costs of alternative solutions; and establish realistic cost, schedule, and -performance goals, for the selected alternative, before either proceeding to full acquisition of the capital project or -useful component or terminating the project. Planning must progress to the point where you are ready to commit -to achieving specific goals for the completion of the acquisition before proceeding to the acquisition phase. -Information gathering activities may include market research of available solutions, architectural drawings, -geological studies, engineering and design studies, and prototypes. Planning may be general to the overall -investment or may be a useful component of a project. Depending on the nature of the project, one or more -planning components may be necessary. - -Program for the purposes of this Exhibit 300 is a group of assets that are planned and managed together to -achieve an overall set of related outcomes. IT Investment is frequently used as a synonym for IT program. - -Project is a temporary endeavor undertaken to accomplish a unique product or service with a defined start and end -point and specific objectives that, when attained, signify completion. Projects are undertaken for development, -modernization, enhancement, disposal, or maintenance of an IT asset. Projects are composed of activities. - -Project Manager (PM) Level of Experience is the specific certification or the number of years of direct project -management experience of the PM. Examples of PM certifications include FAC-P/PM, PMI, or other recognized -certifications. - -Risk Management is a systematic process of identifying, analyzing, and responding to risk. It includes -maximizing the probability and consequences of positive events and minimizes the probability and consequences -of adverse events to overall objectives. - -Solution Architecture is a standardized method of identifying business requirements and viable technology -solutions within the context of a single agency’s enterprise architecture, or a multi-agency sector or government- -wide/international architecture. Solution architecture includes current and future views as well as transition plans -at a number of levels of scope that include applications, systems, segments, enterprise, sector, government-wide, -national, and international. The Federal Solution Architecture Methodology (FSAM) is scheduled for release in -October 2011 to provide the repeatable process for doing solution architecture. - -Page 6 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Shared Service Provider is the provider of a technical solution and/or service that supports the business of -multiple agencies using a shared architecture. - -Unique Investment Identifier as defined in the guidance for Exhibit 53. - -Additional budget terms and definitions are included in the Glossary in Appendix J, "Principles of Budgeting for -Capital Asset Acquisitions" and in the guidance for Exhibit 53 (for IT). - -300.5 How will agencies manage information technology capital assets? - -The Capital Programming Guide of OMB Circular A–11, provides guidance on the principles and techniques for -effective capital programming. Appendix J of this part explains the principles of financing capital asset -acquisitions. Section 9b of OMB Circular A–130 establishes additional requirements for enterprise architectures -(EAs), planning and control of information systems and technology investments and performance management. -Agencies must develop and implement a capital programming process to develop their capital asset portfolio, and -must: - -• Evaluate and select capital assets that will support core mission functions performed by the Federal - -Government, and which demonstrate projected returns on investment that are clearly equal to or better -than alternative uses of available public resources. Specifically for IT, the investments should be -informed by, and address gaps in an agency’s Segment Architecture and map to the agency’s strategic -plan; -Initiate improvements to existing assets or acquisitions of new assets only when no alternative private -sector or governmental source can more efficiently meet the need; -Implement IT reforms from the “25 Point Implementation Plan to Reform Federal Information -Technology Management” such as adhering to modular development principles and the requirement to -establish an Integrated Program Team prior to funding an IT investment; - -• - -• - -• Simplify or otherwise redesign work processes to reduce costs, improve effectiveness, and make - -maximum use of commercial services and off-the-shelf technology; - -• Reduce project risk by avoiding or isolating custom designed components, using components that can be -fully tested or prototyped prior to full implementation or production, and ensuring involvement and -support of users in the design and testing of the asset; - -• Structure major planning and acquisition into useful components with a narrow scope and brief duration, -and that make adequate use of competition and appropriately allocate risk between Government and -contractor. The Agency Head must approve or define the cost, schedule, and performance goals for major -acquisitions, and the agency's Chief Financial Officer must evaluate the proposed cost goals; - -• - -• Ensure a continuous linkage with Federal, agency, and bureau EAs, demonstrating such consistency -through alignment with the agency’s Segment Architecture, compliance with agency business -requirements and standards, as well as identification of milestones, as defined in the agency’s EA -transition strategy; -Institute performance measures and management processes monitoring and comparing actual performance -to planned results. Agencies must use a performance-based acquisition management or earned value -management system, based on the ANSI/EIA Standard 748-B, to obtain timely information regarding the -progress of capital investments where appropriate, as defined in the Federal Acquisition Regulations -(FAR). The system must also measure progress towards useful components in an independently -verifiable basis, in terms of cost, capability of the investment to meet specified requirements, timeliness, -and quality. Agencies are expected to achieve, on average, 90 percent of the cost, schedule and -performance goals for major acquisitions. Through the Tech Stat process, agencies must review major -acquisitions not achieving 90 percent of the goals to determine whether there is a continuing need and -what corrective action, including termination, should be taken; - -• Ensure assets that are IT systems conform to the requirements of OMB Circular No. A–130, - -"Management of Federal Information Resources"; - -Page 7 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -• Ensure assets that are financial management systems conform to the requirements of OMB Circular No. - -A–127; - -• Conduct post-implementation or post-occupancy reviews of capital programming and acquisition - -processes and projects to validate estimated benefits and costs and document effective management -practices, i.e., lessons learned, for broader use; - -• Establish oversight mechanisms requiring periodic review of operational capital assets to determine how - -mission requirements might have changed, and whether the asset continues to fulfill ongoing and -anticipated mission requirements, deliver intended benefits to the agency and customers, and meet user -requirements; and - -• Develop and maintain the following documents all of which may be requested by OMB and are subject to - -delivery within ten business days: - -Investment level Alternative Analysis; -o -Investment level Acquisition Plan; -o -o Earned Value Reports on large projects; -Integrated Program Team Charter; -o -o -Investment Charter; -o Project Charters, as appropriate; and -o Risk Management Plan - -300.6 What other requirements do Exhibits 300s fulfill? - -The Exhibit 300A is designed to coordinate OMB’s collection of agency information for its reports to the -Congress required by the Federal Acquisition Streamlining Act of 1994 (FASA Title V) and the Clinger-Cohen -Act of 1996. The business case (Exhibit 300A) for investment should be demonstrated for supporting the mission -statements, long-term goals and objectives, and annual performance plans developed pursuant to the GPRA. In -addition, the Exhibit 300B establishes reporting requirements through the Federal IT Dashboard to ensure the -proper execution of those investments against the established performance plans. - -300.7 What must I report on Exhibits 300A&B and when? - -All information necessary to complete Exhibit 300A&B should already exist as part of the agency's overall -Capital Planning activities and within project and program specific documentation. The materials used to -populate Exhibit 300A&B should be readily available to OMB upon request. - -Exhibit 300As for major IT investments must be submitted to OMB during the period September 12-16, 2011. -The Exhibit 300A should be fully integrated with your agency’s overall budget submission, including Exhibit -53A. All reporting on IT must be submitted electronically via XML feed to OMB’s Federal IT Dashboard -(http://www.itdashboard.gov/). Additional information regarding the submission process will be posted on the IT -Dashboard including business rules for the validation of data and a schema that describes how data should be -transmitted. As with previous submissions to the IT Dashboard, pre-decisional and procurement sensitive -information will not be displayed to the public. - -As prescribed by OMB, agencies should update their Exhibits 53A and 300A to reflect final Presidential -decisions. - -Exhibit 300B for major IT investments should be initially submitted to OMB by September 23, 2011, in order to -establish cost, schedule and performance targets for FY2012. In line with modular development principles, targets -for projects should aim to deliver functionality within six months. For the purposes of target setting, projects -stretching over a longer period should be broken up into segments of six months or less and reported in Tables B1 -and Table B2 in those smaller pieces. Performance targets for ongoing operations will be established in Table C1. - -Page 8 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Following the establishment of FY 2012 targets, regular updates for metrics and activities will be submitted to the -IT Dashboard at least once by the last business day of each month. The periodicity of performance metrics -updates for ongoing operations will vary according to the nature of the metric as indicated in Table C1. - -If agencies request supplemental funds or reallocate funding within their authority, which include changes to the -agency's portfolio and the performance targets, as part of their supplemental request, agencies should submit new -or revised Exhibits 53A (see Guidance on Exhibit 53), 300A, and 300B. - -Multi-Agency Collaboration Investments (must submit both Exhibit 300A and Exhibit 300B) - -The managing partner (lead agency) will take the lead for completing the multi-agency collaboration investment -Exhibit 300, managing it through the lead agency's capital programming and budget process and submitting the -Exhibit 300 to OMB. The managing partner is also responsible for ensuring the Exhibit 300s include all necessary -information from the partner agencies and has been approved by all necessary partner agencies through the -appropriate governance process. The Executive Committee of each initiative/investment should review and -approve the Exhibit 300s. The tracking of the multi-agency approvals, partner agency funding, and related capital -assets (e.g. migration investments, Centers of Excellence, Shared Service Centers, Supporting components) will -be captured via a MAX E-Gov Funding Tool (similar to the one used in FY2010), all Initiative PMs or their -designated leads should have received details about this in the May 10, 2011 email from OMB. - -For IT assets, partner agencies should report their participation in their Exhibit 53 submissions as appropriate -(see Guidance on Exhibit 53 -). Partner agencies should reference the name and UII of the multi-agency Exhibit -300 in the "Investment Description" field of each Exhibit 53 line item related to the multi-agency Exhibit 300. -Partner agencies should also ensure their activities and participation are included in the appropriate sections of -the multi-agency Exhibit 300. The entire Summary of Funding Total for the investment, including funds -provided by partner agency, should be included in the Exhibit 300s. - -Investments that provide a service to other agencies, but do not receive contributions from partner -agencies should still be reported as Multi-Agency Collaboration Investments. - -Investments such as Lines of Business (LoB) that provide a cross-federal governance structure, but no technical -solution should also report as Multi-Agency Collaboration Investments. As governing bodies, they have oversight -responsibilities of the shared service providers that provide technical solutions and should inform and guide their -performance targets. These Exhibit 300s should not duplicate the costs including in the Exhibit 300 reported by -the shared service provider. The Cost and Schedule table should indicate key activities agreed to under the -investment that indicate an agreed upon roadmap strategy; however, only costs directly attributed to the -investment should be included. - -Managing partners providing multi-agency services should ensure that funding is prioritized to accommodate -building and obtaining approval of a shared architecture. Approval should be obtained through a working -governance model that includes partner agencies, customers and OMB. - -OMB may require additional information from partner agencies related to the multi-agency collaboration -investment Exhibit 300. When necessary, OMB will work with the managing partners to coordinate data requests. - -300.8 How will OMB use the Exhibit 300 A&B? - -The Exhibit 300 A&B is one component of your agency’s total budget justification (see Section 51.2). OMB uses -the Exhibit 300A&B to make both quantitative decisions about budgetary resources consistent with the -Administration’s program priorities, and qualitative assessments about whether the agency’s programming -processes are consistent with OMB policy and guidance. OMB will be evaluating all elements of the budget -submission and will communicate the results of these evaluations in the course of the budget process. If -additional supporting information is necessary, OMB will request from agencies the supporting evidence used to -produce the Exhibit 300A&B. All information necessary to complete an Exhibit 300A&B should already exist as - -Page 9 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -part of the agency's overall Information Technology Investment Management activities and within program and -project specific documentation. The materials used to produce the Exhibit 300A&B should be readily available to -OMB upon request. - -Page 10 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Exhibit 300A: IT Capital Asset Summary -Part I: Summary Information And Justification (All Capital Assets) - -Section A: Overview - -1. Investment Name: -2. Unique Investment Identifier (UII) For IT investment only, see -Section 53.9. - -Section B: Investment Detail - -1. Provide a brief summary of the investment, including a brief description of the related benefit to the mission -delivery and management support areas, and the primary beneficiary(ies) of the investment. Include an -explanation of any dependencies between this investment and other investments. - -[Limit: 2500 char] -2. How does this investment close in part or in whole any identified performance gap in support of the mission -delivery and management support areas? Include an assessment of the program impact if this investment isn't -fully funded. -[(Limit:2500 char] -3. For this investment’s technical features, please identify where any specific technical solutions are required -by legislation, in response to audit findings, or to meet requirements from other sources. Where “Yes” is -indicated, provide a brief description of the technical features required, and any citations regarding specific -mandates for these requirements. - -Legislative Mandate -Audit Finding Resolution -Published Agency Strategic Plan -Other Requirement - -Required by Description - -[Yes/No] - -[Limit: 1000 characters] - -4. Provide a list of this investment's accomplishments in the prior year (PY), including projects or useful -components/project segments completed, new functionality added, or operational efficiency achieved. -[Limit: 1000 char] -5. Provide a list of planned accomplishments for current year (CY) and budget year (BY). -[Limit: 2500 char] -6. Provide brief descriptions of outyear (BY+1, BY+2, BY+3, BY+4 and beyond as necessary) budget requests -for this investment. Briefly describe planned projects and/or useful components proposed. Your justification -should address new functionality, systems integration, technology refreshes, efficiencies to be realized, and any -other planned enhancements to existing assets/systems performance or agency operations. -Fiscal Year -BY+1 -BY+2 -BY+3 -BY+4 and beyond - -Description -[Limit: 500 characters] - -Page 11 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -7. Provide the date of the Charter establishing the required Integrated Program Team (IPT) -for this investment. An IPT must always include, but is not limited to: a qualified fully- -dedicated IT program manager, a contract specialist, an information technology specialist, -a security specialist and a business process owner before OMB will approve this program -investment budget. IT Program Manager, Business Process Owner and Contract Specialist -must be Government Employees. - -(date) - -8. IPT Contact Information - -Name - -Phone Number - -Extension - -Email - -[Limit: -Char] - -250 - -[10 digits 0-9 only] - -[Optional: 6 digits -0-9 only ] - -[Limit one email -only] - -IT Program Manager - -Business Process Owner -Contract Specialist -Information Technology -Specialist -Security Specialist - - -Page 12 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Section C: Summary of Funding (Budget Authority for Capital Assets) - -1. Provide the funding summary for this investment by completing the following table. Include funding - -authority from all sources in millions, and round to three decimal places. Federal personnel costs should be -included only in the rows designated “DME … Govt. FTEs” costs and “O&M Govt. FTE” costs and should -be excluded from the other cost breakouts. Cost levels should be consistent with funding levels in Exhibit 53. -For multi-agency investments, this table should include all funding (both managing and partner agency -contributions). - -For years beyond BY+1, please provide your best estimates for planning purposes, understanding that -estimates for out-year spending will be less certain than estimates for BY+1 or closer. - -For lines in the table that ask for changes in your current submission compared to your most recent previous -submission, please use the President’s Budget as your previous submission. When making comparisons, -please ensure that you compare same-year-to-same-year (e.g., 2011 v. 2011). - -Significant changes from the previous submission should be reflected in an updated investment-level -Alternatives Analysis and is subject to OMB request as discussed in Exhibit 300.5. - -PY-1 & -Prior - -PY -2011 - -CY -2012 - -BY -2013 - -BY+1 -2014 - -BY+2 -2015 - -BY+3 -2016 - -BY+4 & -Beyond - -Planning Costs: - -DME (Excluding Planning) -Costs: -DME (Including Planning) -Govt. FTEs: -Sub-Total DME (Including -Govt. FTE) : -O&M Costs: - -O&M Govt. FTEs: - -Sub-Total O&M Costs -(Including Govt. FTE): -Total Cost (Including Govt. -FTE): - -Total Govt. FTE costs: - -# of FTE rep by costs: - -Total change from prior year -final President’s Budget ($) -Total change from prior year -final President’s Budget (%) -Do not enter information for the dark gray cells. - -2. While some investments are consistent with a defined life cycle model (i.e., an initial period of development -followed by a period of primarily operational spending and an identifiable end point), others represent a -collection of ongoing activities and operations with no known terminal point. In the following table, identify -whether or not this investment uses a defined life cycle model (as defined in OMB Circular A-131) and -provide appropriate investment cost information below. - -a. - -Is this investment consistent with a life cycle model defined in OMB Circular -A-131(i.e., an initial period of development followed by a period of primarily -operational spending and an identifiable end point): - -[Y/N] - -Page 13 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -b - +Prepared for submission to JHEP +Causal gravitational waves as a probe of free streaming +particles and the expansion of the Universe +Anson Hook,a Gustavo Marques-Tavares,a,b and Davide Raccoc +aMaryland Center for Fundamental Physics, University of Maryland, College Park, MD 20742, +U.S.A. +bDepartment of Physics and Astronomy, Johns Hopkins University, Baltimore, MD 21218, U.S.A. +cPerimeterInstituteforTheoreticalPhysics,31CarolineSt.N.,Waterloo,OntarioN2L2Y5,Canada +E-mail: hook@umd.edu, gusmt@umd.edu, dracco@perimeterinstitute.ca +Abstract: The low frequency part of the gravitational wave spectrum generated by local +physics, such as a phase transition or parametric resonance, is largely fixed by causality, +offering a clean window into the early Universe. In this work, this low frequency end of the +spectrum is analyzed with an emphasis on a physical understanding, such as the suppressed +productionofgravitationalwavesduetotheexcitationofanover-dampedharmonicoscillator +andtheirenhancementduetobeingfrozenoutwhileoutsidethehorizon. Duetothedifference +between sub-horizon and super-horizon physics, it is inevitable that there will be a distinct +spectral feature that could allow for the direct measurement of the conformal Hubble rate +at which the phase transition occurred. As an example, free-streaming particles (such as +the gravity waves themselves) present during the phase transition affect the production of +super-horizon modes. This leads to a steeper decrease in the spectrum at low frequencies +as compared to the well-known causal k3 super-horizon scaling of stochastic gravity waves. +If a sizable fraction of the energy density is in free-streaming particles, they even lead to +the appearance of oscillatory features in the spectrum. If the universe was not radiation +dominated when the waves were generated, a similar feature also occurs at the transition +between sub-horizon to super-horizon causality. These features are used to show surprising +consequences, such as the fact that a period of matter domination following the production +of gravity waves actually increases their power spectrum at low frequencies. +1202 +naJ +71 +]hp-peh[ +2v86530.0102:viXra + +Contents +| 1 Introduction | | | | | | 1 | +| -------------- | ------------ | ----------------- | ----------------- | --------- | ------------- | --- | +| 2 The | spectrum | of | causality-limited | | gravity waves | 4 | +| 2.1 | The | physical | intuition | | | 4 | +| 2.2 | Radiation | domination | | | | 7 | +| 2.3 | General | equation | of | state | | 7 | +| 3 The | effect | of free-streaming | | particles | | 9 | +| 3.1 | Implications | | of current | N limits | | 13 | +eff +| 4 Effect | of | non-standard | | expansion | histories | 14 | +| ------------ | --------------- | ------------ | -------- | ---------- | ---------- | --- | +| 4.1 | An intermediate | | period | of matter | domination | 14 | +| 4.2 | Using | the GW | spectrum | to measure | w(t) | 18 | +| 5 Conclusion | | | | | | 19 | +A Analytical description of the GW spectrum with an intermediate period of +| matter | domination | | | | | 20 | +| ------ | ---------- | --- | --- | --- | --- | --- | +1 Introduction +With the first direct detection of Gravitational Waves (GWs) [1], we have gained access to a +new probe of the universe. One especially exciting aspect of GWs is that, because they are so +weakly interacting, they can carry information about the early universe in a very clean and +unpolluted form. The future of GWs is bright as there are many proposed future detectors +such as LISA [2], BBO [3], MAGIS [4] and DECIGO [5]. Using stochastic GW background +1028 +searches, these new GW detectors can probe the universe at a time times earlier than +our current best direct probe, the Cosmic Microwave Background (CMB). +There are many exciting possibilities that stochastic GW backgrounds will allows us +to explore. Stochastic GW backgrounds can be produced by a myriad of early universe +phenomena such as inflation [6–9], reheating/preheating [10–15], phase transitions [16] and +topological defects [17, 18] such as cosmic strings or domain walls, or by scalar perturbations +at second order in perturbation theory [19–23]. The detection of such a background would +teach us details about which of these phenomena were active and what were the relevant +| parameters | governing | | their dynamics. | | | | +| ---------- | --------- | --- | --------------- | --- | --- | --- | +Aside from learning about the actual source of GWs, their discovery also teaches us +about the dynamical evolution of the early universe. There exists a vast literature discussing +– 1 – + +these effects and their detectability in various scenarios [16, 24–44] as well as reviews on the +subject [17, 45]. Intuition for this dependence can be most easily developed focusing on a +situation that has been largely overlooked, the low frequency spectrum of a stochastic GW +background produced by a causal source. This is the case for GWs generated by a phase +transition or parametric resonance for example. The low frequency modes that correspond +to length and time scales much longer than the source duration are largely independent of +the details governing the generation of GWs and are instead fixed by causality. The largest +region in causal contact with itself at the time GWs were generated has a size at most of the +order of the Hubble radius, but is generically smaller. These low frequency modes can only +be affected after entering horizon, making the spectrum sensitive to the change of the Hubble +rate as a function of time, or in other words, to the expansion history of the universe. +In this paper we consider the low frequency part of the power spectrum of a stochastic +background of GWs. These modes are typically referred to as the causal part of the spectrum +in order to emphasize the effect of causality, and under standard assumptions have a k3 +scaling [46]. Nonetheless, as we will see later, there are effects other than causality that +determine their dynamics. We will refer to these modes as the “causality-limited” part of the +spectrum. The shape of the spectrum for modes which are causality-limited but were sub- +horizon at generation (sub-horizon causality) is completely independent from the evolution of +the universe1. Causality arguments affect modes which were super-horizon at generation in a +two-fold way. The first factor is that the production of these modes is highly suppressed. The +reason for this is that they have frequencies ω (cid:28) H (H being the Hubble rate) at the time +of production, and therefore behave like an over-damped harmonic oscillator, making such +modes very hard to excite. As a result, this effect decreases the power at low frequencies. +The second factor is that once excited, their amplitude stays constant while they are super- +horizon, ratherthandecreasing, leadingtoanoverallincreaseinthepoweratlowfrequencies. +The competition between these two effects determines the shape of the low frequency tail of +GWs. +More mathematically, the physical intuition that we will develop is that the initial con- +ditions for causality-limited modes immediately after generation are h = 0 while h˙ (cid:54)= 0, +ij ij +in much the same way that the initial condition of a hammer hitting a string creates no +displacement but a large velocity. If the GW is sub-horizon, ω (cid:29) H, the mode obtains an +amplitude h ∼ h˙/ω. If the mode is super-horizon, friction is important and the mode obtains +a smaller than expected amplitude h ∼ h˙/H. After this initial displacement, over-damped +modes are frozen in place until Hubble crossing at which point they become the standard +under-damped modes seen in GW detectors. +There are several interesting observations that result from an analysis along the previous +lines. The first is that generically there is a difference between sub-horizon and super-horizon +causality. As a result, there will generically be a feature at horizon crossing. Finding such +a feature would give a model-independent direct measurement of the conformal Hubble rate +1They are however sensitive to the total redshift factor between then and now. +– 2 – + +at the time when GWs were created. Upon specifying an expansion history, it is possible to +deduce the Hubble rate, and hence the temperature, at which the GWs were generated. We +willdiscusstwoscenariosinwhichGWmodesthataresuper-horizonorsub-horizon(butata +wavelength larger than the correlation length of the source) at the time of generation behave +differently. +In the first, we study how super-horizon GWs depend on the equation of state parameter +w. We show that while sub-horizon causality forces the spectrum of GWs to fall as k3, that +super-horizon modes fall as k(1+15w)/(1+3w). Thus, as long as w (cid:54)= 1/3, the scaling of GWs +changes between modes that are sub-horizon at the time of generation versus super-horizon. +The change in behavior between sub-horizon modes and super-horizon modes during matter +domination (MD) was previously studied in Ref. [47] and followed up in Refs. [48–50]. We +improve upon these results by developing simple physical intuition for these effects and by +generalizing them to arbitrary equations of state. +Forthesecondscenariowelookattheeffectoffree-streamingparticlesontheproduction +and propagation of GWs. This was originally studied in Ref. [51] for the case where some +species of particles started free-streaming after GWs were produced, and it showed that this +leads to a constant suppression in the amplitude as modes entered the horizon. We focus on +thecontrastingcasewherefree-streamingparticleswerealreadypresentduringthecreationof +theGWs. Wefindthatsub-horizonmodeshavethestandardk3 scaling. Super-horizonmodes +have a scale dependent suppression, and for a sufficiently large fraction of free-streaming +particles there is a a surprising oscillatory feature on top of a suppressed scaling of k4. These +effects come about because in this scenario super-horizon GWs are not frozen out but instead +slowly roll in the potential of the free-streaming particles, suppressing the power in GWs. +This effect is always present, even if at a small level, as the high frequency part of the GW +spectrum itself acts as free-streaming particles. +Another interesting result we find is that an intermediate period of matter domination +after the GW generation actually increases the power at low frequencies. There are well mo- +tivated scenarios where various effects due to reheating generate GWs [10, 11]. In the typical +scenario where reheating occurs at high scales, only the causal part of the GW spectrum is +accessibleasthepeakfrequenciesaretoohigh. Anadditionalfeatureofmanyofthesemodels +is that they generically imply matter domination after the production of GWs. This effect +was previously considered to be undesirable as it dilutes the peak power of the GW. How- +ever, for the low frequencies that are experimentally visible, the effect of a period of matter +domination is to increase the visibility of these scenarios, making it a desirable feature. +In Sec. 2, we discuss the physical intuition behind causality-limited GWs and compare +our physically motivated approximations against exact results. In Sec. 3, we study the effects +of free streaming particles on causal GWs. In Sec. 4, we study how the frequency spectrum +of gravitation waves changes as the universe transitions between matter, radiation, and sub- +horizon modes. Finally, we conclude in Sec. 5. +– 3 – + +| 2 The | spectrum | of causality-limited | | | gravity | waves | | | +| ------- | -------- | -------------------- | --- | --- | ------- | ----- | --- | --- | +| 2.1 The | physical | intuition | | | | | | | +Throughoutthisworkwewillbeexploring“causality-limited”GWs,thatwedefineasfollows. +Consider a source that is active for a short amount of time 1/β. The causality-limited part +of the GW spectrum consists of the waves whose period and wavelength are much longer +than the source’s temporal and spatial correlations respectively. In other words, λ (cid:29) 1/β, +where λ is the wavelength and 1/β is the duration of the process generating the GWs. The +standardexampleofsuchcausality-limitedGWsisthelowfrequencypartoftheGWspectrum +| generated | by a cosmological | | phase | transition. | | | | | +| --------- | ----------------- | --- | ----- | ----------- | --- | --- | --- | --- | +Using conformal time τ and conformal Hubble rate H = a(cid:48)/a, the equation of motion for +a comoving mode k of the graviton h is (we follow a notation close to the one of [46]) +ij +32πGρ +| | | ∂2h | | +k2h | a2 | | | | +| --- | --- | --- | ------- | ------ | ---- | --- | ----------- | ----- | +| | | | ij +2H∂ | τ h ij | ij = | Π | ij ≡ J ij , | (2.1) | +| | | τ | | | | 3 | | | +where Π is the dimensionless anisotropic stress of the sector generating the GWs, ρ is the +ij +energydensityofthatsector, andthenormalizationofΠ followsthatof[46]withaprefactor +ij +of 32πG/3. We will project both h and J onto the two independent (+,×) polarisations, +| | | | | ij | ij | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | +and assume that the respective amplitudes J and h of the two polarisations are equal. We +will also assume that the GWs are produced on a time scale fast compared to Hubble so +that we can approximate H as a constant over its production and the time over which Π is +non-zero to be small. The time at which the phase transition occurs will be denoted by τ . +(cid:63) +The solution to Eq. (2.1), assuming h(τ) = 0 for τ < τ , can be found using Green’s +(cid:63) +| functions | to be | | | | | | | | +| --------- | ----- | --- | ------------------------- | --- | -------- | --------- | -------- | --- | +| | | | (cid:90) e−H(τ−τ(cid:48)) | | | | | | +| | | | | | (cid:16) | (cid:112) | (cid:17) | | +h(k,τ) = dτ(cid:48)√ sin (τ −τ(cid:48)) k2−H2 J(k,τ(cid:48)). (2.2) +k2−H2 +We will be interested in the initial conditions right afterwards τ(cid:48) = τ +(cid:15) assuming that J +(cid:63) +occurs fast so that we can take J(k,τ(cid:48)) = J (k)δ(τ(cid:48)−τ ), and we will drop the k dependence +| | | | | | (cid:63) | (cid:63) | | | +| --- | --- | --- | --- | --- | -------- | -------- | --- | --- | +from here on. The approximation of the source as a Dirac delta simplifies the analysis and +holdsformodeswhoseperiodismuchlongerthanthedurationofthesource, andthuscannot +resolvedetailsofthetimedependence. Therangeofmodesforwhichtheapproximationholds +will vary depending on the exact dynamics of what sources the gravitational wave. Using this +| approximation | and | Eq. (2.2), | we | find that | | | | | +| ------------- | --- | ---------- | ---------- | --------- | ------- | ---------- | -------- | ----- | +| | | h(k,τ | +(cid:15)) | = 0, | ∂ h(k,τ | +(cid:15)) | = J . | (2.3) | +| | | | (cid:63) | | τ | (cid:63) | (cid:63) | | +Afterwards, we have the second order differential equation and initial conditions +| | | | | ∂2h+2H∂ | h+k2h | = 0, | | | +| --- | --- | --- | --- | ------- | ----- | ---- | --- | --- | +| | | | | τ | τ | | | | +(2.4) +| | | | h(k,τ | ) = 0, | ∂ h(k,τ | ) = J | . | | +| --- | --- | --- | ----- | -------- | ------- | ----------------- | --- | --- | +| | | | | (cid:63) | τ | (cid:63) (cid:63) | | | +| | | | | | – 4 – | | | | + +TheobservablerelevantforGWdetectorssuchasLISAisdΩgw/dlogk. Thisisobtained +using +1 +| | | | | (cid:88) | | (cid:10) h(cid:48) | r(x,τ)h(cid:48) | s(x,τ) | (cid:11) | | | +| --- | --- | -------- | --- | -------- | --- | ------------------ | --------------- | ------ | -------- | --- | ----- | +| | | ρgw(x,τ) | | = | | | | | , | | (2.5) | +| | | | | | 32π | Ga2 | ij | ij | | | | +r,s=+,× +which holds only when the relevant modes have entered the horizon at late times and started +| oscillating. | We will | also | take | | | | | | | | | +| ------------ | ------- | ---- | -------------------- | --- | -------- | --------------------- | --- | ------ | --- | --- | ----- | +| | | | (cid:10) | | (cid:11) | | | | | | | +| | | | h(k,τ)h(k(cid:48),τ) | | ≡ | (2π)3δ3(k−k(cid:48))P | | (k,τ), | | | (2.6) | +h +with P (k,τ) the dimensionful power spectrum of GWs. Fourier transforming gives +h +k5P +| | | | dΩgw | | 1 dρgw | 1 | | (k,τ) | | | | +| --- | --- | --- | ---- | --- | ------ | --- | --- | ----- | --- | --- | ----- | +| | | | | = | | = | h | , | | | (2.7) | +c2(2π)3a2G +| | | | dlogk | | ρ c dlogk | ρ | | | | | | +| --- | --- | --- | ----- | --- | --------- | --- | --- | --- | --- | --- | --- | +where ρ = 3H2/(8πG) is the critical energy density. In the second equality we use the c - -Describe why the investment is not consistent with life cycle model -management defined in OMB Circular A-131, and explain how you adapted -your alternatives analysis for this investment? -(Where an agency uses a cost model other than the lifecycle cost model, defined by OMB -Circular A-131, responses from 2c to 2h below should reflect the alternative concept.) -Provide information on what cost model this investment is using and how costs -are captured for what years: - -d. What year did this investment start (use year—e.g., PY-1=2010) - -e. -f. - -g. - -h. - -What year will this investment end (use year—e.g., BY+5=2018) -Estimated Total DME cost (including planning) for the investment life cycle or -other cost model (excluding FTE) -Estimated Total O&M cost the investment life cycle or other cost model -(excluding FTE) -Estimated total Govt. FTE Cost for the investment life cycle or other cost -model - -[Limit: 1000 chars] -(Required if a is N) - -[Limit: 1000 chars] -(Required if a is N) -(Required if a is Y) - -(Required if a is Y) -(Required) - -(Required) - -(Required) - -3a. If the funding levels have changed from the FY 2012 President’s Budget request for PY -or CY, briefly explain those changes: - -[Limit: 500 chars] - -Page 14 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - - Section D: Acquisition/Contract Strategy (All Capital Assets) - -1. Complete or update the table to display all prime contracts (or task orders) for awarded or open solicitations -for this investment (sub-award details is not required). Contracts and/or task orders that have “Ended” should not -be included in the table. Contracts in open solicitation should provide estimated data for all fields (for “Total -Contract Value” the estimated base contract costs and all anticipated option years). Data definitions can be found -at www.usaspending.gov/learn#a2. - -For specifics, please see notes 1 and 2 below the table. - -Field - -Data Description - -Optional for -Awarded Contracts - -Contract/ -Task Order - -(1) Awarded, (2) Pre-award Post-solicitation - -Required only if the contracting agency is different than the -agency submitting the exhibit. Use agency 4 digit code as -used in FPDS. -The unique identifier for each contract, agreement or order. -See http://www.usaspending.gov/learn#a2 - -Required only for IDVs. See -http://www.usaspending.gov/learn#a2 - -This is a code for an agency, but it does not necessarily -represent the agency that issued the contract. Instead, it -serves as part of the unique identification for Federal -Procurement Data System IDV records. -Identifier used to link transactions to solicitation -information. See http://www.usaspending.gov/learn#a2 -Y/N - -Total Value of Contract including base and all options. -Complete using dollars to two decimal places. - -See FAR Part 16. Can be fixed price, cost, cost plus, -incentive, IDV, time and materials, etc - -Y/N Indicates whether the contract is a PBSA as defined -by FAR 37.601. A PBSA describes the requirements in -terms of results rather than the methods of performance of -the work. - -Actual or expected Start Date of Contract/Task Order, the -date that the parties agree will be the starting date for the -contract’s requirements. (YYYY-MM-DD) -(YYYY-MM-DD) - -Contract -Status -Contracting -Agency ID - -Procurement -Instrument -Identifier -(PIID) -Indefinite -Delivery -Vehicle (IDV) -Reference ID -IDV Agency -ID - -Solicitation -ID1 -EVM -Required -Ultimate -Contract -Value1 -Type of -Contract/Tas -k Order -(Pricing) 1 -Is the -contract a -Performance -Based Service -Acquisition -(PBSA)? 1 -Effective -date1 - -Actual or -expected End -Date of -Contract/Tas -k Order1 - -Y - -Y - -Y - -Y - -Y - -Y - -Y - -Y - -Page 15 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Extent -Competed1 - -Contract -Description - -(A) Full and open competition (B) Not available for -competition (C) Not competed (D) Full and open -competition after exclusion of sources (E) Follow-on to -competed action (F) Competed under simplified acquisition -procedures (G) Not competed under simplified acquisition -procedures (CDO) Competitive Delivery Order (NDO) Non- -competitive Delivery Order -A brief description of the goods or services bought (for an -award) or that are available (for an -IDV).www.usaspending.gov/learn?tab=FAQ#2 - -Y - -Y - -1Assuming the PIID or IDV PIID match with USAspending.gov, these data elements will be automatically populated for awarded IT acquisitions -2Assumingthe PIID, IDV PIID, or Solicitation number match with USAspending.gov or FedBizOpps (fbo.gov) this data will be auto populated for awarded -and pre-award, post-solicitation IT acquisitions. - -2. If earned value is not required or will not be a contract requirement -for any of the contracts or task orders above, explain why: - -[char 2500] - -Page 16 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Exhibit 300B: Performance Measurement Report - -The Exhibit 300B is used to provide OMB current fiscal year (FY2012) investment performance data for major -investments. Include in this exhibit, at a minimum, all activities and operations that started in a previous fiscal -year (PY and earlier) and have not completed by the beginning of the current year as well as activities and -operations that are scheduled to start in the current fiscal year. Provide an up-to-date copy of this exhibit to OMB -at least once a month. - -UNDERSTANDING THE EXHIBIT 300B - -On the Exhibit 300B, investments are described as: - -• - -Investment - -o Projects - -• Activities - -o Operations - -Report information about these areas in the following Exhibit 300B sections: - -A.1: General Information. Enter basic information about the major investment. -B.1: Projects. Identify all of the investment’s projects with activities occurring in the current fiscal year. -B.2: Activities. Outline the activities that are performed to achieve the outcome of each project. -B.3: Project Risk. Identify all significant risks to each project’s success. -C.1: Operational Performance Information. Identify performance targets for evaluating operations. -C.2: Operational Risk. Identify all significant risks to investment achieving operational performance targets. - -SECTION A: GENERAL INFORMATION - -1. Investment Name: Agency-provided name of investment, consistent with Exhibit 53. -2. Investment UII: Agency-provided unique investment identifier, consistent with Exhibit 53. - -SECTION B: PROJECT EXECUTION DATA - -Section B addresses planning, DME and significant maintenance projects for the investment. - -In Table B.1, report, at a minimum, all projects with any activities that started in a previous fiscal year (PY and -earlier) and have not completed by the beginning of the current year as well as activities that are scheduled to start -in the current fiscal year, including planning, DME, and maintenance projects. This information should be -updated at least once every month. Include the following data in Table B.1: - -Project -ID - -Project -Name - -Project -Description - -Project -Start -Date - -Project -Completi -on Date - -Table B.1 -Project -Lifecycle -Cost - -PM -Name - -PM Level -of -Experience - -PM -Phone - -[10 digits -0-9 only] - -PM -Extension -[Optional: -6digits 0-9 -only ] - -PM -Email -[Limit one -email -only] - -1. Project ID: An agency-specified number that uniquely identifies the project within this investment. -2. Project Name: Name used by agency to refer specifically to this project. -3. Project Description: Description of project functionality or purpose. -4. Project Start Date: Date of actual start of in-progress projects or planned start of projects which have not - -yet begun (may be before current fiscal year or activities listed in activities table B.2). - -Page 17 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -5. Project Completion Date: Planned date of completion of in-progress projects or actual completion date of -projects which have completed (may be after budget year or of completion date of activities listed in -activities table B.2). - -6. Project Lifecycle Cost: Enter the total cost of all activities related to this project as described in OMB - -Circular No. A-131. (in $ millions) - -7. PM Name: Name of project manager responsible for the success of this project. -8. PM Level of Experience: The years of applicable experience or the status of certification. Available - -selections include: - -1) FAC-P/PM(DAWIA-3)– Senior -2) FAC-P/PM(DAWIA-2)– Mid-Level -3) FAC-P/PM(DAWIA-1)– Entry Level -4) Other certification with 4 or more years PM experience (within the last five years) -5) Other certification with between 2 and 4 years PM experience (within the last five years) -6) Other certification with less than two years PM experience (within the last five years) -7) No certification, but with 4 or more years PM experience (within the last five years) -8) No certification, but with between 2 and 4 years PM experience (within the last five years) -9) No certification, but with less than two years PM experience (within the last five years) - -9. PM Phone: Phone number of project manager responsible for the success of this project. -10. PM Email: Email address of project manager responsible for the success of this project. - -In Table B.2, describe, at a minimum, all activities occurring during the current fiscal year. This table should be -updated once a month at a minimum. In line with modular development principles, activities should be structured -to provide usable functionality in measureable segments that complete at least once every six months or more -often, as described in the 25-Point Implementation Plan to Reform Federal IT. - -A -Project ID - -B -Activity -Name - -C -Activity -Description - -D -Structure -ID - -E -Key -Deliverable/ -Usable -Functionality -[Key -Deliverable, -Usable -Functionality, -N/A] - -Table B.2 - -F -Start Date -Planned - -G -Start Date -Projected - +simplification valid at late times that h(cid:48) = kh, and where an additional k3 comes from the +phase space factor. We can then obtain the k scaling of dΩgw/dlogk by taking k5 and +| multiplying | by the | k scaling | of | h2. | | | | | | | | +| ----------- | ------ | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +The observable of interest depends on h(k,τ), which comes from solving Eq. (2.4). From +that we can see that there are two contributions to the k scaling of (cid:104)hh(cid:105), the k dependence +of the source J (k) and propagation effects. Here is where causality plays a central role: as +(cid:63) +long as we are interested in wavelengths much longer than the typical spatial correlation of +the sources, the Fourier transformed source J ≈ constant (see e.g. [46]), and hence the k de- +(cid:63) +modes.2 +pendence coming from the source is trivial for long wavelength Assuming correlation +k−1 +lengths for the source, , that are small compared to the horizon size, we can separate +source +thesolutionsofthecausality-limitedmodesintotworegimes, k (cid:29) k (cid:29) H (sub-horizon) +| | | | | | | | | source | | (cid:63) | | +| --- | --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | --- | +and k (cid:28) H (super-horizon). Note that in the case of the phase transitions, the existence of +(cid:63) +a causal sub-horizon regime depends on whether all the sources of GWs (including in partic- +ular the sound wave contribution) have a short duration compared to the Hubble time (see +| e.g. [52–55] | for recent | studies | on | this | topic). | | | | | | | +| ------------ | ---------- | ------- | --- | ---- | ------- | --- | --- | --- | --- | --- | --- | +Sub-horizon regime k (cid:29) H : These modes are under-damped by definition. Thus, the +(cid:63) +standard approximation is that of a frictionless solution rescaled by a(τ )/a(τ) to take care +(cid:63) +of Hubble friction as can be seen by using the WKB approximation. This is easily shown to +be +| | | | a(τ | )J | | | | | | | | +| --- | --- | --- | ---- | ----------------- | ------ | --------------- | --- | ---------------------- | --- | --- | ----- | +| | | | | (cid:63) (cid:63) | | | | | | | | +| | | | h ≈ | | sink(τ | −τ (cid:63) ) , | (k | (cid:29) H (cid:63) ). | | | (2.8) | +| | | | a(τ) | k | | | | | | | | +2Moreexplicitly,ifforseparationsxlargerthanalengthscaleλthetwo-pointfunctionP (x)ofthesource +Π +| | | | | | (cid:82) | | (cid:82)λdx4πxsin(kx)P | | | (cid:82)λdx4πx2P | | +| --- | --- | --- | --- | --- | -------- | --- | ---------------------- | --- | --- | ---------------- | --- | +vanishes, then its Fourier transform P (k) = d3xeik·xP (x) = (x) ≈ (x) +| | | | | Π | | Π | 0 | | k Π | 0 | Π | +| ------------- | ----------- | --- | ------ | ------------- | --- | --- | --- | --- | --- | --- | --- | +| is a constant | independent | of | k when | k(cid:28)λ−1. | | | | | | | | +– 5 – + +From this, we immediately see that sub-horizon causality automatically gives +dΩgw +| | | | | | | ∝ k3, | | (k (cid:29) | H ) | (2.9) | +| --- | --- | --- | --- | --- | --- | ----- | --- | ----------- | --- | ----- | +(cid:63) +dlogk +| regardless | of what | | the equation | | of state | of the | universe | | is. | | +| ---------- | ------- | --- | ------------ | --- | -------- | ------ | -------- | --- | --- | --- | +Super-horizon regime k (cid:28) H : There are two competing factors controlling the dy- +(cid:63) +namics of the k (cid:28) H modes. The first is that one is attempting to excite an over-damped +(cid:63) +harmonic oscillator. This effect suppresses the power in these modes. The second is that +super-horizon modes are frozen in place until they enter the horizon, increasing the rela- +tive power in these modes. The competition between these two effects is what controls the +| behavior | of k | (cid:28) H | modes. | | | | | | | | +| -------- | ---- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | +(cid:63) +| Very | quickly | (roughly | | after | a single | e-fold) | the | initial | conditions | | +| ---- | ------- | -------- | --- | ----- | -------- | ------- | ------- | -------- | ---------- | ------ | +| | | | | | h(k,τ ) | = 0 | ∂ h(k,τ | | ) = J | (2.10) | +| | | | | | (cid:63) | | τ | (cid:63) | (cid:63) | | +evolve into +J +(cid:63) +| | | | | h(k,τ | (cid:63) ) ∼ | | , ∂ | τ h(k,τ | (cid:63) ) ≈ 0 | (2.11) | +| --- | --- | --- | --- | ----- | ------------ | --- | --- | ------- | -------------- | ------ | H -Start Date -Actual - -I -Completion -Date -Planned - +(cid:63) +due to Hubble friction, and remain frozen while the mode is super-horizon. The O(1) number +outfrontdependsontheexactequationofstateoftheuniverse. Thisscalingcanbeconfirmed +∂2h +by solving ij +2H (cid:63) ∂ τ h ij = 0 with a non-zero initial velocity. Notice that the amplitude +τ +of the oscillation is suppressed compared to what one would expect based on the amplitudes +for sub-horizon modes: ∼ J (cid:63) /k. This suppression is the effect of attempting to excite an +| over-damped | | harmonic | oscillator. | | | | | | | | +| ----------- | --- | -------- | ----------- | --- | --- | --- | --- | --- | --- | --- | +At this point, the mode is frozen out and its amplitude is constant horizon entry, which +| occurs | at a conformal | | time | τ k given | by | | | | | | +| ------ | -------------- | --- | ---- | --------- | --- | --- | --- | --- | --- | ------ | +| | | | | | | H(τ | ) ≈ | k. | | (2.12) | +k +Afterthemodeentersthehorizon,theGWcanbeapproximatedascompletelyunder-damped +| and redshifts | | away | as a(τ | )/a(τ) | giving | | | | | | +| ------------- | --- | ---- | ------ | ------ | ------ | --- | --- | --- | --- | --- | +k +| | | | | | a(τ ) | J | | | | | +| --- | --- | --- | --- | --- | ----- | -------- | --- | --- | ---------------------- | ------ | +| | | | | | k | (cid:63) | | | | | +| | | | | h ≈ | | sinkτ | , | (k | (cid:28) H (cid:63) ). | (2.13) | +| | | | | | a(τ) | H | | | | | +(cid:63) +We have been cavalier about the phase information just putting in sinkτ, as the phase is +unimportant for computing the power due to averaging over many oscillations. +– 6 – + +| 2.2 Radiation | | domination | | | | | | | | | | | +| ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +It is well known that in radiation domination (RD) there is a simple solution for the GWs. +During RD, a ∼ τ and H = a(cid:48)/a = 1/τ. We can solve Eq. (2.4) exactly in this case and find +| | | | | J τ | | | a(τ )J | | | | | | +| --- | --- | --- | --- | ----------------- | ------ | ------------- | -------- | -------- | --- | ------------ | --- | ------ | +| | | | | (cid:63) (cid:63) | | | (cid:63) | (cid:63) | | | | | +| | | | h = | | sink(τ | −τ (cid:63) ) | = | sink(τ | −τ | (cid:63) ) . | | (2.14) | +| | | | | kτ | | | a(τ) | k | | | | | +Thus, the full solution is given by the solution in a friction-less universe times a red-shifting +| factor of | a(τ )/a(τ). | | | | | | | | | | | | +| --------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +(cid:63) +We now show that the physical intuition given in Sec. 2.1 replicates this exact solution. +The sub-horizon modes given in Eq. (2.8) have the same scalings as the exact solution. The +super-horizon modes given in Eq. (2.13) simplify using k = H(τ ) = H a(τ )/a(τ ) to give +| | | | | | | | | | k | (cid:63) | (cid:63) k | | +| --- | --- | --- | --- | --- | ------- | ---------- | ---- | -------------------- | --- | -------- | ---------- | ------ | +| | | | | | a(τ k ) | J (cid:63) | a(τ | (cid:63) )J (cid:63) | | | | | +| | | | | h ≈ | | sinkτ | ∼ | sinkτ | | | | (2.15) | +| | | | | | a(τ) | H | a(τ) | k | | | | | +(cid:63) +which is again the same scaling as the exact solution, confirming our physical intuition. +| From | these | results | we | can calculate | | how | dΩgw/dlogk | scales | with | k: | | | +| ---- | ----- | ------- | --- | ------------- | ---- | --- | ---------- | ------ | ---- | ---------- | --- | ------ | +| | | | 1 | | dΩgw | | | | | | | | +| | | h(k) | ∼ | ⇒ | | ∼ | k5h(k)2 | ∼ k3, | (k | (cid:28) H | ). | (2.16) | +(cid:63) +| | | | k | | dlogk | | | | | | | | +| --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | +This reproduces the famous fact that causality in a RD universe leads to the spectrum of +| GWs falling | off | as k3 | in the | small | k limit. | | | | | | | | +| ----------- | --- | -------- | ------ | -------- | -------- | --- | --- | --- | --- | --- | --- | --- | +| 2.3 General | | equation | | of state | | | | | | | | | +We now repeat our exercise for a general equation of state p = wρ. In this case, we have +n +| | | | | | a | ∝ τn, | H = | , | | | | (2.17) | +| --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | ------ | +τ +with n = 2/(1+3w), so that for radiation (matter) domination we have n = 1 (n = 2). In +| this case | the exact | | solution | of Eq. | (2.4) | is | | | | | | | +| --------- | --------- | --- | -------- | ---------- | -------------------- | ----- | ------------ | ------ | ----- | --- | --------- | ------ | +| | | | | (cid:16)τ | (cid:17)n−1(cid:104) | | | | | | (cid:105) | | +| | | | kτ2 | (cid:63) | | | | | | | | | +| | h(k,τ) | | = J | | | j (kτ | )y | (kτ)−j | (kτ)y | | (kτ ) , | (2.18) | +| | | | (cid:63) | (cid:63) τ | | n−1 | (cid:63) n−1 | | n−1 | n−1 | (cid:63) | | +where j (x) and y (x) are spherical Bessel functions of first and second kind respectively. +| n | | n | | | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +While this is the exact solution, it is useful to consider two limits. The first is the limit of +| sub-horizon | modes: | | | | | | | | | | | | +| ----------- | ------ | --- | ------ | --- | --------------- | -------- | -------- | ------ | ---------- | -------- | --- | ------ | +| | | | | | a(τ (cid:63) )J | (cid:63) | | | | | | | +| | | | h(k,τ) | ≈ | | sink(τ | −τ | ) , (k | (cid:29) H | ). | | (2.19) | +| | | | | | | | (cid:63) | | | (cid:63) | | | +a(τ) k +| | | | | | | – | 7 – | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +The second is the limit of super-horizon modes at production after they enter the horizon, +| H (cid:29) | k (cid:29) H: | | | | | | | | | | +| ---------- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +(cid:63) +(cid:0) 1(cid:1) +| | | | Γ n− | J τ | (cid:18) 2 | (cid:19)n (cid:16) | nπ(cid:17) | | | | +| --- | --- | ------ | ---- | ----------- | ---------- | ------------------ | ---------- | ---- | ---------------------- | ------ | +| | | | | √2 (cid:63) | (cid:63) | | | | | | +| | | h(k,τ) | ≈ | | | cos | kτ − | , (k | (cid:28) H (cid:63) ). | (2.20) | +| | | | | 2 π | kτ | | 2 | | | | +Again, we will show that the physical intuition given in Sec. 2.1 replicates this exact +solution in the two limits. The sub-horizon solution in Eq. (2.19) has the same parametric +behavior found in Eq. (2.8). For the super-horizon case, note that Eq. (2.13) for a general +| equation | of state | can | be simplified | using | | | | | | | +| -------- | -------- | --- | ------------- | ----- | --- | --- | --- | --- | --- | --- | +(cid:19)1 +| | | | | | | (cid:18) a(τ ) | | | | | +| --- | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | +(cid:63) n +| | | | | H(τ | ) = H | | = | k. | | (2.21) | +| --- | --- | --- | --- | --- | ----- | -------------- | --- | --- | --- | ------ | +| | | | | k | | (cid:63) a(τ ) | | | | | +k +| After some | algebra, | | we find | that | | | | | | | +| ---------- | -------- | --- | ------- | ----- | ------------ | --- | --------------------- | --- | --- | ------ | +| | | | | a(τ | ) J (cid:63) | | J (cid:63) τ (cid:63) | | | | +| | | | | h ≈ k | sinkτ | ∼ | sinkτ | , | | (2.22) | +(kτ)n +| | | | | a(τ) | H (cid:63) | | | | | | +| --- | --- | --- | --- | ---- | ---------- | --- | --- | --- | --- | --- | +reproducing the parametric behavior of the exact solution for the super-horizon modes. +The scaling of dΩgw/dlogk can be directly obtained from the scaling of the solutions. +| For sub-horizon | | modes | k (cid:29) | H | | | | | | | +| --------------- | --- | ----- | ---------- | --- | --- | --- | --- | --- | --- | --- | +(cid:63) +dΩgw +| | | | | ∼ k5h(k)2 | | ∼ k3, | (k | (cid:29) H ). | | (2.23) | +| --- | --- | --- | --- | --------- | --- | ----- | --- | ------------- | --- | ------ | +(cid:63) +dlogk +As expected, as long as the wavelength of the mode is sufficiently larger than the distances +over which the sources are spatially correlated, sub-horizon causality forces these modes to +| fall as | k3. For | the super-horizon | | modes, | we | have | | | | | +| ------- | ------- | ----------------- | --- | ------ | --- | ---- | --- | --- | --- | --- | +dΩgw +| | | | | ∼ k5h(k)2 | | ∼ k5−2n, | (k | (cid:28) H ). | | (2.24) | +| --- | --- | --- | --- | --------- | --- | -------- | --- | ------------- | --- | ------ | +(cid:63) +dlogk +Weseethatthereismorepoweratlowk foranyequationofstatewithw < 1/3. Inparticular, +| for the | case | of MD, | n = 2, we | find that | | | | | | | +| ------- | ---- | ------ | --------- | --------- | --- | --- | --- | --- | --- | --- | +dΩgw +| | | | | ∼ | k | Matter | domination. | | | (2.25) | +| --- | --- | --- | --- | --- | --- | ------ | ----------- | --- | --- | ------ | +dlogk +From the discussion above we find that there is a distinct kink in the spectrum at horizon +crossing as the modes transition from sub-horizon to super-horizon causality. This kink in +the spectrum occurs for all cosmologies, except for exact RD where the spectrum scales as +k3 for long-wavelength modes. Identification of this generic feature in a signal allows for a +model independent measurement of the value of conformal Hubble at which the GWs were +generated. +We can use the exact analytic results of Eq. (2.18) to obtain dΩgw/dlogk. To do this, +– 8 – + +| we take | the late | time | limit | of Eq. | (2.18) | and find | | | | | | +| ------- | -------- | --------- | ------------------ | ------ | ------ | ------------------- | --- | --- | -------- | ------------------- | --- | +| | | (cid:16)τ | (cid:17)n(cid:104) | | | (cid:16) nπ(cid:17) | | | (cid:16) | nπ(cid:17)(cid:105) | | +(cid:63) +| h(k,τ) | ≈ J | τ | | j (kτ | )sin | kτ − | −y | (kτ )cos | kτ | − . | (2.26) | +| -------- | -------- | ----------------- | ----------------- | --------------------------- | -------- | ------------ | --- | ------------ | ------- | ---- | ------ | +| | | (cid:63) (cid:63) | τ | n−1 | (cid:63) | 2 | | n−1 (cid:63) | | 2 | | +| From the | equation | above | we | quickly | see | that | | | | | | +| | | | | (cid:16)τ (cid:17)2n(cid:0) | | | | | | | | +| | | |h|2 | J2τ2 | (cid:63) | | )|2+|y | | )|2(cid:1) | sin2(kτ | | | +| | | = | | | |j n−1 | (kτ (cid:63) | n−1 | (kτ (cid:63) | | +φ), | (2.27) | +| | | | (cid:63) (cid:63) | τ | | | | | | | | +where φ is a constant phase factor. Finally, we arrive at an exact expression for the energy +| in GWs | (after averaging | | over | the | oscillations) | | | | | | | +| ------ | ---------------- | --- | ---- | ------------------------ | ------------- | --------- | ---------- | --- | ---------- | --- | ------ | +| | | | dΩgw | | | k5(cid:0) | | | )|2(cid:1) | | | +| | | | | ∝ k5(cid:104)h2(cid:105) | ∝ | |j | (kτ )|2+|y | | (kτ | | (2.28) | +| | | | | | | n−1 | (cid:63) | n−1 | (cid:63) | | | +dlogk +where we have neglected unimportant proportionality constants. In Fig. 1 we use these exact +analytic results to show how sub-horizon scaling becomes super-horizon scaling for various +| equations | of state. | | | | | | | | | | | +| --------- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +103 +| | | | | | Hubblecrossingatτ | | → | | | | | +| --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | +| | | | 102 | | | ★ | | | | | | +101 +100 +w=0 +10-1 +1/3 +| | | | 10-2 | w | = | | | | | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +1 +w = +10-3 +| | | | 10-1 | | | | 100 | | 101 | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | +Figure 1. Plot of the scaling of dΩgw/dlogk versus k for different equations of state w. The +different cosmologies are all normalized so that their sub-horizon modes are of the same size in order +| to emphasize | their | super-horizon | | differences. | | | | | | | | +| ------------ | ------ | ------------- | -------------- | ------------ | --------- | --- | --- | --- | --- | --- | --- | +| 3 The | effect | of | free-streaming | | particles | | | | | | | +In this section, we show how free-streaming relativistic particles induce not only a dampening +of GWs at horizon crossing but also change the shape of the causality-limited part of the +GW spectrum. We specialize to the case of a radiation-dominated universe with a fraction +ffs = ρfs/ρ in free-streaming relativistic particles since for other equations of state the +total +fraction ffs changes significantly with the expansion of the universe. As was derived in +| | | | | | | – 9 – | | | | | | +| --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | + +Ref. [51], GWs are affected by the presence of particles whose free-streaming length is larger +than the Hubble radius. In this case, the propagation of the free-streaming particles along +| geodesics | is affected | | by and | affects | the | propagation | of | GWs. | | | | +| --------- | ----------- | --- | ------ | ------- | --- | ----------- | --- | ---- | --- | --- | --- | +It is well known that during BBN and CMB a large fraction of the energy density in +radiation was free-streaming, since neutrino interactions freeze out at temperatures below +approximately 2MeV. Little is known about our Universe prior to this era, and therefore the +existence of other relativistic free-streaming species at earlier times remains an interesting +possibility in the early Universe. In fact, since GWs themselves are free-streaming, the high +frequency part of the GW spectrum itself is an irreducible contribution to the population of +free streaming particles. We will see that if such particles were present during the generation +of GWs, they would lead to striking features in the causality-limited part of the spectrum. +| The | equation | of | motion | for | GWs | in this | scenario | is [51] | | | | +| ------------ | -------- | ----------- | ------ | ---------- | ----------------- | ------- | -------- | ------- | -------- | -------- | ----- | +| u2∂2h(u)+2u∂ | | h(u)+u2h(u) | | | = | | | | | | | +| u | | u | | | | | | | | | | +| | | | | (cid:90) u | (cid:18) sin(u−x) | | cos(u−x) | | sin(u−x) | (cid:19) | | +| | | −24ffs | | dx | − | | −3 | | +3 | ∂ h(x), | (3.1) | +| | | | | | (u−x)3 | | (u−x)4 | | (u−x)5 | x | | +u(cid:63) +where u = kτ. As indicated before, we are interested in causality-limited GWs. Thus, we are +| solving | these equations | | subject | | to the | boundary | condition | | | | | +| ------- | --------------- | --- | ------- | ----- | ------------ | -------- | --------- | ------------ | ------------ | --- | ----- | +| | | | | h(k,τ | (cid:63) ) = | 0, | ∂ τ h(k,τ | (cid:63) ) = | J (cid:63) . | | (3.2) | +There are two important effects that free-streaming particles have on GWs: post-pro- +duction dynamics and horizon entry. The effects of free-streaming particles on horizon entry +is the standard effect emphasized in the case of neutrinos. Note that the LHS of Eq. (3.1) is +larger than the RHS by a factor of u2 and hence the largest effect is expected to be at early +(cid:46) +times when u = kτ 1. From this we see that sub-horizon modes are completely unaffected +by free-streaming particles. Conversely, super-horizon modes are hit by a uniform frequency +independent suppression when they enter the horizon, e.g. GW amplitudes are suppressed +by a factor of 0.8 when f = 0.4, as originally shown in Ref. [51]. Amusingly, we will see that +for GWs generated by phase transitions, this frequency-independent suppression is in fact a +| subdominant | effect. | | | | | | | | | | | +| ----------- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +The novel effect here is that free-streaming particles also cause a suppression in the +production of GWs. What we mean by this is that the transition of super-horizon GWs from +a large velocity with a small amplitude to a large amplitude with no velocity discussed in +Eq. (2.11) is significantly affected by the presence of free-streaming particles. The suppressed +production can be computed analytically using the techniques of the previous sections. +We first consider the limit u = kτ (cid:28) 1, where the modes are very super-horizon. In this +| limit, Eq. | (3.1) | can | be simplified | | to | | | | | | | +| ---------- | ----- | --------- | ------------- | ----------- | --- | ------ | -------- | --- | -------- | ------- | ----- | +| | | | | | | | (cid:16) | | (cid:17) | 8f | | +| | u2∂ | 2h(u)+2u∂ | | h(u)+u2h(u) | | | | | | fs | | +| | | | | | | = −ffs | h(u)−h(u | | ) , | ffs = . | (3.3) | +| | | u | | u | | | | | (cid:63) | 5 | | +| | | | | | | – | 10 – | | | | | + +The only difference between this equation and Eq. (3.1) is that we have taken the kτ (cid:28) 1 +limitof theRHS, eliminating theintegral. This approximationallows usto studythe effectof +free-streaming radiation on the evolution of super-horizon GWs. The approximation breaks +down near horizon entry, however the effect of horizon entry has been studied in the past and +approximately amounts to a k-independent reduction of the amplitude, proportional to ffs +for ffs (cid:28) 1, and so it can be treated separately. +The suppression of GW production by free-streaming particles can be directly seen from +theRHSofEq.(3.3),wherethefree-streamingparticlesactlikeaHubblescalemass. Thus,in +contrast to the normal scenario where wave amplitudes freeze out and remain constant while +outside of the horizon, the effective Hubble scale mass is constantly reducing the amplitude +of the GW due to a slow roll type effect. Note that this effect is absent if the wave starts out +with negligible velocity like in previously studied scenarios, in which cases the RHS vanishes. +Eq. (3.3) can be solved exactly with our initial conditions to give +(cid:112) +h(τ) = J -Completion -Date -Projected - -K -Completion -Date Actual - -L -Total Costs -Planned - -M -Total Costs -Projected - -N -Total Costs -Actual - -A. Project ID: An agency-specified number that uniquely identifies the project within this investment. -B. Activity Name: A short description consistent with the critical steps within the agency project - -management methodology. - -C. Activity Description: Describe what work is accomplished by this activity. -D. Structure ID: Agency-specified identifier which indicates work breakdown structure agency uses to - -associate this activity with other activities or a project. Please provide this in the format of “x.x.x.x.x” -where the first string is the Project ID and each following string (separated by periods) matches the -Structure ID of a parent activity. See below for more guidance about parent and child activities expressed -through this structure. - -E. Key Deliverable / Usable Functionality: Indicate whether the completion of this activity provides a key -deliverable or usable functionality. This should only be provided for activities which do not have a child -activity. Use this field to demonstrate this investment’s alignment with the modular development -principles of the 25-Point Implementation Plan to Reform Federal IT. - -F. Start Date Planned: The planned start date for this activity. This is the baseline value. -G. Start Date Projected: When activity has not yet started, enter current planned start date of the activity. -H. Start Date Actual: When activity starts, enter actual start date here. -I. Completion Date Planned: The planned completion date for this activity. This is the baseline value. -J. Completion Date Projected: When activity has not yet completed, enter current planned completion date - -of the activity. - -Page 18 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -K. Completion Date Actual: When activity ends, enter actual completion date here. -L. Total Costs Planned: The planned total cost for this activity. This is the baseline value. -M. Total Costs Projected: When activity has not yet completed, enter current planned total cost of the - -activity. - -N. Total Costs Actual: When activity ends, enter actual total costs for the activity here. - -Note: For programs that are employing Earned Value Management, agencies should reflect “budget at -completion” in the “total costs planned” field; and “estimated at completion” in the “total costs projected field for -projected costs. - -Reporting Parent and Child Activities (WBS Structure) - - “Child” activities may be grouped into “Parent” activities to reflect the work breakdown structure (WBS) the -agency uses to manage the investment. If a work breakdown structure is not used by the agency, please report the -relationship between parent activities and child activities in “Structure ID” using this method. - -When reporting an activity, enter the “Structure ID” as a period-delimited string consisting of the “Project ID” -and each nested parent activity between the project level and the child activity. The “Structure ID” to enter will -vary depending on the activity’s WBS level. - -Example: For child activity 3 which is part of parent activity 10, which in turn is part of parent activity 2, -which in turn is part of Project A, please enter: A.2.10.3 - -• Project A - -o Parent Activity 2 - - Parent Activity 10 - -• Child Activity 3 - -There is no limit to the number of nested “child” and “parent” relationships allowed, and this depth may vary -from activity to activity and from project to project. - -If any of a parent activity's child activities occurs in the current fiscal year, then all child activities of the parent -activity must be reported regardless of their timing. This is to ensure that a complete view of the parent activity is -available. - -All activities with no child activities must have, at a minimum, Project ID, Activity Name, Activity Description, -Structure ID, Start Date Planned, Start Date Projected, Completion Date Planned, Completion Date Projected, -Total Costs Planned, and Total Costs Projected. Completed activities must also have Start Date Actual, -Completion Date Actual, and Total Costs Actual. - -Any parent activities with a child activity must be completely described by the aggregate attributes of its child -activities. In the IT Dashboard, the cost and schedule information for parent activities will be based on the cost -and schedule information of their most detailed reported child activities. Agency-submitted cost and schedule -information is not required for parent activities. - -Project -ID - -A -A -A -A - -Activity Name -Design -Business Requirements -Technical Requirements -Architecture - -Structure -ID - -Start Date -Planned - -Completion -Date Planned - -Planned Total -Costs - -… - -A.2 -A.2.1 -A.2.2 -A.2.3 - -2/1/2012 -2/1/2012 -2/11/2012 -2/21/2012 - -2/29/2012 -2/10/2012 -2/20/2012 -2/29/2012 - -$2.5 -$1.0 -$1.0 -$0.5 - -Data reported in cells marked with a fill color (Structure ID: A.2) is unnecessary. These values will be calculated by aggregating the cost and schedule information -reported in this parent activity’s child activities. - -Page 19 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Project Risk - -Risk assessments should include risk information from all stakeholders and should be performed at the initial -concept stage and then monitored and controlled throughout the life-cycle of the investment. - -In Table B.3, list all significant project related risks for the investment that are currently open and provide risk -assessment information. (It is not necessary to address all 19 OMB Risk Categories). - -1. Project ID: An agency-specified number that uniquely identifies a project within this investment. For - -each identified risk, lists the associated Project ID. - -2. Risk Name: A short description that identifies a risk, the cause of the risk and the effect that the risk may - -have on the project. - -3. Risk Category: Please select the relevant OMB Risk Category for each risk. Risk categories include: 1) -schedule; 2) initial costs; 3) life-cycle costs; 4) technical obsolescence; 5) feasibility; 6) reliability of -systems; 7) dependencies and interoperability between this investment and others; 8) surety (asset -protection) considerations; 9) risk of creating a monopoly for future procurements; 10) capability of -agency to manage the investment; and 11) overall risk of investment failure; 12) organizational and -change management; 13) business; 14) data/info; 15) technology; 16) strategic; 17) security; 18) privacy; -and 19) project resources. - -4. Risk Probability: The likelihood of a negative impact for this risk (Low, Medium, or High likelihood). -5. Risk Impact: Assess level of potential negative impacts for the risk (Low, Medium, or High impact). -6. Mitigation Plan: A short description of the plan or steps to mitigate the identified risk. - -Project ID - -Risk Name - -Risk Category Risk Probability - -Risk Impact - -[char 500] - -[19 Risk -Categories] - -[Low, Medium, -High] - -[Low, Medium, High] - -Mitigation Plan -[char 500] - -Table B.3 - -SECTION C: OPERATIONAL DATA - -Section C addresses operational activities which are not reported as a part of a project in Section B. - -Operational Performance -There are two essential types of operations metrics to be reported (see FEA Reference Model Mapping Quick -Guide): - -1. Results Specific: Provide a minimum of two metrics which measure the effectiveness of the investment in -delivering the desired service or support level; if applicable, at least one metric should reflect customer -results (e.g.; “Service Quality”). - -2. Activities and Technology Specific: Provide a minimum of three –metrics which measure the investment -against its defined process standards or technical service level agreements (SLAs) (e.g.; “Reliability and -Availability”). At least one of these metrics must have a monthly “Reporting Frequency.” - -Provide results specific metrics which are appropriate to the mission of the investment and its business owner or -Customer. Generally these metrics should be provided by the investment’s business owner and will reflect -performance in the broader business activities and not IT-specific functions. The best results specific metrics will -support the business case justification and could be the foundation of a quantitative approach to defining benefits -in a cost-benefit analysis. Unlike in private industry where identified benefits accrue to the organization, -government benefits may accrue to the public. Therefore, results-specific metrics may demonstrate the value - -Page 20 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -realized external to the Federal Government. The table must include a minimum of two results-specific metrics, -one of which should reflect customer results. - -Each metric description should help the user understand what is being measured. In this field, describe the units -used, any calculation algorithm used, and the definition or limits of the population or “universe” measured. - -The unit of measure should be characterized (e.g. number, percentage, dollar value etc) for each metric. Each -metric listed in the table must also indicate how often actual measurements will be reported (monthly, quarterly or -semi-annually), as well as baseline, targets and actual results. The “Actual for PY” should be final actual -measurement from the previous year or the average actual results from the previous year. Describe whether a -successful actual measurement would be “over the target” or be “under the target” in “Measurement Condition.” -“Comment” field is required for performance metrics where target not expected to be met. All data will be -displayed on the IT Dashboard. - -Metric -Description - -Unit of -Measure - -[char 500] - -[char 50] - -FEA -Performance -Measurement -Category -Mapping - -[PRM -Measurement -Category] - -Baseline Target -for PY - -Table C.1 -Actual for -PY - -Target -for CY - -Measurement -Condition - -Reporting -Frequency - -Comment - -Most Recent -Actual -Results - -[numeric] - -[numeric] - -[numeric] - -[numeric] - -[Over target, -Under target] - -[Monthly, -Quarterly, -Semi- -Annual] - -[numeric] - -[Optional -char500] - -Operational Risk - -Risk assessments should include risk information from all stakeholders and should be performed at the initial -concept stage and then monitored and controlled throughout the life-cycle of the investment. - -In Table C.2, list all significant operational related risks for the investment that are currently open and provide -risk assessment information. (It is not necessary to address all 19 OMB Risk Categories). - -1. Risk Name: A short description that identifies a risk, the cause of the risk and the effect that the risk may - -have on the operational activity. - -2. Risk Category: Please select the relevant OMB Risk Category for each risk. Risk categories include: 1) -schedule; 2) initial costs; 3) life-cycle costs; 4) technical obsolescence; 5) feasibility; 6) reliability of -systems; 7) dependencies and interoperability between this investment and others; 8) surety (asset -protection) considerations; 9) risk of creating a monopoly for future procurements; 10) capability of -agency to manage the investment; and 11) overall risk of investment failure; 12) organizational and -change management; 13) business; 14) data/info; 15) technology; 16) strategic; 17) security; 18) privacy; -and 19) project resources. - -3. Risk Probability: The likelihood of a negative impact for this risk (Low, Medium, or High likelihood). -4. Risk Impact: Assess level of potential negative impacts for the risk (Low, Medium, or High impact). -5. Mitigation Plan: A short description provides how to mitigate the risk. - -Risk Name -[char 500] - -Risk Category -[19 Risk -Categories] - -Risk Probability Risk Impact -[Low, Medium, -High] - -[Low, Medium, -High] - -Mitigation Plan -[char 500] - -Table C.2 - -Page 21 of 21 Exhibit 300 - -(2011) - +(cid:63) +u2 +(cid:63) [j (u )y (u)−j (u)y (u )] α = +−1+ 1−4ffs +α (cid:63) α α α (cid:63) +k 2 +(3.4) +J u2 (cid:104) (cid:16) απ(cid:17) (cid:16) απ(cid:17)(cid:105) +≈ − (cid:63) (cid:63) j (u )cos u− −y (u )sin u− , +α (cid:63) α (cid:63) +uk 2 2 +where j and y are spherical Bessel functions and we have taken the large time limit in the +α α +second line. From this, we find that the suppression factor is simply +h amplitude (ffs) ≈ (cid:112) u2j (u )2+u2y (u )2, (3.5) +(cid:63) α (cid:63) (cid:63) α (cid:63) +h +amplitude +(ffs = 0) +which follows directly from Eq. (3.4) (and holds even when α is imaginary). +While Eq. (3.5) is rather un-illuminating in and of itself, it can be simplified in two +interesting limits 4ffs (cid:28) 1 and 4ffs > 1. In the first limit, we find that +h +amplitude +(ffs) +≈ (kτ +(cid:63) +)ffs (cid:0) ffs (cid:28) 1 (cid:1) . (3.6) +h +amplitude +(ffs = 0) +In the other limit, 4ffs > 1, we find +(cid:115) +h amplitude (ffs) (cid:112) +(cid:18)(cid:113) (cid:19) +≈ kτ +(cid:63) +C +1 ++C +2 +sin 4ffs−1log(kτ +(cid:63) +)+C +3 +, (3.7) +h +amplitude +(ffs = 0) +where C +1 +, C +2 +, C +3 +are unenlightening functions of ffs. Interestingly, we find that in this limit +there is an overall suppression of the amplitude that is independent ffs and that there is an +additional oscillatory feature on this suppressed amplitude. The appearance of an oscillation +as ffs increases comes from when the super-horizon modes go from being over-damped to +under-damped even while super-horizon. In this limit, the mass coming from free-streaming +particles overcomes Hubble friction and induces oscillations. +– 11 – + +ToobtaintheshapeoftheGWspectrumasafunctionofk,wenumericallysolveEq.(3.1). +The results for the suppression can be found in Fig. 2, which shows the numerical results as +| | | 100 | | | | | | | 100 | | +| --- | ---- | ----- | ------- | --- | ------- | ------- | ------------- | ----- | ----- | --- | +| | | | | ● ● | ● ● ● ● | ● ● ● ● | ● ● ● ● ● ● ● | ●● ●● | ●●● | | +| | | ● ● ● | ● ● ● ● | | | | ● ● | ● | | | +| | | | | | | | ● ● ● ● | ● | | | +| | | | | | | ● ● | | | | | +| | | | | | | ● ● | ● | | | | +| | | | | | ● ● | | ● | | 1 | | +| | | | | | ● | | ● | | 1 0 - | | +| | | | | ● | ● | | | | | | +| | | | ● | ● | | | ● | | | | +| | | | ● ● | | | | | | | | +| | | | ● | | | | ● | | | | +| | | ● ● | | | | | | | | | +| | 10-1 | ● | | | | ● | | | - 2 | | +| | | | | | | ● | | | 1 0 | | +● +● +● +● +● +| | | | | | ● | | | | 10-3 | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | +● +● +● +| | 10-2 | | | | | | | | 10-4 | | +| --- | ---- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | +● +| | | 10-5 | ● 10-4 | | 10-3 | 10-2 | 10-1 | 100 | | | +| --- | --- | ---- | ------ | --- | ---- | ---- | ---- | --- | --- | --- | +Figure 2. The suppression effect of free streaming particles on the production of GWs, for free +streaming fractions 0.01, 0.1 and 0.4 respectively. The dots are numerical data points while the solid +lines are the analytical approximations. The dotted purple line is a k4 scaling shown to highlight the +| oscillations | taking | place. | | | | | | | | | +| ------------ | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | --- | +well as the analytical approximation, Eq. (3.5). As can be clearly seen, the analytic results +and the numerical agree very well for ffs = 0.01,0.1 and 0.4. This agreement holds despite +the fact that the transition region, where u = kτ ∼ 1, is not accurately captured by our +approximation, and hence, the traditional effect of horizon entry is not entirely taken into +account. +We can re-express our results for the amplitude of the GWs in terms of the observable +| dΩgw/dlogk. | For | sub-horizon | modes, | | we have | the previous | result | | | | +| ----------- | --- | ----------- | ------ | --- | ------- | ------------ | ------ | --- | --- | --- | +dΩgw +k3 +| | | | | ∼ | | (k (cid:29) | H ). | | | (3.8) | +| ----------------- | --- | ------ | ------- | --- | --- | ----------- | -------- | --- | --- | ----- | +| | | | dlogk | | | | (cid:63) | | | | +| For super-horizon | | modes, | we have | | | | | | | | +dΩgw +k3+1 6ffs +| | | | ∼ | 5 | | (k (cid:28) | H , ffs (cid:28) | 1) | | (3.9) | +| ------------------------ | --- | --------- | --- | --- | --- | ----------- | ---------------- | --- | --- | ----- | +| | | dlogk | | | | | (cid:63) | | | | +| for small free-streaming | | fractions | | and | | | | | | | +(cid:18) (cid:18)(cid:113) (cid:19)(cid:19) (cid:18) (cid:19) +| dΩgw | | | | | | | | | 5 | | +| ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +∼ k4 C +C sin 4ffs−1log(k/H )+C k (cid:28) H , ffs (cid:38) (3.10) +| | | 1 2 | | | | (cid:63) 3 | | | (cid:63) | | +| ----- | --- | --- | --- | --- | --- | ---------- | --- | --- | -------- | --- | +| dlogk | | | | | | | | | 32 | | +for large free-streaming fractions, where C are constants. This shows that the presence +1,2,3 +| | | | | | – 12 | – | | | | | +| --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | + +of free-streaming radiation during the generation of GWs leads to a spectrum that decreases +fasteratlowfrequency,withthespectralindexgrowinglinearlyinffs from3untilitsaturates +| at 4 for | ffs ≥ | 5 . | | | | | | | | +| -------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | +32 +As mentioned before, the high k (sub-horizon) modes of the GWs themselves necessarily +play the role of free-streaming particles. The derivation of Eq. (3.1) assumes that the free- +streamingparticlesfollowgeodesics, whichGWsdoaswell. Thus, theonlypossibledifference +betweenthehighfrequencyGWmodesandfree-streamingparticlesisthatthehighfrequency +GW modes were produced by the phase transition itself. Imagine that the GWs are produced +| at a time | (cid:15) after | the phase | transition, | | then Eq. | (3.1) becomes | | | | +| ------------ | -------------- | ----------- | ----------- | ----------------- | -------- | ------------- | -------- | -------- | ------ | +| u2∂2h(u)+2u∂ | | h(u)+u2h(u) | | = | | | | | | +| u | | u | | | | | | | | +| | | (cid:90) | u | (cid:18) sin(u−x) | | cos(u−x) | sin(u−x) | (cid:19) | | +| | −24ffs | | dx | − | | −3 | +3 | ∂ h(x). | (3.11) | +| | | | | (u−x)3 | | (u−x)4 | (u−x)5 | x | | +u(cid:63)+k(cid:15) +As long as H (cid:15) (cid:28) 1, we recover Eq. (3.3) and we see that the high frequency GW modes +(cid:63) +| behave | exactly | like any | other | free-streaming | particles. | | | | | +| ---------------- | ------- | -------- | ------- | -------------- | ---------- | --- | --- | --- | --- | +| 3.1 Implications | | of | current | N eff | limits | | | | | +ResultsfromCMBandBBNmeasurementslimittheamountofnonStandardModelradiation +present in the Universe at those times. These results are presented as limits on the effective +| number | of neutrinos, | | | | | | | | | +| ------ | ------------- | --- | --- | --- | ------------------------- | ----- | --- | --- | --- | +| | | | | | 8 (cid:18) 11 (cid:19)4/3 | (ρ +ρ | ) | | | +ν X +| | | | | N = | | | , | | (3.12) | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | +| | | | | eff | 7 4 | ρ | | | | +γ +where ρ and ρ are respectively the energy density in photons and neutrinos and ρ repre- +| | γ | ν | | | | | | | X | +| --------- | --- | ------------ | --- | ---------- | ------- | ------------- | --- | --- | --- | +| sents any | new | contribution | to | the energy | density | in radiation. | | | | +A combined analysis of BBN and CMB measurements leads to a bound ∆N (cid:46) 0.3 +eff +[56]. This implies that at neutrino decoupling a new type of radiation could contribute to +at most 4.4% of the radiation energy. However, due to the decoupling of massive degrees +of freedom, the relative temperature between the standard model plasma and any decoupled +species changes as the Universe evolves. In particular, a ∆N (cid:46) 0.3 at BBN translates to +eff +| ffs (cid:46) 0.09 | for temperatures | | above | the | weak scale. | | | | | +| ----------------- | ---------------- | --- | ----- | --- | ----------- | --- | --- | --- | --- | +This shows that if below the weak scale there are no modifications to ΛCDM, the maxi- +mum allowed ffs is 0.09, and if a non-zero N is measured in future CMB experiments, the +eff +nature of this radiation at high energies could be tested using GWs.3 Nevertheless larger +ffs are possible with simple modifications of early cosmology, such as late equilibration of +the radiation with the SM with subsequent decay or through an entropy injection into the +standard model plasma, which dilutes the relative energy density in the new species. The +3WhileBBNmeasurementsalonecannotdistinguishfree-streamingradiationfromaninteractingrelativistic +fluid, CMB measurements are now precise enough to differentiate between them (see e.g. [57–59]). Note that +the CMB measurements are only sensitive to the nature of the relativistic species at low energies, O(eV). +– 13 – + +latter case would also lead to a temporary change in the equation of state that changes the +| GW | signal as discussed | in | the next | section. | | | | +| --- | ------------------- | ------------ | -------- | --------- | --------- | --- | --- | +| 4 | Effect of | non-standard | | expansion | histories | | | +In this section we show how the causality-limited spectrum of GWs can be used to probe +the expansion rate of the Universe after the GWs were generated. There are two main +effects of a non-standard expansion history in the spectrum of GWs: an overall re-scaling of +the spectrum due to the change in expansion history; and a change in the power-spectrum +of causality-limited modes that enter the horizon during an era of non-radiation domination. +Thefirstfeaturehasbeenwidelydiscussedintheliterature,whilethesecondhasbeenstudied +in the case of tensor modes for inflation [31, 33–36] and cosmic strings [32, 38, 40, 41, 60] +but has been mostly overlooked with regard to phase transitions (see [47–50] for some recent +| studies | in this direction). | | | | | | | +| ------- | ------------------- | --- | ------ | --------- | ---------- | --- | --- | +| 4.1 | An intermediate | | period | of matter | domination | | | +Inthissubsectionweconsidertheeffectofanintermediateperiodofmatterdomination(MD). +We will compare two scenarios. In the first scenario, GWs are produced at a temperature T +(cid:63) +and the universe is radiation dominated (RD) until CMB. In the second scenario, GWs are +again producedat a temperatureT but there is an intermediate period of matter domination +(cid:63) +between the temperatures Tr→m, where the universe goes from RD to MD, and Tm→r, where +| it goes | from MD | to RD. | | | | | | +| ------- | ------- | ------ | --- | --- | --- | --- | --- | +A summary of the results in this section can be seen in Fig. 3 where we compare the +numerical solutions for the spectrum in both cases. We find that if one is interested in modes +that enter the horizon before Tr→m, then the extra red-shifting induced by the period of MD +simply decreases the amplitude of these modes. If one considers instead the modes that enter +thehorizonafterTm→r thentheiramplitudeincreasesduetothered-shiftingofthefrequencies +movingpowerfromhighfrequenciestolowerfrequencies. Inwhatfollows, wegiveanintuitive +explanation of the results while in App. A we give a more analytical derivation. +Matterdominationgivesanentropydumpthatincreasesthered-shiftbetweenthesource +and present day observers yielding two distinct effects on the spectrum. The first is that the +extra expansion dilutes the energy density, decreasing the overall power. The second is that +the frequencies themselves are red-shifted, an effect that increases the power at low frequency +as it shifts the entire spectrum to lower frequencies. The competition of these two effects +| determines | the | behavior | of the spectrum. | | | | | +| ---------- | --- | -------- | ---------------- | --- | --- | --- | --- | +The net effect of MD can be understood intuitively and is shown visually in Fig. 4. We +first discuss the effect of the dilution of the energy density from the additional red-shifting. +FormodesthatenterthehorizonbeforeTr→m,theoverallpowerisredshiftedbyanadditional +(∆a)−4 +| | due to | the additional | expansion. | | Here we take | | | +| --- | ------ | -------------- | ---------- | ----- | -------------------- | --- | --- | +| | | | | | (cid:18) (cid:19)1/3 | | | +| | | | | (a /a | ) Tr→m | | | +(cid:63) 0 1 +| | | | ∆a | = | = | > 1 | (4.1) | +| --- | --- | --- | --- | ----- | ------ | --- | ----- | +| | | | | (a /a | ) Tm→r | | | +(cid:63) 0 2 +| | | | | | – 14 – | | | +| --- | --- | --- | --- | --- | ------ | --- | --- | + +103 +100 f3 +10-3 +f +10-6 +f3 +RD MD RD +10-9 +10-1 100 101 102 103 104 +Figure 3. The low-frequency spectrum of causality-limited GWs in the case with an intermediate +period of matter domination (light blue) and the case with just radiation domination (orange). The +intermediate period of matter domination is present between the frequencies fm→r to fr→m. The +dashedblacklinesshowthescalingasf (f3)duringMD(RD).Highfrequencymodeshavelesspower +due to the presence of MD while low frequency modes have more power. +Figure 4. A qualitative depiction of the two physical effects determining the shape of the causality- +limited range of the GW spectrum in pure RD (scenario 1) and with an intermediate epoch of MD +(scenario2). Left: impactofthedilutionofρgw withrespecttotheambientenergydensityduringMD +for sub-horizon modes. Right: effect of the different redshift of the scale factor in the two scenarios. +See the text for more details. +to be the difference between scenarios 1 (pure RD) and 2 (intermediate MD) in the amount +of red-shifting between the phase transition and us. For modes that re-enter the horizon after +Tm→r, the modes were frozen out during matter domination and so their overall power is +not modified. Intermediate frequencies interpolate between the two results with a f scaling. +We now discuss the effect of the red-shifting of momenta. The observable GW frequency f +– 15 – + +corresponds to the physical momentum k/a of the GW, which in scenario 2 redshifts more by +a factor ∆a. The f3 fall off of the power spectrum during RD means that when comparing +thetwospectraatthesamephysicalmomenta, thepoweris enhancedbyafactor of(∆a)3 for +all modes entering during RD. In totality, for modes that re-enter the horizon before Tr→m, +these two effects combine to give a total suppression of 1/∆a while for modes that re-enter +the horizon after Tm→r, they combine to give a total enhancement of (∆a)3. Thus we find +that an intermediate period of MD increases the power at low frequencies but suppresses it +at high frequencies. +These estimates can be made more concrete as follows. In the first scenario where the +| universe | was always | radiation | | dominated, | | the | spectrum | of GWs | is | | | | +| -------- | ---------- | --------- | --- | ---------- | --- | ----- | -------- | ------ | ---------- | --- | --- | ----- | +| | | | | dΩgw,1 | | | f3 | | T T | | | | +| | | | | | = | Agw,1 | | f ∼ | (cid:63) 0 | | | (4.2) | +(cid:63),1 +| | | | | dlogf | | | f3 | | M | | | | +| --- | --- | --- | --- | ----- | --- | --- | ---------- | --- | --- | --- | --- | --- | +| | | | | | | | (cid:63),1 | | P | | | | +where Agw,1 is the amplitude of the GW spectrum at its peak and this scaling holds up to +the peak frequency f . Frequencies higher than this are sensitive to the properties of how +(cid:63),1 +| the signal | was generated. | | | | | | | | | | | | +| ---------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +As discussed before, there are two main effects of a period of intermediate matter domi- +nation. Firstly, the amplitude of the peak signal is decreased by (∆a)−4, +| | | | | | | | (cid:18) Tr→m | (cid:19)−4/3 | | | | | +| --- | --- | --- | --- | ----- | --- | ----- | ------------- | ------------ | --- | --- | --- | ----- | +| | | | | Agw,2 | = | Agw,1 | | | . | | | (4.3) | +Tm→r +Secondly, the frequencies are shifted by 1/∆a so that the new peak frequency is +| | | | | | (cid:18) | (cid:19)−1/3 | | (cid:18) | (cid:19)−1/3 | | | | +| --- | --- | --- | ---------- | ---------- | -------- | ------------ | --- | ----------------- | ------------ | --- | --- | ----- | +| | | | | | Tr→m | | T | (cid:63) T 0 Tr→m | | | | | +| | | | f = | f | | | ∼ | | | . | | (4.4) | +| | | | (cid:63),2 | (cid:63),1 | | | | | | | | | +| | | | | | Tm→r | | M | P Tm→r | | | | | +Finally, the universe transitions to and from MD at the critical frequencies +| | | | | | (cid:18) | | (cid:19)−1/3 | | | | | | +| --- | --- | ---- | --- | ----- | -------- | ---- | ------------ | ---- | ----- | --- | --- | ----- | +| | | | | Tr→mT | 0 | Tr→m | | | Tm→rT | 0 | | | +| | | fr→m | ∼ | | | | | fm→r | ∼ | . | | (4.5) | +| | | | | M | | Tm→r | | | | M | | | +| | | | | P | | | | | | P | | | +f3 +From this starting point, the entire spectrum can be found by scaling as as long as the +| Universe | is in RD | and | as f1 | as long | as | it is | in MD. | | | | | | +| -------- | -------------- | --- | ------- | ------- | --- | ----- | ------ | --- | --- | --- | --- | --- | +| The | final spectrum | | is then | | | | | | | | | | + +| | | | | | | (cid:18) | (cid:19)−1/3 | | | | | | +| --- | --- | ------- | --- | --- | ------ | -------- | ------------ | --- | --- | --- | ------ | --- | +| | |  | f 3 | d Ω | g w ,1 | T r | m | | | | | | +| | |  Agw,2 | | = | | → | | | | f | > fr→m | | +  +| | |  | f 3 | d | lo g f | T m | r | | | | | | +| --- | --- | --- | ----------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +| | |   | (cid:63) ,2 | | | | → | | | | | | + +| d Ω | |   | f 3 | f | | | | | | | | | +| --- | ------ | --------- | ----------- | ----- | ----- | --- | -------- | -------- | -------- | --- | ------ | ----- | +| | g w ,2 | | r → | m | | | | | | | | | +| | = | Agw,2 | | | | | | | fr→m | > f | > fm→r | (4.6) | +| d | lo g f | | f 3 | fr | | | | | | | | | +| | |  | | →m | | | | | | | | | +| | |   | (cid:63) ,2 | | | | | | | | | | +| | |  | | | | | | (cid:18) | (cid:19) | | | | +| | |   | f 3 | f | f | 3 | d Ω | T | | | | | +| | |  | r → | m m → | r | | g w ,1 | r→ m | | | | | +| | |   Agw,2 | | | | = | | | | f | < fm→r | | +| | |  | f 3 | f r→ | m f 3 | | d lo g f | T m r | | | | | +| | | | (cid:63) ,2 | | m | → r | | → | | | | | +| | | | | | | – | 16 – | | | | | | + +We see all of the effects that we had mentioned before. At high frequencies, there is a +suppression of 1/∆a = (Tr→m/Tm→r)−1/3 while at low frequencies there is an enhancement +of (∆a)3 ∼ Tr→m/Tm→r. +10-2 10-1 100 101 102 103 +103 +k3 +100 k +10-3 +k3 +10-6 +RD MD +10-9 +10-5 10-4 10-3 10-2 10-1 100 101 +Figure5. ThespectrumofaGWproducedduringmatterdomination. Sub-Hubblecausalityenforces +a k3 scaling, while super-Hubble scaling gives k during matter domination and k3 during radiation +domination. +A limit of the previous scenario is when the period of matter domination extends to the +point in time when the GWs were produced. As shown in Fig. 5, in this case the power +in modes that are sub-horizon when the spectrum is generated scales as k3 while for those +outside it goes as k. Thus even though this also has a transition from a k3 scaling to a k +scaling,theoriginforthek3 scalingisdifferentbetweenthetwoscenarios,sub-horizonphysics +instead of radiation domination. It is then an interesting question to see if it is possible to +differentiate between the two. +As the transition from sub-horizon to MD and the one from RD to MD both lead to +identical scaling away from the transition region, we cannot differentiate between the two +based on their spectrum alone. However, the two transitions are not identical. Thus, in +principle, if one sees the transition between a k3 scaling to a k1 scaling, one could use its +shape to determine whether the k3 scaling was due to radiation domination or modes being +sub-horizon. +WeillustratethedifferencebetweenthesetwoscenariosinFig.6. Fromthis,oneseesthat +thetwodifferenttransitionsareinfactdistinguishable,butthattheirdeviationisrathersmall. +Without extra information from shorter wavelengths that are sensitive to the generation +mechanism for the waves, one would need an extremely precise measurement of the transition +region in order to determine which kind of transition it is. +– 17 – + +| | | 0.03 0.1 | 0.3 1 | 3 10 | | +| --- | --- | -------- | ----- | ---- | --- | +RD +| | | 3 | | 3 | | +| --- | --- | --- | --- | --- | --- | +| | | 2 | | 2 | | +| | | 1 | | 1 | | +MD +| | | 0 | | 0 | | +| --- | --- | -------- | ----- | ---- | --- | +| | | 0.03 0.1 | 0.3 1 | 3 10 | | +Figure 6. Comparison between the evolution of the tilt of GW spectra in two different scenarios: +transition from sub-horizon modes to MD super-horizon modes (green) and from RD super-horizon +modes to MD super-horizon modes (purple). Sub-horizon and RD super-horizon modes both scale as +dΩgw/dlogk ∼ k3 while MD super-horizon modes scale as dΩgw/dlogk ∼ k so that both scenarios +havetheexponenttransitionfrom3to1. Asisshowninthepicture, thetransitionregionisdifferent, +| showing | that the two scenarios | can in principle | be differentiated. | | | +| --------- | ---------------------- | ---------------- | ------------------ | --- | --- | +| 4.2 Using | the GW spectrum | to | measure w(t) | | | +In the early universe, it is highly likely that w was not constant and was a changing function +of time. In this section, we consider how to measure w(t). We will demonstrate that w(t) +can be to a very good approximation read off directly from the slope of dΩgw/dlogk. +In an expanding universe with a constant w, the slope of dΩgw/dlogk for super-horizon +| modes is | given by Eq. (2.24) | and repeated | below for convenience | | | +| -------- | ------------------- | ------------ | --------------------- | --- | --- | +dΩgw 1+15w +| | | | ∼ k . | | (4.7) | +| --- | --- | --- | ----- | --- | ----- | +1+3w +dlogk +When one considers a generic w(τ), there is no longer an exact solution. However, the +intuition developed before still applies. As mentioned before, we can approximate the GW as +having a suppressed production coupled with being frozen out until k = H. These combined +| to give | Eq. (2.13), repeated | below for | the sake of convenience | | | +| ------- | -------------------- | --------- | ----------------------- | --- | --- | +a(τ ) J +k (cid:63) +| | | h = | sinkτ | . | (4.8) | +| --- | --- | --- | ----- | --- | ----- | +a(τ) H +(cid:63) +– 18 – + +| As | long as | w(cid:48)(τ) (cid:28) | H, then | we can proceed | as | before | and obtain | | | | | | +| --- | ------- | --------------------- | ------- | -------------- | --- | -------- | ---------- | --- | --- | --- | --- | ----- | +| | | | | dΩgw | | 1+15w(τ) | | | | | | | +| | | | | | ∼ | k | , | | | | | (4.9) | +1+3w(τ) +dlogk +where, for each momentum k, w(τ) should be taken at the time τ when k re-enters the +horizon, k = H(τ). We see that in this approximation, one can simply read off w(τ) straight +| from | the slope | of dΩgw/dlogk. | | More explicitly | | | | | | | | | +| ---- | --------- | -------------- | --- | ---------------- | --- | --- | -------- | ---- | --- | --- | --- | ------ | +| | | | | dlog(dΩgw/dlogk) | | | 1+15w(τ) | | | | | | +| | | | | | | = | | . | | | | (4.10) | +| | | | | dlogk | | | 1+3w(τ) | | | | | | +| | 10-2 | 10-1 100 | 101 | 102 103 | | | 10-2 | 10-1 | 100 | 101 | 102 | 103 | +| | 3 | RD | | | | 3 | | | | RD | | | +| | 2 | | | | | 2 | | | | | | | +MD +| | 1 | | | | | 1 | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +MD +| | 0 | | | | | 0 | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +10-5 10-4 10-3 10-2 10-1 100 101 10-5 10-4 10-3 10-2 10-1 100 101 +Figure 7. A comparison between the exact equation of state of the universe versus what is measured +from the slope of the GW spectrum. In the solid blue line, we have the exact numerical result while +in the dashed blue line, we have the approximation as measured from the slope of the GW spectrum. +Following this expression, one could in principle extract w(τ) by just measuring the tilt +of the low-frequency tail of the GW spectrum. The approximations made to obtain this +simple expression are not valid during the transition region between matter and radiation +domination. To see if they work in practice, we compared this procedure to several w(τ) +numerically. The results are shown in Fig. 7. We see that this procedure is surprisingly +accurate. +5 Conclusion +In this paper, we have presented a physical picture of how to understand causality-limited +gravitational waves. This picture allows one to easily estimate and predict the shape and be- +havior of low frequency gravity waves. As an example, we showed that generally sub-horizon +and super-horizon modes behave very differently, allowing one to make model independent +– 19 – + +measurements of the conformal Hubble rate at which the gravity waves were created. Addi- +tionally, we showed how an intermediate period of MD has the effect of suppressing power at +high frequencies and enhancing power at low frequencies. +Perhaps the most surprising result is that we showed that free-streaming particles change +the scaling of the GW power spectrum. Unlike the usual case where particles start free- +streaming after GWs are produced, if there are free-streaming particles present when GWs +are produced, they lead to a suppression in the production of super-horizon gravitational +waves, bringing the scaling from k3 down to a maximum of k4. In the limit of a large number +of free-streaming particles, oscillatory features are present demonstrating that interesting +effects unrelated to causality can sculpt the spectrum of gravity waves. +There is much still left to be done with causality-limited gravitational waves. This range +of the spectrum is special because its shape can be calculated from first principles. However, +because causality sends the power to zero at low frequencies, a dedicated analysis would +be needed to see to what extent gravity wave detectors such as LISA would be able to +detect changes on this falling spectrum. On a separate note, since preheating and reheating +generated GWsare typically followed by aperiod ofmatter domination, they are morevisible +than previously expected. It would be interesting to see if these models naturally generate +signals detectable by existing or future gravitational wave detectors. +Acknowledgments +We thank Dani Figueroa and Raman Sundrum for stimulating discussions. We also thank +G´eraldine Servant and Peera Simakachorn for comments on the draft. The authors also +thank the KITP institute (From Inflation to the Hot Big Bang) where part of this work was +conducted, and the support of the NSF under the Grant No. PHY-1748958. This research +was supported in part by the NSF under Grants No. PHY-1914480, PHY-1914731 and by +the Maryland Center for Fundamental Physics (MCFP). Research at Perimeter Institute is +supported in part by the Government of Canada through the Department of Innovation, +Science and Economic Development Canada and by the Province of Ontario through the +Ministry of Colleges and Universities. +A Analytical description of the GW spectrum with an intermediate period +of matter domination +WestartbyusingthephysicalintuitiondescribedinSec.2.1tofindanapproximateanalytical +form of the GW spectrum in cosmologies with a non-standard expansion history between +generation and observation of the GWs. We can approximate Eq. (2.4) for super-horizon +modes by +2n +h(cid:48)(cid:48)+ (cid:63) h(cid:48) = 0 +τ (A.1) +h(τ ) = 0, h(cid:48)(τ ) = J , +(cid:63) (cid:63) (cid:63) +– 20 – + +where n = 2/(1+3w ) is fixed by the equation of state at τ . As long as n > 1/2, these +| | (cid:63) | | (cid:63) | | | | | | (cid:63) | | (cid:63) | | +| ------- | ---------- | ------- | -------- | ------ | --- | --- | --- | --- | -------- | --- | -------- | --- | +| initial | conditions | quickly | | evolve | to | | | | | | | | +J +| | | | | | h(τ) | = | (cid:63) | . | | | | (A.2) | +| --- | --- | --- | --- | --- | ---- | ------ | -------- | --- | --- | --- | --- | ----- | +| | | | | | | (2−1/n | )H | | | | | | +(cid:63) (cid:63) +Afterwards, h remains constant while k (cid:28) H, in analogy with Eq. (2.11), even if the equation +| of state | (EOS) | of | the Universe | | changes. | | | | | | | | +| -------- | ----- | --- | ------------ | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +Assuming a constant EOS during horizon entry, kτ ∼ 1, we must solve +2n +| | | | | h(cid:48)(cid:48)+ | k h(cid:48)+k2h | | = 0, | | | | | | +| --- | --- | --- | --- | ------------------ | --------------- | --- | ---- | --- | --- | --- | --- | --- | +τ +(A.3) +J +| | | | | | | (cid:63) | | h(cid:48)(τ | | | | | +| --- | --- | --- | --- | ----- | ------ | -------- | --- | ----------- | ---- | --- | --- | --- | +| | | | | h(τ i | ) = | | , | i ) | = 0, | | | | +| | | | | | (2−1/n | | )H | | | | | | +(cid:63) (cid:63) +where the initial time, τ (cid:28) 1/k, is chosen such that the equation of state is constant between +i +τ and horizon entry, and n is fixed by the equation of state of the Universe when the mode +| i | | | | k | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +enters the horizon. The solution to this equation in the τ (cid:28) 1/k limit is given by +i +| | | | | | | | (cid:18) | (cid:19)n | | | | | +| ------ | --- | ------------- | -------- | ------ | -------- | -------- | --------- | --------- | ---- | -------- | --- | ----- | +| | | | | | 2J | Γ(n | +1/2) | 2 | k −1 | | | | +| | | | | | (cid:63) | k√ | | | | | | | +| | | | h(τ) = | | | | | | j n | −1 (kτ). | | (A.4) | +| | | | | (2−1/n | )H | | π | kτ | k | | | | +| | | | | | (cid:63) | (cid:63) | | | | | | | +| Taking | the | kτ (cid:29) 1 | limit of | the | equation | above | we arrive | at | | | | | +(cid:18) (cid:19)n +| | | | | J (cid:63) | | Γ(n k√ +1/2) | 2 | k | | | | | +| --- | --- | ---- | --- | ---------- | --- | ------------ | --- | ------ | --- | -------- | --- | ----- | +| | | h(τ) | ≈ | | | | | cos(kτ | | −n π/2). | | (A.5) | +k +| | | | (2−1/n | | (cid:63) )H (cid:63) | | π kτ | | | | | | +| --- | --- | --- | ------ | --- | -------------------- | --- | ---- | --- | --- | --- | --- | --- | +a−1, +Once k (cid:29) H the amplitude falls as h ∝ and hence, focusing only on the amplitude of h +we have +| | | | | | | | | (cid:18) | (cid:19)n | | | | +| --- | --- | --- | --- | ------ | ---------- | -------- | -------- | -------- | --------- | --- | --- | ----- | +| | | | | | J (cid:63) | Γ(n | k√ +1/2) | 2 | k a(τ | r ) | | | +| | | | |h| | ≈ | | | | | | , | | (A.6) | +| | | | | (2−1/n | | )H | | kτ | | a | | | +| | | | | | (cid:63) | (cid:63) | π | r | | | | | +where τ is some fixed reference time at which the EOS was still determined by n . Fixing +| | r | | | | | | | | | | | k | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +the reference time τ allows us to compare different modes that enter horizon during an era +r +of fixed n. +We will now use the result in Eq. (A.6) to compare two scenarios to show the effects +of non-standard expansion histories. As before in the first scenario, GWs are produced at +a temperature T at a scale factor a and the universe is radiation dominated until the +| | | (cid:63) | | | | (cid:63) | | | | | | | +| --- | --- | -------- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | +CMB. In the second scenario, GWs are produced at the same temperature as before, T , but +(cid:63) +there is an intermediate period of MD between the temperatures Tr→m (where the universe +becomes matter dominated) and Tm→r (where the universe becomes radiation dominated). +For simplicity we will take the scale factor at production to be the same a , which implies +(cid:63) +that the scale factor today will differ due to the difference in expansion history. +As discussed in Section 4.1, one of the major effects of an intermediate period of MD is +that the frequency spectrum of GWs is shifted by the additional expansion due to the matter +| | | | | | | – | 21 – | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | + +dominated phase. Since we normalized the scale factor at generation a to be the same in +(cid:63) +both scenarios, the relation between the scale factor and plasma temperature is identical in +both cases up until the temperature reaches Tr→m, when matter domination starts in the +second scenario. The expansion history of both scenarios differ from that point on until +the temperature gets to Tm→r, after that point the scale factors corresponding to the same +temperature (and therefore the scale factors at observation) are related by +| | | | | | | (cid:18) | (cid:19)1/3 | | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ----------- | --- | --- | --- | --- | --- | +Tr→m +| | | | | | a = | a | | . | | | | (A.7) | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | +2 1 +Tm→r +Following Eq. (A.6), the amplitude for modes that have entered the horizon in the first +| scenario | is | | | | | | | | | | | | +| -------- | --- | --- | --- | --- | --------------- | ---- | --- | --------------- | --- | --- | --- | --- | +| | | | | J | (cid:18) am→r,1 | Hm→r | | (cid:19) am→r,1 | | | | | +(cid:63) +| | | | | h ≈ | | | | | . | | | (A.8) | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | +| | | | | H | | k | | a | | | | | +(cid:63) +| Therefore, | the | differential | energy | density | | is given | by | | | | | | +| ---------- | --- | ------------ | ------ | ------- | ------- | -------- | ----------------------- | --- | --------- | --- | --- | ----- | +| | | | | | | (cid:18) | (cid:19)2(cid:16)am→r,1 | | | | | | +| | | | dρ | | H 2 | J | | | (cid:17)4 | | | | +| | | | | 1 | m →r | | (cid:63) | | k3. | | | | +| | | | | = | | | | | | | | (A.9) | +| | | | dlogk | | 2(2π)3G | H | | a | | | | | +(cid:63) +For the second scenario we need to separate between the modes that entered the horizon +before, during and after the period of matter domination. The amplitude for modes that +enteredthehorizonaftertheendofmatterdomination(k < Hm→ram→r,2 )canbeimmediately +| found | from Eq. | (A.6), | | | | | | | | | | | +| ----- | -------- | ------ | --- | --- | -------- | ---- | --- | -------- | --- | --- | --- | --- | +| | | | | | (cid:18) | | | (cid:19) | | | | | +| | | | | J | am→r,2 | Hm→r | | am→r,2 | | | | | +(cid:63) +| | | | | h ≈ | | | | | . | | | (A.10) | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | +| | | | | H | | k | | a | | | | | +(cid:63) +Modes that entered the horizon during matter domination have a different form due to the +| difference | in EOS | and | their amplitude | | is | given | by | | | | | | +| ---------- | ------ | --- | --------------- | --- | -------- | ----- | --- | --------- | --- | --- | --- | ------ | +| | | | | | (cid:18) | | | (cid:19)2 | | | | | +| | | | | J | am→r,2 | Hm→r | | am→r,2 | | | | | +| | | | h | ≈ | (cid:63) | | | | . | | | (A.11) | +| | | | | H | | k | | | a | | | | +(cid:63) +Finally, modes that entered the horizon before matter domination are given by +| | (cid:18) | | (cid:19) | | (cid:32) | | (cid:33) | (cid:18) | | (cid:19) | | | +| --- | ---------- | --- | -------- | --- | -------- | ---- | -------- | -------- | ----------- | -------- | --- | ------ | +| | J ar→mHr→m | | ar→m | | a2 | T 2 | | J | am→r,2 Hm→r | am→r,2 | | | +| h | ≈ (cid:63) | | | ≈ | r→m | r | →m | (cid:63) | | | . | (A.12) | +| | H | k | | a | a2 | T2 | | H | k | a | | | +| | (cid:63) | | | | m→r,2 | m→r | | (cid:63) | | | | | +| | | | | | | – 22 | – | | | | | | + +| From | which | we find | the differential | | energy | density | | | | | | | +| ---- | ----- | ------- | ---------------- | ----------------------- | --------- | ------- | --- | --- | --- | --- | --- | --- | +| | | | (cid:18) | (cid:19)2(cid:16)am→r,2 | | | | | | | | | +| | dρ | H 2 | J | | (cid:17)4 | | | | | | | | +| | 2 | = m →r | | (cid:63) | | k3× | | | | | | | +| | dlogk | 2(2π)3G | H | | a | | | | | | | | +(cid:63) + +| | | | | | | 1 , | k | < am→r,2 | Hm→r | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ---- | --- | --- | --- | + +  +| | | | |  | (cid:16) | | (cid:17)2 | | | | | | +| --- | --- | --- | --- | --- | -------- | ---- | --------- | ---- | --- | -------------- | --- | -------- | +| | | | |  | am→r,2 | Hm→r | , am→r,2 | Hm→r | | < k < ar→mHr→m | | | +| | | | | × | | | | | | | | . (A.13) | +k +| | | | | | (cid:18) | | (cid:19)2 | | | | | | +| --- | --- | --- | --- | --- | -------- | ---------- | --------- | ---------- | --- | --- | --- | --- | +| | | | |   | 2 | 2 | | | | | | | +| | | | |  | a r → | m T r → m | , k | > ar→mHr→m | | | | | +| | | | |   | 2 | 2 | | | | | | | +| | | | | | a m → r | ,2 T m → r | | | | | | | +Comparing this to the differential energy density for the RD only scenario, we can immedi- +(∆a)4 +ately see the dilution from the extra expansion discussed in the main text. Now, this +distribution is in terms of conformal momentum k, which, due to the difference in scale factor +normalization at late times correspond to distinct frequencies in the two cases. In order to +compare the spectra of the two scenarios and their relation to experimental searches, it is +more useful to compute the differential spectrum in terms of frequency, f = k/(2πa): +| | | | | | | 2 | (cid:18) (cid:19)2(cid:16)am→r,1 | | | | | | +| --- | --- | --- | --- | ---- | --- | ---- | -------------------------------- | --- | --------- | --- | --- | ------ | +| | | | | dρ 1 | H | m →r | J (cid:63) | | (cid:17)4 | | | | +| | | | | (f) | = | | | | a3f3. | | | (A.14) | +1 +| | | | dlogf | | | 2G | H (cid:63) | a | | | | | +| --- | --- | --- | ----- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | +Using the fact that scale factors after the end of MD satisfy am→r,1 /a = am→r,2 /a , we find +| | | | | | | | | | | 1 | | 2 | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | + +| | | | | | | | 1 | , | f < | fm→r | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | +  +| | | | (cid:18) | (cid:19)3 | |  | (cid:16) | (cid:17)2 | | | | | +| --- | --- | ------- | -------- | --------- | --- | --- | -------- | --------- | ---- | ---------- | --- | ------ | +| | | dρ | am→r,2 | | dρ |  | fm →r | | | | | | +| | | 2 (f) = | | | 1 | (f) | | , | fm→r | < f < fr→m | , | (A.15) | +f +| | dlogf | | am→r,1 | | dlogf |  | (cid:16) | (cid:17)−4 | | | | | +| --- | ----- | --- | ------ | --- | ----- | --- | --------- | ---------- | --- | ---- | --- | --- | +| | | | | | |  | a | | | | | | +| | | | | | |   | m → r , 2 | , | f > | fr→m | | | +a +| | | | | | | | m → r , 1 | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | +where fr→m = ar→mHr→m/(2πa ) and fm→r = am→r,2 Hm→r/(2πa ), which reproduces the +| | | | | | 2 | | | | | 2 | | | +| ------ | --------- | --- | -------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | +| result | discussed | in | the main | text. | | | | | | | | | +References +[1] LIGO Scientific, Virgo Collaboration, B. Abbott et al., Observation of Gravitational Waves +from a Binary Black Hole Merger, Phys. Rev. Lett. 116 (2016), no. 6 061102, +[arXiv:1602.03837]. +[2] LISA Collaboration, P. Amaro-Seoane et al., Laser Interferometer Space Antenna, +arXiv:1702.00786. +[3] G. M. Harry, P. Fritschel, D. A. Shaddock, W. Folkner, and E. S. Phinney, Laser interferometry +for the big bang observer, Class. Quant. Grav. 23 (2006) 4887–4894. [Erratum: Class. Quant. +Grav.23,7361(2006)]. +[4] MAGIS Collaboration, P. W. Graham, J. M. Hogan, M. A. Kasevich, S. Rajendran, and +R. W. Romani, Mid-band gravitational wave detection with precision atomic sensors, +arXiv:1711.02225. +| | | | | | | – | 23 – | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | + +[5] S. Kawamura et al., The Japanese space gravitational wave antenna: DECIGO, Class. Quant. +Grav. 28 (2011) 094011. +[6] L. Grishchuk, Amplification of gravitational waves in an istropic universe, Sov. Phys. JETP 40 +(1975) 409–415. +[7] A. A. Starobinsky, Spectrum of relict gravitational radiation and the early state of the universe, +JETP Lett. 30 (1979) 682–685. +[8] V. Rubakov, M. Sazhin, and A. Veryaskin, Graviton Creation in the Inflationary Universe and +the Grand Unification Scale, Phys. Lett. B 115 (1982) 189–192. +[9] M. Guzzetti, N. Bartolo, M. Liguori, and S. Matarrese, Gravitational waves from inflation, Riv. +Nuovo Cim. 39 (2016), no. 9 399–495, [arXiv:1605.01615]. +[10] S. Khlebnikov and I. Tkachev, Relic gravitational waves produced after preheating, Phys. Rev. D +56 (1997) 653–660, [hep-ph/9701423]. +[11] R. Easther and E. A. Lim, Stochastic gravitational wave production after inflation, JCAP 04 +(2006) 010, [astro-ph/0601617]. +[12] R. Easther, J. Giblin, John T., and E. A. Lim, Gravitational Wave Production At The End Of +Inflation, Phys. Rev. Lett. 99 (2007) 221301, [astro-ph/0612294]. +[13] J. Garcia-Bellido and D. G. Figueroa, A stochastic background of gravitational waves from +hybrid preheating, Phys. Rev. Lett. 98 (2007) 061302, [astro-ph/0701014]. +[14] J. Garcia-Bellido, D. G. Figueroa, and A. Sastre, A Gravitational Wave Background from +Reheating after Hybrid Inflation, Phys. Rev. D 77 (2008) 043517, [arXiv:0707.0839]. +[15] J. F. Dufaux, A. Bergman, G. N. Felder, L. Kofman, and J.-P. Uzan, Theory and Numerics of +Gravitational Waves from Preheating after Inflation, Phys. Rev. D 76 (2007) 123517, +[arXiv:0707.0875]. +[16] C. Caprini et al., Detecting gravitational waves from cosmological phase transitions with LISA: +an update, JCAP 03 (2020) 024, [arXiv:1910.13125]. +[17] C. Caprini and D. G. Figueroa, Cosmological Backgrounds of Gravitational Waves, Class. +Quant. Grav. 35 (2018), no. 16 163001, [arXiv:1801.04268]. +[18] N. Christensen, Stochastic Gravitational Wave Backgrounds, Rept. Prog. Phys. 82 (2019), no. 1 +016903, [arXiv:1811.08797]. +[19] V. Acquaviva, N. Bartolo, S. Matarrese, and A. Riotto, Second order cosmological perturbations +from inflation, Nucl. Phys. B 667 (2003) 119–148, [astro-ph/0209156]. +[20] S. Mollerach, D. Harari, and S. Matarrese, CMB polarization from secondary vector and tensor +modes, Phys. Rev. D 69 (2004) 063002, [astro-ph/0310711]. +[21] D. Baumann, P. J. Steinhardt, K. Takahashi, and K. Ichiki, Gravitational Wave Spectrum +Induced by Primordial Scalar Perturbations, Phys. Rev. D 76 (2007) 084019, [hep-th/0703290]. +[22] J. R. Espinosa, D. Racco, and A. Riotto, A Cosmological Signature of the SM Higgs Instability: +Gravitational Waves, JCAP 1809 (2018), no. 09 012, [arXiv:1804.07732]. +[23] K. Kohri and T. Terada, Semianalytic calculation of gravitational wave spectrum nonlinearly +– 24 – + +induced from primordial curvature perturbations, Phys. Rev. D 97 (2018), no. 12 123532, +[arXiv:1804.08577]. +[24] N. Seto and J. Yokoyama, Probing the equation of state of the early universe with a space laser +| interferometer, | J. Phys. | Soc. Jap. 72 | (2003) 3082–3086, | [gr-qc/0305096]. | +| --------------- | -------- | ------------ | ----------------- | ---------------- | +[25] L. A. Boyle and P. J. Steinhardt, Probing the early universe with inflationary gravitational +| waves, Phys. | Rev. D 77 | (2008) 063504, | [astro-ph/0512014]. | | +| ------------ | --------- | -------------- | ------------------- | --- | +[26] Y. Watanabe and E. Komatsu, Improved Calculation of the Primordial Gravitational Wave +Spectrum in the Standard Model, Phys. Rev. D 73 (2006) 123515, [astro-ph/0604176]. +[27] L. A. Boyle and A. Buonanno, Relating gravitational wave constraints from primordial +nucleosynthesis, pulsar timing, laser interferometers, and the CMB: Implications for the early +| Universe, | Phys. Rev. D | 78 (2008) 043531, | [arXiv:0708.2279]. | | +| --------- | ------------ | ----------------- | ------------------ | --- | +[28] R. Jinno, T. Moroi, and K. Nakayama, Probing dark radiation with inflationary gravitational +| waves, Phys. | Rev. D 86 | (2012) 123502, | [arXiv:1208.0184]. | | +| ------------ | --------- | -------------- | ------------------ | --- | +[29] C. Caprini et al., Science with the space-based interferometer eLISA. II: Gravitational waves +from cosmological phase transitions, JCAP 04 (2016) 001, [arXiv:1512.06239]. +[30] M. Geller, A. Hook, R. Sundrum, and Y. Tsai, Primordial Anisotropies in the Gravitational +Wave Background from Cosmological Phase Transitions, Phys. Rev. Lett. 121 (2018), no. 20 +201303, [arXiv:1803.10780]. +[31] K. Saikawa and S. Shirai, Primordial gravitational waves, precisely: The role of thermodynamics +| in the Standard | Model, | JCAP 1805 | (2018) 035, [arXiv:1803.01038]. | | +| --------------- | ------ | --------- | ------------------------------- | --- | +[32] Y. Cui, M. Lewicki, D. E. Morrissey, and J. D. Wells, Probing the pre-BBN universe with +gravitational waves from cosmic strings, JHEP 01 (2019) 081, [arXiv:1808.08968]. +[33] R. R. Caldwell, T. L. Smith, and D. G. Walker, Using a Primordial Gravitational Wave +Background to Illuminate New Physics, Phys. Rev. D 100 (2019), no. 4 043513, +[arXiv:1812.07577]. +[34] F. D’Eramo and K. Schmitz, Imprint of a scalar era on the primordial spectrum of gravitational +| waves, Phys. | Rev. Research. | 1 (2019) | 013010, [arXiv:1904.07870]. | | +| ------------ | -------------- | -------- | --------------------------- | --- | +[35] N. Bernal and F. Hajkarim, Primordial Gravitational Waves in Nonstandard Cosmologies, +| Phys. Rev. | D 100 (2019), | no. 6 063502, | [arXiv:1905.10410]. | | +| ---------- | ------------- | ------------- | ------------------- | --- | +[36] D. G. Figueroa and E. H. Tanin, Ability of LIGO and LISA to probe the equation of state of the +| early Universe, | JCAP | 1908 (2019) 011, | [arXiv:1905.11960]. | | +| --------------- | ---- | ---------------- | ------------------- | --- | +[37] P. Auclair et al., Probing the gravitational wave background from cosmic strings with LISA, +| JCAP 04 | (2020) 034, | [arXiv:1909.00819]. | | | +| ------- | ----------- | ------------------- | --- | --- | +[38] C.-F. Chang and Y. Cui, Stochastic Gravitational Wave Background from Global Cosmic +| Strings, Phys. | Dark Univ. | 29 (2020) | 100604, [arXiv:1910.04781]. | | +| -------------- | ---------- | --------- | --------------------------- | --- | +[39] F. Hajkarim and J. Schaffner-Bielich, Thermal History of the Early Universe and Primordial +Gravitational Waves from Induced Scalar Perturbations, Phys. Rev. D 101 (2020), no. 4 043522, +[arXiv:1910.12357]. +[40] Y. Gouttenoire, G. Servant, and P. Simakachorn, Beyond the Standard Models with Cosmic +| Strings, JCAP | 07 (2020) | 032, [arXiv:1912.02569]. | | | +| ------------- | --------- | ------------------------ | --- | --- | +– 25 – + +[41] Y. Gouttenoire, G. Servant, and P. Simakachorn, BSM with Cosmic Strings: Heavy, up to EeV +| mass, Unstable | | Particles, JCAP | 07 (2020) | 016, [arXiv:1912.03245]. | | +| -------------- | --- | --------------- | --------- | ------------------------ | --- | +[42] G. Dom`enech, Induced gravitational waves in a general cosmological background, Int. J. Mod. +| Phys. D | 29 (2020), | no. 03 | 2050028, [arXiv:1912.05583]. | | | +| ------- | ---------- | ------ | ---------------------------- | --- | --- | +[43] S. Blasi, V. Brdar, and K. Schmitz, Fingerprint of Low-Scale Leptogenesis in the Primordial +| Gravitational-Wave | | Spectrum, | arXiv:2004.02889. | | | +| ------------------ | --- | --------- | ----------------- | --- | --- | +[44] G. Dom`enech, S. Pi, and M. Sasaki, Induced gravitational waves as a probe of thermal history of +| the universe, | JCAP | 08 (2020) | 017, [arXiv:2005.12314]. | | | +| ------------- | ---- | --------- | ------------------------ | --- | --- | +[45] R. Allahverdi et al., The First Three Seconds: a Review of Possible Expansion Histories of the +| Early Universe, | | arXiv:2006.16182. | | | | +| --------------- | --- | ----------------- | --- | --- | --- | +[46] C. Caprini, R. Durrer, T. Konstandin, and G. Servant, General Properties of the Gravitational +Wave Spectrum from Phase Transitions, Phys. Rev. D79 (2009) 083519, [arXiv:0901.1661]. +[47] G. Barenboim and W.-I. Park, Gravitational waves from first order phase transitions as a probe +of an early matter domination era and its inverse problem, Phys. Lett. B 759 (2016) 430–438, +[arXiv:1605.03781]. +[48] R.-G. Cai, S. Pi, and M. Sasaki, Universal infrared scaling of gravitational wave background +spectra, arXiv:1909.13728. +[49] H.-K. Guo, K. Sinha, D. Vagie, and G. White, Phase Transitions in an Expanding Universe: +Stochastic Gravitational Waves in Standard and Non-Standard Histories, arXiv:2007.08537. +[50] J. Ellis, M. Lewicki, and V. Vaskonen, Updated predictions for gravitational waves produced in a +| strongly | supercooled | phase | transition, arXiv:2007.15586. | | | +| -------- | ----------- | ----- | ----------------------------- | --- | --- | +[51] S. Weinberg, Damping of tensor modes in cosmology, Phys. Rev. D 69 (2004) 023503, +[astro-ph/0306304]. +[52] R. Jinno and M. Takimoto, Gravitational waves from bubble dynamics: Beyond the Envelope, +| JCAP 01 | (2019) | 060, [arXiv:1707.03111]. | | | | +| ------- | ------ | ------------------------ | --- | --- | --- | +[53] T. Konstandin, Gravitational radiation from a bulk flow model, JCAP 03 (2018) 047, +[arXiv:1712.06869]. +[54] D. Cutting, E. G. Escartin, M. Hindmarsh, and D. J. Weir, Gravitational waves from vacuum +| first order | phase | transitions | II: from thin | to thick walls, | arXiv:2005.13537. | +| ----------- | ----- | ----------- | ------------- | --------------- | ----------------- | +[55] R. Jinno, T. Konstandin, and H. Rubira, A hybrid simulation of gravitational wave production +| in first-order | phase | transitions, | arXiv:2010.00971. | | | +| -------------- | ----- | ------------ | ----------------- | --- | --- | +[56] Planck Collaboration, N. Aghanim et al., Planck 2018 results. VI. Cosmological parameters, +| Astron. | Astrophys. | 641 (2020) | A6, [arXiv:1807.06209]. | | | +| ------- | ---------- | ---------- | ----------------------- | --- | --- | +[57] D. Baumann, D. Green, J. Meyers, and B. Wallisch, Phases of New Physics in the CMB, JCAP +| 01 (2016) | 007, | [arXiv:1508.06342]. | | | | +| --------- | ---- | ------------------- | --- | --- | --- | +[58] C. Brust, Y. Cui, and K. Sigurdson, Cosmological Constraints on Interacting Light Particles, +| JCAP 08 | (2017) | 020, [arXiv:1703.10732]. | | | | +| ------- | ------ | ------------------------ | --- | --- | --- | +[59] N. Blinov and G. Marques-Tavares, Interacting radiation after Planck and its implications for +| the Hubble | Tension, | arXiv:2003.08387. | | | | +| ---------- | -------- | ----------------- | --- | --- | --- | +– 26 – + +[60] Y. Cui, M. Lewicki, D. E. Morrissey, and J. D. Wells, Cosmic Archaeology with Gravitational +Waves from Cosmic Strings, Phys. Rev. D97 (2018), no. 12 123505, [arXiv:1711.03104]. +– 27 – \ No newline at end of file diff --git a/q143/random_k4/random_4.md b/q143/random_k4/random_4.md index 5526b8c35c6c0d7d05b7d26b15fce08f69e7eb06..9509e6da1ad54da1cbe8e93cf49ed222beadcc6c 100644 --- a/q143/random_k4/random_4.md +++ b/q143/random_k4/random_4.md @@ -1,1865 +1,864 @@ -Morgan Stanley Investment Management +Business+More -Asset Management Overview + THE UNDERGRADUATE PROGRAM -July 2006 + Explore ideas, +think strategically, +make an impact. -Page + Business makes things happen +and affects every part of society. +Whatever you want to do, +a degree from Wharton will +give you the knowledge +and experience to do it. 1 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - The information provided herein may include certain non-GAAP financial measures. -The reconciliation of such measures to the comparable GAAP figures are included in -the Company’s Annual Reports on Form 10-K, Quarterly Reports on Form 10-Q and -Current Reports on 8-K, including any amendments thereto, which are available at -www.morganstanley.com. - -This presentation may contain forward-looking statements. You are cautioned not to -place undue reliance on forward-looking statements, which speak only as of the date -on which they are made, which reflect management’s current estimates, projections, -expectations or beliefs and which are subject to risks and uncertainties that may cause -actual results to differ materially. For a discussion of risks and uncertainties that may -affect the future results of the Company, please see “Forward-Looking Statements” -immediately preceding Part I, Item 1, “Competition” and “Regulation” in Part I, Item 1, -“Risk Factors” in Part I, Item 1A and “Certain Factors Affecting Results of Operations” -in Part II, Item 7 of the Company’s Annual Report on Form 10-K for the fiscal year -ended November 30, 2005, “Management’s Discussion and Analysis of Financial -Condition and Results of Operations” in the Company’s Quarterly Reports on Form 10- -Q for the quarterly periods ended February 28, 2006 and May 31, 2006 and other items -throughout the Form 10-K, Form 10-Qs and the Company’s 2006 Current Reports on -Form 8-K. - -Page + Business + Liberal Arts -2 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Table of contents - -Section 1 - -Strengths of MSIM Today - -Section 2 - -MSIM Growth Strategy - -Page - -3 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Section 1 - -Strengths of MSIM Today - -Page - -4 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Is a Key Business of Morgan Stanley - -Morgan Stanley - -STRENGTHS OF MSIM TODAY - -Net revenues -Fiscal year 2005=$26,778 million - -19% - -Asset -Management -11% - -12% - -Profit before taxes -Fiscal year 2005=$7,361 million - -8% - -Asset -Management -14% - -65% - -13% - -58% - -Profit before taxes -$ million - -1,007 - -827 - -482 - -462 - -396 - -Institutional Securities Global Wealth Management - -Asset Management Discover - -Asset Management - -Pre-tax profit margin - -30% - -26% - -35% - -29% - -28% - -25% - -22% - -21% - -Return on average -common equity - -36% - -24% - -23% - -17% - -FY 2003 - -FY 2004 - -FY 2005 - -1st Half -2006 - -(1) - -FY 2003 - -FY 2004 - -FY 2005 - -1st Half 2006 - -(1) - -FY 2005 - -1st Half 2006 - -(1) - -MSIM - -MSIM excl. Private Equity - -MSIM +20 -Morgan Stanley +course units +in business -1. Six months ended May 31 +10 -Page +course units +in liberal arts +and sciences 5 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Key Strengths - -STRENGTHS OF MSIM TODAY - -• Strong brand recognition - -– Morgan Stanley - -– Van Kampen - -• Global footprint - -• Scale - -• Breadth of products - -• Multiple channels of distribution - -• Community of boutiques - -• Full firm commitment and resources - -Page - -6 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Has a Global Footprint - -Products and Distribution - -STRENGTHS OF MSIM TODAY - -New York - -Amsterdam - -Stockholm - -Oakbrook Terrace - -Boston - -London +unrestricted +electives +anywhere at Penn -Paris - -Frankfurt - -Luxembourg - -San Francisco -Menlo Park - -Houston - -Jersey City - -West Conshohocken - -Milan - -Madrid - -Mumbai - -Hong Kong - -Tokyo - -Singapore - -Melbourne - -MSIM Office Present - -MSIM Distribution Coverage - -Page - -7 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - STRENGTHS OF MSIM TODAY - -Assets ($ billion) +2 -1,513 +leadership +course units -1,439 +Penn uses a course unit (CU) system. Most courses are 1 CU. -1,422 + With Wharton’s combination of business and liberal arts -• In June 2006, Pensions and -Investments ranked Morgan -Stanley #11 by worldwide -assets +in the Ivy League, you will gain a strong foundation of skills -• MSIM’s strength is in +and a framework for thinking creatively and strategically to -traditional, actively managed -investments +solve problems. You’ll have the flexibility to follow your interests -MSIM Has The Scale to Succeed +as you engage in hands-on experiences, gain invaluable practice -Top 20 money managers +working in teams, and learn how to use your knowledge and -Pensions and Investments, June 2006 +skills to make a difference. -Ranking +Join our students as they put knowledge into action. -1 +As the leader in business education, we’ll be there to empower -2 +you every step of the way. Are you up for the challenge? 3 -4 + Explore +Business +and More -5 - -6 - -7 +Red Rocks Desert Leadership Trek -8 +Wharton provides innovative interdisciplinary options and active +experiential learning in business, combined with all the resources of Penn. -9 +Start your own business, manage a clothing brand, or even build a new car +—with a degree from Wharton you can go wherever your path leads you +and be more effective in the career you choose. -10 + 4,200 -11 +courses 12 -13 - -14 - -15 - -16 - -17 - -18 - -19 - -20 - -Barclays Global - -State Street Global - -Fidelity Investments - -Vanguard Group +schools to +explore your +interests — +4 undergraduate +and 8 graduate + +Flexibility and Opportunity +Every semester you can take classes at +Wharton and at least one other school +at Penn, including eight of our graduate +schools. In fact, 40% of your required +courses may be taken outside of Wharton. +Why? Because studying business within +social, political, and historical contexts +makes you a better agent of change. +It also means that you will have even more +opportunities and paths to choose from +when you graduate. + +The Wharton BS in Economics— +It’s a Different Way of Learning +Wharton offers a bachelor of science +in economics, but studying business at +Wharton and studying economics in a liberal +arts program are very different. + +If you major in economics at a liberal arts +school, you’ll take courses focused on +theories, history, and modeling. You’ll start +with a theory or abstract idea and then look +for problems to understand how the theory +applies. + +At Wharton you’ll study a range of +business subjects—from accounting to +entrepreneurship to policy. + +Classes focus on problem solving with real- +world examples; you’ll look for solutions +and then examine theories to see how they +might apply. -Legg Mason - -Capital Research - -JPMorgan Asset Mgmt. - -Mellon Financial - -Deutsche Asset Mgmt. - -Northern Trust Global - -Morgan Stanley - -PIMCO - -UBS Global Asset Mgmt. - -AllianceBernstein Inst'l - -AIG Global Investment - -Merrill Lynch - -Prudential Financial +5 -ING + Get the Best of Both Worlds -Wellington Mgmt. +Your education is both Wharton and Penn. Our rigorous and flexible +curriculum combines the best in business education with exploration in the +arts and sciences. Follow your interests wherever they take you. -Goldman Sachs Asset +Gain a Grounding in Business +You’ll get a strong foundation in business by taking courses at Wharton +in diverse subjects like legal studies, marketing, and finance. -958 +Pursue Diverse Interests +You’ll take 10+ courses in the liberal arts and sciences, including a Critical +Writing Seminar, and choose five unrestricted electives from any school at Penn. -864 +Expand Your Options with Interdisciplinary Study +You can apply for one of our globally recognized coordinated +dual-degree programs: -858 +INTERNATIONAL STUDIES & BUSINESS: THE HUNTSMAN PROGRAM -842 +MANAGEMENT & TECHNOLOGY: THE JEROME FISHER PROGRAM -781 +NURSING & HEALTH CARE MANAGEMENT -633 +LIFE SCIENCES & MANAGEMENT: THE ROY & DIANA VAGELOS PROGRAM -618 +Or, once you’re a student, you can create your own course of study— +pursue a minor or a second degree in a completely different subject. -599 + Online — Dual degrees and minors -594 + Wharton Concentrations +We don’t have traditional majors. Instead, you’ll pick a concentration—4 upper-level courses +in a business specialty of your choice: -581 +ACCOUNTING -579 +ACTUARIAL SCIENCE -572 +BEHAVIORAL ECONOMICS -537 +BUSINESS ANALYTICS -531 +BUSINESS ECONOMICS -523 +& PUBLIC POLICY -521 +ENVIRONMENTAL POLICY -496 +& MANAGEMENT -Investment type: +FINANCE -T ra dit io na l +GLOBAL ANALYSIS -A lt e rna t iv e s +MANAGEMENT -Inde x +Entrepreneurship +& Innovation -0 +Multinational Management -250 +Organizational Effectiveness -500 +Strategic Management -750 +MANAGING ELECTRONIC -1,000 +COMMERCE -1,250 +MARKETING -1,500 +HEALTH CARE MANAGEMENT -1,750 +MARKETING & COMMUNICATION -Source: Total assets are from June 2006 Pensions and Investments, using year end 2005 data. Morgan Stanley assets reflect Morgan Stanley consolidated assets -at fiscal year end 2005. Investment types, which are derived from June 2006 Pension and Investments and July 2006 Institutional Investors, are for comparison -purposes +& POLICY -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +MARKETING & OPERATIONS -Page +LEGAL STUDIES & BUSINESS ETHICS -8 +MANAGEMENT - MSIM Organization 2006 +OPERATIONS, INFORMATION -MSIM GROWTH STRATEGY +& DECISIONS -• Investment expertise is +Decision Processes -organized around investment -product areas +Information Systems -• Distribution is organized +Operations Management/ +Management Science -around clients +REAL ESTATE -John Mack -Firm CEO +RETAILING -Bob Scully -Firm Co-President +SOCIAL IMPACT & RESPONSIBILITY -Owen Thomas -MSIM President, COO +STATISTICS -Investments +INDIVIDUALIZED -U.S. & Americas Distribution +The Quadrangle, home to three +of Penn’s 11 College Houses -Non-Americas +30% -• Focused on communicating -effectively among the groups +of Wharton +students earn +more than +one degree -Dennis Shea -Head and CIO, -Global Equity +48% -David Germany -Head and CIO, -Global -Fixed Income +choose more +than one +Wharton +concentration -Stuart Bohart -Head and CIO, -Alternatives +7 -Michael Kiley -Head of Americas -Retail & -Intermediary -Distribution + CONCENTRATIONS +Marketing & Operations +Management and Environmental +Policy & Management -Open -Head of -U.S. Institutional -Distribution +q -Michael Green -Head of MSIM -Non-Americas +"At Penn, I have really been able to utilize my leadership +skills and take initiative when I find something important. +For example, I felt it was necessary for the Black Wharton +Undergraduate Association to have an Ivy League-wide +business conference, and I worked to make that a reality." -Kenneth Winston -Chief Risk Officer +Alexis-Amanda Malcolm | Carthage, New York -• Decentralized, with all -components of the -organization accountable for -financial results +CONCENTRATIONS +Finance and Management -Attracting new talent -Internal +"I am amazed by how many different options there +are, and how diverse the people that you meet are. +Everyone chooses a different path to graduation." -• Head and CIO Global Equity +q -• Head and CIO Alternatives +Jacob Söderstjerna | Malmö, Sweden -• Head of Operations +CONCENTRATIONS +Finance and Management +Life Sciences & Management +dual-degree program -• Two investment teams: +q -External +"Last spring, as part of the Wharton International Program +(WIP), I traveled to Buenos Aires, Argentina, and visited a +wide range of companies to broaden my perspective on how +business is conducted in other regions of the world. +I have also gotten the chance to participate in consulting case +competitions, in which students work in teams to solve some +of the toughest operational or strategic issues for companies." -• CIO London Equity +Emily Zhen | Chicago, Illinois -• Two investment teams in Absolute Return Strategies + q -• Process-Driven team +CONCENTRATIONS +Management and +Behavioral Economics -• Global Mid-Cap Equity Market Neutral team +"Having interests in the marketing, retail, and +entertainment industries, I hope to work with a +marketing agency in a position that allows me to +combine the analytical and creative halves of myself. +The best thing about being a Wharton student is that +your education prepares you to work in all industries, +so choosing only one isn’t necessary!" -• Defined Value team, Absolute Return Strategies +Christopher Richardson | Portland, Maine -• Structured Products and Derivatives team +Challenge +Yourself -• Pension Asset/Liability expertise +q -• Global Infrastructure sector expertise +CONCENTRATIONS +Marketing and +Finance +Minor: French and +Francophone Studies -• Six investment professionals: +"Wharton gives me the freedom and +flexibility to pursue other interests +outside of business. I love being able to +integrate my business education with a +liberal arts education." -• European Senior Loans +Isabel Solares | Mexico City, Mexico -• Alternative Investment Partners +q -• Global Value +CONCENTRATIONS +Finance, Business +Economics & Public Policy, +and Management -• International Value +"Wharton/Penn makes failing impossible. +You can get a low GPA, do not get me wrong, but +you cannot leave this school a failure. Penn is the +kind of place that pushes you to be the very best +you can be because everyone else around you is +reaching for the stars. So, even if you reach for the +stars and land on the moon, you are not a failure." -Page +Bryan Mena-Martinez | Maywood, California 9 -• Senior U.S. Institutional sales professional - -• Head of Morgan Stanley Funds Distribution - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - -• Senior Middle East sales professional - - • MSIM offers a broad array of -traditional products and is -increasing its offering of -alternative products - -• MSIM utilizes multiple - -channels of distribution -globally leveraging the -Morgan Stanley brand and -resources where possible - -MSIM Product and Distribution Balance - -STRENGTHS OF MSIM TODAY - -May 31, 2006 Ending AUM/AUS = $440 billion - -Asset Class - -3% 3% - -5% - -Distribution - -20% - -22% - -17% - -21% - -51% - -Equity - -Fixed Income - -Money Market - -Alternatives - -Real Estate - -Unit Investment Trusts - -23% - -20% - -15% - -Americas Retail-Morgan Stanley Branded - -Americas Retail-Van Kampen Branded - -Americas Intermediary - -U.S. Institutional + Experience +Innovative +Learning + +As the first collegiate business +school, Wharton invented +undergraduate business +education, and we’re still +innovating. Wharton students +learn the latest business ideas +as they’re being created. + +Learn from the Best +Our 225+ professors—the largest, most +published, faculty at any business school— +work directly with leading global businesses, +nonprofit organizations, and government +agencies. + +Our classes are challenging and +intellectually rigorous, but, at the same time, +they’re collaborative and full of interaction. +Professors customize their teaching +methods to the needs of each class, so you’ll +find everything from the Socratic method +and traditional lectures to case studies and +computer simulations. + +All Wharton undergraduate lectures are +taught by professors, not graduate teaching +assistants. Our professors are available to + +Prof. Maurice Schweitzer works with an +Operations, Information & Decisions student + +7+ + +research +programs -Non-U.S. - -Page - -10 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Offers Clients a Broad Product Array - -Key Product Type by Investment Group - -STRENGTHS OF MSIM TODAY - -Global Fixed Income -J. David Germany, CIO -AUM=$166 billion (1) - -• Core/Core Plus -• Long Duration -• Mortgage -• Corporate -• Targeted Duration -• Intermediate Duration - -• High Yield -• Short Duration -• Enhanced LIBOR - -• Customized -• Global/International -• Emerging Markets Debt - -• Institutional Liquidity -• Enhanced Cash -• Retail Money Markets - -Global Equity -Dennis Shea, CIO -AUM=$241 billion (1) - -U.S. Equity -• U.S. Small-, Mid-, Multi-, and Large-Cap - -Value - -• U.S. Small-, Mid-, Multi-, and Large-Cap - -Growth - -• U.S. Large-Cap Core -• Systematic Active Equity -• Global REITs (U.S./European/Asian) -• Sector Funds - -Global/International Equity -• International Value -• International Growth -• International Small-Cap -• Global Emerging Markets -• Global Core -• Global Small- and Large-Cap Value -• Global Systematic Active Equity -• European -• Japan (Small-Cap, Value, Growth, Core) -• Asian (ex. Japan) and Pacific Basin - -Asset Allocation -• Balanced Funds -• Global Tactical Asset Allocation - -– Fully Funded - -– Global Macro - -1. Ending AUM/AUS as of May 31, 2006 - -Alternatives -Stuart Bohart, CIO -AUM=$20 billion (1) - -• Absolute Return Strategies (Hedge Funds) - -– Quantitative Equity Strategies -– Fundamental Equity Strategies -– Global Macro Strategies - -• Alternative Investment Partners - -– Fund of Hedge Funds -– Fund of Private Equity Funds -– Insurance Dedicated Fund of Hedge - -Funds - -– Portable Alpha - -• Senior Loans - -− Collateralized Loan Obligations - -• Structured Products and Derivatives - -• Managed Futures - -• Private Equity - -Unit Trusts -AUS=$13 billion (1) - -• Unit Investment Trusts +12 -Page +programs in +entrepreneurship + + our students. They hold office hours, have +lunch with students, serve as concentration +advisors, and mentor students for research +projects. + +Research Opportunities +Look deeper into the questions that +interest you by engaging in research. +You can customize what you’re learning +and pursue a topic in depth with help +from a faculty scholar through a School- +sponsored program, summer experience, or +independent study. + +Entrepreneurship +Interested in starting your own company? +There isn’t a better place to be if you have +a business idea of your own. Wharton has + +one of the largest programs with faculty, +resources, and know-how to help you make +your idea a reality. + +Technology to Enhance Learning +Wharton provides a host of innovative +resources that redefine how business +is studied and how business insights +are shared. You can manage a billion- +dollar stock portfolio, control a country’s +fishing fleet, or prepare a marketing plan +for a totally new product—all through +simulations. + +Penn Public Policy Internship at +Securities and Exchange Commission + +Undergraduate visit to Google 11 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Utilizes a Multi-Channel Distribution Approach - -STRENGTHS OF MSIM TODAY - -Core Distribution -Channels - -Customer Tiers - -Americas Retail- -Morgan Stanley Branded - -Americas Retail- -Van Kampen Branded - -Americas Intermediary - -U.S. Institutional - -Non-U.S. - -Morgan Stanley Global -Wealth Management + Wharton 398, an intensive business simulation for seniors + +Lead and Work in a Team + +We’re serious about working together to achieve common goals. In fact, +our team-based curriculum makes it difficult to succeed if you’re only in it +for yourself. Our virtual classrooms and our home in Huntsman Hall were +intentionally designed to facilitate teamwork. + +The Leadership Journey +You will be prepared for leadership via a suite +of components that spans all four years. +Wharton 101 introduces business disciplines +and the applications of business to real-world +problems. Additional modules focus on +written and oral communications, teamwork, +and interpersonal skills. Integrative group +projects for seniors build on knowledge +gained from the previous three years. + +Beyond the Classroom +Learning about leadership and teamwork +isn’t something that’s restricted to the +classroom. You might be surprised to find +that your greatest lessons are learned +while working on a new business with +your classmates, or consulting for a local +nonprofit while on a Wharton Leadership +Venture. Getting involved is also a great +way to flex your leadership muscles. And +with over 300 clubs at Penn and over 40 +at Wharton, it’s easy to jump right in. + + 10+ + +Wharton +Leadership +Ventures + +40+ + +cohort +leadership +positions + +Hiking the Inca Trail with Wharton Leadership +Ventures, learning experiences structured to +build your individual leadership capabilities. + +Online — Watch YouTube videos of +Wharton Leadership Ventures + +A Leadership Journey +Integrated Over Four Years + +Freshmen +Discover curricular +opportunities (business +pathways), understand +your own strengths and +leadership potential, and +prepare for the Wharton +experience. + +Sophomores +Develop your writing +and presentation skills +for business. + +Juniors +Build your teamwork +and interpersonal skills. + +Seniors +Participate in integrative/ +capstone projects or +courses that tackle applied +problems. + +Add co-curricular elements, such as the Wharton Industry Exploration Program (WIEP) +trips, Wharton Leadership Ventures, Wharton International Program (WIP), Global Modular +Courses, study abroad, and research programs. -• Financial Advisors - -• Private Wealth -Management -Investment -Representatives - -• Broker Dealers - -• Banks - -• Investment Advisors - -• Intermediaries - -• Banks - -• Financial Planners - -• Consultants -• Corporations -• Public Plans -• Foundations -/Endowments - -• Insurance General - -Accounts/ Financial -Institutions +13 -• Consultants + Choose Your International +Experience +We live in a global society, and that’s +reflected in every part of Wharton and Penn. +Nearly all Wharton courses incorporate +international issues and trends. Students +have a language requirement and are +encouraged to study abroad or participate +in international trips. Wharton has over 20 +semester-length programs with the world’s +top business schools, specifically designed +to fit our curriculum, and Penn offers more +than 100 in the liberal arts. + + Online — Global options + +Make a +Better +World + +Your classmates will come from +around the world, and you’ll do +things like study abroad, minor in +a language, or learn from foreign +business executives through +the Wharton International +Program or a Global Modular +Course. You’ll leave with a global +perspective and become part of +a network of 96,000 Wharton +alumni around the world. + +100+ + +international +exchange +options around +the world -• Corporations +21% -• Public Plans +of Wharton +undergrads +come from +outside the U.S. + + Do Something Meaningful +Wharton supports initiatives that +foster social responsibility in and +out of the classroom. For example, +we offer classes on topics like urban +fiscal policy, nonprofits, and impact +investing, as well as Social Impact +Fellowships for undergrads to work +on research and consulting projects +during the academic year and over +the summer. + + Online — Social impact + +Spotlight: College Town within a City +One of the best things about being a student +at Penn is having the city of Philadelphia at +your doorstep. + +The University of Pennsylvania, America’s +first university, boasts a unified campus that +houses all of Penn’s vibrant undergraduate, +graduate, and professional schools—a unique +interdisciplinary environment in historic +Philadelphia. + +Philly is big enough to have lots to do, yet +small enough to get to know. Penn students +live right in the middle of University City, a +neighborhood with its own stores, restaurants, +and movie theaters. And you are just a hop +away from everything Philadelphia has to +offer, whether you want to go for a run in +Fairmount Park, or try out the restaurants and +shops of Rittenhouse Square. + +q + +Wharton International Program in Paris + +Fisher Fine Arts Library on College Green -• Private Banks +15 -• Intermediaries + Penn Fashion Week, run by Penn Fashion +Collective, Delta Sigma Pi Business +Fraternity, and the Wharton Retail Club + +q + +Freshman orientation in Baker Forum + +Join a Dynamic Community + +You will make friends and build lifelong connections in a dynamic +and diverse community that starts at Penn and extends around the world. +You’ll find hundreds of ways to get involved, impact society, pursue +personal passions, and try out new things in a welcoming environment. + +Cohorts and Clubs +Your community at Wharton starts with +your cohort—a group of 60 students +with whom you’ll share academic and +extracurricular experiences during your first +two years at Penn. As an integrated member +of the Penn community, you will also make +deep connections with students from every +school through your College House, your +athletic teams, Greek life, and the clubs and +organizations you join. + +Support for Diversity +At Penn, you’ll find a diverse community +supportive of a unique mix of cultures and +beliefs, as well as the most international +students of any Ivy League university. +You’ll discover clubs, centers, and events +that reflect the incredible diversity of +our community—from the Penn Arab +Student Society to the LGBT Center to +discussion panels held by Check One, a +group dedicated to supporting multiracial +students. + +Advising and Support +From house deans to Wharton’s dedicated +team of professional academic advisors, our +faculty and staff will be there to assist you +every step of the way. Our large support +network helps you get the most out of your +college experience and find the path that’s +right for you. You’ll never be on your own +at Wharton. + +Alumni—Get Connected for Life +Alumni are constantly coming to campus +to talk to current students about their +experiences and to give advice through our +alumni/student network. No matter what +you do in the future, you’ll have lifelong +connections and friendships with the +business leaders of today and tomorrow. +On top of that, you’ll be part of Penn’s +alumni network, with 304,000 graduates +around the world. + +q Members of Wharton cohort Dinar + +Penn’s Locust Walk, the heart of campus + + Online — Find out +which Wharton and Penn +clubs are right for you + +304,000 +Penn alumni + +10,000 +Penn undergraduates + +2,500 +Wharton undergraduates + +You -Core Investment -Vehicles/Products +6 +Wharton 101 project +team members -Morgan Stanley Brand +60 +Freshman cohort +size -Van Kampen Brand +520 +Average entering +class -Van Kampen Brand +96,000 +Wharton alumni -• Separate Accounts +17 -• Separate Accounts + Open More Doors -• Mutual Funds +A Wharton education prepares you to explore and create solutions to +today’s most significant business and social issues. Our graduates are proof +that you will be able to apply what you learn to any industry and any career. -• Mutual Funds +More Options Now +Each year, our graduates take on diverse +jobs at top companies around the world, +including AB InBev, Amazon, Blackstone, +Goldman Sachs, Google, JP Morgan, +Kraft Foods, McKinsey, Microsoft, and +Time Inc. Some Wharton alumni pursue +entrepreneurial ventures full time, while +others advance their studies in law, medicine, +and other fields. -• Mutual Funds +More Growth Later +The knowledge you gain takes on depth +and insight as you blaze your individual path. +Five years after graduation, our alumni -• Pooled Trust +report advancing in a range of careers, +including brand management, education, +entrepreneurship, financial management, +health care, investment banking, law, +nonprofit management, private equity, and +technology. -• Liquidity Products +Connections for Life +You’ll have ties to leaders changing practice +and influencing every industry—Wharton’s +alumni network, the largest of any business +school at 96,000 strong, and Penn’s, +extending to 304,000 leaders in every field +around the world. -• Institutional Funds + Online — Career Plans Survey Report -• Variable Annuities + Site visit through the Baker Retailing Center -• Platform Business +Wharton Women Conference -• Liquidity Products +Post-Graduation Plans +Three-Year Averages -• Société +89% +Full-Time Employed -• Affiliated Separately -Managed Accounts +4% +in Graduate School -• Variable Annuities +$74,434 +Median Salary -• Alternative -Investments +Top Industries -• Unit Investment Trusts +3% COMMUNICATIONS -• Alternative -Investments +21% CONSULTING -• External Separately -Managed Accounts +32% INVESTMENT BANKING -• Alternative -Investments +18% OTHER FINANCE -• Variable Annuities +3% MANUFACTURING -• Institutional Funds +11% TECHNOLOGY -• Unit Investment Trusts +Business Now, MBA Later? +With Wharton’s rigorous business +education, you can begin making an +impact sooner. Nearly two-thirds +of Wharton undergraduate alumni +find that they never need to return +to school for an MBA in order to +advance their careers. -Morgan Stanley Brand +19 -• Institutional Funds + Clubbing Night, a showcase + for Wharton undergraduate clubs -• Liquidity Products +See Yourself Here -• Société +Wharton seeks students who will jump into the rich academic, cultural, and +social opportunities of the Penn community. Visit the Wharton website for all +the information you need to get started. undergrad.wharton.upenn.edu -d'investissement à -Capital Variable -(SICAV) Funds +We Look Forward +to Meeting You! +u Visit Campus -Sub-Advised +u Attend Penn and Wharton + information sessions -• Mutual Funds +u Take a tour -d'investissement à -Capital Variable -(SICAV) Funds +u Sit in on a class -• Alternative -Investments +Apply +Office of Undergraduate Admissions +University of Pennsylvania +1 College Hall +Philadelphia, PA 19104.6376 ++1.215.898.7507 +info@admissions.upenn.edu -• Investment Trust +www.admissions.upenn.edu -Management -Company (ITM) Funds - -Ending AUM/AUS -As of May 31, 2006 - -Page - -12 - -$96 billion - -$90 billion - -$67 billion - -$98 billion - -$89 billion - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - • Performance is key to driving growth - -MSIM Investment Performance - -• On February 22nd, 2006, Barron’s - -U.S. Mutual Funds Performance - -STRENGTHS OF MSIM TODAY - -annual ranking placed MSIM - -– 30th out of 65 fund families for 1- - -year performance - -– 24th out of 56 for 5-year - -performance - -– 4th out of 40 for 10-year - -performance - -– The 1-year rankings placed MSIM - -outpacing Vanguard, Franklin - -Templeton, JPMorgan and Putnam - -• On July 5th, 2006, Wall Street Journal - -named seven MSIM funds as "Category - -Kings": - -Large-Cap Growth - -– Morgan Stanley Institutional Focus - -Equity - -% Long-Term Fund Assets Above Lipper Median - -67% - -61% - -60% - -48% - -71% - -59% - -61% - -63% - -71% - -71% - -76% - -76% - -1 Year - -3 Year - -5 Year - -Aug'05 - -Nov'05 - -Feb'06 - -May'06 - -% Rated Assets with Morningstar 4/5-Star Rating - -– Morgan Stanley Institutional U.S. - -47% - -45% - -Large Cap Growth - -– Morgan Stanley Growth - -Multicap Growth - -– Morgan Stanley Capital - -Opportunities - -– Morgan Stanley Aggressive Equity - -Pacific Region - -– Morgan Stanley Pacific Growth - -Real Estate - -– Morgan Stanley Institutional - -International Real Estate - -Page - -13 - -39% - -39% - -Aug'05 - -Nov'05 - -Feb'06 - -May'06 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Section 2 - -MSIM Growth Strategy - -Page - -14 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Growth Strategy - -MSIM GROWTH STRATEGY - -Our Strategy for Growth - -Critical Initiatives - -• Increase value of Morgan Stanley’s Asset - -• Build alternatives capability - -Management business - -— PBT growth - -— Multiple (P/E) expansion by - -• Create new traditional products using existing -capabilities and through lift-outs / acquisitions - -• Expand non-U.S. footprint: distribution and - -demonstrating net sales and AUM growth - -products - -• Focus on revenue growth and diversification - -• Invest in Van Kampen - Intermediary and Unit - -by investing in the business - -Investment Trusts growth - -— People - -— Acquisitions - -— Infrastructure - -• Leverage Morgan Stanley franchise and - -MSIM scale and positioning - -• Enhance traditional product strength and build - -Alternatives - -• Performance, performance, performance - -• Stem outflows of Morgan Stanley retail mutual - -funds - -• Rebuild institutional reputation and business, - -particularly in the U.S. - -Page - -15 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Build a Much Broader Alternatives Capability - -What We Have - -What We’ve Done - -What We’re Doing - -MSIM GROWTH STRATEGY - -• The Alternative Investments - -group is composed of 4 distinct -groups: - -– Absolute Return Strategies - -(ARS) - -– Alternative Investment - -Partners (AIP) - -– Senior Loans - -– Structured Products and - -Derivatives (SP&D) - -First Quarter 2006 -• Internal transfer of Structured -Products and Derivatives team - -• Completed lift-out of Process-Driven - -Strategies team - -• Launched / incubated 6 new - -alternative investment products - -Second Quarter 2006 -• Hired Stuart Bohart as the head and - -CIO of Alternatives - -• Completed Oxhead Capital - -acquisition - -• Internal transfer of Defined Value - -team - -• Internal transfer of sector expertise - -to run the global infrastructure -initiative (initiated June 2006) - -• Launched Japanese Equities Market - -Neutral product - -• Re-opened Traxis to new investors - -Absolute Return Strategies - -• Continue efforts on lift-out / -acquisitions, and minority -stakes to build business - -Alternative Investment Partners -• Expand AIP product line and - -introduce new funds - -• Distribution focus on smaller - -mid-market and high net worth - -Senior Loans - -• Growth driven by client focused - -new products through -introduction of funds, -Collateralized Loan Obligations, -Separately Managed Accounts - -Structured Products and -Derivatives - -• Develop custom approaches to - -build structured products -• Introduce Liability Driven - -Investments - -Page - -16 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Enhance Traditional Product Offerings - -Through both Organic Launches and Bolt-on Acquisitions/Lift-Outs -What We Have - -What We’ve Done - -• Many top performing equity -strategies at or near capacity - -• Fixed Income products with -improved performance, and -capacity - -• Good performing strategies - -building three year track records - -• Product gaps in large asset - -gathering categories - -First Quarter 2006 -• Launched / incubated 5 new - -equity products - -Second Quarter 2006 -• Launched / incubated 8 new - -products - -– 7 Equity -– 1 Fixed Income - -MSIM GROWTH STRATEGY - -What We’re Doing - -Organic launches - -• Leverage current talent pool to - -build out strong product -pipeline - -Acquisitions / lift-outs - -• Targeted acquisitions / lift-outs -to complement our existing -product line - -Page - -17 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Expand Non-U.S. Footprint - -Distribution and products - -What We Have - -• $89 billion, or 20% of MSIM - -AUM/AUS is sourced from non- -U.S.(1) - -• MSIM is represented in 13 non- - -U.S. countries - -• Outside of the U.S., portfolio - -managers are located in: - -– Europe -– Japan -– Singapore -– India - -What We’ve Done - -First Quarter 2006 -• Increased distribution focus on - -Alternatives - -• Signed a global agreement with - -Merrill Lynch Global Private -Client division to distribute MSIM -products through Merrill Lynch's -International distribution network -worldwide - -Second Quarter 2006 -Middle East -• Hired sales head -• Leveraging Firm’s Dubai office - -China -• Dedicated sales hire -• Applied (and gained the approval - -in July 2006) for license to -manage domestic Chinese equity -products - -MSIM GROWTH STRATEGY - -What We’re Doing - -Expand the footprint -• Asia - -– Korea -– Australia -– China -– India - -• Europe - -– Middle East -– Switzerland - -• Fill product gaps - -• Explore new client opportunities: - -central banks and direct -wholesaling - -• Evaluate acquisitions in - -– Australia -– China -– Brazil -– Korea -– Taiwan - -1. - -Total MSIM AUM/AUS of $440 billion as of May 31, 2006 - -Page - -18 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Invest in Van Kampen - Intermediary and Unit Investment Trusts Growth -What We Have - -What We’re Doing - -What We’ve Done - -MSIM GROWTH STRATEGY - -• Vibrant $100+ billion fund and Unit - -Investment Trusts complex - -(1) - -First Quarter 2006 -• Launched Van Kampen - -– 6th largest 3rd party fund sponsor -– One of the fastest growing - -(2) - -complexes over the last 4 years - -(2) - -– 2nd largest Unit Investment - -Trusts sponsor - -(3) - -• Strong brand presence within broker - -community - -• Established leader with distinguished -capabilities in several key product -areas: U.S. Large Cap Value / U.S. -Moderate Allocation / High Yield -Municipals / Senior Loans - -• Leader in intermediary platform - -market - -– On 20 of top 25 annuity platforms -– On 22 of top 25 Defined -(5) -Contribution platforms - -– Attractive blend of run rate and - -(4) - -takeover business flows - -International Growth Fund and -Van Kampen Leaders Fund - -• Introduced new retirement - -education and support materials - -• Increase resources for high growth / - -margin businesses: - -– Defined Contribution: - -Doubling field sales force -– Unit Investment Trusts: - -38% increase in field sales force - -• Added Variable Insurance field - -• Expand product offering to add - -sales professionals - -diversity - -– International & Global -– Closed End Funds -– Income Oriented products -– Retail Separately Managed - -Account strategies - -• Position Van Kampen to take - -advantage of retirement markets -from a product and distribution -standpoint - -Second Quarter 2006 -• Launched Van Kampen Equity - -Premium Income - -• Issued 14 new Unit Investment - -Trusts products - -• Introduced new television - -advertising - -1. AUM/AUS as of May 31, 2006 -2. Source: Strategic Insights, May 2006 -3. Source: ICI, May 2006 - -Page - -19 - -4. Source: VARDS, 4th quarter 2005 -5. Source: Pensions and Investments, June 2005 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM GROWTH STRATEGY - -Stem Outflows of Morgan Stanley Retail Mutual Funds -What We Have - -What We’ve Done - -What We’re Doing - -• Morgan Stanley Advisors funds -(excluding Money Market and -Closed End) represents 7% of -total MSIM AUM/AUS (1) - -• Favorable market share in - -Morgan Stanley Global Wealth -Management Group (GWMG) - -• Outflows in Morgan Stanley - -Advisor products - -First Quarter 2006 -• Deployed a new client focused - -field coverage model - -– Assigned dedicated internal -and external wholesalers to -distributing Morgan Stanley -Funds in Morgan Stanley -GWMG - -Second Quarter 2006 -• Hired the head of Morgan -Stanley Funds Distribution - -– A newly created position to - -focus on the Morgan Stanley- -branded funds - -• Preserve favorable market share - -in Morgan Stanley GWMG - -• Actively pursue third party - -distribution of Morgan Stanley -Funds in 401(K), financial -planner, and Registered -Investment Advisor communities - -• Develop distinctive Morgan - -Stanley fund offering - -– Launch distinctive new - -products - -– Rationalize existing product - -line - -1. - -Total MSIM AUM/AUS of $440 billion as of May 31, 2006 - -Page - -20 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Rebuild Institutional Reputation and Business, Particularly in the U.S. -What We Have - -What We’re Doing - -What We’ve Done - -MSIM GROWTH STRATEGY - -• $187 billion in AUM, 43% of - -MSIM total (1) - -• MSIM covers over 100 consulting - -firms - -• Manages assets for 20 of the top -50 U.S. Pension funds, 56 of the -top 200 U.S. Pension funds (2) - -First Quarter 2006 -• Launched / incubated 9 new -products within the channel - -• Conducted road shows with - -consultants and clients - -Second Quarter 2006 - -• Launched / incubated 9 new -products within the channel - -• Internal transfer of Pension Asset -/ Liability management expertise - -• Senior management meetings - -and road shows with consultants -and clients - -• Hosted the annual investment -roundtable of industry thought -leaders (Lyford Cay Conference) - -Client - -• Communicate MSIM story to - -clients and consultants -• Form Institutional Client - -Solutions Group - -• Leverage client service - -People - -• Hire US Institutional leader -• Hire additional salespeople - -Product - -• Develop products to address - -client needs - -• Enhance ability to place - -alternatives - -• Communicate Fixed Income - -turnaround story - -• Retain clients in International - -Value Equity products - -1. As of May 31, 2006 -2. Source: P&I Databook, January 23, 2006. It includes Corporate and Public Funds - -Page - -21 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Summary: Measuring Our Progress - -MSIM GROWTH STRATEGY - -First Year - -Two to Three Years - -Three to Five Years -Three to Five Years - -Business Investment - -Gaining Traction - -Results - -• New management team - -• Filled leadership gaps - -• Reorganized business by - -product and client segment - -• Bottom up business plan - -• Seeded new products - -• Built infrastructure - -• Actively pursued new talent to - -manage money - -• Opened lines of communication - -• Positive flows - -• More complete traditional - -product offering - -• Improved margin - -• Completion of several - -acquisitions - -• Multiple direct hedge fund - -strategies - -• Restructured Morgan Stanley - -Advisor Fund family - -• Leader in the asset - -management business with a -reputation for excellent -performance, innovative -products and superior client -service - -Page - -22 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Measuring Our Performance: Key Success Metrics - -MSIM GROWTH STRATEGY - -• Net revenues - -• Pre-tax margin - -FY 2005 Results (1) - -Aspiration - -$2.9 billion - -Double digit growth - -35% (incl. Private Equity) - -29% (excl. Private Equity) - -30%~35% - -• Assets Under Management - -$431 billion - -$600 billion+ - -• Net Flows - -($19) billion - -Strong and competitive -positive flows - -• FY New products launched - -3 +Penn Is Affordable +Penn's commitment to financial aid opens the +possibility of a Wharton education to students +from all walks of life. We encourage you to apply. -• Retail Product Performance +Penn admits students on the basis +of academic and personal accomplishments, +and practices need-blind admissions for +citizens and permanent residents of the United +States, Canada, and Mexico, and need-aware +admissions for all others. -– % of Morningstar rated assets in 4/5 star +Penn financial aid packages include only grants +and work-study jobs, regardless of family income +level. Students have the option to use loans to +supplement their aid packages. -45% + Online — Watch the Penn Admissions + video, “Apply: Freshmen” -30+ + Hey Day, when juniors are declared +to be seniors—a Penn tradition since 1916 -50%+ +The University of Pennsylvania values diversity and seeks talented students, faculty and staff from diverse backgrounds. +The University of Pennsylvania does not discriminate on the basis of race, color, sex, sexual orientation, gender identity, +religion, creed, national or ethnic origin, citizenship status, age, disability, veteran status or any other legally protected +class status in the administration of its admissions, financial aid, educational or athletic programs, or other University- +administered programs or in its employment practices. -funds +Questions or complaints regarding this policy should be directed to the Executive Director of the Office of Affirmative +Action and Equal Opportunity Programs, Sansom Place East, 3600 Chestnut Street, Suite 228, Philadelphia, PA +19104-6106; or (215) 898-6993 (Voice). -1. Full Year ended November 30, 2005 + Office of Undergraduate Admissions +University of Pennsylvania +1 College Hall +Philadelphia, PA 19104.6376 ++1.215.898.7507 +info@admissions.upenn.edu +www.admissions.upenn.edu -Page +Visit campus -23 +undergrad.wharton.upenn.edu -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +Learn more about us diff --git a/q144/random_k2/question.json b/q144/random_k2/question.json index 44da9d08c6236903debeb6ec17cdc64fb3fb0e96..39b4e08fbd1c23ad94f6cb306bc3d48d4ff5aeff 100644 --- a/q144/random_k2/question.json +++ b/q144/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "germanwingsdigitalcrisisanalysis-150403064828-conversion-gate01_95.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf", - "web_ff8dcb0ea2a037b7.pdf" + "web_5050add9b28048e4.pdf", + "web_5e409f22555df076.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q144/random_k2/random_1.md b/q144/random_k2/random_1.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..b66c1815f8dc909ab75e16a08e2512a86aa07ad4 100644 --- a/q144/random_k2/random_1.md +++ b/q144/random_k2/random_1.md @@ -1,1453 +1,655 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA -8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -i -C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | +1 + +The Nursing Home +Inspection Process + +SUMMARY + +Both the Minnesota Department of Health (MDH) and the U.S. +Department of Health and Human Services share responsibility for +ensuring that Minnesota’s nursing homes provide an acceptable level +of care for their residents. Because the federal government dictates +the overall structure and content of the inspection program, the State +of Minnesota has few opportunities to make significant changes in +how it conducts nursing home inspections. The federal government +mandates how often the state must inspect nursing homes, the steps +the state must follow when conducting inspections, and the standards +the state must apply. Although MDH and other states have asked the +federal government for more flexibility in conducting inspections, the +federal government has not issued any waivers that allow states to +significantly change or implement an alternative inspection program. + +The current nursing home inspection process emerged in the mid-1980s, as + +Congress responded to reports of resident abuse and inadequate enforcement + +of nursing home regulations. In a 1986 report on nursing home quality, the +Institute of Medicine found “serious, even shocking inadequacies” in the +enforcement of regulations.1 As a result of this report and the efforts of advocacy +groups and professional organizations, Congress passed a major reform of nursing +home regulation as part of the Omnibus Budget Reconciliation Act of 1987.2 + +Since that time, Congress and the U.S. Department of Health and Human Services +have periodically modified inspection requirements in response to studies that +have shown continued weak and inconsistent enforcement of nursing home +regulations and quality of care problems. Most significantly, the Nursing Home +Oversight Improvement Program was implemented in 1998, which, among other +things, enhanced federal review of state inspections and required the federal +government to terminate funding for states that fail to conduct adequate +inspections. + +This chapter addresses the following question about how the Minnesota +Department of Health (MDH) inspects nursing homes: + +• What are the respective roles of the Minnesota Department of Health +and U.S. Department of Health and Human Services in conducting +nursing home inspections? + +Institute of Medicine, Committee on Nursing Home Regulation, Improving the Quality of Care + +1 +in Nursing Homes (Washington DC: National Academy Press, 1986), 146. +2 Pub. L. 100-203, Dec. 22, 1987. + + 4 + +NURSING HOME INSPECTIONS + +The federal +government and +states share +responsibility +for ensuring that +nursing homes +provide an +acceptable level +of care to +residents. + +To answer this question, we examined state and federal laws, rules, regulations, +and guidelines related to nursing homes inspections, as well as a wide variety of +research reports by state and federal agencies. We also interviewed state +policymakers, nursing home inspectors and their supervisors, and a sample of +nursing home administrators from throughout the state. + +FEDERAL REGULATION OF INSPECTIONS + +State and federal laws define a nursing home as a facility (or that part of a facility) +that provides health evaluation and treatment services to five or more residents +who do not need an acute care facility (such as a hospital) but who require nursing +supervision or rehabilitation services on an inpatient basis.3 In lay terms, this +means a facility that provides a room, meals, recreational opportunities, and help +with daily living activities such as dressing, eating, bathing, walking, and using +the bathroom. Residents generally have health problems that keep them from +living on their own and may require daily medical attention. + +The federal +government and +states share +responsibility for +ensuring that +nursing homes +provide an +acceptable level of +care to residents. +The Centers for +Medicare and +Medicaid Services +(CMS) in the U. S. +Department of +Health and Human +Services oversees +the inspection +program for +nursing homes that +participate in the +federal Medicare and Medicaid programs.4 The agency sets nursing home +standards; interprets federal regulations, guidelines, and polices; and establishes +and monitors inspection procedures. It contracts with MDH to conduct nursing +home inspections in Minnesota.5 In addition to conducting inspections, MDH +licenses nursing homes for state purposes and certifies their eligibility for +participation in the Medicare and Medicaid programs. Finally, the department is + +Nursing home residents generally need help with many activities of +daily living. + +3 Minn. Stat. (2004), §144A.01, subd. 5-6; and 42 U.S. Code, §1396r, (a) (2000). +4 CMS was formerly called the Health Care Financing Administration. +5 The Minnesota Department of Health contracts with the State Fire Marshall’s Office in the +Minnesota Department of Public Safety to determine facility compliance with the federal Life +Safety Code, which is necessary for participation in the Medicare and Medicaid programs. State +Fire Marshall findings are included in the inspection reports issued by MDH. + + THE NURSING HOME INSPECTION PROCESS + +5 + +responsible for explaining program participation requirements to providers to help +them comply with federal requirements.6 + +Overall, we found that: + +• The federal government sets forth the overall structure and content of +the nursing home inspection program, and Minnesota has very few +opportunities to make significant changes in the program. + +Federal regulations outline both the general parameters of the inspection process +as well as the specifics of how each inspection must be done. They dictate: +(1) how frequently the state must inspect nursing homes, (2) the steps the state +must follow when conducting inspections, and (3) the standards that the state +must apply. We discuss each of these areas in greater detail below. + +Inspection Frequency + +The federal government sets forth how often nursing homes must be inspected: + +Federal law and regulations require that the Minnesota Department of +Health inspect nursing homes every 12 months, on average. + +All nursing facilities must be inspected no later than once every 15 months, with +an average time statewide between inspections of 12 months. Federal regulations +do not allow states to inspect nursing homes with “good” inspection records less +frequently than homes with “bad” records. In addition, CMS requires that at least +10 percent of inspections be “staggered” (started outside of normal business +hours). To meet this requirement, the state must begin some inspections on +weekends or holidays, some in the early morning (before 8:00 AM), and some in +the evening (after 6:00 PM). Furthermore, the federal government requires that +all nursing home inspections be unannounced. + +About 420 Minnesota nursing homes participated in the Medicare and Medicaid +programs during federal fiscal year 2003.7 The department inspected all of these +nursing homes within 14.7 months of their prior inspection, with an average time +between inspections of 12 months.8 In addition, 12 percent of the 403 inspections +conducted were staggered, with 21 inspections beginning before 8:00 AM, 16 +inspections after 6:00 PM, and 10 beginning on a weekend or holiday.9 + +For the most part, nursing home providers, state policymakers, and nursing home +inspectors generally agree that requiring annual inspections of all nursing homes + +6 The department has additional responsibilities related to nursing homes, such as investigating +complaints, which were outside the scope of our evaluation. Nursing home inspectors also inspect +other types of health care facilities, such as hospitals and intermediate care facilities for the mentally +retarded. These activities were likewise outside the scope of our evaluation. +7 The federal fiscal year runs from October 1 through September 30. +8 Centers for Medicare and Medicaid Services, Federal Fiscal Year 2003 State Performance +Standard Review Report (Washington, DC, March 15, 2004), 1. +9 +Ibid., 2. Because nursing homes may go up to 15 months between inspections, the number of +nursing homes that MDH inspected during federal fiscal year 2003 was less than the total number of +nursing homes in the state. + +Nursing homes +must have a +“surprise” +inspection no +later than once +every 15 months. + +(cid:127) + 6 + +NURSING HOME INSPECTIONS + +Because of +federal +requirements, +MDH cannot +inspect nursing +homes with +“good” +inspection +records less +frequently than +those with "bad" +records. + +is, at times, an inefficient use of staff resources. The requirement does not permit +the state to focus efforts on the nursing homes that need oversight the most. To +help increase the efficiency and effectiveness of the inspection process, the +Legislature has repeatedly required the Commissioner of Health to seek federal +permission to implement an alternative inspection process that would change how +often nursing homes must be inspected.10 In response, the department submitted a +proposal to CMS that would have increased the time between “full” inspections +up to 30 months for some homes with “good” compliance records. Other states +have proposed similar approaches, including ones to conduct abbreviated annual +inspections for homes with “good” compliance records. + +To date, CMS has not approved an alternative inspection program put forth by any +state, including Minnesota. According to CMS, the social security law does not +allow states to obtain waivers to implement an alternative inspection program for +nursing homes participating in the Medicare program, although states could +implement an alternative inspection program for homes that only participate in the +Medicaid program. However, because this would involve only a few nursing +homes, it is generally not feasible for states to do so. + +For the last several years, CMS has been studying the feasibility of an alternative +inspection process. Recently, the agency announced that it would be establishing +a few pilot sites around the country to implement a “revamped” inspection +process. Designed to address concerns about inspection consistency and +efficiency, pilot sites will make greater use of computers to make initial +determinations of deficiencies rather than relying on the judgment of inspection +teams. The alternative process will not result in less frequent inspections for +facilities, but may allow inspectors to spend somewhat less time in “good” +facilities and more time in “bad” ones. + +Inspection Steps + +In addition to requiring an inspection no later than once every 15 months: + +• + +Federal regulations require that each nursing home’s annual +inspection be a “standard” or full inspection consisting of seven +federally mandated steps. + +Federal regulations do not allow states to do shorter or abbreviated inspections of +nursing homes with “good” records of compliance or to cut short an inspection +when inspectors do not detect any problems in a facility. On the other hand, state +inspectors must extend the inspection if they suspect that a facility is providing +substandard care to its residents. + +As shown in Table 1.1, the standard inspection consists of seven federally +mandated steps. First, inspectors prepare off-site by reviewing information about +the nursing home and its residents to help identify areas of concern. Immediately +upon arriving at the facility, the inspection team meets with the nursing home +administrator to explain the inspection process and request specific information; + +10 Laws of Minnesota (2000), ch. 312, sec. 2, 5; Laws of Minnesota (1Sp2001), ch. 9, art. 5, sec. 38; +Laws of Minnesota (2002), ch. 379, art. 1, sec. 113; and Laws of Minnesota (2004), ch. 247, +sec. 6. + + THE NURSING HOME INSPECTION PROCESS + +7 + +Table 1.1: The Federal Nursing Home Inspection +Process + +Step 1: Off-site preparation + +Step 2: Entry conference and on-site preparation + +Step 3: + +Initial nursing home tour + +Step 4: Resident sample selection + +Step 5: + +Information gathering + +A. General observation of the facility +B. Kitchen/food service observation +C. Resident review +D. Quality of life assessment +E. Medication pass +F. Quality assessment and assurance review +G. Abuse prevention review + +Step 6: Deficiency determination + +A. Determination of substandard quality of care + +Step 7: Exit conference + +Likewise, +MDH cannot +do abbreviated +inspections in +nursing homes +with "good" +records. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), ch. 7, sec 7200. + +this is followed by a facility tour. Using the information provided by the facility +and what inspectors learned during the tour, the team then selects a sample of +residents to focus on during the information-gathering portion of the inspection. +During this phase, the team meets on a daily basis to compare notes, discuss new +areas of concern, and make adjustments to the inspection as deemed necessary. +Inspectors observe the care and services that facility staff provide to residents, +such as preparing and serving meals, administering medications, and helping, as +necessary, with activities such as bathing, toileting, walking, and grooming. +Inspection team members also interview residents and staff and review resident +records. Once the team is satisfied that they have gathered enough information, it +meets to determine whether the facility has failed to meet any regulatory +requirements. The team prepares a draft inspection report that discusses each +violation of federal regulations (commonly referred to as a deficiency) that the +team has identified, and then meets with nursing home personnel and interested +residents and family members to present its preliminary list of deficiencies. + +After the inspection team leaves the facility, it finalizes the “Statement of +Deficiencies” and submits it to the team’s district supervisor who is responsible +for reviewing the document and submitting a final copy to the facility and CMS. +The facility must submit a “Plan of Correction” within ten days that indicates how +and when it will correct each of the deficiencies that it has received.11 Inspectors +normally conduct an unannounced revisit to verify that the plan of correction has +been implemented and that the deficiencies no longer exist. For the most part, +MDH generally gives a facility 40 days from the end of the inspection to correct +deficiencies before MDH imposes any sanctions on the facility. + +11 Facilities may also dispute a deficiency and request a hearing before MDH or an administrative +law judge within this ten-day period. Chapter 2 discusses how often this happens and the outcome +of such hearings. + +Inspectors spend +much of their +time observing +and talking with +residents and +staff. + + 8 + +NURSING HOME INSPECTIONS + +State law +requires +inspectors to +leave a draft +inspection report +with facilities +when they leave. + +While the state is unable to make significant changes in how inspections are done: + +• Minnesota has expanded the federal nursing home inspection process + +in several ways. + +The state goes beyond federal inspection requirements by adding other tasks, +including requirements to: (1) interview family council members; (2) expand the +number of evening observations nursing home inspectors must make each month; +(3) conduct a “verify and clarify” session with the provider to discuss possible +areas of concern prior to the exit conference; and (4) leave a draft inspection +report with nursing homes after the inspection, with the final report due within +15 days. Some of these activities were added to make the inspection process more +“user friendly” for providers. Others, such as expanding the inspection to include +final interviews with family council members, were at the urging of advocacy +groups. + +For the year ending +September 30, 2004, +MDH inspectors, +working in teams of +three to five +registered nurses, +spent an average of +about 150 hours per +facility to complete +the state and +federally mandated +inspection tasks.12 +As would be +expected, it took +longer to inspect +larger nursing homes +than smaller ones. +For example, a +facility with 40 or +fewer beds averaged about 72 hours per inspection while a facility with 116 to +160 beds averaged 176 hours.13 + +Nursing home inspectors must meet with each facility's resident +council. + +Inspection Standards + +The federal State Operations Manual (SOM) sets forth the federal standards that +inspectors must apply during an inspection as well as guidelines to help them +apply those standards.14 As currently written: + +12 Minnesota Department of Health analysis of data from the Online Survey and Certification +Reporting System, December 2, 2004. State inspectors spent an additional 54 hours per facility, on +average, conducting follow-up inspections to ensure that facilities corrected deficiencies. +13 Minnesota Department of Health, Federal Fiscal Year 2005 Initial Budget Request (St. Paul, +July 15, 2004), unnumbered. +14 Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004). + + THE NURSING HOME INSPECTION PROCESS + +9 + +Inspectors grade +the seriousness of +each deficiency +by assigning it a +letter code. + +• The federal standards and guidelines that state inspection teams must + +use to inspect nursing homes are prescriptive and complex. + +The SOM covers hundreds of pages and contains 274 regulatory standards that +nursing homes must meet at all times. The standards cover 16 different categories +of operation, including administration, dietary services, infection control, life +safety, physical environment, quality of care, quality of life, resident assessment, +and resident rights. Some requirements must be met for each resident and any +violation of these requirements, even for one resident, is a deficiency. For +example, each resident must have a comprehensive care plan. Other requirements +focus on facility systems and are evaluated comprehensively rather than in terms +of a single incident. For example, a facility must have a medication error rate +below 5 percent.15 + +For each deficiency, inspectors must use professional judgment to assess how +many residents or staff are affected by or involved in the deficient practice (scope) +and the amount of actual or potential discomfort or harm involved for residents +(severity). As shown in Table 1.2, these two determinations result in the +inspection team assigning a letter code (A through L) to each deficiency, with +level “A” deficiencies being the least serious. + +Table 1.2: Deficiency Scope and Severity Grid + +Severity + +Scope +Isolated Pattern Widespread + +Level 4: A situation that has caused or is likely to cause +serious resident injury, harm, impairment, or death. + +Level 3: A situation that has caused resident harm. + +Level 2: A situation that has caused minimal discomfort to +a resident OR has the potential to cause resident harm. + +Level 1: A situation that has the potential of causing no + +more than minimal discomfort to a resident. + +J + +G + +D + +A + +K + +H + +E + B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) -M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +L + +I + +F + +C + +NOTE: Harm is defined as a situation that compromises a resident’s ability to maintain or reach his or +her highest practicable physical, mental, or psychosocial well being, as defined by an accurate and +comprehensive assessment, care plan, and provision of services. A nursing home with one or more +quality of life, quality of care, or resident behavior and facility practices deficiencies issued at level “F” +or “H” or above (the shaded area of the grid) is considered to be providing “substandard” care to its +residents. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), Appendix P, V, B-C. + +To determine a deficiency’s scope, inspectors must classify each deficiency in one +of three ways: isolated, pattern, or widespread. Federal guidelines say that a +deficiency is isolated when one or a very limited number of residents or staff are +affected or the situation has occurred only occasionally or in a very limited +number of locations in the facility. For example, if 60 of 70 residents in a facility +are incontinent and the facility failed to provide adequate care or services to +restore or improve bladder function for 2 of these residents, the deficiency should +be classified as isolated. A deficiency represents a pattern when it affects more + +15 However, a single medication error that is considered severe enough may result in a deficiency. + + 10 + +NURSING HOME INSPECTIONS + +than a very limited number of residents or staff, occurs in several locations, or the +same resident has been affected by repeated occurrences of the same deficient +practice. If the above facility did not provide adequate care or services to 10 of its +60 incontinent residents, the resulting deficiency should be issued as a pattern. A +deficiency is identified as widespread when it refers to the entire facility or when +a system failure has affected or has the potential to affect a large number of +residents. For example, a facility failing to provide adequate care or services to +improve or restore bladder function to 30 of its 60 incontinent residents should be +issued a deficiency classified as widespread. + +Inspectors must also determine the severity of a deficiency on a scale from one to +four. Level one refers to deficiencies that have the potential for causing no more +than a minor negative impact on, or minimal physical, mental, or psychosocial +discomfort to, a resident. For example, a facility should receive a level one +deficiency if it failed to post its inspection results or only made them available +upon request. Level two deficiencies are those that have resulted in resident +discomfort or have the potential to harm residents. Federal regulations define +harmful situations as those that compromise residents’ ability to maintain or reach +their highest practicable physical, mental, and psychosocial well being, excluding +situations that are of a “limited consequence” to residents. For example, a nursing +home should receive a level two deficiency if inspectors observed staff failing to +wash their hands properly between caring for residents but no one became +seriously ill as a result. Level three deficiencies are those that have actually +resulted in resident harm. The hand-washing example should be a level three +deficiency if there was evidence that a resident caught a contagious disease as a +result of staff failing to wash their hands properly after providing resident care. +Level four represents immediate jeopardy situations whereby the facility must +undertake immediate corrective action to address problems that have resulted in or +are likely to cause serious injury, harm, impairment, or death to a resident. For +example, if a resident with dementia was found outside during an inspection +heading toward a busy highway and the nursing home did not have a working +system in place to monitor residents with dementia, the facility should be issued a +level four deficiency. + +The “seriousness” of a facility’s deficiencies (their scope and severity) helps +determine the sanctions for nursing homes that fail to correct deficiencies within +an allowable time frame. As shown in Table 1.3, there are three categories of +required sanctions. Generally, nursing homes do not face sanctions for +deficiencies issued at levels “A” through “C.”16 Category 1 sanctions are reserved +for deficiencies issued at levels “D” and “E” and require that facilities implement +a plan of correction developed by the state, have their staff attend a specific +training program, or be subject to state monitoring. Conversely, category 3 +sanctions are reserved for the most serious deficiencies and include the state +assuming management of the facility, terminating the facility’s participation in the +Medicare and Medicaid programs, or closing the facility. Except in instances of +immediate jeopardy to residents (a deficiency issued at level “J” or above) or +when facilities receive level “G” or higher deficiencies in two consecutive +inspections, facilities are generally given an opportunity to correct deficiencies +before any sanctions are imposed—usually 40 days. MDH must deny Medicare + +16 Although the federal government does not require that sanctions be imposed on facilities +for low-level deficiencies (levels “B” and “C”), the state may choose to impose sanctions from +category 1 when facilities fail to correct their deficiencies. + +A deficiency's +letter code helps +determine what +sanctions MDH +could impose on +the facility. + + THE NURSING HOME INSPECTION PROCESS + +11 + +Table 1.3: Required Sanctions for Noncompliance + +Category 1: Deficiencies issued at levels “D” and “E” + +Directed plan of correction; +State monitoring; and/or +Directed in-service training. + +Category 2: Deficiencies issued at levels “F” through “I” + +Denial of payment for new Medicare and Medicaid admissionsa; +Denial of payment for all Medicare and Medicaid residents; +Civil money penalties of $50-$3,000 per day of noncompliance; and/or +Civil money penalties of $1,000-$10,000 per incident of noncompliance. + +Category 3: Deficiencies issued at levels “J” and above + +Temporary management; +Termination from the Medicare/Medicaid programs; and/orb +Facility closure. + +NOTE: The Minnesota Department of Health may impose a category 2 sanction to supplement a +category 1 sanction for deficiencies issued at levels “D” and “E.” In general, a category 1 or 2 sanction +can also be imposed whenever a category 3 sanction is required, and a category 1 sanction may also +be imposed when a category 2 sanction is required. Civil penalties increase to $3,050-$10,000 per +day when they are imposed in addition to a category 3 sanction. The state may also assume +temporary management (a category 3 sanction) when a facility has been issued a level “I” deficiency. +Also, a facility cited for providing substandard care cannot operate a nurse aide training and +competency evaluation program for two years. + +aThe state must deny Medicare and Medicaid payments for new admissions when a facility is not in +substantial compliance within three months of the inspection and when a facility has been cited for +substandard care on three consecutive annual inspections. In the latter situation, state monitoring +must also be imposed. + +bThe state must recommend termination from the Medicare and Medicaid programs when a facility is +not in substantial compliance within six months of the inspection. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), ch. 7, sec. 7210G and 7400. + +and Medicaid reimbursements for new admissions when facilities have not +corrected their deficiencies within three months of the department’s inspection. +Facilities must be terminated from the program if deficiencies are not corrected +within 6 months. + +FUNDING + +In keeping with the high degree of federal involvement in the nursing home +inspection program: + +• + +State funds cover less than 10 percent of the total cost of nursing home +inspections and complaint investigations. + +The federal government is the major source of funding for the inspection +program, with the state contributing less than 10 percent of the total cost for +nursing homes. In fiscal year 2004, MDH spent about $12 million from state and + + 12 + +NURSING HOME INSPECTIONS + +federal sources on activities related to nursing home inspections, including costs +related to investigating complaints against nursing homes.17 The state’s share +(about $1.1 million) is the result of state negotiations with CMS and has +historically been low when compared with that of other states. According to a +2000 analysis of costs by the Health Care Financing Administration, Minnesota +was the only state in the Chicago region that paid less than 10 percent of total +inspection costs.18 Other states paid at least 16 percent, with one state paying +almost 25 percent of total costs. + +17 Cecelia Jackson, “Re: FFY 2004 Nursing Home Expenditures” (December 23, 2004), electronic +mail to jo.vos@state.mn.us. +18 Health Care Financing Administration, “Nursing Home Survey, State Licensure Cost Shares” +(Chicago, May 2000). Minnesota is part of the Chicago region, which also includes Illinois, +Indiana, Michigan, Ohio, and Wisconsin. + diff --git a/q144/random_k2/random_2.md b/q144/random_k2/random_2.md index 69d1305f3a9c9c9fb312a4e7dbd3718670cacdd2..886a7a04c5cccafe307d041d698e8684ec95d131 100644 --- a/q144/random_k2/random_2.md +++ b/q144/random_k2/random_2.md @@ -1,2203 +1,490 @@ -3 -2 -0 -2 +Genetics, DNA, and Heredity -r -a - -M -7 - -] -L -C -. -s -c -[ - -4 -v -1 -7 -1 -1 -1 -. -3 -0 -2 -2 -: -v -i -X -r -a - -Published as a conference paper at ICLR 2023 - -SELF-CONSISTENCY IMPROVES CHAIN OF THOUGHT -REASONING IN LANGUAGE MODELS - -Jason Wei† Dale Schuurmans† Quoc Le† Ed H. Chi† - -Xuezhi Wang†‡ -Sharan Narang† Aakanksha Chowdhery† Denny Zhou†§ -†Google Research, Brain Team -‡xuezhiw@google.com, §dennyzhou@google.com - -ABSTRACT - -Chain-of-thought prompting combined with pre-trained large language models has -achieved encouraging results on complex reasoning tasks. In this paper, we propose -a new decoding strategy, self-consistency, to replace the naive greedy decoding -used in chain-of-thought prompting. It first samples a diverse set of reasoning paths -instead of only taking the greedy one, and then selects the most consistent answer -by marginalizing out the sampled reasoning paths. Self-consistency leverages the -intuition that a complex reasoning problem typically admits multiple different ways -of thinking leading to its unique correct answer. Our extensive empirical evaluation -shows that self-consistency boosts the performance of chain-of-thought prompting -with a striking margin on a range of popular arithmetic and commonsense reasoning -benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), -StrategyQA (+6.4%) and ARC-challenge (+3.9%). - -1 - -INTRODUCTION - -Although language models have demonstrated remarkable success across a range of NLP tasks, their -ability to demonstrate reasoning is often seen as a limitation, which cannot be overcome solely by -increasing model scale (Rae et al., 2021; BIG-bench collaboration, 2021, inter alia). In an effort -to address this shortcoming, Wei et al. (2022) have proposed chain-of-thought prompting, where -a language model is prompted to generate a series of short sentences that mimic the reasoning -process a person might employ in solving a task. For example, given the question “If there are 3 -cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?”, instead -of directly responding with “5”, a language model would be prompted to respond with the entire -chain-of-thought: “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars. The answer is 5.”. It has been observed that chain-of-thought prompting significantly -improves model performance across a variety of multi-step reasoning tasks (Wei et al., 2022). - -In this paper, we introduce a novel decoding strategy called self-consistency to replace the greedy -decoding strategy used in chain-of-thought prompting (Wei et al., 2022), that further improves -language models’ reasoning performance by a significant margin. Self-consistency leverages the -intuition that complex reasoning tasks typically admit multiple reasoning paths that reach a correct -answer (Stanovich & West, 2000). The more that deliberate thinking and analysis is required for a -problem (Evans, 2010), the greater the diversity of reasoning paths that can recover the answer. - -Figure 1 illustrates the self-consistency method with an example. We first prompt the language model -with chain-of-thought prompting, then instead of greedily decoding the optimal reasoning path, we -propose a “sample-and-marginalize” decoding procedure: we first sample from the language model’s -decoder to generate a diverse set of reasoning paths; each reasoning path might lead to a different -final answer, so we determine the optimal answer by marginalizing out the sampled reasoning paths -to find the most consistent answer in the final answer set. Such an approach is analogous to the -human experience that if multiple different ways of thinking lead to the same answer, one has greater -confidence that the final answer is correct. Compared to other decoding methods, self-consistency -avoids the repetitiveness and local-optimality that plague greedy decoding, while mitigating the -stochasticity of a single sampled generation. - -1 - - Published as a conference paper at ICLR 2023 - -Figure 1: The self-consistency method contains three steps: (1) prompt a language model using -chain-of-thought (CoT) prompting; (2) replace the “greedy decode” in CoT prompting by sampling -from the language model’s decoder to generate a diverse set of reasoning paths; and (3) marginalize -out the reasoning paths and aggregate by choosing the most consistent answer in the final answer set. - -Self-consistency is far simpler than prior approaches that either train an additional verifier (Cobbe -et al., 2021) or train a re-ranker given additional human annotations to improve generation quality -(Thoppilan et al., 2022). Instead, self-consistency is entirely unsupervised, works off-the-shelf with -pre-trained language models, requires no additional human annotation, and avoids any additional -training, auxiliary models or fine-tuning. Self-consistency also differs from a typical ensemble -approach where multiple models are trained and the outputs from each model are aggregated, it acts -more like a “self-ensemble” that works on top of a single language model. - -We evaluate self-consistency on a wide range of arithmetic and commonsense reasoning tasks over -four language models with varying scales: the public UL2-20B (Tay et al., 2022) and GPT-3-175B -(Brown et al., 2020), and two densely-activated decoder-only language models: LaMDA-137B -(Thoppilan et al., 2022) and PaLM-540B (Chowdhery et al., 2022). On all four language models, -self-consistency improves over chain-of-thought prompting by a striking margin across all tasks. In -particular, when used with PaLM-540B or GPT-3, self-consistency achieves new state-of-the-art levels -of performance across arithmetic reasoning tasks, including GSM8K (Cobbe et al., 2021) (+17.9% -absolute accuracy gains), SVAMP (Patel et al., 2021) (+11.0%), AQuA (Ling et al., 2017) (+12.2%), -and across commonsense reasoning tasks such as StrategyQA (Geva et al., 2021) (+6.4%) and ARC- -challenge (Clark et al., 2018) (+3.9%). In additional experiments, we show self-consistency can -robustly boost performance on NLP tasks where adding a chain-of-thought might hurt performance -compared to standard prompting (Ye & Durrett, 2022). We also show self-consistency significantly -outperforms sample-and-rank, beam search, ensemble-based approaches, and is robust to sampling -strategies and imperfect prompts. - -2 SELF-CONSISTENCY OVER DIVERSE REASONING PATHS - -A salient aspect of humanity is that people think differently. It is natural to suppose that in tasks -requiring deliberate thinking, there are likely several ways to attack the problem. We propose that -such a process can be simulated in language models via sampling from the language model’s decoder. -For instance, as shown in Figure 1, a model can generate several plausible responses to a math -question that all arrive at the same correct answer (Outputs 1 and 3). Since language models are not -perfect reasoners, the model might also produce an incorrect reasoning path or make a mistake in -one of the reasoning steps (e.g., in Output 2), but such solutions are less likely to arrive at the same -answer. That is, we hypothesize that correct reasoning processes, even if they are diverse, tend to -have greater agreement in their final answer than incorrect processes. - -We leverage this intuition by proposing the following self-consistency method. First, a language -model is prompted with a set of manually written chain-of-thought exemplars (Wei et al., 2022). Next, - -2 - -Language modelQ: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.…Q: Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder for $2 per egg. How much does she make every day?A:She has 16 - 3 - 4 = 9 eggs left. So she makes $2 * 9 = $18 per day. Sample a diverse set of reasoning pathsShe eats 3 for breakfast, so she has 16 - 3 = 13 left. Then she bakes muffins, so she has 13 - 4 = 9 eggs left. So she has 9 eggs * $2 = $18. This means she she sells the remainder for $2 * (16 - 4 - 3) = $26 per day. The answer is $18.The answer is $26.The answer is $18.The answer is $18.Marginalize out reasoning paths to aggregate final answersLanguage modelThis means she uses 3 + 4 = 7 eggs every day. She sells the remainder for $2 per egg, so in total she sells 7 * $2 = $14 per day. The answer is $14.The answer is $14.Greedy decode Published as a conference paper at ICLR 2023 - -GSM8K MultiArith - -AQuA - -SVAMP - -CSQA - -ARC-c - -Greedy decode - -56.5 - -94.7 - -35.8 - -79.0 - -79.0 - -85.2 - -Weighted avg (unnormalized) -Weighted avg (normalized) - -56.3 ± 0.0 -22.1 ± 0.0 - -90.5 ± 0.0 -59.7 ± 0.0 - -35.8 ± 0.0 73.0 ± 0.0 74.8 ± 0.0 82.3 ± 0.0 -15.7 ± 0.0 40.5 ± 0.0 52.1 ± 0.0 51.7 ± 0.0 - -Weighted sum (unnormalized) -Weighted sum (normalized) - -59.9 ± 0.0 -74.1 ± 0.0 - -92.2 ± 0.0 -99.3 ± 0.0 - -38.2 ± 0.0 76.2 ± 0.0 76.2 ± 0.0 83.5 ± 0.0 -48.0 ± 0.0 86.8 ± 0.0 80.7 ± 0.0 88.7 ± 0.0 - -Unweighted sum (majority vote) 74.4 ± 0.1 - -99.3 ± 0.0 - -48.3 ± 0.5 86.6 ± 0.1 80.7 ± 0.1 88.7 ± 0.1 - -Table 1: Accuracy comparison of different answer aggregation strategies on PaLM-540B. - -we sample a set of candidate outputs from the language model’s decoder, generating a diverse set of -candidate reasoning paths. Self-consistency is compatible with most existing sampling algorithms, -including temperature sampling (Ackley et al., 1985; Ficler & Goldberg, 2017), top-k sampling (Fan -et al., 2018; Holtzman et al., 2018; Radford et al., 2019), and nucleus sampling (Holtzman et al., -2020). Finally, we aggregate the answers by marginalizing out the sampled reasoning paths and -choosing the answer that is the most consistent among the generated answers. -In more detail, assume the generated answers ai are from a fixed answer set, ai ∈ A, where -i = 1, . . . , m indexes the m candidate outputs sampled from the decoder. Given a prompt and a -question, self-consistency introduces an additional latent variable ri, which is a sequence of tokens -representing the reasoning path in the i-th output, then couples the generation of (ri, ai) where -ri → ai, i.e., generating a reasoning path ri is optional and only used to reach the final answer ai. As -an example, consider Output 3 from Figure 1: the first few sentences “She eats 3 for breakfast ... So -she has 9 eggs * $2 = $18.” constitutes ri, while the answer 18 from the last sentence, “The answer -is $18”, is parsed as ai.1 After sampling multiple (ri, ai) from the model’s decoder, self-consistency -1(ai = a), -applies a marginalization over ri by taking a majority vote over ai, i.e., arg maxa -or as we defined as the most “consistent” answer among the final answer set. - -(cid:80)m - -i=1 - -In Table 1, we show the test accuracy over a set of reasoning tasks by using different answer -aggregation strategies. In addition to majority vote, one can also weight each (ri, ai) by P (ri, ai | -prompt, question) when aggregating the answers. Note to compute P (ri, ai | prompt, question), we -can either take the unnormalized probability of the model generating (ri, ai) given (prompt, question), -or we can normalize the conditional probability by the output length (Brown et al., 2020), i.e., - -P (ri, ai | prompt, question) = exp - -1 -K - -(cid:80)K - -k=1 log P (tk|prompt,question,t1,...,tk−1), - -(1) - -where log P (tk | prompt, question, t1, . . . , tk−1) is the log probability of generating the k-th token -tk in (ri, ai) conditioned on the previous tokens, and K is the total number of tokens in (ri, ai). -In Table 1, we show that taking the “unweighted sum”, i.e., taking a majority vote directly over ai -yields a very similar accuracy as aggregating using the “normalized weighted sum”. We took a closer -look at the model’s output probabilities and found this is because for each (ri, ai), the normalized -conditional probabilities P (ri, ai | prompt, question) are quite close to each other, i.e., the language -model regards those generations as “similarly likely”.2 Additionally, when aggregating the answers, -the results in Table 1 show that the “normalized” weighted sum (i.e., Equation 1) yields a much -higher accuracy compared to its unnormalized counterpart. For completeness, in Table 1 we also -report the results by taking a “weighted average”, i.e., each a gets a score of its weighted sum divided -by (cid:80)m -i=1 - -1(ai = a), which results in a much worse performance. - -Self-consistency explores an interesting space between open-ended text generation and optimal -text generation with a fixed answer. Reasoning tasks typically have fixed answers, which is why -researchers have generally considered greedy decoding approaches (Radford et al., 2019; Wei et al., -2022; Chowdhery et al., 2022). However, we have found that even when the desired answer is fixed, -introducing diversity in the reasoning processes can be highly beneficial; therefore we leverage - -1The parser is task dependent. For arithmetic reasoning, we parse the first numerical part as the final answer -after the model generates “The answer is ”. For commonsense reasoning, we parse the full string answer as the -final answer after the model generates “The answer is ”. Most generated outputs have a consistent format of -“{Reasoning paths}. The answer is X.” if we prompt the language model in this format. - -2This also means that the language model is not well calibrated and thus cannot distinguish well between -correct solutions and wrong solutions, which also explains why additional re-rankers were trained to better judge -the quality of the solutions in previous work (Cobbe et al., 2021; Thoppilan et al., 2022). - -3 - - Published as a conference paper at ICLR 2023 - -sampling, as commonly used for open-ended text generation (Radford et al., 2019; Brown et al., 2020; -Thoppilan et al., 2022), to achieve this goal. One should note that self-consistency can be applied -only to problems where the final answer is from a fixed answer set, but in principle this approach can -be extended to open-text generation problems if a good metric of consistency can be defined between -multiple generations, e.g., whether two answers agree or contradict each other. - -3 EXPERIMENTS - -We conducted a series of experiments to compare the proposed self-consistency method with existing -approaches on a range of reasoning benchmarks. We find that self-consistency robustly improves -reasoning accuracy for every language model considered, spanning a wide range of model scales. - -3.1 EXPERIMENT SETUP - -Tasks and datasets. We evaluate self-consistency on the following reasoning benchmarks.3 - -• Arithmetic reasoning. For these tasks, we used the Math Word Problem Repository (Koncel- -Kedziorski et al., 2016), including AddSub (Hosseini et al., 2014), MultiArith (Roy & Roth, -2015), and ASDiv (Miao et al., 2020). We also included AQUA-RAT (Ling et al., 2017), a -recently published benchmark of grade-school-math problems (GSM8K; Cobbe et al., 2021), -and a challenge dataset over math word problems (SVAMP; Patel et al., 2021). - -• Commonsense reasoning. For these tasks, we used CommonsenseQA (Talmor et al., 2019), -StrategyQA (Geva et al., 2021), and the AI2 Reasoning Challenge (ARC) (Clark et al., 2018). -• Symbolic Reasoning. We evaluate two symbolic reasoning tasks: last letter concatenation (e.g., -the input is “Elon Musk” and the output should be “nk”), and Coinflip (e.g., a coin is heads-up, -after a few flips is the coin still heads-up?) from Wei et al. (2022). - -Language models and prompts. We evaluate self-consistency over four transformer-based lan- -guage models with varying scales: - -• UL2 (Tay et al., 2022) is an encoder-decoder model trained on a mixture of denoisers with 20- -billion parameters. UL2 is completely open-sourced4 and has similar or better performance than -GPT-3 on zero-shot SuperGLUE, with only 20B parameters and thus is more compute-friendly; -• GPT-3 (Brown et al., 2020) with 175-billion parameters. We use two public engines code-davinci- -001 and code-davinci-002 from the Codex series (Chen et al., 2021) to aid reproducibility;5 -• LaMDA-137B (Thoppilan et al., 2022) is a dense left-to-right, decoder-only language model with -137-billion parameters, pre-trained on a mixture of web documents, dialog data and Wikipedia; -• PaLM-540B (Chowdhery et al., 2022) is a dense left-to-right, decoder-only language model with -540-billion parameters, pre-trained on a high quality corpus of 780 billion tokens with filtered -webpages, books, Wikipedia, news articles, source code, and social media conversations. -We perform all experiments in the few-shot setting, without training or fine-tuning the language -models. For a fair comparison we use the same prompts as in Wei et al. (2022): for all arithmetic -reasoning tasks we use the same set of 8 manually written exemplars; for each commonsense -reasoning task, 4-7 exemplars are randomly chosen from the training set with manually composed -chain-of-thought prompts.6 Full details on the prompts used are given in Appendix A.3. - -Sampling scheme. To sample diverse reasoning paths, we followed similar settings to those -suggested in Radford et al. (2019); Holtzman et al. (2020) for open-text generation. In particular, for -UL2-20B and LaMDA-137B we applied temperature sampling with T = 0.5 and truncated at the -top-k (k = 40) tokens with the highest probability, for PaLM-540B we applied T = 0.7, k = 40, and -for GPT-3 we use T = 0.7 without top-k truncation. We provide an ablation study in Section 3.5 to -show that self-consistency is generally robust to sampling strategies and parameters. - -3By default we use the test split for all datasets if the labels are available for evaluation. For CommonsenseQA -we use the dev split; for StrategyQA we use the question-only set from BIG-bench collaboration (2021): -https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/strategyqa. - -4Model checkpoints at https://github.com/google-research/google-research/tree/master/ul2. -5Public API available at https://openai.com/api/. -6Self-consistency is robust to different sets of prompts and we provide a study in Appendix A.1.2. - -4 - - Published as a conference paper at ICLR 2023 - -3.2 MAIN RESULTS - -We report the results of self-consistency averaged over 10 runs, where we sampled 40 outputs -independently from the decoder in each run. The baseline we compare to is chain-of-thought -prompting with greedy decoding (Wei et al., 2022), referred to as CoT-prompting, which has been -previously used for decoding in large language models (Chowdhery et al., 2022). - -Arithmetic Reasoning The results are shown in Table 2.7 Self-consistency improves the arithmetic -reasoning performance over all four language models significantly over chain-of-thought prompting. -More surprisingly, the gains become more significant when the language model’s scale increases, -e.g., we see +3%-6% absolute accuracy improvement over UL2-20B but +9%-23% for LaMDA- -137B and GPT-3. For larger models that already achieve high accuracy on most tasks (e.g., GPT-3 -and PaLM-540B), self-consistency still contributes significant additional gains with +12%-18% -absolute accuracy on tasks like AQuA and GSM8K, and +7%-11% on SVAMP and ASDiv. With -self-consistency, we achieve new state-of-the-art results on almost all tasks: despite the fact that self- -consistency is unsupervised and task-agnostic, these results compare favorably to existing approaches -that require task-specific training, or fine-tuning with thousands of examples (e.g., on GSM8K). - -Method - -AddSub MultiArith ASDiv - -AQuA - -SVAMP - -GSM8K - -Previous SoTA 94.9a - -60.5a - -75.3b - -37.9c - -57.4d - -35e / 55g - -UL2-20B - -CoT-prompting -Self-consistency 24.8 (+6.6) - -18.2 - -10.7 -15.0 (+4.3) - -16.9 -21.5 (+4.6) 26.9 (+3.3) - -23.6 - -12.6 -19.4 (+6.8) - -4.1 -7.3 (+3.2) - -LaMDA-137B - -PaLM-540B - -GPT-3 -Code-davinci-001 - -CoT-prompting -51.8 -Self-consistency 63.5 (+10.6) 75.7 (+23.9) 58.2 (+9.2) 26.8 (+9.1) - -49.0 - -52.9 - -17.7 - -38.9 -53.3 (+14.4) 27.7 (+10.6) - -17.1 - -CoT-prompting -Self-consistency 93.7 (+1.8) - -91.9 - -94.7 -99.3 (+4.6) - -74.0 -81.9 (+7.9) 48.3 (+12.5) 86.6 (+7.6) - -79.0 - -35.8 - -56.5 -74.4 (+17.9) - -CoT-prompting -59.5 -Self-consistency 67.8 (+10.6) 82.7 (+23.2) 61.9 (+9.2) 25.6 (+6.7) - -52.7 - -57.2 - -18.9 - -39.8 -54.5 (+14.7) 23.4 (+8.8) - -14.6 - -GPT-3 -Code-davinci-002 - -CoT-prompting -Self-consistency 91.6 (+2.2) - -89.4 - -96.2 -100.0 (+3.8) 87.8 (+7.6) 52.0 (+12.2) 86.8 (+11.0) 78.0 (+17.9) - -80.1 - -75.8 - -60.1 - -39.8 +The Basics -Table 2: Arithmetic reasoning accuracy by self-consistency compared to chain-of-thought prompting -(Wei et al., 2022). The previous SoTA baselines are obtained from: a: Relevance and LCA operation -classifier (Roy & Roth, 2015), b: Lan et al. (2021), c: Amini et al. (2019), d: Pi et al. (2022), e: -GPT-3 175B finetuned with 7.5k examples (Cobbe et al., 2021), g: GPT-3 175B finetuned plus an -additional 175B verifier (Cobbe et al., 2021). The best performance for each task is shown in bold. + What is DNA? -Method +It's a history book - -CSQA +a narrative of the journey of our -StrategyQA ARC-e +species through time. -ARC-c +It's a shop manual, with an incredibly detailed blueprint -Letter (4) Coinflip (4) +for building every human cell. -Previous SoTA 91.2a +And it's a transformative textbook of medicine, with -73.9b +insights that will give health care providers immense +new powers to treat, prevent and cure disease." +- Francis Collins -86.4c + What Does DNA Look Like? -75.0c +A -N/A +G -N/A +T -UL2-20B - -51.4 -CoT-prompting -Self-consistency 55.7 (+4.3) 54.9 (+1.6) - -53.3 - -61.6 -69.8 (+8.2) 49.5 (+6.8) - -42.9 - -0.0 -0.0 (+0.0) - -50.4 -50.5 (+0.1) - -LaMDA-137B - -CoT-prompting -57.9 -Self-consistency 63.1 (+5.2) 67.8 (+2.4) - -65.4 - -75.3 -79.3 (+4.0) 59.8 (+4.7) - -55.1 - -8.2 -8.2 (+0.0) - -72.4 -73.5 (+1.1) - -PaLM-540B - -79.0 -CoT-prompting -Self-consistency 80.7 (+1.7) 81.6 (+6.3) - -75.3 - -95.3 -96.4 (+1.1) 88.7 (+3.5) - -85.2 - -65.8 -70.8 (+5.0) 91.2 (+3.0) - -88.2 - -GPT-3 -Code-davinci-001 - -CoT-prompting -46.6 -Self-consistency 54.9 (+8.3) 61.7 (+5.0) - -56.7 - -63.1 -72.1 (+9.0) 53.7 (+10.6) 10.0 (+2.2) 75.9 (+4.5) - -43.1 - -71.4 - -7.8 - -GPT-3 -Code-davinci-002 - -CoT-prompting -79.0 -Self-consistency 81.5 (+2.5) 79.8 (+6.4) - -73.4 - -94.0 -96.0 (+2.0) 87.5 (+3.9) - -83.6 - -70.4 -73.4 (+3.0) 99.5 (+0.5) - -99.0 - -Table 3: Commonsense and symbolic reasoning accuracy by self-consistency compared to chain- -of-thought prompting (Wei et al., 2022). The previous SoTA baselines are obtained from: a: -DeBERTaV3-large + KEAR (Xu et al., 2021b), b: Chowdhery et al. (2022), c: UnifiedQA-FT -(Khashabi et al., 2020). The best performance for each task is shown in bold. - -7The standard deviation of self-consistency is ≤ 0.5 for all tasks and is thus omitted in the table. Please refer - -to Figure 2, Figure 7 and 8 for the standard deviations under varying numbers of sampled paths. - -5 - - Published as a conference paper at ICLR 2023 - -Commonsense and Symbolic Reasoning Table 3 shows the results on commonsense and symbolic -reasoning tasks. Similarly, self-consistency yields large gains across all four language models, and -obtained SoTA results on 5 out of 6 tasks. For symbolic reasoning, we test the out-of-distribution -(OOD) setting where the input prompt contains examples of 2-letters or 2-flips but we test examples -of 4-letters and 4-flips (this setting is more challenging as PaLM-540B or GPT-3 can already achieve -perfect in-distribution accuracy). In this challenging OOD setting, the gain of self-consistency is still -quite significant compared to CoT-prompting with sufficient model sizes. - -To show the effect of the number of sampled reasoning paths, we plot the accuracy (mean and -standard deviation over 10 runs) with respect to varying numbers of sampled paths (1, 5, 10, 20, 40) -in Figure 2. The results show that sampling a higher number (e.g., 40) of reasoning paths leads to a -consistently better performance, further emphasizing the importance of introducing diversity in the -reasoning paths. In Table 4, we show self-consistency yields a richer set of reasoning paths compared -to greedy decoding with a few example questions from two tasks. - -Figure 2: Self-consistency (blue) significantly improves accuracy over CoT-prompting with greedy -decoding (orange) across arithmetic and commonsense reasoning tasks, over LaMDA-137B. Sampling -a higher number of diverse reasoning paths consistently improves reasoning accuracy. - -GSM8K - -StrategyQA - -Question Henry made two stops during his 60-mile bike trip. He first stopped after 20 -miles. His second stop was 15 miles before the end of the trip. How many miles -did he travel between his first and second stops? - -Yes or no: Is Albany, Georgia the most populous -US Albany? - -Greedy -Decode - -He traveled 60 miles in total. He stopped after 20 miles, so he traveled 60 - 20 -= 40 miles between the first and second stops. The answer is 40. - -Albany, Georgia is the most populous US Al- -bany. Thus, the answer is yes. - -Sampled -Path 1 - -Sampled -Path 2 - -Henry travelled a total of 60 miles. His first stop was 20 miles in, and his -second stop was 15 miles before the end. So between his first and second stops -he travelled 60 - 20 -15 = 25 miles. The answer is 25. - -The most populous US Albany is Albany, New -York. Thus, Albany, Georgia is not the most -populous US Albany. So the answer is no. - -He made two stops during a 60-mile trip. The first was 20 miles into the trip. -The second was 15 miles before the end of the trip. This means the second stop -was 60 - 15 = 45 miles into the trip. Since he made the stops in order, the second -stop must have been 45 - 20 = 25 miles after the first stop. The answer is 25. - -Albany, Georgia has a population of about -88,000. Albany, New York has a population of -about 95,000. Thus, Albany, Georgia is not the -most populous US Albany. So the answer is no. - -Table 4: Examples where self-consistency helps repair the errors over greedy decode, on PaLM-540B. -Two sampled reasoning paths that are consistent with the ground truth are shown. - -3.3 SELF-CONSISTENCY HELPS WHEN CHAIN-OF-THOUGHT HURTS PERFORMANCE - -Ye & Durrett (2022) show that sometimes chain-of-thought prompting could hurt performance -compared to standard prompting in few-shot in-context learning. Here we perform a study using -self-consistency to see if it can help fill in the gap, over a set of common NLP tasks, including (1) -Closed-Book Question Answering: BoolQ (Clark et al., 2019), HotpotQA (Yang et al., 2018), and -(2) Natural Language Inference: e-SNLI (Camburu et al., 2018), ANLI (Nie et al., 2020) and RTE -(Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Bentivogli et al., 2009). - -The results over PaLM-540B are shown in Table 5. For some tasks (e.g., ANLI-R1, e-SNLI, RTE), -adding chain-of-thought does hurt performance compared to standard prompting (Brown et al., 2020), -but self-consistency is able to robustly boost the performance and outperform standard prompting, -making it a reliable way to add rationales in few-shot in-context learning for common NLP tasks. - -ANLI R1 / R2 / R3 e-SNLI RTE BoolQ HotpotQA (EM/F1) - -Standard-prompting (no-rationale) -CoT-prompting (Wei et al., 2022) - -69.1 / 55.8 / 55.8 -68.8 / 58.9 / 60.6 - -Self-consistency - -78.5 / 64.5 / 63.4 - -85.8 -81.0 - -88.4 - -84.8 -79.1 - -86.3 - -71.3 -74.2 - -78.4 - -27.1 / 36.8 -28.9 / 39.8 - -33.8 / 44.6 - -Table 5: Compare Standard/CoT prompting with self-consistency on common NLP tasks. - -6 - -0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths56586062Commonsense QA0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -3.4 COMPARE TO OTHER EXISTING APPROACHES - -We conduct a set of additional studies and show that self-consistency significantly outperforms -existing methods including sample-and-rank, beam search, and ensemble-based approaches. - -Comparison to Sample-and-Rank One commonly used approach to improve generation quality is -sample-and-rank, where multiple sequences are sampled from the decoder and then ranked according -to each sequence’s log probability (Adiwardana et al., 2020). We compare self-consistency with -sample-and-rank on GPT-3 code-davinci-001, by sampling the same number of sequences from the -decoder as self-consistency and taking the final answer from the top-ranked sequence. The results are -shown in Figure 3. While sample-and-rank does improve the accuracy with additionally sampled -sequences and ranking, the gain is much smaller compared to self-consistency. - -Figure 3: Self-consistency significantly outperforms sample-and-rank with the same # of samples. - -Comparison to Beam Search In Table 6, we compare self-consistency with beam search decoding -on the UL2-20B model. For a fair comparison we report the accuracy under the same number of -beams and reasoning paths. On both tasks self-consistency outperforms beam search significantly. -Note self-consistency can also adopt beam search to decode each reasoning path (results are shown -as “Self-consistency using beam search”), but its performance is worse compared to self-consistency -with sampling. The reason is that beam search yields a lower diversity in the outputs (Li & Jurafsky, -2016), while in self-consistency the diversity of the reasoning paths is the key to a better performance. - -Beam size / Self-consistency paths - -1 - -AQuA - -MultiArith - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -23.6 -23.6 - -10.7 -10.7 - -19.8 ± 0.3 21.2 ± 0.7 24.6 ± 0.4 24.2 ± 0.5 -19.7 ± 2.5 24.9 ± 2.6 25.3 ± 1.8 26.7 ± 1.0 26.9 ± 0.5 - -12.0 - -11.3 - -11.0 - -10.5 - -11.8 ± 0.0 11.4 ± 0.1 12.3 ± 0.1 10.8 ± 0.1 -9.5 ± 1.2 11.3 ± 1.2 12.3 ± 0.8 13.7 ± 0.9 14.7 ± 0.3 - -5 - -19.3 - -10 - -16.1 - -20 - -15.0 - -40 - -10.2 - -Table 6: Compare self-consistency with beam search decoding on the UL2-20B model. - -Comparison to Ensemble-based Approaches We further compare self-consistency to ensemble- -based methods for few-shot learning. In particular, we consider ensembling by: (1) prompt order -permutation: we randomly permute the exemplars in the prompt 40 times to mitigate model’s -sensitivity to prompt order (Zhao et al., 2021; Lu et al., 2021); and (2) multiple sets of prompts -(Gao et al., 2021): we manually write 3 different sets of prompts. We took majority vote of the -answers from greedy decoding in both approaches as an ensemble. Table 7 shows that compared to -self-consistency, existing ensemble-based approaches achieve a much smaller gain.8 In addition, note -that self-consistency is different from a typical model-ensemble approach, where multiple models -are trained and their outputs are aggregated. Self-consistency acts more like a “self-ensemble” on -top of a single language model. We additionally show the results of ensembling multiple models in -Appendix A.1.3 where the model-ensembles perform much worse compared to self-consistency. - -GSM8K MultiArith - -SVAMP - -ARC-e - -ARC-c - -CoT (Wei et al., 2022) -Ensemble (3 sets of prompts) -Ensemble (40 prompt permutations) -Self-Consistency (40 sampled paths) - -17.1 -18.6 ± 0.5 -19.2 ± 0.1 -27.7 ± 0.2 - -51.8 -57.1 ± 0.7 -60.9 ± 0.2 -75.7 ± 0.3 - -38.9 -42.1 ± 0.6 -42.7 ± 0.1 -53.3 ± 0.2 - -75.3 -76.6 ± 0.1 -76.9 ± 0.1 -79.3 ± 0.3 - -55.1 -57.0 ± 0.2 -57.0 ± 0.1 -59.8 ± 0.2 +C -Table 7: Self-consistency outperforms prompt-order and multi-prompt ensembles on LaMDA-137B. + Every cell in our body has the +same DNA…. -8Self-consistency is compatible with both ensemble approaches and we show the results in Appendix A.1.4. +Eye cell -7 +Karyotype -0510152025303540#Sampled Reasoning Paths12141618202224Accuracy (%)GSM8K0510152025303540#Sampled Reasoning Paths50556065707580Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths303540455055Accuracy (%)ARC (Challenge)Self Consistency (Multi-path)Sample & Rank (Multi-path)Greedy Decode (Single-path) Published as a conference paper at ICLR 2023 +Lung cell -3.5 ADDITIONAL STUDIES +Toe cell -We conducted a number of additional experiments to analyze different aspects of the self-consistency -method, including its robustness to sampling strategies and parameters, and how it works with -imperfect prompts and non-natural-language reasoning paths. + How much DNA is in one cell? -Self-Consistency is Robust to Sampling Strategies and Scaling We show self-consistency is -robust to sampling strategies and parameters, by varying T in temperature sampling (Ackley et al., -1985; Ficler & Goldberg, 2017), k in top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford -et al., 2019), and p in nucleus sampling (Holtzman et al., 2020), over PaLM-540B in Figure 4 (left). -Figure 4 (right) shows that self-consistency robustly improves performance across all scales for the -LaMDA-137B model series. The gain is relatively lower for smaller models due to certain abilities -(e.g., arithmetic) only emerge when the model reaches a sufficient scale (Brown et al., 2020). +Genome = 46 chromosomes -Figure 4: GSM8K accuracy. (Left) Self-consistency is robust to various sampling strategies and -parameters. (Right) Self-consistency improves performance across language model scales. +Genome = approx. 3 billion +base pairs -Self-Consistency Improves Robustness to Imperfect Prompts For few-shot learning with man- -ually constructed prompts, human annotators sometimes make minor mistakes when creating the -prompts. We further study if self-consistency can help improve a language model’s robustness to -imperfect prompts.9 We show the results in Table 8: while imperfect prompts decrease accuracy with -greedy decoding (17.1 → 14.9), self-consistency can fill in the gaps and robustly improve the results. +One base pair is +0.00000000034 meters -Additionally, we found that the consistency (in terms of % of decodes agreeing with the final -aggregated answer) is highly correlated with accuracy (Figure 5, over GSM8K). This suggests that -one can use self-consistency to provide an uncertainty estimate of the model in its generated solutions. -In other words, one can use low consistency as an indicator that the model has low confidence; i.e., -self-consistency confers some ability for the model to “know when it doesn’t know”. +DNA sequence in any two +people is 99.9% identical – +only 0.1% is unique! -Prompt with correct chain-of-thought + What makes one cell different from another? -17.1 - -LaMDA-137B - -Prompt with imperfect chain-of-thought 14.9 -23.4 - -+ Self-consistency (40 paths) - -Prompt with equations - -+ Self-consistency (40 paths) - -PaLM-540B - -Zero-shot CoT (Kojima et al., 2022) - -+ Self-consistency (40 paths) - -5.0 -6.5 - -43.0 -69.2 - -Table 8: Self-consistency works under imperfect prompts, equa- -tion prompts and zero-shot chain-of-thought for GSM8K. - -Figure 5: The consistency is cor- -related with model’s accuracy. - -Self-Consistency Works for Non-Natural-Language Reasoning Paths and Zero-shot CoT We -also tested the generality of the self-consistency concept to alternative forms of intermediate reasoning -like equations (e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there -are 3 + 2 = 5 cars.” to “3 + 2 = 5”). The results are shown in Table 8 (“Prompt with equations”): -self-consistency still improves accuracy by generating intermediate equations; however, compared to -generating natural language reasoning paths, the gain is smaller since the equations are much shorter -and less opportunity remains for generating diversity in the decoding process. In addition, we tested -self-consistency with zero-shot chain-of-thought (Kojima et al., 2022) and show that self-consistency -works for zero-shot CoT as well and improves the results significantly (+26.2%) in Table 8. - -9We use the same prompts as before, but swap all the numbers in the reasoning paths with random numbers -except the final answer, e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars.” to “There are 7 cars in the parking lot already. 6 more arrive. Now there are 7 + 6 = 5 cars.”. - -8 - -0510152025303540#Sampled Reasoning Paths444852566064687276Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.7, k=20T=0.7, no top kp=0.95p=0.9Greedy Decode125102050100200Model size (#param in billions)510152025Accuracy (%)Self ConsistencyGreedy Decode020406080100Consistency (%)020406080100Accuracy (%) Published as a conference paper at ICLR 2023 - -4 RELATED WORK - -Reasoning in language models. Language models are known to struggle in Type 2 tasks, such as -arithmetic, logical and commonsense reasoning (Evans, 2010). Previous work has primarily focused -on specialized approaches for improving reasoning (Andor et al., 2019; Ran et al., 2019; Geva et al., -2020; Pi˛ekos et al., 2021). Compared to prior work, self-consistency is applicable to a wide range of -reasoning tasks without any additional supervision or fine-tuning, while still substantially improving -the performance of the chain-of-thought prompting approach proposed in Wei et al. (2022). -Sampling and re-ranking in language models. Multiple decoding strategies for language models -have been proposed in the literature, e.g., temperature sampling (Ackley et al., 1985; Ficler & -Goldberg, 2017), top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford et al., 2019), -nucleus sampling (Holtzman et al., 2020), minimum Bayes risk decoding (Eikema & Aziz, 2020; Shi -et al., 2022), and typical decoding (Meister et al., 2022). Other work has sought to explicitly promote -diversity in the decoding process (Batra et al., 2012; Li et al., 2016; Vijayakumar et al., 2018). - -Re-ranking is another common approach to improve generation quality in language models (Adiwar- -dana et al., 2020; Shen et al., 2021). Thoppilan et al. (2022) collect additional human annotations -to train a re-ranker for response filtering. Cobbe et al. (2021) train a “verifier” to re-rank generated -solutions, which substantially improves the solve rate on math tasks compared to just fine-tuning the -language model. Elazar et al. (2021) improve the consistency of factual knowledge extraction by -extending pre-training with an additional consistency loss. All these methods require either training -an additional re-ranker or collecting additional human annotation, while self-consistency requires no -additional training, fine-tuning, nor extra data collection. -Extract reasoning paths. Some previous work has considered task-specific approaches for iden- -tifying reasoning paths, such as constructing semantic graphs (Xu et al., 2021a), learning an RNN -to retrieve reasoning paths over the Wikipedia graph (Asai et al., 2020), fine-tuning with human -annotated reasoning paths on math problems (Cobbe et al., 2021), or training an extractor with -heuristic-based pseudo reasoning paths (Chen et al., 2019). More recently, the importance of di- -versity in the reasoning processes has been noticed, but only leveraged via task-specific training, -either through an additional QA model over extracted reasoning paths (Chen et al., 2019), or by the -introduction of latent variables in a commonsense knowledge graph (Yu et al., 2022). Compared to -these approaches, self-consistency is far simpler and requires no additional training. The approach -we propose simply couples the generation of reasoning paths and a final answer by sampling from -the decoder, using aggregation to recover the most consistent answer without additional modules. -Consistency in language models. Some prior work has shown that language models can suffer -from inconsistency in conversation (Adiwardana et al., 2020), explanation generation (Camburu et al., -2020), and factual knowledge extraction (Elazar et al., 2021). Welleck et al. (2020) use “consistency” -to refer to generating an infinite-length sequence in recurrent language models. Nye et al. (2021) -improve the logical consistency of samples from a System 1 model by adding a System 2-inspired -logical reasoning module. In this paper we focus on a slightly different notion of “consistency”, i.e., -utilizing answer consistency among diverse reasoning paths to improve accuracy. - -5 CONCLUSION AND DISCUSSION - -We introduced a simple yet effective method called self-consistency, and observed that it significantly -improves accuracy in a range of arithmetic and commonsense reasoning tasks, across four large -language models with varying scales. Beyond accuracy gains, self-consistency is also useful for -collecting rationales when performing reasoning tasks with language models, and for providing -uncertainty estimates and improved calibration of language model outputs. - -One limitation of self-consistency is that it incurs more computation cost. In practice people can try a -small number of paths (e.g., 5 or 10) as a starting point to realize most of the gains while not incurring -too much cost, as in most cases the performance saturates quickly (Figure 2). As part of future work, -one could use self-consistency to generate better supervised data to fine-tune the model, such that the -model can give more accurate predictions in a single inference run after fine-tuning. In addition, we -observed that language models can sometimes generate incorrect or nonsensical reasoning paths (e.g., -the StrategyQA example in Table 4, the two population numbers are not exactly correct), and further -work is needed to better ground models’ rationale generations. - -9 - - Published as a conference paper at ICLR 2023 - -REPRODUCIBILITY STATEMENT - -In experiments, we included four different language models with varying scales. Two of them are pub- -lic models: UL2 is a completely open-sourced model with model checkpoints available at https:// -github.com/google-research/google-research/tree/master/ul2; GPT-3 is -also a public model with public API available at https://openai.com/api/. For GPT-3, -we have included two public engines (“code-davinci-001” and “code-davinci-002”) to further aid -reproducibility, as Codex is currently free so anyone can reproduce the results. In addition, as our -results make use of LaMDA-137B and PaLM-540B that are not publicly available, we provide the -exact input prompts for all tasks in Appendix A.3 (and note that we do not perform any finetuning -and only apply prompting to off-the-shelf language models). - -ETHICS STATEMENT - -As we stated in the discussion, language models can sometimes generate nonsensical or non-factual -reasoning paths, so one should use language models’ outputs with extra caution. We deal with -reasoning tasks mostly and the generated rationales are only used for inspecting how a model reaches -its answer. One could potentially use the generated rationales to further check why the model makes -certain mistakes or whether the model contains any biases when performing a certain task. For -language model in real-world use, further work is needed to better ground models’ predictions and -improve model’s factuality and safety, to ensure the models do not cause harms to users. - -REFERENCES - -David H. Ackley, Geoffrey E. Hinton, and Terrence J. Sejnowski. A learning algorithm for boltzmann -machines. Cognitive Science, 9(1):147–169, 1985. ISSN 0364-0213. URL https://www. -sciencedirect.com/science/article/pii/S0364021385800124. - -Daniel Adiwardana, Minh-Thang Luong, David R. So, Jamie Hall, Noah Fiedel, Romal Thoppilan, -Zi Yang, Apoorv Kulshreshtha, Gaurav Nemade, Yifeng Lu, and Quoc V. Le. Towards a human-like -open-domain chatbot, 2020. - -Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh -Hajishirzi. MathQA: Towards interpretable math word problem solving with operation-based -In Proceedings of the 2019 Conference of the North American Chapter of the -formalisms. -Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long -and Short Papers), pp. 2357–2367. Association for Computational Linguistics, June 2019. URL -https://aclanthology.org/N19-1245. - -Daniel Andor, Luheng He, Kenton Lee, and Emily Pitler. Giving BERT a calculator: Finding -operations and arguments with reading comprehension. In Proceedings of the 2019 Conference on -Empirical Methods in Natural Language Processing and the 9th International Joint Conference -on Natural Language Processing (EMNLP-IJCNLP), 2019. URL https://aclanthology. -org/D19-1609. - -Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learn- -ing to retrieve reasoning paths over wikipedia graph for question answering. In International -Conference on Learning Representations, 2020. URL https://openreview.net/forum? -id=SJgVHkrYDH. - -Roy Bar-Haim, Ido Dagan, Bill Dolan, Lisa Ferro, Danilo Giampiccolo, Bernardo Magnini, and -Idan Szpektor. The second pascal recognising textual entailment challenge. In Proceedings of the -second PASCAL challenges workshop on recognising textual entailment, 2006. - -Dhruv Batra, Payman Yadollahpour, Abner Guzman-Rivera, and Gregory Shakhnarovich. Diverse -m-best solutions in markov random fields. In Proceedings of the 12th European Conference on -Computer Vision - Volume Part V, ECCV’12, pp. 1–16, Berlin, Heidelberg, 2012. Springer-Verlag. -ISBN 9783642337147. URL https://doi.org/10.1007/978-3-642-33715-4_1. - -10 - - Published as a conference paper at ICLR 2023 - -Luisa Bentivogli, Peter Clark, Ido Dagan, and Danilo Giampiccolo. The fifth pascal recognizing - -textual entailment challenge. In TAC, 2009. - -BIG-bench collaboration. Beyond the imitation game: Measuring and extrapolating the capabil- -In preparation, 2021. URL https://github.com/google/ - -ities of language models. -BIG-bench/. - -Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, -Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel -Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, -Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, -Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, -and Dario Amodei. Language models are few-shot learners. In Advances in Neural Information -Processing Systems, 2020. URL https://proceedings.neurips.cc/paper/2020/ -file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. - -language inference with natural - -Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. - -snli: Natural -gio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, -(eds.), Advances -Information Processing Systems 31, -Curran Associates, -8163-e-snli-natural-language-inference-with-natural-language-explanations. -pdf. - -e- -In S. Ben- -and R. Garnett -pp. 9539–9549. -URL http://papers.nips.cc/paper/ - -language explanations. - -in Neural -Inc., - -2018. - -Oana-Maria Camburu, Brendan Shillingford, Pasquale Minervini, Thomas Lukasiewicz, and Phil -Blunsom. Make up your mind! adversarial generation of inconsistent natural language explanations. -In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. -4157–4165, Online, July 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -acl-main.382. URL https://aclanthology.org/2020.acl-main.382. - -Jifan Chen, Shih-Ting Lin, and Greg Durrett. Multi-hop question answering via reasoning chains. - -CoRR, abs/1910.02610, 2019. URL http://arxiv.org/abs/1910.02610. - -Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared -Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large -language models trained on code. arXiv preprint arXiv:2107.03374, 2021. +DNA = “the life +instructions of the +cell” -Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam -Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, -Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam -Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James -Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Lev- -skaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin -Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, -Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. -Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon -Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark -Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeff Dean, -Slav Petrov, and Noah Fiedel. Palm: Scaling language modeling with pathways, 2022. URL -https://arxiv.org/abs/2204.02311. - -Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina -Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. In NAACL, -2019. - -Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and -Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. -ArXiv, abs/1803.05457, 2018. - -Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, -Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John -Schulman. Training verifiers to solve math word problems, 2021. +Gene = segment of +DNA that tells the +cell how to make a +certain protein. -11 +Allele = one of two +or more different +versions of a gene - Published as a conference paper at ICLR 2023 + Sequence for normal adult hemoglobin: -Ido Dagan, Oren Glickman, and Bernardo Magnini. The pascal recognising textual entailment +Sequence for mutant hemoglobin: -challenge. In Machine Learning Challenges Workshop, pp. 177–190. Springer, 2005. +Wild-type Hemoglobin Protein -Bryan Eikema and Wilker Aziz. Is MAP decoding all you need? the inadequacy of the mode in neural -machine translation. In Proceedings of the 28th International Conference on Computational Lin- -guistics, pp. 4506–4520, Barcelona, Spain (Online), December 2020. International Committee on -Computational Linguistics. URL https://aclanthology.org/2020.coling-main. -398. +Mutant Protein -Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich -Schütze, and Yoav Goldberg. Measuring and improving consistency in pretrained language -models. Transactions of the Association for Computational Linguistics, 9:1012–1031, 2021. doi: -10.1162/tacl_a_00410. URL https://aclanthology.org/2021.tacl-1.60. - -Jonathan St BT Evans. Intuition and reasoning: A dual-process perspective. Psychological Inquiry, - -21(4):313–326, 2010. - -Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Proceedings -of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long -Papers), pp. 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics. -doi: 10.18653/v1/P18-1082. URL https://aclanthology.org/P18-1082. - -Jessica Ficler and Yoav Goldberg. Controlling linguistic style aspects in neural language generation. In -Proceedings of the Workshop on Stylistic Variation, pp. 94–104, Copenhagen, Denmark, September -2017. Association for Computational Linguistics. doi: 10.18653/v1/W17-4912. URL https: -//aclanthology.org/W17-4912. - -Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot -learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguis- -tics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long -Papers), pp. 3816–3830, Online, August 2021. Association for Computational Linguistics. doi: -10.18653/v1/2021.acl-long.295. URL https://aclanthology.org/2021.acl-long. -295. - -Mor Geva, Ankit Gupta, and Jonathan Berant. Injecting numerical reasoning skills into language -models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguis- -tics, 2020. doi: 10.18653/v1/2020.acl-main.89. URL https://aclanthology.org/2020. -acl-main.89. +Normal Red Blood Cell -Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle -use a laptop? A question answering benchmark with implicit reasoning strategies. Transactions of -the Association for Computational Linguistics, 2021. URL https://aclanthology.org/ -2021.tacl-1.21. +Abnormal Red Blood Cell -Danilo Giampiccolo, Bernardo Magnini, Ido Dagan, and Bill Dolan. The third pascal recognizing -textual entailment challenge. In Proceedings of the ACL-PASCAL workshop on textual entailment -and paraphrasing, pp. 1–9. Association for Computational Linguistics, 2007. + The Human Genome Project Goals -Ari Holtzman, Jan Buys, Maxwell Forbes, Antoine Bosselut, David Golub, and Yejin Choi. Learning -In Proceedings of the 56th Annual Meeting of the -to write with cooperative discriminators. -Association for Computational Linguistics (Volume 1: Long Papers), pp. 1638–1649, Melbourne, -Australia, July 2018. Association for Computational Linguistics. doi: 10.18653/v1/P18-1152. -URL https://aclanthology.org/P18-1152. +• To sequence (i.e. determine the exact order of +nucleotides (A,T,G,C) for ALL of the DNA in a +human cell -Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text -degeneration. In International Conference on Learning Representations, 2020. URL https: -//openreview.net/forum?id=rygGQyrFvH. +• To determine which sections of DNA represent -Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to -solve arithmetic word problems with verb categorization. In Proceedings of the 2014 Conference on -Empirical Methods in Natural Language Processing (EMNLP), 2014. doi: 10.3115/v1/D14-1058. -URL https://aclanthology.org/D14-1058. +individual genes (protein-coding units). -12 - - Published as a conference paper at ICLR 2023 - -Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Han- -naneh Hajishirzi. UNIFIEDQA: Crossing format boundaries with a single QA system. In Findings -of the Association for Computational Linguistics: EMNLP 2020, pp. 1896–1907, Online, Novem- -ber 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.findings-emnlp.171. -URL https://aclanthology.org/2020.findings-emnlp.171. - -Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large -language models are zero-shot reasoners. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, -and Kyunghyun Cho (eds.), Advances in Neural Information Processing Systems, 2022. URL -https://openreview.net/forum?id=e2TBb5y0yFf. + The HGP: International effort to decipher +the blueprint of a human being. -Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. MAWPS: -A math word problem repository. In Proceedings of the 2016 Conference of the North American -Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016. -doi: 10.18653/v1/N16-1136. URL https://aclanthology.org/N16-1136. + How It Was Done -Yihuai Lan, Lei Wang, Qiyuan Zhang, Yunshi Lan, Bing Tian Dai, Yan Wang, Dongxiang Zhang, -and Ee-Peng Lim. MWPToolkit: An open-source framework for deep learning-based math word -problem solvers. arXiv preprint arXiv:2109.00799, 2021. URL https://arxiv.org/abs/ -2109.00799. +DNA samples collected from +thousands of volunteers -Jiwei Li and Dan Jurafsky. Mutual information and diverse decoding improve neural machine +Samples sent to Human Genome +Project centers across the world -translation, 2016. URL https://arxiv.org/abs/1601.00372. +Scientists at centers perform DNA +sequencing and analysis -Jiwei Li, Will Monroe, and Dan Jurafsky. A simple, fast diverse decoding algorithm for neural -generation. CoRR, abs/1611.08562, 2016. URL http://arxiv.org/abs/1611.08562. + • February 2001: Draft of the -Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale genera- -tion: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual -Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2017. doi: -10.18653/v1/P17-1015. URL https://aclanthology.org/P17-1015. - -Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. Fantastically -ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. ArXiv, -abs/2104.08786, 2021. - -Clara Meister, Tiago Pimentel, Gian Wiher, and Ryan Cotterell. Typical decoding for natural language - -generation. arXiv preprint arXiv:2202.00666, 2022. - -Shen Yun Miao, Chao Chun Liang, and Keh Yih Su. A diverse corpus for evaluating and developing -English math word problem solvers. In Proceedings of the 58th Annual Meeting of the Asso- -ciation for Computational Linguistics, 2020. URL https://aclanthology.org/2020. -acl-main.92. - -Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adver- -sarial NLI: A new benchmark for natural language understanding. In Proceedings of the 58th -Annual Meeting of the Association for Computational Linguistics. Association for Computational -Linguistics, 2020. - -Maxwell Nye, Michael Henry Tessler, Joshua B. Tenenbaum, and Brenden M. Lake. Improving -coherence and consistency in neural sequence models with dual-system, neuro-symbolic reasoning. -In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan (eds.), Advances in Neural -Information Processing Systems, 2021. URL https://openreview.net/forum?id= -uyKk_avJ-p4. - -Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP models really able to solve simple -math word problems? In Proceedings of the 2021 Conference of the North American Chapter of -the Association for Computational Linguistics: Human Language Technologies, pp. 2080–2094, -Online, June 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main. -168. URL https://aclanthology.org/2021.naacl-main.168. - -Xinyu Pi, Qian Liu, Bei Chen, Morteza Ziyadi, Zeqi Lin, Yan Gao, Qiang Fu, Jian-Guang Lou, and - -Weizhu Chen. Reasoning like program executors, 2022. - -13 - - Published as a conference paper at ICLR 2023 - -Piotr Pi˛ekos, Mateusz Malinowski, and Henryk Michalewski. Measuring and improving BERT’s -mathematical abilities by predicting the order of reasoning. In Proceedings of the 59th Annual Meet- -ing of the Association for Computational Linguistics and the 11th International Joint Conference on -Natural Language Processing (Volume 2: Short Papers), 2021. doi: 10.18653/v1/2021.acl-short.49. -URL https://aclanthology.org/2021.acl-short.49. - -Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language - -models are unsupervised multitask learners. 2019. - -Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John -Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: -Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446, 2021. +sequence published in Nature +(public effort )and Science +(Celera – private company). -Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. NumNet: Machine reading comprehension -with numerical reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural -Language Processing and the 9th International Joint Conference on Natural Language Processing -(EMNLP-IJCNLP), 2019. doi: 10.18653/v1/D19-1251. URL https://aclanthology. -org/D19-1251. +• April, 2003 (50 years after -Subhro Roy and Dan Roth. Solving general arithmetic word problems. In Proceedings of the 2015 -Conference on Empirical Methods in Natural Language Processing, 2015. doi: 10.18653/v1/ -D15-1202. URL https://aclanthology.org/D15-1202. +Watson and Crick structure of +DNA was published) : Full +sequence published and +researchers determined that +within this sequence there was +somewhere between 30,000 +and 40,000 genes. We now +believe there are closer to +25,000 genes -Jianhao Shen, Yichun Yin, Lin Li, Lifeng Shang, Xin Jiang, Ming Zhang, and Qun Liu. Generate -& rank: A multi-task framework for math word problems. In Findings of the Association for -Computational Linguistics: EMNLP 2021, pp. 2269–2279, Punta Cana, Dominican Republic, -November 2021. Association for Computational Linguistics. URL https://aclanthology. -org/2021.findings-emnlp.195. + Still A Lot of Work To Do . . . -Freda Shi, Daniel Fried, Marjan Ghazvininejad, Luke Zettlemoyer, and Sida I. Wang. Natural -language to code translation with execution. In Proceedings of the 2022 Conference on Empirical -Methods in Natural Language Processing, pp. 3533–3546, Abu Dhabi, United Arab Emirates, -December 2022. Association for Computational Linguistics. URL https://aclanthology. -org/2022.emnlp-main.231. +Comparing the +genomes of +humans and +other organisms -Keith E Stanovich and Richard F West. +Discovering +DNA and gene +functions -Individual differences in reasoning: Implications for -the rationality debate? Behavioral and brain sciences, 23(5):645–665, 2000. URL https: -//pubmed.ncbi.nlm.nih.gov/11301544/. +Investigating +interactions between +DNA sequences, gene +products, and +environmental factors -Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. CommonsenseQA: A question -answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of -the North American Chapter of the Association for Computational Linguistics: Human Language -Technologies, Volume 1 (Long and Short Papers), 2019. URL https://aclanthology. -org/N19-1421. +Analyzing +genetic variation +between +individuals and +populations -Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Jason Wei, Xuezhi Wang, Hyung Won Chung, -Dara Bahri, Tal Schuster, Steven Zheng, Denny Zhou, Neil Houlsby, and Donald Metzler. Unifying -language learning paradigms, 2022. URL https://arxiv.org/abs/2205.05131. + How Can We Use This Information? -Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze -Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog -applications. arXiv preprint arXiv:2201.08239, 2022. URL https://arxiv.org/abs/ -2201.08239. +Better understanding of human +disease -Ashwin Vijayakumar, Michael Cogswell, Ramprasaath Selvaraju, Qing Sun, Stefan Lee, David -Crandall, and Dhruv Batra. Diverse beam search for improved description of complex scenes. -Proceedings of the AAAI Conference on Artificial Intelligence, 32, Apr. 2018. URL https: -//ojs.aaai.org/index.php/AAAI/article/view/12340. +Personalized medicine & +Pharmacogenetics -Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc -Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. -Conference on Neural Information Processing Systems (NeurIPS), 2022. URL https://arxiv. -org/pdf/2201.11903. +Greater +insight into +cognitive +function -14 +Insight into +human origins - Published as a conference paper at ICLR 2023 +Identifying genetic susceptibility to disease -Sean Welleck, Ilia Kulikov, Jaedeok Kim, Richard Yuanzhe Pang, and Kyunghyun Cho. Consistency -of a recurrent language model with respect to incomplete decoding. In Proceedings of the 2020 -Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5553–5568, -Online, November 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -emnlp-main.448. URL https://aclanthology.org/2020.emnlp-main.448. + Inheritance of Genes -Weiwen Xu, Yang Deng, Huihui Zhang, Deng Cai, and Wai Lam. Exploiting reasoning chains -for multi-hop science question answering. In Findings of the Association for Computational -Linguistics: EMNLP 2021, pp. 1143–1156, Punta Cana, Dominican Republic, November 2021a. -Association for Computational Linguistics. URL https://aclanthology.org/2021. -findings-emnlp.99. + Gregor Mendel -Yichong Xu, Chenguang Zhu, Shuohang Wang, Siqi Sun, Hao Cheng, Xiaodong Liu, Jianfeng -Gao, Pengcheng He, Michael Zeng, and Xuedong Huang. Human parity on commonsenseqa: -Augmenting self-attention with external attention, 2021b. URL https://arxiv.org/abs/ -2112.03254. +1822-1884 -Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and -Christopher D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answer- -ing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, -pp. 2369–2380, Brussels, Belgium, October-November 2018. Association for Computational Lin- -guistics. doi: 10.18653/v1/D18-1259. URL https://aclanthology.org/D18-1259. +• Augustinian monk who -Xi Ye and Greg Durrett. The unreliability of explanations in few-shot prompting for textual reasoning. -In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho (eds.), Advances in -Neural Information Processing Systems, 2022. URL https://openreview.net/forum? -id=Bct2f8fRd8S. +cross-bred pea plants with +different characteristics +• Observations led to laws -Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, and Meng Jiang. Diversifying -content generation for commonsense reasoning with mixture of knowledge graph experts. In -Findings of Annual Meeting of the Association for Computational Linguistics (ACL), 2022. +regarding the transmission +of hereditary +characteristics from +generation to generation +• Many of the concepts from +his observations still hold +true today! -Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. Calibrate before use: Improving -few-shot performance of language models. In Marina Meila and Tong Zhang (eds.), Proceed- -ings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of -Machine Learning Research. PMLR, 2021. URL https://proceedings.mlr.press/ -v139/zhao21c.html. +Picture from www.nih.nlm.gov -15 + Mendel’s +Laws: +1. Principle of Segregation: Two members of a - Published as a conference paper at ICLR 2023 +gene pair segregate from each other in the +formation of gametes; half the gametes carry +one allele, and the other half carry the other +allele +What it means: each gene has two copies +(alleles) and a parent will give only one copy to +a child. The other parent will give another +copy, and thus the child will receive two copies +(alleles) – one from each parent. Each child +will literally be half-mom and half-dad! -A APPENDIX + Mendel’s +Laws: -A.1 ADDITIONAL EXPERIMENT RESULTS +2. Principle of Independent Assortment: -A.1.1 ROBUSTNESS TO SAMPLING STRATEGIES AND PARAMETERS +Genes for different traits assort +independently of one another in gamete +production +What it means: different genes are +inherited separately. For example, the +gene which codes for eye color is inherited +separately from the gene which codes for +nose shape. -In Figure 6 we ablate the results with respect to different sampling strategies and parameters by -varying T in temperature sampling and k in Top-k sampling, on LaMDA-137B. We show that -self-consistency is robust to various sampling strategies and parameters. + Mendelian Concepts -Figure 6: GSM8K accuracy over LaMDA-137B. Self-consistency works under various sampling -strategies and sampling parameters. +Dominant = only one allele of a gene +necessary to express the trait +Recessive = both alleles of a gene must +be identical to express the trait +Heterozygous = alleles of a particular gene +are non-identical +Homozygous = alleles of a particular gene +are identical -In Figure 7 and Figure 8, we show the results of self-consistency compared with greedy decoding a -single path over LaMDA-137B and PaLM-540B, respectively. Self-consistency improves over greedy -decode by a quite significant margin on both models, on top of high accuracy already achieved by -scaling up model sizes. + Applying Mendelian Concepts: +The Punnett Square -Figure 7: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over LaMDA-137B. Sampling a higher number of diverse reasoning -paths consistently improves reasoning accuracy. +Parent 2 -We further show additional sampled reasoning paths from the LaMDA-137B model in Table 12, and -sampled reasoning paths from the PaLM-540B model in Table 13. We see that the diversity in the -additionally sampled reasoning paths indeed helps the model arrive at a more correct final answer -after aggregation. +Allele 1 -A.1.2 ROBUSTNESS TO DIFFERENT SETS OF PROMPTS +Allele 2 -In Table 9, we further show that self-consistency is quite robust to different sets of input prompts. -We manually wrote 3 different sets of chain-of-thought as prompts to the model. Across all sets of -prompts, self-consistency yields consistent gains over the original CoT approach. +Allele 1 -A.1.3 COMPARED TO MODEL ENSEMBLES +A1/A1 A1/A2 -Additionally, we provide results of directly ensembling the outputs from multiple language models. -The results are shown in Table 10, by greedily decoding sequences from 3 language models and +Parent 1 -16 +Allele 2 -481216202428323640#Sampled Reasoning Paths182022242628Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.5, k=20T=0.5, no top kGreedy Decode0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths4446485052545658ASDiv0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths1416182022242628GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths56586062Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths62636465666768Strategy QA0510152025303540#Sampled Reasoning Paths687072747678ARC (Easy)0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +A2/A1 A2/A2 -Figure 8: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over PaLM-540B. Sampling a higher number of diverse reasoning -paths consistently helps reasoning accuracy. + Homozygous dominant + +Homozygous dominant -Prompt set 1 (used in the main text) +Homozygous dominant + +Heterozygous -Prompt set 2 +A -Prompt set 3 +A -CoT (Wei et al., 2022) -Self-consistency +A -56.5 -74.4 (+17.9) +A -54.6 -72.1 (+17.5) +A -54.0 -70.4 (+16.4) +AA -Table 9: GSM8K accuracy over PaLM-540B. The results show robustness of self-consistency with -respect to different prompts in the input. +AA -taking the majority vote (averaged over 10 runs). Note this is a typical ensemble approach (averaging -over the predictions over multiple models) and it achieves a performance significantly worse than -self-consistency (self-consistency over PaLM-540B gets an accuracy of 74.4%), as lower-capacity -models drag down the performance of higher-capacity models. In addition, this approach is limited in -two ways: 1) It requires multiple models for an ensemble which might not always be available, while -self-consistency only requires one single model to “self-ensemble”; 2) If one of the models is much -weaker, it can actually hurt the final performance. +A -Method +AA -Single model +AA -PaLM-540B, greedy / self-consistency +A -Ensemble of models +AA -LaMDA-137B + PaLM-540B -PaLM-540B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + PaLM-540B + GPT-3 (code-davinci-001, 175B) +AA -GSM8K accuracy +a -56.5 / 74.4 +Aa -36.9 ± 0.5 -36.6 ± 0.4 -16.0 ± 0.8 -33.3 ± 0.7 +Aa -Table 10: Comparison of GSM8K accuracy over multiple-model ensembles. +Homozygous dominant + +Homozygous recessive -A.1.4 COMBINING SELF-CONSISTENCY WITH OTHER ENSEMBLING STRATEGIES +a -Self-consistency is completely compatible with other ensemble strategies, although the gains achieved -by self-consistency are significantly higher than other ensemble strategies (and can “override” the -performance gains achieved by other ensemble strategies). We further performed experiments and -include the results in Table 11 (for a fair comparison, we use 40 sets of prompts, or 40 prompt -permutations to compare with self-consistency with 40 paths, all experiments are based on PaLM- -540B). +a -17 +A -0510152025303540#Sampled Reasoning Chains8688909294Accuracy (%)AddSub0510152025303540#Sampled Reasoning Chains727476788082Accuracy (%)ASDiv0510152025303540#Sampled Reasoning Chains30333639424548Accuracy (%)AQuAGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Chains889092949698Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Chains70.072.575.077.580.082.585.087.5Accuracy (%)SVAMP0510152025303540#Sampled Reasoning Chains505560657075Accuracy (%)GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths7475767778798081Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths7476788082Strategy QA0510152025303540#Sampled Reasoning Paths8890929496ARC (Easy)0510152025303540#Sampled Reasoning Paths788082848688ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +Aa -GSM8K accuracy +Aa -Different sets of prompts (x40) -Prompt-permutation (x40) -Self-consistency (x40) -Self-consistency + different sets of prompts (x40) -Self-consistency + prompt-permutation (x40) +A -58.9 -59.6 -74.4 -75.4 -73.8 +Aa -Table 11: Combining self-consistency with other ensembling strategies. +Aa -A.2 DETAILS ON RESOURCES AND INFERENCE +Heterozygous + Heterozygous -For all four language models we perform prompting-based inference only. For UL2 we use TPU v3 -(2x2 configuration, 4 chips, 8 cores). For GPT-3 models the experiments are done though the public -API.10 For LaMDA-137B we use TPU v3 (8x8 configuration, 64 chips, 128 cores). For PaLM-540B -we use TPU v4 (4x4x12 configuration, 192 chips, 384 cores). Most inference jobs take 1 to 4 hours -(over about 1,000 examples) for each task on UL2 and LaMDA-137B, and about 2 to 12 hours on -PaLM-540B. Some tasks (e.g., commonsense reasoning) take longer but do not exceed 2 days for -each task. +A -For GPT-3 models, we use 128 max tokens for all methods, without frequency penalty or presence -penalty. For all models, we take the generated outputs until the start of the next “Q:" to parse the final -answers, consistent with our prompting format. +a -Question +AA -Greedy Decode +Aa -Sampled Path 1 - -Sampled Path 2 - -[MultiArith] Sam had 10 socks. -If he threw away 3 old ones -that didn’t fit and bought 36 new -ones, how many socks would he -have? - -Sam had 10 socks. He threw -away 3 old ones that didn’t fit. -He bought 36 new ones. That -means he has 10 + 36 - 3 = 41 -socks. The answer is 41. - -[SVAMP] In a school there are -697 girls and the rest are boys. -If there are 228 more girls than -boys. How many boys are there -in that school? - -[AQuA] There are 10 oranges in -a basket. Find the no. of ways in -which 2 oranges are chosen from -the basket? (a) 45 (b) 90 (c) 120 -(d) 150 (e) 180 - -There are 697 girls and the rest -are boys. That means there are -697 + 228 = 925 girls. The -rest are boys. So the number of -boys is 925 - 697 = 228. The -answer is 228. - -There are 10 oranges in a bas- -ket. -So there are 10 ways -in which 2 oranges are chosen -from the basket. - -[CommonsenseQA] The man -laid on the soft moss and looked -up at the trees, where was the -man? (a) niagra falls (b) forest -(c) waterfall (d) ground (e) tree - -The answer must be something -that is soft and mossy. Of the -above choices, only ground is -soft and mossy. So the answer -is (d). - -[ARC] Which of these will most -likely harm a habitat? (a) plant- -ing trees. (b) water pollution. (c) -rainfall. (d) sunlight. - -Planting trees will not harm a -habitat. So the answer is (a). - -[ARC-challenge] Which pro- -cess in the carbon cycle takes the -longest to complete? (a) emis- -sion of waste. (b) respiration in -animals. -(c) photosynthesis in -plants. (d) formation of fossil fu- -els. - -The process of photosynthesis -in plants takes the longest to -complete. So the answer is (c). - -Sam had 10 socks. He -threw away 3 old ones. So -now he has 7 left. Then he -bought 36 new ones. So -now he has 7 + 36 = 43 -socks. The answer is 43. - -He had 10 socks. He threw away -3 old ones that didn’t fit. He -bought 36 new ones. So in total -he has 10 - 3 + 36 = 43 socks. The -answer is 43. - -The number of boys is the -difference between 697 -and 228. This is 697 - 228 -= 469. The answer is 469. - -In the beginning there are 697 -girls and the rest are boys. So, the -number of boys must be 697 - 228 -= 469 boys. The answer is 469. - -There are 10 oranges in a -basket. We need to choose -2 oranges from the basket. -So the number of ways are -10C2. So the answer is (a). - -The -be -answer must -somewhere where moss -is found. Of the above -choices, only forests are -known for having moss. -So the answer is (b). - -The answer is (b), water -pollution, because it will -most likely harm the habi- -tat. The other answers are -not likely to harm the habi- -tat. - -The formation of fossil fu- -els takes the longest time -to complete. So the an- -swer is (d). - -There are 10 oranges in the bas- -ket. If you choose 2 oranges, you -have 10C2 = 45 ways of choosing -2 oranges. So the answer is (a). - -The answer must be a place where -the man is laying down and look- -ing up at trees. Of the above -is a place -choices, only forest -where people lay down and look -at trees. So the answer is (b). - -The answer is (b), water pollution. -Trees are good for a habitat, rain- -fall is necessary for a habitat, and -sunlight is necessary for a habitat. -So the answer is (b). - -The process that takes the longest -time to complete is the forma- -tion of fossil fuels, which hap- -pens over millions of years. So -the answer is (d). - -Table 12: Additional examples where self-consistency helps repair the errors over greedy decode on -LaMDA-137B. Two sampled reasoning paths that are consistent with the ground truth are shown. - -10https://beta.openai.com/docs/api-reference/making-requests - -18 - - Kyle bought a book that is sold -for $19.50. With a 25% discount, -the original price must be higher. -If we call -the original price x, -then the discounted price is 0.75x. -Since this is $19.50, we can write -0.75x = 19.50. Solving this, we -find that the original price is x = -19.50 / 0.75 = $26. The answer is -$26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. One cu- -bic meter contains 1000 litres, so -the tank contains 120 m3 * 1000 -liters/m3 = 120,000 liters. So the -answer is (d). - -Published as a conference paper at ICLR 2023 - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[GSM8K] Kyle bought -last -year’s best-selling book for -$19.50. This is with a 25% dis- -count from the original price. -What was the original price of -the book? - -The book was sold for $19.50. -This is 25% less than the origi- -nal price. This means the original -price was 100%. 25% of 100% -is 25%. -So the original price -was 100% - 25% = 75%. The -original price was 75% of $19.50. -75% of $19.50 is 75% * $19.50 = -$14.625. The answer is $14.625. - -The original price of the book -is $19.50. Since it was dis- -counted by 25%, the original -price was 19.50 / 0.75 = 26. -The answer is 26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. So the -capacity of the tank is 120 m3. So -the answer is (c). - -The volume of the tank is 8m -* 6m* 2.5m = 120m3. This is -equal to 120000 litres. So the -answer is (d) - -[AQuA] The capacity of a -tank of dimensions (8 m x 6 -m x 2.5 m) is (a) 120 litres (b) -1200 litres (c) 12000 litres (d) -120000 litres (e) None of these - -no: -[StrategyQA] Yes -Can the Great Depression be -treated with Prozac? - -or - -The Great Depression was an -economic depression. Prozac is -a drug used to treat depression. -Thus, the Great Depression could -be treated with Prozac. So the an- -swer is yes. - -Prozac is an anti-depressant -medication. The Great Depres- -sion is not a disease. Thus, -Prozac cannot treat the Great -Depression. So the answer is -no. - -The Great Depression is a time pe- -riod. Prozac is used to treat de- -pression (the mental illness). Thus, -the Great Depression can not be -treated with Prozac. So the answer -is no. - -[ARC-easy] Which source of -energy is nonrenewable? (a) -coal. -(c) water. -(d) wood. - -(b) sunlight. - -Coal, -sunlight, and water are -all renewable sources of energy. -Wood is a nonrenewable source of -energy. So the answer is (d). - -Students -[ARC-challenge] -studied the revolution of Earth -around the Sun. During which -month will Florida have the -most sunlight energy avail- -able? (a) March. (b) June. (c) -September. (d) December. +Aa -The Earth is closest to the Sun -in December, hence Florida will -have the most sunlight energy -available in December. So the an- -swer is (d). +aa -A nonrenewable source of en- -ergy is one in which the source -is not replenished in a short -amount of time. Thus, the only -option that is nonrenewable is -coal. So the answer is (a). +A -Coal is a fossil fuel and is formed -from dead organisms from the past. -Coal cannot be produced in a short -period of time, so it is considered -a nonrenewable source of energy. -So the answer is (a). +a -Earth is closest to the Sun on -June 21, the summer solstice. -So the answer is (b). + EXAMPLE -Florida is closer to the sun during -June, so it will have the most sun- -light energy during that month. So -the answer is (b). +• Dan and Kim are going +to have a baby. Kim +has dimples in her +cheeks (a dominant +trait), while Dan does +not. -Table 13: Additional examples where self-consistency helps repair the errors over greedy decode on -PaLM-540B. Two sampled reasoning paths that are consistent with the ground truth are shown. +• What are the chances +Dan and Kim’s baby +will have dimples? -A.3 FULL SETS OF PROMPTS + We know Dan and Kim’s +phenotypes (no dimples/dimples), +but what are their genotypes? -We list the full details of the prompts used for two newly-introduced datasets, AQUA-RAT (Ling -et al., 2017) and AI2 Reasoning Challenge (ARC) (Clark et al., 2018), where we manually composed -the example chain-of-thought in this paper, in Table 14 and Table 15, respectively. +• Dan has the recessive trait (no dimples) -Table 14: Few-shot exemplars for AQUA-RAT. -Q: John found that the average of 15 numbers is 40. If 10 is added to each number then the mean of the -numbers is? Answer Choices: (a) 50 (b) 45 (c) 65 (d) 78 (e) 64 +– He must have two recessive alleles +– Dan’s genotype can be represented as ‘dd’ -A: If 10 is added to each number, then the mean of the numbers also increases by 10. So the new mean -would be 50. The answer is (a). +• Kim has the dominant trait (dimples) -Q: If a / b = 3/4 and 8a + 5b = 22,then find the value of a. Answer Choices: (a) 1/2 (b) 3/2 (c) 5/2 (d) 4/2 (e) -7/2 +– But Kim could be homozygous dominant OR -A: If a / b = 3/4, then b = 4a / 3. So 8a + 5(4a / 3) = 22. This simplifies to 8a + 20a / 3 = 22, which means -44a / 3 = 22. So a is equal to 3/2. The answer is (b). +heterozygous dominant -Q: A person is traveling at 20 km/hr and reached his destiny in 2.5 hr then find the distance? Answer Choices: -(a) 53 km (b) 55 km (c) 52 km (d) 60 km (e) 50 km +– Kim’s genotype can be either ‘DD’ or ‘Dd’ +– Which one is it? -A: The distance that the person traveled would have been 20 km/hr * 2.5 hrs = 50 km. The answer is (e). + More information -Q: How many keystrokes are needed to type the numbers from 1 to 500? Answer Choices: (a) 1156 (b) 1392 -(c) 1480 (d) 1562 (e) 1788 +• What if you knew something about Kim’s -A: There are 9 one-digit numbers from 1 to 9. There are 90 two-digit numbers from 10 to 99. There are 401 -three-digit numbers from 100 to 500. 9 + 90(2) + 401(3) = 1392. The answer is (b). +parents? -19 +• How could that help? - Published as a conference paper at ICLR 2023 + Kim’s Parents -Table 15: Few-shot exemplars for ARC easy/challenge. -Q: George wants to warm his hands quickly by rubbing them. Which skin surface will produce the most -heat? (a) dry palms. (b) wet palms. (c) palms covered with oil. (d) palms covered with lotion. +• As it turns out, Kim’s father has dimples in +both cheeks, while her mother does not +• Her mother must have the recessive trait +and therefore has to have the genotype +‘dd’ -A: Dry surfaces will more likely cause more friction via rubbing than other smoother surfaces, hence dry -palms will produce the most heat. The answer is (a). +• Kim’s father has the dominant trait, but we -Q: Which factor will most likely cause a person to develop a fever? (a) a leg muscle relaxing after exercise. -(b) a bacterial population in the bloodstream. (c) several viral particles on the skin. (d) carbohydrates being -digested in the stomach. +don’t know if he is a homozygote or +heterozygote. He could be ‘DD’ or ‘Dd’ +just like Kim! -A: Option (b), bacterial population is the most likely cause for a person developing fever. The answer is (b). +• But we still know what Kim’s genotype -Q: Which change in the state of water particles causes the particles to become arranged in a fixed position? -(a) boiling. (b) melting. (c) freezing. (d) evaporating. +must be. Why? -A: When water is freezed, the particles are arranged in a fixed position; the particles are still moving for all -other options. The answer is (c). + Kim’s genotype is ‘Dd’ -Q: When a switch is used in an electrical circuit, the switch can (a) cause the charge to build. (b) increase -and decrease the voltage. (c) cause the current to change direction. (d) stop and start the flow of current. +• Kim must have a recessive allele (d), since +that is all she could have inherited from +her mother -A: The function of a switch is to start and stop the flow of a current. The answer is (d). +• Since Kim has dimples, we know she -Table 16: Few-shot exemplars for HotpotQA (closed-book setting). +inherited a dominant allele (D) from her +father +– It doesn’t matter if Kim’s father is DD or Dd; +whichever it is, he passed on a ‘D’ to his +daughter -Q: Which magazine was started first Arthur’s Magazine or First for Women? + What are the chances Kim and Dan’s baby +will have dimples? -A: Arthur’s Magazine started in 1844. First for Women started in 1989. So Arthur’s Magazine was started first. -The answer is Arthur’s Magazine. +Kim -Q: The Oberoi family is part of a hotel company that has a head office in what city? +d -A: The Oberoi family is part of the hotel company called The Oberoi Group. The Oberoi Group has its head -office in Delhi. The answer is Delhi. +D -Q: What nationality was James Henry Miller’s wife? +Dan -A: James Henry Miller’s wife is June Miller. June Miller is an American. The answer is American. +d -Q: The Dutch-Belgian television series that "House of Anubis" was based on first aired in what year? +d -A: "House of Anubis" is based on the Dutch–Belgian television series Het Huis Anubis. Het Huis Anubis is first -aired in September 2006. The answer is 2006. + Kim -As additional information, we also list the exact set of prompts used for all arithmetic reasoning tasks -in Table 17, since there are multiple sets of prompts introduced in Wei et al. (2022). The prompts for -CommonsenseQA and StrategyQA are the same as used in Wei et al. (2022). +d -We provide the exact prompts used for common NLP tasks in the following tables as well, including -NLI (Table 18, Table 19, Table 20) and Closed-Book Question-Answering tasks (Table 16, Table 21). +D -20 +Dd - Published as a conference paper at ICLR 2023 +dd -Table 17: Few-shot exemplars for all arithmetic reasoning tasks, from Wei et al. (2022). -Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done, -there will be 21 trees. How many trees did the grove workers plant today? +Dd -A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted. -So, they must have planted 21 - 15 = 6 trees. The answer is 6. +dd -Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot? +Dan -A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5. +d -Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total? +d -A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74 -chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39. +50% chance the baby will have the genotype ‘Dd’ and have +dimples -Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops -did Jason give to Denny? +50% chance the baby will have the genotype ‘dd’ and not have +dimples -A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of -lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8. + Clinical Application -Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does -he have now? +P -A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so -in total he has 7 + 2 = 9 toys. The answer is 9. + Clinical Application continued . . . -Q: There were nine computers in the server room. Five more computers were installed each day, from -monday to thursday. How many computers are now in the server room? +P -A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 = -20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers. -The answer is 29. + Other things may change us, but we start and +end with family -Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many -golf balls did he have at the end of wednesday? +-Anthony Brandt -A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On -Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33. +If you look deeply into the palm of your hand, +you will see your parents and all generations of +your ancestors. All of them are alive in this +moment. Each is present in your body. You are +the continuation of each of these people. -Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left? +-Thich -A: She bought 5 bagels for $3 each. This means she spent 5 * $3 = $15 on the bagels. She had $23 in -beginning, so now she has $23 - $15 = $8. The answer is 8. +Nhat -21 +Hanh - Published as a conference paper at ICLR 2023 + Questions? -Table 18: Few-shot exemplars for ANLI. + Adapted from a Presentation Created by the University +of North Carolina DNA Day Program -Premise: -"Conceptually cream skimming has two basic dimensions - product and geography." -Based on this premise, can we conclude the hypothesis "Product and geography are what make cream skimming -work." is true? -OPTIONS: -- yes -- no -- it is not possible to tell +• Sponsors -A: Based on "cream skimming has two basic dimensions" we can’t infer that these two dimensions are what -make cream skimming work. The answer is it is not possible to tell. +– NC Glaxo Smith Kline Foundation +– National Human Genome Research Institute +– Sigma XI Scientific Research Society -Premise: -"One of our member will carry out your instructions minutely." -Based on this premise, can we conclude the hypothesis "A member of my team will execute your orders with -immense precision." is true? -OPTIONS: -- yes -- no -- it is not possible to tell +• Collaborative Partners -A: "one of" means the same as "a member of", "carry out" means the same as "execute", and "minutely" means -the same as "immense precision". The answer is yes. - -Premise: -"Fun for adults and children." -Based on this premise, can we conclude the hypothesis "Fun for only children." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "adults and children" contradicts "only children". The answer is no. - -Premise: -"He turned and smiled at Vrenna." -Based on this premise, can we conclude the hypothesis "He smiled at Vrenna who was walking slowly behind -him with her mother." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: the premise does not say anything about "Vrenna was walking". The answer is it is not possible to tell. - -Premise: -"well you see that on television also" -Based on this premise, can we conclude the hypothesis "You can see that on television, as well." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "also" and "as well" mean the same thing. The answer is yes. - -Premise: -"Vrenna and I both fought him and he nearly took us." -Based on this premise, can we conclude the hypothesis "Neither Vrenna nor myself have ever fought him." is -true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "Vrenna and I both" contradicts "neither Vrenna nor myself". The answer is no. - -22 - - Published as a conference paper at ICLR 2023 - -Table 19: Few-shot exemplars for e-SNLI (the rationales are crowd-sourced annotations from the -original dataset). - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is training his horse for a competition." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: The person is not necessarily training his horse. The answer is it is not possible to tell. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is at a diner, ordering an omelette." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: One jumping horse cannot be in a diner ordering food. The answer is no. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is outdoors, on a horse." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: A broken down airplane is outdoors. The answer is yes. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "They are smiling at their parents." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: Just because they are smiling and waving at a camera does not imply their parents or anyone is anyone behind -it. The answer is it is not possible to tell. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "The kids are frowning." is true? OPTIONS: -- yes -- no -- it is not possible to tell -A: One cannot be smiling and frowning at the same time. The answer is no. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "There are children present." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A:The children must be present to see them smiling and waving. The answer is yes. - -23 - - Published as a conference paper at ICLR 2023 - -Table 20: Few-shot exemplars for RTE. - -Premise: -"No Weapons of Mass Destruction Found in Iraq Yet." -Based on this premise, can we conclude the hypothesis "Weapons of Mass Destruction Found in Iraq." is true? - -A: "No Weapons of Mass Destruction Found" contradicts "Weapons of Mass Destruction Found". The answer is -no. - -Premise: -"A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful -gathered in downtown Chicago to mark the installation of new Pope Benedict XVI." -Based on this premise, can we conclude the hypothesis "Pope Benedict XVI is the new leader of the Roman -Catholic Church." is true?’ - -A: "installation of new Pope Benedict XVI." means "Pope Benedict XVI is the new leader". The answer is yes. - -Premise: -"A man is due in court later charged with the murder 26 years ago of a teenager whose case was the first to -be featured on BBC One’s Crimewatch. Colette Aram, 16, was walking to her boyfriend’s house in Keyworth, -Nottinghamshire, on 30 October 1983 when she disappeared. Her body was later found in a field close to her -home. Paul Stewart Hutchinson, 50, has been charged with murder and is due before Nottingham magistrates -later." -Based on this premise, can we conclude the hypothesis "Paul Stewart Hutchinson is accused of having stabbed a -girl." is true? - -A: The premise does not say Paul Stewart Hutchinson "stabbed" this girl. The answer is no. - -Premise: -"Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it -will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients." -Based on this premise, can we conclude the hypothesis "Herceptin can be used to treat breast cancer." is true? +– North Carolina Department of Public Instruction +– Destiny Traveling Science Learning Program +– North Carolina Association for Biomedical Research +– National Human Genome Research Institute +– American Society for Human Genetics -A: "Herceptin was approved to treat breast cancer" implies that "Herceptin can be used to treat breast cancer". -The answer is yes. - -Table 21: Few-shot exemplars for BoolQ (closed-book setting). - -Q: does system of a down have 2 singers? - -A: System of a Down currently consists of Serj Tankian, Daron Malakian, Shavo Odadjian and John Dolmayan. -Serj and Daron do vocals, so the band does have two singers. The answer is yes. - -Q: do iran and afghanistan speak the same language? - -A: Iran and Afghanistan both speak the Indo-European language Persian. The answer is yes. - -Q: is a cello and a bass the same thing? - -A: The cello is played sitting down with the instrument between the knees, whereas the double bass is played -standing or sitting on a stool. The answer is no. - -Q: can you use oyster card at epsom station? - -A: Epsom railway station serves the town of Epsom in Surrey and is not in the London Oyster card zone. The -answer is no. +• North Carolina Educators -24 +– Amy Bradley, Hibriten High School; Cindy Byron, School of Inquiry and Life +Sciences at Asheville (SILSA); Anita Crowley, Lee County High School; +MaryAnne Gore, West Brunswick High School; Terry Howerton, The School of +Biotechnology at Atkins; Michael Kendall, East Bladen High School; Crystal +McDowell; Letitia Myles, E.E. Smith High School; Elizabeth Pressley, Bartlett +Yancey High School; Steven Ross, Harnett Central High School; Murphey +Wellman, North Iredell High School; Anita Willington, East Bladen High School; +Linda Woody, Reidsville High School diff --git a/q144/random_k4/question.json b/q144/random_k4/question.json index 618e9983c82421e452959ff1a60f216217e35114..4051456de7c365b29af8c57e0effa5f31cc5e7c0 100644 --- a/q144/random_k4/question.json +++ b/q144/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "germanwingsdigitalcrisisanalysis-150403064828-conversion-gate01_95.pdf", - "3363cf211564d0408141652e93e983712c4cc6de.pdf", - "web_ff8dcb0ea2a037b7.pdf", - "web_4ae8d7cc08fa2f24.pdf", - "web_365d195edde4e8ba.pdf" + "web_5050add9b28048e4.pdf", + "web_5e409f22555df076.pdf", + "web_8f3d7e741a1d660a.pdf", + "resize_r2_10.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q144/random_k4/random_1.md b/q144/random_k4/random_1.md index 023d319f13d30cc1b5cfcd5e613a62c138b007a8..b66c1815f8dc909ab75e16a08e2512a86aa07ad4 100644 --- a/q144/random_k4/random_1.md +++ b/q144/random_k4/random_1.md @@ -1,1453 +1,655 @@ -Holistic Optimal Label Selection for Robust -Prompt Learning under Partial Labels -Yaqi Zhao, Haoliang Sun, Yating Wang, Yongshun Gong, and Yilong Yin -Shandong University, Jinan, China -Abstract. Promptlearninghasgainedsignificantattentionasaparameter- -efficient approach for adapting large pre-trained vision-language models -todownstreamtasks.However,whenonlypartiallabelsareavailable,its -performanceisoftenlimitedbylabelambiguityandinsufficientsupervi- -soryinformation.Toaddressthisissue,weproposeHolistic Optimal La- -belSelection (HopS),leveragingthegeneralizationabilityofpre-trained -featureencodersthroughtwocomplementarystrategies.First,wedesign -a local density-based filter that selects the top frequent labels from the -nearestneighbors’candidatesetsandusesthesoftmaxscorestoidentify -the most plausible label, capturing structural regularities in the feature -space.Second,weintroduceaglobal selectionobjectivebasedonoptimal -transportthatmapstheuniformsamplingdistributiontothecandidate -label distributions across a batch. By minimizing the expected trans- -portcost,itcandeterminethemostlikelylabelassignments.Thesetwo -strategies work together to provide robust label selection from both lo- -cal and global perspectives. Extensive experiments on eight benchmark -datasets show that HopS consistently improves performance under par- -tialsupervisionandoutperformsallbaselines.Thoseresultshighlightthe -merit of holistic label selection and offer a practical solution for prompt -learning in weakly supervised settings. -Keywords: Partiallabellearning·Promptlearning·Optimaltransport -1 Introduction -Largepre-trainedvision-languagemodels(VLMs),suchasCLIP[26],havedemon- -strated remarkable capabilities across a wide range of downstream tasks [8,19, -27,29]. Among various fine-tuning paradigms [18], prompt tuning has emerged -as an efficient alternative to full model fine-tuning, offering the advantage of -adapting large models with minimal additional parameters [50]. By optimizing -a small set of learnable prompts while keeping the backbone frozen, prompt -learning preserves the generalization strength of the pre-trained model and re- -duces training costs—making it particularly attractive in resource-constrained -or few-shot learning scenarios [47]. -The effectiveness of prompt learning can be significantly compromised in -weakly supervised settings [48], particularly those involving partial labels [20], -whereonlyasubsetofcandidatelabelsisprovidedperinstancewithoutexplicit -6202 -rpA -8 -]VC.sc[ -1v41660.4062:viXra - -| 2 Y. | Zhao | et al. | | | | | | | -| ---- | ---- | ------ | --- | --- | --- | --- | --- | --- | -ground-truthannotations.Thissettingiscommoninreal-worldapplicationssuch -as webly supervised learning [25], human-in-the-loop labeling [39], and open- -world recognition [38]. The main challenge in such scenarios lies in the label -ambiguity, which hampers supervision and often degenerate the generalization -performanceoflearningalgorithms,especiallyforpromptlearningwithfew-shot -| instances | [51]. | | | | | | | | -| ---------- | ----- | -------------- | --- | ------- | -------------- | ------ | ---------------- | --- | -| To address | | the challenges | of | partial | label learning | (PLL), | state-of-the-art | | -methods typically select a plausible label from the candidate set using repre- -sentationslearnedthroughcontrastivelearning(e.g.,PICO[33]).However,such -approachesarenotdirectlyapplicabletopre-trainedmodels,asfine-tuningtheir -vision encoders may compromise their zero-shot capabilities [15]. This presents -a fundamental challenge in filling the gap between reliable label selection with -frozen vision encoders and effective prompt learning with label disambiguation. -| To overcome | | this limitation, | | we propose | a holistic | label | selection | strategy | -| ----------- | --- | ---------------- | --- | ---------- | ---------- | ----- | --------- | -------- | -that fully leverages the representational generalization of pre-trained encoders -for robust prompt learning. Specifically, HopS includes two complementary se- -lection mechanisms. The first is a Local Density-based Filter (LDF), which -estimates label frequency within a k-nearest neighbor (k-NN) structure in the -image feature space and selects the most frequent labels in the neighborhood -for a subset, capturing local semantic regularities in a non-parametric manner. -LDF effectively exploits the zero-shot capabilities of the encoder and avoids -overfitting to noisy labels during the early training stages. Among the subset, -the most plausible label is then identified based on the softmax scores. The -second is a Global Optimal Transport Planner (GOP). It maps a uniform -global prior—representing the overall underlying distribution—to the candidate -label distributions across a mini-batch. By minimizing the expected transport -cost, GOP encourages globally optimal label assignments. The resulting trans- -portplanexplicitlycharacterizeshoweachinstancecontributestotheclass-wise -probability mass in the label distribution, thereby providing a principled and -| interpretable | foundation | for | identifying | and | selecting | the | most relevant | label. | -| -------------- | ---------- | ---------- | ----------- | ------ | ------------- | --- | ------------- | ------- | -| By integrating | | both local | and | global | perspectives, | our | framework | enables | -more accurate and stable label selection, mitigating the impact of label ambi- -guity. We evaluate our method across a variety of vision-language benchmarks -under partial supervision and demonstrate consistent improvements over strong -| baselines. | Our | key contributions | | are as follows: | | | | | -| ---------- | --- | ----------------- | --- | --------------- | --- | --- | --- | --- | -– We provide a comprehensive investigation into leveraging the representa- -| tional | generalization | of | pre-trained | VLMs | for | prompt | learning in | the partial | -| ----------- | -------------- | -------- | ----------- | ---- | --- | ------ | ----------- | ----------- | -| supervision | | setting. | | | | | | | -– We propose a novel holistic label selection framework that combines a local -| density-based | | filter with | a global | optimal | transport | planner. | | | -| ------------- | --- | ----------- | -------- | ------- | --------- | -------- | --- | --- | -– Weempiricallyvalidatetheeffectivenessofourapproach,achievingstate-of- -| the-art | performance | on | multiple | benchmarks. | | | | | -| ------- | ----------- | --- | -------- | ----------- | --- | --- | --- | --- | - -HopS 3 -2 Related Work -2.1 Partial Label Learning -PLL [1,3,13,46] assumes that each training instance is associated with a can- -didate label set, within which only one label is correct but not explicitly spec- -ified. To address the ambiguity introduced by these noisy candidates, subse- -quentresearchhasproposedvariousstrategies,includingconsistencyregulariza- -tion[7,21,40]andcontrastivelearning[33].Thesemethodsaimtoleveragepartial -supervisiontolearnmorediscriminativerepresentationsandimprovegeneraliza- -tion. Expanding on this direction, [12] introduced a dissimilarity propagation- -guided label shrinkage method to refine candidate label sets by eliminating ir- -relevant labels, thereby enhancing supervision quality. Meanwhile, works such -as [44,45] have begun to explore the more challenging instance-dependent PLL -setting, where candidate label sets are generated based on the characteristics -of each instance. More recently, [42] and [31] have continued to leverage can- -didate labels to learn robust representations. Additionally, [20] highlighted the -critical role of feature representations and label denoising in effective PLL. A -related work, SoLar [34], addresses partial label learning under class imbalance -by employing OT to align the estimated long-tailed class prior with a uniform -distribution. In contrast, our method leverages OT to align the overall underly- -inglabeldistributionwiththecandidatelabeldistributions.Furthermore,unlike -conventionalPLLapproaches,ourworkexplorespromptlearningforpre-trained -VLMs, which leverage powerful vision encoders with strong zero-shot capabili- -ties, rather than training encoders from scratch. -2.2 Prompt Learning -Prompt learning has emerged as a parameter-efficient paradigm for adapting -large-scale pre-trained models to downstream tasks [9]. Rather than fine-tuning -theentiremodel,itintroducesauxiliaryinputprompts—eithermanuallycrafted -templates or learnable continuous embeddings—to guide model predictions. In -thevision-languagedomain,CoOp[50]firstdemonstratedthatlearnablecontext -vectors could serve as effective prompts for adapting CLIP-like models, high- -lighting their strong transferability. Since then, several studies have enhanced -the robustness of prompt learning through strategies such as aligning sample -representations in multimodal contexts [32,41], employing mixture-of-expert -prompts[35],andutilizingunsupervisedpromptdistillation[14,17].Inaddition, -prompt learning has achieved significant improvements on various downstream -tasks,includingtext-to-imagegeneration[30]andopen-vocabularysemanticseg- -mentation [16]. While these advances have shown promise in fully supervised -settings, their effectiveness under partial supervision remains unexplored. To -address this gap, we extend robust prompt tuning to the PLL scenario, aiming -tobridgeweaksupervisionwithprompt-basedadaptationinpre-trainedVLMs. - -| 4 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -3 Preliminary -| 3.1 Candidate | | Label | Sets | in PLL | | | | | | -| ------------- | --- | ----- | ---- | ------ | --- | --- | --- | --- | --- | -We define the training dataset as D = {(x ,S )}n , where x ∈ X is an input -| | | | | | | i i | i=1 | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -instance and S i ⊆{1,2,...,C} is the candidate label set for a C-category clas- -sificationtask.Here,allelements{s ,...,s ...,s }inS containoneground- -| | | | | | i1 | il | iL | i | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -truth label and the rest L−1 elements are the confused labels, involving false -positive labels for the current instance. Different levels of label confusion is cor- -| responding | to | the number | of | confused | labels | in each | candidate | set. | | -| ---------- | --- | ------------ | --- | -------- | ------ | --------- | --------- | ---- | --- | -| 3.2 Prompt | | Optimization | | with | The | Candidate | | Set | | -Prompt optimization (PO) in CLIP replaces manual prompt engineering by -learning continuous context vectors in an end-to-end manner, while keeping -the pre-trained model parameters frozen to fully leverage the knowledge en- -codedwithinthem.LetV,T andtdenotethevisionencoder,textencoder,and -learnable prompt parameters, respectively. p(s | x ;t) denotes the predicted -| | | | | | | ij | i | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -probabilityofthej-thlabelwithinthecandidatesetS ,computedviathecosine -i -similarity between the visual and textual embeddings. These embeddings are -derivedfromthevisualencoderV,andthetextencoderTwiththeinputofthe -prompt vector t and the class name embedding v . For PO under partial-label -j -supervision,thestandardcross-entropylossiscomputedoverthecandidatelabel -| set S as | follows: | | | | | | | | | -| -------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -i -C -(cid:88) -| | | L | (x ,S ;t)= | | −s | log(p(s | |x | ;t)) | | -| --- | --- | --- | ---------- | --- | --- | ------- | --- | ---- | --- | -| | | CE | i i | | i,j | | ij | i | | -| | | | | j=1 | | | | | (1) | -(cid:88) -| | | s.t. | | s =1, | ands | =0 | for j | ∈/ S . | | -| --- | --- | ---- | --- | ----- | ---- | --- | ----- | ------ | --- | -| | | | | i,j | | i,j | | i | | -j∈Si -| The prediction | | probability | of | the input | x | is given | by: | | | -| -------------- | --- | ----------- | --- | --------- | --- | -------- | --- | --- | --- | -i -| | | | | exp(cos(V(x | | ),T(t,v | | ))) | | -| --- | --- | ------ | ---- | ----------- | ----------- | ------- | ------- | --- | --- | -| | | | | | | i | | j | | -| | | p(s |x | ;t)= | | | | | . | (2) | -| | | ij | i | (cid:80)C | exp(cos(V(x | | ),T(t,v | ))) | | -| | | | | | | | i | q | | -q=1 -| Despite | the | expressive | power | of | PO, | the inherent | | label ambiguity | within | -| ------- | --- | ---------- | ----- | --- | --- | ------------ | --- | --------------- | ------ | -each candidate set S creates a fundamental gap between partial label learning -i -andfullysupervisedlearning.Thisintroducestwocorechallenges:(1)effectively -disambiguating the candidate labels, and (2) accurately identifying the ground- -| truth label | for | each instance. | | | | | | | | -| ----------- | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | -4 Methodology -To identify the ground-truth label and guide the model in learning effective -promptvectors,weproposeaholisticoptimallabelselectionstrategyforprompt - -| | | | | | | | | HopS | 5 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -learning in VLMs. HopS integrates both local and global perspectives through -twoelaboratedcomponents:alocaldensity-basedfilter,whichselectsthe“local- -consensus” candidate, and a global optimal transport planner, which identifies -the “global-harmony” candidate. These components work in concert to robust -and complementary label guidance for effective prompt optimization under par- -tial supervision. -| 4.1 Local | Density-Based | | Filter | | | | | | | -| --------- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | -The local density of an instance in the image feature space reflects the intrinsic -semantic regularities associated with its candidate labels. To leverage this den- -sity information for label selection, we identify the k-nearest neighbors of each -,Sˆ -instance (x ,S ) in the image feature space, denoted as {(x )}k . Specifi- -| | i i | | | | | | j j | j=1 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -cally, we build an affinity matrix A by computing the cosine similarity between -imagefeaturesofallinstances,wherethefeaturesareextractedusingthefrozen -visionencoderofCLIP.BasedonA,weretrievethetop-kmostsimilarexamples -foreachinstance,therebycapturinglocalsemanticstructure.Oncethek-nearest -neighborsareselected,wethenconstructamultiset-unioncandidatesetforeach -instance as: -| | | | |   | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -k -| | | | | (cid:93) Sˆ | (cid:93) | | | | | -| --- | --- | --- | --- | ----------- | -------- | --- | --- | --- | --- | -| | | | N i | = j | S i | , | | | (3) | -j=1 -where⊎denotesmultisetunion,preservinglabelmultiplicities.Itisworthnoting -that,sincetheaffinitymatrixispre-computedpriortotraining,theretrievalstep -foreachinstancereducestoaneighborsearchwithtimecomplexityO(n·logk). -Given the small number n of training samples in prompt learning, the retrieval -cost is negligible. -| Next, | we compute | the | frequency | of each | category | c in | N i . Let | f(c) | denote | -| ------------ | ---------- | --- | --------- | ----------------- | -------- | ----------- | --------- | ---- | ------ | -| the relative | frequency | of | element | c in the multiset | | N , defined | as: | | | -i -| | | | | |{l|s | =c}| | | | | | -| --- | --- | --- | ----- | ----- | ---- | --- | --- | --- | --- | -| | | | f(c)= | jl | , | | | | (4) | -|N | -i -where |{l | s = c}| counts the number of times element c appears in N , and -| | jl | | | | | | | | i | -| ----------- | ------------ | --- | ----------- | ------ | --------- | --------- | ----------- | --- | --- | -| |N | is the | total number | | of elements | in the | multiset, | including | duplicates. | | | -i -| To enforce | label | consistency | | and suppress | unreliable | candidates, | | we | retain | -| ---------- | ----- | ----------- | --- | ------------ | ---------- | ----------- | --- | --- | ------ | -categories whose frequency exceeds a predefined threshold τ ∈ [0,1], thereby -| forming | a consensus-aware | | mask | set for each | instance | as: | | | | -| ------- | ----------------- | --- | ------- | ------------ | -------- | --- | --- | --- | --- | -| | | | M ={c∈N | |f(c)≥τ}. | | | | | (5) | -| | | | i | i | | | | | | -TherefinedcandidatesetisthenobtainedbyintersectingthemasksetM with -i -the original candidate set S , i.e., M ∩S , ensuring that only labels supported -| | | | i | i i | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -byboththeinstanceanditsneighborsareretained.Topreventdegeneratecases -whereallcandidatelabelsareeliminated,wereverttotheoriginalcandidateset -| S if the | refined set | becomes | empty. | | | | | | | -| -------- | ----------- | ------- | ------ | --- | --- | --- | --- | --- | --- | -i - -| 6 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | --- | --- | --- | --- | -Fig.1: Illustration of the transport cost matrix and the resulting transport plan be- -tween the uniform source instance distribution (Source) and the target candi- -(Target). -| date label | distribution | | | | | | -| ---------- | ------------ | --------- | --------------- | ------------------ | ---------- | ---- | -| Finally, | the most | plausible | candidate label | ylocal is selected | as the one | with | -thehighestpredictionprobabilityintherefinedcandidateset,asdeterminedby -Eq. (2). -| 4.2 Global | Optimal | Transport | Planner | | | | -| ---------- | ------- | --------- | ------- | --- | --- | --- | -Tocomplementthelocalselection,weintroduceagloballyconsistentlabelselec- -tionmechanismviaoptimaltransport(OT).Thisformulationmatchesinstances -with their candidate labels in a way that aligns with both a uniform prior over -instancesandtheempiricallabeldistributionwithinthebatch,selectingforeach -instance the label that receives the highest transport mass. As shown in Fig. 1, -the matrix on the left illustrates the candidate label sets for four instances in a -seven-classclassificationtask,withlightercolorsindicatinglowertransportcosts -and, consequently, higher label credibility. The right matrix shows the resulting -optimal transport plan under the given cost constraints. Here, darker colors in- -dicategreatertransportedmass,reflectingtheplanner’spreferenceforallocating -probabilitymasstomorecredibleclasses.Thetransportprocessstrictlyadheres -to the marginal constraints, ensuring the consistency of source and target dis- -| tributions | before and | after transport. | | | | | -| ---------- | ---------- | ---------------- | --- | --- | --- | --- | -Discrete Transport Formulation. Given a batch of B samples, we define a -uniform source distribution over instences, where r ∈ ∆B is a B-dimensional -probabilitysimplex.Acandidate-awaremarginaldistributionisestimatedbased -on candidate frequencies, denoting as c ∈ ∆C. The element in r and c are -| computed | as: | | | | | | -| -------- | --- | --- | --- | --- | --- | --- | +1 + +The Nursing Home +Inspection Process + +SUMMARY + +Both the Minnesota Department of Health (MDH) and the U.S. +Department of Health and Human Services share responsibility for +ensuring that Minnesota’s nursing homes provide an acceptable level +of care for their residents. Because the federal government dictates +the overall structure and content of the inspection program, the State +of Minnesota has few opportunities to make significant changes in +how it conducts nursing home inspections. The federal government +mandates how often the state must inspect nursing homes, the steps +the state must follow when conducting inspections, and the standards +the state must apply. Although MDH and other states have asked the +federal government for more flexibility in conducting inspections, the +federal government has not issued any waivers that allow states to +significantly change or implement an alternative inspection program. + +The current nursing home inspection process emerged in the mid-1980s, as + +Congress responded to reports of resident abuse and inadequate enforcement + +of nursing home regulations. In a 1986 report on nursing home quality, the +Institute of Medicine found “serious, even shocking inadequacies” in the +enforcement of regulations.1 As a result of this report and the efforts of advocacy +groups and professional organizations, Congress passed a major reform of nursing +home regulation as part of the Omnibus Budget Reconciliation Act of 1987.2 + +Since that time, Congress and the U.S. Department of Health and Human Services +have periodically modified inspection requirements in response to studies that +have shown continued weak and inconsistent enforcement of nursing home +regulations and quality of care problems. Most significantly, the Nursing Home +Oversight Improvement Program was implemented in 1998, which, among other +things, enhanced federal review of state inspections and required the federal +government to terminate funding for states that fail to conduct adequate +inspections. + +This chapter addresses the following question about how the Minnesota +Department of Health (MDH) inspects nursing homes: + +• What are the respective roles of the Minnesota Department of Health +and U.S. Department of Health and Human Services in conducting +nursing home inspections? + +Institute of Medicine, Committee on Nursing Home Regulation, Improving the Quality of Care + +1 +in Nursing Homes (Washington DC: National Academy Press, 1986), 146. +2 Pub. L. 100-203, Dec. 22, 1987. + + 4 + +NURSING HOME INSPECTIONS + +The federal +government and +states share +responsibility +for ensuring that +nursing homes +provide an +acceptable level +of care to +residents. + +To answer this question, we examined state and federal laws, rules, regulations, +and guidelines related to nursing homes inspections, as well as a wide variety of +research reports by state and federal agencies. We also interviewed state +policymakers, nursing home inspectors and their supervisors, and a sample of +nursing home administrators from throughout the state. + +FEDERAL REGULATION OF INSPECTIONS + +State and federal laws define a nursing home as a facility (or that part of a facility) +that provides health evaluation and treatment services to five or more residents +who do not need an acute care facility (such as a hospital) but who require nursing +supervision or rehabilitation services on an inpatient basis.3 In lay terms, this +means a facility that provides a room, meals, recreational opportunities, and help +with daily living activities such as dressing, eating, bathing, walking, and using +the bathroom. Residents generally have health problems that keep them from +living on their own and may require daily medical attention. + +The federal +government and +states share +responsibility for +ensuring that +nursing homes +provide an +acceptable level of +care to residents. +The Centers for +Medicare and +Medicaid Services +(CMS) in the U. S. +Department of +Health and Human +Services oversees +the inspection +program for +nursing homes that +participate in the +federal Medicare and Medicaid programs.4 The agency sets nursing home +standards; interprets federal regulations, guidelines, and polices; and establishes +and monitors inspection procedures. It contracts with MDH to conduct nursing +home inspections in Minnesota.5 In addition to conducting inspections, MDH +licenses nursing homes for state purposes and certifies their eligibility for +participation in the Medicare and Medicaid programs. Finally, the department is + +Nursing home residents generally need help with many activities of +daily living. + +3 Minn. Stat. (2004), §144A.01, subd. 5-6; and 42 U.S. Code, §1396r, (a) (2000). +4 CMS was formerly called the Health Care Financing Administration. +5 The Minnesota Department of Health contracts with the State Fire Marshall’s Office in the +Minnesota Department of Public Safety to determine facility compliance with the federal Life +Safety Code, which is necessary for participation in the Medicare and Medicaid programs. State +Fire Marshall findings are included in the inspection reports issued by MDH. + + THE NURSING HOME INSPECTION PROCESS + +5 + +responsible for explaining program participation requirements to providers to help +them comply with federal requirements.6 + +Overall, we found that: + +• The federal government sets forth the overall structure and content of +the nursing home inspection program, and Minnesota has very few +opportunities to make significant changes in the program. + +Federal regulations outline both the general parameters of the inspection process +as well as the specifics of how each inspection must be done. They dictate: +(1) how frequently the state must inspect nursing homes, (2) the steps the state +must follow when conducting inspections, and (3) the standards that the state +must apply. We discuss each of these areas in greater detail below. + +Inspection Frequency + +The federal government sets forth how often nursing homes must be inspected: + +Federal law and regulations require that the Minnesota Department of +Health inspect nursing homes every 12 months, on average. + +All nursing facilities must be inspected no later than once every 15 months, with +an average time statewide between inspections of 12 months. Federal regulations +do not allow states to inspect nursing homes with “good” inspection records less +frequently than homes with “bad” records. In addition, CMS requires that at least +10 percent of inspections be “staggered” (started outside of normal business +hours). To meet this requirement, the state must begin some inspections on +weekends or holidays, some in the early morning (before 8:00 AM), and some in +the evening (after 6:00 PM). Furthermore, the federal government requires that +all nursing home inspections be unannounced. + +About 420 Minnesota nursing homes participated in the Medicare and Medicaid +programs during federal fiscal year 2003.7 The department inspected all of these +nursing homes within 14.7 months of their prior inspection, with an average time +between inspections of 12 months.8 In addition, 12 percent of the 403 inspections +conducted were staggered, with 21 inspections beginning before 8:00 AM, 16 +inspections after 6:00 PM, and 10 beginning on a weekend or holiday.9 + +For the most part, nursing home providers, state policymakers, and nursing home +inspectors generally agree that requiring annual inspections of all nursing homes + +6 The department has additional responsibilities related to nursing homes, such as investigating +complaints, which were outside the scope of our evaluation. Nursing home inspectors also inspect +other types of health care facilities, such as hospitals and intermediate care facilities for the mentally +retarded. These activities were likewise outside the scope of our evaluation. +7 The federal fiscal year runs from October 1 through September 30. +8 Centers for Medicare and Medicaid Services, Federal Fiscal Year 2003 State Performance +Standard Review Report (Washington, DC, March 15, 2004), 1. +9 +Ibid., 2. Because nursing homes may go up to 15 months between inspections, the number of +nursing homes that MDH inspected during federal fiscal year 2003 was less than the total number of +nursing homes in the state. + +Nursing homes +must have a +“surprise” +inspection no +later than once +every 15 months. + +(cid:127) + 6 + +NURSING HOME INSPECTIONS + +Because of +federal +requirements, +MDH cannot +inspect nursing +homes with +“good” +inspection +records less +frequently than +those with "bad" +records. + +is, at times, an inefficient use of staff resources. The requirement does not permit +the state to focus efforts on the nursing homes that need oversight the most. To +help increase the efficiency and effectiveness of the inspection process, the +Legislature has repeatedly required the Commissioner of Health to seek federal +permission to implement an alternative inspection process that would change how +often nursing homes must be inspected.10 In response, the department submitted a +proposal to CMS that would have increased the time between “full” inspections +up to 30 months for some homes with “good” compliance records. Other states +have proposed similar approaches, including ones to conduct abbreviated annual +inspections for homes with “good” compliance records. + +To date, CMS has not approved an alternative inspection program put forth by any +state, including Minnesota. According to CMS, the social security law does not +allow states to obtain waivers to implement an alternative inspection program for +nursing homes participating in the Medicare program, although states could +implement an alternative inspection program for homes that only participate in the +Medicaid program. However, because this would involve only a few nursing +homes, it is generally not feasible for states to do so. + +For the last several years, CMS has been studying the feasibility of an alternative +inspection process. Recently, the agency announced that it would be establishing +a few pilot sites around the country to implement a “revamped” inspection +process. Designed to address concerns about inspection consistency and +efficiency, pilot sites will make greater use of computers to make initial +determinations of deficiencies rather than relying on the judgment of inspection +teams. The alternative process will not result in less frequent inspections for +facilities, but may allow inspectors to spend somewhat less time in “good” +facilities and more time in “bad” ones. + +Inspection Steps + +In addition to requiring an inspection no later than once every 15 months: + +• + +Federal regulations require that each nursing home’s annual +inspection be a “standard” or full inspection consisting of seven +federally mandated steps. + +Federal regulations do not allow states to do shorter or abbreviated inspections of +nursing homes with “good” records of compliance or to cut short an inspection +when inspectors do not detect any problems in a facility. On the other hand, state +inspectors must extend the inspection if they suspect that a facility is providing +substandard care to its residents. + +As shown in Table 1.1, the standard inspection consists of seven federally +mandated steps. First, inspectors prepare off-site by reviewing information about +the nursing home and its residents to help identify areas of concern. Immediately +upon arriving at the facility, the inspection team meets with the nursing home +administrator to explain the inspection process and request specific information; + +10 Laws of Minnesota (2000), ch. 312, sec. 2, 5; Laws of Minnesota (1Sp2001), ch. 9, art. 5, sec. 38; +Laws of Minnesota (2002), ch. 379, art. 1, sec. 113; and Laws of Minnesota (2004), ch. 247, +sec. 6. + + THE NURSING HOME INSPECTION PROCESS + +7 + +Table 1.1: The Federal Nursing Home Inspection +Process + +Step 1: Off-site preparation + +Step 2: Entry conference and on-site preparation + +Step 3: + +Initial nursing home tour + +Step 4: Resident sample selection + +Step 5: + +Information gathering + +A. General observation of the facility +B. Kitchen/food service observation +C. Resident review +D. Quality of life assessment +E. Medication pass +F. Quality assessment and assurance review +G. Abuse prevention review + +Step 6: Deficiency determination + +A. Determination of substandard quality of care + +Step 7: Exit conference + +Likewise, +MDH cannot +do abbreviated +inspections in +nursing homes +with "good" +records. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), ch. 7, sec 7200. + +this is followed by a facility tour. Using the information provided by the facility +and what inspectors learned during the tour, the team then selects a sample of +residents to focus on during the information-gathering portion of the inspection. +During this phase, the team meets on a daily basis to compare notes, discuss new +areas of concern, and make adjustments to the inspection as deemed necessary. +Inspectors observe the care and services that facility staff provide to residents, +such as preparing and serving meals, administering medications, and helping, as +necessary, with activities such as bathing, toileting, walking, and grooming. +Inspection team members also interview residents and staff and review resident +records. Once the team is satisfied that they have gathered enough information, it +meets to determine whether the facility has failed to meet any regulatory +requirements. The team prepares a draft inspection report that discusses each +violation of federal regulations (commonly referred to as a deficiency) that the +team has identified, and then meets with nursing home personnel and interested +residents and family members to present its preliminary list of deficiencies. + +After the inspection team leaves the facility, it finalizes the “Statement of +Deficiencies” and submits it to the team’s district supervisor who is responsible +for reviewing the document and submitting a final copy to the facility and CMS. +The facility must submit a “Plan of Correction” within ten days that indicates how +and when it will correct each of the deficiencies that it has received.11 Inspectors +normally conduct an unannounced revisit to verify that the plan of correction has +been implemented and that the deficiencies no longer exist. For the most part, +MDH generally gives a facility 40 days from the end of the inspection to correct +deficiencies before MDH imposes any sanctions on the facility. + +11 Facilities may also dispute a deficiency and request a hearing before MDH or an administrative +law judge within this ten-day period. Chapter 2 discusses how often this happens and the outcome +of such hearings. + +Inspectors spend +much of their +time observing +and talking with +residents and +staff. + + 8 + +NURSING HOME INSPECTIONS + +State law +requires +inspectors to +leave a draft +inspection report +with facilities +when they leave. + +While the state is unable to make significant changes in how inspections are done: + +• Minnesota has expanded the federal nursing home inspection process + +in several ways. + +The state goes beyond federal inspection requirements by adding other tasks, +including requirements to: (1) interview family council members; (2) expand the +number of evening observations nursing home inspectors must make each month; +(3) conduct a “verify and clarify” session with the provider to discuss possible +areas of concern prior to the exit conference; and (4) leave a draft inspection +report with nursing homes after the inspection, with the final report due within +15 days. Some of these activities were added to make the inspection process more +“user friendly” for providers. Others, such as expanding the inspection to include +final interviews with family council members, were at the urging of advocacy +groups. + +For the year ending +September 30, 2004, +MDH inspectors, +working in teams of +three to five +registered nurses, +spent an average of +about 150 hours per +facility to complete +the state and +federally mandated +inspection tasks.12 +As would be +expected, it took +longer to inspect +larger nursing homes +than smaller ones. +For example, a +facility with 40 or +fewer beds averaged about 72 hours per inspection while a facility with 116 to +160 beds averaged 176 hours.13 + +Nursing home inspectors must meet with each facility's resident +council. + +Inspection Standards + +The federal State Operations Manual (SOM) sets forth the federal standards that +inspectors must apply during an inspection as well as guidelines to help them +apply those standards.14 As currently written: + +12 Minnesota Department of Health analysis of data from the Online Survey and Certification +Reporting System, December 2, 2004. State inspectors spent an additional 54 hours per facility, on +average, conducting follow-up inspections to ensure that facilities corrected deficiencies. +13 Minnesota Department of Health, Federal Fiscal Year 2005 Initial Budget Request (St. Paul, +July 15, 2004), unnumbered. +14 Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004). + + THE NURSING HOME INSPECTION PROCESS + +9 + +Inspectors grade +the seriousness of +each deficiency +by assigning it a +letter code. + +• The federal standards and guidelines that state inspection teams must + +use to inspect nursing homes are prescriptive and complex. + +The SOM covers hundreds of pages and contains 274 regulatory standards that +nursing homes must meet at all times. The standards cover 16 different categories +of operation, including administration, dietary services, infection control, life +safety, physical environment, quality of care, quality of life, resident assessment, +and resident rights. Some requirements must be met for each resident and any +violation of these requirements, even for one resident, is a deficiency. For +example, each resident must have a comprehensive care plan. Other requirements +focus on facility systems and are evaluated comprehensively rather than in terms +of a single incident. For example, a facility must have a medication error rate +below 5 percent.15 + +For each deficiency, inspectors must use professional judgment to assess how +many residents or staff are affected by or involved in the deficient practice (scope) +and the amount of actual or potential discomfort or harm involved for residents +(severity). As shown in Table 1.2, these two determinations result in the +inspection team assigning a letter code (A through L) to each deficiency, with +level “A” deficiencies being the least serious. + +Table 1.2: Deficiency Scope and Severity Grid + +Severity + +Scope +Isolated Pattern Widespread + +Level 4: A situation that has caused or is likely to cause +serious resident injury, harm, impairment, or death. + +Level 3: A situation that has caused resident harm. + +Level 2: A situation that has caused minimal discomfort to +a resident OR has the potential to cause resident harm. + +Level 1: A situation that has the potential of causing no + +more than minimal discomfort to a resident. + +J + +G + +D + +A + +K + +H + +E + B -| | | | 1 1 | (cid:88) s | | | -| --- | --- | --- | ----- | ---------- | --- | --- | -| | | r = | , c = | il . | | (6) | -| | | i | l | | | | -| | | | B B | |S | | | | -i=1 i -| ThetransportcostmatrixM | | | ∈RB×C | inOTisconstructedfromthesim- | | | -| ----------------------- | --- | --- | ----- | ---------------------------- | --- | --- | -cost -ilaritybetweenvisualandtextualrepresentations.Specifically,givenaninstance -| x , the | cost of assigning | it to class | j is defined | as: | | | -| ------- | ----------------- | ----------- | ------------ | --- | --- | --- | -i -(cid:40) exp(cos(V(xi),T(t,vj))) -| | | 1− (cid:80)C | | , if | j ∈S i | | -| --- | -------- | ------------ | ----------------------- | ---------- | ------ | --- | -| | M [i,j]= | | exp(cos(V(xi),T(t,vq))) | | | (7) | -| | cost | | q=1 | | | | -| | | ∞, | | otherwise. | | | - -HopS 7 -This cost formulation encourages semantic alignment between image-label -pairs,whilediscouragingassignmentstonon-candidateclassesbyassigningthem -infinite cost. -The optimal transport plan P is aims to move probability mass from a uni- -form distribution over instances to a candidate-aware label distribution with -minimal total cost. To ensure differentiability and enable efficient computation, -we incorporate an entropy regularization term H(P)=− -(cid:80)B (cid:80)C -P logP -i=1 j=1 ij ij -following [4]. The resulting optimal transport objective is formulated as: -min ⟨P,M ⟩−εH(P) -cost -P∈RB -+ -×C -(8) -s.t. P1 =r, P⊤1 =c, P =0ifs =0. -C B ij ij -Here, ε is a hyper-parameter controlling the strength of the entropy regulariza- -tion, and 1 and 1 denote all-one vectors of length C and B, respectively. -C B -Approximation Scheme. To efficiently solve the entropy-regularized optimal -transport problem described in Eq. (8) and compute the transport plan, we -adoptthewell-knownSinkhorn-Knoppalgorithm,anefficientiterativemethod. -(cid:18) (cid:19) -M -Pot =diag(α)·exp − cost ·diag(β), (9) -ε -The iterative steps are given as follows: -α←r⊘(Mβ), β ←c⊘(M⊤α). (10) -where α and β are scaling vectors, ⊘ denotes element-wise division, r and c are -the source and target marginal distributions defined above, and M is the Gibbs -kernel [4] derived from the cost matrix. -After computing the optimal transport plan Pot ∈ RB×C, which encodes -the soft assignment between instances and candidate labels, we can select the -best "global-harmony" label yglobal with the largest transport mass. Compared -with "local-consensus" candidate, this strategy ensures that the selected label -not only respects the candidate constraint but also reflects the global semantic -alignment established via the transport plan, thereby promoting consistency -across the entire training set. -4.3 Learning Objective and Procedure -The Objective Function.Afterselectthetwomostplausiblelabelsylocal and -yglobal, we conduct prompt optimization and jointly optimize the cross-entropy -loss. Given an input image x, the overall loss is defined as: -argmin L (p(y |x;t),ylocal)+λL (p(y |x;t),yglobal), (11) -t CE CE -where λ is the weighting coefficient for the two loss components, the prediction -probability p(y |x;t) for each category is computed by Eq. (2). - -| 8 | Y. Zhao et | al. | | | | | | -| --------- | ---------------- | -------- | ------- | --------- | ------------- | -------- | ------- | -| Algorithm | 1 The | Holistic | Optimal | Selection | | | | -| Input: | Training dataset | D={(x | | ,S )}n | , pre-trained | encoders | V and T | -i i i=1 -| Parameter: | k,τ,ε,λ | | | | | | | -| ---------- | ---------- | ------ | --- | --- | --- | --- | --- | -| Output: | The prompt | vector | | | | | | -t -| 1: Extract | and store | all | image features | | R ←V(X). | | | -| ---------- | --------- | --- | -------------- | --- | -------- | --- | --- | -vis -| 2: Calculate | A between | | R by | cosine similarity. | | | | -| ------------ | --------- | --- | ---- | ------------------ | --- | --- | --- | -vis -3: -| for | batch=1,2,... | do | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | -4: -// LDF -5: Choose k neighbors to compute N corresponding to each instance by Eq. (3). -i -| 6: Calculate | the | frequency | f(·) | by Eq. | (4). | | | -| ------------ | --- | --------------- | ---- | ------ | ---- | --------------------- | --------------- | -| 7: Construct | the | consensus-aware | | mask | set | and refined candidate | set by Eq. (5). | -8: Identify Ylocal with the highest prediction probability in the refined candidate -set. -9: // GOP -| 10: Initialize | the | distribution | of | r and | c by Eq. | (6). | | -| -------------- | --- | ------------ | --- | ----- | -------- | ---- | --- | -11: Calculate the cost matrix M with T(t,v ) and R [instance indexes] by Eq. -| | | | | cost | | j vis | | -| --- | --- | --- | --- | ---- | --- | ----- | --- | -(7). -12: -| for | iter=1,2,... | | do | | | | | -| ---------- | ----------------- | --------- | ----------- | --------- | -------- | ----- | --- | -| 13: | // Sinkhorn-Knopp | | Algorithm | | | | | -| 14: | Approximate | transport | plan | by | Eq. (9). | | | -| 15: end | for | | | | | | | -| 16: Gain | the Yglobal | with | the largest | transport | | mass. | | -| 17: Update | with | Eq. | (11). | | | | | -t -18: end for -Training Procedure. As illustrated in Algorithm 1, the HopS performs both -LDF and GOP selection to iteratively refine predictions under partial supervi- -sion, resulting in a more robust and discriminative prompt. The computational -| overhead | is negligible | (see | Tab. | 4). | | | | -| -------- | ------------- | ---- | ---- | --- | --- | --- | --- | -5 Experiments -We conduct experiments on eight datasets, comparing our approach with eight -representative loss functions, zero-shot learning, and 16-shot fully supervised -prompt tuning using CoOp. All methods are evaluated under varying degrees of -labelambiguity.Ourmethodconsistentlyoutperformsallbaselines,highlighting -its effectiveness in prompt learning with partial labels. Extensive experiments -andanalysesfurtherdemonstratethecomplementarityofthetwolabelselection -strategies. -| 5.1 Experimental | | Settings | | | | | | -| ---------------- | --- | -------- | --- | --- | --- | --- | --- | -Basicsettingsareoutlined.Additionaldetails(e.g.,hyperparameters)areinthe -| Appendix | Sec. A. | | | | | | | -| -------- | ------- | --- | --- | --- | --- | --- | --- | -Dataset. We adopt eight datasets: Caltech [5], DTD [2], EuroSAT [11], FGV- -CAircraft [22], Food [23], Flowers [23], OxfordPets [24], and UCF [28]. These - -HopS 9 -datasets encompass a wide spectrum of visual recognition tasks, especially for -challenging fine-grained classification. -Confusion Types. We manually corrupt the datasets into partially labeled -versions using two strategies: random-uniform (rand) and instance-dependent -(insd). (1) In the rand setting, confusing labels are randomly selected into the -candidate set with equal probability for each instance. (2) The insd setting de- -pends on instance-level information. Following [44], we adopt a prototype-based -label confusion strategy, which uses pre-trained image encoders to identify the -top-(L−1)mostsimilarclasses—excludingtheground-truthlabel—ascandidate -labels.Weemploythreevisualbackbones(i.e.,ResNet-18/50[10]trainedonIm- -ageNet and CLIP-ResNet50 [26]) to simulate annotators with varying levels of -domain expertise. -Confusion Levels. The confusion rate, defined as the ratio of the number -of confused labels to L, serves as an indicator of the difficulty in identifying -the ground-truth label within the candidate set. The insd-confusion and rand- -confusionlevelsofthecandidatelabelsarecontrolledbythesizeofthecandidate -label set L, selected from {2,3,4,5,8,9,10}, with the corresponding confusion -rates γ being {0.50,0.67,0.75,0.80,0.88,0.89,0.90}. -c -Prompt Types. The design of prompts can be categorized into two forms, -consistent with CoOp: unified prompt (uni) and classified prompts (cls). The -uni prompt shares a common set of context vectors across all classes, making it -suitable for general classification tasks. In contrast, cls assigns an independent -set of context vectors to each class. -Baselines. We compare HopS with five state-of-the-art partial-label learning -losses:CC[7],RC[7],LWC[37],MSE[6],andEXP[6].Furthermore,weinclude -three well-known robust learning losses, MAE [36], SCE [36], and GCE [49], -as well as three recent partial-label learning methods, Papi [42], CroSel [31], -and SoLar [34], representing recent advances in the field. All hyperparameters -for these methods are carefully tuned following the settings reported in their -original publications. The detailed configurations of all compared methods are -provided in Appendix Tab. 6. -The Learning Framework. To ensure a fair comparison, we adopt the most -influentialcontextoptimizationmethod,CoOp,andconductexperimentsinboth -uni andcls promptstypes.Allresultsarereportedastheaverageoverthreeruns -using three different random seeds. -5.2 Comparison Results -Rand-Confusion Type. We evaluate all methods under two prompt settings: -uni-prompt and cls-prompts. As shown in Fig. 2, HopS achieves SOTA per- -formance across datasets under varying levels of label confusion. It is worth -highlighting that, on Caltech, Flowers, and UCF, HopS achieves performance -comparable to fully supervised 16-shot CoOp and significantly outperforms the -zero-shotcapabilityofCLIPacrossalldatasets.Moreover,HopSmaintainscon- -sistently high performance even under a severe label confusion rate of 0.9. This - -| 10 Y. | Zhao et | al. | | | | | -| ----- | ------- | --- | --- | --- | --- | --- | -Fig.2: HopS achieves the best testing accuracy under uni-prompt (top) and cls- -| prompts (bottom) | | across five | rand confusion | rates. | | | -| ---------------- | -------- | ----------- | -------------- | ---------------- | ------------ | --------------- | -| HopS | achieves | the best | averaged | testing accuracy | across three | confusion rates | -Fig.3: -| (left). The | guiding | effect of LDF | on | GOP (right). | | | -| ----------- | ------- | ------------- | --- | ------------ | --- | --- | -highlights the effectiveness of our holistic label selection strategy in reducing la- -bel ambiguity and providing reliable supervision. The detailed numerical results -| can be found | in Appendix | Tabs. | 7 | and 8. | | | -| ------------ | ----------- | ----- | --- | ------ | --- | --- | -Insd-Confusion Type. Since partial-label losses perform well under the rand -confusionsetting,wefurthercompareHopSagainstthemunderinstance-dependent -partiallabels,whichmorecloselyresemblereal-worldscenarioswherelabelnoise -dependsonfeature-levelsimilarities.AsshowninFig.3(left),HopSconsistently -outperformsothermethodsundertheuni-promptsettingwhenaveragingresults -overResNet-18,ResNet-50,andCLIP-ResNet50simulatingbackbones.Notably, -our method retains a substantial advantage even as the confusion rate rises to -0.8. This demonstrates the robustness of HopS in more challenging and realis- -tic conditions. In particular, this performance gain holds across settings where -candidate labels are derived from annotators with varying levels of domain ex- -pertise (i.e., simulated through different backbone encoders that reflect distinct -data conditions), indicating that HopS is resilient to the type and quality of -label noise. This robustness stems from the complementary nature of LDF and - -HopS 11 -Fig.4:OverlapoflabelsidentifiedbyLDF Fig.5: The performance remains stable -and GOP in HopS with the GT T on the acrossawiderangeofλoneightdatasets -Food (γ =0.67). (directions), and λ=1 works well. -c -GOPcomponentsinHopS.TheseresultsindicatethatHopSiseffectivenotonly -in synthetic settings but also in real-world scenarios with instance-dependent -ambiguity, such as crowd-sourced labels or low-resource domains. The detailed -per-configuration results are reported in Appendix Tabs 9 and 10. -5.3 Further Analysis -Complement Effect of LDF and GOP. To assess the interaction between -the LDF and GOP modules, we compare their individual performance and that -of the holistic model, HopS, in identifying the ground-truth (GT) label. -The training accuracy of LDF, GOP, and HopS over 200 epochs, evaluated un- -der the uni-prompt on the Food dataset at confusion rates of 0.67 and 0.80, -is illustrated in Fig. 3 (right), emphasizing LDF’s role in accelerating early op- -timization by guiding the global module. The four curves compare LDF and -GOP used stand-alone versus used as components within HopS. The symbol ∩ -denotestheproportionofpredictedlabelsfromtheLDForGOPthatareconsis- -tent with the GT label, where T denotes the GT label. Using modules as HopS -components yields faster convergence and a higher, more stable plateau, and -thisadvantagebecomesincreasinglypronouncedastheγ intensifies.Additional -c -results are provided in Appendix Figs. 9-16. -Furthermore, Fig. 4 shows that the proportion of correct labels jointly iden- -tified by both modules (i.e., the brown region) steadily increases, indicating a -strong complementary effect. We also examine the relative contributions of the -twopseudo-labelsourcesbyvaryingthelossweightingcoefficientλ∈{0.5,1.0,2.0}. -As shown in Fig. 5, achieving a balanced combination of local and global sig- -nals leads to consistent improvements in performance across different confu- -sions. In contrast, placing excessive weight on either signal—whether local or -global—results in performance degradation. This finding underscores the criti- -cal importance of maintaining an effective synergy between the two sources of -information. The specific results are provided in Tab. 11 in the Appendix. -EffectofBatchSize.Weinvestigatetheeffectofvaryingbatchsizes,B,within -the set {16,32,64,128,256} for GOP, on the performance of HopS, focusing on -the testing accuracy under varying levels of label confusion: 0.50, 0.75, 0.80, -0.88, and 0.90. As illustrated in Fig. 6, the results compare the performance of -uni-prompt and cls-prompts across different confusion rates. For the 47-class - -| 12 | Y. Zhao et al. | | | | | | -| --- | -------------- | --- | ---------- | ---------------- | ----------- | ----------- | -| | | | | Testing accuracy | (%) of | strategies. | -| | | | Table 1: | | | | -| | | | γ c CoOp | Nei(Hard) | Con Gra | Soft | -| | | | 0.50 84.70 | | 86.20 83.40 | 86.70 | -86.71 -| | | | 0.67 80.00 | 85.60 | 84.30 83.10 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -86.00 -| | | | 0.80 74.60 | 84.10 | 83.80 78.40 | | -| --- | --- | --- | ---------- | ----- | ----------- | --- | -84.80 -closetoC isrecommended. -Fig.6: B -small-scaledatasetDTDandthe100-classlarge-scaledatasetCaltech,theaccu- -racytrendswithrespecttoB remainconsistentacrossdifferentconfusionlevels. -Moreover, the most significant fluctuations in accuracy are observed when the -batch size is approximately equal to the number of classes. Based on this ob- -servation, we recommend choosing a batch size close to the number of classes. -| Results | for other datasets | are | provided in the | Appendix Fig. | 17 and Tab. | 12. | -| ------- | ------------------ | --- | --------------- | ------------- | ----------- | --- | -Effect of LDF. We compare three candidate-searching strategies for label re- -finementinpartial-labellearning:(1)Neighbor-basedrefinement(Nei),whichse- -lectspseudo-labelsbasedonfeaturesimilaritytonearestneighbors;(2)Confidence- -based refinement (Con), which chooses the most confident prediction; and (3) -Graph-based propagation (Gra), which propagates labels through a constructed -similarity graph, enabling global label consistency by considering relationships -between all samples in the dataset and effectively transferring labels across con- -nectednodes.AsshowninTab.1,thesimpleststrategy—neighbor-basedrefine- -ment—yields the best performance, suggesting that in weakly supervised few- -shotsettings,leveraginglocalstructuralsimilaritycanoutperformmorecomplex -propagation methods. One possible reason for this observation is that neighbor- -based methods are inherently more robust to noisy predictions. By grounding -pseudo-labelselectioninlocalfeatureconsistency,thesemethodscanbettercap- -ture semantic regularities in the data. In contrast, graph-based methods might -cause noise accumulation, especially in low-data regimes, where the constructed -| graph may | not accurately | reflect | the underlying | label manifold. | | | -| --------- | -------------- | ------- | -------------- | --------------- | --- | --- | -Additionally,weevaluatetwolabelupdatestrategies:Hard-KNN,whichuses -majorityvoting,andSoft-KNN,whichusessimilarity-weightedvoting.Although -Soft-KNNslightlyoutperformsHard-KNNinsomecases,thegainsaremarginal, -indicatingthatthesimplehardvotingmechanismissufficientandmorecompu- -| tationally | efficient. | | | | | | -| ---------- | ---------- | --- | --- | --- | --- | --- | -EffectofGOP.TheGOPmoduleassumesthatthecandidatelabeldistribution -closely resembles the ground-truth label distribution. To assess this consistency, -we use cosine similarity, which captures the directional alignment between two -distributionsandreflectstheirstructuralsimilarityregardlessofscale.Thismet- -ricprovidesareliablemeasureofhowwellthecandidatelabelsapproximatethe -truesemanticdistribution.AsshowninFig.7,thecosinesimilarityapproaches1 -on EuroSAT under four different confusion rates, indicating a strong alignment -between the distributions. We opt not to use Kullback–Leibler (KL) divergence -due to the prevalence of zero values in the ground-truth distributions, which -| makes KL | computation | unstable | and ill-defined. | | | | -| -------- | ----------- | -------- | ---------------- | --- | --- | --- | - -| | | | | | | | | HopS | 13 | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | -Fig.7:Highcosinesimilaritybetweencan- Fig.8: Testing accuracy under the uni- -didate and ground-truth label distribu- prompt across two noisy conditions with -| tions | across batches. | | | | different | confusion | rates. | | | -| ----- | --------------- | -------- | --- | ---- | ---------------- | --------- | ------ | --------- | ------ | -| | HopS | achieves | the | best | testing accuracy | (%) under | two | confusion | types. | -Table 2: -| | | | Caltech | | | | Flowers | | | -| --- | ------------------------------- | ---- | ------- | --------- | ---------------------------- | --------- | ------- | ---- | ---- | -| | γc HopSCroSelPapiHopSCroSelPapi | | | | HopSCroSelPapiHopSCroSelPapi | | | | | -| | 0.67 89.0 | 84.0 | 55.9 | 50.3 39.0 | 42.0 92.9 | 92.1 78.7 | 82.0 | 78.0 | 67.0 | -| | 0.75 89.0 | 82.7 | 51.8 | 45.3 25.2 | 23.5 92.9 | 90.5 70.4 | 74.0 | 62.0 | 58.7 | -| | 0.80 89.5 | 73.9 | 52.2 | 38.9 11.8 | 13.4 93.2 | 89.0 57.8 | 65.0 | 50.2 | 49.8 | -Comparison with Full-Data PLL Methods. We conducted a systematic -comparisonwithtworecentlyproposedpartial-labellearningmethods,Papi[42] -and CroSel [31]. Unlike the 16-shot setting used in previous experiments, Papi -andCroSelweretrainedontheentiredatasetstofullyexploittheiroptimalper- -formance, while HopS was still trained under the 16-shot few-shot setting. The -experiments were conducted on two challenging datasets, Caltech and Flowers, -both characterized by a large number of categories, with Flowers being more -fine-grained and exhibiting higher intra-class similarity. As shown in Tab. 2, the -left and right sub-columns for each dataset correspond to rand and insd, re- -spectively. HopS’s strong performance across both datasets and three confusion -rates,demonstratingitsrobustnessinhandlingbothdiverseandfine-grainedcat- -egories. Due to space limitations, we report only the results with the ResNet-50 -| backbone; | results | | for other | backbones | are provided | in | Appendix | Tab. | 13. | -| --------- | ------- | --- | --------- | --------- | ------------ | --- | -------- | ---- | --- | -Settings with Missing Ground-Truth. To gain deeper insights into the ro- -bustness of HopS, we evaluate it under more challenging settings where the -ground-truth label may be absent from the candidate set, i.e., the noisy partial -label learning scenario [43]. We conduct 16-shot experiments on two types of -label confusion, rand and insd, as previously described, with each candidate set -S containing three labels. The missing rates of ground-truth labels are 12.5% (2 -| out | of 16 shots) | for | rand | and 25% | (4 out of 16 | shots) | for insd. | | | -| --- | --------------------------- | --- | ---- | ------- | ------------------------------------- | ------ | --------- | --- | --- | -| | AsshowninFig.8,undertheinsd | | | | setting,HopSconsistentlyuncoversmean- | | | | | -ingfulcorrelationswithinthecandidatelabelsets,leadingtosignificantlybetter -performance than other methods. This suggests that HopS is particularly well- -suited for real-world scenarios where label noise in SS is instance-dependent. In -contrast,undertherand settingwheresuchdependenciesareabsent,HopSper- -forms slightly worse than the SOTA robust loss. Nevertheless, its performance - -| 14 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -Table 3: Testing accuracy (%) under under long-tailed partial- Table 4: Compari- -label settings with different confusion types. son of epoch time. -| | | | γ =0.67 | | | γ =0.75 | | Method | Time(s) | -| ------- | ------ | ----- | ----------- | --------- | ----- | ----------- | ----- | ------ | ------- | -| Dataset | Method | | c | | | c | | | | -| | | insde | insds rande | rands | insde | insds rande | rands | EXP | 0.12 | -| | | | | | | | | MSE | 0.11 | -| | SoLar | 9.2 | 7.9 | 16.1 22.6 | 6.4 | 4.4 18.5 | 19.4 | | | -| | | | | | | | | GCE | 0.11 | -| Caltech | HopSu | 46.4 | 50.7 | 71.7 71.2 | 35.6 | 42.0 67.6 | 67.4 | | | -| | HopSc | 50.9 | 52.8 | 74.6 71.4 | 42.4 | 45.9 72.6 | 70.7 | LWC | 0.11 | -| | | | | | | | | MAE | 0.12 | -| | SoLar | 11.1 | 14.0 | 18.1 20.8 | 9.1 | 10.6 17.3 | 20.7 | | | -| | | | | | | | | SCE | 0.11 | -| Flowers | HopSu | 48.3 | 51.4 | 65.4 61.1 | 39.8 | 43.8 61.9 | 53.9 | | | -| | HopSc | 54.4 | 53.1 | 69.9 66.1 | 46.8 | 47.5 69.9 | 59.7 | HopS | 0.12 | -can be improved by integrating robust loss functions (e.g., the SCE loss). Fig- -ure 8 (left) illustrates that HopS, when combined with the SCE loss, achieves -thebestperformanceundertherand setting,highlightingitsflexibilityandcom- -patibility with other robust learning techniques. Detailed values corresponding -| to Fig. | 8 can | be found | in Appendix | | Tabs. | 14 and 15. | | | | -| ------- | ----- | -------- | ----------- | --- | ----- | ---------- | --- | --- | --- | -Challenging Settings with Long-tail Distribution. To further evaluate -robustness under distribution shift, we compare HopS with SoLar [34] on two -representative datasets under two long-tailed patterns: exponentially decayed -class-frequencydistribution(e)andtwo-leveldistribution(s).Assummarizedin -Table3,SoLarsuffersasignificantperformancedropinthedata-limitedfew-shot -setting, as it relies on learning sufficiently strong representations. In contrast, -HopS consistently outperforms SoLar across all long-tailed configurations, re- -| gardless | of whether | | uni-prompts | (HopSu) | | or cls-prompts | (HopSc) | | are used. | -| -------- | ---------- | --- | ----------- | ------- | --- | -------------- | ------- | --- | --------- | -Running Time of Methods. We measured the training time computational -cost of each method, reported as the average time per epoch (in seconds), as -shown in Tab. 4. Notably, although HopS introduces both a local module and -a global module to enhance label identification, the additional computational -overhead remains negligible due to the relatively small dataset size. As a result, -HopS achieves superior performance without incurring a significant increase in -| training | cost. | Additional | results | are | provided | in Tab. | 16 of | Appendix. | | -| -------- | ----- | ---------- | ------- | --- | -------- | ------- | ----- | --------- | --- | -6 Conclusion -In this work, we propose a holistic label selection framework for prompt learn- -ing under partial-label supervision, which improves robustness by combining -a local density-based filter with a global optimal transport planner. By lever- -aging the generalization ability of frozen pre-trained vision-language encoders, -HopS enables effective label disambiguation. Extensive experiments on eight -datasets demonstrate consistent and significant improvements over strong base- -lines, highlighting its practicality and generalizability under weak supervision. - -HopS 15 -References -1. Chen, Y., Patel, V., Chellappa, R., Phillips, P.: Ambiguously labeled learning us- -ing dictionaries. IEEE Transactions on Information Forensics and Security 9(12), -2076–2088 (2014) 3 -2. Cimpoi,M.,Maji,S.,Kokkinos,I.,Mohamed,S.,Vedaldi,A.:Describingtextures -in the wild. In: Computer Vision and Pattern Recognition (2014) 8 -3. Cour, T., Sapp, B., Taskar, B.: Learning from partial labels. Machine Learning -Research 12, 1501–1536 (2011) 3 -4. Cuturi,M.:Sinkhorndistances:Lightspeedcomputationofoptimaltransport.Ad- -vances in Neural Information Processing Systems 26 (2013) 7 -5. Fei-Fei, L., Fergus, R., Perona, P.: Learning generative visual models from few -training examples: An incremental bayesian approach tested on 101 object cate- -gories. In: Computer Vision and Pattern Recognition Workshop (2004) 8 -6. Feng, L., Kaneko, T., Han, B., Niu, G., An, B., Sugiyama, M.: Learning with -multiplecomplementarylabels.In:InternationalConferenceonMachineLearning -(2020) 9 -7. Feng,L.,Lv,J.,Han,B.,Xu,M.,Niu,G.,Geng,X.,An,B.,Sugiyama,M.:Prov- -ably consistent partial-label learning. Advances in Neural Information Processing -Systems 33, 10948–10960 (2020) 3, 9 -8. Frans, K., Soros, L., Witkowski, O.: Clipdraw: Exploring text-to-drawing synthe- -sis through language-image encoders. Advances in Neural Information Processing -Systems 35, 5207–5218 (2022) 1 -9. Guo,Y.,Li,S.,Liu,Z.,Zhang,T.,Chen,C.:Aparameter-efficientandfine-grained -prompt learning for vision-language models. In: Association for Computational -Linguistics. pp. 31346–31359 (2025) 3 -10. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. -In: Computer Vision and Pattern Recognition (2016) 9 -11. Helber, P., Bischke, B., Dengel, A., Borth, D.: Eurosat: A novel dataset and deep -learning benchmark for land use and land cover classification. IEEE Journal of -SelectedTopicsinAppliedEarthObservationsandRemoteSensingpp.2217–2226 -(2019) 8 -12. Jia, Y., Yang, F., Dong, Y.: Partial label learning with dissimilarity propagation -guidedcandidatelabelshrinkage.AdvancesinNeuralInformationProcessingSys- -tems (2023) 3 -13. Jin,R.,Ghahramani,Z.:Learningwithmultiplelabels.AdvancesinNeuralInfor- -mation Processing Systems 15 (2002) 3 -14. Jin,X.,Zhang,H.,Wu,Z.,etal.:Unsupervisedpromptlearningforvision-language -models. In: Computer Vision and Pattern Recognition (2022) 3 -15. Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P.: Fine-tuning can dis- -tort pretrained features and underperform out-of-distribution. In: International -Conference on Learning Representations (2022) 2 -16. Li, J., Lu, Y., Xie, Y., Qu, Y.: Relationship prompt learning is enough for open- -vocabulary semantic segmentation. Advances in Neural Information Processing -Systems (2024) 3 -17. Li, Z., Li, X., Fu, X., Zhang, X., Wang, W., Chen, S., Yang, J.: Promptkd: Unsu- -pervised prompt distillation for vision-language models. In: Computer Vision and -Pattern Recognition (2024) 3 -18. Liu,P.,Yuan,W.,Fu,J.,Jiang,Z.,Hayashi,H.,Neubig,G.:Pre-train,prompt,and -predict:Asystematicsurveyofpromptingmethodsinnaturallanguageprocessing. -ACM Computing Surveys 55(9), 1–35 (2023) 1 - -16 Y. Zhao et al. -19. Luo, Z., Zhao, P., Xu, C., Geng, X., Shen, T., Tao, C., Ma, J., Lin, Q., Jiang, -D.:Lexlip:Lexicon-bottleneckedlanguage-imagepre-trainingforlarge-scaleimage- -textsparseretrieval.In:InternationalConferenceonComputerVision.pp.11206– -11217 (2023) 1 -20. Lv,J.,Liu,Y.,Xia,S.,Xu,N.,Xu,M.,Niu,G.,Zhang,M.,Sugiyama,M.,Geng, -X.: What makes partial-label learning algorithms effective? Advances in Neural -Information Processing Systems (2024) 1, 3 -21. Lv,J.,Xu,M.,Feng,L.,Niu,G.,Geng,X.,Sugiyama,M.:Progressiveidentification -of true labels for partial-label learning. In: International Conference on Machine -Learning (2020) 3 -22. Maji, S., Rahtu, E., Kannala, J., Blaschko, M., Vedaldi, A.: Fine-grained visual -classification of aircraft. arXiv preprint arXiv:1306.5151 (2013) 8 -23. Nilsback, M., Zisserman, A.: Automated flower classification over a large number -of classes. In: Computer Vision, Graphics and Image Processing (2008) 8 -24. Parkhi,O.,Vedaldi,A.,Zisserman,A.,Jawahar,C.:Catsanddogs.In:Computer -Vision and Pattern Recognition (2012) 8 -25. Qin, Y., Chen, X., Shen, Y., Fu, C., Gu, Y., Li, K., Sun, X., Ji, R.: Capro: Webly -supervised learning with cross-modality aligned prototypes. Advances in Neural -Information Processing Systems (2023) 2 -26. Radford, A., Kim, J., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., -Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from -natural language supervision. In: International Conference on Machine Learning -(2021) 1, 9 -27. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution -imagesynthesiswithlatentdiffusionmodels.In:InternationalConferenceonCom- -puter Vision. pp. 10684–10695 (2022) 1 -28. Soomro, K., Zamir, A., Shah, M.: Ucf101: A dataset of 101 human actions classes -from videos in the wild. arXiv preprint arXiv:1212.0402 (2012) 8 -29. Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., -Wang,J.:Alpha-clip:Aclipmodelfocusingonwhereveryouwant.In:International -Conference on Computer Vision. pp. 13019–13029 (2024) 1 -30. Teo, C., Abdollahzadeh, M., Ma, X., Cheung, N.: Fairqueue: Rethinking prompt -learning for fair text-to-image generation. Advances in Neural Information Pro- -cessing Systems (2024) 3 -31. Tian, S., Wei, H., Wang, Y., Feng, L.: Crosel: Cross selection of confident pseudo -labels for partial-label learning. In: Computer Vision and Pattern Recognition -(2024) 3, 9, 13 -32. Tsimpoukelli,M.,Menick,J.,Cabi,S.,Eslami,S.,Vinyals,O.,Hill,F.:Multimodal -few-shot learning with frozen language models. Advances in Neural Information -Processing Systems 34, 200–212 (2021) 3 -33. Wang,H.,Xiao,R.,Li,Y.,Feng,L.,Niu,G.,Chen,G.,Zhao,J.:Pico:Contrastive -label disambiguation for partial label learning. In: International Conference on -Learning Representations (2022) 2, 3 -34. Wang, H., Xia, M., Li, Y., Mao, Y., Feng, L., Chen, G., Zhao, J.: Solar: Sinkhorn -labelrefineryforimbalancedpartial-labellearning.AdvancesinNeuralInformation -Processing Systems 35, 8104–8117 (2022) 3, 9, 14 -35. Wang, R., An, S., Cheng, M., Zhou, T., Hwang, S., Hsieh, C.: One prompt is not -enough:Automatedconstructionofamixture-of-expertprompts.In:International -Conference on Machine Learning (2024) 3 - -HopS 17 -36. Wang,Y.,Liu,W.,Ma,X.,Bailey,J.,Zha,H.,Song,L.:Symmetriccrossentropy -for robust learning with noisy labels. In: International Conference on Computer -Vision (2019) 9 -37. Wen,H.,Cui,J.,Hang,H.,Liu,J.,Wang,Y.,Lin,Z.:Leveragedweightedlossfor -partial label learning. In: International Conference on Machine Learning (2021) 9 -38. Wen, S., Brbic, M.: Cross-domain open-world discovery. In: International Confer- -ence on Machine Learning (2024) 2 -39. Werner, T., Burchert, J., Stubbemann, M., Schmidt-Thieme, L.: A cross-domain -benchmarkforactivelearning.AdvancesinNeuralInformationProcessingSystems -(2024) 2 -40. Wu,D.,Wang,D.,Zhang,M.:Revisitingconsistencyregularizationfordeeppartial -label learning. In: International Conference on Machine Learning (2022) 3 -41. Wu,M.,Cai,X.,Ji,J.,Li,J.,Huang,O.,Luo,G.,Fei,H.,Jiang,G.,Sun,X.,Ji,R.: -Controlmllm: Training-free visual prompt learning for multimodal large language -models. Advances in Neural Information Processing Systems (2024) 3 -42. Xia,S.,Lv,J.,Xu,N.,Niu,G.,Geng,X.:Towardseffectivevisualrepresentations -for partial-label learning. In: Computer Vision and Pattern Recognition (2023) 3, -9, 13 -43. Xu,M.,Lian,Z.,Feng,L.,Liu,B.,Tao,J.:Alim:adjustinglabelimportancemech- -anism for noisy partial label learning. Advances in Neural Information Processing -Systems 36, 38668–38684 (2023) 13 -44. Xu, N., Qiao, C., Geng, X., Zhang, M.: Instance-dependent partial label learning. -Advances in Neural Information Processing Systems 34, 27119–27130 (2021) 3, 9 -45. Yang, F., Cheng, J., Liu, H., Dong, Y., Jia, Y., Hou, J.: Mixed blessing: Class- -wise embedding guided instance-dependent partial label learning. arXiv preprint -arXiv:2406.10502 (2024) 3 -46. Yu, F., Zhang, M.: Maximum margin partial label learning. In: Asian Conference -on Machine Learning (2016) 3 -47. Zeng, F., Cheng, Z., Zhu, F., Wei, H., Zhang, X.: Local-prompt: Extensible local -promptsforfew-shotout-of-distributiondetection.In:InternationalConferenceon -Learning Representations (2025) 1 -48. Zhang,F.,Jiang,W.,Shu,J.,Zheng,F.,Wei,H.,etal.:Onthenoiserobustnessof -in-contextlearningfortextgeneration.AdvancesinNeuralInformationProcessing -Systems (2024) 1 -49. Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural -networks with noisy labels. Advances in Neural Information Processing Systems -(2018) 9 -50. Zhou,K.,Yang,J.,Loy,C.,Liu,Z.:Learningtopromptforvision-languagemodels. -In: International Journal of Computer Vision. pp. 2337–2348 (2022) 1, 3 -51. Zhou,Y.,Xia,X.,Lin,Z.,Han,B.,Liu,T.:Few-shotadversarialpromptlearning -on vision-language models. Advances in Neural Information Processing Systems -37, 3122–3156 (2024) 2 - -18 Y. Zhao et al. -A Experimental Settings -A.1 Confusion Types. -To simulate realistic partial label learning scenarios, we assign each training -instance a candidate label set consisting of the ground-truth label and L-1 con- -fusing labels, thereby introducing controlled ambiguity. The confusion rates γ , -c -definedastheratioofthenumberofconfusedlabelstoL,servesasanindicator -ofthedifficultyinidentifyingtheground-truthlabelwithinthecandidateset.1) -Generationofrand-confusion,randomlysamplesL-1incorrectlabelsforeachin- -stance.Topromotediversity,previouslyusedconfusinglabelsforthesameclass -are avoided when possible. If the number of available incorrect labels is insuf- -ficient, sampling with replacement is applied. 2) Generation of insd-confusion, -intentionalselectsconfusionlabelsbasedonvisualsimilarity.Specifically,image -features are first extracted using a pre-trained model, and a prototype vector -is computed for each class. For each instance, the cosine similarity between its -feature representation and all class prototypes is calculated. Then, the top-(L- -1) most similar labels—excluding the ground-truth—are selected, contributing -to more challenging and informative candidate label sets with visually similar -confusing labels. -A.2 Implementation Details. -A 16-shot training set is randomly sampled from each dataset in a class- -balancedmanner,withafixedrandomseedtoensureexperimentalreproducibil- -ity.Theoriginaltestsetisusedforevaluation.Modeltrainingisconductedusing -the SGD optimizer with a learning rate of 0.002 for a maximum of 200 -epochs. A cosine annealing schedule is applied to the learning rate, with a con- -stant warm-up set to 1e-5 for the first epoch. The batch size is 32, and the -training objective is the standard cross-entropy loss. The model adopts ResNet- -50 as the backbone, configured with random initialization, 16 context tokens, -and the end token position for classification. In terms of module-specific hyper- -parameters, the confidence threshold τ = 0.4 and the number of nearest -neighbors k = 20 are used in the LDF module. For the GOP module, the -cost matrix in the Sinkhorn algorithm is scaled by an entropy regulariza- -tion coefficient ε = 0.05, and the number of Sinkhorn iterations is set to -50. The weighting coefficient for the two loss components is set to λ = 1.0. -Regarding computational resources, all experiments are conducted on a Linux- -based system equipped with eight NVIDIA RTX 4090 GPUs, with each model -requiring approximately 18–30 minutes for training and peaking at 6696 MiB of -memory usage. The software environment includes Python 3.8, PyTorch 1.13.1, -and CUDA 12.0, along with essential libraries such as NumPy 1.21.6, Pandas -1.3.5, scikit-learn 1.0.2, and torchvision 0.14.1. -Dataset. We adopt eight datasets to comprehensively evaluate our method. -Caltech and EuroSAT are general classification datasets covering diverse object - -| | | | | | | | | | HopS 19 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | -and scene categories. In contrast, DTD, FGVCAircraft, Food, Flowers, Oxford- -Pets,andUCFfocusonfine-grainedclassification,emphasizingsubtledifferences -within textures, aircraft models, food dishes, flower species, pet breeds, and hu- -man actions respectively. Table 5 presents the original performance across eight -datasets, evaluated using two models: the zero-shot CLIP model with hand- -0-cliph), -crafted prompts ( which requires no training, and the 16-shot CoOp -model with learnable prompts trained on precise labels. The learnable prompts -inthe16-shotCoOpmodelarefurtherdividedintotwovariants:unifiedprompt -(uni),whichuseasharedcontextvectoracrossallclasses(16-coopu),andclassi- -fiedprompts(cls),whichassigndistinctcontextvectorstoeachclass(16-coopc). -| All | results | are reproduced | | by ourselves | to | ensure a | consistent | comparison. | | -| --- | ------------ | -------------- | --------------------------------- | ------------------ | ----- | -------------- | ---------- | ---------------- | -------- | -| | Table 5: | Testing | accuracy | (%) of models | under | the uni-prompt | | and cls-prompts. | | -| | Dataset | Class | | Hand-craftedprompt | | | 0-cliph | 16-coopu | 16-coopc | -| | EuroSAT | 10 | acenteredsatellitephotoof[class] | | | | 37.6 | 82.9 | 83.3 | -| | OxfordPets | 37 | aphotoofa[class],atypeofpet. | | | | 85.7 | 86.5 | 78.9 | -| | DTD | 47 | | [class]texture. | | | 42.3 | 62.2 | 62.4 | -| | Caltech | 100 | | aphotoofa[class]. | | | 86.3 | 92.0 | 90.4 | -| | FGVCAircraft | 100 | aphotoofa[class],atypeofaircraft. | | | | 17.0 | 31.8 | 37.7 | -| | Food | 101 | aphotoofa[class],atypeoffood. | | | | 77.4 | 74.5 | 70.4 | -| | UCF | 101 | aphotoofapersondoing[class]. | | | | 61.8 | 74.8 | 74.9 | -| | Flowers | 102 | aphotoofa[class],atypeofflower. | | | | 66.1 | 95.2 | 95.1 | -Baselines. We compare HopS with eight state-of-the-art loss functions from -partial-label learning and robust learning. All hyperparameters for these meth- -ods are carefully tuned according to the settings reported in their original pub- -| lications, | | and the configurations | | of | all methods | are | detailed | in Table | 6. | -| ---------- | --- | ---------------------- | --- | --- | ----------- | --- | -------- | -------- | --- | -Inaddition,wecompareHopSwiththreerecentPLLmethods,CroSel,PaPi, -and SoLar, each of which adopts a different strategy for handling label ambigu- -ity. CroSel utilizes cross-entropy loss to train the label selection process, while -incorporating consistency regularization (with data augmentation and MixUp) -to reduce selection noise. PaPi combines a prototypical alignment loss (based -on KL divergence) with cross-entropy loss to optimize class representations and -improveambiguityhandling.SoLarisalong-tailedpartial-labellearningmethod -that improves robustness under class imbalance by leveraging image representa- -| tion | optimization | and | label-distribution | | refinement. | | | | | -| ---- | -------------- | --- | ------------------ | ------- | ----------- | --- | --- | --- | --- | -| B | Comparison | | Results | | | | | | | -| B.1 | Rand-Confusion | | | Labels. | | | | | | -All methods are evaluated under two prompt types: the (uni) prompt and the -(cls) prompts. The corresponding results are summarized in Tables 7 and 8, -where all values are reported as percentages (the percent sign is omitted). The -| best | performance | is | highlighted | in bold, | | and the second-best | | is underlined. | | -| ---- | ----------- | --- | ----------- | -------- | --- | ------------------- | --- | -------------- | --- | - -| 20 | Y. Zhao | et al. | | | | | | | | -| --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -HopS(ours) -Method -| | SCE | MSE | MAE | LWC | GCE | EXP | CC | RC | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLWC=− -LHopS=− -LSCE=α·− -(cid:124) -| | | | | i,j | (cid:88) | LEXP= | LCC=− | | | -| ------ | ----------- | -------- | ----- | ------------- | -------- | ----- | ----- | ----- | --- | -| LDFCro | (cid:88) | | | | LGCE= | | | LRC=− | | -| j | (cid:124) | LMSE= | LMAE= | yij·log(pij)− | | | | | | -| | yij·log(pij | (cid:88) | | | | | | | | -| | Cross j | | | | | | | | | -(cid:123) -| s sEntropy (cid:122) | | yij·log(pij)+β·− | | | | B 1 | | | Table | -| -------------------- | ------------------ | ---------------- | ----------------------- | ------------------ | ----------------------- | ------------ | ------------ | ------------ | ---------------- | -| | E (cid:123) | | | | B 1 | i=1 (cid:88) | B 1 | | Lossformula | -| | ntropy (cid:122) | B 1 | B 1 | | | | B (cid:88) | B | 1 | -| | | (cid:88)(cid:88) | | | i=1j=1 (cid:88)(cid:88) | | i=1 | B i=1j=1 | (cid:88)(cid:88) | -| | LDF | i=1j=1 B | i=1j=1 (cid:88)(cid:88) | | | B exp | | | B | -| | | | B | | | | log | | | -| (cid:125) | )+λ·− | | | | | C | | | C 6: | -| | (cid:125) | C | C | C | 1 | | (cid:32) | (cid:32) | | -| | | (pij−yij) | | | (cid:88) (1−pij)·yij | − | | log(pij)·yij | Partial | -| | | | |pij−yij| | i,j | | j=1 (cid:88) | j=1 (cid:88) | C | | -| (cid:124) | | | | (1−yij)·log(1−pij) | | | C | | | -| | (cid:124) | | | | | | pij·yij | | | -| GOPCro | (cid:88) | | | | | pij·yij | | | | -| j | | (cid:88) | | | | | | | | -| | yij·log(pij Reve j | | | | | | | | label | -| | | pij·log(yij) 2 | | | | | | | | -(cid:33) -| s (cid:123) (cid:122) | rs (cid:123) (cid:122) | | | | | | (cid:33) | | | -| --------------------- | ---------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | -sEntropy eCE -learning -GOP -(cid:125) -(cid:125) ) -| | | | | PLLmethodthatconsiderstrade-offbetweenlossesoncandidateand | | | | | losses | -| ----------------------------------------------------------- | ------------------------------------------------------------ | ---------------------------------------------------- | ------------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- | --------------------------------------------------------- | ------ | -| | Symmetriclosscombiningcross-entropyandreversecross-entropyto | | Robustlosstreatingallclassesequallybyminimizingabsoluteerror. | | | | Classifier-consistentPLLmethodunderthesameassumptionasRC. | Risk-consistentPLLmethodthatassumesthecandidatelabelsetis | | -| Compositelossjointlyoptimizingcross-entropyfromlocaldensity | | PLLmethodthatestimatesriskviameansquarederrorbetween | | | Robustlossgeneralizingcross-entropyandmeanabsoluteerrorto | PLLmethodthatadoptstheexponentiallossasariskestimator. | | | | -and -| filteringandglobaloptimalplanning.(λ=1.0) | | | | | | | | | robust | -| ----------------------------------------- | ------------------------------- | ------------------------------ | --- | --- | ----------------------------- | --- | --- | --------------------------------- | ------ | -| | enhancerobustness.(α=0.1,β=1.0) | | | | | | | uniformlysampledfromthetruelabel. | | -| | | predictionsandcandidatelabels. | | | improvetolerancetolabelnoise. | | | | | -learning -non-candidatelabelsets. -Description -losses - -HopS 21 -| Table | 7: Testing | accuracy | (%) under | uni-prompt | across | five rand | confusion rates. | -| ----- | ---------- | -------- | --------- | ---------- | ------- | --------- | ---------------- | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 82.0 88.2 | 59.6 36.8 | 80.9 32.8 | 86.1 89.0 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -89.4 -| | | 0.75 | 78.7 87.4 | 42.7 24.8 | 76.8 56.0 | 75.7 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | | | | | 89.0 | 89.0 | -| | Caltech | 0.80 | 75.6 89.1 | 55.0 33.4 | 77.4 29.3 | 86.9 86.5 | 89.5 | -| | | 0.88 | 66.0 88.6 | 37.0 15.2 | 72.1 20.4 | 83.0 76.6 | 89.5 | -| | | 0.90 | 65.8 87.1 | 44.5 18.0 | 70.3 23.0 | 80.0 76.2 | 89.5 | -| | | 0.50 | 46.7 62.0 | 37.5 31.7 | 45.3 30.3 | 60.3 60.4 | 62.1 | -| | | 0.75 | 37.1 | 42.1 27.7 | 37.1 26.5 | 54.4 58.2 | 59.8 | -59.9 -| | DTD | 0.80 | 33.8 57.1 | 37.4 31.0 | 35.2 19.9 | 53.6 58.4 | 59.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 32.7 55.8 | 27.5 17.1 | 28.6 22.4 | 48.0 53.4 | 55.8 | -| | | 0.90 | 22.9 48.3 | 26.0 23.8 | 24.2 17.6 | 44.9 50.9 | 55.6 | -| | | 0.50 | 63.5 81.4 | 77.6 60.0 | 64.1 71.5 | 78.0 81.8 | 83.6 | -| | | 0.75 | 60.2 81.1 | 69.5 54.4 | 55.3 50.2 | 75.5 80.7 | | -81.3 -| | EuroSAT | 0.80 | 45.3 72.0 | 71.2 59.4 | 41.6 54.5 | 60.2 72.2 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -74.8 -| | | 0.88 | 46.6 65.7 | 56.3 37.2 | 38.1 53.8 | 52.0 66.8 | 66.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 29.1 44.3 | 42.9 27.4 | 31.0 18.9 | 39.0 51.7 | 53.9 | -| | | 0.50 | 54.2 67.8 | 24.8 18.5 | 51.9 23.2 | 43.8 65.9 | 68.4 | -| | | 0.75 | 46.3 68.0 | 22.8 28.9 | 45.2 20.3 | 48.6 63.8 | | -69.0 -| | Food | 0.80 | 42.3 65.3 | 22.2 19.8 | 41.9 23.0 | 55.6 52.8 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.2 -| | | 0.88 | 33.3 64.4 | 19.9 21.3 | 32.2 11.1 | 44.2 38.9 | 66.0 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.3 62.3 | 20.9 13.1 | 27.6 16.7 | 49.1 30.8 | 65.0 | -| | | 0.50 | 78.0 93.5 | 46.0 32.8 | 80.8 20.0 | 78.2 92.3 | 93.5 | -| | | 0.75 | 74.1 91.7 | 42.0 29.5 | 76.5 23.8 | 80.8 79.3 | | -92.9 -| | Flowers | 0.80 | 70.6 92.7 | 45.0 19.3 | 70.8 23.7 | 75.4 79.3 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -93.2 -| | | 0.88 | 62.0 91.9 | 40.1 20.6 | 60.7 36.7 | 73.0 67.3 | 92.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 56.9 91.8 | 34.3 10.2 | 53.6 33.7 | 78.7 73.9 | 92.5 | -| | | 0.50 | 56.9 72.7 | 28.0 26.1 | 53.4 18.8 | 61.6 70.7 | 71.9 | -| | | 0.75 | 52.8 72.1 | 24.1 11.9 | 56.7 21.3 | 56.6 64.2 | | -72.9 -| | UCF | 0.80 | 50.3 71.3 | 37.1 17.2 | 51.6 20.4 | 58.8 69.1 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -71.7 -| | | 0.88 | 46.2 70.6 | 31.4 12.7 | 39.2 12.0 | 61.1 51.1 | 71.2 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 31.9 68.2 | 36.5 17.0 | 39.1 12.5 | 53.6 52.4 | 69.2 | -| | | 0.50 | 19.5 27.3 | 9.5 6.8 | 22.7 12.0 | 18.3 25.0 | 26.8 | -| | | 0.75 | 17.0 25.5 | 8.3 6.3 | 17.7 5.2 | 19.4 18.5 | 24.6 | -| | FGVCAircraft | 0.80 | 15.4 | 11.2 6.2 | 14.3 5.4 | 19.4 17.6 | 23.8 | -24.7 -| | | 0.88 | 11.2 20.4 | 10.7 5.0 | 11.4 4.4 | 16.8 13.0 | 20.2 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 11.4 20.4 | 10.9 4.1 | 10.0 7.4 | 14.6 10.6 | 19.3 | -| | | 0.50 | 58.9 82.5 | 52.3 38.0 | 60.6 14.7 | 82.0 84.0 | 83.5 | -| | | 0.75 | 49.4 83.0 | 46.9 31.0 | 49.4 46.6 | 76.2 82.6 | 83.8 | -| | OxfordPets | 0.80 | 42.4 80.2 | 49.3 22.7 | 40.9 26.3 | 72.5 80.9 | | -82.6 -| | | 0.88 | 34.8 73.9 | 23.7 39.1 | 31.3 38.1 | 74.2 76.8 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -79.9 -| | | 0.90 | 22.0 74.2 | 40.7 12.3 | 21.9 18.3 | 45.0 66.6 | 80.5 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -| 22 | Y. Zhao | et al. | | | | | | -| ----- | ---------- | -------- | --------- | ----------- | ------- | --------- | ---------------- | -| Table | 8: Testing | accuracy | (%) under | cls-prompts | across | five rand | confusion rates. | -| | Dataset | | γ RC | CC EXP GCE | LWC MAE | MSE SCE | HopS | -c -| | | 0.50 | 89.2 90.5 | 60.9 60.0 | 88.7 46.4 | 85.9 89.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.9 -| | | 0.75 | 87.6 85.8 | 61.3 55.2 | 87.6 57.1 | 79.5 87.3 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -90.4 -| | Caltech | 0.80 | 87.2 83.9 | 66.5 44.5 | 87.5 60.4 | 84.7 88.0 | 90.8 | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 85.0 74.5 | 59.7 48.4 | 85.0 56.9 | 76.7 88.0 | 90.5 | -| | | 0.90 | 83.3 74.3 | 61.3 56.1 | 83.2 60.5 | 74.8 87.5 | 89.7 | -| | | 0.50 | 62.2 62.6 | 52.2 44.6 | 62.0 44.3 | 59.2 61.7 | 61.9 | -| | | 0.75 | 55.4 60.9 | 37.7 33.0 | 55.9 38.6 | 52.0 58.1 | | -61.8 -| | DTD | 0.80 | 53.2 54.6 | 48.7 43.4 | 53.4 36.8 | 48.5 56.7 | 60.9 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.88 | 46.5 55.4 | 39.1 29.4 | 46.5 35.6 | 41.9 52.2 | 57.2 | -| | | 0.90 | 36.3 45.3 | 32.9 33.3 | 36.3 28.7 | 32.9 49.3 | 53.4 | -| | | 0.50 | 76.2 83.9 | 83.8 72.6 | 76.4 76.2 | 79.4 84.5 | 84.9 | -| | | 0.75 | 76.2 81.4 | 69.8 75.3 | 75.0 60.3 | 62.2 | 82.2 | -82.5 -| | EuroSAT | 0.80 | 63.2 76.6 | 73.9 68.1 | 63.2 67.3 | 49.9 74.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -80.6 -| | | 0.88 | 51.0 71.6 | 62.8 49.5 | 54.0 52.9 | 43.5 61.4 | 74.7 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 35.3 46.4 | 46.4 27.5 | 37.9 35.4 | 33.5 44.4 | 54.2 | -| | | 0.50 | 66.9 68.6 | 33.7 31.4 | 66.5 34.7 | 63.5 68.7 | 68.8 | -| | | 0.75 | 60.9 62.9 | 33.5 27.5 | 61.2 26.8 | 58.0 66.4 | | -69.0 -| | Food | 0.80 | 59.2 57.2 | 38.8 24.7 | 59.4 28.8 | 58.9 66.1 | | -| --- | ---- | ---- | --------- | --------- | --------- | --------- | --- | -68.4 -| | | 0.88 | 54.2 57.2 | 34.0 27.7 | 54.4 25.0 | 49.7 65.0 | 68.1 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 49.0 52.6 | 30.8 24.1 | 48.7 24.0 | 48.4 63.8 | 67.1 | -| | | 0.50 | 87.8 89.9 | 39.0 37.3 | 87.9 28.3 | 87.7 95.5 | 95.7 | -| | | 0.75 | 84.8 89.2 | 36.7 29.2 | 84.6 31.1 | 78.9 95.0 | | -95.7 -| | Flowers | 0.80 | 84.7 85.2 | 34.3 30.6 | 85.0 32.7 | 78.3 95.0 | | -| --- | ------- | ---- | --------- | --------- | --------- | --------- | --- | -95.7 -| | | 0.88 | 80.8 76.7 | 38.7 28.6 | 81.1 35.1 | 70.6 94.6 | 95.3 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 76.3 78.5 | 40.1 26.0 | 76.0 25.7 | 76.6 94.2 | 95.0 | -| | | 0.50 | 70.5 71.4 | 39.3 34.3 | 70.3 37.7 | 69.9 71.3 | 74.2 | -| | | 0.75 | 70.3 68.8 | 42.3 31.2 | 65.2 33.0 | 58.3 69.0 | | -73.1 -| | UCF | 0.80 | 65.3 63.3 | 37.9 37.2 | 65.6 34.9 | 59.3 68.5 | | -| --- | --- | ---- | --------- | --------- | --------- | --------- | --- | -73.8 -| | | 0.88 | 49.0 59.0 | 39.3 34.6 | 61.8 32.5 | 60.7 68.0 | 72.3 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 59.3 59.6 | 34.7 24.3 | 59.2 32.5 | 54.3 66.8 | 70.4 | -| | | 0.50 | 28.4 29.6 | 17.2 14.7 | 28.4 12.5 | 27.4 34.9 | 35.8 | -| | | 0.75 | 24.3 26.6 | 13.5 14.3 | 24.2 14.7 | 25.4 32.1 | 32.9 | -| | FGVCAircraft | 0.80 | 23.7 25.1 | 13.7 11.5 | 23.2 11.6 | 24.5 32.2 | | -32.6 -| | | 0.88 | 19.1 21.4 | 13.5 8.3 | 19.6 13.5 | 21.7 28.8 | 28.7 | -| --- | ---------- | ---- | --------- | --------- | --------- | --------- | ---- | -| | | 0.90 | 17.6 22.2 | 13.8 10.5 | 17.8 11.5 | 21.4 28.1 | 26.6 | -| | | 0.50 | 87.8 81.0 | 57.3 45.8 | 71.0 45.1 | 71.7 77.5 | 81.1 | -| | | 0.75 | 84.8 80.3 | 54.0 41.7 | 62.0 37.9 | 64.4 73.9 | 80.5 | -| | OxfordPets | 0.80 | 73.9 | 44.5 38.9 | 56.9 29.5 | 63.4 71.8 | 79.2 | -84.7 -| | | 0.88 | 67.2 | 41.0 34.6 | 44.4 25.5 | 47.6 67.2 | 71.9 | -| --- | --- | ---- | ---- | --------- | --------- | --------- | ---- | -80.8 -| | | 0.90 | 76.3 52.2 | 39.9 30.2 | 38.9 34.2 | 40.9 63.8 | 70.2 | -| --- | --- | ---- | --------- | --------- | --------- | --------- | ---- | - -HopS 23 -B.2 Insd-Confusion Type. -Since partial-label losses perform well under the rand confusion setting, we fur- -ther compare HopS against them under instance-dependent partial labels. The -Insd candidatesets,constructedbasedonvisualsimilarity,arederivedfromfea- -turesextractedbypre-trainedbackbones,includingResNet-18(R18),ResNet-50 -(R50), and CLIP-ResNet50 (CLIP). As shown in Tables 9 and 10, HopS consis- -tently demonstrates significant advantages. -Table 9: Testing accuracy (%) under uni-prompt across two insd confusion rates. -Dataset Insd γc =0.67 γc =0.75 -CC CE EXPGCELWCHopS CC CE EXPGCELWCHopS -CLIP 57.749.2 24.9 13.3 49.5 57.6 41.2 46.8 20.4 9.5 52.3 49.0 -Caltech R18 41.0 54.7 29.2 12.8 54.3 64.1 46.2 45.5 17.1 5.6 44.7 58.2 -R50 28.4 42.2 12.3 8.6 49.5 50.3 31.2 35.9 18.7 9.2 38.7 45.3 -CLIP 39.4 40.4 24.7 25.2 44.7 47.8 28.7 35.5 20.6 12.8 41.6 41.4 -DTD R18 47.3 42.8 19.7 20.0 46.4 48.8 33.7 40.7 12.7 20.8 40.7 47.3 -R50 37.1 42.1 19.1 17.7 44.3 51.7 31.5 37.1 17.3 23.6 40.3 46.2 -CLIP 32.8 37.6 33.6 24.0 36.7 57.5 15.5 21.0 15.3 11.5 27.5 45.8 -EuroSAT R18 34.4 35.3 29.3 35.0 37.3 48.5 23.4 30.1 15.1 17.3 33.5 49.8 -R50 37.2 35.8 32.0 20.0 39.6 52.7 22.0 31.7 27.4 16.4 37.0 37.3 -CLIP 15.7 12.2 8.3 6.2 12.5 17.9 14.911.2 9.3 7.1 12.7 14.0 -FGVCAircraft R18 14.610.0 7.8 5.0 9.5 13.9 13.1 8.3 5.4 6.3 8.9 13.1 -R50 13.7 10.4 7.5 5.1 10.3 14.3 12.8 8.9 4.9 5.0 9.4 12.6 -CLIP 40.2 45.0 23.2 18.8 47.1 54.7 35.8 39.2 15.3 13.5 46.4 48.9 -Food R18 54.4 53.4 17.0 18.2 48.5 59.7 47.9 44.1 13.6 12.3 45.0 59.4 -R50 53.8 47.1 16.1 12.3 50.6 62.5 43.6 35.6 16.7 13.3 46.0 57.0 -CLIP 64.6 32.6 18.4 20.4 48.4 73.9 41.8 40.5 22.0 11.4 37.2 64.8 -Flowers R18 68.5 29.6 21.4 19.6 42.5 82.5 47.5 29.7 20.8 14.5 35.9 75.1 -R50 72.9 44.9 22.3 11.6 37.7 82.0 53.3 21.0 17.8 20.0 36.1 74.0 -CLIP 47.4 40.8 23.0 23.2 47.3 58.5 33.3 37.8 16.5 22.6 41.3 50.3 -OxfordPets R18 49.6 41.3 32.8 18.7 48.5 59.1 42.5 44.3 26.3 12.7 46.1 55.6 -R50 44.1 36.7 25.3 28.0 41.0 53.3 33.8 35.9 18.5 19.2 38.2 51.8 -CLIP 41.2 36.8 17.6 14.3 36.4 49.3 35.0 29.1 14.4 12.9 34.6 38.4 -UCF R18 52.4 41.2 15.4 15.0 49.0 60.9 43.2 32.9 22.2 13.1 40.3 53.5 -R50 48.7 39.4 9.1 14.5 44.1 59.7 40.8 30.1 12.8 9.0 36.0 49.1 -Table 10: Testingaccuracy(%)underuni-promptwithinsd confusionratesγ =0.80. -c -Dataset Insd CC CE EXPGCELWCHopS Dataset CC CE EXPGCELWCHopS -CLIP 46.241.7 4.2 19.6 39.4 45.0 40.631.9 15.4 13.6 38.6 59.2 -Caltech R18 25.4 28.1 6.1 0.5 33.9 49.5 Flowers 36.023.7 15.8 8.8 28.5 69.9 -R50 23.4 37.0 25.4 10.0 31.6 38.9 45.426.5 11.5 12.1 32.1 65.0 -CLIP 25.7 30.9 19.4 7.6 38.1 37.4 36.142.3 22.0 17.3 44.7 48.8 -DTD R18 32.2 31.2 19.1 16.6 37.8 43.1 OxfordPets 35.347.5 27.0 16.0 47.0 55.6 -R50 25.8 32.0 16.7 15.1 34.5 42.5 29.944.6 28.8 18.3 45.0 51.5 -CLIP 33.7 27.2 14.5 15.8 37.8 49.0 15.838.9 11.8 11.4 32.5 43.5 -Food R18 39.041.2 15.4 13.2 42.1 56.5 EuroSAT 14.841.0 12.0 18.9 28.9 40.0 -R50 42.3 43.6 13.2 13.7 44.0 53.7 16.044.4 18.5 9.3 27.0 44.9 -CLIP 27.7 21.9 14.6 10.9 31.3 33.7 11.711.4 8.2 6.7 11.5 13.0 -UCF R18 31.8 32.4 11.6 11.5 37.7 53.2 FGVCAircraft 11.7 8.2 6.3 4.4 8.2 13.7 -R50 27.8 31.3 12.5 7.6 29.2 43.0 11.510.0 5.2 2.3 9.5 13.3 - -| 24 | Y. | Zhao et al. | | | | | -| --- | ------- | ---------------- | --- | --- | --- | --- | -| C | Further | Analysis | | | | | -| C.1 | Mutual | Complementation. | | | | | -To evaluate the interaction between the LDF and GOP modules, we compare -their individual performance with their performance within the overall HopS -model in terms of identifying the ground-truth labels T, considering both the -training and testing phases. Training. Figures 9 – 16 illustrates the identifying -accuracy of LDF, GOP, and HopS over the epochs, highlighting the role of LDF -inacceleratingearly-stageoptimizationbyguidingtheglobalmodule.Moreover, -the proportion of correctly identified labels by each module within HopS ulti- -matelysurpassestheirstandaloneperformance.Testing.Table11demonstrates -that a weighting coefficient of λ = 1.0 achieves a proper balance between local -and global signals, thereby consistently improving performance. Conversely, un- -der high confusion, overemphasizing either component results in performance -| degradation, | | highlighting | the | necessity | of their complementary | interaction. | -| ------------ | ------- | ------------ | -------- | --------- | ---------------------- | --------------- | -| | Fig.9: | Identifying | accuracy | of LDF, | GOP, and HopS | on the Caltech. | -| | Fig.10: | Identifying | accuracy | of | LDF, GOP, and | HopSon the DTD. | -| | | Identifying | accuracy | of LDF, | GOP, and HopS | on the EuroSAT. | -Fig.11: - -HopS 25 -| | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the Food. | -| --- | ----------- | -------- | --- | ------- | -------- | ---- | ------------ | -Fig.12: -| Fig.13: | Identifying | accuracy | of LDF, | GOP, | and HopS | on | the FGVCAircraft. | -| ------- | ----------- | -------- | ------- | --------- | -------- | ------- | ----------------- | -| Fig.14: | Identifying | accuracy | of | LDF, | GOP, and | HopS | on the Flowers. | -| | Identifying | accuracy | of | LDF, GOP, | and | HopS on | the OxfordPets. | -Fig.15: -| Fig.16: | Identifying | accuracy | | of LDF, | GOP, and | HopS | on the UCF. | -| ------- | ----------- | -------- | --- | ------- | -------- | ---- | ----------- | - -| 26 | Y. Zhao | et al. | | | | | | | -| --- | ------- | ----------- | ------- | ---------- | --------- | --- | -------------- | --- | -| | Table | 11: Testing | phases: | (a) module | validity, | (b) | λ sensitivity. | | -.λ -| | 8.98 | 3.65 | 6.43 | | 9.71 2.76 | 5.39 | 3.18 | 0.96 | -| --- | --------- | --------- | --------- | -------------- | --------- | --------- | --------- | -------------- | -| fo | 09.0 4.98 | 2.88 1.45 | 7.35 4.43 | 5.13 2.51 0.41 | 7.36 | 1.46 4.29 | 6.29 0.08 | 2.67 4.76 0.96 | -egnar -| | 88.0 3.09 8.88 | 8.88 2.85 4.85 | 9.65 6.96 5.55 | 3.64 2.02 2.91 | 7.91 2.56 8.66 | 1.46 9.29 2.98 | 8.19 7.97 6.18 | 0.67 3.96 9.76 3.76 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ediw -| | 08.0 5.98 8.98 | 2.98 2.06 9.85 | 9.75 5.47 3.37 | 8.07 4.12 5.12 | 1.62 6.66 9.36 | 9.86 8.19 5.39 | 9.78 7.38 6.48 | 0.48 5.17 8.76 6.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -a -| ssorca | 57.0 1.09 7.09 | 5.98 3.95 3.06 | 3.85 2.08 9.18 | 8.08 9.22 3.32 | 2.42 3.86 5.56 | 2.96 5.39 3.29 | 5.29 7.28 9.97 | 3.38 5.17 9.17 9.17 | -| ------ | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | 05.0 6.98 2.98 | 4.98 2.36 9.26 | 2.26 7.28 5.28 | 2.28 4.72 4.52 | 4.72 6.96 9.96 | 4.86 2.39 1.09 | 8.09 4.38 5.38 | 8.28 1.17 4.17 0.17 | -)%( -| ycarucca | 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 | 5.0 0.1 0.2 5.0 | -| -------- | ------- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- | --------------- | -λ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| --- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| | | DTD | | | dooF | | | FCU | -gnitseT -)b( -.)thgir(slc -| | FDL 9.88 8.88 | 5.88 0.85 2.55 | 8.65 5.67 8.66 | 3.82 8.92 9.52 | 6.32 8.56 9.26 | 3.36 8.49 9.39 | 8.39 7.97 2.67 | 7.26 2.07 4.96 0.07 | -| --- | ------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -| | POG 0.19 | 8.09 | 8.85 8.08 | | 9.86 7.86 | 5.76 5.59 1.59 | 4.08 1.97 | 7.77 1.17 | -| | 5.09 | 8.95 5.85 | 7.57 | 5.24 8.33 1.13 | 3.42 | | 0.59 | 4.27 5.17 | -dna -| | SpoH 7.09 3.09 | 2.09 7.16 9.85 | 3.85 9.87 0.97 | 9.25 9.43 1.23 | 3.82 8.86 4.86 | 6.66 3.59 1.59 | 3.59 8.97 1.87 | 4.86 8.27 3.27 0.17 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)tfel(inu -| | FDL | | | 0.72 | 0.76 | 6.29 | 6.18 | 7.17 | -| --- | --------- | -------------- | -------------- | --------- | --------- | --------- | --------- | -------------- | -| | 8.88 0.88 | 9.68 5.95 2.55 | 5.45 0.47 1.45 | 6.81 4.12 | 5.71 3.66 | 4.26 3.29 | 5.19 8.18 | 1.17 0.86 5.76 | -eht -| | POG 7.98 8.88 | 9.65 0.06 3.65 | 3.65 5.87 9.47 | 5.33 9.91 1.11 | 6.16 3.35 | 4.31 2.19 4.58 | 5.92 5.38 0.18 | 7.18 8.96 4.76 8.11 | -| --- | ------------- | -------------- | -------------- | -------------- | --------- | -------------- | -------------- | ------------------- | -0.3 -rednu -| | SpoH 6.88 0.98 | 7.88 1.95 0.95 | 0.06 2.77 3.57 | 1.44 7.32 2.22 | 3.81 5.76 4.66 | 9.36 5.29 4.29 | 2.39 2.38 6.18 | 6.67 3.07 9.96 0.96 | -| --- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -)%( -| | 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 | 98.0 76.0 08.0 98.0 | -| --- | --------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | ------------------- | -ycarucca c -γ -tfarcriACVGF -| | | | TASoruE | | | | stePdrofxO | | -| ------- | --------------- | --- | ------- | --- | ---- | ------- | ---------- | --- | -| | tesataD hcetlaC | | | | | srewolF | | | -| gnitseT | | DTD | | | dooF | | | FCU | -)a( - -HopS 27 -C.2 Effect of Batch Size. -We investigate the effect of varying batch sizes B ∈{16,32,64,128,256} on the -performance of HopS, focusing on the testing accuracy under varying levels of -label confusion γ ∈ {0.5,0.75,0.80,0.88,0.90}. As shown in Figure 17 and Ta- -c -ble12,theresultscomparetheperformanceunderuni-promptandcls-prompts. -Fig.17: Testing accuracy of the HopS across different B under five confusion rates. - -| 28 | Y. Zhao | et al. | | | | | | -| ----- | ----------- | -------- | ---- | ------------- | ---------- | ------- | ---------------- | -| Table | 12: Testing | accuracy | (%) | of HopS under | uni (left) | and cls | (right) prompts. | -| | Dataset | | γ 16 | 32 64 128 | 256 | 16 32 | 64 128 256 | -c -| | | 0.50 | 89.8 89.4 | 89.4 87.5 | 84.7 90.2 | 90.9 90.7 | 90.8 90.7 | -| --- | ------------ | ---- | --------- | --------- | --------- | --------- | --------- | -| | | 0.75 | 88.9 89.0 | 88.7 87.7 | 85.3 89.6 | 90.1 90.6 | 91.0 90.8 | -| | Caltech | 0.80 | 88.6 89.2 | 89.6 86.6 | 84.6 90.0 | 90.6 90.6 | 90.8 90.9 | -| | | 0.88 | 88.7 89.2 | 88.1 88.0 | 85.7 89.6 | 90.0 90.3 | 90.2 90.2 | -| | | 0.90 | 89.3 89.2 | 89.0 86.7 | 85.5 89.2 | 89.7 90.0 | 90.4 90.5 | -| | | 0.50 | 61.2 61.0 | 60.5 60.9 | 58.0 61.2 | 61.1 61.6 | 61.4 61.8 | -| | | 0.75 | 60.4 58.9 | 58.9 58.9 | 57.3 58.9 | 60.7 59.7 | 59.3 59.3 | -| | EuroSAT | 0.80 | 57.6 58.0 | 58.4 57.1 | 54.9 58.3 | 59.0 57.9 | 58.1 58.6 | -| | | 0.88 | 55.9 55.8 | 56.8 56.3 | 54.1 55.8 | 56.4 56.7 | 56.9 57.1 | -| | | 0.90 | 52.4 53.3 | 52.9 54.0 | 54.1 51.0 | 51.8 52.3 | 52.1 53.1 | -| | | 0.50 | 83.6 82.1 | 81.0 78.5 | 79.8 83.8 | 83.6 83.8 | 84.2 84.4 | -| | | 0.75 | 79.0 80.4 | 77.2 75.3 | 75.7 80.2 | 80.5 81.0 | 80.5 81.3 | -| | UCF | 0.80 | 77.2 76.2 | 71.5 71.4 | 74.5 78.0 | 78.3 78.0 | 77.5 78.1 | -| | | 0.88 | 52.7 61.9 | 47.7 33.1 | 51.7 66.2 | 64.0 66.4 | 53.1 38.1 | -| | | 0.90 | 45.2 46.2 | 26.5 23.6 | 28.7 41.8 | 42.3 44.5 | 28.9 33.3 | -| | | 0.50 | 26.5 26.2 | 25.8 18.8 | 12.0 35.2 | 35.4 34.9 | 34.9 34.8 | -| | | 0.75 | 24.6 23.3 | 22.3 17.4 | 10.7 32.4 | 32.5 32.9 | 32.8 32.3 | -| | Flowers | 0.80 | 23.4 22.6 | 22.5 16.8 | 8.6 31.7 | 32.2 31.9 | 31.9 31.8 | -| | | 0.88 | 18.5 19.1 | 17.6 14.4 | 9.9 29.0 | 28.4 29.0 | 29.1 27.9 | -| | | 0.90 | 17.8 18.1 | 16.7 15.4 | 8.9 26.6 | 26.4 25.8 | 26.7 26.3 | -| | | 0.50 | 67.5 66.8 | 65.4 63.8 | 49.3 68.1 | 68.6 69.2 | 68.9 69.1 | -| | | 0.75 | 67.0 67.1 | 66.6 62.8 | 48.4 67.9 | 68.2 68.6 | 68.9 68.7 | -| | Food | 0.80 | 66.2 67.6 | 66.4 52.8 | 56.0 67.6 | 68.3 68.9 | 68.8 68.8 | -| | | 0.88 | 65.7 65.2 | 65.0 61.7 | 56.9 66.4 | 67.6 68.0 | 67.9 67.8 | -| | | 0.90 | 63.0 63.2 | 62.6 58.6 | 52.0 66.1 | 66.6 67.2 | 67.6 67.1 | -| | | 0.50 | 93.1 92.9 | 92.7 89.7 | 70.0 95.6 | 95.6 95.7 | 95.5 95.5 | -| | | 0.75 | 92.6 92.5 | 92.4 90.2 | 77.0 95.4 | 95.5 95.7 | 95.4 95.4 | -| | OxfordPets | 0.80 | 92.9 91.1 | 92.4 88.4 | 53.3 95.1 | 95.3 95.4 | 95.4 95.1 | -| | | 0.88 | 92.9 91.8 | 92.3 87.7 | 68.1 94.9 | 95.1 95.4 | 95.3 95.2 | -| | | 0.90 | 91.1 92.2 | 89.1 89.4 | 72.8 94.7 | 94.8 95.1 | 94.8 94.8 | -| | | 0.50 | 82.9 82.9 | 81.5 80.3 | 80.4 80.5 | 81.0 80.2 | 80.8 81.3 | -| | | 0.75 | 83.0 82.6 | 81.4 82.0 | 74.1 79.4 | 79.5 80.1 | 80.7 80.8 | -| | DTD | 0.80 | 82.3 82.2 | 82.5 81.9 | 81.1 77.9 | 78.3 79.3 | 79.5 79.4 | -| | | 0.88 | 74.7 78.3 | 79.6 79.4 | 76.4 70.2 | 71.7 72.3 | 73.2 73.8 | -| | | 0.90 | 76.3 77.5 | 76.5 77.8 | 76.8 68.5 | 69.2 70.6 | 72.3 72.9 | -| | | 0.50 | 70.9 70.9 | 71.4 67.6 | 63.0 72.4 | 73.3 73.3 | 73.4 73.2 | -| | | 0.75 | 70.6 71.9 | 71.2 66.2 | 63.7 72.8 | 72.7 72.3 | 73.3 72.8 | -| | FGVCAircraft | 0.80 | 71.0 70.4 | 69.1 67.2 | 62.1 71.8 | 72.3 72.4 | 72.1 72.3 | -| | | 0.88 | 70.9 70.3 | 69.5 65.9 | 64.7 71.3 | 71.9 72.0 | 71.7 72.7 | -| | | 0.90 | 69.0 68.8 | 68.5 66.0 | 64.0 70.0 | 70.3 70.8 | 71.4 70.5 | - -HopS 29 -C.3 Comparison with Full-Data PLL Methods. -Weconductedasystematiccomparisonwithtworecentlyproposedpartial-label -learning methods, Papi and CroSel. Unlike the 16-shot setting used in previous -experiments, Papi and CroSel were trained on the entire datasets to fully ex- -ploit their optimal performance, while HopS was still trained under the 16-shot -few-shot setting. The experiments were performed on two challenging datasets, -CaltechandOxfordFlowers,bothcharacterizedbyalargenumberofcategories -and high intra-class similarity. The results are shown in Table 13. -Table 13: Testing accuracy (%) of HopS under uni-prompt across three backbones. -CLIP R18 R50 -Dataset Method -0.67 0.75 0.80 0.67 0.75 0.80 0.67 0.75 0.80 -HopS 57.6 49.0 45.0 64.1 58.2 49.5 50.3 45.3 38.9 -Caltech CroSel 60.4 43.7 32.2 42.8 33.3 18.8 39.0 25.2 11.8 -Papi 59.7 49.2 40.5 44.3 35.2 15.8 42.0 23.5 13.4 -HopS 73.9 64.8 59.2 82.5 75.1 69.9 82.0 74.0 65.0 -Flowers CroSel 40.4 27.4 11.5 75.0 56.4 45.1 78.0 62.0 50.2 -Papi 35.7 13.7 5.8 67.9 56.1 47.0 67.0 58.7 49.8 -C.4 Settings with Missing Ground-Truth. -Weconduct16-shotexperimentsontwotypesoflabelconfusion,rand andinsd, -as previously described, with each candidate set S containing three labels. The -missing rates of ground-truth labels are 12.5% (2 out of 16 shots) for rand and -25% (4 out of 16 shots) for insd. As shown in Table 14, under the insd setting, -HopS consistently uncovers meaningful correlations within the candidate label -sets, leading to significantly better performance than other methods. This sug- -gests that HopS is particularly well-suited for real-world scenarios where label -noiseinSSisinstance-dependent.Incontrast,undertherand settingwheresuch -dependencies are absent, HopS performs slightly worse than the SOTA robust -loss. Nevertheless, its performance can be improved by integrating robust loss -functions(e.g.,theSCEloss).AsillustratedinTable15,HopScombinedwiththe -SCE loss achieves the best performance under the rand setting, demonstrating -its flexibility and compatibility with other robust learning techniques. -Table 14: Testing accuracy under the uni-prompt with 25.0% missing rate (insd). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.37 0.35 0.16 0.27 0.12 0.27 0.06 0.33 0.37 0.31 -DTD 0.37 0.35 0.19 0.40 0.22 0.31 0.13 0.39 0.41 0.32 -Flowers 0.29 0.45 0.17 0.35 0.13 0.35 0.14 0.51 0.57 0.34 -UCF 0.31 0.41 0.11 0.31 0.11 0.28 0.14 0.45 0.46 0.30 - -30 Y. Zhao et al. -Table 15: Testing accuracy under the uni-prompt with 12.5% missing rate (rand). -Dataset RC CC EXP LWC MAE MSE GCE SCE HopS HopS+SCE -Caltech 0.76 0.84 0.34 0.84 0.38 0.20 0.26 0.86 0.84 0.87 -DTD 0.52 0.51 0.19 0.53 0.23 0.19 0.15 0.54 0.53 0.55 -Flowers 0.53 0.87 0.18 0.73 0.18 0.13 0.15 0.87 0.88 0.89 -UCF 0.50 0.63 0.18 0.57 0.14 0.08 0.10 0.64 0.64 0.64 -C.5 Running Time of Methods. -We measured the computational cost of each method during training, including -the average epoch time (in seconds) and the total training time (in minutes), -as shown in Table 16. Notably, although HopS introduces both a local module -(LDF)andaglobalmodule(GOP)toenhancelabelidentification,theadditional -computational overhead remains negligible due to the relatively small dataset -size. As a result, HopS achieves superior performance without incurring a sig- -nificant increase in training cost. -Table 16: Training efficiency comparison of different methods on Caltech. -Metric RC CC EXP GCE LWC MAE MSE SCE HopS -Average Epoch Time (s) 0.11 0.09 0.12 0.11 0.11 0.12 0.11 0.11 0.12 -Total Time (min) 19.01 14.19 20.75 18.79 18.45 20.09 18.67 18.65 19.85 -C.6 Memory usage of large datasets. -As shown in Table 17, we compare the memory usage (in MiB) of HopS and -Coop at different batch sizes on ImageNet, clearly showing that the memory -costofHopSisonlyslightlyhigherthanthatofCoop.HopSdoesnotrequirean -N2 dense affinity matrix. Instead, the kNN graph is computed sparsely for each -batch, and the features come from a frozen encoder, allowing us to compute the -kNN graph offline. Thus, HopS remains efficient and scalable even on non-few- -shot PLL datasets, due to the sparse computation in its LDF module. -Table 17: Memory usage comparison at different batch sizes on ImageNet -Method 16 32 64 128 256 512 1024 2048 -Coop 18130 18102 18170 18312 19998 22178 19098 19564 -HopS 18582 18582 18622 18764 20456 23312 19576 20430 -△ + 452 + 480 + 452 + 452 + 458 + 1134 + 478 + 866 - -| | | | | | | | HopS | 31 | -| --- | ----------- | -------- | --- | --- | --- | --- | ---- | --- | -| C.7 | Sensitivity | to k and | τ. | | | | | | -We evaluate the sensitivity of LDF by sweeping k ∈ {10,20,40} (with τ = 0.4) -and τ ∈ {0.4,0.5,0.6} (with k = 20) on coarse/fine-grained datasets (Cal- -tech/Flowers), under rand and insd confusion types, using both uni-prompt -(left)andcls-prompts(right).AsshowninTable18,performancevariessmoothly -| with k | and τ, with | no significant | re-tuning | needed | from | Caltech | to Flowers. | | -| ------ | ----------- | -------------- | ----------- | ------ | ---- | ------- | ----------- | --- | -| | | | Sensitivity | of LDF | to | and τ. | | | -| | | Table | 18: | | | k | | | -Dataset Type k=10 k=20 k=40 τ=0.5 τ=0.6 k=10 k=20 k=40 τ=0.5 τ=0.6 -| | CLIP 48.9 | 49.0 | 45.6 | 56.3 53.1 | 52.4 | 60.0 | 48.8 56.8 | 54.0 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | R18 45.3 | 58.2 | 47.9 | 51.4 49.9 | 54.5 | 59.4 | 53.1 65.0 | 62.4 | -Caltech -| | R50 41.7 | 45.3 | 40.0 | 47.3 48.8 | 59.9 | 53.3 | 49.3 50.1 | 54.1 | -| --- | --------- | ---- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 89.4 | 89.0 | 88.4 | 89.4 89.0 | 90.0 | 90.4 | 90.1 90.1 | 90.0 | -| | CLIP 63.1 | 64.8 | 60.6 | 65.8 66.0 | 72.4 | 68.8 | 61.8 69.6 | 68.8 | -| | R18 73.4 | 75.1 | 74.0 | 78.1 76.2 | 83.8 | 80.4 | 74.6 80.0 | 79.9 | -Flowers -| | R50 76.2 | 74.0 | 80.1 | 73.5 74.4 | 80.5 | 78.2 | 81.0 78.1 | 77.9 | -| --- | ----------- | --------- | ---- | --------- | ---- | ---- | --------- | ---- | -| | Rand 91.8 | 92.9 | 92.5 | 91.9 92.0 | 95.5 | 95.7 | 95.5 95.6 | 95.6 | -| C.8 | Settings of | 4/8-Shot. | | | | | | | -As clarified in Figure 18, we additionally conducted 4-shot and 8-shot experi- -ments to evaluate HopS under more limited supervision. The results show that -HopS consistently outperforms the baselines across all eight datasets, further -| demonstrating | its effectiveness | | in low-shot | partial-label | | learning | scenarios. | | -| ------------- | ------------------ | ------- | ----------- | ------------- | -------- | -------- | -------------- | --- | -| | Fig.18: Comparison | results | under | 4/8-shot | settings | across | eight datasets | | \ No newline at end of file + +L + +I + +F + +C + +NOTE: Harm is defined as a situation that compromises a resident’s ability to maintain or reach his or +her highest practicable physical, mental, or psychosocial well being, as defined by an accurate and +comprehensive assessment, care plan, and provision of services. A nursing home with one or more +quality of life, quality of care, or resident behavior and facility practices deficiencies issued at level “F” +or “H” or above (the shaded area of the grid) is considered to be providing “substandard” care to its +residents. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), Appendix P, V, B-C. + +To determine a deficiency’s scope, inspectors must classify each deficiency in one +of three ways: isolated, pattern, or widespread. Federal guidelines say that a +deficiency is isolated when one or a very limited number of residents or staff are +affected or the situation has occurred only occasionally or in a very limited +number of locations in the facility. For example, if 60 of 70 residents in a facility +are incontinent and the facility failed to provide adequate care or services to +restore or improve bladder function for 2 of these residents, the deficiency should +be classified as isolated. A deficiency represents a pattern when it affects more + +15 However, a single medication error that is considered severe enough may result in a deficiency. + + 10 + +NURSING HOME INSPECTIONS + +than a very limited number of residents or staff, occurs in several locations, or the +same resident has been affected by repeated occurrences of the same deficient +practice. If the above facility did not provide adequate care or services to 10 of its +60 incontinent residents, the resulting deficiency should be issued as a pattern. A +deficiency is identified as widespread when it refers to the entire facility or when +a system failure has affected or has the potential to affect a large number of +residents. For example, a facility failing to provide adequate care or services to +improve or restore bladder function to 30 of its 60 incontinent residents should be +issued a deficiency classified as widespread. + +Inspectors must also determine the severity of a deficiency on a scale from one to +four. Level one refers to deficiencies that have the potential for causing no more +than a minor negative impact on, or minimal physical, mental, or psychosocial +discomfort to, a resident. For example, a facility should receive a level one +deficiency if it failed to post its inspection results or only made them available +upon request. Level two deficiencies are those that have resulted in resident +discomfort or have the potential to harm residents. Federal regulations define +harmful situations as those that compromise residents’ ability to maintain or reach +their highest practicable physical, mental, and psychosocial well being, excluding +situations that are of a “limited consequence” to residents. For example, a nursing +home should receive a level two deficiency if inspectors observed staff failing to +wash their hands properly between caring for residents but no one became +seriously ill as a result. Level three deficiencies are those that have actually +resulted in resident harm. The hand-washing example should be a level three +deficiency if there was evidence that a resident caught a contagious disease as a +result of staff failing to wash their hands properly after providing resident care. +Level four represents immediate jeopardy situations whereby the facility must +undertake immediate corrective action to address problems that have resulted in or +are likely to cause serious injury, harm, impairment, or death to a resident. For +example, if a resident with dementia was found outside during an inspection +heading toward a busy highway and the nursing home did not have a working +system in place to monitor residents with dementia, the facility should be issued a +level four deficiency. + +The “seriousness” of a facility’s deficiencies (their scope and severity) helps +determine the sanctions for nursing homes that fail to correct deficiencies within +an allowable time frame. As shown in Table 1.3, there are three categories of +required sanctions. Generally, nursing homes do not face sanctions for +deficiencies issued at levels “A” through “C.”16 Category 1 sanctions are reserved +for deficiencies issued at levels “D” and “E” and require that facilities implement +a plan of correction developed by the state, have their staff attend a specific +training program, or be subject to state monitoring. Conversely, category 3 +sanctions are reserved for the most serious deficiencies and include the state +assuming management of the facility, terminating the facility’s participation in the +Medicare and Medicaid programs, or closing the facility. Except in instances of +immediate jeopardy to residents (a deficiency issued at level “J” or above) or +when facilities receive level “G” or higher deficiencies in two consecutive +inspections, facilities are generally given an opportunity to correct deficiencies +before any sanctions are imposed—usually 40 days. MDH must deny Medicare + +16 Although the federal government does not require that sanctions be imposed on facilities +for low-level deficiencies (levels “B” and “C”), the state may choose to impose sanctions from +category 1 when facilities fail to correct their deficiencies. + +A deficiency's +letter code helps +determine what +sanctions MDH +could impose on +the facility. + + THE NURSING HOME INSPECTION PROCESS + +11 + +Table 1.3: Required Sanctions for Noncompliance + +Category 1: Deficiencies issued at levels “D” and “E” + +Directed plan of correction; +State monitoring; and/or +Directed in-service training. + +Category 2: Deficiencies issued at levels “F” through “I” + +Denial of payment for new Medicare and Medicaid admissionsa; +Denial of payment for all Medicare and Medicaid residents; +Civil money penalties of $50-$3,000 per day of noncompliance; and/or +Civil money penalties of $1,000-$10,000 per incident of noncompliance. + +Category 3: Deficiencies issued at levels “J” and above + +Temporary management; +Termination from the Medicare/Medicaid programs; and/orb +Facility closure. + +NOTE: The Minnesota Department of Health may impose a category 2 sanction to supplement a +category 1 sanction for deficiencies issued at levels “D” and “E.” In general, a category 1 or 2 sanction +can also be imposed whenever a category 3 sanction is required, and a category 1 sanction may also +be imposed when a category 2 sanction is required. Civil penalties increase to $3,050-$10,000 per +day when they are imposed in addition to a category 3 sanction. The state may also assume +temporary management (a category 3 sanction) when a facility has been issued a level “I” deficiency. +Also, a facility cited for providing substandard care cannot operate a nurse aide training and +competency evaluation program for two years. + +aThe state must deny Medicare and Medicaid payments for new admissions when a facility is not in +substantial compliance within three months of the inspection and when a facility has been cited for +substandard care on three consecutive annual inspections. In the latter situation, state monitoring +must also be imposed. + +bThe state must recommend termination from the Medicare and Medicaid programs when a facility is +not in substantial compliance within six months of the inspection. + +SOURCE: Centers for Medicare and Medicaid Services, State Operations Manual (Washington, DC, +May 21, 2004), ch. 7, sec. 7210G and 7400. + +and Medicaid reimbursements for new admissions when facilities have not +corrected their deficiencies within three months of the department’s inspection. +Facilities must be terminated from the program if deficiencies are not corrected +within 6 months. + +FUNDING + +In keeping with the high degree of federal involvement in the nursing home +inspection program: + +• + +State funds cover less than 10 percent of the total cost of nursing home +inspections and complaint investigations. + +The federal government is the major source of funding for the inspection +program, with the state contributing less than 10 percent of the total cost for +nursing homes. In fiscal year 2004, MDH spent about $12 million from state and + + 12 + +NURSING HOME INSPECTIONS + +federal sources on activities related to nursing home inspections, including costs +related to investigating complaints against nursing homes.17 The state’s share +(about $1.1 million) is the result of state negotiations with CMS and has +historically been low when compared with that of other states. According to a +2000 analysis of costs by the Health Care Financing Administration, Minnesota +was the only state in the Chicago region that paid less than 10 percent of total +inspection costs.18 Other states paid at least 16 percent, with one state paying +almost 25 percent of total costs. + +17 Cecelia Jackson, “Re: FFY 2004 Nursing Home Expenditures” (December 23, 2004), electronic +mail to jo.vos@state.mn.us. +18 Health Care Financing Administration, “Nursing Home Survey, State Licensure Cost Shares” +(Chicago, May 2000). Minnesota is part of the Chicago region, which also includes Illinois, +Indiana, Michigan, Ohio, and Wisconsin. + diff --git a/q144/random_k4/random_2.md b/q144/random_k4/random_2.md index 69d1305f3a9c9c9fb312a4e7dbd3718670cacdd2..886a7a04c5cccafe307d041d698e8684ec95d131 100644 --- a/q144/random_k4/random_2.md +++ b/q144/random_k4/random_2.md @@ -1,2203 +1,490 @@ -3 -2 -0 -2 +Genetics, DNA, and Heredity -r -a - -M -7 - -] -L -C -. -s -c -[ - -4 -v -1 -7 -1 -1 -1 -. -3 -0 -2 -2 -: -v -i -X -r -a - -Published as a conference paper at ICLR 2023 - -SELF-CONSISTENCY IMPROVES CHAIN OF THOUGHT -REASONING IN LANGUAGE MODELS - -Jason Wei† Dale Schuurmans† Quoc Le† Ed H. Chi† - -Xuezhi Wang†‡ -Sharan Narang† Aakanksha Chowdhery† Denny Zhou†§ -†Google Research, Brain Team -‡xuezhiw@google.com, §dennyzhou@google.com - -ABSTRACT - -Chain-of-thought prompting combined with pre-trained large language models has -achieved encouraging results on complex reasoning tasks. In this paper, we propose -a new decoding strategy, self-consistency, to replace the naive greedy decoding -used in chain-of-thought prompting. It first samples a diverse set of reasoning paths -instead of only taking the greedy one, and then selects the most consistent answer -by marginalizing out the sampled reasoning paths. Self-consistency leverages the -intuition that a complex reasoning problem typically admits multiple different ways -of thinking leading to its unique correct answer. Our extensive empirical evaluation -shows that self-consistency boosts the performance of chain-of-thought prompting -with a striking margin on a range of popular arithmetic and commonsense reasoning -benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), -StrategyQA (+6.4%) and ARC-challenge (+3.9%). - -1 - -INTRODUCTION - -Although language models have demonstrated remarkable success across a range of NLP tasks, their -ability to demonstrate reasoning is often seen as a limitation, which cannot be overcome solely by -increasing model scale (Rae et al., 2021; BIG-bench collaboration, 2021, inter alia). In an effort -to address this shortcoming, Wei et al. (2022) have proposed chain-of-thought prompting, where -a language model is prompted to generate a series of short sentences that mimic the reasoning -process a person might employ in solving a task. For example, given the question “If there are 3 -cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?”, instead -of directly responding with “5”, a language model would be prompted to respond with the entire -chain-of-thought: “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars. The answer is 5.”. It has been observed that chain-of-thought prompting significantly -improves model performance across a variety of multi-step reasoning tasks (Wei et al., 2022). - -In this paper, we introduce a novel decoding strategy called self-consistency to replace the greedy -decoding strategy used in chain-of-thought prompting (Wei et al., 2022), that further improves -language models’ reasoning performance by a significant margin. Self-consistency leverages the -intuition that complex reasoning tasks typically admit multiple reasoning paths that reach a correct -answer (Stanovich & West, 2000). The more that deliberate thinking and analysis is required for a -problem (Evans, 2010), the greater the diversity of reasoning paths that can recover the answer. - -Figure 1 illustrates the self-consistency method with an example. We first prompt the language model -with chain-of-thought prompting, then instead of greedily decoding the optimal reasoning path, we -propose a “sample-and-marginalize” decoding procedure: we first sample from the language model’s -decoder to generate a diverse set of reasoning paths; each reasoning path might lead to a different -final answer, so we determine the optimal answer by marginalizing out the sampled reasoning paths -to find the most consistent answer in the final answer set. Such an approach is analogous to the -human experience that if multiple different ways of thinking lead to the same answer, one has greater -confidence that the final answer is correct. Compared to other decoding methods, self-consistency -avoids the repetitiveness and local-optimality that plague greedy decoding, while mitigating the -stochasticity of a single sampled generation. - -1 - - Published as a conference paper at ICLR 2023 - -Figure 1: The self-consistency method contains three steps: (1) prompt a language model using -chain-of-thought (CoT) prompting; (2) replace the “greedy decode” in CoT prompting by sampling -from the language model’s decoder to generate a diverse set of reasoning paths; and (3) marginalize -out the reasoning paths and aggregate by choosing the most consistent answer in the final answer set. - -Self-consistency is far simpler than prior approaches that either train an additional verifier (Cobbe -et al., 2021) or train a re-ranker given additional human annotations to improve generation quality -(Thoppilan et al., 2022). Instead, self-consistency is entirely unsupervised, works off-the-shelf with -pre-trained language models, requires no additional human annotation, and avoids any additional -training, auxiliary models or fine-tuning. Self-consistency also differs from a typical ensemble -approach where multiple models are trained and the outputs from each model are aggregated, it acts -more like a “self-ensemble” that works on top of a single language model. - -We evaluate self-consistency on a wide range of arithmetic and commonsense reasoning tasks over -four language models with varying scales: the public UL2-20B (Tay et al., 2022) and GPT-3-175B -(Brown et al., 2020), and two densely-activated decoder-only language models: LaMDA-137B -(Thoppilan et al., 2022) and PaLM-540B (Chowdhery et al., 2022). On all four language models, -self-consistency improves over chain-of-thought prompting by a striking margin across all tasks. In -particular, when used with PaLM-540B or GPT-3, self-consistency achieves new state-of-the-art levels -of performance across arithmetic reasoning tasks, including GSM8K (Cobbe et al., 2021) (+17.9% -absolute accuracy gains), SVAMP (Patel et al., 2021) (+11.0%), AQuA (Ling et al., 2017) (+12.2%), -and across commonsense reasoning tasks such as StrategyQA (Geva et al., 2021) (+6.4%) and ARC- -challenge (Clark et al., 2018) (+3.9%). In additional experiments, we show self-consistency can -robustly boost performance on NLP tasks where adding a chain-of-thought might hurt performance -compared to standard prompting (Ye & Durrett, 2022). We also show self-consistency significantly -outperforms sample-and-rank, beam search, ensemble-based approaches, and is robust to sampling -strategies and imperfect prompts. - -2 SELF-CONSISTENCY OVER DIVERSE REASONING PATHS - -A salient aspect of humanity is that people think differently. It is natural to suppose that in tasks -requiring deliberate thinking, there are likely several ways to attack the problem. We propose that -such a process can be simulated in language models via sampling from the language model’s decoder. -For instance, as shown in Figure 1, a model can generate several plausible responses to a math -question that all arrive at the same correct answer (Outputs 1 and 3). Since language models are not -perfect reasoners, the model might also produce an incorrect reasoning path or make a mistake in -one of the reasoning steps (e.g., in Output 2), but such solutions are less likely to arrive at the same -answer. That is, we hypothesize that correct reasoning processes, even if they are diverse, tend to -have greater agreement in their final answer than incorrect processes. - -We leverage this intuition by proposing the following self-consistency method. First, a language -model is prompted with a set of manually written chain-of-thought exemplars (Wei et al., 2022). Next, - -2 - -Language modelQ: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.…Q: Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder for $2 per egg. How much does she make every day?A:She has 16 - 3 - 4 = 9 eggs left. So she makes $2 * 9 = $18 per day. Sample a diverse set of reasoning pathsShe eats 3 for breakfast, so she has 16 - 3 = 13 left. Then she bakes muffins, so she has 13 - 4 = 9 eggs left. So she has 9 eggs * $2 = $18. This means she she sells the remainder for $2 * (16 - 4 - 3) = $26 per day. The answer is $18.The answer is $26.The answer is $18.The answer is $18.Marginalize out reasoning paths to aggregate final answersLanguage modelThis means she uses 3 + 4 = 7 eggs every day. She sells the remainder for $2 per egg, so in total she sells 7 * $2 = $14 per day. The answer is $14.The answer is $14.Greedy decode Published as a conference paper at ICLR 2023 - -GSM8K MultiArith - -AQuA - -SVAMP - -CSQA - -ARC-c - -Greedy decode - -56.5 - -94.7 - -35.8 - -79.0 - -79.0 - -85.2 - -Weighted avg (unnormalized) -Weighted avg (normalized) - -56.3 ± 0.0 -22.1 ± 0.0 - -90.5 ± 0.0 -59.7 ± 0.0 - -35.8 ± 0.0 73.0 ± 0.0 74.8 ± 0.0 82.3 ± 0.0 -15.7 ± 0.0 40.5 ± 0.0 52.1 ± 0.0 51.7 ± 0.0 - -Weighted sum (unnormalized) -Weighted sum (normalized) - -59.9 ± 0.0 -74.1 ± 0.0 - -92.2 ± 0.0 -99.3 ± 0.0 - -38.2 ± 0.0 76.2 ± 0.0 76.2 ± 0.0 83.5 ± 0.0 -48.0 ± 0.0 86.8 ± 0.0 80.7 ± 0.0 88.7 ± 0.0 - -Unweighted sum (majority vote) 74.4 ± 0.1 - -99.3 ± 0.0 - -48.3 ± 0.5 86.6 ± 0.1 80.7 ± 0.1 88.7 ± 0.1 - -Table 1: Accuracy comparison of different answer aggregation strategies on PaLM-540B. - -we sample a set of candidate outputs from the language model’s decoder, generating a diverse set of -candidate reasoning paths. Self-consistency is compatible with most existing sampling algorithms, -including temperature sampling (Ackley et al., 1985; Ficler & Goldberg, 2017), top-k sampling (Fan -et al., 2018; Holtzman et al., 2018; Radford et al., 2019), and nucleus sampling (Holtzman et al., -2020). Finally, we aggregate the answers by marginalizing out the sampled reasoning paths and -choosing the answer that is the most consistent among the generated answers. -In more detail, assume the generated answers ai are from a fixed answer set, ai ∈ A, where -i = 1, . . . , m indexes the m candidate outputs sampled from the decoder. Given a prompt and a -question, self-consistency introduces an additional latent variable ri, which is a sequence of tokens -representing the reasoning path in the i-th output, then couples the generation of (ri, ai) where -ri → ai, i.e., generating a reasoning path ri is optional and only used to reach the final answer ai. As -an example, consider Output 3 from Figure 1: the first few sentences “She eats 3 for breakfast ... So -she has 9 eggs * $2 = $18.” constitutes ri, while the answer 18 from the last sentence, “The answer -is $18”, is parsed as ai.1 After sampling multiple (ri, ai) from the model’s decoder, self-consistency -1(ai = a), -applies a marginalization over ri by taking a majority vote over ai, i.e., arg maxa -or as we defined as the most “consistent” answer among the final answer set. - -(cid:80)m - -i=1 - -In Table 1, we show the test accuracy over a set of reasoning tasks by using different answer -aggregation strategies. In addition to majority vote, one can also weight each (ri, ai) by P (ri, ai | -prompt, question) when aggregating the answers. Note to compute P (ri, ai | prompt, question), we -can either take the unnormalized probability of the model generating (ri, ai) given (prompt, question), -or we can normalize the conditional probability by the output length (Brown et al., 2020), i.e., - -P (ri, ai | prompt, question) = exp - -1 -K - -(cid:80)K - -k=1 log P (tk|prompt,question,t1,...,tk−1), - -(1) - -where log P (tk | prompt, question, t1, . . . , tk−1) is the log probability of generating the k-th token -tk in (ri, ai) conditioned on the previous tokens, and K is the total number of tokens in (ri, ai). -In Table 1, we show that taking the “unweighted sum”, i.e., taking a majority vote directly over ai -yields a very similar accuracy as aggregating using the “normalized weighted sum”. We took a closer -look at the model’s output probabilities and found this is because for each (ri, ai), the normalized -conditional probabilities P (ri, ai | prompt, question) are quite close to each other, i.e., the language -model regards those generations as “similarly likely”.2 Additionally, when aggregating the answers, -the results in Table 1 show that the “normalized” weighted sum (i.e., Equation 1) yields a much -higher accuracy compared to its unnormalized counterpart. For completeness, in Table 1 we also -report the results by taking a “weighted average”, i.e., each a gets a score of its weighted sum divided -by (cid:80)m -i=1 - -1(ai = a), which results in a much worse performance. - -Self-consistency explores an interesting space between open-ended text generation and optimal -text generation with a fixed answer. Reasoning tasks typically have fixed answers, which is why -researchers have generally considered greedy decoding approaches (Radford et al., 2019; Wei et al., -2022; Chowdhery et al., 2022). However, we have found that even when the desired answer is fixed, -introducing diversity in the reasoning processes can be highly beneficial; therefore we leverage - -1The parser is task dependent. For arithmetic reasoning, we parse the first numerical part as the final answer -after the model generates “The answer is ”. For commonsense reasoning, we parse the full string answer as the -final answer after the model generates “The answer is ”. Most generated outputs have a consistent format of -“{Reasoning paths}. The answer is X.” if we prompt the language model in this format. - -2This also means that the language model is not well calibrated and thus cannot distinguish well between -correct solutions and wrong solutions, which also explains why additional re-rankers were trained to better judge -the quality of the solutions in previous work (Cobbe et al., 2021; Thoppilan et al., 2022). - -3 - - Published as a conference paper at ICLR 2023 - -sampling, as commonly used for open-ended text generation (Radford et al., 2019; Brown et al., 2020; -Thoppilan et al., 2022), to achieve this goal. One should note that self-consistency can be applied -only to problems where the final answer is from a fixed answer set, but in principle this approach can -be extended to open-text generation problems if a good metric of consistency can be defined between -multiple generations, e.g., whether two answers agree or contradict each other. - -3 EXPERIMENTS - -We conducted a series of experiments to compare the proposed self-consistency method with existing -approaches on a range of reasoning benchmarks. We find that self-consistency robustly improves -reasoning accuracy for every language model considered, spanning a wide range of model scales. - -3.1 EXPERIMENT SETUP - -Tasks and datasets. We evaluate self-consistency on the following reasoning benchmarks.3 - -• Arithmetic reasoning. For these tasks, we used the Math Word Problem Repository (Koncel- -Kedziorski et al., 2016), including AddSub (Hosseini et al., 2014), MultiArith (Roy & Roth, -2015), and ASDiv (Miao et al., 2020). We also included AQUA-RAT (Ling et al., 2017), a -recently published benchmark of grade-school-math problems (GSM8K; Cobbe et al., 2021), -and a challenge dataset over math word problems (SVAMP; Patel et al., 2021). - -• Commonsense reasoning. For these tasks, we used CommonsenseQA (Talmor et al., 2019), -StrategyQA (Geva et al., 2021), and the AI2 Reasoning Challenge (ARC) (Clark et al., 2018). -• Symbolic Reasoning. We evaluate two symbolic reasoning tasks: last letter concatenation (e.g., -the input is “Elon Musk” and the output should be “nk”), and Coinflip (e.g., a coin is heads-up, -after a few flips is the coin still heads-up?) from Wei et al. (2022). - -Language models and prompts. We evaluate self-consistency over four transformer-based lan- -guage models with varying scales: - -• UL2 (Tay et al., 2022) is an encoder-decoder model trained on a mixture of denoisers with 20- -billion parameters. UL2 is completely open-sourced4 and has similar or better performance than -GPT-3 on zero-shot SuperGLUE, with only 20B parameters and thus is more compute-friendly; -• GPT-3 (Brown et al., 2020) with 175-billion parameters. We use two public engines code-davinci- -001 and code-davinci-002 from the Codex series (Chen et al., 2021) to aid reproducibility;5 -• LaMDA-137B (Thoppilan et al., 2022) is a dense left-to-right, decoder-only language model with -137-billion parameters, pre-trained on a mixture of web documents, dialog data and Wikipedia; -• PaLM-540B (Chowdhery et al., 2022) is a dense left-to-right, decoder-only language model with -540-billion parameters, pre-trained on a high quality corpus of 780 billion tokens with filtered -webpages, books, Wikipedia, news articles, source code, and social media conversations. -We perform all experiments in the few-shot setting, without training or fine-tuning the language -models. For a fair comparison we use the same prompts as in Wei et al. (2022): for all arithmetic -reasoning tasks we use the same set of 8 manually written exemplars; for each commonsense -reasoning task, 4-7 exemplars are randomly chosen from the training set with manually composed -chain-of-thought prompts.6 Full details on the prompts used are given in Appendix A.3. - -Sampling scheme. To sample diverse reasoning paths, we followed similar settings to those -suggested in Radford et al. (2019); Holtzman et al. (2020) for open-text generation. In particular, for -UL2-20B and LaMDA-137B we applied temperature sampling with T = 0.5 and truncated at the -top-k (k = 40) tokens with the highest probability, for PaLM-540B we applied T = 0.7, k = 40, and -for GPT-3 we use T = 0.7 without top-k truncation. We provide an ablation study in Section 3.5 to -show that self-consistency is generally robust to sampling strategies and parameters. - -3By default we use the test split for all datasets if the labels are available for evaluation. For CommonsenseQA -we use the dev split; for StrategyQA we use the question-only set from BIG-bench collaboration (2021): -https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/strategyqa. - -4Model checkpoints at https://github.com/google-research/google-research/tree/master/ul2. -5Public API available at https://openai.com/api/. -6Self-consistency is robust to different sets of prompts and we provide a study in Appendix A.1.2. - -4 - - Published as a conference paper at ICLR 2023 - -3.2 MAIN RESULTS - -We report the results of self-consistency averaged over 10 runs, where we sampled 40 outputs -independently from the decoder in each run. The baseline we compare to is chain-of-thought -prompting with greedy decoding (Wei et al., 2022), referred to as CoT-prompting, which has been -previously used for decoding in large language models (Chowdhery et al., 2022). - -Arithmetic Reasoning The results are shown in Table 2.7 Self-consistency improves the arithmetic -reasoning performance over all four language models significantly over chain-of-thought prompting. -More surprisingly, the gains become more significant when the language model’s scale increases, -e.g., we see +3%-6% absolute accuracy improvement over UL2-20B but +9%-23% for LaMDA- -137B and GPT-3. For larger models that already achieve high accuracy on most tasks (e.g., GPT-3 -and PaLM-540B), self-consistency still contributes significant additional gains with +12%-18% -absolute accuracy on tasks like AQuA and GSM8K, and +7%-11% on SVAMP and ASDiv. With -self-consistency, we achieve new state-of-the-art results on almost all tasks: despite the fact that self- -consistency is unsupervised and task-agnostic, these results compare favorably to existing approaches -that require task-specific training, or fine-tuning with thousands of examples (e.g., on GSM8K). - -Method - -AddSub MultiArith ASDiv - -AQuA - -SVAMP - -GSM8K - -Previous SoTA 94.9a - -60.5a - -75.3b - -37.9c - -57.4d - -35e / 55g - -UL2-20B - -CoT-prompting -Self-consistency 24.8 (+6.6) - -18.2 - -10.7 -15.0 (+4.3) - -16.9 -21.5 (+4.6) 26.9 (+3.3) - -23.6 - -12.6 -19.4 (+6.8) - -4.1 -7.3 (+3.2) - -LaMDA-137B - -PaLM-540B - -GPT-3 -Code-davinci-001 - -CoT-prompting -51.8 -Self-consistency 63.5 (+10.6) 75.7 (+23.9) 58.2 (+9.2) 26.8 (+9.1) - -49.0 - -52.9 - -17.7 - -38.9 -53.3 (+14.4) 27.7 (+10.6) - -17.1 - -CoT-prompting -Self-consistency 93.7 (+1.8) - -91.9 - -94.7 -99.3 (+4.6) - -74.0 -81.9 (+7.9) 48.3 (+12.5) 86.6 (+7.6) - -79.0 - -35.8 - -56.5 -74.4 (+17.9) - -CoT-prompting -59.5 -Self-consistency 67.8 (+10.6) 82.7 (+23.2) 61.9 (+9.2) 25.6 (+6.7) - -52.7 - -57.2 - -18.9 - -39.8 -54.5 (+14.7) 23.4 (+8.8) - -14.6 - -GPT-3 -Code-davinci-002 - -CoT-prompting -Self-consistency 91.6 (+2.2) - -89.4 - -96.2 -100.0 (+3.8) 87.8 (+7.6) 52.0 (+12.2) 86.8 (+11.0) 78.0 (+17.9) - -80.1 - -75.8 - -60.1 - -39.8 +The Basics -Table 2: Arithmetic reasoning accuracy by self-consistency compared to chain-of-thought prompting -(Wei et al., 2022). The previous SoTA baselines are obtained from: a: Relevance and LCA operation -classifier (Roy & Roth, 2015), b: Lan et al. (2021), c: Amini et al. (2019), d: Pi et al. (2022), e: -GPT-3 175B finetuned with 7.5k examples (Cobbe et al., 2021), g: GPT-3 175B finetuned plus an -additional 175B verifier (Cobbe et al., 2021). The best performance for each task is shown in bold. + What is DNA? -Method +It's a history book - -CSQA +a narrative of the journey of our -StrategyQA ARC-e +species through time. -ARC-c +It's a shop manual, with an incredibly detailed blueprint -Letter (4) Coinflip (4) +for building every human cell. -Previous SoTA 91.2a +And it's a transformative textbook of medicine, with -73.9b +insights that will give health care providers immense +new powers to treat, prevent and cure disease." +- Francis Collins -86.4c + What Does DNA Look Like? -75.0c +A -N/A +G -N/A +T -UL2-20B - -51.4 -CoT-prompting -Self-consistency 55.7 (+4.3) 54.9 (+1.6) - -53.3 - -61.6 -69.8 (+8.2) 49.5 (+6.8) - -42.9 - -0.0 -0.0 (+0.0) - -50.4 -50.5 (+0.1) - -LaMDA-137B - -CoT-prompting -57.9 -Self-consistency 63.1 (+5.2) 67.8 (+2.4) - -65.4 - -75.3 -79.3 (+4.0) 59.8 (+4.7) - -55.1 - -8.2 -8.2 (+0.0) - -72.4 -73.5 (+1.1) - -PaLM-540B - -79.0 -CoT-prompting -Self-consistency 80.7 (+1.7) 81.6 (+6.3) - -75.3 - -95.3 -96.4 (+1.1) 88.7 (+3.5) - -85.2 - -65.8 -70.8 (+5.0) 91.2 (+3.0) - -88.2 - -GPT-3 -Code-davinci-001 - -CoT-prompting -46.6 -Self-consistency 54.9 (+8.3) 61.7 (+5.0) - -56.7 - -63.1 -72.1 (+9.0) 53.7 (+10.6) 10.0 (+2.2) 75.9 (+4.5) - -43.1 - -71.4 - -7.8 - -GPT-3 -Code-davinci-002 - -CoT-prompting -79.0 -Self-consistency 81.5 (+2.5) 79.8 (+6.4) - -73.4 - -94.0 -96.0 (+2.0) 87.5 (+3.9) - -83.6 - -70.4 -73.4 (+3.0) 99.5 (+0.5) - -99.0 - -Table 3: Commonsense and symbolic reasoning accuracy by self-consistency compared to chain- -of-thought prompting (Wei et al., 2022). The previous SoTA baselines are obtained from: a: -DeBERTaV3-large + KEAR (Xu et al., 2021b), b: Chowdhery et al. (2022), c: UnifiedQA-FT -(Khashabi et al., 2020). The best performance for each task is shown in bold. - -7The standard deviation of self-consistency is ≤ 0.5 for all tasks and is thus omitted in the table. Please refer - -to Figure 2, Figure 7 and 8 for the standard deviations under varying numbers of sampled paths. - -5 - - Published as a conference paper at ICLR 2023 - -Commonsense and Symbolic Reasoning Table 3 shows the results on commonsense and symbolic -reasoning tasks. Similarly, self-consistency yields large gains across all four language models, and -obtained SoTA results on 5 out of 6 tasks. For symbolic reasoning, we test the out-of-distribution -(OOD) setting where the input prompt contains examples of 2-letters or 2-flips but we test examples -of 4-letters and 4-flips (this setting is more challenging as PaLM-540B or GPT-3 can already achieve -perfect in-distribution accuracy). In this challenging OOD setting, the gain of self-consistency is still -quite significant compared to CoT-prompting with sufficient model sizes. - -To show the effect of the number of sampled reasoning paths, we plot the accuracy (mean and -standard deviation over 10 runs) with respect to varying numbers of sampled paths (1, 5, 10, 20, 40) -in Figure 2. The results show that sampling a higher number (e.g., 40) of reasoning paths leads to a -consistently better performance, further emphasizing the importance of introducing diversity in the -reasoning paths. In Table 4, we show self-consistency yields a richer set of reasoning paths compared -to greedy decoding with a few example questions from two tasks. - -Figure 2: Self-consistency (blue) significantly improves accuracy over CoT-prompting with greedy -decoding (orange) across arithmetic and commonsense reasoning tasks, over LaMDA-137B. Sampling -a higher number of diverse reasoning paths consistently improves reasoning accuracy. - -GSM8K - -StrategyQA - -Question Henry made two stops during his 60-mile bike trip. He first stopped after 20 -miles. His second stop was 15 miles before the end of the trip. How many miles -did he travel between his first and second stops? - -Yes or no: Is Albany, Georgia the most populous -US Albany? - -Greedy -Decode - -He traveled 60 miles in total. He stopped after 20 miles, so he traveled 60 - 20 -= 40 miles between the first and second stops. The answer is 40. - -Albany, Georgia is the most populous US Al- -bany. Thus, the answer is yes. - -Sampled -Path 1 - -Sampled -Path 2 - -Henry travelled a total of 60 miles. His first stop was 20 miles in, and his -second stop was 15 miles before the end. So between his first and second stops -he travelled 60 - 20 -15 = 25 miles. The answer is 25. - -The most populous US Albany is Albany, New -York. Thus, Albany, Georgia is not the most -populous US Albany. So the answer is no. - -He made two stops during a 60-mile trip. The first was 20 miles into the trip. -The second was 15 miles before the end of the trip. This means the second stop -was 60 - 15 = 45 miles into the trip. Since he made the stops in order, the second -stop must have been 45 - 20 = 25 miles after the first stop. The answer is 25. - -Albany, Georgia has a population of about -88,000. Albany, New York has a population of -about 95,000. Thus, Albany, Georgia is not the -most populous US Albany. So the answer is no. - -Table 4: Examples where self-consistency helps repair the errors over greedy decode, on PaLM-540B. -Two sampled reasoning paths that are consistent with the ground truth are shown. - -3.3 SELF-CONSISTENCY HELPS WHEN CHAIN-OF-THOUGHT HURTS PERFORMANCE - -Ye & Durrett (2022) show that sometimes chain-of-thought prompting could hurt performance -compared to standard prompting in few-shot in-context learning. Here we perform a study using -self-consistency to see if it can help fill in the gap, over a set of common NLP tasks, including (1) -Closed-Book Question Answering: BoolQ (Clark et al., 2019), HotpotQA (Yang et al., 2018), and -(2) Natural Language Inference: e-SNLI (Camburu et al., 2018), ANLI (Nie et al., 2020) and RTE -(Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Bentivogli et al., 2009). - -The results over PaLM-540B are shown in Table 5. For some tasks (e.g., ANLI-R1, e-SNLI, RTE), -adding chain-of-thought does hurt performance compared to standard prompting (Brown et al., 2020), -but self-consistency is able to robustly boost the performance and outperform standard prompting, -making it a reliable way to add rationales in few-shot in-context learning for common NLP tasks. - -ANLI R1 / R2 / R3 e-SNLI RTE BoolQ HotpotQA (EM/F1) - -Standard-prompting (no-rationale) -CoT-prompting (Wei et al., 2022) - -69.1 / 55.8 / 55.8 -68.8 / 58.9 / 60.6 - -Self-consistency - -78.5 / 64.5 / 63.4 - -85.8 -81.0 - -88.4 - -84.8 -79.1 - -86.3 - -71.3 -74.2 - -78.4 - -27.1 / 36.8 -28.9 / 39.8 - -33.8 / 44.6 - -Table 5: Compare Standard/CoT prompting with self-consistency on common NLP tasks. - -6 - -0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths56586062Commonsense QA0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 - -3.4 COMPARE TO OTHER EXISTING APPROACHES - -We conduct a set of additional studies and show that self-consistency significantly outperforms -existing methods including sample-and-rank, beam search, and ensemble-based approaches. - -Comparison to Sample-and-Rank One commonly used approach to improve generation quality is -sample-and-rank, where multiple sequences are sampled from the decoder and then ranked according -to each sequence’s log probability (Adiwardana et al., 2020). We compare self-consistency with -sample-and-rank on GPT-3 code-davinci-001, by sampling the same number of sequences from the -decoder as self-consistency and taking the final answer from the top-ranked sequence. The results are -shown in Figure 3. While sample-and-rank does improve the accuracy with additionally sampled -sequences and ranking, the gain is much smaller compared to self-consistency. - -Figure 3: Self-consistency significantly outperforms sample-and-rank with the same # of samples. - -Comparison to Beam Search In Table 6, we compare self-consistency with beam search decoding -on the UL2-20B model. For a fair comparison we report the accuracy under the same number of -beams and reasoning paths. On both tasks self-consistency outperforms beam search significantly. -Note self-consistency can also adopt beam search to decode each reasoning path (results are shown -as “Self-consistency using beam search”), but its performance is worse compared to self-consistency -with sampling. The reason is that beam search yields a lower diversity in the outputs (Li & Jurafsky, -2016), while in self-consistency the diversity of the reasoning paths is the key to a better performance. - -Beam size / Self-consistency paths - -1 - -AQuA - -MultiArith - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -Beam search decoding (top beam) -Self-consistency using beam search -Self-consistency using sampling - -23.6 -23.6 - -10.7 -10.7 - -19.8 ± 0.3 21.2 ± 0.7 24.6 ± 0.4 24.2 ± 0.5 -19.7 ± 2.5 24.9 ± 2.6 25.3 ± 1.8 26.7 ± 1.0 26.9 ± 0.5 - -12.0 - -11.3 - -11.0 - -10.5 - -11.8 ± 0.0 11.4 ± 0.1 12.3 ± 0.1 10.8 ± 0.1 -9.5 ± 1.2 11.3 ± 1.2 12.3 ± 0.8 13.7 ± 0.9 14.7 ± 0.3 - -5 - -19.3 - -10 - -16.1 - -20 - -15.0 - -40 - -10.2 - -Table 6: Compare self-consistency with beam search decoding on the UL2-20B model. - -Comparison to Ensemble-based Approaches We further compare self-consistency to ensemble- -based methods for few-shot learning. In particular, we consider ensembling by: (1) prompt order -permutation: we randomly permute the exemplars in the prompt 40 times to mitigate model’s -sensitivity to prompt order (Zhao et al., 2021; Lu et al., 2021); and (2) multiple sets of prompts -(Gao et al., 2021): we manually write 3 different sets of prompts. We took majority vote of the -answers from greedy decoding in both approaches as an ensemble. Table 7 shows that compared to -self-consistency, existing ensemble-based approaches achieve a much smaller gain.8 In addition, note -that self-consistency is different from a typical model-ensemble approach, where multiple models -are trained and their outputs are aggregated. Self-consistency acts more like a “self-ensemble” on -top of a single language model. We additionally show the results of ensembling multiple models in -Appendix A.1.3 where the model-ensembles perform much worse compared to self-consistency. - -GSM8K MultiArith - -SVAMP - -ARC-e - -ARC-c - -CoT (Wei et al., 2022) -Ensemble (3 sets of prompts) -Ensemble (40 prompt permutations) -Self-Consistency (40 sampled paths) - -17.1 -18.6 ± 0.5 -19.2 ± 0.1 -27.7 ± 0.2 - -51.8 -57.1 ± 0.7 -60.9 ± 0.2 -75.7 ± 0.3 - -38.9 -42.1 ± 0.6 -42.7 ± 0.1 -53.3 ± 0.2 - -75.3 -76.6 ± 0.1 -76.9 ± 0.1 -79.3 ± 0.3 - -55.1 -57.0 ± 0.2 -57.0 ± 0.1 -59.8 ± 0.2 +C -Table 7: Self-consistency outperforms prompt-order and multi-prompt ensembles on LaMDA-137B. + Every cell in our body has the +same DNA…. -8Self-consistency is compatible with both ensemble approaches and we show the results in Appendix A.1.4. +Eye cell -7 +Karyotype -0510152025303540#Sampled Reasoning Paths12141618202224Accuracy (%)GSM8K0510152025303540#Sampled Reasoning Paths50556065707580Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths303540455055Accuracy (%)ARC (Challenge)Self Consistency (Multi-path)Sample & Rank (Multi-path)Greedy Decode (Single-path) Published as a conference paper at ICLR 2023 +Lung cell -3.5 ADDITIONAL STUDIES +Toe cell -We conducted a number of additional experiments to analyze different aspects of the self-consistency -method, including its robustness to sampling strategies and parameters, and how it works with -imperfect prompts and non-natural-language reasoning paths. + How much DNA is in one cell? -Self-Consistency is Robust to Sampling Strategies and Scaling We show self-consistency is -robust to sampling strategies and parameters, by varying T in temperature sampling (Ackley et al., -1985; Ficler & Goldberg, 2017), k in top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford -et al., 2019), and p in nucleus sampling (Holtzman et al., 2020), over PaLM-540B in Figure 4 (left). -Figure 4 (right) shows that self-consistency robustly improves performance across all scales for the -LaMDA-137B model series. The gain is relatively lower for smaller models due to certain abilities -(e.g., arithmetic) only emerge when the model reaches a sufficient scale (Brown et al., 2020). +Genome = 46 chromosomes -Figure 4: GSM8K accuracy. (Left) Self-consistency is robust to various sampling strategies and -parameters. (Right) Self-consistency improves performance across language model scales. +Genome = approx. 3 billion +base pairs -Self-Consistency Improves Robustness to Imperfect Prompts For few-shot learning with man- -ually constructed prompts, human annotators sometimes make minor mistakes when creating the -prompts. We further study if self-consistency can help improve a language model’s robustness to -imperfect prompts.9 We show the results in Table 8: while imperfect prompts decrease accuracy with -greedy decoding (17.1 → 14.9), self-consistency can fill in the gaps and robustly improve the results. +One base pair is +0.00000000034 meters -Additionally, we found that the consistency (in terms of % of decodes agreeing with the final -aggregated answer) is highly correlated with accuracy (Figure 5, over GSM8K). This suggests that -one can use self-consistency to provide an uncertainty estimate of the model in its generated solutions. -In other words, one can use low consistency as an indicator that the model has low confidence; i.e., -self-consistency confers some ability for the model to “know when it doesn’t know”. +DNA sequence in any two +people is 99.9% identical – +only 0.1% is unique! -Prompt with correct chain-of-thought + What makes one cell different from another? -17.1 - -LaMDA-137B - -Prompt with imperfect chain-of-thought 14.9 -23.4 - -+ Self-consistency (40 paths) - -Prompt with equations - -+ Self-consistency (40 paths) - -PaLM-540B - -Zero-shot CoT (Kojima et al., 2022) - -+ Self-consistency (40 paths) - -5.0 -6.5 - -43.0 -69.2 - -Table 8: Self-consistency works under imperfect prompts, equa- -tion prompts and zero-shot chain-of-thought for GSM8K. - -Figure 5: The consistency is cor- -related with model’s accuracy. - -Self-Consistency Works for Non-Natural-Language Reasoning Paths and Zero-shot CoT We -also tested the generality of the self-consistency concept to alternative forms of intermediate reasoning -like equations (e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there -are 3 + 2 = 5 cars.” to “3 + 2 = 5”). The results are shown in Table 8 (“Prompt with equations”): -self-consistency still improves accuracy by generating intermediate equations; however, compared to -generating natural language reasoning paths, the gain is smaller since the equations are much shorter -and less opportunity remains for generating diversity in the decoding process. In addition, we tested -self-consistency with zero-shot chain-of-thought (Kojima et al., 2022) and show that self-consistency -works for zero-shot CoT as well and improves the results significantly (+26.2%) in Table 8. - -9We use the same prompts as before, but swap all the numbers in the reasoning paths with random numbers -except the final answer, e.g., from “There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + -2 = 5 cars.” to “There are 7 cars in the parking lot already. 6 more arrive. Now there are 7 + 6 = 5 cars.”. - -8 - -0510152025303540#Sampled Reasoning Paths444852566064687276Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.7, k=20T=0.7, no top kp=0.95p=0.9Greedy Decode125102050100200Model size (#param in billions)510152025Accuracy (%)Self ConsistencyGreedy Decode020406080100Consistency (%)020406080100Accuracy (%) Published as a conference paper at ICLR 2023 - -4 RELATED WORK - -Reasoning in language models. Language models are known to struggle in Type 2 tasks, such as -arithmetic, logical and commonsense reasoning (Evans, 2010). Previous work has primarily focused -on specialized approaches for improving reasoning (Andor et al., 2019; Ran et al., 2019; Geva et al., -2020; Pi˛ekos et al., 2021). Compared to prior work, self-consistency is applicable to a wide range of -reasoning tasks without any additional supervision or fine-tuning, while still substantially improving -the performance of the chain-of-thought prompting approach proposed in Wei et al. (2022). -Sampling and re-ranking in language models. Multiple decoding strategies for language models -have been proposed in the literature, e.g., temperature sampling (Ackley et al., 1985; Ficler & -Goldberg, 2017), top-k sampling (Fan et al., 2018; Holtzman et al., 2018; Radford et al., 2019), -nucleus sampling (Holtzman et al., 2020), minimum Bayes risk decoding (Eikema & Aziz, 2020; Shi -et al., 2022), and typical decoding (Meister et al., 2022). Other work has sought to explicitly promote -diversity in the decoding process (Batra et al., 2012; Li et al., 2016; Vijayakumar et al., 2018). - -Re-ranking is another common approach to improve generation quality in language models (Adiwar- -dana et al., 2020; Shen et al., 2021). Thoppilan et al. (2022) collect additional human annotations -to train a re-ranker for response filtering. Cobbe et al. (2021) train a “verifier” to re-rank generated -solutions, which substantially improves the solve rate on math tasks compared to just fine-tuning the -language model. Elazar et al. (2021) improve the consistency of factual knowledge extraction by -extending pre-training with an additional consistency loss. All these methods require either training -an additional re-ranker or collecting additional human annotation, while self-consistency requires no -additional training, fine-tuning, nor extra data collection. -Extract reasoning paths. Some previous work has considered task-specific approaches for iden- -tifying reasoning paths, such as constructing semantic graphs (Xu et al., 2021a), learning an RNN -to retrieve reasoning paths over the Wikipedia graph (Asai et al., 2020), fine-tuning with human -annotated reasoning paths on math problems (Cobbe et al., 2021), or training an extractor with -heuristic-based pseudo reasoning paths (Chen et al., 2019). More recently, the importance of di- -versity in the reasoning processes has been noticed, but only leveraged via task-specific training, -either through an additional QA model over extracted reasoning paths (Chen et al., 2019), or by the -introduction of latent variables in a commonsense knowledge graph (Yu et al., 2022). Compared to -these approaches, self-consistency is far simpler and requires no additional training. The approach -we propose simply couples the generation of reasoning paths and a final answer by sampling from -the decoder, using aggregation to recover the most consistent answer without additional modules. -Consistency in language models. Some prior work has shown that language models can suffer -from inconsistency in conversation (Adiwardana et al., 2020), explanation generation (Camburu et al., -2020), and factual knowledge extraction (Elazar et al., 2021). Welleck et al. (2020) use “consistency” -to refer to generating an infinite-length sequence in recurrent language models. Nye et al. (2021) -improve the logical consistency of samples from a System 1 model by adding a System 2-inspired -logical reasoning module. In this paper we focus on a slightly different notion of “consistency”, i.e., -utilizing answer consistency among diverse reasoning paths to improve accuracy. - -5 CONCLUSION AND DISCUSSION - -We introduced a simple yet effective method called self-consistency, and observed that it significantly -improves accuracy in a range of arithmetic and commonsense reasoning tasks, across four large -language models with varying scales. Beyond accuracy gains, self-consistency is also useful for -collecting rationales when performing reasoning tasks with language models, and for providing -uncertainty estimates and improved calibration of language model outputs. - -One limitation of self-consistency is that it incurs more computation cost. In practice people can try a -small number of paths (e.g., 5 or 10) as a starting point to realize most of the gains while not incurring -too much cost, as in most cases the performance saturates quickly (Figure 2). As part of future work, -one could use self-consistency to generate better supervised data to fine-tune the model, such that the -model can give more accurate predictions in a single inference run after fine-tuning. In addition, we -observed that language models can sometimes generate incorrect or nonsensical reasoning paths (e.g., -the StrategyQA example in Table 4, the two population numbers are not exactly correct), and further -work is needed to better ground models’ rationale generations. - -9 - - Published as a conference paper at ICLR 2023 - -REPRODUCIBILITY STATEMENT - -In experiments, we included four different language models with varying scales. Two of them are pub- -lic models: UL2 is a completely open-sourced model with model checkpoints available at https:// -github.com/google-research/google-research/tree/master/ul2; GPT-3 is -also a public model with public API available at https://openai.com/api/. For GPT-3, -we have included two public engines (“code-davinci-001” and “code-davinci-002”) to further aid -reproducibility, as Codex is currently free so anyone can reproduce the results. In addition, as our -results make use of LaMDA-137B and PaLM-540B that are not publicly available, we provide the -exact input prompts for all tasks in Appendix A.3 (and note that we do not perform any finetuning -and only apply prompting to off-the-shelf language models). - -ETHICS STATEMENT - -As we stated in the discussion, language models can sometimes generate nonsensical or non-factual -reasoning paths, so one should use language models’ outputs with extra caution. We deal with -reasoning tasks mostly and the generated rationales are only used for inspecting how a model reaches -its answer. One could potentially use the generated rationales to further check why the model makes -certain mistakes or whether the model contains any biases when performing a certain task. For -language model in real-world use, further work is needed to better ground models’ predictions and -improve model’s factuality and safety, to ensure the models do not cause harms to users. - -REFERENCES - -David H. Ackley, Geoffrey E. Hinton, and Terrence J. Sejnowski. A learning algorithm for boltzmann -machines. Cognitive Science, 9(1):147–169, 1985. ISSN 0364-0213. URL https://www. -sciencedirect.com/science/article/pii/S0364021385800124. - -Daniel Adiwardana, Minh-Thang Luong, David R. So, Jamie Hall, Noah Fiedel, Romal Thoppilan, -Zi Yang, Apoorv Kulshreshtha, Gaurav Nemade, Yifeng Lu, and Quoc V. Le. Towards a human-like -open-domain chatbot, 2020. - -Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh -Hajishirzi. MathQA: Towards interpretable math word problem solving with operation-based -In Proceedings of the 2019 Conference of the North American Chapter of the -formalisms. -Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long -and Short Papers), pp. 2357–2367. Association for Computational Linguistics, June 2019. URL -https://aclanthology.org/N19-1245. - -Daniel Andor, Luheng He, Kenton Lee, and Emily Pitler. Giving BERT a calculator: Finding -operations and arguments with reading comprehension. In Proceedings of the 2019 Conference on -Empirical Methods in Natural Language Processing and the 9th International Joint Conference -on Natural Language Processing (EMNLP-IJCNLP), 2019. URL https://aclanthology. -org/D19-1609. - -Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learn- -ing to retrieve reasoning paths over wikipedia graph for question answering. In International -Conference on Learning Representations, 2020. URL https://openreview.net/forum? -id=SJgVHkrYDH. - -Roy Bar-Haim, Ido Dagan, Bill Dolan, Lisa Ferro, Danilo Giampiccolo, Bernardo Magnini, and -Idan Szpektor. The second pascal recognising textual entailment challenge. In Proceedings of the -second PASCAL challenges workshop on recognising textual entailment, 2006. - -Dhruv Batra, Payman Yadollahpour, Abner Guzman-Rivera, and Gregory Shakhnarovich. Diverse -m-best solutions in markov random fields. In Proceedings of the 12th European Conference on -Computer Vision - Volume Part V, ECCV’12, pp. 1–16, Berlin, Heidelberg, 2012. Springer-Verlag. -ISBN 9783642337147. URL https://doi.org/10.1007/978-3-642-33715-4_1. - -10 - - Published as a conference paper at ICLR 2023 - -Luisa Bentivogli, Peter Clark, Ido Dagan, and Danilo Giampiccolo. The fifth pascal recognizing - -textual entailment challenge. In TAC, 2009. - -BIG-bench collaboration. Beyond the imitation game: Measuring and extrapolating the capabil- -In preparation, 2021. URL https://github.com/google/ - -ities of language models. -BIG-bench/. - -Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, -Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel -Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, -Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, -Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, -and Dario Amodei. Language models are few-shot learners. In Advances in Neural Information -Processing Systems, 2020. URL https://proceedings.neurips.cc/paper/2020/ -file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. - -language inference with natural - -Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. - -snli: Natural -gio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, -(eds.), Advances -Information Processing Systems 31, -Curran Associates, -8163-e-snli-natural-language-inference-with-natural-language-explanations. -pdf. - -e- -In S. Ben- -and R. Garnett -pp. 9539–9549. -URL http://papers.nips.cc/paper/ - -language explanations. - -in Neural -Inc., - -2018. - -Oana-Maria Camburu, Brendan Shillingford, Pasquale Minervini, Thomas Lukasiewicz, and Phil -Blunsom. Make up your mind! adversarial generation of inconsistent natural language explanations. -In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. -4157–4165, Online, July 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -acl-main.382. URL https://aclanthology.org/2020.acl-main.382. - -Jifan Chen, Shih-Ting Lin, and Greg Durrett. Multi-hop question answering via reasoning chains. - -CoRR, abs/1910.02610, 2019. URL http://arxiv.org/abs/1910.02610. - -Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared -Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large -language models trained on code. arXiv preprint arXiv:2107.03374, 2021. +DNA = “the life +instructions of the +cell” -Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam -Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, -Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam -Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James -Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Lev- -skaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin -Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, -Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. -Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon -Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark -Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeff Dean, -Slav Petrov, and Noah Fiedel. Palm: Scaling language modeling with pathways, 2022. URL -https://arxiv.org/abs/2204.02311. - -Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina -Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. In NAACL, -2019. - -Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and -Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. -ArXiv, abs/1803.05457, 2018. - -Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, -Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John -Schulman. Training verifiers to solve math word problems, 2021. +Gene = segment of +DNA that tells the +cell how to make a +certain protein. -11 +Allele = one of two +or more different +versions of a gene - Published as a conference paper at ICLR 2023 + Sequence for normal adult hemoglobin: -Ido Dagan, Oren Glickman, and Bernardo Magnini. The pascal recognising textual entailment +Sequence for mutant hemoglobin: -challenge. In Machine Learning Challenges Workshop, pp. 177–190. Springer, 2005. +Wild-type Hemoglobin Protein -Bryan Eikema and Wilker Aziz. Is MAP decoding all you need? the inadequacy of the mode in neural -machine translation. In Proceedings of the 28th International Conference on Computational Lin- -guistics, pp. 4506–4520, Barcelona, Spain (Online), December 2020. International Committee on -Computational Linguistics. URL https://aclanthology.org/2020.coling-main. -398. +Mutant Protein -Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich -Schütze, and Yoav Goldberg. Measuring and improving consistency in pretrained language -models. Transactions of the Association for Computational Linguistics, 9:1012–1031, 2021. doi: -10.1162/tacl_a_00410. URL https://aclanthology.org/2021.tacl-1.60. - -Jonathan St BT Evans. Intuition and reasoning: A dual-process perspective. Psychological Inquiry, - -21(4):313–326, 2010. - -Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Proceedings -of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long -Papers), pp. 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics. -doi: 10.18653/v1/P18-1082. URL https://aclanthology.org/P18-1082. - -Jessica Ficler and Yoav Goldberg. Controlling linguistic style aspects in neural language generation. In -Proceedings of the Workshop on Stylistic Variation, pp. 94–104, Copenhagen, Denmark, September -2017. Association for Computational Linguistics. doi: 10.18653/v1/W17-4912. URL https: -//aclanthology.org/W17-4912. - -Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot -learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguis- -tics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long -Papers), pp. 3816–3830, Online, August 2021. Association for Computational Linguistics. doi: -10.18653/v1/2021.acl-long.295. URL https://aclanthology.org/2021.acl-long. -295. - -Mor Geva, Ankit Gupta, and Jonathan Berant. Injecting numerical reasoning skills into language -models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguis- -tics, 2020. doi: 10.18653/v1/2020.acl-main.89. URL https://aclanthology.org/2020. -acl-main.89. +Normal Red Blood Cell -Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle -use a laptop? A question answering benchmark with implicit reasoning strategies. Transactions of -the Association for Computational Linguistics, 2021. URL https://aclanthology.org/ -2021.tacl-1.21. +Abnormal Red Blood Cell -Danilo Giampiccolo, Bernardo Magnini, Ido Dagan, and Bill Dolan. The third pascal recognizing -textual entailment challenge. In Proceedings of the ACL-PASCAL workshop on textual entailment -and paraphrasing, pp. 1–9. Association for Computational Linguistics, 2007. + The Human Genome Project Goals -Ari Holtzman, Jan Buys, Maxwell Forbes, Antoine Bosselut, David Golub, and Yejin Choi. Learning -In Proceedings of the 56th Annual Meeting of the -to write with cooperative discriminators. -Association for Computational Linguistics (Volume 1: Long Papers), pp. 1638–1649, Melbourne, -Australia, July 2018. Association for Computational Linguistics. doi: 10.18653/v1/P18-1152. -URL https://aclanthology.org/P18-1152. +• To sequence (i.e. determine the exact order of +nucleotides (A,T,G,C) for ALL of the DNA in a +human cell -Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text -degeneration. In International Conference on Learning Representations, 2020. URL https: -//openreview.net/forum?id=rygGQyrFvH. +• To determine which sections of DNA represent -Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to -solve arithmetic word problems with verb categorization. In Proceedings of the 2014 Conference on -Empirical Methods in Natural Language Processing (EMNLP), 2014. doi: 10.3115/v1/D14-1058. -URL https://aclanthology.org/D14-1058. +individual genes (protein-coding units). -12 - - Published as a conference paper at ICLR 2023 - -Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Han- -naneh Hajishirzi. UNIFIEDQA: Crossing format boundaries with a single QA system. In Findings -of the Association for Computational Linguistics: EMNLP 2020, pp. 1896–1907, Online, Novem- -ber 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.findings-emnlp.171. -URL https://aclanthology.org/2020.findings-emnlp.171. - -Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large -language models are zero-shot reasoners. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, -and Kyunghyun Cho (eds.), Advances in Neural Information Processing Systems, 2022. URL -https://openreview.net/forum?id=e2TBb5y0yFf. + The HGP: International effort to decipher +the blueprint of a human being. -Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. MAWPS: -A math word problem repository. In Proceedings of the 2016 Conference of the North American -Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016. -doi: 10.18653/v1/N16-1136. URL https://aclanthology.org/N16-1136. + How It Was Done -Yihuai Lan, Lei Wang, Qiyuan Zhang, Yunshi Lan, Bing Tian Dai, Yan Wang, Dongxiang Zhang, -and Ee-Peng Lim. MWPToolkit: An open-source framework for deep learning-based math word -problem solvers. arXiv preprint arXiv:2109.00799, 2021. URL https://arxiv.org/abs/ -2109.00799. +DNA samples collected from +thousands of volunteers -Jiwei Li and Dan Jurafsky. Mutual information and diverse decoding improve neural machine +Samples sent to Human Genome +Project centers across the world -translation, 2016. URL https://arxiv.org/abs/1601.00372. +Scientists at centers perform DNA +sequencing and analysis -Jiwei Li, Will Monroe, and Dan Jurafsky. A simple, fast diverse decoding algorithm for neural -generation. CoRR, abs/1611.08562, 2016. URL http://arxiv.org/abs/1611.08562. + • February 2001: Draft of the -Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale genera- -tion: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual -Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2017. doi: -10.18653/v1/P17-1015. URL https://aclanthology.org/P17-1015. - -Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. Fantastically -ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. ArXiv, -abs/2104.08786, 2021. - -Clara Meister, Tiago Pimentel, Gian Wiher, and Ryan Cotterell. Typical decoding for natural language - -generation. arXiv preprint arXiv:2202.00666, 2022. - -Shen Yun Miao, Chao Chun Liang, and Keh Yih Su. A diverse corpus for evaluating and developing -English math word problem solvers. In Proceedings of the 58th Annual Meeting of the Asso- -ciation for Computational Linguistics, 2020. URL https://aclanthology.org/2020. -acl-main.92. - -Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adver- -sarial NLI: A new benchmark for natural language understanding. In Proceedings of the 58th -Annual Meeting of the Association for Computational Linguistics. Association for Computational -Linguistics, 2020. - -Maxwell Nye, Michael Henry Tessler, Joshua B. Tenenbaum, and Brenden M. Lake. Improving -coherence and consistency in neural sequence models with dual-system, neuro-symbolic reasoning. -In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan (eds.), Advances in Neural -Information Processing Systems, 2021. URL https://openreview.net/forum?id= -uyKk_avJ-p4. - -Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP models really able to solve simple -math word problems? In Proceedings of the 2021 Conference of the North American Chapter of -the Association for Computational Linguistics: Human Language Technologies, pp. 2080–2094, -Online, June 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main. -168. URL https://aclanthology.org/2021.naacl-main.168. - -Xinyu Pi, Qian Liu, Bei Chen, Morteza Ziyadi, Zeqi Lin, Yan Gao, Qiang Fu, Jian-Guang Lou, and - -Weizhu Chen. Reasoning like program executors, 2022. - -13 - - Published as a conference paper at ICLR 2023 - -Piotr Pi˛ekos, Mateusz Malinowski, and Henryk Michalewski. Measuring and improving BERT’s -mathematical abilities by predicting the order of reasoning. In Proceedings of the 59th Annual Meet- -ing of the Association for Computational Linguistics and the 11th International Joint Conference on -Natural Language Processing (Volume 2: Short Papers), 2021. doi: 10.18653/v1/2021.acl-short.49. -URL https://aclanthology.org/2021.acl-short.49. - -Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language - -models are unsupervised multitask learners. 2019. - -Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John -Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: -Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446, 2021. +sequence published in Nature +(public effort )and Science +(Celera – private company). -Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. NumNet: Machine reading comprehension -with numerical reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural -Language Processing and the 9th International Joint Conference on Natural Language Processing -(EMNLP-IJCNLP), 2019. doi: 10.18653/v1/D19-1251. URL https://aclanthology. -org/D19-1251. +• April, 2003 (50 years after -Subhro Roy and Dan Roth. Solving general arithmetic word problems. In Proceedings of the 2015 -Conference on Empirical Methods in Natural Language Processing, 2015. doi: 10.18653/v1/ -D15-1202. URL https://aclanthology.org/D15-1202. +Watson and Crick structure of +DNA was published) : Full +sequence published and +researchers determined that +within this sequence there was +somewhere between 30,000 +and 40,000 genes. We now +believe there are closer to +25,000 genes -Jianhao Shen, Yichun Yin, Lin Li, Lifeng Shang, Xin Jiang, Ming Zhang, and Qun Liu. Generate -& rank: A multi-task framework for math word problems. In Findings of the Association for -Computational Linguistics: EMNLP 2021, pp. 2269–2279, Punta Cana, Dominican Republic, -November 2021. Association for Computational Linguistics. URL https://aclanthology. -org/2021.findings-emnlp.195. + Still A Lot of Work To Do . . . -Freda Shi, Daniel Fried, Marjan Ghazvininejad, Luke Zettlemoyer, and Sida I. Wang. Natural -language to code translation with execution. In Proceedings of the 2022 Conference on Empirical -Methods in Natural Language Processing, pp. 3533–3546, Abu Dhabi, United Arab Emirates, -December 2022. Association for Computational Linguistics. URL https://aclanthology. -org/2022.emnlp-main.231. +Comparing the +genomes of +humans and +other organisms -Keith E Stanovich and Richard F West. +Discovering +DNA and gene +functions -Individual differences in reasoning: Implications for -the rationality debate? Behavioral and brain sciences, 23(5):645–665, 2000. URL https: -//pubmed.ncbi.nlm.nih.gov/11301544/. +Investigating +interactions between +DNA sequences, gene +products, and +environmental factors -Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. CommonsenseQA: A question -answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of -the North American Chapter of the Association for Computational Linguistics: Human Language -Technologies, Volume 1 (Long and Short Papers), 2019. URL https://aclanthology. -org/N19-1421. +Analyzing +genetic variation +between +individuals and +populations -Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Jason Wei, Xuezhi Wang, Hyung Won Chung, -Dara Bahri, Tal Schuster, Steven Zheng, Denny Zhou, Neil Houlsby, and Donald Metzler. Unifying -language learning paradigms, 2022. URL https://arxiv.org/abs/2205.05131. + How Can We Use This Information? -Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze -Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog -applications. arXiv preprint arXiv:2201.08239, 2022. URL https://arxiv.org/abs/ -2201.08239. +Better understanding of human +disease -Ashwin Vijayakumar, Michael Cogswell, Ramprasaath Selvaraju, Qing Sun, Stefan Lee, David -Crandall, and Dhruv Batra. Diverse beam search for improved description of complex scenes. -Proceedings of the AAAI Conference on Artificial Intelligence, 32, Apr. 2018. URL https: -//ojs.aaai.org/index.php/AAAI/article/view/12340. +Personalized medicine & +Pharmacogenetics -Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc -Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. -Conference on Neural Information Processing Systems (NeurIPS), 2022. URL https://arxiv. -org/pdf/2201.11903. +Greater +insight into +cognitive +function -14 +Insight into +human origins - Published as a conference paper at ICLR 2023 +Identifying genetic susceptibility to disease -Sean Welleck, Ilia Kulikov, Jaedeok Kim, Richard Yuanzhe Pang, and Kyunghyun Cho. Consistency -of a recurrent language model with respect to incomplete decoding. In Proceedings of the 2020 -Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5553–5568, -Online, November 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020. -emnlp-main.448. URL https://aclanthology.org/2020.emnlp-main.448. + Inheritance of Genes -Weiwen Xu, Yang Deng, Huihui Zhang, Deng Cai, and Wai Lam. Exploiting reasoning chains -for multi-hop science question answering. In Findings of the Association for Computational -Linguistics: EMNLP 2021, pp. 1143–1156, Punta Cana, Dominican Republic, November 2021a. -Association for Computational Linguistics. URL https://aclanthology.org/2021. -findings-emnlp.99. + Gregor Mendel -Yichong Xu, Chenguang Zhu, Shuohang Wang, Siqi Sun, Hao Cheng, Xiaodong Liu, Jianfeng -Gao, Pengcheng He, Michael Zeng, and Xuedong Huang. Human parity on commonsenseqa: -Augmenting self-attention with external attention, 2021b. URL https://arxiv.org/abs/ -2112.03254. +1822-1884 -Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and -Christopher D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answer- -ing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, -pp. 2369–2380, Brussels, Belgium, October-November 2018. Association for Computational Lin- -guistics. doi: 10.18653/v1/D18-1259. URL https://aclanthology.org/D18-1259. +• Augustinian monk who -Xi Ye and Greg Durrett. The unreliability of explanations in few-shot prompting for textual reasoning. -In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho (eds.), Advances in -Neural Information Processing Systems, 2022. URL https://openreview.net/forum? -id=Bct2f8fRd8S. +cross-bred pea plants with +different characteristics +• Observations led to laws -Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, and Meng Jiang. Diversifying -content generation for commonsense reasoning with mixture of knowledge graph experts. In -Findings of Annual Meeting of the Association for Computational Linguistics (ACL), 2022. +regarding the transmission +of hereditary +characteristics from +generation to generation +• Many of the concepts from +his observations still hold +true today! -Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. Calibrate before use: Improving -few-shot performance of language models. In Marina Meila and Tong Zhang (eds.), Proceed- -ings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of -Machine Learning Research. PMLR, 2021. URL https://proceedings.mlr.press/ -v139/zhao21c.html. +Picture from www.nih.nlm.gov -15 + Mendel’s +Laws: +1. Principle of Segregation: Two members of a - Published as a conference paper at ICLR 2023 +gene pair segregate from each other in the +formation of gametes; half the gametes carry +one allele, and the other half carry the other +allele +What it means: each gene has two copies +(alleles) and a parent will give only one copy to +a child. The other parent will give another +copy, and thus the child will receive two copies +(alleles) – one from each parent. Each child +will literally be half-mom and half-dad! -A APPENDIX + Mendel’s +Laws: -A.1 ADDITIONAL EXPERIMENT RESULTS +2. Principle of Independent Assortment: -A.1.1 ROBUSTNESS TO SAMPLING STRATEGIES AND PARAMETERS +Genes for different traits assort +independently of one another in gamete +production +What it means: different genes are +inherited separately. For example, the +gene which codes for eye color is inherited +separately from the gene which codes for +nose shape. -In Figure 6 we ablate the results with respect to different sampling strategies and parameters by -varying T in temperature sampling and k in Top-k sampling, on LaMDA-137B. We show that -self-consistency is robust to various sampling strategies and parameters. + Mendelian Concepts -Figure 6: GSM8K accuracy over LaMDA-137B. Self-consistency works under various sampling -strategies and sampling parameters. +Dominant = only one allele of a gene +necessary to express the trait +Recessive = both alleles of a gene must +be identical to express the trait +Heterozygous = alleles of a particular gene +are non-identical +Homozygous = alleles of a particular gene +are identical -In Figure 7 and Figure 8, we show the results of self-consistency compared with greedy decoding a -single path over LaMDA-137B and PaLM-540B, respectively. Self-consistency improves over greedy -decode by a quite significant margin on both models, on top of high accuracy already achieved by -scaling up model sizes. + Applying Mendelian Concepts: +The Punnett Square -Figure 7: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over LaMDA-137B. Sampling a higher number of diverse reasoning -paths consistently improves reasoning accuracy. +Parent 2 -We further show additional sampled reasoning paths from the LaMDA-137B model in Table 12, and -sampled reasoning paths from the PaLM-540B model in Table 13. We see that the diversity in the -additionally sampled reasoning paths indeed helps the model arrive at a more correct final answer -after aggregation. +Allele 1 -A.1.2 ROBUSTNESS TO DIFFERENT SETS OF PROMPTS +Allele 2 -In Table 9, we further show that self-consistency is quite robust to different sets of input prompts. -We manually wrote 3 different sets of chain-of-thought as prompts to the model. Across all sets of -prompts, self-consistency yields consistent gains over the original CoT approach. +Allele 1 -A.1.3 COMPARED TO MODEL ENSEMBLES +A1/A1 A1/A2 -Additionally, we provide results of directly ensembling the outputs from multiple language models. -The results are shown in Table 10, by greedily decoding sequences from 3 language models and +Parent 1 -16 +Allele 2 -481216202428323640#Sampled Reasoning Paths182022242628Accuracy (%)T=0.7, k=40T=0.5, k=40T=0.3, k=40T=0.5, k=20T=0.5, no top kGreedy Decode0510152025303540#Sampled Reasoning Paths505560657075Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Paths4446485052545658ASDiv0510152025303540#Sampled Reasoning Paths3336394245485154SVAMP0510152025303540#Sampled Reasoning Paths1416182022242628GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths56586062Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths62636465666768Strategy QA0510152025303540#Sampled Reasoning Paths687072747678ARC (Easy)0510152025303540#Sampled Reasoning Paths505254565860ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +A2/A1 A2/A2 -Figure 8: Self-consistency (blue) significantly improves accuracy across various arithmetic and -commonsense reasoning tasks, over PaLM-540B. Sampling a higher number of diverse reasoning -paths consistently helps reasoning accuracy. + Homozygous dominant + +Homozygous dominant -Prompt set 1 (used in the main text) +Homozygous dominant + +Heterozygous -Prompt set 2 +A -Prompt set 3 +A -CoT (Wei et al., 2022) -Self-consistency +A -56.5 -74.4 (+17.9) +A -54.6 -72.1 (+17.5) +A -54.0 -70.4 (+16.4) +AA -Table 9: GSM8K accuracy over PaLM-540B. The results show robustness of self-consistency with -respect to different prompts in the input. +AA -taking the majority vote (averaged over 10 runs). Note this is a typical ensemble approach (averaging -over the predictions over multiple models) and it achieves a performance significantly worse than -self-consistency (self-consistency over PaLM-540B gets an accuracy of 74.4%), as lower-capacity -models drag down the performance of higher-capacity models. In addition, this approach is limited in -two ways: 1) It requires multiple models for an ensemble which might not always be available, while -self-consistency only requires one single model to “self-ensemble”; 2) If one of the models is much -weaker, it can actually hurt the final performance. +A -Method +AA -Single model +AA -PaLM-540B, greedy / self-consistency +A -Ensemble of models +AA -LaMDA-137B + PaLM-540B -PaLM-540B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + GPT-3 (code-davinci-001, 175B) -LaMDA-137B + PaLM-540B + GPT-3 (code-davinci-001, 175B) +AA -GSM8K accuracy +a -56.5 / 74.4 +Aa -36.9 ± 0.5 -36.6 ± 0.4 -16.0 ± 0.8 -33.3 ± 0.7 +Aa -Table 10: Comparison of GSM8K accuracy over multiple-model ensembles. +Homozygous dominant + +Homozygous recessive -A.1.4 COMBINING SELF-CONSISTENCY WITH OTHER ENSEMBLING STRATEGIES +a -Self-consistency is completely compatible with other ensemble strategies, although the gains achieved -by self-consistency are significantly higher than other ensemble strategies (and can “override” the -performance gains achieved by other ensemble strategies). We further performed experiments and -include the results in Table 11 (for a fair comparison, we use 40 sets of prompts, or 40 prompt -permutations to compare with self-consistency with 40 paths, all experiments are based on PaLM- -540B). +a -17 +A -0510152025303540#Sampled Reasoning Chains8688909294Accuracy (%)AddSub0510152025303540#Sampled Reasoning Chains727476788082Accuracy (%)ASDiv0510152025303540#Sampled Reasoning Chains30333639424548Accuracy (%)AQuAGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Chains889092949698Accuracy (%)MultiArith0510152025303540#Sampled Reasoning Chains70.072.575.077.580.082.585.087.5Accuracy (%)SVAMP0510152025303540#Sampled Reasoning Chains505560657075Accuracy (%)GSM8KGreedy Decode (Single-path)Self Consistency (Multi-path)0510152025303540#Sampled Reasoning Paths7475767778798081Accuracy (%)Commonsense QA0510152025303540#Sampled Reasoning Paths7476788082Strategy QA0510152025303540#Sampled Reasoning Paths8890929496ARC (Easy)0510152025303540#Sampled Reasoning Paths788082848688ARC (Challenge)Greedy Decode (Single-path)Self Consistency (Multi-path) Published as a conference paper at ICLR 2023 +Aa -GSM8K accuracy +Aa -Different sets of prompts (x40) -Prompt-permutation (x40) -Self-consistency (x40) -Self-consistency + different sets of prompts (x40) -Self-consistency + prompt-permutation (x40) +A -58.9 -59.6 -74.4 -75.4 -73.8 +Aa -Table 11: Combining self-consistency with other ensembling strategies. +Aa -A.2 DETAILS ON RESOURCES AND INFERENCE +Heterozygous + Heterozygous -For all four language models we perform prompting-based inference only. For UL2 we use TPU v3 -(2x2 configuration, 4 chips, 8 cores). For GPT-3 models the experiments are done though the public -API.10 For LaMDA-137B we use TPU v3 (8x8 configuration, 64 chips, 128 cores). For PaLM-540B -we use TPU v4 (4x4x12 configuration, 192 chips, 384 cores). Most inference jobs take 1 to 4 hours -(over about 1,000 examples) for each task on UL2 and LaMDA-137B, and about 2 to 12 hours on -PaLM-540B. Some tasks (e.g., commonsense reasoning) take longer but do not exceed 2 days for -each task. +A -For GPT-3 models, we use 128 max tokens for all methods, without frequency penalty or presence -penalty. For all models, we take the generated outputs until the start of the next “Q:" to parse the final -answers, consistent with our prompting format. +a -Question +AA -Greedy Decode +Aa -Sampled Path 1 - -Sampled Path 2 - -[MultiArith] Sam had 10 socks. -If he threw away 3 old ones -that didn’t fit and bought 36 new -ones, how many socks would he -have? - -Sam had 10 socks. He threw -away 3 old ones that didn’t fit. -He bought 36 new ones. That -means he has 10 + 36 - 3 = 41 -socks. The answer is 41. - -[SVAMP] In a school there are -697 girls and the rest are boys. -If there are 228 more girls than -boys. How many boys are there -in that school? - -[AQuA] There are 10 oranges in -a basket. Find the no. of ways in -which 2 oranges are chosen from -the basket? (a) 45 (b) 90 (c) 120 -(d) 150 (e) 180 - -There are 697 girls and the rest -are boys. That means there are -697 + 228 = 925 girls. The -rest are boys. So the number of -boys is 925 - 697 = 228. The -answer is 228. - -There are 10 oranges in a bas- -ket. -So there are 10 ways -in which 2 oranges are chosen -from the basket. - -[CommonsenseQA] The man -laid on the soft moss and looked -up at the trees, where was the -man? (a) niagra falls (b) forest -(c) waterfall (d) ground (e) tree - -The answer must be something -that is soft and mossy. Of the -above choices, only ground is -soft and mossy. So the answer -is (d). - -[ARC] Which of these will most -likely harm a habitat? (a) plant- -ing trees. (b) water pollution. (c) -rainfall. (d) sunlight. - -Planting trees will not harm a -habitat. So the answer is (a). - -[ARC-challenge] Which pro- -cess in the carbon cycle takes the -longest to complete? (a) emis- -sion of waste. (b) respiration in -animals. -(c) photosynthesis in -plants. (d) formation of fossil fu- -els. - -The process of photosynthesis -in plants takes the longest to -complete. So the answer is (c). - -Sam had 10 socks. He -threw away 3 old ones. So -now he has 7 left. Then he -bought 36 new ones. So -now he has 7 + 36 = 43 -socks. The answer is 43. - -He had 10 socks. He threw away -3 old ones that didn’t fit. He -bought 36 new ones. So in total -he has 10 - 3 + 36 = 43 socks. The -answer is 43. - -The number of boys is the -difference between 697 -and 228. This is 697 - 228 -= 469. The answer is 469. - -In the beginning there are 697 -girls and the rest are boys. So, the -number of boys must be 697 - 228 -= 469 boys. The answer is 469. - -There are 10 oranges in a -basket. We need to choose -2 oranges from the basket. -So the number of ways are -10C2. So the answer is (a). - -The -be -answer must -somewhere where moss -is found. Of the above -choices, only forests are -known for having moss. -So the answer is (b). - -The answer is (b), water -pollution, because it will -most likely harm the habi- -tat. The other answers are -not likely to harm the habi- -tat. - -The formation of fossil fu- -els takes the longest time -to complete. So the an- -swer is (d). - -There are 10 oranges in the bas- -ket. If you choose 2 oranges, you -have 10C2 = 45 ways of choosing -2 oranges. So the answer is (a). - -The answer must be a place where -the man is laying down and look- -ing up at trees. Of the above -is a place -choices, only forest -where people lay down and look -at trees. So the answer is (b). - -The answer is (b), water pollution. -Trees are good for a habitat, rain- -fall is necessary for a habitat, and -sunlight is necessary for a habitat. -So the answer is (b). - -The process that takes the longest -time to complete is the forma- -tion of fossil fuels, which hap- -pens over millions of years. So -the answer is (d). - -Table 12: Additional examples where self-consistency helps repair the errors over greedy decode on -LaMDA-137B. Two sampled reasoning paths that are consistent with the ground truth are shown. - -10https://beta.openai.com/docs/api-reference/making-requests - -18 - - Kyle bought a book that is sold -for $19.50. With a 25% discount, -the original price must be higher. -If we call -the original price x, -then the discounted price is 0.75x. -Since this is $19.50, we can write -0.75x = 19.50. Solving this, we -find that the original price is x = -19.50 / 0.75 = $26. The answer is -$26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. One cu- -bic meter contains 1000 litres, so -the tank contains 120 m3 * 1000 -liters/m3 = 120,000 liters. So the -answer is (d). - -Published as a conference paper at ICLR 2023 - -Question - -Greedy Decode - -Sampled Path 1 - -Sampled Path 2 - -[GSM8K] Kyle bought -last -year’s best-selling book for -$19.50. This is with a 25% dis- -count from the original price. -What was the original price of -the book? - -The book was sold for $19.50. -This is 25% less than the origi- -nal price. This means the original -price was 100%. 25% of 100% -is 25%. -So the original price -was 100% - 25% = 75%. The -original price was 75% of $19.50. -75% of $19.50 is 75% * $19.50 = -$14.625. The answer is $14.625. - -The original price of the book -is $19.50. Since it was dis- -counted by 25%, the original -price was 19.50 / 0.75 = 26. -The answer is 26. - -The volume of the tank is 8 m * -6 m * 2.5 m = 120 m3. So the -capacity of the tank is 120 m3. So -the answer is (c). - -The volume of the tank is 8m -* 6m* 2.5m = 120m3. This is -equal to 120000 litres. So the -answer is (d) - -[AQuA] The capacity of a -tank of dimensions (8 m x 6 -m x 2.5 m) is (a) 120 litres (b) -1200 litres (c) 12000 litres (d) -120000 litres (e) None of these - -no: -[StrategyQA] Yes -Can the Great Depression be -treated with Prozac? - -or - -The Great Depression was an -economic depression. Prozac is -a drug used to treat depression. -Thus, the Great Depression could -be treated with Prozac. So the an- -swer is yes. - -Prozac is an anti-depressant -medication. The Great Depres- -sion is not a disease. Thus, -Prozac cannot treat the Great -Depression. So the answer is -no. - -The Great Depression is a time pe- -riod. Prozac is used to treat de- -pression (the mental illness). Thus, -the Great Depression can not be -treated with Prozac. So the answer -is no. - -[ARC-easy] Which source of -energy is nonrenewable? (a) -coal. -(c) water. -(d) wood. - -(b) sunlight. - -Coal, -sunlight, and water are -all renewable sources of energy. -Wood is a nonrenewable source of -energy. So the answer is (d). - -Students -[ARC-challenge] -studied the revolution of Earth -around the Sun. During which -month will Florida have the -most sunlight energy avail- -able? (a) March. (b) June. (c) -September. (d) December. +Aa -The Earth is closest to the Sun -in December, hence Florida will -have the most sunlight energy -available in December. So the an- -swer is (d). +aa -A nonrenewable source of en- -ergy is one in which the source -is not replenished in a short -amount of time. Thus, the only -option that is nonrenewable is -coal. So the answer is (a). +A -Coal is a fossil fuel and is formed -from dead organisms from the past. -Coal cannot be produced in a short -period of time, so it is considered -a nonrenewable source of energy. -So the answer is (a). +a -Earth is closest to the Sun on -June 21, the summer solstice. -So the answer is (b). + EXAMPLE -Florida is closer to the sun during -June, so it will have the most sun- -light energy during that month. So -the answer is (b). +• Dan and Kim are going +to have a baby. Kim +has dimples in her +cheeks (a dominant +trait), while Dan does +not. -Table 13: Additional examples where self-consistency helps repair the errors over greedy decode on -PaLM-540B. Two sampled reasoning paths that are consistent with the ground truth are shown. +• What are the chances +Dan and Kim’s baby +will have dimples? -A.3 FULL SETS OF PROMPTS + We know Dan and Kim’s +phenotypes (no dimples/dimples), +but what are their genotypes? -We list the full details of the prompts used for two newly-introduced datasets, AQUA-RAT (Ling -et al., 2017) and AI2 Reasoning Challenge (ARC) (Clark et al., 2018), where we manually composed -the example chain-of-thought in this paper, in Table 14 and Table 15, respectively. +• Dan has the recessive trait (no dimples) -Table 14: Few-shot exemplars for AQUA-RAT. -Q: John found that the average of 15 numbers is 40. If 10 is added to each number then the mean of the -numbers is? Answer Choices: (a) 50 (b) 45 (c) 65 (d) 78 (e) 64 +– He must have two recessive alleles +– Dan’s genotype can be represented as ‘dd’ -A: If 10 is added to each number, then the mean of the numbers also increases by 10. So the new mean -would be 50. The answer is (a). +• Kim has the dominant trait (dimples) -Q: If a / b = 3/4 and 8a + 5b = 22,then find the value of a. Answer Choices: (a) 1/2 (b) 3/2 (c) 5/2 (d) 4/2 (e) -7/2 +– But Kim could be homozygous dominant OR -A: If a / b = 3/4, then b = 4a / 3. So 8a + 5(4a / 3) = 22. This simplifies to 8a + 20a / 3 = 22, which means -44a / 3 = 22. So a is equal to 3/2. The answer is (b). +heterozygous dominant -Q: A person is traveling at 20 km/hr and reached his destiny in 2.5 hr then find the distance? Answer Choices: -(a) 53 km (b) 55 km (c) 52 km (d) 60 km (e) 50 km +– Kim’s genotype can be either ‘DD’ or ‘Dd’ +– Which one is it? -A: The distance that the person traveled would have been 20 km/hr * 2.5 hrs = 50 km. The answer is (e). + More information -Q: How many keystrokes are needed to type the numbers from 1 to 500? Answer Choices: (a) 1156 (b) 1392 -(c) 1480 (d) 1562 (e) 1788 +• What if you knew something about Kim’s -A: There are 9 one-digit numbers from 1 to 9. There are 90 two-digit numbers from 10 to 99. There are 401 -three-digit numbers from 100 to 500. 9 + 90(2) + 401(3) = 1392. The answer is (b). +parents? -19 +• How could that help? - Published as a conference paper at ICLR 2023 + Kim’s Parents -Table 15: Few-shot exemplars for ARC easy/challenge. -Q: George wants to warm his hands quickly by rubbing them. Which skin surface will produce the most -heat? (a) dry palms. (b) wet palms. (c) palms covered with oil. (d) palms covered with lotion. +• As it turns out, Kim’s father has dimples in +both cheeks, while her mother does not +• Her mother must have the recessive trait +and therefore has to have the genotype +‘dd’ -A: Dry surfaces will more likely cause more friction via rubbing than other smoother surfaces, hence dry -palms will produce the most heat. The answer is (a). +• Kim’s father has the dominant trait, but we -Q: Which factor will most likely cause a person to develop a fever? (a) a leg muscle relaxing after exercise. -(b) a bacterial population in the bloodstream. (c) several viral particles on the skin. (d) carbohydrates being -digested in the stomach. +don’t know if he is a homozygote or +heterozygote. He could be ‘DD’ or ‘Dd’ +just like Kim! -A: Option (b), bacterial population is the most likely cause for a person developing fever. The answer is (b). +• But we still know what Kim’s genotype -Q: Which change in the state of water particles causes the particles to become arranged in a fixed position? -(a) boiling. (b) melting. (c) freezing. (d) evaporating. +must be. Why? -A: When water is freezed, the particles are arranged in a fixed position; the particles are still moving for all -other options. The answer is (c). + Kim’s genotype is ‘Dd’ -Q: When a switch is used in an electrical circuit, the switch can (a) cause the charge to build. (b) increase -and decrease the voltage. (c) cause the current to change direction. (d) stop and start the flow of current. +• Kim must have a recessive allele (d), since +that is all she could have inherited from +her mother -A: The function of a switch is to start and stop the flow of a current. The answer is (d). +• Since Kim has dimples, we know she -Table 16: Few-shot exemplars for HotpotQA (closed-book setting). +inherited a dominant allele (D) from her +father +– It doesn’t matter if Kim’s father is DD or Dd; +whichever it is, he passed on a ‘D’ to his +daughter -Q: Which magazine was started first Arthur’s Magazine or First for Women? + What are the chances Kim and Dan’s baby +will have dimples? -A: Arthur’s Magazine started in 1844. First for Women started in 1989. So Arthur’s Magazine was started first. -The answer is Arthur’s Magazine. +Kim -Q: The Oberoi family is part of a hotel company that has a head office in what city? +d -A: The Oberoi family is part of the hotel company called The Oberoi Group. The Oberoi Group has its head -office in Delhi. The answer is Delhi. +D -Q: What nationality was James Henry Miller’s wife? +Dan -A: James Henry Miller’s wife is June Miller. June Miller is an American. The answer is American. +d -Q: The Dutch-Belgian television series that "House of Anubis" was based on first aired in what year? +d -A: "House of Anubis" is based on the Dutch–Belgian television series Het Huis Anubis. Het Huis Anubis is first -aired in September 2006. The answer is 2006. + Kim -As additional information, we also list the exact set of prompts used for all arithmetic reasoning tasks -in Table 17, since there are multiple sets of prompts introduced in Wei et al. (2022). The prompts for -CommonsenseQA and StrategyQA are the same as used in Wei et al. (2022). +d -We provide the exact prompts used for common NLP tasks in the following tables as well, including -NLI (Table 18, Table 19, Table 20) and Closed-Book Question-Answering tasks (Table 16, Table 21). +D -20 +Dd - Published as a conference paper at ICLR 2023 +dd -Table 17: Few-shot exemplars for all arithmetic reasoning tasks, from Wei et al. (2022). -Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done, -there will be 21 trees. How many trees did the grove workers plant today? +Dd -A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted. -So, they must have planted 21 - 15 = 6 trees. The answer is 6. +dd -Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot? +Dan -A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5. +d -Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total? +d -A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74 -chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39. +50% chance the baby will have the genotype ‘Dd’ and have +dimples -Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops -did Jason give to Denny? +50% chance the baby will have the genotype ‘dd’ and not have +dimples -A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of -lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8. + Clinical Application -Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does -he have now? +P -A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so -in total he has 7 + 2 = 9 toys. The answer is 9. + Clinical Application continued . . . -Q: There were nine computers in the server room. Five more computers were installed each day, from -monday to thursday. How many computers are now in the server room? +P -A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 = -20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers. -The answer is 29. + Other things may change us, but we start and +end with family -Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many -golf balls did he have at the end of wednesday? +-Anthony Brandt -A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On -Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33. +If you look deeply into the palm of your hand, +you will see your parents and all generations of +your ancestors. All of them are alive in this +moment. Each is present in your body. You are +the continuation of each of these people. -Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left? +-Thich -A: She bought 5 bagels for $3 each. This means she spent 5 * $3 = $15 on the bagels. She had $23 in -beginning, so now she has $23 - $15 = $8. The answer is 8. +Nhat -21 +Hanh - Published as a conference paper at ICLR 2023 + Questions? -Table 18: Few-shot exemplars for ANLI. + Adapted from a Presentation Created by the University +of North Carolina DNA Day Program -Premise: -"Conceptually cream skimming has two basic dimensions - product and geography." -Based on this premise, can we conclude the hypothesis "Product and geography are what make cream skimming -work." is true? -OPTIONS: -- yes -- no -- it is not possible to tell +• Sponsors -A: Based on "cream skimming has two basic dimensions" we can’t infer that these two dimensions are what -make cream skimming work. The answer is it is not possible to tell. +– NC Glaxo Smith Kline Foundation +– National Human Genome Research Institute +– Sigma XI Scientific Research Society -Premise: -"One of our member will carry out your instructions minutely." -Based on this premise, can we conclude the hypothesis "A member of my team will execute your orders with -immense precision." is true? -OPTIONS: -- yes -- no -- it is not possible to tell +• Collaborative Partners -A: "one of" means the same as "a member of", "carry out" means the same as "execute", and "minutely" means -the same as "immense precision". The answer is yes. - -Premise: -"Fun for adults and children." -Based on this premise, can we conclude the hypothesis "Fun for only children." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "adults and children" contradicts "only children". The answer is no. - -Premise: -"He turned and smiled at Vrenna." -Based on this premise, can we conclude the hypothesis "He smiled at Vrenna who was walking slowly behind -him with her mother." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: the premise does not say anything about "Vrenna was walking". The answer is it is not possible to tell. - -Premise: -"well you see that on television also" -Based on this premise, can we conclude the hypothesis "You can see that on television, as well." is true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "also" and "as well" mean the same thing. The answer is yes. - -Premise: -"Vrenna and I both fought him and he nearly took us." -Based on this premise, can we conclude the hypothesis "Neither Vrenna nor myself have ever fought him." is -true? -OPTIONS: -- yes -- no -- it is not possible to tell - -A: "Vrenna and I both" contradicts "neither Vrenna nor myself". The answer is no. - -22 - - Published as a conference paper at ICLR 2023 - -Table 19: Few-shot exemplars for e-SNLI (the rationales are crowd-sourced annotations from the -original dataset). - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is training his horse for a competition." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: The person is not necessarily training his horse. The answer is it is not possible to tell. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is at a diner, ordering an omelette." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: One jumping horse cannot be in a diner ordering food. The answer is no. - -Premise: -"A person on a horse jumps over a broken down airplane." -Based on this premise, can we conclude the hypothesis "A person is outdoors, on a horse." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: A broken down airplane is outdoors. The answer is yes. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "They are smiling at their parents." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A: Just because they are smiling and waving at a camera does not imply their parents or anyone is anyone behind -it. The answer is it is not possible to tell. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "The kids are frowning." is true? OPTIONS: -- yes -- no -- it is not possible to tell -A: One cannot be smiling and frowning at the same time. The answer is no. - -Premise: -"Children smiling and waving at camera." -Based on this premise, can we conclude the hypothesis "There are children present." is true? -OPTIONS: -- yes -- no -- it is not possible to tell -A:The children must be present to see them smiling and waving. The answer is yes. - -23 - - Published as a conference paper at ICLR 2023 - -Table 20: Few-shot exemplars for RTE. - -Premise: -"No Weapons of Mass Destruction Found in Iraq Yet." -Based on this premise, can we conclude the hypothesis "Weapons of Mass Destruction Found in Iraq." is true? - -A: "No Weapons of Mass Destruction Found" contradicts "Weapons of Mass Destruction Found". The answer is -no. - -Premise: -"A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful -gathered in downtown Chicago to mark the installation of new Pope Benedict XVI." -Based on this premise, can we conclude the hypothesis "Pope Benedict XVI is the new leader of the Roman -Catholic Church." is true?’ - -A: "installation of new Pope Benedict XVI." means "Pope Benedict XVI is the new leader". The answer is yes. - -Premise: -"A man is due in court later charged with the murder 26 years ago of a teenager whose case was the first to -be featured on BBC One’s Crimewatch. Colette Aram, 16, was walking to her boyfriend’s house in Keyworth, -Nottinghamshire, on 30 October 1983 when she disappeared. Her body was later found in a field close to her -home. Paul Stewart Hutchinson, 50, has been charged with murder and is due before Nottingham magistrates -later." -Based on this premise, can we conclude the hypothesis "Paul Stewart Hutchinson is accused of having stabbed a -girl." is true? - -A: The premise does not say Paul Stewart Hutchinson "stabbed" this girl. The answer is no. - -Premise: -"Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it -will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients." -Based on this premise, can we conclude the hypothesis "Herceptin can be used to treat breast cancer." is true? +– North Carolina Department of Public Instruction +– Destiny Traveling Science Learning Program +– North Carolina Association for Biomedical Research +– National Human Genome Research Institute +– American Society for Human Genetics -A: "Herceptin was approved to treat breast cancer" implies that "Herceptin can be used to treat breast cancer". -The answer is yes. - -Table 21: Few-shot exemplars for BoolQ (closed-book setting). - -Q: does system of a down have 2 singers? - -A: System of a Down currently consists of Serj Tankian, Daron Malakian, Shavo Odadjian and John Dolmayan. -Serj and Daron do vocals, so the band does have two singers. The answer is yes. - -Q: do iran and afghanistan speak the same language? - -A: Iran and Afghanistan both speak the Indo-European language Persian. The answer is yes. - -Q: is a cello and a bass the same thing? - -A: The cello is played sitting down with the instrument between the knees, whereas the double bass is played -standing or sitting on a stool. The answer is no. - -Q: can you use oyster card at epsom station? - -A: Epsom railway station serves the town of Epsom in Surrey and is not in the London Oyster card zone. The -answer is no. +• North Carolina Educators -24 +– Amy Bradley, Hibriten High School; Cindy Byron, School of Inquiry and Life +Sciences at Asheville (SILSA); Anita Crowley, Lee County High School; +MaryAnne Gore, West Brunswick High School; Terry Howerton, The School of +Biotechnology at Atkins; Michael Kendall, East Bladen High School; Crystal +McDowell; Letitia Myles, E.E. Smith High School; Elizabeth Pressley, Bartlett +Yancey High School; Steven Ross, Harnett Central High School; Murphey +Wellman, North Iredell High School; Anita Willington, East Bladen High School; +Linda Woody, Reidsville High School diff --git a/q144/random_k4/random_3.md b/q144/random_k4/random_3.md index 692dc5af339caacaddf265fe3e57e4a7f5067249..d0e4711d45467ad9a4bb5f563733b6d5247d3efa 100644 --- a/q144/random_k4/random_3.md +++ b/q144/random_k4/random_3.md @@ -1,1533 +1,806 @@ -GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS +Music and streaming +market study final report + +Executive summary + +29 November 2022 + + Executive summary + +Introduction + +1. + +2. + +Music streaming has transformed how consumers listen to music. The rise of +music streaming has given consumers easy access to large catalogues of +music covering an array of genres and time periods for a fixed monthly price, +or free with ads. As music streaming services have grown in popularity, +consumer outcomes have improved significantly. Between 2009 and 2021 the +monthly price of individual music streaming subscriptions has fallen by more +than 20% in real terms because the price of these plans has not kept pace +with inflation. At the same time, consumers have gained access to more +music and more innovative and better quality services, for example higher +quality audio, new video content and synced song lyrics. + +Consumers have widely adopted music streaming +– in the UK in 2021 there were 39 million monthly +active users of music streaming services and +there were over 138 billion streams. Streaming is +now the primary means for artists and labels to +distribute music and has been pivotal in securing +the sector’s recovery from piracy. + +1 + + 3. + +4. + +5. + +6. + +7. + +Some parts of the market have improved for artists in recent years, with more +choice about the type of deals with record labels available and more able to +directly release their music on streaming services. Average royalty rates in +major deals for new artists have increased steadily from an average of 19.7% +in 2012 to 23.3% in 2021. For songwriters, the share of revenues going to +publishing rights has increased significantly from 8% in 2008 to 15% in 2021. + +Whilst outcomes are good for consumers and generally improving for +creators, we note that to some extent changes in the sector, precipitated by +streaming, have made it harder for some creators. Reduced barriers to entry +and more choice on how to distribute music has meant there are more artists +than ever and, therefore, creators face more artists and songs to compete +with for streaming revenues. Not only that, but the convenience of streaming +means that older music is enjoying a resurgence of popularity, meaning that +today’s artists need to work harder than ever to grab listeners’ attention. + +These factors may be exacerbated by the fact that it is challenging for music +companies to know who among the growing pool of creators will be +successful. This inherent uncertainty combined with consumer tastes that +tend to tip to a relatively small number of artists means that it is challenging +for creators to succeed. We do not think that these factors arise from how +firms compete in the market. + +We have found that it is unlikely that the outcomes that concern many +stakeholders are primarily driven by competition. Consequently, it is unlikely +that a competition intervention would improve outcomes overall, and release +more money in the system to pay creators more. In such circumstances, there +is a greater risk that a competition intervention will result in unintended +consequences and worse outcomes for both consumers and creators. The +costs, risks and uncertainty created by a market investigation (which could run +for two years) would be imposed on the industry and borne, ultimately, by +consumers. We have therefore decided to not undertake a market +investigation. + +While there is limited potential for a competition intervention to improve +outcomes, there remains a broader policy debate about the optimal +distribution of existing revenues. We think it is a matter for Government and +policymakers to determine whether the current split is appropriate and fair, +and to explore whether wider policy interventions are required, for example +those relating to the copyright framework and how music streaming licensing +rates are set. We hope that our final findings provide insights that will be +helpful for that continuing debate. + +2 + + The effect of digitisation on the music industry + +8. + +9. + +The introduction of the internet made music piracy easier than had previously +been the case. Whilst piracy had always been a problem in the industry, the +internet made it easier to illegally copy works and share them with other +internet users on a large scale. The result was a collapse in music industry +revenues, and therefore creator revenues, as CD and other physical sales +declined. Inflation adjusted UK recorded music revenues fell by around 60% +from £1.9 billion in 2001 to £0.8 billion in 2015. + +Services that allowed consumers to access digital music legally, initially +through paid downloads of songs followed by music streaming, meant that for +the first time music companies and creators could monetise their content on +digital services and stem the flow of revenue losses. Since the introduction of +streaming, adjusting the older revenue figure for inflation music revenues +have increased from £0.8 billion in 2015 to £1.1 billion in 2021, although these +revenues remain below their £1.9 billion peak in 2001. + +UK inflation-adjusted recorded music revenues between 2000 and 2021 by format +type + +10. + +Consumers have embraced legal music +streaming and today music streaming accounts +for more than 80% of music sales. Data published +by Ofcom indicates that 47% of the population +made weekly use of music streaming services in +early 2022. This has nearly doubled since 2017, +but has remained relatively stable since 2020, +indicating that it might be plateauing. Younger +people stream music the most, with 77% of 15- + +3 + + 11. + +12. -GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT -OF INFORMATION TECHNOLOGY CAPITAL ASSETS +34-year-olds streaming music on a weekly basis compared with just 19% of +those aged 55+.1 + +Streaming has changed not only how we listen to music but what we listen to +because for the first time all music, both old and new, is readily available in +one place at no additional cost. This has created opportunities for labels, +publishers, and creators to reach new audiences who may not have heard the +music on its first release which has in turn extended the lifecycle for earning +revenue from songs. This is a benefit to those creators whose music +continues to be listened to, but this development is not necessarily good for all +creators because it means that today’s new music competes with yesterday’s +songs for a share of streaming revenue. + +Digitisation and new online business models continue to create new +opportunities to consume music. Today, besides new online radio stations, a +consumer may hear a piece of music in a video on TikTok, see their favourite +artist on a live stream rather than in concert, or engage with music alongside +new interactive online gaming platforms. Whilst these are different to +streaming services with full catalogues and music available on-demand, they +are new ways that consumers hear and discover music and offer potential +new sources of revenue for the music companies, artists, and songwriters. -Table of Contents +The recorded music sector is concentrated, but that is not driving +the concerns raised by artists -300.1 What is the purpose of the section? -300.2 Does the Exhibit 300 apply to me? -300.3 What background information must I know? -300.4 What special terms should I know? -300.5 How will agencies manage capital assets? -300.6 What other requirements does Exhibit 300 fulfill? -300.7 What must I report on Exhibit 300A&B and when? -300.8 How will OMB use Exhibit 300A&B? +13. -Ex–300A&B IT Capital Asset Summary and Performance Measurement Report +14. -Summary of Changes +Our analysis shows that more creators than ever are releasing music, +doubling from 200,000 in 2014 to 400,000 in 2020. This is, in part, made +possible by innovations in technology and the ability to easily distribute music +online. For example, it is now easier than ever to create and record music +outside the confines of a traditional music studio and share it on streaming +services without the need for a record label. -Introduces new definitions to assist Agencies in the consistent planning, acquisition, and operation of IT -assets +It has long been the case in recorded music that only a very small minority of +artists will achieve the highest level of success. In 2020 over 60% of streams +were of music recorded by only the top 0.4% of artists. To reach the top of the +chart, songs need to be streamed many millions of times each day and many +artists are faced with a situation where their work can be streamed millions of +times, but it does not translate into a significant share of their income. For +example, twelve million streams a year could earn an artist around £12,000, +but less than 1% of artists achieve that number of streams. We have heard -Changes Unique Project Identifier to Unique Investment Identifier +1 Data published by Ofcom (2022), Media Nations: UK 2022. -Introduces Exhibits 300 A&B, “IT Capital Asset Summary and Performance Measurement Report” +4 -300.1 What is the purpose of the section? + from both artists and songwriters that they are unable to make a sustainable +income from music streaming and for many it feels unfair. -Exhibits 300s establish policy for planning, budgeting, acquisition and management of major information -technology (IT) capital investments. The Office of Management and Budget (OMB) provides procedural and -analytic guidelines for implementing specific aspects of these policies described in OMB Circulars and their -associated appendices. +15. We have considered what may be causing these outcomes for artists2 and, in -Exhibit 300s are companions to an agency’s Exhibit 53. Exhibit 300s and the Exhibit 53, together with the -agency’s Enterprise Architecture program, define how to manage the IT Capital Planning and Control Process. -Exhibit 53A is a tool for reporting the funding of the portfolio of all IT investments within a Department while -Exhibit 300A is a tool for detailed justifications of major "IT Investments". Exhibit 300B is for the management -of the execution of those investments through their project life cycle and into their useful life in production. By -integrating the disciplines of architecture, investment management, and project implementation, these programs -provide the foundation for sound IT management practices, end-to-end governance of IT capital assets, and the -alignment of IT investments with an agency’s strategic goals. As architecture-driven IT investments are funded in -the "Invest" phase, they move forward into the implementation phase where system development life cycle -processes are followed and actual versus planned outputs, schedule, and operational performance expenditures are -tracked utilizing performance-based management processes. New for the FY 2013 budget process, Exhibit 300B -requires agencies to provide more detailed benchmarks for the management and performance of projects and -operational assets associated with a major investment. +line with our statutory duties, whether these outcomes are being caused by +competition issues in the market. -300.2 +16. - Does Exhibit 300 apply to me? +The recorded music sector is concentrated, with the 3 major labels holding a +combined share of over 70% of UK streams, and this has persisted for some +time. The market share in terms of streams of independent record companies +(indies) has remained steady at around one quarter for several years. This +share is very fragmented with only 2 indies having a share in excess of 1%. -The policy and budget justification principles in this Exhibit apply to all agencies of the Executive Branch of the -Government subject to Executive Branch review (see Section 25). Exhibit 300A&B must be submitted for each -major IT investment for CIO Council agencies in accordance with this section, parallel requirements for other +Label shares of total UK streams in 2021 -Page 1 of 21 Exhibit 300 +Source: CMA analysis of data from Official Charts.3 -(2011) +2 As described in Chapter 5 of this report, a number of our findings on artists will also relate to songwriters. +Issues specific to songwriters are considered in more detail below. +3 Notes: This pie chart is for illustrative purposes only. These figures are provided in a 5% range where the figure +is below 10%, and a 10% range where the figure is between 10% and 100%. The midpoints of the ranges have +been used to provide an illustration of relative size in the market. Where the sum of these midpoints does not +equal 100%, we have scaled the pie chart so that the area segments represent the share of the sum of the +midpoints. - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS +5 -agencies are addressed in separate guidance. Major IT investments also must be reported on the agency’s Exhibit -53. + 17. -300.3 +The scale of the majors and their global reach means they can offer large +advances which attracts proven and successful artists. In turn, this can make +it difficult for indie labels to attract and retain artists as they become +successful, which can create a barrier to expansion for indie labels. - What background information must I know? +18. Despite the concentrated nature of the market, outcomes for artists as a -The Federal Government must effectively manage its portfolio of capital assets to ensure scarce public resources -are wisely invested. Capital programming integrates the planning, acquisition and management of capital assets -into the budget decision-making process and is intended to assist agencies in improving asset management and in -complying with the results-oriented requirements of: +whole seem to be improving. We accept that this improvement may not be +benefiting all artists, and that for many artists the improvement will seem +insufficient. -• The Clinger-Cohen Act (CCA) of 1996, Public Law 104 – 106, legislatively mandates that IT +19. Our analysis shows that in some respects the options available to artists, -investments be prudently managed. CCA requires federal agencies to focus on the results achieved -through IT investments while streamlining the federal IT procurement process. Congress and OMB have -clearly stated that each agency must actively manage its IT program to provide assurances that technology -expenditures are necessary and shall result in demonstrated improvements in mission effectiveness and -customer service. +particularly new artists, are improving: -• The Federal Acquisition Streamlining Act of 1994, Title V (FASA V), which requires agencies to -establish cost, schedule and measurable performance goals for all major acquisition programs, and -achieve on average 90 percent of those goals. +(a) There is now more choice for artists about which type of deal they would +like to agree, from DIY distribution, A&L services, or more traditional +record deals. -• Security: For IT investments, agencies should maintain up-to-date tracking of systems in the FISMA +(b) Some new artists have greater leverage when negotiating a record deal -inventory to the appropriate IT investment. Costs for security will be collected in both the Exhibit 53A -and 53B. +with a label if they have already built a strong fanbase and online +presence. -• Enterprise Architecture (EA): The IT investment must be included in the agency’s EA and Capital +Average UK artist yearly streaming earnings from majors and average +(mean) royalty rates (2021 prices) in the UK -Planning and Investment Control (CPIC) process and mapped to and supporting the Federal Enterprise -Architecture (FEA). The business case must demonstrate the relationship between the investment and the -business, performance, data, services, application, and technology layers of the agency’s EA. Are this -investment and its assets included in your agency’s target enterprise architecture through the underlying -segment architecture? For detailed guidance regarding segment architecture requirements, please refer to -http://www.whitehouse.gov/omb/e-gov/. See this guidance also regarding the reporting of six digit codes -corresponding to agency segment architectures in Exhibit 53, and, for limited cases determined by the -Chief Architect, reporting an investment alignment with multiple segments. +Source: CMA analysis of data from the majors -• Additional background information for Information Technology can be found in Section 53.2 of the +20. Analysis of new contracts signed by the majors with new artists for multi-track -guidance for Exhibit 53. +deals (ie albums or extended play (EP) records) shows that, between 2012 +and 2021: -300.4 What special terms should I know? +(a) the average gross royalty rate has increased from 19.7% to 23.3%; -Alternatives Analysis refers to an analysis of alternative approaches addressing the performance objectives of an -investment, performed prior to the initial decision to implement a solution, and updated periodically as -appropriate to capture changes in the context for an investment decision. Alternatives analysis details should be -available upon request. +6 -Asset refers to anything that has value to an organization, including, but not limited to: computing device, -information technology (IT) system, IT network, IT circuit, software (both an installed instance and a physical -instance), virtual computing platform (common in cloud and virtualized computing), and related hardware (e.g., -locks, cabinets, keyboards). Assets are the lowest level at which information technology is planned, acquired, -implemented and operated. + (b) the proportion of contracts where labels own copyright of recordings in -Capital assets means land, structures, equipment, intellectual property (e.g., software), and information -technology (including the output of IT service contracts) used by the Federal Government and having an +perpetuity has reduced from 66% to 26.4%; and -Page 2 of 21 Exhibit 300 +(c) the average number of minimum commitment periods (where a period is +defined by a commitment to produce a multi-track output such as an +album) has fallen from 3.8 to 3.4 -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -estimated useful life of two years or more. See Appendix One of the Capital Programming Guide for a more -complete definition of capital assets. - -Capital investment means the planning, development, acquisition of a capital asset and the management and -operation of that asset through its usable life after the initial acquisition. IT Capital investments may consist of -one or more assets, the planning, development and acquisition of which are managed through projects, and which -then provide useful components in an operational (production) environment. - -Capital programming means an integrated process within an agency for planning, budgeting, procurement and -management of the agency’s portfolio of capital assets to achieve agency strategic goals and objectives with the -lowest overall cost and least risk. - -Cost means the expenditure of funds or use of property to acquire, produce, operate or maintain an asset. -Examples include, but are not limited to: sunk costs, operational costs, acquisition costs and disposition costs -(including variable costs such as labor hours). +21. Competition appears to be particularly focused on artists who are already -Cost saving represents the reduction in actual expenditures to achieve a specific objective (as defined in OMB -Circular A–131). Cost savings should be cited in descriptions, and may be included as a benefit in alternative -analyses. +popular or are likely to be. Competition to sign such artists can be very +intense with offers from many labels. Traditional record deals also face +increasing disruption from alternative models, in particular service deals from +artist and label (A&L) service providers, and there are more options than ever +for artists to reach audiences and monetise their work. -Cost avoidance represents results from an action taken in the immediate time frame that will decrease costs in the -future (as defined in OMB Circular A–131). Cost avoidance should be cited in descriptions, and may be included -in alternative analyses. +22. We think that outcomes for artists are driven by factors which are largely -Defense Acquisition Workforce Improvement Act (DAWIA) (P.L. 101-510 of November 5, 1990) established, -for the Department of Defense (DoD), an Acquisition Corps to professionalize the acquisition workforce in DoD -through education, training, and work experience. +unrelated to competition issues in the market. Rather, we think these +outcomes can be attributed to factors more inherent to how music streaming +works. Digitisation has allowed for a huge increase in the number of artists +sharing their music and a vast back catalogue made available via streaming +so there is more music available to stream and consumer tastes tend to tip +towards a relativity small number of artists being successful. In addition, there +is very significant uncertainty about which artists will be successful. The +combination of these market features is likely to result in competition focused +on a relatively limited number of artists and market outcomes where the +majority of the benefits are accrued by a minority of artists. -Dependency means the identification of relationships between projects and operational assets within an -investment, and identification of relationships between investments. Identification of dependencies is critical to -the management of project, program, and portfolio risk. - -Development, Modernization and Enhancement (DME) Costs are costs for projects leading to new IT -assets/systems and projects that change or modify existing IT assets to: substantively improve capability or -performance; implement legislative or regulatory requirements; or to meet an agency leadership request. Capital -costs as part of DME can include hardware, software development and acquisition costs, COTS acquisition costs, -government labor costs, and contracted labor costs for planning, development, acquisition, system integration, and -direct project management and overhead support. - -Disposition Cost for an asset refers to the cost of retiring a capital asset once its useful life is completed or it has -been superseded by a replacement asset, and may be included in operational costs. - -Earned Value Management (EVM) is a project management tool effectively integrating the project scope of -work with schedule and cost elements for optimum project planning and control. The qualities and operating -characteristics of earned value management systems (EVMS) are described in American National Standards -Institute (ANSI)/Electronic Industries Alliance (EIA) Standard –748–1998, Earned Value Management Systems, -approved May 19, 1998. It was reaffirmed on August 28, 2002. Additional information on EVMS is available at -www.acq.osd.mil/pm. - -Enterprise Architecture (EA) is the explicit description and documentation of the current and desired -relationships among business and management processes and information technology of an organization. It -describes the "current architecture" and "target architecture" to include the rules and standards and systems life -cycle information to optimize and maintain the environment which the agency wishes to create and maintain by - -Page 3 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -managing its IT portfolio. The EA must also provide a strategy to enable the agency to support its current state -and also act as the roadmap for transition to its target environment. The EA will define principles and goals and -set direction on such issues as the promotion of interoperability, open systems, public access, end user -satisfaction, and IT security. The agency must support the EA with a complete inventory of agency information -resources, including personnel, equipment, and funds devoted to information resources management and -information technology, at an appropriate level of detail. - -Federal Acquisition Certification for Program and Project Managers (FAC-P/PM) was established to ensure -general training and experience requirements for program and project managers are clearly identified for civilian -agencies. The FAC-P/PM focuses on essential competencies needed for program and project managers; the -program does not include functional or technical competencies, such as those for information technology or -agency-specific competencies. Defense agencies have a similar certification program under the Defense -Acquisition Workforce Improvement Act (DAWIA). Agencies were required to be compliant with FAC-P/PM -starting in FY 2008. Available levels are Entry/Apprentice, Mid/Journeyman and Expert/Advanced for FAC- -P/PM and 1, 2 and 3 for DAWIA. -(http://www.whitehouse.gov/sites/default/files/omb/procurement/workforce/fed_acq_cert_042507.pdf ) -(www.whitehouse.gov/omb/procurement/acq_wk/fac_contracting_program.pdf) - -Federal Enterprise Architecture (FEA) is a business-based documentation and analysis framework for -government-wide improvement. The FEA allows agencies to use standardized methods to describe the -relationship between an agency’s strategic goals, business functions, and enabling technologies at various levels -of scope and complexity. The FEA is comprised of documentation in six domain areas (strategic goals, business -services, data and information, systems and applications, infrastructure, and security) and six reference models -areas that are designed to facilitate standardized analysis, reporting, and the identification of duplicative -investments, gaps, and opportunities for collaboration within and across federal agencies. More information -about the FEA and reference models is available at http://www.whitehouse.gov/omb/e-gov/fea. - -Federal Segment Architecture Methodology (FSAM) is to become the “Federal Solution Architecture -Methodology” in October 2011 and will serve as a scalable and repeatable process for solution architecture at the -application, system, segment, enterprise, sector, government-wide, national, and international levels of scope. -Consistent use of the FSAM should result in more complete and consistent architecture products by helping -architects engage system owners, program offices, and executive sponsors to deliver value-added plans for -improved mission delivery. Specifically, FSAM includes guidance to help architects establish clear relationships -among strategic goals, detailed business / information management requirements, and measurable performance -improvements within each area of the agency’s enterprise architecture. - -Full funding means appropriations are enacted sufficient in total to complete an asset or useful component (see -definition below) of a capital asset before any obligations may be incurred for the component. Incrementally -funding the planning and acquisition of capital assets or (useful components) without certainty if or when future -funding will be available can result in poor planning, inadequate justification of assets acquisition, higher -acquisition costs, project delays, cancellation of projects, the loss of sunk costs, and inadequate funding to -maintain and operate the assets. Budget requests for full acquisition of capital assets must propose full funding -(see Section 31.5). - -Funding means providing the budgetary resources to plan for, acquire, develop, sustain, or operate an asset. - -Information Technology means any equipment or interconnected system or subsystem of equipment that is used -in the automatic acquisition, storage, manipulation, management, movement, control, display, switching, -interchange, transmission or reception of data or information by an executive agency. Information Technology is -related to the terms Capital Asset, IT Investment, Program, Project, Sub-project, Service, and System. - -Integrated Program Team (IPT) means cross-functional or multidisciplinary group of individuals organized and -collectively responsible for the specific purpose of delivering a project/product/or process to an external or - -Page 4 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -internal customer. Each IPT should include experts in program and project management, resource management, -procurement, and systems engineering, security, and other disciplines, as necessary, to evaluate all aspects of the -project. An IPT must always include, but is not limited to: a qualified fully-dedicated IT program manager, a -contract specialist, an Information technology specialist, a security specialist and a business process owner before -OMB will approve this program investment budget. Key members of the IPT will also be co-located during the -most critical junctures of the program, to the maximum extent possible. Agencies should establish integrated -program team members individual performance goals to hold team members accountable for both individual -functional goals and overall program success. IPT should be defined in a program or an IPT charter. - -Interagency acquisition means the use of the Federal Supply Schedules, a multi-agency contract (i.e., a task order -or delivery order contract established by one agency for use by government agencies to obtain supplies and -services, consistent with the Economy Act, 31 U.S.C. 1535), or a government-wide acquisition contract (i.e., a -task-order or delivery-order contract for information technology established by one agency for government-wide -use operated by an executive agent designated by OMB pursuant to Section 11302(3) of the Clinger Cohen Act of -1996). - -Federal IT Dashboard The purpose of the Dashboard is to provide information on the effectiveness of -government IT programs and to support decisions regarding the investment and management of resources. The -Dashboard is now being used by the Administration and Congress to make budget and policy decisions. - -IT Investment means the expenditure of IT resources to accomplish mission objectives. An IT investment may -include a project or projects for the development, modernization, enhancement, or maintenance of a single IT -asset or group of IT assets with related functionality and the subsequent operation of those assets in a production -environment. While each asset or project would have a defined life cycle, an investment that covers a collection -of assets intended to support an ongoing business mission may not have a defined life cycle. - -IT Program Managers and IT Project Managers are defined by OPM in the Job Family Standard for -Administrative Work, in the Information Technology Group (series 2200 in the Federal Classification and Job -Grading Systems). IT Program Managers will be responsible for major investments and will lead the required -Integrated Program Team for the investment. - -Investment Title as defined in Guidance for Exhibit 53 (See Section 53.8). - -Life-cycle costs includes all investment costs (including government FTE), independent of the funding source, i.e. -revolving fund, appropriated fund, working capital fund, trust fund, etc. (see Capital Programming Guide of OMB -Circular A-11 and OMB Circular A-131). - -Maintenance is the activity necessary to keep an asset functioning as designed during its operations and -maintenance phase of an investment. Maintenance costs include costs needed to sustain an IT asset at the current -capability and performance levels including: corrective hardware/software, voice and data communications -maintenance; replacement of damaged or obsolete IT equipment; and associated overhead costs. Where -appropriate, maintenance activities that follow agency defined project management methodologies should be -managed and reported as projects and reported in Section B of the Exhibit 300B. Examples of maintenance -projects include operating system upgrades, technology refreshes, and security patch implementations. - -Major IT Investment means a program requiring special management attention because of its importance to the -mission or function of the agency, a component of the agency, or another organization; has significant program or -policy implications; has high executive visibility; has high development, operating, or maintenance costs; is -funded through other than direct appropriations; or, is defined as major by the agency’s capital planning and -investment control process. OMB may work with the agency to declare other investments as major investments. -Agencies should consult with your OMB agency budget officer or analyst about what investments to consider as -"major". Investments not considered "major" are "non-major." - -Page 5 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Operations mean the day-to-day management of an asset in the production environment and include activities to -operate data centers, help desks, operational centers, telecommunication centers, and end user support services. -Operational activities are reported through Section C of the Exhibit 300B. Operations costs include the expenses -associated with an IT asset that is in the production environment to sustain an IT asset at the current capability -and performance levels including: Federal and contracted labor costs; and costs for the disposal of an asset. - -Operations and Maintenance means the phase of an asset in which the asset is in operations and produces the -same product or provides a repetitive service. Operations and Maintenance (O&M) is synonymous with “steady -state.” - -Performance-based acquisition management means a documented, systematic process for program management, -which includes integration of program scope, schedule and cost objectives, establishment of a baseline plan for -accomplishment of program objectives, and use of earned value techniques for performance measurement during -execution of the program. This includes prototypes and tests to select the most cost effective alternative during -the Planning Phase, the work during the Acquisition Phase, and any developmental, modification, or upgrade -work done during the Operational/Steady State Phase. For operational/steady state systems, an operational -analysis as discussed in Phase IV of the Capital Programming Guide is required. A performance-based -acquisition (as defined in the Federal Acquisition Regulation 37.101) or contract/agreement with a defined quality -assurance plan that includes performance standards/measures should be the basis for monitoring contractor or in- -house performance of this phase. - -Planning means preparing, developing or acquiring the information you will use to: design the asset; assess the -benefits, risks, and risk-adjusted costs of alternative solutions; and establish realistic cost, schedule, and -performance goals, for the selected alternative, before either proceeding to full acquisition of the capital project or -useful component or terminating the project. Planning must progress to the point where you are ready to commit -to achieving specific goals for the completion of the acquisition before proceeding to the acquisition phase. -Information gathering activities may include market research of available solutions, architectural drawings, -geological studies, engineering and design studies, and prototypes. Planning may be general to the overall -investment or may be a useful component of a project. Depending on the nature of the project, one or more -planning components may be necessary. - -Program for the purposes of this Exhibit 300 is a group of assets that are planned and managed together to -achieve an overall set of related outcomes. IT Investment is frequently used as a synonym for IT program. - -Project is a temporary endeavor undertaken to accomplish a unique product or service with a defined start and end -point and specific objectives that, when attained, signify completion. Projects are undertaken for development, -modernization, enhancement, disposal, or maintenance of an IT asset. Projects are composed of activities. - -Project Manager (PM) Level of Experience is the specific certification or the number of years of direct project -management experience of the PM. Examples of PM certifications include FAC-P/PM, PMI, or other recognized -certifications. - -Risk Management is a systematic process of identifying, analyzing, and responding to risk. It includes -maximizing the probability and consequences of positive events and minimizes the probability and consequences -of adverse events to overall objectives. - -Solution Architecture is a standardized method of identifying business requirements and viable technology -solutions within the context of a single agency’s enterprise architecture, or a multi-agency sector or government- -wide/international architecture. Solution architecture includes current and future views as well as transition plans -at a number of levels of scope that include applications, systems, segments, enterprise, sector, government-wide, -national, and international. The Federal Solution Architecture Methodology (FSAM) is scheduled for release in -October 2011 to provide the repeatable process for doing solution architecture. +23. We therefore conclude that a competition intervention, for example a change -Page 6 of 21 Exhibit 300 +to the structure of the market, is unlikely to result in a material increase in +revenues for artists. -(2011) +24. Whilst majors’ profits have been increasing since the lows of piracy, our +profitability analysis has not found evidence of substantial and sustained +excess profits by the majors. This is consistent with our overall finding that - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS +4 As these are only averages across all 3 majors, they do not show how the terms can vary significantly between +artists, reflecting for example the different potential financial rewards and risks based on the characteristics of +individual artists (including by genre, potential, and stage of career). -Shared Service Provider is the provider of a technical solution and/or service that supports the business of -multiple agencies using a shared architecture. - -Unique Investment Identifier as defined in the guidance for Exhibit 53. +7 -Additional budget terms and definitions are included in the Glossary in Appendix J, "Principles of Budgeting for -Capital Asset Acquisitions" and in the guidance for Exhibit 53 (for IT). + there is unlikely to be scope to improve outcomes for artists substantially +through increased competition. -300.5 How will agencies manage information technology capital assets? +25. We understand that Government has taken several steps to address concerns -The Capital Programming Guide of OMB Circular A–11, provides guidance on the principles and techniques for -effective capital programming. Appendix J of this part explains the principles of financing capital asset -acquisitions. Section 9b of OMB Circular A–130 establishes additional requirements for enterprise architectures -(EAs), planning and control of information systems and technology investments and performance management. -Agencies must develop and implement a capital programming process to develop their capital asset portfolio, and -must: +about creator remuneration and has responded to the DCMS Select +Committee recommendations for legislative and policy reform in this area. In +particular, the IPO is conducting a research programme including examining +potential options to strengthen creator rights and remuneration. We are +sharing our final findings with the IPO and DCMS to help inform their work. -• Evaluate and select capital assets that will support core mission functions performed by the Federal +Labels could do more to improve the information they provide to +artists -Government, and which demonstrate projected returns on investment that are clearly equal to or better -than alternative uses of available public resources. Specifically for IT, the investments should be -informed by, and address gaps in an agency’s Segment Architecture and map to the agency’s strategic -plan; -Initiate improvements to existing assets or acquisitions of new assets only when no alternative private -sector or governmental source can more efficiently meet the need; -Implement IT reforms from the “25 Point Implementation Plan to Reform Federal Information -Technology Management” such as adhering to modular development principles and the requirement to -establish an Integrated Program Team prior to funding an IT investment; +26. We heard strong concern from some artists and their representatives that they -• +do not get enough information from record labels on how their earnings from +streaming services are calculated or how the deals that exist between labels +and streaming services may affect what they earn or could earn in future. +Non-disclosure agreements (NDAs) between music streaming services and +labels were cited as a barrier to improving information provision to artists. -• +27. Our analysis shows that some artists are provided with information, such as -• Simplify or otherwise redesign work processes to reduce costs, improve effectiveness, and make +number of streams and the royalties earnt on these streams, which tells them +how much they have earned per stream on music streaming services, and we +saw some positive examples of labels presenting this information in a user- +friendly way. However, this was not consistent across all labels, and we think +best practices could be developed so that information could be presented in a +more straightforward and uniform way with appropriate guidance on how to +interpret the data. This will help artists better understand how they are paid for +streaming and the sources of their income. -maximum use of commercial services and off-the-shelf technology; +28. -• Reduce project risk by avoiding or isolating custom designed components, using components that can be -fully tested or prototyped prior to full implementation or production, and ensuring involvement and -support of users in the design and testing of the asset; +In respect of NDAs, whilst they do prevent certain terms and conditions being +made available to artists and limit access to the ‘source data’ from the music +streaming services, they do not appear to prevent a significant amount of +relevant information being made available to artists about their earnings. By +limiting the access to source data, the NDAs may potentially limit the ability of +artists to verify the accuracy of information. However, this is an issue more to +do with verification of the fulfilment of contractual terms rather than one of +competition to sign artists. -• Structure major planning and acquisition into useful components with a narrow scope and brief duration, -and that make adequate use of competition and appropriately allocate risk between Government and -contractor. The Agency Head must approve or define the cost, schedule, and performance goals for major -acquisitions, and the agency's Chief Financial Officer must evaluate the proposed cost goals; +29. We welcome the work the IPO is undertaking on issues +around transparency for artists, including developing a +code of practice. -• +8 -• Ensure a continuous linkage with Federal, agency, and bureau EAs, demonstrating such consistency -through alignment with the agency’s Segment Architecture, compliance with agency business -requirements and standards, as well as identification of milestones, as defined in the agency’s EA -transition strategy; -Institute performance measures and management processes monitoring and comparing actual performance -to planned results. Agencies must use a performance-based acquisition management or earned value -management system, based on the ANSI/EIA Standard 748-B, to obtain timely information regarding the -progress of capital investments where appropriate, as defined in the Federal Acquisition Regulations -(FAR). The system must also measure progress towards useful components in an independently -verifiable basis, in terms of cost, capability of the investment to meet specified requirements, timeliness, -and quality. Agencies are expected to achieve, on average, 90 percent of the cost, schedule and -performance goals for major acquisitions. Through the Tech Stat process, agencies must review major -acquisitions not achieving 90 percent of the goals to determine whether there is a continuing need and -what corrective action, including termination, should be taken; - -• Ensure assets that are IT systems conform to the requirements of OMB Circular No. A–130, - -"Management of Federal Information Resources"; - -Page 7 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -• Ensure assets that are financial management systems conform to the requirements of OMB Circular No. - -A–127; - -• Conduct post-implementation or post-occupancy reviews of capital programming and acquisition - -processes and projects to validate estimated benefits and costs and document effective management -practices, i.e., lessons learned, for broader use; - -• Establish oversight mechanisms requiring periodic review of operational capital assets to determine how - -mission requirements might have changed, and whether the asset continues to fulfill ongoing and -anticipated mission requirements, deliver intended benefits to the agency and customers, and meet user -requirements; and - -• Develop and maintain the following documents all of which may be requested by OMB and are subject to - -delivery within ten business days: - -Investment level Alternative Analysis; -o -Investment level Acquisition Plan; -o -o Earned Value Reports on large projects; -Integrated Program Team Charter; -o -o -Investment Charter; -o Project Charters, as appropriate; and -o Risk Management Plan - -300.6 What other requirements do Exhibits 300s fulfill? - -The Exhibit 300A is designed to coordinate OMB’s collection of agency information for its reports to the -Congress required by the Federal Acquisition Streamlining Act of 1994 (FASA Title V) and the Clinger-Cohen -Act of 1996. The business case (Exhibit 300A) for investment should be demonstrated for supporting the mission -statements, long-term goals and objectives, and annual performance plans developed pursuant to the GPRA. In -addition, the Exhibit 300B establishes reporting requirements through the Federal IT Dashboard to ensure the -proper execution of those investments against the established performance plans. - -300.7 What must I report on Exhibits 300A&B and when? - -All information necessary to complete Exhibit 300A&B should already exist as part of the agency's overall -Capital Planning activities and within project and program specific documentation. The materials used to -populate Exhibit 300A&B should be readily available to OMB upon request. - -Exhibit 300As for major IT investments must be submitted to OMB during the period September 12-16, 2011. -The Exhibit 300A should be fully integrated with your agency’s overall budget submission, including Exhibit -53A. All reporting on IT must be submitted electronically via XML feed to OMB’s Federal IT Dashboard -(http://www.itdashboard.gov/). Additional information regarding the submission process will be posted on the IT -Dashboard including business rules for the validation of data and a schema that describes how data should be -transmitted. As with previous submissions to the IT Dashboard, pre-decisional and procurement sensitive -information will not be displayed to the public. - -As prescribed by OMB, agencies should update their Exhibits 53A and 300A to reflect final Presidential -decisions. - -Exhibit 300B for major IT investments should be initially submitted to OMB by September 23, 2011, in order to -establish cost, schedule and performance targets for FY2012. In line with modular development principles, targets -for projects should aim to deliver functionality within six months. For the purposes of target setting, projects -stretching over a longer period should be broken up into segments of six months or less and reported in Tables B1 -and Table B2 in those smaller pieces. Performance targets for ongoing operations will be established in Table C1. - -Page 8 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Following the establishment of FY 2012 targets, regular updates for metrics and activities will be submitted to the -IT Dashboard at least once by the last business day of each month. The periodicity of performance metrics -updates for ongoing operations will vary according to the nature of the metric as indicated in Table C1. - -If agencies request supplemental funds or reallocate funding within their authority, which include changes to the -agency's portfolio and the performance targets, as part of their supplemental request, agencies should submit new -or revised Exhibits 53A (see Guidance on Exhibit 53), 300A, and 300B. - -Multi-Agency Collaboration Investments (must submit both Exhibit 300A and Exhibit 300B) - -The managing partner (lead agency) will take the lead for completing the multi-agency collaboration investment -Exhibit 300, managing it through the lead agency's capital programming and budget process and submitting the -Exhibit 300 to OMB. The managing partner is also responsible for ensuring the Exhibit 300s include all necessary -information from the partner agencies and has been approved by all necessary partner agencies through the -appropriate governance process. The Executive Committee of each initiative/investment should review and -approve the Exhibit 300s. The tracking of the multi-agency approvals, partner agency funding, and related capital -assets (e.g. migration investments, Centers of Excellence, Shared Service Centers, Supporting components) will -be captured via a MAX E-Gov Funding Tool (similar to the one used in FY2010), all Initiative PMs or their -designated leads should have received details about this in the May 10, 2011 email from OMB. - -For IT assets, partner agencies should report their participation in their Exhibit 53 submissions as appropriate -(see Guidance on Exhibit 53 -). Partner agencies should reference the name and UII of the multi-agency Exhibit -300 in the "Investment Description" field of each Exhibit 53 line item related to the multi-agency Exhibit 300. -Partner agencies should also ensure their activities and participation are included in the appropriate sections of -the multi-agency Exhibit 300. The entire Summary of Funding Total for the investment, including funds -provided by partner agency, should be included in the Exhibit 300s. - -Investments that provide a service to other agencies, but do not receive contributions from partner -agencies should still be reported as Multi-Agency Collaboration Investments. - -Investments such as Lines of Business (LoB) that provide a cross-federal governance structure, but no technical -solution should also report as Multi-Agency Collaboration Investments. As governing bodies, they have oversight -responsibilities of the shared service providers that provide technical solutions and should inform and guide their -performance targets. These Exhibit 300s should not duplicate the costs including in the Exhibit 300 reported by -the shared service provider. The Cost and Schedule table should indicate key activities agreed to under the -investment that indicate an agreed upon roadmap strategy; however, only costs directly attributed to the -investment should be included. - -Managing partners providing multi-agency services should ensure that funding is prioritized to accommodate -building and obtaining approval of a shared architecture. Approval should be obtained through a working -governance model that includes partner agencies, customers and OMB. - -OMB may require additional information from partner agencies related to the multi-agency collaboration -investment Exhibit 300. When necessary, OMB will work with the managing partners to coordinate data requests. - -300.8 How will OMB use the Exhibit 300 A&B? - -The Exhibit 300 A&B is one component of your agency’s total budget justification (see Section 51.2). OMB uses -the Exhibit 300A&B to make both quantitative decisions about budgetary resources consistent with the -Administration’s program priorities, and qualitative assessments about whether the agency’s programming -processes are consistent with OMB policy and guidance. OMB will be evaluating all elements of the budget -submission and will communicate the results of these evaluations in the course of the budget process. If -additional supporting information is necessary, OMB will request from agencies the supporting evidence used to -produce the Exhibit 300A&B. All information necessary to complete an Exhibit 300A&B should already exist as - -Page 9 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -part of the agency's overall Information Technology Investment Management activities and within program and -project specific documentation. The materials used to produce the Exhibit 300A&B should be readily available to -OMB upon request. - -Page 10 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Exhibit 300A: IT Capital Asset Summary -Part I: Summary Information And Justification (All Capital Assets) - -Section A: Overview - -1. Investment Name: -2. Unique Investment Identifier (UII) For IT investment only, see -Section 53.9. - -Section B: Investment Detail - -1. Provide a brief summary of the investment, including a brief description of the related benefit to the mission -delivery and management support areas, and the primary beneficiary(ies) of the investment. Include an -explanation of any dependencies between this investment and other investments. - -[Limit: 2500 char] -2. How does this investment close in part or in whole any identified performance gap in support of the mission -delivery and management support areas? Include an assessment of the program impact if this investment isn't -fully funded. -[(Limit:2500 char] -3. For this investment’s technical features, please identify where any specific technical solutions are required -by legislation, in response to audit findings, or to meet requirements from other sources. Where “Yes” is -indicated, provide a brief description of the technical features required, and any citations regarding specific -mandates for these requirements. - -Legislative Mandate -Audit Finding Resolution -Published Agency Strategic Plan -Other Requirement - -Required by Description - -[Yes/No] - -[Limit: 1000 characters] - -4. Provide a list of this investment's accomplishments in the prior year (PY), including projects or useful -components/project segments completed, new functionality added, or operational efficiency achieved. -[Limit: 1000 char] -5. Provide a list of planned accomplishments for current year (CY) and budget year (BY). -[Limit: 2500 char] -6. Provide brief descriptions of outyear (BY+1, BY+2, BY+3, BY+4 and beyond as necessary) budget requests -for this investment. Briefly describe planned projects and/or useful components proposed. Your justification -should address new functionality, systems integration, technology refreshes, efficiencies to be realized, and any -other planned enhancements to existing assets/systems performance or agency operations. -Fiscal Year -BY+1 -BY+2 -BY+3 -BY+4 and beyond - -Description -[Limit: 500 characters] - -Page 11 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -7. Provide the date of the Charter establishing the required Integrated Program Team (IPT) -for this investment. An IPT must always include, but is not limited to: a qualified fully- -dedicated IT program manager, a contract specialist, an information technology specialist, -a security specialist and a business process owner before OMB will approve this program -investment budget. IT Program Manager, Business Process Owner and Contract Specialist -must be Government Employees. - -(date) - -8. IPT Contact Information - -Name - -Phone Number - -Extension - -Email - -[Limit: -Char] - -250 - -[10 digits 0-9 only] - -[Optional: 6 digits -0-9 only ] - -[Limit one email -only] - -IT Program Manager - -Business Process Owner -Contract Specialist -Information Technology -Specialist -Security Specialist - - -Page 12 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Section C: Summary of Funding (Budget Authority for Capital Assets) - -1. Provide the funding summary for this investment by completing the following table. Include funding - -authority from all sources in millions, and round to three decimal places. Federal personnel costs should be -included only in the rows designated “DME … Govt. FTEs” costs and “O&M Govt. FTE” costs and should -be excluded from the other cost breakouts. Cost levels should be consistent with funding levels in Exhibit 53. -For multi-agency investments, this table should include all funding (both managing and partner agency -contributions). - -For years beyond BY+1, please provide your best estimates for planning purposes, understanding that -estimates for out-year spending will be less certain than estimates for BY+1 or closer. - -For lines in the table that ask for changes in your current submission compared to your most recent previous -submission, please use the President’s Budget as your previous submission. When making comparisons, -please ensure that you compare same-year-to-same-year (e.g., 2011 v. 2011). - -Significant changes from the previous submission should be reflected in an updated investment-level -Alternatives Analysis and is subject to OMB request as discussed in Exhibit 300.5. - -PY-1 & -Prior - -PY -2011 - -CY -2012 - -BY -2013 - -BY+1 -2014 - -BY+2 -2015 - -BY+3 -2016 - -BY+4 & -Beyond - -Planning Costs: - -DME (Excluding Planning) -Costs: -DME (Including Planning) -Govt. FTEs: -Sub-Total DME (Including -Govt. FTE) : -O&M Costs: - -O&M Govt. FTEs: - -Sub-Total O&M Costs -(Including Govt. FTE): -Total Cost (Including Govt. -FTE): - -Total Govt. FTE costs: - -# of FTE rep by costs: - -Total change from prior year -final President’s Budget ($) -Total change from prior year -final President’s Budget (%) -Do not enter information for the dark gray cells. - -2. While some investments are consistent with a defined life cycle model (i.e., an initial period of development -followed by a period of primarily operational spending and an identifiable end point), others represent a -collection of ongoing activities and operations with no known terminal point. In the following table, identify -whether or not this investment uses a defined life cycle model (as defined in OMB Circular A-131) and -provide appropriate investment cost information below. - -a. - -Is this investment consistent with a life cycle model defined in OMB Circular -A-131(i.e., an initial period of development followed by a period of primarily -operational spending and an identifiable end point): - -[Y/N] - -Page 13 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -b - -c - -Describe why the investment is not consistent with life cycle model -management defined in OMB Circular A-131, and explain how you adapted -your alternatives analysis for this investment? -(Where an agency uses a cost model other than the lifecycle cost model, defined by OMB -Circular A-131, responses from 2c to 2h below should reflect the alternative concept.) -Provide information on what cost model this investment is using and how costs -are captured for what years: - -d. What year did this investment start (use year—e.g., PY-1=2010) - -e. -f. - -g. - -h. - -What year will this investment end (use year—e.g., BY+5=2018) -Estimated Total DME cost (including planning) for the investment life cycle or -other cost model (excluding FTE) -Estimated Total O&M cost the investment life cycle or other cost model -(excluding FTE) -Estimated total Govt. FTE Cost for the investment life cycle or other cost -model - -[Limit: 1000 chars] -(Required if a is N) - -[Limit: 1000 chars] -(Required if a is N) -(Required if a is Y) - -(Required if a is Y) -(Required) - -(Required) - -(Required) - -3a. If the funding levels have changed from the FY 2012 President’s Budget request for PY -or CY, briefly explain those changes: - -[Limit: 500 chars] - -Page 14 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - - Section D: Acquisition/Contract Strategy (All Capital Assets) - -1. Complete or update the table to display all prime contracts (or task orders) for awarded or open solicitations -for this investment (sub-award details is not required). Contracts and/or task orders that have “Ended” should not -be included in the table. Contracts in open solicitation should provide estimated data for all fields (for “Total -Contract Value” the estimated base contract costs and all anticipated option years). Data definitions can be found -at www.usaspending.gov/learn#a2. - -For specifics, please see notes 1 and 2 below the table. - -Field - -Data Description - -Optional for -Awarded Contracts - -Contract/ -Task Order - -(1) Awarded, (2) Pre-award Post-solicitation - -Required only if the contracting agency is different than the -agency submitting the exhibit. Use agency 4 digit code as -used in FPDS. -The unique identifier for each contract, agreement or order. -See http://www.usaspending.gov/learn#a2 - -Required only for IDVs. See -http://www.usaspending.gov/learn#a2 - -This is a code for an agency, but it does not necessarily -represent the agency that issued the contract. Instead, it -serves as part of the unique identification for Federal -Procurement Data System IDV records. -Identifier used to link transactions to solicitation -information. See http://www.usaspending.gov/learn#a2 -Y/N - -Total Value of Contract including base and all options. -Complete using dollars to two decimal places. - -See FAR Part 16. Can be fixed price, cost, cost plus, -incentive, IDV, time and materials, etc - -Y/N Indicates whether the contract is a PBSA as defined -by FAR 37.601. A PBSA describes the requirements in -terms of results rather than the methods of performance of -the work. - -Actual or expected Start Date of Contract/Task Order, the -date that the parties agree will be the starting date for the -contract’s requirements. (YYYY-MM-DD) -(YYYY-MM-DD) - -Contract -Status -Contracting -Agency ID - -Procurement -Instrument -Identifier -(PIID) -Indefinite -Delivery -Vehicle (IDV) -Reference ID -IDV Agency -ID - -Solicitation -ID1 -EVM -Required -Ultimate -Contract -Value1 -Type of -Contract/Tas -k Order -(Pricing) 1 -Is the -contract a -Performance -Based Service -Acquisition -(PBSA)? 1 -Effective -date1 - -Actual or -expected End -Date of -Contract/Tas -k Order1 - -Y - -Y - -Y - -Y - -Y - -Y - -Y - -Y - -Page 15 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Extent -Competed1 - -Contract -Description - -(A) Full and open competition (B) Not available for -competition (C) Not competed (D) Full and open -competition after exclusion of sources (E) Follow-on to -competed action (F) Competed under simplified acquisition -procedures (G) Not competed under simplified acquisition -procedures (CDO) Competitive Delivery Order (NDO) Non- -competitive Delivery Order -A brief description of the goods or services bought (for an -award) or that are available (for an -IDV).www.usaspending.gov/learn?tab=FAQ#2 - -Y - -Y - -1Assuming the PIID or IDV PIID match with USAspending.gov, these data elements will be automatically populated for awarded IT acquisitions -2Assumingthe PIID, IDV PIID, or Solicitation number match with USAspending.gov or FedBizOpps (fbo.gov) this data will be auto populated for awarded -and pre-award, post-solicitation IT acquisitions. - -2. If earned value is not required or will not be a contract requirement -for any of the contracts or task orders above, explain why: - -[char 2500] - -Page 16 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Exhibit 300B: Performance Measurement Report - -The Exhibit 300B is used to provide OMB current fiscal year (FY2012) investment performance data for major -investments. Include in this exhibit, at a minimum, all activities and operations that started in a previous fiscal -year (PY and earlier) and have not completed by the beginning of the current year as well as activities and -operations that are scheduled to start in the current fiscal year. Provide an up-to-date copy of this exhibit to OMB -at least once a month. - -UNDERSTANDING THE EXHIBIT 300B - -On the Exhibit 300B, investments are described as: - -• - -Investment - -o Projects - -• Activities - -o Operations - -Report information about these areas in the following Exhibit 300B sections: - -A.1: General Information. Enter basic information about the major investment. -B.1: Projects. Identify all of the investment’s projects with activities occurring in the current fiscal year. -B.2: Activities. Outline the activities that are performed to achieve the outcome of each project. -B.3: Project Risk. Identify all significant risks to each project’s success. -C.1: Operational Performance Information. Identify performance targets for evaluating operations. -C.2: Operational Risk. Identify all significant risks to investment achieving operational performance targets. - -SECTION A: GENERAL INFORMATION - -1. Investment Name: Agency-provided name of investment, consistent with Exhibit 53. -2. Investment UII: Agency-provided unique investment identifier, consistent with Exhibit 53. - -SECTION B: PROJECT EXECUTION DATA - -Section B addresses planning, DME and significant maintenance projects for the investment. - -In Table B.1, report, at a minimum, all projects with any activities that started in a previous fiscal year (PY and -earlier) and have not completed by the beginning of the current year as well as activities that are scheduled to start -in the current fiscal year, including planning, DME, and maintenance projects. This information should be -updated at least once every month. Include the following data in Table B.1: - -Project -ID - -Project -Name - -Project -Description - -Project -Start -Date - -Project -Completi -on Date - -Table B.1 -Project -Lifecycle -Cost - -PM -Name - -PM Level -of -Experience - -PM -Phone - -[10 digits -0-9 only] - -PM -Extension -[Optional: -6digits 0-9 -only ] - -PM -Email -[Limit one -email -only] - -1. Project ID: An agency-specified number that uniquely identifies the project within this investment. -2. Project Name: Name used by agency to refer specifically to this project. -3. Project Description: Description of project functionality or purpose. -4. Project Start Date: Date of actual start of in-progress projects or planned start of projects which have not - -yet begun (may be before current fiscal year or activities listed in activities table B.2). - -Page 17 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -5. Project Completion Date: Planned date of completion of in-progress projects or actual completion date of -projects which have completed (may be after budget year or of completion date of activities listed in -activities table B.2). - -6. Project Lifecycle Cost: Enter the total cost of all activities related to this project as described in OMB - -Circular No. A-131. (in $ millions) - -7. PM Name: Name of project manager responsible for the success of this project. -8. PM Level of Experience: The years of applicable experience or the status of certification. Available - -selections include: - -1) FAC-P/PM(DAWIA-3)– Senior -2) FAC-P/PM(DAWIA-2)– Mid-Level -3) FAC-P/PM(DAWIA-1)– Entry Level -4) Other certification with 4 or more years PM experience (within the last five years) -5) Other certification with between 2 and 4 years PM experience (within the last five years) -6) Other certification with less than two years PM experience (within the last five years) -7) No certification, but with 4 or more years PM experience (within the last five years) -8) No certification, but with between 2 and 4 years PM experience (within the last five years) -9) No certification, but with less than two years PM experience (within the last five years) - -9. PM Phone: Phone number of project manager responsible for the success of this project. -10. PM Email: Email address of project manager responsible for the success of this project. - -In Table B.2, describe, at a minimum, all activities occurring during the current fiscal year. This table should be -updated once a month at a minimum. In line with modular development principles, activities should be structured -to provide usable functionality in measureable segments that complete at least once every six months or more -often, as described in the 25-Point Implementation Plan to Reform Federal IT. - -A -Project ID - -B -Activity -Name - -C -Activity -Description - -D -Structure -ID - -E -Key -Deliverable/ -Usable -Functionality -[Key -Deliverable, -Usable -Functionality, -N/A] - -Table B.2 - -F -Start Date -Planned - -G -Start Date -Projected - -H -Start Date -Actual - -I -Completion -Date -Planned - -J -Completion -Date -Projected - -K -Completion -Date Actual - -L -Total Costs -Planned - -M -Total Costs -Projected - -N -Total Costs -Actual - -A. Project ID: An agency-specified number that uniquely identifies the project within this investment. -B. Activity Name: A short description consistent with the critical steps within the agency project - -management methodology. - -C. Activity Description: Describe what work is accomplished by this activity. -D. Structure ID: Agency-specified identifier which indicates work breakdown structure agency uses to - -associate this activity with other activities or a project. Please provide this in the format of “x.x.x.x.x” -where the first string is the Project ID and each following string (separated by periods) matches the -Structure ID of a parent activity. See below for more guidance about parent and child activities expressed -through this structure. - -E. Key Deliverable / Usable Functionality: Indicate whether the completion of this activity provides a key -deliverable or usable functionality. This should only be provided for activities which do not have a child -activity. Use this field to demonstrate this investment’s alignment with the modular development -principles of the 25-Point Implementation Plan to Reform Federal IT. - -F. Start Date Planned: The planned start date for this activity. This is the baseline value. -G. Start Date Projected: When activity has not yet started, enter current planned start date of the activity. -H. Start Date Actual: When activity starts, enter actual start date here. -I. Completion Date Planned: The planned completion date for this activity. This is the baseline value. -J. Completion Date Projected: When activity has not yet completed, enter current planned completion date - -of the activity. - -Page 18 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -K. Completion Date Actual: When activity ends, enter actual completion date here. -L. Total Costs Planned: The planned total cost for this activity. This is the baseline value. -M. Total Costs Projected: When activity has not yet completed, enter current planned total cost of the - -activity. - -N. Total Costs Actual: When activity ends, enter actual total costs for the activity here. - -Note: For programs that are employing Earned Value Management, agencies should reflect “budget at -completion” in the “total costs planned” field; and “estimated at completion” in the “total costs projected field for -projected costs. - -Reporting Parent and Child Activities (WBS Structure) - - “Child” activities may be grouped into “Parent” activities to reflect the work breakdown structure (WBS) the -agency uses to manage the investment. If a work breakdown structure is not used by the agency, please report the -relationship between parent activities and child activities in “Structure ID” using this method. - -When reporting an activity, enter the “Structure ID” as a period-delimited string consisting of the “Project ID” -and each nested parent activity between the project level and the child activity. The “Structure ID” to enter will -vary depending on the activity’s WBS level. - -Example: For child activity 3 which is part of parent activity 10, which in turn is part of parent activity 2, -which in turn is part of Project A, please enter: A.2.10.3 - -• Project A - -o Parent Activity 2 - - Parent Activity 10 - -• Child Activity 3 - -There is no limit to the number of nested “child” and “parent” relationships allowed, and this depth may vary -from activity to activity and from project to project. - -If any of a parent activity's child activities occurs in the current fiscal year, then all child activities of the parent -activity must be reported regardless of their timing. This is to ensure that a complete view of the parent activity is -available. - -All activities with no child activities must have, at a minimum, Project ID, Activity Name, Activity Description, -Structure ID, Start Date Planned, Start Date Projected, Completion Date Planned, Completion Date Projected, -Total Costs Planned, and Total Costs Projected. Completed activities must also have Start Date Actual, -Completion Date Actual, and Total Costs Actual. - -Any parent activities with a child activity must be completely described by the aggregate attributes of its child -activities. In the IT Dashboard, the cost and schedule information for parent activities will be based on the cost -and schedule information of their most detailed reported child activities. Agency-submitted cost and schedule -information is not required for parent activities. - -Project -ID - -A -A -A -A - -Activity Name -Design -Business Requirements -Technical Requirements -Architecture - -Structure -ID - -Start Date -Planned - -Completion -Date Planned - -Planned Total -Costs - -… - -A.2 -A.2.1 -A.2.2 -A.2.3 - -2/1/2012 -2/1/2012 -2/11/2012 -2/21/2012 - -2/29/2012 -2/10/2012 -2/20/2012 -2/29/2012 - -$2.5 -$1.0 -$1.0 -$0.5 - -Data reported in cells marked with a fill color (Structure ID: A.2) is unnecessary. These values will be calculated by aggregating the cost and schedule information -reported in this parent activity’s child activities. - -Page 19 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -Project Risk - -Risk assessments should include risk information from all stakeholders and should be performed at the initial -concept stage and then monitored and controlled throughout the life-cycle of the investment. - -In Table B.3, list all significant project related risks for the investment that are currently open and provide risk -assessment information. (It is not necessary to address all 19 OMB Risk Categories). - -1. Project ID: An agency-specified number that uniquely identifies a project within this investment. For - -each identified risk, lists the associated Project ID. - -2. Risk Name: A short description that identifies a risk, the cause of the risk and the effect that the risk may - -have on the project. - -3. Risk Category: Please select the relevant OMB Risk Category for each risk. Risk categories include: 1) -schedule; 2) initial costs; 3) life-cycle costs; 4) technical obsolescence; 5) feasibility; 6) reliability of -systems; 7) dependencies and interoperability between this investment and others; 8) surety (asset -protection) considerations; 9) risk of creating a monopoly for future procurements; 10) capability of -agency to manage the investment; and 11) overall risk of investment failure; 12) organizational and -change management; 13) business; 14) data/info; 15) technology; 16) strategic; 17) security; 18) privacy; -and 19) project resources. - -4. Risk Probability: The likelihood of a negative impact for this risk (Low, Medium, or High likelihood). -5. Risk Impact: Assess level of potential negative impacts for the risk (Low, Medium, or High impact). -6. Mitigation Plan: A short description of the plan or steps to mitigate the identified risk. - -Project ID - -Risk Name - -Risk Category Risk Probability - -Risk Impact - -[char 500] - -[19 Risk -Categories] - -[Low, Medium, -High] - -[Low, Medium, High] - -Mitigation Plan -[char 500] - -Table B.3 - -SECTION C: OPERATIONAL DATA - -Section C addresses operational activities which are not reported as a part of a project in Section B. - -Operational Performance -There are two essential types of operations metrics to be reported (see FEA Reference Model Mapping Quick -Guide): - -1. Results Specific: Provide a minimum of two metrics which measure the effectiveness of the investment in -delivering the desired service or support level; if applicable, at least one metric should reflect customer -results (e.g.; “Service Quality”). - -2. Activities and Technology Specific: Provide a minimum of three –metrics which measure the investment -against its defined process standards or technical service level agreements (SLAs) (e.g.; “Reliability and -Availability”). At least one of these metrics must have a monthly “Reporting Frequency.” - -Provide results specific metrics which are appropriate to the mission of the investment and its business owner or -Customer. Generally these metrics should be provided by the investment’s business owner and will reflect -performance in the broader business activities and not IT-specific functions. The best results specific metrics will -support the business case justification and could be the foundation of a quantitative approach to defining benefits -in a cost-benefit analysis. Unlike in private industry where identified benefits accrue to the organization, -government benefits may accrue to the public. Therefore, results-specific metrics may demonstrate the value - -Page 20 of 21 Exhibit 300 - -(2011) - - GUIDANCE ON EXHIBIT 300—PLANNING, BUDGETING, ACQUISITION, AND MANAGEMENT OF IT CAPITAL ASSETS - -realized external to the Federal Government. The table must include a minimum of two results-specific metrics, -one of which should reflect customer results. - -Each metric description should help the user understand what is being measured. In this field, describe the units -used, any calculation algorithm used, and the definition or limits of the population or “universe” measured. - -The unit of measure should be characterized (e.g. number, percentage, dollar value etc) for each metric. Each -metric listed in the table must also indicate how often actual measurements will be reported (monthly, quarterly or -semi-annually), as well as baseline, targets and actual results. The “Actual for PY” should be final actual -measurement from the previous year or the average actual results from the previous year. Describe whether a -successful actual measurement would be “over the target” or be “under the target” in “Measurement Condition.” -“Comment” field is required for performance metrics where target not expected to be met. All data will be -displayed on the IT Dashboard. - -Metric -Description - -Unit of -Measure - -[char 500] - -[char 50] - -FEA -Performance -Measurement -Category -Mapping - -[PRM -Measurement -Category] - -Baseline Target -for PY - -Table C.1 -Actual for -PY - -Target -for CY - -Measurement -Condition - -Reporting -Frequency - -Comment - -Most Recent -Actual -Results - -[numeric] - -[numeric] - -[numeric] - -[numeric] - -[Over target, -Under target] - -[Monthly, -Quarterly, -Semi- -Annual] - -[numeric] - -[Optional -char500] - -Operational Risk - -Risk assessments should include risk information from all stakeholders and should be performed at the initial -concept stage and then monitored and controlled throughout the life-cycle of the investment. - -In Table C.2, list all significant operational related risks for the investment that are currently open and provide -risk assessment information. (It is not necessary to address all 19 OMB Risk Categories). - -1. Risk Name: A short description that identifies a risk, the cause of the risk and the effect that the risk may - -have on the operational activity. - -2. Risk Category: Please select the relevant OMB Risk Category for each risk. Risk categories include: 1) -schedule; 2) initial costs; 3) life-cycle costs; 4) technical obsolescence; 5) feasibility; 6) reliability of -systems; 7) dependencies and interoperability between this investment and others; 8) surety (asset -protection) considerations; 9) risk of creating a monopoly for future procurements; 10) capability of -agency to manage the investment; and 11) overall risk of investment failure; 12) organizational and -change management; 13) business; 14) data/info; 15) technology; 16) strategic; 17) security; 18) privacy; -and 19) project resources. - -3. Risk Probability: The likelihood of a negative impact for this risk (Low, Medium, or High likelihood). -4. Risk Impact: Assess level of potential negative impacts for the risk (Low, Medium, or High impact). -5. Mitigation Plan: A short description provides how to mitigate the risk. - -Risk Name -[char 500] - -Risk Category -[19 Risk -Categories] - -Risk Probability Risk Impact -[Low, Medium, -High] - -[Low, Medium, -High] - -Mitigation Plan -[char 500] - -Table C.2 - -Page 21 of 21 Exhibit 300 - -(2011) + Publishing revenues from streaming in the UK have grown +significantly, but many songwriters argue they are not paid enough +to make a sustainable income + +30. Each song has two sets of music rights: (1) rights in the underlying song or +‘publishing rights’ which includes the music and lyrics; and (2) rights in the +particular recording of that song, the ‘recording rights’. Without permission to +use both rights, a streaming service cannot legally stream the song. + +31. + +Throughout the study we have heard consistently from groups that represent +songwriters that the publishing right of a song is systematically undervalued +when compared to the recording right. In turn, they argue that this means that +songwriters receive less in royalties for their work and do not make a +sustainable income from streaming. Whilst some songwriters are also +themselves artists and may have access to revenues from recording rights +and other sources of income, that is not always the case. + +32. Songwriters and their representatives have suggested that the undervaluation +of publishing rights is because the majors have market power and interests in +both publishing and recording rights and that it is financially advantageous for +them to suppress publishing revenues in favour of the recording side of their +business, possibly through tacit collusion. We have also heard concerns that +the majors may have the ability and incentive to influence industry outcomes +via their membership of collecting societies (CMOs) who have a role in +negotiating royalties for the use of their members’ publishing rights. + +33. We have examined evidence on the allocation of music streaming revenues, + +and it indicates that: + +(a) The share of revenues going to publishers (publishing share) increased + +from 8% in 2008 to approximately 12% in 2012. Since then it has +increased to 15% in 2021. This is a significant increase across the period. +In contrast, the share of revenues going to record companies in 2021 is at +a broadly similar level to what it was in 2008. + +(b) For the majors, the growth in publishing revenues + +has far exceeded the growth in recording revenues. + +(c) Majors’ growth in publishing revenues is 244% since +2017, which is significantly above the [110-120]% +average across the publishing sector over the same +period. + +9 + + Split in music streaming revenues in 2008 and 2021 + +34. + +Source: Section 4.2.2 of research commissioned by the IPO Music Creators Earnings, the Digital Era, +and On-Demand Streaming Revenues and CMA analysis of data from Apple, Amazon and Spotify.5 + +This evidence is inconsistent with the argument that the majors have tacitly +colluded to suppress the publishing share or that there is otherwise +particularly weak competition to sign songwriters that is leading to a split in +the allocation of music streaming revenues that favours recording rights over +publishing rights. The majors having both a recording and a publishing +business is also not necessarily problematic. For instance, if the majors did +not have a publishing business they might have a stronger incentive to block +increases in the ‘publishing share’ by refusing to accommodate such an +increase through reducing the recording share since any losses to their +recording revenues which occurred would not be mitigated by gains to their +publishing revenues. + +35. We also have not seen clear evidence that CMOs are failing to push for better +terms owing to the majors’ influence or that CMO governance procedures and +processes are failing to mitigate any potential bias in their decision making. If +such regulatory concerns do exist, the proper body to examine them is the +IPO, which has responsibility for monitoring the conduct of CMOs under the +Collective Management of Copyright (EU Directive) Regulations 2016.6 + +5 Note: Whilst the CMA analysis for 2021 is based on data from the largest music streaming services, the IPO +research draws on a range of largely qualitative evidence and therefore (as the IPO research itself +acknowledges) the IPO research estimate of the split of revenues for 2008 is only indicative. Therefore, +comparison between the split between 2008 and 2021 should be treated with caution and taken to be indicative +of the overall trend in the split over time rather than estimates of the exact quantum of change in the split. +6 The conduct of UK CMOs (including the PRS) is governed by the CRM Regulations. The CRM Regulations +designate a National Competent Authority (NCA) which is responsible for monitoring and enforcing compliance + +10 + + 36. Songwriters are concerned that the initial split between publishing and + +recording revenues adopted when streaming began was unfair and +unjustified, for example because it did not reflect the alleged lower costs of +record companies under streaming compared to physical distribution. The +increase in the publishing share is consistent with this concern over the initial +split and there being a subsequent period of market-correction, which may not +yet have fully played out. However, the fact that publishing revenues are +increasing at a greater rate compared to recording revenues, in particular for +the majors, suggests that publishing revenues are not being actively +suppressed because of a distortion or restriction of competition. + +37. We have found that there are inherent difficulties in securing increases to the + +publishing share and, consequently, increasing the amount songwriters are +paid. This difficulty may arise from the fact that music streaming services, +labels and publishers must all reach agreement to change how streaming +revenues are divided. However, they may all have different incentives and so +they may not agree – we call this a ‘licensing negotiation friction’. We think +that agreement between the parties may be particularly challenging when an +increase in the publishing share would necessitate a fall in the share that goes +to record companies due to the strong bargaining position of music +rightsholders, which arises from the need for music streaming services to get +agreement from all key rightsholders to offer a wide range of music. + +38. + +The long-term increase in the publishing share since 2008 has been +accommodated by a fall in the share taken by music streaming services. The +risk is that we may reach a point, or could do soon, where further substantial +increases in the publishing share can only be accommodated by a fall in the +recording share, which labels would be in a strong bargaining position to +resist. + +39. Whilst we think that competition for songwriters has driven up the existing + +publishing share, concerns exist that the current split could still be sub- +optimal, particularly for songwriters. If that is the case, we think that it may +take time for the split to adjust further, if at all, owing to the inherent licensing +negotiation frictions and bargaining power of music rightsholders we have +described. There is also a limit on the extent to which competition to sign +songwriters can drive further increases in the publishing share, particularly if +an increase needs to be accommodated by a fall in the recording share. +Competition policy is not therefore the right tool to reach an optimal split. We +think it is a matter for Government and policymakers to determine whether the + +with the Regulations’ provisions. The NCA functions in the UK are undertaken through the IPO, which has +published guidance on these regulations (see IPO (2021), Guidance on the Collective Management of Copyright +(EU Directive) Regulations 2016). + +11 + + split is appropriate and fair, and to explore what is needed to incentivise song +writing as part of wider policy interventions on this split and other measures, +for example those relating to the copyright framework and how music +streaming licensing rates are set. + +The legal arrangements between major labels and music streaming +services are complex but they do not appear to be significantly +hampering competition and innovation + +40. Major labels rely on music streaming services to distribute their music, + +and streaming services cannot meet consumers’ needs without obtaining +licences to the large catalogue of each major record label and publisher. +Accordingly, music streaming services must negotiate deals with labels +and publishers to access music content. These deals can be exceedingly +complex and will cover financials and other terms. Our market study +uncovered a number of clauses that could plausibly raise competition +concerns. For example a number of agreements contained non- +discrimination clauses which act to prevent the music streaming service +from favouring music content based on price, for example, by giving +more prominence to cheaper music. + +41. However, given the current ‘full catalogue’ business model of music streaming + +– consumers expect to access every major’s repertoire on each streaming +service – there is no credible alternative to each major’s catalogue. Taking +this into account, our view is that the nature of competition between record +companies to supply music to music streaming services is weak but would +remain weak even absent the combined effect of the contractual clauses we +identified as being potentially problematic. Whilst a slight strengthening of +competition might result from the removal of these clauses (individually or in +combination), it is not clear any improvement would be more than marginal. + +42. + +Innovation is intrinsic to a healthy, competitive market and we therefore take +seriously any suggestion that innovation has been hindered. We have +therefore also assessed the impact of contractual clauses on how music +services compete with each other in terms of innovation. To introduce +innovations or changes to services, the music streaming services typically +need to agree with the majors to amend existing contracts, which we heard +can be a long process which may slow the pace of innovation. + +43. We found examples of substantial innovation by music streaming services, +both in terms of the services, such as the introduction of high-quality audio, +and in the price plans available. But we were also given a few examples of +innovations that were slow to market because of the complex negotiations + +12 + + needed to secure licensing agreements. There is a risk that contractual +restrictions may contribute to the slower development of such innovations +than might otherwise be expected or, potentially, preclude innovation +altogether. + +44. While potential competition concerns have been raised with us about the + +effect of agreements between the majors and music streaming services, we +are not persuaded that changing the contractual clauses would significantly +increase innovation. Rather, the problem appears to relate to the need for +music streaming services to agree with multiple rightsholders on what terms +(financial or otherwise) they can use their content, including in new and +innovative ways. It is the sheer volume and complexity of these negotiations +that appear to be the main barrier to even greater innovation. However, these +negotiations appear to be an inherent part of the licensing process – with the +financial terms negotiated depending on the features agreed. + +Competition between music streaming services is currently +delivering good outcomes for consumers, but concerns may arise +in future + +45. + +The music streaming services market is concentrated with a few larger +streaming services such as Spotify, Apple, Amazon, and YouTube (which is +part of Google), alongside a range of other smaller providers. Spotify has the +largest number of monthly active users by some distance, as shown in the +graph below. Music streaming services are popular with consumers and have +grown rapidly – between 2019 and 2021 the number of monthly active users +of music streaming services increased from 32 million to 39 million. Despite +the strong presence of large, well-known firms in the market, and the number +of active users increasing, music streaming services are not making +sustained, excess profits: indeed, our analysis has shown that many services +have low or negative operating margins. + +13 + + Share of UK Monthly Active Users by music streaming service in +December 2021, excluding YouTube’s UUC platform + +Source: CMA analysis of data from music streaming services.7 + +46. We have heard consistently that consumers demand access to a full + +catalogue of music. The result is that the main music streaming services +effectively offer the same music content to consumers. Competition between +the services, therefore, anchors around offering the best experience to +consumers through good design, personalised playlists, and high-quality +audio content, as well as through pricing plans. These services also now +compete with one another by offering non-music content, such as podcasts. + +47. + +For consumers, the monthly price of music +streaming services is either free (ad-funded) or +falling in real terms because the price of +individual subscriptions has remained stable and +not kept pace with inflation. Most services offer a +range of price plans, including family and student +plans, as well as free ad-funded tiers. Streaming +services are also frequently bundled with other + +7 This pie chart is for illustrative purposes only. Monthly Active User shares only account for Spotify, YouTube +Music, Apple, Amazon, Deezer, Soundcloud and Tidal which have a combined streaming share of over 99% +according to CMA analysis of data provided by Official Charts. YouTube Music users include YouTube Music +premium Monthly Active Viewers and YouTube Music ad-funded Daily Active Viewers, meaning this figure will +provide an underestimation of YouTube Music’s actual users. These figures are provided in a 5% range where +the figure is below 10%, and a 10% range where the figure is between 10% and 100%. The midpoints of the +ranges have been used to provide an illustration of relative size in the market. Where the sum of these midpoints +does not equal 100%, we have scaled the pie chart so that the area segments represent the share of the sum of +the midpoints. + +14 + + services, such as mobile phone subscriptions, and accessed via a range of +devices, including smart speakers. + +48. Recorded music is now also costing consumers less overall compared to + +when CDs and other physical formats were more popular, as indicated by UK +recorded music revenues falling by around 40% from £1.9 billion in 2001 to +£1.1 billion in 2021 in real terms. + +49. + +In a market that is expanding, music streaming services mainly compete for +new consumers, rather than encouraging existing customers to switch to their +streaming services. However, music streaming services with ad-funded plans +do actively seek to get customers to upgrade to a paid-for service. + +50. Our analysis shows that consumers cancel their service at above 4% a month +for the major streaming services. Cancellation might occur because free trials +are coming to an end, the user is switching to another service, or because the +user is no longer using any streaming service. The current data on why +consumers may cancel their services is limited, but it suggests that the most +cited reasons are consumers not being able to afford the service or not using +it enough.8 + +51. Switching between music streaming services might be challenging if + +consumers are concerned that they will lose access to their favourite playlists. +There are some nascent music data portability services that support +switching, but demand for them is currently low. Low switching rates are not +necessarily a dynamic that causes immediate concern - for example, in a +growing market. However, as the number of new premium users to compete +for declines in future, there is a risk that prices for music streaming services +will rise significantly for consumers or there may be a deterioration in the +quality of services, if there isn’t a significant threat of switching. Therefore, as +the market reaches maturity we would be concerned if we did not see more +vigorous competition between streaming services (eg through enhanced +efforts to make it seamless for consumers to switch and port their playlists or +musical preferences). Higher rates of switching would imply consumers +exercising choice (and hence competition), although the lack of switching by +itself is not conclusive evidence of a lack of competition. + +52. With more music available, tools that help consumers + +discover new music are more important. Streaming services +compete in bringing music to consumers’ attention and +promoting new artists and songs. We have found that while + +8 This should be treated with caution, however, as only a small proportion of users who cancel respond to the +cancellation survey used to collect this data. + +15 + + the majority of tracks listed on the top new music discovery playlists are +licensed by the majors, the proportion is lower than the majors’ combined +share of total streams. This suggests that artists that are not signed to a major +do have reasonable opportunities to reach new listeners via discovery +playlists and more generally the streaming services have told us that they +design their music recommendation systems with a focus on listener +engagement and user satisfaction. + +53. Overall, whilst music streaming services are currently delivering good + +outcomes for consumers, it is imperative for a sustainable and vibrant market +that services can effectively compete with one another, and we would have +concerns in future if we saw a reduction in competition – for example, if +majors sought commitments from streaming platforms to exclude competitors +(large or small) from discovery or search elements of their service. + +There is a ‘value gap’ between what YouTube and other music +streaming providers pay to rightsholders but it currently amounts +to less than 0.5% of UK recorded music revenues + +54. UUC platforms allow consumers to access music content uploaded by users, + +artists and labels for free (but often with ads), in some cases coupled with +other content such as entertainment videos. These services differ from music +streaming services because any user can upload content, which may include +copyrighted material which they may or may not have permission to share. +Sometimes this content can appear on these platforms before a licence has +been agreed with rightsholders. + +55. UUC platforms have some protection in law through a ‘safe harbour’ provision +which limits the liability they have for hosting illegal content uploaded by users +in some circumstances. However, once they become aware that content is +available without permission from rightsholders, they must remove it or, as is + +16 + + more often the case, allow the rightsholder to grant permission and monetise +the content, for example by sharing ad revenues. + +56. A range of music companies, artists and songwriters have been concerned + +that the asymmetry in the legal regime may result in a loss of revenues (often +called the ‘value gap’) for the music industry. In particular, there is concern +that the safe harbour provisions give UUC platforms greater bargaining power +over rightsholders, leading to rightsholders having to agree worse terms than +they would otherwise. + +57. + +To understand whether a value gap exists and, if so, how significant the gap +is, we have compared the amount that YouTube, as the largest UUC platform, +pays out to rightsholders in comparison to Spotify’s ad-funded service. In +2017, YouTube’s ‘value gap’ in the UK – that is how much more it would have +paid to rightsholders as a proportion of revenue from music content if it had +paid them at the same rate as Spotify’s ad-funded service – was over 20 +percentage points. However, this ‘gap’ has closed over time and in 2021 had +fallen to significantly less than 5 percentage points, or less than £5 million. To +put this figure into context, it is less than 0.5% of the £1,115 million total UK +recorded music revenues in 2021. The evidence therefore suggests that the +extent of YouTube’s ‘value gap’ has decreased in recent years. + +58. Whilst YouTube is the largest UUC service and therefore very important in the +market, there are other UUC platforms that take a different approach to +licensing and may have less effective content management systems. + +59. Some newer UUC platforms incorporate music into their offering, but do not +offer a full streaming service. Examples of this include TikTok, a short-form +video-sharing service, and Twitch, a longer-form live streaming service. These +platforms, alongside YouTube, provide innovative services for consumers and +potential opportunities for creators and music companies to earn further +revenues in addition to those generated by music streaming. + +60. + +The way these services use music differs from traditional music streaming +services: they may use snippets of recordings or only offer specific genres of +music, rather than a full catalogue on an on-demand basis. Therefore, it may +not always be the case that the rates payable for these services will be +comparable to music streaming services, which is an important factor when +assessing if there is a ‘value gap’ between the two offerings. That said, we +have heard examples of refusal to license or difficulties in agreeing licences +that may indicate that safe harbour is among the factors in some commercial +negotiations. We also consider that the quality of content management +systems is important in ensuring that rightsholders are remunerated fairly for +music content uploaded onto UUC platforms, and we have heard some newer + +17 + + emerging platforms may lag behind the more established platforms in this +regard. + +61. Meanwhile, since the exit of the UK from the EU, European legislation + +relevant to UUC has been amended such that it now requires UUC services +using copyrighted content to make their ‘best efforts’ to seek permission to +use that content. We note that the DCMS and IPO are monitoring the practical +impact of these legislative changes on UUC services in the UK. We +encourage them to take account of our broader findings in relation to UUC +services when considering if legislative changes are required in the UK. + +The market is evolving and innovating, but it is vital that it +continues to deliver good outcomes for consumers + +62. + +The music streaming market is changing rapidly, and further technological +advances in the years to come may spark further changes to the way we +listen to music. Our analysis shows that the market is on balance delivering +good outcomes for consumers. However, we would have concerns and may +intervene in the future if aspects of the market change in ways that harm +consumers’ interests. For example, factors that may give rise to concerns +could include: + +(a) if future mergers or acquisitions affect the bargaining power of either + +music companies or music streaming services, which may in turn lead to +worse outcomes for consumers with the CMA likely to pay particularly +close attention to any such merger activity and to investigate whether it +could lead to a substantial lessening of competition. + +(b) shifts in the way consumers access streaming services that influence their + +listening behaviour, for example if there is continued growth in the use of +smart speakers, and whether this could exacerbate barriers to expansion +of streaming services that do not have their own smart speaker +ecosystem. + +(c) greater use of playlists, autoplay and recommendations for music + +discovery and consumption could be a cause of uncertainty and concern +for consumers and artists if their operation, including any underlying +algorithms, is not fair and transparent. + +(d) how difficult it is to switch between music streaming services and whether +this limits the strength of competition between those services when the +market is no longer growing. + +(e) if the level of innovation on the part of streaming services were to + +decrease; or if innovations that would benefit consumers were to be +prohibited by music companies; or if consumers were to be +disadvantaged in other ways, including through significantly higher prices. + +18 diff --git a/q144/random_k4/random_4.md b/q144/random_k4/random_4.md index 5526b8c35c6c0d7d05b7d26b15fce08f69e7eb06..7a50168af73af1ce80ad936866aebdf5198c18c7 100644 --- a/q144/random_k4/random_4.md +++ b/q144/random_k4/random_4.md @@ -1,1865 +1,1797 @@ -Morgan Stanley Investment Management +Q4 FY18 +Financial Update +Salesforce -Asset Management Overview - -July 2006 - -Page +NYSE: CRM +@Salesforce_ir 1 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - The information provided herein may include certain non-GAAP financial measures. -The reconciliation of such measures to the comparable GAAP figures are included in -the Company’s Annual Reports on Form 10-K, Quarterly Reports on Form 10-Q and -Current Reports on 8-K, including any amendments thereto, which are available at -www.morganstanley.com. - -This presentation may contain forward-looking statements. You are cautioned not to -place undue reliance on forward-looking statements, which speak only as of the date -on which they are made, which reflect management’s current estimates, projections, -expectations or beliefs and which are subject to risks and uncertainties that may cause -actual results to differ materially. For a discussion of risks and uncertainties that may -affect the future results of the Company, please see “Forward-Looking Statements” -immediately preceding Part I, Item 1, “Competition” and “Regulation” in Part I, Item 1, -“Risk Factors” in Part I, Item 1A and “Certain Factors Affecting Results of Operations” -in Part II, Item 7 of the Company’s Annual Report on Form 10-K for the fiscal year -ended November 30, 2005, “Management’s Discussion and Analysis of Financial -Condition and Results of Operations” in the Company’s Quarterly Reports on Form 10- -Q for the quarterly periods ended February 28, 2006 and May 31, 2006 and other items -throughout the Form 10-K, Form 10-Qs and the Company’s 2006 Current Reports on -Form 8-K. - -Page + Safe Harbor + +"Safe harbor" statement under the Private Securities Litigation Reform Act of 1995: This presentation contains forward-looking statements about our financial results, which may +include expected GAAP and non-GAAP financial and other operating and non-operating results, including revenue, net income, diluted earnings per share, operating cash flow growth, +operating margin improvement, deferred revenue growth, expected revenue run rate, expected tax rates, stock-based compensation expenses, amortization of purchased intangibles, +amortization of debt discount and shares outstanding. The achievement or success of the matters covered by such forward-looking statements involves risks, uncertainties and +assumptions. If any such risks or uncertainties materialize or if any of the assumptions prove incorrect, the company’s results could differ materially from the results expressed or +implied by the forward-looking statements we make. + +The risks and uncertainties referred to above include -- but are not limited to -- risks associated with the effect of general economic and market conditions; the impact of foreign +currency exchange rate and interest rate fluctuations on our results; our business strategy and our plan to build our business, including our strategy to be the leading provider of +enterprise cloud computing applications and platforms; the pace of change and innovation in enterprise cloud computing services; the competitive nature of the market in which we +participate; our international expansion strategy; our service performance and security, including the resources and costs required to prevent, detect and remediate potential security +breaches; the expenses associated with new data centers and third-party infrastructure providers; additional data center capacity; real estate and office facilities space; our operating +results and cash flows; new services and product features; our strategy of acquiring or making investments in complementary businesses, joint ventures, services, technologies and +intellectual property rights; the performance and fair value of our investments in complementary businesses through our strategic investment portfolio; our ability to realize the +benefits from strategic partnerships and investments; our ability to successfully integrate acquired businesses and technologies; our ability to continue to grow and maintain deferred +revenue and unbilled deferred revenue; our ability to protect our intellectual property rights; our ability to develop our brands; our reliance on third-party hardware, software and +platform providers; our dependency on the development and maintenance of the infrastructure of the Internet; the effect of evolving domestic and foreign government regulations, +including those related to the provision of services on the Internet, those related to accessing the Internet, and those addressing data privacy and import and export controls; the +valuation of our deferred tax assets; the potential availability of additional tax assets in the future; the impact of new accounting pronouncements and tax laws, including the U.S. Tax +Cuts and Jobs Act, and interpretations thereof; uncertainties affecting our ability to estimate our non-GAAP tax rate; the impact of expensing stock options and other equity awards; +the sufficiency of our capital resources; factors related to our outstanding convertible notes, revolving credit facility, term loan and loan associated with 50 Fremont; compliance with +our debt covenants and capital lease obligations; current and potential litigation involving us; and the impact of climate change. + +Further information on these and other factors that could affect the company’s financial results is included in the reports on Forms 10-K, 10-Q and 8-K and in other filings we make +with the Securities and Exchange Commission from time to time. These documents are available on the SEC Filings section of the Investor Information section of the company’s +website at www.salesforce.com/investor. + +Salesforce.com, inc. assumes no obligation and does not intend to update these forward-looking statements, except as required by law. 2 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Table of contents - -Section 1 - -Strengths of MSIM Today - -Section 2 - -MSIM Growth Strategy - -Page - -3 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - Section 1 - -Strengths of MSIM Today + Company Overview +The global leader in CRM -Page +Founded in 1999, public listing (NYSE: CRM) 2004. -4 +#1 CRM provider for the fifth year in a row -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +. - MSIM Is a Key Business of Morgan Stanley +Fastest growing top five enterprise software company with $10.48B in revenue FY18 (25% Y/Y). -Morgan Stanley +1 -STRENGTHS OF MSIM TODAY +Doubled operating cash flow ($1.2B-$2.7B) over the past three years (FY15-FY18) while nearly +doubling revenue ($5.4B-$10.5B). -Net revenues -Fiscal year 2005=$26,778 million +Headquartered in San Francisco, with >29,000 Employees focused on CRM and customer success. -19% +#1 The Future 50 -Asset -Management -11% +Best Places to Work +for LGBTQ Equality -12% +The World’s Most +Innovative Companies -Profit before taxes -Fiscal year 2005=$7,361 million +#1 World’s Most +Innovative Companies -8% +#1 Top 50 +Companies that Care -Asset -Management -14% +1Based on IDC 2017H1 Market Share Revenue Worldwide. Source: IDC, Worldwide Semiannual Software +Tracker, Oct 2017 -65% - -13% - -58% +3 -Profit before taxes -$ million + Financial Overview +Annual Results -1,007 +4 -827 + FY18 Financial Summary -482 +GAAP -462 +Non-GAAP1 -396 +Annual Results -Institutional Securities Global Wealth Management +Increase / (Decrease) y/y -Asset Management Discover +Annual Results -Asset Management +Increase / (Decrease) y/y -Pre-tax profit margin +Revenue -30% +$10,480M -26% +Deferred Revenue -35% +$7,095M -29% +Unbilled Deferred Revenue -28% +N/A 25% -22% - -21% - -Return on average -common equity +28% -36% +N/A -24% +Operating Margin -23% +2.2% -17% +148 bps -FY 2003 +Diluted Earnings Per Share -FY 2004 +$0.172 -FY 2005 +Operating Cash Flow -1st Half -2006 +$2,738M -(1) +(35%) -FY 2003 +27% -FY 2004 +$10,381M CC -FY 2005 +24% CC -1st Half 2006 +$6,965M CC -(1) +25% CC -FY 2005 +$13,300M -1st Half 2006 +48% -(1) +14.5% -MSIM +$1.352 -MSIM excl. Private Equity +N/A -MSIM +128 bps -Morgan Stanley +34%3 -1. Six months ended May 31 +N/A -Page +1The Non-GAAP columns present only non-GAAP financial metrics and the related non-GAAP growth rates as compared to prior periods. Non-GAAP +revenue and non-GAAP deferred revenue represent CC results. Refer to slide 8 for an explanation of non-GAAP CC revenue and to slide 14 for an +explanation of non-GAAP CC deferred revenue. Unbilled Deferred Revenue is an operational measure and represents future billings under our non- +cancelable subscription agreements that have not been invoiced and, accordingly, are not recorded in deferred revenue. Non-GAAP operating margin +and non-GAAP EPS are non-GAAP financial measures. Refer to the Appendix for an explanation of non-GAAP financial measures, and why we believe +these measures can be useful, as well as a reconciliation of non-GAAP financial measures to the most comparable GAAP measures, when applicable. +2Diluted EPS calculation utilizes GAAP revenue results disclosed above. 3Non-GAAP Diluted EPS year-over-year growth is calculated using the EPS +results for the current and prior periods rounded to the nearest whole cent, as presented in the Appendix. 5 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Key Strengths - -STRENGTHS OF MSIM TODAY - -• Strong brand recognition - -– Morgan Stanley - -– Van Kampen + FY18 Financial Results -• Global footprint +$10,480M -• Scale +$8,392M -• Breadth of products +$6,667M -• Multiple channels of distribution +$5,374M -• Community of boutiques +10.7% 12.4% 13.2% 14.5% -• Full firm commitment and resources +1.7% -Page +0.8% -6 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - MSIM Has a Global Footprint - -Products and Distribution +2.2% -STRENGTHS OF MSIM TODAY +-2.7% -New York - -Amsterdam +$2,738M -Stockholm +$2,162M -Oakbrook Terrace - -Boston - -London - -Paris +$1,672M -Frankfurt +$1,181M -Luxembourg +FY15 -San Francisco -Menlo Park +FY16 -Houston +FY17 -Jersey City +FY18 -West Conshohocken +FY15 -Milan +FY16 -Madrid +FY17 -Mumbai +FY18 -Hong Kong + Quarterly Revenue +Operating Margin +Non-GAAP Operating Margin -Tokyo +1Non-GAAP operating margin excludes the effects of stock-based compensation, amortization of +purchased intangibles, and lease termination resulting from purchase of office building. A complete +reconciliation of GAAP to non-GAAP measures can be found in the Appendix and at +www.salesforce.com. -Singapore - -Melbourne - -MSIM Office Present - -MSIM Distribution Coverage - -Page - -7 - -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. - - STRENGTHS OF MSIM TODAY - -Assets ($ billion) - -1,513 - -1,439 - -1,422 - -• In June 2006, Pensions and -Investments ranked Morgan -Stanley #11 by worldwide -assets - -• MSIM’s strength is in - -traditional, actively managed -investments - -MSIM Has The Scale to Succeed - -Top 20 money managers - -Pensions and Investments, June 2006 - -Ranking - -1 - -2 - -3 - -4 - -5 +Operating Cash Flow 6 -7 + Annual Subscription and Support Revenue by Cloud +Complete portfolio of CRM products -8 ++16% y/y -9 ++24% y/y -10 +$3.6B -11 - -12 +$2.9B -13 - -14 - -15 - -16 - -17 - -18 - -19 - -20 ++34% y/y -Barclays Global +$1.9B -State Street Global ++45% y/y -Fidelity Investments +$1.3B -Vanguard Group +FY18 +FY17 +Sales Cloud -Legg Mason +FY18 -Capital Research +FY17 +Service Cloud -JPMorgan Asset Mgmt. +FY18 -Mellon Financial +FY17 +Salesforce +Platform & Other -Deutsche Asset Mgmt. +FY17 -Northern Trust Global +FY18 -Morgan Stanley +Marketing Cloud +& Commerce Cloud2 -PIMCO +37% -UBS Global Asset Mgmt. - -AllianceBernstein Inst'l - -AIG Global Investment - -Merrill Lynch - -Prudential Financial - -ING - -Wellington Mgmt. - -Goldman Sachs Asset - -958 +29% -864 +FY18 -858 +14% -842 +20% -781 +Subscription and Support +Revenue by Cloud1 -633 +1Percentages represent annual cloud revenues as a proportion of total annual Subscription and Support revenue. 2Marketing Cloud +revenue, excluding Commerce Cloud revenue, grew 34% during FY18. -618 +7 -599 + FY18 Revenue by Region +Strong revenue growth across geographies -594 +AMERICAS -581 +EMEA -579 +APAC -572 +$7,579M ++22% y/y ++22% y/y CC1 -537 +$1,904M ++39% y/y ++31% y/y CC1 -531 +$997M ++26% y/y ++27% y/y CC1 -523 +1Non-GAAP revenue CC growth rates as compared to the comparable prior period. We present CC information for revenue to provide a framework for assessing +how our underlying business performed excluding the effects of foreign currency rate fluctuations. To present CC revenue, current and comparative prior period +results for entities reporting in currencies other than United States dollars are converted into United States dollars at the weighted average exchange rate for the +quarter being compared to for growth rate calculations presented, rather than the actual exchange rates in effect during that period. -521 +8 -496 + Financial Overview +Quarterly Results -Investment type: +9 -T ra dit io na l + Q4 FY18 Financial Summary -A lt e rna t iv e s +GAAP -Inde x +Non-GAAP1 -0 +Quarterly Results -250 +Increase y/y -500 +Quarterly Results -750 +Increase y/y -1,000 +Revenue -1,250 +$2,851M -1,500 +Deferred Revenue -1,750 +$7,095M -Source: Total assets are from June 2006 Pensions and Investments, using year end 2005 data. Morgan Stanley assets reflect Morgan Stanley consolidated assets -at fiscal year end 2005. Investment types, which are derived from June 2006 Pension and Investments and July 2006 Institutional Investors, are for comparison -purposes +Unbilled Deferred Revenue -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +N/A -Page +24% -8 +28% - MSIM Organization 2006 +N/A -MSIM GROWTH STRATEGY +Operating Margin -• Investment expertise is +2.7% -organized around investment -product areas +375 bps -• Distribution is organized +Diluted Earnings Per Share -around clients +$0.092 -John Mack -Firm CEO +Operating Cash Flow -Bob Scully -Firm Co-President +$1,051M -Owen Thomas -MSIM President, COO +229% -Investments +49% -U.S. & Americas Distribution +$2,787M CC -Non-Americas +21% CC -• Focused on communicating -effectively among the groups +$6,965M CC -Dennis Shea -Head and CIO, -Global Equity +25% CC -David Germany -Head and CIO, -Global -Fixed Income +$13,300M -Stuart Bohart -Head and CIO, -Alternatives +48% -Michael Kiley -Head of Americas -Retail & -Intermediary -Distribution +13.5% -Open -Head of -U.S. Institutional -Distribution +$0.352 -Michael Green -Head of MSIM -Non-Americas +N/A -Kenneth Winston -Chief Risk Officer +62 bps -• Decentralized, with all -components of the -organization accountable for -financial results +25%3 -Attracting new talent -Internal +N/A -• Head and CIO Global Equity +1The Non-GAAP columns present only non-GAAP financial metrics and the related non-GAAP growth rates as compared to prior periods. Non-GAAP +revenue and non-GAAP deferred revenue represent CC results. Refer to slide 8 for an explanation of non-GAAP CC revenue and to slide 14 for an +explanation of non-GAAP CC deferred revenue. Unbilled Deferred Revenue is an operational measure and represents future billings under our non- +cancelable subscription agreements that have not been invoiced and, accordingly, are not recorded in deferred revenue. Non-GAAP operating margin +and non-GAAP EPS are non-GAAP financial measures. Refer to the Appendix for an explanation of non-GAAP financial measures, and why we believe +these measures can be useful, as well as a reconciliation of non-GAAP financial measures to the most comparable GAAP measures, when applicable. +2Diluted EPS calculation utilizes GAAP revenue results disclosed above. 3Non-GAAP Diluted EPS year-over-year growth is calculated using the EPS +results for the current and prior periods rounded to the nearest whole cent, as presented in the Appendix. -• Head and CIO Alternatives +10 -• Head of Operations + Total Quarterly Revenue and Operating Margin +Balanced top-line and bottom-line growth -• Two investment teams: +$2,851M -External +$1,809M -• CIO London Equity +12.1% -• Two investment teams in Absolute Return Strategies +1.1% -• Process-Driven team +$1,445M -• Global Mid-Cap Equity Market Neutral team +10.9% -• Defined Value team, Absolute Return Strategies +-2.4% -• Structured Products and Derivatives team +$2,294M -• Pension Asset/Liability expertise +12.9% -• Global Infrastructure sector expertise +13.5% -• Six investment professionals: +-1.0% -• European Senior Loans +2.7% -• Alternative Investment Partners +Q4 FY18 ++24% / ++21% CC1 -• Global Value +y/y revenue growth -• International Value +GAAP operating margin -Page ++510 bps -9 +3 year improvement -• Senior U.S. Institutional sales professional +Non-GAAP operating margin2 -• Head of Morgan Stanley Funds Distribution ++260bps -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +3 year improvement -• Senior Middle East sales professional +Q415 Q116 Q216 Q316 Q416 Q117 Q217 Q317 Q417 Q118 Q218 Q318 Q418 - • MSIM offers a broad array of -traditional products and is -increasing its offering of -alternative products + Quarterly Revenue -• MSIM utilizes multiple +Operating Margin -channels of distribution -globally leveraging the -Morgan Stanley brand and -resources where possible +Non-GAAP Operating Margin -MSIM Product and Distribution Balance +2 -STRENGTHS OF MSIM TODAY +1Refer to slide 8 for an explanation of non-GAAP revenue CC growth rate as compared to the +comparable prior period. 2Non-GAAP Operating Margin is the proportion of non-GAAP income from +operations as a percentage of revenue and is a non-GAAP financial measure. Refer to the Appendix for +an explanation of which items are excluded from our non-GAAP financial measures, and why we believe +these measures can be useful, as well as a reconciliation of non-GAAP financial measures to the most +comparable GAAP measures, when applicable. -May 31, 2006 Ending AUM/AUS = $440 billion +11 -Asset Class + Quarterly Subscription and Support Revenue by Cloud +Complete portfolio of CRM products -3% 3% ++16% y/y -5% ++28% y/y -Distribution +$0.9B -20% +$0.8B -22% ++37% y/y -17% +30% -21% +35% -51% ++33% y/y -Equity +Q4 FY18 -Fixed Income +$0.5B -Money Market +$0.4B -Alternatives +Q417 Q418 +Sales Cloud -Real Estate +Q417 Q418 +Service Cloud -Unit Investment Trusts +Q417 Q418 +Salesforce +Platform & Other -23% +Q417 Q418 +Marketing Cloud +& Commerce Cloud 20% 15% -Americas Retail-Morgan Stanley Branded +Subscription and Support +Revenue by Cloud1 -Americas Retail-Van Kampen Branded +1Percentages represent quarterly cloud revenues as a proportion of total quarterly Subscription and Support revenue. 2Marketing Cloud +revenue, excluding Commerce Cloud revenue, grew 31% during Q4 FY18. -Americas Intermediary +12 -U.S. Institutional + Q4 FY18 Revenue by Region +Strong revenue growth across geographies -Non-U.S. +AMERICAS -Page +EMEA -10 +APAC -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +$2,042M ++19% y/y ++19% y/y CC1 - MSIM Offers Clients a Broad Product Array +$536M ++48% y/y ++31% y/y CC1 -Key Product Type by Investment Group +$273M ++27% y/y ++26% y/y CC1 -STRENGTHS OF MSIM TODAY +1Refer to slide 8 for an explanation of Non-GAAP CC revenue. -Global Fixed Income -J. David Germany, CIO -AUM=$166 billion (1) +13 -• Core/Core Plus -• Long Duration -• Mortgage -• Corporate -• Targeted Duration -• Intermediate Duration + Quarterly Deferred Revenue: Billed and Unbilled +Consistent growth on and off the balance sheet -• High Yield -• Short Duration -• Enhanced LIBOR +$9,021M -• Customized -• Global/International -• Emerging Markets Debt +$5,700M -• Institutional Liquidity -• Enhanced Cash -• Retail Money Markets +$3,321M -Global Equity -Dennis Shea, CIO -AUM=$241 billion (1) +$11,392M -U.S. Equity -• U.S. Small-, Mid-, Multi-, and Large-Cap +$7,100M -Value +$4,292M -• U.S. Small-, Mid-, Multi-, and Large-Cap +$14,543M -Growth +$9,000M -• U.S. Large-Cap Core -• Systematic Active Equity -• Global REITs (U.S./European/Asian) -• Sector Funds +$5,543M -Global/International Equity -• International Value -• International Growth -• International Small-Cap -• Global Emerging Markets -• Global Core -• Global Small- and Large-Cap Value -• Global Systematic Active Equity -• European -• Japan (Small-Cap, Value, Growth, Core) -• Asian (ex. Japan) and Pacific Basin +$20,395M -Asset Allocation -• Balanced Funds -• Global Tactical Asset Allocation +$13,300M -– Fully Funded +$7,095M -– Global Macro +Q415 Q116 Q216 Q316 Q416 Q117 Q217 Q317 Q417 Q118 Q218 Q318 Q418 -1. Ending AUM/AUS as of May 31, 2006 +Deferred Revenue -Alternatives -Stuart Bohart, CIO -AUM=$20 billion (1) +Unbilled Deferred Revenue -• Absolute Return Strategies (Hedge Funds) +1 -– Quantitative Equity Strategies -– Fundamental Equity Strategies -– Global Macro Strategies +1Unbilled Deferred Revenue is an operational measure and represents future billings under our non-cancelable subscription +agreements that have not been invoiced and, accordingly, are not recorded in deferred revenue. 2Non-GAAP CC growth rates as +compared to the comparable prior periods. We present CC information for deferred revenue to provide a framework for assessing +how our underlying business performed excluding the effects of foreign currency rate fluctuations. To present CC deferred revenue, +we convert the deferred revenue balances in local currencies in previous comparable periods using the United States dollar currency +exchange rate as on the most recent balance sheet date. -• Alternative Investment Partners +Q4 FY18 +y/y growth -– Fund of Hedge Funds -– Fund of Private Equity Funds -– Insurance Dedicated Fund of Hedge ++40% -Funds +Total Billed and Unbilled +Deferred Revenue -– Portable Alpha ++48% -• Senior Loans +Unbilled Deferred Revenue -− Collateralized Loan Obligations ++28% / ++25% CC1 -• Structured Products and Derivatives +Deferred Revenue -• Managed Futures +14 -• Private Equity + Quarterly Operating Cash Flow +Operating cash flow growth outpaces revenue growth -Unit Trusts -AUS=$13 billion (1) +$337M -• Unit Investment Trusts +$470M -Page +$1,051M -11 +$706M -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +Q4 FY18 +y/y growth - MSIM Utilizes a Multi-Channel Distribution Approach ++49% -STRENGTHS OF MSIM TODAY +operating cash flow -Core Distribution -Channels +Q415 Q116 Q216 Q316 Q416 Q117 Q217 Q317 Q417 Q118 Q218 Q318 Q418 -Customer Tiers +Free Cash Flow -Americas Retail- -Morgan Stanley Branded +1 -Americas Retail- -Van Kampen Branded +Capex -Americas Intermediary +1Free cash flow is a non-GAAP financial measure. Refer to the Appendix for an explanation of non-GAAP +financial measures, and why we believe these measures can be useful, as well as a reconciliation of non- +GAAP free cash flow to the most comparable GAAP measure. -U.S. Institutional +15 -Non-U.S. + Quarterly Cash and Marketable Securities -Morgan Stanley Global -Wealth Management +Q4 FY18 +y/y growth -• Financial Advisors ++105% -• Private Wealth -Management -Investment -Representatives +Total Cash and +Marketable Securities -• Broker Dealers ++228% -• Banks +Marketable Securities -• Investment Advisors ++58% -• Intermediaries +Cash -• Banks +$4,522M -• Financial Planners +$2,725M -• Consultants -• Corporations -• Public Plans -• Foundations -/Endowments +$2,209M -• Insurance General +$1,890M -Accounts/ Financial -Institutions +Q415 Q116 Q216 Q316 Q416 Q117 Q217 Q317 Q417 Q118 Q218 Q318 Q418 -• Consultants +Cash -• Corporations +Marketable Securities -• Public Plans +16 -• Private Banks + Business Overview -• Intermediaries +17 -Core Investment -Vehicles/Products + CRM is Now the Fastest Growing Enterprise Software Segment -Morgan Stanley Brand +$75B -Van Kampen Brand +$60B -Van Kampen Brand +$45B -• Separate Accounts +$30B -• Separate Accounts +$15B -• Mutual Funds +$0B -• Mutual Funds +O/S -• Mutual Funds +ERP -• Pooled Trust +DBMS -• Liquidity Products +CRM -• Institutional Funds +O/S -• Variable Annuities +ERP -• Platform Business +DBMS -• Liquidity Products +CRM -• Société +2016 -• Affiliated Separately -Managed Accounts +2021 -• Variable Annuities +Chart created by Salesforce based on Gartner research. CRM and ERP is part of the Enterprise +Application Software market. DBMS and O/S is part of the Infrastructure Software market. +Source: Gartner, Forecast: Enterprise Software Markets, Worldwide, 2014-2021, 4Q17 Update, 12.15.17 -• Alternative -Investments +O/S* -• Unit Investment Trusts +ERP -• Alternative -Investments +DBMS* -• External Separately -Managed Accounts +CRM -• Alternative -Investments +18 -• Variable Annuities + $72 Billion Market Opportunity Today… -• Institutional Funds +CY16 Market +Share1 -• Unit Investment Trusts +38% -Morgan Stanley Brand +19% -• Institutional Funds +11% -• Liquidity Products +4%2 -• Société +5% -d'investissement à -Capital Variable -(SICAV) Funds +<1% -Sub-Advised +CY16 TAM3 -• Mutual Funds +$12B -d'investissement à -Capital Variable -(SICAV) Funds +$8B -• Alternative -Investments +$7B -• Investment Trust +$5B -Management -Company (ITM) Funds +$26B -Ending AUM/AUS -As of May 31, 2006 +$14B -Page +Sales -12 +Service -$96 billion +Marketing -$90 billion +Commerce -$67 billion +Platform & Other -$98 billion +Analytics -$89 billion +Calculations performed by Salesforce and graphics created by Salesforce based on Gartner research. 1Source: Sales, Service, Marketing, and Commerce +per Gartner, Market Share All Software Markets, Worldwide, 2016, 4.14.17. Platform & other and Analytics market share calculated by Salesforce using the +definitions below. 2Salesforce acquired Demandware on July 11, 2016. Market share includes Demandware’s 2016 commerce contribution of $115.8M. +3Source: Gartner Total Enterprise Software Revenue for All Software Segments and Regions, 2014-2021, 4Q17 Update, 12.15.17. Platform & Other market +defined as Application Platform Software, Application Platform Software as a Service (aPaaS), Other AD, Data Integration Tools, Business Process +Management Suites, Identity Governance and Administration, Mobile Application Development Platforms, Testing, Full Life Cycle API Management, Web +Access Management. Analytics market defined as Modern BI Platforms, Traditional BI Platforms, Data Science Platforms, Analytic Applications. -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +19 - • Performance is key to driving growth + …Growing to $123B in 2021 -MSIM Investment Performance +Market Growth +(CAGR 16-21)1 -• On February 22nd, 2006, Barron’s +12% -U.S. Mutual Funds Performance +15% -STRENGTHS OF MSIM TODAY +16% -annual ranking placed MSIM +15% -– 30th out of 65 fund families for 1- +CY21 TAM1 -year performance +$24B -– 24th out of 56 for 5-year +$14B -performance +$16B -– 4th out of 40 for 10-year +$10B -performance +8% -– The 1-year rankings placed MSIM +$38B -outpacing Vanguard, Franklin +9% -Templeton, JPMorgan and Putnam +$21B -• On July 5th, 2006, Wall Street Journal +Sales -named seven MSIM funds as "Category +Service -Kings": +Marketing -Large-Cap Growth +Commerce -– Morgan Stanley Institutional Focus +Platform & Other -Equity +Analytics -% Long-Term Fund Assets Above Lipper Median +Calculations performed by Salesforce and graphics created by Salesforce based on Gartner research. 1Source: Gartner Total Enterprise Software Revenue +for All Software Segments and Regions, 2014-2021, 4Q17 Update, 12.15.17. Platform & Other market defined as Application Platform Software, +Application Platform Software as a Service (aPaaS), Other AD, Data Integration Tools, Business Process Management Suites, Identity Governance and +Administration, Mobile Application Development Platforms, Testing, Full Life Cycle API Management, Web Access Management. Analytics market +defined as Modern BI Platforms, Traditional BI Platforms, Data Science Platforms, Analytic Applications. Note: data has been updated to reflect +Gartner’s forecast as of December 2017 and therefore are different from what was presented at Salesforce’s Investor Day on November 7, 2017. -67% +20 -61% + The Only Complete Cloud CRM Platform +No other company is better positioned to drive digital transformation -60% +COMMUNITIES +Customer, partner, and employee experiences -48% +SERVICE +Drive service transformations -71% +SALES +Transform sales from +lead to cash to loyalty -59% +COMMERCE +Intelligent, unified shopper +experiences across any channel -61% +MARKETING +Deliver personalized consumer +engagement at scale -63% +COLLABORATION +Quip team collaboration platform -71% +INDUSTRIES +World’s #1 CRM reimagined +for your industry -71% +21 -76% + Salesforce. #1 in CRM +Worldwide CRM applications 2017H1 market share by IDC -76% +19.9% -1 Year +8.4% -3 Year +6.1% -5 Year +2012 -Aug'05 +2013 -Nov'05 +2014 -Feb'06 +2015 -May'06 +2016 -% Rated Assets with Morningstar 4/5-Star Rating +2017 H1 -– Morgan Stanley Institutional U.S. +Source: IDC Worldwide Semiannual Software Tracker, October 2017. -47% +22 -45% + Traction in Target Industries +Industry leaders rely on Salesforce -Large Cap Growth +Financial Services -– Morgan Stanley Growth +HLS -Multicap Growth +Retail -– Morgan Stanley Capital +Travel, Transport & Hospitality -Opportunities +18 OF 20 -– Morgan Stanley Aggressive Equity +Top U.S. & European +Banks1 -Pacific Region +16 OF 20 -– Morgan Stanley Pacific Growth +Top Global +Pharmaceutical Companies -Real Estate +11 OF 20 -– Morgan Stanley Institutional +Top Global +Retailers -International Real Estate +7 OF 10 -Page +Top Hospitality +Companies -13 +Manufacturing -39% +8 OF 10 -39% +Top Discrete +Manufacturers -Aug'05 +Communications & Media -Nov'05 +Consumer Goods -Feb'06 +High Tech -May'06 +8 OF 10 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +Top Global +Telecom Companies - Section 2 +11 OF 15 -MSIM Growth Strategy +Top CPG +Companies -Page +12 OF 15 -14 +Top Tech +Companies -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +1Represents top 10 U.S. banks and top 10 European banks. - Growth Strategy +23 -MSIM GROWTH STRATEGY + The Salesforce Economy +Thriving ecosystem of customers, partners, and developers -Our Strategy for Growth +$859B -Critical Initiatives +in GDP impact by 20201 -• Increase value of Morgan Stanley’s Asset +3.3M -• Build alternatives capability +Salesforce related +jobs created by 20201 -Management business +2 of theTop 10 -— PBT growth +jobs in 2017 are Salesforce-specific -— Multiple (P/E) expansion by +300K -• Create new traditional products using existing -capabilities and through lift-outs / acquisitions +job postings today with Salesforce +related skills3 -• Expand non-U.S. footprint: distribution and +1IDC White Paper sponsored by Salesforce, The Salesforce Economy Forecast: +3.3 Million New Jobs, $859 Billion New Business Revenues to Be Created from +2016 to 2022, October 2017. 2Indeed, Mar 2017 3Burning Glass Report, +“Salesforce Skills Demand”, Aug 2016 -demonstrating net sales and AUM growth +24 -products + Global Events and Ad Campaign +Extending our brand presence globally -• Focus on revenue growth and diversification +World Tour & Events -• Invest in Van Kampen - Intermediary and Unit +Ad Campaign -by investing in the business +Sydney +March 6 -Investment Trusts growth +TrailheaDX +March 28 - 29 -— People +Washington D.C. +April 4 -— Acquisitions +Boston +April 5 -— Infrastructure +Amsterdam +April 12 -• Leverage Morgan Stanley franchise and +New York +April 26 -MSIM scale and positioning +Toronto +May 3 -• Enhance traditional product strength and build +London +May 17 -Alternatives +Paris +June 6 -• Performance, performance, performance +Dreamforce +Sept. 25 - 28 -• Stem outflows of Morgan Stanley retail mutual +25 -funds + 26 -• Rebuild institutional reputation and business, + Appendix -particularly in the U.S. +27 -Page + Seasonality of Financial Metrics +Deferred Revenue and Operating Cash Flow -15 +28 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. + Seasonal Nature of Deferred Revenue and Operating Cash Flow +Quarterly metric seasonality evolving - Build a Much Broader Alternatives Capability +• -What We Have +Deferred revenue primarily consists of billings to customers for our subscription service and we generally invoice our customers in +annual cycles. -What We’ve Done +• We typically issue renewal invoices in advance of the renewal service period, and depending on timing, the initial invoice for the +subscription and services contract and the subsequent renewal invoice may occur in different quarters. This may result in an +increase in deferred revenue. -What We’re Doing +• -MSIM GROWTH STRATEGY +• -• The Alternative Investments +There is a disproportionate weighting towards annual billings in the fourth quarter, primarily as a result of large enterprise account +buying patterns. Our fourth quarter has historically been our strongest quarter for new business and renewals. The year on year +compounding effect of this seasonality in both billing patterns and overall new and renewal business causes the value of invoices +that we generate in the fourth quarter for both new business and renewals to increase as a proportion of our total annual billings. -group is composed of 4 distinct -groups: +Given the year on year compounding effect of this seasonality, the quarterly sequential changes in Deferred Revenue and +Operating Cash Flow have historically become more pronounced. -– Absolute Return Strategies +o As it relates to Deferred Revenue, the sequential change from our third quarter to our fourth quarter has historically increased. The -(ARS) +sequential change in our first three quarters has historically decreased. -– Alternative Investment +o As it relates to Operating Cash Flow, our first quarter and, increasingly, our fourth quarter are our largest collections and operating +cash flow quarters. Our second quarter and third quarter are seasonally smaller in regards to collections and operating cash flow. -Partners (AIP) +29 -– Senior Loans + Q1 Deferred Revenue Sequential Change +Q1 seasonality evolving -– Structured Products and +h +t +w +o +r +g -Derivatives (SP&D) +l -First Quarter 2006 -• Internal transfer of Structured -Products and Derivatives team +a +i +t +n +e +u +q +e +S -• Completed lift-out of Process-Driven +20% -Strategies team +15% -• Launched / incubated 6 new +10% -alternative investment products +5% -Second Quarter 2006 -• Hired Stuart Bohart as the head and +0% -CIO of Alternatives +-5% -• Completed Oxhead Capital +-10% -acquisition +-15% -• Internal transfer of Defined Value +9% -team +8% -• Internal transfer of sector expertise +5% -to run the global infrastructure -initiative (initiated June 2006) +4% -• Launched Japanese Equities Market +Q105 Q106 Q107 Q108 Q109 Q110 Q111 Q112 Q113 Q114 Q115 Q116 Q117 Q118 Q119 +Guide -Neutral product +1 -• Re-opened Traxis to new investors +(2%) -Absolute Return Strategies +(2%) -• Continue efforts on lift-out / -acquisitions, and minority -stakes to build business +(3%) -Alternative Investment Partners -• Expand AIP product line and +(6%) -introduce new funds +(8%) -• Distribution focus on smaller +(7%) -mid-market and high net worth +(8%) -Senior Loans +(8%) -• Growth driven by client focused +(7%) -new products through -introduction of funds, -Collateralized Loan Obligations, -Separately Managed Accounts +(9%) -Structured Products and -Derivatives +(12%) -• Develop custom approaches to +1Based on the mid-point of the guidance provided on February 28, 2017. -build structured products -• Introduce Liability Driven +30 -Investments + Q2 Deferred Revenue Sequential Change +Q2 seasonality evolving -Page +20% -16 +18% -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +12% - Enhance Traditional Product Offerings +11% -Through both Organic Launches and Bolt-on Acquisitions/Lift-Outs -What We Have +9% -What We’ve Done +2% -• Many top performing equity -strategies at or near capacity +3% -• Fixed Income products with -improved performance, and -capacity +2% -• Good performing strategies +Q205 -building three year track records +Q206 -• Product gaps in large asset +Q207 -gathering categories +Q208 -First Quarter 2006 -• Launched / incubated 5 new +Q209 -equity products +0% +Q210 -Second Quarter 2006 -• Launched / incubated 8 new +Q211 -products +Q212 -– 7 Equity -– 1 Fixed Income +Q213 -MSIM GROWTH STRATEGY +Q214 -What We’re Doing +Q215 -Organic launches +3% -• Leverage current talent pool to +1% -build out strong product -pipeline +0% -Acquisitions / lift-outs +h +t +w +o +r +g -• Targeted acquisitions / lift-outs -to complement our existing -product line +l -Page +a +i +t +n +e +u +q +e +S -17 +15% -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +10% - Expand Non-U.S. Footprint +5% -Distribution and products +0% -What We Have +-5% -• $89 billion, or 20% of MSIM +-10% -AUM/AUS is sourced from non- -U.S.(1) +(1%) -• MSIM is represented in 13 non- +Q216 -U.S. countries +Q217 -• Outside of the U.S., portfolio +Q218 -managers are located in: +(4%) -– Europe -– Japan -– Singapore -– India +(5%) -What We’ve Done +31 -First Quarter 2006 -• Increased distribution focus on + Q3 Deferred Revenue Sequential Change +Q3 seasonality evolving -Alternatives +21% -• Signed a global agreement with +8% 8% -Merrill Lynch Global Private -Client division to distribute MSIM -products through Merrill Lynch's -International distribution network -worldwide +6% -Second Quarter 2006 -Middle East -• Hired sales head -• Leveraging Firm’s Dubai office +Q305 -China -• Dedicated sales hire -• Applied (and gained the approval +Q306 -in July 2006) for license to -manage domestic Chinese equity -products +Q307 -MSIM GROWTH STRATEGY +Q308 -What We’re Doing +Q309 -Expand the footprint -• Asia +Q310 -– Korea -– Australia -– China -– India +Q311 -• Europe +(2%) -– Middle East -– Switzerland +Q312 +(2%) -• Fill product gaps +2% -• Explore new client opportunities: +(1%) -central banks and direct -wholesaling +h +t +w +o +r +g -• Evaluate acquisitions in +l -– Australia -– China -– Brazil -– Korea -– Taiwan +a +i +t +n +e +u +q +e +S -1. +25% -Total MSIM AUM/AUS of $440 billion as of May 31, 2006 +20% -Page +15% -18 +10% -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +5% - Invest in Van Kampen - Intermediary and Unit Investment Trusts Growth -What We Have +0% -What We’re Doing +-5% -What We’ve Done +-10% -MSIM GROWTH STRATEGY +-15% -• Vibrant $100+ billion fund and Unit +Q313 -Investment Trusts complex +Q314 -(1) +Q315 -First Quarter 2006 -• Launched Van Kampen +Q316 -– 6th largest 3rd party fund sponsor -– One of the fastest growing +Q317 -(2) +Q318 -complexes over the last 4 years +(3%) -(2) +(3%) -– 2nd largest Unit Investment +(5%) -Trusts sponsor +(6%) -(3) +(9%) -• Strong brand presence within broker +(9%) -community +32 -• Established leader with distinguished -capabilities in several key product -areas: U.S. Large Cap Value / U.S. -Moderate Allocation / High Yield -Municipals / Senior Loans + Q4 Deferred Revenue Sequential Change +Q4 seasonality evolving -• Leader in intermediary platform +h +t +w +o +r +g -market +l -– On 20 of top 25 annuity platforms -– On 22 of top 25 Defined -(5) -Contribution platforms +a +i +t +n +e +u +q +e +S -– Attractive blend of run rate and +60% -(4) +45% -takeover business flows +30% -International Growth Fund and -Van Kampen Leaders Fund +29% -• Introduced new retirement +33% -education and support materials +29% -• Increase resources for high growth / +41% -margin businesses: +Annual +invoicing1 -– Defined Contribution: +35% -Doubling field sales force -– Unit Investment Trusts: +27% -38% increase in field sales force +29% -• Added Variable Insurance field +62% -• Expand product offering to add +59% -sales professionals +50% -diversity +49% 51% -– International & Global -– Closed End Funds -– Income Oriented products -– Retail Separately Managed +44% 45% -Account strategies +15% -• Position Van Kampen to take +0% -advantage of retirement markets -from a product and distribution -standpoint +Q405 -Second Quarter 2006 -• Launched Van Kampen Equity +Q406 -Premium Income +Q407 -• Issued 14 new Unit Investment +Q408 -Trusts products +Q409 -• Introduced new television +Q410 -advertising +Q411 -1. AUM/AUS as of May 31, 2006 -2. Source: Strategic Insights, May 2006 -3. Source: ICI, May 2006 +Q412 -Page +Q413 -19 +Q414 -4. Source: VARDS, 4th quarter 2005 -5. Source: Pensions and Investments, June 2005 +Q415 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +Q416 - MSIM GROWTH STRATEGY +Q417 -Stem Outflows of Morgan Stanley Retail Mutual Funds -What We Have +Q418 -What We’ve Done +33 -What We’re Doing + Operating Cash Flow Sequential Change +OCF seasonality evolving -• Morgan Stanley Advisors funds -(excluding Money Market and -Closed End) represents 7% of -total MSIM AUM/AUS (1) +h +t +w +o +r +g -• Favorable market share in +l -Morgan Stanley Global Wealth -Management Group (GWMG) +a +i +t +n +e +u +q +e +S -• Outflows in Morgan Stanley +700% -Advisor products +600% -First Quarter 2006 -• Deployed a new client focused +500% -field coverage model +400% -– Assigned dedicated internal -and external wholesalers to -distributing Morgan Stanley -Funds in Morgan Stanley -GWMG +300% -Second Quarter 2006 -• Hired the head of Morgan -Stanley Funds Distribution +200% -– A newly created position to +100% -focus on the Morgan Stanley- -branded funds +0% -• Preserve favorable market share +-100% -in Morgan Stanley GWMG +736% -• Actively pursue third party +358% -distribution of Morgan Stanley -Funds in 401(K), financial -planner, and Registered -Investment Advisor communities +74% -• Develop distinctive Morgan +99% -Stanley fund offering +74% -– Launch distinctive new +6% -products +172% -– Rationalize existing product +118% -line +189% -1. +124% -Total MSIM AUM/AUS of $440 billion as of May 31, 2006 +Q114 Q214 Q314 Q414 Q115 Q215 Q315 Q415 Q116 Q216 Q316 Q416 Q117 Q217 Q317 Q417 Q118 Q218 Q318 Q418 +(59%)(47%) -Page +(36%)(23%) -20 +(50%)(48%) -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +(38%) - Rebuild Institutional Reputation and Business, Particularly in the U.S. -What We Have +(76%) -What We’re Doing +(73%)(62%) -What We’ve Done +34 -MSIM GROWTH STRATEGY + Non-GAAP Financial Measures -• $187 billion in AUM, 43% of +This presentation includes information about non-GAAP diluted earnings per share, non-GAAP income from operations, non-GAAP free cash flow, and constant currency revenue and +constant currency deferred revenue growth rates (collectively the “non-GAAP financial measures”). These non-GAAP financial measures are measurements of financial performance that are +not prepared in accordance with U.S. generally accepted accounting principles and computational methods may differ from those used by other companies. Non-GAAP financial measures +are not meant to be considered in isolation or as a substitute for comparable GAAP measures and should be read only in conjunction with the company’s consolidated financial statements +prepared in accordance with GAAP. Management uses both GAAP and non-GAAP measures when planning, monitoring, and evaluating the company’s performance. -MSIM total (1) +The primary purpose of using non-GAAP measures is to provide supplemental information that may prove useful to investors and to enable investors to evaluate the company’s results in the +same way management does. Management believes that supplementing GAAP disclosure with non-GAAP disclosure provides investors with a more complete view of the company’s +operational performance and allows for meaningful period-to-period comparisons and analysis of trends in the company’s business. Further, to the extent that other companies use similar +methods in calculating non-GAAP measures, the provision of supplemental non-GAAP information can allow for a comparison of the company’s relative performance against other +companies that also report non-GAAP operating results. -• MSIM covers over 100 consulting +Non-GAAP diluted earnings per share excludes, to the extent applicable, the impact of the following items: stock-based compensation, amortization of acquisition-related intangibles, +amortization of acquired leases, the net amortization of debt discount on the company’s convertible senior notes, gains/losses on conversions of the company’s convertible senior notes, +gains/losses on sales of land and building improvements, gains/losses on company-initiated acquisitions of entities in which the company held an equity investment, and termination of +office leases, as well as income tax adjustments. These items are excluded because the decisions that give rise to them are not made to increase revenue in a particular period, but instead +for the company’s long-term benefit over multiple periods. -firms +Non-GAAP income from operations excludes the impact of the following items: stock-based compensation, amortization of acquisition-related intangibles, and termination of office leases. -• Manages assets for 20 of the top -50 U.S. Pension funds, 56 of the -top 200 U.S. Pension funds (2) +The company defines the non-GAAP measure free cash flow as GAAP net cash provided by operating activities, less capital expenditures. For this purpose, capital expenditures does not +include our strategic investments, nor does it include any costs or activities related to our purchase of 50 Fremont land and building, and building - leased facilities. -First Quarter 2006 -• Launched / incubated 9 new -products within the channel +Constant currency information is provided as a framework for assessing how our underlying business performed excluding the effect of foreign currency rate fluctuations. To present constant +currency revenue, current and comparative prior period results for entities reporting in currencies other than United States dollars are converted into United States dollars at the weighted +average exchange rate for the quarter being compared to for growth rate calculations presented, rather than the actual exchange rates in effect during that period. To present deferred +revenue on a constant currency basis, we convert the deferred revenue balances in local currencies in previous comparable periods using the United States dollar currency exchange rate as +of the most recent balance sheet date. -• Conducted road shows with +35 -consultants and clients + GAAP to Non-GAAP Financial Reconciliation +Annual Results -Second Quarter 2006 +(in thousands) +Non-GAAP income from operations +GAAP income (loss) from operations +Plus: + Amortization of purchased intangibles + Stock-based expense +Less: + Operating lease termination resulting from purchase of 50 Fremont, net +Non-GAAP income from operations -• Launched / incubated 9 new -products within the channel +Fiscal Year Ended January 31, +2018 +235,768 -• Internal transfer of Pension Asset -/ Liability management expertise +2017 +64,228 -• Senior management meetings +$ -and road shows with consultants -and clients +$ -• Hosted the annual investment -roundtable of industry thought -leaders (Lyford Cay Conference) +$ -Client +2016 +114,923 -• Communicate MSIM story to +2015 +(145,633) -clients and consultants -• Form Institutional Client +$ -Solutions Group +286,885 +997,013 -• Leverage client service +225,277 +820,367 -People +158,070 +593,628 -• Hire US Institutional leader -• Hire additional salespeople +154,973 +564,765 -Product +0 +1,519,666 -• Develop products to address +$ -client needs +0 +1,109,872 -• Enhance ability to place +$ -alternatives +(36,617) +830,004 -• Communicate Fixed Income +$ -turnaround story +0 +574,105 -• Retain clients in International +$ -Value Equity products +Non-GAAP diluted earnings per share +GAAP diluted net income (loss) per share +Plus: +Amortization of purchased intangibles +Amortization of acquired lease intangible +Stock-based expense +Amortization of debt discount, net +Less: +Gains from acquisitions of strategic investments +Income tax effects and adjustments +Non-GAAP diluted earnings per share +Shares used in computing Non-GAAP diluted net income per share -1. As of May 31, 2006 -2. Source: P&I Databook, January 23, 2006. It includes Corporate and Public Funds +Fiscal Year Ended January 31, -Page +$ -21 +2018 +0.17 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +$ - Summary: Measuring Our Progress +2017 +0.26 -MSIM GROWTH STRATEGY +0.39 +0.00 +1.36 +0.04 -First Year +0.32 +0.00 +1.17 +0.04 -Two to Three Years +0.00 +(0.61) +1.35 +734,598 -Three to Five Years -Three to Five Years +$ -Business Investment +(0.02) +(0.76) +1.01 +700,217 -Gaining Traction +$ -Results +36 -• New management team + GAAP to Non-GAAP Financial Reconciliation +Quarterly Results -• Filled leadership gaps +(in thousands) +Non-GAAP income from operations +GAAP income (loss) from operations +Plus: + Amortization of purchased intangibles + Stock-based expense +Non-GAAP income from operations -• Reorganized business by +Three Months Ended January 31, -product and client segment +2018 +77,869 -• Bottom up business plan +$ -• Seeded new products +2017 +(23,345) -• Built infrastructure +$ -• Actively pursued new talent to +2016 +20,560 -manage money +$ -• Opened lines of communication +2015 +(34,816) -• Positive flows +$ -• More complete traditional +68,932 +237,682 +384,483 -product offering +$ -• Improved margin +74,214 +244,387 +295,256 -• Completion of several +$ -acquisitions +39,250 +158,972 +218,782 -• Multiple direct hedge fund +$ -strategies +39,971 +151,802 +156,957 -• Restructured Morgan Stanley +$ -Advisor Fund family +Non-GAAP diluted earnings per share +GAAP diluted net income (loss) per share +Plus: +Amortization of purchased intangibles +Amortization of acquired lease intangible +Stock-based expense +Amortization of debt discount, net +Less: +Gains from acquisitions of strategic investments +Income tax effects and adjustments +Non-GAAP diluted earnings per share +Shares used in computing Non-GAAP diluted net income per share -• Leader in the asset +(in thousands) +Free cash flow analysis, a non-GAAP measure +Operating cash flow +GAAP net cash provided by operating activities +Less: +Capital expenditures +Free cash flow -management business with a -reputation for excellent -performance, innovative -products and superior client -service +Three Months Ended January 31, -Page +$ -22 +2018 +0.09 -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +$ - Measuring Our Performance: Key Success Metrics +2017 +(0.07) -MSIM GROWTH STRATEGY +0.09 +0.00 +0.32 +0.01 -• Net revenues +0.10 +0.00 +0.34 +0.01 -• Pre-tax margin +0.00 +(0.16) +0.35 +749,464 -FY 2005 Results (1) +$ -Aspiration +0.00 +(0.10) +0.28 +711,203 -$2.9 billion +$ -Double digit growth +Three Months Ended January 31, -35% (incl. Private Equity) +2018 -29% (excl. Private Equity) +2017 -30%~35% +2016 -• Assets Under Management +2015 -$431 billion +$ -$600 billion+ +1,051,320 -• Net Flows +$ -($19) billion +706,146 -Strong and competitive -positive flows +$ -• FY New products launched +470,208 -3 +$ -• Retail Product Performance +336,506 -– % of Morningstar rated assets in 4/5 star +(137,759) +913,561 -45% +$ -30+ +(143,974) +562,172 -50%+ +$ -funds +(68,495) +401,713 -1. Full Year ended November 30, 2005 +$ -Page +(85,354) +251,152 -23 +$ -This slide is part of a presentation by Morgan Stanley and is intended to be viewed as part of that presentation. The presentation is based on information -generally available to the public and does not contain any material, non-public information. No representation is made that it is accurate or complete. The -presentation has been prepared solely for informational purposes, is neither an offer to sell nor the solicitation of an offer to buy any security or instrument -and has not been updated since it was originally presented. +37 diff --git a/q159/random_k2/question.json b/q159/random_k2/question.json index 8f16d8cd70cdba62b9af73a2de1dc71c4bbd3778..adadd2f44c837a964e8f3e3382e6ec0a7d3f224c 100644 --- a/q159/random_k2/question.json +++ b/q159/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "f8d3a162ab9507e021d83dd109118b60.pdf", - "web_ed8636d8f0faddf2.pdf", - "web_289916b20eaef9df.pdf" + "web_bf09a64e7d517c98.pdf", + "web_9942a44c5f6232cc.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q159/random_k2/random_1.md b/q159/random_k2/random_1.md index 4df5215b83d8f847c48c411dacd211a9574481da..bdc0cc473d09f25f5a2f7f0b1e46304031484979 100644 --- a/q159/random_k2/random_1.md +++ b/q159/random_k2/random_1.md @@ -1,752 +1,637 @@ -C H A P T E R 1 +Youth in Middle East +and North Africa -Get started with -Web GIS +August 2019 -T his chapter introduces the basic concepts and workflows of Web GIS and the Esri® geo- +Daniella Raz +University of Michigan -spatial cloud. The chapter begins with an overview of Web GIS and its advantages; intro- -duces Esri’s geospatial cloud including ArcGIS® Online and ArcGIS Enterprise; illustrates online, -on-premises, and hybrid deployment patterns; lists the technical evolutions in Web GIS; explains -the content types, hosted feature layers, app types, and user types in Web GIS; and demonstrates -the workflow to build Web GIS apps using the Attachment Viewer instant app template. This -chapter familiarizes you with ArcGIS Online and ArcGIS Enterprise basic operations and work- -flows and introduces flexible ways to build Web GIS apps that you will explore in other chapters. +JULY 2018 Arab Barometer – Wave V +Topic Report – Youth -Objectives +Executive Summary -• Grasp the concept and advantages of Web GIS. -• Understand the deployment patterns of Web GIS. -• Learn the components of the new-generation Web GIS. -• Understand the technical evolutions and trends in Web GIS. -• Understand the concepts of web services and hosted feature layers. -• Publish hosted feature layers from CSV files. -• Learn the workflow for creating web maps and web apps. +Individuals between the ages of 15 and 29 account for roughly 30 percent of the +population of the Middle East and North Africa (MENA).1 The beliefs, attitudes, +and preferences of this large and growing segment of the population has the po- +tential to shape and drive the region’s future. Yet, widespread youth discontent +about their economic prospects poses a significant challenge. Youth unemploy- +ment rates across the region remain high and there is broad frustration with +government efforts to create employment opportunities.2 Though satisfaction +with the state education system varies greatly across countries, youth are less +satisfied than are older generations. -Copyright © 2022 Esri. All rights reserved. 2 +Youth also question if they have the ability to voice their frustrations to the gov- +ernment. In nearly all countries, less than half say the right to freedom of ex- +pression is guaranteed while half or less say they have the right to demonstrate +peacefully in all countries. This combination creates conditions where youth +across the MENA region, far more than older age cohorts, are considering emi- +grating from their country, primarily to Europe, North America, or GCC countries. +Unsurprisingly, many cite economic reasons as the main impetus for leaving. -Chapter 1: Get started with Web GIS +However, youth differ from their older counterparts not only in terms of greater +frustration with the existing system. Relative to older generations youth also +tend to be less religious, less interested in politics, and more likely to be engaged +in social media. Generational differences are particularly stark in Morocco where +youth attitudes diverge substantially and dissatisfaction with the overall state of +affairs is quite salient. -Web GIS and the geospatial cloud +Meanwhile, on gender issues, most youth tend to be supportive of equal rights +for women. Majorities say women have the right to be the head of government +and to receive an education. Yet, the same does not hold true for equal inher- +itance rights with minorities saying women should receive the same as men +in most countries. Additionally, this generation cites low levels of support for +women playing equal roles in society, with most saying that men make better +political leaders and stating that husbands should have the final say in family +matters. -Web GIS is the combination of the web and geographic information systems (GIS). The web -removed the constraint of distance and thus allows people the freedom to interact with GIS apps -globally and access information almost instantly. Web GIS uses web technologies, including but -not limited to Hypertext Transfer Protocol (HTTP), Hypertext Markup Language (HTML), -Uniform Resource Locator (URL), JavaScript, Web Graphics Library (WebGL), and WebSocket. -The first operational GIS was developed in the 1960s by Roger Tomlinson (who developed -the Canada Geographic Information System). Since then, GIS has evolved from a local file-based, -single-computer system to a central database client/server system, often with multiple servers -and many more client computers. The invention of the internet in the 1980s and the growth -of the World Wide Web in the early 1990s laid the foundation for an evolutionary leap toward -Web GIS. In 1993, the Palo Alto Research Center (PARC), a Xerox company, developed a map- -ping web page, which marked the origin of Web GIS. In the 2000s, Web GIS evolved into a new -generation — a system of distributed web services in the cloud you can access anywhere, repre- -sented by ArcGIS. +Youth in MENA appear to be particularly inclined toward China relative to other +global powers. Majorities favor closer economic relations with China in most +countries surveyed, while less than half say the same about relations with the +U.S. in most countries. -Inheriting the power of the web and the cloud, Web GIS offers many advantages: +1https://www.undp.org/content/dam/rbas/report/AHDR%20Reports/AHDR%202016/AHDR%20Final% -• Global reach: You can share your geographic information easily, within your organization +202016/AHDR2016En.pdf, pg. 7 -and with people all over the world. +2https://www.imf.org/external/np/vc/2012/061312.htm?id=186569 -• Large number of users: You can share your app with dozens, or even millions, of users sup- +www.arabbarometer.org -ported by scalable cloud technology. +2 -• Low cost per user: The cost of building one Web GIS app is often lower than building a + Arab Barometer – Wave V +Topic Report – Youth -stand-alone desktop solution and installing it for every user. +These are among the key findings from 11 nationally representative public opin- +ion surveys conducted across the Middle East and North Africa by the Arab +Barometer from September 2018 - April 2019. The surveys included more than +25,000 face-to-face interviews in the respondent’s place of residence and have +a margin of error of ±2 percent in most countries.3 -• Better cross-platform capabilities: Web apps, especially those built with JavaScript, can +Religiosity -run on desktop and mobile browsers running a wide range of operating systems, including -Windows, Mac OS, and Linux, along with iOS, Android, and Windows Phone. +Across the region, youth are less religious than their older counterparts. In fact, in +every country surveyed less than half of youth identify as religious. More specif- +ically, fewer than one-fifth of youth in Algeria (15 percent), Tunisia (16 percent), +and Libya (18 percent), and fewer than one-quarter of youth in Jordan (22 per- +cent), Lebanon (23 percent), and Morocco (24 percent) consider themselves reli- +gious. Higher rates of youth religiosity are found in Palestine (28 percent), Sudan +(32 percent), Yemen (33 percent) Egypt (36 percent), and Iraq (42 percent). -• Easy to use: Web GIS apps typically incorporate simplicity, intuition, and convenience in -their design. Public users can use these apps without a technical background (see figure). -• Easy to maintain: Web clients can benefit from the latest program and data updates each -time they access a web app. The web administrator does not have to update all the clients -separately. +However, trends over time paint a more complicated picture throughout the +region. While indeed less than half of MENA youth identify as religious, in some +countries these rates represent an increase from the last Arab Barometer wave -Copyright © 2022 Esri. All rights reserved. Web GIS and the geospatial cloud +3For details about each survey, please visit: -3 - -GIS professional -using Web GIS - -Casual users - -Developers - -Open data -community - -Web GIS offers a collaborative environment and a pattern for delivering GIS capabilities. GIS -professionals share information to Web GIS and extend geospatial intelligence to users across -organizations and throughout communities. - -GIS is the science of locations — The Science of Where®. The term has two meanings. One mean- -ing is that GIS is itself a science, as the scientific basis for GIS technology. The other meaning -is that GIS has been used for science as an effective tool for making scientific discoveries. The -Science of Where is now recognized as a force for solving problems and understanding our -world. Web GIS takes the science to a whole new scale, transforming how we share and col- -laborate and revealing deeper insight into data. The common patterns of Web GIS applications -include mapping and visualization, data management, field mobility, monitoring, analytics, -design and planning, decision support, constituent engagement, sharing, and collaboration. -Web GIS unlocks and delivers this science to offices and homes and puts GIS technology in the -hands of billions of people. Web GIS provides immense value to government, business, science, -and daily life. In recent years, the concept and importance of spatial location have become more -mainstream, and Web GIS awareness is growing more prominent in many organizations: - -• Government: Web GIS offers an ideal channel for sharing public information services and -delivering open data, an engaging medium for encouraging public participation, and a pow- -erful framework for supporting decision-making. - -• Business: Web GIS helps create novel business models and reshapes existing ones. It - -enhances the power of location-based advertising, business analysis, and volunteered geo- -graphic information, generating tremendous revenue directly and indirectly. -• Science: Web GIS creates research areas and renews existing avenues of research. -• Daily life: Web GIS helps people decide where to eat, stay, shop, and learn how to get from - -place to place. - -Table 1.1 describes patterns of Web GIS applications, including mapping and visualization, data -management, and analytics. - -Copyright © 2022 Esri. All rights reserved. 4 - -Chapter 1: Get started with Web GIS - -Table 1 .1 . Common patterns of Web GIS applications - -Pattern name - -Description - -Mapping and visualization Understand locations and relationships with maps and visual representations. - -Data management - -Collect, organize, and maintain accurate locations and details about assets -and resources. - -Field mobility - -Monitoring - -Analytics - -Manage and enable a mobile workforce to collect and access information in -the field. +https://www.arabbarometer.org/waves/ -Track, manage, and monitor assets and resources in real time. +arab-barometer-wave-v/ -Discover, quantify, and predict trends and patterns to improve outcomes. +www.arabbarometer.org -Design and planning - -Evaluate alternative solutions and create optimal designs. - -Decision support - -Gain situational awareness and enable information-driven decision-making. - -Constituent engagement - -Communicate and collaborate with citizens and external communities -of interest. - -Sharing and collaboration - -Empower everyone to discover, use, make, and share geographic information. - -The geospatial cloud powers Web GIS -Esri’s Web GIS capabilities are provided by ArcGIS Online and ArcGIS Enterprise, which are at -the core of the geospatial cloud. - -• ArcGIS Online is a software-as-a-service (SaaS) offering of Web GIS. ArcGIS Online is - -hosted on the Amazon Web Services (AWS) cloud and Microsoft Azure cloud platforms and -is managed by Esri. There is no hardware infrastructure for an organization to maintain. -ArcGIS Online hosts hundreds of millions of content items and registered users. It serves bil- -lions of maps per day with high performance, scalability, and availability. With the benefits -of cloud computing, ArcGIS Online has been quickly adopted by government and commer- -cial organizations around the world. - -• ArcGIS Enterprise is a Web GIS software product that organizations can acquire and deploy. - -ArcGIS Enterprise includes four basic software components: Portal for ArcGIS, ArcGIS -Server, ArcGIS Data Store, and ArcGIS Web Adaptor. These components support flexible -ways for deployment, from a single-machine deployment to multiple-machine deployments, -highly available (HA) deployment, on Windows, Linux, and the Kubernetes platform, in -on-premises data centers or in AWS, Microsoft Azure, Google Cloud, and many other cloud -platforms. - -Table 1.2 lists these deployment patterns. +3 -Copyright © 2022 Esri. All rights reserved. Web GIS and the geospatial cloud + Arab Barometer – Wave V +Topic Report – Youth + +while in others a decrease. For instance, rates of religiosity have increased in +Egypt (+12), Yemen (+7), Tunisia (+6), Iraq (+6), Morocco (+2), and Libya (+2). In +contrast, youth identify as religious at lower rates in Sudan (-12), Palestine (-9), +Jordan (-7), Algeria (-3) and Lebanon (-2). + +Views on Politics + +In addition to generally low levels of religiosity, youth appear disinterested in +politics, both overall and when compared to older generations. Only three-in- +ten youth in Lebanon (31 percent), Yemen (29 percent), Palestine (29 percent), +Sudan (28 percent), and Egypt (27 percent) say they are interested or very inter- +ested in politics, while no more than a quarter say the same in Jordan (25 per- +cent), Morocco (23 percent), Iraq (21 percent), and Libya (21 percent). Tunisian +and Algerian youth are least politically interested with fewer than two-in-ten +citing interest in politics (17 percent and 15 percent, respectively). + +With the exception of Egypt and Lebanon, youth in the region are less interested +in politics than are older citizens. The difference between age cohorts is widest +in Tunisia and Palestine (10 percentage points). Elsewhere, though differences +are smaller, older citizens nonetheless report greater interest in politics than do +youth. +In Lebanon youth have only marginally greater interest (1 percentage +point), while in Egypt youth are 6 percentage points more likely than older citi- + +www.arabbarometer.org + +4 + + Arab Barometer – Wave V +Topic Report – Youth + +zens to say they are interested or very interested in politics. + +Youth disinterest in the political system could be in part due to their outlook +If young people perceive the most +on the fairness of parliamentary elections. +recent parliamentary elections as unfair, regardless of whether they voted or +were even eligible to vote in them, then their disinterest could perhaps reflect +disillusionment with their country’s political process. With the exception of the +Levant, youth are more likely than older age cohorts to evaluate the last par- +liamentary elections held in their respective country as being neither free nor +fair. In some countries surveyed this difference is stark. For instance, in Morocco +four-in-ten youth characterized the last parliamentary elections as such, com- +pared to fewer than a quarter of older Moroccans (23 percent). In Tunisia, youth +were 12 percentage points more likely than older citizens to characterize the +last parliamentary elections as being neither free nor fair (40 percent versus 28 +percent). + +Internet and Social Media + +Across the region, internet use, defined as going online at least once-in-a-few- +weeks, is substantially greater among youth than older generations. In fact, in +most of the surveyed countries (with the exception of Yemen), more than 90 +percent of youth use the internet, whereas among the older generation internet +usage does not approach such a rate. The highest rates of internet usage are +found in Lebanon (99 percent), followed by Jordan (96 percent), Palestine (95 +percent), Algeria (93 percent), Morocco (93 percent), Tunisia (91 percent), Egypt +(90 percent), Sudan (90 percent), Iraq (89 percent), and Libya (88 percent). The +lowest rate of internet usage among youth is seen in Yemen, where fewer than +eight-in-ten youth say they use the internet at least once every few weeks (77 +percent). The most striking generational differences in internet usage are seen in +Yemen and North Africa (excluding Libya), where internet use among youth ex- +ceeds that of older age cohorts by at least 30 percentage points. In contrast, the +intergenerational gap in internet use is less pronounced in the Levant (Lebanon, +Jordan, Palestine), where it does not exceed 20 percentage points. + +Interestingly, among internet users, social media usage is very high across all +demographics. In fact, youth are not so different from their older counterparts +in this regard. Though Yemeni youth report the lowest rates of internet usage in +the region, those online are also the most likely to engage in social media. Nearly +all Yemeni youth who are online (99 percent) use social media on a typical day. +However, most online youth across the region also engage in social media on a +typical day: Jordan (98 percent), Lebanon (97 percent), Iraq (97 percent), Pales- +tine (96 percent), Libya (96 percent), Egypt (96 percent), Tunisia (96 percent), +Morocco (96 percent), Sudan (94 percent), and Algeria (94 percent). Overall, +youth are consistently more likely than older generations to say they spend two + +www.arabbarometer.org 5 -Table 1 .2 . Esri Web GIS deployment patterns - -Online - -Hybrid - -On-premises or in cloud - -Esri managed - -ArcGIS Online - -ArcGIS Online - -Customer managed - -ArcGIS Enterprise - -ArcGIS Enterprise - -Although on-premises data center technology is not necessarily on the verge of extinction, cloud -computing is an option with many benefits, including scalability, agility, and cost efficiency. -During the COVID-19 pandemic, cloud usage increased rapidly because people use the cloud -platforms to work and shop from home. Today, more organizations are adopting ArcGIS Online -or deploying their ArcGIS Enterprise to the cloud, such as AWS, Microsoft Azure, Google -Cloud, and others (see figure). Many organizations adopt a hybrid model by integrating ArcGIS -Online and ArcGIS Enterprise. - -ArcGIS Online and ArcGIS Enterprise benefit from an effective framework to enforce security -and protect user privacy. They are certified as compliant with many federal and international -security and privacy standards. For more information, see the ArcGIS Trust Center page at -doc .arcgis .com/en/trust/compliance/compliance-tab-intro .htm. - -Built on ArcGIS Online, the Johns Hopkins COVID-19 Dashboard, a real-time pandemic tracker, has become -a global reference for the pandemic and one of the most cited resources to track the spread of coronavirus -around the world. It has received billions of views and remained highly efficient under a massive amount of -web traffic. - -Copyright © 2022 Esri. All rights reserved. - - 6 - -Chapter 1: Get started with Web GIS - -New-generation Web GIS architecture and trends - -ArcGIS Online and ArcGIS Enterprise represent the new generation of Web GIS. Although the -two have some differences (see chapter 5), they generally share similar architecture, similar capa- -bilities, and similar workflows for creating content, as shown in this figure. - -Desktop - -Web - -Device - -ArcGIS Online - -Portal for -ArcGIS - -Server - -Web services - -C - -l -i - -e -n -t - -a -p -p -s - -A -c -c -e -s -s - -S -e -r -v -i -c -e -s - -ArcGIS is a new-generation Web GIS that provides mapping, analysis, data management, and collaboration. - -This new-generation Web GIS architecture is described as follows: - -• At the center of the new Web GIS architecture is a portal—ArcGIS Online or Portal for - -ArcGIS—which represents a gateway for accessing all spatial products in an organization. -The portal helps organize, secure, and facilitate access to geographic information products. -• On the back end, the portal is powered by two components: GIS servers that allow you to -create content and ready-to-use content such as ArcGIS Living Atlas of the World, which -provides tens of thousands of data layers and maps. - -• On the client side, users can search, discover, and use the layers, maps, and apps on desktops, - -in web browsers, and on mobile devices anywhere, anytime. - -Technology evolution and trends in Web GIS -Since its inception, Web GIS has coevolved with geographic science and information technology -(see figure). This evolution and these trends are discussed in greater detail in later chapters. - -Copyright © 2022 Esri. All rights reserved. - New-generation Web GIS architecture and trends + Arab Barometer – Wave V +Topic Report – Youth + +hours or more on social media platforms, although the size of this difference is +often marginal at best. As an example, the widest generational gap is seen in +Morocco, where youth are 10 percentage points more likely to use social me- +dia. However, the difference is much smaller in Yemen (1 percentage point) and +Jordan (2 percentage points), and non-existent in Palestine. + +Though there is little generational difference in overall quantity of social media +usage, there exists a much larger generational discrepancy in terms of moti- +vations for using and trust in social media. Youth are more likely than older +generations to use social media as their primary source of information to fol- +low breaking news, and also to trust this information more than that found in +newspapers or TV news programs. With respect to the former, youth are most +likely to say social media is their primary source for breaking news in Palestine +(76 percent), while around two-thirds say the same in Morocco (67 percent) and +Jordan (63 percent). About half of youth or more in Egypt (58 percent), Tunisia +(55 percent), Iraq (51 percent), and Algeria (49 percent) rely primarily on so- +cial media for breaking news, while less than half do so in Libya (45 percent), +Lebanon (43 percent), Yemen (37 percent) and Sudan (32 percent). Notably, +youth reliance on social media to follow breaking news far exceeds that of their +older counterparts in every country surveyed. The greatest discrepancy between +youth and older generations is seen in Morocco (53 percentage points), Egypt +(44 percentage points), Tunisia (36 percentage points), and Palestine (32 per- +centage points). Elsewhere, though less stark, rates of youth reliance on social + +www.arabbarometer.org + +6 + + Arab Barometer – Wave V +Topic Report – Youth + +media for breaking news still exceed that of older citizens by at least 15 per- +centage points or more. Such trends show that while youth and their older +counterparts report using social media at similar rates overall, their incentives +and motivations for using such platforms differ in important ways. + +Youth in the region vary dramatically in the level of trust in information they re- +ceive on social media. The highest rates of trust among youth are found in Mo- +rocco (77 percent) and Egypt (74 percent), followed by Lebanon (67 percent), +Palestine (56 percent), Iraq (53 percent), and Sudan (51 percent). In contrast, +less than half of youth in Algeria (49 percent), Yemen (47 percent), Tunisia (47 +percent), Libya (41 percent), and Jordan (36 percent) trust social media sources +more than newspapers and TV news. The magnitude of the differences across +age cohorts is also varies widely, ranging from single digit differences in Jordan +(2 percentage points), Libya (5 percentage points), Yemen (3 percentage points), +and Sudan (3 percentage points) to much larger generational gaps in Lebanon +(16 percentage points), Egypt (18 percentage points), and Morocco (32 percent- +age points). Overall, however, youth are more trusting of the information they +receive from social media than are older citizens in every country surveyed. + +Notably, the most used social media platform among youth across the region is +Facebook. The exception to this finding is Yemen, where youth use WhatsApp at +a higher rate than Facebook (46 percent versus 36 percent). Youth in Egypt (94 + +www.arabbarometer.org 7 -Web GIS has coevolved with geographic information science and information technology. - -Web GIS has exemplified the following evolution of stages and trends in technologies: - -• From closed websites to open geospatial web services. Early Web GIS apps were developed -as independent websites. These websites were isolated from one another. It was difficult to -share information and functions between them and difficult to remix the content to create -apps. In the later 1990s, web services technology was conceived. Web services can be thought -of as building blocks that can be shared and remixed in versatile ways for building web apps. -The GIS industry adopted the concept of web services in manufacturing Web GIS products. -For example, ArcGIS Web GIS products fully support the web services architecture. The -server side provides ready-to-use services and allows users to create their own services, -whereas the client side can consume and combine these services to create apps. - -• From one-way to two-way information flow. Early Web GIS products and applications - -mainly supported one-way information flow, which was from server side to client side. Users -were merely the receivers of information. As we entered the 21st century, user-generated -content (UGC) became a significant phenomenon and created a reversed information flow, -from client side to server side. Volunteered geographic information (VGI) is the UGC of a -geospatial nature. For example, ArcGIS facilitates VGI through editable layers, Mobile GIS -apps, and browser-based apps. Users can view maps and perform queries, as well as conduct -field surveys, collect data, and report events they see. - -Copyright © 2022 Esri. All rights reserved. 8 - -Chapter 1: Get started with Web GIS - -• Portal technology is becoming essential. The word portal means gate or entrance. It was -adopted in the mid-1990s to form new terms such as web portal, referring to a website -that serves as the gateway to other websites or web resources. Geoportals are gateways to -geospatial information. Portals have become a core component of Web GIS technology. For -example, ArcGIS Online and Portal for ArcGIS have geoportal capabilities. They facilitate -the management, search, discovery, configuration, security, and remix of GIS data layers and -services. Today, portal collaboration allows different organizations to replicate selected con- -tent as hosts and guests, creating a portal of portals — a distributed Web GIS pattern. ArcGIS -Hub℠ and ArcGIS Enterprise Sites are examples of subject- or initiative-based portals. They -provide an easy-to-configure community engagement platform that organizes people, data, -and tools through information-driven initiatives. Organizations can create websites to share -data, documents, videos, and web maps on a certain subject or for a certain initiative. For -instance, the COVID-19 GIS Hub (coronavirus-resources .esri .com) provides GIS resources, -solutions, and a collaboration platform for the world to combat the coronavirus pandemic. -• Cloud GIS delivers GIS functionality and ready-to-use contents. Cloud GIS, which uses -public and private cloud computing to provide GIS software and contents, has become the -primary way to deliver GIS — not just functionality but also content. Cloud GIS advantages -include vast content and functionality, the flexibility of pay-as-you-go or subscription pric- -ing, reduced complexity, and increased availability of services. Because of these advantages, -cloud GIS, such as ArcGIS Online, has penetrated organizations that have not used GIS -before or been able to afford GIS on their own. - -• Mobile is becoming the pervasive Web GIS client platform. In the post-PC era, mobile - -devices have surpassed desktops and notebooks as the primary platform for accessing online -information. Mobile devices are a part of everyone’s life and work. “Mobile first” is a strategy -of many industries, including the Web GIS industry. Vendors have given extra attention to -Mobile GIS. For example, Esri provides numerous mobile-native apps and mobile-friendly -browser apps to support the needs of users and organizations for Mobile GIS. Mobile GIS -is also associated with many frontiers in Web GIS, such as augmented reality (AR). AR can -superimpose GIS data on top of a user’s camera views and thus augment a user’s sense of -reality. With the rapid advances in Mobile GIS, the vision of using GIS for anything, any- -time, anywhere, and by anyone is being realized faster than we can imagine. - -• Map visualization goes from 2D to 3D and virtual reality. With the increased client-side -graphics processing power and the broader support of WebGL (a JavaScript application pro- -gramming interface [API] that renders interactive 2D and 3D graphics in a browser), Web -GIS products such as ArcGIS Online and ArcGIS Enterprise can smoothly create and display - -Copyright © 2022 Esri. All rights reserved. New-generation Web GIS architecture and trends + Arab Barometer – Wave V +Topic Report – Youth + +percent), Algeria (91 percent), Morocco (90 percent), and Lebanon (90 percent) +are most partial to Facebook. More than eight-in-ten youth in Sudan (85 per- +cent), Libya (82 percent), and Jordan (81 percent), and more than seven-in-ten +youth in Iraq (74 percent) and Palestine (73 percent) use Facebook. + +Education and Employment + +Education System + +Rates of youth satisfaction with state education systems, both overall and as they +compare to older generations, vary greatly throughout the region. Nearly two- +thirds of youth in Palestine (65 percent) and Jordan (65 percent), and over half +of youth in Yemen (54 percent) are satisfied or completely satisfied. However, +fewer than half are satisfied in Sudan (43 percent) and Lebanon (39 percent), +and roughly a third are satisfied in Algeria (34 percent), Libya (31 percent), Egypt +(30 percent), and Tunisia (30 percent). The lowest rates of satisfaction are seen +among Iraqi youth (24 percent) and Moroccan youth (17 percent). Given this +wide variation, there seems to be limited regional consensus about the state +of education. However, one potential regional trend is that across most of the +countries surveyed youth are less satisfied with the education system than are +their older counterparts. +Indeed, in the majority of countries this discrepancy +is small, as seen in Palestine (1 percentage point), Sudan (2 percentage points), +Libya (2 percentage points), Algeria (3 percentage points), Yemen (3 percent- +age points), Iraq (4 percentage points), and Lebanon (5 percentage points). Yet +elsewhere, for instance in Morocco and Egypt, the intergenerational gap is more +sizable (22 percentage points and 10 percentage points, respectively). Excep- +tions to this trend are found only among Tunisian and Jordanian youth, both of +whom voice satisfaction at higher rates than do older citizens (30 percent versus +28 percent and 65 percent versus 59 percent, respectively). + +www.arabbarometer.org + +8 + + Arab Barometer – Wave V +Topic Report – Youth + +Looking to perceived corruption in the education system as one potential rea- +son contributing to low levels of satisfaction, in many countries more than half of +youth believe that rashwa (bribes) is required in order to receive better access to +education services. This belief is most commonly held in Egypt (64 percent), fol- +lowed by Lebanon (62 percent), Sudan (55 percent), Morocco (54 percent), Alge- +ria (53 percent), Yemen (53 percent), Iraq (52 percent), and Tunisia (50 percent). +However, while many youth believe that paying rashwa is necessary, this trend +is not common to all MENA youth as no more than a third of youth in Palestine +(33 percent), Libya (32 percent), and Jordan (29 percent) say the same regard- +ing rashwa to access better educational services. Prevalence of rashwa might +contribute to some dissatisfaction with state education systems, but it does not +account for or explain it in its entirety. + +Employment Opportunities + +Similar to satisfaction rates with state education systems, levels of youth satis- +faction with the government’s performance on creating employment opportu- +nities vary. However, despite this variation, across the region youth approval is +overwhelmingly weak, ranging from a low of five percent in Lebanon to a high +In addition to Lebanon, single-digit approval of +of only 32 percent in Yemen. +government performance on creating employment opportunities is seen in Iraq +(6 percent) and Algeria (9 percent). Fewer than two-in-ten youth in Morocco +(13 percent), Palestine (14 percent), Jordan (15 percent), Tunisia (16 percent), + +www.arabbarometer.org 9 -thematic and photo-realistic 3D web scenes. Representing a big step from 2D online maps, -3D web scenes provide users a more intuitive means to understand their data. Web scenes -are also important for indoor mapping, an ongoing trend in GIS. Even more intuitive than -3D, virtual reality (VR) allows users with certain visual wearables to immerse themselves in -3D city models by transporting themselves into static viewpoints and comparing different -urban design scenarios. This immersive experience brings GIS data and geospatial under- -standing even closer to users. The GIS community can play an important role in building the -metaverse and can assist with getting real work done in the virtual world environment. The -term metaverse refers to the fully immersive 3D expansion of existing internet technologies -to include AR and VR. - -• Data source goes from static to real time and spatiotemporal big data. Many elements in - -Web GIS are of a real-time nature, such as the incidents immediately reported by field crews, -citizens using mobile devices, concurrent measurements from sensor networks, and smart -cities. This massive amount of data presents challenges in real-time intake, processing, analy- -sis, visualization, and storage. ArcGIS GeoEvent™ Server and ArcGIS Velocity use cluster -computing and can ingest millions of sensor readings per second, process them, and store -them in real time. Such products and research allow Web GIS to meet the requirements of -the sensor web and the Internet of Things (IoT). - -• Web GIS is becoming more intelligent. Map visualization is the first step toward data -analysis. ArcGIS smart mapping can analyze the data automatically and suggest the best -mapping style and best defaults. This can help both expert and novice users quickly create -aesthetic and informative maps. Web GIS goes far beyond mapping. Web GIS can perform -big data analysis using distributed computing, aggregate data in the context of both space -and time, extrapolate new ideas from raw data, and bring superior intelligence to business -decisions. More recently, Web GIS has started to use artificial intelligence (AI), especially -machine learning and deep learning. For example, AI significantly improves the accuracy of -online imagery classification. AI can quickly identify the damaged locations from millions of -facilities’ photos and ensure that the damaged facilities are repaired early. - -Copyright © 2022 Esri. All rights reserved. 10 - -Chapter 1: Get started with Web GIS - -Web GIS information model: Users, web layers, maps, and - -apps - -Web GIS elements include users, groups, content items, and metadata (see figure). - -• Users with necessary privileges can create and join groups. -• Users sign in to create and share content items (with certain privileges), which include a - -large variety of data, web layers, tools, web maps, web scenes, and web apps. - -• Content items have metadata, including title, summary, description, categories, and tags, and -additional standards-based metadata. Metadata is indexed so users can search and discover -items more efficiently. - -• Users can keep information to themselves, share with certain groups (rather than individual -users), share with their organizations, or share with everyone — even the public if the organi- -zation allows anonymous access. This allows other users to see and access the items. - -own or join - -Users - -o -w -n - -organize and secure - -Groups - -x -e -d -n - -i - -index - -Items - -Metadata - -The ArcGIS Online and Portal for ArcGIS high-level information model. - -Main types of content items -Web GIS has five main types of content: data, layers, tools, web maps and scenes, and apps. Data -can be used to publish web layers; web layers can be used to create web maps and web scenes; -web maps and scenes and web tools can be used in apps or to create apps, as shown in the figure. - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +Sudan (17 percent), and Egypt (18 percent), and less than a quarter of youth in +Libya (22 percent) approve. + +Moreover, in addition to widespread dissatisfaction with government efforts to +improve employment opportunities, the vast majority of young Arab citizens +also believe that obtaining employment requires having strong connections. +Roughly nine-in-ten youth in Iraq (95 percent), Lebanon (94 percent), Jordan +(94 percent), Tunisia (94 percent), Libya (91 percent), Algeria (91 percent), Sudan +(90 percent), Palestine (90 percent), Egypt (90 percent), and Morocco (88 per- +cent) say that based on recent experience they think that obtaining employment +through wasta (personal connections) happens often or sometimes. Though +comparatively fewer, nearly seven-in-ten Yemeni youth (69 percent) hold the +same perception regarding the connection between wasta and employment. + +Emigration + +Given broad youth dissatisfaction with conditions in their countries, it is perhaps +not surprising that youth across the MENA region have a strong desire to em- +igrate. In six of eleven countries surveyed, roughly half or more of youth want +to leave their homeland. With the exception of Libya, North African and Jorda- +nian youth are most eager to migrate with about half or more citing a desire +to leave. Youth in Morocco cite not only the greatest desire to migrate, but also + +www.arabbarometer.org + +10 + + Arab Barometer – Wave V +Topic Report – Youth + +show the greatest discrepancy with older generation’s desire to do so (70 per- +cent versus 25 percent). Meanwhile, over half of youth in Sudan (62 percent), +Jordan (59 percent), Tunisia (56 percent) and Algeria (56 percent) also wish to +leave their country. Roughly a third of Libyan (31 percent), Yemeni (33 percent), +and Lebanese (35 percent) youth say the same. + +Regional trends are evident with regard to which countries youth are most ea- +ger to migrate. Over two-thirds of youth in French-speaking North Africa want +to migrate to a European country. This is not surprising given both the proximity +of their countries to Europe and, perhaps more importantly, the close linkages +between France and these countries. In contrast, six-in-ten youth in Egypt, and +roughly half of youth in Yemen (48 percent) and Sudan (45 percent), say they +want to emigrate to a GCC country, which is in line with historic patterns of labor +migration to this area. Interestingly, fewer than half of youth across the region +hope to migrate to North America. Jordanian youth are most likely to cite a +desire to migrate to North America (42 percent), whereas Yemeni, Libyan, and +Tunisian youth are least likely (12 percent, 13 percent, and 13 percent). + +The least preferred destination for youth is non-GCC MENA countries. Fewer than +a third of youth across the region have considered emigrating to such countries. +Lebanese and Tunisian youth are least eager (4 percent and 5 percent, respec- +tively) while Yemeni youth are most eager, though still fewer than three-in-ten +Yemeni youth cite an interest in emigration to a non-GCC MENA country (27 + +www.arabbarometer.org 11 -Tools - -configured -and used in - -published to - -added to - -Web, mobile, -and desktop apps - -Data - -Web layers - -Web maps -and scenes - -The main types of content in Web GIS and their relationships. - -These five main content types are described as follows: - -• Data: ArcGIS Online supports data in a variety of formats, including CSV, shapefiles, GPS - -Exchange Format (GPX), JavaScript Object Notation (GeoJSON), photos, imagery, geodata- -bases, and so on. - -• Layers: The Esri geospatial cloud can host layers, including feature layers, tiled layers, vector -tiles, map image layers, image layers, scene layers, CSV layers, tables, and Open Geospatial -Consortium (OGC) standard layers such as Keyhole Markup Language (KML), Web Map -Service (WMS), Web Map Tile Service (WMTS), and Web Feature Service (WFS). - -• Web maps and scenes: A web map, which is 2D, or a scene, which is 3D, comprises one or - -multiple layers and allows sophisticated layer configuration, including styles, pop-ups, labels, -and access permission. - -• Tools: Tools perform analytical functions, such as geocoding, routing, generating PDF files, - -summarizing data, finding hot spots, and analyzing proximity. - -• Apps: GIS apps are lightweight mapcentric programs that can be used on mobile devices, -in web browsers, or on desktops. Developers can program with ArcGIS web APIs to build -web apps. However, you do not have to be a developer to create a web app. ArcGIS provides -many ready-to-use web apps and mobile apps that can be used without programming. - -Apps: The face of Web GIS -Apps bring Web GIS to life. Web GIS end users directly interact with apps. The back-end data, -web layers, and server computation power of Web GIS become live and useful through apps. -ArcGIS provides ready-to-use apps out of the box and provides APIs and software development -kits (SDKs) for developers to create custom apps. - -Copyright © 2022 Esri. All rights reserved. 12 - -Chapter 1: Get started with Web GIS - -The ready-to-use apps include configurable apps (also referred to as ArcGIS Online and Portal -for ArcGIS web app templates), ArcGIS StoryMaps℠ stories, mobile apps, app builders, and -more. ArcGIS organizes these apps into bundles: - -• Essential apps bundle: ArcGIS StoryMaps, ArcGIS Experience Builder, ArcGIS Web -AppBuilder, ArcGIS Dashboards, instant apps, ArcGIS Hub Basic, Map Viewer, Scene -Viewer, and so on - -• Field apps bundle: ArcGIS Field Maps, ArcGIS Survey123, ArcGIS QuickCapture, and so on -• Office apps bundle: ArcGIS for Office, ArcGIS for SharePoint, and so on -• ArcGIS Pro -• ArcGIS Insights℠ -• Add-on apps: ArcGIS Drone2Map®, ArcGIS Navigator, ArcGIS Business Analyst™, and so on - -ArcGIS user types consist of Viewer, Storyteller, Editor, Field Worker, Creator, GIS -Professional, and Insights Analyst (see table 1.3). Different user types have included access to -different app bundles and apps. Apps that are not included with a user type can be acquired and -assigned as add-on licenses. To finish most of the tutorials in this book, a Creator user type with -a Publisher role is required. - -Table 1 .3 . Main ArcGIS user types, privileges, and included apps - -Type - -Viewer - -Storyteller - -Editor - -Privileges - -Can view items -Can’t create, edit, share, or analyze items - -Can create stories and express maps -Can’t create other types of content - -Can view and edit data -Can’t analyze, create, or share items - -Included apps - -Essential apps bundle - -Essential apps bundle - -Essential apps bundle - -Field Worker - -Same as Editor - -+ Field apps bundles - -Creator - -Can edit, create, share content, and do analysis - -+ Office apps bundles - -GIS Professional - -Same as Creator - -Insights Analyst - -Same as GIS Professional - -+ ArcGIS Pro - -Insights - -“+” means including the apps from the user type listed earlier in the table . - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +percent). + +Perceived Civil Liberties + +Of the countries surveyed by the Arab Barometer, Tunisia is the only one cate- +gorized as free by Freedom House, with the remainder of countries identified as +partly free or not free. As such, it is unsurprising that youth in Tunisia are more +confident in their civil liberties like freedom of expression and right to peaceful +protest than are youth in other surveyed countries. However, despite being com- +paratively more confident in the protection of their civil liberties, Tunisian youth +are still quite similar to youth in the rest of the region who report generally low, +though varied, confidence in both freedom of speech and right to peacefully +protest. + +Looking first to freedom of speech, Tunisian youth are indeed most confident in +their freedom of expression with over six-in-ten saying that it is guaranteed to a +great or medium extent in the country (63 percent). More than half of youth in +Jordan (56 percent) and Lebanon (51 percent) say the same. In the rest of the +region fewer than half of respondents feel their freedom of expression is guar- +anteed. About half of youth in Yemen (49 percent), roughly four-in-ten youth +in Palestine (42 percent), Egypt (40 percent), and Morocco (39 percent), and +roughly a third of youth in Algeria (35 percent) and Iraq (33 percent) say that + +www.arabbarometer.org + +12 + + Arab Barometer – Wave V +Topic Report – Youth + +freedom of expression is guaranteed to a great or medium extent. Least confi- +dent of their freedom of expression are youth in Sudan (25 percent) and Libya +(24 percent). + +When comparing youth perceptions of their freedom of speech to the percep- +tions of older age cohorts, there is no clear regional trend. For instance, youth +in Lebanon (+10 points), Jordan (+3 points), Yemen (+4 points) and Palestine +(+4 points) are more confident in their freedom of expression than are older +citizens. Yet, the opposite trend is apparent in Morocco (-16 points), Egypt (-5 +points), and Tunisia (-4 points). +In the remainder of countries surveyed (Libya, +Sudan, Iraq, and Algeria) youth and older citizens do not diverge by more than +a point and rate their freedom of expression similarly. + +Youth perception of their freedom to participate in peaceful protests is also low +across the region, never exceeding half of youth in any country. Though nearly +half of youth in Tunisia (49 percent), Lebanon (49 percent), Jordan (47 percent), +and Yemen (46 percent) feel their freedom to peacefully demonstrate is guaran- +teed to a medium or great extent, fewer youth hold this belief elsewhere in the +region. Four-in-ten Palestinian youth, a third of Moroccan youth, and three-in- +ten Iraqi youth are confident of their freedom to participate in peaceful demon- +strations. Meanwhile, roughly a quarter of youth in Algeria (27 percent), Libya +(24 percent), and Egypt (24 percent) and fewer than two-in-ten in Sudan (19 +percent) say the same. + +www.arabbarometer.org 13 -Basic components of a Web GIS app -Today’s best practices recommend that a GIS app have basemaps, operational layers, and tools -(see figure). ArcGIS supports this practice and allows you to create Web GIS apps. - -+ - -+ - -= - -Basemaps - -Operational -layers - -Tools - -Web GIS app - -The basic components of a Web GIS app. - -The components of a Web GIS app are described as follows: - -• Basemap layers: Basemaps provide a reference or context for your app. ArcGIS provides a -collection of fast-responding 2D and 3D basemaps (see figure). In addition, you can create -and use your own basemaps. - -ArcGIS provides a gallery of basemaps and an elevation service to support 2D and 3D web maps and apps. - -• Operational layers: Operational layers are theme layers that you and other users can use and -interact with. You can use existing layers from ArcGIS Online and ArcGIS Living Atlas of -the World as your operational layers. These layers span a range of subjects and can support - -Copyright © 2022 Esri. All rights reserved. 14 - -Chapter 1: Get started with Web GIS - -maps and apps of almost every subject. You can search these collections and discover layers -that fit your needs. You can also create and use your own operational layers. - -• Tools: Tools perform tasks beyond mapping, including common tasks such as query, geocod- -ing, routing, and more specialized tasks. ArcGIS Online provides extensive spatial analysis -tools for you to ask questions and solve spatial problems. ArcGIS Enterprise provides tools -for standard analysis, big data analysis, and raster analysis, and allows you to publish custom -web tools — in other words, geoprocessing services. - -Web layers or services: The building blocks of Web GIS -Web services technology is at the foundation of today’s Web GIS. The previously mentioned -three types of components of Web GIS apps are typically all based on web services. To make it -easier to understand, web services in ArcGIS are also called web layers or web tools. For exam- -ple, basemaps are often tile services or tile layers, operational layers are often feature services or -feature layers, and many tools are based on geoprocessing services or web tools. - -Early Web GIS was not based on web services; instead, it consisted mostly of stand-alone -websites in which GIS data and functions were available only to their own clients and couldn’t -be reused in other systems. This situation greatly limited the reusability of Web GIS resources. -In the late 1990s, web services technology was conceived. A web service is a program that runs -on a web server and provides programming interfaces for clients to consume over the web. Web -services have many advantages, such as flexibility in being reused and remixed in many web apps. -The GIS industry quickly adopted web services technology and reformed Web GIS products -based on web services architecture in the early 2000s. Today’s Web GIS products are designed to -support the publication, discovery, and use of GIS web services. A GIS service represents a GIS -resource — such as a layer, map, locator, or toolbox — that is located on the server and made avail- -able to web clients. For example, ArcGIS Living Atlas provides collections of ready-to-use GIS -services. The Esri geospatial cloud allows publishers to publish many types of GIS services and -allows users to discover and use GIS services in web maps, web apps, and mobile apps. - -This chapter introduces feature layers and services. Other chapters introduce raster tile ser- -vices, vector tile services, 3D scene services, map services, image services, stream services, geo- -processing services, geocoding services, and more. To support interoperability, ArcGIS supports -OGC standards such as WMS, WMTS, WFS, and KML. - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +Attitudes on Gender + +Arab youth are divided in their views on gender equality. On the one hand, +most are supportive of equal rights for women, with the exception of inheri- +tance. However, there is substantially less support for women enjoying equal +roles in society. This is particularly evident in the political sphere. + +Throughout the region fewer than a third of youth say that a university education +is more important for men than for women. Support for university education +for men compared with women is highest in Yemen and Sudan (27 percent, +each), followed by roughly two-in-ten youth in Jordan (23 percent), Egypt (22 +percent), Algeria (20 percent), Iraq (19 percent), Tunisia (17 percent) and Libya +(17 percent). Even lower rates of support for men’s education taking precedence +over that of women are found in Palestine (14 percent), Lebanon (13 percent), +and Morocco (12 percent). + +Youth also generally hold liberal views on the question of whether women should +have equal rights in the decision to get a divorce. The exception to this trend +is in Sudan, where support for women’s equal rights in deciding to divorce falls +below five-in-ten (46 percent). Support is highest in Lebanon (87 percent), fol- +lowed by Iraq (81 percent), Jordan (80 percent), Tunisia (78 percent), Morocco +(78 percent), Palestine (76 percent), Libya (75 percent) and Algeria (72 percent). +Though still greater than half, lower rates of support are seen in Egypt (64 per- + +www.arabbarometer.org + +14 + + Arab Barometer – Wave V +Topic Report – Youth + +cent) and Yemen (55 percent). Notably, in most countries surveyed, the views +held by youth about women’s right to divorce are not significantly different from +the views held by older citizens. + +In a departure from predominantly liberal views on women’s rights, surprisingly +few youth in the region agree that women’s share of inheritance should be equal +to that of men. Nonetheless, Lebanon is a clear regional outlier to this trend with +over six-in-ten youth saying they agree or strongly agree that inheritance should +be equal between the sexes (62 percent). In contrast, far fewer youth elsewhere +are as supportive. Roughly three-in-ten youth in Iraq (30 percent) and Egypt (28 +percent), and around one-fourth of youth in Morocco (26 percent) and Tunisia +(24 percent) support equal inheritance. Still fewer youth in Algeria (18 percent), +Jordan (18 percent), Libya (16 percent), Sudan (13 percent), and Palestine (8 +percent) say the same. + +A plurality of youth in the region also believe that husbands should have the +final say in all family decisions. Support for this sentiment is greatest in Sudan +(73 percent), Algeria (73 percent), and Iraq (70 percent), followed by Egypt (66 +percent) and Libya (66 percent). Youth across Tunisia (56 percent), Yemen (55 +percent) and the Levant hold similar views to each other regarding women and +family decision-making. +In Lebanon and Jordan 55 percent of youth believe +husbands should have the final say in all family decisions, while in Palestine 50 +percent say the same. Though Moroccan youth are least likely to support limiting + +www.arabbarometer.org 15 -Hosted and nonhosted feature layers -Feature layers are the most used layer type for operational layers. Feature refers to vector data: -points, lines, and polygons. Feature layers can be hosted or nonhosted, depending on where the -underlying data is stored. - -• Hosted refers to a situation in which a web service is based on ArcGIS-managed data: for -instance, the data is stored in the ArcGIS Online cloud database (which is managed auto- -matically by Esri) or in ArcGIS Data Store (which is managed automatically by ArcGIS -Enterprise). - -• Nonhosted refers to a situation in which a web service connects directly to user-managed - -data: for instance, the data is stored in your organization’s own database, and the database is -managed by your own organization. - -Feature layers published to ArcGIS Online are always hosted feature layers. Feature layers pub- -lished in ArcGIS Enterprise can be hosted if the data is stored in ArcGIS Data Store or nonhosted -if your feature layers reference your organization’s own geodatabase — for example, SQL Server, -Oracle, or PostgreSQL. - -Hosted feature layers are often easier to publish because you don’t have to worry about the -underlying database. You don’t have to (and you won’t be able to) directly connect to the database -and manipulate it. To change the database schema and edit the data, you must go through the -hosted feature layer web interface. Nonhosted feature layers require that you have and manage -your own database. You can directly connect to the database and perform various operations -on it. - -This book teaches many ways to publish hosted feature layers—from CSV files, zipped file -geodatabases, ArcGIS Pro, and so on. You can also create empty feature layers and add fields in -ArcGIS Online (see chapter 4). - -Web maps and web scenes: The bridge between layers and apps -Web GIS provides Map Viewer and Scene Viewer for creators to integrate multiple layers, style -these layers, configure layer pop-ups, and make many other configurations. After a web map or -web scene is created, it can be used in one app or reused in many apps. This makes it easier for -creators to create apps. Without web maps and scenes, creators would often have to integrate the -layers directly and configure them in the apps, which can be cumbersome. - -Copyright © 2022 Esri. All rights reserved. 16 - -Chapter 1: Get started with Web GIS - -Paths to building Web GIS applications -The tutorials in this book teach readers how to build Web GIS apps. ArcGIS offers many paths to -this goal, as shown in this figure, which will be used throughout the book. - -ArcGIS offers many ways to build web apps. The green arrows in the figure highlight the workflows and -technology presented in this chapter. - -The figure presents the generic workflow to build Web GIS apps: - -• The data tier contains formats that range from simple CSV files managed with Microsoft - -Excel to sophisticated geodatabases managed with enterprise databases. - -• You can publish desktop resources to ArcGIS Online or ArcGIS Enterprise as web layers, - -tools, or other services. You can also use layers provided by ArcGIS Living Atlas. You can add -layers to web maps and web scenes and configure the styles and pop-ups in these layers. - -• On the client side, you can use various ready-to-use apps or custom apps. The former can be -configured without programming. The latter can be developed using various web APIs or -SDKs. - -Copyright © 2022 Esri. All rights reserved. \ No newline at end of file + Arab Barometer – Wave V +Topic Report – Youth + +women’s roles in family decision-making, nonetheless nearly four-in-ten say that +husbands should have the final say in all family decisions (38 percent). Notably, +though youth in Morocco are far less likely than older generations to support +this notion (-15), youth in Jordan (+6), Lebanon (+7), Libya (+5), and Tunisia (+2) +are more supportive than are their older counterparts. + +In every country surveyed, with the exception of Morocco, over half of youth agree +that, in general, men are better at political leadership than women. Roughly +eight-in-ten youth in Sudan (80 percent) and Jordan (78 percent), and roughly +three-quarters of youth in Algeria (74 percent) hold this belief. Across much of +the rest of the region youth hold this view at similar rates. For instance, around +seven-in-ten youth say men are better at political leadership in Egypt (70 per- +cent), Iraq (69 percent), Yemen (68 percent), Palestine (68 percent), and Libya +(67 percent). Meanwhile, fewer in Tunisia, Lebanon, and Morocco say that men +are better at political leadership than women (52 percent, 51 percent, and 31 +percent, respectively). + +Interestingly, despite unfavorably comparing female political leaders to men and +criticizing women’s role as decision-makers in the family, Arab youth are often +open both to the idea of a woman being a leader of a Muslim country and to +setting aside a certain percentage of elected positions for women as a means +of attaining fairer political representation. More than half of youth support the +notion of a female president or prime minister of a Muslim nation in all coun- + +www.arabbarometer.org + +16 + + Arab Barometer – Wave V +Topic Report – Youth + +tries surveyed, excluding Algeria (37 percent). Roughly half of youth in Sudan +(51 percent) and Libya (54 percent), six-in-ten youth in Yemen (57 percent), Jor- +dan (60 percent), and Palestine (64 percent) and roughly seven-in-ten youth in +Tunisia (67 percent), Egypt (68 percent) and Iraq (69 percent) affirm that women +can become prime ministers or presidents of Muslim countries. Moroccan and +Lebanese youth (77 percent and 79 percent, respectively) are the most accept- +ing of this right. + +The majority of MENA youth also believe that in order to achieve fairer represen- +tation a certain percentage of elected positions should be set aside for women. +Support for this right exceeds half of all youth in all countries surveyed. Youth in +Iraq and Lebanon (79 percent, each) are most supportive, followed by Sudan (76 +percent), Palestine (75 percent), Jordan (72 percent) and Yemen (69 percent). +Excluding Sudan, North African youth show the least support for women’s quo- +tas in elected positions with less than two-thirds of youth in Egypt (66 percent), +Libya (65 percent), Tunisia (63 percent), Morocco (62 percent) and Algeria (53 +percent) saying they agree or strongly agree with this idea. + +International Relations + +In over half of the countries surveyed, youth are most supportive of stronger ties +with China and least interested in stronger ties with the United States among +global powers. Youth in Jordan (70 percent), Tunisia (65 percent), Sudan (62 +percent), Libya (61 percent), Yemen (50 percent), Palestine (49 percent), and +Iraq (47 percent) prefer strengthening ties with China at a higher rate than they +do for ties with the United States, the United Kingdom, or Russia. Exceptions to +this include youth in Morocco, who have a slightly stronger preference for closer +ties with the United Kingdom (63 percent), youth in Lebanon and Algeria, who +are more supportive by a small margin of stronger ties with Russia (38 percent +and 42 percent, respectively), and Egyptian youth, who are equally supportive +of strengthening ties with the United States and Russia (41 percent, each). + +In a majority of countries surveyed youth support for stronger ties with the United +States not only falls below 50 percent, but is also lower than youth support for +strengthening ties with other world powers. Exceptions to the former are Sudan +(60 percent), Morocco (57 percent), and Jordan (55 percent) where youth sup- +port exceeds half. In contrast, fewer than half of youth in Tunisia (47 percent), +and roughly a third of youth in Libya (36 percent), Iraq (34 percent), Palestine, +Lebanon, and Algeria (31 percent, each) say that relations with the United States +should become stronger. Of all countries surveyed Yemeni youth show not only +the least interest in closer ties with the United States (22 percent), but they are +also least interested in strengthening ties with Russia (26 percent) relative to +youth in other countries. + +www.arabbarometer.org + +17 + + Arab Barometer – Wave V +Topic Report – Youth + +Looking to regional powers, one clear trend is that Arab youth are generally less +interested in stronger relations with Iran and more interested in improving re- +lations with Turkey. Though support for stronger ties with Iran never exceeds +four-in-ten youth in any country surveyed, support is highest in Lebanon (38 per- +cent), followed by Sudan (36 percent), Palestine (34 percent), Iraq (32 percent) +and Tunisia (31 percent). In Lebanon and Iraq, sectarian divides might explain +the comparatively greater support for strengthening ties with Iran, a country +with a strong Shi’ite majority. Roughly two-in-ten youth in Morocco (22 per- +cent), Jordan (21 percent), Yemen (20 percent), Algeria (20 percent), and Libya +(18 percent) prefer stronger relations with Iran, while only 8 percent of Egyptian +youth say the same. + +www.arabbarometer.org + +18 + + Arab Barometer – Wave V +Topic Report – Youth + +In comparison to other regional powers (Saudi Arabia, Qatar, and Iran), Arab +youth tend to prefer strengthening relations with Turkey. Youth in Jordan (82 +percent), Palestine (75 percent), Tunisia (64 percent), Yemen (61 percent), Mo- +rocco (56 percent), Algeria (53 percent), and Iraq (45 percent), are all more likely +to want improved relations with Turkey than to do so for relations with Saudi Ara- +bia, Qatar, or Iran. Youth in Sudan prefer strengthening relations with Saudi Ara- +bia, Turkey, and Qatar at similar rates (70 percent, 73 percent, and 74 percent, +respectively), as do youth in Lebanon (41 percent, 41 percent, and 42 percent, +respectively). In Libya and Egypt youth by far prefer improving ties with Saudi +Arabia (55 percent and 58 percent). + +www.arabbarometer.org + +19 + + About Arab Barometer + +The Arab Barometer is a nonpartisan research network that +provides insights into the social, political, and economic attitudes +and values of ordinary citizens across the Arab world. + +We have been conducting rigorous, and nationally representative +face-to-face public opinion surveys on probability samples of the +adult populations across the Arab world since 2006. The margin of +error is ±3 percent. + +The Arab Barometer is the largest repository of publicly available +data on the views of men and women in the MENA region. Our +findings give a voice to the needs and concerns of Arab publics. + +.......... + +.......... + +.......... + +ARABBAROMETER.ORGARABBAROMETER@ARABBAROMETER \ No newline at end of file diff --git a/q159/random_k2/random_2.md b/q159/random_k2/random_2.md index 482f9de9b0ff67f242f61e09321d1c78cb243915..6748665a7c835db276023c66e67f49b0fa9cb861 100644 --- a/q159/random_k2/random_2.md +++ b/q159/random_k2/random_2.md @@ -1,410 +1,493 @@ -IEEEInternationalConferenceonDataMining(ICDM),WorkshoponUserModelingandRecommendation(UMRec),Washington,DC,USA,2025 -| Effectiveness | | | | | of LLMs | | in | Temporal | | | User | | Profiling | | | -| ------------- | --- | --- | ------ | -------------------- | ------- | -------------- | --------------------- | ----------- | ----------- | ---------------------- | --------- | --- | --------- | --- | --- | -| | | | | | for | Recommendation | | | | | | | | | | -| | | | Milad | Sabouri∗, | | Masoud | Mansoury†, | Kun | Lin∗, | Bamshad | Mobasher∗ | | | | | -| | | | | | | | ∗DePaul University, | | USA | | | | | | | -| | | | Email: | msabouri@depaul.edu, | | | klin13@depaul.edu, | | | mobasher@cs.depaul.edu | | | | | | -| | | | | | †Delft | University | of | Technology, | Netherlands | | | | | | | -| | | | | | | Email: | m.mansoury@tudelft.nl | | | | | | | | | -Abstract—Effectively modeling the dynamic nature of user 17% in Recall@10 and 14% in NDCG@10, benefits are -| preferences | is crucial | | for enhancing | | recommendation | | accuracy | | | | | | | | | -| ----------- | ---------- | --- | ------------- | --- | -------------- | --- | -------- | ---- | ---------- | --- | ------- | ------------ | --- | ---------- | ------ | -| | | | | | | | | less | pronounced | in | sparser | environments | | like Video | Games. | -5202 tcO 13 ]RI.sc[ 1v67100.1152:viXra andfosteringtransparencyinrecommendersystems.Traditional -| | | | | | | | | Our | analysis | of these | results | leads | us | to hypothesize | that | -| -------------- | ----- | --------- | --- | --------------- | --- | ------- | ---------- | ---- | -------- | ---------- | ------- | ------------ | ------- | -------------- | ------ | -| user profiling | often | overlooks | | the distinction | | between | transitory | | | | | | | | | -| | | | | | | | | this | nuanced | capability | is | particularly | evident | where | short- | -short-terminterestsandstablelong-termpreferences.Thispaper -examines the capability of leveraging Large Language Models term and long-term preferences are more clearly separable -(LLMs) to capture these temporal dynamics, generating richer (e.g., Movies&TV), versus domains with more stable user -| user representations | | through | | distinct | short-term | and | long-term | | | | | | | | | -| -------------------- | --- | ------- | --- | -------- | ---------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -profiles(e.g.,VideoGames).Thishighlightsacriticaltrade-off -textualsummariesofinteractionhistories.Ourobservationssug- -betweenLLM-enhancedpredictivepowerandassociatedcom- -gestthatwhileLLMstendtoimproverecommendationqualityin -| | | | | | | | | putational | | costs, suggesting | | context-dependent | | application. | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ----------------- | --- | ----------------- | --- | ------------ | --- | -domainswithmoreactiveuserengagement,theirbenefitsappear -less pronounced in sparser environments. This disparity likely Also,beyondpredictiveaccuracy,thisLLM-drivenapproach -stemsfromthevaryingdistinguishabilityofshort-termandlong- inherently offers an intrinsic potential for transparency. The -| term preferences | | across | domains; | the | approach | shows | greater | | | | | | | | | -| ---------------- | --- | ------ | -------- | --- | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -naturallanguageprofilesandlearnedattentionweightsprovide -utilitywherethesetemporalinterestsaremoreclearlyseparable -| | | | | | | | | a pathway | | toward interpretability, | | | conveying | whether | recom- | -| ----------------- | ------------ | -------- | --- | ---------- | ---------- | ---------- | ----------- | ---------- | --- | ------------------------ | ------ | --- | --------- | ---------- | ------ | -| (e.g., Movies&TV) | | compared | | to domains | with | more | stable user | | | | | | | | | -| | | | | | | | | mendations | | stem from | recent | or | enduring | interests. | While | -| profiles | (e.g., Video | Games). | | This | highlights | a critical | trade- | | | | | | | | | -off between enhanced performance and computational costs, fully realizing user-facing explanations is a future direction, -suggesting context-dependent LLM application. Beyond predic- this work contributes insights into the practical capability -| tive capability, | this | LLM-driven | | approach | inherently | | provides | | | | | | | | | -| ---------------- | --------- | ---------- | -------------------- | -------- | ---------- | ---- | ----------- | ----------- | --------- | ---------------- | -------- | ------------- | ---------- | ------------ | ---- | -| | | | | | | | | and | inherent | interpretability | | of LLM-driven | | temporal | user | -| an intrinsic | potential | | for interpretability | | through | | its natural | | | | | | | | | -| | | | | | | | | profiling, | outlining | new | research | | directions | for adaptive | and | -| language | profiles | and | attention | weights. | This | work | contributes | | | | | | | | | -| | | | | | | | | transparent | | recommender | systems. | | | | | -insightsintothepracticalcapabilityandinherentinterpretability -| of LLM-driven | | temporal | user | profiling, | outlining | new | research | | | | | | | | | -| ------------- | --- | -------- | ---- | ---------- | --------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -II. RELATEDWORK -directionsfordevelopingadaptiveandtransparentrecommender -systems. Modeling the dynamic nature of user preferences is a -IndexTerms—TemporalUserProfiling,LargeLanguageMod- -| | | | | | | | | long-standing | | challenge | in | recommender | | systems. | Traditional | -| --- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --------- | --- | ----------- | --- | -------- | ----------- | -els, User Modeling -| | | | | | | | | methods, | such | as averaging | | item embeddings | | [3], | often fail to | -| --- | --- | --- | ------------ | --- | --- | --- | --- | -------- | ---------- | ------------ | ----------- | --------------- | --- | ------------ | ------------- | -| | | | | | | | | capture | the | evolving | distinction | between | | short-term | and long- | -| | | I. | INTRODUCTION | | | | | | | | | | | | | -| | | | | | | | | term | interests. | More | advanced | approaches | | for temporal | and | -Effective recommender systems demand both accurate and sequential recommendation include time-aware matrix factor- -transparent suggestions, a challenge exacerbated by the dy- ization (e.g., TimeSVD++ [4]), and sequential deep learning -namic nature of user preferences. Traditional user profiling, models like GRU4Rec [5] and SASRec [6], which learn user -such as averaging item embeddings, often oversimplifies user dynamics from interaction sequences. Recent work has also -interests by conflating fleeting short-term desires with sta- explored long short-term preference modeling for continuous- -ble long-term tastes, hindering both recommendation quality time sequential recommendation [7]. -and interpretability. This paper examines the capability of Research in recommender systems increasingly emphasizes -leveraging Large Language Models (LLMs) to capture these balancing predictive accuracy with interpretability. Early ef- -temporaldynamics.Ourapproachgeneratesdistinctshort-term forts integrated explanations through content summarization -and long-term textual summaries of user interaction histories, or leveraged user-generated reviews [8]. More recently, Large -which are then encoded via BERT [1] and adaptively fused LanguageModels(LLMs)[9],[10]haveemergedaspowerful -using an attention mechanism [2] to form a unified user tools for generating richer explanations, demonstrating their -representation. capacity to enhance interpretability through personalized nar- -Our observations reveal a nuanced capability of this LLM- rativesandcollaborativeadaptors[11]–[14].Knowledgegraph -driven approach across varying user engagement patterns. (KG)-based approaches [15] have also integrated structured -While it tends to improve recommendation quality in ac- side information for explanations, sometimes combined with -tive domains like Movies&TV, with improvements such as LLM outputs [15]–[18]. - -| | | Fig.1. | Pipeline | for | LLM-Driven | Temporal | | User Profile | Generation | and | Recommendation. | | | | | -| --- | --- | ------ | -------- | --- | ---------- | -------- | --- | ------------ | ---------- | --- | --------------- | --- | --- | --- | --- | -Ourworkintersectsthesetwocriticalareasbyinvestigating (i) LLM-based Temporal Profile Generation: To capture -the utility of leveraging LLMs to explicitly model temporal temporal aspects of user preferences, our approach prompts -user dynamics for enhanced content-based recommendations the LLM twice over each user’s interaction history H , using -u -and inherent transparency. Unlike existing methods that pri- distinct instructions to generate two textual summaries. The -marily use static profiles or sequential models without natural first focuses on recent interactions, producing the short-term -| language | grounding | [6], | [19], | [20], this | study | explores | an | profile: | | | | | | | | -| -------------- | -------------- | --------- | -------- | ---------- | -------- | ------------- | --- | ---------- | ---------- | ------- | ---------- | -------------- | --- | ------ | -------- | -| approach | that generates | | distinct | natural | language | summaries | | | | | | | | | | -| | | | | | | | | | | NLshort | =LLM(H | ,Promptshort). | | | (2) | -| | | | | | | | | | | u | | u | | | | -| for short-term | and | long-term | user | behaviors. | By | encoding | and | | | | | | | | | -| adaptively | fusing | these | semantic | summaries, | | this approach | | | | | | | | | | -| | | | | | | | | The second | summarizes | | persistent | behaviors | | across | the full | -aims to create intrinsically interpretable user profiles. This history, forming the long-term profile: -| unique combination | | transparently | | highlights | the | temporal | ra- | | | | | | | | | -| -------------------- | --------- | ---------------- | ------------ | ----------- | -------- | -------- | -------- | --------------- | ------- | ----------- | ---------------- | ------------- | ------------ | ------------- | ------ | -| tionale behind | | recommendations, | | bridging | advanced | | LLM- | | | | | | | | | -| | | | | | | | | | | NLlong | =LLM(H | ,Promptlong). | | | (3) | -| | | | | | | | | | | u | | u | | | | -| based explainability | | with | nuanced | temporal | dynamics | | of user | | | | | | | | | -| interests. | This work | provides | | preliminary | insights | | into the | | | | | | | | | -| | | | | | | | | These summaries | | enhance | interpretability | | by | separating | tran- | -| practical | utility | of this novel | integration. | | | | | | | | | | | | | -| | | | | | | | | sient and | stable | user | interests. | | | | | -| | | | | | | | | To provide | | a concrete | understanding | | of | our LLM-based | | -| | | III. | METHODOLOGY | | | | | | | | | | | | | -| | | | | | | | | temporal | profile | generation, | | we include | a conceptual | | illus- | -Let U = u ,u ,...,u be users, and I = i ,i ,...,i tration of the prompts and their corresponding output in -| | 1 | 2 | |U| | | | 1 2 | |I| | | | | | | | | | -| --------- | ---- | -------- | ----------- | ---- | -------- | --- | -------- | --------- | --- | ------- | ------------- | -------- | --- | --------- | -------- | -| be items. | Each | user u ∈ | U interacts | with | a subset | | of items | | | | | | | | | -| | | | | | | | | Figure 3. | The | prompts | are carefully | designed | | to elicit | distinct | -I u ⊆ I, with interactions associated to timestamps. We temporal summaries. As shown, the ”Long-term Preference -| define the | interaction | history | as | Hu=(i,t | ):i∈I | | sorted | | | | | | | | | -| ---------- | ----------- | ------- | --- | ------- | ----- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -u,i u Prompt” emphasizes identifying enduring interests and con- -chronologically. -| | | | | | | | | sistent themes | | across | the user’s | entire | history. | In | contrast, | -| --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | ------ | ---------- | ------ | -------- | --- | --------- | -∈Rd) -We aim to learn embeddings for users (e u and items the ”Short-term Preference Prompt” is engineered to focus -| (e ∈Rd) | to predict | user-item | | interaction | likelihood: | | | | | | | | | | | -| ------- | ---------- | --------- | --- | ----------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -i specifically on recent behaviors and fleeting interests. While -thefull,detailedpromptsandtheirconfigurationsareavailable -| | | yˆ | =f(e | ,e ), | | | (1) | | | | | | | | | -| --- | --- | --- | ---- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | u,i | | u i | | | | | | | | | | | | -inourpublicrepository(seesectionIV)toensurereproducibil- -wheref isamulti-layerperceptron(MLP).Thepipeline(Fig- ity, these examples demonstrate the core strategy of using -ure 1) comprises three steps: (i) profile generation via LLMs, natural language instructions to temporally disentangle user -| (ii) semantic | embedding, | | and (iii) | interaction | prediction. | | | preferences. | | | | | | | | -| ------------- | ---------- | --- | --------- | ----------- | ----------- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | - -Fig. 2. A conceptual illustration of the framework’s potential for enhancing transparency. (A) and (B) show the short-term and long-term -textual profiles generated and encoded by our current approach. (C) shows a hypothetical extension wherein the final recommendation is -explicitly justified by both sets of preferences, further strengthening user-facing transparency—an aspect we plan to explore in future work. -TABLEI -DATASETSSTATISTICS -| | | | | | | | | Dataset | Users | Items | | Interactions | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | ----- | ------------ | --- | --- | -| | | | | | | | | Movies&TV | 10,000 | 14,420 | | 202,583 | | | -| | | | | | | | | Games | 10,371 | | 3,790 | 83,842 | | | -TABLEII -USERPROFILESSTATISTICS -| | | | | | | | Dataset | | Mean | Median | Mode | | StdDev | | -| --- | --- | --- | --- | --- | --- | ---------- | ----------- | ----------- | -------- | --------- | --------- | ------ | ---------- | ------ | -| | | | | | | | Movies&TV | | 11.79 | 9.00 | | 6 | 9.80 | | -| | | | | | | | Games | | 4.55 | 3.00 | | 3 | 3.97 | | -| | | | | | | Crucially, | | the learned | weights | | (αshort, | αlong) | explicitly | con- | -| | | | | | | | | | | | u | u | | | -| | | | | | | vey | the model’s | | decision | rationale | regarding | | recent | versus | -| | | | | | | historical | user | interests. | | | | | | | -(iii)InteractionPrediction:Topredictinteractionprobabili- -| | | | | | | ties, | we concatenate | | user embedding | | e | with item | embedding | | -| --- | --- | --- | --- | --- | --- | ----- | -------------- | --- | -------------- | --- | --- | --------- | --------- | --- | -u -| | | | | | | e and | process | through | an | MLP | [21]: | | | | -| --- | --- | --- | --- | --- | --- | ----- | ------- | ------- | --- | --- | ----- | --- | --- | --- | +WHITE PAPER + +In-Store +Shopper +Engagement +Built for the +Future of Retail + + There is an opportunity to +reinvent the moment of truth … + +is your store +ready? + + In-Store Shopper Engagement Built for the Future of Retail + +Introduction + +Now that the offline market is reopening, +it is time for retail brands to upgrade their +brick-and-mortar stores and deliver a +smart retail experience for customers. + +Gartner1 + +Retail Industry Drivers Impacting the In-Store +Experience + +• Consumers have returned to retail stores. And they aren’t just ready to get back to in- +person shopping, they have also accelerated their adoption of new technologies and +expect similar levels of digital technology and content in the physical store as they do in +the digital world. + +• After reporting the results of a recent consumer survey, Gartner concluded, “To maximize +the ROI on digital and capitalize on consumers’ interests in hands-free technology and +physical stores, retail brands should pursue a smart retail strategy and offer an +immersive omnichannel experience for their customers.”1 + +• That’s why in this white paper, we’ll dive into the trends that are impacting the need for a +more content-driven consumer experience, explore what’s possible in the store, and the +business benefits of creating a more memorable shopper journey with 6 ways to stand +out from the crowd. + +3 + + In-Store Shopper Engagement Built for the Future of Retail + +Consumers are Craving More Content, Recipes +and Deals +Analysis by Insider Intelligence, reports that retail sales are anticipated to grow between +4% and 5% each year through 2025. That’s a projection to hit $31.3 trillion in sales.2 + +Brick-and-Mortar vs. eCommerce Sales in US for 2021 + +) +s +n +o + i -| Fig.3. Examples | of | Prompts | for LLM-based | Temporal | User Profile | | | | | | | | | | -| --------------- | --- | ------- | ------------- | -------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(cid:0) (cid:1) -| Generation (Movies&TV | | Domain) | | | | | | | yˆ =MLP | | [e ;e ] | , | | (9) | -| --------------------- | --- | ------- | --- | --- | --- | ---- | ------ | ------------- | ---------- | --- | ----------- | --- | ----------- | --- | -| | | | | | | | | | u,i | | u i | | | | -| | | | | | | with | output | yˆ u,i ∈[0,1] | indicating | | interaction | | likelihood. | The | -(ii)SemanticEmbeddingandAttentionFusion:Eachtextual model is trained via binary cross-entropy loss: -| profile is transformed | | into | a high-dimensional | | embedding via | | | | | | | | | | -| ---------------------- | --- | ---------------------- | ------------------ | --- | ------------- | --- | --- | -------- | --------- | ----- | ----- | ---- | ------- | ---------------- | -| BERT [1]: | | | | | | | 1 | (cid:88) | (cid:104) | | | | (cid:0) | (cid:1)(cid:105) | -| | | | | | | L=− | | | y | logyˆ | +(1−y | )log | 1−yˆ | . | -| | | rshort =BERT(NLshort), | | | | | | | u,i | u,i | | u,i | | u,i | -(4) |D| -| | | u | | u | | | | (u,i,yu,i)∈D | | | | | | | -| --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | ---- | -| | | rlong =BERT(NLlong), | | | | | | | | | | | | (10) | -(5) -u u where D is the training set and y the observed interactions. -u,i -| yielding embeddings | | that | semantically | represent | short-term | | | | | | | | | | -| ------------------- | --- | ---- | ------------ | --------- | ---------- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -| | | | | | | | | | IV. EXPERIMENTS | | | | | | -andlong-terminterests,respectively.Todynamicallyintegrate -| | | | | | | We | evaluate | the | effectiveness | | of the | LLM-driven | | tem- | -| -------------------- | ----------- | --------- | ---- | ---------- | ------------------- | ----- | -------- | --------- | ------------- | --- | ------ | ---------- | ------ | ---- | -| these temporal | embeddings, | | we | apply a | learnable attention | | | | | | | | | | -| | | | | | | poral | user | profiling | framework | | on | two | Amazon | do- | -| layer [2], producing | | a unified | user | embedding: | | | | | | | | | | | -mains[22]—Movies&TVandVideoGames—chosenfortheir -| | | | exp(W | rs hort) | | | | | | | | | | | -| --- | --- | --- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -a u differing user profile density and behavioral variability. This -| αshort | = | | | | , (6) | | | | | | | | | | -| ------ | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -u rshort)+exp(W rlong) allows assessment under diverse recommendation conditions. -| | | exp(W | a | | a u | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -u -| | | | | | | Dataset | statistics | | are in Tables | I | and II. | For LLM | processing, | | -| --- | --- | ----- | ---------- | --- | --- | ------- | ---------- | --- | ------------- | --- | ------- | ------- | ----------- | --- | -| | | αlong | =1−αshort, | | | | | | | | | | | | -u u (7) we retain English item descriptions exceeding 500 characters. -where W ∈R1×d is a learnable parameter vector. The final For benchmarking, we compare against four baselines: Cen- +l +l +i + +M +$ +( + +l + +s +e a -| | | | | | | tric | [3], which | averages | item | embeddings | | without | | temporal | -| --------- | --- | --- | --- | --- | --- | --------- | ---------- | ----------- | ---- | ---------- | --------- | -------- | --- | --------- | -| embedding | is: | | | | | | | | | | | | | | -| | | | | | | modeling. | | Temp-Fusion | is | also | included, | designed | | to assess | -=αshort·rshort+αlong·rlong. -e u (8) temporal fusion’s benefit without LLM-generated semantic -| | | u | u | u | u | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -TABLEIII -COMPARATIVEPERFORMANCEEVALUATIONOFLLM-DRIVENTEMPORALUSERPROFILINGACROSSDIFFERENTDOMAINS -asteriskdenotesastatisticallysignificantimprovementoverthebaseline(Centric)(p<0.05). -| Method | | | | | | Movies&TV | | | | | Video | Games | | | -| ------ | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | ----- | ----- | --- | --- | -Recall@10 NDCG@10 Recall@20 NDCG@20 Recall@10 NDCG@10 Recall@20 NDCG@20 -Centric 0.0113 0.0191 0.0199 0.0269 0.0645 0.0532 0.0932 0.0649 -Popularity 0.0082 0.0145 0.0133 0.0191 0.0397 0.0324 0.0706 0.0453 -| MF | | | 0.0048 | | 0.0085 | | 0.0087 | 0.0124 | 0.0457 | | 0.0370 | 0.0754 | | 0.0491 | -| --- | --- | --- | ------ | --- | ------ | --- | ------ | ------ | ------ | --- | ------ | ------ | --- | ------ | -Temp-Fusion 0.0118 0.0201 0.0207 0.0276 0.0693 0.0589 0.0982 0.0712 -LLM-TP 0.0132∗ 0.0217∗ 0.0223∗ 0.0293∗ 0.0665∗ 0.0547∗ 0.1021∗ 0.0683∗ -| GainofLLM-TPvs.Centric | | | 17% | | 14% | | 12% | 9% | 3% | | 3% | 10% | | 5% | -| ---------------------- | ----------- | ----------- | ---------- | --------- | ------ | ---------- | --------- | --- | --- | --- | --- | --- | --- | --- | -| profiles; | it segments | interaction | | histories | into | short-term | and | | | | | | | | -| long-term | numerical | item | embeddings | | (e.g., | by | averaging | | | | | | | | -withinrecent/historicalwindows)andfusesthemviaattention, -isolatingtheLLM’stextualcontribution.Twostandardcontrol -| baselines | are Popularity | | [23], | a non-personalized | | | ranking | | | | | | | | -| -------------- | -------------- | --- | ------------- | ------------------ | --------- | ---------- | ------------ | --- | --- | --- | --- | --- | --- | --- | -| approach, | and Matrix | | Factorization | | (MF) | [24], | a collabora- | | | | | | | | -| tive filtering | technique | | without | textual | features. | Evaluation | | is | | | | | | | -performedusingarigorousper-usertemporalholdoutprotocol. -| For each | user, we | first | chronologically | | sort | their | interactions | | | | | | | | -| ------------- | ------------- | --------- | --------------- | -------- | ----------- | ------- | ------------ | --- | --- | --- | --- | --- | --- | --- | -| by timestamp. | The | dataset | is | then | split | so that | a user’s | | | | | | | | -| training | data always | precedes | | their | validation | and | test data. | | | | | | | | -| This ensures | that | our model | is | trained | only | on past | behaviors | | | | | | | | -| to predict | future | ones, | thereby | strictly | preventing | | within-user | | | | | | | | -| temporal | data leakage. | | We report | model | performance | | using | | | | | | | | -standardmetrics,includingRecall@KandNDCG@K.Textual -summaries are encoded via SBERT [25] (MiniLM-L6-v2, Fig.4. RelativegainsinRecall@20andNDCG@20ofthefullmodel -384-dim), with profiles generated using GPT-40-mini [26]. over ablation variants on the Movies&TV dataset. -| Interaction | likelihoods | | are predicted | | using | an MLP | (hidden | | | | | | | | -| ----------- | ----------- | --- | ------------- | --- | ----- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | -size128,dropout0.2)trainedwithbinarycross-entropy,batch -size 2048, Adam optimizer, and early stopping (patience=5), smaller K-values. This outcome likely arises from interaction -| executed | on four | NVIDIA | A100 | GPUs. | All | code, | data, and | | | | | | | | -| -------- | ------- | ------ | ---- | ----- | --- | ----- | --------- | --- | --- | --- | --- | --- | --- | --- | -sparsity;ourconjectureisthatloweruseractivityoftenleadsto -| prompts | are available | in | a public | GitHub | repository1. | | | | | | | | | | -| ---------- | -------------- | --- | -------- | ------ | ------------ | --- | --- | ---------------- | ---- | ---------- | ------------ | ---------- | --------- | ------------- | -| | | | | | | | | comparatively | more | stable | preferences, | limiting | | the incremen- | -| | | | | | | | | tal advantage | of | detailed | temporal | modeling | through | textual | -| A. Results | and Discussion | | | | | | | | | | | | | | -| | | | | | | | | differentiation. | This | highlights | | a critical | trade-off | between | -Thissectionpresentstheempiricalresultsofourassessment, enhancedpredictivepowerandassociatedcomputationalcosts, -| summarized | in Table | III, | highlighting | | key | insights | into the | | | | | | | | -| ---------- | -------- | ---- | ------------ | --- | --- | -------- | -------- | ---------- | ----------------- | --- | --- | ------------ | --- | --- | -| | | | | | | | | suggesting | context-dependent | | LLM | application. | | | -capabilityoftheLLM-driventemporaluserprofilingapproach. -| | | | | | | | | Intrinsic | Explainability | | Potential: | | The framework | in- | -| --- | --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | ---------- | --- | ------------- | --- | -TemporalProfilingBenefitsDomainswithHighUserAc- -| | | | | | | | | herently | supports | interpretability | | (Figure | 2). LLM-generated | | -| --- | --- | --- | --- | --- | --- | --- | --- | -------- | -------- | ---------------- | --- | ------- | ----------------- | --- | -tivity:OntheMovies&TVdataset,characterizedbylargerand -| | | | | | | | | natural language | | summaries | (Figure | 2A | and | 2B) clearly | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --------- | ------- | --- | --- | ----------- | -higheruseractivity(averageprofilesize:11.79,std:9.80),the -| | | | | | | | | reflect distinct | short-term | | versus | long-term | user | interests, | -| --------- | --------- | ------------ | --- | ----------- | --- | -------------- | --- | ---------------- | -------------- | --- | --------- | ------------- | ---- | ----------- | -| evaluated | framework | demonstrates | | significant | | effectiveness. | | It | | | | | | | -| | | | | | | | | providing | human-readable | | profiles. | Additionally, | | learned at- | -notably outperforms the Centric baseline (17% Recall@10, tentionweights(αshort,αlong)transparentlyindicatetherelative -| 14% NDCG@10 | | improvements) | | and | Temp-Fusion. | | These ob- | | | u | u | | | | -| ---------------------- | ---------- | ------------- | -------- | ---------------- | ------------ | --------- | ------------ | -------------------- | --------- | ---------- | -------------- | ----------- | ---------- | ----------- | -| | | | | | | | | influence | of these | profiles | in | the unified | user | embedding, | -| servations | underscore | the | value | of incorporating | | | semantically | | | | | | | | -| | | | | | | | | aiding understanding | | of | recommendation | | rationale. | Figure 2C | -| rich, natural-language | | user | profiles | from | LLMs, | | particularly | | | | | | | | -| | | | | | | | | illustrates | how these | components | | could | generate | user-facing | -| where user | behaviors | are | varied | and | frequently | changing. | | | | | | | | | -explanations,anaspectforfutureworktofullyrealizebuilt-in -| Mixed | Effectiveness | | for | Domains | with | Lower | User | | | | | | | | -| ----- | ------------- | --- | --- | ------- | ---- | ----- | ---- | --- | --- | --- | --- | --- | --- | --- | -transparency. -| Activity: | In the | Video | Games | domain, | with | smaller | and less | | | | | | | | -| --------- | ------ | ----- | ----- | ------- | ---- | ------- | -------- | --- | --- | --- | --- | --- | --- | --- | -Insummary,ourexperimentssuggestthattemporalprofiling -| frequent | user interactions | | (mean: | 4.55, | std: | 3.97), | the assess- | | | | | | | | -| -------- | ----------------- | --- | ------ | ----- | ---- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -mentrevealsmixedeffectiveness.Whileourapproachachieves with LLMs is practical and interpretable, especially for do- -the highest Recall@20, Temp-Fusion slightly surpasses it at mains with dynamic user behavior. Future work will focus on -| | | | | | | | | developing | user-facing | interfaces | | and evaluating | | explanations | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | ---------- | --- | -------------- | --- | ------------ | -1https://github.com/milsab/UMRec to enhance transparency and trust. - -TABLEIV LanguageTechnologies,Volume1(LongandShortPapers),J.Burstein, -COMPARISONFORABLATIONVARIANTS(MOVIES&TV) C. Doran, and T. Solorio, Eds. Minneapolis, Minnesota: Association -for Computational Linguistics, Jun. 2019, pp. 4171–4186. [Online]. -AblationVariant Recall@20 NDCG@20 Available:https://aclanthology.org/N19-1423/ -Short-TermOnly(ST) 0.0193 0.0266 [2] A.Waswani,N.Shazeer,N.Parmar,J.Uszkoreit,L.Jones,A.Gomez, -Long-TermOnly(LT) 0.0195 0.0265 L.Kaiser,andI.Polosukhin,“Attentionisallyouneed,”inNIPS,2017. -GeneralPreferences(NoTS) 0.0178 0.0247 [3] M. Polignano, C. Musto, M. de Gemmis, P. Lops, and G. Semeraro, -Dot-ProductScoring(DP) 0.0136 0.0173 “Togetherisbetter:Hybridrecommendationscombininggraphembed- -FullModel(LLM-TP) 0.0223 0.0293 dings and contextualized word representations,” in Proceedings of the -15thACMconferenceonrecommendersystems,2021,pp.187–198. -[4] Y.Koren,“Collaborativefilteringwithtemporaldynamics,”inProceed- -ingsofthe15thACMSIGKDDinternationalconferenceonKnowledge -V. ABLATIONSTUDY -discoveryanddatamining,2009,pp.447–456. -We conducted an ablation study to determine the impact [5] B. Hidasi, A. Karatzoglou, L. Baltrunas, and D. Tikk, “Session- -basedrecommendationswithrecurrentneuralnetworks,”arXivpreprint -of individual components within the evaluated framework -arXiv:1511.06939,2015. -on recommendation accuracy, using the Movies&TV dataset. [6] W.-C. Kang and J. McAuley, “Self-attentive sequential recommenda- -Table IV summarizes the performance metrics, and Fig. 4 tion,” in 2018 IEEE international conference on data mining (ICDM). -IEEE,2018,pp.197–206. -highlighting relative improvements. More results are in the -[7] H.Chi,H.Xu,H.Fu,M.Liu,M.Zhang,Y.Yang,Q.Hao,andW.Wu, -GitHub repository (see Section IV). The ablation results un- “Long short-term preference modeling for continuous-time sequential -derscore the importance of temporal modeling and non-linear recommendation,”arXivpreprintarXiv:2208.00593,2022. -[8] Y.Zhang,X.Chenetal.,“Explainablerecommendation:Asurveyand -scoring.TheGeneralPreferences(NoTS)variant,wherethe -new perspectives,” Foundations and Trends® in Information Retrieval, -LLM generated a single, holistic user profile (distinct from vol.14,no.1,pp.1–101,2020. -Temp-Fusion’s numerical aggregation), showed a consistent [9] W.X.Zhao,K.Zhou,J.Li,T.Tang,X.Wang,Y.Hou,Y.Min,B.Zhang, -J. Zhang, Z. Dong et al., “A survey of large language models,” arXiv -performance drop of over 20% in Recall and 18% in NDCG, -preprintarXiv:2303.18223,2023. -emphasizing the value of LLM-generated temporal distinction [10] Y.Chang,X.Wang,J.Wang,Y.Wu,L.Yang,K.Zhu,H.Chen,X.Yi, -(Fig. 4). The full framework also consistently outperformed C. Wang, Y. Wang et al., “A survey on evaluation of large language -models,” ACM transactions on intelligent systems and technology, -Short-Term Only (ST) and Long-Term Only (LT) variants -vol.15,no.3,pp.1–45,2024. -individually (e.g., +15.2% Recall@20 over ST), confirm- [11] Z. Zhao, W. Fan, J. Li, Y. Liu, X. Mei, Y. Wang, Z. Wen, F. Wang, -ing complementary signals from both temporal components. X. Zhao, J. Tang et al., “Recommender systems in the era of large -language models (llms),” IEEE Transactions on Knowledge and Data -Lastly,replacingMLPscoringwithadotproduct(DPvariant) -Engineering,2024. -resulted in the steepest degradation, highlighting the necessity [12] L. Wu, Z. Zheng, Z. Qiu, H. Wang, H. Gu, T. Shen, C. Qin, C. Zhu, -of non-linear interaction modeling. Overall, Figure 4 and H. Zhu, Q. Liu et al., “A survey on large language models for -recommendation,”WorldWideWeb,vol.27,no.5,p.60,2024. -TableIVdemonstratethattemporaldisentanglementviaLLM- -[13] S. Lubos, T. N. T. Tran, A. Felfernig, S. Polat Erdeniz, and V.-M. -based profile summarization and non-linear embedding fusion Le,“Llm-generatedexplanationsforrecommendersystems,”inAdjunct -collectivelycontributetotheobservedrobustrecommendation Proceedingsofthe32ndACMConferenceonUserModeling,Adaptation -andPersonalization,2024,pp.276–285. -performance within the evaluated framework. -[14] Q. Ma, X. Ren, and C. Huang, “Xrec: Large language models for -explainablerecommendation,”arXivpreprintarXiv:2406.02377,2024. -VI. CONCLUSION -[15] Q. Guo, F. Zhuang, C. Qin, H. Zhu, X. Xie, H. Xiong, and Q. He, -This paper presented an exploratory evaluation of a “A survey on knowledge graph-based recommender systems,” IEEE -Transactions on Knowledge and Data Engineering, vol. 34, no. 8, pp. -lightweight framework that leverages LLM-driven temporal -3549–3568,2020. -user profiling within a pipeline demonstrating interpretability [16] S. Wang, W. Fan, Y. Feng, S. Lin, X. Ma, S. Wang, and D. Yin, -potential. Our evaluation indicates varied capabilities: notable “Knowledgegraphretrieval-augmentedgenerationforllm-basedrecom- -mendation,”arXivpreprintarXiv:2501.02226,2025. -gains in high-activity domains (e.g., Movies&TV) versus -[17] R. Shimizu, M. Matsutani, and M. Goto, “An explainable recommen- -less pronounced benefits in sparser environments (e.g., Video dation framework based on an improved knowledge graph attention -Games). This highlights a critical trade-off between enhanced networkwithmassivevolumesofsideinformation,”Knowledge-Based -Systems,vol.239,p.107970,2022. -performance and computational costs, suggesting context- -[18] G. Shi, X. Deng, L. Luo, L. Xia, L. Bao, B. Ye, F. Du, S. Pan, -dependent LLM application. The framework inherently offers and Y. Li, “Llm-powered explanations: Unraveling recommendations -a promising pathway toward transparency through its natural throughsubgraphreasoning,”arXivpreprintarXiv:2406.15859,2024. -[19] Y.K.Tan,X.Xu,andY.Liu,“Improvedrecurrentneuralnetworksfor -language profiles and attention weights, providing an intrinsic -session-basedrecommendations,”inProceedingsofthe1stworkshopon -potentialforinterpretability.Theseearlyinsightscontributeto deeplearningforrecommendersystems,2016,pp.17–22. -understandingthepracticalcapabilityofLLM-driventemporal [20] Y.Zhu,H.Li,Y.Liao,B.Wang,Z.Guan,H.Liu,andD.Cai,“What -to do next: Modeling user behaviors by time-lstm.” in IJCAI, vol. 17, -user profiling. Future work includes developing user-facing -2017,pp.3602–3608. -explanations, conducting user studies, and exploring adaptive [21] M.-C. Popescu, V. E. Balas, L. Perescu-Popescu, and N. Mastorakis, -LLM strategies. “Multilayer perceptron and neural networks,” WSEAS Transactions on -CircuitsandSystems,vol.8,no.7,pp.579–588,2009. -REFERENCES [22] Y. Hou, J. Li, Z. He, A. Yan, X. Chen, and J. McAuley, “Bridging -language and items for retrieval and recommendation,” arXiv preprint -[1] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- arXiv:2403.03952,2024. -trainingofdeepbidirectionaltransformersforlanguageunderstanding,” [23] F. Ricci, L. Rokach, and B. Shapira, “Introduction to recommender -in Proceedings of the 2019 Conference of the North American systems handbook,” in Recommender systems handbook. Springer, -Chapter of the Association for Computational Linguistics: Human 2011,pp.1–35. - -[24] Y.Koren,R.Bell,andC.Volinsky,“Matrixfactorizationtechniquesfor -recommendersystems,”Computer,vol.42,no.8,pp.30–37,2009. -[25] N.ReimersandI.Gurevych,“Sentence-bert:Sentenceembeddingsusing -| siamese bert-networks,” | in Proceedings | | of the 2019 | Conference | on | -| ----------------------- | ------------------ | -------- | --------------- | ----------- | --- | -| Empirical | Methods in Natural | Language | Processing. | Association | | -| for Computational | Linguistics, | 11 | 2019. [Online]. | Available: | | -https://arxiv.org/abs/1908.10084 -[26] J.Achiam,S.Adler,S.Agarwal,L.Ahmad,I.Akkaya,F.L.Aleman, -| D. Almeida, | J. Altenschmidt, | S. Altman, | S. Anadkat | et al., | “Gpt-4 | -| ----------- | ---------------- | ---------- | ---------- | ------- | ------ | -technicalreport,”arXivpreprintarXiv:2303.08774,2023. \ No newline at end of file +S + +Y +o +Y +e +g +n +a +h +C +% + +Q2 +2018 + +Q3 +2018 + +Q4 +2018 + +Q1 +2019 + +Q2 +2019 + +Q3 +2019 + +Q4 +2019 + +Q1 +2020 + +Q2 +2020 + +Q3 +2020 + +Q4 +2020 + +Q1 +2021 + +Q2 +2021 + +Q3 +2021 + +Q4 +2021 + +B&M YoY Growth + +E-commerce YoY Growth + +E-commerce + +Brick & Mortar + +Source: US Dept of Commerce, Economic Indicators Division, Retail Indicator Branch +Estimated Quarterly U.S. Retail Sales (Not Adjusted): Brick & Mortar and E- + +Brick and mortar vs. E-Commerce Sales US Department of Commerce Retail Indicators Division + +More US consumers are switching brands and retailers now than +in 2020 and 2021, and about 90 percent plan to continue doing so. + +Shopped a different brand recently, +% of respondents + +Shopped a different retailer/store/ +website recently, % of respondents + ++13 percentage +points + ++9 percentage +points + +~90% of consumers +who have switched +brands or retailers +plan to incorporate +these new behaviors +into their routines + +Sep-20 +Source: McKinsey COVID-19 US Consumer Pulse Surveys: Feb 25 – Mar 1, 2022, n = 2,160; Oct 9-15, 2021, n = 2.095; Sept 18-24, 2020, n = +1,026; sampled and weighted to math the US general population aged 18 years and older + +Oct-21 Feb-22 + +Oct-21 Feb-22 + +Sep-20 + +4 + + In-Store Shopper Engagement Built for the Future of Retail + +That said, historically, consumers would get their Sunday +newspapers for coupons in order to get the best deals in +stores. Then, there became a hybrid of clipping coupons +along with digital apps or digital coupons for consumers to +get better offers. Today, and especially coming out of a +post-pandemic world, the in-store experience is even more +digital — with the decline of print newspapers, the shift to +online shopping and the consumer desire for a contactless +checkout experience. + +What’s more, grocers are relying on connected recipes as a +tool to drive Q4 spending. Walmart offers its “Cookshop” +interactive content portal, which currently presents +consumers with shoppable recipes and activities. Online +tutorials made in partnership with major brands like Kraft +or Starbucks, embed “shop your ingredients” buttons and +other shopping list features.6 + +So, retailers need to find the best way to make content, +recipes and deals easily consumable by consumers. In fact, +“When consumers visit an offline store, it is no longer to +simply buy something. Creating an interesting and +immersive experience that online shopping can not rival is +one of the keys to drive foot traffic,” according to Gartner.1 + +92% + +of grocery shoppers used +coupons or deals when shopping. + +60% + +of grocery shoppers said they +planned their shopping trip +based on which stores offer the +best deals. + +51% + +of grocery shoppers said they +used mobile apps to find deals +before going to the grocery store. + +Adoption of the Smart Cart at the ‘Moment +of Truth’ + +• With some of the aforementioned drivers impacting the way consumers are spending, +coupled with inflation and the supply chain bottlenecks still catching up post-COVID – +consumers are looking for deals more than ever right now, and are willing to take on new +technology to do so. + +• As such, the smart cart is becoming more prevalent in retail stores, but operational issues +typically get in the way of success. Cart costs are high (upwards of $250 million), especially +when they involve weigh scales and cameras to make them ‘smart.’ These carts come with +maintenance overhead, can be too small, and don’t stack or nest conveniently. + +• And let’s not forget that long lines during checkout leave retailers open to more shrink + +issues, and consumers open to better technology to solve their payment woes. + +• This leaves retailers with substantial investments in smart cart technology, but low or no + +adoption in stores. + +• However, there is an opportunity for retailers to present these shoppers with relevant +deals (combined with their loyalty information) right in the store aisles for the ideal in- +store experience — and as an added bonus — contactless checkout. + +5 + + In-Store Shopper Engagement Built for the Future of Retail + +EXPLORE WHAT’S POSSIBLE + +While the shopping cart hasn’t seen many innovations over the years + +retailers can engage with shoppers in a way that has never happened + +before by adding a touchscreen right to the cart. + +Not only can it present valuable and relevant information to shoppers on how to +get the best deals and ideas for new items or recipes to try — it can also show +shoppers how to locate those products right in store aisles. + +This is being done every day on social media, but being able to access this content +right on a cart without relying on a personal device connects the channels in a way +that satisfies the future shopper. + +Especially in a world where shoppers are still using their smartphones to conduct +research before heading into the store, adding a mobile and connected +touchscreen is a game-changer for retailers. + +Gartner reports that 95% of retailers are saying they plan to increase their +investments in digital capabilities in 2022.1 + +This means that your competitors are also boosting their technologies to enhance +their brand’s in-store and online shopping experiences with the goal to offer a +cohesive omnichannel experience that is immersive and personalized. + +Brands would be able to reach shoppers in an even more relevant way and be able +to direct them right to their product in the store aisles as they shop (right at the +precise ‘moment of truth.’ The items they need would always be top of mind, and +CPGs can easily connect the ROI back to their ad spend because that anonymized +data would be captured right into the connected screen. + +And, as shoppers look for specific products, they can come across + +personalized ads from brands or even push a recipe or coupon to + +inspire a new or spontaneous purchase (also creating more + +valuable loyalty data for retailers and their CPG suppliers). + +6 + + In-Store Shopper Engagement Built for the Future of Retail + +Did You Know? + +Sylvan Goldman +owner of the grocery chains +Piggly Wiggly and Humpty- +Dumpty. + +What: Goldman invented the shopping cart, arguably the most important innovation in retail +stores even today. + +Where: Oklahoma City + +When: 1937 + +Why: Groceries were getting heavier products (like canned goods), and more refrigerated +items, and were still using hand baskets to carry them. + +How: Goldman enlisted the help of his handyman Fred Young. The two grabbed a few +supplies and spent a night coming up with a prototype of a rolling grocery basket. + +The Result? Shoppers rejected it … at first. Men were offended by it saying it made them +seem weak for not being able to carry groceries in their hands, and women compared it to +pushing a baby stroller. But, with some advertising efforts, the innovation took off, +increasing basket sizes and improving the shopper experience. + +7 + + In-Store Shopper Engagement Built for the Future of Retail + +Business Benefits of Being Memorable + +Creating a truly memorable shopping experience has many business + +benefits, especially when leveraging technology to support it. Adding + +a connected screen, there are many positive implications for + +retailers, brands, and shoppers alike. Consider this: + +81% + +59% + +73% + +35% + +of retail shoppers conduct online research before buying. (Source: GE Capital +Retail Bank). With a connected screen, there’s no need for research prior to a +shopping trip meaning more accurate trips, fewer forgotten items, and no +need for a personal device (plus, more time well spent in the store). + +of shoppers use their mobile devices in-store to compare costs or research +deals and coupons. (Source: HRC Retail Advisory). Knowing a connected +screen is right on the shopping cart means no relying on my mobile device in +the store. + +of retail consumers use multiple channels to shop. To gain a unified +consumer view, marketers must connect all of their channels (Source: +Harvard Business Review). A connected screen allows for recipes, social +content, influencers, etc., to join a shopping trip creating a unified +experience. + +of marketers say poor data quality impacts their ability to target consumers +with the right digital ads. (Source: Forrester). Connected screens provide a +wealth of new data and insights right from the store shelves where the +business benefits spill over to digital advertising channels. + +84% + +of consumers believe retailers should be doing more to integrate their online +and offline channels. (Source: RetailNext). A connected screen would bring +both together into one memorable customer experience. + +40% + +of consumers purchase more from retailers that provide a personalized +shopping experience across channels. Improving the shopper experience will +have a direct impact on your bottom line (Source: RetailNext). The connected +screen converges personalization into brick-and-mortar. + +8 + + In-Store Shopper Engagement Built for the Future of Retail + +6 Ways to Stand Out from the Competition +While many media outlets will continue to talk about the acceleration of digital, retailers are +leaning into the physical store to drive loyalty within their customer experiences, as well as +differentiate themselves from their competition. Shoppers prefer to see, feel and even try +products in the store before making a purchase, so the opportunity for innovation lies +within the brick-and-mortar channel. + +Perhaps Gartner sums it up best, “Retail is experiencing unprecedented change driven by +consumer behaviors, enabled by technology and accelerated by a pandemic.”7 + +Getting started on a business process change or improvement like creating or revamping +your customer experience can be a daunting task. As such, here we have gathered 6 best +practices to help get your store ready for the future of retail (and a more memorable in- +store experience). + +Consumers want deals, but they want relevant deals. Combining loyalty +information about that consumer, plus where they are in the store gives retailers +the ability to drive information that is highly pertinent relative to that consumer. + +Leverage real-time data and insights to act on consumer trends more quickly, +keeping popular products in front of shoppers while they’re in high demand.8 Use +this new stream of data (i.e. heat maps) to understand consumer behavior in the +store (where they stopped, why, how long they stayed, what the interaction was) +and share that with your suppliers. + +Use technology to play a distinct and purposeful role to enhance customer +engagement and loyalty. Be sure it also supports your overall brand purpose.9 +Consumers don’t want to use different apps. Think about the development of your +apps and how you can seamlessly work them into the store environment. + +Create opportunities for speed and convenience (i.e. fast, contactless and +seamless checkout).9 Regardless of size, retailers must focus on checkout, +especially as the rise of contactless or self-service checkout is on the rise. + +Combine shrink avoidance technology, principles and methods — and embed +those right into your carts. + +Use shopping cart-based screens to present circulars to consumers at the moment +when it is most useful - as they enter that aisle where the products are on +promotion. FMI—The Food Industry Association (FMI) said in a report in 2022 that +55% of surveyed retailers found print circulars very effective and also noted that +widely used digital circulars and promotions sent by text are not as effective as +more traditional outreach methods.10 + +9 + + In-Store Shopper Engagement Built for the Future of Retail + +Many times when retailers are ready to implement a new solution or service, they can be +inundated by a ‘rip and replace’ mentality, but it doesn’t have to be that way. Start with small +wins and continue to evolve and enhance as you go with each iteration. Many providers +today can stand up their technologies in a matter of days and weeks so retailers can start +hitting their KPIs immediately and then continue to improve. + +“Even as consumers continue their return to stores, the digital behaviors they’ve picked up +during the pandemic will stick for the long term,” according to Retail Dive.9 “Brands and +retailers that win in the coming years will embrace digital innovation and find more +thoughtful and authentic ways to blend digital and physical engagement.” + +Now, retailers can engage with shoppers in a way that has never been done before – +present valuable information on how to get the best deals, how to locate products and ideas +for new items or recipes to try. This has never been accomplished in a store environment, +but the future of retail is here. + +Shopper Engagement Platform for the Physical +Store Powered by Veeve +Veeve's vision is to create a content delivery and checkout platform that utilizes the +retailers' own Apps to improve the shopper experience and increase the basket size +through in-store, location-driven promotions, advertising, and more. Veeve is delivered by a +Connected Screen: a purpose-built device that brings the power of online shopping, sensor +fusion, and computer vision to physical stores. + +The Connected Screen + +What’s on sale +today? + +Where can I find +salsa? + +Deals & +Coupons + +Location +Tracking + +What’s for +dinner? + +Video +Recipes + +Can I skip +checkout? + +Pay on +Cart + +Visit https://veeve.io/ to learn more. + +Source 1. Gartner, “Time To Prepare Smart Retail For The Post-Pandemic Era” by Wilson Zhao 2. Insider Intelligence, “Global Ecommerce +Forecast” 3. Mastercard SpendingPulse, “Mastercard SpendingPulse Forecasts 7.5%* Growth for U.S. Back-to-School Retail Sales” +4. McKinsey, “How US consumers are feeling, shopping, and spending—and what it means for companies” 5. Inmar Intelligence, “New +Inmar Intelligence Survey Reveals Over Half of Today’s Shoppers Have Switched Loyalty to Less Expensive Grocery Brands Since Start of +Pandemic” 6. PYMNTS, “Grocers See Connected Recipes as Tool to Drive Q4 Spending” 7. Gartner, “Digital Transformation in Retail” +8. Retail Dive, “Retail tech is elevating CX unlocking efficiencies” 9. Retail Dive, “3 pillars driving the future of store experience” +10. FMI The Food Retailing Industry Speaks 2022 + +1 +0 + diff --git a/q159/random_k4/question.json b/q159/random_k4/question.json index aaa73f9ca9f2ccf4dd3e4560846bc8c27cc1af28..dca8249f53314044a07c303cc29f649069a6f0b4 100644 --- a/q159/random_k4/question.json +++ b/q159/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "f8d3a162ab9507e021d83dd109118b60.pdf", - "web_ed8636d8f0faddf2.pdf", - "web_289916b20eaef9df.pdf", - "3becf3cee2ec9db3b4bdae0c4504cf6821a6d8e6.pdf", - "web_f915df7184680905.pdf" + "web_bf09a64e7d517c98.pdf", + "web_9942a44c5f6232cc.pdf", + "web_419331e34a19dd26.pdf", + "web_a28664fd28b776ae.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q159/random_k4/random_1.md b/q159/random_k4/random_1.md index 4df5215b83d8f847c48c411dacd211a9574481da..bdc0cc473d09f25f5a2f7f0b1e46304031484979 100644 --- a/q159/random_k4/random_1.md +++ b/q159/random_k4/random_1.md @@ -1,752 +1,637 @@ -C H A P T E R 1 +Youth in Middle East +and North Africa -Get started with -Web GIS +August 2019 -T his chapter introduces the basic concepts and workflows of Web GIS and the Esri® geo- +Daniella Raz +University of Michigan -spatial cloud. The chapter begins with an overview of Web GIS and its advantages; intro- -duces Esri’s geospatial cloud including ArcGIS® Online and ArcGIS Enterprise; illustrates online, -on-premises, and hybrid deployment patterns; lists the technical evolutions in Web GIS; explains -the content types, hosted feature layers, app types, and user types in Web GIS; and demonstrates -the workflow to build Web GIS apps using the Attachment Viewer instant app template. This -chapter familiarizes you with ArcGIS Online and ArcGIS Enterprise basic operations and work- -flows and introduces flexible ways to build Web GIS apps that you will explore in other chapters. +JULY 2018 Arab Barometer – Wave V +Topic Report – Youth -Objectives +Executive Summary -• Grasp the concept and advantages of Web GIS. -• Understand the deployment patterns of Web GIS. -• Learn the components of the new-generation Web GIS. -• Understand the technical evolutions and trends in Web GIS. -• Understand the concepts of web services and hosted feature layers. -• Publish hosted feature layers from CSV files. -• Learn the workflow for creating web maps and web apps. +Individuals between the ages of 15 and 29 account for roughly 30 percent of the +population of the Middle East and North Africa (MENA).1 The beliefs, attitudes, +and preferences of this large and growing segment of the population has the po- +tential to shape and drive the region’s future. Yet, widespread youth discontent +about their economic prospects poses a significant challenge. Youth unemploy- +ment rates across the region remain high and there is broad frustration with +government efforts to create employment opportunities.2 Though satisfaction +with the state education system varies greatly across countries, youth are less +satisfied than are older generations. -Copyright © 2022 Esri. All rights reserved. 2 +Youth also question if they have the ability to voice their frustrations to the gov- +ernment. In nearly all countries, less than half say the right to freedom of ex- +pression is guaranteed while half or less say they have the right to demonstrate +peacefully in all countries. This combination creates conditions where youth +across the MENA region, far more than older age cohorts, are considering emi- +grating from their country, primarily to Europe, North America, or GCC countries. +Unsurprisingly, many cite economic reasons as the main impetus for leaving. -Chapter 1: Get started with Web GIS +However, youth differ from their older counterparts not only in terms of greater +frustration with the existing system. Relative to older generations youth also +tend to be less religious, less interested in politics, and more likely to be engaged +in social media. Generational differences are particularly stark in Morocco where +youth attitudes diverge substantially and dissatisfaction with the overall state of +affairs is quite salient. -Web GIS and the geospatial cloud +Meanwhile, on gender issues, most youth tend to be supportive of equal rights +for women. Majorities say women have the right to be the head of government +and to receive an education. Yet, the same does not hold true for equal inher- +itance rights with minorities saying women should receive the same as men +in most countries. Additionally, this generation cites low levels of support for +women playing equal roles in society, with most saying that men make better +political leaders and stating that husbands should have the final say in family +matters. -Web GIS is the combination of the web and geographic information systems (GIS). The web -removed the constraint of distance and thus allows people the freedom to interact with GIS apps -globally and access information almost instantly. Web GIS uses web technologies, including but -not limited to Hypertext Transfer Protocol (HTTP), Hypertext Markup Language (HTML), -Uniform Resource Locator (URL), JavaScript, Web Graphics Library (WebGL), and WebSocket. -The first operational GIS was developed in the 1960s by Roger Tomlinson (who developed -the Canada Geographic Information System). Since then, GIS has evolved from a local file-based, -single-computer system to a central database client/server system, often with multiple servers -and many more client computers. The invention of the internet in the 1980s and the growth -of the World Wide Web in the early 1990s laid the foundation for an evolutionary leap toward -Web GIS. In 1993, the Palo Alto Research Center (PARC), a Xerox company, developed a map- -ping web page, which marked the origin of Web GIS. In the 2000s, Web GIS evolved into a new -generation — a system of distributed web services in the cloud you can access anywhere, repre- -sented by ArcGIS. +Youth in MENA appear to be particularly inclined toward China relative to other +global powers. Majorities favor closer economic relations with China in most +countries surveyed, while less than half say the same about relations with the +U.S. in most countries. -Inheriting the power of the web and the cloud, Web GIS offers many advantages: +1https://www.undp.org/content/dam/rbas/report/AHDR%20Reports/AHDR%202016/AHDR%20Final% -• Global reach: You can share your geographic information easily, within your organization +202016/AHDR2016En.pdf, pg. 7 -and with people all over the world. +2https://www.imf.org/external/np/vc/2012/061312.htm?id=186569 -• Large number of users: You can share your app with dozens, or even millions, of users sup- +www.arabbarometer.org -ported by scalable cloud technology. +2 -• Low cost per user: The cost of building one Web GIS app is often lower than building a + Arab Barometer – Wave V +Topic Report – Youth -stand-alone desktop solution and installing it for every user. +These are among the key findings from 11 nationally representative public opin- +ion surveys conducted across the Middle East and North Africa by the Arab +Barometer from September 2018 - April 2019. The surveys included more than +25,000 face-to-face interviews in the respondent’s place of residence and have +a margin of error of ±2 percent in most countries.3 -• Better cross-platform capabilities: Web apps, especially those built with JavaScript, can +Religiosity -run on desktop and mobile browsers running a wide range of operating systems, including -Windows, Mac OS, and Linux, along with iOS, Android, and Windows Phone. +Across the region, youth are less religious than their older counterparts. In fact, in +every country surveyed less than half of youth identify as religious. More specif- +ically, fewer than one-fifth of youth in Algeria (15 percent), Tunisia (16 percent), +and Libya (18 percent), and fewer than one-quarter of youth in Jordan (22 per- +cent), Lebanon (23 percent), and Morocco (24 percent) consider themselves reli- +gious. Higher rates of youth religiosity are found in Palestine (28 percent), Sudan +(32 percent), Yemen (33 percent) Egypt (36 percent), and Iraq (42 percent). -• Easy to use: Web GIS apps typically incorporate simplicity, intuition, and convenience in -their design. Public users can use these apps without a technical background (see figure). -• Easy to maintain: Web clients can benefit from the latest program and data updates each -time they access a web app. The web administrator does not have to update all the clients -separately. +However, trends over time paint a more complicated picture throughout the +region. While indeed less than half of MENA youth identify as religious, in some +countries these rates represent an increase from the last Arab Barometer wave -Copyright © 2022 Esri. All rights reserved. Web GIS and the geospatial cloud +3For details about each survey, please visit: -3 - -GIS professional -using Web GIS - -Casual users - -Developers - -Open data -community - -Web GIS offers a collaborative environment and a pattern for delivering GIS capabilities. GIS -professionals share information to Web GIS and extend geospatial intelligence to users across -organizations and throughout communities. - -GIS is the science of locations — The Science of Where®. The term has two meanings. One mean- -ing is that GIS is itself a science, as the scientific basis for GIS technology. The other meaning -is that GIS has been used for science as an effective tool for making scientific discoveries. The -Science of Where is now recognized as a force for solving problems and understanding our -world. Web GIS takes the science to a whole new scale, transforming how we share and col- -laborate and revealing deeper insight into data. The common patterns of Web GIS applications -include mapping and visualization, data management, field mobility, monitoring, analytics, -design and planning, decision support, constituent engagement, sharing, and collaboration. -Web GIS unlocks and delivers this science to offices and homes and puts GIS technology in the -hands of billions of people. Web GIS provides immense value to government, business, science, -and daily life. In recent years, the concept and importance of spatial location have become more -mainstream, and Web GIS awareness is growing more prominent in many organizations: - -• Government: Web GIS offers an ideal channel for sharing public information services and -delivering open data, an engaging medium for encouraging public participation, and a pow- -erful framework for supporting decision-making. - -• Business: Web GIS helps create novel business models and reshapes existing ones. It - -enhances the power of location-based advertising, business analysis, and volunteered geo- -graphic information, generating tremendous revenue directly and indirectly. -• Science: Web GIS creates research areas and renews existing avenues of research. -• Daily life: Web GIS helps people decide where to eat, stay, shop, and learn how to get from - -place to place. - -Table 1.1 describes patterns of Web GIS applications, including mapping and visualization, data -management, and analytics. - -Copyright © 2022 Esri. All rights reserved. 4 - -Chapter 1: Get started with Web GIS - -Table 1 .1 . Common patterns of Web GIS applications - -Pattern name - -Description - -Mapping and visualization Understand locations and relationships with maps and visual representations. - -Data management - -Collect, organize, and maintain accurate locations and details about assets -and resources. - -Field mobility - -Monitoring - -Analytics - -Manage and enable a mobile workforce to collect and access information in -the field. +https://www.arabbarometer.org/waves/ -Track, manage, and monitor assets and resources in real time. +arab-barometer-wave-v/ -Discover, quantify, and predict trends and patterns to improve outcomes. +www.arabbarometer.org -Design and planning - -Evaluate alternative solutions and create optimal designs. - -Decision support - -Gain situational awareness and enable information-driven decision-making. - -Constituent engagement - -Communicate and collaborate with citizens and external communities -of interest. - -Sharing and collaboration - -Empower everyone to discover, use, make, and share geographic information. - -The geospatial cloud powers Web GIS -Esri’s Web GIS capabilities are provided by ArcGIS Online and ArcGIS Enterprise, which are at -the core of the geospatial cloud. - -• ArcGIS Online is a software-as-a-service (SaaS) offering of Web GIS. ArcGIS Online is - -hosted on the Amazon Web Services (AWS) cloud and Microsoft Azure cloud platforms and -is managed by Esri. There is no hardware infrastructure for an organization to maintain. -ArcGIS Online hosts hundreds of millions of content items and registered users. It serves bil- -lions of maps per day with high performance, scalability, and availability. With the benefits -of cloud computing, ArcGIS Online has been quickly adopted by government and commer- -cial organizations around the world. - -• ArcGIS Enterprise is a Web GIS software product that organizations can acquire and deploy. - -ArcGIS Enterprise includes four basic software components: Portal for ArcGIS, ArcGIS -Server, ArcGIS Data Store, and ArcGIS Web Adaptor. These components support flexible -ways for deployment, from a single-machine deployment to multiple-machine deployments, -highly available (HA) deployment, on Windows, Linux, and the Kubernetes platform, in -on-premises data centers or in AWS, Microsoft Azure, Google Cloud, and many other cloud -platforms. - -Table 1.2 lists these deployment patterns. +3 -Copyright © 2022 Esri. All rights reserved. Web GIS and the geospatial cloud + Arab Barometer – Wave V +Topic Report – Youth + +while in others a decrease. For instance, rates of religiosity have increased in +Egypt (+12), Yemen (+7), Tunisia (+6), Iraq (+6), Morocco (+2), and Libya (+2). In +contrast, youth identify as religious at lower rates in Sudan (-12), Palestine (-9), +Jordan (-7), Algeria (-3) and Lebanon (-2). + +Views on Politics + +In addition to generally low levels of religiosity, youth appear disinterested in +politics, both overall and when compared to older generations. Only three-in- +ten youth in Lebanon (31 percent), Yemen (29 percent), Palestine (29 percent), +Sudan (28 percent), and Egypt (27 percent) say they are interested or very inter- +ested in politics, while no more than a quarter say the same in Jordan (25 per- +cent), Morocco (23 percent), Iraq (21 percent), and Libya (21 percent). Tunisian +and Algerian youth are least politically interested with fewer than two-in-ten +citing interest in politics (17 percent and 15 percent, respectively). + +With the exception of Egypt and Lebanon, youth in the region are less interested +in politics than are older citizens. The difference between age cohorts is widest +in Tunisia and Palestine (10 percentage points). Elsewhere, though differences +are smaller, older citizens nonetheless report greater interest in politics than do +youth. +In Lebanon youth have only marginally greater interest (1 percentage +point), while in Egypt youth are 6 percentage points more likely than older citi- + +www.arabbarometer.org + +4 + + Arab Barometer – Wave V +Topic Report – Youth + +zens to say they are interested or very interested in politics. + +Youth disinterest in the political system could be in part due to their outlook +If young people perceive the most +on the fairness of parliamentary elections. +recent parliamentary elections as unfair, regardless of whether they voted or +were even eligible to vote in them, then their disinterest could perhaps reflect +disillusionment with their country’s political process. With the exception of the +Levant, youth are more likely than older age cohorts to evaluate the last par- +liamentary elections held in their respective country as being neither free nor +fair. In some countries surveyed this difference is stark. For instance, in Morocco +four-in-ten youth characterized the last parliamentary elections as such, com- +pared to fewer than a quarter of older Moroccans (23 percent). In Tunisia, youth +were 12 percentage points more likely than older citizens to characterize the +last parliamentary elections as being neither free nor fair (40 percent versus 28 +percent). + +Internet and Social Media + +Across the region, internet use, defined as going online at least once-in-a-few- +weeks, is substantially greater among youth than older generations. In fact, in +most of the surveyed countries (with the exception of Yemen), more than 90 +percent of youth use the internet, whereas among the older generation internet +usage does not approach such a rate. The highest rates of internet usage are +found in Lebanon (99 percent), followed by Jordan (96 percent), Palestine (95 +percent), Algeria (93 percent), Morocco (93 percent), Tunisia (91 percent), Egypt +(90 percent), Sudan (90 percent), Iraq (89 percent), and Libya (88 percent). The +lowest rate of internet usage among youth is seen in Yemen, where fewer than +eight-in-ten youth say they use the internet at least once every few weeks (77 +percent). The most striking generational differences in internet usage are seen in +Yemen and North Africa (excluding Libya), where internet use among youth ex- +ceeds that of older age cohorts by at least 30 percentage points. In contrast, the +intergenerational gap in internet use is less pronounced in the Levant (Lebanon, +Jordan, Palestine), where it does not exceed 20 percentage points. + +Interestingly, among internet users, social media usage is very high across all +demographics. In fact, youth are not so different from their older counterparts +in this regard. Though Yemeni youth report the lowest rates of internet usage in +the region, those online are also the most likely to engage in social media. Nearly +all Yemeni youth who are online (99 percent) use social media on a typical day. +However, most online youth across the region also engage in social media on a +typical day: Jordan (98 percent), Lebanon (97 percent), Iraq (97 percent), Pales- +tine (96 percent), Libya (96 percent), Egypt (96 percent), Tunisia (96 percent), +Morocco (96 percent), Sudan (94 percent), and Algeria (94 percent). Overall, +youth are consistently more likely than older generations to say they spend two + +www.arabbarometer.org 5 -Table 1 .2 . Esri Web GIS deployment patterns - -Online - -Hybrid - -On-premises or in cloud - -Esri managed - -ArcGIS Online - -ArcGIS Online - -Customer managed - -ArcGIS Enterprise - -ArcGIS Enterprise - -Although on-premises data center technology is not necessarily on the verge of extinction, cloud -computing is an option with many benefits, including scalability, agility, and cost efficiency. -During the COVID-19 pandemic, cloud usage increased rapidly because people use the cloud -platforms to work and shop from home. Today, more organizations are adopting ArcGIS Online -or deploying their ArcGIS Enterprise to the cloud, such as AWS, Microsoft Azure, Google -Cloud, and others (see figure). Many organizations adopt a hybrid model by integrating ArcGIS -Online and ArcGIS Enterprise. - -ArcGIS Online and ArcGIS Enterprise benefit from an effective framework to enforce security -and protect user privacy. They are certified as compliant with many federal and international -security and privacy standards. For more information, see the ArcGIS Trust Center page at -doc .arcgis .com/en/trust/compliance/compliance-tab-intro .htm. - -Built on ArcGIS Online, the Johns Hopkins COVID-19 Dashboard, a real-time pandemic tracker, has become -a global reference for the pandemic and one of the most cited resources to track the spread of coronavirus -around the world. It has received billions of views and remained highly efficient under a massive amount of -web traffic. - -Copyright © 2022 Esri. All rights reserved. - - 6 - -Chapter 1: Get started with Web GIS - -New-generation Web GIS architecture and trends - -ArcGIS Online and ArcGIS Enterprise represent the new generation of Web GIS. Although the -two have some differences (see chapter 5), they generally share similar architecture, similar capa- -bilities, and similar workflows for creating content, as shown in this figure. - -Desktop - -Web - -Device - -ArcGIS Online - -Portal for -ArcGIS - -Server - -Web services - -C - -l -i - -e -n -t - -a -p -p -s - -A -c -c -e -s -s - -S -e -r -v -i -c -e -s - -ArcGIS is a new-generation Web GIS that provides mapping, analysis, data management, and collaboration. - -This new-generation Web GIS architecture is described as follows: - -• At the center of the new Web GIS architecture is a portal—ArcGIS Online or Portal for - -ArcGIS—which represents a gateway for accessing all spatial products in an organization. -The portal helps organize, secure, and facilitate access to geographic information products. -• On the back end, the portal is powered by two components: GIS servers that allow you to -create content and ready-to-use content such as ArcGIS Living Atlas of the World, which -provides tens of thousands of data layers and maps. - -• On the client side, users can search, discover, and use the layers, maps, and apps on desktops, - -in web browsers, and on mobile devices anywhere, anytime. - -Technology evolution and trends in Web GIS -Since its inception, Web GIS has coevolved with geographic science and information technology -(see figure). This evolution and these trends are discussed in greater detail in later chapters. - -Copyright © 2022 Esri. All rights reserved. - New-generation Web GIS architecture and trends + Arab Barometer – Wave V +Topic Report – Youth + +hours or more on social media platforms, although the size of this difference is +often marginal at best. As an example, the widest generational gap is seen in +Morocco, where youth are 10 percentage points more likely to use social me- +dia. However, the difference is much smaller in Yemen (1 percentage point) and +Jordan (2 percentage points), and non-existent in Palestine. + +Though there is little generational difference in overall quantity of social media +usage, there exists a much larger generational discrepancy in terms of moti- +vations for using and trust in social media. Youth are more likely than older +generations to use social media as their primary source of information to fol- +low breaking news, and also to trust this information more than that found in +newspapers or TV news programs. With respect to the former, youth are most +likely to say social media is their primary source for breaking news in Palestine +(76 percent), while around two-thirds say the same in Morocco (67 percent) and +Jordan (63 percent). About half of youth or more in Egypt (58 percent), Tunisia +(55 percent), Iraq (51 percent), and Algeria (49 percent) rely primarily on so- +cial media for breaking news, while less than half do so in Libya (45 percent), +Lebanon (43 percent), Yemen (37 percent) and Sudan (32 percent). Notably, +youth reliance on social media to follow breaking news far exceeds that of their +older counterparts in every country surveyed. The greatest discrepancy between +youth and older generations is seen in Morocco (53 percentage points), Egypt +(44 percentage points), Tunisia (36 percentage points), and Palestine (32 per- +centage points). Elsewhere, though less stark, rates of youth reliance on social + +www.arabbarometer.org + +6 + + Arab Barometer – Wave V +Topic Report – Youth + +media for breaking news still exceed that of older citizens by at least 15 per- +centage points or more. Such trends show that while youth and their older +counterparts report using social media at similar rates overall, their incentives +and motivations for using such platforms differ in important ways. + +Youth in the region vary dramatically in the level of trust in information they re- +ceive on social media. The highest rates of trust among youth are found in Mo- +rocco (77 percent) and Egypt (74 percent), followed by Lebanon (67 percent), +Palestine (56 percent), Iraq (53 percent), and Sudan (51 percent). In contrast, +less than half of youth in Algeria (49 percent), Yemen (47 percent), Tunisia (47 +percent), Libya (41 percent), and Jordan (36 percent) trust social media sources +more than newspapers and TV news. The magnitude of the differences across +age cohorts is also varies widely, ranging from single digit differences in Jordan +(2 percentage points), Libya (5 percentage points), Yemen (3 percentage points), +and Sudan (3 percentage points) to much larger generational gaps in Lebanon +(16 percentage points), Egypt (18 percentage points), and Morocco (32 percent- +age points). Overall, however, youth are more trusting of the information they +receive from social media than are older citizens in every country surveyed. + +Notably, the most used social media platform among youth across the region is +Facebook. The exception to this finding is Yemen, where youth use WhatsApp at +a higher rate than Facebook (46 percent versus 36 percent). Youth in Egypt (94 + +www.arabbarometer.org 7 -Web GIS has coevolved with geographic information science and information technology. - -Web GIS has exemplified the following evolution of stages and trends in technologies: - -• From closed websites to open geospatial web services. Early Web GIS apps were developed -as independent websites. These websites were isolated from one another. It was difficult to -share information and functions between them and difficult to remix the content to create -apps. In the later 1990s, web services technology was conceived. Web services can be thought -of as building blocks that can be shared and remixed in versatile ways for building web apps. -The GIS industry adopted the concept of web services in manufacturing Web GIS products. -For example, ArcGIS Web GIS products fully support the web services architecture. The -server side provides ready-to-use services and allows users to create their own services, -whereas the client side can consume and combine these services to create apps. - -• From one-way to two-way information flow. Early Web GIS products and applications - -mainly supported one-way information flow, which was from server side to client side. Users -were merely the receivers of information. As we entered the 21st century, user-generated -content (UGC) became a significant phenomenon and created a reversed information flow, -from client side to server side. Volunteered geographic information (VGI) is the UGC of a -geospatial nature. For example, ArcGIS facilitates VGI through editable layers, Mobile GIS -apps, and browser-based apps. Users can view maps and perform queries, as well as conduct -field surveys, collect data, and report events they see. - -Copyright © 2022 Esri. All rights reserved. 8 - -Chapter 1: Get started with Web GIS - -• Portal technology is becoming essential. The word portal means gate or entrance. It was -adopted in the mid-1990s to form new terms such as web portal, referring to a website -that serves as the gateway to other websites or web resources. Geoportals are gateways to -geospatial information. Portals have become a core component of Web GIS technology. For -example, ArcGIS Online and Portal for ArcGIS have geoportal capabilities. They facilitate -the management, search, discovery, configuration, security, and remix of GIS data layers and -services. Today, portal collaboration allows different organizations to replicate selected con- -tent as hosts and guests, creating a portal of portals — a distributed Web GIS pattern. ArcGIS -Hub℠ and ArcGIS Enterprise Sites are examples of subject- or initiative-based portals. They -provide an easy-to-configure community engagement platform that organizes people, data, -and tools through information-driven initiatives. Organizations can create websites to share -data, documents, videos, and web maps on a certain subject or for a certain initiative. For -instance, the COVID-19 GIS Hub (coronavirus-resources .esri .com) provides GIS resources, -solutions, and a collaboration platform for the world to combat the coronavirus pandemic. -• Cloud GIS delivers GIS functionality and ready-to-use contents. Cloud GIS, which uses -public and private cloud computing to provide GIS software and contents, has become the -primary way to deliver GIS — not just functionality but also content. Cloud GIS advantages -include vast content and functionality, the flexibility of pay-as-you-go or subscription pric- -ing, reduced complexity, and increased availability of services. Because of these advantages, -cloud GIS, such as ArcGIS Online, has penetrated organizations that have not used GIS -before or been able to afford GIS on their own. - -• Mobile is becoming the pervasive Web GIS client platform. In the post-PC era, mobile - -devices have surpassed desktops and notebooks as the primary platform for accessing online -information. Mobile devices are a part of everyone’s life and work. “Mobile first” is a strategy -of many industries, including the Web GIS industry. Vendors have given extra attention to -Mobile GIS. For example, Esri provides numerous mobile-native apps and mobile-friendly -browser apps to support the needs of users and organizations for Mobile GIS. Mobile GIS -is also associated with many frontiers in Web GIS, such as augmented reality (AR). AR can -superimpose GIS data on top of a user’s camera views and thus augment a user’s sense of -reality. With the rapid advances in Mobile GIS, the vision of using GIS for anything, any- -time, anywhere, and by anyone is being realized faster than we can imagine. - -• Map visualization goes from 2D to 3D and virtual reality. With the increased client-side -graphics processing power and the broader support of WebGL (a JavaScript application pro- -gramming interface [API] that renders interactive 2D and 3D graphics in a browser), Web -GIS products such as ArcGIS Online and ArcGIS Enterprise can smoothly create and display - -Copyright © 2022 Esri. All rights reserved. New-generation Web GIS architecture and trends + Arab Barometer – Wave V +Topic Report – Youth + +percent), Algeria (91 percent), Morocco (90 percent), and Lebanon (90 percent) +are most partial to Facebook. More than eight-in-ten youth in Sudan (85 per- +cent), Libya (82 percent), and Jordan (81 percent), and more than seven-in-ten +youth in Iraq (74 percent) and Palestine (73 percent) use Facebook. + +Education and Employment + +Education System + +Rates of youth satisfaction with state education systems, both overall and as they +compare to older generations, vary greatly throughout the region. Nearly two- +thirds of youth in Palestine (65 percent) and Jordan (65 percent), and over half +of youth in Yemen (54 percent) are satisfied or completely satisfied. However, +fewer than half are satisfied in Sudan (43 percent) and Lebanon (39 percent), +and roughly a third are satisfied in Algeria (34 percent), Libya (31 percent), Egypt +(30 percent), and Tunisia (30 percent). The lowest rates of satisfaction are seen +among Iraqi youth (24 percent) and Moroccan youth (17 percent). Given this +wide variation, there seems to be limited regional consensus about the state +of education. However, one potential regional trend is that across most of the +countries surveyed youth are less satisfied with the education system than are +their older counterparts. +Indeed, in the majority of countries this discrepancy +is small, as seen in Palestine (1 percentage point), Sudan (2 percentage points), +Libya (2 percentage points), Algeria (3 percentage points), Yemen (3 percent- +age points), Iraq (4 percentage points), and Lebanon (5 percentage points). Yet +elsewhere, for instance in Morocco and Egypt, the intergenerational gap is more +sizable (22 percentage points and 10 percentage points, respectively). Excep- +tions to this trend are found only among Tunisian and Jordanian youth, both of +whom voice satisfaction at higher rates than do older citizens (30 percent versus +28 percent and 65 percent versus 59 percent, respectively). + +www.arabbarometer.org + +8 + + Arab Barometer – Wave V +Topic Report – Youth + +Looking to perceived corruption in the education system as one potential rea- +son contributing to low levels of satisfaction, in many countries more than half of +youth believe that rashwa (bribes) is required in order to receive better access to +education services. This belief is most commonly held in Egypt (64 percent), fol- +lowed by Lebanon (62 percent), Sudan (55 percent), Morocco (54 percent), Alge- +ria (53 percent), Yemen (53 percent), Iraq (52 percent), and Tunisia (50 percent). +However, while many youth believe that paying rashwa is necessary, this trend +is not common to all MENA youth as no more than a third of youth in Palestine +(33 percent), Libya (32 percent), and Jordan (29 percent) say the same regard- +ing rashwa to access better educational services. Prevalence of rashwa might +contribute to some dissatisfaction with state education systems, but it does not +account for or explain it in its entirety. + +Employment Opportunities + +Similar to satisfaction rates with state education systems, levels of youth satis- +faction with the government’s performance on creating employment opportu- +nities vary. However, despite this variation, across the region youth approval is +overwhelmingly weak, ranging from a low of five percent in Lebanon to a high +In addition to Lebanon, single-digit approval of +of only 32 percent in Yemen. +government performance on creating employment opportunities is seen in Iraq +(6 percent) and Algeria (9 percent). Fewer than two-in-ten youth in Morocco +(13 percent), Palestine (14 percent), Jordan (15 percent), Tunisia (16 percent), + +www.arabbarometer.org 9 -thematic and photo-realistic 3D web scenes. Representing a big step from 2D online maps, -3D web scenes provide users a more intuitive means to understand their data. Web scenes -are also important for indoor mapping, an ongoing trend in GIS. Even more intuitive than -3D, virtual reality (VR) allows users with certain visual wearables to immerse themselves in -3D city models by transporting themselves into static viewpoints and comparing different -urban design scenarios. This immersive experience brings GIS data and geospatial under- -standing even closer to users. The GIS community can play an important role in building the -metaverse and can assist with getting real work done in the virtual world environment. The -term metaverse refers to the fully immersive 3D expansion of existing internet technologies -to include AR and VR. - -• Data source goes from static to real time and spatiotemporal big data. Many elements in - -Web GIS are of a real-time nature, such as the incidents immediately reported by field crews, -citizens using mobile devices, concurrent measurements from sensor networks, and smart -cities. This massive amount of data presents challenges in real-time intake, processing, analy- -sis, visualization, and storage. ArcGIS GeoEvent™ Server and ArcGIS Velocity use cluster -computing and can ingest millions of sensor readings per second, process them, and store -them in real time. Such products and research allow Web GIS to meet the requirements of -the sensor web and the Internet of Things (IoT). - -• Web GIS is becoming more intelligent. Map visualization is the first step toward data -analysis. ArcGIS smart mapping can analyze the data automatically and suggest the best -mapping style and best defaults. This can help both expert and novice users quickly create -aesthetic and informative maps. Web GIS goes far beyond mapping. Web GIS can perform -big data analysis using distributed computing, aggregate data in the context of both space -and time, extrapolate new ideas from raw data, and bring superior intelligence to business -decisions. More recently, Web GIS has started to use artificial intelligence (AI), especially -machine learning and deep learning. For example, AI significantly improves the accuracy of -online imagery classification. AI can quickly identify the damaged locations from millions of -facilities’ photos and ensure that the damaged facilities are repaired early. - -Copyright © 2022 Esri. All rights reserved. 10 - -Chapter 1: Get started with Web GIS - -Web GIS information model: Users, web layers, maps, and - -apps - -Web GIS elements include users, groups, content items, and metadata (see figure). - -• Users with necessary privileges can create and join groups. -• Users sign in to create and share content items (with certain privileges), which include a - -large variety of data, web layers, tools, web maps, web scenes, and web apps. - -• Content items have metadata, including title, summary, description, categories, and tags, and -additional standards-based metadata. Metadata is indexed so users can search and discover -items more efficiently. - -• Users can keep information to themselves, share with certain groups (rather than individual -users), share with their organizations, or share with everyone — even the public if the organi- -zation allows anonymous access. This allows other users to see and access the items. - -own or join - -Users - -o -w -n - -organize and secure - -Groups - -x -e -d -n - -i - -index - -Items - -Metadata - -The ArcGIS Online and Portal for ArcGIS high-level information model. - -Main types of content items -Web GIS has five main types of content: data, layers, tools, web maps and scenes, and apps. Data -can be used to publish web layers; web layers can be used to create web maps and web scenes; -web maps and scenes and web tools can be used in apps or to create apps, as shown in the figure. - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +Sudan (17 percent), and Egypt (18 percent), and less than a quarter of youth in +Libya (22 percent) approve. + +Moreover, in addition to widespread dissatisfaction with government efforts to +improve employment opportunities, the vast majority of young Arab citizens +also believe that obtaining employment requires having strong connections. +Roughly nine-in-ten youth in Iraq (95 percent), Lebanon (94 percent), Jordan +(94 percent), Tunisia (94 percent), Libya (91 percent), Algeria (91 percent), Sudan +(90 percent), Palestine (90 percent), Egypt (90 percent), and Morocco (88 per- +cent) say that based on recent experience they think that obtaining employment +through wasta (personal connections) happens often or sometimes. Though +comparatively fewer, nearly seven-in-ten Yemeni youth (69 percent) hold the +same perception regarding the connection between wasta and employment. + +Emigration + +Given broad youth dissatisfaction with conditions in their countries, it is perhaps +not surprising that youth across the MENA region have a strong desire to em- +igrate. In six of eleven countries surveyed, roughly half or more of youth want +to leave their homeland. With the exception of Libya, North African and Jorda- +nian youth are most eager to migrate with about half or more citing a desire +to leave. Youth in Morocco cite not only the greatest desire to migrate, but also + +www.arabbarometer.org + +10 + + Arab Barometer – Wave V +Topic Report – Youth + +show the greatest discrepancy with older generation’s desire to do so (70 per- +cent versus 25 percent). Meanwhile, over half of youth in Sudan (62 percent), +Jordan (59 percent), Tunisia (56 percent) and Algeria (56 percent) also wish to +leave their country. Roughly a third of Libyan (31 percent), Yemeni (33 percent), +and Lebanese (35 percent) youth say the same. + +Regional trends are evident with regard to which countries youth are most ea- +ger to migrate. Over two-thirds of youth in French-speaking North Africa want +to migrate to a European country. This is not surprising given both the proximity +of their countries to Europe and, perhaps more importantly, the close linkages +between France and these countries. In contrast, six-in-ten youth in Egypt, and +roughly half of youth in Yemen (48 percent) and Sudan (45 percent), say they +want to emigrate to a GCC country, which is in line with historic patterns of labor +migration to this area. Interestingly, fewer than half of youth across the region +hope to migrate to North America. Jordanian youth are most likely to cite a +desire to migrate to North America (42 percent), whereas Yemeni, Libyan, and +Tunisian youth are least likely (12 percent, 13 percent, and 13 percent). + +The least preferred destination for youth is non-GCC MENA countries. Fewer than +a third of youth across the region have considered emigrating to such countries. +Lebanese and Tunisian youth are least eager (4 percent and 5 percent, respec- +tively) while Yemeni youth are most eager, though still fewer than three-in-ten +Yemeni youth cite an interest in emigration to a non-GCC MENA country (27 + +www.arabbarometer.org 11 -Tools - -configured -and used in - -published to - -added to - -Web, mobile, -and desktop apps - -Data - -Web layers - -Web maps -and scenes - -The main types of content in Web GIS and their relationships. - -These five main content types are described as follows: - -• Data: ArcGIS Online supports data in a variety of formats, including CSV, shapefiles, GPS - -Exchange Format (GPX), JavaScript Object Notation (GeoJSON), photos, imagery, geodata- -bases, and so on. - -• Layers: The Esri geospatial cloud can host layers, including feature layers, tiled layers, vector -tiles, map image layers, image layers, scene layers, CSV layers, tables, and Open Geospatial -Consortium (OGC) standard layers such as Keyhole Markup Language (KML), Web Map -Service (WMS), Web Map Tile Service (WMTS), and Web Feature Service (WFS). - -• Web maps and scenes: A web map, which is 2D, or a scene, which is 3D, comprises one or - -multiple layers and allows sophisticated layer configuration, including styles, pop-ups, labels, -and access permission. - -• Tools: Tools perform analytical functions, such as geocoding, routing, generating PDF files, - -summarizing data, finding hot spots, and analyzing proximity. - -• Apps: GIS apps are lightweight mapcentric programs that can be used on mobile devices, -in web browsers, or on desktops. Developers can program with ArcGIS web APIs to build -web apps. However, you do not have to be a developer to create a web app. ArcGIS provides -many ready-to-use web apps and mobile apps that can be used without programming. - -Apps: The face of Web GIS -Apps bring Web GIS to life. Web GIS end users directly interact with apps. The back-end data, -web layers, and server computation power of Web GIS become live and useful through apps. -ArcGIS provides ready-to-use apps out of the box and provides APIs and software development -kits (SDKs) for developers to create custom apps. - -Copyright © 2022 Esri. All rights reserved. 12 - -Chapter 1: Get started with Web GIS - -The ready-to-use apps include configurable apps (also referred to as ArcGIS Online and Portal -for ArcGIS web app templates), ArcGIS StoryMaps℠ stories, mobile apps, app builders, and -more. ArcGIS organizes these apps into bundles: - -• Essential apps bundle: ArcGIS StoryMaps, ArcGIS Experience Builder, ArcGIS Web -AppBuilder, ArcGIS Dashboards, instant apps, ArcGIS Hub Basic, Map Viewer, Scene -Viewer, and so on - -• Field apps bundle: ArcGIS Field Maps, ArcGIS Survey123, ArcGIS QuickCapture, and so on -• Office apps bundle: ArcGIS for Office, ArcGIS for SharePoint, and so on -• ArcGIS Pro -• ArcGIS Insights℠ -• Add-on apps: ArcGIS Drone2Map®, ArcGIS Navigator, ArcGIS Business Analyst™, and so on - -ArcGIS user types consist of Viewer, Storyteller, Editor, Field Worker, Creator, GIS -Professional, and Insights Analyst (see table 1.3). Different user types have included access to -different app bundles and apps. Apps that are not included with a user type can be acquired and -assigned as add-on licenses. To finish most of the tutorials in this book, a Creator user type with -a Publisher role is required. - -Table 1 .3 . Main ArcGIS user types, privileges, and included apps - -Type - -Viewer - -Storyteller - -Editor - -Privileges - -Can view items -Can’t create, edit, share, or analyze items - -Can create stories and express maps -Can’t create other types of content - -Can view and edit data -Can’t analyze, create, or share items - -Included apps - -Essential apps bundle - -Essential apps bundle - -Essential apps bundle - -Field Worker - -Same as Editor - -+ Field apps bundles - -Creator - -Can edit, create, share content, and do analysis - -+ Office apps bundles - -GIS Professional - -Same as Creator - -Insights Analyst - -Same as GIS Professional - -+ ArcGIS Pro - -Insights - -“+” means including the apps from the user type listed earlier in the table . - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +percent). + +Perceived Civil Liberties + +Of the countries surveyed by the Arab Barometer, Tunisia is the only one cate- +gorized as free by Freedom House, with the remainder of countries identified as +partly free or not free. As such, it is unsurprising that youth in Tunisia are more +confident in their civil liberties like freedom of expression and right to peaceful +protest than are youth in other surveyed countries. However, despite being com- +paratively more confident in the protection of their civil liberties, Tunisian youth +are still quite similar to youth in the rest of the region who report generally low, +though varied, confidence in both freedom of speech and right to peacefully +protest. + +Looking first to freedom of speech, Tunisian youth are indeed most confident in +their freedom of expression with over six-in-ten saying that it is guaranteed to a +great or medium extent in the country (63 percent). More than half of youth in +Jordan (56 percent) and Lebanon (51 percent) say the same. In the rest of the +region fewer than half of respondents feel their freedom of expression is guar- +anteed. About half of youth in Yemen (49 percent), roughly four-in-ten youth +in Palestine (42 percent), Egypt (40 percent), and Morocco (39 percent), and +roughly a third of youth in Algeria (35 percent) and Iraq (33 percent) say that + +www.arabbarometer.org + +12 + + Arab Barometer – Wave V +Topic Report – Youth + +freedom of expression is guaranteed to a great or medium extent. Least confi- +dent of their freedom of expression are youth in Sudan (25 percent) and Libya +(24 percent). + +When comparing youth perceptions of their freedom of speech to the percep- +tions of older age cohorts, there is no clear regional trend. For instance, youth +in Lebanon (+10 points), Jordan (+3 points), Yemen (+4 points) and Palestine +(+4 points) are more confident in their freedom of expression than are older +citizens. Yet, the opposite trend is apparent in Morocco (-16 points), Egypt (-5 +points), and Tunisia (-4 points). +In the remainder of countries surveyed (Libya, +Sudan, Iraq, and Algeria) youth and older citizens do not diverge by more than +a point and rate their freedom of expression similarly. + +Youth perception of their freedom to participate in peaceful protests is also low +across the region, never exceeding half of youth in any country. Though nearly +half of youth in Tunisia (49 percent), Lebanon (49 percent), Jordan (47 percent), +and Yemen (46 percent) feel their freedom to peacefully demonstrate is guaran- +teed to a medium or great extent, fewer youth hold this belief elsewhere in the +region. Four-in-ten Palestinian youth, a third of Moroccan youth, and three-in- +ten Iraqi youth are confident of their freedom to participate in peaceful demon- +strations. Meanwhile, roughly a quarter of youth in Algeria (27 percent), Libya +(24 percent), and Egypt (24 percent) and fewer than two-in-ten in Sudan (19 +percent) say the same. + +www.arabbarometer.org 13 -Basic components of a Web GIS app -Today’s best practices recommend that a GIS app have basemaps, operational layers, and tools -(see figure). ArcGIS supports this practice and allows you to create Web GIS apps. - -+ - -+ - -= - -Basemaps - -Operational -layers - -Tools - -Web GIS app - -The basic components of a Web GIS app. - -The components of a Web GIS app are described as follows: - -• Basemap layers: Basemaps provide a reference or context for your app. ArcGIS provides a -collection of fast-responding 2D and 3D basemaps (see figure). In addition, you can create -and use your own basemaps. - -ArcGIS provides a gallery of basemaps and an elevation service to support 2D and 3D web maps and apps. - -• Operational layers: Operational layers are theme layers that you and other users can use and -interact with. You can use existing layers from ArcGIS Online and ArcGIS Living Atlas of -the World as your operational layers. These layers span a range of subjects and can support - -Copyright © 2022 Esri. All rights reserved. 14 - -Chapter 1: Get started with Web GIS - -maps and apps of almost every subject. You can search these collections and discover layers -that fit your needs. You can also create and use your own operational layers. - -• Tools: Tools perform tasks beyond mapping, including common tasks such as query, geocod- -ing, routing, and more specialized tasks. ArcGIS Online provides extensive spatial analysis -tools for you to ask questions and solve spatial problems. ArcGIS Enterprise provides tools -for standard analysis, big data analysis, and raster analysis, and allows you to publish custom -web tools — in other words, geoprocessing services. - -Web layers or services: The building blocks of Web GIS -Web services technology is at the foundation of today’s Web GIS. The previously mentioned -three types of components of Web GIS apps are typically all based on web services. To make it -easier to understand, web services in ArcGIS are also called web layers or web tools. For exam- -ple, basemaps are often tile services or tile layers, operational layers are often feature services or -feature layers, and many tools are based on geoprocessing services or web tools. - -Early Web GIS was not based on web services; instead, it consisted mostly of stand-alone -websites in which GIS data and functions were available only to their own clients and couldn’t -be reused in other systems. This situation greatly limited the reusability of Web GIS resources. -In the late 1990s, web services technology was conceived. A web service is a program that runs -on a web server and provides programming interfaces for clients to consume over the web. Web -services have many advantages, such as flexibility in being reused and remixed in many web apps. -The GIS industry quickly adopted web services technology and reformed Web GIS products -based on web services architecture in the early 2000s. Today’s Web GIS products are designed to -support the publication, discovery, and use of GIS web services. A GIS service represents a GIS -resource — such as a layer, map, locator, or toolbox — that is located on the server and made avail- -able to web clients. For example, ArcGIS Living Atlas provides collections of ready-to-use GIS -services. The Esri geospatial cloud allows publishers to publish many types of GIS services and -allows users to discover and use GIS services in web maps, web apps, and mobile apps. - -This chapter introduces feature layers and services. Other chapters introduce raster tile ser- -vices, vector tile services, 3D scene services, map services, image services, stream services, geo- -processing services, geocoding services, and more. To support interoperability, ArcGIS supports -OGC standards such as WMS, WMTS, WFS, and KML. - -Copyright © 2022 Esri. All rights reserved. Web GIS information model: Users, web layers, maps, and apps + Arab Barometer – Wave V +Topic Report – Youth + +Attitudes on Gender + +Arab youth are divided in their views on gender equality. On the one hand, +most are supportive of equal rights for women, with the exception of inheri- +tance. However, there is substantially less support for women enjoying equal +roles in society. This is particularly evident in the political sphere. + +Throughout the region fewer than a third of youth say that a university education +is more important for men than for women. Support for university education +for men compared with women is highest in Yemen and Sudan (27 percent, +each), followed by roughly two-in-ten youth in Jordan (23 percent), Egypt (22 +percent), Algeria (20 percent), Iraq (19 percent), Tunisia (17 percent) and Libya +(17 percent). Even lower rates of support for men’s education taking precedence +over that of women are found in Palestine (14 percent), Lebanon (13 percent), +and Morocco (12 percent). + +Youth also generally hold liberal views on the question of whether women should +have equal rights in the decision to get a divorce. The exception to this trend +is in Sudan, where support for women’s equal rights in deciding to divorce falls +below five-in-ten (46 percent). Support is highest in Lebanon (87 percent), fol- +lowed by Iraq (81 percent), Jordan (80 percent), Tunisia (78 percent), Morocco +(78 percent), Palestine (76 percent), Libya (75 percent) and Algeria (72 percent). +Though still greater than half, lower rates of support are seen in Egypt (64 per- + +www.arabbarometer.org + +14 + + Arab Barometer – Wave V +Topic Report – Youth + +cent) and Yemen (55 percent). Notably, in most countries surveyed, the views +held by youth about women’s right to divorce are not significantly different from +the views held by older citizens. + +In a departure from predominantly liberal views on women’s rights, surprisingly +few youth in the region agree that women’s share of inheritance should be equal +to that of men. Nonetheless, Lebanon is a clear regional outlier to this trend with +over six-in-ten youth saying they agree or strongly agree that inheritance should +be equal between the sexes (62 percent). In contrast, far fewer youth elsewhere +are as supportive. Roughly three-in-ten youth in Iraq (30 percent) and Egypt (28 +percent), and around one-fourth of youth in Morocco (26 percent) and Tunisia +(24 percent) support equal inheritance. Still fewer youth in Algeria (18 percent), +Jordan (18 percent), Libya (16 percent), Sudan (13 percent), and Palestine (8 +percent) say the same. + +A plurality of youth in the region also believe that husbands should have the +final say in all family decisions. Support for this sentiment is greatest in Sudan +(73 percent), Algeria (73 percent), and Iraq (70 percent), followed by Egypt (66 +percent) and Libya (66 percent). Youth across Tunisia (56 percent), Yemen (55 +percent) and the Levant hold similar views to each other regarding women and +family decision-making. +In Lebanon and Jordan 55 percent of youth believe +husbands should have the final say in all family decisions, while in Palestine 50 +percent say the same. Though Moroccan youth are least likely to support limiting + +www.arabbarometer.org 15 -Hosted and nonhosted feature layers -Feature layers are the most used layer type for operational layers. Feature refers to vector data: -points, lines, and polygons. Feature layers can be hosted or nonhosted, depending on where the -underlying data is stored. - -• Hosted refers to a situation in which a web service is based on ArcGIS-managed data: for -instance, the data is stored in the ArcGIS Online cloud database (which is managed auto- -matically by Esri) or in ArcGIS Data Store (which is managed automatically by ArcGIS -Enterprise). - -• Nonhosted refers to a situation in which a web service connects directly to user-managed - -data: for instance, the data is stored in your organization’s own database, and the database is -managed by your own organization. - -Feature layers published to ArcGIS Online are always hosted feature layers. Feature layers pub- -lished in ArcGIS Enterprise can be hosted if the data is stored in ArcGIS Data Store or nonhosted -if your feature layers reference your organization’s own geodatabase — for example, SQL Server, -Oracle, or PostgreSQL. - -Hosted feature layers are often easier to publish because you don’t have to worry about the -underlying database. You don’t have to (and you won’t be able to) directly connect to the database -and manipulate it. To change the database schema and edit the data, you must go through the -hosted feature layer web interface. Nonhosted feature layers require that you have and manage -your own database. You can directly connect to the database and perform various operations -on it. - -This book teaches many ways to publish hosted feature layers—from CSV files, zipped file -geodatabases, ArcGIS Pro, and so on. You can also create empty feature layers and add fields in -ArcGIS Online (see chapter 4). - -Web maps and web scenes: The bridge between layers and apps -Web GIS provides Map Viewer and Scene Viewer for creators to integrate multiple layers, style -these layers, configure layer pop-ups, and make many other configurations. After a web map or -web scene is created, it can be used in one app or reused in many apps. This makes it easier for -creators to create apps. Without web maps and scenes, creators would often have to integrate the -layers directly and configure them in the apps, which can be cumbersome. - -Copyright © 2022 Esri. All rights reserved. 16 - -Chapter 1: Get started with Web GIS - -Paths to building Web GIS applications -The tutorials in this book teach readers how to build Web GIS apps. ArcGIS offers many paths to -this goal, as shown in this figure, which will be used throughout the book. - -ArcGIS offers many ways to build web apps. The green arrows in the figure highlight the workflows and -technology presented in this chapter. - -The figure presents the generic workflow to build Web GIS apps: - -• The data tier contains formats that range from simple CSV files managed with Microsoft - -Excel to sophisticated geodatabases managed with enterprise databases. - -• You can publish desktop resources to ArcGIS Online or ArcGIS Enterprise as web layers, - -tools, or other services. You can also use layers provided by ArcGIS Living Atlas. You can add -layers to web maps and web scenes and configure the styles and pop-ups in these layers. - -• On the client side, you can use various ready-to-use apps or custom apps. The former can be -configured without programming. The latter can be developed using various web APIs or -SDKs. - -Copyright © 2022 Esri. All rights reserved. \ No newline at end of file + Arab Barometer – Wave V +Topic Report – Youth + +women’s roles in family decision-making, nonetheless nearly four-in-ten say that +husbands should have the final say in all family decisions (38 percent). Notably, +though youth in Morocco are far less likely than older generations to support +this notion (-15), youth in Jordan (+6), Lebanon (+7), Libya (+5), and Tunisia (+2) +are more supportive than are their older counterparts. + +In every country surveyed, with the exception of Morocco, over half of youth agree +that, in general, men are better at political leadership than women. Roughly +eight-in-ten youth in Sudan (80 percent) and Jordan (78 percent), and roughly +three-quarters of youth in Algeria (74 percent) hold this belief. Across much of +the rest of the region youth hold this view at similar rates. For instance, around +seven-in-ten youth say men are better at political leadership in Egypt (70 per- +cent), Iraq (69 percent), Yemen (68 percent), Palestine (68 percent), and Libya +(67 percent). Meanwhile, fewer in Tunisia, Lebanon, and Morocco say that men +are better at political leadership than women (52 percent, 51 percent, and 31 +percent, respectively). + +Interestingly, despite unfavorably comparing female political leaders to men and +criticizing women’s role as decision-makers in the family, Arab youth are often +open both to the idea of a woman being a leader of a Muslim country and to +setting aside a certain percentage of elected positions for women as a means +of attaining fairer political representation. More than half of youth support the +notion of a female president or prime minister of a Muslim nation in all coun- + +www.arabbarometer.org + +16 + + Arab Barometer – Wave V +Topic Report – Youth + +tries surveyed, excluding Algeria (37 percent). Roughly half of youth in Sudan +(51 percent) and Libya (54 percent), six-in-ten youth in Yemen (57 percent), Jor- +dan (60 percent), and Palestine (64 percent) and roughly seven-in-ten youth in +Tunisia (67 percent), Egypt (68 percent) and Iraq (69 percent) affirm that women +can become prime ministers or presidents of Muslim countries. Moroccan and +Lebanese youth (77 percent and 79 percent, respectively) are the most accept- +ing of this right. + +The majority of MENA youth also believe that in order to achieve fairer represen- +tation a certain percentage of elected positions should be set aside for women. +Support for this right exceeds half of all youth in all countries surveyed. Youth in +Iraq and Lebanon (79 percent, each) are most supportive, followed by Sudan (76 +percent), Palestine (75 percent), Jordan (72 percent) and Yemen (69 percent). +Excluding Sudan, North African youth show the least support for women’s quo- +tas in elected positions with less than two-thirds of youth in Egypt (66 percent), +Libya (65 percent), Tunisia (63 percent), Morocco (62 percent) and Algeria (53 +percent) saying they agree or strongly agree with this idea. + +International Relations + +In over half of the countries surveyed, youth are most supportive of stronger ties +with China and least interested in stronger ties with the United States among +global powers. Youth in Jordan (70 percent), Tunisia (65 percent), Sudan (62 +percent), Libya (61 percent), Yemen (50 percent), Palestine (49 percent), and +Iraq (47 percent) prefer strengthening ties with China at a higher rate than they +do for ties with the United States, the United Kingdom, or Russia. Exceptions to +this include youth in Morocco, who have a slightly stronger preference for closer +ties with the United Kingdom (63 percent), youth in Lebanon and Algeria, who +are more supportive by a small margin of stronger ties with Russia (38 percent +and 42 percent, respectively), and Egyptian youth, who are equally supportive +of strengthening ties with the United States and Russia (41 percent, each). + +In a majority of countries surveyed youth support for stronger ties with the United +States not only falls below 50 percent, but is also lower than youth support for +strengthening ties with other world powers. Exceptions to the former are Sudan +(60 percent), Morocco (57 percent), and Jordan (55 percent) where youth sup- +port exceeds half. In contrast, fewer than half of youth in Tunisia (47 percent), +and roughly a third of youth in Libya (36 percent), Iraq (34 percent), Palestine, +Lebanon, and Algeria (31 percent, each) say that relations with the United States +should become stronger. Of all countries surveyed Yemeni youth show not only +the least interest in closer ties with the United States (22 percent), but they are +also least interested in strengthening ties with Russia (26 percent) relative to +youth in other countries. + +www.arabbarometer.org + +17 + + Arab Barometer – Wave V +Topic Report – Youth + +Looking to regional powers, one clear trend is that Arab youth are generally less +interested in stronger relations with Iran and more interested in improving re- +lations with Turkey. Though support for stronger ties with Iran never exceeds +four-in-ten youth in any country surveyed, support is highest in Lebanon (38 per- +cent), followed by Sudan (36 percent), Palestine (34 percent), Iraq (32 percent) +and Tunisia (31 percent). In Lebanon and Iraq, sectarian divides might explain +the comparatively greater support for strengthening ties with Iran, a country +with a strong Shi’ite majority. Roughly two-in-ten youth in Morocco (22 per- +cent), Jordan (21 percent), Yemen (20 percent), Algeria (20 percent), and Libya +(18 percent) prefer stronger relations with Iran, while only 8 percent of Egyptian +youth say the same. + +www.arabbarometer.org + +18 + + Arab Barometer – Wave V +Topic Report – Youth + +In comparison to other regional powers (Saudi Arabia, Qatar, and Iran), Arab +youth tend to prefer strengthening relations with Turkey. Youth in Jordan (82 +percent), Palestine (75 percent), Tunisia (64 percent), Yemen (61 percent), Mo- +rocco (56 percent), Algeria (53 percent), and Iraq (45 percent), are all more likely +to want improved relations with Turkey than to do so for relations with Saudi Ara- +bia, Qatar, or Iran. Youth in Sudan prefer strengthening relations with Saudi Ara- +bia, Turkey, and Qatar at similar rates (70 percent, 73 percent, and 74 percent, +respectively), as do youth in Lebanon (41 percent, 41 percent, and 42 percent, +respectively). In Libya and Egypt youth by far prefer improving ties with Saudi +Arabia (55 percent and 58 percent). + +www.arabbarometer.org + +19 + + About Arab Barometer + +The Arab Barometer is a nonpartisan research network that +provides insights into the social, political, and economic attitudes +and values of ordinary citizens across the Arab world. + +We have been conducting rigorous, and nationally representative +face-to-face public opinion surveys on probability samples of the +adult populations across the Arab world since 2006. The margin of +error is ±3 percent. + +The Arab Barometer is the largest repository of publicly available +data on the views of men and women in the MENA region. Our +findings give a voice to the needs and concerns of Arab publics. + +.......... + +.......... + +.......... + +ARABBAROMETER.ORGARABBAROMETER@ARABBAROMETER \ No newline at end of file diff --git a/q159/random_k4/random_2.md b/q159/random_k4/random_2.md index 482f9de9b0ff67f242f61e09321d1c78cb243915..6748665a7c835db276023c66e67f49b0fa9cb861 100644 --- a/q159/random_k4/random_2.md +++ b/q159/random_k4/random_2.md @@ -1,410 +1,493 @@ -IEEEInternationalConferenceonDataMining(ICDM),WorkshoponUserModelingandRecommendation(UMRec),Washington,DC,USA,2025 -| Effectiveness | | | | | of LLMs | | in | Temporal | | | User | | Profiling | | | -| ------------- | --- | --- | ------ | -------------------- | ------- | -------------- | --------------------- | ----------- | ----------- | ---------------------- | --------- | --- | --------- | --- | --- | -| | | | | | for | Recommendation | | | | | | | | | | -| | | | Milad | Sabouri∗, | | Masoud | Mansoury†, | Kun | Lin∗, | Bamshad | Mobasher∗ | | | | | -| | | | | | | | ∗DePaul University, | | USA | | | | | | | -| | | | Email: | msabouri@depaul.edu, | | | klin13@depaul.edu, | | | mobasher@cs.depaul.edu | | | | | | -| | | | | | †Delft | University | of | Technology, | Netherlands | | | | | | | -| | | | | | | Email: | m.mansoury@tudelft.nl | | | | | | | | | -Abstract—Effectively modeling the dynamic nature of user 17% in Recall@10 and 14% in NDCG@10, benefits are -| preferences | is crucial | | for enhancing | | recommendation | | accuracy | | | | | | | | | -| ----------- | ---------- | --- | ------------- | --- | -------------- | --- | -------- | ---- | ---------- | --- | ------- | ------------ | --- | ---------- | ------ | -| | | | | | | | | less | pronounced | in | sparser | environments | | like Video | Games. | -5202 tcO 13 ]RI.sc[ 1v67100.1152:viXra andfosteringtransparencyinrecommendersystems.Traditional -| | | | | | | | | Our | analysis | of these | results | leads | us | to hypothesize | that | -| -------------- | ----- | --------- | --- | --------------- | --- | ------- | ---------- | ---- | -------- | ---------- | ------- | ------------ | ------- | -------------- | ------ | -| user profiling | often | overlooks | | the distinction | | between | transitory | | | | | | | | | -| | | | | | | | | this | nuanced | capability | is | particularly | evident | where | short- | -short-terminterestsandstablelong-termpreferences.Thispaper -examines the capability of leveraging Large Language Models term and long-term preferences are more clearly separable -(LLMs) to capture these temporal dynamics, generating richer (e.g., Movies&TV), versus domains with more stable user -| user representations | | through | | distinct | short-term | and | long-term | | | | | | | | | -| -------------------- | --- | ------- | --- | -------- | ---------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -profiles(e.g.,VideoGames).Thishighlightsacriticaltrade-off -textualsummariesofinteractionhistories.Ourobservationssug- -betweenLLM-enhancedpredictivepowerandassociatedcom- -gestthatwhileLLMstendtoimproverecommendationqualityin -| | | | | | | | | putational | | costs, suggesting | | context-dependent | | application. | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ----------------- | --- | ----------------- | --- | ------------ | --- | -domainswithmoreactiveuserengagement,theirbenefitsappear -less pronounced in sparser environments. This disparity likely Also,beyondpredictiveaccuracy,thisLLM-drivenapproach -stemsfromthevaryingdistinguishabilityofshort-termandlong- inherently offers an intrinsic potential for transparency. The -| term preferences | | across | domains; | the | approach | shows | greater | | | | | | | | | -| ---------------- | --- | ------ | -------- | --- | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -naturallanguageprofilesandlearnedattentionweightsprovide -utilitywherethesetemporalinterestsaremoreclearlyseparable -| | | | | | | | | a pathway | | toward interpretability, | | | conveying | whether | recom- | -| ----------------- | ------------ | -------- | --- | ---------- | ---------- | ---------- | ----------- | ---------- | --- | ------------------------ | ------ | --- | --------- | ---------- | ------ | -| (e.g., Movies&TV) | | compared | | to domains | with | more | stable user | | | | | | | | | -| | | | | | | | | mendations | | stem from | recent | or | enduring | interests. | While | -| profiles | (e.g., Video | Games). | | This | highlights | a critical | trade- | | | | | | | | | -off between enhanced performance and computational costs, fully realizing user-facing explanations is a future direction, -suggesting context-dependent LLM application. Beyond predic- this work contributes insights into the practical capability -| tive capability, | this | LLM-driven | | approach | inherently | | provides | | | | | | | | | -| ---------------- | --------- | ---------- | -------------------- | -------- | ---------- | ---- | ----------- | ----------- | --------- | ---------------- | -------- | ------------- | ---------- | ------------ | ---- | -| | | | | | | | | and | inherent | interpretability | | of LLM-driven | | temporal | user | -| an intrinsic | potential | | for interpretability | | through | | its natural | | | | | | | | | -| | | | | | | | | profiling, | outlining | new | research | | directions | for adaptive | and | -| language | profiles | and | attention | weights. | This | work | contributes | | | | | | | | | -| | | | | | | | | transparent | | recommender | systems. | | | | | -insightsintothepracticalcapabilityandinherentinterpretability -| of LLM-driven | | temporal | user | profiling, | outlining | new | research | | | | | | | | | -| ------------- | --- | -------- | ---- | ---------- | --------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -II. RELATEDWORK -directionsfordevelopingadaptiveandtransparentrecommender -systems. Modeling the dynamic nature of user preferences is a -IndexTerms—TemporalUserProfiling,LargeLanguageMod- -| | | | | | | | | long-standing | | challenge | in | recommender | | systems. | Traditional | -| --- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --------- | --- | ----------- | --- | -------- | ----------- | -els, User Modeling -| | | | | | | | | methods, | such | as averaging | | item embeddings | | [3], | often fail to | -| --- | --- | --- | ------------ | --- | --- | --- | --- | -------- | ---------- | ------------ | ----------- | --------------- | --- | ------------ | ------------- | -| | | | | | | | | capture | the | evolving | distinction | between | | short-term | and long- | -| | | I. | INTRODUCTION | | | | | | | | | | | | | -| | | | | | | | | term | interests. | More | advanced | approaches | | for temporal | and | -Effective recommender systems demand both accurate and sequential recommendation include time-aware matrix factor- -transparent suggestions, a challenge exacerbated by the dy- ization (e.g., TimeSVD++ [4]), and sequential deep learning -namic nature of user preferences. Traditional user profiling, models like GRU4Rec [5] and SASRec [6], which learn user -such as averaging item embeddings, often oversimplifies user dynamics from interaction sequences. Recent work has also -interests by conflating fleeting short-term desires with sta- explored long short-term preference modeling for continuous- -ble long-term tastes, hindering both recommendation quality time sequential recommendation [7]. -and interpretability. This paper examines the capability of Research in recommender systems increasingly emphasizes -leveraging Large Language Models (LLMs) to capture these balancing predictive accuracy with interpretability. Early ef- -temporaldynamics.Ourapproachgeneratesdistinctshort-term forts integrated explanations through content summarization -and long-term textual summaries of user interaction histories, or leveraged user-generated reviews [8]. More recently, Large -which are then encoded via BERT [1] and adaptively fused LanguageModels(LLMs)[9],[10]haveemergedaspowerful -using an attention mechanism [2] to form a unified user tools for generating richer explanations, demonstrating their -representation. capacity to enhance interpretability through personalized nar- -Our observations reveal a nuanced capability of this LLM- rativesandcollaborativeadaptors[11]–[14].Knowledgegraph -driven approach across varying user engagement patterns. (KG)-based approaches [15] have also integrated structured -While it tends to improve recommendation quality in ac- side information for explanations, sometimes combined with -tive domains like Movies&TV, with improvements such as LLM outputs [15]–[18]. - -| | | Fig.1. | Pipeline | for | LLM-Driven | Temporal | | User Profile | Generation | and | Recommendation. | | | | | -| --- | --- | ------ | -------- | --- | ---------- | -------- | --- | ------------ | ---------- | --- | --------------- | --- | --- | --- | --- | -Ourworkintersectsthesetwocriticalareasbyinvestigating (i) LLM-based Temporal Profile Generation: To capture -the utility of leveraging LLMs to explicitly model temporal temporal aspects of user preferences, our approach prompts -user dynamics for enhanced content-based recommendations the LLM twice over each user’s interaction history H , using -u -and inherent transparency. Unlike existing methods that pri- distinct instructions to generate two textual summaries. The -marily use static profiles or sequential models without natural first focuses on recent interactions, producing the short-term -| language | grounding | [6], | [19], | [20], this | study | explores | an | profile: | | | | | | | | -| -------------- | -------------- | --------- | -------- | ---------- | -------- | ------------- | --- | ---------- | ---------- | ------- | ---------- | -------------- | --- | ------ | -------- | -| approach | that generates | | distinct | natural | language | summaries | | | | | | | | | | -| | | | | | | | | | | NLshort | =LLM(H | ,Promptshort). | | | (2) | -| | | | | | | | | | | u | | u | | | | -| for short-term | and | long-term | user | behaviors. | By | encoding | and | | | | | | | | | -| adaptively | fusing | these | semantic | summaries, | | this approach | | | | | | | | | | -| | | | | | | | | The second | summarizes | | persistent | behaviors | | across | the full | -aims to create intrinsically interpretable user profiles. This history, forming the long-term profile: -| unique combination | | transparently | | highlights | the | temporal | ra- | | | | | | | | | -| -------------------- | --------- | ---------------- | ------------ | ----------- | -------- | -------- | -------- | --------------- | ------- | ----------- | ---------------- | ------------- | ------------ | ------------- | ------ | -| tionale behind | | recommendations, | | bridging | advanced | | LLM- | | | | | | | | | -| | | | | | | | | | | NLlong | =LLM(H | ,Promptlong). | | | (3) | -| | | | | | | | | | | u | | u | | | | -| based explainability | | with | nuanced | temporal | dynamics | | of user | | | | | | | | | -| interests. | This work | provides | | preliminary | insights | | into the | | | | | | | | | -| | | | | | | | | These summaries | | enhance | interpretability | | by | separating | tran- | -| practical | utility | of this novel | integration. | | | | | | | | | | | | | -| | | | | | | | | sient and | stable | user | interests. | | | | | -| | | | | | | | | To provide | | a concrete | understanding | | of | our LLM-based | | -| | | III. | METHODOLOGY | | | | | | | | | | | | | -| | | | | | | | | temporal | profile | generation, | | we include | a conceptual | | illus- | -Let U = u ,u ,...,u be users, and I = i ,i ,...,i tration of the prompts and their corresponding output in -| | 1 | 2 | |U| | | | 1 2 | |I| | | | | | | | | | -| --------- | ---- | -------- | ----------- | ---- | -------- | --- | -------- | --------- | --- | ------- | ------------- | -------- | --- | --------- | -------- | -| be items. | Each | user u ∈ | U interacts | with | a subset | | of items | | | | | | | | | -| | | | | | | | | Figure 3. | The | prompts | are carefully | designed | | to elicit | distinct | -I u ⊆ I, with interactions associated to timestamps. We temporal summaries. As shown, the ”Long-term Preference -| define the | interaction | history | as | Hu=(i,t | ):i∈I | | sorted | | | | | | | | | -| ---------- | ----------- | ------- | --- | ------- | ----- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -u,i u Prompt” emphasizes identifying enduring interests and con- -chronologically. -| | | | | | | | | sistent themes | | across | the user’s | entire | history. | In | contrast, | -| --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | ------ | ---------- | ------ | -------- | --- | --------- | -∈Rd) -We aim to learn embeddings for users (e u and items the ”Short-term Preference Prompt” is engineered to focus -| (e ∈Rd) | to predict | user-item | | interaction | likelihood: | | | | | | | | | | | -| ------- | ---------- | --------- | --- | ----------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -i specifically on recent behaviors and fleeting interests. While -thefull,detailedpromptsandtheirconfigurationsareavailable -| | | yˆ | =f(e | ,e ), | | | (1) | | | | | | | | | -| --- | --- | --- | ---- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | u,i | | u i | | | | | | | | | | | | -inourpublicrepository(seesectionIV)toensurereproducibil- -wheref isamulti-layerperceptron(MLP).Thepipeline(Fig- ity, these examples demonstrate the core strategy of using -ure 1) comprises three steps: (i) profile generation via LLMs, natural language instructions to temporally disentangle user -| (ii) semantic | embedding, | | and (iii) | interaction | prediction. | | | preferences. | | | | | | | | -| ------------- | ---------- | --- | --------- | ----------- | ----------- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | - -Fig. 2. A conceptual illustration of the framework’s potential for enhancing transparency. (A) and (B) show the short-term and long-term -textual profiles generated and encoded by our current approach. (C) shows a hypothetical extension wherein the final recommendation is -explicitly justified by both sets of preferences, further strengthening user-facing transparency—an aspect we plan to explore in future work. -TABLEI -DATASETSSTATISTICS -| | | | | | | | | Dataset | Users | Items | | Interactions | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --------- | ------ | ------ | ----- | ------------ | --- | --- | -| | | | | | | | | Movies&TV | 10,000 | 14,420 | | 202,583 | | | -| | | | | | | | | Games | 10,371 | | 3,790 | 83,842 | | | -TABLEII -USERPROFILESSTATISTICS -| | | | | | | | Dataset | | Mean | Median | Mode | | StdDev | | -| --- | --- | --- | --- | --- | --- | ---------- | ----------- | ----------- | -------- | --------- | --------- | ------ | ---------- | ------ | -| | | | | | | | Movies&TV | | 11.79 | 9.00 | | 6 | 9.80 | | -| | | | | | | | Games | | 4.55 | 3.00 | | 3 | 3.97 | | -| | | | | | | Crucially, | | the learned | weights | | (αshort, | αlong) | explicitly | con- | -| | | | | | | | | | | | u | u | | | -| | | | | | | vey | the model’s | | decision | rationale | regarding | | recent | versus | -| | | | | | | historical | user | interests. | | | | | | | -(iii)InteractionPrediction:Topredictinteractionprobabili- -| | | | | | | ties, | we concatenate | | user embedding | | e | with item | embedding | | -| --- | --- | --- | --- | --- | --- | ----- | -------------- | --- | -------------- | --- | --- | --------- | --------- | --- | -u -| | | | | | | e and | process | through | an | MLP | [21]: | | | | -| --- | --- | --- | --- | --- | --- | ----- | ------- | ------- | --- | --- | ----- | --- | --- | --- | +WHITE PAPER + +In-Store +Shopper +Engagement +Built for the +Future of Retail + + There is an opportunity to +reinvent the moment of truth … + +is your store +ready? + + In-Store Shopper Engagement Built for the Future of Retail + +Introduction + +Now that the offline market is reopening, +it is time for retail brands to upgrade their +brick-and-mortar stores and deliver a +smart retail experience for customers. + +Gartner1 + +Retail Industry Drivers Impacting the In-Store +Experience + +• Consumers have returned to retail stores. And they aren’t just ready to get back to in- +person shopping, they have also accelerated their adoption of new technologies and +expect similar levels of digital technology and content in the physical store as they do in +the digital world. + +• After reporting the results of a recent consumer survey, Gartner concluded, “To maximize +the ROI on digital and capitalize on consumers’ interests in hands-free technology and +physical stores, retail brands should pursue a smart retail strategy and offer an +immersive omnichannel experience for their customers.”1 + +• That’s why in this white paper, we’ll dive into the trends that are impacting the need for a +more content-driven consumer experience, explore what’s possible in the store, and the +business benefits of creating a more memorable shopper journey with 6 ways to stand +out from the crowd. + +3 + + In-Store Shopper Engagement Built for the Future of Retail + +Consumers are Craving More Content, Recipes +and Deals +Analysis by Insider Intelligence, reports that retail sales are anticipated to grow between +4% and 5% each year through 2025. That’s a projection to hit $31.3 trillion in sales.2 + +Brick-and-Mortar vs. eCommerce Sales in US for 2021 + +) +s +n +o + i -| Fig.3. Examples | of | Prompts | for LLM-based | Temporal | User Profile | | | | | | | | | | -| --------------- | --- | ------- | ------------- | -------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(cid:0) (cid:1) -| Generation (Movies&TV | | Domain) | | | | | | | yˆ =MLP | | [e ;e ] | , | | (9) | -| --------------------- | --- | ------- | --- | --- | --- | ---- | ------ | ------------- | ---------- | --- | ----------- | --- | ----------- | --- | -| | | | | | | | | | u,i | | u i | | | | -| | | | | | | with | output | yˆ u,i ∈[0,1] | indicating | | interaction | | likelihood. | The | -(ii)SemanticEmbeddingandAttentionFusion:Eachtextual model is trained via binary cross-entropy loss: -| profile is transformed | | into | a high-dimensional | | embedding via | | | | | | | | | | -| ---------------------- | --- | ---------------------- | ------------------ | --- | ------------- | --- | --- | -------- | --------- | ----- | ----- | ---- | ------- | ---------------- | -| BERT [1]: | | | | | | | 1 | (cid:88) | (cid:104) | | | | (cid:0) | (cid:1)(cid:105) | -| | | | | | | L=− | | | y | logyˆ | +(1−y | )log | 1−yˆ | . | -| | | rshort =BERT(NLshort), | | | | | | | u,i | u,i | | u,i | | u,i | -(4) |D| -| | | u | | u | | | | (u,i,yu,i)∈D | | | | | | | -| --- | --- | -------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | ---- | -| | | rlong =BERT(NLlong), | | | | | | | | | | | | (10) | -(5) -u u where D is the training set and y the observed interactions. -u,i -| yielding embeddings | | that | semantically | represent | short-term | | | | | | | | | | -| ------------------- | --- | ---- | ------------ | --------- | ---------- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -| | | | | | | | | | IV. EXPERIMENTS | | | | | | -andlong-terminterests,respectively.Todynamicallyintegrate -| | | | | | | We | evaluate | the | effectiveness | | of the | LLM-driven | | tem- | -| -------------------- | ----------- | --------- | ---- | ---------- | ------------------- | ----- | -------- | --------- | ------------- | --- | ------ | ---------- | ------ | ---- | -| these temporal | embeddings, | | we | apply a | learnable attention | | | | | | | | | | -| | | | | | | poral | user | profiling | framework | | on | two | Amazon | do- | -| layer [2], producing | | a unified | user | embedding: | | | | | | | | | | | -mains[22]—Movies&TVandVideoGames—chosenfortheir -| | | | exp(W | rs hort) | | | | | | | | | | | -| --- | --- | --- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -a u differing user profile density and behavioral variability. This -| αshort | = | | | | , (6) | | | | | | | | | | -| ------ | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -u rshort)+exp(W rlong) allows assessment under diverse recommendation conditions. -| | | exp(W | a | | a u | | | | | | | | | | -| --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -u -| | | | | | | Dataset | statistics | | are in Tables | I | and II. | For LLM | processing, | | -| --- | --- | ----- | ---------- | --- | --- | ------- | ---------- | --- | ------------- | --- | ------- | ------- | ----------- | --- | -| | | αlong | =1−αshort, | | | | | | | | | | | | -u u (7) we retain English item descriptions exceeding 500 characters. -where W ∈R1×d is a learnable parameter vector. The final For benchmarking, we compare against four baselines: Cen- +l +l +i + +M +$ +( + +l + +s +e a -| | | | | | | tric | [3], which | averages | item | embeddings | | without | | temporal | -| --------- | --- | --- | --- | --- | --- | --------- | ---------- | ----------- | ---- | ---------- | --------- | -------- | --- | --------- | -| embedding | is: | | | | | | | | | | | | | | -| | | | | | | modeling. | | Temp-Fusion | is | also | included, | designed | | to assess | -=αshort·rshort+αlong·rlong. -e u (8) temporal fusion’s benefit without LLM-generated semantic -| | | u | u | u | u | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -TABLEIII -COMPARATIVEPERFORMANCEEVALUATIONOFLLM-DRIVENTEMPORALUSERPROFILINGACROSSDIFFERENTDOMAINS -asteriskdenotesastatisticallysignificantimprovementoverthebaseline(Centric)(p<0.05). -| Method | | | | | | Movies&TV | | | | | Video | Games | | | -| ------ | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | ----- | ----- | --- | --- | -Recall@10 NDCG@10 Recall@20 NDCG@20 Recall@10 NDCG@10 Recall@20 NDCG@20 -Centric 0.0113 0.0191 0.0199 0.0269 0.0645 0.0532 0.0932 0.0649 -Popularity 0.0082 0.0145 0.0133 0.0191 0.0397 0.0324 0.0706 0.0453 -| MF | | | 0.0048 | | 0.0085 | | 0.0087 | 0.0124 | 0.0457 | | 0.0370 | 0.0754 | | 0.0491 | -| --- | --- | --- | ------ | --- | ------ | --- | ------ | ------ | ------ | --- | ------ | ------ | --- | ------ | -Temp-Fusion 0.0118 0.0201 0.0207 0.0276 0.0693 0.0589 0.0982 0.0712 -LLM-TP 0.0132∗ 0.0217∗ 0.0223∗ 0.0293∗ 0.0665∗ 0.0547∗ 0.1021∗ 0.0683∗ -| GainofLLM-TPvs.Centric | | | 17% | | 14% | | 12% | 9% | 3% | | 3% | 10% | | 5% | -| ---------------------- | ----------- | ----------- | ---------- | --------- | ------ | ---------- | --------- | --- | --- | --- | --- | --- | --- | --- | -| profiles; | it segments | interaction | | histories | into | short-term | and | | | | | | | | -| long-term | numerical | item | embeddings | | (e.g., | by | averaging | | | | | | | | -withinrecent/historicalwindows)andfusesthemviaattention, -isolatingtheLLM’stextualcontribution.Twostandardcontrol -| baselines | are Popularity | | [23], | a non-personalized | | | ranking | | | | | | | | -| -------------- | -------------- | --- | ------------- | ------------------ | --------- | ---------- | ------------ | --- | --- | --- | --- | --- | --- | --- | -| approach, | and Matrix | | Factorization | | (MF) | [24], | a collabora- | | | | | | | | -| tive filtering | technique | | without | textual | features. | Evaluation | | is | | | | | | | -performedusingarigorousper-usertemporalholdoutprotocol. -| For each | user, we | first | chronologically | | sort | their | interactions | | | | | | | | -| ------------- | ------------- | --------- | --------------- | -------- | ----------- | ------- | ------------ | --- | --- | --- | --- | --- | --- | --- | -| by timestamp. | The | dataset | is | then | split | so that | a user’s | | | | | | | | -| training | data always | precedes | | their | validation | and | test data. | | | | | | | | -| This ensures | that | our model | is | trained | only | on past | behaviors | | | | | | | | -| to predict | future | ones, | thereby | strictly | preventing | | within-user | | | | | | | | -| temporal | data leakage. | | We report | model | performance | | using | | | | | | | | -standardmetrics,includingRecall@KandNDCG@K.Textual -summaries are encoded via SBERT [25] (MiniLM-L6-v2, Fig.4. RelativegainsinRecall@20andNDCG@20ofthefullmodel -384-dim), with profiles generated using GPT-40-mini [26]. over ablation variants on the Movies&TV dataset. -| Interaction | likelihoods | | are predicted | | using | an MLP | (hidden | | | | | | | | -| ----------- | ----------- | --- | ------------- | --- | ----- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | -size128,dropout0.2)trainedwithbinarycross-entropy,batch -size 2048, Adam optimizer, and early stopping (patience=5), smaller K-values. This outcome likely arises from interaction -| executed | on four | NVIDIA | A100 | GPUs. | All | code, | data, and | | | | | | | | -| -------- | ------- | ------ | ---- | ----- | --- | ----- | --------- | --- | --- | --- | --- | --- | --- | --- | -sparsity;ourconjectureisthatloweruseractivityoftenleadsto -| prompts | are available | in | a public | GitHub | repository1. | | | | | | | | | | -| ---------- | -------------- | --- | -------- | ------ | ------------ | --- | --- | ---------------- | ---- | ---------- | ------------ | ---------- | --------- | ------------- | -| | | | | | | | | comparatively | more | stable | preferences, | limiting | | the incremen- | -| | | | | | | | | tal advantage | of | detailed | temporal | modeling | through | textual | -| A. Results | and Discussion | | | | | | | | | | | | | | -| | | | | | | | | differentiation. | This | highlights | | a critical | trade-off | between | -Thissectionpresentstheempiricalresultsofourassessment, enhancedpredictivepowerandassociatedcomputationalcosts, -| summarized | in Table | III, | highlighting | | key | insights | into the | | | | | | | | -| ---------- | -------- | ---- | ------------ | --- | --- | -------- | -------- | ---------- | ----------------- | --- | --- | ------------ | --- | --- | -| | | | | | | | | suggesting | context-dependent | | LLM | application. | | | -capabilityoftheLLM-driventemporaluserprofilingapproach. -| | | | | | | | | Intrinsic | Explainability | | Potential: | | The framework | in- | -| --- | --- | --- | --- | --- | --- | --- | --- | --------- | -------------- | --- | ---------- | --- | ------------- | --- | -TemporalProfilingBenefitsDomainswithHighUserAc- -| | | | | | | | | herently | supports | interpretability | | (Figure | 2). LLM-generated | | -| --- | --- | --- | --- | --- | --- | --- | --- | -------- | -------- | ---------------- | --- | ------- | ----------------- | --- | -tivity:OntheMovies&TVdataset,characterizedbylargerand -| | | | | | | | | natural language | | summaries | (Figure | 2A | and | 2B) clearly | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --------- | ------- | --- | --- | ----------- | -higheruseractivity(averageprofilesize:11.79,std:9.80),the -| | | | | | | | | reflect distinct | short-term | | versus | long-term | user | interests, | -| --------- | --------- | ------------ | --- | ----------- | --- | -------------- | --- | ---------------- | -------------- | --- | --------- | ------------- | ---- | ----------- | -| evaluated | framework | demonstrates | | significant | | effectiveness. | | It | | | | | | | -| | | | | | | | | providing | human-readable | | profiles. | Additionally, | | learned at- | -notably outperforms the Centric baseline (17% Recall@10, tentionweights(αshort,αlong)transparentlyindicatetherelative -| 14% NDCG@10 | | improvements) | | and | Temp-Fusion. | | These ob- | | | u | u | | | | -| ---------------------- | ---------- | ------------- | -------- | ---------------- | ------------ | --------- | ------------ | -------------------- | --------- | ---------- | -------------- | ----------- | ---------- | ----------- | -| | | | | | | | | influence | of these | profiles | in | the unified | user | embedding, | -| servations | underscore | the | value | of incorporating | | | semantically | | | | | | | | -| | | | | | | | | aiding understanding | | of | recommendation | | rationale. | Figure 2C | -| rich, natural-language | | user | profiles | from | LLMs, | | particularly | | | | | | | | -| | | | | | | | | illustrates | how these | components | | could | generate | user-facing | -| where user | behaviors | are | varied | and | frequently | changing. | | | | | | | | | -explanations,anaspectforfutureworktofullyrealizebuilt-in -| Mixed | Effectiveness | | for | Domains | with | Lower | User | | | | | | | | -| ----- | ------------- | --- | --- | ------- | ---- | ----- | ---- | --- | --- | --- | --- | --- | --- | --- | -transparency. -| Activity: | In the | Video | Games | domain, | with | smaller | and less | | | | | | | | -| --------- | ------ | ----- | ----- | ------- | ---- | ------- | -------- | --- | --- | --- | --- | --- | --- | --- | -Insummary,ourexperimentssuggestthattemporalprofiling -| frequent | user interactions | | (mean: | 4.55, | std: | 3.97), | the assess- | | | | | | | | -| -------- | ----------------- | --- | ------ | ----- | ---- | ------ | ----------- | --- | --- | --- | --- | --- | --- | --- | -mentrevealsmixedeffectiveness.Whileourapproachachieves with LLMs is practical and interpretable, especially for do- -the highest Recall@20, Temp-Fusion slightly surpasses it at mains with dynamic user behavior. Future work will focus on -| | | | | | | | | developing | user-facing | interfaces | | and evaluating | | explanations | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | ----------- | ---------- | --- | -------------- | --- | ------------ | -1https://github.com/milsab/UMRec to enhance transparency and trust. - -TABLEIV LanguageTechnologies,Volume1(LongandShortPapers),J.Burstein, -COMPARISONFORABLATIONVARIANTS(MOVIES&TV) C. Doran, and T. Solorio, Eds. Minneapolis, Minnesota: Association -for Computational Linguistics, Jun. 2019, pp. 4171–4186. [Online]. -AblationVariant Recall@20 NDCG@20 Available:https://aclanthology.org/N19-1423/ -Short-TermOnly(ST) 0.0193 0.0266 [2] A.Waswani,N.Shazeer,N.Parmar,J.Uszkoreit,L.Jones,A.Gomez, -Long-TermOnly(LT) 0.0195 0.0265 L.Kaiser,andI.Polosukhin,“Attentionisallyouneed,”inNIPS,2017. -GeneralPreferences(NoTS) 0.0178 0.0247 [3] M. Polignano, C. Musto, M. de Gemmis, P. Lops, and G. Semeraro, -Dot-ProductScoring(DP) 0.0136 0.0173 “Togetherisbetter:Hybridrecommendationscombininggraphembed- -FullModel(LLM-TP) 0.0223 0.0293 dings and contextualized word representations,” in Proceedings of the -15thACMconferenceonrecommendersystems,2021,pp.187–198. -[4] Y.Koren,“Collaborativefilteringwithtemporaldynamics,”inProceed- -ingsofthe15thACMSIGKDDinternationalconferenceonKnowledge -V. ABLATIONSTUDY -discoveryanddatamining,2009,pp.447–456. -We conducted an ablation study to determine the impact [5] B. Hidasi, A. Karatzoglou, L. Baltrunas, and D. Tikk, “Session- -basedrecommendationswithrecurrentneuralnetworks,”arXivpreprint -of individual components within the evaluated framework -arXiv:1511.06939,2015. -on recommendation accuracy, using the Movies&TV dataset. [6] W.-C. Kang and J. McAuley, “Self-attentive sequential recommenda- -Table IV summarizes the performance metrics, and Fig. 4 tion,” in 2018 IEEE international conference on data mining (ICDM). -IEEE,2018,pp.197–206. -highlighting relative improvements. More results are in the -[7] H.Chi,H.Xu,H.Fu,M.Liu,M.Zhang,Y.Yang,Q.Hao,andW.Wu, -GitHub repository (see Section IV). The ablation results un- “Long short-term preference modeling for continuous-time sequential -derscore the importance of temporal modeling and non-linear recommendation,”arXivpreprintarXiv:2208.00593,2022. -[8] Y.Zhang,X.Chenetal.,“Explainablerecommendation:Asurveyand -scoring.TheGeneralPreferences(NoTS)variant,wherethe -new perspectives,” Foundations and Trends® in Information Retrieval, -LLM generated a single, holistic user profile (distinct from vol.14,no.1,pp.1–101,2020. -Temp-Fusion’s numerical aggregation), showed a consistent [9] W.X.Zhao,K.Zhou,J.Li,T.Tang,X.Wang,Y.Hou,Y.Min,B.Zhang, -J. Zhang, Z. Dong et al., “A survey of large language models,” arXiv -performance drop of over 20% in Recall and 18% in NDCG, -preprintarXiv:2303.18223,2023. -emphasizing the value of LLM-generated temporal distinction [10] Y.Chang,X.Wang,J.Wang,Y.Wu,L.Yang,K.Zhu,H.Chen,X.Yi, -(Fig. 4). The full framework also consistently outperformed C. Wang, Y. Wang et al., “A survey on evaluation of large language -models,” ACM transactions on intelligent systems and technology, -Short-Term Only (ST) and Long-Term Only (LT) variants -vol.15,no.3,pp.1–45,2024. -individually (e.g., +15.2% Recall@20 over ST), confirm- [11] Z. Zhao, W. Fan, J. Li, Y. Liu, X. Mei, Y. Wang, Z. Wen, F. Wang, -ing complementary signals from both temporal components. X. Zhao, J. Tang et al., “Recommender systems in the era of large -language models (llms),” IEEE Transactions on Knowledge and Data -Lastly,replacingMLPscoringwithadotproduct(DPvariant) -Engineering,2024. -resulted in the steepest degradation, highlighting the necessity [12] L. Wu, Z. Zheng, Z. Qiu, H. Wang, H. Gu, T. Shen, C. Qin, C. Zhu, -of non-linear interaction modeling. Overall, Figure 4 and H. Zhu, Q. Liu et al., “A survey on large language models for -recommendation,”WorldWideWeb,vol.27,no.5,p.60,2024. -TableIVdemonstratethattemporaldisentanglementviaLLM- -[13] S. Lubos, T. N. T. Tran, A. Felfernig, S. Polat Erdeniz, and V.-M. -based profile summarization and non-linear embedding fusion Le,“Llm-generatedexplanationsforrecommendersystems,”inAdjunct -collectivelycontributetotheobservedrobustrecommendation Proceedingsofthe32ndACMConferenceonUserModeling,Adaptation -andPersonalization,2024,pp.276–285. -performance within the evaluated framework. -[14] Q. Ma, X. Ren, and C. Huang, “Xrec: Large language models for -explainablerecommendation,”arXivpreprintarXiv:2406.02377,2024. -VI. CONCLUSION -[15] Q. Guo, F. Zhuang, C. Qin, H. Zhu, X. Xie, H. Xiong, and Q. He, -This paper presented an exploratory evaluation of a “A survey on knowledge graph-based recommender systems,” IEEE -Transactions on Knowledge and Data Engineering, vol. 34, no. 8, pp. -lightweight framework that leverages LLM-driven temporal -3549–3568,2020. -user profiling within a pipeline demonstrating interpretability [16] S. Wang, W. Fan, Y. Feng, S. Lin, X. Ma, S. Wang, and D. Yin, -potential. Our evaluation indicates varied capabilities: notable “Knowledgegraphretrieval-augmentedgenerationforllm-basedrecom- -mendation,”arXivpreprintarXiv:2501.02226,2025. -gains in high-activity domains (e.g., Movies&TV) versus -[17] R. Shimizu, M. Matsutani, and M. Goto, “An explainable recommen- -less pronounced benefits in sparser environments (e.g., Video dation framework based on an improved knowledge graph attention -Games). This highlights a critical trade-off between enhanced networkwithmassivevolumesofsideinformation,”Knowledge-Based -Systems,vol.239,p.107970,2022. -performance and computational costs, suggesting context- -[18] G. Shi, X. Deng, L. Luo, L. Xia, L. Bao, B. Ye, F. Du, S. Pan, -dependent LLM application. The framework inherently offers and Y. Li, “Llm-powered explanations: Unraveling recommendations -a promising pathway toward transparency through its natural throughsubgraphreasoning,”arXivpreprintarXiv:2406.15859,2024. -[19] Y.K.Tan,X.Xu,andY.Liu,“Improvedrecurrentneuralnetworksfor -language profiles and attention weights, providing an intrinsic -session-basedrecommendations,”inProceedingsofthe1stworkshopon -potentialforinterpretability.Theseearlyinsightscontributeto deeplearningforrecommendersystems,2016,pp.17–22. -understandingthepracticalcapabilityofLLM-driventemporal [20] Y.Zhu,H.Li,Y.Liao,B.Wang,Z.Guan,H.Liu,andD.Cai,“What -to do next: Modeling user behaviors by time-lstm.” in IJCAI, vol. 17, -user profiling. Future work includes developing user-facing -2017,pp.3602–3608. -explanations, conducting user studies, and exploring adaptive [21] M.-C. Popescu, V. E. Balas, L. Perescu-Popescu, and N. Mastorakis, -LLM strategies. “Multilayer perceptron and neural networks,” WSEAS Transactions on -CircuitsandSystems,vol.8,no.7,pp.579–588,2009. -REFERENCES [22] Y. Hou, J. Li, Z. He, A. Yan, X. Chen, and J. McAuley, “Bridging -language and items for retrieval and recommendation,” arXiv preprint -[1] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- arXiv:2403.03952,2024. -trainingofdeepbidirectionaltransformersforlanguageunderstanding,” [23] F. Ricci, L. Rokach, and B. Shapira, “Introduction to recommender -in Proceedings of the 2019 Conference of the North American systems handbook,” in Recommender systems handbook. Springer, -Chapter of the Association for Computational Linguistics: Human 2011,pp.1–35. - -[24] Y.Koren,R.Bell,andC.Volinsky,“Matrixfactorizationtechniquesfor -recommendersystems,”Computer,vol.42,no.8,pp.30–37,2009. -[25] N.ReimersandI.Gurevych,“Sentence-bert:Sentenceembeddingsusing -| siamese bert-networks,” | in Proceedings | | of the 2019 | Conference | on | -| ----------------------- | ------------------ | -------- | --------------- | ----------- | --- | -| Empirical | Methods in Natural | Language | Processing. | Association | | -| for Computational | Linguistics, | 11 | 2019. [Online]. | Available: | | -https://arxiv.org/abs/1908.10084 -[26] J.Achiam,S.Adler,S.Agarwal,L.Ahmad,I.Akkaya,F.L.Aleman, -| D. Almeida, | J. Altenschmidt, | S. Altman, | S. Anadkat | et al., | “Gpt-4 | -| ----------- | ---------------- | ---------- | ---------- | ------- | ------ | -technicalreport,”arXivpreprintarXiv:2303.08774,2023. \ No newline at end of file +S + +Y +o +Y +e +g +n +a +h +C +% + +Q2 +2018 + +Q3 +2018 + +Q4 +2018 + +Q1 +2019 + +Q2 +2019 + +Q3 +2019 + +Q4 +2019 + +Q1 +2020 + +Q2 +2020 + +Q3 +2020 + +Q4 +2020 + +Q1 +2021 + +Q2 +2021 + +Q3 +2021 + +Q4 +2021 + +B&M YoY Growth + +E-commerce YoY Growth + +E-commerce + +Brick & Mortar + +Source: US Dept of Commerce, Economic Indicators Division, Retail Indicator Branch +Estimated Quarterly U.S. Retail Sales (Not Adjusted): Brick & Mortar and E- + +Brick and mortar vs. E-Commerce Sales US Department of Commerce Retail Indicators Division + +More US consumers are switching brands and retailers now than +in 2020 and 2021, and about 90 percent plan to continue doing so. + +Shopped a different brand recently, +% of respondents + +Shopped a different retailer/store/ +website recently, % of respondents + ++13 percentage +points + ++9 percentage +points + +~90% of consumers +who have switched +brands or retailers +plan to incorporate +these new behaviors +into their routines + +Sep-20 +Source: McKinsey COVID-19 US Consumer Pulse Surveys: Feb 25 – Mar 1, 2022, n = 2,160; Oct 9-15, 2021, n = 2.095; Sept 18-24, 2020, n = +1,026; sampled and weighted to math the US general population aged 18 years and older + +Oct-21 Feb-22 + +Oct-21 Feb-22 + +Sep-20 + +4 + + In-Store Shopper Engagement Built for the Future of Retail + +That said, historically, consumers would get their Sunday +newspapers for coupons in order to get the best deals in +stores. Then, there became a hybrid of clipping coupons +along with digital apps or digital coupons for consumers to +get better offers. Today, and especially coming out of a +post-pandemic world, the in-store experience is even more +digital — with the decline of print newspapers, the shift to +online shopping and the consumer desire for a contactless +checkout experience. + +What’s more, grocers are relying on connected recipes as a +tool to drive Q4 spending. Walmart offers its “Cookshop” +interactive content portal, which currently presents +consumers with shoppable recipes and activities. Online +tutorials made in partnership with major brands like Kraft +or Starbucks, embed “shop your ingredients” buttons and +other shopping list features.6 + +So, retailers need to find the best way to make content, +recipes and deals easily consumable by consumers. In fact, +“When consumers visit an offline store, it is no longer to +simply buy something. Creating an interesting and +immersive experience that online shopping can not rival is +one of the keys to drive foot traffic,” according to Gartner.1 + +92% + +of grocery shoppers used +coupons or deals when shopping. + +60% + +of grocery shoppers said they +planned their shopping trip +based on which stores offer the +best deals. + +51% + +of grocery shoppers said they +used mobile apps to find deals +before going to the grocery store. + +Adoption of the Smart Cart at the ‘Moment +of Truth’ + +• With some of the aforementioned drivers impacting the way consumers are spending, +coupled with inflation and the supply chain bottlenecks still catching up post-COVID – +consumers are looking for deals more than ever right now, and are willing to take on new +technology to do so. + +• As such, the smart cart is becoming more prevalent in retail stores, but operational issues +typically get in the way of success. Cart costs are high (upwards of $250 million), especially +when they involve weigh scales and cameras to make them ‘smart.’ These carts come with +maintenance overhead, can be too small, and don’t stack or nest conveniently. + +• And let’s not forget that long lines during checkout leave retailers open to more shrink + +issues, and consumers open to better technology to solve their payment woes. + +• This leaves retailers with substantial investments in smart cart technology, but low or no + +adoption in stores. + +• However, there is an opportunity for retailers to present these shoppers with relevant +deals (combined with their loyalty information) right in the store aisles for the ideal in- +store experience — and as an added bonus — contactless checkout. + +5 + + In-Store Shopper Engagement Built for the Future of Retail + +EXPLORE WHAT’S POSSIBLE + +While the shopping cart hasn’t seen many innovations over the years + +retailers can engage with shoppers in a way that has never happened + +before by adding a touchscreen right to the cart. + +Not only can it present valuable and relevant information to shoppers on how to +get the best deals and ideas for new items or recipes to try — it can also show +shoppers how to locate those products right in store aisles. + +This is being done every day on social media, but being able to access this content +right on a cart without relying on a personal device connects the channels in a way +that satisfies the future shopper. + +Especially in a world where shoppers are still using their smartphones to conduct +research before heading into the store, adding a mobile and connected +touchscreen is a game-changer for retailers. + +Gartner reports that 95% of retailers are saying they plan to increase their +investments in digital capabilities in 2022.1 + +This means that your competitors are also boosting their technologies to enhance +their brand’s in-store and online shopping experiences with the goal to offer a +cohesive omnichannel experience that is immersive and personalized. + +Brands would be able to reach shoppers in an even more relevant way and be able +to direct them right to their product in the store aisles as they shop (right at the +precise ‘moment of truth.’ The items they need would always be top of mind, and +CPGs can easily connect the ROI back to their ad spend because that anonymized +data would be captured right into the connected screen. + +And, as shoppers look for specific products, they can come across + +personalized ads from brands or even push a recipe or coupon to + +inspire a new or spontaneous purchase (also creating more + +valuable loyalty data for retailers and their CPG suppliers). + +6 + + In-Store Shopper Engagement Built for the Future of Retail + +Did You Know? + +Sylvan Goldman +owner of the grocery chains +Piggly Wiggly and Humpty- +Dumpty. + +What: Goldman invented the shopping cart, arguably the most important innovation in retail +stores even today. + +Where: Oklahoma City + +When: 1937 + +Why: Groceries were getting heavier products (like canned goods), and more refrigerated +items, and were still using hand baskets to carry them. + +How: Goldman enlisted the help of his handyman Fred Young. The two grabbed a few +supplies and spent a night coming up with a prototype of a rolling grocery basket. + +The Result? Shoppers rejected it … at first. Men were offended by it saying it made them +seem weak for not being able to carry groceries in their hands, and women compared it to +pushing a baby stroller. But, with some advertising efforts, the innovation took off, +increasing basket sizes and improving the shopper experience. + +7 + + In-Store Shopper Engagement Built for the Future of Retail + +Business Benefits of Being Memorable + +Creating a truly memorable shopping experience has many business + +benefits, especially when leveraging technology to support it. Adding + +a connected screen, there are many positive implications for + +retailers, brands, and shoppers alike. Consider this: + +81% + +59% + +73% + +35% + +of retail shoppers conduct online research before buying. (Source: GE Capital +Retail Bank). With a connected screen, there’s no need for research prior to a +shopping trip meaning more accurate trips, fewer forgotten items, and no +need for a personal device (plus, more time well spent in the store). + +of shoppers use their mobile devices in-store to compare costs or research +deals and coupons. (Source: HRC Retail Advisory). Knowing a connected +screen is right on the shopping cart means no relying on my mobile device in +the store. + +of retail consumers use multiple channels to shop. To gain a unified +consumer view, marketers must connect all of their channels (Source: +Harvard Business Review). A connected screen allows for recipes, social +content, influencers, etc., to join a shopping trip creating a unified +experience. + +of marketers say poor data quality impacts their ability to target consumers +with the right digital ads. (Source: Forrester). Connected screens provide a +wealth of new data and insights right from the store shelves where the +business benefits spill over to digital advertising channels. + +84% + +of consumers believe retailers should be doing more to integrate their online +and offline channels. (Source: RetailNext). A connected screen would bring +both together into one memorable customer experience. + +40% + +of consumers purchase more from retailers that provide a personalized +shopping experience across channels. Improving the shopper experience will +have a direct impact on your bottom line (Source: RetailNext). The connected +screen converges personalization into brick-and-mortar. + +8 + + In-Store Shopper Engagement Built for the Future of Retail + +6 Ways to Stand Out from the Competition +While many media outlets will continue to talk about the acceleration of digital, retailers are +leaning into the physical store to drive loyalty within their customer experiences, as well as +differentiate themselves from their competition. Shoppers prefer to see, feel and even try +products in the store before making a purchase, so the opportunity for innovation lies +within the brick-and-mortar channel. + +Perhaps Gartner sums it up best, “Retail is experiencing unprecedented change driven by +consumer behaviors, enabled by technology and accelerated by a pandemic.”7 + +Getting started on a business process change or improvement like creating or revamping +your customer experience can be a daunting task. As such, here we have gathered 6 best +practices to help get your store ready for the future of retail (and a more memorable in- +store experience). + +Consumers want deals, but they want relevant deals. Combining loyalty +information about that consumer, plus where they are in the store gives retailers +the ability to drive information that is highly pertinent relative to that consumer. + +Leverage real-time data and insights to act on consumer trends more quickly, +keeping popular products in front of shoppers while they’re in high demand.8 Use +this new stream of data (i.e. heat maps) to understand consumer behavior in the +store (where they stopped, why, how long they stayed, what the interaction was) +and share that with your suppliers. + +Use technology to play a distinct and purposeful role to enhance customer +engagement and loyalty. Be sure it also supports your overall brand purpose.9 +Consumers don’t want to use different apps. Think about the development of your +apps and how you can seamlessly work them into the store environment. + +Create opportunities for speed and convenience (i.e. fast, contactless and +seamless checkout).9 Regardless of size, retailers must focus on checkout, +especially as the rise of contactless or self-service checkout is on the rise. + +Combine shrink avoidance technology, principles and methods — and embed +those right into your carts. + +Use shopping cart-based screens to present circulars to consumers at the moment +when it is most useful - as they enter that aisle where the products are on +promotion. FMI—The Food Industry Association (FMI) said in a report in 2022 that +55% of surveyed retailers found print circulars very effective and also noted that +widely used digital circulars and promotions sent by text are not as effective as +more traditional outreach methods.10 + +9 + + In-Store Shopper Engagement Built for the Future of Retail + +Many times when retailers are ready to implement a new solution or service, they can be +inundated by a ‘rip and replace’ mentality, but it doesn’t have to be that way. Start with small +wins and continue to evolve and enhance as you go with each iteration. Many providers +today can stand up their technologies in a matter of days and weeks so retailers can start +hitting their KPIs immediately and then continue to improve. + +“Even as consumers continue their return to stores, the digital behaviors they’ve picked up +during the pandemic will stick for the long term,” according to Retail Dive.9 “Brands and +retailers that win in the coming years will embrace digital innovation and find more +thoughtful and authentic ways to blend digital and physical engagement.” + +Now, retailers can engage with shoppers in a way that has never been done before – +present valuable information on how to get the best deals, how to locate products and ideas +for new items or recipes to try. This has never been accomplished in a store environment, +but the future of retail is here. + +Shopper Engagement Platform for the Physical +Store Powered by Veeve +Veeve's vision is to create a content delivery and checkout platform that utilizes the +retailers' own Apps to improve the shopper experience and increase the basket size +through in-store, location-driven promotions, advertising, and more. Veeve is delivered by a +Connected Screen: a purpose-built device that brings the power of online shopping, sensor +fusion, and computer vision to physical stores. + +The Connected Screen + +What’s on sale +today? + +Where can I find +salsa? + +Deals & +Coupons + +Location +Tracking + +What’s for +dinner? + +Video +Recipes + +Can I skip +checkout? + +Pay on +Cart + +Visit https://veeve.io/ to learn more. + +Source 1. Gartner, “Time To Prepare Smart Retail For The Post-Pandemic Era” by Wilson Zhao 2. Insider Intelligence, “Global Ecommerce +Forecast” 3. Mastercard SpendingPulse, “Mastercard SpendingPulse Forecasts 7.5%* Growth for U.S. Back-to-School Retail Sales” +4. McKinsey, “How US consumers are feeling, shopping, and spending—and what it means for companies” 5. Inmar Intelligence, “New +Inmar Intelligence Survey Reveals Over Half of Today’s Shoppers Have Switched Loyalty to Less Expensive Grocery Brands Since Start of +Pandemic” 6. PYMNTS, “Grocers See Connected Recipes as Tool to Drive Q4 Spending” 7. Gartner, “Digital Transformation in Retail” +8. Retail Dive, “Retail tech is elevating CX unlocking efficiencies” 9. Retail Dive, “3 pillars driving the future of store experience” +10. FMI The Food Retailing Industry Speaks 2022 + +1 +0 + diff --git a/q159/random_k4/random_3.md b/q159/random_k4/random_3.md index 98f8c27c9bf7c545c095968ca0e7e8b52e552935..a021009aee38cc778527ef4c4bc6196db6696211 100644 --- a/q159/random_k4/random_3.md +++ b/q159/random_k4/random_3.md @@ -1,161 +1,968 @@ -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ +Becoming a +data-driven +organization -https://doi.org/10.1051/itmconf/20268403023 +The what, why and how -# **Optimizing Retrieval-Augmented Generation for Small Language Models via Output Alignment** + Ongoing digitization is turning everything into data, forcing -## _Runkai_ Dong[1*] +Technological advancements in data analytics are, however, -1School of Computer Science, University of Birmingham, Birmingham, United Kingdom +companies to become more data-driven. While the benefits of -**Abstract.** Traditional Visual Question Answering (VQA) models have a significant limitation: they cannot remember the latest external world knowledge, nor use it. Although Retrieval-Augmented Generation (RAG) can solve the problems of hallucination, current researches are focuses on the massive, computation-heavy models, which prevents us from knowing whether RAG is effective in resource constrained environment. This paper presents a comprehensive study of building a light multi modal RAG (MMRAG) pipeline on consumer-grade hardware. Specifically, this study made a comparative study between two small language models: TinyLlama (1.1B) and Qwen 2.5(3B). The results of research demonstrate that, first, with the access of RAG, the accuracy of outcome results has a significant increase (of about 13%-16%) compared to zero-shot baselines, despite the scale of models. Second, which is the most important, this study identifies a verbosity failure of those instruction-tuned small language models (SLMs), which are more likely to generate output noise and leads to low evaluation marks. To address this, this research developed a post processing protocol, recovering the accuracy of Qwen from 8% to 52.6%. These findings illustrate that for edge-deployed VQA systems, rigorous output alignment is as critical as the retrieval mechanism itself. +making it possible for any type of company in every industry to -## **1 Introduction** +the data-driven organisation are clear (improved performance, -Visual Question Answering (VQA) systems have a main duty: they need to generate answers based on images. But in real life, many questions cannot be answered only given the images, such as figuring out a specific historical person or a rare object. This needs external knowledge. Although Large Language Models (LLMs) are good at this, they still have problems, especially hallucination and lacking up-to-date new world knowledge [1]. To solve these problems, Retrieval-Augmented Generation (RAG) was created. It connects external databases to the model, bridging the gap between what people see and read. +become data-driven. -Recently, this area shows a significant development. Lewis et al. built the first RAG framework for NLP [1]. They showed that using outside memory is better than just using a big frozen model. After that, Fan et al. and Zhao et al. wrote surveys about how RAG is moving towards multimodal tasks [2, 3]. Big models like REVEAL and LLaVA got great results using visual instruction tuning [4, 5]. But these models have a big problem: they need massive computational resources and expensive GPUs to run. This makes it very hard for +more profitability, stronger innovations), there are still some -* Corresponding author’s email: dongrunkai@gmail.com +Discover the basic do’s and don’ts in ‘Becoming a data-driven -© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (https://creativecommons.org/licenses/by/4.0/). +technical and business challenges to overcome. -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ +organisation: the what, why and how’. -https://doi.org/10.1051/itmconf/20268403023 +Table of contents -people with normal computers or edge devices to use them. While some works like MobileVLM tried to fix this for phones, the paper still don’t know enough about how Small Language Models (SLMs) work with RAG [6]. + 1 -This paper tries to fill this gap. This study built a light system on consumer-grade hardware (T5 GPUs). The paper followed the baseline method from Xenos et al. [7], but the paper made a big change in model selection. The paper chose two specific models: TinyLlama (1.1B) [8], which is ultra-lightweight for edge devices, and Qwen 2.5 (3B) [9], which represents the “sweet spot” where models start to become smart. +Why become +data-driven? -This study has some interesting findings. First, this study found that although small models are not good at reasoning, they can have significant improvement with the access of RAG. Additionally, just like Zheng et al. found, the paper noticed a “verbosity bias” (talking too much) in these small models [10]. The paper proposed a cleaning method to fix this, showing that alignment is very important for getting the right score. + You may not have noticed, but everything around us has -## **2 Related Work** +turned into data. Not just our cars or mobile phones, a -The foundation of this research lies in the intersection of VQA and RAG [11, 12]. Traditional VQA approaches store all the knowledge within their model parameters. This is useful for closed-domain tasks, but it fails when facing some open tasks which require broad external knowledge. +growing number of other appliances, machines and ‘things’ -To address this, Lewis et al. introduced RAG in the NLP domain, proposing a hybrid model that retrieves relevant documents to condition the generation process [1]. Afterwards this method was used to solve multimodal tasks. Early multimodal RAG systems used complex structures. More recent approaches, like REVEAL and LLaVA, employ end-to-end visual instruction tuning [4, 5]. While they are smart, they require massive computational resource for training and reasoning, which is not suitable in this study’s edge deployment scenarios. +are generating a constant flux of data. Where we are and -At the same time, people started to show an interest towards effective small models. Works such as MobileVLM prove it is feasible to run visual language models (VLMs) on mobile devices [6]. But, applying RAG to these lightweight architectures still remains underexplored. To fill this gap, this study evaluated how modular RAG pipeline works with models under 3B parameters, such as TinyLlama and Qwen, and focus on the challenges of output alignment and instruction following in smaller architectures [8, 9]. +what we do is used for marketing purposes. Sensors in -## **3 Methodology** +machines tell companies how to improve their output. -## **3.1 Pipeline Architecture** +This flood of data is transforming our world. Companies that -The MM-RAG architecture presented in this study shows a three-stage pipeline, the difference between this pipeline and some monolithic end-to-end models is that this system adopts a modular design. (see Fig. 1) +want to stay ahead must become data-driven. -First, this system uses the BLIP model, converting images into text description. This step is crucial because it is important to convert pixels into text that can be searched from database. Second, this system uses cosine similarity to find the 9 most related Question-Answer pairs in the knowledge base. Third, this system inputs the retrieved results in the prompt and allows the Large Language Model (LLM) do the final synthesis. +The rise of the data-driven organisation -2 +Many organisations worry about staying competitive in the midst of Big +Data, Artificial Intelligence (AI), Machine Learning or the Internet of Things +(IoT). Especially as many of these concepts are already generating value +for many companies. The glue that binds all of these together is data. -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ +What is being data-driven all about? -https://doi.org/10.1051/itmconf/20268403023 +Data-driven organisations process and use ever more data -**==> picture [350 x 167] intentionally omitted <==** +As consultancy firm McKinsey says: -**Fig. 1.** The schematic overview of the proposed lightweight MM-RAG pipeline. (Picture credit: Original) +to improve and speed up their decision-making. The goal of -## **3.2 The Strategic Pivot** +having superior analytics is having superior insights. In data- -**Table 1.** Comparison of language models evaluated +driven organisations, decisions that aren’t supported by data, -|**Model**|**Parameters**|**Role in Study**|**Hardware Status**| -|---|---|---|---| -|Llama-2|13B|Ideal reproduction model|Access Gated| -|Mistral|7B|Initial Pivot|OOM(Failed)| -|Qwen 2.5|3B|Main Subject|Functional| -|TinyLlama|1.1B|Main Subject|Functional| +are considered suspicious. Smarter analytics technologies +now enable every company to become more data-driven. +“Businesses no longer have to go on gut instinct; +they can use data and analytics to make faster +decisions and more accurate forecasts supported +by a mountain of evidence.” -This study faced a big problem with the use of language models. Initially, this study wanted to use Llama-2, but it is a gated model and all access was rejected by community keepers. Then this study had to change the plan. This study rewrote the code to use the Automodel For CausalLM interface instead of Llama For CausalLM [7]. This was a strategic pivot. This change allowed us to easily switch between different models. But this study still had a hardware constraint, as the Mistral 7B model caused “Out-Of-Memory” (OOM) errors because it was too big for the T5 GPU, so this study chose two smaller models: TinyLlama (1.1B) and Qwen 2.5 (3B) (see Table 1). The pivot turned the hardware limitation into a chance to compare different lightweight models. +Becoming a data-driven organization -## **3.3 Output Sanitization** +4 -When this study tested Qwen, a huge problem was found: The model was too “chatty”. Although it could output right answers like “skating”, it added some useless words to the initial answer, like “skating\n\Context”, which leads to marking failure and low output accuracy. To solve this problem, this study developed a rule-based sanitization protocol. It cuts off any text after a newline character and removes redundant words. This step was crucial. It separated the real reasoning from the “noise”, allowing us to see the model’s true performance. As a result, this study observed a significant improvement of accuracy, from 8% to 52.6% + Data-driven organisations +use analytics to become smarter: -## **4 Experiments and results** +1 -3 +They perform +better -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ +The data shows where +they can streamline +their processes. -https://doi.org/10.1051/itmconf/20268403023 +3 -## **4.1 Experimental setup** +They are more +profitable -This study used the OK-VQA validation set (1000 samples) for our tests, and designed a 2x2 experiment matrix: this study compared two models (TinyLlama vs Qwen) in two situations (no RAG vs with RAG). This study only used Exact Match Accuracy to measure success, after applying our cleaning script. +Constant improvements +and better predictions +help to outsmart the +competition and +improve innovation. -## **4.2 Quantitative Analysis** +2 -The experimental data shows that RAG is crucial for small language models. With access to RAG, the accuracy of two models shows a significant increase between 13% and 16%. Model scale also matters, when this study tested the baseline, Qwen (3B) outperformed TinyLlama (1.1B) by 0.8%, which was a small margin. But with the access of RAG, the increase of Qwen(+15.2%) was much higher than the increase of TinyLlama(+13.5%) (see Fig. 2). This shows that Qwen has a better “brain” to use the outside help (context) effectively. +They are +operationally + more predictable -**==> picture [287 x 193] intentionally omitted <==** +Data insights fuel +current and future +decision-making. -**Fig. 2.** Performance comparison (Exact Match Accuracy) across model scales (Picture credit: Original) +These advantages make an organisation more shock-resistant and less +likely to be surprised by the next economy - or technology disruption. -## **4.3 Ablation Study** +Gut feeling no longer makes the difference -To prove that this study’s cleaning script was necessary, this study conducted a test. Without the script, Qwen got a terrible score of only 8%. But when this study used the script, the score jumped to 52.6%. This huge difference tells us something important: the model was not stupid, but it just had a format problem. The model tried too hard to follow the chat format and disrupted the answer structure. So, for chat-based models, alignment of the output format is even more important than the reasoning power itself. +Gut feeling is not good enough anymore to differentiate yourself from +your competitors. To be truly competitive, you will need data. Lots of +relevant data. Luckily, any organisation can set out on the journey to +become data-driven. You no longer need to be a data scientist to work +with data. Citizen Data Scientists are not your professional statistician or +trained analyst, nor your maths wizard or computer scientist, but rather +regular business users who create and use advanced analytical models. -## **4.4 Qualitative Analysis** +Citizen Data Scientists are part of the ongoing wave of democratization +of analytics in every department. These business people have the right +attitude – curious, adventurous, determined – to research and improve +things in your organisation. They want to get their hands on the data +themselves and find new ways to get answers. They’re willing to learn +new methods and use new tools. They often think, “I don’t want to ask a +statistician. I want to try it myself.” -The study also looked at specific examples to see why models failed or succeeded. For visual ability, TinyLlama often makes simple mistakes. For example, it called a “gazebo” a “garage” because it couldn’t see well. On the other hand, Qwen was smarter. In a picture of a person in snow, Qwen correctly guessed “alpine skiing” because it understood the tools (see Fig. 3(a)). However, the paper also saw a “negative RAG” problem. Sometimes, the LLM could give a wrong answer after given the access of RAG. In one case, the model saw a “napkin” +Becoming a data-driven organization -4 +5 -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ + 2,500 PB -https://doi.org/10.1051/itmconf/20268403023 +Every day, the world creates 2,500 +petabytes of data. In the past two years, +mankind has generated more data than +in the preceding 5,000 years combined. -correctly. But the retrieval system brought back text about “forks” and “tableware”. The model got confused and trusted the text more than its eyes, giving the wrong answer “fork”. This shows small models can be easily tricked by bad retrieval results. +Source: IFL Science -**==> picture [181 x 209] intentionally omitted <==** + ZOOM-IN ON +SWISSCOM -**==> picture [171 x 198] intentionally omitted <==** +SWITZERLAND | TELECOM | CUSTOMER SERVICE -**Fig. 3.** Qualitative examples of VQA predictions. (a) Qwen successfully reasons about "alpine skiing". (b) An instance of "Negative RAG" where noisy retrieval ("tableware") misleads the model (Picture credit: Original) +7 x FASTER -## **5 Conclusion** +CUSTOMER SERVICE DATA IS NOW PROCESSED 7X FASTER +MAKING IT FAR MORE USEFUL IN ISSUE SOLVING. -This research successfully validates that implementing MultiModal Retrieval-Augmented Generation on consumer-grade hardware is accessible. Previously, it was believed that this study must use large models, but through the pivot of changing monolithic model to modular small models (1-3B), this study proves that small model combined with RAG can also achieve significant performance improvements. Specifically, for the Qwen 2.5(3B) model, when combined with the sanitization protocol developed in this study, it reached an accuracy of 52.6%, which is competitive among small models. +Swisscom, Switzerland’s biggest telecom operator, found that the analysis +of their customer service data was too slow and required too much +manual work. As such, it did not really help to improve customer service. +Through smarter text analytics, however, relationships and possible +solutions were shown much faster, often almost simultaneously as the +ongoing call center documentation evolved. Reports are now sent daily +instead of weekly or even monthly. -This paper also found some challenges. First, small models are more likely to have verbosity bias. This means that good post-processing scripts are always needed; also, the “Exact Match “evaluation metric is too rigid for chat models. Second, through the “Negative RAG” phenomenon, this study shows that it is necessary to find ways to filter out the retrieved noise in future work. +“We are able to create fully automated daily reports, which +has a direct positive effect on service quality and customer +satisfaction.” -For future work, this study aims to develop a “LLM-as-a-Judge” framework to use Large Language Models as judges. Unlike the rigid exact match system, this approach is more flexible for format changes. Also, 7B parameter model would be implemented to see if the results would be better if access to better GPUs is available in the future. +Albert Labermeier +Senior Marketing Analyst at Swisscom -## **References** +Becoming a data-driven organization -1. P. Lewis et al., Retrieval-augmented generation for knowledge-intensive NLP tasks, in Proc. NeurIPS (2020) +6 -2. W. Fan et al., A survey on RAG meeting LLMs: Towards retrieval-augmented large language models, arXiv preprint arXiv:2405.06211 (2024) + 2 -5 +The road to +becoming +data-driven -ITM Web of Conferences **84** , 03023 (2026) _ATCIDS 2026_ + While the benefits of becoming more data-driven are -https://doi.org/10.1051/itmconf/20268403023 +apparent, in our experience, many companies are still faced -3. P. Zhao et al., Retrieval-augmented generation for AI-generated content: A survey, arXiv preprint arXiv:2402.19473 (2024) +with a few bumps in the road. Luckily, technical advances are -4. Z. Hu et al., REVEAL: Retrieval-augmented visual-language pre-training with multisource multimodal knowledge memory, in Proc. CVPR, 23369–23379 (2023) +bringing data analytics within reach of a growing number of -5. H. Liu et al., Visual instruction tuning, in Proc. NeurIPS (2023) +organisations. -6. X. Chu et al., MobileVLM: A fast, strong and open vision language assistant for mobile devices, arXiv preprint arXiv:2312.16886 (2023) +Changing mindsets -7. A. Xenos et al., A simple baseline for knowledge-based visual question answering, in Proc. EMNLP (2023) +On the road to becoming data-driven, it’s crucial for people to change +their mindset and organisations to change their processes. Doing so, will +help overcome some of these hurdles: -8. P. Zhang et al., "TinyLlama: An open-source small language model," arXiv:2401.02385, (2024) +1 -9. Qwen Team, "Qwen2.5: A comprehensive series of large language models," arXiv:2409.12191, (2024) +2 -10. L. Zheng et al., Judging LLM-as-a-judge with MT-bench and chatbot arena, in Proc. NeurIPS (2023) +3 -11. S. Maekawa et al., Retrieval helps or hurts? A deeper dive into the efficacy of retrieval augmentation to language models, arXiv preprint arXiv:2402.13492 (2024) +4 -12. A. Asai et al., Self-RAG: Learning to retrieve, generate, and critique through selfreflection, in Proc. ICLR (2024) +UNSTRUCTURED DATA -6 +UNCONNECTED SYSTEMS + +LOW DATA QUALITY OR +UNAVAILABLE DATA + +MISALIGNMENT WITH IT + +Data that is not predefined or does + +Organisations often use multiple + +Sometimes, the data quality simply + +Business units shouldn’t have to + +not fit the mould of traditional + +information storage systems side by + +isn’t good enough, because of poor + +depend on IT for data analytics, they + +data models. This includes text + +side with no or difficult connections + +data input or poorly implemented + +should be able to run it themselves. + +documents, pictures, e-mails, sensor + +between them. These systems may + +data connections. It is hard to get + +With IT being under constant + +data, and much more. This data + +even offer conflicting information + +good business intelligence from + +pressure to keep delivering more + +is hard to analyse for traditional + +because they use different sources, + +poor – or plain wrong – data. + +at lower costs, your data analytics + +analytics programs, although it + +processing methods or naming + +contains valuable information. + +conventions. + +requests may end up at the bottom + +of their list. + +But all of these challenges can be overcome by defining +a roadmap towards better data analytics. + +Becoming a data-driven organization + +8 + + ZOOM-IN ON +ASTRAZENECA + +SWEDEN | HEALTHCARE | MANUFACTURING + +VARIATIONS IN THE PRODUCTION PROCESS HAVE BEEN MINIMIZED. + +THE CONTENT OF THE ANALYSES HAS BEEN GREATLY EXPANDED. + +PRODUCTION CYCLES HAVE BECOME LEANER. + +AstraZeneca, a global pharmaceutical company, wanted to make the +production process for its inhalers more cost-effective and qualitative by +better using the production data. Not an easy task, given the system’s 1,700 +parameters and a total data growth of 1.5 million rows per week. + +With the proper analytics system in place, automated data management +for each production batch has become possible, thus allowing for quick +analysis throughout the manufacturing process. Between 50 and 100 +employees – as diverse as engineers, operators and managers – now create +or receive reports from the system while knowledge sharing is greatly +facilitated. + +“Due to the success of the program, several other products +from the same family which are produced in AstraZeneca’s +Swedish operations have now been included under the +system. We plan to use the same system for completely +different product groups as well.” + +Henrik Åkerblom +Process Engineer at AstraZeneca + +80% + +Analysts at Gartner +estimate that 80% of all +enterprise data today is +unstructured + +Source: Gartner + +45% + +According to an IDG +survey, 45% of business +leaders cite ‛unstructured +data’ as their single +biggest hurdle to +overcome in analytics. + +Source: IDG + +Becoming a data-driven organization + +9 + + 3 + +The three +foundations of +better analytics + + Technological improvements within analytics platforms + +enable more companies to become data-driven, as it + +enables organisations to manage their data better, run more + +complex analyses and visualize the outcome in a more + +understandable manner. Getting the technical foundations + +right puts you well on your way. + +Laying the foundation + +There are three foundations to becoming data-driven. + +1 + +2 + +3 + +DATA MANAGEMENT + +This is the data you use as input. A good analytics platform can process any combination of structured, semi-structured and unstructured data. Automated + +connections between your analytics platform and other systems ensure that the most recent data is always available and used. While not every data point will + +be crystal clear from the start, technical advances in Machine Learning and the like already automate data management, for the most part. Similarly, data inputs + +and data comparisons can be automated, neatly breaking down the obstacle of handling semi- and unstructured data. By applying the right governance + +structure, privacy rules can be applied to personally identifiable information. + +ANALYTICS + +Pouring over endless rows of figures and numbers is the heavy lifting of data science. By leaving this task to specialized software, you leave less room for + +human error and create more room to actually start using the results of your analysis. Complex calculations can now be run by a click of a button, making it + +available to any regular business user. + +DATA VISUALIZATION + +The end result of your analytical work should be smarter insights. By visualizing this in different types of graphics and charts, the outcomes become easily + +understandable for everyone at a glance, while reports and dashboards can quickly be set up, thus opening up the insights to a growing number of people + +across the whole organisation. + +Becoming a data-driven organization + +11 + + 74% + +According to a Forrester +study, 74% of all companies +would like to be more data- +driven, but only 29% claim +that they are actually good +at putting this idea into +action. + +Source: Forrester + +The power of analytics that everyone can use + +These three solid foundations enable you to build an analytics platform +that works for everyone. This has major benefits that help eliminate the +obstacles on your data-driven journey. + + You get clear, actionable results, even from imperfect or +unstructured data. Cleaning up your data and perfecting your input +models can come later. + + Your intelligence is easy to view and understand with visuals that +are more captivating than any line of text could ever be. + + Because the platform is easy to use, self-service reduces reliance on +IT. In turn, IT can focus more on its core business. + +1 + +2 + +3 + + ZOOM-IN ON +RABOBANK + +THE NETHERLANDS | BANKING | OPERATIONS + +TRANSPARENCY THROUGHOUT THE ORGANISATION HAS +INCREASED. DATA VISUALIZATION ENABLES THE BANK TO PROVIDE +PERTINENT INFORMATION AND DIRECT CHAIN MANAGERS MORE +EFFECTIVELY. + +The Rabobank Group, a leading global financial services provider serving +more than 10 million customers and headquartered in the Netherlands, +wanted to optimize its operations by improving the financial and +collaborative alignment across its chains. The company discovered that +there was a huge amount of data available from all groups of the bank’s +organisational chain such as departments, business units and local +branches, but there wasn’t one single system that could integrate and +structure all the information efficiently and provide the ability to share +results. + +With data visualization, large amounts of data are presented visually. The +diverse pictorial or graphical options lead to new questions that weren’t +asked before. The bank is now much more flexible in its ability to provide +information and it can direct chain managers more effectively. At the +same time, employees have become more engaged because they can +quickly see the results of what they do. + +“With the knowledge and access to all chain information, +we are able to let go of old business models and replace +them with more dynamic ones.” + +John Lambrechts +Manager Concern Control at Rabobank + +Becoming a data-driven organization + +12 + + 4 + +The +data-driven +journey + + So where do you start your data-driven journey? Anywhere is + +good, as long as it isn’t everywhere. A ‘big bang’ approach is + +risky: it can overcomplicate things or may simply lack focus. + +We recommend a step-by-step approach as the surest way + +forward to success. + +Plotting the course + +1 + +2 + +3 + +Choose your starting point + +The most important datasets have priority + +Expand the reach of the platform + +This can be a team (e.g. the marketing +department) or a specific data source. Consider +collecting data from your CRM system to get a +better insight into customer behavior, or begin +with productivity data from the shop-floor. + +Recorded customer service calls or data from your +finance department could equally be your first +project. It is best if your starting-point is something +you’re already familiar with, and if you have a clear +goal in mind. + +You must make a distinction between must-have +datasets that will work towards your goal and nice- +to-have datasets that are only loosely related. + +Data may be a mix of structured, semi-structured +and unstructured data. Pour it all in and look at +what your analytics platform comes up with and +whether the results are actionable. Consider if +additional data management (e.g. data cleaning) +is needed or whether you can already continue +using the current datasets. + +Once you have value-added results, you can start +expanding the reach of the platform within and +across teams. This can happen in a series of waves +that create more and more buy-in as the results +begin to show more and more benefits. + +Becoming a data-driven organization + +14 + + Start + +expand +within the team + +Scale + +Grow + +expand +beyond the team + +get the entire +organization on-board + +4 + +The data-driven organisation is born + +Once you have moved from a limited number of data sources to an all- +encompassing data management flow; you’ve extended the reach of +data analytics from the few to the many and every key decision is backed +by data, you’ve truly become a data-driven organisation. You‘ll find that, +as you become better at analytics, you’ll move from hindsight to insight + +to foresight. You’ll no longer simply look back at ‘what happened’, but +you’ll steer your gaze to the future. Not only can you track ROI on all +data-enabled projects, you can also run predictive analytics and simulate +‘what-if’ scenarios. The backbone of your business strategy is now +formed by undisputable facts. + +Becoming a data-driven organization + +15 + + ZOOM-IN ON +Eni + +BELGIUM | ENERGY | MARKETING & OPERATIONS + +a +360° +view + +on your +customers + + ‛What-if’ scenarios, allowing them to assess the impact of strategic +decisions, such as changes in price or margin, reduced customer +churn and more. + + User-friendly dashboards that help to keep an eye on the long-term +profitability of the entire customer base. + +“We calculate how much the customer will spend with us +(revenues) and how long they will stay (retention). We also +predict when we might experience payment issues (credit +losses) with them and how much it will cost us to serve their +needs (service costs).” + +Zdravka Jevtimov +Customer Insights Manager at Eni + +CUSTOMER CHURN AND RETENTION CAN NOW BE PREDICTED. + +FUTURE PROFITABILITY OF PRODUCTS, CHANNELS AND +SEGMENTS CAN BE BETTER EVALUATED UP FRONT. + +Eni is an integrated energy company with operations on five continents. +In the highly competitive energy market, it’s crucial to build long-term +relationships with clients. That’s why the company continuously monitors +and analyses the behaviour of its entire client base throughout the +complete customer life cycle. + +The company developed a solid and trustworthy prediction model using +more than 700 parameters offering Eni’s management: + + Valuable information about customers + + Help in evaluating the future profitability of the company’s product +portfolio, sales channels and customer segments + +60% + +56% + +Analyst firm Gartner estimates that + +According to CMO.com, best-in- + +over half of Big Data projects at + +class marketeers are 56% more likely + +companies fail. One big reason for + +to use data and analytics platforms. + +this is that companies often want + +However, only 19% of marketers fully + +to do everything at once instead of + +track all their marketing efforts with + +focusing on smaller projects with a + +data. + +clear end goal or a quick win. + +Source: CMO.com + +Source: Gartner + +Becoming a data-driven organization + +16 + + 5 + +The next level +in data-driven +work + + Being data-driven is not an end-state. It’s the beginning of an exploration of exciting possibilities. As the pace of technology + +innovation keeps accelerating, yesterday’s science fiction becomes today’s reality. Here are some of the elements that will fuel + +the data-driven organisation of the future. + +On the edge of Tomorrow + +Edge analytics + +Transparency + +Security + +IoT + +AI + +Supply chain + +Real-time on site analytics + +Why not be transparent with + +Analytics can be put to work + +With the wealth of data + +Artificial intelligence (AI) + +Experts can use Big Data to + +can track consumers’ in-store + +consumers about the huge + +for data protection as well. + +generated by machines, your + +makes it possible for + +further optimize logistics. + +behavior and pair it with the + +amounts of data that are + +With advanced pattern + +production lines could map + +machines to learn from + +By taking into account + +right kind of offer bundles to + +collected? Some information + +recognition and correlating + +out ways to become even + +experience, adjust to new + +circumstantial factors that + +attract attention and capture + +will always remain sensitive, + +the needs of the individual. + +but offering transparency + +This effectively creates the + +to your customers can be + +segment of one. + +a big win in the branding + +department. According to + +recent surveys, over 80% +of consumers say ethics +matter when they buy. With +the General Data Protection + +Regulation (GDPR), adhering +to privacy rules has become +an absolute must. + +behaviours, risks can be +assessed better and cyber +attacks or real-life security +threats can be prevented +before they even occur. + +more productive, discover + +inputs and perform human- + +influence delivery speed and + +hidden costs and unlikely + +like tasks. AI relies heavily on + +reliability (e.g. traffic flows, + +sources of revenue. The + +deep learning and natural + +accidents, weather patterns, + +Internet of Things (IoT) will +revolutionize production +as well as consumption + +language processing. +Computers are ‘trained’ to + +rain storms), smarter logistics + +could even be used to + +accomplish specific tasks by + +operate more sustainably. + +patterns. And it’s just around + +processing large amounts + +the corner. + +of data and recognizing + +patterns in that data. + +Becoming a data-driven organization + +18 + + ZOOM-IN ON +INTERAMERICAN + +GREECE | INSURANCE | COMPLIANCE + +BE FULLY GDPR-COMPLIANT. +MAKE A BIG STEP TOWARDS BECOMING A DIGITAL-ONLY INSURER. + +For INTERAMERICAN, a leading insurance provider in Greece, trust +is crucial to retaining loyal customers. The company adopted a data +analytics platform to be fully compliant with the EU General Data +Protection Regulation (GDPR) while also supporting the company’s +strategic focus of transforming itself into a digital-only insurer. + +Its data governance initiative improves the availability, completeness and +accuracy of the information and data being managed. Topics covered +are data ownership, data location, data access, data provenance, risk +assessments and proper recovery procedures in case of breaches. All vital +elements in helping give customers peace of mind that their personal +data will be safe. + +“Our organisation is working to transition to the new digital +age and create long-term, trust-based relationships with our +customers. SAS for Data Protection helps us work towards +compliance with the requirements of the new regulation +and foster customer trust.” + +Xenophon Liapakis +CIO at INTERAMERICAN + +41% + +According to a CFI Group +survey, 41% of all consumers +use mobile apps while +shopping. For Millennials, this +figure even rises to 67%. 51% +of those polled said they would +be likely to use apps if they +made the shopping experience +easier and faster. + +Source: CFI Group + +Over half of organisations +surveyed by IDG say that +they are already deploying +analytics to detect cyber +attacks, denial-of-service +attacks and phishing. +However, 59% among them +still said they have been +compromised at least once +per month “because they +were not able to keep up +and fully analyse the data.” + +Source: IDG + +According to Accenture +research, using Big Data +analytics had a net positive +impact on customer service +and demand fulfilment +for nearly half of all supply +chain experts polled. +Other advantages listed +included greater supply +chain integration (36%), +productivity improvements +(33%) and improved cost to +serve (28%). + +Source: Accenture + +46% + +53% + +Becoming a data-driven organization + +19 + + 6 + +Getting in +on the action + + Becoming a data-driven organisation is now within reach +of every company. Cloud solutions have made access to +data analytics platforms much easier: software-as-a-service +(SaaS) enables organisations to no longer build and +maintain everything on premise but rent this for as long as +needed. And with Results-as-a-Service (RaaS), it becomes +even possible to completely ‘outsource’ your analytics. +If you do not have tools and expertise to turn data into +insights, you can still get results. +Your organization provides the data and the business +problem to be solved, and RaaS delivers results you can +act on. + +By starting your data-driven journey in one specific area +of your company, clearly defining your path towards data +analytics, adopting the right mindset and technologies, +and gradually extending the reach and impact throughout +your company, you too can become the type of data-driven +company that is ready for tomorrow’s challenges. +Start your data-driven journey now and, in no time, you’ll +find yourself wondering: ‘How on earth did we run our +business without analytics?’ + +Becoming a data-driven organization + +21 + + Learn more about the what, why and how of +becoming a data-driven organisation + +Read more + +Follow us: + +For more information, contact us + + SOURCES + +https://www.cio.com/article/3204131/analytics/intelligent-analytics-fuels-faster-smarter-decision-making.html + +http://www.gartner.com/newsroom/id/3130017 + +https://www.retailcustomerexperience.com/news/report-says-most-millennials-are-using-mobile-retail-apps/?utm_source=NetWorld%20Alliance&utm_ + +medium=email&utm_campaign=EMNARCE07022014 + +http://www.sustainablebrands.com/news_and_views/stakeholder_trends_insights/sustainable_brands/study_81_consumers_say_they_will_make_ + +http://www.cmo.com/features/articles/2016/5/31/15-mind-blowing-stats-about-data-driven-marketing.html#gs.uH3Iceg + +https://www.csoonline.com/article/3139923/security/how-big-data-is-improving-cyber-security.html + +https://www.forbes.com/sites/louiscolumbus/2015/07/13/ten-ways-big-data-is-revolutionizing-supply-chain-management/#53f4e2769f59 + +https://www.shopify.com/enterprise/94678726-the-need-for-speed-why-customer-service-needs-to-be-faster-than-ever + +https://reprints.forrester.com/#/assets/2/202/’RES127061’/reports + +https://www.sas.com/nl_nl/training/citizen-data-scientist.html + +https://www.sas.com/en_us/insights/articles/analytics/how-to-find-and-equip-citizen-data-scientists.html + +https://www.forbes.com/sites/forbestechcouncil/2017/06/05/the-big-unstructured-data-problem/#28e176dd493a + +SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. +Other brand and product names are trademarks of their respective companies. Copyright © 2017, SAS Institute Inc. All rights reserved. 109150_G66177.1117 + +Becoming a data-driven organization + +22 diff --git a/q159/random_k4/random_4.md b/q159/random_k4/random_4.md index 190981c6c91fcdd99dab32f3d014a0409752abe1..b8b35dc4cf44586c60769a8500f18bf7ea24ecea 100644 --- a/q159/random_k4/random_4.md +++ b/q159/random_k4/random_4.md @@ -1,1494 +1,3959 @@ -Corporate Governance Report -Reliance Industries Limited Corporate Management Governance Financial Notice -Integrated Annual Report 2019-20 Overview Review Statements -“Between my past, the present and the future, there is one common factor: -Relationship and Trust. This is the foundation of our growth.” -Shri Dhirubhai H. Ambani -Founder Chairman -This report is prepared in accordance The essence of Corporate Governance interests are taken into account, before APPROPRIATE GOVERNANCE ETHICS / GOVERNANCE -with the provisions of the Securities lies in promoting and maintaining integrity, making any business decision. RIL has the STRUCTURE WITH DEFINED POLICIES -and Exchange Board of India (Listing transparency and accountability in the distinction of consistently rewarding its ROLES AND RESPONSIBILITIES At RIL, we strive to conduct our business -Obligations and Disclosure Requirements) management’s higher echelons. The shareholders for over four eventful decades The Company has put in place an internal and strengthen our relationships in -Regulations, 2015 (Listing Regulations), and demands of Corporate Governance require from Initial Public Offer (IPO). Since then, governance structure with defined roles a manner that is dignified, distinctive -the report contains the details of Corporate professionals to raise their competence and RIL has moved from one big idea to another and responsibilities of every constituent of and responsible. We adhere to ethical -K. Sethuraman Savithri Parekh Governance systems and processes at capability levels to meet the expectations and these milestones continue to fuel its the system. The Company’s shareholders standards to ensure integrity, transparency, -Reliance Industries Limited (RIL). in managing the enterprise and its relentless pursuit of ever-higher goals. appoint the Board of Directors, which in independence and accountability in dealing -resources effectively with the highest turn governs the Company. The Board with all the stakeholders. Therefore, we -At RIL, Corporate Governance is all about On standalone basis, we have grown by a -standards of ethics. It has thus become has established various Committees have adopted various codes and policies -maintaining a valuable relationship and Compounded Annual Growth Rate (CAGR) -crucial to foster and sustain a culture to discharge its responsibilities in an to carry out our duties in an ethical manner. -trust with all the stakeholders. We consider of Revenues 22.7%, Earnings Before -that integrates all components of good effective manner. The Chairman and Some of these codes and policies are: -stakeholders as partners in our success Interest, Tax, Depreciation and Amortisation -governance by carefully balancing the Managing Director (CMD) provides overall -and remain committed to maximising (EBITDA) before exceptional items 24.3% • Code of Conduct and Our Code -complex inter-relationship among the Board direction and guidance to the Board. In -stakeholders’ value, be it Customers, Local and Net Profit before exceptional items • Code of Conduct for Prohibition of -Jyoti Jain Sridhar Kothandaraman of Directors, Board Committees, Finance, the operations and functioning of the -Communities, Employees, Suppliers, Trade 25.1%. The financial markets have endorsed Insider Trading -Compliance & Assurance teams, Auditors Company, the CMD is assisted by four -Unions, NGOs, Investors & Shareholders and our sterling performance and the market • Code of Practices and Procedures for -and the senior management. Our employee Executive Directors and a core group -Government & Regulatory Authorities. This capitalisation has increased by CAGR of Fair Disclosure of Unpublished Price -satisfaction is reflected in the stability of of senior level executives. The CMD is -approach to value creation emanates from 30.4% during the same period. In terms of Sensitive Information -our senior management, low attrition across responsible for corporate strategy, brand -RIL’s belief that sound governance system, distributing wealth to our shareholders, • Health, Safety and -various levels and substantially higher equity, planning, external contacts and all -based on relationship and trust, is integral apart from having a track record of Environment (HSE) Policy -productivity. Above all, we feel honoured management matters. -to creating enduring value for all. We have a uninterrupted dividend payout, we have • Vigil Mechanism and -to be integral to India’s social development. -defined policy framework for ethical conduct also delivered consistent unmatched The Chairman is responsible for fostering Whistle-blower Policy -Ratnesh Rukhariyar Details of several such initiatives are -of businesses. We believe that any business shareholder returns since listing. The result and promoting the integrity of the • Prevention of Sexual Harassment Policy -available in the Report on Corporate Social -conduct can be ethical only when it rests of our initiative is our ever widening reach Board while nurturing a culture where • Corporate Social Responsibility Policy -Responsibility. -on the six core values viz. Customer Value, and recall. Our shareholder base has grown the Board works harmoniously for the • Policy for selection of Directors and -Ownership Mindset, Respect, Integrity, One At RIL, we believe that as we move closer from 52,000 after the IPO to a consolidated long-term benefit of the Company and all determining Directors’ independence -“At Reliance Industries Limited, -Team and Excellence. towards our aspirations of being a global present base of around 26 lakh. its stakeholders. The Chairman guides • Remuneration Policy for Directors, -we are committed to create an -corporation, our Corporate Governance the Board for effective governance Key Managerial Personnel and -agile organisation promoting For decades, RIL is growing in step -STATEMENT ON COMPANY’S standards must be globally benchmarked. in the Company. other employees -practices that uphold governance with India’s industrial and economic -PHILOSOPHY ON CODE OF Therefore, we have institutionalised the • Dividend Distribution Policy -and ensure business continuity.” development. The Company has helped The Chairman takes a lead role in -GOVERNANCE right building blocks for future growth. • Policy for determining Material Subsidiaries -transform the Indian economy with large managing the Board and facilitating -Corporate Governance encompasses a set The building blocks will ensure that we • Policy on Subsidiary Governance -projects and world-class execution. The effective communication among Directors. -of systems and practices to ensure that achieve our ambition in a prudent and • Policy on Materiality of Related Party -quest to help elevate India’s quality of life The Chairman actively works with the -the Company’s affairs are being managed sustainable manner. RIL not only adheres Transactions and on dealing with -continues and is unabated. It emanates Human Resources, Nomination and -in a manner which ensures accountability, to the prescribed Corporate Governance Related Party Transactions -from a fundamental article of faith: ‘What is Remuneration Committee to plan the -transparency and fairness in all transactions practices as per the Listing Regulations, • Policy on Determination and Disclosure -good for India is good for Reliance’. Board and committees’ composition, -in the widest sense. The objective is to but is also committed to sound Corporate of Materiality of Events and Information -induction of directors to the Board, plan -meet stakeholders’ aspirations and societal Governance principles and practices. It We believe, Corporate Governance is and Web Archival Policy -for directors’ succession and provide -expectations. Good governance practices constantly strives to adopt emerging best not just a destination, but a journey to • Policy for Preservation of Documents -constructive feedback and advice on -stem from the dynamic culture and positive practices being followed worldwide. It is constantly improve sustainable value • Group Risk Management Policy -performance evaluation to directors. -mindset of the organisation. We are our endeavour to achieve higher standards creation. It is an upward-moving target that • Materiality Policy for Commodity exposure -The Company Secretary assists the -committed to meet the aspirations of all and provide oversight and guidance to the we collectively strive towards achieving. • Commodity and Freight Risk -Chairman in management of the Board’s -our stakeholders. This is demonstrated in management in strategy implementation, Our multiple initiatives towards maintaining Management Policy -administrative activities such as meetings, -shareholder returns, high credit ratings, risk management and fulfilment of stated the highest standards of governance are • Foreign Exchange and Derivatives Risk -schedules, agendas, communications and -awards and recognitions, governance goals and objectives. detailed in this Report. Management Policy -documentations. -processes and an entrepreneurial • Investment Governance Policy -Over the years, we have strengthened -performance focussed work environment. • Data Privacy Policy -governance practices. These practices -Additionally, our customers have benefited • Information Security Policy -define the way how business is conducted -from high quality products delivered at • Intellectual Property Policy -and value is generated. Stakeholders’ -extremely competitive prices. -184 Naye India Ka Naya Josh 185 - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | Integrated Annual Report 2019-20 | | Overview Review | Statements | | -| --- | --- | --- | -------------------------------- | --- | --------------- | ---------- | --- | -AUDITS AND INTERNAL CHECKS use and protection of assets, facilitate the (IIRC) framework. The concept of the BOARD OF DIRECTORS of governance. The Board’s actions and framework for the meetings of the Board -AND BALANCES accurate and timely compilation of financial six capitals of business as suggested by BOARD LEADERSHIP decisions are aligned with the Company’s and Committees. These guidelines seek to -S R B C & CO LLP, Chartered Accountants statements and management reports. the framework has been ingrained into At RIL, it is our belief that an enlightened best interests. The Board is committed systematise the decision-making process at -and D T S & Associates LLP, Chartered the Company’s management philosophy Board consciously creates a culture of to the goal of sustainably elevating the the meetings of the Board and Committees -Accountants, are the Statutory Auditors BEST CORPORATE and has become an important enabler for leadership to provide a long-term vision Company’s value creation. The Company in an informed and efficient manner. -of the Company. The Statutory Auditors GOVERNANCE PRACTICES RIL’s value creation story. RIL’s Sustainability and policy approach to improve the quality has defined guidelines and an established -| and the Group Internal Audit Function | RIL strives for highest Corporate | Reporting Journey and its Integrated | | | | | | -| -------------------------------------- | ---------------------------------- | ------------------------------------- | --- | --- | --- | --- | --- | -perform independent reviews of the Governance standards and practices. It, Reporting is covered in Management BOARD COMPOSITION AND CATEGORY OF DIRECTORS -ongoing effectiveness of the Reliance therefore, endeavours to continuously Discussion and Analysis Report. The Company’s policy is to maintain optimum combination of Executive and Non-Executive Directors. -| Management System which integrates | improve and adopt the best of | | | | | | | -| ----------------------------------- | ------------------------------ | --- | --- | --- | --- | --- | --- | -The composition of the Board, Category, DIN and shareholding of Directors are as follows: -| the various components of the systems of | international Corporate Governance codes | SHAREHOLDERS’ | | | | | | -| ----------------------------------------- | ----------------------------------------- | -------------- | --- | ---- | ------------------------ | ----------------------------- | --- | -| internal control. | and practices. Some of the implemented | COMMUNICATIONS | | | | | | -| | | | | Sr. | Director Identification | No. of equity shares held as | | -Name of the Director Category -global governance norms and best The Board recognises the importance of No. Number (DIN) on March 31, 2020 -RISK MANAGEMENT, INTERNAL practices include the following: two-way communication with shareholders, 1 Mukesh D. Ambani 00001695 75,00,000 -CONTROLS AND COMPLIANCE giving a balanced report of results and (Chairman and Managing Director) -• All securities related filings with Promoter Directors -The Company has put in place the “Reliance progress and responding to questions 2 Nita M. Ambani 03115198 75,00,000 -Stock Exchanges are reviewed -Management System” (RMS) as a part of its and issues raised. RIL’s corporate (Non-Executive Director) -every quarter by the Stakeholders’ 3 Yogendra P. Trivedi 00001879 60,400 -| transformation agenda. RMS incorporates | | website (www.ril.com) has information for | | | | | | -| ---------------------------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | -Relationship Committee. -an integrated framework for managing institutional and retail shareholders alike. 4 Prof. Dipak C. Jain 00228513 - -• The Company has independent Board -risks and internal controls. The internal Shareholders seeking information related 5 Dr. Raghunath A. Mashelkar 00074119 - -Committees covering matters related to -financial controls have been documented, to their shareholding may contact the 6 Adil Zainulbhai 06646490 - -Risk Management, HSE, Internal Audit, Non-Executive Directors -embedded and digitised in the business Company directly or through the Company’s 7 Raminder Singh Gujral 07175393 12,000 -Stakeholder Relationship, Directors’ 8 Dr. Shumeet Banerji 02787784 13,500 -| processes. Internal controls are regularly | | Registrar and Transfer Agents, details of | | | | | | -| ------------------------------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | -Remuneration and the nomination 9 Arundhati Bhattacharya 02011213 - -| tested for design, implementation and | | which are available on the Company’s | | | | | | -| ---------------------------------------- | ----------------- | ------------------------------------- | --- | ----------------- | -------- | --- | --- | -| | of Board members. | | | 10 K. V. Chowdary | 08485334 | | - | -| operating effectiveness. RMS is enabled | | website. RIL ensures that complaints | | | | | | -• The Company also has several -through extensive use of technology to of its shareholders are responded to 11 Nikhil R. Meswani 00001620 33,56,748 -other Executive committees of -support the risk management processes, promptly. A comprehensive and informative 12 Hital R. Meswani 00001623 32,23,772 -Executive Directors -senior management who review the 13 P. M. S. Prasad 00012144 6,00,000 -| ensure the ongoing effectiveness of internal | | shareholders’ referencer is available on the | | | | | | -| --------------------------------------------- | --- | --------------------------------------------- | --- | --- | --- | --- | --- | -ongoing effectiveness of operational 14 Pawan Kumar Kapil 02460200 53,000 -| controls in processes, compliance with | | website of the Company. | | | | | | -| --------------------------------------- | ----------------------------------- | ----------------------- | --- | ---------------------------------------------------------------- | --- | --- | --- | -| | and financial risk mitigations and | | | Board members named at Sr. No. 3 to 9 are Independent Directors. | | | | -applicable laws and regulations. -governance practices. -| | | ROLE OF THE COMPANY | | Smt. Nita M. Ambani is the spouse of Shri Mukesh D. Ambani. | | | | -| --------------------------------- | ----------------------------------------- | -------------------- | --- | ----------------------------------------------------------- | --- | --- | --- | -| In conformity with international | • The Group has an independent Internal | | | | | | | -SECRETARY IN OVERALL -standards, the Compliance Function Audit Function that provides risk-based GOVERNANCE PROCESS Shri Nikhil R. Meswani and Shri Hital R. Meswani, are brothers and not related to Promoter Directors. None of the other directors are -ensures compliance activities related assurance across all material areas of related to any other director on the Board. -Functions of of the Company Secretary -| to the Financial, Operational and People | Group Risk and Compliance exposures. | | | | | | | -| ----------------------------------------- | ------------------------------------ | --- | --- | --- | --- | --- | --- | -are discharged by the Group Company Shri Mansingh L. Bhakta (88 years), demitted office as an Independent Director of the Company, w.e.f. August 12, 2019, on account of his -| Management Systems of the various group | • The Company undergoes quarterly | | | | | | | -| ---------------------------------------- | ----------------------------------- | --- | --- | --- | --- | --- | --- | -Secretary and the Joint Company Secretary. advanced age. Other than this, there was no other material reason for demitting office as an Independent Director of the Company. -| entities. This includes covering various | secretarial compliance certification from | | | | | | | -| ----------------------------------------- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -The Company Secretary plays a key role -statutes such as industrial and labour laws, an independent company secretary who DIRECTORS’ PROFILE CODE OF CONDUCT -in ensuring that the Board (including of major subsidiaries, global business -| taxation laws, corporate and securities | is in whole-time practice. | | | | | | | -| ---------------------------------------- | -------------------------- | --- | --- | --- | --- | --- | --- | -committees thereof) procedures are A brief resume of the Directors, nature environment, business strategy and risks The Company has in place a comprehensive -| laws, health, safety and environmental laws, | • The Company has appointed an | | | | | | | -| --------------------------------------------- | -------------------------------- | --- | --- | --- | --- | --- | --- | -followed and regularly reviewed. The of their expertise in specific functional involved. Detailed presentations on the Code of Conduct and Our Code (the Codes) -| etc. All compliance activities are supported | independent firm of Chartered | | | | | | | -| --------------------------------------------- | ------------------------------ | --- | --- | --- | --- | --- | --- | -Company Secretary ensures that all relevant areas etc. are available on the website Company’s business segments are made applicable to the Directors and employees. -| by a robust online compliance monitoring | Accountants to conduct concurrent audit | | | | | | | -| ----------------------------------------- | ---------------------------------------- | --- | --- | --- | --- | --- | --- | -information, details and documents are of the Company. in separate meetings of the Independent The Codes give guidance and support -| system (iRCMS) to ensure ongoing | of share transfer and other incidental | | | | | | | -| --------------------------------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -made available to the Directors and senior Directors from time to time. needed for ethical conduct of business and -| compliance. The ongoing effectiveness | functions carried out by Registrar and | | | | | | | -| -------------------------------------- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -management for effective decision-making FAMILIARISATION PROGRAMMES compliance of law. The Codes reflect the -of compliance management activities is Transfer Agents. Monthly / quarterly updates on relevant -at the meetings. The Company Secretary FOR BOARD MEMBERS core values of the Company viz. Customer -reviewed independently by the Group statutory, regulatory changes and landmark -RIL’S SUSTAINABILITY is primarily responsible to assist and The Board members are provided with Value, Ownership Mindset, Respect, -Audit Function. judicial pronouncements encompassing -REPORTING JOURNEY AND advice the Board in the conduct of affairs necessary documents / brochures, Integrity, One Team and Excellence. -INTEGRATED REPORTING important laws are regularly circulated -The combination of independent of the Company, to ensure compliance reports and internal policies to enable -to the Directors. Visits to various plant -governance, assurance and oversight RIL commenced its Sustainability Reporting with applicable statutory requirements, them to familiarise with the Company’s -locations are organised for the Independent -structures, combined with automated risk in the FY 2004-05, based on the Global to provide guidance to directors and to procedures and practices. -Directors to enable them to understand -| management, controls and compliance | Reporting Initiative’s (GRI) reporting | facilitate convening of meetings. The | | | | | | -| ------------------------------------ | --------------------------------------- | -------------------------------------- | --- | --- | --- | --- | --- | -Periodic presentations are made at the and get acquainted with the operations -| monitoring, ensures the ongoing | guidelines. The reports are available on the | Company Secretary interfaces between the | | | | | | -| -------------------------------- | --------------------------------------------- | ----------------------------------------- | --- | --- | --- | --- | --- | -Board and Committee meetings on of the Company. -| robustness and integrity of financial | website of the Company. | management and regulatory authorities for | | | | | | -| -------------------------------------- | ----------------------- | ------------------------------------------ | --- | --- | --- | --- | --- | -business and performance updates of -reporting, management of internal controls governance matters. The details of such familiarisation -| | RIL published its maiden Integrated Annual | | | the Company including Finance, Sales, | | | | -| --- | ------------------------------------------- | --- | --- | -------------------------------------- | --- | --- | --- | -and ensures compliance with statutory laws, programmes for the Independent -| | Report in the FY 2016-17 aligned with the | | | Marketing of the Company’s major business | | | | -| --- | ------------------------------------------ | --- | --- | ------------------------------------------ | --- | --- | --- | -regulations and company’s policies. These Directors are available on the website -| | International Integrated Reporting Council’s | | | segments, overview of business operations | | | | -| --- | --------------------------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | -provide the foundations that enable optimal of the Company. -| 186 | | | | | | | 187 | -| --- | --- | --- | --- | --- | --- | --- | --- | -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | | Integrated Annual Report 2019-20 | | | Overview Review | Statements | -| ----------------------------------------- | --- | ----------------------------------------- | --- | ------------------------------------------ | -------------------------------- | ------------------------------------------- | ------------------------------------------ | ------------------------------------------ | ---------- | -| | | | | | | SELECTION OF INDEPENDENT | LEAD INDEPENDENT DIRECTOR | The Board provides and evaluates the | | -| | | | | | | DIRECTORS | The Board of Directors of the Company | Company’s strategic direction, management | | -| A copy of the Code of Conduct and Our | | are very important for creating a robust | | The following skills / expertise / | | | | | | -| | | | | | | Considering the requirement of skill sets | way back in October 2005 introduced the | policies and their effectiveness, and | | -| Code are available on the website of the | | future for the Company. The Human | | competencies have been identified for the | | | | | | -| | | | | | | on the Board, eminent people having an | concept of Lead Independent Director and | ensures that shareholders’ long-term | | -| Company. The Codes have been circulated | | Resources, Nomination and Remuneration | | effective functioning of the Company and | | | | | | -| | | | | | | independent standing in their respective | had designated Shri Mansingh L. Bhakta as | interests are being served. | | -| to the Directors and Senior Management | | Committee works along with the Human | | are currently available with the Board: | | | | | | -| | | | | | | field / profession and who can effectively | a Lead Independent Director. The Role of | | | -Personnel, and its compliance is affirmed Resource team of the Company for a The Board has constituted seven main -| | | | | • Leadership / Operational experience | | contribute to the Company’s business | the Lead Independent Director is as under: | | | -| --- | --- | --- | --- | -------------------------------------- | --- | ------------------------------------- | ------------------------------------------ | --- | --- | -by them annually. structured leadership succession plan. Committees, viz. Audit Committee, Human -| | | | | • Strategic Planning | | and policy decisions are considered by | | | | -| --- | --- | --- | --- | --------------------- | --- | --------------------------------------- | --- | --- | --- | -• Presides over meetings of the Resources, Nomination and Remuneration -A declaration on confirmation of compliance CORE SKILLS / EXPERTISE / • Industry Experience, Research & the Human Resources, Nomination and -Independent Directors Committee, Stakeholders’ Relationship -of the Code of Conduct, signed by the Development and Innovation Remuneration Committee, for appointment, -COMPETENCIES AVAILABLE • Ensures adequacy and timely flow of Committee, Corporate Social Responsibility -Company’s Chairman and Managing • Global Business as an Independent Director on the Board. -WITH THE BOARD information to the Independent Directors and Governance Committee, Risk -Director is published in this Report. • Financial, Regulatory / Legal & The Committee inter alia considers -The Board comprises of qualified members • Liaises between the Chairman and Management Committee, Health, Safety -| | | | | Risk Management | | qualification, positive attributes, area of | | | | -| --- | --- | --- | --- | --------------- | --- | -------------------------------------------- | --- | --- | --- | -who possess required skills, expertise and Managing Director, the Management and and Environment Committee and Finance -SUCCESSION PLANNING -| | | | | • Corporate Governance | | expertise and number of Directorship(s) and | | | | -| --- | --- | --- | --- | ----------------------- | --- | -------------------------------------------- | --- | --- | --- | -competencies that allow them to make the Independent Directors Committee and is authorised to constitute -The Company believes that sound Membership(s) held in various committees -effective contributions to the Board and • Presides over meetings of the Board other functional Committees, from time to -succession plans for the senior leadership of other companies by such persons in -its Committees. and Shareholders when the Chairman time, depending on business needs. -accordance with the Company’s Policy for -and Managing Director is not present or -While all the Board members possess the skills identified, their area of core expertise is given below: Selection of Directors and determining The Company’s internal guidelines for -where he is an interested party -Name of the Name of the Directors’ independence. The Board Board / Committee meetings facilitate -Area of Expertise Area of Expertise • Perform such other duties as -Director Director considers the Committee’s recommendation decision-making process at its meetings in -may be delegated to the Lead -Leadership / Operational experience Leadership / Operational experience and takes appropriate decision. an informed and efficient manner. -Independent Director by the Board / -| | Strategic Planning | | | Strategic Planning | | | | | | -| --- | ------------------ | --- | --- | ------------------ | --- | --- | --- | --- | --- | -Arundhati Every Independent Director, at the first Independent Directors. NUMBER OF BOARD MEETINGS -| | Industry Experience, Research & Development | | | Global Business | | | | | | -| --- | -------------------------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | -Bhattacharya meeting of the Board in which he / she Seven Board meetings were held during -Mukesh D. Ambani and Innovation Financial, Regulatory / Legal & Risk Management MEETINGS OF INDEPENDENT -Global Business Corporate Governance participates as a Director and thereafter the financial year, as against the statutory -DIRECTORS -Financial, Regulatory / Legal & Risk Management at the first meeting of the Board in every requirement of four meetings. The details of -| | | | | Leadership / Operational experience | | | The Company’s Independent Directors met | | | -| --- | --- | --- | --- | ----------------------------------- | --- | --- | ---------------------------------------- | --- | --- | -financial year, gives a declaration that Board meetings held are given below: -Corporate Governance Strategic Planning two times during the financial year 2019-20. -| | | | K. V. Chowdary | | | he / she meets the criteria of independence | | | | -| --- | --- | --- | -------------- | --- | --- | -------------------------------------------- | --- | --- | --- | -Leadership / Operational experience Financial, Regulatory / Legal & Risk Management Such meetings were conducted to enable -as provided under the law and that No. of -Industry Experience, Research & Development Corporate Governance Board -| | | | | | | | the Independent Directors to discuss | Date | Directors | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | ---- | ---------- | -Yogendra P. Trivedi and Innovation Leadership / Operational experience he / she is not aware of any circumstance or Strength -matters pertaining to the Company’s affairs Present -Financial, Regulatory / Legal & Risk Management Nita M. Ambani Strategic Planning situation, which exist or may be reasonably -| | | | | | | | and put forth their views. | April 18, 2019 | 14 13 | -| --- | --- | --- | --- | --- | --- | --- | -------------------------- | -------------- | ----- | -Corporate Governance Corporate Governance anticipated, that could impair or impact -July 19, 2019 14 14 -Leadership / Operational experience his / her ability to discharge his / her duties -Leadership / Operational experience BOARD MEETINGS, COMMITTEE October 18, 2019 14 14 -| | Strategic Planning | | | | | with an objective independent judgement | | | | -| --- | ------------------ | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | -Strategic Planning MEETINGS AND PROCEDURES October 25, 2019 14 14 -| | Industry Experience, Research & Development | | | Industry Experience, Research & Development | | and without any external influence. | | | | -| --- | -------------------------------------------- | --- | --- | -------------------------------------------- | --- | ----------------------------------- | --- | --- | --- | -Prof. Dipak C. Jain INSTITUTIONALISED DECISION- December 16, 2019 14 14 -| | and Innovation | | Nikhil R. Meswani | and Innovation | | | | | | -| --- | ----------------------------------- | --- | ----------------- | ----------------------------------------------- | --- | --------------------------------------------- | ------------------------------------------- | ----------------- | ----- | -| | | | | | | In the opinion of the Board, the | MAKING PROCESS | January 17, 2020 | 14 14 | -| | Global Business | | | Global Business | | | | | | -| | | | | | | Independent Directors fulfil the conditions | The Board of Directors is the apex body | February 17, 2020 | 14 13 | -| | Corporate Governance | | | Financial, Regulatory / Legal & Risk Management | | | | | | -| | | | | | | specified in the Listing Regulations and are | constituted by shareholders for overseeing | | | -| | Leadership / Operational experience | | | Corporate Governance | | | | | | -Strategic Planning Leadership / Operational experience independent of the management. the Company’s overall functioning. -| Dr. Raghunath A. | Industry Experience, Research & Development | | | | | | | | | -| ----------------- | -------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Strategic Planning -| Mashelkar | and Innovation | | | | | | | | | -| --------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Industry Experience, Research & Development -| | Financial, Regulatory / Legal & Risk Management | | Hital R. Meswani | and Innovation | | | | | | -| --------------- | ----------------------------------------------- | --- | ---------------- | ----------------------------------------------- | --- | --- | --- | --- | --- | -| | Corporate Governance | | | Global Business | | | | | | -| | Leadership / Operational experience | | | Financial, Regulatory / Legal & Risk Management | | | | | | -| | Strategic Planning | | | Corporate Governance | | | | | | -| | Industry Experience, Research & Development | | | Leadership/ Operational experience | | | | | | -| Adil Zainulbhai | and Innovation | | | Strategic Planning | | | | | | -Global Business -Industry Experience, Research & Development -Financial, Regulatory / Legal & Risk Management -| | | | P. M. S. Prasad | and Innovation | | | | | | -| --------------- | ----------------------------------------------- | --- | --------------- | ----------------------------------------------- | --- | --- | --- | --- | --- | -| | Corporate Governance | | | Global Business | | | | | | -| | Leadership / Operational experience | | | Financial, Regulatory / Legal & Risk Management | | | | | | -| Raminder Singh | Strategic Planning | | | Corporate Governance | | | | | | -| Gujral | Financial, Regulatory / Legal & Risk Management | | | Leadership / Operational experience | | | | | | -| | Corporate Governance | | | Industry Experience, Research & Development | | | | | | -Pawan Kumar Kapil -| | Leadership/ Operational experience | | | and Innovation | | | | | | -| --- | ---------------------------------- | --- | --- | -------------- | --- | --- | --- | --- | --- | -Strategic Planning -Financial, Regulatory / Legal & Risk Management -| Dr. Shumeet Banerji | Global Business | | | | | | | | | -| ------------------- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Financial, Regulatory / Legal & Risk Management -Corporate Governance -188 189 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | Integrated Annual Report 2019-20 | | Overview Review | Statements | -| --- | --- | --- | --- | -------------------------------- | --- | --------------- | ---------- | -Attendance of Directors at Board Meetings, last Annual General Meeting (AGM) and No. of other COMMITTEES -Directorship(s) and Chairmanship(s) / Membership(s) of Committees of each Director in various Companies: DETAILS OF THE COMMITTEES AND OTHER RELATED INFORMATION ARE PROVIDED HEREUNDER: -Composition of Committees of the Company: -No. of -| | | | | | Audit Committee | Human Resources, Nomination and Remuneration Committee | | -| --- | --- | ------------- | --- | ---------------- | --------------- | ------------------------------------------------------ | --- | -| | | No. of other | | Membership(s) / | | | | -Attendance at Category of Directorship and name of 1. Y ogendra P. Trivedi 1. Adil Zainulbhai -| | Directorship(s) | | | Chairmanship(s) | | | | -| --- | ----------------- | --- | ----------------------------- | ---------------- | --- | --- | --- | -| | meetings during | | the other listed Company(s) | | | | | -Name of the Director as on of committees in (Chairman of the Committee) (Chairman of the Committee) -| | 2019-20 | | as on 31-03-2020 | | | | | -| --- | ------- | --- | ---------------- | --- | --- | --- | --- | -31-03-2020 other Company(s) 2. Dr. Raghunath A. Mashelkar 2. Yogendra P. Trivedi -| | | | | as on 31-03-2020 | 3. Adil Zainulbhai | 3. Dr. Raghunath A. Mashelkar | | -| ---------------- | --------- | --- | --- | ---------------- | -------------------------- | ------------------------------- | --- | -| | | | | | 4. Raminder Singh Gujral | 4. Raminder Singh Gujral | | -| | Board AGM | (1) | | (2) | | | | -| | | | | | 5. K. V. Chowdary | 5. Dr. Shumeet Banerji | | -| Mukesh D. Ambani | 7 Yes | 3 | Nil | Nil | | | | -6. K. V. Chowdary -| Mansingh L. Bhakta* | 1 Yes | NA | NA | NA | | | | -| ------------------- | ----- | --- | -------------------------------------- | ------------------ | ------------------------------------ | -------------------------------------------------------- | --- | -| | | | | | Stakeholders’ Relationship Committee | Corporate Social Responsibility and Governance Committee | | -| Yogendra P. Trivedi | 7 Yes | 4 | 1. Z odiac Clothing Company Limited – | 3 (including 2 as | | | | -| | | | | | 1. Y ogendra P. Trivedi | 1. Y ogendra P. Trivedi | | -| | | | Independent Director | Chairman) | | | | -| | | | | | (Chairman of the Committee) | (Chairman of the Committee) | | -2. T he Supreme Industries Limited – -| | | | | | 2. Arundhati Bhattacharya | 2. Dr. Raghunath A. Mashelkar | | -| --- | --- | --- | --- | --- | --------------------------- | ------------------------------- | --- | -Independent Director -| | | | | | 3. K. V. Chowdary | 3. Dr. Shumeet Banerji | | -| --- | --- | --- | --- | --- | ------------------- | ------------------------ | --- | -3. E mami Limited – Independent Director -| | | | | | 4. Nikhil R. Meswani | 4. Nikhil R. Meswani | | -| ------------------- | ----- | --- | --- | --- | ---------------------- | ---------------------- | --- | -| Prof. Dipak C. Jain | 7 Yes | 3 | Nil | 2 | | | | -5. Hital R. Meswani -Dr. Raghunath A. Mashelkar 7 Yes 6 1. G odrej Agrovet Limited – Independent 1 Risk Management Committee Health, Safety and Environment Committee -| | | | Director | | 1. Adil Zainulbhai | 1. H ital R. Meswani | | -| --- | --- | --- | -------- | --- | ---------------------- | ------------------------ | --- | -2. P iramal Enterprises Limited – Independent (Chairman of the Committee) (Chairman of the Committee) -Director -| | | | | | 2. Dr. Shumeet Banerji | 2. Dr. Raghunath A. Mashelkar | | -| --------------- | ----- | --- | --------------------------------------- | ------------------ | ------------------------ | ------------------------------- | --- | -| Adil Zainulbhai | 7 Yes | 7 | 1. Cipla Limited – Independent Director | 8 (including 5 as | | | | -| | | | | | 3. K. V. Chowdary | 3. Arundhati Bhattacharya | | -Chairman) -2. N etwork18 Media & Investments Limited – 4. Hital R. Meswani 4. P. M. S. Prasad -Independent Director 5. P. M. S. Prasad 5. Pawan Kumar Kapil -6. Alok Agarwal -3. T V18 Broadcast Limited – Independent -Chief Financial Officer -Director -7. Srikanth Venkatachari -4. L arsen & Tourbo Limited – Independent -Joint Chief Financial Officer -Director -Finance Committee -| Raminder Singh Gujral | 7 Yes | 2 | 1. A dani Power Limited – Independent Director | 2 | | | | -| --------------------- | ----- | --- | ---------------------------------------------- | --- | --- | --- | --- | -1. M ukesh D. Ambani -| Dr. Shumeet Banerji | 6 Yes | 2 | Nil | Nil | | | | -| ------------------- | ----- | --- | --- | --- | --- | --- | --- | -(Chairman of the Committee) -| Arundhati Bhattacharya | 7 Yes | 5 | 1. P iramal Enterprises Limited – Independent | 3 (including 1 as | | | | -| ---------------------- | ----- | --- | ---------------------------------------------- | ------------------ | --- | --- | --- | -2. Nikhil R. Meswani -| | | | Director | Chairperson) | | | | -| --- | --- | --- | -------- | ------------ | --- | --- | --- | -3. Hital R. Meswani -2. Wipro Limited – Independent Director -3. CRISIL Limited – Independent Director The composition of the Committees is in accordance with the provisions of the Listing Regulations and the Companies Act, 2013. -| K. V. Chowdary# | 5 NA | 2 | 1. C CL Products (India) Limited – Independent | 3 | | | | -| --------------- | ---- | --- | ------------------------------------------------ | --- | --- | --- | --- | -K. Sethuraman, Group Company Secretary and Chief Compliance Officer and Savithri Parekh, Joint Company Secretary and Compliance -Director -Officer, are the secretaries of all the Committees constituted by the Board. -2. D ivi’s Laboratories Limited – Independent -Director -| Nita M. Ambani | 7 Yes | 2 | 1. E IH Limited – Non-Executive Director | Nil | | | | -| ----------------- | ----- | --- | --------------------------------------------- | --------------- | --- | --- | --- | -| Nikhil R. Meswani | 7 Yes | 1 | Nil | 1 (as Chairman) | | | | -| Hital R. Meswani | 7 Yes | 3 | Nil | 1 (as Chairman) | | | | -| P. M. S. Prasad | 7 Yes | 4 | 1. N etwork18 Media & Investments Limited – | 4 | | | | -Non-Executive Director -2. T V18 Broadcast Limited – Non-Executive -Director -| Pawan Kumar Kapil | 7 Yes | 1 | Nil | Nil | | | | -| ----------------- | ----- | --- | --- | --- | --- | --- | --- | -NA – Not Applicable -(1) T he Directorships, held by Directors as mentioned above, do not include directorship(s) in foreign companies and Section 8 companies under the -Companies Act, 2013. -(2) I n accordance with Regulation 26 of the Listing Regulations, Membership(s) / Chairmanship(s) of only Audit Committees and Stakeholders’ Relationship -Committees in all public limited companies have been considered. -* Ceased to be a Director, w.e.f. August 12, 2019. Two meetings were held during his tenure. -# Appointed as a Director, w.e.f. October 18, 2019. Five meetings were held since his appointment. -Video / tele-conferencing facility is offered to facilitate the Directors to participate in the meetings. -The number of Directorship(s) and Committee Membership(s) / Chairmanship(s) of all Directors is / are within the respective limits prescribed under the -Companies Act, 2013 and the Listing Regulations. -190 191 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | | | Integrated Annual Report 2019-20 | | | | | Overview | Review | Statements | -| --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | -------- | ------ | ---------- | -Meetings of Committees held during the year and members’ attendance: Meeting Details Stakeholders’ Relationship Committee Corporate Social Responsibility and -| | | | | | | | | Twelve meetings of the Committee | The terms of reference of the Committee is | | | Governance Committee | | -| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------- | ------------------------------------------- | --- | --- | -------------------- | --- | -Human -Corporate -Resources, were held during the year, as against available on the website of the Company. The terms of reference of the Committee is -| | | | Social Stakeholders’ | | Health, | | Risk | | | | | | | -| --- | --- | --- | ---------------------- | --- | -------- | --- | ----- | --- | --- | --- | --- | --- | --- | -Nomination the statutory requirement of four available on the website of the Company. -Committees of the Audit Responsibility Relationship Safety and Finance Management The Chairman of the Committee was -| | | and | | | | | | meetings. The meetings were held | | | | | | -| --- | --- | ---- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | --- | -Company Committee and (SR) Environment Committee (RM) present at the last Annual General Meeting Meeting Details -| | Remuneration | | | | | | | on April 5, 2019; April 17, 2019; April | | | | | | -| --- | ------------- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | -Governance Committee Committee Committee held on August 12, 2019. Four meetings of the Committee were -| | | (HRNR) | | | | | | 18, 2019; July 18, 2019; July 19, 2019; | | | | | | -| --- | --- | ------- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | -Committee -| | Committee | | | | | | | | | | | held during the year. The meetings were | | -| --- | --------- | --- | --- | --- | --- | --- | --- | ------------------------------------ | --------------- | --- | --- | ---------------------------------------- | --- | -| | | | | | | | | October 14, 2019; October 18, 2019; | Meeting Details | | | | | -held on April 18, 2019; August 14, 2019; -Meetings held 12 4 4 4 4 7 4 October 25, 2019; December 16, 2019; Four meetings of the Committee were -October 17, 2019 and January 16, 2020. -Directors’ Attendance January 15, 2020; January 17, 2020 held during the year, as against statutory -The details of attendance of Committee -| | | | | | | | | and February 17, 2020. The details of | requirement of one meeting. The | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------------- | -------------------------------- | --- | --- | --- | --- | -Mukesh D. Ambani NA NA NA NA NA 7 NA members are given in this Report. -| | | | | | | | | attendance of Committee members are | meetings were held on April 25, 2019; | | | | | -| ------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------ | -------------------------------------------- | --- | --- | ------------------------------------------- | --- | -| Mansingh L. Bhakta | NA | NA | NA | NA | NA | NA | NA | | | | | | | -| | | | | | | | | given in this Report. | August 14, 2019; October 25, 2019 and | | | Risk Management Committee | | -| Yogendra P. Trivedi | 12 | 4 | 4 | 4 | NA | NA | NA | | | | | | | -| | | | | | | | | | January 21, 2020. The details of attendance | | | The terms of reference of the Committee is | | -Human Resources, Nomination and -Prof. Dipak C. Jain NA NA NA NA NA NA NA of Committee members are given available on the website of the Company. -Remuneration Committee -| Dr. Raghunath A. Mashelkar | 11 | 4 | 4 | NA | 4 | NA | NA | | in this Report. | | | | | -| -------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------ | -------------------------------------- | --- | --- | ------------------------------------------- | --- | -| | | | | | | | | Terms of Reference of the Committee | | | | Meeting Details | | -| Adil Zainulbhai | 12 | 4 | NA | NA | NA | NA | 4 | | | | | | | -| | | | | | | | | inter alia include the following: | Investor Grievance Redressal | | | Four meetings of the Committee were | | -| Raminder Singh Gujral | 12 | 4 | NA | NA | NA | NA | NA | | | | | | | -| | | | | | | | | • Formulate the criteria for determining | The number of complaints received and | | | held during the year, as against statutory | | -Dr. Shumeet Banerji NA 3 3 NA NA NA 3 qualifications, positive attributes resolved to the satisfaction of investors requirement of one meeting. The meetings -Arundhati Bhattacharya NA NA NA 4 4 NA NA and independence of a Director, and during the year (out of the investor base of were held on April 17, 2019; July 18, 2019; -| | | | | | | | | recommend to the Board a policy, | 26 lakh) and their break-up is as under: | | | October 17, 2019 and January 16, 2020. | | -| --------------- | --- | --- | --- | --- | --- | --- | --- | --------------------------------- | ----------------------------------------- | --- | --- | --------------------------------------- | --- | -| K. V. Chowdary* | 5 | 1 | NA | 2 | NA | NA | 1 | | | | | | | -Nita M. Ambani NA NA NA NA NA NA NA relating to the remuneration of the The details of attendance of Committee -No. of -| | | | | | | | | Directors, Key Managerial Personnel and | Type of Complaints | | | members are given in this Report. | | -| ----------------- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | ------------------ | --- | ---------- | --------------------------------- | --- | -| Nikhil R. Meswani | NA | NA | 3 | 4 | NA | 7 | NA | | | | Complaints | | | -other employees. -Hital R. Meswani NA NA NA 4 4 7 3 Non-Receipt of Annual Reports 115 Health, Safety and Environment -• Formulate the criteria for evaluation -| | | | | | | | | | Non-Receipt of Dividend | | 110 | Committee | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | --------- | --- | -P. M. S. Prasad NA NA NA NA 3 NA 4 of performance of the Independent -The terms of reference of the Committee is -| | | | | | | | | | Non-Receipt of Interest / | | | 3 | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------------------- | --- | --- | --- | --- | -Pawan Kumar Kapil NA NA NA NA 4 NA NA Directors and the Board of Directors. -| | | | | | | | | | Redemption payments | | | available on the website of the Company. | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | ---------------------------------------- | --- | -• Devise a policy on Board Diversity. -NA – Not a member of the Committee Transfer of securities 383 -| | | | | | | | | • Identify persons who are qualified to | | | | Meeting Details | | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --------------- | --- | -* Appointed as a member of committees w.e.f. October 18, 2019 (5 meetings of Audit Committee, 1 meeting of HRNR Committee, 2 meetings of SR Committee Total 611 -| | | | | | | | | become Directors and who may be | | | | Four meetings of the Committee were | | -| --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | ------------------------------------ | --- | -and 1 meeting of RM Committee were held since his appointment). -| | | | | | | | | appointed in senior management in | | | | held during the year. The meetings were | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ---------------------------------------- | --- | -As on March 31, 2020, no complaints -Procedure at Committee Meetings -• Review with the management, the General accordance with the criteria laid down held on April 25, 2019; August 14, 2019; -were outstanding. -The Company’s guidelines relating to quarterly financial statements before Members of the Audit Committee possess and to recommend to the Board their October 25, 2019 and January 21, 2020. -the Board meetings are applicable to the submission to the Board for approval. requisite qualifications. The representatives appointment and / or removal. The response time for attending to The details of attendance of Committee -Committee meetings. Each Committee has • Review with the management, the of Statutory Auditors are permanent invitees • Recommend to the Board, all investors’ correspondence during financial members are given in this Report. -the authority to engage outside experts, statement of uses / application of funds. to the Audit Committee meetings held remuneration, in whatever form, payable year 2019-20 is as under: -Finance Committee -advisors and counsels to the extent it • Review and monitor the auditor’s quarterly, to approve financial statements. to senior management. -| | | | | | | | | | Particulars | No. | % | The terms of reference of the Committee is | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --- | --- | ------------------------------------------- | --- | -considers appropriate to assist in its independence, performance and The representatives of Statutory Auditors, • Review Human Resource policies -| | | | | | | | | | Total number of | 2,91,909 | 100 | available on the website of the Company. | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | -------- | --- | ---------------------------------------- | --- | -functioning. Minutes of the proceedings of effectiveness of audit process. Executives from Accounts department, and overall human resources -correspondence -Committee meetings are circulated to the • Review the findings of any internal Finance department, Corporate Secretarial of the Company. Meeting Details -received during -respective committee members and placed investigations by the internal auditors department and Internal Audit department Seven meetings of the Committee were -| | | | | | | | | The detailed terms of reference of the | the financial year | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | ------------------- | --- | --- | --- | --- | -before the Board meetings for noting. The into matters where there is suspected attend the Audit Committee meetings. 2019-20 held during the year. The meetings were -Committee is available on the website -composition and terms of reference of all fraud or irregularity or a failure of internal held on April 18, 2019; July 19, 2019; -| | | | | | During the year, all the recommendations | | | of the Company. | Replied within 1 to | 2,91,866 | 99.99 | | | -| --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --------------- | -------------------- | -------- | ----- | --- | --- | -the Committees are in compliance with control systems of a material nature and December 9, 2019; December 16, 2019, -| | | | | | made by the Committee were accepted by | | | | 4 days of receipt | | | | | -| --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | ----------------- | --- | --- | --- | --- | -the Companies Act, 2013 and the Listing reporting the matter to the Board. The Chairman of the Committee was February 3, 2020, February 18, 2020 and -| | | | | | the Board. The Lead Cost Auditor attend | | | | Replied after 4 | 43 | 0.01 | | | -| --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | ---------------- | --- | ---- | --- | --- | -Regulations, as applicable. The composition • Review the functioning of the present at the last Annual General Meeting March 31, 2020. The details of attendance -| | | | | | the Audit Committee meeting where cost | | | | days of receipt | | | | | -| --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --------------- | --- | --- | --- | --- | -of all the Committees is given in this Report. Whistle-blower mechanism / oversee held on August 12, 2019. of Committee members are given -audit report is discussed. -| | | the vigil mechanism. | | | | | | | | | | in this Report. | | -| --------------------- | --- | ------------------------------------------- | --- | --- | --------------------------------------------- | --- | --- | --------------- | ------------------ | --- | --- | --------------- | --- | -| Details of Committees | | | | | | | | Meeting Details | Compliance Officer | | | | | -| | • | Review financial statements, in particular | | | The Internal Auditor reports directly to the | | | | | | | | | -Audit Committee Four meetings of the Committee were K. Sethuraman, Group Company Secretary -| | | the investments made by the Company’s | | | Audit Committee. | | | | | | | | | -| --- | --- | -------------------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Terms of Reference of the Committee held during the year, as against statutory and Chief Compliance Officer and Savithri -unlisted subsidiaries. -| | | | | | The Chairman of the Committee was | | | requirement of one meeting. The meetings | Parekh, Joint Company Secretary and | | | | | -| --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | ----------------------------------------- | ------------------------------------ | --- | --- | --- | --- | -inter alia include the following: -• Recommend appointment, remuneration The detailed terms of reference of the present at the last Annual General Meeting were held on April 17, 2019; July 18, 2019; Compliance Officer, are the Compliance -and terms of appointment of auditors. Committee is available on the website held on August 12, 2019. October 15, 2019 and January 16, 2020. Officers for complying with requirements of -• Approval of payment to statutory of the Company. The details of attendance of Committee Securities Laws. -auditors, including cost auditors, for any members are given in this Report. -other services rendered by them. -192 193 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | Integrated Annual Report 2019-20 | | | | | Overview Review | Statements | -| --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------------- | ---------- | -PERFORMANCE EVALUATION compliance with code of conduct, vision DIRECTORS’ REMUNERATION During the year, there were no other Banerji were appointed as Independent Audit Reports of these subsidiaries -CRITERIA FOR DIRECTORS and strategy, benchmarks established by REMUNERATION POLICY pecuniary relationships or transactions of Directors on the Board of RJIL and they are are available on the website of the -The Human Resources, Nomination and global peers etc., which is in compliance The Company’s Remuneration Policy for Non-Executive Directors with the Company. continuing as such. Company. The Secretarial Audit report -Remuneration Committee has devised a with applicable laws, regulations Directors, Key Managerial Personnel and The Company has not granted any stock of these subsidiaries does not contain -Keeping in view good Corporate -criteria for evaluation of the performance and guidelines. other employees is available on the website options to its Non-Executive Directors. any qualification, reservation or adverse -Governance Prof. Dipak C. Jain and Shri -of the Directors including the Independent of the Company. remark or disclaimer. -Adil Zainulbhai are also on the Board -Directors. The said criteria provides certain FRAMEWORK FOR MONITORING -The Company’s remuneration policy is SUBSIDIARY COMPANIES of Reliance Retail Ventures Limited, an The Company monitors performance of -parameters like attendance, acquaintance -directed towards rewarding performance unlisted subsidiary of the Company. For subsidiary companies, inter alia, by the -with business, communication inter se During the year, Reliance Jio Infocomm -based on review of achievements better administration and governance, following means: -between board members, effective Limited (RJIL), Reliance Retail Limited (RRL) -| | | periodically. The remuneration policy is in | | | | | key subsidiary companies have voluntarily | | | -| --- | --- | -------------------------------------------- | --- | --- | --- | --- | ------------------------------------------ | --- | --- | -participation, domain knowledge, and Reliance Industrial Investments and • Financial statements, in particular -| | | consonance with existing industry practice. | | | | | appointed Independent Directors on their | | | -| --- | --- | ------------------------------------------- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | -Holdings Limited (RIIHL) were material investments made by subsidiary -respective Boards. The composition and -subsidiaries of the Company, as per Listing companies, are reviewed quarterly by -REMUNERATION OF THE MANAGING DIRECTOR AND WHOLE-TIME DIRECTORS FOR THE FINANCIAL effectiveness of Boards of subsidiaries is -Regulations. Post closure of the financial the Company’s Audit Committee. -| YEAR 2019-20 | | | | | | | reviewed by the Company periodically. | | | -| ------------ | --- | --- | --- | --- | --- | --- | -------------------------------------- | --- | --- | -year, Jio Platforms Limited and Reliance • Minutes of Board meetings of subsidiary -| | | | | (` in crore) | | | Governance framework is also ensured | | | -| --- | --- | --- | --- | -------------- | --- | --- | ------------------------------------- | --- | --- | -Global Energy Services (Singapore) Pte. companies are placed before the -through appointment of Managerial -Salary and Retiral Commission Stock Limited have become material subsidiaries Company’s Board regularly. -| Name of the Director | Perquisites | | Total | | | | | | | -| -------------------- | ----------- | --- | ----- | --- | --- | --- | --- | --- | --- | -allowances benefits payable Options Personnel and Secretarial Auditor. A -and RIIHL has ceased to be a material • A statement containing all significant -Mukesh D. Ambani 4.36 0.40 0.71 9.53 15.00 - robust compliance management system -subsidiary of the Company. transactions and arrangements entered -Nikhil R. Meswani 6.32 0.02 0.37 17.28 24.00 - covering all the subsidiaries is also in -into by subsidiary companies is placed -Hital R. Meswani 6.31 0.04 0.37 17.28 24.00 - In terms of the provisions of Regulation 24(1) place. Guidance is provided to subsidiaries -before the Company’s Board. -| | | | | | of the Listing Regulations, appointment | | on matters relating to conduct of Board | | | -| --- | --- | --- | --- | --- | ---------------------------------------- | --- | ---------------------------------------- | --- | --- | -P. M. S. Prasad 10.81* 0.00 0.34 - 11.15 - • Presentations are made to the -Pawan Kumar Kapil 3.60* 0.29 0.15 - 4.04 - of one of the Independent Directors of meeting, training and familiarisation -Company’s Board on business -*includes performance linked incentives for the FY 2018-19 paid in FY 2019-20. the Company on the Board of material programmes for the Independent Directors -performance by the senior management -| | | | | | subsidiaries was applicable only to | | on the Board of subsidiaries. | | | -| --- | --- | --- | --- | --- | ------------------------------------ | --- | ----------------------------- | --- | --- | -In light of the COVID-19 outbreak in India, The Chairman and Managing Director as recommended by Human Resources, on major subsidiaries of the Company. -RJIL and RRL. Prior to RRL and RJIL -which has exacted a huge toll on the had his salary capped at ` 15 crore since Nomination and Remuneration Committee. The Company is in compliance with -becoming material unlisted subsidiaries -societal, economic and industrial health 2008-09 in order to set a personal Regulation 24A of the Listing Regulations. The Company’s Policy for determining -| | | The tenure of office of the Managing | | | of the Company, Prof. Dipak C. Jain was | | | | | -| --- | --- | ------------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | -of the nation, Shri Mukesh D. Ambani, the example of moderation in managerial The Company’s unlisted material Material Subsidiaries is available on the -| | | Director and Whole-time Directors is for | | | appointed as an Independent Director | | | | | -| --- | --- | ----------------------------------------- | --- | --- | ------------------------------------- | --- | --- | --- | --- | -Chairman and Managing Director, has compensation levels. And now, he is subsidiaries RJIL, RRL and RIIHL undergo website of the Company. -| | | 5 (five) years from their respective date of | | | on the Board of RRL and Prof. Dipak C. | | | | | -| --- | --- | --------------------------------------------- | --- | --- | --------------------------------------- | --- | --- | --- | --- | -voluntarily decided to forego his salary. forgoing his salary until the company and Secretarial Audit. Copy of Secretarial -| | | appointments and can be terminated by | | | Jain, Shri Adil Zainulbhai and Dr. Shumeet | | | | | -| ------------------------------------------ | ------------------------------------------- | -------------------------------------- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | -| The Board of Directors noted his decision | all its businesses are fully back to their | | | | | | | | | -either party by giving three months’ notice -| to forego his salary until the impact of | earnings potential. On a similar note, other | | | | | | | | | -| ----------------------------------------- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -GENERAL BODY MEETINGS -in writing. There is no separate provision for -| COVID-19 abates. | Executive Directors have also expressed | | | | | | | | | -| ---------------- | ---------------------------------------- | -------------------------- | --- | --- | ----------------------- | --- | --- | --- | --- | -| | | payment of severance fees. | | | ANNUAL GENERAL MEETINGS | | | | | -their decision to draw remuneration up to -During the preceding three years, the Company’s Annual General Meetings were held at Birla Matushri Sabhagar, 19, Sir Vithaldas -50% of their remuneration entitlement and -Thackersey Marg, Near Bombay Hospital & Medical Research Centre, New Marine Lines, Mumbai - 400 020. -REMUNERATION OF THE NON-EXECUTIVE DIRECTORS FOR THE FINANCIAL YEAR 2019-20 The date and time of the Annual General Meetings held during last three years and the special resolution(s) passed thereat, are as follows: -(` in crore) -| | | | | | Year Date | Time | Special Resolution(s) Passed | | | -| -------------------- | --- | ----------- | ---------- | ----- | ----------------------- | ---------- | --------------------------------------------------------------- | --- | --- | -| Name of the Director | | Sitting Fee | Commission | Total | | | | | | -| | | | | | 2018-19 August 12, 2019 | 11:00 a.m. | i. R e-appoint Shri P. M. S. Prasad as a Whole-time Director | | | -| Mansingh L. Bhakta* | | 0.02 | 0.42 | 0.44 | | | | | | - ii. R e-appoint Shri Raminder Singh Gujral as an Independent Director -| Yogendra P. Trivedi | | 0.33 | 1.15 | 1.48 | | | | | | -| ------------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | -Prof. Dipak C. Jain 0.09 1.15 1.24 2017-18 July 5, 2018 11:00 a.m. i. R e-appoint Shri Adil Zainulbhai as an Independent Director -| Dr. Raghunath A. Mashelkar | | 0.31 | 1.15 | 1.46 | | | | | | -| -------------------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | - ii. O ffer or invitation to subscribe to Redeemable Non-Convertible Debentures on private -| Adil Zainulbhai | | 0.29 | 1.15 | 1.44 | | | | | | -| --------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | -placement -Raminder Singh Gujral 0.25 1.15 1.40 2016-17 July 21, 2017 11:00 a.m. i. R e-appoint Shri Pawan Kumar Kapil as a Whole-time Director -| Dr. Shumeet Banerji | | 0.17 | 1.15 | 1.32 | | | | | | -| ------------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | - ii. R e-appoint Shri Yogendra P. Trivedi as an Independent Director -| Arundhati Bhattacharya | | 0.17 | 1.15 | 1.32 | | | | | | -| ---------------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | - iii. R e-appoint Prof. Ashok Misra as an Independent Director -| K. V. Chowdary** | | 0.14 | 0.52 | 0.66 | | | | | | -| ---------------- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | -Nita M. Ambani 0.07 1.15 1.22 iv. R e-appoint Shri Mansingh L. Bhakta as an Independent Director -Total v. R e-appoint Prof. Dipak C. Jain as an Independent Director -| | | 1.84 | 10.14 | 11.98 | | | | | | -| --- | --- | ---- | ----- | ----- | --- | --- | --- | --- | --- | -*Ceased to be a director w.e.f. August 12, 2019. vi. R e-appoint Dr. Raghunath A. Mashelkar as an Independent Director -**Appointed as a director w.e.f. October 18, 2019. vii. A lteration of the Articles of Association of the Company - viii. O ffer or invitation to subscribe to Redeemable Non-Convertible Debentures on private -placement -194 195 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | Integrated Annual Report 2019-20 | | | Overview Review | Statements | -| --- | --- | --- | -------------------------------- | --- | --- | --------------- | ---------- | -RESOLUTION(S) PASSED THROUGH DETAILS OF NON-COMPLIANCE until the disposal of the appeal. Employees may also report violations to the Presentations to institutional investors SEBI Complaints Redress System -POSTAL BALLOT: BY THE COMPANY, PENALTIES, The prohibition from dealing in equity Chairman of the Audit Committee and there / analysts: Detailed presentations are (SCORES): Investor complaints are -STRICTURES IMPOSED ON THE -No postal ballot was conducted during derivatives in the Futures and Options was no instance of denial of access to the made to institutional investors and financial processed at SEBI in a centralised -the financial year 2019-20. There is no COMPANY BY STOCK EXCHANGE segment expired on March 23, 2018. Audit Committee. The Vigil Mechanism and analysts on the Company’s quarterly, half- web-based complaints redress system. -immediate proposal for passing any OR SEBI, OR ANY STATUTORY The appeal has been heard by SAT Whistle-blower Policy is available on the yearly as well as annual financial results. The salient features of this system are -resolution through postal ballot. AUTHORITY, ON ANY MATTER and is reserved for orders. website of the Company. These presentations are available on the centralised database of all complaints, -RELATED TO CAPITAL MARKETS, website of the Company, as well as sent online upload of Action Taken Reports -DISCLOSURE ON MATERIALLY DURING THE LAST THREE YEARS (b) S EBI had also issued a Show Cause PREVENTION OF SEXUAL -to the Stock Exchanges. No unpublished (ATRs) by concerned companies and online -Notice (SCN) dated November 21, -SIGNIFICANT RELATED PARTY (i) The Securities and Exchange Board HARASSMENT OF WOMEN AT price sensitive information is discussed in viewing by investors of actions taken on the -2017 to the Company in the matter -TRANSACTIONS THAT MAY HAVE of India (SEBI), on August 8, 2014 had WORKPLACE meeting / presentation with institutional complaints and their current status. -concerning trading in the shares of -POTENTIAL CONFLICT WITH passed an adjudication order on a show The Company is committed to provide investors and financial analysts. -Reliance Petroleum Limited by the Designated exclusive email-IDs: The -THE COMPANY’S INTERESTS AT cause notice issued to the Company for a work environment that ensures every -LARGE Company in the year 2007, asking Website: The Company’s website (www.ril. Company has designated the following -| | alleged non-disclosure of the diluted | | | employee is treated with dignity, respect | | | | -| --- | -------------------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | -the Company to show cause as com) contains a separate dedicated section email-IDs exclusively for investor servicing: -The Company’s major related party Earnings per Share in the quarterly and afforded equal treatment. Please refer -| | | to why inquiry should not be held | | | ‘Investor Relations’ where shareholders’ | | | -| --- | --- | ---------------------------------- | --- | --- | ----------------------------------------- | --- | --- | -transactions are generally with its financial results for the quarters ended Human Capital section of Management • For queries on Annual Report: -| | | against the Company in terms of SEBI | | | information is available. | | | -| --- | --- | ------------------------------------- | --- | --- | ------------------------- | --- | --- | -subsidiaries and associates. The related June 2007, September 2007, December Discussion and Analysis Report, investor.relations@ril.com -(Procedure for Holding Inquiry and -party transactions are entered into based 2007, March 2008, June 2008 and for more details. Annual Report: The Annual Report • For queries in respect of shares in -Imposing Penalties by Adjudicating -on considerations of various business September 2008 and imposed monetary containing, inter alia, Audited Financial physical mode: rilinvestor@kfintech.com -Officer) Rules, 1995 and penalty be -exigencies, such as synergy in operations, penalty of ` 13 crore. On an appeal by ADOPTION OF MANDATORY Statement, Audited Consolidated Financial -not imposed under the provisions -sectoral specialisation and the Company’s the Company, the Hon’ble Securities AND DISCRETIONARY Statement, Board’s Report, Auditors’ Shareholders’ Feedback Survey: The -of the SEBI Act, 1992. The Company -long-term strategy for sectoral investments, Appellate Tribunal (SAT), set aside REQUIREMENTS Report and other important information Company sends feedback form seeking -has made written submissions in -optimisation of market share, profitability, SEBI’s order and remanded the matter The Company has complied with all is circulated to the members and others shareholders’ views on various matters -response to the SCN, including on -legal requirements, liquidity and capital for fresh consideration by SEBI. SEBI mandatory requirements of Regulation 34 entitled thereto. The Management relating to investor services and Annual -June 15, 2018, September 12, 2018 -resources of subsidiaries and associates. issued a fresh show cause notice dated of the Listing Regulations. The Company Discussion and Analysis Report forms part Report for improvement in future. -and September 20, 2019. Hearings -April 5, 2016 in the matter alleging has adopted the following discretionary of the Annual Report. The Annual Report is -| All the contracts / arrangements / | | were held before the Adjudicating | | | | | | -| ----------------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | --- | -incorrect disclosure of the diluted requirements of the Listing Regulations: also available in downloadable form on the GENERAL SHAREHOLDER -| transactions entered by the Company | | Officer on September 11, 2018, | | | | | | -| ------------------------------------ | --- | ------------------------------- | --- | --- | --- | --- | --- | -Earnings per Share. The Company website of the Company. INFORMATION -COMMUNICATION TO -| during the financial year with related parties | | January 24, 2020 and March 12, 2020 | | | | | | -| ----------------------------------------------- | --- | ------------------------------------ | --- | --- | --- | --- | --- | -filed a reply to the show cause notice ANNUAL GENERAL MEETING -were in its ordinary course of business and and the matter is part-heard. SHAREHOLDERS Chairman’s Communiqué: Printed copy of -and attended the personal hearing Wednesday, July 15, 2020 at 2:00 p.m. -on an arm’s length basis. Half-yearly reports covering financial the Chairman’s speech is distributed to the -on July 26, 2016. SEBI appointed new (iii) S EBI had issued a show cause through Video Conferencing / Other Audio -| | | | | results were sent to the members at their | shareholders at the Annual General Meeting. | | | -| --- | --- | --- | --- | ------------------------------------------ | -------------------------------------------- | --- | --- | -During the year, the Company had not Adjudicating Officer (AO). The last notice dated November 26, 2015 Visual Means as set out in the Notice -| | | | | registered addresses. In addition to half- | A copy of the Chairman’s speech is also sent | | | -| --- | --- | --- | --- | ------------------------------------------ | --------------------------------------------- | --- | --- | -entered into any contract / arrangement / hearing before the AO was held on to the Company alleging that, the convening the Annual General Meeting. -| | | | | yearly reports, quarterly reports were | to all the shareholders, whose e-mail IDs are | | | -| --------------------------------------------- | ----------------------------------- | --------------------------------------- | --- | --------------------------------------- | ---------------------------------------------- | -------------------------------------------- | --- | -| transaction with related parties which could | November 22, 2018. Further details | Company had not provided the | | | | | | -| | | | | also sent to the members, whose e-mail | registered with the Company / Depository | DIVIDEND PAYMENT DATE | | -| be considered material in accordance with | sought by AO have been provided. | information sought by SEBI regarding | | | | | | -| | | | | IDs are registered with the Company / | Participants. The document is also available | Between July 16, 2020 and July 22, 2020 | | -| the policy of the Company on Materiality of | Adjudication order is awaited. | categorization of the Directors of the | | | | | | -| | | | | Depository Participants. | on the website of the Company. | for electronic transfer to the shareholders | | -| Related Party Transactions. The Company | | Company as on January 07, 2000. | | | | | | -(ii) (a) S EBI had passed an Order under who have furnished bank account details to -has made full disclosure of transactions with The Adjudicating Officer, vide Order AUDIT QUALIFICATION Letters to Investors: Letters were sent to -Section 11B of the SEBI Act, 1992 the Company / its Registrar. -the related parties as set out in Note 32 of dated February 28, 2018, disposed of The Company is in the regime of unmodified the shareholders / debenture holders as -on March 24, 2017 on a show cause -Standalone Financial Statements, forming the adjudication proceedings initiated opinions on financial statements. per records, for claiming unclaimed / unpaid Physical warrants shall be dispatched to -notice dated December 16, 2010 -part of the Annual Report. against the Company without imposition dividend / interest or redemption amount on the shareholders, who have not registered -| | issued to the Company in the matter | | | REPORTING OF INTERNAL AUDITOR | | | | -| --- | ------------------------------------ | --- | --- | ----------------------------- | --- | --- | --- | -of any penalty. debentures / dematerialisation of shares / their ECS mandates, upon normalisation of -There were no materially significant concerning trading in the shares of The Internal Auditor directly reports to the -updating PAN and bank account details. postal services. -related party transactions which could Reliance Petroleum Limited by the WHISTLE-BLOWER POLICY Audit Committee. -The Company has also sent intimations -have potential conflict with interest of the Company in the year 2007, directing FINANCIAL YEAR -| | | The Company promotes safe, ethical | | | to the shareholders holding shares in | | | -| --- | --- | ----------------------------------- | --- | --- | -------------------------------------- | --- | --- | -Company at large. (i) disgorgement of ` 447 crore along MEANS OF COMMUNICATION April 1 to March 31 -| | | and compliant conduct of all its business | | | physical form, informing them about SEBI’s | | | -| --- | ------------------------------------ | ------------------------------------------ | --- | --------------------------------- | ------------------------------------------- | --- | --- | -| | with interest calculated at 12% per | | | Quarterly results: The Company’s | | | | -The Company’s Policy on Materiality of activities and has put in place a mechanism mandate to permit transfer of shares only in FINANCIAL CALENDAR -| | annum from November 29, 2007 till | | | quarterly / half-yearly / annual financial | | | | -| --- | ---------------------------------- | --- | --- | ------------------------------------------- | --- | --- | --- | -Related Party Transactions and on dealing for reporting illegal or unethical behaviour. dematerialised form w.e.f. April 1, 2019. (TENTATIVE) RESULTS FOR THE -| | date of payment; and (ii) prohibiting | | | results are sent to the Stock Exchanges | | | | -| --- | -------------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | -with Related Party Transactions is available The Company has a Vigil Mechanism QUARTER ENDING -the Company from dealing in equity and published in ‘Indian Express’, ‘Financial NSE Electronic Application Processing -on the website of the Company. and Whistle-blower policy under which June 30, 2020 – Third week of July, 2020 -derivatives in the Futures and Options Express’ and ‘Loksatta’. They are also System (NEAPS): NEAPS is a web-based -the employees are encouraged to report -segment of the stock exchanges, available on the website of the Company. application designed by NSE for corporates. September 30, 2020 – Third week -violations of applicable laws and regulations -directly or indirectly for a period of All periodical and other compliance filings of October, 2020 -| | | and the Code of Conduct – without fear of | | News releases, presentations: Official | | | | -| --- | --- | ------------------------------------------ | --- | --------------------------------------- | --- | --- | --- | -one year from March 24, 2017. The are filed electronically on NEAPS. -any retaliation. The reportable matters may news releases and official media releases December 31, 2020 – Fourth week -Company filed an appeal against -be disclosed to the Ethics and Compliance are sent to Stock Exchanges and are also BSE Listing Centre (Listing Centre): BSE’s of January, 2021 -the said Order before SAT. SAT has -| | | Task Force which operates under the | | available on the website of the Company. | Listing Centre is a web-based application | | | -| --- | --- | ------------------------------------ | --- | ---------------------------------------- | ------------------------------------------ | --- | --- | -stayed the direction on disgorgement March 31, 2021 – Fourth week of April, 2021 -| | | supervision of the Audit Committee. | | | designed for corporates. All periodical | | | -| --- | --- | ------------------------------------ | --- | --- | ---------------------------------------- | --- | --- | -and other compliance filings are filed Annual General Meeting – June / July, 2021 -electronically on the Listing Centre. -196 197 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | | | | Integrated Annual Report 2019-20 | | | | | Overview | Review | Statements | | -| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | -------- | ------ | ---------- | --- | -BSE SENSEX VS RIL SHARE PRICE -| LISTING ON STOCK EXCHANGES | | Domestic Custodian | | | CREDIT RATING | | | | | | | | | | | -| -------------------------- | --- | ------------------ | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Equity Shares ICICI Bank Limited The Company’s financial discipline and BSE SENSEX RIL Close Price -BSE Limited (BSE) Empire Complex, 1st Floor, 414, Senapati prudence is reflected in the strong credit 1,600 -42,000 -Phiroze Jeejeebhoy Towers, Dalal Street, Bapat Marg, Lower Parel (West), ratings ascribed by rating agencies as 1,500 -40,000 -Mumbai - 400 001 Mumbai - 400 013 given below. There has been no revision 1,400 -| | | | | | in credit ratings during the financial year | | | | | 38,000 | | | | 1,300 | | -| ------------------- | --- | ----------------------- | --- | --- | -------------------------------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | ----- | --- | -| Scrip Code – 500325 | | PAYMENT OF LISTING FEES | | | | | | | | | | | | | | -2019-20. The details of the Credit Rating are ESB 36,000 LIR -1,200 -| | | Annual listing fee for the financial year | | | | | | | | 34,000 | | | | | | -| --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -National Stock Exchange of India mentioned in Management Discussion and 1,100 -| | | 2020-21 has been paid by the Company | | | | | | | | 32,000 | | | | | | -| ------------- | --- | ------------------------------------- | --- | --- | ---------------- | --- | --- | --- | --- | ------ | --- | --- | --- | ----- | --- | -| Limited (NSE) | | | | | Analysis Report. | | | | | | | | | 1,000 | | -to BSE and NSE. Annual maintenance and -| Exchange Plaza, C-1, Block G, | | | | | | | | | | 30,000 | | | | 900 | | -| ------------------------------------- | --- | ----------------------------------------- | --- | --- | --------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -| | | listing agency fee for the calendar year | | | UTILISATION OF FUNDS | | | | | | | | | | | -| Bandra-Kurla Complex, Bandra (East), | | | | | | | | | | 28,000 | | | | 800 | | -2020 has been paid by the Company to RAISED THROUGH ISSUE OF 9102 RPA 9102 YAM 9102 NUJ 9102 LUJ 9102 GUA 9102 PES 9102 TCO 9102 VON 9102 CED 0202 NAJ 0202 BEF 0202 RAM -Mumbai - 400 051 -NON-CONVERTIBLE DEBENTURES -Luxembourg Stock Exchange. -| Trading Symbol – RELIANCE | | | | | During the financial year 2019-20 no | | | | | | | | | | | -| ------------------------- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -PAYMENT OF DEPOSITORY FEES -funds were raised through issue of -| ISIN: INE002A01018 | | Annual Custody / Issuer fee is being paid by | | | | | | | | | | | | | | -| ------------------ | --- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Non-Convertible Debentures. -| | | the Company within the due date based on | | | | | | | NSE NIFTY VS RIL SHARE PRICE | | | | | | | -| --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | -Global Depository Receipts (GDRs) -| | | invoices received from the Depositories. | | | DEBENTURE TRUSTEE | | | | | | | | | | | -| --- | --- | ---------------------------------------- | --- | --- | ----------------- | --- | --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | NSE NIFTY RIL Close Price | | | | | | | -Luxembourg Stock Exchange -Axis Trustee Services Limited -35A Boulevard Joseph II, FEES PAID TO THE STATUTORY 14,000 1,600 -The Ruby, 2nd Floor, SW, -| L-1840, Luxembourg | | AUDITORS | | | | | | | | | | | | | | -| ------------------ | --- | -------- | --- | --- | ------------------------ | --- | --- | --- | --- | ------ | --- | --- | --- | ----- | --- | -| | | | | | 29, Senapati Bapat Marg, | | | | | 13,000 | | | | 1,500 | | -Total fees for all services paid by the -Overseas Depository Dadar (West), Mumbai – 400 028 12,000 1,400 -Company and its subsidiaries, on a -| The Bank of New York Mellon | | | | | Tel: +91-22-62300451 | | | | | ESN | | | | LIR | | -| ---------------------------- | --- | --- | --- | --- | -------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | ----- | --- | -| | | | | | | | | | | 11,000 | | | | 1,300 | | -consolidated basis, to statutory auditors of -| Corporation | | | | | Fax: +91-22-62300700 | | | | | | | | | | | -| --------------------------------- | --- | --------------------------------------------- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | ------ | --- | --- | --- | ----- | --- | -| | | the Company and other firms in the network | | | | | | | | 10,000 | | | | 1,200 | | -| 240, Greenwich Street, New York, | | | | | E-mail: debenturetrustee@axistrustee.com; | | | | | | | | | | | -| | | entity of which the statutory auditors are a | | | | | | | | 9,000 | | | | 1,100 | | -| NY 10286, USA | | | | | complaints@axistrustee.com | | | | | | | | | | | -part, during the year ended March 31, 2020, -| | | | | | Website Address: www.axistrustee.com | | | | | 8,000 | | | | 1,000 | | -| --- | --- | --- | --- | --- | ------------------------------------ | --- | --- | --- | --- | ----- | --- | --- | --- | ----- | --- | -is ` 46.27 crore. 9102 RPA 9102 YAM 9102 NUJ 9102 LUJ 9102 GUA 9102 PES 9102 TCO 9102 VON 9102 CED 0202 NAJ 0202 BEF 0202 RAM -STOCK MARKET PRICE DATA -SHARE PRICE PERFORMANCE IN COMPARISON TO BROAD BASED INDICES – BSE SENSEX AND NSE NIFTY AS -| Month | National Stock Exchange of India Limited (NSE) | | | | | BSE Limited (BSE) | | | | | | | | | | -| ----- | ---------------------------------------------- | --- | --- | --- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -ON MARCH 31, 2020 -| | High Price (`) | | Low Price (`) | Volume (No.) | High Price (`) | Low Price (`) | | Volume (No.) | | | | | | | | -| ---------- | -------------- | -------- | ------------- | ------------ | -------------- | ------------- | -------- | ------------ | --- | --- | ----------- | -------- | ----------- | --- | ------- | -| April 2019 | | 1,412.40 | 1,321.00 | 16,69,46,655 | 1,410.90 | | 1,321.60 | 1,15,01,774 | | | | | | | | -| | | | | | | | | | | | RIL Share | Sensex | RIL Share | | NIFTY | -May 2019 1,417.50 1,227.50 24,52,03,012 1,417.00 1,227.00 1,97,31,034 Performance on BSE Performance Performance on NSE Performance -June 2019 1,374.25 1,248.65 13,32,79,066 1,373.00 1,248.05 70,48,484 FY2019-20 -18.39% -23.80% -18.30% -26.03% -| July 2019 | 1,300.00 | | 1,162.40 | 15,79,81,147 | 1,299.80 | | 1,163.00 | 85,47,940 | | | | | | | | -| -------------- | -------- | -------- | -------- | ------------ | -------- | --- | -------- | ----------- | -------- | --- | ------- | ------- | ------- | ------- | ------- | -| | | | | | | | | | 2 Years | | 26.01% | -10.62% | | 26.18% | -14.99% | -| August 2019 | | 1,304.45 | 1,095.30 | 23,91,04,397 | 1,304.00 | | 1,095.65 | 1,10,67,323 | | | | | | | | -| | | | | | | | | | 3 Years | | 68.66% | -0.51% | | 68.64% | -6.28% | -| September 2019 | | 1,335.75 | 1,172.65 | 18,15,20,211 | 1,336.00 | | 1,173.25 | 89,95,149 | | | | | | | | -| | | | | | | | | | 5 Years | | 169.78% | 5.40% | 169.67% | | 1.26% | -| October 2019 | | 1,489.65 | 1,281.30 | 16,42,71,981 | 1,489.50 | | 1,281.20 | 75,03,066 | | | | | | | | -| | | | | | | | | | 10 Years | | 107.03% | 68.12% | | 107.35% | 63.79% | -| November 2019 | | 1,584.15 | 1,422.55 | 16,70,40,786 | 1,584.00 | | 1,423.35 | 80,62,936 | | | | | | | | -December 2019 1,617.55 1,508.05 17,14,67,991 1,617.80 1,509.10 1,01,31,329 RIL’s share price on BSE and NSE has been adjusted for the year 2017 and earlier years, on account of issue of bonus shares in -January 2020 1,609.00 1,407.20 19,98,39,592 1,609.50 1,407.45 83,79,170 the FY 2017-18. -| February 2020 | 1,508.00 | | 1325.00 | 18,37,80,216 | 1,507.95 | | 1,325.00 | 98,91,342 | | | | | | | | -| ------------- | -------- | --- | ------- | ------------ | -------- | --- | -------- | ------------ | ------------------------ | --- | --------------------- | --- | ------------------------------------------- | --- | --- | -| March 2020 | 1,369.00 | | 875.65 | 45,49,78,169 | 1,368.75 | | 875.70 | 23,45,49,195 | | | | | | | | -| | | | | | | | | | REGISTRARS AND TRANSFER | | SHARE TRANSFER SYSTEM | | During the year, the Company had obtained, | | | -[Source: This information is compiled from the data available on the websites of BSE and NSE] AGENTS SEBI has mandated that, effective April 1, on half-yearly basis, a certificate, from a -| | | | | | | | | | KFin Technologies Private Limited | | 2019, no share can be transferred in | | Company Secretary in Practice, certifying | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------- | --- | ------------------------------------------- | --- | ---------------------------------------------- | --- | --- | -| | | | | | | | | | (Formerly known as Karvy Fintech | | physical mode. Hence, the Company has | | that all certificates have been issued within | | | -| | | | | | | | | | Private Limited) | | stopped accepting any fresh lodgement | | thirty days of the date of lodgement of the | | | -| | | | | | | | | | Selenium Tower B, Plot 31-32, | | of transfer of shares in physical form. | | transfer (for cases lodged prior to April | | | -| | | | | | | | | | Gachibowli, Financial District, | | The Company had sent communication | | 1, 2019), sub-division, consolidation and | | | -| | | | | | | | | | Nanakramguda, Hyderabad - 500 032 | | to the shareholders encouraging them | | renewal as required under Regulation | | | -| | | | | | | | | | Tel: +91 40 67161700 | | to dematerialise their holding in the | | 40(9) of the Listing Regulations and filed | | | -| | | | | | | | | | Toll Free No.: 1800 425 8998 | | Company. The communication, inter alia, | | a copy of the said certificate with the | | | -| | | | | | | | | | (From 9:00 a.m. to 6:00 p.m.) | | contained procedure for getting the shares | | Stock Exchanges. | | | -| | | | | | | | | | Fax: +91 40 67161680 | | dematerialised. Shareholders holding | | | | | -Trading in equity shares of the Company is -| | | | | | | | | | E-mail: rilinvestor@kfintech.com | | shares in physical form are advised to avail | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --------------------------------------------- | --- | --- | --- | --- | -permitted only in dematerialised form. -| | | | | | | | | | Website: www.kfintech.com | | the facility of dematerialisation. | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | ---------------------------------- | --- | --- | --- | --- | -| 198 | | | | | | | | | | | | | | | 199 | -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | | | | | | Integrated Annual Report 2019-20 | | | | | Overview | Review | Statements | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | -------- | ------ | ---------- | -SHAREHOLDING PATTERN AS ON MARCH 31, 2020 BUILD-UP OF EQUITY SHARE CAPITAL over-the-counter market in the United -| | | | | | | | | | | | The statement showing build-up of equity share capital is available on the website | | | | States of America. | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------------------------------------------------------------------------- | --- | --- | --- | ------------------ | --- | -Sr. -| Category of shareholder | | Number of shareholders | | | | Total number of shares | | | % of (A+B+C) | | | | | | | | -| ----------------------- | --- | ---------------------- | --- | --- | --- | ---------------------- | --- | --- | ------------ | --- | --------------- | --- | --- | --- | --- | --- | -| No. | | | | | | | | | | | of the Company. | | | | | | -RIL GDRs are exempted securities under -(A) Shareholding of Promoter and -US Securities Law. RIL GDR programme -| Promoter Group | | | | | | | | | | | CORPORATE BENEFITS TO INVESTORS | | | | | | -| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | --- | --- | -has been established under Rule 144A and -(1) Indian 52* 309,80,84,968 48.87 (A) DIVIDEND DECLARED FOR THE LAST 10 YEARS -Regulation S of the US Securities Act, 1933. -| (2) Foreign | | | | | 0 | | | 0 | | 0 | | | | | | | -| ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Dividend per -| | | | | | | | | | | | | Date of Dividend | | | Reporting is done under the exempted | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | ------------------------------------- | --- | -Total Shareholding of Promoter and 52 309,80,84,968 48.87 Financial Year Equity Share of -Promoter Group Declaration route of Rule 12g3-2(b) under the US -` 10/- each (`) -Securities Exchange Act, 1934. -| (B) Public Shareholding | | | | | | | | | | | 2009-10 | June 18, 2010 | | 7.00 | | | -| ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ---- | --- | --- | -(1) Institutions 1,819 235,51,81,559 37.15 (post bonus issue 1:1) The Bank of New York Mellon is an -(2) Non-institutions 26,30,300 73,46,36,787 11.59 2010-11 June 3, 2011 8.00 Overseas Depository and ICICI Bank -Total Public Shareholding 26,32,119 308,98,18,346 48.74 2011-12 June 7, 2012 8.50 -Limited is the Domestic Custodian of all the -(C) Shares held by Custodians and -| | | | | | | | | | | | 2012-13 | June 6, 2013 | | 9.00 | Equity Shares underlying the GDRs issued | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | --- | ---- | ----------------------------------------- | --- | -against which Depository Receipts -| | | | | | | | | | | | 2013-14 | June 18, 2014 | | 9.50 | by the Company. | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | --- | ---- | --------------- | --- | -have been issued -| | | | | | | | | | | | 2014-15 | June 12, 2015 | | 10.00 | | | -| ------------------------------- | --- | --- | --- | --- | --- | --- | ------------ | --- | --- | ---- | ------- | -------------- | --- | ----- | ---------------------- | --- | -| (1) Promoter and Promoter Group | | | | | 0 | | | 0 | | 0 | | | | | | | -| | | | | | | | | | | | 2015-16 | March 10, 2016 | | 10.50 | EMPLOYEE STOCK OPTIONS | | -| (2) Public | | | | | 1 | | 15,13,64,196 | | | 2.39 | | | | | | | -| | | | | | | | | | | | 2016-17 | July 21, 2017 | | 11.00 | | | -Total shares held by Custodians and 1 15,13,64,196 2.39 Particulars with regard to Employees’ -| | | | | | | | | | | | 2017-18 | July 5, 2018 | | 6.00 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | --- | ---- | --- | --- | -against which Depository Receipts Stock Options are available on the website -(post bonus issue 1:1) -| have been issued | | | | | | | | | | | | | | | of the Company. | | -| --------------------- | --- | --- | --- | --------- | --- | --- | ------------- | --- | --- | ------ | ------- | --------------- | --- | ---- | --------------- | --- | -| | | | | | | | | | | | 2018-19 | August 12, 2019 | | 6.50 | | | -| Total (A) + (B) + (C) | | | | 26,32,172 | | | 633,92,67,510 | | | 100.00 | | | | | | | -COMMODITY PRICE RISKS / -* As per disclosure under Regulation 30(2) of the SEBI (Substantial Acquisition of Shares and Takeovers) Regulations, 2011, furnished by the promoters. (B) BONUS ISSUES OF FULLY PAID-UP EQUITY SHARES -FOREIGN EXCHANGE RISK AND -CATEGORY-WISE SHAREHOLDING (%) Financial Year Ratio HEDGING ACTIVITIES -The Company is subject to commodity price -| Promoters Institutions | Non-Institutions | GDR Holders | | | | | | | | | | | | | | | -| -------------------------- | ------------------ | ------------ | --- | ---- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ---- | ------------------------------------------------- | --- | -| | | | | | | | | | | | 1980-81 | | | 3:5 | risks due to fluctuation in prices of crude oil, | | -| | | | | | | | | | | | 1983-84 | | | 6:10 | gas and downstream petroleum products. | | -| | | | | 2.39 | | | | | | | 1997-98 | | | 1:1 | | | -Company’s payables and receivables are in -| | | | 11.59 | | | | | | | | 2009-10 | | | 1:1 | | | -| --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | -U.S. Dollars and due to fluctuations in foreign -| | | | | | | | | | | | 2017-18 | | | 1:1 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | -exchange prices, it is subject to foreign -exchange risks. The Company has in place -LIQUIDITY -a robust risk management framework for -The Company’s Equity Shares are among the most liquid and actively traded shares on the -| | | | | | | 48.87 | | | | | | | | | identification and monitoring and mitigation | | -| --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | -Indian Stock Exchanges. RIL shares consistently rank among the top few frequently traded -of commodity price and foreign exchange -shares, both in terms of the number of shares traded, as well as value. -risks. The risks are tracked and monitored -37.15 -| | | | | | | | | | | | Relevant data for the average daily turnover for the financial year 2019-20 is given below: | | | | on a regular basis and mitigation strategies | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------------------------------------------------------------------------------------- | --- | --- | --- | --------------------------------------------- | --- | -are adopted in line with the risk management -| | | | | | | | | | | | Particulars | BSE | NSE | Total | framework. For further details on the above | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --- | --- | ----- | -------------------------------------------- | --- | -risks, please refer the Enterprise Risk -DISTRIBUTION OF SHAREHOLDING BY SIZE AS ON MARCH 31, 2020 Shares (Nos.) 13,98,416 99,81,430 113,79,846 -Management section of the Management -Value (` in crore) -| | | | | | | | | | | | | 157.43 | 1,314.76 | 1,472.19 | Discussion and Analysis Report. | | -| ----------------- | --- | --- | --- | --- | --- | ------- | --- | ------ | ----------------- | --- | --- | ------ | -------- | -------- | ------------------------------- | --- | -| Category (Shares) | | | | | | Holders | | Shares | % of total Shares | | | | | | | | -[Source: This information is compiled from the data available on the websites of BSE and NSE] -| Up to 500 | | | | | | 24,20,204 | | 19,66,29,029 | | 3.10 | | | | | RISK MANAGEMENT | | -| ------------ | --- | --- | --- | --- | --- | --------- | -------- | ------------ | --- | ---- | --------------------------- | ---------------------------------------------- | --- | --- | ----------------------- | --- | -| 501 - 1000 | | | | | | | 1,11,888 | 7,97,76,788 | | 1.26 | | | | | | | -| | | | | | | | | | | | OUTSTANDING GDRS / | surrender of GDRs can either be held by | | | POLICY WITH RESPECT TO | | -| 1001 - 5000 | | | | | | | 87,135 | 17,15,81,459 | | 2.71 | | | | | | | -| | | | | | | | | | | | WARRANTS AND CONVERTIBLE | investors concerned in their name or sold | | | COMMODITIES INCLUDING | | -| 5001 - 10000 | | | | | | | 7,516 | 5,18,42,226 | | 0.82 | | | | | | | -| | | | | | | | | | | | BONDS, CONVERSION DATE AND | off in the Indian secondary markets for cash. | | | THROUGH HEDGING | | -10001 - 20000 2,788 3,85,67,592 0.61 LIKELY IMPACT ON EQUITY To the extent of shares so sold in Indian • COMMODITIES EXPOSURE -| Above 20000 | | | | | | | 2,641 | 580,08,70,416 | | 91.50 | | | | | | | -| ----------- | --- | --- | --- | --- | --- | --------- | ----- | ------------- | --- | ------ | ------------------------------------------ | ---------------------------------------- | --- | --- | ----------------------------------- | --- | -| | | | | | | | | | | | GDRs: Outstanding GDRs as on March 31, | markets, GDRs can be reissued under the | | | The Company is exposed to price | | -| Total | | | | | | 26,32,172 | | 633,92,67,510 | | 100.00 | | | | | | | -| | | | | | | | | | | | 2020 represent 15,13,64,196 equity shares | available head-room. | | | volatility on various Petroleum, | | -| | | | | | | | | | | | constituting 2.39% of Company’s paid-up | | | | Petrochemical and other Energy | | -DEMATERIALISATION OF SHARES -| | | | | | | | | | | | Equity Share Capital. Each GDR represents | RIL GDR PROGRAMME | | | related commodities, as part of its | | -| --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ------------------------------------------ | -------------------------------------- | --- | --- | -------------------------------------- | --- | -| | | | | | | | | | | | two underlying equity shares in the | The Global Depository Receipts of the | | | business operations. Due to the | | -| Mode of Holding | | | | | | | | | | % | | | | | | | -| | | | | | | | | | | | Company. GDR is not a specific time-bound | Company are listed on Luxembourg | | | dynamic markets, prices of such | | -| NSDL | | | | | | | | | | 95.95 | | | | | | | -| | | | | | | | | | | | instrument and can be surrendered at any | Stock Exchange and are traded | | | Commodities fluctuate and can | | -| CDSL | | | | | | | | | | 2.97 | | | | | | | -| | | | | | | | | | | | time and converted into the underlying | on the International Order Book | | | result in Margin Risk. This policy | | -| Physical | | | | | | | | | | 1.08 | | | | | | | -| | | | | | | | | | | | equity shares in the Company. The shares | (London Stock Exchange) and amongst | | | prescribes the guidelines for hedging | | -Total 100.00 so released in favour of the investors upon qualified institutional investors on the Commodities Price risks. -200 201 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | | | | | Integrated Annual Report 2019-20 | | Overview | Review | Statements | | -| --- | --- | --- | --- | --- | --- | -------------------------------- | --- | -------- | ------ | ---------- | --- | -• HEDGING POLICY taken based on various factors including Silvassa Manufacturing Division ANY QUERY ON THE ANNUAL The cumulative amount transferred by the The Company has uploaded on its website, - Exposures are identified and market conditions, physical inventories, 342, Kharadpada, P. O. Naroli – 396 235, REPORT Company to IEPF up to March 31, 2020 the details of unpaid and unclaimed -measured across the Company so that macro-economic situation. These Union Territory of Dadra and Smt. Savithri Parekh is ` 242 crore. amounts lying with the Company as on -appropriate hedging can be done on a decisions and execution are done in line Nagar Haveli, India Joint Company Secretary and date of last Annual General Meeting -In accordance with the provisions of the -Net basis. For Commodities hedging, with the Board approved Commodities Compliance Officer (i.e. August 12, 2019). -Vadodara Manufacturing Division Companies Act, 2013, the Company has -there exist Over The Counter (OTC) Risk Management framework. The Risk Reliance Industries Limited -P. O. Petrochemicals, transferred 18,46,438 equity shares of Details of shares transferred to IEPF -and Exchange markets that offer Management Committee has oversight on 3rd Floor, Maker Chambers IV, -` 10/- each, to the credit of IEPF Authority, -Vadodara – 391 346, Gujarat, India Authority during financial year 2019-20 -financial instruments (derivatives), that all hedging actions taken. 222, Nariman Point, -on August 7, 2019, in respect of which are also available on the website -| enable managing the Price risk. | | | | | Vadodara Composites Division | | Mumbai - 400 021 | | | | | -| ------------------------------- | --- | --- | --- | --- | ---------------------------- | --- | ---------------- | --- | --- | --- | --- | -More details on Risk Management dividend had not been paid or claimed of the Company. -| | | | | | Vadodara - Halol Expressway, | | E-mail: rilagm@ril.com | | | | | -| --- | --- | --- | --- | --- | ------------------------------ | --- | ---------------------- | --- | --- | --- | --- | -Strategic decisions regarding the timing are covered under the Enterprise Risk by the members for seven consecutive -Vill - Asoj, Taluka – Waghodia, The Company has also uploaded these -and the usage of derivatives instruments Management section of the Management years or more as on the cut-off date, i.e. -Vadodara – 391 510, Gujarat, India TRANSFER OF UNPAID / details on the website of the IEPF Authority -such as Swaps / Futures / Options, are Discussion and Analysis Report. UNCLAIMED AMOUNTS July 12, 2019. The Company has initiated -(www.iepf.gov.in). -OIL & GAS necessary action for transfer of shares -Exposure of the Company to commodity risks, which are material is as AND SHARES TO INVESTOR -KG D6 Onshore Terminal in respect of which dividend has not The voting rights on the shares transferred -| under: | | | | | | | EDUCATION AND PROTECTION | | | | | -| ------- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | -Village Gadimoga, Tallarevu Mandal, been paid or claimed by the members to IEPF Authority shall remain frozen till the -FUND -% of such exposure hedged through East Godavari District – 533 463, consecutively since 2012-13. rightful owner claims the shares. -| | | Exposure in | | | | | During the year, the Company has credited | | | | | -| --- | --------- | --------------- | --------------------- | ----- | --------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | -| | | | commodity derivatives | Total | | | | | | | | -| | Exposure | Quantity terms | | | Andhra Pradesh, India | | | | | | | -` 24.89 crore to the Investor Education -| | towards the | towards the | | | | | | | | | | -| --- | ------------ | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Commodity Domestic International Coal Bed Methane Project (CBM) and Protection Fund (IEPF) pursuant to the -| | particular | particular | | | | | | | | | | -| --- | ----------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Name market market Village & P. O.: Lalpur, Tehsil: Burhar, District provisions of the Companies Act, 2013. -| | commodity | commodity (in | | | | | | | | | | -| --- | ---------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Shahdol, Madhya Pradesh – 484 110, India -(` in crore) 1000 Metric Due dates for transfer to IEPF, of unclaimed / unpaid dividends for the financial year 2012-13 and thereafter: -| | | | OTC Exchange OTC | Exchange* | | | | | | | | -| --- | --- | ---- | ---------------- | --------- | -------- | --- | --- | --- | --- | --- | --- | -| | | Ton) | | | TEXTILES | | | | | | | -Crude 2,10,093 72,190 - - 23% 18% 41% Naroda Manufacturing Division FY ended Declaration Date Due Date -| Middle | 1,25,629 | 30,807 | - - 31% | 14% 45% | 103 / 106, Naroda Industrial Estate, Naroda, | | | | | | | -| ----------- | -------- | ------ | ------- | ------- | --------------------------------------------- | --- | -------------- | ------------ | ------------- | --- | --- | -| | | | | | | | March 31, 2013 | June 6, 2013 | July 12, 2020 | | | -| Distillates | | | | | Ahmedabad – 382 330, Gujarat, India | | | | | | | -Light 64,984 15,584 - - 6% 46% 52% March 31, 2014 June 18, 2014 July 24, 2021 -Distillates ADDRESS FOR March 31, 2015 June 12, 2015 July 18, 2022 -Polymer 38,419 5,783 - - - - - March 31, 2016 March 10, 2016 April 15, 2023 -CORRESPONDENCE -| | | | | | | | March 31, 2017 | July 21, 2017 | August 26, 2024 | | | -| ------------------------------------------------------------------- | -------- | -------- | ----- | --- | ----------------------------------- | --- | ------------------------------------- | --------------- | ------------------ | --- | --- | -| Petchem | 35,731 | 6,206 | - - - | - | - FOR SHARES / DEBENTURES HELD | | | | | | | -| | | | | | | | March 31, 2018 | July 5, 2018 | August 4, 2025 | | | -| Intermediate | | | | | IN PHYSICAL FORM | | | | | | | -| | | | | | | | March 31, 2019 | August 12, 2019 | September 11, 2026 | | | -| Polyester | 21,625 | 2,604 | - - - | - | - KFin Technologies Private Limited | | | | | | | -| Total | 4,96,481 | 1,33,174 | | | | | | | | | | -| | | | | | (Formerly known as Karvy Fintech | | EQUITY SHARES IN THE SUSPENSE ACCOUNT | | | | | -| *Includes OTC transactions cleared through International Exchanges. | | | | | Private Limited) | | | | | | | -In terms of Regulation 39 of the Listing Regulations, the Company reports the following details in respect of equity shares lying in the -Selenium Tower B, -suspense account which were issued in demat form and physical form, respectively: -| PLANT LOCATIONS IN INDIA | | | Hoshiarpur Manufacturing | | Plot 31-32, Gachibowli Financial District, | | | | | | | -| ------------------------ | --- | --- | ------------------------- | --- | ------------------------------------------ | --- | --- | ----- | --- | -------- | --- | -| REFINING & MARKETING | | | Division | | Nanakramguda, Hyderabad - 500 032 | | | | | | | -| | | | | | | | | Demat | | Physical | | -DTA Jamnagar Refinery -| | | | Dharamshala Road, V. P. O. Chohal, District | | Tel: +91 40 67161700 | | | | | | | -| --- | --- | --- | -------------------------------------------- | --- | -------------------- | --- | --- | --- | --- | --- | --- | -No. of -| Village Meghpar / Padana, Taluka Lalpur, | | | Hoshiarpur – 146 024, Punjab, India | | Toll Free No.: 1800 425 8998 | | Particulars | | | | | -| ----------------------------------------- | --- | --- | ----------------------------------- | --- | ----------------------------- | --- | ----------- | ------- | ---------------------- | --- | -------- | -| | | | | | | | | No. of | No. of shareholders | | No. of | -| Jamnagar – 361 280, Gujarat, India | | | | | (From 9:00 a.m. to 6:00 p.m.) | | | | | | | -DTA Jamnagar Refinery shareholders equity shares (phase-wise equity shares -Fax: +91 40 67161680 -SEZ Jamnagar Refinery Village Meghpar / Padana, Taluka Lalpur, transfers) -E-mail: rilinvestor@kfintech.com -Unit of Reliance Jamnagar SEZ Jamnagar – 361 280, Gujarat, India Aggregate number of shareholders and the outstanding shares in the 96 2,616 90,058 82,59,125 -| | | | | | Website: www.kfintech.com | | suspense account lying as on April 1, 2019 | | | | | -| --- | --- | --- | --- | --- | ------------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | -Village Meghpar / Padana, Taluka Lalpur, -SEZ Jamnagar Refinery -Jamnagar – 361 280, Gujarat, India Less: Number of shareholders who approached the Company for transfer 0 0 4,228 5,68,598 -Unit of Reliance Jamnagar SEZ FOR SHARES / DEBENTURES of shares (which number is the same as shares transferred from suspense -PETROCHEMICALS Village Meghpar / Padana, Taluka Lalpur, HELD IN DEMAT FORM account during the year) -| Barabanki Manufacturing Division | | | Jamnagar – 361 280, Gujarat, India | | Investors’ concerned Depository | | | | | | | -| -------------------------------- | --- | --- | ---------------------------------- | --- | -------------------------------- | --- | ----------------------------------------------------------------------- | --- | --- | --- | ------ | -| | | | | | | | Add: Number of shareholders and aggregate number of shares transferred | 0 | 0 | 425 | 61,630 | -Dewa Road, P. O. Somaiya Nagar, Nagothane Manufacturing Division Participant(s) and / or KFin Technologies to the Unclaimed Suspense Account during the year -Barabanki – 225 123, Uttar Pradesh, India Private Limited. Less: Number of shares transferred to IEPF Authority during the year 0 0 7,366 3,19,389 -P. O. Petrochemicals Township, -Dahej Manufacturing Division Nagothane – 402 125, Roha Taluka, Aggregate number of shareholders and the outstanding shares in the 96 2,616 78,889 74,32,768 -suspense account lying as on March 31, 2020 -| P. O. Dahej – 392 130, Taluka: Vagra, | | | District Raigad, Maharashtra, India | | | | | | | | | -| --------------------------------------- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -District Bharuch, Gujarat, India -Patalganga Manufacturing -The voting rights on the shares in the suspense account shall remain frozen till the rightful owner claim the shares. -| Hazira Manufacturing Division | | | Division | | | | | | | | | -| ----------------------------------------- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -| Village Mora, P. O. Bhatha, Surat-Hazira | | | B-1 to B-5 & A3, MIDC Industrial Area, | | | | | | | | | -| Road, Surat – 394 510, Gujarat, India | | | Patalganga – 410 220, District Raigad, | | | | | | | | | -Maharashtra, India -| 202 | | | | | | | | | | | 203 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | Integrated Annual Report 2019-20 | | | | Overview Review | Statements | -| --- | --- | -------------------------------- | --- | --- | --- | --------------- | ---------- | -WEBLINKS FOR THE MATTERS REFERRED IN THIS REPORT ARE AS UNDER: COMPLIANCE OF CORPORATE GOVERNANCE REQUIREMENTS SPECIFIED IN REGULATION 17 TO 27 AND -REGULATION 46(2)(b) TO (i) OF LISTING REGULATIONS -| Particulars | Website link | | | | Compliance | | | -| ----------- | ------------ | --- | --- | --- | ----------- | --- | --- | -Sr. -| | | | Particulars | Regulation | Status | Key Compliance observed | | -| --- | --- | --- | ----------- | ---------- | -------- | ----------------------- | --- | -No. -| Policies and Code | | | | | Yes / No / N.A. | | | -| ----------------- | --- | --- | --- | --- | --------------- | --- | --- | -Code of Conduct https://www.ril.com/DownloadFiles/IRStatutory/Code-of-Conduct.pdf 1 Board of Directors 17 Yes • Composition and Appointment of Directors -• Meetings and quorum -| Our Code | http://www.ril.com/DownloadFiles/IRStatutory/ourcode.pdf | | | | | | | -| -------- | -------------------------------------------------------- | --- | --- | --- | --- | --- | --- | -• Review of compliance reports -| Familarisation Programme for Independent | https://www.ril.com/InvestorRelations/Downloads.aspx | | | | | | | -| ----------------------------------------- | ---------------------------------------------------- | --- | --- | --- | --- | --- | --- | -• Plans for orderly succession for appointments -Directors -• Code of Conduct -| Remuneration Policy for Directors, | http://www.ril.com/DownloadFiles/IRStatutory/Remuneration-Policy-for-Directors.pdf | | | | | | | -| ----------------------------------- | ---------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | -• Fees / compensation to Non-Executive Directors -Key Managerial Personnel and other -• Minimum information to be placed before the Board -employees -• Compliance Certificate by Chief Executive Officer and Chief -| Policy for selection of Directors and | http://www.ril.com/DownloadFiles/IRStatutory/Policy-for-Selection-of-Directors.pdf | | | | | | | -| -------------------------------------- | ---------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | -Financial Officer -determining Directors’ independence -• Risk assessment and risk management plan -Policy for determining Material Subsidiaries http://www.ril.com/DownloadFiles/IRStatutory/Material-Subsidiaries.pdf • Performance evaluation of Independent Directors -Policy on Materiality of Related Party http://www.ril.com/DownloadFiles/IRStatutory/Policy-on-Materiality-of-RPT.pdf • Recommendation of Board for each item of special business -| | | | 2 Maximum Number of | 17A | Yes | • Directorships in listed entities | | -| --- | --- | --- | -------------------- | --- | --- | ----------------------------------- | --- | -Transactions and on dealing with Related -Directorships -Party Transactions -| | | | 3 Audit Committee | 18 | Yes | • Composition | | -| ------------------------------------------ | ------------------------------------------------------------------ | --- | ----------------- | --- | --- | -------------- | --- | -| Policy on Determination and Disclosure of | http://www.ril.com/DownloadFiles/IRStatutory/MaterialityPolicy.pdf | | | | | | | -• Meetings and quorum -Materiality of Events and Information and -• Chairperson present at Annual General Meeting -Web Archival Policy -• Role of the Committee -Vigil Mechanism and Whistle- Blower Policy http://www.ril.com/DownloadFiles/IRStatutory/Vigil-Mechanism-and-Whistle-Blower-Policy.pdf 4 Nomination and Remuneration 19 Yes • Composition -Reports Committee • Chairperson present at Annual General Meeting -• Meetings and quorum -| Quarterly, Half-yearly and Annual Financial | http://www.ril.com/InvestorRelations/FinancialReporting.aspx | | | | | | | -| -------------------------------------------- | ------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | -• Role of the Committee -Results (from 2002 to 2020) -Presentation to institutional investors and http://www.ril.com/InvestorRelations/FinancialReporting.aspx 5 Stakeholders Relationship 20 Yes • Composition -analysts (from 1999 to 2020) Committee • Chairperson present at Annual General Meeting -• Meetings -| Annual Report (from 1976 to 2020) | http://www.ril.com/InvestorRelations/FinancialReporting.aspx | | | | | | | -| --------------------------------- | ------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | -• Role of the Committee -Chairman’s Communication (from 2002 to http://www.ril.com/InvestorRelations/Chairman-Communication.aspx 6 Risk Management Committee 21 Yes • Composition -| 2019) | | | | | | • Meetings | | -| ----- | --- | --- | --- | --- | --- | ----------- | --- | -Sustainability Reports http://www.ril.com/Sustainability/CorporateSustainability.aspx • Role of the Committee -| | | | 7 Vigil Mechanism | 22 | Yes | • Vigil Mechanism for Directors and employees | | -| --- | --- | --- | ----------------- | --- | --- | ---------------------------------------------- | --- | -Shareholder Information -• Adequate safeguards against victimisation -| Composition of Board of Directors and | http://www.ril.com/OurCompany/Leadership/BoardOfDirectors.aspx | | | | | | | -| -------------------------------------- | -------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | -• Direct access to Chairperson of Audit Committee -Profile of Directors -| | | | 8 Related party transactions | 23 | Yes | • Policy on Materiality of related party transactions and dealing | | -| --- | --- | --- | ---------------------------- | --- | --- | ------------------------------------------------------------------- | --- | -Composition of various Committees of the http://www.ril.com/OurCompany/Leadership/BoardCommittees.aspx with related party transactions -Board and their terms of reference • Prior approval including omnibus approval of Audit Committee for -ESOS Disclosure under SEBI (Share Based https://www.ril.com/DownloadFiles/IRStatutory/ESOS-2006-Disclosure-2019-20.pdf related party transactions -Employee Benefits) Regulations, 2014 as • Periodical review of related party transactions -| on March 31, 2020 | https://www.ril.com/DownloadFiles/IRStatutory/ESOS-2017-Disclosure-2019-20.pdf | | | | | | | -| ----------------- | ------------------------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | -• Disclosure on related party transactions -Details of unpaid and unclaimed http://www.ril.com/InvestorRelations/ShareholdersInformation.aspx 9 Subsidiaries of the Company 24 Yes • Appointment of Company’s Independent Director on the Board -amounts lying with the Company as on of material subsidiaries -date of last Annual General Meeting • Review of financial statements and investments of subsidiaries by -(i.e. August 12, 2019) and details of shares the Audit Committee -transferred to IEPF during financial year • Minutes of the Board of Directors of the subsidiaries are placed at -| 2019-20. | | | | | | the meeting of the Board of Directors | | -| -------- | --- | --- | --- | --- | --- | ------------------------------------- | --- | -• Significant transactions and arrangements of subsidiaries are -| Secretarial Audit Report of Material | http://www.ril.com/DownloadFiles/IRStatutory/Secretarial-Audit-Reports-of-material-subsidiaries.pdf | | | | | | | -| ------------------------------------- | --------------------------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | -placed at the meeting of the Board of Directors -Unlisted Subsidiary -| | | | 10 Secretarial Audit | 24A | Yes | • Annual Secretarial Audit Report and Annual Secretarial | | -| -------------------------------- | ---------------------------------------------------------------------------------- | --- | -------------------- | --- | --- | ---------------------------------------------------------- | --- | -| Build-up of Equity Share Capital | https://www.ril.com/DownloadFiles/IRStatutory/Build-up-of-Equity-Share-Capital.pdf | | | | | | | -Compliance Report -Shareholders’ Referencer http://www.ril.com/DownloadFiles/IRForms/Shareholders-Referencer.pdf • Secretarial Audit Report of material unlisted subsidiaries -Investor Contacts https://www.ril.com/InvestorRelations/Investor-Contacts.aspx incorporated in India -204 205 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -Integrated Annual Report 2019-20 Overview Review Statements -Compliance CEO AND CFO CERTIFICATION -Sr. -No. Particulars Regulation Status Key Compliance observed The Chairman and Managing Director (CMD) and the Chief Financial Officer (CFO) of the Company give annual certification on financial -Yes / No / N.A. -reporting and internal controls to the Board in terms of Regulation 17(8) of the Listing Regulations, copy of which is attached to this Report. -11 Obligations with respect to 25 Yes • Maximum directorships and tenure -The CMD and the CFO also give quarterly certification on financial results while placing the financial results before the Board in terms of -Independent Directors • Meetings of Independent Directors -Regulation 33(2) of the Listing Regulations. -• Cessation and appointment of Independent Directors -• Familiarisation of Independent Directors -• Declaration from Independent Director that he / she meets the -COMPLIANCE CERTIFICATE OF THE AUDITORS -criteria of independence -• Directors and Officers insurance for all the Independent Directors Certificate from the Company’s Auditors, S R B C & CO LLP and D T S & Associates LLP, Chartered Accountants, confirming compliance -12 Obligations with respect to 26 Yes • Memberships / Chairmanships in Committees with conditions of Corporate Governance, as stipulated under Regulation 34 of the Listing Regulations, is attached to this Report. -employees including Senior • Affirmation on compliance of Code of Conduct by Directors and -Management, Key Managerial Senior Management -Personnel, Directors and • Disclosure of shareholding by Non-Executive Directors CERTIFICATE ON COMPLIANCE WITH CODE OF CONDUCT -Promoters • Disclosures by Senior Management about potential I hereby confirm that the Company has obtained from all the members of the Board and Senior Management Personnel, the affirmation -conflicts of interest that they have complied with the ‘Code of Conduct’ and ‘Our Code’ in respect of the financial year 2019-20. -• No agreement with regard to compensation or profit sharing in -connection with dealings in securities of the Company by Key -Managerial Personnel, Director and Promoter Mukesh D. Ambani -13 Other Corporate Governance 27 Yes • Compliance with discretionary requirements Chairman and Managing Director -requirements • Filing of quarterly, half-yearly and yearly compliance report on -Mumbai, April 30, 2020 -Corporate Governance -14 Website 46(2)(b) to (i) Yes • Terms and conditions of appointment of Independent Directors -• Composition of various Committees of the Board of Directors -• Code of Conduct of Board of Directors and Senior -Management Personnel -• Details of establishment of Vigil Mechanism / -Whistle-blower policy -• Policy on dealing with related party transactions -• Policy for determining material subsidiaries -• Details of familiarisation programmes imparted to -Independent Directors -ANNUAL SECRETARIAL COMPLIANCE REPORT -Pursuant to the SEBI circular no. CIR/CFD/ CMD1/27/2019 dated February 8, 2019, the Company has obtained an Annual Secretarial -Compliance Report from Dr. K. R. Chandratre, Practising Company Secretary, confirming compliance of SEBI Regulations / Circulars / -Guidelines issued thereunder and applicable to the Company. There are no observations or adverse remarks in the said report. -NO DISQUALIFICATION CERTIFICATE FROM COMPANY SECRETARY IN PRACTICE -Certificate from Dr. K. R. Chandratre, Practising Company Secretary, confirming that none of the Directors on the Board of the Company -have been debarred or disqualified from being appointed or continuing as directors of companies by the SEBI, Ministry of Corporate -Affairs, or any such other Statutory Authority, as stipulated under Regulation 34(3) of the Listing Regulations, is attached to this Report. -206 Naye India Ka Naya Josh 207 - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -| | | Integrated Annual Report 2019-20 | | | Overview Review | Statements | -| --- | --- | -------------------------------- | --- | --- | --------------- | ---------- | -NO DISQUALIFICATION CERTIFICATE FROM COMPANY SECRETARY IN PRACTICE CEO / CFO CERTIFICATE -| To: | | | To, | | | | -| --------------------------- | --- | --- | --------------------------- | --- | --- | --- | -| The Members | | | The Board of Directors | | | | -| Reliance Industries Limited | | | Reliance Industries Limited | | | | -3rd Floor, Maker Chambers IV, 222, -Nariman Point, Mumbai - 400 021. 1. W e have reviewed financial statements and the cash flow statement of Reliance Industries Limited (the Company) for the year ended -March 31, 2020 and to the best of our knowledge and belief: -I have examined the relevant registers, records, forms, returns and disclosures received from the Directors of Reliance Industries Limited - i. t hese statements do not contain any materially untrue statement or omit any material fact or contain statements that -having CIN L17110MH1973PLC019786 and having registered office at 3rd Floor, Maker Chambers IV, 222, Nariman Point, Mumbai - 400 021, -might be misleading; -Maharashtra, India (hereinafter referred to as ‘the Company’), produced before me by the Company for the purpose of issuing this -Certificate, in accordance with Regulation 34(3) read with Schedule V Para-C Clause 10(i) of the Securities Exchange Board of India (Listing ii. t hese statements together present a true and fair view of the Company’s affairs and are in compliance with existing accounting -Obligations and Disclosure Requirements) Regulations, 2015. standards, applicable laws and regulations. -In my opinion and to the best of my information and according to the verifications (including Directors Identification Number (DIN) status at 2. T here are, to the best of our knowledge and belief, no transactions entered into by the Company during the year which are fraudulent, -the portal www.mca.gov.in) as considered necessary and explanations furnished to me by the Company & its officers, I hereby certify that illegal or violative of the Company’s Code of Conduct. -none of the Directors on the Board of the Company as stated below for the financial year ending on 31 March 2020, have been debarred -3. W e accept responsibility for establishing and maintaining internal controls for financial reporting and we have evaluated the -or disqualified from being appointed or continuing as Directors of companies by the Securities and Exchange Board of India, Ministry of -effectiveness of Company’s internal control systems pertaining to financial reporting. We have not come across any reportable -Corporate Affairs, or any such other Statutory Authority. -deficiencies in the design or operation of such internal controls. -Sr. Date of appointment 4. W e have indicated to the Auditors and the Audit Committee that: -| Name of the Director | DIN | | | | | | -| -------------------- | --- | -------------- | --- | --- | --- | --- | -| No. | | in the Company | | | | | - i. t here are no significant changes in internal controls over financial reporting during the year; -| 1. Mukesh Dhirubhai Ambani | 00001695 | 01.04.1977 | | | | | -| -------------------------- | -------- | ---------- | --- | --- | --- | --- | -2. Yogendra Premkrishna Trivedi 00001879 16.04.1992 ii. t here are no significant changes in accounting policies during the year; and -| 3. Dipak Chand Jain | 00228513 | 04.08.2005 | | | | | -| ------------------- | -------- | ---------- | --- | --- | --- | --- | - iii. there are no instances of significant fraud of which we have become aware. -| 4. Raghunath Anant Mashelkar | 00074119 | 09.06.2007 | | | | | -| ---------------------------- | -------- | ---------- | ------------------------------ | ----------------------- | ----------------------------- | --- | -| 5. Adil Zainulbhai | 06646490 | 20.12.2013 | | | | | -| 6. Raminder Singh Gujral | 07175393 | 12.06.2015 | | | | | -| | | | Mukesh D. Ambani | Alok Agarwal | Srikanth Venkatachari | | -| 7. Shumeet Banerji | 02787784 | 21.07.2017 | | | | | -| | | | Chairman and Managing Director | Chief Financial Officer | Joint Chief Financial Officer | | -| 8. Arundhati Bhattacharya | 02011213 | 17.10.2018 | | | | | -| 9. Nita Mukesh Ambani | 03115198 | 18.06.2014 | | | | | -10. Nikhil Rasiklal Meswani 00001620 26.06.1986 Mumbai, April 30, 2020 -| 11. Hital Rasiklal Meswani | 00001623 | 04.08.1995 | | | | | -| -------------------------------- | -------- | ---------- | --- | --- | --- | --- | -| 12. Madhusudana Sivaprasad Panda | 00012144 | 21.08.2009 | | | | | -| 13. Pawan Kumar Kapil | 02460200 | 16.05.2010 | | | | | -| 14. Veerayya Chowdary Kosaraju | 08485334 | 18.10.2019 | | | | | -Ensuring the eligibility of the appointment / continuity of every Director on the Board is the responsibility of the management of the -Company. My responsibility is to express an opinion on these, based on my verification. This certificate is neither an assurance as -to the future viability of the Company nor of the efficiency or effectiveness with which the management has conducted the affairs -of the Company. -Dr. K. R. Chandratre -FCS No. 1370, C P No: 5144 -Place: Pune -Date: 30 April 2020 -UDIN: F001370B000192297 -208 209 -Naye India Ka Naya Josh - -Corporate Governance Report (contd) Reliance Industries Limited Corporate Management Governance Financial Notice -Integrated Annual Report 2019-20 Overview Review Statements -INDEPENDENT AUDITOR’S CERTIFICATE ON COMPLIANCE WITH THE CONDITIONS OF CORPORATE -GOVERNANCE AS PER PROVISIONS OF CHAPTER IV OF SECURITIES AND EXCHANGE BOARD OF INDIA -(LISTING OBLIGATIONS AND DISCLOSURE REQUIREMENTS) REGULATIONS, 2015 (AS AMENDED) -To the Members -Reliance Industries Limited -3rd Floor, Maker Chambers IV, 222, Nariman Point, -Mumbai - 400 021, India -1. T he Corporate Governance Report prepared by Reliance Industries Limited (“the Company”), contains details as stipulated in OTHER MATTERS AND RESTRICTION ON USE -regulations 17 to 27 and clauses (b) to (i) of regulation 46(2) and para C and D of Schedule V of Securities and Exchange Board of India 10. T his Report is neither an assurance as to the future viability of the Company nor the efficiency or effectiveness with which the -(Listing Obligations and Disclosure Requirements) Regulations, 2015, as amended (“the Listing Regulations”) (‘applicable criteria’) with management has conducted the affairs of the Company. -respect to Corporate Governance for the year ended March 31, 2020. This report is required by the Company for annual submission to -11. T his Report is addressed to and provided to the members of the Company solely for the purpose of enabling it to comply with its -the Stock exchange and to be sent to the Shareholders of the Company. -obligations under the Listing Regulations and should not be used by any other person or for any other purpose. Accordingly, we do -MANAGEMENT’S RESPONSIBILITY not accept or assume any liability or any duty of care or for any other purpose or to any other party to whom it is shown or into whose -2. T he preparation of the Corporate Governance Report is the responsibility of the Management of the Company including hands it may come without our prior consent in writing. We have no responsibility to update this Report for events and circumstances -the preparation and maintenance of all relevant supporting records and documents. This responsibility also includes the occurring after the date of this Report. -design, implementation and maintenance of internal control relevant to the preparation and presentation of the Corporate -Governance Report. For D T S & Associates LLP For S R B C & CO LLP -Chartered Accountants Chartered Accountants -3. T he Management along with the Board of Directors are also responsible for ensuring that the Company complies with the conditions -ICAI Firm Reg Number: 142412W/ W100595 ICAI Firm Reg Number: 324982E/E300003 -of Corporate Governance as stipulated in the Listing Regulations, issued by the Securities and Exchange Board of India. -AUDITOR’S RESPONSIBILITY per T. P. Ostwal per Vikas Kumar Pansari -4. O ur responsibility is to provide a reasonable assurance in the form of an opinion whether the Company has complied with the Partner Partner -condition of Corporate Governance, as stipulated in the Listing Regulation. Membership No.: 030848 Membership No.: 093649 -5. W e conducted our examination of the Corporate Governance Report in accordance with the Guidance Note on Reports or Certificates UDIN: 20030848AAAAAO8055 UDIN: 20093649AAAAAP4660 -for Special Purposes and the Guidance Note on Certification of Corporate Governance, both issued by the Institute of Chartered Place: Mumbai Place: Mumbai -Accountants of India (“ICAI”). The Guidance Note on Reports or Certificates for Special Purposes requires that we comply with the Date: April 30, 2020 Date: April 30, 2020 -ethical requirements of the Code of Ethics issued by ICAI. -6. W e have complied with the relevant applicable requirements of the Standard on Quality Control (SQC) 1, Quality Control for Firms that -Perform Audits and Reviews of Historical Financial Information, and Other Assurance and Related Services Engagements. -7. T he procedures selected depend on the auditor’s judgement, including the assessment of the risks associated in compliance of -the Corporate Governance Report with the applicable criteria. The procedures includes but not limited to verification of secretarial -records and financial information of the Company and obtained necessary representations and declarations from directors including -independent directors of the Company. -8. T he procedures also include examining evidence supporting the particulars in the Corporate Governance Report on a test basis. -Further, our scope of work under this report did not involve us performing audit tests for the purposes of expressing an opinion on the -fairness or accuracy of any of the financial information or the financial statements of the Company taken as a whole. -OPINION -9. B ased on the procedures performed by us as referred in paragraph 7 and 8 above and according to the information and explanations -given to us, we are of the opinion that the Company has complied with the conditions of Corporate Governance as stipulated in the -Listing Regulations, as applicable for the year ended March 31, 2020, referred to in paragraph 1 above. -210 Naye India Ka Naya Josh 211 \ No newline at end of file +KKANSAS HISTORIC RESOURCES INVENTORY +SURVEY REPORT + +Aggieville Commercial District +Historic Survey +May 2020 + +K-State parade in Aggieville, 1939 Royal Purple + +Brenda R. Spencer +Michelle Spencer +10150 Onaga Road +Wamego, KS 66547 +Brenda@SpencerPreservation.com + +The survey report and inventory forms, which are the subject of this project, have been financed in part with federal +funds from the National Park Service, a division of the United States Department of the Interior, and administered by +the Kansas State Historical Society. The City of Manhattan received a Historic Preservation Fund Grant +through the Kansas State Historical Society for the project. + +Prepared for the +City of Manhattan +May 2020 + + Aggieville Commercial District + +HISTORIC SURVEY REPORT + +MAY 2020 + +I. OVERVIEW + +The Aggieville Community Vision was adopted in the spring of 2017. Lead by the +public input of over 4,200 participants, stakeholders, and Manhattan citizens, it +identifies several redevelopment and public improvement opportunities in the +district. The plan included a recommendation to preserve the historic character of +Aggieville’s historic core and this survey project resulted in part from that +commitment. The plan's implementation is now underway including construction of +a hotel and planning for a parking garage. + +The City of Manhattan was awarded a Historic Preservation Fund Grant by the Kansas +State Historical Society in May 2019. Spencer Preservation was hired by the +City of Manhattan through a competitive bid process in November 2019 to conduct a +historic resource survey of the Aggieville Commercial District. Ben Chmiel, Planner in +the Community Development Department, City of Manhattan was the City point of +contact. Brenda Spencer was the contact from Spencer Preservation. Additional +assistance was provided by Aggieville Historian Dan Walter and by Linda Glasgow of +the Riley County Historical Museum. Chmiel and Spencer met with Kansas State +Historical Society staff on October 24, 2019 to kick off the project. + +The boundary for the survey area was identified prior to applying for the HPF grant +and included commercial and residential properties in the area bounded by +Anderson/ Bluemont Avenue on the north, Fremont Street on the south, 14th Street +on the west and 11th Street on the east. The commercial center at 702 N. 11th Street, +on the east side of the 11th Street was the only property surveyed outside of the +defined boundary. Triangle Park at the northwest corner of the district was included +in the survey. The defined area totaled ninety resources and each was assigned a +sequence number used on report maps and lists, as well as an inventory number in +the Kansas Historic Resources Inventory (KHRI) database. Only two of the resources +had previously been surveyed and therefore had existing records in KHRI; these +records were updated as a part of current project. One of the surveyed properties – +the YMCA/St. Mary’s Hospital Building at 11th and Fremont Streets is listed in the +Register of Kansas Historic Places. A map of the Survey Area with sequence number +of each resource is provided on page 4. + +The survey was completed in May 2020 in the midst of the Covid 19 pandemic. The +findings and recommendations were reviewed with the City of Manhattan project +representative. The project was to include a public meeting with Aggieville business +and property owners. Given the public health concerns and current assembly +challenges, the City of Manhattan opted to delay the public meeting until a more +suitable time for the business community. With concurrence from Kansas State +Historical Society staff, it was agreed that Spencer will work with the City of +Manhattan to host a public meeting as a part of this project within the next 12 +months. + +Brenda R. Spencer +10150 Onaga Road +Wamego, KS 66547 +Brenda@SpencerPreservation + +Historic Preservation Fund Grant +2019-05 +Kansas State Historical Society + +All photos illustrating existing +conditions were taken by Spencer +Preservation in 2020 unless +otherwise noted. + +The survey report and inventory +forms, which are the subject of this +project, have been financed in part +with Federal funds from the +National Park Service, a division of +the United States Department of the +Interior, and administered by the +Kansas State Historical Society. The +contents and opinions, however, do +not necessarily reflect the view or +policies of the United States +Department of the Interior or the +Kansas State Historical Society. + +This program receives Federal funds +from the National Park Service. +Regulations of the U. S. Department +of the Interior strictly prohibit +unlawful discrimination in +departmental Federally Assisted +Programs on the basis of race, color, +national origin, age or handicap. +Any person who believes he or she +has been discriminated against in +any program, activity, or facility +operated by a recipient of Federal +assistance should write to: Director, +Equal Opportunity Program, U. S. +Department of the Interior, National +Park Service, P.O. 37127, +Washington, DC 20013-7127. + + II. METHODOLOGY + +A. Field Work +Spencer Preservation surveyed a total of 90 properties during +initial field work conducted January 5-7, 2020. During this +process Spencer examined each property in the survey area. +Spencer took digital photos of each resource that conform to +KSHS standards for survey documentation, noting +architectural style, primary and secondary materials, +configuration and materials, windows and storefronts, +condition, present use, and significant alterations or +additions. + +Each property in the survey area was given a sequence +number generally working from the northeast corner of the +survey area at 11th Street and Anderson Avenue clockwise +through the area ending at Triangle Park in the northwest +corner of the survey area. + +Riley County GIS maps were used for the survey. The City of +Manhattan Community Development Department prepared a +map with the survey resources individually numbered. +Spencer Preservation developed the block maps used as site +plans for each surveyed property. + +B. Research and Data Collection +Between November 2019 and March 2020 Spencer +Preservation integrated field research and archival data. The +Riley County GIS Database, accessed through Open Records +for Kansas Appraisers (ORKA) online database, was used to +collect basic property information including parcel numbers, +addresses, legal descriptions, and contact information for +current owners. + +Sanborn Fire Insurance Maps for the City of Manhattan were +used to assist with dating building construction and additions. +The Aggieville area was first shown, with the north side of +Moro Street only, on the 1905 Sanborn Map. The 1923 map +was the first to show the entire Aggieville commercial area +and the 1930 map was the first to include the blocks west of +N. Manhattan Avenue. The 1930 map was updated in 1947 +providing the final fire insurance map. The Sanborn Fire +Insurance Maps, accessed online through Kansas City Public +Library, often confirmed building dates of construction or +narrowed the dates to a period between mappings or after +the final map in 1947. The maps were closely scrutinized for +alterations in building footprints to determine date of +additions and when new buildings were built replacing +former structures. + +The Riley County Historical Museum has a booklet titled Fire +Insurance Rates, Manhattan, Riley County, Kansas dated July +16, 1926, with 1929 and 1930 updates/overlays. This booklet +provided valuable data on individual buildings including +current owners and businesses. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Spencer Preservation conducted archival research using the +extensive collections of Kansas State University Morse +Department of Special Collections (KSU SC) and the Riley +County Historical Society and Museum (RCHM). The museum +provided a wealth of information in clipping files, notes, +photographs, Manhattan Telephone Directories, and Polk City +Directories, the latter also available at the Manhattan Public +Library. Kansas State University Royal Purple Yearbooks, +available at KSU Libraries Digital Collections online and the +Manhattan Public Library, provided photographs and business +advertisements that helped date many businesses. + +The 1917 Royal Purple Yearbook contains advertisements for number +of Aggieville businesses including J.L. John’s Confectionery (1123 +Moro Street) and Lisk Twin Photos (1212 Moro Street). + +The earliest local phone directory available from the Riley +County Historical Museum was 1911 with directories +available for most years after 1920. The early Royal Purple +yearbooks documented Aggieville’s first commercial +businesses. The first available Polk City Directory at RCHM or +online is 1938. The phone directories and Polk Directories +were helpful in determining occupancy and longevity of +businesses. + +The 1923 Sanborn Map illustrates the Aggieville commercial area +developing along Moro St. and N. Manhattan Ave. (Sheet 13). + +2 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +The Riley County Historical Museum is the repository of a +Historic Survey conducted in 1979 by volunteers including KSU +students and community members. This survey provided +valuable photos of buildings forty years ago thus helping to +document alterations since that time. Linda Glasgow, Curator +of Archives and Library at RCHM assisted with research at the +museum and supplemented our research during the Covid 19 +shutdown. Glasgow also provided a connection to David +Fiser, two-term Mayor of the City of Manhattan, who grew up +in the 1300 block of Fremont Street. Fiser provided useful +information on residents of the block and the Aggieville area. +Another valuable source of information on Aggieville +businesses was long-time Varney’s employee and historian, +Dan Walter. Walter is author of several Aggieville books and +operates the Aggieville Archives website. He generously +shared his memories, research and personal collection to +supplement our research. Finally, newspapers.com online +was used to fill gaps in research on individual buildings, +particularly those built in the past 50 years. + +Property histories found in the “Historic Function Remarks” +field of the database/survey forms reflect a record of data +found in the above named sources. The sources of historic +data in the survey forms are noted in parentheses using the +abbreviations listed below. A complete bibliography is +provided at the end of this report. + +RCHM-Riley County Historical Society and Museum +KSU SC – Kansas State University Morse Department of Special Collections +KAC/KSU RP – Kansas Agricultural College/KSU Royal Purple Yearbook +MPL – Manhattan Public Library +SB – Sanborn Maps +TD – Telephone Directories +PD – PolK City Directories +FI – 1926 Fire Insurance Booklet at RCHM +O’Brien – Architects & Buildings of Manhattan KS by Dr. Patricia J. O’Brien +Walter – Dan Walter and Aggieville Archives website +Local newspapers including: + +MC – Morning Chronicle +MR – Manhattan Republic +MT – Manhattan Tribune +MM – Manhattan Mercury +MN – Manhattan Nationalist +Col – K-State Collegian +CN – Construction News + +North side of 1200 block of Moro Street. Dan Walter, Aggieville +Archives (1950 Royal Purple). + +D. Data Analysis and Evaluation of Historic Integrity +In May 2020, the team completed an analysis of the survey +data, identified primary periods of historical and architectural +development and prominent characteristics of the surveyed +properties. A map illustrating dates of construction of +surveyed resources is provided in Section IV – Survey Results. + +Few buildings surveyed retain their original appearance; most +have experienced multiple former alterations. In order to be +deemed eligible for listing on federal or state historic +registers, a property must generally retain the physical +features and original materials that define both why a +property is significant and when it gained significance. +Replacement of storefronts, windows, and even entire +facades were common alterations of historic commercial +buildings particularly during local boom years when +merchants had money and were reinvesting in their buildings +to attract new customers. It is the degree to which the past +modifications were in keeping with the building’s historic +character that determines the level of historic integrity a +property retains, or whether the past modifications reflected +a major design change that has gained significance in its own +right (reflecting the period of the alterations). An analysis of +the historic integrity of each building was made based on +existing appearance and obvious alterations, compared to +historic views when possible. Results are summarized below. +The contributing status of each surveyed resource can be +found on the map and list in Sections IV and V. + +C. Data Entry +Survey data was entered into a spreadsheet from which it +was uploaded to the Kansas Historic Resource Inventory +(KHRI) database. The KSHS Survey Coordinator merged the +database populating survey forms for each individual building. +Spencer Preservation then uploaded current photos, +site plans, and available historic views to each survey +form. Following approval by KSHS, the individual survey +forms are now available in the public database online at +http://khri.kansasgis.org/ Select: Manhattan – Aggieville +Commercial District Survey (HPF 2019) under ‘Survey Project.’ + +E. Recommendations and Report +Spencer Preservation analyzed the survey results, +summarized the survey project, and provided +recommendations for future actions in this Survey Report. +Ben Chmiel of the Community Development Department, City +of Manhattan prepared and provided the maps for the survey +report. The report was submitted to the City of Manhattan +and the Kansas State Historical Society. The City of +Manhattan and Spencer Preservation will share the survey +findings in a local public meeting within the next year. + +3 + + MAP OF SURVEY AREA + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +4 + + III. HISTORIC CONTEXT +by Michelle Spencer. + +A College Town on the Prairie +Manhattan was founded at the junction of the Big Blue and +Kansas Rivers by two anti-slavery groups: the New England +Immigrant Aid Society and Hartford Steamboat Colony. The +town was incorporated in 1857 and the arrival of the Union +Pacific Railroad in 1866 made it a regional hub between +Lawrence, Topeka and Ft. Riley. Even as the town was being +settled discussions among the founders arose about the +establishment of an agricultural college.1 Washington +Marlatt, Isaac Goodnow, Samuel Houston, Joseph Denison, +and others joined together in support of a Methodist college +called Bluemont College in Manhattan. When the town +companies agreed, Houston and Goodnow bought fifty shares +of Manhattan town stock & 100 building lots.2 + +The original Bluemont campus was located about three miles +west of downtown surrounded by farmland. A three-story +stone building was erected in 1859 and courses were offered +in 1860 at a cost of $3 to $5 each. At the end of the first year +Bluemont College had fifteen students; not enough to fund +living expenses for more than one overworked instructor.3 +The founders, committed to the idea of a college in +Manhattan, were determined to convince the legislature that +the state college should be located in Manhattan; however, +others in Lawrence and Emporia had similar plans. Goodnow +rallied support and by May 1861 a bill was introduced to +locate the state university in Manhattan. It passed both +houses but was vetoed by the governor. Then in 1862, the +U.S. Congress passed the Morrill Act, which provided grants +of land for the endowment of agricultural colleges. The +supporters of a Manhattan-located college saw their chance +to obtain their initial goal: an agricultural institution. Shortly +thereafter the Trustees of the Bluemont Central College +Association gifted the State of Kansas the Bluemont College +building, the library, and one hundred acres of land.4 +Established as the first land-grant college in the state, Kansas +State Agricultural College (KSAC) admitted men and women +equally; only the second college in the United States to do +so.5 + +The college’s location and specifically the distance from + +1 Julius Terrass Willard. “Bluemont Central College: The Forerunner +of Kansas State College.” Kansas Historical Quarterly (KHQ). May +1945 (Vol. 13 No. 6), 323-357. Transcribed and digitized at +https://www.kshs.org/p/bluemont-central-college/12997. +2 Willard, KHQ, quoting Isaac Goodnow’s diary, dated Sunday, 8 Feb +1857. +3 Ibid. +4 Ibid. +5 Kevin G.W. Olson. Frontier Manhattan: Yankee Settlement to +Kansas Town, 1854-1894. Lawrence, KS: University Press of Kansas, +2013, 145. Olson is quoting Manhattan Independent, 29 Jun 1867. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +downtown required students to board at the college as only a +few farm houses were extant in the school’s vicinity.6 A +poorly-constructed boarding hall was hastily added to the +new campus. Boarding students was not the only problem; +the land was not well-suited for farming; the paramount +necessity for agricultural learning. The Board of Regents and +the Town of Manhattan began to consider how to resolve the +problems of the college’s location. A March 1871 editorial in +the Manhattan Nationalist detailed the issues and proposed +solutions. A $12,000 bond issue was placed before town +residents that would fund the purchase of a “college farm” +comprised of fertile farm land near the town site. In return +for the land, the college regents would expend in one year +$25,000-$30,000 in improving the land in order to establish a +large experimental farm that included fencing, farm +structures, and stock. The newspaper argued that the result +would be the infusion of nearly $40,000 into the local +economy with the employment of numerous local carpenters, +masons and farm laborers.7 Additionally, the resulting +college close to town would be a permanent draw of students +desiring a quality education and cement Manhattan’s role as +one of the premier locations in the state.8 On April 20th the +bond issue was passed by the 1,173 town residents to fund +the purchase of 160 acres of farmland where the college +could prosper, nearer to the city. At the same time the +coursework was revised to reflect the mandates of the Morrill +Act by which courses other than agriculture were required to +be offered.9 Four departments: Agriculture, Military, Literary +and Mechanical offered instruction in scientific and classical +studies in addition to practical and industrial pursuits.10 New +courses were offered in veterinary science, ornithology, +entomology, horticulture, silvaculture (forestry), and +mechanics, many of which utilized the improved college farm +that had been funded by the bond issue.11 + +In 1873 the school moved permanently to the new college +farm (the existing site of KSU). The location was still rural, +two miles northwest of the Union Pacific Railway station +downtown and a mile from the western edge of the +community’s core residential and commercial areas.12 The +influence of the college on the town was readily apparent. In +1875, there were approximately 120 students enrolled when +the college established a weekly newspaper, the Industrialist + +6 Willard, KHQ. +7 “Kansas State Agricultural College.” Manhattan Nationalist. 19 May +1871. 2. +8 Ibid. +9 Manhattan Mercury. 17 Feb 1963. 32. +10 Manhattan Nationalist. 19 May 1871. 2. +11 Ibid. +12 Brenda R. Spencer and Christy Davis. Downtown Manhattan +Historic District Multiple Property Submission. Washington, DC: +National Park Service, 2006. National Parks Service, 2006. + +5 + + (1875-1929).13 After only two issues downtown businesses +were advertising in its pages. The 1875 Annual Report of the +State Board of Agriculture includes an artist’s rendering of the +college campus. The area surrounding campus included a +single boarding house and a few private residences in a vast +open area with a few trees. Two years later the residence of +George Evans was the only building standing on Moro Street +from 11th Street to N. Manhattan Avenue.14 Enrollment +continued to grow, more than doubling to 303 by 1877.15 On +campus, Anderson Hall, the main college building, +constructed in 1879, was expanded twice in the early 1880s. + +As the regional rail hub, Manhattan was the business center +for Riley County. The town did not become a railroad or even +a western town; rather it was described as having “the +distinct character of older eastern cities.”16 The greatest +influence on the nature of the town was the college and its +students. With over 2,100 residents, historian William Cutler +described the town of Manhattan in 1883: “Scattered over +the town are modest cottages and palatial residences, tasty +church edifices, store buildings of large and small +proportions, banks, hotels, mills and elevators, while in the +suburbs are many most beautiful and desirable homes.”17 +One such home, designed by architect George Ropes for local +attorney John E. Hessin at 519 N. 11th Street in 1886, is one of +the few remaining single-family dwellings located in the area +that would become Aggieville. Hessin opened a law practice +in the 1870s with George S. Green. Hessin later served as +County Attorney and City Attorney, as well as a member of +the Board of Regents of KSAC from 1890 to 1896, when he +was elected to the state senate. He made an unsuccessful bid +for Governor in 1898.18 By 1899, his son John Clarke joined +the firm, named Hessin & Hessin.19 By 1890 Manhattan had +installed electricity and improved sidewalks as residential +neighborhoods continued to expand housing for the city’s +3,000 residents.20 + +By 1890 enrollment at KSAC had risen to almost 600.21 +Faculty began to establish homes near the college. Among + +13 In 1896 students felt they were not well represented in the + +Industrialist so they formed their own paper the Student Herald. The +Herald was replaced by the Kansas Aggie in 1913 and finally the +Kansas State Collegian, one year later. +14 Daniel R. Walter. Aggieville 1889-1989: 100 years of the Aggieville +tradition. Manhattan, KS: Custom Family Stories, 2001. 9. +15 Manhattan Nationalist. 9 Feb 1877. +16 Olson, 185. Olson is quoting An Illustrated Sketchbook of Riley +County, Kansas: ‘The Blue Ribbon County.’ Manhattan, KS: The +Manhattan Nationalist, 1881. 32. +17 William G. Cutler. History of the State of Kansas. Chicago: A.T. +Andreas Publishing Co, 1883. +18 Manhattan Mercury. 27 Apr 1898. +19 Manhattan Nationalist. 16 Jun 1929. 10, 41. +20 Olson. 188. +21 James E. Sherow. Manhattan. Charleston, SC: Arcadia Publishing, +2013. 41. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +the first houses built by faculty were Professor George +Failyer’s who built at 11th and Moro Streets, followed by +Professor H.M. Cottrell on the opposite corner. Dean Julius +Terrass Willard built on the south side of Moro St. in the +1890s followed by Engineering Professor O.P. Hood at the +intersection of N. Manhattan Ave. and Moro St. Secretary +Graham then moved in next door to Hood. W.H. Sanders +built the largest residence at N. Manhattan Avenue and +Kearney Street two blocks to the north, east of campus.22 +The college was surrounded by vast open land that was prone +to flooding and was often referred to as the swamp. +Students traversed this area daily with a wood walkway +connecting the college and the town where most roomed in +boarding houses and private homes. The original road from +campus to town was unpaved and impassable much of the +year. For those who could afford it, the Manhattan Transfer +Company provided horse-drawn coach services between +downtown and the campus. The trip took thirty minutes and +cost five cents.23 + +View of Kansas State Agricultural College (Yearbook titled College +Symposium of the Kansas State Agricultural College for the year +1891). + +24 + +An Emerging College Center +Seeds of Aggieville were planted with events that occurred in +the 1890s. First, KSAC began competing in intercollegiate +sports with teams in football (1893), baseball (1894) and +basketball (1901). Sporting events in the twentieth century +would help to shape the character of Aggieville. Initially, +however, two seminal events began on campus - the +construction of a Domestic Science Building housing a +basement kitchen completed in time for the 1897-98 winter +term. Meals were offered to students and faculty at the rate +of twenty-one meals a week for $1.75.25 The second change + +22 “From Swamp to City in 20 Years.” Morning Chronicle. 16 Dec +1923, Aggieville Section. 1. +23 “Manhattan Transfer Company Timecard.” The Student’s Herald. +21 Sep 1899. 8. +24 Olson. 190-191. +25 Julius Terrass Willard. History of Kansas State College of +Agricultural and Applied Science. Manhattan, KS: Kansas State +College Press, 1940. 115. + +6 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +occurred on July 1, 1898, when a bookstore was opened on +campus. The store was heralded by students but local +businesses were not happy with the competition. When the +college presidency changed hands a year later, the bookstore +and dining hall were closed. + +College Barber Shop. Regardless of whether it began as a +laundry and barber shop in 1889 or with the Student Co-op +and a barber shop a decade later, Aggieville soon would +blossom. As the sun set on the nineteenth century, a +commercial center outside downtown began to take shape. + +In September 1899, students took the matter into their own +hands forming a cooperative just off campus that offered +textbooks and school supplies. Stock sold for $2 a share +which entitled the holder to a five percent discount on +books.26 Soon the Student Co-op added food providing +morning and evening meals with a lunch on campus.27 The +die was cast; the area near campus provided commercial +opportunities with a captive student audience to enterprising +Manhattan residents. + +A few businesses claim to be the first commercial enterprise +in what would become Aggieville. It is likely that some +nearby residents offered meals, laundry and rooms in the +1880s. One source claims that as early as 1889 a student +named Hansen operated a laundry service in a small two- +room frame building constructed by George W. Evans on Lot +390 in Ward Four on the south side of the alley directly north +of Moro Street.28 According to the source, Hansen soon +added a barber and retail goods, selling pencils, books and +other student necessities; meanwhile Hansen and Evans lived +in the back room. The first advertisement found for Hansen’s +Laundry is in January 1899.29 The first advertisement found +for a barbershop in the Student’s Herald was in November +1897.30 James Lee’s College Barber Shop offered deals to +students and acted as an agent for the Manhattan Steam +Laundry located downtown. Lee remained for only a year and +was not connected with the firm of the same name that +began on campus in 1919 with Joseph Cooper as proprietor. +Later John Harrison’s College Grocery and Meat Market +(1902) and W.P Barber’s Cleaning and Dye Works (1906) both +located on Moro Street, would each claim to be the first +Aggieville business.31 Barber’s cleaning concern began in a +one-room shop with a bicycle for deliveries, but the business +clearly post-dates the Student Cooperative and James Lee’s + +26 Daniel R. Walter. Aggieville 100 Years. +27 Walter, 100 Years, and Blake Gumprecht. The American College +Town. Boston, MA: University of Massachusetts Press, 2013, 112. +28 “Dr. J. W. Evans Remembers Manhattan of Early Days.” +Manhattan Chronicle. 2 Aug 1945. 5. +29 Student’s Herald. 5 Jan 1899. 2. +30 Student’s Herald. 24 Nov 1898. 4. The other Manhattan +newspapers, particularly the Manhattan Industrialist, which catered +to student news, were thoroughly searched. Similar results were +reported in Blake Gumprecht, “Campus Corners and Aggievilles: The +Distinctive Nature of College Town Commercial Districts.” +Department of Geography, University of New Hampshire, n.d. 4. +31 “The Barber Cleaners Old Aggieville Firm.” Manhattan Mercury. 5 +Oct 1929. Riley County Historical Society & Museum (RCHM) History +of Aggieville Vertical File. + +With a population of over 3,400 the City of Manhattan seized +the initiative of the new century by modernizing its +infrastructure with the construction of a new electric light +plant (1901), the first artificial ice plant (1903), and a +Carnegie Library and county courthouse (1904). Street signs +and numbering on houses and businesses were installed in +order to deliver mail.32 The first flagstone sidewalk started in +downtown and went to the east entrance of the college.33 +The center of activity in Manhattan was Kansas State +Agricultural College. The town had begun to operate on the +rhythm of the college.34 It was in this era that several +businesses opened near campus and before the end of the +decade the area would be known as Aggieville. + +The 1905 Sanborn Map illustrates the earliest development of +Aggieville along Moro Street and near the intersection of N. +Manhattan and Bluemont Avenues. Dwellings in the area +were used as boarding houses, while others took in laundry or +offered meals to students. The Midland Home, operated by +C.L. Evans at 1104 Moro St., was in business by 1905. +Evans sold to M.C. Knight in 1909. Four other boarding +houses (1110 Moro St., 710 and 922 N. Manhattan Ave. and +1201 Bluemont Ave.) were extant on the 1905 Sanborn +Map.35 Scattered among residences were two restaurants, +and three groceries, Hanson’s Laundry, and J.R. Sommer’s +College Barber Shop that operated on the second floor of N.S. +Cloud’s grocery concern named “The New Store.” The +Student Cooperative Bookstore and dining hall was first +located on N. Manhattan Ave. and moved to1214 Moro St. +into a structure built by Mr. Correll in 1903. Five years later +the Co-op would move to a new building and the College +Tailor Shop replaced the Co-op at 1214 Moro Street.36 + +The Student Cooperative was described as a “God-send to the +students who were forced to go through college with a +limited amount of capital.”37 It was a successful endeavor + +32 Historic Preservation Services, LLC. Cultural Resources Survey +Wards 1 and 2 Manhattan, Kansas, Manhattan, KS: City of +Manhattan. 2004. 40. +33 “Dr. J.W. Evans Remembers Manhattan of Early Days.” Morning +Chronicle. 29 Jul 1945. +34 Olson. 189. +35 Kansas City Public Library. Sanborn Fire Insurance Maps for +Manhattan, Riley County, Kansas. 1905, Sheet 4. Accessed online at +www.kclibrary.org/. +36 Student’s Herald. 4 Nov 1908. 5. +37 Gladys Jackson. “Aggieville Past.” Dimensions. 29 Apr 1983. +Jackson is quoting the 1913 industrial version of the Daily Mercury. +RCHM History of Aggieville Vertical File. + +7 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Another is that the name simply evolved from the early KSAC +mascot, the “Aggie.” Clearly the name was extant by 1912 +when G.E. Robinson opened the Aggieville Laundry. The KSAC +mascot changed from Aggies to Wildcats in 1917, but +Aggieville was here to stay. + +and in 1908, the Student Co-op built a new brick building at +the corner of N. Manhattan Avenue and Moro Street- the first +permanent commercial building in Aggieville. The same year, +the Y.M.C.A. building at the corner of Fremont and 11th +Streets was dedicated after more than a year under +construction. Two other commercial businesses established a +foothold near campus in 1908; each would become stalwart +Aggieville merchants. A devastating flood provided +opportunity for Aggieville. In late May 1908, the Big Blue +topped its banks flooding parts of downtown; the Manhattan +Republic noted that “boating was very fine along Second +Street all day.”38 O. William Holt, an enterprising grocer +moved his stock from his downtown location to an empty +building at 1202 Moro Street. After receiving an +overwhelming response he decided to stay.39 One year later +his grocery concern was bought out by L.C. Shafer. In +October 1889, Joseph “Guy” Varney had opened a store on +Poyntz Ave. with his uncle John W. Swingle. Varney soon +bought out his uncle and took over the Fox Bookstore +downtown. In 1908 Varney opened a branch location in +Aggieville.40 Another business central to students was the +Fink Electrical Company at 1126 Moro Street. Fink was +known as “the heart of the student section of the city,” by +providing electric light for studying and was the first place to +sell phonographs and records.41 + +First known masonry business building in Aggieville built for +Student’s Cooperative Bookstore in 1908 (replaced by existing stone +building in 1941 (now Dusty Bookshelf). 1912 Royal Purple. + +It was around this time that Aggieville got its name; although +there are conflicting stories regarding its origin. One is that a +Chicago cookie salesman sent a letter to a local merchant and +not knowing his address simply wrote, “Aggieville, +Manhattan, Kansas.”42 The merchant was amused and placed +the card in his store window and the name caught on. + +38 Manhattan Republic. 9 Jun 1908. 1. +39 Manhattan Mercury. 3 Jun 1909. 7. +40 Manhattan Nationalist. 11 Oct 1889, 1, and Manhattan Mercury. +5 Oct 1929. 43. +41 Jackson. +42 “Aggieville History: A Walk Down Memory Lane.” Royal Purple. +1998. + +“Aggies” Basketball Season, 1916 Royal Purple. + +As the first decade of the century came to a close KSAC could +boast doubling of their facilities to approximately twenty +buildings with functions ranging from halls containing +classrooms, laboratories and a gymnasium to greenhouses, +barns and a plant museum.43 In addition, during +commencement week in 1909, a new streetcar line opened +improving access to the college. The following year KSAC had +2,400 students enrolled and a six-week summer school was +added.44 Aggieville was well positioned to serve students +year round. + +Aggieville Comes of Age +By 1910 city residents approved a $20,000 bond to construct +an interurban railway between Manhattan and Fort Riley + +43 Willard. History of KSAC. 157. +44 Ibid. 544. + +8 + + increasing trade with soldiers from the base.45 In 1913, the +Manhattan Commercial Club placed a full-page ad in local +newspapers outlining the city’s virtues as reasons for home +ownership in Manhattan.46 During this time billiard halls +became illegal as did skating rinks; both were considered +undignified. Blue laws prohibited sporting events and movies +on Sunday until a city referendum in 1934.47 + +In the early years tuition at the college was free leading many +Kansas residents to move to the city to allow their children to +get a college education. Rather than sending their children +alone, advertisements in local newspapers indicate that +families often moved with their children, likely to further +their own economic opportunities in Manhattan.48 +Manhattan’s 1911 City Directory indicates that approximately +one in five households had at least one student sharing the +last name of the head of household and many had three or +four attending KSAC or Manhattan’s business college.49 + +Although businesses were clearly established prior, it was the +1910s when Aggieville began to take its current form. With +the exception of the Student Co-op brick building (later +replaced); most early businesses were located in former +dwellings or frame commercial buildings. The Hessin House +and YMCA on N. 11th Street are the only existing buildings +that predate 1910. + +In 1910, Aggieville and other “west-side” merchants came +together to form the West End Commercial Club with the goal +of mutual improvement of members and the promotion of +good fellowship and high business ideals. The first business +agenda for the eleven charter members was to gain access to +fire protection and postal service; they achieved both. In +November 1910, the group elected J.F. Harrison, president; +Roy Wilson, vice president; L.C. Shafer, secretary and +treasurer; and J.B. Freeman, sergeant at arms.50 The Club +was renamed Aggieville Commercial Club by 1915. + +Local businessman Barney Youngcamp played an important +role in the early development of the business district. He sold +his Elk Barber Shop on Poyntz Avenue in January 1912 and + +45 Nancy Capace. Encyclopedia of Kansas. St. Clair Shores, +MI: Somerset Pub. Inc., 2000. 225. +46 The Morning Chronicle, 30 Aug 1913. +47 Capace. 225. +48 For example, when M.M. West desired to sell the home at 417 +Fremont Street the advertisement stated, “This property is very +desirable particularly for persons who intend sending their children +to Agricultural College.” Abilene Daily Reflector, 27 Oct 1904. 3. +House advertisements noted their proximity to the college. +Manhattan Nationalist, 30 Sep 1897. 6; and Manhattan Mercury, 31 +July 1891. +49 Manhattan City Directory. 1 May 1911. Accessed on 30 Mar 2020 +on Ancestry.com. +50 Manhattan Republic. 10 Nov 1910. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +purchased the Manhattan Bottling Works downtown a few +months later while partnering with J.C. Brannan to build a +new building on Moro Street east of the Co-op Bookstore. +The $12,000 two-story brick building at 1220-1224 Moro St. +would house the “best dance hall in this area of Kansas,” as +well as four business storefronts first occupied by Palace Drug +Store, Kittel’s Varsity Shop, W.H. Hout Tailors, and Askren +Optometrist and Jewelry Store.51 Over the years the building +would become an integral part of the history of numerous +Aggieville businesses. By 1924 the dance hall had become +apartments that remained into the early 1960s. + +Aggieville business advertisements, Morning Chronicle 6 June 1910. + +Others building in 1912 included G.E. Robinson and J.L. John. +Robinson built an-up-to-date laundry on the south side of +Moro Street which appears to be the first business to adopt +the Aggieville name.52 A laundry would remain in the building +at 1219 Moro St. for the next seventy years. John established +his confectionary and ice cream store at 1123 Moro Street, +expanding to a larger corner location at 1201 Moro Street in +1918. The growth of automobile ownership led the city to +pave major roads in the city, moving outward from + +51 Manhattan Mercury. 6 Mar 1912. 3. +52 Manhattan Republic. 11 Apr 1912. 1. + +9 + + downtown.53 In 1913, the first public garage was built in +Aggieville at 1125 Moro St. by Alvin Miller to house the E.B. +Millard Garage.54 In the same year, Mrs. A.S. Wilson had the +existing building at 1127 Moro St. built, occupied by Jennings +and King Plumbing and Heating and the Moore Brothers built +the existing brick building at 1130 Moro St. for Coons’ +Grocery. As classes began in the fall of 1913 Aggieville +businesses included: four grocery stores, two bookstores, a +candy kitchen and ice cream store, one clothing store, a +department store, jewelry store, racket store, furniture shop, +steam laundry, bakery, an electrical supply house, one shoe +shop and two tailors.55 + +A few years earlier, Aggieville businessmen had founded their +own fire department located in a small building in the alley +behind 1218 Moro Street and furnished it with a hose cart, +900 feet of hose, and ladders.56 Fourteen men volunteered +without pay. The fire equipment is seen on the 1912 Sanborn +Map.57 In July 1913, in recognition of the growth and +importance of Aggieville, Manhattan City Council agreed to +expand the fire boundary to include most of Aggieville.58 One +of the first significant fires to occur started in the middle of +the night of November 23, 1914, when a gasoline stove +caught fire at the Thomas Bakery at 1216 Moro Street. The +Aggieville and Manhattan fire departments responded and +were able to keep the fire from spreading to other buildings. +With the expansion of the city’s fire limits and the rapid +response time, the Aggieville fire department was disbanded +in 1917.59 + +Prior to WWI Aggieville expanded with new and bigger +businesses and permanent buildings constructed using the +most modern materials available. Brick and stone were +chosen for fire prevention; a primary reason that many of +those early buildings remain at the heart of Aggieville today. +L.C. Shafer had established his grocery in Aggieville in 1909 +and expanded the business to include a meat market. He +built the existing building at 1203 Moro St. in 1915 to house +his business (adding a second floor in 1923). By 1926, Shafer +sold the business to W.F. Long and got into the real estate +business.60 + +53 “Manhattan Recognized as a Model Town Due to Her Many +Modern Improvements.” Manhattan Mercury. 5 Oct 1929. +54 Manhattan Republic. 24 Jul 1913. 1. +55 “Aggieville, the Thriving College Suburb.” The Daily Mercury +Industrial Edition. 30 Aug 1913. +56 Charles R. Tannehill. An Early History of the Manhattan Fire +Department: The Volunteer Period. Manhattan, KS: Falling Wood +Press. 40. +57 Sanborn Fire Insurance Maps for Manhattan, Riley County, Kansas. +1912, Sheet 10. +58 Manhattan Republic. 24 Jun 1913. +59 Tannehill. 41. +60 Manhattan Mercury. 6 Aug 1926. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +J.L. John’s Confectionery built at 1201 Moro St. in 1918 and L.C. +Shafer’s Grocery at 1203 Moro St. was built in 1915 as a one-story +building (R) and expanded to two-story in 1923 (Dan Walter, +Aggieville Archives). + +Several early businesses built new buildings during this time. +John F. Harrison who had the biggest grocery store in +Aggieville “felt the need” to expand in early 1911 into an +61 +adjacent lot and put new fronts on his existing stores. +years later, Harrison built a two-story masonry building to +replace the frame structure. The new building named +Harrison Hall, would house his grocery and meat market on +the ground floor and a dance hall on the second floor. +Dances were sponsored by university clubs and civic groups. +The Avalon Ballroom, as it was called by the 1930s, would +host every kind of dance from sock hops and swing dances to +big bands. + +Four + +The Olson Brothers expanded their downtown shoe repair +business by adding a shoe repair department in Elmer Kittell’s +Varsity Shop located in the north room of the Student Co-op +building. The Olson’s kept two men in their 3 +Street shop +and added one for the Varsity location. Olson’s Shoe Repair +had expanded to 1214 Moro St. by the late 1930s; the fourth +and fifth generations continue to operate the business in +Aggieville, constructing a new building in 1983. + +62 + +rd + +Guy Varney had expanded his business to Aggieville in 1908 +and in 1916, built the existing building at 623 N. Manhattan +Avenue to house his College Bookstore.63 The following year +Varney sold his downtown location and the Aggieville store +would expand to meet the needs of the Aggieville and KSU +communities for the next one hundred years. The 1913 +Industrial Edition of the Manhattan Mercury noted that the +bookstore carried stationery, wallpaper, pictures, cut glass +and china, Eastman Kodak supplies and Spalding Sporting +Goods, all in addition to books. The new building was +designed for apartments above but was converted to offices +on the upper floor early on. When Guy Varney died in 1920 + +61 Manhattan Republic. 23 Feb 1911. +62 Manhattan Republic. 12 Jan 1911. +63 Manhattan Republic 17 Aug 1916. 1. + +10 + + his wife Grace ran the store for the next sixteen years until +her son Ted took over the business. Throughout the history of +the store, Varney’s was adept at fulfilling the customers’ +needs and wants with extra goods and services that included +offering public school books; a lending library of over 500 +fiction books and a picture framing service during the late +1920s and early 1930s; office furniture and equipment +necessary for a “modern office;” as well as everything a +college student desired.64 + +While single family homes were being replaced by business +structures on Moro Street and N. Manhattan Avenue, the +1300 block of Fremont Street south of the City Park saw +extensive residential development. The land belonged to +Albert Dickens, a graduate of KSAC (1893), who joined the +Horticulture Department faculty in 1899. After the college +established a Forestry Division in 1909, Dickens was named +State Forester, in addition to his other departmental duties.65 +Upon his death in 1930, the horticultural building was +renamed Dickens Hall in his honor.66 Beginning in 1911, lots +were sold in the Dickens Place Addition, many to college +faculty to build single family homes. Extant houses in the +1300 block of Fremont Street that were built during this time +include the following: + +1314 - The first house on the block was constructed in 1911 +at 1314 Fremont Street by Professor J.W. Searson. Searson +was a renowned English Professor at KSAC and author of +numerous Kansas Readers used for elementary school +instruction. The house had a number of residents and owners +over the next few decades before having another long-term +resident. Professor Homer E. Socolofsky, a KSAC +alumnus, taught history for forty-five years, living at 1314 +from at least 1960 until 1982. Even after it was converted to +apartments in 1965, the Socolofsky family remained.67 His +son Robert was Riley County Attorney in the early 1980s.68 + +1320 – The second home built along Fremont St. occurred at +1320 in 1911 by John L. Coons, Aggieville grocer, who sold the +property to V.L. Polson in 1912.69 The house changed +ownership frequently in its early history. Some of the owners +worked at the university including Dean of the Music +Department Ira Pratt, who bought the house in 1923 from +A.M Patterson.70 + +64 “A Book Store of Departments.” Manhattan Mercury. 5 Oct 1929. +43. +65 Willard, History of KSAC. 165. +66 Ibid. +67 Open Records for Kansas Appraisers (ORKA) accessed online at +https://www.kansasgis.org/orka/ +and Manhattan Mercury. 29 Aug 1982. 41. +68 Manhattan Mercury. 7 Dec 1980. 1. +69 Riley County. Tax Records for Manhattan, Kansas. Riley County +Historical Society & Museum. +70 Manhattan Republic. 29 Mar 1923. 1. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +1326 - R.J. Barnett, principal of the K-State Publicity +Department, and his wife owned the house at 1326 Fremont +Street in 1911. By 1915, Bernice Houser had bought the +property and ran a boarding house there until 1952 when her +health failed.71 The house was later converted to apartments. + +1328 – The house at 1328 Fremont Street was built in 1911 +by Andrey Abraham Potter. Potter joined the KSAC faculty in +1905 as Assistant Professor of Mechanical Engineering, +becoming the Dean of Engineering and Director of KSAC’s +Engineering Experiment Station in 1913.72 Potter left +Manhattan for Purdue University in 1920, after having +“developed cleaner coal for state institutions in Kansas with a +method that removed sulfur from coal in one of the country's +first clean-air programs.”73 After several short-term owners, +Basketball Coach Jack Gardner moved his family to 1328 by +January 1950, selling after Gardner resigned in the summer of +1953 to take a coaching job in Utah.74 In 1939, Gardner was +named KSU Basketball Coach in a move that excited +basketball fans and helped to spur the movement for the +building of a new field house.75 After the successful 1947- +1948 season when the team won the Big 7 Championship, the +team became known as “Jack Gardner’s Cats.”76 + +1318 - Frances Hertslet built the home at 1318 Fremont +Street in 1912. By 1913, Mary P. Van Zile owned the house +and lived there until at least 1920. Mrs. Van Zile, a graduate +of KSAC, returned to the school to become the head of the +Domestic Science Department in August 1908.77 By 1911, she +was named Dean of Women, a position she held until 1940.78 +Van Zile was known as a confidant and cupid, but also +determined a dress code for women on campus, established +female curfews and was well-like by male and female +students.79 In 1926, the first women’s dormitory was +christened Van Zile Hall in her honor. The house at 1318 +Fremont St. had multiple owners following Mrs. Van Zile until +it was purchased by Lud Fiser in 1948.80 Fiser, Manhattan +High School Football and Track Coach, was named interim +Head Coach of K-State Football in 1945. He remained as + +71 Manhattan: Aggieville Commercial District Survey. 2020. Accessed +online at https://khri.kansasgis.org/index.cfm and “Bernice Houser.” +Manhattan Mercury. 1 Aug 1955. 2. +72 Willard, History of KSAC. 139. +73 “Andrey Potter, 97; An Aide at Purdue.” The New York Times. 7 +Nov 1979. B7. +74 Manhattan Mercury. 10 June 1953. 10. +75 Manhattan Mercury. 20 Jul 1939. 1, and Manhattan Republic. 26 +Dec 1940. 5. +76 Manhattan Mercury. 3 Aug 1948. 1. +77 Manhattan Nationalist, 6 Aug 1908. 6. +78 Manhattan Mercury. 10 May 1911. 1. +79 Marien Makemson. “If It Wasn’t for George Evans…” Manhattan +Mercury. 17 Aug 1986. 92. +80 Manhattan Mercury. 15 Jun 1948. 7. + +11 + + assistant coach the following year as well as head Baseball +Coach until 1948. He then served as Director of the +Manhattan Chamber of Commerce for 28 years (1948-1975). +His son Dave grew up in the house and after college, made +Manhattan his home. Dave Fiser served the Manhattan +community in various roles including City Commissioner and +Mayor in the 1980s. The house remains a single-family +residence today. + +1310 - The house at 1310 Fremont was built in 1921 by Rev. +R.A. Edson at a cost of $6,000.81 Roger Allerheilgen, a long- +time barber in Aggieville lived here ca. 1970s and walked to +work each day. Allerheilgen worked for seventeen years at +two Aggieville barbers (Ray's and Campus Barber Shop) +before opening his own shop. He retired in 2013 from his +shop, Campus Hair Styling at 1206 Moro, having cut hair in +Aggieville for forty-eight years.82 + +509 N. Manhattan – Located just north of the corner of N. +Manhattan Avenue and Fremont Street, this parcel was one +of the last in the Dickens Place Addition to be built upon. +Dickens still owned the lot in 1920. By 1923 Sanborn Maps +illustrate the house had been built. Mrs. Grace C. Creighton +is listed at this address by 1921 and remained in 1940. The +house was converted to apartments at an unknown date. + +WWI and the Roaring ‘20s +The U.S. entry into war in April 1917 led to a few changes on +campus and Aggieville. At KSAC new hastily-constructed +frame buildings appeared on campus to house the newly- +formed Student Army Training Corps. These facilities included +barracks, a mess hall, and other necessities including +a barber shop.83 Following the war enrollment grew to 3,400; +a thirty percent increase over 1910.84 In 1919, Anderson Hall +was remodeled to include a basement “recreation center” +that in addition to providing students a place to gather in +small groups, was large enough for dances and social +events.85 Space in the recreation center was leased to Joseph +Cooper for a barber shop and to alumni Elmer F. Kittell, who +ran a canteen – a “lunch counter and soda fountain” for a +year before selling to KSAC senior N. S. Spangler.86 In August +1924, the business concerns were evicted from campus, +leading Spangler to build his own building at 1421 Anderson +Avenue across from campus, where he operated the College +Canteen until at least 1959.87 + +81 Manhattan Mercury. 30 Dec 1921. 4. +82 Manhattan Mercury. 25 Nov 1974. 2, and 07 Apr 1996. 23. +83 Willard, History of KSAC. 252. +84 “Enrollment Summary by Academic Year.” Office of the Registrar. +Kansas State University. Accessed online at https://www.k- +state.edu/registrar/statistics/enrbyyr.html. +85 Willard, History of KSAC. 258. +86 Kansas Industrialist. 28 Jun 1920. 1, and Manhattan Republic. 28 +Apr 1921. 1. +87 Willard, History of KSAC. 252 and Gumprecht. 119. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +The 1920s were a boom time for Manhattan and Aggieville. +Manhattan was serviced by four bus lines with twenty-five +buses daily. In 1922, the city grew faster than any other +urban area in Kansas to a population of 10,000, without +students.88 The previous year, the city had issued 150 +building permits for residences valued at $262,535.89 +Manhattanites enjoyed twenty-five miles of asphalt-paved +streets with fifty miles of paved sidewalks.90 Aggieville got its +first service station constructed by H.P. Wareham at 1231 +Moro Street in 1922, and leased to Long Oil until 1927, when +Long built a new superstation down the block.91 Another +firm, the Wilson/Gillman Brothers Garage was built at 1111- +1113 Moro Street in 1924. + +Looking east on Moro Street from N. Manhattan Avenue in 1923. +Trolley track and lines remained at the time and new street lights +were in place (Riley County Historical Museum). + +In May 1923, it was reported that $40,000 worth of +construction was underway in Aggieville.92 Streets were +widened to accommodate street car service and electric +lighting while students were away and businesses readied for +fall including W.C.A. Meseke who built a modern one story +brick building at 1121 Moro Street to house his new furniture +store.93 Walter’s Plumbing built a new building adjacent at +1119 Moro St. and L.C. Shafer added a second story to his +grocery building to accommodate four apartments at 1203 +Moro Street. + +New businesses provided all forms of necessities for modern +life. The S.S. Prentice Dry Goods Company opened in the +Wilson Building at 1127 Moro in 1924. Four years earlier, +George W. Given opened the Given Clothing Company in the +Youngcamp Building while construction was underway on his +new building at 1205 Moro. Given Clothing Co. was an +upscale men’s store featuring Ed V. Price & Co. clothing + +88 Manhattan Republic. 4 Oct 1923. 5. +89 Manhattan Mercury. 30 Dec 1921. 1. +90 Royal Purple. 1919. +91 Manhattan Republic. 25 May 1922. 2. +92 Manhattan Republic. 23 May 1923. +93 Manhattan Nationalist. 8 Feb 1923. + +12 + + “made to individual measure.”94 O. M. Hixson’s photographic +studio, named the Royal Studio, established in Manhattan in +1898, was located at 1101 Moro St. in the 1920s and early +1930s. The business occupied three subsequent locations on +Moro Street and was purchased by Laurence Blaker in 1939. + +Not all businesses in the district were focused on college +clientele. In 1924, W.H. Schellenberger announced the +opening of Aggieville Hardware and Electric Company. One of +the biggest hardware stores in the city created the motto, “If +you can’t get something at Aggieville Hardware, it can’t be +bought.”95 I.W. Johnsmeyer built the existing building at 610 +N. 12th Street at a cost of $4,000 to house the Gold Medal +Bakery relocating from a frame structure on Moro Street.96 +Later named Brumm’s Gold Medal Bakery, the business +remained into the 1950s. In August 1923, Aggieville got its +first bank at 1131 Moro Street. J.L. John, owner of the +Creamery and Confectionery at 1201 Moro Street, had a +building constructed for the College State Bank with a club +room for the I.O.O.F. on the second floor.97 + +Other new business buildings included W.P. Barber’s modern +cleaning and dyeing plant on N. 12th Street and Harry Miller’s +new building on N. Manhattan Avenue, both buildings +designed by H.B. Winter. Paul Dooley, a jeweler who moved +into the Miller Building, planned to focus on college trade. +Next door was barber Joe Cooper, who continued his focus +on the college clientele while Larsen’s Tailor and Cleaners, +hoped to capitalize on both students and local residents. +Abernathy Grocery, also in the Miller Building, kept a +complete stock to entice all customers. + +As 1923 came to a close, Aggieville was a thriving four-block +business district comprised of brilliantly lighted streets with +sixty-eight flourishing businesses. The Morning Chronicle +boasted Aggieville had grown from a swamp to a shining +city.98 The district remained nestled in a residential area +illustrated by the 1923 Sanborn Map, which shows four +dwellings remaining in the 1100 and 1200 blocks of Moro +Street. Those two business blocks were flanked by Laramie +Street and Bluemont Avenue, largely residential streets, +making it clear why businesses catered to local residents as +well as students. In September 1924, the Morning Chronicle +declared that Aggieville had grown from a “few straggling +little sickly businesses a half dozen years ago to a flourishing +business district with nearly every line of business enterprise + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +represented.”99 Downtown businesses including Stevenson’s +Clothing, Palace Drug, Duckwall’s and R.H. Brown Music Co., +seized the opportunity to broaden their market and opened +branch stores in Aggieville. Duckwall’s noted that in addition +to students, they expected to reach households in the +western residential district.100 One newspaper crowed, +“There are enough eating places in Aggieville to take care of +the entire population of Manhattan, comfortably.”101 + +Aggieville article and business advertisements, The Morning +Chronicle, September 9, 1924. + +By 1925, the city’s population had reached 10,112.102 +Manhattan experienced its greatest building spurt in history – +over $1M for the year including $200,000 for a new dormitory +and five new business buildings in Aggieville. Brick buildings +continued to replace frame structures as seen by the +new J.A. Cress building at 1218 Moro Street and Dr. +LaShelle’s building down the block at 1204. C. A. Swenson +built the existing building at 1206 Moro St. in 1925; initially +designed and constructed to house W.B. Roper’s Barber Shop +and Duckwall’s.103 + +One of the biggest signs of modernization was announced in +October 1925. Aggieville was getting a movie theater. Harry +Miller bought the land south of the College Bookstore and +College Drug Store and asked Manhattan architect H.B. +Winter to prepare plans for the $100,000 theater.104 The +Miller Theater opened May 1, 1926, with a special section in +the Morning Chronicle boasting the new theater was one of +five like it in the United States and would feature first run +movies, opening with “Paris at Midnight.”105 The theater had + +94 Royal Purple Ads 1920, 1923, & 1927. +95 Manhattan Chronicle.11 Dec 1924. 4. +96 Manhattan Republic. 22 Feb 1923. 5 and Manhattan Tribune. 22 +Feb 1923. 4. +97 Independent Order of Odd Fellows. Manhattan Republic. 1 Jan +1931. 1. +98 “From Swamp to City in 20 Years.” Morning Chronicle. 16 Dec +1923. Aggieville Section, 1. + +99 “Aggieville Definitely Dons Long Pants.” Morning Chronicle. 9 Sep +1924. 4. +100 Ibid. +101 Ibid. +102 Manhattan Republic. 1 Oct 1925. 7. +103 Manhattan Mercury. 24 Apr 1925. 1 & 28 Jul 1925. 1. +104 Manhattan Mercury. 2 Oct 1925. 1. +105 Morning Chronicle. 1 May 1926. + +13 + + 800 seats and a live orchestra that played before and after the +movie. KSAC football games were broadcast live, attracting +local sports fans. In 1927, Miller sold the theater to +W.R. and Ella C. Browning, who remodeled the interior +adding a balcony to seat 400 and raising the height of the loft +to better accommodate scenery.106 The name was changed +to the Varsity Theater in 1931 and to the Sosna Theater in +1935, when Sam Sosna bought the business. Sosna added a +marquee with electric neon sign.107 + +In 1926, H.P. Wareham built a one-story commercial building +at 1221-1227 Moro Street east of his service station. +Designed to house four businesses, initial occupants included +an Aggieville branch of Kenney and Petrich Drugs called +Aggieville Rexall, the Rogers and Bell Clothing Company, and +the Piggly Wiggly Grocery Store. The Walters Brothers built a +one-story building on N. 12th Street extending the bank +building south to the alley. Other new construction included +two buildings built by brothers Forrest and Harold Forrester. +The first at 1216 Moro Street housed a clothing and shoe +store and then in 1929, the second at 704 N. Manhattan +Avenue, housed their drug store.108 The Forresters owned +the Palace Drug Store then located at 1224 Moro Street in the +Youngcamp Building, a sister store to the Downtown Palace +Drug located in the Gillett Hotel Annex.109 In 1929, Palace +Drug moved to the new building. Called The Uptown Palace, +the grand opening was held January 7, 1930.110 The Morning +Chronicle called the Palace the “official rendezvous of both +faculty members and students.”111 + +New construction was not limited to private enterprise. In +1926, KSAC got a new library building, the first dormitory and +spent $40,000 on improvements repairing and repainting +several buildings and building new sidewalks.112 Two years +later, streetcars were replaced by buses and by 1931, all the +tracks were removed.113 Aggieville hosted KSAC events +beginning with the annual Ag Fair, which was held on the +first Saturday in May during the 1920s. Local businesses +sponsored the event including White House Grocery & +Market, A.V. Laundry, Fink Electric Co, the Elite Textile Shop, +and Palace Drug Store.114 Sadly, the fair would be a casualty + +106 RCHM History of Aggieville Vertical File, 23 Jun 1927 and +Aggieville Commercial District Survey. +107 Manhattan Mercury. 29 Jan 1936. 1. +108 Manhattan Republic. 4 Oct 1923. 5. +109 Manhattan Daily Mercury. 1913 Industrialist edition. RCHM +History of Aggieville Vertical File. +110 Daniel R. Walter and Ralph L. Miller. 2012. The drug stores of +Aggieville: ... and a few other tangents along the way. Manhattan, +KS: Able Printing Company. 2012. +111 Morning Chronicle. Aggieville Section, 3. +112 Manhattan Mercury & Daily News. 31 Aug 1926. +113 Willard. History of KSAC. 347. +114 Official Fair Book of the Third Annual Ag Fair. 5 May 1923. RCHM +History of Aggieville Vertical File. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +of the Depression, but would be replaced by other events +such as the Aggieville Jamboree in the 1950s. As the 1920s +drew to a close, Manhattan celebrated her Diamond Jubilee, +boasting of forty miles of paved roads and fifty miles of brick +or asphalt sidewalks although it was noted that the west side +of town lacked storm sewers requiring the city to undertake +mud removal on streets including Moro.115 + +From Financial Struggle to Post-War Stability +As the 1930s began, Manhattan residents had begun to +experience financial strain that would deepen in the decade +to come. Building costs soared to seventy-five percent above +what they had been in 1929, forcing most businesses to delay +new construction or remodeling.116 Residential construction, +including fraternity and sorority houses, continued. Eighty- +nine building permits were issued in 1930; only fifteen of +which were non-residential structures.117 As cars gained +importance, more space was needed to house and repair +automobiles. Wilson/Gillman Brothers Garage at 1111-1113 +Moro Street expanded to the adjacent lot at 1115-1117 Moro +St. adding a 50’ by 100’ garage to the service station. Nearby, +Floyd McKowan opened a filling station at the southwest +corner of 11th and Moro Streets one year later.118 The station +changed owners but would remain an Aggieville staple +through the 1970s. + +The first bank opened in Aggieville in 1923 but failed in the Great +Depression (Dan Walter, Aggieville Archives). + +The Depression hit Aggieville and affected students as it did +everyone across the United States. On January 1, 1931, the +seven-year old College State Bank went into receivership, +affecting local business loans and capital. The bank was + +115 “Manhattan Recognized as a Model Town Due to Her Many +Modern Improvements.” Manhattan Mercury. 5 Oct 1929. +116 Manhattan Republic. 1 Jan 1931. +117 Ibid. +118 Ibid. + +14 + + founded by L.E. Grigg, W.C. Grigg, F.E. Freudenberg, F.H. +Walters, B. Youngcamp, S. Wagaman and H.S. Ramey.119 +With the election of President Franklin D. Roosevelt came +new programs such as the National Youth Authority, which +employed 2,289 KSAC students and provided the college with +over $232,000 worth of funding.120 Other New Deal +programs funded art and building projects on campus. By +1939 the campus had grown to thirty-four buildings including +greenhouses and pavilions. But when war came in the next +decade all construction at the college, and most in Aggieville, +would cease. + +Recreational and social pursuits were important even during +the Depression. Team sports became popular and leagues +were formed. Aggieville’s baseball team, cheered on by +locals, won the city league in 1938. As the 1930s came to a +close the city of 11,000 inhabitants had nearly six million +dollars in annual retail sales, creating an optimistic outlook +for the future.121 The 1940s would bring many changes to +Aggieville, including a new chapter for one of the district’s +first businesses. The Student Co-op Bookstore that opened in +1899 was sold to long-time manager Ray Pollum. Pollum built +a new building (the existing stone facade) on the same corner +site and changed the name to Campus Bookstore.122 The +building housed a bookstore until 1970. Then in 1991, Dusty +Bookshelf, a used bookstore, would revive the tradition in the +same building. + +Many changes occurred on campus after the declaration of +war in December 1941. Most courses were abbreviated so +that students could finish their degrees in two or three years, +allowing them to support the war industry or join the +military. By 1943, there were 1,400 Reserve Officer Training +Corps Cadets and 500 Air Force trainees at KSAC.123 The +Home Economics Department achieved its highest enrollment +and the Van Zile Dormitory housed soldiers.124 + +Following the war, as veterans returned and took advantage +of the GI Bill, money was still tight and local veterans formed +a grocery co-op.125 Other GIs opened new businesses on +Moro Street including George Hampton’s City Typewriter Co. +at 1206, Hill’s Linoleum and Rug Co at 1123, McNeil Music +Store at 1112, Brownie’s Coffee Shop at 1108, Fran +Schneider’s School of Dance at 1203, Chef Café No. 2 at 1201 +and Marshall Stover’s Rent-a-Bike south of the theater on N. + +119 Ibid. +120 Willard. 1940. 332. +121 R.L. Polk & Co. Polk’s Manhattan (Riley County, KS) City Directory. +Kansas City, MO: R.L. Polk & Co., 1939. +122 In 1967 Pollum sold to William Kammer III, who changed the +name to Bill’s Campus Book Store. The name was changed back to +Campus Book Store in 1970 with new owners. +123 Manhattan Mercury. 17 Feb 1963. 34. +124 Ibid. +125 Ibid. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Manhattan Avenue.126 Other established businesses such as +the Royal Studio, moved into new, expanded, or remodeled +spaces. Owner Lawrence Blaker who bought the business in +1939 was an Army veteran and built the existing building at +1201 Moro St. in 1946. Jim and Tom Farrell bought the +Sinclair Station at 1215 Moro St., later building a new station +on the same site (1955). DeYoung’s Electrical Shop moved to +1127 Moro St.; Bottger’s IGA moved to 1223 Moro St.; and +Duckwall’s expanded to occupy the entire building at 617-619 +N. Manhattan Ave. where it remained until closing in 1969. +By 1946 at least ten veteran-owned firms had opened in +Aggieville.127 + +In 1946 George Scheu bought the bank building at 1131 Moro +Street undertaking a major remodeling for the National Cash +Register Co., a branch of their district office in Topeka.128 +Gillman’s Garage at 1111 Moro St. became Watson Transfer +Co. in 1946, a year later moving next door to 1115-1117 Moro +St., where it remained through 1965.129 Mayflower Transit +took Watson’s place in 1111 Moro Street.130 + +1947 view of N. Manhattan Avenue with Sosna Theater, Bottger’s +IGA, Duckwall’s and College Drug Store with Manhattan Transit Bus +on street (Dan Walter, Aggieville Archives). + +The mood of community was high as the 1950s began. The +summer of 1951 would bring a devastating flood to many +downtown Manhattan businesses. Aggieville retailers +immediately offered help including storing and “hosting” +downtown retailers’ goods for sale, as Backman-Ballard +Sporting Goods did for Stevenson’s Clothing.131 Friendly + +126 “More than 20 New Businesses in Aggieville.” Morning Chronicle +[likely source]. 5 Sep 1946. RCHM Aggieville History Vertical File. +127 Ibid. +128 Ibid. +129 Manhattan: Aggieville Commercial District Survey. 2020. +Accessed online at https://khri.kansasgis.org/index.cfm. +130 R.L. Polk & Co. Polk’s Manhattan (Riley County, KS) City Directory. +1965. +131 Manhattan Mercury. 20 Jul 1951. 5. + +15 + + competition took place between the downtown district and +Aggieville, often supporting benevolent organizations such as +the March of Dimes. In January 1956, Aggieville and +Downtown squared off to see who could raise the most +money in a Mile of Dimes event.132 + +New businesses continued to open in Aggieville and a few +new buildings were constructed. Miller’s Launderette, the +first “help-yourself” laundry in Aggieville, built a new building +at 1129 Moro.133 The building continued as a self-serve +laundry through the early 1980s. Aggie Barber Shop opened +at 613 N. 12th Street in 1956 and remains in Aggieville today. +In 1952 Phil and Billie Woodward who had Woodward’s +Department Store downtown, established Woody’s Men’s +Store in Aggieville. It later changed locations, but remained +in the district until 1988. + +The post-war boom seemed to offer unlimited opportunities +for Aggieville businesses. The population of Manhattan +almost doubled from 11,359 in 1940 to 19,056 in 1950.134 In +the following decades growth rates slowed from the post-war +high, but would continue to rise by almost twenty percent +between 1960 and 1970, then 18.4 percent from 1970 to +1980.135 Enrollment at KSAC would almost triple in the five +years following the war, from 2,206 in 1945 to 5,907 by 1950; +the number continued to increase every year until 1981.136 +But difficult times were coming. Although the Aggieville area +had been spared physical damage in the 1951 flood, +Manhattan residents looked to the suburbs where flooding +was less likely and cheap land was prevalent.137 As auto- +bound Manhattanites moved westward, developers followed +with new shopping options. Big competition would arrive +soon from campus and the suburbs. + +The Era of Change +In 1955 Manhattan had 556 businesses in the city of +Manhattan, eighty-seven of which were in Aggieville and 422 +were downtown. The other sixty-seven were in outlying +areas.138 In 1961, the West Loop Shopping Center opened, +anchored by a new Dillion’s Grocery Store. The area was +annexed by city in 1962.139 Heavy traffic, limited parking and + +132 Manhattan Mercury. 20 Jan 1956. 1. +133 Manhattan Mercury. 13 May 1947. 3 Aug 1947, & 13 Aug 1947. +134 U.S. Census Bureau, Decennial Census, 1900-2010. Online at +Kansas University Institute for Policy and Social Research, +http://www.ipsr.ku.edu/ksdata/ksah/population/2pop33.pdf +135 South Arterial Construction, Ft. Riley Blvd to US-24, Manhattan: +Environmental Impact Statement. United States: n.p., 1985. +136 “Enrollment Summary by Academic Year.” Office of the Registrar. +Kansas State University. Accessed online at https://www.k- +state.edu/registrar/statistics/enrbyyr.html +137 Spencer & Davis. 60. +138 The First One Hundred Years: A History of the City of Manhattan, +Kansas 1855-1955. +139 Manhattan Mercury. 3 May 1962. 1. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +DeYoung’s Radio Service moved to 1127 Moro St. from the Wareham +Building and Henry Miller built the existing building at 1129 Moro St. +in 1947 to house a launderette (Riley County Historical Museum). + +small storefronts were inherent in downtown and Aggieville, +while West Loop offered free parking, easy access, and large +stores with a liberal selection of goods. Within a decade, +West Loop would grow from a small group of stores to a +“town within a town.” +opened off East Highway 24, offering discount shopping for +everything from fashionable clothes to colorful household +142 +accessories, brand name appliances and sporting goods. + +Then in October 1971, Walmart + +141 + +140 + +In response, the City of Manhattan adopted a Land Use Plan +in 1968 committing to maintain downtown as the central +shopping core as opposed to allowing development of +suburban shopping malls. The plan would lead to +redevelopment and revitalization programs in downtown and +Aggieville. One of the first steps was the alleviation of traffic +congestion throughout the city. The plan called for the +widening of N. Manhattan Avenue linking Anderson and +Bluemont Avenues on the north edge of Aggieville. The +University granted an easement to the City for the creation of +a park comprised of a triangular parcel resulting from the new +road across the southeast corner of campus. The +easement prohibited use of the parcel for roadway or +buildings, requiring it be maintained as a park and that the +stone fence along the south and east borders of campus be +re-laid to reflect the new campus boundary. Triangle Park, a +green space with mature trees, remained largely undeveloped +at the time but would become the setting for Aggieville +and community events in the coming years. + +In further attempt to compete with new suburban centers +and chain retailers boasting free on-site parking, a number of +the existing public parking lots were created by the City of + +140 Spencer & Davis. 60. +141 Spencer & Davis. 60 and RCHM Vertical File “Shopping Centers- +West Loop.” +142 Manhattan Mercury. 25 Aug 1971 and 12 Oct 1971. 8. + +16 + + Manhattan by use of eminent domain in the 1960s and 70s. +The first lot located in the 1200 block of Anderson Avenue +was created in 1960 (now under construction for a hotel). +This lot included the mid-block walkway on the north side of +Moro Street by Olson’s Shoe Shop. Three additional lots +including the north side of the 1100 block of Moro St., the +south side of the 1200 block of Moro St., and the northwest +corner of Laramie Street and N. Manhattan Avenue (south of +theater) were acquired in 1978 and developed as public +parking lots.143 These lots vastly expanded the district’s +available parking. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +casual dining. While Aggieville had long been home to +numerous diners, cafés and lunch counters, students and local +residents alike were increasingly attracted to new options +in ethnic cuisine and convenience alternatives such as pizza +and burgers. Pizza Hut was founded in Kansas in 1958 in +Wichita after a local woman had read an article in the +Saturday Evening Post about the growing popularity of +pizza.146 Prior to WWII there had been less than 500 pizza +restaurants in the U.S.; by 1957 there were more than +20,000.147 Aggieville got its first pizzeria in the late 1950s - +Pizza Villa at 712 N. Manhattan Avenue. Then, Bernie Butler +opened Pizza Hut in the existing building at 1121 Moro Street +in 1960.148 The restaurant would celebrate its 30th, 40th, and +50th anniversaries as the oldest Pizza Hut still in its original +location. Other casual dining and fast food restaurants would +follow including House of Burgers, Rusty’s Drive-in, JD’s Pizza, +Raoul's Mexican Restaurant, Hibachi Hut, Piñata Mexican +Restaurant, and Hardees in the 1970s.149 Between the +opening of the Student Union, new dorms that offered meal +plans, and changing eating habits, the culinary landscape +forever changed in Aggieville. + +Advertisement for “Independent Dining Hall” in Aggieville, 1940 +Royal Purple. + +The Board of Regents announced in 1950 that it would begin +to raise funds for a Student Union on campus. A committee +was formed and an architect hired, but it would be two years +before ground was broken and four more before the Union +was open on March 8, 1956. For decades prior, Aggieville had +essentially functioned as the student union catering to +students’ every need and desire. In 1950, most students still +lived off campus and ate many of their meals in Aggieville +restaurants who offered weekly meal plans taking over from +boarding houses in previous decades. Other restaurants +catered to the lunch crowd including: AV Snack Shack, +Wildcat Grille, Mar Café, and Dolly’s K-Lunch, where waiting +lines sometimes were longer than lunch time breaks +allowed.144 + +The first known chain restaurant to open in Aggieville was +Dairy Queen who opened in 1949 in the former Yeager Dairy +building at 712 N. Manhattan Avenue.145 However, nationally +Americans were still as yet unaccustomed to the concept of + +143 City of Manhattan records, Community Development +Department. +144 Gumprecht. 119. +145 Manhattan Mercury. 1 May 1949. Dairy Inn replaced Dairy +Queen at this location by 1956. + +1968 view of Moro Street, looking west with pizza and hamburger +signs prominently displayed (KSU Morse Department of Special +Collections). + +Other social norms were related to the changing role of +women and expanding popularity of alcoholic beverages. +During the 1950s and early 1960s there were three taverns in +Aggieville. Their manner was genteel and generally, they +were patronized by men. The bars catered to fraternities and +sorority girls were forbidden to enter. Not only were women +not seen in bars; they were rarely seen in establishments + +146 + +Wichita Entrepreneur Marguerite Mollohan suggested to +brothers Dan and Frank Carney that they should open a pizza +restaurant in one of her rental buildings. Brenda Spencer and +Michelle Spencer. Grandview Terrace Apartments National Register +Nomination. Washington, DC: National Parks Service, 2016. +147 Richard Gehman. “Crazy about Pizza.” The Saturday Evening +Post. 30 Nov 1957. 54. +148 Polk’s Manhattan City Directory. 1966. +149 Polk’s Manhattan City Directories. 1970s. + +17 + + such as recreation clubs or pool halls.150 Women formerly had +to be in their rooms by ten p.m. and were forbidden to leave +town without permission but these restrictions changed +in the 1960s. By 1965 ROTC was no longer mandatory for +men and women no longer had a curfew.151 In 1969, a co-ed +dorm opened on campus.152 In Aggieville the greatest change +would come from an acceptance of college men and women +consuming alcohol. + +Opening in 1939, Slim’s Shamrock Tavern (later became Kite’s) was +among Aggieville’s first bars (1940 Royal Purple). + +The number of bars in Aggieville was steady from 1950 +through the late 1960s. The bar at 1201 Moro Street has +changed colorful owners numerous times since it opened as +the Shamrock Tavern, Aggieville’s first bar in 1939. Initially +operated by Slim & Marie Redeker, Shamrock Tavern was +bought by Keith “Kite” Thomas and in 1954 Shamrock’s +turned into Kite’s.153 Murphy’s Bar opened 712 N. 12th Street +1950 becoming Aggie Lounge, a cowboy western bar serving +exclusively Budweiser beer in 1963. The bar remains today, +owned by Brian Gieber, son of original owner Tony Gieber.154 +The recreational Hole-in-One Club formerly on Moro Street +had moved to 710 N. Manhattan Avenue and in 1959 became +The Open. A year later Keith Thomas bought it and it was +renamed The Dugout. Thomas remodeled the building in +1968 and changed the name to Mr. K’s.155 + +150 Gumprecht. 121. +151 Gumprecht. 125. +152 Kansas State University Housing and Dining Services Historical +Timeline. Accessed online at https://housing.k-state.edu/about- +us/history/index.html and Gumprecht. 125. +153 Ibid. +154 K-Stater Magazine. Fall 2016. 16. +155 RCHM Aggieville Vertical Files. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +By 1970, Me and Ed’s Tavern opened at 712 N. Manhattan +Avenue, later called Spanky’s Bar and then Rockin K’s in 1980. +Auntie Mae’s Parlor opened at 616 N. 12th Street in the 1970s. +The bar survived changes in drinking laws and student trends, +staying true to its speakeasy-style since 1974.156 +Reflecting the social change occurring across the country, +Aggieville’s role transitioned from a full service neighborhood +to a student-centric nighttime entertainment district. By +1978, the number of bars in Aggieville had increased to +thirteen.157 Many of the bars replaced long-time businesses. +Duckwall’s, College Cleaners, Dodd’s Furniture and Courson +Chiropractic all closed and were replaced with bars.158 + +The district would begin to feel the heat of the social issues +boiling up across the country. Political protests were held on +campus and would spill over into Aggieville. Acts of +vandalism such as breaking windows became more +prevalent.159 In December 1968, arsonists burned Nichols +Gymnasium on campus. Aggieville’s image was further +altered when the drug scene arrived. A few new businesses +opened that sold drug paraphernalia while traditional +businesses struggled to survive amid harassment and petty +crime. As the atmosphere became tense, families chose not +to shop in Aggieville and the neighborhood became almost +exclusively a student hangout. With the growing number of +bars came an increase in public drunkenness and violence. + +The counterculture era was redolent with highs and lows for +Aggieville businesses. Some firms chose to relocate such as +Reed and Elliott’s Jewelry Store in 1975, which had been in +Aggieville since 1938. Others like Doebele’s IGA (1976) simply +closed. In contrast, one of the success stories was when +Woody’s Ladies Shop joined Woody’s Men’s Shop in +Aggieville. Woodward's purchased Roth Rock’s Lady Shop +619 N. Manhattan in 1960. In March 1963, Woody's Ladies +Shop moved to 1225 Moro. The store would remain in +Aggieville, later under different ownership, until 2002. + +Aggieville businesses did not give up. They tried to cater to +locals as well as students through advertising and +promotions. In the early 1950s, a group of Aggieville +businesses labeled “Uptown Shopping Center in Aggieville” +offered free parking and extended hours on Thursday +evenings to broaden shopping opportunities for working +families. +In 1978, a St. Patrick's Day Parade and Road Race +was started to draw customers into Aggieville at a time when + +160 + +156 K-Stater Magazine. Fall 2016. 16. +157 Gumprecht. 127. +158 Manhattan: Aggieville Commercial District Survey. Accessed +online at https://khri.kansasgis.org/index.cfm and Gumprecht. +159 Manhattan Mercury. 5 Mar 1968. 2. +160 Two full-page advertisements were in one of the local Manhattan +papers (untitled and undated) and included Backman-Ballard +Sporting Goods, which was in business 1951-1954. + +18 + + KSU students were gone on spring break. The tradition still +holds today. In 2014, the event attracted 10,000 people with +over 1,000 runners participating.161 Some stalwart businesses +survived the upheaval. Aggieville bookstores remained +popular. The newsstand then bookstore at 1131 Moro St. +changed hands but remained in business through the mid- +1980s. Varney’s Bookstore changed its name from the +University Bookstore in the early 1970s when the university +opened its own bookstore. The business continued to grow +expanding to the rear in 1963 and adding floor space from +adjacent business buildings on the south in 1975, 1989 and +again in 1991. + +Where students lived was also changing. At the start of the +1950s, most students lived in boarding or rental houses or +fraternity or sorority houses. Only one permanent dormitory, +the 130-bed women’s Van Zile Hall, existed on the 1947 +Sanborn Map. In 1957, Jardine Hall opened, providing +married student housing and within a decade several campus +buildings were added for new dorms.162 In Aggieville, only a +handful of apartment houses were listed in the 1951 City +Directory, two of which were the on second-floor of the +Youngcamp Building and above the Shamrock Tavern at 1201- +1203 Moro Street. Three apartments listed on Laramie Street +were likely converted single-family homes; however +apartment listings south and east of campus were growing.163 + +1965 Aerial view of Aggieville prior to connection of Bluemont and +Anderson Avenues. The area south and west of the core are still +largely residential (KSU Morse Department of Special Collections). + +The first known purpose-built apartment building in +Aggieville, The Lamplighter Apartments, opened in December +1964.164 Other multi-unit apartment buildings followed in the +1970s including Villa II Apartments at 526 N. 14th Street; + +161 Dan Walter, 125 Years of Aggieville Tradition. Manhattan, KS: +Able Printing Company, 2014. 50. +162 Kansas State University Housing and Dining Services Historical +Timeline. Accessed online at https://housing.k-state.edu/about- +us/history/index.html +163 R.L. Polk & Co. Polk’s Manhattan City Directory. 1951. +164 Manhattan Mercury. 5 Nov 1964. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Kimberley Gold Key Apartments (1972) at 1200 Fremont +Street; Schumann Apartments (1971) at 1119 Laramie; and +The Villager (1975) at 1114 Fremont Street.165 + +Lamplight Apartments constructed in 1964 on N. Manhattan Avenue +(1979 Historic Survey Photo, RCHM). + +Modern Times: the 1980s and 1990s +The decades of the 1980s and 1990s brought further change +to the face and character of Aggieville largely through new +development following the 1978 adoption of the Manhattan's +new Land Use Plan that included an Aggieville Improvement +Plan. The plan assigned C-3 (commercial) zoning to a defined +area that spanned generally from Bluemont Avenue to +Laramie Street and 11th Street to N. Manhattan Avenue AND +identified a targeted area for future expansion to include the +south side of Laramie Street between 11th and 14th Streets +and west to 14th Street between Anderson Avenue and +Laramie Street.166 + +After development of several purpose-built apartment +buildings in the 1970s, the rezoning led to commercial +development in the areas south and west of Aggieville’s +traditional core. Many of the new buildings were +concentrated around Laramie Street including a mixed-use +building at 12th & Laramie Streets in 1980. The 25,000 square +foot commercial center called the Laramie Plaza Shopping +Center opened in 1985 with space for 15-20 businesses. Also +new individual business buildings included a billiard hall/bar +and Pizza Hut offices at the 12 +intersection. + +and Laramie Streets + +th + +New construction on the edge of the historic core included a +small commercial center on the east side of 11 Street in +th +1983 and Nautilus Towers on the northwest corner of 11 +and Moro Streets in 1984. The 1922 gas station at N. + +th + +165 Manhattan Mercury. 19 Apr 1971. 14; 11 Dec 1972. 8; and 26 Oct +1975. 38. +166 Manhattan Mercury. 16 Oct 1983. C1. + +19 + + Manhattan Avenue and Moro Street, long occupied by a +liquor store, was demolished and a new multi-business +building constructed in 1986. Toward the end of the decade, +development pushed west with construction of a new strip +center on Anderson Avenue west of Varney’s Bookstore, +anchored by Kwik Shop and Kinko’s. These commercial +centers provided an opportunity for small businesses to offer +goods and services to the growing student population that +topped 20,000 by 1990.167 + +1987 construction of commercial center at 14th Street and Anderson +Avenue (KSU Morse Department of Special Collections). + +Some long-term businesses closed during the 1980s including +Aggieville Hardware and Blaker Studio Royal. Additionally, +two devastating fires occurred in the district. The Sigma Chi +Fraternity in the 1200 block of Fremont St. was destroyed by +fire resulting in construction of the existing brick building in +1982. The Youngcamp Building at 1220-1224 Moro St. the +first two-story masonry building in Aggieville (built in 1914), +had a devastating fire in 1987 that resulted in loss of the roof +and second floor. The ground floor was salvaged and a new +roof installed, resulting in the existing one-story building. + +Two challenges that came from outside the district would +have bearing on Aggieville for years to come. The first came +when Congress passed the National Minimum Drinking Age +Act. When it took effect on July 1, 1987, many KSU students +could no longer be served alcohol, dramatically affecting the +local bar scene.168 In 1983, there were thirteen bars in +Aggieville, eight of which served the eighteen-year-olds; four +clubs and the Last Chance Saloon that did both. By mid- +August 1986, there were six private clubs, the Last Chance +Salon and five bars serving 3.2 beer.169 The change in +drinking age occurred on the heels of two embarrassing +events when out-of-control students crowded Aggieville +following K-State football victories over in-state rival KU +resulted in vandalism and property damage. Ultimately + +167 “Enrollment Summary by Academic Year.” Office of the Registrar. +Kansas State University. Accessed online at https://www.k- +state.edu/registrar/statistics/enrbyyr.html +168 “Where Have All the Kids Gone…Aggieville.” Manhattan Mercury. +17 Aug 1986. 93. +169 Ibid. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +student attitudes towards drinking and a new generation +raised amid anti-drug campaigns would reduce or at least +moderate the market for alcohol consumption. + +Simultaneous with the change in drinking age, Aggieville was +forced to call on their experiences competing with suburban +development of the 1960-1970s again with the opening of +Manhattan Town Center, a downtown mall, in 1987. + +As bars closed or turned into private clubs and customers had +numerous new shopping options, Aggieville businesses looked +for ways of marketing to customers, new and old. The +Aggieville Business Association, long-dedicated to promoting +business in the district, knew they needed a plan to help +combat these new challenges. The City-supported Aggieville +Redevelopment Plan was designed by local architect Ken +Ebert to enhance the district’s streetscape with new +decorative sidewalks and street lighting.170 The project +included improvements to Triangular Park, a prominent +physical and social component of the Aggieville district. The +park received picnic tables, benches, bike racks, and trash +receptacles, and expanded seating areas along the N. +Manhattan Avenue sidewalk to encourage regular use of the +green space. 171 The improvements maintained most mature +trees and green space while adding a small plaza in the center +of the park and a wood pergola with backdrop to +accommodate Aggieville and community events. In addition +to physical improvements, the Business Association amplified +their marketing efforts to present a cohesive image. They +continued long-established events like the St. Patrick’s Day +Parade and hosted new events including a 100th birthday +celebration in 1989 to attract residents and visitors. + +St. Patrick’s Day Parade (Dan Walter, 1990s). + +170 Marien Makemson. “If It Wasn’t for George Evans…” Manhattan +Mercury. 17 Aug 1986. 92 and Aggieville Seeks Aid in Upgrading. +Manhattan Mercury. 25 Nov 1985. RCHM History of Aggieville +Vertical File. +171 Walter, Dan. “How Triangle Park Came to Be.” Manhattan +Mercury, 7 Oct 1999. Kansas State University Morse Department of +Special Collections Aggieville Vertical File. + +20 + + When Aggieville celebrated its 100th birthday the Manhattan +Mercury quoted Aggieville Historian Dan Walter, remarking +on the wide variety of businesses that have called the district +home through the years: + +You could buy a car, bicycle, skateboard, surfboard or +boat…You could get your eyes checked, teeth fixed, hair +styled, or nails polished…You could pick up a hammer +and nails or get groceries for the week…You could buy a +pizza or a snack or a waterbed….no one can question that +many businesses have come and gone, but each one, in +its own special way, added to the richness of the +Aggieville tradition---a tradition of strength through +diversity.172 + +Special Aggieville Tabloid, October 1990 (KSU Morse Department of +Special Collections). + +Rather than traditional goods and services, the area moved +toward an eclectic mix of businesses that included Krystallos, +Accessories on the Park, Pathfinder, and The Palace, the +former drugstore turned gift boutique, which kept the booths +from the soda fountain, the ice box and original ceramic tile +floor.173 Espresso Royale, a specialty coffee shop opened at +616 N. 12th Street in 1987, Aggieville’s first in a new type of +drinking establishment. There was a concerted effort to +move from conventional retail shops and bars to a diversified + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +mix with a broader appeal.174 + +The new Aggieville was not without loss. On Valentine’s Day +1998, a fire broke out in the 1100 block of Moro destroying +three businesses as well as damaging others. Harrison Hall, +built 1915 by John F. Harrison to house his grocery business, +was destroyed by the fire. The former two-story brick +building housed a number of prominent businesses over the +years including Pines Cafeteria, the Avalon Ballroom, Roberts +Furniture and Brothers Tavern. At the time of the fire, the +building was occupied by Club Karrigan, Greek’s Pizzeria and +Adventure Travel. The two buildings flanking Harrison Hall, +housing the Futon Store and Speed-Wash were heavily +damaged. Both businesses replaced their storefronts within +the year. The lot where Harrison had stood would remain +vacant for the nearly a decade. Then in the summer of 1998 +more than seventy years of movies in Aggieville came to an +end when the Campus Theater closed.175 The theater had +been in operation since 1926 as the Miller, Varsity and Sosna +Theater. Another Varsity Theater opened in 1969 at 1215 +Moro and had closed in the mid-1980s, although briefly +showing independent films in the mid-1990s. + +The New Millennium and Beyond +The Aggieville Business Association remains strong. A new +tradition was established in 2003 with a New Year’s Eve ball +drop that drew 8,000 its first year. The Mayor’s Christmas +tree continues to be located in Triangle Park and the area’s +holiday season kicks off with an annual lighting ceremony. A +large housing development, Laramie Village, was +constructed on the south side of the 1200 block in 2004 and +New Hope Community Center was built at N. Manhattan Ave. +and Fremont St. in 2013. The east end of Moro St. was +transformed with new buildings including the decade-long +vacant site of Harrison Hall. Additionally, Cozy Inn +Hamburgers constructed a new building in 2009, a mixed-use +development was built on the southwest corner of 11th and +Moro Streets in 2011 and Popeye’s Restaurant was built on +Bluemont Ave. in 2016. + +A few Aggieville cornerstones have been lost in recent years. +Woody’s Ladies Shop closed in 2002 after forty-two years and +three owners.176 Long-time bar, Aggie Station, and Krystallos, +a popular specialty shop, were lost when the building at 710 +N. Manhattan burned in 2003. A new multi-use building was +constructed on the site housing Eighteen Sixty-Three, a new +club with living units above. The Hibachi Hut, established by +John & Margaret Anderson in 1959, closed in 2013. Ballard’s +Sporting Goods, opened in Aggieville in 1961 also closed its +doors in 2013. The business began as a partnership between + +172 Dan Walter. “Aggieville: Strength through Diversity.” Manhattan +Mercury. Special Aggieville 100th Anniversary section. 1989. +173 Karen Rainey. “Shopping Bag Appeal.” Dimensions. 29 Apr 1983. +RCHM History of Aggieville Vertical File. + +174 Becky Howard. “Economics Change in Aggieville.” Manhattan +Mercury. 17 Aug 1986. 93. +175 K-State Collegian, 22 Jul 1998. +176 Manhattan Mercury. 11 Aug 2002. + +21 + + Eddie Backman and Charlie Ballard in 1951, later returning to +Aggieville operated by Sonny Ballard. Pizza Hut, the longest +in its original location shuttered in 2015 and Varney’s +Bookstore, a cornerstone in Aggieville since 1908, closed in +2016. + +Two early Aggieville businesses remain: Haynes Beauty Salon +(formerly Campus Barber and Beauty Shops) and Olson’s +Shoe Repair. Also beating all the odds in the modern +marketplace is the used bookstore, The Dusty Bookshelf. In +business since 1986, the store overcame a massive fire in +2017, rebuilding in the same location and retaining the 1941 +stone facade, surviving as a mainstay in the district. + +In 2017, more than 4,000 local stakeholders created a new +plan for the future called the Aggieville Community Vision. +The plan identified several redevelopment and public +improvement opportunities in the district including a +recommendation to preserve the historic character of +Aggieville’s core. The district continues to foster new +commercial ventures. Property values in the area have +soared amidst new development that has included a luxury +hotel at N. Manhattan and Bluemont Avenues. A second +hotel is currently under construction on Bluemont Ave. and a +parking garage is planned south of the old theater on N. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Manhattan Avenue, both projects emerging from the +Community Vision. + +Considered the first shopping center in Kansas, Aggieville has +experienced monumental change since its advent more than +a century ago. One thing that has not changed in the district +is the uniqueness. Aggieville is not just a collection of +neighborhood stores, bars, and restaurants that happen to sit +near a college campus. It is a place where Kansas State +students, alumni, and the Manhattan community come to +embrace the district’s historic past and a prosperous vibrant +future. + +2017 View from east end of Moro Street (Dan Walter, Aggieville +Archives). + +22 + + IV. SURVEY RESULTS + +A. Dates of Construction +Surveyed properties reflect construction in all but one decade +from the 1880s to the 2010s. The oldest resource surveyed is +the Hessin House at 519 N. 11th Street built in 1886; the +newest is Popeye’s Restaurant at 1115 Bluemont Avenue in +2016. A map illustrating the dates of construction follows this +Section on page 24. + +Hessin House at 519 N. 11th Street built in 1886 (1936 Royal Purple). + +The limestone residence built in 1886 for local attorney John +E. Hessin is not only the oldest resource in the survey area, it +is the only existing building constructed before 1900. +Elaborate stone homes are located throughout Manhattan’s +historic Wards 1 & 2 but are rarer in Ward 4. In 1907 the +YMCA building was constructed at 1100 Fremont Street later +becoming Parkview then St. Mary’s Hospital (Listed on KHR in +2019). Both of these early structures later served as +fraternity houses. + +The first permanent masonry commercial structure built in +Aggieville was a brick building at N. Manhattan Ave. and +Moro St. for the Student’s Cooperative Bookstore in 1908 +however that building was replaced in 1941. No commercial +buildings remain from the first decade of the twentieth +century. + +Existing buildings reflect the growth and development of +Aggieville with a majority of historic resources built in the +1910s and 1920s as Aggieville transitioned into a viable +commercial center with permanent buildings built to house +businesses that provided all types of goods and services. + +Sixteen percent of the buildings surveyed were built in the +1910s including five of the frame houses in the 1300 block of +Fremont St., six commercial buildings on Moro St. and one on +N. Manhattan Ave. Commercial buildings included the Wilson +and Miller Buildings at 1125-1127 Moro St., Aggieville +Laundry at 1219 Moro St., and Varney’s College Bookstore at +623 N. Manhattan Ave. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Reflecting the area’s largest period of growth, thirty-one +percent of the surveyed resources were built in the 1920s. +This era saw development not only along Moro St. and N. +Manhattan Ave. but included existing buildings on N. 12th +Street - W.P. Barber’s Cleaners (712) and Gold Medal Bakery +(610). The Great Depression and WWII all but brought new +construction to a halt. Four existing buildings were +constructed 1930-1941 and three in the post-war period from +1946-1955. Among them were the new Co-op Bookstore +(now Dusty Bookshelf building) in 1941, Blaker Studio Royal in +1946, and Farrell’s new Sinclair Station in 1955. + +The 1960s also saw the addition of four existing resources, +each a prominent addition to Aggieville. The first purpose- +built apartment building – The Lamplighter, was constructed +in 1964 on N. Manhattan Avenue. Aggieville saw the +establishment of Triangle Park and the first permanent +commercial building fronting Bluemont Avenue with +Cinderella Dry Cleaners built in 1967 (now Coco Bolos). +Perhaps most representative of the changing times was +construction in 1968-69 of Handi-Corner Shopping Village, +Aggieville’s first shopping center in the 1100 block of Laramie +Street. + +Modern development continued into the 1970s with ten +existing buildings constructed including four additional +apartment buildings, two commercial buildings, a bank, and +Shop Quick Convenience Shop were all built fronting +Bluemont Avenue (replacing former dwellings). Nineteen +percent of the existing buildings were constructed in the final +twenty years of the twentieth century. Much of this +development followed adoption in 1983 of a new land-use +plan that expanded the Aggieville “district” south to Fremont +Street and west to 14th Street. Aggieville has continued to +grow, albeit more slowly than the 1980s-1990s, with nine +percent of the existing buildings built since 2000. Thirty-nine +percent of the surveyed resources were constructed after +1970 and thus less than fifty-years old, the basic threshold for +historic consideration. + +Triangle Park was established in 1966 when KSU granted an +easement to the City of Manhattan for use as a community park. +Manhattan Mercury, October 1990 (KSU Morse Department SC). + +23 + + MAP ILLUSTRATING DATES OF CONSTRUCTION + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +24 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Late 19th and Early 20th Century American Movements +Commercial Style +The prominent style of commercial buildings built in the early +decades of the twentieth century is classified as the +Commercial Style. The Commercial Style became popular +during the Progressive Era and is characterized by its +simplicity compared to its predecessors built in the Victorian +era. This distinction is usually seen in storefront materials +that included brick or tile bulkheads and masonry piers +instead of cast-iron columns, wood storefronts and ornate +metal cornices and window hoods. Commercial Style +buildings typically had brick facades with a stepped or shaped +parapet. Aggieville has lost two prominent two-story +buildings to fire and only five two-story buildings remain on +Moro St. or N. Manhattan Ave. One-story Commercial Style +buildings are the predominant style of historic commercial +buildings in the core area. Reflecting Aggieville’s primary +development into a commercial district in the early decades +of the twentieth century, thirty-nine of the surveyed +resources (42%) are classified under the Commercial Style, by +far the most common in the core commercial area. + +B. Architectural Styles - Properties in the survey area reflect a +variety of architectural styles although like many Kansas +buildings, architectural designs are often not classic examples +of a style but rather a Midwestern or local interpretation of +popular styles. Spencer Preservation assigned each surveyed +building an architectural style and where appropriate, a +commercial building form/classification. Richard Longstreth’s +The Buildings of Main Street, A Guide to Commercial +Architecture provides the industry standard for classification +of commercial building types based on form and was used to +guide these selections. National Register Bulletin 16A– How +to Complete the National Register Registration Form provides +guidance for identifying architectural style and building +function. The KHRI inventory form also provides a listing of +styles based generally on National Register typology to +identify architectural styles. Following is an overview of the +primary architectural styles found in the survey area. + +Late Victorian and Late 19th and 20th Century Revival Styles +The two oldest surveyed resources were both architect- +designed buildings representative of the periods in which +they were constructed. The Hessin House, designed by +architect George Ropes, was constructed in 1886 in the Late +Victorian style common among stone houses of this era. +South of the Hessin House on 11th Street, is the YMCA +building, designed by Holland and Squires and built in 1907. +The symmetrical facade with cut stone detailing and formal +central entrance are features typical of the Classical Revival +style. No commercial buildings in the survey area reflect the +Late Victorian or Revival Styles. + +These three one-story brick buildings located at 1123- 1127 Moro St. +are among the district’s earliest existing commercial buildings +built in 1912. They are classic representatives of the +Commercial Style with brick facades featuring corbelling +on the upper facade as the only ornament. + +This photo of the YMCA was published in the 1910 Royal Purple +Yearbook. The building later became Parkview then St. Mary’s +Hospital and served that function until a new hospital was built on +Sunset Avenue in the 1950s. The building became home to Delta +Sigma Phi fraternity in 1955 who built the existing Modern addition +on the west. The building continues to serve the fraternity today. + +The two-story brick buildings at 1201 -1203 Moro St. are among the +few two-story buildings in the core area. Both are classic +representatives of the Commercial Style built in +1918 and 1923 respectively. + +25 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Built in 1926, the Miller Building on N. Manhattan Ave. is also a +Commercial Style building with multiple storefronts designed to +house several separate businesses. + +Built in 1926, the Wareham Building in the 1200 block of Moro St. +exhibits the classic form of a Commercial Style building with brick +piers framing multiple storefront bays. The use of green clay tile at +the parapet illustrates Spanish/Mission Revival style influences. + +Minimal Commercial (Early to Mid-Twentieth Century) +This classification used in the KHRI database represents +simple buildings with plain facades one to three stories in +height. Similar to Commercial Style buildings, this sub-type is +characterized by lack of ornament, not even the brick +corbelling or shaped parapets seen in Commercial Style +resources. The brick facades can be asymmetrical in +composition and were often blond or tan brick versus red. +Storefronts and other masonry openings were typically +punched openings with embedded steel lintels (not visible) or +very plain brick lintels and concrete sills. + +Commercial buildings built in this style were often concrete +block with brick or stone veneer facades and had aluminum +storefronts. Masonry veneer typically surrounded the +storefront, door, and window openings often without defined +piers or base. Only two existing commercial facades reflect +this style. + +Miller Launderette built in 1938 at 1129 Moro St. reflects the +Minimal Commercial style with its blond brick facade, minimal +ornamentation, and punched storefront opening. + +Modern Movement +This general architectural classification reflects a rejection of +the classically-ordered compositions and detailed +ornamentation that distinguished the Late Victorian and +Revival styles that preceded it. The style began in the 1920s +with Art Deco and continued into the 1970s. The movement +looked to completely new forms that embraced +advancements in construction techniques and materials such +as aluminum and masonry cladding. + +Art Deco was defined by its use of stylized ornament that +incorporated geometric forms on simplified building forms +with smooth facades and linear accents. Later Modern styles +that coincided with the Great Depression saw a complete +rejection of all form of ornament seen in styles such as +International and Brutalism. Few of these early Modern +styles are currently represented in Aggieville but one building +stands out – The Palace Drug Store built in 1929. + +The Palace Drug Store was designed by Salina architect Charles +Shaver and built in 1929. It is classified as a Commercial Style +building but embodies Art Deco detailing with its sunburst transom +surround (repeated in glass of transom) and shaped parapet (1979 +Historic Survey, RCHM). + +26 + + Post WWII - Other Modern Movement +Many of the simplified designs built in the cautious period +following World War II fall under the Modern Movement +stylistic classification. Most of the post-WWII buildings +reflecting the Modern Movement incorporate simple clean +lines with plain parapets and no cornice. Many utilize a +change in materials or textures for articulation of the facade. +Window openings are commonly small individual windows or +bands of horizontal windows. Designs utilize banding and +canopies to create horizontal emphasis on otherwise +unadorned facades. + +Given the National Park Service’s basic age threshold of fifty +years for initial consideration for historic designation, the +Post-WWII-era Modern buildings are one of the largest +growing stylistic categories among preservation efforts +nationally. However, there are only three examples of this +style among the commercial buildings in Aggieville including +the Cinderella Dry Cleaners building, now Coco Bolos, the +Lamplighter Apartments and Blaker Studio. + +Other +Although the Modern Movement spanned into the 1970s, the +Postmodern/Neoeclectic Style emerged around that time and +continued into the twenty-first century. Many postmodern +designs returned to traditional building forms and Classical +detailing. As new materials and technology entered the +construction market, new products such as fiberglass, EIFS +(Exterior Insulated Finish System), and metal sheathing made +reproduction of traditional forms and detailing affordable. +However while contemporary buildings often borrow from +established styles, they rarely reflect the underlying ordering +and articulation of historic styles. Other historic forms such as +the mansard roof also reemerged in the 1970s as a popular +form in commercial buildings. Aggieville has four small +apartment buildings in this style. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Blaker Studio Royal at 1200 Moro St. was built in 1946. This concrete +block building features an asymmetrical facade of smooth ashlar +stone and light-colored brick. The design had glass block bays +on the east street facade, likely to provide indirect lighting in the +photo studio. The building retained its historic appearance in this +1979 photo but has since been remodeled, losing much of its mid- +century character (1979 Historic Survey, RCHM). + +Residential +While most of Aggieville’s purpose-built apartment buildings +are not yet fifty years old, the resources that were +constructed as single-family dwellings reflect earlier styles. +The residences along Fremont Street are all frame structures +dating to the 1910s-1920s. The exception is the duplex at +1124-1126 Fremont St., a one-story dwelling built in 1938. +The dwellings in the 1300 block are all two-story and reflect a +variety of stylistic influences including Craftsman/Bungalow, +Stick, National Folk, and Queen Anne. Some of these original +dwellings have been converted to apartments and most of +the dwellings have experienced physical modifications that +have compromised their historic integrity. + +Schuman Apartments at 1119 Laramie St. was built in 1971 and +retains its original form and detailing today. It is a typical example +of the Mansard style seen throughout the Midwest in the 1970s. + +ca. 1960s view of dwellings in 1300 block of Fremont St. from SW +corner of City Park (Manhattan Christian College Photo, Dan Walter). + +C. Commercial Building Types - In addition to architectural +style, historic commercial buildings are often classified by a +typology based on building form that was developed by +Richard Longstreth and detailed in his book The Buildings of +Main Street. The most common of these classifications are +the One-Part Commercial Block and Two-Part Commercial + +27 + + Block, referring to one- and two-story buildings with +traditional commercial storefronts and having distinct first +and second floors. These classifications are based on the +building form, the configuration of the parts of the building +facade versus the type of detailing and ornament that defines +architectural style. The building forms are identified for the +surveyed historic commercial resources and noted in the KHRI +survey record. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Register of Historic Places or Register of Kansas Historic +Places. Using this evaluation of integrity, combined with the +National Park Service’s basic fifty-year age threshold for +historic consideration, a preliminary determination of +eligibility was made for listing in federal and state historic +registers. The 1970 date of construction was used as the first +basic threshold for historic significance. + +There are a variety of reasons that a building could be +identified as a non-contributor to a historic district or not +eligible for listing including: built less than fifty years ago; +removal or infill of glass storefront on traditional historic +commercial building thereby changing the relationship of the +building to the street front and passersby; infill or downsizing +of upper windows on the front facade altering the +proportions of the historic facade; or installation of siding +covering street facades (obscuring historic fabric). + +These three buildings on the north side of the 1200 block of Moro St. +illustrate the common One- and Two-Part Commercial Block building +classifications based on building form (1955 Centennial Parade +Photo, RCHM). + +D. Architects – Manhattan has long benefitted in many ways +from having Kansas State University as a part of the +community and that influence is seen by faculty in the College +of Architecture designing buildings in the community. Dr +Patricia J. O’Brien wrote the book The Architects & Buildings +of Manhattan, Kansas (Riley County Historical Society, 2008), +which outlines numerous local architects with ties to the +university and prominent regional and state architects all +who designed buildings in Manhattan. Designers of existing +Aggieville buildings include Holland & Squires, George Ropes, +Charles Shaver, and local architect Henry B. Winter. +Contemporary firms include Ken Ebert and BBN Architects of +Manhattan. Known builders of existing Aggieville buildings +include William Bates, Mont Green, Chas (Dutch) Hooper, +Fred S. Hopper, Clarence Johnson, O.D. Milligan, the Moore +Brothers, and McCullough Construction. + +th + +and Moro Streets is a + +The J.L. John’s Building at the corner of 12 +good example of a traditional historic commercial building. The +building retains its historic shaped parapet and brick corbelling on +the parapet as well as paired windows on the upper facades. +The 1918 building was constructed with multiple storefronts at street +level and apartments on the upper floor. The storefronts have been +modified and replaced on numerous occasions, the latest introducing +operable vertical bi-fold and horizontal-light operable windows +which are not compatible with the building’s historic character. +However, the storefronts maintain historic proportions and fill the +original openings. With the intact upper facade, the building retains +sufficient integrity to convey its historic commercial design and +function. Similar windows on a one-story building may diminish the +historic character to a degree that the building is not eligible for +historic listing. + +E. Integrity/Eligibility for Listing +After analyzing the survey results based on dates of +construction and architectural styles, each surveyed property +was evaluated for its level of historic integrity – the degree to +which the building continues to portray its historic design and +function. Generally a property must retain the physical +features and original materials that define both why a +property is historically significant and when it gained +significance in order to be eligible for listing in the National + +Many of the commercial buildings have received new +storefronts and a few have received entire new facades. +When new facades reflect an overall design change such that +a 1920s building now reflects a 1990s design, the potential +historic status is based on the date of the replacement +facade. Replacement storefronts that do not maintain the +components and proportions of traditional historic +commercial buildings can compromise integrity to a level that +makes a building ineligible for historic listing. + +28 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +View of north side 1200 block Moro St. (1926 Royal Purple) + +Aluminum slipcover extant in 1979 Historic Survey photo ( RCHM). + +Two-story buildings can often tolerate storefront +modifications that might render a one-story building +ineligible for listing because an intact upper facade with +original windows can overshadow alterations such as +replacement storefronts still conveying a significant level of +historic integrity. + +Although seemingly cosmetic, two common alterations +among Aggieville buildings have altered or obscured the +historic character of individual historic commercial buildings. +The first is use of operable windows at the storefront, often +with horizontal lights. The number and orientation of glass +panes in a storefront is historically significant reflecting the +transition from float to plate glass and technological +developments including tempered and tinted glass. +Horizontal glass lights have no historic basis in commercial +storefronts and therefore negatively impacts the historic +integrity of storefronts. Secondly, it has long been common +practice for Aggieville businesses to expand into neighboring +buildings occupying two or more multiple storefronts. While +this practice alone does not affect historic integrity, the +elimination of doors on storefronts and prominent signs or +awnings that span multiple buildings, changes the scale of +individual buildings and diminish the degree to which these +buildings reflect their historic character. + +Photos on the left illustrate the progression throughout the +one hundred-year history of the Cress and Forrester Buildings +at 1218 A & B Moro Street + +2020 street view of three buildings (1210-1214 Moro St.) occupied by +a single business (with the center building shared by Campus +Barbershop). The business signs and awnings align with each +individual storefront/building yet the commonality conveys that all +of the buildings are occupied by one business. + +2013 view after slipcover was removed (Dan Walter). + +2020 view with awnings spanning four buildings (occupied by a +single business) and operable storefront windows installed, most +having small horizontal lights. + +29 + + If the non-contributing designation is based on age of the +building or reversible changes such as removal of siding that +is covering historic fabric, a building’s contributing status can +be re-evaluated by the Kansas State Historic Preservation +Office. The contributing/non-contributing designation applies +only within a potential historic district; it does not mean that a +historic district is currently in place. An official determination +of eligibility for individual listing on historic registers +requires assessment of a building’s interior as well +as exterior. Although interior investigation was not included +in the scope of this survey, a few of the buildings may be +eligible for individual listing in the Register of Kansas Historic +Places. None of the existing buildings appear to be eligible +for individual listing on the National Register of Historic +Places. + +The contributing status of surveyed properties is based on the +professional evaluation and opinion of Spencer Preservation. +These determinations were reviewed by the Kansas State +Historic Preservation Office as a part of the survey project. + +A total of ninety resources were surveyed and fifty-three +percent were deemed to potentially be contributors to a +historic district including those potentially eligible for +individual listing. The survey area included forty-one +contributing, thirty-seven non-contributing and the old +YMCA/St. Marys Hospital which is already listed on Register +of Kansas Historic Places. + +Built for Coons’ Grocery in 1912, the corner building (on left) is one of +the earliest extant buildings in Aggieville and retains a high degree of +integrity. The brick facades to the east are contemporary facades +and buildings constructed after the 1998 loss of Harrison Hall to a +fire. The new construction is generally compatible in scale and +materials with area historic commercial buildings. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +1964 View looking SW on N. Manhattan Ave. (Dan Walter, Aggieville +Archives). + +Current view looking SW on N. Manhattan Ave. with former Varney’s +buildings and theater. Prior to closing in 2016, Varneys had +expanded into three buildings on the south. The existing storefronts +and first-floor facades on the northern three buildings date to a 1989 +remodel that compromised the historic integrity of the buildings. +The building(s) between the theater and the bookstore were +originally two separate buildings occupied by College Drug/Miller +Pharmacy and Duckwall’s/Bottger’s IGA. Other than replacement of +the stone panel on the parapet, the remodeling of the theater for +Varney’s bookstore in 2013 retained the upper facade and theater +marquee – historic features that distinguish the original theater +building. Even though these buildings have lost some historic +integrity, the bookstore and theater would be contributors to a +historic district and are long-time businesses that hold an important +place in Aggieville’s history. + +Not Eligible/Non-contributing status means only that a +building is not yet 50 years old and/or it does not retain a +level of historic integrity to convey its original design and +construction or that former historic alterations that have +gained historic significance in their own right. This does not +mean that the buildings or the Aggieville area as a whole +possess no historic value, they just do not meet the +established criteria for listing in the state or national +registers. The Aggieville area and many of its buildings are +locally significant as an important part of the Manhattan +and Kansas State community and are an integral part of our +heritage that is worth preserving. + +30 + + MAP ILLUSTRATING ELIGIBILITY & CONTRIBUTING STATUS FOR HISTORIC LISTING + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +31 + + V. LIST OF SURVEYED PROPERTIES + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +# +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 +13 +14 +15 +16 +17 +18 +19 +20 +21 +22 +23 +24 +25 +26 +27 +28 +29 +30 +31 +32 +33 +34 +35 +36 +37 +38 +39 +40 +41 +42 +43 +44 +45 +46 +47 +48 +49 +50 +51 +52 +53 +54 + +KHRI Number +161-2902 +161-2903 +161-2904 +161-2905 +161-2906 +161-2907 +161-2908 +161-2909 +161-2910 +161-2911 +161-2912 +161-2913 +161-2914 +161-2915 +161-2916 +161-2917 +161-2918 +161-2919 +161-2920 +161-2921 +161-2922 +161-2923 +161-2924 +161-2925 +161-2926 +161-2927 +161-2928 +161-2929 +161-2930 +161-2931 +161-2932 +161-2933 +161-2934 +161-2935 +161-2936 +161-2937 +161-2938 +161-2939 +161-2940 +161-2941 +161-2942 +161-2943 +161-2944 +161-2945 +161-2946 +161-2947 +161-2948 +161-2949 +161-2950 +161-2951 +161-2952 +161-2953 +161-2954 +161-2955 + +1101 +1115 + +1127-31 + +Address +Bluemont +Bluemont +Bluemont + +714 N 12th +1219 +1227 + +Bluemont +Bluemont + +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro + +718-22 N Manhattan +716 N Manhattan +712 N Manhattan +708 N Manhattan +706 N Manhattan +704 N Manhattan +700 N Manhattan +1224 +1218 +1218 +1218 +1212 +1214 +1210 +1208 +1206 +1204 +1202 +1200 +1130 +1128 +1126 +1120 +1118 +1116 +702 N 11th +1231 +1127 +1219 +1215 +1209-13 +1207 +1205 +1203 +1201 +1131 +616 N 12th +1129 +1127 +1125 +1123 +1121 +1119 +1115 +1111 +1109 +1101 +1100-30 + +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Laramie + +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro + +Historic Name +Kansas State Bank +Popeye's Restaurant +Shop Quik (1973)/Strip Center +W.P. Barber Cleaners +Piñata Mexican Restaurant +Cinderella Dry Cleaners +Miller Building +Lutz Building +Paul Dooley Jewelers +Eighteen 63 (Non-Historic) +Porter's (Non-Historic) +Palace Drug +Student's Cooperative Bookstore +Youngcamp Building +Cress Building +Forrester Building +Royal Café +Olson's Shoe Repair Building +Olson's Shoe Repair +Smart Shop (1930s-70s) +Aggieville Grocery +C.A. Swenson Building +LaShelle Building +College Tailor/Royal Studio +Blaker Studio Royal +Moore Building +Backman Cleaners +Hill Building +Commercial Building +Fair Building +Hunam Building +Strip Center +McCullough Building +Wareham Building +Aggieville Laundry +Commercial Building +Farrell's Aggieville Station +Shafer Building +Given Clothing Co. +L.C. Shafer Grocery Building +J.L. John’s Creamery Building +College State Bank +Walters Bros. Building +Miller Launderette +Wilson Building +Alvin Miller Building +Original John’s Candy Shop +Meseke Building +Walter’s Plumbing Building +Gilman Garage/Service Station +Wilson Garage +Cozy Inn Hamburgers +Mixed Use Building +Handi Corner Shopping Village + +Status + +Eligible + +Eligible + +Eligible + +Date +1978 Not Eligible +2016 Not Eligible +1973 Not Eligible +1924 Not Eligible +1979 Not Eligible +1967 Contributing +1924 Contributing +1924 Not Eligible +1924 Not Eligible +2004 Not Eligible +1997 Not Eligible +1929 +1941 Contributing +1912 Not Eligible +1925 Contributing +1925 Contributing +1920 Not Eligible +1920 Not Eligible +1983 Not Eligible +1920 Not Eligible +1920 Not Eligible +1925 Contributing +1925 Contributing +1920 +1946 Not Eligible +1913 +1940 Contributing +1978 Not Eligible +2007 Not Eligible +1998 Not Eligible +2011 Not Eligible +1983 Not Eligible +1986 Not Eligible +1926 +1912 Not Eligible +1992 Not Eligible +1955 Not Eligible +1922 Not Eligible +1920 Not Eligible +1915 Contributing +1918 +1923 Contributing +1926 Contributing +1947 Contributing +1913 Not Eligible +1913 Contributing +1912 Contributing +1923 Contributing +1923 Contributing +1930 Not Eligible +1924 Contributing +2010 Not Eligible +2011 Not Eligible +1969 Contributing + +Eligible + +Eligible + +32 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +KHRI # +161-2956 + +Address + +606 N 12th + +Historic Name +Gold Medal Bakery + +Status +Date +1923 Contributing + +161-2957 +161-2958 +161-2959 +161-2960 +161-2961 +161-2962 +161-3490-00020 +161-2897 +161-2963 +161-2964 +161-2965 +161-2966 +161-2967 +161-2968 +161-2969 +161-2970 +161-2971 +161-2972 +161-2973 +161-2974 +161-2975 +161-2976 +161-2977 +161-2978 +161-2979 +161-2980 +161-2981 +161-2982 +161-2983 +161-2984 +161-2985 +161-2986 +161-2987 +161-2988 +161-2989 + +Laramie +Laramie + +606 N Manhattan +1215 +1201 +513 N 12th +1125 +1119 +519 N 11th +1100 +1114 + +Laramie +Laramie + +Fremont +Fremont +Fremont + +1124 -26 + +500-08 N 12th + +Fremont +Fremont + +1200 +1224 +623 N Manhattan +621 N Manhattan +617-19 N Manhattan +613 N Manhattan +531 N Manhattan +525 N Manhattan +519 N Manhattan +509 N Manhattan +1300 +1310 +1314 +1318 +1320 +1326 +1328 + +Fremont +Fremont +Fremont +Fremont +Fremont +Fremont +Fremont +Fremont + +1334-36 + +516 N 14th +526 N 14th +1315 +1328 +1317-37 +1399 + +Laramie +Laramie +Anderson +Anderson + +Hardees +Laramie Village Apartments +Mixed Use Building +Pizza Hut Offices +Laramie Plaza +Schumann Apartments +John Hessin House +St. Mary's Hospital +The Villager Apartments +Duplex +Five-Plex +Kimberly Gold Key Apartments +Sigma Chi Fraternity House +College Book Store +College Drug Store +Duckwall’s/Bottger's IGA +Miller Theater +Ken's/Gilly's Restaurant +Lamplighter Apartments +Westbrook Apartments +Grace C. Creighton House +New Hope Community Center +Rev. R.A Edson House +Prof. J. W. Searson House +Frances Hertslet House +V.L. Polson House +R.J. Barnett House +Dean A.A. Potter House +Duplex +Seneca Apartments +Villa II Apartments +Fast Eddy's Billiard Parlor +Burger King +Commercial Strip Center +Triangle Park + +Eligible + +Eligible +Eligible + +1973 Not Eligible +2004 Not Eligible +1980 Not Eligible +1984 Not Eligible +1984 Not Eligible +1971 Contributing +1886 +1907 +1975 Not Eligible +1938 Contributing +1999 Not Eligible +1972 Not Eligible +1982 Not Eligible +1916 Contributing +1923 Not Eligible +1928 Not Eligible +1926 +1971 Not Eligible +1964 +1982 Not Eligible +1920 Not Eligible +2013 Not Eligible +1921 Not Eligible +1912 Contributing +1912 Contributing +1912 Not Eligible +1911 Not Eligible +1911 Not Eligible +1998 Not Eligible +1985 Not Eligible +1971 Not Eligible +1983 Not Eligible +1988 Not Eligible +1987 Not Eligible +Yes +1966 + +Eligible + +# +56 + +57 +58 +59 +60 +61 +62 +63 +64 +65 +66 +67 +68 +69 +70 +71 +72 +73 +74 +75 +76 +77 +78 +79 +80 +81 +82 +83 +84 +85 +86 +87 +88 +89 +90 +91 + +90 resources surveyed +including YMCA/St. Mary’s +Hospital that is listed in +Kansas Register + +Contributing +Resources +38% Total + +34 + +KEY: +Eligible for listing on Register of Kansas Historic Places AND contributing to district +Contributing to significance of potential historic district +Not Eligible for individual listing on state register or contributing to district due to level of historic integrity + +33 + + VI. RECOMMENDATIONS +The primary purpose of the survey project was to document +Aggieville’s historic resources. The following +recommendations identify opportunities to protect and +capitalize on the area’s historic character balancing historic +preservation and future development. + +A. Designation of Local or Kansas Register Historic District +It is recommended that consideration be given to establishing +a state or local historic district in Aggieville’s historic core. +The proposed district is illustrated on the map at the end of +this section and a list of the properties in the proposed +district follows in Section VII. The distribution of contributing +properties in the survey area helped to define the proposed +boundary. The period of significance for the district could +span from the earliest contributing structure in 1912 to 1970, +picking up the 1960s developments including Triangle Park +and Handi-Corner Shopping Village, Aggieville’s first strip +center. However, if local support is lacking for the inclusion +of the modern resources, the district boundary could be +limited to the traditional core. + +The recommended boundary for the proposed historic district +encompasses forty-eight resources, twenty-six of +which were deemed to be eligible for individual listing in the +Register of Kansas Historic Places or contributors to a state +register historic district. The proposed boundary results in +the required majority of contributing buildings at fifty-four +percent based on the 2020 analysis. The contributing status +and proposed boundary is the professional opinion of +Spencer Preservation and has been reviewed with the Kansas +State Historical Society prior to publication. The boundary +for a designated historic district should meet municipal +objectives and local support. Consensus of a majority of +property owners is highly recommended prior to official +submission of a nomination. The Historic Preservation Fund +that helped to finance this survey project is also a source for +the development and submission of a district nomination to +the state register. + +Typically, the prime motivation for formation of a historic +district is to provide property owners access to financial +incentives that are available through the Kansas State +Historical Society for designated historic buildings. Buildings +that have been determined to be contributors to a historic +district, as well as buildings that are individually listed in the +state register, are eligible for the incentives which include the +Heritage Trust Fund Grant Program and Kansas 25 percent +Historic Tax Credit that can help finance required +maintenance and rehabilitation projects.177 + +177 No resources in the survey areas were deemed eligible for listing +on the National Register of Historic Places therefore, the Federal +Historic Tax Credit is not available as a financial incentive. + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +It is recommended that the City of Manhattan work with the +Aggieville Business Association to determine the level of +interest and support among local building owners within the +proposed district. If access to state financial incentives is not +a primary motivation for Aggieville property owners, a locally- +designated historic district would also be a consideration. +Local districts can be used to provide recognition and +protection of historic resources but would not make +contributing resources eligible for KSHS financial incentives. +Most local districts include some form of regulation including +guidelines for demolition, infill construction and rehabilitation +of historic buildings. The City has already made +some modifications to Aggieville zoning to help preserve the +scale and building materials in the historic core. Additional +parameters and specific incentives can be customized to +meet local enthusiasm and acceptance. + +Manhattan has formerly made some zoning modifications regarding +building materials and scale; additional guidelines, parameters and +incentives could be a part of a district or overlay. + +B. Conservation Overlay District +If the City and Aggieville community do not support +designation of a state or local historic district, consideration +should be given to the establishment of a conservation +district that can be used to protect the historic character of +Aggieville without placing a higher level of regulatory review +that comes with designation of traditional historic districts. A +Conservation Overlay District (COD) is a planning strategy +designed to help ensure preservation of the historic and +culturally significant areas in a community while promoting +economic development. Conservation Overlay Districts are +typically established in areas with distinctive physical +characteristics that have preservation or conservation as a +primary goal. These areas might not merit designation as a +historic district but are priorities based on their economic + +34 + + potential, significance to the community and character.178 + +The City of Topeka uses Neighborhood Conservation Districts +as a means to protect neighborhood character by addressing +the appropriateness of design of new construction and the +compatibility of exterior improvements and additions.179 + +Preservation North Carolina defines Neighborhood +Conservation Overlay Districts (NCOD) as a zoning tool used +to preserve, revitalize, protect and enhance significant older +areas. Both conservation and historic districts are overlay +districts; however, a NCOD typically regulates fewer aspects +and focuses more on significant character-defining features +such as lot size, building height, setbacks and streetscapes. +Unlike historic districts, NCODs rarely consider specific +elements such as buildings materials, colors, and decorative +details.180 + +The National Trust has published a report on Neighborhood +Conservation Districts (NDC) available separate from this +survey report. The report summarized common elements of +successful conservation programs: +1) making sure NCDs offer a clear alternative to stricter forms of + +2) + +3) +4) + +historic preservation regulation; +developing guidelines with neighborhood participation and +acceptance; +a streamlined, flexible review process that is user-friendly; and +predictability about outcomes through consistent application +and enforcement.181 + +The conservation overlay may be an effective tool to preserve +and protect Aggieville’s unique characteristics through a +more flexible alternative than traditional historic districts. +While some conservation overlay districts include incentives +such as property tax abatements for substantial +rehabilitation, technical assistance and waivers of local taxes +and permitting fees, the conservation overlay does not make +properties within the district eligible for state preservation +incentives outlined above. + +C. Design Guidelines +It is recommended that consideration be given to establishing +a set of design guidelines for future rehabilitation and new + +(cid:120) + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +construction in Aggieville’s historic core. Guidelines are an +effective way to protect character of historic buildings and +manage infill/new construction. The Historic Preservation +Fund is a source of potential funding for the development of +design guidelines. + +Terracotta detailing on Forrester Building at 1218 Moro Street. + +D. Capitalize on Rich History and Alumni/Community Interest +Aggieville was likely a part of every K-State graduate’s +university life. The survey project recorded not only physical +information on the buildings, but also a history of businesses +that occupied the buildings. This part of the survey data +could be used as a fun trivia game targeting alumni. +Scavenger hunts for historic details on buildings focus the +community and visitors attention to Aggieville’s built +environment. Pursuing new marketing activities based on +architectural character of the historic buildings and on the +history of long-time businesses is a good way to promote +recognition and appreciation. + +Sample Trivia Questions: +(cid:120) What is the oldest existing structure in the Aggieville survey area? The +George Hessin House was built in 1886, occupied by Kappa Sigma +Fraternity in the 1920s-40s. + +(cid:120) Which Aggieville building was built by H.P. Wareham, the same year + +that he built the Wareham Hotel downtown? The Wareham Building at +1121 – 1127 Moro St. was built in 1926 with original occupants +including: Rogers & Bell Clothing, Aggieville Rexall Drug, Vermont +Luncheonette, and Piggly Wiggly Grocery Store. +Did you know that Reed & Elliot Jewelers started in Aggieville as Reed’s +Time Shop? What year did they move downtown? 1975. + +178 Rebecca Lubens and Julia Miller. (2002-2003). Protecting Older +Neighborhoods Through Conservation District Programs, +Preservation Law Reporter. 1001-1042. +179 City of Topeka website accessed online-20 May 2020 at +https://www.topeka.org/planning/neighborhood-conservation- +districts/ +180 Preservation North Carolina Website accessed online 17 May +2020 at https://www.presnc.org/neighborhood-conservation- +overlay-districts/ +181 Neighborhood Conservation Districts: Planning and +Administrative Practices. National Trust for Historic Preservation, +prepared for the City of Detroit December 2018, 12. + +(cid:120) What was Aggieville’s first strip center and when was it built? Handi +Corner Shopping Village built in 1968-69 in 1100 block of Laramie St. + +This image is from a 1963 advertisement is an sample of Aggieville’s +long-standing cooperative marketing (Manhattan Mercury, 17 +February 1963). + +35 + + MAP ILLUSTRATING PROPOSED HISTORIC DISTRICT + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +36 + + VII. LIST OF PROPERTIES IN POTENTIAL HISTORIC DISTRICT + +AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +# +6 +7 +8 +9 +10 +11 +12 +13 +14 +15 +16 +17 +18 +19 +20 +21 +22 +23 +24 +25 +26 +27 +33 +34 +35 +36 +37 +38 +39 +40 +41 +42 +43 +44 +45 +46 +47 +48 +49 +50 +51 +54 +56 +70 +71 +72 +73 +91 + +1227 +718-22 +716 +712 +708 +706 +704 +700 +1224 +1218 A +B +1218 +1218 +C +1212 +1214 +1210 +1208 +1206 +1204 +1202 +1200 +1130 +1128 +1231 +1127 +1219 +1215 +1209-13 +1207 +1205 +1203 +1201 +1131 +616 +1129 +1127 +1125 +1123 +1121 +1119 +1115 +1111 +1100-30 +606 +623 +621 +617-19 +613 +1399 + +Address +Bluemont + +N Manhattan +N Manhattan +N Manhattan +N Manhattan +N Manhattan +N Manhattan +N Manhattan +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +12th +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Moro +Laramie +12th +N +N Manhattan +N Manhattan +N Manhattan +N Manhattan + +N + +Anderson + +Historic Name +Cinderella Dry Cleaners + +Ave. +Ave. Miller Building +Ave. +Ave. +Ave. +Ave. +Ave. +Ave. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +St. +Ave. +Ave. +Ave. +Ave. Miller Theater +Triangle Park +Ave. + +Lutz Building +Paul Dooley Jewelers +Eighteen 63 (Non-Historic) +Porter's (Non-Historic) +Palace Drug +Student's Cooperative Bookstore +Youngcamp Building +Cress Building +Forrester Building +Royal Café +Olson's Shoe Repair Building +Olson's Shoe Repair (Non-Historic) +Smart Shop (1930s-70s) +Aggieville Grocery +C.A. Swenson Building +LaShelle Building +College Tailor/Royal Studio +Blaker Studio Royal +Moore Building +Backman Cleaners +McCullough Building (Non-Historic) +Wareham Building +Aggieville Laundry +Commercial Building (Non-Historic) +Farrell's Aggieville Station +Shafer Building +Given Clothing Co. +L.C. Shafer Grocery Building +J.L. John’s Creamery Building +College State Bank +Walters Bros. Building +Miller Launderette +Wilson Building +Alvin Miller Building +Original Johns Candy Shop +Meseke Building +Walter’s Plumbing Building +Gilman Garage/Service Station +Wilson Garage +Handi Corner Shopping Village +Gold Medal Bakery +College Book Store +College Drug Store +Duckwall’s/Bottger's IGA + +KEY: +Eligible for listing on Register of Kansas Historic Places AND contributing to potential historic district +Contributing to significance of potential historic district +Not Eligible for individual listing or contributing to district due to level of historic integrity +Not Eligible* Non-contributing status based on reversible changes and could be re-evaluated + +Present Function +CoCo Bolos +Haynes Salon, Rock-a-Belly Deli +Bombay Bites +Gyroville +Eighteen 63; 4 loft apartments +Porter's (Bar) +Varsity Donuts +Dusty Bookshelf +Chartier Building +Johnny Kaw Sports Bar +Johnny Kaw Sports Bar +Johnny Kaw Sports Bar +Jimmy John's +Olson's Shoe Repair/Sister of Sound +O'Malley's Alley Bar +O'Malley's Pub +Campus Hair/Jameson Irish Whiskey +Part of O'Malley's +Alm's Group RE +Tanners Bar & Grill +So Long Saloon/Taco Lucha +On the Wild Side +Buffalo Wild Wings/Radinas +Acme//Cold Stone +Bluestem Bistro +former Keltic Star +Johnny Kaw's Yard Bar +Orange Sky Yoga +Public Hall +Juicy's Vapor Lounge/Apts +Kite's/Aggie Barber/apts +Fat Shack/Lofts +Auntie Maes/Shaggeiville +Tubby's #2 +Tubby's +Aggieville Office +Former Goose Restaurant #1 +Former Goose #2 +Hi Lo/Al's Pizzeria #1 +Al's/Hi Lo in front/Tate’s in rear +Tate's on Moro/police station +Handi-Corner Strip Center +formerly Fuzzy's Tacos +Former Pie Five +Part of Rally House +Part of Rally House +Rally House +Triangle Park + +Status +Contributing +Contributing +Not Eligible +Not Eligible +Not Eligible +Not Eligible +Eligible +Contributing +Not Eligible +Contributing +Contributing +Not Eligible* +Not Eligible* +Not Eligible +Not Eligible* +Not Eligible* +Contributing +Contributing +Eligible +Not Eligible +Eligible +Contributing +Not Eligible +Eligible +Not Eligible +Not Eligible +Not Eligible +Not Eligible* +Not Eligible* +Contributing +Eligible +Contributing +Contributing +Contributing +Not Eligible +Contributing +Contributing +Contributing +Contributing +Not Eligible +Contributing +Contributing +Contributing +Contributing +Not Eligible +Not Eligible +Eligible +Eligible + +37 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +IX. BIBLIOGRAPHY +Aggieville Historic Survey Records and Photos, Riley County Historical Society & Museum. 1979. unpublished. +Capace, Nancy. Encyclopedia of Kansas. St. Clair Shores, MI: Somerset Pub. Inc., 2000. +City of Manhattan records, Community Development Department. +City of Topeka. Website. Accessed online 20 May 2020 at https://www.topeka.org/planning/neighborhood-conservation- + +districts/ + +Cutler, William G. History of the State of Kansas. Chicago: A.T. Andreas Publishing Co, 1883. +Fire Insurance Rates, Manhattan, Riley County, KS, 16 July 1926 with 1929 and 1930 updates/overlays. n.p. Riley County + +Historic Society and Museum (RCHM) Manhattan Directories Vertical File. +Fiser, David. Manhattan, KS, Telephone interview by Brenda Spencer. 13 Apr 2020. +Gumprecht, Blake. The American College Town. Boston, MA: University of Massachusetts Press, 2013. + +. “Campus Corners and Aggievilles: The Distinctive Nature of College Town Commercial Districts.” TD, Department of + +Geography, University of New Hampshire, n.d. + +Historic Preservation Services, LLC. Cultural Resources Survey Wards 1 and 2 Manhattan, Kansas, 2004. +Jackson, Gladys. “Aggieville Past.” Dimensions. 29 Apr 1983. Riley County Historic Society and Museum (RCHM) History of + +Aggieville Vertical File. + +Jones, Carolyn. 2014. The first one hundred years: a history of the city of Manhattan, Kansas, 1855-1955. Manhattan, KS: + +Manhattan Tribune-News, 1955. + +Kansas City Public Library. Sanborn Fire Insurance Maps for Manhattan, Riley County, Kansas. 1905, 1912, 1923, 1930, and + +1947 accessed online at www.kclibrary.org/ + +Kansas State University Housing and Dining Services Historical Timeline. Accessed online at https://housing.k- + +state.edu/about-us/history/index.html + +Kansas State University Libraries Digital Collections online at https://www.lib.k-state.edu/digital-collections +Kansas State University Morse Department of Special Collections. Aggieville Clipping Files and Photos. +Kansas State University Registrar’s Office. “Enrollment Summary by Academic Year.” Office of the Registrar. Kansas State + +University. Accessed online at https://www.k-state.edu/registrar/statistics/enrbyyr.html + +Longstreth, Richard. The Buildings of Main Street: a Guide to American Commercial Architecture. Walnut Creek, CA: AltaMira + +Press, 2000. + +Lubens, Rebecca and Julia Miller. “Protecting Older Neighborhoods through Conservation District Programs.” Preservation + +Law Reporter, Jan-Mar 2002-2003. 1001-1042. + +Makemson, Marien. “If It Wasn’t for George Evans…” Manhattan Mercury. 17 Aug 1986. 92. +Manhattan City and Telephone Directories. Riley County Historical Society & Museum and Manhattan Public Library. + +Manhattan City Directory, 1911. Transcribed list from RCHM and online at Ancestry.com. +Wareham Phone Directories 1914, 1920. +United Telephone Directory, Manhattan, 1927-1949. +Manhattan Phone Directories, 1921-1937, 1952. + +National Trust for Historic Preservation. “Neighborhood Conservation Districts: Planning and Administrative Practices. + +Prepared for the City of Detroit, December 2018. + +Newspapers accessed from Riley County Historical Society & Museum, Kansas State University Morse Department of Special + +Collections Clipping Files and on Newspapers.com. +The Daily Mercury Industrial Edition. +The Industrialist. +K-State Collegian. +The Manhattan Independent. +The Manhattan Mercury & Weekly Mercury. +The Manhattan Nationalist. +The Manhattan Republic. +The Manhattan Tribune. +The Morning Chronicle. +Student’s Herald. + +O’Brien, Patricia J. The Architects & Buildings of Manhattan Kansas. Manhattan, KS: Hawley Printing, 2008. +Olson, Kevin G.W. Frontier Manhattan: Yankee Settlement to Kansas Town, 1854-1894. Lawrence, KS: University Press of + +Kansas, 2013. + +38 + + AGGIEVILLE COMMERCIAL DISTRICT Historic Survey Report 2020 + +Open Records for Kansas Appraisers (ORKA) accessed online on multiple dates through the survey project at + +https://www.kansasgis.org/orka/ + +Polk, R.L. & Co. Polk’s Manhattan (Riley County, KS) City Directory. Kansas City, MO: R.L. Polk & Co. Multiple years accessed at + +Manhattan Public Library, Riley County Historical Society & Museum, and online at Ancestry.com. +Preservation North Carolina Website. Accessed online on 17 May 2020 at https://www.presnc.org/neighborhood- + +conservation-overlay-districts/ + +Riley County. Tax Records. Riley County Historical Society & Museum. +Riley County Historic Society and Museum. History of Aggieville Vertical File and Photographs. +Royal Purple (K-State Yearbook). Numerous years. Manhattan Public Library and online at + +https://archive.org/search.php?query=collection%3Akansasstateuniversitylibrariesyearbooks&sort= -date + +Sherow, James E. Manhattan. Charleston, SC: Arcadia Publishing, 2013. 41. +Shortridge, James. Cities on the Plains. Lawrence, KS: University of Kansas Press. 2004. +Spencer, Brenda R. and Christy Davis. Downtown Manhattan Historic District Multiple Property Submission. Washington, DC: + +National Parks Service, 2006. + +Tannehill, Charles R. An Early History of the Manhattan Fire Department: The Volunteer Period. Manhattan, KS: Falling Wood + +Press. 40. + +U.S. Census Bureau, Decennial Census, 1900-2010. Online at Kansas University Institute for Policy and Social Research, + +http://www.ipsr.ku.edu/ksdata/ksah/population/2pop33.pdf. + +Walter, Daniel R. 125 years of Aggieville tradition. Manhattan, KS: Able Printing Company. 2014. + +. Aggieville 1889-1989: 100 years of the Aggieville tradition. Manhattan, KS: Custom Family Stories. 2001. +. Aggievillearchives.com. +. Aggieville images: 125 years of Aggieville. Manhattan, KS: Able Printing Company. 2015. +. Aggieville through the years: 1880's to the year 2001! Manhattan, KS: Custom Family Stories. 2001. +. E-mail Correspondence with Brenda Spencer. Jan-Apr 2020. +. The Varney's Book Story: 1890-1974. Manhattan, KS: Able Printing Company. 2009. +. and Ralph L. Miller. 2012. The Drug Stores of Aggieville: ... and a Few Other Tangents along the Way. Manhattan, KS: + +Able Printing Company. 2012. + +Walters, J.D. The History of the Kansas State Agricultural College. Manhattan, KS: KSAC Printing Dept., 1909. Manhattan + +Kansas Public Library. + +Willard, Julius Terrass. “Bluemont Central College: The Forerunner of Kansas State College.” Kansas Historical Quarterly + +(KHQ). May 1945 (Vol. 13 No. 6). 323-357. Transcribed and digitized at https://www.kshs.org/p/bluemont-central- +college/12997 + +. History of Kansas State College of Agricultural and Applied Science. Manhattan, KS: Kansas State College Press, 1940. + +Manhattan Kansas Public Library. + +39 + diff --git a/q166/random_k2/question.json b/q166/random_k2/question.json index c5d6ffc5135c48917e653360219596a0fa8f39d4..575f773eb739a2a7968823850cd3d20bf08815bd 100644 --- a/q166/random_k2/question.json +++ b/q166/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "stereo_headset.pdf", - "web_c53505eb5c4f8264.pdf", - "web_fe44495665edf65a.pdf" + "5fdcf120b846746985684fa54250aab44639b30c.pdf", + "web_2dad2d6279d3265e.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q166/random_k2/random_1.md b/q166/random_k2/random_1.md index c3a1aac8ea3031fefb7fe76173d9b3ff3594440a..2e9f5241e58113455522088c7a6110de8838d130 100644 --- a/q166/random_k2/random_1.md +++ b/q166/random_k2/random_1.md @@ -1,2180 +1,1208 @@ -HALF YEAR REPORT -JANUARY – JUNE - -HALF YEAR REPORT 2021 -HALF YEAR -REPORT -TABLE OF CONTENTS -FINANCIAL HIGHLIGHTS (IFRS) ______________________________________________________________________________________________________________________ 1 -BUSINESS PERFORMANCE __________________________________________________________________________________________________________________________ 2 -ECONOMIC AND SECTOR DEVELOPMENT ____________________________________________________________________________________________________________________ 2 -INCOME STATEMENT _______________________________________________________________________________________________________________________________________ 2 -STATEMENT OF FINANCIAL POSITION AND STATEMENT OF CASH FLOWS ____________________________________________________________________________________ 4 -BUSINESS PERFORMANCE BY SEGMENT _____________________________________________________________________________________________________________ 6 -EMEA ______________________________________________________________________________________________________________________________________________________ 6 -NORTH AMERICA ___________________________________________________________________________________________________________________________________________ 7 -GREATER CHINA ___________________________________________________________________________________________________________________________________________ 7 -ASIA-PACIFIC ______________________________________________________________________________________________________________________________________________ 8 -LATIN AMERICA ____________________________________________________________________________________________________________________________________________ 8 -OUTLOOK ___________________________________________________________________________________________________________________________________________ 9 -GLOBAL ECONOMIC GROWTH TO RECOVER IN 2021 __________________________________________________________________________________________________________ 9 -SPORTING GOODS INDUSTRY TO RECOVER NOTABLY IN 2021 _________________________________________________________________________________________________ 9 -ADIDAS INCREASES OUTLOOK FOR FULL YEAR 2021 __________________________________________________________________________________________________________ 9 -RISKS AND OPPORTUNITIES _______________________________________________________________________________________________________________________________ 10 -CONSOLIDATED STATEMENT OF FINANCIAL POSITION_______________________________________________________________________________________________ 11 -CONDENSED CONSOLIDATED INCOME STATEMENT _________________________________________________________________________________________________ 13 -CONSOLIDATED STATEMENT OF COMPREHENSIVE INCOME _________________________________________________________________________________________ 14 -CONSOLIDATED STATEMENT OF CHANGES IN EQUITY ______________________________________________________________________________________________ 15 -CONSOLIDATED STATEMENT OF CASH FLOWS ______________________________________________________________________________________________________ 16 -EXPLANATORY NOTES TO THE CONDENSED INTERIM CONSOLIDATED FINANCIAL STATEMENTS (IFRS) AS AT JUNE 30, 2021 _________________________ 18 -RESPONSIBILITY STATEMENT ______________________________________________________________________________________________________________________ 32 -To enhance readability, registered trademarks as well as references to rounding differences are omitted in this publication. - - HALF YEAR REPORT 2021 - -| | | 1 | | | | 2 | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -FINANCIAL HIGHLIGHTS (IFRS) -| | | | | | | | First half year | First half year | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | ---------------- | ------------------ | --- | -| | | | | | | | | 2021 | 2020 Change | | -| | | | | | | | | | | | -Operating Highlights (€ in millions) -| Net sales | | | | | | | 10,345 | | 7,733 | 34% | -| ------------------------- | --- | --- | --- | --- | --- | --- | -------- | ------- | ------- | ---- | -| Gross profit | | | | | | | | 5,362 | 3,898 | 38% | -| | | | | | | | | | | | -| Other operating expenses | | | | | | | | 4,154 | 4,156 | 0% | -| | | | | | | | | | | | -| EBITDA | | | | | | | | 1,804 | 425 | 324% | -| | | | | | | | | | | | -| Operating profit / (loss) | | | | | | | | 1,248 | (215) | n.a. | -| | | | | | | | | | | | -Net income / (loss) from continuing operations 890 (217) n.a. -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable to shareholders 955 (264) n.a. -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -| | | | | | | | | | | | -Key Ratios -| Gross margin | | | | | | | 51.8% | 50.4% | | 1.4pp | -| ------------ | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | ----- | -Other operating expenses in % of net sales 40.2% 53.8% (13.6pp) -| | | | | | | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | ------- | -------- | -------- | ----- | -| Operating margin | | | | | | | 12.1% | (2.8%) | 14.8pp | | -| | | | | | | | | | | | -| Effective tax rate | | | | | | | 24.5% | 22.4% | | 2.1pp | -| | | | | | | | | | | | -Net income attributable to shareholders in % of net sales 9.2% (3.4%) 12.6pp -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Average operating working capital in % of net sales1 21.4% 23.2% (1.8pp) -| | | | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | --- | --- | ------- | -------- | ---------- | ----- | -| Equity ratio2 | | | | | | | 32.5% | 30.7% | | 1.8pp | -| | | | | | | | | | | | -| Adjusted net borrowings3/EBITDA4 | | | | | | | | 0.9 | 2.3 | (1.4) | -| | | | | | | | | | | | -| Financial leverage | | | | | | | 44.3% | 80.1% | (35.8pp) | | -| | | | | | | | | | | | -| Return on equity2 | | | | | | | 13.4% | (4.2%) | 17.7pp | | -| | | | | | | | | | | | -| | | | | | | | | | | | -Balance Sheet and C ash Flow Data (€ in millions) -| Total assets | | | | | | | 21,877 | 20,301 | | 8% | -| ------------------------------------ | --- | --- | --- | --- | --- | --- | -------- | -------- | ------- | ----- | -| Inventories | | | | | | | | 4,054 | 5,213 | (22%) | -| | | | | | | | | | | | -| Receivables and other current assets | | | | | | | | 4,257 | 3,694 | 15% | -| | | | | | | | | | | | -| Operating working capital | | | | | | | | 4,213 | 4,506 | (7%) | -| | | | | | | | | | | | -| Shareholders’ equity | | | | | | | | 7,106 | 6,230 | 14% | -| | | | | | | | | | | | -| Capital expenditure | | | | | | | | 183 | 217 | (16%) | -| | | | | | | | | | | | -Net cash generated from / (used in) operating activities 1,077 (819) n.a. -| | | | | | | | | | | | -| ------------------------------ | --- | --- | --- | --- | --- | --- | --- | ------ | -------- | ---- | -| | | | | | | | | | | | -| Per Share of Common Stock (€) | | | | | | | | | | | -| Basic earnings | | | | | | | | 4.52 | (0.97) | n.a. | -| Diluted earnings | | | | | | | | 4.52 | (0.97) | n.a. | -| | | | | | | | | | | | -Net cash generated from / (used in) operating activities 5.52 (4.19) n.a. -| | | | | | | | | | | | -| ---------------------------- | --- | --- | --- | --- | --- | --- | -------- | -------- | ---- | ---- | -| Dividend | | | | | | | | 3.00 | – | n.a. | -| | | | | | | | | | | | -| Share price at end of period | | | | | | | 313.90 | 233.60 | | 34% | -| | | | | | | | | | | | -| | | | | | | | | | | | -| Other (at end of period) | | | | | | | | | | | -| Number of employees | | | | | | | 58,007 | 61,445 | | (6%) | -Number of shares outstanding 195,066,060 195,032,889 0% -| | | | | | | | | | | | -| ---------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | ------------- | ---- | --- | -| Average number of shares | | | | | | | 195,066,060 | 195,266,358 | | 0% | -| | | | | | | | | | | | -| 1 Twelve-month trailing average. | | | | | | | | | | | -2 Based on shareholders' equity. -3 Adjusted net borrowings = short-term borrowings + long-term borrowings and future cash used in lease and pension liabilities – cash and cash equivalents and short-term financial assets. -4 EBITDA of last twelve months. -All income and expenses of the Reebok business are reported as discontinued operations as of the first quarter 2021. The prior year figures of the consolidated income statement and the -consolidated statement of cash flows have been restated to show the discontinued operations separately from continuing operations. - -1 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -BUSINESS PERFORMANCE -ECONOMIC AND SECTOR DEVELOPMENT -GLOBAL ECONOMY RECOVERS IN FIRST HALF OF 20211 -Around the globe, economies have been recovering from the enormous negative shock the coronavirus -pandemic had caused. The pace of this recovery, however, remains uneven. Developed economies have -been benefiting from substantial vaccination progress and fiscal support, while world trade continues to -be constrained by supply bottlenecks and travel restrictions. Economic activity is being dampened among -developing economies by a partial withdrawal of macroeconomic support and somewhat limited access to -vaccines amid an emergence of new coronavirus variants. Against this backdrop, consumer confidence -rebounded despite elevated unemployment rates and heightened policy uncertainty. -SPORTING GOODS INDUSTRY RETURNS TO GROWTH IN THE FIRST SIX MONTHS OF 2021 -In the first half of 2021, the global sporting goods industry recovered from negative impacts caused by -lockdown and quarantine measures as a result of the coronavirus pandemic. The return of major amateur -and professional sports events, as well as elevated health awareness, drove demand for sporting goods, -and consumers returned to retail stores in all parts of the world. The industry also benefited from excess -household spending, while coronavirus pandemic-related disruptions, supply chain challenges as well as -the geo-political situation weighed on the recovery of the sector. -INCOME STATEMENT -FOCUS ON CONTINUING OPERATIONS -Due to the planned divestiture of the Reebok business, all related income and expenses are reported as -discontinued operations at the end of June 2021. All figures related to the 2020 financial year in this report -refer to the company’s continuing operations unless otherwise stated. -ADIDAS WITH STRONG FINANCIAL PERFORMANCE IN THE FIRST HALF OF 2021 -In the first half of 2021, revenues increased 40% on a currency-neutral basis. In euro terms, revenues -grew 34% to € 10.345 billion (2020: € 7.733 billion). This growth was achieved against the backdrop of -prolonged lockdowns in many parts of the world, industry-wide supply chain challenges and the geo- -political situation. From a category perspective, growth was driven by exceptional increases in all strategic -performance categories amid the return of major sports events, while the lifestyle category recorded -excellent growth as well. From a regional perspective, on a currency-neutral basis, sales were driven by -exceptional increases in all strategic growth markets EMEA, North America and Greater China. Both Asia- -Pacific and Latin America grew at strong double-digit rates as well. -Gross margin improved 1.4 percentage points to 51.8% in the first six months of 2021 (2020: 50.4%). While -negative currency fluctuations as well as the less favorable channel mix weighed on the gross margin -development, this was more than offset by a significantly better pricing mix due to lower promotional -activity as well as the non-recurrence of last year’s inventory allowances and purchase order cancellation -costs. -1 Source: World Bank, Global Economic Prospects. -2 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -Royalty and commission income increased 29% to € 31 million (2020: € 24 million), while other operating -income declined 50% to € 10 million (2020: € 20 million). -| KEY FINANCIAL HIGHLIGHTS | | | | | | | -| ------------------------------------- | --- | --- | ------------ | ------------ | --------- | --- | -| | | | First half | First half | | | -| | | | | | | | -| | | | year 2021 | year 2020 | Change | | -| Operating Highlights (€ in millions) | | | | | | | -| Net sales1 | | | | 10,345 | 7,733 | 34% | - -| Operating profit/(loss)1 | | | | 1,248 | (215) | n.a. | -| ------------------------ | --- | --- | --- | ------- | ------- | ---- | -| | | | | | | | -Net income/(loss) from continuing operations1 890 (217) n.a. -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Net income/(loss) attributable to shareholders2 955 (264) n.a. -| | | | | | | | -| ------------- | --- | --- | --- | ------------- | ----- | ----- | -| Key Ratios | | | | | | | -| Gross margin1 | | | | 51.8% 50.4% | | 1.4pp | - -Other operating expenses in % of net sales1 40.2% 53.8% (13.6pp) -| | | | | | | | -| ----------------- | --- | --- | --- | -------------- | -------- | --- | -| Operating margin1 | | | | 12.1% (2.8%) | 14.8pp | | -| | | | | | | | -Per Share of Common Stock (€) -| | | | | | | | -| ----------------- | --- | --- | --- | ------ | -------- | ---- | -| Diluted earnings2 | | | | 4.89 | (1.35) | n.a. | -| | | | | | | | -| | | | | | | | -| | | | | | | | -1 Figures reflect continuing operations as a result of the planned divestiture of the Reebok business. - -| 2 Includes continuing and discontinued operations. | | | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | -Other operating expenses, including depreciation and amortization, consist of marketing and point-of-sale -as well as operating overhead expenses. In the first half of 2021, other operating expenses were flat at -€ 4.154 billion (2020: € 4.156 billion). As a percentage of sales, other operating expenses decreased -13.6 percentage points to 40.2% (2020: 53.8%). Marketing and point-of-sale expenses amounted to -€ 1.157 billion (2020: € 1.174 billion), a 1% decline compared to the prior year. The company significantly -increased its digital marketing investments to support its DTC storytelling and e-commerce business, -while physical activities remained restrained in many parts of the world during a significant portion of the -first six months of the year. As a percentage of sales, marketing and point-of sale expenses declined by 4.0 -percentage points to 11.2% in the first half of 2021 (2020: 15.2%). Operating overhead expenses remained -flat at € 2.997 billion (2020: € 2.983 billion). Efficiency measures and the non-recurrence of extraordinarily -high purchase order cancellation costs in the prior year period were offset by increased logistics costs -resulting from the significant e-commerce growth as well as temporary stranded costs related to the -intended divestiture of Reebok in the amount of around € 120 million. As a percentage of sales, operating -overhead expenses decreased 9.6 percentage points to 29.0% (2020: 38.6%). In the first six months of -2021, adidas recorded an operating profit of € 1.248 billion (2020: loss of € 215 million), resulting in an -operating margin of 12.1% (2020: -2.8%). -Financial income was down 67% to € 6 million (2020: € 17 million) while financial expenses decreased -9% to € 75 million (2020: € 82 million). Consequently, net financial expenses amounted to € 69 million -compared to € 65 million in the prior year’s period. The company recorded income taxes of € 289 million -resulting in a tax rate of 24.5%. During the six month period, the company incurred net income from -continuing operations of € 890 million (2020: loss of € 217 million), resulting in basic and diluted earnings -per share (EPS) from continuing operations of € 4.52 (2020: € -0.97). ► SEE FINANCIAL HIGHLIGHTS -In the first half of 2021, adidas’ gain from discontinued operations net of tax amounted to € 72 million -(2020: loss of € 74 million). As a result, the net income attributable to shareholders, which, in addition to -the net income from continuing operations, includes the gain from discontinued operations, came in at -€ 955 million (2020: loss of € 264 million). Consequently, both basic and diluted EPS from continuing and -discontinued operations were € 4.89 (2020: € -1.35). -3 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -The total number of shares outstanding remained stable in the first half of 2021. The average number of -shares used in the calculation of EPS was 195,066,060. ► SEE FINANCIAL HIGHLIGHTS -STATEMENT OF FINANCIAL POSITION AND STATEMENT OF CASH FLOWS -Changes in the statement of financial position are discussed in relation to the respective positions at the -end of June 2020. -PLANNED DIVESTITURE OF THE REEBOK BUSINESS IMPACTS BALANCE SHEET ITEMS -At June 30, 2021, all assets and liabilities of the Reebok business are presented as assets and liabilities -classified as held for sale due to a concrete plan to sell that business. At the end of the first half of 2021, -assets of € 1,674 million and liabilities of € 667 million were allocated to the Reebok business. However, a -restatement of the 2020 balance sheet items is not permitted under IFRS. -ASSETS -At the end of June 2021, total assets were up 8% to € 21.877 billion compared to the prior year (2020: -€ 20.301 billion), mainly driven by an increase in current assets. -Total current assets increased 29% to € 14.136 billion at the end of June 2021 (2020: € 10.931 billion). -Cash and cash equivalents were up 106% to € 4.151 billion (2020: € 2.018 billion), as net cash generated -from operating activities was only partly offset by net cash used for investing and financing activities. -Inventories decreased 22% to € 4.054 billion versus the prior year level of € 5.213 billion. This decline -reflects strong sell-through of the company’s products and successful inventory management across all -markets. On a currency-neutral basis, inventories also decreased 22%. Accounts receivable increased by -24% to € 2.324 billion (2020: € 1.869 billion), reflecting the company’s strong top-line growth. On a -currency-neutral basis, receivables were up 26%. Other current assets rose 1% to € 1.057 billion (2020: -€ 1.051 billion). Assets classified as held for sale increased to € 1.674 billion due to the planned -divestiture of the Reebok business. ► SEE NOTE 04 -Total non-current assets decreased 17% to € 7.741 billion at the end of June 2021 (2020: € 9.369 billion). -Fixed assets were down 19% to € 6.248 billion (2020: € 7.739 billion). Deferred tax assets were down 1% to -€ 1.154 billion (2020: € 1.166 billion). -STRUCTURE OF STATEMENT OF FINANCIAL POSITION1 (IN % OF TOTAL ASSETS) -| | | | June 30, 2021 | June 30, 2020 | | -| ----------------------- | --- | --- | ---------------- | ---------------- | --- | -| | | | | | | -| Assets (€ in millions) | | | 21,877 | 20,301 | | - -| Cash and cash equivalents | | | | 19.0 | 9.9 | -| ------------------------- | --- | --- | --- | ------ | --- | - -| Accounts receivable | | | | 10.6 | 9.2 | -| ------------------- | --- | --- | --- | ------ | ---- | -| | | | | | | -| Inventories | | | | 18.5 | 25.7 | -| | | | | | | -| Fixed assets | | | | 28.6 | 38.1 | -| | | | | | | -| Other assets | | | | 23.3 | 17.1 | -| | | | | | | - 1 For absolute figu res, see adidas AG Consolidated Statement of Financial Position. -LIABILITIES AND EQUITY -Total current liabilities remained virtually flat at € 9.161 billion at the end of June 2021 (2020: -€ 9.137 billion). Accounts payable were down 16% to € 2.165 billion (2020: € 2.575 billion), mainly due to -the normalization of payment terms. On a currency-neutral basis, accounts payable also decreased 16%. -Short-term borrowings declined 44% to € 682 million at the end of June 2021 (2020: € 1,217 million), -mainly reflecting lower borrowings compared to prior year, partly offset by the reclassification of a -4 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -eurobond (nominal value € 600 million) due to its maturity in October 2021. Liabilities classified as held for -sale increased to € 667 million due to the planned divestiture of the Reebok business. -Total non-current liabilities increased 14% to € 5.357 billion at the end of June 2021 (2020: € 4.693 billion). -Long-term borrowings increased 55% to € 2.473 billion compared to the prior year (2020: € 1.599 billion), -reflecting various bond placements in the second half of 2020. The non-current lease liability declined 7% -to € 2.206 billion (2020: € 2.374 billion) mainly due to the reclassification related to the planned divestiture -of the Reebok business. -The total number of shares outstanding remained fairly stable and slightly increased by 33,171 shares to -195,066,060 at the end of June 2021 compared to the prior year. ► SEE FINANCIAL HIGHLIGHTS -Shareholders’ equity increased 14% to € 7.106 billion at the end of June 2021 (2020: € 6.230 billion) mainly -due to the increase in net income. Consequently, the company’s equity ratio increased 1.8 percentage -points from 30.7% to 32.5%. -STRUCTURE OF STATEMENT OF FINANCIAL POSITION1 (IN % OF TOTAL LIABILITIES AND EQUITY) -| | | | June 30, 2021 | June 30, 2020 | | -| --------------------------------------- | --- | --- | ---------------- | ---------------- | --- | -| | | | | | | -| Liabilities and equity (€ in millions) | | | 21,877 | 20,301 | | - -| Short-term borrowings | | | | 3.1 | 6.0 | -| --------------------- | --- | --- | --- | ----- | --- | - -| Accounts payable | | | | 9.9 | 12.7 | -| -------------------- | --- | --- | --- | -------- | -------- | -| | | | | | | -| Long-term borrowings | | | | 11.3 | 7.9 | -| | | | | | | -| Other liabilities | | | | 42.0 | 41.6 | -| Total equity | | | | 33.6 | 31.9 | -| | | | | | | - 1 For absolute fig ures, see adidas AG Consolidated Statement of Financial Position. -OPERATING WORKING CAPITAL -Operating working capital decreased 7% to € 4.213 billion at the end of June 2021 (2020: € 4.506 billion). -On a currency-neutral basis, operating working capital was down 5%. Average operating working capital -as a percentage of sales decreased 1.8 percentage points to 21.4% (2020: 23.2%), reflecting the company’s -accelerating top-line momentum and focus on tight inventory management. ► SEE FINANCIAL HIGHLIGHTS -LIQUIDITY ANALYSIS -In the first half of 2021, net cash generated from operating activities was € 1.172 billion (2020 net cash -used: € 824 million). This development was driven by the operating profit generated during the period as -well as the decrease in operating working capital. Net cash used in investing activities rose to € 86 million -(2020 net cash generated: € 51 million). The majority of investing activities in the first half of 2021 related -to spending for property, plant and equipment, such as investments into controlled space and the further -development of the company’s headquarter infrastructure. Net cash used in financing activities totaled -€ 945 million (2020 net cash generated: € 601 million), mainly due to the dividend paid to shareholders. -Also as a result of these developments, cash and cash equivalents increased € 2.133 billion from € 2.018 -billion at the end of June 2020 to € 4.151 billion at the end of June 2021. Adjusted net borrowings at June -30, 2021, amounted to € 3.146 billion, representing a decline of € 1.842 billion compared to adjusted net -borrowings of € 4.988 billion at the end of June 2020. This development was mainly due to cash generated -from operating activities, partly offset by an increase in long-term borrowings. The company’s ratio of -adjusted net borrowings over EBITDA amounted to 0.9. ► SEE FINANCIAL HIGHLIGHTS -5 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -BUSINESS PERFORMANCE BY SEGMENT -At the beginning of 2021, we launched our new strategy ‘Own the Game’ for the period until 2025. As part -of this strategy, we are focusing our growth efforts on the three strategic markets Greater China, EMEA -and North America. To be able to execute this strategy successfully, adidas has changed its organizational -structure. Since January 1, 2021, adidas manages Greater China as a separate market. The remaining -Asia-Pacific (APAC) market now comprises Japan, South Korea, Southeast Asia and the Pacific region. The -change reflects the increasing importance of Greater China as a growth market for the company. In -addition, adidas created the EMEA (Europe, Middle East and Africa) market. To better leverage economies -of scale, the company has integrated the former markets Europe, Russia/CIS and Emerging Markets into -the newly formed EMEA market. The markets North America and Latin America remain unchanged. -Compared to the first half of 2020 when adidas’ top line was negatively impacted by the very large number -of temporary store closures as result of the coronavirus pandemic, sales in all segments continued to -recover and grew at a double-digit rate in both Performance and Lifestyle in the first half of 2021. In most -segments, lower inventory write-offs and a more favorable pricing mix due to less promotional activity had -a positive effect on the gross margin development. At the same time, a less favorable channel and -category mix, negative currency developments and higher sourcing costs negatively impacted the gross -margin development in most segments. Operating expenses as a percentage of sales declined in all -segments, resulting in higher operating margins across all segments. -EMEA -Sales in EMEA increased 41% on a currency-neutral basis. In euro terms, sales increased 38% to -€ 3.680 billion. -Gross margin in EMEA decreased 0.1 percentage points to 50.8%, as lower inventory write-offs and a more -favorable pricing mix due to less promotional activity were more than offset by a less favorable channel -and category mix, negative currency developments, and higher sourcing costs. Operating expenses grew -3% to € 1.082 billion, reflecting higher marketing expenditure. As a percentage of sales, operating -expenses were down 10.1 percentage points to 29.4%. The operating margin increased 10.0 percentage -points to 21.5%. Operating profit in EMEA increased 157% to € 792 million. -| EMEA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | -------------------- | ----------- | --- | -| Net sales | | | | | 3,680 | 2,670 | 38% | 41% | -| Gross profit | | | | | 1,870 | 1,360 | 38% | – | -| | | | | | | | | | -| Gross margin | | | | 50.8% | | 50.9% (0.1pp) | | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 792 | 308 | 157% | – | -| | | | | | | | | | -| Segmental operating margin | | | | 21.5% | | 11.5% 10.0pp | | – | -| | | | | | | | | | -| | | | | | | | | | - -6 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -NORTH AMERICA -Sales in North America increased 38% on a currency-neutral basis. In euro terms, sales rose 27% to -€ 2.406 billion. -Gross margin in North America increased 4.2 percentage points to 45.5%. Lower inventory write-offs and a -more favorable pricing mix due to less promotional activity more than offset a less favorable channel and -category mix, as well as higher sourcing costs. Operating expenses decreased 12% to € 661 million, -reflecting a decline in both marketing expenditure and operating overhead costs. Operating expenses as a -percentage of sales decreased 12.0 percentage points to 27.5%. The operating margin increased 16.2 -percentage points to 18.6%. Operating profit in North America rose 893% to € 447 million. -| NORTH AMERICA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 | Change neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | ------------ | ------------------- | --- | -| Net sales | | | | | 2,406 | 1,892 | 27% | 38% | -| Gross profit | | | | | 1,094 | 781 | 40% | – | -| | | | | | | | | | -| Gross margin | | | | 45.5% | | 41.3% | 4.2pp | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 447 | 45 | 893% | – | -| Segmental operating margin | | | | 18.6% | | 2.4% | 16.2pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -GREATER CHINA -Sales in Greater China increased 38% on a currency-neutral basis. In euro terms, sales rose 37% to -€ 2.405 billion. -Gross margin in Greater China was down 2.1 percentage points to 54.0%. Lower inventory write-offs were -more than offset by a less favorable channel, category and pricing mix, negative currency developments -and higher sourcing costs. Operating expenses were up 21% to € 542 million, reflecting an increase in -both marketing expenditure and operating overhead costs. Operating expenses as a percentage of sales -declined 2.9 percentage points to 22.5%. The operating margin increased 0.8 percentage points to 31.5%. -Operating profit in Greater China rose 40% to € 758 million. -| GREATER CHINA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 | Change neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | ------------ | ------------------- | --- | -| Net sales | | | | | 2,405 | 1,757 | 37% | 38% | -| Gross profit | | | | | 1,299 | 987 | 32% | – | -| | | | | | | | | | -| Gross margin | | | | 54.0% | | 56.2% | (2.1pp) | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 758 | 540 | 40% | – | -| | | | | | | | | | -| Segmental operating margin | | | | 31.5% | | 30.7% | 0.8pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -7 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ASIA-PACIFIC -Sales in Asia-Pacific increased 26% on a currency-neutral basis. In euro terms, sales rose 21% to -€ 1.135 billion. -Gross margin in Asia-Pacific declined 0.3 percentage points to 52.4%. Lower inventory write-offs were -more than offset by a less favorable channel, category and pricing mix, negative currency developments -and higher sourcing costs. Operating expenses were down 7% to € 317 million, reflecting a decline in both -marketing expenditure and operating overhead costs. As a percentage of sales, operating expenses -decreased 8.5 percentage points to 27.9%. The operating margin increased 8.1 percentage points to -25.1%. Operating profit in Asia-Pacific rose 78% to € 285 million. -| ASIA-PACIFIC AT A GLANCE € IN MILLIONS | | | | | | | | | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | -------------------- | ----------- | --- | -| Net sales | | | | | 1,135 | 939 | 21% | 26% | -| Gross profit | | | | | 595 | 494 | 20% | – | -| | | | | | | | | | -| Gross margin | | | | 52.4% | | 52.6% (0.3pp) | | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 285 | 160 | 78% | – | -| Segmental operating margin | | | | 25.1% | | 17.1% | 8.1pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -LATIN AMERICA -Sales in Latin America increased 81% on a currency-neutral basis. In euro terms, sales rose 59% to -€ 645 million. -Gross margin in Latin America increased 1.5 percentage points to 48.2%, as lower inventory write-offs and -a more favorable pricing mix due to less promotional activity more than offset a less favorable channel -mix, negative currency developments and higher sourcing costs. Operating expenses were down 3% to -€ 190 million, reflecting a decline in operating overhead costs. Operating expenses as a percentage of -sales were down 18.5 percentage points to 29.4%. The operating margin increased 19.1 percentage points -to 18.8%. Operating profit in Latin America rose to € 121 million. -| LATIN AMERICA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ----- | -------------------- | ----------- | --- | -| Net sales | | | | | 645 | 406 | 59% | 81% | -| Gross profit | | | | | 311 | 189 | 64% | – | -| | | | | | | | | | -| Gross margin | | | | 48.2% | | 46.7% | 1.5pp | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 121 | (1) | n.a. | – | -| | | | | | | | | | -| Segmental operating margin | | | | 18.8% | | (0.2%) 19.1pp | | – | -| | | | | | | | | | -| | | | | | | | | | -8 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -OUTLOOK2 -GLOBAL ECONOMIC GROWTH TO RECOVER IN 20213 -Global GDP is forecast to grow 5.6% in 2021, with the recovery projected to be more robust than initially -expected. Due to rapid vaccination progress in major economies, the recovery is expected to be unevenly -distributed and, in some cases, delayed. It largely reflects sharp rebounds in some major economies amid -highly unequal vaccine access, with many poorer countries facing subdued prospects. The strength of the -near-term global recovery is largely attributable to a few major economies, such as the United States and -China, with many emerging market and developing economies lagging behind. The United States and -China are each expected to contribute over one-quarter of global growth in 2021. -SPORTING GOODS INDUSTRY TO RECOVER NOTABLY IN 2021 -In the absence of any major delay in the pandemic recovery or other macroeconomic shocks, we expect -the global sporting goods industry to recover notably in 2021. Consumer confidence is forecast to rebound -amid widespread vaccination and effective pandemic management as well as major sports events -returning to the global stage. While extended lockdown periods, quarantine measures and physical store -closures are still forecast to have an adverse impact in the short term, the sporting goods industry is -projected to remain fundamentally attractive in the long term, as existing global trends are accelerating. -Sports-inspired apparel and footwear (‘athleisure’) keeps evolving as a structural component of the -broader fashion landscape, complemented by increasing sports participation rates as well as rising -awareness for health and wellness. In addition, sustainability is expected to further gain in importance -amid growing environmental awareness of consumers. Moreover, purchasing behavior is forecast to -further shift toward online and social channels, and consumer insights generation along with the creation -of premium shopping experiences become increasingly important. For the sporting goods sector, risks -related to disruptions caused by the coronavirus pandemic, supply chain challenges, re-escalating trade -tensions as well as the geo-political situation remain. -ADIDAS INCREASES OUTLOOK FOR FULL YEAR 2021 -While adidas continues to be impacted by Covid-19-related lockdowns, industry-wide supply-chain -challenges and the geo-political situation, the company has increased its top- and bottom-line outlook for -2021. -Given the accelerating top-line momentum, the company now expects currency-neutral sales to increase -at a rate of up to 20% year-over-year in 2021, driven by strong double-digit improvements in all markets. -This new outlook reflects sales growth of up to 7% in the second half of the year compared to the 2020 -level, which will be fueled by an array of innovative product releases, including the re-introduction of NMD, -one of adidas’ most successful franchises in recent history. In addition, the company’s Futurecraft 4D -2 This Management Report contains forward-looking statements that reflect Management’s current view with respect to the future development of adidas. The -outlook is based on estimates that we have made on the basis of all the information available to us at the time of completion of this First Half Year Report. In -addition, such forward-looking statements are subject to uncertainties as described in the Risk and Opportunity Report of the adidas 2020 Annual Report which -are beyond the control of the company. In case the underlying assumptions turn out to be incorrect or described risks or opportunities materialize, actual -results and developments may materially deviate (negatively or positively) from those expressed by such statements. adidas does not assume any obligation to -update any forward-looking statements made in this Management Report beyond statutory disclosure obligations. -3 Source: World Bank, Global Economic Prospects. -9 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -portfolio will be significantly expanded by the introduction of the 4D FWD Pulse. The scaling of the well- -established ZX franchise will continue with the introduction of the ZX 5K Boost. The Forum franchise, after -having been successfully incubated throughout the first two quarters of the year, will quadruple in size -during the second half of the year driven by several iterations as well as dedicated marketing activities. In -addition, major sport events, including the Olympic Games in Japan, the start of the club football season in -Europe and the kick-off to the NFL season in the US, as well as grassroots activities around the world will -provide ideal platforms to tell brand and product stories in front of global and local audiences. -The company’s full-year gross margin forecast continues to be for a level of around 52.0%. The operating -margin is now expected to increase to a level of between 9.5% and 10% (previously: 9% to 10%). Net -income from continuing operations is now projected to increase to a level of between € 1.4 billion and -€ 1.5 billion (previously: € 1.25 billion to € 1.45 billion). The company’s profitability outlook continues to -include temporary stranded costs related to the intended divestiture of Reebok. In 2021, these costs are -expected to amount to around € 250 million on the operating profit level and to impact net income from -continuing operations by approximately € 200 million. The medium-term growth outlook is not impacted -by these costs as adidas anticipates that only around 30% of the € 250 million will reoccur in 2022 and that -by 2023 the stranded costs will be fully eliminated. -RISKS AND OPPORTUNITIES -Uncertainties regarding the further development of the coronavirus pandemic remain high, in particular in -light of rising infection rates in Asia and Europe. Compared to the assessment in the 2020 Annual Report, -the company’s risk profile has slightly deteriorated due to an increased risk related to consumer demand -as a consequence of the geo-political situation and product supply due to additional supply chain -challenges. Macroeconomic, regulatory and currency risks could also have a high impact on the -company’s financial results. While we still see major opportunities related to consumer demand and -product offering as well as data analytics in the mid to long term, we consider upside potential related to -short-term opportunities as limited. -Nevertheless, given the company’s strong risk-bearing capacity due to its current liquidity position and -financial health, Management does not foresee any material jeopardy to the viability of the company as a -going concern. We remain confident that the earnings strength forms a solid foundation for our business -development and provides the necessary resources to pursue future opportunities. -10 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF FINANCIAL -POSITION -ADIDAS AG CONSOLIDATED STATEMENT OF FINANCIAL POSITION (IFRS) € IN MILLIONS - June 30, 2021 June 30, 2020 Change in % December 31, 2020 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | -Assets -| Cash and cash equivalents | | | | | 4,151 | 2,018 | 105.7 | 3,994 | -| ---------------------------------- | --- | --- | --- | --------- | --------- | ------- | -------- | ------- | -| Short-term financial assets | | | | | – | 6 | n.a. | 0 | -| | | | | | | | | | -| Accounts receivable | | | | | 2,324 | 1,869 | 24.4 | 1,952 | -| | | | | | | | | | -| Other current financial assets | | | | | 810 | 653 | 24.0 | 702 | -| | | | | | | | | | -| Inventories | | | | | 4,054 | 5,213 | (22.2) | 4,397 | -| | | | | | | | | | -| Income tax receivables | | | | | 66 | 121 | (45.5) | 109 | -| | | | | | | | | | -| Other current assets | | | | | 1,057 | 1,051 | 0.6 | 999 | -| | | | | | | | | | -| Assets classified as held for sale | | | | | 1,674 | – | n.a. | 0 | -| | | | | | | | | | -| Total current assets | | | | 14,136 | 10,931 | | 29.3 | 12,154 | -| | | | | | | | | | -| | | | | | – | – | | | -Property, plant and equipment 2,065 2,293 (9.9) 2,157 -| | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------- | ------- | -------- | ----- | -| Right-of-use assets | | | | | 2,430 | 2,733 | (11.1) | 2,430 | -| | | | | | | | | | -| Goodwill | | | | | 1,199 | 1,258 | (4.7) | 1,208 | -| | | | | | | | | | -| Trademarks | | | | | 16 | 820 | (98.0) | 750 | -| | | | | | | | | | -| Other intangible assets | | | | | 261 | 284 | (7.9) | 252 | -| | | | | | | | | | -| Long-term financial assets | | | | | 276 | 351 | (21.4) | 353 | -| | | | | | | | | | -Other non-current financial assets 243 340 (28.7) 414 -| Deferred tax assets | | | | | 1,154 | 1,166 | (1.0) | 1,233 | -| ------------------------ | --- | --- | --- | --- | ------- | ------- | -------- | ----- | -| | | | | | | | | | -| Other non-current assets | | | | | 96 | 124 | (22.5) | 103 | -| | | | | | | | | | -Total non-current assets 7,741 9,369 (17.4) 8,899 -| | | | | | | | | | -| ------------- | --- | --- | --- | --------- | --------- | --- | ------ | ------- | -| Total assets | | | | 21,877 | 20,301 | | 7.8 | 21,053 | -| | | | | | | | | | - -11 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ADIDAS AG CONSOLIDATED STATEMENT OF FINANCIAL POSITION (IFRS) € IN MILLIONS - June 30, 2021 June 30, 2020 Change in % December 31, 2020 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | -Liabilities and equity -| Short-term borrowings | | | | | 682 | 1,217 | (44.0) | 686 | -| --------------------------------------- | --- | --- | --- | --- | -------- | -------- | -------- | ------ | -| Accounts payable | | | | | 2,165 | 2,575 | (15.9) | 2,390 | -| | | | | | | | | | -| Current lease liabilities | | | | | 480 | 639 | (24.9) | 563 | -| | | | | | | | | | -| Other current financial liabilities | | | | | 340 | 265 | 28.5 | 446 | -| | | | | | | | | | -| Income taxes | | | | | 686 | 588 | 16.5 | 562 | -| | | | | | | | | | -| Other current provisions | | | | | 1,433 | 1,373 | 4.3 | 1,609 | -| | | | | | | | | | -| Current accrued liabilities | | | | | 2,260 | 1,933 | 16.9 | 2,172 | -| | | | | | | | | | -| Other current liabilities | | | | | 450 | 547.0 | (17.6) | 398 | -| | | | | | | | | | -| Liabilities classified as held for sale | | | | | 667 | – | n.a. | – | -| | | | | | | | | | -| Total current liabilities | | | | | 9,161 | 9,137 | 0.3 | 8,827 | -| | | | | | | | | | -| Long-term borrowings | | | | | 2,473 | 1,599 | 54.7 | 2,482 | -| | | | | | | | | | -Non-current lease liabilities 2,206 2,374 (7.1) 2,159 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other non-current financial liabilities 98 46 113.0 115 -| | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | ----- | ----- | ------- | --- | -| Pensions and similar obligations | | | | | 234 | 245 | (4.5) | 284 | -| | | | | | | | | | -Deferred tax liabilities 189 224 (15.7) 241 -| Other non-current provisions | | | | | 134 | 179 | (25.4) | 229 | -| ------------------------------- | --- | --- | --- | --- | ----- | ----- | -------- | --- | -| | | | | | | | | | -| Non-current accrued liabilities | | | | | 6 | 9 | (35.4) | 8 | -| | | | | | | | | | -| Other non-current liabilities | | | | | 18 | 16 | 11.9 | 17 | -| | | | | | | | | | -Total non-current liabilities 5,357 4,693 14.2 5,535 -| | | | | | | | | | -| ------------------------- | --- | --- | --- | --- | -------- | ------------- | ------- | ------ | -| Share capital | | | | | 195 | 195 | 0.0 | 195 | -| | | | | | | | | | -| Reserves | | | | | (190) | (2) 8,799.1 | | (474) | -| | | | | | | | | | -| Retained earnings | | | | | 7,101 | 6,037 | 17.6 | 6,733 | -| | | | | | | | | | -| Shareholders’ equity | | | | | 7,106 | 6,230 | 14.0 | 6,454 | -| | | | | | | | | | -| Non-controlling interests | | | | | 253 | 240 | 5.2 | 237 | -| | | | | | | | | | -| | | | | | | | | | -| Total equity | | | | | 7,358 | 6,471 | 13.7 | 6,691 | -| | | | | | | | | | -Total liabilities and equity 21,877 20,301 7.8 21,053 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | - -12 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | - -| | | 1 | | | 2 | | | | 3 | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONDENSED CONSOLIDATED INCOME STATEMENT -ADIDAS AG CONDENSED CONSOLIDATED INCOME STATEMENT (IFRS) € IN MILLIONS1 -| | | | | | | | | | | Second | | Second | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | ------- | --- | - First half First half quarter quarter - year 2021 year 2020 Change 2021 2020 Change -| Net sales | | | | | | 10,345 | | 7,733 | 33.8% | 5,077 | | 3,352 | 51.5% | -| --------- | --- | --- | --- | --- | --- | -------- | --- | ----- | ------- | ------- | --- | ------- | ----- | -Cost of sales 4,983 3,835 30.0% 2,446 1,631 49.9% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Gross profit 5,362 3,898 37.5% 2,632 1,720 53.0% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 51.8% 50.4% 1.4pp 51.8% 51.3% 0.5pp -Royalty and commission income 31 24 29.5% 17 9 95.7% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other operating income 10 20 (50.2%) 2 18 (86.1%) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other operating expenses 4,154 4,156 (0.0%) 2,107 2,010 4.9% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 40.2% 53.8% (13.6pp) 41.5% 60.0% (18.5pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Marketing and point-of-sale expenses 1,157 1,174 (1.4%) 616 525 17.2% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 11.2% 15.2% (4.0pp) 12.1% 15.7% (3.5pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Operating overhead expenses2 2,997 2,983 0.5% 1,492 1,484 0.5% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 29.0% 38.6% (9.6pp) 29.4% 44.3% (14.9pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Operating profit / (loss) 1,248 (215) n.a. 543 (263) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 12.1% (2.8%) 14.8pp 10.7% (7.8%) 18.5pp -| Financial income | | | | | | | 6 | 17 | (67.2%) | | 4 | 13 | (70.0%) | -| ------------------ | --- | --- | --- | --- | --- | --- | ---- | --- | --------- | --- | ---- | ---- | ------- | -| | | | | | | | | | | | | | | -| Financial expenses | | | | | | | 75 | 82 | (8.7%) | | 37 | 42 | (11.0%) | -| | | | | | | | | | | | | | | -Income / (loss) before taxes 1,179 (279) n.a. 510 (292) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 11.4% (3.6%) 15.0pp 10.0% (8.7%) 18.8pp -| Income taxes | | | | | | | 289 | (63) | | n.a. 123 | | (49) | n.a. | -| ------------ | --- | --- | --- | --- | --- | --- | ----- | ---- | --- | ---------- | --- | ------ | ---- | -| | | | | | | | | | | | | | | -(% of / income / (loss) before taxes) 24.5% 22.4% 2.1pp 24.0% 16.9% 7.1pp -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) from continuing operations 890 (217) n.a. 387 (243) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 8.6% (2.8%) 11.4pp 7.6% (7.2%) 14.9pp -Gain / (loss) from discontinued operations, net of tax 72 (74) n.a. 20 (74) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) 962 (291) n.a. 407 (317) n.a. -(% of net sales) 9.3% (3.8%) 13.1pp 8.0% (9.4%) 17.5pp -Income / (loss) attributable to shareholders 955 (264) n.a. 397 (295) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 9.2% (3.4%) 12.6pp 7.8% (8.8%) 16.6pp -Net income / (loss) attributable to non-controlling interests 7 (27) n.a. 11 (22) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Basic earnings per share from continuing operations (in €) 4.52 (0.97) n.a. 1.93 (1.13) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Diluted earnings per share from continuing operations (in €) 4.52 (0.97) n.a. 1.93 (1.13) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | | | -Basic earnings per share from continuing and discontinued -| | | | | | | | 4.89 | (1.35) | | n.a. 2.03 | | (1.51) | n.a. | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------- | --- | -------------- | --- | ---------- | ----- | - operations (in €) -| D il u | t e d e a r | n i n g s per share from continuing and discontinued | | | | | | | | | | | | -| ------ | -------------- | ------------------------------------------------------ | --- | --- | --- | --- | ------- | ------- | --- | ------------- | --- | --------- | ----- | -| | | ) | | | | | 4.89 | (1.35) | | n.a. 2.03 | | (1.51) | n.a. | -| op e | r a t io n s | (i n € | | | | | | | | | | | | - 1 Comparative informa tion for 2020 has been adjusted due to the classification of th e R eebok business a s discontinued o pe rations. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -2 Aggregated distribution and selling expenses, general and administration expenses, sundry expenses and impairment losses (net) on accounts receivable and contract assets. - -13 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF COMPREHENSIVE -INCOME -ADIDAS AG CONSOLIDATED STATEMENT OF COMPREHENSIVE INCOME (IFRS) € IN MILLIONS -| | | | | | First half | | First half | | Second | | Second | -| --- | --- | --- | --- | ---- | ----------- | --- | ----------- | ----------------- | ------- | ------------- | ------- | -| | | | | | year 2021 | | year 2020 | quarter 2021 | | | | -| | | | | | | | | | | quarter 2020 | | -Net income / (loss) after taxes 962 (291) 407 (317) -Items of other comprehensive income that -will not be reclassified subsequently to -profit or loss -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Remeasurements of defined benefit plans -| | | | | | | 43 | | 0 | | 6 | (19) | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | ---- | -(IAS 19), net of tax1 -| Net (loss) / gain on other equity investments | | | | | | | | | | | | -| ---------------------------------------------- | --- | --- | --- | --- | --- | ----- | --- | ----- | --- | --- | --- | -| | | | | | | (1) | | (2) | | 3 | (2) | -(IFRS 9), net of tax -| Subtotal of items of other comprehensive | | | | | | | | | | | | -| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -income that will not be reclassified 41 (1) 10 (21) -subsequently to profit or loss -Items of other comprehensive income that -| will be reclassified to profit or loss when | | | | | | | | | | | | -| -------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -specific conditions are met -Net gain / (loss) on cash flow hedges and -| | | | | | | 100 | | 142 | | (6) | (120) | -| --- | --- | --- | --- | --- | --- | ----- | --- | ----- | --- | ----- | ----- | -net foreign investment hedges, net of tax -| Net (loss) / gain on cost of hedging reserve – | | | | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | ----- | --- | -| | | | | | | (5) | | 9 | | (2) | 9 | -options, net of tax -| Net gain / (loss) on cost of hedging reserve – | | | | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | --- | --- | ---- | --- | ------ | --- | --- | --- | -| | | | | | | 10 | | (17) | | 5 | 23 | -forward contracts, net of tax -Currency translation differences 145 (177) 8 (34) - -Subtotal of items of other comprehensive -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -income that will be reclassified to profit or 251 (44) 5 (123) -loss when specific conditions are met -| Other comprehensive income | | | | | | 292 | | (45) | | 14 | (143) | -| --------------------------- | --- | --- | --- | --- | --- | ------ | --- | ------- | --- | ----- | ------ | -| | | | | | | | | | | | | -Total comprehensive income 1,254 (336) 422 (460) -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Attributable to shareholders of adidas AG 1,239 (311) 413 (432) -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Attributable to non-controlling interests 15 (24) 8 (28) - 1 Includes actuarial gains or losses relating to defined benefit obli ga tions, return on plan as sets (excluding inter e st income) and the ass e t c eiling effect. - -14 - -| | | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF CHANGES IN -EQUITY -ADIDAS AG CONSOLIDATED STATEMENT OF CHANGES IN EQUITY (IFRS) € IN MILLIONS -Cost of - Cumulative Cost of hedging -| | | | | | currency | | hedging | reserve - | | | | Share- | | Non- | | -| --- | --- | --- | --- | --- | --------- | --- | -------- | ---------- | --- | --- | --- | ------- | --- | ----- | --- | -Share Capital translation Hedging reserve - forward Other Retained holders’ controlling Total - capital reserve differences reserve options contracts reserves earnings equity interests equity -Balance at - 196 887 (470) (150) (10) 6 (218) 6,555 6,796 261 7,058 -December 31, 2019 -Other comprehensive -| | | | | | (180) | | 142 | 9 | (17) | (1) | | (48) | | 3 | (45) | -| --- | --- | --- | --- | --- | ------- | --- | ----- | --- | ------ | --- | --- | ------ | --- | --- | ---- | -income -Net income (264) (264) (27) (291) - -Total comprehensive - (180) 142 9 (17) (1) (264) (311) (24) (336) -income -Repurchase of -| | | | (1) | | | | | | | | (256) | (257) | | | (257) | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | --- | ----- | -adidas AG shares -| Repurchase of | | | | | | | | | | | | | | | | -| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -adidas AG shares due to -| equity-settled share- | | | (0) | | | | | | | | (14) | (14) | | | (14) | -| --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | --- | --- | ---- | -based payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Reissuance of treasury -shares due to e quity- -| settled share-based | | | 0 | | | | | | | | 17 | | 17 | | 17 | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | -payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Equity-settled share- -| based pa yment | | | | | | | | | | | (1) | | (1) | | (1) | -| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -First-time consolidation -due to obtaining control -| | | | | | | | | | | | | | | 3 | 3 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -in accordance with -IFRS 10 -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Balance at 195 887 (650) (9) (1) (11) (220) 6,037 6,230 240 6,471 -June 30, 2020 -Balance at - 195 887 (850) (250) (3) (23) (235) 6,733 6,454 237 6,691 -December 31, 2020 -Other comprehensive -| | | | | | 137 | | 100 | (5) | 10 | 41 | | 284 | | 9 | 292 | -| --- | --- | --- | --- | --- | ----- | --- | ----- | ----- | ---- | --- | --- | ----- | --- | --- | --- | -income -| Net income | | | | | | | | | | | 955 | 955 | | 7 | 962 | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------ | --- | ---- | ---- | - -Total comprehensive - 137 100 (5) 10 41 955 1,239 15 1,254 -income -Repurchase of -adidas AG shares due to -| | | | (1) | | | | | | | | (15) | (15) | | | (15) | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | --- | --- | ---- | -equity-settled share- -based payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Reissuance of treasury -shares due to e quity- -| | | | 1 | | | | | | | | 14 | | 15 | | 15 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | -settled share-based -payment -| | | | | | | | | | | | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | --- | ----- | -| Dividend payment | | | | | | | | | | | (585) | (585) | | | (585) | - -Equity-settled share- -based payment (1) (1) (1) -| Balance at | | | | | | | | | | | | | | | | -| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - 195 887 (712) (150) (7) (14) (194) 7,101 7,106 253 7,358 -June 30, 2021 - -15 - -| | | | | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| | 1 | | | 2 | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF CASH FLOWS -ADIDAS AG CONSOLIDATED STATEMENT OF CASH FLOWS (IFRS) € IN MILLIONS -| | | | | | | | | First half | | First half | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------ | -| | | | | | | | | year 2021 | | year 2020 | -| | | | | | | | | | | | -Operating activities: -| Income / (loss) before taxes | | | | | | | | | 1,179 | (279) | -| ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | -| | | | | | | | | | | | -Adjustments for: -Depreciation, amortization and impairment losses 568 645 -| Reversals of impairment losses | | | | | | | | | (7) | (1) | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ---- | -| | | | | | | | | | | | -| Interest income | | | | | | | | | (5) | (16) | -| | | | | | | | | | | | -| Interest expense | | | | | | | | | 69 | 77 | -| | | | | | | | | | | | -| Unrealized foreign exchange losses, net | | | | | | | | | 9 | 9 | -| | | | | | | | | | | | -Losses on sale of property, plant and equipment and intangible assets, net 4 3 -| | | | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | -| Other non-cash (income) /expense | | | | | | | | | (6) | 0 | -| | | | | | | | | | | | -Operating profit before working capital changes 1,810 438 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Increase) / decrease in receivables and other assets (553) 610 - -| Decrease / (increase) in inventories | | | | | | | | | 13 | (1,142) | -| ------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------- | -| | | | | | | | | | | | -Decrease in accounts payable and other liabilities (82) (539) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Cash generated from / (used in) operations before taxes 1,189 (632) -| | | | | | | | | | | | -| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| Income taxes paid | | | | | | | | | (112) | (186) | - -Net cash generated from / (used in) operating activities – continuing operations 1,077 (819) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash generated from / (used in) operating activities – discontinued operations 95 (6) - -Net cash generated from / (used in) operating activities 1,172 (824) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | -Investing activities: -Purchase of trademarks and other intangible assets (53) (33) -| Purchase of property, plant and equipment | | | | | | | | | (130) | (176) | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| | | | | | | | | | | | -| Proceeds from sale of property, plant and equipment | | | | | | | | | 0 | 2 | -| | | | | | | | | | | | -| Proceeds from sale of a disposal group | | | | | | | | | 12 | – | -| | | | | | | | | | | | -| Proceeds from disposal of discontinued operations | | | | | | | | | 29 | – | -| | | | | | | | | | | | -| Proceeds from sale of short-term financial assets | | | | | | | | | 0 | 288 | -| | | | | | | | | | | | -Purchase of investments and other long-term assets 55 (37) -| | | | | | | | | | | | -| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Interest received | | | | | | | | | 5 | 16 | -| | | | | | | | | | | | -Net cash (used in) / generated from investing activities – continuing operations (81) 59 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash used in investing activities – discontinued operations (5) (8) - -Net cash (used in) / generated from investing activities 86 51 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | -| | | | | | | | | | | | -16 - -| | | | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| | 1 | | | 2 | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ADIDAS AG CONSOLIDATED STATEMENT OF CASH FLOWS (IFRS) € IN MILLIONS -| | | | | | | | First half | | First half | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | ---------- | ----------- | -| | | | | | | | year 2021 | year 2020 | | - -| Financing activities: | | | | | | | | | | -| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| Reverse transaction of buyback of Eurobonds | | | | | | | | – | 11 | -| Interest paid | | | | | | | | (54) | (66) | -| | | | | | | | | | | -| Repayments of lease liabilities | | | | | | | | (267) | (226) | -| | | | | | | | | | | -| Dividend paid to shareholders of adidas AG | | | | | | | | (585) | – | -| | | | | | | | | | | -| Repurchase of treasury shares | | | | | | | | – | (257) | -| | | | | | | | | | | -Repurchase of treasury shares due to share-based payments (15) (14) -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Proceeds from reissuance of treasury shares due to share-based payments 12 12 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Repayments of) / proceeds from short-term borrowings (17) 1,165 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash (used in) / generated from financing activities – continuing operations (926) 624 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash used in financing activities – discontinued operations (19) (23) - -Net cash (used in) / generated from financing activities (945) 601 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| Effect of exchange rates on cash | | | | | | | | 15 | (29) | -| --------------------------------- | --- | --- | --- | --- | --- | --- | --- | ----- | ----- | -| | | | | | | | | | | -Increase / (decrease) of cash and cash equivalents 157 (201) - -Cash and cash equivalents at beginning of year 3,994 2,220 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Cash and cash equivalents at end of period 4,151 2,018 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -17 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -EXPLANATORY NOTES TO THE CONDENSED INTERIM -CONSOLIDATED FINANCIAL STATEMENTS (IFRS) AS -AT JUNE 30, 2021 -01 GENERAL -The interim consolidated financial statements of adidas AG and its subsidiaries (collectively ‘adidas,’ the -‘Group’ or ‘the company’) for the first half year ending June 30, 2021, are prepared in compliance with -International Financial Reporting Standards (IFRS), as endorsed by the European Union (EU). The company -applied all International Financial Reporting Standards issued by the International Accounting Standards -Board (IASB) and Interpretations of the IFRS Interpretations Committee effective as at June 30, 2021, -insofar as they have already been adopted into European law. -These interim consolidated financial statements were prepared in compliance with International -Accounting Standard IAS 34 ‘Interim Financial Reporting.’ Accordingly, these interim consolidated -financial statements do not include all of the information and notes required for consolidated financial -statements at financial year-end. Therefore, these interim consolidated financial statements should be -read in conjunction with the 2020 annual consolidated financial statements. The accounting policies as -well as principles, practices and presentation applied in the consolidated financial statements for the year -ending December 31, 2020, also apply to the interim consolidated financial statements for the first half -year ending June 30, 2021. -The following new accounting standards and interpretations as well as amendments to existing standards, -which were issued by the IASB and are effective in the EU for financial years beginning on January 1, 2021, -have been applied for the first time: -Interest Rate Benchmark Reform Phase 2 – Amendments to IFRS 9, IAS 39, IFRS 7, IFRS 4 and IFRS 16 -─ -Amendments to IFRS 4 Incurance Contracts - deferral of IFRS 9 -─ -These do not have any material impact on the interim consolidated financial statements of adidas AG. -Further information can be found in the 2020 Annual Report. -New accounting standards and interpretations as well as amendments to existing standards that are not -yet effective in the EU or effective for financial years beginning after January 1, 2021, are not expected to -have any material impact on the consolidated financial statements. -The amendment to IFRS 16 ‘Covid-19-Related Rent Concessions’ published by the IASB on May 28, 2020, -and effective for financial years beginning on or after June 1, 2020, was originally limited to rent -concessions that were due on or before June 30, 2021. The IASB has decided to extend the period during -which the concessions can be applied to financial years beginning on or after April 1, 2021. This decision -will not impact adidas, as adidas has not applied and will not apply the option provided in this amendment -regarding the treatment of rent concessions relating to the coronavirus pandemic. The option allows -lessees to treat rental concessions in connection with the coronavirus pandemic not as lease -modifications within the meaning of IFRS 16 but to realize the effects of the rental concessions in the -periods in which they are granted. -18 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -The preparation of financial statements in conformity with IFRS requires the use of assumptions and -estimates that affect reported amounts and related disclosures. The first half of 2020 was significantly -impacted by the coronavirus pandemic. Further information can be found in Note 8 of the first half year -report 2020. Compared to the first half of 2020, business development in the first half of 2021 was less -impacted by the effects of the coronavirus pandemic. Nevertheless, due to the ongoing pandemic, it is still -difficult to predict the resulting effects on assets and liabilities as well as income and expenses. Estimates -and assumptions relevant to the financial statements were made to the best of our knowledge, based on -current events and measures. In particular, judgments and discretion are used in assessing the -recoverability of deferred tax assets, in impairment tests for non-financial assets, in determining -allowances for inventories and accounts receivable, and in the calculation of provisions. -There were no changes in the scope of consolidation in the first half of 2021. -The interim consolidated financial statements and the interim Group management report were not audited -in accordance with § 317 German Commercial Code (Handelsgesetzbuch – HGB) nor reviewed in -accordance with § 115 section 5 German Securities Trading Act (Wertpapierhandelsgesetz – WpHG) by an -auditor. -Costs that are incurred unevenly during the financial year are anticipated or deferred in the interim -consolidated financial statements only if it would be also appropriate to anticipate or defer such costs at -the end of the financial year. -The results of operations for the first half year ending June 30, 2021, are not necessarily indicative of -results to be expected for the entire year. -The interim consolidated financial statements are presented in euros (€) and, unless otherwise stated, all -values are presented in millions of euros (€ in millions). Due to rounding principles, numbers presented -may not sum up exactly to totals provided. -02 REVENUE -The sales of the company in certain product categories are seasonal, and therefore, revenues and -attributable earnings may vary within the financial year. In the years before the coronavirus pandemic, -sales and earnings tended to be strongest in the first and third quarters of the financial year because -these coincided with the launch of the spring/summer and fall/winter collections, respectively. In the first -half of 2021, sales stabilized but variations may still occur due to the ongoing coronavirus pandemic. Shifts -in the share of sales and attributable earnings of particular product categories or in the regional -composition may occur throughout the year. A disaggregation of revenue into product categories is -contained in these Notes. ► SEE NOTE 08 -03 DISCONTINUED OPERATIONS -On February 16, 2021, the company decided to begin a formal process aimed at divesting Reebok. Due to -the concrete plans to sell Reebok and the approval by the respective committees, the Reebok operating -business is reported as discontinued operations and classified as a disposal group held for sale at the end -of the first quarter. The fair value was determined based on the basis of existing non-binding purchase -price offers at this point of time. The transaction is expected to be completed later in 2021. -19 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -Reebok was previously neither classified as assets held for sale nor as discontinued operations. The prior -year figures of the consolidated income statement and the consolidated statement of cash flows have been -restated to show the discontinued operations separately from continuing operations. -The results of the Reebok operations are shown as discontinued operations in the consolidated income -statement: -DISCONTINUED OPERATION REEBOK -First half First half -| € in millions | | | year 2021 | year 2020 | | -| --------------------- | --- | --- | ------------ | ------------ | ----- | -| Operational business | | | | | | -| Net Sales | | | | 823 | 600 | -| Expenses | | | | (666) | (658) | -| | | | | | | -Gain / (loss) from operating activities before taxes 157 (58) -| | | | | | | -| ------------ | --- | --- | --- | ------ | ---- | -| Income taxes | | | | (78) | (12) | -Gain / (loss) from operating activities, net of tax 78 (69) -| | | | | | | -| ----------------------------- | --- | --- | --- | ------ | --- | -| Transaction costs, net of tax | | | | (10) | – | -Gain / (loss) from discontinued operations, net of tax 68 (69) -| | | | | | | -| --- | --- | --- | --- | --- | --- | -Basic earnings per share from discontinued operations (€) 0.35 (0.36) -Diluted earnings per share from discontinued operations (€) 0.35 (0.36) - Rounding differences may arise in percentages and totals for figures presented in millions as calculation is alway s b ased on the figures s ta ted in thousands. -Gains from discontinued operations for the first half year ending June 30, 2021, in an amount of -€ 72 million (2020: losses of € 74 million) are entirely attributable to the shareholders of adidas AG. The -gains from discontinued operations also include besides Reebok an amount of € 4 million (2020: loss -€ 5 million) relating to divestitures in previous years. -Total net loss of cumulative foreign currency translation differences for discontinued operations reflected -within equity amount to € 225 million. For the first half year 2021 the loss within equity amount to -€ 125 million. For 2020 the gain within equity amount to € 368 million. -The income tax expense related to discontinued operations amounts to € 76 million. This amount includes -the revaluation of deferred tax assets and liabilities as a result of the tax rate change in the UK -(€ 34 million). - -20 - - HALF YEAR REPORT 2021 - -| 1 | 2 | 3 | | | -| --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -04 ASSETS AND LIABILITIES HELD FOR SALE -At June 30, 2021, the disposal group Reebok was stated at carrying amount and comprised the following -major classes of assets and liabilities: -GROUP OF ASSETS AND LIABILITIES -Assets classified as held-for-sale € in millions June 30, 2021 -| | | | | | -| ------------------- | --- | --- | --- | --- | -| Accounts receivable | | | | 207 | -| Inventories | | | | 388 | - -| Other current financial assets | | | | 36 | -| ------------------------------ | --- | --- | --- | --- | - -| Other current assets | | | | 42 | -| --------------------- | --- | --- | --- | ---- | -| | | | | | -| Total current assets | | | | 673 | - -| Long-term financial assets | | | | 12 | -| ----------------------------- | --- | --- | --- | --- | -| Property, plant and equipment | | | | 66 | -| | | | | | -| Right-of-use assets | | | | 108 | -| | | | | | -| Goodwill | | | | 27 | - -| Trademarks | | | | 757 | -| ------------------ | --- | --- | --- | --- | -| | | | | | -| Deferred tax asset | | | | 24 | - -| Other non-current financial assets | | | | 6 | -| ---------------------------------- | --- | --- | --- | --- | - -| Total non-current assets | | | | 1,000 | -| ------------------------- | --- | --- | --- | -------- | -| Total assets | | | | 1,673 | -Liabilities classified as held for sale € in millions June 30, 2021 -| | | | | | -| ----------------------- | --- | --- | --- | --- | -| Accounts payable | | | | 174 | -| Current lease liability | | | | 29 | - -| Other current provisions | | | | 60 | -| ------------------------ | --- | --- | --- | --- | - -| Other current accrued liabilities | | | | 112 | -| ---------------------------------------------- | --- | --- | --- | ---- | -| | | | | | -| Other current financial liabilities | | | | 12 | -| | | | | | -| Other current liabilities | | | | 18 | -| | | | | | -| Total current liabilities | | | | 405 | -| | | | | | -| Non-current lease liability | | | | 123 | -| Provision for pensions and similar obligations | | | | 6 | -| | | | | | -| Deferred tax liability | | | | 108 | -| | | | | | -| Other non-current provisions | | | | 18 | - -| Other non-current accrued liabilities | | | | 2 | -| ---------------------------------------- | --- | --- | --- | ---- | -| | | | | | -| Other non-current financial liablilities | | | | 4 | -| | | | | | -| Other non-current liablilities | | | | 2 | -| | | | | | -| Total non-current liabilities | | | | 263 | -| | | | | | -| Total liabilities | | | | 667 | - -At June 30, 2021, the company assessed the fair value of the disposal group on the basis of existing non- -binding purchase price offers at this point of time. There was no indication for any impairment of the -disposal group. -Further information on the allocation of goodwill is contained in these Notes. ► SEE NOTE 09 -| | | | | | -| --- | --- | --- | --- | --- | -21 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | | | 3 | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -05 SHAREHOLDERS’ EQUITY -During the period from January 1, 2021, to June 30, 2021, the nominal capital of adidas AG remained -unchanged. Consequently, on June 30, 2021, the nominal capital of adidas AG amounted to € 200,416,186 -divided into 200,416,186 registered no-par-value shares. -In the 2016 financial year, adidas AG introduced an employee stock purchase plan in favor of employees of -adidas AG and its affiliated companies. -adidas AG purchased shares in connection with this employee stock purchase plan in the first six months -of the 2021 financial year. More details on the repurchase of adidas AG shares and use of treasury shares -in connection with the employee stock purchase plan in the first six months of the 2021 financial year are -set out in the tables ‘Repurchase of adidas AG shares and use of treasury shares in the context of the -employee stock purchase plan2021’ and ‘Repurchase of adidas AG shares and use of treasury shares in -the context of the employee stock purchase plan 2021/matching shares’. -REPURCHASE OF ADIDAS AG SHARES AND USE OF TREASURY SHARES IN THE CONTEXT OF THE EMPLOYEE STOCK PURCHASE PLAN 2021 -Total price -| | | | | | in € | Average | | Amount | | Amount | | | -| --- | --- | --- | --- | ----------- | -------- | --------- | ------ | ---------- | --- | ---------- | --- | --- | -| | | | | (excluding | | purchase | | in the | | in the | | | -| | | | | incidental | | | price | nominal | | nominal | | | -Purchase Number of purchasing per share capital capital Issuance date -date shares costs) in € in € in % to employees -| January 8, | | | | | | | | | | | January 12, | | -| ----------- | --- | ------ | --- | ------------ | --- | ------ | --- | ------ | --- | ---- | ------------ | ----- | -| | | 23,652 | | 6,992,862.30 | | 295.66 | | 23,652 | | 0.01 | | | -| 2021 | | | | | | | | | | | | 2021 | -| April 9, | | | | | | | | | | | April 13, | | -| | | 24,032 | | 6,757,458.75 | | 281.19 | | 24,032 | | 0.01 | | | -| 2021 | | | | | | | | | | | | 2021 | -| | | | | | | | | | | | | | - -REPURCHASE OF ADIDAS AG SHARES AND USE OF TREASURY SHARES IN THE CONTEXT OF THE EMPLOYEE STOCK PURCHASE PLAN 2021/ MATCHING -SHARES -Total price -| | | | | | in € | Average | | Amount | | Amount | | | -| --- | --- | --- | --- | ----------- | -------- | --------- | ------ | ---------- | --- | ---------- | --- | --- | -| | | | | (excluding | | purchase | | in the | | in the | | | -| | | | | incidental | | | price | nominal | | nominal | | | -Purchase Number of purchasing per share capital capital Issuance date -date shares costs) in € in € in % to employees -| January 8, | | | | | | | | | | | January 12, | | -| ----------- | --- | ----- | --- | ------------ | --- | ------ | --- | ----- | --- | ----- | ------------ | ----- | -| | | 2,843 | | 840,550.80 | | 295.66 | | 2,843 | | 0.001 | | | -| 2021 | | | | | | | | | | | | 2021 | -| April 9, | | | | | | | | | | | April 13, | | -| | | 3,817 | | 1,073,286.45 | | 281.19 | | 3,817 | | 0.002 | | | -| 2021 | | | | | | | | | | | | 2021 | -| | | | | | | | | | | | | | - -On May 12, 2021, the Annual General Meeting of adidas AG approved the proposal of the Executive Board -and Supervisory Board on the appropriation of retained earnings for the 2020 financial year in the amount -of € 3.00 per share. The dividend was paid on May 18, 2021. Based on the number of dividend-entitled -shares at the time of the Annual General Meeting, this results in a dividend distribution of € 585,198,180. -On June 30, 2021 adidas AG held a total of 5,350,126 treasury shares, corresponding to a notional amount -of € 5,350,126 in the nominal capital and consequently 2.67% of the nominal capital. In accordance with -§ 71b German Stock Corporation Act (Aktiengesetz - AktG), the treasury shares held directly or indirectly -do not confer any rights to the company. -22 - -| | | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -06 FINANCIAL INSTRUMENTS -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | | | | Category | | | | June 30, 2021 | | | | | December 31, 2020 | | -| --- | --- | --- | --- | --- | --------- | --- | --- | --- | -------------- | --- | --- | --- | --- | ------------------ | --- | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Financial assets -Cash and cash equivalents -Cash and cash equivalen ts Amortized 2,383 – – – – 1,762 – – – – -| | | | | | cost | | | | | | | | | | | -| --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Fair value -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -through -| Cash equivalents | | | | | | 1,768 | 1,768 | – | 1,768 | – | 2,232 | 2,232 | | – 2,232 | – | -| ---------------- | --- | --- | --- | --- | --- | ----- | ----- | --- | ----- | --- | ----- | ----- | --- | ------- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Short-term financial assets | | | | | | | – – | – | – | – | | 0 | 0 | – | 0 – | -| --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Accounts receivable | | | | | cost | 2,324 | – | – | – | – | 1,952 | | – | – | – – | -| ------------------- | --- | --- | --- | --- | ----- | ----- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -Other current financ ial assets - -| Derivatives used in hedge | | | | | Hedge | | | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - accounting 119 119 – 119 – 163 163 – 163 – -accounting -| | | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -| Derivatives not used in hedge | | | | | through | | | | | | | | | | | -| | | | | | | | 32 32 | – | 32 | – | | 32 | 32 | – | 32 – | -| accounting | | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Promissory notes | | | | | | 147 | 147 | – | – | 147 | | 6 | 6 | – | – 6 | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Other financial assets | | | | | cost | 511 | – | – | – | – | 511 | | – | – | – – | -| ---------------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Long-term financial asse ts - -Fair value -| | | | | | through | | | | | | | | | | | -| --- | --- | --- | --- | --- | --------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Other equity investments profit or 89 89 – – 89 89 89 – – 89 -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Other equity investments | | | | | other | | 73 73 | – | – | 73 | | 80 | 80 | – | – 80 | -| ------------------------ | --- | --- | --- | --- | ------ | --- | ----- | --- | --- | --- | --- | --- | --- | --- | ---- | -compre- -hensive -| | | | | | income | | | | | | | | | | | -| --- | --- | --- | --- | --- | ------ | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Other investments | | | | | | | 26 25 | – | 251 | – | | 35 | 37 | – | 372 – | -| ----------------- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | ----- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Other investments | | | | | cost | | 87 – | – | – | – | 149 | | – | – | – – | -| ----------------- | --- | --- | --- | --- | ----- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -Loans Amort iz e d 0 0 – – – 0 0 – – – -| | | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other non -current financial assets -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | - -| Derivatives used in hedge | | | | | Hedge | | | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -accounting accounting 18 18 – 18 – 14 14 – 14 – -| | | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -| Derivatives not used in hedge | | | | | through | | | | | | | | | | | -| | | | | | | 108 | 108 | – | 108 | – | | 99 | 99 | – | 99 – | -| accounting | | | | | profit or | | | | | | | | | | | -| | | | | | loss | | | | | | | | | | | - -Fair value -through -| Promissory notes | | | | | | | – – | – | – | – | 166 | 166 | | – | – 166 | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Earn-out components | | | | | | | – – | – | – | – | | 12 | 12 | – | – 12 | -| ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | - -23 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | | | Category | | | | June 30, 2021 | | | | | December 31, 2020 | | -| --- | --- | --- | --- | --------- | --- | --- | --- | -------------- | --- | --- | --- | --- | ------------------ | --- | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -Amortized -| Other financial assets | | | | cost | 117 | – | – | – | – | 114 | – | – | – | – | -| ---------------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Assets classified as held for sale Amort iz e d 1,674 – – – – – – – – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Financial assets per level – – – 2,070 309 – – – 2,577 352 - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Financial liabilities -Short-term borrowings -Amortized -Bank borrowings cost 82 – – – – 87 – – – – - Amort iz e d 600 600 600 – – 599 605 605 – – -| Eurobond | | | | | | | | | | | | | | | -| -------- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | c o s t | | | | | | | | | | | -Accounts pay able Amort iz e d 2,165 – – – – 2,390 – – – – - -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Current accrued l iabilities Amort iz e d 801 – – – – 939 – – – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -C u r r e nt a c c r u e d li a bilities for Amort iz e d 801 – – – – 743 – – – – -| cu s t o m | e r d i s c o u n | t s | | c o s t | | | | | | | | | | | -| ---------- | ------------------ | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other current financial liabilities -| | | | | | | | | | | | | | | | -| -------------------------- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Derivatives used in hedge | | | | Hedge | | | | | | | | | | | - accounting 157 157 – 157 – 258 258 – 258 – -accounting -| | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -| Derivatives not used in hedge | | | | through | | | | | | | | | | | -| | | | | | | 26 26 | – | 26 | – | 24 | 24 | – | 24 | – | -| accounting | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Other financial liabilities | | | | cost | 158 | – | – | – | – | 164 | – | – | – | – | -| --------------------------- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Current lease liabilities n.a. 480 – – – – 563 – – – – - -Long-term borrowings -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Bank borrowings | | | | | | 91 – | – | – | – | 103 | – | – | – | – | -| --------------- | --- | --- | ----- | -------- | --- | ------ | --- | --- | --- | ---- | --- | --- | --- | --- | -| | | | | cost | | | | | | | | | | | -| | | | Amort | iz e d | | | | | | | | | | | -Eurobond 1,889 1,944 1,944 – – 1,888 1,987 1,987 – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | Amort | iz e d | | | | | | | | | | | -Convertibl e bond 493 628 628 – – 491 631 631 – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Non-current accr ued liabilities Amort iz e d 1 0 – – – 2 0 – – – - -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -O th e r n o n-current financial – - -lia b il it ie s -D e r iv a t iv e s used in hedge H e d g e 5 5 – 5 – 26 26 – 26 – -n -| ac c | o u t in g | | acco | un t in g | | | | | | | | | | | -| ------------------------------- | ---------- | --- | ---- | ----------- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -| | | | | Fair value | | | | | | | | | | | -| Derivatives not used in hedge | | | | through | | | | | | | | | | | -| | | | | | | 34 34 | – | 34 | – | 85 | 85 | – | 85 | – | -| accounting | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Other financial liabilities | | | | cost | | – – | – | – | – | 4 | – | – | – | – | -| --------------------------- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Non-current lease liabili ties n.a. 2,206 2,206 – – – 2,159 – – – – - -Amortized -Liabilities classified as held for sale cost 667 – – – – – – – – – -Financial liabilities per level – – 3,172 222 – – – 3,2230 393 – - -Thereof: aggregated by category -according to IFRS 9 -Financial assets at fair value -| through profit or loss | | | | | 2,170 | | | | | 2,670 | | | | | -| ----------------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -(FVTPL) -Thereof: designated as such -| upon initial recognition | | | | | | – | | | | – | | | | | -| -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Fair Value Option – FVO) -| | | | | | | 87 | | | | 87 | | | | | -| --- | --- | --- | --- | --- | --- | --------- | --- | --- | ----- | ----- | ----- | --- | ----- | --- | -Thereof: held for trading (FAHfT) - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -24 - -| | | | HALF YEAR REPORT 2021 | | | | | -| ---- | --- | --- | ---------------------- | --- | --- | --- | --- | - -| | 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | Category | | June 30, 2021 | | | December 31, 2020 | -| --- | --- | ----------- | --- | -------------- | --- | --- | ------------------ | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -Financial assets at fair value -| through other | | | 73 | | | 80 | | -| -------------- | --- | --- | --- | --- | --- | --- | --- | -comprehensive income (FVOCI) -Thereof: debt instruments – – - -Thereof: equity investments -(without recycling to 73 80 -profit and loss) -Financial assets at amortized cost -| | | | 7,096 | | | 4,489 | | -| --- | --- | --- | ----- | --- | --- | ----- | --- | -(AC) -Financial liabilities at fair value -thro ugh profit or loss -| | | | 119 | | | 108 | | -| --- | --- | --- | --- | --- | --- | --- | --- | -(FVTPL) -| | | | – | | | – | | -| --- | --- | --- | ------------ | --- | ------- | -------- | --------- | -Thereof: held for trading (FLHfT) - -Financial liabilities at amortized -cost (AC) 7,747 7,409 -| | | | | | | | | -| --- | --- | --- | ----------- | --- | ------- | ------- | --------- | - 1 Net gains in the amount of € 1 million and gains in the amount of € 0 million due to currenc y translation differences were recogn ized in equity i n 2021 -2 Net gains in the amount of € 0 million and losses in the amount of € 2 million due to currency translation differences were recognized in equity in 2020 - -Level 1 is based on quoted prices in active markets for identical assets or liabilities. -Level 2 is based on inputs other than quoted prices included within Level 1 that are observable for the assets or liabilities, either directly (i.e. as prices) or indirectly (i.e. derived from prices). -Level 3 is based on inputs for the assets or liabilities that are not based on observable market data (unobservable inputs). - -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -25 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -RECONCILIATION OF FAIR VALUE HIERARCHY LEVEL 3 IN 2020 € IN MILLIONS -| | | | | | | | | | Realized | | Unrealized | | | | -| --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --------- | --- | ----------- | -------------- | ----------- | --- | -| | | | | | Fair value | | | | | | | Currency | Fair value | | - Jan. 1,2020 Additions Disposals Gains Losses Gains Losses translation Dec. 31, 2020 -Investments in other equity instruments held -| | | | | | | 84 | – | – | – | – | 2 | – | – | 87 | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -for trading (FAHfT) -I n v e s t m ents in othe r equity instruments 2 – – – – – – – 2 -| ( F V T P | L ) | | | | | | | | | | | | | | -| --------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -I n v e s tm ents in other equity instruments 78 3 – – – – (2) – 79 -| ( F V O C | I) | | | | | | | | | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Promissory notes (FVTPL) 182 – (1) – (3) 9 – (15) 171 - -Earn-out components – assets (FVTPL) 45 – (41) – – 12 – (4) 12 -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | | | | -RECONCILIATION OF FAIR VALUE HIERARCHY LEVEL 3 IN 2021 € IN MILLIONS -| | | | | | | | | | Realized | | Unrealized | | | | -| --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --------- | --- | ----------- | -------------- | --- | --- | -| | | | | | Fair value | | | | | | | Currency | | | -Fair value - Jan. 1,2021 Additions Disposals Gains Losses Gains Losses translation June 30, 2021 -Investments in other equity instruments held -| | | | | | | 87 | – | – | – | – | – | – | – | 87 | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -for trading (FAHfT) -| I n v e s t | m ents in othe r equity instruments | | | | | | | | | | | | | | -| ----------- | ------------------------------------ | --- | --- | --- | --- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | --- | -| | | | | | | 2 | – | – | – | – | – | – | – | 2 | -( F V T P L ) -I n v e s tm ents in other equity instruments -| | I) | | | | | 79 | 1 | (6) | – | – | – | (1) | – | 73 | -| --- | --- | --- | --- | --- | --- | ----- | ---- | ------ | ---- | ---- | ---- | ------ | ---- | --- | -( F V O C -Promissory notes (FVTPL) 171 – (20) – – – (9) 5 147 - -Earn-out components – assets (FVTPL) 12 – (21) 9 – – – – – -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -The valuation methods used in measuring Level 1, Level 2 and Level 3 fair values remain unchanged and -can be found in the Notes to the 2020 consolidated financial statements. -NET GAINS/(LOSSES) ON FINANCIAL INSTRUMENTS RECOGNIZED IN THE CONSOLIDATED INCOME STATEMENT € IN MILLIONS -| | | | | | | | | | | Period ending | | Year ending | | | -| --- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------- | ----- | --- | -| | | | | | | | | | | | | | | | -| | | | | | | | | | | June 30, 2021 | | Dec. 31, 2020 | | | -| | Financial assets classified at | | | | | | | | | | | | | | -| | | | | | | | | | | | (3) | | (114) | | -amortized cost (AC) -| | Financial assets at fair value through | | | | | | | | | | | | | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | 1 | 18 | | -profit or loss (FVTPL) -| | Thereof: designated as such upon | | | | | | | | | | | – | – | | -| --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ----- | --- | -initial recognition -| | Thereof: classified as held for trading | | | | | | | | | | | – | 2 | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | - -Equity instruments at fair value through -| | | | | | | | | | | | | – | – | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -profit or loss (FVTPL) -| | Equity instruments at fair value through | | | | | | | | | | | | | | -| --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | – | – | | -other comprehensive income (FVOCI) -| | Financial liabilities at amortized cost (AC) | | | | | | | | | | | 1 | 39 | | -| --- | -------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ----- | --- | - -Financial liabilities at fair value through -| | | | | | | | | | | | | – | – | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -profit or loss (FVTPL) -| | Thereof: designated as such upon | | | | | | | | | | | | | | -| --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | – | – | | -initial recognition -| | Thereof: classified as held for trading | | | | | | | | | | | – | – | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------ | --- | - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -26 - -| | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| 1 | | 2 | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -07 EARNINGS PER SHARE -Basic earnings per share are calculated by dividing the net income from continuing operations attributable -to shareholders by the weighted average number of shares outstanding during the year, excluding -ordinary shares purchased by adidas and held as treasury shares. If negative earnings per share are -reported, according to IAS 33.41 no anti-dilutive effect may be taken into account. -EARNINGS PER SHARE - Continuing operations Discontinued operations Total - First half year First half year First half year First half year First half year -First half year -| | | 2021 | 2020 | 2021 | 2020 | 2021 | 2020 | -| --- | --- | ------- | ------- | ------- | ------- | ------- | ----- | -Net income / (loss) from -| continuing operations | | 890 | (217) | – | – | – | – | -| ---------------------- | --- | ----- | ------- | --- | --- | --- | --- | -(€ in millions) -| Net income / (loss) attributable | | | | | | | | -| --------------------------------- | --- | --- | --- | --- | --- | --- | --- | -to non-control ling interests -| | | 7 | (27) | – | – | – | – | -| --- | --- | --- | ------ | --- | --- | --- | --- | -(€ in millions) -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable 883 (190) 72 (74) 955 (264) -to shareholde rs (€ in millions) -Weighted average number of shares 195,066,060 195,266,358 195,066,060 195,266,358 195,066,060 195,266,358 -Basic earnings per share (€) 4.52 (0.97) 0.37 (0.38) 4.89 (1.35) -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable - to shareholders (€ in millions) 883 (190) 72 (74) 955 (264) -Net income / (loss) used to -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -determine diluted earnings 883 (190) 72 (74) 955 (264) -per share (€ in millions) -Weighted average number of shares 195,066,060 195,266,358 195,066,060 195,266,358 195,066,060 195,266,358 -Dilutive effect of share-based -| | | 6,163 | – | – | – | 6,163 | – | -| --- | --- | -------- | ---- | ---- | ---- | -------- | --- | -payments -| Weighted average number of | | | | | | | | -| --------------------------- | --- | --- | --- | --- | --- | --- | --- | -shares for diluted earnings - 195,072,223 195,266,358 195,066,060 195,266,358 195,072,223 195,266,358 -per share -Diluted earnings per share (€) 4.52 (0.97) 0.37 (0.38) 4.89 (1.35) - -27 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -08 SEGMENTAL INFORMATION -adidas operates predominantly in one industry segment – the design, distribution and marketing of athletic -and sports lifestyle products. -As at June 30, 2021, following the company’s internal management reporting by markets and in -accordance with the definition of IFRS 8 ‘Operating Segments,’ five operating segments were identified: -EMEA, North America, Asia-Pacific, Greater China and Latin America. -In order to be able to successfully execute our new strategy ‘Own the Game’ for the period until 2025, -adidas has changed its organizational structure. Since January 1, 2021, adidas manages Greater China as -a separate market and integrated Russia/CIS and Emerging Markets into the European market. -The operating segment Reebok North America which was reported in the internal management reporting -until February 16, 2021, is not monitored anymore due to the reclassification of Reebok business as -discontinued operations on February 16th, 2021. Therefore the segment North America only includes the -business activities of adidas. Comparative segmental information has been retrospectively adjusted. Also -the Reebok business activities in the other operating segments have been removed in the segment -information for 2021 and 2020. -Each market comprises all wholesale, retail and e-commerce business activities relating to the -distribution and sale of products of the adidas brand to retail customers and end consumers. -Other Businesses includes the business activities of the Y-3 label and other subordinated businesses -which are not monitored separately by the chief operating decision maker. Also, certain centralized -corporate functions do not meet the definition of IFRS 8 for an operating segment. This includes, in -particular, functions such as Global Brands and Global Sales (central brand and distribution -management), central treasury, global sourcing as well as other headquarter functions. Assets, liabilities, -income and expenses relating to these corporate functions are presented in the reconciliations. -The chief operating decision-maker for adidas has been defined as the entire Executive Board of -adidas AG. -Net sales represents revenue from contracts with customers. There are no intersegment sales between -the reportable segments. Accounting and valuation policies applied for reporting segmental information -are the same as those used for adidas. -The results of the operating segments are reported in the line item ‘Segmental operating profit.’ This is -defined as gross profit minus other operating expenses plus royalty and commission income and other -operating income attributable to the segment or group of segments, however without considering -headquarter costs and central expenses for marketing. -Segmental assets include accounts receivable as well as inventories. Only these items are reported to the -chief operating decision-maker on a regular basis. -Segmental liabilities only contain accounts payable from operating activities as there are no other liability -items reported regularly to the chief operating decision-maker. -28 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -SEGMENTAL INFORMATION € IN MILLIONS -| | | | Net sales | Segmental | Segmental | Segmental | -| --- | --- | --- | ---------- | ---------- | ---------- | ---------- | - (thir d parties)1 opera tin g profit1 assets2 li abilities2 - - 2021 2020 2021 2020 2021 2020 2021 2020 -EMEA 3,680 2,670 792 308 2,458 2,390 (186) (176) -North America 2,406 1,892 447 45 1,540 1,555 (51) (41) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Greater China 2,405 1,757 758 540 1,219 1,240 (230) (266) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Asia-Pacific 1,135 939 285 160 600 699 (43) (72) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Latin America 645 406 121 (1) 507 532 (91) (55) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Reportable segments 10,272 7,665 2,404 1,053 6,325 6,415 (601) (609) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Other Businesses 73 68 17 10 52 59 8 (2) -Total 10,345 7,733 2,420 1,063 6,377 6,474 (593) (611) - 1 First half year. -2 At June 30. - -OPERATING PROFIT € IN MILLIONS -First half First half -| | | | | | | | -| --- | --- | --- | --- | ------------ | ------------ | --- | -| | | | | year 2021 | year 2020 | | - -| Operating profit for reportable segments | | | | | 2,404 | 1,053 | -| ---------------------------------------- | --- | --- | --- | --- | -------- | ------ | -| Operating profit for Other Businesses | | | | | 17 | 10 | -| | | | | | | | -| Segmental operating profit | | | | | 2,420 | 1,063 | -| | | | | | | | -| HQ | | | | | (745) | (586) | - -| Central expenditure for marketing | | | | | (373) | (429) | -| --------------------------------- | --- | --- | --- | --- | --------- | --------- | -| Consolidation | | | | | (54) | (263) | -| | | | | | | | -| Operating profit/ (loss) | | | | | 1,248 | (215) | -| | | | | | | | -| Financial income | | | | | 6 | 17 | - -| Financial expenses | | | | | (75) | (82) | -| ------------------ | --- | --- | --- | --- | ------ | ---- | -| | | | | | | | -Income/ (loss) before taxes from continuing operations 1,179 (279) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | - -NET SALES (THIRD PARTIES) € IN MILLIONS -First half First half -| | | | | | | | -| --- | --- | --- | --- | ------------ | ------------ | --- | -| | | | | year 2021 | year 2020 | | - -| Footwear | | | | | 5,766 | 4,495 | -| -------- | --- | --- | --- | --------- | ------- | ------------ | -| Apparel | | | | | 4,037 | 2,812 | -| | | | | | | | -| Hardware | | | | | 542 | 426 | -| | | | | | | | -| Total | | | | 10,345 | | 7,733 | - -| 09 | GOODWILL AND TRADEMARKS | | | | | | -| --- | ------------------------ | --- | --- | --- | --- | --- | -Following the company’s internal management reporting by markets, the number of cash-generating units -decreased to a total of six, effective January 1, 2021. In the first quarter 2021, the number of cash- -generating units further decreased to a total of five as the cash-generating unit North America Reebok is -classified as a disposal group and shown in ‘Assets/liabilities classified as held for sale.’ -Due to the changes in segmental reporting, the carrying amounts of acquired goodwill have been -reallocated to the new groups of cash-generating units in the first quarter 2021 as follows: -29 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ALLOCATION OF GOODWILL -| | | | | | | | Goodwill (€ in millions) | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | --- | - -| | | | | | | | | | (Re-) | | (Re-) | | -| --- | --- | --- | --- | --- | --- | ------------ | --- | ----------- | ------ | ----------- | ------ | --- | -| | | | | | | Aggregation | | allocation | | allocation | | | - Dec. 31, 2020 EMEA Asia-Pacific adidas Golf Jan. 1, 2021 -| Europe | | | | | 593 | | (593) | | | | | n.a. | -| ---------------- | --- | --- | --- | --- | ------ | --- | ----- | --- | ------- | --- | ------- | ------ | -| Emerging Markets | | | | | 76 | | (76) | | | | | n.a. | -| | | | | | | | | | | | | | -| EMEA | | | | | n.a. | | 669 | | | | 25 | 694 | -| | | | | | | | | | | | | | -| North America | | | | | n.a. | | | | | | 77 | 77 | -| | | | | | | | | | | | | | -| Greater China | | | | | n.a. | | | | 269 | | 10 | 280 | -| | | | | | | | | | | | | | -| Asia-Pacific | | | | | 361 | | | | (269) | | 66 | 157 | -| | | | | | | | | | | | | | -| adidas Golf | | | | | 178 | | | | | | (178) | n.a. | -| | | | | | | | | | | | | | -| Total | | | | | 1,208 | | - | | - | | - | 1,208 | -| | | | | | | | | | | | | | - -Due to the change in the composition of the company’s operating segments and associated cash- -generating units respectively, adidas assessed at January 1, 2021 whether goodwill or Reebok trademark -impairments were required. The underlying value drivers and key assumptions for impairment testing -purposes remained in principle unchanged compared to the impairment test performed for the -consolidated financial statements as at December 31, 2020. In this context, there was no need for goodwill -or Reebok trademark impairment. -On February 16, 2021, the company decided to begin a formal process aimed at divesting Reebok. Reebok -is reported as discontinued operations and classified as a disposal group held for sale as at June 30, 2021, -accordingly. As a result, the goodwill allocated to the group of cash generating units Europe, Middle East -and Africa (EMEA), North America, Greater China and Asia-Pacific (APAC) was split and reallocated -between adidas and Reebok cash-generating units based on relative values (fair values), respectively. -RECONCILIATION OF GOODWILL, NET € IN MILLIONS -| | | | | | | | North | | Greater | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | --- | --- | --- | - EMEA America China Asia-Pacific Total -| January 1, 2021 | | | | | 694 | | 77 | | 280 | | 157 | 1,208 | -| --------------------- | --- | --- | --- | --- | ---- | --- | ----- | --- | ------ | --- | ------ | ------ | -| Reebok Disposal Group | | | | | (23) | | - | | (3) | | (1) | (27) | -Currency translation -| | | | | | 11 | | 0 | | 5 | | 2 | 18 | -| --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ---- | --- | -differences -| June 30, 2021 | | | | | 682 | | 77 | | 282 | | 158 | 1,199 | -| -------------- | --- | --- | --- | --- | ----- | --- | ------ | --- | ------- | --- | ------- | ------- | - -In the first quarter 2021, the reallocated goodwill as well as the Reebok trademark in an amount of € 757 -million were initially measured according to IAS 36 ‘Impairment of Assets’ and goodwill allocated to -Reebok as well as the Reebok trademark were subsequently transferred to ’Assets/liabilities classified as -held for sale’ due to the concrete plans to divest Reebok. -| | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -30 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -010 EVENTS AFTER THE BALANCE SHEET DATE -The seven-year € 600 million Eurobond, originally due on October 8, 2021, was already repaid by adidas on -July 8, 2021. -With the approval of the Supervisory Board, the Executive Board of adidas AG has decided to launch a new -share buyback program in the second half of 2021. Starting on July 1, 2021, the company plans to buy back -shares worth up to € 550 million until the end of the year. Until the finalization of these interim -consolidated statements on July 26, 2021, adidas AG purchased a total of 226,850 shares for a total price -of € 72,683,346. -Besides this, there were no other major events between the end of the first half of 2021 and the finalization -of these interim consolidated financial statements on July 26, 2021, that might have a material influence -on the assets, liabilities, financial position, and profit or loss of the company. -Herzogenaurach, July 26, 2021 -The Executive Board of adidas AG -31 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -RESPONSIBILITY STATEMENT -To the best of our knowledge, and in accordance with the applicable reporting principles for interim -financial reporting, the interim consolidated financial statements give a true and fair view of the assets, -liabilities, financial position and profit or loss of the Group, and the interim Group management report -includes a fair review of the development and performance of the business and the position of the Group, -together with a description of the material opportunities and risks associated with the expected -development of the Group for the remaining months of the financial year. -Herzogenaurach, July 26, 2021 -KASPER RORSTED ROLAND AUSCHEL BRIAN GREVY -CHIEF EXECUTIVE OFFICER GLOBAL SALES GLOBAL BRANDS -HARM OHLMEYER AMANDA RAJKUMAR MARTIN SHANKLAND -CHIEF FINANCIAL OFFICER GLOBAL HUMAN RESOURCES GLOBAL OPERATIONS -32 - -CONTACT -ADIDAS AG -ADI-DASSLER-STR. 1 -91074 HERZOGENAURACH -GERMANY -Tel + 49 (0) 91 32 84 – 0 -► ADIDAS-GROUP.COM -adidas ist a member of DIRK -(German Investor Relations Association) -INVESTOR RELATIONS -investor.relations@adidas.com -► ADIDAS–GROUP.DE/INVESTORS -© ADIDAS 2021 -33 \ No newline at end of file +DUET: Joint Exploration of User–Item Profiles in Recommendation System +YueChen1,*,YifeiSun1,*,LuWang2,†,FangkaiYang2,PuZhao2,MinjieHong3,YifeiDong4,MinghuaHe2, +NanHu2,JianjinZhang2,ZhiweiDai2,YuefengZhan2,WeihaoHan2,HaoSun2, +QingweiLin2,WeiweiDeng2,FengSun2,QiZhang2,SaravanRajmohan2,DongmeiZhang2 +1PekingUniversity2Microsoft3ZhejiangUniversity4KTHRoyalInstituteofTechnology +*Equalcontribution†Correspondingauthor +Abstract +| | | | User History: | Item History: | User History: | Item History: | +| --- | --- | --- | -------------------- | ------------------------- | -------------------- | ------------------------- | +| | | | • Misfits (Punk) | • Review: | • Misfits (Punk) | • Review: | +| | | | • I r o n M a iden | " P o l i s he d P o p " | • I r o n M a iden | " P o l i s he d P o p " | +Traditionalrecommendationsystemsrepresent ( M e t al ) • R e v i e w : " G r e a t ( M e t al ) • R e v i e w : " G r e a t +6202 rpA 51 ]RI.sc[ 1v10831.4062:viXra • • Parliament Funk Bass" +| usersanditemsasdensevectorsandlearnto | | | Parliament | Funk Bass" | | | +| ------------------------------------- | --- | --- | ----------- | ------------------ | ------ | ------------------ | +| | | | (Funk) | • Meta: "80s Rock" | (Funk) | • Meta: "80s Rock" | +aligntheminasharedlatentspaceforrelevance +| | | | | Duet | Independent | | +| --- | --- | --- | --- | ----- | ------------ | --- | +estimation. RecentLLM-basedrecommenders (Joint Alignment) Generation +| instead leverage | natural-language | representa- | | | | | +| ---------------- | ---------------- | ----------- | ------------- | ------------- | ------------- | ------------- | +| | | | User Profile: | Item History: | User Profile: | Item History: | +tionsthatareeasiertointerpretandintegrate Enthusiast of A peak-era FUNK- Fan of HEAVY A highly rated POP- +| | | | FUNK and SOUL. | ROCK album with | METAL and | ROCK album with | +| --- | --- | --- | --------------- | ---------------- | ---------- | ---------------- | +withdownstreamreasoningmodules. Thispa- Values technical genre-blurring PUNK. Prefers polished +| | | | mastery | creativity. | dark themes. | production. | +| --- | --- | --- | ------- | ----------- | ------------ | ----------- | +textual +per studies how to construct effective Semantic Alignment ✘ Semantic Mismatch +profilesforusersanditems,andhowtoalign +themforrecommendation. Acentraldifficulty Figure1: DUETalignsrawuseranditemdatabytrans- +| | | | forming them | into textual | profiles within | a shared se- | +| --- | --- | --- | ------------ | ------------ | --------------- | ------------ | +isthatthebestprofileformatisnotknownapri- +manticspace. +ori: manuallydesignedtemplatescanbebrittle +| andmisalignedwithtaskobjectives. | | Moreover, | | | | | +| -------------------------------- | --- | --------- | ------------ | ------------ | -------------------- | --- | +| | | | to introduce | semantically | rich, human-readable | | +generatinguseranditemprofilesindependently +mayproducedescriptionsthatareindividually representationsforrecommendation(Wangetal., +plausible yet semantically inconsistent for a 2025;Zhang,2024;Baoetal.,2023;Hongetal., +| specificuser–itempair. | WeproposeDUET,an | | | | | | +| ---------------------- | ---------------- | --- | -------------------------- | --- | ------------------- | --- | +| | | | 2025a,b,c;Wangetal.,2024). | | Anaturaldirectionis | | +interaction-awareprofilegeneratorthatjointly +toreplacelatentvectorswithtextualuseranditem +producesuseranditemprofilesconditionedon +profilesthatcanbeinspected,edited,andreusedby +| both user history | and item evidence. | DUET | | | | | +| ----------------- | ------------------ | ---- | --- | --- | --- | --- | +downstreamcomponentsincorporatedwithLLMs. +| follows a three-stage | procedure: | it first turns | | | | | +| --------------------- | ---------- | -------------- | --- | --- | --- | --- | +However,existingLLM-basedapproachesremain +rawhistoriesandmetadataintocompactcues, +then expands these cues into paired profile limited in two important ways. First, directly +promptsandthengenerateprofiles,andfinally promptinganLLMwithrawuseranditemhisto- +optimizesthegenerationpolicywithreinforce- +riestoobtainrecommendationsoftenyieldsnoisy +mentlearningusingdownstreamrecommenda- andincompletesignals,especiallywhenhistories +| tionperformanceasfeedback. | | Experimentson | | | | | +| -------------------------- | --- | ------------- | --------- | ------------------------ | --- | ------------- | +| | | | are long, | sparse, or heterogeneous | | (Wang et al., | +threereal-worlddatasetsshowthatDUETcon- +| | | | 2025). Second, | profile-based | methods | typically | +| --- | --- | --- | -------------- | ------------- | ------- | --------- | +sistentlyoutperformsstrongbaselines,demon- +relyonmanuallydesignedtemplatesorhandcrafted +stratingthebenefitsoftemplate-freeprofileex- +attributes,whichrequiressubstantialhumanengi- +plorationandjointuser–itemtextualalignment. +Project code page: https://github.com/duet- neering and constrains the representation space. +review/duet_code. More fundamentally, many approaches generate +userprofilesanditemprofilesindependently,with- +1 Introduction +| | | | out modeling | how user | preferences | and item se- | +| --- | --- | --- | ------------ | -------- | ----------- | ------------ | +Traditional recommendation systems represent manticsinteractatdecisiontime(Yangetal.,2023; +Xietal.,2024). +usersanditemsasdensevectorsandlearntoalign +theminasharedlatentspaceforrelevanceestima- Toaddressthesechallenges,wepropose DUET, +tion (Covington et al., 2016; Wu, 2023). While ajointuser–itemprofilegeneratorthattakesboth +effective,suchembeddingsareopaque: theyoffer userhistoryanditemhistoryasinputandproduces +limitedinterpretabilityandmakeitdifficulttoan- a paired set of profiles for the interaction. Cru- +alyzewhyanitemisrecommended. Recentwork cially, DUET doesnotrequireprofiletemplates: it +thereforeleverageslargelanguagemodels(LLMs) istrainedwithreinforcementlearningusingfeed- + +back from downstream recommendation perfor- asseeninworkslikeCRESDUP(Chenetal.,2007) +mance,enablingittoexploreanddiscovereffective andUPCSim(Widiyaningtyasetal.,2021). While +profileformatsautomatically. Figure1illustrates foundational,thesemethodswerelimitedbytheir +whyjointprofilingmatters. Theuserhaslistened rigid,hand-engineeredfeatures. Morerecently,the +toMisfits(punk),IronMaiden(metal),andParlia- adventofLargeLanguageModels(LLMs)hasen- +ment(funk),whilethecandidatealbumisdescribed abledashifttowardsgeneratingprofilesinnatural +byreviewssuchas“polishedpop”and“greatfunk language. Studies such as KAR (Xi et al., 2024), +bass”andametatag“’80srock.” Consideringboth GPG(Zhang,2024),PALR(Yangetal.,2023),and +sides together, DUET can reconcile these signals LettinGo (Wang et al., 2025) leverage LLMs to +intoacompatibleinterpretation(e.g.,highlighting createtextualuserprofilesfrombehavioraldata. A +the user’s funk/soul affinity and the item’s funk- keylimitationoftheseapproachesistwofold: they +rockcharacter). Incontrast,independentlygener- relyonstaticorpre-definedtemplatesforprofile +atedprofilesmayamplifydifferentfacets,summa- generation,andtheyoftenfocusexclusivelyonuser +rizingtheuseras“heavymetal/punk”whilesum- profiles,neglectingtherich,expressiveinformation +marizingtheitemas“pop-rock”,whichresultingin inherentinitemsandthecomplexdynamicsofuser- +asemanticallymismatchedpairthatobscuresthe iteminteractions. Incontrast,ourworkintroduces +truerelevancesignal. anewparadigmwherebothuseranditemprofiles +DUETproceedsinthreestages. First,rawhisto- arenotfixedbutaredynamicallyexploredandop- +ries and metadata are distilled into minimal cues timizedinasharedsemanticspacetodirectlyalign +thatcapturecompactbutinformativesignals. Sec- withrecommendationperformance. +ond, the model expands cues into richer profile +prompts to generate textual profiles, allowing ex- 2.2 ReinforcementLearningforLLM-Based +plorationoveralternativeprofilestructuresandem- RecommendationSystems +phases. Third,theresultingprofilesareconsumed +ReinforcementLearning(RL),particularlythrough +bydownstreamrecommenders,andtheirtaskfeed- +techniqueslikeRLHF,hasbecomeacoremethod +backisusedtooptimizetheprofilegenerationpol- +for aligning LLMs with specific objectives. This +icy. By coupling both sides in a shared semantic +approach has been adapted for recommendation +space,DUETlearnsuserprofilesthatreflectwhat +tasks (Wang et al., 2025; Lin et al., 2025; Deng +kindsofitemsauserprefersanditemprofilesthat +etal.,2025),butexistingeffortsoftenfacekeylimi- +reflectwhatkindsofusersanitemappealsto. +tations. Theyfrequentlyrelyonofflinerewardmod- +Ourcontributionsareasfollows: +els(Jeongetal.,2023;Chenetal.,2025;Heetal., +• Werepresentusersanditemsasnatural-language 2025; Liu et al., 2025) that do not adapt in real- +profiles and align them in a shared semantic timetosystemfeedback,asetupthatrisksissues +space,extendingtheclassicvector-basedalign- like reward hacking (Skalse et al., 2025). Other +mentprincipletointerpretabletextualrepresenta- methods(Sunetal.,2024;Luetal.,2024)restrict +tions. themselvestoofflinepreferencetuning(e.g.,DPO), +• We introduce an exploration-based framework which can easily overfit on static datasets. Our +thatstartsfromcue-basedinitialization,expands framework, DUET,overcomesthesechallengesby +cuesintocandidateprofiles,andjointlyoptimizes integrating RL into a closed-loop system where +user and item profiles with downstream recom- downstreamrecommendationperformanceserves +mendationfeedbackviareinforcementlearning, asthereal-timerewardsignal,allowingforthedy- +avoidingrigidtemplates. namicandinteractiverefinementoftextualprofiles. +• Extensiveexperimentsacrossmultiplereal-world +datasets show that DUET consistently outper- 3 Method +formsstrongbaselines,validatingbothjointpro- +filingandfeedback-drivenprofileoptimization. DUET is a closed-loop framework that trans- +forms raw user–item interaction histories into +2 RelatedWork performance-aligned textual profiles through +learnedrepresentationstrategies. AsshowninFig- +2.1 ProfilesinRecommendation +ure 2, DUET consists of three stages. (1) Cue- +Earlyrecommendationsystemsprimarilyreliedon Based Initialization distills interaction histories +pre-definedprofilesbasedonstructuredattributes, intoconciseevidence-basedcues. (2)JointExplo- + +CUE-BASED INITIALIZATION JOINT EXPLORATION ON-POLICY OPTIMIZATION +| | | | | Item | DU | | | Duet | | | | +| ----------- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | --- | +| ItemHistory | | | | | ET | | | | | | | +Initial CUE +Rated: Baba Is You (5★) +User: “Prefers retro puzzle games” +Reviewed: “I love logic- +Item: “Retro-style indie puzzle game” +| | | | based puzzle games.” | | | Profile Prompt | | | | | | +| --- | --- | --- | -------------------- | --- | --- | -------------- | --- | --- | --- | --- | --- | +Rated: The Witness (5★) +User: “Describe the user's typical +| | | | Rated: Fez (4★) | | puzzle-game preferences … | | | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒1 𝑃𝑟𝑜𝑓𝑖𝑙𝑒2 | … | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒𝑛 | | +| --- | --- | --- | --------------- | --- | ------------------------- | --- | --- | ----------------- | --- | -------- | --- | +Item: “… item's core mechanics… +| | | | | | | | Profile | Downstream Task Env | | | | +| ----------- | --- | --- | ----------------------- | ---- | --------------------------------- | --- | ------- | --------------------- | --- | --- | --- | +| | | | | User | User: “A player who enjoys logic- | | | | | | | +| UserHistory | | | | | intensive indie puzzle games, | | | | | | | +| | | | Category: Indie Puzzle | | | | | Ranking correctness | | | | +prefers retro visual aesthetics…” +| | | | Game | | | | | &format Reward | | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Reviews: +Item: “A retro-style indie puzzle +| | | | “Challenging logic | | game featuring challenging | | | Group Computation | | | | +| --- | --- | --- | -------------------- | ------------------------ | ------------------------------- | --- | --- | ------------------- | --- | ---- | --- | +| | | | mechanics.” | | logic mechanics that appeal to | | | | | | | +| | | | “ R e t r o p ix | e l- a rt st y l e . ” | hardcore puzzle players.” | | | 𝐴𝑑𝑣1 𝐴𝑑𝑣2 | | | | +| | | | “ H a r d b u | t re w a r d i n g . ” | | | | | … | 𝐴𝑑𝑣𝑛 | | +| | | | | Figure2: | OverviewoftheDUETframework. | | | | | | | +ration via Adaptive Profile Prompt Discovery 3.2 Cue-BasedInitialization +| jointly | explores | user-item’s | | profile prompts | that | | | | | | | +| ---------- | -------- | ----------- | -------- | --------------- | -------- | -------- | --------- | -------- | -------- | --- | --------- | +| | | | | | | Raw user | histories | and item | metadata | | are often | +| define how | user | and item | profiles | should | be writ- | | | | | | | +noisy,redundant,andnotdirectlysuitableforpro- +ten. (3)OptimizationviaOn-policyExploration +| | | | | | | fileconstruction. | | Toaddressthis,DUETintroduces | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ---------------------------- | --- | --- | --- | +jointlyoptimizesuseranditemprofilesunderdown- +theconceptofcues: +concisehypothesesthatsum- +streamrecommendationfeedback. +| | | | | | | marize minimal | | but informative | | aspects | of users | +| ------------------- | ------ | --- | ------------------------- | ------- | -------- | -------------- | ------------ | --------------- | -------------- | ------- | -------- | +| All three | stages | are | realized | through | a single | | | | | | | +| | | | | | | and items. | These | cues act | as lightweight | | seeds, | +| passinputandoutput: | | | cueextraction,self-prompt | | | | | | | | | +| | | | | | | which are | deliberately | underspecified | | so | that the | +construction,andprofilegenerationareproduced +systemcansubsequentlyexplorericherprofilefor- +inasinglesequence-to-sequencegenerationpass +mats. +atinferencetime,enablingefficientdeployment. +| | | | | | | Definition1(Cue). | | Acueisaminimaltextualhy- | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------------------ | --- | --- | --- | +3.1 ProblemFormulation +pothesisderivedfromhistoricaldatathathighlights +WeformulateprofilegenerationinDUETasanon- one potential aspect of a user’s preference or an +policyreinforcementlearningproblem,motivated item’scharacteristic. Ratherthanaimingforcom- +bytheabsenceofanytextualgroundtruthdefining pleteness, cues capture partial but salient signals +an optimal user or item profile. Profile quality is thatserveasstartingpointsforprofileexploration. +evaluatedsolelybyitsfunctionalutilityinafixed +| | | | | | | To extract | cues | automatically, | | the | LLM is | +| --- | --- | --- | --- | --- | --- | ---------- | ---- | -------------- | --- | --- | ------ | +recommendationenvironment. +| | | | | | | prompted | to summarize | minimal | | but informative | | +| ---- | ---------- | --- | ------------ | ------ | ------- | ------------------------ | ------------ | ------- | ----------------- | --------------- | --- | +| DUET | is modeled | as | a generative | policy | π θ in- | | | | | | | +| | | | | | | aspectsofuseroritemdata. | | | Forexample,givena | | | +teractingwithafrozendownstreamrecommender. +user’sinteractionhistory,themodelisguidedwith +| For each | user–item | pair, | the | state is | defined as | | | | | | | +| -------- | --------- | ----- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | +instructionssuchas: +| s = {H | ,H },whereH | | andH | denotetheuser | | | | | | | | +| ------ | ----------- | --- | ---- | ------------- | --- | --- | --- | --- | --- | --- | --- | +| | u i | | u | i | | | | | | | | +and item interaction histories. An action corre- CueExtractionPrompt +spondstoasingle-passjointgeneration +“Fromthehistorybelow,analyzetheuser’s +a = {C u ,S u ,P u ,C i ,S i ,P i }, historicalinteractionstounderstandprefer- +ences,ratingbehavior,reviewsentimentor +| where | (C ,C | ) are | cues | distilled | from his- | | | | | | | +| ----- | ----- | ----- | ---- | --------- | --------- | --- | --- | --- | --- | --- | --- | +u i +| | | | | | | anyotherdimension. | | Keepthedescription | | | | +| -------- | ------------ | ----------- | --- | ----------- | ------- | ------------------ | --- | ------------------ | --- | --- | --- | +| tory, (S | u ,S i ) are | constructed | | user-item’s | profile | | | | | | | +prompts,and(P ,P )arethefinaltextualprofiles. conciseandavoidfullsentences.” +u i +| Thepolicyπ | | (a | s)definesajointdistribution | | | | | | | | | | +| ---------- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +θ +over the entire generation sequence. As an on- This lightweight guidance allows the LLM to +policy agent, DUET is optimized using rewards maprawhistoriesandmetadataintocompacttex- +from its own sampled generations rather than by tualcues. Thedetaildexampleofcuecanbefound +| imitatingfixedsummaries. | | | | | | inAppendixA.3. | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | + +| 3.3 | JointExplorationviaAdaptiveProfile | | | | | | Single-PassInput | | | | | | +| --- | ---------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +PromptDiscovery Task:Generatestructuredprofilesfortheuseranditemfrom +historicaldata. +Aftercueextraction,DUETdoesnotdirectlysum- Combined Context: [User History H u ] + [Item +| | | | | | | | History | H i ]+[Avg | Ratings] | | | | +| ------------------------------------ | --- | --- | --- | --- | --- | ---------- | ------- | ---------- | -------- | --- | --- | --- | +| marizeuserinterestsoritemattributes. | | | | | | Instead,it | | | | | | | +“First,extractaconciseCueforbothuseranditem.Second, +| explores | | the space | of profile | construction | | strate- | | | | | | | +| -------- | --- | --------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | +basedonthecues,constructaProfilePromptthatdefines +gies—natural-languagepromptsthatdefinethefor- thedescriptionlogic.Finally,generatetheProfileguidedby +theconstructedProfilePrompt.” +mat,abstractionlevel,andattributeselectionlogic +↓DUETForwardPass↓ +| usedtogenerateprofiles. | | | | Explorationistherefore | | | | | | | | | +| -------------------------------------------- | --- | --- | --- | ---------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| performedoverhowprofilesshouldbeconstructed, | | | | | | | Single-PassOutput | | | | | | +User +ratherthanoversuperficialtextualparaphrases. +Cue:“prefersnostalgiclogic-basedchallenges” +For each user–item pair, DUET intro- Profileprompt: “Focuson1990svisualaestheticsand +| duces | an | explicit | intermediate | | variable, | the | strategicdepth.” | | | | | | +| ----- | --- | -------- | ------------ | --- | --------- | --- | ---------------- | --------- | --- | ----- | ----------- | ------------ | +| | | | | | | | Profile: | “A player | who | seeks | retro-style | visual charm | +constructed_prompt,whichservesasadiscrete +pairedwithdeepstrategicreasoning.” +| and | interpretable | | profile | prompt | S. | This profile | | | | | | | +| --- | ------------- | --- | ------- | ------ | --- | ------------ | --- | --- | --- | --- | --- | --- | +Item +promptisashortnatural-languageinstruction(e.g., +Cue:“retro-styleindiepuzzlewithhighdifficulty” +“Describetheuser’stypicalgamingpreferencesand +| | | | | | | | Profile | prompt: | “Describe | | pixel-art | graphics and | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --------- | --- | --------- | ------------ | +engagementpatterns”)thatspecifieswhataspects intellectualdifficultytomatchlogicpreference.” +Profile:“A2Dexperiencefeaturingpixelatednostalgia +| to | describe | and | how to | organize | them, | without | | | | | | | +| --- | -------- | --- | ------ | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +andchallengingmechanicsthatdemandlogicaldeduction.” +| containingtheprofilecontentitself. | | | | | | Conditioned | | | | | | | +| ---------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +onS,themodelgeneratesthefinaluseranditem Figure3: Single-passgenerationinDUET: cueextrac- +| profiles. | | | | | | | tion,profileprompt(constructedprompt),andprofile | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | --- | --- | +generationareproducedinonepassforbothuserand +| Exploration | | | is driven | by treating | | the profile | | | | | | | +| ----------- | --- | --- | --------- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +item. +| promptS | | asastochasticactionsampledfromthe | | | | | | | | | | | +| ------- | --- | --------------------------------- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| policyπ | | (S | Cue). | Duringtraining,theagentsam- | | | | | | | | | | +θ +plesdifferentprofilepromptinstructions,executes +| | | | | | | | andintellectualdifficulty”tomatch. | | | | | RLreinforces | +| ---------------------------- | --- | --- | --- | --- | ---------------- | --- | ---------------------------------- | --- | --- | --- | --- | ------------ | +| themtogenerateprofilepairs(P | | | | | ,P ),andreceives | | | | | | | | +u i +thissharedsemanticdirection,suppressingirrele- +| a reward | | based | on downstream | | recommendation | | | | | | | | +| --------- | --- | ----------------------------------- | ------------- | --- | -------------- | --- | ------------ | ------------- | -------- | --- | ----- | ---------------- | +| | | | | | | | vant signals | and | forcing | the | final | profiles to con- | +| accuracy. | | Thepolicyisoptimizedtoreinforcepro- | | | | | | | | | | | +| | | | | | | | verge | into a shared | semantic | | space | of nostalgia | +filepromptsthatconsistentlyyieldhigherrewards, +andlogic,significantlyimprovingrecommendation +enablingthemodeltoactivelysearchforeffective +accuracy. +profileconstructionformats. +Thisprocessisrealizedasaunifiedgeneration +3.4 OptimizationviaOn-policyExploration +pass, +| | | | | | | | On-policy | optimization. | | | We train | the profile | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | --- | -------- | ----------- | +O = [ Cue → profilepromptS → Profile], generator π in an on-policy manner against a +| | (cid:124)(cid:123)(cid:122)(cid:125) | | | | | (cid:124) (cid:123)(cid:122) (cid:125) | | θ | | | | | +| --- | ------------------------------------ | --- | --------- | ------------------ | --------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | (cid:124) | (cid:123)(cid:122) | (cid:125) | | | | | | | | +Context ProfileConstructionPrompt Execution frozendownstreammodelf, whichservesasthe +| | | | | | | (1) | environment | critic. | For | each | sampled | user–item | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | --- | ---- | ------- | --------- | +wheretheprofilepromptS parameterizesthepro- pair (u,i), the policy generates (P ,P ) and re- +u i +filegenerator. Atinferencetime,DUETcollapses ceivesascalarrewardbasedonthepredictionac- +thelearnedprofilepromptdistributionandgreed- curacy of f(P ,P ). The policy parameters are +u i +ilyexecutestheoptimalprofilepromptinasingle updatedtoreinforcegenerationsthatleadtolower +| forwardpass,introducingnoadditionallatency. | | | | | | | predictionerror. | | | | | | +| ------------------------------------------- | --- | ---------- | ------ | ------------ | -------- | ------- | --------------------------- | --- | --- | --- | --- | ------------- | +| Example | | of profile | prompt | Exploration. | | As | | | | | | | +| | | | | | | | Continuousfractionalreward. | | | | | Usingdiscrete | +| illustrated | | in Figure | 3, | DUET | resolves | this by | | | | | | | +integerratingsasrewardsleadstosparseandunsta- +| identifying | | a shared | context | in | a single | forward | | | | | | | +| ----------- | --- | -------- | ------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +blefeedback,asnear-misspredictions(e.g.,predict- +pass. ThepolicyextractsacoordinatedCue(e.g., +ing4foragroundtruth5)receivethesamepenalty +“prefersretropuzzlegames”)andconstructsapro- +| | | | | | | | assevereerrors. | | Toprovidedenseandinformative | | | | +| ---- | ------ | ---- | ------------------- | --- | --- | -------- | --------------- | --- | ---------------------------- | --- | --- | --- | +| file | prompt | (the | constructed_prompt) | | | that in- | | | | | | | +feedback,wedefineacontinuousfractionalreward: +structsthemodeltohighlight“nostalgicaesthetics +| and | strategic | logic” | for | the user, | while | simulta- | | | | | | | +| ------- | ---------- | ------ | ---------- | ---------- | ----- | -------- | --- | ------- | --- | ---- | ------ | -------- | +| | | | | | | | | | | | |y −yˆ | | | +| | | | | | | | | R (u,i) | | = 1− | ui | ui , (2) | +| neously | describing | | the item’s | “pixel-art | | graphics | | perf | | | | | +M + +where yˆ ui = f(P u ,P i ) is the predicted relevance ingsettingbasedondownstreampredictedrating +score and M is the maximum rating gap (e.g., scores. For each observed user–item interaction, +M = 4 for a 1–5 scale). This reward provides we randomly sample nine items that the user has +fine-grainedgradientsthatencourageincremental not interacted with to form a candidate set of ten +improvementsinrecommendationaccuracy. items, which are ranked in descending order ac- +We adopt Group Relative Policy Optimization cordingtotheirpredictedratings. Theground-truth +(GRPO) (DeepSeek-AI et al., 2025) to optimize interacted item is treated as the only positive in- +theprofilegeneratorundertheabovereward. The stance,andweadoptNDCG@Kastheevaluation +downstreamrecommenderf isfrozenthroughout metric, with K set to 1, 5, and 10, to assess how +training, which yields a stable policy optimiza- effectivelythelearnedprofilessupportcorrectitem +| tionsettingandpreventsrewarddriftorfeedback- | | | | ranking. | | | | | +| -------------------------------------------- | --- | --- | --- | --------- | ---------- | ---------- | ---- | ------- | +| | | | | Baselines | We compare | our method | with | several | +inducedrepresentationcollapse. +| | | | | representative | baselines. | 10H | directly | uses the | +| --- | --- | --- | --- | -------------- | ---------- | --- | -------- | -------- | +4 Experiment mostrecentinteractionhistoriesforpredictionwith- +| | | | | outconstructingexplicitprofiles. | | | KAR(Xietal., | | +| --- | --- | --- | --- | -------------------------------- | --- | --- | ------------ | --- | +4.1 ExperimentalSettings +2024)augmentsrecommendationmodelswithex- +Datasets. Experiments were conducted on three ternalreasoningknowledgeaboutuserpreferences +widelyusedreal-worlddatasets. AmazonMusic andfactualknowledgeaboutitemsextractedfrom +(Music) and Amazon Book (Book) are derived LLMs,whicharetransformedintotask-compatible +fromtheAmazonProductdataset1,whileYelpis representations. RLMRec(Renetal.,2024)lever- +fromtheYelpOpendataset2. Alldatasetsinclude agesLLMstolearnsemanticuseranditemrepre- +userreviews,ratings,andrichtextualinformation. sentationsfromtextualsignalsandalignsthemwith +WeusedthefullAmazonMusicdataset,butonly collaborativerelationalinformationthroughcross- +subsets (latest two months for Book and six for viewrepresentationlearning. PALR(Yangetal., +Yelp). Data was split by timestamp into training, 2023)fine-tunesalargelanguagemodelasarank- +validation,andtestsetstopreventinformationleak- ing component that selects preferred items from +age(Jietal.,2023). retrievedcandidatesexpressedinnaturallanguage. +EvaluationMetrics LG (LettinGo) (Wang et al., 2025) explores di- +For each observed user–item interaction (i.e., verseuserprofilecandidateswithLLMsandaligns +a review record), we construct the evaluation profilegenerationwithdownstreamrecommenda- +instance based on the user’s interaction history tionperformanceviapreferenceoptimization. Rea- +strictly prior to the corresponding timestamp. son4Rec(Fangetal.,2025)introducesadelibera- +tiverecommendationframeworkthatincorporates +Specifically,theuser’srecenthistoricalinteractions +beforethecurrentinteractionareusedtogenerate explicitstep-wisereasoningoveruserpreferences +auserprofile,whilehistoricalreviewsfromother toguideratingprediction. +| users are used | to generate | an item profile. | The | | | | | | +| -------------- | ----------- | ---------------- | --- | --- | --- | --- | --- | --- | +4.2 MainResults +downstreamrecommendationsystemthenpredicts +theratingofthetargetitemconditionedontwoin- Table 1 presents a comparison of our proposed +puts: thegenerateduserprofile,andthegenerated method against five baselines on three datasets: +| | | | | Amazon | Music, Amazon | Books, | and | Yelp. We | +| ------------ | ------------------------------- | --- | --- | ------ | ------------- | ------ | --- | -------- | +| itemprofile. | Anexceptionisthe10Hbaseline,for | | | | | | | | +which the downstream model directly consumes use Qwen3-8B (Team, 2025) and LLaMA3-8B +therawrecentinteractionhistories. (Dubey et al., 2024) as both the profile generator +andthedownstreamrecommendationmodel,with +| We evaluate | performance | using four | widely | | | | | | +| --------------- | ----------- | ------------------ | ------- | ------------ | ----------- | ----- | ----------- | ------- | +| | | | | a prediction | temperature | of 0. | In addition | to rat- | +| adopted metrics | (Wang | et al., 2025; Fang | et al., | | | | | | +2025): MeanAbsoluteError(MAE),RootMean ingpredictionresults,wealsoreportranking-based +SquareError(RMSE),Accuracy,andF1score. evaluation results in Table ??, which assess the +effectivenessofthelearnedprofilesfromadown- +Inadditiontoratingprediction,wefurtherevaluate +thegenerateduseranditemprofilesunderarank- streamrankingperspective. +| | | | | Overallsuperiorityoverstrongbaselines. | | | | As | +| --- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | +1https://cseweb.ucsd.edu/~jmcauley/datasets/ shown in Table 1, our method consistently out- +amazon/links.html +| | | | | performs | the strongest | baselines | across | all three | +| --- | --- | --- | --- | -------- | ------------- | --------- | ------ | --------- | +2https://business.yelp.com/data/resources/ +| open-dataset/ | | | | datasetsunderbothQwen3-8BandLLaMA3-8B. | | | | | +| ------------- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | + +Method Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) +(%) (%) (%) +Qwen3(8B) +10H 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +KAR(Xietal.,2024) 0.7396 1.2184 55.34 48.67 0.7483 1.1380 58.65 60.29 0.7098 1.0923 56.17 58.78 +RLMRec(Renetal.,2024) 0.8197 1.3312 47.15 42.46 0.7438 1.1069 54.89 57.65 0.7812 1.1584 52.86 55.93 +PALR(Yangetal.,2023) 0.7994 1.2876 48.53 43.19 0.6075 0.9531 57.35 56.77 0.7485 1.1187 54.24 56.38 +LG(Wangetal.,2025) 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +R4Rec(Fangetal.,2025) 0.7028 1.1523 55.69 47.73 0.5654 0.9635 58.69 54.67 0.6397 1.0098 58.47 56.84 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +LLaMA3(8B) +10H 1.0864 1.9532 22.09 27.30 0.7917 1.3346 38.13 46.87 0.8064 1.3866 37.15 45.27 +KAR(Xietal.,2024) 0.6427 1.1668 54.51 47.98 0.5726 0.9033 57.53 59.92 0.5892 0.9614 55.87 58.21 +RLMRec(Renetal.,2024) 0.7428 1.3572 46.74 42.11 0.6076 0.9886 53.78 57.42 0.6226 0.9477 52.12 55.79 +PALR(Yangetal.,2023) 0.7238 1.3265 47.72 43.29 0.5823 0.9222 56.73 59.31 0.5977 0.8855 55.06 57.62 +LG(Wangetal.,2025) 0.6196 1.1289 56.03 51.24 0.5204 0.9369 61.92 59.50 0.5543 0.7967 58.95 60.39 +R4Rec(Fangetal.,2025) 0.7586 1.0418 55.80 53.00 0.5442 0.7722 60.86 54.88 0.6029 0.8345 59.70 56.35 +Ours 0.5367 0.9687 60.87 54.74 0.4680 0.8277 63.30 60.60 0.5092 0.9500 63.42 58.12 +Table1: PerformanceonthreedatasetsusingQwen3(8B)andLLaMA3(8B). +Method Yelp AmazonMusic AmazonBooks +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +Table2: RankingperformanceunderEASE-based(Steck,2019)hardnegatives. +Under Qwen3-8B, our method achieves an accu- our approach reaches an NDCG@1 of 0.5619 +racyof61.23%onYelp,67.96%onAmazonMu- and an NDCG@10 of 0.7443, clearly outper- +sicand64.38%onAmazonBooks,surpassingLG forming the strongest baseline methods. Similar +(LettinGo)(Wangetal.,2025)by5.05%,5.59% trendsareobservedonAmazonMusic,whereour +and5.03%,respectively. Similarimprovementsare methodattains0.5347atNDCG@1and0.7331at +observed under LLaMA3-8B, indicating that the NDCG@10, as well as on Amazon Books with +gainsarestableacrossbackbonemodels. NDCG@1 of 0.4866 and NDCG@10 of 0.7107. +Theseresultsindicatethatthelearnedprofilesen- +Advantages over fixed-structure and delib- +able more accurate identification and ordering of +erative baselines. Compared with strong fixed- +relevantitemswithincandidatesets,leadingtosu- +structureordeliberativemethodssuchasKAR(Xi +perior ranking quality in downstream recommen- +etal.,2024)andReason4Rec(Fangetal.,2025), +dation. Wefurtherevaluaterankingperformance +our approach consistently achieves lower predic- +under a more challenging setting using EASE- +tion error and higher accuracy. For example, on +based (Steck, 2019) hard negatives, and observe +Amazon Books with Qwen3-8B, our method re- +consistentimprovements;detailedresultsarepro- +duces MAE to 0.4612, compared to 0.7098 for +videdinAppendixB.2. +KARand0.6397forReason4Rec,demonstrating +moreeffectiveabstractionofuserpreferenceswith- +4.3 AblationStudy +outrelyingonpredefinedreasoningtemplates. +Consistent improvements in downstream Effectivenessofprofilegeneration,cue&strategy, +ranking. Table??reportstherankingresultsun- anduser–itemjointoptimization. Table3reports +der the Qwen3-8B backbone. Our method con- the ablation results under different design con- +sistentlyachievesthebestperformanceacrossall figurations. Starting from the history-only base- +three datasets and evaluation cutoffs. On Yelp, line (10H), introducing explicit profile genera- + +| Method | | | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | --- | ---- | --- | ------ | ----------- | --- | --- | ----------- | ------ | +| | | | MAE | RMSE | Acc | F1 MAE | RMSE | Acc | F1 | MAE RMSE | Acc F1 | +10H(HistoryOnly) 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +Profile 0.7218 1.1863 55.48 48.09 0.6597 1.0218 58.67 57.48 0.6764 1.0469 57.14 57.68 +Profile+Cue&Strategy 0.7085 1.1654 55.83 48.54 0.5708 0.9897 58.91 55.53 0.6389 1.0108 58.43 56.88 +Profile+JointOpt.(LG(Wang 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +etal.,2025)) +Profile+Cue&Strategy+ 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +JointOpt.(Ours) +Table3: AblationstudyondifferentdesignconfigurationsinDUETusingQwen3(8B). +| Method | | Yelp | | | | AmazonMusic | | | | AmazonBooks | | +| ------ | -------- | ---- | --- | ----- | --- | ----------- | --- | ----- | --- | ----------- | --------- | +| | MAE RMSE | | Acc | F1(%) | MAE | RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | +| | | | (%) | | | | (%) | | | | (%) | +10H+30P 0.5126 0.9485 61.23 55.18 0.3883 0.7494 67.96 63.89 0.4612 0.9089 65.13 59.97 +10H+50P 0.4909 0.9207 62.43 56.24 0.3924 0.7543 67.88 63.87 0.4553 0.9023 64.62 59.52 +10H+70P 0.4987 0.9326 61.98 55.81 0.3937 0.7564 68.22 64.12 0.4608 0.9068 64.38 59.27 +Table4: Impactofhistoricalinteractionlengthonprofilequality(usingQwen3(8B)). +tionyieldssubstantialperformanceimprovements jointoptimizationarecombined. Underthisfull +across all datasets. For example, on Yelp, MAE configuration, accuracy reaches 61.23% on Yelp, +isreducedfrom1.1235to0.7218andaccuracyin- 67.96%onAmazonMusic,and64.38%onAma- +creases from 23.17% to 55.48%, confirming that zon Books, with corresponding MAE values of +textualprofilesprovidesignificantlymoreinforma- 0.5126, 0.3937, and 0.4612, respectively. Com- +tiverepresentationsthanrawinteractionhistories. pared with the LettinGo-style configuration, this +| | | | | | | setting | yields | consistent | | improvements | across all | +| ------ | ---------------- | --- | ----- | --- | ------ | --------- | ------ | -------------------------------------- | --- | ------------ | ---------- | +| Adding | the cue&strategy | | layer | on | top of | pro- | | | | | | +| | | | | | | datasets. | | Overall,theablationresultsindicatethat | | | | +filegenerationleadstomodestbutconsistentgains. +whileprofilegenerationandjointoptimization +OnAmazonMusic,MAEfurtherdecreasesfrom +0.6597to0.5708,whileaccuracyslightlyimproves contributesubstantiallytoperformancegains,inte- +gratingcue&strategydiscoveryfurtherenhances +| from58.67%to58.91%. | | | Althoughthenumerical | | | | | | | | | +| ------------------- | --- | --- | -------------------- | --- | --- | --- | ------------- | --- | -------- | --------- | ------------ | +| | | | | | | the | effectiveness | | of joint | user–item | optimization | +improvementsintroducedbystrategyalonearelim- +ited, these results suggest that strategy discovery byprovidingastructuredspaceforexplorationand +refinement. +primarilyreshapeshowpreferenceinformationis +abstractedandexpressed,ratherthandirectlyopti- Impactofhistoricalinteractionlengthonprofile +mizingpredictionaccuracyinisolation. +| | | | | | | quality. | | Table | 4 analyzes | the | impact of histori- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ----- | ---------- | --- | ------------------ | +Whenenablingjointoptimizationwithoutstrat- cal interaction length on profile quality. Across +thethreedatasets,varyingthenumberofhistorical +| egy (i.e., | the LettinGo | (Wang | et | al., 2025)-style | | | | | | | | +| ---------- | ------------ | ----- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | +configuration),performanceimprovesmorenotice- interactions leads to only moderate performance +ablyacrossdatasets. OnAmazonMusic,accuracy differences, indicating that our method does not +| | | | | | | strongly | | depend | on long | histories. | In particular, | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------ | ------- | ---------- | -------------- | +increasesto62.37%,comparedto58.67%withpro- +filegenerationalone. Inourimplementation,this usingamoderatehistorylength(e.g.,30–50inter- +settingcorrespondstoareproductionofLettinGo, actions)alreadyachievescompetitiveorbestper- +whereprofilegenerationandjointoptimizationare formanceonmostmetrics,whilefurtherincreasing +thehistorylengthprovideslimitedadditionalgains. +| appliedwithoutthestrategylayer. | | | | Whiletheorig- | | | | | | | | +| ------------------------------- | ------- | --------- | --- | ------------- | ---------- | --- | --- | --- | --- | --- | --- | +| inal method | focuses | primarily | on | user | profiling, | | | | | | | +OnYelpandAmazonBooks,extendingthehis- +weextendtheoptimizationtobothuseranditem +torylengthbeyondthisrangedoesnotconsistently +profilestoensureafaircomparison. improveaccuracyorF1scoreandmayslightlyde- +Thestrongestandmostconsistentperformance gradeperformance,suggestingthatexcessivehis- +is achieved when cue&strategy discovery and toricalinteractionscanintroducenoisyorlessrele- + +vantsignals. Incontrast,AmazonMusicexhibits maintaining comparable coverage. Compared to +relativelystableperformanceacrossdifferenthis- existing methods, which either rely on extractive +torylengths,indicatingthatuserpreferencesinthis compression(e.g.,KAR(Xietal.,2024))orgener- +domainarelesssensitivetohistorytruncation. atefree-formdescriptionswithweakergrounding +Necessity of RL-based Optimization. We com- (e.g.,RLMRec(Renetal.,2024)andPALR(Yang +pare the full model with a variant without RL etal.,2023)), DUET consistentlyattainsstronger +(DUET W/O RL) to isolate the effect of RL. As alignment without sacrificing coverage. For in- +showninTable5,removingRLleadstosubstantial stance,LG(Wangetal.,2025)achievesrelatively +performancedegradationacrossalldatasets(e.g., highalignmentbutexhibitslessstablegrounding +YelpAccuracydropsfrom61.23%to48.53%),indi- acrossdatasets,whileRLMRecattainshighercov- +catingthatthegainscannotbeattributedtoprompt erageatthecostofweakersemanticalignment. In +| designalone. | | | | | contrast,DUETmaintainsbothhighalignmentand | | | | | | +| ------------ | --- | --- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | +Without RL, the generator reduces to a static mid-to-highcoverage,suggestingamoreeffective +mapping from interaction history to textual pro- balancebetweensemanticabstractionandevidence +| files,lackingadaptiveselectionofrelevantsignals. | | | | | preservation. | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | +Incontrast,RLenablesoptimizingprofileconstruc- +4.5 CaseStudy +tionunderrewardfeedback,resultinginmoredis- +Figure4presentsarepresentativecasestudythat +| criminative | representations.These | | results | demon- | | | | | | | +| ----------- | --------------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | +strate that RL-based optimization is essential for demonstrates how semantically aligned user and +| DUET | . | | | | itemprofiles | enableaccurate | | prediction | | thattran- | +| ---- | --- | --- | --- | --- | ------------ | -------------- | --- | ---------- | --- | --------- | +scendsthelimitationsofsparserawinteractionhis- +4.4 SemanticAnalysisofGeneratedProfiles tories. In the user profile, initially scattered and +To better understand the source of performance fragmentedpreferencecuesaresystematicallydis- +gains, we analyze whether the generated profiles tilled into a stable and coherent preference struc- +exhibitmeaningfulsemanticpropertiesratherthan turefavoringfunk,soul,andprogressiverock(or- +servingasintermediatetextualartifacts. Weintro- ange), coupled with a pronounced emphasis on +duce two complementary metrics to characterize musicalcomplexityandhistoricalsignificance(pur- +semanticcompatibilityandgrounding. ple). Thecorrespondingitemprofileexhibitsstrik- +Semantic Alignment. We measure the ingsymmetry,characterizingthealbumasanexem- +embedding-levelsimilaritybetweengenerateduser plaryfunk-rockcomposition(orange)andposition- +anditemprofilesusingall-mpnet-base-v2from ingitasadefiningandculturallyinfluentialrelease +Sentence-Transformers (Reimers and Gurevych, of the 1970s (purple). This example shows 519 +2019). Foreachuser–itempair,wecomputecosine that the learned profiles capture meaningful pref- +| similarity: | | | | | erence–attributecorrespondencethatisdifficultto | | | | | | +| -------------------------------- | -------------------------------- | --- | ------------ | --- | ----------------------------------------------- | --------- | ------- | ---------- | ---- | ----------- | +| | | | e ·e | | recoverfromindividualreviewsalone. | | | | | | +| | | | u i | | | | | | | | +| | Align(u,i) | = | | (3) | | | | | | | +| | | | ∥e ∥∥e ∥ | | | | | | | | +| | | | u i | | 5 Conclusion | | | | | | +| wheree | ande denotetheembeddingvectorsof | | | | | | | | | | +| | u i | | | | In this | paper, we | propose | DUET | , a | closed-loop | +| thegenerateduseranditemprofiles. | | | Highervalues | | | | | | | | +| | | | | | framework | for | jointly | generating | user | and item | +indicatestrongersemanticcompatibilitybetween +| | | | | | textualprofilesforrecommendation. | | | | | Unlikeprior | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | +modeleduserpreferencesanditemcharacteristics. +| | | | | | methods | that rely | on | fixed templates | | or indepen- | +| -------- | --------------- | --- | ---------- | ------ | ------- | ----------- | --------- | --------------- | --- | -------------- | +| Coverage | (Faithfulness). | | We measure | token- | | | | | | | +| | | | | | dently | constructed | profiles, | DUET | | treats profile | +levelgroundingas: +| | | | | | generation | as an | exploration | problem | | and aligns | +| --- | --- | --- | --- | --- | ---------- | ----- | ----------- | ------- | --- | ---------- | +|Tokens(profile)∩Tokens(history)| representations directly with downstream recom- +Cov = +| | | |Tokens(profile)| | | | mendationperformance. | | | | | | +| --- | --- | ----------------- | --- | --- | --------------------- | --- | ---- | ---------- | --------- | ---- | +| | | | | (4) | Specifically, | | DUET | integrates | cue-based | ini- | +Thismetricquantifieshowmuchofthegenerated tialization, adaptive strategy construction, and +profileissupportedbyhistoricaltextualevidence. feedback-drivenjointoptimizationtoproduceflex- +We report coverage separately for user and item ible yet task-aligned user–item profiles. By opti- +| profiles. | | | | | mizing | both profiles | in | a shared | semantic | space, | +| --------- | --- | --- | --- | --- | ------ | ------------- | --- | -------- | -------- | ------ | +AsshowninTable6,DUETachievesthehigh- theframeworkreducessemanticmismatchandcap- +est semantic alignment across all datasets while turesinteraction-relevantsignalsthataredifficultto + +Setting Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1 MAE RMSE Acc F1 MAE RMSE Acc F1 +(%) (%) (%) (%) (%) (%) +DUETw/oRL 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +DUET(full) 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table5: EffectofRL-basedoptimization(usingQwen3(8B)). +User History: +Andre' S Grindle "Tour of U.S.A.", Rating: 5.0. If your looking for heavy metals dim orgins look no further. +Andre' S Grindle "Misfits", Rating:4.0. Actually the opening title track is more of a direct pop/rock type song. But the rock n roll +returns in a big way on "Hay Fever". +............ +Item History: +Worlds Away, Rating: 4.0, Review: The CD is wonderful, and is a seminal piece of music for both the group and for those of us who +were 'of age' in the '70s and '80s. Pablo Cruise was - and is - a group that makes beautiful music like few others, and this +shipper/seller made it all the more special with a fast ship! THANKS!!! +Worlds Away, Rating: 5.0, I love this CD! I'd recommend the Pablo Cruise Millenium Collection to get to know this band. Worlds Away to +me represents the best of breezy '70s pop. Five stars! +............ +USER PROFILE: +Andre Grindle is a dedicated music enthusiast with a deep appreciation for funk, soul, and progressive rock, particularly drawn to artists +who blend technical mastery with genre-defining innovation. His reviews consistently highlight albums that showcase musical complexity, +lyrical depth, and historical significance, often referencing artists like Stevie Wonder, Prince, and Rush. He values authenticity and +artistic vision, frequently praising albums that resist trend-chasing and prioritize creative integrity. His detailed, introspective reviews +suggest a preference for nuanced analysis over superficial praise, with a focus on emotional resonance, production quality, and the +artist's evolution. While he is generous with high ratings (often 5.0), he is also willing to critique perceived shortcomings, as seen in his +4.0 for "Test for Echo" and 3.0 for "The Color of Success." His engagement is characterized by a blend of personal anecdotes, technical +appreciation, and historical context, indicating a preference for music that transcends mere entertainment. +Item PROFILE: +"Worlds Away" by Pablo Cruise is celebrated as a peak-era funk-rock album that exemplifies the band's ability to merge polished +production with genre-blurring creativity. Users consistently praise its tight musicianship, genre-defying sound, and the band's ability to +craft catchy, emotionally resonant tracks. The album's appeal lies in its seamless fusion of rock, funk, and pop, with standout tracks like +"Love Will Find a Way" and "Worlds Away" receiving particular acclaim for their infectious melodies and lyrical depth. While some users +note its dated production compared to modern standards, the majority highlight its historical significance as a defining work of the +1970s rock-funk movement. The album resonates most with fans of 1970s progressive rock, genre-defying artists, and listeners seeking +well-crafted, emotionally engaging music. Its high average rating reflects widespread admiration for its artistic ambition and enduring +appeal, though a minority acknowledges its limitations in terms of modern production quality. +Figure4: Illustrationofthemutualcorrespondencebetweenuseranditem. Thehighlightedregionsdemonstrate +thatuserpreferencessummarizedintheuserprofilealignwiththekeyattributesextractedintheitemprofile,which +providescomplementaryinformationbeyondrawhistoriesandthusimprovespredictionaccuracy. +Method Yelp AmazonMusic AmazonBooks +Align UserCov. ItemCov. Align UserCov. ItemCov. Align UserCov. ItemCov. +10H 0.3902 /(1.00) /(1.00) 0.3577 /(1.00) /(1.00) 0.2604 /(1.00) /(1.00) +KAR(Xietal.,2024) 0.4932 0.1745 0.1823 0.4807 0.2170 0.2391 0.5702 0.1877 0.1544 +RLMRec(Renetal.,2024) 0.4010 0.2646 0.3773 0.3931 0.3526 0.4889 0.4436 0.2994 0.3506 +PALR(Yangetal.,2023) 0.4715 0.3675 0.1917 0.4938 0.2197 0.2390 0.5216 0.2050 0.2267 +LG(Wangetal.,2025) 0.5709 0.2378 0.2682 0.5109 0.2546 0.3749 0.5506 0.3883 0.3364 +R4Rec(Fangetal.,2025) 0.4882 0.2348 0.2330 0.4208 0.3328 0.3595 0.4946 0.2831 0.2949 +Ours 0.6382 0.2880 0.3429 0.5947 0.4002 0.4482 0.7287 0.3457 0.3127 +Table6: Semanticalignmentandcoverageofgeneratedprofiles. Cov. denotesthefractionoftokensgroundedin +historicaltext. +recoverfromrawhistoriesorstaticpromptsalone. reinforcementlearning–basedoptimization. These +resultssuggestthatadaptive,interaction-awaretex- +Experiments on multiple real-world datasets tualprofilesprovideapromisingdirectionformore +demonstratethatDUETconsistentlyoutperforms effective interpretable and performance-oriented +strongbaselinesunderdifferentbackbonemodels, recommendationsystems. +validating the effectiveness of joint profiling and + +| Limitations | | | | | | | References | | | | | | | +| ----------- | --- | --- | --- | --- | --- | --- | ---------- | ----- | ------ | ---- | ------ | ------ | ----- | +| | | | | | | | Keqin Bao, | Jizhi | Zhang, | Yang | Zhang, | Wenjie | Wang, | +Despiteitseffectiveness,ourapproachhasseveral FuliFeng,andXiangnanHe.2023. Tallrec: Anef- +fectiveandefficienttuningframeworktoalignlarge +| limitations. | First, | the | proposed | framework | | relies | | | | | | | | +| ------------ | ------ | --- | -------- | --------- | --- | ------ | -------------------------------- | --- | --- | --- | --- | ---------- | --- | +| | | | | | | | languagemodelwithrecommendation. | | | | | InProceed- | | +onlargelanguagemodelsforbothprofilegenera- ingsofthe17thACMConferenceonRecommender +tion and downstream recommendation, which in- Systems,pages1007–1014. +| troduces | additional | computational | | | overhead | dur- | | | | | | | | +| -------- | ---------- | ------------- | --- | --- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | +TingChen,Wei-LiHan,Hai-DongWang,Yi-XunZhou, +| ingtrainingandinference. | | | Whileourexperiments | | | | | | | | | | | +| ------------------------ | --- | --- | ------------------- | --- | --- | --- | ------------------------- | --- | --- | --- | ---------------- | --- | --- | +| | | | | | | | BinXu,andBin-YuZang.2007. | | | | Contentrecommen- | | | +showconsistentgainsacrossdifferentbackbones, dationsystembasedonprivatedynamicuserprofile. +the overall efficiency may be constrained when In2007Internationalconferenceonmachinelearn- +| | | | | | | | ing and | cybernetics, | | volume | 4, pages | 2112–2118. | | +| --------------------------------- | --- | --- | --- | --- | ---------- | --- | ------- | ------------ | --- | ------ | -------- | ---------- | --- | +| scalingtoverylargeuseroritemsets. | | | | | Second,the | | | | | | | | | +IEEE. +| quality | of the generated | | profiles | is | inherently | de- | | | | | | | | +| ------- | ----------------- | --- | -------- | --- | ------------- | --- | --------- | ------- | --- | ----------- | ----- | --- | ----- | +| pendent | on the underlying | | LLMs | | and prompting | | | | | | | | | +| | | | | | | | Yue Chen, | Minghua | | He, Fangkai | Yang, | Pu | Zhao, | +strategies. Variationsinmodelcapacityorprompt LuWang,YuKang,YifeiDong,YuefengZhan,Hao +Sun,QingweiLin,SaravanRajmohan,andDongmei +sensitivitymayleadtodifferencesinprofilestabil- +| | | | | | | | Zhang. | 2025. | Warriormath: | | Enhancing | the | mathe- | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | ------------ | --- | --------- | --- | ------ | +ity,whichwedonotexplicitlycontrolinthecurrent +maticalabilityoflargelanguagemodelswithadefect- +design. Finally,ourevaluationfocusesontext-rich +| | | | | | | | awareframework. | | | Preprint,arXiv:2508.01245. | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | -------------------------- | --- | --- | --- | +recommendationscenarioswheresufficienthistor- +| | | | | | | | Paul Covington, | | Jay | Adams, and | Emre | Sargin. | 2016. | +| ------------ | --- | ---------- | --- | ----------------- | --- | --- | --------------- | --- | --- | ---------- | ---- | ------- | ----- | +| ical reviews | are | available. | | The effectiveness | | of | | | | | | | | +Deepneuralnetworksforyoutuberecommendations. +| the proposed | strategy | | in domains | with | extremely | | | | | | | | | +| ------------ | -------- | --- | ---------- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +InProceedingsofthe10thACMconferenceonrec- +sparsetextualsignalsorinnon-textualmodalities +ommendersystems,pages191–198. +remainstobefurtherexplored. +DeepSeek-AI,DayaGuo,DejianYang,HaoweiZhang, +JunxiaoSong,RuoyuZhang,RunxinXu,QihaoZhu, +ShirongMa,PeiyiWang,XiaoBi,XiaokangZhang, +Data,Privacy,andEthicsConsiderations Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhi- +| | | | | | | | hong | Shao, | Zhuoshu | Li, Ziyi | Gao, and | 181 | others. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | ------- | ------------- | --------- | --- | ------- | +| | | | | | | | 2025. | Deepseek-r1: | | Incentivizing | reasoning | | capa- | +Alldatasetsusedinthisworkarepubliclyreleased bilityinllmsviareinforcementlearning. Preprint, +arXiv:2501.12948. +bytheiroriginalowners,andwefollowtheofficial +termsofuseassociatedwitheachdataset,utilizing +JiaxinDeng,ShiyaoWang,KuoCai,LejianRen,Qigen +thedatasolelyforresearchpurposesaspermitted +| | | | | | | | Hu, Weifeng | | Ding, | Qiang Luo, | and | Guorui | Zhou. | +| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ----- | ---------- | --- | ------ | ----- | +bytheirrespectivelicenses;forreproducibility,the 2025. Onerec: Unifyingretrieveandrankwithgen- +officialdatasetsourcesarecitedinthecorrespond- erativerecommenderanditerativepreferencealign- +ment. Preprint,arXiv:2502.18965. +ingsectionsofthispaper,andwedonotredistribute +anyrawuser-generatedcontentsuchasreviewtext AbhimanyuDubey,AbhinavJauhri,AbhinavPandey, +| aspartofthiswork. | | Thisresearchdoesnotattempt | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +AbhishekKadian,AhmadAl-Dahle,AieshaLetman, +to identify or infer the identity of any individual Akhil Mathur, Alan Schelten, Amy Yang, Angela +| | | | | | | | Fan,and1others.2024. | | | Thellama3herdofmodels. | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | ---------------------- | --- | --- | --- | +user,andpriortotraininganddownstreamprofile +arXivpreprintarXiv:2407.21783. +| generation, | we | apply text | filtering | | procedures | to | | | | | | | | +| ----------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +remove direct personal identifiers where applica- YiFang,WenjieWang,YangZhang,FengbinZhu,Qi- +ble;themodelsaretrainedsolelyonpubliclyavail- fanWang,FuliFeng,andXiangnanHe.2025. Rea- +son4rec:Largelanguagemodelsforrecommendation +abledata,andnoadditionalprivateorproprietary +| | | | | | | | withdeliberativeuserpreferencealignment. | | | | | Preprint, | | +| ---------------- | --- | -------------- | --- | ------------ | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --------- | --- | +| user information | | is introduced. | | Furthermore, | | we | | | | | | | | +arXiv:2502.02061. +| do not release | generated | | user | profiles | or | any per- | | | | | | | | +| -------------- | --------- | --- | ---- | -------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +MinghuaHe,YueChen,FangkaiYang,PuZhao,Wenjie +useroutputsproducedduringthemodelingprocess, +Yin,YuKang,QingweiLin,SaravanRajmohan,and +andallreportedresultsarepresentedinaggregate +| | | | | | | | DongmeiZhang.2025. | | | Execoder:Empoweringlarge | | | | +| -------- | ---------- | -------- | --- | ----- | ---- | ------- | ------------------ | --- | --- | ------------------------ | --- | --- | --- | +| form for | evaluation | purposes | | only. | From | a secu- | | | | | | | | +languagemodelswithexecutabilityrepresentationfor +rityperspective,theassociatedcoderepositoryhas codetranslation. Preprint,arXiv:2501.18460. +beenscannedforpotentialsecrets,andanyinternal +MinjieHong,ZirunGuo,YanXia,ZehanWang,Ziang +endpointsorconfidentialconfigurationinformation +| | | | | | | | Zhang, | Tao | Jin, and | Zhou Zhao. | 2025a. | Apo: | En- | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | ---------- | ------ | ---- | --- | +havebeenremovedpriortorelease. +hancingreasoningabilityofmllmsviaasymmetric +| | | | | | | | policyoptimization. | | | Preprint,arXiv:2506.21655. | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | -------------------------- | --- | --- | --- | + +Minjie Hong, Yan Xia, Zehan Wang, Jieming Zhu, JoarSkalse,NikolausH.R.Howe,DmitriiKrashenin- +Ye Wang, Sihang Cai, Xiaoda Yang, Quanyu Dai, nikov, and David Krueger. 2025. Defining +ZhenhuaDong,ZhimengZhang,and1others.2025b. and characterizing reward hacking. Preprint, +| Eager-llm: | Enhancinglargelanguagemodelsasrec- | | | | | arXiv:2209.13085. | | | | | | +| ---------- | ---------------------------------- | --------- | ----------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| ommenders | through | exogenous | behavior-semantic | | | | | | | | | +integration. InProceedingsoftheACMonWebCon- Harald Steck. 2019. Embarrassingly shallow autoen- +ference2025,pages2754–2762. codersforsparsedata. InProceedingsofTheWorld +WideWebConference(WWW),pages3251–3257. +MinjieHong,ZetongZhou,ZirunGuo,ZiangZhang, +| Ruofan | Hu, Weinan | Gan, | Jieming | Zhu, and | Zhou | | | | | | | +| ------ | ---------- | ---- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | +ChaoSun,YaoboLiang,YamingYang,ShilinXu,Tian- +| Zhao.2025c. | Generativereasoningrecommendation | | | | | | | | | | | +| ----------- | --------------------------------- | --- | --- | --- | --- | ---------------------------- | -------- | ---- | ------ | -------- | ------------- | +| | | | | | | mengYang,andYunhaiTong.2024. | | | | | Rlrf4rec: Re- | +| viallms. | Preprint,arXiv:2510.20815. | | | | | | | | | | | +| | | | | | | inforcement | learning | from | recsys | feedback | for en- | +JihwanJeong,YinlamChow,GuyTennenholtz,Chih- hancedrecommendationreranking. arXivpreprint +| Wei Hsu, | Azamat | Tulepbergenov, | | Mohammad | | arXiv:2410.05939. | | | | | | +| ------------------------------------ | ------------ | --------------- | ---------- | -------- | --------- | ----------------- | --- | --------------------- | --- | --- | --------- | +| Ghavamzadeh, | and | Craig | Boutilier. | 2023. | Fac- | | | | | | | +| | | | | | | QwenTeam.2025. | | Qwen3technicalreport. | | | Preprint, | +| tual and | personalized | recommendations | | using | lan- | | | | | | | +| guagemodelsandreinforcementlearning. | | | | | Preprint, | arXiv:2505.09388. | | | | | | +arXiv:2310.06176. +| | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | +| --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | +Yitong Ji, Aixin Sun, Jie Zhang, and Chenliang Li. bert, Amjad Almahairi, Yasmine Babaei, Nikolay +| 2023. A | critical | study | on data leakage | in | recom- | | | | | | | +| ------- | -------- | ----- | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti +| mendersystemofflineevaluation. | | | | ACMTrans.Inf. | | | | | | | | +| ------------------------------ | --- | --- | --- | ------------- | --- | ------------------------ | --- | --- | ------- | --- | --------- | +| | | | | | | Bhosale,and1others.2023. | | | Llama2: | | Openfoun- | +Syst.,41(3):75:1–75:27. +| | | | | | | dation and | fine-tuned | chat | models. | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | ---------- | ---------- | ---- | ------- | ----- | -------- | +arXiv:2307.09288. +| JiachengLin,TianWang,andKunQian.2025. | | | | | Rec-r1: | | | | | | | +| ------------------------------------- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +Bridginggenerativelargelanguagemodelsanduser- +| | | | | | | LeandrovonWerra, | | YounesBelkada, | | LewisTunstall, | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | -------------- | --- | -------------- | --- | +centricrecommendationsystemsviareinforcement +EdwardBeeching,TristanThrush,NathanLambert, +| learning. | Preprint,arXiv:2503.24289. | | | | | | | | | | | +| --------- | -------------------------- | --- | --- | --- | --- | -------- | ------ | ----------- | ------------- | --- | ------------ | +| | | | | | | Shengyi | Huang, | Kashif | Rasul, | and | Quentin Gal- | +| | | | | | | louédec. | TRL: | Transformer | Reinforcement | | Learn- | +AiweiLiu,MinghuaHe,ShaoxunZeng,SijunZhang, +| LinhaoZhang,ChuhanWu,WeiJia,YuanLiu,Xiao | | | | | | ing. | | | | | | +| ---------------------------------------- | --- | --- | ----------------------- | --- | --- | ---- | --- | --- | --- | --- | --- | +| Zhou,andJieZhou.2025. | | | Wedlm:Reconcilingdiffu- | | | | | | | | | +sionlanguagemodelswithstandardcausalattention LuWang,DiZhang,FangkaiYang,PuZhao,Jianfeng +forfastinference. Preprint,arXiv:2512.22737. Liu,YuefengZhan,HaoSun,QingweiLin,Weiwei +| | | | | | | Deng, Dongmei | | Zhang, | Feng | Sun, and | Qi Zhang. | +| --- | --- | --- | --- | --- | --- | ------------- | --- | ------ | ---- | -------- | --------- | +HongtaoLiu,FangzhaoWu,WenjunWang,Xianchen +| | | | | | | 2025. Lettingo: | | Exploreuserprofilegenerationfor | | | | +| --- | --- | --- | --- | --- | --- | --------------- | --- | ------------------------------- | --- | --- | --- | +Wang,PengfeiJiao,ChuhanWu,andXingXie.2019. +| | | | | | | recommendationsystem. | | | InProceedingsofthe31st | | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | +Nrpa: Neuralrecommendationwithpersonalizedat- ACMSIGKDDConferenceonKnowledgeDiscovery +tention. In Proceedings of the 42nd International andDataMiningV.2,KDD’25,NewYork,NY,USA. +ACMSIGIRConferenceonResearchandDevelop- AssociationforComputingMachinery. +mentinInformationRetrieval,pages1233–1236. +YeWang,JiahaoXun,MinjieHong,JiemingZhu,Tao +| Wensheng | Lu, Jianxun | Lian, | Wei Zhang, | Guanghua | | | | | | | | +| -------- | ----------- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | +Jin,WangLin,HaoyuanLi,LinjunLi,YanXia,Zhou +Li,MingyangZhou,HaoLiao,andXingXie.2024. +| | | | | | | Zhao,and1others.2024. | | | Eager: | Two-streamgen- | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ------ | -------------- | --- | +Aligninglargelanguagemodelsforcontrollablerec- +arXivpreprintarXiv:2403.05063. erativerecommenderwithbehavior-semanticcollab- +ommendations. +| | | | | | | oration. | InProceedingsofthe30thACMSIGKDD | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------- | --- | --- | --- | --- | +RadaMihalceaandPaulTarau.2004. Textrank: Bring- ConferenceonKnowledgeDiscoveryandDataMin- +| | | InEMNLP. | | | | ing,pages3245–3254. | | | | | | +| --- | --- | -------- | --- | --- | --- | ------------------- | --- | --- | --- | --- | --- | +ingorderintotext. +| Zhaopeng | Qiu, Xian | Wu, | Jingyue Gao, | and | Wei Fan. | | | | | | | +| -------- | --------- | --- | ------------ | --- | -------- | -------- | --------------- | --- | -------- | -------- | --- | +| | | | | | | Triyanna | Widiyaningtyas, | | Indriana | Hidayah, | and | +2021. U-BERT:pre-traininguserrepresentationsfor +| | | | | | | TeguhBAdji.2021. | | Userprofilecorrelation-based | | | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | +improved recommendation. In AAAI, pages 4320– similarity(upcsim)algorithminmovierecommenda- +| 4327.AAAIPress. | | | | | | tionsystem. | JournalofBigData,8(1):52. | | | | | +| --------------------------------- | --- | --- | --- | -------------- | --- | ----------- | ----------------------------------- | --- | --- | --- | --- | +| NilsReimersandIrynaGurevych.2019. | | | | Sentence-bert: | | | | | | | | +| | | | | | | J.Wu.2023. | Computationalunderstandingofuserin- | | | | | +Sentenceembeddingsusingsiamesebert-networks. +terfaces. +InProceedingsofthe2019ConferenceonEmpirical +MethodsinNaturalLanguageProcessing(EMNLP), +pages3982–3992. Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, +| | | | | | | Hong Zhu, | Jieming | Zhu, | Bo Chen, | Ruiming | Tang, | +| --- | --- | --- | --- | --- | --- | --------- | ------- | ---- | -------- | ------- | ----- | +Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi WeinanZhang,andYongYu.2024. Towardsopen- +Cheng,JunfengWang,DaweiYin,andChaoHuang. world recommendation with knowledge augmenta- +2024. Representationlearningwithlargelanguage tionfromlargelanguagemodels. InProceedingsof +modelsforrecommendation. InProceedingsofthe the18thACMConferenceonRecommenderSystems, +| ACMonWebConference2024,pages3464–3475. | | | | | | pages12–22. | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | + +A ExperimentSetup +FanYang,ZhengChen,ZiyanJiang,EunahCho,Xiao- +| jiangHuang,andYanbinLu.2023. | | Palr:Personaliza- | | | | | | | | +| ------------------------------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| tionawarellmsforrecommendation. | | arXivpreprint | A.1 DataCuration | | | | | | | +arXiv:2305.07622. +Jiarui Zhang. 2024. Guided profile generation im- Table7: Statisticaldetailsoftheevaluationdatasets. +| proves personalization | with llms. | arXiv preprint | | | | | | | | +| ---------------------- | ---------- | -------------- | ---------- | ------ | ----------- | --- | ------------ | ------ | ------ | +| arXiv:2409.13093. | | | Dataset | #Train | #Valid | | #Test #User | | #Item | +| | | | Music | 43,071 | 3,271 | | 1,296 | 4,183 | 2,660 | +| | | | Book | 71,972 | 6,144 | | 5,541 13,863 | | 13,515 | +| | | | Yelp | 51,497 | 4,757 | | 4,328 | 8,453 | 13,426 | +| | | | We conduct | | experiments | | on three | widely | used | +real-worlddatasets: +| | | | • AmazonMusic(Music): | | | | Thisreferstothe“Dig- | | | +| --- | --- | --- | --------------------- | --- | ------ | ------ | -------------------- | --- | ------ | +| | | | ital Music” | | subset | of the | well-known | | Amazon | +Productdataset3,whichrecordsrichuserreviews, +ratings,andtextualinformationaboutitems,such +| | | | as titles, | across | a | broad | range | of product | cate- | +| --- | --- | --- | ---------- | ------ | --- | ----- | ----- | ---------- | ----- | +gories,ontheAmazonplatform.2 +| | | | • AmazonBook(Book): | | | Thisreferstothe“Book” | | | | +| --- | --- | --- | ------------------- | --- | --- | --------------------- | --- | --- | --- | +subsetoftheAmazonProductdataset. +dataset4, +| | | | • Yelp: | This | refers | to the | Yelp | Open | | +| --- | --- | --- | ------- | -------- | -------------- | -------- | ------- | ------------- | --------- | +| | | | which | includes | user | reviews, | ratings | | for busi- | +| | | | nesses | such | as restaurants | | and | retail shops, | as | +wellastextualinformationaboutthebusinesses. +Itiswidelyusedinrecommendationtasks(Qiu +etal.,2021). +WeusetheentireMusicdatasetforexperiments, +| | | | while for | the Book | | and Yelp | datasets, | we | utilize | +| --- | --- | --- | ------------------------------- | -------- | --- | -------- | --------- | ---------- | ------- | +| | | | onlyasubsetduetotheirlargesize. | | | | | FortheBook | | +dataset,weusedatafromthelasttwomonths,and +fortheYelpdataset,weusedatafromthelastsix +months. Foreachdataset,wesplititintotraining, +| | | | validation, | and | test | sets based | on | the timestamps | | +| --- | --- | --- | ---------------- | --- | -------- | ---------- | ---- | -------------- | --- | +| | | | of interactions, | | ensuring | that | test | interactions | oc- | +curafteralltrainingandvalidationinteractionsto +preventinformationleakage(Jietal.,2023). +| | | | Regarding | | data | filtering, | following | | prior | +| --- | --- | --- | --------- | --- | ----------- | ---------- | --------- | -------- | ----- | +| | | | work (Liu | et | al., 2019), | we | adopt | a 5-core | set- | +tingtofilterthedataandexcludecold-startusers +| | | | and items—those | | not | appearing | | in the | training | +| --- | --- | --- | --------------- | --- | ---------- | --------- | ---- | ------ | -------- | +| | | | set—from | the | validation | and | test | sets. | The sta- | +tisticaldetailsoftheprocesseddatasetareprovided +inTable7. +3https://cseweb.ucsd.edu/~jmcauley/datasets/ +amazon/links.html. +4https://business.yelp.com/data/resources/ +open-dataset/. + +| A.1.1 | ImplementationDetails | | | | | | RLMRecPrompt | | | | +| ----- | --------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | +Inourexperiments,weprimarilyemployQwen3- Role: Businessrecommendationassistant +Task: Determinebusinesstypesauserislikely +| 8B(Team,2025)andLLaMA38BInstruct | | | | | | (Dubey | | | | | +| ------------------------------------------ | --- | ----- | --- | --- | ------- | ---------- | ------------ | --- | --- | --- | +| etal.,2024;Touvronetal.,2023)asboththerec- | | | | | | | toenjoy | | | | +| ommendation | | model | and | the | profile | generation | InputFormat: | | | | +•Title: Businessname +model. Thetrainingprocessisimplementedusing +theTRL(vonWerraetal.). Keyhyperparameters, •Categories: Businesscategories +suchasbatchsizeandlearningrate,aredetermined •Sentiment: Usersentimenttowardbusiness +throughgridsearchtoachieveoptimalperformance. +OutputRequirements: +Moredetailscanbefoundinourcode. +1. JSONformatonly +2. Structure: +A.2 BaselinePrompts +{ +| | | | | | | | "summarization": | "Types | of businesses | user | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------ | ------------- | ---- | +KARPrompt +likelyenjoys"(≤100words), +Task: Analyzeuserpreferencesbasedonbusi- +| | | | | | | | "reasoning": | "Brief explanation | | for summa- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------ | --- | ---------- | +nessreviewinghistory +rization"(nowordlimit) +Input: {user_history}-User’sbusinessreview- +} +inghistorywithsentimentsovertime +3. NoadditionaltextoutsideJSON +Instructions: +| 1. | Analyze | the | user’s | preferences | considering | | | | | | +| --- | ------- | --- | ------ | ----------- | ----------- | --- | ------ | ----------- | --- | ------ | +| | | | | | | | Input: | INTERACTION | | ITEMS: | +businessnamesandcategories +{user_history} +| 2. | Take | into account | | sentiment | patterns | over | | | | | +| --- | ---- | ------------ | --- | --------- | -------- | ---- | --- | --- | --- | --- | +time +| 3. | Provideclearexplanationsbasedonreview- | | | | | | LGPrompt | | | | +| --- | -------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | +inghistorydetails +Youwillserveasanassistanttohelpmegener- +| 4. | Consider | other | pertinent | | factors | that may | | | | | +| --- | -------- | ----- | --------- | --- | ------- | -------- | --- | --- | --- | --- | +ateauserprofilebasedonthisuser’ssentiments +influencepreferences +historytobetterunderstandthisusers’interest +andthuspredicthis/hersentimentaboutatarget +| PALRPrompt | | | | | | | item. Iwillprovideyouwithsomebehaviorhis- | | | | +| ---------- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --------------- | --- | +| | | | | | | | toryoftheuserinthisformat: | | [itemattributes | | +Task: Summarizeuserpreferencesusingkey- +| | | | | | | | andsentiment]. | Theuserprofileyougenerate | | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ------------------------- | --- | --- | +words. +shouldcontainasmuchusefulcontentaspossi- +| Input: | | {user_history} | | - historical | businesses | | | | | | +| ------ | --- | -------------- | --- | ------------ | ---------- | --- | --- | --- | --- | --- | +bletohelppredicttheuser’ssentimenttowards +withusersentiments. +anewbusiness. +| Output | | Format: | An | itemized | list ranked | by | | | | | +| ------ | --- | ------- | --- | -------- | ----------- | --- | --- | --- | --- | --- | +USERHISTORY:{user_history}. +importance. +PROFILEYOUGENERATE: +Template: +| | • KEY_WORD_1: | | | | | "HIS- | | | | | +| --- | ------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +R4RecPrompt(Reasoner) +| | TORY_BUSINESS_1", | | | | | "HIS- | | | | | +| --- | ----------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +UserReviewHistory +TORY_BUSINESS_2" +⟨H u organizedasbelow⟩ +| | • KEY_WORD_2: | | | | | "HIS- | 1. TitleofItem1 | | | | +| ------------- | ---------------- | --- | --- | --- | --- | ----- | ---------------- | ----------------------- | ----------- | --- | +| | TORY_BUSINESS_3" | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| | | | | | | | NegativeAspects: | [Aspect1],[Aspect2],... | | | +| Instructions: | | | | | | | User Preference | Elements: | [Preference | 1], | +[Preference2],... +| | 1. Extract | key | preference | | indicators | from | | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | ---- | --- | --- | --- | --- | +2. TitleofItem2 +userinteractionhistory. +| | | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------------- | --- | --- | +2. Rankkeywordsbyimportance. NegativeAspects: [Aspect1],[Aspect2],... +| | | | | | | | User Preference | Elements: | [Preference | 1], | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --------- | ----------- | --- | + +| [Preference2],... | | | | | | | cantlyworsethanDUETacrossalldatasets. | | | | | This | +| ----------------- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | ---- | +indicatesthatcoherentsummariesaloneareinsuf- +... +| | | | | | | | ficient, | and that | the gains | of DUET | | arise from | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --------- | ------- | --- | ---------- | +ItemReviewHistorybyOtherUsers +⟨H organizedinthesameformatasabove⟩ learnedsemanticabstractionratherthanextractive +i +compression. +| Task: | Analyze | whether | | the user | will | like the | | | | | | | +| --------- | ------- | ------- | --- | ---------- | ----------- | -------- | --- | --- | --- | --- | --- | --- | +| new Music | | i based | on | the user’s | preferences | | | | | | | | +B.2 RankingunderHardNegativeSampling +| andtheitem’sfeatures. | | | | Provideyourrationale | | | | | | | | | +| --------------------- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Toconstructamorechallengingrankingscenario, +inoneconciseparagraph. +wereplacerandomnegativeswithhardnegatives +| | | | | | | | generatedbyacollaborativefilteringmodel. | | | | | Specif- | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ------- | +A.3 ExampleofCue +ically,foreachuser,weretrievehigh-scoringitems +Thefollowingexamplesillustratehowrawsignals +| | | | | | | | from an | EASE(Steck, | 2019) | model | that | the user | +| --------------------- | --- | --- | ---------------- | --- | --- | ---- | ------- | ----------- | ----- | ----- | ---- | -------- | +| aredistilledintocues. | | | Asshownonthetop, | | | user | | | | | | | +hasnotinteractedwith,andcombinethemwiththe +cuesemphasizehistoricalpreferences,whileitem +ground-truthitemtoformthecandidateset. +cues highlight metadata and user-group patterns. Table 9 reports the results. Compared to ran- +| Together, | they | provide | minimal | | but informative | | | | | | | | +| --------- | ---- | ------- | ------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +domsampling,performancedecreasesforallmeth- +hypothesesforprofileexploration. +| | | | | | | | ods due | to increased | difficulty, | while | DUET | con- | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | ----------- | ---------------- | ---- | ------ | +| | | | | | | | sistently | maintains | the | best performance | | across | +ExamplesofUserCues +| | | | | | | | datasets. | Thisindicatesthattheimprovementsare | | | | | +| ------- | ----- | ------ | ------ | --- | ------- | -------- | --------- | ----------------------------------- | --- | --- | --- | --- | +| “enjoys | retro | puzzle | games” | — | derived | from re- | | | | | | | +robustandnotlimitedtotrivialrankingscenarios. +peatedengagementwithclassictitles. +“prefersconciseproductreviews”—inferredfrom B.3 RobustnessunderPreferenceDiversity +apatternofshort,directcomments. +WefurtheranalyzetherobustnessofDUETunder +“tendstogivehighratingsbutrarelycomments”— +| | | | | | | | varyinglevelsofuserpreferencediversity. | | | | | Weuse | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | ----- | +highlightingconsistencybutlimitedfeedback. +thevarianceofhistoricalratingsasaproxyforpref- +| | | | | | | | erencestability: | | lowvarianceindicatesconsistent | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ------------------------------ | --- | --- | --- | +ExamplesofItemCues +| | | | | | | | preferences, | while | high | variance | corresponds | to | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ----- | ---- | -------- | ----------- | --- | +“lightweighttrail-runningshoes”—derivedfrom +diverseorpotentiallyconflictingsignals. +productmetadata. +Wepartitionusersintothreegroupsbasedonper- +| “popular | among | budget-conscious | | | users” | — in- | | | | | | | +| -------- | ----- | ---------------- | --- | --- | ------ | ----- | --- | --- | --- | --- | --- | --- | +centilethresholds(bottom33%,middle33%,top +ferredfrompurchasepatterns. +33%ofratingvariance)andevaluateperformance +| “stylized | with | retro | aesthetics” | | — extracted | from | | | | | | | +| --------- | ---- | ----- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | +withineachgroup. +itemdescriptions. +| | | | | | | | As shown | in | Table 10, | performance | | degrades | +| --- | --- | --- | --- | --- | --- | --- | -------- | --- | --------- | ----------- | --- | -------- | +smoothlyaspreferencediversityincreasesacross +| | | | | | | | all datasets. | Importantly, | | the degradation | | is grad- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------ | --- | --------------- | --- | -------- | +B AdditionalExperiments +ualratherthancatastrophic,indicatingthatDUET +B.1 Non-LLMBaselineviaExtractive +remainsstableunderheterogeneousornoisyinter- +Summarization +actionhistories. +| To examine | | whether | the | gains | of DUET | stem | | | | | | | +| ---------- | --- | ------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | +C TheUseofLargeLanguageModels +| from improved | | semantic | | representations | | rather | | | | | | | +| ------------- | --- | ---------------- | --- | --------------- | --------- | ------ | --- | --- | --- | --- | --- | --- | +| than generic | | text generation, | | we | introduce | a non- | | | | | | | +WeusedaLargeLanguageModel(LLM)onlyas +LLMbaselinebasedonextractivesummarization. a writing assistant to polish the language of the +| Specifically, | | we apply | TextRank | | (Mihalcea | and | | | | | | | +| ------------- | --- | -------- | -------- | --- | --------- | --- | ---------- | ------ | ------- | ----------- | --- | --------- | +| | | | | | | | manuscript | (e.g., | grammar | refinement, | | style ad- | +Tarau,2004)toselectsalientsentencesfromuser +| | | | | | | | justment,andclarityimprovement). | | | | Theresearch | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | ----------- | --- | +historiesandconstructuserprofileswithoutusing ideas,methodologydesign,experiments,andanal- +anygenerativemodel. ysiswereentirelyconceived,implemented,andval- +| The | extracted | summaries | | are | then fed | into the | | | | | | | +| --- | --------- | --------- | --- | --- | -------- | -------- | --- | --- | --- | --- | --- | --- | +idatedbytheauthorswithoutrelianceontheLLM. +samedownstreampredictorforratingestimation. +| | | | | | | | The LLM | did not | contribute | to research | | ideation, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | ---------- | ----------- | --- | --------- | +Thisbaselineisolatestheeffectofreadablesumma- experimentaldesign,orresultinterpretation. +rizationfromrepresentationlearning. +Table8showsthatTextRankimprovesoversim- +| ple history | truncation | | (10H), | but | remains | signifi- | | | | | | | +| ----------- | ---------- | --- | ------ | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | + +| Method | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | -------- | ---- | ----- | --- | ----------- | --------- | --- | ----------- | --------- | +| | MAE RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | MAE | RMSE | Acc F1(%) | +| | | (%) | | | | (%) | | | (%) | +10H 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +TextRank 0.8104 1.1323 50.51 30.23 0.5914 0.8223 61.19 29.68 0.6328 0.8854 59.59 29.48 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table8: Comparisonwithanon-LLMextractivesummarizationbaseline(TextRank). +| Method | | | Yelp | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | ---- | --- | ----------- | --- | --- | ----------- | --- | +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +| | Table9: | RankingperformanceunderEASE-based(Steck,2019)hardnegatives. | | | | | | | | +| ------------- | ------- | ----------------------------------------------------------- | --- | --- | ----------- | --- | --- | ----------- | --- | +| VarianceGroup | | Yelp | | | AmazonMusic | | | AmazonBooks | | +#Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 +| | | | (%) | (%) | | (%) (%) | | | (%) (%) | +| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ------- | +Stable 1530 0.4017 0.8664 71.76 68.96 432 0.2368 0.5990 82.56 86.89 1876 0.3171 0.8403 76.64 72.56 +Moderate 1354 0.4773 0.8309 60.09 46.26 436 0.3921 0.6628 64.16 47.02 1827 0.4495 0.8059 66.73 55.45 +Diverse 1444 0.6632 1.1198 51.13 45.95 428 0.5536 0.9607 57.10 59.27 1838 0.6199 1.0603 49.52 46.35 +Table10: PerformanceofDUETunderdifferentlevelsofuserpreferencediversity(measuredbyratingvariance). +#Samp. denotesthenumberofsamplesineachgroup. \ No newline at end of file diff --git a/q166/random_k2/random_2.md b/q166/random_k2/random_2.md index 7f06b970eb191d018409053c543e59811fc6cbd9..b408d57341c4f4ff52ea6f83833a4f9892d91ec0 100644 --- a/q166/random_k2/random_2.md +++ b/q166/random_k2/random_2.md @@ -1,1186 +1,1349 @@ - TABLE OF -CONTENTS - -Is Graduate School Right for You?............................. - -4 - -What to Expect from Graduate School .................... - -5 - -Why UTRGV?................................................................ - -6 - -Institutional Accreditations ...................................... - -Graduate Programs ..................................................... - -Graduate Admissions ................................................ - -Research Centers & Key Research Areas ................. - -Costs & Financing ....................................................... - -Important Contact Information ............................... - -8 - -10 - -12 - -14 - -16 - -18 - -2 - -3 - - IS GRADUATE -SCHOOL -Right for You? -Right for You? - -Your undergraduate education gave you a strong foundation of knowledge and skills to start -your career. These days, however, to become an expert in your field, a graduate education is -often required. - -Specialized skills that you’ll learn and practice in graduate school are the key to career growth, -a better income, and unlocking your full potential. More and more, employers look for hiring -candidates to hold a graduate degree that demonstrates their expertise. - -Even if you’re currently satisfied with your career, you might have a passion for lifelong -learning or want to take a deeper dive into a particular subject. Investing in self-development is -never a bad decision, and a graduate degree can create surprising benefits for you in the future. - -As a graduate student, you’ll connect with supportive faculty who are leaders in their fields, -as well as other motivated graduate students who share similar goals and interests. You have -the flexibility to choose the pathway that’s right for you, including research, accelerated online -master’s programs, on-campus master’s or doctoral programs, and certificates. - -WHAT TO -EXPECT FROM -Graduate School? -Graduate School? - -• You can expect to be more responsible for your success than you were as a an - -• You’ll need to plan ahead to ensure you don’t miss out on required classes. - -undergraduate. You may even help with research or teaching. - -Since some classes are only offered, for example, in the spring, you will prepare -a graduation plan to guide you towards on-time completion of your program. - -• You’ll take rigorous classes alongside driven classmates like yourself. Being in - -graduate school raises the bar for expectations. You’ll need to maintain at least -a 3.0 GPA. - -• Unlike large undergraduate classes with hundreds of students, graduate - -classes are smaller. You will be on a first-name basis with your professors, who -will be both instructors and valuable mentors. - -• As an undergraduate, you may have assisted graduate students or faculty with -their research. As a graduate student, depending on your program, you’ll be -expected to carry out your own research, present it, and publish it. - -• If you’re taking classes on campus, you can expect most of them to be in the -evening. Many graduate students have jobs and families, so most on-campus -graduate classes are offered after 5 pm, twice a week. Online students need to -prioritize and allocate time for classwork in their schedules. - -• You can expect to be proud of your hard work and for pursuing your passion. - -Get ready to positively transform your community and the world. - -4 - -5 - - UTRGV? - -WhyWhy - -A high-quality education at an affordable price - -Graduate school at UTRGV is the perfect fit for - -ambitious leaders, dreamers, researchers, educators, - -and creators. - -We welcome you to join our vibrant community, -whether on campus or online. This is your opportunity -to leap forward in your career with an excellent -graduate education that’s both rigorous and affordable. - -Pursuing a graduate degree is a commitment to your -future, and we understand that our students have -different learning styles and real-world responsibilities. -We have options to suit your lifestyle and give you the -flexibility to choose between 100% online or on-campus -programs. All students receive the same high level of -training and support from faculty and staff. - -UTRGV offers unique research and resources through -the STARGATE partnership with SpaceX, the Coastal -Studies Laboratory on South Padre Island, and UT -Health RGV clinics from our School of Medicine. - -The Valley itself has a low cost of living that’s 10.2% -below the national average, with easy access to the -beautiful Gulf Coast. - -Demand for trained professionals in the Rio Grande -Valley, the United States, and around the world -continues to grow, and UTRGV equips graduate students -with the necessary competitive edge. - -Established in - -2015 - -4th Largest - -institution in -the UT System - -Enrollment Numbers 2020: -Graduate - -Overall - -32,618 - -5,122 - -UTRGV RANKINGS -UTRGV -RANKINGS -19th -6th - -1st - -In net price among -national universities by -Washington Monthly (2018) - -In the nation for top master’s -universities by Washington -Monthly (2019) - -Best online college in Texas by -OnlineColleges.com (2019) - -3rd - -Most affordable university in -America by BestValueSchools.com -(2018) - -1st - -4th - -19th - -79th -6th - -Best college in Texas by -BestColleges.com (2018) - -In net price among -national universities by -Washington Monthly -(2018) - -In the nation for top -master’s universities by -Washington Monthly -(2019) - -Best online college -Among 300+ national universities -in Texas by -by Washington Monthly (2018) -OnlineColleges.com -(2019) - -3rd - -in the nation among public -institutions for least student -Most affordable -debt according to U.S. News -university in America -& World Report (2020) -by BestValueSchools.com -(2018) - -1st - -4th - -3rd -79th - -Best college in Texas -by BestColleges.com -(2018) - -America’s Best Bank for the -Buck Colleges: South by -Washington Monthly (2020) - -Among 300+ national -universities by -Washington Monthly -(2018) - -QUALITY . FLEXIBILITY . AFFORDABILITY - -• According to the U.S. Bureau of Labor Statistics, 101 occupations require a graduate degree for entry. - -• The number of jobs requiring a master’s degree is projected to increase by 17% between 2016 and 2026. - -• Students who obtain master’s degrees earn, on average, $12,000 more in entry-level positions. - -• UTRGV offers graduate programs that align with the fastest growing careers. - -% change -2018-2028 - -Medium annual wage (2018) - -2018 Fastest Growing Occupations - -Occupational therapy assistants - -Information security analysts - -Physician assistants - -Statisticians - -Nurse practitioners - -Speech-language pathologists - -Physical therapist assistants - -Genetic counselors - -Mathematicians - -Operations research analysts - -Software developers, applications - -Health specialties teachers, postsecondary - - 33.1 - - 31.6 - - 31.1 - - 30.7 - - 28.2 - - 27.3 - - 27.1 - - 27.0 - - 26.0 - - 25.6 - - 25.6 - - 23.2 - - $60,220 - - $98,350 - - $108,610 - - $87,780 - - $107,030 - - $77,510 - - $58,040 - - $80,370 - - $101,900 - - $83,390 - - $103,620 - - $97,370 - - $44,630 - - $50,090 - - $87,930 - - $60,280 - - $63,120 - - $73,490 - -Substance abuse, behavioral disorder, and mental health counselors - - 22.5 - -Marriage and family therapists - -Physical therapists - -Respiratory therapists - -Market research analysts and marketing specialists - -Nursing instructors and teachers, postsecondary - - 22.3 - - 21.9 - - 20.8 - - 20.4 - - 20.0 - -Bureau of Labor Statistics, U.S. Department of Labor, Occupational Outlook Handbook, [Fastest Growing Occupations], on the Internet -at [https://www.bls.gov/ooh/fastest-growing.htm] (visited [October 2019]).”” * U.S. Bureau of Labor Statistics - -55+ - -ON-CAMPUS AND ONLINE -MASTER’S PROGRAMS -16-week-long semesters - -Full-time students (9 hours/semester) -can finish in as little as 2 years - -25 - -ACCELERATED ONLINE - -PROGRAMS -7-week-long modules - -Full-time students (6 hours/module) -can finish in as little as 1 year - -6DOCTORAL PROGRAMS - -40CERTIFICATE PROGRAMS - -SCHOLARSHIPS & ASSISTANTSHIPS -AVAILABLE - -$0 - -APPLICATION FEE - -Except: Occupational Therapy, Physician Assistant -Studies, Communication Sciences and Disorders, -Psychiatric Mental Health Nurse Practitioner, and -Physician Assistant Studies - Bridge. - -6 - -7 - - INSTITUTIONAL -Accreditations -Accreditations - -The University of Texas Rio Grande Valley (UTRGV) is accredited by the Southern Association of Colleges and - -Schools Commission on Colleges to award bachelor’s, master’s, and doctoral degrees. - -Individual Program Accreditations and Recognitions - -• Robert C. Vackar College of Business & Entrepreneurship - AACSB International, The Association to Advance - -• Department of Physician Assistant Studies - Accreditation Review Commission on Education for the - -Collegiate Schools of Business - -Physician Assistant (ARC-PA) - -• Educational Leadership (EdD) and Curriculum and - -Instruction (EdD) - -• School of Rehabilitation Services & Counseling - Council for Accreditation of Counseling and Related - - Admitted to the Carnegie Project on the Education - -Educational Programs - -Doctorate (CPED) - -• Social Work (MSSW) - Council on Social Work Education (CSWE) - -• Music (MM) - National Association of School of Music (NASM) - -• Department of Occupational Therapy - Accreditation Council for Occupational Therapy - -Education (ACOTE) - -• Department of Communication Sciences & Disorders - Council on Academic Accreditation in Audiology and -Speech-Language Pathology of the American Speech- -Language-Hearing Association - -• Master’s Degree in Nursing and Post-Graduate - -Psychiatric Mental Health Nurse Practitioner Certificate - - Commision on Collegiate Nursing Education - -• Counseling (MEd) - Council for Accreditation of Counseling and Related - -Educational Programs - -8 - -9 - - GRADUATE PROGRAMS - -COLLEGE OF EDUCATION -AND P-16 INTEGRATION -utrgv.edu/cep -Bilingual Education (MEd)2 -Board Certified Behavior Analyst - -(Certificate) - -Certificado Avanzado en Educación - Bilingüe 2 -Counseling (MEd) -Curriculum & Instruction (MEd) - Content Specialization in English (MEd)3 - Content Specialization in Mathematics (MEd)3 - Content Specialization in Elementary Math and - -Science Education (MEd)3 - - Content Specialization in Mathematics - -Education (MEd)3 - -Curriculum and Instruction (EdD) -Digital Literacy Leader (Certificate)2 -Early Childhood Education (MEd) -Educational Leadership (EdD) -Educational Leadership (MEd)4 -Educational Technology (MEd)3 -E-Learning (Certificate)3 -Higher Education Teaching (Certificate)3 -Online Instructional Designer (Certificate)3 -Principal Credential Certification (6 Credit Hours) - -(Certificate) 4 - -Principal Credential Certification (21 Credit Hours) -Reading and Literacy (MEd) 2 -Reading Specialist Certification (Certificate) 2 -School Psychology (MA) -Secondary Bilingual Education (Certificate) 2 -Special Education (MEd)3 -Superintendent Certification (Certificate) -Teacher Leadership (MEd)4 -Technology Leadership in Education - -(Certificate)3 - -TxVSN Digital Literacies (Certificate)2 - -COLLEGE OF ENGINEERING -AND COMPUTER SCIENCE -utrgv.edu/coec -Civil Engineering (MS) -Computer Science (MS) -Electrical Engineering (MSE) -Engineering Management (MS) 4 -Information Technology (MS) -Manufacturing Engineering (MSE) -Materials (Certificate) -Mechanical Engineering (MSE) -Mechanics and Design (Certificate) -Thermal Fluid Science (Certificate) - -COLLEGE OF FINE ARTS -utrgv.edu/cofa -Art (MFA) -Interdisciplinary Studies (MA) - Concentration in Art History -Creative Writing (MFA) -Design (Certificate) -Latin American Art History (Certificate) -Music (MM) - -COLLEGE OF HEALTH PROFESSIONS -utrgv.edu/coha -Bilingual Counseling (Certificate)2 -Bioethics (MS)2 -Clinical Rehabilitation Counseling (MS) -Clinical Laboratory Sciences (Certificate)3 -Communication Sciences and Disorders (MS) -Cooperative Pharmacy Program (with UT Austin) -Exercise Science (MS) -Health Sciences (MS)3 - Concentration in Clinical Laboratory Sciences3 - Concentration in Healthcare Administration3 - Concentration in Health Professions Education - -and Technology3 - - Concentration in Healthcare Informatics3 - Concentration in Nutrition3 -Health Science Nutrition (Certificate)3 -Healthcare Informatics (Certificate)3 -Kinesiology (MS)3 -Occupational Therapy (MS) -Physician Assistant Studies (MPAS) -Physician Assistant Studies - Bridge - Program (MPAS)2 -Rehabilitation Counseling (PhD) - -ROBERT C. VACKAR COLLEGE OF -BUSINESS & ENTREPRENEURSHIP -utrgv.edu/cobe -Accountancy (MAcc)4 -Advanced Business Administration - -(Certificate)4 - -Business Administration (MBA) - General4 - Specialization in Accounting (MBA)4 - Specialization in Business Analytics (MBA)4 - Specialization in Entrepreneurship (MBA)4 - Specialization in Financial Planning (MBA)4 - - Specialization in Health Care Administration - -(MBA)4 - - Specialization in International Trade (MBA)4 - Specialization in Marketing (MBA)4 -Business Administration (PhD) -Business Administration in Spanish (MBA)3 -Business Analytics (Certificate) -Business Analytics (MS)3 -Entrepreneurship (Certificate) -Healthcare Administration Leadership - -(Certificate) - -International Sustainability Management - -(Internal Certificate)5 - -User Experience and User Interface Design (UX/UI) - -(Certificate)3 - -SCHOOL OF NURSING -utrgv.edu/nursing -Family Nurse Practitioner (MSN) -Nursing Administration (MSN)3 -Nursing Education (MSN)3 -Psychiatric Mental Health Nurse - Practitioner (Certificate)2 - -SCHOOL OF SOCIAL WORK -utrgv.edu/social work -Social Work (63-Hour) (MSSW) -Social Work (MSSW)4 - Concentration in Administration and - -Community Practice (Advanced Standing) -(MSSW)4 - - Concentration in Direct Practice with Latino -Individuals, Families, and Groups (Advanced -Standing) (MSSW)4 - -COLLEGE OF LIBERAL ARTS -utrgv.edu/cla -Advanced Placement Spanish Literature - -(Certificate)2 - -Board Certified Behavior Analyst - -(Certificate) - -Clinical Psychology (MA, PhD) -Communication (MA) -Communication Training and Consulting - -(Certificate) - -Court Interpreting (Certificate)2 -Criminal Justice (MS)4 -Disaster Studies (MA) -English (MA) -English as a Second Language (MA) - -COLLEGE OF LIBERAL ARTS cont. -Experimental Psychology (MA) -Gender and Women’s Studies (Certificate) -Healthcare Interpreting (Certificate)2 -History (MA) -Interdisciplinary Studies (MA) - Concentration in Anthropology - Concentration in English - Concentration in History - Concentration in Mexican American Studies -Literary Translation (Certificate)2 -Localization and Audiovisual Translation - -(Certificate)2 - -Media Relations and Strategic - Communication (Certificate) -Mexican American Studies (Certificate) -Political Science (MA) -Public Affairs (MPA)4 -Secondary English Language Arts (Certificate) -Sociology (MS) -Spanish (MA) -Spanish Translation and Interpreting - -(MA, Certificate) 2 - -COLLEGE OF SCIENCES -utrgv.edu/cos -Agricultural, Environmental and Sustainability - -Sciences (MS) - -Applied Statistics and Data Science (MS) - -Biochemistry and Molecular Biology (MS) - -Biology (MS) - -Chemistry (MS) - -Cooperative PhD in Physics (with UT Arlington) - -Mathematics (Accelerated BS/MS - - 5-Year Program, MS) - -Ocean, Coastal, and Earth Sciences (MS) - -Physics (MS, PhD) -Interdisciplinary Studies (MS) - - Concentration in Science and Technology - -Instructional method is traditional -(face-to-face) unless otherwise noted. -2 Online (16-week Semesters) -3 Accelerated Online (7-Week Courses) -4 On-Campus & Accelerated Online Options -5 Online (16-week semester) and Accelerated - Online Options - - Graduate -Graduate -ADMISSIONS - -While most of our graduate programs have rolling admissions, ideally, you should begin the application -process several months in advance to ensure that you submit all documents on time and to have the best -chance for scholarships and assistantships. -Admissions Checklist: - -Apply online: utrgv.edu/gradapply -Get a certified English translation of your educational records - -Send your official college or university transcripts - -Submit official English Proficiency exam scores. (TOEFL, IELTS or Duolingo English Test) -Send your official GRE or GMAT scores (if applicable)*Note: Due to the impacts of the coronavirus pandemic on -testing centers across the world, we are temporarily suspending the GRE and GMAT test score requirements for - -most graduate programs until Fall 2021. -Other documents requested by the program (varies by program). Review requirements at utrgv.edu/grad -The application fee is waived for all graduate programs, except for Physician Assistant Studies, - -Occupational Therapy, Communication Sciences & Disorders, Physician Assistant Studies Bridge Program, - -and Psychiatric Mental Health Nurse Practitioner - -*If you will be pursuing your degree as an *International Student, please complete the following steps with International -Admissions & Student Services to finalize your admissions process.*International Student is defined as someone -who is not a U.S. citizen or U.S. resident and has a visa status or will require one to pursue their studies in a U.S. higher -education institution. - -Submit financial documentation - -Submit immigration documents - -To check the status of your application and review the final decision, visit the self-service center at utrgv.edu/selfservice. - -FAQ FREQUENTLY ASKED QUESTIONS - -Does my program require the GRE or GMAT? -Not all graduate programs require an entrance exam. -Visit utrgv.edu/graduate to find out -if your program requires one. - -*Note: Due to the impacts of the coronavirus pandemic - -on testing centers across the world, we are temporarily - -suspending the GRE and GMAT test score requirements for - -most graduate programs until Fall 2021. - -Where can I register for the GRE? What is UTRGV’s GRE -code? -You can register to take the GRE at ets.org/gre and send us -your scores using the GRE Code 6570. - -Where can I register for the GMAT? What is UTRGV’s -GMAT code? -You can register to take the GMAT at mba.com and send us -your scores using the GMAT Code 9P3. - -I attended a legacy institution (UT Pan American, UT -Brownsville, or Texas Southmost College). -Do I need to request those transcripts to be sent to -UTRGV? -No, we have access to all transcripts from our legacy -institutions and can upload them to your application. - -I started a graduate application. How do I log back -into my application? -Visit utrgv.edu/selfservice to log in with your username and -password that you previously -created. Do not create another application. - -Where do I send my official transcripts? -You can get them sent electronically to gradapps@utrgv.edu -or mail them to: The University of Texas Rio Grande Valley -The Graduate College -Marialice Shary Shivers Bldg. 1.158 -1201 W. Univesity Dr., Edinburg, TX 78539-2999 - -What are the minimum English Proficiency Exam -scores? -TOEFL minimum: 550 paper based, 79 internet based -IELTS minimum: 6.5 -Duolingo English Test score minimum of 105 -(Applicants with a score of 105 or above may also be -required to participate in a Zoom or Skype interview before -receiving a final decision.) - -How long is a TOEFL or IELTS score valid? -The TOEFL and the IELTS exam scores have a two year -validity period. - -Do you have exemptions for the English Proficiency -Exam (TOELF or IELTS)? -To view a list of exempt countries visit UTRGV.edu/grad - -Do I need to get my transcripts evaluated by an -evaluation agency? -We accept evaluations from the Foreign Credentials Services -of America (FCSA) or the World Education Services (WES). -However, if you have not had your transcript evaluated -by FCSA or WES, you can send us your transcripts to be -evaluated by UTRGV. - -How many credit hours do I need to be enrolled if I -have a F1 visa? -Nine credit hours per semester is the minimum; please -contact the International Admissions and Student Services -Office at international@utrgv.edu for -more details. - -Is proof of health insurance required for -international students? -Medical insurance is mandatory for any enrolled -international students on non-immigrant visa J-1, J-2, F-1, -F-2. An insurance waiver is available for students with -medical coverage. Although, medical insurance is strongly -recommended, F-1/F-2 Mexican students are exempt from -this health insurance requirement. For information on -waivers, contact the International Admissions and Student -Services office at international@utrgv.edu - -Who should I contact to assist me with inquires on -the student visa process? -International Admissions and Student Services -1-(956) 882-7092 | 1- (956) 665-2922 -utrgv.edu/iass | international@utrgv.edu - -12 - -13 - - Research -CENTERS - -At UTRGV, we have 15 research centers and -institutes - -• Center for Advanced Radio Astronomy (CARA) -• Center for Excellence in STEM Education -• University Transportation Center for Railway Safety (UTCRS) -• Texas Manufacturing Assistance Center (TMAC) -• South Texas Industrial Assessment Center -• Spacecraft Tracking and Astronomical Research into Gigahertz - Astrophysical Transient Emission (STARGATE) -• Nanotechnology Center of Excellence -• The Institute for Neuroscience -• Center for Vector-Borne Disease (CVBD) -• Coastal Studies Laboratory -• Center for Sustainable Agriculture & Rural Advancement (SARA) -• Center for Gravitational Wave Astronomy (CGWA) -• The South Texas Diabetes and Obesity Institute (STDOI) -• Dr. Cristina V. Torres Memorial Astronomical Observatory -• Center for Innovation and Commercialization - -KEY RESEARCH -Areas - -• Sustainable Agriculture - - and Food Systems - -• Diabetes, Obesity, and Human - - Genetics - -• Gravitational Wave and - -• Advanced Renewable - - Advanced Radio Astronomy - - Energy - -• Space Science Technology - -• Cell and Molecular - - (STARGATE) - -• Material Science and - - Engineering - -• Nanoscience and - - Nanotechnology - -• Optics - -• Coastal Resilience - - and Disaster Studies - -• Ocean Intelligence - - Biology - -• Biophysics - -• Computational Biology - -• Applied Statistics and - - Data Science - -• Mathematics (nonlinear - - waves, fluid dynamics, - - solitons, integrable - - systems) - -14 - -15 - - Costs & -FINANCING - -Scholarships* -UTRGV has several scholarship opportunities that use the same scholarship application. Scholarships are highly -competitive, so be sure to apply early. Visit utrgv.edu/applyscholarships for more information and to start your -application. - -Graduate College Scholarships - -Each semester the Graduate College may offer a variety of scholarships, dependent upon funding. To view - -what scholarships are currently being offered visit utrgv.edu/gradscholarships. - -Assistantships - -Assistantships give students the opportunity to work on campus (up to 20-hour work weeks) and are - -based on college/institutional demands and funds. Three types of assistantships are offered: - -• Teaching - -• Research - -• Graduate - -Please see your program director about the different assistantships available in your - -field of study. Visit careers.utrgv.edu for additional graduate assistantship opportunities. - -Texas Resident Tuition Rate -The Texas resident tuition rate may be granted to students who receive a teaching or -research assistantship in their discipline or who are awarded a competitive scholarship -of over $1,000. - -Payment Options -UTRGV provides payment plans to assist with the cost of tuition and fees. Eligible students may elect one of two payment -assistance options during the term of registration: - -• Installment Plan: This plan is available to any student unable to pay the full amount or partial amount due by the - university due dates. Installment plans consist of 4 payments. - Note: This is not available for summer sessions and accelerated online students. - -• Emergency Tuition and Fee Loan: This plan defers the payment of qualified tuition and fee charges to a future date. - -You can set up payment plans online through your my.utrgv.edu account within the ASSIST application. For additional -information, contact student_billing@utrgv.edu or call 956-665-7824 or 956-665-2191. - -Estimated Cost of Attendance: Fall 2020 - Summer 2021 - -Online and On-Campus (16-Week Semester) Programs - -Residency - -Texas Resident - -Non-Resident/International - -Per 3-Credit-Hour Course - -Per 9-Credit-Hour Semester - -$1,481.39 - -$2,708.39 - -$3,894.17 - -$7,575.17 - -Accelerated Online (7-Week Module) Programs - -Program - -Per 3-Credit-Hour Course - -Total Cost - -Technology Leadership in Education (Certificate) - -$1,374.99 - -E-Learning (Certificate) - -$1,374.99 - -Online Instructional Designer (Certificate) - -$1,374.99 - -Principal Certification (Certificate) - -$1,374.99 - -$4,125 - -$5,500 - -$5,500 - -$9,625 - -Curriculum & Instruction (MEd) -Educational Leadership (MEd) -Educational Technology (MEd) -Special Education (MEd) - -Social Work 33-Hour (MSSW) - -Social Work (MSSW) (63 Hr) - -Accountancy (MACC) - -Business Analytics (MS) - -Health Sciences (MS) - -$1,374.99 - -$13,750 - -$1,333.32 - -$1,333.32 - -$1,416.66 - -$1,416.66 - -$1,333.32 - -$14,667 - -$27,999 - -$14,170 - -$15,584 - -$16,000 - -12 - -16 - -13 -17 - - Accelerated Online (7-Week Module) Programs - -Program - -Per 3-Credit-Hour Course - -Total Cost - -Health Sciences (Certificate) - -Kinesiology (MS) - -Nursing (MSN) - -Public Affairs (MPA) - -Criminal Justice (MS) - -Business Administration (MBA) - -Business Analytics (MS) - - International Sustainability Management -(Certificate) - -Entrepreneurship (Certificate - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,416.66 - -$1,416.66 - -$1,416.66 - -$1,416.66 - -User Experience and User Interface Design - (UX/UI) (Certificate) - -$1,416.66 - -$8,000 - -$16,000 - -$16,000 - -$17,000 - -$16,000 - -$9,625 - -$15,583 - -$5,666 - -$5,666 - -$5,666 - -IMPORTANT CONTACT -Information -Information - -GRADUATE SCHOLARSHIPS -gradscholarships@utrgv.edu - -CONTINUING EDUCATION -continuinged@utrgv.edu - -RESIDENCE LIFE -home@utrgv.edu - -INTERNATIONAL ADMISSIONS -international@utrgv.edu - -LANGUAGE INSTITUTE -international@utrgv.edu - -18 - -19 - +Undergraduate +Brochure +2025 + +About HKU Business School +HKU Business School strives to nurture first-class business leaders and foster academic +and relevant research to serve the needs of Hong Kong, Mainland China and the rest of +the world in the new Asia-led economy. As Asia’s premier international business school, +we engage leading scholars from all corners of the globe and they instil in students global +CONTENTS knowledge with an Asian perspective. +With the highest proportion of non-local undergraduate students across all Faculties at +HKU, four of our undergraduate programmes ranked amongst the University’s Top 10 +programmes in 2024#, HKU Business School attracts the best and brightest students +from Hong Kong and beyond. +| About HKU Business School | P. 1 | | +| ------------------------- | ---- | --- | +HKU Business School is fully accredited by the European Quality Improvement Systems +| Explore Your Options | P. 4 | | +| -------------------- | ---- | --- | +(EQUIS) and the Association to Advance Collegiate Schools of Business (AACSB). +| Enrich Your Life | P. 16 | | +| ------------------------------------------------------------------------------------- | ----- | --- | +| 2025 JUPAS Admission | P. 36 | | +# Based on average JUPAS admission scores in the best 5 subjects in HKDSE +| 2025 Non-JUPAS Admission | P. 38 | | +| -------------------------- | ----- | ------ | +| 2025 Dual Degree Admission | P. 39 | | +| Develop Your Career | P. 40 | VISION | +| Freshmen Sharing | P. 41 | | +To be a leading, globally-impactful academic institution of business and +economics +| Our Global Presence | P. 44 | | +| ------------------- | ----- | --- | +| Useful Websites | P. 45 | | +UNIQUE PROPOSITION +Deeply rooted in Hong Kong +Fully engaged with China +Truly international +MISSION STATEMENT +To provide world-leading business and economics education that +nurtures dynamic and responsible leaders +To advance the frontiers of knowledge through innovative and +globally-significant research +To create positive change in the world by partnering with communities +01 + +| Dean’s Message | | | | | | | | | | | Greetings from | | | | +| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Associate Dean (Undergraduate) +| | | | | | At HKU | Business | School, | we have | 11 | diverse | | | | | +| --- | --- | --- | --- | --- | -------------- | --------- | ----------- | --------- | ------ | -------- | --- | --- | --- | --- | +| | | | | | undergraduate | | programmes | and | three | dual | | | | | +Business education today is an evolving core that +| | | | | | bachelor’s | degrees | in | partnership | | with | | | | | +| --- | --- | --- | --- | --- | ----------- | -------- | --- | ------------ | --- | ----- | --- | --- | --- | --- | +builds upon learning from all academic disciplines and +world-renowned universities. Our new programmes +| | | | | | | | | | | | draws inspiration | from | both | successful and | +| --- | --- | --- | --- | --- | -------------- | ----------- | ---------- | ---- | --------- | --- | ------------------- | ----- | ----- | ---------------- | +| | | | | | in big data | analytics, | including | the | Bachelor | of | | | | | +unsuccessful business practices. Our undergraduate +| | | | | | Business | Administration | | in Accounting | | Data | | | | | +| --- | --- | --- | --- | --- | --------- | --------------- | --- | --------------- | --- | ----- | --- | --- | --- | --- | +curricula in six core academic areas equip students +Analytics, the Bachelor of Business Administration +with knowledge in diverse aspects of business and +(Business Analytics), and the Bachelor of Science in +economics, and prepare them to be creative leaders +Marketing Analytics and Technology, are tailored to +of tomorrow with global perspectives. +| | | | | | meet the | growing | market | demand | for | skilled | | | | | +| --- | --- | --- | --- | --- | --------------- | -------- | --------- | ---------------- | ---- | -------- | ----------------------------------------------------- | --- | --- | --- | +| | | | | | professionals | in | business | data analytics | | and AI | A hallmark of our School that separates us from the | | | | +| | | | | | transformation. | | | | | | others is our truly multi-disciplinary approach. HKU | | | | +I'm thrilled to introduce you to our dynamic and +| | | | | | | | | | | | Business | School’s tremendous | | resources offer | +| ----------------- | --------- | ----------- | --- | ------ | -------------------------------------------------- | --- | --- | --- | --- | --- | --------- | --------------------- | --- | ----------------- | +| forward-thinking | academic | community, | | where | We believe in learning by doing. Our partnerships | | | | | | | | | | +I hope that you will share my enthusiasm for the +exciting opportunities and transformative students with a wide array of choices to combine +with over 300 institutions and corporations in more programmes and my admiration for the people who +studies in economics and finance, accounting and +experiences await you. than 40 countries provide you with unparalleled have contributed to their development. I look forward +finance, accounting and data analytics, business and +experiential and enrichment opportunities. Whether to having you as a member of the HKU Business +As a part of the University of Hong Kong, which law, business and data analytics, and international +proudly ranks 17th globally in the 2025 QS World it's internships, exchange programmes, or industry School family and a future leader in the world of +business and global management, as well as dual +projects, these experiences will sharpen your skills business. +University Rankings, HKU Business School stands out degrees in collaboration with world-renowned +and broaden your mindset, preparing you to tackle +| as a beacon of excellence in business and economics | | | | | | | | | | | universities. | | | | +| ---------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | +real-world challenges and build a successful career. Warm regards, +in Asia. As the youngest Faculty of the University, we +Our faculty members engage you with their inspiring +| have been growing rather quickly to become the top | | | | | Joining HKU Business School means becoming part of | | | | | | | | | | +| --------------------------------------------------- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +teaching and bring their impactful research into the +| business | school in Asia | in faculty | strength | and | a community that supports your growth and success. | | | | | | | | | | +| --------- | ----------------- | ------------ | --------- | ---- | --------------------------------------------------- | --- | --- | --- | --- | --- | ----------- | ----------- | ----------- | ------------ | +| | | | | | | | | | | | classroom. | They open | your mind | to diverse | +research productivity. +You’ll be surrounded by brilliant minds, engaging in +| | | | | | | | | | | | perspectives, | sharpen your | curiosity | to explore | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | -------------- | ---------- | ------------ | +stimulating discussions, and working on projects that +Our mission is to inspire and develop the next innovative ideas, and instil in you an appetite for +can make a real impact. The friendships and networks +generation of dynamic and responsible business acquiring new knowledge and extraordinary +you build here will last a lifetime and open doors to a +leaders. We offer a vibrant and challenging Professor Derek Chan +experiences throughout your academic journey and +myriad of career opportunities. Associate Dean (Undergraduate) +environment where you’ll be encouraged to push career development. Our students are among the +your limits, think critically, and innovate. Our team of If you're ready to embark on an exciting journey of brightest and most creative of their peers. Our alumni +| leading | scholars from | around | the | world and | | | | | | | | | | | +| -------- | --------------- | ------- | ---- | ----------- | --------------------------------------------------- | --- | --- | --- | --- | --- | ---------------------------------------------------------- | --- | --- | --- | +| | | | | | learning and personal development, I invite you to | | | | | | have gone on to forge brilliant careers in all sectors of | | | | +experienced and dedicated teachers brings a wealth explore what HKU Business School has to offer. I look the local, regional and international business +of knowledge and a global perspective to our forward to welcoming you to our community, where community, and formed a global network that you will +classrooms. Our wide range of top-quality degree future leaders are nurtured and dreams take flight. join for the rest of your life. +| programmes | is designed | to | equip | you with | | | | | | | | | | | +| ----------- | ------------- | --- | ------ | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Warm regards, +| cutting-edge | knowledge | and | technical | skills to | | | | | | | | | | | +| ------------- | --------------- | ------------------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| navigate | and excel in | the fast-changing | | global | | | | | | | | | | | +business landscape. +Professor Hongbin Cai +Dean +Chair of Economics +02 03 + +Explore Your Options HKU – Sciences Po Dual Degree Programme +A Wide Selection of Programmes Students will spend the first two years at Sciences Po in France and the third and fourth years +at HKU. Programmes at Sciences Po are taught in English (no French requirements for being +admitted) with intensive French language classes for all non-francophone students. +6755 Upon successful completion of the dual degree programme, students will be awarded the +Bachelor of Business Administration [BBA] +Bachelor of Business Administration (BBA) degree from HKU and a Bachelor’s degree from +Sciences Po. +Separate application is required for applying to the dual degree programme, in addition to the +The BBA programme combines strong functional training with intensive exposure to application under the JUPAS or International/Non-JUPAS Admissions Scheme. +| communication | skills, computer | applications, | and other social science | and/or science | +| -------------- | ------------------ | -------------- | ---------------------------- | ---------------- | +subjects. Students can choose to major in the following area(s): +Find out more: +https://admissions.hku.hk/programmes/university-collaborative- +programmes/hku-sciences-po-dual-degree-programme +| Entrepreneurship, | Finance | Human Resource | Information Systems | Marketing | +| ------------------ | ------- | -------------- | -------------------- | --------- | +Management HKU – University of British Columbia Dual +| Design and | | | and Analytics | | +| ---------- | --- | --- | ------------- | --- | + Innovation* +Degree Programme +BBA Curriculum Structure +Students will spend the first and final years at HKU and the second and third years at the +Credits University of British Columbia (UBC) in Canada, and graduate with a Bachelor of Business +Components +Administration from HKU and a Bachelor of Commerce from UBC upon successful completion +| A | University Graduation Requirements | | | 54 | +| --- | ---------------------------------- | --- | --- | --- | +of the programme. +| | Language Enhancement Courses | | | 18 | +| --- | ---------------------------- | --- | --- | --- | + Common Core Curriculum 36 Separate application is required for applying to the dual degree programme, in addition to the + Non-Credit Bearing Courses as required by the University - application under the JUPAS or International/Non-JUPAS Admissions Scheme. +| B | BBA Major | | | 78 | +| --- | -------------------- | --- | --- | --- | +| | Faculty Core Courses | | | 24 | +Find out more: +| | Disciplinary Core Courses / Disciplinary Electives | | | 48 | +| --- | -------------------------------------------------- | --- | --- | --- | + Capstone Course 6 https://admissions.hku.hk/programmes/university- +https://ug.hkubs.hku.hk/programme/bba collaborative-programmes/hku-university-of-british-columbia- +| C | Free Electives / Optional Major / Optional Minor(s) | | | 108 | +| --- | --------------------------------------------------- | --- | --- | --- | +dual-degree-programme +Total 240 +* Candidates must undergo a selection process arranged by the Programme Coordinator for declaring the EDI major. +Find out more: +04 05 + +Bachelor of Business Administration in +Accounting Data Analytics [BBA(ADA)] +Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares students for a +professional qualification in accounting. It provides students with an option to focus on data +analytics aspect in accounting instead of having a minor in finance. It is an excellent choice for +students who are interested in STEM with accounting and business applications and wish to +pursue their career in the relevant fields. The intensive accounting training enables +BBA(ADA) students to distinguish themselves from students of the other data analytics / +STEM programmes. Students will be equipped with comprehensive accounting and business +knowledge that prepares them to obtain professional qualifications in accounting. They will +also acquire computer programming and data analytics skills to meet the needs of the +accounting and business professions in today’s digital transformation economy. It is +accredited as an approved accountancy programme by: +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia +:erom +tuo +dniF +ada-abb/emmargorp/kh.ukh.sbukh.gu//:sptth +Bachelor of Business Administration in +Accounting and Finance [BBA(Acc&Fin)] +The BBA(Acc&Fin) programme prepares students for a professional qualification in +accounting and provides the broad-based education in accounting and finance which is +necessary to progress toward a leadership role in the financial sector. Students are required +to take professional core in accounting and a minor in finance, with an option to second major +in finance. Tailor-made professional preparation programmes and internship opportunities +with renowned accounting firms and financial institutions are available for talented students. +It is accredited as an approved accountancy programme by: +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia +https://ug.hkubs.hku.hk/programme/bba-acc-and-fin +Find +out +more: +6781 6781 +BBA(Acc&Fin) Curriculum Structure +Components Credits +A University Graduation Requirements 54 +Language Enhancement Courses 18 BBA(ADA) Curriculum Structure +Common Core Curriculum 36 +Non-Credit Bearing Courses as required by the University - Components Credits +B Professional Core in Accounting 108 A University Graduation Requirements 54 +Faculty Core Courses 24 Language Enhancement Courses 18 +Disciplinary Core Courses 78 Common Core Curriculum 36 +Capstone Course 6 Non-Credit Bearing Courses as required by the University - +C Minor in Finance* 36 B Professional Core in Accounting Data Analytics 156 +Disciplinary Core Courses 12 Faculty Core Courses 24 +Disciplinary Electives 24 Disciplinary Core Courses 114 +Disciplinary Electives 12 +D Free Electives / Optional Minor 42 Capstone Course 6 +Total 240 C Free Electives 30 +Total 240 +*Students may opt to pursue a Major in Finance: +Components Credits +A University Graduation Requirements 54 +B Professional Core in Accounting 108 +C Major in Finance 78 +Disciplinary Core Courses 30 +Disciplinary Electives 18 +Free Electives 24 +Capstone Course 6 +Total 240 +06 07 + +| | 6793 | | 6808 | | | +| --- | ---- | --- | ---- | --- | --- | +Bachelor of Business Administration Bachelor of Business Administration (Law) +(Business Analytics) [BBA(BA)] and Bachelor of Laws [BBA(Law)&LLB] +The BBA(BA) programme is crafted to meet the growing industry demand for talents in the Jointly offered by the HKU Business School and the Faculty of Law, the BBA(Law)&LLB +business analytics field. Its curriculum offers students a comprehensive coverage of both integrated double-degree programme blends two strategic disciplines into a single +technical and managerial skill sets. Students will learn a wide spectrum of knowledge from programme for the bright and highly motivated. It is an excellent choice for students seeking +the disciplines of information technology, data science, business statistics and management. a broader education with parallel concentration on legal and business studies. Graduates of +Core courses include programming language, database management, introduction to the programme will be awarded two degrees — BBA(Law) by the HKU Business School, and +business analytics, management information systems, decision and risk analysis, etc. They LLB by the Faculty of Law. After successful completion of the LLB programme, students may +help students build the foundations for advanced topics in senior years. Students also have apply for admission to the Postgraduate Certificate in Laws (PCLL) programme provided that +the option to choose from analytics technique courses and business application courses. The they meet the relevant admission requirements. +aim is to provide students with state-of-the-art experiences in artificial intelligence, big data, +The programme offers business stream and accounting stream to students in their business +social media, and other latest development. Such design helps train well-rounded students +studies. The accounting stream is accredited by: +who will be more adaptable in their career development. The programme is suitable for +students who have a passion for problem-solving through analytics, and who aim to develop +The Hong Kong Institute of Certified Public Accountants (HKICPA) +their career path in analytics. +The Association of Chartered Certified Accountants (ACCA) +| | BBA(BA) Curriculum Structure | | CPA Australia | | | +| --- | ---------------------------- | --- | ------------- | --- | --- | +BBA(Law)&LLB Curriculum Structure (Business Stream) +Credits +Components +| | | | Components | Credits | | +| ---------------------------------------- | ---------------------------------------------------------- | --- | ---------------------------------------------------------- | ------- | ------------------------------------------------- | +| | A University Graduation Requirements | 54 | | | | +| | Language Enhancement Courses | 18 | | | | +| | | | A University Graduation Requirements | 36 | | +| | Common Core Curriculum | 36 | | | | +| | | | Language Enhancement Courses | 12 | | +| | Non-Credit Bearing Courses as required by the University | - | | | | +| | | | Common Core Curriculum | 24 | | +| | | | Non-Credit Bearing Courses as required by the University | - | | +| | B Major in Business Analytics | 96 | | | | +| | Faculty Core Courses | 24 | | | | +| | | | B BBA Major | 78 | | +| | Disciplinary Core Courses | 42 | | | | +| | | | Faculty Core Courses | 24 | | +| | Disciplinary Electives (Analytics Techniques) | 12 | | | | +| | | | Disciplinary Core Courses / Disciplinary Electives | 48 | bll-dna-wal-abb/emmargorp/kh.ukh.sbukh.gu//:sptth | +| | Disciplinary Electives (Business Applications) | 12 | | | | +| | | | Capstone Course | 6 | | +| | Capstone Course | 6 | | | | +| | | | C Professional Core in Law | 156 | | +| https://ug.hkubs.hku.hk/programme/bba-ba | C Free Electives / Optional Major / Optional Minor(s) | 90 | | | | +| | | | Disciplinary Core Courses | 108 | | +| | | | Disciplinary Electives | 42 | | +Total 240 +| | | | Capstone Course | 6 | | +| --- | --- | --- | ---------------------------------- | --- | --- | +| | | | D Free Electives / Optional Minor | 30 | | +Total 300 +BBA(Law)&LLB Curriculum Structure (Accounting Stream) +Credits +Components +| Find out more: | | | | | :erom tuo dniF | +| -------------- | --- | --- | ------------------------------------- | --- | -------------- | +| | | | A University Graduation Requirements | 36 | | +| | | | B Professional Core in Accounting | 108 | | +| | | | Faculty Core Courses | 24 | | +| | | | Disciplinary Core Courses | 78 | | +| | | | Capstone Course | 6 | | +| | | | C Professional Core in Law | 156 | | +Total 300 +| 08 | | | | | 09 | +| --- | --- | --- | --- | --- | --- | + +| 6896 | | | | | 6767 | | +| ---- | --- | --- | --- | --- | ---- | --- | +Bachelor of Business Administration in Bachelor of Economics [BEcon] +International Business and Global +Management [BBA(IBGM)] +BBA(IBGM) is an intensive, premier programme in international business and global The BEcon programme trains students in economic theories, statistical tools, the application +management featuring a unique curriculum designed to evolve with and reflect the of theories to solve practical economic problems in trade, banking and the labour market, and +ever-changing demands of a globalised world. Exclusive elements such as overseas evaluation of the regional economic performance of China, Hong Kong and other economies. +exchange studies, a leadership boot camp, an Asia-Pacific study tour, an international field Students will be furnished with solid theoretical and empirical economic knowledge to +trip and global enrichment opportunities are strategically integrated into the curriculum, analyse and understand human behaviour and social interactions. The programme allows +whilst students are given the flexibility to pursue an in-depth inquiry into another specialised students a high degree of flexibility in planning their future career specialisation. +area of study as a second major. This cross-functional and personalised approach aims to +foster a selected group of outstanding students to become the world’s future leaders +BEcon Curriculum Structure +| equipped with | global perspectives, | cultural sensitivity, | social responsibility | and key | | | +| --------------- | ---------------------- | ----------------------- | ----------------------- | --------- | --- | --- | +Credits +| competencies to succeed in the international business environment. | | | | | Components | | +| ------------------------------------------------------------------ | --- | --- | --- | --- | ------------------------------------- | --- | +| | | | | | A University Graduation Requirements | 54 | +BBA(IBGM) Curriculum Structure Language Enhancement Courses 18 +| | | | | | Common Core Curriculum | 36 | +| --- | --- | --- | --- | --- | ------------------------ | --- | +Credits +Components Non-Credit Bearing Courses as required by the University - +| A | University Graduation Requirements | | | 54 | | | +| --- | ---------------------------------- | --- | --- | --- | --------------------------- | --- | +| | | | | | B Major in Economics | 78 | +| | Language Enhancement Courses | | | 18 | Faculty Core Courses | 24 | +| | Common Core Curriculum | | | 36 | Disciplinary Core Courses | 30 | + Non-Credit Bearing Courses as required by the University - Disciplinary Electives 18 +| | | | | | Capstone Course | 6 | +| --- | ----------------------------------------------------- | --- | --- | --- | ------------------------------------------------------ | --- | +| B | Major in International Business and Global Management | | | 96 | | | +| | Faculty Core Courses | | | 24 | | | +| | | | | | C Free Electives / Optional Major / Optional Minor(s) | 108 | +| | Disciplinary Core Courses | | | 54 | | | +| | Disciplinary Electives | | | 12 | | | +Total 240 +| | Capstone Course | | | 6 | | | +| --- | ---------------------------- | --- | --- | ----- | --- | --- | +| C | Second Major | | | 72-90 | | | +| D | Free Electives | | | 0-18* | | | +| E | Compulsory Overseas Exchange | | | - | | | +noceb/emmargorp/kh.ukh.sbukh.gu//:sptth +| | | | Total | 240 | | | +| --- | --- | --- | ----- | --- | --- | --- | +*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. +http://ibgm.business.hku.hk +Find out more: :erom tuo dniF +10 11 + +| 6767 | | 6860 | | | +| ---- | --- | ---- | --- | --- | +Bachelor of Economics and Finance Bachelor of Finance in Asset Management +[BEcon&Fin] and Private Banking [BFin(AMPB)] +BEcon&Fin is designed to meet the interests of students who are seeking a balanced study of The BFin(AMPB) programme aims to prepare students to enter the rapidly growing fields of +both economics and finance. The programme recognises the significance of economics both asset management and private wealth management in Hong Kong and Asia Pacific, or in +in its own right and as a foundation for the study of finance, and the significance of finance as banking and finance generally. During the first two years, students complete introductory +a career path for students who are keen on joining this profession. Students will receive solid courses in accounting, economics, mathematics, statistics, corporate finance, and +training in theories and applications of economics and finance. They will be equipped with the investments and portfolio analysis. In subsequent years, students take more specified and +skills to analyse human behaviour and social interaction, as well as to solve problems in the industry related courses including equity valuation and investment in specific asset classes +financial sphere. e.g. private equity, venture capital, hedge funds and digital assets, understanding of financial +statements and doing financial analysis, lending and credit in private wealth and corporate +BEcon&Fin Curriculum Structure +bank, financial regulations and compliance, current issues in finance, ESG in investment, +Components Credits institutional consulting and family office. The programme contains practical-oriented courses +taught by teachers with extensive experience in the finance industry and offers an array of +A University Graduation Requirements 54 career-oriented enrichment activities and networking opportunities with industry +| Language Enhancement Courses | 18 | | | | +| ------------------------------ | --- | --- | --- | --- | +practitioners. +| Common Core Curriculum | 36 | | | | +| ---------------------------------------------------------- | --- | --- | --- | --- | +| Non-Credit Bearing Courses as required by the University | - | | | | +BFin(AMPB) Curriculum Structure +| B Major in Economics and Finance | 96 | | | | +| --------------------------------- | --- | --- | --- | --- | +Credits +| Faculty Core Courses | 24 | Components | | | +| ------------------------------------------------------ | --- | ---------------------------------------------------------- | --- | --- | +| Disciplinary Core Courses | 42 | | | | +| | | A University Graduation Requirements | 54 | | +| Economics Disciplinary Electives | 12 | | | | +| | | Language Enhancement Courses | 18 | | +| Finance Disciplinary Electives | 12 | | | | +| | | Common Core Curriculum | 36 | | +| Capstone Course | 6 | | | | +| | | Non-Credit Bearing Courses as required by the University | - | | +| C Free Electives / Optional Major / Optional Minor(s) | 90 | | | | +| | | B Major in Asset Management and Private Banking | 96 | | +| | | Faculty Core Courses | 24 | | +Total 240 +| | | Disciplinary Core Courses | 48 | | +| --- | --- | --------------------------- | --- | --- | +| | | Disciplinary Electives | 18 | | +| | | Capstone Course | 6 | | +https://ug.hkubs.hku.hk/programme/becon-and-fin HKU - Peking University “Future Leaders” Dual +| | | C Free Electives / Optional Major / Optional Minor(s) | 90 | bpma-nfib/emmargorp/kh.ukh.sbukh.gu//:sptth | +| --- | --- | ------------------------------------------------------ | --- | -------------------------------------------- | +Bachelor’s Degree in Economics, Finance and +240 +Management Total +The programme combines the strengths of two leading business schools in Hong Kong and +Mainland China and is designed to nurture first-class business leaders and cultivate their +global perspective with intimate understanding of China. Students will spend the first two +years at HKU and the third and fourth years at Peking University (PKU), and be awarded a +Bachelor of Economics and Finance (BEcon&Fin) degree from HKU and a Bachelor of +Management (BMgmt) degree from PKU upon successful completion of the dual degree +programme. +Find out more: :erom tuo dniF +Separate application is required for applying to the dual degree programme, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. +Find out more: +https://ug.hkubs.hku.hk/admission/degree-programmes/hku-pku- +dual-degree +12 13 + +| 6846 | | | | | 6884 | | +| ---- | --- | --- | --- | --- | ---- | --- | +Bachelor of Science in Marketing Bachelor of Science in Quantitative +Analytics and Technology [BSc(MAT)] Finance [BSc(QFin)] +Marketing is critical to all businesses. It dictates whether a company will be successful. BSc(QFin) is a premium finance programme. Built on a solid foundation in economics and +Technology, big data, and digitization are fundamentally changing how marketers operate. finance, the programme equips students with the essential quant skills, including +The BSc(MAT) programme equips students with quantitative skills, technology know-how, mathematics, statistics, and programming, that are required in more advanced and +and business marketing acumen to pursue a career in the digital economy and modern sophisticated analyses, such as valuation models of structured products, risk management, +businesses. It targets both STEM students with an interest in business as well as business simulation and data analysis, financial engineering, and quantitative trading. +students interested in the technology space. Students will gain a solid foundation in +The programme goes beyond merely equipping students with quant skills. More importantly, +cutting-edge technical areas including computer programming, data science, and technology +it stresses on combining these skills with an understanding of the financial markets, the macro +by excelling in courses from the Faculty of Engineering and Faculty of Science. In addition, +environment, and the practical situations where financial managers work, with a balanced mix +students will master digital business, marketing and leadership through courses at the HKU +of quant and business courses. To nurture students’ soft skills and market acumen, the +Business School. This interdisciplinary approach will enable students to engage in a variety of +programme provides a wide range of value-enhancing activities, such as Round Tables with +jobs and functions such as the design and launch of new technology products, targeting +Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training workshops, +| consumers, digital | marketing, forecast | and predicting | markets, and | building a | | | +| -------------------- | --------------------- | ---------------- | -------------- | ------------ | --- | --- | +simulation trading workshops, and an international field trip to other international financial +technology-focused strategy. The programme aims to nurture future leaders in the field of +centres such as New York, London and Zurich. +technology and marketing who are able to use the latest technology, science, and data +analytics to support business and marketing decisions in the digital world. +BSc(QFin) Curriculum Structure +| BSc(MAT) Curriculum Structure | | | | | Components | Credits | +| ----------------------------- | --- | --- | --- | ------- | ------------------------------------- | ------- | +| Components | | | | Credits | | | +| | | | | | A University Graduation Requirements | 54 | +| | | | | | Language Enhancement Courses | 18 | +A University Graduation Requirements 54 Common Core Curriculum 36 + Language Enhancement Courses 18 Non-Credit Bearing Courses as required by the University - +| Common Core Curriculum | | | | 36 | | | +| ------------------------ | --- | --- | --- | --- | --- | --- | + Non-Credit Bearing Courses as required by the University - B Major in Quantitative Finance 96 +| | | | | | Faculty Core Courses | 24 | +| --- | --- | --- | --- | --- | ---------------------- | --- | +B Major in Marketing Analytics and Technology 96 Disciplinary Core Courses 54 +| Faculty Core Courses | | | | 24 | Disciplinary Electives | 12 | +| ------------------------------------------------------------ | --- | --- | --- | --- | ------------------------ | --- | +| Disciplinary Core Courses | | | | 42 | Capstone Course | 6 | +| Disciplinary Electives (Marketing Technology and Strategy) | | | | 12 | | | + nfiq-csb/emmargorp/kh.ukh.sbukh.gu//:sptth +https://ug.hkubs.hku.hk/programme/bsc-mat Disciplinary Electives (Marketing Analytics) 12 C Free Electives / Optional Major / Optional Minor(s) 90 +| Capstone Course | | | | 6 | | | +| ----------------- | --- | --- | --- | --- | --- | --- | +Total 240 +| C Free Electives / Optional Major / Optional Minor(s) | | | | 90 | | | +| ------------------------------------------------------ | --- | --- | ----- | --- | --- | --- | +| | | | Total | 240 | | | +Find out more: :erom tuo dniF +14 15 + +| Enrich Your Life | | | | | | | | International Field Trips | +| ---------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------- | +To enhance students' exposure to global businesses, the School organises overseas field trips +for students visiting corporate firms and mingling with C-suite Executives in some of the +HKU Business School fosters a culture of student initiative and implementation of ideas, both +world's most prominent financial centres, e.g. Australia, France, Japan, Singapore, Spain, +within and outside the campus, as it is these real-world engagements that set us apart. Zurich, New York City and London. +Experiences in learning beyond the classroom, extending into realms outside academia, +enable students to cultivate an understanding of social dynamics, cultural diversity, and +Visiting Programmes +collaborative intellect. These experiences contribute significantly to their personal growth, +Through the Yale Visiting International Students Programme, students are sent to spend a +interpersonal skills, and social development, shaping them into global citizens, lifelong +year at Yale University where they actively engage themselves in a multicultural society. +learners, and future business leaders. +With the generosity of the donors, some of our students are selected for the Junior Year +Abroad Programme and Visiting Students Programme to study as visiting students at +Pembroke College and St Peter’s College of University of Oxford respectively with +Overseas Enrichment Experiences substantial scholarships. The intellectual ambience of these top universities is truly +inspirational and conducive to learning. +Business Exchange Programme / +Overseas Career Opportunities +HKUWW Undergraduate Student Exchange Programme +Being the youngest Faculty in the University, our exchange Overseas career opportunities provide exceptional learning experiences to our students. In +collaboration with the DCAF - Geneva Centre for Security Sector Governance, a Traineeship +activities are phenomenal. Each year, the School plans to send +Project will be launched in 2025, and selected students will be sent to Geneva for a 6-month +| over 400 | students | for overseas | exchange | to top-notch | | | | | +| --------------------------------------- | --------- | -------------- | --------- | -------------- | -------------------- | --- | --------------------- | --------------------------------------------------------------------------------- | +| | | | | | 400+ | | 400+ | project related to the humanitarian business, unquestionably rewarding exposure. | +| universities to broaden their horizons. | | | | | | | | | +| | | | | | exchange partners | | Business School | | +| | | | | | across 40+ countries | | students on exchange | | +Beta Gamma Sigma Regional Leadership Conference +Business School Exchange Partners * Beta Gamma Sigma (BGS) is the international honour society that serves business +programmes accredited by the Association to Advance Collegiate Schools of Business +(AACSB International). With the establishment of the HKU BGS Chapter, the School registers +Denmark students annually for the BGS Regional Leadership Conference held in the U.S. where they +Canada +McGill University Copenhagen Business School can meet and bond with many other fellow students from the world's best business schools. +University of British Columbia Aarhus University The Leadership Summit is eye-opening and challenging which leads to a great journey of +| University of Toronto | | | | | | | | self-discovery. | +| --------------------- | --- | --- | --- | --- | --- | --- | --- | --------------- | +Western University +Simon Fraser University +Belgium +| | | | KU Leuven | | The Netherlands | | | | +| --- | --- | --- | --------- | --- | --------------- | --- | --- | --- | +University of Amsterdam +| | U.K. | | | | Erasmus University Rotterdam | | | | +| --- | ---- | --- | --- | --- | ---------------------------- | --- | --- | --- | +King's College London +| | University of Manchester | | | | Italy | Japan | | | +| --- | ------------------------ | --- | --- | --- | ------------------ | ---------------- | --- | --- | +| | | | | | Bocconi University | Kyoto University | | | +University of Edinburgh +| | University of Warwick | | France | | | | | | +| --- | --------------------- | --- | ------- | --- | --- | --- | --- | --- | +Toulouse School of Economics +| | | | ESSEC Business School | | | Singapore | | | +| --- | --- | --- | --------------------- | --- | --- | --------- | --- | --- | +National University of Singapore +Nanyang Technological University +Germany +U.S. +| New York University | | | Frankfurt School of Finance & Management | | | | | | +| --------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | +| University of Southern California | | | University of Mannheim | | | | | | +Australia +| University of Washington | | | Munich Business School | | | University of Sydney | | | +| ------------------------ | --- | --- | ---------------------- | --- | --- | -------------------- | --- | --- | +University of Georgia +Monash University +University of New South Wales +Middle East +Reichman University +* This list is non-exclusive. +16 17 + +Student Sharing +| | | | | | | Ng, Hoi Kiu | BEcon&Fin | | | | | +| --- | --- | --- | --- | --- | --- | ----------- | --------- | --- | --- | --- | --- | +Exchange at Peking University Guanghua School of Management +(China), 2023-24 +| | | | Kwok, Sin Ue | BBA(Law)&LLB | | The curriculum at Peking University was incredibly comprehensive. Each | | | | | | +| --- | --- | --- | ------------- | ------------ | --- | ----------------------------------------------------------------------- | --- | --- | --- | --- | --- | +Exchange at University of Edinburgh +class provided me with valuable knowledge and fundamental skills +(United Kingdom), 2023-24 +essential for my professional development in the business field. Engaging +| | | | University of Edinburgh has rich history, coupled | | | in discussions with both local and international students broadened my | | | | | | +| --- | --- | --- | -------------------------------------------------- | --- | --- | ----------------------------------------------------------------------- | --- | --- | --- | --- | --- | +| | | | with its modern facilities and renowned faculty, | | | horizons and deepened my understanding of China's business landscape. | | | | | | +Living in Beijing also allowed me to explore the city's rich history and +provided an ideal environment for my academic +| | | | growth and personal development. The campus | | | modernity, providing numerous invaluable learning experiences and | | | | | | +| --- | --- | --- | ---------------------------------------------- | -------------------- | ------------ | ------------------------------------------------------------------------- | ------------------------- | ------------------- | --------------- | --- | --- | +| | | | offered a wide range of resources and support | | | connections | to this vibrant urban | center. I highly | recommend all | | | +| | | | services for | exchange students, | making my | university students to participate in an exchange program at least once, | | | | | | +| | | | transition smooth | and enjoyable. | I actively | as it offers numerous important takeaways. | | | | | | +participated in welcoming events organized by the +university, where I met a group of international +| Hung, Lok Yin | BBA(Acc&Fin) | | | | | | | | | | | +| -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +friends. The combination of academic excellence, +Exchange at Aarhus University (Denmark), +| | | | cultural immersion, | and the | development of | | | | | | | +| --------------------------------------------------- | --- | --- | -------------------------------------------- | ---------------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| 2023-24 | | | essential life | skills has prepared | me for the | | | | | | | +| Studying at AU is surprisingly similar to studying | | | challenges and opportunities that lie ahead. | | | | | | | | | +at HKU, with the major difference being the +workload, which enabled me to delve into Danish +society from various angles. I met numerous +people and was captivated by the concept of +"hygge", embodying the essence of cherishing +| cozy moments | with loved ones. | I deeply | | | | | | | | | | +| -------------- | ------------------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +appreciate the notion that happiness can be +| found in simplicity | and purity. | Overall, the | | | | | | | | | | +| ---------------------- | ------------- | -------------- | --- | --- | --- | ----------- | --------- | --- | --- | --- | --- | +| | | | | | | Yu, Er Han | BBA(IBGM) | | | | | +exchange experience has been transformative for +Exchange at University of California, Irvine +me. Despite its brevity, it was an unforgettable +(United States), 2023-24 +and life-changing event that will forever be So, Yiu Hei Michael BEcon&Fin +cherished as a pivotal part of my youth. This journey not only broadened my academic International Committee of the Red Cross +| | | | | | | horizons but also enriched my personal and cultural | | | Traineeship, 2023-24 | | | +| --- | --- | --- | --- | --- | --- | ---------------------------------------------------- | -------------------------- | ---------------- | --------------------------------------------------- | --- | --- | +| | | | | | | insights, | marking a pivotal | chapter in my | During my ICRC traineeship, I had the privilege of | | | +| | | | | | | educational | voyage. The distinctive | academic | | | | +collaborating closely with the Geneva Centre for +| | | | | | | framework | facilitated in-depth | exploration of | | | | +| --- | --- | --- | --- | --- | --- | ---------- | ---------------------- | ---------------- | ----------------- | ----------- | --------------------- | +| | | | | | | | | | Security Sector | Governance | (DCAF) on a joint | +subjects and enhanced interaction with both faculty +| | | | | | | | | | project. Under | the supervision | of the ICRC's | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------- | ---------------- | +and peers. Throughout this period, overcoming +| | | | | | | | | | Economic | Adviser, I assisted | in the day-to-day | +| --- | --- | ----------------- | --- | --- | --- | ----------------------------------------------- | --- | --- | ----------- | ---------------------- | -------------------- | +| | | Lau, Sui Yu BBA | | | | | | | | | | +| | | | | | | challenges—from adjusting to a new educational | | | management | of this initiative, | which involved | +Exchange at University of New South Wales (Australia), 2023-24 +| | | | | | | system to | assimilating into a | diverse cultural | | | | +| --- | --- | --------------------------------------------------------------------- | --- | --- | --- | ------------------------------------------------- | ---------------------- | ------------------ | ---------- | ------------------------- | ----------- | +| | | | | | | | | | fostering | external communication, | conducting | +| | | UNSW has provided me with resources and incentives to motivate me to | | | | environment—fostered resilience and a heightened | | | | | | +research on conflict and security reporting trends, +sense of self-awareness, reinforcing my dedication +work on practical career skills, such as Excel training and presentation skills. and liaising with a diverse range of stakeholders. +| | | | | | | to pursuing | a career in globalized | business. | I | | | +| --- | --- | --- | --- | --- | --- | ------------- | -------------------------- | ---------- | --- | --- | --- | +The school also showcased innovation in its teaching style, being receptive This cross-organizational partnership underscored +to student feedback and staying abreast of current global topics. It has wholeheartedly recommend this exchange program the importance of adopting interdisciplinary +| | | been a valuable opportunity for me to venture beyond my comfort zone | | | | to my peers, as it ensures holistic development and | | | | | | +| --- | --- | --------------------------------------------------------------------- | --- | --- | --- | ---------------------------------------------------- | --- | --- | ----------- | ----------------- | ------------------- | +| | | | | | | | | | approaches | to address the | complex interplay | +| | | and expand my horizons outside the classroom. While this journey was | | | | creates lasting memories. | | | | | | +between business and human rights. The experience +challenging, it stands out as the most memorable experience of my life. significantly enhanced my ability to navigate +| | | | | | | | | | multicultural | environments | and work effectively | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | ------------- | ----------------------- | +with high-level international organizations. +18 19 + +Corporate Exposure +| | | | | | Vaidya, Samhith | BEcon&Fin | | | | | +| --- | --- | --- | --- | --- | ---------------- | --------- | --- | --- | --- | --- | +HKU-Accenture Business Consulting Programme 2023-24 +HKU Business School has a strong tie with corporate partners. To nurture students' business The HKU-Accenture Business Consulting Programme was +everything I needed - an opportunity to meet like-minded people +and market acumen, enrichment programmes are organised through the joint efforts of +corporate partnerships. and be mentored to think out of the box in a systematic manner. +The skills I take home, the friends I've made, and the mentors I +Enrichment Programmes in collaboration with Corporate Partners connected with are not just the biggest take aways, but also the +ability to apply my learnings from the programme onto my courses, +HKU Business School has been collaborating with corporate partners on initiatives to spice up +interviews, and case competitions. +students' career exploring journey. The partnership with Accenture enables students to gain +exclusive insights into business consulting through the HKU-Accenture Business Consulting +Programme. +"Venture into Future Career" has also been rolled out since 2020-21 in collaboration with +Cyberport and the Chartered Institute of Management Accountants. Students are given a +platform to mingle with professionals from different industries via a series of activities like +coffee chats and guest talks. This definitely helps students explore their career choices and +sort out their career direction. +Jiang, Yiguo Amory +| HKU Business School Mentorship Programme | | | | | | | | | BBA | | +| ---------------------------------------- | --- | --- | --- | --- | -------------- | --------- | --- | ------------------------------- | --- | --- | +| | | | | | Li, Mingfang | BBA(IBGM) | | HKU Business School Mentorship | | | +Leveraging the School's extensive network in the business and finance industry, we have +| | | | | | Venture into Future Career (VFC) | | | Programme 2023-24 | | | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | ----------------- | --- | --- | +developed our own mentorship programme to furnish students with opportunities to +Programme 2023-24 +There is so much I have learned from my mentor. +| enhance their | skills and knowledge | essential | for their future personal | and career | | | | | | | +| --------------- | ----------------------- | ---------- | ----------------------------- | ------------ | -------------- | ------------------------- | --------- | -------------------------------------------------- | --- | --- | +| | | | | | Participating | in the VFC programmre | provided | He provided numerous opportunities to participate | | | +development. Our students are empowered to connect to the real business world through +profound insights as professionals from diverse +direct and active interaction with seasoned professionals, executives and entrepreneurs. in the events organized by his company. I believe +| | | | | | industries | shared their personal | experiences, | that these hands-on experiences in the industry | | | +| --- | --- | --- | --- | --- | ------------------------------------------------- | ------------------------ | --------------- | ------------------------------------------------- | ------------------------- | ------------ | +| | | | | | enhancing | our understanding. | The dedicated | are the most valuable. Through my involvement in | | | +| | | | | | sessions on CV enhancement and interview skills | | | real events, | I had the opportunity | to witness | +| | | | | | before engaging with guest speakers and pursuing | | | firsthand how everything operated, particularly | | | +internships were particularly beneficial. Learning +observing how my mentor handled a variety of +| | | | | | from entrepreneurs | and seasoned | professionals | situations. There is no better way to learn about the | | | +| --- | --- | --- | --- | --- | -------------------- | -------------- | ------------------ | ------------------------------------------------------ | --- | --- | +| | | | | | was invaluable, | fostering | a rich learning | PR industry than by actively engaging in it myself. | | | +environment where networking, CV refinement, and +experiential learning were key. A standout moment +| | | | | | was the | engaging coffee | chat with industry | | | | +| --- | --- | --- | --- | --- | --------- | ----------------- | --------------------- | ----------------------- | ---------- | --- | +| | | | | | | | | Dong, Christina W Y | BFin(AMPB) | | +professional Mr. Darron Sun. The key lesson learned +| | | | | | was the | uniqueness of each | career journey, | HKU Business School Mentorship Programme | | | +| --- | --- | --- | --- | --- | ------------ | --------------------- | -------------------- | ----------------------------------------- | --- | --- | +| | | | | | emphasizing | the importance | of avoiding | 2023-24 | | | +| | | | | | comparisons | and highlighting | the potential for | | | | +My mentor invited me to participate in the annual +industry transitions or role changes regardless of +conference of the Institute of ESG and Benchmark, +| | | | | | one's career stage. | | | where I gained valuable insights into the growth of | | | +| --- | --- | --- | --- | --- | ------------------- | --- | --- | ---------------------------------------------------- | --- | --- | +established companies in ESG, greatly enhancing +| | | | | | | | | my understanding | in ESG-related | courses and | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------ | ---------------- | ------------- | +inspiring ideas for upcoming cases. Furthermore, +| | | | | | | | | my mentor | arranged a visit | to his company, | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | ------------------- | ------------------ | +providing updates on company valuation and +IPO-related systems. Exploring the latest ESG +| | | | | | | | | regulations | from HKEX has | sparked fresh | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------ | ---------------- | --------------- | +perspectives for my career trajectory planning. +20 21 + +Experiential Learning Programmes +Experiential learning is also embedded in our course curriculum. Business Consulting +Practicum offers business and economics students hands-on experience of working with +business clients in solving "real-life" problems relating to their studies, under the guidance of +the faculty teacher, Consultant-in-Residence and professional mentors. Business Case +Analysis and Presentation enables students to communicate and ask for practical advice +from senior executives through in-classes case presentations. Impact Lab gives students an +excellent opportunity to work directly for social ventures under the guidance of a teacher and +professional mentors. +Tsang, Tze Yeung Matthew BEcon&Fin +An intensive academic and cultural exchange experience, Doing Business in China, is +Business Consulting Practicum 2024 +offered to students where they can interact with peers from Peking University, learn about +the latest business and social environment of China, and explore the rich Chinese cultural BCP has been a rewarding experience that +heritage. Students can also acquire an in-depth understanding of the development and allowed me to apply my academic knowledge +latest economics, finance, creativity, innovation and entrepreneurship trends in the to real-world business challenges. Working +fast-growing Greater Bay Area by joining the Creativity, Innovation & Entrepreneurship directly with my mentor and client has +in China Programme. pushed me to think critically and creatively in +ways I hadn't experienced in the classroom. +I'm confident that the problem-solving, +communication, and client-facing skills I have +developed will serve me well when pursuing +my future career in consulting or another +business-oriented field. +Kaur, Suchleen BBA(Acc&Fin) +Business Consulting Practicum 2024 +The opportunity for real-time interaction and feedback with the +client had propelled me to join BCP. I was always curious about +how a consultant would address and communicate with the client. +BCP has taught me about proper business etiquette and equipped +me the necessary skills to negotiate conflicting opinions during +team discussions. Throughout the process, our mentor and +professors provided great help and insights. Their invaluable +feedback and questions helped nudge us in the right direction, +taking a step back to look at the big picture and most importantly, +prioritise the client’s needs. +22 23 + +Ou, Tanyalak BBA +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2024 +The combination of thought-provoking lectures, insightful guest +speaker sessions, and dynamic debate challenges created a richly +immersive learning environment. I was constantly challenged to +think critically, conduct thorough research, and articulate my ideas +and arguments effectively. It was an experience that equipped me +with invaluable insights and skills for navigating the complexities of +China's dynamic business landscape. +Yeung, Yui Pong BBA(Law)&LLB +Doing Business in China Programme 2024 +Participating in the DBIC Programme at PKU +Guanghua School of Management was an +extraordinary and transformative experience. The +lectures were particularly inspiring, covering a wide +range of topics that are crucial for understanding +the Chinese business landscape. From discussions +on demographics to exploring real estate finance +and the digital economy, and delving into the realm +of digital marketing in China, each session offered +Nan, Shan BEcon&Fin +valuable unique perspectives. Moreover, I was +"Creativity, Innovation & Entrepreneurship in China" +particularly impressed with the office visit to Swire +Summer Virtual Programme 2024 +and Kuaishou, which was a fantastic experience +Joining the CIEC programme ignited my entrepreneurial spirit and +understanding how the two companies navigate +enabled me to accumulate knowledge in doing business +the Chinese market and achieve rapid growth. +and making decisions in life. We learned about China’s economic growth +and cultural richness and grasped the role of innovation in boosting the +economy. “Social networking is critical in the business world”, CIEC +establishes a great networking platform for students, where +like-minded peers’ ideas collide, allowing us to explore the world +through a completely new lens. +24 25 + +Case Competitions +Darmawan, Jocelyn Victoria BSc(QFin) +The School provides ample opportunities to refine students’ presentation Bloomberg Global Trading Challenge 2023 +and analytical skills through a myriad of local as well as overseas (Grand Prize Winner) +competitions. From training to delicate analysis of business cases to a It was such an amazing experience to win the +variety of business scenarios, students are equipped with different Challenge, and I would like to give a shout out to my +skillsets to compete with candidates from the world's leading universities. teammates who have dedicated a lot of their time and +energy into researching and strategizing our portfolio +allocation. I learned a lot about how the stock market +reacts to different macroeconomic events, and how +to capitalize on opportunities to achieve a high +portfolio return. I would also like to extend my thanks +to Bloomberg, HKU Business School, and the faculty +advisor for their unwavering support. +Lau, Chun Ho Darren BBA(Acc&Fin) +EY Young Tax Professional of the Year 2023 +(1st Runner-up) +The EY Young Tax Professional of the Year 2023 was a +defining moment in our university lives, as it allowed +us to apply critical thinking to complex, cross-border +tax issues and showcase our insights before esteemed +CHAN, Chun Yan Natalie BBA(Law)&LLB +professionals. This rewarding journey not only +Lazaridis International Case Conference 2023 +broadened our understanding of the international tax +(2nd Place) +landscape but also granted us a platform to showcase +Participating in the LazICC 2023 and securing 2nd our abilities and potential. In essence, we eagerly +Place was a transformative experience that left an anticipate building upon this foundation to evolve into +indelible mark on my personal and professional world-class, game-changing professionals. +growth. Throughout the competition, I had the +privilege of working closely with a dedicated team. +The complex business cases enabled our team +to sharpen our analytical skills, while reinforcing +the significance of time management and +communication abilities. The international aspect of +the competition provided us with the unique +opportunity to network with individuals from around +the world. I am immensely thankful for our coach and +faculty members who supported us along the way. +26 27 + +Luo, Regina BBA(IBGM) +HKICPA Business Case Competition 2023 (2nd Runner-up) +Taking part in this competition was an extraordinary +experience that allowed me to expand my business +knowledge and fostered my growth in several soft skills. It +was a fresh new experience as we learnt to think outside the +box and make decisions based on different platforms by +putting ourselves in the shoes of the gaming company. +Communication and presentation skills were another key +takeaway from the competition for me. It was truly a +wonderful and invaluable experience. +Students' achievements in case competitions +(2022-2024 awards) +Overseas / Mainland +Amplify Trading Global Markets Competition 2022 Finalist +Asian Business Case Competition 2022 1st Runner-up +Bloomberg Global Trading Challenge 2023 Grand Prize Winner +Regional Prize Winner: Asia & Oceania +Yao, Chuanjie Rosie BBA(BA) +CBS Case Competition 2023 Winner +HKGCC Business Case Competition 2023/24 +(Champion - Swire Pacific track and CGMA Global Business Challenge 2023 - North Asia Final Champion +1st Runner-up - Chinachem Group track) Champions' Trophy Case Competition 2024 Finalist +China Mainland, Hong Kong and Macao Digital Economy Innovation and 2nd Runner-up (Gold Medal) +Participating in the HKGCC Case Competition was +Entrepreneurship Competition 2024 +an incredible experience that has left an indelible +Climathon2022@Hangzhou Championship +mark on my personal and professional growth. +Deloitte Digital Camp 2023 Champion +Through our innovative initiatives for the case +Deloitte Digital Camp 2024 Second Place +companies, we aimed to make a positive impact +Deloitte Tax Championship 2023 - National Round Merit Award +on society and the environment. I deeply +appreciate my teammates’ collective effort, Deloitte Tax Championship 2023 - Regional Round Top 3 teams +collaboration, and synergy which were the driving EY Young Tax Professional of the Year 2022 - National round Champion +forces behind our success. I extend my deepest Family Office Association Hong Kong (FOAHK) 2023 Case Competition 2nd Runner-up +gratitude to our supervisor, whose guidance and Global Trading Challenge 2022 World Top 10% +support were invaluable throughout the Häagen Dazs Challenge 2022 First Prize +competition. This experience has left me inspired HSBC Global Private Banking Case Challenge 2023 1st Runner-up +and even more determined to contribute to the +HSBC Global Private Banking Case Challenge 2022 1st Runner-up +business world and create meaningful change. +Investing in the Future University Competition 2023 by Pictet Asset 2nd Runner-up +Management +Lazaridis International Case Conference 2023 1st Runner-up +Marshall International Case Competition 2024 Finalist +Scotiabank International Case Competition 2022 Winner +Scotiabank International Case Competition 2023 Winner +Shaping the Future Nansha Innovation Challenge 2022 1st Runner-up +Most Creative Award +UW Global Business Case Competition 2023 Finalist +Youth-Storm Microsoft Business Analytics Challenge 2022 National Champion +28 29 + +| Local | | Local | | +| ----- | --- | ----- | --- | +100WF Hong Kong JumpStart University Investment Hackathon 2023 1st Runner-up HKU-Airstar Bank Joint University Case Competition 2024 Champion +| ACCA Business Competition 2022-23 | Champion | | 2nd Runner-up | +| --------------------------------- | ------------------------- | ------------------------------------------------ | -------------- | +| | Creative Excellence Award | HKU-SaSa Joint University Case Competition 2023 | 1st Runner-up | +| | Digital Excellence Award | | Best Presenter | +| | Best Budgeting Award | Hong Kong Joint University Case Competition 2022 | Champion | +ACCA Hong Kong Business Competition 2023 Outstanding Team for start-up Filix Medtech 2nd Runner-up +| | Team Excellence Award | HSBC Life Insurance Innovation Competition 2023 | 1st Runner-up | +| --- | ------------------------- | -------------------------------------------------------- | ---------------------------- | +| | Merit Award for start-up | | Best Use of Technology Award | +| | MakeOmnia | International Macro Business Simulation Competition 2022 | Champion | +Advancing Greener Tomorrow Competition 2022 Champion International MBS Business Management Simulation Competition 2023 Champion +BOCHK Challenge 2022 Gold Award for Innovative Design 1st Runner-up +BOCHK Challenge 2023 Bronze Award for Innovative Design ISM-HK Student Case Competition 2024 First place +CFA Institute Research Challenge 2022-23 Hong Kong Local Final 2nd Runner-up Third Place +CGMA Global Business Challenge 2023 - Semi-final Round (Hong Kong) Champion Best presenter +Finalist J.P. Morgan Private Bank Challenge 2022 Hong Kong Champion +Deloitte USHINE Case Competition 2024 Champion KPMG Ideation Challenge Semi-Final (Hong Kong Regional Final) 2022 Winner +ESG Challenge 2023 Champion LEAD to the Future - Business Case Competition 2022 1st Runner-up +EY Young Tax Professional of the Year 2022 - Hong Kong Region Champion 2nd Runner-up +| | 1st Runner-up | | 3rd Runner-up | +| --- | ------------- | --- | ------------- | +Finalist L'Oréal Brandstorm - Brandstorm 2022 (Hong Kong Final) Tech Track Winner +EY Young Tax Professional of the Year 2023 - Hong Kong Region 1st Runner-up MBS Simulated Business Management Competition 2024 Champion +| | 2nd Runner-up | | 1st Runner-up | +| --- | ------------- | --- | ------------- | +Female Entrepreneur Business Simulation Competition 2022 2nd Runner-up New Century Elite Development Programme 2023 2nd Runner-up +FSDC Career Day 2022 - Pitch Perfect Video Competition Champion Polymer Capital Management Inter-varsity ESG Case Challenge 2022 Champion +HKGCC Business Case Competition 2023/24 Chevalier - Champion Quest for Securities & Investment Elites (QSIE) Case Analysis Competition 2022 1st Runner-up +| | Chevalier - 1st Runner-up | | Outstanding Award | +| --- | ------------------------- | --- | ----------------- | +Chinachem Group - Champion The Hong Kong Future Leader Awards 2022 Best Performing Team Award +| | Swire Pacific - Champion | Venture Capital On Campus (VCOC) - Investment Challenge 2023 | Winner Team | +| --- | ------------------------ | ------------------------------------------------------------ | ----------- | +Chinachem - 1st Runner-up +| HKGCC Pitch Perfect Programme 2024 | Overall Champion | | | +| ---------------------------------- | ---------------- | --- | --- | +Winner +| HKICPA Business Case Competition 2023 | 2nd Runner-up | | | +| ------------------------------------- | ------------- | --- | --- | +| HKSI Institute Case Competition 2022 | Finalist | | | +Top Performer Award +Best ESG Initiative Award +| HKSI Institute Case Competition 2023 | Champion | | | +| ------------------------------------ | -------- | --- | --- | +1st Runner-up +Merit & Finalist +Semi-finalist +| HKSI Institute Case Competition 2024 | Champion | | | +| ------------------------------------ | -------- | --- | --- | +1st Runner-up +Best performer of Roma Risk +Advisory Limited +Merit +30 31 + +Internship Experience +Kim, Chaeyeon B F in(AMPB) +J.P. Morgan - Equity Research Summer Analyst +This internship allowed me to immerse myself in the exhilarating world of financial +Cheung, Lok Wan BBA analysis. I contributed to drafting research reports and also worked on an individual +UBS AG - Global Wealth Management project. The internship was not just about the tasks at hand; it was a journey of +Summer Analyst personal and professional growth. I had the privilege of meeting various exceptionally +Working at an international firm as a talented individuals within the firm. Through engaging with these professionals, I +junior leverages my personal growth and gained valuable insights into the financial markets. This experience was not only a +builds a strong team network. Finance is testament to my passion for finance but also a stepping stone towards achieving +not just about numbers; we are doing a excellence in my future career. +huge people business to deliver +professional services and connect the +world to a better place. +Lee, Dohoon B E c on&Fin +HKU – Neuberger Berman +Ho, Yau Wai Athena B BA(IS) +Private Equity Internship +Bloomberg - Analytics and Sales Summer Intern +Programme 2023-24 +I am grateful to have the opportunity to work with +I believe what sets Neuberger Berman HK office Lee, Wing Ting Stephanie B B A(Acc&Fin) +many talented and incredible managers, team +apart as an employer especially to the interns, is the HKU – Neuberger Berman Private Equity +leaders, mentors, coordinators, project leads and +culture that genuinely values the juniors and nurtures Internship Programme 2023-24 +colleagues who provided professional advice, +Chung, Lok Yin B Fin(AMPB) guidance and support for my personal growth and their development. And I believe such culture starts During my enriching internship at Neuberger +Raffles Family Office - Summer Intern development, especially in my presentation skills, all the way from the top, Mr. Kent Chen, down to the Berman, I immersed myself in a dynamic deal +communication skills and market knowledge. entire team. As interns, we are given full exposure to environment within the Southeast Asia team, +During my time at Raffles Family Office, I developed whichever transaction we are staffed on including engaging in primary and co-investment deals that +a spreadsheet that automatically retrieves pertinent I am also glad to collaborate with my brilliant intern the email chains, data rooms, and the shared drive. broadened my deal type exposure. Working on +data from Bloomberg when an equity ticker is batchmates on various projects about terminal In fact, thanks to Kent, I was also fortunate to attend deals in diverse regions like India, Australia, and +entered, facilitating relationship managers in functions and market news. a forum in January where we got to listen in on Japan significantly expanded my market +producing client meeting materials. Despite my keynote speeches and panel discussions, and understanding. A pivotal aspect of my internship +initial lack of familiarity with the Bloomberg network with private equity and venture capital was gaining comprehensive insights into the +Terminal and VBA, I was encouraged by my professionals that flew in from across Asia. This was private equity industry and the Southeast Asia +manager (mentor) to visit the Bloomberg office to one of those experiences that really motivated market. By actively contributing to deal processes, +deepen my understanding. I even employed GPT to me once again to pursue a career in this field. drafting investment memos, and interacting with +assist in generating VBA codes. This experience +senior professionals, I acquired valuable +underscored the importance of embracing new +knowledge of prevailing investment trends across +technologies and tools, highlighting that consistent +multiple markets. +effort and exploration can lead to valuable Luo, Wenhao B E c on&Fin +outcomes in this tech-driven era. +HKU - MindWorks Venture Capital Internship +Hui, Matthew B B A(Acc&Fin) Programme 2023-24 +Ares Management Corporation - Intern +My experience with the HKU - MindWorks Venture Capital Internship Programme was +(Investment Operation) +truly enriching. As an investment intern, I had the opportunity to work closely with +My journey at Ares has significantly enhanced my understanding of the alternative investment industry, especially industry professionals and gained hands-on experience in the venture capital field. I +in the private credit sector; I have the opportunity to assist in the operations of multiple funds, monitoring and evaluated potential investment opportunities, conducted market research, and assisted +repatriating fund cash flow, managing FX and forwards / options hedging for risk management. I was also with due diligence. The internship allowed me to apply my academic knowledge in a +honoured to participate in structuring and launching a billion-sized private credit fund, completing due diligence real-world setting, fostering significant personal and professional growth. I am grateful +and KYC checks for external parties and communicating with LPs and external stakeholders. I'm truly thankful to for the invaluable connections, industry insights, and practical skills I acquired through +work with wonderful colleagues and for their mentorship and guidance during my period in Ares. this experience, which have better prepared me for my future career aspirations. +32 33 + +Mo, Cheuk Lam B E c on&Fin +HSBC - Wealth and Personal Banking Intern +During my internship, I immersed myself in the mortgage industry, +| enhancing | my understanding | of key facets. | Highlights include | | | | | | +| ------------ | ------------------ | -------------------- | -------------------- | --- | --- | --- | --- | --- | +| involvement | in developing | equitable mortgage | products, gaining | | | | | | +insights into risk assessment, understanding compliance standards' +| impact, and | collaborating | across departments | for streamlined | | | | | | +| ------------- | -------------- | -------------------- | ----------------- | --- | --- | --- | --- | --- | +mortgage operations. This experience has equipped me to tackle future challenges in the mortgage sector, +providing a solid foundation for my professional growth, and I look forward to leveraging this knowledge in Wong, Chloe Miriam B B A (Law)&LLB +upcoming endeavors. Recipient of HKEX Foundation Scholarship 2023-24 +I am very grateful to the HKEX Foundation for this Scholarship. +Thanks to this generous award, I have been able to pursue valuable +learning opportunities that both align with my existing passions and +| | Wong, Ho Yeung B B A (Acc&Fin) | | | | | broaden my horizons. | | | +| --- | ---------------------------------------- | --- | --- | --- | --- | -------------------- | --- | --- | +Bloomberg - Analytics and Sales Summer Intern +My experience as an analytics and sales summer intern at Bloomberg was incredibly +fulfilling. I was involved in a range of projects across different asset classes and had +the opportunity to interact with clients from various sectors, including International Su, Zihan B E c o n&Fin +Banks on the Sell side, Private Equity firms, and Hedge Funds on the Buy Side. I am Recipient of Hong Kong Jockey Club +grateful for the support I received from HKU Business School, both in terms of Scholarship 2022-23 to 2023-24 +academic support and extracurricular activities prior to the internship. The HKJC Scholarship not only recognizes my +past achievements but also encourages me to +move forward and explore more. I am incredibly +grateful for the Jockey Club's generosity and will +Our Student Scholars continue to pass this love on to the broader +community and society. +Chiu, Yi Nap B S c ( Q Fin) +Recipient of Shanghai Commercial Bank +Ho, Hiu Lam B B A(IBGM) +Scholarship 2023-24 +Recipient of HKU Business School - BOC Life +| | | | | | | The Shanghai | Commercial Bank | Scholarship | +| --- | --- | --- | --- | --- | --- | -------------- | ----------------- | ------------ | +Future Leader Scholarship 2023-24 +certainly made my university life more vibrant. +| | | | The scholarship | provided immense | support | | | | +| --- | --- | --- | ----------------- | ------------------ | -------- | --- | --- | --- | +Without it, I could not have pursued my dream as +towards my overseas exchange experience and +a dancer, continued to stay in Lee Shau Kee Hall +has undoubtedly empowered me to develop a +while having the chance to explore the world! +| | | | global mindset | and gain new | perspectives | | | | +| --- | --- | --- | ---------------- | --------------- | ------------- | --- | --- | --- | +through immersing myself in a distinct cultural +environment. I am grateful that the scholarship +Gandhi, Devansh B B A +| | | | has been | effective in | | | | | +| --- | --- | --- | ---------- | -------------- | --- | --- | --- | --- | +Recipient of Faculty Student Contribution and +| | | | accelerating | my growth | | | | | +| --- | --- | --- | ------------- | ----------- | --- | --- | --- | --- | +Innovation Award 2023-24 +| | | | personally, | academically | | | | | +| --- | --- | --- | ------------ | ------------- | --- | --- | --- | --- | +Receiving the Award is a profound honour, marking +and professionally. +a significant milestone in my journey. It motivates +me to strive further to build diverse communities +of innovation. This award acknowledges my past +| efforts and | fuels my dedication | to continue | | | | | | | +| ------------- | ---------------------- | --------------- | --- | --- | --- | --- | --- | --- | +| making a | meaningful impact | in the world | of | | | | | | +entrepreneurship. +34 35 + +For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) +2025 +| | | | In addition | to the minimum | entrance requirements | above, candidates | applying | for the BBA(BA), | +| --- | --- | --- | ------------- | ----------------- | ----------------------- | ------------------- | --------- | ------------------- | +JUPAS Admission BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the +HKDSE examination: +Admissions criteria BBA(BA) / BSc(MAT) - Level 3 or above in one of the following subjects: +(HKDSE) +| | | | | | Combined | | Integrated | | +| --- | --- | --- | --- | ------- | -------------- | --------- | ---------- | ------- | +| | | | | Biology | BAFS Chemistry | Economics | ICT | Physics | +| | | | | | Science | | Science | | +Local candidates applying on the strength of the +Hong Kong Diploma of Secondary Education (HKDSE) +| are required | to fulfil | the minimum entrance | BBA(BA) | | | | | | +| -------------- | ----------- | ----------------------- | ------- | --- | --- | --- | --- | --- | +requirements for HKU Business School programmes, +| as shown below: | | | BSc(MAT) | | | | | | +| --------------- | --- | --- | -------- | --- | --- | --- | --- | --- | +Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not +strictly required. +Level 4 or above in English Language; +Level 3 or above in Chinese Language1; +| | | | BBA(Law)&LLB | Level 5 or above in | | Level 4 or above in | | | +| --- | --- | --- | -------------- | -------------------- | --- | -------------------- | --- | --- | +Level 3 or above in Mathematics; +| | | | | English Language | | Chinese Language1 & Mathematics | | | +| --- | --- | --- | --- | ---------------- | --- | ------------------------------- | --- | --- | +Grade “Attained (A)” in Citizenship and +Social Development2; and +Level 3 or above in two Elective Subjects. +| | | | BBA(IBGM) | Level 5 or above in | | | Level 4 or above in | | +| --- | --- | --- | ---------- | --------------------------------------------------------- | --- | --- | -------------------- | --- | +| | | | | English Language | | | Mathematics | | +| | | | BSc(QFin) | Level 3 or above in Extended Module 1 or 2 of Mathematics | | | | | +Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis. +Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered. +Application +For details of the 2025 JUPAS exercise, please visit www.jupas.edu.hk. +36 37 + +2025 Non-JUPAS Admission University and Programme Entrance Requirements +Due to the complex combination of education background of applicants, each application will +be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the +Selection process and entrance requirements standards and requirements on the website of the Admissions Office of the University before +making an application: +Eligibility +(1) Admissions standards (International Qualifications or National Qualifications); +(1) Local students applying for admission on qualifications other than the HKDSE examination; +(2) Programme specific subject requirements; +(2) Non-Local students with qualifications other than the mainland China’s National Joint +(3) English language requirement; and +College Entrance Examination (NJCEE/Gao Kao) +(4) Second language requirement (i.e. any language other than English). +Application Procedures +In particular, please note that the expected lower boundary for admissions listed on the +All non-JUPAS applications should be submitted via the online application system by noon of +website is provided for reference purposes only. Admissions is competitive in nature and +November 27, 2024 (Hong Kong Time) for the first-round evaluation. Scanned copies of the +achieving the expected lower boundary does not guarantee admission to the programme. +following documents should also be uploaded by the application deadline. +The actual admission grade or score will be dependent on the overall quality of applications +(1) Official prediction of grades or scores from your current high school (if available); received in the 2025 admissions exercise. +(2) All internal and public examination results during your senior secondary education; and +Announcement of Result +(3) At least one academic reference letter Offer results will be released from December 2024 onwards. All application status will be +updated in the online application system once an admissions decision has been made. +Late submission of supporting documents will delay the consideration of application. +Applicants should check their application status online regularly. +Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other +programmes will be arranged if necessary. More Details +For further information about the application procedures and +admissions standards, please visit the Admissions Office’s website at +https://admissions.hku.hk/apply/international-qualifications. +2025 Dual Degree Admission +Separate application is required for applying to the dual degree programmes, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. For details of the +dual degrees and the application procedures, please visit the Admissions Office’s website at +https://admissions.hku.hk/programmes/university-collaborative-programmes. +38 39 + +| Develop Your Career | | | | | | Freshmen Sharing | | | | | | +| ------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +Upon graduation, over 88% of our students begin their careers in the commerce and industry sector. Others either +| move on to further | studies or explore | opportunities | with the | civil service, education | institutions and | | | | | | | +| ---------------------- | --------------------- | -------------- | ---------- | --------------------------- | ------------------ | --- | --- | --- | --- | --- | --- | +community/social services. I chose HKU Business School for its prestigious +reputation and focus on experiential learning. I +Career Prospects believe HKUBS is the best place to help me realize my +| | | | | | | dream. BBA(Acc&Fin) | perfectly aligns | with my | With the understanding that the economy serves as | | | +| --- | --- | --- | --- | --- | --- | --------------------- | ------------------ | --------- | -------------------------------------------------- | --- | --- | +Companies/organisations that have offered full-time positions to our graduates include but not limited to the passion for numbers and financial analysis. I look a fundamental and crucial part of all other subjects, +| | | | | | | forward to engaging | in hands-on | learning | and recognizing | Hong Kong | as a central and | +| --- | --- | --- | --- | --- | --- | ---------------------- | ------------- | --------- | ----------------- | ----------- | -------------------- | +following: +| | | | | | | experiences and collaborating with diverse peers. I | | | international | commercial hub, | I chose to study | +| ------------------------- | --- | --- | ---------------------- | --- | --- | --------------------------------------------------------- | --- | --- | ------------------------------------------------ | ----------------- | -------------------- | +| Accenture Company Limited | | | Hang Seng Bank Limited | | | | | | | | | +| | | | | | | anticipate a vibrant university life filled with diverse | | | BEcon&Fin at HKU. However, after my first month | | | +Bank of China (Hong Kong) Limited HKSAR Government experiences and networking opportunities. The here, I discovered that what truly strengthens my +| | | | | | | dynamic atmosphere at HKU inspires me, and I am | | | decision is the university’s ability to cultivate my | | | +| --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | ----------------------------------------------------- | --- | --- | +BDO International Limited Hong Kong Monetary Authority excited to explore opportunities that will enhance my autonomy, give me more space for self-learning, and +| | | | | | | academic journey and prepare me for a successful | | | provide numerous opportunities to meet diverse | | | +| -------------- | --- | --- | -------------------- | --- | --- | ----------------------------------------------------- | --- | --- | --------------------------------------------------- | --- | --- | +| Bloomberg L.P. | | | JPMorgan Chase & Co. | | | | | | | | | +| | | | | | | career. HKU's commitment to innovation truly inspire | | | individuals and those who share the same dreams as | | | +Citibank (Hong Kong) Limited KPMG me. I’m excited to explore new ideas and contribute mine. According to my MBTI test, I have “I” +| | | | | | | to both the academic and social communities. | | | personality, but here I have found many ways to | | | +| --------------------------------------------- | --- | --- | -------------- | --- | --- | -------------------------------------------- | --- | --- | ------------------------------------------------ | --- | --- | +| Crédit Agricole Corporate and Investment Bank | | | Morgan Stanley | | | | | | | | | +express myself. I feel like I am challenging and +improving myself every day! +| Credit Suisse Group AG | | | PricewaterhouseCoopers Limited | | | | | | | | | +| ------------------------------------- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| Deloitte Advisory (Hong Kong) Limited | | | Standard Chartered Bank (Hong Kong) Limited | | | | | | | | | +| Deloitte Touche Tohmatsu Limited | | | The Hongkong and Shanghai Banking Corporation | | | | | | | | | +Yang, Huimiao Murphy +Limited (HSBC) +Deutsche Bank +BBA(Acc&Fin) +| | | | UBS Group AG | | | | | | | Chen, Xinqiao Vicky | | +| --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | ------------------- | --- | +Ernst & Young +BEcon&Fin +Goldman Sachs (Asia) LLC +Full +| employment | | 57 | | | | | | | | | | +| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +% +98.9% +18th +of our 2023 graduates have received +at least 2 job offers upon graduation +| for the consecutive year | | | | | | | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +37,166 +$ +was the average salary of +our 2023 graduates +40 41 + +Seduced by HKU's scholarly prowess and dynamic +I chose to study IBGM at HKU Business School because of its strong +campus spirit, my path led to HKU Business School, +reputation as an innovative and established institution. Moreover, the +renowned for its cutting-edge business education. +| | | | As an Economics and Finance major, I’ve embarked | | | programme’s | emphasis on | developing well-rounded | future leaders | +| --- | --- | --- | ----------------------------------------------------- | --- | --- | --------------------------------------------------------------------------------- | ------------- | --------------------------- | ---------------- | +| | | | on my university journey brimming with intellectual | | | through global | exposure | and strategic management | training was | +| | | | rigor, leadership prospects, and a tapestry of peers | | | extremely attractive to me. I expect to have a lot of fun in university, gaining | | | | +valuable experience and developing in all areas of my life. So far, my +broadening my worldview. Every day, I’m thrilled to +The main reason I have chosen the IBGM programme expectations are being fulfilled, as HKU is truly an incredibly fun and +push my limits in academic debates and research +at HKU Business School is due to the extraordinary +| | | | projects, immerse myself in Hong Kong's unique | | | diverse place filled with unique opportunities. | | | | +| ------------------------ | ------------ | ---------- | ----------------------------------------------- | --- | --- | ----------------------------------------------- | --- | --- | --- | +| exchange opportunities | it offers. | With the | | | | | | | | +blend of Eastern and Western cultures, and build an +opportunities of studying abroad in North America +extensive network by engaging with peers and +and Europe, and visiting companies in Asia, I believe +faculty from diverse backgrounds. In addition, HKU +my global perspective will be cultivated by gaining +| | | | offers me | a platform to | give back through | | | | | +| ------------- | -------------------- | ------------- | ----------- | ---------------- | -------------------- | --- | --- | --- | --- | +| exposure to | some of the most | economically | | | | | | | | +co-curricular engagement, readying myself for a +developed regions. Additionally, I appreciate the Wong, James Dylan Cheng Yan +robust finance career. Look forward to joining the +| flexibility provided | by the IBGM | programme in | | | | | | | | +| ---------------------- | -------------- | -------------- | ----------------------------------------------- | --- | --- | --- | --- | --------- | --- | +| | | | HKU Business School community and experiencing | | | | | BBA(IBGM) | | +choosing interdisciplinary courses for a second major +an unforgettable university life! +or minor, allowing me to expand my interest in +Psychology. Overall, I believe the experiences and +alumni connections I will gain from the HKU Business +School will be extremely useful in preparing me for +my future career. +| | | | | Qu, Hongru Larina | | When it comes to HKU, the first thing that springs to | | | | +| --- | --- | --- | --- | ----------------- | --- | ------------------------------------------------------ | --- | --- | --- | +most people's minds might be its extremely high +BEcon&Fin +| | | | | | | academic | standing. For me, | it is not only the | | +| --- | --- | --- | --- | --- | --- | --------- | -------------------- | ----------------------- | --- | +academic excellence but also the unique campus +atmosphere that attracted me to become a member +Ng, Shui Yee Sophie +| | | | | | | of HKU. | HKU is indeed | an international and | | +| --- | --- | --- | --- | --- | --- | --------- | ---------------- | ----------------------- | --- | +BBA(IBGM) all-inclusive university, where I can experience great +diversity and meet elites from all around the world. I +am really eager to meet and collaborate with them. I +believe such a rich environment and atmosphere will +be highly conducive to my personal development +| | | I chose HKU because I was truly fascinated by the opportunity to become a | | | | and the expansion of my interpersonal networks. | | | | +| --- | --- | -------------------------------------------------------------------------- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | +student at one of Asia’s leading universities, known for its top-notch +educational standards. I am excited to study BEcon&Fin in the heart of a +major financial center and look forward to exploring the various internship +opportunities that HKU Business School offers. All my professors are very +enthusiastic and sociable, which motivates me to work hard. HKU exceeds +| | | my expectations, and I am glad to be here! | | | | | | Lui, Ka Wai Susie | | +| --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | ----------------- | --- | +BEcon&Fin +Kotlova, Tatiana Tanya +BEcon&Fin +42 43 + +Our Global Presence +Beijing +Israel - +Tel Aviv +Shenzhen +Hong Kong Useful Websites +The University of Hong Kong (HKU) HKU Business School +Vietnam - Undergraduate Admissions Overview of Undergraduate Programmes +Ho Chi Minh City https://admissions.hku.hk/ https://ug.hkubs.hku.hk/programme +Enrichment Programmes +HKU – Entrance Requirements +https://ug.hkubs.hku.hk/student-enrichment +For JUPAS Candidates +https://admissions.hku.hk/apply/jupas Other HKU Websites +For non-JUPAS Candidates Mainland Affairs Office (MAO) +https://admissions.hku.hk/apply/international-qualifications https://mainlandaffairs.hku.hk/ +Centre of Development and Resources for +Students (CEDARS) +https://www.cedars.hku.hk/ +44 45 + +HKU Business School +4/F, K.K. Leung Building +The University of Hong Kong +Pokfulam Road, Hong Kong SAR +(852) 3917 5343 +(852) 2549 3735 +fbe.ugenquiry@hku.hk +www.hkubs.hku.hk +HKU Business School News and Events +HKU Business School +HKU Business School +香港大学经管学院 +香港大学经管学院 +HKU Business School Undergraduate Admissions News and Events +@hkubsug +HKU Business School - UG Admissions +The information in this brochure is accurate as of October 2024 and is subject to change. \ No newline at end of file diff --git a/q166/random_k4/question.json b/q166/random_k4/question.json index a94f51e075d7db7b3f385b3b5b4badf9d7c4ffe7..a3fa4033a3b3cf8b060be4e14934d070585cb61c 100644 --- a/q166/random_k4/question.json +++ b/q166/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "stereo_headset.pdf", - "web_c53505eb5c4f8264.pdf", - "web_fe44495665edf65a.pdf", - "web_6c246a9c943f3cd0.pdf", - "web_86f9823f6b6edb2b.pdf" + "5fdcf120b846746985684fa54250aab44639b30c.pdf", + "web_2dad2d6279d3265e.pdf", + "web_a43524926d478681.pdf", + "web_66d0bc5ff83e607b.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q166/random_k4/random_1.md b/q166/random_k4/random_1.md index c3a1aac8ea3031fefb7fe76173d9b3ff3594440a..2e9f5241e58113455522088c7a6110de8838d130 100644 --- a/q166/random_k4/random_1.md +++ b/q166/random_k4/random_1.md @@ -1,2180 +1,1208 @@ -HALF YEAR REPORT -JANUARY – JUNE - -HALF YEAR REPORT 2021 -HALF YEAR -REPORT -TABLE OF CONTENTS -FINANCIAL HIGHLIGHTS (IFRS) ______________________________________________________________________________________________________________________ 1 -BUSINESS PERFORMANCE __________________________________________________________________________________________________________________________ 2 -ECONOMIC AND SECTOR DEVELOPMENT ____________________________________________________________________________________________________________________ 2 -INCOME STATEMENT _______________________________________________________________________________________________________________________________________ 2 -STATEMENT OF FINANCIAL POSITION AND STATEMENT OF CASH FLOWS ____________________________________________________________________________________ 4 -BUSINESS PERFORMANCE BY SEGMENT _____________________________________________________________________________________________________________ 6 -EMEA ______________________________________________________________________________________________________________________________________________________ 6 -NORTH AMERICA ___________________________________________________________________________________________________________________________________________ 7 -GREATER CHINA ___________________________________________________________________________________________________________________________________________ 7 -ASIA-PACIFIC ______________________________________________________________________________________________________________________________________________ 8 -LATIN AMERICA ____________________________________________________________________________________________________________________________________________ 8 -OUTLOOK ___________________________________________________________________________________________________________________________________________ 9 -GLOBAL ECONOMIC GROWTH TO RECOVER IN 2021 __________________________________________________________________________________________________________ 9 -SPORTING GOODS INDUSTRY TO RECOVER NOTABLY IN 2021 _________________________________________________________________________________________________ 9 -ADIDAS INCREASES OUTLOOK FOR FULL YEAR 2021 __________________________________________________________________________________________________________ 9 -RISKS AND OPPORTUNITIES _______________________________________________________________________________________________________________________________ 10 -CONSOLIDATED STATEMENT OF FINANCIAL POSITION_______________________________________________________________________________________________ 11 -CONDENSED CONSOLIDATED INCOME STATEMENT _________________________________________________________________________________________________ 13 -CONSOLIDATED STATEMENT OF COMPREHENSIVE INCOME _________________________________________________________________________________________ 14 -CONSOLIDATED STATEMENT OF CHANGES IN EQUITY ______________________________________________________________________________________________ 15 -CONSOLIDATED STATEMENT OF CASH FLOWS ______________________________________________________________________________________________________ 16 -EXPLANATORY NOTES TO THE CONDENSED INTERIM CONSOLIDATED FINANCIAL STATEMENTS (IFRS) AS AT JUNE 30, 2021 _________________________ 18 -RESPONSIBILITY STATEMENT ______________________________________________________________________________________________________________________ 32 -To enhance readability, registered trademarks as well as references to rounding differences are omitted in this publication. - - HALF YEAR REPORT 2021 - -| | | 1 | | | | 2 | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -FINANCIAL HIGHLIGHTS (IFRS) -| | | | | | | | First half year | First half year | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | ---------------- | ------------------ | --- | -| | | | | | | | | 2021 | 2020 Change | | -| | | | | | | | | | | | -Operating Highlights (€ in millions) -| Net sales | | | | | | | 10,345 | | 7,733 | 34% | -| ------------------------- | --- | --- | --- | --- | --- | --- | -------- | ------- | ------- | ---- | -| Gross profit | | | | | | | | 5,362 | 3,898 | 38% | -| | | | | | | | | | | | -| Other operating expenses | | | | | | | | 4,154 | 4,156 | 0% | -| | | | | | | | | | | | -| EBITDA | | | | | | | | 1,804 | 425 | 324% | -| | | | | | | | | | | | -| Operating profit / (loss) | | | | | | | | 1,248 | (215) | n.a. | -| | | | | | | | | | | | -Net income / (loss) from continuing operations 890 (217) n.a. -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable to shareholders 955 (264) n.a. -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -| | | | | | | | | | | | -Key Ratios -| Gross margin | | | | | | | 51.8% | 50.4% | | 1.4pp | -| ------------ | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | ----- | -Other operating expenses in % of net sales 40.2% 53.8% (13.6pp) -| | | | | | | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | ------- | -------- | -------- | ----- | -| Operating margin | | | | | | | 12.1% | (2.8%) | 14.8pp | | -| | | | | | | | | | | | -| Effective tax rate | | | | | | | 24.5% | 22.4% | | 2.1pp | -| | | | | | | | | | | | -Net income attributable to shareholders in % of net sales 9.2% (3.4%) 12.6pp -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Average operating working capital in % of net sales1 21.4% 23.2% (1.8pp) -| | | | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | --- | --- | ------- | -------- | ---------- | ----- | -| Equity ratio2 | | | | | | | 32.5% | 30.7% | | 1.8pp | -| | | | | | | | | | | | -| Adjusted net borrowings3/EBITDA4 | | | | | | | | 0.9 | 2.3 | (1.4) | -| | | | | | | | | | | | -| Financial leverage | | | | | | | 44.3% | 80.1% | (35.8pp) | | -| | | | | | | | | | | | -| Return on equity2 | | | | | | | 13.4% | (4.2%) | 17.7pp | | -| | | | | | | | | | | | -| | | | | | | | | | | | -Balance Sheet and C ash Flow Data (€ in millions) -| Total assets | | | | | | | 21,877 | 20,301 | | 8% | -| ------------------------------------ | --- | --- | --- | --- | --- | --- | -------- | -------- | ------- | ----- | -| Inventories | | | | | | | | 4,054 | 5,213 | (22%) | -| | | | | | | | | | | | -| Receivables and other current assets | | | | | | | | 4,257 | 3,694 | 15% | -| | | | | | | | | | | | -| Operating working capital | | | | | | | | 4,213 | 4,506 | (7%) | -| | | | | | | | | | | | -| Shareholders’ equity | | | | | | | | 7,106 | 6,230 | 14% | -| | | | | | | | | | | | -| Capital expenditure | | | | | | | | 183 | 217 | (16%) | -| | | | | | | | | | | | -Net cash generated from / (used in) operating activities 1,077 (819) n.a. -| | | | | | | | | | | | -| ------------------------------ | --- | --- | --- | --- | --- | --- | --- | ------ | -------- | ---- | -| | | | | | | | | | | | -| Per Share of Common Stock (€) | | | | | | | | | | | -| Basic earnings | | | | | | | | 4.52 | (0.97) | n.a. | -| Diluted earnings | | | | | | | | 4.52 | (0.97) | n.a. | -| | | | | | | | | | | | -Net cash generated from / (used in) operating activities 5.52 (4.19) n.a. -| | | | | | | | | | | | -| ---------------------------- | --- | --- | --- | --- | --- | --- | -------- | -------- | ---- | ---- | -| Dividend | | | | | | | | 3.00 | – | n.a. | -| | | | | | | | | | | | -| Share price at end of period | | | | | | | 313.90 | 233.60 | | 34% | -| | | | | | | | | | | | -| | | | | | | | | | | | -| Other (at end of period) | | | | | | | | | | | -| Number of employees | | | | | | | 58,007 | 61,445 | | (6%) | -Number of shares outstanding 195,066,060 195,032,889 0% -| | | | | | | | | | | | -| ---------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | ------------- | ---- | --- | -| Average number of shares | | | | | | | 195,066,060 | 195,266,358 | | 0% | -| | | | | | | | | | | | -| 1 Twelve-month trailing average. | | | | | | | | | | | -2 Based on shareholders' equity. -3 Adjusted net borrowings = short-term borrowings + long-term borrowings and future cash used in lease and pension liabilities – cash and cash equivalents and short-term financial assets. -4 EBITDA of last twelve months. -All income and expenses of the Reebok business are reported as discontinued operations as of the first quarter 2021. The prior year figures of the consolidated income statement and the -consolidated statement of cash flows have been restated to show the discontinued operations separately from continuing operations. - -1 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -BUSINESS PERFORMANCE -ECONOMIC AND SECTOR DEVELOPMENT -GLOBAL ECONOMY RECOVERS IN FIRST HALF OF 20211 -Around the globe, economies have been recovering from the enormous negative shock the coronavirus -pandemic had caused. The pace of this recovery, however, remains uneven. Developed economies have -been benefiting from substantial vaccination progress and fiscal support, while world trade continues to -be constrained by supply bottlenecks and travel restrictions. Economic activity is being dampened among -developing economies by a partial withdrawal of macroeconomic support and somewhat limited access to -vaccines amid an emergence of new coronavirus variants. Against this backdrop, consumer confidence -rebounded despite elevated unemployment rates and heightened policy uncertainty. -SPORTING GOODS INDUSTRY RETURNS TO GROWTH IN THE FIRST SIX MONTHS OF 2021 -In the first half of 2021, the global sporting goods industry recovered from negative impacts caused by -lockdown and quarantine measures as a result of the coronavirus pandemic. The return of major amateur -and professional sports events, as well as elevated health awareness, drove demand for sporting goods, -and consumers returned to retail stores in all parts of the world. The industry also benefited from excess -household spending, while coronavirus pandemic-related disruptions, supply chain challenges as well as -the geo-political situation weighed on the recovery of the sector. -INCOME STATEMENT -FOCUS ON CONTINUING OPERATIONS -Due to the planned divestiture of the Reebok business, all related income and expenses are reported as -discontinued operations at the end of June 2021. All figures related to the 2020 financial year in this report -refer to the company’s continuing operations unless otherwise stated. -ADIDAS WITH STRONG FINANCIAL PERFORMANCE IN THE FIRST HALF OF 2021 -In the first half of 2021, revenues increased 40% on a currency-neutral basis. In euro terms, revenues -grew 34% to € 10.345 billion (2020: € 7.733 billion). This growth was achieved against the backdrop of -prolonged lockdowns in many parts of the world, industry-wide supply chain challenges and the geo- -political situation. From a category perspective, growth was driven by exceptional increases in all strategic -performance categories amid the return of major sports events, while the lifestyle category recorded -excellent growth as well. From a regional perspective, on a currency-neutral basis, sales were driven by -exceptional increases in all strategic growth markets EMEA, North America and Greater China. Both Asia- -Pacific and Latin America grew at strong double-digit rates as well. -Gross margin improved 1.4 percentage points to 51.8% in the first six months of 2021 (2020: 50.4%). While -negative currency fluctuations as well as the less favorable channel mix weighed on the gross margin -development, this was more than offset by a significantly better pricing mix due to lower promotional -activity as well as the non-recurrence of last year’s inventory allowances and purchase order cancellation -costs. -1 Source: World Bank, Global Economic Prospects. -2 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -Royalty and commission income increased 29% to € 31 million (2020: € 24 million), while other operating -income declined 50% to € 10 million (2020: € 20 million). -| KEY FINANCIAL HIGHLIGHTS | | | | | | | -| ------------------------------------- | --- | --- | ------------ | ------------ | --------- | --- | -| | | | First half | First half | | | -| | | | | | | | -| | | | year 2021 | year 2020 | Change | | -| Operating Highlights (€ in millions) | | | | | | | -| Net sales1 | | | | 10,345 | 7,733 | 34% | - -| Operating profit/(loss)1 | | | | 1,248 | (215) | n.a. | -| ------------------------ | --- | --- | --- | ------- | ------- | ---- | -| | | | | | | | -Net income/(loss) from continuing operations1 890 (217) n.a. -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Net income/(loss) attributable to shareholders2 955 (264) n.a. -| | | | | | | | -| ------------- | --- | --- | --- | ------------- | ----- | ----- | -| Key Ratios | | | | | | | -| Gross margin1 | | | | 51.8% 50.4% | | 1.4pp | - -Other operating expenses in % of net sales1 40.2% 53.8% (13.6pp) -| | | | | | | | -| ----------------- | --- | --- | --- | -------------- | -------- | --- | -| Operating margin1 | | | | 12.1% (2.8%) | 14.8pp | | -| | | | | | | | -Per Share of Common Stock (€) -| | | | | | | | -| ----------------- | --- | --- | --- | ------ | -------- | ---- | -| Diluted earnings2 | | | | 4.89 | (1.35) | n.a. | -| | | | | | | | -| | | | | | | | -| | | | | | | | -1 Figures reflect continuing operations as a result of the planned divestiture of the Reebok business. - -| 2 Includes continuing and discontinued operations. | | | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | -Other operating expenses, including depreciation and amortization, consist of marketing and point-of-sale -as well as operating overhead expenses. In the first half of 2021, other operating expenses were flat at -€ 4.154 billion (2020: € 4.156 billion). As a percentage of sales, other operating expenses decreased -13.6 percentage points to 40.2% (2020: 53.8%). Marketing and point-of-sale expenses amounted to -€ 1.157 billion (2020: € 1.174 billion), a 1% decline compared to the prior year. The company significantly -increased its digital marketing investments to support its DTC storytelling and e-commerce business, -while physical activities remained restrained in many parts of the world during a significant portion of the -first six months of the year. As a percentage of sales, marketing and point-of sale expenses declined by 4.0 -percentage points to 11.2% in the first half of 2021 (2020: 15.2%). Operating overhead expenses remained -flat at € 2.997 billion (2020: € 2.983 billion). Efficiency measures and the non-recurrence of extraordinarily -high purchase order cancellation costs in the prior year period were offset by increased logistics costs -resulting from the significant e-commerce growth as well as temporary stranded costs related to the -intended divestiture of Reebok in the amount of around € 120 million. As a percentage of sales, operating -overhead expenses decreased 9.6 percentage points to 29.0% (2020: 38.6%). In the first six months of -2021, adidas recorded an operating profit of € 1.248 billion (2020: loss of € 215 million), resulting in an -operating margin of 12.1% (2020: -2.8%). -Financial income was down 67% to € 6 million (2020: € 17 million) while financial expenses decreased -9% to € 75 million (2020: € 82 million). Consequently, net financial expenses amounted to € 69 million -compared to € 65 million in the prior year’s period. The company recorded income taxes of € 289 million -resulting in a tax rate of 24.5%. During the six month period, the company incurred net income from -continuing operations of € 890 million (2020: loss of € 217 million), resulting in basic and diluted earnings -per share (EPS) from continuing operations of € 4.52 (2020: € -0.97). ► SEE FINANCIAL HIGHLIGHTS -In the first half of 2021, adidas’ gain from discontinued operations net of tax amounted to € 72 million -(2020: loss of € 74 million). As a result, the net income attributable to shareholders, which, in addition to -the net income from continuing operations, includes the gain from discontinued operations, came in at -€ 955 million (2020: loss of € 264 million). Consequently, both basic and diluted EPS from continuing and -discontinued operations were € 4.89 (2020: € -1.35). -3 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -The total number of shares outstanding remained stable in the first half of 2021. The average number of -shares used in the calculation of EPS was 195,066,060. ► SEE FINANCIAL HIGHLIGHTS -STATEMENT OF FINANCIAL POSITION AND STATEMENT OF CASH FLOWS -Changes in the statement of financial position are discussed in relation to the respective positions at the -end of June 2020. -PLANNED DIVESTITURE OF THE REEBOK BUSINESS IMPACTS BALANCE SHEET ITEMS -At June 30, 2021, all assets and liabilities of the Reebok business are presented as assets and liabilities -classified as held for sale due to a concrete plan to sell that business. At the end of the first half of 2021, -assets of € 1,674 million and liabilities of € 667 million were allocated to the Reebok business. However, a -restatement of the 2020 balance sheet items is not permitted under IFRS. -ASSETS -At the end of June 2021, total assets were up 8% to € 21.877 billion compared to the prior year (2020: -€ 20.301 billion), mainly driven by an increase in current assets. -Total current assets increased 29% to € 14.136 billion at the end of June 2021 (2020: € 10.931 billion). -Cash and cash equivalents were up 106% to € 4.151 billion (2020: € 2.018 billion), as net cash generated -from operating activities was only partly offset by net cash used for investing and financing activities. -Inventories decreased 22% to € 4.054 billion versus the prior year level of € 5.213 billion. This decline -reflects strong sell-through of the company’s products and successful inventory management across all -markets. On a currency-neutral basis, inventories also decreased 22%. Accounts receivable increased by -24% to € 2.324 billion (2020: € 1.869 billion), reflecting the company’s strong top-line growth. On a -currency-neutral basis, receivables were up 26%. Other current assets rose 1% to € 1.057 billion (2020: -€ 1.051 billion). Assets classified as held for sale increased to € 1.674 billion due to the planned -divestiture of the Reebok business. ► SEE NOTE 04 -Total non-current assets decreased 17% to € 7.741 billion at the end of June 2021 (2020: € 9.369 billion). -Fixed assets were down 19% to € 6.248 billion (2020: € 7.739 billion). Deferred tax assets were down 1% to -€ 1.154 billion (2020: € 1.166 billion). -STRUCTURE OF STATEMENT OF FINANCIAL POSITION1 (IN % OF TOTAL ASSETS) -| | | | June 30, 2021 | June 30, 2020 | | -| ----------------------- | --- | --- | ---------------- | ---------------- | --- | -| | | | | | | -| Assets (€ in millions) | | | 21,877 | 20,301 | | - -| Cash and cash equivalents | | | | 19.0 | 9.9 | -| ------------------------- | --- | --- | --- | ------ | --- | - -| Accounts receivable | | | | 10.6 | 9.2 | -| ------------------- | --- | --- | --- | ------ | ---- | -| | | | | | | -| Inventories | | | | 18.5 | 25.7 | -| | | | | | | -| Fixed assets | | | | 28.6 | 38.1 | -| | | | | | | -| Other assets | | | | 23.3 | 17.1 | -| | | | | | | - 1 For absolute figu res, see adidas AG Consolidated Statement of Financial Position. -LIABILITIES AND EQUITY -Total current liabilities remained virtually flat at € 9.161 billion at the end of June 2021 (2020: -€ 9.137 billion). Accounts payable were down 16% to € 2.165 billion (2020: € 2.575 billion), mainly due to -the normalization of payment terms. On a currency-neutral basis, accounts payable also decreased 16%. -Short-term borrowings declined 44% to € 682 million at the end of June 2021 (2020: € 1,217 million), -mainly reflecting lower borrowings compared to prior year, partly offset by the reclassification of a -4 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -eurobond (nominal value € 600 million) due to its maturity in October 2021. Liabilities classified as held for -sale increased to € 667 million due to the planned divestiture of the Reebok business. -Total non-current liabilities increased 14% to € 5.357 billion at the end of June 2021 (2020: € 4.693 billion). -Long-term borrowings increased 55% to € 2.473 billion compared to the prior year (2020: € 1.599 billion), -reflecting various bond placements in the second half of 2020. The non-current lease liability declined 7% -to € 2.206 billion (2020: € 2.374 billion) mainly due to the reclassification related to the planned divestiture -of the Reebok business. -The total number of shares outstanding remained fairly stable and slightly increased by 33,171 shares to -195,066,060 at the end of June 2021 compared to the prior year. ► SEE FINANCIAL HIGHLIGHTS -Shareholders’ equity increased 14% to € 7.106 billion at the end of June 2021 (2020: € 6.230 billion) mainly -due to the increase in net income. Consequently, the company’s equity ratio increased 1.8 percentage -points from 30.7% to 32.5%. -STRUCTURE OF STATEMENT OF FINANCIAL POSITION1 (IN % OF TOTAL LIABILITIES AND EQUITY) -| | | | June 30, 2021 | June 30, 2020 | | -| --------------------------------------- | --- | --- | ---------------- | ---------------- | --- | -| | | | | | | -| Liabilities and equity (€ in millions) | | | 21,877 | 20,301 | | - -| Short-term borrowings | | | | 3.1 | 6.0 | -| --------------------- | --- | --- | --- | ----- | --- | - -| Accounts payable | | | | 9.9 | 12.7 | -| -------------------- | --- | --- | --- | -------- | -------- | -| | | | | | | -| Long-term borrowings | | | | 11.3 | 7.9 | -| | | | | | | -| Other liabilities | | | | 42.0 | 41.6 | -| Total equity | | | | 33.6 | 31.9 | -| | | | | | | - 1 For absolute fig ures, see adidas AG Consolidated Statement of Financial Position. -OPERATING WORKING CAPITAL -Operating working capital decreased 7% to € 4.213 billion at the end of June 2021 (2020: € 4.506 billion). -On a currency-neutral basis, operating working capital was down 5%. Average operating working capital -as a percentage of sales decreased 1.8 percentage points to 21.4% (2020: 23.2%), reflecting the company’s -accelerating top-line momentum and focus on tight inventory management. ► SEE FINANCIAL HIGHLIGHTS -LIQUIDITY ANALYSIS -In the first half of 2021, net cash generated from operating activities was € 1.172 billion (2020 net cash -used: € 824 million). This development was driven by the operating profit generated during the period as -well as the decrease in operating working capital. Net cash used in investing activities rose to € 86 million -(2020 net cash generated: € 51 million). The majority of investing activities in the first half of 2021 related -to spending for property, plant and equipment, such as investments into controlled space and the further -development of the company’s headquarter infrastructure. Net cash used in financing activities totaled -€ 945 million (2020 net cash generated: € 601 million), mainly due to the dividend paid to shareholders. -Also as a result of these developments, cash and cash equivalents increased € 2.133 billion from € 2.018 -billion at the end of June 2020 to € 4.151 billion at the end of June 2021. Adjusted net borrowings at June -30, 2021, amounted to € 3.146 billion, representing a decline of € 1.842 billion compared to adjusted net -borrowings of € 4.988 billion at the end of June 2020. This development was mainly due to cash generated -from operating activities, partly offset by an increase in long-term borrowings. The company’s ratio of -adjusted net borrowings over EBITDA amounted to 0.9. ► SEE FINANCIAL HIGHLIGHTS -5 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -BUSINESS PERFORMANCE BY SEGMENT -At the beginning of 2021, we launched our new strategy ‘Own the Game’ for the period until 2025. As part -of this strategy, we are focusing our growth efforts on the three strategic markets Greater China, EMEA -and North America. To be able to execute this strategy successfully, adidas has changed its organizational -structure. Since January 1, 2021, adidas manages Greater China as a separate market. The remaining -Asia-Pacific (APAC) market now comprises Japan, South Korea, Southeast Asia and the Pacific region. The -change reflects the increasing importance of Greater China as a growth market for the company. In -addition, adidas created the EMEA (Europe, Middle East and Africa) market. To better leverage economies -of scale, the company has integrated the former markets Europe, Russia/CIS and Emerging Markets into -the newly formed EMEA market. The markets North America and Latin America remain unchanged. -Compared to the first half of 2020 when adidas’ top line was negatively impacted by the very large number -of temporary store closures as result of the coronavirus pandemic, sales in all segments continued to -recover and grew at a double-digit rate in both Performance and Lifestyle in the first half of 2021. In most -segments, lower inventory write-offs and a more favorable pricing mix due to less promotional activity had -a positive effect on the gross margin development. At the same time, a less favorable channel and -category mix, negative currency developments and higher sourcing costs negatively impacted the gross -margin development in most segments. Operating expenses as a percentage of sales declined in all -segments, resulting in higher operating margins across all segments. -EMEA -Sales in EMEA increased 41% on a currency-neutral basis. In euro terms, sales increased 38% to -€ 3.680 billion. -Gross margin in EMEA decreased 0.1 percentage points to 50.8%, as lower inventory write-offs and a more -favorable pricing mix due to less promotional activity were more than offset by a less favorable channel -and category mix, negative currency developments, and higher sourcing costs. Operating expenses grew -3% to € 1.082 billion, reflecting higher marketing expenditure. As a percentage of sales, operating -expenses were down 10.1 percentage points to 29.4%. The operating margin increased 10.0 percentage -points to 21.5%. Operating profit in EMEA increased 157% to € 792 million. -| EMEA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | -------------------- | ----------- | --- | -| Net sales | | | | | 3,680 | 2,670 | 38% | 41% | -| Gross profit | | | | | 1,870 | 1,360 | 38% | – | -| | | | | | | | | | -| Gross margin | | | | 50.8% | | 50.9% (0.1pp) | | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 792 | 308 | 157% | – | -| | | | | | | | | | -| Segmental operating margin | | | | 21.5% | | 11.5% 10.0pp | | – | -| | | | | | | | | | -| | | | | | | | | | - -6 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -NORTH AMERICA -Sales in North America increased 38% on a currency-neutral basis. In euro terms, sales rose 27% to -€ 2.406 billion. -Gross margin in North America increased 4.2 percentage points to 45.5%. Lower inventory write-offs and a -more favorable pricing mix due to less promotional activity more than offset a less favorable channel and -category mix, as well as higher sourcing costs. Operating expenses decreased 12% to € 661 million, -reflecting a decline in both marketing expenditure and operating overhead costs. Operating expenses as a -percentage of sales decreased 12.0 percentage points to 27.5%. The operating margin increased 16.2 -percentage points to 18.6%. Operating profit in North America rose 893% to € 447 million. -| NORTH AMERICA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 | Change neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | ------------ | ------------------- | --- | -| Net sales | | | | | 2,406 | 1,892 | 27% | 38% | -| Gross profit | | | | | 1,094 | 781 | 40% | – | -| | | | | | | | | | -| Gross margin | | | | 45.5% | | 41.3% | 4.2pp | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 447 | 45 | 893% | – | -| Segmental operating margin | | | | 18.6% | | 2.4% | 16.2pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -GREATER CHINA -Sales in Greater China increased 38% on a currency-neutral basis. In euro terms, sales rose 37% to -€ 2.405 billion. -Gross margin in Greater China was down 2.1 percentage points to 54.0%. Lower inventory write-offs were -more than offset by a less favorable channel, category and pricing mix, negative currency developments -and higher sourcing costs. Operating expenses were up 21% to € 542 million, reflecting an increase in -both marketing expenditure and operating overhead costs. Operating expenses as a percentage of sales -declined 2.9 percentage points to 22.5%. The operating margin increased 0.8 percentage points to 31.5%. -Operating profit in Greater China rose 40% to € 758 million. -| GREATER CHINA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 | Change neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | ------------ | ------------------- | --- | -| Net sales | | | | | 2,405 | 1,757 | 37% | 38% | -| Gross profit | | | | | 1,299 | 987 | 32% | – | -| | | | | | | | | | -| Gross margin | | | | 54.0% | | 56.2% | (2.1pp) | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 758 | 540 | 40% | – | -| | | | | | | | | | -| Segmental operating margin | | | | 31.5% | | 30.7% | 0.8pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -7 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ASIA-PACIFIC -Sales in Asia-Pacific increased 26% on a currency-neutral basis. In euro terms, sales rose 21% to -€ 1.135 billion. -Gross margin in Asia-Pacific declined 0.3 percentage points to 52.4%. Lower inventory write-offs were -more than offset by a less favorable channel, category and pricing mix, negative currency developments -and higher sourcing costs. Operating expenses were down 7% to € 317 million, reflecting a decline in both -marketing expenditure and operating overhead costs. As a percentage of sales, operating expenses -decreased 8.5 percentage points to 27.9%. The operating margin increased 8.1 percentage points to -25.1%. Operating profit in Asia-Pacific rose 78% to € 285 million. -| ASIA-PACIFIC AT A GLANCE € IN MILLIONS | | | | | | | | | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ------- | -------------------- | ----------- | --- | -| Net sales | | | | | 1,135 | 939 | 21% | 26% | -| Gross profit | | | | | 595 | 494 | 20% | – | -| | | | | | | | | | -| Gross margin | | | | 52.4% | | 52.6% (0.3pp) | | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 285 | 160 | 78% | – | -| Segmental operating margin | | | | 25.1% | | 17.1% | 8.1pp | – | -| | | | | | | | | | - -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -LATIN AMERICA -Sales in Latin America increased 81% on a currency-neutral basis. In euro terms, sales rose 59% to -€ 645 million. -Gross margin in Latin America increased 1.5 percentage points to 48.2%, as lower inventory write-offs and -a more favorable pricing mix due to less promotional activity more than offset a less favorable channel -mix, negative currency developments and higher sourcing costs. Operating expenses were down 3% to -€ 190 million, reflecting a decline in operating overhead costs. Operating expenses as a percentage of -sales were down 18.5 percentage points to 29.4%. The operating margin increased 19.1 percentage points -to 18.8%. Operating profit in Latin America rose to € 121 million. -| LATIN AMERICA AT A GLANCE € IN MILLIONS | | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -Change -| | | | | First half | | First half | | | -| --- | --- | --- | --- | ---------------- | --- | -------------- | --- | --- | -(currency- -| | | | | year 2021 | | year 2020 Change | neutral) | | -| -------------------------- | --- | --- | --- | ------------ | ----- | -------------------- | ----------- | --- | -| Net sales | | | | | 645 | 406 | 59% | 81% | -| Gross profit | | | | | 311 | 189 | 64% | – | -| | | | | | | | | | -| Gross margin | | | | 48.2% | | 46.7% | 1.5pp | – | -| | | | | | | | | | -| Segmental operating profit | | | | | 121 | (1) | n.a. | – | -| | | | | | | | | | -| Segmental operating margin | | | | 18.8% | | (0.2%) 19.1pp | | – | -| | | | | | | | | | -| | | | | | | | | | -8 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -OUTLOOK2 -GLOBAL ECONOMIC GROWTH TO RECOVER IN 20213 -Global GDP is forecast to grow 5.6% in 2021, with the recovery projected to be more robust than initially -expected. Due to rapid vaccination progress in major economies, the recovery is expected to be unevenly -distributed and, in some cases, delayed. It largely reflects sharp rebounds in some major economies amid -highly unequal vaccine access, with many poorer countries facing subdued prospects. The strength of the -near-term global recovery is largely attributable to a few major economies, such as the United States and -China, with many emerging market and developing economies lagging behind. The United States and -China are each expected to contribute over one-quarter of global growth in 2021. -SPORTING GOODS INDUSTRY TO RECOVER NOTABLY IN 2021 -In the absence of any major delay in the pandemic recovery or other macroeconomic shocks, we expect -the global sporting goods industry to recover notably in 2021. Consumer confidence is forecast to rebound -amid widespread vaccination and effective pandemic management as well as major sports events -returning to the global stage. While extended lockdown periods, quarantine measures and physical store -closures are still forecast to have an adverse impact in the short term, the sporting goods industry is -projected to remain fundamentally attractive in the long term, as existing global trends are accelerating. -Sports-inspired apparel and footwear (‘athleisure’) keeps evolving as a structural component of the -broader fashion landscape, complemented by increasing sports participation rates as well as rising -awareness for health and wellness. In addition, sustainability is expected to further gain in importance -amid growing environmental awareness of consumers. Moreover, purchasing behavior is forecast to -further shift toward online and social channels, and consumer insights generation along with the creation -of premium shopping experiences become increasingly important. For the sporting goods sector, risks -related to disruptions caused by the coronavirus pandemic, supply chain challenges, re-escalating trade -tensions as well as the geo-political situation remain. -ADIDAS INCREASES OUTLOOK FOR FULL YEAR 2021 -While adidas continues to be impacted by Covid-19-related lockdowns, industry-wide supply-chain -challenges and the geo-political situation, the company has increased its top- and bottom-line outlook for -2021. -Given the accelerating top-line momentum, the company now expects currency-neutral sales to increase -at a rate of up to 20% year-over-year in 2021, driven by strong double-digit improvements in all markets. -This new outlook reflects sales growth of up to 7% in the second half of the year compared to the 2020 -level, which will be fueled by an array of innovative product releases, including the re-introduction of NMD, -one of adidas’ most successful franchises in recent history. In addition, the company’s Futurecraft 4D -2 This Management Report contains forward-looking statements that reflect Management’s current view with respect to the future development of adidas. The -outlook is based on estimates that we have made on the basis of all the information available to us at the time of completion of this First Half Year Report. In -addition, such forward-looking statements are subject to uncertainties as described in the Risk and Opportunity Report of the adidas 2020 Annual Report which -are beyond the control of the company. In case the underlying assumptions turn out to be incorrect or described risks or opportunities materialize, actual -results and developments may materially deviate (negatively or positively) from those expressed by such statements. adidas does not assume any obligation to -update any forward-looking statements made in this Management Report beyond statutory disclosure obligations. -3 Source: World Bank, Global Economic Prospects. -9 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -portfolio will be significantly expanded by the introduction of the 4D FWD Pulse. The scaling of the well- -established ZX franchise will continue with the introduction of the ZX 5K Boost. The Forum franchise, after -having been successfully incubated throughout the first two quarters of the year, will quadruple in size -during the second half of the year driven by several iterations as well as dedicated marketing activities. In -addition, major sport events, including the Olympic Games in Japan, the start of the club football season in -Europe and the kick-off to the NFL season in the US, as well as grassroots activities around the world will -provide ideal platforms to tell brand and product stories in front of global and local audiences. -The company’s full-year gross margin forecast continues to be for a level of around 52.0%. The operating -margin is now expected to increase to a level of between 9.5% and 10% (previously: 9% to 10%). Net -income from continuing operations is now projected to increase to a level of between € 1.4 billion and -€ 1.5 billion (previously: € 1.25 billion to € 1.45 billion). The company’s profitability outlook continues to -include temporary stranded costs related to the intended divestiture of Reebok. In 2021, these costs are -expected to amount to around € 250 million on the operating profit level and to impact net income from -continuing operations by approximately € 200 million. The medium-term growth outlook is not impacted -by these costs as adidas anticipates that only around 30% of the € 250 million will reoccur in 2022 and that -by 2023 the stranded costs will be fully eliminated. -RISKS AND OPPORTUNITIES -Uncertainties regarding the further development of the coronavirus pandemic remain high, in particular in -light of rising infection rates in Asia and Europe. Compared to the assessment in the 2020 Annual Report, -the company’s risk profile has slightly deteriorated due to an increased risk related to consumer demand -as a consequence of the geo-political situation and product supply due to additional supply chain -challenges. Macroeconomic, regulatory and currency risks could also have a high impact on the -company’s financial results. While we still see major opportunities related to consumer demand and -product offering as well as data analytics in the mid to long term, we consider upside potential related to -short-term opportunities as limited. -Nevertheless, given the company’s strong risk-bearing capacity due to its current liquidity position and -financial health, Management does not foresee any material jeopardy to the viability of the company as a -going concern. We remain confident that the earnings strength forms a solid foundation for our business -development and provides the necessary resources to pursue future opportunities. -10 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF FINANCIAL -POSITION -ADIDAS AG CONSOLIDATED STATEMENT OF FINANCIAL POSITION (IFRS) € IN MILLIONS - June 30, 2021 June 30, 2020 Change in % December 31, 2020 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | -Assets -| Cash and cash equivalents | | | | | 4,151 | 2,018 | 105.7 | 3,994 | -| ---------------------------------- | --- | --- | --- | --------- | --------- | ------- | -------- | ------- | -| Short-term financial assets | | | | | – | 6 | n.a. | 0 | -| | | | | | | | | | -| Accounts receivable | | | | | 2,324 | 1,869 | 24.4 | 1,952 | -| | | | | | | | | | -| Other current financial assets | | | | | 810 | 653 | 24.0 | 702 | -| | | | | | | | | | -| Inventories | | | | | 4,054 | 5,213 | (22.2) | 4,397 | -| | | | | | | | | | -| Income tax receivables | | | | | 66 | 121 | (45.5) | 109 | -| | | | | | | | | | -| Other current assets | | | | | 1,057 | 1,051 | 0.6 | 999 | -| | | | | | | | | | -| Assets classified as held for sale | | | | | 1,674 | – | n.a. | 0 | -| | | | | | | | | | -| Total current assets | | | | 14,136 | 10,931 | | 29.3 | 12,154 | -| | | | | | | | | | -| | | | | | – | – | | | -Property, plant and equipment 2,065 2,293 (9.9) 2,157 -| | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------- | ------- | -------- | ----- | -| Right-of-use assets | | | | | 2,430 | 2,733 | (11.1) | 2,430 | -| | | | | | | | | | -| Goodwill | | | | | 1,199 | 1,258 | (4.7) | 1,208 | -| | | | | | | | | | -| Trademarks | | | | | 16 | 820 | (98.0) | 750 | -| | | | | | | | | | -| Other intangible assets | | | | | 261 | 284 | (7.9) | 252 | -| | | | | | | | | | -| Long-term financial assets | | | | | 276 | 351 | (21.4) | 353 | -| | | | | | | | | | -Other non-current financial assets 243 340 (28.7) 414 -| Deferred tax assets | | | | | 1,154 | 1,166 | (1.0) | 1,233 | -| ------------------------ | --- | --- | --- | --- | ------- | ------- | -------- | ----- | -| | | | | | | | | | -| Other non-current assets | | | | | 96 | 124 | (22.5) | 103 | -| | | | | | | | | | -Total non-current assets 7,741 9,369 (17.4) 8,899 -| | | | | | | | | | -| ------------- | --- | --- | --- | --------- | --------- | --- | ------ | ------- | -| Total assets | | | | 21,877 | 20,301 | | 7.8 | 21,053 | -| | | | | | | | | | - -11 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ADIDAS AG CONSOLIDATED STATEMENT OF FINANCIAL POSITION (IFRS) € IN MILLIONS - June 30, 2021 June 30, 2020 Change in % December 31, 2020 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | -Liabilities and equity -| Short-term borrowings | | | | | 682 | 1,217 | (44.0) | 686 | -| --------------------------------------- | --- | --- | --- | --- | -------- | -------- | -------- | ------ | -| Accounts payable | | | | | 2,165 | 2,575 | (15.9) | 2,390 | -| | | | | | | | | | -| Current lease liabilities | | | | | 480 | 639 | (24.9) | 563 | -| | | | | | | | | | -| Other current financial liabilities | | | | | 340 | 265 | 28.5 | 446 | -| | | | | | | | | | -| Income taxes | | | | | 686 | 588 | 16.5 | 562 | -| | | | | | | | | | -| Other current provisions | | | | | 1,433 | 1,373 | 4.3 | 1,609 | -| | | | | | | | | | -| Current accrued liabilities | | | | | 2,260 | 1,933 | 16.9 | 2,172 | -| | | | | | | | | | -| Other current liabilities | | | | | 450 | 547.0 | (17.6) | 398 | -| | | | | | | | | | -| Liabilities classified as held for sale | | | | | 667 | – | n.a. | – | -| | | | | | | | | | -| Total current liabilities | | | | | 9,161 | 9,137 | 0.3 | 8,827 | -| | | | | | | | | | -| Long-term borrowings | | | | | 2,473 | 1,599 | 54.7 | 2,482 | -| | | | | | | | | | -Non-current lease liabilities 2,206 2,374 (7.1) 2,159 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other non-current financial liabilities 98 46 113.0 115 -| | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | ----- | ----- | ------- | --- | -| Pensions and similar obligations | | | | | 234 | 245 | (4.5) | 284 | -| | | | | | | | | | -Deferred tax liabilities 189 224 (15.7) 241 -| Other non-current provisions | | | | | 134 | 179 | (25.4) | 229 | -| ------------------------------- | --- | --- | --- | --- | ----- | ----- | -------- | --- | -| | | | | | | | | | -| Non-current accrued liabilities | | | | | 6 | 9 | (35.4) | 8 | -| | | | | | | | | | -| Other non-current liabilities | | | | | 18 | 16 | 11.9 | 17 | -| | | | | | | | | | -Total non-current liabilities 5,357 4,693 14.2 5,535 -| | | | | | | | | | -| ------------------------- | --- | --- | --- | --- | -------- | ------------- | ------- | ------ | -| Share capital | | | | | 195 | 195 | 0.0 | 195 | -| | | | | | | | | | -| Reserves | | | | | (190) | (2) 8,799.1 | | (474) | -| | | | | | | | | | -| Retained earnings | | | | | 7,101 | 6,037 | 17.6 | 6,733 | -| | | | | | | | | | -| Shareholders’ equity | | | | | 7,106 | 6,230 | 14.0 | 6,454 | -| | | | | | | | | | -| Non-controlling interests | | | | | 253 | 240 | 5.2 | 237 | -| | | | | | | | | | -| | | | | | | | | | -| Total equity | | | | | 7,358 | 6,471 | 13.7 | 6,691 | -| | | | | | | | | | -Total liabilities and equity 21,877 20,301 7.8 21,053 -| | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | - -12 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | - -| | | 1 | | | 2 | | | | 3 | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONDENSED CONSOLIDATED INCOME STATEMENT -ADIDAS AG CONDENSED CONSOLIDATED INCOME STATEMENT (IFRS) € IN MILLIONS1 -| | | | | | | | | | | Second | | Second | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | ------- | --- | - First half First half quarter quarter - year 2021 year 2020 Change 2021 2020 Change -| Net sales | | | | | | 10,345 | | 7,733 | 33.8% | 5,077 | | 3,352 | 51.5% | -| --------- | --- | --- | --- | --- | --- | -------- | --- | ----- | ------- | ------- | --- | ------- | ----- | -Cost of sales 4,983 3,835 30.0% 2,446 1,631 49.9% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Gross profit 5,362 3,898 37.5% 2,632 1,720 53.0% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 51.8% 50.4% 1.4pp 51.8% 51.3% 0.5pp -Royalty and commission income 31 24 29.5% 17 9 95.7% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other operating income 10 20 (50.2%) 2 18 (86.1%) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other operating expenses 4,154 4,156 (0.0%) 2,107 2,010 4.9% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 40.2% 53.8% (13.6pp) 41.5% 60.0% (18.5pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Marketing and point-of-sale expenses 1,157 1,174 (1.4%) 616 525 17.2% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 11.2% 15.2% (4.0pp) 12.1% 15.7% (3.5pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Operating overhead expenses2 2,997 2,983 0.5% 1,492 1,484 0.5% -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 29.0% 38.6% (9.6pp) 29.4% 44.3% (14.9pp) -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Operating profit / (loss) 1,248 (215) n.a. 543 (263) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 12.1% (2.8%) 14.8pp 10.7% (7.8%) 18.5pp -| Financial income | | | | | | | 6 | 17 | (67.2%) | | 4 | 13 | (70.0%) | -| ------------------ | --- | --- | --- | --- | --- | --- | ---- | --- | --------- | --- | ---- | ---- | ------- | -| | | | | | | | | | | | | | | -| Financial expenses | | | | | | | 75 | 82 | (8.7%) | | 37 | 42 | (11.0%) | -| | | | | | | | | | | | | | | -Income / (loss) before taxes 1,179 (279) n.a. 510 (292) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 11.4% (3.6%) 15.0pp 10.0% (8.7%) 18.8pp -| Income taxes | | | | | | | 289 | (63) | | n.a. 123 | | (49) | n.a. | -| ------------ | --- | --- | --- | --- | --- | --- | ----- | ---- | --- | ---------- | --- | ------ | ---- | -| | | | | | | | | | | | | | | -(% of / income / (loss) before taxes) 24.5% 22.4% 2.1pp 24.0% 16.9% 7.1pp -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) from continuing operations 890 (217) n.a. 387 (243) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 8.6% (2.8%) 11.4pp 7.6% (7.2%) 14.9pp -Gain / (loss) from discontinued operations, net of tax 72 (74) n.a. 20 (74) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) 962 (291) n.a. 407 (317) n.a. -(% of net sales) 9.3% (3.8%) 13.1pp 8.0% (9.4%) 17.5pp -Income / (loss) attributable to shareholders 955 (264) n.a. 397 (295) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(% of net sales) 9.2% (3.4%) 12.6pp 7.8% (8.8%) 16.6pp -Net income / (loss) attributable to non-controlling interests 7 (27) n.a. 11 (22) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Basic earnings per share from continuing operations (in €) 4.52 (0.97) n.a. 1.93 (1.13) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Diluted earnings per share from continuing operations (in €) 4.52 (0.97) n.a. 1.93 (1.13) n.a. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | | | -Basic earnings per share from continuing and discontinued -| | | | | | | | 4.89 | (1.35) | | n.a. 2.03 | | (1.51) | n.a. | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------- | --- | -------------- | --- | ---------- | ----- | - operations (in €) -| D il u | t e d e a r | n i n g s per share from continuing and discontinued | | | | | | | | | | | | -| ------ | -------------- | ------------------------------------------------------ | --- | --- | --- | --- | ------- | ------- | --- | ------------- | --- | --------- | ----- | -| | | ) | | | | | 4.89 | (1.35) | | n.a. 2.03 | | (1.51) | n.a. | -| op e | r a t io n s | (i n € | | | | | | | | | | | | - 1 Comparative informa tion for 2020 has been adjusted due to the classification of th e R eebok business a s discontinued o pe rations. -| | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -2 Aggregated distribution and selling expenses, general and administration expenses, sundry expenses and impairment losses (net) on accounts receivable and contract assets. - -13 - - HALF YEAR REPORT 2021 - -| 1 | | | 2 | | | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF COMPREHENSIVE -INCOME -ADIDAS AG CONSOLIDATED STATEMENT OF COMPREHENSIVE INCOME (IFRS) € IN MILLIONS -| | | | | | First half | | First half | | Second | | Second | -| --- | --- | --- | --- | ---- | ----------- | --- | ----------- | ----------------- | ------- | ------------- | ------- | -| | | | | | year 2021 | | year 2020 | quarter 2021 | | | | -| | | | | | | | | | | quarter 2020 | | -Net income / (loss) after taxes 962 (291) 407 (317) -Items of other comprehensive income that -will not be reclassified subsequently to -profit or loss -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Remeasurements of defined benefit plans -| | | | | | | 43 | | 0 | | 6 | (19) | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | ---- | -(IAS 19), net of tax1 -| Net (loss) / gain on other equity investments | | | | | | | | | | | | -| ---------------------------------------------- | --- | --- | --- | --- | --- | ----- | --- | ----- | --- | --- | --- | -| | | | | | | (1) | | (2) | | 3 | (2) | -(IFRS 9), net of tax -| Subtotal of items of other comprehensive | | | | | | | | | | | | -| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -income that will not be reclassified 41 (1) 10 (21) -subsequently to profit or loss -Items of other comprehensive income that -| will be reclassified to profit or loss when | | | | | | | | | | | | -| -------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -specific conditions are met -Net gain / (loss) on cash flow hedges and -| | | | | | | 100 | | 142 | | (6) | (120) | -| --- | --- | --- | --- | --- | --- | ----- | --- | ----- | --- | ----- | ----- | -net foreign investment hedges, net of tax -| Net (loss) / gain on cost of hedging reserve – | | | | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | ----- | --- | -| | | | | | | (5) | | 9 | | (2) | 9 | -options, net of tax -| Net gain / (loss) on cost of hedging reserve – | | | | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | --- | --- | ---- | --- | ------ | --- | --- | --- | -| | | | | | | 10 | | (17) | | 5 | 23 | -forward contracts, net of tax -Currency translation differences 145 (177) 8 (34) - -Subtotal of items of other comprehensive -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -income that will be reclassified to profit or 251 (44) 5 (123) -loss when specific conditions are met -| Other comprehensive income | | | | | | 292 | | (45) | | 14 | (143) | -| --------------------------- | --- | --- | --- | --- | --- | ------ | --- | ------- | --- | ----- | ------ | -| | | | | | | | | | | | | -Total comprehensive income 1,254 (336) 422 (460) -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Attributable to shareholders of adidas AG 1,239 (311) 413 (432) -| | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Attributable to non-controlling interests 15 (24) 8 (28) - 1 Includes actuarial gains or losses relating to defined benefit obli ga tions, return on plan as sets (excluding inter e st income) and the ass e t c eiling effect. - -14 - -| | | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF CHANGES IN -EQUITY -ADIDAS AG CONSOLIDATED STATEMENT OF CHANGES IN EQUITY (IFRS) € IN MILLIONS -Cost of - Cumulative Cost of hedging -| | | | | | currency | | hedging | reserve - | | | | Share- | | Non- | | -| --- | --- | --- | --- | --- | --------- | --- | -------- | ---------- | --- | --- | --- | ------- | --- | ----- | --- | -Share Capital translation Hedging reserve - forward Other Retained holders’ controlling Total - capital reserve differences reserve options contracts reserves earnings equity interests equity -Balance at - 196 887 (470) (150) (10) 6 (218) 6,555 6,796 261 7,058 -December 31, 2019 -Other comprehensive -| | | | | | (180) | | 142 | 9 | (17) | (1) | | (48) | | 3 | (45) | -| --- | --- | --- | --- | --- | ------- | --- | ----- | --- | ------ | --- | --- | ------ | --- | --- | ---- | -income -Net income (264) (264) (27) (291) - -Total comprehensive - (180) 142 9 (17) (1) (264) (311) (24) (336) -income -Repurchase of -| | | | (1) | | | | | | | | (256) | (257) | | | (257) | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | --- | ----- | -adidas AG shares -| Repurchase of | | | | | | | | | | | | | | | | -| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -adidas AG shares due to -| equity-settled share- | | | (0) | | | | | | | | (14) | (14) | | | (14) | -| --------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | --- | --- | ---- | -based payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Reissuance of treasury -shares due to e quity- -| settled share-based | | | 0 | | | | | | | | 17 | | 17 | | 17 | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | -payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Equity-settled share- -| based pa yment | | | | | | | | | | | (1) | | (1) | | (1) | -| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -First-time consolidation -due to obtaining control -| | | | | | | | | | | | | | | 3 | 3 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -in accordance with -IFRS 10 -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Balance at 195 887 (650) (9) (1) (11) (220) 6,037 6,230 240 6,471 -June 30, 2020 -Balance at - 195 887 (850) (250) (3) (23) (235) 6,733 6,454 237 6,691 -December 31, 2020 -Other comprehensive -| | | | | | 137 | | 100 | (5) | 10 | 41 | | 284 | | 9 | 292 | -| --- | --- | --- | --- | --- | ----- | --- | ----- | ----- | ---- | --- | --- | ----- | --- | --- | --- | -income -| Net income | | | | | | | | | | | 955 | 955 | | 7 | 962 | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------ | --- | ---- | ---- | - -Total comprehensive - 137 100 (5) 10 41 955 1,239 15 1,254 -income -Repurchase of -adidas AG shares due to -| | | | (1) | | | | | | | | (15) | (15) | | | (15) | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | --- | --- | ---- | -equity-settled share- -based payment -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Reissuance of treasury -shares due to e quity- -| | | | 1 | | | | | | | | 14 | | 15 | | 15 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | -settled share-based -payment -| | | | | | | | | | | | | | | | | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --- | --- | ----- | -| Dividend payment | | | | | | | | | | | (585) | (585) | | | (585) | - -Equity-settled share- -based payment (1) (1) (1) -| Balance at | | | | | | | | | | | | | | | | -| ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - 195 887 (712) (150) (7) (14) (194) 7,101 7,106 253 7,358 -June 30, 2021 - -15 - -| | | | | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| | 1 | | | 2 | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CONSOLIDATED STATEMENT OF CASH FLOWS -ADIDAS AG CONSOLIDATED STATEMENT OF CASH FLOWS (IFRS) € IN MILLIONS -| | | | | | | | | First half | | First half | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------ | -| | | | | | | | | year 2021 | | year 2020 | -| | | | | | | | | | | | -Operating activities: -| Income / (loss) before taxes | | | | | | | | | 1,179 | (279) | -| ---------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ------- | -| | | | | | | | | | | | -Adjustments for: -Depreciation, amortization and impairment losses 568 645 -| Reversals of impairment losses | | | | | | | | | (7) | (1) | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ---- | -| | | | | | | | | | | | -| Interest income | | | | | | | | | (5) | (16) | -| | | | | | | | | | | | -| Interest expense | | | | | | | | | 69 | 77 | -| | | | | | | | | | | | -| Unrealized foreign exchange losses, net | | | | | | | | | 9 | 9 | -| | | | | | | | | | | | -Losses on sale of property, plant and equipment and intangible assets, net 4 3 -| | | | | | | | | | | | -| -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | --- | -| Other non-cash (income) /expense | | | | | | | | | (6) | 0 | -| | | | | | | | | | | | -Operating profit before working capital changes 1,810 438 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Increase) / decrease in receivables and other assets (553) 610 - -| Decrease / (increase) in inventories | | | | | | | | | 13 | (1,142) | -| ------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------- | -| | | | | | | | | | | | -Decrease in accounts payable and other liabilities (82) (539) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Cash generated from / (used in) operations before taxes 1,189 (632) -| | | | | | | | | | | | -| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| Income taxes paid | | | | | | | | | (112) | (186) | - -Net cash generated from / (used in) operating activities – continuing operations 1,077 (819) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash generated from / (used in) operating activities – discontinued operations 95 (6) - -Net cash generated from / (used in) operating activities 1,172 (824) -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | -Investing activities: -Purchase of trademarks and other intangible assets (53) (33) -| Purchase of property, plant and equipment | | | | | | | | | (130) | (176) | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| | | | | | | | | | | | -| Proceeds from sale of property, plant and equipment | | | | | | | | | 0 | 2 | -| | | | | | | | | | | | -| Proceeds from sale of a disposal group | | | | | | | | | 12 | – | -| | | | | | | | | | | | -| Proceeds from disposal of discontinued operations | | | | | | | | | 29 | – | -| | | | | | | | | | | | -| Proceeds from sale of short-term financial assets | | | | | | | | | 0 | 288 | -| | | | | | | | | | | | -Purchase of investments and other long-term assets 55 (37) -| | | | | | | | | | | | -| ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Interest received | | | | | | | | | 5 | 16 | -| | | | | | | | | | | | -Net cash (used in) / generated from investing activities – continuing operations (81) 59 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash used in investing activities – discontinued operations (5) (8) - -Net cash (used in) / generated from investing activities 86 51 -| | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | -| | | | | | | | | | | | -16 - -| | | | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| | 1 | | | 2 | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ADIDAS AG CONSOLIDATED STATEMENT OF CASH FLOWS (IFRS) € IN MILLIONS -| | | | | | | | First half | | First half | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | ---------- | ----------- | -| | | | | | | | year 2021 | year 2020 | | - -| Financing activities: | | | | | | | | | | -| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------- | ----- | -| Reverse transaction of buyback of Eurobonds | | | | | | | | – | 11 | -| Interest paid | | | | | | | | (54) | (66) | -| | | | | | | | | | | -| Repayments of lease liabilities | | | | | | | | (267) | (226) | -| | | | | | | | | | | -| Dividend paid to shareholders of adidas AG | | | | | | | | (585) | – | -| | | | | | | | | | | -| Repurchase of treasury shares | | | | | | | | – | (257) | -| | | | | | | | | | | -Repurchase of treasury shares due to share-based payments (15) (14) -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Proceeds from reissuance of treasury shares due to share-based payments 12 12 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Repayments of) / proceeds from short-term borrowings (17) 1,165 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash (used in) / generated from financing activities – continuing operations (926) 624 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Net cash used in financing activities – discontinued operations (19) (23) - -Net cash (used in) / generated from financing activities (945) 601 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| Effect of exchange rates on cash | | | | | | | | 15 | (29) | -| --------------------------------- | --- | --- | --- | --- | --- | --- | --- | ----- | ----- | -| | | | | | | | | | | -Increase / (decrease) of cash and cash equivalents 157 (201) - -Cash and cash equivalents at beginning of year 3,994 2,220 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Cash and cash equivalents at end of period 4,151 2,018 -| | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -17 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -EXPLANATORY NOTES TO THE CONDENSED INTERIM -CONSOLIDATED FINANCIAL STATEMENTS (IFRS) AS -AT JUNE 30, 2021 -01 GENERAL -The interim consolidated financial statements of adidas AG and its subsidiaries (collectively ‘adidas,’ the -‘Group’ or ‘the company’) for the first half year ending June 30, 2021, are prepared in compliance with -International Financial Reporting Standards (IFRS), as endorsed by the European Union (EU). The company -applied all International Financial Reporting Standards issued by the International Accounting Standards -Board (IASB) and Interpretations of the IFRS Interpretations Committee effective as at June 30, 2021, -insofar as they have already been adopted into European law. -These interim consolidated financial statements were prepared in compliance with International -Accounting Standard IAS 34 ‘Interim Financial Reporting.’ Accordingly, these interim consolidated -financial statements do not include all of the information and notes required for consolidated financial -statements at financial year-end. Therefore, these interim consolidated financial statements should be -read in conjunction with the 2020 annual consolidated financial statements. The accounting policies as -well as principles, practices and presentation applied in the consolidated financial statements for the year -ending December 31, 2020, also apply to the interim consolidated financial statements for the first half -year ending June 30, 2021. -The following new accounting standards and interpretations as well as amendments to existing standards, -which were issued by the IASB and are effective in the EU for financial years beginning on January 1, 2021, -have been applied for the first time: -Interest Rate Benchmark Reform Phase 2 – Amendments to IFRS 9, IAS 39, IFRS 7, IFRS 4 and IFRS 16 -─ -Amendments to IFRS 4 Incurance Contracts - deferral of IFRS 9 -─ -These do not have any material impact on the interim consolidated financial statements of adidas AG. -Further information can be found in the 2020 Annual Report. -New accounting standards and interpretations as well as amendments to existing standards that are not -yet effective in the EU or effective for financial years beginning after January 1, 2021, are not expected to -have any material impact on the consolidated financial statements. -The amendment to IFRS 16 ‘Covid-19-Related Rent Concessions’ published by the IASB on May 28, 2020, -and effective for financial years beginning on or after June 1, 2020, was originally limited to rent -concessions that were due on or before June 30, 2021. The IASB has decided to extend the period during -which the concessions can be applied to financial years beginning on or after April 1, 2021. This decision -will not impact adidas, as adidas has not applied and will not apply the option provided in this amendment -regarding the treatment of rent concessions relating to the coronavirus pandemic. The option allows -lessees to treat rental concessions in connection with the coronavirus pandemic not as lease -modifications within the meaning of IFRS 16 but to realize the effects of the rental concessions in the -periods in which they are granted. -18 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -The preparation of financial statements in conformity with IFRS requires the use of assumptions and -estimates that affect reported amounts and related disclosures. The first half of 2020 was significantly -impacted by the coronavirus pandemic. Further information can be found in Note 8 of the first half year -report 2020. Compared to the first half of 2020, business development in the first half of 2021 was less -impacted by the effects of the coronavirus pandemic. Nevertheless, due to the ongoing pandemic, it is still -difficult to predict the resulting effects on assets and liabilities as well as income and expenses. Estimates -and assumptions relevant to the financial statements were made to the best of our knowledge, based on -current events and measures. In particular, judgments and discretion are used in assessing the -recoverability of deferred tax assets, in impairment tests for non-financial assets, in determining -allowances for inventories and accounts receivable, and in the calculation of provisions. -There were no changes in the scope of consolidation in the first half of 2021. -The interim consolidated financial statements and the interim Group management report were not audited -in accordance with § 317 German Commercial Code (Handelsgesetzbuch – HGB) nor reviewed in -accordance with § 115 section 5 German Securities Trading Act (Wertpapierhandelsgesetz – WpHG) by an -auditor. -Costs that are incurred unevenly during the financial year are anticipated or deferred in the interim -consolidated financial statements only if it would be also appropriate to anticipate or defer such costs at -the end of the financial year. -The results of operations for the first half year ending June 30, 2021, are not necessarily indicative of -results to be expected for the entire year. -The interim consolidated financial statements are presented in euros (€) and, unless otherwise stated, all -values are presented in millions of euros (€ in millions). Due to rounding principles, numbers presented -may not sum up exactly to totals provided. -02 REVENUE -The sales of the company in certain product categories are seasonal, and therefore, revenues and -attributable earnings may vary within the financial year. In the years before the coronavirus pandemic, -sales and earnings tended to be strongest in the first and third quarters of the financial year because -these coincided with the launch of the spring/summer and fall/winter collections, respectively. In the first -half of 2021, sales stabilized but variations may still occur due to the ongoing coronavirus pandemic. Shifts -in the share of sales and attributable earnings of particular product categories or in the regional -composition may occur throughout the year. A disaggregation of revenue into product categories is -contained in these Notes. ► SEE NOTE 08 -03 DISCONTINUED OPERATIONS -On February 16, 2021, the company decided to begin a formal process aimed at divesting Reebok. Due to -the concrete plans to sell Reebok and the approval by the respective committees, the Reebok operating -business is reported as discontinued operations and classified as a disposal group held for sale at the end -of the first quarter. The fair value was determined based on the basis of existing non-binding purchase -price offers at this point of time. The transaction is expected to be completed later in 2021. -19 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | 3 | | | -| --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -Reebok was previously neither classified as assets held for sale nor as discontinued operations. The prior -year figures of the consolidated income statement and the consolidated statement of cash flows have been -restated to show the discontinued operations separately from continuing operations. -The results of the Reebok operations are shown as discontinued operations in the consolidated income -statement: -DISCONTINUED OPERATION REEBOK -First half First half -| € in millions | | | year 2021 | year 2020 | | -| --------------------- | --- | --- | ------------ | ------------ | ----- | -| Operational business | | | | | | -| Net Sales | | | | 823 | 600 | -| Expenses | | | | (666) | (658) | -| | | | | | | -Gain / (loss) from operating activities before taxes 157 (58) -| | | | | | | -| ------------ | --- | --- | --- | ------ | ---- | -| Income taxes | | | | (78) | (12) | -Gain / (loss) from operating activities, net of tax 78 (69) -| | | | | | | -| ----------------------------- | --- | --- | --- | ------ | --- | -| Transaction costs, net of tax | | | | (10) | – | -Gain / (loss) from discontinued operations, net of tax 68 (69) -| | | | | | | -| --- | --- | --- | --- | --- | --- | -Basic earnings per share from discontinued operations (€) 0.35 (0.36) -Diluted earnings per share from discontinued operations (€) 0.35 (0.36) - Rounding differences may arise in percentages and totals for figures presented in millions as calculation is alway s b ased on the figures s ta ted in thousands. -Gains from discontinued operations for the first half year ending June 30, 2021, in an amount of -€ 72 million (2020: losses of € 74 million) are entirely attributable to the shareholders of adidas AG. The -gains from discontinued operations also include besides Reebok an amount of € 4 million (2020: loss -€ 5 million) relating to divestitures in previous years. -Total net loss of cumulative foreign currency translation differences for discontinued operations reflected -within equity amount to € 225 million. For the first half year 2021 the loss within equity amount to -€ 125 million. For 2020 the gain within equity amount to € 368 million. -The income tax expense related to discontinued operations amounts to € 76 million. This amount includes -the revaluation of deferred tax assets and liabilities as a result of the tax rate change in the UK -(€ 34 million). - -20 - - HALF YEAR REPORT 2021 - -| 1 | 2 | 3 | | | -| --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -04 ASSETS AND LIABILITIES HELD FOR SALE -At June 30, 2021, the disposal group Reebok was stated at carrying amount and comprised the following -major classes of assets and liabilities: -GROUP OF ASSETS AND LIABILITIES -Assets classified as held-for-sale € in millions June 30, 2021 -| | | | | | -| ------------------- | --- | --- | --- | --- | -| Accounts receivable | | | | 207 | -| Inventories | | | | 388 | - -| Other current financial assets | | | | 36 | -| ------------------------------ | --- | --- | --- | --- | - -| Other current assets | | | | 42 | -| --------------------- | --- | --- | --- | ---- | -| | | | | | -| Total current assets | | | | 673 | - -| Long-term financial assets | | | | 12 | -| ----------------------------- | --- | --- | --- | --- | -| Property, plant and equipment | | | | 66 | -| | | | | | -| Right-of-use assets | | | | 108 | -| | | | | | -| Goodwill | | | | 27 | - -| Trademarks | | | | 757 | -| ------------------ | --- | --- | --- | --- | -| | | | | | -| Deferred tax asset | | | | 24 | - -| Other non-current financial assets | | | | 6 | -| ---------------------------------- | --- | --- | --- | --- | - -| Total non-current assets | | | | 1,000 | -| ------------------------- | --- | --- | --- | -------- | -| Total assets | | | | 1,673 | -Liabilities classified as held for sale € in millions June 30, 2021 -| | | | | | -| ----------------------- | --- | --- | --- | --- | -| Accounts payable | | | | 174 | -| Current lease liability | | | | 29 | - -| Other current provisions | | | | 60 | -| ------------------------ | --- | --- | --- | --- | - -| Other current accrued liabilities | | | | 112 | -| ---------------------------------------------- | --- | --- | --- | ---- | -| | | | | | -| Other current financial liabilities | | | | 12 | -| | | | | | -| Other current liabilities | | | | 18 | -| | | | | | -| Total current liabilities | | | | 405 | -| | | | | | -| Non-current lease liability | | | | 123 | -| Provision for pensions and similar obligations | | | | 6 | -| | | | | | -| Deferred tax liability | | | | 108 | -| | | | | | -| Other non-current provisions | | | | 18 | - -| Other non-current accrued liabilities | | | | 2 | -| ---------------------------------------- | --- | --- | --- | ---- | -| | | | | | -| Other non-current financial liablilities | | | | 4 | -| | | | | | -| Other non-current liablilities | | | | 2 | -| | | | | | -| Total non-current liabilities | | | | 263 | -| | | | | | -| Total liabilities | | | | 667 | - -At June 30, 2021, the company assessed the fair value of the disposal group on the basis of existing non- -binding purchase price offers at this point of time. There was no indication for any impairment of the -disposal group. -Further information on the allocation of goodwill is contained in these Notes. ► SEE NOTE 09 -| | | | | | -| --- | --- | --- | --- | --- | -21 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | | | 3 | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -05 SHAREHOLDERS’ EQUITY -During the period from January 1, 2021, to June 30, 2021, the nominal capital of adidas AG remained -unchanged. Consequently, on June 30, 2021, the nominal capital of adidas AG amounted to € 200,416,186 -divided into 200,416,186 registered no-par-value shares. -In the 2016 financial year, adidas AG introduced an employee stock purchase plan in favor of employees of -adidas AG and its affiliated companies. -adidas AG purchased shares in connection with this employee stock purchase plan in the first six months -of the 2021 financial year. More details on the repurchase of adidas AG shares and use of treasury shares -in connection with the employee stock purchase plan in the first six months of the 2021 financial year are -set out in the tables ‘Repurchase of adidas AG shares and use of treasury shares in the context of the -employee stock purchase plan2021’ and ‘Repurchase of adidas AG shares and use of treasury shares in -the context of the employee stock purchase plan 2021/matching shares’. -REPURCHASE OF ADIDAS AG SHARES AND USE OF TREASURY SHARES IN THE CONTEXT OF THE EMPLOYEE STOCK PURCHASE PLAN 2021 -Total price -| | | | | | in € | Average | | Amount | | Amount | | | -| --- | --- | --- | --- | ----------- | -------- | --------- | ------ | ---------- | --- | ---------- | --- | --- | -| | | | | (excluding | | purchase | | in the | | in the | | | -| | | | | incidental | | | price | nominal | | nominal | | | -Purchase Number of purchasing per share capital capital Issuance date -date shares costs) in € in € in % to employees -| January 8, | | | | | | | | | | | January 12, | | -| ----------- | --- | ------ | --- | ------------ | --- | ------ | --- | ------ | --- | ---- | ------------ | ----- | -| | | 23,652 | | 6,992,862.30 | | 295.66 | | 23,652 | | 0.01 | | | -| 2021 | | | | | | | | | | | | 2021 | -| April 9, | | | | | | | | | | | April 13, | | -| | | 24,032 | | 6,757,458.75 | | 281.19 | | 24,032 | | 0.01 | | | -| 2021 | | | | | | | | | | | | 2021 | -| | | | | | | | | | | | | | - -REPURCHASE OF ADIDAS AG SHARES AND USE OF TREASURY SHARES IN THE CONTEXT OF THE EMPLOYEE STOCK PURCHASE PLAN 2021/ MATCHING -SHARES -Total price -| | | | | | in € | Average | | Amount | | Amount | | | -| --- | --- | --- | --- | ----------- | -------- | --------- | ------ | ---------- | --- | ---------- | --- | --- | -| | | | | (excluding | | purchase | | in the | | in the | | | -| | | | | incidental | | | price | nominal | | nominal | | | -Purchase Number of purchasing per share capital capital Issuance date -date shares costs) in € in € in % to employees -| January 8, | | | | | | | | | | | January 12, | | -| ----------- | --- | ----- | --- | ------------ | --- | ------ | --- | ----- | --- | ----- | ------------ | ----- | -| | | 2,843 | | 840,550.80 | | 295.66 | | 2,843 | | 0.001 | | | -| 2021 | | | | | | | | | | | | 2021 | -| April 9, | | | | | | | | | | | April 13, | | -| | | 3,817 | | 1,073,286.45 | | 281.19 | | 3,817 | | 0.002 | | | -| 2021 | | | | | | | | | | | | 2021 | -| | | | | | | | | | | | | | - -On May 12, 2021, the Annual General Meeting of adidas AG approved the proposal of the Executive Board -and Supervisory Board on the appropriation of retained earnings for the 2020 financial year in the amount -of € 3.00 per share. The dividend was paid on May 18, 2021. Based on the number of dividend-entitled -shares at the time of the Annual General Meeting, this results in a dividend distribution of € 585,198,180. -On June 30, 2021 adidas AG held a total of 5,350,126 treasury shares, corresponding to a notional amount -of € 5,350,126 in the nominal capital and consequently 2.67% of the nominal capital. In accordance with -§ 71b German Stock Corporation Act (Aktiengesetz - AktG), the treasury shares held directly or indirectly -do not confer any rights to the company. -22 - -| | | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -06 FINANCIAL INSTRUMENTS -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | | | | Category | | | | June 30, 2021 | | | | | December 31, 2020 | | -| --- | --- | --- | --- | --- | --------- | --- | --- | --- | -------------- | --- | --- | --- | --- | ------------------ | --- | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Financial assets -Cash and cash equivalents -Cash and cash equivalen ts Amortized 2,383 – – – – 1,762 – – – – -| | | | | | cost | | | | | | | | | | | -| --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Fair value -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -through -| Cash equivalents | | | | | | 1,768 | 1,768 | – | 1,768 | – | 2,232 | 2,232 | | – 2,232 | – | -| ---------------- | --- | --- | --- | --- | --- | ----- | ----- | --- | ----- | --- | ----- | ----- | --- | ------- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Short-term financial assets | | | | | | | – – | – | – | – | | 0 | 0 | – | 0 – | -| --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Accounts receivable | | | | | cost | 2,324 | – | – | – | – | 1,952 | | – | – | – – | -| ------------------- | --- | --- | --- | --- | ----- | ----- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -Other current financ ial assets - -| Derivatives used in hedge | | | | | Hedge | | | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - accounting 119 119 – 119 – 163 163 – 163 – -accounting -| | | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -| Derivatives not used in hedge | | | | | through | | | | | | | | | | | -| | | | | | | | 32 32 | – | 32 | – | | 32 | 32 | – | 32 – | -| accounting | | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Promissory notes | | | | | | 147 | 147 | – | – | 147 | | 6 | 6 | – | – 6 | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Other financial assets | | | | | cost | 511 | – | – | – | – | 511 | | – | – | – – | -| ---------------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Long-term financial asse ts - -Fair value -| | | | | | through | | | | | | | | | | | -| --- | --- | --- | --- | --- | --------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Other equity investments profit or 89 89 – – 89 89 89 – – 89 -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Other equity investments | | | | | other | | 73 73 | – | – | 73 | | 80 | 80 | – | – 80 | -| ------------------------ | --- | --- | --- | --- | ------ | --- | ----- | --- | --- | --- | --- | --- | --- | --- | ---- | -compre- -hensive -| | | | | | income | | | | | | | | | | | -| --- | --- | --- | --- | --- | ------ | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Other investments | | | | | | | 26 25 | – | 251 | – | | 35 | 37 | – | 372 – | -| ----------------- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | ----- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Amortized -| Other investments | | | | | cost | | 87 – | – | – | – | 149 | | – | – | – – | -| ----------------- | --- | --- | --- | --- | ----- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -Loans Amort iz e d 0 0 – – – 0 0 – – – -| | | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other non -current financial assets -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | - -| Derivatives used in hedge | | | | | Hedge | | | | | | | | | | | -| -------------------------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -accounting accounting 18 18 – 18 – 14 14 – 14 – -| | | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -| Derivatives not used in hedge | | | | | through | | | | | | | | | | | -| | | | | | | 108 | 108 | – | 108 | – | | 99 | 99 | – | 99 – | -| accounting | | | | | profit or | | | | | | | | | | | -| | | | | | loss | | | | | | | | | | | - -Fair value -through -| Promissory notes | | | | | | | – – | – | – | – | 166 | 166 | | – | – 166 | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | -Fair value -through -| Earn-out components | | | | | | | – – | – | – | – | | 12 | 12 | – | – 12 | -| ------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | -profit or -loss -| | | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | --- | --- | --- | ----- | - -23 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | | | Category | | | | June 30, 2021 | | | | | December 31, 2020 | | -| --- | --- | --- | --- | --------- | --- | --- | --- | -------------- | --- | --- | --- | --- | ------------------ | --- | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -Amortized -| Other financial assets | | | | cost | 117 | – | – | – | – | 114 | – | – | – | – | -| ---------------------- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Assets classified as held for sale Amort iz e d 1,674 – – – – – – – – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Financial assets per level – – – 2,070 309 – – – 2,577 352 - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Financial liabilities -Short-term borrowings -Amortized -Bank borrowings cost 82 – – – – 87 – – – – - Amort iz e d 600 600 600 – – 599 605 605 – – -| Eurobond | | | | | | | | | | | | | | | -| -------- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | c o s t | | | | | | | | | | | -Accounts pay able Amort iz e d 2,165 – – – – 2,390 – – – – - -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Current accrued l iabilities Amort iz e d 801 – – – – 939 – – – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -C u r r e nt a c c r u e d li a bilities for Amort iz e d 801 – – – – 743 – – – – -| cu s t o m | e r d i s c o u n | t s | | c o s t | | | | | | | | | | | -| ---------- | ------------------ | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Other current financial liabilities -| | | | | | | | | | | | | | | | -| -------------------------- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| Derivatives used in hedge | | | | Hedge | | | | | | | | | | | - accounting 157 157 – 157 – 258 258 – 258 – -accounting -| | | | | Fair value | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | ----------- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -| Derivatives not used in hedge | | | | through | | | | | | | | | | | -| | | | | | | 26 26 | – | 26 | – | 24 | 24 | – | 24 | – | -| accounting | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Other financial liabilities | | | | cost | 158 | – | – | – | – | 164 | – | – | – | – | -| --------------------------- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Current lease liabilities n.a. 480 – – – – 563 – – – – - -Long-term borrowings -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Bank borrowings | | | | | | 91 – | – | – | – | 103 | – | – | – | – | -| --------------- | --- | --- | ----- | -------- | --- | ------ | --- | --- | --- | ---- | --- | --- | --- | --- | -| | | | | cost | | | | | | | | | | | -| | | | Amort | iz e d | | | | | | | | | | | -Eurobond 1,889 1,944 1,944 – – 1,888 1,987 1,987 – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | Amort | iz e d | | | | | | | | | | | -Convertibl e bond 493 628 628 – – 491 631 631 – – -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Non-current accr ued liabilities Amort iz e d 1 0 – – – 2 0 – – – - -| | | | | c o s t | | | | | | | | | | | -| --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -O th e r n o n-current financial – - -lia b il it ie s -D e r iv a t iv e s used in hedge H e d g e 5 5 – 5 – 26 26 – 26 – -n -| ac c | o u t in g | | acco | un t in g | | | | | | | | | | | -| ------------------------------- | ---------- | --- | ---- | ----------- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -| | | | | Fair value | | | | | | | | | | | -| Derivatives not used in hedge | | | | through | | | | | | | | | | | -| | | | | | | 34 34 | – | 34 | – | 85 | 85 | – | 85 | – | -| accounting | | | | profit or | | | | | | | | | | | -loss -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -Amortized -| Other financial liabilities | | | | cost | | – – | – | – | – | 4 | – | – | – | – | -| --------------------------- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Non-current lease liabili ties n.a. 2,206 2,206 – – – 2,159 – – – – - -Amortized -Liabilities classified as held for sale cost 667 – – – – – – – – – -Financial liabilities per level – – 3,172 222 – – – 3,2230 393 – - -Thereof: aggregated by category -according to IFRS 9 -Financial assets at fair value -| through profit or loss | | | | | 2,170 | | | | | 2,670 | | | | | -| ----------------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | -(FVTPL) -Thereof: designated as such -| upon initial recognition | | | | | | – | | | | – | | | | | -| -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(Fair Value Option – FVO) -| | | | | | | 87 | | | | 87 | | | | | -| --- | --- | --- | --- | --- | --- | --------- | --- | --- | ----- | ----- | ----- | --- | ----- | --- | -Thereof: held for trading (FAHfT) - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ----- | --- | ----- | --- | ----- | --- | -24 - -| | | | HALF YEAR REPORT 2021 | | | | | -| ---- | --- | --- | ---------------------- | --- | --- | --- | --- | - -| | 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -CARRYING AMOUNTS OF FINANCIAL INSTRUMENTS AND THEIR FAIR VALUES INCLUDING HIERARCHY ACCORDING TO IFRS 13 € IN MILLIONS -| | | Category | | June 30, 2021 | | | December 31, 2020 | -| --- | --- | ----------- | --- | -------------- | --- | --- | ------------------ | - Carrying Fair Carrying Fair - amount value Level 1 Level 2 Level 3 amount value Level 1 Level 2 Level 3 -Financial assets at fair value -| through other | | | 73 | | | 80 | | -| -------------- | --- | --- | --- | --- | --- | --- | --- | -comprehensive income (FVOCI) -Thereof: debt instruments – – - -Thereof: equity investments -(without recycling to 73 80 -profit and loss) -Financial assets at amortized cost -| | | | 7,096 | | | 4,489 | | -| --- | --- | --- | ----- | --- | --- | ----- | --- | -(AC) -Financial liabilities at fair value -thro ugh profit or loss -| | | | 119 | | | 108 | | -| --- | --- | --- | --- | --- | --- | --- | --- | -(FVTPL) -| | | | – | | | – | | -| --- | --- | --- | ------------ | --- | ------- | -------- | --------- | -Thereof: held for trading (FLHfT) - -Financial liabilities at amortized -cost (AC) 7,747 7,409 -| | | | | | | | | -| --- | --- | --- | ----------- | --- | ------- | ------- | --------- | - 1 Net gains in the amount of € 1 million and gains in the amount of € 0 million due to currenc y translation differences were recogn ized in equity i n 2021 -2 Net gains in the amount of € 0 million and losses in the amount of € 2 million due to currency translation differences were recognized in equity in 2020 - -Level 1 is based on quoted prices in active markets for identical assets or liabilities. -Level 2 is based on inputs other than quoted prices included within Level 1 that are observable for the assets or liabilities, either directly (i.e. as prices) or indirectly (i.e. derived from prices). -Level 3 is based on inputs for the assets or liabilities that are not based on observable market data (unobservable inputs). - -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -25 - -| | | | | | | HALF YEAR REPORT 2021 | | | | | | | | | -| ---- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | 1 | | | | 2 | | | | 3 | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -RECONCILIATION OF FAIR VALUE HIERARCHY LEVEL 3 IN 2020 € IN MILLIONS -| | | | | | | | | | Realized | | Unrealized | | | | -| --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --------- | --- | ----------- | -------------- | ----------- | --- | -| | | | | | Fair value | | | | | | | Currency | Fair value | | - Jan. 1,2020 Additions Disposals Gains Losses Gains Losses translation Dec. 31, 2020 -Investments in other equity instruments held -| | | | | | | 84 | – | – | – | – | 2 | – | – | 87 | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -for trading (FAHfT) -I n v e s t m ents in othe r equity instruments 2 – – – – – – – 2 -| ( F V T P | L ) | | | | | | | | | | | | | | -| --------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -I n v e s tm ents in other equity instruments 78 3 – – – – (2) – 79 -| ( F V O C | I) | | | | | | | | | | | | | | -| --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Promissory notes (FVTPL) 182 – (1) – (3) 9 – (15) 171 - -Earn-out components – assets (FVTPL) 45 – (41) – – 12 – (4) 12 -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | | | | -RECONCILIATION OF FAIR VALUE HIERARCHY LEVEL 3 IN 2021 € IN MILLIONS -| | | | | | | | | | Realized | | Unrealized | | | | -| --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --------- | --- | ----------- | -------------- | --- | --- | -| | | | | | Fair value | | | | | | | Currency | | | -Fair value - Jan. 1,2021 Additions Disposals Gains Losses Gains Losses translation June 30, 2021 -Investments in other equity instruments held -| | | | | | | 87 | – | – | – | – | – | – | – | 87 | -| --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -for trading (FAHfT) -| I n v e s t | m ents in othe r equity instruments | | | | | | | | | | | | | | -| ----------- | ------------------------------------ | --- | --- | --- | --- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | --- | -| | | | | | | 2 | – | – | – | – | – | – | – | 2 | -( F V T P L ) -I n v e s tm ents in other equity instruments -| | I) | | | | | 79 | 1 | (6) | – | – | – | (1) | – | 73 | -| --- | --- | --- | --- | --- | --- | ----- | ---- | ------ | ---- | ---- | ---- | ------ | ---- | --- | -( F V O C -Promissory notes (FVTPL) 171 – (20) – – – (9) 5 147 - -Earn-out components – assets (FVTPL) 12 – (21) 9 – – – – – -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -The valuation methods used in measuring Level 1, Level 2 and Level 3 fair values remain unchanged and -can be found in the Notes to the 2020 consolidated financial statements. -NET GAINS/(LOSSES) ON FINANCIAL INSTRUMENTS RECOGNIZED IN THE CONSOLIDATED INCOME STATEMENT € IN MILLIONS -| | | | | | | | | | | Period ending | | Year ending | | | -| --- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------- | ----- | --- | -| | | | | | | | | | | | | | | | -| | | | | | | | | | | June 30, 2021 | | Dec. 31, 2020 | | | -| | Financial assets classified at | | | | | | | | | | | | | | -| | | | | | | | | | | | (3) | | (114) | | -amortized cost (AC) -| | Financial assets at fair value through | | | | | | | | | | | | | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | 1 | 18 | | -profit or loss (FVTPL) -| | Thereof: designated as such upon | | | | | | | | | | | – | – | | -| --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ----- | ----- | --- | -initial recognition -| | Thereof: classified as held for trading | | | | | | | | | | | – | 2 | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | - -Equity instruments at fair value through -| | | | | | | | | | | | | – | – | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -profit or loss (FVTPL) -| | Equity instruments at fair value through | | | | | | | | | | | | | | -| --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | – | – | | -other comprehensive income (FVOCI) -| | Financial liabilities at amortized cost (AC) | | | | | | | | | | | 1 | 39 | | -| --- | -------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ----- | --- | - -Financial liabilities at fair value through -| | | | | | | | | | | | | – | – | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | --- | -profit or loss (FVTPL) -| | Thereof: designated as such upon | | | | | | | | | | | | | | -| --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | | | | | | | | | | | | – | – | | -initial recognition -| | Thereof: classified as held for trading | | | | | | | | | | | – | – | | -| --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------ | --- | - -| | | | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -26 - -| | | HALF YEAR REPORT 2021 | | | | | | -| ---- | --- | ---------------------- | --- | --- | --- | --- | --- | - -| 1 | | 2 | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -07 EARNINGS PER SHARE -Basic earnings per share are calculated by dividing the net income from continuing operations attributable -to shareholders by the weighted average number of shares outstanding during the year, excluding -ordinary shares purchased by adidas and held as treasury shares. If negative earnings per share are -reported, according to IAS 33.41 no anti-dilutive effect may be taken into account. -EARNINGS PER SHARE - Continuing operations Discontinued operations Total - First half year First half year First half year First half year First half year -First half year -| | | 2021 | 2020 | 2021 | 2020 | 2021 | 2020 | -| --- | --- | ------- | ------- | ------- | ------- | ------- | ----- | -Net income / (loss) from -| continuing operations | | 890 | (217) | – | – | – | – | -| ---------------------- | --- | ----- | ------- | --- | --- | --- | --- | -(€ in millions) -| Net income / (loss) attributable | | | | | | | | -| --------------------------------- | --- | --- | --- | --- | --- | --- | --- | -to non-control ling interests -| | | 7 | (27) | – | – | – | – | -| --- | --- | --- | ------ | --- | --- | --- | --- | -(€ in millions) -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable 883 (190) 72 (74) 955 (264) -to shareholde rs (€ in millions) -Weighted average number of shares 195,066,060 195,266,358 195,066,060 195,266,358 195,066,060 195,266,358 -Basic earnings per share (€) 4.52 (0.97) 0.37 (0.38) 4.89 (1.35) -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -Net income / (loss) attributable - to shareholders (€ in millions) 883 (190) 72 (74) 955 (264) -Net income / (loss) used to -| | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | -determine diluted earnings 883 (190) 72 (74) 955 (264) -per share (€ in millions) -Weighted average number of shares 195,066,060 195,266,358 195,066,060 195,266,358 195,066,060 195,266,358 -Dilutive effect of share-based -| | | 6,163 | – | – | – | 6,163 | – | -| --- | --- | -------- | ---- | ---- | ---- | -------- | --- | -payments -| Weighted average number of | | | | | | | | -| --------------------------- | --- | --- | --- | --- | --- | --- | --- | -shares for diluted earnings - 195,072,223 195,266,358 195,066,060 195,266,358 195,072,223 195,266,358 -per share -Diluted earnings per share (€) 4.52 (0.97) 0.37 (0.38) 4.89 (1.35) - -27 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -08 SEGMENTAL INFORMATION -adidas operates predominantly in one industry segment – the design, distribution and marketing of athletic -and sports lifestyle products. -As at June 30, 2021, following the company’s internal management reporting by markets and in -accordance with the definition of IFRS 8 ‘Operating Segments,’ five operating segments were identified: -EMEA, North America, Asia-Pacific, Greater China and Latin America. -In order to be able to successfully execute our new strategy ‘Own the Game’ for the period until 2025, -adidas has changed its organizational structure. Since January 1, 2021, adidas manages Greater China as -a separate market and integrated Russia/CIS and Emerging Markets into the European market. -The operating segment Reebok North America which was reported in the internal management reporting -until February 16, 2021, is not monitored anymore due to the reclassification of Reebok business as -discontinued operations on February 16th, 2021. Therefore the segment North America only includes the -business activities of adidas. Comparative segmental information has been retrospectively adjusted. Also -the Reebok business activities in the other operating segments have been removed in the segment -information for 2021 and 2020. -Each market comprises all wholesale, retail and e-commerce business activities relating to the -distribution and sale of products of the adidas brand to retail customers and end consumers. -Other Businesses includes the business activities of the Y-3 label and other subordinated businesses -which are not monitored separately by the chief operating decision maker. Also, certain centralized -corporate functions do not meet the definition of IFRS 8 for an operating segment. This includes, in -particular, functions such as Global Brands and Global Sales (central brand and distribution -management), central treasury, global sourcing as well as other headquarter functions. Assets, liabilities, -income and expenses relating to these corporate functions are presented in the reconciliations. -The chief operating decision-maker for adidas has been defined as the entire Executive Board of -adidas AG. -Net sales represents revenue from contracts with customers. There are no intersegment sales between -the reportable segments. Accounting and valuation policies applied for reporting segmental information -are the same as those used for adidas. -The results of the operating segments are reported in the line item ‘Segmental operating profit.’ This is -defined as gross profit minus other operating expenses plus royalty and commission income and other -operating income attributable to the segment or group of segments, however without considering -headquarter costs and central expenses for marketing. -Segmental assets include accounts receivable as well as inventories. Only these items are reported to the -chief operating decision-maker on a regular basis. -Segmental liabilities only contain accounts payable from operating activities as there are no other liability -items reported regularly to the chief operating decision-maker. -28 - - HALF YEAR REPORT 2021 - -| 1 | | 2 | | 3 | | | -| --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -SEGMENTAL INFORMATION € IN MILLIONS -| | | | Net sales | Segmental | Segmental | Segmental | -| --- | --- | --- | ---------- | ---------- | ---------- | ---------- | - (thir d parties)1 opera tin g profit1 assets2 li abilities2 - - 2021 2020 2021 2020 2021 2020 2021 2020 -EMEA 3,680 2,670 792 308 2,458 2,390 (186) (176) -North America 2,406 1,892 447 45 1,540 1,555 (51) (41) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Greater China 2,405 1,757 758 540 1,219 1,240 (230) (266) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Asia-Pacific 1,135 939 285 160 600 699 (43) (72) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Latin America 645 406 121 (1) 507 532 (91) (55) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Reportable segments 10,272 7,665 2,404 1,053 6,325 6,415 (601) (609) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | -Other Businesses 73 68 17 10 52 59 8 (2) -Total 10,345 7,733 2,420 1,063 6,377 6,474 (593) (611) - 1 First half year. -2 At June 30. - -OPERATING PROFIT € IN MILLIONS -First half First half -| | | | | | | | -| --- | --- | --- | --- | ------------ | ------------ | --- | -| | | | | year 2021 | year 2020 | | - -| Operating profit for reportable segments | | | | | 2,404 | 1,053 | -| ---------------------------------------- | --- | --- | --- | --- | -------- | ------ | -| Operating profit for Other Businesses | | | | | 17 | 10 | -| | | | | | | | -| Segmental operating profit | | | | | 2,420 | 1,063 | -| | | | | | | | -| HQ | | | | | (745) | (586) | - -| Central expenditure for marketing | | | | | (373) | (429) | -| --------------------------------- | --- | --- | --- | --- | --------- | --------- | -| Consolidation | | | | | (54) | (263) | -| | | | | | | | -| Operating profit/ (loss) | | | | | 1,248 | (215) | -| | | | | | | | -| Financial income | | | | | 6 | 17 | - -| Financial expenses | | | | | (75) | (82) | -| ------------------ | --- | --- | --- | --- | ------ | ---- | -| | | | | | | | -Income/ (loss) before taxes from continuing operations 1,179 (279) -| | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | - -NET SALES (THIRD PARTIES) € IN MILLIONS -First half First half -| | | | | | | | -| --- | --- | --- | --- | ------------ | ------------ | --- | -| | | | | year 2021 | year 2020 | | - -| Footwear | | | | | 5,766 | 4,495 | -| -------- | --- | --- | --- | --------- | ------- | ------------ | -| Apparel | | | | | 4,037 | 2,812 | -| | | | | | | | -| Hardware | | | | | 542 | 426 | -| | | | | | | | -| Total | | | | 10,345 | | 7,733 | - -| 09 | GOODWILL AND TRADEMARKS | | | | | | -| --- | ------------------------ | --- | --- | --- | --- | --- | -Following the company’s internal management reporting by markets, the number of cash-generating units -decreased to a total of six, effective January 1, 2021. In the first quarter 2021, the number of cash- -generating units further decreased to a total of five as the cash-generating unit North America Reebok is -classified as a disposal group and shown in ‘Assets/liabilities classified as held for sale.’ -Due to the changes in segmental reporting, the carrying amounts of acquired goodwill have been -reallocated to the new groups of cash-generating units in the first quarter 2021 as follows: -29 - - HALF YEAR REPORT 2021 - -| 1 | | | | 2 | | | | | 3 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) - -ALLOCATION OF GOODWILL -| | | | | | | | Goodwill (€ in millions) | | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | --- | - -| | | | | | | | | | (Re-) | | (Re-) | | -| --- | --- | --- | --- | --- | --- | ------------ | --- | ----------- | ------ | ----------- | ------ | --- | -| | | | | | | Aggregation | | allocation | | allocation | | | - Dec. 31, 2020 EMEA Asia-Pacific adidas Golf Jan. 1, 2021 -| Europe | | | | | 593 | | (593) | | | | | n.a. | -| ---------------- | --- | --- | --- | --- | ------ | --- | ----- | --- | ------- | --- | ------- | ------ | -| Emerging Markets | | | | | 76 | | (76) | | | | | n.a. | -| | | | | | | | | | | | | | -| EMEA | | | | | n.a. | | 669 | | | | 25 | 694 | -| | | | | | | | | | | | | | -| North America | | | | | n.a. | | | | | | 77 | 77 | -| | | | | | | | | | | | | | -| Greater China | | | | | n.a. | | | | 269 | | 10 | 280 | -| | | | | | | | | | | | | | -| Asia-Pacific | | | | | 361 | | | | (269) | | 66 | 157 | -| | | | | | | | | | | | | | -| adidas Golf | | | | | 178 | | | | | | (178) | n.a. | -| | | | | | | | | | | | | | -| Total | | | | | 1,208 | | - | | - | | - | 1,208 | -| | | | | | | | | | | | | | - -Due to the change in the composition of the company’s operating segments and associated cash- -generating units respectively, adidas assessed at January 1, 2021 whether goodwill or Reebok trademark -impairments were required. The underlying value drivers and key assumptions for impairment testing -purposes remained in principle unchanged compared to the impairment test performed for the -consolidated financial statements as at December 31, 2020. In this context, there was no need for goodwill -or Reebok trademark impairment. -On February 16, 2021, the company decided to begin a formal process aimed at divesting Reebok. Reebok -is reported as discontinued operations and classified as a disposal group held for sale as at June 30, 2021, -accordingly. As a result, the goodwill allocated to the group of cash generating units Europe, Middle East -and Africa (EMEA), North America, Greater China and Asia-Pacific (APAC) was split and reallocated -between adidas and Reebok cash-generating units based on relative values (fair values), respectively. -RECONCILIATION OF GOODWILL, NET € IN MILLIONS -| | | | | | | | North | | Greater | | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | --- | --- | --- | - EMEA America China Asia-Pacific Total -| January 1, 2021 | | | | | 694 | | 77 | | 280 | | 157 | 1,208 | -| --------------------- | --- | --- | --- | --- | ---- | --- | ----- | --- | ------ | --- | ------ | ------ | -| Reebok Disposal Group | | | | | (23) | | - | | (3) | | (1) | (27) | -Currency translation -| | | | | | 11 | | 0 | | 5 | | 2 | 18 | -| --- | --- | --- | --- | --- | --- | --- | ---- | --- | ---- | --- | ---- | --- | -differences -| June 30, 2021 | | | | | 682 | | 77 | | 282 | | 158 | 1,199 | -| -------------- | --- | --- | --- | --- | ----- | --- | ------ | --- | ------- | --- | ------- | ------- | - -In the first quarter 2021, the reallocated goodwill as well as the Reebok trademark in an amount of € 757 -million were initially measured according to IAS 36 ‘Impairment of Assets’ and goodwill allocated to -Reebok as well as the Reebok trademark were subsequently transferred to ’Assets/liabilities classified as -held for sale’ due to the concrete plans to divest Reebok. -| | | | | | | | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -30 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -010 EVENTS AFTER THE BALANCE SHEET DATE -The seven-year € 600 million Eurobond, originally due on October 8, 2021, was already repaid by adidas on -July 8, 2021. -With the approval of the Supervisory Board, the Executive Board of adidas AG has decided to launch a new -share buyback program in the second half of 2021. Starting on July 1, 2021, the company plans to buy back -shares worth up to € 550 million until the end of the year. Until the finalization of these interim -consolidated statements on July 26, 2021, adidas AG purchased a total of 226,850 shares for a total price -of € 72,683,346. -Besides this, there were no other major events between the end of the first half of 2021 and the finalization -of these interim consolidated financial statements on July 26, 2021, that might have a material influence -on the assets, liabilities, financial position, and profit or loss of the company. -Herzogenaurach, July 26, 2021 -The Executive Board of adidas AG -31 - -HALF YEAR REPORT 2021 -1 2 3 -AT A GLANCE INTERIM GROUP MANAGEMENT REPORT INTERIM CONSOLIDATED FINANCIAL STATEMENTS -(IFRS) -RESPONSIBILITY STATEMENT -To the best of our knowledge, and in accordance with the applicable reporting principles for interim -financial reporting, the interim consolidated financial statements give a true and fair view of the assets, -liabilities, financial position and profit or loss of the Group, and the interim Group management report -includes a fair review of the development and performance of the business and the position of the Group, -together with a description of the material opportunities and risks associated with the expected -development of the Group for the remaining months of the financial year. -Herzogenaurach, July 26, 2021 -KASPER RORSTED ROLAND AUSCHEL BRIAN GREVY -CHIEF EXECUTIVE OFFICER GLOBAL SALES GLOBAL BRANDS -HARM OHLMEYER AMANDA RAJKUMAR MARTIN SHANKLAND -CHIEF FINANCIAL OFFICER GLOBAL HUMAN RESOURCES GLOBAL OPERATIONS -32 - -CONTACT -ADIDAS AG -ADI-DASSLER-STR. 1 -91074 HERZOGENAURACH -GERMANY -Tel + 49 (0) 91 32 84 – 0 -► ADIDAS-GROUP.COM -adidas ist a member of DIRK -(German Investor Relations Association) -INVESTOR RELATIONS -investor.relations@adidas.com -► ADIDAS–GROUP.DE/INVESTORS -© ADIDAS 2021 -33 \ No newline at end of file +DUET: Joint Exploration of User–Item Profiles in Recommendation System +YueChen1,*,YifeiSun1,*,LuWang2,†,FangkaiYang2,PuZhao2,MinjieHong3,YifeiDong4,MinghuaHe2, +NanHu2,JianjinZhang2,ZhiweiDai2,YuefengZhan2,WeihaoHan2,HaoSun2, +QingweiLin2,WeiweiDeng2,FengSun2,QiZhang2,SaravanRajmohan2,DongmeiZhang2 +1PekingUniversity2Microsoft3ZhejiangUniversity4KTHRoyalInstituteofTechnology +*Equalcontribution†Correspondingauthor +Abstract +| | | | User History: | Item History: | User History: | Item History: | +| --- | --- | --- | -------------------- | ------------------------- | -------------------- | ------------------------- | +| | | | • Misfits (Punk) | • Review: | • Misfits (Punk) | • Review: | +| | | | • I r o n M a iden | " P o l i s he d P o p " | • I r o n M a iden | " P o l i s he d P o p " | +Traditionalrecommendationsystemsrepresent ( M e t al ) • R e v i e w : " G r e a t ( M e t al ) • R e v i e w : " G r e a t +6202 rpA 51 ]RI.sc[ 1v10831.4062:viXra • • Parliament Funk Bass" +| usersanditemsasdensevectorsandlearnto | | | Parliament | Funk Bass" | | | +| ------------------------------------- | --- | --- | ----------- | ------------------ | ------ | ------------------ | +| | | | (Funk) | • Meta: "80s Rock" | (Funk) | • Meta: "80s Rock" | +aligntheminasharedlatentspaceforrelevance +| | | | | Duet | Independent | | +| --- | --- | --- | --- | ----- | ------------ | --- | +estimation. RecentLLM-basedrecommenders (Joint Alignment) Generation +| instead leverage | natural-language | representa- | | | | | +| ---------------- | ---------------- | ----------- | ------------- | ------------- | ------------- | ------------- | +| | | | User Profile: | Item History: | User Profile: | Item History: | +tionsthatareeasiertointerpretandintegrate Enthusiast of A peak-era FUNK- Fan of HEAVY A highly rated POP- +| | | | FUNK and SOUL. | ROCK album with | METAL and | ROCK album with | +| --- | --- | --- | --------------- | ---------------- | ---------- | ---------------- | +withdownstreamreasoningmodules. Thispa- Values technical genre-blurring PUNK. Prefers polished +| | | | mastery | creativity. | dark themes. | production. | +| --- | --- | --- | ------- | ----------- | ------------ | ----------- | +textual +per studies how to construct effective Semantic Alignment ✘ Semantic Mismatch +profilesforusersanditems,andhowtoalign +themforrecommendation. Acentraldifficulty Figure1: DUETalignsrawuseranditemdatabytrans- +| | | | forming them | into textual | profiles within | a shared se- | +| --- | --- | --- | ------------ | ------------ | --------------- | ------------ | +isthatthebestprofileformatisnotknownapri- +manticspace. +ori: manuallydesignedtemplatescanbebrittle +| andmisalignedwithtaskobjectives. | | Moreover, | | | | | +| -------------------------------- | --- | --------- | ------------ | ------------ | -------------------- | --- | +| | | | to introduce | semantically | rich, human-readable | | +generatinguseranditemprofilesindependently +mayproducedescriptionsthatareindividually representationsforrecommendation(Wangetal., +plausible yet semantically inconsistent for a 2025;Zhang,2024;Baoetal.,2023;Hongetal., +| specificuser–itempair. | WeproposeDUET,an | | | | | | +| ---------------------- | ---------------- | --- | -------------------------- | --- | ------------------- | --- | +| | | | 2025a,b,c;Wangetal.,2024). | | Anaturaldirectionis | | +interaction-awareprofilegeneratorthatjointly +toreplacelatentvectorswithtextualuseranditem +producesuseranditemprofilesconditionedon +profilesthatcanbeinspected,edited,andreusedby +| both user history | and item evidence. | DUET | | | | | +| ----------------- | ------------------ | ---- | --- | --- | --- | --- | +downstreamcomponentsincorporatedwithLLMs. +| follows a three-stage | procedure: | it first turns | | | | | +| --------------------- | ---------- | -------------- | --- | --- | --- | --- | +However,existingLLM-basedapproachesremain +rawhistoriesandmetadataintocompactcues, +then expands these cues into paired profile limited in two important ways. First, directly +promptsandthengenerateprofiles,andfinally promptinganLLMwithrawuseranditemhisto- +optimizesthegenerationpolicywithreinforce- +riestoobtainrecommendationsoftenyieldsnoisy +mentlearningusingdownstreamrecommenda- andincompletesignals,especiallywhenhistories +| tionperformanceasfeedback. | | Experimentson | | | | | +| -------------------------- | --- | ------------- | --------- | ------------------------ | --- | ------------- | +| | | | are long, | sparse, or heterogeneous | | (Wang et al., | +threereal-worlddatasetsshowthatDUETcon- +| | | | 2025). Second, | profile-based | methods | typically | +| --- | --- | --- | -------------- | ------------- | ------- | --------- | +sistentlyoutperformsstrongbaselines,demon- +relyonmanuallydesignedtemplatesorhandcrafted +stratingthebenefitsoftemplate-freeprofileex- +attributes,whichrequiressubstantialhumanengi- +plorationandjointuser–itemtextualalignment. +Project code page: https://github.com/duet- neering and constrains the representation space. +review/duet_code. More fundamentally, many approaches generate +userprofilesanditemprofilesindependently,with- +1 Introduction +| | | | out modeling | how user | preferences | and item se- | +| --- | --- | --- | ------------ | -------- | ----------- | ------------ | +Traditional recommendation systems represent manticsinteractatdecisiontime(Yangetal.,2023; +Xietal.,2024). +usersanditemsasdensevectorsandlearntoalign +theminasharedlatentspaceforrelevanceestima- Toaddressthesechallenges,wepropose DUET, +tion (Covington et al., 2016; Wu, 2023). While ajointuser–itemprofilegeneratorthattakesboth +effective,suchembeddingsareopaque: theyoffer userhistoryanditemhistoryasinputandproduces +limitedinterpretabilityandmakeitdifficulttoan- a paired set of profiles for the interaction. Cru- +alyzewhyanitemisrecommended. Recentwork cially, DUET doesnotrequireprofiletemplates: it +thereforeleverageslargelanguagemodels(LLMs) istrainedwithreinforcementlearningusingfeed- + +back from downstream recommendation perfor- asseeninworkslikeCRESDUP(Chenetal.,2007) +mance,enablingittoexploreanddiscovereffective andUPCSim(Widiyaningtyasetal.,2021). While +profileformatsautomatically. Figure1illustrates foundational,thesemethodswerelimitedbytheir +whyjointprofilingmatters. Theuserhaslistened rigid,hand-engineeredfeatures. Morerecently,the +toMisfits(punk),IronMaiden(metal),andParlia- adventofLargeLanguageModels(LLMs)hasen- +ment(funk),whilethecandidatealbumisdescribed abledashifttowardsgeneratingprofilesinnatural +byreviewssuchas“polishedpop”and“greatfunk language. Studies such as KAR (Xi et al., 2024), +bass”andametatag“’80srock.” Consideringboth GPG(Zhang,2024),PALR(Yangetal.,2023),and +sides together, DUET can reconcile these signals LettinGo (Wang et al., 2025) leverage LLMs to +intoacompatibleinterpretation(e.g.,highlighting createtextualuserprofilesfrombehavioraldata. A +the user’s funk/soul affinity and the item’s funk- keylimitationoftheseapproachesistwofold: they +rockcharacter). Incontrast,independentlygener- relyonstaticorpre-definedtemplatesforprofile +atedprofilesmayamplifydifferentfacets,summa- generation,andtheyoftenfocusexclusivelyonuser +rizingtheuseras“heavymetal/punk”whilesum- profiles,neglectingtherich,expressiveinformation +marizingtheitemas“pop-rock”,whichresultingin inherentinitemsandthecomplexdynamicsofuser- +asemanticallymismatchedpairthatobscuresthe iteminteractions. Incontrast,ourworkintroduces +truerelevancesignal. anewparadigmwherebothuseranditemprofiles +DUETproceedsinthreestages. First,rawhisto- arenotfixedbutaredynamicallyexploredandop- +ries and metadata are distilled into minimal cues timizedinasharedsemanticspacetodirectlyalign +thatcapturecompactbutinformativesignals. Sec- withrecommendationperformance. +ond, the model expands cues into richer profile +prompts to generate textual profiles, allowing ex- 2.2 ReinforcementLearningforLLM-Based +plorationoveralternativeprofilestructuresandem- RecommendationSystems +phases. Third,theresultingprofilesareconsumed +ReinforcementLearning(RL),particularlythrough +bydownstreamrecommenders,andtheirtaskfeed- +techniqueslikeRLHF,hasbecomeacoremethod +backisusedtooptimizetheprofilegenerationpol- +for aligning LLMs with specific objectives. This +icy. By coupling both sides in a shared semantic +approach has been adapted for recommendation +space,DUETlearnsuserprofilesthatreflectwhat +tasks (Wang et al., 2025; Lin et al., 2025; Deng +kindsofitemsauserprefersanditemprofilesthat +etal.,2025),butexistingeffortsoftenfacekeylimi- +reflectwhatkindsofusersanitemappealsto. +tations. Theyfrequentlyrelyonofflinerewardmod- +Ourcontributionsareasfollows: +els(Jeongetal.,2023;Chenetal.,2025;Heetal., +• Werepresentusersanditemsasnatural-language 2025; Liu et al., 2025) that do not adapt in real- +profiles and align them in a shared semantic timetosystemfeedback,asetupthatrisksissues +space,extendingtheclassicvector-basedalign- like reward hacking (Skalse et al., 2025). Other +mentprincipletointerpretabletextualrepresenta- methods(Sunetal.,2024;Luetal.,2024)restrict +tions. themselvestoofflinepreferencetuning(e.g.,DPO), +• We introduce an exploration-based framework which can easily overfit on static datasets. Our +thatstartsfromcue-basedinitialization,expands framework, DUET,overcomesthesechallengesby +cuesintocandidateprofiles,andjointlyoptimizes integrating RL into a closed-loop system where +user and item profiles with downstream recom- downstreamrecommendationperformanceserves +mendationfeedbackviareinforcementlearning, asthereal-timerewardsignal,allowingforthedy- +avoidingrigidtemplates. namicandinteractiverefinementoftextualprofiles. +• Extensiveexperimentsacrossmultiplereal-world +datasets show that DUET consistently outper- 3 Method +formsstrongbaselines,validatingbothjointpro- +filingandfeedback-drivenprofileoptimization. DUET is a closed-loop framework that trans- +forms raw user–item interaction histories into +2 RelatedWork performance-aligned textual profiles through +learnedrepresentationstrategies. AsshowninFig- +2.1 ProfilesinRecommendation +ure 2, DUET consists of three stages. (1) Cue- +Earlyrecommendationsystemsprimarilyreliedon Based Initialization distills interaction histories +pre-definedprofilesbasedonstructuredattributes, intoconciseevidence-basedcues. (2)JointExplo- + +CUE-BASED INITIALIZATION JOINT EXPLORATION ON-POLICY OPTIMIZATION +| | | | | Item | DU | | | Duet | | | | +| ----------- | --- | --- | --- | ---- | --- | --- | --- | ---- | --- | --- | --- | +| ItemHistory | | | | | ET | | | | | | | +Initial CUE +Rated: Baba Is You (5★) +User: “Prefers retro puzzle games” +Reviewed: “I love logic- +Item: “Retro-style indie puzzle game” +| | | | based puzzle games.” | | | Profile Prompt | | | | | | +| --- | --- | --- | -------------------- | --- | --- | -------------- | --- | --- | --- | --- | --- | +Rated: The Witness (5★) +User: “Describe the user's typical +| | | | Rated: Fez (4★) | | puzzle-game preferences … | | | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒1 𝑃𝑟𝑜𝑓𝑖𝑙𝑒2 | … | 𝑃𝑟𝑜𝑓𝑖𝑙𝑒𝑛 | | +| --- | --- | --- | --------------- | --- | ------------------------- | --- | --- | ----------------- | --- | -------- | --- | +Item: “… item's core mechanics… +| | | | | | | | Profile | Downstream Task Env | | | | +| ----------- | --- | --- | ----------------------- | ---- | --------------------------------- | --- | ------- | --------------------- | --- | --- | --- | +| | | | | User | User: “A player who enjoys logic- | | | | | | | +| UserHistory | | | | | intensive indie puzzle games, | | | | | | | +| | | | Category: Indie Puzzle | | | | | Ranking correctness | | | | +prefers retro visual aesthetics…” +| | | | Game | | | | | &format Reward | | | | +| --- | --- | --- | ---- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Reviews: +Item: “A retro-style indie puzzle +| | | | “Challenging logic | | game featuring challenging | | | Group Computation | | | | +| --- | --- | --- | -------------------- | ------------------------ | ------------------------------- | --- | --- | ------------------- | --- | ---- | --- | +| | | | mechanics.” | | logic mechanics that appeal to | | | | | | | +| | | | “ R e t r o p ix | e l- a rt st y l e . ” | hardcore puzzle players.” | | | 𝐴𝑑𝑣1 𝐴𝑑𝑣2 | | | | +| | | | “ H a r d b u | t re w a r d i n g . ” | | | | | … | 𝐴𝑑𝑣𝑛 | | +| | | | | Figure2: | OverviewoftheDUETframework. | | | | | | | +ration via Adaptive Profile Prompt Discovery 3.2 Cue-BasedInitialization +| jointly | explores | user-item’s | | profile prompts | that | | | | | | | +| ---------- | -------- | ----------- | -------- | --------------- | -------- | -------- | --------- | -------- | -------- | --- | --------- | +| | | | | | | Raw user | histories | and item | metadata | | are often | +| define how | user | and item | profiles | should | be writ- | | | | | | | +noisy,redundant,andnotdirectlysuitableforpro- +ten. (3)OptimizationviaOn-policyExploration +| | | | | | | fileconstruction. | | Toaddressthis,DUETintroduces | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ---------------------------- | --- | --- | --- | +jointlyoptimizesuseranditemprofilesunderdown- +theconceptofcues: +concisehypothesesthatsum- +streamrecommendationfeedback. +| | | | | | | marize minimal | | but informative | | aspects | of users | +| ------------------- | ------ | --- | ------------------------- | ------- | -------- | -------------- | ------------ | --------------- | -------------- | ------- | -------- | +| All three | stages | are | realized | through | a single | | | | | | | +| | | | | | | and items. | These | cues act | as lightweight | | seeds, | +| passinputandoutput: | | | cueextraction,self-prompt | | | | | | | | | +| | | | | | | which are | deliberately | underspecified | | so | that the | +construction,andprofilegenerationareproduced +systemcansubsequentlyexplorericherprofilefor- +inasinglesequence-to-sequencegenerationpass +mats. +atinferencetime,enablingefficientdeployment. +| | | | | | | Definition1(Cue). | | Acueisaminimaltextualhy- | | | | +| --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------------------ | --- | --- | --- | +3.1 ProblemFormulation +pothesisderivedfromhistoricaldatathathighlights +WeformulateprofilegenerationinDUETasanon- one potential aspect of a user’s preference or an +policyreinforcementlearningproblem,motivated item’scharacteristic. Ratherthanaimingforcom- +bytheabsenceofanytextualgroundtruthdefining pleteness, cues capture partial but salient signals +an optimal user or item profile. Profile quality is thatserveasstartingpointsforprofileexploration. +evaluatedsolelybyitsfunctionalutilityinafixed +| | | | | | | To extract | cues | automatically, | | the | LLM is | +| --- | --- | --- | --- | --- | --- | ---------- | ---- | -------------- | --- | --- | ------ | +recommendationenvironment. +| | | | | | | prompted | to summarize | minimal | | but informative | | +| ---- | ---------- | --- | ------------ | ------ | ------- | ------------------------ | ------------ | ------- | ----------------- | --------------- | --- | +| DUET | is modeled | as | a generative | policy | π θ in- | | | | | | | +| | | | | | | aspectsofuseroritemdata. | | | Forexample,givena | | | +teractingwithafrozendownstreamrecommender. +user’sinteractionhistory,themodelisguidedwith +| For each | user–item | pair, | the | state is | defined as | | | | | | | +| -------- | --------- | ----- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | +instructionssuchas: +| s = {H | ,H },whereH | | andH | denotetheuser | | | | | | | | +| ------ | ----------- | --- | ---- | ------------- | --- | --- | --- | --- | --- | --- | --- | +| | u i | | u | i | | | | | | | | +and item interaction histories. An action corre- CueExtractionPrompt +spondstoasingle-passjointgeneration +“Fromthehistorybelow,analyzetheuser’s +a = {C u ,S u ,P u ,C i ,S i ,P i }, historicalinteractionstounderstandprefer- +ences,ratingbehavior,reviewsentimentor +| where | (C ,C | ) are | cues | distilled | from his- | | | | | | | +| ----- | ----- | ----- | ---- | --------- | --------- | --- | --- | --- | --- | --- | --- | +u i +| | | | | | | anyotherdimension. | | Keepthedescription | | | | +| -------- | ------------ | ----------- | --- | ----------- | ------- | ------------------ | --- | ------------------ | --- | --- | --- | +| tory, (S | u ,S i ) are | constructed | | user-item’s | profile | | | | | | | +prompts,and(P ,P )arethefinaltextualprofiles. conciseandavoidfullsentences.” +u i +| Thepolicyπ | | (a | s)definesajointdistribution | | | | | | | | | | +| ---------- | --- | -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +θ +over the entire generation sequence. As an on- This lightweight guidance allows the LLM to +policy agent, DUET is optimized using rewards maprawhistoriesandmetadataintocompacttex- +from its own sampled generations rather than by tualcues. Thedetaildexampleofcuecanbefound +| imitatingfixedsummaries. | | | | | | inAppendixA.3. | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | + +| 3.3 | JointExplorationviaAdaptiveProfile | | | | | | Single-PassInput | | | | | | +| --- | ---------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +PromptDiscovery Task:Generatestructuredprofilesfortheuseranditemfrom +historicaldata. +Aftercueextraction,DUETdoesnotdirectlysum- Combined Context: [User History H u ] + [Item +| | | | | | | | History | H i ]+[Avg | Ratings] | | | | +| ------------------------------------ | --- | --- | --- | --- | --- | ---------- | ------- | ---------- | -------- | --- | --- | --- | +| marizeuserinterestsoritemattributes. | | | | | | Instead,it | | | | | | | +“First,extractaconciseCueforbothuseranditem.Second, +| explores | | the space | of profile | construction | | strate- | | | | | | | +| -------- | --- | --------- | ---------- | ------------ | --- | ------- | --- | --- | --- | --- | --- | --- | +basedonthecues,constructaProfilePromptthatdefines +gies—natural-languagepromptsthatdefinethefor- thedescriptionlogic.Finally,generatetheProfileguidedby +theconstructedProfilePrompt.” +mat,abstractionlevel,andattributeselectionlogic +↓DUETForwardPass↓ +| usedtogenerateprofiles. | | | | Explorationistherefore | | | | | | | | | +| -------------------------------------------- | --- | --- | --- | ---------------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| performedoverhowprofilesshouldbeconstructed, | | | | | | | Single-PassOutput | | | | | | +User +ratherthanoversuperficialtextualparaphrases. +Cue:“prefersnostalgiclogic-basedchallenges” +For each user–item pair, DUET intro- Profileprompt: “Focuson1990svisualaestheticsand +| duces | an | explicit | intermediate | | variable, | the | strategicdepth.” | | | | | | +| ----- | --- | -------- | ------------ | --- | --------- | --- | ---------------- | --------- | --- | ----- | ----------- | ------------ | +| | | | | | | | Profile: | “A player | who | seeks | retro-style | visual charm | +constructed_prompt,whichservesasadiscrete +pairedwithdeepstrategicreasoning.” +| and | interpretable | | profile | prompt | S. | This profile | | | | | | | +| --- | ------------- | --- | ------- | ------ | --- | ------------ | --- | --- | --- | --- | --- | --- | +Item +promptisashortnatural-languageinstruction(e.g., +Cue:“retro-styleindiepuzzlewithhighdifficulty” +“Describetheuser’stypicalgamingpreferencesand +| | | | | | | | Profile | prompt: | “Describe | | pixel-art | graphics and | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | --------- | --- | --------- | ------------ | +engagementpatterns”)thatspecifieswhataspects intellectualdifficultytomatchlogicpreference.” +Profile:“A2Dexperiencefeaturingpixelatednostalgia +| to | describe | and | how to | organize | them, | without | | | | | | | +| --- | -------- | --- | ------ | -------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +andchallengingmechanicsthatdemandlogicaldeduction.” +| containingtheprofilecontentitself. | | | | | | Conditioned | | | | | | | +| ---------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +onS,themodelgeneratesthefinaluseranditem Figure3: Single-passgenerationinDUET: cueextrac- +| profiles. | | | | | | | tion,profileprompt(constructedprompt),andprofile | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | --- | --- | --- | +generationareproducedinonepassforbothuserand +| Exploration | | | is driven | by treating | | the profile | | | | | | | +| ----------- | --- | --- | --------- | ----------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +item. +| promptS | | asastochasticactionsampledfromthe | | | | | | | | | | | +| ------- | --- | --------------------------------- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| policyπ | | (S | Cue). | Duringtraining,theagentsam- | | | | | | | | | | +θ +plesdifferentprofilepromptinstructions,executes +| | | | | | | | andintellectualdifficulty”tomatch. | | | | | RLreinforces | +| ---------------------------- | --- | --- | --- | --- | ---------------- | --- | ---------------------------------- | --- | --- | --- | --- | ------------ | +| themtogenerateprofilepairs(P | | | | | ,P ),andreceives | | | | | | | | +u i +thissharedsemanticdirection,suppressingirrele- +| a reward | | based | on downstream | | recommendation | | | | | | | | +| --------- | --- | ----------------------------------- | ------------- | --- | -------------- | --- | ------------ | ------------- | -------- | --- | ----- | ---------------- | +| | | | | | | | vant signals | and | forcing | the | final | profiles to con- | +| accuracy. | | Thepolicyisoptimizedtoreinforcepro- | | | | | | | | | | | +| | | | | | | | verge | into a shared | semantic | | space | of nostalgia | +filepromptsthatconsistentlyyieldhigherrewards, +andlogic,significantlyimprovingrecommendation +enablingthemodeltoactivelysearchforeffective +accuracy. +profileconstructionformats. +Thisprocessisrealizedasaunifiedgeneration +3.4 OptimizationviaOn-policyExploration +pass, +| | | | | | | | On-policy | optimization. | | | We train | the profile | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------- | --- | --- | -------- | ----------- | +O = [ Cue → profilepromptS → Profile], generator π in an on-policy manner against a +| | (cid:124)(cid:123)(cid:122)(cid:125) | | | | | (cid:124) (cid:123)(cid:122) (cid:125) | | θ | | | | | +| --- | ------------------------------------ | --- | --------- | ------------------ | --------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | +| | | | (cid:124) | (cid:123)(cid:122) | (cid:125) | | | | | | | | +Context ProfileConstructionPrompt Execution frozendownstreammodelf, whichservesasthe +| | | | | | | (1) | environment | critic. | For | each | sampled | user–item | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | --- | ---- | ------- | --------- | +wheretheprofilepromptS parameterizesthepro- pair (u,i), the policy generates (P ,P ) and re- +u i +filegenerator. Atinferencetime,DUETcollapses ceivesascalarrewardbasedonthepredictionac- +thelearnedprofilepromptdistributionandgreed- curacy of f(P ,P ). The policy parameters are +u i +ilyexecutestheoptimalprofilepromptinasingle updatedtoreinforcegenerationsthatleadtolower +| forwardpass,introducingnoadditionallatency. | | | | | | | predictionerror. | | | | | | +| ------------------------------------------- | --- | ---------- | ------ | ------------ | -------- | ------- | --------------------------- | --- | --- | --- | --- | ------------- | +| Example | | of profile | prompt | Exploration. | | As | | | | | | | +| | | | | | | | Continuousfractionalreward. | | | | | Usingdiscrete | +| illustrated | | in Figure | 3, | DUET | resolves | this by | | | | | | | +integerratingsasrewardsleadstosparseandunsta- +| identifying | | a shared | context | in | a single | forward | | | | | | | +| ----------- | --- | -------- | ------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +blefeedback,asnear-misspredictions(e.g.,predict- +pass. ThepolicyextractsacoordinatedCue(e.g., +ing4foragroundtruth5)receivethesamepenalty +“prefersretropuzzlegames”)andconstructsapro- +| | | | | | | | assevereerrors. | | Toprovidedenseandinformative | | | | +| ---- | ------ | ---- | ------------------- | --- | --- | -------- | --------------- | --- | ---------------------------- | --- | --- | --- | +| file | prompt | (the | constructed_prompt) | | | that in- | | | | | | | +feedback,wedefineacontinuousfractionalreward: +structsthemodeltohighlight“nostalgicaesthetics +| and | strategic | logic” | for | the user, | while | simulta- | | | | | | | +| ------- | ---------- | ------ | ---------- | ---------- | ----- | -------- | --- | ------- | --- | ---- | ------ | -------- | +| | | | | | | | | | | | |y −yˆ | | | +| | | | | | | | | R (u,i) | | = 1− | ui | ui , (2) | +| neously | describing | | the item’s | “pixel-art | | graphics | | perf | | | | | +M + +where yˆ ui = f(P u ,P i ) is the predicted relevance ingsettingbasedondownstreampredictedrating +score and M is the maximum rating gap (e.g., scores. For each observed user–item interaction, +M = 4 for a 1–5 scale). This reward provides we randomly sample nine items that the user has +fine-grainedgradientsthatencourageincremental not interacted with to form a candidate set of ten +improvementsinrecommendationaccuracy. items, which are ranked in descending order ac- +We adopt Group Relative Policy Optimization cordingtotheirpredictedratings. Theground-truth +(GRPO) (DeepSeek-AI et al., 2025) to optimize interacted item is treated as the only positive in- +theprofilegeneratorundertheabovereward. The stance,andweadoptNDCG@Kastheevaluation +downstreamrecommenderf isfrozenthroughout metric, with K set to 1, 5, and 10, to assess how +training, which yields a stable policy optimiza- effectivelythelearnedprofilessupportcorrectitem +| tionsettingandpreventsrewarddriftorfeedback- | | | | ranking. | | | | | +| -------------------------------------------- | --- | --- | --- | --------- | ---------- | ---------- | ---- | ------- | +| | | | | Baselines | We compare | our method | with | several | +inducedrepresentationcollapse. +| | | | | representative | baselines. | 10H | directly | uses the | +| --- | --- | --- | --- | -------------- | ---------- | --- | -------- | -------- | +4 Experiment mostrecentinteractionhistoriesforpredictionwith- +| | | | | outconstructingexplicitprofiles. | | | KAR(Xietal., | | +| --- | --- | --- | --- | -------------------------------- | --- | --- | ------------ | --- | +4.1 ExperimentalSettings +2024)augmentsrecommendationmodelswithex- +Datasets. Experiments were conducted on three ternalreasoningknowledgeaboutuserpreferences +widelyusedreal-worlddatasets. AmazonMusic andfactualknowledgeaboutitemsextractedfrom +(Music) and Amazon Book (Book) are derived LLMs,whicharetransformedintotask-compatible +fromtheAmazonProductdataset1,whileYelpis representations. RLMRec(Renetal.,2024)lever- +fromtheYelpOpendataset2. Alldatasetsinclude agesLLMstolearnsemanticuseranditemrepre- +userreviews,ratings,andrichtextualinformation. sentationsfromtextualsignalsandalignsthemwith +WeusedthefullAmazonMusicdataset,butonly collaborativerelationalinformationthroughcross- +subsets (latest two months for Book and six for viewrepresentationlearning. PALR(Yangetal., +Yelp). Data was split by timestamp into training, 2023)fine-tunesalargelanguagemodelasarank- +validation,andtestsetstopreventinformationleak- ing component that selects preferred items from +age(Jietal.,2023). retrievedcandidatesexpressedinnaturallanguage. +EvaluationMetrics LG (LettinGo) (Wang et al., 2025) explores di- +For each observed user–item interaction (i.e., verseuserprofilecandidateswithLLMsandaligns +a review record), we construct the evaluation profilegenerationwithdownstreamrecommenda- +instance based on the user’s interaction history tionperformanceviapreferenceoptimization. Rea- +strictly prior to the corresponding timestamp. son4Rec(Fangetal.,2025)introducesadelibera- +tiverecommendationframeworkthatincorporates +Specifically,theuser’srecenthistoricalinteractions +beforethecurrentinteractionareusedtogenerate explicitstep-wisereasoningoveruserpreferences +auserprofile,whilehistoricalreviewsfromother toguideratingprediction. +| users are used | to generate | an item profile. | The | | | | | | +| -------------- | ----------- | ---------------- | --- | --- | --- | --- | --- | --- | +4.2 MainResults +downstreamrecommendationsystemthenpredicts +theratingofthetargetitemconditionedontwoin- Table 1 presents a comparison of our proposed +puts: thegenerateduserprofile,andthegenerated method against five baselines on three datasets: +| | | | | Amazon | Music, Amazon | Books, | and | Yelp. We | +| ------------ | ------------------------------- | --- | --- | ------ | ------------- | ------ | --- | -------- | +| itemprofile. | Anexceptionisthe10Hbaseline,for | | | | | | | | +which the downstream model directly consumes use Qwen3-8B (Team, 2025) and LLaMA3-8B +therawrecentinteractionhistories. (Dubey et al., 2024) as both the profile generator +andthedownstreamrecommendationmodel,with +| We evaluate | performance | using four | widely | | | | | | +| --------------- | ----------- | ------------------ | ------- | ------------ | ----------- | ----- | ----------- | ------- | +| | | | | a prediction | temperature | of 0. | In addition | to rat- | +| adopted metrics | (Wang | et al., 2025; Fang | et al., | | | | | | +2025): MeanAbsoluteError(MAE),RootMean ingpredictionresults,wealsoreportranking-based +SquareError(RMSE),Accuracy,andF1score. evaluation results in Table ??, which assess the +effectivenessofthelearnedprofilesfromadown- +Inadditiontoratingprediction,wefurtherevaluate +thegenerateduseranditemprofilesunderarank- streamrankingperspective. +| | | | | Overallsuperiorityoverstrongbaselines. | | | | As | +| --- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | +1https://cseweb.ucsd.edu/~jmcauley/datasets/ shown in Table 1, our method consistently out- +amazon/links.html +| | | | | performs | the strongest | baselines | across | all three | +| --- | --- | --- | --- | -------- | ------------- | --------- | ------ | --------- | +2https://business.yelp.com/data/resources/ +| open-dataset/ | | | | datasetsunderbothQwen3-8BandLLaMA3-8B. | | | | | +| ------------- | --- | --- | --- | -------------------------------------- | --- | --- | --- | --- | + +Method Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) MAE RMSE Acc F1(%) +(%) (%) (%) +Qwen3(8B) +10H 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +KAR(Xietal.,2024) 0.7396 1.2184 55.34 48.67 0.7483 1.1380 58.65 60.29 0.7098 1.0923 56.17 58.78 +RLMRec(Renetal.,2024) 0.8197 1.3312 47.15 42.46 0.7438 1.1069 54.89 57.65 0.7812 1.1584 52.86 55.93 +PALR(Yangetal.,2023) 0.7994 1.2876 48.53 43.19 0.6075 0.9531 57.35 56.77 0.7485 1.1187 54.24 56.38 +LG(Wangetal.,2025) 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +R4Rec(Fangetal.,2025) 0.7028 1.1523 55.69 47.73 0.5654 0.9635 58.69 54.67 0.6397 1.0098 58.47 56.84 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +LLaMA3(8B) +10H 1.0864 1.9532 22.09 27.30 0.7917 1.3346 38.13 46.87 0.8064 1.3866 37.15 45.27 +KAR(Xietal.,2024) 0.6427 1.1668 54.51 47.98 0.5726 0.9033 57.53 59.92 0.5892 0.9614 55.87 58.21 +RLMRec(Renetal.,2024) 0.7428 1.3572 46.74 42.11 0.6076 0.9886 53.78 57.42 0.6226 0.9477 52.12 55.79 +PALR(Yangetal.,2023) 0.7238 1.3265 47.72 43.29 0.5823 0.9222 56.73 59.31 0.5977 0.8855 55.06 57.62 +LG(Wangetal.,2025) 0.6196 1.1289 56.03 51.24 0.5204 0.9369 61.92 59.50 0.5543 0.7967 58.95 60.39 +R4Rec(Fangetal.,2025) 0.7586 1.0418 55.80 53.00 0.5442 0.7722 60.86 54.88 0.6029 0.8345 59.70 56.35 +Ours 0.5367 0.9687 60.87 54.74 0.4680 0.8277 63.30 60.60 0.5092 0.9500 63.42 58.12 +Table1: PerformanceonthreedatasetsusingQwen3(8B)andLLaMA3(8B). +Method Yelp AmazonMusic AmazonBooks +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +Table2: RankingperformanceunderEASE-based(Steck,2019)hardnegatives. +Under Qwen3-8B, our method achieves an accu- our approach reaches an NDCG@1 of 0.5619 +racyof61.23%onYelp,67.96%onAmazonMu- and an NDCG@10 of 0.7443, clearly outper- +sicand64.38%onAmazonBooks,surpassingLG forming the strongest baseline methods. Similar +(LettinGo)(Wangetal.,2025)by5.05%,5.59% trendsareobservedonAmazonMusic,whereour +and5.03%,respectively. Similarimprovementsare methodattains0.5347atNDCG@1and0.7331at +observed under LLaMA3-8B, indicating that the NDCG@10, as well as on Amazon Books with +gainsarestableacrossbackbonemodels. NDCG@1 of 0.4866 and NDCG@10 of 0.7107. +Theseresultsindicatethatthelearnedprofilesen- +Advantages over fixed-structure and delib- +able more accurate identification and ordering of +erative baselines. Compared with strong fixed- +relevantitemswithincandidatesets,leadingtosu- +structureordeliberativemethodssuchasKAR(Xi +perior ranking quality in downstream recommen- +etal.,2024)andReason4Rec(Fangetal.,2025), +dation. Wefurtherevaluaterankingperformance +our approach consistently achieves lower predic- +under a more challenging setting using EASE- +tion error and higher accuracy. For example, on +based (Steck, 2019) hard negatives, and observe +Amazon Books with Qwen3-8B, our method re- +consistentimprovements;detailedresultsarepro- +duces MAE to 0.4612, compared to 0.7098 for +videdinAppendixB.2. +KARand0.6397forReason4Rec,demonstrating +moreeffectiveabstractionofuserpreferenceswith- +4.3 AblationStudy +outrelyingonpredefinedreasoningtemplates. +Consistent improvements in downstream Effectivenessofprofilegeneration,cue&strategy, +ranking. Table??reportstherankingresultsun- anduser–itemjointoptimization. Table3reports +der the Qwen3-8B backbone. Our method con- the ablation results under different design con- +sistentlyachievesthebestperformanceacrossall figurations. Starting from the history-only base- +three datasets and evaluation cutoffs. On Yelp, line (10H), introducing explicit profile genera- + +| Method | | | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | --- | ---- | --- | ------ | ----------- | --- | --- | ----------- | ------ | +| | | | MAE | RMSE | Acc | F1 MAE | RMSE | Acc | F1 | MAE RMSE | Acc F1 | +10H(HistoryOnly) 1.1235 1.9478 23.17 27.54 0.9102 1.4021 39.26 46.58 0.9314 1.4527 37.63 45.19 +Profile 0.7218 1.1863 55.48 48.09 0.6597 1.0218 58.67 57.48 0.6764 1.0469 57.14 57.68 +Profile+Cue&Strategy 0.7085 1.1654 55.83 48.54 0.5708 0.9897 58.91 55.53 0.6389 1.0108 58.43 56.88 +Profile+JointOpt.(LG(Wang 0.6632 1.1047 56.18 48.95 0.4737 0.8834 62.37 57.09 0.5821 0.9416 59.35 60.57 +etal.,2025)) +Profile+Cue&Strategy+ 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +JointOpt.(Ours) +Table3: AblationstudyondifferentdesignconfigurationsinDUETusingQwen3(8B). +| Method | | Yelp | | | | AmazonMusic | | | | AmazonBooks | | +| ------ | -------- | ---- | --- | ----- | --- | ----------- | --- | ----- | --- | ----------- | --------- | +| | MAE RMSE | | Acc | F1(%) | MAE | RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | +| | | | (%) | | | | (%) | | | | (%) | +10H+30P 0.5126 0.9485 61.23 55.18 0.3883 0.7494 67.96 63.89 0.4612 0.9089 65.13 59.97 +10H+50P 0.4909 0.9207 62.43 56.24 0.3924 0.7543 67.88 63.87 0.4553 0.9023 64.62 59.52 +10H+70P 0.4987 0.9326 61.98 55.81 0.3937 0.7564 68.22 64.12 0.4608 0.9068 64.38 59.27 +Table4: Impactofhistoricalinteractionlengthonprofilequality(usingQwen3(8B)). +tionyieldssubstantialperformanceimprovements jointoptimizationarecombined. Underthisfull +across all datasets. For example, on Yelp, MAE configuration, accuracy reaches 61.23% on Yelp, +isreducedfrom1.1235to0.7218andaccuracyin- 67.96%onAmazonMusic,and64.38%onAma- +creases from 23.17% to 55.48%, confirming that zon Books, with corresponding MAE values of +textualprofilesprovidesignificantlymoreinforma- 0.5126, 0.3937, and 0.4612, respectively. Com- +tiverepresentationsthanrawinteractionhistories. pared with the LettinGo-style configuration, this +| | | | | | | setting | yields | consistent | | improvements | across all | +| ------ | ---------------- | --- | ----- | --- | ------ | --------- | ------ | -------------------------------------- | --- | ------------ | ---------- | +| Adding | the cue&strategy | | layer | on | top of | pro- | | | | | | +| | | | | | | datasets. | | Overall,theablationresultsindicatethat | | | | +filegenerationleadstomodestbutconsistentgains. +whileprofilegenerationandjointoptimization +OnAmazonMusic,MAEfurtherdecreasesfrom +0.6597to0.5708,whileaccuracyslightlyimproves contributesubstantiallytoperformancegains,inte- +gratingcue&strategydiscoveryfurtherenhances +| from58.67%to58.91%. | | | Althoughthenumerical | | | | | | | | | +| ------------------- | --- | --- | -------------------- | --- | --- | --- | ------------- | --- | -------- | --------- | ------------ | +| | | | | | | the | effectiveness | | of joint | user–item | optimization | +improvementsintroducedbystrategyalonearelim- +ited, these results suggest that strategy discovery byprovidingastructuredspaceforexplorationand +refinement. +primarilyreshapeshowpreferenceinformationis +abstractedandexpressed,ratherthandirectlyopti- Impactofhistoricalinteractionlengthonprofile +mizingpredictionaccuracyinisolation. +| | | | | | | quality. | | Table | 4 analyzes | the | impact of histori- | +| --- | --- | --- | --- | --- | --- | -------- | --- | ----- | ---------- | --- | ------------------ | +Whenenablingjointoptimizationwithoutstrat- cal interaction length on profile quality. Across +thethreedatasets,varyingthenumberofhistorical +| egy (i.e., | the LettinGo | (Wang | et | al., 2025)-style | | | | | | | | +| ---------- | ------------ | ----- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | +configuration),performanceimprovesmorenotice- interactions leads to only moderate performance +ablyacrossdatasets. OnAmazonMusic,accuracy differences, indicating that our method does not +| | | | | | | strongly | | depend | on long | histories. | In particular, | +| --- | --- | --- | --- | --- | --- | -------- | --- | ------ | ------- | ---------- | -------------- | +increasesto62.37%,comparedto58.67%withpro- +filegenerationalone. Inourimplementation,this usingamoderatehistorylength(e.g.,30–50inter- +settingcorrespondstoareproductionofLettinGo, actions)alreadyachievescompetitiveorbestper- +whereprofilegenerationandjointoptimizationare formanceonmostmetrics,whilefurtherincreasing +thehistorylengthprovideslimitedadditionalgains. +| appliedwithoutthestrategylayer. | | | | Whiletheorig- | | | | | | | | +| ------------------------------- | ------- | --------- | --- | ------------- | ---------- | --- | --- | --- | --- | --- | --- | +| inal method | focuses | primarily | on | user | profiling, | | | | | | | +OnYelpandAmazonBooks,extendingthehis- +weextendtheoptimizationtobothuseranditem +torylengthbeyondthisrangedoesnotconsistently +profilestoensureafaircomparison. improveaccuracyorF1scoreandmayslightlyde- +Thestrongestandmostconsistentperformance gradeperformance,suggestingthatexcessivehis- +is achieved when cue&strategy discovery and toricalinteractionscanintroducenoisyorlessrele- + +vantsignals. Incontrast,AmazonMusicexhibits maintaining comparable coverage. Compared to +relativelystableperformanceacrossdifferenthis- existing methods, which either rely on extractive +torylengths,indicatingthatuserpreferencesinthis compression(e.g.,KAR(Xietal.,2024))orgener- +domainarelesssensitivetohistorytruncation. atefree-formdescriptionswithweakergrounding +Necessity of RL-based Optimization. We com- (e.g.,RLMRec(Renetal.,2024)andPALR(Yang +pare the full model with a variant without RL etal.,2023)), DUET consistentlyattainsstronger +(DUET W/O RL) to isolate the effect of RL. As alignment without sacrificing coverage. For in- +showninTable5,removingRLleadstosubstantial stance,LG(Wangetal.,2025)achievesrelatively +performancedegradationacrossalldatasets(e.g., highalignmentbutexhibitslessstablegrounding +YelpAccuracydropsfrom61.23%to48.53%),indi- acrossdatasets,whileRLMRecattainshighercov- +catingthatthegainscannotbeattributedtoprompt erageatthecostofweakersemanticalignment. In +| designalone. | | | | | contrast,DUETmaintainsbothhighalignmentand | | | | | | +| ------------ | --- | --- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | +Without RL, the generator reduces to a static mid-to-highcoverage,suggestingamoreeffective +mapping from interaction history to textual pro- balancebetweensemanticabstractionandevidence +| files,lackingadaptiveselectionofrelevantsignals. | | | | | preservation. | | | | | | +| ------------------------------------------------ | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | +Incontrast,RLenablesoptimizingprofileconstruc- +4.5 CaseStudy +tionunderrewardfeedback,resultinginmoredis- +Figure4presentsarepresentativecasestudythat +| criminative | representations.These | | results | demon- | | | | | | | +| ----------- | --------------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | +strate that RL-based optimization is essential for demonstrates how semantically aligned user and +| DUET | . | | | | itemprofiles | enableaccurate | | prediction | | thattran- | +| ---- | --- | --- | --- | --- | ------------ | -------------- | --- | ---------- | --- | --------- | +scendsthelimitationsofsparserawinteractionhis- +4.4 SemanticAnalysisofGeneratedProfiles tories. In the user profile, initially scattered and +To better understand the source of performance fragmentedpreferencecuesaresystematicallydis- +gains, we analyze whether the generated profiles tilled into a stable and coherent preference struc- +exhibitmeaningfulsemanticpropertiesratherthan turefavoringfunk,soul,andprogressiverock(or- +servingasintermediatetextualartifacts. Weintro- ange), coupled with a pronounced emphasis on +duce two complementary metrics to characterize musicalcomplexityandhistoricalsignificance(pur- +semanticcompatibilityandgrounding. ple). Thecorrespondingitemprofileexhibitsstrik- +Semantic Alignment. We measure the ingsymmetry,characterizingthealbumasanexem- +embedding-levelsimilaritybetweengenerateduser plaryfunk-rockcomposition(orange)andposition- +anditemprofilesusingall-mpnet-base-v2from ingitasadefiningandculturallyinfluentialrelease +Sentence-Transformers (Reimers and Gurevych, of the 1970s (purple). This example shows 519 +2019). Foreachuser–itempair,wecomputecosine that the learned profiles capture meaningful pref- +| similarity: | | | | | erence–attributecorrespondencethatisdifficultto | | | | | | +| -------------------------------- | -------------------------------- | --- | ------------ | --- | ----------------------------------------------- | --------- | ------- | ---------- | ---- | ----------- | +| | | | e ·e | | recoverfromindividualreviewsalone. | | | | | | +| | | | u i | | | | | | | | +| | Align(u,i) | = | | (3) | | | | | | | +| | | | ∥e ∥∥e ∥ | | | | | | | | +| | | | u i | | 5 Conclusion | | | | | | +| wheree | ande denotetheembeddingvectorsof | | | | | | | | | | +| | u i | | | | In this | paper, we | propose | DUET | , a | closed-loop | +| thegenerateduseranditemprofiles. | | | Highervalues | | | | | | | | +| | | | | | framework | for | jointly | generating | user | and item | +indicatestrongersemanticcompatibilitybetween +| | | | | | textualprofilesforrecommendation. | | | | | Unlikeprior | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | ----------- | +modeleduserpreferencesanditemcharacteristics. +| | | | | | methods | that rely | on | fixed templates | | or indepen- | +| -------- | --------------- | --- | ---------- | ------ | ------- | ----------- | --------- | --------------- | --- | -------------- | +| Coverage | (Faithfulness). | | We measure | token- | | | | | | | +| | | | | | dently | constructed | profiles, | DUET | | treats profile | +levelgroundingas: +| | | | | | generation | as an | exploration | problem | | and aligns | +| --- | --- | --- | --- | --- | ---------- | ----- | ----------- | ------- | --- | ---------- | +|Tokens(profile)∩Tokens(history)| representations directly with downstream recom- +Cov = +| | | |Tokens(profile)| | | | mendationperformance. | | | | | | +| --- | --- | ----------------- | --- | --- | --------------------- | --- | ---- | ---------- | --------- | ---- | +| | | | | (4) | Specifically, | | DUET | integrates | cue-based | ini- | +Thismetricquantifieshowmuchofthegenerated tialization, adaptive strategy construction, and +profileissupportedbyhistoricaltextualevidence. feedback-drivenjointoptimizationtoproduceflex- +We report coverage separately for user and item ible yet task-aligned user–item profiles. By opti- +| profiles. | | | | | mizing | both profiles | in | a shared | semantic | space, | +| --------- | --- | --- | --- | --- | ------ | ------------- | --- | -------- | -------- | ------ | +AsshowninTable6,DUETachievesthehigh- theframeworkreducessemanticmismatchandcap- +est semantic alignment across all datasets while turesinteraction-relevantsignalsthataredifficultto + +Setting Yelp AmazonMusic AmazonBooks +MAE RMSE Acc F1 MAE RMSE Acc F1 MAE RMSE Acc F1 +(%) (%) (%) (%) (%) (%) +DUETw/oRL 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +DUET(full) 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table5: EffectofRL-basedoptimization(usingQwen3(8B)). +User History: +Andre' S Grindle "Tour of U.S.A.", Rating: 5.0. If your looking for heavy metals dim orgins look no further. +Andre' S Grindle "Misfits", Rating:4.0. Actually the opening title track is more of a direct pop/rock type song. But the rock n roll +returns in a big way on "Hay Fever". +............ +Item History: +Worlds Away, Rating: 4.0, Review: The CD is wonderful, and is a seminal piece of music for both the group and for those of us who +were 'of age' in the '70s and '80s. Pablo Cruise was - and is - a group that makes beautiful music like few others, and this +shipper/seller made it all the more special with a fast ship! THANKS!!! +Worlds Away, Rating: 5.0, I love this CD! I'd recommend the Pablo Cruise Millenium Collection to get to know this band. Worlds Away to +me represents the best of breezy '70s pop. Five stars! +............ +USER PROFILE: +Andre Grindle is a dedicated music enthusiast with a deep appreciation for funk, soul, and progressive rock, particularly drawn to artists +who blend technical mastery with genre-defining innovation. His reviews consistently highlight albums that showcase musical complexity, +lyrical depth, and historical significance, often referencing artists like Stevie Wonder, Prince, and Rush. He values authenticity and +artistic vision, frequently praising albums that resist trend-chasing and prioritize creative integrity. His detailed, introspective reviews +suggest a preference for nuanced analysis over superficial praise, with a focus on emotional resonance, production quality, and the +artist's evolution. While he is generous with high ratings (often 5.0), he is also willing to critique perceived shortcomings, as seen in his +4.0 for "Test for Echo" and 3.0 for "The Color of Success." His engagement is characterized by a blend of personal anecdotes, technical +appreciation, and historical context, indicating a preference for music that transcends mere entertainment. +Item PROFILE: +"Worlds Away" by Pablo Cruise is celebrated as a peak-era funk-rock album that exemplifies the band's ability to merge polished +production with genre-blurring creativity. Users consistently praise its tight musicianship, genre-defying sound, and the band's ability to +craft catchy, emotionally resonant tracks. The album's appeal lies in its seamless fusion of rock, funk, and pop, with standout tracks like +"Love Will Find a Way" and "Worlds Away" receiving particular acclaim for their infectious melodies and lyrical depth. While some users +note its dated production compared to modern standards, the majority highlight its historical significance as a defining work of the +1970s rock-funk movement. The album resonates most with fans of 1970s progressive rock, genre-defying artists, and listeners seeking +well-crafted, emotionally engaging music. Its high average rating reflects widespread admiration for its artistic ambition and enduring +appeal, though a minority acknowledges its limitations in terms of modern production quality. +Figure4: Illustrationofthemutualcorrespondencebetweenuseranditem. Thehighlightedregionsdemonstrate +thatuserpreferencessummarizedintheuserprofilealignwiththekeyattributesextractedintheitemprofile,which +providescomplementaryinformationbeyondrawhistoriesandthusimprovespredictionaccuracy. +Method Yelp AmazonMusic AmazonBooks +Align UserCov. ItemCov. Align UserCov. ItemCov. Align UserCov. ItemCov. +10H 0.3902 /(1.00) /(1.00) 0.3577 /(1.00) /(1.00) 0.2604 /(1.00) /(1.00) +KAR(Xietal.,2024) 0.4932 0.1745 0.1823 0.4807 0.2170 0.2391 0.5702 0.1877 0.1544 +RLMRec(Renetal.,2024) 0.4010 0.2646 0.3773 0.3931 0.3526 0.4889 0.4436 0.2994 0.3506 +PALR(Yangetal.,2023) 0.4715 0.3675 0.1917 0.4938 0.2197 0.2390 0.5216 0.2050 0.2267 +LG(Wangetal.,2025) 0.5709 0.2378 0.2682 0.5109 0.2546 0.3749 0.5506 0.3883 0.3364 +R4Rec(Fangetal.,2025) 0.4882 0.2348 0.2330 0.4208 0.3328 0.3595 0.4946 0.2831 0.2949 +Ours 0.6382 0.2880 0.3429 0.5947 0.4002 0.4482 0.7287 0.3457 0.3127 +Table6: Semanticalignmentandcoverageofgeneratedprofiles. Cov. denotesthefractionoftokensgroundedin +historicaltext. +recoverfromrawhistoriesorstaticpromptsalone. reinforcementlearning–basedoptimization. These +resultssuggestthatadaptive,interaction-awaretex- +Experiments on multiple real-world datasets tualprofilesprovideapromisingdirectionformore +demonstratethatDUETconsistentlyoutperforms effective interpretable and performance-oriented +strongbaselinesunderdifferentbackbonemodels, recommendationsystems. +validating the effectiveness of joint profiling and + +| Limitations | | | | | | | References | | | | | | | +| ----------- | --- | --- | --- | --- | --- | --- | ---------- | ----- | ------ | ---- | ------ | ------ | ----- | +| | | | | | | | Keqin Bao, | Jizhi | Zhang, | Yang | Zhang, | Wenjie | Wang, | +Despiteitseffectiveness,ourapproachhasseveral FuliFeng,andXiangnanHe.2023. Tallrec: Anef- +fectiveandefficienttuningframeworktoalignlarge +| limitations. | First, | the | proposed | framework | | relies | | | | | | | | +| ------------ | ------ | --- | -------- | --------- | --- | ------ | -------------------------------- | --- | --- | --- | --- | ---------- | --- | +| | | | | | | | languagemodelwithrecommendation. | | | | | InProceed- | | +onlargelanguagemodelsforbothprofilegenera- ingsofthe17thACMConferenceonRecommender +tion and downstream recommendation, which in- Systems,pages1007–1014. +| troduces | additional | computational | | | overhead | dur- | | | | | | | | +| -------- | ---------- | ------------- | --- | --- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | +TingChen,Wei-LiHan,Hai-DongWang,Yi-XunZhou, +| ingtrainingandinference. | | | Whileourexperiments | | | | | | | | | | | +| ------------------------ | --- | --- | ------------------- | --- | --- | --- | ------------------------- | --- | --- | --- | ---------------- | --- | --- | +| | | | | | | | BinXu,andBin-YuZang.2007. | | | | Contentrecommen- | | | +showconsistentgainsacrossdifferentbackbones, dationsystembasedonprivatedynamicuserprofile. +the overall efficiency may be constrained when In2007Internationalconferenceonmachinelearn- +| | | | | | | | ing and | cybernetics, | | volume | 4, pages | 2112–2118. | | +| --------------------------------- | --- | --- | --- | --- | ---------- | --- | ------- | ------------ | --- | ------ | -------- | ---------- | --- | +| scalingtoverylargeuseroritemsets. | | | | | Second,the | | | | | | | | | +IEEE. +| quality | of the generated | | profiles | is | inherently | de- | | | | | | | | +| ------- | ----------------- | --- | -------- | --- | ------------- | --- | --------- | ------- | --- | ----------- | ----- | --- | ----- | +| pendent | on the underlying | | LLMs | | and prompting | | | | | | | | | +| | | | | | | | Yue Chen, | Minghua | | He, Fangkai | Yang, | Pu | Zhao, | +strategies. Variationsinmodelcapacityorprompt LuWang,YuKang,YifeiDong,YuefengZhan,Hao +Sun,QingweiLin,SaravanRajmohan,andDongmei +sensitivitymayleadtodifferencesinprofilestabil- +| | | | | | | | Zhang. | 2025. | Warriormath: | | Enhancing | the | mathe- | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | ------------ | --- | --------- | --- | ------ | +ity,whichwedonotexplicitlycontrolinthecurrent +maticalabilityoflargelanguagemodelswithadefect- +design. Finally,ourevaluationfocusesontext-rich +| | | | | | | | awareframework. | | | Preprint,arXiv:2508.01245. | | | | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | -------------------------- | --- | --- | --- | +recommendationscenarioswheresufficienthistor- +| | | | | | | | Paul Covington, | | Jay | Adams, and | Emre | Sargin. | 2016. | +| ------------ | --- | ---------- | --- | ----------------- | --- | --- | --------------- | --- | --- | ---------- | ---- | ------- | ----- | +| ical reviews | are | available. | | The effectiveness | | of | | | | | | | | +Deepneuralnetworksforyoutuberecommendations. +| the proposed | strategy | | in domains | with | extremely | | | | | | | | | +| ------------ | -------- | --- | ---------- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +InProceedingsofthe10thACMconferenceonrec- +sparsetextualsignalsorinnon-textualmodalities +ommendersystems,pages191–198. +remainstobefurtherexplored. +DeepSeek-AI,DayaGuo,DejianYang,HaoweiZhang, +JunxiaoSong,RuoyuZhang,RunxinXu,QihaoZhu, +ShirongMa,PeiyiWang,XiaoBi,XiaokangZhang, +Data,Privacy,andEthicsConsiderations Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhi- +| | | | | | | | hong | Shao, | Zhuoshu | Li, Ziyi | Gao, and | 181 | others. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------ | ------- | ------------- | --------- | --- | ------- | +| | | | | | | | 2025. | Deepseek-r1: | | Incentivizing | reasoning | | capa- | +Alldatasetsusedinthisworkarepubliclyreleased bilityinllmsviareinforcementlearning. Preprint, +arXiv:2501.12948. +bytheiroriginalowners,andwefollowtheofficial +termsofuseassociatedwitheachdataset,utilizing +JiaxinDeng,ShiyaoWang,KuoCai,LejianRen,Qigen +thedatasolelyforresearchpurposesaspermitted +| | | | | | | | Hu, Weifeng | | Ding, | Qiang Luo, | and | Guorui | Zhou. | +| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ----- | ---------- | --- | ------ | ----- | +bytheirrespectivelicenses;forreproducibility,the 2025. Onerec: Unifyingretrieveandrankwithgen- +officialdatasetsourcesarecitedinthecorrespond- erativerecommenderanditerativepreferencealign- +ment. Preprint,arXiv:2502.18965. +ingsectionsofthispaper,andwedonotredistribute +anyrawuser-generatedcontentsuchasreviewtext AbhimanyuDubey,AbhinavJauhri,AbhinavPandey, +| aspartofthiswork. | | Thisresearchdoesnotattempt | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +AbhishekKadian,AhmadAl-Dahle,AieshaLetman, +to identify or infer the identity of any individual Akhil Mathur, Alan Schelten, Amy Yang, Angela +| | | | | | | | Fan,and1others.2024. | | | Thellama3herdofmodels. | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | ---------------------- | --- | --- | --- | +user,andpriortotraininganddownstreamprofile +arXivpreprintarXiv:2407.21783. +| generation, | we | apply text | filtering | | procedures | to | | | | | | | | +| ----------- | --- | ---------- | --------- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +remove direct personal identifiers where applica- YiFang,WenjieWang,YangZhang,FengbinZhu,Qi- +ble;themodelsaretrainedsolelyonpubliclyavail- fanWang,FuliFeng,andXiangnanHe.2025. Rea- +son4rec:Largelanguagemodelsforrecommendation +abledata,andnoadditionalprivateorproprietary +| | | | | | | | withdeliberativeuserpreferencealignment. | | | | | Preprint, | | +| ---------------- | --- | -------------- | --- | ------------ | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --------- | --- | +| user information | | is introduced. | | Furthermore, | | we | | | | | | | | +arXiv:2502.02061. +| do not release | generated | | user | profiles | or | any per- | | | | | | | | +| -------------- | --------- | --- | ---- | -------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +MinghuaHe,YueChen,FangkaiYang,PuZhao,Wenjie +useroutputsproducedduringthemodelingprocess, +Yin,YuKang,QingweiLin,SaravanRajmohan,and +andallreportedresultsarepresentedinaggregate +| | | | | | | | DongmeiZhang.2025. | | | Execoder:Empoweringlarge | | | | +| -------- | ---------- | -------- | --- | ----- | ---- | ------- | ------------------ | --- | --- | ------------------------ | --- | --- | --- | +| form for | evaluation | purposes | | only. | From | a secu- | | | | | | | | +languagemodelswithexecutabilityrepresentationfor +rityperspective,theassociatedcoderepositoryhas codetranslation. Preprint,arXiv:2501.18460. +beenscannedforpotentialsecrets,andanyinternal +MinjieHong,ZirunGuo,YanXia,ZehanWang,Ziang +endpointsorconfidentialconfigurationinformation +| | | | | | | | Zhang, | Tao | Jin, and | Zhou Zhao. | 2025a. | Apo: | En- | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | -------- | ---------- | ------ | ---- | --- | +havebeenremovedpriortorelease. +hancingreasoningabilityofmllmsviaasymmetric +| | | | | | | | policyoptimization. | | | Preprint,arXiv:2506.21655. | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | -------------------------- | --- | --- | --- | + +Minjie Hong, Yan Xia, Zehan Wang, Jieming Zhu, JoarSkalse,NikolausH.R.Howe,DmitriiKrashenin- +Ye Wang, Sihang Cai, Xiaoda Yang, Quanyu Dai, nikov, and David Krueger. 2025. Defining +ZhenhuaDong,ZhimengZhang,and1others.2025b. and characterizing reward hacking. Preprint, +| Eager-llm: | Enhancinglargelanguagemodelsasrec- | | | | | arXiv:2209.13085. | | | | | | +| ---------- | ---------------------------------- | --------- | ----------------- | --- | --- | ----------------- | --- | --- | --- | --- | --- | +| ommenders | through | exogenous | behavior-semantic | | | | | | | | | +integration. InProceedingsoftheACMonWebCon- Harald Steck. 2019. Embarrassingly shallow autoen- +ference2025,pages2754–2762. codersforsparsedata. InProceedingsofTheWorld +WideWebConference(WWW),pages3251–3257. +MinjieHong,ZetongZhou,ZirunGuo,ZiangZhang, +| Ruofan | Hu, Weinan | Gan, | Jieming | Zhu, and | Zhou | | | | | | | +| ------ | ---------- | ---- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | +ChaoSun,YaoboLiang,YamingYang,ShilinXu,Tian- +| Zhao.2025c. | Generativereasoningrecommendation | | | | | | | | | | | +| ----------- | --------------------------------- | --- | --- | --- | --- | ---------------------------- | -------- | ---- | ------ | -------- | ------------- | +| | | | | | | mengYang,andYunhaiTong.2024. | | | | | Rlrf4rec: Re- | +| viallms. | Preprint,arXiv:2510.20815. | | | | | | | | | | | +| | | | | | | inforcement | learning | from | recsys | feedback | for en- | +JihwanJeong,YinlamChow,GuyTennenholtz,Chih- hancedrecommendationreranking. arXivpreprint +| Wei Hsu, | Azamat | Tulepbergenov, | | Mohammad | | arXiv:2410.05939. | | | | | | +| ------------------------------------ | ------------ | --------------- | ---------- | -------- | --------- | ----------------- | --- | --------------------- | --- | --- | --------- | +| Ghavamzadeh, | and | Craig | Boutilier. | 2023. | Fac- | | | | | | | +| | | | | | | QwenTeam.2025. | | Qwen3technicalreport. | | | Preprint, | +| tual and | personalized | recommendations | | using | lan- | | | | | | | +| guagemodelsandreinforcementlearning. | | | | | Preprint, | arXiv:2505.09388. | | | | | | +arXiv:2310.06176. +| | | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | +| --- | --- | --- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | +Yitong Ji, Aixin Sun, Jie Zhang, and Chenliang Li. bert, Amjad Almahairi, Yasmine Babaei, Nikolay +| 2023. A | critical | study | on data leakage | in | recom- | | | | | | | +| ------- | -------- | ----- | --------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti +| mendersystemofflineevaluation. | | | | ACMTrans.Inf. | | | | | | | | +| ------------------------------ | --- | --- | --- | ------------- | --- | ------------------------ | --- | --- | ------- | --- | --------- | +| | | | | | | Bhosale,and1others.2023. | | | Llama2: | | Openfoun- | +Syst.,41(3):75:1–75:27. +| | | | | | | dation and | fine-tuned | chat | models. | arXiv | preprint | +| --- | --- | --- | --- | --- | --- | ---------- | ---------- | ---- | ------- | ----- | -------- | +arXiv:2307.09288. +| JiachengLin,TianWang,andKunQian.2025. | | | | | Rec-r1: | | | | | | | +| ------------------------------------- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | +Bridginggenerativelargelanguagemodelsanduser- +| | | | | | | LeandrovonWerra, | | YounesBelkada, | | LewisTunstall, | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | -------------- | --- | -------------- | --- | +centricrecommendationsystemsviareinforcement +EdwardBeeching,TristanThrush,NathanLambert, +| learning. | Preprint,arXiv:2503.24289. | | | | | | | | | | | +| --------- | -------------------------- | --- | --- | --- | --- | -------- | ------ | ----------- | ------------- | --- | ------------ | +| | | | | | | Shengyi | Huang, | Kashif | Rasul, | and | Quentin Gal- | +| | | | | | | louédec. | TRL: | Transformer | Reinforcement | | Learn- | +AiweiLiu,MinghuaHe,ShaoxunZeng,SijunZhang, +| LinhaoZhang,ChuhanWu,WeiJia,YuanLiu,Xiao | | | | | | ing. | | | | | | +| ---------------------------------------- | --- | --- | ----------------------- | --- | --- | ---- | --- | --- | --- | --- | --- | +| Zhou,andJieZhou.2025. | | | Wedlm:Reconcilingdiffu- | | | | | | | | | +sionlanguagemodelswithstandardcausalattention LuWang,DiZhang,FangkaiYang,PuZhao,Jianfeng +forfastinference. Preprint,arXiv:2512.22737. Liu,YuefengZhan,HaoSun,QingweiLin,Weiwei +| | | | | | | Deng, Dongmei | | Zhang, | Feng | Sun, and | Qi Zhang. | +| --- | --- | --- | --- | --- | --- | ------------- | --- | ------ | ---- | -------- | --------- | +HongtaoLiu,FangzhaoWu,WenjunWang,Xianchen +| | | | | | | 2025. Lettingo: | | Exploreuserprofilegenerationfor | | | | +| --- | --- | --- | --- | --- | --- | --------------- | --- | ------------------------------- | --- | --- | --- | +Wang,PengfeiJiao,ChuhanWu,andXingXie.2019. +| | | | | | | recommendationsystem. | | | InProceedingsofthe31st | | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ---------------------- | --- | --- | +Nrpa: Neuralrecommendationwithpersonalizedat- ACMSIGKDDConferenceonKnowledgeDiscovery +tention. In Proceedings of the 42nd International andDataMiningV.2,KDD’25,NewYork,NY,USA. +ACMSIGIRConferenceonResearchandDevelop- AssociationforComputingMachinery. +mentinInformationRetrieval,pages1233–1236. +YeWang,JiahaoXun,MinjieHong,JiemingZhu,Tao +| Wensheng | Lu, Jianxun | Lian, | Wei Zhang, | Guanghua | | | | | | | | +| -------- | ----------- | ----- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | +Jin,WangLin,HaoyuanLi,LinjunLi,YanXia,Zhou +Li,MingyangZhou,HaoLiao,andXingXie.2024. +| | | | | | | Zhao,and1others.2024. | | | Eager: | Two-streamgen- | | +| --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ------ | -------------- | --- | +Aligninglargelanguagemodelsforcontrollablerec- +arXivpreprintarXiv:2403.05063. erativerecommenderwithbehavior-semanticcollab- +ommendations. +| | | | | | | oration. | InProceedingsofthe30thACMSIGKDD | | | | | +| --- | --- | --- | --- | --- | --- | -------- | ------------------------------- | --- | --- | --- | --- | +RadaMihalceaandPaulTarau.2004. Textrank: Bring- ConferenceonKnowledgeDiscoveryandDataMin- +| | | InEMNLP. | | | | ing,pages3245–3254. | | | | | | +| --- | --- | -------- | --- | --- | --- | ------------------- | --- | --- | --- | --- | --- | +ingorderintotext. +| Zhaopeng | Qiu, Xian | Wu, | Jingyue Gao, | and | Wei Fan. | | | | | | | +| -------- | --------- | --- | ------------ | --- | -------- | -------- | --------------- | --- | -------- | -------- | --- | +| | | | | | | Triyanna | Widiyaningtyas, | | Indriana | Hidayah, | and | +2021. U-BERT:pre-traininguserrepresentationsfor +| | | | | | | TeguhBAdji.2021. | | Userprofilecorrelation-based | | | | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | ---------------------------- | --- | --- | --- | +improved recommendation. In AAAI, pages 4320– similarity(upcsim)algorithminmovierecommenda- +| 4327.AAAIPress. | | | | | | tionsystem. | JournalofBigData,8(1):52. | | | | | +| --------------------------------- | --- | --- | --- | -------------- | --- | ----------- | ----------------------------------- | --- | --- | --- | --- | +| NilsReimersandIrynaGurevych.2019. | | | | Sentence-bert: | | | | | | | | +| | | | | | | J.Wu.2023. | Computationalunderstandingofuserin- | | | | | +Sentenceembeddingsusingsiamesebert-networks. +terfaces. +InProceedingsofthe2019ConferenceonEmpirical +MethodsinNaturalLanguageProcessing(EMNLP), +pages3982–3992. Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, +| | | | | | | Hong Zhu, | Jieming | Zhu, | Bo Chen, | Ruiming | Tang, | +| --- | --- | --- | --- | --- | --- | --------- | ------- | ---- | -------- | ------- | ----- | +Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi WeinanZhang,andYongYu.2024. Towardsopen- +Cheng,JunfengWang,DaweiYin,andChaoHuang. world recommendation with knowledge augmenta- +2024. Representationlearningwithlargelanguage tionfromlargelanguagemodels. InProceedingsof +modelsforrecommendation. InProceedingsofthe the18thACMConferenceonRecommenderSystems, +| ACMonWebConference2024,pages3464–3475. | | | | | | pages12–22. | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | + +A ExperimentSetup +FanYang,ZhengChen,ZiyanJiang,EunahCho,Xiao- +| jiangHuang,andYanbinLu.2023. | | Palr:Personaliza- | | | | | | | | +| ------------------------------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| tionawarellmsforrecommendation. | | arXivpreprint | A.1 DataCuration | | | | | | | +arXiv:2305.07622. +Jiarui Zhang. 2024. Guided profile generation im- Table7: Statisticaldetailsoftheevaluationdatasets. +| proves personalization | with llms. | arXiv preprint | | | | | | | | +| ---------------------- | ---------- | -------------- | ---------- | ------ | ----------- | --- | ------------ | ------ | ------ | +| arXiv:2409.13093. | | | Dataset | #Train | #Valid | | #Test #User | | #Item | +| | | | Music | 43,071 | 3,271 | | 1,296 | 4,183 | 2,660 | +| | | | Book | 71,972 | 6,144 | | 5,541 13,863 | | 13,515 | +| | | | Yelp | 51,497 | 4,757 | | 4,328 | 8,453 | 13,426 | +| | | | We conduct | | experiments | | on three | widely | used | +real-worlddatasets: +| | | | • AmazonMusic(Music): | | | | Thisreferstothe“Dig- | | | +| --- | --- | --- | --------------------- | --- | ------ | ------ | -------------------- | --- | ------ | +| | | | ital Music” | | subset | of the | well-known | | Amazon | +Productdataset3,whichrecordsrichuserreviews, +ratings,andtextualinformationaboutitems,such +| | | | as titles, | across | a | broad | range | of product | cate- | +| --- | --- | --- | ---------- | ------ | --- | ----- | ----- | ---------- | ----- | +gories,ontheAmazonplatform.2 +| | | | • AmazonBook(Book): | | | Thisreferstothe“Book” | | | | +| --- | --- | --- | ------------------- | --- | --- | --------------------- | --- | --- | --- | +subsetoftheAmazonProductdataset. +dataset4, +| | | | • Yelp: | This | refers | to the | Yelp | Open | | +| --- | --- | --- | ------- | -------- | -------------- | -------- | ------- | ------------- | --------- | +| | | | which | includes | user | reviews, | ratings | | for busi- | +| | | | nesses | such | as restaurants | | and | retail shops, | as | +wellastextualinformationaboutthebusinesses. +Itiswidelyusedinrecommendationtasks(Qiu +etal.,2021). +WeusetheentireMusicdatasetforexperiments, +| | | | while for | the Book | | and Yelp | datasets, | we | utilize | +| --- | --- | --- | ------------------------------- | -------- | --- | -------- | --------- | ---------- | ------- | +| | | | onlyasubsetduetotheirlargesize. | | | | | FortheBook | | +dataset,weusedatafromthelasttwomonths,and +fortheYelpdataset,weusedatafromthelastsix +months. Foreachdataset,wesplititintotraining, +| | | | validation, | and | test | sets based | on | the timestamps | | +| --- | --- | --- | ---------------- | --- | -------- | ---------- | ---- | -------------- | --- | +| | | | of interactions, | | ensuring | that | test | interactions | oc- | +curafteralltrainingandvalidationinteractionsto +preventinformationleakage(Jietal.,2023). +| | | | Regarding | | data | filtering, | following | | prior | +| --- | --- | --- | --------- | --- | ----------- | ---------- | --------- | -------- | ----- | +| | | | work (Liu | et | al., 2019), | we | adopt | a 5-core | set- | +tingtofilterthedataandexcludecold-startusers +| | | | and items—those | | not | appearing | | in the | training | +| --- | --- | --- | --------------- | --- | ---------- | --------- | ---- | ------ | -------- | +| | | | set—from | the | validation | and | test | sets. | The sta- | +tisticaldetailsoftheprocesseddatasetareprovided +inTable7. +3https://cseweb.ucsd.edu/~jmcauley/datasets/ +amazon/links.html. +4https://business.yelp.com/data/resources/ +open-dataset/. + +| A.1.1 | ImplementationDetails | | | | | | RLMRecPrompt | | | | +| ----- | --------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | +Inourexperiments,weprimarilyemployQwen3- Role: Businessrecommendationassistant +Task: Determinebusinesstypesauserislikely +| 8B(Team,2025)andLLaMA38BInstruct | | | | | | (Dubey | | | | | +| ------------------------------------------ | --- | ----- | --- | --- | ------- | ---------- | ------------ | --- | --- | --- | +| etal.,2024;Touvronetal.,2023)asboththerec- | | | | | | | toenjoy | | | | +| ommendation | | model | and | the | profile | generation | InputFormat: | | | | +•Title: Businessname +model. Thetrainingprocessisimplementedusing +theTRL(vonWerraetal.). Keyhyperparameters, •Categories: Businesscategories +suchasbatchsizeandlearningrate,aredetermined •Sentiment: Usersentimenttowardbusiness +throughgridsearchtoachieveoptimalperformance. +OutputRequirements: +Moredetailscanbefoundinourcode. +1. JSONformatonly +2. Structure: +A.2 BaselinePrompts +{ +| | | | | | | | "summarization": | "Types | of businesses | user | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------ | ------------- | ---- | +KARPrompt +likelyenjoys"(≤100words), +Task: Analyzeuserpreferencesbasedonbusi- +| | | | | | | | "reasoning": | "Brief explanation | | for summa- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ------------------ | --- | ---------- | +nessreviewinghistory +rization"(nowordlimit) +Input: {user_history}-User’sbusinessreview- +} +inghistorywithsentimentsovertime +3. NoadditionaltextoutsideJSON +Instructions: +| 1. | Analyze | the | user’s | preferences | considering | | | | | | +| --- | ------- | --- | ------ | ----------- | ----------- | --- | ------ | ----------- | --- | ------ | +| | | | | | | | Input: | INTERACTION | | ITEMS: | +businessnamesandcategories +{user_history} +| 2. | Take | into account | | sentiment | patterns | over | | | | | +| --- | ---- | ------------ | --- | --------- | -------- | ---- | --- | --- | --- | --- | +time +| 3. | Provideclearexplanationsbasedonreview- | | | | | | LGPrompt | | | | +| --- | -------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | +inghistorydetails +Youwillserveasanassistanttohelpmegener- +| 4. | Consider | other | pertinent | | factors | that may | | | | | +| --- | -------- | ----- | --------- | --- | ------- | -------- | --- | --- | --- | --- | +ateauserprofilebasedonthisuser’ssentiments +influencepreferences +historytobetterunderstandthisusers’interest +andthuspredicthis/hersentimentaboutatarget +| PALRPrompt | | | | | | | item. Iwillprovideyouwithsomebehaviorhis- | | | | +| ---------- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --------------- | --- | +| | | | | | | | toryoftheuserinthisformat: | | [itemattributes | | +Task: Summarizeuserpreferencesusingkey- +| | | | | | | | andsentiment]. | Theuserprofileyougenerate | | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ------------------------- | --- | --- | +words. +shouldcontainasmuchusefulcontentaspossi- +| Input: | | {user_history} | | - historical | businesses | | | | | | +| ------ | --- | -------------- | --- | ------------ | ---------- | --- | --- | --- | --- | --- | +bletohelppredicttheuser’ssentimenttowards +withusersentiments. +anewbusiness. +| Output | | Format: | An | itemized | list ranked | by | | | | | +| ------ | --- | ------- | --- | -------- | ----------- | --- | --- | --- | --- | --- | +USERHISTORY:{user_history}. +importance. +PROFILEYOUGENERATE: +Template: +| | • KEY_WORD_1: | | | | | "HIS- | | | | | +| --- | ------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +R4RecPrompt(Reasoner) +| | TORY_BUSINESS_1", | | | | | "HIS- | | | | | +| --- | ----------------- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | +UserReviewHistory +TORY_BUSINESS_2" +⟨H u organizedasbelow⟩ +| | • KEY_WORD_2: | | | | | "HIS- | 1. TitleofItem1 | | | | +| ------------- | ---------------- | --- | --- | --- | --- | ----- | ---------------- | ----------------------- | ----------- | --- | +| | TORY_BUSINESS_3" | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| | | | | | | | NegativeAspects: | [Aspect1],[Aspect2],... | | | +| Instructions: | | | | | | | User Preference | Elements: | [Preference | 1], | +[Preference2],... +| | 1. Extract | key | preference | | indicators | from | | | | | +| --- | ---------- | --- | ---------- | --- | ---------- | ---- | --- | --- | --- | --- | +2. TitleofItem2 +userinteractionhistory. +| | | | | | | | PositiveAspects: | [Aspect1],[Aspect2],... | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------------- | --- | --- | +2. Rankkeywordsbyimportance. NegativeAspects: [Aspect1],[Aspect2],... +| | | | | | | | User Preference | Elements: | [Preference | 1], | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --------- | ----------- | --- | + +| [Preference2],... | | | | | | | cantlyworsethanDUETacrossalldatasets. | | | | | This | +| ----------------- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | ---- | +indicatesthatcoherentsummariesaloneareinsuf- +... +| | | | | | | | ficient, | and that | the gains | of DUET | | arise from | +| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --------- | ------- | --- | ---------- | +ItemReviewHistorybyOtherUsers +⟨H organizedinthesameformatasabove⟩ learnedsemanticabstractionratherthanextractive +i +compression. +| Task: | Analyze | whether | | the user | will | like the | | | | | | | +| --------- | ------- | ------- | --- | ---------- | ----------- | -------- | --- | --- | --- | --- | --- | --- | +| new Music | | i based | on | the user’s | preferences | | | | | | | | +B.2 RankingunderHardNegativeSampling +| andtheitem’sfeatures. | | | | Provideyourrationale | | | | | | | | | +| --------------------- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +Toconstructamorechallengingrankingscenario, +inoneconciseparagraph. +wereplacerandomnegativeswithhardnegatives +| | | | | | | | generatedbyacollaborativefilteringmodel. | | | | | Specif- | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | ------- | +A.3 ExampleofCue +ically,foreachuser,weretrievehigh-scoringitems +Thefollowingexamplesillustratehowrawsignals +| | | | | | | | from an | EASE(Steck, | 2019) | model | that | the user | +| --------------------- | --- | --- | ---------------- | --- | --- | ---- | ------- | ----------- | ----- | ----- | ---- | -------- | +| aredistilledintocues. | | | Asshownonthetop, | | | user | | | | | | | +hasnotinteractedwith,andcombinethemwiththe +cuesemphasizehistoricalpreferences,whileitem +ground-truthitemtoformthecandidateset. +cues highlight metadata and user-group patterns. Table 9 reports the results. Compared to ran- +| Together, | they | provide | minimal | | but informative | | | | | | | | +| --------- | ---- | ------- | ------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | +domsampling,performancedecreasesforallmeth- +hypothesesforprofileexploration. +| | | | | | | | ods due | to increased | difficulty, | while | DUET | con- | +| --- | --- | --- | --- | --- | --- | --- | --------- | ------------ | ----------- | ---------------- | ---- | ------ | +| | | | | | | | sistently | maintains | the | best performance | | across | +ExamplesofUserCues +| | | | | | | | datasets. | Thisindicatesthattheimprovementsare | | | | | +| ------- | ----- | ------ | ------ | --- | ------- | -------- | --------- | ----------------------------------- | --- | --- | --- | --- | +| “enjoys | retro | puzzle | games” | — | derived | from re- | | | | | | | +robustandnotlimitedtotrivialrankingscenarios. +peatedengagementwithclassictitles. +“prefersconciseproductreviews”—inferredfrom B.3 RobustnessunderPreferenceDiversity +apatternofshort,directcomments. +WefurtheranalyzetherobustnessofDUETunder +“tendstogivehighratingsbutrarelycomments”— +| | | | | | | | varyinglevelsofuserpreferencediversity. | | | | | Weuse | +| --- | --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | ----- | +highlightingconsistencybutlimitedfeedback. +thevarianceofhistoricalratingsasaproxyforpref- +| | | | | | | | erencestability: | | lowvarianceindicatesconsistent | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | ------------------------------ | --- | --- | --- | +ExamplesofItemCues +| | | | | | | | preferences, | while | high | variance | corresponds | to | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ----- | ---- | -------- | ----------- | --- | +“lightweighttrail-runningshoes”—derivedfrom +diverseorpotentiallyconflictingsignals. +productmetadata. +Wepartitionusersintothreegroupsbasedonper- +| “popular | among | budget-conscious | | | users” | — in- | | | | | | | +| -------- | ----- | ---------------- | --- | --- | ------ | ----- | --- | --- | --- | --- | --- | --- | +centilethresholds(bottom33%,middle33%,top +ferredfrompurchasepatterns. +33%ofratingvariance)andevaluateperformance +| “stylized | with | retro | aesthetics” | | — extracted | from | | | | | | | +| --------- | ---- | ----- | ----------- | --- | ----------- | ---- | --- | --- | --- | --- | --- | --- | +withineachgroup. +itemdescriptions. +| | | | | | | | As shown | in | Table 10, | performance | | degrades | +| --- | --- | --- | --- | --- | --- | --- | -------- | --- | --------- | ----------- | --- | -------- | +smoothlyaspreferencediversityincreasesacross +| | | | | | | | all datasets. | Importantly, | | the degradation | | is grad- | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------ | --- | --------------- | --- | -------- | +B AdditionalExperiments +ualratherthancatastrophic,indicatingthatDUET +B.1 Non-LLMBaselineviaExtractive +remainsstableunderheterogeneousornoisyinter- +Summarization +actionhistories. +| To examine | | whether | the | gains | of DUET | stem | | | | | | | +| ---------- | --- | ------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | +C TheUseofLargeLanguageModels +| from improved | | semantic | | representations | | rather | | | | | | | +| ------------- | --- | ---------------- | --- | --------------- | --------- | ------ | --- | --- | --- | --- | --- | --- | +| than generic | | text generation, | | we | introduce | a non- | | | | | | | +WeusedaLargeLanguageModel(LLM)onlyas +LLMbaselinebasedonextractivesummarization. a writing assistant to polish the language of the +| Specifically, | | we apply | TextRank | | (Mihalcea | and | | | | | | | +| ------------- | --- | -------- | -------- | --- | --------- | --- | ---------- | ------ | ------- | ----------- | --- | --------- | +| | | | | | | | manuscript | (e.g., | grammar | refinement, | | style ad- | +Tarau,2004)toselectsalientsentencesfromuser +| | | | | | | | justment,andclarityimprovement). | | | | Theresearch | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | ----------- | --- | +historiesandconstructuserprofileswithoutusing ideas,methodologydesign,experiments,andanal- +anygenerativemodel. ysiswereentirelyconceived,implemented,andval- +| The | extracted | summaries | | are | then fed | into the | | | | | | | +| --- | --------- | --------- | --- | --- | -------- | -------- | --- | --- | --- | --- | --- | --- | +idatedbytheauthorswithoutrelianceontheLLM. +samedownstreampredictorforratingestimation. +| | | | | | | | The LLM | did not | contribute | to research | | ideation, | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------- | ---------- | ----------- | --- | --------- | +Thisbaselineisolatestheeffectofreadablesumma- experimentaldesign,orresultinterpretation. +rizationfromrepresentationlearning. +Table8showsthatTextRankimprovesoversim- +| ple history | truncation | | (10H), | but | remains | signifi- | | | | | | | +| ----------- | ---------- | --- | ------ | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | + +| Method | | Yelp | | | AmazonMusic | | | AmazonBooks | | +| ------ | -------- | ---- | ----- | --- | ----------- | --------- | --- | ----------- | --------- | +| | MAE RMSE | Acc | F1(%) | MAE | RMSE | Acc F1(%) | MAE | RMSE | Acc F1(%) | +| | | (%) | | | | (%) | | | (%) | +10H 0.8283 1.3893 48.53 41.15 0.7322 1.1430 57.18 52.48 0.8741 1.4760 51.83 50.35 +TextRank 0.8104 1.1323 50.51 30.23 0.5914 0.8223 61.19 29.68 0.6328 0.8854 59.59 29.48 +Ours 0.5126 0.9485 61.23 55.18 0.3937 0.7564 67.96 63.89 0.4612 0.9089 64.38 59.27 +Table8: Comparisonwithanon-LLMextractivesummarizationbaseline(TextRank). +| Method | | | Yelp | | AmazonMusic | | | AmazonBooks | | +| ------ | --- | --- | ---- | --- | ----------- | --- | --- | ----------- | --- | +NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 NDCG@1 NDCG@5 NDCG@10 +10H 0.1823 0.2815 0.4928 0.1875 0.3796 0.5153 0.1841 0.3146 0.4263 +KAR(Xietal.,2024) 0.2156 0.3298 0.5412 0.3018 0.4896 0.6015 0.2965 0.4715 0.5834 +RLMRec(Renetal.,2024) 0.2419 0.3472 0.5587 0.3371 0.5434 0.6162 0.2748 0.4526 0.5719 +PALR(Yangetal.,2023) 0.2494 0.3563 0.5691 0.3395 0.5247 0.6115 0.2627 0.4634 0.5538 +LG(Wangetal.,2025) 0.3187 0.4685 0.5814 0.4012 0.5674 0.6489 0.3795 0.5189 0.6284 +R4Rec(Fangetal.,2025) 0.2575 0.3792 0.5526 0.2928 0.5912 0.6343 0.3013 0.4928 0.5959 +Ours 0.3390 0.4873 0.6008 0.5123 0.6165 0.7025 0.4288 0.5638 0.6599 +| | Table9: | RankingperformanceunderEASE-based(Steck,2019)hardnegatives. | | | | | | | | +| ------------- | ------- | ----------------------------------------------------------- | --- | --- | ----------- | --- | --- | ----------- | --- | +| VarianceGroup | | Yelp | | | AmazonMusic | | | AmazonBooks | | +#Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 #Samp. MAE RMSE Acc F1 +| | | | (%) | (%) | | (%) (%) | | | (%) (%) | +| --- | --- | --- | --- | --- | --- | ------- | --- | --- | ------- | +Stable 1530 0.4017 0.8664 71.76 68.96 432 0.2368 0.5990 82.56 86.89 1876 0.3171 0.8403 76.64 72.56 +Moderate 1354 0.4773 0.8309 60.09 46.26 436 0.3921 0.6628 64.16 47.02 1827 0.4495 0.8059 66.73 55.45 +Diverse 1444 0.6632 1.1198 51.13 45.95 428 0.5536 0.9607 57.10 59.27 1838 0.6199 1.0603 49.52 46.35 +Table10: PerformanceofDUETunderdifferentlevelsofuserpreferencediversity(measuredbyratingvariance). +#Samp. denotesthenumberofsamplesineachgroup. \ No newline at end of file diff --git a/q166/random_k4/random_2.md b/q166/random_k4/random_2.md index 7f06b970eb191d018409053c543e59811fc6cbd9..b408d57341c4f4ff52ea6f83833a4f9892d91ec0 100644 --- a/q166/random_k4/random_2.md +++ b/q166/random_k4/random_2.md @@ -1,1186 +1,1349 @@ - TABLE OF -CONTENTS - -Is Graduate School Right for You?............................. - -4 - -What to Expect from Graduate School .................... - -5 - -Why UTRGV?................................................................ - -6 - -Institutional Accreditations ...................................... - -Graduate Programs ..................................................... - -Graduate Admissions ................................................ - -Research Centers & Key Research Areas ................. - -Costs & Financing ....................................................... - -Important Contact Information ............................... - -8 - -10 - -12 - -14 - -16 - -18 - -2 - -3 - - IS GRADUATE -SCHOOL -Right for You? -Right for You? - -Your undergraduate education gave you a strong foundation of knowledge and skills to start -your career. These days, however, to become an expert in your field, a graduate education is -often required. - -Specialized skills that you’ll learn and practice in graduate school are the key to career growth, -a better income, and unlocking your full potential. More and more, employers look for hiring -candidates to hold a graduate degree that demonstrates their expertise. - -Even if you’re currently satisfied with your career, you might have a passion for lifelong -learning or want to take a deeper dive into a particular subject. Investing in self-development is -never a bad decision, and a graduate degree can create surprising benefits for you in the future. - -As a graduate student, you’ll connect with supportive faculty who are leaders in their fields, -as well as other motivated graduate students who share similar goals and interests. You have -the flexibility to choose the pathway that’s right for you, including research, accelerated online -master’s programs, on-campus master’s or doctoral programs, and certificates. - -WHAT TO -EXPECT FROM -Graduate School? -Graduate School? - -• You can expect to be more responsible for your success than you were as a an - -• You’ll need to plan ahead to ensure you don’t miss out on required classes. - -undergraduate. You may even help with research or teaching. - -Since some classes are only offered, for example, in the spring, you will prepare -a graduation plan to guide you towards on-time completion of your program. - -• You’ll take rigorous classes alongside driven classmates like yourself. Being in - -graduate school raises the bar for expectations. You’ll need to maintain at least -a 3.0 GPA. - -• Unlike large undergraduate classes with hundreds of students, graduate - -classes are smaller. You will be on a first-name basis with your professors, who -will be both instructors and valuable mentors. - -• As an undergraduate, you may have assisted graduate students or faculty with -their research. As a graduate student, depending on your program, you’ll be -expected to carry out your own research, present it, and publish it. - -• If you’re taking classes on campus, you can expect most of them to be in the -evening. Many graduate students have jobs and families, so most on-campus -graduate classes are offered after 5 pm, twice a week. Online students need to -prioritize and allocate time for classwork in their schedules. - -• You can expect to be proud of your hard work and for pursuing your passion. - -Get ready to positively transform your community and the world. - -4 - -5 - - UTRGV? - -WhyWhy - -A high-quality education at an affordable price - -Graduate school at UTRGV is the perfect fit for - -ambitious leaders, dreamers, researchers, educators, - -and creators. - -We welcome you to join our vibrant community, -whether on campus or online. This is your opportunity -to leap forward in your career with an excellent -graduate education that’s both rigorous and affordable. - -Pursuing a graduate degree is a commitment to your -future, and we understand that our students have -different learning styles and real-world responsibilities. -We have options to suit your lifestyle and give you the -flexibility to choose between 100% online or on-campus -programs. All students receive the same high level of -training and support from faculty and staff. - -UTRGV offers unique research and resources through -the STARGATE partnership with SpaceX, the Coastal -Studies Laboratory on South Padre Island, and UT -Health RGV clinics from our School of Medicine. - -The Valley itself has a low cost of living that’s 10.2% -below the national average, with easy access to the -beautiful Gulf Coast. - -Demand for trained professionals in the Rio Grande -Valley, the United States, and around the world -continues to grow, and UTRGV equips graduate students -with the necessary competitive edge. - -Established in - -2015 - -4th Largest - -institution in -the UT System - -Enrollment Numbers 2020: -Graduate - -Overall - -32,618 - -5,122 - -UTRGV RANKINGS -UTRGV -RANKINGS -19th -6th - -1st - -In net price among -national universities by -Washington Monthly (2018) - -In the nation for top master’s -universities by Washington -Monthly (2019) - -Best online college in Texas by -OnlineColleges.com (2019) - -3rd - -Most affordable university in -America by BestValueSchools.com -(2018) - -1st - -4th - -19th - -79th -6th - -Best college in Texas by -BestColleges.com (2018) - -In net price among -national universities by -Washington Monthly -(2018) - -In the nation for top -master’s universities by -Washington Monthly -(2019) - -Best online college -Among 300+ national universities -in Texas by -by Washington Monthly (2018) -OnlineColleges.com -(2019) - -3rd - -in the nation among public -institutions for least student -Most affordable -debt according to U.S. News -university in America -& World Report (2020) -by BestValueSchools.com -(2018) - -1st - -4th - -3rd -79th - -Best college in Texas -by BestColleges.com -(2018) - -America’s Best Bank for the -Buck Colleges: South by -Washington Monthly (2020) - -Among 300+ national -universities by -Washington Monthly -(2018) - -QUALITY . FLEXIBILITY . AFFORDABILITY - -• According to the U.S. Bureau of Labor Statistics, 101 occupations require a graduate degree for entry. - -• The number of jobs requiring a master’s degree is projected to increase by 17% between 2016 and 2026. - -• Students who obtain master’s degrees earn, on average, $12,000 more in entry-level positions. - -• UTRGV offers graduate programs that align with the fastest growing careers. - -% change -2018-2028 - -Medium annual wage (2018) - -2018 Fastest Growing Occupations - -Occupational therapy assistants - -Information security analysts - -Physician assistants - -Statisticians - -Nurse practitioners - -Speech-language pathologists - -Physical therapist assistants - -Genetic counselors - -Mathematicians - -Operations research analysts - -Software developers, applications - -Health specialties teachers, postsecondary - - 33.1 - - 31.6 - - 31.1 - - 30.7 - - 28.2 - - 27.3 - - 27.1 - - 27.0 - - 26.0 - - 25.6 - - 25.6 - - 23.2 - - $60,220 - - $98,350 - - $108,610 - - $87,780 - - $107,030 - - $77,510 - - $58,040 - - $80,370 - - $101,900 - - $83,390 - - $103,620 - - $97,370 - - $44,630 - - $50,090 - - $87,930 - - $60,280 - - $63,120 - - $73,490 - -Substance abuse, behavioral disorder, and mental health counselors - - 22.5 - -Marriage and family therapists - -Physical therapists - -Respiratory therapists - -Market research analysts and marketing specialists - -Nursing instructors and teachers, postsecondary - - 22.3 - - 21.9 - - 20.8 - - 20.4 - - 20.0 - -Bureau of Labor Statistics, U.S. Department of Labor, Occupational Outlook Handbook, [Fastest Growing Occupations], on the Internet -at [https://www.bls.gov/ooh/fastest-growing.htm] (visited [October 2019]).”” * U.S. Bureau of Labor Statistics - -55+ - -ON-CAMPUS AND ONLINE -MASTER’S PROGRAMS -16-week-long semesters - -Full-time students (9 hours/semester) -can finish in as little as 2 years - -25 - -ACCELERATED ONLINE - -PROGRAMS -7-week-long modules - -Full-time students (6 hours/module) -can finish in as little as 1 year - -6DOCTORAL PROGRAMS - -40CERTIFICATE PROGRAMS - -SCHOLARSHIPS & ASSISTANTSHIPS -AVAILABLE - -$0 - -APPLICATION FEE - -Except: Occupational Therapy, Physician Assistant -Studies, Communication Sciences and Disorders, -Psychiatric Mental Health Nurse Practitioner, and -Physician Assistant Studies - Bridge. - -6 - -7 - - INSTITUTIONAL -Accreditations -Accreditations - -The University of Texas Rio Grande Valley (UTRGV) is accredited by the Southern Association of Colleges and - -Schools Commission on Colleges to award bachelor’s, master’s, and doctoral degrees. - -Individual Program Accreditations and Recognitions - -• Robert C. Vackar College of Business & Entrepreneurship - AACSB International, The Association to Advance - -• Department of Physician Assistant Studies - Accreditation Review Commission on Education for the - -Collegiate Schools of Business - -Physician Assistant (ARC-PA) - -• Educational Leadership (EdD) and Curriculum and - -Instruction (EdD) - -• School of Rehabilitation Services & Counseling - Council for Accreditation of Counseling and Related - - Admitted to the Carnegie Project on the Education - -Educational Programs - -Doctorate (CPED) - -• Social Work (MSSW) - Council on Social Work Education (CSWE) - -• Music (MM) - National Association of School of Music (NASM) - -• Department of Occupational Therapy - Accreditation Council for Occupational Therapy - -Education (ACOTE) - -• Department of Communication Sciences & Disorders - Council on Academic Accreditation in Audiology and -Speech-Language Pathology of the American Speech- -Language-Hearing Association - -• Master’s Degree in Nursing and Post-Graduate - -Psychiatric Mental Health Nurse Practitioner Certificate - - Commision on Collegiate Nursing Education - -• Counseling (MEd) - Council for Accreditation of Counseling and Related - -Educational Programs - -8 - -9 - - GRADUATE PROGRAMS - -COLLEGE OF EDUCATION -AND P-16 INTEGRATION -utrgv.edu/cep -Bilingual Education (MEd)2 -Board Certified Behavior Analyst - -(Certificate) - -Certificado Avanzado en Educación - Bilingüe 2 -Counseling (MEd) -Curriculum & Instruction (MEd) - Content Specialization in English (MEd)3 - Content Specialization in Mathematics (MEd)3 - Content Specialization in Elementary Math and - -Science Education (MEd)3 - - Content Specialization in Mathematics - -Education (MEd)3 - -Curriculum and Instruction (EdD) -Digital Literacy Leader (Certificate)2 -Early Childhood Education (MEd) -Educational Leadership (EdD) -Educational Leadership (MEd)4 -Educational Technology (MEd)3 -E-Learning (Certificate)3 -Higher Education Teaching (Certificate)3 -Online Instructional Designer (Certificate)3 -Principal Credential Certification (6 Credit Hours) - -(Certificate) 4 - -Principal Credential Certification (21 Credit Hours) -Reading and Literacy (MEd) 2 -Reading Specialist Certification (Certificate) 2 -School Psychology (MA) -Secondary Bilingual Education (Certificate) 2 -Special Education (MEd)3 -Superintendent Certification (Certificate) -Teacher Leadership (MEd)4 -Technology Leadership in Education - -(Certificate)3 - -TxVSN Digital Literacies (Certificate)2 - -COLLEGE OF ENGINEERING -AND COMPUTER SCIENCE -utrgv.edu/coec -Civil Engineering (MS) -Computer Science (MS) -Electrical Engineering (MSE) -Engineering Management (MS) 4 -Information Technology (MS) -Manufacturing Engineering (MSE) -Materials (Certificate) -Mechanical Engineering (MSE) -Mechanics and Design (Certificate) -Thermal Fluid Science (Certificate) - -COLLEGE OF FINE ARTS -utrgv.edu/cofa -Art (MFA) -Interdisciplinary Studies (MA) - Concentration in Art History -Creative Writing (MFA) -Design (Certificate) -Latin American Art History (Certificate) -Music (MM) - -COLLEGE OF HEALTH PROFESSIONS -utrgv.edu/coha -Bilingual Counseling (Certificate)2 -Bioethics (MS)2 -Clinical Rehabilitation Counseling (MS) -Clinical Laboratory Sciences (Certificate)3 -Communication Sciences and Disorders (MS) -Cooperative Pharmacy Program (with UT Austin) -Exercise Science (MS) -Health Sciences (MS)3 - Concentration in Clinical Laboratory Sciences3 - Concentration in Healthcare Administration3 - Concentration in Health Professions Education - -and Technology3 - - Concentration in Healthcare Informatics3 - Concentration in Nutrition3 -Health Science Nutrition (Certificate)3 -Healthcare Informatics (Certificate)3 -Kinesiology (MS)3 -Occupational Therapy (MS) -Physician Assistant Studies (MPAS) -Physician Assistant Studies - Bridge - Program (MPAS)2 -Rehabilitation Counseling (PhD) - -ROBERT C. VACKAR COLLEGE OF -BUSINESS & ENTREPRENEURSHIP -utrgv.edu/cobe -Accountancy (MAcc)4 -Advanced Business Administration - -(Certificate)4 - -Business Administration (MBA) - General4 - Specialization in Accounting (MBA)4 - Specialization in Business Analytics (MBA)4 - Specialization in Entrepreneurship (MBA)4 - Specialization in Financial Planning (MBA)4 - - Specialization in Health Care Administration - -(MBA)4 - - Specialization in International Trade (MBA)4 - Specialization in Marketing (MBA)4 -Business Administration (PhD) -Business Administration in Spanish (MBA)3 -Business Analytics (Certificate) -Business Analytics (MS)3 -Entrepreneurship (Certificate) -Healthcare Administration Leadership - -(Certificate) - -International Sustainability Management - -(Internal Certificate)5 - -User Experience and User Interface Design (UX/UI) - -(Certificate)3 - -SCHOOL OF NURSING -utrgv.edu/nursing -Family Nurse Practitioner (MSN) -Nursing Administration (MSN)3 -Nursing Education (MSN)3 -Psychiatric Mental Health Nurse - Practitioner (Certificate)2 - -SCHOOL OF SOCIAL WORK -utrgv.edu/social work -Social Work (63-Hour) (MSSW) -Social Work (MSSW)4 - Concentration in Administration and - -Community Practice (Advanced Standing) -(MSSW)4 - - Concentration in Direct Practice with Latino -Individuals, Families, and Groups (Advanced -Standing) (MSSW)4 - -COLLEGE OF LIBERAL ARTS -utrgv.edu/cla -Advanced Placement Spanish Literature - -(Certificate)2 - -Board Certified Behavior Analyst - -(Certificate) - -Clinical Psychology (MA, PhD) -Communication (MA) -Communication Training and Consulting - -(Certificate) - -Court Interpreting (Certificate)2 -Criminal Justice (MS)4 -Disaster Studies (MA) -English (MA) -English as a Second Language (MA) - -COLLEGE OF LIBERAL ARTS cont. -Experimental Psychology (MA) -Gender and Women’s Studies (Certificate) -Healthcare Interpreting (Certificate)2 -History (MA) -Interdisciplinary Studies (MA) - Concentration in Anthropology - Concentration in English - Concentration in History - Concentration in Mexican American Studies -Literary Translation (Certificate)2 -Localization and Audiovisual Translation - -(Certificate)2 - -Media Relations and Strategic - Communication (Certificate) -Mexican American Studies (Certificate) -Political Science (MA) -Public Affairs (MPA)4 -Secondary English Language Arts (Certificate) -Sociology (MS) -Spanish (MA) -Spanish Translation and Interpreting - -(MA, Certificate) 2 - -COLLEGE OF SCIENCES -utrgv.edu/cos -Agricultural, Environmental and Sustainability - -Sciences (MS) - -Applied Statistics and Data Science (MS) - -Biochemistry and Molecular Biology (MS) - -Biology (MS) - -Chemistry (MS) - -Cooperative PhD in Physics (with UT Arlington) - -Mathematics (Accelerated BS/MS - - 5-Year Program, MS) - -Ocean, Coastal, and Earth Sciences (MS) - -Physics (MS, PhD) -Interdisciplinary Studies (MS) - - Concentration in Science and Technology - -Instructional method is traditional -(face-to-face) unless otherwise noted. -2 Online (16-week Semesters) -3 Accelerated Online (7-Week Courses) -4 On-Campus & Accelerated Online Options -5 Online (16-week semester) and Accelerated - Online Options - - Graduate -Graduate -ADMISSIONS - -While most of our graduate programs have rolling admissions, ideally, you should begin the application -process several months in advance to ensure that you submit all documents on time and to have the best -chance for scholarships and assistantships. -Admissions Checklist: - -Apply online: utrgv.edu/gradapply -Get a certified English translation of your educational records - -Send your official college or university transcripts - -Submit official English Proficiency exam scores. (TOEFL, IELTS or Duolingo English Test) -Send your official GRE or GMAT scores (if applicable)*Note: Due to the impacts of the coronavirus pandemic on -testing centers across the world, we are temporarily suspending the GRE and GMAT test score requirements for - -most graduate programs until Fall 2021. -Other documents requested by the program (varies by program). Review requirements at utrgv.edu/grad -The application fee is waived for all graduate programs, except for Physician Assistant Studies, - -Occupational Therapy, Communication Sciences & Disorders, Physician Assistant Studies Bridge Program, - -and Psychiatric Mental Health Nurse Practitioner - -*If you will be pursuing your degree as an *International Student, please complete the following steps with International -Admissions & Student Services to finalize your admissions process.*International Student is defined as someone -who is not a U.S. citizen or U.S. resident and has a visa status or will require one to pursue their studies in a U.S. higher -education institution. - -Submit financial documentation - -Submit immigration documents - -To check the status of your application and review the final decision, visit the self-service center at utrgv.edu/selfservice. - -FAQ FREQUENTLY ASKED QUESTIONS - -Does my program require the GRE or GMAT? -Not all graduate programs require an entrance exam. -Visit utrgv.edu/graduate to find out -if your program requires one. - -*Note: Due to the impacts of the coronavirus pandemic - -on testing centers across the world, we are temporarily - -suspending the GRE and GMAT test score requirements for - -most graduate programs until Fall 2021. - -Where can I register for the GRE? What is UTRGV’s GRE -code? -You can register to take the GRE at ets.org/gre and send us -your scores using the GRE Code 6570. - -Where can I register for the GMAT? What is UTRGV’s -GMAT code? -You can register to take the GMAT at mba.com and send us -your scores using the GMAT Code 9P3. - -I attended a legacy institution (UT Pan American, UT -Brownsville, or Texas Southmost College). -Do I need to request those transcripts to be sent to -UTRGV? -No, we have access to all transcripts from our legacy -institutions and can upload them to your application. - -I started a graduate application. How do I log back -into my application? -Visit utrgv.edu/selfservice to log in with your username and -password that you previously -created. Do not create another application. - -Where do I send my official transcripts? -You can get them sent electronically to gradapps@utrgv.edu -or mail them to: The University of Texas Rio Grande Valley -The Graduate College -Marialice Shary Shivers Bldg. 1.158 -1201 W. Univesity Dr., Edinburg, TX 78539-2999 - -What are the minimum English Proficiency Exam -scores? -TOEFL minimum: 550 paper based, 79 internet based -IELTS minimum: 6.5 -Duolingo English Test score minimum of 105 -(Applicants with a score of 105 or above may also be -required to participate in a Zoom or Skype interview before -receiving a final decision.) - -How long is a TOEFL or IELTS score valid? -The TOEFL and the IELTS exam scores have a two year -validity period. - -Do you have exemptions for the English Proficiency -Exam (TOELF or IELTS)? -To view a list of exempt countries visit UTRGV.edu/grad - -Do I need to get my transcripts evaluated by an -evaluation agency? -We accept evaluations from the Foreign Credentials Services -of America (FCSA) or the World Education Services (WES). -However, if you have not had your transcript evaluated -by FCSA or WES, you can send us your transcripts to be -evaluated by UTRGV. - -How many credit hours do I need to be enrolled if I -have a F1 visa? -Nine credit hours per semester is the minimum; please -contact the International Admissions and Student Services -Office at international@utrgv.edu for -more details. - -Is proof of health insurance required for -international students? -Medical insurance is mandatory for any enrolled -international students on non-immigrant visa J-1, J-2, F-1, -F-2. An insurance waiver is available for students with -medical coverage. Although, medical insurance is strongly -recommended, F-1/F-2 Mexican students are exempt from -this health insurance requirement. For information on -waivers, contact the International Admissions and Student -Services office at international@utrgv.edu - -Who should I contact to assist me with inquires on -the student visa process? -International Admissions and Student Services -1-(956) 882-7092 | 1- (956) 665-2922 -utrgv.edu/iass | international@utrgv.edu - -12 - -13 - - Research -CENTERS - -At UTRGV, we have 15 research centers and -institutes - -• Center for Advanced Radio Astronomy (CARA) -• Center for Excellence in STEM Education -• University Transportation Center for Railway Safety (UTCRS) -• Texas Manufacturing Assistance Center (TMAC) -• South Texas Industrial Assessment Center -• Spacecraft Tracking and Astronomical Research into Gigahertz - Astrophysical Transient Emission (STARGATE) -• Nanotechnology Center of Excellence -• The Institute for Neuroscience -• Center for Vector-Borne Disease (CVBD) -• Coastal Studies Laboratory -• Center for Sustainable Agriculture & Rural Advancement (SARA) -• Center for Gravitational Wave Astronomy (CGWA) -• The South Texas Diabetes and Obesity Institute (STDOI) -• Dr. Cristina V. Torres Memorial Astronomical Observatory -• Center for Innovation and Commercialization - -KEY RESEARCH -Areas - -• Sustainable Agriculture - - and Food Systems - -• Diabetes, Obesity, and Human - - Genetics - -• Gravitational Wave and - -• Advanced Renewable - - Advanced Radio Astronomy - - Energy - -• Space Science Technology - -• Cell and Molecular - - (STARGATE) - -• Material Science and - - Engineering - -• Nanoscience and - - Nanotechnology - -• Optics - -• Coastal Resilience - - and Disaster Studies - -• Ocean Intelligence - - Biology - -• Biophysics - -• Computational Biology - -• Applied Statistics and - - Data Science - -• Mathematics (nonlinear - - waves, fluid dynamics, - - solitons, integrable - - systems) - -14 - -15 - - Costs & -FINANCING - -Scholarships* -UTRGV has several scholarship opportunities that use the same scholarship application. Scholarships are highly -competitive, so be sure to apply early. Visit utrgv.edu/applyscholarships for more information and to start your -application. - -Graduate College Scholarships - -Each semester the Graduate College may offer a variety of scholarships, dependent upon funding. To view - -what scholarships are currently being offered visit utrgv.edu/gradscholarships. - -Assistantships - -Assistantships give students the opportunity to work on campus (up to 20-hour work weeks) and are - -based on college/institutional demands and funds. Three types of assistantships are offered: - -• Teaching - -• Research - -• Graduate - -Please see your program director about the different assistantships available in your - -field of study. Visit careers.utrgv.edu for additional graduate assistantship opportunities. - -Texas Resident Tuition Rate -The Texas resident tuition rate may be granted to students who receive a teaching or -research assistantship in their discipline or who are awarded a competitive scholarship -of over $1,000. - -Payment Options -UTRGV provides payment plans to assist with the cost of tuition and fees. Eligible students may elect one of two payment -assistance options during the term of registration: - -• Installment Plan: This plan is available to any student unable to pay the full amount or partial amount due by the - university due dates. Installment plans consist of 4 payments. - Note: This is not available for summer sessions and accelerated online students. - -• Emergency Tuition and Fee Loan: This plan defers the payment of qualified tuition and fee charges to a future date. - -You can set up payment plans online through your my.utrgv.edu account within the ASSIST application. For additional -information, contact student_billing@utrgv.edu or call 956-665-7824 or 956-665-2191. - -Estimated Cost of Attendance: Fall 2020 - Summer 2021 - -Online and On-Campus (16-Week Semester) Programs - -Residency - -Texas Resident - -Non-Resident/International - -Per 3-Credit-Hour Course - -Per 9-Credit-Hour Semester - -$1,481.39 - -$2,708.39 - -$3,894.17 - -$7,575.17 - -Accelerated Online (7-Week Module) Programs - -Program - -Per 3-Credit-Hour Course - -Total Cost - -Technology Leadership in Education (Certificate) - -$1,374.99 - -E-Learning (Certificate) - -$1,374.99 - -Online Instructional Designer (Certificate) - -$1,374.99 - -Principal Certification (Certificate) - -$1,374.99 - -$4,125 - -$5,500 - -$5,500 - -$9,625 - -Curriculum & Instruction (MEd) -Educational Leadership (MEd) -Educational Technology (MEd) -Special Education (MEd) - -Social Work 33-Hour (MSSW) - -Social Work (MSSW) (63 Hr) - -Accountancy (MACC) - -Business Analytics (MS) - -Health Sciences (MS) - -$1,374.99 - -$13,750 - -$1,333.32 - -$1,333.32 - -$1,416.66 - -$1,416.66 - -$1,333.32 - -$14,667 - -$27,999 - -$14,170 - -$15,584 - -$16,000 - -12 - -16 - -13 -17 - - Accelerated Online (7-Week Module) Programs - -Program - -Per 3-Credit-Hour Course - -Total Cost - -Health Sciences (Certificate) - -Kinesiology (MS) - -Nursing (MSN) - -Public Affairs (MPA) - -Criminal Justice (MS) - -Business Administration (MBA) - -Business Analytics (MS) - - International Sustainability Management -(Certificate) - -Entrepreneurship (Certificate - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,333.32 - -$1,416.66 - -$1,416.66 - -$1,416.66 - -$1,416.66 - -User Experience and User Interface Design - (UX/UI) (Certificate) - -$1,416.66 - -$8,000 - -$16,000 - -$16,000 - -$17,000 - -$16,000 - -$9,625 - -$15,583 - -$5,666 - -$5,666 - -$5,666 - -IMPORTANT CONTACT -Information -Information - -GRADUATE SCHOLARSHIPS -gradscholarships@utrgv.edu - -CONTINUING EDUCATION -continuinged@utrgv.edu - -RESIDENCE LIFE -home@utrgv.edu - -INTERNATIONAL ADMISSIONS -international@utrgv.edu - -LANGUAGE INSTITUTE -international@utrgv.edu - -18 - -19 - +Undergraduate +Brochure +2025 + +About HKU Business School +HKU Business School strives to nurture first-class business leaders and foster academic +and relevant research to serve the needs of Hong Kong, Mainland China and the rest of +the world in the new Asia-led economy. As Asia’s premier international business school, +we engage leading scholars from all corners of the globe and they instil in students global +CONTENTS knowledge with an Asian perspective. +With the highest proportion of non-local undergraduate students across all Faculties at +HKU, four of our undergraduate programmes ranked amongst the University’s Top 10 +programmes in 2024#, HKU Business School attracts the best and brightest students +from Hong Kong and beyond. +| About HKU Business School | P. 1 | | +| ------------------------- | ---- | --- | +HKU Business School is fully accredited by the European Quality Improvement Systems +| Explore Your Options | P. 4 | | +| -------------------- | ---- | --- | +(EQUIS) and the Association to Advance Collegiate Schools of Business (AACSB). +| Enrich Your Life | P. 16 | | +| ------------------------------------------------------------------------------------- | ----- | --- | +| 2025 JUPAS Admission | P. 36 | | +# Based on average JUPAS admission scores in the best 5 subjects in HKDSE +| 2025 Non-JUPAS Admission | P. 38 | | +| -------------------------- | ----- | ------ | +| 2025 Dual Degree Admission | P. 39 | | +| Develop Your Career | P. 40 | VISION | +| Freshmen Sharing | P. 41 | | +To be a leading, globally-impactful academic institution of business and +economics +| Our Global Presence | P. 44 | | +| ------------------- | ----- | --- | +| Useful Websites | P. 45 | | +UNIQUE PROPOSITION +Deeply rooted in Hong Kong +Fully engaged with China +Truly international +MISSION STATEMENT +To provide world-leading business and economics education that +nurtures dynamic and responsible leaders +To advance the frontiers of knowledge through innovative and +globally-significant research +To create positive change in the world by partnering with communities +01 + +| Dean’s Message | | | | | | | | | | | Greetings from | | | | +| -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | +Associate Dean (Undergraduate) +| | | | | | At HKU | Business | School, | we have | 11 | diverse | | | | | +| --- | --- | --- | --- | --- | -------------- | --------- | ----------- | --------- | ------ | -------- | --- | --- | --- | --- | +| | | | | | undergraduate | | programmes | and | three | dual | | | | | +Business education today is an evolving core that +| | | | | | bachelor’s | degrees | in | partnership | | with | | | | | +| --- | --- | --- | --- | --- | ----------- | -------- | --- | ------------ | --- | ----- | --- | --- | --- | --- | +builds upon learning from all academic disciplines and +world-renowned universities. Our new programmes +| | | | | | | | | | | | draws inspiration | from | both | successful and | +| --- | --- | --- | --- | --- | -------------- | ----------- | ---------- | ---- | --------- | --- | ------------------- | ----- | ----- | ---------------- | +| | | | | | in big data | analytics, | including | the | Bachelor | of | | | | | +unsuccessful business practices. Our undergraduate +| | | | | | Business | Administration | | in Accounting | | Data | | | | | +| --- | --- | --- | --- | --- | --------- | --------------- | --- | --------------- | --- | ----- | --- | --- | --- | --- | +curricula in six core academic areas equip students +Analytics, the Bachelor of Business Administration +with knowledge in diverse aspects of business and +(Business Analytics), and the Bachelor of Science in +economics, and prepare them to be creative leaders +Marketing Analytics and Technology, are tailored to +of tomorrow with global perspectives. +| | | | | | meet the | growing | market | demand | for | skilled | | | | | +| --- | --- | --- | --- | --- | --------------- | -------- | --------- | ---------------- | ---- | -------- | ----------------------------------------------------- | --- | --- | --- | +| | | | | | professionals | in | business | data analytics | | and AI | A hallmark of our School that separates us from the | | | | +| | | | | | transformation. | | | | | | others is our truly multi-disciplinary approach. HKU | | | | +I'm thrilled to introduce you to our dynamic and +| | | | | | | | | | | | Business | School’s tremendous | | resources offer | +| ----------------- | --------- | ----------- | --- | ------ | -------------------------------------------------- | --- | --- | --- | --- | --- | --------- | --------------------- | --- | ----------------- | +| forward-thinking | academic | community, | | where | We believe in learning by doing. Our partnerships | | | | | | | | | | +I hope that you will share my enthusiasm for the +exciting opportunities and transformative students with a wide array of choices to combine +with over 300 institutions and corporations in more programmes and my admiration for the people who +studies in economics and finance, accounting and +experiences await you. than 40 countries provide you with unparalleled have contributed to their development. I look forward +finance, accounting and data analytics, business and +experiential and enrichment opportunities. Whether to having you as a member of the HKU Business +As a part of the University of Hong Kong, which law, business and data analytics, and international +proudly ranks 17th globally in the 2025 QS World it's internships, exchange programmes, or industry School family and a future leader in the world of +business and global management, as well as dual +projects, these experiences will sharpen your skills business. +University Rankings, HKU Business School stands out degrees in collaboration with world-renowned +and broaden your mindset, preparing you to tackle +| as a beacon of excellence in business and economics | | | | | | | | | | | universities. | | | | +| ---------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | +real-world challenges and build a successful career. Warm regards, +in Asia. As the youngest Faculty of the University, we +Our faculty members engage you with their inspiring +| have been growing rather quickly to become the top | | | | | Joining HKU Business School means becoming part of | | | | | | | | | | +| --------------------------------------------------- | --- | --- | --- | --- | --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +teaching and bring their impactful research into the +| business | school in Asia | in faculty | strength | and | a community that supports your growth and success. | | | | | | | | | | +| --------- | ----------------- | ------------ | --------- | ---- | --------------------------------------------------- | --- | --- | --- | --- | --- | ----------- | ----------- | ----------- | ------------ | +| | | | | | | | | | | | classroom. | They open | your mind | to diverse | +research productivity. +You’ll be surrounded by brilliant minds, engaging in +| | | | | | | | | | | | perspectives, | sharpen your | curiosity | to explore | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | -------------- | ---------- | ------------ | +stimulating discussions, and working on projects that +Our mission is to inspire and develop the next innovative ideas, and instil in you an appetite for +can make a real impact. The friendships and networks +generation of dynamic and responsible business acquiring new knowledge and extraordinary +you build here will last a lifetime and open doors to a +leaders. We offer a vibrant and challenging Professor Derek Chan +experiences throughout your academic journey and +myriad of career opportunities. Associate Dean (Undergraduate) +environment where you’ll be encouraged to push career development. Our students are among the +your limits, think critically, and innovate. Our team of If you're ready to embark on an exciting journey of brightest and most creative of their peers. Our alumni +| leading | scholars from | around | the | world and | | | | | | | | | | | +| -------- | --------------- | ------- | ---- | ----------- | --------------------------------------------------- | --- | --- | --- | --- | --- | ---------------------------------------------------------- | --- | --- | --- | +| | | | | | learning and personal development, I invite you to | | | | | | have gone on to forge brilliant careers in all sectors of | | | | +experienced and dedicated teachers brings a wealth explore what HKU Business School has to offer. I look the local, regional and international business +of knowledge and a global perspective to our forward to welcoming you to our community, where community, and formed a global network that you will +classrooms. Our wide range of top-quality degree future leaders are nurtured and dreams take flight. join for the rest of your life. +| programmes | is designed | to | equip | you with | | | | | | | | | | | +| ----------- | ------------- | --- | ------ | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Warm regards, +| cutting-edge | knowledge | and | technical | skills to | | | | | | | | | | | +| ------------- | --------------- | ------------------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| navigate | and excel in | the fast-changing | | global | | | | | | | | | | | +business landscape. +Professor Hongbin Cai +Dean +Chair of Economics +02 03 + +Explore Your Options HKU – Sciences Po Dual Degree Programme +A Wide Selection of Programmes Students will spend the first two years at Sciences Po in France and the third and fourth years +at HKU. Programmes at Sciences Po are taught in English (no French requirements for being +admitted) with intensive French language classes for all non-francophone students. +6755 Upon successful completion of the dual degree programme, students will be awarded the +Bachelor of Business Administration [BBA] +Bachelor of Business Administration (BBA) degree from HKU and a Bachelor’s degree from +Sciences Po. +Separate application is required for applying to the dual degree programme, in addition to the +The BBA programme combines strong functional training with intensive exposure to application under the JUPAS or International/Non-JUPAS Admissions Scheme. +| communication | skills, computer | applications, | and other social science | and/or science | +| -------------- | ------------------ | -------------- | ---------------------------- | ---------------- | +subjects. Students can choose to major in the following area(s): +Find out more: +https://admissions.hku.hk/programmes/university-collaborative- +programmes/hku-sciences-po-dual-degree-programme +| Entrepreneurship, | Finance | Human Resource | Information Systems | Marketing | +| ------------------ | ------- | -------------- | -------------------- | --------- | +Management HKU – University of British Columbia Dual +| Design and | | | and Analytics | | +| ---------- | --- | --- | ------------- | --- | + Innovation* +Degree Programme +BBA Curriculum Structure +Students will spend the first and final years at HKU and the second and third years at the +Credits University of British Columbia (UBC) in Canada, and graduate with a Bachelor of Business +Components +Administration from HKU and a Bachelor of Commerce from UBC upon successful completion +| A | University Graduation Requirements | | | 54 | +| --- | ---------------------------------- | --- | --- | --- | +of the programme. +| | Language Enhancement Courses | | | 18 | +| --- | ---------------------------- | --- | --- | --- | + Common Core Curriculum 36 Separate application is required for applying to the dual degree programme, in addition to the + Non-Credit Bearing Courses as required by the University - application under the JUPAS or International/Non-JUPAS Admissions Scheme. +| B | BBA Major | | | 78 | +| --- | -------------------- | --- | --- | --- | +| | Faculty Core Courses | | | 24 | +Find out more: +| | Disciplinary Core Courses / Disciplinary Electives | | | 48 | +| --- | -------------------------------------------------- | --- | --- | --- | + Capstone Course 6 https://admissions.hku.hk/programmes/university- +https://ug.hkubs.hku.hk/programme/bba collaborative-programmes/hku-university-of-british-columbia- +| C | Free Electives / Optional Major / Optional Minor(s) | | | 108 | +| --- | --------------------------------------------------- | --- | --- | --- | +dual-degree-programme +Total 240 +* Candidates must undergo a selection process arranged by the Programme Coordinator for declaring the EDI major. +Find out more: +04 05 + +Bachelor of Business Administration in +Accounting Data Analytics [BBA(ADA)] +Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares students for a +professional qualification in accounting. It provides students with an option to focus on data +analytics aspect in accounting instead of having a minor in finance. It is an excellent choice for +students who are interested in STEM with accounting and business applications and wish to +pursue their career in the relevant fields. The intensive accounting training enables +BBA(ADA) students to distinguish themselves from students of the other data analytics / +STEM programmes. Students will be equipped with comprehensive accounting and business +knowledge that prepares them to obtain professional qualifications in accounting. They will +also acquire computer programming and data analytics skills to meet the needs of the +accounting and business professions in today’s digital transformation economy. It is +accredited as an approved accountancy programme by: +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia +:erom +tuo +dniF +ada-abb/emmargorp/kh.ukh.sbukh.gu//:sptth +Bachelor of Business Administration in +Accounting and Finance [BBA(Acc&Fin)] +The BBA(Acc&Fin) programme prepares students for a professional qualification in +accounting and provides the broad-based education in accounting and finance which is +necessary to progress toward a leadership role in the financial sector. Students are required +to take professional core in accounting and a minor in finance, with an option to second major +in finance. Tailor-made professional preparation programmes and internship opportunities +with renowned accounting firms and financial institutions are available for talented students. +It is accredited as an approved accountancy programme by: +The Hong Kong Institute of Certified Public Accountants (HKICPA) +The Association of Chartered Certified Accountants (ACCA) +CPA Australia +https://ug.hkubs.hku.hk/programme/bba-acc-and-fin +Find +out +more: +6781 6781 +BBA(Acc&Fin) Curriculum Structure +Components Credits +A University Graduation Requirements 54 +Language Enhancement Courses 18 BBA(ADA) Curriculum Structure +Common Core Curriculum 36 +Non-Credit Bearing Courses as required by the University - Components Credits +B Professional Core in Accounting 108 A University Graduation Requirements 54 +Faculty Core Courses 24 Language Enhancement Courses 18 +Disciplinary Core Courses 78 Common Core Curriculum 36 +Capstone Course 6 Non-Credit Bearing Courses as required by the University - +C Minor in Finance* 36 B Professional Core in Accounting Data Analytics 156 +Disciplinary Core Courses 12 Faculty Core Courses 24 +Disciplinary Electives 24 Disciplinary Core Courses 114 +Disciplinary Electives 12 +D Free Electives / Optional Minor 42 Capstone Course 6 +Total 240 C Free Electives 30 +Total 240 +*Students may opt to pursue a Major in Finance: +Components Credits +A University Graduation Requirements 54 +B Professional Core in Accounting 108 +C Major in Finance 78 +Disciplinary Core Courses 30 +Disciplinary Electives 18 +Free Electives 24 +Capstone Course 6 +Total 240 +06 07 + +| | 6793 | | 6808 | | | +| --- | ---- | --- | ---- | --- | --- | +Bachelor of Business Administration Bachelor of Business Administration (Law) +(Business Analytics) [BBA(BA)] and Bachelor of Laws [BBA(Law)&LLB] +The BBA(BA) programme is crafted to meet the growing industry demand for talents in the Jointly offered by the HKU Business School and the Faculty of Law, the BBA(Law)&LLB +business analytics field. Its curriculum offers students a comprehensive coverage of both integrated double-degree programme blends two strategic disciplines into a single +technical and managerial skill sets. Students will learn a wide spectrum of knowledge from programme for the bright and highly motivated. It is an excellent choice for students seeking +the disciplines of information technology, data science, business statistics and management. a broader education with parallel concentration on legal and business studies. Graduates of +Core courses include programming language, database management, introduction to the programme will be awarded two degrees — BBA(Law) by the HKU Business School, and +business analytics, management information systems, decision and risk analysis, etc. They LLB by the Faculty of Law. After successful completion of the LLB programme, students may +help students build the foundations for advanced topics in senior years. Students also have apply for admission to the Postgraduate Certificate in Laws (PCLL) programme provided that +the option to choose from analytics technique courses and business application courses. The they meet the relevant admission requirements. +aim is to provide students with state-of-the-art experiences in artificial intelligence, big data, +The programme offers business stream and accounting stream to students in their business +social media, and other latest development. Such design helps train well-rounded students +studies. The accounting stream is accredited by: +who will be more adaptable in their career development. The programme is suitable for +students who have a passion for problem-solving through analytics, and who aim to develop +The Hong Kong Institute of Certified Public Accountants (HKICPA) +their career path in analytics. +The Association of Chartered Certified Accountants (ACCA) +| | BBA(BA) Curriculum Structure | | CPA Australia | | | +| --- | ---------------------------- | --- | ------------- | --- | --- | +BBA(Law)&LLB Curriculum Structure (Business Stream) +Credits +Components +| | | | Components | Credits | | +| ---------------------------------------- | ---------------------------------------------------------- | --- | ---------------------------------------------------------- | ------- | ------------------------------------------------- | +| | A University Graduation Requirements | 54 | | | | +| | Language Enhancement Courses | 18 | | | | +| | | | A University Graduation Requirements | 36 | | +| | Common Core Curriculum | 36 | | | | +| | | | Language Enhancement Courses | 12 | | +| | Non-Credit Bearing Courses as required by the University | - | | | | +| | | | Common Core Curriculum | 24 | | +| | | | Non-Credit Bearing Courses as required by the University | - | | +| | B Major in Business Analytics | 96 | | | | +| | Faculty Core Courses | 24 | | | | +| | | | B BBA Major | 78 | | +| | Disciplinary Core Courses | 42 | | | | +| | | | Faculty Core Courses | 24 | | +| | Disciplinary Electives (Analytics Techniques) | 12 | | | | +| | | | Disciplinary Core Courses / Disciplinary Electives | 48 | bll-dna-wal-abb/emmargorp/kh.ukh.sbukh.gu//:sptth | +| | Disciplinary Electives (Business Applications) | 12 | | | | +| | | | Capstone Course | 6 | | +| | Capstone Course | 6 | | | | +| | | | C Professional Core in Law | 156 | | +| https://ug.hkubs.hku.hk/programme/bba-ba | C Free Electives / Optional Major / Optional Minor(s) | 90 | | | | +| | | | Disciplinary Core Courses | 108 | | +| | | | Disciplinary Electives | 42 | | +Total 240 +| | | | Capstone Course | 6 | | +| --- | --- | --- | ---------------------------------- | --- | --- | +| | | | D Free Electives / Optional Minor | 30 | | +Total 300 +BBA(Law)&LLB Curriculum Structure (Accounting Stream) +Credits +Components +| Find out more: | | | | | :erom tuo dniF | +| -------------- | --- | --- | ------------------------------------- | --- | -------------- | +| | | | A University Graduation Requirements | 36 | | +| | | | B Professional Core in Accounting | 108 | | +| | | | Faculty Core Courses | 24 | | +| | | | Disciplinary Core Courses | 78 | | +| | | | Capstone Course | 6 | | +| | | | C Professional Core in Law | 156 | | +Total 300 +| 08 | | | | | 09 | +| --- | --- | --- | --- | --- | --- | + +| 6896 | | | | | 6767 | | +| ---- | --- | --- | --- | --- | ---- | --- | +Bachelor of Business Administration in Bachelor of Economics [BEcon] +International Business and Global +Management [BBA(IBGM)] +BBA(IBGM) is an intensive, premier programme in international business and global The BEcon programme trains students in economic theories, statistical tools, the application +management featuring a unique curriculum designed to evolve with and reflect the of theories to solve practical economic problems in trade, banking and the labour market, and +ever-changing demands of a globalised world. Exclusive elements such as overseas evaluation of the regional economic performance of China, Hong Kong and other economies. +exchange studies, a leadership boot camp, an Asia-Pacific study tour, an international field Students will be furnished with solid theoretical and empirical economic knowledge to +trip and global enrichment opportunities are strategically integrated into the curriculum, analyse and understand human behaviour and social interactions. The programme allows +whilst students are given the flexibility to pursue an in-depth inquiry into another specialised students a high degree of flexibility in planning their future career specialisation. +area of study as a second major. This cross-functional and personalised approach aims to +foster a selected group of outstanding students to become the world’s future leaders +BEcon Curriculum Structure +| equipped with | global perspectives, | cultural sensitivity, | social responsibility | and key | | | +| --------------- | ---------------------- | ----------------------- | ----------------------- | --------- | --- | --- | +Credits +| competencies to succeed in the international business environment. | | | | | Components | | +| ------------------------------------------------------------------ | --- | --- | --- | --- | ------------------------------------- | --- | +| | | | | | A University Graduation Requirements | 54 | +BBA(IBGM) Curriculum Structure Language Enhancement Courses 18 +| | | | | | Common Core Curriculum | 36 | +| --- | --- | --- | --- | --- | ------------------------ | --- | +Credits +Components Non-Credit Bearing Courses as required by the University - +| A | University Graduation Requirements | | | 54 | | | +| --- | ---------------------------------- | --- | --- | --- | --------------------------- | --- | +| | | | | | B Major in Economics | 78 | +| | Language Enhancement Courses | | | 18 | Faculty Core Courses | 24 | +| | Common Core Curriculum | | | 36 | Disciplinary Core Courses | 30 | + Non-Credit Bearing Courses as required by the University - Disciplinary Electives 18 +| | | | | | Capstone Course | 6 | +| --- | ----------------------------------------------------- | --- | --- | --- | ------------------------------------------------------ | --- | +| B | Major in International Business and Global Management | | | 96 | | | +| | Faculty Core Courses | | | 24 | | | +| | | | | | C Free Electives / Optional Major / Optional Minor(s) | 108 | +| | Disciplinary Core Courses | | | 54 | | | +| | Disciplinary Electives | | | 12 | | | +Total 240 +| | Capstone Course | | | 6 | | | +| --- | ---------------------------- | --- | --- | ----- | --- | --- | +| C | Second Major | | | 72-90 | | | +| D | Free Electives | | | 0-18* | | | +| E | Compulsory Overseas Exchange | | | - | | | +noceb/emmargorp/kh.ukh.sbukh.gu//:sptth +| | | | Total | 240 | | | +| --- | --- | --- | ----- | --- | --- | --- | +*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. +http://ibgm.business.hku.hk +Find out more: :erom tuo dniF +10 11 + +| 6767 | | 6860 | | | +| ---- | --- | ---- | --- | --- | +Bachelor of Economics and Finance Bachelor of Finance in Asset Management +[BEcon&Fin] and Private Banking [BFin(AMPB)] +BEcon&Fin is designed to meet the interests of students who are seeking a balanced study of The BFin(AMPB) programme aims to prepare students to enter the rapidly growing fields of +both economics and finance. The programme recognises the significance of economics both asset management and private wealth management in Hong Kong and Asia Pacific, or in +in its own right and as a foundation for the study of finance, and the significance of finance as banking and finance generally. During the first two years, students complete introductory +a career path for students who are keen on joining this profession. Students will receive solid courses in accounting, economics, mathematics, statistics, corporate finance, and +training in theories and applications of economics and finance. They will be equipped with the investments and portfolio analysis. In subsequent years, students take more specified and +skills to analyse human behaviour and social interaction, as well as to solve problems in the industry related courses including equity valuation and investment in specific asset classes +financial sphere. e.g. private equity, venture capital, hedge funds and digital assets, understanding of financial +statements and doing financial analysis, lending and credit in private wealth and corporate +BEcon&Fin Curriculum Structure +bank, financial regulations and compliance, current issues in finance, ESG in investment, +Components Credits institutional consulting and family office. The programme contains practical-oriented courses +taught by teachers with extensive experience in the finance industry and offers an array of +A University Graduation Requirements 54 career-oriented enrichment activities and networking opportunities with industry +| Language Enhancement Courses | 18 | | | | +| ------------------------------ | --- | --- | --- | --- | +practitioners. +| Common Core Curriculum | 36 | | | | +| ---------------------------------------------------------- | --- | --- | --- | --- | +| Non-Credit Bearing Courses as required by the University | - | | | | +BFin(AMPB) Curriculum Structure +| B Major in Economics and Finance | 96 | | | | +| --------------------------------- | --- | --- | --- | --- | +Credits +| Faculty Core Courses | 24 | Components | | | +| ------------------------------------------------------ | --- | ---------------------------------------------------------- | --- | --- | +| Disciplinary Core Courses | 42 | | | | +| | | A University Graduation Requirements | 54 | | +| Economics Disciplinary Electives | 12 | | | | +| | | Language Enhancement Courses | 18 | | +| Finance Disciplinary Electives | 12 | | | | +| | | Common Core Curriculum | 36 | | +| Capstone Course | 6 | | | | +| | | Non-Credit Bearing Courses as required by the University | - | | +| C Free Electives / Optional Major / Optional Minor(s) | 90 | | | | +| | | B Major in Asset Management and Private Banking | 96 | | +| | | Faculty Core Courses | 24 | | +Total 240 +| | | Disciplinary Core Courses | 48 | | +| --- | --- | --------------------------- | --- | --- | +| | | Disciplinary Electives | 18 | | +| | | Capstone Course | 6 | | +https://ug.hkubs.hku.hk/programme/becon-and-fin HKU - Peking University “Future Leaders” Dual +| | | C Free Electives / Optional Major / Optional Minor(s) | 90 | bpma-nfib/emmargorp/kh.ukh.sbukh.gu//:sptth | +| --- | --- | ------------------------------------------------------ | --- | -------------------------------------------- | +Bachelor’s Degree in Economics, Finance and +240 +Management Total +The programme combines the strengths of two leading business schools in Hong Kong and +Mainland China and is designed to nurture first-class business leaders and cultivate their +global perspective with intimate understanding of China. Students will spend the first two +years at HKU and the third and fourth years at Peking University (PKU), and be awarded a +Bachelor of Economics and Finance (BEcon&Fin) degree from HKU and a Bachelor of +Management (BMgmt) degree from PKU upon successful completion of the dual degree +programme. +Find out more: :erom tuo dniF +Separate application is required for applying to the dual degree programme, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. +Find out more: +https://ug.hkubs.hku.hk/admission/degree-programmes/hku-pku- +dual-degree +12 13 + +| 6846 | | | | | 6884 | | +| ---- | --- | --- | --- | --- | ---- | --- | +Bachelor of Science in Marketing Bachelor of Science in Quantitative +Analytics and Technology [BSc(MAT)] Finance [BSc(QFin)] +Marketing is critical to all businesses. It dictates whether a company will be successful. BSc(QFin) is a premium finance programme. Built on a solid foundation in economics and +Technology, big data, and digitization are fundamentally changing how marketers operate. finance, the programme equips students with the essential quant skills, including +The BSc(MAT) programme equips students with quantitative skills, technology know-how, mathematics, statistics, and programming, that are required in more advanced and +and business marketing acumen to pursue a career in the digital economy and modern sophisticated analyses, such as valuation models of structured products, risk management, +businesses. It targets both STEM students with an interest in business as well as business simulation and data analysis, financial engineering, and quantitative trading. +students interested in the technology space. Students will gain a solid foundation in +The programme goes beyond merely equipping students with quant skills. More importantly, +cutting-edge technical areas including computer programming, data science, and technology +it stresses on combining these skills with an understanding of the financial markets, the macro +by excelling in courses from the Faculty of Engineering and Faculty of Science. In addition, +environment, and the practical situations where financial managers work, with a balanced mix +students will master digital business, marketing and leadership through courses at the HKU +of quant and business courses. To nurture students’ soft skills and market acumen, the +Business School. This interdisciplinary approach will enable students to engage in a variety of +programme provides a wide range of value-enhancing activities, such as Round Tables with +jobs and functions such as the design and launch of new technology products, targeting +Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training workshops, +| consumers, digital | marketing, forecast | and predicting | markets, and | building a | | | +| -------------------- | --------------------- | ---------------- | -------------- | ------------ | --- | --- | +simulation trading workshops, and an international field trip to other international financial +technology-focused strategy. The programme aims to nurture future leaders in the field of +centres such as New York, London and Zurich. +technology and marketing who are able to use the latest technology, science, and data +analytics to support business and marketing decisions in the digital world. +BSc(QFin) Curriculum Structure +| BSc(MAT) Curriculum Structure | | | | | Components | Credits | +| ----------------------------- | --- | --- | --- | ------- | ------------------------------------- | ------- | +| Components | | | | Credits | | | +| | | | | | A University Graduation Requirements | 54 | +| | | | | | Language Enhancement Courses | 18 | +A University Graduation Requirements 54 Common Core Curriculum 36 + Language Enhancement Courses 18 Non-Credit Bearing Courses as required by the University - +| Common Core Curriculum | | | | 36 | | | +| ------------------------ | --- | --- | --- | --- | --- | --- | + Non-Credit Bearing Courses as required by the University - B Major in Quantitative Finance 96 +| | | | | | Faculty Core Courses | 24 | +| --- | --- | --- | --- | --- | ---------------------- | --- | +B Major in Marketing Analytics and Technology 96 Disciplinary Core Courses 54 +| Faculty Core Courses | | | | 24 | Disciplinary Electives | 12 | +| ------------------------------------------------------------ | --- | --- | --- | --- | ------------------------ | --- | +| Disciplinary Core Courses | | | | 42 | Capstone Course | 6 | +| Disciplinary Electives (Marketing Technology and Strategy) | | | | 12 | | | + nfiq-csb/emmargorp/kh.ukh.sbukh.gu//:sptth +https://ug.hkubs.hku.hk/programme/bsc-mat Disciplinary Electives (Marketing Analytics) 12 C Free Electives / Optional Major / Optional Minor(s) 90 +| Capstone Course | | | | 6 | | | +| ----------------- | --- | --- | --- | --- | --- | --- | +Total 240 +| C Free Electives / Optional Major / Optional Minor(s) | | | | 90 | | | +| ------------------------------------------------------ | --- | --- | ----- | --- | --- | --- | +| | | | Total | 240 | | | +Find out more: :erom tuo dniF +14 15 + +| Enrich Your Life | | | | | | | | International Field Trips | +| ---------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------- | +To enhance students' exposure to global businesses, the School organises overseas field trips +for students visiting corporate firms and mingling with C-suite Executives in some of the +HKU Business School fosters a culture of student initiative and implementation of ideas, both +world's most prominent financial centres, e.g. Australia, France, Japan, Singapore, Spain, +within and outside the campus, as it is these real-world engagements that set us apart. Zurich, New York City and London. +Experiences in learning beyond the classroom, extending into realms outside academia, +enable students to cultivate an understanding of social dynamics, cultural diversity, and +Visiting Programmes +collaborative intellect. These experiences contribute significantly to their personal growth, +Through the Yale Visiting International Students Programme, students are sent to spend a +interpersonal skills, and social development, shaping them into global citizens, lifelong +year at Yale University where they actively engage themselves in a multicultural society. +learners, and future business leaders. +With the generosity of the donors, some of our students are selected for the Junior Year +Abroad Programme and Visiting Students Programme to study as visiting students at +Pembroke College and St Peter’s College of University of Oxford respectively with +Overseas Enrichment Experiences substantial scholarships. The intellectual ambience of these top universities is truly +inspirational and conducive to learning. +Business Exchange Programme / +Overseas Career Opportunities +HKUWW Undergraduate Student Exchange Programme +Being the youngest Faculty in the University, our exchange Overseas career opportunities provide exceptional learning experiences to our students. In +collaboration with the DCAF - Geneva Centre for Security Sector Governance, a Traineeship +activities are phenomenal. Each year, the School plans to send +Project will be launched in 2025, and selected students will be sent to Geneva for a 6-month +| over 400 | students | for overseas | exchange | to top-notch | | | | | +| --------------------------------------- | --------- | -------------- | --------- | -------------- | -------------------- | --- | --------------------- | --------------------------------------------------------------------------------- | +| | | | | | 400+ | | 400+ | project related to the humanitarian business, unquestionably rewarding exposure. | +| universities to broaden their horizons. | | | | | | | | | +| | | | | | exchange partners | | Business School | | +| | | | | | across 40+ countries | | students on exchange | | +Beta Gamma Sigma Regional Leadership Conference +Business School Exchange Partners * Beta Gamma Sigma (BGS) is the international honour society that serves business +programmes accredited by the Association to Advance Collegiate Schools of Business +(AACSB International). With the establishment of the HKU BGS Chapter, the School registers +Denmark students annually for the BGS Regional Leadership Conference held in the U.S. where they +Canada +McGill University Copenhagen Business School can meet and bond with many other fellow students from the world's best business schools. +University of British Columbia Aarhus University The Leadership Summit is eye-opening and challenging which leads to a great journey of +| University of Toronto | | | | | | | | self-discovery. | +| --------------------- | --- | --- | --- | --- | --- | --- | --- | --------------- | +Western University +Simon Fraser University +Belgium +| | | | KU Leuven | | The Netherlands | | | | +| --- | --- | --- | --------- | --- | --------------- | --- | --- | --- | +University of Amsterdam +| | U.K. | | | | Erasmus University Rotterdam | | | | +| --- | ---- | --- | --- | --- | ---------------------------- | --- | --- | --- | +King's College London +| | University of Manchester | | | | Italy | Japan | | | +| --- | ------------------------ | --- | --- | --- | ------------------ | ---------------- | --- | --- | +| | | | | | Bocconi University | Kyoto University | | | +University of Edinburgh +| | University of Warwick | | France | | | | | | +| --- | --------------------- | --- | ------- | --- | --- | --- | --- | --- | +Toulouse School of Economics +| | | | ESSEC Business School | | | Singapore | | | +| --- | --- | --- | --------------------- | --- | --- | --------- | --- | --- | +National University of Singapore +Nanyang Technological University +Germany +U.S. +| New York University | | | Frankfurt School of Finance & Management | | | | | | +| --------------------------------- | --- | --- | ---------------------------------------- | --- | --- | --- | --- | --- | +| University of Southern California | | | University of Mannheim | | | | | | +Australia +| University of Washington | | | Munich Business School | | | University of Sydney | | | +| ------------------------ | --- | --- | ---------------------- | --- | --- | -------------------- | --- | --- | +University of Georgia +Monash University +University of New South Wales +Middle East +Reichman University +* This list is non-exclusive. +16 17 + +Student Sharing +| | | | | | | Ng, Hoi Kiu | BEcon&Fin | | | | | +| --- | --- | --- | --- | --- | --- | ----------- | --------- | --- | --- | --- | --- | +Exchange at Peking University Guanghua School of Management +(China), 2023-24 +| | | | Kwok, Sin Ue | BBA(Law)&LLB | | The curriculum at Peking University was incredibly comprehensive. Each | | | | | | +| --- | --- | --- | ------------- | ------------ | --- | ----------------------------------------------------------------------- | --- | --- | --- | --- | --- | +Exchange at University of Edinburgh +class provided me with valuable knowledge and fundamental skills +(United Kingdom), 2023-24 +essential for my professional development in the business field. Engaging +| | | | University of Edinburgh has rich history, coupled | | | in discussions with both local and international students broadened my | | | | | | +| --- | --- | --- | -------------------------------------------------- | --- | --- | ----------------------------------------------------------------------- | --- | --- | --- | --- | --- | +| | | | with its modern facilities and renowned faculty, | | | horizons and deepened my understanding of China's business landscape. | | | | | | +Living in Beijing also allowed me to explore the city's rich history and +provided an ideal environment for my academic +| | | | growth and personal development. The campus | | | modernity, providing numerous invaluable learning experiences and | | | | | | +| --- | --- | --- | ---------------------------------------------- | -------------------- | ------------ | ------------------------------------------------------------------------- | ------------------------- | ------------------- | --------------- | --- | --- | +| | | | offered a wide range of resources and support | | | connections | to this vibrant urban | center. I highly | recommend all | | | +| | | | services for | exchange students, | making my | university students to participate in an exchange program at least once, | | | | | | +| | | | transition smooth | and enjoyable. | I actively | as it offers numerous important takeaways. | | | | | | +participated in welcoming events organized by the +university, where I met a group of international +| Hung, Lok Yin | BBA(Acc&Fin) | | | | | | | | | | | +| -------------- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +friends. The combination of academic excellence, +Exchange at Aarhus University (Denmark), +| | | | cultural immersion, | and the | development of | | | | | | | +| --------------------------------------------------- | --- | --- | -------------------------------------------- | ---------------------- | ---------------- | --- | --- | --- | --- | --- | --- | +| 2023-24 | | | essential life | skills has prepared | me for the | | | | | | | +| Studying at AU is surprisingly similar to studying | | | challenges and opportunities that lie ahead. | | | | | | | | | +at HKU, with the major difference being the +workload, which enabled me to delve into Danish +society from various angles. I met numerous +people and was captivated by the concept of +"hygge", embodying the essence of cherishing +| cozy moments | with loved ones. | I deeply | | | | | | | | | | +| -------------- | ------------------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +appreciate the notion that happiness can be +| found in simplicity | and purity. | Overall, the | | | | | | | | | | +| ---------------------- | ------------- | -------------- | --- | --- | --- | ----------- | --------- | --- | --- | --- | --- | +| | | | | | | Yu, Er Han | BBA(IBGM) | | | | | +exchange experience has been transformative for +Exchange at University of California, Irvine +me. Despite its brevity, it was an unforgettable +(United States), 2023-24 +and life-changing event that will forever be So, Yiu Hei Michael BEcon&Fin +cherished as a pivotal part of my youth. This journey not only broadened my academic International Committee of the Red Cross +| | | | | | | horizons but also enriched my personal and cultural | | | Traineeship, 2023-24 | | | +| --- | --- | --- | --- | --- | --- | ---------------------------------------------------- | -------------------------- | ---------------- | --------------------------------------------------- | --- | --- | +| | | | | | | insights, | marking a pivotal | chapter in my | During my ICRC traineeship, I had the privilege of | | | +| | | | | | | educational | voyage. The distinctive | academic | | | | +collaborating closely with the Geneva Centre for +| | | | | | | framework | facilitated in-depth | exploration of | | | | +| --- | --- | --- | --- | --- | --- | ---------- | ---------------------- | ---------------- | ----------------- | ----------- | --------------------- | +| | | | | | | | | | Security Sector | Governance | (DCAF) on a joint | +subjects and enhanced interaction with both faculty +| | | | | | | | | | project. Under | the supervision | of the ICRC's | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------------- | ----------------- | ---------------- | +and peers. Throughout this period, overcoming +| | | | | | | | | | Economic | Adviser, I assisted | in the day-to-day | +| --- | --- | ----------------- | --- | --- | --- | ----------------------------------------------- | --- | --- | ----------- | ---------------------- | -------------------- | +| | | Lau, Sui Yu BBA | | | | | | | | | | +| | | | | | | challenges—from adjusting to a new educational | | | management | of this initiative, | which involved | +Exchange at University of New South Wales (Australia), 2023-24 +| | | | | | | system to | assimilating into a | diverse cultural | | | | +| --- | --- | --------------------------------------------------------------------- | --- | --- | --- | ------------------------------------------------- | ---------------------- | ------------------ | ---------- | ------------------------- | ----------- | +| | | | | | | | | | fostering | external communication, | conducting | +| | | UNSW has provided me with resources and incentives to motivate me to | | | | environment—fostered resilience and a heightened | | | | | | +research on conflict and security reporting trends, +sense of self-awareness, reinforcing my dedication +work on practical career skills, such as Excel training and presentation skills. and liaising with a diverse range of stakeholders. +| | | | | | | to pursuing | a career in globalized | business. | I | | | +| --- | --- | --- | --- | --- | --- | ------------- | -------------------------- | ---------- | --- | --- | --- | +The school also showcased innovation in its teaching style, being receptive This cross-organizational partnership underscored +to student feedback and staying abreast of current global topics. It has wholeheartedly recommend this exchange program the importance of adopting interdisciplinary +| | | been a valuable opportunity for me to venture beyond my comfort zone | | | | to my peers, as it ensures holistic development and | | | | | | +| --- | --- | --------------------------------------------------------------------- | --- | --- | --- | ---------------------------------------------------- | --- | --- | ----------- | ----------------- | ------------------- | +| | | | | | | | | | approaches | to address the | complex interplay | +| | | and expand my horizons outside the classroom. While this journey was | | | | creates lasting memories. | | | | | | +between business and human rights. The experience +challenging, it stands out as the most memorable experience of my life. significantly enhanced my ability to navigate +| | | | | | | | | | multicultural | environments | and work effectively | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | ------------- | ----------------------- | +with high-level international organizations. +18 19 + +Corporate Exposure +| | | | | | Vaidya, Samhith | BEcon&Fin | | | | | +| --- | --- | --- | --- | --- | ---------------- | --------- | --- | --- | --- | --- | +HKU-Accenture Business Consulting Programme 2023-24 +HKU Business School has a strong tie with corporate partners. To nurture students' business The HKU-Accenture Business Consulting Programme was +everything I needed - an opportunity to meet like-minded people +and market acumen, enrichment programmes are organised through the joint efforts of +corporate partnerships. and be mentored to think out of the box in a systematic manner. +The skills I take home, the friends I've made, and the mentors I +Enrichment Programmes in collaboration with Corporate Partners connected with are not just the biggest take aways, but also the +ability to apply my learnings from the programme onto my courses, +HKU Business School has been collaborating with corporate partners on initiatives to spice up +interviews, and case competitions. +students' career exploring journey. The partnership with Accenture enables students to gain +exclusive insights into business consulting through the HKU-Accenture Business Consulting +Programme. +"Venture into Future Career" has also been rolled out since 2020-21 in collaboration with +Cyberport and the Chartered Institute of Management Accountants. Students are given a +platform to mingle with professionals from different industries via a series of activities like +coffee chats and guest talks. This definitely helps students explore their career choices and +sort out their career direction. +Jiang, Yiguo Amory +| HKU Business School Mentorship Programme | | | | | | | | | BBA | | +| ---------------------------------------- | --- | --- | --- | --- | -------------- | --------- | --- | ------------------------------- | --- | --- | +| | | | | | Li, Mingfang | BBA(IBGM) | | HKU Business School Mentorship | | | +Leveraging the School's extensive network in the business and finance industry, we have +| | | | | | Venture into Future Career (VFC) | | | Programme 2023-24 | | | +| --- | --- | --- | --- | --- | --------------------------------- | --- | --- | ----------------- | --- | --- | +developed our own mentorship programme to furnish students with opportunities to +Programme 2023-24 +There is so much I have learned from my mentor. +| enhance their | skills and knowledge | essential | for their future personal | and career | | | | | | | +| --------------- | ----------------------- | ---------- | ----------------------------- | ------------ | -------------- | ------------------------- | --------- | -------------------------------------------------- | --- | --- | +| | | | | | Participating | in the VFC programmre | provided | He provided numerous opportunities to participate | | | +development. Our students are empowered to connect to the real business world through +profound insights as professionals from diverse +direct and active interaction with seasoned professionals, executives and entrepreneurs. in the events organized by his company. I believe +| | | | | | industries | shared their personal | experiences, | that these hands-on experiences in the industry | | | +| --- | --- | --- | --- | --- | ------------------------------------------------- | ------------------------ | --------------- | ------------------------------------------------- | ------------------------- | ------------ | +| | | | | | enhancing | our understanding. | The dedicated | are the most valuable. Through my involvement in | | | +| | | | | | sessions on CV enhancement and interview skills | | | real events, | I had the opportunity | to witness | +| | | | | | before engaging with guest speakers and pursuing | | | firsthand how everything operated, particularly | | | +internships were particularly beneficial. Learning +observing how my mentor handled a variety of +| | | | | | from entrepreneurs | and seasoned | professionals | situations. There is no better way to learn about the | | | +| --- | --- | --- | --- | --- | -------------------- | -------------- | ------------------ | ------------------------------------------------------ | --- | --- | +| | | | | | was invaluable, | fostering | a rich learning | PR industry than by actively engaging in it myself. | | | +environment where networking, CV refinement, and +experiential learning were key. A standout moment +| | | | | | was the | engaging coffee | chat with industry | | | | +| --- | --- | --- | --- | --- | --------- | ----------------- | --------------------- | ----------------------- | ---------- | --- | +| | | | | | | | | Dong, Christina W Y | BFin(AMPB) | | +professional Mr. Darron Sun. The key lesson learned +| | | | | | was the | uniqueness of each | career journey, | HKU Business School Mentorship Programme | | | +| --- | --- | --- | --- | --- | ------------ | --------------------- | -------------------- | ----------------------------------------- | --- | --- | +| | | | | | emphasizing | the importance | of avoiding | 2023-24 | | | +| | | | | | comparisons | and highlighting | the potential for | | | | +My mentor invited me to participate in the annual +industry transitions or role changes regardless of +conference of the Institute of ESG and Benchmark, +| | | | | | one's career stage. | | | where I gained valuable insights into the growth of | | | +| --- | --- | --- | --- | --- | ------------------- | --- | --- | ---------------------------------------------------- | --- | --- | +established companies in ESG, greatly enhancing +| | | | | | | | | my understanding | in ESG-related | courses and | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------ | ---------------- | ------------- | +inspiring ideas for upcoming cases. Furthermore, +| | | | | | | | | my mentor | arranged a visit | to his company, | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | ------------------- | ------------------ | +providing updates on company valuation and +IPO-related systems. Exploring the latest ESG +| | | | | | | | | regulations | from HKEX has | sparked fresh | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------ | ---------------- | --------------- | +perspectives for my career trajectory planning. +20 21 + +Experiential Learning Programmes +Experiential learning is also embedded in our course curriculum. Business Consulting +Practicum offers business and economics students hands-on experience of working with +business clients in solving "real-life" problems relating to their studies, under the guidance of +the faculty teacher, Consultant-in-Residence and professional mentors. Business Case +Analysis and Presentation enables students to communicate and ask for practical advice +from senior executives through in-classes case presentations. Impact Lab gives students an +excellent opportunity to work directly for social ventures under the guidance of a teacher and +professional mentors. +Tsang, Tze Yeung Matthew BEcon&Fin +An intensive academic and cultural exchange experience, Doing Business in China, is +Business Consulting Practicum 2024 +offered to students where they can interact with peers from Peking University, learn about +the latest business and social environment of China, and explore the rich Chinese cultural BCP has been a rewarding experience that +heritage. Students can also acquire an in-depth understanding of the development and allowed me to apply my academic knowledge +latest economics, finance, creativity, innovation and entrepreneurship trends in the to real-world business challenges. Working +fast-growing Greater Bay Area by joining the Creativity, Innovation & Entrepreneurship directly with my mentor and client has +in China Programme. pushed me to think critically and creatively in +ways I hadn't experienced in the classroom. +I'm confident that the problem-solving, +communication, and client-facing skills I have +developed will serve me well when pursuing +my future career in consulting or another +business-oriented field. +Kaur, Suchleen BBA(Acc&Fin) +Business Consulting Practicum 2024 +The opportunity for real-time interaction and feedback with the +client had propelled me to join BCP. I was always curious about +how a consultant would address and communicate with the client. +BCP has taught me about proper business etiquette and equipped +me the necessary skills to negotiate conflicting opinions during +team discussions. Throughout the process, our mentor and +professors provided great help and insights. Their invaluable +feedback and questions helped nudge us in the right direction, +taking a step back to look at the big picture and most importantly, +prioritise the client’s needs. +22 23 + +Ou, Tanyalak BBA +"Creativity, Innovation & Entrepreneurship in China" +Summer Virtual Programme 2024 +The combination of thought-provoking lectures, insightful guest +speaker sessions, and dynamic debate challenges created a richly +immersive learning environment. I was constantly challenged to +think critically, conduct thorough research, and articulate my ideas +and arguments effectively. It was an experience that equipped me +with invaluable insights and skills for navigating the complexities of +China's dynamic business landscape. +Yeung, Yui Pong BBA(Law)&LLB +Doing Business in China Programme 2024 +Participating in the DBIC Programme at PKU +Guanghua School of Management was an +extraordinary and transformative experience. The +lectures were particularly inspiring, covering a wide +range of topics that are crucial for understanding +the Chinese business landscape. From discussions +on demographics to exploring real estate finance +and the digital economy, and delving into the realm +of digital marketing in China, each session offered +Nan, Shan BEcon&Fin +valuable unique perspectives. Moreover, I was +"Creativity, Innovation & Entrepreneurship in China" +particularly impressed with the office visit to Swire +Summer Virtual Programme 2024 +and Kuaishou, which was a fantastic experience +Joining the CIEC programme ignited my entrepreneurial spirit and +understanding how the two companies navigate +enabled me to accumulate knowledge in doing business +the Chinese market and achieve rapid growth. +and making decisions in life. We learned about China’s economic growth +and cultural richness and grasped the role of innovation in boosting the +economy. “Social networking is critical in the business world”, CIEC +establishes a great networking platform for students, where +like-minded peers’ ideas collide, allowing us to explore the world +through a completely new lens. +24 25 + +Case Competitions +Darmawan, Jocelyn Victoria BSc(QFin) +The School provides ample opportunities to refine students’ presentation Bloomberg Global Trading Challenge 2023 +and analytical skills through a myriad of local as well as overseas (Grand Prize Winner) +competitions. From training to delicate analysis of business cases to a It was such an amazing experience to win the +variety of business scenarios, students are equipped with different Challenge, and I would like to give a shout out to my +skillsets to compete with candidates from the world's leading universities. teammates who have dedicated a lot of their time and +energy into researching and strategizing our portfolio +allocation. I learned a lot about how the stock market +reacts to different macroeconomic events, and how +to capitalize on opportunities to achieve a high +portfolio return. I would also like to extend my thanks +to Bloomberg, HKU Business School, and the faculty +advisor for their unwavering support. +Lau, Chun Ho Darren BBA(Acc&Fin) +EY Young Tax Professional of the Year 2023 +(1st Runner-up) +The EY Young Tax Professional of the Year 2023 was a +defining moment in our university lives, as it allowed +us to apply critical thinking to complex, cross-border +tax issues and showcase our insights before esteemed +CHAN, Chun Yan Natalie BBA(Law)&LLB +professionals. This rewarding journey not only +Lazaridis International Case Conference 2023 +broadened our understanding of the international tax +(2nd Place) +landscape but also granted us a platform to showcase +Participating in the LazICC 2023 and securing 2nd our abilities and potential. In essence, we eagerly +Place was a transformative experience that left an anticipate building upon this foundation to evolve into +indelible mark on my personal and professional world-class, game-changing professionals. +growth. Throughout the competition, I had the +privilege of working closely with a dedicated team. +The complex business cases enabled our team +to sharpen our analytical skills, while reinforcing +the significance of time management and +communication abilities. The international aspect of +the competition provided us with the unique +opportunity to network with individuals from around +the world. I am immensely thankful for our coach and +faculty members who supported us along the way. +26 27 + +Luo, Regina BBA(IBGM) +HKICPA Business Case Competition 2023 (2nd Runner-up) +Taking part in this competition was an extraordinary +experience that allowed me to expand my business +knowledge and fostered my growth in several soft skills. It +was a fresh new experience as we learnt to think outside the +box and make decisions based on different platforms by +putting ourselves in the shoes of the gaming company. +Communication and presentation skills were another key +takeaway from the competition for me. It was truly a +wonderful and invaluable experience. +Students' achievements in case competitions +(2022-2024 awards) +Overseas / Mainland +Amplify Trading Global Markets Competition 2022 Finalist +Asian Business Case Competition 2022 1st Runner-up +Bloomberg Global Trading Challenge 2023 Grand Prize Winner +Regional Prize Winner: Asia & Oceania +Yao, Chuanjie Rosie BBA(BA) +CBS Case Competition 2023 Winner +HKGCC Business Case Competition 2023/24 +(Champion - Swire Pacific track and CGMA Global Business Challenge 2023 - North Asia Final Champion +1st Runner-up - Chinachem Group track) Champions' Trophy Case Competition 2024 Finalist +China Mainland, Hong Kong and Macao Digital Economy Innovation and 2nd Runner-up (Gold Medal) +Participating in the HKGCC Case Competition was +Entrepreneurship Competition 2024 +an incredible experience that has left an indelible +Climathon2022@Hangzhou Championship +mark on my personal and professional growth. +Deloitte Digital Camp 2023 Champion +Through our innovative initiatives for the case +Deloitte Digital Camp 2024 Second Place +companies, we aimed to make a positive impact +Deloitte Tax Championship 2023 - National Round Merit Award +on society and the environment. I deeply +appreciate my teammates’ collective effort, Deloitte Tax Championship 2023 - Regional Round Top 3 teams +collaboration, and synergy which were the driving EY Young Tax Professional of the Year 2022 - National round Champion +forces behind our success. I extend my deepest Family Office Association Hong Kong (FOAHK) 2023 Case Competition 2nd Runner-up +gratitude to our supervisor, whose guidance and Global Trading Challenge 2022 World Top 10% +support were invaluable throughout the Häagen Dazs Challenge 2022 First Prize +competition. This experience has left me inspired HSBC Global Private Banking Case Challenge 2023 1st Runner-up +and even more determined to contribute to the +HSBC Global Private Banking Case Challenge 2022 1st Runner-up +business world and create meaningful change. +Investing in the Future University Competition 2023 by Pictet Asset 2nd Runner-up +Management +Lazaridis International Case Conference 2023 1st Runner-up +Marshall International Case Competition 2024 Finalist +Scotiabank International Case Competition 2022 Winner +Scotiabank International Case Competition 2023 Winner +Shaping the Future Nansha Innovation Challenge 2022 1st Runner-up +Most Creative Award +UW Global Business Case Competition 2023 Finalist +Youth-Storm Microsoft Business Analytics Challenge 2022 National Champion +28 29 + +| Local | | Local | | +| ----- | --- | ----- | --- | +100WF Hong Kong JumpStart University Investment Hackathon 2023 1st Runner-up HKU-Airstar Bank Joint University Case Competition 2024 Champion +| ACCA Business Competition 2022-23 | Champion | | 2nd Runner-up | +| --------------------------------- | ------------------------- | ------------------------------------------------ | -------------- | +| | Creative Excellence Award | HKU-SaSa Joint University Case Competition 2023 | 1st Runner-up | +| | Digital Excellence Award | | Best Presenter | +| | Best Budgeting Award | Hong Kong Joint University Case Competition 2022 | Champion | +ACCA Hong Kong Business Competition 2023 Outstanding Team for start-up Filix Medtech 2nd Runner-up +| | Team Excellence Award | HSBC Life Insurance Innovation Competition 2023 | 1st Runner-up | +| --- | ------------------------- | -------------------------------------------------------- | ---------------------------- | +| | Merit Award for start-up | | Best Use of Technology Award | +| | MakeOmnia | International Macro Business Simulation Competition 2022 | Champion | +Advancing Greener Tomorrow Competition 2022 Champion International MBS Business Management Simulation Competition 2023 Champion +BOCHK Challenge 2022 Gold Award for Innovative Design 1st Runner-up +BOCHK Challenge 2023 Bronze Award for Innovative Design ISM-HK Student Case Competition 2024 First place +CFA Institute Research Challenge 2022-23 Hong Kong Local Final 2nd Runner-up Third Place +CGMA Global Business Challenge 2023 - Semi-final Round (Hong Kong) Champion Best presenter +Finalist J.P. Morgan Private Bank Challenge 2022 Hong Kong Champion +Deloitte USHINE Case Competition 2024 Champion KPMG Ideation Challenge Semi-Final (Hong Kong Regional Final) 2022 Winner +ESG Challenge 2023 Champion LEAD to the Future - Business Case Competition 2022 1st Runner-up +EY Young Tax Professional of the Year 2022 - Hong Kong Region Champion 2nd Runner-up +| | 1st Runner-up | | 3rd Runner-up | +| --- | ------------- | --- | ------------- | +Finalist L'Oréal Brandstorm - Brandstorm 2022 (Hong Kong Final) Tech Track Winner +EY Young Tax Professional of the Year 2023 - Hong Kong Region 1st Runner-up MBS Simulated Business Management Competition 2024 Champion +| | 2nd Runner-up | | 1st Runner-up | +| --- | ------------- | --- | ------------- | +Female Entrepreneur Business Simulation Competition 2022 2nd Runner-up New Century Elite Development Programme 2023 2nd Runner-up +FSDC Career Day 2022 - Pitch Perfect Video Competition Champion Polymer Capital Management Inter-varsity ESG Case Challenge 2022 Champion +HKGCC Business Case Competition 2023/24 Chevalier - Champion Quest for Securities & Investment Elites (QSIE) Case Analysis Competition 2022 1st Runner-up +| | Chevalier - 1st Runner-up | | Outstanding Award | +| --- | ------------------------- | --- | ----------------- | +Chinachem Group - Champion The Hong Kong Future Leader Awards 2022 Best Performing Team Award +| | Swire Pacific - Champion | Venture Capital On Campus (VCOC) - Investment Challenge 2023 | Winner Team | +| --- | ------------------------ | ------------------------------------------------------------ | ----------- | +Chinachem - 1st Runner-up +| HKGCC Pitch Perfect Programme 2024 | Overall Champion | | | +| ---------------------------------- | ---------------- | --- | --- | +Winner +| HKICPA Business Case Competition 2023 | 2nd Runner-up | | | +| ------------------------------------- | ------------- | --- | --- | +| HKSI Institute Case Competition 2022 | Finalist | | | +Top Performer Award +Best ESG Initiative Award +| HKSI Institute Case Competition 2023 | Champion | | | +| ------------------------------------ | -------- | --- | --- | +1st Runner-up +Merit & Finalist +Semi-finalist +| HKSI Institute Case Competition 2024 | Champion | | | +| ------------------------------------ | -------- | --- | --- | +1st Runner-up +Best performer of Roma Risk +Advisory Limited +Merit +30 31 + +Internship Experience +Kim, Chaeyeon B F in(AMPB) +J.P. Morgan - Equity Research Summer Analyst +This internship allowed me to immerse myself in the exhilarating world of financial +Cheung, Lok Wan BBA analysis. I contributed to drafting research reports and also worked on an individual +UBS AG - Global Wealth Management project. The internship was not just about the tasks at hand; it was a journey of +Summer Analyst personal and professional growth. I had the privilege of meeting various exceptionally +Working at an international firm as a talented individuals within the firm. Through engaging with these professionals, I +junior leverages my personal growth and gained valuable insights into the financial markets. This experience was not only a +builds a strong team network. Finance is testament to my passion for finance but also a stepping stone towards achieving +not just about numbers; we are doing a excellence in my future career. +huge people business to deliver +professional services and connect the +world to a better place. +Lee, Dohoon B E c on&Fin +HKU – Neuberger Berman +Ho, Yau Wai Athena B BA(IS) +Private Equity Internship +Bloomberg - Analytics and Sales Summer Intern +Programme 2023-24 +I am grateful to have the opportunity to work with +I believe what sets Neuberger Berman HK office Lee, Wing Ting Stephanie B B A(Acc&Fin) +many talented and incredible managers, team +apart as an employer especially to the interns, is the HKU – Neuberger Berman Private Equity +leaders, mentors, coordinators, project leads and +culture that genuinely values the juniors and nurtures Internship Programme 2023-24 +colleagues who provided professional advice, +Chung, Lok Yin B Fin(AMPB) guidance and support for my personal growth and their development. And I believe such culture starts During my enriching internship at Neuberger +Raffles Family Office - Summer Intern development, especially in my presentation skills, all the way from the top, Mr. Kent Chen, down to the Berman, I immersed myself in a dynamic deal +communication skills and market knowledge. entire team. As interns, we are given full exposure to environment within the Southeast Asia team, +During my time at Raffles Family Office, I developed whichever transaction we are staffed on including engaging in primary and co-investment deals that +a spreadsheet that automatically retrieves pertinent I am also glad to collaborate with my brilliant intern the email chains, data rooms, and the shared drive. broadened my deal type exposure. Working on +data from Bloomberg when an equity ticker is batchmates on various projects about terminal In fact, thanks to Kent, I was also fortunate to attend deals in diverse regions like India, Australia, and +entered, facilitating relationship managers in functions and market news. a forum in January where we got to listen in on Japan significantly expanded my market +producing client meeting materials. Despite my keynote speeches and panel discussions, and understanding. A pivotal aspect of my internship +initial lack of familiarity with the Bloomberg network with private equity and venture capital was gaining comprehensive insights into the +Terminal and VBA, I was encouraged by my professionals that flew in from across Asia. This was private equity industry and the Southeast Asia +manager (mentor) to visit the Bloomberg office to one of those experiences that really motivated market. By actively contributing to deal processes, +deepen my understanding. I even employed GPT to me once again to pursue a career in this field. drafting investment memos, and interacting with +assist in generating VBA codes. This experience +senior professionals, I acquired valuable +underscored the importance of embracing new +knowledge of prevailing investment trends across +technologies and tools, highlighting that consistent +multiple markets. +effort and exploration can lead to valuable Luo, Wenhao B E c on&Fin +outcomes in this tech-driven era. +HKU - MindWorks Venture Capital Internship +Hui, Matthew B B A(Acc&Fin) Programme 2023-24 +Ares Management Corporation - Intern +My experience with the HKU - MindWorks Venture Capital Internship Programme was +(Investment Operation) +truly enriching. As an investment intern, I had the opportunity to work closely with +My journey at Ares has significantly enhanced my understanding of the alternative investment industry, especially industry professionals and gained hands-on experience in the venture capital field. I +in the private credit sector; I have the opportunity to assist in the operations of multiple funds, monitoring and evaluated potential investment opportunities, conducted market research, and assisted +repatriating fund cash flow, managing FX and forwards / options hedging for risk management. I was also with due diligence. The internship allowed me to apply my academic knowledge in a +honoured to participate in structuring and launching a billion-sized private credit fund, completing due diligence real-world setting, fostering significant personal and professional growth. I am grateful +and KYC checks for external parties and communicating with LPs and external stakeholders. I'm truly thankful to for the invaluable connections, industry insights, and practical skills I acquired through +work with wonderful colleagues and for their mentorship and guidance during my period in Ares. this experience, which have better prepared me for my future career aspirations. +32 33 + +Mo, Cheuk Lam B E c on&Fin +HSBC - Wealth and Personal Banking Intern +During my internship, I immersed myself in the mortgage industry, +| enhancing | my understanding | of key facets. | Highlights include | | | | | | +| ------------ | ------------------ | -------------------- | -------------------- | --- | --- | --- | --- | --- | +| involvement | in developing | equitable mortgage | products, gaining | | | | | | +insights into risk assessment, understanding compliance standards' +| impact, and | collaborating | across departments | for streamlined | | | | | | +| ------------- | -------------- | -------------------- | ----------------- | --- | --- | --- | --- | --- | +mortgage operations. This experience has equipped me to tackle future challenges in the mortgage sector, +providing a solid foundation for my professional growth, and I look forward to leveraging this knowledge in Wong, Chloe Miriam B B A (Law)&LLB +upcoming endeavors. Recipient of HKEX Foundation Scholarship 2023-24 +I am very grateful to the HKEX Foundation for this Scholarship. +Thanks to this generous award, I have been able to pursue valuable +learning opportunities that both align with my existing passions and +| | Wong, Ho Yeung B B A (Acc&Fin) | | | | | broaden my horizons. | | | +| --- | ---------------------------------------- | --- | --- | --- | --- | -------------------- | --- | --- | +Bloomberg - Analytics and Sales Summer Intern +My experience as an analytics and sales summer intern at Bloomberg was incredibly +fulfilling. I was involved in a range of projects across different asset classes and had +the opportunity to interact with clients from various sectors, including International Su, Zihan B E c o n&Fin +Banks on the Sell side, Private Equity firms, and Hedge Funds on the Buy Side. I am Recipient of Hong Kong Jockey Club +grateful for the support I received from HKU Business School, both in terms of Scholarship 2022-23 to 2023-24 +academic support and extracurricular activities prior to the internship. The HKJC Scholarship not only recognizes my +past achievements but also encourages me to +move forward and explore more. I am incredibly +grateful for the Jockey Club's generosity and will +Our Student Scholars continue to pass this love on to the broader +community and society. +Chiu, Yi Nap B S c ( Q Fin) +Recipient of Shanghai Commercial Bank +Ho, Hiu Lam B B A(IBGM) +Scholarship 2023-24 +Recipient of HKU Business School - BOC Life +| | | | | | | The Shanghai | Commercial Bank | Scholarship | +| --- | --- | --- | --- | --- | --- | -------------- | ----------------- | ------------ | +Future Leader Scholarship 2023-24 +certainly made my university life more vibrant. +| | | | The scholarship | provided immense | support | | | | +| --- | --- | --- | ----------------- | ------------------ | -------- | --- | --- | --- | +Without it, I could not have pursued my dream as +towards my overseas exchange experience and +a dancer, continued to stay in Lee Shau Kee Hall +has undoubtedly empowered me to develop a +while having the chance to explore the world! +| | | | global mindset | and gain new | perspectives | | | | +| --- | --- | --- | ---------------- | --------------- | ------------- | --- | --- | --- | +through immersing myself in a distinct cultural +environment. I am grateful that the scholarship +Gandhi, Devansh B B A +| | | | has been | effective in | | | | | +| --- | --- | --- | ---------- | -------------- | --- | --- | --- | --- | +Recipient of Faculty Student Contribution and +| | | | accelerating | my growth | | | | | +| --- | --- | --- | ------------- | ----------- | --- | --- | --- | --- | +Innovation Award 2023-24 +| | | | personally, | academically | | | | | +| --- | --- | --- | ------------ | ------------- | --- | --- | --- | --- | +Receiving the Award is a profound honour, marking +and professionally. +a significant milestone in my journey. It motivates +me to strive further to build diverse communities +of innovation. This award acknowledges my past +| efforts and | fuels my dedication | to continue | | | | | | | +| ------------- | ---------------------- | --------------- | --- | --- | --- | --- | --- | --- | +| making a | meaningful impact | in the world | of | | | | | | +entrepreneurship. +34 35 + +For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) +2025 +| | | | In addition | to the minimum | entrance requirements | above, candidates | applying | for the BBA(BA), | +| --- | --- | --- | ------------- | ----------------- | ----------------------- | ------------------- | --------- | ------------------- | +JUPAS Admission BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the +HKDSE examination: +Admissions criteria BBA(BA) / BSc(MAT) - Level 3 or above in one of the following subjects: +(HKDSE) +| | | | | | Combined | | Integrated | | +| --- | --- | --- | --- | ------- | -------------- | --------- | ---------- | ------- | +| | | | | Biology | BAFS Chemistry | Economics | ICT | Physics | +| | | | | | Science | | Science | | +Local candidates applying on the strength of the +Hong Kong Diploma of Secondary Education (HKDSE) +| are required | to fulfil | the minimum entrance | BBA(BA) | | | | | | +| -------------- | ----------- | ----------------------- | ------- | --- | --- | --- | --- | --- | +requirements for HKU Business School programmes, +| as shown below: | | | BSc(MAT) | | | | | | +| --------------- | --- | --- | -------- | --- | --- | --- | --- | --- | +Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not +strictly required. +Level 4 or above in English Language; +Level 3 or above in Chinese Language1; +| | | | BBA(Law)&LLB | Level 5 or above in | | Level 4 or above in | | | +| --- | --- | --- | -------------- | -------------------- | --- | -------------------- | --- | --- | +Level 3 or above in Mathematics; +| | | | | English Language | | Chinese Language1 & Mathematics | | | +| --- | --- | --- | --- | ---------------- | --- | ------------------------------- | --- | --- | +Grade “Attained (A)” in Citizenship and +Social Development2; and +Level 3 or above in two Elective Subjects. +| | | | BBA(IBGM) | Level 5 or above in | | | Level 4 or above in | | +| --- | --- | --- | ---------- | --------------------------------------------------------- | --- | --- | -------------------- | --- | +| | | | | English Language | | | Mathematics | | +| | | | BSc(QFin) | Level 3 or above in Extended Module 1 or 2 of Mathematics | | | | | +Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis. +Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered. +Application +For details of the 2025 JUPAS exercise, please visit www.jupas.edu.hk. +36 37 + +2025 Non-JUPAS Admission University and Programme Entrance Requirements +Due to the complex combination of education background of applicants, each application will +be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the +Selection process and entrance requirements standards and requirements on the website of the Admissions Office of the University before +making an application: +Eligibility +(1) Admissions standards (International Qualifications or National Qualifications); +(1) Local students applying for admission on qualifications other than the HKDSE examination; +(2) Programme specific subject requirements; +(2) Non-Local students with qualifications other than the mainland China’s National Joint +(3) English language requirement; and +College Entrance Examination (NJCEE/Gao Kao) +(4) Second language requirement (i.e. any language other than English). +Application Procedures +In particular, please note that the expected lower boundary for admissions listed on the +All non-JUPAS applications should be submitted via the online application system by noon of +website is provided for reference purposes only. Admissions is competitive in nature and +November 27, 2024 (Hong Kong Time) for the first-round evaluation. Scanned copies of the +achieving the expected lower boundary does not guarantee admission to the programme. +following documents should also be uploaded by the application deadline. +The actual admission grade or score will be dependent on the overall quality of applications +(1) Official prediction of grades or scores from your current high school (if available); received in the 2025 admissions exercise. +(2) All internal and public examination results during your senior secondary education; and +Announcement of Result +(3) At least one academic reference letter Offer results will be released from December 2024 onwards. All application status will be +updated in the online application system once an admissions decision has been made. +Late submission of supporting documents will delay the consideration of application. +Applicants should check their application status online regularly. +Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other +programmes will be arranged if necessary. More Details +For further information about the application procedures and +admissions standards, please visit the Admissions Office’s website at +https://admissions.hku.hk/apply/international-qualifications. +2025 Dual Degree Admission +Separate application is required for applying to the dual degree programmes, in addition to the +application under the JUPAS or International/Non-JUPAS Admissions Scheme. For details of the +dual degrees and the application procedures, please visit the Admissions Office’s website at +https://admissions.hku.hk/programmes/university-collaborative-programmes. +38 39 + +| Develop Your Career | | | | | | Freshmen Sharing | | | | | | +| ------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | +Upon graduation, over 88% of our students begin their careers in the commerce and industry sector. Others either +| move on to further | studies or explore | opportunities | with the | civil service, education | institutions and | | | | | | | +| ---------------------- | --------------------- | -------------- | ---------- | --------------------------- | ------------------ | --- | --- | --- | --- | --- | --- | +community/social services. I chose HKU Business School for its prestigious +reputation and focus on experiential learning. I +Career Prospects believe HKUBS is the best place to help me realize my +| | | | | | | dream. BBA(Acc&Fin) | perfectly aligns | with my | With the understanding that the economy serves as | | | +| --- | --- | --- | --- | --- | --- | --------------------- | ------------------ | --------- | -------------------------------------------------- | --- | --- | +Companies/organisations that have offered full-time positions to our graduates include but not limited to the passion for numbers and financial analysis. I look a fundamental and crucial part of all other subjects, +| | | | | | | forward to engaging | in hands-on | learning | and recognizing | Hong Kong | as a central and | +| --- | --- | --- | --- | --- | --- | ---------------------- | ------------- | --------- | ----------------- | ----------- | -------------------- | +following: +| | | | | | | experiences and collaborating with diverse peers. I | | | international | commercial hub, | I chose to study | +| ------------------------- | --- | --- | ---------------------- | --- | --- | --------------------------------------------------------- | --- | --- | ------------------------------------------------ | ----------------- | -------------------- | +| Accenture Company Limited | | | Hang Seng Bank Limited | | | | | | | | | +| | | | | | | anticipate a vibrant university life filled with diverse | | | BEcon&Fin at HKU. However, after my first month | | | +Bank of China (Hong Kong) Limited HKSAR Government experiences and networking opportunities. The here, I discovered that what truly strengthens my +| | | | | | | dynamic atmosphere at HKU inspires me, and I am | | | decision is the university’s ability to cultivate my | | | +| --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | --- | ----------------------------------------------------- | --- | --- | +BDO International Limited Hong Kong Monetary Authority excited to explore opportunities that will enhance my autonomy, give me more space for self-learning, and +| | | | | | | academic journey and prepare me for a successful | | | provide numerous opportunities to meet diverse | | | +| -------------- | --- | --- | -------------------- | --- | --- | ----------------------------------------------------- | --- | --- | --------------------------------------------------- | --- | --- | +| Bloomberg L.P. | | | JPMorgan Chase & Co. | | | | | | | | | +| | | | | | | career. HKU's commitment to innovation truly inspire | | | individuals and those who share the same dreams as | | | +Citibank (Hong Kong) Limited KPMG me. I’m excited to explore new ideas and contribute mine. According to my MBTI test, I have “I” +| | | | | | | to both the academic and social communities. | | | personality, but here I have found many ways to | | | +| --------------------------------------------- | --- | --- | -------------- | --- | --- | -------------------------------------------- | --- | --- | ------------------------------------------------ | --- | --- | +| Crédit Agricole Corporate and Investment Bank | | | Morgan Stanley | | | | | | | | | +express myself. I feel like I am challenging and +improving myself every day! +| Credit Suisse Group AG | | | PricewaterhouseCoopers Limited | | | | | | | | | +| ------------------------------------- | --- | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| Deloitte Advisory (Hong Kong) Limited | | | Standard Chartered Bank (Hong Kong) Limited | | | | | | | | | +| Deloitte Touche Tohmatsu Limited | | | The Hongkong and Shanghai Banking Corporation | | | | | | | | | +Yang, Huimiao Murphy +Limited (HSBC) +Deutsche Bank +BBA(Acc&Fin) +| | | | UBS Group AG | | | | | | | Chen, Xinqiao Vicky | | +| --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | ------------------- | --- | +Ernst & Young +BEcon&Fin +Goldman Sachs (Asia) LLC +Full +| employment | | 57 | | | | | | | | | | +| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +% +98.9% +18th +of our 2023 graduates have received +at least 2 job offers upon graduation +| for the consecutive year | | | | | | | | | | | | +| ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +37,166 +$ +was the average salary of +our 2023 graduates +40 41 + +Seduced by HKU's scholarly prowess and dynamic +I chose to study IBGM at HKU Business School because of its strong +campus spirit, my path led to HKU Business School, +reputation as an innovative and established institution. Moreover, the +renowned for its cutting-edge business education. +| | | | As an Economics and Finance major, I’ve embarked | | | programme’s | emphasis on | developing well-rounded | future leaders | +| --- | --- | --- | ----------------------------------------------------- | --- | --- | --------------------------------------------------------------------------------- | ------------- | --------------------------- | ---------------- | +| | | | on my university journey brimming with intellectual | | | through global | exposure | and strategic management | training was | +| | | | rigor, leadership prospects, and a tapestry of peers | | | extremely attractive to me. I expect to have a lot of fun in university, gaining | | | | +valuable experience and developing in all areas of my life. So far, my +broadening my worldview. Every day, I’m thrilled to +The main reason I have chosen the IBGM programme expectations are being fulfilled, as HKU is truly an incredibly fun and +push my limits in academic debates and research +at HKU Business School is due to the extraordinary +| | | | projects, immerse myself in Hong Kong's unique | | | diverse place filled with unique opportunities. | | | | +| ------------------------ | ------------ | ---------- | ----------------------------------------------- | --- | --- | ----------------------------------------------- | --- | --- | --- | +| exchange opportunities | it offers. | With the | | | | | | | | +blend of Eastern and Western cultures, and build an +opportunities of studying abroad in North America +extensive network by engaging with peers and +and Europe, and visiting companies in Asia, I believe +faculty from diverse backgrounds. In addition, HKU +my global perspective will be cultivated by gaining +| | | | offers me | a platform to | give back through | | | | | +| ------------- | -------------------- | ------------- | ----------- | ---------------- | -------------------- | --- | --- | --- | --- | +| exposure to | some of the most | economically | | | | | | | | +co-curricular engagement, readying myself for a +developed regions. Additionally, I appreciate the Wong, James Dylan Cheng Yan +robust finance career. Look forward to joining the +| flexibility provided | by the IBGM | programme in | | | | | | | | +| ---------------------- | -------------- | -------------- | ----------------------------------------------- | --- | --- | --- | --- | --------- | --- | +| | | | HKU Business School community and experiencing | | | | | BBA(IBGM) | | +choosing interdisciplinary courses for a second major +an unforgettable university life! +or minor, allowing me to expand my interest in +Psychology. Overall, I believe the experiences and +alumni connections I will gain from the HKU Business +School will be extremely useful in preparing me for +my future career. +| | | | | Qu, Hongru Larina | | When it comes to HKU, the first thing that springs to | | | | +| --- | --- | --- | --- | ----------------- | --- | ------------------------------------------------------ | --- | --- | --- | +most people's minds might be its extremely high +BEcon&Fin +| | | | | | | academic | standing. For me, | it is not only the | | +| --- | --- | --- | --- | --- | --- | --------- | -------------------- | ----------------------- | --- | +academic excellence but also the unique campus +atmosphere that attracted me to become a member +Ng, Shui Yee Sophie +| | | | | | | of HKU. | HKU is indeed | an international and | | +| --- | --- | --- | --- | --- | --- | --------- | ---------------- | ----------------------- | --- | +BBA(IBGM) all-inclusive university, where I can experience great +diversity and meet elites from all around the world. I +am really eager to meet and collaborate with them. I +believe such a rich environment and atmosphere will +be highly conducive to my personal development +| | | I chose HKU because I was truly fascinated by the opportunity to become a | | | | and the expansion of my interpersonal networks. | | | | +| --- | --- | -------------------------------------------------------------------------- | --- | --- | --- | ----------------------------------------------- | --- | --- | --- | +student at one of Asia’s leading universities, known for its top-notch +educational standards. I am excited to study BEcon&Fin in the heart of a +major financial center and look forward to exploring the various internship +opportunities that HKU Business School offers. All my professors are very +enthusiastic and sociable, which motivates me to work hard. HKU exceeds +| | | my expectations, and I am glad to be here! | | | | | | Lui, Ka Wai Susie | | +| --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | ----------------- | --- | +BEcon&Fin +Kotlova, Tatiana Tanya +BEcon&Fin +42 43 + +Our Global Presence +Beijing +Israel - +Tel Aviv +Shenzhen +Hong Kong Useful Websites +The University of Hong Kong (HKU) HKU Business School +Vietnam - Undergraduate Admissions Overview of Undergraduate Programmes +Ho Chi Minh City https://admissions.hku.hk/ https://ug.hkubs.hku.hk/programme +Enrichment Programmes +HKU – Entrance Requirements +https://ug.hkubs.hku.hk/student-enrichment +For JUPAS Candidates +https://admissions.hku.hk/apply/jupas Other HKU Websites +For non-JUPAS Candidates Mainland Affairs Office (MAO) +https://admissions.hku.hk/apply/international-qualifications https://mainlandaffairs.hku.hk/ +Centre of Development and Resources for +Students (CEDARS) +https://www.cedars.hku.hk/ +44 45 + +HKU Business School +4/F, K.K. Leung Building +The University of Hong Kong +Pokfulam Road, Hong Kong SAR +(852) 3917 5343 +(852) 2549 3735 +fbe.ugenquiry@hku.hk +www.hkubs.hku.hk +HKU Business School News and Events +HKU Business School +HKU Business School +香港大学经管学院 +香港大学经管学院 +HKU Business School Undergraduate Admissions News and Events +@hkubsug +HKU Business School - UG Admissions +The information in this brochure is accurate as of October 2024 and is subject to change. \ No newline at end of file diff --git a/q166/random_k4/random_3.md b/q166/random_k4/random_3.md index 97ff7db9e8854ead7c943ee255fc2632b139760b..481e365c46bf38073a1bbbbafb7d7f3283256258 100644 --- a/q166/random_k4/random_3.md +++ b/q166/random_k4/random_3.md @@ -1,595 +1,180 @@ -FusionCharts +AMAZFIT ACTIVE +Product Manual +A2211 -FusionCharts + 01 Package Contents -White Paper +Smartwatch with band × 1 -Principles of Data Visualization - -What We See in a Visual +Charging base × 1 - Principles of Data Visualization - What We See in a Visual +Instruction Manual × 1 -Table of Contents +1 -Why We Visualize Data ............................................................................................................................................................ 3 + 02 About the Watch -The Importance of Visualizations in Business ................................................................................................... ............ 3 +Power button -Two Goals in Data Visualization ........................................................................................................................................... 4 +Heart rate sensor - Explain Data to Solve Specific Problems .................................................................................................................... 4 +Charging contacts - Explore Large Data Sets for Better Understanding ................................................................................................. 5 - -Eyesight and Memory Working in Parallel ....................................................................................................................... 6 - - The Role of Memory in Vision ........................................................................................................... ............................. 7 - -Preattentive Attributes Used by Our Working Memory .............................................................................................. 8 - -Forming Analytical Patterns Out of Preattentive Attributes ...................................................................................... 9 - -Using the Gestalt Principles to Bring Out Patterns in Visualizations ..................................................................... 11 - -An Example from Recorded Future ........................................................................................................ .......................... 12 - -Gestalt Principles ........................................................................................................... ......................................................... 13 - -Preattentive Attributes ........................................................................................................... .............................................. 14 - -Conclusion ........................................................................................................... ..................................................................... 14 - -About FusionCharts ........................................................................................................... .................................................... 15 +Note: Press and hold the power button to turn on the watch. If the watch cannot be turned on, +please charge it and then try again. 2 -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -Why We Visualize Data - -We visualize information to meet a very basic need - to tell a story. It’s one of the most primitive forms - -of communication known to man, having its origins in cave drawings dated as early as 30,000 B.C., even - -before written communication, which emerged in 3,000 B.C. Vision is the single most important faculty - -we use to communicate information. - -With the passing of time, we’ve found new ways to visualize information. Today, we’re familiar with the - -basic chart types like the line chart, bar chart, and pie chart. However, we rarely stop to think about - -why they’re more effective than bland tables, text, and numbers. Further, we can’t easily spot instances - -when they’re done wrong, and can be improved. This white paper aims to give you a foundational - -understanding of how we process visual information. You’ll be able to use this knowledge when de- - -signing your next dashboard, and have an informed opinion on how to communicate even more clearly - -and powerfully using visualizations. Even if you don’t design visualizations yourself, these principles - -will equip you with skills to better analyze visualizations you come across in the newspapers, or in your - -daily work reports. Let’s start with an understanding of why data visualizations are so important in a - -business context today. - -The Importance of Visualizations in Business - -A visual can communicate more information than a table in a much smaller space. This trait of visuals - -makes them more effective than tables for presenting data. For example, notice the table below, and - -try to spot the month with the highest sales. - -Month -Sales + 03 Pairing the Watch -Jan -45 +Zepp is a must-have app for this watch. It provides scientific sleep and workout +guidance and personalized health services to help you start a healthy lifestyle. -Feb -56 +1. Use your mobile phone to scan the QR code on the right side to download + and install the app, or search for the app in the Google Play Store or Apple + App Store and then download and install the the latest version of the app. +2. Open the app, register an account and log in. +3. Pair the watch as prompted by the app. -Mar -36 - -Apr -58 - -May -75 - -Jun -62 - -This data when visualized gives you the same information in a second or two. - -www.fusioncharts.com -FusionCharts | www.fusioncharts.com - -FusionCharts +Note: +1. For better user experience, please use the most recent version of the app. +2. The operating system needs to be Android 7.0, iOS 14.0, or above. +3. Do not pair the watch directly using your mobile phone’s Bluetooth. Follow the steps in the app to pair + your watch correctly. 3 -5 - -FusionCharts - - Principles of Data Visualization - What We See in a Visual - -Edward Tufte, a data visualization expert, says ‘Graphical excellence is that which gives to the viewer - -the greatest number of ideas in the shortest time with the least ink in the smallest space.’ This trait of - -visualizations is what makes them vital to businesses. - -Two Goals in Data Visualization - -In a business environment, visualizations can have two broad goals, which sometimes overlap. - - Explanatory +Download Zepp App - Exploratory + 04 Charging the Watch -Explain Data to Solve Specific Problems +1. Connect the watch's charging cable to a power adapter or a computer's USB port. +2. Put the watch into the charging base. Pay attention to the direction and position of the watch, and + make sure that the metal contacts on the back of the watch fit closely with the charging base. +3. When charging s tarts, the watch screen displays the charging progress. -Visuals that are meant to direct the viewer along a defined path are explanatory in nature. The bulk of - -business dashboards that we come across in day-to-day scenarios fall in this category. - -Here, the viewer starts with a question in mind. For example, ‘which country has the highest value?’. - -She then views the visualization, and finds an answer to the question. +Note: +1. Please use the charging base that comes with the watch. + Make sure that the charging base is dry before charging. +2. Please use a 1A or above power adapter. 4 -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -This type of visualization is used in many scenarios for the following tasks: - -Answer a question. E.g., How much sales did we have last quarter? - -Support a decision. E.g., We need to stock more football jerseys as they were sold out on - -most days last week - -Communicate information. E.g., Revenue is on track for this quarter - -Increase efficiency. E.g., ‘Technical specifications’ is the most viewed section in the product - -page. It should be given more visibility. - -Explore Large Data Sets for Better Understanding - -Exploratory visuals offer the viewer many dimensions to a data set, or compares multiple data sets with - -each other. They invite the viewer to explore the visual, ask questions along the way, and find answers to - -those questions. For example, below is a visualization of the State of the Union address of recent Presi- -dents.[1] - -Here, the viewer starts by familiarizing herself with the visualization, then identifies an area of interest. + 05 Disassembling and Assembling the Watch Strap -For example, which President spoke more about jobs? She then explores the ‘Jobs’ section of the visuali- +Please refer to the illustrations to disassemble or assemble the watch strap. -zation, and finds her answer. She could then move on to exploring other areas of the visualization. - -FusionCharts - -FusionCharts +Note: Once the strap is assembled, pull the strap properly to ensure it is installed successfully. 5 - Principles of Data Visualization - What We See in a Visual - -Exploratory analysis can be cyclical without a specific end point. Viewers can find many insights from a - -single visualization, and interact with it to gain understanding rather than make a specific decision. This - -type of visualization can accomplish the following tasks: - - Pose new questions - - Explore and discover - -Though they’re not as popular as the previous category, exploratory visualizations have gained promi- - -nence in recent years with the rise of big data. The high volume of data, and varied data sets that have - -become common today lend themselves easily to exploratory analysis. - -Now that we’re aware of what makes visualizations intrinsic to business, let’s dive into the mechanics of - -how we process visual information. We’ll understand the role of memory in perceiving visual information, - -and how to apply that understanding as we work with visualizations. - -Eyesight and Memory Working in Parallel - -If you’re familiar with big data terminology you would have come across the term ‘massively parallel - -processing’ (MPP), as most commonly seen in MapReduce technology that powers Hadoop. It breaks data - -down into small units, and processes each of the units in parallel. A similar process has always been used - -by us to process information with our vision as well. When we look at a visual, our eyes and our brain - -work in parallel to take in new information, and break it into small chunks. Then both the eyes and the - -brain process the chunks in parallel to find meaning. Let’s look at an example to better understand this. - -Let’s say we walk into a supermarket to buy oranges. Our eyes first scan the layout of the supermarket. - -At the same time, our brain processes the various sections of the layout, and instructs the eyes to zone - -in on the fruits section. It does this by sending signals about how fruits look from memory. The eyes then - -break the entire scanned area into parts, and scans each part to spot the fruits section. The same process - -is repeated till we zero in on the oranges in the fruits section. This process of visualizing information is - -performed by the eyes and memory working in parallel. + 06 Wearing the Watch + +1. In order to ensure the accuracy of heart rate, blood oxygen, + and other measurements, please wear your watch at a distance + of at least one finger away from the carpal, and keep the strap + at a comfortable snugness. +2. During workout, please wear it as tightly as possible to ensure + that the watch does not shake. Wear your watch at a distance + of at least one finger away from the carpal to obtain more accurate + heart rate data. After workout, you can loosen the strap appropriately + to ensure a comfortable wear experience. +3. When measuring blood oxygen saturation, wear the watch correctly. Avoid wearing the watch on the + wrist joint, keep your arm flat, maintain comfortable (appropriately tight) fitting between the watch and + the skin of your wrist, and keep your arm still throughout the measurement process. + +Note: +When affected by external factors (arm sag, arm sway, arm hair, tattoo, etc.), the measurement result may +be inaccurate or the measurement may fail and provide no output. 6 -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -The Role of Memory in Vision - -While that’s an overview of how we process visual information, let’s discuss the vital role of memory in - -our vision. There are two types of memory that come into play when we process visual information. - - Long-term memory + 07 Basic Parameters - Working memory +Product Name: Smart Watch Operating temperature: 0°C to 45°C +Product model: A2211 Water resistance rating: 5 ATM +Input: DC 5V 500mA MAX Bluetooth frequency: 2402–2480 MHz +Bluetooth version: V5.2 +Bluetooth output power: <12 dBm +View product certification information on the watch's Settings > System > Regulatory page. +Device requirements: Devices installed with Android 7.0 or iOS 14.0 or above OS versions. -Long-term Memory - -This type of memory is what makes us always expect the units to be marked off on the X and Y axes - -of a chart, or the date range selector at the top of a dashboard. This type of memory is formed by past - -interactions and experiences. It needs to be considered when designing the layout of a dashboard or - -visualization. There should be good reason to go contrary to long-term memory when deciding the basic - -structure of a dashboard or visualization. A discussion of long-term memory is out of the scope of this - -white paper. Instead, we’ll focus on the other type - working memory. - -Working Memory - -When we look at a line chart, or notice a number in a dashboard, we use our working memory to store - -just the information we need at the moment. This type of memory breaks the entire visual into small - -chunks of information, in a process appropriately called ‘chunking.’ Surprisingly, for all the complexity of - -our brain, our working memory can hold only about 3 chunks of information at any given time. When - -perceiving a complex visual, we’re constantly replacing the 3 available slots in our working memory. - -This is why when designing a visual or a dashboard, one of our goals is to limit the number of prominent - -chunks of information that we want the user to notice, and pack as much information as possible into - -each chunk. While we want to avoid information overload, a sound understanding of this concept can - -help us direct the attention of the viewer in a natural way. - -Next, Let’s discuss working memory in depth. Just as any language has alphabets, which are combined to - -form words, the language of data visualization consists of certain basic building blocks called preatten- - -tive attributes. - -FusionCharts - -FusionCharts +The Bluetooth® word mark and logos are registered trademarks owned by +Bluetooth SIG, Inc. and any use of such marks by Anhui Huami Information +Technology Co., Ltd. is under license. Other trademarks and trade names +are those of their respective owners. 7 - Principles of Data Visualization - What We See in a Visual - -Preattentive Attributes Used by Our Working -Memory - -Colin Ware, Director of the Data Visualization Research Lab at the University of New Hampshire, terms - -the basic building blocks of the visualization process as ‘Preattentive attributes.’ These attributes are what - -immediately catch our eye when we look at a visualization. They can be perceived in less than 10 millisec- - -onds, even before we make a conscious effort to notice them. Here’s a list of the preattentive attributes: - -Form -Orientation Line Length Line Width Size - -Shape Curvature Added Marks Enclosure - -Color -Intensity Hue 2-D Position - -Spatial Position + 08 Battery Safety -These attributes come into play when we analyze any visualization. Of this list, only Position, and Length - -can used to perceive quantitative data with precision. The other attributes are useful for perceiving other - -types of data such as categorical, or relational data. For example, both the pie chart and the bar chart be- - -low show the same data. But you can’t easily tell from the pie chart which is the biggest pie. That’s more - -clearly visible in the bar chart as it calls on the preattentive attribute of length. +1. This device is equipped with a built-in battery that cannot be removed or replaced. Do not disassemble + or modify the battery by yourself. +2. Disposal of a battery into fire or a hot oven, or mechanically crushing or cutting of a battery, that can + result in an explosion. +3. Leaving a battery in an extremely high temperature surrounding environment can result in an + explosion or the leakage of flammable liquid or gas. +4. A battery subjected to extremely low air pressure may result in an explosion or the leakage of + flammable liquid or gas. +5. Do not drop, hit, squeeze, or puncture the battery, and avoid exposing the battery to large external + pressure, which may cause short circuit and overheating inside the battery. 8 -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -Considering preattentive attributes can help when deciding which chart type to use for our data. While - -preattentive attributes are what we immediately identify in a visual, they aren’t the only things we notice. - -We go on to form patterns out of the preattentive attributes. - -Forming Analytical Patterns Out of Preatten- -tive Attributes - -If preattentive attributes are the alphabets of visual language, analytical patterns are the words we form - -using them. We immediately identify the preattentive attributes in a visualization. We then combine the - -preattentive attributes to seek out analytical patterns in the visual. Here are the basic analytical patterns - -that we identify when looking at a visual: - -FusionCharts - -FusionCharts + 09 Safety Notice + +1. Do not allow children or pets to bite or swallow the product or its accessories, as this may cause injury. +2. Do not place this product under excessively high or low temperatures, which may cause the product to + catch fire or explode. +3. Do not place this product near heat sources or open flame, such as ovens and electric heaters. +4. Some people may experience allergic skin reactions to plastics, leathers, fibers, and other materials, and + symptoms such as redness, swelling, and inflammation will occur after long-term contact with this + product's components. If you experience such symptoms, please discontinue use and consult your + doctor. +5. Do not use unauthorized or incompatible power adapters or data cables, which may damage the watch + and accessories, or cause fire, explosion, or other hazards. +6. The radio waves generated by this product may affect the normal operation of implanted medical + devices or personal medical devices, such as pacemakers and hearing aids. If you use any such medical + devices, consult the manufacturer for relevant usage restrictions. +7. Do not dispose of this device or its accessories as ordinary household waste. Please make sure you + dispose of or recycle this device and its accessories in accordance with local laws and regulations. 9 - Principles of Data Visualization - What We See in a Visual - -These patterns are an intrinsic part of our vision, and are frequently used when we analyze and describe - -a chart. However, as a word of caution, we are hard-wired to look for patterns in any visual information - -we notice. Sometimes we do this even when there isn’t an apparent connection or pattern in the visual. - -We’ve seen how preattentive attributes and patterns enable us to process visual information. However, - -when crafting visualizations we often want to highlight certain patterns over others. In these cases, Ge- - -stalt’s Principles come in handy. + 10 Repair and Maintenance + +1. Avoid using detergents such as soap, hand sanitizer, bath foam, or lotion for cleaning to prevent chemical + residues from irritating the skin, corroding the device, or degrading the water resistance of the device. +2. After wearing the device while bathing (for devices that support bathing), swimming (for devices that + support swimming), or sweating, please promptly clean and dry the device. +3. The leather strap is not waterproof. Avoid getting the strap wet, as this may shorten its service life. +4. When using light-colored watch straps, avoid contact with dark clothing to prevent staining. +5. Do not wear your watch during snorkeling, hot showers, hot springs, saunas (steam rooms), diving, + underwater diving, water skiing, and other activities involving high-speed water currents. +6. Do not place the device in direct sunlight for long periods. Excessively high or low ambient temperature + may cause device failure. +7. Handle with care during transportation. Keep the device dry and away from humidity. +8. Do not wear the watch in MRI examinations and other highly magnetic environments. Otherwise, the + watchmay be damaged. 10 -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -Using the Gestalt Principles to Bring Out -Patterns in Visualizations - -Gestalt principles describe how our mind organizes individual elements into groups. We can use these - -principles to highlight patterns that are important, and downplay other patterns. The image below - -illustrates the principles of Gestalt which are relevant to visualization. - -Proximity - -Similarity - -Enclosure - -Symmetry - -Closure - -Continuity + 11 Troubleshooting -Connection +Please check the FAQ on the website: https://support.amazfit.com/en. +If the product is not in use for a long period of time, it is recommended that you power off the product before +storing it. Recharge the battery every 6 months to 100% to prevent battery damage by over-discharge due to +long-term storage. -Figure & ground +12 Warranty and Return Policy -Here’s what we notice from each of the illustrations: +The Amazfit Limited Warranty covers Amazfit products against manufacturing defects beginning on the +original purchase date. The warranty period is 12 months or another specified period as the applicable +consumer laws in the consumer’s country of purchase require. Our warranty is in addition to rights provided +by applicable consumer laws. -Proximity: We see three rows of dots instead of four columns of dots because they are closer - -horizontally than vertically. - -Similarity: We see similar looking objects as part of the same group. - -Enclosure: We group the first four and and last four dots as two rows instead of eight dots. - -Closure: We automatically close the square and circle instead of seeing three disconnected - -paths. - -Continuity: We see one continuous path instead of three arbitrary ones. - -Connection: We group the connected dots as belonging to the same group. - -Symmetry: We see three pairs of symmetrical brackets rather than six individual brackets. - -Figure & ground: We either notice the two faces, or the vase. Whichever we notice becomes - -the figure, and the other the ground - -FusionCharts - -FusionCharts +Official website: en.amazfit.com +Support: support.amazfit.com/en 11 - Principles of Data Visualization - What We See in a Visual - -These principles can help us perform many tasks such as reduce the noise from charts, choose the ideal - -aspect ratio, and show relationships between elements more clearly. The Gestalt principles are a power- - -ful tool to have when crafting visualizations. - -Now that we have a good understanding of how visualizations work, let’s apply these concepts to a real- - -world example. - -An Example from Recorded Future - -We’ll consider the following visualization from Recorded Future, a web intelligence company.[2] This -visualization compares the mentions of Apple, Google, and Microsoft across the web. The data here is - -forward-looking, and considers only mentions related to the next five years. - -This visualization features two chart types - An area chart, which is grayed out in the background, and a - -bubble chart, which is color-coded in the foreground. Let’s analyze this simple visualization, and identify - -which elements from this white paper it uses. - -12 - -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -Gestalt Principles - -Figure & ground -The first thing you notice when looking at this visualization is that the bubbles stand out against the - -backdrop of the area charts. This shows the Gestalt principle of figure & ground. - -Proximity -Zoning in on the bubbles shows 3 distinct groups of bubbles. We can identify this easily because of - -how close the bubbles are to each other. - -Similarity -Further, we notice that the bubbles are of three colors - green, purple, and blue. This similarity brings - -out the grouping even more clearly. - -Analytical Patterns - -Going up, going down, remaining flat -This pattern is most visible in the area chart. We notice that the overall trend for Apple is upward, while - -Google’s Microsoft’s stays flat. - -Wide, narrow -There are many peaks along the area chart, but one of them for Microsoft is particularly noticeable in - -Jan 2015. This peak is narrower than the other peaks. - -Tightly, loosely distributed -We notice the loose distribution of bubbles for all three companies around 2018 and 2019. - -Normal, abnormal -Zoning in on the bubbles for Microsoft, we notice an outlier for Microsoft in mid-2015. The other bub- - -bles fall within a more normal range. - -FusionCharts - -FusionCharts - -13 - - Principles of Data Visualization - What We See in a Visual - -Preattentive Attributes - -Spatial position -We use the preattentive attribute of position to track the rise and fall of the area chart. Similarly, we - -notice the abnormal bubble in Microsoft’s chart because of it’s higher position compared to the other - -bubbles. - -Size / Area -The bubbles vary in size. Their size corresponds to the number of web mentions for a particular topic. - -This makes it easy to spot the important mentions, and explore them in detail. - -Hue / Color -As mentioned earlier, the color of the bubbles makes it easy to classify them into three groups. This em- - -ploys the preattentive attribute of Hue. - -Intensity -Finally, the low intensity of the area chart places it in the background, giving priority to the bubbles. - -Being aware of the preattentive attributes, analytical patterns, and Gestalt Principles can make a visual - -come alive to us. Considering we analyze visual information on a daily basis, and how important it is to - -business, it pays to know how visualizations work. - -Conclusion - -In this white paper, we dissected the process of visual perception. First, we saw that all visualizations -have a goal - explanatory, or exploratory. Second, preattentive attributes are the basic building blocks of - -a visualization, and are identified by us almost immediately. Third, preattentive attributes lead us to spot - -patterns in a visual. Finally, when crafting visualizations we could use the Gestalt Principles to prioritize - -important patterns, and downplay the noisy ones. - -In conclusion, it’s obvious that we are naturally hard-wired to visualize information in a certain way. Un- - -derstanding those basic principles of data visualization will help us craft outstanding visualizations, and - -tell compelling stories, much to the delight our colleagues, and end users. - -14 - -FusionCharts FusionCharts Principles of Data Visualization - What We See in a Visual - -About FusionCharts - -FusionCharts Suite XT is the industry’s leading enterprise-grade charting component with delightful - -JavaScript (HTML5) charts that work across devices and browsers (including IE 6, 7 and 8). Using it, you - -can create your first chart in under 15 minutes and then add advanced reporting capabilities like drill- - -down and zoom in a couple of hours after that. It comes with extensive docs, demos and personalized - -tech support that makes sure implementation is a breeze for you. - -22,000 customers and 500,000 developers in 120 countries including organizations like NASA, Micro- - -soft, Cisco, GE, AT&T and World Bank use FusionCharts Suite XT to go from data to delight in minutes. - -Learn more about how you can add delight to your products at www.fusioncharts.com - -FusionCharts - -FusionCharts - -15 - - Principles of Data Visualization - What We See in a Visual - -Reference: - -1. The Washington Post. History Through the President’s Words [1] - -2. Recorded Future. Apple, Google, and Microsoft during the next five years [2] - -Further Reading: - -1. Tufte, Edward. (2001). The Visual Display of Quantitative Information, 2nd Edition. Graphics - -2. Press.Ware, Colin. (2008). Visual Thinking. Morgan Kaufmann. - -3. Ware, Colin. (2012). Information Visualization, 3rd Edition. Morgan Kaufmann. - -4. Few, Stephen. (2009). Now You See It: Simple Visualization Techniques for Quantitative Analysis. Analytics Press. - -16 - -FusionCharts FusionCharts \ No newline at end of file diff --git a/q166/random_k4/random_4.md b/q166/random_k4/random_4.md index 0f60fe9853149c46914426d19786d3f880c8bfa5..e00065ccdc6e5b4ef75f2cc878ffa2f24cc910f1 100644 --- a/q166/random_k4/random_4.md +++ b/q166/random_k4/random_4.md @@ -1,1857 +1,611 @@ -5 -2 -0 -2 - -n -u -J -9 - -] -I - -A -. -s -c -[ - -1 -v -6 -4 -4 -7 -0 -. -6 -0 -5 -2 -: -v -i -X -r -a - -Fact in Fragments: Deconstructing Complex Claims via -LLM-based Atomic Fact Extraction and Verification - -Liwen Zhenga, Chaozhuo Lia, Zheng Liub, Feiran Huangc, Haoran Jiaa, -Zaisheng Yed, Xi Zhanga - -aKey Laboratory of Trustworthy Distributed Computing and Service (MoE), -Beijing University of Posts and Telecommunications, Beijing, 100876, China -bBeijing Academy of Artificial Intelligence, Beijing, China -cJinan University, Guangzhou, 510632, China -dDepartment of Gastric Surgery, Clinical Oncology School of Fujian Medical University, -Fujian Cancer Hospital(Fujian Branch of Fudan University Shanghai Cancer -Center), Fuzhou, 350014, China - -Abstract - -Fact verification plays a vital role in combating misinformation by assess- -ing the veracity of claims through evidence retrieval and reasoning. How- -ever, traditional methods struggle with complex claims requiring multi-hop -reasoning over fragmented evidence, as they often rely on static decompo- -sition strategies and surface-level semantic retrieval, which fail to capture -the nuanced structure and intent of the claim. This results in accumu- -lated reasoning errors, noisy evidence contamination, and limited adaptabil- -ity to diverse claims, ultimately undermining verification accuracy in com- -plex scenarios. To address this, we propose Atomic Fact Extraction and -Verification (AFEV), a novel framework that iteratively decomposes com- -plex claims into atomic facts, enabling fine-grained retrieval and adaptive -reasoning. AFEV dynamically refines claim understanding and reduces error -propagation through iterative fact extraction, reranks evidence to filter noise, -and leverages context-specific demonstrations to guide the reasoning process. -Extensive experiments on five benchmark datasets demonstrate that AFEV -achieves state-of-the-art performance in both accuracy and interpretability. - -Email addresses: zhenglw@bupt.edu.cn (Liwen Zheng), lichaozhuo@bupt.edu.cn - -(Chaozhuo Li), zhengliu1026@gmail.com (Zheng Liu), huangfr@jnu.edu.cn (Feiran -Huang), jiahaoran@bupt.edu.cn (Haoran Jia), yzs1986@fjmu.edu.cn (Zaisheng Ye), -zhangx@bupt.edu.cn (Xi Zhang) - -Preprint submitted to Nuclear Physics B - -June 10, 2025 - - Keywords: Fact Verification, Evidence Retrieval, Multi-hop Reasoning, -Atomic Fact Extraction - -1. Introduction - -Fact verification is a critical task aimed at assessing the veracity of claims -and typically involves two fundamental stages: evidence retrieval and claim -verification. During the evidence retrieval phase, relevant pieces of informa- -tion are identified from large-scale corpora. In the subsequent verification -phase, the retrieved evidence is synthesized and analyzed to determine the -veracity of the claim [1]. Fact verification serves as an essential pillar in -the modern information ecosystem, mitigating the spread of misinforma- -tion, fostering the integrity of knowledge systems, and supporting informed -decision-making in economic and political domains. - -Despite significant progress in fact verification, existing approaches pri- -marily focus on verifying relatively simple claims, in which the authenticity -can be determined through a single piece of evidence [2]. However, real-world -claims are often more complex, requiring reasoning over multiple interrelated -pieces of evidence to reach a reliable conclusion [3, 4]. Multi-hop verification -methods attempt to address this challenge by aggregating and reasoning over -multiple evidence snippets, demonstrating promising performance in han- -dling complex claims [5, 6]. Nevertheless, the multi-hop evidence retrieval -process frequently introduces a large number of loosely connected sentences, -leading to a fragmented context where critical clues may be obscured. Con- -ventional reasoning models often struggle to directly address complex verifi- -cation tasks, as they exhibit limitations in modeling long-range dependencies -and performing multi-step inference. This necessitates a strong reasoning -capability to synthesize dispersed information effectively — a strength that -large language models (LLMs) are particularly well-equipped to provide. The -emergence of large language models presents a transformative opportunity -for fact verification, particularly in verifying complex claims that demand -sophisticated multi-hop reasoning [7]. LLMs possess exceptional comprehen- -sion abilities and extensive world knowledge, enabling them to extract fine- -grained relationships among entities and infer implicit connections across -multiple pieces of evidence, which enhances their potential to overcome the -limitations of conventional methods. - -Despite the impressive generative capabilities of LLMs, directly applying -them to fact verification presents numerous challenges. A significant limi- - -2 - - Figure 1: Comparison of different claim decomposition strategies. - -tation stems from inherent issues such as training data biases, parametric -overfitting, and knowledge gaps, which collectively contribute to the phe- -nomenon of “hallucination” —instances where models generate plausible but -factually inaccurate or ungrounded content [8]. Therefore, direct reliance on -LLMs for factual verification is inherently imprudent and risky. Some cur- -rent LLM-based fact verification methods adopt an Iterative Retrieval and -Verification strategy. By incorporating external knowledge into the model’s -context and leveraging feedback mechanisms to validate the factuality, these -methods achieve improved accuracy in certain scenarios [9]. Nevertheless, -they still exhibit significant limitations when tackling complex fact verifica- -tion problems that require deeper reasoning and logical integration [10]. To -overcome these challenges, an emerging line of research focuses on atomic -fact extraction from complex claims. As depicted in Figure 1(a), this ap- -proach simplifies the problem by breaking down intricate statements into -simpler atomic facts, facilitating targeted retrieval and verification of in- -dividual components. This decomposition not only enhances the model’s -ability to handle nuanced logical relationships but also reduces the risk of -cascading errors during reasoning. - -While existing methods have shown promise in mitigating hallucination -risks in large language models, they remain weakly coupled with strong claim -understanding. Most approaches emphasize acquiring external knowledge, -while overlooking the need to improve comprehension of the original claim, -which limits their performance on complex cases [10]. These limitations - -3 - - manifest in the following ways: 1) Accumulated Errors from Static Decom- -position Strategies: Although decomposing complex claims into atomic facts -can simplify verification, conventional static strategies—whether rule-based -or zero-shot LLM-driven—fail to adapt dynamically to the semantic gran- -ularity and contextual dependencies inherent in claims. These approaches -tend to prioritize syntactic fragmentation over a contextual understanding of -claim intent and structure. Furthermore, the absence of explicit supervision -during decomposition amplifies error propagation, especially in multi-hop -reasoning scenarios. 2) Noisy Evidence Contamination in Surface-Level Se- -mantic Alignment: Semantic similarity-based fact verification frameworks -are inherently vulnerable to context-agnostic retrieval, where surface-level -alignment overrides deeper claim-evidence compatibility. This often leads to -the inclusion of extraneous evidence fragments and, in multi-hop reasoning, -overemphasis on isolated subconcepts while neglecting causal dependencies -embedded in the original claim. 3) Constraints of Static Prompting on Claim -Understanding and Reasoning: Most existing LLM-based verification meth- -ods rely on static, complex prompts to guide reasoning. However, verifying -diverse claim types requires varied reasoning strategies and deeper contextual -understanding—capabilities that static prompts often fail to support. This -rigidity limits the model’s capacity to grasp claim intent and structure, lead- -ing to shallow or misaligned reasoning. Consequently, the model struggles to -generalize across verification scenarios, hindering overall performance. - -To address the aforementioned challenges, this paper proposes an Atomic -Fact Extraction and Verification (AFEV) framework for reliable fact verifi- -cation, which progressively decomposes complex claims into simpler atomic -facts. As illustrated in Figure 1(b), the LLM-enabled extractor transforms -intricate reasoning tasks into manageable subtasks, allowing for more pre- -cise, targeted verification of individual components. To mitigate the error -accumulation caused by static decomposition strategies, the framework em- -ploys an iterative atomic fact extraction strategy. This strategy leverages the -understanding of previously verified atomic facts to guide the subsequent de- -composition process, ensuring that claims are decomposed more coherently -with each iteration. By integrating fine-grained external evidence and dy- -namic demonstrations, the proposed approach ensures accurate validation of -atomic facts while maintaining both interpretability and reliability through- -out the verification pipeline. Specifically, the pretrained evidence re-ranker is -utilized to filter out irrelevant or noisy information for each atomic fact, pri- -oritizing highly relevant evidence and reducing the impact of distractive data - -4 - - on the verification results. In addition, instead of relying on static prompting -methods, our framework dynamically retrieves relevant instances specific to -each atomic fact, tailoring the reasoning process to the unique characteris- -tics of the fact under verification. During the atomic fact verification phase, -the reasoner generates both a factuality label for each atomic fact and a -corresponding rationale, guided by dynamic demonstrations and reranked -evidence. This process leverages LLMs’ capacity for adaptive reasoning and -interpretable decision-making. Finally, the reasoner synthesizes the verified -atomic facts to form a comprehensive judgment of overall factuality. Experi- -mental results on several benchmark datasets validate the superiority of this -method. - -The main contributions of this paper include: - -• This paper tackles the challenge of understanding and verifying com- -plex claims by introducing a novel full-cycle fact verification framework. -It iteratively decomposes and refines intricate claims into atomic facts, -thereby strengthening claim comprehension and markedly improving -both accuracy and interpretability. - -• The proposed AFEV framework comprises three key modules: the Dy- -namic Atomic Fact Extraction module, which is built upon LLMs and -iteratively decomposes complex claims into atomic facts, thereby reduc- -ing reasoning complexity and mitigating error propagation; the Refined -Evidence Retrieval module, which retrieves fine-grained evidence and -dynamically samples demonstrations to guide more reliable reasoning; -and the Adaptive Atomic Fact Verification module, which harnesses -the advanced reasoning capabilities of LLMs to generate verification -results and rationales for each atomic fact, further enhancing subse- -quent decomposition and verification processes. - -• Extensive experiments on five fact verification datasets demonstrate the -superior performance and effectiveness of AFEV in handling complex -claims. - -2. Related Work - -Recent research on multi-hop fact verification employs iterative evidence -retrieval and reasoning mechanisms to model inference paths and capture - -5 - - critical clues for accurate claim verification [11]. HESM [12] employs a hierar- -chical multi-hop evidence retrieval mechanism; however, its heavy reliance on -entity links constrains its applicability across different scenarios. MDR [13] -and Baleen [5] leverage dense retrieval to link claims with previously retrieved -evidence, enhancing the original query. RECOMP [14] enhances retrieval by -explicitly capturing semantic-level interactions across evidence pieces. Dur- -ing the claim verification stage, existing methods construct multi-hop evi- -dence graphs and perform graph reasoning to extract key clues for assess- -ing claim veracity [15]. Additionally, some approaches introduce learnable -perturbation mechanisms to remove redundant information, enabling more -fine-grained reasoning [16]. - -Existing multi-hop fact verification methods are somewhat effective in -handling complex claims. However, their supervised training paradigm relies -heavily on large-scale annotated data, limiting both scalability and cross- -domain adaptability. -In contrast, Large Language Models (LLMs) offer -a paradigm shift for claim verification by leveraging deep language under- -standing, multimodal fusion, and logical reasoning. Moreover, the Retrieval -Augmented Generation (RAG) framework, which dynamically integrates ex- -ternal evidence with internal knowledge, has been shown to mitigate halluci- -nation issues in LLMs, thereby enhancing the accuracy and interpretability of -fact-checking [17, 18]. Self-RAG [8] introduces a confidence-based dynamic -evidence retrieval decision mechanism, while FFRR [19] innovatively con- -structs dual-dimensional feedback signals, including document-level seman- -tic integrity and query-level relevance, to iteratively optimize the retrieval -model. Meanwhile, CRAG [20] incorporates a lightweight retrieval qual- -ity assessment module that dynamically triggers completion or re-retrieval -operations based on probabilistic thresholds. Furthermore, some studies en- -hance fact-checking by constructing adversarial argument spaces, strengthen- -ing model robustness in reasoning under conflicting evidence and advancing -the paradigm toward dialectical reasoning [21, 22]. - -Although the aforementioned methods refine verification results through -intrinsic feedback mechanisms, they still face challenges in handling com- -plex factuality tasks. To address this issue, recent studies have focused on -dynamic iterative retrieval-verification frameworks, which integrate external -knowledge into the model’s context and leverage feedback mechanisms for -factuality validation [10, 23]. RAC [9] decomposes complex claims into in- -dependently verifiable atomic fact units and employs reinforcement learning -to optimize retrieval queries and weight evidence credibility, enabling pro- - -6 - - gressive factual correction. FIRE [24] introduces an iterative retrieval and -verification mechanism, incorporating an Adaptive Retrieval Trigger that dy- -namically generates multi-granularity retrieval instructions based on evidence -completeness metrics. Notably, the effectiveness of these approaches hinges -on the precision of claim decomposition. Consequently, ongoing research ex- -plores methods for breaking down complex claims into more verifiable atomic -facts [25]. Nevertheless, existing static claim decomposition strategies strug- -gle to maintain information completeness and coherence, leading to broken -reasoning chains and exacerbating redundancy, which in turn increases the -risk of error accumulation during the inference process. - -3. Problem Definition - -1, ei - -2, . . . , ei - -The proposed framework for fact verification employs an iterative atomic -fact extraction process, which decomposes a complex claim C into a set of -atomic facts F = {F1, F2, . . . , FT }. For each atomic fact Fi, relevant evidence -Ei = {ei -k} is retrieved from an external corpus D, ensuring that -the verification process is grounded in factual information. To enhance the -reasoning module’s precision, dynamic demonstrations Ai = {ai -d} -are constructed by retrieving highly relevant claims associated with Fi. Each -demonstration ai consists of a claim, its ground-truth label, corresponding -gold evidence, and a rationale, providing the reasoning module with contex- -tual guidance and improving its ability to handle nuanced verification tasks. -The verification process for each atomic fact Fi is formally defined as: - -2, . . . , ai - -1, ai - -yi, ri = Reasoner(Fi, Ei, Ai), - -(1) - -where yi represents the verification result and ri denotes the rationale gener- -ated by the reasoning module. Finally, the verification results and rationales -for all atomic facts are aggregated to yield the overall verification label y∗ -for the original claim C. - -4. Methodology - -As illustrated in Figure 2, the proposed AFEV model performs fact veri- -fication through a three-stage framework: 1) the Dynamic Atomic Fact Ex- -traction module iteratively decomposes the original claim into atomic facts, -simplifying complex reasoning into manageable sub-problems; 2) the Re- -fined Evidence Retrieval module utilizes a pretrained evidence reranker to - -7 - - Figure 2: Framework of the proposed AFEV model. - -eliminate irrelevant information and retrieve highly relevant instances as dy- -namic reasoning demonstrations; and 3) the Adaptive Atomic Fact Verifica- -tion module leverages the retrieved evidence and demonstrations to validate -atomic facts, extracting key evidence to support the verification process and -provide interpretable rationales for the results. - -4.1. Dynamic Atomic Fact Extraction - -The Dynamic Atomic Fact Extraction module serves as the foundational -component of our fact verification framework, designed to decompose com- -plex claims into simpler, atomic facts. This decomposition is critical for -transforming intricate reasoning tasks into more manageable subtasks, en- -abling targeted and precise verification of individual facts. To address the -inherent limitations of static decomposition strategies, we propose an itera- -tive decomposition approach, which dynamically refines the extraction pro- -cess based on previously verified facts, thereby mitigating the risk of error -propagation and enhancing the overall reliability of the verification pipeline. -Given a complex claim C, our approach incrementally extracts atomic -facts, denoted as F = {F1, F2, . . . , FT }, in a sequential manner. Instead of a -static decomposition, where all atomic facts are generated all at once with- - -8 - - out incorporating verification feedback, our framework adopts an iterative -extraction strategy that dynamically refines the process based on previously -extracted and verified facts. - -At each iteration t, the model determines whether the extraction process -should terminate or continue to extract the next atomic fact Ft. This deci- -sion is based on assessing the coverage of the original claim C by the set of -extracted atomic facts F1:t−1, ensuring that the decomposition remains both -comprehensive and relevant. If the extracted facts collectively provide suffi- -cient coverage of the claim, the process terminates. Otherwise, the system -proceeds with extracting the next atomic fact Ft. - -When the extraction continues, Ft is generated by the LLM-based Ex- -tractor, which dynamically refines its output based on the understanding of -previously extracted and verified facts. Specifically, the extractor generates -predictions conditioned on the atomic facts from earlier iterations, denoted -as F1:t−1 = {F1, F2, . . . , Ft−1}, as well as their corresponding verification out- -comes. This iterative refinement process helps the model adapt its decom- -position strategy, minimizing redundancy while ensuring that the extracted -facts remain faithful to the claim structure. By incorporating verification -results into the extraction loop, the framework reduces error propagation, -improving both accuracy and robustness. - -The fact extraction process can be formalized as a function where the -extractor takes as input the original claim C, along with the previously ex- -tracted facts and their verification results, to generate the next atomic fact: - -Ft = Extractor(C, F1:t−1, y1:t−1, r1:t−1). - -(2) - -Here, F1:t−1 represents the set of atomic facts extracted up to iteration t, -while y1:t−1 and r1:t−1 captures their corresponding verification results and ra- -tionales. This formulation ensures that each newly extracted fact is informed -by prior knowledge, facilitating a structured and adaptive claim decompo- -sition process. Through this iterative methodology, the extraction process -continuously refines its understanding of the claim while progressively veri- -fying its individual components, leading to a more interpretable and reliable -fact verification pipeline. - -4.2. Refined Evidence Retrieval - -To ensure reliable verification and mitigate hallucination risks, external -evidence Et is retrieved for each Ft. It is obtained by Evidence Retrieval and - -9 - - Reranking, which is designed to ensure that the evidence used for verifying -atomic facts is both highly relevant and free from noise. To further improve -accurate reasoning, we incorporate Dynamic Instance Retrieval to identify -related instances that align with each atomic fact Ft. This allows the model -to leverage highly relevant dynamic demonstrations and enhance the overall -verification result. - -4.2.1. Evidence Retrieval and Reranking - -The retrieval process begins by identifying a preliminary set of candidate -evidence E′ -t for each atomic fact Ft from a large corpus D. To capture -semantic relevance, we compute the cosine similarity between Ft and each -candidate ej ∈ D. The top-k′ candidates with the highest similarity scores -are selected as the initial retrieval set: - -score(ej, Ft) = - -f (ej) · f (Ft) -∥f (ej)∥ ∥f (Ft)∥ - -, - -(3) - -E′ - -t = {ej ∈ D | j ∈ arg topk′ score(ej, Ft)} , -where f (·) is a text encoder that produces dense vector embeddings. This -retrieval step ensures coarse filtering of the corpus, yielding a candidate set -for downstream reranking and reasoning. - -(4) - -However, due to the potential for irrelevant or noisy information in the -initial retrieval stage, a pre-trained reranker is employed to filter and reorder -the evidence. The reranker evaluates each piece of retrieved evidence and -constructs a ranked subset of evidence Et = {e1 - -t } as follows: - -t , . . . , ek - -t , e2 - -Et = Reranker(E′ - -t, Ft). - -(5) - -To optimize the reranker, we formulate a supervised training objective -based on relevance judgments obtained from LLMs. During training, instead -of randomly sampling, we utilize LLMs to select positive and negative sam- -ples from the initial retrieved evidence set E′ -t to enhance the model’s discrim- -inative ability. Specifically, LLMs assess how each piece of evidence supports -the verification of fact Fi, assigning relevance scores accordingly. The highest- -scoring evidence is chosen as the positive sample e+, and the m lowest-scoring -pieces of evidence are selected as negative samples E− = {e− -m}. -Subsequently, we apply an InfoNCE loss Lr as the learning objective of the -reranker: - -2 , . . . , e− - -1 , e− - -La = − - -(cid:32) - -log - -1 -N - -N -(cid:88) - -i=1 - -exp (f (Fi) · f (e+)) /τ -j=1 exp (cid:0)f (Fi) · f (e− - -j )(cid:1) /τ - -(cid:80)m - -(cid:33) - -, - -(6) - -10 - - where N is number of facts in each training batch, f (·) is the encoder and τ -is the temperature coefficient. - -This retrieval and reranking process ensures that only the most contribu- -tive evidence is retained, thereby reducing the impact of distractive or irrel- -evant information on the verification results. - -4.2.2. Dynamic Instance Retrieval - -In addition to retrieving evidence, the module dynamically identifies re- -lated claims and instances At that align with the atomic fact Ft. These in- -stances serve as dynamic demonstrations, providing in-context prompts that -enhance the model’s ability to reason about Ft by offering analogous exam- -ples and contextual guidance. The Instance Retriever component searches for -claims that share logical or semantic similarities with Ft from the training set, -ensuring that the reasoning process is tailored to the unique characteristics of -the fact under verification. This is achieved by computing the semantic sim- -ilarity between Ft and the claims using a pre-trained language model, with -claims that achieve high similarity scores being prioritized for retrieval. The -retrieved instances At = {at -d} are then used to guide the reasoning -module, enabling it to generate more accurate and interpretable verification -results. The dynamic instance retrieval process can be formalized as follows: - -2, . . . , at - -1, at - -sim(aj, Ft) = - -f (aj) · f (Ft) -∥f (aj)∥ ∥f (Ft)∥ - -, - -(7) - -At = {aj ∈ C | j ∈ arg topd sim(aj, Ft)} , -where f (·) is a text encoder and At represents the set of dynamically retrieved -demonstrations for Ft. - -(8) - -By incorporating these contextually relevant examples, the reasoning -module is better equipped to handle complex and nuanced verification tasks, -improving both the accuracy and interpretability of the final results. This -approach not only enhances the model’s reasoning capabilities but also en- -sures that the verification process is robust and adaptable to diverse claim -structures and contexts. - -4.3. Adaptive Atomic Fact Verification - -The Adaptive Atomic Fact Verification Module constitutes the final and -most critical stage of our fact verification framework, tasked with validating -each atomic fact Ft using the retrieved external evidence Et and dynamically - -11 - - retrieved instances At. The primary objective of this module is to generate -a factuality label yt (e.g., true, false, or unverifiable). To mitigate the po- -tential mismatch in granularity between the fine-grained atomic fact Ft and -the coarser instances, which may otherwise provide suboptimal contextual -guidance, we incorporate the original claim C from which Ft is derived as an -additional input to the verification module. This addition helps maintain a -certain level of granularity alignment between the guidance and the fact. - -In addition to the factuality label, the Reasoner produces a detailed ratio- -nale rt, which highlights the key pieces of evidence and reasoning steps that -led to the conclusion. This rationale not only enhances the transparency and -interpretability of the verification process but also provides valuable context -for subsequent atomic fact extraction. For instance, the rationale may reveal -implicit information (e.g., entity names or relationships) that refines the de- -composition of the next atomic fact, leading to more precise and contextually -grounded verification. This iterative feedback mechanism is crucial for han- -dling complex claims, as it ensures that each step builds upon the insights -gained from previous verifications. The reasoning process can be formalized -as follows: - -yt, rt = Reasoner(Ft, C, Et, At). - -(9) - -By integrating evidence, demonstrations, and interpretable reasoning, -this module not only delivers accurate verification outcomes but also fosters -interpretability, making it a reliable system for handling complex claims. - -Once all atomic facts F = {F1, F2, . . . , FT } have been verified, the final -step is to aggregate the results to produce a comprehensive conclusion about -the overall factuality of the original claim C. The aggregation process com- -bines the individual verification results yt for each atomic fact Ft into a final -claim validation y∗. The aggregation process can be formalized as follows: - -y∗ = Aggregate (cid:0){Ft}T - -t=1, {yt}T - -t=1, {rt}T - -t=1 - -(cid:1) , - -(10) - -where y∗ represents the final claim validation, and {yt}T -factuality labels for all atomic facts. - -t=1 denotes the set of - -Overall, by integrating external evidence Et, the Atomic Fact Verification -module ensures that verification is anchored in reliable and relevant evidence, -reducing the risk of hallucination. The use of dynamic demonstrations At -further enhances the adaptability of the reasoning process, allowing the model -to handle a wide range of claim types and complexities. Additionally, the -generation of interpretable rationales rt provides transparency, enhancing - -12 - - the transparency and interpretability of the verification process. Finally, -the aggregation process ensures that the final claim validation y∗ is both -comprehensive and coherent, reflecting the overall truthfulness of the claim -through its atomic verifications. - -4.4. Computational Complexity Analysis - -At the t-th iteration, we first invoke an LLM to generate the atomic -fact Ft, conditioned on the original claim, previously generated facts F1:t−1, -rationales r1:t−1, labels y1:t−1, and a task-specific prompt. While invoking -an LLM incurs non-negligible cost, the inherently compact nature of atomic -facts and rationales enables us to control input size by limiting their lengths, -thereby reducing the total number of tokens and improving inference effi- -ciency. Moreover, since no gradient computation is involved and each claim -can be processed independently, this process is parallel and highly scalable -during inference. Similarly, in the verification stage for each atomic fact, -we observe controllable inference costs and efficient LLM-driven reasoning, -making the process both practical and scalable. - -To improve efficiency in evidence retrieval, we adopt a two-stage ap- -proach. A bi-encoder retriever is first used to efficiently retrieve k′ coarse -candidate evidence snippets, with time complexity approximately O(log N ) -for a corpus of size N . These candidates are then reranked by a cross-encoder -reranker with time complexity O(k′d), where d is the embedding dimension. -Compared to using the reranker directly for precise retrieval, this cascaded -strategy significantly reduces computational cost while maintaining retrieval -quality. Additionally, to reduce computational overhead during instance re- -trieval, we avoid searching over the full training set. Instead, we construct -a candidate pool by randomly sampling a subset of training instances at a -1:1 ratio with the test set, preserving diversity while substantially improving -retrieval efficiency. - -In summary, our framework is designed with several efficiency- and cost- -oriented strategies that enable scalable deployment while maintaining man- -ageable inference overhead. - -5. Experiments - -5.1. Experimental Setup - -This section offers a detailed description of the datasets utilized for train- -ing and evaluation, the assessment metrics applied, the baseline models used - -13 - - for comparison, and the implementation details. - -5.1.1. Datasets. - -In this study, we leverage multiple publicly available fact verification -datasets for model training and evaluation, including HOVER [26], Poli- -tiHop [3], RAWFC [27], LIAR [28] and LIAR-PLUS [29]. Specifically, the -LIAR and LIAR-PLUS datasets are annotated with six distinct categories of -truthfulness labels, enabling a fine-grained classification of claims based on -their veracity. In contrast, HOVER, PolitiHop, and RAWFC are structured -around three classification categories, providing a more general framework -for fact verification tasks. - -Table 1: Dataset Sizes for LIAR-PLUS, LIAR, HOVER, PolitiHop and RAWFC. - -Train Dev - -Test - -LIAR-PLUS 10,240 -10,269 -LIAR -18,171 -HOVER -592 -PolitiHop -1,612 -RAWFC - -1,284 -1,284 -4,000 -141 -200 - -1,267 -1,283 -4,000 -200 -200 - -5.1.2. Baselines. - -To validate the effectiveness of the proposed AFEV model, we employ -two distinct sets of existing models as baselines, ensuring a comprehensive -evaluation. The first set of baselines comprises state-of-the-art multi-hop fact -verification methods. As demonstrated in Table 2, Pipeline [30] serves as a -general-purpose baseline, offering a broad perspective on the task. IB [31], -TSS [32], and LR [33] represent single-granular models, which focus exclu- -sively on retrieving sentence-level evidence during the retrieval phase. Build- -ing upon insights from CURE [34], we further enhance our baseline com- -parison by introducing a suite of multi-granular models. These models are -designed to capture both sentence-level and token-level rationales, enabling -a more nuanced and precise verification process. Specifically, DeClarE [35], -FRESH [36], SHAP [37], L-INTGRAD [38], DIFFMASK [39], CURE [34], -and VMASK [40] are integrated into our evaluation framework. This dual- -level approach not only enhances the model’s ability to discern subtle nuances -in the evidence but also underscores the superiority of our proposed method -in handling complex multi-hop reasoning tasks. - -14 - - The second set of baselines consists of advanced LLM-based reasoning -methods, which leverage the powerful capabilities of large language models -for fact verification. As illustrated in Table 3, these baselines include Stan- -dard Prompt [41], Vanilla CoT [42], Search-Augmented CoT [43], ReAct [44], -RAFTS [21], FFRR [19] and HiSS [43]. All of these methods rely heavily -on LLMs but employ distinct reasoning strategies to achieve fact verifica- -tion. Specifically, Standard Prompt directly queries the LLM to return the -verification label for the claim. In contrast, the other baselines adopt more so- -phisticated, step-by-step verification strategies, enabling the models to break -down complex claims into intermediate reasoning steps. To ensure a more -granular and comprehensive comparison, we also incorporate several strong -supervised models into our evaluation framework. These models include -CNN [45], RNN [46], DeClarE [35], SentHAN [47], SBERT [48], GenFE [49], -and CofCED [50]. These supervised models represent a diverse range of ar- -chitectures and techniques, providing a robust benchmark for evaluating the -performance of our proposed method. - -5.1.3. Evaluation Metrics. - -For the multi-hop fact verification baselines, we conduct experiments on -the LIAR-PLUS, HOVER, and PolitiHop datasets, ensuring alignment with -established benchmarks in the field. To maintain consistency with prior -works, particularly CURE [34], we employ Label Accuracy and the F1 score -as the primary evaluation metrics. These metrics are well-suited for assess- -ing the model’s ability to correctly classify claims while balancing precision -and recall, which is critical for multi-hop reasoning tasks that require the -integration of multiple evidence pieces. - -For evaluating LLM-based reasoning methods, we utilize the LIAR and -RAWFC datasets, which are widely recognized for evaluating the reason- -ing capabilities of language models. To ensure comparability with HiSS and -RAFTS, we adopt precision, recall, and the F1 score as the evaluation met- -rics. These metrics provide a comprehensive assessment of the model’s per- -formance. - -5.1.4. Implementation Details - -Our framework employs GPT-3.5 to construct both the atomic fact ex- -tractor and the reasoner, leveraging its advanced natural language under- -standing and generation capabilities. For the evidence retriever, we utilize a -supervised fine-tuned language model, which is specifically trained to iden- - -15 - - Figure 3: Prompt for dynamic atomic fact extraction. - -tify and retrieve relevant evidence from large corpora. To further enhance -the precision of evidence retrieval, we use GPT-3.5 to sample both positive -and negative examples, which are then used to train the evidence reranker. -This approach ensures that the reranker can effectively distinguish between -high-quality and low-quality evidence, improving the overall reliability of the -verification process. Additionally, to ensure a fair and consistent comparison, -all LLM-based baselines reported in Table 3 also use GPT-3.5 as their back- -bone. This alignment guarantees that the performance differences observed -are solely attributable to the methodological advancements of our approach, -rather than discrepancies in the underlying model capabilities. - -After extensive parameter tuning, we set the final number of evidence re- -trieval candidates to 5 and the number of reranked evidence to 2. Addition- -ally, for each atomic fact to be verified, we dynamically select 1 demonstration -to guide the reasoning process. This dynamic demonstration selection ensures -that the model adapts to the specific context of each claim, enhancing its -ability to handle diverse and complex verification scenarios. Moreover, Fig- -ure 3 and Figure 4 provide illustrative examples of the prompts employed for -dynamic atomic fact extraction and adaptive fact verification, respectively. - -5.2. Overall Performance - -Compared to single-granular models, multi-granular methods, particu- -larly CURE and VMASK, demonstrate the ability to capture fine-grained - -16 - - Figure 4: Prompt for adaptive fact verification. - -evidence. These methods not only aggregate a richer set of clues by leveraging -both sentence-level and token-level rationales, but also reduce the reasoning -burden on the model through the explicit use of token-level evidence, which -provides more precise and localized information. As a result, as illustrated -in Table 2, multi-granular methods generally outperform traditional single- -granular models in fact verification tasks, achieving higher accuracy. How- -ever, our proposed AFEV method, despite relying on sentence-level single- -granular evidence for each atomic fact verification, achieves state-of-the-art -performance. This remarkable result highlights that while fine-grained ev- -idence can enhance performance, the design of the retrieval and reasoning -mechanisms also plays a critical role in achieving superior results. - -Since AFEV utilizes LLMs for atomic fact extraction and verification, we -further compare it with LLM-based step-by-step baselines to highlight its -superiority. We choose to evaluate on LIAR and RAWFC because they rep- -resent particularly challenging verification scenarios where traditional super- -vised pipelines often struggle. LIAR lacks annotated evidence, and RAWFC -provides only weakly structured evidence, making both datasets better suited -for assessing the effectiveness of LLM-based approaches. As shown in Ta- -ble 3, despite the remarkable reasoning capabilities of LLMs, models relying -on simple reasoning strategies still underperform compared to supervised fact -verification models. This performance gap may stem from limitations in evi- - -17 - - Table 2: Overall performance on LIAR-PLUS, HOVER and PolitiHop. The improvement -over the best baseline method (i.e., VMASK) is statistically significant (sign test, p-value < -0.01). The best results are highlighted in bold, and the second-best results are underlined. - -LIAR-PLUS - -HOVER - -PolitiHop - -Models - -Pipeline -IB -TSS -LR - -DeClarE -FRESH -SHAP -L-INTGRAD -DIFFMASK -CURE -VMASK - -LA - -58.11 -62.52 -62.39 -76.52 - -47.73 -43.45 -76.39 -71.72 -58.50 -82.10 -82.62 - -Macro-F1 - -53.93 -60.48 -61.72 -75.19 - -21.54 -41.37 -75.31 -69.84 -48.03 -80.78 -81.46 - -LA - -62.55 -56.78 -53.68 -51.10 - -50.83 -60.28 -59.83 -50.03 -71.53 -76.98 -74.38 - -Macro-F1 - -62.44 -56.74 -51.11 -40.50 - -50.76 -60.14 -58.18 -53.86 -71.30 -76.89 -73.69 - -LA - -65.96 -68.79 -65.25 -70.21 - -69.50 -61.70 -59.57 -69.50 -67.38 -69.50 -72.34 - -Macro-F1 - -41.73 -54.89 -43.34 -47.12 - -27.34 -44.35 -40.71 -27.34 -44.71 -32.36 -55.80 - -AFEV - -83.73 - -83.12 - -78.87 - -78.76 - -74.14 - -57.69 - -dence retrieval quality and the inherent hallucination issues of LLMs, which -can lead to unreliable or factually incorrect outputs. To address these chal- -lenges, methods like RAFTS and HiSS have introduced enhanced retrieval -and reasoning strategies, guiding LLMs to achieve improved performance in -fact verification tasks. However, our proposed model surpasses even these ad- -vanced baselines, achieving state-of-the-art results on both the RAWFC and -LIAR datasets. This superior performance demonstrates the effectiveness -of our approach in integrating evidence retrieval, reranking, and reasoning -while effectively addressing the limitations of LLMs. - -5.3. Performance on Evidence Retrieval - -Evidence retrieval is an indispensable component of fact verification tasks, -as it directly impacts the quality and reliability of the verification process. -Since AFEV retrieves sentence-level evidence, we aim to validate the effec- -tiveness of our retrieve-then-rerank strategy by comparing its performance -with single-granular models, following the methodology of prior work [34]. -Specifically, we conduct experiments on the HOVER, LIAR-PLUS and Poli- -tiHop dataset, which is well-suited for evaluating multi-hop retrieval capabil- - -18 - - Table 3: Overall performance on LIAR and RAWFC. The improvement over the best -baseline method (i.e., RAFTS) is statistically significant (sign test, p-value < 0.01). - -Models - -CNN -RNN -DeClarE -SentHAN -SBERT -GenFE -CofCED - -Standard Prompt -Vanilla CoT -Search-Augmented CoT -ReAct -HiSS -FFRR -RAFTS - -RAWFC - -LIAR - -P - -38.8 -41.4 -43.4 -45.7 -51.1 -44.3 -53.0 - -48.5 -42.4 -47.2 -51.2 -53.4 -56.5 -62.8 - -R - -38.5 -42.1 -43.5 -45.5 -46.0 -44.8 -51.0 - -48.5 -46.6 -51.4 -48.5 -54.4 -57.4 -52.6 - -F1 - -38.6 -41.7 -43.4 -45.6 -48.4 -44.5 -52.0 - -48.5 -44.4 -49.2 -49.8 -53.9 -57.0 -57.3 - -P - -22.6 -24.4 -22.9 -22.6 -24.1 -28.0 -29.5 - -29.1 -22.6 -27.5 -33.2 -46.8 -34.5 -47.1 - -R - -22.4 -21.2 -20.6 -20.0 -22.1 -26.2 -29.6 - -25.1 -24.2 -23.6 -29.0 -31.3 -32.6 -37.9 - -F1 - -22.5 -22.7 -21.7 -21.2 -23.1 -27.1 -29.5 - -27.0 -23.7 -25.4 -31.0 -37.5 -33.5 -42.0 - -AFEV - -63.3 - -57.6 - -60.2 - -48.2 - -40.3 - -43.9 - -ities. As shown in Table 4, AFEV achieves state-of-the-art retrieval perfor- -mance across all datasets, outperforming existing single-granular baselines. -This superior performance underscores the effectiveness of our retrieve-then- -rerank strategy, which combines comprehensive retrieval with precise rerank- -ing to identify the most relevant and high-quality evidence. - -5.4. Ablation Study - -To thoroughly evaluate the contribution of each component in AFEV, -we conduct a series of ablation experiments on the HOVER dataset. These -experiments are designed to isolate and analyze the impact of key modules -within our framework, providing insights into their individual and collective -roles in achieving superior performance. The ablation settings are described -as follows: - --w/o Atomic Fact Extraction: We skip the decomposition of initial -complex claims and directly verify the claims without breaking them down -into atomic facts. This ablation is designed to demonstrate the necessity of - -19 - - Table 4: Retrieval performance comparison. - -LIAR-PLUS - -HOVER - -PolitiHop - -Models - -F1 - -Precision Recall - -0.6677 -Pipeline -0.3777 -IB -0.4324 -TSS -0.6242 -LR -CURE -0.6789 -AFEV 0.6952 - -0.9427 -Pipeline -0.6236 -IB -0.6883 -TSS -0.9419 -LR -CURE -0.9376 -AFEV 0.9423 - -0.6390 -Pipeline -0.4180 -IB -0.4272 -TSS -0.5699 -LR -CURE -0.6947 -AFEV 0.7021 - -0.7450 -0.3927 -0.6349 -0.6776 -0.8072 -0.8342 - -0.9028 -0.7018 -0.9026 -0.9029 -0.9045 -0.9096 - -0.5986 -0.5106 -0.5177 -0.5674 -0.6584 -0.6737 - -0.6564 -0.3967 -0.3469 -0.6381 -0.6329 -0.6427 - -0.9900 -0.5783 -0.5755 -0.9988 -0.9877 -0.9923 - -0.8234 -0.3902 -0.4044 -0.6657 -0.8403 -0.8137 - -atomic fact extraction in handling complex claims, particularly in multi-hop -reasoning scenarios where decomposing claims into simpler units is essential -for accurate verification. - --w/o Iterative Extraction: The iterative decomposition strategy is -abandoned, and the initial complex claims are decomposed into atomic facts -in a single step. This setting aims to highlight the importance of iterative -refinement in accurately identifying and extracting atomic facts, as a one- -shot decomposition may overlook nuanced dependencies and relationships -within the claim. - --w/o Rationales: The use of rationales generated during intermediate -steps of the iterative decomposition process is removed. Specifically, the -model does not leverage the explanatory information about verification re- -sults from previous steps to guide the extraction of subsequent atomic facts. -This ablation is intended to evaluate the role of explanatory feedback in -improving the precision and coherence of atomic fact extraction. - -w/o Reranking: We disable the fine-tuned reranker and directly use -the initially retrieved evidence without further filtering. This setting aims +The University of Tokyo + +Outline of University + +The University of Tokyo ( UTokyo) was established +in 1877 as the first national university in Japan. As +a leading research university, UTokyo offers courses +in essentially all academic disciplines at both under- +graduate and graduate levels and conducts research +across the full spectrum of academic activity. With +around 4,000 faculties and about 28,000 students, +evenly divided between undergraduate and graduate +students, about 10% of which are international stu- +dents, UTokyo is known for the excellence of its facul- +ty and students and ever since its foundation. + +International Activities + +UTokyo’s initiative of “Constructing a Global Campus Model” is +funded by MEXT as part of its Top Global University Project. This ini- +tiative has the following six characteristics: +1. World-class, cutting-edge research taking place in a wide range of +academic fields +2. A comprehensively reformed education system suitable for the +modern Globalization Era +3. A wide range of degree programs and courses throughout the +curriculum available in English +4. A high level of research and teaching carried out in Japanese, and +courses in languages other than English taking place systematically +5. A diverse environment where academic and administrative staff +and students of the University can teach, research and work togeth- +er regardless of culture, language, gender, and age while mutually +understanding each other +6. An organization put into place that supports the Global Campus, and where staff with high levels of +knowledge and competency is placed throughout the University + +UTokyo offers a broad spectrum of unique experiences and opportunities, and invites students from all +over the world to participate while experiencing Japanese life and culture through the following programs +for example: +1. Undergraduate courses offered in English: Global Science Course (GSC), Programs in English at Komaba +(PEAK) +2. A variety of short-term programs are available over the summer holidays. +3. Student Exchange Programs with partner universities + +Academics Law, Medicine, Engineering, Letters, Science, Agriculture, Economics, Arts and Sciences, Education, +Phamaceutical Sciences, Mathematical Sciences, Frontier Sciences, Information Science and Technology, +Interdisciplinary Information Studies, Public Policy +Number of Students Undergraduate 14,040 / Graduate 14,292 / International Students 4,135 +Contact Details 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8654, Japan / intlst.adm@gs.mail.u-tokyo.ac.jp / https:// +www.u-tokyo.ac.jp/ja/index.html (English) https://www.u-tokyo.ac.jp/en/index.html (Chinese) https://www. +u-tokyo.ac.jp/zh/index.html (Korean) https://www.u-tokyo.ac.jp/ko/index.html +Student Exchange intex-ut.adm@gs.mail.u-tokyo.ac.jp 20 - to highlight the importance of reranking in reducing noise and improving -the quality of evidence, as unfiltered evidence may contain irrelevant or low- -quality content that hinders accurate verification. - -w/o Dynamic Demonstrations: The dynamic demonstrations are re- -placed with static ones, which remain fixed across different atomic facts. -This ablation is designed to highlight the benefits of dynamic demonstration -selection, which adapts to the specific context of each atomic fact, enhancing -the relevance and effectiveness of the reasoning process. - -w/o Evidence Retrieval: We disable evidence retrieval from the ex- -ternal corpus but still sample dynamic demonstrations to guide the verifica- -tion process. This ablation aims to isolate the contribution of external evi- -dence retrieval, demonstrating how the model performs when relying solely -on demonstrations and internal reasoning capabilities. - -w/o Demonstrations: This variant disables the similar instances used -for in-context demonstrations. This setting evaluates the role of demonstration- -based reasoning in providing contextual guidance and improving the model’s -ability to handle complex claims by leveraging analogous examples. - -w/o External Evidence Retrieval: We completely remove both evi- -dence retrieval and demonstration sampling, relying exclusively on the rea- -soner’s internal parametric knowledge (e.g., GPT-3.5) to verify each atomic -fact. This ablation evaluates the model’s ability to perform fact verifica- -tion without any external evidence or contextual guidance, highlighting the -limitations of relying solely on the LLMs’ inherent knowledge. - -Table 5: Performance of ablation study on HOVER. - -Models - -LA Macro-F1 - -Atomic Fact -Extraction - --w/o Atomic Fact Extraction --w/o Iterative Extraction --w/o Rationales - -External Evidence -Retrieval - --w/o Reranking --w/o Dynamic Demonstrations - -Atomic Fact -Verification - --w/o Evidence Retrieval --w/o Demonstrations --w/o External Evidence Retrieval - -77.04 -77.66 -78.74 - -77.97 -78.35 - -77.35 -78.20 -77.16 - -76.96 -77.81 -78.82 - -78.04 -78.51 - -77.29 -78.36 -77.09 - -AFEV - -78.87 - -78.76 + Tokyo Medical and Dental University + +Tokyo Medical and Dental University (TMDU) +is the only comprehensive medical university +in Japan that has both undergraduate and +graduate programs in medicine and dentist- +ry. The undergraduate programs are offered +through the Faculty of Medicine and Faculty +of Dentistry, and prepare students to be med- +ical and dental professionals. Our graduate +programs in Medical and Dental Sciences and +Health Care Sciences produce leaders who go +into research, education or a combination of +such work. Many professors teach and conduct +cutting edge research in their respective Faculty or at +one of our specialized research institutes, such as the +Medical Research Institute or the Institute of Biomate- +rials and Bioengineering. Today, 3,225 students pursue +medical, dental, and life science degrees at TMDU, with +nearly equal numbers of students in undergraduate and +graduate programs. Undergraduates begin their career +at the College of Liberal Arts and Sciences to ensure a +well-rounded education. After moving on to the Faculty +of Medicine or Faculty of Dentistry, students concentrate +on a curriculum that is increasingly integrated between +medical and dental sciences and utilizes an inter-profes- +sional educational approach, just as graduate students at +TMDU have traditionally enjoyed. + +TMDU currently has 367 international students, who are +predominantly enrolled in the graduate school. In fact, +despite its relatively small size, TMDU has the largest +number of international students in the medical and den- +tal fields of any national or public university in Japan. + +TMDU’s Vision is “Cultivating Professionals with Knowledge and Humanity, thereby Contributing to +People’s Well-being”. We aim to deploy these professionals domestically and globally, and become a +world-leading integrated medical university. + +Academics Medicine, Dentistry, Health Care Sciences, Oral Health Care Sciences +Number of Students Undergraduate 1,486 / Graduate 1,739 / International Students 367 +Contact Details 1-5-45, Yushima, Bunkyo-ku, Tokyo 113-8510, Japan / kouhou.adm@tmd.ac.jp / +81 3-3813-6111 +/ http://www.tmd.ac.jp/english/ +Student Exchange iss.adm@tmd.ac.jp / +81 3-3813-6111 +Research Exchange uraoffice.adm@tmd.ac.jp / +81 3-3813-6111 21 - As shown in Table 5, the removal or modification of each component leads -to a noticeable decline in overall model performance, underscoring their im- -portance within the overall framework. Specifically, the result of -w/o Atomic -Fact Extraction demonstrates that decomposing complex claims into atomic -facts significantly enhances model performance for fact verification. It not -only reduces the reasoning complexity by breaking down complex claims into -manageable units but also facilitates the retrieval of more comprehensive -and relevant external evidence. In addition, the proposed iterative extrac- -tion strategy further improves performance by mitigating error accumulation -throughout the decomposition process, and this incremental refinement is -critical for handling complex claims with multiple layers of reasoning. - -Retrieving evidence from the external knowledge corpus is essential for -mitigating hallucination issues inherent in LLMs, as it provides factual ground- -ing and reduces reliance on the model’s parametric knowledge alone. There- -fore, removing evidence retrieval in -w/o Evidence Retrieval results in a sig- -nificant decrease in verification performance. Additionally, the results of -w/o Demonstrations and -w/o Dynamic Demonstrations emphasize the im- -portance of context-aware demonstration selection. Dynamic demonstrations -are further adapt to the specific context of each atomic fact, ensuring that -the reasoning process is guided by the most relevant and informative exam- -ples. This adaptability enhances the model’s ability to handle diverse and -nuanced verification scenarios. - -5.5. Hyperparameter Sensitivity Analysis - -We conduct a hyperparameter sensitivity analysis on the HOVER dataset -to evaluate the impact of two critical parameters: k, the number of evidence -pieces selected after evidence reranking, and d, the number of dynamically -sampled similar instances used as reasoning demonstrations. - -As depicted in Figure 5(a), the results indicate that when k is too small, -the model suffers from insufficient evidence, leading to incomplete or inac- -curate verification results. Conversely, when k is too large, the model is -exposed to excessive noise, as irrelevant or low-quality evidence degrades -performance. These findings suggest that selecting 1–2 pieces of evidence -per atomic fact strikes the best balance, as decomposing complex claims into -atomic facts reduces the amount of information to verify, requiring fewer yet -more precise evidence pieces. - -Similarly, the number of demonstrations d plays a crucial role in guid- -ing the reasoning process. Using too few demonstrations (e.g., d = 1) often + Tokyo University of Foreign Studies + +Tokyo University of Foreign Studies ( TUFS ) is +the oldest instituition in Japan that is devoted +to international studies. It began as Bansho +Shirabesho ( Institution for Research of Foreign +Documents ), a government translation bureau +founded in 1857. It was then established as an +independent educational and research insti- +tution with the name Tokyo Gaikokugo Gakko +( Tokyo School of Foreign Languages) in 1899. +In 1999, the University celebrated the 100th +anniversary of its “independence”. + +The present campus is located in Fuchu City, + +a suburb in the western part of Tokyo. Students belong- +ing to the schools of Language and Culture Studies (1480 +students), International and Area Studies (1500 students), +and the Graduate School of Global Studies (148 students +enroll in the Master’s Course and 40 students in the Doc- +toral Course every year) study about languages and cul- +tures of the world. + +In addition to this, the campus also holds the Research +Institute for Languages and Cultures of Asia and Africa +and the Japanese Language Center for International Stu- +dents. + +TUFS was selected as the ‘Top Global University’ in +2014 by the Ministry of Education, Culture, Sports, Sci- +ence and Technology; MEXT ( Type B : Toward Globaliza- +tion). + +[Educational Objectives] + +TUFS offers comprehensive research and education in +the languages, cultures and societies of various areas in +the world including Japan. We aim to develop human re- +sources, who will be able to address global issues. By collaborating with people from various cultural +backgrounds, acquiring and utilizing a broad education and a variety of views with a flexible thought +process, rich sensibility, and advanced language ability, TUFS students will be armed with both knowl- +edge and adaptability to meet all modern global challenges. + +Academics Language and Culture Studies, International and Area Studies, Global Studies +Number of Students Undergraduate 3,865 / Graduate 555 / International Students 753 +Contact Details 3-11-1 Asahi-cho, Fuchu-shi, Tokyo 183-8534, Japan / soumu-koho@tufs.ac.jp / +81 42-330-5111 +/ http://www.tufs.ac.jp/english/ +Student Exchange ryugakushien@tufs.ac.jp / +81 42-330-5113 +Research Exchange kokusai-kyoten@tufs.ac.jp / +81 42-330-5594 22 - 0.78 - -0.76 - -0.74 - -0.72 - -0.70 - -HOVER(LA) - -1 - -2 - -3 -k - -4 - -5 - -0.78 - -0.76 - -0.74 - -0.72 - -0.70 - -1 - -2 - -HOVER(LA) - -4 - -5 - -3 -d - -(a) Accuracy vs. k - -(b) Accuracy vs. d - -Figure 5: Hyperparameter sensitivity analysis. - -results in insufficient contextual guidance, while too many (e.g., d > 2) intro- -duce redundancy and unnecessary complexity. Results in Figure 5(b) show -that 1-2 demonstrations per atomic fact are sufficient for reasoning perfor- -mance. This is attributed to the reduced reasoning complexity of atomic -facts, which requires less contextual support. - -5.6. Efficiency Study - -Table 6: Performance of efficiency study on HOVER. - -Models - -LA Macro-F1 Time/(h) - -One-hop -Extraction - --w/o Iterative Extraction --w/o Iterative Extraction and Reranking - -Iterative -Extraction - --w/o Reranking -AFEV - -77.04 -76.84 - -77.97 -78.87 - -76.96 -76.86 - -78.04 -78.76 - -0.86 -0.71 - -0.79 -0.94 - -To assess both the computational efficiency of our model, we evaluate -AFEV and its ablations on the HOVER test set. As summarized in Table 6, --w/o Iterative Extraction replaces the iterative decomposition strategy with a -one-shot approach. While this variant slightly improves runtime, it results in -a noticeable decline in verification accuracy. In contrast, -w/o Reranking re- -tains iterative decomposition but omits the evidence reranking stage. Despite -relying on less precise evidence retrieval, it outperforms -w/o Iterative Ex- -traction, highlighting the value of accurate atomic claim generation through -iterative decomposition. Besides, -w/o Reranking also achieves lower run- -time than -w/o Iterative Extraction, indicating that iterative decomposition + Tokyo Gakugei University + +History + +Tokyo Gakugei University was formally char- +tered as a university in 1949 when four teacher +training institutions in Tokyo from the pre-war +era were merged. + +In 1966 the Faculty of Liberal Arts was +changed to Faculty of Education, and a Gradu- +ate School of Education (master’s program) was +established to provide education and research +in a wide range of fields from humanities and +social sciences to natural science, physical edu- +cation, and the arts. + +In 2015, the Teacher Training Division was reorganized +as the School Education Division, and the Liberal Arts +Division was reorganized as the Educational Specialist +Division. + +Core Objectives + +Tokyo Gakugei University plays a leading role in the +transformation of society through education. Our mission +is to respond to the characteristics of contemporary soci- +ety, including the growth of a knowledge-based society, +increasing globalization, the diversification of education- +al challenges, and the exhaustion of local communities, +by cultivating human resources who contribute to educa- +tion for the children of the next generation to give them +four key strengths: the strength to work together to solve +challenges, the strength that comes from valuing diversi- +ty, the strength to reflect on and express themselves, and +the strength to create a new society. We have been a full +curriculum education university that has helped to train +valuable educators in a wide variety of education practic- + +es and subject curricula in primary and secondary education. + +Now, recognizing MEXT’s Third Medium-term Objective Period as a major turning point for Japa- +nese society and education, we aim to act as a nationwide center to lead Japanese education and to +be an international hub that will bring the results of Japanese education to a wider audience around +the world. + +Academics Education +Number of Students Undergraduate 4,615 / Graduate 927 / International Students 243 +Contact Details 4-1-1 Nukui-kita-machi, Koganei-shi, Tokyo 184-8501, Japan / ryuugaku@u-gakugei.ac.jp / +81 +42-329-7763 / http://www.u-gakugei.ac.jp/english/ +Student Exchange ryuugaku@u-gakugei.ac.jp / +81 42-329-7763 +Research Exchange kokusai@u-gakugei.ac.jp / +81 42-329-7849 23 - does not substantially compromise efficiency. This alleviates concerns that -its iterative nature would significantly hinder overall runtime performance. - -5.7. Case Study - -Figure 6: The reasoning process of AFEV for a specific case. - -In this section, we present a case study to systematically illustrate the -execution mechanism of the proposed AFEV model. As depicted in Figure 6, -the given claim contains multiple pieces of information, making it challeng- -ing to verify in a single step. Under our framework, AFEV first extracts the + Tokyo University of +Agriculture and Technology + +Tokyo University of Agriculture and Technolo- +gy ( TUAT ) was founded in 1874 when the Agri- +cultural Training Institute and Silkworm Diseas- +es Experiment Section were established under +the Industrial Encouragement Department of +Ministry of Home Affairs. The two institutions +respectively evolved into the Faculty of Agri- +culture and the Faculty of Engineering of TUAT +in 1949. Likewise, TUAT has a long history and +tradition of over 140 years. + +TUAT strives to tackle the challenges towards + +“the realization of sustainable development of society”, the needs which became prominent in the +20th century. Through education and research based on flexible ideas in agriculture, engineering and +the combined fields, TUAT promotes its philosophy to contribute to the progress of science and tech- +nology in harmony with the world peace, society, and nature, as well as to develop human resources +with capability to challenge problems and to creation of knowledge. + +[Strengths and plans] + +In the Third Medium-term Plan of TUAT, starting from the +2016 fiscal year, the President’s vision calls for achievement +of “globally-recognized research university”. We aim to +play a vital role as a university that can strengthen Japanese +presence in the world. By actively promoting following func- +tions, TUAT strives to become competitive with international +universities which maintain outstanding achievements, and +to promote education research and social implementation +with excellence. +1. Globally-competitive research ability. +2. Education and research with interactive capacity in the +international community. +3. Leadership role in the Japanese industries to strengthen +presence in the international community. +4. Training of advanced level of innovation leaders. + +Academics Agriculture, Engineering +Number of Students Undergraduate 3,867 / Graduate 2,124 / International Students 349 +Contact Details 3-8-1 Harumi-cho, Fuchu-shi, Tokyo 183-8538, Japan / kokusai@cc.tuat.ac.jp / +81 42-367-5913 / +https://www.tuat.ac.jp/en/ +Student Exchange kokusai@cc.tuat.ac.jp / +81 42-367-5913 +Research Exchange urac@ml.tuat.ac.jp 24 - initial atomic fact (Fact1) from the original claim and retrieves key external -evidence to support its verification, along with dynamic demonstrations to -guide the reasoning process. Subsequently, leveraging the verification result -and rationale of Fact1, the model extracts the second atomic fact (Fact2). -Unlike direct decomposition, this iterative strategy refines Fact2 by replacing -“that football club” with a more specific entity, “FC Barcelona.” This im- -proves clarity and facilitates more accurate evidence retrieval and reasoning. -Following a similar verification process as Fact1, the model easily obtains the -verification result and rationale for Fact2. - -Building on the enhanced context from Fact1 and Fact2, the extracted -Fact3 further incorporates additional information not explicitly present in -the original claim, such as “Lionel Messi,” which is derived from the ratio- -nale of Fact2. This demonstrates the high rationality and superiority of the -proposed iterative extraction and reasoning strategy. Finally, once all sub- -information in the original claim has been fully extracted and verified, the -extractor automatically terminates the decomposition process and synthe- -sizes the verification results of all atomic facts to produce the final claim -verification result. - -This case study highlights the effectiveness of our framework in handling -complex claims through iterative atomic fact extraction, evidence retrieval, -and dynamic demonstration-guided reasoning, ensuring both accurate and -interpretable verification results. - -6. Conclusion - -This paper presents AFEV, a novel framework for fact verification that -addresses the challenges of complex claims through dynamic atomic fact ex- -traction, refined evidence retrieval and adaptive atomic fact verification. By -decomposing intricate claims into simpler atomic facts and leveraging exter- -nal evidence alongside contextually relevant demonstrations, AFEV signifi- -cantly enhances the accuracy and interpretability of the verification process. -The integration of iterative extraction and adaptive reasoning mechanisms -ensures that each subcomponent of the claim is meticulously verified and con- -textualized, ultimately leading to a more reliable overall result. Extensive -experiments on five benchmark datasets demonstrate that AFEV achieves -state-of-the-art performance, underscoring its effectiveness in handling nu- -anced verification tasks. + Tokyo University of the Arts + +Tokyo University of the Arts ( Tokyo Geidai) +was formed by the merger of the Tokyo Fine +Arts School and the Tokyo Music School, both +founded in 1887 as pioneers of modern art and +music education in Japan. Today, the university +consists of the Faculties of Fine Arts and Music +and the Graduate School of Film and New Me- +dia, as well as the Graduate School of Global +Arts, thus offering a wide range of courses in +the field of the arts. + +To fully demonstrate its capability and attrac- +tion as a globally unparalleled comprehensive arts university and +rigorously promote the cultivation of outstanding artists who can +play an active role in the world, we strive to enhance, to a greater +extent, our education and research environments as well as sup- +port for our students under the “Tokyo Geidai NEXT 10 Vision.” + +Specifically, we will focus on the following: + +(1) Promote initiatives to attract prominent artists from overseas, +as well as diverse specialists and practitioners +(2) Expand opportunities to acquire international experience +through means such as sending students to overseas universities, +art festivals, and music festivals +(3) Enhance practical education and research programs in art or- +ganizations, businesses, etc. +(4) Enhance the university’s education and research environment +to ensure thorough, individualized instruction and small-group +learning + +(5) Expand specialist staff who support hands-on guidance and practical training in workshops, studios, etc. +(6) Provide world-class large workshop facilities, musical instruments, and filming and editing equipment + +As the only national arts university in Japan, Tokyo Geidai will continue efforts to nurture the next +generation of outstanding artists in order to promote the power of Japanese art and culture and to +contribute towards the global development of art through a close collaboration with relevant organi- +zations including the Japanese Ministry of Education, Culture, Sports, Science and Technology and the +Agency for Cultural Affairs. + +Academics Fine Arts, Music, Film and New Media, Global Arts +Number of Students Undergraduate 2,108 / Graduate 1,449 / International Students 329 +Contact Details 12-8 Ueno Park, Taito Ward, Tokyo 110-8714, Japan / intl-tua@ml.geidai.ac.jp / +81 50-5525- +2785 / http://global.geidai.ac.jp/e/ +Student Exchange intl-tua@ml.geidai.ac.jp / +81 50-5525-2785 +Research Exchange intl-tua@ml.geidai.ac.jp / +81 50-5525-2785 25 - 7. Acknowledgements - -This work was supported by the Natural Science Foundation of China - -[No. 62372057]. + Tokyo Institute of Technology + +Tokyo Tech is the top national university for +science and technology in Japan with a history +spanning almost 140 years. Of the approximately +10,000 students at the Ookayama, Suzukakedai, +and Tamachi Campuses, half are in their bachelor’s +degree program while the other half are in mas- +ter’s and doctoral degree programs. International +students number 1,700. There are 1,100 faculty +and 600 administrative and technical staff mem- +bers. + +Tokyo Tech has launched a new education sys- +tem in April 2016 and undergraduate and grad- +uate schools are now joined for the first time in +Japan. Through this new system, Tokyo Tech seeks +to develop talented people in the fields of science +and technology with the expertise and skills to +lead. + +While leading the world in several research fields +already, Tokyo Tech is carrying out plans to further +strengthen its research. + +Today the world faces many issues on a global +level and needs the outcomes of scientific and +technological research to solve them. + +The Institute will become a research and educa- +tional hub which will attract outstanding interna- +tional researchers and students to work on these +problems. + +Tokyo Tech is the best sci-tech university in Ja- +pan, and embraces the challenges of becoming +one of the world’s top 10 research universities with +spirit, enthusiasm and confidence. The Tokyo Tech +community continues to strive for the betterment +of society. + +Academics Science, Engineering, Materials and Chemical Technology, Computing, Life Science and Technology, +Environment and Society +Number of Students Undergraduate 4,828 / Graduate 5,384 / International Students 1,693 +Contact Details 2-12-1 Ookayama, Meguro-ku, Tokyo 152-8550, Japan / pr@jim.titech.ac.jp / +81 3-3726-1111 / +http://www.titech.ac.jp/english/index.html +Student Exchange ryu.kor3@jim.titech.ac.jp / +81 3-5734-3027 +Research Exchange kenkik.kik2@jim.titech.ac.jp / +81 3-5734-3803 -References - -[1] J. Thorne, A. Vlachos, C. Christodoulopoulos, A. Mittal, FEVER: a -large-scale dataset for fact extraction and VERification, in: Proceedings -of the 2018 Conference of the North American Chapter of the Associ- -ation for Computational Linguistics: Human Language Technologies, -Volume 1 (Long Papers), Association for Computational Linguistics, -New Orleans, Louisiana, 2018, pp. 809–819. - -[2] Z. Guo, M. Schlichtkrull, A. Vlachos, A survey on automated fact- -checking, Transactions of the Association for Computational Linguistics -10 (2022) 178–206. - -[3] W. Ostrowski, A. Arora, P. Atanasova, I. Augenstein, Multi-hop fact -checking of political claims, in: Proceedings of the Thirtieth Interna- -tional Joint Conference on Artificial Intelligence, IJCAI-21, Interna- -tional Joint Conferences on Artificial Intelligence Organization, 2021, -pp. 3892–3898, main Track. - -[4] L. Zheng, C. Li, X. Zhang, Y.-M. Shang, F. Huang, H. Jia, Evidence -retrieval is almost all you need for fact verification, in: Findings of the -Association for Computational Linguistics ACL 2024, 2024, pp. 9274– -9281. +26 -[5] O. Khattab, C. Potts, M. A. Zaharia, Baleen: Robust multi-hop rea- -soning at scale via condensed retrieval, in: Advances in Neural Informa- -tion Processing Systems 34: Annual Conference on Neural Information -Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, -2021, pp. 27670–27682. + Tokyo University of +Marine Science and Technology + +Tokyo University of Marine Science and Tech- +nology ( TUMSAT ) was founded in 2003 through +the merger of Tokyo University of Mercantile +Marine and Tokyo University of Fisheries. These +two former universities were founded in 1875 +and 1888, respectively, and TUMSAT prides +itself on history and tradition of 140 years. + +TUMSAT has inherited from its predecessors +its personality, characteristics and traditions, +and plays a key role as Japan’s only marine +university through its education and research +activities based on the motto “Voices from the Ocean,” supporting the development of Japan as a +maritime nation. + +In April 2017, we created the School of Marine Resources and Environment, to help address the +pressing issues that we face, nurture talent that can lead the way in the creation of new marine in- +dustry, make efficient use of marine resources, and remain internationally competitive. Based on an +understanding of marine science from the atmosphere to the seafloor, the school fosters professionals +who can work globally in the exploration of oceanic energy +and seafloor resources, and the preservation and restoration +of the marine environment. By developing such experts, we +hope to contribute to the creation of marine industries. + +To respond to the demands of globalization, we are devising +an English education program, with lectures in English and +abroad study programs, aiming to nurture marine specialists +who can be active worldwide. Moreover, we are implement- +ing the OQEANOUS program, selected as the Re-inventing +Japan Project in 2016, which is a student exchange program +including a double degree program by the credit transfer sys- +tem. Based on agreements with 101 universities and institutes +across 31 countries and regions, we are developing interna- +tional exchanges (as of March 2019). Having international +students accounting for almost 30 percent of the graduate +student body, we are one of the most internationally active +universities in Japan. + +Academics Marine Life Science, Marine Technology, Marine Resources and Environment +Number of Students Undergraduate 1,990 / Graduate 720 / International Students 255 +Contact Details 4-5-7 Konan, Minato-ku,Tokyo 108-8477, Japan / ks-koku@o.kaiyodai.ac.jp / +81 3-5463-0675 / +https://www.kaiyodai.ac.jp/english/ +Student Exchange iss@o.kaiyodai.ac.jp / +81 3-5463-4052 +Research Exchange ks-koku@o.kaiyodai.ac.jp / +81 3-5463-0675 -[6] L. Zheng, C. Li, L. Zhang, H. Jia, S. Wang, Z. Liu, X. Zhang, Mrr- -fv: Unlocking complex fact verification with multi-hop retrieval and -reasoning, Proceedings of the AAAI Conference on Artificial Intelligence -39 (2025) 26066–26074. doi:10.1609/aaai.v39i24.34802. +27 -26 + Ochanomizu University - [7] H. Li, X. Han, H. Wang, Y. Wang, M. Wang, R. Xing, Y. Geng, Z. Zhai, -P. Nakov, T. Baldwin, Loki: An open-source tool for fact verification, in: -Proceedings of the 31st International Conference on Computational Lin- -guistics: System Demonstrations, Association for Computational Lin- -guistics, Abu Dhabi, UAE, 2025, pp. 28–36. - -[8] A. Asai, Z. Wu, Y. Wang, A. Sil, H. Hajishirzi, Self-rag: Learn- -ing to retrieve, generate, and critique through self-reflection (2023). -arXiv:2310.11511. -URL https://arxiv.org/abs/2310.11511 - -[9] C. Li, J. Flanigan, RAC: efficient LLM factuality correction with re- -trieval augmentation, CoRR abs/2410.15667 (2024). arXiv:2410.15667. - -[10] B. He, N. Chen, X. He, L. Yan, Z. Wei, J. Luo, Z. Ling, Retrieving, -rethinking and revising: The chain-of-verification can improve retrieval -augmented generation, in: Findings of the Association for Computa- -tional Linguistics: EMNLP 2024, Miami, Florida, USA, November 12- -16, 2024, Association for Computational Linguistics, 2024, pp. 10371– -10393. - -[11] Y. Zhu, J. Si, Y. Zhao, H. Zhu, D. Zhou, Y. He, EXPLAIN, EDIT, -GENERATE: Rationale-sensitive counterfactual data augmentation for -multi-hop fact verification, in: Proceedings of the 2023 Conference on -Empirical Methods in Natural Language Processing, Association for -Computational Linguistics, Singapore, 2023, pp. 13377–13392. - -[12] S. Subramanian, K. Lee, Hierarchical Evidence Set Modeling for au- -tomated fact extraction and verification, in: Proceedings of the 2020 -Conference on Empirical Methods in Natural Language Processing -(EMNLP), Association for Computational Linguistics, Online, 2020, pp. -7798–7809. - -[13] W. Xiong, X. L. Li, S. - -Iyer, J. Du, P. Lewis, W. Y. Wang, -Y. Mehdad, W.-t. Yih, S. Riedel, D. Kiela, et al., Answering complex -open-domain questions with multi-hop dense retrieval, arXiv preprint -arXiv:2009.12756 (2020). - -[14] F. Xu, W. Shi, E. Choi, Recomp: Improving retrieval-augmented lms -with compression and selective augmentation (2023). arXiv:2310.04408. +Mission -27 +“Ochanomizu University will support all women, +regardless of age or nationality, in protecting their +individual dignity and rights, freely developing +their unique qualities and capabilities, and pursu- +ing personal learning so as to satisfy their intellec- +tual appetites.” - [15] C. Zhang, L. Zhang, D. Zhou, Causal walk: Debiasing multi-hop fact -verification with front-door adjustment, CoRR abs/2403.02698 (2024). -arXiv:2403.02698. +As a pioneer of women’s education in Japan, +Ochanomizu University offers programs that will +develop women capable of being opinion lead- +ers in politics, economics, academia, culture, and +other fields on the international stage. These pro- +grams – that of the “21st century Ochanomizu University model” – +focus on three areas: +1. Development of higher education for women into the future +2. Centralization of research and 21st century liberal arts education +3. Social contribution and international exchange -[16] J. Si, Y. Zhu, D. Zhou, Exploring faithful rationale for multi-hop fact -verification via salience-aware graph learning, CoRR abs/2212.01060 -(2022). arXiv:2212.01060. +Cutting-edge research and development based on new ways of -[17] H. Wang, K. Shu, Explainable claim verification via knowledge-grounded -reasoning with large language models, in: Findings of the Association -for Computational Linguistics: EMNLP 2023, Association for Compu- -tational Linguistics, Singapore, 2023, pp. 6288–6304. +thinking is a unique feature of research at Ochanomizu University. -[18] L. Pan, X. Lu, M.-Y. Kan, P. Nakov, QACheck: A demonstration sys- -tem for question-guided multi-hop fact-checking, in: Proceedings of the -2023 Conference on Empirical Methods in Natural Language Process- -ing: System Demonstrations, Association for Computational Linguis- -tics, Singapore, 2023, pp. 264–273. +Admission and Support Information for International Students +. Admission: There are special entrance examinations. Check the +website; http://www.ao.ocha.ac.jp/index.html +.Support Information: +1. Scholarship +2. Exemption and Extension for tuition fee +3. Tutor System +4. Peer Sport System, Mentor System -[19] X. Zhang, W. Gao, Reinforcement retrieval leveraging fine-grained feed- -back for fact checking news claims with black-box LLM, in: Proceedings -of the 2024 Joint International Conference on Computational Linguis- -tics, Language Resources and Evaluation, LREC/COLING 2024, 20-25 -May, 2024, Torino, Italy, ELRA and ICCL, 2024, pp. 13861–13873. +Message from University -[20] S. Yan, J. Gu, Y. Zhu, Z. Ling, Corrective retrieval augmented genera- +Approximately 240 internaitonal students from over 30 countries are currently enrolled in Ochan- +omizu University for study research in various fields. In order for international students to get accus- +tomed to the Japanese culture as well as to encourage their friendship with Japanese and other inter- +national students, we hold various events as follows: -tion, CoRR abs/2401.15884 (2024). arXiv:2401.15884. +International Festival / Japanese traditional flower arrangement class / Kimono class/Japanese tradi- -[21] Z. Yue, H. Zeng, L. Shang, Y. Liu, Y. Zhang, D. Wang, Retrieval aug- -mented fact verification by synthesizing contrastive arguments, in: Pro- -ceedings of the 62nd Annual Meeting of the Association for Computa- -tional Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thai- -land, August 11-16, 2024, Association for Computational Linguistics, -2024, pp. 10331–10343. +tional calligraphy class / Summer Program (taught in Japanese/English). +We are looking forward to welcoming you at Ochanomizu University. -[22] B. Wang, J. Ma, H. Lin, Z. Yang, R. Yang, Y. Tian, Y. Chang, Ex- -plainable fake news detection with large language model via defense -among competing wisdom, in: Proceedings of the ACM on Web Confer- -ence 2024, WWW 2024, Singapore, May 13-17, 2024, ACM, 2024, pp. -2452–2463. +Academics Letters, Education, Science, Human Life and Environmental Sciences, Humanities and Sciences +Number of Students Undergraduate 2,046 / Graduate 851 / International Students 240 +Contact Details 2-1-1 Otsuka, Bunkyo-ku, Tokyo 112-8610, Japan / ryu@cc.ocha.ac.jp / +81 3-5978-5722 / http:// +www.ocha.ac.jp/en/index.html +Student Exchange ryunai@cc.ocha.ac.jp / +81 3-5978-5143 +Research Exchange kenkyo-TL@cc.ocha.ac.jp / +81 3-5978-5502 / +81 3-5978-5163 28 - [23] Z. Shi, S. Zhang, W. Sun, S. Gao, P. Ren, Z. Chen, Z. Ren, Generate- -then-ground in retrieval-augmented generation for multi-hop question -answering, in: Proceedings of the 62nd Annual Meeting of the Associa- -tion for Computational Linguistics (Volume 1: Long Papers), ACL 2024, -Bangkok, Thailand, August 11-16, 2024, Association for Computational -Linguistics, 2024, pp. 7339–7353. - -[24] Z. Xie, R. Xing, Y. Wang, J. Geng, H. Iqbal, D. Sahnan, I. Gurevych, -P. Nakov, FIRE: fact-checking with iterative retrieval and verification, -CoRR abs/2411.00784 (2024). arXiv:2411.00784. - -[25] A. Gunjal, G. Durrett, Molecular facts: Desiderata for decontextualiza- -tion in LLM fact verification, in: Findings of the Association for Com- -putational Linguistics: EMNLP 2024, Miami, Florida, USA, November -12-16, 2024, Association for Computational Linguistics, 2024, pp. 3751– -3768. - -[26] Y. Jiang, S. Bordia, Z. Zhong, C. Dognin, M. Singh, M. Bansal, HoVer: -A dataset for many-hop fact extraction and claim verification, in: Find- -ings of the Association for Computational Linguistics: EMNLP 2020, -Association for Computational Linguistics, Online, 2020, pp. 3441–3460. - -[27] Z. Yang, J. Ma, H. Chen, H. Lin, Z. Luo, Y. Chang, A coarse-to-fine -cascaded evidence-distillation neural network for explainable fake news -detection, in: Proceedings of the 29th International Conference on Com- -putational Linguistics, International Committee on Computational Lin- -guistics, Gyeongju, Republic of Korea, 2022, pp. 2608–2621. - -[28] W. Y. Wang, “liar, liar pants on fire”: A new benchmark dataset for -fake news detection, in: Proceedings of the 55th Annual Meeting of the -Association for Computational Linguistics (Volume 2: Short Papers), -Association for Computational Linguistics, Vancouver, Canada, 2017, -pp. 422–426. - -[29] T. Alhindi, S. Petridis, S. Muresan, Where is your evidence: Improv- -ing fact-checking by justification modeling, in: Proceedings of the First -Workshop on Fact Extraction and VERification (FEVER), Association -for Computational Linguistics, Brussels, Belgium, 2018, pp. 85–90. + The University of Electro-Communications + +The mission of the University of Electro-Com- +munications (UEC ) is to cultivate academic ex- +cellences that bring about bright future of the +earth and humanity, as well as create leading +knowledge and skills. In order to achieve these +goals, UEC has established three fundamental +principles: “Pursuing education and research at +the cutting-edge of science and technology for +the benefit of the all humankind”; “Cultivating +talented researchers and engineers who will be +successful internationally to take the initiative in +various fields”; and “Creating engagement and +cooperation with society through practical applications +of the achievements of pioneering research activities in +science and technology”. + +As an action guideline for the realization of these prin- +ciples, UEC declared the new UEC Vision in April, 2019. +The essence of the Vision is “Providing an environment +where global and unique students and researchers can +gather to pursue their activities in a borderless manner”, +“Training human resources capable of leading the ad- +vancement of science and technology, with interdisci- +plinary and pluralistic thinking power, while keeping their +own expertise, within the dynamism of comprehensive +communication sciences”, and “Creation of fusion-bor- +der based academic disciplines that are not bound by +established concepts”. + +On December 8, 2018, UEC celebrated the centennial +anniversary of its foundation under the slogan “ ひ ら け +(hirake ), INNOVATION !”. The first word of the slogan is +a Japanese Hiragana, “ ひらけ ”, which has many meaning such as open, start, create, lead, develop, +expand, and so on. + +In the next 100 years, we will continue to create new value to contribute to the development of a +sustainable society, and cultivate innovation leaders with global perspectives under the slogan “ひらけ, +INNOVATION ! ”. + +Academics Informatics and Engineering +Number of Students Undergraduate 3,455 / Graduate 1,383 / International Students 312 +Contact Details 1-5-1 Chofugaoka, Chofu, Tokyo 182-8585, Japan / kouhou-k@office.uec.ac.jp / +81 42-443-5019 +/ http://www.uec.ac.jp/eng/ +Student Exchange kokusai-k@office.uec.ac.jp / +8142-443-5115 +Research Exchange kokusai-k@office.uec.ac.jp / +8142-443-5115 29 - [30] J. DeYoung, S. Jain, N. F. Rajani, E. Lehman, C. Xiong, R. Socher, -B. C. Wallace, ERASER: A benchmark to evaluate rationalized NLP -models, in: Proceedings of the 58th Annual Meeting of the Association -for Computational Linguistics, Association for Computational Linguis- -tics, Online, 2020, pp. 4443–4458. - -[31] B. Paranjape, M. Joshi, J. Thickstun, H. Hajishirzi, L. Zettlemoyer, -An information bottleneck approach for controlling conciseness in ratio- -nale extraction, in: Proceedings of the 2020 Conference on Empirical -Methods in Natural Language Processing (EMNLP), Association for -Computational Linguistics, Online, 2020, pp. 1938–1952. - -[32] M. Glockner, I. Habernal, I. Gurevych, Why do you think that? explor- -ing faithful sentence-level rationales without supervision, in: Findings -of the Association for Computational Linguistics: EMNLP 2020, Asso- -ciation for Computational Linguistics, Online, 2020, pp. 1080–1095. - -[33] S. Carton, S. Kanoria, C. Tan, What to learn, and how: To- -ward effective learning from rationales, CoRR abs/2112.00071 (2021). -arXiv:2112.00071. - -[34] J. Si, Y. Zhu, D. Zhou, Consistent multi-granular rationale extraction for -explainable multi-hop fact verification, CoRR abs/2305.09400 (2023). -arXiv:2305.09400. - -[35] K. Popat, S. Mukherjee, A. Yates, G. Weikum, DeClarE: Debunking -fake news and false claims using evidence-aware deep learning, in: Pro- -ceedings of the 2018 Conference on Empirical Methods in Natural Lan- -guage Processing, Association for Computational Linguistics, Brussels, -Belgium, 2018, pp. 22–32. - -[36] S. Jain, S. Wiegreffe, Y. Pinter, B. C. Wallace, Learning to faithfully -rationalize by construction, in: Proceedings of the 58th Annual Meet- -ing of the Association for Computational Linguistics, Association for -Computational Linguistics, Online, 2020, pp. 4459–4473. - -[37] S. M. Lundberg, S. Lee, A unified approach to interpreting model pre- -dictions, in: Advances in Neural Information Processing Systems 30: -Annual Conference on Neural Information Processing Systems 2017, De- -cember 4-9, 2017, Long Beach, CA, USA, 2017, pp. 4765–4774. + Hitotsubashi University + +Throughout the 140 years of its history, Hitot- +subashi University has steadily developed into +one of Japan’s leading research universities in the +field of social sciences. The university’s particular +strengths lie in academic research with practical +applications. Such studies include research direct- +ed toward providing solutions and redesigning +systems related to society, the economy and the +law in Japan and worldwide, research to help im- +prove business management, and much more. + + Hitotsubashi University offers a wide range of +educational opportunities. Students can specialize +in almost any field of the social sciences, including +management, economics, finance, law, sociology +and history. Through its unique zemi (seminar-style) +teaching system that focuses on small groups of +about 7 or 8 students, Hitotsubashi University pro- +vides an academic environment that values and +nurtures each and every student. In addition to +their respective specialized disciplines, students +are encouraged to take courses from other depart- +ments to broaden their knowledge. The university +also hosts international students from all over the +world. Over 850 students from 53 countries are +now enrolled at Hitotsubashi, creating a university +community with a distinctly cosmopolitan flavor. + + In an increasingly globalized world, Hitotsubashi +University vows to build upon its long-standing +tradition and commitment to progress, and to con- +tinue to fulfill its role at the center of cutting-edge +research and education. + +Academics Management, Business, MBA, Economics, Law, Politics, Sociology, Linguistics +Number of Students Undergraduate 4,402 / Graduate 2,002 / International Students 904 +Contact Details 2-1 Naka, Kunitachi, Tokyo 186-8601, Japan / wwwadm@dm.hit-u.ac.jp / +81 42-580-8000 / +http://www.hit-u.ac.jp/eng/ +Student Exchange (Inbound) edu-gs.g2@dm.hit-u.ac.jp (Outbound) edu-gs.g@dm.hit-u.ac.jp / (Inbound) +81 42- +580-8162 (Outbound) +81 42-580-8764 +Research Exchange res-kk.g@dm.hit-u.ac.jp / +81 42-580-8055 30 - [38] P. K. Mudrakarta, A. Taly, M. Sundararajan, K. Dhamdhere, Did the -model understand the question?, in: Proceedings of the 56th Annual -Meeting of the Association for Computational Linguistics (Volume 1: -Long Papers), Association for Computational Linguistics, Melbourne, -Australia, 2018, pp. 1896–1906. - -[39] N. De Cao, M. S. Schlichtkrull, W. Aziz, I. Titov, How do decisions -emerge across layers in neural models? interpretation with differentiable -masking, in: Proceedings of the 2020 Conference on Empirical Methods -in Natural Language Processing (EMNLP), Association for Computa- -tional Linguistics, Online, 2020, pp. 3243–3255. - -[40] H. Chen, Y. Ji, Learning variational word masks to improve the inter- -pretability of neural text classifiers, in: Proceedings of the 2020 Confer- -ence on Empirical Methods in Natural Language Processing (EMNLP), -Association for Computational Linguistics, Online, 2020, pp. 4236–4251. - -[41] T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhari- -wal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, -A. Herbert-Voss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. M. -Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, -S. Gray, B. Chess, J. Clark, C. Berner, S. McCandlish, A. Radford, -I. Sutskever, D. Amodei, Language models are few-shot learners, CoRR -abs/2005.14165 (2020). arXiv:2005.14165. - -[42] J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. Chi, -Q. Le, D. Zhou, Chain-of-thought prompting elicits reasoning in large -language models (2023). arXiv:2201.11903. - -[43] X. Zhang, W. Gao, Towards llm-based fact verification on news claims -with a hierarchical step-by-step prompting method, in: Proceedings of -the 13th International Joint Conference on Natural Language Processing -and the 3rd Conference of the Asia-Pacific Chapter of the Association -for Computational Linguistics, IJCNLP 2023 -Volume 1: Long Papers, -Nusa Dua, Bali, November 1 - 4, 2023, Association for Computational -Linguistics, 2023, pp. 996–1011. - -[44] S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, Y. Cao, -React: Synergizing reasoning and acting in language models (2023). -arXiv:2210.03629. + National Graduate Institute for +Policy Studies + +Located in the heart of the fascinating city of +Tokyo, The National Graduate Institute for Policy +Studies (GRIPS) is an international premier policy +school with the aim of contributing to the better- +ment of democratic governance around the world. +We excel at providing interdisciplinary education +for future leaders in the public sector and conduct +research on contemporary policy issues to gener- +ate innovative solutions. + +Founded in 1997 as a stand-alone graduate insti- +tute, GRIPS is comprised of world-class academics +and distinguished practitioners with expertise in +public sector policy formulation and management. +Around 20% of the faculty and over 60% of stu- +dents are recruited from outside Japan. Our vi- +brant, diverse student body consists of almost 400 +members hailing from 65 countries and regions – +all with the ambition to advance good governance +across the globe or contribute to policy related +research. + +We offer a diverse array of Master’s and Doctoral +Programs, from which students cultivate the ability +to analyze issues and suggest solutions, develop +interdisciplinary knowledge and skills that span +related fields, and gain practical expertise. In ad- +dition to our degree programs, we also offer exec- +utive level short term training programs across a +wide range of themes. + +Since our inception, our achievements in pro- +moting good governance are considerable and +far-reaching. Today, our impressive Alumni network +of over 4,200 strong are actively shaping policy in +more than 110 countries around the world. + +Academics Policy Studies +Number of Students Undergraduate 0 / Graduate 380 / International Students 264 +Contact Details 7-22-1 Roppongi, Minato-ku, Tokyo 106-8677, Japan / kouhou@grips.ac.jp / +81 3-6439-6100 / +http://www.grips.ac.jp/en/ 31 - [45] W. Y. Wang, ”liar, liar pants on fire”: A new benchmark dataset for -fake news detection, in: Proceedings of the 55th Annual Meeting of -the Association for Computational Linguistics, ACL 2017, Vancouver, -Canada, July 30 - August 4, Volume 2: Short Papers, Association for -Computational Linguistics, 2017, pp. 422–426. - -[46] H. Rashkin, E. Choi, J. Y. Jang, S. Volkova, Y. Choi, Truth of varying -shades: Analyzing language in fake news and political fact-checking, in: -Proceedings of the 2017 Conference on Empirical Methods in Natural -Language Processing, EMNLP 2017, Copenhagen, Denmark, September -9-11, 2017, Association for Computational Linguistics, 2017, pp. 2931– -2937. - -[47] J. Ma, W. Gao, S. R. Joty, K. Wong, Sentence-level evidence embed- -ding for claim verification with hierarchical attention networks, in: Pro- -ceedings of the 57th Conference of the Association for Computational -Linguistics, ACL 2019, Florence, Italy, July 28- August 2, 2019, Volume -1: Long Papers, Association for Computational Linguistics, 2019, pp. -2561–2571. - -[48] N. Kotonya, F. Toni, Explainable automated fact-checking for pub- -lic health claims, in: Proceedings of the 2020 Conference on Empir- -ical Methods in Natural Language Processing, EMNLP 2020, Online, -November 16-20, 2020, Association for Computational Linguistics, 2020, -pp. 7740–7754. - -[49] P. Atanasova, J. G. Simonsen, C. Lioma, I. Augenstein, Generating fact -checking explanations, in: Proceedings of the 58th Annual Meeting of -the Association for Computational Linguistics, ACL 2020, Online, July -5-10, 2020, Association for Computational Linguistics, 2020, pp. 7352– -7364. - -[50] Z. Yang, J. Ma, H. Chen, H. Lin, Z. Luo, Y. Chang, A coarse-to-fine -cascaded evidence-distillation neural network for explainable fake news -detection, in: Proceedings of the 29th International Conference on Com- -putational Linguistics, COLING 2022, Gyeongju, Republic of Korea, -October 12-17, 2022, International Committee on Computational Lin- -guistics, 2022, pp. 2608–2621. - -32 - diff --git a/q175/random_k2/question.json b/q175/random_k2/question.json index d389e0592fbfcf7f70177c53b78e3f534714bcfa..c967e9d886463a985413eacad6619e96d6c19bfa 100644 --- a/q175/random_k2/question.json +++ b/q175/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "f86d073b0d735ac873a65d906ba82758.pdf", - "web_39b6cadc4e9ca03e.pdf", - "58a44062a5b5795816b8b85c9c8c3a95c53d18a7.pdf" + "resize_r2_0.pdf", + "web_660e5c500c425b81.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q175/random_k2/random_1.md b/q175/random_k2/random_1.md index 2570a92a07cb79b4daa90af30bf44b6e8b43f56a..052cee2fff6d2fb0526b1d43400b3c01f2587dfb 100644 Binary files a/q175/random_k2/random_1.md and b/q175/random_k2/random_1.md differ diff --git a/q175/random_k2/random_2.md b/q175/random_k2/random_2.md index bd35c3422555cd95beb556bbd614d1deaa052c77..e3ed939376d2f6fad348f021570f136f5c225f39 100644 --- a/q175/random_k2/random_2.md +++ b/q175/random_k2/random_2.md @@ -1,961 +1,1527 @@ -Logical Reasoning in Large Language Models: A Survey -HanmengLiu1†, ZhizhangFu1†, MengruDing1, RuoxiNing1 -ChaoliZhang2, XiaozhangLiu3 and YueZhang1∗ -1 WestlakeUniversity -2 ZhejiangNormalUniversity -3 HainanUniversity -{liuhanmeng,zhangyue}@westlake.edu.cn,{fuzhizhang.fzz,dingmengru2021}@gmail.com, -ruoxining@outlook.com,chaolizcl@zjnu.edu.cn,lxzh@hainanu.edu.cn -Abstract theirreasoningisincreasinglyvital. Asaresult,post-training -LLM for reasoning has garnered a surge of interest in both -Withtheemergenceofadvancedreasoningmodels industry and research[OpenAI, 2024; DeepSeek-AI, 2025; -like OpenAI o3 and DeepSeek-R1, large language Muennighoffetal.,2025]. -models(LLMs)havedemonstratedremarkablerea- Despite growing research, existing surveys [Plaat et al., -soning capabilities. However, their ability to per- 2024; Sun et al., 2023; Yu et al., 2024] often conflate log- -form rigorous logical reasoning remains an open -ical reasoning with general-purpose heuristic strategies like -question. This survey synthesizes recent advance- Chain-of-Thought (CoT) [Xia et al., 2024]. There has been -mentsinlogicalreasoningwithinLLMs, acritical -a lack of a literature review dedicated to LLM and formal -area of AI research. It outlines the scope of logi- -symbolic logic. This survey provides a comprehensive re- -calreasoninginLLMs,itstheoreticalfoundations, viewoflogicalreasoninginlargelanguagemodels(LLMs), -andthebenchmarksusedtoevaluatereasoningpro- -with a focus on formal and symbolic logic-based reasoning -ficiency. We analyze existing capabilities across -ratherthangeneralheuristicapproaches. Webeginbydefin- -differentreasoningparadigms—deductive,induc- -ing logical reasoning in AI, distinguishing it from general- -tive,abductive,andanalogical—andassessstrate- -purpose reasoning, and categorizing key paradigms, includ- -gies to enhance reasoning performance, including -ing deductive, inductive, abductive, and analogical reason- -data-centrictuning,reinforcementlearning,decod- ing. Additionally,weanalyzeexistingbenchmarksandeval- -ingstrategies,andneuro-symbolicapproaches.The -uationmethodologies,identifyinggapsinassessingsymbolic -reviewconcludeswithfuturedirections,emphasiz- -inference, consistency, and robustness. We further explore -ing the need for further exploration to strengthen -state-of-the-art techniques for enhancing logical reasoning, -logicalreasoninginAIsystems. -such as instruction fine-tuning, logic-informed pre-training, -reinforcement learning, inference-time decoding strategies, -1 Introduction andhybridneuro-symbolicmethods. Weexaminerecentad- -vancesinneuro-symbolicintegration,alongwithapplications -Logical reasoning is a fundamental challenge to arti- -of theorem provers, logic solvers, and formal verification -ficial intelligence (AI) and natural language processing -frameworksinLLMs. Finally,wehighlightopenchallenges -(NLP) [Newell and Simon, 1956; McCarthy and Hayes, -inscalability, reasoningconsistency, explainability, andeffi- -1981;McCarthy,1959]. Whileearlyformallogic-basedrea- -ciency, proposing future directions for multi-modal reason- -soningapproachesfacedlimitationsinscalabilityandadapt- -ing, hybrid architectures, and improved evaluation frame- -ability [Pereira, 1982; Cann, 1993], data-driven models -works. Thestructureofthesubsequentchaptersisillustrated -became the dominant method since the 1980s [McCarthy, -inFigure1. -1989].Recently,pre-trainedLargeLanguageModels(LLMs) -and their emergent logical reasoning abilities have attracted -2 LogicinArtificialIntelligence -increasing attention [Liu et al., 2023b; Xu et al., 2023]. -Logical reasoning is a cornerstone of artificial intelligence -Logical reasoning integrates LLMs with inference struc- -(AI),enablingmachinestosimulatehumanthoughtprocesses -turing, enabling multistep deduction and abstraction, and -improving interpretability and reliability [Shi et al., 2021; and solve complex problems. At its core, logical reasoning -Staceyetal.,2022;Rajaramanetal.,2023]. Italsostrength- appliesstructuredrulestoderiveconclusionsfrompremises, -providingarigorousframeworkfordecision-makingandin- -ens generalization, helping models handle novel scenarios -beyond their training data [Haruta et al., 2020]. As LLMs -ference[Sunetal.,2023]. -become integral to domains like legal analysis and scien- -2.1 HistoryofLogicReasoningResearch -tific discovery, ensuring the correctness and verifiability of -Logical reasoning can be traced back to ancient Greece, -* Correspondingauthor. where Aristotle’s syllogisms laid the foundation for classi- -† Equalcontribution. cal logic. During the Middle Ages, scholars refined these -5202 -beF -31 -]IA.sc[ -1v00190.2052:viXra - -| | | | | | C o n T R o L | [L iu e ta l ., 2 | 0 2 1 a ] , F O L IO | [ H a n e t a | l ., 2 0 2 4 a ] , L | o g i cNLI[Tianetal.,2021], | | | | | -| --- | --- | --- | ------------------------------ | --- | ---------------- | ----------------- | -------------------- | ----------------- | -------------------- | --------------------------- | --- | --- | --- | --- | -| | | | NaturalLanguageInference(§3.1) | | R u lt eT a k er | [C la rk e t a l. | , 2 0 2 1 ] , L og | iB e nc h [ P a r | m a r e t a l ., 2 0 | 2 3 ] | | | | | -Types&history(§2) -Task&Benchmarks(§3) ReadingComprehension(§3.2) L o g iQ A [L iu e t a l . , 2 0 2 3 a ] , R eC l o r [ Y u e t al . , 20 2 0 ], A R - LS A T [ Y u e t al.,2020],CLUTRR[Sinhaetal.,2019], -| | | | | | G S M [ Co | bb e e t a l . , 2 0 2 | 1 ; L i et al . , | 2 0 2 4a ] L I N G | O L Y [ B ea n | e t al. , 2 02 4 ] | | | | | -| --- | ----------------------- | --- | ----------------------------- | --- | ------------------------------------------------------------------------ | ---------------------- | ----------------- | ------------------ | -------------- | ------------------ | --- | --- | --- | --- | -| | | | Benchmarksandtestsuites(§3.3) | | GLoRE[liuetal.,2023d],LogiGLUE[Luoetal.,2024],LogiTorch[Helweetal.,2022] | | | | | | | | | | -| | | | DeductiveReasoning(§4.1) | | [Saparovetal.,2023],[Yuanetal.,2023],[Rybetal.,2022] | | | | | | | | | | -| | | | InductiveReasoning(§4.2) | | [Yangetal.,2024b],[Bowenetal.,2024],[Sullivan,2024] | | | | | | | | | | -| | Evaluation&Analysis(§4) | | AbductiveReasoning(§4.3) | | TrueDetective[DelandFishel,2023],[Nguyenetal.,2023] | | | | | | | | | | -gninosaeRlacigoLMLL -| | | | AnalogicalReasoning(§4.4) | | ANALOGICAL[Wijesiriwardeneetal.,2023],[PetersenandvanderPlas,2023],[Qinetal.,2024] | | | | | | | | | | -| --- | --- | --- | ----------------------------- | --- | --------------------------------------------------------------------------------------------- | --- | ------------------------------------------------------ | ------------------- | ------------------------------------------------ | ------------------ | ------------------------ | ---------------------------- | --- | --- | -| | | | OverallAnalysis&Metrics(§4.5) | | [Liuetal.,2023b],[Xuetal.,2023],[Liuetal.,2024c],[Gandarelaetal.,2024],[Thatikondaetal.,2025] | | | | | | | | | | -| | | | | | | | F O | L I O [ H a n e t | a l. , 2 0 2 4 a] , | P -F O LI O [ H | an e t a l., 2 0 2 4 | b ] , | | | -| | | | | | Expert-CuratedDatasets | | L ea | n D oj o [ Y a n g | e t a l ., 2 0 23 | ] , S ym b ol- L | L M [ X u e t a l. , | 2 0 24a] | | | -| | | | Data-CentricApproaches(§5.1) | | SyntheticDatasets | | RulteTaker[Clarketal.,2021],FLD×2[Morishitaetal.,2024] | | | | | | | | -| | | | | | | | L og | iC o T [ L i u e | t a l . , 2 0 2 3 c ],LogicPro[Jiangetal.,2024], | | | | | | -| | | | | | LLM-distilledDatasets | | PO | D A [ W a n g e | t a l . , 2 0 2 4 b ] | | | | | | -| | | | | | InstructionFine-Tuning | | L o gi | C o T [ L iu e | t al ., 2 0 2 3 c ] | , L og i P T [ Fe | n g e t a l. , 2 0 2 4 | ] , P G L [Wangetal.,2024a], | | | -| | | | | | | | Sy m | b o l -L L M [ X | u e t a l. , 2 0 | 2 4 a] , T P C L | [ W a n g e t a l ., 2 0 | 2 4 b ] , | | | -| | | | | | | | [Ji a | o e t a l . , 2 0 2 | 4 ] , [ X i e t a | l. , 2 0 2 4 ] , M | a rc o - o 1 [Z h a o et | a l. , 2 0 2 4 ] , | | | -| | | | | | ReinforcementLearning | | | | [ | 5] | | [D I,2025] | | | -| | | | Model-CentricApproaches(§5.2) | | | | D e e | ps e e k - R 1 - Z | e r o D e e p S ee | k - A I , 2 0 2 | , D e e p se ek - R 1 | e e p S e e k -A | | | -EnhancementMethods(§5) -GoT[Leietal.,2023],ChainofLogic[Servantezetal.,2024], -| | | | | | | | S e le | c ti o n- In fe re | n c e [ C r e s w e | ll e t a l ., 2 0 2 | 3 ] , [ M a l o n et a | l. , 2 0 2 4 ] , | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ------------------ | ------------------- | ------------------- | ---------------------- | ---------------- | --- | --- | -Discussion(§6) Inference-TimeDecoding M a ie u t ic P ro m p ti n g [ J u n g e t a l ., 2 0 2 2 ] , L o g i c- o f - tho ug h t [ L i u e t al.,2024a], -DetermLR[Sunetal.,2024],Neurologic[Luetal.,2021], -Formal-LLM[Lietal.,2024b] -| | | | | | [ Z a y ya d an | d A d i , 2 0 24 ] | , L eanDojo[Yangetal.,2023],LQOT[Liuetal.,2024b],[Ouyangetal.,2023], | | | | | | | | -| --- | --- | --- | ---------------------------------- | --- | --------------- | ---------------------- | -------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -| | | | ExternalKnowledgeUtilization(§5.3) | | K n o w R A | [M a i e t a l ., 2 02 | 5 ] | | | | | | | | -LINC[Olaussonetal.,2023],LOGICLLAMA[Yangetal.,2024a], -| | | | Neuro-SymbolicApproaches(§5.4) | | CLOVER[Ryuetal.,2024],LOGIC-LM[Panetal.,2023],LogicAgent[Liuetal.,2024a], | | | | | | | | | | -| --- | --- | --- | ------------------------------ | --- | ------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLM-TRes[Toroghietal.,2024],SymbCoT[Xuetal.,2024c],Aristotle[Xuetal.,2024b] -Figure1:Thestructureofthissurvey -theories, and in the 17th century, Leibniz’s universal lan- soningisvalid, theconclusionmustalsobetrue. Forexam- -guageandcalculusratiocinatorbridgedlogicwithmathemat- ple,giventhepremises“Allapplesarered”and“Thisfruitis -ics, foreshadowing modern computational logic. The 19th an apple” one can deduce that “This fruit is red” Deductive -century saw George Boole’s Boolean algebra, which trans- reasoning is fundamental in fields such as mathematics and -formedlogicintoamathematicalframework,layingthefoun- formallogic,wherecertaintyandrigorareparamount. -dationfordigitalcomputing. -| | | | | | | Inductive | | Reasoning. | | Unlike | deductive | | reasoning, | induc- | -| -------- | --------------- | --- | ------ | ------ | ------------ | --------- | --------- | ---------- | ----- | ------- | ----------- | --- | ---------- | -------- | -| The 20th | century ushered | in | modern | logic, | with Russell | | | | | | | | | | -| | | | | | | tive | reasoning | | draws | general | conclusions | | based on | specific | -andWhitehead’sPrincipiaMathematicaformalizingcomplex -| | | | | | | observations | | or | evidence. | | While the | conclusions | are | often | -| ------------------ | ------ | ------------- | --- | -------------- | --------- | ------------------- | --- | --- | --------- | ----------------------------- | --------- | ----------- | --- | ----- | -| logical systems. | By the | mid-century, | AI | pioneers | like John | | | | | | | | | | -| | | | | | | consideredprobable, | | | | theyarenotguaranteedtobetrue. | | | | For | -| McCarthy leveraged | logic | for knowledge | | representation | and | | | | | | | | | | -instance,observingthatallswansseensofararewhitemight -| automated | theorem proving, | leading | to | logic | programming | | | | | | | | | | -| ------------------ | ------------------------------- | ------- | --- | ----- | ----------- | ---- | ------ | --------- | --- | ---------- | ---- | ---- | --------- | ------ | -| | | | | | | lead | to the | inductive | | conclusion | that | “All | swans are | white” | -| andknowledgebases. | The1970sintroducednon-monotonic | | | | | | | | | | | | | | -Inductivereasoningiswidelyusedinscientificdiscoveryand -| logic, enabling | AI to handle | commonsense | | reasoning. | The | | | | | | | | | | -| --------------- | ------------ | ----------- | --- | ---------- | --- | ----------- | --- | ---------------- | --- | --- | ----- | -------- | ---------- | --- | -| | | | | | | data-driven | | decision-making, | | | where | patterns | and trends | are | -1980ssawlogicalreasoningintegratewithknowledgerepre- -inferredfromempiricaldata. -| sentation, advancing | expert | systems | for | real-world | applica- | | | | | | | | | | -| -------------------- | ------ | ------- | --- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tions. The1990ssawtheriseofknowledgegraphs,structur- -| | | | | | | Abductive | | Reasoning. | | This | form | of reasoning | seeks | the | -| --- | --- | --- | --- | --- | --- | --------- | --- | ---------- | --- | ---- | ---- | ------------ | ----- | --- | -ingvastknowledgeforcomplexreasoningtasks. -mostplausibleexplanationorcauseforasetofobservations, -In the 21st century, neuro-symbolic approaches have often in the presence of incomplete information. Abductive -mergeddeeplearningwithlogicalinference,resultingintools reasoning is particularly useful in diagnostic tasks and real- -like DeepLogic [Cingillioglu and Russo, 2019] and SAT- world problem-solving. For example, seeing wet spots on -Net [Wang et al., 2019]. Logical reasoning remains a cor- thestreetmightleadonetoinferthat“Ithasrecentlyrained” -nerstoneofAIresearch,evolvingfromphilosophytomodern While abductive conclusions are not certain, they provide a -computing. As AI advances, logical reasoning continues to practicalbasisforhypothesisgenerationanddecision-making -| shape intelligent | systems, | ensuring | structured, | | interpretable, | underuncertainty. | | | | | | | | | -| ----------------- | -------- | -------- | ----------- | --- | -------------- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | -androbustdecision-making. -| | | | | | | Analogical | | Reasoning. | | | Analogical | reasoning | | involves | -| --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | --- | --- | ---------- | --------- | --- | -------- | -2.2 TypesofLogicalReasoning drawing comparisons between similar situations or domains -| | | | | | | to | make | inferences | or | solve | problems. | By | identifying | par- | -| --- | --- | --- | --- | --- | --- | --- | ---- | ---------- | --- | ----- | --------- | --- | ----------- | ---- | -Logicalreasoningcanbebroadlycategorizedintofourmain -| | | | | | | allelsbetweendifferentscenarios, | | | | | thistypeofreasoningen- | | | | -| --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | -types,eachservingdistinctpurposesandapplications: -| | | | | | | ables | creative | problem-solving | | | and | knowledge | transfer. | For | -| --- | --- | --- | --- | --- | --- | ----- | -------- | --------------- | --- | --- | --- | --------- | --------- | --- | -DeductiveReasoning. Thistypeofreasoningderivesspe- example,understandingthatplanetsorbitthesuninelliptical -cificconclusionsfromgeneralprinciplesorpremises. Itop- pathsmightleadonetoanalogicallyreasonthatotherceles- -eratesundertherulethatifallpremisesaretrueandtherea- tialbodies, suchascomets, exhibitsimilarorbitalcharacter- - -Dataset Language QuestionType Size Source Neutral labels. It isolates FOL-based inference from com- -LogiQA Zh/En Multichoice 15,937 Exam-based monsensereasoning,enablingpreciseevaluationofreasoning -| ReClor | En | Multichoice | | | 6,138 | Exam-based | | | | | | | | -| ------ | --- | ----------- | --- | --- | ----- | ---------- | --- | --- | --- | --- | --- | --- | --- | -AR-LSAT En Multichoice 2,064 Exam-based accuracyandgeneralization. -| CLUTRR | En | Question-answering | | | 6,016 | Rule-based | | | | | | | | -| ------- | --- | ------------------ | --- | --- | ----- | --------------- | ------------ | --- | ------------- | --- | ----------- | ------- | -------- | -| | | | | | | | ProofWriter | | [Tafjord | et | al., 2021] | extends | Rule- | -| GSM | En | Mathwordproblems | | | 19K | Exam-based | | | | | | | | -| | | | | | | | Taker [Clark | | et al., 2021] | by | introducing | CWA | (closed- | -| LINGOLY | En | Question-answering | | | 1,133 | Expert-designed | | | | | | | | -ConTRoL En ternaryclassification 8,325 Exam-based world assumption) and OWA (open-world assumption) to -FOLIO En binaryclassification 1,351 Expert-designed handlenegationandopen-worldreasoning. ItincludesBirds- -| LogicNLI | En | ternaryclassification | | | 30K | Exam-based | | | | | | | | -| -------- | --- | --------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -ProofWriter En binaryclassification - Exam-based Electricity (handcrafted domain theories) and ParaRules -LogicBench En binaryclassification 1,270 Rule-based (crowdsourced paraphrased rules) for systematic evaluation -| GLoRE | Zh/En | Miscellaneous | | 17tasks | | Miscellaneous | | | | | | | | -| -------- | ----- | ------------- | --- | ------- | --- | ------------- | ----------------- | --- | ------ | ---------- | ---------- | --- | ---------- | -| | | | | | | | of generalization | | across | linguistic | variations | and | real-world | -| LogiGLUE | En | Miscellaneous | | 24tasks | | Miscellaneous | | | | | | | | -LogiTorch En Miscellaneous 16tasks Miscellaneous knowledgedomains. -BIG-Bench En Miscellaneous 7tasks Miscellaneous [Parmar 2023] -| | | | | | | | LogicBench | | | et al., | is | a GPT-3-generated | | -| --- | --- | --- | --- | --- | --- | --- | ---------- | -------- | -------- | ------- | ---------- | ----------------- | -------- | -| | | | | | | | dataset | covering | 25 types | of | reasoning, | including | proposi- | -Table1:MainDatasetsandBenchmarksofLogicalReasoningTask. -| | | | | | | | tional logic, | FOL, | and | non-monotonic | logic. | It | consists of | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ---- | --- | ------------- | ------ | --- | ----------- | -istics. Analogicalreasoningisparticularlyvaluableinfields -1,270testentrieslabeledasYesorNo. -likeeducation,design,andinnovation. -3.2 MachineReadingComprehension(MRC) -3 TasksandBenchmarks -MRCevaluateslogicalreasoningbyrequiringmodelstoan- -| Logical reasoning | | datasets | and | benchmarks | are | essential for | | | | | | | | -| ----------------- | --- | -------- | --- | ---------- | --- | ------------- | -------------- | --- | -------- | ------- | -------- | ----- | -------- | -| | | | | | | | swer questions | | based on | a given | passage. | Tasks | are com- | -evaluatingthereasoningcapabilitiesoflargelanguagemod- -| | | | | | | | monlyformattedasmultiple-choice, | | | | spanextraction, | | orfree | -| ----------------------------- | ----- | -------- | --- | -------------- | --- | ---------- | -------------------------------- | ---- | --------------- | --- | --------------- | ------------ | ------ | -| els (LLMs). | These | datasets | can | be categorized | | into three | | | | | | | | -| | | | | | | | response, | with | multiple-choice | | QA being | particularly | effec- | -| typesbasedontheirdatasources: | | | | | | | tiveduetoitsstandardization. | | | | | | | -Rule-based Datasets [Tafjord et al., 2021; Sinha et al., [Liu 2023a] -| | | | | | | | LogiQA | | et al., | | is sourced | from | the Chinese | -| --------- | ------------- | --- | --------- | ----- | ------- | ---------- | ------ | --- | ------- | --- | ---------- | ---- | ----------- | -| 2019] are | automatically | | generated | using | logical | rules, en- | | | | | | | | -CivilServiceExam,containing15,937entriesinChineseand -| abling large-scale | | data collection. | | However, | ensuring | diver- | | | | | | | | -| ------------------ | --- | ------------------- | --- | -------- | -------- | ---------- | -------- | ---------- | ------- | ------- | --------- | --- | --------- | -| | | | | | | | English. | It targets | complex | logical | reasoning | and | is widely | -| sity is crucial | | to avoid repetitive | | patterns | and | comprehen- | | | | | | | | -usedforevaluatingLLMs. -sivelyevaluatereasoningcapabilities. ReClor[Yuetal.,2020],derivedfromtheGMAT,features -| Expert-Designed | | Datasets | [Han | et | al., 2024a] | are con- | | | | | | | | -| --------------- | --- | -------- | ---- | --- | ----------- | -------- | --- | --- | --- | --- | --- | --- | --- | -6,138Englishentrieswithfour-optionmultiple-choiceques- -| structed by | domain | experts, | ensuring | high | precision | and ac- | | | | | | | | -| ----------- | ------ | -------- | -------- | ---- | --------- | ------- | --- | --- | --- | --- | --- | --- | --- | -tions. -| curacy. Although | | typically | smaller | than | crowd-sourced | cor- | | | | | | | | -| ---------------- | ---------- | --------- | ------- | ---- | ------------- | ---- | ---------- | ----- | ------- | ---------- | -------- | ------ | --------- | -| | | | | | | | AR-LSAT | [Wang | et | al., 2022] | is based | on | the LSAT, | -| pora, their | meticulous | design | makes | them | indispensable | for | | | | | | | | -| | | | | | | | containing | 2,064 | entries | spanning | ordering | games, | grouping | -in-depthlogicalreasoningevaluation. -games,andallocationgames,eachwithfiveoptions. -Exam-BasedDatasets[Liuetal.,2021b;Yuetal.,2020; -CLUTRR[Sinhaetal.,2019]focusesoninductivereason- -Wangetal.,2022]originatefromstandardizedtestquestions -| | | | | | | | ing, requiring | | models to | infer | kinship relationships | | in short | -| ---------------------- | --- | -------------- | ------- | ----- | -------- | --------- | -------------- | --------------------------------------------- | --------- | ----- | --------------------- | --- | -------- | -| (e.g., Chinese | | National Civil | Service | Exam, | LSAT, | GRE), | | | | | | | | -| | | | | | | | narratives. | Itcontains6,016entries,combiningentityextrac- | | | | | | -| offering high-quality, | | expert-crafted | | logic | problems | at scale. | | | | | | | | -tionandlogicalinference. -Thesedatasetsarewidelyusedtoevaluatereasoninginreal- -GSMevaluatesmathematicalreasoningcapabilities,com- -worldscenarios. -| | | | | | | | prising | two datasets: | | GSM8K | [Cobbe | et al., | 2021] (8.5K | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | --- | ----- | ------ | ------- | ----------- | -Table1summarizesimportantdatasetsforlogicalreason- -| | | | | | | | grade school | math | problems) | | and GSM-PLUS | | [Li et al., | -| ---------- | --------- | ----- | ----- | ------- | ------- | -------- | ------------ | ---- | --------- | --- | ------------ | --- | ----------- | -| ing, which | typically | cover | tasks | such as | Natural | Language | | | | | | | | -2024a](10,552),whichisaugmentedwithmathematicalper- -Inference(NLI)(§3.1)andMachineReadingComprehension -turbationsforrobustnessevaluation. -(MRC)(§3.2). -| | | | | | | | LINGOLY | | [Bean et | al., 2024] | uses Linguistic | | Olympiad | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | -------- | ---------- | --------------- | --- | -------- | -3.1 NaturalLanguageInference(NLI) puzzlestoevaluatein-contextpatternidentificationandgen- -| | | | | | | | eralization | in low-resource | | or | extinct languages. | | It contains | -| ------------- | --- | --------- | ---------- | --------- | --- | ------------ | ----------- | --------------- | --- | --- | ------------------ | --- | ----------- | -| NLI evaluates | | whether a | hypothesis | logically | | follows from | | | | | | | | -1,133problemsacross6formatsand5difficultylevels,cov- -| apremise,directlyassessingamodel’sreasoningability. | | | | | | La- | | | | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -eringover90languages. -belstypicallyfallintobinary(Entailment,Non-entailment)or -| ternary (Entailment, | | Contradiction, | | Neutral) | classifications. | | | | | | | | | -| -------------------- | --- | -------------- | --- | -------- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -3.3 BenchmarkSuites -SomedatasetsuseTrueandFalselabelsinstead. -ConTRoL[Liuetal., 2021a]isderivedfromrecruitment Benchmarksuitesstandardizeevaluationandfacilitatemodel -exams (e.g., bank entry, U.S. police selection), containing comparisoninlogicalreasoningresearch. -8,325 entries with Correct, Incorrect, and Can’t Say labels, GLoRE[liuetal.,2023d]isafew-shotandzero-shottest- -correspondingtoEntailment,Contradiction,andNeutral. ingplatform,including17test-onlydatasetstoassessgener- -FOLIO[Hanetal.,2024a]isanexpert-constructeddataset alizationinlow-datascenarios. -for First-Order Logic (FOL) reasoning, consisting of 1,351 LogiGLUE [Luo et al., 2024] consists of 24 logical -entrieslabeledasTrueorFalse,makingitarigorousbench- reasoning tasks, standardizing datasets into a sequence-to- -markforformallogicalinference. sequence format for uniform input processing. It provides -| | [Tian | | 2021] | | | | | | | | | | | -| --- | ----- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LogicNLI et al., contains 30K entries gener- bothtestandtrainingsets,enablingextensivemodeltraining -atedusinglogicalrules,withEntailment,Contradiction,and andtargetedevaluations. - -LogiTorch[Helweetal.,2022]isaPyTorch-basedlibrary ical sub-proofs without examples, generalization, and sensi- -for natural language logical reasoning, offering 16 datasets, tivitytosyntacticvariations[Saparovetal.,2023;Yuanetal., -2023;Rybetal.,2022]. -modelarchitectures,andanaccessibleAPIforquickevalua- -tion. -BIG-bench [Srivastava et al., 2022] is a collaborative 4.2 InductiveReasoning -| benchmarkwith7tasksdedicatedtologicalreasoning, | | | | such | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | ---- | --------- | ---------- | --- | ----- | ----------- | --- | ---- | ------------ | -| | | | | | Inductive | reasoning, | | which | generalizes | | from | specific in- | -asLogicGridPuzzleandLogicalFallacyDetection. -stancestobroaderrules,isessentialfortaskslikehypothesis -generationandpatternrecognition.WhileYangetal.[2024b] -| | | | | | find that | pre-trained | models | | can serve | as | effective | “reason- | -| --- | --- | --- | --- | --- | ----------------- | ----------- | ---------- | --------- | --------- | ------------ | --------- | ----------- | -| | | | | | ers,” Bowen | et | al. [2024] | show | that | even | advanced | LLMs | -| | | | | | struggle | with | simple | inductive | tasks | in their | symbolic | set- | -| | | | | | tings. Similarly, | | Sullivan | [2024] | | demonstrates | | that Trans- | -formermodels,evenafterfine-tuning,failtolearnfundamen- -| | | | | | tal logical | principles, | | indicating | limited | inductive | | reasoning | -| --- | --- | --- | --- | --- | ----------- | ----------- | --- | ---------- | ------- | --------- | --- | --------- | -capabilities. -| | | | | | 4.3 AbductiveReasoning | | | | | | | | -| --- | --- | --- | --- | --- | ---------------------- | ---------- | --- | ----- | ----- | -------- | --------- | ------ | -| | | | | | Abductive | reasoning, | | which | seeks | the most | plausible | expla- | -nationsforobservedphenomena,iscrucialinfieldslikelaw -| | | | | | and medicine. | | Del and | Fishel | [2023] | highlights | | the chal- | -| --- | --- | --- | --- | --- | ------------- | ------------ | ------- | ---------- | --------- | ------------ | ---------- | --------- | -| | | | | | lenges LLMs | | face in | generating | plausible | | hypotheses | from | -| | | | | | incomplete | information. | | In | the legal | domain, | | Nguyen et | -| | | | | | al. [2023] | show | that | despite | strong | performance, | | models | -strugglewithabductivereasoning,underscoringthecomplex- -ityofthisparadigm. -| | | | | | 4.4 AnalogicalReasoning | | | | | | | | -| --- | --- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | -Analogicalreasoning,whichinfersunknowninformationby -| | | | | | comparing | it with | known | information, | | is | vital for | tasks re- | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ------------ | --- | --- | --------- | --------- | -quiringcreativityandknowledgetransfer.Wijesiriwardeneet -| (a) A multi-choice | reading | comprehension | example from | the | | | | | | | | | -| ------------------ | ------- | ------------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -al.[2023]introducedANALOGICAL,abenchmarkforlong- -LogiQAdataset. -textanalogicalreasoning.Theyfindthatasanalogycomplex- -| | | | | | ity increases, | | LLMs struggle | | to recognize | | analogical | pairs. | -| --- | --- | --- | --- | --- | -------------- | --- | ------------- | --- | ------------ | --- | ---------- | ------ | -PetersenandvanderPlas[2023]showthatmodelscanlearn -analogicalreasoningwithminimaldata,approachinghuman -| | | | | | performance. | | However, | Qin | et al. | [2024] | question | whether | -| --- | --- | --- | --- | --- | ------------ | -------- | ------------- | --- | ---------- | ------- | ----------- | ------- | -| | | | | | LLMs truly | rely | on analogical | | reasoning, | | discovering | that | -| | | | | | random | examples | in prompts | | often | achieve | comparable | per- | -formancetorelevantexamples. -| | | | | | 4.5 OverallAnalysisandMetrics | | | | | | | | -| --- | --- | --- | --- | --- | ----------------------------- | ------- | -------- | --- | ----- | ----------- | --- | --------- | -| | | | | | Liu et al. | [2023b] | evaluate | | GPT-4 | and ChatGPT | | on bench- | -(b) AnNLIexamplefromtheConTRoLdataset. marks like LogiQA and ReClor, showing that while GPT- -| | | | | | 4 outperforms | | ChatGPT, | both | of them | struggle | | with out-of- | -| --- | --- | --- | --- | --- | ------------- | --- | -------- | ---- | ------- | -------- | --- | ------------ | -Figure2:ExampletestsofLogicalreasoninginNLPtasks. -| | | | | | distribution | tasks. | Xu | et al. | [2023] | introduce | | the NeuLR | -| --- | --- | --- | --- | --- | ------------ | ------ | --- | ------ | ------ | --------- | --- | --------- | -4 Evaluations -datasetandproposeaframeworkevaluatingLLMsacrosssix -| | | | | | dimensions: | correctness, | | rigor, | self-awareness, | | | proactivity, | -| --------------------- | --- | -------------- | -------- | ------ | ----------- | ------------ | --- | ------ | --------------- | --- | --- | ------------ | -| The rapid development | | of pre-trained | language | models | | | | | | | | | -guidance,andabsenceofhallucinations. -| (PLMs) necessitates | rigorous | evaluation | of their logical | rea- | | | | | | | | | -| ------------------- | -------- | ---------- | ---------------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -soning capabilities. This section examines four reasoning Metrics for Evaluating Logical Reasoning. Traditional -paradigms—deductive, inductive, abductive, and analogi- metrics like accuracy and F1 score are insufficient for as- -cal—whileanalyzingevaluationapproachesandmetrics. sessing logical reasoning. Recent studies have introduced -| | | | | | nuanced | metrics | such | as consistency | | (invariance | | to logi- | -| --- | --- | --- | --- | --- | ------- | ------- | ---- | -------------- | --- | ----------- | --- | -------- | -4.1 DeductiveReasoning -callyequivalentinputs),generalization(performanceonout- -Deductivereasoning,derivingspecificconclusionsfromgen- of-distribution data), and explainability (clarity of reason- -eralpremises,iscrucialforautomatedtheoremproving. De- ing steps). Thatikonda et al. [2025] find that combining -spite LLMs performing well on tasks like compositional BERTScorewithtraditionalmetricsimprovesalignmentwith -[2024c] -proofs, standard benchmarks, and encoding entailment rela- human judgments. Liu et al. propose a framework -tionships, they struggle with extended reasoning, hypothet- for measuring logical consistency, showing that BERTScore - -aligns better with human rankings than LLM-based evalua- 5.2 Model-CentricApproaches -| torslikeGPT-4. | | Gandarelaetal.[2024]emphasizestheneed | | | | | | | | | | | | | -| -------------- | --- | ------------------------------------- | --- | --- | --- | --- | --- | ------------- | ---------- | ------- | ----- | --------- | --- | ----- | -| | | | | | | | | Model-Centric | approaches | enhance | LLMs’ | reasoning | | capa- | -for metrics that account for the expressivity of logical theo- bilitiesbyoptimizingmodelparametersanddecodingstrate- -ries,particularlyininductivereasoning. gies. Theformalobjectiveis: -| 5 EnhancementMethods | | | | | | | | | (θ∗,S∗)=argmaxR(M | | | | | | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | ----- | --- | --- | -| | | | | | | | | | | | | θ ,S) | | (2) | -θ,S -| Enhancing | LLMs’ | logical | reasoning | | remains | crucial. | This | | | | | | | | -| --------- | ----- | ------- | --------- | --- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -where: -| sectionfocusesoncorestrategies: | | | | Data-CentricApproaches | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(§5.1), Model-Centric Approaches (§5.2), External Knowl- • θ: learnablemodelparameters. -edge Utilization (§5.3), and Neuro-Symbolic Reasoning M modelwithparametersθ. -• θ : -(§5.4). -• S: decodingstrategy(e.g.,chain-of-thoughtprompting, -| 5.1 Data-CentricApproaches | | | | | | | | verification-baseddecoding). | | | | | | | -| -------------------------- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | -Data-centricapproachesenhanceLLMs’reasoningcapabili- • R: reasoningperformancemetric. -Thisformulationhighlightsthejointoptimizationofmodel -| ties by utilizing | | meticulously | | curated | training | datasets. | For- | | | | | | | | -| ----------------- | --- | ------------ | --- | ------- | -------- | --------- | ---- | --- | --- | --- | --- | --- | --- | --- | -mally,thiscanbeexpressedas: parametersθanddecodingstrategyS. Practicalimplementa- -tionscanbecategorizedas: -| | | D∗ =argmaxR(M | | | ) | | (1) | | | | | | | | -| --- | --- | ------------- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | ------------ | --- | --- | --- | -| | | | | | D | | | • InstructionFine-Tuning: | | | optimizingθ. | | | | +Issue Brief + +MTV/AP-NORC: COMPARING THE +POLITICAL VIEWS OF YOUNG +PEOPLE AND THEIR +PARENTS’ +GENERATION + +Young people and their parents share a negative outlook on +the current political landscape in America and are largely +dissatisfied with the government’s handling of a wide +variety of key issues, according to a new poll of Americans +age 15-34, and parents of children age 15-26 by MTV and The +Associated Press-NORC Center for Public Affairs Research. + +Young people’s opinions about their generation’s ability to +impact the government have shifted in recent months, +following student-driven activism. Between the first poll +fielded between late February and early March 2018, +shortly after the shooting in Parkland, Florida, and this +second poll conducted roughly two months later, young +people age 15-34, and particularly those age 15-22, feel +more politically empowered. The proportion overall who +believe that elected officials care what they think rose +from a quarter to a third, and there was a similar rise in +the numbers who think people like them can impact the +government. + +The generation of parents of children age 15-26 is less +racially and ethnically diverse and includes a greater +proportion of Republicans than young people age 15-26, +and the survey demographics reflect this. Parents are +more likely to approve of President Donald Trump (42 +percent vs. 26 percent) and are somewhat more optimistic +(59 percent vs. 45 percent) than the younger generation. +Still, the findings show that young people and parents are +on the same page about politics in general. Majorities say +they tend to agree with one another on many key issues +and that their political discussions are generally amicable. + +© 2017 iStock/Nikada + +Three Things You Should Know +About the MTV/AP-NORC Poll of +Americans Age 15-34 and Parents +of 15-26 Year Olds: + +1) + +In the wake of recent protests and +demonstrations, 46 percent of young +people age 15-34 believe that people +like them can affect the government +at least a moderate amount, an +increase from 37 percent in the March +2018 poll. For 15-22 year olds, this rose +15 points, to 48 percent. + +2) Nearly two-thirds of young people +age 15-26 and parents of this age +group discuss politics at least once a +month. Majorities say these +conversations are civil and that they +are at least sometimes able to change +others’ political views when +disagreements do occur. + +3) More than two-thirds across both + +generations are disillusioned with the +two major political parties and feel a +third is needed. + +© Copyright 2018. The Associated Press and NORC + +June 2018 1 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Both young people age 15-26 and parents of this age group give the federal government a grade of F +more often than any other letter grade when assessing its performance on a wide variety of issues, +including gun control, the cost of higher education, foreign relations, immigration, environment and +climate change, health care, income inequality, taxes, sexual harassment and assault, abortion, racism, +and the treatment of citizens by police and the justice system. The economy and job creation fare +better, most commonly receiving grades of C from both groups. + +This survey is the second in a multiwave series of polls highlighting the voices of the youngest +generation of voters age 15-34, with this installment including a sample of parents of 15-26 year olds to +allow for comparisons between generations. This nationwide poll by MTV and The AP-NORC Center +surveyed a total of 939 young people, including 148 teens age 15-17, 410 adults age 18-26, and 381 adults +age 27-34, as well as 607 parents of young people age 15-26. + +AFTER RECENT DEMONSTRATIONS, YOUNG PEOPLE FEEL MORE +EMPOWERED IN POLITICS. + +Since the first MTV/AP-NORC Youth Political Pulse survey in March 2018, overall outlook among the +larger group of young people including those age 15-34 remains stable. Similar to the previous survey, +just under half are optimistic and a slight majority are pessimistic about the future of the United States. +Four in 10 have felt angry or anxious about the state of the country in the past month, while 1 in 10 +have felt excited or positive. + +Young peoples’ overall standpoint is fairly negative, but there is one area where young people have +grown somewhat more optimistic. After many recent protests and demonstrations, more now than in +March say elected officials care at least a moderate amount what people like them think (34 percent vs. +25 percent) and that people like them can affect the government at least a moderate amount (46 +percent vs. 37 percent). + +Since March 2018, more young people feel efficacious in American government. + +t +n +e +c +r +e +P + +100 + +90 + +80 + +70 + +60 + +50 + +40 + +30 + +20 + +10 + +0 + +37 + +25 + +46 + +34 + +March 2018 + +May 2018 + +Elected officials care at least a moderate amount + +People like me can affect government at least a moderate amount + +Questions: How much do elected public officials care what people like you think? +How much can people like you affect what the government does? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people age 15-22 in particular are increasingly optimistic when it comes to political efficacy. In +March, just a third said that people like them can affect what the government does at least a moderate +amount. Now, 48 percent of those age 15-22 say the same. + +© Copyright 2018. The Associated Press and NORC + +June 2018 2 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young peoples’ priorities have also shifted. When asked to describe in their own words what issue +facing the country concerns them the most, 21 percent cited gun laws in March. Just 6 percent mention +gun issues now.1 The most common worries are now the economy (16 percent) and social inequality (14 +percent). + +BOTH YOUNG PEOPLE AND PARENTS SAY POLITICS HAVE BECOME +DYSFUNCTIONAL, AND BOTH ARE DISSATISFIED WITH THE TWO-PARTY +SYSTEM. + +Young people age 15-26 and parents of young people age 15-26 alike have mixed outlooks on the future +of the United States. Young people are more pessimistic than parents (55 percent vs. 40 percent), +although these differences are mainly attributable to the parents being more Republican and less +racially and ethnically diverse than young people. + +Both generations report relatively negative feelings about the state of the country over the past month. +Substantial minorities of both groups have felt anxious or angry in the past month, while few have felt +positive or excited. Young people are more likely than parents to say they feel anxious (43 percent vs. +29 percent) and less likely to feel positive (8 percent vs. 17 percent) or excited (8 percent vs. 15 percent). + +Young people age 15-26 are more anxious and less positive or excited about the state of America than +are parents of young people. + +y + +l +e +m +e +r +t +x +e + +r +o + +y +r +e +v +y +a +s + +o +h +w + +t +n +e +c +r +e +P + +100 + +90 + +80 + +70 + +60 + +50 + +40 + +30 + +20 + +10 + +0 + +43 + +29 + +40 + +36 + +17 + +8 + +15 + +8 + +Anxious + +Angry + +Positive + +Excited + +Young people age 15-26 + +Parents of young people age 15-26 + +Question: In the past month… How _____ have you felt about the state of the country? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people and parents have similar assessments of the political system and their own efficacy +within it. Majorities of both groups say American politics have been dysfunctional over the past +month. Sixty-four percent of young people and 67 percent of parents feel that politics have not been +functioning very well or at all. + +Young people see the current two-party system as inadequate, and parents agree. Less than a third +think the Republican and Democratic parties do an adequate job of representing the American people, +while more than two-thirds say a third party is needed. + +1 Interviews for this survey were conducted April 23-May 9, 2018, prior to the May 18 shooting in Santa Fe, Texas. + +© Copyright 2018. The Associated Press and NORC + +June 2018 3 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +YOUNG PEOPLE AND PARENTS ARE SIMILARLY ENGAGED WITH THE +MIDTERM ELECTIONS. + +When it comes to the upcoming midterm elections in November, parents of young people age 15-26 are +more likely than young people age 15-26 to be regularly reading or watching news about the elections +(66 percent vs. 45 percent). However, older adults tend to consume news more often than young +people in general2 and may be hearing more about the midterms as a result. + +The two generations are equally likely to engage with the November elections in other ways, such as +posting or commenting about them on social media (25 percent of parents vs. 25 percent of young +people age 15-26), sharing political memes (18 percent of parents vs. 24 percent of young people age 15- +26), participating in events (14 percent of parents vs. 16 percent of young people age 15-26), or +volunteering with a candidate or issue campaign (11 percent of parents vs. 9 percent of young people +age 15-26). + +BOTH GENERATIONS HAVE VARIED POLITICAL CONCERNS AND ARE +STRONGLY DISAPPOINTED WITH THE GOVERNMENT’S PERFORMANCE +ON KEY ISSUES. + +Economic issues top long lists of current concerns for both young people and parents. When asked to +say in their own words what concerns them most about the issues currently facing the country, 15 +percent of young people age 15-26 and 17 percent of parents of young people age 15-26 say they are +most concerned about the economy. The two groups place similar priorities on a range of other issues, +although social inequality ranks second highest for young people (13 percent) and fails to make the top +five most worrisome issues for parents (6 percent). + +2 The Media Insight Project. 2014. The Personal News Cycle. +http://www.apnorc.org/PDFs/Media%20Insight/News%20Cycle/The_Personal_News_Cycle_Final.pdf + +© Copyright 2018. The Associated Press and NORC + +June 2018 4 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and parents report a wide range of top concerns of issues facing the country. + +The economy + +15 + +17 + +Social inequality + +War or threat of nuclear war + +13 + +6 + +11 +10 + +Immigration + +Gun laws + +Other issues + +Political climate + +Concerns about President Trump, leadership + +Foreign policy + +The environment + +Health care + +Terrorism, homeland security, national security + +Crime and safety + +Education + +Nothing + +Everything + +7 +7 + +7 + +4 + +6 + +9 + +5 + +8 + +5 + +7 + +5 +6 + +4 +5 + +3 + +6 + +2 + +4 + +2 +3 + +2 +2 + +1 +0 + +0 +1 + +0 + +10 + +20 + +30 + +40 + +50 + +60 + +70 + +80 + +90 + +100 + +Percent who offered each response + +Young people age 15-26 + +Parents of young people age 15-26 + +Question: Of all of the issues facing the country right now, what concerns you the most? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people age 15-26 believe the federal government is doing a poor job addressing a majority of key +issues. They most commonly give the government a failing grade for 15 of 17 key issues. However, +young people rate performance on the economy and job creation as average, most commonly giving a +grade of C. + +© Copyright 2018. The Associated Press and NORC + +June 2018 5 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Parents most commonly give the same grades as young people age 15-26 on most issues, with a few +exceptions. They give Cs rather than Fs for LGBT rights, the quality of education, terrorism and +homeland security, and taxes. + +Young people age 15-26 and parents give low marks to the federal government on its handling of many +issues. + +Young People +Age 15-26 + +Parents of Young +People Age 15-26 + +The environment and climate change + +Gun control + +Racism + +Immigration + +Cost of higher education + +Treatment of citizens by police and justice system + +Abortion + +Income inequality + +Sexual harassment and assault + +The quality of education + +Taxes + +Foreign relations + +Health care policy + +The rights of LGBT Americans + +Terrorism and homeland security + +The economy + +Job creation + +A + +B + +C + D -| where: | | | | | | | | • ReinforcementLearning: | | | optimizingθ. | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | ------------ | --- | --- | --- | -• D: trainingdatasets. • Inference-TimeDecoding: optimizingS. -• M : modeltrainedonD. Model-Centric approaches focus on directly improving + +F A + +B + +C + D -| | | | | | | | | the model’s | reasoning | capabilities | by optimizing | | its internal | | -| ---- | ----------- | --------- | --- | ------ | --------------- | --- | ----- | ----------- | ------------ | ------------ | ------------- | --- | ------------ | --- | -| • R: | performance | evaluator | | (e.g., | LLM-as-a-judge, | | rule- | | | | | | | | -| | | | | | | | | mechanisms | and decoding | strategies, | making | | them comple- | | -basedmetrics). -mentarytodata-centricapproaches. -| This formulation | | highlights | | the central | role | of | dataset op- | | | | | | | | -| ---------------- | --- | ---------- | --- | ----------- | ---- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | -InstructionFine-Tuning -| timization | in data-centric | | approaches. | | In | practice, | data- | | | | | | | | -| --------------- | --------------- | --------- | ----------- | ----- | ----- | --------- | --------- | ----------------------------------------- | ----------- | ----- | ----------- | -------------- | --- | ------ | -| | | | | | | | | Instruction | Fine-Tuning | (IFT) | adapts LLMs | through | | super- | -| centric methods | | typically | involve | three | types | of | datasets: | | | | | | | | -| | | | | | | | | visedlearningontask-specificinstructions. | | | | Forexample,Liu | | | -expert-curateddatasets,syntheticdatasets,andLLM-distilled -etal.[2023c]designmulti-grainedinstructionsspanningdi- -datasets. -| | | | | | | | | verse levels | of abstraction | and | complexity. | Similarly, | | Feng | -| -------------- | --- | ----------- | --- | --------- | ------ | ------------ | ------- | ------------- | -------------- | --------- | ----------- | ---------- | ------------ | ------ | -| Expert-Curated | | Datasets. | | The FOLIO | | series [Han | et al., | [2024] | | | | | | | -| | | | | | | | | et al. | IFT models | to mimic | logical | solvers | by | repli- | -| 2024a; Han | et | al., 2024b] | | establish | formal | verification | | | | | | | | | -| | | | | | | | | cating formal | deduction | reasoning | processes. | | In addition, | | -throughFOLannotations,withP-FOLIOextendingthecom- Xuetal.[2024a]implementtwo-stagesymbolicfine-tuning -plexity of reasoning chains for enhanced training. Lean- through Injection (injecting symbolic knowledge) and Infu- -Dojo[Yangetal.,2023]provides98k+human-proventheo- sion(balancingsymbolandNLreasoning). -| rempairsformathematicalreasoning. | | | | | Additionally,Symbol- | | | | | | | | | et | -| --------------------------------- | --- | --- | --- | --- | -------------------- | --- | --- | ----------- | ----- | ------------ | ------------ | --- | ---- | --- | -| | | | | | | | | To overcome | IFT’s | over-fitting | limitations, | | Wang | | -LLM[Xuetal.,2024a]systematicallyorganizes34symbolic -| | | | | | | | | al. [2024b] | enforce | contrastive | learning | | between | fac- | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | ----------- | -------- | --- | ------- | ---- | -reasoning tasks to capture inter-symbol relationships across tual/counterfactual paths with IFT. Further, Wang et -20distinctsymbolicfamilies. al. [2024a] augment Llamas with a Program-Guided Learn- -ingFrameworkandlogic-specificarchitectureadjustments. -| Synthetic | Datasets. | | Rule-based | | synthetic | data | remains | | | | | | | | -| --------- | --------- | --- | ---------- | --- | --------- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Recently,Muennighoffetal.[2025]proposes1,achieving -| fundamental | for | data generation. | | RuleTaker | | [Clark | et al., | | | | | | | | -| ----------- | --- | ---------------- | --- | --------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | -2021]formalizesthisthroughathree-phasepipeline: behav- test-time scaling through IFT on 1,000 meticulously crafted -ior formalization, example synthesis and linguistic equiva- long CoT samples. Combined with budget-forcing tech- -lents generation. Similarly, Morishita et al. [2024] devel- nique, itsignificantlyenhances thereasoningcapabilityofa -Qwen2.5-32B-Instructmodel,allowingextrapolatingbeyond -| ops Formal | Logic | Deduction | | Diverse | (FLD | ×2 ), | a synthetic | | | | | | | | -| ---------- | ----- | --------- | --- | ------- | ---- | ----- | ----------- | --- | --- | --- | --- | --- | --- | --- | -itsperformancewithouttest-timeintervention. -datasetbasedonsymbolictheoryandpreviousempiricalin- -sights. -ReinforcementLearning -LLM-Distilled Datasets. Researchers employ advanced Reinforcementlearning(RL)hasbecomepivotalinoptimiz- -models such as GPT-4 for intermediate reasoning step dis- ing large language models (LLMs), particularly since the -tillation. LogiCoT [Liu et al., 2023c] augments existing breakthroughofReinforcementLearningfromHumanFeed- -datasets with GPT4-generated reasoning chains, while Log- back (RLHF). Jiao et al. [2024] leverage RL for planning- -icPro[Jiangetal.,2024]combinesalgorithmicproblemswith basedreasoningoptimization,whileXietal.[2024]develop -code solutions to create variable-guided reasoning data. To R3,achievingprocesssupervisionbenefitsthroughoutcome- -| advance, | Wang | et al. [2024b] | | propose | PODA, | which | gener- | onlysupervision. | | | | | | | -| -------- | ---- | -------------- | --- | ------- | ----- | ----- | ------ | ---------------- | --- | --- | --- | --- | --- | --- | -atescontrastiveanalysesofcorrect/incorrectoptionsthrough The success of large-scale RL in OpenAI-o1 [OpenAI, -2024]hasinspirednumerousstudies.RLalgorithmstraino1- -premise-orientedaugmentation,enablingreasoningpathdif- -ferentiationviacontrastivelearning. stylemodelstoenhanceChain-of-Thought(CoT)reasoning, - -| addressing | issues | like | formulaic | outputs | | and limited | long- | | | | | | | | | -| ---------- | ------ | ---- | --------- | ------- | --- | ----------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | -form reasoning. For instance, Zhao et al. [2024] integrate (M∗,K∗)=argmaxR(M,K) (3) -M,K -| CoT instruction | | fine-tuning | | with Monte | Carlo | Tree | Search | | | | | | | | | -| --------------- | -------- | ----------- | ---------- | ---------- | ----- | ------------ | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | -| (MCTS) | decoding | for | multi-path | reasoning | | exploration. | In | where: | | | | | | | | -contrast,Zhangetal.[2024]employMCTStogeneratecode- -| | | | | | | | | • M: | the neural | model, | which | includes | both | the | model’s | -| --------- | ---- | --------------- | --- | ----------- | --- | ----- | ---------- | ---------- | ---------- | ------ | ------------ | ---------- | ---- | ----------- | ------- | -| reasoning | data | for instruction | | fine-tuning | | (IFT) | and Direct | | | | | | | | | -| | | | | | | | | parameters | | and | its decoding | strategies | | (generally, | the | -PreferenceOptimization(DPO). -model’sparametersremainunchanged). -| A significant | | breakthrough | | comes | | from | DeepSeek- | | | | | | | | | -| ------------- | --- | ------------ | --- | ----- | --- | ---- | --------- | ---- | --------- | --- | ----------- | --------- | --------- | --- | ------ | -| | | | | | | | | • K: | knowledge | | integration | strategy, | including | | knowl- | -R1[DeepSeek-AI,2025],whichpioneersanovelRLstrategy -| | | | | | | | | edge | source | curation, | structured | | knowledge | representa- | | -| ---------- | ------- | ---------- | --- | ----------------- | --- | --- | ------- | ---- | ------ | --------- | ---------- | --- | --------- | ----------- | --- | -| to enhance | logical | reasoning. | | DeepSeek-R1-Zero, | | | trained | | | | | | | | | -tion,retrieval-augmentedmechanisms,etc. -purelythroughRLwithoutIFT,demonstratesimpressiverea- -R: -soning capabilities but faces challenges in readability and • reasoningperformanceevaluator(e.g.,factualaccu- -language consistency. To address this, DeepSeek-R1 intro- racy,logicalconsistency). -ducesminimallong-CoTIFTdataasacoldstartbeforeRL, -ZayyadandAdi[2024]andYangetal.[2023]extractdata -| achieving | a balance | between | | usability | and | reasoning | per- | | | | | | | | | -| --------- | --------- | ----------- | ------------ | --------- | ------------ | --------- | ------- | ---------- | -------------- | --- | ----- | ----- | ------ | ------- | ----- | -| | | | | | | | | from Lean, | a mathematical | | proof | tool, | to aid | theorem | prov- | -| formance. | By | iteratively | synthesizing | | high-quality | | reason- | | | | | | | | | -ing. Incontrast,“Logic-Query-of-Thoughts”(LQOT)[Liuet -ing data through RL, DeepSeek-R1 overcomes limitations al.,2024b]decomposescomplexlogicalproblemsintoeasier -| imposed | by human | annotators, | | addressing | | issues | such as | | | | | | | | | -| ------- | -------- | ----------- | --- | ---------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -sub-questionsbeforeintegratingknowledgegraphs. -| mechanistic | responses, | | repetitive | patterns, | | and | insufficient | | | | | | | | | -| ----------- | ---------- | --- | ---------- | --------- | ---------- | --- | ------------ | ------------------ | --- | -------------- | ------- | ------- | ---------- | ---------- | ---------- | -| | | | | | | | | In reading | | comprehension, | | Ouyang | et | al. [2023] | con- | -| long-chain | reasoning. | | This | approach | represents | | a potential | | | | | | | | | -| | | | | | | | | struct supergraphs | | to | address | complex | contextual | | reasoning, | -paradigmshiftinlogicalreasoningoptimization,pushingthe -whileKnowRA[Maietal.,2025]autonomouslydetermines -boundariesofwhatLLMscanachieveinstructuredreasoning -| | | | | | | | | whether | to accept | external | knowledge | | to assist | document- | | -| ---------------------- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --------- | -------- | --------- | --- | --------- | --------- | --- | -| tasks. | | | | | | | | levelrelationextraction. | | | | | | | | -| Inference-TimeDecoding | | | | | | | | 5.4 Neuro-SymbolicApproaches | | | | | | | | -We categorize logical reasoning enhancement methods dur- Neural-symbolic hybrid methods represent a burgeoning re- -| ing inference-time | | into | inference-time | | | scaling | and con- | | | | | | | | | -| ------------------ | --- | ---- | -------------- | --- | --- | ------- | -------- | ----------- | ---- | ---- | ---------- | --- | -------- | ----------- | --- | -| | | | | | | | | search area | that | aims | to combine | the | powerful | representa- | | -straineddecoding. -tionalcapabilitiesofdeeplearningwiththeprecisionandin- -| Inference-time | | scaling | employs | computational | | | augmenta- | | | | | | | | | -| -------------- | --- | ------- | ------- | ------------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -terpretabilityofsymbolicreasoning. -tion without parameter updates. One common approach is Formally, a neural-symbolic hybrid system aims to opti- -decoding with structured outputs and modular workflows. mize both the neural model M and the symbolic solver P -| GoT [Lei | et al., | 2023] | creates | structured | | reasoning | nodes | | | | | | | | | -| ------------ | ------- | ----- | ---------- | ---------- | ---------- | ---------- | ----- | ------------- | -------------------------------------------- | --- | ------------ | --- | ----------- | --- | --------- | -| | | | | | | | | (whereP | representsthesymbolicreasoningprocess)tomax- | | | | | | | -| to improve | complex | | multi-step | logical | reasoning. | | Simi- | | | | | | | | | -| | | | | | | | | imize logical | reasoning | | performance. | | The overall | | objective | -| larly, Chain | of | Logic | [Servantez | et | al., 2024] | introduces | a | | | | | | | | | -canbeexpressedas: -Decomposition-Recompositionstructureforlegalreasoning. -In other contexts, researchers design more complex modu- (M∗,P∗)=argmaxR(P(M(x))), -lar workflows for better performance [Creswell et al., 2023; M,P -Malonetal.,2024]. -where: -| Another | inference-time | | scaling | approach | | involves | stimu- | | | | | | | | | -| ------- | -------------- | --- | ------- | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -• M: Theneuralmodel,whichincludesboththemodel’s -| lating autonomous | | reasoning, | | guiding | LLMs | to | iteratively | | | | | | | | | -| ----------------- | -------- | ---------- | -------- | ------- | --------- | ----- | ----------- | ----------------------------------- | --- | --- | --- | --- | --- | -------------- | --- | -| | | | | | | | | parametersanditsdecodingstrategies. | | | | | | Itmapstheinput | | -| refine their | answers. | | Maieutic | | Prompting | [Jung | et al., | | | | | | | | | -x(e.g.,naturallanguage)intoasymbolicrepresentation -2022]eliminatescontradictionsthroughrecursivereasoning. -zwithinaformallanguageL: -| Similarly, | Logic-of-Thoughts | | | [Liu | et al., | 2024a] | and De- | | | | | | | | | -| ---------- | ----------------- | --- | --- | ---- | ------- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -termLR[Sunetal.,2024]progressivelyapproachtheanswers -| | | | | | | | | | | | z =M(x), | z | ∈L. | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -inaniterativestyle. -Constraineddecodingmethods,ontheotherhand,focuson • P: The symbolic solver, which operates on the sym- -improvingthecontrollabilityandreliabilityofreasoningpro- bolic representation z produced by M to generate the -cesses. Neurologic[Luetal.,2021]enforcespredicatelogic finaloutputy: -| | | | | [Li | | 2024b] | | | | | y =P(z). | | | | | -| ------------ | ----- | ---------- | --- | --- | ------- | ------ | ---------- | --- | --- | --- | -------- | --- | --- | --- | --- | -| constraints, | while | Formal-LLM | | | et al., | | integrates | | | | | | | | | -automataforconstrainingplangeneration. -| | | | | | | | | • R: | Thereasoningperformancemetric, | | | | | whichevaluates | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------------------------------ | --- | --- | --- | --- | -------------- | --- | -theabilitytoperformlogicalreasoningtasks. -5.3 ExternalKnowledgeUtilization -Theoptimizationprocessinvolvestwokeydirections: -LLMsoftengenerateincorrectanswersduetohallucinations -| | | | | | | | | • Improving | | M: including | | refining | the model’s | | parame- | -| --------------- | --- | ------- | ----- | ---- | --- | ------- | ---------- | ----------- | --- | ------------ | --- | -------- | ----------- | --- | ------- | -| when performing | | complex | tasks | such | as | logical | reasoning, | | | | | | | | | -tersanddecodingstrategiestoproducesymbolicrepre- -makingitnecessarytoincorporateexternalknowledgetoas- -sist in producing accurate responses. Formally, the optimal sentationsthatarebothaccurateandcompatiblewithP. -integrationofexternalknowledgecanbeformulatedasajoint • Enhancing P: involving improving the symbolic -| optimizationproblem: | | | | | | | | solver’sabilitytoprocess. | | | | | | | | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | - -By jointly optimizing M and P, neural-symbolic hybrid bottlenecks with large knowledge bases or complex rule de- -systemsaimtoleveragethestrengthsofbothneuralnetworks pendencies. Conversely, data-driven methods (e.g., instruc- -| | | | | | | | | | | | [Parmar | | 2024]) | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | ------ | --- | -andsymbolicreasoningtoachievesuperiorlogicalreasoning tion tuning on LogicBench et al., achieve -capabilities. Itisworthnotingthatinearlierneural-symbolic broadertaskcoveragebutfailtogeneralizebeyondsyntactic -pipelines,P isoftenimplementedasafixedexternallogical patterns. How can we reconcile transparent reasoning with -reasoningengine,andthusisgenerallynotoptimized. How- black-box model performance? Hybrid architectures offer -ever,inadvancedpractice,LLMsareincreasinglybeingused promisebutintroducecomputationaloverhead,limitingprac- -| toperformtheroleofP,enablingdiverseoptimization. | | | | | | | ticaldeployment. | | | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | -Fundamentally, these methods involve translating prob- ExistingbenchmarkslikeLogiQA[Liu -EvaluationRigor. -| lems into | symbolic | representations | | with | LLMs, | and ex- | | | | | | | | | -| --------------- | -------- | --------------- | ------- | ----- | ----- | ----------- | -------------- | ---- | ---------- | ----------- | ------- | ------- | --------------- | ------- | -| | | | | | | | et al., 2021b] | | and ReClor | [Yu | et al., | 2020] | conflate | reason- | -| ternal symbolic | | solvers | solving | them. | For | example, in | | | | | | | | | -| | | | | | | | ing ability | with | pattern | recognition | | through | multiple-choice | | -LINC[Olaussonetal.,2023],LLMsconvertnaturallanguage -| | | | | | | | formats. | While | efforts | like | NeuLR | [Xu et | al., 2023] | curate | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------- | ---- | ----- | ------ | ---------- | ------ | -(NL) into first-order logic (FOL) expressions, and utilize an “neutral” content to isolate reasoning from domain knowl- -externaltheoremproverforsymbolicdeductiveinference. -| | | | | | | | edge,theylackscopeforholisticevaluation. | | | | | | Currentmetrics | | -| ------- | ------- | -------- | --------- | -------------- | --- | ------ | ---------------------------------------- | --- | --------------------------------------- | --- | --- | --- | -------------- | --- | -| Further | efforts | focus on | improving | NL-to-symbolic | | trans- | | | | | | | | | -| | | | | | | | (e.g., accuracy, | | BLEU)failtoassessconsistency(invariance | | | | | | -lation. Oneprevailingapproachisdirectlyoptimizingtrans- -tologicallyequivalentinputs)orsoundness(adherencetofor- -| lation through | | training | [Yang | et al., | 2024a] | or decoding | | | | | | | | | -| -------------- | ---- | -------------- | ----- | --------- | ------ | ----------- | -------------------- | --- | ----------- | ------------------------------- | --- | ---- | ---------- | ---- | -| | | | | | | | malproofstructures). | | | Whatdefinesagoldstandardforlog- | | | | | -| strategies | [Ryu | et al., 2024], | | while the | other | depends on | | | | | | | | | -| | | | | | | | ical reasoning | | evaluation? | Benchmarks | | must | prioritize | sys- | -verification or correction mechanisms [Yang et al., 2024a; tematictestingofcoreprinciples(e.g.,transitivity,contrapo- -Panetal.,2023]. -sition)overtask-specificperformance. -Buildinguponthese,recentadvancementsaddressthetra- -ditional pipeline limitations by fully integrating LLMs into Future Directions. Addressing these challenges requires -reasoning processes. Logic Agent (LA) [Liu et al., 2024a] hybrid architectures that dynamically integrate neural and -| | | | | | | | symbolic | components, | | such | as | differentiable | | theorem | -| ------------- | -------- | ------- | -------- | ----------- | ----- | ---------- | ----------------------------------------- | ----------- | --- | ---- | --- | -------------- | ------------- | ------- | -| replaces | external | solvers | with | rule-guided | LLM | inference | | | | | | | | | -| | | | | | | | provers,tobalancescalabilityandprecision. | | | | | | Equallyimpor- | | -| chains, while | LLM-TRes | | [Toroghi | et al., | 2024] | implements | | | | | | | | | -tantisthedevelopmentofevaluationframeworksthatstress- -self-containedverifiablereasoningwithoutexternalsymbolic -solvers. SymbCoT[Xuetal.,2024c]coordinatestranslation, test models on perturbed logical statements (e.g., negated -planning,solvingandverificationentirelythroughLLMs.Xu premises, swapped quantifiers) to isolate reasoning from -[2024b] memorization. Multimodal reasoning, which grounds infer- -| et al. | propose | Aristotle, | | which | further | systematizes | | | | | | | | | -| ------------ | --------- | ----------- | --- | ------- | ------- | ------------ | ---------------- | ------- | ------------- | ------ | ---------- | ------ | ------------------ | --- | -| | | | | | | | ence in | diverse | modalities | (text, | images, | code), | presents | un- | -| the symbolic | reasoning | pipeline | | through | three | LLM-driven | | | | | | | | | -| | | | | | | | tapped potential | | for enhancing | | robustness | | and interpretabil- | | -| components: | Logical | Decomposer, | | Logical | Search | Router, | | | | | | | | | -andLogicalResolver. ity. Finally, interdisciplinary collaboration—leveraging in- -| | | | | | | | sights from | formal | logic, | cognitive | | science, | and | machine | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ------ | ------------ | --------- | --------- | -------- | ---- | ------- | -| | | | | | | | learning—will | | be essential | | to design | systems | that | reason | -6 Discussion -| | | | | | | | withandaboutuncertainty. | | | | UntilLLMsreliablydisentangle | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ---------------------------- | --- | --- | --- | -Theintegrationoflogicalreasoningintolargelanguagemod- logicfromlexicon,theirdeploymentinhigh-stakesdomains -els(LLMs)remainsacriticalchallenge,markedbypersistent -| | | | | | | | will remain | precarious. | | Bridging | this | gap | demands | rigorous | -| --- | --- | --- | --- | --- | --- | --- | ----------- | ----------- | --- | -------- | ---- | --- | ------- | -------- | -gapsbetweenheuristicperformanceandformallogicalrigor. -| | | | | | | | benchmarks, | scalable | | hybrid | methods, | and | a redefinition | of | -| --- | --- | --- | --- | --- | --- | --- | ----------- | -------- | --- | ------ | -------- | --- | -------------- | --- | -Below,weanalyzethreeunresolvedtensionsdominatingthe -evaluationparadigms. -fieldandoutlinefuturedirections. -7 Conclusion -| Robustness | vs. | Generalization. | | LLMs | exhibit | inconsis- | | | | | | | | | -| ---------- | --- | --------------- | --- | ---- | ------- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -tentperformanceinstructuredreasoningtaskssuchasdeduc- -| | | | | | | | This survey | synthesizes | | the | rapid advancements | | and | persis- | -| -------------- | --- | --------- | ---------- | --- | ----------- | ----- | ----------- | ----------- | --- | --- | ------------------ | --- | --- | ------- | -| tive inference | and | abductive | hypothesis | | generation. | While | | | | | | | | | -tentchallengesinlogicalreasoningforlargelanguagemodels -modelsfine-tunedondatasetslikeFOLIO[Hanetal.,2024a] -(LLMs). WhileLLMsdemonstrateimpressiveheuristicrea- -| excel in | controlled | settings, | they | struggle | with | adversarial | | | | | | | | | -| -------- | ---------- | --------- | ---- | -------- | ---- | ----------- | ------- | -------- | ------- | ------------------ | --- | --- | ---------- | --- | -| | | | | | | | soning, | rigorous | logical | inference—spanning | | | deductive, | in- | -perturbations or semantically equivalent rephrasings. This ductive, abductive, and analogical paradigms—remains in- -| inconsistency | arises | from | their | reliance | on surface-level | sta- | | | | | | | | | -| --------------------- | --------------------------------------- | ------ | ----- | --------------------- | ---------------- | ----------- | --------------------- | ------------- | -------------- | --------- | -------------- | ------------ | --------------- | --------- | -| | | | | | | | consistent | due | to limitations | | in robustness, | | generalization, | | -| tistical correlations | | rather | than | causal relationships, | | coupled | | | | | | | | | -| | | | | | | | and interpretability. | | We | analyzed | strategies | | to enhance | rea- | -| with limited | out-of-distribution | | | generalization. | | A key ques- | | | | | | | | | -| | | | | | | | soning, | including | neuro-symbolic | | | integration, | data-centric | | -| tionpersists: | canLLMsachievehuman-likerobustnesswith- | | | | | | | | | | | | | | -| | | | | | | | tuning, | reinforcement | | learning, | test-time | scaling, | | and other | -out sacrificing cross-domain adaptability? Current methods improved decoding methods, and highlighted benchmarks -prioritizenarrowtaskperformance,leavingreal-worldappli- -| | | | | | | | like FOLIO | and | LogiQA | for | systematic | evaluation. | | Future | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------ | --- | ---------- | ----------- | --- | ------ | -cabilityuncertain. -progresshingesonhybridarchitecturesthatunifyneuraland -Interpretabilityvs.Performance. Acentraltensionliesin symbolicreasoning,robustevaluationframeworks,andscal- -balancingneuralscalabilitywithsymbolicprecision. Neuro- able methods for cross-domain and multimodal inference. -symbolic approaches like Logic-LM [Pan et al., 2023] and Addressing these challenges will advance LLMs toward re- -| Symbol-LLM[Xuetal., | | | 2024a]embedformallogicsolvers | | | | | | | | | | | | -| ------------------- | --- | --- | ----------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -liable,interpretablereasoningcriticalforreal-worldapplica- -| into neural | architectures, | | improving | interpretability | | through | tions. | | | | | | | | -| ------------------- | -------------- | ----------------------------------- | --------- | ---------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| step-by-stepproofs. | | However,thesemethodsfacescalability | | | | | | | | | | | | | - -| References | | | | | | | [Liuetal.,2021b] | | | | -| ---------- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | -JianLiu,LeyangCui,HanmengLiu,DandanHuang,Yile -| | | | | | | | Wang,andYueZhang. | Logiqa: | achallengedatasetformachinereading | | -| ---------------- | ------ | ------- | ---- | --------- | ------------ | ----- | ----------------- | ------- | ---------------------------------- | --- | -| [Beanetal.,2024] | Andrew | M Bean, | Simi | Hellsten, | Harry Mayne, | Jabez | | | | | -Magomere,EthanAChi,etal. Lingoly: Abenchmarkofolympiad-level comprehensionwithlogicalreasoning.2021. -linguisticreasoningpuzzlesinlow-resourceandextinctlanguages. arXiv [Liuetal.,2023a] HanmengLiu,JianLiu,LeyangCui,ZhiyangTeng,Nan -preprintarXiv:2406.06196,2024. -| | | | | | | | Duan,etal. Logiqa2.0—animproveddatasetforlogicalreasoninginnat- | | | | -| ----------------- | ----------------------------------- | --- | --- | --- | --- | ----- | --------------------------------------------------------------- | ----------------------------------- | --- | --- | -| | | | | | | | urallanguageunderstanding. | IEEE/ACMTransactionsonAudio,Speech, | | | -| [Bowenetal.,2024] | ChenBowen,RuneSætre,andYusukeMiyao. | | | | | Acom- | | | | | -andLanguageProcessing,pages2947–2962,2023. -prehensiveevaluationofinductivereasoningcapabilitiesandproblemsolv- -| inginlargelanguagemodels. | | | InProc.ofACLFindings,pages323–339, | | | | [Liuetal.,2023b] | | | | -| ------------------------- | --- | --- | ---------------------------------- | --- | --- | --- | ---------------- | --- | --- | --- | -HanmengLiu,RuoxiNing,ZhiyangTeng,JianLiu,Qiji -2024. Zhou,andYueZhang. Evaluatingthelogicalreasoningabilityofchatgpt -| [Cann,1993] | RonnieCann. | Formalsemantics: | | anintroduction. | | Cambridge | andgpt-4,2023. | | | | -| ----------- | ----------- | ---------------- | --- | --------------- | --- | --------- | -------------- | --- | --- | --- | -UniversityPress,UnitedStates,1993. -| | | | | | | | [Liuetal.,2023c] | HanmengLiu,ZhiyangTeng,LeyangCui,ChaoliZhang, | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --------------------------------------------- | --- | --- | -[CingilliogluandRusso,2019] Nuri Cingillioglu and Alessandra Russo. QijiZhou,andYueZhang. Logicot: Logicalchain-of-thoughtinstruction -Deeplogic:Towardsend-to-enddifferentiablelogicalreasoning,2019. tuning.InProc.ofEMNLPFindings,pages2908–2921,2023. -[Clarketal.,2021] PeterClark,OyvindTafjord,andKyleRichardson.Trans- [liuetal.,2023d] Hanmeng liu, Zhiyang Teng, Ruoxi Ning, Jian Liu, Qiji -formersassoftreasonersoverlanguage.InProc.ofIJCAI,2021. Zhou,andYueZhang. Glore: Evaluatinglogicalreasoningoflargelan- -guagemodels,2023. -| [Cobbeetal.,2021] | Karl | Cobbe, | Vineet | Kosaraju, | Mohammad | Bavarian, | | | | | -| ----------------- | ---- | ------ | ------ | --------- | -------- | --------- | --- | --- | --- | --- | -[Liuetal.,2024a] -MarkChen,HeewooJun,etal.Trainingverifierstosolvemathwordprob- Hanmeng Liu, Zhiyang Teng, Chaoli Zhang, and Yue -lems.arXivpreprintarXiv:2110.14168,2021. Zhang.Logicagent:Enhancingvaliditywithlogicruleinvocation,2024. -[Creswelletal.,2023] -AntoniaCreswell, MurrayShanahan, andIrinaHig- [Liuetal.,2024b] LihuiLiu,ZihaoWang,RuizhongQiu,YikunBan,Eunice -gins. Selection-inference: Exploiting large language models for inter- Chan,etal. Logicqueryofthoughts: Guidinglargelanguagemodelsto -pretablelogicalreasoning.InProc.ofICLR,2023. answercomplexlogicquerieswithknowledgegraphs,2024. -[DeepSeek-AI,2025] DeepSeek-AI. DeepSeek-R1:IncentivizingReasoning [Liuetal.,2024c] YinhongLiu,ZhijiangGuo,TianyaLiang,EhsanShareghi, -CapabilityinLLMsviaReinforcementLearning.Technicalreport,2025. IvanVulic´,andNigelCollier. Aligningwithlogic: Measuring,evaluating -andimprovinglogicalconsistencyinlargelanguagemodels.arXivpreprint -| [DelandFishel,2023] | | MaksymDelandMarkFishel.Truedetective:Adeep | | | | | | | | | -| ------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | -arXiv:2410.02205,2024. -abductivereasoningbenchmarkundoableforGPT-3andchallengingfor -GPT-4. InProceedingsofthe12thJointConferenceonLexicalandCom- [Luetal.,2021] Ximing Lu, Peter West, Rowan Zellers, Ronan Le Bras, -putationalSemantics(*SEM2023),pages314–322,2023. Chandra Bhagavatula, and Yejin Choi. NeuroLogic decoding: -[Fengetal.,2024] (un)supervisedneuraltextgenerationwithpredicatelogicconstraints. In -JiazhanFeng,RuochenXu,JunhengHao,HiteshiSharma, Proc.ofNAACL,pages4288–4299,2021. -YelongShen,etal.Languagemodelscanbedeductivesolvers.InProc.of -ACLFindings,pages4026–4042,2024. [Luoetal.,2024] ManLuo, ShrinidhiKumbhar, Mingshen, MihirParmar, -| | | | | | | | NeerajVarshney,etal. | Towardslogiglue:Abriefsurveyandabenchmark | | | -| --------------------- | --- | --------------------------------------------- | --- | --- | --- | --- | -------------------- | ----------------------------------------- | --- | --- | -| [Gandarelaetal.,2024] | | Joa˜oPedroGandarela,DaniloSCarvalho,andAndre´ | | | | | | | | | -foranalyzinglogicalreasoningcapabilitiesoflanguagemodels,2024. -| Freitas. Inductivelearningoflogicaltheorieswithllms: | | | | | Acomplexity- | | | | | | -| ---------------------------------------------------- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | -gradedanalysis.arXivpreprintarXiv:2408.16779,2024. [Maietal.,2025] Chengcheng Mai, Yuxiang Wang, Ziyu Gong, Hanxiang -Wang,andYihuaHuang.Knowra:Knowledgeretrievalaugmentedmethod -| [Hanetal.,2024a] | Simeng | Han, | Hailey Schoelkopf, | | Yilun Zhao, | Zhenting | | | | | -| ---------------- | ------ | ---- | ------------------ | --- | ----------- | -------- | --- | --- | --- | --- | -Qi,MartinRiddell,etal. FOLIO:Naturallanguagereasoningwithfirst- fordocument-levelrelationextractionwithcomprehensivereasoningabili- -ties,2025. -orderlogic.InProc.ofEMNLP,pages22017–22031,2024. -[Malonetal.,2024] -[Hanetal.,2024b] SimengHan,AaronYu,RuiShen,ZhentingQi,Martin ChristopherMalon,MartinMin,XiaodanZhu,etal.Ex- -ploringtheroleofreasoningstructuresforconstructingproofsinmulti- -| Riddell,etal. | P-FOLIO:Evaluatingandimprovinglogicalreasoningwith | | | | | | | | | | -| ------------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -InProc.ofEMNLPFindings, stepnaturallanguagereasoningwithlargelanguagemodels. InProc.of -abundanthuman-writtenreasoningchains. EMNLP,pages15299–15312,2024. -pages16553–16565,2024. -[Harutaetal.,2020] [McCarthyandHayes,1981] J.McCarthyandP.J.Hayes. Somephilosophi- -IzumiHaruta,KojiMineshima,andDaisukeBekki.Log- calproblemsfromthestandpointofartificialintelligence. InReadingsin -| icalinferenceswithcomparativesandgeneralizedquantifiers. | | | | | | InProc.of | | | | | -| -------------------------------------------------------- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | -ArtificialIntelligence,pages431–450.1981. -ACL,pages263–270,2020. -[McCarthy,1959] -[Helweetal.,2022] ChadiHelwe,Chloe´Clavel,andFabianSuchanek.Log- JohnMcCarthy. Programswithcommonsense. InPro- -ceedingsoftheTeddingtonConferenceontheMechanizationofThought -itorch: Apytorch-basedlibraryforlogicalreasoningonnaturallanguage. -| InProc.ofEMNLP,2022. | | | | | | | Processes,1959. | | | | -| -------------------- | --- | ------------- | ---- | --------- | -------- | ---------- | --------------- | ----------------------------------------------------- | --- | --- | -| | | | | | | | [McCarthy,1989] | JohnMcCarthy.Artificialintelligence,logicandformaliz- | | | -| [Jiangetal.,2024] | Jin | Jiang, Yuchen | Yan, | Yang Liu, | Yonggang | Jin, Shuai | | | | | -Peng,etal. Logicpro:Improvingcomplexlogicalreasoningviaprogram- ingcommonsense. PhilosophicalLogicandArtificialIntelligence,pages -161–190,1989. -guidedlearning.arXivpreprintarXiv:2409.12929,2024. -[Jiaoetal.,2024] Fangkai Jiao, Chengwei Qin, Zhengyuan Liu, Nancy F. [Morishitaetal.,2024] TerufumiMorishita,GakuMorio,AtsukiYamaguchi, -| | | | | | | | andYasuhiroSogawa. | Enhancingreasoningcapabilitiesofllmsviaprin- | | | -| ------------------- | ----------- | --------------------------------------------- | ------------- | -------- | --------- | ----- | --------------------------- | -------------------------------------------- | --- | --- | -| Chen,andShafiqJoty. | | Learningplanning-basedreasoningbytrajectories | | | | | | | | | -| | | | | | | | cipledsyntheticlogiccorpus. | InProc.ofNeurIPS,pages73572–73604, | | | -| collection | and process | reward | synthesizing. | In Proc. | of EMNLP, | pages | 2024. | | | | -334–350,2024. -[Muennighoffetal.,2025] -| [Jungetal.,2022] | | | | | | | | NiklasMuennighoff,ZitongYang,WeijiaShi,Xi- | | | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------ | --- | --- | -JaehunJung,LianhuiQin,SeanWelleck,FaezeBrahman, angLisaLi,LiFei-Fei,etal. s1:Simpletest-timescaling. arXivpreprint -| ChandraBhagavatula,etal. | | Maieuticprompting:Logicallyconsistentrea- | | | | | | | | | -| ------------------------ | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -arXiv:2501.19393,2025. -soningwithrecursiveexplanations.InProc.ofEMNLP,pages1266–1279, -2022. [NewellandSimon,1956] A. Newell and H. Simon. The logic theory -[Leietal.,2023] BinLei,ChunhuaLiao,CaiwenDing,etal.Boostinglogical machine–acomplexinformationprocessingsystem. IRETransactionson -InformationTheory,1956. -| reasoninginlargelanguagemodelsthroughanewframework: | | | | | | Thegraph | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -ofthought.arXivpreprintarXiv:2308.08614,2023. [Nguyenetal.,2023] Ha-Thanh Nguyen, Randy Goebel, Francesca Toni, -| | | | | | | | KostasStathis,andKenSatoh. | Howwelldosotalegalreasoningmodels | | | -| --------------- | ---------------------------------------------- | --- | --- | --- | --- | --- | -------------------------- | --------------------------------- | --- | --- | -| [Lietal.,2024a] | QintongLi,LeyangCui,XueliangZhao,LingpengKong, | | | | | | | | | | -supportabductivereasoning?,2023. -| andWeiBi. | GSM-plus: | Acomprehensivebenchmarkforevaluatingthe | | | | | | | | | -| --------------------------------------------- | --------- | --------------------------------------- | --- | --- | ------------- | --- | -------------------- | -------------- | -------- | ------------------- | -| robustnessofLLMsasmathematicalproblemsolvers. | | | | | InProc.ofACL, | | [Olaussonetal.,2023] | | | | -| | | | | | | | | Theo Olausson, | Alex Gu, | Ben Lipkin, Cedegao | -pages2961–2984,2024. Zhang,ArmandoSolar-Lezama,etal. LINC:Aneurosymbolicapproach -[Lietal.,2024b] ZelongLi,WenyueHua,HaoWang,HeZhu,andYongfeng forlogicalreasoningbycombininglanguagemodelswithfirst-orderlogic -provers.InProc.ofEMNLP,pages5153–5176,2023. -| Zhang. Formal-llm: | | Integratingformallanguageandnaturallanguagefor | | | | | | | | | -| ------------------ | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -controllablellm-basedagents.arXivpreprintarXiv:2402.00798,2024. [OpenAI,2024] OpenAI. LearningtoreasonwithLLMs. Technicalreport, -2024. -| [Liuetal.,2021a] | HanmengLiu,LeyangCui,JianLiu,andYueZhang.Nat- | | | | | | | | | | -| ---------------- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -[Ouyangetal.,2023] -urallanguageinferenceincontext-investigatingcontextualreasoningover SiruOuyang,ZhuoshengZhang,andHaiZhao. Fact- -longtexts.Proc.ofAAAI,pages13388–13396,2021. drivenlogicalreasoningformachinereadingcomprehension,2023. - -| [Panetal.,2023] | | | | | | | [Tianetal.,2021] | | | | | | -| --------------- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | -LiangmingPan, AlonAlbalak, XinyiWang, andWilliam JidongTian,YitianLi,WenqingChen,LiqiangXiao,Hao -Wang. Logic-LM: Empowering large language models with symbolic He, andYaohuiJin. Diagnosingthefirst-orderlogicalreasoningability -solversforfaithfullogicalreasoning. InProc.ofEMNLPFindings,pages throughLogicNLI.InProc.ofEMNLP,pages3738–3747,2021. -3806–3824,2023. -| | | | | | | | [Toroghietal.,2024] | Armin Toroghi, | | Willis Guo, | Ali Pesaranghader, | and | -| --- | --- | --- | --- | --- | --- | --- | ------------------- | -------------- | --- | ----------- | ------------------ | --- | -[Parmaretal.,2023] MihirParmar, NeerajVarshney, NisargPatel, Santosh ScottSanner. Verifiable, debuggable, andrepairablecommonsenselogi- -Mashetty, ManLuo, etal. Logicbench: Abenchmarkforevaluationof calreasoningviallm-basedtheoryresolution. InProc.ofEMNLP,pages -| logicalreasoning,2023. | | | | | | | 6634–6652,2024. | | | | | | -| ---------------------- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -[Parmaretal.,2024] MihirParmar,NisargPatel,NeerajVarshney,Mutsumi [Wangetal.,2019] Po-Wei Wang, Priya L. Donti, Bryan Wilder, and Zico -Nakamura,ManLuo,etal. Logicbench:Towardssystematicevaluationof Kolter.Satnet:Bridgingdeeplearningandlogicalreasoningusingadiffer- -logicalreasoningabilityoflargelanguagemodels. InProc.ofACL,pages entiablesatisfiabilitysolver,2019. -13679–13707,2024. -| | | | | | | | [Wangetal.,2022] | Siyuan Wang, | Zhongkun | | Liu, Wanjun | Zhong, Ming | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------------ | -------- | --- | ----------- | ----------- | -[Pereira,1982] -FernandoCarlosNevesPereira. Logicfornaturallanguage Zhou,ZhongyuWei,etal.Fromlsat:Theprogressandchallengesofcom- -analysis.1982. plexreasoning. IEEE/ACMTransactionsonAudio,Speech,andLanguage -Processing,2022. -| [PetersenandvanderPlas,2023] | | | MollyPetersenandLonnekevanderPlas. | | | | | | | | | | -| ---------------------------- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Canlanguagemodelslearnanalogicalreasoning?investigatingtrainingob- [Wangetal.,2024a] ChenWang,XudongLi,HaoranLiu,XinyueWu,and -jectivesandcomparisonstohumanperformance.InProc.ofEMNLP,pages WantingHe. Efficientlogicalreasoninginlargelanguagemodelsthrough -16414–16425,2023. program-guidedlearning.AuthoreaPreprints,2024. -[Plaatetal.,2024] AskePlaat,AnnieWong,SuzanVerberne,JoostBroekens, [Wangetal.,2024b] Chenxu Wang, Ping Jian, and Zhen Yang. Thought- -| NikivanStein,andThomasBack. | | | | Reasoningwithlargelanguagemodels, | | | | | | | | | -| --------------------------- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -pathcontrastivelearningviapremise-orienteddataaugmentationforlogical -asurvey.arXivpreprintarXiv:2407.11511,2024. readingcomprehension.arXivpreprintarXiv:2409.14495,2024. -[Qinetal.,2024] -Chengwei Qin, Wenhan Xia, Tan Wang, Fangkai Jiao, [Wijesiriwardeneetal.,2023] Thilini Wijesiriwardene, Ruwan Wickrama- -YuchenHu,etal. Relevantorrandom: Canllmstrulyperformanalogi- rachchi,BimalGajera,ShreeyashGowaikar,ChandanGupta,etal. ANA- -calreasoning?,2024. LOGICAL-anovelbenchmarkforlongtextanalogyevaluationinlarge -[Rajaramanetal.,2023] KanagasabaiRajaraman,SaravananRajamanickam, languagemodels.InProc.ofACLFindings,pages3534–3549,2023. -andWeiShi. Investigatingtransformer-guidedchainingforinterpretable [Xietal.,2024] ZhihengXi,WenxiangChen,BoyangHong,SenjieJin,Rui -naturallogicreasoning.InProc.ofACLFindings,pages9240–9253,2023. -Zheng,etal.Traininglargelanguagemodelsforreasoningthroughreverse -curriculumreinforcementlearning.InProc.ofICML,2024. -| [Rybetal.,2022] | Samuel | Ryb, | Mario | Giulianelli, | Arabella | Sinclair, and | | | | | | | -| --------------- | ------ | ---- | ----- | ------------ | -------- | ------------- | --- | --- | --- | --- | --- | --- | -RaquelFerna´ndez. AnaLog: Testinganalyticalanddeductivelogiclearn- [Xiaetal.,2024] YuXia, RuiWang, XuLiu, MingyanLi, TongYu, etal. -abilityinlanguagemodels.InProceedingsofthe11thJointConferenceon Beyondchain-of-thought:Asurveyofchain-of-xparadigmsforllms.arXiv -LexicalandComputationalSemantics,pages55–68,2022. preprintarXiv:2404.15676,2024. -[Ryuetal.,2024] HyunRyu,GyeongmanKim,HyeminSLee,andEunho [Xuetal.,2023] FangzhiXu,QikaLin,JiaweiHan,TianzheZhao,JunLiu, -Yang. Divide and translate: Compositional first-order logic transla- andErikCambria. Arelargelanguagemodelsreallygoodlogicalreason- -| tion and | verification | for | complex | logical | reasoning. | arXiv preprint | | | | | | | -| -------- | ------------ | --- | ------- | ------- | ---------- | -------------- | --- | --- | --- | --- | --- | --- | -ers?acomprehensiveevaluationandbeyond,2023. -arXiv:2410.08047,2024. -| | | | | | | | [Xuetal.,2024a] | FangzhiXu,ZhiyongWu,QiushiSun,SiyuRen,FeiYuan, | | | | | -| ------------------- | --- | -------- | -------- | ------- | ------- | ------------- | ------------------ | ---------------------------------------------- | ------------ | --- | -------------- | ------------- | -| [Saparovetal.,2023] | | Abulhair | Saparov, | Richard | Yuanzhe | Pang, Vishakh | | | | | | | -| | | | | | | | et al. Symbol-LLM: | Towards | foundational | | symbol-centric | interface for | -Padmakumar,NitishJoshi,MehranKazemi,etal. Testingthegeneralde- largelanguagemodels.InProc.ofACL,pages13091–13116,2024. -ductivereasoningcapacityoflargelanguagemodelsusingoodexamples. -InProc.ofNeurIPS,pages3083–3105,2023. [Xuetal.,2024b] Jundong Xu, Hao Fei, Meng Luo, Qian Liu, Liangming -| | | | | | | | Pan, etal. | Aristotle: Masteringlogicalreasoningwithalogic-complete | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ------------------------------------------------------- | --- | --- | --- | --- | -[Servantezetal.,2024] Sergio Servantez, Joe Barrow, Kristian Hammond, decompose-search-resolveframework. arXivpreprintarXiv:2412.16953, -| andRajivJain. | Chainoflogic: | | Rule-basedreasoningwithlargelanguage | | | | | | | | | | -| ------------- | ------------- | --- | ------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -2024. -models.InProc.ofACLFindings,pages2721–2733,2024. -| | | | | | | | [Xuetal.,2024c] | JundongXu,HaoFei,LiangmingPan,QianLiu,Mong-Li | | | | | -| --- | --- | --- | --- | --- | --- | --- | --------------- | --------------------------------------------- | --- | --- | --- | --- | -[Shietal.,2021] JihaoShi,XiaoDing,LiDu,TingLiu,andBingQin.Neural Lee, andWynneHsu. Faithfullogicalreasoningviasymbolicchain-of- -naturallogicinferenceforinterpretablequestionanswering. InProc.of thought.InProc.ofACL,pages13326–13365,2024. -EMNLP,pages3673–3684,2021. [Yangetal.,2023] Kaiyu Yang, Aidan M. Swope, Alex Gu, Rahul Chala- -[Sinhaetal.,2019] KoustuvSinha,ShagunSodhani,JinDong,JoellePineau, mala, Peiyang Song, et al. Leandojo: theorem proving with retrieval- -andWilliamL.Hamilton. Clutrr: Adiagnosticbenchmarkforinductive augmentedlanguagemodels.InProc.ofICONIP,2023. -| reasoningfromtext. | | EmpiricalMethodsofNaturalLanguageProcessing | | | | | | | | | | | -| ------------------ | --- | ------------------------------------------- | --- | --- | --- | --- | ----------------- | --------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Yangetal.,2024a] | YuanYang,SihengXiong,AliPayani,EhsanShareghi, | | | | | -(EMNLP),2019. -| | | | | | | | andFaramarzFekri. | Harnessingthepoweroflargelanguagemodelsfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | ------------------------------------------ | --- | --- | --- | --- | -[Srivastavaetal.,2022] AarohiSrivastava,AbhinavRastogi,AbhishekRao, naturallanguagetofirst-orderlogictranslation. InProc.ofACL,pages -| AbuAwalMdShoeb,AbubakarAbid,etal. | | | | | Beyondtheimitationgame: | | 6942–6959,2024. | | | | | | -| --------------------------------- | --- | --- | --- | --- | ----------------------- | --- | --------------- | --- | --- | --- | --- | --- | -Quantifyingandextrapolatingthecapabilitiesoflanguagemodels. arXiv [Yangetal.,2024b] Zonglin Yang, Li Dong, Xinya Du, Hao Cheng, Erik -preprintarXiv:2206.04615,2022. -Cambria,etal.Languagemodelsasinductivereasoners.InProc.ofEACL, -| [Staceyetal.,2022] | | | | | | | pages209–225,2024. | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | -JoeStacey,PasqualeMinervini,HaimDubossarsky,and -| MarekRei. | Logicalreasoningwithspan-levelpredictionsforinterpretable | | | | | | | | | | | | -| --------- | --------------------------------------------------------- | --- | --- | --- | --- | --- | -------------- | ------------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Yuetal.,2020] | WeihaoYu,ZihangJiang,YanfeiDong,andJiashiFeng.Re- | | | | | -androbustNLImodels.InProc.ofEMNLP,pages3809–3823,2022. -clor:Areadingcomprehensiondatasetrequiringlogicalreasoning.InProc. -| [Sullivan,2024] | MichaelSullivan.Itisnottruethattransformersareinductive | | | | | | ofICLR,2020. | | | | | | -| --------------- | ------------------------------------------------------- | --- | --- | --- | --- | -------------- | ------------ | --- | --- | --- | --- | --- | -| learners: | ProbingNLImodelswithexternalnegation. | | | | | InProc.ofEACL, | | | | | | | -pages1924–1945,2024. [Yuetal.,2024] FeiYu,HongboZhang,PrayagTiwari,andBenyouWang. -| | | | | | | | Naturallanguagereasoning,asurvey. | | | ACMComputingSurveys,pages1– | | | -| --------------- | ----------------------------------------------- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --------------------------- | --- | --- | -| [Sunetal.,2023] | JiankaiSun,ChuanyangZheng,EnzeXie,ZhengyingLiu, | | | | | | 39,2024. | | | | | | -arXiv -RuihangChu,etal. Asurveyofreasoningwithfoundationmodels. [Yuanetal.,2023] ZhangdieYuan,SongboHu,IvanVulic´,AnnaKorhonen, -preprintarXiv:2312.11562,2023. -| | | | | | | | andZaiqiaoMeng. | Canpretrainedlanguagemodels(yet)reasondeduc- | | | | | -| --------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------ | -------------------------------------------- | --- | --- | --- | --- | -| [Sunetal.,2024] | | | | | | | tively? InProc.ofEACL,pages1447–1462,2023. | | | | | | -HongdaSun,WeikaiXu,WeiLiu,JianLuan,BinWang, -| etal. Determlr: | | Augmentingllm-basedlogicalreasoningfromindetermi- | | | | | | | | | | | -| --------------- | --- | ------------------------------------------------- | --- | --- | --- | --- | ------------------- | ---- | ------ | --------- | ----------- | -------- | -| | | | | | | | [ZayyadandAdi,2024] | Majd | Zayyad | and Yossi | Adi. Formal | language | -nacytodeterminacy.InProc.ofACL,pages9828–9862,2024. -knowledgecorpusforretrievalaugmentedgeneration,2024. -| [Tafjordetal.,2021] | | Oyvind | Tafjord, | Bhavana | Dalvi, | and Peter Clark. | | | | | | | -| ------------------- | ---------- | ------ | ------------- | ------- | ------ | -------------------- | ----------------- | --------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Zhangetal.,2024] | YuxiangZhang,ShangxiWu,YuqiYang,JiangmingShu, | | | | | -| ProofWriter: | Generating | | implications, | proofs, | and | abductive statements | | | | | | | -overnaturallanguage.InProc.ofACLFindings,pages3621–3634,2021. JinlinXiao,etal. o1-coder: ano1replicationforcoding. arXivpreprint -arXiv:2412.00154,2024. -| [Thatikondaetal.,2025] | | Ramya | Keerthy | Thatikonda, | | Wray Buntine, and | [Zhaoetal.,2024] | | | | | | -| ---------------------- | --- | ----- | ------- | ----------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | -YuZhao,HuifengYin,BoZeng,HaoWang,TianqiShi, -EhsanShareghi. Assessingthealignmentoffolclosenessmetricswithhu- etal.Marco-o1:Towardsopenreasoningmodelsforopen-endedsolutions. -manjudgement.arXivpreprintarXiv:2501.08613,2025. -arXivpreprintarXiv:2411.14405,2024. \ No newline at end of file + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +F + + + + + + + + + + + + + + + + + + + + + + + + Indicates the most commonly given grade, or mode. The frequency of each grade is indicated with darker shading for +more common responses, according to the key below. + +0% + +50% + +Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of +the following issues? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +PARENT/CHILD DISCUSSIONS ABOUT POLITICS ARE USUALLY CORDIAL, +BUT IT CAN BE DIFFICULT TO SWAY THE OTHER GENERATION WHEN +VIEWS DIFFER. + +Young people age 15-26 and their parents have fairly regular discussions with one another about +politics and social issues, and few see these discussions as a source of conflict. Across the two groups, +64 percent of young people and parents say they discuss politics with each other at least once a +month, 20 percent say they have these discussions less often, and 16 percent say they never talk about +politics or social issues together. + +Among those who discuss political and social issues with their parents or children, a majority say they +usually see eye to eye in these conversations, and about a third say they debate diplomatically. Fewer +say the discussion turns into a major conflict or requires changing the subject to avoid arguments. + +© Copyright 2018. The Associated Press and NORC + +June 2018 6 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Most parents and their children see eye to eye on politics and social issues. + +Someone changes the +subject quickly to avoid +getting into disagreements. +9% + +It turns into +World War III. +5% + +We debate things +diplomatically. +32% + +We usually see +eye to eye. +53% + +Question: When you talk with [him/her/them] about political and social issues, how would you describe how the discussion +goes? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +When the two generations were asked about political discussions with a particular family member, +parents’ accounts are similar regardless of the specific relationship, such as mother-daughter or father- +son. Sons reporting on discussions with fathers are more likely than other pairings to say they talk +about politics at least once a week, but least likely to say these discussions tend to go smoothly. Only +42 percent of sons say they usually see eye to eye with their fathers, while 49 percent of sons say the +same about their mothers, 54 percent of daughters say so about their fathers, and 62 percent say so +about their mothers. + +Those who identify as political independents (36 percent) are less likely than Democrats (60 percent) +or Republicans (53 percent) to say they usually see eye to eye in parent/child discussions about +political and social issues. + +When asked about potential differences in opinion on specific issues, both generations report high +levels of agreement. In their discussions, at least 6 in 10 say there is agreement when discussing each +of 18 topics. Agreement levels are highest on Trump’s job performance and on the issues of racism, +sexual harassment and assault, and health care policy, and lowest on LGBT rights, abortion, and +foreign relations. + +While agreement is high, there also seems to be a fair amount of ambiguity about family members’ +political positions. On 13 of the 18 issues asked about, at least 15 percent say they are unsure if they +tend to agree or disagree. + +© Copyright 2018. The Associated Press and NORC + +June 2018 7 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and their parents tend to be on the same page with many political and social issues. + +President Trump's job performance + +Racism + +Sexual harassment and assault + +Health care policy + +The quality of education + +Treatment of citizens by police and justice system + +Cost of higher education + +The economy + +Taxes + +The environment and climate change + +Income inequality + +Immigration + +Job creation + +Terrorism and homeland security + +Gun control + +The rights of LGBT Americans + +Foreign relations + +Abortion + +76 + +76 + +76 + +75 + +74 + +74 + +73 + +72 + +71 + +71 + +70 + +70 + +69 + +69 + +69 + +62 + +62 + +61 + +13 + +8 + +9 + +10 + +9 + +8 + +13 + +15 + +13 + +9 + +8 + +11 + +9 + +11 + +17 + +19 + +13 + +16 + +8 + +9 + +14 + +17 + +15 + +13 + +19 + +19 + +19 + +20 + +23 + +17 + +22 + +20 + +14 + +19 + +24 + +22 + +Tend to agree + +Tend to disagree + +Not sure + +Questions: When talking about the following issues, do you tend to agree or disagree with [him/her/them]? +When talking with [your family member] about Donald Trump and how he’s handling his job as president, do you tend to +agree or disagree with [him/her/them]? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +The two generations report different issues as the biggest sources of disagreement. Among those who +say they generally don’t see eye to eye when talking about political and social issues, young people +age 15-26 say they tend to disagree with their parents most about racism (37 percent), LGBT rights (36 +percent), and how Trump is handling his presidency (33 percent). Parents of young people age 15-26 +report that the largest sources of disagreement are gun control (27 percent), LGBT rights (25 percent), +and how Trump is handling his presidency (21 percent). + +When there are disagreements, few young people and parents are able to persuade one another to +change their views, and the findings are consistent between the generations. Eleven percent overall +say they are always or often able to persuade their family member to change their views. Fifty-three +percent acknowledge that they are sometimes able to persuade their family member. Just 6 percent +say the reverse is true, that they are often or always convinced to change their views by their family +member. Another 44 percent say their family member is sometimes able to persuade them to change +their own views. + +© Copyright 2018. The Associated Press and NORC + +June 2018 8 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Few young people or parents find it easy to persuade one another to change their political and social +views. + +How often you are able to persuade your + family member to change their views + +11 + +53 + +36 + +How often your family member is able + to persuade you to change your views + +6 + +44 + +49 + +0 + +10 + +20 + +30 + +40 + +50 + +60 + +70 + +80 + +90 + +100 + +Percent + +Always or often + +Sometimes + +Rarely or never + +Questions: In general, when you disagree with [him/her/that person] about political or social issues, how often would you +say that you’re able to persuade [him/her/them] to change [his/her/their views]? +In general, when you disagree with [him/her/that person] about political or social issues, how often would you say that [he +is/she is/they are] able to change your views? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +YOUNG PEOPLE ARE PERCEIVED AS MORE IDEALISTIC THAN THE OLDER +GENERATION, AND PARENTS FEEL THEIR CHILDREN’S GENERATION IS +LESS INFORMED THAN THEIR OWN. + +When asked to compare the generations, young people age 15-26 and parents of this age group tend to +agree on who is the more patriotic and rigid generation, but there is some disagreement about who is +more informed. + +About two-thirds of both young people and parents say the older generation is more patriotic. More +than half of each group feel the older generation is more rigid. + +Among both groups, the largest proportion view the parents’ generation as more politically active and +moral than the younger generation, and the younger generation as more idealistic. + +Young people are divided on whether their generation is more uninformed, their parents’ generation is +more uninformed, or whether the two are about equal. Parents believe the younger generation is more +uninformed than their own. + +© Copyright 2018. The Associated Press and NORC + +June 2018 9 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and parents agree that the older generation is more patriotic and more rigid. + +Which generation is more…? + +Patriotic + +Moral + +Rigid + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +Compassionate + +Young people +say… + +Parents say… + +Politically active + +Young people +say… + +Idealistic + +Uninformed + +Parents say… + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +64 + +10 + +25 + +20 + +40 + +60 + +80 + +100 + +65 + +7 + +27 + +20 + +44 + +20 + +20 + +20 + +20 + +53 + +57 + +52 + +40 + +60 + +16 + +40 + +40 + +40 + +40 + +60 + +13 + +60 + +60 + +19 + +60 + +28 + +34 + +40 + +60 + +23 + +40 + +40 + +60 + +33 + +60 + +80 + +39 + +80 + +33 + +80 + +100 + +100 + +100 + +17 + +25 + +80 + +100 + +28 + +80 + +37 + +80 + +40 + +80 + +100 + +100 + +100 + +25 + +80 + +100 + +32 + +28 + +40 + +60 + +80 + +100 + +30 + +40 + +28 + +23 + +32 + +19 + +20 + +20 + +20 + +20 + +40 + +60 + +80 + +100 + +47 + +30 + +40 + +60 + +32 + +40 + +43 + +40 + +60 + +60 + +80 + +34 + +80 + +37 + +80 + +100 + +100 + +100 + +20 + +38 + +20 + +41 + +20 + +37 + +20 + +Parents' generation + +Young people's generation + +Both about equally + +Question: Thinking about people in your generation, compared to people in [his/her/their] generation, do you think the +following words describe your generation better, describe [his/her/their] generation better, or describe both generations +about equally well? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +© Copyright 2018. The Associated Press and NORC + +June 2018 10 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Republicans are more likely than Democrats to describe their own generation as patriotic (51 percent +vs. 32 percent) and are less likely to describe their own generation as idealistic (27 percent vs. 39 +percent). + +FOR SOME, POLITICS SEEPS INTO INTERACTIONS WITH EXTENDED +FAMILY. + +A significant minority of both generations have had political disagreements with relatives that have +led to conflict. Twenty percent have unfriended or blocked a family member on social media due to a +disagreement over politics, and 21 percent have themselves been unfriended or blocked by a family +member. Fewer than 1 in 5 have had a holiday gathering ruined (11 percent), have decided not to attend +a family event (12 percent), or have had a family member decide to not attend a family event (17 +percent) due to political disagreements. + +While there are no differences by generation, Democrats are more likely than Republicans to have had +political disagreements with relatives affect their social media behavior. Democrats are more likely +than Republicans to say they’ve unfriended or blocked a family member on social media (26 percent +vs. 12 percent) and to have had a family member unfriend or block them on social media (27 percent vs. +15 percent). + +EVEN AMONG THOSE WHO APPROVE OF TRUMP, THE REPORT CARD IS +MEDIOCRE. + +Young people age 15-34 give poor marks to the federal government on its handling of many public +policy issues. Even among Trump’s supporters, the report card is still only average. Young people who +approve of Trump’s job performance most commonly give the federal government a C for all issues +except terrorism and homeland security and treatment of citizens by the police and justice system, +which earn a B, and abortion, which earns an F. For those who disapprove of Trump, F is the most +common grade for all issues except job creation, which most commonly gets graded a C. + +Most common grades for the federal government’s handling of issues. + +Approve of Trump + +Disapprove of Trump + +Gun control + +Cost of higher education + +Foreign relations + +Immigration + +The environment and climate change + +Health care policy + +Income inequality + +The economy + +Taxes + +Sexual harassment and assault + +Abortion + +Racism + +The rights of LGBT Americans + +Terrorism and homeland security + +Treatment of citizens by police and justice system + +The quality of education + +Job creation + +C + +C + +C + +C + +C + +C + +C + +C + +C + +C + +F + +C + +C + +B + +B + +C + +C + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +C + +Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of +the following issues? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +© Copyright 2018. The Associated Press and NORC + +June 2018 11 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Looking at grades by political party, the results are similar. Democrats’ most common grade across +most issues is an F. However, Democrats give better marks to the federal government’s efforts on job +creation, most commonly grading it a C. + +Republicans are slightly more mixed, but most commonly give issues a C. For abortion, they most +commonly give a grade of F. + +STUDY METHODOLOGY + +This survey was conducted by The Associated Press-NORC Center for Public Affairs Research and +with funding from MTV and NORC at the University of Chicago. Staff from NORC at the University of +Chicago, The Associated Press, and MTV collaborated on all aspects of the study. + +Data were collected using AmeriSpeak®, NORC’s probability-based panel designed to be +representative of the U.S. household population. During the initial recruitment phase of the panel, +randomly selected U.S. households were sampled with a known, non-zero probability of selection from +the NORC National Sample Frame and then contacted by U.S. mail, email, telephone, and field +interviewers (face-to-face). The panel provides sample coverage of approximately 97 percent of the U.S. +household population. Those excluded from the sample include people with P.O. Box only addresses, +some addresses not listed in the USPS Delivery Sequence File, and some newly constructed dwellings. + +Interviews for this survey were conducted between April 23 and May 9, 2018, with young people age +15-34 and parents of young people age 15-26 representing the 50 states and the District of Columbia. +Interviews were conducted in both English and Spanish, depending on respondent preference. Adult +panel members age 18-34 were randomly drawn from AmeriSpeak, and 791 completed the survey—780 +via the web and 11 via telephone. Teen panel members age 15-17 were drawn from AmeriSpeak Teen, +and 148 completed the survey—145 via the web and 3 via telephone. Adult panel members who are +known parents of 15-17 year olds were randomly drawn from AmeriSpeak, and 104 completed the +survey—101 via the web and 3 via telephone. In addition, panel members age 30-76 were randomly +drawn to screen for parents of young adults age 18-26, and 503 completed the survey—469 via the web +and 34 via telephone. + +Among young people age 15-34, the final stage completion rate is 25.9 percent, the weighted household +panel response rate is 33.7 percent, and the weighted household panel retention rate is 87.2 percent, for +a cumulative response rate of 7.6 percent. The overall margin of sampling error among young people +age 15-34 is +/- 4.3 percentage points at the 95 percent confidence level, including the design effect. The +overall margin of sampling error among young people age 15-26 is +/- 5.6 percentage points at the 95 +percent confidence level, including the design effect. The margin of sampling error may be higher for +subgroups. + +Among parents of young people age 15-26, the screener completion rate is 27.9 percent, with an +incidence rate of 25.2. The interview completion rate is 95.6 percent, and the final stage completion rate +is 26.7 percent. The weighted household panel response rate is 33.7 percent, and the weighted +household panel retention rate is 87.2 percent, for a cumulative response rate of 7.8 percent. The +overall margin of sampling error is +/- 6.7 percentage points at the 95 percent confidence level, +including the design effect. The margin of sampling error may be higher for subgroups. + +Once the sample has been selected and fielded, and all the study data have been collected and made +final, a poststratification process is used to adjust for any survey nonresponse as well as any +noncoverage or under- and oversampling resulting from the study specific sample design. +Poststratification variables included age, gender, census division, race/ethnicity, and education. + +© Copyright 2018. The Associated Press and NORC + +June 2018 12 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Weighting variables were obtained from the 2017 Current Population Survey. The weighted data +reflect the U.S. population of young people age 15-34 and parents of young people age 15-26. + +All analyses were conducted using STATA (version 14), which allows for adjustment of standard errors +for complex sample designs. All differences reported between subgroups of the U.S. population are at +the 95 percent level of statistical significance, meaning that there is only a 5 percent (or less) +probability that the observed differences could be attributed to chance variation in sampling. + +A comprehensive listing of all study questions, complete with tabulations of top-level results for each +question, is available on The AP-NORC Center for Public Affairs Research website: www.apnorc.org. + +For more information, email info@apnorc.org. + +CONTRIBUTING RESEARCHERS + +From NORC at the University of Chicago + +From The Associated Press + +Becky Reimer +Emily Alvarez +Liz Kantor +Trevor Tompson +Jennifer Benz + +Emily Swanson + +ABOUT THE ASSOCIATED PRESS-NORC CENTER FOR PUBLIC AFFAIRS +RESEARCH + +The AP-NORC Center for Public Affairs Research taps into the power of social science research and the +highest-quality journalism to bring key information to people across the nation and throughout the +world. + +■ + +The Associated Press (AP) is the world’s essential news organization, bringing fast, unbiased news +to all media platforms and formats. + +■ NORC at the University of Chicago is one of the oldest and most respected, independent research + +institutions in the world. + +The two organizations have established The AP-NORC Center for Public Affairs Research to conduct, +analyze, and distribute social science research in the public interest on newsworthy topics, and to use +the power of journalism to tell the stories that research reveals. + +The founding principles of The AP-NORC Center include a mandate to carefully preserve and protect +the scientific integrity and objectivity of NORC and the journalistic independence of AP. All work +conducted by the Center conforms to the highest levels of scientific integrity to prevent any real or +perceived bias in the research. All of the work of the Center is subject to review by its advisory +committee to help ensure it meets these standards. The Center will publicize the results of all studies +and make all datasets and study documentation available to scholars and the public. + +ABOUT MTV + +MTV is a global youth culture brand inspired by music. For more information, check out +mtvpress.com. MTV is a unit of Viacom Inc. (NASDAQ: VIAB, VIA) + +© Copyright 2018. The Associated Press and NORC + +June 2018 13 + diff --git a/q175/random_k4/question.json b/q175/random_k4/question.json index 0d22a7dc83b9c356eb52456882699b38ab028f3c..915422a14865a92089409d54fee636ce55fba64f 100644 --- a/q175/random_k4/question.json +++ b/q175/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "f86d073b0d735ac873a65d906ba82758.pdf", - "web_39b6cadc4e9ca03e.pdf", - "58a44062a5b5795816b8b85c9c8c3a95c53d18a7.pdf", - "web_ebb5e31663fc625e.pdf", - "web_46dbd475b3ebe8e7.pdf" + "resize_r2_0.pdf", + "web_660e5c500c425b81.pdf", + "web_7f91d188bac86e23.pdf", + "web_f3420c2dcb7612a6.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q175/random_k4/random_1.md b/q175/random_k4/random_1.md index 2570a92a07cb79b4daa90af30bf44b6e8b43f56a..052cee2fff6d2fb0526b1d43400b3c01f2587dfb 100644 Binary files a/q175/random_k4/random_1.md and b/q175/random_k4/random_1.md differ diff --git a/q175/random_k4/random_2.md b/q175/random_k4/random_2.md index bd35c3422555cd95beb556bbd614d1deaa052c77..e3ed939376d2f6fad348f021570f136f5c225f39 100644 --- a/q175/random_k4/random_2.md +++ b/q175/random_k4/random_2.md @@ -1,961 +1,1527 @@ -Logical Reasoning in Large Language Models: A Survey -HanmengLiu1†, ZhizhangFu1†, MengruDing1, RuoxiNing1 -ChaoliZhang2, XiaozhangLiu3 and YueZhang1∗ -1 WestlakeUniversity -2 ZhejiangNormalUniversity -3 HainanUniversity -{liuhanmeng,zhangyue}@westlake.edu.cn,{fuzhizhang.fzz,dingmengru2021}@gmail.com, -ruoxining@outlook.com,chaolizcl@zjnu.edu.cn,lxzh@hainanu.edu.cn -Abstract theirreasoningisincreasinglyvital. Asaresult,post-training -LLM for reasoning has garnered a surge of interest in both -Withtheemergenceofadvancedreasoningmodels industry and research[OpenAI, 2024; DeepSeek-AI, 2025; -like OpenAI o3 and DeepSeek-R1, large language Muennighoffetal.,2025]. -models(LLMs)havedemonstratedremarkablerea- Despite growing research, existing surveys [Plaat et al., -soning capabilities. However, their ability to per- 2024; Sun et al., 2023; Yu et al., 2024] often conflate log- -form rigorous logical reasoning remains an open -ical reasoning with general-purpose heuristic strategies like -question. This survey synthesizes recent advance- Chain-of-Thought (CoT) [Xia et al., 2024]. There has been -mentsinlogicalreasoningwithinLLMs, acritical -a lack of a literature review dedicated to LLM and formal -area of AI research. It outlines the scope of logi- -symbolic logic. This survey provides a comprehensive re- -calreasoninginLLMs,itstheoreticalfoundations, viewoflogicalreasoninginlargelanguagemodels(LLMs), -andthebenchmarksusedtoevaluatereasoningpro- -with a focus on formal and symbolic logic-based reasoning -ficiency. We analyze existing capabilities across -ratherthangeneralheuristicapproaches. Webeginbydefin- -differentreasoningparadigms—deductive,induc- -ing logical reasoning in AI, distinguishing it from general- -tive,abductive,andanalogical—andassessstrate- -purpose reasoning, and categorizing key paradigms, includ- -gies to enhance reasoning performance, including -ing deductive, inductive, abductive, and analogical reason- -data-centrictuning,reinforcementlearning,decod- ing. Additionally,weanalyzeexistingbenchmarksandeval- -ingstrategies,andneuro-symbolicapproaches.The -uationmethodologies,identifyinggapsinassessingsymbolic -reviewconcludeswithfuturedirections,emphasiz- -inference, consistency, and robustness. We further explore -ing the need for further exploration to strengthen -state-of-the-art techniques for enhancing logical reasoning, -logicalreasoninginAIsystems. -such as instruction fine-tuning, logic-informed pre-training, -reinforcement learning, inference-time decoding strategies, -1 Introduction andhybridneuro-symbolicmethods. Weexaminerecentad- -vancesinneuro-symbolicintegration,alongwithapplications -Logical reasoning is a fundamental challenge to arti- -of theorem provers, logic solvers, and formal verification -ficial intelligence (AI) and natural language processing -frameworksinLLMs. Finally,wehighlightopenchallenges -(NLP) [Newell and Simon, 1956; McCarthy and Hayes, -inscalability, reasoningconsistency, explainability, andeffi- -1981;McCarthy,1959]. Whileearlyformallogic-basedrea- -ciency, proposing future directions for multi-modal reason- -soningapproachesfacedlimitationsinscalabilityandadapt- -ing, hybrid architectures, and improved evaluation frame- -ability [Pereira, 1982; Cann, 1993], data-driven models -works. Thestructureofthesubsequentchaptersisillustrated -became the dominant method since the 1980s [McCarthy, -inFigure1. -1989].Recently,pre-trainedLargeLanguageModels(LLMs) -and their emergent logical reasoning abilities have attracted -2 LogicinArtificialIntelligence -increasing attention [Liu et al., 2023b; Xu et al., 2023]. -Logical reasoning is a cornerstone of artificial intelligence -Logical reasoning integrates LLMs with inference struc- -(AI),enablingmachinestosimulatehumanthoughtprocesses -turing, enabling multistep deduction and abstraction, and -improving interpretability and reliability [Shi et al., 2021; and solve complex problems. At its core, logical reasoning -Staceyetal.,2022;Rajaramanetal.,2023]. Italsostrength- appliesstructuredrulestoderiveconclusionsfrompremises, -providingarigorousframeworkfordecision-makingandin- -ens generalization, helping models handle novel scenarios -beyond their training data [Haruta et al., 2020]. As LLMs -ference[Sunetal.,2023]. -become integral to domains like legal analysis and scien- -2.1 HistoryofLogicReasoningResearch -tific discovery, ensuring the correctness and verifiability of -Logical reasoning can be traced back to ancient Greece, -* Correspondingauthor. where Aristotle’s syllogisms laid the foundation for classi- -† Equalcontribution. cal logic. During the Middle Ages, scholars refined these -5202 -beF -31 -]IA.sc[ -1v00190.2052:viXra - -| | | | | | C o n T R o L | [L iu e ta l ., 2 | 0 2 1 a ] , F O L IO | [ H a n e t a | l ., 2 0 2 4 a ] , L | o g i cNLI[Tianetal.,2021], | | | | | -| --- | --- | --- | ------------------------------ | --- | ---------------- | ----------------- | -------------------- | ----------------- | -------------------- | --------------------------- | --- | --- | --- | --- | -| | | | NaturalLanguageInference(§3.1) | | R u lt eT a k er | [C la rk e t a l. | , 2 0 2 1 ] , L og | iB e nc h [ P a r | m a r e t a l ., 2 0 | 2 3 ] | | | | | -Types&history(§2) -Task&Benchmarks(§3) ReadingComprehension(§3.2) L o g iQ A [L iu e t a l . , 2 0 2 3 a ] , R eC l o r [ Y u e t al . , 20 2 0 ], A R - LS A T [ Y u e t al.,2020],CLUTRR[Sinhaetal.,2019], -| | | | | | G S M [ Co | bb e e t a l . , 2 0 2 | 1 ; L i et al . , | 2 0 2 4a ] L I N G | O L Y [ B ea n | e t al. , 2 02 4 ] | | | | | -| --- | ----------------------- | --- | ----------------------------- | --- | ------------------------------------------------------------------------ | ---------------------- | ----------------- | ------------------ | -------------- | ------------------ | --- | --- | --- | --- | -| | | | Benchmarksandtestsuites(§3.3) | | GLoRE[liuetal.,2023d],LogiGLUE[Luoetal.,2024],LogiTorch[Helweetal.,2022] | | | | | | | | | | -| | | | DeductiveReasoning(§4.1) | | [Saparovetal.,2023],[Yuanetal.,2023],[Rybetal.,2022] | | | | | | | | | | -| | | | InductiveReasoning(§4.2) | | [Yangetal.,2024b],[Bowenetal.,2024],[Sullivan,2024] | | | | | | | | | | -| | Evaluation&Analysis(§4) | | AbductiveReasoning(§4.3) | | TrueDetective[DelandFishel,2023],[Nguyenetal.,2023] | | | | | | | | | | -gninosaeRlacigoLMLL -| | | | AnalogicalReasoning(§4.4) | | ANALOGICAL[Wijesiriwardeneetal.,2023],[PetersenandvanderPlas,2023],[Qinetal.,2024] | | | | | | | | | | -| --- | --- | --- | ----------------------------- | --- | --------------------------------------------------------------------------------------------- | --- | ------------------------------------------------------ | ------------------- | ------------------------------------------------ | ------------------ | ------------------------ | ---------------------------- | --- | --- | -| | | | OverallAnalysis&Metrics(§4.5) | | [Liuetal.,2023b],[Xuetal.,2023],[Liuetal.,2024c],[Gandarelaetal.,2024],[Thatikondaetal.,2025] | | | | | | | | | | -| | | | | | | | F O | L I O [ H a n e t | a l. , 2 0 2 4 a] , | P -F O LI O [ H | an e t a l., 2 0 2 4 | b ] , | | | -| | | | | | Expert-CuratedDatasets | | L ea | n D oj o [ Y a n g | e t a l ., 2 0 23 | ] , S ym b ol- L | L M [ X u e t a l. , | 2 0 24a] | | | -| | | | Data-CentricApproaches(§5.1) | | SyntheticDatasets | | RulteTaker[Clarketal.,2021],FLD×2[Morishitaetal.,2024] | | | | | | | | -| | | | | | | | L og | iC o T [ L i u e | t a l . , 2 0 2 3 c ],LogicPro[Jiangetal.,2024], | | | | | | -| | | | | | LLM-distilledDatasets | | PO | D A [ W a n g e | t a l . , 2 0 2 4 b ] | | | | | | -| | | | | | InstructionFine-Tuning | | L o gi | C o T [ L iu e | t al ., 2 0 2 3 c ] | , L og i P T [ Fe | n g e t a l. , 2 0 2 4 | ] , P G L [Wangetal.,2024a], | | | -| | | | | | | | Sy m | b o l -L L M [ X | u e t a l. , 2 0 | 2 4 a] , T P C L | [ W a n g e t a l ., 2 0 | 2 4 b ] , | | | -| | | | | | | | [Ji a | o e t a l . , 2 0 2 | 4 ] , [ X i e t a | l. , 2 0 2 4 ] , M | a rc o - o 1 [Z h a o et | a l. , 2 0 2 4 ] , | | | -| | | | | | ReinforcementLearning | | | | [ | 5] | | [D I,2025] | | | -| | | | Model-CentricApproaches(§5.2) | | | | D e e | ps e e k - R 1 - Z | e r o D e e p S ee | k - A I , 2 0 2 | , D e e p se ek - R 1 | e e p S e e k -A | | | -EnhancementMethods(§5) -GoT[Leietal.,2023],ChainofLogic[Servantezetal.,2024], -| | | | | | | | S e le | c ti o n- In fe re | n c e [ C r e s w e | ll e t a l ., 2 0 2 | 3 ] , [ M a l o n et a | l. , 2 0 2 4 ] , | | | -| --- | --- | --- | --- | --- | --- | --- | ------ | ------------------ | ------------------- | ------------------- | ---------------------- | ---------------- | --- | --- | -Discussion(§6) Inference-TimeDecoding M a ie u t ic P ro m p ti n g [ J u n g e t a l ., 2 0 2 2 ] , L o g i c- o f - tho ug h t [ L i u e t al.,2024a], -DetermLR[Sunetal.,2024],Neurologic[Luetal.,2021], -Formal-LLM[Lietal.,2024b] -| | | | | | [ Z a y ya d an | d A d i , 2 0 24 ] | , L eanDojo[Yangetal.,2023],LQOT[Liuetal.,2024b],[Ouyangetal.,2023], | | | | | | | | -| --- | --- | --- | ---------------------------------- | --- | --------------- | ---------------------- | -------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -| | | | ExternalKnowledgeUtilization(§5.3) | | K n o w R A | [M a i e t a l ., 2 02 | 5 ] | | | | | | | | -LINC[Olaussonetal.,2023],LOGICLLAMA[Yangetal.,2024a], -| | | | Neuro-SymbolicApproaches(§5.4) | | CLOVER[Ryuetal.,2024],LOGIC-LM[Panetal.,2023],LogicAgent[Liuetal.,2024a], | | | | | | | | | | -| --- | --- | --- | ------------------------------ | --- | ------------------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LLM-TRes[Toroghietal.,2024],SymbCoT[Xuetal.,2024c],Aristotle[Xuetal.,2024b] -Figure1:Thestructureofthissurvey -theories, and in the 17th century, Leibniz’s universal lan- soningisvalid, theconclusionmustalsobetrue. Forexam- -guageandcalculusratiocinatorbridgedlogicwithmathemat- ple,giventhepremises“Allapplesarered”and“Thisfruitis -ics, foreshadowing modern computational logic. The 19th an apple” one can deduce that “This fruit is red” Deductive -century saw George Boole’s Boolean algebra, which trans- reasoning is fundamental in fields such as mathematics and -formedlogicintoamathematicalframework,layingthefoun- formallogic,wherecertaintyandrigorareparamount. -dationfordigitalcomputing. -| | | | | | | Inductive | | Reasoning. | | Unlike | deductive | | reasoning, | induc- | -| -------- | --------------- | --- | ------ | ------ | ------------ | --------- | --------- | ---------- | ----- | ------- | ----------- | --- | ---------- | -------- | -| The 20th | century ushered | in | modern | logic, | with Russell | | | | | | | | | | -| | | | | | | tive | reasoning | | draws | general | conclusions | | based on | specific | -andWhitehead’sPrincipiaMathematicaformalizingcomplex -| | | | | | | observations | | or | evidence. | | While the | conclusions | are | often | -| ------------------ | ------ | ------------- | --- | -------------- | --------- | ------------------- | --- | --- | --------- | ----------------------------- | --------- | ----------- | --- | ----- | -| logical systems. | By the | mid-century, | AI | pioneers | like John | | | | | | | | | | -| | | | | | | consideredprobable, | | | | theyarenotguaranteedtobetrue. | | | | For | -| McCarthy leveraged | logic | for knowledge | | representation | and | | | | | | | | | | -instance,observingthatallswansseensofararewhitemight -| automated | theorem proving, | leading | to | logic | programming | | | | | | | | | | -| ------------------ | ------------------------------- | ------- | --- | ----- | ----------- | ---- | ------ | --------- | --- | ---------- | ---- | ---- | --------- | ------ | -| | | | | | | lead | to the | inductive | | conclusion | that | “All | swans are | white” | -| andknowledgebases. | The1970sintroducednon-monotonic | | | | | | | | | | | | | | -Inductivereasoningiswidelyusedinscientificdiscoveryand -| logic, enabling | AI to handle | commonsense | | reasoning. | The | | | | | | | | | | -| --------------- | ------------ | ----------- | --- | ---------- | --- | ----------- | --- | ---------------- | --- | --- | ----- | -------- | ---------- | --- | -| | | | | | | data-driven | | decision-making, | | | where | patterns | and trends | are | -1980ssawlogicalreasoningintegratewithknowledgerepre- -inferredfromempiricaldata. -| sentation, advancing | expert | systems | for | real-world | applica- | | | | | | | | | | -| -------------------- | ------ | ------- | --- | ---------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tions. The1990ssawtheriseofknowledgegraphs,structur- -| | | | | | | Abductive | | Reasoning. | | This | form | of reasoning | seeks | the | -| --- | --- | --- | --- | --- | --- | --------- | --- | ---------- | --- | ---- | ---- | ------------ | ----- | --- | -ingvastknowledgeforcomplexreasoningtasks. -mostplausibleexplanationorcauseforasetofobservations, -In the 21st century, neuro-symbolic approaches have often in the presence of incomplete information. Abductive -mergeddeeplearningwithlogicalinference,resultingintools reasoning is particularly useful in diagnostic tasks and real- -like DeepLogic [Cingillioglu and Russo, 2019] and SAT- world problem-solving. For example, seeing wet spots on -Net [Wang et al., 2019]. Logical reasoning remains a cor- thestreetmightleadonetoinferthat“Ithasrecentlyrained” -nerstoneofAIresearch,evolvingfromphilosophytomodern While abductive conclusions are not certain, they provide a -computing. As AI advances, logical reasoning continues to practicalbasisforhypothesisgenerationanddecision-making -| shape intelligent | systems, | ensuring | structured, | | interpretable, | underuncertainty. | | | | | | | | | -| ----------------- | -------- | -------- | ----------- | --- | -------------- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | -androbustdecision-making. -| | | | | | | Analogical | | Reasoning. | | | Analogical | reasoning | | involves | -| --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | --- | --- | ---------- | --------- | --- | -------- | -2.2 TypesofLogicalReasoning drawing comparisons between similar situations or domains -| | | | | | | to | make | inferences | or | solve | problems. | By | identifying | par- | -| --- | --- | --- | --- | --- | --- | --- | ---- | ---------- | --- | ----- | --------- | --- | ----------- | ---- | -Logicalreasoningcanbebroadlycategorizedintofourmain -| | | | | | | allelsbetweendifferentscenarios, | | | | | thistypeofreasoningen- | | | | -| --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | -types,eachservingdistinctpurposesandapplications: -| | | | | | | ables | creative | problem-solving | | | and | knowledge | transfer. | For | -| --- | --- | --- | --- | --- | --- | ----- | -------- | --------------- | --- | --- | --- | --------- | --------- | --- | -DeductiveReasoning. Thistypeofreasoningderivesspe- example,understandingthatplanetsorbitthesuninelliptical -cificconclusionsfromgeneralprinciplesorpremises. Itop- pathsmightleadonetoanalogicallyreasonthatotherceles- -eratesundertherulethatifallpremisesaretrueandtherea- tialbodies, suchascomets, exhibitsimilarorbitalcharacter- - -Dataset Language QuestionType Size Source Neutral labels. It isolates FOL-based inference from com- -LogiQA Zh/En Multichoice 15,937 Exam-based monsensereasoning,enablingpreciseevaluationofreasoning -| ReClor | En | Multichoice | | | 6,138 | Exam-based | | | | | | | | -| ------ | --- | ----------- | --- | --- | ----- | ---------- | --- | --- | --- | --- | --- | --- | --- | -AR-LSAT En Multichoice 2,064 Exam-based accuracyandgeneralization. -| CLUTRR | En | Question-answering | | | 6,016 | Rule-based | | | | | | | | -| ------- | --- | ------------------ | --- | --- | ----- | --------------- | ------------ | --- | ------------- | --- | ----------- | ------- | -------- | -| | | | | | | | ProofWriter | | [Tafjord | et | al., 2021] | extends | Rule- | -| GSM | En | Mathwordproblems | | | 19K | Exam-based | | | | | | | | -| | | | | | | | Taker [Clark | | et al., 2021] | by | introducing | CWA | (closed- | -| LINGOLY | En | Question-answering | | | 1,133 | Expert-designed | | | | | | | | -ConTRoL En ternaryclassification 8,325 Exam-based world assumption) and OWA (open-world assumption) to -FOLIO En binaryclassification 1,351 Expert-designed handlenegationandopen-worldreasoning. ItincludesBirds- -| LogicNLI | En | ternaryclassification | | | 30K | Exam-based | | | | | | | | -| -------- | --- | --------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -ProofWriter En binaryclassification - Exam-based Electricity (handcrafted domain theories) and ParaRules -LogicBench En binaryclassification 1,270 Rule-based (crowdsourced paraphrased rules) for systematic evaluation -| GLoRE | Zh/En | Miscellaneous | | 17tasks | | Miscellaneous | | | | | | | | -| -------- | ----- | ------------- | --- | ------- | --- | ------------- | ----------------- | --- | ------ | ---------- | ---------- | --- | ---------- | -| | | | | | | | of generalization | | across | linguistic | variations | and | real-world | -| LogiGLUE | En | Miscellaneous | | 24tasks | | Miscellaneous | | | | | | | | -LogiTorch En Miscellaneous 16tasks Miscellaneous knowledgedomains. -BIG-Bench En Miscellaneous 7tasks Miscellaneous [Parmar 2023] -| | | | | | | | LogicBench | | | et al., | is | a GPT-3-generated | | -| --- | --- | --- | --- | --- | --- | --- | ---------- | -------- | -------- | ------- | ---------- | ----------------- | -------- | -| | | | | | | | dataset | covering | 25 types | of | reasoning, | including | proposi- | -Table1:MainDatasetsandBenchmarksofLogicalReasoningTask. -| | | | | | | | tional logic, | FOL, | and | non-monotonic | logic. | It | consists of | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ---- | --- | ------------- | ------ | --- | ----------- | -istics. Analogicalreasoningisparticularlyvaluableinfields -1,270testentrieslabeledasYesorNo. -likeeducation,design,andinnovation. -3.2 MachineReadingComprehension(MRC) -3 TasksandBenchmarks -MRCevaluateslogicalreasoningbyrequiringmodelstoan- -| Logical reasoning | | datasets | and | benchmarks | are | essential for | | | | | | | | -| ----------------- | --- | -------- | --- | ---------- | --- | ------------- | -------------- | --- | -------- | ------- | -------- | ----- | -------- | -| | | | | | | | swer questions | | based on | a given | passage. | Tasks | are com- | -evaluatingthereasoningcapabilitiesoflargelanguagemod- -| | | | | | | | monlyformattedasmultiple-choice, | | | | spanextraction, | | orfree | -| ----------------------------- | ----- | -------- | --- | -------------- | --- | ---------- | -------------------------------- | ---- | --------------- | --- | --------------- | ------------ | ------ | -| els (LLMs). | These | datasets | can | be categorized | | into three | | | | | | | | -| | | | | | | | response, | with | multiple-choice | | QA being | particularly | effec- | -| typesbasedontheirdatasources: | | | | | | | tiveduetoitsstandardization. | | | | | | | -Rule-based Datasets [Tafjord et al., 2021; Sinha et al., [Liu 2023a] -| | | | | | | | LogiQA | | et al., | | is sourced | from | the Chinese | -| --------- | ------------- | --- | --------- | ----- | ------- | ---------- | ------ | --- | ------- | --- | ---------- | ---- | ----------- | -| 2019] are | automatically | | generated | using | logical | rules, en- | | | | | | | | -CivilServiceExam,containing15,937entriesinChineseand -| abling large-scale | | data collection. | | However, | ensuring | diver- | | | | | | | | -| ------------------ | --- | ------------------- | --- | -------- | -------- | ---------- | -------- | ---------- | ------- | ------- | --------- | --- | --------- | -| | | | | | | | English. | It targets | complex | logical | reasoning | and | is widely | -| sity is crucial | | to avoid repetitive | | patterns | and | comprehen- | | | | | | | | -usedforevaluatingLLMs. -sivelyevaluatereasoningcapabilities. ReClor[Yuetal.,2020],derivedfromtheGMAT,features -| Expert-Designed | | Datasets | [Han | et | al., 2024a] | are con- | | | | | | | | -| --------------- | --- | -------- | ---- | --- | ----------- | -------- | --- | --- | --- | --- | --- | --- | --- | -6,138Englishentrieswithfour-optionmultiple-choiceques- -| structed by | domain | experts, | ensuring | high | precision | and ac- | | | | | | | | -| ----------- | ------ | -------- | -------- | ---- | --------- | ------- | --- | --- | --- | --- | --- | --- | --- | -tions. -| curacy. Although | | typically | smaller | than | crowd-sourced | cor- | | | | | | | | -| ---------------- | ---------- | --------- | ------- | ---- | ------------- | ---- | ---------- | ----- | ------- | ---------- | -------- | ------ | --------- | -| | | | | | | | AR-LSAT | [Wang | et | al., 2022] | is based | on | the LSAT, | -| pora, their | meticulous | design | makes | them | indispensable | for | | | | | | | | -| | | | | | | | containing | 2,064 | entries | spanning | ordering | games, | grouping | -in-depthlogicalreasoningevaluation. -games,andallocationgames,eachwithfiveoptions. -Exam-BasedDatasets[Liuetal.,2021b;Yuetal.,2020; -CLUTRR[Sinhaetal.,2019]focusesoninductivereason- -Wangetal.,2022]originatefromstandardizedtestquestions -| | | | | | | | ing, requiring | | models to | infer | kinship relationships | | in short | -| ---------------------- | --- | -------------- | ------- | ----- | -------- | --------- | -------------- | --------------------------------------------- | --------- | ----- | --------------------- | --- | -------- | -| (e.g., Chinese | | National Civil | Service | Exam, | LSAT, | GRE), | | | | | | | | -| | | | | | | | narratives. | Itcontains6,016entries,combiningentityextrac- | | | | | | -| offering high-quality, | | expert-crafted | | logic | problems | at scale. | | | | | | | | -tionandlogicalinference. -Thesedatasetsarewidelyusedtoevaluatereasoninginreal- -GSMevaluatesmathematicalreasoningcapabilities,com- -worldscenarios. -| | | | | | | | prising | two datasets: | | GSM8K | [Cobbe | et al., | 2021] (8.5K | -| --- | --- | --- | --- | --- | --- | --- | ------- | ------------- | --- | ----- | ------ | ------- | ----------- | -Table1summarizesimportantdatasetsforlogicalreason- -| | | | | | | | grade school | math | problems) | | and GSM-PLUS | | [Li et al., | -| ---------- | --------- | ----- | ----- | ------- | ------- | -------- | ------------ | ---- | --------- | --- | ------------ | --- | ----------- | -| ing, which | typically | cover | tasks | such as | Natural | Language | | | | | | | | -2024a](10,552),whichisaugmentedwithmathematicalper- -Inference(NLI)(§3.1)andMachineReadingComprehension -turbationsforrobustnessevaluation. -(MRC)(§3.2). -| | | | | | | | LINGOLY | | [Bean et | al., 2024] | uses Linguistic | | Olympiad | -| --- | --- | --- | --- | --- | --- | --- | ------- | --- | -------- | ---------- | --------------- | --- | -------- | -3.1 NaturalLanguageInference(NLI) puzzlestoevaluatein-contextpatternidentificationandgen- -| | | | | | | | eralization | in low-resource | | or | extinct languages. | | It contains | -| ------------- | --- | --------- | ---------- | --------- | --- | ------------ | ----------- | --------------- | --- | --- | ------------------ | --- | ----------- | -| NLI evaluates | | whether a | hypothesis | logically | | follows from | | | | | | | | -1,133problemsacross6formatsand5difficultylevels,cov- -| apremise,directlyassessingamodel’sreasoningability. | | | | | | La- | | | | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -eringover90languages. -belstypicallyfallintobinary(Entailment,Non-entailment)or -| ternary (Entailment, | | Contradiction, | | Neutral) | classifications. | | | | | | | | | -| -------------------- | --- | -------------- | --- | -------- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | -3.3 BenchmarkSuites -SomedatasetsuseTrueandFalselabelsinstead. -ConTRoL[Liuetal., 2021a]isderivedfromrecruitment Benchmarksuitesstandardizeevaluationandfacilitatemodel -exams (e.g., bank entry, U.S. police selection), containing comparisoninlogicalreasoningresearch. -8,325 entries with Correct, Incorrect, and Can’t Say labels, GLoRE[liuetal.,2023d]isafew-shotandzero-shottest- -correspondingtoEntailment,Contradiction,andNeutral. ingplatform,including17test-onlydatasetstoassessgener- -FOLIO[Hanetal.,2024a]isanexpert-constructeddataset alizationinlow-datascenarios. -for First-Order Logic (FOL) reasoning, consisting of 1,351 LogiGLUE [Luo et al., 2024] consists of 24 logical -entrieslabeledasTrueorFalse,makingitarigorousbench- reasoning tasks, standardizing datasets into a sequence-to- -markforformallogicalinference. sequence format for uniform input processing. It provides -| | [Tian | | 2021] | | | | | | | | | | | -| --- | ----- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -LogicNLI et al., contains 30K entries gener- bothtestandtrainingsets,enablingextensivemodeltraining -atedusinglogicalrules,withEntailment,Contradiction,and andtargetedevaluations. - -LogiTorch[Helweetal.,2022]isaPyTorch-basedlibrary ical sub-proofs without examples, generalization, and sensi- -for natural language logical reasoning, offering 16 datasets, tivitytosyntacticvariations[Saparovetal.,2023;Yuanetal., -2023;Rybetal.,2022]. -modelarchitectures,andanaccessibleAPIforquickevalua- -tion. -BIG-bench [Srivastava et al., 2022] is a collaborative 4.2 InductiveReasoning -| benchmarkwith7tasksdedicatedtologicalreasoning, | | | | such | | | | | | | | | -| ----------------------------------------------- | --- | --- | --- | ---- | --------- | ---------- | --- | ----- | ----------- | --- | ---- | ------------ | -| | | | | | Inductive | reasoning, | | which | generalizes | | from | specific in- | -asLogicGridPuzzleandLogicalFallacyDetection. -stancestobroaderrules,isessentialfortaskslikehypothesis -generationandpatternrecognition.WhileYangetal.[2024b] -| | | | | | find that | pre-trained | models | | can serve | as | effective | “reason- | -| --- | --- | --- | --- | --- | ----------------- | ----------- | ---------- | --------- | --------- | ------------ | --------- | ----------- | -| | | | | | ers,” Bowen | et | al. [2024] | show | that | even | advanced | LLMs | -| | | | | | struggle | with | simple | inductive | tasks | in their | symbolic | set- | -| | | | | | tings. Similarly, | | Sullivan | [2024] | | demonstrates | | that Trans- | -formermodels,evenafterfine-tuning,failtolearnfundamen- -| | | | | | tal logical | principles, | | indicating | limited | inductive | | reasoning | -| --- | --- | --- | --- | --- | ----------- | ----------- | --- | ---------- | ------- | --------- | --- | --------- | -capabilities. -| | | | | | 4.3 AbductiveReasoning | | | | | | | | -| --- | --- | --- | --- | --- | ---------------------- | ---------- | --- | ----- | ----- | -------- | --------- | ------ | -| | | | | | Abductive | reasoning, | | which | seeks | the most | plausible | expla- | -nationsforobservedphenomena,iscrucialinfieldslikelaw -| | | | | | and medicine. | | Del and | Fishel | [2023] | highlights | | the chal- | -| --- | --- | --- | --- | --- | ------------- | ------------ | ------- | ---------- | --------- | ------------ | ---------- | --------- | -| | | | | | lenges LLMs | | face in | generating | plausible | | hypotheses | from | -| | | | | | incomplete | information. | | In | the legal | domain, | | Nguyen et | -| | | | | | al. [2023] | show | that | despite | strong | performance, | | models | -strugglewithabductivereasoning,underscoringthecomplex- -ityofthisparadigm. -| | | | | | 4.4 AnalogicalReasoning | | | | | | | | -| --- | --- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | -Analogicalreasoning,whichinfersunknowninformationby -| | | | | | comparing | it with | known | information, | | is | vital for | tasks re- | -| --- | --- | --- | --- | --- | --------- | ------- | ----- | ------------ | --- | --- | --------- | --------- | -quiringcreativityandknowledgetransfer.Wijesiriwardeneet -| (a) A multi-choice | reading | comprehension | example from | the | | | | | | | | | -| ------------------ | ------- | ------------- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -al.[2023]introducedANALOGICAL,abenchmarkforlong- -LogiQAdataset. -textanalogicalreasoning.Theyfindthatasanalogycomplex- -| | | | | | ity increases, | | LLMs struggle | | to recognize | | analogical | pairs. | -| --- | --- | --- | --- | --- | -------------- | --- | ------------- | --- | ------------ | --- | ---------- | ------ | -PetersenandvanderPlas[2023]showthatmodelscanlearn -analogicalreasoningwithminimaldata,approachinghuman -| | | | | | performance. | | However, | Qin | et al. | [2024] | question | whether | -| --- | --- | --- | --- | --- | ------------ | -------- | ------------- | --- | ---------- | ------- | ----------- | ------- | -| | | | | | LLMs truly | rely | on analogical | | reasoning, | | discovering | that | -| | | | | | random | examples | in prompts | | often | achieve | comparable | per- | -formancetorelevantexamples. -| | | | | | 4.5 OverallAnalysisandMetrics | | | | | | | | -| --- | --- | --- | --- | --- | ----------------------------- | ------- | -------- | --- | ----- | ----------- | --- | --------- | -| | | | | | Liu et al. | [2023b] | evaluate | | GPT-4 | and ChatGPT | | on bench- | -(b) AnNLIexamplefromtheConTRoLdataset. marks like LogiQA and ReClor, showing that while GPT- -| | | | | | 4 outperforms | | ChatGPT, | both | of them | struggle | | with out-of- | -| --- | --- | --- | --- | --- | ------------- | --- | -------- | ---- | ------- | -------- | --- | ------------ | -Figure2:ExampletestsofLogicalreasoninginNLPtasks. -| | | | | | distribution | tasks. | Xu | et al. | [2023] | introduce | | the NeuLR | -| --- | --- | --- | --- | --- | ------------ | ------ | --- | ------ | ------ | --------- | --- | --------- | -4 Evaluations -datasetandproposeaframeworkevaluatingLLMsacrosssix -| | | | | | dimensions: | correctness, | | rigor, | self-awareness, | | | proactivity, | -| --------------------- | --- | -------------- | -------- | ------ | ----------- | ------------ | --- | ------ | --------------- | --- | --- | ------------ | -| The rapid development | | of pre-trained | language | models | | | | | | | | | -guidance,andabsenceofhallucinations. -| (PLMs) necessitates | rigorous | evaluation | of their logical | rea- | | | | | | | | | -| ------------------- | -------- | ---------- | ---------------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -soning capabilities. This section examines four reasoning Metrics for Evaluating Logical Reasoning. Traditional -paradigms—deductive, inductive, abductive, and analogi- metrics like accuracy and F1 score are insufficient for as- -cal—whileanalyzingevaluationapproachesandmetrics. sessing logical reasoning. Recent studies have introduced -| | | | | | nuanced | metrics | such | as consistency | | (invariance | | to logi- | -| --- | --- | --- | --- | --- | ------- | ------- | ---- | -------------- | --- | ----------- | --- | -------- | -4.1 DeductiveReasoning -callyequivalentinputs),generalization(performanceonout- -Deductivereasoning,derivingspecificconclusionsfromgen- of-distribution data), and explainability (clarity of reason- -eralpremises,iscrucialforautomatedtheoremproving. De- ing steps). Thatikonda et al. [2025] find that combining -spite LLMs performing well on tasks like compositional BERTScorewithtraditionalmetricsimprovesalignmentwith -[2024c] -proofs, standard benchmarks, and encoding entailment rela- human judgments. Liu et al. propose a framework -tionships, they struggle with extended reasoning, hypothet- for measuring logical consistency, showing that BERTScore - -aligns better with human rankings than LLM-based evalua- 5.2 Model-CentricApproaches -| torslikeGPT-4. | | Gandarelaetal.[2024]emphasizestheneed | | | | | | | | | | | | | -| -------------- | --- | ------------------------------------- | --- | --- | --- | --- | --- | ------------- | ---------- | ------- | ----- | --------- | --- | ----- | -| | | | | | | | | Model-Centric | approaches | enhance | LLMs’ | reasoning | | capa- | -for metrics that account for the expressivity of logical theo- bilitiesbyoptimizingmodelparametersanddecodingstrate- -ries,particularlyininductivereasoning. gies. Theformalobjectiveis: -| 5 EnhancementMethods | | | | | | | | | (θ∗,S∗)=argmaxR(M | | | | | | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | ----- | --- | --- | -| | | | | | | | | | | | | θ ,S) | | (2) | -θ,S -| Enhancing | LLMs’ | logical | reasoning | | remains | crucial. | This | | | | | | | | -| --------- | ----- | ------- | --------- | --- | ------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | -where: -| sectionfocusesoncorestrategies: | | | | Data-CentricApproaches | | | | | | | | | | | -| ------------------------------- | --- | --- | --- | ---------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(§5.1), Model-Centric Approaches (§5.2), External Knowl- • θ: learnablemodelparameters. -edge Utilization (§5.3), and Neuro-Symbolic Reasoning M modelwithparametersθ. -• θ : -(§5.4). -• S: decodingstrategy(e.g.,chain-of-thoughtprompting, -| 5.1 Data-CentricApproaches | | | | | | | | verification-baseddecoding). | | | | | | | -| -------------------------- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --- | --- | --- | -Data-centricapproachesenhanceLLMs’reasoningcapabili- • R: reasoningperformancemetric. -Thisformulationhighlightsthejointoptimizationofmodel -| ties by utilizing | | meticulously | | curated | training | datasets. | For- | | | | | | | | -| ----------------- | --- | ------------ | --- | ------- | -------- | --------- | ---- | --- | --- | --- | --- | --- | --- | --- | -mally,thiscanbeexpressedas: parametersθanddecodingstrategyS. Practicalimplementa- -tionscanbecategorizedas: -| | | D∗ =argmaxR(M | | | ) | | (1) | | | | | | | | -| --- | --- | ------------- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | ------------ | --- | --- | --- | -| | | | | | D | | | • InstructionFine-Tuning: | | | optimizingθ. | | | | +Issue Brief + +MTV/AP-NORC: COMPARING THE +POLITICAL VIEWS OF YOUNG +PEOPLE AND THEIR +PARENTS’ +GENERATION + +Young people and their parents share a negative outlook on +the current political landscape in America and are largely +dissatisfied with the government’s handling of a wide +variety of key issues, according to a new poll of Americans +age 15-34, and parents of children age 15-26 by MTV and The +Associated Press-NORC Center for Public Affairs Research. + +Young people’s opinions about their generation’s ability to +impact the government have shifted in recent months, +following student-driven activism. Between the first poll +fielded between late February and early March 2018, +shortly after the shooting in Parkland, Florida, and this +second poll conducted roughly two months later, young +people age 15-34, and particularly those age 15-22, feel +more politically empowered. The proportion overall who +believe that elected officials care what they think rose +from a quarter to a third, and there was a similar rise in +the numbers who think people like them can impact the +government. + +The generation of parents of children age 15-26 is less +racially and ethnically diverse and includes a greater +proportion of Republicans than young people age 15-26, +and the survey demographics reflect this. Parents are +more likely to approve of President Donald Trump (42 +percent vs. 26 percent) and are somewhat more optimistic +(59 percent vs. 45 percent) than the younger generation. +Still, the findings show that young people and parents are +on the same page about politics in general. Majorities say +they tend to agree with one another on many key issues +and that their political discussions are generally amicable. + +© 2017 iStock/Nikada + +Three Things You Should Know +About the MTV/AP-NORC Poll of +Americans Age 15-34 and Parents +of 15-26 Year Olds: + +1) + +In the wake of recent protests and +demonstrations, 46 percent of young +people age 15-34 believe that people +like them can affect the government +at least a moderate amount, an +increase from 37 percent in the March +2018 poll. For 15-22 year olds, this rose +15 points, to 48 percent. + +2) Nearly two-thirds of young people +age 15-26 and parents of this age +group discuss politics at least once a +month. Majorities say these +conversations are civil and that they +are at least sometimes able to change +others’ political views when +disagreements do occur. + +3) More than two-thirds across both + +generations are disillusioned with the +two major political parties and feel a +third is needed. + +© Copyright 2018. The Associated Press and NORC + +June 2018 1 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Both young people age 15-26 and parents of this age group give the federal government a grade of F +more often than any other letter grade when assessing its performance on a wide variety of issues, +including gun control, the cost of higher education, foreign relations, immigration, environment and +climate change, health care, income inequality, taxes, sexual harassment and assault, abortion, racism, +and the treatment of citizens by police and the justice system. The economy and job creation fare +better, most commonly receiving grades of C from both groups. + +This survey is the second in a multiwave series of polls highlighting the voices of the youngest +generation of voters age 15-34, with this installment including a sample of parents of 15-26 year olds to +allow for comparisons between generations. This nationwide poll by MTV and The AP-NORC Center +surveyed a total of 939 young people, including 148 teens age 15-17, 410 adults age 18-26, and 381 adults +age 27-34, as well as 607 parents of young people age 15-26. + +AFTER RECENT DEMONSTRATIONS, YOUNG PEOPLE FEEL MORE +EMPOWERED IN POLITICS. + +Since the first MTV/AP-NORC Youth Political Pulse survey in March 2018, overall outlook among the +larger group of young people including those age 15-34 remains stable. Similar to the previous survey, +just under half are optimistic and a slight majority are pessimistic about the future of the United States. +Four in 10 have felt angry or anxious about the state of the country in the past month, while 1 in 10 +have felt excited or positive. + +Young peoples’ overall standpoint is fairly negative, but there is one area where young people have +grown somewhat more optimistic. After many recent protests and demonstrations, more now than in +March say elected officials care at least a moderate amount what people like them think (34 percent vs. +25 percent) and that people like them can affect the government at least a moderate amount (46 +percent vs. 37 percent). + +Since March 2018, more young people feel efficacious in American government. + +t +n +e +c +r +e +P + +100 + +90 + +80 + +70 + +60 + +50 + +40 + +30 + +20 + +10 + +0 + +37 + +25 + +46 + +34 + +March 2018 + +May 2018 + +Elected officials care at least a moderate amount + +People like me can affect government at least a moderate amount + +Questions: How much do elected public officials care what people like you think? +How much can people like you affect what the government does? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people age 15-22 in particular are increasingly optimistic when it comes to political efficacy. In +March, just a third said that people like them can affect what the government does at least a moderate +amount. Now, 48 percent of those age 15-22 say the same. + +© Copyright 2018. The Associated Press and NORC + +June 2018 2 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young peoples’ priorities have also shifted. When asked to describe in their own words what issue +facing the country concerns them the most, 21 percent cited gun laws in March. Just 6 percent mention +gun issues now.1 The most common worries are now the economy (16 percent) and social inequality (14 +percent). + +BOTH YOUNG PEOPLE AND PARENTS SAY POLITICS HAVE BECOME +DYSFUNCTIONAL, AND BOTH ARE DISSATISFIED WITH THE TWO-PARTY +SYSTEM. + +Young people age 15-26 and parents of young people age 15-26 alike have mixed outlooks on the future +of the United States. Young people are more pessimistic than parents (55 percent vs. 40 percent), +although these differences are mainly attributable to the parents being more Republican and less +racially and ethnically diverse than young people. + +Both generations report relatively negative feelings about the state of the country over the past month. +Substantial minorities of both groups have felt anxious or angry in the past month, while few have felt +positive or excited. Young people are more likely than parents to say they feel anxious (43 percent vs. +29 percent) and less likely to feel positive (8 percent vs. 17 percent) or excited (8 percent vs. 15 percent). + +Young people age 15-26 are more anxious and less positive or excited about the state of America than +are parents of young people. + +y + +l +e +m +e +r +t +x +e + +r +o + +y +r +e +v +y +a +s + +o +h +w + +t +n +e +c +r +e +P + +100 + +90 + +80 + +70 + +60 + +50 + +40 + +30 + +20 + +10 + +0 + +43 + +29 + +40 + +36 + +17 + +8 + +15 + +8 + +Anxious + +Angry + +Positive + +Excited + +Young people age 15-26 + +Parents of young people age 15-26 + +Question: In the past month… How _____ have you felt about the state of the country? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people and parents have similar assessments of the political system and their own efficacy +within it. Majorities of both groups say American politics have been dysfunctional over the past +month. Sixty-four percent of young people and 67 percent of parents feel that politics have not been +functioning very well or at all. + +Young people see the current two-party system as inadequate, and parents agree. Less than a third +think the Republican and Democratic parties do an adequate job of representing the American people, +while more than two-thirds say a third party is needed. + +1 Interviews for this survey were conducted April 23-May 9, 2018, prior to the May 18 shooting in Santa Fe, Texas. + +© Copyright 2018. The Associated Press and NORC + +June 2018 3 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +YOUNG PEOPLE AND PARENTS ARE SIMILARLY ENGAGED WITH THE +MIDTERM ELECTIONS. + +When it comes to the upcoming midterm elections in November, parents of young people age 15-26 are +more likely than young people age 15-26 to be regularly reading or watching news about the elections +(66 percent vs. 45 percent). However, older adults tend to consume news more often than young +people in general2 and may be hearing more about the midterms as a result. + +The two generations are equally likely to engage with the November elections in other ways, such as +posting or commenting about them on social media (25 percent of parents vs. 25 percent of young +people age 15-26), sharing political memes (18 percent of parents vs. 24 percent of young people age 15- +26), participating in events (14 percent of parents vs. 16 percent of young people age 15-26), or +volunteering with a candidate or issue campaign (11 percent of parents vs. 9 percent of young people +age 15-26). + +BOTH GENERATIONS HAVE VARIED POLITICAL CONCERNS AND ARE +STRONGLY DISAPPOINTED WITH THE GOVERNMENT’S PERFORMANCE +ON KEY ISSUES. + +Economic issues top long lists of current concerns for both young people and parents. When asked to +say in their own words what concerns them most about the issues currently facing the country, 15 +percent of young people age 15-26 and 17 percent of parents of young people age 15-26 say they are +most concerned about the economy. The two groups place similar priorities on a range of other issues, +although social inequality ranks second highest for young people (13 percent) and fails to make the top +five most worrisome issues for parents (6 percent). + +2 The Media Insight Project. 2014. The Personal News Cycle. +http://www.apnorc.org/PDFs/Media%20Insight/News%20Cycle/The_Personal_News_Cycle_Final.pdf + +© Copyright 2018. The Associated Press and NORC + +June 2018 4 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and parents report a wide range of top concerns of issues facing the country. + +The economy + +15 + +17 + +Social inequality + +War or threat of nuclear war + +13 + +6 + +11 +10 + +Immigration + +Gun laws + +Other issues + +Political climate + +Concerns about President Trump, leadership + +Foreign policy + +The environment + +Health care + +Terrorism, homeland security, national security + +Crime and safety + +Education + +Nothing + +Everything + +7 +7 + +7 + +4 + +6 + +9 + +5 + +8 + +5 + +7 + +5 +6 + +4 +5 + +3 + +6 + +2 + +4 + +2 +3 + +2 +2 + +1 +0 + +0 +1 + +0 + +10 + +20 + +30 + +40 + +50 + +60 + +70 + +80 + +90 + +100 + +Percent who offered each response + +Young people age 15-26 + +Parents of young people age 15-26 + +Question: Of all of the issues facing the country right now, what concerns you the most? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +Young people age 15-26 believe the federal government is doing a poor job addressing a majority of key +issues. They most commonly give the government a failing grade for 15 of 17 key issues. However, +young people rate performance on the economy and job creation as average, most commonly giving a +grade of C. + +© Copyright 2018. The Associated Press and NORC + +June 2018 5 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Parents most commonly give the same grades as young people age 15-26 on most issues, with a few +exceptions. They give Cs rather than Fs for LGBT rights, the quality of education, terrorism and +homeland security, and taxes. + +Young people age 15-26 and parents give low marks to the federal government on its handling of many +issues. + +Young People +Age 15-26 + +Parents of Young +People Age 15-26 + +The environment and climate change + +Gun control + +Racism + +Immigration + +Cost of higher education + +Treatment of citizens by police and justice system + +Abortion + +Income inequality + +Sexual harassment and assault + +The quality of education + +Taxes + +Foreign relations + +Health care policy + +The rights of LGBT Americans + +Terrorism and homeland security + +The economy + +Job creation + +A + +B + +C + D -| where: | | | | | | | | • ReinforcementLearning: | | | optimizingθ. | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | ------------ | --- | --- | --- | -• D: trainingdatasets. • Inference-TimeDecoding: optimizingS. -• M : modeltrainedonD. Model-Centric approaches focus on directly improving + +F A + +B + +C + D -| | | | | | | | | the model’s | reasoning | capabilities | by optimizing | | its internal | | -| ---- | ----------- | --------- | --- | ------ | --------------- | --- | ----- | ----------- | ------------ | ------------ | ------------- | --- | ------------ | --- | -| • R: | performance | evaluator | | (e.g., | LLM-as-a-judge, | | rule- | | | | | | | | -| | | | | | | | | mechanisms | and decoding | strategies, | making | | them comple- | | -basedmetrics). -mentarytodata-centricapproaches. -| This formulation | | highlights | | the central | role | of | dataset op- | | | | | | | | -| ---------------- | --- | ---------- | --- | ----------- | ---- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | -InstructionFine-Tuning -| timization | in data-centric | | approaches. | | In | practice, | data- | | | | | | | | -| --------------- | --------------- | --------- | ----------- | ----- | ----- | --------- | --------- | ----------------------------------------- | ----------- | ----- | ----------- | -------------- | --- | ------ | -| | | | | | | | | Instruction | Fine-Tuning | (IFT) | adapts LLMs | through | | super- | -| centric methods | | typically | involve | three | types | of | datasets: | | | | | | | | -| | | | | | | | | visedlearningontask-specificinstructions. | | | | Forexample,Liu | | | -expert-curateddatasets,syntheticdatasets,andLLM-distilled -etal.[2023c]designmulti-grainedinstructionsspanningdi- -datasets. -| | | | | | | | | verse levels | of abstraction | and | complexity. | Similarly, | | Feng | -| -------------- | --- | ----------- | --- | --------- | ------ | ------------ | ------- | ------------- | -------------- | --------- | ----------- | ---------- | ------------ | ------ | -| Expert-Curated | | Datasets. | | The FOLIO | | series [Han | et al., | [2024] | | | | | | | -| | | | | | | | | et al. | IFT models | to mimic | logical | solvers | by | repli- | -| 2024a; Han | et | al., 2024b] | | establish | formal | verification | | | | | | | | | -| | | | | | | | | cating formal | deduction | reasoning | processes. | | In addition, | | -throughFOLannotations,withP-FOLIOextendingthecom- Xuetal.[2024a]implementtwo-stagesymbolicfine-tuning -plexity of reasoning chains for enhanced training. Lean- through Injection (injecting symbolic knowledge) and Infu- -Dojo[Yangetal.,2023]provides98k+human-proventheo- sion(balancingsymbolandNLreasoning). -| rempairsformathematicalreasoning. | | | | | Additionally,Symbol- | | | | | | | | | et | -| --------------------------------- | --- | --- | --- | --- | -------------------- | --- | --- | ----------- | ----- | ------------ | ------------ | --- | ---- | --- | -| | | | | | | | | To overcome | IFT’s | over-fitting | limitations, | | Wang | | -LLM[Xuetal.,2024a]systematicallyorganizes34symbolic -| | | | | | | | | al. [2024b] | enforce | contrastive | learning | | between | fac- | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | ------- | ----------- | -------- | --- | ------- | ---- | -reasoning tasks to capture inter-symbol relationships across tual/counterfactual paths with IFT. Further, Wang et -20distinctsymbolicfamilies. al. [2024a] augment Llamas with a Program-Guided Learn- -ingFrameworkandlogic-specificarchitectureadjustments. -| Synthetic | Datasets. | | Rule-based | | synthetic | data | remains | | | | | | | | -| --------- | --------- | --- | ---------- | --- | --------- | ---- | ------- | --- | --- | --- | --- | --- | --- | --- | -Recently,Muennighoffetal.[2025]proposes1,achieving -| fundamental | for | data generation. | | RuleTaker | | [Clark | et al., | | | | | | | | -| ----------- | --- | ---------------- | --- | --------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | -2021]formalizesthisthroughathree-phasepipeline: behav- test-time scaling through IFT on 1,000 meticulously crafted -ior formalization, example synthesis and linguistic equiva- long CoT samples. Combined with budget-forcing tech- -lents generation. Similarly, Morishita et al. [2024] devel- nique, itsignificantlyenhances thereasoningcapabilityofa -Qwen2.5-32B-Instructmodel,allowingextrapolatingbeyond -| ops Formal | Logic | Deduction | | Diverse | (FLD | ×2 ), | a synthetic | | | | | | | | -| ---------- | ----- | --------- | --- | ------- | ---- | ----- | ----------- | --- | --- | --- | --- | --- | --- | --- | -itsperformancewithouttest-timeintervention. -datasetbasedonsymbolictheoryandpreviousempiricalin- -sights. -ReinforcementLearning -LLM-Distilled Datasets. Researchers employ advanced Reinforcementlearning(RL)hasbecomepivotalinoptimiz- -models such as GPT-4 for intermediate reasoning step dis- ing large language models (LLMs), particularly since the -tillation. LogiCoT [Liu et al., 2023c] augments existing breakthroughofReinforcementLearningfromHumanFeed- -datasets with GPT4-generated reasoning chains, while Log- back (RLHF). Jiao et al. [2024] leverage RL for planning- -icPro[Jiangetal.,2024]combinesalgorithmicproblemswith basedreasoningoptimization,whileXietal.[2024]develop -code solutions to create variable-guided reasoning data. To R3,achievingprocesssupervisionbenefitsthroughoutcome- -| advance, | Wang | et al. [2024b] | | propose | PODA, | which | gener- | onlysupervision. | | | | | | | -| -------- | ---- | -------------- | --- | ------- | ----- | ----- | ------ | ---------------- | --- | --- | --- | --- | --- | --- | -atescontrastiveanalysesofcorrect/incorrectoptionsthrough The success of large-scale RL in OpenAI-o1 [OpenAI, -2024]hasinspirednumerousstudies.RLalgorithmstraino1- -premise-orientedaugmentation,enablingreasoningpathdif- -ferentiationviacontrastivelearning. stylemodelstoenhanceChain-of-Thought(CoT)reasoning, - -| addressing | issues | like | formulaic | outputs | | and limited | long- | | | | | | | | | -| ---------- | ------ | ---- | --------- | ------- | --- | ----------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | -form reasoning. For instance, Zhao et al. [2024] integrate (M∗,K∗)=argmaxR(M,K) (3) -M,K -| CoT instruction | | fine-tuning | | with Monte | Carlo | Tree | Search | | | | | | | | | -| --------------- | -------- | ----------- | ---------- | ---------- | ----- | ------------ | ------ | ------ | --- | --- | --- | --- | --- | --- | --- | -| (MCTS) | decoding | for | multi-path | reasoning | | exploration. | In | where: | | | | | | | | -contrast,Zhangetal.[2024]employMCTStogeneratecode- -| | | | | | | | | • M: | the neural | model, | which | includes | both | the | model’s | -| --------- | ---- | --------------- | --- | ----------- | --- | ----- | ---------- | ---------- | ---------- | ------ | ------------ | ---------- | ---- | ----------- | ------- | -| reasoning | data | for instruction | | fine-tuning | | (IFT) | and Direct | | | | | | | | | -| | | | | | | | | parameters | | and | its decoding | strategies | | (generally, | the | -PreferenceOptimization(DPO). -model’sparametersremainunchanged). -| A significant | | breakthrough | | comes | | from | DeepSeek- | | | | | | | | | -| ------------- | --- | ------------ | --- | ----- | --- | ---- | --------- | ---- | --------- | --- | ----------- | --------- | --------- | --- | ------ | -| | | | | | | | | • K: | knowledge | | integration | strategy, | including | | knowl- | -R1[DeepSeek-AI,2025],whichpioneersanovelRLstrategy -| | | | | | | | | edge | source | curation, | structured | | knowledge | representa- | | -| ---------- | ------- | ---------- | --- | ----------------- | --- | --- | ------- | ---- | ------ | --------- | ---------- | --- | --------- | ----------- | --- | -| to enhance | logical | reasoning. | | DeepSeek-R1-Zero, | | | trained | | | | | | | | | -tion,retrieval-augmentedmechanisms,etc. -purelythroughRLwithoutIFT,demonstratesimpressiverea- -R: -soning capabilities but faces challenges in readability and • reasoningperformanceevaluator(e.g.,factualaccu- -language consistency. To address this, DeepSeek-R1 intro- racy,logicalconsistency). -ducesminimallong-CoTIFTdataasacoldstartbeforeRL, -ZayyadandAdi[2024]andYangetal.[2023]extractdata -| achieving | a balance | between | | usability | and | reasoning | per- | | | | | | | | | -| --------- | --------- | ----------- | ------------ | --------- | ------------ | --------- | ------- | ---------- | -------------- | --- | ----- | ----- | ------ | ------- | ----- | -| | | | | | | | | from Lean, | a mathematical | | proof | tool, | to aid | theorem | prov- | -| formance. | By | iteratively | synthesizing | | high-quality | | reason- | | | | | | | | | -ing. Incontrast,“Logic-Query-of-Thoughts”(LQOT)[Liuet -ing data through RL, DeepSeek-R1 overcomes limitations al.,2024b]decomposescomplexlogicalproblemsintoeasier -| imposed | by human | annotators, | | addressing | | issues | such as | | | | | | | | | -| ------- | -------- | ----------- | --- | ---------- | --- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -sub-questionsbeforeintegratingknowledgegraphs. -| mechanistic | responses, | | repetitive | patterns, | | and | insufficient | | | | | | | | | -| ----------- | ---------- | --- | ---------- | --------- | ---------- | --- | ------------ | ------------------ | --- | -------------- | ------- | ------- | ---------- | ---------- | ---------- | -| | | | | | | | | In reading | | comprehension, | | Ouyang | et | al. [2023] | con- | -| long-chain | reasoning. | | This | approach | represents | | a potential | | | | | | | | | -| | | | | | | | | struct supergraphs | | to | address | complex | contextual | | reasoning, | -paradigmshiftinlogicalreasoningoptimization,pushingthe -whileKnowRA[Maietal.,2025]autonomouslydetermines -boundariesofwhatLLMscanachieveinstructuredreasoning -| | | | | | | | | whether | to accept | external | knowledge | | to assist | document- | | -| ---------------------- | --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --------- | -------- | --------- | --- | --------- | --------- | --- | -| tasks. | | | | | | | | levelrelationextraction. | | | | | | | | -| Inference-TimeDecoding | | | | | | | | 5.4 Neuro-SymbolicApproaches | | | | | | | | -We categorize logical reasoning enhancement methods dur- Neural-symbolic hybrid methods represent a burgeoning re- -| ing inference-time | | into | inference-time | | | scaling | and con- | | | | | | | | | -| ------------------ | --- | ---- | -------------- | --- | --- | ------- | -------- | ----------- | ---- | ---- | ---------- | --- | -------- | ----------- | --- | -| | | | | | | | | search area | that | aims | to combine | the | powerful | representa- | | -straineddecoding. -tionalcapabilitiesofdeeplearningwiththeprecisionandin- -| Inference-time | | scaling | employs | computational | | | augmenta- | | | | | | | | | -| -------------- | --- | ------- | ------- | ------------- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -terpretabilityofsymbolicreasoning. -tion without parameter updates. One common approach is Formally, a neural-symbolic hybrid system aims to opti- -decoding with structured outputs and modular workflows. mize both the neural model M and the symbolic solver P -| GoT [Lei | et al., | 2023] | creates | structured | | reasoning | nodes | | | | | | | | | -| ------------ | ------- | ----- | ---------- | ---------- | ---------- | ---------- | ----- | ------------- | -------------------------------------------- | --- | ------------ | --- | ----------- | --- | --------- | -| | | | | | | | | (whereP | representsthesymbolicreasoningprocess)tomax- | | | | | | | -| to improve | complex | | multi-step | logical | reasoning. | | Simi- | | | | | | | | | -| | | | | | | | | imize logical | reasoning | | performance. | | The overall | | objective | -| larly, Chain | of | Logic | [Servantez | et | al., 2024] | introduces | a | | | | | | | | | -canbeexpressedas: -Decomposition-Recompositionstructureforlegalreasoning. -In other contexts, researchers design more complex modu- (M∗,P∗)=argmaxR(P(M(x))), -lar workflows for better performance [Creswell et al., 2023; M,P -Malonetal.,2024]. -where: -| Another | inference-time | | scaling | approach | | involves | stimu- | | | | | | | | | -| ------- | -------------- | --- | ------- | -------- | --- | -------- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -• M: Theneuralmodel,whichincludesboththemodel’s -| lating autonomous | | reasoning, | | guiding | LLMs | to | iteratively | | | | | | | | | -| ----------------- | -------- | ---------- | -------- | ------- | --------- | ----- | ----------- | ----------------------------------- | --- | --- | --- | --- | --- | -------------- | --- | -| | | | | | | | | parametersanditsdecodingstrategies. | | | | | | Itmapstheinput | | -| refine their | answers. | | Maieutic | | Prompting | [Jung | et al., | | | | | | | | | -x(e.g.,naturallanguage)intoasymbolicrepresentation -2022]eliminatescontradictionsthroughrecursivereasoning. -zwithinaformallanguageL: -| Similarly, | Logic-of-Thoughts | | | [Liu | et al., | 2024a] | and De- | | | | | | | | | -| ---------- | ----------------- | --- | --- | ---- | ------- | ------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -termLR[Sunetal.,2024]progressivelyapproachtheanswers -| | | | | | | | | | | | z =M(x), | z | ∈L. | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -inaniterativestyle. -Constraineddecodingmethods,ontheotherhand,focuson • P: The symbolic solver, which operates on the sym- -improvingthecontrollabilityandreliabilityofreasoningpro- bolic representation z produced by M to generate the -cesses. Neurologic[Luetal.,2021]enforcespredicatelogic finaloutputy: -| | | | | [Li | | 2024b] | | | | | y =P(z). | | | | | -| ------------ | ----- | ---------- | --- | --- | ------- | ------ | ---------- | --- | --- | --- | -------- | --- | --- | --- | --- | -| constraints, | while | Formal-LLM | | | et al., | | integrates | | | | | | | | | -automataforconstrainingplangeneration. -| | | | | | | | | • R: | Thereasoningperformancemetric, | | | | | whichevaluates | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | ------------------------------ | --- | --- | --- | --- | -------------- | --- | -theabilitytoperformlogicalreasoningtasks. -5.3 ExternalKnowledgeUtilization -Theoptimizationprocessinvolvestwokeydirections: -LLMsoftengenerateincorrectanswersduetohallucinations -| | | | | | | | | • Improving | | M: including | | refining | the model’s | | parame- | -| --------------- | --- | ------- | ----- | ---- | --- | ------- | ---------- | ----------- | --- | ------------ | --- | -------- | ----------- | --- | ------- | -| when performing | | complex | tasks | such | as | logical | reasoning, | | | | | | | | | -tersanddecodingstrategiestoproducesymbolicrepre- -makingitnecessarytoincorporateexternalknowledgetoas- -sist in producing accurate responses. Formally, the optimal sentationsthatarebothaccurateandcompatiblewithP. -integrationofexternalknowledgecanbeformulatedasajoint • Enhancing P: involving improving the symbolic -| optimizationproblem: | | | | | | | | solver’sabilitytoprocess. | | | | | | | | -| -------------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | - -By jointly optimizing M and P, neural-symbolic hybrid bottlenecks with large knowledge bases or complex rule de- -systemsaimtoleveragethestrengthsofbothneuralnetworks pendencies. Conversely, data-driven methods (e.g., instruc- -| | | | | | | | | | | | [Parmar | | 2024]) | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------- | --- | ------ | --- | -andsymbolicreasoningtoachievesuperiorlogicalreasoning tion tuning on LogicBench et al., achieve -capabilities. Itisworthnotingthatinearlierneural-symbolic broadertaskcoveragebutfailtogeneralizebeyondsyntactic -pipelines,P isoftenimplementedasafixedexternallogical patterns. How can we reconcile transparent reasoning with -reasoningengine,andthusisgenerallynotoptimized. How- black-box model performance? Hybrid architectures offer -ever,inadvancedpractice,LLMsareincreasinglybeingused promisebutintroducecomputationaloverhead,limitingprac- -| toperformtheroleofP,enablingdiverseoptimization. | | | | | | | ticaldeployment. | | | | | | | | -| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | -Fundamentally, these methods involve translating prob- ExistingbenchmarkslikeLogiQA[Liu -EvaluationRigor. -| lems into | symbolic | representations | | with | LLMs, | and ex- | | | | | | | | | -| --------------- | -------- | --------------- | ------- | ----- | ----- | ----------- | -------------- | ---- | ---------- | ----------- | ------- | ------- | --------------- | ------- | -| | | | | | | | et al., 2021b] | | and ReClor | [Yu | et al., | 2020] | conflate | reason- | -| ternal symbolic | | solvers | solving | them. | For | example, in | | | | | | | | | -| | | | | | | | ing ability | with | pattern | recognition | | through | multiple-choice | | -LINC[Olaussonetal.,2023],LLMsconvertnaturallanguage -| | | | | | | | formats. | While | efforts | like | NeuLR | [Xu et | al., 2023] | curate | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------- | ---- | ----- | ------ | ---------- | ------ | -(NL) into first-order logic (FOL) expressions, and utilize an “neutral” content to isolate reasoning from domain knowl- -externaltheoremproverforsymbolicdeductiveinference. -| | | | | | | | edge,theylackscopeforholisticevaluation. | | | | | | Currentmetrics | | -| ------- | ------- | -------- | --------- | -------------- | --- | ------ | ---------------------------------------- | --- | --------------------------------------- | --- | --- | --- | -------------- | --- | -| Further | efforts | focus on | improving | NL-to-symbolic | | trans- | | | | | | | | | -| | | | | | | | (e.g., accuracy, | | BLEU)failtoassessconsistency(invariance | | | | | | -lation. Oneprevailingapproachisdirectlyoptimizingtrans- -tologicallyequivalentinputs)orsoundness(adherencetofor- -| lation through | | training | [Yang | et al., | 2024a] | or decoding | | | | | | | | | -| -------------- | ---- | -------------- | ----- | --------- | ------ | ----------- | -------------------- | --- | ----------- | ------------------------------- | --- | ---- | ---------- | ---- | -| | | | | | | | malproofstructures). | | | Whatdefinesagoldstandardforlog- | | | | | -| strategies | [Ryu | et al., 2024], | | while the | other | depends on | | | | | | | | | -| | | | | | | | ical reasoning | | evaluation? | Benchmarks | | must | prioritize | sys- | -verification or correction mechanisms [Yang et al., 2024a; tematictestingofcoreprinciples(e.g.,transitivity,contrapo- -Panetal.,2023]. -sition)overtask-specificperformance. -Buildinguponthese,recentadvancementsaddressthetra- -ditional pipeline limitations by fully integrating LLMs into Future Directions. Addressing these challenges requires -reasoning processes. Logic Agent (LA) [Liu et al., 2024a] hybrid architectures that dynamically integrate neural and -| | | | | | | | symbolic | components, | | such | as | differentiable | | theorem | -| ------------- | -------- | ------- | -------- | ----------- | ----- | ---------- | ----------------------------------------- | ----------- | --- | ---- | --- | -------------- | ------------- | ------- | -| replaces | external | solvers | with | rule-guided | LLM | inference | | | | | | | | | -| | | | | | | | provers,tobalancescalabilityandprecision. | | | | | | Equallyimpor- | | -| chains, while | LLM-TRes | | [Toroghi | et al., | 2024] | implements | | | | | | | | | -tantisthedevelopmentofevaluationframeworksthatstress- -self-containedverifiablereasoningwithoutexternalsymbolic -solvers. SymbCoT[Xuetal.,2024c]coordinatestranslation, test models on perturbed logical statements (e.g., negated -planning,solvingandverificationentirelythroughLLMs.Xu premises, swapped quantifiers) to isolate reasoning from -[2024b] memorization. Multimodal reasoning, which grounds infer- -| et al. | propose | Aristotle, | | which | further | systematizes | | | | | | | | | -| ------------ | --------- | ----------- | --- | ------- | ------- | ------------ | ---------------- | ------- | ------------- | ------ | ---------- | ------ | ------------------ | --- | -| | | | | | | | ence in | diverse | modalities | (text, | images, | code), | presents | un- | -| the symbolic | reasoning | pipeline | | through | three | LLM-driven | | | | | | | | | -| | | | | | | | tapped potential | | for enhancing | | robustness | | and interpretabil- | | -| components: | Logical | Decomposer, | | Logical | Search | Router, | | | | | | | | | -andLogicalResolver. ity. Finally, interdisciplinary collaboration—leveraging in- -| | | | | | | | sights from | formal | logic, | cognitive | | science, | and | machine | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ------ | ------------ | --------- | --------- | -------- | ---- | ------- | -| | | | | | | | learning—will | | be essential | | to design | systems | that | reason | -6 Discussion -| | | | | | | | withandaboutuncertainty. | | | | UntilLLMsreliablydisentangle | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | --- | ---------------------------- | --- | --- | --- | -Theintegrationoflogicalreasoningintolargelanguagemod- logicfromlexicon,theirdeploymentinhigh-stakesdomains -els(LLMs)remainsacriticalchallenge,markedbypersistent -| | | | | | | | will remain | precarious. | | Bridging | this | gap | demands | rigorous | -| --- | --- | --- | --- | --- | --- | --- | ----------- | ----------- | --- | -------- | ---- | --- | ------- | -------- | -gapsbetweenheuristicperformanceandformallogicalrigor. -| | | | | | | | benchmarks, | scalable | | hybrid | methods, | and | a redefinition | of | -| --- | --- | --- | --- | --- | --- | --- | ----------- | -------- | --- | ------ | -------- | --- | -------------- | --- | -Below,weanalyzethreeunresolvedtensionsdominatingthe -evaluationparadigms. -fieldandoutlinefuturedirections. -7 Conclusion -| Robustness | vs. | Generalization. | | LLMs | exhibit | inconsis- | | | | | | | | | -| ---------- | --- | --------------- | --- | ---- | ------- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -tentperformanceinstructuredreasoningtaskssuchasdeduc- -| | | | | | | | This survey | synthesizes | | the | rapid advancements | | and | persis- | -| -------------- | --- | --------- | ---------- | --- | ----------- | ----- | ----------- | ----------- | --- | --- | ------------------ | --- | --- | ------- | -| tive inference | and | abductive | hypothesis | | generation. | While | | | | | | | | | -tentchallengesinlogicalreasoningforlargelanguagemodels -modelsfine-tunedondatasetslikeFOLIO[Hanetal.,2024a] -(LLMs). WhileLLMsdemonstrateimpressiveheuristicrea- -| excel in | controlled | settings, | they | struggle | with | adversarial | | | | | | | | | -| -------- | ---------- | --------- | ---- | -------- | ---- | ----------- | ------- | -------- | ------- | ------------------ | --- | --- | ---------- | --- | -| | | | | | | | soning, | rigorous | logical | inference—spanning | | | deductive, | in- | -perturbations or semantically equivalent rephrasings. This ductive, abductive, and analogical paradigms—remains in- -| inconsistency | arises | from | their | reliance | on surface-level | sta- | | | | | | | | | -| --------------------- | --------------------------------------- | ------ | ----- | --------------------- | ---------------- | ----------- | --------------------- | ------------- | -------------- | --------- | -------------- | ------------ | --------------- | --------- | -| | | | | | | | consistent | due | to limitations | | in robustness, | | generalization, | | -| tistical correlations | | rather | than | causal relationships, | | coupled | | | | | | | | | -| | | | | | | | and interpretability. | | We | analyzed | strategies | | to enhance | rea- | -| with limited | out-of-distribution | | | generalization. | | A key ques- | | | | | | | | | -| | | | | | | | soning, | including | neuro-symbolic | | | integration, | data-centric | | -| tionpersists: | canLLMsachievehuman-likerobustnesswith- | | | | | | | | | | | | | | -| | | | | | | | tuning, | reinforcement | | learning, | test-time | scaling, | | and other | -out sacrificing cross-domain adaptability? Current methods improved decoding methods, and highlighted benchmarks -prioritizenarrowtaskperformance,leavingreal-worldappli- -| | | | | | | | like FOLIO | and | LogiQA | for | systematic | evaluation. | | Future | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------ | --- | ---------- | ----------- | --- | ------ | -cabilityuncertain. -progresshingesonhybridarchitecturesthatunifyneuraland -Interpretabilityvs.Performance. Acentraltensionliesin symbolicreasoning,robustevaluationframeworks,andscal- -balancingneuralscalabilitywithsymbolicprecision. Neuro- able methods for cross-domain and multimodal inference. -symbolic approaches like Logic-LM [Pan et al., 2023] and Addressing these challenges will advance LLMs toward re- -| Symbol-LLM[Xuetal., | | | 2024a]embedformallogicsolvers | | | | | | | | | | | | -| ------------------- | --- | --- | ----------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -liable,interpretablereasoningcriticalforreal-worldapplica- -| into neural | architectures, | | improving | interpretability | | through | tions. | | | | | | | | -| ------------------- | -------------- | ----------------------------------- | --------- | ---------------- | --- | ------- | ------ | --- | --- | --- | --- | --- | --- | --- | -| step-by-stepproofs. | | However,thesemethodsfacescalability | | | | | | | | | | | | | - -| References | | | | | | | [Liuetal.,2021b] | | | | -| ---------- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | -JianLiu,LeyangCui,HanmengLiu,DandanHuang,Yile -| | | | | | | | Wang,andYueZhang. | Logiqa: | achallengedatasetformachinereading | | -| ---------------- | ------ | ------- | ---- | --------- | ------------ | ----- | ----------------- | ------- | ---------------------------------- | --- | -| [Beanetal.,2024] | Andrew | M Bean, | Simi | Hellsten, | Harry Mayne, | Jabez | | | | | -Magomere,EthanAChi,etal. Lingoly: Abenchmarkofolympiad-level comprehensionwithlogicalreasoning.2021. -linguisticreasoningpuzzlesinlow-resourceandextinctlanguages. arXiv [Liuetal.,2023a] HanmengLiu,JianLiu,LeyangCui,ZhiyangTeng,Nan -preprintarXiv:2406.06196,2024. -| | | | | | | | Duan,etal. Logiqa2.0—animproveddatasetforlogicalreasoninginnat- | | | | -| ----------------- | ----------------------------------- | --- | --- | --- | --- | ----- | --------------------------------------------------------------- | ----------------------------------- | --- | --- | -| | | | | | | | urallanguageunderstanding. | IEEE/ACMTransactionsonAudio,Speech, | | | -| [Bowenetal.,2024] | ChenBowen,RuneSætre,andYusukeMiyao. | | | | | Acom- | | | | | -andLanguageProcessing,pages2947–2962,2023. -prehensiveevaluationofinductivereasoningcapabilitiesandproblemsolv- -| inginlargelanguagemodels. | | | InProc.ofACLFindings,pages323–339, | | | | [Liuetal.,2023b] | | | | -| ------------------------- | --- | --- | ---------------------------------- | --- | --- | --- | ---------------- | --- | --- | --- | -HanmengLiu,RuoxiNing,ZhiyangTeng,JianLiu,Qiji -2024. Zhou,andYueZhang. Evaluatingthelogicalreasoningabilityofchatgpt -| [Cann,1993] | RonnieCann. | Formalsemantics: | | anintroduction. | | Cambridge | andgpt-4,2023. | | | | -| ----------- | ----------- | ---------------- | --- | --------------- | --- | --------- | -------------- | --- | --- | --- | -UniversityPress,UnitedStates,1993. -| | | | | | | | [Liuetal.,2023c] | HanmengLiu,ZhiyangTeng,LeyangCui,ChaoliZhang, | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --------------------------------------------- | --- | --- | -[CingilliogluandRusso,2019] Nuri Cingillioglu and Alessandra Russo. QijiZhou,andYueZhang. Logicot: Logicalchain-of-thoughtinstruction -Deeplogic:Towardsend-to-enddifferentiablelogicalreasoning,2019. tuning.InProc.ofEMNLPFindings,pages2908–2921,2023. -[Clarketal.,2021] PeterClark,OyvindTafjord,andKyleRichardson.Trans- [liuetal.,2023d] Hanmeng liu, Zhiyang Teng, Ruoxi Ning, Jian Liu, Qiji -formersassoftreasonersoverlanguage.InProc.ofIJCAI,2021. Zhou,andYueZhang. Glore: Evaluatinglogicalreasoningoflargelan- -guagemodels,2023. -| [Cobbeetal.,2021] | Karl | Cobbe, | Vineet | Kosaraju, | Mohammad | Bavarian, | | | | | -| ----------------- | ---- | ------ | ------ | --------- | -------- | --------- | --- | --- | --- | --- | -[Liuetal.,2024a] -MarkChen,HeewooJun,etal.Trainingverifierstosolvemathwordprob- Hanmeng Liu, Zhiyang Teng, Chaoli Zhang, and Yue -lems.arXivpreprintarXiv:2110.14168,2021. Zhang.Logicagent:Enhancingvaliditywithlogicruleinvocation,2024. -[Creswelletal.,2023] -AntoniaCreswell, MurrayShanahan, andIrinaHig- [Liuetal.,2024b] LihuiLiu,ZihaoWang,RuizhongQiu,YikunBan,Eunice -gins. Selection-inference: Exploiting large language models for inter- Chan,etal. Logicqueryofthoughts: Guidinglargelanguagemodelsto -pretablelogicalreasoning.InProc.ofICLR,2023. answercomplexlogicquerieswithknowledgegraphs,2024. -[DeepSeek-AI,2025] DeepSeek-AI. DeepSeek-R1:IncentivizingReasoning [Liuetal.,2024c] YinhongLiu,ZhijiangGuo,TianyaLiang,EhsanShareghi, -CapabilityinLLMsviaReinforcementLearning.Technicalreport,2025. IvanVulic´,andNigelCollier. Aligningwithlogic: Measuring,evaluating -andimprovinglogicalconsistencyinlargelanguagemodels.arXivpreprint -| [DelandFishel,2023] | | MaksymDelandMarkFishel.Truedetective:Adeep | | | | | | | | | -| ------------------- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | -arXiv:2410.02205,2024. -abductivereasoningbenchmarkundoableforGPT-3andchallengingfor -GPT-4. InProceedingsofthe12thJointConferenceonLexicalandCom- [Luetal.,2021] Ximing Lu, Peter West, Rowan Zellers, Ronan Le Bras, -putationalSemantics(*SEM2023),pages314–322,2023. Chandra Bhagavatula, and Yejin Choi. NeuroLogic decoding: -[Fengetal.,2024] (un)supervisedneuraltextgenerationwithpredicatelogicconstraints. In -JiazhanFeng,RuochenXu,JunhengHao,HiteshiSharma, Proc.ofNAACL,pages4288–4299,2021. -YelongShen,etal.Languagemodelscanbedeductivesolvers.InProc.of -ACLFindings,pages4026–4042,2024. [Luoetal.,2024] ManLuo, ShrinidhiKumbhar, Mingshen, MihirParmar, -| | | | | | | | NeerajVarshney,etal. | Towardslogiglue:Abriefsurveyandabenchmark | | | -| --------------------- | --- | --------------------------------------------- | --- | --- | --- | --- | -------------------- | ----------------------------------------- | --- | --- | -| [Gandarelaetal.,2024] | | Joa˜oPedroGandarela,DaniloSCarvalho,andAndre´ | | | | | | | | | -foranalyzinglogicalreasoningcapabilitiesoflanguagemodels,2024. -| Freitas. Inductivelearningoflogicaltheorieswithllms: | | | | | Acomplexity- | | | | | | -| ---------------------------------------------------- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | -gradedanalysis.arXivpreprintarXiv:2408.16779,2024. [Maietal.,2025] Chengcheng Mai, Yuxiang Wang, Ziyu Gong, Hanxiang -Wang,andYihuaHuang.Knowra:Knowledgeretrievalaugmentedmethod -| [Hanetal.,2024a] | Simeng | Han, | Hailey Schoelkopf, | | Yilun Zhao, | Zhenting | | | | | -| ---------------- | ------ | ---- | ------------------ | --- | ----------- | -------- | --- | --- | --- | --- | -Qi,MartinRiddell,etal. FOLIO:Naturallanguagereasoningwithfirst- fordocument-levelrelationextractionwithcomprehensivereasoningabili- -ties,2025. -orderlogic.InProc.ofEMNLP,pages22017–22031,2024. -[Malonetal.,2024] -[Hanetal.,2024b] SimengHan,AaronYu,RuiShen,ZhentingQi,Martin ChristopherMalon,MartinMin,XiaodanZhu,etal.Ex- -ploringtheroleofreasoningstructuresforconstructingproofsinmulti- -| Riddell,etal. | P-FOLIO:Evaluatingandimprovinglogicalreasoningwith | | | | | | | | | | -| ------------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -InProc.ofEMNLPFindings, stepnaturallanguagereasoningwithlargelanguagemodels. InProc.of -abundanthuman-writtenreasoningchains. EMNLP,pages15299–15312,2024. -pages16553–16565,2024. -[Harutaetal.,2020] [McCarthyandHayes,1981] J.McCarthyandP.J.Hayes. Somephilosophi- -IzumiHaruta,KojiMineshima,andDaisukeBekki.Log- calproblemsfromthestandpointofartificialintelligence. InReadingsin -| icalinferenceswithcomparativesandgeneralizedquantifiers. | | | | | | InProc.of | | | | | -| -------------------------------------------------------- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | -ArtificialIntelligence,pages431–450.1981. -ACL,pages263–270,2020. -[McCarthy,1959] -[Helweetal.,2022] ChadiHelwe,Chloe´Clavel,andFabianSuchanek.Log- JohnMcCarthy. Programswithcommonsense. InPro- -ceedingsoftheTeddingtonConferenceontheMechanizationofThought -itorch: Apytorch-basedlibraryforlogicalreasoningonnaturallanguage. -| InProc.ofEMNLP,2022. | | | | | | | Processes,1959. | | | | -| -------------------- | --- | ------------- | ---- | --------- | -------- | ---------- | --------------- | ----------------------------------------------------- | --- | --- | -| | | | | | | | [McCarthy,1989] | JohnMcCarthy.Artificialintelligence,logicandformaliz- | | | -| [Jiangetal.,2024] | Jin | Jiang, Yuchen | Yan, | Yang Liu, | Yonggang | Jin, Shuai | | | | | -Peng,etal. Logicpro:Improvingcomplexlogicalreasoningviaprogram- ingcommonsense. PhilosophicalLogicandArtificialIntelligence,pages -161–190,1989. -guidedlearning.arXivpreprintarXiv:2409.12929,2024. -[Jiaoetal.,2024] Fangkai Jiao, Chengwei Qin, Zhengyuan Liu, Nancy F. [Morishitaetal.,2024] TerufumiMorishita,GakuMorio,AtsukiYamaguchi, -| | | | | | | | andYasuhiroSogawa. | Enhancingreasoningcapabilitiesofllmsviaprin- | | | -| ------------------- | ----------- | --------------------------------------------- | ------------- | -------- | --------- | ----- | --------------------------- | -------------------------------------------- | --- | --- | -| Chen,andShafiqJoty. | | Learningplanning-basedreasoningbytrajectories | | | | | | | | | -| | | | | | | | cipledsyntheticlogiccorpus. | InProc.ofNeurIPS,pages73572–73604, | | | -| collection | and process | reward | synthesizing. | In Proc. | of EMNLP, | pages | 2024. | | | | -334–350,2024. -[Muennighoffetal.,2025] -| [Jungetal.,2022] | | | | | | | | NiklasMuennighoff,ZitongYang,WeijiaShi,Xi- | | | -| ---------------- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------ | --- | --- | -JaehunJung,LianhuiQin,SeanWelleck,FaezeBrahman, angLisaLi,LiFei-Fei,etal. s1:Simpletest-timescaling. arXivpreprint -| ChandraBhagavatula,etal. | | Maieuticprompting:Logicallyconsistentrea- | | | | | | | | | -| ------------------------ | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -arXiv:2501.19393,2025. -soningwithrecursiveexplanations.InProc.ofEMNLP,pages1266–1279, -2022. [NewellandSimon,1956] A. Newell and H. Simon. The logic theory -[Leietal.,2023] BinLei,ChunhuaLiao,CaiwenDing,etal.Boostinglogical machine–acomplexinformationprocessingsystem. IRETransactionson -InformationTheory,1956. -| reasoninginlargelanguagemodelsthroughanewframework: | | | | | | Thegraph | | | | | -| --------------------------------------------------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -ofthought.arXivpreprintarXiv:2308.08614,2023. [Nguyenetal.,2023] Ha-Thanh Nguyen, Randy Goebel, Francesca Toni, -| | | | | | | | KostasStathis,andKenSatoh. | Howwelldosotalegalreasoningmodels | | | -| --------------- | ---------------------------------------------- | --- | --- | --- | --- | --- | -------------------------- | --------------------------------- | --- | --- | -| [Lietal.,2024a] | QintongLi,LeyangCui,XueliangZhao,LingpengKong, | | | | | | | | | | -supportabductivereasoning?,2023. -| andWeiBi. | GSM-plus: | Acomprehensivebenchmarkforevaluatingthe | | | | | | | | | -| --------------------------------------------- | --------- | --------------------------------------- | --- | --- | ------------- | --- | -------------------- | -------------- | -------- | ------------------- | -| robustnessofLLMsasmathematicalproblemsolvers. | | | | | InProc.ofACL, | | [Olaussonetal.,2023] | | | | -| | | | | | | | | Theo Olausson, | Alex Gu, | Ben Lipkin, Cedegao | -pages2961–2984,2024. Zhang,ArmandoSolar-Lezama,etal. LINC:Aneurosymbolicapproach -[Lietal.,2024b] ZelongLi,WenyueHua,HaoWang,HeZhu,andYongfeng forlogicalreasoningbycombininglanguagemodelswithfirst-orderlogic -provers.InProc.ofEMNLP,pages5153–5176,2023. -| Zhang. Formal-llm: | | Integratingformallanguageandnaturallanguagefor | | | | | | | | | -| ------------------ | --- | ---------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -controllablellm-basedagents.arXivpreprintarXiv:2402.00798,2024. [OpenAI,2024] OpenAI. LearningtoreasonwithLLMs. Technicalreport, -2024. -| [Liuetal.,2021a] | HanmengLiu,LeyangCui,JianLiu,andYueZhang.Nat- | | | | | | | | | | -| ---------------- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -[Ouyangetal.,2023] -urallanguageinferenceincontext-investigatingcontextualreasoningover SiruOuyang,ZhuoshengZhang,andHaiZhao. Fact- -longtexts.Proc.ofAAAI,pages13388–13396,2021. drivenlogicalreasoningformachinereadingcomprehension,2023. - -| [Panetal.,2023] | | | | | | | [Tianetal.,2021] | | | | | | -| --------------- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | --- | --- | -LiangmingPan, AlonAlbalak, XinyiWang, andWilliam JidongTian,YitianLi,WenqingChen,LiqiangXiao,Hao -Wang. Logic-LM: Empowering large language models with symbolic He, andYaohuiJin. Diagnosingthefirst-orderlogicalreasoningability -solversforfaithfullogicalreasoning. InProc.ofEMNLPFindings,pages throughLogicNLI.InProc.ofEMNLP,pages3738–3747,2021. -3806–3824,2023. -| | | | | | | | [Toroghietal.,2024] | Armin Toroghi, | | Willis Guo, | Ali Pesaranghader, | and | -| --- | --- | --- | --- | --- | --- | --- | ------------------- | -------------- | --- | ----------- | ------------------ | --- | -[Parmaretal.,2023] MihirParmar, NeerajVarshney, NisargPatel, Santosh ScottSanner. Verifiable, debuggable, andrepairablecommonsenselogi- -Mashetty, ManLuo, etal. Logicbench: Abenchmarkforevaluationof calreasoningviallm-basedtheoryresolution. InProc.ofEMNLP,pages -| logicalreasoning,2023. | | | | | | | 6634–6652,2024. | | | | | | -| ---------------------- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | -[Parmaretal.,2024] MihirParmar,NisargPatel,NeerajVarshney,Mutsumi [Wangetal.,2019] Po-Wei Wang, Priya L. Donti, Bryan Wilder, and Zico -Nakamura,ManLuo,etal. Logicbench:Towardssystematicevaluationof Kolter.Satnet:Bridgingdeeplearningandlogicalreasoningusingadiffer- -logicalreasoningabilityoflargelanguagemodels. InProc.ofACL,pages entiablesatisfiabilitysolver,2019. -13679–13707,2024. -| | | | | | | | [Wangetal.,2022] | Siyuan Wang, | Zhongkun | | Liu, Wanjun | Zhong, Ming | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | ------------ | -------- | --- | ----------- | ----------- | -[Pereira,1982] -FernandoCarlosNevesPereira. Logicfornaturallanguage Zhou,ZhongyuWei,etal.Fromlsat:Theprogressandchallengesofcom- -analysis.1982. plexreasoning. IEEE/ACMTransactionsonAudio,Speech,andLanguage -Processing,2022. -| [PetersenandvanderPlas,2023] | | | MollyPetersenandLonnekevanderPlas. | | | | | | | | | | -| ---------------------------- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Canlanguagemodelslearnanalogicalreasoning?investigatingtrainingob- [Wangetal.,2024a] ChenWang,XudongLi,HaoranLiu,XinyueWu,and -jectivesandcomparisonstohumanperformance.InProc.ofEMNLP,pages WantingHe. Efficientlogicalreasoninginlargelanguagemodelsthrough -16414–16425,2023. program-guidedlearning.AuthoreaPreprints,2024. -[Plaatetal.,2024] AskePlaat,AnnieWong,SuzanVerberne,JoostBroekens, [Wangetal.,2024b] Chenxu Wang, Ping Jian, and Zhen Yang. Thought- -| NikivanStein,andThomasBack. | | | | Reasoningwithlargelanguagemodels, | | | | | | | | | -| --------------------------- | --- | --- | --- | --------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -pathcontrastivelearningviapremise-orienteddataaugmentationforlogical -asurvey.arXivpreprintarXiv:2407.11511,2024. readingcomprehension.arXivpreprintarXiv:2409.14495,2024. -[Qinetal.,2024] -Chengwei Qin, Wenhan Xia, Tan Wang, Fangkai Jiao, [Wijesiriwardeneetal.,2023] Thilini Wijesiriwardene, Ruwan Wickrama- -YuchenHu,etal. Relevantorrandom: Canllmstrulyperformanalogi- rachchi,BimalGajera,ShreeyashGowaikar,ChandanGupta,etal. ANA- -calreasoning?,2024. LOGICAL-anovelbenchmarkforlongtextanalogyevaluationinlarge -[Rajaramanetal.,2023] KanagasabaiRajaraman,SaravananRajamanickam, languagemodels.InProc.ofACLFindings,pages3534–3549,2023. -andWeiShi. Investigatingtransformer-guidedchainingforinterpretable [Xietal.,2024] ZhihengXi,WenxiangChen,BoyangHong,SenjieJin,Rui -naturallogicreasoning.InProc.ofACLFindings,pages9240–9253,2023. -Zheng,etal.Traininglargelanguagemodelsforreasoningthroughreverse -curriculumreinforcementlearning.InProc.ofICML,2024. -| [Rybetal.,2022] | Samuel | Ryb, | Mario | Giulianelli, | Arabella | Sinclair, and | | | | | | | -| --------------- | ------ | ---- | ----- | ------------ | -------- | ------------- | --- | --- | --- | --- | --- | --- | -RaquelFerna´ndez. AnaLog: Testinganalyticalanddeductivelogiclearn- [Xiaetal.,2024] YuXia, RuiWang, XuLiu, MingyanLi, TongYu, etal. -abilityinlanguagemodels.InProceedingsofthe11thJointConferenceon Beyondchain-of-thought:Asurveyofchain-of-xparadigmsforllms.arXiv -LexicalandComputationalSemantics,pages55–68,2022. preprintarXiv:2404.15676,2024. -[Ryuetal.,2024] HyunRyu,GyeongmanKim,HyeminSLee,andEunho [Xuetal.,2023] FangzhiXu,QikaLin,JiaweiHan,TianzheZhao,JunLiu, -Yang. Divide and translate: Compositional first-order logic transla- andErikCambria. Arelargelanguagemodelsreallygoodlogicalreason- -| tion and | verification | for | complex | logical | reasoning. | arXiv preprint | | | | | | | -| -------- | ------------ | --- | ------- | ------- | ---------- | -------------- | --- | --- | --- | --- | --- | --- | -ers?acomprehensiveevaluationandbeyond,2023. -arXiv:2410.08047,2024. -| | | | | | | | [Xuetal.,2024a] | FangzhiXu,ZhiyongWu,QiushiSun,SiyuRen,FeiYuan, | | | | | -| ------------------- | --- | -------- | -------- | ------- | ------- | ------------- | ------------------ | ---------------------------------------------- | ------------ | --- | -------------- | ------------- | -| [Saparovetal.,2023] | | Abulhair | Saparov, | Richard | Yuanzhe | Pang, Vishakh | | | | | | | -| | | | | | | | et al. Symbol-LLM: | Towards | foundational | | symbol-centric | interface for | -Padmakumar,NitishJoshi,MehranKazemi,etal. Testingthegeneralde- largelanguagemodels.InProc.ofACL,pages13091–13116,2024. -ductivereasoningcapacityoflargelanguagemodelsusingoodexamples. -InProc.ofNeurIPS,pages3083–3105,2023. [Xuetal.,2024b] Jundong Xu, Hao Fei, Meng Luo, Qian Liu, Liangming -| | | | | | | | Pan, etal. | Aristotle: Masteringlogicalreasoningwithalogic-complete | | | | | -| --- | --- | --- | --- | --- | --- | --- | ---------- | ------------------------------------------------------- | --- | --- | --- | --- | -[Servantezetal.,2024] Sergio Servantez, Joe Barrow, Kristian Hammond, decompose-search-resolveframework. arXivpreprintarXiv:2412.16953, -| andRajivJain. | Chainoflogic: | | Rule-basedreasoningwithlargelanguage | | | | | | | | | | -| ------------- | ------------- | --- | ------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | -2024. -models.InProc.ofACLFindings,pages2721–2733,2024. -| | | | | | | | [Xuetal.,2024c] | JundongXu,HaoFei,LiangmingPan,QianLiu,Mong-Li | | | | | -| --- | --- | --- | --- | --- | --- | --- | --------------- | --------------------------------------------- | --- | --- | --- | --- | -[Shietal.,2021] JihaoShi,XiaoDing,LiDu,TingLiu,andBingQin.Neural Lee, andWynneHsu. Faithfullogicalreasoningviasymbolicchain-of- -naturallogicinferenceforinterpretablequestionanswering. InProc.of thought.InProc.ofACL,pages13326–13365,2024. -EMNLP,pages3673–3684,2021. [Yangetal.,2023] Kaiyu Yang, Aidan M. Swope, Alex Gu, Rahul Chala- -[Sinhaetal.,2019] KoustuvSinha,ShagunSodhani,JinDong,JoellePineau, mala, Peiyang Song, et al. Leandojo: theorem proving with retrieval- -andWilliamL.Hamilton. Clutrr: Adiagnosticbenchmarkforinductive augmentedlanguagemodels.InProc.ofICONIP,2023. -| reasoningfromtext. | | EmpiricalMethodsofNaturalLanguageProcessing | | | | | | | | | | | -| ------------------ | --- | ------------------------------------------- | --- | --- | --- | --- | ----------------- | --------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Yangetal.,2024a] | YuanYang,SihengXiong,AliPayani,EhsanShareghi, | | | | | -(EMNLP),2019. -| | | | | | | | andFaramarzFekri. | Harnessingthepoweroflargelanguagemodelsfor | | | | | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | ------------------------------------------ | --- | --- | --- | --- | -[Srivastavaetal.,2022] AarohiSrivastava,AbhinavRastogi,AbhishekRao, naturallanguagetofirst-orderlogictranslation. InProc.ofACL,pages -| AbuAwalMdShoeb,AbubakarAbid,etal. | | | | | Beyondtheimitationgame: | | 6942–6959,2024. | | | | | | -| --------------------------------- | --- | --- | --- | --- | ----------------------- | --- | --------------- | --- | --- | --- | --- | --- | -Quantifyingandextrapolatingthecapabilitiesoflanguagemodels. arXiv [Yangetal.,2024b] Zonglin Yang, Li Dong, Xinya Du, Hao Cheng, Erik -preprintarXiv:2206.04615,2022. -Cambria,etal.Languagemodelsasinductivereasoners.InProc.ofEACL, -| [Staceyetal.,2022] | | | | | | | pages209–225,2024. | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | -JoeStacey,PasqualeMinervini,HaimDubossarsky,and -| MarekRei. | Logicalreasoningwithspan-levelpredictionsforinterpretable | | | | | | | | | | | | -| --------- | --------------------------------------------------------- | --- | --- | --- | --- | --- | -------------- | ------------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Yuetal.,2020] | WeihaoYu,ZihangJiang,YanfeiDong,andJiashiFeng.Re- | | | | | -androbustNLImodels.InProc.ofEMNLP,pages3809–3823,2022. -clor:Areadingcomprehensiondatasetrequiringlogicalreasoning.InProc. -| [Sullivan,2024] | MichaelSullivan.Itisnottruethattransformersareinductive | | | | | | ofICLR,2020. | | | | | | -| --------------- | ------------------------------------------------------- | --- | --- | --- | --- | -------------- | ------------ | --- | --- | --- | --- | --- | -| learners: | ProbingNLImodelswithexternalnegation. | | | | | InProc.ofEACL, | | | | | | | -pages1924–1945,2024. [Yuetal.,2024] FeiYu,HongboZhang,PrayagTiwari,andBenyouWang. -| | | | | | | | Naturallanguagereasoning,asurvey. | | | ACMComputingSurveys,pages1– | | | -| --------------- | ----------------------------------------------- | --- | --- | --- | --- | --- | --------------------------------- | --- | --- | --------------------------- | --- | --- | -| [Sunetal.,2023] | JiankaiSun,ChuanyangZheng,EnzeXie,ZhengyingLiu, | | | | | | 39,2024. | | | | | | -arXiv -RuihangChu,etal. Asurveyofreasoningwithfoundationmodels. [Yuanetal.,2023] ZhangdieYuan,SongboHu,IvanVulic´,AnnaKorhonen, -preprintarXiv:2312.11562,2023. -| | | | | | | | andZaiqiaoMeng. | Canpretrainedlanguagemodels(yet)reasondeduc- | | | | | -| --------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------ | -------------------------------------------- | --- | --- | --- | --- | -| [Sunetal.,2024] | | | | | | | tively? InProc.ofEACL,pages1447–1462,2023. | | | | | | -HongdaSun,WeikaiXu,WeiLiu,JianLuan,BinWang, -| etal. Determlr: | | Augmentingllm-basedlogicalreasoningfromindetermi- | | | | | | | | | | | -| --------------- | --- | ------------------------------------------------- | --- | --- | --- | --- | ------------------- | ---- | ------ | --------- | ----------- | -------- | -| | | | | | | | [ZayyadandAdi,2024] | Majd | Zayyad | and Yossi | Adi. Formal | language | -nacytodeterminacy.InProc.ofACL,pages9828–9862,2024. -knowledgecorpusforretrievalaugmentedgeneration,2024. -| [Tafjordetal.,2021] | | Oyvind | Tafjord, | Bhavana | Dalvi, | and Peter Clark. | | | | | | | -| ------------------- | ---------- | ------ | ------------- | ------- | ------ | -------------------- | ----------------- | --------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | [Zhangetal.,2024] | YuxiangZhang,ShangxiWu,YuqiYang,JiangmingShu, | | | | | -| ProofWriter: | Generating | | implications, | proofs, | and | abductive statements | | | | | | | -overnaturallanguage.InProc.ofACLFindings,pages3621–3634,2021. JinlinXiao,etal. o1-coder: ano1replicationforcoding. arXivpreprint -arXiv:2412.00154,2024. -| [Thatikondaetal.,2025] | | Ramya | Keerthy | Thatikonda, | | Wray Buntine, and | [Zhaoetal.,2024] | | | | | | -| ---------------------- | --- | ----- | ------- | ----------- | --- | ----------------- | ---------------- | --- | --- | --- | --- | --- | -YuZhao,HuifengYin,BoZeng,HaoWang,TianqiShi, -EhsanShareghi. Assessingthealignmentoffolclosenessmetricswithhu- etal.Marco-o1:Towardsopenreasoningmodelsforopen-endedsolutions. -manjudgement.arXivpreprintarXiv:2501.08613,2025. -arXivpreprintarXiv:2411.14405,2024. \ No newline at end of file + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +F + + + + + + + + + + + + + + + + + + + + + + + + Indicates the most commonly given grade, or mode. The frequency of each grade is indicated with darker shading for +more common responses, according to the key below. + +0% + +50% + +Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of +the following issues? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +PARENT/CHILD DISCUSSIONS ABOUT POLITICS ARE USUALLY CORDIAL, +BUT IT CAN BE DIFFICULT TO SWAY THE OTHER GENERATION WHEN +VIEWS DIFFER. + +Young people age 15-26 and their parents have fairly regular discussions with one another about +politics and social issues, and few see these discussions as a source of conflict. Across the two groups, +64 percent of young people and parents say they discuss politics with each other at least once a +month, 20 percent say they have these discussions less often, and 16 percent say they never talk about +politics or social issues together. + +Among those who discuss political and social issues with their parents or children, a majority say they +usually see eye to eye in these conversations, and about a third say they debate diplomatically. Fewer +say the discussion turns into a major conflict or requires changing the subject to avoid arguments. + +© Copyright 2018. The Associated Press and NORC + +June 2018 6 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Most parents and their children see eye to eye on politics and social issues. + +Someone changes the +subject quickly to avoid +getting into disagreements. +9% + +It turns into +World War III. +5% + +We debate things +diplomatically. +32% + +We usually see +eye to eye. +53% + +Question: When you talk with [him/her/them] about political and social issues, how would you describe how the discussion +goes? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +When the two generations were asked about political discussions with a particular family member, +parents’ accounts are similar regardless of the specific relationship, such as mother-daughter or father- +son. Sons reporting on discussions with fathers are more likely than other pairings to say they talk +about politics at least once a week, but least likely to say these discussions tend to go smoothly. Only +42 percent of sons say they usually see eye to eye with their fathers, while 49 percent of sons say the +same about their mothers, 54 percent of daughters say so about their fathers, and 62 percent say so +about their mothers. + +Those who identify as political independents (36 percent) are less likely than Democrats (60 percent) +or Republicans (53 percent) to say they usually see eye to eye in parent/child discussions about +political and social issues. + +When asked about potential differences in opinion on specific issues, both generations report high +levels of agreement. In their discussions, at least 6 in 10 say there is agreement when discussing each +of 18 topics. Agreement levels are highest on Trump’s job performance and on the issues of racism, +sexual harassment and assault, and health care policy, and lowest on LGBT rights, abortion, and +foreign relations. + +While agreement is high, there also seems to be a fair amount of ambiguity about family members’ +political positions. On 13 of the 18 issues asked about, at least 15 percent say they are unsure if they +tend to agree or disagree. + +© Copyright 2018. The Associated Press and NORC + +June 2018 7 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and their parents tend to be on the same page with many political and social issues. + +President Trump's job performance + +Racism + +Sexual harassment and assault + +Health care policy + +The quality of education + +Treatment of citizens by police and justice system + +Cost of higher education + +The economy + +Taxes + +The environment and climate change + +Income inequality + +Immigration + +Job creation + +Terrorism and homeland security + +Gun control + +The rights of LGBT Americans + +Foreign relations + +Abortion + +76 + +76 + +76 + +75 + +74 + +74 + +73 + +72 + +71 + +71 + +70 + +70 + +69 + +69 + +69 + +62 + +62 + +61 + +13 + +8 + +9 + +10 + +9 + +8 + +13 + +15 + +13 + +9 + +8 + +11 + +9 + +11 + +17 + +19 + +13 + +16 + +8 + +9 + +14 + +17 + +15 + +13 + +19 + +19 + +19 + +20 + +23 + +17 + +22 + +20 + +14 + +19 + +24 + +22 + +Tend to agree + +Tend to disagree + +Not sure + +Questions: When talking about the following issues, do you tend to agree or disagree with [him/her/them]? +When talking with [your family member] about Donald Trump and how he’s handling his job as president, do you tend to +agree or disagree with [him/her/them]? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +The two generations report different issues as the biggest sources of disagreement. Among those who +say they generally don’t see eye to eye when talking about political and social issues, young people +age 15-26 say they tend to disagree with their parents most about racism (37 percent), LGBT rights (36 +percent), and how Trump is handling his presidency (33 percent). Parents of young people age 15-26 +report that the largest sources of disagreement are gun control (27 percent), LGBT rights (25 percent), +and how Trump is handling his presidency (21 percent). + +When there are disagreements, few young people and parents are able to persuade one another to +change their views, and the findings are consistent between the generations. Eleven percent overall +say they are always or often able to persuade their family member to change their views. Fifty-three +percent acknowledge that they are sometimes able to persuade their family member. Just 6 percent +say the reverse is true, that they are often or always convinced to change their views by their family +member. Another 44 percent say their family member is sometimes able to persuade them to change +their own views. + +© Copyright 2018. The Associated Press and NORC + +June 2018 8 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Few young people or parents find it easy to persuade one another to change their political and social +views. + +How often you are able to persuade your + family member to change their views + +11 + +53 + +36 + +How often your family member is able + to persuade you to change your views + +6 + +44 + +49 + +0 + +10 + +20 + +30 + +40 + +50 + +60 + +70 + +80 + +90 + +100 + +Percent + +Always or often + +Sometimes + +Rarely or never + +Questions: In general, when you disagree with [him/her/that person] about political or social issues, how often would you +say that you’re able to persuade [him/her/them] to change [his/her/their views]? +In general, when you disagree with [him/her/that person] about political or social issues, how often would you say that [he +is/she is/they are] able to change your views? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +YOUNG PEOPLE ARE PERCEIVED AS MORE IDEALISTIC THAN THE OLDER +GENERATION, AND PARENTS FEEL THEIR CHILDREN’S GENERATION IS +LESS INFORMED THAN THEIR OWN. + +When asked to compare the generations, young people age 15-26 and parents of this age group tend to +agree on who is the more patriotic and rigid generation, but there is some disagreement about who is +more informed. + +About two-thirds of both young people and parents say the older generation is more patriotic. More +than half of each group feel the older generation is more rigid. + +Among both groups, the largest proportion view the parents’ generation as more politically active and +moral than the younger generation, and the younger generation as more idealistic. + +Young people are divided on whether their generation is more uninformed, their parents’ generation is +more uninformed, or whether the two are about equal. Parents believe the younger generation is more +uninformed than their own. + +© Copyright 2018. The Associated Press and NORC + +June 2018 9 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Young people and parents agree that the older generation is more patriotic and more rigid. + +Which generation is more…? + +Patriotic + +Moral + +Rigid + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +Compassionate + +Young people +say… + +Parents say… + +Politically active + +Young people +say… + +Idealistic + +Uninformed + +Parents say… + +Young people +say… + +Parents say… + +Young people +say… + +Parents say… + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +0 + +64 + +10 + +25 + +20 + +40 + +60 + +80 + +100 + +65 + +7 + +27 + +20 + +44 + +20 + +20 + +20 + +20 + +53 + +57 + +52 + +40 + +60 + +16 + +40 + +40 + +40 + +40 + +60 + +13 + +60 + +60 + +19 + +60 + +28 + +34 + +40 + +60 + +23 + +40 + +40 + +60 + +33 + +60 + +80 + +39 + +80 + +33 + +80 + +100 + +100 + +100 + +17 + +25 + +80 + +100 + +28 + +80 + +37 + +80 + +40 + +80 + +100 + +100 + +100 + +25 + +80 + +100 + +32 + +28 + +40 + +60 + +80 + +100 + +30 + +40 + +28 + +23 + +32 + +19 + +20 + +20 + +20 + +20 + +40 + +60 + +80 + +100 + +47 + +30 + +40 + +60 + +32 + +40 + +43 + +40 + +60 + +60 + +80 + +34 + +80 + +37 + +80 + +100 + +100 + +100 + +20 + +38 + +20 + +41 + +20 + +37 + +20 + +Parents' generation + +Young people's generation + +Both about equally + +Question: Thinking about people in your generation, compared to people in [his/her/their] generation, do you think the +following words describe your generation better, describe [his/her/their] generation better, or describe both generations +about equally well? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +© Copyright 2018. The Associated Press and NORC + +June 2018 10 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Republicans are more likely than Democrats to describe their own generation as patriotic (51 percent +vs. 32 percent) and are less likely to describe their own generation as idealistic (27 percent vs. 39 +percent). + +FOR SOME, POLITICS SEEPS INTO INTERACTIONS WITH EXTENDED +FAMILY. + +A significant minority of both generations have had political disagreements with relatives that have +led to conflict. Twenty percent have unfriended or blocked a family member on social media due to a +disagreement over politics, and 21 percent have themselves been unfriended or blocked by a family +member. Fewer than 1 in 5 have had a holiday gathering ruined (11 percent), have decided not to attend +a family event (12 percent), or have had a family member decide to not attend a family event (17 +percent) due to political disagreements. + +While there are no differences by generation, Democrats are more likely than Republicans to have had +political disagreements with relatives affect their social media behavior. Democrats are more likely +than Republicans to say they’ve unfriended or blocked a family member on social media (26 percent +vs. 12 percent) and to have had a family member unfriend or block them on social media (27 percent vs. +15 percent). + +EVEN AMONG THOSE WHO APPROVE OF TRUMP, THE REPORT CARD IS +MEDIOCRE. + +Young people age 15-34 give poor marks to the federal government on its handling of many public +policy issues. Even among Trump’s supporters, the report card is still only average. Young people who +approve of Trump’s job performance most commonly give the federal government a C for all issues +except terrorism and homeland security and treatment of citizens by the police and justice system, +which earn a B, and abortion, which earns an F. For those who disapprove of Trump, F is the most +common grade for all issues except job creation, which most commonly gets graded a C. + +Most common grades for the federal government’s handling of issues. + +Approve of Trump + +Disapprove of Trump + +Gun control + +Cost of higher education + +Foreign relations + +Immigration + +The environment and climate change + +Health care policy + +Income inequality + +The economy + +Taxes + +Sexual harassment and assault + +Abortion + +Racism + +The rights of LGBT Americans + +Terrorism and homeland security + +Treatment of citizens by police and justice system + +The quality of education + +Job creation + +C + +C + +C + +C + +C + +C + +C + +C + +C + +C + +F + +C + +C + +B + +B + +C + +C + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +F + +C + +Question: What letter grade, A through F, would you give the federal government for how well they are addressing each of +the following issues? +Source: MTV/AP-NORC poll conducted April 23-May 9, 2018, with 607 parents and 558 children age 15-26 nationwide + +© Copyright 2018. The Associated Press and NORC + +June 2018 11 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Looking at grades by political party, the results are similar. Democrats’ most common grade across +most issues is an F. However, Democrats give better marks to the federal government’s efforts on job +creation, most commonly grading it a C. + +Republicans are slightly more mixed, but most commonly give issues a C. For abortion, they most +commonly give a grade of F. + +STUDY METHODOLOGY + +This survey was conducted by The Associated Press-NORC Center for Public Affairs Research and +with funding from MTV and NORC at the University of Chicago. Staff from NORC at the University of +Chicago, The Associated Press, and MTV collaborated on all aspects of the study. + +Data were collected using AmeriSpeak®, NORC’s probability-based panel designed to be +representative of the U.S. household population. During the initial recruitment phase of the panel, +randomly selected U.S. households were sampled with a known, non-zero probability of selection from +the NORC National Sample Frame and then contacted by U.S. mail, email, telephone, and field +interviewers (face-to-face). The panel provides sample coverage of approximately 97 percent of the U.S. +household population. Those excluded from the sample include people with P.O. Box only addresses, +some addresses not listed in the USPS Delivery Sequence File, and some newly constructed dwellings. + +Interviews for this survey were conducted between April 23 and May 9, 2018, with young people age +15-34 and parents of young people age 15-26 representing the 50 states and the District of Columbia. +Interviews were conducted in both English and Spanish, depending on respondent preference. Adult +panel members age 18-34 were randomly drawn from AmeriSpeak, and 791 completed the survey—780 +via the web and 11 via telephone. Teen panel members age 15-17 were drawn from AmeriSpeak Teen, +and 148 completed the survey—145 via the web and 3 via telephone. Adult panel members who are +known parents of 15-17 year olds were randomly drawn from AmeriSpeak, and 104 completed the +survey—101 via the web and 3 via telephone. In addition, panel members age 30-76 were randomly +drawn to screen for parents of young adults age 18-26, and 503 completed the survey—469 via the web +and 34 via telephone. + +Among young people age 15-34, the final stage completion rate is 25.9 percent, the weighted household +panel response rate is 33.7 percent, and the weighted household panel retention rate is 87.2 percent, for +a cumulative response rate of 7.6 percent. The overall margin of sampling error among young people +age 15-34 is +/- 4.3 percentage points at the 95 percent confidence level, including the design effect. The +overall margin of sampling error among young people age 15-26 is +/- 5.6 percentage points at the 95 +percent confidence level, including the design effect. The margin of sampling error may be higher for +subgroups. + +Among parents of young people age 15-26, the screener completion rate is 27.9 percent, with an +incidence rate of 25.2. The interview completion rate is 95.6 percent, and the final stage completion rate +is 26.7 percent. The weighted household panel response rate is 33.7 percent, and the weighted +household panel retention rate is 87.2 percent, for a cumulative response rate of 7.8 percent. The +overall margin of sampling error is +/- 6.7 percentage points at the 95 percent confidence level, +including the design effect. The margin of sampling error may be higher for subgroups. + +Once the sample has been selected and fielded, and all the study data have been collected and made +final, a poststratification process is used to adjust for any survey nonresponse as well as any +noncoverage or under- and oversampling resulting from the study specific sample design. +Poststratification variables included age, gender, census division, race/ethnicity, and education. + +© Copyright 2018. The Associated Press and NORC + +June 2018 12 + + MTV/AP-NORC: COMPARING THE POLITICAL VIEWS OF YOUNG PEOPLE AND THEIR PARENTS’ GENERATION + +Weighting variables were obtained from the 2017 Current Population Survey. The weighted data +reflect the U.S. population of young people age 15-34 and parents of young people age 15-26. + +All analyses were conducted using STATA (version 14), which allows for adjustment of standard errors +for complex sample designs. All differences reported between subgroups of the U.S. population are at +the 95 percent level of statistical significance, meaning that there is only a 5 percent (or less) +probability that the observed differences could be attributed to chance variation in sampling. + +A comprehensive listing of all study questions, complete with tabulations of top-level results for each +question, is available on The AP-NORC Center for Public Affairs Research website: www.apnorc.org. + +For more information, email info@apnorc.org. + +CONTRIBUTING RESEARCHERS + +From NORC at the University of Chicago + +From The Associated Press + +Becky Reimer +Emily Alvarez +Liz Kantor +Trevor Tompson +Jennifer Benz + +Emily Swanson + +ABOUT THE ASSOCIATED PRESS-NORC CENTER FOR PUBLIC AFFAIRS +RESEARCH + +The AP-NORC Center for Public Affairs Research taps into the power of social science research and the +highest-quality journalism to bring key information to people across the nation and throughout the +world. + +■ + +The Associated Press (AP) is the world’s essential news organization, bringing fast, unbiased news +to all media platforms and formats. + +■ NORC at the University of Chicago is one of the oldest and most respected, independent research + +institutions in the world. + +The two organizations have established The AP-NORC Center for Public Affairs Research to conduct, +analyze, and distribute social science research in the public interest on newsworthy topics, and to use +the power of journalism to tell the stories that research reveals. + +The founding principles of The AP-NORC Center include a mandate to carefully preserve and protect +the scientific integrity and objectivity of NORC and the journalistic independence of AP. All work +conducted by the Center conforms to the highest levels of scientific integrity to prevent any real or +perceived bias in the research. All of the work of the Center is subject to review by its advisory +committee to help ensure it meets these standards. The Center will publicize the results of all studies +and make all datasets and study documentation available to scholars and the public. + +ABOUT MTV + +MTV is a global youth culture brand inspired by music. For more information, check out +mtvpress.com. MTV is a unit of Viacom Inc. (NASDAQ: VIAB, VIA) + +© Copyright 2018. The Associated Press and NORC + +June 2018 13 + diff --git a/q175/random_k4/random_3.md b/q175/random_k4/random_3.md index 7df7bc0e2df92b538e19432c90fe123f528730a0..89a0e0d7330b5b5f606ca129bb50b359ffb607a6 100644 --- a/q175/random_k4/random_3.md +++ b/q175/random_k4/random_3.md @@ -1,1755 +1,299 @@ -REPORT ON GEOLOGICAL FIELDWORK IN +Wikidata's SPARQL -TIRUVANNAMALAI, UTHANGARAI, +introduction for beginners -TIRUPATTUR, SALEM, PERAMBALUR, AND +CC BY-SA +Lucie-Aimée Kaffee -ARIYALLUR AREAS OF TAMIL NADU, INDIA +Frimelle +@lucielinde -Department of Geology, School of Earth Sciences + What is turtle? -Central University of Tamil Nadu, +RDF triple language -Thiruvarur, Tamil Nadu +Subject Predicate Object -Batch 2020- 2022 +as Web URI -14th - 19th March, 2022 +Alan Turing (Q7251) instance of (P31) human (Q5) -1 | P a g e +RDF + - Team Members +turtle +@prefix wd: -Faculty Members +@prefix wdt: -1. Dr. Balamurugan Guru +wd:Q7251 wdt:P31 wd:Q5 -2. Dr. Bankim Chandra Mahanta + What is SPARQL? -Students +RDF query language -Reg. No Student Name -P202801 Abhishek Roushan -P202802 Anish Kumar S -P202804 Anusmita Mandal -P202805 Bhagyashree Senapati -P202807 Devashish Murmu -P202809 Girindra Bora -P202810 Harikrishnan S -P202811 -Jananisri R A -P202812 Mydhili S B -P202813 Nivash R -P202815 Priyakshi Borgohain -P202816 Ramavath Lalu -P202817 Roshini C -P202818 Sachin Subhajit Jena -P202819 Subham Dutta -P202820 Tapas Ranjan Sahoo +to form complex queries with SELECT and retrieve data from an endpoint -Field Support and Guidance +Return all items, which are instance of human, limited to 200 items. -1. Geological Survey of India +PREFIX wd: -2. Tamil Nadu Mineral and Mines +PREFIX wdt: -3. Ponguru Magnesite Mines, Salem +SELECT ?subject WHERE { ?subject wdt:P31 wd:Q5 . } LIMIT 200 -4. Periyar University, Salem + Exploring Wikidata’s linked data -5. Govt. Arts College (7-Arts), Salem +Show all data for Alan Turing (Q7251) limited to 100 -6. Ramco Cement +On Wikidata Query Service Beta -2 | P a g e +Extract of the result - Table of Contents -List of Figures ......................................................................................................................................... 6 -Introduction ..................................................................................................................................... 8 -1. -Geology of Tamil Nadu .......................................................................................................... 8 +predicate -1.1 +rdfs:label -1.2 +rdfs:label -1.3 +object -Climate and Rainfall ............................................................................................................... 8 +Alan Turing -Scope and purpose of the field visit ......................................................................................... 9 +ﺞﻧروﺗ نﻻآ -1.4 Method of Investigation ........................................................................................................ 10 +skos:altLabel -1.4.1 Measuring Attitude of Beds .......................................................................................... 10 +Алан Цюрынг -1.4.2 + -1.4.3 + -1.4.4 +Turing’s field of work (P101) is computer science (Q21198) -1.4.5 + Exploring Wikidata’s linked data -1.4.6 +Show all fields of work of Alan Turing -Collecting Sample ......................................................................................................... 10 +On Wikidata Query Service Beta -Identifying Mineral Constituents................................................................................... 10 + On Wikidata Query Service Beta -Describing Lithology..................................................................................................... 10 +Result Result -Taking Photographs ...................................................................................................... 10 +object -GPS ............................................................................................................................... 10 + -1.5 + -Schedule for field visit .......................................................................................................... 10 +label -2. Detailed Filed Visit: DAY 1 (15/03/2022) .................................................................................... 12 -Area Name: Northern part of Southern Granulite Terrain, Harur-Uttangarai Molybdenum Belt -12 +computer science -2.1 +cryptanalysis -Location 1 (sheared zone) ............................................................................................................. 12 +Turing’s field of work (P101) is computer science (Q21198) and cryptanalysis (Q897511) -Location 2 and 3 (Non-sheared zone) .......................................................................................... 12 + Exploring Wikidata’s linked data -2.1.1 +Data linked to Alan Turing (Q7251) limited to 100 -Economic Importance ................................................................................................... 14 +On Wikidata Query Service Beta -2.2 +Extract of the result -Visit to GSI Drilling Site ....................................................................................................... 15 +subject -2.2.1 +predicate -Economics Important .................................................................................................... 15 + -2.3 + -Area Name: Mylonite Rock .................................................................................................. 16 + -Location 1 ..................................................................................................................................... 16 + -2.4 + -Area Name: GSI Camp Site .................................................................................................. 16 + -3. DAY 2 (16/3/2022) ....................................................................................................................... 17 -3.1 -Area Name: Tamin Vermiculite Mines and Exploration Plant Tamil Nadu Minerals and Mine -Corporation, Sebathur. ...................................................................................................................... 17 +* -Location 1 ..................................................................................................................................... 17 + -3.1.1 + Truthy and full statements -General Geology ........................................................................................................... 17 +represents statements in two formats -3.1.2 Mineralogical Characteristics ........................................................................................ 17 +truthy +contain only the statements of best rank -3.1.3 +best rank are all statements of preferred rank, if those exist or all statements of normal rank -Economic Importance ................................................................................................... 17 +qualifiers are ignored -3.2 -Area Name: Ultramafic Pyroxenite/Syenite Suite (Location 2), Core Syenite / Ultramafic -Alkaline Suite (Location 3), Kanjanur (Location 4) ......................................................................... 20 +Prefix wdt: (http://www.wikidata.org/prop/direct/) - e.g. wdt:P101 -Location 2 ..................................................................................................................................... 20 +full statements +represent all data about a statement -Location 3 ..................................................................................................................................... 20 +represented in a separate node with the unique id of the statement - e.g. q92741-7ED21209-4E23-4EB1-886A-772CC6A5350F -Location 2 ..................................................................................................................................... 20 +Prefix p: (http://www.wikidata.org/prop/) - e.g. p:P101 -3.2.1 + Full statements -General Geology ........................................................................................................... 20 +qualifier and references are objects of the full statement -3.2.2 Mineralogical Characteristics ........................................................................................ 20 +wds:Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 a wikibase:Statement ; -3.2.3 +wikibase:rank wikibase:NormalRank ; -Economic Importance ................................................................................................... 20 +psv:P1082 wdv:0419376cf1da9a8efcbb5ab7c87ded58 ; -4. DAY 3 (17/3/2022) ...................................................................................................................... 23 -Area Name: Sangakari Granite, Tuthipalayam ...................................................................... 23 +ps:P1082 10000 ; -4.1 +pqv:P585 wdv:04a483576f9618f0263dd20cb503ad32 ; -3 | P a g e +pq:P585 1680-01-01T00:00:00Z . - Location 1 ..................................................................................................................................... 23 +Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 (one statement for population of Berlin) is type of statement ; -4.1.1 +has a NormalRank ; (this statement wouldn’t show up in truthy since population has a prefered rank) -General Geology ........................................................................................................... 23 +the population (P1082) is connected to a full value ; -4.1.2 Mineralogical Characteristics ........................................................................................ 24 +the population is 10000 ; -4.1.3 +the qualifier for point (P585) in time is connected to a full value ; -Economic Importance ................................................................................................... 24 +the qualifier point in time (P585) implies this statement is valid for 1680 . -4.2 + Full statements -Area Name: Near Water Treatment Plant, Tidumal (Opp to Dhasampalam, TNWB) .......... 24 +qualifier and references are objects of the full statement -Location 2 ..................................................................................................................................... 24 +wds:Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 a wikibase:Statement ; -4.2.1 +wikibase:rank wikibase:NormalRank ; -General Geology ........................................................................................................... 24 +psv:P1082 wdv:0419376cf1da9a8efcbb5ab7c87ded58 ; -4.2.2 Mineral Characteristics ................................................................................................. 24 +ps:P1082 10000 ; -4.2.3 +pqv:P585 wdv:04a483576f9618f0263dd20cb503ad32 ; -Economic Importance ................................................................................................... 25 +pq:P585 1680-01-01T00:00:00Z . -4.3 +Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 (one statement for population of Berlin) is type of statement ; -Area Name: Sittampundi ....................................................................................................... 25 +has a NormalRank ; (this statement wouldn’t show up in truthy since population has a prefered rank) -Location 3 ..................................................................................................................................... 25 +the population (P1082) is connected to a full value ; -4.3.1 +the population is 10000 ; -General Geology ........................................................................................................... 25 +the qualifier for point (P585) in time is connected to a full value ; -4.3.2 Mineralogical Characteristics ........................................................................................ 25 +the qualifier point in time (P585) implies this statement is valid for 1680 . -4.3.3 + Full statements -Economic Importance ................................................................................................... 25 +qualifier and references are objects of the full statement -4.4 +wds:Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 a wikibase:Statement ; -Area Name: Thotiyanthotam ................................................................................................. 25 +wikibase:rank wikibase:NormalRank ; -Location 4 ..................................................................................................................................... 25 +psv:P1082 wdv:0419376cf1da9a8efcbb5ab7c87ded58 ; -4.4.1 +ps:P1082 10000 ; -General Geology ........................................................................................................... 26 +pqv:P585 wdv:04a483576f9618f0263dd20cb503ad32 ; -4.4.2 Mineral Characteristics ................................................................................................. 26 +pq:P585 1680-01-01T00:00:00Z . -4.5 +Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 (one statement for population of Berlin) is type of statement ; -Area Name: Nallur ................................................................................................................ 27 +has a NormalRank ; (this statement wouldn’t show up in truthy since population has a prefered rank) -Location 5 ..................................................................................................................................... 27 +the population (P1082) is connected to a full value ; -4.5.1 +the population is 10000 ; -General Geology ........................................................................................................... 27 +the qualifier for point (P585) in time is connected to a full value ; -4.5.2 Mineralization Characteristics ....................................................................................... 27 +the qualifier point in time (P585) implies this statement is valid for 1680 . -4.5.3 + Full statements -Economic Importance ................................................................................................... 27 +qualifier and references are objects of the full statement -5. DAY 4 (18/3/2022) ....................................................................................................................... 29 -Area Name: Ponguru Magnesite Mine .................................................................................. 29 +wds:Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 a wikibase:Statement ; -5.1 +wikibase:rank wikibase:NormalRank ; -Location 1 ..................................................................................................................................... 29 +psv:P1082 wdv:0419376cf1da9a8efcbb5ab7c87ded58 ; -5.1.1 +ps:P1082 10000 ; -5.1.2 +pqv:P585 wdv:04a483576f9618f0263dd20cb503ad32 ; -General Geology ........................................................................................................... 29 +pq:P585 1680-01-01T00:00:00Z . -Economic Importance ................................................................................................... 29 +Q64-01f1575b-4bf3-982b-34d6-102fbe912ae2 (one statement for population of Berlin) is type of statement ; -5.2 +has a NormalRank ; (this statement wouldn’t show up in truthy since population has a prefered rank) -5.3 +the population (P1082) is connected to a full value ; -5.4 +the population is 10000 ; -Department of Geology, Periyar University ........................................................................ 30 +the qualifier for point (P585) in time is connected to a full value ; -Area Name: Government Arts College, Salem-7 .................................................................. 31 +the qualifier point in time (P585) implies this statement is valid for 1680 . -Area Name: Yercaud, Sevarayan .......................................................................................... 32 + Example Query -Location 2 ..................................................................................................................................... 32 +Who’s birthday is today? -5.4.1 +P569: date of birth -General Geology ........................................................................................................... 32 +On Wikidata Query Service Beta -5.4.2 Mineral Characteristics ................................................................................................. 32 + Ressources -6. DAY 5 (18/3/2022) ....................................................................................................................... 34 -6.1 -Area Name: Uttatur Terrain, Cauvery Basin (Location 1), Karai Formation (Location 2), -Sathanur Petrified Wood (Location 3), Alathur-Ariyalur (Location 4) ............................................ 34 +Wikidata Query Service Beta -Location 1 ..................................................................................................................................... 34 +http://wdqs-beta.wmflabs.org/ -Location 2 ..................................................................................................................................... 34 +Mediawiki: RDF Dump Format -Location 3 ..................................................................................................................................... 34 +https://www.mediawiki.org/wiki/Wikibase/Indexing/RDF_Dump_Format -4 | P a g e +Prefixes used for Wikidata - Location 4 ..................................................................................................................................... 34 +https://www.mediawiki.org/wiki/Wikibase/Indexing/RDF_Dump_Format#Prefixes_used -6.1.1 +Mediawiki: Query Examples -Geology ......................................................................................................................... 34 +https://www.mediawiki.org/wiki/Wikibase/Indexing/SPARQL_Query_Examples -6.1.2 Mineralogical Characteristics: ....................................................................................... 34 +github: wikidata query rdf docs -6.1.3 +https://github.com/wikimedia/wikidata-query-rdf/tree/master/docs -Economic Importance: .................................................................................................. 34 +Wikidata turtle dumps -6.2 +https://dumps.wikimedia.org/wikidatawiki/entities/ -Area Name: National Fossil Park, Sathanur ......................................................................... 37 +W3C RDF Primer -Location ........................................................................................................................................ 37 - -6.3 - -Area Name: Disconformity between Karai shale and Kolakanatham sandstone .................. 38 - -Location ........................................................................................................................................ 38 - -7. DAY 6 (19/3/2022) ....................................................................................................................... 39 -Area Name: The Ramco Cements Limited (Office) (Location 1), Cement Mine (Location 2), -7.1 -Karambiyam (Location 4), Factory (Location 4) .............................................................................. 39 - -Location 1 ..................................................................................................................................... 39 - -Location 2 ..................................................................................................................................... 39 - -Location 3 ..................................................................................................................................... 39 - -Location 4 ..................................................................................................................................... 39 - -7.1.1 - -General Geology: .......................................................................................................... 39 - -7.1.2 Mineralogical Characteristics: ....................................................................................... 39 - -7.1.3 - -Economic Importance: .................................................................................................. 39 - -8. Conclusion .................................................................................................................................... 41 -Acknowledgement ................................................................................................................................ 42 - -5 | P a g e - - List of Figures - -Figure 1 Geological Map of Tamil Nadu (Geological Survey of India, 2006) ........................................ 9 - -Figure 2 (a) Dharmapuri Rift Zone Shear Zone showing the Quartz Vein, (b) Molybdenum (green - -colour) Present in Hornblende Gneiss .................................................................................................. 13 - -Figure 3 (a) Two sets of prominent foliation plane, (b) Quartz veins (Fractured body filled with quartz - -within a rock), (c) Cavity Filling (Hydrothermal type of deposit where the deposition of minerals takes - -place within the cavity), (d) Molybdenum with trace amounts of Pyrite (Mo is a strategic mineral) .. 14 - -Figure 4 (a) Diamond drill bit with the casing, (b) Core Sample in a box pattern, (c) Drilling site for - -Molybdenum, (d) Mylonitized Epidotite Hornblende Gneiss. .............................................................. 15 - -Figure 5 (a) & (b) Core samples after drilling kept in a book pattern at GSI camp, Uttangarai, (c) visible - -gneissose structure. ............................................................................................................................... 16 - -Figure 6 (a) Mica Mine, (b) Control Machine, Air Compressors ......................................................... 18 - -Figure 7 Mica of Grade 3, 4 and 5 ........................................................................................................ 18 - -Figure 8 (a) Grade 5 (Refined), (b) Grade 5 (Raw), (c) Grade 4 (Refined), (d) Grade 3 (Refined), (e) - -Grade 3 (Raw), (f) Waste Material ....................................................................................................... 19 - -Figure 9 (a) Syenite (Coarse-grained intrusive igneous rock similar to granite but is deficient in quartz), - -(b) Syenite (Coarse-grained intrusive igneous rock similar to granite but is deficient in quartz), (c) - -Crocodile weathering (Occurs in partially serpentinized rocks) ........................................................... 21 - -Figure 10 (a) Syenite boundary rock for the Ultramafic Syenite suite, (b) Syenite rock of core area in - -Ultramafic Syenite suite ........................................................................................................................ 22 - -Figure 11 Granite with hornblende and mica minerals ......................................................................... 23 - -Figure 12 Eclogite Garnet (Metamorphic rock hosted in the matrix of sodium-rich pyroxene) ........... 24 - -Figure 13 Anorthosite (Rock composed predominantly calcium-rich plagioclase feldspar) ................ 25 - -Figure 14 Structural Mapping in Anorthosite Complex ....................................................................... 26 - -Figure 15 Hands-on experience with Structural Mapping .................................................................... 26 - -Figure 16 Chromite Deposit .................................................................................................................. 27 - -Figure 17 (a) Pot hole (Depressions formed due to Eddy currents), (b) Ptygmatic Fold (Irregular gold - -found where particular competent layers are enclosed in a matrix), (c) Boudinage (Structure formed by - -extension and is cylindrical or lens like in shape) ................................................................................ 28 - -Figure 18 Mine visit and general insight of the Ponguru Magnesite Mine ........................................... 30 - -Figure 19 Different grades of Anorthosite Sample ............................................................................... 30 - -Figure 20 Discussion with Prof. S. Anbazhagan, Geology Department, Periyar University ................ 31 - -Figure 21 Interaction with the Department of Geology, Government Arts College Salem-7............... 31 - -Figure 22 (a) Rain print and Mud Crack (Sedimentary Structures), (b) Weathered Bauxite (Bauxite is a - -residual rock which high content of aluminium and here it is a low-grade bauxite deposit) ................ 32 - -Figure 23 Yercaud Bauxite mines ......................................................................................................... 33 - -6 | P a g e - - Figure 24 (a) Dr. Kumarvel, GSI explaining the Badland Topography, its mechanism of formation (b) - -Disconformity between Karai shale and Kolakanatham sandstone, (c) Dr. Kumarvel, GSI eplaining the - -idea of mapping and usage of toposheet during field study. ................................................................. 35 - -Figure 25 (a) Karai Formation (mostly clay dominated), (b) Charnockite (Metamorphic rock which - -mainly consists of orthopyroxene-bearing quartz-feldspar formed at high P/T condition and is found in - -granulite facies), (c) Charnockite (Metamorphic rock which mainly consists of orthopyroxene-bearing - -quartz-feldspar formed at high P/T condition and is found in granulite facies), (d) Badland Topography - -(Dry terrain where different agents extensively erode soft terrain ....................................................... 36 - -Figure 26 (a) Phosphatic Uttatur potatoes, (b) Sand dyke in Badland topography ............................... 37 - -Figure 27 (a) Petrified Tree Trunk, (b) Petrified wood of Villupuram district ..................................... 38 - -Figure 28 Disconformity between Karai shale and Kolakanatham sandstone indicates alternative - -transgression and regression. ................................................................................................................ 38 - -Figure 29 Ammonite with dendritic pattern (Shelled cephalopods belonging to the phylum Mollusca - -with some imprints of leaves) ............................................................................................................... 40 - -Figure 30 Interaction with the Geologist, Ramco Cement, Ariyalur .................................................... 40 - -7 | P a g e - - 1. Introduction - -1.1 Geology of Tamil Nadu - -The area is 80% covered by Archean to late Proterozoic crystalline rocks, while the remainder is - -covered by Phanerozoic sedimentary rocks, mainly along the coastal strip and in some inland river - -valleys. The hard rock terrain consists predominantly of charnockite and Khondalite groups and its - -migmatitic descendants, supracrustal sequences of the Sathyamangalam and Kolar Groups and the - -Peninsular Gneiss Complex (Bhavani Group) intruded by ultramafic complexes, basal dykes, granites - -and syenites. During the Archean and Proterozoic epochs, crystalline rocks were derived through a - -complex evolutionary history of multiple deformations, anatexis, intrusions, and multiphase - -metamorphic events. Carboniferous to Mio-Pliocene sedimentary rocks and coastal belts include - -fluvial, fluviomarine, marine sequences, and Quaternary sediments. Especially in the northern and - -central parts, the Pre-Cambrian terrane of Tamil Nadu is highly fractured and deeply faulted. - -Phanerozoic sediments are relatively unaffected and show well-preserved bedding planes with dips - -ranging from horizontal to 100°. On the other hand, the crystalline rocks had undergone multiphase - -deformation and metamorphism, resulting in well-developed foliation, i.e., S1, following the - -lithographic contacts in many places. Subsequent deformations have also induced the local development - -of new S-tissues. The NE regional structural trend is NNESSW, characterized by long linear canoe- - -shaped folds. Although the northwest is characterized by - -a regional structural trend near NS and - -multiphase folding has been recognized on-site, distinct regional structures have not been interpreted. - -In the central part, the regional structural trend EW often differs from ENE-WSW and WNW-ESE. - -South of the Tambaraparani River, i.e., the southernmost sector, shows pronounced NWSE structural - -granulation. Several shear zones were also recognized in the Precambrian. The terrain includes the - -Moyyar Shear Zone, Bhavani, Salem, Attur, Cauvery, Dharmapuri, Gangavalli and Achankovil. - -1.2 Climate and Rainfall - -The climate of Tamil Nadu is mainly tropical, i.e., it ranges from dry sub-humid to semi-arid and - -depends primarily on monsoon rains and hence is prone to drought when the monsoons fail. In May - -and June, which are the hottest months, daily maximum temperatures in Chennai rise on average - -around 100°F (38°C), while minimum temperatures drop on average around 80°F (20 °C). In December - -and January, the coldest months, temperatures generally rise by around 70°F (21°C) daily to mid-80s F - -(about 30°C). i.e., about 945 mm (37.2 in), of which 48% through the northeast monsoon and 32% - -through the southwest monsoon. - -Therefore, the state has two distinct rainy periods: The southwest monsoon from June to September, - -with strong southwest winds; the Northeast monsoon from October to December, with prevailing - -winds from the northeast; Mountainous and hilly areas. The extreme western part of the state mainly - -receives the most precipitation, while the southern and lower southeast regions receive the least - -precipitation. As the state depends entirely on rainfall to replenish its water resources, monsoon failures - -8 | P a g e - - lead to severe water shortages and drought. Tamil Nadu is divided into seven agroclimatic zones: - -northeast, northwest, west, south, high rainfall, high altitude hills and the Cauveri delta, which is the - -most fertile agricultural zone. - -Figure 1 Geological Map of Tamil Nadu (Geological Survey of India, 2006) - -1.3 Scope and purpose of the field visit - -The main aim of the field visit is to gain geological field exposure to explorations, mining, and different - -lithology. - -The objectives of the geological field visit are as follows: - -• To study and understand the litho units mutual relationship, characters, and associates. - -• To analyze the structures of the regions and their relationship to the regional tectonic - -settings. - -• To collect samples of various litho- units for further petrographic studies and other analyses. - -• To visit the real-time field survey and exploration sites. - -9 | P a g e - - • To visit the mining exploration, processing, grade classification and final product. - -• To study the stratigraphic sequence of the investigated area with the regional geology. - -• To prepare an overall geological perspective of the study area. - -1.4 Method of Investigation - -The following methodologies have been used during the fieldwork. - -1.4.1 Measuring Attitude of Beds - -While the exposure investigation was taken, the bed's strike, dip direction, and dip amount were - -measured using the Brunton compass and Clinometers. - -1.4.2 Collecting Sample - -Representative rock samples have been collected from all the study areas. - -1.4.3 Identifying Mineral Constituents - -The constituent minerals and grain length of rock specimens have been identified with the assistance of a - -lens. - -1.4.4 Describing Lithology - -The lithology of the exposures has been defined and accorded in the notebook. - -1.4.5 Taking Photographs - -Photographs of nicely uncovered geomorphic features, structures, vegetation and topography have - -been taken in every phase. All those records received using the above strategies from exposures have - -been cited in the notebook. - -1.4.6 GPS - -The global positioning system is a device which can determine the exact position of location of field - -investigation (as well as of the user) by signals from various satellites situated in space. The Latitude, - -Longitude and Elevation of its position can be known from it. It has some Expected Positional Error - -(EPE or accuracy) which tells accuracy of the data. In the area of no remarkable topography, it is an - -excellent way to find location instead of clinometers back bearing. - -1.5 Schedule for the field visit - -The field visit was initially planned for two weeks; however, due to the Covid-19 issue and research - -dissertation, the field visit was carried out for six days. The field visit started with the reconnaissance - -survey, sample collection, drilling operation, litho-log preparation, and ore estimation in the Uttangarai - -region with the support of the Geological Survey of India. In the following days, the field visit was - -carried out in the mica mines, Samalpatti carbonatite area, Sittampondi anorthosite complex, magnesite - -mine, Periyar University and Government Arts College of Salem, and Yercaud. The field visit ended - -with extensive work on the Gondwana and Cretaceous formations of Ariyalur and Perambalur Area - -with the support of GSI and Ramco Cement. - -10 | P a g e - - Table 1: Schedule of the Field Visit - -11 | P a g e - - 2. Detailed Filed Visit: DAY 1 (15/03/2022) - -2.1 Area Name: Northern part of Southern Granulite Terrain, Harur-Uttangarai - -Molybdenum Belt - -Location 1 (sheared zone) - -Latitude: 12° 9' 51'' - -Location 2 and 3 (Non-sheared zone) - -Longitude: 78° 27' 23'' - -Latitude: 12° 9' 55'' - -Longitude: 78° 27' 27'' - -The area is situated in the Dharmapuri and Krishnagiri districts of Tamil Nadu, bounded by the - -Uthangarai- Krishnagiri belt on the western side and the Madras belt on the eastern side. The NNE- - -SSW trending shear zone of 200 km long 50 km vast Mo province coincides with the Dharmapuri Rift - -Suture Zone (DRSZ)/ Dharmapuri Rift Zone (DRZ) bounded by Harur lineament in the east and - -Dharmapuri West lineament in the west. - -Dharmapuri Alkaline Carbonatite Belt, between Dharmapuri and Krishnagiri district, Tamil Nadu, - -is 200 km long (strike) and 40 km wide. It is bounded on the eastern side by Charnockite, the western - -side by older granulite terrain and; the central part by gneissic high grade and metamorphic terrain. This - -shear zone falls under the southern granulite terrain with a toposheet number 57L/8. On the northern - -side, the area is abrupting the Palakkad zone, and on the south side is the Cauveri shear zone. This shear - -zone is a failure rift shear zone mainly due to rifting and is trending parallel to NNE-SSW. Moreover, - -the attitude of one of the outcrops shows a trend of N60°E/40°/SE, which is controlled by structural - -fabric like sigmoidal. Also, the fracture plane is dipping towards the south and has mainly two to three - -prominent cleavages. Also, the area is a core part of the belt that indicates a highly mineralized zone has - -a host rock of Epidote Horblende Gneiss. Because of various activities like chloritization of the host - -rock and action of fluid along with P/T condition at greater depth, Mylonite is formed in this area. - -The area to the east is primarily composed of a charnokite cluster of fine to medium-grained rocks - -that are banded intermediate charnockite with linear bands of 2 pyroxene granulites, stone iron ore and - -pink quartzo-feldspathic granulite cut by little bodies of gabbro-norite-anorthosite-dunite-peridotite and - -younger basic dykes. Epidote amphibole gneiss rock with the variable composition of granodiorite - -to quartzdiorite is the characteristic rock sort that occupies most of the Mo province and is made by - -the gradual regression of charnockite. Enclaves of intermediate charnockite, pyroxene granulite, - -meta-pyroxenite, meta-gabbro, and - -conformable bands of quartzo-feldspathic gneiss occur at - -intervals epidote hornblende gneiss throughout the belt. Gabbro and dolerite embankments cross the - -terrain in WNW-ESE and NW-SE directions. - -The Mo belt also characteristically includes the Neoproterozoic ultramafic carbonatite-alkaline plutons - -as oval-shaped complexes, namely Odugattur, Rasimalai, Elagiri, Koratti, Samalpatti and Pakkanadu. - -12 | P a g e - - The alkaline complexes show well-preserved planar igneous features. Nonconforming bodies of - -felsites, aplitic syenite, pegmatoidal syenite and quartz veining also occur in this belt. Younger quartz - -reefs intersect all of the above rock types along strike from NNE-SSW. - -Figure 2 (a) Dharmapuri Rift Zone Shear Zone showing the Quartz Vein, (b) Molybdenum (green - -colour) Present in Hornblende Gneiss - -The lithology of the terrain to the east is the Charnokite group of rock-pyroxene granulite, magnetite - -quartzite, anorthosite-dunite-peridotite bodies and younger basic dykes, and to the west, Peninsular - -Gneissic Complex of older supracrustal of Satyamangalam group. The Presence of Epidote-hornblende - -gneiss with composition granodiorite to quartz-diorite is the major rock type present in the middle of - -the terrain and occupies a significant part of deposits for Mo province. The landscape is formed by - -retrograde metamorphism with Continental tectonics, so the presence of Epidote-hornblende gneiss is - -formed due to the progressive continuing retrogression of Charnokites. The belt consists of an alkali - -complex that shows a well-preserved igneous planer feature with variations of different veins. And the - -younger quartz dyke cut the alkali complex along the NNE-SSW direction and developed parallel - -foliation with a steep dip towards NW. The Mo mineralization occurs in the structural fabric and shear - -fabric in the sericitized two types of gneiss such as quartz feldspathic gneiss and hornblende gneiss. - -The shear zone is marked by the presence of homogeneous massive quartz with multiple shears with - -colour variation due to alteration because of the work of hydrothermal activity and can find some S-C - -fabrics. Towards the south side of the homogeneous massive quartz, the drilling was going on for Mo. - -The drilling is controlled by the Diamond drilling with an inclined dip of 45°. - -13 | P a g e - - Figure 3 (a) Two sets of prominent foliation plane, (b) Quartz veins (Fractured body filled with quartz - -within a rock), (c) Cavity Filling (Hydrothermal type of deposit where the deposition of minerals takes - -place within the cavity), (d) Molybdenum with trace amounts of Pyrite (Mo is a strategic mineral) - -2.1.1 Economic Importance - -In this shear zone, mainly Molybdenum (Mo) exploration is done with the help of the G2 stage of t h e - -investigation. This zone is characterized by well-developed quartz veins and sericitised gneiss, on - -the other hand. We know Mo is of fine flake- like granular aggregates with disseminated grains and - -is found in the fractures of quartz veins, so the significant concentration of Mo is located in the contact - -zones of quartz chains and the hydrothermally altered rocks within the veins. Thus these veins act as - -a carrier pattern for fluids, and the quartz vein type of deposit is the main type of deposit in this - -14 | P a g e - - zone. Also, Pyrite is seen as a secondary mineral due to sulphidisation. - -2.2 Visit to GSI Drilling Site - -The Harur Uttangarai Belt has been divided into 20 blocks, each 1.65 km long, for ease of operation. - -The blocks (1 to 11) are in the south of the Ponnaiyar River and are under the Harur Sector, and the - -blocks (12 to 20) are in the north of the Ponnaiyar River and are under the Uttangarai Sector. - -After detailed mapping and trenching, the continuity of the MSZ (Moyar Shear Zone) strike was - -established and then the underground exploration began by drilling that too in inclined holes for about - -45°. The first level holes were drilled to intersect the MSZ at 50m vertical depth. Generally, four stages - -of exploration are carried out, such as the G1, G2, G3 and G4 phases, of which G4 is the basic level. - -The G3 stage reduces the drilling interval by increasing the level. The G1 stage is the final level of - -exploration where the main part of the drilling is carried out, and less than 40° is generally avoided. - -Also, we have interpreted some data from one of the core samples to gain knowledge of how a sample - -is evaluated after collecting or extraction. - -Figure 4 (a) Diamond drill bit with the casing, (b) Core Sample in a box pattern, (c) Drilling site for - -Molybdenum, (d) Mylonitized Epidotite Hornblende Gneiss. - -2.2.1 Economics Important - -The primary ore deposits of the area are Mo with a base metal of Pb, Zn, Au, Co and Ni. The grade- - -15 | P a g e - - wise Mo can be explored up to a maximum of 4000 ppm-5000 ppm. The crustal abundance of Mo is - -1.5 ppm. - -2.3 Area Name: Mylonite Rock - -Location 1 - -Latitude: 12° 12' 41'' - -Longitude: 78° 28' 52'' - -The area is the core part of the Harur-Uttangarai Molybdenum Belt, which indicates the highly - -mineralized zone having Epidote Hornblende Gneiss host rock. Due to fluid activity and chloritization - -of the host rock with temperature variation at deeper depth, it changed to Mylonite. Hence, the rock - -type is called Mylonitized Epidotite Hornblende Gneiss. - -Figure 5 (a) & (b) Core samples after drilling kept in a book pattern at GSI camp, Uttangarai, (c) visible - -gneissose structure. - -2.4 Area Name: GSI Camp Site - -The area is the GSI campsite. Here the collection of core samples was there. The Harur-Uttangarai - -Molybdenum Belt was mapped by a Geochemical and Pedogeochemical survey. There are four stages - -of exploration cum drilling G1, G2, G3 and G4. The stages are called grades. The volume will reduce, - -the grade will increase closer to the ore body, and complexity will resolve in the highest exploration - -stage. The exploration is done up to G1. After that, mining will start. The core samples followed the - -book pattern for ease of calculation of core recovery, which was found to be 35%. The core sample is - -of the same composition as they have collected from the Harur- Uttangarai Molybdenum Belt. - -16 | P a g e - - 3. DAY 2 (16/3/2022) - -3.1 Area Name: TAMIN Vermiculite Mines and Exploration Plant Tamil Nadu Minerals - -and Mine Corporation, Sebathur. - -Location 1 - -Latitude: 12° 25' 12'' - -General Information - -Longitude: 78° 31' 36'' - -Total Area: 23.71 Hector - -Recovery: 9% - -Name of Mica: Vermiculite Mica - -Per day production: Approximately 5 tons - -Direction: NE-SW - -Grade: 3,4,5 - -Geological Reserves: 1895875 tons - -Working Principle - -▪ Step-1: Collection of raw material by plucking method- Grade1 and 2 - -▪ Step-2: Separate the material based on size - -▪ Step-3: Fry/roast the material with oil at the temperature of 6000 °C - -3.1.1 General Geology - -This mica mine is located in the village named Sebathur and has a total extension area of about 23.71 - -hectares. The direction of formation of the mineral deposit is NE-SW. The lithology of - -the mine - -area is Kankar, Carbonatite, Pyroxene, Calc Granulite, Syenite, Gneisses, and Vermiculite. - -3.1.2 Mineralogical Characteristics - -Mica is one of a group of hydrated potassium aluminium silicate minerals. It is a layered silicate species - -with a two-dimensional sheet or layered structure. Among the essential rock-forming minerals, mica is - -found in all three major rock types. Out of 28 known Mica groups, Vermiculite is predominantly found - -in this mica mine. - -3.1.3 Economic Importance - -This mine mainly consists of Vermiculite mica, of which the total geological reserve is about 18 lakhs - -95 thousand 875 tonnes. The per day recovery percentage is 9% from the raw material. Per tonne, - -mica is about 21 thousand, including GST, and per day production is exfoliated as 5 tonnes - -(approx.) The mica is refined into different grades, i.e. based on the size as per the customer's - -requirement. Basically, there are three grades, such as Grade 3, 4, and 5, depending on the size, i.e. - -Grade 5 is the finest of all. - -17 | P a g e - - Figure 6 (a) Mica Mine, (b) Control Machine, Air Compressors - -Figure 7 Mica of Grade 3, 4 and 5 - -18 | P a g e - - Figure 8 (a) Grade 5 (Refined), (b) Grade 5 (Raw), (c) Grade 4 (Refined), (d) Grade 3 (Refined), (e) - -Grade 3 (Raw), (f) Waste Material - -19 | P a g e - - Table 2 Different Grade Sizes - -Grade-1 - -Grade-2 - -Grade-3 - -Grade-4 - -Grade-5 - -Above 12 mm - -6 mm-12 mm - -3 mm-6 mm - -1 mm-3 mm - -Below 1 mm - -3.2 Area Name: Ultramafic Pyroxenite/Syenite Suite (Location 2), Core Syenite / - -Ultramafic Alkaline Suite (Location 3), Kanjanur (Location 4) - -Location 2 - -Latitude: 12° 20' 55'' - -Location 3 - -Latitude: 12° 17' 39'' - -Location 2 - -Latitude: 12° 15' 56'' - -3.2.1 General Geology - -Longitude: 78° 31' 12'' - -Longitude: 78° 25' 32'' - -Longitude: 78° 28' 03'' - -This area is mainly bounded with syenite and is trending towards NNE-SSW (L2). It is characterized - -by a Trachytic structure where plagioclase minerals seem like floating, i.e. plagioclase is of - -bigger grain size present in a fine grain matrix. Because of the shear zone activity, the veins of - -deposition are present in the area. - -3.2.2 Mineralogical Characteristics - -In Locations 2 and 3, within the syenite rock, there is a presence of Carbonatite in a few places. - -Carbonatites are mainly of two types i.e Fe-rich and Mg-rich, i.e., they are - -composed of - -ferromagnesian minerals and alkali feldspar. A particular group of alkali syenites is characterized by - -the presence of a feldspathoid mineral such as nepheline, leucite, and sodalite. In location 4, the hill - -was moderately covered with weathered Dunite Rock exhibiting a Crocodile weathering pattern. - -3.2.3 Economic Importance - -Mainly syenite is used for construction purposes such as flooring, interior decoration, building stone, - -etc. - -20 | P a g e - - Figure 9 (a) Syenite (Coarse-grained intrusive igneous rock similar to granite but is deficient in quartz), - -(b) Syenite (Coarse-grained intrusive igneous rock similar to granite but is deficient in quartz), (c) - -Crocodile weathering (Occurs in partially serpentinized rocks) - -21 | P a g e - - Figure 10 (a) Syenite boundary rock for the Ultramafic Syenite suite, (b) Syenite rock of core area in - -Ultramafic Syenite suite - -22 | P a g e - - 4. DAY 3 (17/3/2022) - -4.1 Area Name: Sangakari Granite, Tuthipalayam - -Location 1 - -Latitude: 11° 28' 37'' - -4.1.1 General Geology - -Longitude: 77° 59' 12'' - -The Sangakari granite is in the form of a batholith which may be extended more than 100 km. It is a - -large body of igneous rock formed beneath the Earth's surface due to the intrusion and solidification of - -magma. It is mainly composed of coarse-grained rocks such as granite (in this area) or granodiorite with - -a surface exposure of 100 km2 or more. It is formed when a number of plutons converge together to - -form a massive expanse of granitic rock. - -The Sangakari granite, also called Sankari – Tiruchengode (ST) granite, covering an area of about 520 - -km2 is emplaced at the intersection of the NE-SW trending Mettur lineament with the E-W trending - -Moyar-Bhavani-Attur Lineament. This area is formed by a deep crustal fault resulting from the surface - -to the upper mantle. It is geologically younger as it is derived from the Sangagiri batholith. The age of - -the Sangagiri batholith is more or less similar in age, i.e. 2.8 billion years as Pakkanadu pegmatite. This - -granite is generally massive, and no planar fabric is developed even at the contact. The rock is composed - -of mineral hornblende, orthoclase, biotite and mica. - -Figure 11 Granite with hornblende and mica minerals - -23 | P a g e - - 4.1.2 Mineralogical Characteristics - -Granite is a coarse-medium-grained intrusive igneous rock rich in quartz and feldspar. It is a common - -plutonic rock of the Earth's crust formed by the cooling of magma at great depths. In the area, we - -observed hornblende and biotite mica grains, giving us an idea of the rock being an Amphibolite. - -4.1.3 Economic Importance - -Granite is mainly used for construction purposes. - -4.2 Area Name: Near Water Treatment Plant, Tidumal (Opp to Dhasampalam, TNWB) - -Location 2 - -Latitude: 11° 13' 8'' - -4.2.1 General Geology - -Longitude: 77° 56' 13'' - -The area is mostly covered with exposures of high- grade metamorphic rocks, dominantly Eclogite - -Garnet. The general trend of the outcrops falls around 234°/77°/161° towards SE. - -Figure 12 Eclogite Garnet (Metamorphic rock hosted in the matrix of sodium-rich pyroxene) - -4.2.2 Mineral Characteristics - -It is a high P/T metamorphic rock that is essentially composed of more than 75% of garnet, mainly - -pyrope or Mg-rich almandine and omphacite or Na-Ca-Al-Mg clinopyroxene. Also, Rutile, kyanite, - -24 | P a g e - - and quartz are typically present. Thus, they are found primarily on blueschist facies. - -4.2.3 Economic Importance - -In medical industries and geological simulations. - -4.3 Area Name: Sittampundi - -Location 3 - -Latitude: 11° 13' 10'' - -4.3.1 General Geology - -Longitude: 77° 56' 05'' - -The Anorthosite complex in southern India formed around 2.5 Ga, implying lower crustal stabilization - -of the Dharwar Craton. The parental magma of Anorthosite was contaminated by ancient mafic lower - -crust. - -Figure 13 Anorthosite (Rock composed predominantly calcium-rich plagioclase feldspar) - -4.3.2 Mineralogical Characteristics - -Anorthosite is mainly a light- coloured, coarse-grained plutonic rock primarily composed of - -plagioclase, usually labradorite or bytownite, often with trace amounts of pyroxene. Other minerals like - -Olivine, amphibole, ilmenite, magnetite, and spinel are also found. - -4.3.3 Economic Importance - -This particular complex is essential for its corundum deposits excavated as Ruby for - -the gemstone - -industries. - -4.4 Area Name: Thotiyanthotam - -Location 4 - -Latitude: 11° 14' 29'' - -Longitude: 77° 57' 39'' - -25 | P a g e - - 4.4.1 General Geology - -Presence of Chromite. - -4.4.2 Mineral Characteristics - -It is an oxide mineral of the spinel group which contains Mg, Fe, Al and Cr in different proportions. - -Figure 14 Structural Mapping in Anorthosite Complex - -Figure 15 Hands-on experience with Structural Mapping - -26 | P a g e - - Figure 16 Chromite Deposit - -4.5 Area Name: Nallur - -Location 5 - -Latitude: 11° 55' 31'' - -Longitude: 77° 56' 54'' - -This area comes under the migmatite complex. Migmatite is considered to both igneous as well as - -metamorphic origins. The rock types may be quartzo-feldspathic gneiss, garnetiferous granite - -gneiss, garnet biotite gneiss with associated migmatites, hornblende gneiss, hornblende-biotite and - -quartz mica gneiss, biotite gneiss and epidote-chlorite biotite gneiss are grouped under migmatites and - -gneisses. - -4.5.1 General Geology - -It is a Migmatitic Complex formed by multiple deformations because of repeated metamorphism. - -The complex is formed as dykes, sills and ultramafic bodies. The rock type consists of feldspar and - -quartz-like igneous minerals and gneiss. The exposure had many folded patterns caused during lava flow - -along with some boudinage structures. Also, it was an old river bed interpreted from numerous potholes. - -4.5.2 Mineralization Characteristics - -The rock is mainly composed of alternating layers of quartz, feldspar and biotite and is a type of high- - -grade metamorphic rock formed under extreme P/T conditions. - -4.5.3 Economic Importance - -As a building stone and sometimes as a polished ornament. - -27 | P a g e - - Figure 17 (a) Pot hole (Depressions formed due to Eddy currents), (b) Ptygmatic Fold (Irregular gold - -found where particular competent layers are enclosed in a matrix), (c) Boudinage (Structure formed by - -extension and is cylindrical or lens like in shape) - -28 | P a g e - - 5. DAY 4 (18/3/2022) - -5.1 Area Name: Ponguru Magnesite Mine - -Location 1 - -Latitude: 11° 41' 52'' - -Longitude: 78° 07' 19'' - -• - -• - -• - -• - -• - -• - -• - -• - -➢ - -➢ - -➢ - -Name of Mines- Sri Ponguru Magnesite Mines - -Mineral- Magnesite associate with dunite - -Area- 77.5Hector - -Types of deposit- vein type deposit - -Method of mining- Open cast mining - -Rate of recovery- 6% - -Drilling types- Two types of Jack Hammer drilling and Rotary cum Percussive drilling - -Explosive – NH4NH3 Uses - -Used as refractory material - -Use as a catalyst and filler in the production of rubber - -Also used as fertilizers - -5.1.1 General Geology - -This mine was established in 1963 in Jagir, Ammapalyan Salem Taluk district, where Magnesite - -deposits are found in an area of 200 acres. The method of mining that is operated is Open cast - -Drilling. The height of a particular bench of 6 m is maintained along with the width. Mainly in - -this mine, the chemical explosive component is used, and here it has a heat capacity of about 1500°C. - -Mainly two types of drilling occur, i.e. Jack and hammer and Wagon drill. Also, other drilling like - -Rotary drilling, Percussive and Rotary-Percussive is being performed. Segregation by hand plucking - -is also seen inside the mine after blasting in a potential area. - -The country rock of the mine is dunite, i.e. the deposit occurs in an Ultrabasic Rock. Generally, - -magnesite is found in banded iron formations and sedimentary rocks and the abundance we observed - -the occurrence of magnesite mainly in a vein type of deposit. It is a magnesium carbonate mineral - -that usually forms during the alteration of Mg-rich rocks or carbonate rocks by chemical weathering. - -It is found in the Calcite group of minerals. - -5.1.2 Economic Importance - -The recovery percentage of the mine is only 6%, and if the deposit has less than 3% of silica, then only - -it is considered a high-grade deposit. Also, the dump should not be more than 30 m to have a - -substantial deposit. Thus, it is mainly used in steel furnaces and Ceramics. - -29 | P a g e - - Figure 18 Mine visit and general insight of the Ponguru Magnesite Mine - -5.2 Department of Geology, Periyar University - -At Periyar University, we interacted well with Dr. S Anbazhagan, an expert - -in Remote Sensing and - -GIS Applications, Planetary Remote Sensing, Natural Disaster Management and mitigation. During - -the discussion, he explained the Lunar Soil Simulant for Chandrayan II mission programme, where the - -Geology Department of Periyar University plays a significant role by contributing to the National - -Planetary Exploration Programme. A rover schedules the southern side of the Lunar highland; based - -on this; a satellite image has been created by IRS P6, which is then compared with the - -Anorthosite Complex of the Sittampundi area. He shared his work experience and involvement with that - -project and how he collected samples and prepared the patent. Also, samples of Anorthosite have been - -shown to us, which have been collected for the ongoing research work. - -Figure 19 Different grades of Anorthosite Sample - -30 | P a g e - - Figure 20 Discussion with Prof. S. Anbazhagan, Geology Department, Periyar University - -5.3 Area Name: Government Arts College, Salem-7 - -We had a good discussion with Dr. M. Jayabalan (Associate Professor and Head) and Dr. V. - -Thirukumaran (Associate Professor). Dr. Jayabalan and Dr. Thirukumaran have shared the work - -experience of their field visit and explained the geological features like structure, fossils content and - -lithology of Salem district and nearby areas such as Sittampundi Anorthite Complex, Samalpatti - -carbonatite and different mines present over Salem such as Yercaud Bauxite mine and Ponguru - -Magnesite Mines. They showed their collection of different megascopic rock samples and fossils from - -other states. - -Figure 21 Interaction with the Department of Geology, Government Arts College Salem-7 - -31 | P a g e - - 5.4 Area Name: Yercaud, Sevarayan - -Location 2 - -Latitude: 11° 48' 53'' - -Longitude: 78° 13' 12'' - -Bauxite is formed as a residual deposit; it is found in Shevaroy of Yercaud. Here we can find that the - -parent rock is charnockite and granulite; these rocks are formed at the lower crust and then uplifted - -because of their low density. The entire Yercaud hill is made up of these rocks. - -The pink colour in the bauxite is due to the Iron content in the rock; high the iron, the more the colour. - -The bauxite formed here is due to weathering. Both chemical and physical weathering work in the - -formation of bauxite from charnockite. - -5.4.1 General Geology - -The deposit mainly takes place at a certain height as it is temperature appropriate and is studied in a - -paleoclimatic condition. Mostly the structure of the rock is oolitic. Other structures like Rain prints and - -mud cracks are observed in the area. - -Figure 22 (a) Rain print and Mud Crack (Sedimentary Structures), (b) Weathered Bauxite (Bauxite is a - -residual rock which high content of aluminium and here it is a low-grade bauxite deposit) - -5.4.2 Mineral Characteristics - -Low-grade bauxite is found in the area with the parent rock of Charnockite and Granulite. From - -Laterite, primarily Geothite is formed, and finally, it gives rise to bauxite, a residual rock, i.e., it - -forms when laterite soils are severely leached of silica and other soluble materials in a wet tropical - -32 | P a g e - - or subtropical climate. It is a mixture of hydrous aluminium oxides, aluminium hydroxides, clay - -minerals, and insoluble materials such as quartz, hematite, magnetite, siderite, and goethite. The - -aluminium minerals in bauxite can include gibbsite Al(OH)3, boehmite AlO(OH), and diaspore, - -AlO(OH). - -Figure 23 Yercaud Bauxite mines - -33 | P a g e - - 6. DAY 5 (18/3/2022) - -6.1 Area Name: Uttatur Terrain, Cauvery Basin (Location 1), Karai Formation - -(Location 2), Sathanur Petrified Wood (Location 3), Alathur-Ariyalur (Location 4) - -Location 1 -Latitude: 11° 04' 20'' - -Location 2 -Latitude: 11° 07' 29'' - -Location 3 -Latitude: 11° 09' 40'' - -Location 4 -Latitude: 11° 07' 18'' - -6.1.1 Geology - -Longitude: 78° 52' 09'' - -Longitude: 78° 54' 26'' - -Longitude: 78° 58' 35'' - -Longitude: 78° 56' 02'' - -It is Precambrian Basement in the southern granulite terrain. Mainly this area is Karai formation - -of Uttatur Group where major transgression has occurred below 150 m water depth and ranges from - -Lower Cretaceous to Upper Cretaceous. It falls under the Cauvery Basin, a sedimentary basin and sub- - -divided into Pondicherry, Tiruchirapalli and Ariyalur sub-basins. The trend of one of the outcrops that - -we measured in the area is N40E/N130/SE. In the area, we observed two types of terrains, i.e. hard rock - -and soft rock and the presence of soft rock, mostly clay, indicates that the area was filled with water or - -is used as a water reservoir. In location 2, we mainly observed badland topography which is calcic - -rich and plays a critical role because there cannot be any activity. Also, we observed some sand dykes - -that usually form late than the mudstone. In Location 3, we observed petrified wood about 90my old, - -formed in a shallow marine to fluvial interphase. In Location 4, we observed contact zones, i.e. the - -formation of contact between Karai and Kollokantham sandstone. We know there are four - -unconformities, and in the area we mainly observed only disconformity, which is an unconformity - -between parallel layers of sedimentary rocks representing a period of erosion or non-deposition. - -6.1.2 Mineralogical Characteristics: - -Karai Formation is predominantly composed of clay. The parent rock in the area is Charnockite which - -has quartz, feldspar and pyroxene. Especially in Location 2, we observed that it has mudstone with - -intercalation of siltstone. The area was filled with phosphatic nodules or commonly known as Uttatur - -Potato, along with the presence of gypsum. - -6.1.3 Economic Importance: - -The area is abundantly filled with fossils, mainly belemnites, indicating the area to be a marine - -environment. It is an important food source for many marine creatures and played an important role in - -restructuring marine ecosystems after the Triassic–Jurassic extinction event. The presence of phosphatic - -nodules, spherical in shape, contains fossils in it, supplies nutrients, is highly organic and is upwelling - -34 | P a g e - - in nature. - -The Mesozoic rocks of southern India are exposed mainly in the Cauvery Basin, divided into three sub- - -basin, and Ariyalur is one of them. The importance of these Cretaceous deposits lies in the development - -of their diverse lithofacies, abundance and variety of fossils and wide geographical distribution in the - -region. The Cauvery basin cover 1/3rd part of the landmass and 2/3rd of the offshore. The main structural - -framework of the Cauvery basin is of the horst and graben type. The area is located at Uttattur in the - -Perambalur district. The Uttattur Group is divisible into Lower Arogyapuram Formation (Albian), - -Middle Maruvattur Formation (Late Albian) made up of limestone, mud and clay and Upper Karai - -Formation (Late Albian – early Turonian) consisting of gypseous clay and sandstone with lenses of - -limestone. The presence of alternative shale, limestone and clay are due to sea level changes in 110ma - -with many transgression and regression. The area is dominant in Non-conformity. - -Badlands are dry terrain, where softer sedimentary rocks are clay-rich soils that have been extensively - -Eroded by wind and water. They are characterized by steep slopes, minimum human habituation, - -minimum vegetation and high drainage density. The general dip of the area is toward NE at a low angle. - -The terrain highlands are due to their highly eroding nature. This area belongs to a past enclosed bay - -and experienced a huge deposition of salt evaporation. It is a highly eroded area, and a greater number - -of gullies are formed, joining the mainstream, or these sediments are rich in gypsum; the region is - -favourable for cotton cultivation. Here, abundantly found phosphatic nodules known as ‘Uttatur - -potatoes’ with an outer shell of chalk and inner phosphate forming a potato-like structure. - -Figure 24 (a) Dr. Kumarvel, GSI explaining the Badland Topography, its mechanism of formation (b) - -Disconformity between Karai shale and Kolakanatham sandstone, (c) Dr. Kumarvel, GSI eplaining the - -idea of mapping and usage of toposheet during field study. - -35 | P a g e - - Figure 25 (a) Karai Formation (mostly clay dominated), (b) Charnockite (Metamorphic rock which - -mainly consists of orthopyroxene-bearing quartz-feldspar formed at high P/T condition and is found in - -granulite facies), (c) Charnockite (Metamorphic rock which mainly consists of orthopyroxene-bearing - -quartz-feldspar formed at high P/T condition and is found in granulite facies), (d) Badland Topography - -(Dry terrain where different agents extensively erode soft terrain - -These marine sediments belong to the Uttatur formation. The Uttatur beds consist of fine silts, - -calcareous shales and sandy Clays containing ferruginous, phosphatic and calcareous nodules. The clay - -36 | P a g e - - is often streaked with yellow and brown ferruginous stains. Phosphatic minerals first cover the area, - -and then the area is reached by lime. The area is called Badlands topography. Sand dyke is found in the - -area. - -Figure 26 (a) Phosphatic Uttatur potatoes, (b) Sand dyke in Badland topography - -6.2 Area Name: National Fossil Park, Sathanur - -Location - -Latitude: 11° 09' 49'' - -Longitude: 78° 58' 08'' - -Sathanur is a National Geo-heritage monument of India located at Sathanur in Perambalur district, - -Tamil Nadu. The petrified tree trunk is believed to be over 120 million years old, and it was found by - -M.S. Krishnan in 1940. It is considered to be evidence of the presence of a sea during the cretaceous - -period in this area. This fossil is of a conifer and measures around 18 metres long. The sand deposition - -is preserved, and the tree's tissue is removed and replaced by silica materials, but the cell structure is - -still sustained and now formed as a rock. This fossil tree is a rare National asset collection of specimens. - -Two fossil woods of 20ma are also present in that park, collected from Thriuvakkarai of Villupuram - -district. - -37 | P a g e - - Figure 27 (a) Petrified Tree Trunk, (b) Petrified wood of Villupuram district - -6.3 Area Name: Disconformity between Karai shale and Kolakanatham sandstone - -Location - -Latitude: 11° 07' 18'' - -Longitude: 78° 56' 02'' - -An exposure showed the abrupt contact zone between Karai- shale and Kolakanatham- sandstone. - -The exposure was so clear that we could recognize the shale and sandstone contact zone; in other - -words, it was a disconformity. - -Figure 28 Disconformity between Karai shale and Kolakanatham sandstone indicates alternative - -transgression and regression. - -38 | P a g e - - 7. DAY 6 (19/3/2022) - -7.1 Area Name: The Ramco Cements Limited (Office) (Location 1), Cement Mine - -(Location 2), Karambiyam (Location 4), Factory (Location 4) - -Location 1 - -Latitude: 11° 07' 12'' - -Location 2 - -Latitude: 11° 09' 31'' - -Location 3 - -Latitude: 11° 10' 34'' - -Location 4 - -Latitude: 11° 10' 25'' - -Longitude: 79° 08' 51'' - -Longitude: 79° 10' 45'' - -Longitude: 79° 05' 49'' - -Longitude: 79° 05' 42'' - -The area is a part of Ramco Cement limited; it is an open-cast mine of limestone associated with fossils - -deposit. The Ariyalur Ramco cement factory is of sedimentary hosted in the case of Virudhunagar and - -Thoothukudi of meta sediment deposits. The lithology of the area is Limestone, ferruginous Limestone, - -Calc Sandstone and Conglomerate. The primary fossil contents are Ammonite, Gryphaea, - -Stigmatophygus, and Pecten. The major composition for Cement is limestone CaCo3(96%), Iron ore - -Fe2O3(2%), Al2O3 (2%) and 1% of burning material. - -7.1.1 General Geology: - -The Ramco Cement has 6-7 plants, of which three are present in the state of Tamil Nadu itself, and this - -mine, i.e. situated in Ariyalur, is of sedimentary type. The Ariyalu Ramco Cement has two areas, i.e. - -Virudhanagar, which is sedimentary hosted, and Thoothukudi, which is meta sediment hosted. The area - -is a metasedimentary basin, part of Ramco Cement limited, and is associated with many fossils. The - -major type of mining in this area is open cast mining. - -Location 3, the Karambiyam area belongs to the Tiruchirapalli group and is a protected area where we - -observed prominent ammonite fossils. - -7.1.2 Mineralogical Characteristics: - -This area mainly consists of limestone and calcite; the parent rock is mostly Pyroxene Garnet. It is of - -the upper Cretaceous period with a lithology of Limestone, Ferruginous Limestone, Calc Sandstone, - -and Conglomerate. - -The Cement composition is mostly limestone of about 96%, Iron Ore of 2%, Aluminium oxide of 2% - -and burning material of 1%. - -7.1.3 Economic Importance: - -Many fossils, mainly of the Cretaceous period, are found in the area, such as Ammonite, Gryphaea, - -Pecten, and Stigmatophygus, which help in excellent stratigraphy and indicate the presence of marine - -sedimentary rock. - -39 | P a g e - - Figure 29 Ammonite with dendritic pattern (Shelled cephalopods belonging to the phylum Mollusca - -with some imprints of leaves) - -Figure 30 Interaction with the Geologist, Ramco Cement, Ariyalur - -40 | P a g e - - 8. Conclusion - -Geological fieldwork is an integral part of the curriculum, which helps in enhancing theoretical - -knowledge in the practical world. - -Our field was mainly based in geological domain areas of Tamil Nadu, where we visited some of the - -mines, such as Mica Mines and Magnesite Mines. We also observed various rock types, with few - -primary and secondary structures, along with some economically essential minerals and fossils. We - -also visited Periyar University and Arts and Science College, Salem, where we were introduced to - -various new ideas and learned the importance of the field area. - -Although Geologists are not perfect, as there is a saying that each one of the Geologist has got their - -fault but we the budding Geologist, still tried our best to complete the fieldwork successfully in time - -with much enthusiasm and fill ourselves with much knowledge and holding a belief of learning - -more in the near future. - -41 | P a g e - - Acknowledgement - -Firstly, we would like to thank God, who gave me this opportunity to participate and perform on such - -a promising trip. This fieldwork was conducted from 14th to 19th March 2022. We would like to express - -our gratitude to the authorities of the Central University of Tamil Nadu for permitting us to conduct the - -Geological Field Work in and around different districts of Tamil Nadu. Our utmost thanks to Dr.Guru - -Balamurugan, Head of the Department of Geology and Dr. Bankim Chandra Mahanta, Assistant - -Professor, Department of Geology, for making all the necessary arrangements for the field study. - -We express our earnest gratitude to the Geological Survey of India (GSI), Periyar University and - -Government College of Arts and Science, Ramco Cements, Ariyalur, Ponguru Magnesite, Salem, - -Sevattur Mica Mine for giving us their valuable time and guidance throughout the fieldwork and - -providing us the knowledge and helping us in understanding all the necessary terms related to the - -fieldwork. - -At last, we would like to thank all the faculty members of the Department of Geology, CUTN, for their - -continuous support. - -42 | P a g e +http://www.w3.org/2007/02/turtle/primer/ diff --git a/q175/random_k4/random_4.md b/q175/random_k4/random_4.md index 87d1d0d34b01f6239ce3914346674fc098f98e9c..c628675fbe529d89838de1ae7c364d79b8d4ce01 100644 --- a/q175/random_k4/random_4.md +++ b/q175/random_k4/random_4.md @@ -1,1731 +1,1223 @@ -DRDO -NEWSLETTER -A Monthly Bulletin of Defence Research and Development Organisation -https://www.drdo.gov.in ISSN: 0I9S7S1N-:4 0397911-4391 APRIL 2026 | VOLUME 46 | ISSUE 4 -IN-FLIGHT RELEASE TRIALS OF -INDIGENOUS AIR DROPPABLE CONTAINER -‘ADC-150’ FROM THE P8I AIRCRAFT - -DRDO NEWSLETTER -Editor-in-Chief: Kiran Chauhan -Associate Editor-in-Chief: Sudhanshu Bhushan -APRIL 2026 | VOLUME 46 | ISSUE 4 -Editor: Dipti Arora -Design & Pre-press: Raj Kumar 46th Year of Publication -Printing: Rajesh Kr Singh -Distribution: Pratyaksh Sharma -LABORATORY CORRESPONDENTS -Agra : Shri Gayasuddin Quraishi, Aerial Delivery Research and Development establishment (ADRDE) -Ahmednagar : Shri RA Shaikh, Vehicle Research and Development Establishment (VRDE) -Ambernath : Dr Ganesh S Dhole, Naval Materials Research Laboratory (NMRL) -Balasore : Shri Mrinal Goswami, Integrated Test Range (ITR) -Shri Ratnakar S Mohapatra, Proof & Experimental Establishment (PXE) -Bengaluru : Shri Sriram GN, Aeronautical Development Establishment (ADE) -Smt MR Bhuvaneswari, Centre for Airborne Systems (CABS) -Smt Faheema AGJ, Centre for Artificial Intelligence & Robotics (CAIR) -Dr Josephine Nirmala M, Combat Aircraft Systems Development & Integration Centre (CASDIC) -Shri Jitender Kumar & Shri Vedang Chauhan, Centre for Military Airworthiness & Certification (CEMILAC) -Dr Sanchita Sil & Dr Sudhir S Kamble, Defence Bioengineering & Electromedical Laboratory (DEBEL) -Dr V Senthil, Gas Turbine Research Establishment (GTRE) -Smt Saima Bashir, Electronics & Radar Development Establishment (LRDE) -Ms Mita Jana, Microwave Tube Research & Development Centre (MTRDC) -Chandigarh : Dr Pal Dinesh Kumar, Terminal Ballistics Research Laboratory (TBRL) -: Dr Anuja Kumari, Defence Geoinformatics Research Establishment (DGRE) -Chennai : Shri K Anbazhagan, Combat Vehicles Research & Development Establishment (CVRDE) -Dehradun : Shri Sachin Oberai, Defence Electronics Applications Laboratory (DEAL) -Shri JP Singh, Instruments Research & Development Establishment (IRDE) -Delhi : Shri Vikas Kashyap, Centre for Personnel Talent Management (CEPTAM) -Shri Vishal Dwivedi, Centre for Fire, Explosive & Environment Safety (CFEES) -Dr KP Mishra, Defence Institute of Physiology & Allied Sciences (DIPAS) -Shri Santosh Kumar Choudhury, Defence Institute of Psychological Research (DIPR) -Dr Navin Kumar Soni, Institute of Nuclear Medicine and Allied Sciences (INMAS) -Dr Sujata Dash, Institute for Systems Studies & Analyses (ISSA) -Ms Sugandha Aggarwal, Recruitment & Assessment Centre (RAC) -Shri Ashok Kumar, Scientific Analysis Group (SAG) -Dr Rupesh Kumar Chaubey, Solid State Physics Laboratory (SSPL) -Gwalior : Dr MK Meghvansi, Defence R&D Establishment (DRDE) -Haldwani : Dr Atul Grover, Defence Institute of Bio-Energy Research (DIBER) -Hyderabad : Dr A Nagendranath, Advanced Systems Laboratory (ASL) -Shri Srinivas Juluru, Defence Research and Development Laboratory (DRDL) -Shri Ch Narasimhachari, Defence Electronics Research Laboratory (DLRL) -Shri S Shashi Nath, Defence Metallurgical Research Laboratory (DMRL) -Jagdalpur : Shri Khilawan Singh, SF Complex (SFC) -Jodhpur : Shri DK Tripathi, Defence Laboratory (DL) -Kanpur : Dr Mohit Katiyar, Defence Materials & Stores Research & Development Establishment (DMSRDE) -Kochi : Smt Letha MM, Naval Physical & Oceanographic Laboratory (NPOL) -Leh : Dr Dorjey Angchok, Defence Institute of High Altitude Research (DIHAR) -Mussoorie : Shri Sunil Bhandari, Institute of Technology Management (ITM) -Mysuru : Dr M Palmurugan, Defence Institute of Bio-defence Technologies (DIBT) -Nasik : Shri Ashutosh Sharma, Advanced Centre for Energetic Materials (ACEM) -Pune : Shri Ajay K Pandey, Armament Research and Development Establishment (ARDE) -Dr Ganesh Shankar Dombe, High Energy Materials Research Laboratory (HEMRL) -Dr Anoop Anand, Research and Development Establishment (E) (R&DE) -Tezpur : Dr KS Nakhuru, Defence Research Laboratory (DRL) -Visakhapatnam : Smt Jyotsna Rani, Naval Science & Technological Laboratory (NSTL) -APRIL 2026 2 - -DRDO NEWSLETTER -Contents -COVER STORY~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~4 -INNOVATIONS~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~5 -INFRA DEVELOPMENT~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~5 -MOU~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~6 -EVENTS~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~7 -HRD ACTIVITIES~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~19 -RAJBHASHA ACTIVITIES~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~28 -SPORTS ACTIVITIES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~30 -VISITS~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~31 -Please mail your feedback and suggestions at: -director.desidoc@gov.in; drdonl.desidoc@gov.in; -Contact at: 011-23902403; 23902472; Fax: 011-23819151 -3 APRIL 2026 - -COVER STORY DRDO NEWSLETTER -IN-FLIGHT RELEASE TRIALS OF INDIGENOUS AIR DROPPABLE -CONTAINER ‘ADC-150’ FROM THE P8I AIRCRAFT -Defence Research & for providing quick response Bengaluru, provided the flight -Development Organization to naval vessels under distress, clearance and certification, and -(DRDO) and the Indian Navy needing critical stores/equipment, Defence Research & Development -jointly conducted four successful medical assistance, etc. at blue sea Laboratory (DRDL), Hyderabad -in-flight release trials of the deployed far from the coast. provided the instrumentation -indigenous Air Droppable The Naval Science & support for the trials. To meet -Container ‘ADC-150’ from the Technological Laboratory the requirement of the Indian -P8I aircraft between February 21 (NSTL), Visakhapatnam is the Navy, the ADC-150 system for the -to March 01, 2026, at different nodal laboratory for the activity. P8I aircraft was developed and -extreme release conditions. Aerial Delivery Research & qualified in a short timeframe. As -Indigenously designed and Development Establishment all the developmental flight trials -developed to deliver 150 kg (ADRDE), Agra has developed have been completed successfully, -payload, the Air Droppable the parachute system and Centre the system is expected to be -Container enhances the naval for Military Airworthiness inducted into the Indian Navy -operational logistics capabilities & Certification (CEMILAC), soon. -APRIL 2026 4 - -INNOVATIONS/INFRA DRDO NEWSLETTER -INDIGENOUS TACTILE SENSOR FOR STRATEGIC ROBOTIC -APPLICATIONS -Research and Development N and are well evaluated for handed over by DIAT to R&D(E) -Establishment (E) (R&D(E)), thousands of operation cycles for on 9 March 2026 in an event -Pune, along with Defence Institute their repeatability and sensitivity. that was graced by the presence -of Advanced Technology (DIAT), a The sensor arrays are supported of Dr Samir V Kamat, Secretary -University supported by Ministry by an electronic data acquisition DD R&D & Chairman DRDO -of Defence (MoD), has developed system and python programming and Dr BHVS Narayana Murthy, -an indigenous tactile sensor for software. The sensors are Vice Chancellor, DIAT. The -strategic robotic applications. developed by the group led by Prof. event included demonstration -These sensors are capable to Sangeeta Kale from DIAT under of the developed sensor product, -handle delicate and lightweight the guidance of Dr SE Talole and which will be now integrated -payloads along with heavy objects Dr Kiran Akella from R&D(E). with the humanoid system being -using its gripping and slip- The sensor arrays along with developed under STAR project of -detection capability. The sensors its electronics interface were R&D(E). -shall give the future robots a -sense of touch. The technology is a -precursor to develop artificial skin -for future robots. Currently being -deployed in DRDO humanoid -robot grippers under the STAR -project, the sensors can grip and -hold objects using piezo-resistive -materials which are made of -polymer composites. They handle -forces in the range of 1 N to 50 -INFRA DEVELOPMENT -INAUGURATION OF NEW TELEMETRY INFRASTUCTURE -AT ITR, CHANDIPUR -A new Telemetry Centre design of the project and wished breakthrough for the telemetry -with state-of-the-art facilities more innovation and technological team of ITR. -was inaugurated in Integrated -Test Range (ITR), Chandipur. -Shri U Raja Babu, Director General -(MSS) in the august presence of -Shri Anindya Biswas, Director, -RCI, Shri K Suchendar, Director, -ITR, Shri M Veera Sankar, IDSE -& Chief Engineer, inaugurated the -facility. During the inauguration -Chief Guest Shri Raja Babu lauded -the planning and forward-thinking -5 APRIL 2026 - -MOU DRDO NEWSLETTER -DGRE INKS MOU WITH SAC-ISRO -Defence Geoinformatics DGRE and Nilesh M Desai, The MoU will strengthen the -Research Establishment Director, SAC signed the pact at ongoing and future research -(DGRE), Chandigarh, and Space SAC (ISRO) in Ahmedabad on 25 projects in Himalayan and -Applications Centre (SAC), ISRO, February 2026 for 5 years. Antarctica regions. -| Ahmedabad, | signed | an MoU | for | -| ----------------- | ---------------- | --------------- | --------- | -| the development | | of geospatial | | -| technologies | for monitoring | | and | -| modelling | of snow, | terrain | and | -| glacier | products, | and | analysis | -| & assessment | of | mountain | geo- | -| hazards | (avalanches, | landslides, | | -GLOF). Dr PK Satyawali, Director, -DIPAS SIGNS MOU WITH INDIAN INSTITUTE OF -TECHNOLOGY-ROPAR -The Defence Institute of IIT-Ropar. This collaboration augmentation, high-altitude -Physiology and Allied Science is poised to drive innovation sustenance, smart energy, and -(DIPAS), Delhi, has signed a for human performance efficient waste management. -| Memorandum | of | Understanding | | -| ----------- | --- | -------------- | --- | -(MoU) with the Indian Institute of -Technology-Ropar (IIT-Ropar) for -| collaborative | research, | academic | | -| -------------- | ---------- | -------------- | --- | -| exchange, | and | strengthening | | -technological innovation utilising -| each other’s | research | expertise | | -| -------------- | --------- | ---------- | --- | -on 10 February 2026. The signing -ceremony was held at IIT-Ropar -in the presence of Director, DIPAS -Shri Devakanta Pahad Singh, OS, -and Prof. Rajeev Ahuja, Director, -MOU SIGNED BY ITM -| Institute | of | Technology | | -| ------------ | ---------- | ----------- | ---- | -| Management | (ITM), | Mussoorie | | -| has signed | two MoUs | with | IIT | -Roorkee and IIT Bombay for design -and delivery of joint certification -| programmes, | short | term | courses | -| ------------ | ------ | ----- | -------- | -and customized training for DRDO -employees during its Raising Day. -APRIL 2026 6 - -| EVENTS | | | | | | | | DRDO NEWSLETTER | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | -NATIONAL SCIENCE DAY CELEBRATION -DEBEL, Bengaluru -National Institute of Technical March 2026. The Centre Head, -Teachers Training & Research DIHAR, Dr OP Chaurasia, Sc 'G', -| The | National | Science | Day | | | | | | | | | -| ---- | --------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -(NITTTR), Chandigarh, delivered virtually chaired the celebration. -2026 (NSD 2026) was celebrated -a talk on ‘Scientific Approach and Dr Vijay K Bharti, Sc ‘F’, delivered -| at Defence | Bioengineering | | & | | | | | | | | | -| ------------ | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Management for the Benefit of the NSD oration on “Study on -| Electromedical | | Laboratory | | | | | | | | | | -| --------------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -the Nation’. Director DGRE, Dr the Effectiveness of Yaks in -(DEBEL), Bengaluru on 9 March -PK Satyawali also addressed the Logistics Support in Super High- -| 2026. | The NSD | oration | was | | | | | | | | | -| ------ | ----------- | -------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -gathering about the instrumental Altitude Areas" an R&D study on -| delivered | by Shri | Srinivasan | K, | | | | | | | | | -| ---------- | --------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -role of DRDO in celebrating endurance exercise physiology -| Sc ‘E’ | on ‘Liner-less | Composite | | | | | | | | | | -| -------- | ---------------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -‘National Science Day’, Women’s and load-carrying capacity to -| Pressure | Vessels: | Challenges | | | | | | | | | | -| --------- | --------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -in Strategic Role for Defence R&D develop pack animal technology -| and Opportunities | | for | Pressure | | | | | | | | | -| ------------------- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -as well as ‘Viksit Bharat’. for high-altitude mountain area -| Containment’. | He | presented | an | | | | | | | | | -| -------------- | ----- | ---------- | --- | --------------------------- | --- | --- | --- | ------------ | --- | --------- | ---- | -| | | | | He applauded DGRE services | | | | deployment. | He | received | the | -overview of potential applications -towards geo-hazard management commendation certificate and -| of pressure | cylinder | in | defence, | | | | | | | | | -| ------------- | --------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -across entire Himalaya and medal for his work. Thereafter, -| critical | technology | involved | and | | | | | | | | | -| --------- | ----------- | --------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -motivated DGRE fraternity to brainstorming and discussion -| fabrication | process. | The | event | | | | | | | | | -| ------------ | --------- | ---- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -think out of the box to develop were held on how S&T activities -| was followed | by presentation | | of | | | | | | | | | -| -------------- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -various technologies and geo- and technology development will -science day medal and certificate -informatics products to address help India in achieving “Viksit -to the orator by Director, DEBEL. -the armed forces requirement Bharat" (Developed India) by -with changing global scenario. 2047 as a common national goal. -Shri Chitesh Kumar Sharma, Sc ‘D’ -delivered a technical talk on ‘‘Rock -| | | | | Slope | Stability | Analysis | using | | | | | -| --- | --- | --- | --- | ----------- | ---------- | --------- | ------------- | --- | --- | --- | --- | -| | | | | Tree-based | Ensemble | | Machine | | | | | -| | | | | Learning | Models". | | Shri Sharma | | | | | -was awarded with the medal and -certificate for his oration. -| DGRE, Chandigarh | | | | | | | | DMSRDE, Kanpur | | | | -| ---------------- | -------- | ---- | ----- | --- | --- | --- | --- | -------------- | ---------- | --- | ---- | -| | | | | | | | | Defence | Materials | | and | -| National | Science | Day | 2026 | | | | | | | | | -Stores Research & Development -| was | celebrated | at | Defence | | | | | | | | | -| --------------- | ----------- | --- | --------- | --- | --- | --- | --- | -------------- | ----------- | ---------- | ----- | -| Geoinformatics | | | Research | | | | | Establishment | | (DMSRDE), | | -| | | | | | | | | Kanpur, | celebrated | NSD | 2026 | -| Establishment | | | (DGRE), | | | | | | | | | -on 27 February 2026. The theme -Chandigarh, on 2 March 2026 with -| | | | | DIHAR, Leh | | | | for NSD | 2026 | was | “Women | -| ----------- | ------- | --- | --------- | ---------- | --- | --- | --- | ------------- | ----------- | ---- | ------- | -| the theme | ‘Women | in | Science: | | | | | | | | | -| | | | | | | | | in Science: | Catalysing | | Viksit | -Catalyzing Viksit Bharat’. The The Defence Institute of High -Chief Guest for the occasion, Prof. Bharat”. The Chief Guest Prof. -| | | | | Altitude | Research | | (DIHAR), | | | | | -| --- | --- | --- | --- | --------- | --------- | --- | --------- | ----------- | --------- | --- | ------ | -| | | | | | | | | RS Verma, | Director | | MNNIT | -(Dr) Bhola Ram Gurjar, Director, -Leh, celebrated NSD 2026 on 2 -| | | | | | | 7 | | | | | APRIL 2026 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | - -| EVENTS | | | | | | | | | DRDO NEWSLETTER | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | -competitions were organised. Dr Petroleum (IIP), CSIR, Dehradun. -Allahabad, Prayagraj delivered a -talk on “Bio-inspired Materials”. AP Dash, OS & Director, HEMRL, He delivered a keynote address -On the occasion, Dr Kingsuk welcomed the Chief Guest and on ‘Journey Towards Sustainable -Mukhopadhyay, OS & Director informed about the significance of Energy and Environment’. Shri -DMSRDE addressed the the celebration and how important Abhijit Chakraborty, Sc ‘F’, -is the scientific temperament and delivered the oration on ‘Optical -| DMSRDE | fraternity | | and | spoke | | | | | | | | | -| ------- | ----------- | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -about the importance of NSD and scientific progress for the country to Telescope: Technique for RSO -history behind its celebration. become a developed nation. tional Detection and Way Ahead’. On -Dr Ajit Shankar Singh, Sc ‘E’ importance. Shri PC Rao delivered this auspicious occasion, Dr Ajay -delivered the NSD oration on the NSD oration on “Science of Kumar congratulated the scientific -Insensitive Munitions: Simulation community of IRDE for its quest -| ‘Phthalonitrile | | (PN) | Resins | for | | | | | | | | | -| ---------------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -High Temperature Applications’. & Experimental Study”. The Chief to achieve excellence and self- -The Chief Guest presented the Guest distributed awards to the reliance. -NSD medal and certificate to Dr winners of various competitions. -Ajit Shankar Singh, Sc ‘E’. Dr Dr SK Sahoo delivered the vote of -thanks. -| Debmalya | Roy, | Sc | 'G' & | Head | | | | | | | | | -| --------- | ----- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | -NSD coordinated the event. -ISSA, Delhi -| HEMRL, Pune | | | | | | | | | The | Institute | for | Systems | -| ----------- | ----------- | ------- | ---------- | --- | -------------- | --- | --------- | --- | ---------------------------------- | ---------- | ----- | -------- | -| | | | | | IRDE, Dehradun | | | | Studies & Analyses (ISSA), Delhi, | | | | -| High | | Energy | Materials | | | | | | | | | | -| | | | | | | | | | celebrated | NSD | 2026 | on 27 | -| Research | Laboratory | | (HEMRL), | | | | | | | | | | -| | | | | | Instruments | | Research | & | February 2026. The NSD oration | | | | -| Pune, | celebrated | | NSD 2026 | on | | | | | | | | | -Development Establishment was given by Ms Pooja, Sc ‘B’ -27 February 2026. Padma Shri (IRDE), Dehradun, celebrated -| | | | | | | | | | on “A | Modelling | Approach | to | -| ----------------- | ------------ | ------------ | -------------- | ------- | ----------- | --- | --------------------- | ------- | ------- | ---------- | --------- | --- | -| Awardee | | Dr | Krishnamurthy | | | | | | | | | | -| | | | | | NSD 2026 | on | 2 March | 2026. | | | | | -| Balasubramanian, | | | DS | & | | | | | | | | | -| | | | | | The theme | of | NSD 2026 | was | | | | | -| Director, | Non-Ferrous | | Materials | | | | | | | | | | -| | | | | | ‘Women | in | Science: Catalysing | | | | | | -| Technology | | Development | | Centre | | | | | | | | | -Viksit Bharat’ which focuses the -(NFTDC), Hyderabad, graced the crucial role of women in fostering -occasion as the Chief Guest. Shri -| | | | | | innovation | and | propelling | India | | | | | -| --- | --- | --- | --- | --- | ----------- | ---- | ----------- | ------ | --- | --- | --- | --- | -Bikashhose, Sc ‘G’, delivered the -| | | | | | towards | becoming | a | developed | | | | | -| -------- | --------- | --- | ------------- | --- | -------- | --------- | ------- | ---------- | --- | --- | --- | --- | -| welcome | address. | | He informed | | | | | | | | | | -| | | | | | nation. | Dr Ajay | Kumar, | OS & | | | | | -that various competitions such as -| | | | | | Director, | IRDE, | inaugurated | the | | | | | -| --- | --- | --- | --- | --- | ---------- | ------ | ------------ | ---- | --- | --- | --- | --- | -science crossword, essay writing, event. The Chief Guest of the -| poster, | scientific | | presentations, | | | | | | | | | | -| -------- | ----------- | -------- | --------------- | ----- | ----------- | --- | --------- | ------ | --- | --- | --- | --- | -| | | | | | event was | Dr | Harender | Singh | | | | | -| online | and | offline | science | quiz | | | | | | | | | -Bisht, Director, Indian Institute of -| APRIL 2026 | | | | | | | 8 | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| EVENTS | | | | | | DRDO NEWSLETTER | | | | -| ------ | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | -Chemical Warfare in Theatre-level of the day and elaborated about 2026. Dr Khilawan Choudhary, Sc -Wargame”. the theme “Women in Science ‘F’, was the orator of the event. He -The oration was focussed Catalyzing Viksit Bharat”. delivered a talk on ‘Data-Driven -on elaborating a modelling Dr Das delivered an inspiring Lifetime Estimation of MOSFET -approach for evaluation of extent lecture on “Radar–The Third using Supervised Learning -of dispersion of chemical agents Eye of Modern Warfare: From Algorithms’. He received a medal -and their resultant causalities in Isolated Nodes to Multi-Domain and commendation certificate -formations deployed in a theatre Ecosystems” highlighting the from Dr SK Datta, OS & Centre -thereby affecting effectiveness of importance of autonomous Head, MTRDC. -formation. Smt Deepakshi Shah, sensing, multi-sensor data fusion, -OS & Director presented medal AI & intelligent superiority. He -and commendation certificate to gave the roadmap for futuristic -| the orator. | | | Radar R&D in the evolving global | | | | | | | -| ------------ | --- | --- | --------------------------------- | ---------- | ------------- | --- | --- | --- | --- | -| | | | scenario. | This was | followed by | | | | | -LRDE, Bengaluru -NSD oration on “Electromagnetic -| The NSD 2026 was celebrated | | | Compatibility: Quest for Harmony | | | | | | | -| ---------------------------- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -| at Electronics | | & | Radar from DC to Cosmic Rays” by Shri | | | | | | | -Development Establishment Shiva Shastha Rai A, Sc ‘F’. DG -(LRDE), Bengaluru, on 9 March (ECS) presented the medal and -2026. Dr BK Das, DS & DG (ECS) commendation certificate to the -was the Chief Guest and Shri M orator. The event concluded with NMRL, Ambernath -Sheik Althaf, OS & Director, LRDE vote of thanks by Shri SD Suresh, -The NSD 2026 was celebrated -| presided | over the | function. | The Sc ‘G’. | | | | | | | -| ----------------- | ---------- | -------------- | ------------ | --- | --- | ---------- | ---------- | --- | --------- | -| | | | | | | at Naval | Materials | | Research | -| event commenced | | with welcome | | | | | | | | -MTRDC, Bengaluru -Laboratory (NMRL), Ambernath, -| address by | Shri | Abid Hussain | | | | | | | | -| ------------ | ----- | -------------- | --- | --- | --- | --- | --- | --- | --- | -on 27 February 2026 with great -VA, Sc ‘H’ & Associate Director National Science Day 2026 was -(Planning). Director, LRDE celebrated at the Microwave Tube enthusiasm. The theme of the -program was "Women in Science: -greeted the scientific community Research & Development Centre -| | | | | | | Catalysing | Viksit | Bharat". | All | -| --- | --- | --- | --- | --- | --- | ----------- | ------- | --------- | ---- | -and briefed about the significance (MTRDC), Bengaluru, on 2 March -| | | | | | | officers | and | staff of | NMRL | -| --- | --- | --- | --- | --- | --- | --------- | ---- | ---------- | ----- | -participated in the event. Dr Suman -| | | | | | | Roy Choudhury, | | OS & | Director, | -| --- | --- | --- | --- | --- | --- | ---------------- | --- | ------ | ---------- | -NMRL, addressed the gathering, -| | | | | | | emphasizing | the | importance | of | -| --- | --- | --- | --- | --- | --- | ------------ | -------------- | ----------- | ---- | -| | | | | | | scientific | advancements. | | The | -Science Day Oration Award 2026 -| | | | | | | was conferred | | upon Shri | Subir | -| --- | --- | --- | --- | --- | --- | --------------- | --- | ----------- | ------ | -Barman, Sc 'F', who delivered a -lecture on "Synergistic Multilayer -Coating for Corrosion Protection -| | | | | | | and Hydrogen | | Management | | -| --- | --- | --- | --- | --- | --- | ------------------- | ---------- | ------------ | --------- | -| | | | | | | in Fuel | Tanks: | Accelerated | | -| | | | | | | and Field | Studies". | As | part of | -| | | | | | | the celebrations, | | a science | day | -oration and quiz competition were -organized for students of 8th and -| | | | | 9 | | | | | APRIL 2026 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---------- | - -| EVENTS | | | | | | | | DRDO NEWSLETTER | | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | -PXE, Chandipur -9th standard from nearby schools. covered ocean variability in -A total of 11 schools participated, the IOR strongly influences -| | | | | | | | | The | NSD | 2026 | was observed | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | ----- | -------------- | --- | -with approximately 42 students sonar performance by altering -| | | | | | | | | at Proof | | & | Experimental | | -| ---------------------------- | --- | --- | ---------------------- | --- | --- | --- | ------- | ---------- | --- | --- | ------------- | --- | -| taking part in the program. | | | the four-dimensional | | | | (space– | | | | | | -Establishment (PXE), Chandipur, -| | | | time) | structure | of | temperature, | | | | | | | -| --- | --- | --- | ------ | ---------- | --- | ------------- | --- | ------------- | --- | ------ | ------ | ------ | -| | | | | | | | | on 2 March | | 2026. | Prof. | Ashok | -salinity, and sound speed, which -Kumar Ganguli, Director, IISER, -| | | | control | acoustic | | propagation | | | | | | | -| --- | --- | --- | -------- | --------- | --- | ------------ | --- | --- | --- | --- | --- | --- | -Berhampur, graced this occasion -| | | | and transmission | | | loss. | Dr Das | | | | | | -| --- | --- | --- | ------------------ | --- | --- | ------ | -------- | --- | --- | --- | --- | --- | -as the Chief Guest. Shri PK Sahani -presented the citation to Dr Anand -Sc 'E' briefed the purpose behind -P in the presence of Dr D Seshagiri, -| | | | | | | | | the celebration | | of | the Day. | Shri | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | ---------- | ----- | -OS & Director, NPOL. -| | | | | | | | | Devanandh | | S, Sc | 'B' received | a | -| --- | --- | --- | ------------------- | --- | --- | --- | --- | --------------------------------- | --- | ------- | -------------- | --- | -| | | | NSTL, Visakhapatnam | | | | | medal and a certificate from the | | | | | -Chief Guest for his oration titled -Naval Science & Technological -| | | | | | | | | ‘Applications | | of | Active | Shape | -| --- | --- | --- | ----------- | --- | -------- | ----------- | --- | -------------- | --------- | ------ | ----------- | ------ | -| | | | Laboratory | | (NSTL), | Visakhapat- | | | | | | | -| | | | | | | | | Model | in Test | Range | Scenario’. | | -NPOL, Kochi -nam celebrated NSD on 2 March -Shri Subodha Kumar Nayak, OS -| | | | 2026. | The | Chief | Guest | Prof. | | | | | | -| --- | --- | --- | ------ | ---- | ------ | ------ | ------ | --- | --- | --- | --- | --- | -Naval Physical & Oceanographic -| | | | | | | | | & Director, | | PXE in | his | address | -| ----------- | -------- | ------ | --------- | -------- | --- | ----- | -------- | ------------- | --- | -------- | ---- | -------- | -| | | | GP Raja | Sekhar, | | Vice | Chancel- | | | | | | -| Laboratory | (NPOL), | Kochi | | | | | | | | | | | -| | | | | | | | | highlighted | | role | of | Women | -lor, Andhra University; Guest of -observed NSD on 2 March 2026 scientists in research sphere. -Honour Shri J Venkata Rao, EO -by arranging an invited talk and Shri RS Mohapatra, Sc 'F' & -a science day oration. Dr BK Das, (Simhachalam Devasthanam); Dr -| | | | | | | | | Additional | | Director | introduced | | -| --- | --- | --- | -------- | --- | ----------- | --- | ------- | ----------- | --- | --------- | ----------- | --- | -| | | | Abraham | | Varughese, | | OS & | | | | | | -DS & DG (ECS), the Chief Guest -the Chief Guest to the gathering. -| | | | Director, | NSTL | | addressed | the | | | | | | -| --- | --- | --- | ---------- | ----- | --- | ---------- | ----- | --- | --- | --- | --- | --- | -of the occasion delivered a talk -| | | | | | | | | Prof. | Ganguli | in | his | speech | -| --- | --- | --- | --- | --- | --- | --- | --- | ------ | -------- | --- | ---- | ------- | -gathering. Dr Varughese expressed -on ‘Changing War Scenarios and emphasized the impact of science -| | | | importance | | of the | creativity | of | | | | | | -| --- | --- | --- | ----------- | --- | -------- | ----------- | ---- | --- | --- | --- | --- | --- | -Our Preparation Through Best in our lives and his inspiring -Practices’. He asserted the need Indian teams to build a better -scientific journey. As a part of the -| | | | future | where | innovation | | has no | | | | | | -| --- | --- | --- | ------- | ------ | ----------- | --- | -------- | --- | --- | --- | --- | --- | -for indigenisation in all fields and -celebration, students from various -bounds and where Viksit Bharat -explained about the best practices -schools visited the laboratory on -stands as a beacon of progress and -| to be followed. | | | | | | | | 27 February 2026. | | | | | -| --------------- | --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | --- | -atmanirbhar Bharat. Shri P Naga -| The | Science Day | oration | | | | | | | | | | | -| ---- | ------------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -was by Dr Anand P, Sc ‘F’ on Sai Babu, Sc 'C' was awarded with -a medal and a commendation cer- -| ‘Towards | Transparent | Ocean: | | | | | | | | | | | -| --------- | ------------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -tificate for his work on “Develop- -Addressing 4-D Ocean Variability -| | | | ment | of Solid-State | | Lithium-Ion | | | | | | | -| -------------- | ----------- | --- | ------- | ---------------- | --------------- | ------------ | ----- | --- | --- | --- | --- | --- | -| for Enhanced | Underwater | | | | | | | | | | | | -| | | | Cell”. | On | the occasion, | | NSTL | | | | | | -Domain Awareness’. The oration -| | | | conducted | | quiz | and | elocution | | | | | | -| --- | --- | --- | ------------- | --- | ----- | ---- | ----------- | --- | --- | --- | --- | --- | -| | | | competitions | | for | the | school | | | | | | -students around Visakhapatnam. -| APRIL 2026 | | | | | 10 | | | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| EVENTS | | | | | | | DRDO NEWSLETTER | | | | -| ------ | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | -RAISING DAY CELEBRATIONS -ADRDE, Agra and staffs for their tireless efforts, emergency preparedness, R&D -specifically mentioning their role in radiation countermeasures, -Aerial Delivery Research & in the various project activities. radiation decontamination, and -| Development | | Establishment | | | | | | | | | -| ------------ | --- | -------------- | --- | --- | --- | --- | --- | --- | --- | --- | -allied biomedical technologies. -INMAS, Delhi -(ADRDE), Agra celebrated its 58th -| | | | | | | | Dr Kaushik | commended | | the | -| --- | --- | --- | --- | --- | --- | --- | ------------ | ---------- | --- | ---- | -Raising Day on 20 January 2026, -| | | | | | Institute | of Nuclear Medicine | contributions | of INMAS | for | the | -| --- | --- | --- | --- | --- | ---------- | ---------------------- | -------------- | ---------- | ---- | ---- | -marking a milestone of excellence, and Allied Sciences (INMAS), launching of the ‘Vikiran Prehri -self-relieance, and service to the Delhi, celebrated its 65th Raising Permanent Flagship Programme,' -nation. The day served as a tribute -Day on 13 February 2026. Dr and the national program ‘Vikiran -| to the | scientists | and | technical | | | | | | | | -| -------- | ----------- | ---- | ---------- | --- | --- | --- | --- | --- | --- | --- | -BK Das, DS & DG (ECS), was Prehri’ for medical teams of MHA -| staff whose | | innovations | | have | | | | | | | -| ------------- | --- | ------------ | --- | ----- | --- | --- | --- | --- | --- | --- | -the Chief Guest on the occasion. attached with VVIPs to deal -transformed India into a global Dr Chandrika Kaushik, DS & with nuclear and radiological -defense power. Dr K Rajalakshmi DG (PC & SI), was the Guest of exigencies. -Menon, OS & DG (Aero) gave an -| | | | | | Honour. Dr Jaya Shanti, Director, | | Dr Kaushik | | appreciated | | -| ---------- | ------ | --------- | --- | ------ | ---------------------------------- | --- | ------------ | --- | ------------ | --- | -| inspiring | video | message. | | After | | | | | | | -DRDO-Industry Academia Centre INMAS for developing two de- -the message a glimpse of ADRDE -of Excellence, IIT Delhi, also coporporating agents. The Chief -various landmark achievements of attended the celebrations. The Guest, Dr Das, in his address -2025 was presented. other dignitaries from corporate appreciated the contributions -| Dr | Manmohan | Parida, | OS | & | | | | | | | -| --- | --------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | -HQrs and Directors from Delhi- of INMAS in the fields of R&N -| Director, | DRDE, | Gwalior | graced | | | | | | | | -| ---------- | ------ | -------- | ------- | --- | --- | --- | --- | --- | --- | --- | -based laboratories also graced the emergency management, training, -the occasion as the Chief Guest. -| | | | | | occasion. | | and biomedical research. | | | | -| --- | --- | --- | --- | --- | ---------- | --- | ------------------------ | --- | --- | --- | -In a powerful keynote address Dr Sudhir Chandna, OS & The dignitaries also released the -he emphasized the importance Director, INMAS, welcomed 24th volume of 'Nucleus,' an in- -| of innovation, | | technology | | and | | | | | | | -| ---------------- | --- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | -the guests and highlighted house annual bulletin of INMAS, -| community | | service. | Dr Manoj | | | | | | | | -| ---------- | --- | --------- | ---------- | --- | --- | --- | --- | --- | --- | --- | -the momentous efforts made and the 28th edition of the Hindi -| Kumar, | OS & | Director, | ADRDE | | | | | | | | -| ------- | ------ | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -by INMAS through the live magazine 'Prastuti'. Mementoes -commended the ADRDE officers demonstration-cum-exhibition were also conferred to employees -of major core capabilities of on successful completion of 25 -| | | | | | INMAS in radiation and nuclear | | years of service in DRDO. | | | | -| --- | --- | --- | --- | --- | ------------------------------- | --- | ------------------------- | --- | ---------- | --- | -| | | | | | | 11 | | | APRIL 2026 | | - -| EVENTS | | | | | | | | DRDO NEWSLETTER | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | -IRDE, Dehradun -| The | Instruments | | Research | | | | | | | | -| --------------- | ------------ | -------------- | --------- | --- | --- | --- | --- | --- | --- | --- | -| & Development | | Establishment | | | | | | | | | -(IRDE), Dehradun, celebrated its -66th Raising Day on 18 February -2026 with full zeal and vivacity. Dr -Ajay Kumar, OS & Director, IRDE -| welcomed | the | august | gathering | | | | | | | | -| ------------------- | ---- | ------- | ---------- | ------ | --- | --- | --- | --- | --- | --- | -| and congratulated | | | officers, | staff | | | | | | | -and their families on Raising Day -of IRDE. He conveyed his thanks -| and appreciated | | all | employees | | | | | | | | -| ----------------- | ---- | ----------- | ---------- | --- | --- | --- | --- | --- | --- | --- | -| of IRDE | for | completion | | of | | | | | | | -projects during the year 2025. He -strongly believed that the projects -as the Guest of Honour. Shri SP from Technology Development -targeted for the year 2026 will be -Dobhal, Director, ITM delivered to Technology Management and -| completed | with | the | committed | | | | | | | | -| ---------- | ----- | ---- | ---------- | --- | --- | --- | --- | --- | --- | --- | -the welcome address and also stressed that ITM should take -| workforce | of | IRDE | within | the | | | | | | | -| ---------- | --- | ----- | ------- | ---- | --- | --- | --- | --- | --- | --- | -highlighted overall performance a lead. He also released ITM -| designated | time | frame. | Dr | Ajay | | | | | | | -| ----------- | ----- | ------- | --- | ----- | --- | --- | --- | --- | --- | --- | -and achievements of institute Training Calendar 2026-27, -| Kumar | awarded | commendation | | | | | | | | | -| ------ | -------- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | -during the preceding year. The STEMS Project Compendium -certificates and badges to thirty- -Guest of Honour highlighted and ITM Hindi magazine, 'Srijan' -| one IRDE | employees | | from | all | | | | | | | -| ---------- | ---------- | --- | ----- | ---- | --- | --- | --- | --- | --- | --- | -the importance of additional followed by a curtain raiser -the cadres for their dedication to -infrastructure, personal well- event of “Society for Technology -work and showcasing great skill in -| | | | | | being and | MoU’s with | India’s | Research | & Management” | | -| --- | --- | --- | --- | --- | ----------- | ------------ | -------- | --------- | --------------- | --- | -their respective areas of activity. -| | | | | | leading institutes. | | | (STREAM). | | | -| ------------- | ---------- | -------------- | --------- | ----- | ------------------- | ------------ | -------- | ------------------------------- | ---------- | ---------- | -| Dr Ajay | Kumar | also | honoured | | | | | | | | -| | | | | | Dr Kamat, | in his | keynote | On the | occasion, | lab-level | -| Project | Directors | | and | team | | | | | | | -| | | | | | address | highlighted | the | Commendation Certificates were | | | -| members | of | ten projects | | with | | | | | | | -| | | | | | importance | of ITM | towards | presented and an employee was | | | -| commendation | | certificates | | for | | | | | | | -training of DRDO employees and felicitated on completing 25 -| successfully | | completion | | and | | | | | | | -| ------------- | --- | ----------- | --- | ---- | ---------------------------------- | --- | --- | --------------------------- | --- | --- | -| | | | | | to devote attention on transition | | | years of service in DRDO. | | | -closing of projects during the year -2025. As part of the celebration, -| various | sports | events | | were | | | | | | | -| ---------- | ------- | -------- | --- | ----- | --- | --- | --- | --- | --- | --- | -| conducted | and | winners | | were | | | | | | | -awarded by Dr Sudhir Khare, OS. -ITM, Mussoorie -| Institute | | of | Technology | | | | | | | | -| ----------- | ---- | ------- | ----------- | ---- | --- | --- | --- | --- | --- | --- | -| Management | | (ITM), | Mussoorie | | | | | | | | -| celebrated | its | 65th | Raising | Day | | | | | | | -on 28 February 2026. The event -was graced by Dr Samir V Kamat, -| Secretary | DD | R&D | & Chairman | | | | | | | | -| ---------- | ---------- | ------ | ------------ | --- | --- | --- | --- | --- | --- | --- | -| DRDO, | as Chief | Guest | and | Dr | | | | | | | -Mayank Dwivedi, OS & DG (HR) -| APRIL 2026 | | | | | | 12 | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -EVENTS DRDO NEWSLETTER -INTERNATIONAL WOMEN’S DAY CELEBRATIONS -DRDE, Gwalior -The Defence Research & -Development Establishment -(DRDE), Gwalior, celebrated -International Women's Day -2026 (IWD 2026) on 9 March -2026 with a one-day workshop -focused on work–life balance, -bringing together women -officers, staff members, students, -and contractual employees of -the laboratory. The workshop -featured Dr Kamalpreet Likhari, -Senior Ophthalmologist from -RJN Eye Institute, Gwalior, who -delivered an engaging talk on defence technologies for national Psychological Research (DIPR), -stress management, maintaining security. In his address, Dr Parida Delhi, delivered a talk on -a healthy lifestyle, and enhancing highlighted the laboratory’s 'Managing Emotions'. The talk -productivity while balancing important role in strengthening was focused on how to effectively -professional and personal India’s national security through manage emotions through -responsibilities. The valedictory advanced defence research, various practical techniques -event was graced by Guest of while appreciating the dedication with an ultimate goal of peaceful -Honour Mrs Anu Beniwal, IPS, and achievements of women and successful life. She engaged -Additional Superintendent of researchers and staff at DRDE. everyone in activities related to -Police, Gwalior; Dr Manmohan managing emotions. Director, -ISSA, Delhi -Parida, Director, DRDE; and Mrs ISSA appreciated the speaker and -Priti Chhanda Parida. The Institute for Systems mentioned that the topic of the -Welcoming the guests and Studies & Analyses (ISSA), Delhi, talk was relevant since it touches -participants, Dr Manisha Sathe, celebrated IWD 2026 on 9 March everyone on daily basis and the -Sc ‘G’, emphasized the growing 2026. Dr Sandhya Verma, Sc need to manage emotions is the -contributions of women scientists 'F', from Defence Institute of key to a happy society. All women -in advancing chemical and -biological defence technologies. -Addressing the gathering, Mrs -Beniwal spoke about the vital -role of women in nation-building -and public service, encouraging -greater participation of women -in both science and societal -leadership. She also lauded the -contribution of women to the to the -scientific community of DRDE in -developing indigenous chem bio- -13 APRIL 2026 - -| EVENTS | | | | | | | | | DRDO NEWSLETTER | | | | | -| ------ | --- | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | -employees of ISSA and Group NPOL, Kochi participated in LCA program and -Directors attended the program. the products developed by them. -| | | | | | Naval | | Physical | and | | | | | | -| --------------- | --- | --- | --- | --- | -------------- | --- | --------- | ----------- | ---------------------------------- | --- | --- | --- | --- | -| NMRL, Ambernath | | | | | | | | | She shared her vast experience in | | | | | -| | | | | | Oceanographic | | | Laboratory | | | | | | -developing Tejas. The second part -| | | | | | (NPOL), | Kochi | celebrated | IWD | | | | | | -| -------------- | --- | -------- | --- | ---- | -------- | ------ | ----------- | ---- | --- | --- | --- | --- | --- | -| International | | Women’s | | Day | | | | | | | | | | -of the talk by Dr Padmavati was -2026 on 16 March 2026. NPOL -was celebrated at Naval Materials more technical on “Development of -| | | | | | organized | an | invited | talk by | | | | | | -| --- | --- | --- | --- | --- | ---------- | --- | -------- | --------- | --- | --- | --- | --- | --- | -Research Laboratory (NMRL), Large-Scale Airborne Embedded -Ambernath, with immense zeal Dr Padmavati, OS & Technical -Application for a Fighter Aircraft: -Director (Avionics), ADA. It was -and happiness on 9 March, 2026. -a Case Study”. She has described -| | | | | | a journey | starting | with | genesis | | | | | | -| ------------- | --- | ----------- | --- | --------- | ----------------- | --------- | ------------ | -------- | ----------------- | --- | --- | ------ | ------ | -| Mrs Debyani | | Chattaraj, | | Sc 'F', | | | | | | | | | | -| | | | | | | | | | the development | | of | large | Scale | -| | | | | | of Aeronautical | | Development | | | | | | | -Chairperson of the Women’s airborne embedded Application -| | | | | | Agency | and | Light | Compact | | | | | | -| --- | --- | --- | --- | --- | ------- | ---- | ------ | -------- | --- | --- | --- | --- | --- | -Cell, highlighted the significance, in detail. Dr D Seshagiri, OS -| history | and theme | | of the | day. | | | | | | | | | | -| --------------- | ----------- | ----------- | ---------- | ----- | --- | --- | --- | --- | -------------- | --- | ------ | --------- | -------- | -| | | | | | | | | | & Director, | | NPOL, | | offered | -| Dr Suman | Roy | Choudhury, | | OS | | | | | | | | | | -| | | | | | | | | | felicitations | on | the | occasion | of | -| and Director, | | NMRL, | addressed | | | | | | | | | | | -IWD 2026. Ms Manjula P Raj, Sc -the gathering and emphasized 'G' & Chairperson, NPOL Women -| the importance | | | of women’s | | | | | | Cell welcomed the gathering. | | | | | -| ---------------- | --- | --- | ------------ | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --- | -empowerment, rights and equality -NSTL, Visakhapatnam -| in scientific | | organizations. | | The | | | | | | | | | | -| --------------- | --- | --------------- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -Guest of Honour, Ms T Shirisha, -| | | | | | | | | | International | | | Women’s | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --- | --- | -------- | --- | -Sc 'G', RCI, Hyderabad, delivered Day 2026 was organized in a -a motivational talk and presented befitting manner on 6 March -| a lecture | on | “Naval | Anti-Ship | | | | | | | | | | | -| ----------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | ----- | ---------- | --- | -------- | --- | -| | | | | | | | | | 2026 | at Naval | | Science | & | -Missile- Short Range”, along with -Technological Laboratory (NSTL), -a short film on missile testing. The -| | | | | | | | | | Visakhapatnam. | | The | event | was | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | ---- | ------ | ---- | -Chief Guest, Dr Tessy Vincent, graced by a distinguished group of -OS, BARC, Trombay delivered an Aircraft (LCA). She explained the dignitaries including Dr Chandrika -inspiring lecture on “Backend of design challenges faced in Tejas -Kaushik, DS & DG (PC&SI); Prof. -the Nuclear Fuel Cycle” and shared MK1 and gave an overall picture of -M Shashi, Principal (AU, College -her professional experiences with the DRDO laboratories, academic -| | | | | | | | | | of Engineering); | | Dr | Abraham | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | --- | -------- | --- | -the audience. institutions and industries Varughese, OS & Director NSTL; -Smt R Sasi Sri, First Lady of Naval -Systems & Materials Cluster; Smt -Ilin Dinah Abraham, First Lady of -NSTL & President MKM; Smt and -Smt K Sangeeta, IDAS, IFA (R&D). -| | | | | | | | | | In her | address, | she | highlighted | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | --------- | --------------- | -------------- | ------ | -| | | | | | | | | | the presence | | of committees | | like | -| | | | | | | | | | the Women | | Cell | and Internal | | -| | | | | | | | | | Complaint | Cell | (ICC) | at | NSTL, | -which focus on women’s welfare -and safety in the workplace. -| | | | | | | | | | Dr Varughese | | | congratulated | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | -------------- | ---------- | --- | -------------- | ------ | -| | | | | | | | | | the women | employees | | for | their | -significant role to the organization. -He encouraged women to embrace -| APRIL 2026 | | | | | | | 14 | | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -| EVENTS | | | | | | DRDO NEWSLETTER | | | -| ------ | --- | --- | --- | --- | --- | --------------- | --- | --- | -a collection of Hindi articles were -| | | | | | | released | by the | distinguished | -| --- | --- | --- | --- | --- | --- | ------------- | -------- | -------------- | -| | | | | | | dignitaries. | An | exhibition | -showcasing journey of 130 years -| | | | | | | of PXE, | the culture | of Odisha | -| --- | --- | --- | --- | --- | --- | --------- | ------------- | ----------- | -through food, art and craft along -| | | | | | | with stalls | showcasing | advanced | -| --- | --- | --- | --- | --- | --- | ------------- | ----------- | ---------------- | -| | | | | | | technologies | was | inaugurated | -| | | | | | | by the | esteemed | guests. The | -| | | | | | | workshop | hosted | a session for | -presentation of best paper, article -and poem by authors followed by -felicitations of women leaders of -DRDO playing the roles of Director -leadership roles in their respective accompanied by Prof. Geetanjali General, Laboratory Director, -fields. The Chief Guest Dr Kaushik, Das, Vice Chancellor of Berhampur Program Director and Project -| | | | | | | Director. | During | the technical | -| --- | --- | --- | --- | --- | --- | ---------- | ------- | --------------- | -highlighted, Senior ladies are to University Odisha as the Guest -| | | | | | | visit to | PXE technical | area, all | -| --- | --- | --- | --- | --- | --- | ---------- | --------------- | ----------- | -mentor the younger generation of Honour. Dr Samir V Kamat, -coming into the organization. Secretary DD R&D & Chairman the participants experienced live -Prof. Shashi emphasised that both DRDO, Prof. (Dr) Prateek Kishore, firing and got opportunity to fire -men and women take equal roles DG (ACE), Dr K Rajalakshmi small arms. The plenary session -| | | | | | | covered | "Challenges | in Women | -| --- | --- | --- | --- | --- | --- | -------- | ------------ | ---------- | -and responsibilities towards their Menon, DG (Aero), Shri Subodha -Leadership” followed by an active -| families. | | | Kumar Nayak, | Director, | PXE, | | | | -| --------- | --- | --- | -------------- | ---------- | ----- | --- | --- | --- | -As part of the celebrations, a Chairperson of ICC-Ms Sweta discussion between audience and -free medical camp was organized Singh, Sc 'G' & Convenor, IWD- the esteemed speakers. An invited -for all women fraternity of NSTL. 2026 Ms Smrutisudha Sahoo, Sc talk on “Cognition and Emotion” -| | | | | | | provided | clarity | on handling | -| --- | --- | --- | --- | --- | --- | --------- | -------- | ------------- | -More than 120 women participated 'F' graced the occasion. IWD 2026 -| | | | | | | emotions | in daily | life. This was | -| --- | --- | --- | --- | --- | --- | --------- | ---------- | ----------------- | -in the health check-up. A charity Souvenir ‘SUSHREE’ containing -event was also organized on the all innovative technical, non- followed by ceremonious handing -occasion. technical articles along with over of ‘Deeplakshmi’ baton to the -poems and a souvenir ‘KSHITIJ’, next torchbearer DRDE, Gwalior. -PXE, Chandipur -| A workshop | for | IWD 2026 | | | | | | | -| --------------- | ----------- | -------------- | --- | --- | --- | --- | --- | --- | -| was conducted | at | Proof and | | | | | | | -| Experimental | | Establishment | | | | | | | -| (PXE), | Chandipur, | under the | | | | | | | -theme ‘SUSHREE’, i.e. “Scientific -Upliftment and Strengthening Her -| Research | Endeavor | Earnestly” | | | | | | | -| --------- | ------------ | ------------ | --- | --- | --- | --- | --- | --- | -| during | 9-10 March | 2026. The | | | | | | | -| workshop | started | with beach | | | | | | | -walkathon at Chandipur sea beach. -| In the | inaugural | session, Mrs | | | | | | | -| ----------- | ------------ | -------------- | --- | --- | --- | --- | --- | ---------- | -| Anu Garg, | IAS, Chief | Secretary | | | | | | | -| of Odisha | was the | Chief Guest | | | | | | | -| | | | | 15 | | | | APRIL 2026 | - -| EVENTS | | | | | | | DRDO NEWSLETTER | | | | | -| ------ | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | -RAC, Delhi engineering, entrepreneurship, organization's commitment to -nutrition, and health, etc. through women's empowerment and -| The | Recruitment | | | and | | | | | | | | -| ----------- | ------------ | --- | ------- | ------ | ------------------------------- | --- | ------------- | --- | --- | --- | --- | -| | | | | | invited talks by eminent women | | inclusivity. | | | | | -| Assessment | Centre | | (RAC), | Delhi | | | | | | | | -professionals, Ms Chitralekha The program was graced by Chief -celebrated IWD 2026 and National -Vaidya, founder & CEO Guest Prof. (Dr) Renu Vij, Vice -| Safety | Day | on 9 March | | 2026. | | | | | | | | -| ------- | ---- | ------------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | -Varshasookt Pvt. Ltd. Mumbai; Chancellor of Panjab University, -| Ms Veena | Gupta, | | motivational | | | | | | | | | -| ---------- | ------- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | -who delivered an inspiring address -speaker, shared detailed insights -| | | | | | | | highlighting | | the | importance | | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ---- | ----------- | --- | -on occupational safety and health -| | | | | | | | of education, | | leadership | | and | -| ------------- | -------------- | ------------ | ----------- | ---- | --- | --- | --------------- | ----------- | ----------- | ------- | -------- | -| activities, | self-defence, | | industrial | | | | | | | | | -| | | | | | | | resilience | among | | women. | The | -| safety, | and | management. | | She | | | | | | | | -| | | | | | | | Guest | of Honour | | Dr | Radhika | -| also shared | | her views | regarding | | | | | | | | | -Trikha, CEO of iHub-AWaDH IIT- -| the | increasing | | participation | | | | | | | | | -| ---------- | ----------- | ----- | -------------- | ---- | --- | --- | ------- | ------------ | --------- | ------------ | --------- | -| | | | | | | | Ropar, | shared | valuable | | insights | -| of women | in | both | private | and | | | | | | | | -| | | | | | | | on | innovation, | | technology, | | -government sectors, their effective -| | | | | | | | startups | and | the | increasing | | -| --------------- | --- | -------------- | --------- | --- | --- | --- | ---------------- | ------ | ------------ | ----------- | ---- | -| leadership, | | and their | valuable | | | | | | | | | -| | | | | | | | role of | women | in | research | and | -| contributions. | | Additionally, | | | a | | | | | | | -| | | | | | | | entrepreneurial | | ecosystems. | | Sh | -quiz competition was organized at Ms Jilma Peruvangat, co-founder -| | | | | | | | J Santhosarajan | | OS | & | Director, | -| --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | --- | ---------- | -the office to mark the occasion of Kozhnosys, Pune; Ms Varsha -TBRL, in his address, emphasized -National Safety Day. The winning Pherwani, Clinical Nutritionist, -| | | | | | | | the importance | | of | rights, | justice | -| --- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | -------- | -------- | -participants were duly awarded. Pune; Dr Gyatri Ekbote from -and meaningful action for women -The event concluded with a pledge Pune. The workshop conveyed, -| | | | | | | | & girls. | He | highlighted | | that | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------------ | --- | ----- | -reaffirming the commitment of all IWD theme 2026 "Give To Gain” -| | | | | | | | courage, | dedication | | and | hard | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ---- | ----- | -personnel toward personal and by elaborating several ways for -work are key factors that enable -| industrial safety. | | | | | individuals | and organizations | to | | | | | -| ------------------ | --- | --- | --- | --- | ------------ | ------------------- | ------ | --- | -------- | ----------- | --- | -| | | | | | | | women | to | achieve | remarkable | | -practice it for maximum benefit. -| R&DE(E), Pune | | | | | | | success | in | various | fields. | The | -| ------------- | -------- | --------- | --- | --- | ---------------- | --- | -------- | ----- | --------- | -------- | ---------- | -| | | | | | TBRL, Chandigarh | | program | also | featured | | talks by | -| A | one-day | workshop | | on | | | | | | | | -Advocate Baljit Singh, about legal -"Empowering Women for Better Terminal Ballistics Research -awareness, discussed various acts -| Tomorrow” | | was | conducted | | Laboratory | (TBRL), Chandigarh | | | | | | -| ---------- | --- | ---- | ---------- | --- | ----------- | -------------------- | --- | --- | --- | --- | --- | -and women's rights. Dr Parminder -on 5th March 2026 at R&DE celebrated IWD 2026 on 10 -| | | | | | | | Kaur, | Gynae | & infertility | | expert, | -| --- | --- | --- | --- | --- | --- | --- | ------ | ------ | --------------- | --- | -------- | -(Engrs), Pune, as a part of IWD March 2026 at Ramgarh Ranges. -| | | | | | | | addressed | | important | | aspects | -| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---------- | --- | -------- | -2026. The workshop has covered The event saw the presence of -| | | | | | | | related | to women's | | health, | well- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------ | --- | -------- | ----- | -various aspects, like women more than 200 women scientists, -being and empowerment. -empowerment, environmental officers & staff, reflecting the -| APRIL 2026 | | | | | | 16 | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | - -EVENTS DRDO NEWSLETTER -RELEASE OF THE DRDO HR MANUAL-POLICIES AND -PROCEDURES–2025 -“DRDO HR Manual (Policies function as a comprehensive ready ees and administrators in ensur- -and Procedures)–2025” was reckoner for all DRDO officials, ing regulatory compliance while -released by the Secretary DD R&D enabling clarity, consistency, and enabling consistent and informed -& Chairman DRDO on 27 February informed decision-making, and decision-making. -2026 in the presence of Dr Mayank it would greatly assist readers Its implementation repre- -Dwivedi, OS & DG (HR); Dr Sanjai in the effective understanding sents an important step to- -Kumar Dwivedi, Director, DHRD; and implementation of the ward strengthening institutional -Dr SK Singh, Director, DMS; Shri organisation’s HR policies and processes and fostering a work -SP Dobhal, Director, ITM; Smt procedures. culture that promotes accounta- -Anita Mohindra, SC ‘G’, ITM; As an authoritative reference, bility, professional growth, and -and Dr Mehbuba Begam, Sc ‘F’ & the Manual will assist employ- organizational excellence. -Head (Training), DHRD, marking -a significant milestone for the -organisation. -Dr Kamat appreciated and -congratulated Dr Dwivedi and the -entire publication and editorial -committee for their sincere -efforts, meticulous coordination, -and dedicated commitment in -preparing this comprehensive -document. He expressed -confidence that the Manual would -MEETING ON ‘SEVA SANKALP RESOLUTION’ ADOPTED -BY THE UNION CABINET AT DMSRDE -A meeting on ‘Seva Sankalp project/task with honest intent the laboratory will form the -Resolution’ as per directives of will paved the path towards foundation for advancing India -DMS, DRDO HQrs, New Delhi was building ‘Viksit Bharat by 2047’. as a self-reliant nation in defence -organised in the Auditorium on 3 The positive work culture in sector. -March 2026. The event was chaired -by Dr Kingsuk Mukhopadhyay, -OS & Director, DMSRDE. All -the officers including Additional -Directors and Divisional Heads -attended the meeting. -The Seva Sankalp Resolution -in hindi was read by Shri Amit -Saraiya, Sc 'F' & Head TC & HRD. -Director, DMSRDE talked about -the implementation of rules and -regulations in executing any -17 APRIL 2026 - -| EVENTS | | | | | | DRDO NEWSLETTER | | | -| ------ | --- | --- | --- | --- | --- | --------------- | --- | --- | -MOCK FIRE DRILL AT CVRDE -As part of 55th National Safety -demonstrating the real fire incident -Day/Month a Fire mock drill was -| | | | | | | and use | of fire tender | and fire | -| ---------- | ----------- | -------- | --- | --- | --- | --------- | ----------------- | ---------- | -| conducted | at CVRDE, | Chennai | | | | | | | -extinguishers to mitigate different -on 6 March 2026. Shri J Rajesh -| | | | | | | fire scenarios. | Each | fire tender | -| ------- | ----------------- | ------- | --- | --- | --- | ----------------- | ----- | ------------- | -| Kumar, | OS & Director, | CVRDE, | | | | | | | -having one leading Fire Man and -inaugurated the event. The theme -Four Fire Men demonstrated their -of the event was “Engage, Educate -skills to put-off the fire. -and Empower People to Enhance -| | | | | | | During | his address, | Shri T | -| --- | --- | --- | --- | --- | --- | ------- | -------------- | -------- | -Safety”. Various fire teams from -Panneer Selvam, OS & Associate -| five organisations | namely | Tamil | | | | | | | -| -------------------- | ------- | ------ | --- | --- | --- | --------- | ------------- | --------- | -| | | | | | | Director | (Arjun MBT) | narrated | -Avadi and Heavy Vehicles Factory -| Nadu | Fire & Rescue | Service, | | | | | | | -| ----- | ---------------- | --------- | --- | --- | --- | --- | --- | --- | -the importance of safety culture in -Ordnance Depot, Ordnance of Avadi participated in the drill. -| | | | The mock | drill was | conducted, | the day today activities. | | | -| --- | --- | --- | ---------- | ----------- | ----------- | ------------------------- | --- | --- | -Clothing Factory, Engine Factory, -NATIONAL SAFETY DAY CELEBRATIONS AT DRDL -| 55th National Safety Day was | | | The Chief | Guest | delivered an | | | | -| ----------------------------- | --- | --- | ----------- | ------ | -------------- | --- | --- | --- | -organized at Defence Research and insightful talk on “Personality -Development Laboratory (DRDL), Development-Positive Thinking -Hyderabad, on the 6 March 2026. and its Implication on Behavioural -Dr Ankathi Raju, DS & Director, Safety". To promote the awareness -DRDL along with management on safety issues, the event -council welcomed the Chief Guest included competitions like poster- -Shri Burra Venkatesham, IAS making, slogan-writing, poems -(Retd.), Chairman, Telangana and quiz. All 320 participants -Public Service Commission. were administered the safety and -health pledge. -AWARENESS PROGRAM ON THE OCCASION OF -NATIONAL SAFETY WEEK AT ITR -| To Mark | the National | Safety | | | | | | | -| --------- | -------------- | ------- | --- | --- | --- | --- | --- | --- | -safety banners in and outside "Engage, Educate, & Empower -Day-Week-2026 and to foster the -| | | | of ITR premises with the theme | | | People to Enhance Safety". | | | -| --- | --- | --- | ------------------------------- | --- | --- | --------------------------- | --- | --- | -idea of importance of safety in the -organisation a two days awareness -| program | was conducted | at | | | | | | | -| ----------- | --------------- | ------- | --- | --- | --- | --- | --- | --- | -| Integrated | Test Range | (ITR), | | | | | | | -Chandipur, during 18-19 March -| 2026. | During the | celebration, | | | | | | | -| ------ | ------------ | ------------- | --- | --- | --- | --- | --- | --- | -essay and poster making, slogan -writing & quiz competitions were -| conducted | along with | display of | | | | | | | -| ---------- | ------------ | ------------ | --- | --- | --- | --- | --- | --- | -| APRIL 2026 | | | | 18 | | | | | - -HRD ACTIVITIES DRDO NEWSLETTER -COURSES AT DGRE, CHANDIGARH -Training at Basic Training Centre, ITBP, Panchkula -Defence Geoinformatics was aimed to acquaint newly of DGRE, Snow avalanche -Research Establishment inducted ITBP personnel towards formation, types of avalanches, -(DGRE), Chandigarh, conducted the enhancement of operational failure mechanisms, methods -comprehensive awareness preparedness with practical of snow avalanche mitigation, -training on ‘Snow and Avalanche skills for safe movement in high- safety precautions during any -Safety and Rescue Methods’ at altitude snow-bound avalanche- movement in avalanche-prone -Basic Training Centre Bhanu, prone regions. During the course, areas including tools and -Ramgarh, Panchkula (Haryana) newly trained ITBP personnel’s equipment’s used for avalanche -on 12 February 2026 and during participated and were briefed safety rescue and hands on -18-20 February 2026. The training about the role and responsibilities experience in ground. -Course on Emerging Technologies for Early warnings of Geo-Hazards -Defence Geoinformatics talks on hazard monitoring of emerging technologies— -Research Establishment (DGRE), technologies, data processing recognizing that solutions must be -Chandigarh, organized a course techniques, modeling approaches, context-specific, affordable, and -on “Emerging Technologies for risk communication strategies. sustainable for mountain regions. -Early Warnings of Geo Hazards” They have discussed both the Dr PK Srivastava, Sc ‘G’, was the -during 18-20 February 2026. opportunities and the limitations Course Director. -The objective of the course was to -enhance the knowledge on geo- -hazards and effective designing of -early warning system, application -of emerging AI-ML, sensor and -remote sensing technologies -for monitoring, detection and -early warning of geo-hazards in -Himalaya. A total of 21 participants -attended the course. In-house -experts of DGRE and eminent -speakers from various institutes -and academia delivered scientific -19 APRIL 2026 - -HRD ACTIVITIES DRDO NEWSLETTER -COURSE ON TQM AT IIM-MUMBAI -Directorate of Human Resource evolution of QM, beginning with fundamental of statistics required -Development (DHRD), DRDO the history of quality and quality for quality and process control -HQrs, Delhi, organized a three- philosophies. Participants were enabling them to understand -day Management Development introduced to the seven basic data-driven decision making in -Programme (MDP) on Total quality control tools, along with the operational environments. -Quality Management (TQM), for -Sc ‘C’, Sc ’D’, and Sc ’E’ at IIM, -Mumbai during 19-21 January -2026. A total of 25 participants -from various DRDO laboratories -attended the course. The program -was conducted to enhance -participant’s understanding of -modern quality practices and -their application in organizational -processes. The program provided -a comprehensive overview of the -COURSE ON CLOAKING MATERIALS AT DMSRDE -Defence Materials and designing defence systems and materials. The course concluded -Stores Research & Development products. During the course, with a valedictory address by Dr -Establishment (DMSRDE), eminent speakers from DRDO and DS Bag, Associate Director. Dr -Kanpur, organised a course on various institutes and academia Himanshu Bhushan Baskey, Sc -‘Stealth & Cloaking Materials for delivered scientific talks on recent ‘F’ was Course Director and Shri -Defence Platforms’ during 16-18 technological advancement Sanjeev Kumar, Sc ‘F’ was Deputy -February 2026. The course was in naval warfare and cloaking Course Director. -inaugurated by the Chief Guest -Dr Kingsuk Mukhopadhyay, OS -& Director, DMSRDE and the -Guest of Honour Dr Sampat R -Vadera, Distinguished Professor, -IIT-Jodhpur & Ex. Director DLJ. -The Chief Guest in his address -spoke about importance of -stealth in present scenario for -SKILL DEVELOPMENT PROGRAM ON AI-ML -APPLICATIONS IN DEFENCE SECTOR AT DTTC -A two-day specialized workshop during 11-12 February 2026. The students. The workshop was -under the skill development program was designed to provide inaugurated by the Chief Guest -program on “AI/ML Applications concept-oriented and hands- Dr Kingsuk Mukhopadhyay, OS & -in Defence Sector was organized on training for both industry Director, DMSRDE and the Guest -by DTTC (DMSRDE), Lucknow professionals and engineering of Honour Shri Alok Kumar Dixit, -APRIL 2026 20 - -HRD ACTIVITIES DRDO NEWSLETTER -Associate Director. In his welcome -address, Dr Ashish Dubey, Sc 'G', -talked about DTTC and its role in -supporting MSMEs and startups, -and briefed the participants about -| the Centre’s | | technical | facilities | | | | -| -------------- | --- | ---------- | ----------- | --- | --- | --- | -and mandate. -| A total | of | 23 | participants | | | | -| --------- | ----------- | ---------- | ------------- | --- | --- | --- | -| various | institute, | primarily | from | | | | -CSE background took part in the -program. - -INDUSTRY INTERACTION ON IMPLEMENTATION OF -E-CERTIFICATION -The Centre for Military & Public Interface) giving overview Issues Faced by D&D agencies", -Airworthiness & Certification of the program, followed by which was delivered by Shri Ram -(CEMILAC), Bengaluru, has Director (Systems) addressing the Singh, Sc 'G'. This was followed -developed an e-Certification gathering, wherein she elaborated by an extensive Q&A session, -System (e-CS) to facilitate a the aim and objectives of the wherein participants discussed the -web-based interface for design interaction along with its benefits. technical know-how, application & -agencies over the Internet and The technical session consisted troubleshooting aspects with the -the Intranet. The interaction of lecture on “Operational Aspects think-tank of CEMILAC e-portal -on e-certification system was of Design Interface and Various team. -| conducted | at | CEMILAC | on 14 | | | | -| -------------------- | ---------------- | -------------- | ----------- | --- | --- | --- | -| February | 2026. | Approximately | | | | | -| seventy | representatives | | from | | | | -| DRDO, | HAL, | BEL, | NAL, ADA, | | | | -| Services | and | various | private | | | | -| industries | participated | | along | | | | -| with more | than | three | hundred | | | | -| fifty participants | | joining | through | | | | -| Webex. | The | interaction | was | | | | -inaugurated by Director (Material -TALK ON MICROBES AT R&DE (E) -| Research | and | Development | | Pune, delivered | the lecture. | She | -| -------------- | ---- | ------------ | ------------ | ----------------- | ------------------ | ---- | -| Establishment | | | (Engineers) | spoke about | the significance | | -R&DE (E), Pune, organized a and benefits of microbes in our -talk on ‘Microbes: the Invisible daily life, especially their role -Superheroes’ as a part of their ‘The in acting as fundamental agents -Forum’ lecture series on 12 March for digestion, food production, -2026. Dr Jyutika Rajwade, Sc 'F' agriculture, medical industry, and -of Agharkar Research Institute, environmental nutrient recycling. -21 APRIL 2026 - -HRD ACTIVITIES DRDO NEWSLETTER -ORIENTATION PROGRAM FOR LATERALLY INDUCTED -SCIENTISTS -An orientation program for participants to go beyond their call Director, CFEES was the Chief -laterally inducted scientists of duty and contribute in nation Guest during the valedictory -was organized by Institute of building. During the course, function of the course. He -Technology Management (ITM), sessions were delivered on various felicitated the course participants -Mussoorie, during 5-16 January topics by eminent speakers from with certificates. Shri Akshay -2026 at Metcalfe House, Delhi. DRDO laboratories and academia. Lakhanpal, Sc ‘D was the Course -The course was designed by ITM Shri Arvind Kumar, OS & Director. -and DHRD for Scientist (C/D/E/F) -who have been laterally inducted -into DRDO. A total 51 participants -from different clusters of DRDO -| participated | | in the course. | The | -| ------------- | --- | ----------------- | ---- | -objective of the program was to -lay the foundations for the newly -| inducted | scientists | to | facilitate | -| ---------------- | ----------- | ------------ | ----------- | -| their | seamless | integration | into | -| the mainstream | | of | DRDO and | -helping them in amalgamating in -the organization by understanding -the systems, procedures in vogue -| and to | establish | peer | group | -| -------- | ---------- | ----- | ------ | -networking. -| The | Chief | Guest | V Adm | -| --------- | ------ | ---------- | --------- | -| Kalidoss | | Srinivas, | (Retd.), | -AVSM, NSM, VSM, DG (ATVP), -| in his | addressed | motivated | the | -| -------- | ------------ | ---------- | ---- | -COURSES AT ITM, MUSSOORIE -Certification Course on Project in Managed Environment -| The | PRIME-2 | course | was | -| ---- | -------- | ------- | ---- | -designed to equip future Project -| Directors | and | Deputy | Project | -| ----------- | ------------ | ------------ | ----------- | -| Directors | from | DRDO | with | -| essential | skills | in project | and | -| R&D | management | | conducted | -| by the | Institute | of | Technology | -| Management | | (ITM), | Mussoorie, | -| during | 19 January | 2026 | to 13 | -February 2026. Shri LC Mangal, -| DS & | DG (TM), | in his | address, | -| ------ | ---------- | -------- | --------- | -shared his critical views on all the -APRIL 2026 22 - -HRD ACTIVITIES DRDO NEWSLETTER -important parameters of project During the valedictory function, importance of efficient project -management practices within the Chief Guest Dr Ravindra management but also what all -DRDO and discussed that what all Singh, OS & DG (R&M) engaged new could be done to improve -is required and critical for success with the participants virtually and the efficiency in delivering DRDO -of any R&D project. discussed with them not only the projects. -Course on STEMS–25.2 -The mandatory training Industry collaborations in timely Dr Samir V Kamat, Secretary -programme ‘Specialized Training execution of the projects. Shri DDR&D & Chairman DRDO -on Essential Management Dobhal, Director ITM, deliberated was the Chief Guest during the -Skills’ (STEMS–25.2) for upon the new training policy valedictory function. -newly promoted Scientist 'G's 2025 of DRDO and importance He highlighted the -was organized by Institute of of networking among senior importance of technological -Technology Management (ITM), scientists across laboratories and innovation, adaptability and -Mussoorie, during 16-27 February clusters of DRDO through STEMS strategic foresight in a rapidly -2026 at ITM, Mussoorie. A course. Dr Goswami highlighted changing global conflicts and -total of 67 Scientists 'G' from the roadmap of DRDO during the war scenarios. -different laboratories/Centres/ session in the course. -Directorates of DRDO participated -in the course. The course was -inaugurated by the Chief Guest -Dr Sudhir Kamath, Director DIA- -RCoE IISc Bengaluru, the Guest -of Honour Dr Sumit Goswami, -Director, DP & C and Director -ITM Shri SP Dobhal. The Chief -Guest deliberated upon the -importance of DRDO-Academia- -BRAINSTORMING SESSION WITH DRDO SCIENTISTS -(PSYCHOLOGISTS) IN/AF -O/o SA to CAS conducted Experts highlighted the need for focus was given for developing -a brainstorming session with evidence-based interventions, accessible counselling modules -psychologists from IAF on 25 community awareness for IAF fraternity. The discussion -February 2026 at Air Force programme and the integration also explored early intervention -Station, New Delhi. The of mind education initiative into strategies for workplace wellness -interactive event brought together IAF training curricula. Special to foster a resilient workforce. -leading psychologists of IAF to -discuss the possible framework -for enhancement of wellness in -IAF. The session was chaired by -Smt Asha Tripathi, SA to CAS -who emphasized the growing -importance of psychological -resilience for service personnel. -23 APRIL 2026 - -HRD ACTIVITIES DRDO NEWSLETTER -AI IMMERSION PROGRAM AT NMRL -An AI Immersion Program Language Models, along with driven research capabilities within -was conducted at Naval Materials demonstrations of AI/ML tools. DRDO and associated defence -Research Laboratory (NMRL), The initiative aims to enhance AI- establishments. -| Ambernath, | during | 5-7 January | -| ---------------- | -------------------- | ------------- | -| 2026, | with participation | from | -| scientists | of NMRL | and R& | -| DE (E), | Pune, | along with | -| representatives | from | the Indian | -Navy, Mazagon Dock Shipbuilders -Limited, Mumbai, and NG India. -| The program | was | delivered by | -| ------------- | ---- | -------------- | -experts from GITAA Pvt. Ltd. in -collaboration with IITM Pravartak -| Technologies | Foundation | at IIT | -| ------------- | ----------- | -------- | -Madras, covering fundamentals of -Data Science, Machine Learning, -| AI, Generative-AI, | | and Large | -| -------------------- | --- | ----------- | -INTERNATIONAL CONFERENCE ON INTELLIGENT -SYSTEMS, SMART AND GREEN TECHNOLOGIES -| The | IEEE | International | -| ------------- | --------------- | -------------- | -| Conference | on | Intelligent | -| Systems, | Smart | and Green | -| Technologies | (ICISSGT-2026) | | -| for Defence | and | National | -| Development | was | organised | -under the joint aegis of IEEE Vizag -| Bay Section | and Naval | Science | -| ------------------- | -------------- | ----------- | -| and Technological | | Laboratory | -| (NSTL), | Visakhapatnam | during | -6-7 February 2026. In his opening -| address, | Dr RV Hara | Prasad, | -| ------------ | -------------- | -------- | -| DG (NS&M), | highlighted | that | -intelligent and autonomous advancing cutting-edge research, participation from senior DRDO -| technologies, | supported | by | -| -------------- | ---------- | --- | -testing and evaluation in support officials, scientists, academicians, -| sustainable | and energy-efficient | | -| ------------ | ---------------------- | --- | -of naval and defence systems. industry experts, and research -solution, are becoming key Prof. Debabrata Das, Director scholars from across the country. -enablers for next-generation IIIT, Bangalore delivered a Eminent experts delivered -defence systems. keynote address on “Information keynote addresses and technical -Dr Abraham Varughese, OS & -Scheduling Protocol Evolution presentations covering defence -| Director, | NSTL, highlighting | the | -| ---------- | -------------------- | ---- | -and 6G Network Slicing“ during oriented and dual-use applications -role of the laboratory, remarked the conference. of intelligent, smart, and green -that’s remains committed to ICISSGT-2026 witnessed technologies. -APRIL 2026 24 - -HRD ACTIVITIES DRDO NEWSLETTER -DIPAS ORGANIZES USER INTERACTIVE WORKSHOP -The Defence Institute of from the armed forces (including importance of structured follow- -Physiology and Allied Science MGS Branch, Armoured Corps, up activities, including one-on-one -(DIPAS), Delhi, organized a one- CE, AFMS, RVC, etc.) and CAPF engagements with user agencies, -day user-interactive workshop, (including CRPF, ITBP, SSB, BSF, to ensure clearer requirement -“Technologies for Habitability CoBRA forces) participated in mapping, enhance responsiveness, -and Sustenance in Extreme the workshop. DIPAS showcased and maintain strong, ongoing -Environments," on 27 February its products to the users. The partnerships with industry and -2026. Director, DIPAS, in his workshop highlighted the operational stakeholders. -inaugural address, stated the -importance of user interaction -and feedback in maintaining -the operational relevance of -technological interventions. -Shri Sangeet Singla, IDAS, IFA -(R&D), was invited as the Chief -Guest along with Dr Arindam -Basu (ex-DG, NITRA) as the -Guest of Honour. Director LIC, -DRDO Hqrs was also attended the -workshop. Over 25 participants -WORKSHOP ON BEST PRACTICES IN RECRUITMENT -AND ASSESSMENT AT RAC -The Recruitment and The inaugural ceremony emphasized tapping the talent -Assessment Centre (RAC), Delhi, was graced by Dr VK Saraswat, having ‘intellectual capability,' -organized a three-day workshop on Member NITI Aayog & Former providing the focused induction -best practices in recruitment and Secretary DRDO, as Chief Guest, training, and thereafter placing -assessment during 25-27 February and Dr Mayank Dwivedi, DG them in the right job. -2026. Subject experts from (HR), as Guest of Honour while Dr The event was successfully -various prestigious organisations Samir V Kamat, Secretary DDR&D organised under the able guidance -like DoAE, ISRO, BEL, TCS, SSB, & Chairman DRDO, joined the of Smt KV Prabha, OS & Director -Advisor (Cyber), DHTI, National ceremony as Distinguished Guest RAC, and Dr Atul Dev, Sc 'H', as -Law University, etc. were invited through VC mode. Dr Saraswat Convener of the workshop. -as eminent speakers to share their -respective organisational practices -on AI tools for R&A, psychological -tools, cybersecurity & DPDP Act, -legal aspects in R&A, etc. The -participants included scientists -from various DRDO laboratories -and other allied organizations, -including ADA, AFMC, ATVP, -CME, WESEE, DOAE, and SSB. -25 APRIL 2026 - -| HRD ACTIVITIES | | | | | | DRDO NEWSLETTER | | -| -------------- | --- | --- | --- | --- | --- | --------------- | --- | -CBRN EMERGENCY MANAGEMENT COURSE FOR NDRF -A Specialised CBRN Emergency -Management Training Course for -| the National | Disaster | Response | | | | | | -| -------------- | --------------- | ---------- | --- | --- | --- | --- | --- | -| Force | (NDRF) was | conducted | | | | | | -| during | 19-23 January | 2026 at | | | | | | -| Institute | of Nuclear | Medicine | | | | | | -| and Allied | Sciences | (INMAS), | | | | | | -Delhi. Sixty NDRF personnel from -| various | battalions | participated | | | | | | -| -------- | ----------- | ------------- | --- | --- | --- | --- | --- | -in the program. In the inaugural -session, Dr Himanshu Ojha, with special focus on medical by competency evaluation -Sc ‘F’, provided an overview of management, decontamination, exercise. The valedictory session -the course and fundamentals psychological impact, and stress highlighted mutual appreciation -| | | | management. | Hands-on | training | and reaffirmed | continued | -| --------- | --------- | -------------- | ------------ | --------- | --------- | ---------------- | ---------- | -| of CBRN | defence. | Experts from | | | | | | -using advanced equipment and an collaboration to strengthen -| DRDO | laboratories | and other | | | | | | -| ----- | ------------- | ----------- | --- | --- | --- | --- | --- | -agency as NIDM, delivered on-field radiation source search NDRF’s CBRN preparedness and -lectures on diverse CBRN aspects, exercise were conducted, followed response capabilities. -MHA-DRDO COLLABORATION: SIXTH APEX COUNCIL -MEETING -The sixth annual meeting of Border Surveillance System accorded for two new projects, -the Apex Council Steering the (BOSS). She also informed viz. Border Domination Games -MHA-DRDO collaboration was about prototype development Software (Seema Prahari) and -held on 2 March 2026 at DRDO of Digital Hand Held Radio Sets Mortar Locating Radar, assigned -HQrs, New Delhi. The meeting and Radio Frequency-Counter to DRDO laboratories ISSA and -was chaired by the Union Home Drone Systems by DRDO LRDE, respectively. The Council -Secretary Shri Govind Mohan, laboratories DEAL and DLRL, reviewed progress of various -IAS & Co-chaired by the Secretary respectively. These projects had advanced technological solutions -DD R&D & Chairman, Dr Samir V received AoN in earlier council presented by Shri Sangita Rao -Kamat. It witnessed participation meetings. Additionally, AoNs were Achary Addanki, Director DLIC. -| from | senior officers | of MHA, | | | | | | -| ----- | ----------------- | --------- | --- | --- | --- | --- | --- | -CAPFs, Delhi Police, NSCS, NDRF -and BPR&D. Dr (Mrs) Chandrika -Kaushik, DS & DG (PC&SI) and -| Member | Secretary | of the Apex | | | | | | -| ------- | ---------- | -------------- | --- | --- | --- | --- | --- | -Council, highlighted the progress -across various DRDO laboratories, -| including | QR formulation, | joint | | | | | | -| ---------- | ----------------- | ------ | --- | --- | --- | --- | --- | -trials and successful collaboration. -| The key | achievements | included | | | | | | -| -------------- | --------------- | --------- | --- | --- | --- | --- | --- | -| field | evaluation of | Man-Pack | | | | | | -| SDR, | Compact Trans | Horizon | | | | | | -| Communication | System | and | | | | | | -| APRIL 2026 | | | | 26 | | | | - -HRD ACTIVITIES DRDO NEWSLETTER -FORMER DIRECTOR DONATES BOOKS TO DGRE -| DGRE, | Chandigarh, | | had | | | -| --------------- | ------------ | --- | ----- | --- | --- | -| the privilege | of hosting | a | book | | | -dedication ceremony in honor of -| Dr Ashwagosha | | Ganju, Former | | | | -| --------------- | --- | --------------- | --- | --- | --- | -Director, DGRE, who generously -| donated | a valuable | collection | of | | | -| ------------ | ------------ | -------------- | --- | --- | --- | -| books (300 | Nos) | to the DGRE | | | | -library. The donated books cover -| important | domains | relevant | to | | | -| ------------- | ---------- | ----------------- | ----- | --- | --- | -| snow and | avalanche | research | | | | -| terrain | analysis, | and scientific | | | | -| advancement. | Dr | PK Satyawali, | | | | -| Director, | DGRE, | expressed | deep | | | -| appreciation | for | this thoughtful | | | | -gesture. -SPECIAL SALARY ACCOUNT PACKAGE CAMP AT DMS -A special salary account package -camp was successfully organized -| by DMS, | DRDO | HQrs on | 24 | | | -| --------- | ----- | --------- | --- | --- | --- | -February 2026 in association with -twelve Public Sector Banks (PSBs). -The camp was conducted in line -with the initiative of the Ministry -| of Finance, | which | has launched | | | | -| -------------- | ------- | -------------- | --- | --- | --- | -| the "Special | Salary | Account | | | | -Package for Central Government -| Employees" | across | all 12 Public | | | | -| ----------- | ------- | ---------------- | --- | --- | --- | -Sector Banks. The initiative aims -| to provide | enhanced | banking | | | | -| ------------ | --------- | -------- | --- | --- | --- | -| benefits, | improved | service | | | | -delivery, and customized financial -solutions to cent. govt. employees. -TALK ON FINANCIAL AWARENESS AT NPOL -| Naval | Physical | | and were | conducted | by SBI, Union | -| ------ | --------- | --- | ---------- | ------------ | ---------------- | -Oceanographic Laboratory Bank of India and IDBI on the -(NPOL), Kochi, organized financial subject “Benefits of Central Govt. -awareness class titled “Importance Salary Account Packages”. Miny G, -of Financial Planning in Service Sc ‘H’ and Director (Technology), -Life” on 17 February 2026. Dr delivered the felicitation address -Remadevi M, Sc ‘G’ welcomed the and highlighted the importance -gathering and briefly introduced and benefits of financial planning -| the topic. | The awareness | | class in service life. | | | -| ------------ | --------------- | --- | ----------------------- | --- | --- | -27 APRIL 2026 - -RAJBHASHA ACTIVITIES DRDO NEWSLETTER -RASHTRIYA KARMA YOGI JAN SEVA PROGRAM AT CABS -A one-day Rashtriya Karma and completed the one-day building for a Viksit Bharat 2047 -Yogi Jan Seva Training Program training programme and got their and key learnings. Ms P Santhya, -was conducted at Centre for certificates. OS & Director, CABS, appreciated -Airborne Systems (CABS), All the participants contributed the untiring efforts put in by Ms -Bengaluru during the months with enthusiasm while sharing Anasuya Sahu, Sc 'F', CABS, Ms -of September 2025 to January their role models, karmayogi J Gayathri, Sc 'F', CAIR and Shri -2026 in six batches. A total of moments, shared their Suresh Babu, Sc 'E', GTRE who were -287 employees were registered contributions towards nation the master trainers. -RAJBHASHA ACTIVITIES -RAJBHASHA WORKSHOP IN DRPPI -The Directorate of Rajbhasha, digital and technical environment. their digital skills in Hindi. More -Parliament & Public Interface The workshop witnessed than 50 participants have attended -(DRPPI) conducted a one-day participation from scientists, the workshop. The workshop was -workshop on 11 February 2026 at officers, and administrative staff coordinated by Ms Arun Kamal (AD -DRDO HQrs. Shri Vipin Kumar who were interested in enhancing OL). -| Kaushik, | OS & Director, | | DRPPI, | | -| ---------- | ----------------- | -------- | --------- | ----- | -| welcomed | the guest | speaker | | Shri | -| Rajendra | Prasad | Verma, | Deputy | | -| Director, | Central | Hindi | Training | | -Institute, Pune. The session began -| with | an introductory | | address | | -| ------------- | ----------------- | ----------- | -------- | --- | -| highlighting | the | importance | | of | -Hindi in the modern technological -era and the role of technology in -the progress of Hindi by Director -DRPPI. The workshop, highlighting -| the important | technical | | tools | for | -| --------------- | --------------- | --- | ------ | ---- | -| Hindi, | was organised | | with | the | -objective of promoting the effective -| use of | the Hindi | language | in | the | -| -------- | ----------- | --------- | --- | ---- | -APRIL 2026 28 - -RAJBHASHA ACTIVITIES DRDO NEWSLETTER -SECOND JOINT OFFICIAL LANGUAGE S&T SEMINAR -2025-26 OF SSS CLUSTER -The Second Joint Official The theme of the seminar physiological monitoring, -Language Scientific & Technical was “Contribution of Soldier rajbhasha and scientific areas -Seminar 2025-26 of Soldier Support Systems in Modern of DRDO. A total number of 54 -Support System (SSS) cluster Bharat” with sub-themes on bio- scientific papers and 11 posters on -was organized by Defence technology, life-support systems, the above areas were presented -Bioengineering & Electromedical human adaptation in space during the seminar by authors -Laboratory (DEBEL), Bengaluru and underwater environments, from different clusters of DRDO. -on 13 February 2026. O/o DG (SSS), -Laboratories/Establishments/ -Centers of SSS cluster were Co- -organizers of this seminar. The -seminar was inaugurated in the -presence of Dr UK Singh, DS & DG -(SSS) as Chief Patron, Shri Sanjeev -Singh, Chief Commissioner of -Income Tax (TDS), Bengaluru as -the Chief Guest and Shri Sangeet, -IFA (R&D) as the Guest of Honor. -RAJBHASHA SEMINAR OF ECS CLUSTER AT IRDE -Instruments Research & Implementation of Rajbhasha’. dignitaries released IRDE annual -Development Establishment The Chief Guest of the Seminar Hindi magazine ‘Sankalp’. In the -(IRDE), Dehradun, organised was Dr BK Das, DS & DG (ECS). valedictory session, outstanding -‘All India S&T Joint Rajbhasha Shri Lal Chand Mangal, DS & paper presenters were awarded -Seminar’ of Electronics and DG (TM); Dr Ravindra Singh, OS with commendation letters and -Communication Systems (ECS) & DG (R&M); In his inaugural delegates of different labs were -cluster of DRDO during 25- address, Dr BK Das highlighted awarded with mementos. A total -26 February 2026. The theme the importance of using Hindi of 194 participants attended this -of the seminar was ‘DRDO’s for writing scientific articles and seminar. Dr Sudhir Khare, OS & -Contribution and Challenges encouraged the participants to Additional Director, IRDE was the -for Modern Warfare Skills’. The continue promoting Hindi in Chief Seminar Coordinator while -sub-them was ‘Contribution of their academic and professional Dr Lalit Mohan Pant, Sc ‘G’, IRDE -Modern Technology in Effective endeavours. Dr Das and other was the Seminar Coordinator. -29 APRIL 2026 - -RAJBHASHA ACTIVITIES DRDO NEWSLETTER -HINDI WORKSHOP AT ISSA -The Institute for Systems -Studies & Analyses (ISSA), -Delhi, conducted one-day Hindi -workshop on 23 February -2026. Shri Kamlesh Kamal, -Public Relation Officer, ITBP -was invited to deliver a lecture -on “Karyalay Hindi: Shuddh -Prayog”. He elaborated the -fundamental principles of word -etymology, word formation in -Hindi and analyzed the underlying -meanings of linguistic structures. -He also discussed, common -errors frequently observed in -official noting, drafting and -correspondence. -SPORTS ACTIVITIES -DRDO NATIONAL LAWN TENNIS TOURNAMENT -Centre for Artificial Intelligence Doubles and Mixed Doubles 33 Prasanna Kumar, OS & Associate -& Robotics (CAIR), Bengaluru, players (6 Females and 27 Males) Director, CAIR distributed the -organized DRDO National Lawn from 4 different zones of DRDO prizes to the winners and runners. -Tennis Tournament for the year participated in the event. Shri -2025-26 during 5-8 January -2026. The opening ceremony -was held on 5 January 2026. Dr -Narayan Panigrahi, Sc ‘H’, CAIR -inaugurated the event in the -presence of Dr Rakesh Sahu, Sc ‘F’, -EMO and Ms Taniza Roy, Sc ‘F’, -LRDE, Chairperson, South Zone -Sports Council. The tournament -includes various categories of -events: Team Championship; -Women’s Singles and Men’s -Singles; Men’s Singles (veteran) -and Doubles (veteran); Men’s -APRIL 2026 30 - -VISITS DRDO NEWSLETTER -VISITORS TO DRDO LABORATORIES -CAIR, Bengaluru -Parliamentary Standing -Committee on Defence (SCoD), -led by Hon’ble Chairman Shri -Radha Mohan Singh visited -Centre for Artificial Intelligence -& Robotics (CAIR), Bengaluru, on -19 January 2026. Chairman and -member delegation comprising -of 16 Hon'ble Members of -Parliament and 7 Officials from -Lok Sabha Secretariat and -Ministry of Defence undertook -a study visit to CAIR. Dr Samir -V Kamat, Secretary DD R&D & -Chairman DRDO, Smt Sheena -Rani R, DS & DG (MCC), Shri -Ravindra Singh, OS & DG (R&M), -Shri Vipin Kumar Kaushik, Senior Scientists on Indigenization of Insensitive Munition (IM) -Director DPARO&M, Dr Rituraj of Pyro Cartridges, Indigenous technology for General Purpose -Kumar, OS & Director CAIR, Shri development of Insensitive (GP) bomb. -Prassana Kumar, OS & Associate Munitions (IM), Signal Flares He interacted with senior -Director, Shri Sanjeev Gupta, Sc (Green and Red) & Anti-Sonar scientists and discussed on various -'G' welcomed the Chairman and Decoy, IR Flare, Smoke Flares aspects of projects related to -member delegation. (Red & Green), Development of Indian Navy. He appreciated the -The committee witnessed Propellants, Igniters and Pyro contribution of HEMRL to Indian -several CAIR developed products Cartridges and Development Navy. -and technologies, through -demonstrations made by CAIR -scientists. Chairperson, SCoD -appreciated the presentation -made and the clarifications -provided to the queries. -HEMRL, Pune -Admiral Karambir Singh, -PVSM, AVSM (Retd.) Hon'ble -Advisor to Secretary DDR&D -and Chairman DRDO visited -High Energy Materials Research -Laboratory, Pune, on 27 -February 2026. During the visit, -presentations were made by the -31 APRIL 2026 - -VISITS DRDO NEWSLETTER -IRDE, Dehradun -Shri Ravinesh Kumar, IDAS, -PCDA (R&D) visited Instruments -Research & Development -Establishment (IRDE), Dehradun, -on 19 February 2026. He was -welcomed by Dr Ajay Kumar, OS -& Director, IRDE. Dr Ajay Kumar -presented major R&D initiatives -taken by IRDE and technologies -developed by IRDE. -He also briefed about various -products of IRDE which have been -developed and inducted into the -Services. Shri Ravinesh Kumar -appreciated the efforts of IRDE -in developing state-of-the-art -systems. -MTRDC, Bengaluru -A group of student officers -from Army Air Defence (AAD) -on 3 February 2026 and a group -of 28 Air force personnel from 16 -TS on 25 February 2026 visited -Microwave Tube Research & -Development Centre (MTRDC), -Bengaluru, as part of their -industrial demonstrative tour -to know about EMI/EMC and -technical electronic warfare -related facilities. -the troops, emphasizing discipline, opportunity for personnel to raise -NMRL, Ambernath -vigilance and professional concerns and receive guidance. -Col. Abhishek Kumar, SC, standards in the discharge of DSC A total of 30 NMRL DSC Jawans -Director DSC (MG&G), Mumbai duties. The interaction provided an participated in the program. -Area, visited Naval Materials -Research Laboratory (NMRL), -Ambernath on 12 February 2026. -During the visit, he conducted -physical fitness tests of the DSC -troops to assess their operational -readiness and discipline. -A “Sainik Sammelan” was -organized wherein he addressed -APRIL 2026 32 -Published by DESIDOC, Metcalfe House, Delhi-110 054 \ No newline at end of file +Online BSc Degree +and Diploma Program +PROGRAMMING AND DATA SCIENCE + +Batch 1: Starts January 2021 + +For latest updates, visit - www.onlinedegree.iitm.ac.in + +Brochure Last Updated: 2nd November 2020 + + Contents + +Introduction + +Why learn Programming & Data Science + +About IIT Madras + +Our Experience in Online Education + +Academics + +Overall Structure of the programme + +Term Structure + +Course Structure + +Assessments + + Online Assignments + + Three Quizzes + + End Term Exam + + Pass Criteria for Each Course + + Learners Who Do Not Pass a Course + +Exam Cities + +Fee Structure + +Foundational Level + +Diploma Level + + Diploma in Programming + + Diploma in Data Science + +Degree Level + +Admissions + +Important Dates + +Two Admission Paths + + Regular Entry + + Diploma Only Entry + +Mandatory Requirements + +Regular Entry + + Eligibility to Apply + + Application Process + + Qualifier Process + + Qualifier Exam and Pass Criteria + +Contact Us + +3 + +3 + +4 + +4 + +5 + +5 + +6 + +6 + +7 + +7 + +7 + +7 + +8 + +8 + +8 + +9 + +10 + +10 + +11 + +11 + +12 + +13 + +13 + +13 + +13 + +13 + +13 + +14 + +14 + +14 + +14 + +15 + +16 + + Introduction + +IIT Madras, India’s top technical institute, welcomes you to the world’s first online BSc +Degree program in Programming and Data Science. For the first time, you can work +towards an undergraduate degree / diploma from an IIT regardless of your age or +location, and with a wide range of academic backgrounds. + +“ + +IIT Madras is pioneering a completely online program to +produce graduates who will be employable in the sector +that presently has high demand - Programming and Data +Science. The program uses an innovative combination of +online learning and in-person assessment and is flexible and +affordable. I believe the program will be a gamechanger +in the field of education in India in the coming years and +inspire other top-ranking institutions to launch many such +online initiatives. + +- Prof. Bhaskar Ramamurthi, Director, IIT Madras + +Why Programming & Data Science? + +Decision making is increasingly becoming +data driven within the commercial sector +and the government. There is a large +quantity of data readily available for this +at present. Smart devices connected to +the internet bring more input sources, +apart from those generated by humans +as they interact with services and one +another. Data driven decision making +involves the analysis of large volumes +of data to identify patterns and build +predictive models. This requires a +combination of skills ranging from +computing, statistics and mathematics, +and is broadly labelled data science. + +INCREASING DEMAND + +BETTER JOB OPPORTUNITIES + +UPSKILL + +As formal academic programs in data +science are only just emerging, there +is a huge gap between the demand for +data scientists and the supply of suitable +qualified applicants in the job market. +The throughput of traditional classroom +programs is limited and will not be able +to provide the numbers needed to meet +the upcoming requirements. Online +teaching has the potential to train much +larger numbers, but delivering scale +without diluting quality has proven to be +a challenge until recently. + +Our program therefore aims to create +a thoroughly trained programmer, +proficient in application development, +data sciences and machine learning. + +3 + + About IIT Madras + +Indian Institute of Technology Madras is one of the +foremost institutes of national importance for higher +technical education and research. + +In 1956, the German Government offered +technical assistance in establishing +an institute of higher education in +engineering in India. The first Indo- +German agreement in Bonn, West +Germany for the establishment of the +Indian Institute of Technology at Madras +was signed in 1959. The Institute with its +campus in Chennai, previously known +as Madras, was formally inaugurated +in 1959 by Prof. Humayun Kabir, Union +Minister for Scientific Research and +Cultural Affairs. + +The Institute today has sixteen academic +departments and several advanced +research centres in various disciplines +of engineering and pure sciences. A +faculty of international repute, a brilliant +student community, excellent technical +& supporting staff and an effective +administration have all contributed to the +pre-eminent status of IIT Madras. + +Our Experience in Online Education + +IITM is well-equipped to provide an +online degree course. We have been +ranked number 1 Engineering Institute +for the past 5 years and we have been +ranked number 1 Overall for the past 2 +years by NIRF. + +Furthermore, IITM began the NPTEL +program, the country’s first online +content portal as an inter-IIT consortium +project in 2001 and nptel.ac.in  is one +of the largest online course portals in +the world. In 2014, IITM also created +NPTEL online courses - the country’s first +Massive Open Online Course (MOOC). +Today, NPTEL is considered the largest +MOOC provider in India. + +MHRD launched the SWAYAM MOOC +platform in 2017 and NPTEL has been +the largest contributor to it. NPTEL +also manages swayam.gov.in  - the +national MOOCs portal, as well as the +network of more than 3800 SWAYAM +local chapter colleges. With IITM +continuing to lead the NPTEL project, it +makes them the perfect choice to offer +an online degree program. + +2001 +NPTEL +Program + +2014 +NPTEL +Online +Courses + +2017 +SWAYAM +MOOC + +2020 +IITM +Online +Degree + +4 + + Overall Structure + +There are THREE LEVELS in our Online +Degree program and to get a BSc +Degree in Programming and Data +Science from IIT Madras, a learner +has to successfully complete all three +levels. + +1. +FOUNDATIONAL LEVEL + +2. +DIPLOMA LEVEL + +3. +DEGREE LEVEL + +There is also the flexibility to exit at +any level. Depending on the courses +completed, the learner can receive +Foundational Certificate from Centre +for Continuing Education, IIT Madras +or Diploma(s) from IIT Madras. (Pg 13 +for Admissions related information) + +We are currently formulating a +Diploma Only Entry (Pg 13) option for +those who are interested in doing only +Diploma Level courses and exiting +with one or both Diploma(s) from IIT +Madras. + +Total Courses: 31 +116 credits + +Completion time: 3 - 6 years +based on preferred pace & performance + +Fees: Based on courses registered +*Refer Fee Structure (Pg 9). + +Online Courses & Assignments +12 weeks of coursework for each course + +Quizzes and Exams +3 Quizzes and one End Term Exam for +each course - in person and invigilated. + +ACADEMICS + +5 + + Term Structure + +Every year is divided into three +terms - January Term, May Term and +September Term. + +Before progressing to the next level, +the learner has to clear all the courses +in the current level. + +In a term, a learner may register +for a course only if they have +already successfully completed the +prerequisites for that course. + +The learner may be allowed to +register for upto 4 courses in a term +depending on their performance in +the previous exam(s). + +A learner has to clear all Foundational +level courses successfully, before +registering for any Diploma Level +courses. Similarly, learners have +to clear Diploma Level courses +successfully before registering for any +Degree Level courses. + +Course Structure + +The duration of the core courses +in Foundational, Diploma and +Degree Level is 12 weeks and the +courses carry 4 credits each. +The Skill Enhancement courses +and Electives in Degree Level +have varying credits and may +be of different duration. + +The fee for each term depends on the +total credits one registers for in that +term. Refer Fee Structure (Pg 9). + +The recorded video content released +every week will comprise a series of +lectures, sample problems and step +by step instructions on how to solve +them. + +The duration of each video would be +15 to 30 minutes amounting to a total +of about 2 - 3 hours of video content +each week in each course. + +Weekly assignment(s), which will be +released along with the videos, will +have to be completed and submitted +online in the specified period of time. + +There will be 3 quizzes for each +course, one each at the end of weeks +4, 7 and 10. Each course further +culminates in an End Term Exam. All +the quizzes and the end term exams +have to be completed by the learner, +in person, at designated exam centres, +held under invigilation. + +All in-person, invigilated quizzes and +exams will be scheduled for weekends, +subject to availability of dates with +exam partner. + +6 + + Assessments + +There are 3 types of assessments for each course - online weekly assignments, +monthly in-person invigilated quizzes and an in-person invigilated end term exam. + +If the Average Assignment Score for a +course is less than 40/100, the learner +will not be eligible to write the end +term exam for that course. + +Online Assignments + +A 12-week course will have at least one +assignment every week. + +Each assignment has to be submitted +online within the specified due date. + +From the 12 weekly assignment scores +for a course, a learner’s best 8 scores +will be considered to calculate the +Average Assignment Score. + +Three Quizzes +IN PERSON, INVIGILATED + +There will be 3 quizzes for each course, one each at the end of Weeks 4, 7 and 10. +Every quiz will be based on content covered up to that point in the course. These +quizzes have to be attempted in person, in the specified exam city (Pg 9). + +Each quiz will be conducted in a single +session and will consist of sections of +45 minutes duration for every course +chosen. + +It is mandatory to appear for at least +one out of the three quizzes to be +eligible to write the end term exam. + +The best two out of the three quiz +scores will be considered to calculate +the Average Quiz Score (Q) out of 100. + +40% of Q will be used to calculate Total +Course Score T and course grade. + +End Term Exam + +IN PERSON, INVIGILATED + +At the end of a term, there will be an +end term exam of 3 hours duration for +each course. + +This exam score (out of 100) is called +End Term Score (E) which accounts for +60% of the Total Course Score (T). + +To be eligible to write the End Term +Exam, a learner’s average assignment +score should be >= 40/100 AND they +should have appeared for at least one +out of the three quizzes. + +7 + + Pass Criteria for Each Course + +Total Course Score (T) = 40% of Average quiz score (Q) + 60% of End Term Score (E) + +A learner is deemed to have passed a course only if the below mentioned two +conditions are met: + +Total Course Score (T) >= 50/100 AND + +End Term exam score (E) >= 40/100 + +These criteria are the same for all learners participating in the course. + +An absolute grading system, as described in the below table, will be followed: + +Condition satisfied by ‘T’ - Total Course Score (Out Of 100) +and ‘E’ - End Semester Score (Out Of 60) + +Letter Grade + +(T ≥ 90) AND (E ≥ 24) + +(90 > T ≥ 80) AND (E ≥ 24) + +(80 > T ≥ 70) AND (E ≥ 24) + +(70 > T ≥ 60) AND (E ≥ 24) + +(60 > T ≥ 50) AND (E ≥ 24) + +S + +A + +B + +C + +D + +(T < 50) OR (E < 24) OR Average assignment score <40/100 OR No quiz attended + +U (Fail) + +Learners Who Do Not Pass a Course + +For any learner, if Total Course Score (T) < 50/100 or average assignment +score < 40/100 or no quiz attended, they have to repeat the entire course. This +includes submission of assignments and redoing all the subsequent quizzes and +end term exam. Fees for repeating the course is the same as the course fees. + +If Total Course Score (T) >= 50/100 but End Term Score (E) < 40/100, the learner has +2 options: + +Attend only the end term exam whenever it is conducted in the subsequent +term. The fee to repeat an end term exam will be ₹1000 for foundational and +₹2000 for diploma / degree level courses. + +They can repeat the entire course by paying the full course fees and +completing all assignments, invigilated quizzes and the end term exam. + +Exam Cities + +The Invigilated Quizzes and End Term +exams are scheduled to be conducted +in a limited number of cities across +India. Learners can select 3 options +among the cities, and they will be +expected to attend the exam in the +City / Centre that has been allotted +to them. The allotted city will be from +these 3 options. + +The list of exam cities where +invigilated exams will be scheduled +may change later depending on the +location of majority of our learners. +Here is the link to view the current +exam cities . + +8 + + Fee Structure + +For details about application fees, check Application Process (Pg 14) section. + +Each term, pay only for the courses you register for! + +Foundational Level - ₹4000 per course. +Skill Enhancement courses in Diploma / Degree Level - ₹5,000 per course. +All other courses in Diploma / Degree Level - ₹10,000 per course. + +Fee waivers depend on category of learner and family income: + +Family Income > 5LPA + +Family Income > 1LPA +and <=5LPA + +Family Income <=1LPA + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +General + +Full Fee + +OBC + +Full Fee + +- + +- + +50% waiver + +EWS + Family +Income + +75% waiver + +EWS + Family +Income + +50%waiver + +OBC-NCL + +Family Income + +75% waiver + +OBC-NCL + +Family Income + +SC / ST + +50% waiver + +SC / ST + +50% waiver + +SC / ST + +75% waiver + +PwD + +50% waiver + +PwD + +50% waiver + +PwD + +75% waiver + +SC / ST + Family +Income + +EWS / OBC-NCL + +PwD + Family Income + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +The term family income for the purpose of availing fee waivers includes the income +of the candidate, the income of his/her parents and spouse, also the income of his/ +her siblings and children below the age of 18 years. + +OBC-NCL / EWS / Family Income certificates need to be obtained in the prescribed +format. Only those who clear Qualifier Exam and wish to avail fee waiver based on +family income need to submit the Family Income certificate. (Note that Family Income +certificate is valid only for one year and will need to be submitted afresh each year to +continue availing corresponding fee waiver.) +Format for OBC-NCL certificate  +Format for EWS certificate  +Format for Family Income certificate  + +Tentative Fee for Each Term, depending on your end Goal and possible paths to +reach the goal: + + Foundational Certificate  + + Diploma in Programming  | Diploma in Data Science  | Both Diplomas  + + BSc in Programming and Data Science  + +*Note: If a learner does not pass a course in the term they registered for it - refer +section about Learners who do not pass a course for details on whether they will +need to repeat entire course or only repeat the end term exam. Fee to repeat an entire +course will be the same as the full course fee. Fee to only repeat the end term exam is +₹1000 for foundational level courses and ₹2000 for diploma / degree level courses. + +9 + + Foundational Level + +The Foundational Level comprises +courses in Mathematics, Statistics, Basics +of Programing and Python, and English. +These courses have been chosen to +ensure that the learner who passes these +successfully is well prepared to proceed +to the Diploma Level courses. + +8 courses +32 credits + +1 - 3 years +10 hrs/course/week + +₹32,000* +*Refer Fee Structure (Pg 9). + +Requirements for registration + +The learner should apply for and clear the Qualifier Process (Pg 14). + +Options on successful completion + +Learners have the following two options when they successfully complete all 8 +Foundational Level courses: + +Exit: The learner may exit with a Foundational Certificate from Centre for +Continuing Education, IIT Madras. + +Proceed to next level: The learner can join the Diploma Level (Pg 10). + +Courses in Foundational Level + +Mathematics for Data Science I + +Statistics for Data Science I + +Mathematics for Data Science II + +Statistics for Data Science II + +Computational Thinking + +Programming in Python + +English I + +English II + +Diploma Level + +There are two sections in the Diploma +Level with courses for Diploma in +Programming and courses for Diploma +in Data Science. Each of these diplomas +comprise 5 core courses and 1 skill +enhancement course. These Diploma +Level courses assume that the learner is +well-versed with fundamentals. + +Requirements for registration + +6 + 6 courses +22 + 22 credits + +1 - 3 years +10 hrs/course/week + +₹55,000* + ₹55,000* +*Refer Fee Structure (Pg 9). + +The learner should have cleared all 8 Foundational Level (Pg 10) courses or should +have joined via Diploma Only Entry (Pg 13). + +10 + + Options on successful completion + +Learners have the following options based on the courses completed in this level: + +They may exit with a Diploma in Programming from IIT Madras. + +They may exit with a Diploma in Data Science from IIT Madras. + +They may exit with both Diplomas from IIT Madras. + +The learner can join the Degree Level (Pg 11) if they have completed all 12 +Diploma Level courses AND have not entered the program via Diploma Only +Entry option. + +Diploma in Programming + +The Diploma in Programming lays a +sturdy foundation in Databases and +Programming concepts with Data +Structures and Algorithms. The learner +goes on to apply these in the building +of a web application by the end of the +diploma. + +Courses for Diploma in Programming + +6 courses +22 credits + +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Database Management Systems + +Modern Application Development I + +Programming, Data Structures and +Algorithms using Python + +Modern Application Development II + +Programming Concepts using Java + +Skill Enhancement I + +Diploma in Data Science + +The Diploma in Data Science exposes +the learner to the holistic approach of +gathering, analysing, and interpreting +data for a variety of problems. The +courses on Business Data lays down the +context and the need for the data, while +the Machine Learning courses equip +the learner to use and analyse this data +towards impactful conclusions. + +Courses for Diploma in Data Science + +6 courses +22 credits + +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Machine Learning Foundations + +Business Data Management + +Machine Learning Theory + +Business Analytics + +Machine Learning Practice + +Skill Enhancement II + +11 + + Degree Level + +The Degree Level has courses that +complete the requirements for the BSc +degree. + +Requirements for registration +The learner should have come in via +Regular Entry (Pg 14), cleared all 8 +Foundational Level (Pg 11) courses and +all 12 Diploma Level (Pg 11) courses. + +11 courses +40 credits + +1 - 3 years +10 hrs/course/week + +₹1,00,000* +*Refer Fee Structure (Pg 9). + +Exit: Once the learner successfully completes 40 credits from Degree Level courses, +they can exit with a BSc Degree in Programming & Data Science from +IIT Madras. + +Credit Requirements +Learners have to complete a total of 40 credits at the degree level: + +4 credits from mandatory course titled “Strategies for Professional Growth” +4 credits from 2 Skill Enhancement Courses of 2 credits each. These are to be +completed on NPTEL and the credits transfered to this degree program. +8 credits from 2 Core Courses of 4 credits each. There are 3 pairs of core course +options to pick from. +24 credits from Elective Courses - 6 courses of 4 credits each or more courses of +lesser credits. + +Core Courses +Learner has to do a pair of core courses where each course is of 4 credits. Following +are the options for Core Courses: + +Core Course Option I + +Core Course Option II + +Core Course Option III + +Operating Systems + +Software Testing + +Computer Architecture + +Software Engineering + +AI: Search Methods for +Problem Solving + +Deep Learning + +Elective Courses +A total of 24 credits need to be obtained from Elective Courses. Listed below are 9 +courses of 4 credits each, including the 3 pairs of core courses. Excluding the 2 courses +that the learner has already picked as core courses, he / she can opt for the elective +courses in the following two ways: + +a. Obtain all 24 credits by picking 6 of the remaining 7 courses listed below. OR + +b. Obtain upto 12 credits from NPTEL (accepted courses list will be provided soon) and +obtain remaining credits by picking from a subset of the 7 courses below. + +Operating Systems + +Software Testing + +Computer Architecture + +Software Engineering + +Topics in Application +Development + +Topics in Machine +Learning + +AI: Search Methods for +Problem Solving + +Deep Learning + +Topics in Data Science + +12 + + ADMISSIONS + +Important Dates + +Applications Closed (Regular Entry) + +- 15 September 2020 + +Weekly release of Term 1 Qualifier courses +content starts + +- 5 October 2020 + +Qualifier Exam Hall Ticket (only for those who +get minimum required marks in Assignments) + +- 2 November 2020 + +Qualifier Exam (only for those with Hall Ticket)* + +- 22 November, 2020* + +Qualifier Exam Results + +- 7 December, 2020 + +Registration for Term 1 (only for those who clear +Qualifier Exam) + +- 7 to 27 December, 2020 + +Foundational Level Batch 1 starts + +- 4 January, 2021 + +*Dates for in-person exams are subject to government restrictions prevalent at that time. + +Two Admission Paths +1. Regular Entry + +Anyone joining the program via Regular Entry will join at the Foundational Level. + +The Foundational Level courses allow for interested applicants with a variety of +educational backgrounds to learn the basics (Math, Statistics, Computational Thinking, +Python programming and English) which help them take up the Diploma Level and +then the Degree Level courses. + +There are three exit options for those who enter the program via Regular Entry: +Exit 1. Foundational Level certificate from Centre for Continuing Education, IIT Madras. +Exit 2. Diploma in Programming AND/OR Diploma in Data Science from IIT Madras +Exit 3. BSc Degree in Programming and Data Science from IIT Madras + +View Regular Entry (Pg 14) for details on Eligibility, Application Process, Qualifier +Process, Qualifier Exam and Passing Criteria. + +2. Diploma Only Entry + +The Diploma Only Entry is primarily aimed at working professionals or learners who +already possess the fundamentals taught in Foundational Level courses and wish to +exit with one or two Diplomas (Diploma in Programming AND/OR Diploma in Data +Science) from IIT Madras. + +Learners entering via Diploma Only Entry can only do Diploma Level courses and +cannot proceed to the Degree Level courses. + +We have plans to open applications for Diploma Only Entry in May / June 2021. All +details related to Diploma Only Entry will be provided soon on our website . + +13 + + Mandatory Requirements + +Access to good internet connection as +well as a laptop/desktop device will be +a key requirement to learn effectively. + +Learner should be able to travel to +assigned exam centres for quizzes and +exams, each term. + +Regular Entry +Eligibility to Apply + +Eligibility for those who cleared class 12 or equivalent in 2018 or earlier: + +1. + +2. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2018 or earlier. + +Eligibility for those who cleared class 12 or equivalent in 2019 or 2020: + +1. + +2. + +3. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2019 or 2020. + +The candidate should be: +(i) currently enrolled / previously enrolled in any Bachelor’s Degree program in +campus mode OR +(ii) completed 1 year Diploma (campus mode only) OR +(iii) completed CA Inter + +Note that applicants who cleared class 12 or equivalent in 2019 or 2020 may apply for the +qualfier process with eligibility criteria 1 & 2 alone, but will be required to fulfil criterion 3 to +join the foundational courses after clearing the qualifying exam. + +Application Process + +Anyone who is eligible may apply by filling in the application form, uploading +required documents and paying the application fee. Check out Application +Process  section on our website for the latest updates and to start applying. + +Application fee - ₹3,000 (For applicants belonging to PwD (>= 40% disability) / SC / ST +category - ₹1,500; For applicants belonging to PwD (>= 40% disability) and also SC / +ST category - ₹750) + +Qualifier Process + +All applicants will have to go through a Qualifier Process, like a trial month, +wherein they will get access to 4 weeks of content for the four foundational level +courses - English I, Mathematics for Data Science I, Statistics for Data Science I and +Computational Thinking. + +Video content of 2 - 3 hours duration will be released for each course on a weekly +basis. Along with the content, the first three weeks will also include an assignment +for each course that need to be completed and submitted online within the given +deadline. + +Average assignment score in each course will be calculated. Only those who get +minimum required average assignment scores in all four courses will be allowed to +write the Qualifier Exam (refer next page). + +14 + + Minimum Average Assignment +Score required in each course + +General Learner + +SC / ST / PwD with 40% disability + +PwD with 40% disability & SC / ST + +OBC-NCL / EWS + +40% + +30% + +30% + +35% + +Qualifier Exam and Passing Criteria + +Only learners who get minimum required average assignment scores in all four +courses will be eligible for the in-person invigilated Qualifier Exam conducted at the +end of 4 weeks of coursework. Only these candidates will receive hall tickets. + +The Qualifier Exam is of 3 hours duration and covers all 4 subjects with 45 minutes +allotted per course. + +To pass the Qualifier Exam, the learner has to get a minimum Total Qualifier Exam +Score and a minimum Qualifier Exam Score in each course individually. Refer table +below. + +Only those who pass the Qualifier Exam will be eligible to register for Foundational +Level courses. + +Minimum required +Qualifier Exam Score +in each course + +Minimum required +Total Qualifier +Exam Score + +General Learner + +SC / ST / PwD with 40% disability + +PwD with 40% disability & SC / ST + +OBC-NCL / EWS + +40% + +30% + +30% + +35% + +50% + +40% + +40% + +45% + +Note: Relaxations in pass criteria indicated for various categories of learners, is +applicable ONLY for the qualifier process. There will be no relaxations in terms of +grades / pass criteria once registered into the program. + +Though everyone who passes the Qualifier Exam will be allowed to register for +Foundational Level, there will be limitations on the number of courses a learner will +be allowed to register for in the first two terms based on the Total Qualifier Exam +Score (S). + +Total Qualifier Exam Score (S) + +Number of courses allowed to +register for in first two terms + +Minimum required <= S < 50% + +up to 2 courses + +50% <= S < 70% + +up to 3 courses + +S >= 70% + +up to 4 course + +Maximum number of courses a learner will be allowed to register for in later terms +will depend similarly on the average of course scores in the previous two terms. + +15 + + Contact Us + +Online Degree Office, 3rd Floor, ICSR Building, + +IIT Madras, Chennai - 600036 + +7850-999966 (Mon-Fri 9am-6pm) + +support@onlinedegree.iitm.ac.in + +www.onlinedegree.iitm.ac.in  + +iitmadrasonline + +16 + diff --git a/q181/random_k2/question.json b/q181/random_k2/question.json index 683ff1357d61526c21eb7cae485da9a7376deae3..fb6b1644936e7305f3ce70361d5af7c4f9318418 100644 --- a/q181/random_k2/question.json +++ b/q181/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "b3m5kaeqm2w8n4bwcesw-140602121350-phpapp02_95.pdf", - "web_e9f399fea7556d59.pdf", - "web_b0b861dfece3f50d.pdf" + "web_17b633ba09ee110f.pdf", + "web_18e98cd838edeb2e.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q181/random_k2/random_1.md b/q181/random_k2/random_1.md index d3d5a7efa18017d2cc9d64bd996d8e5da4aa2173..261b89724e814d1be04e011ae16d4387de2cce1a 100644 --- a/q181/random_k2/random_1.md +++ b/q181/random_k2/random_1.md @@ -1,3125 +1,767 @@ -User Guide - -Lenovo ThinkBook 14 2-in-1 Gen 4 - - Read this first - -Before using this documentation and the product it supports, ensure that you read and understand the -following: - -• Generic Safety and Compliance Notices - -• Safety and Warranty Guide - -• Setup Guide - -First Edition (February 2024) - -© Copyright Lenovo 2024. - -LIMITED AND RESTRICTED RIGHTS NOTICE: If data or software is delivered pursuant to a General Services -Administration “GSA” contract, use, reproduction, or disclosure is subject to restrictions set forth in Contract No. GS- -35F-05925. - - Contents - -About this guide . . . . . . . . . . . . . iii - -. - -. - -. - -. - -. . - -Left - -Base . - -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. -Microphones . . -Infrared LED . . -Camera shutter . -. . -Camera . -. -. . -Camera light -. . -. -Screen . -. . -. -Antennas -. . -. -. -. . -. . -Keyboard . -. . -Touchpad . -. . -. -. . - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -Multi-purpose USB Type-C connector . -. -. . -. -Charging light . . -. -USB Standard-A connector -. -. . -Always-on connector . -. -. . -. -HDMI connector . -. -. -. . -. -Combo audio jack . -. -. -. . -. -. -Speakers -. . -. -. . -. -. -. -. -. . -. -. . -. -. -Speakers -. -. . -. -. . -. -. -. . -Power light . -. -. . -. -Power button . . -. -. -. . -. -Fingerprint sensor . -. -microSD card slot . -. . -. -. -USB Standard-A connector -. -. . -Security lock slot -. -. . -. . - -Chapter 1. Meet your computer . . . . . 1 -. 1 -Front . -. 2 -. 2 -. 2 -. 2 -. 2 -. 2 -. 2 -. 3 -. 3 -. 4 -. 5 -. 5 -. 5 -. 6 -. 6 -. 6 -. 6 -. 6 -. 7 -. 7 -. 7 -. 8 -. 8 -. 8 -. 8 -. 8 -. 9 - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -Avoid constant body contact with specific hot -. 9 -. -. -. -. . -sections . -. 9 -. -. -Air vents (intake) . -. -. -. -Novo button hole . -. 9 -. 11 -. -. -. . -. -. 11 -. -. -Air vents (outlet) . -. 12 -. 14 -. 14 - -. . -. . -. . -. . -. . -. . -Statement on USB transfer rate . . -. . -Operating environment . . - -. . -. . -. . -. . -. . -Features and specifications . . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. - -Bottom . . - -. -. -. -. -. -. - -Right . - -Rear - -. . - -. . - -. -. - -. - -. - -. - -. - -. - -. - -Chapter 2. Get started with your -computer . . . . . . . . . . . . . . . . 15 -. 15 -Connect to a network . -. 15 -. 15 -. 16 - -. -Connect to a Wi-Fi network . -Establish a wired connection . -. - -Unique Lenovo apps . - -. . -. . -. . -. . - -. . -. . -. . -. . - -. -. -. -. - -. -. -. -. - -. . - -. . - -. - -. - -. +FOR RELEASE MARCH 18, 2021 -Work with Windows . +BY Rakesh Kochhar -The Novo Button menu . . +FOR MEDIA OR OTHER INQUIRIES: -Interact with your computer . -. +Rakesh Kochhar, Senior Researcher -. +Stefan Cornibert, Communications Manager -. . +Tanya Arditi, Communications Manager -. . +202.419.4372 -. -. -. -. -. -. -. -. +www.pewresearch.org -. -. -. -. -. -. -. -. +RECOMMENDED CITATION -. . -. . -. . -. . -. . -. . -. . -. . - -. -Lenovo Vantage . -Lenovo Smart Meeting . -. - -. . -. . -. . -Windows help information . . -. . -Open the Novo Button menu . -. . -. . - -. -. -. -. -. -. -. -Hotkeys . . -. -. . -Touch operations supported by Windows . -Connect to an external display . -. -Protect your privacy using the camera -. . -shutter -. -. . -Enroll your fingerprints . -Lenovo Slim Pen (for selected models) . - -. . -. . - -. -. -. - -. . - -. . - -. . - -. -. - -. -. - -. - -. - -. - -. - -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. - -. 16 -. 16 -. 18 -. 19 -. 19 -. 20 -. 20 -. 20 -. 21 -. 23 - -. 24 -. 24 -. 24 - -. . -. . - -Chapter 3. Explore your computer . . 27 -. 27 -. -Intelligent features . -. -. 27 -Gesture controls. -. -. 27 -Smart Noise Cancelling . -. 28 -. -Super Resolution . . -. 28 -. -. . -Eye Care Mode . -. 29 -. -. . -. -. 29 -. 30 -. 30 -. 31 - -. . -. . -. . -. . -. . -. . -. -Rechargeable battery pack . -Set power button behavior. . -. . -A power plan . -System operation modes . . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -Manage power . - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. . - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. -. - -. . - -. . - -. . - -. . - -. . - -. . - -. 33 - -Chapter 4. Help and support -. -Frequently asked questions . - -. . . . . 33 -. 33 -. -. . -How do I partition my storage drive . . -. 33 -. -What should I do if my computer stops -responding . -. . -What should I do if I spill liquid on the -. -computer -Where can I get the latest device drivers and -UEFI/BIOS . -. -. -. . -Why does my computer start automatically -when I open the lid . . -. -. -. . -. -. . -. -. -. -. -. +Pew Research Center, March, 2021, “The Pandemic Stalls +Growth in the Global Middle Class, Pushes Poverty Up Sharply” -. -. . -. -. . -. -. . -. -. . -. -. . -. -Before you contact Lenovo . -. -Lenovo Customer Support Center . -. + 1 -. -. -Self-help resources . -What is a CRU? . -. -. -CRUs for your product model -. -Call Lenovo . . +PEW RESEARCH CENTER -. 33 -. 33 -. 35 -. 35 -. 36 -. 36 -. 36 -. 37 +About Pew Research Center -. . -. . -. . -. . -. . -. . -. . -. . +Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes +and trends shaping America and the world. It does not take policy positions. The Center conducts +public opinion polling, demographic research, content analysis and other data-driven social +science research. It studies U.S. politics and policy; journalism and media; internet, science and +technology; religion and public life; Hispanic trends; global attitudes and trends; and U.S. social +and demographic trends. All of the Center’s reports are available at www.pewresearch.org. Pew +Research Center is a subsidiary of The Pew Charitable Trusts, its primary funder. -Purchase additional services. +© Pew Research Center 2021 -. -. -. -. -. -. -. -. +www.pewresearch.org -. -. -. -. -. -. + 2 -. 33 +PEW RESEARCH CENTER -. 33 +How we did this -. . +The COVID-19 pandemic has created an economic crisis, shuttering businesses and cutting jobs +around the world. This report looks at how the downturn has affected the standard of living +globally and in seven major regions. -. . +The focus is on the distribution of people across five income tiers in 2020: poor, low income, +middle income, upper-middle income and high income. See the methodology and an earlier Pew +Research Center report for more on the definition and meaning of these tiers in a global context. -. . +The key data source for the analysis is the World Bank’s PovcalNet database, which provides +access to household survey data on either income or consumption for more than 160 countries. At +the regional level, the latest year for which comprehensive data are available ranges from 2014 for +South Asia to 2018 for East Asia and the Pacific and three other regions. The global regions are as +defined by the World Bank. -. . +These benchmark estimates of income are extrapolated to 2020 using World Bank estimates of +output growth through 2020. One projection is based on the World Bank’s January 2020 forecasts +of economic growth in 2020, and the other is based on its January 2021 estimates of growth in +2020. The difference between these two measures is used to represent the effect of the pandemic +on the income distribution in each region. Regional estimates are summed to yield estimates of +the global distribution. -. +www.pewresearch.org -. + 3 -. +PEW RESEARCH CENTER -. +BY Rakesh Kochhar -. +The COVID-19 pandemic is having a deep effect on the global economy. In January 2020, as +reports of the novel coronavirus were emerging, the World Bank forecasted that the global +economy would expand by 2.5% that year. In January 2021, with the pandemic still holding much +of the world in its grip, the World Bank estimated that the global economy contracted by 4.3% in +2020, a turnabout of 6.8 percentage points. -. +The economic downturn is likely to have +diminished living standards around the world, +pushing millions out of the global middle class +and swelling the ranks of the poor. At the same +time, the path to a recovery is clouded with +uncertainties. -. +A new Pew Research Center analysis finds that +the global middle class encompassed 54 million +fewer people in 2020 than the number +projected prior to the onset of the pandemic. +Meanwhile, the number of poor is estimated to +have been 131 million higher because of the +recession. -. +The drop-off in the global middle class was +centered in South Asia and in East Asia and the +Pacific, and it stalled the expansion seen in the +years preceding the pandemic. South Asia, +specifically India, along with Sub-Saharan +Africa, accounted for most of the increase in +poverty, reversing years of progress on this +front. -Appendix A. Function descriptions -for Lenovo hotkeys . . . . . . . . . . 39 +The COVID-19 downturn curbed growth +in the global middle class, increased +poverty sharply in 2020 -© Copyright Lenovo 2024 +Estimated change in the number of people in each +income tier due to the global recession in 2020, in +millions -i +High income - Appendix B. Notices and -trademarks . . . . . . . . . . . . . . . 41 +-62 -ii +million -User Guide +Upper-middle +income - About this guide +-36 -• This guide applies to Lenovo product model(s) listed below. Illustrations in this guide may look slightly +Middle income -different from your product model. +-54 + +Low income + ++21 -Model name +Poor -Machine type (MT) ++131 -ThinkBook 14 2-in-1 G4 IML +Notes: The poor live on $2 or less daily, low income on $2.01-$10, +middle income on $10.01-$20, upper-middle income on $20.01- +$50, and high income on more than $50; figures expressed in 2011 +purchasing power parities in 2011 prices. The estimates show the +difference in the number of people in an income tier based on pre- +pandemic projections and post-pandemic estimates. The term +“post-pandemic” refers to the period since the onset of the +pandemic in January 2020.It is assumed there is no change in the +income distribution in a region from the benchmark year for the +projection to 2020. +Source: Pew Research Center analysis of data from the World Bank +PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes +Poverty Up Sharply” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 4 + +PEW RESEARCH CENTER + +As defined in this report (and in previous Pew Research Center analyses), people who are middle +income live on $10.01-$20 a day, which translates to an annual income of about $14,600 to +$29,200 for a family of four. This is modest by the standards of advanced economies. In fact, it +straddles the official poverty line in the United States − about $23,000 for a family of four in 2020 +(expressed in 2011 prices). By global standards, the poor live on $2 or less a day, or no more than +$2,920 annually for a family of four. + +The number of people in the global high-income tier (more than $50 daily) is estimated to have +decreased by 62 million in 2020, erasing about half of the gain since 2011, with most of the change + +Who is middle class (or middle income)? +This report uses the terms “middle income” and “middle class” interchangeably, a common practice among +economists who tend to define the middle class in terms of income or consumption. But being middle class +can connote more than income, be it a college education, white-collar work, economic security, owning a +home, or having certain social and political values. Class could also simply be a matter of self-identification. +The interplay among these many factors is examined in studies by Hout (2007) and Savage et al. (2013), +among others. + +The population in each global region is divided into five groups: poor, low income, middle income, upper- +middle income and high income. The poor live on $2 or less daily, low income on $2.01-$10, middle income +on $10.01-$20, upper-middle income on $20.01-$50 and high income on more than $50. All dollar figures +are expressed in 2011 prices and purchasing power parity dollars, currency exchange rates adjusted for +differences in the prices of goods and services across countries. -ThinkBook 14 2-in-1 G4 IML 1 +The assignment to a group, or income tier, is based on a household’s daily per capita income or +consumption, a simple way of controlling for differences in household size. In the source data, some +countries only report income data while others only report consumption data. The terms “income” and +“consumption” are used interchangeably for the sake of convenience. -ThinkBook 14 2-in-1 G4 IML 2 +A middle-income, or middle-class, threshold of $10 follows the practice of other researchers. A similar +threshold has been used by the World Bank (2007, 2015), researchers at the Organization for Economic +Cooperation and Development (Kharas, 2010), the development community (Birdsall, Lustig and Meyer, +2013) and others. There is evidence that the $10 threshold, which is about five times higher than the World +Bank’s global poverty line, is associated with economic security and “insulates” people from falling back +into poverty. An earlier Pew Research Center report discusses the choice of income thresholds in greater +detail. -21MX +The global middle-class standard is modest by the standards of advanced economies. In 2020, the official +poverty line for a family of four in the U.S. stood at about $15.90 per person per day (in 2011 prices). Thus, +many who are poor by the U.S. definition would meet the global middle-income standard or, conversely, +many who are in the global middle class would be poor by U.S. standards. -21MX +www.pewresearch.org -21MX + 5 -• For further compliance information, refer to the Generic Safety and Compliance Notices at +PEW RESEARCH CENTER -https://pcsupport.lenovo.com/docs/generic_notices. +emanating from advanced economies. Meanwhile, the upper-middle income population ($20.01- +$50 daily) fell by 36 million, while the low-income population ($2.01-$10 daily) is estimated to +have increased by 21 million. -• This guide may contain information about accessories, features, and software that are not available on all +These estimates are based on the Center’s analysis of World Bank data on the distribution of +people by either income or consumption levels in seven major global regions. The latest available +estimates of income for each region, such as for 2018 for Europe and Central Asia and 2015 for the +Middle East and North Africa, are extrapolated +to 2020 using two sets of growth estimates: the +forecasts issued by the World Bank in January +2020, prior to the pandemic, and the estimates +issued in January 2021, with the pandemic’s +economic impact in plainer sight. The +difference between these two measures is taken +to represent the impact of the COVID-19 +downturn on the standard of living of people +globally. -models. +Globally, most are either poor or low +income, and the pandemic likely drove +the numbers higher in 2020 -• This guide contains instructions that are based on the Windows operating system. These instructions are +Estimated number of people in each income tier in 2020, +before and after the global recession, in millions + +Pre-pandemic projection + +Post-pandemic estimate + +High income + +531 million + +593 + +Prior to the emergence of COVID-19, some 1.38 +billion people were expected to be counted in +the global middle class in 2020. But the +pandemic is estimated to have driven this +number down to 1.32 billion. The share of +people in the middle class globally is estimated +to have been 17.1% in 2020, instead of +potentially 17.8%. The erosion in the middle +class might have been deeper if not for the fact +that China – which is home to more than one- +third of the global middle class – evaded an +economic contraction, even though growth +there was slower than anticipated. + +The number of global poor is estimated to have +risen to 803 million in 2020, much greater than +the 672 million initially expected. The global +poverty rate, which had been in steady decline +this century, is likely to have increased to + +Upper-middle +Upper-middle +income +income + +Middle +income + +Low income + +1,140 + +1,176 + +1,324 + +1,378 + +Poor + +803 + +672 + +3,956 -not applicable if you install and use other operating systems. +3,936 + +Notes: The poor live on $2 or less daily, low income on $2.01-$10, +middle income on $10.01-$20, upper-middle income on $20.01- +$50, and high income on more than $50; figures expressed in 2011 +purchasing power parities in 2011 prices. It is assumed there is no +change in the income distribution in a region from the benchmark +year for the projection to 2020. The term “post-pandemic” refers to +the period since the onset of the pandemic in January 2020. +Source: Pew Research Center analysis of data from the World Bank +PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes +Poverty Up Sharply” -• Microsoft® makes periodic feature changes to the Windows® operating system through Windows Update. -As a result, the operating system related instructions may become outdated. Refer to Microsoft resources -for the latest information. +PEW RESEARCH CENTER -• The content of the guide is subject to change without notice. To obtain the latest version, go to +www.pewresearch.org -https://support.lenovo.com. + 6 + +PEW RESEARCH CENTER + +10.4%, nearly reverting to the rate in 2017, instead of sinking to a new low of 8.7%, as previously +expected. + +The steep rise in global poverty is driven by the fact that many who were in the low-income tier +prior to the pandemic lived on the margin of poverty. In 2017, 877 million people lived on $2 to $3 +a day, according to World Bank estimates. Thus, large numbers were susceptible to falling back +into poverty prior to the pandemic, especially in South Asia and Sub-Saharan Africa. -© Copyright Lenovo 2024 +At the other end of the economic spectrum, the global high-income population likely fell to 531 +million in 2020 from the 593 million originally expected. Consequently, the expected share in the +high-income tier is 6.8%, not 7.6%. A vast majority of the global high-income population lives in +advanced economies, and this group of countries also experienced a sharp downturn in 2020. +Ironically, this bolstered the global middle-income population as people in advanced economies +slipped down the ladder from higher income tiers. -iii +The reversal in global living standards in 2020 comes on the heels of notable progress earlier in +the decade. From 2011 to 2019, the global middle-class population increased from 899 million to +1.34 billion, or by 54 million people annually, on average. The pandemic is estimated to have +erased a year of growth, +leaving the global middle-class +population nearly unchanged +from 2019 to 2020. + +The global middle class expanded rapidly from 2011 to +2019, and poverty fell sharply + +Number of people in each income tier, in millions + +Global poverty had abated at +an average rate of 49 million +annually from 2011 to 2019, +with the number of global +poor falling from 1.10 billion +to 691 million during this +period. The pandemic, by +adding 131 million to the ranks +of the poor, has set the +progress on poverty back by +several years, and the poverty +rate is estimated to have +increased from 9.0% in 2019 +to 10.4% in 2020. - iv +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on +$10.01-$20, upper-middle income on $20.01-$50, and high income on more than $50; +figures expressed in 2011 purchasing power parities in 2011 prices. It is assumed there is +no change in the income distribution in a region from the benchmark year for the projection +to 2019. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -User Guide +PEW RESEARCH CENTER - Chapter 1. Meet your computer +www.pewresearch.org -Front + 7 -No. +PEW RESEARCH CENTER -1 +From 2011 to 2019, the global high-income population had increased from 459 million to 576 +million, growing by an average of 15 million annually. Thus, the setback of 62 million due to the +pandemic in 2020 also reverses several years of growth in the high-income population. The share +of the global population in the high-income tier is expected to have decreased from 7.5% in 2019 +to 6.8% in 2020. -2 +The effect of the COVID-19 downturn on the middle class and other income tiers shows distinct +regional differences. Asia is expected to have accounted for most of the shrinking of the global +middle class in 2020, with the middle-class population falling by 32 million in South Asia and by +19 million in East Asia and the Pacific. Meanwhile, the middle class in advanced economies is +estimated to have increased by 16 million, in part due to the drop-off in the high-income tier in +those countries. -3 +South Asia is estimated to have experienced the largest decrease in the middle +class and the greatest increase in poverty due to the COVID-19 downturn in 2020 + +Estimated change in the number of people in each income tier due to the global recession in 2020, in millions + +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on $10.01-$20, upper-middle income on $20.01-$50, and +high income on more than $50; figures expressed in 2011 purchasing power parities in 2011 prices. The estimates show the diff erence in the +number of people expected in an income tier based on projections made before and after the onset of the pandemic in January 2020. It is +assumed there is no change in the income distribution in a region from the benchmark year for the projection to 2020. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 8 + +PEW RESEARCH CENTER + +The projected rise in global poverty in the pandemic is concentrated in South Asia (an increase of +78 million) and Sub-Saharan Africa (40 million), historically among the poorest regions in the +world. The high-income population in advanced economies is estimated to have shrunk by 47 +million in 2020, accounting for most of the falloff globally. It should be noted that the global cutoff +for the high-income tier – more than $50 a day – is not much higher than the median income +across advanced economies, as an estimated 40% of the population in these countries is in the +global high-income tier. Thus, the contraction of the high-income tier in advanced economies is +more a reflection of the reduction in the standard of living of the typical person in these countries +than outcomes for the wealthy. + +The contribution of a region to global changes overall and the shift in its own income distribution +depends mostly on two factors: the sizes of income tiers in the region prior to the COVID-19 +recession and the impact of the downturn on economic growth in the region. + +Prior to the pandemic, nearly half of the global middle-class population – 672 million of 1.38 +billion – was estimated to live in East Asia and the Pacific, a region that includes China. Latin +America and the Caribbean accounted for 191 million, and Europe and Central Asia was home to +183 million. South Asia accounted for 123 million more. + +In view of these magnitudes, the estimated loss (relative to pre-pandemic expectations) of 32 +million people from the middle class in South Asia is notable for two reasons. First, the region +contributed the most to the shrinking of the global middle class in 2020 despite having a middle- +income population smaller than several other regions. Second, the percentage decrease in South +Asia’s middle-income population from what was previously expected (26%) is much greater than +in any other region. Although East Asia and the Pacific shed 19 million from the middle-income +tier, that represented only 3% of its middle class. -4 +Poverty is most prevalent in Sub-Saharan Africa. Some 494 million people in Sub-Saharan Africa, +out of a total population of 1.14 billion, were expected to be living in poverty ahead of the +pandemic in 2020. South Asia, with a total population of 1.86 billion, was expected to be home to +104 million poor. Nonetheless, South Asia is estimated to have added nearly double the amount of +people to the ranks of the globally poor as Sub-Saharan Africa in the pandemic. The percentage +increase in poverty in South Asia (75%) dwarfs the increase in Sub-Saharan Africa (8%). -5 +The number of poor in other regions was smaller, estimated to range from 6 million in Europe and +Central Asia to 27 million in Latin America and the Caribbean ahead of the pandemic in 2020. +But, since that moment, the increase in poverty in these regions is notable. The Middle East and +North Africa and East Asia and the Pacific are each estimated to have seen increases in poverty of -6 +www.pewresearch.org -7 + 9 -* for selected models +PEW RESEARCH CENTER -Description +Prior to the pandemic, most of the global middle class lived in East Asia and the +Pacific, the poor in Sub-Saharan Africa, high-income earners in advanced economies -Microphones +Estimated number of people in each income tier prior to the global recession in 2020, by region, in millions -Infrared LED* +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on $10.01-$20, upper-middle income on $20.01-$50, and +high income on more than $50; figures expressed in 2011 purchasing power parities in 2011 prices. It is assumed there is no change in the +income distribution in a region from the benchmark year for the projection to 2020. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -Camera shutter +PEW RESEARCH CENTER -Camera +more than 20% due to the recession. The rise in poverty in both Europe and Central Asia and +Latin America and the Caribbean is not far behind, with each experiencing a 17% increase in +poverty. -Camera light +A large majority of the global high-income population – 489 million of 593 million – lived in +advanced economies before the global recession took hold. The estimated decrease of 47 million, +or 10%, in this population in advanced economies accounts for most of the shrinking of the global +high-income tier. -Screen +In proportional terms, the drop off in the high-income tier is also notable in Latin America and the +Caribbean (15%), Europe and Central Asia (12%) and East Asia and the Pacific (12%). The losses in +the Middle East and North Africa, South Asia and Sub-Saharan Africa are relatively small in +magnitude (each less than 1 million people) but also represent double-digit shares of the high- +income populations in those regions. -Antennas +www.pewresearch.org -© Copyright Lenovo 2024 + 10 -1 +PEW RESEARCH CENTER - Microphones +The relatively outsized role of +South Asia in the contraction of +the global middle class and the +expansion of poverty is the +result of it seeing the sharpest +reduction in economic growth +in the pandemic. In January +2020, the World Bank +forecasted that South Asia +would experience a 4.3% +increase in gross domestic +product (GDP) per capita in +2020. But, in January 2021, +output in South Asia was +estimated to have decreased by +7.8% in 2020, a retreat of 12.1 +percentage points from +potential. -The microphones are the PC's built-in sound input devices. They capture your voice and ambient sound and -convert them into digital form. Microphones are essential components when you use your PC for video -conferencing or voice recording. +South Asia estimated to have seen the sharpest drop +in output growth in 2020, East Asia and the Pacific +the least -Infrared LED +Growth in real gross domestic product per capita in 2020, forecasted and +estimated, in percent -The infrared LED generates and emits near-infrared waves that are received and used by a camera (or a -dedicated infrared camera) for facial recognition. +Global -Camera shutter +Advanced economies -The camera shutter is a sliding cap that you can move to block the camera lens. +Latin America and the Caribbean -Note: The camera shutter is designed for privacy protection. When the camera lens is blocked, the camera -function is disabled. +East Asia and the Pacific -Camera +Europe and the Central Asia -The built-in camera captures visible light and converts it to digital signals. It is used for video recording and -video conferencing. +South Asia -The camera for some models can also detect near-infrared waves. For those models, an IR LED is also -included to emit near-infrared waves. They are used together to achieve face-based authentication. +Middle East and North Africa -Camera light +Sub-Saharan Africa -The camera light indicates whether the camera is activated. +Jan. 2020 +forecast -Table 1. Camera light status and description +Jan. 2021 +estimate Difference -Camera light status + 2.5% -Description +1.0 -On +0.8 -Off +5.1 -Screen +2.3 -The camera is activated. +4.3 -The camera is not activated. +0.7 -The screen of the built-in display is where text, graphics, and videos are displayed. +0.3 -The screen is touch-enabled, allowing you to interact with your PC by intuitively touching buttons, icons, and -menu items displayed on the screen. Touch-enabled screens also support multi-finger gestures. +-4.3 -Antennas +-5.8 -The antennas transmit and receive radio waves to allow data to be transferred between your PC and a Wi-Fi -network device or a Bluetooth device. +-7.8 -Note: The antennas are hidden inside the PC. +0.4 -2 +-3.2 -User Guide +-7.8 - Base +-6.5 -No. +-6.1 -1 +-6.8 -2 +-6.8 -Keyboard +-8.6 -Description +-4.7 -Keyboard +-5.5 -Touchpad +-12.1 -The keyboard is the PC's built-in input device for typing characters. It also includes keys that can make you -more productive when interacting with the PC, apps, and the Windows operating system. +-7.2 -Table 2. Productivity keys +-6.4 -Key +Notes: Percentages shown are for the change in real GDP per capita at market prices. The +estimate for advanced economies is computed as the percent change in GDP less the +percent change in population. +Source: World Bank, Global Economic Prospects, January 2021 and January 2020 editions. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -Location on the keyboard +PEW RESEARCH CENTER -Function +The setback to economic +growth in other regions is +smaller, ranging from 4.7 percentage points in East Asia and the Pacific to 8.6 points in Latin +America and the Caribbean. East Asia and the Pacific is the only region in which growth is +estimated to have stayed positive in 2020, with GDP per capita inching up by 0.4% even after the +onset of the pandemic. Otherwise, the latest estimates of the change in output per capita in 2020 +ranged from -3.2% in Europe and Central Asia to -7.8% in Latin America and the Caribbean. -Function keys (F1 through -F12) +www.pewresearch.org -In the top row + 11 -Hotkeys +PEW RESEARCH CENTER -In the top row +Acknowledgments -Fn key +This report is a collaborative effort based on the input and analysis of the following individuals. -In the lower-left corner +Rakesh Kochhar, Senior Researcher +Kim Parker, Director, Social Trends Research +Richard Wike, Director, Global Attitudes Research +James Bell, Vice President, Global Strategies +J.J. Moncus, Research Assistant +Aidan Connaughton, Research Assistant +David Kent, Senior Copy Editor +Travis Mitchell, Digital Producer +Michael Keegan, Senior Information Graphics Designer +Shannon Greenwood, Digital Producer -Windows key +www.pewresearch.org -In the lower-left corner + 12 -Executing predefined commands or -functions for selected apps +PEW RESEARCH CENTER -• Changing some of the PC's frequently +Methodology -used settings +The starting point for the analysis is household survey-based estimates of regional income +distributions from the World Bank’s PovcalNet database. The global regions are as defined by the +World Bank. In this report, “advanced economies” refers to the group of countries listed as “other +high income” by the World Bank. -• Opening an app or a Windows gadget +These data were used to estimate a benchmark distribution of the population in each of seven +major global regions across five income tiers. The income tiers are defined by the daily per capita +income or consumption of people in a region as follows: poor ($2 or less daily), low income +($2.01-$10), middle income ($10.01-$20), upper-middle income ($20.01-$50) and high income +(more than $50). All dollar figures are expressed in 2011 prices and purchasing power parity +dollars. -• Toggling the functions of dual-function +The year for which the benchmark estimates were drawn was the latest one in which the +household surveys covered at least 75% of the region’s population. This varied across regions as +follows: 2018 in advanced economies, Latin America and the Caribbean, East Asia and the Pacific, +and Europe and Central Asia; 2017 in Sub-Saharan Africa; 2015 in the Middle East and North +Africa; and 2014 for South Asia. -keys +Extrapolating the benchmark estimates to 2020 required the following three steps: -• Used with several letter keys to change +1. Extrapolate benchmark-year incomes to 2020 +2. Estimate income distributions in 2020 based on extrapolated incomes +3. Determine the population in each income tier in 2020 -some of the PC's frequently used -settings +Extrapolating incomes from benchmark years to 2020 -Used with certain keys to change some -frequently used Windows settings +Household incomes are extrapolated to 2020 using estimates of the growth in real gross domestic +product (GDP) per capita in a region. This, in turn, required several sets of data. -Hotkeys share keys with function keys. To toggle the functions for these keys, hold down the Fn key or turn -on the FnLock switch. +The first is the level of GDP per capita in the benchmark year, say 2017 for Sub-Saharan Africa, +and in 2019, the latest year for which data are available in the World Bank’s World Development +Indicators database. In this database, GDP per capita is expressed in constant 2017 international +dollars. The result is an estimate of the percentage change in per capita income in a region from +the benchmark year to 2019. -Chapter 1. Meet your computer +Next, the change in income from 2019 to 2020 is approximated by two measures of growth in real +GDP per capita (at market prices) issued by the World Bank in its reports on global economic -3 +www.pewresearch.org - Note: Keyboard layout varies by language and geographical location. The keyboard on your PC may look -slightly different from the product illustrations of this publication. + 13 -Touchpad +PEW RESEARCH CENTER -The touchpad is the PC's built-in pointing device, which provides the basic functionality of an external -mouse. Slide your finger on the touchpad to move the pointer on the screen and tap or double-tap to select -or execute a screen item. +prospects. The first measure of growth in 2020 is a forecast released in January 2020, prior to the +emergence of the pandemic. The second measure is an estimate released in January 2021, with the +economic effects of the pandemic more apparent. The result is two estimates of how much per +capita income changed in each region from the benchmark year to 2020, one in a counterfactual +“pre-pandemic world” and the other in a “post-pandemic” world, the period since the emergence +of the coronavirus. -The touchpad also supports Windows multi-finger gestures, which provide shortcuts to frequently used apps -and functions. +In a slight variation to this procedure, the change in per capita income in 2020 for advanced +economies is estimated as the difference in the growth in GDP and the growth in population. This +is because estimates of the change in GDP per capita in 2020 for this group of countries is not +directly reported in the source data. -4 +Estimating income distributions in 2020 based on extrapolated incomes -User Guide +The estimated change in income from the benchmark year to 2020 is used to reset the boundaries +of the income tiers. For example, based on the pre-pandemic forecast, it was estimated that per +capita income in East Asia and the Pacific would have increased by 10.5% from 2018 to 2020. As a +result, people living on between $1.82 to $2 a day in East Asia and the Pacific in 2018 would have +seen their income increase to more than $2 a day in 2020. Thus, they would no longer be poor in +2020; only people with incomes of $1.81 or less in 2018 would still be poor in 2020. Thus, $1.81 is +used as the poverty boundary to derive an estimate of the share who were poor in East Asia and +the Pacific in 2020 in the hypothetical pandemic-free world. - Left +With the emergence of the pandemic, the growth in per capita income in East Asia and the Pacific +in 2020 is estimated to have been 5.6%. In this circumstance, only people with incomes from +$1.90 to $2 a day in 2018 would have emerged from poverty in 2020. Thus, $1.89 is used as the +poverty boundary to estimate the state of poverty in East Asia and the Pacific in 2020 in the post- +pandemic scenario. -No. +A similar procedure is used to reset the boundaries of all income tiers in all regions. The +estimation of income distributions for 2020 was greatly facilitated by the “povcalnet” command in +Stata developed by World Bank researchers. (Castaneda Aguilar, R.A., C. Lakner, E.B. Prydz, J.S. +Lopez, R. Wu and Q. Zhao, Estimating Global Poverty in Stata: The Povcalnet Command, +September 2019). -1, 5 +www.pewresearch.org + + 14 + +PEW RESEARCH CENTER + +Determining the population in each income tier in 2020 -2 +The estimates derived from PovcalNet yield the percentages of people in a region who are poor, +low income and so on. These percentages are applied to the estimated population of each region to +determine the number who fall into each income tier. -3 +The overall population in a region in 2020 is derived from two sources: World Bank data on +population through 2019, as listed on PovcalNet, and United Nations forecasts of the change in +population from 2019 to 2020. This exercise is conducted at the country level, some 218 in all, and +the results are summed to obtain an estimate of the population in each region in 2020. -4 +Key assumptions and potential uncertainties + +As is the case with any projection, the estimates reported in this analysis are subject to a degree of +uncertainty. One issue is that final end-of-year estimates of real GDP growth are not yet available +for most countries, if any. Thus, the January 2021 estimates from the World Bank that are used in +the analysis are likely to be revised in the coming months. In its projections of the increase in +extreme poverty ($1.90 or less a day) in 2020, the World Bank assumes a global contraction of +either 5% or 8%. The resulting change in extreme poverty in 2020 is estimated to range from 88 +million to 115 million. + +The methodology in this analysis assumes that incomes change at the same rate for all people, +whether poor, low income, middle income, upper-middle income or high income. In other words, +it is assumed that there is no change in inequality from the benchmark year to 2020. If the +COVID-19 recession has worsened inequality, the increase in the number of poor is likely greater +than estimated in this analysis, and the decrease in the number who are high income is likely less +than estimated. The middle class may have shrunk by more than projected. + +Although World Bank researchers have simulated the relationship between inequality and the +poverty rate, more precise measures of changes in inequality in the COVID-19 era will not be +forthcoming until newer household survey data become available. Also, the extent to which social +spending by governments, especially in advanced economies, may have ameliorated the effects of +the COVID-19 downturn is not yet clear. -6 - -7 - -Description - -Multi-purpose USB Type-C® connector - -Charging light - -USB Standard-A connector (always-on) - -HDMI™ connector - -Combo audio jack - -Speaker - -Multi-purpose USB Type-C connector - -This USB Type-C® connector is the PC's power input connector. Use the included power adapter and this -connector to supply power to the PC. - -When this connector is not used by the included power adapter, it can also be used to connect: - -• Storage or peripheral devices that follow the universal serial bus (USB) specification for data transfer and - -device interconnection - -• Display devices - -Note: When connecting display devices, you need to use appropriate cables and adapters (if needed) -according to the connection capabilities of the display device. - -• Thunderbolt™-enabled docks or devices (for the connector with an icon next to it) - -Charging light - -The charging light indicates whether the PC is plugged into an electrical outlet. When the PC is plugged into -an electrical outlet, the color of the light indicates whether the battery is fully charged (or will shortly be fully -charged). - -Chapter 1. Meet your computer - -5 - - Table 3. Charging light statuses and descriptions - -Light status - -Off - -On, amber - -On, white - -Plugged in? - -Battery charge level - -No - -Yes - -Yes - -/ - -1%–90% - -91%–100% - -USB Standard-A connector - -The USB Standard-A connector is used to connect storage or peripheral devices that follow the universal -serial bus (USB) specification for data transfer and device interconnection. - -Always-on connector - -A USB connector with a battery icon ( -USB device connected to this type of connector even when the PC is powered off, in sleep mode, or in -hibernation mode. - -) supports the always-on function. The PC can supply power to a - -The always-on function can be turned on and off in: - -• The PC's firmware setup utility, or - -• Lenovo Vantage or Lenovo PC Manager - -HDMI connector - -The HDMI connector is used to connect an external display device, such as a television, a projector, or a -monitor. - -Combo audio jack - -The combo audio jack is used to connect single-plug headsets, headphones, or external speakers. - -Speakers - -The speakers are the PC's built-in sound output devices. - -6 - -User Guide - - Right - -No. - -1 - -2 - -3 - -4 - -5 - -6 - -Description - -Speaker - -Power light - -Power button (with an embedded fingerprint sensor) - -microSD™ card slot - -USB Standard-A connector - -Security lock slot - -Speakers - -The speakers are the PC's built-in sound output devices. - -Power light - -The power light indicates the current power state of the PC: whether it is powered on, powered off, in sleep -mode, or in hibernation mode. - -When the PC is powered on, this light can also indicate low battery by blinking rapidly. - -Table 4. Power light status and description - -Light status - -White (solid on) - -White (blinking rapidly) - -Power state - -Powered on - -Powered on - -White (blinking slowly) - -In sleep mode - -Off - -Powered off or in hibernation mode - -Battery charge level - -21%–100% - -1%–20% - -/ - -/ - -If the PC's power button includes an embedded fingerprint sensor, the power light turns solid green to -prompt you to enroll or scan your fingerprint. - -Chapter 1. Meet your computer - -7 - - Power button - -Press the power button to turn on your PC. - -Note: By default, on a Windows PC, pressing the power button when the PC is turned on will put the PC into -sleep mode. - -Fingerprint sensor - -The fingerprint sensor scans your finger to unlock your PC or verify your identity. - -Note: You need to enroll one or more fingerprints before using them for identity verification. On a Windows -PC, go to Settings ➙ Accounts ➙ Sign-in options to enroll your fingerprints. - -microSD card slot - -The microSD card slot is used to insert a microSD, microSDHC, or microSDXC memory card to transfer data -between the memory card and your PC. - -USB Standard-A connector - -The USB Standard-A connector is used to connect storage or peripheral devices that follow the universal -serial bus (USB) specification for data transfer and device interconnection. - -Security lock slot - -The security lock slot is used to attach a compatible cable lock. The cable lock typically includes a loop at -one end and can be used to secure the PC to a stationary object, preventing theft when the PC is left -unattended for a short period of time in cafes, shops, libraries, and other public places. - -8 - -User Guide - - Bottom - -No. - -1 - -2 - -Description - -Air vents (intake) - -Novo button hole - -Avoid constant body contact with specific hot sections -CAUTION: -When the computer is operating, it should be placed on a hard and flat surface with its bottom area -not in contact with user’s bare skin. Under normal operating conditions, the temperature of the -bottom surface will remain within an acceptable range as defined in IEC 62368-1, but such -temperatures can still be high enough to cause discomfort or harm to the user if directly touched for -over one minute at a time. As such, it is recommended that users avoid prolonged direct contact with -the bottom of the computer. - -Air vents (intake) - -The air vents allow air to be sucked inside of the PC to cool the internal components. - -Important: When the PC is operating, do not place it on a bed, sofa, carpet, or other flexible surfaces. -Otherwise, the air vents will be blocked and the PC may overheat, reducing performance or causing the PC -to be unresponsive or even shut down. - -Novo button hole - -When the PC is powered off, you can press the Novo button to display the Novo button menu. From the -menu, you can then choose to: - -• Open the firmware setup utility - -• Display the boot device selection menu - -• Display the Windows advanced startup options page - -Chapter 1. Meet your computer - -9 - - Note: The Novo button is rarely used during normal PC operations. To prevent users from accidentally -pressing it, the Novo button is placed in a recessed hole. You can use a straightened paper clip to press this -button. - -10 - -User Guide - - Rear - -No. - -1 - -Description - -Air vents (outlet) - -Air vents (outlet) - -The air vents allow hot air to be discharged out of the PC. - -Important: When the PC is operating, do not place it on a bed, sofa, carpet, or other flexible surfaces. -Otherwise, the air vents will be blocked and the PC may overheat, reducing performance or causing the PC -to be unresponsive or even shut down. - -Chapter 1. Meet your computer - -11 - - Features and specifications - -Dimensions - -Width - -Depth - -Thickness - -ac power adapter - -Input - -Output - -Power - -Rechargeable battery pack - -Capacity - -Cell type - -312.85 mm - -223.5 mm - -• Thinnest: 16.85 mm - -• Thickest: 18.7 mm - -100 V ac–240 V ac, 50 Hz–60 Hz - -20 V dc, 3.25 A - -65 W - -60 Wh - -Lithium-polymer - -Number of cells - -4 - -Note: The battery capacity is the typical or average capacity as measured in a specific test environment. -Capacities measured in other environments may differ but are no lower than the rated capacity (see product -label). - -Memory - -Type - -Installation - -Number of slots - -Mass storage device - -Type - -Slot type - -Interface - -Display - -DDR5 - -Slot mounted - -2 - -Solid-state drive (SSD) - -• M.2 (2242), or - -• M.2 (2280) - -PCIe Gen 4 - -Screen size (diagonal) - -14 inches - -Resolution - -Refresh rate - -1920 × 1200 pixels - -60 Hz - -12 - -User Guide - - Connectors and slots - -USB Standard-A connector - -• Quantity: 2 - -• Maximum power output: 5 V, 0.9 A - -• Supported signaling protocols: - -– USB 2.0 480 Mbps - -– SuperSpeed USB 5 Gbps - -Multi-purpose USB Type-C -connector - -• Quantity: 2 - -Note: The always-on connector supports maximum power output of 5 V and 2 A. - -• Maximum power output: 15 W (5 V, 3 A) - -• Maximum power input: 65 W - -• Supported signaling protocols: - -– USB 2.0 480 Mbps - -– SuperSpeed USB 5 Gbps - -– SuperSpeed USB 10 Gbps - -– DisplayPort™ 1.4 - -HDMI connector - -Combo audio jack - -– Thunderbolt 4 41.25 Gbps (for the connector with an icon next to it) - -Supported signaling protocol: Transition minimized differential signaling (TMDS) - -• Diameter: 3.5 mm - -• Supported plugs: - -– 3-pole, TRS - -– 4-pole, TRRS (CTIA and OMTP) - -microSD slot - -Supported card types: - -• microSD™ - -• microSD High Capacity (microSDHC™) - -• microSD Extended Capacity (microSDXC™) - -Security lock slot - -Kensington Nano Security Slot™ - -Note: Data rates and performance ratings are dependent on connected devices and cables if they are used. -USB Type-C connectors that are DisplayPort 1.4 compliant through DisplayPort alternate mode provide a -maximum output resolution of 5120 x 3200, at a 60 Hz frame rate and color depth of 24 bpp (bits per pixel). -The actual maximum output resolution is dependent on the connected display device and the cable being -used. - -Networking - -Wi-Fi® - -Bluetooth - -• Wi-Fi 6, or - -• Wi-Fi 6E - -Bluetooth 5.2 - -Note: Support of Bluetooth 5.3 may require a future operating system update. - -Chapter 1. Meet your computer - -13 - - Statement on USB transfer rate - -Depending on many factors such as the processing capability of the host and peripheral devices, file -attributes, and other factors related to system configuration and operating environments, the actual transfer -rate using the various USB connectors on this device will vary and will be slower than the data rate listed -below for each corresponding device. - -USB device - -USB 3.2 Gen 1 - -USB 3.2 Gen 2 - -Data rate (Gbit/s) - -5 - -10 - -Operating environment - -Maximum altitude (without pressurization) - -3048 m (10 000 ft) - -Temperature - -• At altitudes up to 2438 m (8000 ft) - -– Operating: 5°C to 35°C (41°F to 95°F) - -– Storage: 5°C to 43°C (41°F to 109°F) - -• At altitudes above 2438 m (8000 ft) - -– Maximum temperature when operating under the unpressurized condition: 31.3°C (88°F) - -Note: When you charge the battery, its temperature must be no lower than 10°C (50°F). - -Relative humidity - -• Operating: 8% to 95% at wet-bulb temperature 23°C (73°F) - -• Storage: 5% to 95% at wet-bulb temperature 27°C (81°F) - -14 - -User Guide - - Chapter 2. Get started with your computer - -Connect to a network - -Connect to a Wi-Fi network - -Ensure that you have a secure Wi-Fi network account and the required credentials. - -Step 1. Select the network icon - -on the bottom right of your display. - -Step 2. Select an available network, and then select Connect. If you want to be automatically connected to - -this Wi-Fi network the next time you start your computer, select Connect automatically before -selecting Connect. - -Step 3. - -Input your credentials if necessary, and then follow the on-screen instructions to connect to the -disired Wi-Fi network. - -Establish a wired connection - -Step 1. Plug an Ethernet cable into the Ethernet connector on your computer. - -Step 2. Plug the other end of the Ethernet cable into a network wall jack or a router. - -Note: If your computer does not include an Ethernet connector, you can purchase a USB-C to Ethernet -adapter from Lenovo at https://www.lenovo.com/accessories. - -© Copyright Lenovo 2024 - -15 - - Unique Lenovo apps - -Lenovo Vantage -Lenovo Vantage is a one-stop solution to help you update your computer, configure hardware settings, and -access personalized support. - -If your computer is pre-installed with Lenovo Vantage, type Vantage in the Windows search box to launch -this app. - -Notes: - -• Available features may vary depending on your computer model. - -• You can download the latest version of this app from Microsoft Store. - -Lenovo Smart Meeting -Lenovo Smart Meeting is a video conferencing app with multiple features for enhancing your professional -image, protecting your privacy, and reducing your computer's power consumption. - -If you want your settings in this app to also take effect on other mainstream video conferencing apps, such -as Microsoft Teams and Zoom, ensure that you select Lenovo Virtual Camera in the app. - -Access the app - -Type Lenovo Smart Meeting in the Windows search box and then press Enter. - -Explore key features - -• Smart appearance - -– Video enhancer: Adjust the brightness automatically for better image quality during the video call. - -– Face framing: Keep your face centered automatically during the video call when you move around. - -– Customized background: Blur or customize your background during the video call to protect your - -privacy. - -16 - -User Guide - - – Dual Camera Fusion: Show the image of any participant that is visible from the front or rear of the - -camera on the screen when collaborating with a Lenovo Magic Bay 4K Webcam. - -Note: You can purchase the Lenovo Magic Bay 4K Webcam from Lenovo at https://www.lenovo.com/ -accessories. - -– Temporary Avatar: Create and display a temporary portrait of you as if you were still in the video - -conference when you are temporarily away. - -• Intelligent sensing - -– Presence detection: Keep your computer awake when you are facing the screen and lock the - -computer screen automatically when you leave. You can customize the screen-off time from 10 to 60 -seconds. - -– Privacy protection: Blur your screen and display a warning message when a shoulder surfer is - -detected. - -Chapter 2. Get started with your computer - -17 - - – Adaptive dimming: Track your face movements to recognize your focus of attention. When you are - -facing away, the display dims to conserve power. You can customize the brightness value and screen -dimming timer. - -Notes: - -• Lenovo does not collect any personal data from this app. - -• The available features vary depending on the computer model. - -• Lenovo Smart Meeting makes periodic feature updates to keep improving your experience. The - -description described here might be different from that on your actual user interface. - -Work with Windows - -The following table contains Windows frequently used settings. You can learn the basics and start working -with Windows right away. - -To configure settings, type the corresponding keywords in the Windows search box and select the best -match. Follow the on-screen instructions to customize settings. - -Table 5. Basic settings - -Functions - -Descriptions - -Control Panel - -View or change Windows settings, including hardware and software setup -and configuration. - -Connect to Wi-Fi -networks - -For models with a wireless LAN module, you can connect your computer to -a Wi-Fi network. Click the network icon in the Windows notification area, -and then select a network for connection. -Note: The wireless LAN module on your computer may support different -standards. For some countries or regions, use of 802.11ax may be disabled -according to local regulations. - -Search by -keywords - -Control Panel - -Wi-Fi - -Airplane mode - -Airplane mode is a convenient setting to turn all wireless communications of -your computer on and off. You may need to turn it on when boarding an -airplane. - -Airplane mode - -18 - -User Guide - - Table 5. Basic settings (continued) - -Functions - -Descriptions - -Night light mode - -Adjust color -temperature - -Windows -Updates - -Night light is a switch in Windows that you can turn on and off. When it is -turned on, your screen shows warmer colors and the amount of emitted -blue light is reduced. Enabling night light reduces the chances of developing -eye fatigue or eye strain. -Note: For more guidance on reducing eye fatigue or eye strain, go to -https://www.lenovo.com/us/en/compliance/visual-fatigue. - -If night light mode is turned on, you can adjust the color temperature of the -screen. -Note: Selected Lenovo computers are low blue light certified using night -light mode and color temperature settings available in the pre-installed -operating system. These computers are tested with night light turned on -and color temperature set to 48 or greater. - -Microsoft periodically releases feature and security updates to the Windows -operating system. Updates that are applicable to your Windows version are -downloaded automatically when your computer is connected to the -Internet. When updates are downloaded, you are prompted to restart the -computer to install these updates. You can also manually check if there are -available updates for the installed version of Windows. -Attention: Only use Windows Update to download and install updates. -Updates downloaded from other sources may include security risks. - -Table 6. Optional settings - -Functions - -Descriptions - -Search by -keywords - -Night light - -Night light - -Windows Updates - -Search by -keywords - -Sign-in options - -Set up face -recognition* - -Enroll your -fingerprint - -For models with a Windows Hello-compatible infrared camera, you can -set up face recognition to sign in with your face. - -For models with a fingerprint sensor, you can enroll your fingerprint for -Windows sign-in. - -Sign-in options - -* for selected models - -Windows help information - -If the on-screen instructions cannot solve your problem, refer to the following to get the online Windows help -information. - -• Type Get Help or Tips in the Windows search box and then press Enter. When the app opens, type the - -problem description and select the matched result. - -• Visit Microsoft support Web site: https://support.microsoft.com. Enter what you are looking for in the - -search box and get search results. - -The Novo Button menu - -The Novo Button menu can be displayed before the operating system starts. From the menu, you can -choose to - -• Open the BIOS/UEFI setup utility - -• Open the boot device selection menu - -• Open the Windows startup options screen - -Chapter 2. Get started with your computer - -19 - - Note: From the Windows startup options screen, you can then choose to - -Start your computer using a recovery drive - -Reset your computer - -Open advanced options screen - -Open the Novo Button menu - -For Lenovo computers with a Novo button, you can press the button to open the Novo Button menu. - -Step 1. Turn off the computer. - -Step 2. Open the LCD screen and press the Novo button. - -Note: Alternatively, turn off the computer. Press Fn and the power button to open the Novo Button menu. - -Interact with your computer - -Hotkeys - -Look at the top row of the keyboard of your Lenovo computer and you’ll find keys with printed icons. These -keys are called hotkeys and they provide shortcuts to functions that you’ll use frequently. The function that -each hotkey provides is denoted by the icon printed on the key. - -Hotkeys share keys with function keys (F1–F12) and other keys. Depending on the FnLock setting, you use -hotkey functions by pressing the key directly or in combination with the Fn key. - -Note: The FnLock switch does not apply to hotkeys not found in the first row of the keyboard. To use these -hotkeys, always hold down the Fn key while pressing the key. - -Figure 1. Location of the Fn key - -Related topics - -Appendix A “Function descriptions for Lenovo hotkeys” on page 39 - -The FnLock switch -The FnLock is an electronic switch that affects how you use hotkey functions. To turn it on and off, press Fn -+ Esc. - -Note: The Esc key is in the upper left corner of the keyboard. It has an LED that indicates the status of the -FnLock switch. - -20 - -User Guide - - Figure 2. Locations of the FnLock key and the Fn key - -A Lenovo keyboard usually contains hotkeys in the top row. These hotkeys share keys with the function keys -(F1–F12) and other keys. For these dual-function keys, the icons or characters denoting the primary functions -are printed on top of the icons and characters denoting the secondary functions. - -• A: an icon or character denoting the primary function - -• B: an icon or character denoting the secondary function - -Figure 3. The layout of a dual-function key - -Table 7. FnLock and dual-function keys - -FnLock -(Esc) -LED - -Off - -On - -FnLock -status - -Pressing the hotkey alone - -Pressing the hotkey while holding down -the Fn key - -Disabled - -Primary function - -Enabled - -Secondary function - -Secondary function - -Primary function - -Touch operations supported by Windows - -For computers with a touch-enabled screen, you can touch the screen directly with your fingers and interact -with your computer in a more natural way. The following table lists frequently used touch operations that are -supported by the Windows operating system. - -Chapter 2. Get started with your computer - -21 - - Touch operation - -Used for - -Opening apps, documents, and other functions on the -taskbar or the Start menu - -Tap - -Tap twice quickly - -Tap, hold and lift - -Pinch - -Spread - -Opening apps that are pinned on the desktop - -Opening the context menu - -Zooming out - -Zooming in - -Opening the widgets panel - -Swipe in from the left edge - -22 - -User Guide - - Touch operation - -Used for - -Opening the notification center - -Swipe in from the right edge - -The Windows operating system also supports 3- and 4-finger gestures on the screen and the touchpad. You -can set them up in Settings ➙ Bluetooth & devices. - -Connect to an external display - -Connect to a wired display -Connect your computer to the desired display with an appropriate cable. - -Step 1. Connect one end of the display cable to the HDMI connector or a multi-purpose USB Type-C - -connector on your computer. - -Step 2. Connect the other end of the cable to the display. - -Connect to a wireless display -Ensure that: - -• Both your computer and the display support Miracast® technology. - -• The display is connected to the same Wi-Fi network as your computer and is discoverable. - -Step 1. Press Windows key + K. - -The computer searches for wireless display devices and audio devices and lists the results. - -Step 2. Select the display you want to connect to, and then follow the on-screen instructions. - -Change display settings -Step 1. Right-click on a blank area on the desktop, and then select Display settings. - -Your computer shows the Display window. - -Step 2. Select the display for which you want to change the settings. - -Chapter 2. Get started with your computer - -23 - - Step 3. Change the display settings as necessary. - -Set the display mode -Step 1. Press - -or Fn + - -. - -Your computer shows a list of display modes, with the current mode highlighted. - -Step 2. Select a display mode from the list. - -Protect your privacy using the camera shutter - -The camera shutter is a mechanical component that prevents any attempt from capturing your image, thus -protecting your privacy. To cover the camera lens, slide the camera shutter to the left. When you want to use -the camera, slide the camera shutter to the right. - -If you slide the camera shutter to the left during a video call , people on the video call will not be able to see -you. If you slide the camera shutter back to the right, they will be able to see you again. - -Enroll your fingerprints - -You can use the biometric fingerprint option to sign in to Windows quickly if your computer includes a -fingerprint sensor. This sign-in option provides a reliable and secure way of identity verification. - -Step 1. Select Start ➙ Settings ➙ Accounts ➙ Sign-in options. - -Step 2. Under Ways to sign in, select Fingerprint recognition (Windows Hello) to set up sign-in with a - -fingerprint sensor. - -Step 3. Follow the on-screen instructions to enroll your fingerprints. - -Notes: - -• You need to set up a PIN code before you are allowed to use this sign-in option. - -• It is recommended that you enroll multiple fingerprints in case of any injuries to your fingers. - -Lenovo Slim Pen (for selected models) - -The Lenovo Slim Pen is referred to as the pen in this section. - -Depending on the model, your computer might come with a pen. The pen enables you to write and draw in a -natural manner. - -Notes: - -• The pen is not waterproof. Take care not to drop the pen in water. - -24 - -User Guide - - • The pen contains pressure sensors. Do not push the tip of the pen for an extended period of time. Do not - -subject the pen to shock or vibration. - -Before first use -Charge the pen before first use. - -Step 1. Open the pen cap. - -Step 2. Connect the pen to a charger with a USB Type-C cable. - -Step 3. Disconnect the charger cable. Place the pen cap back. - -Note: You can check the pen’s battery level in the Lenovo Pen Settings. - -Use the pen - -Figure 4. Lenovo Slim Pen - -Chapter 2. Get started with your computer - -25 - - Item - -Component - -Description - -Pen cap - -Top button - -Open the pen cap to charge the pen. - -To right-click, press and hold the top button and tap the screen in supported -applications. - -Bottom button - -To erase text or drawings, press and hold the bottom button and rub at the -targeted items with the pen tip in supported applications. - -Pen tip - -Write or draw on the touch screen. - -1 - -2 - -3 - -4 - -Notes: - -• The default function of each button may vary in different applications. - -• You can view or change the functions in the Lenovo Pen Settings. - -Store the pen -When the pen is not in use, attach it to the lower right side of your computer. - -Lenovo Pen Settings -Lenovo Pen Settings is a pre-installed app for setting up Lenovo digital pens. This app allows you to perform -the following functions: - -• Reassign pen button functions - -• Adjust pen sensitivity - -• Check the pen’s battery level - -To open the Lenovo Pen Settings, type the app name in the Windows search box and select the matched -result. If the software is not pre-installed, you can download and install the latest drivers from https:// -support.lenovo.com. - -26 - -User Guide - - Chapter 3. Explore your computer - -Intelligent features - -Your computer may be pre-installed with one of Lenovo Vantage or Lenovo PC Manager but not all. Most -features described here can be enabled or disabled in one of these apps. Other features may be enabled in a -standalone app. - -Notes: - -• Software features are subject to change. Please refer to your actual product. - -• You may need to complete online update to the apps for the features to take effect. - -Gesture controls - -Lenovo unique touch gestures -In addition to touch gestures supported by the Windows operating system, some Lenovo computers may -support additional touch gestures. You can perform these gestures on the touchpad of your computer. - -Gesture - -Function - -Increases/Decreases display brightness when you watch -videos in full screen mode. -Note: This gesture is not supported by all video players -and may conflict with similar third-party gesture -functions. - -Slide up/down on the left side of the touchpad -Note: For computers with a touch-enabled screen, you -can also perform this gesture on the left side of the -screen. - -Increases/Decreases system volume when you watch -videos in full screen mode. -Note: This gesture is not supported by all video players -and may conflict with similar third-party gesture -functions. - -Slide up/down on the right side of the touchpad -Note: For computers with a touch-enabled screen, you -can also perform this gesture on the right side of the -screen. - -Smart Noise Cancelling - -Smart Noise Cancelling is a noise reduction feature available on some Lenovo product models. By filtering -out input and output noises, Smart Noise Cancelling enhances your audio experience. - -© Copyright Lenovo 2024 - -27 - - Function - -Description - -Remarks - -Voice Recognition: Your computer captures multiple voices in -a way that reflects their original spatial positions. - -Microphone -noise cancelling - -Only My Voice: This option requires you to record your voice so -that your computer captures this voice only and tries to -eliminate other voices. -Note: To remove your voice record, select REMOVE MY -VOICE. - -Normal: Your computer focuses on the voice of the person -facing it and reduces ambient sounds. - -• This function takes effect - -only when built-in -microphones/arrays or 3.5 -mm jack wired microphones -are used as the input media. - -• To disable this function, - -select Off. - -Multiple Voices: Your computer captures multiple voices from -an expanded range in front of the computer. - -Speaker noise -cancelling - -Your computer filters out other sounds to play only human -voices. - -Meeting noise -cancelling - -When this function is selected, your computer uses special -algorithms for noise reduction when you are using conferencing -applications. - -These functions are not -applicable to scenarios like -listening to music and watching -videos. - -Notes: - -• Depending on its hardware, your computer may not support all the functions and options described - -above. - -• You can view and customize this feature under Device Settings in Lenovo Vantage or Lenovo PC - -Manager. - -Super Resolution - -By utilizing the capabilities and potentials of Intel processors, Super Resolution helps you play videos with a -higher resolution than the original. It works especially well in cases where the source video has a poor -resolution. - -For most players, Super Resolution can be enabled or disabled in Lenovo Vantage or Lenovo PC Manager, -but for some specific players, you might need to enable this feature manually. - -Eye Care Mode - -Eye Care Mode intelligently adjusts the color temperature of the screen and can reduce the chances of -developing eye fatigue or eye strain. - -28 - -User Guide - - Manage power - -Use the information in this section to achieve the best balance between performance and power efficiency. - -Rechargeable battery pack - -Your computer includes a built-in, rechargeable battery pack that makes mobile computing a reality. When -the computer is plugged into an electrical outlet, the battery charges. If you use the computer when you -don't have access to an electrical outlet, the battery discharges to supply electricity that the computer -system requires for operation. - -You can charge the battery any time you want. The battery packs of Lenovo computers support multiple -charging modes that are suitable for different power usage habits. You can switch the battery’s active -charging mode in Lenovo Vantage or Lenovo PC Manager. - -Battery charging is also affected by its temperature. The recommended temperature range for charging the -battery is between 10°C (50°F) and 35°C (95°F). - -Note: - -You can check the battery temperature in Lenovo Vantage. - -To maximize the life of the battery, once the battery is fully charged, it must discharge to 94% or lower -before it will be allowed to recharge again. - -Normal mode -Normal mode is the most basic charging mode. In normal mode, it typically takes 2 to 4 hours for the battery -to charge from 0% to 100%. - -Rapid charge mode -If you want the battery to be charged faster than in normal mode, switch the battery charging to rapid charge -mode. The following table lists the typical time needed for batteries in rapid charge mode to be charged to -80% and 100% respectively. - -Table 8. Reference charge time for batteries in rapid charge mode - -Mode - -Time needed to charge from 0% to 80% - -Time needed to charge from 0% to 100% - -Rapid charge - -Less than 1 hour - -Less than 2 hours - -Conservation mode -If your computer is constantly plugged into an electrical outlet, consider switching the battery charging to -conservation mode. In conservation mode, the battery will not be fully charged. Instead, the battery’s charge -will be kept within 75%–80%. This is beneficial to the long-term health of the battery. - -Note: If you want the battery to be fully charged before bringing the computer to work, disable conservation -mode by switching the battery charging to normal or rapid charge mode. - -Nighttime charge optimization -Some people follow a regular pattern when using their computers. They finish their workday with the -computer at a low battery charge level. They plug in their computers at night and need the battery to be fully -charged the next morning so they can unplug the computer and bring it to work. These activities happen at -approximately the same time each day. If this sounds like you, consider enabling nighttime charge -optimization for the battery. - -Chapter 3. Explore your computer - -29 - - Nighttime charge optimization affects battery charging during the night hours, the time when you’re usually -asleep. When it is enabled, the computer regularly adapts its charging behavior based on observation of -when you plug in the computer at night and unplug it in the morning. During the nighttime, the battery is -charged to a particular range and is kept within that range for an extended period, before being further -charged to 100%. Nighttime charge optimization ensures safe charging during the night and is beneficial to -the long-term health of the battery. - -Note: With nighttime charge optimization enabled, if you break your routine one day by unplugging the -computer much earlier than usual in the morning, you may find that the battery is not fully charged. - -If the battery pack of your computer supports nighttime charge optimization, it can be enabled in Lenovo -Vantage or Lenovo PC Manager. - -Recover full battery capacity -If your computer is constantly plugged in to an electrical outlet and the battery rarely discharges, the battery -may not be charged to its full capacity even if the battery meter reports 100% charge. You can recover the -battery’s full charging potential simply by discharging and re-charging the battery. - -Step 1. Unplug the computer and use it until the battery charge drops below 20%. - -Step 2. Plug in the computer and charge the battery to 100%. - -Set power button behavior - -By default, pressing the power button puts the computer to sleep mode. However, you can change the -power button behavior in Windows Control Panel. - -Step 1. Type Control Panel in the Windows search box and then press Enter. Open the control panel and - -view by large or small icons. - -Step 2. Select the power options and then click choose what the power button does. - -A power plan - -A power plan is a collection of power-saving settings made available by an operating system. With a power -plan, you can set idle time-outs for different hardware components to enter a low-power state. The default -power plan and some of its idle time-out settings are listed below for computer models pre-installed with -Windows. - -The listed settings are active when the computer is connected to an electrical outlet. If your computer -includes a built-in battery pack, a different collection of time-out settings is set to take effect when the -computer is operating on battery power. - -• Default power plan: Balanced - -• Turn off the display: after 5 minutes - -• Put the computer to sleep: after 5 minutes - -Note: To wake the computer from the sleep state, press the power button or any key on the keyboard. - -Change or customize a power plan -This operation is applicable to computers with pre-installed Windows. - -Step 1. Type power plan in the Windows search box and then press Enter. - -Step 2. Customize a power plan of your preference. - -30 - -User Guide - - System operation modes - -Lenovo has preset several modes in which your computer can operate. The maximum attainable -performance, power consumption, and speed limit for the heat sink fan vary between the operation modes. -Consider the following conditions when you want to switch operation modes. - -• The environment where you use your computer, and - -• The tasks running on your computer - -You can switch the operation mode in the pre-installed app Lenovo Vantage or Lenovo PC Manager. As a -shortcut, you can also use the key combination Fn + Q. Three modes are usually available for most Lenovo -computers. The following table lists the operation modes and the recommended conditions for each mode. - -Note: The operation modes listed in the table are descriptive and may not be the same as those displayed -by the app. - -Table 9. Operation modes and their recommended usage conditions - -Operation mode - -Recommended conditions - -High Performance - -• Your computer is plugged into an electrical outlet. - -• You want the best performance, and - -• You don't care if the fan makes a little noise. - -Auto (Balance) - -You plan to frequently switch between different computer tasks over a period of time. - -Power Saving (Quiet) - -• Your computer is operating on battery power, or - -• You want the computer to be as quiet as possible. - -Note: In Auto (Balanced) mode, the computer dynamically switches between High Performance mode and -Power Saving (Quiet) mode depending on the tasks running on the computer. - -Chapter 3. Explore your computer - -31 - - 32 - -User Guide - - Chapter 4. Help and support - -Frequently asked questions - -How do I partition my storage drive - -Refer to https://support.lenovo.com/solutions/ht503851. - -What should I do if my computer stops responding - -Press and hold the power button until the computer turns off. Then restart the computer. - -What should I do if I spill liquid on the computer - -1. Carefully unplug the ac power adapter and turn off the computer immediately. The more quickly you - -stop the current from passing through the computer the more likely you will reduce damage from short -circuits. - -Attention: Although you might lose some data or work by turning off the computer immediately, leaving -the computer on might make your computer unusable. - -2. Wait until you are certain that all the liquid is dry before turning on your computer. - -CAUTION: -Do not try to drain out the liquid by turning over the computer. If your computer has keyboard -drainage holes on the bottom, the liquid will be drained out through the holes. - -Where can I get the latest device drivers and UEFI/BIOS - -• Lenovo Vantage or Lenovo PC Manager - -• Lenovo Support Web site at https://support.lenovo.com - -Why does my computer start automatically when I open the lid - -Your computer may have Flip to Start enabled. Many Lenovo notebook computers include a sensor that can -detect the angle at which the lid is opened. When you open the lid, the sensor can detect this behavior. If Flip -to Start is enabled, the computer will respond by starting up automatically. - -If you don’t like this feature, you can disable it. Flip to Start can be enabled and disabled in: - -• Lenovo Vantage or Lenovo PC Manager - -• Firmware Setup Utility - -Self-help resources - -Use the following self-help resources to learn more about the computer and troubleshoot problems. - -© Copyright Lenovo 2024 - -33 - - Resources - -How to access? - -Troubleshooting and frequently asked questions - -• https://www.lenovo.com/tips - -• https://forums.lenovo.com - -Accessibility information - -https://www.lenovo.com/accessibility - -Reset or restore Windows - -• Use Lenovo recovery options. - -1. Go to - -https://support.lenovo.com/ -HowToCreateLenovoRecovery. - -2. Follow the on-screen instructions. - -• Use Windows recovery options. - -1. Go to https://pcsupport.lenovo.com. - -2. Detect your computer or manually select your - -computer model. - -3. Click Troubleshoot & Diagnose ➙ Custom -Troubleshooting ➙ Operating System -Diagnostics and then follow the on-screen -instructions. - -Use Lenovo Vantage or Lenovo PC Manager to: - -• Download and install the latest drivers and firmware. - -• Configure hardware settings. - -Use Windows Search. - -• Diagnose computer hardware problems. - -• Check the computer warranty status. - -Product documentation: - -• Generic Safety and Compliance Notices - -• Safety and Warranty Guide - -• Setup Guide - -• This User Guide - -• Regulatory Notice - -Lenovo Support Web site with the latest support -information of the following: - -• Drivers and software - -• Diagnostic solutions - -• Product and service warranty - -• Product and parts details - -• Knowledge base and frequently asked questions - -Windows help information - -1. Go to https://support.lenovo.com. - -2. Detect your computer or select computer model - -manually. - -3. Select Guides & Manuals and filter out the - -documentation you want. - -Visit https://support.lenovo.com - -• Use Get Help or Tips. - -• Use Windows Search or the Cortana® personal - -assistant. - -• Microsoft support Web site: -https://support.microsoft.com - -34 - -User Guide - - What is a CRU? - -Customer replaceable units (CRUs) are parts that can be replaced by the customer. A Lenovo computer may -contain the following types of CRUs: - -Self-service CRU - -Optional-service CRU - -Parts that can be replaced easily by customer -themselves or by trained service technicians at an -additional cost. - -Parts that can be replaced by customers with a -greater skill level. Trained service technicians can -also provide service to install or replace the parts -under the type of warranty designated for the -customer’s machine. - -If you intend to install a CRU, Lenovo will ship the CRU to you. You might be required to return the defective -part that is replaced by the CRU. When return is required: (1) return instructions, a prepaid shipping label, -and a container will be included with the replacement CRU; and (2) you might be charged for the -replacement CRU if Lenovo does not receive the defective CRU within thirty (30) days of your receipt of the -replacement CRU. For full details, see the Lenovo Limited Warranty at -https://www.lenovo.com/warranty/llw_02. - -CRUs for your product model - -The table below lists the CRUs and CRU types that are defined for your product model. - -Part - -Power cord* - -ac power adapter - -Lenovo Slim Pen* - -* for selected models - -Notes: - -Self-service CRU - -Optional-service CRU - -X - -X - -X - -• CRU replacement instruction is provided in one or more of the following publications and is available from - -Lenovo at any time upon your request. - -the product User Guide - -the printed publications that came with the product - -• Replacement of any parts not listed above, including the built-in rechargeable battery, must be done by a -Lenovo-authorized repair facility or technician. Go to https://support.lenovo.com/partnerlocation for more -information. - -Chapter 4. Help and support - -35 - - Call Lenovo - -If you have tried to correct the problem yourself and still need help, you can call Lenovo Customer Support -Center. - -Before you contact Lenovo - -Record product information and problem details before you contact Lenovo. - -Product information - -• Product name - -Problem symptoms and details - -• What is the problem? Is it continuous or intermittent? - -• Machine type and serial number - -• Any error message or error code? - -• What operating system are you using? Which version? - -• Which software applications were running at the time - -of the problem? - -• Can the problem be reproduced? If so, how? - -Note: The product name and serial number can usually be found on the bottom of the computer, either -printed on a label or etched on the cover. - -Lenovo Customer Support Center - -During the warranty period, you can call Lenovo Customer Support Center for help. - -Telephone numbers - -For a list of the Lenovo Support phone numbers for your country or region, go to -https://pcsupport.lenovo.com/supportphonelist. - -Note: Phone numbers are subject to change without notice. If the number for your country or region is not -provided, contact your Lenovo reseller or Lenovo marketing representative. - -Services available during the warranty period - -• Problem determination - Trained personnel are available to assist you with determining if you have a - -hardware problem and deciding what action is necessary to fix the problem. - -• Lenovo hardware repair - If the problem is determined to be caused by Lenovo hardware under warranty, - -trained service personnel are available to provide the applicable level of service. - -• Engineering change management - Occasionally, there might be changes that are required after a product -has been sold. Lenovo or your reseller, if authorized by Lenovo, will make selected Engineering Changes -(ECs) that apply to your hardware available. - -Services not covered - -• Replacement or use of parts not manufactured for or by Lenovo or nonwarranted parts -• Identification of software problem sources -• Configuration of UEFI/BIOS as part of an installation or upgrade -• Changes, modifications, or upgrades to device drivers -• Installation and maintenance of network operating systems (NOS) -• Installation and maintenance of programs - -For the terms and conditions of the Lenovo Limited Warranty that apply to your Lenovo hardware product, -see “Warranty information” in the Safety and Warranty Guide that comes with your computer. - -36 - -User Guide - - Purchase additional services - -During and after the warranty period, you can purchase additional services from Lenovo at -https://pcsupport.lenovo.com/warrantyupgrade. - -Service availability and service name might vary by country or region. - -Chapter 4. Help and support - -37 - - 38 - -User Guide - - Appendix A. Function descriptions for Lenovo hotkeys - -Hotkey functions are denoted by the icons printed on the keys. The following table describes the functions -for all Lenovo hotkeys. The keyboard of your Lenovo computer should only support a subset of the listed -hotkeys. - -Table 10. Hotkey icons and denoted functions - -Hotkey icon - -Function description - -Mutes/Unmutes sound. - -Increases volume. - -Decreases volume. - -Enables/Disables the microphone. - -Increases screen brightness. - -Decreases screen brightness. - -Selects and sets up display devices. - -Enables/Disables airplane mode. (Windows operating systems) - -Displays the Lenovo Smart Key quick launch panel. - -Enables/Disables the touchpad. - -Displays all open apps as thumbnails. (Windows operating systems) - -Opens the Calculator app. (Windows operating systems) - -Opens the Snipping tool. (Windows operating systems) - -Opens the Settings app. (Windows operating systems) - -Locks the screen. (Windows operating systems) - -Opens the clipboard history. (Windows operating systems) - -Puts the computer to sleep. - -Opens the context menu of the current active app. - -Opens an associated voice app, voice service, or Microsoft Cortana. - -Adjusts the keyboard backlight. - -Increases the brightness of the keyboard backlight. - -Decreases the brightness of the keyboard backlight. - -Answers an incoming Microsoft Teams call. - -Rejects an incoming Microsoft Teams call. - -Accesses Lenovo Support resources. - -Starts/Pauses audio or video playback. - -Stops audio or video playback. - -Plays the previous audio or video in the playlist. - -© Copyright Lenovo 2024 - -39 - - Table 10. Hotkey icons and denoted functions (continued) - -Hotkey icon - -Function description - -Plays the next audio or video in the playlist. - -Switches the computer’s active operation mode. - -Blurs the background for videoconferencing. - -Switches the computer’s active audio mode. - -Turns on/off night light mode. (Windows operating systems) - -Note: Hotkey functions may change if the associated app or the Windows operating system is updated. - -Related topics - -“Hotkeys” on page 20 - -40 - -User Guide - - Appendix B. Notices and trademarks - -Notices - -Lenovo may not offer the products, services, or features discussed in this document in all countries. Consult -your local Lenovo representative for information on the products and services currently available in your -area. Any reference to a Lenovo product, program, or service is not intended to state or imply that only that -Lenovo product, program, or service may be used. Any functionally equivalent product, program, or service -that does not infringe any Lenovo intellectual property right may be used instead. However, it is the user's -responsibility to evaluate and verify the operation of any other product, program, or service. - -Lenovo may have patents or pending patent programs covering subject matter described in this document. -The furnishing of this document does not give you any license to these patents. You can send license -inquiries, in writing, to: -Lenovo (United States), Inc. -8001 Development Drive -Morrisville, NC 27560 -U.S.A. -Attention: Lenovo Director of Licensing - -LENOVO PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS -OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, -MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some jurisdictions do not allow -disclaimer of express or implied warranties in certain transactions, therefore, this statement may not apply to -you. - -Changes are made periodically to the information herein; these changes will be incorporated in new editions -of the publication. To provide better service, Lenovo reserves the right to improve and/or modify the -products and software programs described in the manuals included with your computer, and the content of -the manual, at any time without additional notice. - -The software interface and function and hardware configuration described in the manuals included with your -computer might not match exactly the actual configuration of the computer that you purchase. For the -configuration of the product, refer to the related contract (if any) or product packing list, or consult the -distributor for the product sales. Lenovo may use or distribute any of the information you supply in any way it -believes appropriate without incurring any obligation to you. - -The products described in this document are not intended for use in implantation or other life support -applications where malfunction may result in injury or death to persons. The information contained in this -document does not affect or change Lenovo product specifications or warranties. Nothing in this document -shall operate as an express or implied license or indemnity under the intellectual property rights of Lenovo or -third parties. All information contained in this document was obtained in specific environments and is -presented as an illustration. The result obtained in other operating environments may vary. - -Lenovo may use or distribute any of the information you supply in any way it believes appropriate without -incurring any obligation to you. - -Any references in this publication to non-Lenovo Web sites are provided for convenience only and do not in -any manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the -materials for this Lenovo product, and use of those Web sites is at your own risk. - -© Copyright Lenovo 2024 - -41 - - Any performance data contained herein was determined in a controlled environment. Therefore, the result -obtained in other operating environments may vary significantly. Some measurements may have been made -on development-level systems and there is no guarantee that these measurements will be the same on -generally available systems. Furthermore, some measurements may have been estimated through -extrapolation. Actual results may vary. Users of this document should verify the applicable data for their -specific environment. - -This document is copyrighted by Lenovo and is not covered by any open source license, including any Linux -agreement(s) which may accompany software included with this product. Lenovo may update this document -at any time without notice. - -For the latest information or any questions or comments, contact or visit the Lenovo Web site: -https://support.lenovo.com. - -Trademarks - -Lenovo and the Lenovo logo are trademarks of Lenovo. Thunderbolt is a trademark of Intel Corporation or its -subsidiaries. Microsoft, Windows, and Cortana are trademarks of the Microsoft group of companies. SD, -microSD, microSDHC, and microSDXC marks and logos are trademarks of SD-3C LLC. Kensington Nano -Security Slot is a registered trademark of ACCO Brands. DisplayPort is a trademark of the Video Electronics -Standards Association. The terms HDMI and HDMI High-Definition Multimedia Interface, and the HDMI logo -are trademarks or registered trademarks of HDMI Licensing Administrator, Inc. in the United States and other -countries. Wi-Fi is a registered trademark of Wi-Fi Alliance. USB Type-C is a registered trademark of USB -Implementers Forum. All other trademarks are the property of their respective owners. - -42 - -User Guide +Another assumption in this analysis is that personal incomes change at the same rate as GDP per +capita. Empirically, it is not unusual to observe that personal income or consumption increases at +a slower rate than GDP per capita. However, the relationship between the two measures varies +widely across countries and it is not clear what the relative rates of growth might be at the regional +or global level. The World Bank often assumes a “pass-through” rate of 0.85, i.e., only 85% of + +www.pewresearch.org + + 15 + +PEW RESEARCH CENTER + +national output growth is passed through to household income or consumption. If a pass-through +rate of 0.85 is assumed for this analysis the estimated change in the number of people in each +income tier is as follows: Poor (+117 million), low income (+11 million), middle income (-45 +million), upper-middle income (-30 million) and high income (-53 million). + +From the point of view of measuring global poverty, a key issue is that the latest available +benchmark survey for India was conducted in 2011. As the second most populous country in the +world, India accounts for a large share of the global population in poverty. In 2011, 969 million +people lived in extreme poverty globally, on $1.90 or less daily, and 285 million of that number +were from India. The World Bank has applied a variety of estimation techniques to update the +poverty counts for India, but uncertainty lingers in these updates for India and, by implication, in +current measures of global poverty. + +www.pewresearch.org diff --git a/q181/random_k4/question.json b/q181/random_k4/question.json index 270a09d35fe0563e58f9b6cb0a39cb35c14943a1..6569192c9b87991f26b6b76730095929b9731d2c 100644 --- a/q181/random_k4/question.json +++ b/q181/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "b3m5kaeqm2w8n4bwcesw-140602121350-phpapp02_95.pdf", - "web_e9f399fea7556d59.pdf", - "web_b0b861dfece3f50d.pdf", - "arxiv_2405_20834.pdf", - "da82ba1659f104383b88b82d6ffaa0cd0c0d177b.pdf" + "web_17b633ba09ee110f.pdf", + "web_18e98cd838edeb2e.pdf", + "resize_r2_16.pdf", + "web_55eb3e1bfd6dcec7.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q181/random_k4/random_1.md b/q181/random_k4/random_1.md index d3d5a7efa18017d2cc9d64bd996d8e5da4aa2173..261b89724e814d1be04e011ae16d4387de2cce1a 100644 --- a/q181/random_k4/random_1.md +++ b/q181/random_k4/random_1.md @@ -1,3125 +1,767 @@ -User Guide - -Lenovo ThinkBook 14 2-in-1 Gen 4 - - Read this first - -Before using this documentation and the product it supports, ensure that you read and understand the -following: - -• Generic Safety and Compliance Notices - -• Safety and Warranty Guide - -• Setup Guide - -First Edition (February 2024) - -© Copyright Lenovo 2024. - -LIMITED AND RESTRICTED RIGHTS NOTICE: If data or software is delivered pursuant to a General Services -Administration “GSA” contract, use, reproduction, or disclosure is subject to restrictions set forth in Contract No. GS- -35F-05925. - - Contents - -About this guide . . . . . . . . . . . . . iii - -. - -. - -. - -. - -. . - -Left - -Base . - -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. -Microphones . . -Infrared LED . . -Camera shutter . -. . -Camera . -. -. . -Camera light -. . -. -Screen . -. . -. -Antennas -. . -. -. -. . -. . -Keyboard . -. . -Touchpad . -. . -. -. . - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -Multi-purpose USB Type-C connector . -. -. . -. -Charging light . . -. -USB Standard-A connector -. -. . -Always-on connector . -. -. . -. -HDMI connector . -. -. -. . -. -Combo audio jack . -. -. -. . -. -. -Speakers -. . -. -. . -. -. -. -. -. . -. -. . -. -. -Speakers -. -. . -. -. . -. -. -. . -Power light . -. -. . -. -Power button . . -. -. -. . -. -Fingerprint sensor . -. -microSD card slot . -. . -. -. -USB Standard-A connector -. -. . -Security lock slot -. -. . -. . - -Chapter 1. Meet your computer . . . . . 1 -. 1 -Front . -. 2 -. 2 -. 2 -. 2 -. 2 -. 2 -. 2 -. 3 -. 3 -. 4 -. 5 -. 5 -. 5 -. 6 -. 6 -. 6 -. 6 -. 6 -. 7 -. 7 -. 7 -. 8 -. 8 -. 8 -. 8 -. 8 -. 9 - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. -Avoid constant body contact with specific hot -. 9 -. -. -. -. . -sections . -. 9 -. -. -Air vents (intake) . -. -. -. -Novo button hole . -. 9 -. 11 -. -. -. . -. -. 11 -. -. -Air vents (outlet) . -. 12 -. 14 -. 14 - -. . -. . -. . -. . -. . -. . -Statement on USB transfer rate . . -. . -Operating environment . . - -. . -. . -. . -. . -. . -Features and specifications . . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -. . -. . -. . -. . -. . -. . -. . -. . - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. - -Bottom . . - -. -. -. -. -. -. - -Right . - -Rear - -. . - -. . - -. -. - -. - -. - -. - -. - -. - -. - -Chapter 2. Get started with your -computer . . . . . . . . . . . . . . . . 15 -. 15 -Connect to a network . -. 15 -. 15 -. 16 - -. -Connect to a Wi-Fi network . -Establish a wired connection . -. - -Unique Lenovo apps . - -. . -. . -. . -. . - -. . -. . -. . -. . - -. -. -. -. - -. -. -. -. - -. . - -. . - -. - -. - -. +FOR RELEASE MARCH 18, 2021 -Work with Windows . +BY Rakesh Kochhar -The Novo Button menu . . +FOR MEDIA OR OTHER INQUIRIES: -Interact with your computer . -. +Rakesh Kochhar, Senior Researcher -. +Stefan Cornibert, Communications Manager -. . +Tanya Arditi, Communications Manager -. . +202.419.4372 -. -. -. -. -. -. -. -. +www.pewresearch.org -. -. -. -. -. -. -. -. +RECOMMENDED CITATION -. . -. . -. . -. . -. . -. . -. . -. . - -. -Lenovo Vantage . -Lenovo Smart Meeting . -. - -. . -. . -. . -Windows help information . . -. . -Open the Novo Button menu . -. . -. . - -. -. -. -. -. -. -. -Hotkeys . . -. -. . -Touch operations supported by Windows . -Connect to an external display . -. -Protect your privacy using the camera -. . -shutter -. -. . -Enroll your fingerprints . -Lenovo Slim Pen (for selected models) . - -. . -. . - -. -. -. - -. . - -. . - -. . - -. -. - -. -. - -. - -. - -. - -. - -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. - -. 16 -. 16 -. 18 -. 19 -. 19 -. 20 -. 20 -. 20 -. 21 -. 23 - -. 24 -. 24 -. 24 - -. . -. . - -Chapter 3. Explore your computer . . 27 -. 27 -. -Intelligent features . -. -. 27 -Gesture controls. -. -. 27 -Smart Noise Cancelling . -. 28 -. -Super Resolution . . -. 28 -. -. . -Eye Care Mode . -. 29 -. -. . -. -. 29 -. 30 -. 30 -. 31 - -. . -. . -. . -. . -. . -. . -. -Rechargeable battery pack . -Set power button behavior. . -. . -A power plan . -System operation modes . . - -. . -. . -. . -. . -. . -. . -. . -. . -. . -. . - -Manage power . - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. -. -. -. -. -. -. -. -. -. - -. . - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. - -. -. - -. . - -. . - -. . - -. . - -. . - -. . - -. 33 - -Chapter 4. Help and support -. -Frequently asked questions . - -. . . . . 33 -. 33 -. -. . -How do I partition my storage drive . . -. 33 -. -What should I do if my computer stops -responding . -. . -What should I do if I spill liquid on the -. -computer -Where can I get the latest device drivers and -UEFI/BIOS . -. -. -. . -Why does my computer start automatically -when I open the lid . . -. -. -. . -. -. . -. -. -. -. -. +Pew Research Center, March, 2021, “The Pandemic Stalls +Growth in the Global Middle Class, Pushes Poverty Up Sharply” -. -. . -. -. . -. -. . -. -. . -. -. . -. -Before you contact Lenovo . -. -Lenovo Customer Support Center . -. + 1 -. -. -Self-help resources . -What is a CRU? . -. -. -CRUs for your product model -. -Call Lenovo . . +PEW RESEARCH CENTER -. 33 -. 33 -. 35 -. 35 -. 36 -. 36 -. 36 -. 37 +About Pew Research Center -. . -. . -. . -. . -. . -. . -. . -. . +Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes +and trends shaping America and the world. It does not take policy positions. The Center conducts +public opinion polling, demographic research, content analysis and other data-driven social +science research. It studies U.S. politics and policy; journalism and media; internet, science and +technology; religion and public life; Hispanic trends; global attitudes and trends; and U.S. social +and demographic trends. All of the Center’s reports are available at www.pewresearch.org. Pew +Research Center is a subsidiary of The Pew Charitable Trusts, its primary funder. -Purchase additional services. +© Pew Research Center 2021 -. -. -. -. -. -. -. -. +www.pewresearch.org -. -. -. -. -. -. + 2 -. 33 +PEW RESEARCH CENTER -. 33 +How we did this -. . +The COVID-19 pandemic has created an economic crisis, shuttering businesses and cutting jobs +around the world. This report looks at how the downturn has affected the standard of living +globally and in seven major regions. -. . +The focus is on the distribution of people across five income tiers in 2020: poor, low income, +middle income, upper-middle income and high income. See the methodology and an earlier Pew +Research Center report for more on the definition and meaning of these tiers in a global context. -. . +The key data source for the analysis is the World Bank’s PovcalNet database, which provides +access to household survey data on either income or consumption for more than 160 countries. At +the regional level, the latest year for which comprehensive data are available ranges from 2014 for +South Asia to 2018 for East Asia and the Pacific and three other regions. The global regions are as +defined by the World Bank. -. . +These benchmark estimates of income are extrapolated to 2020 using World Bank estimates of +output growth through 2020. One projection is based on the World Bank’s January 2020 forecasts +of economic growth in 2020, and the other is based on its January 2021 estimates of growth in +2020. The difference between these two measures is used to represent the effect of the pandemic +on the income distribution in each region. Regional estimates are summed to yield estimates of +the global distribution. -. +www.pewresearch.org -. + 3 -. +PEW RESEARCH CENTER -. +BY Rakesh Kochhar -. +The COVID-19 pandemic is having a deep effect on the global economy. In January 2020, as +reports of the novel coronavirus were emerging, the World Bank forecasted that the global +economy would expand by 2.5% that year. In January 2021, with the pandemic still holding much +of the world in its grip, the World Bank estimated that the global economy contracted by 4.3% in +2020, a turnabout of 6.8 percentage points. -. +The economic downturn is likely to have +diminished living standards around the world, +pushing millions out of the global middle class +and swelling the ranks of the poor. At the same +time, the path to a recovery is clouded with +uncertainties. -. +A new Pew Research Center analysis finds that +the global middle class encompassed 54 million +fewer people in 2020 than the number +projected prior to the onset of the pandemic. +Meanwhile, the number of poor is estimated to +have been 131 million higher because of the +recession. -. +The drop-off in the global middle class was +centered in South Asia and in East Asia and the +Pacific, and it stalled the expansion seen in the +years preceding the pandemic. South Asia, +specifically India, along with Sub-Saharan +Africa, accounted for most of the increase in +poverty, reversing years of progress on this +front. -Appendix A. Function descriptions -for Lenovo hotkeys . . . . . . . . . . 39 +The COVID-19 downturn curbed growth +in the global middle class, increased +poverty sharply in 2020 -© Copyright Lenovo 2024 +Estimated change in the number of people in each +income tier due to the global recession in 2020, in +millions -i +High income - Appendix B. Notices and -trademarks . . . . . . . . . . . . . . . 41 +-62 -ii +million -User Guide +Upper-middle +income - About this guide +-36 -• This guide applies to Lenovo product model(s) listed below. Illustrations in this guide may look slightly +Middle income -different from your product model. +-54 + +Low income + ++21 -Model name +Poor -Machine type (MT) ++131 -ThinkBook 14 2-in-1 G4 IML +Notes: The poor live on $2 or less daily, low income on $2.01-$10, +middle income on $10.01-$20, upper-middle income on $20.01- +$50, and high income on more than $50; figures expressed in 2011 +purchasing power parities in 2011 prices. The estimates show the +difference in the number of people in an income tier based on pre- +pandemic projections and post-pandemic estimates. The term +“post-pandemic” refers to the period since the onset of the +pandemic in January 2020.It is assumed there is no change in the +income distribution in a region from the benchmark year for the +projection to 2020. +Source: Pew Research Center analysis of data from the World Bank +PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes +Poverty Up Sharply” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 4 + +PEW RESEARCH CENTER + +As defined in this report (and in previous Pew Research Center analyses), people who are middle +income live on $10.01-$20 a day, which translates to an annual income of about $14,600 to +$29,200 for a family of four. This is modest by the standards of advanced economies. In fact, it +straddles the official poverty line in the United States − about $23,000 for a family of four in 2020 +(expressed in 2011 prices). By global standards, the poor live on $2 or less a day, or no more than +$2,920 annually for a family of four. + +The number of people in the global high-income tier (more than $50 daily) is estimated to have +decreased by 62 million in 2020, erasing about half of the gain since 2011, with most of the change + +Who is middle class (or middle income)? +This report uses the terms “middle income” and “middle class” interchangeably, a common practice among +economists who tend to define the middle class in terms of income or consumption. But being middle class +can connote more than income, be it a college education, white-collar work, economic security, owning a +home, or having certain social and political values. Class could also simply be a matter of self-identification. +The interplay among these many factors is examined in studies by Hout (2007) and Savage et al. (2013), +among others. + +The population in each global region is divided into five groups: poor, low income, middle income, upper- +middle income and high income. The poor live on $2 or less daily, low income on $2.01-$10, middle income +on $10.01-$20, upper-middle income on $20.01-$50 and high income on more than $50. All dollar figures +are expressed in 2011 prices and purchasing power parity dollars, currency exchange rates adjusted for +differences in the prices of goods and services across countries. -ThinkBook 14 2-in-1 G4 IML 1 +The assignment to a group, or income tier, is based on a household’s daily per capita income or +consumption, a simple way of controlling for differences in household size. In the source data, some +countries only report income data while others only report consumption data. The terms “income” and +“consumption” are used interchangeably for the sake of convenience. -ThinkBook 14 2-in-1 G4 IML 2 +A middle-income, or middle-class, threshold of $10 follows the practice of other researchers. A similar +threshold has been used by the World Bank (2007, 2015), researchers at the Organization for Economic +Cooperation and Development (Kharas, 2010), the development community (Birdsall, Lustig and Meyer, +2013) and others. There is evidence that the $10 threshold, which is about five times higher than the World +Bank’s global poverty line, is associated with economic security and “insulates” people from falling back +into poverty. An earlier Pew Research Center report discusses the choice of income thresholds in greater +detail. -21MX +The global middle-class standard is modest by the standards of advanced economies. In 2020, the official +poverty line for a family of four in the U.S. stood at about $15.90 per person per day (in 2011 prices). Thus, +many who are poor by the U.S. definition would meet the global middle-income standard or, conversely, +many who are in the global middle class would be poor by U.S. standards. -21MX +www.pewresearch.org -21MX + 5 -• For further compliance information, refer to the Generic Safety and Compliance Notices at +PEW RESEARCH CENTER -https://pcsupport.lenovo.com/docs/generic_notices. +emanating from advanced economies. Meanwhile, the upper-middle income population ($20.01- +$50 daily) fell by 36 million, while the low-income population ($2.01-$10 daily) is estimated to +have increased by 21 million. -• This guide may contain information about accessories, features, and software that are not available on all +These estimates are based on the Center’s analysis of World Bank data on the distribution of +people by either income or consumption levels in seven major global regions. The latest available +estimates of income for each region, such as for 2018 for Europe and Central Asia and 2015 for the +Middle East and North Africa, are extrapolated +to 2020 using two sets of growth estimates: the +forecasts issued by the World Bank in January +2020, prior to the pandemic, and the estimates +issued in January 2021, with the pandemic’s +economic impact in plainer sight. The +difference between these two measures is taken +to represent the impact of the COVID-19 +downturn on the standard of living of people +globally. -models. +Globally, most are either poor or low +income, and the pandemic likely drove +the numbers higher in 2020 -• This guide contains instructions that are based on the Windows operating system. These instructions are +Estimated number of people in each income tier in 2020, +before and after the global recession, in millions + +Pre-pandemic projection + +Post-pandemic estimate + +High income + +531 million + +593 + +Prior to the emergence of COVID-19, some 1.38 +billion people were expected to be counted in +the global middle class in 2020. But the +pandemic is estimated to have driven this +number down to 1.32 billion. The share of +people in the middle class globally is estimated +to have been 17.1% in 2020, instead of +potentially 17.8%. The erosion in the middle +class might have been deeper if not for the fact +that China – which is home to more than one- +third of the global middle class – evaded an +economic contraction, even though growth +there was slower than anticipated. + +The number of global poor is estimated to have +risen to 803 million in 2020, much greater than +the 672 million initially expected. The global +poverty rate, which had been in steady decline +this century, is likely to have increased to + +Upper-middle +Upper-middle +income +income + +Middle +income + +Low income + +1,140 + +1,176 + +1,324 + +1,378 + +Poor + +803 + +672 + +3,956 -not applicable if you install and use other operating systems. +3,936 + +Notes: The poor live on $2 or less daily, low income on $2.01-$10, +middle income on $10.01-$20, upper-middle income on $20.01- +$50, and high income on more than $50; figures expressed in 2011 +purchasing power parities in 2011 prices. It is assumed there is no +change in the income distribution in a region from the benchmark +year for the projection to 2020. The term “post-pandemic” refers to +the period since the onset of the pandemic in January 2020. +Source: Pew Research Center analysis of data from the World Bank +PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes +Poverty Up Sharply” -• Microsoft® makes periodic feature changes to the Windows® operating system through Windows Update. -As a result, the operating system related instructions may become outdated. Refer to Microsoft resources -for the latest information. +PEW RESEARCH CENTER -• The content of the guide is subject to change without notice. To obtain the latest version, go to +www.pewresearch.org -https://support.lenovo.com. + 6 + +PEW RESEARCH CENTER + +10.4%, nearly reverting to the rate in 2017, instead of sinking to a new low of 8.7%, as previously +expected. + +The steep rise in global poverty is driven by the fact that many who were in the low-income tier +prior to the pandemic lived on the margin of poverty. In 2017, 877 million people lived on $2 to $3 +a day, according to World Bank estimates. Thus, large numbers were susceptible to falling back +into poverty prior to the pandemic, especially in South Asia and Sub-Saharan Africa. -© Copyright Lenovo 2024 +At the other end of the economic spectrum, the global high-income population likely fell to 531 +million in 2020 from the 593 million originally expected. Consequently, the expected share in the +high-income tier is 6.8%, not 7.6%. A vast majority of the global high-income population lives in +advanced economies, and this group of countries also experienced a sharp downturn in 2020. +Ironically, this bolstered the global middle-income population as people in advanced economies +slipped down the ladder from higher income tiers. -iii +The reversal in global living standards in 2020 comes on the heels of notable progress earlier in +the decade. From 2011 to 2019, the global middle-class population increased from 899 million to +1.34 billion, or by 54 million people annually, on average. The pandemic is estimated to have +erased a year of growth, +leaving the global middle-class +population nearly unchanged +from 2019 to 2020. + +The global middle class expanded rapidly from 2011 to +2019, and poverty fell sharply + +Number of people in each income tier, in millions + +Global poverty had abated at +an average rate of 49 million +annually from 2011 to 2019, +with the number of global +poor falling from 1.10 billion +to 691 million during this +period. The pandemic, by +adding 131 million to the ranks +of the poor, has set the +progress on poverty back by +several years, and the poverty +rate is estimated to have +increased from 9.0% in 2019 +to 10.4% in 2020. - iv +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on +$10.01-$20, upper-middle income on $20.01-$50, and high income on more than $50; +figures expressed in 2011 purchasing power parities in 2011 prices. It is assumed there is +no change in the income distribution in a region from the benchmark year for the projection +to 2019. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -User Guide +PEW RESEARCH CENTER - Chapter 1. Meet your computer +www.pewresearch.org -Front + 7 -No. +PEW RESEARCH CENTER -1 +From 2011 to 2019, the global high-income population had increased from 459 million to 576 +million, growing by an average of 15 million annually. Thus, the setback of 62 million due to the +pandemic in 2020 also reverses several years of growth in the high-income population. The share +of the global population in the high-income tier is expected to have decreased from 7.5% in 2019 +to 6.8% in 2020. -2 +The effect of the COVID-19 downturn on the middle class and other income tiers shows distinct +regional differences. Asia is expected to have accounted for most of the shrinking of the global +middle class in 2020, with the middle-class population falling by 32 million in South Asia and by +19 million in East Asia and the Pacific. Meanwhile, the middle class in advanced economies is +estimated to have increased by 16 million, in part due to the drop-off in the high-income tier in +those countries. -3 +South Asia is estimated to have experienced the largest decrease in the middle +class and the greatest increase in poverty due to the COVID-19 downturn in 2020 + +Estimated change in the number of people in each income tier due to the global recession in 2020, in millions + +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on $10.01-$20, upper-middle income on $20.01-$50, and +high income on more than $50; figures expressed in 2011 purchasing power parities in 2011 prices. The estimates show the diff erence in the +number of people expected in an income tier based on projections made before and after the onset of the pandemic in January 2020. It is +assumed there is no change in the income distribution in a region from the benchmark year for the projection to 2020. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” + +PEW RESEARCH CENTER + +www.pewresearch.org + + 8 + +PEW RESEARCH CENTER + +The projected rise in global poverty in the pandemic is concentrated in South Asia (an increase of +78 million) and Sub-Saharan Africa (40 million), historically among the poorest regions in the +world. The high-income population in advanced economies is estimated to have shrunk by 47 +million in 2020, accounting for most of the falloff globally. It should be noted that the global cutoff +for the high-income tier – more than $50 a day – is not much higher than the median income +across advanced economies, as an estimated 40% of the population in these countries is in the +global high-income tier. Thus, the contraction of the high-income tier in advanced economies is +more a reflection of the reduction in the standard of living of the typical person in these countries +than outcomes for the wealthy. + +The contribution of a region to global changes overall and the shift in its own income distribution +depends mostly on two factors: the sizes of income tiers in the region prior to the COVID-19 +recession and the impact of the downturn on economic growth in the region. + +Prior to the pandemic, nearly half of the global middle-class population – 672 million of 1.38 +billion – was estimated to live in East Asia and the Pacific, a region that includes China. Latin +America and the Caribbean accounted for 191 million, and Europe and Central Asia was home to +183 million. South Asia accounted for 123 million more. + +In view of these magnitudes, the estimated loss (relative to pre-pandemic expectations) of 32 +million people from the middle class in South Asia is notable for two reasons. First, the region +contributed the most to the shrinking of the global middle class in 2020 despite having a middle- +income population smaller than several other regions. Second, the percentage decrease in South +Asia’s middle-income population from what was previously expected (26%) is much greater than +in any other region. Although East Asia and the Pacific shed 19 million from the middle-income +tier, that represented only 3% of its middle class. -4 +Poverty is most prevalent in Sub-Saharan Africa. Some 494 million people in Sub-Saharan Africa, +out of a total population of 1.14 billion, were expected to be living in poverty ahead of the +pandemic in 2020. South Asia, with a total population of 1.86 billion, was expected to be home to +104 million poor. Nonetheless, South Asia is estimated to have added nearly double the amount of +people to the ranks of the globally poor as Sub-Saharan Africa in the pandemic. The percentage +increase in poverty in South Asia (75%) dwarfs the increase in Sub-Saharan Africa (8%). -5 +The number of poor in other regions was smaller, estimated to range from 6 million in Europe and +Central Asia to 27 million in Latin America and the Caribbean ahead of the pandemic in 2020. +But, since that moment, the increase in poverty in these regions is notable. The Middle East and +North Africa and East Asia and the Pacific are each estimated to have seen increases in poverty of -6 +www.pewresearch.org -7 + 9 -* for selected models +PEW RESEARCH CENTER -Description +Prior to the pandemic, most of the global middle class lived in East Asia and the +Pacific, the poor in Sub-Saharan Africa, high-income earners in advanced economies -Microphones +Estimated number of people in each income tier prior to the global recession in 2020, by region, in millions -Infrared LED* +Notes: The poor live on $2 or less daily, low income on $2.01-$10, middle income on $10.01-$20, upper-middle income on $20.01-$50, and +high income on more than $50; figures expressed in 2011 purchasing power parities in 2011 prices. It is assumed there is no change in the +income distribution in a region from the benchmark year for the projection to 2020. +Source: Pew Research Center analysis of data from the World Bank PovcalNet database. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -Camera shutter +PEW RESEARCH CENTER -Camera +more than 20% due to the recession. The rise in poverty in both Europe and Central Asia and +Latin America and the Caribbean is not far behind, with each experiencing a 17% increase in +poverty. -Camera light +A large majority of the global high-income population – 489 million of 593 million – lived in +advanced economies before the global recession took hold. The estimated decrease of 47 million, +or 10%, in this population in advanced economies accounts for most of the shrinking of the global +high-income tier. -Screen +In proportional terms, the drop off in the high-income tier is also notable in Latin America and the +Caribbean (15%), Europe and Central Asia (12%) and East Asia and the Pacific (12%). The losses in +the Middle East and North Africa, South Asia and Sub-Saharan Africa are relatively small in +magnitude (each less than 1 million people) but also represent double-digit shares of the high- +income populations in those regions. -Antennas +www.pewresearch.org -© Copyright Lenovo 2024 + 10 -1 +PEW RESEARCH CENTER - Microphones +The relatively outsized role of +South Asia in the contraction of +the global middle class and the +expansion of poverty is the +result of it seeing the sharpest +reduction in economic growth +in the pandemic. In January +2020, the World Bank +forecasted that South Asia +would experience a 4.3% +increase in gross domestic +product (GDP) per capita in +2020. But, in January 2021, +output in South Asia was +estimated to have decreased by +7.8% in 2020, a retreat of 12.1 +percentage points from +potential. -The microphones are the PC's built-in sound input devices. They capture your voice and ambient sound and -convert them into digital form. Microphones are essential components when you use your PC for video -conferencing or voice recording. +South Asia estimated to have seen the sharpest drop +in output growth in 2020, East Asia and the Pacific +the least -Infrared LED +Growth in real gross domestic product per capita in 2020, forecasted and +estimated, in percent -The infrared LED generates and emits near-infrared waves that are received and used by a camera (or a -dedicated infrared camera) for facial recognition. +Global -Camera shutter +Advanced economies -The camera shutter is a sliding cap that you can move to block the camera lens. +Latin America and the Caribbean -Note: The camera shutter is designed for privacy protection. When the camera lens is blocked, the camera -function is disabled. +East Asia and the Pacific -Camera +Europe and the Central Asia -The built-in camera captures visible light and converts it to digital signals. It is used for video recording and -video conferencing. +South Asia -The camera for some models can also detect near-infrared waves. For those models, an IR LED is also -included to emit near-infrared waves. They are used together to achieve face-based authentication. +Middle East and North Africa -Camera light +Sub-Saharan Africa -The camera light indicates whether the camera is activated. +Jan. 2020 +forecast -Table 1. Camera light status and description +Jan. 2021 +estimate Difference -Camera light status + 2.5% -Description +1.0 -On +0.8 -Off +5.1 -Screen +2.3 -The camera is activated. +4.3 -The camera is not activated. +0.7 -The screen of the built-in display is where text, graphics, and videos are displayed. +0.3 -The screen is touch-enabled, allowing you to interact with your PC by intuitively touching buttons, icons, and -menu items displayed on the screen. Touch-enabled screens also support multi-finger gestures. +-4.3 -Antennas +-5.8 -The antennas transmit and receive radio waves to allow data to be transferred between your PC and a Wi-Fi -network device or a Bluetooth device. +-7.8 -Note: The antennas are hidden inside the PC. +0.4 -2 +-3.2 -User Guide +-7.8 - Base +-6.5 -No. +-6.1 -1 +-6.8 -2 +-6.8 -Keyboard +-8.6 -Description +-4.7 -Keyboard +-5.5 -Touchpad +-12.1 -The keyboard is the PC's built-in input device for typing characters. It also includes keys that can make you -more productive when interacting with the PC, apps, and the Windows operating system. +-7.2 -Table 2. Productivity keys +-6.4 -Key +Notes: Percentages shown are for the change in real GDP per capita at market prices. The +estimate for advanced economies is computed as the percent change in GDP less the +percent change in population. +Source: World Bank, Global Economic Prospects, January 2021 and January 2020 editions. +“The Pandemic Stalls Growth in the Global Middle Class, Pushes Poverty Up Sharply” -Location on the keyboard +PEW RESEARCH CENTER -Function +The setback to economic +growth in other regions is +smaller, ranging from 4.7 percentage points in East Asia and the Pacific to 8.6 points in Latin +America and the Caribbean. East Asia and the Pacific is the only region in which growth is +estimated to have stayed positive in 2020, with GDP per capita inching up by 0.4% even after the +onset of the pandemic. Otherwise, the latest estimates of the change in output per capita in 2020 +ranged from -3.2% in Europe and Central Asia to -7.8% in Latin America and the Caribbean. -Function keys (F1 through -F12) +www.pewresearch.org -In the top row + 11 -Hotkeys +PEW RESEARCH CENTER -In the top row +Acknowledgments -Fn key +This report is a collaborative effort based on the input and analysis of the following individuals. -In the lower-left corner +Rakesh Kochhar, Senior Researcher +Kim Parker, Director, Social Trends Research +Richard Wike, Director, Global Attitudes Research +James Bell, Vice President, Global Strategies +J.J. Moncus, Research Assistant +Aidan Connaughton, Research Assistant +David Kent, Senior Copy Editor +Travis Mitchell, Digital Producer +Michael Keegan, Senior Information Graphics Designer +Shannon Greenwood, Digital Producer -Windows key +www.pewresearch.org -In the lower-left corner + 12 -Executing predefined commands or -functions for selected apps +PEW RESEARCH CENTER -• Changing some of the PC's frequently +Methodology -used settings +The starting point for the analysis is household survey-based estimates of regional income +distributions from the World Bank’s PovcalNet database. The global regions are as defined by the +World Bank. In this report, “advanced economies” refers to the group of countries listed as “other +high income” by the World Bank. -• Opening an app or a Windows gadget +These data were used to estimate a benchmark distribution of the population in each of seven +major global regions across five income tiers. The income tiers are defined by the daily per capita +income or consumption of people in a region as follows: poor ($2 or less daily), low income +($2.01-$10), middle income ($10.01-$20), upper-middle income ($20.01-$50) and high income +(more than $50). All dollar figures are expressed in 2011 prices and purchasing power parity +dollars. -• Toggling the functions of dual-function +The year for which the benchmark estimates were drawn was the latest one in which the +household surveys covered at least 75% of the region’s population. This varied across regions as +follows: 2018 in advanced economies, Latin America and the Caribbean, East Asia and the Pacific, +and Europe and Central Asia; 2017 in Sub-Saharan Africa; 2015 in the Middle East and North +Africa; and 2014 for South Asia. -keys +Extrapolating the benchmark estimates to 2020 required the following three steps: -• Used with several letter keys to change +1. Extrapolate benchmark-year incomes to 2020 +2. Estimate income distributions in 2020 based on extrapolated incomes +3. Determine the population in each income tier in 2020 -some of the PC's frequently used -settings +Extrapolating incomes from benchmark years to 2020 -Used with certain keys to change some -frequently used Windows settings +Household incomes are extrapolated to 2020 using estimates of the growth in real gross domestic +product (GDP) per capita in a region. This, in turn, required several sets of data. -Hotkeys share keys with function keys. To toggle the functions for these keys, hold down the Fn key or turn -on the FnLock switch. +The first is the level of GDP per capita in the benchmark year, say 2017 for Sub-Saharan Africa, +and in 2019, the latest year for which data are available in the World Bank’s World Development +Indicators database. In this database, GDP per capita is expressed in constant 2017 international +dollars. The result is an estimate of the percentage change in per capita income in a region from +the benchmark year to 2019. -Chapter 1. Meet your computer +Next, the change in income from 2019 to 2020 is approximated by two measures of growth in real +GDP per capita (at market prices) issued by the World Bank in its reports on global economic -3 +www.pewresearch.org - Note: Keyboard layout varies by language and geographical location. The keyboard on your PC may look -slightly different from the product illustrations of this publication. + 13 -Touchpad +PEW RESEARCH CENTER -The touchpad is the PC's built-in pointing device, which provides the basic functionality of an external -mouse. Slide your finger on the touchpad to move the pointer on the screen and tap or double-tap to select -or execute a screen item. +prospects. The first measure of growth in 2020 is a forecast released in January 2020, prior to the +emergence of the pandemic. The second measure is an estimate released in January 2021, with the +economic effects of the pandemic more apparent. The result is two estimates of how much per +capita income changed in each region from the benchmark year to 2020, one in a counterfactual +“pre-pandemic world” and the other in a “post-pandemic” world, the period since the emergence +of the coronavirus. -The touchpad also supports Windows multi-finger gestures, which provide shortcuts to frequently used apps -and functions. +In a slight variation to this procedure, the change in per capita income in 2020 for advanced +economies is estimated as the difference in the growth in GDP and the growth in population. This +is because estimates of the change in GDP per capita in 2020 for this group of countries is not +directly reported in the source data. -4 +Estimating income distributions in 2020 based on extrapolated incomes -User Guide +The estimated change in income from the benchmark year to 2020 is used to reset the boundaries +of the income tiers. For example, based on the pre-pandemic forecast, it was estimated that per +capita income in East Asia and the Pacific would have increased by 10.5% from 2018 to 2020. As a +result, people living on between $1.82 to $2 a day in East Asia and the Pacific in 2018 would have +seen their income increase to more than $2 a day in 2020. Thus, they would no longer be poor in +2020; only people with incomes of $1.81 or less in 2018 would still be poor in 2020. Thus, $1.81 is +used as the poverty boundary to derive an estimate of the share who were poor in East Asia and +the Pacific in 2020 in the hypothetical pandemic-free world. - Left +With the emergence of the pandemic, the growth in per capita income in East Asia and the Pacific +in 2020 is estimated to have been 5.6%. In this circumstance, only people with incomes from +$1.90 to $2 a day in 2018 would have emerged from poverty in 2020. Thus, $1.89 is used as the +poverty boundary to estimate the state of poverty in East Asia and the Pacific in 2020 in the post- +pandemic scenario. -No. +A similar procedure is used to reset the boundaries of all income tiers in all regions. The +estimation of income distributions for 2020 was greatly facilitated by the “povcalnet” command in +Stata developed by World Bank researchers. (Castaneda Aguilar, R.A., C. Lakner, E.B. Prydz, J.S. +Lopez, R. Wu and Q. Zhao, Estimating Global Poverty in Stata: The Povcalnet Command, +September 2019). -1, 5 +www.pewresearch.org + + 14 + +PEW RESEARCH CENTER + +Determining the population in each income tier in 2020 -2 +The estimates derived from PovcalNet yield the percentages of people in a region who are poor, +low income and so on. These percentages are applied to the estimated population of each region to +determine the number who fall into each income tier. -3 +The overall population in a region in 2020 is derived from two sources: World Bank data on +population through 2019, as listed on PovcalNet, and United Nations forecasts of the change in +population from 2019 to 2020. This exercise is conducted at the country level, some 218 in all, and +the results are summed to obtain an estimate of the population in each region in 2020. -4 +Key assumptions and potential uncertainties + +As is the case with any projection, the estimates reported in this analysis are subject to a degree of +uncertainty. One issue is that final end-of-year estimates of real GDP growth are not yet available +for most countries, if any. Thus, the January 2021 estimates from the World Bank that are used in +the analysis are likely to be revised in the coming months. In its projections of the increase in +extreme poverty ($1.90 or less a day) in 2020, the World Bank assumes a global contraction of +either 5% or 8%. The resulting change in extreme poverty in 2020 is estimated to range from 88 +million to 115 million. + +The methodology in this analysis assumes that incomes change at the same rate for all people, +whether poor, low income, middle income, upper-middle income or high income. In other words, +it is assumed that there is no change in inequality from the benchmark year to 2020. If the +COVID-19 recession has worsened inequality, the increase in the number of poor is likely greater +than estimated in this analysis, and the decrease in the number who are high income is likely less +than estimated. The middle class may have shrunk by more than projected. + +Although World Bank researchers have simulated the relationship between inequality and the +poverty rate, more precise measures of changes in inequality in the COVID-19 era will not be +forthcoming until newer household survey data become available. Also, the extent to which social +spending by governments, especially in advanced economies, may have ameliorated the effects of +the COVID-19 downturn is not yet clear. -6 - -7 - -Description - -Multi-purpose USB Type-C® connector - -Charging light - -USB Standard-A connector (always-on) - -HDMI™ connector - -Combo audio jack - -Speaker - -Multi-purpose USB Type-C connector - -This USB Type-C® connector is the PC's power input connector. Use the included power adapter and this -connector to supply power to the PC. - -When this connector is not used by the included power adapter, it can also be used to connect: - -• Storage or peripheral devices that follow the universal serial bus (USB) specification for data transfer and - -device interconnection - -• Display devices - -Note: When connecting display devices, you need to use appropriate cables and adapters (if needed) -according to the connection capabilities of the display device. - -• Thunderbolt™-enabled docks or devices (for the connector with an icon next to it) - -Charging light - -The charging light indicates whether the PC is plugged into an electrical outlet. When the PC is plugged into -an electrical outlet, the color of the light indicates whether the battery is fully charged (or will shortly be fully -charged). - -Chapter 1. Meet your computer - -5 - - Table 3. Charging light statuses and descriptions - -Light status - -Off - -On, amber - -On, white - -Plugged in? - -Battery charge level - -No - -Yes - -Yes - -/ - -1%–90% - -91%–100% - -USB Standard-A connector - -The USB Standard-A connector is used to connect storage or peripheral devices that follow the universal -serial bus (USB) specification for data transfer and device interconnection. - -Always-on connector - -A USB connector with a battery icon ( -USB device connected to this type of connector even when the PC is powered off, in sleep mode, or in -hibernation mode. - -) supports the always-on function. The PC can supply power to a - -The always-on function can be turned on and off in: - -• The PC's firmware setup utility, or - -• Lenovo Vantage or Lenovo PC Manager - -HDMI connector - -The HDMI connector is used to connect an external display device, such as a television, a projector, or a -monitor. - -Combo audio jack - -The combo audio jack is used to connect single-plug headsets, headphones, or external speakers. - -Speakers - -The speakers are the PC's built-in sound output devices. - -6 - -User Guide - - Right - -No. - -1 - -2 - -3 - -4 - -5 - -6 - -Description - -Speaker - -Power light - -Power button (with an embedded fingerprint sensor) - -microSD™ card slot - -USB Standard-A connector - -Security lock slot - -Speakers - -The speakers are the PC's built-in sound output devices. - -Power light - -The power light indicates the current power state of the PC: whether it is powered on, powered off, in sleep -mode, or in hibernation mode. - -When the PC is powered on, this light can also indicate low battery by blinking rapidly. - -Table 4. Power light status and description - -Light status - -White (solid on) - -White (blinking rapidly) - -Power state - -Powered on - -Powered on - -White (blinking slowly) - -In sleep mode - -Off - -Powered off or in hibernation mode - -Battery charge level - -21%–100% - -1%–20% - -/ - -/ - -If the PC's power button includes an embedded fingerprint sensor, the power light turns solid green to -prompt you to enroll or scan your fingerprint. - -Chapter 1. Meet your computer - -7 - - Power button - -Press the power button to turn on your PC. - -Note: By default, on a Windows PC, pressing the power button when the PC is turned on will put the PC into -sleep mode. - -Fingerprint sensor - -The fingerprint sensor scans your finger to unlock your PC or verify your identity. - -Note: You need to enroll one or more fingerprints before using them for identity verification. On a Windows -PC, go to Settings ➙ Accounts ➙ Sign-in options to enroll your fingerprints. - -microSD card slot - -The microSD card slot is used to insert a microSD, microSDHC, or microSDXC memory card to transfer data -between the memory card and your PC. - -USB Standard-A connector - -The USB Standard-A connector is used to connect storage or peripheral devices that follow the universal -serial bus (USB) specification for data transfer and device interconnection. - -Security lock slot - -The security lock slot is used to attach a compatible cable lock. The cable lock typically includes a loop at -one end and can be used to secure the PC to a stationary object, preventing theft when the PC is left -unattended for a short period of time in cafes, shops, libraries, and other public places. - -8 - -User Guide - - Bottom - -No. - -1 - -2 - -Description - -Air vents (intake) - -Novo button hole - -Avoid constant body contact with specific hot sections -CAUTION: -When the computer is operating, it should be placed on a hard and flat surface with its bottom area -not in contact with user’s bare skin. Under normal operating conditions, the temperature of the -bottom surface will remain within an acceptable range as defined in IEC 62368-1, but such -temperatures can still be high enough to cause discomfort or harm to the user if directly touched for -over one minute at a time. As such, it is recommended that users avoid prolonged direct contact with -the bottom of the computer. - -Air vents (intake) - -The air vents allow air to be sucked inside of the PC to cool the internal components. - -Important: When the PC is operating, do not place it on a bed, sofa, carpet, or other flexible surfaces. -Otherwise, the air vents will be blocked and the PC may overheat, reducing performance or causing the PC -to be unresponsive or even shut down. - -Novo button hole - -When the PC is powered off, you can press the Novo button to display the Novo button menu. From the -menu, you can then choose to: - -• Open the firmware setup utility - -• Display the boot device selection menu - -• Display the Windows advanced startup options page - -Chapter 1. Meet your computer - -9 - - Note: The Novo button is rarely used during normal PC operations. To prevent users from accidentally -pressing it, the Novo button is placed in a recessed hole. You can use a straightened paper clip to press this -button. - -10 - -User Guide - - Rear - -No. - -1 - -Description - -Air vents (outlet) - -Air vents (outlet) - -The air vents allow hot air to be discharged out of the PC. - -Important: When the PC is operating, do not place it on a bed, sofa, carpet, or other flexible surfaces. -Otherwise, the air vents will be blocked and the PC may overheat, reducing performance or causing the PC -to be unresponsive or even shut down. - -Chapter 1. Meet your computer - -11 - - Features and specifications - -Dimensions - -Width - -Depth - -Thickness - -ac power adapter - -Input - -Output - -Power - -Rechargeable battery pack - -Capacity - -Cell type - -312.85 mm - -223.5 mm - -• Thinnest: 16.85 mm - -• Thickest: 18.7 mm - -100 V ac–240 V ac, 50 Hz–60 Hz - -20 V dc, 3.25 A - -65 W - -60 Wh - -Lithium-polymer - -Number of cells - -4 - -Note: The battery capacity is the typical or average capacity as measured in a specific test environment. -Capacities measured in other environments may differ but are no lower than the rated capacity (see product -label). - -Memory - -Type - -Installation - -Number of slots - -Mass storage device - -Type - -Slot type - -Interface - -Display - -DDR5 - -Slot mounted - -2 - -Solid-state drive (SSD) - -• M.2 (2242), or - -• M.2 (2280) - -PCIe Gen 4 - -Screen size (diagonal) - -14 inches - -Resolution - -Refresh rate - -1920 × 1200 pixels - -60 Hz - -12 - -User Guide - - Connectors and slots - -USB Standard-A connector - -• Quantity: 2 - -• Maximum power output: 5 V, 0.9 A - -• Supported signaling protocols: - -– USB 2.0 480 Mbps - -– SuperSpeed USB 5 Gbps - -Multi-purpose USB Type-C -connector - -• Quantity: 2 - -Note: The always-on connector supports maximum power output of 5 V and 2 A. - -• Maximum power output: 15 W (5 V, 3 A) - -• Maximum power input: 65 W - -• Supported signaling protocols: - -– USB 2.0 480 Mbps - -– SuperSpeed USB 5 Gbps - -– SuperSpeed USB 10 Gbps - -– DisplayPort™ 1.4 - -HDMI connector - -Combo audio jack - -– Thunderbolt 4 41.25 Gbps (for the connector with an icon next to it) - -Supported signaling protocol: Transition minimized differential signaling (TMDS) - -• Diameter: 3.5 mm - -• Supported plugs: - -– 3-pole, TRS - -– 4-pole, TRRS (CTIA and OMTP) - -microSD slot - -Supported card types: - -• microSD™ - -• microSD High Capacity (microSDHC™) - -• microSD Extended Capacity (microSDXC™) - -Security lock slot - -Kensington Nano Security Slot™ - -Note: Data rates and performance ratings are dependent on connected devices and cables if they are used. -USB Type-C connectors that are DisplayPort 1.4 compliant through DisplayPort alternate mode provide a -maximum output resolution of 5120 x 3200, at a 60 Hz frame rate and color depth of 24 bpp (bits per pixel). -The actual maximum output resolution is dependent on the connected display device and the cable being -used. - -Networking - -Wi-Fi® - -Bluetooth - -• Wi-Fi 6, or - -• Wi-Fi 6E - -Bluetooth 5.2 - -Note: Support of Bluetooth 5.3 may require a future operating system update. - -Chapter 1. Meet your computer - -13 - - Statement on USB transfer rate - -Depending on many factors such as the processing capability of the host and peripheral devices, file -attributes, and other factors related to system configuration and operating environments, the actual transfer -rate using the various USB connectors on this device will vary and will be slower than the data rate listed -below for each corresponding device. - -USB device - -USB 3.2 Gen 1 - -USB 3.2 Gen 2 - -Data rate (Gbit/s) - -5 - -10 - -Operating environment - -Maximum altitude (without pressurization) - -3048 m (10 000 ft) - -Temperature - -• At altitudes up to 2438 m (8000 ft) - -– Operating: 5°C to 35°C (41°F to 95°F) - -– Storage: 5°C to 43°C (41°F to 109°F) - -• At altitudes above 2438 m (8000 ft) - -– Maximum temperature when operating under the unpressurized condition: 31.3°C (88°F) - -Note: When you charge the battery, its temperature must be no lower than 10°C (50°F). - -Relative humidity - -• Operating: 8% to 95% at wet-bulb temperature 23°C (73°F) - -• Storage: 5% to 95% at wet-bulb temperature 27°C (81°F) - -14 - -User Guide - - Chapter 2. Get started with your computer - -Connect to a network - -Connect to a Wi-Fi network - -Ensure that you have a secure Wi-Fi network account and the required credentials. - -Step 1. Select the network icon - -on the bottom right of your display. - -Step 2. Select an available network, and then select Connect. If you want to be automatically connected to - -this Wi-Fi network the next time you start your computer, select Connect automatically before -selecting Connect. - -Step 3. - -Input your credentials if necessary, and then follow the on-screen instructions to connect to the -disired Wi-Fi network. - -Establish a wired connection - -Step 1. Plug an Ethernet cable into the Ethernet connector on your computer. - -Step 2. Plug the other end of the Ethernet cable into a network wall jack or a router. - -Note: If your computer does not include an Ethernet connector, you can purchase a USB-C to Ethernet -adapter from Lenovo at https://www.lenovo.com/accessories. - -© Copyright Lenovo 2024 - -15 - - Unique Lenovo apps - -Lenovo Vantage -Lenovo Vantage is a one-stop solution to help you update your computer, configure hardware settings, and -access personalized support. - -If your computer is pre-installed with Lenovo Vantage, type Vantage in the Windows search box to launch -this app. - -Notes: - -• Available features may vary depending on your computer model. - -• You can download the latest version of this app from Microsoft Store. - -Lenovo Smart Meeting -Lenovo Smart Meeting is a video conferencing app with multiple features for enhancing your professional -image, protecting your privacy, and reducing your computer's power consumption. - -If you want your settings in this app to also take effect on other mainstream video conferencing apps, such -as Microsoft Teams and Zoom, ensure that you select Lenovo Virtual Camera in the app. - -Access the app - -Type Lenovo Smart Meeting in the Windows search box and then press Enter. - -Explore key features - -• Smart appearance - -– Video enhancer: Adjust the brightness automatically for better image quality during the video call. - -– Face framing: Keep your face centered automatically during the video call when you move around. - -– Customized background: Blur or customize your background during the video call to protect your - -privacy. - -16 - -User Guide - - – Dual Camera Fusion: Show the image of any participant that is visible from the front or rear of the - -camera on the screen when collaborating with a Lenovo Magic Bay 4K Webcam. - -Note: You can purchase the Lenovo Magic Bay 4K Webcam from Lenovo at https://www.lenovo.com/ -accessories. - -– Temporary Avatar: Create and display a temporary portrait of you as if you were still in the video - -conference when you are temporarily away. - -• Intelligent sensing - -– Presence detection: Keep your computer awake when you are facing the screen and lock the - -computer screen automatically when you leave. You can customize the screen-off time from 10 to 60 -seconds. - -– Privacy protection: Blur your screen and display a warning message when a shoulder surfer is - -detected. - -Chapter 2. Get started with your computer - -17 - - – Adaptive dimming: Track your face movements to recognize your focus of attention. When you are - -facing away, the display dims to conserve power. You can customize the brightness value and screen -dimming timer. - -Notes: - -• Lenovo does not collect any personal data from this app. - -• The available features vary depending on the computer model. - -• Lenovo Smart Meeting makes periodic feature updates to keep improving your experience. The - -description described here might be different from that on your actual user interface. - -Work with Windows - -The following table contains Windows frequently used settings. You can learn the basics and start working -with Windows right away. - -To configure settings, type the corresponding keywords in the Windows search box and select the best -match. Follow the on-screen instructions to customize settings. - -Table 5. Basic settings - -Functions - -Descriptions - -Control Panel - -View or change Windows settings, including hardware and software setup -and configuration. - -Connect to Wi-Fi -networks - -For models with a wireless LAN module, you can connect your computer to -a Wi-Fi network. Click the network icon in the Windows notification area, -and then select a network for connection. -Note: The wireless LAN module on your computer may support different -standards. For some countries or regions, use of 802.11ax may be disabled -according to local regulations. - -Search by -keywords - -Control Panel - -Wi-Fi - -Airplane mode - -Airplane mode is a convenient setting to turn all wireless communications of -your computer on and off. You may need to turn it on when boarding an -airplane. - -Airplane mode - -18 - -User Guide - - Table 5. Basic settings (continued) - -Functions - -Descriptions - -Night light mode - -Adjust color -temperature - -Windows -Updates - -Night light is a switch in Windows that you can turn on and off. When it is -turned on, your screen shows warmer colors and the amount of emitted -blue light is reduced. Enabling night light reduces the chances of developing -eye fatigue or eye strain. -Note: For more guidance on reducing eye fatigue or eye strain, go to -https://www.lenovo.com/us/en/compliance/visual-fatigue. - -If night light mode is turned on, you can adjust the color temperature of the -screen. -Note: Selected Lenovo computers are low blue light certified using night -light mode and color temperature settings available in the pre-installed -operating system. These computers are tested with night light turned on -and color temperature set to 48 or greater. - -Microsoft periodically releases feature and security updates to the Windows -operating system. Updates that are applicable to your Windows version are -downloaded automatically when your computer is connected to the -Internet. When updates are downloaded, you are prompted to restart the -computer to install these updates. You can also manually check if there are -available updates for the installed version of Windows. -Attention: Only use Windows Update to download and install updates. -Updates downloaded from other sources may include security risks. - -Table 6. Optional settings - -Functions - -Descriptions - -Search by -keywords - -Night light - -Night light - -Windows Updates - -Search by -keywords - -Sign-in options - -Set up face -recognition* - -Enroll your -fingerprint - -For models with a Windows Hello-compatible infrared camera, you can -set up face recognition to sign in with your face. - -For models with a fingerprint sensor, you can enroll your fingerprint for -Windows sign-in. - -Sign-in options - -* for selected models - -Windows help information - -If the on-screen instructions cannot solve your problem, refer to the following to get the online Windows help -information. - -• Type Get Help or Tips in the Windows search box and then press Enter. When the app opens, type the - -problem description and select the matched result. - -• Visit Microsoft support Web site: https://support.microsoft.com. Enter what you are looking for in the - -search box and get search results. - -The Novo Button menu - -The Novo Button menu can be displayed before the operating system starts. From the menu, you can -choose to - -• Open the BIOS/UEFI setup utility - -• Open the boot device selection menu - -• Open the Windows startup options screen - -Chapter 2. Get started with your computer - -19 - - Note: From the Windows startup options screen, you can then choose to - -Start your computer using a recovery drive - -Reset your computer - -Open advanced options screen - -Open the Novo Button menu - -For Lenovo computers with a Novo button, you can press the button to open the Novo Button menu. - -Step 1. Turn off the computer. - -Step 2. Open the LCD screen and press the Novo button. - -Note: Alternatively, turn off the computer. Press Fn and the power button to open the Novo Button menu. - -Interact with your computer - -Hotkeys - -Look at the top row of the keyboard of your Lenovo computer and you’ll find keys with printed icons. These -keys are called hotkeys and they provide shortcuts to functions that you’ll use frequently. The function that -each hotkey provides is denoted by the icon printed on the key. - -Hotkeys share keys with function keys (F1–F12) and other keys. Depending on the FnLock setting, you use -hotkey functions by pressing the key directly or in combination with the Fn key. - -Note: The FnLock switch does not apply to hotkeys not found in the first row of the keyboard. To use these -hotkeys, always hold down the Fn key while pressing the key. - -Figure 1. Location of the Fn key - -Related topics - -Appendix A “Function descriptions for Lenovo hotkeys” on page 39 - -The FnLock switch -The FnLock is an electronic switch that affects how you use hotkey functions. To turn it on and off, press Fn -+ Esc. - -Note: The Esc key is in the upper left corner of the keyboard. It has an LED that indicates the status of the -FnLock switch. - -20 - -User Guide - - Figure 2. Locations of the FnLock key and the Fn key - -A Lenovo keyboard usually contains hotkeys in the top row. These hotkeys share keys with the function keys -(F1–F12) and other keys. For these dual-function keys, the icons or characters denoting the primary functions -are printed on top of the icons and characters denoting the secondary functions. - -• A: an icon or character denoting the primary function - -• B: an icon or character denoting the secondary function - -Figure 3. The layout of a dual-function key - -Table 7. FnLock and dual-function keys - -FnLock -(Esc) -LED - -Off - -On - -FnLock -status - -Pressing the hotkey alone - -Pressing the hotkey while holding down -the Fn key - -Disabled - -Primary function - -Enabled - -Secondary function - -Secondary function - -Primary function - -Touch operations supported by Windows - -For computers with a touch-enabled screen, you can touch the screen directly with your fingers and interact -with your computer in a more natural way. The following table lists frequently used touch operations that are -supported by the Windows operating system. - -Chapter 2. Get started with your computer - -21 - - Touch operation - -Used for - -Opening apps, documents, and other functions on the -taskbar or the Start menu - -Tap - -Tap twice quickly - -Tap, hold and lift - -Pinch - -Spread - -Opening apps that are pinned on the desktop - -Opening the context menu - -Zooming out - -Zooming in - -Opening the widgets panel - -Swipe in from the left edge - -22 - -User Guide - - Touch operation - -Used for - -Opening the notification center - -Swipe in from the right edge - -The Windows operating system also supports 3- and 4-finger gestures on the screen and the touchpad. You -can set them up in Settings ➙ Bluetooth & devices. - -Connect to an external display - -Connect to a wired display -Connect your computer to the desired display with an appropriate cable. - -Step 1. Connect one end of the display cable to the HDMI connector or a multi-purpose USB Type-C - -connector on your computer. - -Step 2. Connect the other end of the cable to the display. - -Connect to a wireless display -Ensure that: - -• Both your computer and the display support Miracast® technology. - -• The display is connected to the same Wi-Fi network as your computer and is discoverable. - -Step 1. Press Windows key + K. - -The computer searches for wireless display devices and audio devices and lists the results. - -Step 2. Select the display you want to connect to, and then follow the on-screen instructions. - -Change display settings -Step 1. Right-click on a blank area on the desktop, and then select Display settings. - -Your computer shows the Display window. - -Step 2. Select the display for which you want to change the settings. - -Chapter 2. Get started with your computer - -23 - - Step 3. Change the display settings as necessary. - -Set the display mode -Step 1. Press - -or Fn + - -. - -Your computer shows a list of display modes, with the current mode highlighted. - -Step 2. Select a display mode from the list. - -Protect your privacy using the camera shutter - -The camera shutter is a mechanical component that prevents any attempt from capturing your image, thus -protecting your privacy. To cover the camera lens, slide the camera shutter to the left. When you want to use -the camera, slide the camera shutter to the right. - -If you slide the camera shutter to the left during a video call , people on the video call will not be able to see -you. If you slide the camera shutter back to the right, they will be able to see you again. - -Enroll your fingerprints - -You can use the biometric fingerprint option to sign in to Windows quickly if your computer includes a -fingerprint sensor. This sign-in option provides a reliable and secure way of identity verification. - -Step 1. Select Start ➙ Settings ➙ Accounts ➙ Sign-in options. - -Step 2. Under Ways to sign in, select Fingerprint recognition (Windows Hello) to set up sign-in with a - -fingerprint sensor. - -Step 3. Follow the on-screen instructions to enroll your fingerprints. - -Notes: - -• You need to set up a PIN code before you are allowed to use this sign-in option. - -• It is recommended that you enroll multiple fingerprints in case of any injuries to your fingers. - -Lenovo Slim Pen (for selected models) - -The Lenovo Slim Pen is referred to as the pen in this section. - -Depending on the model, your computer might come with a pen. The pen enables you to write and draw in a -natural manner. - -Notes: - -• The pen is not waterproof. Take care not to drop the pen in water. - -24 - -User Guide - - • The pen contains pressure sensors. Do not push the tip of the pen for an extended period of time. Do not - -subject the pen to shock or vibration. - -Before first use -Charge the pen before first use. - -Step 1. Open the pen cap. - -Step 2. Connect the pen to a charger with a USB Type-C cable. - -Step 3. Disconnect the charger cable. Place the pen cap back. - -Note: You can check the pen’s battery level in the Lenovo Pen Settings. - -Use the pen - -Figure 4. Lenovo Slim Pen - -Chapter 2. Get started with your computer - -25 - - Item - -Component - -Description - -Pen cap - -Top button - -Open the pen cap to charge the pen. - -To right-click, press and hold the top button and tap the screen in supported -applications. - -Bottom button - -To erase text or drawings, press and hold the bottom button and rub at the -targeted items with the pen tip in supported applications. - -Pen tip - -Write or draw on the touch screen. - -1 - -2 - -3 - -4 - -Notes: - -• The default function of each button may vary in different applications. - -• You can view or change the functions in the Lenovo Pen Settings. - -Store the pen -When the pen is not in use, attach it to the lower right side of your computer. - -Lenovo Pen Settings -Lenovo Pen Settings is a pre-installed app for setting up Lenovo digital pens. This app allows you to perform -the following functions: - -• Reassign pen button functions - -• Adjust pen sensitivity - -• Check the pen’s battery level - -To open the Lenovo Pen Settings, type the app name in the Windows search box and select the matched -result. If the software is not pre-installed, you can download and install the latest drivers from https:// -support.lenovo.com. - -26 - -User Guide - - Chapter 3. Explore your computer - -Intelligent features - -Your computer may be pre-installed with one of Lenovo Vantage or Lenovo PC Manager but not all. Most -features described here can be enabled or disabled in one of these apps. Other features may be enabled in a -standalone app. - -Notes: - -• Software features are subject to change. Please refer to your actual product. - -• You may need to complete online update to the apps for the features to take effect. - -Gesture controls - -Lenovo unique touch gestures -In addition to touch gestures supported by the Windows operating system, some Lenovo computers may -support additional touch gestures. You can perform these gestures on the touchpad of your computer. - -Gesture - -Function - -Increases/Decreases display brightness when you watch -videos in full screen mode. -Note: This gesture is not supported by all video players -and may conflict with similar third-party gesture -functions. - -Slide up/down on the left side of the touchpad -Note: For computers with a touch-enabled screen, you -can also perform this gesture on the left side of the -screen. - -Increases/Decreases system volume when you watch -videos in full screen mode. -Note: This gesture is not supported by all video players -and may conflict with similar third-party gesture -functions. - -Slide up/down on the right side of the touchpad -Note: For computers with a touch-enabled screen, you -can also perform this gesture on the right side of the -screen. - -Smart Noise Cancelling - -Smart Noise Cancelling is a noise reduction feature available on some Lenovo product models. By filtering -out input and output noises, Smart Noise Cancelling enhances your audio experience. - -© Copyright Lenovo 2024 - -27 - - Function - -Description - -Remarks - -Voice Recognition: Your computer captures multiple voices in -a way that reflects their original spatial positions. - -Microphone -noise cancelling - -Only My Voice: This option requires you to record your voice so -that your computer captures this voice only and tries to -eliminate other voices. -Note: To remove your voice record, select REMOVE MY -VOICE. - -Normal: Your computer focuses on the voice of the person -facing it and reduces ambient sounds. - -• This function takes effect - -only when built-in -microphones/arrays or 3.5 -mm jack wired microphones -are used as the input media. - -• To disable this function, - -select Off. - -Multiple Voices: Your computer captures multiple voices from -an expanded range in front of the computer. - -Speaker noise -cancelling - -Your computer filters out other sounds to play only human -voices. - -Meeting noise -cancelling - -When this function is selected, your computer uses special -algorithms for noise reduction when you are using conferencing -applications. - -These functions are not -applicable to scenarios like -listening to music and watching -videos. - -Notes: - -• Depending on its hardware, your computer may not support all the functions and options described - -above. - -• You can view and customize this feature under Device Settings in Lenovo Vantage or Lenovo PC - -Manager. - -Super Resolution - -By utilizing the capabilities and potentials of Intel processors, Super Resolution helps you play videos with a -higher resolution than the original. It works especially well in cases where the source video has a poor -resolution. - -For most players, Super Resolution can be enabled or disabled in Lenovo Vantage or Lenovo PC Manager, -but for some specific players, you might need to enable this feature manually. - -Eye Care Mode - -Eye Care Mode intelligently adjusts the color temperature of the screen and can reduce the chances of -developing eye fatigue or eye strain. - -28 - -User Guide - - Manage power - -Use the information in this section to achieve the best balance between performance and power efficiency. - -Rechargeable battery pack - -Your computer includes a built-in, rechargeable battery pack that makes mobile computing a reality. When -the computer is plugged into an electrical outlet, the battery charges. If you use the computer when you -don't have access to an electrical outlet, the battery discharges to supply electricity that the computer -system requires for operation. - -You can charge the battery any time you want. The battery packs of Lenovo computers support multiple -charging modes that are suitable for different power usage habits. You can switch the battery’s active -charging mode in Lenovo Vantage or Lenovo PC Manager. - -Battery charging is also affected by its temperature. The recommended temperature range for charging the -battery is between 10°C (50°F) and 35°C (95°F). - -Note: - -You can check the battery temperature in Lenovo Vantage. - -To maximize the life of the battery, once the battery is fully charged, it must discharge to 94% or lower -before it will be allowed to recharge again. - -Normal mode -Normal mode is the most basic charging mode. In normal mode, it typically takes 2 to 4 hours for the battery -to charge from 0% to 100%. - -Rapid charge mode -If you want the battery to be charged faster than in normal mode, switch the battery charging to rapid charge -mode. The following table lists the typical time needed for batteries in rapid charge mode to be charged to -80% and 100% respectively. - -Table 8. Reference charge time for batteries in rapid charge mode - -Mode - -Time needed to charge from 0% to 80% - -Time needed to charge from 0% to 100% - -Rapid charge - -Less than 1 hour - -Less than 2 hours - -Conservation mode -If your computer is constantly plugged into an electrical outlet, consider switching the battery charging to -conservation mode. In conservation mode, the battery will not be fully charged. Instead, the battery’s charge -will be kept within 75%–80%. This is beneficial to the long-term health of the battery. - -Note: If you want the battery to be fully charged before bringing the computer to work, disable conservation -mode by switching the battery charging to normal or rapid charge mode. - -Nighttime charge optimization -Some people follow a regular pattern when using their computers. They finish their workday with the -computer at a low battery charge level. They plug in their computers at night and need the battery to be fully -charged the next morning so they can unplug the computer and bring it to work. These activities happen at -approximately the same time each day. If this sounds like you, consider enabling nighttime charge -optimization for the battery. - -Chapter 3. Explore your computer - -29 - - Nighttime charge optimization affects battery charging during the night hours, the time when you’re usually -asleep. When it is enabled, the computer regularly adapts its charging behavior based on observation of -when you plug in the computer at night and unplug it in the morning. During the nighttime, the battery is -charged to a particular range and is kept within that range for an extended period, before being further -charged to 100%. Nighttime charge optimization ensures safe charging during the night and is beneficial to -the long-term health of the battery. - -Note: With nighttime charge optimization enabled, if you break your routine one day by unplugging the -computer much earlier than usual in the morning, you may find that the battery is not fully charged. - -If the battery pack of your computer supports nighttime charge optimization, it can be enabled in Lenovo -Vantage or Lenovo PC Manager. - -Recover full battery capacity -If your computer is constantly plugged in to an electrical outlet and the battery rarely discharges, the battery -may not be charged to its full capacity even if the battery meter reports 100% charge. You can recover the -battery’s full charging potential simply by discharging and re-charging the battery. - -Step 1. Unplug the computer and use it until the battery charge drops below 20%. - -Step 2. Plug in the computer and charge the battery to 100%. - -Set power button behavior - -By default, pressing the power button puts the computer to sleep mode. However, you can change the -power button behavior in Windows Control Panel. - -Step 1. Type Control Panel in the Windows search box and then press Enter. Open the control panel and - -view by large or small icons. - -Step 2. Select the power options and then click choose what the power button does. - -A power plan - -A power plan is a collection of power-saving settings made available by an operating system. With a power -plan, you can set idle time-outs for different hardware components to enter a low-power state. The default -power plan and some of its idle time-out settings are listed below for computer models pre-installed with -Windows. - -The listed settings are active when the computer is connected to an electrical outlet. If your computer -includes a built-in battery pack, a different collection of time-out settings is set to take effect when the -computer is operating on battery power. - -• Default power plan: Balanced - -• Turn off the display: after 5 minutes - -• Put the computer to sleep: after 5 minutes - -Note: To wake the computer from the sleep state, press the power button or any key on the keyboard. - -Change or customize a power plan -This operation is applicable to computers with pre-installed Windows. - -Step 1. Type power plan in the Windows search box and then press Enter. - -Step 2. Customize a power plan of your preference. - -30 - -User Guide - - System operation modes - -Lenovo has preset several modes in which your computer can operate. The maximum attainable -performance, power consumption, and speed limit for the heat sink fan vary between the operation modes. -Consider the following conditions when you want to switch operation modes. - -• The environment where you use your computer, and - -• The tasks running on your computer - -You can switch the operation mode in the pre-installed app Lenovo Vantage or Lenovo PC Manager. As a -shortcut, you can also use the key combination Fn + Q. Three modes are usually available for most Lenovo -computers. The following table lists the operation modes and the recommended conditions for each mode. - -Note: The operation modes listed in the table are descriptive and may not be the same as those displayed -by the app. - -Table 9. Operation modes and their recommended usage conditions - -Operation mode - -Recommended conditions - -High Performance - -• Your computer is plugged into an electrical outlet. - -• You want the best performance, and - -• You don't care if the fan makes a little noise. - -Auto (Balance) - -You plan to frequently switch between different computer tasks over a period of time. - -Power Saving (Quiet) - -• Your computer is operating on battery power, or - -• You want the computer to be as quiet as possible. - -Note: In Auto (Balanced) mode, the computer dynamically switches between High Performance mode and -Power Saving (Quiet) mode depending on the tasks running on the computer. - -Chapter 3. Explore your computer - -31 - - 32 - -User Guide - - Chapter 4. Help and support - -Frequently asked questions - -How do I partition my storage drive - -Refer to https://support.lenovo.com/solutions/ht503851. - -What should I do if my computer stops responding - -Press and hold the power button until the computer turns off. Then restart the computer. - -What should I do if I spill liquid on the computer - -1. Carefully unplug the ac power adapter and turn off the computer immediately. The more quickly you - -stop the current from passing through the computer the more likely you will reduce damage from short -circuits. - -Attention: Although you might lose some data or work by turning off the computer immediately, leaving -the computer on might make your computer unusable. - -2. Wait until you are certain that all the liquid is dry before turning on your computer. - -CAUTION: -Do not try to drain out the liquid by turning over the computer. If your computer has keyboard -drainage holes on the bottom, the liquid will be drained out through the holes. - -Where can I get the latest device drivers and UEFI/BIOS - -• Lenovo Vantage or Lenovo PC Manager - -• Lenovo Support Web site at https://support.lenovo.com - -Why does my computer start automatically when I open the lid - -Your computer may have Flip to Start enabled. Many Lenovo notebook computers include a sensor that can -detect the angle at which the lid is opened. When you open the lid, the sensor can detect this behavior. If Flip -to Start is enabled, the computer will respond by starting up automatically. - -If you don’t like this feature, you can disable it. Flip to Start can be enabled and disabled in: - -• Lenovo Vantage or Lenovo PC Manager - -• Firmware Setup Utility - -Self-help resources - -Use the following self-help resources to learn more about the computer and troubleshoot problems. - -© Copyright Lenovo 2024 - -33 - - Resources - -How to access? - -Troubleshooting and frequently asked questions - -• https://www.lenovo.com/tips - -• https://forums.lenovo.com - -Accessibility information - -https://www.lenovo.com/accessibility - -Reset or restore Windows - -• Use Lenovo recovery options. - -1. Go to - -https://support.lenovo.com/ -HowToCreateLenovoRecovery. - -2. Follow the on-screen instructions. - -• Use Windows recovery options. - -1. Go to https://pcsupport.lenovo.com. - -2. Detect your computer or manually select your - -computer model. - -3. Click Troubleshoot & Diagnose ➙ Custom -Troubleshooting ➙ Operating System -Diagnostics and then follow the on-screen -instructions. - -Use Lenovo Vantage or Lenovo PC Manager to: - -• Download and install the latest drivers and firmware. - -• Configure hardware settings. - -Use Windows Search. - -• Diagnose computer hardware problems. - -• Check the computer warranty status. - -Product documentation: - -• Generic Safety and Compliance Notices - -• Safety and Warranty Guide - -• Setup Guide - -• This User Guide - -• Regulatory Notice - -Lenovo Support Web site with the latest support -information of the following: - -• Drivers and software - -• Diagnostic solutions - -• Product and service warranty - -• Product and parts details - -• Knowledge base and frequently asked questions - -Windows help information - -1. Go to https://support.lenovo.com. - -2. Detect your computer or select computer model - -manually. - -3. Select Guides & Manuals and filter out the - -documentation you want. - -Visit https://support.lenovo.com - -• Use Get Help or Tips. - -• Use Windows Search or the Cortana® personal - -assistant. - -• Microsoft support Web site: -https://support.microsoft.com - -34 - -User Guide - - What is a CRU? - -Customer replaceable units (CRUs) are parts that can be replaced by the customer. A Lenovo computer may -contain the following types of CRUs: - -Self-service CRU - -Optional-service CRU - -Parts that can be replaced easily by customer -themselves or by trained service technicians at an -additional cost. - -Parts that can be replaced by customers with a -greater skill level. Trained service technicians can -also provide service to install or replace the parts -under the type of warranty designated for the -customer’s machine. - -If you intend to install a CRU, Lenovo will ship the CRU to you. You might be required to return the defective -part that is replaced by the CRU. When return is required: (1) return instructions, a prepaid shipping label, -and a container will be included with the replacement CRU; and (2) you might be charged for the -replacement CRU if Lenovo does not receive the defective CRU within thirty (30) days of your receipt of the -replacement CRU. For full details, see the Lenovo Limited Warranty at -https://www.lenovo.com/warranty/llw_02. - -CRUs for your product model - -The table below lists the CRUs and CRU types that are defined for your product model. - -Part - -Power cord* - -ac power adapter - -Lenovo Slim Pen* - -* for selected models - -Notes: - -Self-service CRU - -Optional-service CRU - -X - -X - -X - -• CRU replacement instruction is provided in one or more of the following publications and is available from - -Lenovo at any time upon your request. - -the product User Guide - -the printed publications that came with the product - -• Replacement of any parts not listed above, including the built-in rechargeable battery, must be done by a -Lenovo-authorized repair facility or technician. Go to https://support.lenovo.com/partnerlocation for more -information. - -Chapter 4. Help and support - -35 - - Call Lenovo - -If you have tried to correct the problem yourself and still need help, you can call Lenovo Customer Support -Center. - -Before you contact Lenovo - -Record product information and problem details before you contact Lenovo. - -Product information - -• Product name - -Problem symptoms and details - -• What is the problem? Is it continuous or intermittent? - -• Machine type and serial number - -• Any error message or error code? - -• What operating system are you using? Which version? - -• Which software applications were running at the time - -of the problem? - -• Can the problem be reproduced? If so, how? - -Note: The product name and serial number can usually be found on the bottom of the computer, either -printed on a label or etched on the cover. - -Lenovo Customer Support Center - -During the warranty period, you can call Lenovo Customer Support Center for help. - -Telephone numbers - -For a list of the Lenovo Support phone numbers for your country or region, go to -https://pcsupport.lenovo.com/supportphonelist. - -Note: Phone numbers are subject to change without notice. If the number for your country or region is not -provided, contact your Lenovo reseller or Lenovo marketing representative. - -Services available during the warranty period - -• Problem determination - Trained personnel are available to assist you with determining if you have a - -hardware problem and deciding what action is necessary to fix the problem. - -• Lenovo hardware repair - If the problem is determined to be caused by Lenovo hardware under warranty, - -trained service personnel are available to provide the applicable level of service. - -• Engineering change management - Occasionally, there might be changes that are required after a product -has been sold. Lenovo or your reseller, if authorized by Lenovo, will make selected Engineering Changes -(ECs) that apply to your hardware available. - -Services not covered - -• Replacement or use of parts not manufactured for or by Lenovo or nonwarranted parts -• Identification of software problem sources -• Configuration of UEFI/BIOS as part of an installation or upgrade -• Changes, modifications, or upgrades to device drivers -• Installation and maintenance of network operating systems (NOS) -• Installation and maintenance of programs - -For the terms and conditions of the Lenovo Limited Warranty that apply to your Lenovo hardware product, -see “Warranty information” in the Safety and Warranty Guide that comes with your computer. - -36 - -User Guide - - Purchase additional services - -During and after the warranty period, you can purchase additional services from Lenovo at -https://pcsupport.lenovo.com/warrantyupgrade. - -Service availability and service name might vary by country or region. - -Chapter 4. Help and support - -37 - - 38 - -User Guide - - Appendix A. Function descriptions for Lenovo hotkeys - -Hotkey functions are denoted by the icons printed on the keys. The following table describes the functions -for all Lenovo hotkeys. The keyboard of your Lenovo computer should only support a subset of the listed -hotkeys. - -Table 10. Hotkey icons and denoted functions - -Hotkey icon - -Function description - -Mutes/Unmutes sound. - -Increases volume. - -Decreases volume. - -Enables/Disables the microphone. - -Increases screen brightness. - -Decreases screen brightness. - -Selects and sets up display devices. - -Enables/Disables airplane mode. (Windows operating systems) - -Displays the Lenovo Smart Key quick launch panel. - -Enables/Disables the touchpad. - -Displays all open apps as thumbnails. (Windows operating systems) - -Opens the Calculator app. (Windows operating systems) - -Opens the Snipping tool. (Windows operating systems) - -Opens the Settings app. (Windows operating systems) - -Locks the screen. (Windows operating systems) - -Opens the clipboard history. (Windows operating systems) - -Puts the computer to sleep. - -Opens the context menu of the current active app. - -Opens an associated voice app, voice service, or Microsoft Cortana. - -Adjusts the keyboard backlight. - -Increases the brightness of the keyboard backlight. - -Decreases the brightness of the keyboard backlight. - -Answers an incoming Microsoft Teams call. - -Rejects an incoming Microsoft Teams call. - -Accesses Lenovo Support resources. - -Starts/Pauses audio or video playback. - -Stops audio or video playback. - -Plays the previous audio or video in the playlist. - -© Copyright Lenovo 2024 - -39 - - Table 10. Hotkey icons and denoted functions (continued) - -Hotkey icon - -Function description - -Plays the next audio or video in the playlist. - -Switches the computer’s active operation mode. - -Blurs the background for videoconferencing. - -Switches the computer’s active audio mode. - -Turns on/off night light mode. (Windows operating systems) - -Note: Hotkey functions may change if the associated app or the Windows operating system is updated. - -Related topics - -“Hotkeys” on page 20 - -40 - -User Guide - - Appendix B. Notices and trademarks - -Notices - -Lenovo may not offer the products, services, or features discussed in this document in all countries. Consult -your local Lenovo representative for information on the products and services currently available in your -area. Any reference to a Lenovo product, program, or service is not intended to state or imply that only that -Lenovo product, program, or service may be used. Any functionally equivalent product, program, or service -that does not infringe any Lenovo intellectual property right may be used instead. However, it is the user's -responsibility to evaluate and verify the operation of any other product, program, or service. - -Lenovo may have patents or pending patent programs covering subject matter described in this document. -The furnishing of this document does not give you any license to these patents. You can send license -inquiries, in writing, to: -Lenovo (United States), Inc. -8001 Development Drive -Morrisville, NC 27560 -U.S.A. -Attention: Lenovo Director of Licensing - -LENOVO PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS -OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, -MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some jurisdictions do not allow -disclaimer of express or implied warranties in certain transactions, therefore, this statement may not apply to -you. - -Changes are made periodically to the information herein; these changes will be incorporated in new editions -of the publication. To provide better service, Lenovo reserves the right to improve and/or modify the -products and software programs described in the manuals included with your computer, and the content of -the manual, at any time without additional notice. - -The software interface and function and hardware configuration described in the manuals included with your -computer might not match exactly the actual configuration of the computer that you purchase. For the -configuration of the product, refer to the related contract (if any) or product packing list, or consult the -distributor for the product sales. Lenovo may use or distribute any of the information you supply in any way it -believes appropriate without incurring any obligation to you. - -The products described in this document are not intended for use in implantation or other life support -applications where malfunction may result in injury or death to persons. The information contained in this -document does not affect or change Lenovo product specifications or warranties. Nothing in this document -shall operate as an express or implied license or indemnity under the intellectual property rights of Lenovo or -third parties. All information contained in this document was obtained in specific environments and is -presented as an illustration. The result obtained in other operating environments may vary. - -Lenovo may use or distribute any of the information you supply in any way it believes appropriate without -incurring any obligation to you. - -Any references in this publication to non-Lenovo Web sites are provided for convenience only and do not in -any manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the -materials for this Lenovo product, and use of those Web sites is at your own risk. - -© Copyright Lenovo 2024 - -41 - - Any performance data contained herein was determined in a controlled environment. Therefore, the result -obtained in other operating environments may vary significantly. Some measurements may have been made -on development-level systems and there is no guarantee that these measurements will be the same on -generally available systems. Furthermore, some measurements may have been estimated through -extrapolation. Actual results may vary. Users of this document should verify the applicable data for their -specific environment. - -This document is copyrighted by Lenovo and is not covered by any open source license, including any Linux -agreement(s) which may accompany software included with this product. Lenovo may update this document -at any time without notice. - -For the latest information or any questions or comments, contact or visit the Lenovo Web site: -https://support.lenovo.com. - -Trademarks - -Lenovo and the Lenovo logo are trademarks of Lenovo. Thunderbolt is a trademark of Intel Corporation or its -subsidiaries. Microsoft, Windows, and Cortana are trademarks of the Microsoft group of companies. SD, -microSD, microSDHC, and microSDXC marks and logos are trademarks of SD-3C LLC. Kensington Nano -Security Slot is a registered trademark of ACCO Brands. DisplayPort is a trademark of the Video Electronics -Standards Association. The terms HDMI and HDMI High-Definition Multimedia Interface, and the HDMI logo -are trademarks or registered trademarks of HDMI Licensing Administrator, Inc. in the United States and other -countries. Wi-Fi is a registered trademark of Wi-Fi Alliance. USB Type-C is a registered trademark of USB -Implementers Forum. All other trademarks are the property of their respective owners. - -42 - -User Guide +Another assumption in this analysis is that personal incomes change at the same rate as GDP per +capita. Empirically, it is not unusual to observe that personal income or consumption increases at +a slower rate than GDP per capita. However, the relationship between the two measures varies +widely across countries and it is not clear what the relative rates of growth might be at the regional +or global level. The World Bank often assumes a “pass-through” rate of 0.85, i.e., only 85% of + +www.pewresearch.org + + 15 + +PEW RESEARCH CENTER + +national output growth is passed through to household income or consumption. If a pass-through +rate of 0.85 is assumed for this analysis the estimated change in the number of people in each +income tier is as follows: Poor (+117 million), low income (+11 million), middle income (-45 +million), upper-middle income (-30 million) and high income (-53 million). + +From the point of view of measuring global poverty, a key issue is that the latest available +benchmark survey for India was conducted in 2011. As the second most populous country in the +world, India accounts for a large share of the global population in poverty. In 2011, 969 million +people lived in extreme poverty globally, on $1.90 or less daily, and 285 million of that number +were from India. The World Bank has applied a variety of estimation techniques to update the +poverty counts for India, but uncertainty lingers in these updates for India and, by implication, in +current measures of global poverty. + +www.pewresearch.org diff --git a/q181/random_k4/random_2.md b/q181/random_k4/random_2.md index 8f865df3772bc41afe1f8a2c8625c0b0956ae175..c346d7f5a27964259c5644ba47c0ac6027670487 100644 --- a/q181/random_k4/random_2.md +++ b/q181/random_k4/random_2.md @@ -1,2302 +1,1018 @@ -LINC: A Neurosymbolic Approach for Logical Reasoning by Combining -| | | Language | | Models | with | First-Order | | Logic | Provers | | | -| ---------------- | -------------------- | -------- | -------- | -------- | -------------------- | ----------------- | --- | ----- | ---------------- | --- | --- | -| TheoX.Olausson*1 | | | | AlexGu*1 | BenjaminLipkin*2 | | | | CedegaoE.Zhang*2 | | | -| | ArmandoSolar-Lezama1 | | | | JoshuaB.Tenenbaum1,2 | | | | RogerLevy2 | | | -| | | | {theoxo, | gua, | lipkinb, | cedzhang}@mit.edu | | | | | | -1MITCSAIL2MITBCS -∗Equalcontribution. -| | | Abstract | | | | 1 Introduction | | | | | | -| ------- | ---------- | -------- | ----------- | --------- | --- | -------------- | -------- | --- | --- | -------------- | ------ | -| | | | | | | Widespread | adoption | | of | large language | models | -| Logical | reasoning, | i.e., | deductively | inferring | | | | | | | | -the truth value of a conclusion from a set of (LLMs)suchasGPT-3(Brownetal.,2020),GPT- -4202 beF 41 ]LC.sc[ 2v46151.0132:viXra -| | | | | | | 4 (OpenAI, | 2023), | | and PaLM | (Chowdhery | et al., | -| --- | --- | --- | --- | --- | --- | ---------- | ------ | --- | -------- | ---------- | ------- | -premises,isanimportanttaskforartificialin- -telligencewithwidepotentialimpactsonsci- 2022)haveledtoaseriesofremarkablesuccesses -ence, mathematics, and society. While many in tasks ranging from text summarization to pro- -prompting-basedstrategieshavebeenproposed gramsynthesis. Someofthesesuccesseshaveen- -toenableLargeLanguageModels(LLMs)to -couragedthehypothesisthatsuchmodelsareable -| dosuchreasoningmoreeffectively, | | | | theystill | | | | | | | | -| ------------------------------- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -toflexiblyandsystematicallyreason(Huangand -| appear | unsatisfactory, | | often failing | in subtle | | | | | | | | -| ------ | --------------- | --- | ------------- | --------- | --- | --- | --- | --- | --- | --- | --- | -and unpredictable ways. In this work, we Chang, 2022), especially when using prompting -strategiesthatexplicitlyencourageverbalizingin- -| investigate | the | validity | of instead | reformulat- | | | | | | | | -| ----------- | --- | -------- | ---------- | ----------- | --- | --- | --- | --- | --- | --- | --- | -ingsuchtasksasmodularneurosymbolicpro- termediate reasoning steps before generating the -LINC: -gramming, which we call Logical In- final answer (Nye et al., 2021; Wei et al., 2022; -ference via Neurosymbolic Computation. In Kojimaetal.,2022;Wangetal.,2023b). However, -LINC,theLLMactsasasemanticparser,trans- -thisreasoningabilityappearstobeunreliablefor -latingpremisesandconclusionsfromnatural -tasksthatrequirereasoningoutofdomain(Liang -| language | to | expressions | in first-order | | logic. | | | | | | | -| -------- | --- | ----------- | -------------- | --- | ------ | ------------- | ------- | --- | ------- | -------------------- | --- | -| | | | | | | et al., 2022; | Saparov | | et al., | 2023), understanding | | -Theseexpressionsarethenoffloadedtoanex- -| | | | | | | negation | (Anil | et al., | 2022), | and following | long | -| --- | --- | --- | --- | --- | --- | -------- | ----- | ------- | ------ | ------------- | ---- | -ternaltheoremprover,whichsymbolicallyper- -formsdeductiveinference. Leveragingthisap- reasoningchains(Dzirietal.,2023). Furthermore, -proach, we observe significant performance whilethestandardapproachof“scalingup”seems -gains on FOLIO and a balanced subset of toimproveperformanceacrosssomereasoningdo- -ProofWriterforthreedifferentmodelsinnearly mains,otherdomains,e.g.,reasoninginvolvinguse -| all experimental | | conditions | we | evaluate. | On | | | | | | | -| ---------------- | --- | ---------- | --- | ------------- | --- | --------- | -------- | ----------- | ---- | -------------------- | ------- | -| | | | | | | of Modus | Tollens, | | show | no such improvements | | -| ProofWriter, | | augmenting | the | comparatively | | | | | | | | -| | | | | | | (McKenzie | et | al., 2022). | | These findings | suggest | -smallopen-sourceStarCoder+(15.5Bparame- -| | | | | | | that such | models | may | be | relying on approximate | | -| --- | --- | --- | --- | --- | --- | --------- | ------ | --- | --- | ---------------------- | --- | -ters)withLINCevenoutperformsGPT-3.5and -GPT-4withChain-of-Thought(CoT)prompt- heuristicsbasedonsurface-levelstatisticalpatterns -ing by an absolute 38% and 10%, respec- in reasoning tasks, rather than consistent, gener- -tively. When used with GPT-4, LINC scores alizablerepresentationsandstrategies(Srivastava -CoT -26% higher than on ProofWriter while etal.,2023;Creswelletal.,2023). -performingcomparativelyonFOLIO.Further -| | | | | | | At the | same | time, | the ability | to accurately | and | -| --- | --- | --- | --- | --- | --- | ------ | ---- | ----- | ----------- | ------------- | --- | -analysisrevealsthatalthoughbothmethodson -soundlyperformlogicalreasoningisimportantfor -averagesucceedroughlyequallyoftenonthis -dataset,theyexhibitdistinctandcomplemen- AIandNLPduetoitsimpactondownstreamtasks. -taryfailuremodes. Wethusprovidepromising For example: retrieval-augmented chatbots may -evidenceforhowlogicalreasoningovernatural becomemoretruthfulifitcanbeverifiedthattheir -languagecanbetackledthroughjointlylever- -answerslogicallyfollowfromtheretrievedfacts; -| agingLLMsalongsidesymbolicprovers. | | | | | All | | | | | | | -| ---------------------------------- | --- | --- | --- | --- | --- | ----------- | ------ | --- | ------- | ---------- | --------- | -| | | | | | | data-driven | models | | capable | of logical | reasoning | -correspondingcodeispubliclyavailable.1 -| | | | | | | may speed | up | progress | across | mathematics | and | -| --- | --- | --- | --- | --- | --- | ------------ | ------- | -------- | --------- | ----------- | ------- | -| | | | | | | the sciences | through | | automated | theorem | proving | -∗ -Authororderrandomized;allreservetherighttolist -andknowledgediscovery;andAItutoringsystems -theirnamefirst. -https://github.com/benlipkin/linc which ensure internal logical consistency might +USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 1 of 27 + +[PUBLISH] + +In the +United States Court of Appeals +For the Eleventh Circuit + +____________________ + +No. 22-13963 + +____________________ + +UNITED STATES OF AMERICA, + +versus + +SHADON DEMETRIC EDWARDS, + + Plaintiff-Appellee, + + Defendant-Appellant. + +____________________ + +Appeal from the United States District Court +for the Southern District of Florida +D.C. Docket No. 0:19-cr-60373-AHS-1 +____________________ + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 2 of 27 + +2 + +Opinion of the Court + +22-13963 + +Before ROSENBAUM, NEWSOM, and MARCUS, Circuit Judges. + +NEWSOM, Circuit Judge: + +Shadon Edwards pleaded guilty to possessing a firearm and +ammunition despite having previously been convicted of a crime +punishable by more than a year in prison, in violation of 18 U.S.C. +§ 922(g). On appeal, Edwards challenges his § 922(g) conviction +and 180-month sentence on three grounds. First, he contends that +his guilty plea wasn’t knowing and voluntary because the district +court failed to inform him that, by pleading, he would waive his +right to appeal the denial of an earlier suppression motion. Second, +he asserts that the district court wrongly enhanced his sentence un- +der the Armed Career Criminal Act because the government didn’t +prove beyond a reasonable doubt, as required by Erlinger v. United +States, 602 U.S. 821 (2024), that he had earlier been convicted of +three crimes that occurred on different occasions. Third, he argues +that § 922(g) exceeds Congress’s Commerce Clause authority. + +After careful review, and with the benefit of oral argument, + +we affirm the district court’s judgment. + +I + +A + +Shadon Edwards was suspected of being a high-level organ- +izer in a drug-trafficking organization responsible for distributing +illegal narcotics throughout south Florida. After obtaining an ar- +rest warrant for Edwards, law-enforcement officials observed him +driving around Pompano Beach. The officers followed and, after + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 3 of 27 + +22-13963 + +Opinion of the Court + +3 + +seeing Edwards park, approached and took him into custody. A +search incident to Edwards’s arrest revealed that he had a Glock 43 +nine-millimeter semi-automatic pistol, loaded with six rounds of +ammunition. + +B + +A grand jury indicted Edwards on a single count: the know- +ing possession of a firearm and ammunition by one who had pre- +viously been convicted of a crime punishable by more than a year +in prison, in violation of 18 U.S.C. § 922(g)(1).1 Edwards filed a +motion to suppress the gun and ammunition found in his car, +which the district court denied. He then pleaded guilty pursuant +to a written plea agreement that neither expressly preserved nor +expressly waived his right to appeal. In connection with his plea, +Edwards also signed a factual proffer describing his offense conduct +and some of his prior criminal history, including, as particularly rel- +evant here, convictions “for trafficking +in oxycodone (10- +022573CF10A), aggravated battery with a deadly weapon (10- +022679CF10A), and robbery with a deadly weapon/aggravated bat- +tery (97-000465CF10B).” Factual Proffer ¶ 2, Dkt. No. 81. + +At the change-of-plea hearing, the district judge engaged in +a colloquy with Edwards to assess several matters, including +whether he was competent to plead guilty and whether he under- +stood the consequences of doing so. At the hearing, there was no + +1 The indictment also mentioned the Armed Career Criminal Act, 18 U.S.C. +§ 924(e)(1), albeit without further elaboration. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 4 of 27 + +4 + +Opinion of the Court + +22-13963 + +discussion—one way or the other—about whether, by pleading, +Edwards would forfeit his right to appeal the earlier denial of his +suppression motion. Edwards pleaded guilty, and the judge found +that his plea was “entered into freely and voluntarily with a know- +ing, intelligent waiver of [his] rights.” Tr. of Change of Plea at +18:20–22, Dkt. No. 105. + +In computing Edwards’s offense level for sentencing pur- +poses, the presentence investigation report applied an enhance- +ment pursuant to U.S.S.G. § 4B1.4(a), which states that “[a] defend- +ant who is subject to an enhanced sentence under the provisions of +18 U.S.C. § 924(e) is an armed career criminal.” Section 924(e), the +Armed Career Criminal Act, in turn prescribes a mandatory mini- +mum 15-year sentence for anyone who violates § 922(g) and “has +three previous convictions” for “violent felon[ies] or . . . serious +drug offense[s], or both, committed on occasions different from +one another.” 18 U.S.C. § 924(e). In particular, Edwards’s PSI +listed the same three predicate convictions as had his factual prof- +fer: + +The defendant is an armed career criminal because he +is subject to an enhanced sentence under the provi- +sions of 18 U.S.C. § 924(e), pursuant to § 4B1.4(a). +The defendant has the prior felony convictions as +noted in docket numbers 10-022573CF110A [i.e., for +trafficking in oxycodone], 10-022679CF10A [i.e., for +aggravated battery with a deadly weapon], and 97- +000465CF10B [i.e., +for robbery with a deadly +weapon/aggravated battery]. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 5 of 27 + +22-13963 + +Opinion of the Court + +5 + +PSI ¶ 20. With the ACCA enhancement, Edwards’s statutory min- +imum term of imprisonment was 15 years, so the PSI increased the +low end of Edwards’s range from 168 months to 180 months in +prison. Edwards didn’t object to the PSI’s recommended sen- +tence.2 + +At the sentencing hearing, the district judge noted that the +PSI’s recommended sentence reflected an ACCA-enhanced 15-year +mandatory minimum, and Edwards agreed to it. The judge there- +fore imposed a sentence of 180 months in prison, followed by a +two-year term of supervised release. Again, Edwards didn’t object. + +During the sentencing hearing, Edwards’s lawyer asked the +judge to appoint separate counsel to appeal the earlier denial of the +motion to suppress the handgun evidence. Seemingly in an effort +to recall the nature of Edwards’s plea, which Edwards had entered +several months earlier, the judge asked the lawyers whether there +was a formal agreement. After Edwards’s counsel explained that +he didn’t “remember there being an appellate waiver,” the judge +suggested that because “the plea agreement . . . does not mention +an appellate waiver,” Edwards had “the right . . . to take an appeal” + +2 Edwards made two factual objections to the PSI—both are unrelated to this +appeal. See Tr. of Sentencing Hr’g at 3:20–25, Dkt. No. 106 (“[Defense Coun- +sel]: There were two very minor—they’re not even—they don’t even affect +his guidelines—two minor objections, I would say, to the PSI, and they’re mi- +nor in and of itself. That’s about it. Other than that, we’re just asking for the +Court to sentence him to 15 years and we all go about our business.”). + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 6 of 27 + +6 + +Opinion of the Court + +22-13963 + +of the suppression ruling. Tr. of Sentencing Hr’g at 24:18–22, Dkt. +No. 106. + +Edwards subsequently filed a notice of appeal indicating his +intent to seek review of the denial of the suppression motion. The +district court granted Edwards’s lawyer’s motion to withdraw and +assigned new counsel, who filed an amended notice of appeal. + +II + +Before us, Edwards challenges his conviction and 180-month +sentence on three grounds: He argues (1) that his guilty plea wasn’t +knowing and voluntary because he didn’t realize that by pleading +he would waive his right to appeal the district court’s denial of his +suppression motion; (2) that the district court erred in enhancing +his sentence under ACCA because the government didn’t prove, as +required by Erlinger v. United States, 602 U.S. 821 (2024), that his +predicate offenses occurred on different occasions; and (3) that +§ 922(g)(1) exceeds Congress’s Commerce Clause power. We will +address each issue in turn. + +A + +First up, the voluntariness of Edwards’s plea. When, as here, +a defendant fails to object during his plea colloquy, we review the +district court’s determination that his plea was knowing and volun- +tary only for plain error. See United States v. Monroe, 353 F.3d 1346, +1349 (11th Cir. 2003) (citing United States v. Vonn, 535 U.S. 55, 59 +(2002)). “To establish plain error, a defendant must show that (1) +there was an error, (2) the error was plain, and (3) the error affects +substantial rights.” United States v. Roosevelt Coats, 8 F.4th 1228, 1235 + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 7 of 27 + +22-13963 + +Opinion of the Court + +7 + +(11th Cir. 2021). An error about the voluntariness of a plea affects +a defendant’s “substantial rights” if he shows a reasonable proba- +bility that the error influenced his “‘decision to plead.’” United +States v. Steiger, 99 F.4th 1316, 1326 (11th Cir. 2024) (en banc) (quot- +ing United States v. Dominguez Benitez, 542 U.S. 74, 84 (2004)). If the +defendant makes the first three showings, we have the “discretion +to recognize an unpreserved error but only if (4) the error seriously +affects the fairness, integrity or public reputation of judicial pro- +ceedings.” Roosevelt Coats, 8 F.4th at 1235 (citation modified). + +Before a district judge accepts a guilty plea, he must “address +the defendant personally in open court and determine that the plea +is voluntary.” Fed. R. Crim. P. 11(b)(2). In particular, the judge +must ensure that the defendant understands certain facts, including +“the terms of any plea-agreement provision waiving the right to +appeal or to collaterally attack the sentence.” Id. 11(b)(1)(N); see +also, e.g., United States v. Lejarde-Rada, 319 F.3d 1288, 1289 (11th Cir. +2003) (noting that the Rule 11 colloquy aims to ensure, among +other things, that the defendant “know[s] and understand[s] the +consequences of his guilty plea” (citation modified)). + +Importantly for present purposes, a knowing and voluntary +guilty plea “waives all non-jurisdictional defects,” United States v. +Pierre, 120 F.3d 1153, 1155 (11th Cir. 1997) (citation modified), in- +cluding, as relevant here, the allegedly erroneous denial of a mo- +tion to suppress evidence. Accordingly, a “defendant who wishes +to preserve appellate review of a non-jurisdictional defect while at +the same time pleading guilty can do so only by entering a + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 8 of 27 + +8 + +Opinion of the Court + +22-13963 + +‘conditional plea’ in accordance with Fed. R. Crim. P. 11(a)(2).” Id. +All here agree that Edwards’s plea was “unconditional.” Br. of Ap- +pellant at 31. + +Edwards insists, though, that his guilty plea wasn’t knowing +and voluntary because he didn’t understand one of the conse- +quences of pleading—namely, that he would lose the right to ap- +peal the denial of his motion to suppress the handgun evidence. +He asserts that the district court failed to inform him of that fact +during the Rule 11 colloquy and, indeed, even suggested at his later +sentencing hearing that he could pursue the appeal. Edwards relies +principally on our decision in Pierre to argue that because he be- +lieved that his plea agreement preserved the right to appeal the de- +nial of the suppression motion—but in fact didn’t—his plea wasn’t +knowing and voluntary. + +In Pierre, we considered the appeal of a defendant who, hav- +ing pleaded guilty, sought to argue that his speedy-trial rights had +been violated. 120 F.3d at 1154. We vacated the conviction upon +determining that, at the time he pleaded guilty, the defendant didn’t +understand the consequences of doing so: “Because Pierre en- +tered—and the district court accepted—this guilty plea only on the +reasonable (but mistaken) belief that [he] had preserved the speedy +trial issues for appeal, his plea was, as a matter of law, not knowing +and voluntary.” Id. at 1156. We based our decision on a review of +“[t]he plea colloquy,” which “unequivocally indicate[d] that Pierre +intended to plead guilty on the condition” that he retained his ap- +pellate rights. Id. at 1155. For example, we cited statements at the + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 9 of 27 + +22-13963 + +Opinion of the Court + +9 + +plea hearing demonstrating that Pierre had pleaded guilty “only af- +ter having been assured by the district court judge that [] he had +preserved the speedy trial issues for appeal.” Id. In particular, the +district judge had said: “So I’m going to deny your renewed motion +to dismiss. Now, if you’re ready to go to trial, we have a jury com- +ing up. If you feel that you wish to plead guilty and preserve these +legal issues, you may do so.” Id. We also cited statements demon- +strating that “Pierre’s counsel [had] explicitly tried to preserve the +speedy trial issue.” Id. Specifically, Pierre’s counsel had stated: “As +long as my client is assured by the court, which you have done, that +these issues are protected for purposes of appeal, [Pierre] may be +willing to wish to do that in the future. But for the purposes of +trial, he is entering a plea. It is in his best interest at this time.” Id. + +Here, Edwards hasn’t pointed to any similar evidence from +the plea hearing—no statement by him, his lawyer, or the district +judge—indicating that he didn’t understand the consequences of +pleading guilty. To be sure, Edwards now says that once he lost +the suppression issue, he wanted to plead and “go onto the sentenc- +ing part” so that he could appeal the denial of the motion. See Br. +of Appellant at 29–30 (quoting Tr. of Change of Plea at 10:4–12). +But at the plea hearing, the suppression issue was mentioned only +once, in passing. The district judge asked Edwards’s counsel how +he “got ready for today’s hearing.” Tr. of Change of Plea at 8:5–6. +Edwards’s lawyer responded: “[I]n anticipation of today’s hearing, +[my client] and I had a scheduled visit last week where we went +over the entire plea agreement and . . . the entire factual proffer + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 10 of 27 + +10 + +Opinion of the Court + +22-13963 + +together.” Id. at 8:18–21. Edwards’s lawyer then added the follow- +ing regarding the suppression motion—its only mention: + +[Defense Counsel]: As the Court noted earlier, we +also had a motion to suppress in this case. The facts +are rather simple. I mean, he was pulled over and he +was arrested and he was in possession of a firearm +and he was a convicted felon. It’s not a—this is not +factually difficult. It was just whether or not they had +requisite legal rights to effectuate that stop. That was +the whole issue. And this Court, unfortunately, ruled +against us, but that’s the nature of the case. + +Id. at 8:22–9:4. + +No one suggested at that time that Edwards wanted to—or +thought he could—appeal the denial of the suppression motion. +And Edwards’s counsel certainly didn’t tie the entry of his plea to +the preservation of the right to appeal, as the lawyer had in Pierre. +To the contrary, Edwards’s lawyer seemed resigned to the district +court’s ruling. + +The preservation issue didn’t resurface for another three +months, at the sentencing hearing. Edwards asserts before us that +a colloquy at that hearing shows that he intended all along to ap- +peal the denial of the motion to suppress. Br. of Appellant at 30– +31. We disagree. + +Here’s the relevant sentencing-hearing exchange, in full: + +Defense Counsel: He did tell me, Judge, he wants me +to try to get counsel appointed to just appeal the + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 11 of 27 + +22-13963 + +Opinion of the Court + +11 + +ruling on the motion to suppress. So I told him that I +would file the notice. And if the Court at some point +needs to have a hearing as to whether he is in forma +pauperis and is entitled to counsel for that appeal, I +guess we’ll have a hearing. But that’s what he re- +quested of me. + +The Court: Sure. On this case did we have a formal +plea agreement? I can’t remember now. + +Defense Counsel: I don’t remember – – + +Government: No, I think he pled open, Your Honor. + +Defense Counsel: Yeah. Yeah, I don’t remember +there being an appellate waiver. I don’t recall. + +The Court: Right. I just wanted to advise him if there +was. So the plea agreement at docket entry 80 does +not mention an appellate waiver. Certainly, sir, you +do have the right, if you want, to take an appeal of +that. Just don’t miss your timelines on that. Okay? + +Mr. Edwards: All right. Thank you, sir. + +Tr. of Sentencing Hr’g at 24:7–25:1. + +Unsurprisingly, Edwards focuses on the district court’s state- +ment that he “ha[d] the right, if [he] want[ed], to take an appeal” +from the order denying the suppression motion. Id. at 24:23–24. +But that remark is both too little and too late. As for the too-little +part, the context makes clear that in the months that had passed +since the plea hearing, everyone (including the district judge) had + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 12 of 27 + +12 + +Opinion of the Court + +22-13963 + +lost track of the details of Edwards’s plea—whether it was a formal +plea agreement, whether it contained an appeal waiver, etc. At the +sentencing hearing, the judge assumed that because the agreement +didn’t contain an appeal waiver, Edwards could appeal the suppres- +sion order. But the judge hardly made any guarantees. + +And as for the too-late: All of the evidence to which Ed- +wards now points comes from the sentencing hearing, which, +again, took place some three months after he pleaded guilty. That +makes this case quite different from Pierre, in which the defendant +could—and did—point to strong indications from the plea hearing +itself that he misunderstood, at the time, what appellate rights he +was waiving. There is simply no basis in the record here for imput- +ing backwards, as it were, whatever was said at sentencing to show +genuine confusion at the earlier plea hearing.3 + +Edwards has failed to show that the district court plainly +erred in accepting his plea as knowing and voluntary.4 We there- +fore affirm the district court’s judgment in that respect. + +3 Edwards separately complains that the district judge failed to explicitly warn +him that he couldn’t appeal the denial of his motion to suppress. But nothing +in Rule 11 or our caselaw requires a judge to proactively inform a defendant +that by entering an unconditional plea he waives his right to appeal all non- +jurisdictional matters. Edwards ultimately concedes as much, saying only that +one circuit court has called this the “preferred practice.” Reply Br. at 3–4 (cit- +ing United States v. Lampazianie, 251 F.3d 519, 526 (5th Cir. 2001)). Perhaps, +but that is no basis for finding plain error. + +4 Unrelatedly, Edwards argues that his plea wasn’t knowingly and voluntarily +entered “on the basis of Erlinger as well.” Letter Br. of Appellant at 5 n.4, Dkt. +No. 37. In a footnote in a supplemental brief, Edwards asserts that had he + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 13 of 27 + +22-13963 + +Opinion of the Court + +13 + +B + 1 -Input -| Step 1: | | | | Premise: | All rectangles have four sides. | | | | | | | Output | -| -------- | --- | --- | --- | ---------- | --------------------------------- | --- | --- | --- | --- | --- | --- | ------ | -| Semantic | | | | Premise: | All four-sided things are shapes. | | | | | | | | -True -| Parser | | | | Conclusion: | Are all rectangles shapes? | | | | | | | | -| ------ | --- | --- | --- | ------------ | -------------------------- | -------- | --- | --- | --- | -------- | --- | --- | -| | | | | | | Sample 1 | | | | Sample 1 | | | - all x. (rectangle(x) -> foursides(x)) - all x. (foursides(x) -> isshape(x)) -| | all x. (rectangle(x) -> isshape(x)) | | | | | | | Step 2: | | True | | | -| --- | -------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | ------- | --- | ---- | --- | --- | -Step 3: -Logic -K-Maj -| | | | | | | Sample 2 | | Theorem | | Sample 2 | | | -| --- | --- | --- | --- | --- | --- | -------- | --- | ------- | --- | -------- | --- | --- | - all x. (rectangle(x) -> foursides(x)) Voting - all x. (foursides(x) -> isshape(x))) Prover Error - all x. (rectangle(x) -> isshape(x)) -| | | | | | | Sample 3 | | | | Sample 3 | | | -| --- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | -------- | --- | --- | - all x. (rectangle(x) -> foursidedthings(x)) -| | all x. (foursidedthings(x) -> isshape(x)) | | | | | | | | | True | | | -| --- | -------------------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | - all x. (rectangle(x) -> isshape(x)) -| | | | | | | Sample N | | | | Sample N | | | -| --- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | -------- | --- | --- | - all x. (rectangle(x) -> foursides(x)) - all x. (foursidedthings(x) -> isshape(x)) Unknown - all x. (rectangle(x) -> isshape(x)) -Figure 1: This figure showcases the essence of our approach. Starting from a problem in natural language, in -Step1,theLLMsemanticparsersampleslogicformulasexpressingestimatesofthesemantics. Itispossiblethat -someofthesemightcontainerrors,e.g.,thesecondexampleshowsasyntaxerrorinvolvinganextraparenthesis, -whereasthefourthexamplehighlightsasemanticerrorcausedbymismatchedpredicates. InStep2,thesearethen -eachoffloadedtoanautomatedtheoremprover,filteringoutsyntaxerrors,andproducinglabelsfortheremaining -samples. InStep3,theremainingcandidateoutputsarepassedthroughamajority-votesievetoarriveatthebest -estimateforasingleoutputlabel. -make for better educational platforms, teaching thesolverastray,leadingtoanincorrectconclusion. -studentstothinkmoreclearlyandrigorously. The Asitisnotclearwhetherthetaskofformalization -questionofhowtoenablestate-of-the-artLLMsto ismoreorlessdifficultthanthatofend-to-endnat- -becomemorereliablelogicalreasonersisthusone ural language reasoning, our core interest in this -ofgreatimportance,withfar-reachingimplications. work is to compare and contrast our neurosym- -bolicapproachtoexistingreasoningstrategieslike -| In | this | work, | we analyze | LINC: | Logical | | | | | | | | -| --- | ---- | ----- | ---------- | ----- | ------- | --- | ----------------- | --- | --- | ----------------- | --- | -------- | -| | | | | | | | Chain-of-Thought. | | | Our contributions | | are thus | -InferenceviaNeurosymbolicComputation(Fig.1). -| LINC, | | | | | | | three-fold: | | | | | | -| ----- | ------- | --------- | --- | ---------- | ------- | --- | ----------- | --- | --- | --- | --- | --- | -| In | logical | reasoning | | is tackled | through | a | | | | | | | -modular, two-step neurosymbolic process. First, • First,weproposeLINC,atwo-stageneurosym- -thelanguagemodelconvertsthenaturallanguage bolic approach for logical reasoning tasks -| premises | and | desired | conclusion | into | first-order | | (Sec.2). | | | | | | -| ------------------------------------------- | ------ | --------------- | ---------- | ---------- | ----------- | --- | --------- | ------------ | ---------- | --------- | ------ | -------------- | -| logic(FOL)expressions(Enderton,2001;Barker- | | | | | | | | | | LINC | | | -| | | | | | | | • Second, | | we compare | | to | three baseline | -| Plummer | et | al., 2011). | Second, | a symbolic | | FOL | | | | | | | -| | | | | | | | LLM | strategies | | (Fig. 2), | across | three mod- | -| theorem | prover | algorithmically | | determines | | the | | | | | | | -| | | | | | | | els | (StarCoder+, | | GPT-3.5, | GPT-4) | and two | -truthvalueoftheconclusiongiventheformalized -| | | | | | | | datasets | | (FOLIO | and ProofWriter) | | (Sec. 4). | -| --------- | ---------------------------------- | --- | --- | --- | --- | --- | -------- | --- | ------ | ---------------- | --- | --------- | -| premises. | Inpractice,wealsoincorporateathird | | | | | | | | | | | | -WefindthatLINCsignificantlyimprovesper- -| majority | voting | step, | which | is shown | to improve | | | | | | | | -| -------- | ------ | ----- | ----- | -------- | ---------- | --- | --- | --- | --- | --- | --- | --- | -formanceovereverybaselineinallexperimen- -LINCisanaturalextensionofrecent -performance. -talconditionsexceptforGPT-4onFOLIO. -workaugmentinglanugagemodelswithsymbolic -tools such as calculators or interpreters (Schick • Third,weprovideathorougherroranalysisof -| | | | | | | | | LINC | Chain-of-Thought, | | | | -| ------------ | --- | --- | --- | --- | --- | --- | ---- | ---- | ----------------- | --- | --- | --------- | -| etal.,2023). | | | | | | | both | | and | | | identify- | -LINChasakeyadvantage: ingthreehigh-levelfailuremodesofeach. We -thelanguagemodel -discoverthatthesefailuremodesaredistinct, -itselfnolongerneedstoperformanydeductiverea- -highlightingthepotentialforasynergyofthe -| soning, | which | is offloaded | | to the theorem | prover. | | | | | | | | -| ------- | ----- | ------------ | --- | -------------- | ------- | --- | --- | --- | --- | --- | --- | --- | -twomethods(Sec.5). -| However, | therearealsocleardrawbacks: | | | | thefor- | | | | | | | | -| -------- | --------------------------- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | -malization from natural language to first-order Overall, we present strong evidence for the po- -logic must perfectly capture all relevant informa- tentialoffutureneurosymboliclogicalreasoning -tioncontainedinthepremises,andanylossofin- systemsbasedonintegratinglanguagemodelsand -| formationintheformalizationproceduremaylead | | | | | | | theoremprovers. | | | | | | -| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | --- | --- | - -| 2 LINC:LogicalInferencevia | | | | | | 3 Experiments | | | | | | | -| -------------------------- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | -NeurosymbolicComputation -Inthissection,wepresentourexperimentalsetup, -themodelsweuse,andthethreebaselinestowhich -| Ourneurosymbolicapproachtoend-to-endlogical | | | | | | wecompareLINC. | | | | | | | -| ------------------------------------------- | --- | --- | --- | ---------------- | --- | -------------- | --- | ----------------------------- | --- | --- | --- | --- | -| reasoningconsistsoftwostages. | | | | Inthefirststage, | | | | | | | | | -| | | | | | | Datasets: | | Ourexperimentsusetasksfromtwo | | | | | -theLLMactsasasemanticparser,translatingNL -| | | | | | | existing | datasets: | FOLIO | (Han | et | al., 2022) | and | -| --- | --- | --- | --- | --- | --- | -------- | --------- | ----- | ---- | --- | ---------- | --- | -statementsintoFOLexpressionsinoursupported -ProofWriter(Tafjordetal.,2021),bothofwhich -| logiclanguage. | Inthesecondstage,theseexpres- | | | | | | | | | | | | -| -------------- | ----------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -havebeenshowntobechallengingforoff-the-shelf -| sions are | parsed | from | the text | generated | by the | | | | | | | | -| --------- | -------- | ------ | -------- | ------------ | ------ | --------------------------------------- | ------------------ | --- | ------------ | --- | --------- | --- | -| | | | | | | LLMs(Hanetal.,2022;Creswelletal.,2023). | | | | | | FO- | -| LLM and | then get | passed | to | an automated | theo- | | | | | | | | -| | | | | | | LIO is | an expert-written, | | open-domain, | | logically | | -remprover;weuseProver9,ahigh-performance -complexanddiversedatasetfornaturallanguage -| prover widely | used | in | the logic | community | (Mc- | | | | | | | | -| ----------------- | ---- | --- | --------- | --------- | -------- | --------- | ------- | --------------- | ------ | -------- | --- | -------- | -| | | | | | | reasoning | with | first-order | logic. | We | use | its val- | -| Cune, 2005–2010). | | The | external | solver | then ex- | | | | | | | | -| | | | | | | idation | set for | our evaluation. | | However, | | of the | -ecutesasymbolicdeductionalgorithm,whichei- -204samplesinthevalidationset,wediscoverthat -| ther returns | a value | from | the | set {True, | False, | | | | | | | | -| ------------ | ------- | ---- | --- | ---------- | ------ | --- | --- | --- | --- | --- | --- | --- | -22haveerrors(detailsinAppendixC),leavingus -Uncertain}orraisesanexceptionduetoimproper -| | | | | | | with182examplesforourevaluation. | | | | | ProofWriter, | | -| ---------- | ------ | ------ | ----- | ----- | ---------- | -------------------------------- | --- | --- | --- | --- | ------------ | --- | -| FOL syntax | (e.g., | if the | model | fails | to balance | | | | | | | | -meanwhile,isasyntheticallygenerateddatasetfor -paranthesesintheformulae). -| | | | | | | logical | reasoning | over | natural | language. | | For our | -| --- | --- | --- | --- | --- | --- | ------- | --------- | ---- | ------- | --------- | --- | ------- | -Atitscore,thestrengthofthisapproachliesin evaluation,weusetheOWA(Open-WorldAssump- -the reformulation of the problem space. End-to- tion)portionofProofWriter,sincethissettingbest -endNL-basedreasoningallowsforoperationover -| | | | | | | matches | that | of FOLIO. | Since | we | are running | a | -| -------- | -------- | ---------- | ------ | --- | ---------- | ------- | ---- | --------- | ----- | --- | ----------- | --- | -| a highly | flexible | expression | space, | but | leaves the | | | | | | | | -largenumberofexperiments,werandomlyselect -LLMwiththedifficulttaskofperformingexplicit 360 data points to evaluate on in order to reduce -deductiveinferenceoverexpressionsinthisspace. costs. Wesampletheseinsuchawaythatthere- -| Using LINC, | we | instead | trade | off the | flexible ex- | | | | | | | | -| ----------- | --- | ------- | ----- | ------- | ------------ | --- | --- | --- | --- | --- | --- | --- | -sultingdatasetisbalancedacrossboththenumber -pressionspaceofNLforsyntacticallystrictlogic ofreasoningstepsintheshortestgroundtruthproof -formulas, allowing us to leverage symbolic algo- (depth0-5;50sampleseach)andacrossthethree -rithmswithprovableguaranteesthatthedeductive labels(True/False/Uncertain;120sampleseach; -chainswillbecorrectwithrespecttothesemantics -20eachperdepth). -oftheintermediaterepresentation. Makingeffec- In-contextlearningexamples: Wehand-pick -tiveuseofthisreformulationthusrequiresthelogic eight diverse samples from the FOLIO training -expressionsgeneratedbytheLLMtobe1)syntac- -| | | | | | | set to be | used | as few-shot | | in-context | examples. | | -| -------------- | ---- | ---- | -------- | -------- | ------ | --------- | ---- | ----------- | --- | ---------- | --------- | --- | -| tically valid, | such | that | they are | accepted | by the | | | | | | | | -BecauseProofWriterdoesnotcomewithground -prover, and 2) semantically valid, such that their truthFOLstatements,weusetheseeightsamples -| evaluationresultsinthecorrectconclusion. | | | | | Inour | | | | | | | | -| ---------------------------------------- | --- | --- | --- | --- | ----- | -------- | ------------ | --- | -------- | --- | ------ | ----- | -| | | | | | | for both | evaluations. | | Compared | to | FOLIO, | ques- | -experiments,wemitigatetheserisksbyusingaK- -tionsinProofWritergenerallyhavemorepremises -waymajorityvotingprocedure,whichisdiscussed per question (in our validation sets: an average -furtherinSec.3. -| | | | | | | of 5.3 in | FOLIO | vs. | 18.8 in | ProofWriter). | | Thus, | -| --- | --- | --- | --- | --- | --- | --------- | ----- | --- | ------- | ------------- | --- | ----- | -The significance of this problem space refor- ourevaluationonFOLIOisanin-distributiontask, -whereasProofWriterrequiresgeneralizingout-of- -mulationcan—beyondthenumericalincreasesin -performance observed across our experiments— distributiontoreasoningoverconsiderablylarger -perhapsbestbeseenthroughanin-depthcompari- setsofpremisesthanaregivenintheprompt. -sonofhowLINCandtraditionalend-to-endLLM -| | | | | | | Majority | | voting: | K-way | majority | voting, | in | -| ----------------------------------------- | --- | --- | --- | --- | --- | -------- | --------------------- | ------- | ----- | ------------ | ------- | --- | -| reasoningapproachessuchasChain-of-Thought | | | | | | whichK | | | | | | | -| | | | | | | | samplesaretakeni.i.d. | | | fromthemodel | | | -(CoT)fail. Toforeshadowourlatteranalysis,we and the mode is used as the final prediction, has -CoT, LINC -find that compared to has worse recall previouslybeenshowntoimprovetheperformance -butbetterprecisiononTrue/Falsepredictions. -| | | | | | We | ofprompting-basedstrategiesinlogicalreasoning | | | | | | | -| --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | -discussthisfurtherinSec.5andhighlightthatthis tasks (Wang et al., 2023b). We implement such -suggeststhatLINC,aswellasneurosymboliccom- a strategy in our work, with reported accuracies -putationmoregenerally,hasthepotentialtoreduce reflectingK=10-waymajorityvoting,unlessoth- -LLMoverconfidenceandhallucination. erwise stated. In the case of ties between two la- - -All rectangles have four sides. -All four-sided things are shapes. -Are all rectangles shapes? -* N ICL -Examples -⧺ -< A < E N / V S E A W V L E A U R L A : U T A E T T > r E u > e Naïve < N F N F N F A < E L O L O L O N / V : L : L : L S E A : : : W V L E A U R L A A a A a A a : U T l l l l r l A E l l l l e l T T > r E r x f x a x u > e . o . l . e c u l t ( r ( ( a r f r - r n e o e e s g c u c c i l t r t t d e a s a a e s n i n n d g d g g h l e l l t a e s e e h v ( ( ( s i e x x x n ) ) ) g s f s h o a - - - u a p > > > r r e e s f i i s ? o s s S i s u s s c d h r h h r e a s a a s p i p p a . e d e e t s e ( ( c . s x x h ( ) ) x ) ) p ) a ) d < S s a m A < E i i r u N / V n d e s S E A c e t W V L e s s E A U , h b R L A a a e : U T l a p A E l n e s T T > d s h r E r , a u > e a p e c l t e t l h s a e , n f n g o s l u a o e r l - s l T s r i h r u d a e e e v c ! d e t a t f n h o g i u l n r e C g s o s T < N F N F N F < E L O L O L O / V : L : L : L E A : : : V L A U L A A a A a A a U T l l l l r l A E l l l l e l T > E r x f x a x > e . o . l . c u l t ( r ( ( a r f r - r n e o e e s g c u c c i l t r t t d e a s a a e s n i n n d g d g g h l e l l t a e s e e h v ( ( ( s i e x x x n ) ) ) g s f s h o a - - - u a p > > > r r e e s f i i s ? o s s i s u s s d h r h h e a s a a s p i p p . e d e e s e ( ( L . s x x I ( ) ) N x ) ) C ) ) -⧺ -All dogs are mammals. ⧺ = string concatenation -Harry is a dog. -Is Harry a mammal? - -… -Figure 2: This figure outlines the string concatenation workflow for each of our conditions. We start with the -originalproblem,provideICLexamplesthroughanintermediatemarkuplanguage,andfinallyappendtheproblem -toevaluate. Atthisstage,weallowthemodeltoautoregressivelysampleuntilproducingastoptoken. -bels, we arbitrarily select the first of the two to IntheScratchpadbaseline(Nyeetal.,2021),the -have been generated. We report the effect of K model is asked to first generate FOL expressions -onperformanceacrossourconditionsandbriefly correspondingtothepremises,andthengenerate -discusstrendsinAppendixH. thelabel. ThisbaselineisthusanablationofLINC, -Models: We use three models pre-trained on -whereweusetheLLMinsteadofProver9asthe -bothnaturallanguageandcode: GPT-3.5(Ouyang logicsolver. Finally,intheCoTbaseline,weusethe -et al., 2022), GPT-42 (OpenAI, 2023), and Star- standard technique of CoT prompting (Wei et al., -Coder+3 (Lietal.,2023)withadecodingtemper- 2022; Kojima et al., 2022; Wang et al., 2023b), -ature of T = 0.8 for all experiments. We defer wherethemodelisaskedtogeneratestep-by-step -model,hyperparameter,andhardwaredetailstoAp- naturallanguagereasoningtoarriveattheconclu- -pendixB.WeoptforStarCoder+forthreereasons: sion. The prompts we use for all approaches can -firstly,unliketheothermodelsweconsider,itisa befoundinAppendixD. -free,open-accessmodel. Secondly,ithasadataset -4 Results&Discussion -search functionality4, with which we verify that -FOLIOandProofWriterarenotinStarCoder+’s -OurmainresultsareshowninFigure3. Eachbar -trainingset,givingfurtherassuranceinthevalidity representseitherLINCoroneofthethreebaselines, -ofourfindings. Thirdly,withits15.5Bparameters while each group of bars indicates the language -itislikelyconsiderablysmallerthanGPT-3.5and modelused(in{StarCoder+,GPT-3.5,GPT-4}). -GPT-45, allowing us to compare performance at WenotefirstthatintheFOLIOdomain(Figure -differentmodelscales. 3a), StarCoder+—the smallest model we experi- -Controlled baselines: We compare LINC to ment with—benefits the most from LINC, achiev- -threebaselines,whichwecallNaïve,Scratchpad, -ingameanaccuracythatis14.2pointshigherthan -and Chain-of-Thought (CoT), as illustrated in the closest controlled baseline (56.0% vs. 41.8% -Fig. 2. In the Naïve baseline, the model is given with CoT). We find that Scratchpad, where the -the natural language premises and is asked to di- intermediate logical formulae are still generated -rectlygeneratethelabel(True/False/Uncertain). -but the call to the symbolic solver is ablated and -replacedbythemodel’sownprediction,doesnot -2Weusegpt-3.5-turbo-16k-0613andgpt-4-0613. -appeartobenefitperformance;forStarCoder+,nei- -3https://huggingface.co/bigcode/starcoderplus -ther Scratchpad nor the Naïve baseline perform -4See https://huggingface.co/spaces/bigcode/ -in-the-stack and https://huggingface.co/spaces/ betterthansimplydeterministicallypredictingthe -bigcode/search. mostcommonlabel(“Uncertain”). ForGPT-3.5, -5Although the exact size of these models has not been thetrendissimilar,althoughthegapbetweenLINC -madepublic,theircommonpredecessorGPT-3wasknownto -have175Bparameters;seeBrownetal.(2020). andtheclosestbaselineshrinks(62.6%vs. 54.9% - -| 100% | | | | | | 100% | | | | -| ---- | ---------- | --- | ---------------- | ---------- | ----- | ---- | ---------- | ---------------- | ----------- | -| | Naive | | Chain-of-Thought | | | | Naive | Chain-of-Thought | 96.4% 98.3% | -| | Scratchpad | | LINC (ours) | | | | Scratchpad | LINC (ours) | | -| 80% | | | | 75.3% | | 80% | | 82.5% | | -| | | | | 69.8%68.7% | 72.5% | | | | 72.2% | -62.6% -| 60% | | 56.0% | 54.9% | | | 60% | | | 53.1% 55.8% | -| --- | --- | ----- | ----- | --- | --- | --- | --- | --- | ----------- | -48.4%47.8% -| | | 41.8% | | | | | | 43.6% | | -| --- | ---------- | ----- | --------- | ----- | --- | --- | --------------- | --------------- | ----- | -| | | | | | | | 37.8%38.1%38.6% | 36.4% | | -| 40% | 34.6%32.4% | | | | | 40% | | 33.1% | | -| 20% | | | | | | 20% | | | | -| | StarCoder+ | | GPT-3.5 | GPT-4 | | | StarCoder+ | GPT-3.5 | GPT-4 | -| | | | (a)FOLIO. | | | | | (b)ProofWriter. | | -Figure3: ResultsofeachmodelontheFOLIOandProofWriterdatasets. Accuraciesareforbootstrapped10-way -majorityvoteforallmodels. Errorbarsare±1bootstrappedstandarddeviation. Dotted,blacklineistheaccuracy -obtainedbyalwaysguessingthemostcommonlabelinthedataset. -averageaccuracies). ForGPT-4,thetrendreverses: tionofthenecessaryproofdepthinFigure4. We -LINC underperforms CoT. However, we perform notefirstthatStarCoder+’sperformanceremains -a McNemar’s test (McNemar, 1947) to get the p- flatandclosetochancewithallthreebaselinemeth- -| | | | LINC | CoT | | | Meanwhile,withLINCtheperfor- | | | -| ----- | --- | ---------- | ---- | ----------- | --- | -------------- | ---------------------------- | --- | --- | -| value | on | this GPT-4 | vs. | comparison, | | ods(Figure4a). | | | | -and we find that the difference is not significant manceremainsfarabovechance,althoughitdrops -(p = 0.58). Meanwhile,forourbalancedsubsetof somewhatasnecessaryproofdepthincreases;this -ProofWriter,weseesignificantperformancegains performancedropsuggeststhatStarCoder+strug- -across the board (Figure 3b); particularly so for glessomewhatwiththeNL-to-FOLtranslationtask -GPT-3.5andGPT-4,whichachievemeanaccura- as the problem at hand gets larger. For GPT-3.5, -ciesof96.4%and98.3%whenpairedwithLINC. allbaselinesperformabovechanceatproofdepth -In light of the high accuracies obtained with 0 (i.e., where the conclusion can immediately be -LINC on ProofWriter, we offer two plausible rea- reachedfromthepremises),butthenquicklydrop -sons why LINC is particularly favorable on this backdown(Figure4b). WhileChain-of-Thought -dataset. Firstly,ProofWriteris—unlikeFOLIO— prompting allows the model to complete some -completelysyntheticallygenerated,withrelatively depth-1 tasks, even this strategy then performs -shortsentences,perhapslendingitselfparticularly equivalentlytochance(within1standarddeviation) -LINC, -well to being formalized in FOL. However, it is for higher depths. When augmented with -Scratchpad -noteworthy that the mode does not however, GPT-3.5 is able to achieve near-perfect -seemtoimproveperformanceovertheNaïvebase- performanceacrossallproofdepths,providingevi- -line, indicating that even if the NL-to-FOL task denceforthescalabilityofthisapproachtolonger -were particularly easy in this domain, this is not deductive chains. Finally, for GPT-4 we observe -somethingthatthemodelisitselfcapableoflever- muchstrongerperformancefromthebaselines;in -agingtoimproveitspredictions. Thesecondreason particular,CoTperformsabovechanceforallproof -mightbethatthebaselinestrategiesstruggleinthis depths,andallbaselinesperformwellforshallow -out-of-distributionsetting,inwhichthemodelmust proofs(Figure4c). However,evenwithGPT-4the -generalizetoalargersetofpremises(withpoten- performance drops as the necessary proof depth -tiallylongerdeductivechains)thanthosefoundin increaseswitheveryconfigurationexceptforLINC, -theprompt. Thisdistributionshiftmakesitharder whichperformsnearoratceilingthroughthemax- -imumproofdepthavailableinthedataset. -forthemodeltoignoreirrelevantpremisesinthe -| question | and | carry | out all deductive | chains | cor- | | | | | -| -------- | --- | ----- | ----------------- | ------ | ---- | --- | --- | --- | --- | -Meanwhile,withLINC,thesymbolicsolver -| rectly. | | | | | | 5 ErrorAnalysis | | | | -| -------- | ------- | ---------- | -------- | --- | -------- | --------------- | --- | --- | --- | -| robustly | handles | irrelevant | premises | and | long de- | | | | | -ductivechains,sincetheLLMonlyneedstotrans- HavingestablishedthatLINCcanimproveperfor- -lateeachsentenceintoFOL. mance in many settings, we now move on to our -Totestthislastexplanationfurther,weploteach finalresearchquestion: Howdothefailuremodes -model’sperformanceacrossProofWriterasafunc- of LINCcomparetothoseofin-contextreasoning - -| 1.0 | | | | 1.0 | | | | | 1.0 | | | | | -| -------- | --- | --- | --- | -------- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | -| 0.8 | | | | 0.8 | | | | | 0.8 | | | | | -| ycaruccA | | | | ycaruccA | | | | | ycaruccA | | | | | -| 0.6 | | | | 0.6 | | | | | 0.6 | | | | | -| 0.4 | | | | 0.4 | | | | | 0.4 | | | | | -| 0.2 | | | | 0.2 | | | | | 0.2 | | | | | -Chance Scratchpad LINC (ours) Chance Scratchpad LINC (ours) Chance Scratchpad LINC (ours) -Naive Chain-of-Thought Naive Chain-of-Thought Naive Chain-of-Thought -| 0.0 | 0 1 | 2 3 4 | 5 | 0.0 | 0 1 | 2 3 | 4 | 5 | 0.0 0 | 1 | 2 | 3 | 4 5 | -| --- | -------------- | ----------- | --- | --- | ----------- | --- | --- | --- | ----- | --------- | ----------- | --- | --- | -| | | Proof Depth | | | Proof Depth | | | | | | Proof Depth | | | -| | (a)StarCoder+. | | | | (b)GPT-3.5. | | | | | (c)GPT-4. | | | | -Figure4: AccuracypernecessaryproofdepthinProofWriter. Accuraciesreportedareforbootstrapped10-way -majorityvote,andshadedareascover±1bootstrappedstandarddeviation. Black,dottedlinesreflecttheexpected -successrateofguessingarandomlabel,whichis1/3inallsubsetsperourexperimentdesign. -WefocusoncomparingGPT-4+CoTvs. onesymbolNaziGermanPolitician,causingthe -methods? -GPT-4+LINConFOLIO,sincetheiroverallperfor- informationthathewasindependentlyNazi,Ger- -mance is very similar (75.3% vs. 72.5% average man, or a politician to be lost. As a result, LINC -Uncertain -accuracy). We leave an analysis of StarCoder+’s predicted instead of the ground truth -| predictionsonFOLIOtoAppendixG. | | | | | | labelTrue. | | | | | | | | -| ------------------------------ | --- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | -5.1 QualitativeAnalysis Premise: Heinrich Scmidt was a Nazi German -politician. -Qualitatively,wefindthatLINCandCoThavecom- -| | | | | | | FOL:NaziGermanPolitician | | | | (HeinrichSchmidt) | | | | -| ------- | --------- | -------------- | --- | ------- | ------- | -------------------------------------- | --- | --- | --- | ----------------- | --- | ------- | --- | -| pletely | different | failure modes. | | We give | a high- | | | | | | | | | -| | | | | | | Conclusion(Prover9:Uncertain):Heinrich | | | | | | Schmidt | was | -leveloverviewandabbreviatedexamplesofeach -German. -failuremodehere,leavingfulldetailedexamples -| toAppendixE. | | | | | | FOL:German(HeinrichSchmidt) | | | | | | | | -| ------------ | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | -First,wedetailthefailuremodesforLINC: -| | | | | | | L3: | FOL | contains | syntax | | errors. | Across | all | -| --- | --- | ---------------- | -------- | --- | -------- | --- | --- | -------- | ------ | --- | ------- | ------ | --- | -| L1: | FOL | fails to capture | implicit | | informa- | | | | | | | | | -tionnotmentionedinthepremises. Often,there generations, we find that the FOL expressions -| | | | | | | sometimes | | contain | syntax | errors: | 38% | for | Star- | -| --- | --- | --- | --- | --- | --- | --------- | --- | ------- | ------ | ------- | --- | --- | ----- | -isobviousinformationnotexplicitlylistedinthe -CoderPlus,24%forGPT-3.5,and13%forGPT-4. -premisesthatisnecessarytoexplicitlyencodein -FOLinordertosuccessfullymakeadesireddeduc- The most common error is that the same symbol -| | | | | | | is used | with | multiple | arities. | | As an | example, | if | -| --- | --- | --- | --- | --- | --- | ------- | ---- | -------- | -------- | --- | ----- | -------- | --- | -tion. Forexample,inthesnippetbelowonemust -| | | | | | | Summer(July4)andLoves(Alex, | | | | | Summer)were | | | -| --- | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --- | --- | ----------- | --- | --- | -encodeinFOLtheimplicitassumptionthatHarry -Summer -isaperson(Person(Harry)). both present in a FOL translation, would -haveamultiplearityviolation. -| P re m | ise 1 : W | h e n a person reads | a book, | that | person | | | | | | | | | -| ------ | --------- | -------------------- | ------- | ---- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | -Next,wedetailthreefailuremodesforCoT: -| ga i ns | k n o w | l e dg e. | | | | | | | | | | | | -| ------- | ------- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -C1: CoTconcludessomethingdifferentthan -| FOL:all | x. | all y. (Person(x) | | & Reads(x, | y) & | | | | | | | | | -| ------- | --- | ----------------- | --- | ---------- | ---- | --- | --- | --- | --- | --- | --- | --- | --- | -Book(y) -> Gains(x, Knowledge)) it suggests. One common mistake we often see -withtheCoTapproachisthatitgeneratesachainof -| Premise2: | | Harry read the | book "Walden" | | by Henry | | | | | | | | | -| --------- | --- | -------------- | ------------- | --- | -------- | --------- | --- | ---------- | ------------ | --- | --- | ---- | ------- | -| Thoreau. | | | | | | reasoning | | indicating | uncertainty, | | but | then | asserts | -FOL:Reads(Harry, Walden) thattheconclusionisfalse. Anexampleofthisis -Conclusion (Prover9: Uncertain): Harry gains shownbelow,whereGPT-4suggeststhereis“no -| knowledge. | | | | | | evidence” | | for the | conclusion | | but then | deduces | it | -| ---------------- | --- | ---------- | --- | --- | --- | --------- | --- | ------- | ---------- | --- | -------- | ------- | --- | -| FOL:Gains(Harry, | | Knowledge) | | | | false. | | | | | | | | -L2: FOLfailstocaptureinformationexplic- -(GPT-4IncorrectCoT):Let’sthinkstepbystep.Wewant -itlymentionedinthepremisesduetothechoice to evaluate if platypus are reptiles. We know from the -premisesthatplatypusaremammalsandthattheyhaveno -| ofrepresentation. | | Evenwheninformationisex- | | | | | | | | | | | | -| ----------------- | --- | ------------------------ | --- | --- | --- | ------ | --- | --------- | ---- | ------- | ---- | ----- | -------- | -| | | | | | | teeth. | The | fact that | some | mammals | have | teeth | does not | -plicitlywritteninthepremises,thechoiceofhow -contradictthefactthatplatypus,althoughmammals,donot -theNLisrepresentedinFOLcanleadtolostinfor- haveteeth.Moreover,thestatementabouthumanshaving -teethisunrelatedtotheclassificationofplatypus.There’s -| mation. | Intheexamplebelow, | | thefactthatHein- | | | | | | | | | | | -| ------- | ------------------ | --- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -noevidenceinthepremisesthatwouldsuggestplatypusare -richwasaNaziGermanpoliticianiscapturedby -reptiles.Therefore,thestatementisfalse. - -T -ru -e -F a -lse -U -n ce -rta -in -E -rro -r -Tr u -e -Fals -e -ert ai -n -Err -or -nc -U -Reference Label -lebaL -detciderP -100 T -ru -39 4 1 0 e -80 -F a -0 31 0 0 -lse -60 U -n ce -rta -23 18 62 0 40 in -E -20 rro -1 1 2 0 r -0 -Tr u -e -Fals -e -ert ai -n -Err -or -nc -U -Reference Label -(a)ConfusionmatrixforLINC. -lebaL -detciderP -100 -50 1 7 0 -80 -3 38 9 0 -60 -10 15 49 0 40 +We next consider the Armed Career Criminal Act enhance- +ment. Because Edwards never objected to the enhancement, we +review only for plain error. See United States v. Jones, 743 F.3d 826, +828 (11th Cir. 2014); Fed. R. Crim. P. 52(b). While a conviction un- +der § 922(g) ordinarily yields a maximum sentence of 10 years in +prison, 18 U.S.C. § 924(a)(2), ACCA mandates a minimum 15-year +sentence for an offender who has three prior convictions for “vio- +lent felon[ies]” or “serious drug offense[s]” that were “committed +on occasions different from one another,” id. § 924(e)(1). Edwards +contends that the district court erred in enhancing his sentence un- +der ACCA because the government failed to prove beyond a rea- +sonable doubt that his three predicate convictions occurred on dif- +ferent occasions. + +In Wooden v. United States, the Supreme Court explained that +ACCA “contains both a three-offense requirement and a three-occa- +sion requirement.” 595 U.S. 360, 368 (2022). Although ACCA + +“known that he was entitled to a jury determination as to whether his offenses +were committed on different occasions, he may not have pled guilty—or he +may have pled guilty to the 18 U.S.C. § 922(g)(1) charge, and sought a trial on +the occasions issue.” Id. We’ll address the merits of Edwards’s Erlinger argu- +ment in the next Part, but we decline to consider it vis-à-vis the voluntariness +of his plea. “We have long held that an appellant abandons a claim when he +either makes only passing references to it or raises it in a perfunctory manner +without supporting arguments and authority.” Sapuppo v. Allstate Floridian Ins. +Co., 739 F.3d 678, 680–81 (11th Cir. 2014). + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 14 of 27 + +14 + +Opinion of the Court + +22-13963 + +initially focused only on the number of offenses, Congress later +amended it to specify that qualifying offenses must also occur on +different occasions. See Anti-Drug Abuse Act of 1988, Pub. L. 100– +690, § 7056 (codified as amended at 18 U.S.C. § 924(e)(1)) (inserting +“committed on occasions different from one another,” after “or +both”). Congress “added the occasions clause only after a court +applied ACCA to an offender . . . convicted of multiple counts of +robbery arising from a single criminal episode.” Wooden, 595 U.S.at +371–73 (citing Petty v. United States, 481 U.S. 1034, 1034–1035 +(1987)). The Wooden Court emphasized that while each ACCA- +qualifying offense often occurs separately, “multiple crimes may +occur on one occasion even if not at the same moment.” Id. at 366. +“[T]reating each temporally distinct offense as its own occasion” as +a per se matter, the Court explained, would improperly “collaps[e] +two separate statutory conditions.” Id. at 368. + +To operationalize the different-occasions requirement, the +Wooden Court described a “range of circumstances [that] may be +relevant to identifying episodes of criminal activity”—e.g., the of- +fenses’ timing, location, and character. Id. at 369. Applying those +considerations—which we’ll explore in detail soon enough—the +Court concluded that Wooden’s ten burglaries didn’t occur on dif- +ferent occasions because (1) he committed them “in a single unin- +terrupted course of conduct,” (2) he committed them at “one loca- +tion,” and (3) the offenses were “essentially identical” and “inter- +twined.” Id. at 370. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 15 of 27 + +22-13963 + +Opinion of the Court + +15 + +Approximately four months after Wooden was decided, Ed- +wards entered his unconditional guilty plea in this case. His written +plea agreement recites that “[t]he defendant . . . understands and +acknowledges that the Court must impose a statutory mandatory +minimum term of imprisonment of fifteen (15) years.” Plea Agree- +ment ¶ 3, Dkt. No. 80. The agreement was accompanied by a +signed factual proffer, which recited that “[t]he defendant has at +least seven (7) felony convictions” and specifically enumerated +three of them—“for trafficking in oxycodone (10-022573CF10A), +aggravated battery with a deadly weapon (10-022679CF10A), and +(97- +robbery with a deadly weapon/aggravated battery +000465CF10B).” Factual Proffer ¶ 2. The same three convictions +were cited in Edwards’s presentence investigation report as +grounds for enhancing his sentence under ACCA. See PSI ¶ 20 +(“The defendant is an armed career criminal because he is subject +to an enhanced sentence under the provisions of 18 U.S.C. § 924(e), +pursuant to § 4B1.4(a). The defendant has the prior felony convic- +in docket numbers 10-022573CF110A, 10- +tions as noted +022679CF10A, and 97-000465CF10B.”). And the district court in +fact used those three convictions to enhance Edwards’s sentence. +The judge explained that “looking at the offense level computa- +tion” in the PSI, the “Armed Career Criminal enhancement gets +[the offense level] up to 33.” Tr. of Sentencing Hr’g at 4:19–23 (cit- +ing PSI ¶ 20). + +Edwards didn’t object to the ACCA enhancement on any +ground, including that the offenses underlying the ACCA predi- +cates weren’t committed on different occasions. He didn’t object + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 16 of 27 + +16 + +Opinion of the Court + +22-13963 + +to the PSI’s ACCA-enhanced recommended sentence, see supra at 5 +n.2, nor did he object to the enhancement during the sentencing +hearing. To the contrary, Edwards’s counsel suggested at the sen- +tencing hearing that the ACCA-enhanced 15-year sentence was ap- +propriate: + +[Defense Counsel]: My client has been aware for +months that he’s facing a statutory minimum sen- +tence of 15 years, which is 180 months in federal +prison. He’s been aware of it. He has not gone to +trial on this particular case. He has not testified in this +particular case. He got acceptance of responsibility, +even though it doesn’t really affect him. So he’s just +asking the Court to accept both parties’ recommen- +dations, sentence him to 15 years, and he’ll move on +with his life. + +Tr. of Sentencing Hr’g at 16:5–13; see also id. at 21:14–17 (“Court: +[N]ow that sentence has been imposed, do either of you object to +the Court’s finding of fact or the manner in which the sentence was +pronounced? [Defense Counsel]: No, Judge.”). Hearing no objec- +tion to the applicability of the ACCA enhancement based on Ed- +wards’s three prior convictions, the district court sentenced him to +the 15-year statutory minimum. + +Then came Erlinger, in which the Supreme Court held that +because the determination that a defendant’s prior crimes were +committed on different occasions “‘increase[s] the prescribed range +of penalties to which a criminal defendant is exposed,’” it “must be +resolved by a unanimous jury beyond a reasonable doubt (or freely + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 17 of 27 + +22-13963 + +Opinion of the Court + +17 + +admitted in a guilty plea).” 602 U.S. at 834 (quoting Apprendi v. New +Jersey, 530 U.S. 466, 490 (2000)).5 In light of Erlinger, Edwards con- +tends on appeal that because a jury didn’t find beyond a reasonable +doubt that his three ACCA predicate offenses were committed on +different occasions, the district court committed plain error when +it imposed an enhanced 15-year minimum-mandatory sentence. + +2 + +The government concedes that in light of Erlinger, Edwards +has satisfied the first two prongs of the plain-error standard—an +“error” that is “plain”—because, it says, the different-occasions re- +quirement was neither proven to a jury beyond a reasonable doubt +nor expressly admitted by Edwards in his plea. See Br. of Appellee +at 28.6 + +5 To be clear, Erlinger didn’t add a different-occasions element to the statute. +As already explained, Congress codified the different-occasions requirement +some 35 years ago. And almost four months before Edwards entered his plea, +and eight months before his sentencing hearing, the Supreme Court in Wooden +reiterated the significance of the different-occasions element and clarified the +circumstances under which offenses will (and won’t) satisfy it. Erlinger, rather, +resolved the “who decides” question—namely, by holding that the different- +occasions issue must be resolved either by a jury beyond a reasonable doubt +or by the defendant himself in a guilty plea. See 602 U.S. at 834. District courts, +the Erlinger Court held, “may not assume the jury’s factfinding function for +themselves” to conclude that offenses occurred on different occasions. Id. + +6 The second half of that disjunction isn’t self-evidently correct. Here, Ed- +wards at least arguably admitted in his plea that his three convictions satisfied +ACCA’s different-occasions requirement. As already explained, Edwards ad- +mitted that the three convictions enumerated in both the factual proffer and +the PSI qualified him for the ACCA enhancement. And by pleading to the + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 18 of 27 + +18 + +Opinion of the Court + +22-13963 + +Still, to prevail on plain error, Edwards must show, at step +three, that the Erlinger error affected his “substantial rights.”� +“[U]nlike harmless error—where the government carries the bur- +den—the onus of establishing prejudice under plain error rests with +the defendant.” United States v. Margarita Garcia, 906 F.3d 1255, +1267 (11th Cir. 2018). That burden is “heavy”—and, in particular, +“the burden of showing prejudice to meet the third-prong require- +ment is anything but easy.” Id. (citation modified). + +To demonstrate prejudice, a defendant must “show a rea- +sonable probability that, but for the error, the outcome of the pro- +ceeding would have been different.” United States v. Corbett, 921 F.3d +1032, 1037 (11th Cir. 2019) (quoting Molina-Martinez v. United States, +578 U.S. 189, 194 (2016)). “[W]e define the ‘outcome’ that must +have been affected by considering the nature of the rule violated +by the district court and the role that rule plays in the judicial pro- +cess.” Steiger, 99 F.4th at 1326. When, as here, “the proceeding is +sentencing, we ask whether the sentence would be different but for +erroneous sentencing enhancements or an inaccurate sentencing +guidelines range.” Id. (citation modified). Edwards thus “has the +difficult burden of showing there is a ‘reasonable probability’ that +he would have received a lighter sentence but for the error.” Jones, + +whole—i.e., ACCA applicability, writ large, based on those three convic- +tions—one could argue that he implicitly pleaded to the constituent parts, in- +cluding the applicability of the different-occasions requirement. But the gov- +ernment hasn’t pressed that argument, so we won’t pursue it. + +� + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 19 of 27 + +22-13963 + +Opinion of the Court + +19 + +743 F.3d at 830.7 In other words, a showing that Edwards would +have been entitled to a jury determination on the separate-occa- +sions inquiry is not, in and of itself, enough to establish that the +error affected his “substantial rights” under the plain-error frame- +work. Rather, to prevail, Edwards must further show a reasonable +probability that a jury would have concluded that he committed +the three predicate crimes on fewer than three occasions. + +As already explained, the district court enhanced Edwards’s +sentence under ACCA on the basis of three prior convictions: +(1) for a December 20, 1996, robbery with a deadly weapon and + +7 Both parties seem to misunderstand—albeit in different ways—how the sub- +stantial-rights prong operates in this case. For the first time at oral argument, +Edwards asserted that under Greer v. United States, 593 U.S. 508 (2021), he can +show prejudice if there is a reasonable probability that had he known about +the Erlinger rule, he wouldn’t have pleaded guilty and would have instead in- +sisted on going to trial. See Oral Arg. at 7:46, 28:22. That is incorrect. True, +when a defendant challenges the voluntariness of his plea, the reasonable- +probability standard is appropriately focused on whether he would have +pleaded guilty. Greer, 593 U.S. at 508. Accordingly, that formulation would +have governed Edwards’s voluntariness challenge had we found the first two +prongs of the plain-error standard satisfied. See supra at 7. But it doesn’t gov- +ern his challenge to the applicability of the ACCA enhancement. + +For its part, the government contends that the requisite reasonable +probability doesn’t exist here because Edwards hasn’t shown either “that the +three qualifying offenses were not committed on different occasions” or that +“the government would have been unable to prove that fact had he made a +timely objection.” Br. of Appellee at 31. But that formulation erects too high +a bar. To be sure, a defendant may seek to demonstrate prejudice in those +ways—and they would likely be persuasive—but he isn’t required to do so in +order to show an effect on his substantial rights. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 20 of 27 + 20 -0 0 0 0 -0 -(b)ConfusionmatrixforChain-of-Thought. -C -o -rre -ct -In -co rre -ct -ect ect -orr orr C nc -I -LINC Outcome -emoctuO -ToC -100 L IN 1.0 + +Opinion of the Court + +22-13963 + +aggravated battery, for which the arrest occurred on April 1, 1997; +(2) for an undated instance of drug trafficking, for which the arrest +occurred on December 28, 2010; and (3) for an October 1, 2010, +aggravated battery with a deadly weapon, for which the arrest oc- +curred on December 30, 2010. Edwards concedes that it is fair to +assume that the 1996 robbery was committed on a different occa- +sion than the other two—the nearly 14-year gap is hard to ignore. +But, he maintains, the government didn’t prove that the drug-traf- +ficking and aggravated-battery offenses, both with arrest dates in +2010, were committed on different occasions. + +The PSI contains no details about the drug-trafficking of- +fense other than the arrest date. Although “[t]he circumstances of +this arrest were requested from the Broward County Clerk of +Court,” no report was ever produced. Letter Br. of Appellant at 7, +Dkt. No. 37 (citing PSI ¶ 34). Edwards contends that, “based on the +lack of facts in the record concerning the 2010 drug offense,” a jury +likely would have found that the drug-trafficking and aggravated- +battery offenses were committed on the same occasion and, there- +fore, that he wasn’t subject to the ACCA enhancement.8 Id. at 8. +We cannot agree. + +8 Edwards’s other arguments improperly employ a harmless-error-like stand- +ard, pursuant to which the government bears the burden. See Letter Br. of +Appellant at 5 (“[B]ecause the government did not prove that his offenses oc- +curred on three separate occasions,” “Edwards is entitled to be resentenced +without the ACCA enhancement.”); see also id. at 6 (“[T]he ‘facts’ in the record +here—even the ‘unobjected to’ facts in the PSI—are insufficient for a jury to + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 21 of 27 + +22-13963 + +Opinion of the Court + +21 + +In our review for plain error, we consider “the whole rec- +ord” to determine “the effect of any error on [the defendant’s] sub- +stantial rights,” including “proceedings that both precede and post- +date the errors about which [the defendant] complains.” United +States v. Reed, 941 F.3d 1018, 1021 (11th Cir. 2019) (citation modi- +fied); see also Greer v. United States, 593 U.S. 503, 511 (2021) (explain- +ing that “an appellate court conducting plain-error review may +consider the entire record—not just the record from the particular +proceeding where the error occurred”). So we will examine the en- +tire record for evidence bearing on what the Wooden Court called +the “range of circumstances” that “may be relevant to identifying +episodes of criminal activity.” Wooden, 595 U.S. at 369; see also Er- +linger, 602 U.S. at 828 (noting the “intensely factual nature of this +inquiry”). The Court hoped that, “[f]or the most part,” applying +the different-occasions requirement would be “straightforward and +intuitive.” Wooden, 595 U.S. at 369. But for the tough cases, the +Court offered guidance to help determine whether offenses consti- +tute single or multiple episodes of criminal activity. In particular, +the Court emphasized considerations such as “[1] whether the de- +fendant’s past offenses were ‘committed close in time,’ [2] whether +they were committed near to or far from one another, and +[3] whether the offenses were ‘similar or intertwined’ in purpose +and character.” Erlinger, 602 U.S. at 828 (quoting Wooden, 595 U.S. +at 369). + +conclude beyond a reasonable doubt that the offenses occurred on different +occasions.”). + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 22 of 27 + +22 + +Opinion of the Court + +22-13963 + +Candidly, the record doesn’t shed much light on the first two +Wooden factors. With respect to timing, while the PSI establishes +that the aggravated-battery offense occurred on October 1, 2010, it +doesn’t specify the date of the drug-trafficking offense. Accord- +ingly, we can’t make any firm judgments about whether those two +crimes were “committed close in time, in an uninterrupted course +of conduct” or if they were “separated by substantial gaps in time +or significant intervening events.” Wooden, 595 U.S. at 369. The +record likewise reveals nothing about the “[p]roximity of location” +that would permit an assessment of whether the two crimes oc- +curred far from one another, which would of course make it “less +likely they are components of the same criminal event.” Id. + +That leaves the offenses’ “purpose and character”—and +there, we have more to go on. Edwards’s drug-related offense in- +volved trafficking in oxycodone, possession of cocaine with intent +to deliver or sell, and possession of drug paraphernalia. By con- +trast, his aggravated-battery arose out of domestic dispute with a +former romantic partner. According to the arrest affidavit— + +[O]n October 1, 2010, the defendant picked up a small +flowerpot, forcefully grabbed the victim by her hair, +and then smashed the pot against the victim’s head. +The victim sustained a laceration to the left side of +her head. The victim and the defendant had been in- +volved in a romantic relationship for two years and +lived together for seven months. The victim was +transported to the hospital where she received multi- +ple staples on her head. The defendant was later ar- +rested on December 30, 2010. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 23 of 27 + +22-13963 + +Opinion of the Court + +23 + +PSI ¶ 35. At least on the face of things, Edwards’s drug-trafficking +and aggravated battery offenses seem to have nothing in com- +mon—to be totally unrelated. And there’s certainly nothing in the +record to suggest that the two crimes “share a common scheme or +purpose,” such that they are likely to “compose one occasion.” +Wooden, 595 U.S. at 369. Accordingly, we think the third, “purpose +and character” factor significantly diminishes the probability that +the drug-trafficking and aggravated-battery offenses occurred on +the same occasion. + +Other facts likewise indicate, on balance, that those crimes +occurred on different occasions. First, Edwards was arrested for +the offenses two days apart—for the drug-trafficking crime on De- +cember 28, 2010, and for the aggravated-battery crime on Decem- +ber 30, 2010. Second, the criminal cases bear non-consecutive +docket numbers—indicating, at the very least, that the charges +were filed on different dates. And third, the two crimes were adju- +dicated almost three years apart—the aggravated battery on July +21, 2011, and the drug trafficking on April 29, 2014. We think it at +least relevant—and certainly not irrelevant—that the two offenses +were never combined into a single arrest or case, despite ample op- +portunities to do so. + +We also note the conspicuous absence of any evidence to +the contrary. Even on appeal, and with an opportunity to supple- +ment the record, Edwards hasn’t presented any evidence that +would demonstrate, or even strongly suggest, that the drug-traf- +ficking and aggravated-battery offenses were committed on the + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 24 of 27 + +24 + +Opinion of the Court + +22-13963 + +same occasion. He doesn’t assert, for instance, that the two crimes +took place on the same date, were committed in the same location, +or involved the same victims. Rather, Edwards simply says that +there isn’t sufficient evidence in the record for the government to +prove beyond a reasonable doubt that the ACCA enhancement ap- +plies. But on plain-error review, where the burden rests with Ed- +wards, and with the balance of the record evidence against him, +that’s not enough. + +To be sure, Edwards isn’t required to show by a preponder- +ance of the evidence that, but for the Erlinger error, the result +would have been different. See Dominguez Benitez, 542 U.S. at 83 +n.9 (“The reasonable-probability standard is not the same as, and +should not be confused with, a requirement that a defendant prove +by a preponderance of the evidence that but for error things would +have been different.”). And we acknowledge Edwards’s argument +that the lack of smoking-gun evidence about the locations and +dates of the crimes makes it possible that a jury could have found +the different-occasions requirement not satisfied. But a mere pos- +sibility isn’t a reasonable probability. See United States v. Marcus, +560 U.S. 258, 263 (2010) (deeming a standard pegged to “any possi- +bility, no matter how unlikely,” of a different outcome “irreconcilable +with our ‘plain error’ precedent” (citation modified)). And again, +the record evidence that we do have—the drug-trafficking and ag- +gravated-battery offenses’ divergent “purpose[s] and character[s],” +the different arrest and adjudication dates, the non-consecutive +docket numbers, etc.—indicates that they were indeed “committed +on occasions different from one another.” 18 U.S.C. § 924(e)(1). + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 25 of 27 + +22-13963 + +Opinion of the Court + +25 + +Finally, we note that Edwards hasn’t explained how there +could be a “reasonable probability”—more than a mere possibil- +ity—of a different outcome here. As we’ve observed, see supra at +24, Edwards hasn’t put forward any theory that pushes this case +over the reasonable-probability hurdle. Far from showing a rea- +sonable probability, Edwards hasn’t even speculated to how the Er- +linger error might have affected the outcome. For example, he +hasn’t argued that the aggravated-battery offense occurred in the +course of a dispute over the drug trafficking, such that a jury could +have found that those two offenses occurred on the same occasion. +And “[w]here the effect of an alleged error is so uncertain, a defend- +ant cannot meet his burden of showing that the error actually af- +fected his substantial rights.” Jones v. United States, 527 U.S. 373, +394–95 (1999). + +We therefore hold that Edwards hasn’t met his plain-error +burden to show that his “substantial rights” were affected, and we +affirm the district court’s decision to apply the ACCA enhance- +ment. + C -1 -80 108 29 0.8 -C o -S T 0.22 1 -60 c ra -tc 0.6 -h -p -a 40 d 0.21 0.54 1 -0.4 -24 21 N -20 a iv -e 0.14 0.52 0.56 1 -0.2 -0 -C T d e -LI N C o c h p a N ai v at -cr -S -(c) Comparing the consistency of LINC vs. (d)Similaritybetweenincorrectpredictionsofeachmethod, -Chain-of-Thought. i.e.,(A wrong == B wrong) / (A wrong or B wrong). -Figure5: AnalyzingandcomparingthemistakesmadebyGPT-4ontheFOLIOdataset. -C2: CoTmakesincorrectlogicaldeductions. (GPT-4IncorrectCoT):Let’sthinkstepbystep.Wewant -WhiletherearemanyinstanceswhereCoTmakes toevaluateifMikehasanF1visa.Fromthepremises,we -know that Mike is an international student and needs to -incorrectlogicaldeductions,onecommondeduc- -applyforCPTifhewantstoworkintheUS.Wealsoknow -tion mistake is the structure “if B is true and thataninternationalstudentintheUSwithanF1visaneeds -A → B, then A is true”. In the example shown toapplyforCPTorOPTifthestudentwantstoworkinthe -US.So,itisclearthatMikehasanF1visasincehefitsthe -below,weseethisincorrectdeductionwhereAis -criteria.Therefore,theconclusionistrue. -“MikehasanF1visa”andBis“Mikeneedstoap- -plyforCPTorOPTifhewantstoworkintheUS”. C3: CoTfailstofindcomplexpathsofreason- -We note that making this type of logical mistake ing. We find that with all three models, CoT fails -isoftencalledthefallacyoftheconverse,andthat whenthepathofreasoningnecessarytomakethe -humansarenotimmunetoit,either. deduction is complex. Sometimes, CoT has diffi- -cultygettingstarted,andothertimes,itgetsstuck -Premise1: InternationalstudentsinUShaveeitheranF1 -inthemiddleofareasoningchain. -visaoraJ1visa. -Premise2:AninternationalstudentinUSwithanF1visa 5.2 QuantitativeAnalysis -needstoapplyforCPTorOPTifthestudentwantstowork -intheUS. 1. ComparedtoCoT,LINChasworserecallbut -betterprecisiononTrue/Falsepredictions. To -Premise3:Mikeisaninternationalstudent. -seethis,weplottedtheconfusionmatricesinFig- -Premise4:MikeneedstoapplyforCPTifhewantstowork -ure 5b (CoT) and Figure 5a (LINC). Looking just -intheUS. -at the distributions of predicted labels of the two -Conclusion(GroundTruth: Uncertain): MikehasanF1 -visa. -methods,weseethatCoTpredicts32%True,27% - -| False, | and41%Uncertain, | | | whileLINCpredicts | | | | | | | | | -| ------------------------------------- | ---------------- | --- | --- | ----------------- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | | vs. thenumberofinstanceswhereatleastoneof | | | | | | -| 24%True,17%False,and57%Uncertain(with | | | | | | | themiswrong. | | | | | | -2%ofpredictionsthrowinganerror). Notably,we Figure5dshowsthepairwisesimilaritybetween -observethatLINCpredictsUncertainmuchmore -| | | | | | | | our four | methods, | highlighting | | that | the similar- | -| ---------- | --- | ---- | -------- | --------- | --- | ------ | ----------- | -------- | -------------- | --- | ---- | ------------- | -| frequently | | than | CoT (57% | vs. 41%). | To | under- | | LINC’s | | | | | -| | | | | | | | ity between | | mispredictions | | | and the other | -stand why, note that the translation from natural methods’mispredictions(0.14,0.21,0.22)ismuch -languagetoFOLisalossyprocess: recallthatin lower than the similarity between any pair of the -L1andL2,wesawthattheinformationconveyed in-context reasoning methods (0.52, 0.54, 0.56). -throughtheFOLissometimesasubsetoftheinfor- -| | | | | | | | These results | suggest | | that for | GPT-4 | on FOLIO, | -| --- | --- | --- | --- | --- | --- | --- | ------------- | ------- | --- | -------- | ----- | --------- | -mationintheoriginalpremises. Removingpieces LINCistheonlymethodweevaluatewhichsignifi- -ofcrucialinformationthatwereonthecriticalpath cantlyaltersthewaysinwhichthemodelfailsto -todeducingTrue/Falsemaythenleaveanuncer- -reason. -| tainconclusion. | | | Atthesametime,whiletheFOL | | | | | | | | | | -| ---------------------------------------------- | --- | --- | ------------------------- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | -| translationssometimesdonotretainalloftheinfor- | | | | | | | 6 RelatedWork | | | | | | -mationintheNL,theyrarelycontainfalseinforma- -tionthatwasnotprovidedintheoriginalpremises. Reasoning in LLMs: Our work contributes to -| | | | | | | | the wider | literature | on | eliciting | natural | language | -| --- | --- | --- | --- | --- | --- | --- | --------- | ---------- | --- | --------- | ------- | -------- | -Therefore,LINC’sprecisionwhenpredictingTrue -| | False | | | | | | reasoning | capabilities | | in models. | | Although we | -| --- | ----- | ------- | ---------- | -------- | --- | ------- | --------- | ------------ | --- | ---------- | --- | ----------- | -| or | | is very | high (93%) | compared | to | that of | | | | | | | -CoT(81%),butthiscomesatthecostoflowerre- havefocusedhereoncomparinganeurosymbolic -| | | | | | | | approach | to Scratchpad | | (Nye | et | al., 2021) and | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------- | --- | ---- | --- | -------------- | -callonTrue/Falsepredictions(60%forLINCvs. -| 75%forCoT). | | | | | | | Chain-of-Thought | | prompting | | (Wei | et al., 2022; | -| ----------- | ---- | --- | ---------- | --- | --------- | --- | ---------------- | ------- | --------- | ---- | ------- | ------------- | -| | LINC | | CoT | | | | Kojima | et al., | 2022; | Wang | et al., | 2023b), many | -| | 2. | and | mispredict | on | different | ex- | | | | | | | -amples. EarlierinSec. 5.1,wesawthatLINCand other similar or related techniques have been de- -velopedinrecentyears;theseincludeleast-to-most -CoTexhibitdifferentfailuremodes,whichsuggests -prompting(Zhouetal.,2023),selection-inference -| theyshouldfailondifferentexamples. | | | | | Indeed,we | | | | | | | | -| ---------------------------------- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | -findthatthisisthecaseinourexperimentsonFO- (Creswelletal.,2023),backwardchaining(Tafjord -etal.,2022;Kazemietal.,2023),andself-taught -| LIO: | Figure | 5c | shows a | 2 × 2 confusion | | matrix | | | | | | | -| ---- | ------ | --- | ------- | --------------- | --- | ------ | ------------------------------ | --- | --- | --- | --- | ----------- | -| | | | | | | | reasoning(Zelikmanetal.,2022). | | | | | Someofthese | -whichcompareswhetherornoteachmethod’spre- -| | | | | | | | techniques | have | been | formalized | | under the lan- | -| ------- | --- | ------------ | --- | ------- | -------- | ------ | ---------- | ---- | ---- | ---------- | --- | -------------- | -| diction | | was correct. | We | observe | that out | of the | | | | | | | -24 + 29 + 21 = 74 samples where at least one guage model cascades framework (Dohan et al., -2022). -methodmakesanincorrectprediction,only21are -| | | | | | | | Semantic | parsing: | | The | notion | of a semantic | -| --- | --- | --- | --- | --- | --- | --- | -------- | -------- | --- | --- | ------ | ------------- | -shared. Onacloserexaminationofthese21sam- -ples,wefindthat16areambiguousorincorrectin parserrestsonalongtraditionofresearch(Kamath -theirspecification(detailsinAppendixE.4),sothe andDas,2019)whoseaimistomapfragmentsof -naturallanguageintouseful,symbolicmeaningrep- -twomethodsonlyagreeon5well-formedsamples. -ThissuggeststhatLINCandCoTarecomplementary resentations(ZelleandMooney,1996;Zettlemoyer -methodswhichfailunderdistinctcircumstances. andCollins,2005;Berantetal.,2013;Liangetal., -| | | | | | | | 2013;Wongetal.,2023). | | | Unlikeearlierworksin | | | -| --- | ----------------- | --- | --- | ---------- | --------- | --- | --------------------- | --- | --- | -------------------- | --- | --- | -| | 3. Mispredictions | | of | in-context | reasoning | | | | | | | | -thistradition,weusealanguagemodeltogenerate -| baselines | | are more | similar | to each | other | than | | | | | | | -| --------- | --- | -------- | ------- | ------- | ----- | ---- | --- | --- | --- | --- | --- | --- | -theyarewithmispredictionsofLINC.Asanex- thesemanticparse,whichisamethodunderactive -tensionofthepreviousanalysis,wenextinvestigate investigationinrecentyears(ShinandVanDurme, -2022;Drozdovetal.,2022;Luetal.,2022;Wang -thecorrelationbetweenthemispredictionsofeach -| pair | of methods. | | To do so, | we define | a similarity | | etal.,2023a). | | | | | | -| ---- | ----------- | --- | --------- | --------- | ------------ | --- | ------------- | --- | --- | --- | --- | --- | -score between two methods A and B as follows: Neurosymbolic approaches for reasoning: -GivenadatasetD withN rowsandgroundtruth Methodswhichcombineneuralnetworkswithsym- -| | | }N | }N | | | | bolictechniqueshaveseenbroaduptakeindomains | | | | | | -| -------- | --- | ------- | --- | --------------- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | --- | -| labels{A | | i and{B | i | fromtwomethodsA | | | | | | | | | -| | | i=1 | | i=1 | | | | | | | | | -andB,wedefine adjacent to logical reasoning, such as generating -| | | | (cid:80)N | | | | outputs | consistent | with | a pre-existing | | symbolic | -| --- | --- | --- | --------- | ----- | ------ | --- | ------- | ---------- | ---- | -------------- | --- | -------- | -| | | | | 1[A = | B ̸= R | ] | | | | | | | -sim (A,B) ≜ i=1 i i i knowledge base (Marra et al., 2019; Manhaeve -| | D | | (cid:80)N | | | | | | | | | | -| --- | --- | --- | --------- | ---- | ------ | --- | --- | --- | --- | --- | --- | --- | -| | | | 1[A | ̸= R | orB ̸= | R ] | | | | | | | -i=1 i i i i et al., 2018; Zhang et al., 2023a) and performing -Inwords,sim (A,B)measuresthenumberofin- algorithmicreasoningoversymbolicallygrounded -D -stanceswhereAandBarewronginidenticalways inputs (Ebrahimi et al., 2021; Ibarz et al., 2022; - -Velicˇkovic´ etal.,2022). Asforlogicalreasoning themodeldoesnotneedtolearnhoworwhereto -withLLMsinparticular,therehavebeenafewdif- invokethetool: instead,thetoolispredefinedand -ferentproposalsforwhenandhowtobestcombine is applied after the generation step finishes. For -theLLMwithasymboliccomponent. Zhangetal. example,Gaoetal.(2023)andDrorietal.(2022) -(2022)finetunealanguagemodeltosynthesizepo- solvemathematicalreasoningtasksbygenerating -tential facts paired with likelihoods and then use PythonprogramsandusingthePythoninterpreter -ahandwrittendifferentiablesymbolicreasonerin as the tool, Liu et al. (2023) approach physical -ordertodeduceotherfacts. WeirandVanDurme reasoning tasks with a physical simulator as the -(2022) relax thesolver byinstead training neural tool, and Wong et al. (2023) tackle cognitively- -“entailment”modelstodecideifandhowagivenin- inspiredprobabilisticreasoningtaskswithChurch -ferenceruleappliesateachstage. Concurrentlyto (aprobabilisticprogramminglanguage)asthetool. -thiswork,Logic-LM(Panetal.,2023)andSATLM In the second class, the model must learn to in- -(Yeetal.,2023)proposeneurosymbolicapproaches voke the tool by itself, meaning that the model -whichhavemuchincommonwithLINC.However, mustgenerateexplicitAPIcallstothetoolwhich -otherthanthemodelsanddatasetsconsidered,their are then executed when those calls are decoded -contributions have a few key differences to ours. (Schick et al., 2023; Thoppilan et al., 2022; Yao -First,weplaceparticularemphasisonestablishing etal.,2022;Chengetal.,2023). Ourworkbelongs -anin-depthunderstandingoftherelativebenefits to the former class, with the task at hand being -and drawbacks of a neurosymbolic approach to logicalreasoningandthetoolavailableforusebe- -reasoningwhencomparedtotraditionalin-context ingaFOLsolver(Prover9). Wereferthereaderto -reasoningstrategieslikeChain-of-Thought. Sec- Mialonetal.(2023)foramorethoroughsurveyof -ond, Logic-LM employs a self-refinement strat- recentworkinthetool-usageliterature. -egy, which has shown promise across code gen- -7 Conclusion -erationandNLPtasks(Zhangetal.,2023b;Chen -etal.,2023a;Pengetal.,2023;Madaanetal.,2023; -In this work, we present LINC: Logical Inference -Olaussonetal.,2023)butwhichwedonotconsider -viaNeurosymbolicComputation,aneurosymbolic -here. Third,SATLMstudiesarithmeticreasoning -approachforscalablelogicalreasoningwithlarge -in addition to logical reasoning, showcasing the -languagemodels. OurexperimentsshowthatLINC -versatilityoftheneurosymbolicapproach. Fourth, -leadstosignificantperformancegainsinnearlyev- -andfinally,weuseanFOLrepresentationthatwe -erysettingweconsider,andthatitsupportsgener- -believeiseasierforhumanstoreadandmodelsto -alizationtosettingswherethemodelhastoreason -learn. We highly encourage interested readers to -aboutamuchlargersetofpremisesthanitisshown -studythesetwocontemporaryworksindetail. -inthein-contextlearningexamples. Furthermore, -Autoformalization: Theideaofautomatically carryingoutaquantitativeandqualitativeanalysis -translating natural language into structured sym- of the mistakes made by LINC, we find evidence -bolic representations that programs can reason thatitmaycomplementpurelyin-contextreasoning -abouthasgainedpopularityinthedomainofformal strategiessuchasChain-of-Thoughtprompting, -mathematics,leadingtoautoformalizationsystems sincetheydiffergreatlyinthetypesandfrequen- -forseveraltheoremproversincludingMizar(Wang ciesofmistakesmade. Thisworkthussupportsthe -etal.,2018,2020),Lean3(Azerbayevetal.,2023), efficacy of neurosymbolic approaches to natural -andIsabelle(Wuetal.,2022). Outsideformalmath- languagereasoning,settingthestageforcontinued -ematics,autoformalizationhasalsobeenappliedto advancesincombininglargelanguagemodelsand -translatingnaturallanguageintosystemspecifica- symbolic reasoning engines; we discuss several -tionlanguagessuchastemporallogic(Hahnetal., promisingfuturedirectionsinAppendixA. -2022;Cosleretal.,2023;Chenetal.,2023b). -8 Limitations -Toolusage: Ourworkisheavilyinspiredbyre- -cent work on tool usage. The central idea in this Narrowscopeoflogicalreasoningtaskconsid- -lineofresearchistoaugmentlanguagemodelswith ered: Inthiswork,wefocusexclusivelyononeas- -externaltoolssuchascalculators,codeinterpreters pectoflogicalreasoning: predictingthetruthvalue -and information retrieval systems. We further di- of a conclusion given a set of natural language -videtheseworksintotwoclasses. Inthefirstclass, premises. Here, we consider a setting where the - -premisesandconclusionareexpressedinrelatively dleproblemsrequiringhigher-orderlogics(Miller -short statements, which makes the formalization andNadathur,1986;Higginbotham,1998). Also, -tasktractable. Inparticular,ProofWriter’snatural in many settings it is desirable to work with non- -language statements are synthetically generated, classicallogics(Priest,2008;Burgess,2009). Al- -so they can be easily and accurately parsed into ternativetheoremproverswouldbeneededforsuch -FOL. FOLIO reflects a more naturalistic dataset, problems. AmethodlikeLINCcanbenaturallyex- -soweseeahigherfailurerateinLINC’ssemantic -| | | | | | | tended | to those | settings, | but | exactly | how | well it | -| --- | --- | --- | --- | --- | --- | ------ | -------- | --------- | --- | ------- | --- | ------- | -parsingstep. However,theformalizationtaskbe- worksthererequiresfurtherinvestigations. -comesmoredifficultifthepremisesareinlonger Computational costs: Implementing our ap- -paragraphform,suchasinquestionansweringor proachwithbothGPTmodelsandtheStarCoder+ -contradictiondetectionfromcontextpassages. This modelrequiresnon-trivialresources. Theformer -| is because | the same | piece | of | information | can be | | | | | | | | -| ---------- | -------- | ----- | --- | ----------- | ------ | -------- | -------- | --- | ------ | ------------ | --- | ------- | -| | | | | | | requires | reliance | on | costly | API requests | | and the | -formalizedinavarietyofways,andthereisalot latterdedicatedGPUsforinference. Especiallyas -ofinformationthatmustbepragmaticallyinferred weusemajorityvoting,manygenerationsmustbe -toarriveattheproperconclusion. madeforeachquery,increasingthecomputational -requirements. -| Generalizability | | of | qualitative | evaluation: | | | | | | | | | -| ---------------- | ------------ | ---- | ----------- | ----------- | ---- | --- | --- | --- | --- | --- | --- | --- | -| | | LINC | | CoT | | | | | | | | | -| While | we find that | | and | produce | com- | | | | | | | | -plementarymistakesforournaturallanguagerea- 9 Acknowledgements -soningtask,itisunclearifthisresultalsoholdstrue -| | | | | | | T.X. Olausson | | is supported | | by the | Defense | Ad- | -| ---------- | ---------- | ---- | --- | -------- | ---------- | ------------- | --- | ------------ | --- | ------ | ------- | --- | -| in similar | scenarios, | such | as | the ones | considered | | | | | | | | -vancedResearchProjectsAgency(DARPA)under -| inPAL(Gaoetal.,2023),Logic-LM,and | | | | | SATLM. | | | | | | | | -| --------------------------------- | ---------- | ---------- | --- | --------------- | ------ | ----------------------------------- | --- | --- | --- | --- | --- | --- | -| | | | | | | theASKEMprogram,awardHR00112220042. | | | | | | A. | -| This is | due to the | difference | | in intermediate | lan- | | | | | | | | -GuissupportedbytheNationalScienceFounda- -| guageandoveralllogicalreasoningtask. | | | | | However, | | | | | | | | -| ------------------------------------ | ---- | --- | -------- | --------- | -------- | ---------- | -------- | --- | --------- | ---------- | -------- | ----- | -| | | | | | | tion (NSF) | Graduate | | Research | Fellowship | | under | -| we hypothesize | that | it | will and | encourage | future | | | | | | | | -| | | | | | | Grant No. | 2141064. | | B. Lipkin | | and C.E. | Zhang | -investigationinthisdirection. -| | | | | | | aresupportedbyMITPresidentialFellowships. | | | | | | A. | -| --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | -Moresophisticatedreasoningtechniques: Re- Solar-LezamaissupportedbytheNationalScience -| cent work | has proposed | | more | sophisticated | tech- | | | | | | | | -| --------- | ------------------------ | --- | ---- | ------------- | ---------- | -------------------- | ----- | --- | ----- | ------------------- | --- | ------- | -| | | | | | | Foundation | (NSF) | and | Intel | Corporation | | through | -| niques | beyond chain-of-thought, | | | such | as tree of | | | | | | | | -| | | | | | | NSFGrantCCF:2217064. | | | | J.B.Tenenbaumissup- | | | -thoughts (Yao et al., 2023), program of thoughts, portedbyAFOSRGrant#FA9550-22-1-0387and -(Chenetal.,2022),orusingretrievalinchain-of- -| | | | | | | the MIT-IBM | | Watson | AI | Lab. R.P. | Levy | is sup- | -| ------------------------------------- | --- | --- | --- | --- | ----- | ----------- | --- | ------ | --- | --------- | ---- | ------- | -| thoughtprompting(Yasunagaetal.,2023). | | | | | These | | | | | | | | -portedbyagrantfromtheSimonsFoundationto -have potential to improve and eliminate some of theSimonsCenterfortheSocialBrainatMIT. -thefailuremodesofthetraditionalCoTmethod. -| | | | | | In | Wethankouranonymousreviewersfortheirin- | | | | | | | -| --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -addition,ideassuchasself-repairmayalsoserve -| | | | | | | sightfulfeedbackandrecommendations. | | | | | | Wethank | -| --------------------------- | --- | --- | --- | --------------- | --- | ----------------------------------- | --- | --- | --- | --- | --- | ------- | -| toimprovethesefailuremodes. | | | | Itremainsfuture | | | | | | | | | -themembersoftheComputerAidedProgramming, -work to do a more thorough investigation of the Computational Psycholinguistics, and Computa- -| efficacy | of these | techniques, | | though there | is also | | | | | | | | -| -------- | -------- | ----------- | --- | ------------ | ------- | ---------------- | --- | ------- | ------ | --- | ---------------- | --- | -| | | | | | | tional Cognitive | | Science | groups | | for constructive | | -preliminaryevidencethattheystilllackreasoning -| | | | | | | commentaryatvariousstagesofthisproject. | | | | | | We | -| --- | --- | --- | --- | --- | --- | --------------------------------------- | --- | --- | --- | --- | --- | --- | -capabilities(Huangetal.,2023). -thankYoonKimforhelpfulsuggestionsandcom- -Scalability: It is unclear how well LINC will mentsonpiecesoftheinitialprojectproposal. In -perform as the number of premises scales. First, addition,wethankZhaofengWuandSimengHan -onemistakeinformalizationcanleadtoanincor- fordiscussionsregardingtheFOLIOdataset. -rectdeduction,andmorepremisesleadtoahigher -| probability | of errors. | | Second, | in the | deduction | | | | | | | | -| ----------- | ---------- | --- | ------- | ------ | --------- | --- | --- | --- | --- | --- | --- | --- | -References -| stage, while | many | fast | algorithms | (e.g., | forward- | | | | | | | | -| ------------ | ---- | ---- | ---------- | ------ | -------- | --- | --- | --- | --- | --- | --- | --- | -andbackward-chaining)existforlogicaldeduction, -| | | | | | | Cem Anil, | Yuhuai | Wu, | Anders | Andreassen, | | Aitor | -| -------------------------------- | --- | --- | --- | ------------- | --- | ---------- | ------ | ------ | ------ | ----------- | --------- | ----- | -| thegeneralproblemisstillNP-hard. | | | | Therefore,the | | | | | | | | | -| | | | | | | Lewkowycz, | | Vedant | Misra, | Vinay | Ramasesh, | Am- | -theoremprovermaytakealongtimeinpractice. broseSlone,GuyGur-Ari,EthanDyer,andBehnam -| | | | | | | Neyshabur. | | 2022. | Exploring | length | generalization | | -| ---------------------------------- | --- | --- | --- | --- | ------ | ---------- | -------- | ------- | --------- | ----------- | -------------- | --------- | -| Otherlogicsbeyondfirst-orderlogic: | | | | | Inthis | | | | | | | | -| | | | | | | in large | language | models. | | In Advances | | in Neural | -work, we exclusively focus on first-order logic. InformationProcessingSystems,volume35,pages -| However, | FOL is | not | expressive | enough | to han- | | | | | | | | -| -------- | ------ | --- | ---------- | ------ | ------- | ------------ | --- | ------------- | --- | --- | --- | --- | -| | | | | | | 38546–38556. | | (Citedonpg.1) | | | | | - -Zhangir Azerbayev, Bartosz Piotrowski, Hailey language modeling with pathways. arXiv preprint -Schoelkopf,EdwardWAyers,DragomirRadev,and arXiv:2204.02311. (Citedonpg.1) -| Jeremy | Avigad. | 2023. | ProofNet: | Autoformalizing | | | | | | | | -| ------ | ------- | ----- | --------- | --------------- | --- | --- | --- | --- | --- | --- | --- | -andformallyprovingundergraduate-levelmathemat- -| | | | | | | | Matthias | Cosler, Christopher | | Hahn, | Daniel Men- | -| --- | --- | --- | --- | --- | --- | --- | -------- | ------------------- | --- | ----- | ----------- | -ics. arXivpreprintarXiv:2302.12433. (Citedonpg. doza,FrederikSchmitt,andCarolineTrippel.2023. -| 9) | | | | | | | nl2spec: | Interactivelytranslatingunstructurednatu- | | | | -| --- | --- | --- | --- | --- | --- | --- | -------- | ----------------------------------------- | --- | --- | --- | -rallanguagetotemporallogicswithlargelanguage -David Barker-Plummer, Jon Barwise, and John models. InComputerAidedVerification,pages383– -| Etchemendy. | | 2011. Language, | | proof, | and logic, | 2 | | | | | | -| ----------- | ------------------------------------ | --------------- | --- | ------ | ---------- | --- | ----------------------------------- | --- | --- | --- | -------- | -| | | | | | | | 396,Cham.SpringerNatureSwitzerland. | | | | (Citedon | -| edition. | CenterfortheStudyofLanguageandInfor- | | | | | | | | | | | -pg.9) -| mation. | (Citedonpg.2) | | | | | | | | | | | -| ------- | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -AntoniaCreswell,MurrayShanahan,andIrinaHiggins. -| Mohammad | Bavarian, | Heewoo | | Jun, Nikolas | Tezak, | | | | | | | -| -------- | --------- | ------ | --- | ------------ | ------ | --- | ------------------------------------------------- | --- | --- | --- | --- | -| | | | | | | | 2023. Selection-inference:Exploitinglargelanguage | | | | | -JohnSchulman,ChristineMcLeavey,JerryTworek, -| | | | | | | | models | for interpretable | logical | reasoning. | In The | -| -------- | ----- | ----- | --------- | -------- | --- | ---- | ------ | ----------------- | ------- | ---------- | ------ | -| and Mark | Chen. | 2022. | Efficient | training | of | lan- | | | | | | -EleventhInternationalConferenceonLearningRep- -| guage | models | to fill in | the middle. | | arXiv preprint | | | | | | | -| ----------------- | ------ | -------------- | ----------- | --- | -------------- | --- | ------------- | ----------------- | --- | --- | --- | -| | | | | | | | resentations. | (Citedonpg.1,3,8) | | | | -| arXiv:2207.14255. | | (Citedonpg.15) | | | | | | | | | | -JonathanBerant,AndrewChou,RoyFrostig,andPercy David Dohan, Winnie Xu, Aitor Lewkowycz, Ja- -Liang. 2013. Semantic parsing on Freebase from cob Austin, David Bieber, Raphael Gontijo Lopes, -question-answerpairs. InProceedingsofthe2013 Yuhuai Wu, Henryk Michalewski, Rif A Saurous, -Conference on Empirical Methods in Natural Lan- JaschaSohl-Dickstein,etal.2022. Languagemodel -| | | | | | | | cascades. | arXivpreprintarXiv:2207.10342. | | | (Cited | -| --- | --- | --- | --- | --- | --- | --- | --------- | ------------------------------ | --- | --- | ------ | -guageProcessing,pages1533–1544,Seattle,Wash- -| ington,USA.AssociationforComputationalLinguis- | | | | | | | onpg.8) | | | | | -| ---------------------------------------------- | --- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | -tics. (Citedonpg.8) -IddoDrori,SarahZhang,ReeceShuttleworth,Leonard -Tom Brown, Benjamin Mann, Nick Ryder, Melanie Tang, Albert Lu, Elizabeth Ke, Kevin Liu, Linda -Subbiah,JaredDKaplan,PrafullaDhariwal,Arvind -| | | | | | | | Chen,SunnyTran,NewmanCheng,etal.2022. | | | | A | -| --- | --- | --- | --- | --- | --- | --- | ------------------------------------- | --- | --- | --- | --- | -Neelakantan,PranavShyam,GirishSastry,Amanda -neuralnetworksolves,explains,andgeneratesuni- -Askell,etal.2020. Languagemodelsarefew-shot versitymathproblemsbyprogramsynthesisandfew- -learners. InAdvancesinneuralinformationprocess- shotlearningathumanlevel. ProceedingsoftheNa- -ingsystems,volume33,pages1877–1901. (Citedon tionalAcademyofSciences,119(32):e2123433119. -pg.1,4) -(Citedonpg.9) -| JohnPBurgess.2009. | | Philosophicallogic. | | | Princeton | | | | | | | -| ------------------ | --- | ------------------- | --- | --- | --------- | --- | --------------- | --------- | -------- | ---- | -------- | -| | | | | | | | Andrew Drozdov, | Nathanael | Schärli, | Ekin | Akyürek, | -| UniversityPress. | | (Citedonpg.10) | | | | | | | | | | -NathanScales,XinyingSong,XinyunChen,Olivier -| | | | | | | | Bousquet, | and Denny | Zhou. | 2022. Compositional | | -| ----- | ----- | -------- | --- | ----- | ----- | --- | --------- | --------- | ----- | ------------------- | --- | -| Wenhu | Chen, | Xueguang | Ma, | Xinyi | Wang, | and | | | | | | -William W Cohen. 2022. Program of thoughts semanticparsingwithlargelanguagemodels. arXiv -| | | | | | | | preprintarXiv:2209.15003. | | (Citedonpg.8) | | | -| ---------- | ----------------------------------- | --------- | ------ | ----- | -------- | --- | ------------------------- | --- | ------------- | --- | --- | -| prompting: | Disentanglingcomputationfromreason- | | | | | | | | | | | -| ing for | numerical | reasoning | tasks. | arXiv | preprint | | | | | | | -NouhaDziri,XimingLu,MelanieSclar,XiangLorraine -| arXiv:2211.12588. | | (Citedonpg.10) | | | | | | | | | | -| ----------------- | --- | -------------- | --- | --- | --- | --- | -------------- | -------------- | --- | ---------- | ----- | -| | | | | | | | Li, LiweiJian, | BillYuchenLin, | | PeterWest, | Chan- | -Xinyun Chen, Maxwell Lin, Nathanael Schärli, and dra Bhagavatula, Ronan Le Bras, Jena D Hwang, -DennyZhou.2023a. Teachinglargelanguagemod- etal.2023. Faithandfate: Limitsoftransformerson -elstoself-debug. arXivpreprintarXiv:2304.05128. compositionality. arXivpreprintarXiv:2305.18654. -| (Citedonpg.9) | | | | | | | (Citedonpg.1) | | | | | -| ------------- | ----- | ----- | ------- | ---- | ------ | --- | ------------- | --- | --- | --- | --- | -| Yongchao | Chen, | Rujul | Gandhi, | Yang | Zhang, | and | | | | | | -MonirehEbrahimi,AaronEberhart,FedericoBianchi, -| Chuchu | Fan. | 2023b. NL2TL: | | Transforming | | natu- | | | | | | -| ------ | ---- | ------------- | --- | ------------ | --- | ----- | ---------------------- | --- | ------------------------ | --- | --- | -| | | | | | | | andPascalHitzler.2021. | | Towardsbridgingtheneuro- | | | -rallanguagestotemporallogicsusinglargelanguage symbolic gap: Deep deductive reasoners. Applied -| models. | arXivpreprintarXiv:2305.07766. | | | | (Citedon | | | | | | | -| ------- | ------------------------------ | --- | --- | --- | -------- | --- | -------------------------- | --- | --- | ------------- | --- | -| | | | | | | | Intelligence,51:6326–6348. | | | (Citedonpg.8) | | -pg.9) -| | | | | | | | HerbertBEnderton.2001. | | Amathematicalintroduction | | | -| --------- | --------- | ------- | ----- | ----------- | ------------ | --- | ---------------------- | ----------------------- | ------------------------- | --- | --- | -| Zhoujun | Cheng, | Tianbao | Xie, | Peng | Shi, Chengzu | | | | | | | -| | | | | | | | tologic. | Elsevier. (Citedonpg.2) | | | | -| Li, Rahul | Nadkarni, | | Yushi | Hu, Caiming | Xiong, | | | | | | | -DragomirRadev,MariOstendorf,LukeZettlemoyer, -| | | | | | | | Luyu Gao, | Aman Madaan, | Shuyan | Zhou, | Uri Alon, | -| -------------------------- | --- | --- | --- | ---------------- | --- | --- | --------------- | ------------ | ------------------------- | ----- | --------- | -| NoahA.Smith,andTaoYu.2023. | | | | Bindinglanguage | | | | | | | | -| | | | | | | | PengfeiLiu, | YimingYang, | JamieCallan, | | andGra- | -| modelsinsymboliclanguages. | | | | InTheEleventhIn- | | | | | | | | -| | | | | | | | hamNeubig.2023. | | PAL:Program-aidedlanguage | | | -ternationalConferenceonLearningRepresentations. -| | | | | | | | models. | In International | Conference | | on Machine | -| --- | --- | --- | --- | --- | --- | --- | ------- | ---------------- | ---------- | --- | ---------- | -(Citedonpg.9) -| | | | | | | | Learning,pages10764–10799.PMLR. | | | | (Citedonpg. | -| -------------------------------------------- | ------ | ------ | ------- | ---- | -------- | --- | ------------------------------- | --- | --- | --- | ----------- | -| AakankshaChowdhery,SharanNarang,JacobDevlin, | | | | | | | 9,10) | | | | | -| Maarten | Bosma, | Gaurav | Mishra, | Adam | Roberts, | | | | | | | -Paul Barham, Hyung Won Chung, Charles Sutton, Herbert P Grice. 1975. Logic and conversation. In -Sebastian Gehrmann, et al. 2022. Palm: Scaling Speechacts,pages41–58.Brill. (Citedonpg.15) - -Christopher Hahn, Frederik Schmitt, Julia J Tillman, Percy Liang, Rishi Bommasani, Tony Lee, Dimitris -NiklasMetzger,JulianSiber,andBerndFinkbeiner. Tsipras, Dilara Soylu, Michihiro Yasunaga, Yian -2022. Formalspecificationsfromnaturallanguage. Zhang,DeepakNarayanan,YuhuaiWu,AnanyaKu- -arXivpreprintarXiv:2206.01962. (Citedonpg.9) mar, et al. 2022. Holistic evaluation of language -| | | | | | | models. | arXivpreprintarXiv:2211.09110. | | | (Citedon | -| --------------------------------------------- | -------- | ----------- | ------- | -------- | --------- | ------------ | ------------------------------ | --------- | -------------- | -------- | -| SimengHan,HaileySchoelkopf,YilunZhao,Zhenting | | | | | | pg.1) | | | | | -| Qi, Martin | Riddell, | Luke | Benson, | Lucy | Sun, Eka- | | | | | | -| | | | | | | Percy Liang, | Michael | I Jordan, | and Dan Klein. | 2013. | -| terina Zubova, | | Yujie Qiao, | Matthew | Burtell, | et al. | | | | | | -2022. Folio: Naturallanguagereasoningwithfirst- Learning dependency-based compositional seman- -arXivpreprintarXiv:2209.00840. tics. Computational Linguistics, 39(2):389–446. -| orderlogic. | | | | | (Cited | | | | | | -| ----------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | -(Citedonpg.8) -onpg.3) -RuiboLiu,JasonWei,ShixiangShaneGu,Te-YenWu, -| James Higginbotham. | | 1998. | On | higher-order | logic | | | | | | -| ------------------- | --- | ----- | --- | ------------ | ----- | --- | --- | --- | --- | --- | -SoroushVosoughi,ClaireCui,DennyZhou,andAn- -| andnaturallanguage. | | InProceedingsoftheBritish | | | | | | | | | -| --------------------------- | --------- | ------------------------- | --- | -------------- | --- | -------------------------------- | ---------- | ---------- | ---------------- | ----------- | -| | | | | | | drewM.Dai.2023. | | Mind’seye: | Groundedlanguage | | -| Academy,volume95,pages1–27. | | | | (Citedonpg.10) | | | | | | | -| | | | | | | modelreasoningthroughsimulation. | | | InTheEleventh | | -| | | | | | | International | Conference | on | Learning | Representa- | -| Jie Huang | and Kevin | Chen-Chuan | | Chang. 2022. | To- | | | | | | -tions. (Citedonpg.9) -| wardsreasoninginlargelanguagemodels: | | | | | Asurvey. | | | | | | -| ------------------------------------ | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | -arXivpreprintarXiv:2212.10403. (Citedonpg.1) XuantaoLu,JingpingLiu,ZhouhongGu,HanwenTong, -| | | | | | | ChenhaoXie, | JunyangHuang, | | YanghuaXiao, | and | -| --- | --- | --- | --- | --- | --- | ----------- | ------------- | --- | ------------ | --- | -Jie Huang, Xinyun Chen, Swaroop Mishra, Wenguang Wang. 2022. Parsing natural language -Huaixiu Steven Zheng, Adams Wei Yu, Xiny- intopropositionalandfirst-orderlogicwithdualre- -ingSong,andDennyZhou.2023. Largelanguage inforcement learning. In Proceedings of the 29th -models cannot self-correct reasoning yet. arXiv InternationalConferenceonComputationalLinguis- -| preprintarXiv:2310.01798. | | | (Citedonpg.10) | | | | | | | | -| ------------------------- | --- | --- | -------------- | --- | --- | -------------------- | ------------------------------- | ------------- | --- | --- | -| | | | | | | tics,pages5419–5431. | | (Citedonpg.8) | | | -| | | | | | | AmanMadaan, | NiketTandon,PrakharGupta,Skyler | | | | -BorjaIbarz,VitalyKurin,GeorgePapamakarios,Kyr- -| | | | | | | Hallinan, | Luyu Gao, | Sarah | Wiegreffe, | Uri Alon, | -| ---------------- | --- | ----- | -------- | ------ | -------- | --------- | --------- | ----- | ---------- | --------- | -| iacos Nikiforou, | | Mehdi | Bennani, | Róbert | Csordás, | | | | | | -AndrewJosephDudzik,MatkoBošnjak,AlexVitvit- Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, -skyi,YuliaRubanova,etal.2022. Ageneralistneural etal.2023. Self-refine:Iterativerefinementwithself- -algorithmiclearner. InLearningonGraphsConfer- feedback. arXivpreprintarXiv:2303.17651. (Cited -onpg.9) -| ence,pages2–1.PMLR. | | | (Citedonpg.8) | | | | | | | | -| ------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | --- | -RobinManhaeve,SebastijanDumancic,AngelikaKim- -| AishwaryaKamathandRajarshiDas.2019. | | | | | Asurvey | | | | | | -| ----------------------------------- | --- | ------------------------ | --- | --- | ------- | --------------------- | -------------------------------- | ------------------- | --- | --- | -| | | | | | | mig, ThomasDemeester, | | andLucDeRaedt.2018. | | | -| onsemanticparsing. | | InAutomatedKnowledgeBase | | | | | | | | | -| | | | | | | DeepProbLog: | Neuralprobabilisticlogicprogram- | | | | -| Construction(AKBC). | | (Citedonpg.8) | | | | | | | | | -ming. InAdvancesinneuralinformationprocessing -| | | | | | | systems,volume31. | | (Citedonpg.8) | | | -| -------------- | --- | ------- | ---- | -------------- | --- | ----------------- | --- | ------------- | --- | --- | -| Mehran Kazemi, | | Najoung | Kim, | Deepti Bhatia, | Xin | | | | | | -Xu,andDeepakRamachandran.2023. LAMBADA: Giuseppe Marra, Francesco Giannini, Michelangelo -Backwardchainingforautomatedreasoninginnat- -| | | | | | | Diligenti,andMarcoGori.2019. | | | Integratinglearn- | | -| -------------- | ------------------ | -------------- | --- | ------------- | ------ | ----------------------------------- | ---------- | --- | ----------------- | ------- | -| ural language. | | In Proceedings | | of the 61st | Annual | | | | | | -| | | | | | | ingandreasoningwithdeeplogicmodels. | | | | InJoint | -| Meeting | of the Association | | for | Computational | Lin- | | | | | | -| | | | | | | European | Conference | on | Machine Learning | and | -guistics(Volume1: LongPapers),pages6547–6568, KnowledgeDiscoveryinDatabases,pages517–532. -Toronto,Canada.AssociationforComputationalLin- Springer. (Citedonpg.8) -| guistics. | (Citedonpg.8) | | | | | | | | | | -| --------- | ------------- | --- | --- | --- | --- | ---------- | ---------- | --- | ----------- | ------ | -| | | | | | | W. McCune. | 2005–2010. | | Prover9 and | mace4. | -DenisKocetkov,RaymondLi,LoubnaBenAllal,JiaLi, |http://www.cs.unm.edu/ mccune/prover9/|. (Cited -onpg.3) -ChenghaoMou,CarlosMuñozFerrandis,YacineJer- -nite,MargaretMitchell,SeanHughes,ThomasWolf, -| | | | | | | Ian McKenzie, | Alexander | Lyzhov, | Alicia | Parrish, | -| ---------- | ---------------------------------- | --- | --- | --- | --- | ------------- | --------- | ------- | ------ | -------- | -| etal.2022. | TheStack:3TBofpermissivelylicensed | | | | | | | | | | -AmeyaPrabhu,AaronMueller,NajoungKim,Sam -| source code. | | arXiv | preprint | arXiv:2211.15533. | | | | | | | -| -------------- | --- | ----- | -------- | ----------------- | --- | -------------------------- | --- | --- | ----------------- | --- | -| (Citedonpg.15) | | | | | | Bowman,andEthanPerez.2022. | | | Theinversescaling | | -prize. (Citedonpg.1) -TakeshiKojima,ShixiangShaneGu,MachelReid,Yu- Quinn McNemar. 1947. Note on the sampling error -| takaMatsuo,andYusukeIwasawa.2022. | | | | | Largelan- | | | | | | -| --------------------------------- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | -ofthedifferencebetweencorrelatedproportionsor -| guagemodelsarezero-shotreasoners. | | | | InAdvances | | | | | | | -| --------------------------------- | --- | --- | --- | ---------- | --- | ------------ | ---------------------------- | --- | --- | ------ | -| | | | | | | percentages. | Psychometrika,12(2):153–157. | | | (Cited | -inneuralinformationprocessingsystems,volume35, -onpg.5) -| pages22199–22213. | | (Citedonpg.1,4,8) | | | | | | | | | -| ----------------- | --- | ----------------- | --- | --- | --- | --- | --- | --- | --- | --- | -GrégoireMialon,RobertoDessì,MariaLomeli,Christo- -Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas forosNalmpantis,RamPasunuru,RobertaRaileanu, -Muennighoff,DenisKocetkov,ChenghaoMou,Marc Baptiste Rozière, Timo Schick, Jane Dwivedi-Yu, -Marone,ChristopherAkiki,JiaLi,JennyChim,etal. AsliCelikyilmaz,etal.2023. Augmentedlanguage -2023. Starcoder: maythesourcebewithyou! arXiv models: asurvey. arXivpreprintarXiv:2302.07842. -| preprintarXiv:2305.06161. | | | (Citedonpg.4,15) | | | (Citedonpg.9) | | | | | -| ------------------------- | --- | --- | ---------------- | --- | --- | ------------- | --- | --- | --- | --- | - -DaleA.MillerandGopalanNadathur.1986. Someuses Abulhair Saparov, Richard Yuanzhe Pang, Vishakh -ofhigher-orderlogicincomputationallinguistics. In Padmakumar,NitishJoshi,SeyedMehranKazemi, -Proceedingsofthe24thAnnualMeetingonAssoci- Najoung Kim, and He He. 2023. Testing the -ationforComputationalLinguistics,page247–256, general deductive reasoning capacity of large lan- -NewYork,USA.AssociationforComputationalLin- guagemodelsusingoodexamples. arXivpreprint -| guistics. | (Citedonpg.10) | | | | | | arXiv:2305.15269. | | (Citedonpg.1) | | | | | -| --------- | -------------- | --- | --- | --- | --- | --- | ----------------- | --- | ------------- | --- | --- | --- | --- | -MaxwellNye,AndersJohanAndreassen,GuyGur-Ari, TimoSchick,JaneDwivedi-Yu,RobertoDessì,Roberta -Henryk Michalewski, Jacob Austin, David Bieber, Raileanu,MariaLomeli,LukeZettlemoyer,Nicola -| | | | | | | | Cancedda,andThomasScialom.2023. | | | | | Toolformer: | | -| ----- | ------ | ----- | ---------- | --- | ------- | ------ | ------------------------------- | --- | --- | --- | --- | ----------- | --- | -| David | Dohan, | Aitor | Lewkowycz, | | Maarten | Bosma, | | | | | | | | -Languagemodelscanteachthemselvestousetools. -| DavidLuan,etal.2021. | | | Showyourwork: | | | Scratch- | | | | | | | | -| -------------------- | ------------------------------ | --- | ------------- | --- | ---- | -------- | ------------------------------ | --- | ----- | ---- | --------------- | --- | --------- | -| | | | | | | | arXivpreprintarXiv:2302.04761. | | | | (Citedonpg.2,9) | | | -| pads | for intermediate | | computation | | with | language | | | | | | | | -| models. | arXivpreprintarXiv:2112.00114. | | | | | (Citedon | | | | | | | | -| | | | | | | | Noam Shazeer. | | 2019. | Fast | transformer | | decoding: | -pg.1,4,8) -| | | | | | | | One write-head | | is all | you | need. | arXiv | preprint | -| ------ | --------- | ------- | --- | ------- | ------------- | --------- | ----------------- | --- | -------------- | --- | ------ | ----- | -------- | -| | | | | | | | arXiv:1911.02150. | | (Citedonpg.15) | | | | | -| Theo X | Olausson, | Jeevana | | Priya | Inala, | Chenglong | | | | | | | | -| Wang, | Jianfeng | Gao, | and | Armando | Solar-Lezama. | | | | | | | | | -| | | | | | | | Richard Shin | and | Benjamin | Van | Durme. | 2022. | Few- | -2023. Isself-repairasilverbulletforcodegenera- -shotsemanticparsingwithlanguagemodelstrained -| tion? | arXivpreprintarXiv:2306.09896. | | | | | (Citedon | | | | | | | | -| ----- | ------------------------------ | --- | --- | --- | --- | -------- | ------- | ---------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | | oncode. | InProceedingsofthe2022Conferenceof | | | | | | -pg.9) -theNorthAmericanChapteroftheAssociationfor -| | | | | | | | ComputationalLinguistics: | | | HumanLanguageTech- | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------------- | --- | --- | ------------------ | --- | --- | --- | -OpenAI.2023. GPT-4technicalreport. (Citedonpg.1, nologies, pages 5417–5425, Seattle, United States. -4,15) -| | | | | | | | Association | for | Computational | | Linguistics. | | (Cited | -| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ------------- | --- | ------------ | --- | ------ | -onpg.8) -LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, -CarrollWainwright,PamelaMishkin,ChongZhang, -| | | | | | | | Aarohi Srivastava, | | Abhinav | Rastogi, | | Abhishek | Rao, | -| --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | ------- | -------- | --- | -------- | ---- | -SandhiniAgarwal,KatarinaSlama,AlexRay,etal. AbuAwalMdShoeb,AbubakarAbid,AdamFisch, -2022. Training languagemodelsto followinstruc- AdamRBrown,AdamSantoro,AdityaGupta,Adrià -tionswithhumanfeedback. InAdvancesinNeural Garriga-Alonso, et al. 2023. Beyond the imitation -InformationProcessingSystems,volume35,pages -| | | | | | | | game: | Quantifying | and | extrapolating | | the | capabili- | -| ------------ | ---- | ---------------- | -------- | --- | ----- | --------- | --------------------- | ----------- | ------------- | --------------------- | --- | --- | --------- | -| 27730–27744. | | (Citedonpg.4,15) | | | | | | | | | | | | -| | | | | | | | tiesoflanguagemodels. | | | TransactionsonMachine | | | | -| | | | | | | | LearningResearch. | | (Citedonpg.1) | | | | | -| Liangming | Pan, | Alon | Albalak, | | Xinyi | Wang, and | | | | | | | | -William Yang Wang. 2023. Logic-lm: Empower- RobertStalnaker.2002. Commonground. Linguistics -ing large language models with symbolic solvers andphilosophy,25(5/6):701–721. (Citedonpg.15) -| for faithful | | logical | reasoning. | | arXiv | preprint | | | | | | | | -| ------------ | --- | ------- | ---------- | --- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | -OyvindTafjord,BhavanaDalvi,andPeterClark.2021. -| arXiv:2305.12295. | | | (Citedonpg.9) | | | | | | | | | | | -| ----------------- | --- | --- | ------------- | --- | --- | --- | --------------------------------------- | --- | ---------- | ------------- | --- | ------- | ------- | -| | | | | | | | ProofWriter: | | Generating | implications, | | proofs, | and | -| | | | | | | | abductivestatementsovernaturallanguage. | | | | | | InFind- | -GuilhermePenedo,QuentinMalartic,DanielHesslow, -ingsoftheAssociationforComputationalLinguistics: -| Ruxandra | Cojocaru, | | Alessandro | | Cappelli, | Hamza | | | | | | | | -| -------- | --------- | --- | ---------- | --- | --------- | ----- | --- | --- | --- | --- | --- | --- | --- | -Alobeidli, Baptiste Pannier, Ebtesam Almazrouei, ACL-IJCNLP2021,pages3621–3634,Online.Asso- -| | | | | | | | ciationforComputationalLinguistics. | | | | | (Citedonpg. | | -| --------------------- | --- | --- | --- | -------------------- | --- | --- | ----------------------------------- | --- | --- | --- | --- | ----------- | --- | -| andJulienLaunay.2023. | | | | TheRefinedWebdataset | | | | | | | | | | -3) -| for Falcon | | LLM: | outperforming | | curated | corpora | | | | | | | | -| ---------- | --------- | ---- | ------------- | ----- | ------- | -------- | --- | --- | --- | --- | --- | --- | --- | -| with | web data, | and | web data | only. | arXiv | preprint | | | | | | | | -OyvindTafjord,BhavanaDalviMishra,andPeterClark. -| arXiv:2306.01116. | | | (Citedonpg.15) | | | | | | | | | | | -| ----------------- | --- | --- | -------------- | --- | --- | --- | ----------------------------- | --------- | --------- | --- | --------------- | ---- | -------- | -| | | | | | | | 2022. | Entailer: | Answering | | questions | with | faithful | -| | | | | | | | andtruthfulchainsofreasoning. | | | | InProceedingsof | | | -BaolinPeng,MichelGalley,PengchengHe,HaoCheng, -the2022ConferenceonEmpiricalMethodsinNat- -YujiaXie,YuHu,QiuyuanHuang,LarsLiden,Zhou -uralLanguageProcessing,pages2078–2093,Abu -| Yu,WeizhuChen,andJianfengGao.2023. | | | | | | Check | | | | | | | | -| ---------------------------------- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | -Dhabi,UnitedArabEmirates.AssociationforCom- -| yourfactsandtryagain: | | | Improvinglargelanguage | | | | | | | | | | | -| --------------------- | --- | --- | ---------------------- | --- | --- | --- | ---------------------- | --- | --- | ------------- | --- | --- | --- | -| | | | | | | | putationalLinguistics. | | | (Citedonpg.8) | | | | -modelswithexternalknowledgeandautomatedfeed- -| back. | arXivpreprintarXiv:2302.12813. | | | | | (Citedon | | | | | | | | -| ----- | ------------------------------ | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -RomalThoppilan,DanielDeFreitas,JamieHall,Noam -| pg.9) | | | | | | | Shazeer, | Apoorv | Kulshreshtha, | | Heng-Tze | | Cheng, | -| ----- | --- | --- | --- | --- | --- | --- | -------- | ----------- | ------------- | ------ | -------- | --- | ---------- | -| | | | | | | | Alicia | Jin, Taylor | Bos, | Leslie | Baker, | Yu | Du, et al. | -Gabriel Poesia, Alex Polozov, Vu Le, Ashish Tiwari, 2022. LaMDA:Languagemodelsfordialogapplica- -GustavoSoares,ChristopherMeek,andSumitGul- -| | | | | | | | tions. | arXivpreprintarXiv:2201.08239. | | | | | (Citedon | -| ---------- | --- | ------------ | --- | ---------------------- | --- | --- | ------ | ------------------------------ | --- | --- | --- | --- | -------- | -| wani.2022. | | Synchromesh: | | Reliablecodegeneration | | | | | | | | | | -pg.9) -InInternational -frompre-trainedlanguagemodels. -ConferenceonLearningRepresentations. (Citedon Petar Velicˇkovic´, Adrià Puigdomènech Badia, David -| pg.15) | | | | | | | Budden,RazvanPascanu,AndreaBanino,MishaDa- | | | | | | | -| ------ | --- | --- | --- | --- | --- | --- | ------------------------------------------ | --- | --- | --- | --- | --- | --- | -shevskiy,RaiaHadsell,andCharlesBlundell.2022. -GrahamPriest.2008. Anintroductiontonon-classical The clrs algorithmic reasoning benchmark. In In- -logic: From if to is. Cambridge University Press. ternationalConferenceonMachineLearning,pages -| (Citedonpg.10) | | | | | | | 22084–22102.PMLR. | | | (Citedonpg.9) | | | | -| -------------- | --- | --- | --- | --- | --- | --- | ----------------- | --- | --- | ------------- | --- | --- | --- | - -BailinWang,ZiWang,XuezhiWang,YuanCao,RifA MichihiroYasunaga,XinyunChen,YujiaLi,Panupong -Saurous,andYoonKim.2023a. Grammarprompting Pasupat,JureLeskovec,PercyLiang,EdHChi,and -fordomain-specificlanguagegenerationwithlarge DennyZhou.2023. Largelanguagemodelsasana- -languagemodels. arXivpreprintarXiv:2305.19234. logicalreasoners. arXivpreprintarXiv:2310.01714. -| (Citedonpg.8) | | | | | | | (Citedonpg.10) | | | | | | -| ------------- | --- | --- | --- | --- | --- | --- | -------------- | --- | --- | --- | --- | --- | -QingxiangWang,ChadBrown,CezaryKaliszyk,and XiYe,QiaochuChen,IsilDillig,andGregDurrett.2023. -JosefUrban.2020. Explorationofneuralmachine Satisfiability-aidedlanguagemodelsusingdeclara- -translation in autoformalization of mathematics in tive prompting. arXiv preprint arXiv:2305.09656. -| mizar. | In Proceedings | | of the | 9th ACM | SIGPLAN | | (Citedonpg.9) | | | | | | -| ------ | -------------- | --- | ------ | ------- | ------- | --- | ------------- | --- | --- | --- | --- | --- | -InternationalConferenceonCertifiedProgramsand -EricZelikman,YuhuaiWu,JesseMu,andNoahGood- -| Proofs,pages85–98. | | | (Citedonpg.9) | | | | | | | | | | -| ------------------ | ----- | ------ | ------------- | --- | ----- | ------ | ---------- | --------------------------------- | --- | --- | --------- | ---- | -| | | | | | | | man. 2022. | STaR: Bootstrapping | | | reasoning | with | -| | | | | | | | reasoning. | InAdvancesinNeuralInformationPro- | | | | | -| Qingxiang | Wang, | Cezary | Kaliszyk, | and | Josef | Urban. | | | | | | | -2018. First experiments with neural translation of cessing Systems, volume 35, pages 15476–15488. -| informaltoformalmathematics. | | | | InIntelligentCom- | | | (Citedonpg.8) | | | | | | -| ---------------------------- | --- | ---------------------------- | --- | ----------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | -| puterMathematics: | | 11thInternationalConference, | | | | | | | | | | | -CICM2018,Hagenberg,Austria,August13-17,2018, John M Zelle and Raymond J Mooney. 1996. Learn- -ingtoparsedatabasequeriesusinginductivelogic -| Proceedings11,pages255–270.Springer. | | | | | (Citedon | | | | | | | | -| ------------------------------------ | --- | --- | --- | --- | -------- | --- | ------------ | ------------------------------ | --- | --- | --- | --- | -| | | | | | | | programming. | InProceedingsoftheNationalCon- | | | | | -pg.9) -ferenceonArtificialIntelligence,pages1050–1055. -| XuezhiWang,JasonWei,DaleSchuurmans,QuocVLe, | | | | | | | (Citedonpg.8) | | | | | | -| ------------------------------------------- | ------------- | --- | ------------------- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | -| EdHChi, | SharanNarang, | | AakankshaChowdhery, | | | | | | | | | | -andDennyZhou.2023b. Self-consistencyimproves LukeSZettlemoyerandMichaelCollins.2005. Learn- -| | | | | | | | ing to | map sentences | to logical | form: | Structured | | -| ---------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | ------------- | ---------- | ----- | ---------- | --- | -| chainofthoughtreasoninginlanguagemodels. | | | | | | In | | | | | | | -classificationwithprobabilisticcategorialgrammars. -TheEleventhInternationalConferenceonLearning -| | | | | | | | In Proceedings | of the | Twenty-First | | Conference | on | -| ---------------- | --- | ---------------------- | --- | --- | --- | --- | -------------- | ------ | ------------ | --- | ---------- | --- | -| Representations. | | (Citedonpg.1,3,4,8,23) | | | | | | | | | | | -UncertaintyinArtificialIntelligence,UAI’05,page -JasonWei,XuezhiWang,DaleSchuurmans,Maarten 658–666, Arlington, Virginia, USA. AUAI Press. -| Bosma,FeiXia,EdChi,QuocVLe,DennyZhou, | | | | | | | (Citedonpg.8) | | | | | | -| ------------------------------------- | ------------------------------------ | -------- | ------- | --- | -------- | --- | ------------- | ----------- | -------- | ------- | --------- | ----- | -| etal.2022. | Chain-of-thoughtpromptingelicitsrea- | | | | | | | | | | | | -| | | | | | | | Hanlin Zhang, | Ziyang Li, | Jiani | Huang, | Mayur | Naik, | -| soning | in large | language | models. | In | Advances | in | | | | | | | -| | | | | | | | and Eric | Xing. 2022. | Improved | logical | reasoning | | -NeuralInformationProcessingSystems,volume35, -oflanguagemodelsviadifferentiablesymbolicpro- -| pages24824–24837. | | | (Citedonpg.1,4,8) | | | | | | | | | | -| ----------------- | --- | --- | ----------------- | --- | --- | --- | --------- | ------------------------------ | --- | --- | --- | ---- | -| | | | | | | | gramming. | InFirstWorkshoponPre-training: | | | | Per- | -spectives,Pitfalls,andPathsForwardatICML2022. -| NathanielWeirandBenjaminVanDurme.2022. | | | | | | Dy- | | | | | | | -| ---------------------------------------------- | --- | ------ | ------- | --- | ----- | -------- | ------------- | --- | --- | --- | --- | --- | -| namicgenerationofinterpretableinferencerulesin | | | | | | | (Citedonpg.9) | | | | | | -| a neuro-symbolic | | expert | system. | | arXiv | preprint | | | | | | | -HonghuaZhang,MeihuaDang,NanyunPeng,andGuy -| arXiv:2209.07662. | | (Citedonpg.9) | | | | | | | | | | | -| ----------------- | --- | ------------- | --- | --- | --- | --- | --------------------------- | --- | -------------------------- | ------------------- | --- | --- | -| | | | | | | | VandenBroeck.2023a. | | Tractablecontrolforautore- | | | | -| | | | | | | | gressivelanguagegeneration. | | | InInternationalCon- | | | -LionelWong,GabrielGrand,AlexanderKLew,NoahD -ferenceonMachineLearning,pages40932–40945. -| Goodman, | Vikash | K | Mansinghka, | | Jacob Andreas, | | | | | | | | -| ----------------------------------------- | ------- | --- | ----------- | --------------- | -------------- | ----- | ------------ | --------------------------------------- | ------- | ------ | --- | -------- | -| | | | | | | | PMLR. | (Citedonpg.8) | | | | | -| andJoshuaBTenenbaum.2023. | | | | Fromwordmod- | | | | | | | | | -| els to world | models: | | Translating | from | natural | lan- | | | | | | | -| | | | | | | | Kechi Zhang, | Zhuo Li, | Jia Li, | Ge Li, | and | Zhi Jin. | -| guagetotheprobabilisticlanguageofthought. | | | | | | arXiv | | | | | | | -| | | | | | | | 2023b. | Self-edit: Fault-awarecodeeditorforcode | | | | | -| preprintarXiv:2306.12672. | | | | (Citedonpg.8,9) | | | | | | | | | -| | | | | | | | generation. | InProceedingsofthe61stAnnualMeet- | | | | | -ingoftheAssociationforComputationalLinguistics -YuhuaiWu,AlbertQiaochuJiang,WendaLi,Markus -| | | | | | | | (Volume1: | LongPapers),pages769–787,Toronto, | | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | --------------------------------- | --- | --- | --- | --- | -Rabe,CharlesStaats,MatejaJamnik,andChristian -Canada.AssociationforComputationalLinguistics. -| Szegedy. | 2022. | Autoformalization | | | with large | lan- | | | | | | | -| ------------ | ----- | ----------------------------- | --- | --- | ---------- | ---- | ------------- | --- | --- | --- | --- | --- | -| guagemodels. | | InAdvancesinNeuralInformation | | | | | (Citedonpg.9) | | | | | | -ProcessingSystems,volume35,pages32353–32368. -| | | | | | | | Denny Zhou, | Nathanael | Schärli, | Le Hou, | Jason | Wei, | -| --- | --- | --- | --- | --- | --- | --- | ----------- | --------- | -------- | ------- | ----- | ---- | -(Citedonpg.9) -| | | | | | | | Nathan | Scales, Xuezhi | Wang, | Dale | Schuurmans, | | -| --- | --- | --- | --- | --- | --- | --- | ------ | -------------- | ----- | ---- | ----------- | --- | -ClaireCui,OlivierBousquet,QuocVLe,andEdH -| Shunyu Yao, | Dian | Yu, | Jeffrey | Zhao, | Izhak | Shafran, | | | | | | | -| ----------- | ------------ | --- | ------- | --------- | ---------- | -------- | -------------- | ------------- | --------- | ------- | ------- | ------ | -| | | | | | | | Chi. 2023. | Least-to-most | prompting | | enables | com- | -| Thomas | L Griffiths, | | Yuan | Cao, | and | Karthik | | | | | | | -| | | | | | | | plex reasoning | in large | language | models. | | In The | -| Narasimhan. | 2023. | | Tree of | thoughts: | Deliberate | | | | | | | | -EleventhInternationalConferenceonLearningRep- -| problemsolvingwithlargelanguagemodels. | | | | | | arXiv | | | | | | | -| -------------------------------------- | ------- | ----- | ---- | -------------- | ------- | ----- | ------------- | ------------- | --- | --- | --- | --- | -| | | | | | | | resentations. | (Citedonpg.8) | | | | | -| preprintarXiv:2305.10601. | | | | (Citedonpg.10) | | | | | | | | | -| Shunyu Yao, | Jeffrey | Zhao, | Dian | Yu, | Nan Du, | Izhak | | | | | | | -Shafran,KarthikNarasimhan,andYuanCao.2022. -React: Synergizingreasoningandactinginlanguage -| models. | arXivpreprintarXiv:2210.03629. | | | | (Citedon | | | | | | | | -| ------- | ------------------------------ | --- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -pg.9) - -A FutureDirections -dateforthepreviouslymentionedlargernumberof -| | | | | | | | premises | involved | in | a typical | question). | | For the | -| ------ | ----- | ----- | ---------- | --- | ------------ | --- | ---------- | -------- | --- | --------- | ---------- | --- | ------- | -| Of the | error | types | we catalog | in | our analysis | | | | | | | | | -| | | | | | | | StarCoder+ | model, | we | allow | generation | up | until | -LINC, -| of | the key | opportunity | | for improvement | | is | | | | | | | | -| --- | ------- | ----------- | --- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -the8192contextwindowlength,sincethismodel -moreeleganthandlingofnaturalisticlanguageuse. -| | | | | | | | is run locally. | | In either | case, | decoding | is | halted | -| --------- | ------ | -------- | --- | -------- | ------ | ---- | --------------- | --- | --------- | ----- | -------- | --- | ------ | -| While the | errors | observed | | with CoT | result | from | | | | | | | | -earlywheneverthestoptokenispro- -LINC, -| faulty deductive | | inferences, | | in the | case of | | | | | | | | | -| ---------------- | --- | ----------- | --- | ------ | ------- | --- | ------ | ---------------------------------- | --- | --- | --- | --- | --- | -| | | | | | | | duced. | Alllocalexperimentswereexecutedona | | | | | | -allerrorshavebeenlocalizedtothesemanticpars- -clusterequippedwithNVIDIAA100GPUs. -| ingprocedure. | | Thisprocessflowsprimarilyunim- | | | | | | | | | | | | -| ------------- | --- | ------------------------------ | --- | --- | --- | --- | --- | ------- | --- | --- | ---------------------- | --- | --- | -| | | | | | | | GPT | Models: | We | use | the gpt-3.5-turbo-16k- | | | -pededintheevaluationofthesyntheticProofWriter -0613andgpt-4-0613checkpointsoftheGPT-3.5 -dataset,yetleavesroomforimprovementwiththe -(Ouyangetal.,2022)andGPT-4(OpenAI,2023) -| naturalistic | FOLIO. | | In follow-up | work, | we | hope | | | | | | | | -| ------------ | ------ | --- | ------------ | ----- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | -models,respectively,invokingbothmodelsviathe -| to deeply | explore | naturalistic | | evaluation | settings, | | | | | | | | | -| --------- | ------- | ------------ | --- | ---------- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -OpenAIAPI. -aswhendatagetthemostmessyisalsowhereim- -| | | | | | | | StarCoder+: | | StarCoder+(15.5B)6 | | | isaversion | | -| ------------------- | --- | --- | ------------- | --- | ----- | --- | ----------- | --- | ------------------ | --- | --- | ---------- | --- | -| provementsbecomethe | | | mostvaluable. | | Here, | we | | | | | | | | -ofStarCoderBase(Lietal.,2023)whichhasbeen -| propose | three | strategies | for | further | improvement | | | | | | | | | -| ------- | ----- | ---------- | --- | ------- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | -finetunedon600Btokensfromacombinationof(1) -onnaturalisticsettings. -FalconRefinedWeb(Penedoetal.,2023)(filtered -First,innaturalisticcommunication,“obvious” -versionofCommonCrawl),(2)TheStackv1.2(Ko- -informationisoftenleftoutofexplicitproductions, -| | | | | | | | cetkov et | al., 2022), | | and (3) | a Wikipedia | | dataset. | -| --- | --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ------- | ----------- | --- | -------- | -lefttobeinferredinthe“commonground”ofthe -Itsbasemodel,StarCoderBase,isanopen-access -communicativeact(Grice,1975;Stalnaker,2002). -modelwithaGPT-2architectureusingmulti-query -| Implicit | premise | rediscovery | | through | controlled | | | | | | | | | -| -------- | ------- | ----------- | --- | ------- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | -attention(Shazeer,2019)andfill-in-the-middleob- -explorationonthelogicalneighborhoodoftheex- -| | | | | | | | jective(Bavarianetal.,2022). | | | | StarCoderBasehas | | | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | ---------------- | --- | --- | -istingexplicitpremisespromisestobeapowerful -| | | | | | | | a8192contextwindowandistrainedon1T | | | | | | code | -| --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | ---- | -strategyforimprovingperformanceinunderspeci- -tokensofpermissivelylicensedtextfromGitHub -fiedsettings. -across80programminglanguages(Lietal.,2023). -| Second, | while | a | number | of samples | are | lost to | | | | | | | | -| ------- | ----- | --- | ------ | ---------- | --- | ------- | ------ | ---------- | ------- | --- | ---------------- | --- | --- | -| | | | | | | | We use | StarCoder+ | instead | | of StarCoderBase | | be- | -syntaxerrors,recentworkhasproposedrestricting -| | | | | | | | cause it | is finetuned | | on natural | language, | | which | -| ------------ | --- | ----- | ----- | ------ | ---------- | --- | ------------------------------------- | ------------ | --- | ---------- | --------- | --- | ----- | -| the sampling | | space | of an | LLM to | that which | is | | | | | | | | -| | | | | | | | shouldimprovetheperformanceonourtask. | | | | | | We | -consistentwithtermexpansionsinacontext-free- -runStarCoder+withbf16precisiontoreduceits -| grammar(CFG)(Poesiaetal.,2022). | | | | | Doingsoin | | | | | | | | | -| ------------------------------- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | -memoryfootprint. -thissettingwouldeliminateallsyntaxerrors. -Third, sometimes, the translation process to C FOLIODatasetPreprocessing -FOLislossy,throwingawayvaluableinformation -| | | | | | | | We use | the publicly | | available | FOLIO | dataset | on | -| ------- | ------ | -------- | --------- | --- | ------- | --- | ------ | ------------ | --- | --------- | ----- | ------- | --- | -| present | in the | original | sentence. | We | propose | im- | | | | | | | | -https://github.com/Yale-LILY/FOLIO. -| provingthefaithfulnessofFOLtranslationsbyask- | | | | | | | | | | | | | We | -| --------------------------------------------- | --- | --- | --- | --- | --- | --- | ------ | -------------- | --- | ------- | ---- | ------------ | --- | -| | | | | | | | choose | representative | | samples | from | the training | | -ingtheLLMtotranslatetheFOLbacktonatural -splitofthedatasettobeourfew-shotexamplesand -| languageandcomparingwiththeoriginal. | | | | | Forward | | | | | | | | | -| ------------------------------------ | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | -translationsthatrankhighlywhenback-translated use the validation split of the dataset in our eval- -| | | | | | | | uation. | The testing | split | is | not publicly | available. | | -| --- | --- | --- | --- | --- | --- | --- | ------- | ----------- | ----- | --- | ------------ | ---------- | --- | -wouldbethosewhichhaveeffectivelycapturedthe -Theoriginaldatasethas204validationexamples. -intricaciesofaparticularsentence’ssemantics. -However,wediscoveredthatthereareerrorsin22 -| Overall, | we | believe | that | shifting | to | evalua- | | | | | | | | -| -------- | ---- | ------------ | ---- | --------- | ------------ | ------- | ------------- | --- | -------------------------- | --------- | --- | ------------- | --- | -| | | | | | | | ofthesamples. | | Weremovethesesamplesforour | | | | | -| tions on | more | naturalistic | | datasets, | and incorpo- | | | | | | | | | -| | | | | | | | evaluation | and | use the | remaining | | 182 examples. | | -ratingstrategiessuchasthosepresentedhere,will -Theerrorsasfollows: -helppavethepathforwardforneurosymbolicap- -proachestoformalreasoning. -• In4samples,oneormoreofthegroundtruth -| | | | | | | | FOL | expressions | | have | unbalanced | parenthe- | | -| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ---- | ---------- | --------- | --- | -B ModelDetailsandParameters -ses(samples3,109,110,111). -| We use | a decoding | | temperature | T | = 0.8 | for all | | | | | | | | -| ------ | ---------- | --- | ----------- | --- | ----- | ------- | --- | --- | --- | --- | --- | --- | --- | -• In8samples,thelabelobtainedbyexecuting -models. ForGPT-3.5andGPT-4,welimitthemax- -| | | | | | | | the | ground-truth | | FOL | expressions | does | not | -| ----------- | --- | --------- | --- | ----------- | ------- | --- | --- | ------------ | --- | --- | ----------- | ---- | --- | -| imum number | | of tokens | | to generate | to 1024 | for | | | | | | | | -FOLIO and 4096 for ProofWriter (to accommo- 6https://huggingface.co/bigcode/starcoderplus - -main/data/v0.0/folio-train.jsonl. -| match | the | provided | ground | | truth label. | We | | | | | | | We do | -| ----- | --- | -------- | ------ | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | ----- | -double-checked this, first by executing the notoptimizeforthechoiceoffew-shotexamples, -FOLexpressionsthroughProver9andsecond andthisistheonlysetofexamplesweevaluated -bycheckingitmanually. (samples6,28,30, with, so it is likely that there exist better choices -48,113,115,139,140). forfew-shotexamplesthatwouldleadtoimproved -performanceacrosstheboard. -• In10samples,thenumberofpremisesdoes -not match the number of FOL expressions D.1 FOLIO,1-shot(baseline) -| (samples | | 10, 11, | 12, 88, | 106, | 107, | 108, 174, | | | | | | | | -| -------- | --- | ------- | ------- | ---- | ---- | --------- | --- | --------- | ---------------- | --- | ----- | ----- | --- | -| | | | | | | | The | following | is a first-order | | logic | (FOL) | | -175,176). -(cid:44)→ problem. -| | | | | | | | The | problem | is to determine | | whether | the | | -| --- | --- | --- | --- | --- | --- | --- | --- | ------- | --------------- | --- | ------- | --- | --- | -The sample numbers above refer to the (cid:44)→ conclusion follows from the premises. -| | | | | | | | The | premises | are given | in the | form | of a set | of | -| ---------- | --- | --- | ---------- | --- | ------------ | --- | --- | --------------------- | --------- | ---------------- | ---- | -------- | --- | -| line index | in | the | validation | | file located | at | | | | | | | | -| | | | | | | | | (cid:44)→ first-order | | logic sentences. | | | | -https://github.com/Yale-LILY/FOLIO/blob/ -| | | | | | | | The | conclusion | is given | in | the form | of a single | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | -------- | --- | -------- | ----------- | --- | -main/data/v0.0/folio-validation.jsonl. (cid:44)→ first-order logic sentence. -| | | | | | | | The | task is | to evaluate | the | conclusion | as | 'True', | -| --- | --- | --- | --- | --- | --- | --- | --- | ------------------ | ----------- | ----------- | ---------- | --------- | ------- | -| | | | | | | | | (cid:44)→ 'False', | or | 'Uncertain' | | given the | | -D FOLIOFew-ShotPrompts -(cid:44)→ premises. -| The methodologies | | | we investigate | | do | not re- | | | | | | | | -| ----------------- | ---------- | ---------- | --------------- | --- | --------- | --------- | ---------- | ----------- | --------------- | --- | --------------------- | --- | --- | -| quire any | finetuning | on | domain-specific | | | data. In- | | | | | | | | -| | | | | | | | All | dispensable | things | are | environment-friendly. | | | -| stead, we | use | in-context | learning | | (ICL) | with pre- | | | | | | | | -| | | | | | | | All | woodware | is dispensable. | | | | | -| trained | models. | We | prompt | | the model | with | | | | | | | | -| | | | | | | | All | paper | is woodware. | | | | | -a set of instructions and 1-8 ICL examples, No good things are bad. -| | | | | | | | All | environment-friendly | | things | are | good. | | -| ----------- | -------------------------------- | ----------- | ---------- | --- | ----------- | ----- | ----------- | -------------------- | --------- | ------ | ----- | ------------ | --- | -| which | adhere | to a | structured | | text format | de- | | | | | | | | -| | | | | | | | A worksheet | | is either | paper | or is | environment- | | -| signed | to scaffold | generations | | | and ease | post- | | (cid:44)→ friendly. | | | | | | -| processing. | Inparticular,webegineachICLexam- | | | | | | | | | | | | | - -| ple with | each | of the | NL premises | | wrapped | in an | | | | | | | | -| -------- | ---- | ------ | ----------- | --- | ------- | ----- | ----------- | --- | ------------------- | --- | --- | --- | --- | -| | | | | | | | A worksheet | | is not dispensable. | | | | | -... -| HTML-style | | tag | | | | | | | | | | | | -| ---------- | --- | ------ | ---------- | --- | ------- | --- | ------------- | --- | --- | --- | --- | --- | --- | -| followed | by | the NL | conclusion | | wrapped | in | | | | | | | | -Uncertain -| .... | | | | | The | requi- | | | | | | | | -| ----------------------------- | --- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | - -siteevaluationstepsforeachevaluationparadigm - -arethenoutlinedinasubsequentsectionwrapped ...premises for sample here, one premise per -| .... | | | | | Followingthein- | | | line | | | | | | -| ------------------------- | --- | --- | --- | --- | --------------- | --- | --- | ---- | --- | --- | --- | --- | --- | -(cid:44)→ - -clusionofICLexamples,atestexampleisadded, - -withtheandsections. -| | | | | | | | ...conclusion | | for sample | here | | | | -| -------------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | --- | ---------- | ---- | --- | --- | --- | -| Then,thetagisopened,andtheLM | | | | | | | | | | | | | | - -| is allowed | to | proceed | with | causal | generation | un- | | | | | | | | -| ---------- | --- | ------- | ---- | ------ | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | - -| til the | | tag | is | generated. | Upon | gen- | | | | | | | | -| ------- | --- | --- | --- | ---------- | ---- | ---- | --- | --- | --------- | --- | ---- | --- | --- | -| | | | | | | | | | Listing1: | | todo | | | - -| eration | of this | stop token, | | the | | block | | | | | | | | -| ------- | ------- | ----------- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | -issegmentedforpost-processingaccordingtothe -D.2 FOLIO,1-shot(scratchpad) -| | | | ({naïve, | | scratchpad, | | | | | | | | | -| --------------------------------------- | --------- | --------- | ----------------- | -------- | ----------- | -------- | --- | -------------------- | ---------------- | ---------------- | -------- | --------- | --- | -| method | being | evaluated | | | | | | | | | | | | -| chain-of-thought, | | | neuro-symbolic}). | | | | | | | | | | | -| | | | | | | | The | following | is a first-order | | logic | (FOL) | | -| Forthefew-shotexamples,weusesamplesfrom | | | | | | | | (cid:44)→ problem. | | | | | | -| | | | | | | | The | problem | is to determine | | whether | the | | -| the publicly | available | | FOLIO | training | | set. We | | | | | | | | -| | | | | | | | | (cid:44)→ conclusion | follows | | from the | premises. | | -| | | | | | | | The | premises | are given | in the | form | of a set | of | -| select a | set of | diverse | samples | that | are | balanced | | | | | | | | -| | | | | | | | | first-order | | logic sentences. | | | | -(cid:44)→ -across labels. Since the FOLIO training set does The conclusion is given in the form of a single -notcomewithFOLexpressionsfortheconclusions first-order logic sentence. -(cid:44)→ -| | | | | | | | The | task is | to translate | each | of | the premises | | -| -------- | ---------- | --- | -------- | --- | -------- | --- | --- | ------- | ------------ | ---- | --- | ------------ | --- | -| or chain | of thought | | prompts, | we | manually | add | | | | | | | | -| | | | | | | | | and | conclusions | into | FOL | expressions, | and | -(cid:44)→ -bothforeachsample. Forthek-shotsetting(k<8), then to evaluate the conclusion as 'True -(cid:44)→ -weusethefirstk samplesfromthefollowinglist ', 'False', or 'Uncertain' given the -(cid:44)→ -premises. -| of sample | indices: | 126, | 24, | 61, | 276, | 149, 262, | | (cid:44)→ | | | | | | -| ---------------------------------------- | -------- | ------ | -------- | --- | ------ | ----------- | ---------- | --------- | --- | --- | --- | --- | --- | -| 264, 684. | Here, | sample | i refers | | to the | ith line in | | | | | | | | -| https://github.com/Yale-LILY/FOLIO/blob/ | | | | | | | | | | | | | | - -All dispensable things are environment-friendly. -All woodware is dispensable. Let's think step by step. We want to evaluate if -All paper is woodware. (cid:44)→ a worksheet is not dispensable. From -No good things are bad. (cid:44)→ premise 6, we know that a worksheet is -All environment-friendly things are good. (cid:44)→ either paper or is environment-friendly. -A worksheet is either paper or is environment- (cid:44)→ If it is paper, then from premise 3, a -(cid:44)→ friendly. (cid:44)→ worksheet is woodware, and from premise - (cid:44)→ 2, a worksheet is dispensable. If it is - (cid:44)→ environment-friendly, we know it is good -A worksheet is not dispensable. (cid:44)→ from premise 5, but we know nothing about - (cid:44)→ whether it is dispensable. Therefore, we - (cid:44)→ don't know if a worksheet is dispensible -TEXT: All dispensable things are environment- (cid:44)→ or not, so the statement is uncertain. -(cid:44)→ friendly. ANSWER: Uncertain -FOL: all x. (Dispensable(x) -> -(cid:44)→ EnvironmentFriendly(x)) -TEXT: All woodware is dispensable. -FOL: all x. (Woodware(x) -> Dispensable(x)) ...premises for sample here, one premise per -TEXT: All paper is woodware. (cid:44)→ line -FOL: all x. (Paper(x) -> Woodware(x)) -TEXT: No good things are bad. -FOL: all x. (Good(x) -> -Bad(x)) ...conclusion for sample here -TEXT: All environment-friendly things are good. -FOL: all x. (EnvironmentFriendly(x) -> Good(x)) -TEXT: A worksheet is either paper or is -(cid:44)→ environment-friendly. -FOL: ((Paper(Worksheet) & -EnvironmentFriendly( D.4 FOLIO,1-shot(neurosymbolic) -(cid:44)→ Worksheet)) | (-Paper(Worksheet) & -(cid:44)→ EnvironmentFriendly(Worksheet))) The following is a first-order logic (FOL) -TEXT: A worksheet is not dispensable. (cid:44)→ problem. -FOL: -Dispensable(Worksheet) The problem is to determine whether the -ANSWER: Uncertain (cid:44)→ conclusion follows from the premises. - The premises are given in the form of a set of -(cid:44)→ first-order logic sentences. - The conclusion is given in the form of a single -...premises for sample here, one premise per (cid:44)→ first-order logic sentence. -(cid:44)→ line The task is to translate each of the premises - (cid:44)→ and conclusions into FOL expressions, so - (cid:44)→ that the expressions can be evaluated by -...conclusion for sample here (cid:44)→ a theorem solver to determine whether the - (cid:44)→ conclusion follows from the premises. - Expressions should be adhere to the format of -(cid:44)→ the Python NLTK package logic module. -D.3 FOLIO,1-shot(chainofthought) - -The following is a first-order logic (FOL) All dispensable things are environment-friendly. -(cid:44)→ problem. All woodware is dispensable. -The problem is to determine whether the All paper is woodware. -(cid:44)→ conclusion follows from the premises. No good things are bad. -The premises are given in the form of a set of All environment-friendly things are good. -(cid:44)→ first-order logic sentences. A worksheet is either paper or is environment- -The conclusion is given in the form of a single (cid:44)→ friendly. -(cid:44)→ first-order logic sentence. -The task is to translate each of the premises -(cid:44)→ and conclusions into FOL expressions, A worksheet is not dispensable. - - - TEXT: All dispensable things are environment- -All dispensable things are environment-friendly. (cid:44)→ friendly. -All woodware is dispensable. FOL: all x. (Dispensable(x) -> -All paper is woodware. (cid:44)→ EnvironmentFriendly(x)) -No good things are bad. TEXT: All woodware is dispensable. -All environment-friendly things are good. FOL: all x. (Woodware(x) -> Dispensable(x)) -A worksheet is either paper or is environment- TEXT: All paper is woodware. -(cid:44)→ friendly. FOL: all x. (Paper(x) -> Woodware(x)) - TEXT: No good things are bad. - FOL: all x. (Good(x) -> -Bad(x)) -A worksheet is not dispensable. TEXT: All environment-friendly things are good. - FOL: all x. (EnvironmentFriendly(x) -> Good(x)) - -TEXT: A worksheet is either paper or is (Incorrect) Translation 2 (StarCoderPlus): (( -(cid:44)→ environment-friendly. (cid:44)→ Rabbit(Animal) & -Squirrel(Animal)) | (- -FOL: ((Paper(Worksheet) & -EnvironmentFriendly( (cid:44)→ Rabbit(Animal) & Squirrel(Animal))) -(cid:44)→ Worksheet)) | (-Paper(Worksheet) & (Incorrect) Translation 3 (GPT-4): all x. (( -(cid:44)→ EnvironmentFriendly(Worksheet))) (cid:44)→ Animal(x) & Rabbit(x)) | (Animal(x) & -TEXT: A worksheet is not dispensable. (cid:44)→ Squirrel(x))) -FOL: -Dispensable(Worksheet) - Premise: a person either studys or teaches -(Ground Truth) Translation: all x. (Study(x) | - (cid:44)→ Teaches(x)) -...premises for sample here, one premise per (Incorrect) Translation 1 (StarCoderPlus): -(cid:44)→ line (cid:44)→ Studys(Person) | Teaches(Person) - (Incorrect) Translation 2 (StarCoderPlus): (( - (cid:44)→ Study(Person) & -Teach(Person)) | (-Study -...conclusion for sample here (cid:44)→ (Person) & Teach(Person))) - (Correct) Translation 3 (GPT-4): all x. (Studies - (cid:44)→ (x) | Teaches(x)) -Premise: A man is either kind or evil. -E FOLIOErrorAnalysis (Ground Truth) Translation: all x. (Kind(x) & - -(cid:44)→ Evil(x)) | (-Kind(x) & Evil(x)) -(Incorrect) Translation 1 (GPT-3.5): ((Man(x) & - -E.1 Ambiguityof“Either”statements -(cid:44)→ Kind(x)) | (-Man(x) & Evil(x))) -Dependingonthecontext,thephrase“eitherxor (Incorrect) Translation 2 (StarCoderPlus): Kind( -y”couldmeanx XOR y,x OR y,orbeambiguous. (cid:44)→ AMan) | Evil(AMan) -(Incorrect) Translation 3 (StarCoderPlus): (Kind -Throughoutourexperiments,wefoundthatmodels (cid:44)→ (x) | Evil(x)) -hadmanycreativelyincorrectwaysoftranslating -Premise: Ben is either from The Simpsons or -these statements. One reoccurring error was that -(cid:44)→ funny. -statementsthatclearlyintendedx XOR y(suchas, -(Ground Truth) Translation: (Simpsons(Ben) & - -“an animal is either a rabbit or a squirrel”) were (cid:44)→ Funny(Ben)) | (-Simpsons(Ben) & Funny(Ben -translated into x OR y. We tried to account for (cid:44)→ )) -(Correct) Translation 1 (StarCoderPlus): (( -this into account by including multiple samples (cid:44)→ Simpsons(Ben) & -Funny(Ben)) | (-Simpsons -with this construct in the few shot examples (see (cid:44)→ (Ben) & Funny(Ben))) -(Incorrect) Translation 2 (GPT-3.5): ( -Sec. D.4). However, the models still handle this -(cid:44)→ FromTheSimpsons(Ben) | Funny(Ben)) -constructinconsistentlyandincorrectly. (Incorrect) Translation 3 (GPT-4): -Inaddition,wefindthatthroughouttheFOLIO (cid:44)→ FromTheSimpsons(Ben) | Funny(Ben) -dataset,bymatchingthenaturallanguagepremises -totheFOLpremises,wefindnoconsistentorpre- E.2 GPT-4LINCFailureModes -dictable pattern as to how “either x or y” state- -L1: FOL fails to capture implicit information -ments are translated. For example, “an animal is -notmentionedinthepremises. Threeexamples -either a rabbi or a squirrel” is translated as all -oferrorsfromtheFOLIOvalidationsetareshown -x. Rabbit(x) | Squirrel(x),whilewebelieve -below. ThefirsttwooccurringinbothGPT-3.5and -thisinstanceshouldclearlybeXOR.Therefore,we -GPT-4,andthelatteronlyoccursinGPT-3.5and -believethatsomeofthesesamplesareinherently -interestingly, is correct in GPT-4. In Example 1, -ambiguousormalformed. -tomakethecorrectconclusion,wemustencodein -To highlight model behavior on these exam- -FOLthatHarryisaperson(Person(Harry))and -ples, four representative examples from the FO- -that Walden is a book (Book("Walden")). Harry -LIO validation set are shown below; examples -being a person is implicit, but “Walden” being a -have multiple translations because we used tem- -bookisexplicitlymentionedinpremise4butfails -peratureT = 0.8. Here,Correct/Incorrectindicate -to be explicitly encoded by the model. In Exam- -whether the translations match the ground truth -ple 2, we must encode that KiKi is an animal to -(which doesn’t necessarily match how we would -make the correct deduction. One can argue that -translateit). -this example is ambiguous, but from the context, -Premise: an animal is either a rabbit or a -mostwouldmakethisinference. InExample3,we -(cid:44)→ squirrel -(Ground Truth) Translation: all x. (Rabbit(x) | needaclausethatsaysLGAandLGAarethesame -(cid:44)→ Squirrel(x)) airport(SameAirport(LGA, LGA)). -(Correct) Translation 1 (GPT-3.5): all x. ( -(cid:44)→ Animal(x) -> (Rabbit(x) | Squirrel(x))) Example 1 (GPT-4) - -| Premises: | | | | | | choice | of | representation. | | We | show | four | illus- | -| --------- | --- | --- | --- | --- | --- | ------ | --- | --------------- | --- | --- | ---- | ---- | ------ | -Books contain tons of knowledge. trative examples: the first two examples occur -| When a person | reads | a book, | that | person | gains | | | | | | | | | -| ------------- | ----- | ------- | ---- | ------ | ----- | ------- | ------- | --- | ---------- | --- | ----- | -------- | --- | -| | | | | | | in both | GPT-3.5 | | and GPT-4, | | while | the last | two | -(cid:44)→ knowledge. -If a person gains knowledge, they become smarter. examples only occur in GPT-3.5 and are correct -| (cid:44)→ | | | | | | whenusingGPT-4. | | | Thefirstexampleisdescribed | | | | | -| ---------- | -------- | -------- | --- | ----- | -------- | --------------- | ---- | ----- | -------------------------- | --- | ----- | ---- | ---- | -| Harry read | the book | "Walden" | by | Henry | Thoreau. | | | | | | | | | -| | | | | | | in the | main | text. | In example | | 2, we | know | that | -Conclusion (Ground Truth: True) Barutin Cove is a cove, but this lost information -| Harry is | smarter than | before. | | | | | | | | | | | | -| -------- | ------------ | ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -isnotencodedintheFOL,sowecannotconclude -| FOLs: | | | | | | Cove(BarutinCove), | | | although | | this is | crucial | in- | -| --------------- | ------------- | ----------- | ---------- | ----------- | --------- | ------------------------------- | --- | --- | -------- | --- | ----------- | ------- | --- | -| all x. (Book(x) | -> | Contains(x, | | Knowledge)) | | | | | | | | | | -| | | | | | | formationformakingthededuction. | | | | | InExample3, | | | -| all x. all | y. (Person(x) | | & Reads(x, | y) | & Book(y) | | | | | | | | | --> Gains(x, Knowledge)) wemustdeducethatTom’slicenseplateisalicense -(cid:44)→ -all x. (Gains(x, Knowledge) -> Smarter(x)) plate(FromIstanbul(TomPlate)),whichisobvi- -| Reads(Harry, | Walden) | | | | | | | | | | | | | -| ------------ | ------- | --- | --- | --- | --- | --- | ------ | ----------- | --- | --- | ------ | ------ | ----- | -| | | | | | | ous | but is | not encoded | by | the | model. | In the | final | -Conclusion (Prover9 Deduction: Uncertain) example,theideathatanobjecthasfoursidesisex- -| Smarter(Harry) | | | | | | pressedbothasFourSided(x)andFourSides(x), | | | | | | | | -| -------------- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -sothemodelisunabletomakethecorrectdeduc- -| Example 2 | (GPT-4) | | | | | | | | | | | | | -| ------------ | ----------- | ---- | ---------- | --- | -------- | ------- | --- | ------- | --- | --- | --- | --- | --- | -| Premises: | | | | | | tion. | | | | | | | | -| If an animal | barks, | then | Jane does | not | like it. | | | | | | | | | -| All dogs | are animals | that | bark. | | | | | | | | | | | -| | | | | | | Example | 1 | (GPT-4) | | | | | | -| If an animal | jumps, | then | Jane likes | it. | | | | | | | | | | -Premises: -| If an animal | has legs, | then | it | jumps. | | | | | | | | | | -| --------------- | --------------- | ------ | --------- | ------ | --------- | ---------- | -------------------- | ---------- | ---------- | -------- | ----------- | -------- | --- | -| | | | | | | Heinrich | Scmidt | was | a Nazi | German | politician. | | | -| If an animal | is terricolous, | | then | it | has legs. | | | | | | | | | -| | | | | | | Heeinrich | | Scmidt was | also | a member | of | the | | -| If KiKi neither | jumps | nor | has legs, | then | KiKi is | | | | | | | | | -| | | | | | | | (cid:44)→ Prussian | State | Parliament | | and | the Nazi | | -| | terricolous | or has | legs. | | | | | | | | | | | -| (cid:44)→ | | | | | | | (cid:44)→ Reichstag. | | | | | | | -| Conclusion | (Ground | Truth: | True) | | | | | | | | | | | -| | | | | | | Conclusion | | (Ground | Truth: | True): | | | | -| KiKi neither | barks | nor is | a dog. | | | | | | | | | | | -| | | | | | | Heinrich | Schmidt | was | either | German | or | Russian. | | -FOLs: -FOLs: -| all x. (Animal(x) | & | Barks(x) | -> | NotLike(Jane, | x) | | | | | | | | | -| ----------------- | --- | -------- | --- | ------------- | --- | --------------------------------- | --- | --- | --- | --- | ------- | --- | --- | -| | | | | | | GermanPolitician(HeinrichSchmidt) | | | | | & Nazi( | | | -) -| (cid:44)→ | | | | | | | (cid:44)→ HeinrichSchmidt) | | | | | | | -| ----------------- | ------------ | -------- | ----------- | ---------- | --- | ---------------------------------------------- | -------------------------- | --- | --- | --- | --- | --- | --- | -| all x. (Dog(x) | -> Animal(x) | | & Barks(x)) | | | | | | | | | | | -| | | | | | | MemberPrussianStateParliament(HeinrichSchmidt) | | | | | | | & | -| all x. (Animal(x) | & | Jumps(x) | -> | Like(Jane, | x)) | | | | | | | | | -(cid:44)→ MemberNaziReichstag(HeinrichSchmidt) -| all x. (Animal(x) | & | HasLegs(x) | | -> Jumps(x)) | | | | | | | | | | -| ----------------- | --- | -------------- | --- | ------------ | ---------- | ---------- | --- | -------- | ---------- | --- | ----------- | --- | --- | -| all x. (Animal(x) | & | Terricolous(x) | | -> | HasLegs(x) | | | | | | | | | -| | | | | | | Conclusion | | (Prover9 | Deduction: | | Uncertain): | | | -) -| (cid:44)→ | | | | | | German(HeinrichSchmidt) | | | | | Russian( | | | | -| ------------- | ----------------- | --- | --- | ---------------- | --- | ----------------------- | --- | --- | --- | ---------- | --- | --- | --- | -| (-Jumps(KiKi) | & -HasLegs(KiKi)) | | | -> (Terricolous( | | | | | | | | | | -(cid:44)→ HeinrichSchmidt) -| KiKi) | | HasLegs(KiKi)) | | | | | | | | | | | | | -| ----- | ---------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(cid:44)→ -| | | | | | | Example | 2 | (GPT-4) | | | | | | -| ---------- | -------- | ---------- | --- | ---------- | --- | ------- | --- | ------- | --- | --- | --- | --- | --- | -| Conclusion | (Prover9 | Deduction: | | Uncertain) | | | | | | | | | | -Premises: -| -Barks(KiKi) | & -Dog(KiKi) | | | | | | | | | | | | | -| ------------ | ------------ | --- | --- | --- | --- | ------- | -------------------- | -------- | ------------- | --- | --------- | ------- | --- | -| | | | | | | Barutin | Cove | is named | after | the | Bulgarian | | | -| | | | | | | | (cid:44)→ settlement | | of Barutin. | | | | | -| Example 3 | (GPT-3.5): | | | | | | | | | | | | | -| | | | | | | Barutin | Cove | is on | the southwest | | coast | of Snow | | -Premises: -(cid:44)→ Island. -| Susan flies | to LGA | airport. | | | | | | | | | | | | -| ------------- | ----------- | -------- | ------- | --- | -------- | --- | ------------------- | -------- | ------- | ------- | --------- | ------- | --- | -| | | | | | | The | South | Shetland | Islands | include | Snow | Island, | | -| The departure | and arrival | | can not | be | the same | | | | | | | | | -| | | | | | | | (cid:44)→ Greenwich | | Island, | and | Deception | Island. | | -airport. -| (cid:44)→ | | | | | | The | South | Shetland | Islands | are | considered | part | | -| ------------ | -------- | -------- | ------ | --- | --- | ---------- | ------------ | ----------- | -------- | --------- | ---------- | ----- | ---- | -| John flies | from LGA | airport. | | | | | | | | | | | | -| | | | | | | | (cid:44)→ of | Antarctica. | | | | | | -| | | | | | | If A | is located | in | B and | B is | located | in C, | then | -| Conclusion | (Ground | Truth: | False) | | | | | | | | | | | -| | | | | | | | (cid:44)→ A | is located | in | C. | | | | -| Susan flies | from LGA | airport. | | | | | | | | | | | | -| | | | | | | Conclusion | | (Ground | Truth: | True) | | | | -| FOLs: | | | | | | There | is at | least | one cove | in | Antarctica | named | | -| Flies(Susan, | LGA) | | | | | | after | a place | in | Bulgaria. | | | | -(cid:44)→ -| all x. all | y. (Departure(x) | | & Arrival(y) | | & - | | | | | | | | | -| ------------------------ | ---------------- | --- | ------------ | --- | --- | ----- | --- | --- | --- | --- | --- | --- | --- | -| (cid:44)→ SameAirport(x, | | y)) | | | | FOLs: | | | | | | | | -Flies(John, LGA) NamedAfter(BarutinCove, Barutin) & In(Barutin, -Bulgaria) -(cid:44)→ -Conclusion (Prover9 Deduction: Uncertain): On(BarutinCove, SnowIsland) -| Flies(Susan, | LGA) | | | | | In(SnowIsland, | | SouthShetlandIslands) | | | | & In( | | -| ------------ | ---- | --- | --- | --- | --- | -------------- | ---------------- | --------------------- | --- | --------------------- | --- | ----- | --- | -| | | | | | | | GreenwichIsland, | | | SouthShetlandIslands) | | | & | -(cid:44)→ -| | | | | | | | In(DeceptionIsland, | | | SouthShetlandIslands) | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | --------------------- | --- | --- | --- | -(cid:44)→ -L2: FOL fails to capture information ex- In(SouthShetlandIslands, Antarctica) -plicitly mentioned in the premises due to the all x. all y. all z. ((In(x, y) & In(y, z)) -> - -| | (cid:44)→ In(x, | z)) | | | | | Premises: | | | | | | | -| --- | --------------- | --- | --- | --- | --- | --- | --------- | --- | ------------ | --- | ------- | ------- | ------- | -| | | | | | | | There | are | four seasons | in | a year: | Spring, | Summer, | -Conclusion (Prover9 Deduction: Uncertain) (cid:44)→ Fall, and Winter. -exists x. (Cove(x) & In(x, Antarctica) & exists All students who want to have a long vacation -(cid:44)→ y. (In(y, Bulgaria) & NamedAfter(x, y))) (cid:44)→ love summer the most. -| Example | 3 | (GPT-3.5) | | | | | FOLs: | | | | | | | -| --------- | --- | --------- | --- | --- | --- | --- | ----- | ----------- | --- | ---------- | --- | ----------- | ------ | -| Premises: | | | | | | | all | x. (Year(x) | -> | (Spring(x) | | | Summer(x) | | Fall | -All vehicle registration plates in Istanbul (cid:44)→ (x) | Winter(x))) -(cid:44)→ begin with the number 34. all x. (Student(x) & WantsLongVacation(x) -> -Plates that do not begin with the number 34 are Loves(x, Summer)) -(cid:44)→ -| | not | from | Istanbul. | | | | | | | | | | | -| --- | --- | ---- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(cid:44)→ -| Joe's | vehicle | registration | | plate | is from | | Prover9 | Error: | | | | | | -| ----- | --------- | ------------ | --- | ----- | ------- | --- | ------- | --------- | ------- | --- | ---- | ------------- | --- | -| | Istanbul. | | | | | | The | following | symbols | are | used | with multiple | | -(cid:44)→ -Tom's license plate begins with the number 35. arities: Summer/1, Summer/0. -(cid:44)→ -| If a | license | plate | begins | with | the number | 35, | | | | | | | | -| ---- | ------- | ------- | ------ | ----- | ---------- | ------ | ------- | --- | --- | --- | --- | --- | --- | -| | then | it does | not | begin | with the | number | Example | 2 | | | | | | -(cid:44)→ -| | 34. | | | | | | Premises: | | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | -(cid:44)→ -| | | | | | | | Badults | is | a British | Sitcom | series, | starring | | -| ---------- | --- | ------- | ------ | ------- | --- | --- | ------- | ------- | --------- | -------- | ------- | -------- | --- | -| Conclusion | | (Ground | Truth: | False): | | | | members | of | Pappy's. | | | | -(cid:44)→ -Tom's license plate is from Istanbul. Badults piloted in July 2013 on BBC Three. -| FOLs: | | | | | | | FOLs: | | | | | | | -| ----- | --- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | -all x. (VehicleRegistrationPlate(x) -> exists x. (Badults(x) & BritishSitcom(x) & -| | BeginsWith(x, | | 34)) | | | | | StarringMembersOf(x, | | | Pappys)) | | | -| --- | ------------- | --- | ---- | --- | --- | --- | --- | -------------------- | --- | --- | -------- | --- | --- | -| | (cid:44)→ | | | | | | | (cid:44)→ | | | | | | -all x. (-BeginsWith(x, 34) -> -FromIstanbul(x)) Piloted(Badults, July2013, BBCThree) -FromIstanbul(JoePlate) -| BeginsWith(TomPlate, | | | 35) | | | | Prover9 | Error: | | | | | | -| -------------------- | --- | --- | --- | --- | --- | --- | ------- | ------ | --- | --- | --- | --- | --- | -all x. (LicensePlate(x) & BeginsWith(x, 35) -> - The following symbols are used with multiple -| | BeginsWith(x, | | 34)) | | | | | arities: | | Badults/1, | Badults/0. | | | -| ---------- | ------------- | -------- | ---------- | --- | ----------- | --- | ----- | --------- | ----- | ---------- | ---------- | ------ | ----------- | -| | (cid:44)→ | | | | | | | (cid:44)→ | | | | | | -| Conclusion | | (Prover9 | Deduction: | | Uncertain): | | | | | | | | | -| | | | | | | | While | rarer, | there | are | also | syntax | errors. For | -FromIstanbul(TomPlate) -illustration,twoexamplesfromGPT-4areshown -| Example | 4 | (GPT-3.5) | | | | | below: | | | | | | | -| ------- | --- | --------- | --- | --- | --- | --- | ------ | --- | --- | --- | --- | --- | --- | -Premises: -| | | | | | | | Premise: | Ailton | Silva | foes | not | play for | a | -| ---------- | ---------- | ----------- | ----------- | ------- | --- | --- | -------- | ------------ | ------------------- | ------- | -------------- | ------------- | --- | -| All | squares | have | four sides. | | | | | | | | | | | -| | | | | | | | | football | | club. | | | | -| All | four-sided | things | are | shapes. | | | | (cid:44)→ | | | | | | -| | | | | | | | FOL: | -Exists | x. (FootballClub(x) | | | & PlaysFor( | | -| | | | | | | | | AiltonSilva, | | x)) | | | | -| Conclusion | | (Ground | Truth: | True): | | | | (cid:44)→ | | | | | | -| | | | | | | | Error: | Exists | should | not | be capitalized | | | -| All | squares | are shapes. | | | | | | | | | | | | -| | | | | | | | Premise: | The | security | deposit | | can be either | | -FOLs: -| | | | | | | | | equal | to | one month's | rent | or more. | | -| --- | ---------------- | --- | ---------------- | --------- | --- | --- | ---- | --------- | ------------ | ----------- | ---- | -------------- | --- | -| all | x. (Square(x) | | -> FourSides(x)) | | | | | (cid:44)→ | | | | | | -| | | | | | | | FOL: | all x. | all y. | (Rent(x, | y) | -> (Deposit(x) | = | -| all | x. (FourSided(x) | | -> | Shape(x)) | | | | | | | | | | -| | | | | | | | | y | | Deposit(x) | > | y)) | | | -(cid:44)→ -| | | | | | | | Error: | > is | not allowed | | in Prover9's | FOL | | -| ---------- | ------------------------ | -------- | ------------ | --- | ---------- | --------- | ------- | ----- | ----------- | -------- | ------------ | ---------- | -------- | -| Conclusion | | (Prover9 | Deduction: | | Uncertain) | | | | | | | | | -| all | x. (Square(x) | | -> Shape(x)) | | | | | | | | | | | -| | | | | | | | L4: | There | are | mistakes | | in the FOL | trans- | -| | | | | | | | lation. | For | GPT-4, | there | are | very few | transla- | -| L3: | FOLcontainssyntaxerrors. | | | | | Thisisone | | | | | | | | -tionmistakes,butwestillfindacoupleinstances -reasonthatweobservemajorityvotinghelpssig- -whereGPT-3.5failstocapturethesemanticmean- -nificantlyforneurosymbolicmethods,asmultiple -| | | | | | | | ing | of the | NL examples | | properly. | Two | examples | -| --- | --- | --- | --- | --- | --- | --- | --- | ------ | ----------- | --- | --------- | --- | -------- | -generationsincreasethelikelihoodthatatoneleast -areshownhere: -| oneFOLrepresentationsuccessfullycompiles. | | | | | | As | | | | | | | | -| ----------------------------------------- | --- | -------- | ----- | -------- | ------ | ----- | --------- | -------- | ----------------------- | ------- | ----- | --- | --- | -| | | | | | | | Premise: | Platypus | | have no | teeth | | | -| stated | in | the main | text, | the most | common | error | | | | | | | | -| | | | | | | | Incorrect | | FOL: -NoTeeth(Platypus) | | | | | -isthatthesamesymbolisusedwithmultipleari- -ties. Twoexamplesofthisareshownbelow: inthe Premise: Imperium doesn't have a feud with a -| | | | | | | | | professional | | wrestling | | stable | that | -| ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------ | --- | --------- | --- | ------ | ---- | -| firstexample,theprimitiveSummerisbothusedas | | | | | | | | (cid:44)→ | | | | | | -| | | | | | | | | includes | | Ivy Nile. | | | | -(cid:44)→ -a function representing that a year has a summer Incorrect FOL: -all x. (Includes(x, IvyNile) -> - -| | | | | | | | | HasFeud(Imperium, | | | x)) | | | -| --- | --------- | -------- | ----- | ------- | --------- | ---------- | --- | ----------------- | --- | --- | --- | --- | --- | -| (in | the first | premise) | and | as a | constant | represent- | | (cid:44)→ | | | | | | -| ing | that a | student | loves | summer. | A similar | phe- | | | | | | | | -nomenon occurs in the second example with the E.3 GPT-4CoTFailureModes -symbolBadults. -C1: CoTconcludessomethingdifferentthanit -| Example | 1 | | | | | | suggests. | | We show | two | examples | of | this below | -| ------- | --- | --- | --- | --- | --- | --- | --------- | --- | ------- | --- | -------- | --- | ---------- | - -(bothofwhichLINCsolvescorrectly). (cid:44)→ needs to apply for CPT if he wants to -(cid:44)→ work in the US. We also know that an -Example 1 (cid:44)→ international student in the US with an -Premises: (cid:44)→ F1 visa needs to apply for CPT or OPT if -Plungers suck. (cid:44)→ the student wants to work in the US. So, -Vacuums suck. (cid:44)→ it is clear that Mike has an F1 visa -Vampires suck. (cid:44)→ since he fits the criteria. Therefore, -Space is a vacuum. (cid:44)→ the conclusion is true. -A duster is a household appliance that doesn't -(cid:44)→ suck. Example 2 -Premises: -Conclusion: If people perform in school talent shows often, -Space is a vampire. (cid:44)→ then they attend and are very engaged -(cid:44)→ with school events. -(GPT-4 Incorrect CoT): Let's think step by step. People either perform in school talent shows -(cid:44)→ We want to evaluate if Space is a (cid:44)→ often or are inactive and disinterested -(cid:44)→ vampire. We know that Space is a vacuum (cid:44)→ members of their community. -(cid:44)→ and that vacuums suck. However, while it' If people chaperone high school dances, then -(cid:44)→ s true that vampires also suck, this (cid:44)→ they are not students who attend the -(cid:44)→ doesn't mean that everything that sucks (cid:44)→ school. -(cid:44)→ is a vampire. Therefore, there is no All people who are inactive and disinterested -(cid:44)→ reason to believe that Space is a vampire (cid:44)→ members of their community chaperone high -(cid:44)→ based on these premises, so the (cid:44)→ school dances. -(cid:44)→ statement is false. All young children and teenagers who wish to -ANSWER: False (cid:44)→ further their academic careers and -(cid:44)→ educational opportunities are students -Example 2 (cid:44)→ who attend the school. -Premises: Bonnie either both attends and is very engaged -Some mammals have teeth. (cid:44)→ with school events and is a student who -Platypus have no teeth. (cid:44)→ attends the school, or she neither -Platypus are mammals. (cid:44)→ attends and is very engaged with school -Humans have teeth. (cid:44)→ events nor is a student who attends the -(cid:44)→ school. -Conclusion: -Platypus are reptiles. Conclusion (Ground Truth: Uncertain): -Bonnie performs in school talent shows often. -(GPT-4 Incorrect CoT): Let's think step by step. -(cid:44)→ We want to evaluate if platypus are (GPT-4 Incorrect CoT): Let's think step by step. -(cid:44)→ reptiles. We know from the premises that (cid:44)→ We want to evaluate if Bonnie performs -(cid:44)→ platypus are mammals and that they have (cid:44)→ in school talent shows often. Based on -(cid:44)→ no teeth. The fact that some mammals have (cid:44)→ the last premise, if Bonnie is a student -(cid:44)→ teeth does not contradict the fact that (cid:44)→ who attends the school and is very -(cid:44)→ platypus, although mammals, do not have (cid:44)→ engaged with school events, this is only -(cid:44)→ teeth. Moreover, the statement about (cid:44)→ possible if she performs in school talent -(cid:44)→ humans having teeth is unrelated to the (cid:44)→ shows often, as per the first premise. -(cid:44)→ classification of platypus. There's no (cid:44)→ If she doesn't attend and is not engaged -(cid:44)→ evidence in the premises that would (cid:44)→ with school events, she is an inactive -(cid:44)→ suggest platypus are reptiles. Therefore, (cid:44)→ and disinterested member of her community -(cid:44)→ the statement is false. (cid:44)→ . But the fourth premise states that -(cid:44)→ these individuals chaperone high school -C2: CoTmakesincorrectlogicaldeductions. (cid:44)→ dances, meaning they cannot be students, -(cid:44)→ which would contradict the last premise. -Example 1 (cid:44)→ Therefore, Bonnie performs in school -Premises: (cid:44)→ talent shows often, so the statement is -International students in US have either an F1 (cid:44)→ true. -(cid:44)→ visa or a J1 visa. -An international student in US with an F1 visa -(cid:44)→ needs to apply for CPT or OPT if the C3: CoTfailstofindcomplexpathsofreason- -(cid:44)→ student wants to work in the US. ing. Wehighlighttwoexamplesbelow: inthefirst -Mike is an international student. -example, the ground truth is false. To make this -Mike needs to apply for CPT if he wants to work -(cid:44)→ in the US. deduction,onemustreasonthatifaGreyhoundis -aBoeing707,thenitisaplane,whichmeansitis -Conclusion (Ground Truth: Uncertain): -empty, which means it cannot transport multiple -Mike has an F1 visa. -passengers,whichmeansitisnotanairline,which -(GPT-4 Incorrect CoT): Let's think step by step. -meanstherearenoGreyhoundplanes,whichisa -(cid:44)→ We want to evaluate if Mike has an F1 -contradiction. LookingattheCoTgenerations,the -(cid:44)→ visa. From the premises, we know that -(cid:44)→ Mike is an international student and firstCoTattemptgivesupafterfailingtofindany - -linkbetweenGreyhoundandBoeing707. Thesec- No young person teaches. -Jerry neither teaches nor is a manager. -ondgenerationattemptstomakedeductionsfrom -Rose is young or a student. -thepremises. Inthiscase,noneofthe10CoTrea- -soningchainsbeginwiththecorrectstepofstarting Conclusion (Ground Truth: True) -Rose is a human or Jerry is a manager. -fromthenegationoftheconclusionanddeducing -itfalse. (GPT-4 Incorrect CoT, 1): Let's think step by -Inthesecondexample, tomakethecorrectde- (cid:44)→ step. We want to evaluate if Rose is a -(cid:44)→ human or Jerry is a manager. Let's first -duction,weneedtostartfromthefactthatRoseis -(cid:44)→ look at Rose. From premise 6, we know -youngorastudent. IfRoseisyoung,thentheydo (cid:44)→ that Rose is either young or a student. -not teach, which means they study, which means (cid:44)→ If Rose is a student, then from premise 1 -(cid:44)→ we know that Rose is a human. If Rose is -theyareastudent,whichmeanstheyareahuman. (cid:44)→ young, we don't know from the premises -IfRoseisastudent,thentheyarehumans. Neither (cid:44)→ whether she is a human. Therefore, it is -oftheCoTgenerationisabletomakeprogresson (cid:44)→ uncertain whether Rose is a human. Now -(cid:44)→ let's look at Jerry. From premise 5, we -the deduction from the information that Rose is (cid:44)→ know that Jerry is not a manager. -young. In addition, the first CoT generation also (cid:44)→ Therefore, Jerry is not a manager. So, it -(cid:44)→ is uncertain whether Rose is a human, -hasalogicalerroratthelaststep,whereitasserts -(cid:44)→ but we know that Jerry is not a manager, -that“AorFalse”isFalsewhenthetruthnessofA (cid:44)→ so the statement is false. -isuncertain. -(GPT-4 Incorrect CoT, 2): Let's think step by -Example 1 (cid:44)→ step. We want to evaluate if Rose is a -Premises: (cid:44)→ human or Jerry is a manager. First, we -If something can transport multiple passengers, (cid:44)→ know that Jerry neither teaches nor is a -(cid:44)→ then they are not empty. (cid:44)→ manager, which means Jerry is not a -All airlines can transport multiple passengers. (cid:44)→ manager. Therefore, the only way for the -All planes are empty. (cid:44)→ conclusion to be true is if Rose is a -All Boeing 707s are planes. (cid:44)→ human. We know that Rose is either young -If Greyhound is not an airline, then there are (cid:44)→ or a student. If she's a student, since -(cid:44)→ no Greyhound planes. (cid:44)→ all students are humans, then she is a -(cid:44)→ human. However, if Rose is young, we don' -Conclusion (Ground Truth: False) (cid:44)→ t have enough information to determine -A Greyhound is a Boeing 707. (cid:44)→ whether she's a human. Therefore, it's -(cid:44)→ uncertain whether Rose is a human or -(GPT-4 Incorrect CoT, 1): Let's think step by (cid:44)→ Jerry is a manager. -(cid:44)→ step. We want to evaluate if a Greyhound -(cid:44)→ is a Boeing 707. Looking at our premises, -(cid:44)→ we have no direct or indirect E.4 SharedmistakesbetweenGPT-4CoTand -(cid:44)→ information linking Greyhounds to Boeing LINC -(cid:44)→ 707s. Given the information available, we -(cid:44)→ cannot make any certain conclusion about AsshowninFig. 5c,thereare21commonerrors -(cid:44)→ a Greyhound being a Boeing 707. betweenGPT-4CoTandGPT-4LINC.Afteranin- -(cid:44)→ Therefore, it is uncertain whether a -(cid:44)→ Greyhound is a Boeing 707. depthanalysisoftheexamples,weseethat16of -theseariseduetoinherenterrorsinthedataset: -(GPT-4 Incorrect CoT, 2): Let's think step by -(cid:44)→ step. We want to evaluate if a Greyhound -(cid:44)→ is a Boeing 707. The first three premises • 2ofthesesamplescontainthesentencefrag- -(cid:44)→ don't provide any information about ment “Either Zaha Hadid’s design style or -(cid:44)→ Greyhound or Boeing 707s. The fourth KellyWearstler’sdesignstyle.” asapremise. -(cid:44)→ premise tells us that all Boeing 707s are -(cid:44)→ planes but doesn't tell us anything This premise is likely intended to mean all -(cid:44)→ about Greyhound. The fifth premise tells designstylesareoneofthesetwostyles,but -(cid:44)→ us that if Greyhound is not an airline, -this is hard for the model to grasp from just -(cid:44)→ then there are no Greyhound planes, but -(cid:44)→ this still doesn't imply that a Greyhound thefragment(sample41,42). -(cid:44)→ is a Boeing 707. Therefore, we do not -(cid:44)→ have enough information to conclude • 2ofthesesamplescontainthesentencefrag- -(cid:44)→ whether a Greyhound is a Boeing 707 or -ment“EitherfemaletennisplayersatRoland -(cid:44)→ not. So, the statement is uncertain. -Garros2022ormaletennisplayersatRoland -Example 2 -Garros2022.” asapremise(sample43,45). -Premises: -All students are humans. -• 2ofthesesampleshaveanambiguoususeof -If someone studies, then they are a student. -A person either studies or teaches. “either”: “BeniseitherfromTheSimpsonsor - -| | | | | | | F ProofwriterStarCoder+Errors | | | | | | | -| --- | --- | --- | --- | --- | --- | ----------------------------- | --- | --- | --- | --- | --- | --- | -funny,”whichinthiscaseambiguouslymeans -XOR.Webelievethissentenceisanunnatural -Wealsoanalyzed,qualitatively,theerrorsthatStar- -usageof“either”(sample142,143). -| | | | | | | Coder+ | makes | on the | ProofWriter | | dataset. | Since | -| --- | --- | --- | --- | --- | --- | ------ | ----- | ------ | ----------- | --- | -------- | ----- | -premisesfromtheProofWriterdatasetarerelatively -• In4samples,thereisanamethatisimplicitly -| | | | | | | simpleanddirect,lossytranslationisrarer. | | | | | | Moster- | -| ---------- | --- | ------- | --------- | ------- | ---- | ---------------------------------------- | --- | --- | --- | --- | --- | ------- | -| an animal, | but | this is | not clear | (sample | 126, | | | | | | | | -rorsareduetosyntaxorimproperrepresentations. -| 127, 128, | 199). | Also, | in samples | | 126-128, | | | | | | | | -| --------- | ----- | ----- | ---------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | -thereisastatement“IfRockisneitheraflynor Weshowfiverepresentativeerrorsbelow: -a bird, then Rock neither flies nor breathes.” Premise: All quiet people are round. -| | | | | | | StarCoder+ | FOL: | all x. | Quiet(x) | -> | Round(x) | | -| --- | --- | --- | --- | --- | --- | ---------- | ---- | ------ | -------- | --- | -------- | --- | -thatshouldlikelysay“IfRockneitherfliesnor -| | | | | | | Ground | Truth FOL: | all | x. (Quiet(x) | | -> Round(x)) | | -| ---------- | ----- | -------- | -------- | ------------ | --- | ------ | ---------- | --- | ------------ | --- | ------------ | --- | -| is a bird, | ...”. | With the | original | formulation, | | | | | | | | | -everythingcouldbeuncertainbecausenothing Premise: If someone is quiet and nice then they -| | | | | | | (cid:44)→ | are cold. | | | | | | -| -------- | ------ | ---- | --------- | ---- | -------- | ---------- | --------- | ------ | --------- | --- | -------- | --- | -| is known | in the | case | that Rock | is a | fly (and | | | | | | | | -| | | | | | | Starcoder+ | FOL: | all x. | (Quiet(x) | \& | Nice(x)) | -> | -thisisareasonableinterpretation). -(cid:44)→ Cold(x) -| | | | | | | Ground | Truth FOL: | all | x. ((Quiet(x) | | \& Nice(x)) | | -| --- | --- | --- | --- | --- | --- | --------- | ----------- | --- | ------------- | --- | ----------- | --- | -| | | | | | | (cid:44)→ | -> Cold(x)) | | | | | | -• In5samples,theground-truthFOLrepresen- -tationofthenaturallanguagepremiseisincor- -| | | | | | | Premise: | If something | | is furry | then | it is | quiet. | -| --- | --- | --- | --- | --- | --- | -------- | ------------ | --- | -------- | ---- | ----- | ------ | -rect,causingthelabeltobeincorrect(samples Starcoder+ FOL: Furry(x) -> Quiet(x) -| | | | | | | Ground | Truth FOL: | all | x. (Furry(x) | | -> Quiet(x)) | | -| --- | --- | --- | --- | --- | --- | ------ | ---------- | --- | ------------ | --- | ------------ | --- | -29,79,85,86,87). -| | | | | | | Premise: | The cow | eats | the dog. | | | | -| --- | --- | --- | --- | --- | --- | -------- | ------- | ---- | -------- | --- | --- | --- | -• There is a sample where the conclusion is Starcoder+ FOL: Cow(Eats(Dog)) -likely mis-worded and should be “Barutin Ground Truth FOL: Eats(Cow, Dog) -| Cove is | not located | in | Antarctica.” | | instead | | | | | | | | -| ------- | ----------- | --- | ------------ | --- | ------- | -------- | ------- | -------- | --- | --- | --- | --- | -| | | | | | | Premise: | The cow | is blue. | | | | | -of“BarutinisnotlocatedinAntarctica.” This Starcoder+ FOL: Cow(x) \& Blue(x) -| | | | | | | Ground | Truth FOL: | Blue(Cow) | | | | | -| --------------------------- | --- | ---------------------------- | --- | ------------ | ----- | --------------------------------- | ---------- | --------- | --- | --- | --- | --- | -| changesthegroundtruthlabel. | | | | (sample121). | | | | | | | | | -| The remaining | | 5 samples | are | examples | where | | | | | | | | -| | | | | | | G StarCoderPlusFOLIOErrorAnalysis | | | | | | | -| bothmethodsfail. | | Inallcases,CoTfailstofindthe | | | | | | | | | | | -correctreasoningchain,asthepremises/reasoning For StarCoder+, we see a slightly different trend. -Meanwhile,LINC -pathisconvolutedandcomplex. InFig. 6a, weseethesamepatternasforGPT-4, -failsasfollows: withamajorityofuncertainpredictions. InFig. 6b, -however,weseethatCoTforStarCoder+primarily -• Insample104,LINCgeneratesseparatesym- -| | | | | | | predictstrue. | Thisislikelybecausethemodelwas | | | | | | -| -------- | ------------- | --- | --- | ------------ | --- | ------------- | ------------------------------ | ---- | ---- | ---- | ----- | ------- | -| bols for | NotLiveAtHome | | and | -LiveAtHome, | | | | | | | | | -| | | | | | | trained | on much | more | code | than | text, | and may | -whicharesemanticallyequivalent. nothavepickedupsophisticatedtextualchain-of- -| | | | | | | thought | reasoning | capabilities. | | In | Fig. | 6c, we | -| --- | --- | --- | --- | --- | --- | ------- | --------- | ------------- | --- | --- | ---- | ------ | -Insample149,LINCgeneratesFOLsthatmiss -| • | | | | | | | | | | | | CoT | -| --- | --- | --- | --- | --- | --- | ------- | -------- | -------------- | --- | ------- | --- | --- | -| | | | | | | can see | that the | mispredictions | | between | | and | -implicit information; i.e. that people with LINCdiffermuchmoreforStarCoder+thanGPT- -namesRoseandJerryarehumans. -| | | | | | | 4. Finally, | in | Fig. | 6d, we | see the | same | trends | -| --- | --- | --- | --- | --- | --- | ----------- | ---- | ----- | ------ | ---------------- | ---- | ------ | -| | | | | | | as we saw | with | GPT-4 | but | more pronounced, | | as | -• Insamples172and173,LINCdoesnotgener- -thesimilaritybetweenmispredictionsinLINCand -ateasinglevalidsetofFOLpremisesbecause -| | | | | | | those in | the baseline | | methods | is even | lower | than | -| --- | --- | --- | --- | --- | --- | -------- | ------------ | --- | ------- | ------- | ----- | ---- | -allgeneratedexpressionshaveamismatched -theywereforGPT-4. -arityissue. -| | | | | | | H TheeffectofK-waymajorityvotingon | | | | | | | -| --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | -Insample193,LINCfailstoexpressthecon- -| • | | | | | | LINCandCoT | | | | | | | -| ------- | ---------- | --- | ---------- | --- | -------- | ---------- | --- | --- | --- | --- | --- | --- | -| clusion | correctly: | the | conclusion | is | “If Yuri | | | | | | | | -is not an American basketball player, then In all our experiments, we use 10-way majority -Yuri is a basketball player.”, but the FOLs voting, inspired by prior work which found that -outputted are either -American(Yuri) Chain-of-Thoughtpromptingbenefitedtherefrom -& -BasketballPlayer(Yuri) (Wang et al., 2023b). However, one might won- -| | -American(Yuri) | | | | & | | | | | | | | -| --- | --------------- | --- | --- | --- | --- | ------- | ------ | --------------- | --- | ----- | ---- | ---- | -| or | | | | | | der how | robust | the performance | | gains | seen | with | -BasketballPlayer(Yuri) LINCaretotheprecisevalueofK. Figure7thus - -| | T | | | | | 100 | | T | | | | | 100 | -| --------------- | --- | --- | --------------- | --- | --- | --- | --------------- | --- | --- | --------------- | --- | ---- | --- | -| | ru | | | | | | | ru | | | | | | -| | e | | | | | | | | e | | | | | -| | | 31 | 7 | 6 | 0 | | | | 61 | 38 | | 55 0 | | -| | | | | | | 80 | | | | | | | 80 | -| lebaL detciderP | F a | | | | | | lebaL detciderP | F a | | | | | | -| | lse | | | | | | | lse | | | | | | -| | | 1 | 16 | 0 | 0 | | | | 1 | 12 | | 7 0 | | -| | U | | | | | 60 | | U | | | | | 60 | -| | n | | | | | | | n | | | | | | -| | ce | | | | | | | ce | | | | | | -| | rta | | | | | | | rta | | | | | | -| | in | | | | | | | in | | | | | | -| | | 25 | 29 | 55 | 0 | 40 | | | 1 | 4 | | 3 0 | 40 | -| | E | | | | | | | E | | | | | | -| | rro | | | | | | | rro | | | | | | -| | | | | | | 20 | | | | | | | 20 | -| | r | 6 | 2 | 4 | 0 | | | | r 0 | 0 | | 0 0 | | -| | | | | | | 0 | | | | | | | 0 | -| | | e | e | n | or | | | | e | e | | n or | | -| | | u | Fals | ai | | | | | u | Fals | ai | | | -| | | Tr | | ert | Err | | | | Tr | | ert | Err | | -| | | | | nc | | | | | | | nc | | | -| | | | | U | | | | | | | U | | | -| | | | Reference Label | | | | | | | Reference Label | | | | -(a)ConfusionmatrixforLINC. (b)ConfusionmatrixforChain-of-Thought. -| | | | | | | 100 | | L | | | | | 1.0 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -IN -| | C | | | | | | | C | | | | | | -| ----------- | ----- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| | o rre | | | | | | | | 1 | | | | | -| | ct | | | | | 80 | | | | | | | 0.8 | -| | | 37 | | 39 | | | | | | | | | | -| emoctuO ToC | | | | | | | | C o | | | | | | -T -| | | | | | | 60 | S c | | 0.11 | 1 | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | -| | | | | | | | | ra | | | | | 0.6 | -tc -h -| | In | | | | | | | p | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | -| | co | | | | | 40 | | a d | | | | | | -| | rre | | | | | | | | 0.1 | 0.62 | | 1 | | -0.4 -ct -| | | 65 | | 41 | | | | N | | | | | | -| --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | -| | | | | | | 20 | | a iv | | | | | | -e -| | | | | | | | | | 0.21 | 0.44 | 0.48 | 1 | 0.2 | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | ---- | ---- | --- | --- | -0 -| | | ect | | ect | | | | | C | T | | d e | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------ | --- | -| | | | | | | | | | N | o | | a ai v | | -| | | orr | | orr | | | | | LI | C | h p | N | | -| | | C | | nc | | | | | | | c | | | -| | | | | I | | | | | | | at | | | -cr -| | | | LINC Outcome | | | | | | | | S | | | -| --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -(c) Comparing the consistency of LINC vs (d)Similaritybetweenincorrectpredictionsofeachmethod, -Chain-of-Thought. i.e.,(A wrong == B wrong) / (A wrong or B wrong). -Figure6: AnalyzingandcomparingthemistakesmadebyStarCoder+ontheFOLIOdataset. -shows,foreachmodelequippedwitheitherLINC -Chain-of-Thought, -| or | | | | how the | accuracy | on FO- | | | | | | | | -| --- | ------ | ---- | --- | --------------- | -------- | ------ | --- | ----------------------------- | --- | --------------------------- | --- | ---------------------- | --- | -| LIO | varies | with | K ∈ | {1,2,3,...,10}. | We | note | | 1.0 | | | | | | -| | | | | | | | | tes tset tuo-dleh no ycaruccA | | StarCoder+-Chain-of-Thought | | GPT-4-Chain-of-Thought | | -that,generallyspeaking,LINCmakesgooduseof StarCoder+-LINC (ours) GPT-4-LINC (ours) -| | | | | | | | | | | GPT-3.5-Chain-of-Thought | | Chance | | -| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | ------ | --- | -0.8 -increasedvaluesofK. Thisisespeciallytruefor GPT-3.5-LINC (ours) -theweakermodels;thesearemorepronetogener- -0.6 -atingsyntacticallyinvalidFOLexpressions,which -| causethesolvertoreturnanError. | | | | | Takingthema- | | | | | | | | | -| ------------------------------ | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | -0.4 -jorityvoteovermanysamplesthuslessenstherisk -| ofpredictingError,whichisofcoursealwaysthe | | | | | | | | 0.2 | | | | | | -| ------------------------------------------ | ------ | -------- | --- | ----------- | ------ | -------- | --- | --- | --- | --- | --- | --- | --- | -| wrong | label. | Notably, | | our results | do not | indicate | | | | | | | | -| | CoT | | | | | | | 0.0 | | | | | | -that benefits from majority voting in this do- 1 2 3 4 5 6 7 8 9 10 -main. Futureworkisneededtoestablishhowthis K-way majority vote -relatestothefindingsinthepreviouslymentioned Figure 7: Accuracy on FOLIO per value of K (Ap- -| priorwork. | | | | | | | pendixH).Shadedareasare±1standarddeviationover | | | | | | | -| ---------- | --- | --- | --- | --- | --- | --- | ---------------------------------------------- | ------------ | --- | -------- | ---- | --------------- | --- | -| | | | | | | | 1000 | bootstrapped | | samples. | Note | that increasing | K | -benefitsLINC(solidlines;shadingincolor)butnotCoT -| | | | | | | | (dashedlines; | | shadingingray)inourexperimentson | | | | | -| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | -------------------------------- | --- | --- | --- | --- | -thisdataset. \ No newline at end of file + +Finally, we address Edwards’s argument that Congress ex- +ceeded its Commerce Clause authority when it enacted 18 U.S.C. +§ 922(g).9 + +9 We review a district court’s determination regarding the constitutionality of +a statute for plain error when, as here, the issue wasn’t presented below. See +United States v. Valois, 915 F.3d 717, 729 n.7 (11th Cir. 2019). + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 26 of 27 + +26 + +Opinion of the Court + +22-13963 + +We have repeatedly held that § 922(g) is facially constitu- +tional under the Commerce Clause because it contains an express +jurisdictional requirement. See, e.g., United States v. Stancil, 4 F.4th +1193, 1200 (11th Cir. 2021); United States v. Jordan, 635 F.3d 1181, +1189 (11th Cir. 2011); United States v. Scott, 263 F.3d 1270, 1273–74 +(11th Cir. 2001) (“[T]he jurisdictional element of the statute, i.e., +the requirement that the felon ‘possess in or affecting commerce, +any firearm or ammunition,’ immunizes § 922(g)(1) from [a] facial +constitutional attack.”); United States v. Dupree, 258 F.3d 1258, 1259– +60 (11th Cir. 2001); United States v. McAllister, 77 F.3d 387, 390 (11th +Cir. 1996). + +We have also rejected as-applied challenges to § 922(g), +holding that the government demonstrates the required “minimal +nexus” to interstate commerce when it proves that the firearms +were manufactured outside the state where the offense took place +and, thus, necessarily traveled in interstate commerce. See, e.g., Jor- +dan, 635 F.3d at 1189 (holding that § 922(g) is constitutional as ap- +plied to “a defendant who possessed a firearm only intrastate” be- +cause the firearm had moved in interstate commerce); see also +United States v. Wright, 607 F.3d 708, 715–16 (11th Cir. 2010). Here, +Edwards admitted in his factual proffer that the firearm and ammu- +nition that he possessed in Florida were manufactured out of state +and had therefore moved in interstate commerce prior to his pos- +session of them. The government thus satisfied the minimal-nexus +requirement here. + + USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 27 of 27 + +22-13963 + +Opinion of the Court + +27 + +Under the prior-panel-precedent rule, we’re obliged to fol- +low our existing decisions unless overruled by this Court en banc +or abrogated by the Supreme Court. United States v. White, 837 +F.3d 1225, 1228 (11th Cir. 2016). Because neither has occurred +here, we affirm the judgment of the district court that § 922(g) +doesn’t violate the Commerce Clause. + +III + +For the foregoing reasons, we AFFIRM the district court’s + +judgment. + +AFFIRMED. + diff --git a/q181/random_k4/random_3.md b/q181/random_k4/random_3.md index ed155355c8065855b0ca8f084a2bdb28a25cae49..12427a1be709195a0087a38811902bf043f32efc 100644 --- a/q181/random_k4/random_3.md +++ b/q181/random_k4/random_3.md @@ -1,527 +1,918 @@ -# **Retrieval Meets Reasoning: Even High-school Textbook Knowledge Benefits Multimodal Reasoning** +Graduate Student Handbook for -**Cheng Tan**[1] _[∗]_ **Jingxuan Wei**[2] _[∗]_ **Linzhuang Sun**[2] **Zhangyang Gao**[1] **Siyuan Li**[1] **Bihui Yu**[2] **Ruifeng Guo**[2] **Stan Z. Li**[1] _[†]_ 1Westlake University 2Shenyang Institute of Computing Technology, Chinese Academy of Sciences +The John E. Walker Department of + +Economics 2025-2026 + +Overview + +The Department faculty members teach and publish regularly. This manual outlines the policies and +procedures in the Ph.D. program. -## **Abstract** +The goal of the Ph.D. program is to educate students in the basics of microeconomic theory, +macroeconomic theory, econometrics, and in their application to a Primary and Secondary Field of +study, culminating in the writing and defending of the Ph.D. dissertation. Our training is rigorous, +with an emphasis on the application, implementation, and interpretation of mathematically rigorous +models to a wide variety of phenomena that lend themselves to logical analysis. -Large language models equipped with retrieval-augmented generation (RAG) represent a burgeoning field aimed at enhancing answering capabilities by leveraging external knowledge bases. Although the application of RAG with language-only models has been extensively explored, its adaptation into multimodal vision-language models remains nascent. Going beyond mere answer generation, the primary goal of multimodal RAG is to cultivate the models’ ability to reason in response to relevant queries. To this end, we introduce a novel multimodal RAG framework named RMR ( **R** etrieval **M** eets **R** easoning). The RMR framework employs a bi-modal retrieval module to identify the most relevant question-answer pairs, which then serve as scaffolds for the multimodal reasoning process. This training-free approach not only encourages the model to engage deeply with the reasoning processes inherent in the retrieved content but also facilitates the generation of answers that are precise and richly interpretable. Surprisingly, utilizing solely the ScienceQA dataset, collected from elementary and high school science curricula, RMR significantly boosts the performance of various vision-language models across a spectrum of benchmark datasets, including A-OKVQA, MMBench, and SEED. These outcomes highlight the substantial potential of our multimodal retrieval and reasoning mechanism to improve the reasoning capabilities of vision-language models. +The program is designed to start in the Fall semester. The program requires 42 hours of graded +coursework and an additional 18 hours of thesis hours. Knowledge is assessed via a combination of +grades in courses, Qualifying Examinations in Core subjects (microeconomic theory, macroeconomic +theory, and econometrics), and Qualifying Examinations and/or papers in the Primary Field. Aside +from coursework, competence is assessed via students’ written work in the form of second and third +year papers, presented in the Workshop of their choice, and, ultimately, the Dissertation itself (18 +thesis hours). Ph.D. students have written dissertations on a wide range of topics, which can be +viewed here, https://www.clemson.edu/graduate/students/theses-and-dissertations/index.html. +The dissertations written by Clemson Ph.D. students are testament to the fact that virtually any topic +can be analyzed within the framework of economics, and that our faculty are ready and willing to +help them do so. -## **1 Introduction** +Our goal is to provide students the tools to prepare them for jobs both academic and non-academic, +private and public-sector, and domestic and international. Students must demonstrate competence +by passing core course sequences in the initial year of study and the subsequent Qualifying +Examination in Economic Theory, as well as satisfying a Minimum Competency Requirement in +econometrics. In their second and third years, students choose two concentrations from the fields of +economic growth, environmental economics, financial economics, +industrial organization, +international economics, labor economics, macroeconomics, and public economics, among others. +Students begin developing a research program by writing and submitting papers in their second and +third years of study. In addition, all students beyond the first year must attend a weekly workshop in +their chosen field(s), where they will have the opportunity to present their research and receive +feedback from the faculty and other graduate students. Successful students will have chosen a main +Ph.D. advisor no later than the start of their fourth year and will write and defend their dissertation +within four or five years. -While deep learning and its applications have been widely explored in recent years [64, 23, 44, 41, 61, 43, 60, 42, 62], retrieval-augmented generation (RAG) has rapidly emerged as a cornerstone in the development of large language models (LLMs), enabling them to enhance their capabilities by leveraging external knowledge bases [77, 30, 11]. Integrating LLMs with RAG has found its most impactful application within language-centric models, where the dynamic interplay between retrieved content and answer generation significantly elevates the quality and relevance of responses [10, 86, 10]. While early works have demonstrated that incorporating directly retrieved information into language models can improve the quality of the generated content [29], subsequent developments have involved refinement and mitigate the potential noise associated with the raw retrieval results [71, 78, 2], thus ensuring that the content generated is not only accurate but also contextually enriched. +1 -Although the integration of RAG with large language models has been extensively explored, its adaptation to multimodal scenarios that encompass both visual and textual inputs remains relatively nascent [12, 76, 74]. Notably, in the domain of visual question answering (VQA), where queries comprise concise textual prompts paired with complex visual data, the requirements for integration diverge significantly from traditional text-centric approaches [74, 17]. Traditional RAG systems, + Contact Information -> _∗_ Equal contribution. +The John E. Walker Department of Economics is located at: -> _†_ Corresponding author. +225 Walter T. Cox Blvd., Suite N-320 +College of Business +864-656-3481 +864-656-4192 (fax) -Preprint. Under review. +The Graduate Program Coordinator for the Ph.D. program is: -initially designed for text-heavy applications, encounter substantial challenges when applied directly to multimodal tasks. In these contexts, textual data often provides insufficient contextual cues, failing to bridge the interpretative demands of rich visual information. This fundamental limitation is critical: models trained predominantly on textual data struggle to effectively capture the nuanced complexity of visual information, leading to significant gaps in the model’s ability to accurately interpret and reason about visual content. For instance, as illustrated in Figure 1, even when the three most pertinent pieces of information are retrieved in response to a query, the model may still fail to engage with the underlying reasoning processes embedded within the retrieved content. Instead of synthesizing insights from these inputs, the model may default to merely replicating answers, which can result in inaccuracies. This highlights a critical shortfall in current multimodal RAG systems, i.e., their inability to fully leverage the cognitive reasoning demanded by complex multimodal data. +Professor: Curtis Simon +225 Walter T Cox Blvd., Suite N-312F +864-656-3966 +cjsmn@clemson.edu -**==> picture [389 x 147] intentionally omitted <==** +Admission Requirements -**----- Start of picture text -----**
-Question 𝑋!"#$% Retrieval library
Which i in column 3?(A) the school Retrieved 𝑋! Retrieved 𝑋# Retrieved 𝑋"
(B) the park Which i in row A? Which i in column 1? Which i in row B?
(C) the pond (A) the pond (A) the police department (A) the theater
(D) the gas station (B) the school (B) the pond (B) the fire department
(C) the fire department (C) the school (C) the grocery store
(D) the gas station (D) the fire department (D) the fast-food restaurant
Result
The answer is (B). The answer is (B). The answer is (B).
The answer is (B).
**----- End of picture text -----**
+The Department of Economics requires that students have demonstrated academically distinguished +records at the undergraduate and, if applicable, the graduate level. To apply to any of these programs, +follow the instructions at the Graduate School page at www.grad.clemson.edu/admission/index.php +and complete the online application. +Coursework -Figure 1: limitations of multimodal retrieval enhancement with simple question-answer pairs. +Students need not have majored in Economics but must have coursework at least through +Intermediate Microeconomic Theory (the equivalent of our ECON 3140) and mathematics at least +through multivariate calculus (Clemson’s MATH 2070). Applicants to the Ph.D. program must have +taken the GRE examination; we look for quantitative scores in the 160s and verbal scores in the 150s. -Building on the foundational understanding that RAG significantly enhances the capabilities of large language models, we hypothesize that the ultimate purpose of multimodal RAG extends beyond merely instructing models to generate direct answers. Instead, our goal is to equip models with the ability to engage in cognitive reasoning, akin to human thought processes when confronted with complex, context-rich questions. This perspective underscores the necessity for multimodal RAG to be inherently flexible and open-ended, facilitating deep contemplation and robust reasoning whether the tasks are unimodal or multimodal, and independent of the data modality being retrieved. +Transfer of Credit -To realize this vision, we develop a comprehensive multimodal RAG framework, named RMR ( **R** etrieval **M** eets **R** easoning), which seamlessly integrates multimodal retrieval capabilities with in-context learning (ICL). This framework begins by employing a bi-modality retrieval module to fetch the most pertinent question-answer pairs, which may be unimodal or multimodal. It then integrates these elements into the model’s reasoning process, guiding it through the provided rationales associated with each retrieved item. Following this retrieval phase, the model autonomously learns coherent rationales that reflect a deep and meaningful engagement with the given problem. +As a rule, the Department does not allow transfer of credit for Ph.D. courses taken at other institutions, +a rule that is stricter than that of the University as a whole. The reason for this policy is that in our +experience, even students from well-ranked programs do not always have the knowledge that our +Department expects in order to pass the Qualifying Examination in EconomicTheory. -Remarkably, even when retrieval is limited to the ScienceQA dataset [49], which covers only elementary and high school science curricula, our Retrieval Meets Reasoning (RMR) framework demonstrates substantial enhancements across a variety of open-source multimodal models. Notably, the LLaVA model registers a +7.66% improvement, Qwen-VL achieves a +9.93% increase, InternLMXComposer2-VL records a +5.33% improvement, Gemini achieves a +33.67% increase. Furthermore, when evaluated against diverse datasets such as A-OKVQA [58], MMBench [47], and SEEDBench [37], RMR consistently delivers significant performance gains by leveraging the specialized knowledge embedded in the high-school curriculum from the ScienceQA dataset. +Deadlines -The main contributions of this work are as summarized as follows: +Formally, the deadline for application to the Ph.D. program is January 15 of the Spring semester. In +practice, the Department of Economics has evaluated applications on a rolling basis, that is, both +(shortly) prior to the formal deadline, and afterward throughout the Spring semester. That said, +applicants who desire full consideration for funding in the Ph.D. program should have completed their +application by January 15; applications received after that date will still be considered for funding, but +do not have the same chances. -- We introduce a comprehensive multimodal RAG framework, Retrieval Meets Reasoning (RMR), designed to enhance the reasoning capabilities of multimodal models, enabling them to generate answers through cognitive processes rather than merely replicating responses. +2 -- We develop a bi-modality retrieval module that effectively bridges the gap between unimodal and multimodal data, ensuring robust and accurate retrieval outcomes. + Tuition and Fees -- Despite its training-free manner, RMR demonstrates significant effectiveness across a variety of multimodal models and datasets, showcasing its capability to improve multimodal reasoning tasks. +recent -2 +Tuition and fees are set by Board of Trustees during their July meeting for the upcoming academic +year. The most +at +on +https://www.clemson.edu/graduate/finance-tuition/. The Department of Economics is classified +as a Tier 3 Program. Students on Graduate Assistantship have their tuition waived but must pay Full +Time Graduate Assistant Fees each semester. A student must take at least nine- credit hours per +semester in the fall and spring to be full-time. -## **2 Related Work** +information -**RAG in LLMs** Despite recent advancements in deep learning and large language or vision models [70, 63, 9], retrieval-augmented generation (RAG) improves language model capabilities by integrating external knowledge into the generation process [38, 22, 8, 14]. The development of RAG was initiated with the introduction of dense retrievers, which radically transformed how responses are generated by utilizing externally sourced information [36, 77, 87]. Initial efforts primarily focused on refining the interaction between the retriever and the generator [36, 24], leading to the production of contextually enriched responses. As a reliable enhancement, RAG has been foundational in allowing Large Language Models (LLMs) to exploit the vast reservoir of knowledge they encompass for various applications, including question answering, dialogue systems, and summarization [25, 71, 5]. +available -Recent advancements in RAG have been directed at addressing specific challenges, such as reducing hallucination phenomena and integrating outdated and obscure long-tail knowledge. An active retrieval mechanism that adaptively selects the most relevant knowledge pieces is employed to provide up-to-date information for the generation process [30]. Building on this, ActiveRAG [73] is proposed to incorporate an active learning mechanism that not only retrieves pertinent information but also synthesizes it with existing knowledge, markedly enhancing the model’s ability to handle knowledge-intensive tasks by dynamically integrating information. Although the development in RAG has significantly enhanced the synergy between retrieval and generation, the focus has primarily been on text-based applications, with limited exploration into multimodal scenarios. +tuition -**RAG for ICL** In-context learning (ICL) has revolutionized the functionality of LLMs, enabling them to adapt to new tasks by leveraging a few contextual examples provided directly within their input. This shift towards using retrieved demonstrations to facilitate ICL has increased the flexibility of LLMs across various applications [19, 72, 85]. The technique of demonstration retrieval, which involves selecting few-shot examples specifically tailored to the query, not only boosts task performance but also helps to mitigate biases that arise from manual or random selection of demonstrations. A key development has been the optimization of retrieval objectives, which ensures that the demonstrations are both pertinent to the query and diverse enough to offer a comprehensive context [72]. +fees -Expanding ICL into multimodal tasks represents a significant advancement, tackling the complex challenge of integrating textual and visual data. The extension of ICL into multimodal tasks represents a significant leap forward, addressing the inherent complexity of integrating textual and non-textual data. MM-Retrieval [45] is a concurrent work that introduces a retrieval-augmented multi-modal CoT reasoning approach, which dynamically selects demonstration examples by leveraging cross-modal and intra-modal similarities. However, it operates by employing modality-specific retrievers to gather demonstrations, which are then directly structured into a chain-of-thought format. +and -## **3 Retrieval Meets Reasoning** +is -## **3.1 Preliminaries** +All graduate students are required to have health insurance. Students who are already covered by +policies that meet the University’s requirements are eligible for waivers. All others are covered by the +Clemson University Student Insurance Plan. A description of the plan +is available at +https://www.clemson.edu/graduate/students/health-insurance.html. Fees are included with tuition +and fees for Fall and Spring semesters (summer premium is included in the Spring semester) and +students are automatically enrolled. Fees for the Clemson University Student Insurance are subsidized +by +at +is +https://www.clemson.edu/campus-life/student-health/insurance-and-billing.html. -In vision-language tasks, the objective is to develop a mapping _F_ Θ : _X →Y_ , where _X_ represents multimodal inputs that include both textual and visual elements, and _Y_ denotes the corresponding outputs. Formally, given a dataset _D_ = _{X , Y}_ , each input _X ∈X_ can be decomposed into _X_ = ( _T, I_ ), where _T_ denotes the text component and _I_ denotes the image component. In certain cases, one of them may be absent, i.e., _T_ = _∅_ or _I_ = _∅_ , resulting in modality-incomplete inputs. +the Graduate School. A description of -The primary goal is to accurately predict the output _Y ∈Y_ for a given input _X_ . This task can be expressed as identifying the answer _Y_ that maximizes the conditional probability _p_ ( _Y | T, I_ ) given the text component _T_ and the image component _I_ : +available -**==> picture [263 x 16] intentionally omitted <==** +subsidy -Here, _p_ ( _Y[′] | T, I_ ) denotes the probability of a candidate answer _Y[′]_ given the inputs _T_ and _I_ . For the supervised learning setting, the model optimal parameters Θ _[∗]_ are those that minimize the loss function _L_ , which quantifies the discrepancy between the predicted answer _F_ Θ( _X_ ) and the ground truth answer _Y_ . This optimization is: Θ _[∗]_ = arg minΘ _L_ ( _F_ Θ( _X_ ) _, Y_ ). However, in our approach, we employ an in-context learning strategy based on pre-trained large language models (LLMs). This +the -3 +for -method uses the retrieved content as contextual information without re-training the model, enabling it to obtain strong reasoning ability effectively. +I. -## **3.2 Bi-modality Retrieval Module** +Requirements for the PhD -To unify the retrieval module in multimodal models for both text and visual modality, we propose a bi-modality retrieval module based on the Contrastive Language-Image Pre-training (CLIP) [54] framework, as shown in Figure 2. This module is designed to handle various cases where the input may consist of complete image-text pairs, image-only inputs, or text-only inputs. The core idea is to create a robust embedding representation that can effectively capture the relevant information across different modalities and retrieve the most pertinent examples to guide the reasoning process. +Credit Hours -**==> picture [397 x 186] intentionally omitted <==** +The Ph.D. in Economics requires 60 hours of coursework, plus 18 thesis hours. A full-time course +load is considered to be 9 hours per semester, and students should, through the dissertation, generally +be enrolled full time. Core Economic Theory and Econometrics courses are taken during the first +two semesters. Starting with the second year in the program, PhD students are expected to take two +Field courses each semester. Students must demonstrate proficiency in two Fields, one of which is to +be designated Primary and the otherSecondary. -**----- Start of picture text -----**
-embedding library
Question : Birds are
Image retrieve retrieve warm-blooded.blooded … Warm-
image Encoder … Rationale : Birds, mammals, fish,
reptiles, and amphibians are groups of
animals...
Answer : The answer is albatross.
Which property do
these two objects have
in common? Select the
better answer. Question : Which ocean
is highlighted?
image-text pair
Rationale : Oceans are huge bodies of
What is the source of the Text salt water. The world has five oceans. All
allusionbelow? Myrain thethinkssentenceMr. Encoder ofworld ocean. This is the Atlantic Ocean…the oceans are connected, making one
Harper is a Luddite ScienceQA-based
because he doesn't own a Knowledge Library Answer : The answer is the Atlantic Ocean.
cell phone.
text …
**----- End of picture text -----**
+Required Courses +A. Economic Theory -Figure 2: The overall architecture and the retrieval mechanism of the bi-modality retrieval module. +Ph.D. students are required to take courses in mathematical economics (ECON 8040), microeconomic +theory (ECON 8010 and 8020), and macroeconomic theory (ECON 8050). -**Embedding Representation** The retrieval module begins by computing embeddings for the inputs using the CLIP model, which is adept at handling both textual and visual data. The embedding strategy is adaptive, accommodating the varying availability of modalities within the input. For imagetext pairs ( _T, I_ ), we calculate the mean pooling of CLIP’s text embedding _**h** T_ = CLIP _T_ ( _T_ ) _∈_ R _[d]_ and image embedding _**h** I_ = CLIP _I_ ( _I_ ) _∈_ R _[d]_ , where _d_ denotes the embedding dimension. Here, CLIP _T_ ( _·_ ) and CLIP _I_ ( _·_ ) denote the text and image encoders of CLIP, respectively. The combined embedding _**h** X_ is then computed as the average of these text and image embeddings. For imageonly inputs, we use the CLIP image embedding _**h** I_ , and for text-only inputs, we use the CLIP text embedding _**h** T_ . Thus, the item embedding _**h** X ∈_ R _[d]_ for any input _X_ is defined as: +B. Econometrics -**==> picture [277 x 39] intentionally omitted <==** +Ph.D. students are required to take two econometrics courses (ECON 8060, ECON 8070) during +their first year, and a third course (ECON 8080, 9090, others possible) during subsequent years. -This approach ensures robust embeddings regardless of the presence or absence of textual and visual components, thereby providing a flexible and consistent representation of multimodal inputs. +3 -**High-school Knowledge Library** We construct a comprehensive knowledge embedding library _H_ = _{_ _**h**[i] X[}][N] i_ =1[using][the][ScienceQA][dataset,][which][is][derived][from][elementary][and][high][school] textbooks. This dataset is particularly valuable because it includes detailed rationales for each answer, forming question-rationale-answer triplets ( _Qi, Ri, Ai_ ) for each sample. The structured nature of this dataset provides a rich source of contextual information that is essential for training models to understand and reason about both textual and visual data comprehensively. + Qualifying Examination in Economic Theory -**Retrieval Mechanism** The retrieval mechanism employed for identifying relevant triplets operates based on the cosine similarity between the query embedding _**h**[query] X_ and each triplet embedding _**h**[i] X_ +1. Ph.D. students are expected to sit for the Qualifying Examination during the first week of +classes in their third semester in the program, and once at the beginning of each subsequent +semester if they do not pass the Examination. -4 +2. The Qualifying Examination will be comprised of material covered in the courses ECON +8010 (Micro I), 8020 (Micro II), 8040 (Applied Math Econ), and 8050 (Macro Theory). -stored in the library. The cosine similarity is defined as: +3. The exam will be no longer than four hours in duration. -**==> picture [270 x 28] intentionally omitted <==** +4. Every effort will be made to inform students whether they pass or fail the Qualifying -For each query, the top- _k_ triplets with the highest similarity scores are retrieved. This process ensures that the most contextually relevant examples are selected. The retrieval process can be expressed as: +Examination no later than thirty-one days after taking the Examination. -_R_ ( _Xquery_ ) = _{_ ( _Qi, Ri, Ai_ ) _| i ∈_ Top- _k_ �sim( _**h**[query] X ,_ _**h**[i] X_[)] _[,][ ∀][i][ ∈{]_[1] _[, ..., N][}]_ � _},_ (4) where _R_ ( _Xquery_ ) denotes the set of retrieved triplets for the given query _Xquery_ , and Top- _k_ ( _·_ ) represents the function that selects the top- _k_ items based on the cosine similarity scores. +5. Students on stipend who do not pass the Qualifying Examination in their first attempt may -## **3.3 Learn to Reasoning from the Retrieved Content** +receive a reduced stipend in the subsequent semester. -Given the retrieved question-rationale-answer (QRA) data, we organize them into a structured format to teach the model how to reason, as illustrated in Figure 3. This section details the process in leveraging the retrieved content to enhance the model’s reasoning capabilities. +6. Students who fail the Qualifying Examination in their first or subsequent attempt must retake +the examination at its next administration, which will be the first week of the following +semester. -**==> picture [357 x 230] intentionally omitted <==** +7. A student on stipend who successfully passes the Qualifying Examination on this second try +will be eligible to have their stipend restored retroactively to the start of the relevant academic +year. -**----- Start of picture text -----**
-Query 𝓧𝒒𝒖𝒆𝒓𝒚 Result 𝓨𝒒𝒖𝒆𝒓𝒚
Question The rationale is: Magnet sizes affect the magnitude of the
Think about the magnetic force between the magnets in each pair. magnetic force … But Magnet B is smaller in Pair 2 than in Pair 1. So the answer is (C) The magnitude of the magnetic force is
Which of the following statements is true? smaller in Pair 2.
(A) The magnitude of the magnetic force is the same in both pairs.
(B) The magnitude of the magnetic force is smaller in Pair 1.
(C) The magnitude of the magnetic force is smaller in Pair 2.
Multimodal large language model
ScienceQA-based
Knowledge Library Content 𝓒(𝓧𝒒𝒖𝒆𝒓𝒚)
Retrieved 𝓡(𝓧𝒒𝒖𝒆𝒓𝒚)
For the question: ( 𝑸𝟏 ) Think about the
Question Question … 𝑿𝟏 𝑿𝟐 … magnetic force between the magnets in each pair. Which of the following statements is true? Referencing picture The rationale is: ( 𝑹𝟏 () 𝑰 Magnet sizes affect the 𝟏 ) 1.
Think about the magnetic force between the magnets in each pair. magnitude of the magnetic force
Which of the following statements is true? The answer is: ( 𝑨𝟏 ) The magnitude of the
magnetic force is greater in Pair 2.
(A) The magnitude of the magnetic force is greater in Pair 2. …
(B) The magnitude of the magnetic force is the same in both pairs.
(C) The magnitude of the magnetic force is greater in Pair 1. For the Current question: Think about ...
Referencing picture 6.
Imitate the process of generating rationales
based on the previous question, providing a
rationale and answer with the option's letter
from the given choices.
**----- End of picture text -----**
+8. A student who twice fails to pass the Qualifying Examination must take the Qualifying +Examination at its next administration. Such students who are on stipend will only be eligible +for non-reduced stipends if they successfully pass the examination this third time. Any +student failing the Qualifying Examination for the third time (1) will become ineligible for +Departmental support, including tuition, starting the subsequent semester, and (2) will be +advised to complete their studies as soon as possible. In most cases, students will have acquired +credits sufficient to earn an MA (en-route). +9. Students who neglect to take or retake the Qualifying Examination at the next available -Figure 3: The reasoning process from the retrieved content. The model uses the organized context from retrieved question-rationale-answer triplets to generate answers. +opportunity will be considered to have failed the Qualifying Examination. -For a given input _Xquery_ = ( _Tquery, Iquery_ ), suppose we retrieve the top- _k_ relevant QRA triplets _R_ ( _Xquery_ ) = _{_ ( _Qi, Ri, Ai_ ) _}[k] i_ =1[.][These][retrieved][triplets][are][used][to][form][a][context] _[C]_[(] _[X][query]_[)] which provides a structured set of examples to guide the model’s reasoning process. The context _C_ ( _Xquery_ ) is constructed as follows: +Minimum Competency in Econometrics -**==> picture [307 x 11] intentionally omitted <==** +1. Minimum Competency in Econometrics is established by taking and passing a three-course +sequence from among PhD-eligible Econometrics courses with a grade point average (GPA) +of 2.5 or higher – called the “Econometrics GPA.” -where each example is a concatenation of the question, rationale, and answer: +2. PhD-eligible Econometrics courses include ECON 8060, ECON 8070, ECON 8080, ECON +9090, and a variety of special topics courses listed under ECON 9000. Neither ECON 8600 +nor ECON 8700, nor any other courses designed for the MS/MA program are PhD-eligible. +If you are in doubt as to whether a course is PhD-eligible, please feel free to contract the +Graduate Coordinator. -**==> picture [318 x 12] intentionally omitted <==** +4 -where _⊕_ denotes the concatenation operation. The context _C_ ( _Xquery_ ) is then used to guide the model’s reasoning process, enabling it to learn from the retrieved content and generate accurate and contextually enriched answers. The model is prompted to reason based on the structured examples provided in the context, thereby predicting the answer _Yquery_ given the input _Xquery_ and the context _C_ ( _Xquery_ ). We define the conditional probability of generating the answer _Yquery_ as: + 3. All students must take and pass ECON 8060 and ECON 8070 with a grade of C or better. -**==> picture [337 x 11] intentionally omitted <==** +4. Students must also take a third course in econometrics and pass it with a grade of B or better. +This third course can be either Advanced Cross-Section (ECON 8080), Time Series (Econ +9090) Econometrics, or another suitable 9000-level course. All students, regardless of +specialization, are strongly advised to take ECON 8080. -where _F_ Θ represents the LLMs parameterized by Θ. The final answer _Yquery_ is obtained by: +5. Students who do not satisfy the GPA requirement after taking three PhD-eligible +Econometrics courses may satisfy the Minimum Competency requirement either by (1) taking +additional eligible courses as listed in item 2 above or (2) retaking a course to obtain a higher +grade the second time around. The Department will compute the Econometrics GPA based +on the three highest grades. NOTE: The Graduate School computes a student’s overall GPA +based on all courses taken; earning a higher grade the second time around does not +“overwrite” the grade earned the first time around. -**==> picture [293 x 17] intentionally omitted <==** +6. Students who fail to establish Minimum Competency in Econometrics by the end of their +fourth semester in the program will be advised to complete their studies with an MA degree. +Students on funding who fail to meet this standard will become ineligible for Departmental +support. -5 +Primary Field -## **4 Experiments** +The Department of Economics offers eight Primary Fields: +• Environmental and Natural Resource Economics +• Financial Economics +• Growth and Development +Industrial Organization +• +International Economics +• +• Labor Economics +• Monetary and Macroeconomics +• Public Economics -**Datasets** We employed four multimodal reasoning benchmarks: (i) _ScienceQA_ [49], a multimodal question dataset that includes over 21k multiple-choice questions, 3 subjects, 26 topics, 127 categories, and 379 distinct skills. (ii) _A-OKVQA_ [59], a knowledge-based multimodal dataset that includes 25k questions with extensive commonsense and world knowledge. (iii) _MMBench_ [47], a dataset comprising 2,974 multiple-choice questions covering 20 ability dimensions. (iv) _SEED-Bench_ [37], a large-scale dataset that includes 19k multiple-choice questions, and 12 evaluation dimensions, including both spatial and temporal understanding. We built the high-school knowledge library using the training data from ScienceQA, which employs over 12k question-rationale-answer triplets data. +The Primary Field requirement is satisfied by (1) taking the required courses and (2) either taking and +passing a written and/or oral Field Examination or writing a satisfactory Field Paper. Students should +consult the instructors of the Field to determine the precise requirements for their Field. Students +who fail to meet the Field requirements by the start of their fourth year in the program will be +ineligible for any departmental financial support. -**Baselines** For ScienceQA, we compared our approach against strong baselines across four categories, excluding vision LLMs that specifically fine-tune or train on ScienceQA for a fair comparison: (i) heuristic and expert-guided choices, such as random choice and human evaluation [49]; (ii) small multimodal visual question answering models, which include MCAN [80], Top-Down [1], BAN [34], DFAF [21], ViLT [35], Patch-TRM [51], and VisualBERT [40]; (iii) zero-shot instruction-tuned large language models like GPT-3.5 [6] and its CoT-enhanced variants [49], in addition to ChatGPT, GPT-4 [53], and Chameleon [50]; (iv) strong vision LLMs like LLaVA-1.5 [46], Qwen-VL [3], InternLM-XComposer2-VL [20], and Gemini [56]. Regarding the A-OKVQA dataset, the baselines include state-of-the-art approaches, such as Pythia [4], ViLBERT [48], LXMERT [65], KRISP [52], GPV-2 [31], BLIP-2 [39], PICa [75], IPVR [13], PromptCap [27], Prophet [79], PaLI-3-VPD [28], PaLI-X-VPD [28], and Gemini [56]. For the MMBench [47] and SEED-Bench [37] datasets, we mainly focus on augmenting Gemini with RMR for the convenient access of the Gemini API. +Secondary Field -## **4.1 Results on ScienceQA** +Students demonstrate proficiency in most Secondary Fields by taking a designated two-course sequence +(Econometrics is the sole exception). All Primary Fields may also serve as Secondary Fields. The +Department also offers two Fields as Secondary Fieldsonly: -Table 1: The comparison on ScienceQA dataset. Question classes: NAT = natural science, SOC = social science, LAN = language science, TXT = text context, IMG = image context, NO = no context, G1-6 = grades 1-6, G7-12 = grades 7-12. +• Anti-trust and Regulation -|Model
Size|NAT
SOC
LAN|NAT
SOC
LAN|NAT
SOC
LAN|TXT|IMG
NO|IMG
NO|G1-6 G7-12|G1-6 G7-12|AVG| -|---|---|---|---|---|---|---|---|---|---| -|Random Choice [49]
-
Human[49]
-|40.28 46.13 29.25
90.23 84.97 87.48|||47.75
89.60|40.08 33.66
87.50 88.10||39.35 40.67
91.59 82.42||39.83
88.40| -|MCAN [80]
95M
Top-Down [1]
70M
BAN [34]
112M
DFAF [21]
74M
ViLT [35]
113M
Patch-TRM [51]
90M
VisualBERT [40]
111M
UnifedQABase [33]
223M
UnifedQABase w/ CoT[49]
223M|56.08 46.23 58.09
59.50 54.33 61.82
60.88 46.57 66.64
64.03 48.82 63.55
60.48 63.89 60.27
65.19 46.79 65.55
59.33 69.18 61.18
68.16 69.18 74.91
71.00 76.04 78.91|||59.43
62.90
62.61
65.88
63.20
66.96
62.71
63.78
66.42|51.17 55.40
54.88 59.79
52.60 65.51
54.49 64.11
61.38 57.00
55.28 64.95
62.17 58.54
61.38 77.84
66.53 81.81||51.65 59.72
57.27 62.16
56.83 63.94
57.12 67.17
60.72 61.90
58.04 67.50
62.96 59.92
72.98 65.00
77.06 68.82||54.54
59.02
59.37
60.72
61.14
61.42
61.87
70.12
74.11| -|GPT-3.5 [49]
173B
GPT-3.5 w/ CoT [49]
173B
ChatGPT w/ CoT [53]
-
GPT-4 w/ CoT [53]
-
Chameleon + ChatGPT [50]
-
Chameleon + GPT-4[50]
-|74.64 69.74 76.00
75.44 70.87 78.09
78.82 70.98 83.18
85.48 72.44 **90.27**
81.62 70.64 84.00
89.83 74.13 89.82|||74.44
74.68
77.37
82.65
79.77
88.27|67.28 77.42
67.43 79.93
67.92 86.13
71.49 92.89
70.80 86.62
77.64 92.13||76.80 68.89
78.23 69.68
80.72 74.03
86.66 79.04
81.86 76.53
88.03 83.72||73.97
75.17
78.31
83.99
79.93
86.54| -|LLaVA-1.5 [46]
13B|70.12 76.72 67.64|||70.48|71.89 68.92||76.06
61.5||70.86| -|LLaVA-1.5+RMR
13B|78.11
84.25
74.73|||79.81|78.33
74.36||82.05
72.18||78.52| -|_+Improvement_
-|+_7.99_
+_7.53_
+_7.09_|||+_9.33_|+_6.44_
+_5.44_||+_5.99_
+_10.68_||+_7.66_| -|Qwen-VL [3]
/|67.01 66.37 59.36|||68.52|67.03 57.00||69.75 56.16||64.89| -|Qwen-VL+RMR
/|70.07
86.16
75.36|||70.77|76.50
72.68||78.67
67.90||74.82| -|_+Improvement_
/|+_3.06_
+_19.79_
+_16.00_|||+_2.25_|+_9.47_
+_15.68_||+_8.92_
+_11.74_||+_9.93_| -|InternLM-XComposer2-VL [20]
/|88.19|93.48|78.64|88.47|89.14|80.77|88.66|83.52|86.82| -|InternLM-XComposer2-VL+RMR
/|**94.85**|**97.19**|82.55|**95.11**|**96.03**|85.23|**93.98**|88.86|**92.15**| -|_+Improvement_
/|+_6.66_|+_3.71_|+_3.91_|+_6.64_|+_6.89_|+_4.46_|+_5.32_|+_5.34_|+_5.33_| -|Gemini [56]
/|59.68|74.24|41.73|57.72|64.01|47.46|65.20|45.29|58.08| -|Gemini+RMR
/|91.79|94.26|89.64|91.40|89.69|**91.01**|92.84|**89.78**|91.75| -|_+Improvement_
/
+_32.11_
+_20.02_
+_47.91_
+_33.68_
+_25.68_
+_43.55_
+_27.64_
+_44.49_
+_33.67_|||||||||| +• Econometrics, by establishing Proficiency in Econometrics (see below) +5 + Proficiency in Econometrics -6 +Students satisfy the requirements for Proficiency in Econometrics by earning a grade of B or better in +each of the following courses: ECON 8060, ECON 8070, and two other Advanced Econometrics +courses. Students who achieve Proficiency in Econometrics satisfy the requirements to declare +Econometrics as a Secondary Field (see below). -We set the number of retrieved examples _k_ to 3 by default. Table 1 illustrates the performance of RMR compared to various strong baselines across the ScienceQA dataset. Notably, our approach demonstrates significant improvements in reasoning capabilities without the need for fine-tuning or re-training on the target dataset, adhering to a zero-shot training setting. Despite the relatively low baseline performance of the vanilla Gemini model, which achieved an average accuracy of 58.08%, the integration of RMR remarkably enhances its reasoning capabilities, achieving an impressive average accuracy of 91.75%. This represents a substantial improvement of +33.67% over the baseline. Similarly, for the robust InternLM-XComposer2-VL model, our RMR framework yields a notable improvement of +5.33%, showcasing its effectiveness in boosting the reasoning capabilities of strong multimodal models. These results underscore the effectiveness of our proposed RMR framework in enhancing the performance of vision-language models across various question classes and contexts, demonstrating its potential to significantly advance the field of multimodal reasoning. +Second and Third-Year Papers -## **4.2 Results on A-OKVQA** +All second- and third-year students must turn in a paper at the end of the Spring semester to the +Graduate Program Advisor. Monetary prizes will be awarded for the best second year papers. It is +crucial that students receive faculty input about their paper topic well before the paper is due. +Students who fail to turn in a second or third-year paper will be ineligible for any departmental financial +support. A committee formed by the Graduate Program Director will award prizes for the best +papers, with advice from the instructors of the Field courses. -Table 2 presents the performance comparison for the direct-answer task on the A-OKVQA dataset. The table includes a variety of vision-language model combinations across different architectures. The models range from earlier architectures like Pythia and ViLBERT, to more recent and powerful systems such as BLIP-2, PaLM-CoT, and PaLI-X-VPD. Our RMR framework, when integrated with the Gemini model, demonstrates a substantial improvement in performance. The baseline Gemini model achieves a direct-answer accuracy of 44.2%, but with the incorporation of RMR, this accuracy increases significantly to 63.1%. This represents an impressive improvement of +18.9%. +Field Workshops -This improvement highlights the capability of the RMR framework to effectively augment the reasoning and answering performance of existing multimodal models, showcasing its potential to set new benchmarks in visual question answering. The results indicate that RMR not only competes with but also surpasses state-of-the-art methods across diverse model architectures and parameter sizes. +Students in the second year and beyond are required to attend and earn passing grades in Applied +Economics Field Workshops. Generally, these workshops are designed to introduce students to the +process of doing research. Students have the opportunity to develop their own ideas, as well as read, +examine critically, and comment on the work of others. -Table 2: The comparison of the direct-answer task on the A-OKVQA dataset. +Starting in their second year, students must enroll in Workshops in Applied Economics (courses +numbered 9820, sections vary). These workshops give students the opportunity to present their own +research and learn about and evaluate the research of others. The workshops are designed to involve +the student in all key aspects of the scientific process: defining a problem and the questions of interest, +devising a research strategy to shed light on the problem, communication of the findings, and learning +how to scrutinize and critique research, whether one’s own or that of others. The Workshops are an +ideal vehicle for discovering projects suitable for transformation into a PhD dissertation. Grading is +pass-fail. The criteria for passing, as well as procedures and other policies for each workshop will be +determined by the faculty leading that workshop. -|Model|Vision Model|Text Model|Parameters|Direct-answer| -|---|---|---|---|---| -|Pythia [4]|ResNet [26]|BERT [32]|70M|25.2| -|ViLBERT [48]|Faster R-CNN [57]|BERT [32]|300M|30.6| -|LXMERT [65]|Transformer [67]|Transformer [67]|220M|30.7| -|KRISP [52]|Faster R-CNN [57]|BERT [32]|200M|33.7| -|GPV-2[31]|VinVL[83]|T5-Base[55]|300M|48.6| -|BLIP-2 [39]|CLIP-VIT-LARGE [54]|FlanT5XXL [16]|11B|53.2| -|PaLM-CoT [69]|-|PaLM [15]|540B|41.5| -|PICa [75]|VinVL [83]|GPT-3 [7]|175B|42.4| -|IPVR [13]|Faster-RCNN [57]|OPT [84]|66B|46.4| -|PromptCap [27]|Ofa [68]|GPT-3 [7]|175B|56.3| -|Prophet [79]|MCAN [81]|GPT-3 [7]|175B|58.2| -|PaLI-3-VPD generalist [28]|SigLIP [82]|UL2 [66]|5B|56.5| -|PaLI-X-VPDgeneralist[28]|ViT[18]|UL2[66]|55B|62.7| -|Gemini [56]|-|-|-|44.2| -|Gemini+RMR|-|-|-|**63.1**| -|_+Improvement_|-|-|-|+_18.9_| +The Dissertation +The Ph.D. degree in economics requires 18 thesis hours, writing of a Ph.D. dissertation, and oral +defense of said dissertation. Timely completion of the Ph.D. degree requires that students choose a +dissertation topic and thesis advisor as soon as possible after completing their required courses.The +dissertation may (but need not) grow out of research initiated in a student’s 2nd year paper and/or +third year workshop. Regardless of the topic chosen, the dissertation must demonstrate to the student’s +mastery of the subject, an agenda for future research, and an ability to independently pursue that +research and other lines of scholarly inquiry. +1. Students are eligible to enroll for dissertation (ECON 9910) hours once they have passed the -## **4.3 Results on MMBench and SEED-Bench** +Qualifying Examination in Economic Theory. -Figure 4 illustrates the comparative performance of the Gemini model and its enhanced version, Gemini+RMR, on the MMBench-Dev and MMBench-Test datasets across various categories. The categories evaluated include overall performance, conceptual problems (CP), fine-grained perception of cross instances (FP-C), fine-grained perception of single instances, attribute reasoning (AR), logical reasoning (LR), and relation reasoning (RR). For the MMBench-Dev dataset, Gemini+RMR shows marked improvements in all categories compared to the baseline Gemini model. The overall performance of Gemini+RMR is 70.96%, compared to 57.9% for Gemini, representing a substantial improvement. Specifically, Gemini+RMR scores 76.01% in CP, 72.70% in FP-S, 65.03% in FP-C, 75.38% in AR, 55.93% in LR, and 68.70% in RR. These results indicate that the RMR framework significantly enhances the model’s capability to handle a wide range of multimodal reasoning tasks. +2. All PhD-eligible students should have found a Ph.D. Dissertation Advisor no later than +December of their fourth year in the program. The student, in consultation with the Advisor, +should assemble a committee that includes three additional faculty members, for a total of +four members. -7 +6 -Similarly, on the MMBench-Test dataset, Gemini+RMR outperforms the baseline Gemini model across all categories. The overall performance of Gemini+RMR is 67.26%, compared to 57.23% for Gemini. The specific category improvements are as follows: 69.16% in CP, 69.10% in FP-S, 61.54% in FP-C, 77.43% in AR, 53.18% in LR, and 63.98% in RR. These improvements further demonstrate the effectiveness of the RMR framework in enhancing the model’s reasoning capabilities across diverse evaluation metrics. Overall, the integration of the RMR framework with the Gemini model leads to significant performance improvements in both the MMBench-Dev and MMBench-Test datasets, underscoring the efficacy of our approach in advancing multimodal reasoning performance. + 3. The chair of this committee and at least two additional members of the committee must be -**==> picture [396 x 136] intentionally omitted <==** +tenure track faculty in the John E. Walker Department of Economics. -**----- Start of picture text -----**
-(a) Performance comparison on MMBench-Dev (b) Performance comparison on MMBench-Test
**----- End of picture text -----**
+II. +Sample Four-Year Curriculum -Figure 4: Comparative performance of Gemini and Gemini+RMR on the MMBench-Dev and MMBench-Test datasets. +The PhD requires at least 42 hours of graded coursework and at least 18 hours of dissertation. +Required Core courses are marked with an asterisk (*). ECON 90X0 (marked with #) may be +substituted for ECON 8080. Pass-Fail (P/F) courses do not count towards the 42 graded credit +hour requirement, and no P/F course counts towards the GPA. This sample curriculum does not +include the Qualifying Examination in Economic Theory, required of all students in the second year +(and above), given in August and January, or Field requirements (check with the relevant Professors +for those). -Figure 5 presents the comparative performance of the Gemini model and its enhanced version, Gemini+RMR, across various evaluation dimensions on the SEED-Bench dataset. The evaluation categories include overall performance, scene understanding (SU), instance identity (IId), instance attributes (IA), instance location (IL), instance counting (IC), instance interaction (IIn), visual reasoning (VR), spatial relation (SR), and text understanding (TU). The results show significant performance improvements across all categories when integrating the RMR framework with the Gemini model. The overall performance of Gemini+RMR is 64.72%, compared to 56.53% for Gemini. +Year 1: Fall -**==> picture [199 x 122] intentionally omitted <==** +*Math Camp – Starts August 1 +*ECON 8010 (3 Hours) -- Microeconomic Theory I +*ECON 8040 (3 Hours) -- Applied Math Econ +*ECON 8060 (3 Hours) -- Econometrics I +*ECON 9810 (1 Hour P/F) -- Introductory Workshop -Figure 5: Performance on SEED-Bench dataset. +Year 1: Spring -## **4.4 Ablation Study** +*ECON 8020 (3 Hours) -- Microeconomic Theory II +*ECON 8050 (3 Hours) -- Macroeconomic Theory I +*ECON 8070 (3 Hours) -- Econometrics II +*ECON 9810 (1 Hour P/F) -- Introductory Workshop -**Effect of Retrieval Size** We investigate the impact of varying the retrieval size on the performance of the RMR framework. The retrieval size _|R_ ( _Xquery_ ) _|_ is defined as the number of question-rationaleanswer (QRA) triplets retrieved for each query _Xquery_ . Table 3 presents the results of our ablation study on the ScienceQA dataset, showing the performance of Gemini with different retrieval sizes. +Year 2: Fall -Table 3: The impact of retrieval size on the performance of RMR on ScienceQA. +*ECON 8080 (3 Hours) Advanced Cross Section Econometrics +Field Course 1 (3 Hours) +Field Course 2 (3 Hours) +*ECON 9820 (3 Hours P/F) Field Workshop -|Model|NAT|SOC|LAN|TXT|IMG|NO|G1-6|G7-12|AVG| -|---|---|---|---|---|---|---|---|---|---| -|Gemini w/o retrieval|59.68|74.24|41.73|57.72|64.01|47.46|65.20|45.29|58.08| -|Gemini w/_|R_(_Xquery_)_|_= 1|88.41|89.43|84.82|87.34|84.53|87.11|89.72|84.05|87.69| -|Gemini w/_|R_(_Xquery_)_|_= 2|87.79|91.00|84.55|86.75|84.93|86.90|88.88|85.37|87.62| -|Gemini w/_|R_(_Xquery_)_|_= 3|91.79|94.26|89.64|91.40|89.69|91.01|92.84|89.78|91.75| -|Gemini w/_|R_(_Xquery_)_|_= 4|88.54|92.80|84.27|88.03|86.37|86.41|89.76|85.76|88.33| -|Gemini w/_|R_(_Xquery_)_|_= 5|87.83|89.09|84.45|86.46|83.59|86.69|89.24|83.59|87.22| +Year 2: Spring +#ECON 90X0 (3 Hours) Advanced Econometrics Course +Field Course 1 (3 Hours) +Field Course 2 (3 Hours) +*ECON 9820 (3 Hours P/F) Field Workshop +*Second Year Paper Due -8 +Year 3: Fall -The results indicate that the optimal retrieval size is _|R_ ( _Xquery_ ) _|_ = 3, which achieves the highest average accuracy of 91.75%. Increasing the retrieval size beyond this value leads to a slight decrease in performance, which may be attributed to (i) the inclusion of less relevant examples that could potentially introduce noise into the reasoning process and (ii) the longer context length that may hinder the model’s ability to effectively reason over the retrieved content. +Field Course 1 (3 Hours) +Field Course 2 (3 Hours) +*ECON 9820 (3 Hours P/F) Field Workshop +*ECON 9910 (3 Hours P/F) Dissertation Research -**Data modality** We compare the performance of the RMR framework using different data modalities: all available data (denoted as "-All"), only text-image pairs (denoted as "-T&I"), and only text data (denoted as "-T"). Table 4 presents the results of this ablation study on ScienceQA. RMR framework consistently enhances the performance of the models regardless of the data modality used, which can be attributed to the universality of its bi-modality retrieval module. +Year 3: Spring -Table 4: The impact of single vs. multi-modal retrieval on ScienceQA. +Field Course 1 (3 Hours) +Field Course 2 (3 Hours) +*ECON 9820 (3 Hours P/F) Field Workshop +*ECON 9910 (3 Hours P/F) Dissertation Research +*Third Year Paper Due -|Model|NAT|SOC|LAN|TXT|IMG|NO|G1-6|G7-12|AVG| -|---|---|---|---|---|---|---|---|---|---| -|LLaVA-All|70.12|76.72|67.64|70.48|71.89|68.92|76.06|61.5|70.86| -|w/ RMR|78.11|84.25|74.73|79.81|78.33|74.36|82.05|72.18|78.52| -|LLaVA-T&I|69.56|74.74|86.36|69.69|71.89|-|78.03|56.97|71.89| -|w/ RMR|75.19|82.33|95.45|75.58|78.33|-|82.58|68.03|78.33| -|LLaVA-T|70.76|88.80|66.86|71.74|-|68.92|73.90|64.37|69.92| -|w/ RMR|81.50|96.00|73.86|86.57|-|74.36|81.47|74.81|78.69| -|Qwen-VL-All|67.01|66.37|59.36|68.52|67.03|57.00|69.75|56.16|64.89| -|w/ RMR|70.07|86.16|75.36|70.77|76.50|72.68|78.67|67.90|74.82| -|Qwen-VL-T&I|67.25|66.10|77.27|65.23|67.03|-|73.06|52.38|67.03| -|w/ RMR|70.31|85.47|90.91|68.50|76.50|-|82.44|62.07|76.50| -|Qwen-VL-T|66.73|68.00|58.62|73.76|-|57.00|66.10|58.56|62.95| -|w/ RMR|69.80|90.40|74.72|74.40|-|72.68|74.52|71.58|73.29| -|InternLM-All|88.19|93.48|78.64|88.47|89.14|80.77|88.66|83.52|86.82| -|w/ RMR|94.85|97.19|82.55|95.11|96.03|85.23|93.98|88.86|92.15| -|InternLM-T&I|85.77|93.98|97.73|86.32|89.14|-|90.83|85.03|89.14| -|w/ RMR|95.29|97.25|95.45|95.31|96.03|-|96.78|94.22|96.03| -|InternLM-T|90.99|90.40|77.84|91.89|-|80.77|86.25|82.56|84.71| -|w/ RMR|94.34|96.80|82.01|94.8|-|85.23|90.89|85.47|88.62| -|Gemini-All|59.68|74.24|41.73|57.72|64.01|47.46|65.20|45.29|58.08| -|w/ RMR|91.79|94.26|89.64|91.40|89.69|91.01|92.84|89.78|91.75| -|Gemini-T&I|55.75|76.83|68.18|54.89|64.01|-|69.91|49.66|64.01| -|w/ RMR|86.85|93.72|97.73|87.03|89.69|-|91.18|86.05|89.69| -|Gemini-T|64.24|58.40|40.62|62.23|-|47.46|60.00|42.52|52.70| -|w/ RMR|97.51|97.60|89.30|98.35|-|91.01|94.67|92.14|93.62| +7 + Year 4: Fall and Spring +*ECON 9820 (3 Hours P/F) Field Workshop +*ECON 9910 (6 Hours P/F) Dissertation Research +Oral Dissertation Defense (Spring) -## **5 Conclusion and Limitation** +III. -In this work, we introduce RMR, a multimodal RAG framework designed to enhance the reasoning capabilities of vision LLMs. By leveraging a bi-modality retrieval module, RMR retrieves the most relevant question-rationale-answer triplets from a high-school knowledge library constructed using the ScienceQA dataset. The retrieved triplets are then utilized to form a structured context that guides the model’s reasoning process. Extensive experiments on multiple multimodal reasoning benchmarks demonstrate that RMR significantly improves the performance of various vision LLMs. +Fields and Requirements -Despite the promising results, our work has several limitations that warrant further investigation. First, while RMR operates in a training-free manner, which offers convenience and effectiveness, developing a trainable multimodal RAG model could potentially further enhance the reasoning capabilities of vision-language models by allowing the model to adapt more precisely to specific datasets and tasks. Additionally, the high-school knowledge library constructed by ScienceQA may not be comprehensive enough to cover all scenarios, especially for domain-specific questions. +Regularly offered fields (with annual or biennial course sequences) include Financial Econ, Industrial +Organization, Labor, Public, Growth & Development, and International. Field requirements and +selected faculty follow. Students should check with the Faculty members of the appropriateApplied +Economics Field Workshops to determine the current requirements for a Primary Field, that is, +whether a Field Exam, Field Paper, or both are required. +Anti-trust and Regulation (Secondary Field only) -9 +Antitrust (ECON 8250), Regulation (ECON 8260) -## **References** +Applied Econometrics (Secondary Field only) -- [1] Peter Anderson, Xiaodong He, Chris Buehler, Damien Teney, Mark Johnson, Stephen Gould, and Lei Zhang. Bottom-up and top-down attention for image captioning and visual question answering. In _Proceedings of the IEEE conference on computer vision and pattern recognition_ , pages 6077–6086, 2018. +Students who achieve Proficiency in Econometrics are considered to have satisfied the +requirements to declare Econometrics as a Secondary Field (see above). -- [2] Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. Self-rag: Learning to retrieve, generate, and critique through self-reflection. _arXiv preprint arXiv:2310.11511_ , 2023. +Environmental / Natural Resource Economics -- [3] Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond. 2023. +Two courses from Advanced Natural Resource Economics (ECON/APEC 8090), +Economics of Environmental Quality (ECON 8110), Seminar in Resource Economics +(ECON 9040) -- [4] Stella Biderman, Hailey Schoelkopf, Quentin Gregory Anthony, Herbie Bradley, Kyle O’Brien, Eric Hallahan, Mohammad Aflah Khan, Shivanshu Purohit, USVSN Sai Prashanth, Edward Raff, et al. Pythia: A suite for analyzing large language models across training and scaling. In _International Conference on Machine Learning_ , pages 2397–2430. PMLR, 2023. +Financial Economics -- [5] Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens. In _International conference on machine learning_ , pages 2206–2240. PMLR, 2022. +Financial Economics (ECON 8550) plus either Time Series Econometrics (ECON 9090) +or Computational Finance (MTHSC 9820) -- [6] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. _Advances in neural information processing systems_ , 33:1877–1901, 2020. +Growth and Development -- [7] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. _Advances in neural information processing systems_ , 33:1877–1901, 2020. +Economic Development (ECON 8310), Advanced Economic Growth (ECON 9160) -- [8] Deng Cai, Yan Wang, Lemao Liu, and Shuming Shi. Recent advances in retrieval-augmented text generation. In _Proceedings of the 45th international ACM SIGIR conference on research and development in information retrieval_ , pages 3417–3419, 2022. +Industrial Organization -- [9] Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng-Ann Heng, and Stan Z Li. A survey on generative diffusion models. _IEEE Transactions on Knowledge and Data Engineering_ , 2024. +Organization of Industry (ECON 8240), and either Antitrust (ECON 8250), Theory +of Regulation (ECON 8260), Property Rights (ECON 8270), or Advanced IO (ECON +9240) -- [10] Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, et al. A survey on evaluation of large language models. _ACM Transactions on Intelligent Systems and Technology_ , 15:1–45, 2024. +International Economics -- [11] Jiawei Chen, Hongyu Lin, Xianpei Han, and Le Sun. Benchmarking large language models in retrieval-augmented generation. In _Proceedings of the AAAI_ , volume 38, pages 17754–17762, 2024. +International Trade (ECON 8400), Int’l Finance (ECON 8410) -- [12] Wenhu Chen, Hexiang Hu, Xi Chen, Pat Verga, and William Cohen. Murag: Multimodal retrieval-augmented generator for open question answering over images and text. In _Proceedings of EMNLP_ , pages 5558–5570, 2022. +Labor Economics -- [13] Zhenfang Chen, Qinhong Zhou, Yikang Shen, Yining Hong, Hao Zhang, and Chuang Gan. See, think, confirm: Interactive prompting between vision and language models for knowledge-based visual reasoning. _arXiv preprint arXiv:2301.05226_ , 2023. +Labor Economics (ECON 8160), Advanced Labor (ECON 9170) -- [14] Xin Cheng, Di Luo, Xiuying Chen, Lemao Liu, Dongyan Zhao, and Rui Yan. Lift yourself up: Retrieval-augmented text generation with self-memory. _Advances in Neural Information Processing Systems_ , 36, 2024. +Monetary Economics -- [15] Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. _arXiv preprint arXiv:2204.02311_ , 2022. +Monetary Economics (ECON 9500), Int’l Finance (ECON 8410) -- [16] Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. Scaling instruction-finetuned language models. _arXiv preprint arXiv:2210.11416_ , 2022. +Public Economics / Public Choice -10 +Public Finance (ECON 8200) and either Public Choice (ECON 8210) or Empirical +Public Economics (ECON 9200) -- [17] Wanqing Cui, Keping Bi, Jiafeng Guo, and Xueqi Cheng. More: Multi-modal retrieval augmented generative commonsense reasoning. _arXiv preprint arXiv:2402.13625_ , 2024. +8 -- [18] Mostafa Dehghani, Josip Djolonga, Basil Mustafa, Piotr Padlewski, Jonathan Heek, Justin Gilmer, Andreas Peter Steiner, Mathilde Caron, Robert Geirhos, Ibrahim Alabdulmohsin, et al. Scaling vision transformers to 22 billion parameters. In _International Conference on Machine Learning_ , pages 7480–7512. PMLR, 2023. + IV. -- [19] Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, and Zhifang Sui. A survey on in-context learning. _arXiv preprint arXiv:2301.00234_ , 2022. +Standards of Performance -- [20] Xiaoyi Dong, Pan Zhang, Yuhang Zang, Yuhang Cao, Bin Wang, Linke Ouyang, Xilin Wei, Songyang Zhang, Haodong Duan, Maosong Cao, et al. Internlm-xcomposer2: Mastering free-form text-image composition and comprehension in vision-language large model. _arXiv preprint arXiv:2401.16420_ , 2024. +Annual Review: The Graduate Student Progress Report -- [21] Peng Gao, Zhengkai Jiang, Haoxuan You, Pan Lu, Steven CH Hoi, Xiaogang Wang, and Hongsheng Li. Dynamic fusion with intra-and inter-modality attention flow for visual question answering. In _Proceedings of the IEEE/CVF conference on computer vision and pattern recognition_ , pages 6639–6648, 2019. +Each year, Ph.D. students must turn in a progress report that includes information on the year started +in the Ph.D. program, date of passage of the Qualifying Examination in Economic Theory, the year +in which the student satisfied the Minimum Sufficiency Econometrics requirement, graduate assistant +duties, Primary and Second field progress, research activities (workshops attended, papers presented, +and progress on the dissertation), and honors and awards received. The information collected may +change according to Departmental needs. In addition, Faculty are requested to evaluate the quality of +the services of Graduate Research or Teaching Assistants assigned to them on a scale of 1 to 5, where +1 indicates poor or very poor performance and 5 indicates performance beyond expectations. -- [22] Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, and Haofen Wang. Retrieval-augmented generation for large language models: A survey. _arXiv preprint arXiv:2312.10997_ , 2023. +Each Ph.D. student is ranked annually on a scale of 1-5. The Department, in the strongest terms, +discourages making any analogies with academic grading scales. The purpose of the ranking is not to +distinguish finely between students on academic or service merit, but to offer the broadest of +information on how it sees things going. The rankings should be interpreted as follows. -- [23] Zhangyang Gao, Cheng Tan, Lirong Wu, and Stan Z Li. Simvp: Simpler yet better video prediction. In _Proceedings of the IEEE/CVF conference on computer vision and pattern recognition_ , pages 3170–3180, 2022. +1. Student should depart the Ph.D. program immediately. -- [24] Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. Retrieval augmented language model pre-training. In _International conference on machine learning_ , pages 3929–3938. PMLR, 2020. +2. Student is making less than satisfactory progress, and funding (if applicable) is in jeopardy. -- [25] Junxian He, Graham Neubig, and Taylor Berg-Kirkpatrick. Efficient nearest neighbor language models. _arXiv preprint arXiv:2109.04212_ , 2021. +3. Student is making satisfactory (good or very good) progress. -- [26] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In _Proceedings of the IEEE conference on computer vision and pattern recognition_ , pages 770–778, 2016. +4. Student performance is excellent. (Example: good papers or dissertation chapter written, -- [27] Yushi Hu, Hang Hua, Zhengyuan Yang, Weijia Shi, Noah A Smith, and Jiebo Luo. Promptcap: Prompt-guided task-aware image captioning. _arXiv preprint arXiv:2211.09699_ , 2022. +work presented at conference.) -- [28] Yushi Hu, Otilia Stretcu, Chun-Ta Lu, Krishnamurthy Viswanathan, Kenji Hata, Enming Luo, Ranjay Krishna, and Ariel Fuxman. Visual program distillation: Distilling tools and programmatic reasoning into vision-language models. _arXiv preprint arXiv:2312.03052_ , 2023. +5. Student performance is outstanding. (Example: sole-authored paper is published in -- [29] Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Atlas: Few-shot learning with retrieval augmented language models. _Journal of Machine Learning Research_ , 24:1–43, 2023. +reputable journal, or student receives fully funded scholarship from reputablesource.) -- [30] Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. Active retrieval augmented generation. _arXiv preprint arXiv:2305.06983_ , 2023. +The Department has no firm targets but expects 75-85% of students to earn a grade of 3, 15-25% a +grade of 4, and 5-10% a grade below 3. The Department expects, too, that students may earn just one +or two grades above 3 during their Ph.D. studies; that a student who earns a 4 one year may earn 3s +in subsequent years. -- [31] Amita Kamath, Christopher Clark, Tanmay Gupta, Eric Kolve, Derek Hoiem, and Aniruddha Kembhavi. Webly supervised concept expansion for general purpose vision models. In _European Conference on Computer Vision_ , pages 662–681. Springer, 2022. +Criteria for Renewal of Assistantships -- [32] Jacob Devlin Ming-Wei Chang Kenton and Lee Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In _Proceedings of naacL-HLT_ , volume 1, page 2, 2019. +The Department awards both one-time, non-renewable assistantships and renewable assistantships. +Renewal of renewable Assistantships and Fellowships naturally requires that funds be available for +the Department to spend. Beyond that, it is not possible to list every possible criterion by which the +decision to renew an Assistantship will be made. At a minimum, each student: -- [33] Daniel Khashabi, Sewon Min, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and Hannaneh Hajishirzi. UNIFIEDQA: Crossing format boundaries with a single QA system. In Trevor Cohn, Yulan He, and Yang Liu, editors, _Findings of the Association for Computational Linguistics: EMNLP 2020_ , pages 1896–1907, Online, November 2020. Association for Computational Linguistics. +1. Must be in good academic standing, not on academic probation, and have maintained a -11 +GPR of 3.0 or better. -- [34] Jin-Hwa Kim, Jaehyun Jun, and Byoung-Tak Zhang. Bilinear attention networks. _Advances in neural information processing systems_ , 31, 2018. +2. Must pass the Qualifying Examination in economic theory in timely fashion – see -- [35] Wonjae Kim, Bokyung Son, and Ildoo Kim. Vilt: Vision-and-language transformer without convolution or region supervision. In _International Conference on Machine Learning_ , pages 5583–5594. PMLR, 2021. +Section III. -- [36] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. _Advances in Neural Information Processing Systems_ , 33:9459–9474, 2020. +3. Must meet the Econometrics Minimum Competency Requirement in timely fashion. -- [37] Bohao Li, Rui Wang, Guangzhi Wang, Yuying Ge, Yixiao Ge, and Ying Shan. Seedbench: Benchmarking multimodal llms with generative comprehension. _arXiv preprint arXiv:2307.16125_ , 2023. +9 -- [38] Huayang Li, Yixuan Su, Deng Cai, Yan Wang, and Lemao Liu. A survey on retrieval-augmented text generation. _arXiv preprint arXiv:2202.01110_ , 2022. + 4. Must be on track to complete their degree in a timely fashion, taking all courses required +of them on time, and complete the Primary and Secondary fields no later than the end of +the third year. -- [39] Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping languageimage pre-training with frozen image encoders and large language models. _arXiv preprint arXiv:2301.12597_ , 2023. +5. Must submit the second and third-year papers on time. -- [40] Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. What does bert with vision look at? In _Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics_ , pages 5265–5275, 2020. +6. Must have chosen a Dissertation Advisor no later than the start of the fourth year. -- [41] Siyuan Li, Weiyang Jin, Zedong Wang, Fang Wu, Zicheng Liu, Cheng Tan, and Stan Z Li. Semireward: A general reward model for semi-supervised learning. _arXiv preprint arXiv:2310.03013_ , 2023. +7. Should have written at least one chapter of their Dissertation by the end of the fourth year. -- [42] Siyuan Li, Zicheng Liu, Juanxi Tian, Ge Wang, Zedong Wang, Weiyang Jin, Di Wu, Cheng Tan, Tao Lin, Yang Liu, et al. Switch ema: A free lunch for better flatness and sharpness. _arXiv preprint arXiv:2402.09240_ , 2024. +8. Should generally expect to receive a maximum of five years of funding. However, +students whose work and accomplishments at the end of the fifth year are unusually +strong may, contingent on the availability of funds and the judgment of the +Department Head, be offered one additional year of funding. On the other hand, students +whose progress falls short may receive reduced funding in their fifth year. Students who +demonstrate particularly slow progress may receive minimal funding, or no funding at all beyond +the fourth year. -- [43] Siyuan Li, Zedong Wang, Zicheng Liu, Cheng Tan, Haitao Lin, Di Wu, Zhiyuan Chen, Jiangbin Zheng, and Stan Z Li. Moganet: Multi-order gated aggregation network. In _The Twelfth International Conference on Learning Representations_ , 2023. +9. Must comply with Departmental requests for materials and information on time. -- [44] Siyuan Li, Luyuan Zhang, Zedong Wang, Di Wu, Lirong Wu, Zicheng Liu, Jun Xia, Cheng Tan, Yang Liu, Baigui Sun, et al. Masked modeling for self-supervised representation learning on vision and beyond. _arXiv preprint arXiv:2401.00897_ , 2023. +10. Must satisfactorily carry out any and all duties assigned to them. For international -- [45] Bingshuai Liu, Chenyang Lyu, Zijun Min, Zhanyu Wang, Jinsong Su, and Longyue Wang. Retrieval-augmented multi-modal chain-of-thoughts reasoning for large language models. _arXiv preprint arXiv:2312.01714_ , 2023. +students, this includes demonstrating proficiency in the English language. -- [46] Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. _arXiv preprint arXiv:2304.08485_ , 2023. +11. In the rare case that a student fails to satisfactorily carry out his or her duties, he or she +may have their funding reduced and be placed on Departmental Probation. Such a student +will have one semester in which to demonstrate their willingness and ability to provide a +satisfactory level of service. If, unfortunately, this does not occur, that student will have +their funding reduced to a minimal level or rescinded entirely. -- [47] Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. Mmbench: Is your multi-modal model an all-around player? _arXiv preprint arXiv:2307.06281_ , 2023. +Criteria for Renewal of Fellowships -- [48] Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. _Advances in neural information processing systems_ , 32, 2019. +Fellowship recipients occupy positions of particularly high honor and responsibility. Conditional on +satisfactory performance, Fellowship recipients can generally count on support for the first three +years of their study. Starting at the end of the third year, the performance of Fellowship recipients +will be carefully reviewed on a semester-by-semester basis, and the decision of whether to continue +to fund a Fellowship recipient will be made. Funds freed up in the course of the evaluation process +may be reallocated. In addition to satisfying the criteria for Renewal of Assistantships, Fellowship +recipients are expected to demonstrate superior performance in their coursework, scholarship, and, +when applicable, performance of duties assigned. It is not possible to list all of the factors that will +be considered in the determination of superior performance but will certainly include the quality of +written work to date, including journal publications and book publications, research grants obtained, +and presentations at professional meetings. -- [49] Pan Lu, Swaroop Mishra, Tanglin Xia, Liang Qiu, Kai-Wei Chang, Song-Chun Zhu, Oyvind Tafjord, Peter Clark, and Ashwin Kalyan. Learn to explain: Multimodal reasoning via thought chains for science question answering. _Advances in Neural Information Processing Systems_ , 35:2507–2521, 2022. +10 -- [50] Pan Lu, Baolin Peng, Hao Cheng, Michel Galley, Kai-Wei Chang, Ying Nian Wu, Song-Chun Zhu, and Jianfeng Gao. Chameleon: Plug-and-play compositional reasoning with large language models. _arXiv preprint arXiv:2304.09842_ , 2023. + Procedures -- [51] Pan Lu, Liang Qiu, Jiaqi Chen, Tony Xia, Yizhou Zhao, Wei Zhang, Zhou Yu, Xiaodan Liang, and Song-Chun Zhu. Iconqa: A new benchmark for abstract diagram understanding and visual language reasoning. _arXiv preprint arXiv:2110.13214_ , 2021. +Each student is responsible for informing themselves about the procedures and policies of the +University and Graduate School. This section is not meant to substitute for the information available +on the University websites, but hopefully will serve as a reasonable introduction to help students get +off to a smoother start. Some useful Graduate School Policies sites are as follows: +Policies and procedures: https://www.clemson.edu/graduate/students/policies- +procedures/index.html +Forms: https://www.clemson.edu/graduate/academics/forms-and-requests/ -12 +Deadlines: +https://www.clemson.edu/graduate/academics/deadlines- +for-completing-your-program.html -- [52] Kenneth Marino, Xinlei Chen, Devi Parikh, Abhinav Gupta, and Marcus Rohrbach. Krisp: Integrating implicit and symbolic knowledge for open-domain knowledge-based vqa. In _Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition_ , pages 14111–14121, 2021. +Although the information in this section is correct at the time of this writing, students are responsible +for any and all subsequent changes. Students must consult the Graduate School Announcements and +the links found on the web page of the Graduate School for up-to-date information. -- [53] OpenAI. Gpt-4 technical report, 2023. +Form GS2 Plan of Study -- [54] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In _International conference on machine learning_ , pages 8748–8763. PMLR, 2021. +The GS2 form serves the purposes of both planning and informing the Graduate School about courses +taken. As such, the form must be filed at the start of your academic career, and should be updated at +key points in time, and in particular, when you plan on applying for a degree. The last time a student +files the GS-2 form is particularly crucial because Enrolled Student Services determines a student’s +eligibility for graduation by checking each course listed against the student’s transcript. The GS2 is +currently available at https://www.clemson.edu/graduate/academics/plan-of-study.html. Each +student’s course of study will vary but will have certain elements in common. A sample curriculum +for a PhD student is shown on the following page. -- [55] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. _The Journal of Machine Learning Research_ , 21(1):5485–5551, 2020. +The MA and MS Degree En Route to the PhD -- [56] Machel Reid, Nikolay Savinov, Denis Teplyashin, Dmitry Lepikhin, Timothy Lillicrap, Jeanbaptiste Alayrac, Radu Soricut, Angeliki Lazaridou, Orhan Firat, Julian Schrittwieser, et al. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. _arXiv preprint arXiv:2403.05530_ , 2024. +Students in the Ph.D. are eligible to receive an MA or MS degree upon the successful completion of 30 +credit hours provided that they have earned passing grades in all Ph.D. core courses: ECON 8010, +8020, 8050, 8060, 8070, and 8080 or 9090. The Master of Science in Economic Analytics is a STEM- +certified degree. Completion of the MS degree allows international students with an F-1 VISA to apply +for a 24-month extension of their post-completion optional practical training (OPT). -- [57] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. _Advances in neural information processing systems_ , 28, 2015. +Students who apply for an MA or MS En Route must submit separate GS2 forms for the MA/MS and +PhD. The GS2 submitted for the MA/MS should include 30 graded credit hours, with the remaining +12 or more graded credit hours listed on the GS2 submitted for the PhD. -- [58] Dustin Schwenk, Apoorv Khandelwal, Christopher Clark, Kenneth Marino, and Roozbeh Mottaghi. A-okvqa: A benchmark for visual question answering using world knowledge. In _ECCV_ , pages 146–162. Springer, 2022. +11 -- [59] Dustin Schwenk, Apoorv Khandelwal, Christopher Clark, Kenneth Marino, and Roozbeh Mottaghi. A-okvqa: A benchmark for visual question answering using world knowledge. In _European Conference on Computer Vision_ , pages 146–162. Springer, 2022. + Sample GS2 for Economics PhD -- [60] Cheng Tan, Zhangyang Gao, Siyuan Li, and Stan Z Li. Simvp: Towards simple yet powerful spatiotemporal predictive learning. _arXiv preprint arXiv:2211.12509_ , 2022. +ECON 8010* -- [61] Cheng Tan, Zhangyang Gao, Lirong Wu, Yongjie Xu, Jun Xia, Siyuan Li, and Stan Z Li. Temporal attention unit: Towards efficient spatiotemporal predictive learning. In _Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition_ , pages 18770–18782, 2023. +ECON 8040* -- [62] Cheng Tan, Siyuan Li, Zhangyang Gao, Wenfei Guan, Zedong Wang, Zicheng Liu, Lirong Wu, and Stan Z Li. Openstl: A comprehensive benchmark of spatio-temporal predictive learning. _Advances in Neural Information Processing Systems_ , 36:69819–69831, 2023. +ECON 8060* -- [63] Cheng Tan, Jingxuan Wei, Zhangyang Gao, Linzhuang Sun, Siyuan Li, Xihong Yang, and Stan Z Li. Boosting the power of small multimodal reasoning models to match larger models with self-consistency training. _arXiv preprint arXiv:2311.14109_ , 2023. +ECON 9810* -- [64] Cheng Tan, Jun Xia, Lirong Wu, and Stan Z Li. Co-learning: Learning from noisy labels with self-supervision. In _Proceedings of the 29th ACM International Conference on Multimedia_ , pages 1405–1413, 2021. +ECON 8020* -- [65] Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from transformers. In _Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing_ , 2019. +ECON 8050* -- [66] Yi Tay, Mostafa Dehghani, Vinh Q Tran, Xavier Garcia, Jason Wei, Xuezhi Wang, Hyung Won Chung, Siamak Shakeri, Dara Bahri, Tal Schuster, et al. Ul2: Unifying language learning paradigms. _arXiv preprint arXiv:2205.05131_ , 2022. +ECON 8070* -- [67] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. _Advances in neural information processing systems_ , 30, 2017. +ECON 9810* -- [68] Peng Wang, An Yang, Rui Men, Junyang Lin, Shuai Bai, Zhikang Li, Jianxin Ma, Chang Zhou, Jingren Zhou, and Hongxia Yang. Ofa: Unifying architectures, tasks, and modalities through a simple sequence-to-sequence learning framework. In _International Conference on Machine Learning_ , pages 23318–23340. PMLR, 2022. +ECON 8980 -13 +ECON 9010 -- [69] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. _Advances in Neural Information Processing Systems_ , 35:24824–24837, 2022. +ECON 8080*# -- [70] Jingxuan Wei, Cheng Tan, Zhangyang Gao, Linzhuang Sun, Siyuan Li, Bihui Yu, Ruifeng Guo, and Stan Z Li. Enhancing human-like multi-modal reasoning: A new challenging dataset and comprehensive framework. _arXiv preprint arXiv:2307.12626_ , 2023. +ECON 9820* -- [71] Fangyuan Xu, Weijia Shi, and Eunsol Choi. Recomp: Improving retrieval-augmented lms with compression and selective augmentation. _arXiv preprint arXiv:2310.04408_ , 2023. +ECON 8160# -- [72] Xin Xu, Yue Liu, Panupong Pasupat, Mehran Kazemi, et al. In-context learning with retrieved demonstrations for language models: A survey. _arXiv preprint arXiv:2401.11624_ , 2024. +ECON 8240# -- [73] Zhipeng Xu, Zhenghao Liu, Yibin Liu, Chenyan Xiong, Yukun Yan, Shuo Wang, Shi Yu, Zhiyuan Liu, and Ge Yu. Activerag: Revealing the treasures of knowledge via active learning. _arXiv preprint arXiv:2402.13547_ , 2024. +ECON 9000# -- [74] Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, and Zhen-Hua Ling. Corrective retrieval augmented generation. _arXiv preprint arXiv:2401.15884_ , 2024. +ECON 9820* -- [75] Zhengyuan Yang, Zhe Gan, Jianfeng Wang, Xiaowei Hu, Yumao Lu, Zicheng Liu, and Lijuan Wang. An empirical study of gpt-3 for few-shot knowledge-based vqa. In _Proceedings of the AAAI Conference on Artificial Intelligence_ , volume 36, pages 3081–3089, 2022. +ECON 9170# -- [76] Michihiro Yasunaga, Armen Aghajanyan, Weijia Shi, Rich James, Jure Leskovec, Percy Liang, Mike Lewis, Luke Zettlemoyer, and Wen-tau Yih. Retrieval-augmented multimodal language modeling. In _Proceedings of ICML_ , pages 39755–39769, 2023. +ECON 8250# -- [77] Wenhao Yu. Retrieval-augmented generation across heterogeneous knowledge. In _Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Student Research Workshop_ , pages 52–58, 2022. +ECON 8220 -- [78] Wenhao Yu, Hongming Zhang, Xiaoman Pan, Kaixin Ma, Hongwei Wang, and Dong Yu. Chain-of-note: Enhancing robustness in retrieval-augmented language models. _arXiv preprint arXiv:2311.09210_ , 2023. +ECON 9820* -- [79] Zhou Yu, Xuecheng Ouyang, Zhenwei Shao, Meng Wang, and Jun Yu. Prophet: Prompting large language models with complementary answer heuristics for knowledge-based visual question answering. _arXiv preprint arXiv:2303.01903_ , 2023. +ECON 8260 -- [80] Zhou Yu, Jun Yu, Yuhao Cui, Dacheng Tao, and Qi Tian. Deep modular co-attention networks for visual question answering. In _Proceedings of the IEEE/CVF conference on computer vision and pattern recognition_ , pages 6281–6290, 2019. +ECON 9910* -- [81] Zhou Yu, Jun Yu, Yuhao Cui, Dacheng Tao, and Qi Tian. Deep modular co-attention networks for visual question answering. In _Proceedings of the IEEE/CVF conference on computer vision and pattern recognition_ , pages 6281–6290, 2019. +ECON 9820* -- [82] Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In _Proceedings of the IEEE/CVF International Conference on Computer Vision_ , pages 11975–11986, 2023. +ECON 9910* -- [83] Pengchuan Zhang, Xiujun Li, Xiaowei Hu, Jianwei Yang, Lei Zhang, Lijuan Wang, Yejin Choi, and Jianfeng Gao. Vinvl: Revisiting visual representations in vision-language models. In _Proceedings of the IEEE/CVF conference on computer vision and pattern recognition_ , pages 5579–5588, 2021. +ECON 9820* -- [84] Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, et al. Opt: Open pre-trained transformer language models. _arXiv preprint arXiv:2205.01068_ , 2022. +ECON 9910* -- [85] Yuanhan Zhang, Kaiyang Zhou, and Ziwei Liu. What makes good examples for visual in-context learning? _Advances in Neural Information Processing Systems_ , 36, 2024. +ECON 9820* -- [86] Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, and Mengnan Du. Explainability for large language models: A survey. _ACM Transactions on Intelligent Systems and Technology_ , 15:1–38, 2024. +Micro I -- [87] Penghao Zhao, Hailin Zhang, Qinhan Yu, Zhengren Wang, Yunteng Geng, Fangcheng Fu, Ling Yang, Wentao Zhang, and Bin Cui. Retrieval-augmented generation for ai-generated content: A survey. _arXiv preprint arXiv:2402.19473_ , 2024. +Math Econ -14 +Econometrics I -## **A Qualitative Analysis** +Workshop -_**Rationale**_ : An object has different properties. A property of an object can tell you how it looks, feels, tastes, or smells. Properties can also tell you how an object will behave when something happens to it. Different objects can have properties in common. Look at each object. For each object, decide if it has that property. A sticky object can attach or stick to other things. The toothpaste is sticky, but the soccer shorts and the water slide are not. This color is blue. All three objects are blue. A hard object does not change shape when pressed or squeezed. The water slide is hard, but the toothpaste and the soccer shorts are not. The property that all three objects have in common is blue. +Micro II -**==> picture [347 x 269] intentionally omitted <==** +Macro I -**----- Start of picture text -----**
-Query 𝓧𝒒𝒖𝒆𝒓𝒚 looks, feels, tastes, or smells. Properties can also tell you how an object will behave
something happens to it. Different objects can have properties in common. Look at
Question : Which property do these three objects object. For each object, decide if it has that property. A sticky object can attach or stick
have in common? other things. The toothpaste is sticky, but the soccer shorts and the water slide are not. This
Options : color is blue. All three objects are blue. A hard object does not change shape when pressed
or squeezed. The water slide is hard, but the toothpaste and the soccer shorts are not.
(A) Blue. property that all three objects have in common is blue.
(B) Hard. Answer : The answer is (A) Blue.
(C) Sticky.
Multimodal large language model
𝑿𝟏 𝑿𝟐 𝑿𝟑
Question : Which property do these three objects Question : Which property matches this Question : Which property do these two
have in common? object? objects have in common?
Options : Options : Options :
(A) Fragile. (A) Blue. (A) Bouncy.
(B) Blue. (B) Hard. (B) Soft.
(C) Sticky Rationale : An object has different
Rationale : An object has different properties. A Rationale : Look at the object. Think properties. A property of an object can
property of an object can tell you how it looks, about each property. A hard object tell you how it looks, feels, tastes, or
feels, tastes, or smells. Properties can also tell keeps its shape when you squeeze it. smells. Different objects can have the
you how an object will behave when something The jeans are not hard. Blue is a color. same properties. Look at each object.
happens to it. Different objects can have This color is blue. The jeans are blue. For each object, decide if it has that
properties in common. Look at each object. For Answer : The answer is (A) Blue. property. A soft object changes shape
each object, decide if it has that property. A when you squeeze it. Both objects are
sticky object can attach or stick to other things. soft. A bouncy object will bounce back
The toothpaste is sticky, but the jeans and the from the floor if you drop it. Neither of
track suit are not. Blue is a color. This color is the objects are bouncy. The property
blue. All three objects are blue. A fragile object that both objects have in common is soft.
will break into pieces if you drop it. The jeans
andthat all threethe trackobjectssuit arehave innot fragile.commonTheis blue.property Answer : The answer is (B) Soft.
Answer : The answer is (B) Blue.
**----- End of picture text -----**
+Econometrics II +Workshop -Figure 6: The retrieved data of an image-text pair example. +Macro II -_**Rationale**_ : Guide words appear on each page of a dictionary. They tell you the first word and Query 𝓧𝒒𝒖𝒆𝒓𝒚 last word on the page. The other words on the page come between the guide words in alphabetical order. To put words in alphabetical order, put them in order by their first letters. If _**Question**_ : Which word would you find on a the first letters are the same, look at the second letters. If the second letters are the same, look dictionary page with the following guide at the third letters, and so on. If one word is shorter, and there are no more letters to compare, words? imitate - iron then the shorter word comes first in alphabetical order. For example, be comes before bed. Put _**Options**_ : the words in alphabetical order. Since inherit is between the guide words imitate - iron, it would be found on that page. (A) Itch. _**Answer**_ : The answer is (B) Inherit. (B) Inherit. Multimodal large language model 𝑿𝟏 𝑿𝟐 𝑿𝟑 _**Question**_ : Which word would you find on a _**Question**_ : Would you find the word instruct _**Question**_ : Which word would you find on dictionary page with the following guide on a dictionary page with the following guide a dictionary page with the following guide words? idea - inspire words? imp - item words? impose - issue _**Options**_ : _**Options**_ : _**Options**_ : (A) Issue. (A) No. (A) Ill. (B) Immense. (B) Yes. (B) Increase. _**Rationale**_ : Guide words appear on each _**Rationale**_ : Guide words appear on each page _**Rationale**_ : Guide words appear on each page of a dictionary. They tell you the first of a dictionary. They tell you the first word page of a dictionary. They tell you the first word and last word on the page. The other and last word on the page. The other words word and last word on the page. The other words on the page come between the guide on the page come between the guide words in words on the page come between the guide words in alphabetical order. To put words alphabetical order. To put words in words in alphabetical order. To put words in alphabetical order, put them in order by alphabetical order, put them in order by their in alphabetical order, put them in order by their first letters. If the first letters are the first letters. If the first letters are the same, their first letters. If the first letters are the same, look at the second letters. If the look at the second letters. If the second letters same, look at the second letters. If the second letters are the same, look at the third are the same, look at the third letters, and so second letters are the same, look at the letters, and so on. If one word is shorter, on. If one word is shorter, and there are no third letters, and so on. If one word is and there are no more letters to compare, more letters to compare, then the shorter shorter, and there are no more letters to then the shorter word comes first in word comes first in alphabetical order. For compare, then the shorter word comes first alphabetical order. For example, be comes example, be comes before bed. Put the words in alphabetical order. For example, be before bed. Put the words in alphabetical in alphabetical order. Since instruct is comes before bed. Put the words in order. Since immense is between the guide between the guide words imp - item, it would alphabetical order. Since increase is words idea - inspire, it would be found on be found on that page. between the guide words impose - issue, it that page. _**Answer**_ : The answer is (B) Yes. would be found on that page. _**Answer**_ : The answer is (B) immense. _**Answer**_ : The answer is (B) Increase. +Micro III -Figure 7: The retrieved data of a text-only example. +Econometrics III -15 +Workshop -**==> picture [133 x 58] intentionally omitted <==** +Labor I (Secondary) -**----- Start of picture text -----**
-Query 𝓧𝒒𝒖𝒆𝒓𝒚
Question : Is marble a mineral or a rock?
Options :
(A) Rock.
(B) Mineral.
**----- End of picture text -----**
+IO (Primary) +Econometrics IV -_**Rationale**_ : Minerals are the building blocks of rocks. A rock can be made of one or more minerals. Minerals and rocks have the following properties: Property | Mineral | Rock It is a solid. | Yes | Yes It is formed in nature. | Yes | Yes It is not made by organisms. | Yes | Yes It is a pure substance. | Yes... Different types of minerals have different crystal structures, but all minerals have a fixed crystal structure. This means that the atoms and molecules in different pieces of the same type of mineral are always arranged the same way.\nHowever, rocks do not have a fixed crystal structure. So, the arrangement of atoms or molecules in different pieces of the same type of rock may be different! The properties of marble match the properties of a rock. So, marble is a rock. _**Answer**_ : The answer is (A) Blue. +Workshop -Multimodal large language model 𝑿𝟏 𝑿𝟐 𝑿𝟑 _**Question**_ : Is celestine a mineral or a rock? _**Question**_ : What type of rock is marble? _**Question**_ : Is soapstone a mineral or _**Options**_ : _**Options**_ : a rock? _**Options**_ : (A) Mineral. (A) Sedimentary. (B) Rock. (B) Metamorphic. (A) Rock. (B) Mineral. _**Rationale**_ : Minerals are the building blocks of (C) Igneous. rocks. A rock can be made of one or more _**Rationale**_ : Igneous rock is formed when _**Rationale**_ : Minerals are the building minerals. Minerals and rocks have the following melted rock cools and hardens into solid blocks of rocks. A rock can be made of properties: Property | Mineral | Rock. It is a solid. rock. This type of change can occur at one or more minerals. Minerals and rocks | Yes | Yes. It is formed in nature. | Yes | Yes. It Earth‘s surface or below it. Sedimentary have the following properties: Property | is not made by organisms. | Yes | Yes. It is a pure rock is formed when layers of sediment Mineral | Rock It is a solid. | Yes | Yes It substance. | Yes | No It has a fixed crystal are pressed together, or compacted, to is formed in nature. | Yes | Yes It is not structure. | Yes | No. You can use these properties make rock... Like other metamorphic made by organisms. | Yes | Yes It is a to tell whether a substance is a mineral, a rock, or rocks, it forms when a rock is changed by pure substance. | Yes | No It has a fixed neither. Look closely at the last three properties: high temperature and pressure. Heat and crystal structure. | Yes | No …So, the Minerals and rocks are not made by organisms. pressure can change the type and arrangement of atoms or molecules in Organisms make their own body parts. For arrangement of minerals in a rock. This different pieces of the same type of rock example, snails and clams make their shells. change forms a new rock with different may be different! The properties of Because they are made by organisms, body parts properties. Marble can form when soapstone match the properties of a rock. cannot be minerals or rocks.. Celestine has all the sedimentary rocks such as limestone are So, soapstone is a rock. properties of a mineral. So, celestine is a mineral. changed by heat and pressure. _**Answer**_ : The answer is (B) Soft. _**Answer**_ : The answer is (A) Mineral. _**Answer**_ : The answer is (A) Mineral. +Labor II (Secondary) -Figure 8: The retrieved data of an image-text pair example. +Antitrust (Primary) -**==> picture [365 x 316] intentionally omitted <==** +Public -**----- Start of picture text -----**
-Query 𝓧𝒒𝒖𝒆𝒓𝒚 Rationale first base. A compound: A simple sentencesentenceis ais two simplesentence withsentences joined byonly one subject a commaand predicate.and a conjunction such as and, but,The pitcher threw the ball to
or, or so. The pitcher threw the ball, and the batter hit it. Some simple sentences have a compound subject or a
Question : Which is a compound sentence? compound predicate, but they are not compound sentences… The introductory phrase is part of the predicate. In
Options : the winter, Farmer Ben wears his heavy coat. This is a simple sentence. There is one subject, Farmer Ben, and
one predicate, wears his heavy coat in the winter. The first sentence is the compound sentence. It is made up of
(A) Dillon liked the sea otters, but the jellyfish two simple sentences joined by a comma and the conjunction but. Dillon liked the sea otters, but the jellyfish
were his favorite. were his favorite.
Answer : The answer is (A) Dillon liked the sea otters, but the jellyfish were his favorite.
(B) The artist prepared a canvas for a new oil
painting. Multimodal large language model
𝑿𝟏 𝑿𝟐 𝑿𝟑
Question : Which is a compound sentence? Question : Which is a compound sentence? Question : Which is a compound sentence?
Options : Options : Options :
(A) The artist prepared a canvas for a new oil (A) The camera isn't working, so the battery must (A) The tailor measures the length of the pant leg.
painting. be dead. (B) Desmond liked the sea otters, but the
(B) Devon shot the arrow, but she missed the (B) The artist prepared a canvas for a new oil jellyfish were his favorite.
target. painting.
Rationale : A simple sentence is a sentence Rationale : A simple sentence is a sentence with Rationale : A simple sentence is a sentence with
with only one subject and predicate. The only one subject and predicate. The pitcher threw only one subject and predicate. The pitcher threw
pitcher threw the ball to first base. A the ball to first base. A compound sentence is the ball to first base. A compound sentence is two
compound sentence is two simple sentences two simple sentences joined by a comma and a simple sentences joined by a comma and a
joined by a comma and a conjunction such as conjunction such as and, but, or, or so. The conjunction such as and, but, or, or so. The pitcher
and, but, or, or so. The pitcher threw the ball, pitcher threw the ball, and the batter hit it. Some threw the ball, and the batter hit it. Some simple
and the batter hit it. Some simple sentences simple sentences have a compound subject or a sentences have a compound subject or a
have a compound subject or a compound compound predicate, but they are not compound compound predicate, but they are not compound
predicate, but they are not compound sentences…Some simple sentences have sentences…Some simple sentences have
sentences…Some simple sentences have introductory phrases, but they are not compound introductory phrases, but they are not compound
introductory phrases, but they are not sentences. The introductory phrase is part of the sentences. The introductory phrase is part of the
compound sentences. The introductory phrase predicate. In the winter, Farmer Ben wears his predicate. In the winter, Farmer Ben wears his
is part of the predicate. In the winter, Farmer heavy coat. This is a simple sentence. There is heavy coat. This is a simple sentence. There is one
Ben wears his heavy coat. This is a simple one subject, Farmer Ben, and one predicate, subject, Farmer Ben, and one predicate, wears his
sentence. There is one subject, Farmer Ben, wears his heavy coat in the winter. The first heavy coat in the winter. The second sentence is
and one predicate, wears his heavy coat in the sentence is the compound sentence. It is made up the compound sentence. It is made up of two
winter. The second sentence is the compound of two simple sentences joined by a comma and simple sentences joined by a comma and the
sentence. It is made up of two simple sentences the conjunction so. The camera isn't working, so conjunction but. Desmond liked the sea otters, but
joined by a comma and the conjunction but. the battery must be dead. the jellyfish were his favorite.
Devon shot the arrow, but she missed the Answer : The answer is (A) The camera isn't Answer : The answer is (B) Desmond liked the sea
target.
Answer : The answer is (B) Devon shot the working, so the battery must be dead. otters, but the jellyfish were his favorite.
arrow, but she missed the target.
**----- End of picture text -----**
+Workshop +Regulation -Figure 9: The retrieved data of a text-only example. +Dissertation -16 +Workshop -We present qualitative examples to illustrate the effectiveness of the RMR framework. Figure 6 showcases an example where the model is asked, "Which property do these three objects have in common?" The retrieved data plays a crucial role in guiding the model’s reasoning process. _X_ 1 provides a similar question involving different objects. The similarity in the question structure helps the model understand the type of reasoning required, offering a blueprint for approaching the problem. By seeing how the retrieved example addresses the comparison of different objects, the model can apply a similar strategy to the current query. _X_ 2 includes a related object, enhancing the model’s comprehension of various object properties. The additional context provided by _X_ 2 helps the model draw connections between the current objects and previously encountered ones, broadening its understanding and improving its ability to identify common properties. Similar to _X_ 1, _X_ 3 also presents a question about the common properties of different objects, albeit with a wider variety. The diversity in examples reinforces the model’s reasoning skills, helping it generalize its reasoning ability across different contexts and object sets. +Dissertation -Figure 7 depicts a text-only example where the model is asked, "Which word would you find on a dictionary page with the following guide words?". Three similar questions are retrieved with different words, providing the model with a diverse set of examples to learn from. The retrieved examples help the model understand the structure of the question and the type of reasoning required to answer it. By observing the different words in the retrieved examples, the model can learn to identify the commonalities between the guide words and the target words, enhancing its reasoning capabilities. +Workshop -Figure 8 presents an image and text example where the model is asked, "Is marble a mineral or a rock?". The retrieved data provides relevant context that helps the model formulate a more accurate and informed response. The retrieved data _X_ 1 contains the question, "Is celestine a mineral or a rock?" This question is analogous to the query about marble, offering the model a direct parallel that aids in understanding how to distinguish between minerals and rocks. The similar structure helps the model apply the reasoning used for celestine to marble, reinforcing the process of categorizing geological substances. _X_ 2 poses the more specific question, "What type of rock is marble?" It provides the model with additional information about the classification of marble within the broader category of rocks. This specific context helps the model not only affirm that marble is a rock but also understand its specific type, thereby enriching the model’s geological knowledge base. _X_ 3 includes the question, "Is soapstone a mineral or a rock?" Similar to _X_ 1, it presents another instance of the mineral vs. rock distinction. The inclusion of different substances like soapstone reinforces the model’s ability to generalize the reasoning process across various materials, ensuring a robust understanding of the mineral and rock classification. +Dissertation -Figure 9 presents a text-only example where the model is asked, "Which is a compound sentence?" The retrieved examples all ask the same question but with different options, which provide the model with varied contexts and sentence structures to learn from. The consistent retrieval of questions on compound sentences helps the model understand the syntactic characteristics that define compound sentences. By comparing and analyzing the different options presented in the retrieved examples, the model can improve its ability to identify compound sentences accurately. +Workshop -## **B Broader Impact** +3 -Here, we outline several broader impacts of our work: +3 -**Hallucination** Similar to other large language models, RMR-integrated models like LLaVA might generate outputs that are not grounded in factual information or the input data provided. This phenomenon, known as hallucination, may raise concerns. Ensuring the reliability and accuracy of outputs in such sensitive domains is crucial, and further research is needed to mitigate these risks. +3 -**Biases** The RMR framework inherits biases from its base models, including the vision encoder (CLIP) and the vision LLMs (LLaVA, Qwen-VL, InternLM, and Gemini). These biases may be reflected in the retrieved examples and the model’s reasoning process. It is essential to address these biases to ensure fair and unbiased reasoning capabilities. +1 -**Energy Consumption** Although our approach is training-free, running inference still incurs energy consumption. The computational resources required for processing large volumes of data. Efforts should be made to optimize inference efficiency and explore sustainable computing practices to minimize the energy footprint of using RMR-enhanced models. +3 -17 +3 + +3 + +1 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +3 + +6 + +3 + +6 + +3 + +6 + +3 + +Fall 2022 + +Fall 2022 + +Fall 2022 + +Fall 2022 + +Spring 2023 + +Spring 2023 + +Spring 2023 + +Spring 2023 + +Fall 2023 + +Fall 2023 + +Fall 2023 + +Fall 2023 + +Spring 2024 + +Spring 2024 + +Spring 2024 + +Spring 2024 + +Fall 2024 + +Fall 2024 + +Fall 2024 + +Fall 2024 + +Spring 2025 + +Spring 2025 + +Spring 2025 + +Fall 2025 + +Fall 2025 + +Spring 2026 + +Spring 2026 + +#ECON 8080 may be replaced with ECON 9090, typically offered in the Spring. Primary and Secondary Fields will +vary depending on the student. Check with the instructors of those Fields to determine requirements above and +beyond coursework, for example, whether a paper must be submitted or whether the student must pass a qualifying +Field examination. Most students choose a 9820 Workshop corresponding to their Primary Field, but this is not a +hard-and-fast requirement. + +12 + + Form GS4: Graduation + +The Graduate Diploma Application, GS4, announces to the Graduate School that a student anticipates +completing the dissertation and intends to graduate. This form must be filed in the first two or three weeks +of the semester in which the student wishes to graduate. The chair of the dissertation committee must +confirm from the Student Progress Report that the student has passed the Qualifying Examination in +Economic Theory, has satisfied the Minimum Competency Requirement in Econometrics, and has +satisfied the Primary and Secondary Field Requirements prior to signing the GS4 form. The practical +impact of this rule is that all field requirements must be met in the semester prior to graduating. + +Form GS5: Preparation for Graduation + +The GS5 form signifies that the student has passed the Qualifying Examination in Economic Theory, +has satisfied the Minimum Sufficiency in Econometrics requirement, and has satisfied the Primary and +Secondary Field requirements. Student progress is monitored by the Graduate Program Coordinator, +as well as the student’s PhD Advisory Committee. + +Form GS7: Thesis Defense + +The defense must be scheduled early enough to allow time – typically, at least two weeks -- for +committee-required revisions to the dissertation paper before the GS7 deadline. The final PhD +examination must be passed at least three weeks prior to the commencement in which the student +graduates. The defense will be administered by the PhD dissertation committee, and members of the +faculty and the Dean of the Graduate School are invited to attend. The results of the exam are +communicated to the Graduate School on form GS7, within five days of the exam. + +The final draft of the thesis should be given to committee members well in advance of, and no less +than three weeks prior to the defense. This will give committee members time to propose +modifications. Revisions can be made after the defense. The Graduate Announcements make clear +that: “It should be understood that a vote to pass a student on his/her performance at the +thesis/dissertation defense (form GS7) does not imply final approval of the thesis or dissertation. +Approval of the thesis or dissertation is given by faculty signing the approval page. The approval page +should not be signed by the committee until the student has made all revisions as instructed by the +committee.” The practical impact of this rule is that necessary revisions can be made to improve the +dissertation subsequent to the dissertation defense, and further, that the degree will ultimately be +awarded only when such revisions are made to the satisfaction of thecommittee. + +Written notification of the defense is due in Enrolled Student Services at least 10 business days prior +to the defense, which typically amounts to 2 calendar weeks. The information must include the +student’s name, program of study, title of thesis paper, major advisor, date, time, and location. The +student arranges the date, time, and place for the defense in consultation with the major advisor and +other members of the advisory committee. Copies of the thesis paper must be delivered to the +student's advisory committee at least two weeks before the defense. The student’s major professor and +advisory committee conduct the oral examination, but all faculty members are invited to attend. After +the defense, revisions in the thesis paper by the student must be approved by the major advisor and +committee members. Students may have two attempts to pass the exam. A GS7 must be signed and +submitted no later than the end of the penultimate week of the fall or spring semester to indicate +passage of the final examination. + +13 + + Academic Probation + +Unfortunately, students occasionally perform below the standards set by the Graduate School. There +are two basic requirements: a GPA of 3.0 or better and carrying out research to a satisfactory standard. +This section is no more than an introduction; for details students should consult the Policies and +Procedures of the Graduate School. + +There are two levels of academic probation: R1 for those on probation for the first time, and R2 for +those beyond the first time. Students who find themselves on probation because their GPR has fallen +below 3.0 are expected to raise their GPR to 3.0 or better within NINE (9) credit hours. Although +students who fail to do so may be dismissed from their program at that point, those who make +sufficient progress may be given another chance to do so within the next nine credit hours. Students +who are placed on academic probation must file a GSR1 or GSR2 Plan for Success, which must be +signed by the Graduate Program Coordinator and approved by the Graduate School. It is the +student’s responsibility to contact the Coordinator. +Graduate work in economics, particularly at the PhD level, often challenges students at a level that +they have never experienced before. Students accustomed to excelling as undergraduates, or even at +the MA level, can be overwhelmed by the amount and difficulty of material, especially during the first +year. It cannot be overstated how important it is for each student to examine themselves and their +performance on a continuous basis. + +In many cases, a rough start can, with hard work, be overcome. A poor midterm exam grade might +be offset by a good grade on the final exam, and a subpar first-semester performance might be +redeemed by good performance during the second. A student who consistently fails to perform well +should consider seriously their ultimate goals and how best to achieve as many of them as possible. In +some cases, this may involve dropping a course by the drop date – please see the academic calendar at +http://www.registrar.clemson.edu/html/acad_cal.htm – to prevent an F from becoming a part of +their academic record (October 28 for the Fall 2022 academic year). In other cases, this may +involve switching from the PhD program to the MA program, provided that permission is granted by +the Graduate School. + +In addition to maintaining a 3.0 GPA or better, graduate students must demonstrate academic progress +by passing the Qualifying Examination in Economic Theory, demonstrate Minimum Competency in +Econometrics, satisfy Primary and Secondary Field requirements, and submit second- and third-year +papers, and, ultimately, write and defend a PhD dissertation. Students who fail to satisfy these +requirements risk dismissal from the Program. + +14 diff --git a/q181/random_k4/random_4.md b/q181/random_k4/random_4.md index 6927512bd90f46ab9e09b90b6a7a3c1b4e782552..7d30cced73b1f30da245c4cf6780b4baf04ace71 100644 --- a/q181/random_k4/random_4.md +++ b/q181/random_k4/random_4.md @@ -1,598 +1,553 @@ -## **SAGE: Selective Attention-Guided Extraction for Token-Efficient Document Indexing** + GUI DE -Xinzhi Wang Peter Baile Chen Purdue University MIT wang6171@purdue.edu peterbc@mit.edu Jun Chen Michael Cafarella Independent Researcher MIT chjuncn@gmail.com michjc@csail.mit.edu +Social Media +Marketing Strategy +Eight easy steps to develop your +social media presence -## **ABSTRACT** + Contents -Large language models with long context windows can answer complex questions directly from full-length academic, technical, and policy documents, but passing entire documents is often costly, slow, and can degrade answer quality while increasing the risk of unnecessary data leakage. This paper targets the common setting of answering many heterogeneous questions over long document(s), where fixed position heuristics and standard retrieval-augmented generation (RAG) can fail due to document structure variability and weak query-chunk semantic similarity, which often requires taskand domain-specific tuning of embedding retrievers. We propose Selective Attention-Guided Extraction (SAGE), a training-free, plugand-play context reduction framework that uses a lightweight local LLM to perform a single prefilling pass and convert language model attention signals into a query-specific relevance heatmap at configurable granularities. SAGE further introduces _differential attention_ strategies to better isolate question-relevant evidence, then selects the top-scoring units under a user-defined token budget and forwards only this reduced context to a downstream LLM for answer generation. SAGE surpasses traditional reduction techniques across multiple long-document QA benchmarks, notably securing a top-4 rank on QuALITY-hard while constrained to a 10% context budget. This enables a 90% reduction in tokens with competitive accuracy, without the need for model fine-tuning or complex calibration. +Step 1: Set social media marketing goals that align to business objectives -## **PVLDB Reference Format:** +Step 2: Learn everything you can about your audience -Xinzhi Wang, Peter Baile Chen, Gerardo Vitagliano, Matthew Russo, Jun Chen, Michael Cafarella, Samuel Madden, and Chunwei Liu. SAGE: Selective Attention-Guided Extraction for Token-Efficient Document Indexing . PVLDB, 14(1): XXX-XXX, 2020. doi:XX.XX/XXX.XX +Step 3: Research the competition -## **PVLDB Artifact Availability:** +Step 4: Conduct a social media audit -The source code, data, and/or other artifacts have been made available at https://github.com/Tranway1/AttentiveTrim.git. +Step 5: Set up accounts and improve existing profiles -This work is licensed under the Creative Commons BY-NC-ND 4.0 International License. Visit https://creativecommons.org/licenses/by-nc-nd/4.0/ to view a copy of this license. For any use beyond those covered by this license, obtain permission by emailing info@vldb.org. Copyright is held by the owner/author(s). Publication rights licensed to the VLDB Endowment. Proceedings of the VLDB Endowment, Vol. 14, No. 1 ISSN 2150-8097. doi:XX.XX/XXX.XX +Step 6: Find inspiration -Gerardo Vitagliano Matthew Russo MIT MIT gerarvit@csail.mit.edu mdrusso@csail.mit.edu Samuel Madden Chunwei Liu MIT Purdue University madden@csail.mit.edu chunwei@purdue.edu +Step 7: Create a social media content calendar -## **1 INTRODUCTION** +Step 8: Test, evaluate, and adjust your strategy -As LLM-based applications take on increasingly complex workloads, the demand for long-context question answering has grown rapidly [23, 25, 27, 29, 37, 42, 49]. Although recent models can ingest entire documents end-to-end [27], _more context is not always better_ . Long inputs often mix a small amount of salient evidence with substantial irrelevant text, which can distract the model and degrade answer accuracy [13, 28]. The practical overhead is also substantial: under pay-per-token commercial APIs, processing full documents directly increases monetary cost, and even in self-hosted settings, it wastes compute on non-essential text. +3 -These issues are exacerbated when using smaller local models. Once an input exceeds the context window of a model, the document can no longer be processed in a single forward pass, requiring chunking or multi-stage pipelines that increase latency and memory overhead while complicating deployment. +4 -Beyond efficiency and quality, minimizing unnecessary data exposure is a practical constraint in many real-world settings. Enterprise reports, proprietary documents, and internal analyses often contain sensitive information that should not be fully revealed to downstream or business-facing LLMs. Together, these considerations motivate _context reduction_ : extracting a compact subset of the input that preserves answer quality while bounding LLM cost and limiting unnecessary information exposure. +5 -A natural first idea for context reduction is to take advantage of where answers usually appear in a document. Prior work [23] and our early observations suggest that some question types do show rough structural patterns. For example, in research papers, answers are often found in sections such as the abstract, introduction, evaluation, or conclusion. This motivates validation-based positional heuristics: using a small set of documents with known answers, we can locate the answer spans, map them to token positions, and build a relevance heatmap over the document. This heatmap can then be used to guide context extraction for new documents. +6 -However, these patterns are only rough, and validation-based heatmaps have clear limitations. First, they rely on a strong LLM to accurately locate answer spans in long documents, which can be unreliable when documents are lengthy or vary in structure. Second, answer locations can differ a lot across question types. As shown in Figure 1, authorship-related evidence is usually concentrated near the beginning of a paper, while evidence for contributions is spread much more broadly. Third, for reasoning-heavy questions, the relevant evidence is often distributed across several sections rather than confined to a single region. For instance, a paper’s main +7 -**==> picture [193 x 140] intentionally omitted <==** +8 -**----- Start of picture text -----**
-20
10
0
0.0 0.2 0.4 0.6 0.8 1.0
Character position
(a) Who are the authors of the paper
4
2
0
0.0 0.2 0.4 0.6 0.8 1.0
Character position
(b) What is the main contribution of the paper
Frequency
Frequency
**----- End of picture text -----**
+8 +10 -**Figure 1: Heatmaps of ground-truth answer distributions in the Paper dataset. Author queries (a) cluster at the start, while contribution queries (b) disperse throughout.** +GUIDE / Social Media Marketing Strategy -contribution may be introduced in the abstract, motivated in the introduction, explained in detail later, and summarized again in the conclusion. In these cases, a single average position or fixed region is unlikely to capture all the evidence needed to answer the question correctly. +#239584107 -Retrieval-Augmented Generation offers an alternative by retrieving relevant evidence before generation, reducing the need to pass entire documents to the model. RAG systems span dense bi-encoders as well as token-level late interaction retrievers such as ColBERT [17] and ColBERTv2 [38], with recent refinements that further improve token-level scoring [15]. With the emergence of long-context LLMs, RAG has also moved toward retrieving larger and more coherent units [12] and organizing evidence hierarchically [39]. Yet retrieval still relies on embedding similarity, which does not guarantee that selected chunks contain the answer. This gap is most visible when queries lack discriminative keywords or when the supportive context is only weakly aligned with the query in the embedding space. As a result, retrieval may miss critical evidence or include misleading context. For example, for a query such as “Who are the authors of this paper?”, the chunks most similar in embedding space are often in the reference section (where author names appear in citations) rather than near the title or header, where the correct answer is typically found. This failure mode becomes especially painful when many questions must be answered over the same document. +2 -These limitations point to a more adaptive, model-driven approach. Instead of inferring relevance indirectly via validation heatmaps or embedding similarity, we ask " _can relevance be obtained directly from the model itself?_ " Transformer-based LLMs compute attention distributions during the prefilling phase that reflect how strongly different parts of the context contribute to answering a given question. Importantly, these signals are available without generating any output. We only run the prefill pass and read the attention weights, avoiding the dominant cost of the decoding phase [1]. We find that these attention signals provide a fine-grained, query-specific relevance signal that can be exploited without external supervision. + Related resource -Building on this insight, we propose **S** elective **A** ttention- **G** uided **E** xtraction (SAGE), an attention-driven context reduction framework that uses a lightweight local LLM as a preprocessing filter. +The Social Media Metrics +that Really Matter—and +How to Track Them -Given a document and a query, the local model computes tokenlevel attention over the document, producing a relevance heatmap that can be aggregated at configurable granularities such as tokens, sentences, or sections. We then select the highest-scoring units to form a compact context that can be passed to any downstream LLM for answer generation under a user-provided token budget. +Step 1 -A practical challenge is that raw attention can be noisy: generic document elements such as headers, boilerplate text, and formatting cues may consistently attract attention even when they are not query-relevant. Recent work suggests that contrastive signals can help separate shared background patterns from task-specific evidence [7, 22, 47]. Inspired by this line of work, SAGE introduces _differential attention_ . We subtract attention patterns induced by a contrasting query to cancel persistent noise and better isolate query-specific evidence. +Set social media marketing goals that align to +business objectives -This design offers several system-level advantages. First, SAGE reduces the original document to a compact context that retains only the content most relevant to the query. Second, it provides explicit budget control via a user-specified token budget, bounding inference cost regardless of document length. Third, unlike validation-based heatmap construction, SAGE does not require ground truth answers, validation datasets, or query-specific calibration. More broadly, the approach is training-free and can be applied directly with off-the-shelf LLMs, without model modification or fine-tuning, making it readily applicable to new questions and domains. In addition, SAGE leverages KV-cache reuse to improve efficiency by avoiding repeated encoding of the same document chunks across queries. +The more specific your strategy is, the more effective the execution will be. +Set SMART goals and track the right metrics to set yourself up for success. -We evaluate SAGE on four benchmarks spanning different modalities. On QuALITY-hard [35], SAGE ranks **4th** on the public leaderboard [34] while using only a **10%** context budget. On _Paper_ and _Notice_ [23], SAGE consistently outperforms strong embedding-based RAG baselines across token budgets, and we analyze the effects of attention model size and differential-attention variants. Finally, we show that SAGE extends to semi-structured tables without datasetspecific tuning, achieving accuracy comparable to full-table input while substantially reducing the table content passed to the generator. Looking ahead, we expect the same attention-guided principle to extend to additional modalities, including more complex semistructured tables and visual documents. +Set SMART goals -To summarize, our contribution includes: +The first step to creating a social media marketing strategy is to establish +your objectives and goals. Without goals, you have no way to measure your +success or your social media return on investment (ROI). -- We study _budgeted context reduction_ in the setting of _multiple questions per document_ , and highlight limitations of validationbased positional heuristics and embedding only retrieval for dispersed evidence and weak lexical cues (Section 1). +Each of your goals should be: -- We propose **S** elective **A** ttention- **G** uided **E** xtraction (SAGE), a training-free pipeline that uses a lightweight local LLM to compute token-level relevance and select compact context under an explicit token budget (Section 3). +Specific -- We introduce _differential attention_ to filter out query-independent noise and improve relevance estimation, and study alternative reference queries, including a fixed baseline prompt and a farthest question contrast (Section 3.2). +Measurable -- We evaluate SAGE on four benchmarks spanning different regimes and modalities, showing consistent improvements over strong baselines in token accuracy trade-offs (Section 4). +Attainable -2 +Relevant -- We demonstrate an extension of SAGE to table selection, achieving accuracy comparable to full-table input while reducing the _average row usage_ exposed to the generator (Section 4.8). +Time-bound -solely on attention signals to extract contiguous windows without training. +An example of a SMART goal for your business might be "Grow our Instagram +audience by 50 new followers per week." -## **2.3 KV Cache Compression** +With SMART goals, you’ll make sure your goals actually lead to real business +results, rather than just lofty ideals. -## **2 RELATED WORK** +GUIDE / Social Media Marketing Strategy -To mitigate the high overhead of long-context LLM inference, prior work spans several distinct but complementary directions. +3 -## **2.1 Long-Context RAG** + Track meaningful metrics -Retrieval-Augmented Generation (RAG) avoids full document ingestion by conditioning generation only on retrieved evidence. While early systems relied on dense single-vector encoders, late interaction retrievers like ColBERT [17], ColBERTv2 [38], and TRIAL [15] improve localization by retaining token-level structure. As longcontext LLMs emerged, methods like LongRAG [12] and RAPTOR [39] adapted by retrieving larger, coherent multi-thousand-token units or hierarchical summaries. +While vanity metrics like retweets and likes can be fun to share and easy to +track, it’s hard to prove their real value for your business. Instead, focus on +targets such as leads generated, web referrals, and conversion rate. -However, simply retrieving more chunks is not a panacea. Jin et al. [13] shows that increasing the number of retrieved passages can actively distract the reader model with “hard negatives”, chunks of text that are highly similar to the query in embedding space but actually contain irrelevant or misleading information. This risk of degraded reasoning motivates the need for refinement pipelines like LongRefiner [14] and OP-RAG [48]. +You may want to track different goals for different channels, or even different uses +of each channel. For example, you can use paid campaigns to increase brand + awareness, but measure acquisition and engagement for organic social posts. -SAGE is highly complementary to these directions. While RAG pipelines select candidate documents or chunks, SAGE acts as a budget-constrained filter to preserve only the highest-utility evidence _within_ those texts with a finer granularity. +Make sure to align your social media goals with your overall marketing +strategy. This will make it easier for you to show the value of your work and +get executive buy-in and investment. -## **2.2 Prompt Compression** +Start developing your social media marketing plan by writing down at least +three social media goals. -When retrieval is insufficient, _prompt compression_ reduces the effective prompt length while preserving necessary information [21]. One approach compresses text into continuous _soft prompts_ [21] using learned vectors or reusable tokens, as seen in CC [43] and GIST [31]. ICAE [6] distills text using a frozen LLM, xRAG [4] injects document embeddings via a fusion bridge, and MIPRO [33] optimizes prompt programs via meta-optimization. However, soft prompts are opaque, require domain-specific training [21], and often cause semantic distortion [24]. +Goals -To avoid meaning distortion, _hard prompt_ methods select or prune discrete tokens. Early methods like PoWER-BERT [8] and TR-BERT [46] prune tokens based on learned signals but often adopt fixed lengths on training data. Adaptive methods tailor pruning dynamically. SpAtten [41] uses proportional lengths, LTP [18] applies learned thresholds, and LLMLingua[10] scores and removes low-utility tokens. LongLLMLingua [11] further mitigates the risk of dropping crucial evidence by adding question-aware signals. Similarly, task-specific importance metrics are now actively used to compress dense token streams in multimodal settings [40]. +1. -To prevent the fragmentation of local syntax caused by token dropping, recent _context reduction_ systems extract contiguous sentences. CPC [24] retains top-ranked sentences using a contextaware encoder, while ZenDB [23] leverages hierarchical structure in templated documents. Unlike aggressive compressors that suffer from knowledge overwriting [9], SAGE strictly preserves the original source text, avoiding structural assumptions by relying +2. -A complementary line of work reduces memory and latency during decoding by optimizing the KV cache without altering the input prompt. FastGen [5] and SnapKV [20] profile attention patterns to compress caching for heavy-hitting tokens. For dynamic cache management, _𝐻_ 2 _𝑂_ [51] formulates KV eviction as a submodular optimization problem, while NACL [3] improves eviction robustness via proxy-token estimation. Beyond eviction, RetrievalAttention [26] manages cache access via approximate nearest-neighbor search, and SepLLM [2] leverages separator tokens as compact anchors. Other optimizations include dynamic cache budget allocation across layers [52] and concatenating precomputed caches across repeated contexts, as seen in KVLink [45]. While effective at reducing decoding costs, most KV-cache methods still require ingesting the full context during the computationally expensive prefilling stage. In contrast, SAGE reduces the input context _before_ generation, lowering both prefilling and decoding costs while providing explicit budget control and supporting privacy-preserving deployments. +3. -## **2.4 Differential and Contrastive Attention** +Step 2 -Several studies highlight the critical role of attention structure for extractive tasks and long-context reasoning [44, 50]. Historically, differential attention was originally introduced in image processing via the Differential Attention Network (DAN), which contrasts supporting and opposing exemplars to better align model focus with human attention [36]. In modern LLMs, methods like Differential Transformer [47] and Grouped Differential Attention [22] introduce fundamental architectural changes, computing attention as the difference between softmax maps to cancel common-mode noise. Contrastive signals are also leveraged to isolate task-relevant visual regions in multimodal models [7] and to suppress generic text generation during the decoding phase [19]. +Related resource -In contrast to approaches that require heavy model modifications, specialized training regimens (e.g., DAN [36]), or decoding-time interventions, SAGE uses differential reasoning as a training-free filter for input context reduction. By applying this contrastive signal purely during the prefilling stage without altering the underlying Transformer, SAGE provides a lightweight, plug-and-play solution to neutralize query-agnostic noise and isolate discriminative evidence prior to generation. +How to build audience +personas -## **3 METHODOLOGY** +Learn everything you can about your audience -Building upon the need for efficient, noise-resistant context management, this section details the architecture of SAGE. Our primary design objective is to extract the most informative text spans from a long document while strictly adhering to a user-specified token budget, ensuring that the generation LLM receives a high-quality, continuous context. +if you’re not engaged in social media listening, you’re creating your business +strategy with blinders on—and you’re missing out on mountains of actionable +insights from real people who are actively talking about you or your industry online. -As illustrated in Figure 2, the SAGE pipeline processes a userprovided **document** , **query** , and **token budget** across four primary stages. Stage _○_ 1 computes a query-aware document attention score through four substeps. First, in step (a), we partition the +Here’s how to start listening and building your understanding of your +audience and their needs. -3 +Create audience personas -**==> picture [506 x 284] intentionally omitted <==** +tKnowing who your audience is and what they want to see on social is key to +creating content that they will like, comment on, and share. This knowledge +also critical for planning how to develop your social media fans into +customers for your business. -**----- Start of picture text -----**
-③ Unit-Window
② Differential Attention Selection ( § 3.3 ) Based on the following
Budget ( § 3.2 ) Select text windows with context, answer the
User (a) Find contrast question: highest average question concisely.
𝑞 [−] attention under a budget
Document (b) Query normalization: Context: …
ത 1 Assemble selected context
(1000 Tokens) 𝐒(q) = 𝑇𝑞 ∗𝐒 (q) into a prompt Question: …Answer:
(c) Compute differential
Query 𝒒 attention
① Compute Token Level 𝑑 𝑞= 𝐒 [ത] 𝑞 − 𝐒 [ത] 𝑞 [−] ④ Reduced Context
Prompt ( § 3.3 )
Attention Score ( § 3.1 ) SAGE
(a) Chunked Doc. Processing ( § 3.1.1 ) Query → context attention (c) Chunk-Length Normalization ( § 3.1.2 )
Context 𝐶1 Context: 𝐶1 Aggregate over heads & layers Length-based normalization: ǁ𝑠𝑖 = 𝑠𝑖 ∗ 𝑇default𝑇𝑖
(300 Tokens) Question: Answer: 𝑞 𝑇default = 300 Generate an
Context 𝐶2 Context: 𝐶2 (b) Attention Scale Bias ( § 3.1.2 ) Long chunk: ǁ𝑠1 = 𝑠1 ∗ 𝑇default𝑇1 = 𝑠1 ∗ 300300 [= 𝑠][1] answer using a generation
(300 Tokens) Question: Answer: 𝑞 • Softmax distributes fixed scale• Fewer tokens share the mass Short chunk: ǁ𝑠4 = 𝑠4 ∗ 𝑇4 LLM
𝑇default
Context 𝐶3 Context: 𝐶3 ⇒ higher attention scale
(300 Tokens) Question: Answer: 𝑞 𝑠1 𝑠4 (d) Reconstruct full-doc. Attention ( § 3.1.3 ) Generated
(100 Tokens)Context 𝐶4 Question: Context: 𝑞𝐶4 𝑠𝑠32 Misaligned attention scale Concatenating chunk attention scores: 𝐒= 𝑐𝑜𝑛𝑐𝑎𝑡 ǁ𝑠1, ǁ𝑠2, ǁ𝑠3, ǁ𝑠4 Answer
Answer: Lower attention scale Higher attention scale
**----- End of picture text -----**
+Try creating audience personas. For example, a retail brand might create +different personas based on demographics, buying motivations, common +buying objections, and the emotional needs of each type of customer. +GUIDE / Social Media Marketing Strategy -**Figure 2: Overview of the SAGE pipeline.** _○_ **1** _**Attention computation**_ **utilizes chunked processing and length normalization to form a document-level score.** _○_ **2** _**Differential attention**_ **refines relevance by filtering structural noise.** _○_ **3** _**Unit-window selection**_ **extracts top text spans under the budget.** _○_ **4 Spans are assembled into a** _**reduced-context prompt**_ **for final answer generation.** +4 -document into chunks that fit within the context window of a lightweight local LLM, computing token-level attention scores for each chunk paired with the query. Because these raw scores can be skewed by chunk length, we identify attention scale misalignment across chunks in step (b) and apply _chunk-length normalization_ in step (c) to remove the scale difference caused by varying chunk sizes. Finally, in step (d), we map these normalized scores back to their original positions to reconstruct a cohesive, full document attention score. + Personas sharpen your marketing tactics. Luxury buyers, for example, +might not respond to Facebook ads with sales. But they might respond to +Facebook ads with exclusive in-store events to be the first to see a new line of +clothing. With personas, you’ll have the customer insights you need to create +campaigns that speak to the real desires and motivations of your buyers. -In Stage _○_ 2 , we first apply _query normalization_ so scores are on a consistent scale across different questions. Then, _differential attention_ is used to filter out noise and highlight query-relevant signals. Unit-window selection (Stage _○_ 3 ) leverages these refined scores to extract the most useful, coherent text spans within the allocated budget. Finally, in Stage _○_ 4 , these selected spans are assembled into a reduced-context prompt and passed to the generation model to produce the final answer. In the following subsections, we discuss each stage in detail. +Gather real-world data -## **3.1 Chunked Attention Computation and Normalization** +Don’t make assumptions. Social media analytics can also provide a ton of +valuable information about who your followers are, where they live, which +languages they speak, and how they interact with your brand on social. These +insights allow you to refine your strategy and better target your social ads. -_3.1.1 Chunked Attention Computation._ As we try to extract the parts of a long document that are truly relevant for answering a query, we first need a query-aware signal of token importance. Selfattention offers a natural choice. When a Transformer processes the question, each query token allocates attention to the earlier context tokens it relies on. We therefore use a lightweight local +Jugnoo, an Uber-like service for auto-rickshaws in India, used Facebook +Analytics to learn that 90 percent of their users who referred other +customers were between 18 and 34 years old, and that 65 percent of that +group was using Android. They used this information to target their ads, +resulting in a 40 percent lower cost per referral. -LLM to measure how much attention the query assigns to each document token, yielding an efficient relevance score for guiding later selection. Given a document _𝐷_ and a query _𝑞_ , let _𝑇𝐷_ and _𝑇𝑞_ denote the number of tokens in the tokenized document and query, respectively. +Step 3 -A direct forward pass over the concatenation of the full document and query is often infeasible, because _𝑇𝐷_ + _𝑇𝑞_ may exceed the context window of the model. We therefore partition the document into _𝑛_ contiguous chunks: +Research the competition -**==> picture [76 x 9] intentionally omitted <==** +Odds are, your competitors are already using social media—and that means +you can learn from what they’re already doing. -Each chunk _𝐶𝑖_ contains _𝑇𝑖_ tokens, where _𝑇𝑖_ ≤ _𝑇_ default, and _𝑇_ default is a predefined chunk size chosen to fit the local model context window. In practice, all chunks except possibly the last one have length _𝑇_ default. +Conduct a competitive analysis -For each chunk _𝐶𝑖_ , we construct the prompt +A competitive analysis allows you to understand who the competition is and +what they’re doing well (and not so well). You’ll get a good sense of what’s +expected in your industry, which will help you set some social media targets +of your own. -Context: _𝐶𝑖_ ; Question: _𝑞_ ; Answer: +This analysis will also help you spot opportunities. For example, maybe +one of your competitors dominates on Facebook, but has put little effort +into Twitter or Instagram. You might want to focus on the networks where +your audience is underserved, rather than trying to win fans away from a +dominant player. -and run a forward pass through the local LLM to extract selfattention scores. Such a forward pass only requires prefilling, and no decoding or token generation is performed. This is sufficient, as our goal is to read off how the model routes attention when interpreting the question, not to produce an answer at the current stage. As the context is placed before the query in the input sequence, each query token can attend to all tokens in _𝐶𝑖_ . We aggregate these attention scores across query tokens, layers, and heads to obtain a relevance score for each context token. +Engage in social listening -4 +Social listening is another way to keep track of the competition. -For the _𝑡_ -th token in chunk _𝐶𝑖_ , the raw attention score is +As you track your competitors’ accounts and relevant industry keywords, you +may notice strategic shifts in the way competitors use their social accounts. +Or you might spot a specific post or campaign that really hits the mark—or +one that bombs. -**==> picture [180 x 29] intentionally omitted <==** +Keep an eye on this information and use to it evaluate your own goals and plans. -where _𝑐𝑖,𝑡_ denotes the _𝑡_ -th token in chunk _𝐶𝑖_ , and _𝐴ℓ,ℎ_ ( _𝑞𝑘,𝑐𝑖,𝑡_ ) is the self-attention scores assigned to _𝑐𝑖,𝑡_ when processing query token _𝑞𝑘_ at layer _ℓ_ and head _ℎ_ . +Related resources -However, we cannot use the raw score _𝑠𝑖,𝑡_ directly. Chunks might have different lengths, most notably the last chunk, as the example shown in Figure 2 Step _○_ 1 section (a) and (b). This variation changes the scale of the resulting attention values. We therefore need normalization. +How to conduct a +competitor audit -_3.1.2 Chunk-Length Normalization._ In Transformer self-attention, the scores for each query token are produced by a softmax over all visible previous tokens. As a result, each query token distributes a fixed total attention mass across the available context tokens. When we aggregate attention across query tokens, layers, and heads to form _𝑠𝑖,𝑡_ , the attention score is influenced not only by token relevance but also by how many tokens share this fixed attention mass. +Getting started with +social listening -This dynamic introduces a structural bias during chunked processing. If two chunks are equally relevant but differ in length, the shorter chunk will inherently exhibit a higher attention scale simply because the same total attention mass is distributed across fewer tokens. As a result, tokens in shorter chunks can appear artificially inflated in importance compared to those in longer chunks. +Watch: How to set up +social listening streams -To make the attention scale aligned across chunks, we apply a simple chunk-length based normalization: +GUIDE / Social Media Marketing Strategy -**==> picture [156 x 21] intentionally omitted <==** +5 -where _𝑇𝑖_ is the length of chunk _𝐶𝑖_ and _𝑇_ default is the default chunk length. + Related resource -This normalization reduces the scores of shorter chunks in proportion to their length, preventing the inflation caused by higher attention scale. As illustrated in Figure 2 Stage _○_ 1 , Step (a) and (b), the normalization is especially important for the final chunk, which is often shorter than _𝑇_ default. +Hootsuite’s social media +audit template -_3.1.3 Reconstructing Full Document Attention._ After computing normalized attention scores for all chunks, we reconstruct a full document attention score by concatenating chunk-level vectors in their original order. Let s˜ _𝑖_ denote the normalized attention vector for chunk _𝐶𝑖_ . We define the full document attention vector as +Step 4 -**==> picture [171 x 9] intentionally omitted <==** +Conduct a social media audit -The resulting vector S is aligned with the original document token order and serves as the full document relevance score for the next stage. +Conducting a social media audit helps you assess how well your current +social media use works for you. -_3.1.4 KV-cache reuse for efficient attention computation._ Computing attention independently for every chunk–query pair would require repeatedly encoding the same chunk content for different queries. To eliminate this redundancy, we reuse the model’s key–value (KV) cache. +Evaluate your current efforts -We decompose the chunk–query input into two parts. The _prefix_ is the context section, while the _suffix_ is the query section. The +If you’re already using social media tools, you need to take a step back and look +at what you’ve already accomplished. Ask yourself the following questions: -**==> picture [118 x 80] intentionally omitted <==** +•• What’s working? -**==> picture [119 x 76] intentionally omitted <==** +•• What’s not working? -**==> picture [239 x 18] intentionally omitted <==** +•• Who’s connecting with you on social? -**----- Start of picture text -----**
-(a) Raw attention without differ- (b) Differential attention with its
ential attention farthest query
**----- End of picture text -----**
+•• Which social media sites does your target market use? +•• How does your social media presence compare to that of your competitors? -**Figure 3: Effect of differential attention on token-level relevance. (a) Without differential attention, generic structural tokens "Introduction" receive high attention. (b) With differential attention, generic content attention is canceled, reducing noise in the attention score.** +Once you gather all this information in one place, you’ll have a good starting +point for planning how to improve your results. -query text is inserted after the Question: marker and varies across queries. +Your audit should give you a clear picture of what purpose each of your +social accounts serves. If the purpose of an account isn’t clear, think about +whether it’s worth keeping. It may be a valuable account that just needs +a strategic redirection, or it may be an outdated account that’s no longer +worth your while. -For each chunk _𝐶𝑖_ , we encode the prefix once and store the resulting KV states. These cached states represent the hidden representations of the chunk tokens and can be reused for multiple queries. We index cache entries using a hash of the chunk token IDs and maintain the cache per document. When switching to a new document, all cached states are cleared to prevent cross-document reuse. +To help you decide, ask yourself the following questions: -When processing a query _𝑞_ for chunk _𝐶𝑖_ , we retrieve the cached KV states and run the model only on the query suffix, supplying the cached states as past key values. During this step, each query token attends to the cached context tokens, allowing the model to produce the same attention scores as if the full prefix–suffix sequence had been processed jointly. We then aggregate these attention scores to compute _𝑠𝑖,𝑡_ without re-encoding the chunk. +1. Is my audience here? -This cache-based execution preserves the exact attention computation while substantially improving efficiency. Each chunk is encoded once and can be reused across many queries, reducing both computation and memory overhead. +2. If so, how are they using this platform? -## **3.2 Differential Attention** +3. Can I use this account to help achieve meaningful business goals? -A full document attention map is not yet sufficient for context selection. Raw attention is a useful relevance score, but it often highlights content that is broadly noticeable rather than truly informative for the target query, such as section headers and formatting markers. Figure 3 illustrates this behavior. In Figure 3a, the model correctly attends to useful evidence, but it also gives high attention to the header token “Introduction,” which is noticeable in the document structure but not necessarily relevant to the query. Because of this, attention-guided selection may waste budget on text that looks important but is not actually useful for answering the question. This observation suggests a simple way to separate signal from background. Many of these high-attention regions are query invariant. They receive attention for a wide range of questions because they are prominent or central to the document structure. If we can estimate this shared background attention, we can subtract it away and keep what remains distinctive to the target query. +Asking these tough questions now will help keep your social media strategy +on track as you grow your social presence. -5 +Look for impostor accounts -**==> picture [242 x 181] intentionally omitted <==** +During your audit process, you may discover fraudulent accounts using your +business name or the names of your products—that is, accounts that you +and your business don’t own. -**----- Start of picture text -----**
-Each query token distributes total attention
Document D mass 1 over 𝐷.
Summing across query tokens scales with
Query 𝑞 Query 𝑞 [−] query length; we average over query tokens .
What is the What is the Differential attention
title of the publication year score for query q
paper? and venue of
𝑑 𝑞= 𝑺 [ഥ] 𝑞− 𝑺 [ഥ] 𝑞 [−]
the paper?
Compute
attention scale
for 𝑞 and 𝑞 [−]
1
Query-to- Query-to- ഥ𝑺 𝑞= 𝑇𝑞 𝑺 𝑞𝑘
document document
attention attention 1 −
𝐒 𝑞𝑘 𝐒 𝑞𝑘− ഥ𝑺 𝑞 [−] = 𝑇𝑞− [ 𝑺] 𝑞𝑘
Query Normalization
Unified attention scale Moderate attention scale High attention scale
**----- End of picture text -----**
+These imposter accounts can be harmful to your brand (never mind +capturing followers that should be yours), so be sure to report them. You +may want to get your social accounts verified to ensure your fans and +followers know they are dealing with the real you. +GUIDE / Social Media Marketing Strategy -**Figure 4: Query normalization for differential attention. Each query token distributes a fixed total attention scale over** _𝐷_ **,** S( _𝑞_ ) **is the aggregation over all heads and layers across tokens with query length. We therefore average the Queryto-document attention scale to obtain** S[¯] ( _𝑞_ ) = _𝑇_ 1 _𝑞_[S][(] _[𝑞]_[)] **[and]** ¯S( _𝑞_[−] ) = _𝑇𝑞_ 1[−][S][(] _[𝑞]_[−][)] **[, then compute]** _[ 𝑑]_[(] _[𝑞]_[)][=][ ¯S][(] _[𝑞]_[) −][¯S][(] _[𝑞]_[−][)] **[.]** +6 -As illustrated in Figure 3b, we compute attention for the target query and for a contrasting query, then take their difference. Regions that are highly attended under both queries are treated as background and largely cancel out; regions that matter primarily for the query remain prominent, yielding a cleaner, more query-specific relevance score. + Related resources -This framing raises two practical questions that we address next. First, how to perform the subtraction properly? Second, how do we construct an effective contrasting query that captures documentwide noise while minimally overlapping with the true evidence needed for the target query? +Watch: How to optimize +your social profiles -One practical challenge is that we cannot directly subtract the two aggregated attention scores when the queries have different lengths. Each query token spreads a fixed amount of attention mass over the document, so a longer query naturally accumulates a larger total attention magnitude even when the overall attention pattern is similar. To make the two scores comparable, we first normalize attention by query length before taking the difference. +We’ve also got step-by-step +guides for each network +to walk you through the +process: -_3.2.1 Query Normalization._ Figure 4 illustrates why query normalization is necessary. Consider two queries over the same document: a shorter target query _𝑞_ , "What is the title of the paper?" and a longer contrasting query _𝑞_[−] , "What is the publication year and venue of the paper?". Even if both queries attend to the same generic regions of the document, the longer query contributes more total attention mass simply because it has more query tokens, causing a misalignment in attention scale. As a result, directly subtracting the aggregated scores S( _𝑞_ ) and S( _𝑞_[−] ) would confuse query length with relevance. +Create a Facebook +business page -To remove this bias, we convert each reconstructed query-todocument attention score into an average per-query-token score. Using the full document attention vector from Equation (3), we write S( _𝑞_ ) and S( _𝑞_[−] ) for the reconstructed attention scores obtained with the target query _𝑞_ and the contrasting query _𝑞_[−] , respectively. +Create an Instagram +business account -We then normalize each score by its query length: +Create a Twitter business +account -**==> picture [151 x 21] intentionally omitted <==** +Create a Snapchat +account -**==> picture [196 x 10] intentionally omitted <==** +Create a LinkedIn +Company Page -**==> picture [159 x 22] intentionally omitted <==** +Create a Pinterest +business account -We then define the differential attention score for the target query as +Create a YouTube channel -**==> picture [162 x 10] intentionally omitted <==** +Create a WeChat business +account -This normalization unifies the attention scale of the two queries before subtraction, so the difference reflects relevance rather than query length. In practice, it stabilizes differential attention when query lengths vary and prevents longer queries from dominating the difference purely due to scale. +Don’t let this list overwhelm +you. Remember, it’s better to +use fewer channels well than +to stretch yourself thin trying +to maintain a presence on +every social network. -The resulting score d( _𝑞_ ) highlights regions that are more strongly attended by the target query than by the contrasting query. Tokens that attract attention under both queries tend to cancel out and move closer to zero, while tokens that are uniquely relevant to the target query receive larger positive values. +Finally, here’s a quick +reference guide for image +sizes for every network. -_3.2.2 Applying Differential Attention in SAGE._ To make differential attention effective, the contrasting query should attend to document regions that differ from those needed by the target query. This improves separation between target-specific evidence and shared background attention, and therefore increases the denoising effect of subtraction. +Step 5 -- In our implementation, we consider three variants for analysis: +Set up accounts and improve existing profiles -- **Raw query (** _𝑞_ **).** The user’s original question, without differential subtraction. +Decide which networks you’ll focus on, and then set up and optimize your +accounts. -- **Fixed contrast query.** A fixed prompt shared across the dataset. For this paper, we used _“Please repeat the context.”_ . +Determine which networks to use (and how to use them) -- **Farthest query (** _𝑞_[−] **).** A query selected to be maximally dissimilar to the target query. When multiple questions are available for the same document, we choose the farthest query as the one whose embedding is most distant from the target query embedding. By single differential attention calculation, we got target-specific evidence revealed for both queries. +As you decide which social channels to use, you’ll also need to define your +strategy for each network. For example, you might decide to use Twitter for +customer service, Facebook for customer acquisition, and Instagram for +engaging existing customers. -We compare these variants empirically in Section 4.6 and show that using a semantically distant contrast query yields a cleaner relevance score and improves downstream selection quality. +It’s a good exercise to create mission statements for each network. These +one-sentence declarations will help you focus on a very specific goal for +each account on each social network. -## **3.3 Unit Window Selection with Budget Control** +For example, you could decide that: -Given the unified attention scores from the earlier stages, SAGE converts token-level relevance signals into a set of coherent text snippets that the generation model can consume under a strict token budget. +•• Facebook is best for acquiring new customers via paid advertising. -Although attention is computed at the token level, selecting only isolated high-scoring tokens produces a fragmented context that is poorly suited for generation. LLMs generally benefit from contiguous spans that preserve local semantics. We therefore perform budget-aware selection over fixed-length windows and merge overlapping windows into coherent snippets. To reduce the effect of spiky token-level scores, we also apply a light local smoothing step before window scoring so that nearby tokens in the same evidence span receive more consistent scores. +•• -6 +Instagram is where you build brand affinity with existing customers. -We define a user-specified budget _𝑏_ ∈(0 _,_ 1] as the fraction of the document to retain. The total number of selected tokens is constrained to ⌈ _𝑏𝑇𝐷_ ⌉. We use a base window ratio _𝑤_ = 0 _._ 02 (2% of the document length) and define the window length _𝐿_ as +•• Twitter is where you engage press and industry influencers. -**==> picture [88 x 27] intentionally omitted <==** +•• LinkedIn is where you engage existing employees and attract new talent. -When _𝑏 < 𝑤_ , selecting a single window preserves the most relevant contiguous region under the budget. When the budget is larger than _𝑤_ , fixing the window size allows SAGEto select multiple evidence regions distributed across the document. +•• YouTube is where you support existing customers with education and -We slide a length _𝐿_ window over the full document attention matrix and compute an aggregate score for each window. Windows are ranked by score, and selection proceeds greedily under the budget constraint. If a newly selected window overlaps with previously selected spans, the spans are merged. Otherwise, the window is added as a new span. This process continues until the token budget is exhausted. +video help content. -This strategy avoids two common failure modes. First, selecting a single long span may introduce substantial irrelevant text while missing important evidence in other parts of the document. At the other extreme, selecting many disjoint high-scoring tokens or tiny spans yields a fragmented context that is difficult for the generation model to use. Unit-window selection balances these extremes by concentrating the budget on high utility regions while preserving local coherence. +•• Snapchat is where you distribute content with the goal of building brand -Finally, the selected snippets are assembled in an original logical order into a reduced context prompt and then passed to a generation model to produce the final answer. +awareness with younger consumers. -## **4 EVALUATION** +If you can’t create a solid mission statement for a particular social network, +you may want to reconsider whether that network is worth it. -We evaluate SAGE on four datasets that stress attention guided context reduction from different angles, including document length, structure, modality, and question format: QuALITY [35], Paper [23], Notice [23], and AIT-QA [16]. We report task performance under varying token budgets and compare SAGE against RAG baselines on each dataset. For evaluation, we use an LLM as a judge to provide a binary correctness decision. We also compute cosine similarity for all questions to measure semantic alignment between SAGE’s generated responses and the ground truth. Together, these metrics provide a complementary view of output quality. +Set up (and optimize) your accounts -Rather than treating each dataset as an isolated benchmark, we use them to examine complementary aspects of long context understanding, including reasoning across distributed evidence, robustness to domain variation, efficiency under tight token budgets, and adaptability to inputs. +Once you’ve decided which networks to focus on, it’s time to create your +profiles—or improve existing profiles so they align with your strategic plan. -We structure our evaluation around the following guiding questions: +In general, make sure you fill out all profile fields, use keywords people will +use to search for your business, and use images that are correctly sized for +each network. -- (Q1) Does SAGE consistently outperform retrieval-based pipelines that use strong embedding models under aligned token budgets? +GUIDE / Social Media Marketing Strategy -- (Q2) Can SAGE handle reasoning-intensive questions when relevant evidence is hard to find by keywords or semantic similarity? +7 -- (Q3) How does token budget size affect accuracy? + Step 6 -- (Q4) How does the size of the attention model influence context selection quality and accuracy? +Find inspiration -- (Q5) How does SAGE compare with retrieval pipelines in runtime cost and computational overhead? +While it’s important that your brand be distinctive and unique, you can still +draw inspiration from other businesses that are great on social. -- (Q6) Does SAGE remain effective across different domains and document structures without dataset-specific tuning? +Social network success stories -- (Q7) How do different differential attention strategies influence context selection quality? +All of the social networks feature success stories that highlight how brands +are using their tools effectively. You can usually find these on the business +section of the social network’s website. (For example, take a look at the +Facebook business success stories.) These case studies offer valuable +insights you can apply to your own goals for each social network. -- (Q8) Can SAGE remain effective under budgets smaller than a single retrieval chunk, where chunk-based methods become inflexible? +Award-winning accounts and campaigns -- (Q9) Can attention-guided selection extend to other modalities while preserving task accuracy? +For examples of brands that are at the top of their social media game, check +out the winners of The Facebook Awards or The Shorty Awards. -## **4.1 Datasets** +Step 7 -The four datasets varied in different modalities, context lengths, and QA formats. They also varied in their primary focus and the number of questions, allowing us to evaluate SAGE comprehensively across diverse settings. Table 1 summarizes these key characteristics. +Create a social media content calendar -**QuALITY** [35] is a long passage, multiple-choice benchmark for reading comprehension and reasoning. We evaluate on **QuALITYhard** , a difficult subset constructed by labeling questions that annotators could not reliably answer under time-constrained search, but could answer in an untimed setting. As a result, QuALITY-hard emphasizes questions that require integrating evidence across the passage rather than relying on localized keyword matches. +Sharing great content is essential, of course, but it’s equally important to have +a plan in place for when you’ll share content to get the maximum impact. -**Paper** and **Notice** are derived from ZenDB [23] and provide multi-question workloads over the same document. _Paper_ contains very long scientific publications with both localized metadata queries and open-ended semantic questions, such as the main contribution. This complexity makes it a challenging setting for retrieval and a highly useful testbed for analyzing differential attention. _Notice_ consists of regulatory notices written in formal, repetitive legal language. Its questions target specific fields whose answers are typically localized within the document. +Your social media content calendar also needs to account for the time you’ll +spend interacting with your audience (although you need to allow for some +spontaneous engagement as well). -Finally, **AIT-QA** [16] evaluates whether attention-guided selection extends beyond free text to semi-structured tables. It contains 515 questions over 116 tables with complex headers and domainspecific terminology. +Create a posting schedule -## **4.2 Baselines** +Your social media content calendar lists the dates and times at which you will +publish types of content on each channel. It’s the perfect place to plan all of your +social media activities—from images and link sharing to blog posts and videos. -We compare SAGE against a standard RAG pipeline. We instantiate the RAG baselines with top-performing embedding models from the MTEB leaderboard [30, 32], including _Octen-Embedding4B_ , _Qwen3-Embedding-8B_ , and _UAE-Large-V1_ . These three models cover a strong range of retrieval quality and computational cost, and represent competitive, widely used choices in modern RAG systems. +Your calendar ensures your posts are spaced out appropriately and +published at the optimal times. It should include both your day-to-day posts +and your content for social media campaigns. -Retrieval is performed at the document level. For each query, RAG searches only among chunks belonging to the corresponding document, rather than across the entire corpus. For each document, we tokenize the text and split it into overlapping chunks of fixed length. We compute embeddings for all chunks and retrieve the top- _𝑘_ chunks most similar to the query using cosine similarity. Retrieved chunks are reordered according to their original positions in the document, deduplicated in overlapping regions, concatenated, +Related resources -7 +How to create a social +media content calendar -**Table 1: Summary of evaluation datasets. The four datasets provide complementary test settings across different perspectives.** +Watch: How to save time +with bulk scheduling -|**Dataset**|**Modality**|**Context**|**QA format**|**Main focus**|**# Questions**| -|---|---|---|---|---|---| -|**QuALITY-**|Text (passage)|Long-|Multiple-choice|Multi-hop and Reasoning intensive|hard subset of QuALITY| -|**hard**||context||reading comprehension|| -|**Paper**|Text (scientifc|Very long-|Open ended QA|Mixed factual and semantic extrac-|9 per document| -|(ZenDB)|papers)|context||tion; long-document analysis|| -|**Notice**|Text (regulatory|Long
con-|Open ended QA|Localized factual extraction under|6 per document| -|(ZenDB)|notices)|text||aligned token budgets|| -|**AIT-QA**|Table
(airline|Semi-|QA|Table QA over complex headers and|515 (over 116 tables)| -||SEC flings)|structured||domain-specifc content|| +GUIDE / Social Media Marketing Strategy +8 + Plot your content mix -and then fed to the answer-generation model. To isolate the effect of context selection, we use the same answer generation model and decoding configuration for both RAG and SAGE. In particular, generation is conditioned on the selected context and the query, with the temperature fixed to 0.0 to eliminate stochastic variation. We vary _𝑘_ to assess how retrieval depth impacts performance and, when feasible, adjust the resulting prompt length to match SAGE’s token budget, enabling fair, token-aligned comparisons. +Make sure your calendar reflects the mission statement you’ve assigned to +each social profile, so that everything you post is working to support your +business goals. For example, you might decide that: -## **4.3 Evaluation Metrics** +•• 50 percent of content will drive traffic back to your blog -We evaluate answer quality using two complementary criteria: _LLM-based judgment_ and _semantic similarity_ . These metrics are chosen to reflect the heterogeneous answer formats across datasets, ranging from discrete multiple choice and numeric outputs to long, open-ended textual responses. +•• 25 percent of content will be curated from other sources -_LLM-Based Judgment._ Evaluating long-context QA is challenging because many questions admit multiple valid phrasings, and small surface differences do not necessarily indicate incorrect reasoning. We therefore use an LLM as a judge to apply explicit, taskaware grading rules and produce a binary correctness decision. For QuALITY, answers are discrete, so correctness is evaluated by exact match. For the other datasets, the judge directly decides whether an answer is correct, allowing harmless rephrasing while penalizing missing or incorrect core content. +•• 20 percent of content will support enterprise goals (selling, lead -As answer formats and grading criteria differ across datasets, we design a separate judging prompt for each dataset. Each prompt specifies the expected answer format and the criteria for correctness, and is shared by SAGE and all baselines within that dataset. To ensure the judge behaves as intended, we manually inspected the prompts and audited a substantial sample of judged instances (about 200 per dataset) to verify that the outputs follow the stated rules. +generation, and so on) -_Semantic Similarity Score._ Binary correctness can hide meaningful differences among partially correct answers, especially for long, open-ended responses where the format may vary even if the core idea aligns with the ground truth. To better characterize answer quality, we additionally compute a semantic similarity score between the generated answer and the reference. Specifically, we embed both texts and report their cosine similarity in [0,1], which is robust to paraphrasing and reflects partial semantic overlap. We analyze similarity at the distribution level using Cumulative Distribution Functions (CDFs) to compare how often each method produces high-similarity answers under different context budgets. +•• 5 percent of content will be about your employees and company culture -## **4.4 Experimental Setup** +Placing these different post types in your content calendar will help ensure +you maintain the ratio you’ve planned. If you’re starting from scratch and +you’re simply not sure what types of content to post, try the 80-20 rule: -To enable a fair, controlled comparison, we use separate models for attention computation, retrieval, answer generation, and evaluation, while keeping all other settings and pipeline components fixed. +•• 80 percent of your posts should inform, educate, or entertain your audience -_Attention Models._ For SAGE, attention scores are computed using three language models of increasing scale: _LLaMA-3.2-1B_ , _Qwen38B_ , and _Qwen3-14B_ . These models are used exclusively for attention estimation and do not participate in answer generation. This setup allows us to study the impact of model capacity on attention quality while keeping the downstream generation fixed. +•• 20 percent can directly promote your brand -_Answer Generation and Evaluation._ All methods, including RAG baselines, use _Gemini-2.5-Flash-Lite_ for answer generation. By fixing the generation model and decoding configuration, we isolate the effect of context selection. Unless stated otherwise, SAGE uses the farthest query differential attention variant when constructing the reduced context for generation. LLM-based evaluation is performed using _GPT-4o-mini_ across all datasets. +You could also try the social media rule of thirds: -_Embeddings._ Different embedding models are used at different stages to balance efficiency and semantic fidelity. For farthest question selection in differential attention, we use _all-MiniLM-L6-v2_ due to its low computational cost and reliable relative distance estimates. For semantic similarity evaluation, we use _Qwen3-Embedding-0.6B_ , which provides stronger semantic representations while remaining efficient at scale. +•• One-third of your social content promotes your business, converts -## **4.5 QuALITY** +readers, and generates profit -QuALITY-hard [35] is designed to stress long-context reasoning. The questions are typically unambiguous, but answering them often requires combining evidence from multiple parts of a passage rather than matching a single keyword-rich context. This makes QuALITY-hard a direct test of whether attention-guided context reduction can preserve reasoning quality when relevant evidence is not reliably surfaced by retrieval similarity. In this section, we focus on answering (Q1) to (Q5). +•• One-third of your social content shares ideas and stories from thought -_4.5.1 Effectiveness compared to strong RAG baselines (Q1, Q2)._ Figure 5 shows that SAGE outperforms on QuALITY-hard across a range of tested token budgets. The strength of SAGE comes from better context selection, not from using a large attention model. For instance, at a 10% token budget, SAGE with Qwen3-14B achieves similar accuracy compared to RAG configurations that use nearly four times as many tokens. Moreover, even the smallest attention +leaders in your industry or like-minded businesses -8 +•• One-third of your social content involves personal interactions with your -**==> picture [242 x 172] intentionally omitted <==** +audience -**----- Start of picture text -----**
-0.70
0.65
0.60
0.55
0.50
SAGE
0.45 LLaMA-3.2-1B
Qwen3-8B
0.40 QRAG Baselineswen3-14B
Qwen3-Embedding-8B
0.35 UAE-Large-V1
Octen-Embedding-4B
0.30
5 10 15 20 25 30 35 40
Avg. Token Usage (%)
Avg. Accuracy
**----- End of picture text -----**
+Once you have your calendar set, use scheduling tools or bulk scheduling to +prepare your posting in advance rather than updating constantly throughout +the day. This allows you to focus on crafting the language and format of your +posts, rather than writing them on the fly whenever you have time. +GUIDE / Social Media Marketing Strategy -**Figure 5: Accuracy on QuALITY-hard versus average token usage. We compare SAGE using attention models of different sizes with RAG baselines using different retrieval embeddings. Even the 1B attention model outperforms RAG, and performance improves as model size increases.** +9 -model, _LLaMA-3.2-1B_ , outperforms all strong RAG baselines across the evaluated budget range. SAGE works well as it selects evidence based on attention signals that reflect how the question connects to different parts of the passage, which helps when evidence is spread across the document. Overall, these results address (Q1) and (Q2): Across all evaluated settings, SAGE consistently outperforms RAG baselines at similar or lower token usage, including RAG configurations using strong retrieval embeddings. + Step 8 -_4.5.2 More token budget is not always better (Q3)._ As the token budget increases, both SAGE and RAG generally improve in Figure 5. However, at larger budgets, the marginal gains diminish, and the curves begin to plateau. Concretely, once the prompt already contains the key evidence, adding more context yields few additional correctly answered questions. This pattern suggests that QuALITY-hard is not primarily bottlenecked by raw context length. Instead, performance depends largely on retrieval quality and on whether the prompt contains the _right_ evidence. The observation is consistent with prior findings that adding more retrieved context does not necessarily translate into better long-context QA accuracy [13, 28]. Taken together, these results answer (Q3): accuracy saturates beyond moderate budgets, indicating that evidence selection matters more than simply providing more tokens. +Test, evaluate, and adjust your strategy -Token usage between SAGE and RAG is not strictly aligned on QuALITY. QuALITY documents vary substantially in length, approximately 1.5K to 6K words without punctuation [35]. Because RAG retrieves fixed-size chunks, the final prompt length depends on the number of retrieved chunks and the amount of overlap among them, which makes it difficult to hit an exact token budget in practice. In contrast, SAGE directly optimizes for a user-specified budget and can reliably meet it, while chunk-based retrieval may overshoot or undershoot the desired token limit. +Your social media strategy is a hugely important document for your +business, and you can’t assume you’ll get it exactly right on the first try. -_4.5.3 Ablation on attention model size (Q4)._ As we increase the attention model size from _LLaMA-3.2-1B_ to _Qwen3-8B_ and _Qwen314B_ , accuracy improves consistently across budgets. The effect +As you start to implement your plan and track your results, you may find +that some strategies don’t work as well as you’d anticipated, while others are +working even better than expected. -is strongest under tight constraints, where selection mistakes are most costly. At a 5% token budget, accuracy increases from **0.36** (1B) to **0.46** (8B) and **0.48** (14B), meaning that larger attention models enable SAGE to answer substantially more questions correctly while operating under the same strict budget. At higher budgets the gap shrinks, again reflecting diminishing returns once most key evidence is already included. +Track your data -Importantly, SAGE does not require a large attention model to be effective: even the 1B scorer remains competitive and already outperforms the strongest RAG baselines in Figure 5. Overall, these results answer (Q4), increasing attention model capacity improves selection quality and yields higher QA accuracy, with the largest gains under tight budgets. +In addition to using the analytics tools available within each social network, +use UTM parameters to track visitors as they move through your website, so +you can see exactly which social posts drive the most traffic to your website. -_4.5.4 Runtime and computational overhead (Q5)._ We further examine the runtime on QuALITY-hard. For RAG, the dominant overhead comes from computing embeddings for all **8,798** chunks, which takes **98.2s** (UAE-Large-V1), **370.7s** (Qwen3-Embedding-8B), and **838.6s** (Octen-Embedding-4B). For SAGE, using _LLaMA-3.2-1B_ as the attention model, we answer **3,797** queries in **690.9s** , with **3,011** cache hits that reuse document encodings across queries. Overall, SAGE is slower than the UAE and Qwen3 RAG baselines, but faster than the Octen baseline. Taken together, these results answer (Q5). SAGE operates in a similar runtime regime to retrieval pipelines while achieving materially higher QuALITY-hard accuracy. +Re-evaluate, test, and do it all again -_4.5.5 Public leaderboard: competitive ranking under strict budgets._ On the public QuALITY leaderboard [34], SAGE ranks **4th** on QuALITY-hard while using only a **10%** context budget with _Qwen38B_ . This demonstrates that the accuracy gains we observe in controlled experiments translate to a competitive end-to-end system under strict token constraints. Notably, SAGE achieves this ranking with substantially lower token usage than many other highperforming systems. Many of those approaches rely on heavier pipelines, such as fine-tuning and longer inputs, which typically lead to higher inference latency. +When data starts coming in, use it to reevaluate your strategy regularly. +You can also use this information to test different posts, campaigns, and +strategies against one another. Constant testing allows you to understand +what works and what doesn’t, so you can refine your strategy in real time. -## **4.6 Paper** +Surveys can also be a great way to find out how well your strategy is working. +Ask your social media followers, email list, and website visitors whether +you’re meeting their needs and expectations on social media. You can even +ask them what they’d like to see more of—and then make sure to deliver on +what they tell you. -We next evaluate SAGE on the _Paper_ dataset from ZenDB [23]. Compared to QuALITY-hard, Paper pushes context reduction into a more realistic long document setting. Scientific papers are substantially longer and follow a clear section structure. Their questions vary from asking localized metadata, such as venue and year, to open-ended semantic queries like the main contribution of a paper. This combination lets us test whether the advantages of SAGE extend beyond narrative passages. Concretely, _Paper_ answers (Q1), (Q2), (Q3), and (Q7), and serves as evidence for (Q6). Without any dataset-specific tuning, SAGE remains effective on Paper, indicating that the benefits of attention-guided context reduction extend beyond narrative passages to structurally complex scientific writing. +Things change fast on social media. New networks emerge, while others +go through significant demographic shifts. Your business will go through +periods of change as well. All this means that your social media strategy +should be a living document that you look at regularly and adjust as needed. +Refer to it often to keep you on track, but don’t be afraid to make changes +so that it better reflects new goals, tools, or plans. -_4.6.1 LLM-judge accuracy: (Q1) to (Q3)._ Figure 6 reports LLMjudge accuracy over average token usage for several representative questions. Overall, SAGE achieves higher accuracy than RAG baselines at similar budgets ((Q1)), and its performance increases smoothly with additional context, indicating that the extra tokens selected by attention are generally helpful rather than distracting. +When you update your social strategy, make sure to let everyone on your +social team know, so they can all work together to help your business make +the most of your social media accounts. -9 +GUIDE / Social Media Marketing Strategy -**==> picture [506 x 167] intentionally omitted <==** +10 -**----- Start of picture text -----**
-SAGE Farthest RAG Qwen3 Embedding 8B RAG Uae Large V1 RAG Bflhc Octen Embedding 4B
1.0 1.0 1.0 1.0
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0.0 0.0 0.0 0.0
0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15
Average tokens used (%) Average tokens used (%) Average tokens used (%) Average tokens used (%)
(a) What is the number of authors (b) What is the main contribution (c) What is the publication year (d) What is the title of the paper
of the paper and venue of the paper
Accuracy Accuracy Accuracy Accuracy
**----- End of picture text -----**
+ Bonus: Download our social media +strategy template +Does this all feel a little overwhelming? The truth is that building your social +media strategy is a substantial job. It should be, since it’s such an important +document for your business. But it doesn’t have to be complicated. +We’ve created a template to guide you through the whole process of +creating your social media marketing plan. Visit our blog to download it (plus +six other social media templates that can save you hours of work). -**Figure 6: LLM-judge accuracy on the Paper dataset. Accuracy vs. average token usage over documents for SAGE and RAG baselines across representative questions.** +Take the next leap in your social media career -**==> picture [506 x 150] intentionally omitted <==** +Hootsuite Academy -**----- Start of picture text -----**
-SAGE 0.5% SAGE 10.0% Best RAG 1.5% Best RAG 15.5%
SAGE 1.0% SAGE 15.1% Best RAG 10.9%
1.0 1.0 1.0 1.0
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0.0 0.0 0.0 0.0
1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00
Similarity Score Similarity Score Similarity Score Similarity Score
(a) What is the number of au-(b) What is the main contribution (c) What is the publication year (d) What is the title of the paper
thors of the paper and venue of the paper
P(Similarity > x) P(Similarity > x) P(Similarity > x) P(Similarity > x)
**----- End of picture text -----**
+Earn industry-recognized social media certifications that will make you stand +out from the crowd. +Begin with our free comprehensive training on social marketing. All of our +courses are delivered online and taught by expert industry practitioners. +Take the free course here. -**Figure 7: Semantic similarity CDFs on the Paper dataset. CDFs of cosine similarity between generated and ground-truth answers, comparing SAGE to the best RAG baseline at matched average token usage (including the 0.5% budget). Results use Qwen3-8B.** +Build a professional workflow with Hootsuite -The difference is most visible on reasoning-intensive questions where evidence is not well captured by keyword similarity, addressing (Q2). For example, on _“What is the main contribution of the paper?”_ , SAGE improves rapidly with budget and remains consistently above RAG across the displayed range. In other words, as we allow more context, the extra tokens selected by SAGE more often contain the missing evidence needed to answer the question correctly, whereas retrieving more chunks improves correctness more slowly. +From scheduling Instagram posts to advanced ROI measurement, Hootsuite’s +flexible platform helps you execute every aspect of your social media strategy. +We’re the world’s most widely used platform for managing social media. -For questions with more localized answers, such as publication year and venue in Figure 6c, both methods benefit from additional context, but SAGE reaches high accuracy with fewer tokens. For metadata queries such as title and author count, SAGE again achieves consistently stronger accuracy under the same budgets, reflecting more reliable selection of document regions that contain the relevant fields. +Get started with the plan that fits your needs here. -We also evaluated larger budgets up to 40%. However, for most questions, performance flattens out at moderate budgets, with only +Did you find this guide helpful? -a small subset continuing to improve at higher budgets. This answers (Q3) and motivates focusing Figure 6 on the most informative range up to 15%, which captures the most informative setting where differences between methods are most pronounced. +Tell us whether we should praise our writers or shout at them. Either way, +we appreciate your honest feedback. Take our 4-minute survey here. -_4.6.2 Semantic similarity: (Q1) to (Q3)._ We complement judge accuracy with semantic similarity distributions. Figure 7 plots CDFs of cosine similarity between generated answers and ground truth references. For a similarity threshold _𝑥_ , the y-axis reports the fraction of questions whose similarity exceeds _𝑥_ . Curves that lie higher indicate better overall answer quality. We plot the similarity threshold from 1 to 0 on the x-axis to make high similarity behavior easier to compare. We also evaluated ROUGE. It shows the same overall trends. For clarity, we report cosine similarity. +GUIDE / Social Media Marketing Strategy -Across representative questions, SAGE dominates the best performing RAG baseline in the middle-to-high similarity region, reinforcing (Q1) by showing improvements in overall answer quality rather than only binary pass rates. The largest gains again appear on open-ended reasoning questions such as “main contribution”. - -10 - -**==> picture [506 x 160] intentionally omitted <==** - -**----- Start of picture text -----**
-Raw (b=0.005) Raw (b=0.15) Fixed Question (b=0.005) Fixed Question (b=0.15) Farthest (b=0.005) Farthest (b=0.15)
Raw (b=0.1) Raw (b=0.4) Fixed Question (b=0.1) Fixed Question (b=0.4) Farthest (b=0.1) Farthest (b=0.4)
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00
Similarity Score Similarity Score
(a) what is the authors of the paper (b) what is the main contribution of
the paper
P(Similarity > x) P(Similarity > x)
**----- End of picture text -----**
- - -**Figure 8: Effect of differential attention on the Paper dataset. Semantic similarity CDFs compare Raw attention with differential attention variants under different token budgets. Experiments are conducted using Qwen3-8B.** - -This supports our answer to (Q2). Under the same budget, SAGE produces a higher fraction of answers that closely match the reference semantics, while increasing retrieval depth yields smaller shifts in the distribution. The CDFs also answer (Q3): increasing the budget generally shifts curves upward, but the improvements flatten out at larger budgets. - -Taken together, the judge and similarity views agree that SAGE converts budget into answer quality more effectively than retrieval, especially for reasoning-intensive semantic questions. - -_4.6.3 Different Differential Attention Variant (Q7)._ We study the differential attention variant on _Paper_ . Figure 8 compares raw attention with fixed contrast query and farthest query. Across budgets and representative questions, both differential variants shift the similarity CDFs upward relative to raw attention, meaning they produce a larger fraction of high-quality answers under the same token budget. This directly answers (Q7): differential attention strengthens context selection by filtering out noise and producing cleaner, query-specific relevance scores. - -Between the two variants, farthest question performs best overall. Its CDF curves dominate the fixed contrast variant across most similarity thresholds, with especially clear gains on reasoning questions (Figure 8b). In practice, prefer an informative contrast question when available. Otherwise, use fixed contrast, which still delivers strong performance. - -## **4.7 Notice** - -We next evaluate on the _Notice_ dataset from ZenDB [23]. Notice documents have less variation in length, which enables more controlled comparisons with aligned token usage between SAGE and RAG. Extremely small budgets that fall below a single retrieval chunk are studied on this dataset. As Notice has a markedly different document style, the results below provide additional evidence for (Q6), showing SAGEremains effective without specific tuning. - -_4.7.1 Accuracy and semantic quality under aligned budgets._ Figure 9 reports LLM-judge accuracy versus average token usage for representative questions, and Figure 10 complements this view with semantic similarity distributions. Across questions, SAGE matches - -**Table 2: AIT-QA results comparing full-table RAG and Top-4 row selection methods.** - -||**Method**
Full Table
SAGE|**Value**
0.88
0.87|**Avg. Row Usage**
100%
51%| -|---|---|---|---| - - - -or outperforms strong RAG baselines at aligned token usage, while also supporting substantially smaller budgets (Q1). - -_4.7.2 Budgets below a single retrieval chunk (Q8)._ Notice highlights a practical limitation of chunk-based retrieval. Even retrieving a single chunk incurs a fixed token cost, which limits its flexibility under tight budgets. In contrast, SAGE selects evidence at finer granularity and remains effective even when the budget is smaller than a single retrieved chunk. This answers (Q8): SAGE can operate at extremely low budgets that chunk-based RAG cannot meaningfully support. - -_A failure mode at larger budgets._ An instructive exception appears for _“What are the state abbreviation and ZIP code of the company?”_ in Figure 9d. Here, SAGE performs best at small budgets, but accuracy decreases as the budget grows and approaches the RAG baselines. The same trend appears in the similarity CDF (Figure 10d). This behavior reflects a limitation of greedy attention-based selection in repetitive documents. With a small budget, SAGE typically captures the query-relevant ZIP code. As the budget increases, additional sections containing other ZIP codes may be included, introducing competing candidates and reducing precision. More broadly, this highlights that adding more context is not always beneficial when documents contain repeated entities and near duplicate fields. - -## **4.8 AIT-QA** - -We finally evaluate whether attention-guided selection extends beyond free-form text to semi-structured inputs, addressing (Q9), using the _AIT-QA_ dataset [16]. Each document in AIT-QA is a structured table represented in JSON format. As a result, neither standard RAG nor token-based context reduction can be applied - -11 - -**==> picture [506 x 167] intentionally omitted <==** - -**----- Start of picture text -----**
-SAGE Farthest RAG Qwen3 Embedding 8B RAG Uae Large V1 RAG Bflhc Octen Embedding 4B
1.0 1.0 1.0 1.0
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0.0 0.0 0.0 0.0
0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60
Average tokens used (%) Average tokens used (%) Average tokens used (%) Average tokens used (%)
(a) What are the violation items (b) What is the date of the notice (c) What is the name of the com-(d) What are the state abbreviation
pany and ZIP code of the company
Accuracy Accuracy Accuracy Accuracy
**----- End of picture text -----**
- - -**Figure 9: LLM-judge accuracy on the Notice dataset. Accuracy versus average token usage for SAGE and RAG baselines across representative questions. SAGE supports budgets smaller than a single RAG chunk. For question (d), accuracy decreases as token usage increases; we discuss this failure mode in Section 4.7.** - -**==> picture [506 x 154] intentionally omitted <==** - -**----- Start of picture text -----**
-SAGE 5.0% SAGE 38.1% SAGE 74.2% Best RAG 22.8% Best RAG 53.0% Best RAG 74.0%
SAGE 22.1% SAGE 53.1% Best RAG 7.6% Best RAG 37.9%
1.0 1.0 1.0 1.0
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0.0 0.0 0.0 0.0
1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00 1.00 0.75 0.50 0.25 0.00
Similarity Score Similarity Score Similarity Score Similarity Score
(a) What are the violation items (b) What is the date of the notice (c) What is the name of the com-(d) What are the state abbreviation
pany and zip code of the company
P(Similarity > x) P(Similarity > x) P(Similarity > x) P(Similarity > x)
**----- End of picture text -----**
- - -**Figure 10: Semantic similarity CDFs on the Notice dataset. CDFs compare SAGE and the best performing RAG baseline under matched average token budgets across representative questions. Results are obtained using Qwen3-8B for attention estimation.** - -directly. If a table is split into text chunks or partial windows would break row–column structure and substantially hinder reasoning. - -To handle this setting, we adapt SAGE to operate at the _row level_ . We compute query-aware attention over the full table in its original JSON representation. For each row, we aggregate tokenlevel attention into a single relevance score and then select the top- _𝑘_ rows. The reduced prompt includes only the selected rows along with the column headers. - -Table 2 compares this row-selection variant with a full-table baseline. Selecting only the top rows achieves accuracy comparable to using the full table, while substantially reducing the amount of table content passed to the generation model. This answers (Q9), SAGE can preserve task accuracy on semi-structured tables while enabling budget-aware context reduction. - -These findings show that SAGE remains effective beyond unstructured text, highlighting its potential as a flexible, budget-aware context selection framework that transfers across data types without additional fine-tuning (Q6). - -## **5 CONCLUSION** - -We presented SAGE, a training-free, plug-and-play context reduction framework that leverages a lightweight local LLM to extract highly compact, fine-grained, and query-relevant subsets from long documents under strict token budgets. Across four complementary benchmarks, SAGE consistently outpaces strong RAG baselines in token efficiency, demonstrating its largest gains on complex, reasoning-intensive questions. Furthermore, by utilizing differential attention, the framework remains highly effective even on substantially longer, noisy scientific papers. Beyond standard text, our results on AIT-QA confirm that SAGE naturally extends to structured inputs, preserving critical tabular data while drastically reducing the exposed context. Moving forward, we aim to enhance disambiguation mechanisms for complex structured data, integrate multi-stage retrieval, and expand SAGE to process richer multimodal documents. - -12 - -## **REFERENCES** - -- [1] Amey Agrawal, Nitin Kedia, Anmol Agarwal, Jayashree Mohan, Nipun Kwatra, Souvik Kundu, Ramachandran Ramjee, and Alexey Tumanov. 2025. On Evaluating Performance of LLM Inference Serving Systems. _arXiv preprint arXiv:2507.09019_ (2025). - -- [2] Guoxuan Chen, Han Shi, Jiawei Li, Yihang Gao, Xiaozhe Ren, Yimeng Chen, Xin Jiang, Zhenguo Li, Weiyang Liu, and Chao Huang. 2024. Sepllm: Accelerate large language models by compressing one segment into one separator. _arXiv preprint arXiv:2412.12094_ (2024). - -- [3] Yilong Chen, Guoxia Wang, Junyuan Shang, Shiyao Cui, Zhenyu Zhang, Tingwen Liu, Shuohuan Wang, Yu Sun, Dianhai Yu, and Hua Wu. 2024. Nacl: A general and effective kv cache eviction framework for llm at inference time. In _Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . 7913–7926. - -- [4] Xin Cheng, Xun Wang, Xingxing Zhang, Tao Ge, Si-Qing Chen, Furu Wei, Huishuai Zhang, and Dongyan Zhao. 2024. xrag: Extreme context compression for retrieval-augmented generation with one token. _Advances in Neural Information Processing Systems_ 37 (2024), 109487–109516. - -- [5] Suyu Ge, Yunan Zhang, Liyuan Liu, Minjia Zhang, Jiawei Han, and Jianfeng Gao. 2023. Model tells you what to discard: Adaptive kv cache compression for llms. _arXiv preprint arXiv:2310.01801_ (2023). - -- [6] Tao Ge, Jing Hu, Lei Wang, Xun Wang, Si-Qing Chen, and Furu Wei. 2023. Incontext autoencoder for context compression in a large language model. _arXiv preprint arXiv:2307.06945_ (2023). - -- [7] Yuyao Ge, Shenghua Liu, Yiwei Wang, Lingrui Mei, Baolong Bi, Xuanshan Zhou, Jiayu Yao, Jiafeng Guo, and Xueqi Cheng. 2025. Focusing by Contrastive Attention: Enhancing VLMs’ Visual Reasoning. _arXiv preprint arXiv:2509.06461_ (2025). - -- [8] Saurabh Goyal, Anamitra Roy Choudhury, Saurabh Raje, Venkatesan Chakaravarthy, Yogish Sabharwal, and Ashish Verma. 2020. Power-bert: Accelerating bert inference via progressive word-vector elimination. In _International Conference on Machine Learning_ . PMLR, 3690–3699. - -- [9] Ruishan Guo, Yibing Liu, Guoxin Ma, Yan Wang, Yueyang Zhang, Long Xia, Kecheng Chen, Zhiyuan Sun, and Daiting Shi. 2026. When Less is More: The LLM Scaling Paradox in Context Compression. _arXiv preprint arXiv:2602.09789_ (2026). - -- [10] Huiqiang Jiang, Qianhui Wu, Chin-Yew Lin, Yuqing Yang, and Lili Qiu. 2023. Llmlingua: Compressing prompts for accelerated inference of large language models. In _Proceedings of the 2023 conference on empirical methods in natural language processing_ . 13358–13376. - -- [11] Huiqiang Jiang, Qianhui Wu, Xufang Luo, Dongsheng Li, Chin-Yew Lin, Yuqing Yang, and Lili Qiu. 2024. Longllmlingua: Accelerating and enhancing llms in long context scenarios via prompt compression. In _Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . 1658–1677. - -- [12] Ziyan Jiang, Xueguang Ma, and Wenhu Chen. 2024. Longrag: Enhancing retrievalaugmented generation with long-context llms. _arXiv preprint arXiv:2406.15319_ (2024). - -- [13] Bowen Jin, Jinsung Yoon, Jiawei Han, and Sercan O Arik. 2024. Long-context llms meet rag: Overcoming challenges for long inputs in rag. _arXiv preprint arXiv:2410.05983_ (2024). - -- [14] Jiajie Jin, Xiaoxi Li, Guanting Dong, Yuyao Zhang, Yutao Zhu, Yongkang Wu, Zhonghua Li, Ye Qi, and Zhicheng Dou. 2025. Hierarchical document refinement for long-context retrieval-augmented generation. In _Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)_ . 3502–3520. - -- [15] Hyukkyu Kang, Injung Kim, and Wook-Shin Han. 2025. TRIAL: Token Relations and Importance Aware Late-interaction for Accurate Text Retrieval. In _Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing_ . 16875–16888. - -- [16] Yannis Katsis, Saneem Chemmengath, Vishwajeet Kumar, Samarth Bharadwaj, Mustafa Canim, Michael Glass, Alfio Gliozzo, Feifei Pan, Jaydeep Sen, Karthik Sankaranarayanan, et al. 2022. Ait-qa: Question answering dataset over complex tables in the airline industry. In _Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Industry Track_ . 305–314. - -- [17] Omar Khattab and Matei Zaharia. 2020. Colbert: Efficient and effective passage search via contextualized late interaction over bert. In _Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval_ . 39–48. - -- [18] Sehoon Kim, Sheng Shen, David Thorsley, Amir Gholami, Woosuk Kwon, Joseph Hassoun, and Kurt Keutzer. 2022. Learned token pruning for transformers. In _Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining_ . 784–794. - -- [19] Xiang Lisa Li, Ari Holtzman, Daniel Fried, Percy Liang, Jason Eisner, Tatsunori B Hashimoto, Luke Zettlemoyer, and Mike Lewis. 2023. Contrastive decoding: Open-ended text generation as optimization. In _Proceedings of the 61st annual_ - -_meeting of the association for computational linguistics (volume 1: Long papers)_ . 12286–12312. - -- [20] Yuhong Li, Yingbing Huang, Bowen Yang, Bharat Venkitesh, Acyr Locatelli, Hanchen Ye, Tianle Cai, Patrick Lewis, and Deming Chen. 2024. Snapkv: Llm knows what you are looking for before generation. _Advances in Neural Information Processing Systems_ 37 (2024), 22947–22970. - -- [21] Zongqian Li, Yinhong Liu, Yixuan Su, and Nigel Collier. 2025. Prompt compression for large language models: A survey. In _Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)_ . 7182–7195. - -- [22] Junghwan Lim, Sungmin Lee, Dongseok Kim, Wai Ting Cheung, Beomgyu Kim, Taehwan Kim, Haesol Lee, Junhyeok Lee, Dongpin Oh, and Eunhwan Park. 2025. Grouped Differential Attention. _arXiv preprint arXiv:2510.06949_ (2025). - -- [23] Yiming Lin, Madelon Hulsebos, Ruiying Ma, Shreya Shankar, Sepanta Zeighami, Aditya G Parameswaran, and Eugene Wu. 2025. Querying templatized document collections with large language models. In _2025 IEEE 41st International Conference on Data Engineering (ICDE)_ . IEEE, 2422–2435. - -- [24] Barys Liskavets, Maxim Ushakov, Shuvendu Roy, Mark Klibanov, Ali Etemad, and Shane K Luke. 2025. Prompt compression with context-aware sentence encoding for fast and improved llm inference. In _Proceedings of the AAAI Conference on Artificial Intelligence_ , Vol. 39. 24595–24604. - -- [25] Chunwei Liu, Matthew Russo, Michael Cafarella, Lei Cao, Peter Baile Chen, Zui Chen, Michael Franklin, Tim Kraska, Samuel Madden, Rana Shahout, et al. [n.d.]. Palimpzest: Optimizing ai-powered analytics with declarative query processing. - -- [26] Di Liu, Meng Chen, Baotong Lu, Huiqiang Jiang, Zhenhua Han, Qianxi Zhang, Qi Chen, Chengruidong Zhang, Bailu Ding, Kai Zhang, et al. 2024. Retrievalattention: Accelerating long-context llm inference via vector retrieval. _arXiv preprint arXiv:2409.10516_ (2024). - -- [27] Jiaheng Liu, Dawei Zhu, Zhiqi Bai, Yancheng He, Huanxuan Liao, Haoran Que, Zekun Wang, Chenchen Zhang, Ge Zhang, Jiebin Zhang, et al. [n.d.]. A comprehensive survey on long context language modeling. 2025. _URL https://api. semanticscholar. org/CorpusID_ 277271533 ([n. d.]). - -- [28] Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. Lost in the middle: How language models use long contexts. _Transactions of the Association for Computational Linguistics_ 12 (2024), 157–173. - -- [29] Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, et al. 2024. Mmlongbench-doc: Benchmarking long-context document understanding with visualizations. _Advances in Neural Information Processing Systems_ 37 (2024), 95963–96010. - -- [30] mteb. [n.d.]. MTEB leaderboard - a hugging face space by mteb. https:// huggingface.co/spaces/mteb/leaderboard - -- [31] Jesse Mu, Xiang Li, and Noah Goodman. 2024. Learning to compress prompts with gist tokens. _Advances in Neural Information Processing Systems_ 36 (2024). - -- [32] Niklas Muennighoff, Nouamane Tazi, Loïc Magne, and Nils Reimers. 2023. Mteb: Massive text embedding benchmark. In _Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics_ . 2014–2037. - -- [33] Krista Opsahl-Ong, Michael J Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, and Omar Khattab. 2024. Optimizing instructions and demonstrations for multi-stage language model programs. In _Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing_ . 9340–9366. - -- [34] Richard Yuanzhe Pang, Alicia Parrish, Nitish Joshi, Nikita Nangia, Jason Phang, Angelica Chen, Vishakh Padmakumar, Johnny Ma, Jana Thompson, He He, and Samuel Bowman. [n.d.]. QuALITY Leaderboard — nyu-mll.github.io. https://nyumll.github.io/quality/. [Accessed 01-03-2026]. - -- [35] Richard Yuanzhe Pang, Alicia Parrish, Nitish Joshi, Nikita Nangia, Jason Phang, Angelica Chen, Vishakh Padmakumar, Johnny Ma, Jana Thompson, He He, and Samuel Bowman. 2022. QuALITY: Question Answering with Long Input Texts, Yes!. In _Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies_ . Association for Computational Linguistics, Seattle, United States, 5336–5358. https://aclanthology.org/2022.naacl-main.391 - -- [36] Badri Patro and Vinay P Namboodiri. 2018. Differential attention for visual question answering. In _Proceedings of the IEEE conference on computer vision and pattern recognition_ . 7680–7688. - -- [37] Rui Qian, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Shuangrui Ding, Dahua Lin, and Jiaqi Wang. 2024. Streaming long video understanding with large language models. _Advances in Neural Information Processing Systems_ 37 (2024), 119336– 119360. - -- [38] Keshav Santhanam, Omar Khattab, Jon Saad-Falcon, Christopher Potts, and Matei Zaharia. 2022. Colbertv2: Effective and efficient retrieval via lightweight late interaction. In _Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies_ . 3715–3734. - -- [39] Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D Manning. 2024. Raptor: Recursive abstractive processing for tree-organized retrieval. In _The Twelfth International Conference on Learning Representations_ . - -13 - -- [40] Kele Shao, TAO Keda, Kejia Zhang, Sicheng Feng, Mu Cai, Yuzhang Shang, Haoxuan You, Can Qin, Yang Sui, and Huan Wang. [n.d.]. A Survey of Token Compression for Efficient Multimodal Large Language Models. _Transactions on Machine Learning Research_ ([n. d.]). - -- [41] Hanrui Wang, Zhekai Zhang, and Song Han. 2021. Spatten: Efficient sparse attention architecture with cascade token and head pruning. In _2021 IEEE international symposium on high-performance computer architecture (HPCA)_ . IEEE, 97–110. - -- [42] Yuetian Weng, Mingfei Han, Haoyu He, Xiaojun Chang, and Bohan Zhuang. 2024. Longvlm: Efficient long video understanding via large language models. In _European Conference on Computer Vision_ . Springer, 453–470. - -- [43] David Wingate, Mohammad Shoeybi, and Taylor Sorensen. 2022. Prompt compression and contrastive conditioning for controllability and toxicity reduction in language models. In _Findings of the Association for Computational Linguistics: EMNLP 2022_ . 5621–5634. - -- [44] Wenhao Wu, Yizhong Wang, Guangxuan Xiao, Hao Peng, and Yao Fu. 2024. Retrieval head mechanistically explains long-context factuality. _arXiv preprint arXiv:2404.15574_ (2024). - -- [45] Jingbo Yang, Bairu Hou, Wei Wei, Yujia Bao, and Shiyu Chang. 2025. Kvlink: Accelerating large language models via efficient kv cache reuse. _arXiv preprint arXiv:2502.16002_ (2025). - -- [46] Deming Ye, Yankai Lin, Yufei Huang, and Maosong Sun. 2021. Tr-bert: Dynamic token reduction for accelerating bert inference. In _Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies_ . 5798–5809. - -- [47] Tianzhu Ye, Li Dong, Yuqing Xia, Yutao Sun, Yi Zhu, Gao Huang, and Furu Wei. 2024. Differential transformer. _arXiv preprint arXiv:2410.05258_ (2024). - -- [48] Tan Yu, Anbang Xu, and Rama Akkiraju. 2024. In defense of rag in the era of long-context language models. _arXiv preprint arXiv:2409.01666_ (2024). - -- [49] Peiyuan Zhang, Kaichen Zhang, Bo Li, Guangtao Zeng, Jingkang Yang, Yuanhan Zhang, Ziyue Wang, Haoran Tan, Chunyuan Li, and Ziwei Liu. 2024. Long context transfer from language to vision. _arXiv preprint arXiv:2406.16852_ (2024). - -- [50] Zhenyu Zhang, Runjin Chen, Shiwei Liu, Zhewei Yao, Olatunji Ruwase, Beidi Chen, Xiaoxia Wu, and Zhangyang Wang. 2024. Found in the middle: How language models use long contexts better via plug-and-play positional encoding. _Advances in Neural Information Processing Systems_ 37 (2024), 60755–60775. - -- [51] Zhenyu Zhang, Ying Sheng, Tianyi Zhou, Tianlong Chen, Lianmin Zheng, Ruisi Cai, Zhao Song, Yuandong Tian, Christopher Ré, Clark Barrett, et al. 2024. H2o: Heavy-hitter oracle for efficient generative inference of large language models. _Advances in Neural Information Processing Systems_ 36 (2024). - -- [52] Xiabin Zhou, Wenbin Wang, Minyan Zeng, Jiaxian Guo, Xuebo Liu, Li Shen, Min Zhang, and Liang Ding. 2024. DynamicKV: Task-aware adaptive KV cache compression for long context LLMs. _arXiv preprint arXiv:2412.14838_ (2024). - -14 +11 diff --git a/q188/random_k2/question.json b/q188/random_k2/question.json index b195b3ba4f1eced7f3eadac4f2c41d1fae86d236..f9c0507c78785750bd282dcfac68ca71338299c1 100644 --- a/q188/random_k2/question.json +++ b/q188/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "2305.14160v4.pdf", - "web_ab2f104fb9b1c2fe.pdf", - "web_6a55c713826db217.pdf" + "web_99fd395a0c29c14c.pdf", + "web_93a0de1d8da7a0cf.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q188/random_k2/random_1.md b/q188/random_k2/random_1.md index 510efcdbde69ac509e9a3fa23225deb77871706d..178a074cfe698dd3ddc22a4fffd3c36000aa45a9 100644 --- a/q188/random_k2/random_1.md +++ b/q188/random_k2/random_1.md @@ -1,1380 +1,1420 @@ -ARYAMAN -FINANCIAL SERVICES LIMITED +Encyclopedic VQA: +Visualquestionsaboutdetailedpropertiesoffine-grainedcategories +ThomasMensink†,∗ JasperUijlings†,∗ LluisCastrejon∗ ArushiGoel‡ +mensink@google.com jrru@google.com lluisc@google.com goel.arushi@gmail.com +FelipeCadar‡ HowardZhou∗ FeiSha∗ Andre´ Araujo∗ +cadar@dcc.ufmg.br howardzhou@google.com fsha@google.com andrearaujo@google.com +VittorioFerrari∗ +vittoferrari@google.com +Abstract +We propose Encyclopedic-VQA, a large scale visual +question answering (VQA) dataset featuring visual ques- +tions about detailed properties of fine-grained categories +and instances. It contains 221k unique question+answer +pairs each matched with (up to) 5 images, resulting +in a total of 1M VQA samples. Moreover, our dataset +comes with a controlled knowledge base derived from +Wikipedia, marking the evidence to support each answer. Figure1:Oneofthetwoanswersaboveiswrong,doyouknow +Empirically, we show that our dataset poses a hard whichone? Encyclopedicquestionsaboutdetailedpropertiesof +challenge for large vision+language models as they per- fine-grainedentitiesaredifficult.Notonlyforhumans,butalsofor +form poorly on our dataset: PaLI [14] is state-of-the-art largeVLMs.PaLI[14]failstoanswerbothquestionscorrectly. +on OK-VQA [37], yet it only achieves 13.0% accuracy +on our dataset. Moreover, we experimentally show +Answeringthesequestionscorrectlyrequiresknowledgeof +that progress on answering our encyclopedic questions +detailed properties (i.e. symbol of which city, year of au- +can be achieved by augmenting large models with a +tomation) of fine-grained categories (‘Pinus Pinea’) or in- +mechanism that retrieves relevant information from the +stances (‘Point Reyes Lighthouse’). We hypothesize that +knowledge base. An oracle experiment with perfect re- +this type of encyclopedic knowledge is hard for VLMs to +trievalachieves87.0%accuracyonthesingle-hopportion +properlyencodeinitsmodelparametersbecausesuchlong- +of our dataset, and an automatic retrieval-augmented +tail information occurs rarely in its training data (i.e. the +prototypeyields48.8%.Webelievethatourdatasetenables +web). Additionally,VLMsproducesuchincorrectanswers +future research on retrieval-augmented vision+language +generally with high confidence, while these answers are +models. It is available at https://github.com/ +hardforuserstoverifysincethemodeldoesnotprovideany +google-research/google-research/tree/ +explanations. Theanswerto Fig.1(left) iscorrect; (right) +master/encyclopedic_vqa. +shouldbe1975. PaLI[14]predictsbothincorrectly. +Bothproblemscanbeaddressedbyretrieval-augmented +1.Introduction models, which base their predictions on knowledge re- +trieved from a database. These models recently gained +Recently, large Vision+Language models (VLMs) have popularity in NLP (e.g. [9, 23, 30, 33]), and some early +demonstrated impressive performance on Visual Question multi-modal models also exist [22, 26, 36, 52]. Retrieval- +Answering (VQA) benchmarks [7, 14, 24, 57]. However, augmentedmodelsarewell-suitedforencyclopedicknowl- +Fig.1showstwotypicalexampleswheresuchmodelsfail. edge since retrieving the correct entry from a knowledge +†Equalcontribution.‡WorkdoneduringinternshipatGoogle. base greatly facilitates constructing the right answer. Fur- +∗GoogleResearch thermore, the retrieved piece of knowledge provides attri- +3202 +luJ +42 +]VC.sc[ +2v42290.6032:viXra -T= ARYAM AN FINANCIAL SERVICES LIMITED = -| BOARD OF DIRECTORS | | P.V.R Murthy | Director | -| ------------------- | --- | ------------------------ | -------- | -| | | Virender Ganda - | Director | -| | | Mohan Datari - | Director | -| | | M.V. Ramana - | Director | -| | | Suhas Ganpule - | Director | -| REGISTERED OFFICE | | : 401, Chiranjiv Towers | | -43, Nehru Place, -New Delhi 110019 -| CORPORATE | OFFICE | : 208, Maker Chamber V, | | -| ---------- | ------- | ------------------------ | --- | -2nd -Floor, Nariman Point, -Mumbai-400021. -| AUDITORS | | : M/s. Thakur Vaidynath Aiyar & Co. | | -| --------- | --- | ------------------------------------ | --- | -Chartered Accountants -212, Deendayal Upadhyay Marg. -New Delhi. -| BANKERS | | : State Bank Of India | | -| -------- | --- | ---------------------- | --- | -The Vysya Bank Limited -State Bank of Patiala -Indusind Bank Ltd. -| Share Transfer Agent | | : Adroit Corporate Services Pvt. Ltd. | | -| --------------------- | --- | -------------------------------------- | --- | -19, Jaferbhoy Industrial Estate. lsl Floor. -Makwana Road, Marol Naka, -Andheri (W), Mumbai - 400059. +Question Type +skramdnaL +Templated Automatic Automatic - multi-answer 2-Hop +Q: Who founded this monastery? Q: When was the first permanent Q: What fish can be found in this lake? Q: What amusement park is located in +settlement made at this valley? the city where this square is located? +A: Prince Constantin Brâncoveanu A: 1864 A: trout, lake char A: Tivoli Gardens +C: Horezu monastery C: Clover valley C: Úlfljótsvatn C: Rådhuspladsen, Copenhagen +dlroW +larutaN +Q: How old does this reptile become? Q: How many feet tall does this tree Q: Where is this bird found? Q: How many national park service maintained +grow to? sites are in the state where this plant grows? +A: 40 years A: 7 to 13 A: Colombia, Venezuela, Ecuador A: 24 +C: Gila monster C: Acacia paradoxa C: Boissonneaua C: Chorizanthe rigida +Figure2: ExampleVQAannotationsfordifferentquestiontypes. EachexampleconsistsofanimageI,aquestionQandtheanswer +A.WealsoshowthecategoryCofthesubjectofthequestion.Asattribution,weprovideasectionwithintheWikipediapageofCwhich +supportstheanswer.OurEncyclopedic-VQAdatasethasatotalof1M(I,Q,A)triplets. +OK-VQA A-OKVQA FVQA KVQA S3VQA +Q: What sort of vehicle uses Q: What does the man who Q: What furniture in this image Q: Who is to the left of Hillary Q: What is the process by +this item? sits have trouble doing? can I lie on? Clinton? which this insect develops? +A: firetruck A: walking A: sofa A: Aamir Khan A: complete metamorphosis +Q: When was this piece of Q: What could block the Q: Which transportation way in Q: Who among the people in the Q: Where was this weapon used +sporting equipment invented? washer’s door? this image is cheaper than taxi? image is the eldest? to attack the French camp? +A: 1926 A: stove A: bus A: Person in the left A: Crecy +Figure3: TypicalexamplesofVQAdatasetswhichrequireknowledgebeyondtheimage. +bution to the answer by design. This increases model in- quiringknowledgeofdetailedproperties,butmostlyabout +terpretabilityandthereforehumantrust,hasapplicationsto coarserbasic-levelcategories[45](e.g.Fig.3bottom-left). +fairnessandhelpsdiagnosinganddebuggingmodelerrors. Thereareafewotherdatasetswhichtargetdetailedproper- +To drive progress on handling encyclopedic knowledge ties[27,47]. Butboth[27,47]lackattributionannotations, +and attribution in VQA we need a suitable dataset. The [27]isverysmall(Tab.1right-mostcolumn),and[47]fo- +popular OK-VQA [37] and A-OKVQA [46] datasets fo- cusesonlyoncelebrities(Fig.3fourthcolumn). Henceno +cus on questions requiring knowledge outside the query currentVQAdatasetisfullysatisfactory. +image. However, the majority of questions require com- In this paper we introduce the Encyclopedic-VQA +monsense knowledge (e.g. Fig. 3 top-row, two left-most dataset, which offers several attractive features (Fig. 2, +examples), a type of knowledge for which current VLMs Tab.1). Ourdatasetasksquestionsaboutfine-grainedcat- +are powerful in both theory and practice [7, 14, 24, 57]. egories from iNaturalist 2021 [51] (Fig. 2 bottom row) +These datasets [37, 46] also include some questions re- andinstancesfromtheGoogleLandmarksDatasetv2[56] -r -7= ARYAMANFINANCIALSERVICES LIMITED -NOTICE -"RESOLVED THAT in accordance with the -| | | | | provision of section | | 257 and all other | | -| --- | --- | --- | --- | --------------------- | --- | ------------------ | --- | -NOTICE is hereby given that the 11™ ANNUAL -| | | | | applicable provisions, if any, of | | | the | -| --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | -GENERAL MEETING of the Members of the -ARYAMAN FINANCIAL SERVICES companies Act 1956 or any statutory -modifications or reenactment thereof, Shri -| LIMITED will be held on | | Thursday, 29th | | | | | | -| ------------------------ | --- | -------------- | --- | --- | --- | --- | --- | -Suhas S Ganpule, who was appointed as -| day December 2005 at 10.00 a.m. at its | | | | | | | | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Additional Director in pursuant to section -registered office 401, Chiranjiv Towers, 43, -Nehru Place, New Delhi 110 019 to transact 260 of companies Act, 1956 be and hereby -appointed as Director of the company -the following business: -subject to retirement by rotation under the -ORDINARY BUSINESS provisions of Article of Association of the -company". -1. To receive, consider and adopt the Balance -Sheet as at June 30, 2005 and the Profit and 6. To consider and if thought fit to pass with -Loss Account for the year ended on that date or without any modification, the -| | | | | following | resolution | as Ordinary | | -| --- | --- | --- | --- | ---------- | ----------- | ------------ | --- | -and the Reports of Directors and Auditors -Resolution- -thereon. -"RESOLVED THAT in accordance with the -2. To appoint a director in place of Shri Mohan -| | | | | provision of section 257 and | | all other | | -| --- | --- | --- | --- | ----------------------------- | --- | --------- | --- | -Datari, who retires by rotation, and being -| | | | | applicable | provisions, | if any, of | the | -| ------------ | ---------------- | ---- | --- | ----------- | ------------------------- | ----------- | --- | -| eligible, | offers himself | for | re- | | | | | -| appointment. | | | | companies | Act 1956 or any statutory | | | -modifications or reenactment thereof, Shri -Manmohan Baliga, who was appointed as -| 3. To appoint | Auditors and | to fix | their | | | | | -| --------------- | ------------- | -------- | ----- | ------------------------------------------ | --- | --- | --- | -| remuneration. | | | | Additional Director in pursuant to section | | | | -260 of companies Act, 1956 be and hereby -| | | | | appointed | as Director of the company | | | -| --- | --- | --- | --- | ---------- | -------------------------- | --- | --- | -SPECIAL BUSINESS -subject to retirement by rotation under the -4. To consider and if thought fit to pass with provisions of Article of Association of the -company". -or without any modification, the following -resolution as Ordinary Resolution- -7. To consider and if thought fit to pass with -" RESOLVED THAT in accordance with the or without any modification, the -| | | | | following | resolution | as Ordinary | | -| ---------- | ----------- | -------- | ---------- | ---------- | ----------- | ------------ | --- | -| provision | of section | 257 and | all other | | | | | -Resolution- -| applicable | provisions, | if any, | of the | | | | | -| ----------- | ------------ | ----------------- | ------- | --- | --- | --- | --- | -| companies | Act 1956 | or any statutory | | | | | | -modifications or reenactment thereof, Shri "RESOLVED THAT in suppersession of the -earlier resolutions passed under Section -Munnangi Venkat Ramanna, who was -293 (l)(d) of the Companies Act, 1956, the -| appointed | as Additional | Director | in | | | | | -| ---------- | --------------- | --------- | --- | --- | --- | --- | --- | -pursuant to section 260 of the Companies Board of Directors of the Company be and -is hereby authorised to borrow from time -Act, 1956 be and is hereby appointed as -to time all such sums of money as they -| Director | of the | company subject | to | | | | | -| --------- | -------- | ----------------- | --- | --- | --- | --- | --- | -may deem necessary for the purpose of -retirement by rotation under the provisions -of Article of Association of the company". business of the Company -| | | | | notwithstanding | that moneys to be | | | -| --- | --- | --- | --- | ---------------- | ----------------- | --------------- | --- | -| | | | | borrowed | together | with the moneys | | -5. To consider and if thought fit to pass with -or without any modification, the following already borrowed by the Company (apart -resolution as Ordinary Resolution- from temporary loans (including working +OK-VQA[37] A-OKVQA[46] FVQA[53] S3VQA[27] KVQA[47] Encyclopedic-VQA(ours) +Trulymultimodal +| | | | | ++ | | ++ | ++ | ++ | | ++ | | ++ | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| Encyclopedic | | | | ± | | - | - | + | | ++ | | ++ | | +| (cid:44)→Detailedproperties | | | | + | | ± | - | ++ | | ++ | | ++ | | +| (cid:44)→fine-grainedcategories/instances | | | | - | | - | - | + | | ++ | | ++ | | +| Controlledknowledgebase | | | | -- | | -- | + | ± | | ± | | ++ | | +| (cid:44)→answersupportedbyKB | | | | -- | | -- | ++ | ++ | | ++ | | ++ | | +(cid:44)→KBprovided +| | | | | -- | | -- | ++ | -- | | ++ | | ++ | | +| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| (cid:44)→KBfree-form | | | | -- | | -- | -- | -- | | -- | | ++ | | +| (cid:44)→attribution | | | | -- | | ± | ++ | -- | | -- | | ++ | | +| Scale | | | | ± | | + | -- | -- | | ++ | | ++ | | +| Two-hop | | | | -- | | -- | -- | -- | | ++ | | + | | +Subject various various various various celebrities fine-grainedspecies,landmarks +| NumberoftextquestionsQ | | | | 14k | | 25k | 6k | 7k | | 183k | | 221k | | +| -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ------ | --- | +| NumberofimagesI | | | | 14k | | 24k | 2k | 7k | | 25k | | 514k | | +| NumberofuniqueVQAtriplets(I,Q,A) | | | | 14k | | 25k | 6k | 7k | | 183k | | 1,036k | | +Table1: ComparisonofrecentVQAdatasets.WecomparerecentVQAdatasetswithourproposeddatasetonVQAdesignprinciples. +(Fig. 2 top row). We construct questions about detailed that specific part of the knowledge base which supports it. +properties based on Wikipedia (e.g. founder of building, Hencetheknowledgebaseisanintegralpartofourdataset, +maximumageofanimal). Asaconsequence,allquestions whichenablesmeasuringwhetheramodelanswersaques- +in our dataset are about encyclopedic knowledge. Further- tion correctly for the right reason. For generality we want +more, we provide a controlled knowledge base suited to thisknowledgebasetobefree-formtextandtocontainim- +answerthesequestions: 2MWikipediapagesconsistingof ages. (4) Scale. The dataset should be large. Dataset size +free-formtextandimages[48]. Wealsomarkground-truth hasalwaysmatteredandthisisevenmoretruewiththein- +attributionforeachansweratthegranularitylevelofasec- creasinglylargeVLMmodels(e.g.[6,14,28,60,39]). (5) +tion within a Wikipedia page. Importantly, our dataset is Two-hop. Aportionofourquestionsshouldrequireknowl- +collected at scale: we have 221k unique question+answer edge from multiple different documents from the knowl- +pairseachmatchedwitharound5images,resultinginato- edgebase. Includingsuchcomplextwo-hopquestions[58] +talof1Mexamples.Thismakesourdatasetthelargestofits leavessubstantialheadroomforfuturemodeldevelopment. +| kind. Finally, | manyof | ourquestions | | arecomplex | | two-hop | | | | | | | | +| -------------- | ------ | ------------ | --- | ---------- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +3.RelatedWork +questions[58],whichrequiremultipledifferentdocuments +fromtheknowledgebasetosolve(Fig.2right). +VisualQuestionAnswering(VQA).DAQUAR[35],FM- +| We validate | the usefulness | | of our | dataset | through | sev- | | | | | | | | +| ----------- | -------------- | --- | ------ | ------- | ------- | ---- | --- | --- | --- | --- | --- | --- | --- | +IQA[20],VisualMadlibs[59],VQAv1[8]andVQAv2[21] +| eralexperiments. | Inparticular,wedemonstratethatalarge | | | | | | | | | | | | | +| ---------------- | ------------------------------------ | ------ | ---------------- | ----------- | ------ | ---------- | --------- | -------- | ----------- | -------------- | --- | ---------- | ------ | +| | | | | | | | are early | VQA | datasets. | These datasets | | mostly ask | visual | +| VLM (PaLI | [14]) which | yields | state-of-the-art | | | results on | | | | | | | | +| | | | | | | | questions | that can | be answered | based | on | the query | image | +| OK-VQA | [37] performs | poorly | on | our dataset | (13.0% | ac- | | | | | | | | +andgenericknowledgesuchasbasic-levelcategoryrecog- +| curacy). | Next we demonstrate | | through | | an oracle | experi- | | | | | | | | +| -------- | ------------------- | --- | ------- | --- | --------- | ------- | --- | --- | --- | --- | --- | --- | --- | +nition(e.g.‘cat’),countingitems,colors,etc. +mentthatretrieval-augmentedmodelscanyield87.0%ac- +| | | | | | | | Knowledge-based | | VQA. | Tab. | 1 and | Fig. 3 compare | | +| ------- | --------------- | --- | ------ | ----- | ------ | --------- | --------------- | ----- | ------------ | ---- | ------- | -------------- | --- | +| curacy. | Finally, we use | an | online | image | search | engine to | | | | | | | | +| | | | | | | | datasets | which | are designed | to | require | knowledge | not | +buildanautomaticretrieval-augmentedprototype,reaching +| | | | | | | | presentintheimage[27,37,46,47,53]. | | | | | Inparticular,OK- | | +| ---------------- | ----------------------- | -------- | ----- | -------- | ----------- | --------- | ---------------------------------- | ---------- | --------------------------------- | ---- | -------- | ---------------- | --- | +| 48.8% accuracy. | We | conclude | that | (1) our | dataset | poses a | | | | | | | | +| | | | | | | | VQA [37] | and | A-OKVQA | [46] | mostly | require common- | | +| strong challenge | in VQA | and | is in | fact too | hard | for stan- | | | | | | | | +| | | | | | | | senseknowledge. | | Inaddition,somequestions(18%)inA- | | | | | +| dard VLMs; | (2) retrieval-augmented | | | VLMs | demonstrate | a | | | | | | | | +| | | | | | | | OKVQA | do require | knowledge | of | detailed | properties, | but | +strongpotentialforaddressingencyclopedicknowledge;(3) +| | | | | | | | about basic-level | | categories. | Finally, | 3% | of the questions | | +| --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ----------- | -------- | --- | ---------------- | --- | +ourresultsleavesignificantheadroomforfurtherresearchto +| | | | | | | | requireknowledgeaboutphysics. | | | | Inthispaperwecreatea | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------- | --- | --- | --- | -------------------- | --- | --- | +improveretrieval-augmentedVLMs. +datasetwithquestionsexclusivelyaboutdetailedproperties +| | | | | | | | of fine-grained | | categories | and instances | | (Fig. 2). | We be- | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | ---------- | ------------- | --- | --------- | ------ | +2.Designprinciplesforourdataset +| | | | | | | | lieve that | such encyclopedic | | questions | truly | require | access | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------------- | --- | --------- | ----- | ------- | ------ | +We create our VQA dataset with the following desired toaknowledgebasetobeanswered. Infact,wereleasethe +properties in mind: (1) Truly Multimodal. The questions knowledgebasealongwithourdataset,whereasnoexplicit +should not be answerable without the image or the textual knowledgebasewasinvolvedinthecreationof[37,46]. +question [21]. (2) Encyclopedic. The questions should SomeexistingVQAdatasetsaresupportedbyaknowl- +be about detailed properties of fine-grained categories or edgebase[27,47,53]. FVQA[53]isaboutcommonsense +instances; a type of questions which are problematic for knowledge. KVQA [47] asks detailed properties about +vanilla VLMs [7, 14, 24, 57]. (3) Controlled knowledge celebrities. Both[47,53]arebasedonastructuredknowl- +base. Each answer in our dataset should be attributable to edge base (RDF triplets). Instead, our knowledge base is -7= ARYAMANFINANCIALSERVICES LIMITED -capital facilities) obtained from the remuneration, costs, charges expenses and -Company's bankers in the ordinary course all other monies payable by the Company -of business) (hereinafter referred to as the to the concerned Lenders, and/or Agents -"Borrowings") shall exceed the aggregate and Trustees for debentures provided that -of the paid-up capital of the Company and the amount of loans (other than temporary -its free reserves, that is to say reserves not loans (including working capital facilities) -set apart for any specific purpose provided obtained from the Company's bankers in -that the total amount of the Borrowings the ordinary course of buinsess) shall not -by the Board of Directors shall not exceed exceed Rs. 50 Crores." -Rs. 25 crores (Rupees Twenty Five Crores -Only)." "RESOLVED FURTHER THAT the Board of -Directors of the Company be and is hereby -8. To consider and if thought fit to pass with authorised to finalise with any of the -or without any modification, the following Lenders, or other persons, jointly or -resolution as Ordinary Resolution- severally, the documents for creating -aforesaid mortgage/charge and to do all -"RESOLVED THAT in supersession of the such acts, deeds, matters and things as may -earlier resolutions passed under section be considered necessary, desirable or -293(l)(a) of the Companies Act, 1956 (the expedient for implementing the Resolution -"Act"), consent of the Company be and is and to resolve any question, difficulty or -hereby accorded in terms of section doubt which may arise in relation thereto -293(l)(a) and other applicable provisions, or otherwise considered by the Board of -if any, of the Act, to mortgaging and/or Directors to be in the best interest of the -charging by the Board of Directors of the Company." -Company of all or any of the immovable -and/or movable properties of the By Order of the Board -Company, wheresoever situated, both -present and future, or the whole or Place: Mumbai -substantially the whole of the undertaking Date: December 01, 2005 -or undertakings of the Company on such For Aryaman Financial Services Limited -terms, in such form and in such manner -as the Board of Directors may think fit, -together with power to take over the Prasanna Chandwaskar -management of the business and concern Company Secretary -of the Company in certain events in favour -of Banks, Financial Institutions, other A MEMBER ENTITLED TO ATTEND AND -investing agencies and trustees for the VOTE IS ENTITLED TO APPOINT ONE -holders of Debentures / other instruments, OR MORE PROXIES TO ATTEND AND -for securing, inter alia, any loans (both VOTE INSTEAD OF HIMSELF AND A -rupee loans as well as foreign currency PROXY NEED NOT BE A MEMBER OF -loans) and/or advances already obtained THE COMPANY. Proxies, in order to be -or debts already incurred or that may effective, should be duly stamped, -hereafter be obtained or incurred from any completed, signed and deposited at the -of the lenders and/or to secure any Registered Office of the Company not less -debentures issued/that may be issued, than 48 hours before the Meeting. -and/or any financial obligations/ -commitment hereinafter collectively The Register of Members and Share Transfer -referred to as the "Loans") and all interest, Books of the Company will remain -compound / additional interest, closed from Monday, December 26, 2005 -commitment charges, Trustees' to Thursday, December 29, 2005 (both +| readilyavailablefree-formWikipediatext,whichhasmore | | | | | | | | 4.Dataset | | | | | | | +| --------------------------------------------------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | +| information. | Furthermore,KVQAisexclusivelyaboutpeo- | | | | | | | | | | | | | | +Wenowdetailtheconstructionofourdatasetwhilefol- +| ple, which | requires | specialized | | methods | based | on | face de- | | | | | | | | +| ---------- | ---------------- | ----------- | ------------ | ------- | ----- | ------- | -------- | ----------- | ------------ | ---------- | -------- | --------- | ------------ | ------- | +| | | | | | | | | lowing the | design | principles | | (Sec. 2). | To achieve | scale | +| tection | and recognition. | | In contrast, | | our | dataset | offers a | | | | | | | | +| | | | | | | | | we automate | construction | | whenever | | possible. We | use hu- | +broaderrangeoftopics,includinganimals,plantsandland- +manannotatorstoensurequalityandtoprovideinformation +| marks. | S3VQA | [27] | is the most | related | work | to | ours. It | | | | | | | | +| ------ | ----- | ---- | ----------- | ------- | ---- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | +| | | | | | | | | whichwecouldnotgetautomatically. | | | | | Wesimplifyhuman | | +buildsoncategoriesofOpenImages[32],somefine-grained +annotationtasksasmuchaspossible,whichincreasesboth +| and others | more | basic-level. | | They automatically | | generate | | | | | | | | | +| ---------- | ---- | ------------ | --- | ------------------ | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +thequalityandefficiencyoftheirwork. +| questions | from | Wikipedia | articles, | | which | by construction | | | | | | | | | +| --------- | ---- | --------- | --------- | --- | ----- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +WedefineaunitforourVQAtaskasatriplet(I,Q,A). +| are about | detailed | properties. | | Then | they let | experts | manu- | | | | | | | | +| --------------------------------------- | --------- | ----------- | -------- | ---- | ---------- | ------------ | ----- | -------------------------------------- | --- | --- | --- | ------------------------ | --------- | --- | +| | | | | | | | | Themulti-modalquestionfeaturesanimageI | | | | | andaccom- | | +| allyselectandrephraserelevantquestions. | | | | | | Ourpaperalso | | | | | | | | | +| | | | | | | | | panyingtextualquestionQ. | | | | ThesubjectofthequestionQ | | | +| automatically | generates | | QA pairs | from | Wikipedia. | | How- | | | | | | | | +appearsinI +| | | | | | | | | | andisacoreconceptofourwork. | | | | Wereferto | | +| --------------------------------- | --- | --- | --- | --- | --------------- | --- | --- | --- | --------------------------- | --- | --- | --- | --------- | --- | +| ever,wegobeyond[27]inseveralways: | | | | | (1)ourdatasetis | | | | | | | | | | +thecategoryofthesubjectasC(e.g.‘HorezuMonastery’or +muchlarger(7kvs1MVQAtriplets),(2)weincludemore +| | | | | | | | | ‘GilaMonster’inFig. | | | 2-left). | TheanswerAispurelytex- | | | +| --------- | --------- | ---------- | ------ | ------ | ------- | ---- | -------- | --------------------------------------------------- | --- | ----- | -------- | ---------------------- | ---------- | --------- | +| complex | multi-hop | questions, | | (3) we | release | the | knowl- | | | | | | | | +| | | | | | | | | tual. Ourdatasetalsorecordstheevidenceforeachanswer | | | | | | | +| edge base | along | with | the QA | pairs. | This | also | includes | | | | | | | | +| | | | | | | | | in the knowledge | | base, | as a | section | within the | Wikipedia | +ground-truthanswerattributionintheformofthesupport- +| | | | | | | | | pageforC | wheretheanswerisfound. | | | | Incontrasttomany | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ---------------------- | --- | --- | --- | ---------------- | --- | +ingWikipediasection. +| | | | | | | | | existing | VQA | datasets, | we | record only | a single | answer, | +| --- | -------- | ------- | ---- | ------------- | --- | ------------ | --- | -------- | --- | --------- | --- | ----------- | -------- | ------- | +| The | InfoSeek | dataset | [15] | is concurrent | | to our work. | It | | | | | | | | +sinceitisunambiguousgiventheknowledgebase. +| also targets | detailed | properties | | of fine-grained | | categories, | | | | | | | | | +| ------------ | ----------- | ------------ | --- | --------------- | ------- | ----------- | --- | ------- | --- | ------- | ----- | ------------ | --- | ---------- | +| | | | | | | | | To make | our | dataset | truly | multi-modal, | we | always re- | +| but does | not include | multi-answer | | and | two-hop | questions, | | | | | | | | | +fertothesubjectofthequestionbyitssupercategory(e.g. +andwascollectedusingadifferentannotationprotocol. +| | | | | | | | | ‘this monastery’, | | ‘this | reptile’, | as opposed | to ‘the | Horezu | +| ---------------------------- | --- | --- | --- | ------------------ | --- | --- | --- | ----------------- | ---- | ----- | --------- | ---------- | ------------ | ------ | +| Automaticquestiongeneration. | | | | AfewVQAdatasetsare | | | | | | | | | | | +| | | | | | | | | Monastery’, | ‘the | Gila | monster’, | etc.). | This ensures | visual | +generated from image captions [44, 13]. While this en- recognitionisrequiredtosolveourtask,astherewouldbe +ablesmuchlargerscale,itresultsinsimplevisualquestions. +manypotentialanswerstothetextualquestionalone(e.g.a +| We build | a portion | of | our dataset | using | a | similar | pipeline | | | | | | | | +| -------- | --------- | --- | ----------- | ----- | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | --- | +differentanswerforeachmonasteryintheworld). +| as [13], | but apply | it to | Wikipedia | pages | instead | to | obtain | | | | | | | | +| -------- | --------- | ----- | --------- | ----- | ------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +4.1.SupportingDatasets +| high-quality | questions | | on detailed | properties. | | As | another | | | | | | | | +| ------------ | --------- | --- | ----------- | ----------- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +difference,weverifythemwithhumanannotators. We build on two of the largest existing datasets with +| Multi-Hop | Reasoning | | in NLP. | The | NLP | community | has | | | | | | | | +| --------- | --------- | --- | ------- | --- | --- | --------- | --- | ----------------------- | --- | --- | ------------------------------ | --- | --- | --- | +| | | | | | | | | fine-grainedcategories: | | | iNaturalist2021(iNat21)[51]and | | | | +many text-only QA datasets (e.g. [19, 29, 31, 38, 42, 50, Google Landmarks Dataset V2 (GLDv2) [56]. iNat21 is +55, 58]). Notably, HotpotQA [58] introduces the concept afine-grainedvisualclassificationdatasetcontaining2.7M +| of‘bridgeentity’whichlinkstworelatedentities. | | | | | | Theyuse | | | | | | | | | +| --------------------------------------------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +imagesdepicting10,000species,groupedinto11supercat- +thistoshowanannotatortworelatedWikipediaparagraphs egories: plants, insects, birds, mammals, etc. GLDv2 is +| from different | pages | and | ask | to create | questions | which | re- | | | | | | | | +| -------------- | ----- | --- | --- | --------- | --------- | ----- | --- | ---------- | ----------- | --- | ------- | ---- | --------- | --------- | +| | | | | | | | | a landmark | recognition | | dataset | with | 4M images | depicting | +quires knowledge from both paragraphs. We use ‘bridge 200k landmarks. Each landmark is sourced from Wiki- +entities’toautomaticallychaintwoquestionstogetherinto media Commons [5], enabling us to mine the existing +acompoundtwo-hopquestion. +category-hierarchy(allowingtodefinesupercategories,like +Retrieval-augmented models. Our dataset seems partic- bridges, castles, lighthouses, lakes etc.), and to link them +ularly suited for retrieval-augmented models. While pio- to Wikipedia articles. We use these provided annotations +neered in NLP [9, 23, 30, 33], several recent works use to speed up the annotation of our dataset, to identify rele- +retrieval for VQA. KRISP [36] leverages triplets encoding vantknowledgecategories,andtoassignimagesItotextual +facts, categorical knowledge and object relationships in a questionsQautomatically. +graph-based reasoning framework. KAT [22] uses Wiki- We create our controlled knowledge base starting from +data triplets, and a reasoning module cross-attends them the WIT dataset [48], which contains 37M image-snippet +withGPT-3answers,theresultbeingfedintoadecoderfor pairs with 11M unique images from Wikipedia. We select +answer generation. InFactuality [52] leverages index im- allimageslinkedtoanEnglishsnippetandthenextendWIT +agestolinkentitiespresentinthequeryimageandWikidata to include the full corresponding Wikipedia article (snap- +triplets to gather facts, which are fed into a UNITER [16] shot of 13 August 2022). This spans 2M English articles. +reasoning module. REVEAL [26]’s external knowledge is We then identify which categories of iNat21 and GLDv2 +composedofimage-textpairs,questionansweringpairsand map one-to-one to a single Wikipedia article. We found +knowledgegraphtriplets,whichareusedtoassistagenera- such unique mappings for 80% of the iNat21 and 50% of +tormoduleinproducinganswers. the GLDv2 categories. We only create questions for those -F ARYAMAN FIN ANCIAL SERVICES LIMITED -Place : Mumbai -| EXPLANATORY | | STATEMENT | | AS | | | -| -------------------- | --- | ---------- | --------- | ------------------------- | ---------- | --- | -| | | | | Date : December 01, 2005 | ForAryaman | | -| REQUIRED BY SECTION | | | 173(2) OF | | | | -THE COMPANIES ACT, 1956. Financial Services Limited -Prasanna Chandwaskar -ITEM NO. 4, 5 & 6 -| Shri Munnangi V Ramana, Shri Suhas S | | | | Company Secretary | | | -| ------------------------------------ | --- | --- | --- | ----------------- | --- | --- | -Ganpule and Shri Baliga were appointed as -DIRECTOR'S REPORT -Additional Directors in accordance with the -provisions of section -| 260 of the Companies | | Act, 1956 at Board | | TO THE MEMBERS | | | -| --------------------- | --- | ------------------ | --- | -------------- | --- | --- | -Meeting held on December 20, 2004, May -Your Directors present the Eleventh -02,2005 and December 1,2005 respectively. -Annual Report and audited accounts of -The Company has received notice from the -your -shareholder of the Company under section -Company for the financial year ended -| 257 of the Companies | | Act, 1957 for | | the | | | -| --------------------- | --- | -------------- | --- | --- | --- | --- | -June 30, 2005. -appointment of Shri Ramana, Shri Ganpule -and Shri Baliga as Director of the Company, -| | | | | FINANCIAL RESULTS | (Rs. in lakhs) | | -| --- | --- | --- | --- | ----------------- | -------------- | --- | -at the ensuing Annual General Meeting. -| | | | | Particulars | 30th June, | 31s1 March, | -| --------------------- | --- | --------------- | --- | ----------- | ----------- | ----------- | -| The Board recommends | | the approval of | | | | | -| | | | | | 2005 | 2004 | -resolution by way of simple majority. -| | | | | Income from operations | 27.67 | 81.44 | -| --- | --- | --- | --- | ----------------------- | ----- | ----- | -None of the directors except Shri Munnangi Less- Expenditure 35.21 86.73 -(5.29) -V Ramana, Shri Suhas Ganpule and Shri Profit before Depreciation (7.54) -(6.49) -Baliga, are in any way concerned or Less -Depreciation (5.03) -| | | | | Profit before tax | (12.57) | (11.78) | -| --- | --- | --- | --- | ------------------ | ------- | ------- | -interested in the said resolution at 4, 5 & 6 -| respectively. | | | | Provision for tax | | | -| ------------- | --- | --- | --- | ----------------- | --- | --- | -(269.64) -| | | | | Balance carried to | (281.42) | | -| --- | --- | --- | --- | ------------------- | -------- | -------- | -| | | | | Balance Sheet | (293.99) | (281.42) | -ITEM NO. 7 & 8 -The Board of Directors in its meeting held -DIVIDEND -| on 30th August, | | 2005 has approved | | the | | | -| ---------------- | --- | ------------------ | --- | --- | --- | --- | -Due to absence of profits, the Director -| expansion | of the Company's | | existing | | | | -| ---------- | ----------------- | --- | -------- | --- | --- | --- | -do not recommend any dividend for the -activities. -year. -| Considering | the expansion | | proposals | | | | -| ------------ | -------------- | --- | --------- | --- | --- | --- | -DEPOSITS -| entailing borrowings | | the limits specified | | | | | -| --------------------- | --- | -------------------- | --- | --- | --- | --- | -pursuant to Sections 293 (1) (a) and (d) of -| | | | | Your company did not accept/ | | hold any | -| --- | --- | --- | --- | ----------------------------- | --- | -------- | -the Companies Act, 1956 vide the existing -deposit from public/ shareholders during -resolutions passed by the members of the -the year under review. -Company would be inadequate and hence -the same needs to be suitably increased. -DIRECTORS -The Board recommends the approval of -Shri Mohan Datari, Director of the company -resolution by way of simple majority. -retire by rotation and offers himself for -reappointment. -None of the directors of the Company is -concerned or interested in the resolution. -The shareholders are requested to appoint to -Shri Munnangi V Ramana, Shri Suhas S -By Order of the Board -Ganpule Shri Mohan Baliga as Directors +Monastery Horezu Monastery Templated question based upon super category +Horezu Monastery +Question: Who founded this monastery? +The Horezu Monastery or Hurezi Monastery was founded in 1690 by Prince +| | | | Constantin Brâncoveanu in the town of Horezu, Wallachia, Romania. It is | | | | | Answer: | | | | | +| --- | --- | --- | ------------------------------------------------------------------------ | --- | --- | --- | --- | ------- | --- | --- | --- | --- | +considered to be a masterpiece of "Brâncovenesc style", known for its +| | | | architectural purity and balance, the richness of its sculpted detail, its | | | | | Evidence: | | Copy selected evidence | | | +| --- | --- | --- | --------------------------------------------------------------------------- | --- | --- | --- | --- | --------- | --- | ---------------------- | --- | --- | +| | | | treatment of religious compositions, its votive portraits, and its painted | | | | | | | Not enough evidence | | | +decorative works. +Plant Acacia paradoxa Acacia paradoxa Automated question from wikipedia section +Are the following question/answer pairs relevant and correct? +Description +The large shrub or tree up to 2 to 4 metres (7 to 13 ft) tall and has a similar Question: How many feet tall does the acacia Yes +paradoxa grow to? +| | | | width, it has ribbed branchlets that are often arched downward. It is dense | | | | | Answer: | 7 to 13 | | | No | +| --- | --- | --- | ---------------------------------------------------------------------------- | --- | --- | --- | --- | -------- | ------- | --- | --- | --- | +with foliage; the leaves are actually enlarged petioles known as phyllodes. +They are crinkly and the new ones are covered in hairs. The erect phyllodes +| | | | | | | | | Question: | What are the new leaves of this plant | | | Yes | +| --- | --- | --- | ------------------------------------------------------------------------- | --- | --- | --- | --- | ---------- | -------------------------------------- | --- | --- | --- | +| | | | are asymettric and have a lanceolate shape and are around 30 millimetres | | | | | | covered in? | | | | +(1.18 in) in length and 7 mm (0.276 in) wide. Answer: hairs No +Figure4: Datacollectionfortemplatedandautomaticallygeneratedsingle-hopquestions. Top: Expertscreatetemplatedquestions +Q based on a super category (e.g. Monastery). Annotators are given the full Wikipedia page for a particular C, and asked to provide +theanswersAandtheevidence. Bottom: QuestionsareautomaticallygeneratedfromaWikipediasectionofC andvalidatedbyhuman +annotators.Inbothprocesses,theannotatorneverseestheimageIcorrespondingtoC. +categorieswithauniquelyidentifiableWikipediaarticle. resolvethespecificcategoryC andamodelsimplyremem- +The combination of iNat21 & GLDv2 with Wikipedia beringQ+Apairscannotsolvethechallenge. +as knowledge base enables creating (I,Q,A) triplets cov- Automaticallygeneratedsingle-hop. Weincreasethedi- +ering the long-tail: questions about detailed properties (on versity of questions in the dataset with automatically gen- +| Wikipedia | text) | of fine-grained | categories | (species | from | | | | | | | | +| --------- | ----- | --------------- | ---------- | -------- | ---- | ---------------- | --- | ----- | ----------- | -------- | --- | -------- | +| | | | | | | erated questions | | (Fig. | 4). Similar | to [27], | we | feed the | +iNat21) and instances (landmarks from GLDv2). Hence, Wikipedia article of a category to a question generation +thissatisfiestheencyclopedicdesignprinciple. model [13]. It processes a section of an article at a time, +| | | | | | | producing | a large | number | of Q+A | pairs | (typically | 100s), | +| ----------------------- | --- | --- | --- | --- | --- | --------------------------------- | ------- | ------ | ------ | ---------------------- | ---------- | ------ | +| 4.2.Single-hopquestions | | | | | | byinvertingstatementsfromthetext. | | | | | | | +| | | | | | | Weincludetwofilteringsteps. | | | | First,werequirethecat- | | | +Theanswertoasingle-hopquestioncanbefoundinthe +| | | | | | | egory name | of | C to be | used in the | question. | This | reduces | +| ------------------------------ | --- | --- | ------------------- | --- | --- | ---------- | --- | ------- | ----------- | --------- | ---- | ------- | +| WikipediaarticleofthecategoryC | | | whichisthesubjectof | | | | | | | | | | +thequestion.Weconstructsuchquestionsintwoalternative thenumberofquestionsby20×,andensuresthattheques- +| | | | | | | tion is about | C | (and not | about another | entity | found | in the | +| --- | --- | --- | --- | --- | --- | ------------- | --- | -------- | ------------- | ------ | ----- | ------ | +ways: templatedandautomaticallygenerated. +| | | | | | | snippet | of text). | Second, | to increase | diversity | we | remove | +| -------------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --------- | --------- | ----------- | ---------- | ------------ | ------ | +| Templatedsingle-hop. | | Weuseofthesupercategoryanno- | | | | | | | | | | | +| | | | | | | near identical | | questions | and limit | the number | of questions | | +tationsavailableinthesupportingdatasets(e.g.reptilesfor +perWikipediasection. +| iNat21, monasteries | | for GLDv2). | For each | super | category | | | | | | | | +| ------------------- | --- | ----------- | -------- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | +wemanuallydefineseveralquestions,likeWhofoundedthis The human annotation task is now extremely simple: +validatewhethertheQ+Apairisrelevantandcorrect,given +| monastery?. | Wethenaskhumanannotatorstoanswerthese | | | | | | | | | | | | +| ----------- | ------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +questionsforaparticularspecies/landmarkC (e.g.Horezu theWikipediasectionusedtogenerateit. AquestionQis +| | | | | | | relevant | if it is | is well-formed, | makes | sense | semantically, | | +| --- | --- | --- | --- | --- | --- | -------- | -------- | --------------- | ----- | ----- | ------------- | --- | +Monastery),showingthemtherelevantWikipediapage(see +Fig. 4). Hence, theannotatorsdonothavetorecognizeC and is expected to have an unambiguous answer. The an- +in an image, which would require expert knowledge. In- swer A is correct if it is supported by the given Wikipedia +| | | | | | | section. | Finally, | wecandirectlyusetheWikipediasection | | | | | +| --- | --- | --- | --- | --- | --- | -------- | -------- | ----------------------------------- | --- | --- | --- | --- | +stead,theyjusthavetoidentifywhethertheanswerisinthe +Wikipediaarticle. Ifyes,theymarktheevidenceforit(for usedtogeneratetheQApairasevidencefortheanswer(at- +tributionforfree!) +| attribution). | If | no, we discard | that question | (as | it cannot | | | | | | | | +| ------------- | --- | -------------- | ------------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +be answered with our knowledge base and it would break About 80% of the questions sent to annotators are val- +thecontrolledknowledgebaseprinciple). Westartwith10 idated positively. The questions contain (by design) the +questions per category, resulting in 4.3 questions on aver- name of the category C, and we rephrase them using a +ageperspecieswithavalidanswerforiNat21,and3.4per carefully-promptedFLAN[18]versionofPaLM[17]tore- +| | | | | | | place the | mention | of | C by its super | category, | which | pro- | +| --- | --- | --- | --- | --- | --- | --------- | ------- | --- | -------------- | --------- | ----- | ---- | +landmarkforGLDv2(seeexamplesinFig.2firstcolumn). +E´glise +NowwehaveatextualquestionQandananswerA. To duces Q. For example, the question Where is the +form a complete multi-modal question, we pair Q with an Saint-Cannat located? is rephrased by PaLM to Where is +| | | | | | | thischurchlocated? | | AnimageI | | fromC | | | +| --- | --- | --- | --- | --- | --- | ------------------ | --- | -------- | --- | ----- | --- | --- | +imageI ofC fromthesupportingdatasets. NotethatanyQ fromasupporting +occursmultipletimesinthedatasetwithdifferentanswers. datasetisthenassociatedtothisquestion. +HencebydesigntheaccompanyingimageI isnecessaryto Multi-answer questions. For many properties of a cat- -ARYAMAN FINANCIAL SERVICES LIMITED -of the company, they were appointed as The Board wish to place record their -additional Directors under section 260 of the gratitude for their continued co-operation, -Companies Act, 1956 on your Board and the assistance and guidance extended by the -Company has received notice under section banks, clients, stock exchange, members -257 of the Companies Act, 1956 for their and associates. -candidature for appointment of Directors -of the Company. CONSERVATION OF ENERGY -TECHNICAL ABSORBATION, FOREIGN -AUDITORS EXCHANGE EARNING AND OUTGO -M/s Thakur Vidyanath Aiyar & Co, The company is not engaged in any -Chartered Accountants retire at the manufacturing activity and therefore -forthcoming Annual General Meeting and provisions of section 217 (1) (e) of the -eligible for the reappointment. companies Act, 1956 is not applicable. -Comments made by the Auditors in their -report are self-explanatory and therefore CORPORATE GOVERNANCE: -do not call for any further explanations. -During the year under report, your -DIRECTORS' RESPONSIBILITY STATEMENT Company has evolved Corporate -Governance practices. The required -As required under section 217 (2 A A) of the structures for governance have been in -companies, Act 1956 your Directors state:- place and the meetings are periodically -i. While preparing annual accounts the held in compliance with the ideal practices -applicable accounting standards have been prescribed under the code. -followed. -ii. Company has selected such accounting Management Discussion and Analysis -policies and applied them consistently and Report is separately given in the Annual -made judgment that are responsible and Report. A separate report on Corporate -prudent which give true and fair view of Governance is annexed hereto as a part of -affairs of the company. this report. A certificate from the Auditors -iii. Company has taken proper and sufficient of the Company regarding compliance of -care for the maintenance of adequate conditions of Corporate Governance as -accounting records in accordance with the prescribed under Clause 49 of the Listing -provisions of the Act for safeguarding the Agreement is attached to this report. -assets of the company and the preventing -and detecting fraud and other irregularities. ACKNOWLEDGEMENT -iv. Company has prepared accounts on -Your Directors would like to express deep -ongoing concern basis. -sense of appreciation for the assistance arid -co-operation received from the Financial -PERSONNEL -Institutions, Banks, Government -Relations between the management and Authorities and Shareholders and for the -employees were company cordial. Your devoted services by the Executives, Staff -Directors wish to place on record and Workers of the Company. -appreciation of service rendered by all the -staff members. There is no employee who For and on the behalf of the Board -draws salary in excess of the limits as -Director -prescribed under section 217 (2A) of the -companies (Particular of Employees Rules) Place: Mumbai -1975 as amended. Date: December 01, 2005 +Q: What is the main competitor for food for this animal? Q: What is the population size of this animal? Chained two hop question with bridge: spotted hyena +What is the population +size of the main +competitor for food of +this animal? +Q: What is the main competitor for food for this animal? Q: What is the population size of this animal? Chained question with bridge: spotted hyena +| | Lion | | | Spotted hyena | | | | +| --- | ---- | --- | --- | ------------- | --- | --- | --- | +Predatory competition +| | | | | The spotted hyena (Crocuta crocuta), also | | What is the population | | +| --- | --- | --- | --- | ------------------------------------------ | --- | ----------------------- | --- | +Lions and spotted hyenas occupy a similar known as the laughing hyena ... It is listed size of the main +as being of least concern by the IUCN on +ecological niche and compete for prey and a c c o u n t o f it s w id e s p r e a d ra n g e a n d la r g e competitor for food of +c a r ri o n ; a re v ie w o f d a t a a c ro ss s e v e ral n u m b e r s e s t im a te d b e t w e e n 2 7 , 0 0 0 a n d this animal? +| | s tu d i e s i n d ic | a te s a d i e t a r y o ve rl a p o f | | | | | | +| --- | -------------------------------------- | ---------------------------------------------- | --- | --------------------------------------- | --- | --- | --- | +| | 58.6%. Lions typically ignore hyenas… | | | 47,000 individuals. The species is, ... | | | | +Figure 5: Illustrationoftwo-hopquestiongenerationusingbridgeentities. Thebridgeentity,e.g.spottedhyena,isananswertoa +single-hopquestionwhichhasitsownentryintheknowledgebase. Hencewecanaskasecondsingle-hopquestionaboutit. Thesetwo +questionsarethenchainedintoatwo-hopquestion. +egory C the response should be a list with multiple an- pair is combined with (up to) 5 images I showing differ- +swers. For example What fish can be found in this lake? ent instances of the same category C from the supporting +orWhereisthisbirdfound?,couldbeansweredwithalist datasets. Thisincreasesvisualdiversityandintotalweuse +offish(orcountries,respectively;Fig.2,thirdcolumn).For 514kuniqueimages. Thereareintotal15ktextualsingle- +these kinds of questions we define the multi-answer ques- hoptemplatedquestions,and158kautomaticallygenerated +tion type. We filter the automatically generated QA pairs questions. Moreover,thedatasetcontains25kmulti-answer +for questions likely to have multiple answers, and extract questionsand22ktwo-hopquestions. +| aninitiallist. | Thenweasktheannotatorstocompleteand | | | | | | | +| -------------- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +Lookingindetailatthe221kQ+Apairs,175kofthem +verifythelistwithallthepossibleanswers. +haveauniqueQ(i.e.nostringmatchwithanyotherques- +4.3.Two-hopquestions tion). Moreover, there are 95k unique answers, of which +| | | | | 73koccuronlyonce. | Theremaining22kanswersfollowa | | | +| --- | --- | --- | --- | ----------------- | ----------------------------- | --- | --- | +Two-hopquestionsaredefinedasrequiringtworeason- Zipf’slaw:11kanswersoccurtwice,2kanswersoccur10+ +| ing steps | to obtain the | final answer. | Here we focus | on | | | | +| --------- | ------------- | ------------- | ------------- | --- | --- | --- | --- | +times,andlessthan400answersoccurmorethan50times. +chained two-hopquestions,whichrequiretwoconsecutive +retrievalsteps. Toconstructsuchquestionsatscale,weuse Train / Val / Test splits. To allow for evaluating differ- +thenotionofabridgeentity[58]: iftheanswertoasingle- entpropertiesofthedataset, wecarefullydesignitstrain/- +hop question is an entity with its own Wikipedia page, it val/test splits. First, there is no overlap in the images I +servesasabridgeentity,aboutwhichwecanaskasecond usedinthetrain,val,ortestsplits. ForGLDv2,wesample +single-hopquestion,seeFig.5. images for all our splits from their train-clean split, which +We automatically identify bridge entities from the avoids most noisy labels in their dataset. For iNat21, we +single-hopanswers(Sec.4.2),thenmanuallydiscardwords sampleourtrainimagesfromtheirtrainset,andourvaland +like yes, blue, and heavy which would yield very artificial test images from their validation set (their test set annota- +multi-hop questions. We then automatically generate and tionsarenotpubliclyavailable). Next, abouttwo-thirdsof +validate questions for these bridge entities using our auto- thequestionsQandhalfoftheanswersAinourvalandtest +maticallygeneratedsingle-hoppipeline. splits do not occur in our train split. Finally, roughly 17% +Wecreatethefinaltwo-hopquestionautomaticallyusing ofthesubjectcategoriesC inourvalandtestsplitsarenot +PaLM, by feeding it the two single-hop questions with an used in our train split. This allows for analyzing different +appropriateprompt.Unfortunately,thesetwo-hopquestions desirablepropertiesofVQAmodelsonourdataset. +areoftenincorrect.Therefore,wecreateasecondpromptto +validatethetwo-hopquestion:PaLMisaskedtoanswerthe +two-hopquestionusingthetwoinitialsingle-hopquestions +| with answers | as context. | If the predicted | answer is | iden- | | | | +| ------------ | ----------- | ---------------- | --------- | ----- | ------------------ | -------------------- | --- | +| | | | | | Numberof(Q,A)pairs | Total(I,Q,A)triplets | | +tical to the answer of the second single-hop question, then Train Val Test Train Val Test +the two-hop question is validated and kept in our dataset; Templated 13,928 400 1,000 66,535 1,827 1,000 +| | | | | Automatic | 153,441 1,750 | 2,750 737,114 | 8,025 2,750 | +| --------------------------------- | ---------------- | ------------------- | ------------ | ----------- | ------------- | ------------- | ----------- | +| otherwise, | it is discarded. | This increases | the accuracy | of | | | | +| | | | | MultiAnswer | 23,929 400 | 1,000 112,736 | 1,844 1,000 | +| thetwo-hopquestionssubstantially. | | Examplesofvalidated | | | | | | +| | | | | TwoHop | 21,040 400 | 1,000 99,866 | 1,895 1,000 | +two-hopquestionsaregiveninFig.2(right). +| | | | | Total | 212,338 2,950 | 5,750 1,016,251 | 13,591 5,750 | +| --- | --- | --- | --- | ----- | ------------- | --------------- | ------------ | +4.4.Datasetstatistics +Table2: +| | | | | DatasetStatistics. | | Wereportthenumberof(question, | | +| --- | --- | --- | --- | ------------------ | --- | ----------------------------- | --- | +Our dataset contains in total 1M (I,Q,A) triplets answer)pairs,and(image,question,answer)tripletsfordifferent +(Tab. 2). These are derived from a total of 221k textual question types. In total our dataset contains 1M VQA triplets, +makingitthelargestofitskind. +| Q+Apairsfrom16.7kdifferentcategoriesC. | | | EachQ+A | | | | | +| -------------------------------------- | --- | --- | ------- | --- | --- | --- | --- | -P= ARYAMANFINANCIALSERVICES LIMITED -dependant upon conditions of primary -| MANAGEMENT DISCUSSION | | | AND | | | | | | -| ---------------------- | --- | --- | --- | -------------- | --- | -------- | ----------- | --- | -| | | | | and secondary | | capital | market. Any | | -ANALYSIS REPORT. -adverse effect on the capital market could -| | | | | affect the performance and | | | | profitability | -| ------------- | ---------- | --- | --- | --------------------------- | --- | --- | --- | ------------- | -| 1. INDUSTRY | STRUCTURE | | AND | | | | | | -of your Company. -DEVELOPMENTS -Merchant Banking and Financial Services -6. INTERNAL CONTROL SYSTEMS -market is fast growing over the last few -The company has an adequate system of -years. Merchant Banker services includes -internal controls that ensure that all assets -providing merchant banking services like -Public Issue Management i.e, acting as are protected against loss from -unauthorized use or disposition and all -| manager / | lead manager to the public | | | | | | | | -| ---------- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | -transaction are recorded and reported in -| issues, right issues, open | | offer | under | | | | | | -| --------------------------- | ----------- | ------ | ------- | ----------- | ----- | ------------------ | --- | --- | -| | | | | conformity | with | generally accepted | | | -| Takeover | Code, Buy | Back | Scheme, | | | | | | -accounting principles. -Portfolio Management Services, private -placement of debt and equity instruments. -7. FINANCIAL PERFORMANCE -The present position of stock market in -India has opened new era for finance The company has achieved a turnover of -| | | | | Rs 27.67 during | | the year as | | against | -| --- | --- | --- | --- | ---------------- | --- | ------------ | --- | ------- | -industry. -| | | | | Rs.81.44 | lacs | in the | corresponding | | -| --- | --- | --- | --- | --------- | ----- | -------- | ------------- | --- | -previous year. During the year company -2. OPPORTUNITY AND THREATS -| Opportunities: | | | | incurred a profit/ | | | loss of Rs. 12.57 as | | -| -------------- | --- | --- | --- | ------------------- | --- | --- | -------------------- | --- | -againstRs.il.78 lacs in the corresponding -Aryaman Financial Services Limited is -previous year. -SEBI registered Category - I. Considering -present capital marVet conditions, your -company has tremendous opportunities in 8. HUMAN RESOURSE AND -INDUSTRIAL RELATIONS -its business activities especially merchant -The industrial relation remain co ordinal -banking activities like issue management, -during the year under review. -portfolio management services, corporate -debt restructuring etc. -9. CAUTIONARY STATEMENT -Statement in this report on management -Threats: -| | | | | Discussion | and analysis describing the | | | | -| --- | --- | --- | --- | ----------- | --------------------------- | --- | --- | --- | -With the number of players coming down, -| | | | | company's | objectives, | | projections, | | -| --- | --- | --- | --- | ---------- | ------------ | --- | ------------ | --- | -we expect marginal competition. -estimates, expectations or prediction are -| | | | | based | on certain | | assumption | and | -| --- | --- | --- | --- | ------ | ------------ | --- | ----------- | --- | -3. SEGMENTWISE PERFORMANCE -Your Company is dealing in one segment expectation of future events. Actual result -| | | | | could | differ | materially from | | those | -| --- | --- | --- | --- | ------ | ------- | ---------------- | --- | ----- | -mainly in Merchant Banking activity. -| | | | | expressed | or implied. The company | | | | -| --- | --- | --- | --- | ---------- | --------------------------- | --- | --- | --- | -| | | | | assume | no responsibility to amend, | | | | -4. OUTLOOK -The growing economy and vibrant capital modify or revise any of the statements on -the basis of subsequent developments, -market conditions promise good scope for -information or events. -financial service sector. Your company is -well prepared to seize the opportunities as -and when they come. The management is For on behalf of the Board -| generally | optimistic | about | future | | | | | | -| ---------- | ----------- | ------ | ------ | --- | --- | --- | --- | --- | -Director -operations. -| 5. RISKS AND CINCERNS | | | Place: Mumbai | | | | | | -| ---------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | -Date: 01.12.2005 -| Company's | activities | are | mainly | | | | | | -| ---------- | ----------- | ---- | ------ | --- | --- | --- | --- | --- | +Dataset varietyofQ disparityofQ Model Retrieval OK-VQAAccuracy +| | | #uniquebigrams | | | avg.cosinedist. | | | | KAT[22] | | ✓ | | 53.1 | | | +| ----------- | --- | -------------- | --- | ----- | --------------- | ----- | ----- | ------------- | -------- | ---- | ------------------- | --- | ---- | ------- | ----- | +| | | | | | | | | REVEAL[26] | | | ✓ | | 59.1 | | | +| FVQA[53] | | | | 7.9k | | 0.620 | | | | | | | | | | +| | | | | | | | | PromptCap[25] | | | - | | 60.4 | | | +| KVQA[47] | | | | 12.8k | | 0.504 | | | | | | | | | | +| | | | | | | | | | PaLI[14] | | - | | 64.5 | | | +| OK-VQA[37] | | | | 19.3k | | 0.843 | | | | | | | | | | +| S3VQA[27] | | | | 19.7k | | 0.805 | | | | | | | | | | +| | | | | | | | Table | 4: | OK-VQA: | PaLI | is state-of-the-art | | on | OK-VQA. | It is | +| A-OKVQA[46] | | | | 33.9k | | 0.856 | | | | | | | | | | +thereforeangoodcandidatetoapplytoourEncyclopedic-VQA. +| Encyclopedic-VQA | | | 257.9k | | | 0.833 | | | | | | | | | | +| -------------------------------- | ------------ | ------------ | ------------------------- | -------- | -------- | -------- | ------------- | ----------- | --- | ---------- | ---- | ---- | --- | ----- | ------- | +| Table | 3: Diversity | of questions | Q | measured | in terms | of vari- | 5.Experiments | | | | | | | | | +| etyanddisparity(higherisbetter). | | | Disparitynumbersaremostly | | | | | | | | | | | | | +| | | | | | | | | We evaluate | | PaLI [14], | PaLM | [17] | and | GPT-3 | [10] on | +takenfrom[46],butwereproducedtheirA-OKVQAresulttoval- +idate our re-implementation. Our dataset has the largest variety our dataset. We apply all models directly and in retrieval- +andisclosetothebestonesondisparity. augmented settings. For evaluation we use the test split +| | | | | | | | of | our dataset | | on the | single-hop | templated | | and automat- | | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----------- | ------------ | --------- | ---------- | ----------- | ----------- | ------------ | --- | +| | | | | | | | ically | generated | | questions | (except | | for Sec. | 5.6 where | we | +| | | | | | | | also | evaluate | multi-answer | | | and two-hop | questions). | | We | +4.5.QuestionDiversity +measureaccuracyasthepercentageofquestionswherethe +| | | | | | | | predicted | | model | answer | matches | the | ground-truth | | answer. | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----- | ------ | ------- | --- | ------------ | --- | ------- | +WewanttocomparethediversityofourVQAquestions +to other datasets. However, diversity is a broad concept All ground-truth answers and model predictions are pre- +| | | | | | | | processed | | following | standard | | VQA | practices | [21] | to facil- | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --------- | -------- | --- | --- | --------- | ---- | --------- | +whichmanifeststhroughacombinationofthreebasicprop- +| | | | | | | | itate | checking | their | correctness | | (remove | articles, | punctua- | | +| ------- | -------- | ---------- | ----------- | ----- | ---- | -------- | ----- | -------- | ----- | ----------- | --- | ------- | --------- | -------- | --- | +| erties: | variety, | disparity, | and balance | [49]. | Here | we focus | | | | | | | | | | +onvarietyanddisparity. tion, etc.). Furthermore, weusethe’BERTMatching’cri- +| | | | | | | | terion | BEM | [12] | to determine | | whether | a predicted | | answer | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | ---- | ------------ | --- | ------- | ----------- | --- | ------ | +Variety. This refers to the semantic variety spanned by is correct given the question and the ground-truth answer. +| the VQA | questions. | One | simple | measure | of variety | is the | | | | | | | | | | +| ------- | ---------- | --- | ------ | ------- | ---------- | ------ | --- | ---------- | --- | ------ | --- | ---------------- | --- | ------ | ------ | +| | | | | | | | BEM | evaluation | | allows | for | more flexibility | | in the | answer | +numberofuniquetextualquestionsQ(175kinourdataset). +| | | | | | | | formulation | | than | classical | exact | matching, | | coming | much | +| ------------------ | ---- | ------------------------------- | --------- | ---- | --------- | ------- | ----------- | --- | ----- | --------- | ----- | --------------- | --- | ------ | ----- | +| However, | this | ignores the | fact that | some | questions | are se- | | | | | | | | | | +| | | | | | | | closer | to | human | judgement | | of correctness, | | as is | shown | +| manticallysimilar. | | Anothermeasureisthenumberoftop- | | | | | | | | | | | | | | +in[12]andinouruserstudyinAppendixB.Weconsideran +icscoveredbythequestions,forwhichwecanusethenum- answercorrectwhenitsBEMscoreis≥0.5. Ifanexample +| ber of | categories | C (16.7k | in our | dataset). | However, | each | | | | | | | | | | +| ------ | ---------- | -------- | ------ | --------- | -------- | ---- | --- | -------- | ------------ | --- | -------- | --- | ---------- | --- | --- | +| | | | | | | | has | multiple | ground-truth | | answers, | | we compute | the | BEM | +datasettypicallyusesadifferentsetofcategoriesatdiffer- +scoreforeachandchoosethemaximum. +entlevelsofgranularity,makingtheircategorycountshard +to compare directly. So instead, we consider the number 5.1.LargeModelswithoutRetrieval +| of unique | bigrams | across | all questions | Q, | as a | reasonable | | | | | | | | | | +| --------- | ------- | ------ | ------------- | --- | ---- | ---------- | ----- | --- | ------- | --- | ----------- | --- | ---- | --------- | --- | +| | | | | | | | PaLI. | To | measure | the | performance | | of a | large VLM | on | +approximationofthesemanticspacespannedbyallQ. +Encyclopedic-VQA +| | | | | | | | | | | we | use | PaLI [14]. | It yields | the | state- | +| ---------- | --- | ------------- | ------------- | --- | ------------- | --- | ---------- | --- | -------- | --- | ------ | ---------- | --------- | --- | ------ | +| Disparity. | | This measures | how different | | the questions | are | | | | | | | | | | +| | | | | | | | of-the-art | | accuracy | on | OK-VQA | [37] | (64.5%), | and | out- | +from each other. We follow the measure introduced in A- performs retrieval-augmented models such as KAT [22] +| OKVQA | [46]: | (1) we | first project | all | questions | Q into | | | | | | | | | | +| -------- | ----- | -------------- | ------------- | ---------- | --------- | ------ | ------- | --- | ---------- | --- | ---- | ------- | --- | ------- | ------ | +| | | | | | | | (53.1%) | | and REVEAL | | [26] | (59.1%) | by | a large | margin | +| a common | | semantic space | using | a publicly | available | [3] | | | | | | | | | | +(Tab.4). Thisdemonstratesthatthetypeofknowledgere- +Sentence-BERT model [43]; then (2) we measure the av- quired to solve OK-VQA can be captured by large VLMs. +eragecosinedistancebetweenallquestionpairs(definedas +| | | | | | | | PaLI | is therefore | | a good | candidate | | to verify | whether | large | +| --------- | ---------- | --- | ----------- | ---- | ----- | --------- | ---- | ------------ | --- | ------ | --------- | --- | --------- | ------- | ----- | +| 1− cosine | similarity | by | [46]). Note | that | while | [46] pro- | | | | | | | | | | +VLMscapturealsoencyclopedicknowledge.WeuseaPaLI +posedthisasagenericmeasureofdiversity,itactuallyonly +modelwith17Bparameterspre-trainedonahugeamountof +measuresdisparityasdefinedin[49]. +dataincludingWikipedia(soithasseentheknowledgebase +| Results. | | | | | | | containingtheanswersforourdataset). | | | | | | OurPaLImodelis | | | +| -------- | --- | -------------- | --- | --------- | --- | ---------- | ----------------------------------- | --- | --- | --- | --- | --- | -------------- | --- | --- | +| | We | report variety | and | disparity | for | knowledge- | | | | | | | | | | +basedVQAdatasetsinTab.3. Ourdatasetoffersthelargest additionallyfine-tunedonOK-VQAandthereforeparticu- +questionvarietybyanorderofmagnitude.Importantly,this larlysuitedforVQAtasks. WefeedthismodelQ,I inputs +is not only due to dataset size: while Encyclopedic-VQA foreachtestsampletoproducemodelanswers. +andKVQA[47]arethelargestdatasetsandhaveacompa- Tab.5(firstrow)showsthatPaLIhasanaccuracyofonly +rable number of questions, KVQA has much less variety. 13.0%. This low performance indicates that PaLI either +In terms of disparity, our dataset is close to the best ones fails to recognise fine-grained categories and instances, or +OK-VQA[37]andA-OKVQA[46],andbetterthanFVQA, failstoprovidedetailedproperties,orboth. Henceencyclo- +| KVQAandS3VQA. | | | | | | | pedicknowledgeistrulydifficultforthismodel. | | | | | | | | | +| ------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -r ARYAMAN FINANCIAL SERVICES LIMITED -c) The Balance Sheet, the Profit and loss account -AUDITORS' REPORT -and Cash Flow statement dealt with by this -report are in agreement with the books of -ARYAMAN FINANCIAL SERVICES LTD. -account. -| | | | d) | In our opinion, the Balance Sheet, the | | Profit | -| --- | --- | --- | --- | --------------------------------------- | --- | ------ | -We have audited the attached Balance Sheet of -and Loss Account and the Cash Flow Statement -ARYAMAN FINANCIAL SERVICES LTD as -at 30th June, 2005 and the Profit and Loss dealt with by this report comply with the -accounting standards referred to in subsection -Account and also the Cash Flow Statement for -| | | | | (3C) of section | 211 of the Companies Act, | | -| --- | --- | --- | --- | ---------------- | ------------------------- | --- | -the year ended on that date annexed thereto. -1956; -These financial statements are the responsibility -| of the Company's | | management. | Our | | | | -| ------------------- | --- | ------------ | --- | ------------------------------------------- | --- | --- | -| | | | e) | On the basis of the written representations | | | -responsibility is to express an opinion on these -received from the directors, as on 30th June. -financial statements based on our audit. -2005, and taken on record by the Board of -We have conducted our audit in accordance with Directors, we report that none of the directors -is disqualified as on 30th June, 2005 from being -auditing standards generally accepted in India. -appointed as a director in terms of clause (g) -| These Standards | require that we plan and | | | | | | -| ---------------- | ------------------------ | --- | --- | --- | --- | --- | -perform the audit to obtain reasonable assurance of sub-section (1) of section 274 of the -Companies Act, 1956; -about whether the financial statements are free -of material misstatement. An audit includes -| | | | f) | In our opinion and to the best of our information | | | -| --- | --- | --- | --- | ------------------------------------------------- | --- | --- | -examining, on a test basis, evidence supporting -the amounts and disclosures in the financial and according to the explanations given to us, -the said accounts give the information required -statements. An audit also includes assessing the -by the Companies Act, 1956, in the manner so -| accounting principles used and | | significant | | | | | -| ------------------------------- | --- | ----------- | --- | ------------- | ---------------- | ------------- | -| | | | | required and | give a true and | fair view in | -| estimates made by management, | | as well as | | | | | -evaluating the overall financial statement conformity with the accounting principles -presentation. We believe that our audit provides generally accepted in India : -a reasonable basis for our opinion. -| | | | a. | In the case of the Balance Sheet, of the state of | | | -| --- | --- | --- | --- | ------------------------------------------------- | --- | --- | -30'h June. 2005 ; -As required by the Companies (Auditors' affairs of the Company as at -and -| Report) Order, | 2003 issued by the Central | | | | | | -| --------------- | --------------------------- | --- | --- | --- | --- | --- | -Government of India in terms of subsection -| | | | | b. In the case of the Profit and loss account, of | | | -| --- | --- | --- | --- | -------------------------------------------------- | --- | --- | -(4A) of Section 227 of the Companies Act, -1956, we enclose in the Annexure a statement the loss for the year ended on that date. -on the matter specified in paragraph 4 and 5 of -| | | | c. | In the case of the Cash Flow Statement, of the | | | -| --- | --- | --- | --- | ---------------------------------------------- | --- | --- | -the said order. -| | | | | cash flow for the year ended on | | that date. | -| ---------------------------------------- | --- | --- | --- | -------------------------------- | --- | ---------- | -| Further to our comments in the Annexure | | | | | | | -referred to above, we report that: -For THAKUR VAIDYANATH AIYAR & CO. -Chartered Accountants. -| a) We have obtained all the | | information | and | | | | -| ----------------------------- | --------- | ------------ | ------- | --- | --- | --- | -| explanations, | which to | the best | of our | | | | -Place:- Mumbai -| knowledge and belief were necessary | | | for the | | | | -| ------------------------------------ | --- | --- | ------- | --- | --- | --- | -purpose of our audit. -| | | | ( | C.V. Parameswar) | | | -| --- | --- | --- | --- | ---------------- | --- | --- | -b) In our opinion, proper books of account as -Partner -required by law have been kept by the Company -Dated :- 01-12-2005 -so far as appears from our examination of those -M. No. 11541 -books; +Model +| | System | | Retrieval | Extrainput | | PaLI[14] | PaLM[17] | GPT-3[10] | | +| --- | ------------ | --- | --------- | ---------- | --- | -------- | -------- | --------- | --- | +| | Vanillamodel | | - | | - | 13.0% | 19.7% | 15.5% | | +| | | | SubjectC | | - | 16.7% | 31.0% | 26.9% | | +| | Oracle | | KBArticle | | - | 29.7% | 78.4% | 77.4% | | +| | | | KBSection | | - | 48.8% | 87.0% | 82.1% | | +| | | | KBArticle | | - | 21.4% | 48.0% | 44.9% | | +Lens +| | | | KBSection | | - | 28.1% | 48.8% | 44.6% | | +| --- | --- | --- | --------- | --- | --- | ----- | ----- | ----- | --- | +Captions +| | PromptCap | | - | | | 17.8% | 29.7% | 25.6% | | +| --- | --------- | --- | --- | --- | --- | ----- | ----- | ----- | --- | +NNtrainsamples +| Table 5: Accuracy | on single-hop | questions. | | | | | | | | +| ----------------- | ------------- | ---------- | --- | ------------ | -------- | --- | -------------- | --------------------------- | --------- | +| | | | We | report model | accuracy | for | our single-hop | templated and automatically | generated | +questions. WhilelargemodelsstrugglewithourEncyclopedic-VQAdataset,ourexperimentsshowthepromiseofaugmentingthemwith +aretrievalmechanism. +PaLM and GPT-3. Models with a larger language un- trieval component shows great potential for predicting de- +derstanding component [11, 10, 17, 40] might be bet- tailedproperties,and(3)sincePaLMandGPT-3workmuch +ter at memorizing detailed properties and extracting infor- better than PaLI, having a strong language understanding +mation from textual knowledge base entries. Therefore, modelisimportantforextractingtherightinformationfrom +we experiment with PaLM 2 [17] (text-bison@001 thefree-formtextKBarticle. +model, available through the PaLM API) and GPT-3 [10] OracleretrievalofKBsection. Finally,tounderstandthe +(text-davinci-003, accessible through the OpenAI importanceofretrievinginformationthatismoreprecisely +API).Botharetrainedonmassiveamountsoftextincluding localizedthananentireWikipediapage, weprovidetothe +Wikipedia. GPT-3 is especially large, with 175B parame- modelstheground-truthsectionwhichsupportstheanswer +ters. Here we feed each model only the text questions Q (Oracle-KBSectioninTab.5). Again,accuracyimproves +as thery cannot consume images. This measures how well to48.8%forPaLI,87.0%forPaLMand82.1%forGPT-3. +our dataset can be solved by language alone. These mod- Thisdemonstratestheimportanceofprovidingexactinfor- +elsreachamodestaccuracy(19.7%,15.5%,Tab.5). While mationtothelanguagecomponent. Asabonus,thesmaller +they only takes textual inputs, they can be made multi- the retrieved document is upon which the model bases its +modal by adding a visual retrieval mechanism, as we do answer,themoreverifiableandinterpretablethisansweris. +inthefollowingsections. +5.3.LargeModelswithVisualRetrieval +5.2.LargeModelswithOracleRetrieval +Lens-basedretrievalofKBarticle. Togobeyondtheora- +Oracle retrieval of subject C. This experiment tests cledemonstrationabove,asproof-of-conceptweproposeto +whether the most difficult aspect of Encyclopedic-VQA is augmentlargemodelswitharealretrievalsystembasedon +recognizingfine-grainedcategoriesandinstancesintheim- GoogleLens[1]. GoogleLensisanimageretrievalsystem +age. Therefore we provide along with the test question its whichindexesahugeamountofwebimages.Givenaquery +corresponding subject category C as an additional textual image,itfindsotherimagesbasedontheirvisualsimilarity +inputtoourmodels(i.e.aspartoftheprompt;Oracle-Sub- andrelevancetoobjectsitrecognizesinthequeryimage. It +ject C in Tab. 5). While PaLI shows small improvements returnsthemostsimilarindexedimagesalongwithanentity +over its non-retrieval version, PaLM and GPT-3 improve prediction based on these top-ranked images. To augment +considerably to 31.0% and 26.9%, respectively. However, oursystem,wesendGoogleLensthequeryimageI toob- +thisisstillratherlowperformance. Weconcludethatdeter- tain its entity prediction. We then find the best matching +miningthefine-grainedcategoryorinstanceisonlypartof KB article for this entity in our knowledge base. Finally, +whatmakesourdatasetdifficult. wefeedtheretrievedKBarticleasprompttoourmodelsas +| OracleretrievalofKB | article. | Wenowgoa | | stepfurther | | inSec.5.2. | | | | +| ------------------- | -------- | -------- | --- | ----------- | --- | ---------- | --- | --- | --- | +andprovidethefullground-truthWikipediaarticleaboutC Results are shown as Lens - KB Article in Tab. 5. All +tothemodelsintheprompt(Oracle,KBArticleinTab.5). models greatly outperform their non-retrieval augmented +This time, results increase dramatically: 29.7% accuracy versions. PaLMandGPT-3inparticularmorethandouble, +for PaLI, 78.4% for PaLM and 77.4% for GPT-3. This to48.0%and44.9%respectively. +demonstrates that (1) memorizing detailed properties is a Lens-basedretrievalofKBsection. GivenaretrievedKB +hard challenge for vanilla large VLMs, (2) adding a re- article, wenowaimtoselectthemostrelevantKBsection -17= ARYAMANFINANCIALSERVICES LIMITED -have not observed any continuing failure to -ANNEXURE REFERRED TO IN -PARAGRAPH 1 OF OUR REPORT OF correct major weakness and on the basis of -such checks carried out by us during the course -EVEN DATE ON THE ACCOUNTS FOR -THE YEAR ENDED 30th JUNE, 2005. of our audit, the company has an adequate -| | | | internal control | procedures commensurate | | | -| --- | --- | --- | ----------------- | ----------------------- | --- | --- | -In respect of Fixed Assets with the size and nature of its business. -The company has maintained proper records v. In respect of transactions covered under -a) -showing full particulars including quantitative Section 301 of the Companies Act. 1956 -details and situation of fixed assets. -| | | | a) In our | opinion and | according | to the | -| --- | --- | --- | ------------ | ------------- | ---------- | ------- | -b) As explained to us, the fixed assets have been -physically verified by the management during information and explanations given to us, the -the year and no material discrepancy was transactions that need to be entered into the -register maintained under Section 301 of the -noticed on such verification. -.) The company has not disposed / discarded any Act, have been so entered. -of its fixed assets during the year. -b) As per the information and explanations given -to us, there are no transactions of purchase -ii. In respect of inventories -The company is in the business of providing and sales of services have been made in -financial services and in view of this Para 4 (ii) pursuance of contracts or arrangements -| | | | entered | in the register maintained under | | | -| --- | --- | --- | -------- | -------------------------------- | --- | --- | -(a), (b) & (c) of the said order are not applicable -| | | | Section | 301 of the companies | | Act. 1956. | -| --- | --- | --- | -------- | --------------------- | --- | ---------- | -to the company. -aggregating during the year to Rs.5 Lakhs or -iii. In respect of loans, Secured and unsecured, more in respect of each party. -| Granted or taken | by the company | to / from | | | | | -| ----------------- | --------------- | ---------- | --- | --- | --- | --- | -Companies, firms or other parties listed in the vi. As per the information and explanations given -register maintained under Section 301 of the to us, the company has not accepted any -companies Act. 1956 : deposits from the public within the meaning -of Section 58A and 58AA of the Act and rules -a) The company has not granted any loan to framed there under. -Companies, Firms or other parties listed in the -register maintained unHer section 301 of The vii In our opinion, the company has an internal -audit system, which is commensurate with the -| Companies Act, | 1956. | | | | | | -| --------------- | ----- | --- | --- | --- | --- | --- | -b) In view of our comment Para 4 (iii) (a) the size and nature of its business. -clauses (b), (c) and (d) are not applicable to the -| company. | | | viii We have | been informed that | | the central | -| -------- | --- | --- | ------------- | -------------------- | --- | ------------ | -government has not prescribed maintenance -| c) The company has not taken any loan | | from | | | | | -| --------------------------------------- | --- | ---- | --- | --- | --- | --- | -Companies, Firms or other parties listed in the of cost records under Section 209( 1 )(d) of the -register maintained under section 301 of The companies Act. 1956. -| Companies Act, | 1956. | | | | | | -| --------------- | ----- | --- | --- | --- | --- | --- | -ix In respect of statutory dues -| cO In view of our comment | Para 4 (iii) (c) | the | | | | | -| --------------------------- | ----------------- | --- | --- | --- | --- | --- | -clauses (f) and (g) are not applicable to the -| company. | | | a) The company has been generally regular in | | | | -| -------- | --- | --- | --------------------------------------------- | --- | --- | --- | -depositing undisputed statutory dues including -iv. In our opinion and according to information Provident fund, Investors education and -and explanations given to us there are adequate Protection fund, Employees state insurance. -control procedures commensurate with the size Income Tax, Sales Tax, Wealth Tax, Service -| | | | Tax, Custom Duty, Excise Duty, Cess | | | and | -| --- | --- | --- | ------------------------------------ | --- | --- | --- | -of the company and the nature of its business -| | | | other statutory dues with the | | | appropriate | -| --- | --- | --- | ------------------------------ | --- | --- | ----------- | -with regard to purchase of fixed assets and sale -of services. During the course of our audit, we authorities. According to the information and +| | | | Accuracy | | | | 5.5.ComparisontoPromptCap | | | | | | | | +| ------------- | ------- | --- | --------- | ----------- | --- | ----------- | ------------------------- | ----- | ------------------- | ------ | ----------- | ------- | --------- | ------- | +| | | | w/correct | w/incorrect | | KBretrieval | | | | | | | | | +| | | | | | | | We compare | | our retrieval-based | | | methods | to | Prompt- | +| LensRetrieval | Overall | | retrieval | retrieval | | accuracy | | | | | | | | | +| | | | | | | | Cap [25], | which | has | strong | performance | | on OK-VQA | | +| KBArticle | 48.0% | | 77.7% | 21.2% | | 47.4% | | | | | | | | | +(Tab.4). PromptCapisasystemwithmultiplecomponents. +| KBSection | 48.8% | | 82.3% | 20.7% | | 45.6% | | | | | | | | | +| --------- | ----- | --- | ----- | ----- | --- | ----- | -------- | ---------- | ----- | ----- | -------- | --- | ------ | ----- | +| | | | | | | | It has a | captioning | model | which | consumes | | both I | and Q | +andgeneratesanimagecaptiontailoredtoanswertheques- +Table6:Attributionforretrieval-augmentedVLMs.Wereport +| | | | | | | | tion Q. | This caption | is | then | passed | to GPT-3 | as | context | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | --- | ---- | ------ | -------- | --- | ------- | +modelaccuracyforPaLM[17]conditionalonretrievalresults.Re- +trievalsuccessratesaresimilarforarticlesandsections,butPaLM toanswerQ. Additionally,PromptCapperformsin-context +hasbetteraccuracywhenaugmentedwithsections. learning: for a given test question, it uses the CLIP em- +| | | | | | | | beddings | [39] | of Q and | I to | find the | 32 nearest | | training | +| --- | --- | --- | --- | --- | --- | --- | --------- | ----------------------------------------- | -------- | ---- | -------- | ---------- | --- | -------- | +| | | | | | | | examples. | Thenitincludestheir32correspondingPrompt- | | | | | | | +Capcaptions,questionsandanswersintheGPT-3prompt. +| withinitbasedonQ. | | TodosowequeryPaLMwithaspe- | | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +cialpromptfeedingoneKBsectionatatime,alongwithQ In[25]theunderlyingLLMwasGPT-3. Inthisexperiment +while asking ‘can the answer to this question be found in wealsoapplyPromptCapwithPaLMandPaLI.Resultsin +thistext?’. WeretainallsectionsforwhichPaLManswers Tab. 5 show that PromptCap indeed does better than using +‘Yes’ (usually only one), and consider their concatenation PaLI, PaLM, or GPT-3 alone. However, it still performs +| | | | | | | | substantially | worse | than | our retrieval-augmented | | | methods. | | +| ------------------------------- | --- | --- | --- | --------------------- | --- | --- | ------------- | ----- | ---- | ----------------------- | --- | --- | -------- | --- | +| asthefinalretrieved‘KBsection’. | | | | Finally,wefeedthisre- | | | | | | | | | | | +trievedsectionintheprompttoallmodels(PaLI,PaLMor Thisfurtherconfirmsthehardchallengeourdatasetposes. +GPT-3)asinSec.5.2. +5.6.Multi-answerandtwo-hopquestions +| Results | are shown | as | Lens - | KB Section | in | Tab. 5. Per- | | | | | | | | | +| ------- | --------- | --- | ------ | ---------- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | +formance is now even higher for PaLI (28.1%) and PaLM To evaluate multi-answer questions, we convert the +| | | | | | | | model prediction | | into | a set | of strings | and | compute | the | +| -------- | ------- | ----- | ---- | -------- | ---------- | -------- | --------------------------- | --- | ---- | ----- | ---------- | ---- | ------- | ------- | +| (48.8%), | whereas | GPT-3 | does | not seem | to benefit | further. | | | | | | | | | +| | | | | | | | intersection-over-the-union | | | (IoU) | between | this | set | and the | +Thesenumbersareroughlyhalfwaybetweenversionswith- +outretrievalandwithoracleretrieval. Weconcludethatre- set of ground-truth answers. If IoU >= 0.5 then we con- +| | | | | | | | sider that | prediction | as | correct. | If not, | then | we use | BEM | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---------- | --- | -------- | ------- | ---- | ------ | --- | +trievalaugmentationworksinpractice,andthatourdataset +leavessignificantheadroomforfutureresearchonretrieval- to determine the equivalence of the prediction list string +| | | | | | | | and the | ground-truth | list. | For | two-hop | questions, | | we use | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | ----- | --- | ------- | ---------- | --- | ------ | +augmentedVLMs. +| | | | | | | | BEM as | described | for | single-hop | templated | | and automati- | | +| ---------- | ---------- | --- | --- | -------- | ------ | -------- | ------ | --------- | --- | ---------- | --------- | --- | ------------- | --- | +| CLIP-based | retrieval. | | KAT | [22] and | REVIVE | [34] are | | | | | | | | | +callygeneratedquestions. +| two retrieval-augmented | | | VQA | systems | which | use frozen | | | | | | | | | +| ----------------------- | ---------- | --- | ------- | ---------- | ----- | ----------- | ---- | --------- | --- | ------- | --------- | ------- | --- | ----- | +| | | | | | | | PaLI | with Lens | KB | Section | Retrieval | obtains | an | accu- | +| CLIP [39] | embeddings | to | perform | retrieval. | | In Appendix | | | | | | | | | +racyof9.2%formulti-answerquestionsand14.7%fortwo- +| C we explore | there | whether | KAT | and | REVIVE | could po- | | | | | | | | | +| ------------ | ----- | ------- | --- | --- | ------ | --------- | -------------- | --- | ---------- | ---- | ---- | ---------- | --- | --------- | +| | | | | | | | hop questions. | | Similarly, | PaLM | with | KB Section | | retrieval | +tentiallysucceedonEncyclopedic-VQA. +| | | | | | | | achieves | 33.6% | and 22.8% | for | multi-answer | | and | two-hop | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ------------- | --------- | ----- | ------------ | ----- | --- | ------- | +| | | | | | | | questions, | respectively. | | GPT-3 | obtains | 32.1% | for | multi- | +answerand18.7%fortwo-hopquestions,inbetweenPaLI +5.4.AttributionforRetrieval-Augmentedmodels +andPaLM.Thesemodestperformances,especiallyontwo- +hopquestions,confirmthechallengeofourproposedtasks, +| The attribution | | annotations | | of our dataset | | enable mea- | | | | | | | | | +| --------------- | --- | ----------- | --- | -------------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +andhighlighttheusefulnessofEncyclopedic-VQAtomea- +suringwhetherretrieval-augmentedmodelsgivethecorrect +| | | | | | | | sure progress | on | designing | retrieval | | mechanisms | for | such | +| ------------------------ | --- | --- | --------------------------- | --- | --- | --- | ------------- | --- | --------- | --------- | --- | ---------- | --- | ---- | +| answerfortherightreason. | | | Tab.6measuresattributionfor | | | | | | | | | | | | +complextypesofquestions. +thePaLMmodelswithLensretrieval. +| First, | we observe | that | Google | Lens | retrieves | the cor- | | | | | | | | | +| ------ | ---------- | ---- | ------ | ---- | --------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +6.Conclusions +| rect KB | Article 47.4% | | of the | time. Furthermore, | | we re- | | | | | | | | | +| ------- | ------------- | --- | ------ | ------------------ | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | +port 45.6% retrieval accuracy at the finer level of a KB We introduced Encyclopedic-VQA, a large-scale VQA +Section. Thisdemonstratesthatgivenacorrectlyretrieved dataset about detailed properties of fine-grained categories +KBArticle,oursystemalmostalwaysfindsthecorrectSec- and instances, which includes a knowledge base with an- +tion within it. More importantly, if the retrieved KB Sec- swerattributions. Wedemonstratedthatourdatasetistruly +tionisincorrect,accuracydropsdrasticallyto20.7%,close difficultforstandardVLMs. Additionally,weshowedwith +tothenon-augmentedvariant(bothwhenusingPaLM).In both an oracle experiment and a prototype automatic sys- +contrast, if the correct KB Section is found, accuracy is tem that augmenting these models with a retrieval compo- +82.3%.Thus,retrievingthecorrectdocument,andevenbet- nent substantially improves results, yet leaving headroom +terthecorrectsection,isessentialtoperformingwellonour for even further improvements. Therefore our dataset en- +| dataset. | | | | | | | ablesfutureresearchonretrieval-augmentedVLMs. | | | | | | | | +| -------- | --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -p ARYAMAN FINANCIAL SERVICES LIMITED -given any guarantee for loans taken by -explanation given to us, no undisputed amounts -payable in respect of the aforesaid dues were others from banks or financial -| outstanding as it 30th June, 2005 for a period of | | | institutions. | | | -| ------------------------------------------------- | --- | --- | ------------- | --- | --- | -more than six months from the date they became -payable except Profession Tax dues of Rs. 10507 xvi. The company has not raised any term -- and Income Tax(TDS Payable) Rs.1,06,864/- loans during the year. -xvii. On the basis of our examination of -b) Based on the information made available to us -and explanation given to us, the details of dues balance sheet it appears that the funds -which have not been deposited on account of raised on short term basis are not used -for long term basis and vice versa. -dispute are given below : -Particulars Assessment year Forum where -Amount For which the matter dispute is xviii. The company has not made any -(Rs. In Lakhs) preferential allotment of shares during -| Pertains to pending | | | the year to parties | and companies | | -| ------------------- | ------- | ----- | ---------------------------------------- | ------------- | --- | -| Income Tax | 1997-98 | | covered in the register maintained under | | | -| CIT (Appeals) | | 26.53 | section 301 of the companies Act. 1956. | | | -x. The accumulated loss of the company is not xix. The company has not issued any -more than 50% of its net worth and the company debentures during the year and therefore -has incurred cash loss during financial year para 4 (xix) of the order is not applicable -| covered by our audit and the company had not | | | to the company. | | | -| -------------------------------------------- | --- | --- | --------------- | --- | --- | -incurred cash loss in the immediately preceding -| financial year. | | | xx. The company has not raised any money | | | -| --------------- | --- | --- | ---------------------------------------- | --- | --- | -by way public issue during the year. -xi. Based on our audit procedures and according -to the information and explanation given to us xxi. During the course of our examination of -we are of the opinion that the company has not the books and records of the company -defaulted in repayments of dues to the financial carried out in accordance with generally -accepted auditing practices in India and -institution, banks or debenture holders. -| | | | according | to information | and | -| --- | --- | --- | ---------- | ---------------- | --- | -xii. In our opinion and according to the information explanation given to us we have neither -and explanations given to us no loans and come across any instance of fraud on or -advances have been granted by the company by the company during the year, nor have -on the basis of security by way of pledge of the we been informed of such case by the -| shares, debenture & other securities. | | | management. | | | -| ------------------------------------- | --- | ------------ | ----------- | --- | --- | -| xiii. The company is not a chit fund | | or a nidhi / | | | | -mutual benefit funds / society. Therefore para ForThakur Vaidyanath Aiyar & Co. -4 (xiii) is not applicable to the company. Chartered Accountants -xiv. Based on information and explanation given to -| us, the company is not dealing or trading in | | | Place: Mumbai | | | -| ---------------------------------------------- | ----------- | ---------- | ----------------- | --- | --- | -| shares, securities, | debentures | and other | | | | -| investments and hence the requirements of para | | | (C.V. Parameswar) | | | -4 (xiv) of the order are not applicable to the -| company. | | | Partner | | | -| -------- | --- | --- | ----------------- | ----------- | --- | -| | | | Date: 01-12-2005 | M.No. 11541 | | -xv. In our opinion and according to the information -& explanation given to us, the company has not +References [12] Jannis Bulian, Christian Buck, Wojciech Gajewski, Ben- +| | | | | | | | jamin | Boerschinger, | | and Tal | Schuster. | Tomayto, | | tomahto. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------- | --- | ------- | --------- | -------- | --- | -------- | +[1] GoogleLens. https://lens.google.com-Webinterfaceavail- beyondtoken-levelanswerequivalenceforquestionanswer- +| ableathttps://images.google.com. | | | | 8,13 | | | | | | | | | | | +| -------------------------------- | --------- | ----- | --------- | --------- | ------------ | --- | -------------- | --- | ------------------------------ | --- | --- | --- | --- | -------- | +| | | | | | | | ingevaluation. | | arXivpreprintarXiv:2202.07654, | | | | | 2022. 7, | +| [2] Google | universal | image | embedding | challenge | leaderboard. | | | | | | | | | | +13 +https://www.kaggle.com/competitions/google-universal- [13] Soravit Changpinyo, Doron Kukliansky, Idan Szpektor, Xi +image-embedding/leaderboard. 14 Chen, NanDing, andRaduSoricut. Allyoumayneedfor +[3] Huggingface. https://huggingface.co/sentence- VQAareimagecaptions. InNAACL,2022. 4,5 +transformers/multi-qa-MiniLM-L6-cos-v1. 7 [14] XiChen,XiaoWang,SoravitChangpinyo,AJPiergiovanni, +[4] inat challenge 2021 - fgvc8 - leaderboard. Piotr Padlewski, Daniel Salz, Sebastian Goodman, Adam +https://www.kaggle.com/competitions/inaturalist- Grycner,BasilMustafa,LucasBeyer,AlexanderKolesnikov, +| | | | | | | | Joan | Puigcerver, | Nan | Ding, | Keran | Rong, | Hassan | Akbari, | +| ----------------- | --- | --- | --- | --- | --- | --- | ---- | ----------- | --- | ----- | ----- | ----- | ------ | ------- | +| 2021/leaderboard. | | 14 | | | | | | | | | | | | | +[5] WikimediaCommons. https://commons.wikimedia.org. 4 GauravMishra,LintingXue,AshishThapliyal,JamesBrad- +| | | | | | | | bury, | Weicheng | Kuo, | Mojtaba | Seyedhosseini, | | | Chao Jia, | +| --- | --- | --- | --- | --- | --- | --- | ----- | -------- | ---- | ------- | -------------- | --- | --- | --------- | +[6] ArmenAghajanyan,BernieHuang,CandaceRoss,Vladimir +| | | | | | | | Burcu | Karagol | Ayan, | Carlos | Riquelme, | | Andreas | Steiner, | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------- | ----- | ------ | --------- | --- | ------- | -------- | +Karpukhin,HuXu,NamanGoyal,DmytroOkhonko,Man- +| | | | | | | | Anelia | Angelova, | Xiaohua | | Zhai, Neil | Houlsby, | | and Radu | +| --- | --- | --- | --- | --- | --- | --- | ------ | --------- | ------- | --- | ---------- | -------- | --- | -------- | +darJoshi,GargiGhosh,MikeLewis,andLukeZettlemoyer. +| | | | | | | | Soricut. | PaLI: | A Jointly-Scaled | | | Multilingual | Language- | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ---------------- | --- | --- | ------------ | --------- | --- | +CM3:Acausalmaskedmultimodalmodeloftheinternet.In +| | | | | | | | ImageModel. | | arXiv,2022. | 1,2,3,7,8 | | | | | +| ----------------- | ------------ | ---------- | ------------- | ------------- | ----------- | --------- | ----------------------------- | --------- | ------------ | ---------- | -------------- | --------- | -------- | --------- | +| arXiv,2022. | | 3 | | | | | | | | | | | | | +| | | | | | | | [15] Yang | Chen, | Hexiang | Hu, Yi | Luan, | Haitian | Sun, | Soravit | +| [7] Jean-Baptiste | | Alayrac, | Jeff Donahue, | | Pauline | Luc, An- | | | | | | | | | +| | | | | | | | Changpinyo, | | Alan Ritter, | | and Ming-Wei | | Chang. | Can | +| toine | Miech, | Iain Barr, | Yana | Hasson, Karel | Lenc, | Arthur | | | | | | | | | +| | | | | | | | pre-trained | | vision and | language | | models | answer | visual | +| Mensch, | Katie | Millican, | Malcolm | Reynolds, | Roman | Ring, | | | | | | | | | +| | | | | | | | information-seekingquestions? | | | | InarXiv,2023. | | 4 | | +| Eliza | Rutherford, | Serkan | Cabi, | Tengda Han, | Zhitao | Gong, | | | | | | | | | +| | | | | | | | [16] Yen-Chun | Chen, | Linjie | Li, | Licheng | Yu, Ahmed | | El Kholy, | +| Sina | Samangooei, | Marianne | Monteiro, | Jacob | Menick, | Se- | | | | | | | | | +| | | | | | | | Faisal | Ahmed, | Zhe | Gan, | Yu Cheng, | and | Jingjing | Liu. | +| bastian | Borgeaud, | Andrew | Brock, | Aida | Nematzadeh, | Sa- | | | | | | | | | +| | | | | | | | UNITER: | UNiversal | | Image-TExt | Representation | | | Learning. | +| hand | Sharifzadeh, | Mikolaj | Binkowski, | | Ricardo | Barreira, | | | | | | | | | +| | | | | | | | InECCV,2020. | | 4 | | | | | | +| Oriol | Vinyals, | Andrew | Zisserman, | and | Karen | Simonyan. | | | | | | | | | +| | | | | | | | [17] Aakanksha | | Chowdhery, | Sharan | | Narang, | Jacob | Devlin, | +Flamingo:aVisualLanguageModelforFew-ShotLearning. +| | | | | | | | Maarten | Bosma, | Gaurav | | Mishra, | Adam | Roberts, | Paul | +| ------------- | ------ | --------- | -------- | ------ | --- | -------- | ------- | --------- | ------ | ------ | ------- | ------ | -------- | ------ | +| NeurIPS,2022. | | 1,2,3 | | | | | | | | | | | | | +| | | | | | | | Barham, | Hyung | Won | Chung, | Charles | | Sutton, | Sebas- | +| [8] Stanislaw | Antol, | Aishwarya | Agrawal, | Jiasen | Lu, | Margaret | | | | | | | | | +| | | | | | | | tian | Gehrmann, | Parker | | Schuh, | Kensen | Shi, | Sasha | +Mitchell, Dhruv Batra, C. Lawrence Zitnick, and Devi Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker +Parikh. VQA: visual question answering. In ICCV, 2015. Barnes, Yi Tay, Noam Shazeer, Vinodkumar Prabhakaran, +3,13 +| | | | | | | | EmilyReif, | | NanDu, | BenHutchinson, | | ReinerPope, | | James | +| ------------- | --------- | ------ | --- | ------- | ------ | --------- | ---------- | ----- | ------- | -------------- | ------- | ----------- | --- | -------- | +| [9] Sebastian | Borgeaud, | Arthur | | Mensch, | Jordan | Hoffmann, | | | | | | | | | +| | | | | | | | Bradbury, | Jacob | Austin, | | Michael | Isard, | Guy | Gur-Ari, | +TrevorCai,ElizaRutherford,KatieMillican,Georgevanden Pengcheng Yin, Toju Duke, Anselm Levskaya, Sanjay +Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier +Clark,DiegodeLasCasas,AureliaGuy,JacobMenick,Ro- Garcia,VedantMisra,KevinRobinson,LiamFedus,Denny +manRing,TomHennigan,SaffronHuang,LorenMaggiore, +| | | | | | | | Zhou, | Daphne | Ippolito, | David | Luan, | | Hyeontaek | Lim, | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------ | --------- | ----- | ----- | --- | --------- | ---- | +ChrisJones,AlbinCassirer,AndyBrock,MichelaPaganini, +| | | | | | | | Barret | Zoph, | Alexander | Spiridonov, | | Ryan | Sepassi, | David | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | --------- | ----------- | --- | ---- | -------- | ----- | +GeoffreyIrving, OriolVinyals, SimonOsindero, KarenSi- Dohan, Shivani Agrawal, Mark Omernick, Andrew M. +monyan,JackW.Rae,ErichElsen,andLaurentSifre. Im- Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, +provinglanguagemodelsbyretrievingfromtrillionsofto- AitorLewkowycz, EricaMoreira, RewonChild, Oleksandr +| kens. | arXiv:2112.04426,2022. | | | 1,4 | | | | | | | | | | | +| ----- | ---------------------- | --- | --- | --- | --- | --- | -------- | --------- | --- | ------------ | --- | ----- | ------ | ----- | +| | | | | | | | Polozov, | Katherine | | Lee, Zongwei | | Zhou, | Xuezhi | Wang, | +[10] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- Brennan Saeta, Mark Diaz, Orhan Firat, Michele Catasta, +biah,JaredDKaplan,PrafullaDhariwal,ArvindNeelakan- JasonWei,KathyMeier-Hellstern,DouglasEck,JeffDean, +tan,PranavShyam,GirishSastry,AmandaAskell,etal.Lan- Slav Petrov, and Noah Fiedel. PaLM: Scaling Language +guagemodelsarefew-shotlearners. Advancesinneuralin- ModelingwithPathways. arXiv,2022. 5,7,8,9,14 +formationprocessingsystems,33:1877–1901,2020. +| | | | | | | 7,8 | [18] Hyung | Won | Chung, | Le Hou, | Shayne | | Longpre, | Barret | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------ | ------- | ------ | --- | -------- | ------ | +[11] TomB.Brown,BenjaminMann,NickRyder,MelanieSub- Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, +biah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakan- Mostafa Dehghani, Siddhartha Brahma, et al. Scaling +tan, Pranav Shyam, Girish Sastry, Amanda Askell, Sand- instruction-finetuned language models. arXiv preprint +| | | | | | | | arXiv:2210.11416,2022. | | | 5,14 | | | | | +| ------------- | --- | ------------------- | --- | -------- | -------- | --- | ---------------------- | --- | --- | ---- | --- | --- | --- | --- | +| hini Agarwal, | | Ariel Herbert-Voss, | | Gretchen | Krueger, | Tom | | | | | | | | | +Henighan,RewonChild,AdityaRamesh,DanielM.Ziegler, [19] Matthew Dunn, Levent Sagun, Mike Higgins, V. Ugur +JeffreyWu,ClemensWinter,ChristopherHesse,MarkChen, Guney, Volkan Cirik, and Kyunghyun Cho. SearchQA: A +Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, newQ&Adatasetaugmentedwithcontextfromasearchen- +JackClark,ChristopherBerner,SamMcCandlish,AlecRad- gine. InarXiv,2017. 4 +ford,IlyaSutskever,andDarioAmodei. LanguageModels [20] HaoyuanGao, JunhuaMao, JieZhou, ZhihengHuang, Lei +areFew-ShotLearners. arXiv,2020. 8 Wang,andWeiXu. Areyoutalkingtoamachine? Dataset -ARYAMAN FINANCIAL SERVICES LIMITED -CERTIFICATE -To the Members of Aryaman Financial Services Ltd. -We have examined the compliance of conditions of Corporate Governance by Aryaman Financial -Services Ltd. for the year 30th June, 2005 as stipulated in clause 49 of the Listing Agreement -executed by the said Company with stock exchanges. -The Compliance of conditions of Corporate Governance is the responsibility of the Management. -Our examination was limited to procedures and implementation thereof, adopted by the Company -for ensuring the compliance of the conditions of the Corporate Governance, It is neither an audit -nor an expression of opinion on the financial statements of the Company. -In our opinion and to the best of our information and according to the explanations given to us -we invite your attention to the following. -1. Audit Committee / Remuneration Committee / Shareholders / Investors Grievance Committee as -required by clause 49 of the Listing Agreement is not formed till the end of the accounting year. -2. Means of Communication regarding publication of Quarterly results in the news papers are not -complied with. -Subject to above we certify that the Company has complied with the conditions of Corporate -Governance as stipulated in the above-mentioned Listing Agreement. -We have to state that no investor grievance is pending for a period exceeding one month against the -company as per the records maintained by the registrar & share transfer agents of the company. -We further state that such compliance is neither an assurance as to the future viability of the Company -nor the efficiency or effectiveness with which the Management has conducted the affairs of the -Company. -For and on behalf of -Thakur Vaidyanath Aiyar & Co. -Chartered Accountants. -C.V.Parameswar -Partner. -Date : 1st December, 2005 M.No.11541 -Place : Mumbai +andmethodsformultilingualimagequestion. InNeurIPS, [34] YuanzeLin,YujiaXie,DongdongChen,YichongXu,Chen- +| 2015. | 3 | | | | | | guangZhu,andLuYuan. | | | Revive:Regionalvisualrepresen- | | | +| ----- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --------------- | ------------------------------ | --------------- | ------- | +| | | | | | | | tation matters | in | knowledge-based | | visual question | answer- | +[21] YashGoyal,TejasKhot,DouglasSummers-Stay,DhruvBa- +tra, and Devi Parikh. Making the V in VQA matter: Ele- ing. arXivpreprintarXiv:2206.01201,2022. 9,13,14 +vating the role of image understanding in Visual Question [35] Mateusz Malinowski and Mario Fritz. A multi-world ap- +Answering. InCVPR,2017. 3,7,13 proachtoquestionansweringaboutreal-worldscenesbased +InNeurIPS,2014. +[22] Liangke Gui, Borui Wang, Qiuyuan Huang, Alex Haupt- onuncertaininput. 3 +mann, Yonatan Bisk, and Jianfeng Gao. KAT: A Knowl- [36] KennethMarino,XinleiChen,DeviParikh,AbhinavGupta, +edgeAugmentedTransformerforVision-and-Language. In and Marcus Rohrbach. KRISP: Integrating Implicit and +NAACL,2022. 1,4,7,9,13,14 Symbolic Knowledge for Open-Domain Knowledge-Based +[23] KelvinGuu,KentonLee,ZoraTung,PanupongPasupat,and VQA. InProc.CVPR,2021. 1,4 +Ming-WeiChang.REALM:Retrieval-AugmentedLanguage [37] Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and +ModelPre-Training. InProc.ICML,2020. 1,4 RoozbehMottaghi. OK-VQA:Avisualquestionanswering +| [24] Yaru | Hao, | Haoyu Song, | Li | Dong, Shaohan | Huang, | Zewen | | | | | | | +| ------------------------------------ | ---- | ----------- | --- | ------------- | ------ | -------- | ---------- | --------- | -------- | ---------- | --- | ----------- | +| | | | | | | | benchmark | requiring | external | knowledge. | In | CVPR, 2019. | +| Chi,WenhuiWang,ShumingMa,andFuruWei. | | | | | | Language | 1,2,3,7,13 | | | | | | +modelsaregeneral-purposeinterfaces. InarXiv,2022. 1,2, [38] TodorMihaylov,PeterClark,TusharKhot,andAshishSab- +3 +| | | | | | | | harwal. | Can a | suit of armor | conduct | electricity? | a new | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----- | ------------- | ------- | ------------ | ----- | +[25] YushiHu,HangHua,ZhengyuanYang,WeijiaShi,NoahA +datasetforopenbookquestionanswering.InEMNLP,2018. +| Smith, | and | Jiebo Luo. | Promptcap: | Prompt-guided | | task- | 4 | | | | | | +| --------------------- | --- | ---------- | ------------------------------ | ------------- | ----- | ---------- | ------------------ | ------- | ------------- | -------- | -------------- | -------------- | +| awareimagecaptioning. | | | arXivpreprintarXiv:2211.09699, | | | | | | | | | | +| | | | | | | | [39] Alec Radford, | | Jong Wook | Kim, | Chris Hallacy, | Aditya | +| 2022. | 7,9 | | | | | | | | | | | | +| | | | | | | | Ramesh, | Gabriel | Goh, Sandhini | | Agarwal, | Girish Sastry, | +| [26] Ziniu | Hu, | Ahmet | Iscen, | Chen Sun, | Zirui | Wang, Kai- | | | | | | | +| | | | | | | | Amanda | Askell, | Pamela | Mishkin, | Jack Clark, | Gretchen | +Wei Chang, Yizhou Sun, Cordelia Schmid, David A. Krueger, and Ilya Sutskever. Learning transferable visual +Ross, and Alireza Fathi. REVEAL: Retrieval-Augmented modelsfromnaturallanguagesupervision. InICML,2021. +| Visual-Language | | Pre-Training | | with Multi-Source | | Multi- | | | | | | | +| --------------- | --- | ------------ | --- | ----------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +3,9,13 +| modalKnowledgeMemory. | | | | InCVPR,2023. | 1,4,7 | | | | | | | | +| --------------------- | --- | --- | --- | ------------ | ----- | --- | --------------- | ------------------ | --- | --- | ---------- | ----------- | +| | | | | | | | [40] JackW.Rae, | SebastianBorgeaud, | | | TrevorCai, | KatieMilli- | +[27] Aman Jain, Mayank Kothyari, Vishwajeet Kumar, Preethi can,JordanHoffmann,FrancisSong,JohnAslanides,Sarah +Jyothi,GaneshRamakrishnan,andSoumenChakrabarti.Se- Henderson, Roman Ring, Susannah Young, Eliza Ruther- +| lect, | substitute, | search: | A | new benchmark | for | knowledge- | | | | | | | +| ----- | ----------- | ------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | +ford,TomHennigan,JacobMenick,AlbinCassirer,Richard +| augmentedvisualquestionanswering. | | | | InSIGIR,2021. | | 2,3, | | | | | | | +| --------------------------------- | --- | --- | --- | ------------- | --- | ---- | -------------- | --- | ------------------ | --- | --------- | ---------- | +| | | | | | | | Powell, George | | van den Driessche, | | Lisa Anne | Hendricks, | +4,5,7 +| | | | | | | | Maribeth | Rauh, | Po-Sen Huang, | | Amelia Glaese, | Johannes | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------- | --- | -------------- | -------- | +[28] ChaoJia,YinfeiYang,YeXia,Yi-TingChen,ZaranaParekh, Welbl, Sumanth Dathathri, Saffron Huang, Jonathan Ue- +HieuPham,QuocV.Le,Yun-HsuanSung,ZhenLi,andTom sato, John Mellor, Irina Higgins, Antonia Creswell, Nat +Duerig.Scalingupvisualandvision-languagerepresentation +| | | | | | | | McAleese, | Amy | Wu, Erich | Elsen, | Siddhant | Jayakumar, | +| --------------------------------- | --- | --- | --- | ------------ | --- | --- | --------- | --- | --------- | ------ | -------- | ---------- | +| learningwithnoisytextsupervision. | | | | InICML,2021. | | 3 | | | | | | | +ElenaBuchatskaya,DavidBudden,EsmeSutherland,Karen +[29] MandarJoshi,EunsolChoi,DanielS.Weld,andLukeZettle- Simonyan, MichelaPaganini, LaurentSifre, LenaMartens, +moyer. TriviaQA: A large scale distantly supervised chal- Xiang Lorraine Li, Adhiguna Kuncoro, Aida Nematzadeh, +lengedatasetforreadingcomprehension. InACL,2017. 4 Elena Gribovskaya, Domenic Donato, Angeliki Lazaridou, +[30] UrvashiKhandelwal,OmerLevy,DanJurafsky,LukeZettle- ArthurMensch,Jean-BaptisteLespiau,MariaTsimpoukelli, +moyer,andMikeLewis. Generalizationthroughmemoriza- NikolaiGrigorev,DougFritz,ThibaultSottiaux,MantasPa- +tion: Nearestneighborlanguagemodels. InICLR,2020. 1, jarskas, Toby Pohlen, Zhitao Gong, Daniel Toyama, Cy- +| 4 | | | | | | | priendeMassond’Autume,YujiaLi,TayfunTerzi,Vladimir | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------------- | --- | --- | --- | --- | --- | +Mikulik,IgorBabuschkin,AidanClark,DiegodeLasCasas, +| [31] Tushar | Khot, | Peter | Clark, | Michal Guerquin, | Peter | Jansen, | | | | | | | +| ----------- | ----- | ----- | ------ | ---------------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +and Ashish Sabharwal. QASC: A dataset for question an- Aurelia Guy, Chris Jones, James Bradbury, Matthew John- +sweringviasentencecomposition. InAAAI,2020. 4 son, Blake Hechtman, Laura Weidinger, Iason Gabriel, +[32] Alina Kuznetsova, Hassan Rom, Neil Alldrin, Jasper Ui- WilliamIsaac,EdLockhart,SimonOsindero,LauraRimell, +| | | | | | | | Chris Dyer, | Oriol | Vinyals, | Kareem | Ayoub, | Jeff Stanway, | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ----- | -------- | ------ | ------ | ------------- | +jlings,IvanKrasin,JordiPont-Tuset,ShahabKamali,Stefan +| | | | | | | | Lorrayne | Bennett, | Demis | Hassabis, | Koray | Kavukcuoglu, | +| ----------------------- | ------ | -------- | --- | --------------------------- | -------- | -------- | ------------------- | -------- | ------- | --------- | ------- | ------------ | +| Popov, | Matteo | Malloci, | Tom | Duerig, and | Vittorio | Ferrari. | | | | | | | +| | | | | | | | and GeoffreyIrving. | | Scaling | Language | Models: | Methods, | +| TheOpenImagesDatasetV4: | | | | Unifiedimageclassification, | | | | | | | | | +object detection, and visual relationship detection at scale. Analysis&InsightsfromTrainingGopher. arXiv,2022. 8 +IJCV,2020. 4 [41] ColinRaffel,NoamShazeer,AdamRoberts,KatherineLee, +SharanNarang,MichaelMatena,YanqiZhou,WeiLi,,and +| [33] Patrick | Lewis, | Ethan | Perez, | Aleksandra | Piktus, | Fabio | | | | | | | +| ------------ | ------ | ----- | ------ | ---------- | ------- | ----- | --- | --- | --- | --- | --- | --- | +Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich PeterJLiu. Exploringthelimitsoftransferlearningwitha +Ku¨ttler, Mike Lewis, Wen tau Yih, Tim Rockta¨schel, Se- unifiedtext-to-texttransformer. JMLR,2021. 13 +bastian Riedel, and Douwe Kiela. Retrieval-Augmented [42] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and +Generation for Knowledge-Intensive NLP Tasks. In Proc. PercyLiang.SQuAD:100,000+questionsformachinecom- +| NeurIPS,2020. | | 1,4 | | | | | prehensionoftext. | | InEMNLP,2016. | | 4 | | +| ------------- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------- | --- | --- | --- | -1= ARYAMANFINANCIALSERVICES LIMITED -CORPORATE GOVERNANCE REPORT - 2005 -PHILOSOPHY ON CORPORATE GOVERNANCE : -Board of Directors of the Company acts as a trustee and assumes fiduciary responsibility of -protecting the interests of the shareholders and other stakeholders of the Company. Board -supports the broad principles of Corporate Governance. In order to attain the highest level of -good Corporate Governance practice, Board lays strong emphasis on transparency, -accountability and integrity. -Board of Directors -Composition and category of directors; attendance of each director at the board meetings and -the last AGM; number of other Public Companies on the Board or Committees of which a -director is a member or chairperson; and number of board meetings, dates on which held, are -as under: -Board Meeting Attended during the Year 2004-05 Attendance at the Last AGM -Name of Director Category (attended) -Shri PVR Murthy* E & NI 3 Yes -Shri Virendra Ganda NE & I 8 Yes -Shri Mohan Dattary NE & 1 8 Yes -Shri Munnangi V Ramana** NE & I 5 N.A. -Shri Suhas Ganpule*** NE & II N.A. -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -* Resigned w.e.f. December 24, 2004 , -** Appointed w.e.f December 20, 2004 -*** Appointed w.e.f. May 02, 2005 -8 Meetings of the board of directors were held during the financial year 2004-05. One each on -April 30,2004, July 29,2004, October 28,2004, December 20,2004, December 24,2005, January -28, 2005, March 22, 2005 and May 18, 2005 -AUDIT COMMITTEE -Terms of reference : -The terms of reference stipulated by the Board of Directors to the Audit Committee are as -contained under clause 49 of the Listing Agreement. -Composition: -The Audit Committee was constituted by the Board of Directors in the meeting held on -December 01, 2005. The composition of the Committee and attendance of the members is -given hereunder: +[43] NilsReimersandIrynaGurevych. Sentence-bert: Sentence [60] Lu Yuan, Dongdong Chen, Yi-Ling Chen, Noel Codella, +embeddingsusingsiamesebert-networks. InEMNLP,2019. Xiyang Dai, Jianfeng Gao, Houdong Hu, Xuedong Huang, +| 7 | | | | | | | | BoxinLi,ChunyuanLi,CeLiu,MengchenLiu,ZichengLiu, | | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | +[44] MengyeRen,RyanKiros,andRichardS.Zemel. Exploring YumaoLu,YuShi,LijuanWang,JianfengWang,BinXiao, +modelsanddataforimagequestionanswering. InNeurIPS, ZhenXiao,JianweiYang,MichaelZeng,LuoweiZhou,and +| 2015. | 4 | | | | | | | PengchuanZhang. | Florence: ANewFoundationModelfor | +| ------------ | ------ | ---------- | --- | ------------------ | --- | --- | ---------- | --------------- | -------------------------------- | +| | | | | | | | | ComputerVision. | arXiv,2021. 3 | +| [45] Eleanor | Rosch. | Principles | | of categorization. | | | In Eleanor | | | +RoschandB.B.Lloyd,editors,CognitionandCategoriza- +| tion,pages27–48.LawrenceErlbaumAssociates,1978. | | | | | | | 2 | | | +| ----------------------------------------------- | -------- | ------ | ------- | ----------- | --- | ----------- | ------ | --- | --- | +| [46] Dustin | Schwenk, | Apoorv | | Khandelwal, | | Christopher | Clark, | | | +| Kenneth | Marino, | and | Roozbeh | Mottaghi. | | A-OKVQA: | A | | | +benchmarkforvisualquestionansweringusingworldknowl- +| edge. | InECCV,2022. | | 2,3,7,13 | | | | | | | +| ------------------------ | ------------ | --------------------------------------- | ------------ | ------------ | -------------------- | ------- | --------- | --- | --- | +| [47] Sanket | Shah, | Anand | Mishra, | | Naganand | Yadati, | and | | | +| Partha | Pratim | Talukdar. | KVQA: | | Knowledge-aware | | visual | | | +| questionanswering. | | | InAAAI,2019. | | 2,3,7 | | | | | +| [48] Krishna | Srinivasan, | | Karthik | Raman, | Jiecao | Chen, | Michael | | | +| Bendersky,andMarcNajork. | | | | Wit: | Wikipedia-basedimage | | | | | +| text | dataset | for multimodal | | multilingual | | machine | learning. | | | +| SIGIR,2021. | | 3,4 | | | | | | | | +| [49] AndyStirling. | | Ageneralframeworkforanalysisngdiversity | | | | | | | | +| in science, | | technology, | and | society. | Journal | of | the Royal | | | +| SocietyInterface,2007. | | | 7 | | | | | | | +[50] AlonTalmorandJonathanBerant.Thewebasaknowledge- +| baseforansweringcomplexquestions. | | | | | InNAACL,2018. | | 4 | | | +| --------------------------------- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | +[51] GrantvanHorn,ElijahCole,SaraBeery,KimberlyWilber, +| SergeBelongie,andOisinMacAodha. | | | | | BenchmarkingRep- | | | | | +| ------------------------------- | --- | -------- | ----------- | ----- | ---------------- | ------------ | --- | --- | --- | +| resentation | | Learning | for Natural | World | Image | Collections. | | | | +| InCVPR,2021. | | 2,4 | | | | | | | | +[52] PeterVickers,NikolaosAletras,EmilioMonti,andLoicBar- +| rault. | InFactuality: | | EfficientIntegrationofRelevantFacts | | | | | | | +| --------------------------- | ------------- | --- | ----------------------------------- | ----------- | --- | --- | --- | --- | --- | +| forVisualQuestionAnswering. | | | | InACL,2021. | | 1,4 | | | | +[53] PengWang,QiWu,ChunhuaShen,AnthonyDick,andAn- +| tonvandenHengel. | | | FVQA:Fact-basedvisualquestionan- | | | | | | | +| ---------------- | ---------------------- | ------ | -------------------------------- | ---- | ----------- | --- | ------- | --- | --- | +| swering. | IEEETrans.onPAMI,2018. | | | | 3,7 | | | | | +| [54] Jason | Wei, | Xuezhi | Wang, | Dale | Schuurmans, | | Maarten | | | +Bosma,BrianIchter,FeiXia,EdChi,QuocLe,andDenny +| Zhou. | Chain-of-Thought | | | Prompting | Elicits | Reasoning | in | | | +| -------------------- | ---------------- | ------ | --------------- | --------- | ------- | --------- | ------- | --- | --- | +| LargeLanguageModels. | | | InNeurIPS,2022. | | | 14 | | | | +| [55] Johannes | Welbl, | Pontus | Stenetorp, | | and | Sebastian | Riedel. | | | +Constructingdatasetsformulti-hopreadingcomprehension +| acrossdocuments. | | | InACL,2018. | | 4 | | | | | +| ---------------- | --- | ---------- | ----------- | -------- | ------- | ------ | ----- | --- | --- | +| [56] T. Weyand, | | A. Araujo, | B. | Cao, and | J. Sim. | Google | land- | | | +marksdatasetv2-alarge-scalebenchmarkforinstance-level +| recognitionandretrieval. | | | InCVPR,2020. | | | 2,4 | | | | +| ------------------------ | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | +[57] ZhengyuanYang,ZheGan,JianfengWang,XiaoweiHu,Yu- +| maoLu,ZichengLiu,andLijuanWang. | | | | | | Anempiricalstudy | | | | +| ------------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | +| ofgpt-3forfew-shotknowledge-basedvqa. | | | | | | InAAAI,2022. | | | | +1,2,3 +| [58] Zhilin | Yang, | Peng | Qi, Saizheng | | Zhang, | Yoshua | Bengio, | | | +| ------------------------------- | ------ | ------------ | -------------- | ------------- | ------------ | ----------- | ----------- | --- | --- | +| William | Cohen, | Ruslan | Salakhutdinov, | | and | Christopher | D. | | | +| Manning. | | HotpotQA: | A | dataset | for diverse, | | explainable | | | +| multi-hopquestionanswering. | | | | InEMNLP,2018. | | | 3,4,6 | | | +| [59] Licheng | Yu, | Eunbyung | Park, | Alexander | | C. | Berg, and | | | +| Tamara | L. | Berg. Visual | madlibs: | | Fill in | the blank | image | | | +| generationandquestionanswering. | | | | | InICCV,2015. | | 3 | | | -r ARYAMANFINANCIALSERVICES LIMITED -| Name of Member | Member/ Chairman | Category | -| ----------------------- | ----------------- | -------- | -| Shri Baliga | Chairman | NE & I | -| Shri Mohan Datari | Member | NE & I | -| Shri Munnangi V Ramana | Member | NE & I | -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -The Company Secretary of the Company, Shri Prasanna Chandwaskar is the Secretary of the -Committee. Director Finance, Internal Auditors and Statutory Auditors are the permanent invitees -to the Audit Committee Meeting. -REMUNERATION COMMITTEE -The Remuneration Committee was constituted by the Board of Directors in its meeting held on -December 01, 2005. No meeting of the Committee was held during the year. The terms of -reference, compositions are as under: -Terms of reference: To recommend appointment of Managerial Personnel and consider the -remuneration payable to them and review thereof from time to time. -Composition: -| Name of Member | Member/ Chairman | | -| ----------------------- | ---------------- | --- | -| Shri Mohan Datari | Chairman | | -| Shri Baliga | Member | | -| Shri Munnangi V Ramana | Member | | -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -REMUNERATION TO EXECUTIVE DIRECTORS: -The details of remuneration paid/ payable to the directors for the Financial Year 2004-05 are as under: -Shri PVR Murthy - Managing Director (Resigned w.e.f. December 24, 2004) -(Rs.) -| Salary & Allowances | 60000/- | | -| -------------------- | ------- | --- | -| Perquisites | 42000/- | | -Remuneration to Non-Executive Directors: -The Non-executive directors do not draw any remuneration from the Company. Non-executive -independent directors are entitled to the sitting fees for every meeting of the Board and Audit -Committee attended. -Non-executive independent directors are paid Rs. 1500/- for attending Board Meeting and Rs. -1OOO/- for Audit Committee. +A.Correctnessofourdataset astrictevaluationmeasureandbecausetheirquestionsare +| | | | | | | | more | open-ended | | (they | are not supported | | by a controlled | | +| ------- | --- | ------ | --------- | ------- | ------- | --------- | -------------- | ---------- | --- | ------- | ----------------- | --------- | --------------- | ------- | +| We rely | on | iNat21 | and GLDv2 | to have | clearly | identifi- | | | | | | | | | +| | | | | | | | knowledgebase, | | | Sec.4), | thosedatasets | collect10 | | ground- | +ablecategoriesonthetestimagesofourdataset,andtoob- +truthanswersperquestion,tocoversomeofthevariability +taintheirground-truthlabels.Furthermore,weonlyusecat- in answer formulation. This then enables a more relaxed +egorieswhichunambiguouslymaptoWikipediaarticlesby +| | | | | | | | evaluation, | | as the | model | answer need | only | match | some of | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ----------- | --------------- | ------ | ----- | -------------- | ---- | ------------- | ------- | +| explicitlyseekingforone-to-onemappings(Sec.4.1). | | | | | | The | | | | | | | | | +| | | | | | | | the | 10 ground-truth | | ones | (three matches | | for a perfect | an- | +mainremainingpossiblesourceoferroriswhetheraques- +swer). Nevertheless,itisinstructivetoverifywhetherexact +| tion can | be answered | | given | the corresponding | | Wikipedia | | | | | | | | | +| -------- | ----------- | --- | ----- | ----------------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +matchingwouldworkonourdataset,wherewehaveasin- +| article, | and whether | | the recorded | ground-truth | | answer is | | | | | | | | | +| -------- | ----------- | ----- | ------------ | -------------- | --- | --------- | --- | ------------ | --- | ------- | -------- | -------- | -------- | ------ | +| | | | | | | | gle | ground-truth | | answer. | We start | from the | publicly | avail- | +| indeed | correct. | Human | annotators | systematically | | vali- | | | | | | | | | +ableexactmatchingimplementationof[22]andincludead- +| date these | aspects | for | every | single question+answer | | pair | | | | | | | | | +| ---------------- | ------- | --------------------------------- | ----- | ---------------------- | --- | ---- | -------------------------------------------- | --- | --------- | ----- | -------- | ------ | ----------- | --- | +| | | | | | | | ditionalrelaxationsfornumbercomparisons(e.g. | | | | | | numbers | | +| (Sec.4.2and4.3). | | Yet,weperformhereanadditionaluser | | | | | | | | | | | | | +| | | | | | | | over | 100 | - usually | years | - may be | off by | one, ranges | are | +studytogetannumericalestimateofqualityinthissense. +| | | | | | | | correct | if | they | partially | overlap, etc). | We | find that | exact | +| --- | --- | --- | --- | --- | --- | --- | ------- | --- | ---- | --------- | -------------- | --- | --------- | ----- | +In this study, we randomly sample 100 questions from matchjudgementsareequivalenttohumanjudgementsonly +thetestset.Thenweasksixexpertstoeachanswer50ques- +| | | | | | | | in | 68% of | the cases. | Where | they | differ, exact | match | is al- | +| --- | --- | --- | --- | --- | --- | --- | --- | ------ | ---------- | ----- | ---- | ------------- | ----- | ------ | +tionseach,giventhecorrespondingWikipediapageasref- +| | | | | | | | ways | overly | strict, | rejecting | answers | that | a human | would | +| --- | --- | --- | --- | --- | --- | --- | ---- | ------ | ------- | --------- | ------- | ---- | ------- | ----- | +erence. Theseexpertwerenotinvolvedintheoriginaldata judge as correct. This demonstrates that exact matching +collectionprocess.Thisprocessresultsinthreeanswersper +doesnotworkwellforourdataset,andjustifiesourchoice +question. If the majority of the 3 expert answers matches of BEM as the evaluation measure to check whether pre- +ourcollectedground-truthaccordingtoBEM[12],wecon- +| | | | | | | | dictedmodelanswersmatchground-truthonesinSec. | | | | | | | 5. | +| ---------- | -------- | --- | ------------- | ----- | --- | --------- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | +| sider that | question | to | be answerable | given | the | Wikipedia | | | | | | | | | +page, andourground-truthto becorrect. Wefind thatthis C. CLIP retrieval in existing retrieval- +holdsformostofourquestions(86%). +augmentedVQAsystems[22,34] +| To put | this number | | in context, | we also | estimate | the an- | | | | | | | | | +| ------ | ----------- | --- | ----------- | ------- | -------- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +swerabilityofA-OKVQA[46]withrespecttotheirevalua- KAT[22]andREVIVE[34]aretworetrieval-augmented +tionmetric.A-OKVQAfollowspreviousVQAdatasetsand VQA systems which use frozen CLIP [39] embeddings to +provides10ground-truthanswersperquestion. Apredicted performretrieval. Morespecifically,theyfirstencodetheir +answer is counted as correct if it matches with 3 ground- text-only knowledge base (extracted from Wikidata) using +truth answers (exact string matching [8, 46], after normal- thelanguagetowerofCLIP.Attesttime,givenaVQAques- +izationsuchasremovingpunctuationandarticles,convert- tion,theyencodeitsimageI withCLIPandthencompare +ingnumberstodigits,etc.). Sowecanconsideraquestion itsembeddingtotheknowledgebaseembeddingstoretrieve +is answerable if there are at least 3 equivalent answers out the most similar entries. The top few most similar entries +ofthe10.Wefindthat86%oftheA-OKVQAquestionsare are then passed on to a T5 model [41] which is trained to +answerable. +| | | | | | | | producethefinalanswer. | | | | Notehowcorrectretrievaliscru- | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | ----------------------------- | --- | --- | --- | +To conclude, this user study demonstrates that our cialtoperformwellonEncyclopedic-VQA,astheretrived +Encyclopedic-VQAdatasetisofveryhighquality. knowledge base entries should contain the answer for the +| | | | | | | | overall | system | | to succeed. | Hence, | we now | test the | CLIP- | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------ | --- | ----------- | ------ | ------ | -------- | ----- | +B.QualityofBEM[12]evaluationmeasure based retrieval component of KAT [22] and REVIVE [34] +inisolation,toestimatewhethertheywouldbeabletosuc- +| We want | to | understand | how | well the | BERT | Matching | | | | | | | | | +| ------- | --- | ---------- | --- | -------- | ---- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +ceedonEncyclopedic-VQA. +(BEM)[12]evaluationmeasuremirrorshumanjudgments. WerepresentWikipediaarticlesastheCLIPembedding +| To do so | we build | on | the user | study above | and | ask an ex- | | | | | | | | | +| -------- | -------- | --- | -------- | ----------- | --- | ---------- | --- | ----- | --------- | ----------- | ------- | ------------- | --- | ------- | +| | | | | | | | of | their | title and | description | strings | concatenated. | | At test | +pert human to judge whether each answer from the user time, given a VQA question, we use CLIP to embed the +studymatchesthecollectedground-truthanswerornot.The +judgehasaccesstoboththequestionQandthecorrespond- +| ing Wikipedia | | page. | We find | that the | BEM | judgements | | | | | Recall | | | | +| ------------- | ---------- | ----- | ------- | ----------------- | --- | ------------ | --- | ------ | --- | --- | ------ | --- | --- | --- | +| equal human | judgements | | in | the vast majority | | of the cases | | | | | | | | | +| | | | | | | | | Method | | @1 | @5 | @10 | @20 | | +(96%). BEMisastricterjudgein3%ofthecases,thehu- +man in 1%. This demonstrates that BEM correlates very CLIP[39] 3.3% 7.7% 12.1% 16.5% +highlywithhumanjudgementofcorrectnessofananswer, Lens[1] 47.4% 62.5% 64.7% 65.2% +confirmingwhatreportedin[12]forotherdatasets. +Exact Match. Many VQA datasets perform exact string Table7: RecallresultsforCLIPandLens.Wereporttherecall +inretrievingtherightKBarticlewithinthetop-Kdocuments. +| matchingfortheirevaluation[8,21,37,46]. | | | | | | Becauseitis | | | | | | | | | +| --------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | -T ARYAMANFINANCIALSERVICES LIMITED -SHARE TRANSFER AND INVESTOR GRIEVANCE COMMITTEE: -The Share Transfer Committee was constituted by the Board of Directors at its meeting held -on December 01, 2005. The Committee consist of Shri Munnangi V Ramana and Shri -Baliga, all being a non-executive director. The Committee meets once in a fortnight. -Compliance Officer: Shri Prasanna Chandwaskar, Company Secretary. -Number of Shareholders complaints received during the year : -During the Year under review, total no Investor complaints were received and no complaint is pending -as at November 30, 2005 for a period over 30 days. -MANAGEMENT: -a. Management Discussion and Analysis -Management Discussion and Analysis of various businesses of the Company is separately given -in the Annual Report. -b. Disclosures by management to the Board -All details relating to financial and commercial transactions where directors may have a pecuniary -interest are provided to the Board, and the interested directors neither participate in the discussion, -nor do they vote on such matters. -GENERAL BODY MEETINGS: -The details of Annual General Meetings held in the last three years are given hereunder. -Meeting Day & Date of Meeting Time Place -10* Annual Gen. Meeting Thursday, September 30,2004 11.30a.m. 401. Chirnjiv Towers. -43. Nehru Place. -New Delhi-1)0 0)9 -9th Annual General Meeting Tuesday, September 30,2003 11.30 a.m. A- 293. Shivalik. Geetanjali -Panchsheel Road. -New Delhi-1100! 7 -8ch Annual General Meeting Monday, December 30,2002 11.30a.m. B-47. Shivalik, Geetanjali -Panchsheel Road. -New Delhi-110017 -DISCLOSURE -There are no related party transactions. -Details of information on appointment of new directors -Reappointment Directorship held in other Companies Committee positions in various Companies -Name of the Director Appointment -Shri Munnangi V Ramana December 20, 2004 -Shri Suhas Ganpule May 02, 2005 +imageItoformthequeryforretrieval,proceedingasin[22, E.DatasetStatistics +| 34]. Results | in | Tab. 7 demonstrate | | that recall | is | low: the | | | | | | | | | +| ------------ | --- | ------------------ | --- | ----------- | --- | -------- | --- | -------- | ---- | ------- | ------ | -------- | -------- | ------ | +| | | | | | | | | In Table | 8 we | provide | a more | detailed | overview | of the | +correctWikipediaarticlecorrespondingtothesubjectofthe +| | | | | | | | datasetstatistics. | | | FormoredetailsseeSection4ofthemain | | | | | +| -------- | ------------ | ------ | -------------- | ---- | ---- | ------ | ------------------ | --- | --- | ---------------------------------- | --- | --- | --- | --- | +| question | is retrieved | in the | first position | only | 3.3% | of the | | | | | | | | | +paper. +| time. Even | within | the top-20 | results, | retrieval | accuracy | is | | | | | | | | | +| ----------------- | ------ | ------------------------------ | -------- | --------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| stillonlyat16.5%. | | ThissuggestthatKAT/REVIVEwould | | | | | | | | | | | | | +F.Promptsfordatasetcreation +notworkwellonourdataset. +Lens retrieval. In Sec. 5.3 we use Google Lens to iden- AsdiscussedinSection4ofthepaper,weusedaFLAN +tify the subject C of a VQA question in our datatset, i.e. [18] version of the PaLM [17] model to (1) rephrase au- +a iNat21 or GLDv2 category, which works well (Tab. 7). tomatically generated single-hop questions (both single- +Lens is an image-based system which indexes billions of answer and multi-answer), and to (2) combine two single- +| images on | the web | and retrieves | | relevant | ones | based on | | | | | | | | | +| --------- | ------- | ------------- | --- | -------- | ---- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +hopquestionsintoatwo-hopquestion.Generally,wefound +their visual similarity. Hence, it has a greater chance to that the model tends to work better when fed with very +findawebimageresemblingaVQAtestimage,thanwhen specific(andsometimesredundant)instructions,combined +restricting the search only to Wikipedia. Attached to the withchain-of-thoughtprompts[54]wherepossible. Addi- +image is often various meta-data which enables to deter- tionally,weaddeddifficultcasesasexamplesintheprompt, +minethenameofthesubjectcategory(whichwethenuseto +| | | | | | | | to | guide | the model’s | behavior | | in more | challenging | situa- | +| --------------------------- | --- | ------------- | ----------------------- | ------- | --------- | ----- | ------ | ----- | ----------- | -------- | --- | ------- | ----------- | ------ | +| findtherightWikipediapage). | | | Recognizinglandmarksand | | | | tions. | | | | | | | | +| species from | the | natural world | are | typical | use-cases | so we | | | | | | | | | +F.1.Rephrasingautomaticallygeneratedsingle-hop +| can expect | Lens | to work well | for | recognizing | the | subjects | | | | | | | | | +| ---------- | ---- | ------------ | --- | ----------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +questions +| C inourdataset. | | However,academicresultsoniNat21[4] | | | | | | | | | | | | | +| --------------- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +andGLDv2[2]suggeststhatspecializedsystemsworkvery +| | | | | | | | | We replaced | | the name | of the | category | C | in the ques- | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --- | -------- | ------ | -------- | --- | ------------ | +wellonthesedatasets,suggestingthatagoodimage-based tionbyitssupercategory. Thisrephrasingprocesswasrun +classifierwouldmakeaviablesubstituteforLens. +forallautomaticallygeneratedsingle-hopquestions(single +Generally, the core points of our paper are that (1) En- andmulti-answer). Forexample,thequestionWhereisthe +| cyclopedic-VQA | | poses a | hard challenge | | for large | LLMs | E´glise | | | | | | | | +| -------------- | --- | ------- | -------------- | --- | --------- | ---- | ------- | ------------ | --- | -------- | --- | ------------- | --- | ----------- | +| | | | | | | | | Saint-Cannat | | located? | | was rephrased | | to Where is | +and VLMs, and (2) solving it requires augmenting the thischurchlocated? Afterseveraliterations,wedecidedto +LLM/VLM with a retrieval component to access a knowl- add three examples in the prompt, using chain-of-thought +| edgebase. | Theexactchoiceofretrievalcomponentisflex- | | | | | | | | | | | | | | +| --------- | ----------------------------------------- | --- | --- | --- | --- | --- | --------- | --- | ---- | ------------ | --- | ----- | --------- | ------ | +| | | | | | | | reasoning | | with | three steps. | The | model | generally | writes | +ibleandlikelysubjecttofurtherexploration. the three required steps in the output, and we parse the +| | | | | | | | rephrased | | question | right | after the | third | step written | by the | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | -------- | ----- | --------- | ----- | ------------ | ------ | +D.Qualitativeresults model. Sometimes, the model would return an incorrectly +| | | | | | | | rephrased | | question. | We observe | | that the | main | failure case | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --------- | ---------- | --- | -------- | ---- | ------------ | +We present qualitative results for some of the methods is when the model simply copies the input question to the +we experimented with in Figures 6 and 7. These illustrate output. Thus, we filter out any rephrased question that is +caseswherethedifferentsetupsmaywork,andcaseswhere +| | | | | | | | identical | | to the | original question | | fed in | its input. | We sam- | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ------ | ----------------- | --- | ------ | ---------- | ------- | +allofthemfail. Forexample,givenatextualquestionWhat pled100rephrasedquestionstomanuallyassessthequality +does this reptile eat?, PaLM can reasonably guess an an- of the rephrasings, which we found to be correct in 90% +swer without any additional context: insects – since that’s ofthecases. ThefinalpromptisgiveninPrompt1,where +whatmanyreptileseat. “$C$” and “$Q$” are placeholders for the category C and +thetextual(pre-rephrasing)questionQ. +Formorespecificinformation,relatingtomoredetailed +properties(e.g.,numberofeggsaspecificreptilelays,dates, +| | | | | | | | F.2. | Chaining | | single-hop | questions | | into | two-hop | +| --- | --- | --- | --- | --- | --- | --- | ---- | -------- | --- | ---------- | --------- | --- | ---- | ------- | +howbigaspecificfishmaybecome),largemodelsmaystill +questions +| makereasonableguesses, | | | butgenerallythisleadstoincor- | | | | | | | | | | | | +| ---------------------- | ------------------------------------ | --- | ----------------------------- | --- | --- | --- | --- | ---------- | --- | ----------------- | --- | ----------- | --- | ----------- | +| rectanswers. | Augmentinglargemodelswithcontextfrom | | | | | | | | | | | | | | +| | | | | | | | | We created | | two-hop questions | | by chaining | | two single- | +retrieved knowledge improves the accuracy in these cases hopquestions,wheretheanswertothefirstsingle-hopques- +substantially,leadingtopreciseanswersthatareattributable tionservesasabridgeentitythatmakesaconnectiontothe +tothepieceofknowledgethatwasretrieved. +| | | | | | | | second | single-hop | | question. | For | example, | given | the first | +| --- | --- | --- | --- | --- | --- | --- | ------ | ---------- | --- | --------- | --- | -------- | ----- | --------- | +Finally, we also see cases where the generated answer single-hopquestion(SQ1)Whatisthemaincompetitorfor +is incorrect, even if the correct piece of knowledge is re- food for this animal? with answer (SA1) Spotted hyena, +trieved. Thisindicatesthatinsomecaseslargemodelsmay andthesecondsingle-hopquestion(SQ2)Whatisthepop- +stillhavedifficultiesusingretrievedknowledgetogenerate ulation size of this animal? with answer (SA2) Between +accurateanswers. 27,000 and 47,000 individuals, we generated What is the -F= ARYAMAN FINANCIAL SERVICES LIMITED -Details of non-compliance -There was no non compliance by the Company, nor there was any penalties or strictures -imposed on the company by the Stock Exchange or SEBI or any authority on any matter -related to capital market during the last 3 years. -MEANS OF COMMUNICATION -The quarterly results are not published in the News papers -Management Discussion and Analysis form part of this Annual Report. -GENERAL SHAREHOLDER INFORMATION -AGM will be held on Thursday, December 29, 2005 at 10.00 a.m. at the registered office. -Financial Year of the Company : April 01, 2004 to June 30, 2005 -Date of Book Closure: December 26, 2005 to December 29, 2005 (Both days inclusive) -Dividend payment date: Not applicable. -Listing on Stock Exchanges: Mumbai Stock Exchange Limited, Delhi Stock Exchange and -Ahmedabad Stock Exchange Limited. -| Stock Code: | Bombay Stock Exchange Limited | | : 530245 | | -| ------------ | ------------------------------ | --- | -------- | --- | -Market Price Data: High & Low price of equity shares on Stock Exchange shares of the Company -are infrequently traded on the exchanges. -| Month | | High Amount (Rs.) | Low Amount (Rs.) | | -| ----------- | --- | ------------------ | ---------------- | ---- | -| April-2004 | | 1.21 | | 0.82 | -May-2004 -| June-2004 | | 0.80 | | 0.69 | -| ---------- | --- | ----- | --- | ---- | -July-2004 -| August - 2004 | | 0.69 | | 0.63 | -| ----------------- | --- | ----- | --- | ---- | -| September - 2004 | | 0.82 | | 0.75 | -| October - 2004 | | 1.70 | | 0.90 | -| November - 2004 | | 2.02 | | 2.01 | -| December - 2004 | | 2.69 | | 1.93 | -| January - 2005 | | 4.60 | | 2.76 | -| February - 2005 | | 5.50 | | 2.10 | -| March-2005 | | 3.10 | | 2.00 | -| April-2005 | | 3.96 | | 2.01 | -| May-2005 | | 7.32 | | 3.15 | -| June-2005 | | 6.90 | | 4.06 | -Performance in comparison to broad-based indices i.e. BSE Sensex is as under: +Qualitative results: PaLI +| | In which direction from the | What is Mary handling | | | | | | | +| --- | ---------------------------- | ---------------------- | --- | --- | --- | --- | --- | --- | +In what month of 1944 was this What was this church fitted with +| Question | german-czech border is this | St.Dominic in the facade of this | | | | | | | +| -------- | ---------------------------- | --------------------------------- | --------- | --- | ----------------- | --- | -------- | --- | +| | | | | | church destroyed? | | in 1914? | | +| | mountain located? | | building? | | | | | | +No +| | north | | child | | december | | window | | +| --- | ----- | --- | ----- | --- | -------- | --- | ------ | --- | +Retrieval +Lens +| | north | | rosary | | december | | bell | | +| --- | ----- | --- | ------ | --- | -------- | --- | ---- | --- | +Section +Oracle +| | north | | rosary | | may | | steeple | | +| --- | ----- | --- | ------ | --- | --- | --- | ------- | --- | +Section +Ground +| | north | | the rosary | | may | | electricity | | +| --- | ----- | --- | ---------- | --- | --- | --- | ----------- | --- | +Truth +Figure 6: +PaLIqualitativeresults. Wepresent4multi-modalquestionsandtheanswersproducedwithdifferentexperimentalsetups, +usingthePaLImodel: withoutretrieval(“NoRetrieval”), Lens-basedretrievalofKBsection(“LensSection”), OracleretrievalofKB +secQtionu(“aOrlaictleaSetcitvione”) .rAeddsitiounalltlys,w:e Fprolvaidenth-ePgroaunLd-tMruthanswerinthelastrow.Weshowcaseswhereeachofthesetupscan +producethecorrectanswer,aswellasanexamplewhereallmethodsfail(theright-mostone). +| | | How many eggs does this | | | | How big does this fish typically | | | +| --- | --- | ------------------------ | --- | --- | --- | --------------------------------- | --- | --- | +Question What does this reptile eat? What does this animal eat? +| | | reptile typically lay? | | | | | become? | | +| --- | --- | ---------------------- | --- | --- | --- | --- | ------- | --- | +No +| | insects | | 10-15 | | insects | | 6 cm | | +| --- | ------- | --- | ----- | --- | ------- | --- | ---- | --- | +Retrieval +Lens +| | insects | | 3-6 | | plankton | | 13 cm | | +| --- | ------- | --- | --- | --- | -------- | --- | ----- | --- | +Section +| Oracle | insects | | 3-6 | | mussels | | 13 cm | | +| ------- | ------- | --- | --- | --- | ------- | --- | ----- | --- | +Section +Ground +| | insects | | 3-6 | | mussels | | 25 cm | | +| --- | ------- | --- | --- | --- | ------- | --- | ----- | --- | +Truth +Figure7: PaLMqualitativeresults. Wepresent4multi-modalquestionsandtheanswersproducedwithdifferentexperimentalsetups, +usingthePaLMmodel: withoutretrieval(“NoRetrieval”),Lens-basedretrievalofKBsection(“LensSection”),OracleretrievalofKB +section(“OracleSection”).Additionally,weprovidetheground-truthanswerinthelastrow.Weshowcaseswhereeachofthesetupscan +producethecorrectanswer,aswellasanexamplewhereallmethodsfail(theright-mostone). +| population | size of the main competitor | for food | of this an- | individuals. | | | | | +| ---------- | --------------------------- | -------- | ----------- | ------------ | --- | --- | --- | --- | +imal? Inthisexample,theentityspottedhyenaservesasa +| | | | | After | several iterations, | we crafted | a prompt | with 4 ex- | +| --------------------------------------- | --- | --- | ----------- | ----- | ------------------- | ---------- | -------- | ---------- | +| bridgebetweenthetwosingle-hopquestions. | | | Notethatthe | | | | | | +answertothetwo-hopquestionisidenticaltotheanswerto amplesofvaryingdifficulty(Prompt2). Theoutputofthe +| | | | | model | is taken as the chained | two-hop | question. | Some- | +| --- | --- | --- | --- | ----- | ----------------------- | ------- | --------- | ----- | +thesecondsingle-hopquestion,Between27,000and47,000 +times,though,themodelwouldreturnanincorrecttwo-hop -| r ARYAMANFINANCIALSERVICES LIMITED | | | = | -| ----------------------------------- | ---------- | ----------------------------- | ---- | -| Month | BSE Index | Closing price of Shares (Rs.) | | -| April - 2004 | 5655.09 | | 0.82 | -| May - 2004 | 4759.62 | | | -| June - 2004 | 4795.46 | | 0.69 | -| July-2004 | 5170.32 | | | -| August-2004 | 5192.08 | | 0.69 | -| September - 2004 | 5583.61 | | 0.82 | -| October - 2004 | 5672.27 | | 0.69 | -| November - 2004 | 6234.29 | | 2.02 | -| December - 2004 | 6602.69 | | 2.31 | -| January-2005 | 6555.94 | | 3.83 | -| February - 2005 | 6713.86 | | 2.75 | -| March - 2005 | 6492.82 | | 2.00 | -| April-2005 | 6154.44 | | 3.79 | -| May-2005 | 6715.11 | | 6.68 | -| June-2005 | 7193.85 | | 4.06 | -Registrar and Transfer Agent: The Company has appointed Registrar and Transfer agent to handle the -share transfer work and to solve the complaints of Shareholders. Name, Address and telephone number -of Registrar and Transfer Agent is given hereunder: -M/s. Adroit Corporate Services Pvt Ltd. -Unit No.3 & 4, Near Maharashtra Weigh Bridge, -Link Road, Saki Naka, Andheri (E), -Mumbai-400072, -Tele. No.: +91-22 - 288590942 / 28503748 / 26924437 -Fax No. : +91 -22-26924438 -Share Transfer System: -Shares sent for transfer in physical form are registered by the Company's Registrars and Share -Transfer Agents within 15 days from the receipt of the documents, if the same are found in order. -Shares under objection are returned within two weeks. +NumberofQ+Apairs NumberofCategories %UniqueC Total(I,Q,A)triplets +| | | | Train | Val | Test | Train | Val | Test | Val | Test | Train | Val | Test | | +| --- | --- | --- | ----- | --- | ---- | ----- | --- | ---- | --- | ---- | ----- | --- | ---- | --- | +Templated 8,133 200 500 1,983 114 300 26% 21% 40,665 1,000 500 +12taNi Automatic 96,317 1,000 1,500 4,061 343 562 5% 6% 481,585 5,000 1,500 +MultiAnswer 11,262 200 500 3,585 118 318 31% 22% 56,310 1,000 500 +| | | TwoHop | 10,819 | 200 | 500 | 2,499 | 131 | 359 | 11% | 6% | 54,095 | 1,000 | 500 | | +| --- | --- | --------- | ------ | --- | --- | ----- | --- | --- | --- | --- | ------ | ----- | --- | --- | +| | | Templated | 5,795 | 200 | 500 | 1,808 | 135 | 268 | 27% | 28% | 25,870 | 827 | 500 | | +2vDLG +Automatic 57,124 750 1,250 1,965 138 208 9% 9% 255,529 3,025 1,250 +MultiAnswer 12,667 200 500 6,449 151 339 45% 56% 56,426 844 500 +| | | TwoHop | 10,221 | 200 | 500 | 1,405 | 99 | 154 | 11% | 11% | 45,771 | 895 | 500 | | +| --- | --- | ------ | ------ | --- | --- | ----- | --- | --- | --- | --- | ------ | --- | --- | --- | +Total 212,338 2,950 5,750 16,249 1,071 2,152 14% 17% 1,016,251 13,591 5,750 +Table8: DatasetStatistics. Wereportthenumberofquestion,answer,categoriesandtripletsforbothsupportingdatasetsanddifferent +questiontypes.Intotalourdatasetcontains1MVQAtriplets,makingitthelargestofitskind. +question. Forthisreason, wedesignedasecondpromptto evantsectionswithPaLM.Whennotavailable,werevertto +| validatethetwo-hopquestionprovidedastheinitialoutput | | | | | | | | usingPrompt4. | | | | | | | +| ---------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | +(Prompt3).Themodelisaskedtoanswerthetwo-hopques- +G.2.Promptforrelevantsectionidentification +tionusingthetwoinitialsingle-hopquestionswithanswers +ascontext.Ifthepredictedanswerisidenticaltotheanswer +| | | | | | | | | We use | Prompt | 8 to | identify | relevant | sections | in a | +| --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | ---- | -------- | -------- | -------- | ---- | +of the second single-hop question, then the two-hop ques- Wikipedia article retrieved by Lens. To do so, we query +tionisvalidatedandkeptinourdataset;otherwise,itisdis- +PaLMforeachsection$S$intheLensretrievedWikipedia +carded. Finally,wefiltersomecommonfailurecases,such article for a question $Q$. Note that we use a few exam- +aswhenthemodeloutputstheexactfirstsingle-hoporsec- plesintheprompttoconditionPaLMtogenerateayes/no +| ond single-hop | | question | as the | chained | two-hop | question. | | | | | | | | | +| -------------- | --- | -------- | ------ | ------- | ------- | --------- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | +| | | | | | | | | answer. | TheanswerproducedbyPaLMisconvertedintoa | | | | | | +We sampled 100 chained two-hop questions to manually stringandmatchedtoeitheryesorno. Whennomatchis +| assess their | quality, | which | we | found | to be correct | in 88% | | | | | | | | | +| ------------ | -------- | ----- | --- | ----- | ------------- | ------ | --- | --------- | ------ | ------------ | ------ | -------- | --- | --------- | +| | | | | | | | | found, we | assume | that section | is not | relevant | to | the ques- | +ofthecases. Inthefinalprompts(Prompts2-3)“$SQ1$”/ tion. Ifmorethanonesectionisidentifiedasrelevantfora +“$SA1$”areplaceholdersforthefirstsingle-hopquestion/ question, the input $S$ to Prompt 7 becomes the concate- +answer,“$SQ2$”/“$SA2$”areplaceholdersforthesecond +nationofallrelevantsections. +single-hopquestion/answer,and“$Q$”istheplaceholder +forthegeneratedtwo-hopquestion. +G.Promptsforevaluation +InSection5weanalyzetheperformanceoflargemodels +| inourdataset. | Similarlytothedatasetcreationprocess,we | | | | | | | | | | | | | | +| ------------- | --------------------------------------- | --- | ------------------------- | ----------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +| use diverse | prompts | to | adapt | and improve | the | behavior | of | | | | | | | | +| thesemodels. | Inparticular, | | weusepromptstoincorporate | | | | | | | | | | | | +retrievalresultsandtoidentifyrelevantsectionsforarticles +retrievedwithLens. +G.1.Promptsforquestionanswering +| We use | textual | Prompts | 4-7 | to produce | | answers with | | | | | | | | | +| -------------- | -------- | ------------- | -------- | ------------ | ------------ | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| large models. | Note | that | we | use the | same prompts | when | | | | | | | | | +| using PaLI | or PaLM. | | However, | we always | | use the ques- | | | | | | | | | +| tion image | I as | an additional | | input to | the model | when us- | | | | | | | | | +| ing PaLI. | Note | that we | use $Q$ | to refer | to the | textual part | | | | | | | | | +| of a question, | $C$ | to denote | | the question | subject | C name, | | | | | | | | | +$Art$todenotethefulltextofaWikipediaarticleand$S$ +| todenoteasectionofaWikipediaarticle. | | | | | Weusethesame | | | | | | | | | | +| --------------------------------------- | --- | --- | ------------ | ----- | ------------ | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| retrievalpromptsforLensandOraclesetups. | | | | | | However,for | | | | | | | | | +| Lens experiments, | | we | only include | $Art$ | or | $S$ if avail- | | | | | | | | | +able,aswemightnotretrieveentitieswithLensorfindrel- -Distribution of SharehoHingas on November 30, 2005 -| | | | No. of shares | | % of Shares | -| -------------- | ------------- | ---------- | -------------- | --- | ----------- | -| No. of Equity | No. of share | % of Share | | | | -| Shares held | holders | holders | held | | held | -| Upto - 500 | 1456 | 83.06 | 246601 | | 5.33 | -| 501 - 1,000 | 129 | 7.36 | 105951 | | 2.29 | -| 1,001 -2,000 | 58 | 3.31 | 86550 | | 1.87 | -| 2,001 -3,000 | 32 | 1.83 | 79743 | | 1 .72 | -| 3,001 - 4,000 | 13 | 0.74 | 46550 | | 1.01 | -| 4,001 - 5,000 | 11 | 0.63 | 51229 | | 1.11 | -| 5,001 - 10,000 | 23 | 1.31 | 163419 | | 3.53 | -10,001 - above -| | 31 | 1.77 | 3849627 | | 83.15 | -| ----- | ---- | ------ | -------- | --- | ------ | -| Total | r*3 | 100.00 | 4629670 | | 100.00 | -Dematerialisation of shares and liquidity: 13.60% have been dematerialised. -Outstanding GDR's/ADR's/Convertible- Nil. -Particular of shares held on physical / Electronic form as on November 30, 2005. -| Category | Holders | No. of Shares | | % of sharers | | -| ---------- | -------- | -------------- | --- | ------------ | --- | -| Paper mode | 1214 | 3999970 | | 86.40 | | -| NSDL | | | | 9.23 | | -| | 422 | 427307 | | | | -| CDSL | 117 | 202393 | | 4.37 | | -| Total | 1753 | 4629670 | | 100 | | -Address for correspondence: -| Aryaman Financial | Services Limited | | | | | -| ----------------- | ---------------- | --- | --- | --- | --- | -401, Chiranjiv Towers, -43, Nehru Place, New Delhi 110 019 +| Prompt1: | Rephrasingautomatically-generatedsinglehopquestion | | | | | | | +| -------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | +In this task, please rephrase the question by replacing the entity name by the word "this", +| followed | by the type | of the | entity. | See the | examples | below: | | +| -------- | ----------- | ------ | ------- | ------- | -------- | ------ | --- | +EXAMPLE 1: +| entity | name: eiffel | tower | | | | | | +| --------- | ------------- | -------- | ------ | ------------ | --- | ------ | --------- | +| question: | How tall | is the | eiffel | tower? | | | | +| step 1 | (find type of | entity): | The | eiffel tower | is | a type | of: tower | +step 2 (write the word "this" followed by the type obtained in "step 1"): this tower +| step 3 | (final rephrased | question): | | How tall | is this | tower? | | +| ------ | ---------------- | ---------- | --- | -------- | ------- | ------ | --- | +EXAMPLE 2: +| entity | name: salmon | | | | | | | +| --------- | ------------- | -------- | ----------- | --------- | ------ | ------- | ---- | +| question: | Which country | is | the largest | producer | of | salmon? | | +| step 1 | (find type of | entity): | The | salmon is | a type | of: | fish | +step 2 (write the word "this" followed by the type obtained in "step 1"): this fish +step 3 (final rephrased question): Which country is the largest producer of this fish? +EXAMPLE 3: +| entity | name: Grand | Beach Provincial | | Park | | | | +| ------ | ----------- | ---------------- | --- | ---- | --- | --- | --- | +question: grand beach provincial park is located on the east side of what lake? +step 1 (find type of entity): The Grand Beach Provincial Park is a type of: park +step 2 (write the word "this" followed by the type obtained in "step 1"): this park +step 3 (final rephrased question): this park is located on the east side of what lake? +Note that the entity name should not be part of the rephrased question. +| Please | make sure to | write out | all | 3 steps as | in the | examples | above. | +| ------ | ------------ | --------- | --- | ---------- | ------ | -------- | ------ | +Based on the above examples, provide a rephrased question for the following case: +| entity | name: $C$ | | | | | | | +| --------- | --------- | --- | --- | --- | --- | --- | --- | +| question: | $Q$ | | | | | | | -ARYAMAN FINANCIAL SERVICES LIMITED -BALANCE SHEET AS AT 30th JUNE - 2005 -| | Schedule | | 30th JUNE, 2005 | | 31st MARCH, 2004 | -| --- | -------- | --- | --------------- | --- | ---------------- | -(Rs) -(Rs.) -SOURCES OF FUNDS -Shareholders' Funds -| a)Share Capital | 1 | 53,619,074 | | 53.619,074 | | -| -------------------- | --- | ----------- | ---------- | ----------- | ---------- | -| b)Reserves & Surplus | 2 | (3.296.920) | | (2,041.813) | | -| | | | 50,320,154 | | 51,577,261 | -Loan Funds -SECURED LOANS -| Deferred Tax Liability (Net) | | | 8,746,000 | | 8,746,000 | -| ---------------------------- | --- | --- | ---------- | --- | ---------- | -| | | | 59,066,154 | | 60,323,261 | -APPLICATION OF FUNDS -FIXED ASSETS -| Gross Block | | 33,129,228 | | 33,129,228 | | -| ----------------- | --- | ---------- | ---------- | ---------- | ---------- | -| Less:Depreciation | | 4.105.964 | | 3,602.375 | | -| Net Block | | | 29,023,264 | | 29,526,853 | -INVESTMENTS -| CURRENT ASSETS.LOANS AND ADVANCES | 4 | 38,829,511 | | 38,376,313 | | -| ---------------------------------- | --- | ---------- | --- | ---------- | --- | -LESS:CURRENT LIABILITIES & PROVISIONS 5 8,786,621 7,781,752 -| NET CURRENT ASSETS | | | 30,042,890 | | 30,594,561 | -| -------------------------- | --- | --- | ---------- | --- | ---------- | -| MISCELLANEOUS EXPENDITURE | 6 | | | | 201,847 | -(To the extent not written-off or adjusted) 59,066,154 60,323,261 -| Notes on Accounts | 12 | | | | | -| ----------------- | --- | --- | --- | --- | --- | -The Schdules referred to above form an -integral part of the Balance Sheet -As per our report of even date -for Thakur Vaidyanath & Aiyar & Co. -for and On behalf of the Board -Chartered Accountatnts -| C.V. Parameswar | | | | DIRECTOR | DIRECTOR | -| --------------- | --- | --- | --- | -------- | -------- | -Partner -M.No.11541 -Place: Mumbai -Dated: 01/12/2005 +| Prompt2: | Chainingtwosingle-hopquestionsintoatwo-hopquestion | | | | | | | +| -------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | +EXAMPLE 1: +| question | 1: in which | city | is this | building | located? | | | +| -------- | ---------------- | ------ | ------- | ----------- | -------- | ---------- | --- | +| answer | 1: San Francisco | | | | | | | +| question | 2: what | is the | average | temperature | in San | Francisco? | | +| answer | 2: 15 Celsius | | | | | | | +combined question: What is the average temperature in the city where this building is located? +EXAMPLE 2: +| question | 1: what | is the | predator | of this | animal? | | | +| -------- | ---------------- | ------ | --------- | ------- | ----------- | --- | --- | +| answer | 1: Lion | | | | | | | +| question | 2: What | is the | weight of | a lion | on average? | | | +| answer | 2: 190 kilograms | | | | | | | +combined question: What is the average weight of the predator of this animal? +EXAMPLE 3: +| question | 1: In what | country | is this | plant | found? | | | +| -------- | ------------- | ---------------- | -------- | ----- | --------- | ------------ | --- | +| answer | 1: Australia | | | | | | | +| question | 2: What | does australia’s | | size | give it a | wide variety | of? | +| answer | 2: landscapes | and | climates | | | | | +combined question: What does the size of the country where this plant is found give it a wide +variety of? +EXAMPLE 4: +| question | 1: What | country | is this | plant | the national | flower | of? | +| -------- | --------------- | ------- | ------- | ----- | ------------ | ------ | --- | +| answer | 1: South Africa | | | | | | | +question 2: south africa is a member of the commonwealth of nations and what other +organization? +| answer | 2: the G20 | | | | | | | +| ------ | ---------- | --- | --- | --- | --- | --- | --- | +combined question: The country that this plant is the national flower of is a member of the +| commonwealth | of nations | and | what | other | organization? | | | +| ------------ | ---------- | --- | ---- | ----- | ------------- | --- | --- | +Based on the above 4 examples, provide a combined question for the following case, such that +the answer to the combined question is the same as the answer to question 2: +| question | 1: $SQ1$ | | | | | | | +| -------- | ------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | +| answer | 1: $SA1$ | | | | | | | +| question | 2: $SQ2$ | | | | | | | +| answer | 2: $SA2$ | | | | | | | +| combined | question: | | | | | | | +| Prompt3: | Validatingtwo-hopquestiongiventheoriginalsingle-hopquestions | | | | | | | +| question | 1: $SQ1$ | | | | | | | +| answer | 1: $SA1$ | | | | | | | +| question | 2: $SQ2$ | | | | | | | +| answer | 2: $SA2$ | | | | | | | +Based on the questions and answers above, please answer the following question: $Q$ -| IF ARYAMANFINANCIALSERVICES LIMITED | | | | | | • | -| ------------------------------------- | --- | --- | --- | --- | --- | --- | -PROFIT AND LOSS ACCOUNT FOR THE YEAR ENDED 30th JUNE.2Q05 -| | Schedule | | 30th JUNE, 2005 | | | 31st MARCH, 2004 | -| --- | -------- | --- | --------------- | --- | --- | ----------------- | -| | (Rs.) | | | | | (Rs.) | -INCOME -| Merchant Banking | | 2,760,000 | | | 4,200,295 | | -| ---------------------------- | --- | --------- | ----- | --- | --------- | --- | -| Other Income from Operation | 7 | | 7,238 | | 3,466,748 | | -| Project Consultancy | | | | | 225,000 | | -128,647 -Proft / Loss on Sale of Shares -Proft / Loss on Sale of Invetments -| Prior Period Adjustment | | | | | 123,291 | | -| ----------------------- | --- | --- | --- | --------- | ------- | --------- | -| | | | | 2,767,238 | | 8,143,981 | -EXPENDITURE -| Personnel expenses | 8 | | 204,459 | | 631,341 | | -| -------------------------------------- | --- | --------- | ------- | ----------- | --------- | ----------- | -| Administation and other Expenses | 9 | 1,239,747 | | | 2,139,310 | | -| Finance Charges | 10 | | 29,864 | | 1,823,941 | | -| Legal and Consultancy charges | 11 | 1,778,215 | | | 1,660,499 | | -| Auditors' Remuneration | | | 66,250 | | 55,000 | | -| Miscellaneous Expenditure written off | | 201,847 | | | 244,276 | | -| Prior Period Adjustment | | | 374 | | | | -| Proft / Loss on Sale of Asset | | | | | 805,776 | | -| Bad Debts written off | | | | | 1,313,316 | | -| | | | | 3,520,756 | | 8.673.456 | -| Profit before Depreciation | | | | (753,518) | | (529,477) | -| Depreciation | | | | 503,589 | | 648.453 | -| Profitless) after Depreciation | | | | (1,257,107) | | (1,177,930) | -Provision for Deferred Tax -| Profit/(loss) after Tax | | | | (1,257,107) | | (1,177,930) | -| ------------------------ | --- | --- | --- | ------------ | --- | ------------ | -| Balance as per Last year | | | | (28,141,813) | | (26,963,883) | -| | | | | (29,398,920) | | (28,141,813) | -in -APPROPRIATIONS -Profit/(loss) after appropriations carried to Balance Sheet (28,141,813) -| Notes on Accounts | 12 | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | -The Schdules referred to above form an -As per our report of even date -for Thakur Vaidyanath & Aiyar & Co. For and on behalf of the Board -Chartered Accountatnts -C.V. Parameswar -| | | | | DIRECTOR | DIRECTOR | | -| --- | --- | --- | --- | --------- | -------- | --- | -Partner -M.No.11541 -Place: Mumbai -Dated: 01/12/2005 - -7= ARYAMANFINANCIALSERVICES LIMITED -SCHEDULES TO BALANCE SHEET -| | 30th JUNE, 2005 | 31st MARCH, 2004 | | -| --- | --------------- | ---------------- | ----- | -| | (Rs.) | | (Rs.) | -SCHEDULE 1 -SHARE CAPITAL -AUTHORISED -| 50,00,000 Equity Shares of Rs.10/- each | | 50000000 | 50000000 | -| --------------------------------------- | --- | -------- | -------- | -ISSUED.SUBSCRIBED AND PAID-UP -46,29,670 Equity Shares of Rs. 10.00 each 46,296,700 46,296,700 -| Deposit againstlssue of Shares | 7,322,374 | | 7,322,374 | -| ------------------------------ | --------- | ---------- | ---------- | -| | | 53,619,074 | 53,619,074 | -SCHEDULE 2 -RESERVES AND SURPLUS -a) GENERAL RESERVE -| | 26,100,000 | | 26,100,000 | -| --- | ---------- | --- | ---------- | -Balance as per Last Year -| | 26,100,000 | | 26,100,000 | -| ------------------------- | ------------ | ----------- | ------------ | -| b) Proft and Loss account | (29,398,920) | | (28,141,813) | -| | | (3,298,920) | (2,041,813) | -SCHEDULE 4 -CURRENT ASSETS.LOANS AND ADVANCES -RENT ASSETS -Interest Accrued but not Due -Sundry Debtors: -(unsecured,considered good) --Debts outstanding for a period exceeding 6 months 16,877,836 16,945,245 -| -Others | 1,576,863 | | 122,865 | -| ------- | ---------- | ---------- | ---------- | -| | 18,454,699 | | 17,068,110 | -| | | 18,454,699 | 17,068,110 | -CASH AND BANK BALANCE: -| -Cash Balance on Hand | | 20,119 | 507,832 | -| --------------------- | --- | ------ | ------- | --Bank Balance: -| -With Scheduled Banks on Current Account | | 22,706 | 116,047 | -| ---------------------------------------- | --- | ------ | ------- | --With Scheduled Banks on Deposit Account -| | | 18,497,524 | 17,691,988 | -| --- | --- | ---------- | ---------- | -LOANS AND ADVANCES -(Unsecured,considered good) -| Shares Held Against Portfolio | 32,681,045 | | 32,681,045 | -| ----------------------------- | ---------- | ----------- | ---------- | -| Portfolio Advances | 33,682,872 | | 33,546,366 | -| | | (1,001,827) | (865,321) | -Shares Held against Portfolio Commitments 19,414,282 19,414,282 -Advances recoverable in cash or in kind or value to be -| received | | 1,141,872 | 1,351,204 | -| ---------------- | --- | ---------- | ---------- | -| Security Deposit | | 777,660 | 784,160 | -| | | 20,331,987 | 20,684,325 | -| | | 38,829,511 | 38,376,313 | -SCHEDULE S -CURRENT LIABILITIES AND PROVISIONS -alCURRENT LIABILITIES -| 1. Sundry Creditors | 693,403 | | 562,128 | -| ---------------------------- | --------- | --------- | --------- | -| 2. Other Current Liabilities | 8,093,218 | | 7,219,624 | -| | | 8,786,621 | 7,781,752 | -blPROVISIONS -| | | 8,786,621 | 7,781,752 | -| --- | --- | --------- | --------- | -SCHEDULES -MISCELLANEOUS EXPENDITURE -extent not written off or adjusted) -| a)Public Issue Expenses | | | 188,152 | -| ----------------------- | --- | --- | ------- | -b)Preliminary and Pre-operative Expenses -13,695 -201,847 - -SCHEDULE 3 -FIX ASSETS -| DESCRIPTION | | GROSS BLOCK | | | | | DEPRECIATION | | | | -| ----------- | --- | ----------- | --- | --- | --- | --- | ------------ | --- | --- | --- | -NET PL^^K -| | | | | | | | | | ASON31ST | ASON31ST | -| --- | --------- | ---------- | ------- | ---------- | -------- | --------- | --------------- | -------- | --------- | -------- | -| | ASON31ST | ADDITIONS | | DELETIONS | ASON3OTH | ASON31ST | FOR SALESWDJ. | ASON31ST | | | -| | | | | | | | | | MARCH | MARCH | -| | MARCH | | DURING | DURING | JUNE | MARCH | THE DURING | MARCH | | | -2004 THEYEAR THEYEAR 2006 2004 YEAR THEYEAR 2006 2006 2006 -A) OWN USE -LAND-FREEHOLD 8,342,965 - - 8,342,965 - 8,342,965 8,342,965 -BUILDING 22,166,276 - - 22,166,276 3,096,043 451,390 - 3,547,433 18,618,843 19,070,233 -FIXTURE 627,420 - - 627,420 409,355 34,999 - 444,354 183,066 218,065 -VEHICLES -AIRCONDmONERS 235,050 - - 235,050 92,457 13,948 - 106,405 128,645 142,593 -OFFICE EQUIPMENT 6,341 - - 6,341 261 376 - 637 5,704 6,080 -COMPUTERS 18,150 - - 18,150 4,259 2,876 - 7,135 11,015 13,891 -B) LEASED -| PLANT AND MACHINERY | | | | | | . | | | - | | -| ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| LEASE ADJUSTMENT ACCOUNT | | - | | | | | | | | | -C) CAPITAL WORK-IN-PROGRESS -j -| BUILDING AND EQUIPMENT | 72,500 | | - | - | 72,500 | | | | 72,500 | 72,500 | -| ---------------------- | ------- | --- | --- | --- | ------ | --- | --- | --- | ------- | ------ | -! ADVANCE FOR COCHIN STOCK -. -EXCHW<£ CORPORATE MEMBERSHIP 1,660,526 - - 1,660,526 1,660,526 1,660,526 -TOTAL 33,129,228 - - 33,129,228 3,602,375 503,589 - 4,105,964 29,023,264 29,526,853 -PREVIOUS YEAR 39,881,271 25,436 6,777,479 33,129,228 5,232,890 648,453 2,278,968 3,602,375 29,526,853 34,648,381 - -fj= ARYAMANFINANCIALSERVICES LIMITED - -^,1 -SCHEDULES TO PROFIT AND LOSS ACCOUNT -| ACCOUNT NAME | | 30th JUNE, 2005 | | 31st MARCH, 2004 | | -| ------------- | --- | --------------- | ----- | ---------------- | ----- | -| | | | (Rs.) | | (Rs.) | -SCHEDULE 7 -OTHER INCOME FROM OPERATIONS -| INTEREST | | | - | | 2,248 | -| ----------------------------------------- | ------ | --- | ----- | --- | --------- | -| LEGAL, PROFESSIONAL & CONSULTANCY CHARGES | | | _ | | 2,817,500 | -| POJECT CONSULTANCY CHARGES | | | - | | | -| MISCELLANEOUS | INCOME | | 7,238 | | | -| COMMISSION RECIVED | | | - | | 647,000 | -| | | | 7,238 | | 3,466,748 | -SCHEDULE 8 -PERSONNEL EXPENSES -| SALARY AND ALLOWANCES | | | 165,000 | | 547,000 | -| --------------------- | --- | --- | ------- | --- | ------- | -| | | | 39,459 | | 84,341 | -STAFF WELFARE -| | | | 204,459 | | 631,341 | -| --- | --- | --- | ------- | --- | ------- | -SCHEDULE 9 -ADMINISTRATIVE AND OTHER EXPENSES -| TELEPHONE CHARGES | | | 222,113 | | 548,578 | -| ------------------------------ | --- | --- | ------- | --- | ------- | -| COURIER CHARGES | | | 8,652 | | 25,502 | -| POSTAGE AND TELEGRAM | | | 20 | | 1,311 | -| RENT.RATES AND TAXES | | | 125,125 | | 294,500 | -| AIRS AND MAINTENANCE-BUILDING | | | 712 | | 3,018 | -| REPAIRS AND MAINTENANCE-OTHERS | | | 192,937 | | 240,738 | -| BOOKS AND PERIODICALS | | | | | 3,541 | -975 -| OFFICE EXPENSES | | | 191,306 | | 256,233 | -| ----------------------- | --- | --- | ------- | --- | ------- | -| BUSINESS PROMOTION | | | 55,340 | | 76,527 | -| 1 EGAL EXPENSES | | | - | | | -| ELECTRICITY CHARGES | | | 41,763 | | 125,089 | -| PRINTING AND STATIONERY | | | 27,892 | | 76,898 | -| TRAVELLING EXPENSES | | | 145,433 | | 333,007 | -| INSURANCE | | | 3,070 | | 9,723 | -| DEMAT CHARGES | | | 34,881 | | 33,623 | -| LISTING CHARGES | | | 15,000 | | 7,500 | -| OTHERS EXPENSES | | | | | 103,523 | -174,528 -| | | | 1,239,747 | | 2,139,310 | -| --- | --- | --- | --------- | --- | --------- | -SCHEDULE 10 -FINANCE CHARGES -| BANK CHARGES | | | 7,490 | | 21,752 | -| --------------------- | --- | --- | ------ | --- | --------- | -| LEASE RENT | | | 22,374 | | 134,244 | -| INTEREST | | | - | | 92,945 | -| OTHER FINANCE CHARGES | | | - | | 1,575,000 | -1,823,941 -29,864 -SCHEDULE 11 -LEGAL AMD CONSULTANCY CHARGES -| LEGAL.PROFESSIONAL AND CONSULTANCY | | | | | 1,572,975 | -| ---------------------------------- | --- | --- | --- | --- | --------- | -1,672,805 -| MEMBERSHIP CHARGES | | | 104,160 | | 81,944 | -| --------------------- | --- | --- | --------- | --- | --------- | -| DIRECTOR SITTING FEES | | | - | | - | -| STAMP PAPERS | | | 1,250 | | 5,580 | -| INCOME TAX PAID | | | - | | | -| MARKETING FEES | | | - | | | -| | | | 1,778,215 | | 1,660,499 | - -IT ARYAMANFINANCIALSERVICES LIMITED -SCHEDULE 14 -NOTES TO ACCOUNTS -SIGNIFICANT ACCOUNTING POLICIES: - -Accounts are prepared on Historical Cost Convention accruing for incomes, expenditures, assets -and liabilities. -a. Income from Project Consultancy is accrued considering stage wise completion of work, wherever -agreed upon or feasible, otherwise it is accrued on percentage basis based on the quantum of work -completed. -b. Income from Merchant Banking is accrued on percentage basis based on quantum of work completed. -a. FIXED ASSETS: -Fixed Assets are shown at cost less accumulated depreciation. Cost of asset includes all expenses -related to acquisition of the asset. -b. DEPRECIATION: -Owned Assets: Depreciation is charged on Straight Line Method at the rates and in the manner -provided in Schedule XIV of the Companies Act, 1956. -Leased Assets: On leased assets the minimum statutory depreciation is charged along with which -the stipulated lease Equalisation Account is created, to ensure full depreciation of the leased assets -within the tenure of the lease period by way of matching annual charges based on the lease rental -accruals. -Investments are stated at cost. -Miscellaneous Expenditure are written off as follows: -a. Preliminary Expenses, Pre-Operative Expenses (other than under (b)below) & Public Issue Expenses -- Pro-rata over 10 years. -b. Pre-Operative Expenses relating to Merchant Banking Division and Registrar & Share Transfer -Agents - Pro-rata oYar 5 years. -d. Computer Software -Equally over 3 years. -Taxes on Income -The Company provides for deferred tax using liability method, based on tax effect of timing -difference resulting from the recognition of items in the financial statements and in estimating its -current income tax provision. -Deferred Tax Assets arising from temporary timing differences are recognized to the extent there - -IF ARYAMANFINANCIALSERVICES LIMITED -Taxes on Income -The Company provides for deferred tax using liability method, based on tax effect of timing -difference resulting from the recognition of items in the financial statements and in estimating its -current income tax provision. -Deferred Tax Assets arising from temporary timing differences are recognized to the extent there -is reasonable certainity that the assets can be realized in future. -NOTES -1. Estimated amount of contracts to be executed on capital accounts Rs. 2,48,500/- -(P.Y. Rs. 2,48,500) -2. Contingent Liability RS.42.36 Lakhs (P.Y. RS. 42.36 Lakhs) -3. During the previous year the company assigned an office space which was pending registration -and a liability there against of Rs. 39.83 lacs to a party. The Company had filed a suit in the High -Court against the seller of the property for specific performance of the contract i.e. in arranging to -handover the original title papers making the bank as the third party to the suit. The company had -assigned the assets along with the liabilities to a party with the understanding that no liability -should accrue to the company with effect from 15-03- 2004 and the pending legal case shall be -pursued by the assignee. -4. Some of the debit/credit balances are subject to confirmation and reconciliation. -5. In view of the number of employees being below the stipulated numbers, the Payment of Gratuity -Act is not applicable to the company for the year. The same is the case with respect to payment -of Bonus Act. -6. Auditors Remuneration: - -| Amount (Rs.) | Amount (Rs.) | | -| ------------ | ------------ | --- | -(P.Y.) -| Audit Fees-Statutory | 43750 | 35000 | -| -------------------- | ----- | ----- | -| -Tax Audit | 12500 | 10000 | -| Certification | 10000 | 10000 | -NIL -| Out of pocket expenses | NIL | | -| ---------------------- | ----- | ----- | -| TOTAL | 66250 | 55000 | -7. Deferred Tax Liability / Assets -| Deferred Tax Liabilities | Amount (Rs.) | | -| ------------------------ | ------------ | --- | -| -At 1st April, 2004 | 87,46,000 | | -- Additional Adjustments for Current year -87,46,000 - -8. Directors' Remuneration : - -| | Amount | (Rs.) | Amount | (Rs.) | -| --- | ------- | ----- | ------- | ----- | -(P.Y.) -| Salary | | 60000 | | 360000 | -| -------------------- | --- | ------ | --- | ------ | -| Provident Fund | | NIL | | NIL | -| Perquisites | | 42000 | | 252000 | -| Legal & Professional | | 495000 | | NIL | -| TOTAL | | 597000 | | 612000 | -9. Quantitative Details of Shares under Portfolio Advance: -| | | No. | Amount | (Rs.) | -| --- | --- | --- | ------- | ----- | -Equity Shares of Rs. 107- each Fully Paid-up -| Opening Stock | | 1655784 | P.Y.(1555784) | | -| -------------- | --- | -------- | ------------------ | --- | -| | | 52095327 | P.Y.(5109S327) | | -| Purchase | | NIL | P.Y. (I,0ft,fl00) | | -| | | NIL | P.Y. (10,00,000) | | -| Sale | | NIL | P.Y. (NIL) | NIL | -| | | P.Y. | (NIL) | | -10. Income/Expenditure in foreign currency: -| | Amount | (Rs.) | Amount | (Rs.) | -| --- | ------- | ----- | ------- | ----- | -(P.Y.) -| Expenditure | Rs. NIL | | Rs. NIL | | -| ----------- | ------- | --- | ------- | --- | -| Income | Rs. NIL | | Rs. NIL | | -11. EARNING PER SHARE -| | Share Price | | Share Price | | -| --- | ----------- | --- | ----------- | --- | -As at -As at -| | JUNE 30, 2005 | | March 31. 2004 | | -| ----------------------------- | ------------- | ----- | -------------- | ----- | -| Nominal value of Equity Share | Rs.10/- | | Rs.10/- | | -| Earning per Share | (Rs. | 0.27) | (Rs. | 0.25) | -12. Related Party Disclosure -Mr. P. V.R. Murthy, Director of the company was covered by the definition of the Key Management -Personnel upto December 2004. Remuneration of Rs 5,52,000/- is paid under his contract of -employment with the company. Additionally, as at 30th June, 2005 there is no other related party -relationships as envisaged in Accounting Standard 18 pertaining to Related Party Disclosures. -13. Current year figures are for 15 months as compared to previous year figures for 12 months. -14. Previous years' figures have been regrouped and rearranged wherever necessary. - -F ARYAMANFINANCIALSERVICES LIMITED -Additional information as required under Part IV of Schedule VI to the Companies Act, 1956 -Balance Sheet Abstract and Company's General Business Profile -I Registration Details -Registration No. 59009 -State Code No. 55 -Balance Sheet Date 30.06.2005 -II Capital Raised During the year (Rs. In Thousands) -Public Issue -Nil -Rights Issue Nil -Bonus Issue -Nil -Private Placement - Equity Shares -Nil -III Position of Mobillisation and Deployment of Funds (Rs. In Thousands) -Total Liabilities 59066 -Total Assets 59066 -Sources of Funds -Paid - up Capital 53619 -Reserves & Surplus -3299 -Secured Loans -Nil -Unsecured Loans -Deferred Tax liability (Net) 8746 -Application of Funds -Net Fixed Assets 29023 -Investments Nill -Net Current Assets 30043 -Misc. Expenditure 0 -Accumulated Losses Nil -IV Performance of the Company (Rs. In Thousands) -Turnover (Gross Receipts) 2767 -Total Expenditure 4024 -Profit (+) / Loss (-) before Tax -12 5 7 -Profit (+) / Loss (-) after Tax -1257 -Earning per share (Rs.) Nil -Dividend (%) Nil -V Generic names of Principal Products , Services of the C ompany -Item Code No. (ITC Code) -Not Applicable -Production Description -1 Fjnanda| Sen/jces -2. Project Consultancy -3. Merchant Banking -Code No. for the services rendered by the Company is not available in the Publication of Indian Trade -classification for ITC Code of Products by Ministry of Commerce, Directorate General of Commercial -Intelligence and Statistics, Calcutta 700 001 -As per our Report of even date -For Thakur Vaidyanath Aiyar & Co. For and on behalf of the Board -Chartered Accountants -C.V. Parameswar Director -Partner -M.No. 11541 Director -Place:- Mumbai -Dated:-01-12-2005 - -P= ARYAMAN FINANCIAL SERVICES LIMITED -ARYAMAN FINANCIAL SERVICES LIMITED -Registered Office : 401, Chiranjiv Towers, 43, Nehru Place, New Delhi 110019 -Corporate office : 208, Maker Chambers-v, Nariman Point, Mumbai 400 021. -ATTENDANCE SLIP -Eleventh Annual General Meeting Scheduled on Thursday, 29th Dec.2005 AT -401, Chirajiv Towers, 43, Nehru Place, New Delhi 110019. at 10,00 am. -PLEASE COMPLETE THIS ATTENDANCES SLIP AND HAND IT OVER AT THE ENTRANCE OF THE MEETING HALL -NAME & ADDRESS OF THE SHAREHOLDER L.F.NO -I, hereby record my presence at the ELEVENTH ANNUAL GENERAL MEETING of the Company -Scheduled on Thursday , 29th, Dec. 2005 at 401, Chiranjiv Towers, New Dilhi at 10. am. -SIGNATURE OF THE SHAREHOLDER OR THE PROXY ATTENDING THE MEETING. -If Shareholders, please sing here if Proxy, please sing here -NOTE : The copy of Annual Report may please be brought at the Meeting Hall -(TEAR HEAR) -ARYAMAN FINANCIAL SERVICES LIMITED -Registered Office : 401, Chiranjiv Towers, 43, Nehru Place, New Delhi 110019. -Corporate Office : 208, Maker Chambers -v, Nariman Point, Mumbai 400 021 -PROXY FORM -L.F. No. -of being a member / members of ARYAMAN FINANCIAL SERVICES LIMITED -hereby appoint ........ - ...... -of --------------------------------------------------------- failing him / her- ----------------------------------------------------o f -or -as my / our proxy to vote fo>- me/ us on behalf at the Eleven Annual General Meeting of the Company to -be held at 401 , Chiranjiv Tower, New Delhi - 110117 on Thursday 29th Dec. 2005 -as witness my/ our hand(s) this ----------------------------- day of - ................... — 2005 -(Signature of the Shareholder) -NOTE :The Proxy From must be returned so as to reach the Registered Office of the Company note less -than 48 hours before the time for holding the aforesaid meeting. - -BOOK-POST -To, -// undelivered please return to .- -Adroit Corporate Services Pvt. Ltd. -Unit, Aryaman Financial Services Ltd. -19, Jaferbhoy Industrial Estate, 1st Floor, -Makwana Road, Marol Naka, -Andheri (W), Mumbai - 400059 \ No newline at end of file +| Prompt4: | Questiononlyevaluation | | | | | | +| ---------- | --------------------------- | ------ | -------------- | -------- | ------------ | --- | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt5: | Retrievalwithentityname | | | | | | +| Entity | name: | $C$ | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt6: | RetrievalwithKBArticle | | | | | | +| Context: | $Art$ | | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt7: | RetrievalwithKBSection | | | | | | +| Context: | $S$ | | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt8: | LensretrievaltogetKBSection | | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | Is | this | fungus edible? | | | | +Text: This compound induces mammalian cells (specifically, the cell line HL60 to differentiate +into granulocyte- or macrophage-like cells. The fungus also contains the mycotoxin muscarine, +and the antifungal metabolite strobilurin D. Despite the presence of these toxins, some guides +| list this | fungus | safe | for human | consumption. | | | +| ---------- | ------ | ------ | --------- | --------------- | ------------- | --- | +| The answer | is: | yes | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | In | which | season | does this plant | give flowers? | | +Text: This cactus has stems about 1/2-1 inch wide with 6-9 edges. Its flowers are white, up +to 30 centimetres in diameter with a scent redolent of vanilla. The flowers open after sundown, +closing and wasting after a few hours. By 9 am the next day they are gone. +| The answer | is: | no | | | | | +| ---------- | ------ | ------ | ----------- | -------- | ------------ | --- | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | What | is | the habitat | of this | animal? | | +Text: X is native to Europe and North Africa through to Central Asia. It is introduced to the +United States and parts of South America. It widespread across the northeastern United States +and eastern Canada, and can be found outside, or more commonly inside houses. It is thought to +| have been | introduced | | into America | from | Europe by English | colonists. | +| ---------- | ---------- | ------ | ------------ | -------- | ----------------- | ---------- | +| The answer | is: | yes | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | $Q$ | | | | | | +Text: $S$ +| The answer | is: | | | | | | +| ---------- | --- | --- | --- | --- | --- | --- | \ No newline at end of file diff --git a/q188/random_k2/random_2.md b/q188/random_k2/random_2.md index 5c5eba2e1fd980c21e8cbb1cd873801d5dedd049..a334b8d779333a86838d9effd3dd99b5ea050aa5 100644 --- a/q188/random_k2/random_2.md +++ b/q188/random_k2/random_2.md @@ -1,2541 +1,704 @@ - CONTENTS +SOCIAL MEDIA MARKETING -01 -04 - 16 -36 -38 -39 -40 -41 -44 -45 + SOCIAL MEDIA MARKETING -/ About HKU Business School +W H A T T H I S S E S S I O N W I L L C O V E R -/ Explore Your Options +1 -/ Enrich Your Life +2 -/ 2026 JUPAS Admissions +3 -/ 2026 Non-JUPAS Admissions +4 -/ 2026 Dual Degree Admissions - -/ Develop Your Career - -/ Freshmen Sharing - -/ Our Global Presence - -/ Useful Websites - -About HKU Business School - -HKU Business School strives to nurture first-class business leaders and foster academic and relevant -research to serve the needs of Hong Kong, Chinese Mainland and the rest of the world in the new Asia-led -economy. As Asia’s premier international business school, we engage leading scholars from all corners of the -globe and they instil in students global knowledge with an Asian perspective. - -With the highest proportion of non-local undergraduate students across all Faculties at HKU, four of our -undergraduate programmes ranked amongst the University’s Top 10 programmes in 2025#, HKU Business -School attracts the best and brightest students from Hong Kong and beyond. - -HKU Business School is fully accredited by the European Quality Improvement Systems (EQUIS) and the -Association to Advance Collegiate Schools of Business (AACSB). - -# Based on average JUPAS admission scores in the best 5 subjects in HKDSE - -VISION -To be a leading, -globally-impactful academic -institution of business and -economics. - -UNIQUE -PROPOSITION - -Deeply rooted in Hong Kong -Fully engaged with China -Truly international - -MISSION -STATEMENT - -To provide world-leading -business and economics -education that nurtures -dynamic and responsible -leaders. -To advance the frontiers of -knowledge through innovative -and globally-significant -research. -To create positive change in the -world by partnering with -communities. - -01 - - Dean’s Message - -I am delighted to introduce you to the HKU Business -School, a dynamic and forward-thinking academic -community where transformative opportunities and -unparalleled experiences await. - -As a part of the University of Hong Kong, which -proudly ranks 11th globally in the 2026 QS World -University Rankings, HKU Business School stands out -as a beacon of excellence in business and economics -in Asia. This year marks the 25th anniversary of our -Faculty, we continue to march forward as a leading, -impactful business school, championing -globally -research, and -leadership, pioneering -thought -educational excellence. - -of - -dynamic, - -challenging - -responsible, - -Our mission is to inspire and develop the next -generation -and -globally-minded business leaders. We offer a vibrant -and -learning environment where -students are encouraged to push their limits, think -critically, and innovate. Our team of leading scholars -from around the world, and experienced and -dedicated teachers as well as mentors bring a wealth -of knowledge, global perspectives, and personalised -guidance beyond the classroom. Our full spectrum of -top-quality degree programmes, including 11 diverse -undergraduate programmes and four dual bachelor’s -degrees -in partnership with world-renowned -universities, is designed to equip students with -cutting-edge knowledge and technical skills to -in the fast-changing global -navigate and excel -business landscape. - -In response to the transformative power of artificial -intelligence, we are proud to launch the Second Major -and Minor programmes in “Artificial Intelligence in -Business” -to our -(AIB), offered exclusively -undergraduate students. These programmes are -designed to equip future business -leaders with -practical proficiency in AI technologies, strategic -insight to drive cross-sector innovation, and a critical -understanding of the ethical and societal implications -of AI adoption. Through a rigorous curriculum that -integrates business fundamentals with cutting-edge -AI applications, students will be empowered to tackle -complex challenges and create meaningful impact in -a data-driven world. - -Greetings from -Associate Dean (Undergraduate) - -inspiration - -Business education today is an evolving domain that -builds upon learning from all academic disciplines and -draws -successful and -from both -unsuccessful business practices. Our undergraduate -curricula in six core academic areas equip students -with diverse knowledge across business and -economics spheres, and prepare them to be creative -leaders of tomorrow with global perspectives. - -A hallmark of our School that separates us from the -others is our truly multi-disciplinary approach. HKU -Business School’s extensive resources offer students -in -a wide range of choices, combining studies -economics and finance, accounting and finance, -accounting and data analytics, business and law, -business and data analytics, international business -and global management, as well as dual degrees in -collaboration with world-renowned universities. - -Our faculty members will engage you with their -inspiring teaching and how they bring their impactful -research into the classroom. They will open your mind -to diverse perspectives, -ignite your curiosity to -explore innovative ideas, and instil in you an appetite -for acquiring new knowledge and extraordinary -experiences throughout your academic journey and -career development. Our students are among the -brightest and most creative of their peers. Our alumni -have gone on to forge brilliant careers in all sectors of -the -international business -community, and have formed a global network that -you will be a part of for the rest of your life. - -regional and - -local, - -I hope that you will share my enthusiasm for the -programmes and my admiration for the people who -have contributed to their development. I look forward -to having you as a member of the HKU Business -School family and a future leader in the world of -business. - -Warm regards, - -Professor Derek Chan -Associate Dean (Undergraduate) - -Through our extensive partnerships with over 300 -institutions and corporations across more than 40 -countries, as well as our unique network of centres in -Beijing, Shenzhen, Shanghai, Vietnam, and Barcelona, -our students gain unparalleled global experiential -and enrichment opportunities. Whether through -internships, overseas exchange programmes, or -hands-on -these experiences -sharpen practical skills, broaden perspectives, and -prepare students to confidently tackle real-world -challenges. - -industry projects, - -Joining HKU Business School means becoming part of -a community that supports your growth and success. -You will collaborate with brilliant minds, engaging in -stimulating discussions, and working on impactful -projects that make a difference. The friendships, -networks, and experiences you build here will last a -to a myriad of career -lifetime, open doors -opportunities, and empower you to overcome -challenges, pursue your passions with confidence, -and contribute positively to your community and the -world at large. - -If you are ready to embark on an exciting journey of -learning, discovery, and personal development, I -invite you to explore what HKU Business School has to -look forward to welcoming you to our -offer. I -community, where future leaders are nurtured and -dreams take flight. - -Warm regards, - -Professor Hongbin Cai -Dean -Chair of Economics - -0302 Explore Your Options -A Wide Selection of Programmes - -Bachelor of Business Administration [BBA] - -The BBA programme combines strong functional training with intensive exposure -to communication skills, computer applications, and other social science and/or -science subjects. Students can choose to major in the following area(s): - -Entrepreneurship, -Design and - Innovation* - -Finance - -Human Resource -Management - -Information Systems -and Analytics - -Marketing - -BBA Curriculum Structure +5 -Components - -A - -B - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -BBA Major -Faculty Core Courses -Disciplinary Core Courses / Disciplinary Electives -Capstone Course - -C - -Free Electives / Optional Major / Optional Minor(s) - -Total - -Credits - -54 -12 -36 -6 -- - -78 -24 -48 -6 - -108 - -240 - -*Candidates must undergo a selection process arranged by the Programme Coordinator for -declaring the EDI major. - -HKU – Sciences Po Dual Degree Programme - -Students will spend the first two years at Sciences Po in France and the third -and fourth years at HKU. Programmes at Sciences Po are taught in English (no -French requirements for being admitted) with intensive French language -classes for all non-francophone students. - -Upon successful completion of the dual degree programme, students will be -awarded the Bachelor of Business Administration (BBA) degree from HKU and -a Bachelor’s degree from Sciences Po. - -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. - -Find out more: -https://admissions.hku.hk/programmes/university- -collaborative-programmes/hku-sciences-po-dual- -degree-programme - -HKU – University of British Columbia -Dual Degree Programme - -Students will spend the first and final years at HKU and the second and third -years at the University of British Columbia (UBC) in Canada, and graduate with -a Bachelor of Business Administration from HKU and a Bachelor of Commerce -from UBC upon successful completion of the programme. - -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. - -Find out more: -https://admissions.hku.hk/programmes/university- -collaborative-programmes/hku-university-of-british- -columbia-dual-degree-programme - -Find out more: -https://ug.hkubs.hku.hk/programme/bba - -0504 - - Bachelor of Business Administration in Accounting and -Finance [BBA(Acc&Fin)] - -Bachelor of Business Administration in Accounting Data -Analytics [BBA(ADA)] - -The BBA(Acc&Fin) programme prepares students for a professional qualification -in accounting and provides the broad-based education in accounting and finance -which is necessary to progress toward a leadership role in the financial sector. -Students are required to take professional core in accounting and a minor in -finance, with an option to second major in finance. Tailor-made professional -internship opportunities with renowned -preparation programmes and -accounting firms and financial institutions are available for talented students. It -is accredited as an approved accountancy programme by: - -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia - -BBA(Acc&Fin) Curriculum Structure - -Components - -Credits - -A - -B - -C - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -Professional Core in Accounting -Faculty Core Courses -Disciplinary Core Courses -Capstone Course - -Minor in Finance* -Disciplinary Core Courses -Disciplinary Electives - -D. - -Free Electives / Optional Minor - -*Students may opt to pursue a Major in Finance: - -Components - -A - -B - -C - -University Graduation Requirements - -Professional Core in Accounting - -Major in Finance -Disciplinary Core Courses -Disciplinary Electives -Free Electives -Capstone Course - -54 -12 -36 -6 -- - -108 -24 -78 -6 - -36 -12 -24 - -42 - -Total - -240 - -Credits - -54 - -108 - -78 -30 -18 -24 -6 - -Total - -240 - -Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares -students for a professional qualification in accounting. It provides students -with an option to focus on data analytics aspect in accounting instead of having -a minor in finance. It is an excellent choice for students who are interested in -STEM with accounting and business applications and wish to pursue their -career in the relevant fields. The intensive accounting training enables -BBA(ADA) students to distinguish themselves from students of the other data -analytics/STEM programmes. Students will be equipped with comprehensive -accounting and business knowledge that prepares them to obtain professional -qualifications in accounting. They will also acquire computer programming -and data analytics skills to meet the needs of the accounting and business -professions in today’s digital transformation economy. It is accredited as an -approved accountancy programme by: - -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia - -BBA(ADA) Curriculum Structure - -Components - -A - -B - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -Professional Core in Accounting Data Analytics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course - -C - -Free Electives - -Total - -Credits - -54 -12 -36 -6 -- - -156 -24 -114 -12 6 -30 - -240 +What is social media marketing and why is it important -Find out more: -https://ug.hkubs.hku.hk/programme/bba-acc-and-fin +Social media landscape -Find out more: -https://ug.hkubs.hku.hk/programme/bba-ada +Social media marketing – use cases and benefits -0706 +What does good look like? - Bachelor of Business Administration (Business Analytics) -[BBA(BA)] +How to execute a campaign + a club success story -Bachelor of Business Administration (Law) and Bachelor of -Laws [BBA(Law)&LLB] +Do’s and Don’ts + additional resources -The BBA(BA) programme is crafted to meet the growing industry demand for -talents in the business analytics field. Its curriculum offers students a -comprehensive coverage of both technical and managerial skill sets. Students -will learn a wide spectrum of knowledge from the disciplines of information -technology, data science, business statistics and management. Core courses -include programming language, database management, introduction to -business analytics, management information systems, decision and risk -analysis, etc. They help students build the foundations for advanced topics in -senior years. Students also have the option to choose from analytics technique -courses and business application courses. The aim is to provide students with -state-of-the-art experiences in artificial intelligence, big data, social media, -and other latest development. Such design helps train well-rounded students -who will be more adaptable in their career development. The programme is -suitable for students who have a passion for problem-solving through -analytics, and who aim to develop their career path in analytics. +2 -BBA(BA) Curriculum Structure + WHAT IS SOCIAL MEDIA MARKETING? -Components +D E S C R I P T I O N -Credits +The use of social +networks, content +sharing apps, +messaging platforms, +blogs, and forums to +connect with and enable +meaningful +conversations about +your brand, product or +service, to ultimately +drive your business +objective. -A +3 -B + WHY IS SOCIAL IMPORTANT? -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +C U S T O M E R E X P E C T A T I O N S H A V E C H A N G E D -Major in Business Analytics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives (Analytics Techniques) -Disciplinary Electives (Business Applications) -Capstone Course +2X -C +customers are twice as likely to +share a negative experience with a +business than a positive one -Free Electives / Optional Major / Optional Minor(s) +90% -54 -12 -36 -6 -- - -96 -24 -42 -12 -12 -6 +will look at online +reviews before making +a purchase -90 +74% -Jointly offered by the HKU Business School and the Faculty of Law, the -BBA(Law)&LLB integrated double-degree programme blends two strategic -disciplines into a single programme for the bright and highly motivated. It is an -excellent choice for students seeking a broader education with parallel -concentration on legal and business studies. Graduates of the programme will be -awarded two degrees — BBA(Law) by the HKU Business School, and LLB by the -Faculty of Law. After successful completion of the LLB programme, students -may apply for admission to the Postgraduate Certificate in Laws (PCLL) -programme provided that they meet the relevant admission requirements. +trust suggestions from +”friends” on social media -The programme offers business stream and accounting stream to students in -their business studies. The accounting stream is accredited by: +67% -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia +of people will spend money after +getting recommendations from +their “friends” online -BBA(Law)&LLB Curriculum Structure (Business Stream) +Sources: White House Office of Consumer Affairs, McKinsey, Experian |24|7, Gigya, CEI -Components +4 -Credits + WHY IS SOCIAL IMPORTANT? -A +N E W T R E N D S A R E E M E R G I N G C H A N G I N G T H E U S A G E A N D I M P O R T A N C E O F S O C I A L -B +Understanding the social media space and managing your social presence amplifies your marketing +strategy. Social media delivers invaluable insight into your brand awareness, customer sentiment, +marketplace trends, and your competitor’s actions, whilst enabling you to reach more prospects than any +other marketing channel. -C +Every negative social interaction has a cost -University Graduation Requirements -Language Enhancement Course -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Customers are content creators -BBA Major -Faculty Core Courses -Disciplinary Core Courses / Disciplinary Electives -Capstone Course +Brands are turning online influencers into advocates -Professional Core in Law -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Social is now a prominent point of purchase -D. +Brand focus is shifting to measuring quality of interactions -Free Electives / Optional Minor +Reviews can increase sales and add credibility -Total +Social is multifunctional – used for marketing, listening, +response, customer care, troubleshooting -240 +5 -Total + SOCIAL MEDIA MARKETING LANDSCAPE -BBA(Law)&LLB Curriculum Structure (Accounting Stream) +T H E K E Y P L A Y E R S -Components +The social media landscape is large and +complex and cluttered with competing +platforms, but the most heavily utilised +for marketing are the big 4: -A +Facebook (FB, Instagram, +Messenger) -B +Google (YouTube) -University Graduation Requirements - -Professional Core in Accounting -Faculty Core Courses -Disciplinary Core Courses -Capstone Course - -C - -Professional Core in Law - -Total - -36 -6 -24 -6 -- +Twitter -78 -24 -48 -6 +LinkedIn -156 -108 -42 6 -30 + SOCIAL MEDIA MARKETING LANDSCAPE -300 +D I F F E R E N T C H A N N E L S A R E U S E D F O R D I F F E R E N T P U R P O S E S -Credits +Social networks +Relationship networks help brands and people connect to share +information and ideas. With a large & regular user base, they are a must- +have and the ”mass-market” approach to social marketing. -36 +Use for: customer care, education, daily comms, breaking news -108 -24 -78 -6 +Messaging & chat apps -156 +Most widely used apps that enable messaging, video and voice calls. Business +can conduct e-commerce conversations via live agents or chatbots. -300 +Use for: customer care, conversational commerce, payments, status updates, +to replace SMS -Find out more: -https://ug.hkubs.hku.hk/programme/bba-ba +Media & content sharing +networks -Find out more: -https://ug.hkubs.hku.hk/programme/bba-law-and-llb +Find and share media online, including photos, video, and live video or other +curated content. -0908 +Consumer review +networks +Manage online reviews and reputation. Positive reviews bring social proof to +your value proposition. Negative reviews provide you with an opportunity to +resolve an issue publicly. - Bachelor of Business Administration in International Business -and Global Management [BBA(IBGM)] +Use for: brand awareness, lead generation, audience engagement, targeted +advertising, influencer marketing -Bachelor of Economics [BEcon] +Use for: customer care, reputation management, market research -BBA(IBGM) is an intensive, premier programme in international business and -global management featuring a unique curriculum designed to evolve with and -reflect the ever-changing demands of a globalised world. Exclusive elements -such as overseas exchange studies, a leadership boot camp, an Asia-Pacific -study tour, an international field trip and global enrichment opportunities are -strategically integrated into the curriculum, whilst students are given the -flexibility to pursue an in-depth inquiry into another specialised area of study -as a second major. This cross-functional and personalised approach aims to -foster a selected group of outstanding students to become the world’s future -leaders equipped with global perspectives, cultural sensitivity, social -responsibility and key competencies to succeed in the international business -environment. +Blogs & forums +Publish, find, discuss, and share news, information, and opinions. Usually allow +users to remain anonymous, leading to more honest opinions. -BBA(IBGM) Curriculum Structure +Interest-based networks +Help connect with others around a shared interest or hobby and tend to focus +solely on a single subject and provide a dedicated experience for users. -Components +Use for: market research, influencer marketing, product advertising, SEO +results -A +Use for: targeted marketing, brand awareness, trend watching -B +7 -C + SOCIAL MARKETING USE CASES -D +H O W T O U S E S O C I A L T O M E E T Y O U R O B J E C T I V E S -E +Manage +own page -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Generate +insights -Major in International Business and Global Management -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Nurture community +& reputation -Second Major +Engage with +customers -Free Electives +Grow through +amplification -Compulsory Overseas Exchange +Draft and publish organic +content to owned +page(s) -Total +Define and manage +content strategy and +calendar -Credits +Manage content +approvals -54 -12 -36 -6 -- +Listen/capture social +conversations about LTA, +coaches, or venues and +identify trends -96 -24 -54 -12 -6 +Keep up with competitors +and stay informed of their +social engagement, +content and campaigns -72-90 +Understand target +audience demographics, +sentiment, and channel +preferences -0-18* +Grow your community of +social followers and +supporters -- +Find and nurture +advocates and +influencers who can +credibly promote your +brand to an extended +audience -240 +Identify and manage +potential crises -*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. +Respond to consumer +problems when they +have questions or +concerns -The BEcon programme trains students in economic theories, statistical tools, -the application of theories to solve practical economic problems in trade, -banking and the labour market, and evaluation of the regional economic -performance of Chinese Mainland, Hong Kong and other economies. Students -will be furnished with solid theoretical and empirical economic knowledge to -analyse and understand human behaviour and social interactions. The -programme allows students a high degree of flexibility in planning their future -career specialisation. +Be part of the +conversations about your +venues, coaches or +brand -BEcon Curriculum Structure +React with relevant +content – use what you +see and hear to help +inform future marketing +strategy -Components +Push paid content to +custom audiences to reach +your marketing objectives -A +Identify and amplify +relevant user content for +authentic engagement -B +Social commerce – +increasing number of +businesses are using +social as a payment +channel -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +8 -Major in Economics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course + THE BENEFITS OF SOCIAL -C +W H A T S O C I A L M E D I A M A R K E T I N G C A N D O F O R Y O U -Free Electives / Optional Major / Optional Minor(s) +1. -Total +Strengthen your +brand reputation -Credits +2. -54 -12 -36 -6 -- +Grow your +business -78 -24 -30 -18 -6 +3. -108 +Care for your +customers -240 +Know & act on what is being said about +you +• Things escalate quickly on social. By -Find out more: -http://ibgm.business.hku.hk +listening to key phrases that relate to your +brand or activity you can be the first to +know when a news story breaks and react +accordingly. -Find out more: -https://ug.hkubs.hku.hk/programme/becon +• Social insights also give you an unbiased -1110 +view of your brand awareness and +popularity. - Bachelor of Economics and Finance [BEcon&Fin] +Identify trends +• Trends appear on social media first. Social -BEcon&Fin is designed to meet the interests of students who are seeking a -balanced study of both economics and finance. The programme recognises -the significance of economics both in its own right and as a foundation for the -study of finance, and the significance of finance as a career path for students -who are keen on joining this profession. Students will receive solid training in -theories and applications of economics and finance. They will be equipped -with the skills to analyse human behaviour and social interaction, as well as to -solve problems in the financial sphere. +can make or break a product. -BEcon&Fin Curriculum Structure +• Listening to industry thought-leadership +supports your research and innovation -Components +Stay ahead of competition +• Your competitors are on social. Analysing -Credits +their social presence delivers unique +insight into their plans and achievements +• Regular benchmarking of your competitors -A +on social helps you keep up with them -B +Amplify owned content +• Social enables you to publish targeted -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +content to channels that could be seen by +large volumes of users -Major in Economics and Finance -Faculty Core Courses -Disciplinary Core Courses -Economics Disciplinary Electives -Finance Disciplinary Electives -Capstone Course +• The power of influencer marketing -C +provides a human voice to your product -Free Electives / Optional Major / Optional Minor(s) +Generate earned media +• Customers are 56% more likely to buy a +product or service after seeing a positive +customer-generated post of it -54 -12 -36 -6 -- +• By reposting user-generated content on +your owned channels, you increase the +trust in your brand and products -96 -24 -42 -12 -12 -6 +Deliver incremental sales +• Advertising on social media is cheaper +than traditional media (such as TV ads) +but with a massive reach -90 +• The main difference is that you will be -Total +able to target finely who sees what advert, +increasing your conversion while keeping +costs down -240 +Create experiences +• Social is more than just advertising, it’s an +educational, emotional and experiential +way to engage your customers -HKU - Peking University “Future -Leaders” Dual Bachelor’s Degree in -Economics, Finance and Management -The programme combines the strengths of two leading business schools in -Hong Kong and Chinese Mainland and is designed to nurture first-class business -leaders and cultivate their global perspective with intimate understanding of -China. Students will spend the first two years at HKU and the third and fourth -years at Peking University (PKU), and be awarded a Bachelor of Economics and -Finance (BEcon&Fin) degree from HKU and a Bachelor of Management [BMgmt] -degree from PKU upon successful completion of the dual degree programme. +• Customers no longer want to phone when -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. +they have an issue, they want an +immediate answer and turn to social. -Find out more: -https://ug.hkubs.hku.hk/admission/degree-program -mes/hku-pku-dual-degree +Reduce cost to serve +• With customised chatbots and the use of +artificial intelligence on the rise, in-app +messaging is becoming the norm for +customer care -Bachelor of Finance in Asset Management and Private -Banking [BFin(AMPB)] -The BFin(AMPB) programme aims to prepare students to enter the rapidly -growing fields of asset management and private wealth management in Hong -Kong and Asia Pacific, or in banking and finance generally. During the first two -years, students complete introductory courses in accounting, economics, -mathematics, statistics, corporate finance, and investments and portfolio -analysis. In subsequent years, students take more specified and industry -related courses including equity valuation and investment in specific asset -classes e.g. private equity, venture capital, hedge funds and digital assets, -understanding of financial statements and doing financial analysis, lending -and credit in private wealth and corporate bank, financial regulations and -compliance, current -institutional -in finance, ESG -consulting and family office. The programme contains practical-oriented -courses taught by teachers with extensive experience in the finance industry -and offers an array of career-oriented enrichment activities and networking -opportunities with industry practitioners. +• Technology solutions enable improved -investment, +customer service through social -issues +Build a community +• Customers trust each other more than +you. Providing them with a platform +where they can interact and share +feedback will increase their trust and +loyalty to your brand, coaches, or venues. -in +9 -BFin(AMPB) Curriculum Structure + WHAT DOES GOOD LOOK LIKE? -Components +H O W T O W I N W I T H S O C I A L M E D I A -Credits +A great social media strategy includes the following pillars: -A +SOCIAL MEDIA +PILLARS -B +Emotional content -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Build rapport with your customers. Forget transactional posts +and focus on posting content that showcases how you will make +a difference in your customers’ life -Major in Asset Management and Private Banking -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Consistency -C +Your content should match your brand identity and objectives +and be consistent across channels. -Free Electives / Optional Major / Optional Minor(s) +Regular & well- +timed posts -54 -12 -36 -6 -- +Just like any other channel, there is a better time to post content, +and that varies depending on the network and audience. Keep +your customers interested by posting often and sharing varied +content -96 -24 -42 -24 -6 +Two-way +engagement -90 +Social is about more than just pushing out content. Allow your +customers to interact with you and make sure you respond to +create a differentiated experience -Total +Optimisation -240 +Test, learn, adjust, repeat. Iteration is key to success and social +is no different. Plan for A/B testing and track success of +individual campaigns and performance across platforms -Find out more: -https://ug.hkubs.hku.hk/programme/becon-and-fin +Personalisation -Find out more: -https://ug.hkubs.hku.hk/programme/bfin-ampb +Strategic use of +channels -1312 +Use social insights to know who your customers are and adapt +your messaging, including graphics and tone of voice to target +the right customers with the right message at the right time - Bachelor of Science in Marketing Analytics and Technology -[BSc(MAT)] +Not all channels will be relevant to your brand. Whilst a presence +on the Big 4 is a must, focusing your budget and efforts on the +channels where your customers are the most active will deliver +better value for money -Bachelor of Science in Quantitative Finance [BSc(QFin)] +Overall strategy +alignment -Marketing is critical to all businesses. It dictates whether a company will be -successful. Technology, Big Data, and digitization are fundamentally changing -how marketers operate. The BSc(MAT) programme equips students with -quantitative skills, technology know-how, and business marketing acumen to -pursue a career in the digital economy and modern businesses. It targets both -STEM students with an interest in business as well as business students -interested in the technology space. Students will gain a solid foundation in -cutting-edge technical areas including computer programming, data science, -and technology by excelling in courses from the Faculty of Engineering and -Faculty of Science. In addition, students will master digital business, -marketing and leadership through courses at the HKU Business School. This -interdisciplinary approach will enable students to engage in a variety of jobs -and functions such as the design and launch of new technology products, -targeting consumers, digital marketing, forecast and predicting markets, and -building a technology-focused strategy. The programme aims to nurture -future leaders in the field of technology and marketing who are able to use the -latest technology, science, and data analytics to support business and -marketing decisions in the digital world. +Driving revenue and engagement from social does require +investment. Incorporate social as its own channel into your +overall strategy, budget and resources -BSc(MAT) Curriculum Structure +10 -Components + WHAT DOES GOOD LOOK LIKE? -Credits +P U T T I N G S O C I A L M E D I A P I L L A R S I N T O P R A C T I S E -A +Conversational & Engaging Content – +Royal Academy & Aldi -B +Influencer & Interest-based Marketing – +Gymshark & Wilson -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Royal Academy uses relatable and +humourous posts and interactive +campaigns such as their +#RAdailydoodle to keep their 441K +Twitter followers engaged, drawing +increased user participation and +personalised well-crafted responses -Major in Marketing Analytics and Technology -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives (Marketing Technology and Strategy) -Disciplinary Electives (Marketing Analytics) -Capstone Course +Aldi UK celebrated their birthday with +a simple, yet creative invitation to +engage competitors, generating a ton +of interest on Twitter and a positive +response, well-tracked via their +unique hashtag +#Aldi30thBirthdayParty -C +Gymshark sponsors key Instagram +influencers to capitalise on their large +audiences (500k+). They also use paid +social as a primary means to drive +sales, and in 2019 saw a £73m increase +in turnover. In 2017, their Black Friday +posts drew a 6.6x return on ad spend. -Free Electives / Optional Major / Optional Minor(s) +Influencer marketing can also reap +benefits on a smaller scale. Like the +Germany-based Tennis Siblings Alex +& Sofi, who are sponsored by Wilson +and adidas and use their modest +following of 15K to promote products +and local tennis tournaments. -54 -12 -36 -6 -- +11 -96 -24 -42 -12 -12 -6 + EXECUTING A CAMPAIGN - WHERE TO START -90 +G E T T I N G T H E B A S I C S R I G H T -Total +2 -240 +Identify your audience +and grow followers -BSc(QFin) is a premium finance programme. Built on a solid foundation in -economics and finance, the programme equips students with the essential -quant skills, including mathematics, statistics, and programming, that are -required in more advanced and sophisticated analyses, such as valuation -models of structured products, risk management, simulation and data -analysis, financial engineering, and quantitative trading. +Start following your members and +people who “look like” them (e.g. +have similar demo, interests, +location, etc.) -The programme goes beyond merely equipping students with quant skills. -More importantly, it stresses on combining these skills with an understanding of -the financial markets, the macro environment, and the practical situations -where financial managers work, with a balanced mix of quant and business -courses. To nurture students’ soft skills and market acumen, the programme -provides a wide range of value-enhancing activities, such as Round Tables with -Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training -workshops, simulation trading workshops, and an international field trip to -other international financial centres such as New York, London and Zurich. +Plan a content calendar and post a +variety of engaging content +including text, photo, and video. -BSc(QFin) Curriculum Structure +Track performance to see what +content resonates best with your +audience. -Components +3 -Credits +Launch and track your +campaign -A +You can begin with a simple +campaign – such as amplifying a +piece of existing content that +performed well with your existing +followers. -B +Set up campaigns by objective +(page likes, clicks, etc.) and set a +daily or monthly budget. -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Use built-in dashboards from the +social platforms to review +performance and optimise by ad +frequency, time of day, ad format, +and more. -Major in Quantitative Finance -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +1 -C +Set up your account +page(s) -Free Electives / Optional Major / Optional Minor(s) +Establish your presence by +setting up business account / club +page across the relevant social +platforms -54 -12 -36 -6 -- +Include essential business +information for search purposes, +including: Location, Contact info, +and Website URL + +#Tip: use hashtags to +start a conversation with +your members and easily +follow what they say about +you and your club + +4 + +Scale up and add on +to master social + +Once you have your first campaign +under your belt, report on your +performance and measure against +your original business KPIs. + +Refresh your content and launch +new campaigns for different target +audiences. + +Explore and introduce more +complex functionality available to +you, including new data sources, +more robust tracking (e.g. FB pixel), +new ad formats, or new social +platforms. -96 -24 -54 12 -6 - -90 - -Total - -240 - -Find out more: -https://ug.hkubs.hku.hk/programme/bsc-mat - -Find out more: -https://ug.hkubs.hku.hk/programme/bsc-qfin - -1514 - - Enrich Your Life - -initiative and -HKU Business School fosters a culture of student -implementation of ideas, both within and outside the campus, as it is -these real-world engagements that set us apart. Experiences in learning -beyond the classroom, extending into realms outside academia, enable -students to cultivate an understanding of social dynamics, cultural -diversity, and collaborative -intellect. These experiences contribute -significantly to their personal growth, interpersonal skills, and social -development, shaping them into global citizens, lifelong learners, and -future business leaders. - -Overseas Enrichment Experiences - -Business Exchange Programme / -HKUWW Undergraduate Student Exchange Programme -Even as the youngest Faculty at the University, our exchange -activities are nothing but phenomenal. Each year, over 400 -students participate in overseas exchanges at leading universities -to broaden their horizons. - -Business School -Exchange Partners * - -Canada - -McGill University -University of British Columbia -University of Toronto -Western University -Simon Fraser University - -U.K. - -King's College London -University of Manchester -University of Edinburgh -University of Warwick - -The Netherlands - -University of Amsterdam -Erasmus University Rotterdam - -400+ -exchange partners -across 40+ countries - -400+ -Business School -students on exchange - -Denmark - -Copenhagen Business School -Aarhus University - -Germany - -Frankfurt School of Finance & Management -University of Mannheim -Munich Business School - -Belgium - -KU Leuven - -Italy - -Bocconi University - -France - -U.S. - -Toulouse School of Economics -ESSEC Business School - -New York University -University of Southern California -University of Washington -University of Georgia - -Middle East - -Reichman University - -Singapore - -National University of Singapore -Nanyang Technological University - -* This list is non-exclusive. - -International Field Trips -To enhance students' exposure to global businesses, the School organises -overseas field trips for students visiting corporate firms and mingling with -C-suite Executives in some of the world's most prominent financial -centres, e.g. Australia, France, Japan, London, New York City, Singapore, -Spain, and Zurich. - -Visiting Programmes -Through the Yale Visiting International Students Programme, students -are sent to spend a year at Yale University, where they actively engage -themselves in a multicultural society. With the generosity of the donors, -some of our students are selected for the Junior Year Abroad Programme -and Visiting Students Programme to study as visiting students at -Pembroke College and St Peter’s College of University of Oxford -respectively with substantial scholarships. The intellectual ambience of -these top universities is truly inspirational and conducive to learning. - -Beta Gamma Sigma Regional Leadership Conference -Beta Gamma Sigma (BGS) is the international honour society that -serves business programmes accredited by the Association to Advance -Collegiate Schools of Business (AACSB International). With the -establishment of the HKU BGS Chapter, the School registers students -annually for the BGS Regional Leadership Conference held in the U.S., -where they can meet and bond with many other fellow students from the -world's best business schools. The Leadership Summit is eye-opening -and challenging, which leads to a great journey of self-discovery. - -Japan - -Kyoto University - -Australia - -University of Sydney -Monash University -University of New South Wales - -1716 Student Sharing - -Chen, Hsuan-yu -Exchange at National University of Singapore (Singapore), 2024-25 - -As a third-year undergraduate student pursuing a double major in -Accounting and Finance, I consider it a tremendous privilege to have -participated in a one-semester exchange at the National University of -Singapore (NUS) through the Business Exchange Programme. This -invaluable opportunity has not only enriched my academic journey but -also broadened my global perspective through immersive cultural -experiences and meaningful cross-cultural interactions. As I move -forward in my academic and professional journey, I carry with me not -just the knowledge gained from my courses, but the invaluable life -lessons experienced from Exchange. - -Lee, Tsz To Derek -Exchange at King’s College London (United Kingdom), 2024-25 - -I am incredibly grateful to have this valuable opportunity to study -abroad at King’s College London. The programme provided me a chance -to venture outside of my comfort zone. Not only can you learn how to be -more independent, you can also boost interpersonal skills given the -many opportunities to socialise with others, which is essential for your -future career. The experience is remarkable. I think you will regret it if -you miss out on the overseas exchange programme. - -Cheung, Nga Ting Caroline -Exchange at University of New South Wales (Australia), 2024-25 - -This exchange programme has changed me in many ways. It has helped -me become more outgoing, allowing me to make new friends and -explore different cities. Living independently has been a significant part -of this journey. From exploring new cities to forming lasting friendships, -each adventure has enriched my life in ways I never expected. This -opportunity has not only broadened my horizons but also deepened my -appreciation for diversity and connection among people from different -backgrounds. Each memory I have created during this time is a treasure -that I will carry with me forever. - -Ho, Ying Tung Crystal -Exchange at Copenhagen Business School (Denmark), 2024-25 - -In the Fall semester, I participated in a business exchange programme at -Denmark’s Copenhagen Business School (CBS). This experience was -truly transformative, enriching me both academically and personally -while allowing me to forge lifelong memories and friendships. An -exchange experience abroad truly broadens your horizons and offers -insights beyond what textbooks and lectures can provide. You will gain -invaluable perspectives and undergo personal growth like never before. -I wholeheartedly encourage you to muster your courage and embark on -an exchange journey - you won’t regret it! - -Whitfield III, Gerald Bruce Trey -Exchange at New York University (United States), 2024-25 - -I found the exchange experience to be extremely enriching. A major -highlight for me was exploring the lifestyle and professional sides of the -city. Because of my initiative socially, and in groups such as church, I was -able to get well-acquainted with the local rhythms and lifestyle, and -also made many friends that I’d consider friends for a lifetime. I am -deeply appreciative of being given this rewarding experience, which -significantly shaped both my future and the person I am today. - -Ip, Wing Tung Alla -Student Representative for the BGS Regional Leadership -Conference 2025 in Highland Heights, Kentucky - -Attending the BGS Leadership Conference at Northern Kentucky -University proved transformative. I received profound insights into -leadership, communication and cultural intelligence through immersive -workshops, case competitions and networking opportunities. The event -prompted reflections on personal leadership styles and encouraged -collaboration across diverse peers. As I prepare to enter the workforce, -I intend to apply these learned principles of trust-building, cultural -sensitivity and collaborative leadership to navigate challenges so I can -shape my professional journey with purpose and authenticity. - -1918 Corporate Exposure - -HKU Business School has a strong tie with corporate partners. To nurture students' business -and market acumen, enrichment programmes are organised through the joint efforts of -corporate partnerships. - -Enrichment Programmes in collaboration with Corporate Partners -HKU Business School has been collaborating with corporate partners on initiatives to spice up -students' career exploration journey. The partnership with Accenture enables students to -gain exclusive insights into business consulting through the HKU-Accenture Business -Consulting Programme. - -To provide students with a unique opportunity to explore current business trends and -developments in the Greater Bay Area (GBA), our School has co-organised a tailored study tour -with The Association of Chartered Certified Accountants (ACCA). This initiative aims to -enhance students’ readiness for future career development and personal success. Students will -have the chance to connect with seasoned accounting and business professionals in the GBA. - -HKU Business School Mentorship Programme -Leveraging the School's extensive network in the business and finance industry, we have -developed our own mentorship programme to furnish students with opportunities to -enhance their skills and knowledge essential for their future personal and career -development. Our students are empowered to connect to the real business world through -direct and active interaction with seasoned professionals, executives and entrepreneurs. - -Lee, Lok Yiu Angel -HKU Business School Mentorship Programme 2024-25 - -My mentor offered -invaluable career advice, highlighting key -considerations for entering specific fields and preparing for interviews. -Her experiences and mindset during job searches resonated with me as -I navigated something similar. She emphasised the importance of -open-mindedness and eagerness to learn in the workplace. I kept her -words at heart during my internship. I also enjoyed catching up with her -to share my new job experiences. - -Choi, Han Na Hannah -HKUBS x Deloitte China Mentorship Programme 2024-25 - -As my mentor has been working in Deloitte China for more than 10 -years, I was most curious about how he maintains a work-life balance -and how he stays dedicated and professional at work. Through our -lengthy conversations over lunch, he passed on valuable insights on -developing soft skills and adopting an appropriate work attitude at an -early stage. He is more than an audit mentor but also a life coach to me. - -Chan, Chin Ching Jacky -BSc(QFin) Executive Mentorship Programme 2024-25 - -Throughout the mentorship programme, I have received -precious guidance that has greatly influenced my career -path. My mentor highlighted the importance of identifying -a company that aligns with my goals and would provide -opportunities for me to develop the necessary skill sets. He -also shared practical advice on building and managing -professional and personal relationships. These lessons -have broadened my perspective and also equipped me -with the tools to handle real-world situations more -effectively and confidently. - -Wong, Wan Tat -HKUBS x ACCA Greater Bay Area Study Tour 2024-25 - -My greatest takeaway from this study tour was the power of peer -learning and collaboration. Being immersed in an environment with a -diverse group of individuals, all focused on similar goals, created an -incredible opportunity to learn from each other’s experiences and -perspectives. The collaborative projects and informal discussions were -invaluable in broadening my horizon and developing new approaches -to the accounting field in Chinese Mainland. I am leaving with a strong -network of contacts and a renewed appreciation for teamwork. - -Leung, Ho Lun Maverick -HKU-Accenture Business Consulting Programme 2024-25 - -I visited - -The HKU-Accenture Business Consulting Programme immersed me in -an authentic consulting experience through working with a real client – -their distribution centres, conducted market -HKTV. -questionnaires, and presented the final deliverables - all within the -eight-week programme. I was honoured to meet Managing Directors -from Accenture who shared crucial skills in conducting a consulting -project. It’s truly one of the best programmes for anyone considering a -consulting career! - -2120 Experiential Learning Programmes - -Experiential learning is embedded in our course curriculum. The Business Consulting -Practicum offers business and economics students hands-on experience in working with -clients to solve "real-life" problems relevant to their studies, under the guidance of faculty -members and professional mentors. The Business Case Analysis and Presentation enables -students to communicate and seek practical advice from senior executives through in-class -case presentations. The Impact Lab provides students with an excellent opportunity to work -directly with social ventures, again under the guidance of teachers and professional mentors. -Additionally, students can gain an in-depth understanding of the latest trends in economics, -finance, creativity, innovation, and entrepreneurship in the rapidly growing Greater Bay Area -by participating in the Creativity, Innovation & Entrepreneurship in China Programme. - -Chaabane, Adam Andrea -"Creativity, Innovation & Entrepreneurship in China" -Summer Virtual Programme 2025 - -I have learned that prominent businessmen and professionals in Hong -Kong are people like me who were once attending business school. I -most enjoyed the guest -lectures because they allowed direct -interactions with inspiring figures. The virtual experience enabled me to -follow this programme easily from my home country. I definitely -recommend it for a better understanding of Hong Kong and GBA’s -markets and opportunities that arise in cross-border business. It would -be great to learn more about opportunities for foreigners in the region. - -Kaur, Suchleen -Business Consulting Practicum 2024 - -The opportunity for real-time interaction and feedback with the client had -propelled me to join BCP. I was always curious about how a consultant -would address and communicate with the client. BCP has taught me -about proper business etiquette and equipped me the necessary skills to -negotiate conflicting opinions during team discussions. Throughout the -process, our mentor and professors provided great help and insights. -Their invaluable feedback and questions helped nudge us in the right -direction, taking a step back to look at the big picture and most -importantly, prioritise the client’s needs. - -Zhou, Qingzhuo -"Creativity, Innovation & Entrepreneurship in China" -Summer Virtual Programme 2025 - -My greatest takeaway from the CIEC 2025 programme was the unique -window to gain a comprehensive understanding of the Greater Bay -Area’s dynamic business landscape. I explored the latest trends in -business, finance, and innovation from guest talks and virtual lectures -with industry leaders. I particularly appreciate the coverage on a -diverse range of topics and the opportunity to earn HKU credits. A -highlight was the cultural exploration segment which offered an -unparalleled perspective on the Greater Bay Area. I recommend this -programme without hesitation. - -Tsang, Tze Yeung Matthew -Business Consulting Practicum 2024 - -BCP has been a rewarding experience that allowed me to apply my -academic knowledge to real-world business challenges. Working -directly with my mentor and client has pushed me to think critically and -creatively in ways I hadn't experienced in the classroom. I'm confident -that the problem-solving, communication, and client-facing skills I have -developed will serve me well when pursuing my future career in -consulting or another business-oriented field. - -2322 Cheung, Ho Ning Jolie -UW Global Business Case Competition 2025 (Champion) - -in UW GBCC was an - -Competing -incredibly rewarding and -unforgettable experience. We had the opportunity to work with -participants from different universities. -It was a fantastic -opportunity to exchange perspectives and learn from different -problem-solving approaches. Working under tight timeframes -pushed us to think creatively. UW GBCC has not only elevated my -professional capabilities but further solidified my belief in business -as a powerful vehicle for creating positive change in the world. - -Case Competitions - -The School provides ample opportunities to refine students' presentation and analytical skills -through a myriad of local as well as overseas competitions. From training to delicate analysis -of business cases to a variety of business scenarios, students are equipped with different -skillsets to compete with candidates from the world's leading universities. - -Lo, Nikkia -NHH International Case Competition 2024 (Champion) - -My week at NHH International Case Competition has been an -unforgettable rollercoaster of emotions. From the excitement of -exploring a new Nordic city, the exhaustion of our 24-hour case -competition, to the nervous anticipation of pitching on a global stage -with a -livestream, every moment was extraordinary. Meeting -like-minded peers around the world was an incredible experience that -forged new friendships. Thank you, HKU Business School, along with -our professors, coaches and staff for bringing this opportunity to us! - -Xia, Yunchu Kevin -The McDonough Business Strategy Challenge 2025 (First Place) - -Winning first place at MBSC2025 was incredible and deeply rewarding. -My team was highly engaged. We were constantly bouncing ideas off -each other and refining our strategy. There were moments where we hit -roadblocks, but our team rallied, challenged each other constructively -and developed a cohesive solution. Connecting with like-minded -participants, judges and NGOs was also invaluable. MBSC not only -sharpened my analytical and presentation skills but also reaffirmed my -commitment to leveraging business for good. I am so grateful for the -experience and very eager to apply these lessons moving forward! - -2524 Liu, Yunqi Iris -Deloitte Tax Championship 2024 (Outstanding Team Award) - -This competition has been a fantastic learning experience. It -showed me that tax isn’t just about the technical details - it’s also -deeply connected to social trends and global events like climate -change, geopolitical conflicts and national policies. This broader -perspective has enhanced my understanding of the industry and -its importance in today’s world. I want to express my sincere -gratitude to my advisors for their guidance and recognition of my -team’s hard work. I am grateful for the opportunity to learn and -grow with such an amazing group of people. - -Pan, Ho Man Vivian -HKGCC Business Case Competition 2024/25 (Champion) - -in - -the HKGCC Case Competition was a - -Participating -truly -transformative experience that greatly enhanced my strategic -thinking, teamwork and problem-solving skills. The competition -challenged me to analyse intricate business problems, think critically -under pressure and develop innovative solutions. Working closely -with people of various majors allowed me to appreciate differing -opinions that were at times conflicting, find ways to build consensus -and synthesize ideas effectively. - -Students' achievements in case competitions -(2023-2025 awards) - -Overseas / Chinese Mainland - -Asia Pacific Sustainable Marketing Case Competition 2025 - -Third Place - -Bloomberg Global Trading Challenge 2023 - -CBS Case Competition 2023 - -CFA Institute Research Challenge 2025 - Asia Pacific Regional Final - -CFA Institute Research Challenge 2025 - Sub-Regional Final - -CGMA Global Business Challenge 2023 - North Asia Final - -Champions' Trophy Case Competition 2024 - -Grand Prize Winner -Regional Prize Winner: Asia & Oceania - -Winner - -Winner - -Winner - -Champion - -Finalist - -China Mainland, Hong Kong and Macao Digital Economy Innovation and -Entrepreneurship Competition 2024 - -2nd Runner-up (Gold Medal) - -Deloitte Digital Camp 2024 - -Deloitte Digital Camp 2023 - -Second Place - -Champion - -Deloitte Tax Championship 2024 - National Round - -Outstanding Team Award - -Deloitte Tax Championship 2024 - Regional Round - -Deloitte Tax Championship 2023 - National Round - -Deloitte Tax Championship 2023 - Regional Round - -Top 3 teams - -Merit Award - -Top 3 teams - -Family Office Association Hong Kong (FOAHK) 2023 Case Competition - -2nd Runner-up - -HSBC Global Private Banking Case Challenge 2023 - -Investing in the Future University Competition 2023 -by Pictet Asset Management - -John Molson Undergraduate Case Competition 2025 - -Lazaridis International Case Conference 2023 - -Marshall International Case Competition 2024 - -McDonough Business Strategy Challenge 2025 - -NHH International Case Competition 2024 - -Scotiabank International Case Competition 2023 - -Sydney International Business Competition 2025 - -Thammasat Undergraduate Business Challenge 2024 - -The Third KPMG ESG Case Study Competition - -UW Global Business Case Competition 2025 - -UW Global Business Case Competition 2023 - -1st Runner-up - -2nd Runner-up - -Finalist - -1st Runner-up - -Finalist - -First Place - -Champion - -Winner - -Third Place - -1st Runner-up - -Third Prize - -Champion - -Finalists - -27 - -26 Local - -Local - -100WF Hong Kong JumpStart University Investment Hackathon 2023 - -1st Runner-up - -HKSI Institute Case Competition 2024 - -Champion - -ACCA Hong Kong Business Competition 2023 - -Outstanding Team for start-up Filix -Medtech - -Team Excellence Award - -Merit Award for start-up MakeOmnia - -Alvarez & Marsal Hong Kong Business Case Competition 2024 - -Champion & Best Presenter - -HKSI Institute Case Competition 2023 - -Best Performer of Roma Risk Advisory -Limited - -1st Runner-up - -Merit - -Champion - -1st Runner-up - -Merit & Finalist - -Semi-finalist - -Champion - -2nd Runner-up - -Best Presenter - -1st Runner-up - -HKU-Airstar Bank Joint University Case Competition 2024 - -HKU-SaSa Joint University Case Competition 2023 - -Hong Kong MBS Digital Economy Business Simulation Competition 2025 - -Third Place - -HSBC Global Private Banking Case Challenge 2024 - -HSBC Life Insurance Competition 2025 - -HSBC Life Insurance Innovation Competition 2023 - -1st Runner-up - -Top 3 Winner - -1st Runner-up - -Best Use of Technology Award - -International MBS Business Management Simulation Competition 2023 - -Champion - -ISM-HK Student Case Competition 2025 - -ISM-HK Student Case Competition 2024 - -MBS Simulated Business Management Competition 2024 - -New Century Elite Development Programme 2023 - -Venture Capital On Campus (VCOC) - Investment Challenge 2023 - -1st Runner-up - -Champion - -Third Place - -First place - -Best presenter - -Third Place - -Champion - -1st Runner-up - -2nd Runner-up - -Winner Team - -29 - -AmplifyME APAC M&A APAC Challenge 2025 - -Third Place - -BOCHK Challenge 2023 - -Bronze Award for Innovative Design - -CGMA Global Business Challenge 2025 (Regional Final) - -1st Runner-up - -CGMA Global Business Challenge 2023 (Semi-final Round) - -Deloitte USHINE Case Competition 2024 - -ESG Challenge 2023 - -EY Young Tax Professional of the Year 2023 - Hong Kong Region - -Champion - -Finalist - -Champion - -Champion - -1st Runner-up - -2nd Runner-up - -Family Office Association Hong Kong (FOAHK) 2025 Case Competition - -Champion - -HKCGI Corporate Governance Paper Competition and -Presentation Awards 2024 - -HKGCC Business Case Competition 2023/24 - -HKGCC Business Case Competition 2024/25 - -28 - -HKGCC Pitch Perfect Programme 2024 - -HKICPA Business Case Competition 2023 - -HKSI Institute Case Competition 2025 - -1st Runner-up - -1st Runner-up (Paper Writing Competition) - -1st Runner-up (Paper Presentation -Competition) - -Audience’s Favourite Team Award - -Chevalier - Champion - -Chinachem Group - Champion - -Swire Pacific - Champion - -Chevalier - 1st Runner-up - -Chinachem - 1st Runner-up - -HKGCC - Champion - -Jardine Matheson - Champion - -SOCAM - Champion - -Jardine Matheson - 1st Runner-up - -MTR - 2nd Runner-up - -Overall Champion - -Winner - -2nd Runner-up - -Champion - -1st Runner-up - -2nd Runner-up - -Merit - - Chhawchharia, Mihika Amit -HSBC - Summer Analyst program - -internship, - -I collaborated closely with various -Throughout my -departments of the bank and -interacted regularly with senior -management, gaining comprehensive exposure to the bank’s operations. -This experience allowed me to hone my professional skills, including -communication, leadership, financial analysis, and market expertise. I -was honored to be ranked as the top intern in my cohort and secured a -return offer to join HSBC as a full-time graduate analyst in their Global -Banking division. I am deeply grateful for the knowledge and support I -received at HKU, which enabled me to thrive in a prestigious global -institution alongside some of the brightest talents in the industry. - -Internship Experience - -Lee, Chung Hong Calvin -HSBC - Global Risk Management Intern - -Reflecting on my time at HSBC, I realise that I have grown on a -professional and personal level. I have developed a deeper passion for -the banking industry and a greater appreciation for the critical role that -risk management and compliance play. I wore many hats and gained -precious insights into the banking industry. I was responsible for -overseeing compliance risks across 32 markets in Asia Pacific. I am -confident that the lessons I have learned at HSBC will guide me every -step of the way. - -Hui, Matthew -HSBC - International Banking (ISB) Intern - Corporate Coverage - -The comprehensive exposure to Corporate Banking was invaluable. From -analysing credit risk for lending applications to understanding diverse -product offerings for MNC clients, each day brought new learning -opportunities. I'm truly grateful for the experience of shadowing and -supporting different RMs in the team. Together with the sharing and -introductions from product partners, I learned how to identify clients' -needs and gained perspective on effectively reaching out and pitching to -clients. This experience taught me how to view banking relationships -through a strategic lens - understanding not just immediate needs but -trends. The -also anticipating -professional mindset, strategic thinking, and collaborative spirit fostered -during this internship will undoubtedly shape my future career path. - -future requirements and market - -Nagar, Tanvi -HSBC - International Banking Intern - -immense - -My internship experience was nothing short of incredible. Not only did I -grow immensely in my technical and professional skills but also -underwent an -journey of personal growth. I received -unparalleled support from the team and my mentors and gained an -in-depth insight into the field of banking and finance in Hong Kong. I -formed meaningful connections and grew my network. This experience -has truly shaped me and continues to propel my drive to pursue a career -in Banking in the years to come. - -3130 Sharma, Seerat -KPMG - Financial Services Audit Intern -(COOP3000 Cooperative Education Programme) - -I had the privilege of contributing to audit engagements for leading -clients in the banking and insurance sectors, where I conducted -substantive testing, supported risk assessments and navigated audit -methodologies using advanced tools. This role sharpened my analytical -mindset, strengthened my collaboration skills and offered a front-row -view into the dynamic world of financial auditing. It’s been a truly -rewarding journey that has solidified my passion for a career in -professional services. - -Xu, Aijing -CMB International Capital Corporation - Wealth Management Intern - -During my internship, I actively engaged with the product team in -market analysis and client pitching focused on mutual funds. I gained -valuable insights into various asset classes and macroeconomic -trends. Analyzing mutual fund performances and participating in -client meetings allowed me to contribute to investment portfolio -strategies. Collaborating with -team enhanced my -communication skills and understanding of client needs. Overall, this -experience deepened my knowledge of wealth management and -strengthened my passion for finance. - -the sales - -Shi, Yu Cheung Karson -Bank of China - Credit Summer Intern (Risk Management) - -During the internship, the financial techniques and business knowledge -I learned from HKU Business School courses helped a lot in company -financial analysis. By applying knowledge taught in the financial -accounting module, I assisted in evaluating loan demand. Through -cross-departmental collaborations, I also realised how classroom -simulations mirrored real-world risk mitigation challenges. This -experience reinforced the value of integrating academic theories with -dynamic market practices. - -Hu, Wen Qing -PwC Assurance Practice Intern - -As an assurance intern in PwC, I assisted my seniors with sampling -financial statement line items, reconciling balances, and documented -relevant non-financial information. It showed me how demanding the -auditing industry was, and allowed me to set my expectations for a -career in auditing as an accounting student. - -Wirawen, Karen -HKU – Neuberger Berman Private Equity Internship Programme - -You never truly understand private equity until you’re immersed in it -and my internship at Neuberger Berman was deeply enriching. Coming -in with limited knowledge, I gained direct exposure by working on -various deals across sectors and geographies, including Japan, Korea -and Southeast Asia. From drafting memos for the -investment -committee to attending annual general meetings and taking notes on -calls, I progressed from simple tasks to meaningful responsibilities. The -seniors were welcoming, always open to teaching, providing feedback -and offering valuable career advice. - -Chung, Marcus Lok Sum -Morgan Stanley Industrial Placement Intern - -During my internship with Sales & Trading (Institutional Equity -Division), I had the amazing opportunity to rotate across trading -and structuring desks where I was exposed to the fast-paced -environment especially during periods of high market volatility. To -see how traders manage positions and mitigate risks first-hand -was incredibly insightful. I also contributed to constructing trading -strategies and seeing the tangible impact of our work was deeply -rewarding. Morgan Stanley definitely provided me with a broader -perspective and understanding of market dynamics, liquidity, -client interactions and cross-asset opportunities. - -3332 Our Student Scholars - -Chan, Kwan Yin Cheryl -Recipient of Louisa Cheang Inspiration Scholarship 2024-25 - -I would like to express my gratitude for the Louisa Cheang Inspiration -Scholarship, which has significantly contributed to my personal and -professional growth. In the past academic year, this scholarship enabled -in the full-time Registered Visiting Student -me to participate -Programme at St Catherine’s College, University of Oxford. This -immersive experience enhanced my legal studies through engaging -discussions with scholars and participation in mooting competitions, -sharpening my research and advocacy skills. Additionally, I plan to -utilise this scholarship to pursue overseas postgraduate studies, -allowing me to deepen my understanding of the complexities of law in -a global context. - -Hussain, Iman -Recipient of FBE Student Contribution and -Innovation Award 2024-25 / -HKU Business School Future Leader Scholarship 2024-25 - -As someone who believes in learning through real-world experience, -the FBE Student Contribution and Innovation Award and the HKU -Business School Future Leader Scholarship have been life-changing. -They not only enabled me to go on an exchange semester to Cardiff, -where I explored new cultures and perspectives but also shaped my -entrepreneurial journey. - -With this support, I’ve been able to work on building a social -enterprise that aligns with my passion for creating meaningful impact. -I’m incredibly thankful for these opportunities to grow and make a -difference. - -Chow, Yee Shun Eason -Recipient of the Jardine HKU Scholarship 2024-25 - -It is an incredible honour to have received the Jardine HKU Scholarship -which is enriching my university journey! This scholarship allowed me to -pursue my many goals – from attending international training camps as -a handball player to participating in overseas exchanges without -financial worries. I have had the privilege of undertaking a summer -internship at the Jardine Group, learning from seasoned professionals in -my future industry. The Jardine Foundation also hosted exciting -networking activities and site visits which helped me build lasting -connections with fellow scholars in a vibrant community. - -Shih, Pui Hei Iris -Recipient of the Hong Kong Jockey Club Scholarships - -Undergraduate Scholarship 2024-25 - -This scholarship has been transformative to my university life as it has -allowed me to fully embrace and explore the world beyond the -classroom. It provided the financial stability I needed to focus on my -studies and presented opportunities for me to participate in diverse -extracurricular activities. These experiences broadened my perspective -and opened me to different points of view. - -Nurgali, Adilzhan -Recipient of BOC Hong Kong - -One Belt One Road Scholarship 2024-25 - -It means a lot to me to be awarded the BOC Hong Kong - One Belt One -Road Scholarship. It was not -just about financial support but -recognition of the hard work I have put in. As an international student -soon heading for exchange, this award gives me the freedom to explore -new cultures and experiences without extra stress. I am truly grateful to -be part of such an inclusive and supportive community. - -3534 2026 JUPAS Admissions - -Admissions criteria (HKDSE) - -Local candidates applying on the strength of the Hong Kong Diploma of -Secondary Education (HKDSE) are required to fulfil the minimum entrance -requirements for HKU Business School programmes as shown below: - -Level 4 -Level 4 - -EngEng ≥ -Level 4 or above in English Language -ChinChin ≥ -Level 3 or above in Chinese Language1 -MathMath ≥ -Level 3 or above in Mathematics - -Level 3 -Level 3 - -Level 3 -Level 3 - -AA - -CSDCSD = -Grade “Attained (A)” in -Citizenship and Social Development2 - -2x - -Elective -Elective -Subjects -Subjects - -≥ - -Level 3 -Level 3 - -Level 3 or above in two Elective Subjects - -For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) -In addition to the minimum entrance requirements above, candidates applying for the BBA(BA), -BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the -HKDSE examination: - -BBA(BA) / BSc(MAT) -≥ -Level 3 or above in one of the following subjects: - -Level 3 - -Biology - -BAFS - -Chemistry - -Economics - -ICT - -Physics - -BBA(BA) - -BSc(MAT) - -Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not -strictly required. - -BBA(Law)&LLB / BBA(IBGM) / BSc(QFin) - -BBA(Law)&LLB - -BBA(IBGM) - -BSc(QFin) - -EngEng ≥ - -Level 5 -Level 5 - -Chin, Math -Chin, Math ≥ - -Level 4 -Level 4 - -EngEng ≥ - -Level 5 -Level 5 - -+ - -+ - -M1/M2 -M1/M2 ≥ - -Level 3 -Level 3 - -MathMath ≥ - -Level 4 -Level 4 - -Application - -For details of the 2026 JUPAS exercise, please visit www.jupas.edu.hk. -Programme Scoring Formula + CLUB SUCCESS STORY -For more information on the Programme Scoring Formula and Subject Weightings, please -visit the Admissions Office’s website at https://admissions.hku.hk/apply/jupas. +N O R B U R Y P A R K L T C -Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis.Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered.Level 5 or above inEnglish LanguageLevel 4 or above inChinese Language1 & MathematicsLevel 5 or above inEnglish LanguageLevel 4 or above inMathematicsLevel 3 or above in Extended Module 1 or 2 of Mathematics3736 2026 Non-JUPAS Admissions +SITUATION -Selection process and entrance requirements +Norbury Park LTC was established in 1889 in the London Borough of Croydon. It +has four floodlit tarmac courts and a small management committee made up +entirely of volunteers. Club membership was at an all-time low so the committee +put together an action plan to significantly increase membership within a year. -Eligibility -(1) Local students applying for admission on international qualifications other than the +APPROACH -HKDSE examination; +The club recruited a volunteer Marketing and Communications Officer and decided +to run a Nature Valley Big Tennis Weekend (open day) to showcase the club’s +facilities to the local community and engage prospects. -(2) Non-local students with international qualifications other than Chinese Mainland's +Using free resources and content available on My Tennis Toolkit, they created a +Facebook event and advert with a call-to-action to sign up for taster sessions. -National Joint College Entrance Examination (NJCEE/Gao Kao). +OUTCOME -Application Procedures -All non-JUPAS applications should be submitted via the online application system by noon of -November 26, 2025 (Hong Kong Time) for the first-round evaluation. Scanned copies of the -following documents should also be uploaded by the application deadline. +£24 +spent on +Facebook +campaign -(1) Official prediction of grades or scores from your current high school (if available); +Reaching +~3000 target +audience (local +and interested in +tennis) -(2) All internal and public examination results during your senior secondary education; and +145 declared +engagements +on FB event +page -(3) Academic reference letter (optional) +Converted to 69 +bookings – +roughly 80% of +all total +attendees! -Late submission of supporting documents will delay the consideration of application. +Drove an overall increase in club membership by 10% -Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other -programmes will be arranged if necessary. +13 -University and Programme Entrance Requirements -Due to the complex combination of education background of applicants, each application will -be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the -standards and requirements on the website of the Admissions Office of the University before -making an application: + MEASURING PERFORMANCE -(1) Admissions standards (International Qualifications or National Qualifications); +A Q U I C K L O O K A T F A C E B O O K A D M A N A G E R -(2) Programme specific subject requirements; and +#Tip: Don’t be intimidated by complexity. +There are tons of available resources online +that offer step-by-step instructions for setting +up a campaign. -(3) English language requirement +Source: Adspresso by Hootsuiite -In particular, please note that the expected lower boundary for admissions listed on the -website is provided for reference purposes only. Admissions is competitive in nature and -achieving the expected lower boundary does not guarantee admission to the programme. -The actual admission grade or score will be dependent on the overall quality of applications -received in the 2026 admissions exercise. +14 -Announcement of Result -Offer results will be released from December 2025 onwards. All application status will be -updated in the online application system once an admissions decision has been made. -Applicants should check their application status online regularly. + DO’S AND DON'TS -More Details -For further information about the application procedures and admissions -standards, -at -https://admissions.hku.hk/apply/international-qualifications. +F O R S O C I A L M E D I A M A R K E T I N G -the Admissions Office’s website +DO -please +Have a social media presence! Today’s consumers expect to be able to find +brands on social media -visit +Pick the right social platforms that best align with your brand, audience and your +marketing goals – whether it’s Instagram, YouTube, or Twitter -2026 Dual Degree Admissions +Take advantage of free resources and training available to you – including My +Tennis Toolkit, Facebook Blueprint, and Google Marketing Platform -Separate application is required for applying to the dual degree programmes, in addition to the -application under the JUPAS or International/Non-JUPAS Admissions Scheme. +Follow your competitors. See what others are doing well, and find out how you +can differentiate yourself and appeal to those audiences -For details of the dual degrees and the application procedures, please visit -the Admissions Office’s website at -https://admissions.hku.hk/programmes/university-collaborative-programmes. +Get creative! Use social media to find influencers or other unique ways to connect +and engage with your audience -3938 Develop Your Career +DON’T -Freshmen Sharing +…forget to publish content regularly. By maintaining your presence you’ll stay top of +mind with members when they are looking for a coach or venue -Upon graduation, over 82% of our students begin their careers in the commerce and industry sector. Others either -move on to further studies or explore opportunities with the civil service, education institutions and -community/social services. +…be afraid to launch a social media marketing campaign. You can start small and +build and refine as you learn -Career Prospects +15 -Companies/organisations that have offered full-time positions to our graduates include but not limited to the -following: + ADDITIONAL RESOURCES -Accenture Company Limited -Bank of China (Hong Kong) Limited -BDO International Limited -Bloomberg L.P. -Citibank (Hong Kong) Limited -Crédit Agricole Corporate and Investment Bank -Credit Suisse Group AG -Deloitte Advisory (Hong Kong) Limited -Deloitte Touche Tohmatsu Limited -Deutsche Bank -Ernst & Young -Goldman Sachs (Asia) LLC -Hang Seng Bank Limited -HKSAR Government -Hong Kong Monetary Authority -JPMorgan Chase & Co. -KPMG -Morgan Stanley -PricewaterhouseCoopers Limited -Standard Chartered Bank (Hong Kong) Limited -The Hongkong and Shanghai Banking Corporation -Limited (HSBC) -UBS Group AG - -Full employment -for the 19th - -consecutive year - -98.7% - -was the average salary of -our 2024 graduates - -$36,770 - -of our 2024 graduates have -received at least 2 job offers -upon graduation - -57% - -Kantbek, Yerassyl -BBA - -I chose HKU Business School because Hong Kong is -one of the world’s most competitive financial centres, -offering a dynamic environment to grow. HKU -combines -real-world -rigorous academics with -experiences, such as Asia-focused case studies and -competitions like HICS, which I plan to join in Year 2. I -am particularly drawn to programs in marketing and -finance, as well as opportunities for entrepreneurship -through the Techno-Entrepreneurship Core and -iDendron incubator. With a diverse student body and -global exchange options, I look forward to learning -from peers, connecting with mentors, and developing -skills to make a meaningful impact in business. - -The main reason I chose HKU, particularly HKU -Business School, lies in its global connections and -strong partnerships with other institutions and -respected organizations, which I believe will be -advantageous for future employment opportunities. -I expect to embrace challenges over the course of my -four years at the university, as these will be vital for -my intellectual and social maturity. The first time I -visited the campus, I was amazed by the diversity of -the students and -facilities, which -demonstrated that the university is truly a place for -everyone. One of the many great opportunities -offered here is the chance to travel for studies, which -I would rank among my highest interests. - -the open - -Pernice, Audrey Yiwen -BEcon&Fin - -4140 The main reason I chose BBA(IBGM) is its flexibility in allowing -students to choose a second major, as well as the numerous -overseas exchange opportunities. I am also fortunate to have -been admitted as a member of the HKU International Case -Society (HICS), where I get to know people from other programs -who are also interested in the realm of business. HKU Business -School provides me with the perfect platform to strike a balance -between my work and life. - -Ho, Lik Hang Gordon -BBA(IBGM) - -Lo, Hoi Tung Reacker -BSc(MAT) - -I chose HKU Business School for its professional background and -diverse opportunities. The multidisciplinary programme allows me to -pursue Marketing Analytics and Technology, aligning with my varied -interests in business and STEM. As a freshman, I am eager to embrace -exciting experiences such as internships and exchanges. I look -forward to gaining insights from fellow students and professionals -through upcoming career events. With the strong support from HKU, -I am committed to developing my entrepreneurial skills and pursuing -my dreams without regrets! - -Drawn by HKU’s global prestige and the excellence of HKU Business -School, I chose to pursue Quantitative Finance at HKU to deepen my -passion for quantitative trading and data-driven decision-making. -The school’s flexible curriculum empowers me to explore -complementary fields such as computer science, statistics, and risk -management. As a target school for top investment banks, HKU -also offers unparalleled career exposure. HKU provides the ideal -environment to grow academically and professionally, and I am -excited to explore the many opportunities ahead. - -Tsang, Chun Shing Patrick -BSc(QFin) - -Fung, Yat Hang Rex -BFin(AMPB) - -My aspiration for BFin(AMPB) was ignited early in high school by its -unparalleled, practitioner-focused design, a curriculum that truly -bridges finance theory with industry practice. This conviction was -solidified through the heartfelt guidance of our Programme -Director, Professor Anna Wong, who mentored me even before I -officially joined HKU, while I was an apprentice at a US bank. - -Now, experiencing BFin(AMPB) firsthand, the transition from the -professional world to the classroom feels seamless. It is this -powerful integration of theory and real-world practice that makes -AMPB incomparable. HKUBS has instantly instilled in me a sense of -ownership, this is not just a school, but a launchpad for impact, and -I am thrilled to begin the next chapter of my growth here. - -I was drawn to BBA(IBGM) because of HKU’s -world-renowned academic reputation and its vibrant, -international community. I believe a truly “global -business” education extends beyond the classroom, -through experiential learning opportunities, daily -interactions with local and non-local peers, and -participation in hall and student society cultures, I will -gain a more dynamic and holistic multidisciplinary -perspective on the business world. Over the coming -four years, I am eager to harness these opportunities -to become a well-rounded leader and effective -problem-solver for my future career on the global -business stage. - -Yan, Sung Hei Christopher -BBA(IBGM) - -4342 Our Global Presence - -Israel - -Tel Aviv - -Beijing - -Shenzhen - -Hong Kong - -Vietnam - -Ho Chi Minh City - -Useful Websites -The University of Hong Kong (HKU) - -Undergraduate Admissions -https://admissions.hku.hk/ +H E L P F U L L I N K S T O G E T Y O U S T A R T E D -HKU – Entrance Requirements +LTA Marketing Guides and materials: +https://www.lta.org.uk/workforce-venues/tennis-venue-support/membership-growth-and- +retention/membership-growth/ -For JUPAS Candidates -https://admissions.hku.hk/apply/jupas +My Tennis Toolkit (includes creative assets for social media): +https://www.lta.org.uk/globalassets/venue/my-tennis-toolkit-usage-guide.pdf -For non-JUPAS Candidates -https://admissions.hku.hk/apply/international-qualifications +Facebook Blueprint - free online courses for business marketing: +https://www.facebook.com/business/learn -HKU Business School +Instagram Advertising: +https://business.instagram.com/advertising -Overview of Undergraduate Programmes -https://ug.hkubs.hku.hk/programme +Twitter Advertising: +https://business.twitter.com/en/advertising.html +https://blog.hootsuite.com/twitter-ads/ -Enrichment Programmes -https://ug.hkubs.hku.hk/student-enrichment +LinkedIn Advertising: +https://business.linkedin.com/marketing-solutions/ads +https://blog.hootsuite.com/linkedin-ads-guide/ -Other HKU Websites +Measurement / Tracking: +Google Analytics Academy: https://analytics.google.com/analytics/academy/ +Google Ad Manager: https://skillshop.exceedlms.com/student/path/34897-drive-advertising-revenue- +with-google-ad-manager -Mainland Affairs Office (MAO) -https://mainlandaffairs.hku.hk/ +16 -Centre of Development and Resources for -Students (CEDARS) -https://www.cedars.hku.hk/ + ANY QUESTIONS? -44 +THANK YOU -45 +17 diff --git a/q188/random_k4/question.json b/q188/random_k4/question.json index f651ba0cf288248b7e568785f7c05b9ab9ebd051..fc8cca38eb80b361b0ff900f91959703eb250205 100644 --- a/q188/random_k4/question.json +++ b/q188/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "2305.14160v4.pdf", - "web_ab2f104fb9b1c2fe.pdf", - "web_6a55c713826db217.pdf", - "web_0592b193cfe894c0.pdf", - "web_99fc489414634c09.pdf" + "web_99fd395a0c29c14c.pdf", + "web_93a0de1d8da7a0cf.pdf", + "web_f77c9a6a0c40e754.pdf", + "web_2c09f8e84be01107.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q188/random_k4/random_1.md b/q188/random_k4/random_1.md index 510efcdbde69ac509e9a3fa23225deb77871706d..178a074cfe698dd3ddc22a4fffd3c36000aa45a9 100644 --- a/q188/random_k4/random_1.md +++ b/q188/random_k4/random_1.md @@ -1,1380 +1,1420 @@ -ARYAMAN -FINANCIAL SERVICES LIMITED +Encyclopedic VQA: +Visualquestionsaboutdetailedpropertiesoffine-grainedcategories +ThomasMensink†,∗ JasperUijlings†,∗ LluisCastrejon∗ ArushiGoel‡ +mensink@google.com jrru@google.com lluisc@google.com goel.arushi@gmail.com +FelipeCadar‡ HowardZhou∗ FeiSha∗ Andre´ Araujo∗ +cadar@dcc.ufmg.br howardzhou@google.com fsha@google.com andrearaujo@google.com +VittorioFerrari∗ +vittoferrari@google.com +Abstract +We propose Encyclopedic-VQA, a large scale visual +question answering (VQA) dataset featuring visual ques- +tions about detailed properties of fine-grained categories +and instances. It contains 221k unique question+answer +pairs each matched with (up to) 5 images, resulting +in a total of 1M VQA samples. Moreover, our dataset +comes with a controlled knowledge base derived from +Wikipedia, marking the evidence to support each answer. Figure1:Oneofthetwoanswersaboveiswrong,doyouknow +Empirically, we show that our dataset poses a hard whichone? Encyclopedicquestionsaboutdetailedpropertiesof +challenge for large vision+language models as they per- fine-grainedentitiesaredifficult.Notonlyforhumans,butalsofor +form poorly on our dataset: PaLI [14] is state-of-the-art largeVLMs.PaLI[14]failstoanswerbothquestionscorrectly. +on OK-VQA [37], yet it only achieves 13.0% accuracy +on our dataset. Moreover, we experimentally show +Answeringthesequestionscorrectlyrequiresknowledgeof +that progress on answering our encyclopedic questions +detailed properties (i.e. symbol of which city, year of au- +can be achieved by augmenting large models with a +tomation) of fine-grained categories (‘Pinus Pinea’) or in- +mechanism that retrieves relevant information from the +stances (‘Point Reyes Lighthouse’). We hypothesize that +knowledge base. An oracle experiment with perfect re- +this type of encyclopedic knowledge is hard for VLMs to +trievalachieves87.0%accuracyonthesingle-hopportion +properlyencodeinitsmodelparametersbecausesuchlong- +of our dataset, and an automatic retrieval-augmented +tail information occurs rarely in its training data (i.e. the +prototypeyields48.8%.Webelievethatourdatasetenables +web). Additionally,VLMsproducesuchincorrectanswers +future research on retrieval-augmented vision+language +generally with high confidence, while these answers are +models. It is available at https://github.com/ +hardforuserstoverifysincethemodeldoesnotprovideany +google-research/google-research/tree/ +explanations. Theanswerto Fig.1(left) iscorrect; (right) +master/encyclopedic_vqa. +shouldbe1975. PaLI[14]predictsbothincorrectly. +Bothproblemscanbeaddressedbyretrieval-augmented +1.Introduction models, which base their predictions on knowledge re- +trieved from a database. These models recently gained +Recently, large Vision+Language models (VLMs) have popularity in NLP (e.g. [9, 23, 30, 33]), and some early +demonstrated impressive performance on Visual Question multi-modal models also exist [22, 26, 36, 52]. Retrieval- +Answering (VQA) benchmarks [7, 14, 24, 57]. However, augmentedmodelsarewell-suitedforencyclopedicknowl- +Fig.1showstwotypicalexampleswheresuchmodelsfail. edge since retrieving the correct entry from a knowledge +†Equalcontribution.‡WorkdoneduringinternshipatGoogle. base greatly facilitates constructing the right answer. Fur- +∗GoogleResearch thermore, the retrieved piece of knowledge provides attri- +3202 +luJ +42 +]VC.sc[ +2v42290.6032:viXra -T= ARYAM AN FINANCIAL SERVICES LIMITED = -| BOARD OF DIRECTORS | | P.V.R Murthy | Director | -| ------------------- | --- | ------------------------ | -------- | -| | | Virender Ganda - | Director | -| | | Mohan Datari - | Director | -| | | M.V. Ramana - | Director | -| | | Suhas Ganpule - | Director | -| REGISTERED OFFICE | | : 401, Chiranjiv Towers | | -43, Nehru Place, -New Delhi 110019 -| CORPORATE | OFFICE | : 208, Maker Chamber V, | | -| ---------- | ------- | ------------------------ | --- | -2nd -Floor, Nariman Point, -Mumbai-400021. -| AUDITORS | | : M/s. Thakur Vaidynath Aiyar & Co. | | -| --------- | --- | ------------------------------------ | --- | -Chartered Accountants -212, Deendayal Upadhyay Marg. -New Delhi. -| BANKERS | | : State Bank Of India | | -| -------- | --- | ---------------------- | --- | -The Vysya Bank Limited -State Bank of Patiala -Indusind Bank Ltd. -| Share Transfer Agent | | : Adroit Corporate Services Pvt. Ltd. | | -| --------------------- | --- | -------------------------------------- | --- | -19, Jaferbhoy Industrial Estate. lsl Floor. -Makwana Road, Marol Naka, -Andheri (W), Mumbai - 400059. +Question Type +skramdnaL +Templated Automatic Automatic - multi-answer 2-Hop +Q: Who founded this monastery? Q: When was the first permanent Q: What fish can be found in this lake? Q: What amusement park is located in +settlement made at this valley? the city where this square is located? +A: Prince Constantin Brâncoveanu A: 1864 A: trout, lake char A: Tivoli Gardens +C: Horezu monastery C: Clover valley C: Úlfljótsvatn C: Rådhuspladsen, Copenhagen +dlroW +larutaN +Q: How old does this reptile become? Q: How many feet tall does this tree Q: Where is this bird found? Q: How many national park service maintained +grow to? sites are in the state where this plant grows? +A: 40 years A: 7 to 13 A: Colombia, Venezuela, Ecuador A: 24 +C: Gila monster C: Acacia paradoxa C: Boissonneaua C: Chorizanthe rigida +Figure2: ExampleVQAannotationsfordifferentquestiontypes. EachexampleconsistsofanimageI,aquestionQandtheanswer +A.WealsoshowthecategoryCofthesubjectofthequestion.Asattribution,weprovideasectionwithintheWikipediapageofCwhich +supportstheanswer.OurEncyclopedic-VQAdatasethasatotalof1M(I,Q,A)triplets. +OK-VQA A-OKVQA FVQA KVQA S3VQA +Q: What sort of vehicle uses Q: What does the man who Q: What furniture in this image Q: Who is to the left of Hillary Q: What is the process by +this item? sits have trouble doing? can I lie on? Clinton? which this insect develops? +A: firetruck A: walking A: sofa A: Aamir Khan A: complete metamorphosis +Q: When was this piece of Q: What could block the Q: Which transportation way in Q: Who among the people in the Q: Where was this weapon used +sporting equipment invented? washer’s door? this image is cheaper than taxi? image is the eldest? to attack the French camp? +A: 1926 A: stove A: bus A: Person in the left A: Crecy +Figure3: TypicalexamplesofVQAdatasetswhichrequireknowledgebeyondtheimage. +bution to the answer by design. This increases model in- quiringknowledgeofdetailedproperties,butmostlyabout +terpretabilityandthereforehumantrust,hasapplicationsto coarserbasic-levelcategories[45](e.g.Fig.3bottom-left). +fairnessandhelpsdiagnosinganddebuggingmodelerrors. Thereareafewotherdatasetswhichtargetdetailedproper- +To drive progress on handling encyclopedic knowledge ties[27,47]. Butboth[27,47]lackattributionannotations, +and attribution in VQA we need a suitable dataset. The [27]isverysmall(Tab.1right-mostcolumn),and[47]fo- +popular OK-VQA [37] and A-OKVQA [46] datasets fo- cusesonlyoncelebrities(Fig.3fourthcolumn). Henceno +cus on questions requiring knowledge outside the query currentVQAdatasetisfullysatisfactory. +image. However, the majority of questions require com- In this paper we introduce the Encyclopedic-VQA +monsense knowledge (e.g. Fig. 3 top-row, two left-most dataset, which offers several attractive features (Fig. 2, +examples), a type of knowledge for which current VLMs Tab.1). Ourdatasetasksquestionsaboutfine-grainedcat- +are powerful in both theory and practice [7, 14, 24, 57]. egories from iNaturalist 2021 [51] (Fig. 2 bottom row) +These datasets [37, 46] also include some questions re- andinstancesfromtheGoogleLandmarksDatasetv2[56] -r -7= ARYAMANFINANCIALSERVICES LIMITED -NOTICE -"RESOLVED THAT in accordance with the -| | | | | provision of section | | 257 and all other | | -| --- | --- | --- | --- | --------------------- | --- | ------------------ | --- | -NOTICE is hereby given that the 11™ ANNUAL -| | | | | applicable provisions, if any, of | | | the | -| --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | -GENERAL MEETING of the Members of the -ARYAMAN FINANCIAL SERVICES companies Act 1956 or any statutory -modifications or reenactment thereof, Shri -| LIMITED will be held on | | Thursday, 29th | | | | | | -| ------------------------ | --- | -------------- | --- | --- | --- | --- | --- | -Suhas S Ganpule, who was appointed as -| day December 2005 at 10.00 a.m. at its | | | | | | | | -| --------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -Additional Director in pursuant to section -registered office 401, Chiranjiv Towers, 43, -Nehru Place, New Delhi 110 019 to transact 260 of companies Act, 1956 be and hereby -appointed as Director of the company -the following business: -subject to retirement by rotation under the -ORDINARY BUSINESS provisions of Article of Association of the -company". -1. To receive, consider and adopt the Balance -Sheet as at June 30, 2005 and the Profit and 6. To consider and if thought fit to pass with -Loss Account for the year ended on that date or without any modification, the -| | | | | following | resolution | as Ordinary | | -| --- | --- | --- | --- | ---------- | ----------- | ------------ | --- | -and the Reports of Directors and Auditors -Resolution- -thereon. -"RESOLVED THAT in accordance with the -2. To appoint a director in place of Shri Mohan -| | | | | provision of section 257 and | | all other | | -| --- | --- | --- | --- | ----------------------------- | --- | --------- | --- | -Datari, who retires by rotation, and being -| | | | | applicable | provisions, | if any, of | the | -| ------------ | ---------------- | ---- | --- | ----------- | ------------------------- | ----------- | --- | -| eligible, | offers himself | for | re- | | | | | -| appointment. | | | | companies | Act 1956 or any statutory | | | -modifications or reenactment thereof, Shri -Manmohan Baliga, who was appointed as -| 3. To appoint | Auditors and | to fix | their | | | | | -| --------------- | ------------- | -------- | ----- | ------------------------------------------ | --- | --- | --- | -| remuneration. | | | | Additional Director in pursuant to section | | | | -260 of companies Act, 1956 be and hereby -| | | | | appointed | as Director of the company | | | -| --- | --- | --- | --- | ---------- | -------------------------- | --- | --- | -SPECIAL BUSINESS -subject to retirement by rotation under the -4. To consider and if thought fit to pass with provisions of Article of Association of the -company". -or without any modification, the following -resolution as Ordinary Resolution- -7. To consider and if thought fit to pass with -" RESOLVED THAT in accordance with the or without any modification, the -| | | | | following | resolution | as Ordinary | | -| ---------- | ----------- | -------- | ---------- | ---------- | ----------- | ------------ | --- | -| provision | of section | 257 and | all other | | | | | -Resolution- -| applicable | provisions, | if any, | of the | | | | | -| ----------- | ------------ | ----------------- | ------- | --- | --- | --- | --- | -| companies | Act 1956 | or any statutory | | | | | | -modifications or reenactment thereof, Shri "RESOLVED THAT in suppersession of the -earlier resolutions passed under Section -Munnangi Venkat Ramanna, who was -293 (l)(d) of the Companies Act, 1956, the -| appointed | as Additional | Director | in | | | | | -| ---------- | --------------- | --------- | --- | --- | --- | --- | --- | -pursuant to section 260 of the Companies Board of Directors of the Company be and -is hereby authorised to borrow from time -Act, 1956 be and is hereby appointed as -to time all such sums of money as they -| Director | of the | company subject | to | | | | | -| --------- | -------- | ----------------- | --- | --- | --- | --- | --- | -may deem necessary for the purpose of -retirement by rotation under the provisions -of Article of Association of the company". business of the Company -| | | | | notwithstanding | that moneys to be | | | -| --- | --- | --- | --- | ---------------- | ----------------- | --------------- | --- | -| | | | | borrowed | together | with the moneys | | -5. To consider and if thought fit to pass with -or without any modification, the following already borrowed by the Company (apart -resolution as Ordinary Resolution- from temporary loans (including working +OK-VQA[37] A-OKVQA[46] FVQA[53] S3VQA[27] KVQA[47] Encyclopedic-VQA(ours) +Trulymultimodal +| | | | | ++ | | ++ | ++ | ++ | | ++ | | ++ | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| Encyclopedic | | | | ± | | - | - | + | | ++ | | ++ | | +| (cid:44)→Detailedproperties | | | | + | | ± | - | ++ | | ++ | | ++ | | +| (cid:44)→fine-grainedcategories/instances | | | | - | | - | - | + | | ++ | | ++ | | +| Controlledknowledgebase | | | | -- | | -- | + | ± | | ± | | ++ | | +| (cid:44)→answersupportedbyKB | | | | -- | | -- | ++ | ++ | | ++ | | ++ | | +(cid:44)→KBprovided +| | | | | -- | | -- | ++ | -- | | ++ | | ++ | | +| -------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| (cid:44)→KBfree-form | | | | -- | | -- | -- | -- | | -- | | ++ | | +| (cid:44)→attribution | | | | -- | | ± | ++ | -- | | -- | | ++ | | +| Scale | | | | ± | | + | -- | -- | | ++ | | ++ | | +| Two-hop | | | | -- | | -- | -- | -- | | ++ | | + | | +Subject various various various various celebrities fine-grainedspecies,landmarks +| NumberoftextquestionsQ | | | | 14k | | 25k | 6k | 7k | | 183k | | 221k | | +| -------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---- | --- | ------ | --- | +| NumberofimagesI | | | | 14k | | 24k | 2k | 7k | | 25k | | 514k | | +| NumberofuniqueVQAtriplets(I,Q,A) | | | | 14k | | 25k | 6k | 7k | | 183k | | 1,036k | | +Table1: ComparisonofrecentVQAdatasets.WecomparerecentVQAdatasetswithourproposeddatasetonVQAdesignprinciples. +(Fig. 2 top row). We construct questions about detailed that specific part of the knowledge base which supports it. +properties based on Wikipedia (e.g. founder of building, Hencetheknowledgebaseisanintegralpartofourdataset, +maximumageofanimal). Asaconsequence,allquestions whichenablesmeasuringwhetheramodelanswersaques- +in our dataset are about encyclopedic knowledge. Further- tion correctly for the right reason. For generality we want +more, we provide a controlled knowledge base suited to thisknowledgebasetobefree-formtextandtocontainim- +answerthesequestions: 2MWikipediapagesconsistingof ages. (4) Scale. The dataset should be large. Dataset size +free-formtextandimages[48]. Wealsomarkground-truth hasalwaysmatteredandthisisevenmoretruewiththein- +attributionforeachansweratthegranularitylevelofasec- creasinglylargeVLMmodels(e.g.[6,14,28,60,39]). (5) +tion within a Wikipedia page. Importantly, our dataset is Two-hop. Aportionofourquestionsshouldrequireknowl- +collected at scale: we have 221k unique question+answer edge from multiple different documents from the knowl- +pairseachmatchedwitharound5images,resultinginato- edgebase. Includingsuchcomplextwo-hopquestions[58] +talof1Mexamples.Thismakesourdatasetthelargestofits leavessubstantialheadroomforfuturemodeldevelopment. +| kind. Finally, | manyof | ourquestions | | arecomplex | | two-hop | | | | | | | | +| -------------- | ------ | ------------ | --- | ---------- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +3.RelatedWork +questions[58],whichrequiremultipledifferentdocuments +fromtheknowledgebasetosolve(Fig.2right). +VisualQuestionAnswering(VQA).DAQUAR[35],FM- +| We validate | the usefulness | | of our | dataset | through | sev- | | | | | | | | +| ----------- | -------------- | --- | ------ | ------- | ------- | ---- | --- | --- | --- | --- | --- | --- | --- | +IQA[20],VisualMadlibs[59],VQAv1[8]andVQAv2[21] +| eralexperiments. | Inparticular,wedemonstratethatalarge | | | | | | | | | | | | | +| ---------------- | ------------------------------------ | ------ | ---------------- | ----------- | ------ | ---------- | --------- | -------- | ----------- | -------------- | --- | ---------- | ------ | +| | | | | | | | are early | VQA | datasets. | These datasets | | mostly ask | visual | +| VLM (PaLI | [14]) which | yields | state-of-the-art | | | results on | | | | | | | | +| | | | | | | | questions | that can | be answered | based | on | the query | image | +| OK-VQA | [37] performs | poorly | on | our dataset | (13.0% | ac- | | | | | | | | +andgenericknowledgesuchasbasic-levelcategoryrecog- +| curacy). | Next we demonstrate | | through | | an oracle | experi- | | | | | | | | +| -------- | ------------------- | --- | ------- | --- | --------- | ------- | --- | --- | --- | --- | --- | --- | --- | +nition(e.g.‘cat’),countingitems,colors,etc. +mentthatretrieval-augmentedmodelscanyield87.0%ac- +| | | | | | | | Knowledge-based | | VQA. | Tab. | 1 and | Fig. 3 compare | | +| ------- | --------------- | --- | ------ | ----- | ------ | --------- | --------------- | ----- | ------------ | ---- | ------- | -------------- | --- | +| curacy. | Finally, we use | an | online | image | search | engine to | | | | | | | | +| | | | | | | | datasets | which | are designed | to | require | knowledge | not | +buildanautomaticretrieval-augmentedprototype,reaching +| | | | | | | | presentintheimage[27,37,46,47,53]. | | | | | Inparticular,OK- | | +| ---------------- | ----------------------- | -------- | ----- | -------- | ----------- | --------- | ---------------------------------- | ---------- | --------------------------------- | ---- | -------- | ---------------- | --- | +| 48.8% accuracy. | We | conclude | that | (1) our | dataset | poses a | | | | | | | | +| | | | | | | | VQA [37] | and | A-OKVQA | [46] | mostly | require common- | | +| strong challenge | in VQA | and | is in | fact too | hard | for stan- | | | | | | | | +| | | | | | | | senseknowledge. | | Inaddition,somequestions(18%)inA- | | | | | +| dard VLMs; | (2) retrieval-augmented | | | VLMs | demonstrate | a | | | | | | | | +| | | | | | | | OKVQA | do require | knowledge | of | detailed | properties, | but | +strongpotentialforaddressingencyclopedicknowledge;(3) +| | | | | | | | about basic-level | | categories. | Finally, | 3% | of the questions | | +| --- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ----------- | -------- | --- | ---------------- | --- | +ourresultsleavesignificantheadroomforfurtherresearchto +| | | | | | | | requireknowledgeaboutphysics. | | | | Inthispaperwecreatea | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------- | --- | --- | --- | -------------------- | --- | --- | +improveretrieval-augmentedVLMs. +datasetwithquestionsexclusivelyaboutdetailedproperties +| | | | | | | | of fine-grained | | categories | and instances | | (Fig. 2). | We be- | +| --- | --- | --- | --- | --- | --- | --- | --------------- | --- | ---------- | ------------- | --- | --------- | ------ | +2.Designprinciplesforourdataset +| | | | | | | | lieve that | such encyclopedic | | questions | truly | require | access | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ----------------- | --- | --------- | ----- | ------- | ------ | +We create our VQA dataset with the following desired toaknowledgebasetobeanswered. Infact,wereleasethe +properties in mind: (1) Truly Multimodal. The questions knowledgebasealongwithourdataset,whereasnoexplicit +should not be answerable without the image or the textual knowledgebasewasinvolvedinthecreationof[37,46]. +question [21]. (2) Encyclopedic. The questions should SomeexistingVQAdatasetsaresupportedbyaknowl- +be about detailed properties of fine-grained categories or edgebase[27,47,53]. FVQA[53]isaboutcommonsense +instances; a type of questions which are problematic for knowledge. KVQA [47] asks detailed properties about +vanilla VLMs [7, 14, 24, 57]. (3) Controlled knowledge celebrities. Both[47,53]arebasedonastructuredknowl- +base. Each answer in our dataset should be attributable to edge base (RDF triplets). Instead, our knowledge base is -7= ARYAMANFINANCIALSERVICES LIMITED -capital facilities) obtained from the remuneration, costs, charges expenses and -Company's bankers in the ordinary course all other monies payable by the Company -of business) (hereinafter referred to as the to the concerned Lenders, and/or Agents -"Borrowings") shall exceed the aggregate and Trustees for debentures provided that -of the paid-up capital of the Company and the amount of loans (other than temporary -its free reserves, that is to say reserves not loans (including working capital facilities) -set apart for any specific purpose provided obtained from the Company's bankers in -that the total amount of the Borrowings the ordinary course of buinsess) shall not -by the Board of Directors shall not exceed exceed Rs. 50 Crores." -Rs. 25 crores (Rupees Twenty Five Crores -Only)." "RESOLVED FURTHER THAT the Board of -Directors of the Company be and is hereby -8. To consider and if thought fit to pass with authorised to finalise with any of the -or without any modification, the following Lenders, or other persons, jointly or -resolution as Ordinary Resolution- severally, the documents for creating -aforesaid mortgage/charge and to do all -"RESOLVED THAT in supersession of the such acts, deeds, matters and things as may -earlier resolutions passed under section be considered necessary, desirable or -293(l)(a) of the Companies Act, 1956 (the expedient for implementing the Resolution -"Act"), consent of the Company be and is and to resolve any question, difficulty or -hereby accorded in terms of section doubt which may arise in relation thereto -293(l)(a) and other applicable provisions, or otherwise considered by the Board of -if any, of the Act, to mortgaging and/or Directors to be in the best interest of the -charging by the Board of Directors of the Company." -Company of all or any of the immovable -and/or movable properties of the By Order of the Board -Company, wheresoever situated, both -present and future, or the whole or Place: Mumbai -substantially the whole of the undertaking Date: December 01, 2005 -or undertakings of the Company on such For Aryaman Financial Services Limited -terms, in such form and in such manner -as the Board of Directors may think fit, -together with power to take over the Prasanna Chandwaskar -management of the business and concern Company Secretary -of the Company in certain events in favour -of Banks, Financial Institutions, other A MEMBER ENTITLED TO ATTEND AND -investing agencies and trustees for the VOTE IS ENTITLED TO APPOINT ONE -holders of Debentures / other instruments, OR MORE PROXIES TO ATTEND AND -for securing, inter alia, any loans (both VOTE INSTEAD OF HIMSELF AND A -rupee loans as well as foreign currency PROXY NEED NOT BE A MEMBER OF -loans) and/or advances already obtained THE COMPANY. Proxies, in order to be -or debts already incurred or that may effective, should be duly stamped, -hereafter be obtained or incurred from any completed, signed and deposited at the -of the lenders and/or to secure any Registered Office of the Company not less -debentures issued/that may be issued, than 48 hours before the Meeting. -and/or any financial obligations/ -commitment hereinafter collectively The Register of Members and Share Transfer -referred to as the "Loans") and all interest, Books of the Company will remain -compound / additional interest, closed from Monday, December 26, 2005 -commitment charges, Trustees' to Thursday, December 29, 2005 (both +| readilyavailablefree-formWikipediatext,whichhasmore | | | | | | | | 4.Dataset | | | | | | | +| --------------------------------------------------- | -------------------------------------- | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | +| information. | Furthermore,KVQAisexclusivelyaboutpeo- | | | | | | | | | | | | | | +Wenowdetailtheconstructionofourdatasetwhilefol- +| ple, which | requires | specialized | | methods | based | on | face de- | | | | | | | | +| ---------- | ---------------- | ----------- | ------------ | ------- | ----- | ------- | -------- | ----------- | ------------ | ---------- | -------- | --------- | ------------ | ------- | +| | | | | | | | | lowing the | design | principles | | (Sec. 2). | To achieve | scale | +| tection | and recognition. | | In contrast, | | our | dataset | offers a | | | | | | | | +| | | | | | | | | we automate | construction | | whenever | | possible. We | use hu- | +broaderrangeoftopics,includinganimals,plantsandland- +manannotatorstoensurequalityandtoprovideinformation +| marks. | S3VQA | [27] | is the most | related | work | to | ours. It | | | | | | | | +| ------ | ----- | ---- | ----------- | ------- | ---- | --- | -------- | -------------------------------- | --- | --- | --- | --- | --------------- | --- | +| | | | | | | | | whichwecouldnotgetautomatically. | | | | | Wesimplifyhuman | | +buildsoncategoriesofOpenImages[32],somefine-grained +annotationtasksasmuchaspossible,whichincreasesboth +| and others | more | basic-level. | | They automatically | | generate | | | | | | | | | +| ---------- | ---- | ------------ | --- | ------------------ | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +thequalityandefficiencyoftheirwork. +| questions | from | Wikipedia | articles, | | which | by construction | | | | | | | | | +| --------- | ---- | --------- | --------- | --- | ----- | --------------- | --- | --- | --- | --- | --- | --- | --- | --- | +WedefineaunitforourVQAtaskasatriplet(I,Q,A). +| are about | detailed | properties. | | Then | they let | experts | manu- | | | | | | | | +| --------------------------------------- | --------- | ----------- | -------- | ---- | ---------- | ------------ | ----- | -------------------------------------- | --- | --- | --- | ------------------------ | --------- | --- | +| | | | | | | | | Themulti-modalquestionfeaturesanimageI | | | | | andaccom- | | +| allyselectandrephraserelevantquestions. | | | | | | Ourpaperalso | | | | | | | | | +| | | | | | | | | panyingtextualquestionQ. | | | | ThesubjectofthequestionQ | | | +| automatically | generates | | QA pairs | from | Wikipedia. | | How- | | | | | | | | +appearsinI +| | | | | | | | | | andisacoreconceptofourwork. | | | | Wereferto | | +| --------------------------------- | --- | --- | --- | --- | --------------- | --- | --- | --- | --------------------------- | --- | --- | --- | --------- | --- | +| ever,wegobeyond[27]inseveralways: | | | | | (1)ourdatasetis | | | | | | | | | | +thecategoryofthesubjectasC(e.g.‘HorezuMonastery’or +muchlarger(7kvs1MVQAtriplets),(2)weincludemore +| | | | | | | | | ‘GilaMonster’inFig. | | | 2-left). | TheanswerAispurelytex- | | | +| --------- | --------- | ---------- | ------ | ------ | ------- | ---- | -------- | --------------------------------------------------- | --- | ----- | -------- | ---------------------- | ---------- | --------- | +| complex | multi-hop | questions, | | (3) we | release | the | knowl- | | | | | | | | +| | | | | | | | | tual. Ourdatasetalsorecordstheevidenceforeachanswer | | | | | | | +| edge base | along | with | the QA | pairs. | This | also | includes | | | | | | | | +| | | | | | | | | in the knowledge | | base, | as a | section | within the | Wikipedia | +ground-truthanswerattributionintheformofthesupport- +| | | | | | | | | pageforC | wheretheanswerisfound. | | | | Incontrasttomany | | +| --- | --- | --- | --- | --- | --- | --- | --- | -------- | ---------------------- | --- | --- | --- | ---------------- | --- | +ingWikipediasection. +| | | | | | | | | existing | VQA | datasets, | we | record only | a single | answer, | +| --- | -------- | ------- | ---- | ------------- | --- | ------------ | --- | -------- | --- | --------- | --- | ----------- | -------- | ------- | +| The | InfoSeek | dataset | [15] | is concurrent | | to our work. | It | | | | | | | | +sinceitisunambiguousgiventheknowledgebase. +| also targets | detailed | properties | | of fine-grained | | categories, | | | | | | | | | +| ------------ | ----------- | ------------ | --- | --------------- | ------- | ----------- | --- | ------- | --- | ------- | ----- | ------------ | --- | ---------- | +| | | | | | | | | To make | our | dataset | truly | multi-modal, | we | always re- | +| but does | not include | multi-answer | | and | two-hop | questions, | | | | | | | | | +fertothesubjectofthequestionbyitssupercategory(e.g. +andwascollectedusingadifferentannotationprotocol. +| | | | | | | | | ‘this monastery’, | | ‘this | reptile’, | as opposed | to ‘the | Horezu | +| ---------------------------- | --- | --- | --- | ------------------ | --- | --- | --- | ----------------- | ---- | ----- | --------- | ---------- | ------------ | ------ | +| Automaticquestiongeneration. | | | | AfewVQAdatasetsare | | | | | | | | | | | +| | | | | | | | | Monastery’, | ‘the | Gila | monster’, | etc.). | This ensures | visual | +generated from image captions [44, 13]. While this en- recognitionisrequiredtosolveourtask,astherewouldbe +ablesmuchlargerscale,itresultsinsimplevisualquestions. +manypotentialanswerstothetextualquestionalone(e.g.a +| We build | a portion | of | our dataset | using | a | similar | pipeline | | | | | | | | +| -------- | --------- | --- | ----------- | ----- | --- | ------- | -------- | --- | --- | --- | --- | --- | --- | --- | +differentanswerforeachmonasteryintheworld). +| as [13], | but apply | it to | Wikipedia | pages | instead | to | obtain | | | | | | | | +| -------- | --------- | ----- | --------- | ----- | ------- | --- | ------ | --- | --- | --- | --- | --- | --- | --- | +4.1.SupportingDatasets +| high-quality | questions | | on detailed | properties. | | As | another | | | | | | | | +| ------------ | --------- | --- | ----------- | ----------- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +difference,weverifythemwithhumanannotators. We build on two of the largest existing datasets with +| Multi-Hop | Reasoning | | in NLP. | The | NLP | community | has | | | | | | | | +| --------- | --------- | --- | ------- | --- | --- | --------- | --- | ----------------------- | --- | --- | ------------------------------ | --- | --- | --- | +| | | | | | | | | fine-grainedcategories: | | | iNaturalist2021(iNat21)[51]and | | | | +many text-only QA datasets (e.g. [19, 29, 31, 38, 42, 50, Google Landmarks Dataset V2 (GLDv2) [56]. iNat21 is +55, 58]). Notably, HotpotQA [58] introduces the concept afine-grainedvisualclassificationdatasetcontaining2.7M +| of‘bridgeentity’whichlinkstworelatedentities. | | | | | | Theyuse | | | | | | | | | +| --------------------------------------------- | --- | --- | --- | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +imagesdepicting10,000species,groupedinto11supercat- +thistoshowanannotatortworelatedWikipediaparagraphs egories: plants, insects, birds, mammals, etc. GLDv2 is +| from different | pages | and | ask | to create | questions | which | re- | | | | | | | | +| -------------- | ----- | --- | --- | --------- | --------- | ----- | --- | ---------- | ----------- | --- | ------- | ---- | --------- | --------- | +| | | | | | | | | a landmark | recognition | | dataset | with | 4M images | depicting | +quires knowledge from both paragraphs. We use ‘bridge 200k landmarks. Each landmark is sourced from Wiki- +entities’toautomaticallychaintwoquestionstogetherinto media Commons [5], enabling us to mine the existing +acompoundtwo-hopquestion. +category-hierarchy(allowingtodefinesupercategories,like +Retrieval-augmented models. Our dataset seems partic- bridges, castles, lighthouses, lakes etc.), and to link them +ularly suited for retrieval-augmented models. While pio- to Wikipedia articles. We use these provided annotations +neered in NLP [9, 23, 30, 33], several recent works use to speed up the annotation of our dataset, to identify rele- +retrieval for VQA. KRISP [36] leverages triplets encoding vantknowledgecategories,andtoassignimagesItotextual +facts, categorical knowledge and object relationships in a questionsQautomatically. +graph-based reasoning framework. KAT [22] uses Wiki- We create our controlled knowledge base starting from +data triplets, and a reasoning module cross-attends them the WIT dataset [48], which contains 37M image-snippet +withGPT-3answers,theresultbeingfedintoadecoderfor pairs with 11M unique images from Wikipedia. We select +answer generation. InFactuality [52] leverages index im- allimageslinkedtoanEnglishsnippetandthenextendWIT +agestolinkentitiespresentinthequeryimageandWikidata to include the full corresponding Wikipedia article (snap- +triplets to gather facts, which are fed into a UNITER [16] shot of 13 August 2022). This spans 2M English articles. +reasoning module. REVEAL [26]’s external knowledge is We then identify which categories of iNat21 and GLDv2 +composedofimage-textpairs,questionansweringpairsand map one-to-one to a single Wikipedia article. We found +knowledgegraphtriplets,whichareusedtoassistagenera- such unique mappings for 80% of the iNat21 and 50% of +tormoduleinproducinganswers. the GLDv2 categories. We only create questions for those -F ARYAMAN FIN ANCIAL SERVICES LIMITED -Place : Mumbai -| EXPLANATORY | | STATEMENT | | AS | | | -| -------------------- | --- | ---------- | --------- | ------------------------- | ---------- | --- | -| | | | | Date : December 01, 2005 | ForAryaman | | -| REQUIRED BY SECTION | | | 173(2) OF | | | | -THE COMPANIES ACT, 1956. Financial Services Limited -Prasanna Chandwaskar -ITEM NO. 4, 5 & 6 -| Shri Munnangi V Ramana, Shri Suhas S | | | | Company Secretary | | | -| ------------------------------------ | --- | --- | --- | ----------------- | --- | --- | -Ganpule and Shri Baliga were appointed as -DIRECTOR'S REPORT -Additional Directors in accordance with the -provisions of section -| 260 of the Companies | | Act, 1956 at Board | | TO THE MEMBERS | | | -| --------------------- | --- | ------------------ | --- | -------------- | --- | --- | -Meeting held on December 20, 2004, May -Your Directors present the Eleventh -02,2005 and December 1,2005 respectively. -Annual Report and audited accounts of -The Company has received notice from the -your -shareholder of the Company under section -Company for the financial year ended -| 257 of the Companies | | Act, 1957 for | | the | | | -| --------------------- | --- | -------------- | --- | --- | --- | --- | -June 30, 2005. -appointment of Shri Ramana, Shri Ganpule -and Shri Baliga as Director of the Company, -| | | | | FINANCIAL RESULTS | (Rs. in lakhs) | | -| --- | --- | --- | --- | ----------------- | -------------- | --- | -at the ensuing Annual General Meeting. -| | | | | Particulars | 30th June, | 31s1 March, | -| --------------------- | --- | --------------- | --- | ----------- | ----------- | ----------- | -| The Board recommends | | the approval of | | | | | -| | | | | | 2005 | 2004 | -resolution by way of simple majority. -| | | | | Income from operations | 27.67 | 81.44 | -| --- | --- | --- | --- | ----------------------- | ----- | ----- | -None of the directors except Shri Munnangi Less- Expenditure 35.21 86.73 -(5.29) -V Ramana, Shri Suhas Ganpule and Shri Profit before Depreciation (7.54) -(6.49) -Baliga, are in any way concerned or Less -Depreciation (5.03) -| | | | | Profit before tax | (12.57) | (11.78) | -| --- | --- | --- | --- | ------------------ | ------- | ------- | -interested in the said resolution at 4, 5 & 6 -| respectively. | | | | Provision for tax | | | -| ------------- | --- | --- | --- | ----------------- | --- | --- | -(269.64) -| | | | | Balance carried to | (281.42) | | -| --- | --- | --- | --- | ------------------- | -------- | -------- | -| | | | | Balance Sheet | (293.99) | (281.42) | -ITEM NO. 7 & 8 -The Board of Directors in its meeting held -DIVIDEND -| on 30th August, | | 2005 has approved | | the | | | -| ---------------- | --- | ------------------ | --- | --- | --- | --- | -Due to absence of profits, the Director -| expansion | of the Company's | | existing | | | | -| ---------- | ----------------- | --- | -------- | --- | --- | --- | -do not recommend any dividend for the -activities. -year. -| Considering | the expansion | | proposals | | | | -| ------------ | -------------- | --- | --------- | --- | --- | --- | -DEPOSITS -| entailing borrowings | | the limits specified | | | | | -| --------------------- | --- | -------------------- | --- | --- | --- | --- | -pursuant to Sections 293 (1) (a) and (d) of -| | | | | Your company did not accept/ | | hold any | -| --- | --- | --- | --- | ----------------------------- | --- | -------- | -the Companies Act, 1956 vide the existing -deposit from public/ shareholders during -resolutions passed by the members of the -the year under review. -Company would be inadequate and hence -the same needs to be suitably increased. -DIRECTORS -The Board recommends the approval of -Shri Mohan Datari, Director of the company -resolution by way of simple majority. -retire by rotation and offers himself for -reappointment. -None of the directors of the Company is -concerned or interested in the resolution. -The shareholders are requested to appoint to -Shri Munnangi V Ramana, Shri Suhas S -By Order of the Board -Ganpule Shri Mohan Baliga as Directors +Monastery Horezu Monastery Templated question based upon super category +Horezu Monastery +Question: Who founded this monastery? +The Horezu Monastery or Hurezi Monastery was founded in 1690 by Prince +| | | | Constantin Brâncoveanu in the town of Horezu, Wallachia, Romania. It is | | | | | Answer: | | | | | +| --- | --- | --- | ------------------------------------------------------------------------ | --- | --- | --- | --- | ------- | --- | --- | --- | --- | +considered to be a masterpiece of "Brâncovenesc style", known for its +| | | | architectural purity and balance, the richness of its sculpted detail, its | | | | | Evidence: | | Copy selected evidence | | | +| --- | --- | --- | --------------------------------------------------------------------------- | --- | --- | --- | --- | --------- | --- | ---------------------- | --- | --- | +| | | | treatment of religious compositions, its votive portraits, and its painted | | | | | | | Not enough evidence | | | +decorative works. +Plant Acacia paradoxa Acacia paradoxa Automated question from wikipedia section +Are the following question/answer pairs relevant and correct? +Description +The large shrub or tree up to 2 to 4 metres (7 to 13 ft) tall and has a similar Question: How many feet tall does the acacia Yes +paradoxa grow to? +| | | | width, it has ribbed branchlets that are often arched downward. It is dense | | | | | Answer: | 7 to 13 | | | No | +| --- | --- | --- | ---------------------------------------------------------------------------- | --- | --- | --- | --- | -------- | ------- | --- | --- | --- | +with foliage; the leaves are actually enlarged petioles known as phyllodes. +They are crinkly and the new ones are covered in hairs. The erect phyllodes +| | | | | | | | | Question: | What are the new leaves of this plant | | | Yes | +| --- | --- | --- | ------------------------------------------------------------------------- | --- | --- | --- | --- | ---------- | -------------------------------------- | --- | --- | --- | +| | | | are asymettric and have a lanceolate shape and are around 30 millimetres | | | | | | covered in? | | | | +(1.18 in) in length and 7 mm (0.276 in) wide. Answer: hairs No +Figure4: Datacollectionfortemplatedandautomaticallygeneratedsingle-hopquestions. Top: Expertscreatetemplatedquestions +Q based on a super category (e.g. Monastery). Annotators are given the full Wikipedia page for a particular C, and asked to provide +theanswersAandtheevidence. Bottom: QuestionsareautomaticallygeneratedfromaWikipediasectionofC andvalidatedbyhuman +annotators.Inbothprocesses,theannotatorneverseestheimageIcorrespondingtoC. +categorieswithauniquelyidentifiableWikipediaarticle. resolvethespecificcategoryC andamodelsimplyremem- +The combination of iNat21 & GLDv2 with Wikipedia beringQ+Apairscannotsolvethechallenge. +as knowledge base enables creating (I,Q,A) triplets cov- Automaticallygeneratedsingle-hop. Weincreasethedi- +ering the long-tail: questions about detailed properties (on versity of questions in the dataset with automatically gen- +| Wikipedia | text) | of fine-grained | categories | (species | from | | | | | | | | +| --------- | ----- | --------------- | ---------- | -------- | ---- | ---------------- | --- | ----- | ----------- | -------- | --- | -------- | +| | | | | | | erated questions | | (Fig. | 4). Similar | to [27], | we | feed the | +iNat21) and instances (landmarks from GLDv2). Hence, Wikipedia article of a category to a question generation +thissatisfiestheencyclopedicdesignprinciple. model [13]. It processes a section of an article at a time, +| | | | | | | producing | a large | number | of Q+A | pairs | (typically | 100s), | +| ----------------------- | --- | --- | --- | --- | --- | --------------------------------- | ------- | ------ | ------ | ---------------------- | ---------- | ------ | +| 4.2.Single-hopquestions | | | | | | byinvertingstatementsfromthetext. | | | | | | | +| | | | | | | Weincludetwofilteringsteps. | | | | First,werequirethecat- | | | +Theanswertoasingle-hopquestioncanbefoundinthe +| | | | | | | egory name | of | C to be | used in the | question. | This | reduces | +| ------------------------------ | --- | --- | ------------------- | --- | --- | ---------- | --- | ------- | ----------- | --------- | ---- | ------- | +| WikipediaarticleofthecategoryC | | | whichisthesubjectof | | | | | | | | | | +thequestion.Weconstructsuchquestionsintwoalternative thenumberofquestionsby20×,andensuresthattheques- +| | | | | | | tion is about | C | (and not | about another | entity | found | in the | +| --- | --- | --- | --- | --- | --- | ------------- | --- | -------- | ------------- | ------ | ----- | ------ | +ways: templatedandautomaticallygenerated. +| | | | | | | snippet | of text). | Second, | to increase | diversity | we | remove | +| -------------------- | --- | ---------------------------- | --- | --- | --- | -------------- | --------- | --------- | ----------- | ---------- | ------------ | ------ | +| Templatedsingle-hop. | | Weuseofthesupercategoryanno- | | | | | | | | | | | +| | | | | | | near identical | | questions | and limit | the number | of questions | | +tationsavailableinthesupportingdatasets(e.g.reptilesfor +perWikipediasection. +| iNat21, monasteries | | for GLDv2). | For each | super | category | | | | | | | | +| ------------------- | --- | ----------- | -------- | ----- | -------- | --- | --- | --- | --- | --- | --- | --- | +wemanuallydefineseveralquestions,likeWhofoundedthis The human annotation task is now extremely simple: +validatewhethertheQ+Apairisrelevantandcorrect,given +| monastery?. | Wethenaskhumanannotatorstoanswerthese | | | | | | | | | | | | +| ----------- | ------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +questionsforaparticularspecies/landmarkC (e.g.Horezu theWikipediasectionusedtogenerateit. AquestionQis +| | | | | | | relevant | if it is | is well-formed, | makes | sense | semantically, | | +| --- | --- | --- | --- | --- | --- | -------- | -------- | --------------- | ----- | ----- | ------------- | --- | +Monastery),showingthemtherelevantWikipediapage(see +Fig. 4). Hence, theannotatorsdonothavetorecognizeC and is expected to have an unambiguous answer. The an- +in an image, which would require expert knowledge. In- swer A is correct if it is supported by the given Wikipedia +| | | | | | | section. | Finally, | wecandirectlyusetheWikipediasection | | | | | +| --- | --- | --- | --- | --- | --- | -------- | -------- | ----------------------------------- | --- | --- | --- | --- | +stead,theyjusthavetoidentifywhethertheanswerisinthe +Wikipediaarticle. Ifyes,theymarktheevidenceforit(for usedtogeneratetheQApairasevidencefortheanswer(at- +tributionforfree!) +| attribution). | If | no, we discard | that question | (as | it cannot | | | | | | | | +| ------------- | --- | -------------- | ------------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +be answered with our knowledge base and it would break About 80% of the questions sent to annotators are val- +thecontrolledknowledgebaseprinciple). Westartwith10 idated positively. The questions contain (by design) the +questions per category, resulting in 4.3 questions on aver- name of the category C, and we rephrase them using a +ageperspecieswithavalidanswerforiNat21,and3.4per carefully-promptedFLAN[18]versionofPaLM[17]tore- +| | | | | | | place the | mention | of | C by its super | category, | which | pro- | +| --- | --- | --- | --- | --- | --- | --------- | ------- | --- | -------------- | --------- | ----- | ---- | +landmarkforGLDv2(seeexamplesinFig.2firstcolumn). +E´glise +NowwehaveatextualquestionQandananswerA. To duces Q. For example, the question Where is the +form a complete multi-modal question, we pair Q with an Saint-Cannat located? is rephrased by PaLM to Where is +| | | | | | | thischurchlocated? | | AnimageI | | fromC | | | +| --- | --- | --- | --- | --- | --- | ------------------ | --- | -------- | --- | ----- | --- | --- | +imageI ofC fromthesupportingdatasets. NotethatanyQ fromasupporting +occursmultipletimesinthedatasetwithdifferentanswers. datasetisthenassociatedtothisquestion. +HencebydesigntheaccompanyingimageI isnecessaryto Multi-answer questions. For many properties of a cat- -ARYAMAN FINANCIAL SERVICES LIMITED -of the company, they were appointed as The Board wish to place record their -additional Directors under section 260 of the gratitude for their continued co-operation, -Companies Act, 1956 on your Board and the assistance and guidance extended by the -Company has received notice under section banks, clients, stock exchange, members -257 of the Companies Act, 1956 for their and associates. -candidature for appointment of Directors -of the Company. CONSERVATION OF ENERGY -TECHNICAL ABSORBATION, FOREIGN -AUDITORS EXCHANGE EARNING AND OUTGO -M/s Thakur Vidyanath Aiyar & Co, The company is not engaged in any -Chartered Accountants retire at the manufacturing activity and therefore -forthcoming Annual General Meeting and provisions of section 217 (1) (e) of the -eligible for the reappointment. companies Act, 1956 is not applicable. -Comments made by the Auditors in their -report are self-explanatory and therefore CORPORATE GOVERNANCE: -do not call for any further explanations. -During the year under report, your -DIRECTORS' RESPONSIBILITY STATEMENT Company has evolved Corporate -Governance practices. The required -As required under section 217 (2 A A) of the structures for governance have been in -companies, Act 1956 your Directors state:- place and the meetings are periodically -i. While preparing annual accounts the held in compliance with the ideal practices -applicable accounting standards have been prescribed under the code. -followed. -ii. Company has selected such accounting Management Discussion and Analysis -policies and applied them consistently and Report is separately given in the Annual -made judgment that are responsible and Report. A separate report on Corporate -prudent which give true and fair view of Governance is annexed hereto as a part of -affairs of the company. this report. A certificate from the Auditors -iii. Company has taken proper and sufficient of the Company regarding compliance of -care for the maintenance of adequate conditions of Corporate Governance as -accounting records in accordance with the prescribed under Clause 49 of the Listing -provisions of the Act for safeguarding the Agreement is attached to this report. -assets of the company and the preventing -and detecting fraud and other irregularities. ACKNOWLEDGEMENT -iv. Company has prepared accounts on -Your Directors would like to express deep -ongoing concern basis. -sense of appreciation for the assistance arid -co-operation received from the Financial -PERSONNEL -Institutions, Banks, Government -Relations between the management and Authorities and Shareholders and for the -employees were company cordial. Your devoted services by the Executives, Staff -Directors wish to place on record and Workers of the Company. -appreciation of service rendered by all the -staff members. There is no employee who For and on the behalf of the Board -draws salary in excess of the limits as -Director -prescribed under section 217 (2A) of the -companies (Particular of Employees Rules) Place: Mumbai -1975 as amended. Date: December 01, 2005 +Q: What is the main competitor for food for this animal? Q: What is the population size of this animal? Chained two hop question with bridge: spotted hyena +What is the population +size of the main +competitor for food of +this animal? +Q: What is the main competitor for food for this animal? Q: What is the population size of this animal? Chained question with bridge: spotted hyena +| | Lion | | | Spotted hyena | | | | +| --- | ---- | --- | --- | ------------- | --- | --- | --- | +Predatory competition +| | | | | The spotted hyena (Crocuta crocuta), also | | What is the population | | +| --- | --- | --- | --- | ------------------------------------------ | --- | ----------------------- | --- | +Lions and spotted hyenas occupy a similar known as the laughing hyena ... It is listed size of the main +as being of least concern by the IUCN on +ecological niche and compete for prey and a c c o u n t o f it s w id e s p r e a d ra n g e a n d la r g e competitor for food of +c a r ri o n ; a re v ie w o f d a t a a c ro ss s e v e ral n u m b e r s e s t im a te d b e t w e e n 2 7 , 0 0 0 a n d this animal? +| | s tu d i e s i n d ic | a te s a d i e t a r y o ve rl a p o f | | | | | | +| --- | -------------------------------------- | ---------------------------------------------- | --- | --------------------------------------- | --- | --- | --- | +| | 58.6%. Lions typically ignore hyenas… | | | 47,000 individuals. The species is, ... | | | | +Figure 5: Illustrationoftwo-hopquestiongenerationusingbridgeentities. Thebridgeentity,e.g.spottedhyena,isananswertoa +single-hopquestionwhichhasitsownentryintheknowledgebase. Hencewecanaskasecondsingle-hopquestionaboutit. Thesetwo +questionsarethenchainedintoatwo-hopquestion. +egory C the response should be a list with multiple an- pair is combined with (up to) 5 images I showing differ- +swers. For example What fish can be found in this lake? ent instances of the same category C from the supporting +orWhereisthisbirdfound?,couldbeansweredwithalist datasets. Thisincreasesvisualdiversityandintotalweuse +offish(orcountries,respectively;Fig.2,thirdcolumn).For 514kuniqueimages. Thereareintotal15ktextualsingle- +these kinds of questions we define the multi-answer ques- hoptemplatedquestions,and158kautomaticallygenerated +tion type. We filter the automatically generated QA pairs questions. Moreover,thedatasetcontains25kmulti-answer +for questions likely to have multiple answers, and extract questionsand22ktwo-hopquestions. +| aninitiallist. | Thenweasktheannotatorstocompleteand | | | | | | | +| -------------- | ----------------------------------- | --- | --- | --- | --- | --- | --- | +Lookingindetailatthe221kQ+Apairs,175kofthem +verifythelistwithallthepossibleanswers. +haveauniqueQ(i.e.nostringmatchwithanyotherques- +4.3.Two-hopquestions tion). Moreover, there are 95k unique answers, of which +| | | | | 73koccuronlyonce. | Theremaining22kanswersfollowa | | | +| --- | --- | --- | --- | ----------------- | ----------------------------- | --- | --- | +Two-hopquestionsaredefinedasrequiringtworeason- Zipf’slaw:11kanswersoccurtwice,2kanswersoccur10+ +| ing steps | to obtain the | final answer. | Here we focus | on | | | | +| --------- | ------------- | ------------- | ------------- | --- | --- | --- | --- | +times,andlessthan400answersoccurmorethan50times. +chained two-hopquestions,whichrequiretwoconsecutive +retrievalsteps. Toconstructsuchquestionsatscale,weuse Train / Val / Test splits. To allow for evaluating differ- +thenotionofabridgeentity[58]: iftheanswertoasingle- entpropertiesofthedataset, wecarefullydesignitstrain/- +hop question is an entity with its own Wikipedia page, it val/test splits. First, there is no overlap in the images I +servesasabridgeentity,aboutwhichwecanaskasecond usedinthetrain,val,ortestsplits. ForGLDv2,wesample +single-hopquestion,seeFig.5. images for all our splits from their train-clean split, which +We automatically identify bridge entities from the avoids most noisy labels in their dataset. For iNat21, we +single-hopanswers(Sec.4.2),thenmanuallydiscardwords sampleourtrainimagesfromtheirtrainset,andourvaland +like yes, blue, and heavy which would yield very artificial test images from their validation set (their test set annota- +multi-hop questions. We then automatically generate and tionsarenotpubliclyavailable). Next, abouttwo-thirdsof +validate questions for these bridge entities using our auto- thequestionsQandhalfoftheanswersAinourvalandtest +maticallygeneratedsingle-hoppipeline. splits do not occur in our train split. Finally, roughly 17% +Wecreatethefinaltwo-hopquestionautomaticallyusing ofthesubjectcategoriesC inourvalandtestsplitsarenot +PaLM, by feeding it the two single-hop questions with an used in our train split. This allows for analyzing different +appropriateprompt.Unfortunately,thesetwo-hopquestions desirablepropertiesofVQAmodelsonourdataset. +areoftenincorrect.Therefore,wecreateasecondpromptto +validatethetwo-hopquestion:PaLMisaskedtoanswerthe +two-hopquestionusingthetwoinitialsingle-hopquestions +| with answers | as context. | If the predicted | answer is | iden- | | | | +| ------------ | ----------- | ---------------- | --------- | ----- | ------------------ | -------------------- | --- | +| | | | | | Numberof(Q,A)pairs | Total(I,Q,A)triplets | | +tical to the answer of the second single-hop question, then Train Val Test Train Val Test +the two-hop question is validated and kept in our dataset; Templated 13,928 400 1,000 66,535 1,827 1,000 +| | | | | Automatic | 153,441 1,750 | 2,750 737,114 | 8,025 2,750 | +| --------------------------------- | ---------------- | ------------------- | ------------ | ----------- | ------------- | ------------- | ----------- | +| otherwise, | it is discarded. | This increases | the accuracy | of | | | | +| | | | | MultiAnswer | 23,929 400 | 1,000 112,736 | 1,844 1,000 | +| thetwo-hopquestionssubstantially. | | Examplesofvalidated | | | | | | +| | | | | TwoHop | 21,040 400 | 1,000 99,866 | 1,895 1,000 | +two-hopquestionsaregiveninFig.2(right). +| | | | | Total | 212,338 2,950 | 5,750 1,016,251 | 13,591 5,750 | +| --- | --- | --- | --- | ----- | ------------- | --------------- | ------------ | +4.4.Datasetstatistics +Table2: +| | | | | DatasetStatistics. | | Wereportthenumberof(question, | | +| --- | --- | --- | --- | ------------------ | --- | ----------------------------- | --- | +Our dataset contains in total 1M (I,Q,A) triplets answer)pairs,and(image,question,answer)tripletsfordifferent +(Tab. 2). These are derived from a total of 221k textual question types. In total our dataset contains 1M VQA triplets, +makingitthelargestofitskind. +| Q+Apairsfrom16.7kdifferentcategoriesC. | | | EachQ+A | | | | | +| -------------------------------------- | --- | --- | ------- | --- | --- | --- | --- | -P= ARYAMANFINANCIALSERVICES LIMITED -dependant upon conditions of primary -| MANAGEMENT DISCUSSION | | | AND | | | | | | -| ---------------------- | --- | --- | --- | -------------- | --- | -------- | ----------- | --- | -| | | | | and secondary | | capital | market. Any | | -ANALYSIS REPORT. -adverse effect on the capital market could -| | | | | affect the performance and | | | | profitability | -| ------------- | ---------- | --- | --- | --------------------------- | --- | --- | --- | ------------- | -| 1. INDUSTRY | STRUCTURE | | AND | | | | | | -of your Company. -DEVELOPMENTS -Merchant Banking and Financial Services -6. INTERNAL CONTROL SYSTEMS -market is fast growing over the last few -The company has an adequate system of -years. Merchant Banker services includes -internal controls that ensure that all assets -providing merchant banking services like -Public Issue Management i.e, acting as are protected against loss from -unauthorized use or disposition and all -| manager / | lead manager to the public | | | | | | | | -| ---------- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | -transaction are recorded and reported in -| issues, right issues, open | | offer | under | | | | | | -| --------------------------- | ----------- | ------ | ------- | ----------- | ----- | ------------------ | --- | --- | -| | | | | conformity | with | generally accepted | | | -| Takeover | Code, Buy | Back | Scheme, | | | | | | -accounting principles. -Portfolio Management Services, private -placement of debt and equity instruments. -7. FINANCIAL PERFORMANCE -The present position of stock market in -India has opened new era for finance The company has achieved a turnover of -| | | | | Rs 27.67 during | | the year as | | against | -| --- | --- | --- | --- | ---------------- | --- | ------------ | --- | ------- | -industry. -| | | | | Rs.81.44 | lacs | in the | corresponding | | -| --- | --- | --- | --- | --------- | ----- | -------- | ------------- | --- | -previous year. During the year company -2. OPPORTUNITY AND THREATS -| Opportunities: | | | | incurred a profit/ | | | loss of Rs. 12.57 as | | -| -------------- | --- | --- | --- | ------------------- | --- | --- | -------------------- | --- | -againstRs.il.78 lacs in the corresponding -Aryaman Financial Services Limited is -previous year. -SEBI registered Category - I. Considering -present capital marVet conditions, your -company has tremendous opportunities in 8. HUMAN RESOURSE AND -INDUSTRIAL RELATIONS -its business activities especially merchant -The industrial relation remain co ordinal -banking activities like issue management, -during the year under review. -portfolio management services, corporate -debt restructuring etc. -9. CAUTIONARY STATEMENT -Statement in this report on management -Threats: -| | | | | Discussion | and analysis describing the | | | | -| --- | --- | --- | --- | ----------- | --------------------------- | --- | --- | --- | -With the number of players coming down, -| | | | | company's | objectives, | | projections, | | -| --- | --- | --- | --- | ---------- | ------------ | --- | ------------ | --- | -we expect marginal competition. -estimates, expectations or prediction are -| | | | | based | on certain | | assumption | and | -| --- | --- | --- | --- | ------ | ------------ | --- | ----------- | --- | -3. SEGMENTWISE PERFORMANCE -Your Company is dealing in one segment expectation of future events. Actual result -| | | | | could | differ | materially from | | those | -| --- | --- | --- | --- | ------ | ------- | ---------------- | --- | ----- | -mainly in Merchant Banking activity. -| | | | | expressed | or implied. The company | | | | -| --- | --- | --- | --- | ---------- | --------------------------- | --- | --- | --- | -| | | | | assume | no responsibility to amend, | | | | -4. OUTLOOK -The growing economy and vibrant capital modify or revise any of the statements on -the basis of subsequent developments, -market conditions promise good scope for -information or events. -financial service sector. Your company is -well prepared to seize the opportunities as -and when they come. The management is For on behalf of the Board -| generally | optimistic | about | future | | | | | | -| ---------- | ----------- | ------ | ------ | --- | --- | --- | --- | --- | -Director -operations. -| 5. RISKS AND CINCERNS | | | Place: Mumbai | | | | | | -| ---------------------- | --- | --- | ------------- | --- | --- | --- | --- | --- | -Date: 01.12.2005 -| Company's | activities | are | mainly | | | | | | -| ---------- | ----------- | ---- | ------ | --- | --- | --- | --- | --- | +Dataset varietyofQ disparityofQ Model Retrieval OK-VQAAccuracy +| | | #uniquebigrams | | | avg.cosinedist. | | | | KAT[22] | | ✓ | | 53.1 | | | +| ----------- | --- | -------------- | --- | ----- | --------------- | ----- | ----- | ------------- | -------- | ---- | ------------------- | --- | ---- | ------- | ----- | +| | | | | | | | | REVEAL[26] | | | ✓ | | 59.1 | | | +| FVQA[53] | | | | 7.9k | | 0.620 | | | | | | | | | | +| | | | | | | | | PromptCap[25] | | | - | | 60.4 | | | +| KVQA[47] | | | | 12.8k | | 0.504 | | | | | | | | | | +| | | | | | | | | | PaLI[14] | | - | | 64.5 | | | +| OK-VQA[37] | | | | 19.3k | | 0.843 | | | | | | | | | | +| S3VQA[27] | | | | 19.7k | | 0.805 | | | | | | | | | | +| | | | | | | | Table | 4: | OK-VQA: | PaLI | is state-of-the-art | | on | OK-VQA. | It is | +| A-OKVQA[46] | | | | 33.9k | | 0.856 | | | | | | | | | | +thereforeangoodcandidatetoapplytoourEncyclopedic-VQA. +| Encyclopedic-VQA | | | 257.9k | | | 0.833 | | | | | | | | | | +| -------------------------------- | ------------ | ------------ | ------------------------- | -------- | -------- | -------- | ------------- | ----------- | --- | ---------- | ---- | ---- | --- | ----- | ------- | +| Table | 3: Diversity | of questions | Q | measured | in terms | of vari- | 5.Experiments | | | | | | | | | +| etyanddisparity(higherisbetter). | | | Disparitynumbersaremostly | | | | | | | | | | | | | +| | | | | | | | | We evaluate | | PaLI [14], | PaLM | [17] | and | GPT-3 | [10] on | +takenfrom[46],butwereproducedtheirA-OKVQAresulttoval- +idate our re-implementation. Our dataset has the largest variety our dataset. We apply all models directly and in retrieval- +andisclosetothebestonesondisparity. augmented settings. For evaluation we use the test split +| | | | | | | | of | our dataset | | on the | single-hop | templated | | and automat- | | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----------- | ------------ | --------- | ---------- | ----------- | ----------- | ------------ | --- | +| | | | | | | | ically | generated | | questions | (except | | for Sec. | 5.6 where | we | +| | | | | | | | also | evaluate | multi-answer | | | and two-hop | questions). | | We | +4.5.QuestionDiversity +measureaccuracyasthepercentageofquestionswherethe +| | | | | | | | predicted | | model | answer | matches | the | ground-truth | | answer. | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ----- | ------ | ------- | --- | ------------ | --- | ------- | +WewanttocomparethediversityofourVQAquestions +to other datasets. However, diversity is a broad concept All ground-truth answers and model predictions are pre- +| | | | | | | | processed | | following | standard | | VQA | practices | [21] | to facil- | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --------- | -------- | --- | --- | --------- | ---- | --------- | +whichmanifeststhroughacombinationofthreebasicprop- +| | | | | | | | itate | checking | their | correctness | | (remove | articles, | punctua- | | +| ------- | -------- | ---------- | ----------- | ----- | ---- | -------- | ----- | -------- | ----- | ----------- | --- | ------- | --------- | -------- | --- | +| erties: | variety, | disparity, | and balance | [49]. | Here | we focus | | | | | | | | | | +onvarietyanddisparity. tion, etc.). Furthermore, weusethe’BERTMatching’cri- +| | | | | | | | terion | BEM | [12] | to determine | | whether | a predicted | | answer | +| --- | --- | --- | --- | --- | --- | --- | ------ | --- | ---- | ------------ | --- | ------- | ----------- | --- | ------ | +Variety. This refers to the semantic variety spanned by is correct given the question and the ground-truth answer. +| the VQA | questions. | One | simple | measure | of variety | is the | | | | | | | | | | +| ------- | ---------- | --- | ------ | ------- | ---------- | ------ | --- | ---------- | --- | ------ | --- | ---------------- | --- | ------ | ------ | +| | | | | | | | BEM | evaluation | | allows | for | more flexibility | | in the | answer | +numberofuniquetextualquestionsQ(175kinourdataset). +| | | | | | | | formulation | | than | classical | exact | matching, | | coming | much | +| ------------------ | ---- | ------------------------------- | --------- | ---- | --------- | ------- | ----------- | --- | ----- | --------- | ----- | --------------- | --- | ------ | ----- | +| However, | this | ignores the | fact that | some | questions | are se- | | | | | | | | | | +| | | | | | | | closer | to | human | judgement | | of correctness, | | as is | shown | +| manticallysimilar. | | Anothermeasureisthenumberoftop- | | | | | | | | | | | | | | +in[12]andinouruserstudyinAppendixB.Weconsideran +icscoveredbythequestions,forwhichwecanusethenum- answercorrectwhenitsBEMscoreis≥0.5. Ifanexample +| ber of | categories | C (16.7k | in our | dataset). | However, | each | | | | | | | | | | +| ------ | ---------- | -------- | ------ | --------- | -------- | ---- | --- | -------- | ------------ | --- | -------- | --- | ---------- | --- | --- | +| | | | | | | | has | multiple | ground-truth | | answers, | | we compute | the | BEM | +datasettypicallyusesadifferentsetofcategoriesatdiffer- +scoreforeachandchoosethemaximum. +entlevelsofgranularity,makingtheircategorycountshard +to compare directly. So instead, we consider the number 5.1.LargeModelswithoutRetrieval +| of unique | bigrams | across | all questions | Q, | as a | reasonable | | | | | | | | | | +| --------- | ------- | ------ | ------------- | --- | ---- | ---------- | ----- | --- | ------- | --- | ----------- | --- | ---- | --------- | --- | +| | | | | | | | PaLI. | To | measure | the | performance | | of a | large VLM | on | +approximationofthesemanticspacespannedbyallQ. +Encyclopedic-VQA +| | | | | | | | | | | we | use | PaLI [14]. | It yields | the | state- | +| ---------- | --- | ------------- | ------------- | --- | ------------- | --- | ---------- | --- | -------- | --- | ------ | ---------- | --------- | --- | ------ | +| Disparity. | | This measures | how different | | the questions | are | | | | | | | | | | +| | | | | | | | of-the-art | | accuracy | on | OK-VQA | [37] | (64.5%), | and | out- | +from each other. We follow the measure introduced in A- performs retrieval-augmented models such as KAT [22] +| OKVQA | [46]: | (1) we | first project | all | questions | Q into | | | | | | | | | | +| -------- | ----- | -------------- | ------------- | ---------- | --------- | ------ | ------- | --- | ---------- | --- | ---- | ------- | --- | ------- | ------ | +| | | | | | | | (53.1%) | | and REVEAL | | [26] | (59.1%) | by | a large | margin | +| a common | | semantic space | using | a publicly | available | [3] | | | | | | | | | | +(Tab.4). Thisdemonstratesthatthetypeofknowledgere- +Sentence-BERT model [43]; then (2) we measure the av- quired to solve OK-VQA can be captured by large VLMs. +eragecosinedistancebetweenallquestionpairs(definedas +| | | | | | | | PaLI | is therefore | | a good | candidate | | to verify | whether | large | +| --------- | ---------- | --- | ----------- | ---- | ----- | --------- | ---- | ------------ | --- | ------ | --------- | --- | --------- | ------- | ----- | +| 1− cosine | similarity | by | [46]). Note | that | while | [46] pro- | | | | | | | | | | +VLMscapturealsoencyclopedicknowledge.WeuseaPaLI +posedthisasagenericmeasureofdiversity,itactuallyonly +modelwith17Bparameterspre-trainedonahugeamountof +measuresdisparityasdefinedin[49]. +dataincludingWikipedia(soithasseentheknowledgebase +| Results. | | | | | | | containingtheanswersforourdataset). | | | | | | OurPaLImodelis | | | +| -------- | --- | -------------- | --- | --------- | --- | ---------- | ----------------------------------- | --- | --- | --- | --- | --- | -------------- | --- | --- | +| | We | report variety | and | disparity | for | knowledge- | | | | | | | | | | +basedVQAdatasetsinTab.3. Ourdatasetoffersthelargest additionallyfine-tunedonOK-VQAandthereforeparticu- +questionvarietybyanorderofmagnitude.Importantly,this larlysuitedforVQAtasks. WefeedthismodelQ,I inputs +is not only due to dataset size: while Encyclopedic-VQA foreachtestsampletoproducemodelanswers. +andKVQA[47]arethelargestdatasetsandhaveacompa- Tab.5(firstrow)showsthatPaLIhasanaccuracyofonly +rable number of questions, KVQA has much less variety. 13.0%. This low performance indicates that PaLI either +In terms of disparity, our dataset is close to the best ones fails to recognise fine-grained categories and instances, or +OK-VQA[37]andA-OKVQA[46],andbetterthanFVQA, failstoprovidedetailedproperties,orboth. Henceencyclo- +| KVQAandS3VQA. | | | | | | | pedicknowledgeistrulydifficultforthismodel. | | | | | | | | | +| ------------- | --- | --- | --- | --- | --- | --- | ------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | -r ARYAMAN FINANCIAL SERVICES LIMITED -c) The Balance Sheet, the Profit and loss account -AUDITORS' REPORT -and Cash Flow statement dealt with by this -report are in agreement with the books of -ARYAMAN FINANCIAL SERVICES LTD. -account. -| | | | d) | In our opinion, the Balance Sheet, the | | Profit | -| --- | --- | --- | --- | --------------------------------------- | --- | ------ | -We have audited the attached Balance Sheet of -and Loss Account and the Cash Flow Statement -ARYAMAN FINANCIAL SERVICES LTD as -at 30th June, 2005 and the Profit and Loss dealt with by this report comply with the -accounting standards referred to in subsection -Account and also the Cash Flow Statement for -| | | | | (3C) of section | 211 of the Companies Act, | | -| --- | --- | --- | --- | ---------------- | ------------------------- | --- | -the year ended on that date annexed thereto. -1956; -These financial statements are the responsibility -| of the Company's | | management. | Our | | | | -| ------------------- | --- | ------------ | --- | ------------------------------------------- | --- | --- | -| | | | e) | On the basis of the written representations | | | -responsibility is to express an opinion on these -received from the directors, as on 30th June. -financial statements based on our audit. -2005, and taken on record by the Board of -We have conducted our audit in accordance with Directors, we report that none of the directors -is disqualified as on 30th June, 2005 from being -auditing standards generally accepted in India. -appointed as a director in terms of clause (g) -| These Standards | require that we plan and | | | | | | -| ---------------- | ------------------------ | --- | --- | --- | --- | --- | -perform the audit to obtain reasonable assurance of sub-section (1) of section 274 of the -Companies Act, 1956; -about whether the financial statements are free -of material misstatement. An audit includes -| | | | f) | In our opinion and to the best of our information | | | -| --- | --- | --- | --- | ------------------------------------------------- | --- | --- | -examining, on a test basis, evidence supporting -the amounts and disclosures in the financial and according to the explanations given to us, -the said accounts give the information required -statements. An audit also includes assessing the -by the Companies Act, 1956, in the manner so -| accounting principles used and | | significant | | | | | -| ------------------------------- | --- | ----------- | --- | ------------- | ---------------- | ------------- | -| | | | | required and | give a true and | fair view in | -| estimates made by management, | | as well as | | | | | -evaluating the overall financial statement conformity with the accounting principles -presentation. We believe that our audit provides generally accepted in India : -a reasonable basis for our opinion. -| | | | a. | In the case of the Balance Sheet, of the state of | | | -| --- | --- | --- | --- | ------------------------------------------------- | --- | --- | -30'h June. 2005 ; -As required by the Companies (Auditors' affairs of the Company as at -and -| Report) Order, | 2003 issued by the Central | | | | | | -| --------------- | --------------------------- | --- | --- | --- | --- | --- | -Government of India in terms of subsection -| | | | | b. In the case of the Profit and loss account, of | | | -| --- | --- | --- | --- | -------------------------------------------------- | --- | --- | -(4A) of Section 227 of the Companies Act, -1956, we enclose in the Annexure a statement the loss for the year ended on that date. -on the matter specified in paragraph 4 and 5 of -| | | | c. | In the case of the Cash Flow Statement, of the | | | -| --- | --- | --- | --- | ---------------------------------------------- | --- | --- | -the said order. -| | | | | cash flow for the year ended on | | that date. | -| ---------------------------------------- | --- | --- | --- | -------------------------------- | --- | ---------- | -| Further to our comments in the Annexure | | | | | | | -referred to above, we report that: -For THAKUR VAIDYANATH AIYAR & CO. -Chartered Accountants. -| a) We have obtained all the | | information | and | | | | -| ----------------------------- | --------- | ------------ | ------- | --- | --- | --- | -| explanations, | which to | the best | of our | | | | -Place:- Mumbai -| knowledge and belief were necessary | | | for the | | | | -| ------------------------------------ | --- | --- | ------- | --- | --- | --- | -purpose of our audit. -| | | | ( | C.V. Parameswar) | | | -| --- | --- | --- | --- | ---------------- | --- | --- | -b) In our opinion, proper books of account as -Partner -required by law have been kept by the Company -Dated :- 01-12-2005 -so far as appears from our examination of those -M. No. 11541 -books; +Model +| | System | | Retrieval | Extrainput | | PaLI[14] | PaLM[17] | GPT-3[10] | | +| --- | ------------ | --- | --------- | ---------- | --- | -------- | -------- | --------- | --- | +| | Vanillamodel | | - | | - | 13.0% | 19.7% | 15.5% | | +| | | | SubjectC | | - | 16.7% | 31.0% | 26.9% | | +| | Oracle | | KBArticle | | - | 29.7% | 78.4% | 77.4% | | +| | | | KBSection | | - | 48.8% | 87.0% | 82.1% | | +| | | | KBArticle | | - | 21.4% | 48.0% | 44.9% | | +Lens +| | | | KBSection | | - | 28.1% | 48.8% | 44.6% | | +| --- | --- | --- | --------- | --- | --- | ----- | ----- | ----- | --- | +Captions +| | PromptCap | | - | | | 17.8% | 29.7% | 25.6% | | +| --- | --------- | --- | --- | --- | --- | ----- | ----- | ----- | --- | +NNtrainsamples +| Table 5: Accuracy | on single-hop | questions. | | | | | | | | +| ----------------- | ------------- | ---------- | --- | ------------ | -------- | --- | -------------- | --------------------------- | --------- | +| | | | We | report model | accuracy | for | our single-hop | templated and automatically | generated | +questions. WhilelargemodelsstrugglewithourEncyclopedic-VQAdataset,ourexperimentsshowthepromiseofaugmentingthemwith +aretrievalmechanism. +PaLM and GPT-3. Models with a larger language un- trieval component shows great potential for predicting de- +derstanding component [11, 10, 17, 40] might be bet- tailedproperties,and(3)sincePaLMandGPT-3workmuch +ter at memorizing detailed properties and extracting infor- better than PaLI, having a strong language understanding +mation from textual knowledge base entries. Therefore, modelisimportantforextractingtherightinformationfrom +we experiment with PaLM 2 [17] (text-bison@001 thefree-formtextKBarticle. +model, available through the PaLM API) and GPT-3 [10] OracleretrievalofKBsection. Finally,tounderstandthe +(text-davinci-003, accessible through the OpenAI importanceofretrievinginformationthatismoreprecisely +API).Botharetrainedonmassiveamountsoftextincluding localizedthananentireWikipediapage, weprovidetothe +Wikipedia. GPT-3 is especially large, with 175B parame- modelstheground-truthsectionwhichsupportstheanswer +ters. Here we feed each model only the text questions Q (Oracle-KBSectioninTab.5). Again,accuracyimproves +as thery cannot consume images. This measures how well to48.8%forPaLI,87.0%forPaLMand82.1%forGPT-3. +our dataset can be solved by language alone. These mod- Thisdemonstratestheimportanceofprovidingexactinfor- +elsreachamodestaccuracy(19.7%,15.5%,Tab.5). While mationtothelanguagecomponent. Asabonus,thesmaller +they only takes textual inputs, they can be made multi- the retrieved document is upon which the model bases its +modal by adding a visual retrieval mechanism, as we do answer,themoreverifiableandinterpretablethisansweris. +inthefollowingsections. +5.3.LargeModelswithVisualRetrieval +5.2.LargeModelswithOracleRetrieval +Lens-basedretrievalofKBarticle. Togobeyondtheora- +Oracle retrieval of subject C. This experiment tests cledemonstrationabove,asproof-of-conceptweproposeto +whether the most difficult aspect of Encyclopedic-VQA is augmentlargemodelswitharealretrievalsystembasedon +recognizingfine-grainedcategoriesandinstancesintheim- GoogleLens[1]. GoogleLensisanimageretrievalsystem +age. Therefore we provide along with the test question its whichindexesahugeamountofwebimages.Givenaquery +corresponding subject category C as an additional textual image,itfindsotherimagesbasedontheirvisualsimilarity +inputtoourmodels(i.e.aspartoftheprompt;Oracle-Sub- andrelevancetoobjectsitrecognizesinthequeryimage. It +ject C in Tab. 5). While PaLI shows small improvements returnsthemostsimilarindexedimagesalongwithanentity +over its non-retrieval version, PaLM and GPT-3 improve prediction based on these top-ranked images. To augment +considerably to 31.0% and 26.9%, respectively. However, oursystem,wesendGoogleLensthequeryimageI toob- +thisisstillratherlowperformance. Weconcludethatdeter- tain its entity prediction. We then find the best matching +miningthefine-grainedcategoryorinstanceisonlypartof KB article for this entity in our knowledge base. Finally, +whatmakesourdatasetdifficult. wefeedtheretrievedKBarticleasprompttoourmodelsas +| OracleretrievalofKB | article. | Wenowgoa | | stepfurther | | inSec.5.2. | | | | +| ------------------- | -------- | -------- | --- | ----------- | --- | ---------- | --- | --- | --- | +andprovidethefullground-truthWikipediaarticleaboutC Results are shown as Lens - KB Article in Tab. 5. All +tothemodelsintheprompt(Oracle,KBArticleinTab.5). models greatly outperform their non-retrieval augmented +This time, results increase dramatically: 29.7% accuracy versions. PaLMandGPT-3inparticularmorethandouble, +for PaLI, 78.4% for PaLM and 77.4% for GPT-3. This to48.0%and44.9%respectively. +demonstrates that (1) memorizing detailed properties is a Lens-basedretrievalofKBsection. GivenaretrievedKB +hard challenge for vanilla large VLMs, (2) adding a re- article, wenowaimtoselectthemostrelevantKBsection -17= ARYAMANFINANCIALSERVICES LIMITED -have not observed any continuing failure to -ANNEXURE REFERRED TO IN -PARAGRAPH 1 OF OUR REPORT OF correct major weakness and on the basis of -such checks carried out by us during the course -EVEN DATE ON THE ACCOUNTS FOR -THE YEAR ENDED 30th JUNE, 2005. of our audit, the company has an adequate -| | | | internal control | procedures commensurate | | | -| --- | --- | --- | ----------------- | ----------------------- | --- | --- | -In respect of Fixed Assets with the size and nature of its business. -The company has maintained proper records v. In respect of transactions covered under -a) -showing full particulars including quantitative Section 301 of the Companies Act. 1956 -details and situation of fixed assets. -| | | | a) In our | opinion and | according | to the | -| --- | --- | --- | ------------ | ------------- | ---------- | ------- | -b) As explained to us, the fixed assets have been -physically verified by the management during information and explanations given to us, the -the year and no material discrepancy was transactions that need to be entered into the -register maintained under Section 301 of the -noticed on such verification. -.) The company has not disposed / discarded any Act, have been so entered. -of its fixed assets during the year. -b) As per the information and explanations given -to us, there are no transactions of purchase -ii. In respect of inventories -The company is in the business of providing and sales of services have been made in -financial services and in view of this Para 4 (ii) pursuance of contracts or arrangements -| | | | entered | in the register maintained under | | | -| --- | --- | --- | -------- | -------------------------------- | --- | --- | -(a), (b) & (c) of the said order are not applicable -| | | | Section | 301 of the companies | | Act. 1956. | -| --- | --- | --- | -------- | --------------------- | --- | ---------- | -to the company. -aggregating during the year to Rs.5 Lakhs or -iii. In respect of loans, Secured and unsecured, more in respect of each party. -| Granted or taken | by the company | to / from | | | | | -| ----------------- | --------------- | ---------- | --- | --- | --- | --- | -Companies, firms or other parties listed in the vi. As per the information and explanations given -register maintained under Section 301 of the to us, the company has not accepted any -companies Act. 1956 : deposits from the public within the meaning -of Section 58A and 58AA of the Act and rules -a) The company has not granted any loan to framed there under. -Companies, Firms or other parties listed in the -register maintained unHer section 301 of The vii In our opinion, the company has an internal -audit system, which is commensurate with the -| Companies Act, | 1956. | | | | | | -| --------------- | ----- | --- | --- | --- | --- | --- | -b) In view of our comment Para 4 (iii) (a) the size and nature of its business. -clauses (b), (c) and (d) are not applicable to the -| company. | | | viii We have | been informed that | | the central | -| -------- | --- | --- | ------------- | -------------------- | --- | ------------ | -government has not prescribed maintenance -| c) The company has not taken any loan | | from | | | | | -| --------------------------------------- | --- | ---- | --- | --- | --- | --- | -Companies, Firms or other parties listed in the of cost records under Section 209( 1 )(d) of the -register maintained under section 301 of The companies Act. 1956. -| Companies Act, | 1956. | | | | | | -| --------------- | ----- | --- | --- | --- | --- | --- | -ix In respect of statutory dues -| cO In view of our comment | Para 4 (iii) (c) | the | | | | | -| --------------------------- | ----------------- | --- | --- | --- | --- | --- | -clauses (f) and (g) are not applicable to the -| company. | | | a) The company has been generally regular in | | | | -| -------- | --- | --- | --------------------------------------------- | --- | --- | --- | -depositing undisputed statutory dues including -iv. In our opinion and according to information Provident fund, Investors education and -and explanations given to us there are adequate Protection fund, Employees state insurance. -control procedures commensurate with the size Income Tax, Sales Tax, Wealth Tax, Service -| | | | Tax, Custom Duty, Excise Duty, Cess | | | and | -| --- | --- | --- | ------------------------------------ | --- | --- | --- | -of the company and the nature of its business -| | | | other statutory dues with the | | | appropriate | -| --- | --- | --- | ------------------------------ | --- | --- | ----------- | -with regard to purchase of fixed assets and sale -of services. During the course of our audit, we authorities. According to the information and +| | | | Accuracy | | | | 5.5.ComparisontoPromptCap | | | | | | | | +| ------------- | ------- | --- | --------- | ----------- | --- | ----------- | ------------------------- | ----- | ------------------- | ------ | ----------- | ------- | --------- | ------- | +| | | | w/correct | w/incorrect | | KBretrieval | | | | | | | | | +| | | | | | | | We compare | | our retrieval-based | | | methods | to | Prompt- | +| LensRetrieval | Overall | | retrieval | retrieval | | accuracy | | | | | | | | | +| | | | | | | | Cap [25], | which | has | strong | performance | | on OK-VQA | | +| KBArticle | 48.0% | | 77.7% | 21.2% | | 47.4% | | | | | | | | | +(Tab.4). PromptCapisasystemwithmultiplecomponents. +| KBSection | 48.8% | | 82.3% | 20.7% | | 45.6% | | | | | | | | | +| --------- | ----- | --- | ----- | ----- | --- | ----- | -------- | ---------- | ----- | ----- | -------- | --- | ------ | ----- | +| | | | | | | | It has a | captioning | model | which | consumes | | both I | and Q | +andgeneratesanimagecaptiontailoredtoanswertheques- +Table6:Attributionforretrieval-augmentedVLMs.Wereport +| | | | | | | | tion Q. | This caption | is | then | passed | to GPT-3 | as | context | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | --- | ---- | ------ | -------- | --- | ------- | +modelaccuracyforPaLM[17]conditionalonretrievalresults.Re- +trievalsuccessratesaresimilarforarticlesandsections,butPaLM toanswerQ. Additionally,PromptCapperformsin-context +hasbetteraccuracywhenaugmentedwithsections. learning: for a given test question, it uses the CLIP em- +| | | | | | | | beddings | [39] | of Q and | I to | find the | 32 nearest | | training | +| --- | --- | --- | --- | --- | --- | --- | --------- | ----------------------------------------- | -------- | ---- | -------- | ---------- | --- | -------- | +| | | | | | | | examples. | Thenitincludestheir32correspondingPrompt- | | | | | | | +Capcaptions,questionsandanswersintheGPT-3prompt. +| withinitbasedonQ. | | TodosowequeryPaLMwithaspe- | | | | | | | | | | | | | +| ----------------- | --- | -------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +cialpromptfeedingoneKBsectionatatime,alongwithQ In[25]theunderlyingLLMwasGPT-3. Inthisexperiment +while asking ‘can the answer to this question be found in wealsoapplyPromptCapwithPaLMandPaLI.Resultsin +thistext?’. WeretainallsectionsforwhichPaLManswers Tab. 5 show that PromptCap indeed does better than using +‘Yes’ (usually only one), and consider their concatenation PaLI, PaLM, or GPT-3 alone. However, it still performs +| | | | | | | | substantially | worse | than | our retrieval-augmented | | | methods. | | +| ------------------------------- | --- | --- | --- | --------------------- | --- | --- | ------------- | ----- | ---- | ----------------------- | --- | --- | -------- | --- | +| asthefinalretrieved‘KBsection’. | | | | Finally,wefeedthisre- | | | | | | | | | | | +trievedsectionintheprompttoallmodels(PaLI,PaLMor Thisfurtherconfirmsthehardchallengeourdatasetposes. +GPT-3)asinSec.5.2. +5.6.Multi-answerandtwo-hopquestions +| Results | are shown | as | Lens - | KB Section | in | Tab. 5. Per- | | | | | | | | | +| ------- | --------- | --- | ------ | ---------- | --- | ------------ | --- | --- | --- | --- | --- | --- | --- | --- | +formance is now even higher for PaLI (28.1%) and PaLM To evaluate multi-answer questions, we convert the +| | | | | | | | model prediction | | into | a set | of strings | and | compute | the | +| -------- | ------- | ----- | ---- | -------- | ---------- | -------- | --------------------------- | --- | ---- | ----- | ---------- | ---- | ------- | ------- | +| (48.8%), | whereas | GPT-3 | does | not seem | to benefit | further. | | | | | | | | | +| | | | | | | | intersection-over-the-union | | | (IoU) | between | this | set | and the | +Thesenumbersareroughlyhalfwaybetweenversionswith- +outretrievalandwithoracleretrieval. Weconcludethatre- set of ground-truth answers. If IoU >= 0.5 then we con- +| | | | | | | | sider that | prediction | as | correct. | If not, | then | we use | BEM | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ---------- | --- | -------- | ------- | ---- | ------ | --- | +trievalaugmentationworksinpractice,andthatourdataset +leavessignificantheadroomforfutureresearchonretrieval- to determine the equivalence of the prediction list string +| | | | | | | | and the | ground-truth | list. | For | two-hop | questions, | | we use | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------------ | ----- | --- | ------- | ---------- | --- | ------ | +augmentedVLMs. +| | | | | | | | BEM as | described | for | single-hop | templated | | and automati- | | +| ---------- | ---------- | --- | --- | -------- | ------ | -------- | ------ | --------- | --- | ---------- | --------- | --- | ------------- | --- | +| CLIP-based | retrieval. | | KAT | [22] and | REVIVE | [34] are | | | | | | | | | +callygeneratedquestions. +| two retrieval-augmented | | | VQA | systems | which | use frozen | | | | | | | | | +| ----------------------- | ---------- | --- | ------- | ---------- | ----- | ----------- | ---- | --------- | --- | ------- | --------- | ------- | --- | ----- | +| | | | | | | | PaLI | with Lens | KB | Section | Retrieval | obtains | an | accu- | +| CLIP [39] | embeddings | to | perform | retrieval. | | In Appendix | | | | | | | | | +racyof9.2%formulti-answerquestionsand14.7%fortwo- +| C we explore | there | whether | KAT | and | REVIVE | could po- | | | | | | | | | +| ------------ | ----- | ------- | --- | --- | ------ | --------- | -------------- | --- | ---------- | ---- | ---- | ---------- | --- | --------- | +| | | | | | | | hop questions. | | Similarly, | PaLM | with | KB Section | | retrieval | +tentiallysucceedonEncyclopedic-VQA. +| | | | | | | | achieves | 33.6% | and 22.8% | for | multi-answer | | and | two-hop | +| --- | --- | --- | --- | --- | --- | --- | ---------- | ------------- | --------- | ----- | ------------ | ----- | --- | ------- | +| | | | | | | | questions, | respectively. | | GPT-3 | obtains | 32.1% | for | multi- | +answerand18.7%fortwo-hopquestions,inbetweenPaLI +5.4.AttributionforRetrieval-Augmentedmodels +andPaLM.Thesemodestperformances,especiallyontwo- +hopquestions,confirmthechallengeofourproposedtasks, +| The attribution | | annotations | | of our dataset | | enable mea- | | | | | | | | | +| --------------- | --- | ----------- | --- | -------------- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | +andhighlighttheusefulnessofEncyclopedic-VQAtomea- +suringwhetherretrieval-augmentedmodelsgivethecorrect +| | | | | | | | sure progress | on | designing | retrieval | | mechanisms | for | such | +| ------------------------ | --- | --- | --------------------------- | --- | --- | --- | ------------- | --- | --------- | --------- | --- | ---------- | --- | ---- | +| answerfortherightreason. | | | Tab.6measuresattributionfor | | | | | | | | | | | | +complextypesofquestions. +thePaLMmodelswithLensretrieval. +| First, | we observe | that | Google | Lens | retrieves | the cor- | | | | | | | | | +| ------ | ---------- | ---- | ------ | ---- | --------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +6.Conclusions +| rect KB | Article 47.4% | | of the | time. Furthermore, | | we re- | | | | | | | | | +| ------- | ------------- | --- | ------ | ------------------ | --- | ------ | --- | --- | --- | --- | --- | --- | --- | --- | +port 45.6% retrieval accuracy at the finer level of a KB We introduced Encyclopedic-VQA, a large-scale VQA +Section. Thisdemonstratesthatgivenacorrectlyretrieved dataset about detailed properties of fine-grained categories +KBArticle,oursystemalmostalwaysfindsthecorrectSec- and instances, which includes a knowledge base with an- +tion within it. More importantly, if the retrieved KB Sec- swerattributions. Wedemonstratedthatourdatasetistruly +tionisincorrect,accuracydropsdrasticallyto20.7%,close difficultforstandardVLMs. Additionally,weshowedwith +tothenon-augmentedvariant(bothwhenusingPaLM).In both an oracle experiment and a prototype automatic sys- +contrast, if the correct KB Section is found, accuracy is tem that augmenting these models with a retrieval compo- +82.3%.Thus,retrievingthecorrectdocument,andevenbet- nent substantially improves results, yet leaving headroom +terthecorrectsection,isessentialtoperformingwellonour for even further improvements. Therefore our dataset en- +| dataset. | | | | | | | ablesfutureresearchonretrieval-augmentedVLMs. | | | | | | | | +| -------- | --- | --- | --- | --- | --- | --- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | -p ARYAMAN FINANCIAL SERVICES LIMITED -given any guarantee for loans taken by -explanation given to us, no undisputed amounts -payable in respect of the aforesaid dues were others from banks or financial -| outstanding as it 30th June, 2005 for a period of | | | institutions. | | | -| ------------------------------------------------- | --- | --- | ------------- | --- | --- | -more than six months from the date they became -payable except Profession Tax dues of Rs. 10507 xvi. The company has not raised any term -- and Income Tax(TDS Payable) Rs.1,06,864/- loans during the year. -xvii. On the basis of our examination of -b) Based on the information made available to us -and explanation given to us, the details of dues balance sheet it appears that the funds -which have not been deposited on account of raised on short term basis are not used -for long term basis and vice versa. -dispute are given below : -Particulars Assessment year Forum where -Amount For which the matter dispute is xviii. The company has not made any -(Rs. In Lakhs) preferential allotment of shares during -| Pertains to pending | | | the year to parties | and companies | | -| ------------------- | ------- | ----- | ---------------------------------------- | ------------- | --- | -| Income Tax | 1997-98 | | covered in the register maintained under | | | -| CIT (Appeals) | | 26.53 | section 301 of the companies Act. 1956. | | | -x. The accumulated loss of the company is not xix. The company has not issued any -more than 50% of its net worth and the company debentures during the year and therefore -has incurred cash loss during financial year para 4 (xix) of the order is not applicable -| covered by our audit and the company had not | | | to the company. | | | -| -------------------------------------------- | --- | --- | --------------- | --- | --- | -incurred cash loss in the immediately preceding -| financial year. | | | xx. The company has not raised any money | | | -| --------------- | --- | --- | ---------------------------------------- | --- | --- | -by way public issue during the year. -xi. Based on our audit procedures and according -to the information and explanation given to us xxi. During the course of our examination of -we are of the opinion that the company has not the books and records of the company -defaulted in repayments of dues to the financial carried out in accordance with generally -accepted auditing practices in India and -institution, banks or debenture holders. -| | | | according | to information | and | -| --- | --- | --- | ---------- | ---------------- | --- | -xii. In our opinion and according to the information explanation given to us we have neither -and explanations given to us no loans and come across any instance of fraud on or -advances have been granted by the company by the company during the year, nor have -on the basis of security by way of pledge of the we been informed of such case by the -| shares, debenture & other securities. | | | management. | | | -| ------------------------------------- | --- | ------------ | ----------- | --- | --- | -| xiii. The company is not a chit fund | | or a nidhi / | | | | -mutual benefit funds / society. Therefore para ForThakur Vaidyanath Aiyar & Co. -4 (xiii) is not applicable to the company. Chartered Accountants -xiv. Based on information and explanation given to -| us, the company is not dealing or trading in | | | Place: Mumbai | | | -| ---------------------------------------------- | ----------- | ---------- | ----------------- | --- | --- | -| shares, securities, | debentures | and other | | | | -| investments and hence the requirements of para | | | (C.V. Parameswar) | | | -4 (xiv) of the order are not applicable to the -| company. | | | Partner | | | -| -------- | --- | --- | ----------------- | ----------- | --- | -| | | | Date: 01-12-2005 | M.No. 11541 | | -xv. In our opinion and according to the information -& explanation given to us, the company has not +References [12] Jannis Bulian, Christian Buck, Wojciech Gajewski, Ben- +| | | | | | | | jamin | Boerschinger, | | and Tal | Schuster. | Tomayto, | | tomahto. | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------------- | --- | ------- | --------- | -------- | --- | -------- | +[1] GoogleLens. https://lens.google.com-Webinterfaceavail- beyondtoken-levelanswerequivalenceforquestionanswer- +| ableathttps://images.google.com. | | | | 8,13 | | | | | | | | | | | +| -------------------------------- | --------- | ----- | --------- | --------- | ------------ | --- | -------------- | --- | ------------------------------ | --- | --- | --- | --- | -------- | +| | | | | | | | ingevaluation. | | arXivpreprintarXiv:2202.07654, | | | | | 2022. 7, | +| [2] Google | universal | image | embedding | challenge | leaderboard. | | | | | | | | | | +13 +https://www.kaggle.com/competitions/google-universal- [13] Soravit Changpinyo, Doron Kukliansky, Idan Szpektor, Xi +image-embedding/leaderboard. 14 Chen, NanDing, andRaduSoricut. Allyoumayneedfor +[3] Huggingface. https://huggingface.co/sentence- VQAareimagecaptions. InNAACL,2022. 4,5 +transformers/multi-qa-MiniLM-L6-cos-v1. 7 [14] XiChen,XiaoWang,SoravitChangpinyo,AJPiergiovanni, +[4] inat challenge 2021 - fgvc8 - leaderboard. Piotr Padlewski, Daniel Salz, Sebastian Goodman, Adam +https://www.kaggle.com/competitions/inaturalist- Grycner,BasilMustafa,LucasBeyer,AlexanderKolesnikov, +| | | | | | | | Joan | Puigcerver, | Nan | Ding, | Keran | Rong, | Hassan | Akbari, | +| ----------------- | --- | --- | --- | --- | --- | --- | ---- | ----------- | --- | ----- | ----- | ----- | ------ | ------- | +| 2021/leaderboard. | | 14 | | | | | | | | | | | | | +[5] WikimediaCommons. https://commons.wikimedia.org. 4 GauravMishra,LintingXue,AshishThapliyal,JamesBrad- +| | | | | | | | bury, | Weicheng | Kuo, | Mojtaba | Seyedhosseini, | | | Chao Jia, | +| --- | --- | --- | --- | --- | --- | --- | ----- | -------- | ---- | ------- | -------------- | --- | --- | --------- | +[6] ArmenAghajanyan,BernieHuang,CandaceRoss,Vladimir +| | | | | | | | Burcu | Karagol | Ayan, | Carlos | Riquelme, | | Andreas | Steiner, | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------- | ----- | ------ | --------- | --- | ------- | -------- | +Karpukhin,HuXu,NamanGoyal,DmytroOkhonko,Man- +| | | | | | | | Anelia | Angelova, | Xiaohua | | Zhai, Neil | Houlsby, | | and Radu | +| --- | --- | --- | --- | --- | --- | --- | ------ | --------- | ------- | --- | ---------- | -------- | --- | -------- | +darJoshi,GargiGhosh,MikeLewis,andLukeZettlemoyer. +| | | | | | | | Soricut. | PaLI: | A Jointly-Scaled | | | Multilingual | Language- | | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ---------------- | --- | --- | ------------ | --------- | --- | +CM3:Acausalmaskedmultimodalmodeloftheinternet.In +| | | | | | | | ImageModel. | | arXiv,2022. | 1,2,3,7,8 | | | | | +| ----------------- | ------------ | ---------- | ------------- | ------------- | ----------- | --------- | ----------------------------- | --------- | ------------ | ---------- | -------------- | --------- | -------- | --------- | +| arXiv,2022. | | 3 | | | | | | | | | | | | | +| | | | | | | | [15] Yang | Chen, | Hexiang | Hu, Yi | Luan, | Haitian | Sun, | Soravit | +| [7] Jean-Baptiste | | Alayrac, | Jeff Donahue, | | Pauline | Luc, An- | | | | | | | | | +| | | | | | | | Changpinyo, | | Alan Ritter, | | and Ming-Wei | | Chang. | Can | +| toine | Miech, | Iain Barr, | Yana | Hasson, Karel | Lenc, | Arthur | | | | | | | | | +| | | | | | | | pre-trained | | vision and | language | | models | answer | visual | +| Mensch, | Katie | Millican, | Malcolm | Reynolds, | Roman | Ring, | | | | | | | | | +| | | | | | | | information-seekingquestions? | | | | InarXiv,2023. | | 4 | | +| Eliza | Rutherford, | Serkan | Cabi, | Tengda Han, | Zhitao | Gong, | | | | | | | | | +| | | | | | | | [16] Yen-Chun | Chen, | Linjie | Li, | Licheng | Yu, Ahmed | | El Kholy, | +| Sina | Samangooei, | Marianne | Monteiro, | Jacob | Menick, | Se- | | | | | | | | | +| | | | | | | | Faisal | Ahmed, | Zhe | Gan, | Yu Cheng, | and | Jingjing | Liu. | +| bastian | Borgeaud, | Andrew | Brock, | Aida | Nematzadeh, | Sa- | | | | | | | | | +| | | | | | | | UNITER: | UNiversal | | Image-TExt | Representation | | | Learning. | +| hand | Sharifzadeh, | Mikolaj | Binkowski, | | Ricardo | Barreira, | | | | | | | | | +| | | | | | | | InECCV,2020. | | 4 | | | | | | +| Oriol | Vinyals, | Andrew | Zisserman, | and | Karen | Simonyan. | | | | | | | | | +| | | | | | | | [17] Aakanksha | | Chowdhery, | Sharan | | Narang, | Jacob | Devlin, | +Flamingo:aVisualLanguageModelforFew-ShotLearning. +| | | | | | | | Maarten | Bosma, | Gaurav | | Mishra, | Adam | Roberts, | Paul | +| ------------- | ------ | --------- | -------- | ------ | --- | -------- | ------- | --------- | ------ | ------ | ------- | ------ | -------- | ------ | +| NeurIPS,2022. | | 1,2,3 | | | | | | | | | | | | | +| | | | | | | | Barham, | Hyung | Won | Chung, | Charles | | Sutton, | Sebas- | +| [8] Stanislaw | Antol, | Aishwarya | Agrawal, | Jiasen | Lu, | Margaret | | | | | | | | | +| | | | | | | | tian | Gehrmann, | Parker | | Schuh, | Kensen | Shi, | Sasha | +Mitchell, Dhruv Batra, C. Lawrence Zitnick, and Devi Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker +Parikh. VQA: visual question answering. In ICCV, 2015. Barnes, Yi Tay, Noam Shazeer, Vinodkumar Prabhakaran, +3,13 +| | | | | | | | EmilyReif, | | NanDu, | BenHutchinson, | | ReinerPope, | | James | +| ------------- | --------- | ------ | --- | ------- | ------ | --------- | ---------- | ----- | ------- | -------------- | ------- | ----------- | --- | -------- | +| [9] Sebastian | Borgeaud, | Arthur | | Mensch, | Jordan | Hoffmann, | | | | | | | | | +| | | | | | | | Bradbury, | Jacob | Austin, | | Michael | Isard, | Guy | Gur-Ari, | +TrevorCai,ElizaRutherford,KatieMillican,Georgevanden Pengcheng Yin, Toju Duke, Anselm Levskaya, Sanjay +Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier +Clark,DiegodeLasCasas,AureliaGuy,JacobMenick,Ro- Garcia,VedantMisra,KevinRobinson,LiamFedus,Denny +manRing,TomHennigan,SaffronHuang,LorenMaggiore, +| | | | | | | | Zhou, | Daphne | Ippolito, | David | Luan, | | Hyeontaek | Lim, | +| --- | --- | --- | --- | --- | --- | --- | ----- | ------ | --------- | ----- | ----- | --- | --------- | ---- | +ChrisJones,AlbinCassirer,AndyBrock,MichelaPaganini, +| | | | | | | | Barret | Zoph, | Alexander | Spiridonov, | | Ryan | Sepassi, | David | +| --- | --- | --- | --- | --- | --- | --- | ------ | ----- | --------- | ----------- | --- | ---- | -------- | ----- | +GeoffreyIrving, OriolVinyals, SimonOsindero, KarenSi- Dohan, Shivani Agrawal, Mark Omernick, Andrew M. +monyan,JackW.Rae,ErichElsen,andLaurentSifre. Im- Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, +provinglanguagemodelsbyretrievingfromtrillionsofto- AitorLewkowycz, EricaMoreira, RewonChild, Oleksandr +| kens. | arXiv:2112.04426,2022. | | | 1,4 | | | | | | | | | | | +| ----- | ---------------------- | --- | --- | --- | --- | --- | -------- | --------- | --- | ------------ | --- | ----- | ------ | ----- | +| | | | | | | | Polozov, | Katherine | | Lee, Zongwei | | Zhou, | Xuezhi | Wang, | +[10] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- Brennan Saeta, Mark Diaz, Orhan Firat, Michele Catasta, +biah,JaredDKaplan,PrafullaDhariwal,ArvindNeelakan- JasonWei,KathyMeier-Hellstern,DouglasEck,JeffDean, +tan,PranavShyam,GirishSastry,AmandaAskell,etal.Lan- Slav Petrov, and Noah Fiedel. PaLM: Scaling Language +guagemodelsarefew-shotlearners. Advancesinneuralin- ModelingwithPathways. arXiv,2022. 5,7,8,9,14 +formationprocessingsystems,33:1877–1901,2020. +| | | | | | | 7,8 | [18] Hyung | Won | Chung, | Le Hou, | Shayne | | Longpre, | Barret | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ------ | ------- | ------ | --- | -------- | ------ | +[11] TomB.Brown,BenjaminMann,NickRyder,MelanieSub- Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, +biah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakan- Mostafa Dehghani, Siddhartha Brahma, et al. Scaling +tan, Pranav Shyam, Girish Sastry, Amanda Askell, Sand- instruction-finetuned language models. arXiv preprint +| | | | | | | | arXiv:2210.11416,2022. | | | 5,14 | | | | | +| ------------- | --- | ------------------- | --- | -------- | -------- | --- | ---------------------- | --- | --- | ---- | --- | --- | --- | --- | +| hini Agarwal, | | Ariel Herbert-Voss, | | Gretchen | Krueger, | Tom | | | | | | | | | +Henighan,RewonChild,AdityaRamesh,DanielM.Ziegler, [19] Matthew Dunn, Levent Sagun, Mike Higgins, V. Ugur +JeffreyWu,ClemensWinter,ChristopherHesse,MarkChen, Guney, Volkan Cirik, and Kyunghyun Cho. SearchQA: A +Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, newQ&Adatasetaugmentedwithcontextfromasearchen- +JackClark,ChristopherBerner,SamMcCandlish,AlecRad- gine. InarXiv,2017. 4 +ford,IlyaSutskever,andDarioAmodei. LanguageModels [20] HaoyuanGao, JunhuaMao, JieZhou, ZhihengHuang, Lei +areFew-ShotLearners. arXiv,2020. 8 Wang,andWeiXu. Areyoutalkingtoamachine? Dataset -ARYAMAN FINANCIAL SERVICES LIMITED -CERTIFICATE -To the Members of Aryaman Financial Services Ltd. -We have examined the compliance of conditions of Corporate Governance by Aryaman Financial -Services Ltd. for the year 30th June, 2005 as stipulated in clause 49 of the Listing Agreement -executed by the said Company with stock exchanges. -The Compliance of conditions of Corporate Governance is the responsibility of the Management. -Our examination was limited to procedures and implementation thereof, adopted by the Company -for ensuring the compliance of the conditions of the Corporate Governance, It is neither an audit -nor an expression of opinion on the financial statements of the Company. -In our opinion and to the best of our information and according to the explanations given to us -we invite your attention to the following. -1. Audit Committee / Remuneration Committee / Shareholders / Investors Grievance Committee as -required by clause 49 of the Listing Agreement is not formed till the end of the accounting year. -2. Means of Communication regarding publication of Quarterly results in the news papers are not -complied with. -Subject to above we certify that the Company has complied with the conditions of Corporate -Governance as stipulated in the above-mentioned Listing Agreement. -We have to state that no investor grievance is pending for a period exceeding one month against the -company as per the records maintained by the registrar & share transfer agents of the company. -We further state that such compliance is neither an assurance as to the future viability of the Company -nor the efficiency or effectiveness with which the Management has conducted the affairs of the -Company. -For and on behalf of -Thakur Vaidyanath Aiyar & Co. -Chartered Accountants. -C.V.Parameswar -Partner. -Date : 1st December, 2005 M.No.11541 -Place : Mumbai +andmethodsformultilingualimagequestion. InNeurIPS, [34] YuanzeLin,YujiaXie,DongdongChen,YichongXu,Chen- +| 2015. | 3 | | | | | | guangZhu,andLuYuan. | | | Revive:Regionalvisualrepresen- | | | +| ----- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --------------- | ------------------------------ | --------------- | ------- | +| | | | | | | | tation matters | in | knowledge-based | | visual question | answer- | +[21] YashGoyal,TejasKhot,DouglasSummers-Stay,DhruvBa- +tra, and Devi Parikh. Making the V in VQA matter: Ele- ing. arXivpreprintarXiv:2206.01201,2022. 9,13,14 +vating the role of image understanding in Visual Question [35] Mateusz Malinowski and Mario Fritz. A multi-world ap- +Answering. InCVPR,2017. 3,7,13 proachtoquestionansweringaboutreal-worldscenesbased +InNeurIPS,2014. +[22] Liangke Gui, Borui Wang, Qiuyuan Huang, Alex Haupt- onuncertaininput. 3 +mann, Yonatan Bisk, and Jianfeng Gao. KAT: A Knowl- [36] KennethMarino,XinleiChen,DeviParikh,AbhinavGupta, +edgeAugmentedTransformerforVision-and-Language. In and Marcus Rohrbach. KRISP: Integrating Implicit and +NAACL,2022. 1,4,7,9,13,14 Symbolic Knowledge for Open-Domain Knowledge-Based +[23] KelvinGuu,KentonLee,ZoraTung,PanupongPasupat,and VQA. InProc.CVPR,2021. 1,4 +Ming-WeiChang.REALM:Retrieval-AugmentedLanguage [37] Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and +ModelPre-Training. InProc.ICML,2020. 1,4 RoozbehMottaghi. OK-VQA:Avisualquestionanswering +| [24] Yaru | Hao, | Haoyu Song, | Li | Dong, Shaohan | Huang, | Zewen | | | | | | | +| ------------------------------------ | ---- | ----------- | --- | ------------- | ------ | -------- | ---------- | --------- | -------- | ---------- | --- | ----------- | +| | | | | | | | benchmark | requiring | external | knowledge. | In | CVPR, 2019. | +| Chi,WenhuiWang,ShumingMa,andFuruWei. | | | | | | Language | 1,2,3,7,13 | | | | | | +modelsaregeneral-purposeinterfaces. InarXiv,2022. 1,2, [38] TodorMihaylov,PeterClark,TusharKhot,andAshishSab- +3 +| | | | | | | | harwal. | Can a | suit of armor | conduct | electricity? | a new | +| --- | --- | --- | --- | --- | --- | --- | ------- | ----- | ------------- | ------- | ------------ | ----- | +[25] YushiHu,HangHua,ZhengyuanYang,WeijiaShi,NoahA +datasetforopenbookquestionanswering.InEMNLP,2018. +| Smith, | and | Jiebo Luo. | Promptcap: | Prompt-guided | | task- | 4 | | | | | | +| --------------------- | --- | ---------- | ------------------------------ | ------------- | ----- | ---------- | ------------------ | ------- | ------------- | -------- | -------------- | -------------- | +| awareimagecaptioning. | | | arXivpreprintarXiv:2211.09699, | | | | | | | | | | +| | | | | | | | [39] Alec Radford, | | Jong Wook | Kim, | Chris Hallacy, | Aditya | +| 2022. | 7,9 | | | | | | | | | | | | +| | | | | | | | Ramesh, | Gabriel | Goh, Sandhini | | Agarwal, | Girish Sastry, | +| [26] Ziniu | Hu, | Ahmet | Iscen, | Chen Sun, | Zirui | Wang, Kai- | | | | | | | +| | | | | | | | Amanda | Askell, | Pamela | Mishkin, | Jack Clark, | Gretchen | +Wei Chang, Yizhou Sun, Cordelia Schmid, David A. Krueger, and Ilya Sutskever. Learning transferable visual +Ross, and Alireza Fathi. REVEAL: Retrieval-Augmented modelsfromnaturallanguagesupervision. InICML,2021. +| Visual-Language | | Pre-Training | | with Multi-Source | | Multi- | | | | | | | +| --------------- | --- | ------------ | --- | ----------------- | --- | ------ | --- | --- | --- | --- | --- | --- | +3,9,13 +| modalKnowledgeMemory. | | | | InCVPR,2023. | 1,4,7 | | | | | | | | +| --------------------- | --- | --- | --- | ------------ | ----- | --- | --------------- | ------------------ | --- | --- | ---------- | ----------- | +| | | | | | | | [40] JackW.Rae, | SebastianBorgeaud, | | | TrevorCai, | KatieMilli- | +[27] Aman Jain, Mayank Kothyari, Vishwajeet Kumar, Preethi can,JordanHoffmann,FrancisSong,JohnAslanides,Sarah +Jyothi,GaneshRamakrishnan,andSoumenChakrabarti.Se- Henderson, Roman Ring, Susannah Young, Eliza Ruther- +| lect, | substitute, | search: | A | new benchmark | for | knowledge- | | | | | | | +| ----- | ----------- | ------- | --- | ------------- | --- | ---------- | --- | --- | --- | --- | --- | --- | +ford,TomHennigan,JacobMenick,AlbinCassirer,Richard +| augmentedvisualquestionanswering. | | | | InSIGIR,2021. | | 2,3, | | | | | | | +| --------------------------------- | --- | --- | --- | ------------- | --- | ---- | -------------- | --- | ------------------ | --- | --------- | ---------- | +| | | | | | | | Powell, George | | van den Driessche, | | Lisa Anne | Hendricks, | +4,5,7 +| | | | | | | | Maribeth | Rauh, | Po-Sen Huang, | | Amelia Glaese, | Johannes | +| --- | --- | --- | --- | --- | --- | --- | -------- | ----- | ------------- | --- | -------------- | -------- | +[28] ChaoJia,YinfeiYang,YeXia,Yi-TingChen,ZaranaParekh, Welbl, Sumanth Dathathri, Saffron Huang, Jonathan Ue- +HieuPham,QuocV.Le,Yun-HsuanSung,ZhenLi,andTom sato, John Mellor, Irina Higgins, Antonia Creswell, Nat +Duerig.Scalingupvisualandvision-languagerepresentation +| | | | | | | | McAleese, | Amy | Wu, Erich | Elsen, | Siddhant | Jayakumar, | +| --------------------------------- | --- | --- | --- | ------------ | --- | --- | --------- | --- | --------- | ------ | -------- | ---------- | +| learningwithnoisytextsupervision. | | | | InICML,2021. | | 3 | | | | | | | +ElenaBuchatskaya,DavidBudden,EsmeSutherland,Karen +[29] MandarJoshi,EunsolChoi,DanielS.Weld,andLukeZettle- Simonyan, MichelaPaganini, LaurentSifre, LenaMartens, +moyer. TriviaQA: A large scale distantly supervised chal- Xiang Lorraine Li, Adhiguna Kuncoro, Aida Nematzadeh, +lengedatasetforreadingcomprehension. InACL,2017. 4 Elena Gribovskaya, Domenic Donato, Angeliki Lazaridou, +[30] UrvashiKhandelwal,OmerLevy,DanJurafsky,LukeZettle- ArthurMensch,Jean-BaptisteLespiau,MariaTsimpoukelli, +moyer,andMikeLewis. Generalizationthroughmemoriza- NikolaiGrigorev,DougFritz,ThibaultSottiaux,MantasPa- +tion: Nearestneighborlanguagemodels. InICLR,2020. 1, jarskas, Toby Pohlen, Zhitao Gong, Daniel Toyama, Cy- +| 4 | | | | | | | priendeMassond’Autume,YujiaLi,TayfunTerzi,Vladimir | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------------------------------------- | --- | --- | --- | --- | --- | +Mikulik,IgorBabuschkin,AidanClark,DiegodeLasCasas, +| [31] Tushar | Khot, | Peter | Clark, | Michal Guerquin, | Peter | Jansen, | | | | | | | +| ----------- | ----- | ----- | ------ | ---------------- | ----- | ------- | --- | --- | --- | --- | --- | --- | +and Ashish Sabharwal. QASC: A dataset for question an- Aurelia Guy, Chris Jones, James Bradbury, Matthew John- +sweringviasentencecomposition. InAAAI,2020. 4 son, Blake Hechtman, Laura Weidinger, Iason Gabriel, +[32] Alina Kuznetsova, Hassan Rom, Neil Alldrin, Jasper Ui- WilliamIsaac,EdLockhart,SimonOsindero,LauraRimell, +| | | | | | | | Chris Dyer, | Oriol | Vinyals, | Kareem | Ayoub, | Jeff Stanway, | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ----- | -------- | ------ | ------ | ------------- | +jlings,IvanKrasin,JordiPont-Tuset,ShahabKamali,Stefan +| | | | | | | | Lorrayne | Bennett, | Demis | Hassabis, | Koray | Kavukcuoglu, | +| ----------------------- | ------ | -------- | --- | --------------------------- | -------- | -------- | ------------------- | -------- | ------- | --------- | ------- | ------------ | +| Popov, | Matteo | Malloci, | Tom | Duerig, and | Vittorio | Ferrari. | | | | | | | +| | | | | | | | and GeoffreyIrving. | | Scaling | Language | Models: | Methods, | +| TheOpenImagesDatasetV4: | | | | Unifiedimageclassification, | | | | | | | | | +object detection, and visual relationship detection at scale. Analysis&InsightsfromTrainingGopher. arXiv,2022. 8 +IJCV,2020. 4 [41] ColinRaffel,NoamShazeer,AdamRoberts,KatherineLee, +SharanNarang,MichaelMatena,YanqiZhou,WeiLi,,and +| [33] Patrick | Lewis, | Ethan | Perez, | Aleksandra | Piktus, | Fabio | | | | | | | +| ------------ | ------ | ----- | ------ | ---------- | ------- | ----- | --- | --- | --- | --- | --- | --- | +Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich PeterJLiu. Exploringthelimitsoftransferlearningwitha +Ku¨ttler, Mike Lewis, Wen tau Yih, Tim Rockta¨schel, Se- unifiedtext-to-texttransformer. JMLR,2021. 13 +bastian Riedel, and Douwe Kiela. Retrieval-Augmented [42] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and +Generation for Knowledge-Intensive NLP Tasks. In Proc. PercyLiang.SQuAD:100,000+questionsformachinecom- +| NeurIPS,2020. | | 1,4 | | | | | prehensionoftext. | | InEMNLP,2016. | | 4 | | +| ------------- | --- | --- | --- | --- | --- | --- | ----------------- | --- | ------------- | --- | --- | --- | -1= ARYAMANFINANCIALSERVICES LIMITED -CORPORATE GOVERNANCE REPORT - 2005 -PHILOSOPHY ON CORPORATE GOVERNANCE : -Board of Directors of the Company acts as a trustee and assumes fiduciary responsibility of -protecting the interests of the shareholders and other stakeholders of the Company. Board -supports the broad principles of Corporate Governance. In order to attain the highest level of -good Corporate Governance practice, Board lays strong emphasis on transparency, -accountability and integrity. -Board of Directors -Composition and category of directors; attendance of each director at the board meetings and -the last AGM; number of other Public Companies on the Board or Committees of which a -director is a member or chairperson; and number of board meetings, dates on which held, are -as under: -Board Meeting Attended during the Year 2004-05 Attendance at the Last AGM -Name of Director Category (attended) -Shri PVR Murthy* E & NI 3 Yes -Shri Virendra Ganda NE & I 8 Yes -Shri Mohan Dattary NE & 1 8 Yes -Shri Munnangi V Ramana** NE & I 5 N.A. -Shri Suhas Ganpule*** NE & II N.A. -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -* Resigned w.e.f. December 24, 2004 , -** Appointed w.e.f December 20, 2004 -*** Appointed w.e.f. May 02, 2005 -8 Meetings of the board of directors were held during the financial year 2004-05. One each on -April 30,2004, July 29,2004, October 28,2004, December 20,2004, December 24,2005, January -28, 2005, March 22, 2005 and May 18, 2005 -AUDIT COMMITTEE -Terms of reference : -The terms of reference stipulated by the Board of Directors to the Audit Committee are as -contained under clause 49 of the Listing Agreement. -Composition: -The Audit Committee was constituted by the Board of Directors in the meeting held on -December 01, 2005. The composition of the Committee and attendance of the members is -given hereunder: +[43] NilsReimersandIrynaGurevych. Sentence-bert: Sentence [60] Lu Yuan, Dongdong Chen, Yi-Ling Chen, Noel Codella, +embeddingsusingsiamesebert-networks. InEMNLP,2019. Xiyang Dai, Jianfeng Gao, Houdong Hu, Xuedong Huang, +| 7 | | | | | | | | BoxinLi,ChunyuanLi,CeLiu,MengchenLiu,ZichengLiu, | | +| --- | --- | --- | --- | --- | --- | --- | --- | ------------------------------------------------ | --- | +[44] MengyeRen,RyanKiros,andRichardS.Zemel. Exploring YumaoLu,YuShi,LijuanWang,JianfengWang,BinXiao, +modelsanddataforimagequestionanswering. InNeurIPS, ZhenXiao,JianweiYang,MichaelZeng,LuoweiZhou,and +| 2015. | 4 | | | | | | | PengchuanZhang. | Florence: ANewFoundationModelfor | +| ------------ | ------ | ---------- | --- | ------------------ | --- | --- | ---------- | --------------- | -------------------------------- | +| | | | | | | | | ComputerVision. | arXiv,2021. 3 | +| [45] Eleanor | Rosch. | Principles | | of categorization. | | | In Eleanor | | | +RoschandB.B.Lloyd,editors,CognitionandCategoriza- +| tion,pages27–48.LawrenceErlbaumAssociates,1978. | | | | | | | 2 | | | +| ----------------------------------------------- | -------- | ------ | ------- | ----------- | --- | ----------- | ------ | --- | --- | +| [46] Dustin | Schwenk, | Apoorv | | Khandelwal, | | Christopher | Clark, | | | +| Kenneth | Marino, | and | Roozbeh | Mottaghi. | | A-OKVQA: | A | | | +benchmarkforvisualquestionansweringusingworldknowl- +| edge. | InECCV,2022. | | 2,3,7,13 | | | | | | | +| ------------------------ | ------------ | --------------------------------------- | ------------ | ------------ | -------------------- | ------- | --------- | --- | --- | +| [47] Sanket | Shah, | Anand | Mishra, | | Naganand | Yadati, | and | | | +| Partha | Pratim | Talukdar. | KVQA: | | Knowledge-aware | | visual | | | +| questionanswering. | | | InAAAI,2019. | | 2,3,7 | | | | | +| [48] Krishna | Srinivasan, | | Karthik | Raman, | Jiecao | Chen, | Michael | | | +| Bendersky,andMarcNajork. | | | | Wit: | Wikipedia-basedimage | | | | | +| text | dataset | for multimodal | | multilingual | | machine | learning. | | | +| SIGIR,2021. | | 3,4 | | | | | | | | +| [49] AndyStirling. | | Ageneralframeworkforanalysisngdiversity | | | | | | | | +| in science, | | technology, | and | society. | Journal | of | the Royal | | | +| SocietyInterface,2007. | | | 7 | | | | | | | +[50] AlonTalmorandJonathanBerant.Thewebasaknowledge- +| baseforansweringcomplexquestions. | | | | | InNAACL,2018. | | 4 | | | +| --------------------------------- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | +[51] GrantvanHorn,ElijahCole,SaraBeery,KimberlyWilber, +| SergeBelongie,andOisinMacAodha. | | | | | BenchmarkingRep- | | | | | +| ------------------------------- | --- | -------- | ----------- | ----- | ---------------- | ------------ | --- | --- | --- | +| resentation | | Learning | for Natural | World | Image | Collections. | | | | +| InCVPR,2021. | | 2,4 | | | | | | | | +[52] PeterVickers,NikolaosAletras,EmilioMonti,andLoicBar- +| rault. | InFactuality: | | EfficientIntegrationofRelevantFacts | | | | | | | +| --------------------------- | ------------- | --- | ----------------------------------- | ----------- | --- | --- | --- | --- | --- | +| forVisualQuestionAnswering. | | | | InACL,2021. | | 1,4 | | | | +[53] PengWang,QiWu,ChunhuaShen,AnthonyDick,andAn- +| tonvandenHengel. | | | FVQA:Fact-basedvisualquestionan- | | | | | | | +| ---------------- | ---------------------- | ------ | -------------------------------- | ---- | ----------- | --- | ------- | --- | --- | +| swering. | IEEETrans.onPAMI,2018. | | | | 3,7 | | | | | +| [54] Jason | Wei, | Xuezhi | Wang, | Dale | Schuurmans, | | Maarten | | | +Bosma,BrianIchter,FeiXia,EdChi,QuocLe,andDenny +| Zhou. | Chain-of-Thought | | | Prompting | Elicits | Reasoning | in | | | +| -------------------- | ---------------- | ------ | --------------- | --------- | ------- | --------- | ------- | --- | --- | +| LargeLanguageModels. | | | InNeurIPS,2022. | | | 14 | | | | +| [55] Johannes | Welbl, | Pontus | Stenetorp, | | and | Sebastian | Riedel. | | | +Constructingdatasetsformulti-hopreadingcomprehension +| acrossdocuments. | | | InACL,2018. | | 4 | | | | | +| ---------------- | --- | ---------- | ----------- | -------- | ------- | ------ | ----- | --- | --- | +| [56] T. Weyand, | | A. Araujo, | B. | Cao, and | J. Sim. | Google | land- | | | +marksdatasetv2-alarge-scalebenchmarkforinstance-level +| recognitionandretrieval. | | | InCVPR,2020. | | | 2,4 | | | | +| ------------------------ | --- | --- | ------------ | --- | --- | --- | --- | --- | --- | +[57] ZhengyuanYang,ZheGan,JianfengWang,XiaoweiHu,Yu- +| maoLu,ZichengLiu,andLijuanWang. | | | | | | Anempiricalstudy | | | | +| ------------------------------------- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --- | +| ofgpt-3forfew-shotknowledge-basedvqa. | | | | | | InAAAI,2022. | | | | +1,2,3 +| [58] Zhilin | Yang, | Peng | Qi, Saizheng | | Zhang, | Yoshua | Bengio, | | | +| ------------------------------- | ------ | ------------ | -------------- | ------------- | ------------ | ----------- | ----------- | --- | --- | +| William | Cohen, | Ruslan | Salakhutdinov, | | and | Christopher | D. | | | +| Manning. | | HotpotQA: | A | dataset | for diverse, | | explainable | | | +| multi-hopquestionanswering. | | | | InEMNLP,2018. | | | 3,4,6 | | | +| [59] Licheng | Yu, | Eunbyung | Park, | Alexander | | C. | Berg, and | | | +| Tamara | L. | Berg. Visual | madlibs: | | Fill in | the blank | image | | | +| generationandquestionanswering. | | | | | InICCV,2015. | | 3 | | | -r ARYAMANFINANCIALSERVICES LIMITED -| Name of Member | Member/ Chairman | Category | -| ----------------------- | ----------------- | -------- | -| Shri Baliga | Chairman | NE & I | -| Shri Mohan Datari | Member | NE & I | -| Shri Munnangi V Ramana | Member | NE & I | -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -The Company Secretary of the Company, Shri Prasanna Chandwaskar is the Secretary of the -Committee. Director Finance, Internal Auditors and Statutory Auditors are the permanent invitees -to the Audit Committee Meeting. -REMUNERATION COMMITTEE -The Remuneration Committee was constituted by the Board of Directors in its meeting held on -December 01, 2005. No meeting of the Committee was held during the year. The terms of -reference, compositions are as under: -Terms of reference: To recommend appointment of Managerial Personnel and consider the -remuneration payable to them and review thereof from time to time. -Composition: -| Name of Member | Member/ Chairman | | -| ----------------------- | ---------------- | --- | -| Shri Mohan Datari | Chairman | | -| Shri Baliga | Member | | -| Shri Munnangi V Ramana | Member | | -E - Executive, NE - Non Executive, I - Independent, NI - Non Independent -REMUNERATION TO EXECUTIVE DIRECTORS: -The details of remuneration paid/ payable to the directors for the Financial Year 2004-05 are as under: -Shri PVR Murthy - Managing Director (Resigned w.e.f. December 24, 2004) -(Rs.) -| Salary & Allowances | 60000/- | | -| -------------------- | ------- | --- | -| Perquisites | 42000/- | | -Remuneration to Non-Executive Directors: -The Non-executive directors do not draw any remuneration from the Company. Non-executive -independent directors are entitled to the sitting fees for every meeting of the Board and Audit -Committee attended. -Non-executive independent directors are paid Rs. 1500/- for attending Board Meeting and Rs. -1OOO/- for Audit Committee. +A.Correctnessofourdataset astrictevaluationmeasureandbecausetheirquestionsare +| | | | | | | | more | open-ended | | (they | are not supported | | by a controlled | | +| ------- | --- | ------ | --------- | ------- | ------- | --------- | -------------- | ---------- | --- | ------- | ----------------- | --------- | --------------- | ------- | +| We rely | on | iNat21 | and GLDv2 | to have | clearly | identifi- | | | | | | | | | +| | | | | | | | knowledgebase, | | | Sec.4), | thosedatasets | collect10 | | ground- | +ablecategoriesonthetestimagesofourdataset,andtoob- +truthanswersperquestion,tocoversomeofthevariability +taintheirground-truthlabels.Furthermore,weonlyusecat- in answer formulation. This then enables a more relaxed +egorieswhichunambiguouslymaptoWikipediaarticlesby +| | | | | | | | evaluation, | | as the | model | answer need | only | match | some of | +| ------------------------------------------------ | --- | --- | --- | --- | --- | --- | ----------- | --------------- | ------ | ----- | -------------- | ---- | ------------- | ------- | +| explicitlyseekingforone-to-onemappings(Sec.4.1). | | | | | | The | | | | | | | | | +| | | | | | | | the | 10 ground-truth | | ones | (three matches | | for a perfect | an- | +mainremainingpossiblesourceoferroriswhetheraques- +swer). Nevertheless,itisinstructivetoverifywhetherexact +| tion can | be answered | | given | the corresponding | | Wikipedia | | | | | | | | | +| -------- | ----------- | --- | ----- | ----------------- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | --- | +matchingwouldworkonourdataset,wherewehaveasin- +| article, | and whether | | the recorded | ground-truth | | answer is | | | | | | | | | +| -------- | ----------- | ----- | ------------ | -------------- | --- | --------- | --- | ------------ | --- | ------- | -------- | -------- | -------- | ------ | +| | | | | | | | gle | ground-truth | | answer. | We start | from the | publicly | avail- | +| indeed | correct. | Human | annotators | systematically | | vali- | | | | | | | | | +ableexactmatchingimplementationof[22]andincludead- +| date these | aspects | for | every | single question+answer | | pair | | | | | | | | | +| ---------------- | ------- | --------------------------------- | ----- | ---------------------- | --- | ---- | -------------------------------------------- | --- | --------- | ----- | -------- | ------ | ----------- | --- | +| | | | | | | | ditionalrelaxationsfornumbercomparisons(e.g. | | | | | | numbers | | +| (Sec.4.2and4.3). | | Yet,weperformhereanadditionaluser | | | | | | | | | | | | | +| | | | | | | | over | 100 | - usually | years | - may be | off by | one, ranges | are | +studytogetannumericalestimateofqualityinthissense. +| | | | | | | | correct | if | they | partially | overlap, etc). | We | find that | exact | +| --- | --- | --- | --- | --- | --- | --- | ------- | --- | ---- | --------- | -------------- | --- | --------- | ----- | +In this study, we randomly sample 100 questions from matchjudgementsareequivalenttohumanjudgementsonly +thetestset.Thenweasksixexpertstoeachanswer50ques- +| | | | | | | | in | 68% of | the cases. | Where | they | differ, exact | match | is al- | +| --- | --- | --- | --- | --- | --- | --- | --- | ------ | ---------- | ----- | ---- | ------------- | ----- | ------ | +tionseach,giventhecorrespondingWikipediapageasref- +| | | | | | | | ways | overly | strict, | rejecting | answers | that | a human | would | +| --- | --- | --- | --- | --- | --- | --- | ---- | ------ | ------- | --------- | ------- | ---- | ------- | ----- | +erence. Theseexpertwerenotinvolvedintheoriginaldata judge as correct. This demonstrates that exact matching +collectionprocess.Thisprocessresultsinthreeanswersper +doesnotworkwellforourdataset,andjustifiesourchoice +question. If the majority of the 3 expert answers matches of BEM as the evaluation measure to check whether pre- +ourcollectedground-truthaccordingtoBEM[12],wecon- +| | | | | | | | dictedmodelanswersmatchground-truthonesinSec. | | | | | | | 5. | +| ---------- | -------- | --- | ------------- | ----- | --- | --------- | --------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | +| sider that | question | to | be answerable | given | the | Wikipedia | | | | | | | | | +page, andourground-truthto becorrect. Wefind thatthis C. CLIP retrieval in existing retrieval- +holdsformostofourquestions(86%). +augmentedVQAsystems[22,34] +| To put | this number | | in context, | we also | estimate | the an- | | | | | | | | | +| ------ | ----------- | --- | ----------- | ------- | -------- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +swerabilityofA-OKVQA[46]withrespecttotheirevalua- KAT[22]andREVIVE[34]aretworetrieval-augmented +tionmetric.A-OKVQAfollowspreviousVQAdatasetsand VQA systems which use frozen CLIP [39] embeddings to +provides10ground-truthanswersperquestion. Apredicted performretrieval. Morespecifically,theyfirstencodetheir +answer is counted as correct if it matches with 3 ground- text-only knowledge base (extracted from Wikidata) using +truth answers (exact string matching [8, 46], after normal- thelanguagetowerofCLIP.Attesttime,givenaVQAques- +izationsuchasremovingpunctuationandarticles,convert- tion,theyencodeitsimageI withCLIPandthencompare +ingnumberstodigits,etc.). Sowecanconsideraquestion itsembeddingtotheknowledgebaseembeddingstoretrieve +is answerable if there are at least 3 equivalent answers out the most similar entries. The top few most similar entries +ofthe10.Wefindthat86%oftheA-OKVQAquestionsare are then passed on to a T5 model [41] which is trained to +answerable. +| | | | | | | | producethefinalanswer. | | | | Notehowcorrectretrievaliscru- | | | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------- | --- | --- | --- | ----------------------------- | --- | --- | --- | +To conclude, this user study demonstrates that our cialtoperformwellonEncyclopedic-VQA,astheretrived +Encyclopedic-VQAdatasetisofveryhighquality. knowledge base entries should contain the answer for the +| | | | | | | | overall | system | | to succeed. | Hence, | we now | test the | CLIP- | +| --- | --- | --- | --- | --- | --- | --- | ------- | ------ | --- | ----------- | ------ | ------ | -------- | ----- | +B.QualityofBEM[12]evaluationmeasure based retrieval component of KAT [22] and REVIVE [34] +inisolation,toestimatewhethertheywouldbeabletosuc- +| We want | to | understand | how | well the | BERT | Matching | | | | | | | | | +| ------- | --- | ---------- | --- | -------- | ---- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +ceedonEncyclopedic-VQA. +(BEM)[12]evaluationmeasuremirrorshumanjudgments. WerepresentWikipediaarticlesastheCLIPembedding +| To do so | we build | on | the user | study above | and | ask an ex- | | | | | | | | | +| -------- | -------- | --- | -------- | ----------- | --- | ---------- | --- | ----- | --------- | ----------- | ------- | ------------- | --- | ------- | +| | | | | | | | of | their | title and | description | strings | concatenated. | | At test | +pert human to judge whether each answer from the user time, given a VQA question, we use CLIP to embed the +studymatchesthecollectedground-truthanswerornot.The +judgehasaccesstoboththequestionQandthecorrespond- +| ing Wikipedia | | page. | We find | that the | BEM | judgements | | | | | Recall | | | | +| ------------- | ---------- | ----- | ------- | ----------------- | --- | ------------ | --- | ------ | --- | --- | ------ | --- | --- | --- | +| equal human | judgements | | in | the vast majority | | of the cases | | | | | | | | | +| | | | | | | | | Method | | @1 | @5 | @10 | @20 | | +(96%). BEMisastricterjudgein3%ofthecases,thehu- +man in 1%. This demonstrates that BEM correlates very CLIP[39] 3.3% 7.7% 12.1% 16.5% +highlywithhumanjudgementofcorrectnessofananswer, Lens[1] 47.4% 62.5% 64.7% 65.2% +confirmingwhatreportedin[12]forotherdatasets. +Exact Match. Many VQA datasets perform exact string Table7: RecallresultsforCLIPandLens.Wereporttherecall +inretrievingtherightKBarticlewithinthetop-Kdocuments. +| matchingfortheirevaluation[8,21,37,46]. | | | | | | Becauseitis | | | | | | | | | +| --------------------------------------- | --- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | --- | --- | -T ARYAMANFINANCIALSERVICES LIMITED -SHARE TRANSFER AND INVESTOR GRIEVANCE COMMITTEE: -The Share Transfer Committee was constituted by the Board of Directors at its meeting held -on December 01, 2005. The Committee consist of Shri Munnangi V Ramana and Shri -Baliga, all being a non-executive director. The Committee meets once in a fortnight. -Compliance Officer: Shri Prasanna Chandwaskar, Company Secretary. -Number of Shareholders complaints received during the year : -During the Year under review, total no Investor complaints were received and no complaint is pending -as at November 30, 2005 for a period over 30 days. -MANAGEMENT: -a. Management Discussion and Analysis -Management Discussion and Analysis of various businesses of the Company is separately given -in the Annual Report. -b. Disclosures by management to the Board -All details relating to financial and commercial transactions where directors may have a pecuniary -interest are provided to the Board, and the interested directors neither participate in the discussion, -nor do they vote on such matters. -GENERAL BODY MEETINGS: -The details of Annual General Meetings held in the last three years are given hereunder. -Meeting Day & Date of Meeting Time Place -10* Annual Gen. Meeting Thursday, September 30,2004 11.30a.m. 401. Chirnjiv Towers. -43. Nehru Place. -New Delhi-1)0 0)9 -9th Annual General Meeting Tuesday, September 30,2003 11.30 a.m. A- 293. Shivalik. Geetanjali -Panchsheel Road. -New Delhi-1100! 7 -8ch Annual General Meeting Monday, December 30,2002 11.30a.m. B-47. Shivalik, Geetanjali -Panchsheel Road. -New Delhi-110017 -DISCLOSURE -There are no related party transactions. -Details of information on appointment of new directors -Reappointment Directorship held in other Companies Committee positions in various Companies -Name of the Director Appointment -Shri Munnangi V Ramana December 20, 2004 -Shri Suhas Ganpule May 02, 2005 +imageItoformthequeryforretrieval,proceedingasin[22, E.DatasetStatistics +| 34]. Results | in | Tab. 7 demonstrate | | that recall | is | low: the | | | | | | | | | +| ------------ | --- | ------------------ | --- | ----------- | --- | -------- | --- | -------- | ---- | ------- | ------ | -------- | -------- | ------ | +| | | | | | | | | In Table | 8 we | provide | a more | detailed | overview | of the | +correctWikipediaarticlecorrespondingtothesubjectofthe +| | | | | | | | datasetstatistics. | | | FormoredetailsseeSection4ofthemain | | | | | +| -------- | ------------ | ------ | -------------- | ---- | ---- | ------ | ------------------ | --- | --- | ---------------------------------- | --- | --- | --- | --- | +| question | is retrieved | in the | first position | only | 3.3% | of the | | | | | | | | | +paper. +| time. Even | within | the top-20 | results, | retrieval | accuracy | is | | | | | | | | | +| ----------------- | ------ | ------------------------------ | -------- | --------- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| stillonlyat16.5%. | | ThissuggestthatKAT/REVIVEwould | | | | | | | | | | | | | +F.Promptsfordatasetcreation +notworkwellonourdataset. +Lens retrieval. In Sec. 5.3 we use Google Lens to iden- AsdiscussedinSection4ofthepaper,weusedaFLAN +tify the subject C of a VQA question in our datatset, i.e. [18] version of the PaLM [17] model to (1) rephrase au- +a iNat21 or GLDv2 category, which works well (Tab. 7). tomatically generated single-hop questions (both single- +Lens is an image-based system which indexes billions of answer and multi-answer), and to (2) combine two single- +| images on | the web | and retrieves | | relevant | ones | based on | | | | | | | | | +| --------- | ------- | ------------- | --- | -------- | ---- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +hopquestionsintoatwo-hopquestion.Generally,wefound +their visual similarity. Hence, it has a greater chance to that the model tends to work better when fed with very +findawebimageresemblingaVQAtestimage,thanwhen specific(andsometimesredundant)instructions,combined +restricting the search only to Wikipedia. Attached to the withchain-of-thoughtprompts[54]wherepossible. Addi- +image is often various meta-data which enables to deter- tionally,weaddeddifficultcasesasexamplesintheprompt, +minethenameofthesubjectcategory(whichwethenuseto +| | | | | | | | to | guide | the model’s | behavior | | in more | challenging | situa- | +| --------------------------- | --- | ------------- | ----------------------- | ------- | --------- | ----- | ------ | ----- | ----------- | -------- | --- | ------- | ----------- | ------ | +| findtherightWikipediapage). | | | Recognizinglandmarksand | | | | tions. | | | | | | | | +| species from | the | natural world | are | typical | use-cases | so we | | | | | | | | | +F.1.Rephrasingautomaticallygeneratedsingle-hop +| can expect | Lens | to work well | for | recognizing | the | subjects | | | | | | | | | +| ---------- | ---- | ------------ | --- | ----------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +questions +| C inourdataset. | | However,academicresultsoniNat21[4] | | | | | | | | | | | | | +| --------------- | --- | ---------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +andGLDv2[2]suggeststhatspecializedsystemsworkvery +| | | | | | | | | We replaced | | the name | of the | category | C | in the ques- | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | --- | -------- | ------ | -------- | --- | ------------ | +wellonthesedatasets,suggestingthatagoodimage-based tionbyitssupercategory. Thisrephrasingprocesswasrun +classifierwouldmakeaviablesubstituteforLens. +forallautomaticallygeneratedsingle-hopquestions(single +Generally, the core points of our paper are that (1) En- andmulti-answer). Forexample,thequestionWhereisthe +| cyclopedic-VQA | | poses a | hard challenge | | for large | LLMs | E´glise | | | | | | | | +| -------------- | --- | ------- | -------------- | --- | --------- | ---- | ------- | ------------ | --- | -------- | --- | ------------- | --- | ----------- | +| | | | | | | | | Saint-Cannat | | located? | | was rephrased | | to Where is | +and VLMs, and (2) solving it requires augmenting the thischurchlocated? Afterseveraliterations,wedecidedto +LLM/VLM with a retrieval component to access a knowl- add three examples in the prompt, using chain-of-thought +| edgebase. | Theexactchoiceofretrievalcomponentisflex- | | | | | | | | | | | | | | +| --------- | ----------------------------------------- | --- | --- | --- | --- | --- | --------- | --- | ---- | ------------ | --- | ----- | --------- | ------ | +| | | | | | | | reasoning | | with | three steps. | The | model | generally | writes | +ibleandlikelysubjecttofurtherexploration. the three required steps in the output, and we parse the +| | | | | | | | rephrased | | question | right | after the | third | step written | by the | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | -------- | ----- | --------- | ----- | ------------ | ------ | +D.Qualitativeresults model. Sometimes, the model would return an incorrectly +| | | | | | | | rephrased | | question. | We observe | | that the | main | failure case | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | --------- | ---------- | --- | -------- | ---- | ------------ | +We present qualitative results for some of the methods is when the model simply copies the input question to the +we experimented with in Figures 6 and 7. These illustrate output. Thus, we filter out any rephrased question that is +caseswherethedifferentsetupsmaywork,andcaseswhere +| | | | | | | | identical | | to the | original question | | fed in | its input. | We sam- | +| --- | --- | --- | --- | --- | --- | --- | --------- | --- | ------ | ----------------- | --- | ------ | ---------- | ------- | +allofthemfail. Forexample,givenatextualquestionWhat pled100rephrasedquestionstomanuallyassessthequality +does this reptile eat?, PaLM can reasonably guess an an- of the rephrasings, which we found to be correct in 90% +swer without any additional context: insects – since that’s ofthecases. ThefinalpromptisgiveninPrompt1,where +whatmanyreptileseat. “$C$” and “$Q$” are placeholders for the category C and +thetextual(pre-rephrasing)questionQ. +Formorespecificinformation,relatingtomoredetailed +properties(e.g.,numberofeggsaspecificreptilelays,dates, +| | | | | | | | F.2. | Chaining | | single-hop | questions | | into | two-hop | +| --- | --- | --- | --- | --- | --- | --- | ---- | -------- | --- | ---------- | --------- | --- | ---- | ------- | +howbigaspecificfishmaybecome),largemodelsmaystill +questions +| makereasonableguesses, | | | butgenerallythisleadstoincor- | | | | | | | | | | | | +| ---------------------- | ------------------------------------ | --- | ----------------------------- | --- | --- | --- | --- | ---------- | --- | ----------------- | --- | ----------- | --- | ----------- | +| rectanswers. | Augmentinglargemodelswithcontextfrom | | | | | | | | | | | | | | +| | | | | | | | | We created | | two-hop questions | | by chaining | | two single- | +retrieved knowledge improves the accuracy in these cases hopquestions,wheretheanswertothefirstsingle-hopques- +substantially,leadingtopreciseanswersthatareattributable tionservesasabridgeentitythatmakesaconnectiontothe +tothepieceofknowledgethatwasretrieved. +| | | | | | | | second | single-hop | | question. | For | example, | given | the first | +| --- | --- | --- | --- | --- | --- | --- | ------ | ---------- | --- | --------- | --- | -------- | ----- | --------- | +Finally, we also see cases where the generated answer single-hopquestion(SQ1)Whatisthemaincompetitorfor +is incorrect, even if the correct piece of knowledge is re- food for this animal? with answer (SA1) Spotted hyena, +trieved. Thisindicatesthatinsomecaseslargemodelsmay andthesecondsingle-hopquestion(SQ2)Whatisthepop- +stillhavedifficultiesusingretrievedknowledgetogenerate ulation size of this animal? with answer (SA2) Between +accurateanswers. 27,000 and 47,000 individuals, we generated What is the -F= ARYAMAN FINANCIAL SERVICES LIMITED -Details of non-compliance -There was no non compliance by the Company, nor there was any penalties or strictures -imposed on the company by the Stock Exchange or SEBI or any authority on any matter -related to capital market during the last 3 years. -MEANS OF COMMUNICATION -The quarterly results are not published in the News papers -Management Discussion and Analysis form part of this Annual Report. -GENERAL SHAREHOLDER INFORMATION -AGM will be held on Thursday, December 29, 2005 at 10.00 a.m. at the registered office. -Financial Year of the Company : April 01, 2004 to June 30, 2005 -Date of Book Closure: December 26, 2005 to December 29, 2005 (Both days inclusive) -Dividend payment date: Not applicable. -Listing on Stock Exchanges: Mumbai Stock Exchange Limited, Delhi Stock Exchange and -Ahmedabad Stock Exchange Limited. -| Stock Code: | Bombay Stock Exchange Limited | | : 530245 | | -| ------------ | ------------------------------ | --- | -------- | --- | -Market Price Data: High & Low price of equity shares on Stock Exchange shares of the Company -are infrequently traded on the exchanges. -| Month | | High Amount (Rs.) | Low Amount (Rs.) | | -| ----------- | --- | ------------------ | ---------------- | ---- | -| April-2004 | | 1.21 | | 0.82 | -May-2004 -| June-2004 | | 0.80 | | 0.69 | -| ---------- | --- | ----- | --- | ---- | -July-2004 -| August - 2004 | | 0.69 | | 0.63 | -| ----------------- | --- | ----- | --- | ---- | -| September - 2004 | | 0.82 | | 0.75 | -| October - 2004 | | 1.70 | | 0.90 | -| November - 2004 | | 2.02 | | 2.01 | -| December - 2004 | | 2.69 | | 1.93 | -| January - 2005 | | 4.60 | | 2.76 | -| February - 2005 | | 5.50 | | 2.10 | -| March-2005 | | 3.10 | | 2.00 | -| April-2005 | | 3.96 | | 2.01 | -| May-2005 | | 7.32 | | 3.15 | -| June-2005 | | 6.90 | | 4.06 | -Performance in comparison to broad-based indices i.e. BSE Sensex is as under: +Qualitative results: PaLI +| | In which direction from the | What is Mary handling | | | | | | | +| --- | ---------------------------- | ---------------------- | --- | --- | --- | --- | --- | --- | +In what month of 1944 was this What was this church fitted with +| Question | german-czech border is this | St.Dominic in the facade of this | | | | | | | +| -------- | ---------------------------- | --------------------------------- | --------- | --- | ----------------- | --- | -------- | --- | +| | | | | | church destroyed? | | in 1914? | | +| | mountain located? | | building? | | | | | | +No +| | north | | child | | december | | window | | +| --- | ----- | --- | ----- | --- | -------- | --- | ------ | --- | +Retrieval +Lens +| | north | | rosary | | december | | bell | | +| --- | ----- | --- | ------ | --- | -------- | --- | ---- | --- | +Section +Oracle +| | north | | rosary | | may | | steeple | | +| --- | ----- | --- | ------ | --- | --- | --- | ------- | --- | +Section +Ground +| | north | | the rosary | | may | | electricity | | +| --- | ----- | --- | ---------- | --- | --- | --- | ----------- | --- | +Truth +Figure 6: +PaLIqualitativeresults. Wepresent4multi-modalquestionsandtheanswersproducedwithdifferentexperimentalsetups, +usingthePaLImodel: withoutretrieval(“NoRetrieval”), Lens-basedretrievalofKBsection(“LensSection”), OracleretrievalofKB +secQtionu(“aOrlaictleaSetcitvione”) .rAeddsitiounalltlys,w:e Fprolvaidenth-ePgroaunLd-tMruthanswerinthelastrow.Weshowcaseswhereeachofthesetupscan +producethecorrectanswer,aswellasanexamplewhereallmethodsfail(theright-mostone). +| | | How many eggs does this | | | | How big does this fish typically | | | +| --- | --- | ------------------------ | --- | --- | --- | --------------------------------- | --- | --- | +Question What does this reptile eat? What does this animal eat? +| | | reptile typically lay? | | | | | become? | | +| --- | --- | ---------------------- | --- | --- | --- | --- | ------- | --- | +No +| | insects | | 10-15 | | insects | | 6 cm | | +| --- | ------- | --- | ----- | --- | ------- | --- | ---- | --- | +Retrieval +Lens +| | insects | | 3-6 | | plankton | | 13 cm | | +| --- | ------- | --- | --- | --- | -------- | --- | ----- | --- | +Section +| Oracle | insects | | 3-6 | | mussels | | 13 cm | | +| ------- | ------- | --- | --- | --- | ------- | --- | ----- | --- | +Section +Ground +| | insects | | 3-6 | | mussels | | 25 cm | | +| --- | ------- | --- | --- | --- | ------- | --- | ----- | --- | +Truth +Figure7: PaLMqualitativeresults. Wepresent4multi-modalquestionsandtheanswersproducedwithdifferentexperimentalsetups, +usingthePaLMmodel: withoutretrieval(“NoRetrieval”),Lens-basedretrievalofKBsection(“LensSection”),OracleretrievalofKB +section(“OracleSection”).Additionally,weprovidetheground-truthanswerinthelastrow.Weshowcaseswhereeachofthesetupscan +producethecorrectanswer,aswellasanexamplewhereallmethodsfail(theright-mostone). +| population | size of the main competitor | for food | of this an- | individuals. | | | | | +| ---------- | --------------------------- | -------- | ----------- | ------------ | --- | --- | --- | --- | +imal? Inthisexample,theentityspottedhyenaservesasa +| | | | | After | several iterations, | we crafted | a prompt | with 4 ex- | +| --------------------------------------- | --- | --- | ----------- | ----- | ------------------- | ---------- | -------- | ---------- | +| bridgebetweenthetwosingle-hopquestions. | | | Notethatthe | | | | | | +answertothetwo-hopquestionisidenticaltotheanswerto amplesofvaryingdifficulty(Prompt2). Theoutputofthe +| | | | | model | is taken as the chained | two-hop | question. | Some- | +| --- | --- | --- | --- | ----- | ----------------------- | ------- | --------- | ----- | +thesecondsingle-hopquestion,Between27,000and47,000 +times,though,themodelwouldreturnanincorrecttwo-hop -| r ARYAMANFINANCIALSERVICES LIMITED | | | = | -| ----------------------------------- | ---------- | ----------------------------- | ---- | -| Month | BSE Index | Closing price of Shares (Rs.) | | -| April - 2004 | 5655.09 | | 0.82 | -| May - 2004 | 4759.62 | | | -| June - 2004 | 4795.46 | | 0.69 | -| July-2004 | 5170.32 | | | -| August-2004 | 5192.08 | | 0.69 | -| September - 2004 | 5583.61 | | 0.82 | -| October - 2004 | 5672.27 | | 0.69 | -| November - 2004 | 6234.29 | | 2.02 | -| December - 2004 | 6602.69 | | 2.31 | -| January-2005 | 6555.94 | | 3.83 | -| February - 2005 | 6713.86 | | 2.75 | -| March - 2005 | 6492.82 | | 2.00 | -| April-2005 | 6154.44 | | 3.79 | -| May-2005 | 6715.11 | | 6.68 | -| June-2005 | 7193.85 | | 4.06 | -Registrar and Transfer Agent: The Company has appointed Registrar and Transfer agent to handle the -share transfer work and to solve the complaints of Shareholders. Name, Address and telephone number -of Registrar and Transfer Agent is given hereunder: -M/s. Adroit Corporate Services Pvt Ltd. -Unit No.3 & 4, Near Maharashtra Weigh Bridge, -Link Road, Saki Naka, Andheri (E), -Mumbai-400072, -Tele. No.: +91-22 - 288590942 / 28503748 / 26924437 -Fax No. : +91 -22-26924438 -Share Transfer System: -Shares sent for transfer in physical form are registered by the Company's Registrars and Share -Transfer Agents within 15 days from the receipt of the documents, if the same are found in order. -Shares under objection are returned within two weeks. +NumberofQ+Apairs NumberofCategories %UniqueC Total(I,Q,A)triplets +| | | | Train | Val | Test | Train | Val | Test | Val | Test | Train | Val | Test | | +| --- | --- | --- | ----- | --- | ---- | ----- | --- | ---- | --- | ---- | ----- | --- | ---- | --- | +Templated 8,133 200 500 1,983 114 300 26% 21% 40,665 1,000 500 +12taNi Automatic 96,317 1,000 1,500 4,061 343 562 5% 6% 481,585 5,000 1,500 +MultiAnswer 11,262 200 500 3,585 118 318 31% 22% 56,310 1,000 500 +| | | TwoHop | 10,819 | 200 | 500 | 2,499 | 131 | 359 | 11% | 6% | 54,095 | 1,000 | 500 | | +| --- | --- | --------- | ------ | --- | --- | ----- | --- | --- | --- | --- | ------ | ----- | --- | --- | +| | | Templated | 5,795 | 200 | 500 | 1,808 | 135 | 268 | 27% | 28% | 25,870 | 827 | 500 | | +2vDLG +Automatic 57,124 750 1,250 1,965 138 208 9% 9% 255,529 3,025 1,250 +MultiAnswer 12,667 200 500 6,449 151 339 45% 56% 56,426 844 500 +| | | TwoHop | 10,221 | 200 | 500 | 1,405 | 99 | 154 | 11% | 11% | 45,771 | 895 | 500 | | +| --- | --- | ------ | ------ | --- | --- | ----- | --- | --- | --- | --- | ------ | --- | --- | --- | +Total 212,338 2,950 5,750 16,249 1,071 2,152 14% 17% 1,016,251 13,591 5,750 +Table8: DatasetStatistics. Wereportthenumberofquestion,answer,categoriesandtripletsforbothsupportingdatasetsanddifferent +questiontypes.Intotalourdatasetcontains1MVQAtriplets,makingitthelargestofitskind. +question. Forthisreason, wedesignedasecondpromptto evantsectionswithPaLM.Whennotavailable,werevertto +| validatethetwo-hopquestionprovidedastheinitialoutput | | | | | | | | usingPrompt4. | | | | | | | +| ---------------------------------------------------- | --- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | +(Prompt3).Themodelisaskedtoanswerthetwo-hopques- +G.2.Promptforrelevantsectionidentification +tionusingthetwoinitialsingle-hopquestionswithanswers +ascontext.Ifthepredictedanswerisidenticaltotheanswer +| | | | | | | | | We use | Prompt | 8 to | identify | relevant | sections | in a | +| --- | --- | --- | --- | --- | --- | --- | --- | ------ | ------ | ---- | -------- | -------- | -------- | ---- | +of the second single-hop question, then the two-hop ques- Wikipedia article retrieved by Lens. To do so, we query +tionisvalidatedandkeptinourdataset;otherwise,itisdis- +PaLMforeachsection$S$intheLensretrievedWikipedia +carded. Finally,wefiltersomecommonfailurecases,such article for a question $Q$. Note that we use a few exam- +aswhenthemodeloutputstheexactfirstsingle-hoporsec- plesintheprompttoconditionPaLMtogenerateayes/no +| ond single-hop | | question | as the | chained | two-hop | question. | | | | | | | | | +| -------------- | --- | -------- | ------ | ------- | ------- | --------- | --- | ------- | --------------------------------------- | --- | --- | --- | --- | --- | +| | | | | | | | | answer. | TheanswerproducedbyPaLMisconvertedintoa | | | | | | +We sampled 100 chained two-hop questions to manually stringandmatchedtoeitheryesorno. Whennomatchis +| assess their | quality, | which | we | found | to be correct | in 88% | | | | | | | | | +| ------------ | -------- | ----- | --- | ----- | ------------- | ------ | --- | --------- | ------ | ------------ | ------ | -------- | --- | --------- | +| | | | | | | | | found, we | assume | that section | is not | relevant | to | the ques- | +ofthecases. Inthefinalprompts(Prompts2-3)“$SQ1$”/ tion. Ifmorethanonesectionisidentifiedasrelevantfora +“$SA1$”areplaceholdersforthefirstsingle-hopquestion/ question, the input $S$ to Prompt 7 becomes the concate- +answer,“$SQ2$”/“$SA2$”areplaceholdersforthesecond +nationofallrelevantsections. +single-hopquestion/answer,and“$Q$”istheplaceholder +forthegeneratedtwo-hopquestion. +G.Promptsforevaluation +InSection5weanalyzetheperformanceoflargemodels +| inourdataset. | Similarlytothedatasetcreationprocess,we | | | | | | | | | | | | | | +| ------------- | --------------------------------------- | --- | ------------------------- | ----------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | +| use diverse | prompts | to | adapt | and improve | the | behavior | of | | | | | | | | +| thesemodels. | Inparticular, | | weusepromptstoincorporate | | | | | | | | | | | | +retrievalresultsandtoidentifyrelevantsectionsforarticles +retrievedwithLens. +G.1.Promptsforquestionanswering +| We use | textual | Prompts | 4-7 | to produce | | answers with | | | | | | | | | +| -------------- | -------- | ------------- | -------- | ------------ | ------------ | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| large models. | Note | that | we | use the | same prompts | when | | | | | | | | | +| using PaLI | or PaLM. | | However, | we always | | use the ques- | | | | | | | | | +| tion image | I as | an additional | | input to | the model | when us- | | | | | | | | | +| ing PaLI. | Note | that we | use $Q$ | to refer | to the | textual part | | | | | | | | | +| of a question, | $C$ | to denote | | the question | subject | C name, | | | | | | | | | +$Art$todenotethefulltextofaWikipediaarticleand$S$ +| todenoteasectionofaWikipediaarticle. | | | | | Weusethesame | | | | | | | | | | +| --------------------------------------- | --- | --- | ------------ | ----- | ------------ | ------------- | --- | --- | --- | --- | --- | --- | --- | --- | +| retrievalpromptsforLensandOraclesetups. | | | | | | However,for | | | | | | | | | +| Lens experiments, | | we | only include | $Art$ | or | $S$ if avail- | | | | | | | | | +able,aswemightnotretrieveentitieswithLensorfindrel- -Distribution of SharehoHingas on November 30, 2005 -| | | | No. of shares | | % of Shares | -| -------------- | ------------- | ---------- | -------------- | --- | ----------- | -| No. of Equity | No. of share | % of Share | | | | -| Shares held | holders | holders | held | | held | -| Upto - 500 | 1456 | 83.06 | 246601 | | 5.33 | -| 501 - 1,000 | 129 | 7.36 | 105951 | | 2.29 | -| 1,001 -2,000 | 58 | 3.31 | 86550 | | 1.87 | -| 2,001 -3,000 | 32 | 1.83 | 79743 | | 1 .72 | -| 3,001 - 4,000 | 13 | 0.74 | 46550 | | 1.01 | -| 4,001 - 5,000 | 11 | 0.63 | 51229 | | 1.11 | -| 5,001 - 10,000 | 23 | 1.31 | 163419 | | 3.53 | -10,001 - above -| | 31 | 1.77 | 3849627 | | 83.15 | -| ----- | ---- | ------ | -------- | --- | ------ | -| Total | r*3 | 100.00 | 4629670 | | 100.00 | -Dematerialisation of shares and liquidity: 13.60% have been dematerialised. -Outstanding GDR's/ADR's/Convertible- Nil. -Particular of shares held on physical / Electronic form as on November 30, 2005. -| Category | Holders | No. of Shares | | % of sharers | | -| ---------- | -------- | -------------- | --- | ------------ | --- | -| Paper mode | 1214 | 3999970 | | 86.40 | | -| NSDL | | | | 9.23 | | -| | 422 | 427307 | | | | -| CDSL | 117 | 202393 | | 4.37 | | -| Total | 1753 | 4629670 | | 100 | | -Address for correspondence: -| Aryaman Financial | Services Limited | | | | | -| ----------------- | ---------------- | --- | --- | --- | --- | -401, Chiranjiv Towers, -43, Nehru Place, New Delhi 110 019 +| Prompt1: | Rephrasingautomatically-generatedsinglehopquestion | | | | | | | +| -------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | +In this task, please rephrase the question by replacing the entity name by the word "this", +| followed | by the type | of the | entity. | See the | examples | below: | | +| -------- | ----------- | ------ | ------- | ------- | -------- | ------ | --- | +EXAMPLE 1: +| entity | name: eiffel | tower | | | | | | +| --------- | ------------- | -------- | ------ | ------------ | --- | ------ | --------- | +| question: | How tall | is the | eiffel | tower? | | | | +| step 1 | (find type of | entity): | The | eiffel tower | is | a type | of: tower | +step 2 (write the word "this" followed by the type obtained in "step 1"): this tower +| step 3 | (final rephrased | question): | | How tall | is this | tower? | | +| ------ | ---------------- | ---------- | --- | -------- | ------- | ------ | --- | +EXAMPLE 2: +| entity | name: salmon | | | | | | | +| --------- | ------------- | -------- | ----------- | --------- | ------ | ------- | ---- | +| question: | Which country | is | the largest | producer | of | salmon? | | +| step 1 | (find type of | entity): | The | salmon is | a type | of: | fish | +step 2 (write the word "this" followed by the type obtained in "step 1"): this fish +step 3 (final rephrased question): Which country is the largest producer of this fish? +EXAMPLE 3: +| entity | name: Grand | Beach Provincial | | Park | | | | +| ------ | ----------- | ---------------- | --- | ---- | --- | --- | --- | +question: grand beach provincial park is located on the east side of what lake? +step 1 (find type of entity): The Grand Beach Provincial Park is a type of: park +step 2 (write the word "this" followed by the type obtained in "step 1"): this park +step 3 (final rephrased question): this park is located on the east side of what lake? +Note that the entity name should not be part of the rephrased question. +| Please | make sure to | write out | all | 3 steps as | in the | examples | above. | +| ------ | ------------ | --------- | --- | ---------- | ------ | -------- | ------ | +Based on the above examples, provide a rephrased question for the following case: +| entity | name: $C$ | | | | | | | +| --------- | --------- | --- | --- | --- | --- | --- | --- | +| question: | $Q$ | | | | | | | -ARYAMAN FINANCIAL SERVICES LIMITED -BALANCE SHEET AS AT 30th JUNE - 2005 -| | Schedule | | 30th JUNE, 2005 | | 31st MARCH, 2004 | -| --- | -------- | --- | --------------- | --- | ---------------- | -(Rs) -(Rs.) -SOURCES OF FUNDS -Shareholders' Funds -| a)Share Capital | 1 | 53,619,074 | | 53.619,074 | | -| -------------------- | --- | ----------- | ---------- | ----------- | ---------- | -| b)Reserves & Surplus | 2 | (3.296.920) | | (2,041.813) | | -| | | | 50,320,154 | | 51,577,261 | -Loan Funds -SECURED LOANS -| Deferred Tax Liability (Net) | | | 8,746,000 | | 8,746,000 | -| ---------------------------- | --- | --- | ---------- | --- | ---------- | -| | | | 59,066,154 | | 60,323,261 | -APPLICATION OF FUNDS -FIXED ASSETS -| Gross Block | | 33,129,228 | | 33,129,228 | | -| ----------------- | --- | ---------- | ---------- | ---------- | ---------- | -| Less:Depreciation | | 4.105.964 | | 3,602.375 | | -| Net Block | | | 29,023,264 | | 29,526,853 | -INVESTMENTS -| CURRENT ASSETS.LOANS AND ADVANCES | 4 | 38,829,511 | | 38,376,313 | | -| ---------------------------------- | --- | ---------- | --- | ---------- | --- | -LESS:CURRENT LIABILITIES & PROVISIONS 5 8,786,621 7,781,752 -| NET CURRENT ASSETS | | | 30,042,890 | | 30,594,561 | -| -------------------------- | --- | --- | ---------- | --- | ---------- | -| MISCELLANEOUS EXPENDITURE | 6 | | | | 201,847 | -(To the extent not written-off or adjusted) 59,066,154 60,323,261 -| Notes on Accounts | 12 | | | | | -| ----------------- | --- | --- | --- | --- | --- | -The Schdules referred to above form an -integral part of the Balance Sheet -As per our report of even date -for Thakur Vaidyanath & Aiyar & Co. -for and On behalf of the Board -Chartered Accountatnts -| C.V. Parameswar | | | | DIRECTOR | DIRECTOR | -| --------------- | --- | --- | --- | -------- | -------- | -Partner -M.No.11541 -Place: Mumbai -Dated: 01/12/2005 +| Prompt2: | Chainingtwosingle-hopquestionsintoatwo-hopquestion | | | | | | | +| -------- | -------------------------------------------------- | --- | --- | --- | --- | --- | --- | +EXAMPLE 1: +| question | 1: in which | city | is this | building | located? | | | +| -------- | ---------------- | ------ | ------- | ----------- | -------- | ---------- | --- | +| answer | 1: San Francisco | | | | | | | +| question | 2: what | is the | average | temperature | in San | Francisco? | | +| answer | 2: 15 Celsius | | | | | | | +combined question: What is the average temperature in the city where this building is located? +EXAMPLE 2: +| question | 1: what | is the | predator | of this | animal? | | | +| -------- | ---------------- | ------ | --------- | ------- | ----------- | --- | --- | +| answer | 1: Lion | | | | | | | +| question | 2: What | is the | weight of | a lion | on average? | | | +| answer | 2: 190 kilograms | | | | | | | +combined question: What is the average weight of the predator of this animal? +EXAMPLE 3: +| question | 1: In what | country | is this | plant | found? | | | +| -------- | ------------- | ---------------- | -------- | ----- | --------- | ------------ | --- | +| answer | 1: Australia | | | | | | | +| question | 2: What | does australia’s | | size | give it a | wide variety | of? | +| answer | 2: landscapes | and | climates | | | | | +combined question: What does the size of the country where this plant is found give it a wide +variety of? +EXAMPLE 4: +| question | 1: What | country | is this | plant | the national | flower | of? | +| -------- | --------------- | ------- | ------- | ----- | ------------ | ------ | --- | +| answer | 1: South Africa | | | | | | | +question 2: south africa is a member of the commonwealth of nations and what other +organization? +| answer | 2: the G20 | | | | | | | +| ------ | ---------- | --- | --- | --- | --- | --- | --- | +combined question: The country that this plant is the national flower of is a member of the +| commonwealth | of nations | and | what | other | organization? | | | +| ------------ | ---------- | --- | ---- | ----- | ------------- | --- | --- | +Based on the above 4 examples, provide a combined question for the following case, such that +the answer to the combined question is the same as the answer to question 2: +| question | 1: $SQ1$ | | | | | | | +| -------- | ------------------------------------------------------------ | --- | --- | --- | --- | --- | --- | +| answer | 1: $SA1$ | | | | | | | +| question | 2: $SQ2$ | | | | | | | +| answer | 2: $SA2$ | | | | | | | +| combined | question: | | | | | | | +| Prompt3: | Validatingtwo-hopquestiongiventheoriginalsingle-hopquestions | | | | | | | +| question | 1: $SQ1$ | | | | | | | +| answer | 1: $SA1$ | | | | | | | +| question | 2: $SQ2$ | | | | | | | +| answer | 2: $SA2$ | | | | | | | +Based on the questions and answers above, please answer the following question: $Q$ -| IF ARYAMANFINANCIALSERVICES LIMITED | | | | | | • | -| ------------------------------------- | --- | --- | --- | --- | --- | --- | -PROFIT AND LOSS ACCOUNT FOR THE YEAR ENDED 30th JUNE.2Q05 -| | Schedule | | 30th JUNE, 2005 | | | 31st MARCH, 2004 | -| --- | -------- | --- | --------------- | --- | --- | ----------------- | -| | (Rs.) | | | | | (Rs.) | -INCOME -| Merchant Banking | | 2,760,000 | | | 4,200,295 | | -| ---------------------------- | --- | --------- | ----- | --- | --------- | --- | -| Other Income from Operation | 7 | | 7,238 | | 3,466,748 | | -| Project Consultancy | | | | | 225,000 | | -128,647 -Proft / Loss on Sale of Shares -Proft / Loss on Sale of Invetments -| Prior Period Adjustment | | | | | 123,291 | | -| ----------------------- | --- | --- | --- | --------- | ------- | --------- | -| | | | | 2,767,238 | | 8,143,981 | -EXPENDITURE -| Personnel expenses | 8 | | 204,459 | | 631,341 | | -| -------------------------------------- | --- | --------- | ------- | ----------- | --------- | ----------- | -| Administation and other Expenses | 9 | 1,239,747 | | | 2,139,310 | | -| Finance Charges | 10 | | 29,864 | | 1,823,941 | | -| Legal and Consultancy charges | 11 | 1,778,215 | | | 1,660,499 | | -| Auditors' Remuneration | | | 66,250 | | 55,000 | | -| Miscellaneous Expenditure written off | | 201,847 | | | 244,276 | | -| Prior Period Adjustment | | | 374 | | | | -| Proft / Loss on Sale of Asset | | | | | 805,776 | | -| Bad Debts written off | | | | | 1,313,316 | | -| | | | | 3,520,756 | | 8.673.456 | -| Profit before Depreciation | | | | (753,518) | | (529,477) | -| Depreciation | | | | 503,589 | | 648.453 | -| Profitless) after Depreciation | | | | (1,257,107) | | (1,177,930) | -Provision for Deferred Tax -| Profit/(loss) after Tax | | | | (1,257,107) | | (1,177,930) | -| ------------------------ | --- | --- | --- | ------------ | --- | ------------ | -| Balance as per Last year | | | | (28,141,813) | | (26,963,883) | -| | | | | (29,398,920) | | (28,141,813) | -in -APPROPRIATIONS -Profit/(loss) after appropriations carried to Balance Sheet (28,141,813) -| Notes on Accounts | 12 | | | | | | -| ------------------ | --- | --- | --- | --- | --- | --- | -The Schdules referred to above form an -As per our report of even date -for Thakur Vaidyanath & Aiyar & Co. For and on behalf of the Board -Chartered Accountatnts -C.V. Parameswar -| | | | | DIRECTOR | DIRECTOR | | -| --- | --- | --- | --- | --------- | -------- | --- | -Partner -M.No.11541 -Place: Mumbai -Dated: 01/12/2005 - -7= ARYAMANFINANCIALSERVICES LIMITED -SCHEDULES TO BALANCE SHEET -| | 30th JUNE, 2005 | 31st MARCH, 2004 | | -| --- | --------------- | ---------------- | ----- | -| | (Rs.) | | (Rs.) | -SCHEDULE 1 -SHARE CAPITAL -AUTHORISED -| 50,00,000 Equity Shares of Rs.10/- each | | 50000000 | 50000000 | -| --------------------------------------- | --- | -------- | -------- | -ISSUED.SUBSCRIBED AND PAID-UP -46,29,670 Equity Shares of Rs. 10.00 each 46,296,700 46,296,700 -| Deposit againstlssue of Shares | 7,322,374 | | 7,322,374 | -| ------------------------------ | --------- | ---------- | ---------- | -| | | 53,619,074 | 53,619,074 | -SCHEDULE 2 -RESERVES AND SURPLUS -a) GENERAL RESERVE -| | 26,100,000 | | 26,100,000 | -| --- | ---------- | --- | ---------- | -Balance as per Last Year -| | 26,100,000 | | 26,100,000 | -| ------------------------- | ------------ | ----------- | ------------ | -| b) Proft and Loss account | (29,398,920) | | (28,141,813) | -| | | (3,298,920) | (2,041,813) | -SCHEDULE 4 -CURRENT ASSETS.LOANS AND ADVANCES -RENT ASSETS -Interest Accrued but not Due -Sundry Debtors: -(unsecured,considered good) --Debts outstanding for a period exceeding 6 months 16,877,836 16,945,245 -| -Others | 1,576,863 | | 122,865 | -| ------- | ---------- | ---------- | ---------- | -| | 18,454,699 | | 17,068,110 | -| | | 18,454,699 | 17,068,110 | -CASH AND BANK BALANCE: -| -Cash Balance on Hand | | 20,119 | 507,832 | -| --------------------- | --- | ------ | ------- | --Bank Balance: -| -With Scheduled Banks on Current Account | | 22,706 | 116,047 | -| ---------------------------------------- | --- | ------ | ------- | --With Scheduled Banks on Deposit Account -| | | 18,497,524 | 17,691,988 | -| --- | --- | ---------- | ---------- | -LOANS AND ADVANCES -(Unsecured,considered good) -| Shares Held Against Portfolio | 32,681,045 | | 32,681,045 | -| ----------------------------- | ---------- | ----------- | ---------- | -| Portfolio Advances | 33,682,872 | | 33,546,366 | -| | | (1,001,827) | (865,321) | -Shares Held against Portfolio Commitments 19,414,282 19,414,282 -Advances recoverable in cash or in kind or value to be -| received | | 1,141,872 | 1,351,204 | -| ---------------- | --- | ---------- | ---------- | -| Security Deposit | | 777,660 | 784,160 | -| | | 20,331,987 | 20,684,325 | -| | | 38,829,511 | 38,376,313 | -SCHEDULE S -CURRENT LIABILITIES AND PROVISIONS -alCURRENT LIABILITIES -| 1. Sundry Creditors | 693,403 | | 562,128 | -| ---------------------------- | --------- | --------- | --------- | -| 2. Other Current Liabilities | 8,093,218 | | 7,219,624 | -| | | 8,786,621 | 7,781,752 | -blPROVISIONS -| | | 8,786,621 | 7,781,752 | -| --- | --- | --------- | --------- | -SCHEDULES -MISCELLANEOUS EXPENDITURE -extent not written off or adjusted) -| a)Public Issue Expenses | | | 188,152 | -| ----------------------- | --- | --- | ------- | -b)Preliminary and Pre-operative Expenses -13,695 -201,847 - -SCHEDULE 3 -FIX ASSETS -| DESCRIPTION | | GROSS BLOCK | | | | | DEPRECIATION | | | | -| ----------- | --- | ----------- | --- | --- | --- | --- | ------------ | --- | --- | --- | -NET PL^^K -| | | | | | | | | | ASON31ST | ASON31ST | -| --- | --------- | ---------- | ------- | ---------- | -------- | --------- | --------------- | -------- | --------- | -------- | -| | ASON31ST | ADDITIONS | | DELETIONS | ASON3OTH | ASON31ST | FOR SALESWDJ. | ASON31ST | | | -| | | | | | | | | | MARCH | MARCH | -| | MARCH | | DURING | DURING | JUNE | MARCH | THE DURING | MARCH | | | -2004 THEYEAR THEYEAR 2006 2004 YEAR THEYEAR 2006 2006 2006 -A) OWN USE -LAND-FREEHOLD 8,342,965 - - 8,342,965 - 8,342,965 8,342,965 -BUILDING 22,166,276 - - 22,166,276 3,096,043 451,390 - 3,547,433 18,618,843 19,070,233 -FIXTURE 627,420 - - 627,420 409,355 34,999 - 444,354 183,066 218,065 -VEHICLES -AIRCONDmONERS 235,050 - - 235,050 92,457 13,948 - 106,405 128,645 142,593 -OFFICE EQUIPMENT 6,341 - - 6,341 261 376 - 637 5,704 6,080 -COMPUTERS 18,150 - - 18,150 4,259 2,876 - 7,135 11,015 13,891 -B) LEASED -| PLANT AND MACHINERY | | | | | | . | | | - | | -| ------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| LEASE ADJUSTMENT ACCOUNT | | - | | | | | | | | | -C) CAPITAL WORK-IN-PROGRESS -j -| BUILDING AND EQUIPMENT | 72,500 | | - | - | 72,500 | | | | 72,500 | 72,500 | -| ---------------------- | ------- | --- | --- | --- | ------ | --- | --- | --- | ------- | ------ | -! ADVANCE FOR COCHIN STOCK -. -EXCHW<£ CORPORATE MEMBERSHIP 1,660,526 - - 1,660,526 1,660,526 1,660,526 -TOTAL 33,129,228 - - 33,129,228 3,602,375 503,589 - 4,105,964 29,023,264 29,526,853 -PREVIOUS YEAR 39,881,271 25,436 6,777,479 33,129,228 5,232,890 648,453 2,278,968 3,602,375 29,526,853 34,648,381 - -fj= ARYAMANFINANCIALSERVICES LIMITED - -^,1 -SCHEDULES TO PROFIT AND LOSS ACCOUNT -| ACCOUNT NAME | | 30th JUNE, 2005 | | 31st MARCH, 2004 | | -| ------------- | --- | --------------- | ----- | ---------------- | ----- | -| | | | (Rs.) | | (Rs.) | -SCHEDULE 7 -OTHER INCOME FROM OPERATIONS -| INTEREST | | | - | | 2,248 | -| ----------------------------------------- | ------ | --- | ----- | --- | --------- | -| LEGAL, PROFESSIONAL & CONSULTANCY CHARGES | | | _ | | 2,817,500 | -| POJECT CONSULTANCY CHARGES | | | - | | | -| MISCELLANEOUS | INCOME | | 7,238 | | | -| COMMISSION RECIVED | | | - | | 647,000 | -| | | | 7,238 | | 3,466,748 | -SCHEDULE 8 -PERSONNEL EXPENSES -| SALARY AND ALLOWANCES | | | 165,000 | | 547,000 | -| --------------------- | --- | --- | ------- | --- | ------- | -| | | | 39,459 | | 84,341 | -STAFF WELFARE -| | | | 204,459 | | 631,341 | -| --- | --- | --- | ------- | --- | ------- | -SCHEDULE 9 -ADMINISTRATIVE AND OTHER EXPENSES -| TELEPHONE CHARGES | | | 222,113 | | 548,578 | -| ------------------------------ | --- | --- | ------- | --- | ------- | -| COURIER CHARGES | | | 8,652 | | 25,502 | -| POSTAGE AND TELEGRAM | | | 20 | | 1,311 | -| RENT.RATES AND TAXES | | | 125,125 | | 294,500 | -| AIRS AND MAINTENANCE-BUILDING | | | 712 | | 3,018 | -| REPAIRS AND MAINTENANCE-OTHERS | | | 192,937 | | 240,738 | -| BOOKS AND PERIODICALS | | | | | 3,541 | -975 -| OFFICE EXPENSES | | | 191,306 | | 256,233 | -| ----------------------- | --- | --- | ------- | --- | ------- | -| BUSINESS PROMOTION | | | 55,340 | | 76,527 | -| 1 EGAL EXPENSES | | | - | | | -| ELECTRICITY CHARGES | | | 41,763 | | 125,089 | -| PRINTING AND STATIONERY | | | 27,892 | | 76,898 | -| TRAVELLING EXPENSES | | | 145,433 | | 333,007 | -| INSURANCE | | | 3,070 | | 9,723 | -| DEMAT CHARGES | | | 34,881 | | 33,623 | -| LISTING CHARGES | | | 15,000 | | 7,500 | -| OTHERS EXPENSES | | | | | 103,523 | -174,528 -| | | | 1,239,747 | | 2,139,310 | -| --- | --- | --- | --------- | --- | --------- | -SCHEDULE 10 -FINANCE CHARGES -| BANK CHARGES | | | 7,490 | | 21,752 | -| --------------------- | --- | --- | ------ | --- | --------- | -| LEASE RENT | | | 22,374 | | 134,244 | -| INTEREST | | | - | | 92,945 | -| OTHER FINANCE CHARGES | | | - | | 1,575,000 | -1,823,941 -29,864 -SCHEDULE 11 -LEGAL AMD CONSULTANCY CHARGES -| LEGAL.PROFESSIONAL AND CONSULTANCY | | | | | 1,572,975 | -| ---------------------------------- | --- | --- | --- | --- | --------- | -1,672,805 -| MEMBERSHIP CHARGES | | | 104,160 | | 81,944 | -| --------------------- | --- | --- | --------- | --- | --------- | -| DIRECTOR SITTING FEES | | | - | | - | -| STAMP PAPERS | | | 1,250 | | 5,580 | -| INCOME TAX PAID | | | - | | | -| MARKETING FEES | | | - | | | -| | | | 1,778,215 | | 1,660,499 | - -IT ARYAMANFINANCIALSERVICES LIMITED -SCHEDULE 14 -NOTES TO ACCOUNTS -SIGNIFICANT ACCOUNTING POLICIES: - -Accounts are prepared on Historical Cost Convention accruing for incomes, expenditures, assets -and liabilities. -a. Income from Project Consultancy is accrued considering stage wise completion of work, wherever -agreed upon or feasible, otherwise it is accrued on percentage basis based on the quantum of work -completed. -b. Income from Merchant Banking is accrued on percentage basis based on quantum of work completed. -a. FIXED ASSETS: -Fixed Assets are shown at cost less accumulated depreciation. Cost of asset includes all expenses -related to acquisition of the asset. -b. DEPRECIATION: -Owned Assets: Depreciation is charged on Straight Line Method at the rates and in the manner -provided in Schedule XIV of the Companies Act, 1956. -Leased Assets: On leased assets the minimum statutory depreciation is charged along with which -the stipulated lease Equalisation Account is created, to ensure full depreciation of the leased assets -within the tenure of the lease period by way of matching annual charges based on the lease rental -accruals. -Investments are stated at cost. -Miscellaneous Expenditure are written off as follows: -a. Preliminary Expenses, Pre-Operative Expenses (other than under (b)below) & Public Issue Expenses -- Pro-rata over 10 years. -b. Pre-Operative Expenses relating to Merchant Banking Division and Registrar & Share Transfer -Agents - Pro-rata oYar 5 years. -d. Computer Software -Equally over 3 years. -Taxes on Income -The Company provides for deferred tax using liability method, based on tax effect of timing -difference resulting from the recognition of items in the financial statements and in estimating its -current income tax provision. -Deferred Tax Assets arising from temporary timing differences are recognized to the extent there - -IF ARYAMANFINANCIALSERVICES LIMITED -Taxes on Income -The Company provides for deferred tax using liability method, based on tax effect of timing -difference resulting from the recognition of items in the financial statements and in estimating its -current income tax provision. -Deferred Tax Assets arising from temporary timing differences are recognized to the extent there -is reasonable certainity that the assets can be realized in future. -NOTES -1. Estimated amount of contracts to be executed on capital accounts Rs. 2,48,500/- -(P.Y. Rs. 2,48,500) -2. Contingent Liability RS.42.36 Lakhs (P.Y. RS. 42.36 Lakhs) -3. During the previous year the company assigned an office space which was pending registration -and a liability there against of Rs. 39.83 lacs to a party. The Company had filed a suit in the High -Court against the seller of the property for specific performance of the contract i.e. in arranging to -handover the original title papers making the bank as the third party to the suit. The company had -assigned the assets along with the liabilities to a party with the understanding that no liability -should accrue to the company with effect from 15-03- 2004 and the pending legal case shall be -pursued by the assignee. -4. Some of the debit/credit balances are subject to confirmation and reconciliation. -5. In view of the number of employees being below the stipulated numbers, the Payment of Gratuity -Act is not applicable to the company for the year. The same is the case with respect to payment -of Bonus Act. -6. Auditors Remuneration: - -| Amount (Rs.) | Amount (Rs.) | | -| ------------ | ------------ | --- | -(P.Y.) -| Audit Fees-Statutory | 43750 | 35000 | -| -------------------- | ----- | ----- | -| -Tax Audit | 12500 | 10000 | -| Certification | 10000 | 10000 | -NIL -| Out of pocket expenses | NIL | | -| ---------------------- | ----- | ----- | -| TOTAL | 66250 | 55000 | -7. Deferred Tax Liability / Assets -| Deferred Tax Liabilities | Amount (Rs.) | | -| ------------------------ | ------------ | --- | -| -At 1st April, 2004 | 87,46,000 | | -- Additional Adjustments for Current year -87,46,000 - -8. Directors' Remuneration : - -| | Amount | (Rs.) | Amount | (Rs.) | -| --- | ------- | ----- | ------- | ----- | -(P.Y.) -| Salary | | 60000 | | 360000 | -| -------------------- | --- | ------ | --- | ------ | -| Provident Fund | | NIL | | NIL | -| Perquisites | | 42000 | | 252000 | -| Legal & Professional | | 495000 | | NIL | -| TOTAL | | 597000 | | 612000 | -9. Quantitative Details of Shares under Portfolio Advance: -| | | No. | Amount | (Rs.) | -| --- | --- | --- | ------- | ----- | -Equity Shares of Rs. 107- each Fully Paid-up -| Opening Stock | | 1655784 | P.Y.(1555784) | | -| -------------- | --- | -------- | ------------------ | --- | -| | | 52095327 | P.Y.(5109S327) | | -| Purchase | | NIL | P.Y. (I,0ft,fl00) | | -| | | NIL | P.Y. (10,00,000) | | -| Sale | | NIL | P.Y. (NIL) | NIL | -| | | P.Y. | (NIL) | | -10. Income/Expenditure in foreign currency: -| | Amount | (Rs.) | Amount | (Rs.) | -| --- | ------- | ----- | ------- | ----- | -(P.Y.) -| Expenditure | Rs. NIL | | Rs. NIL | | -| ----------- | ------- | --- | ------- | --- | -| Income | Rs. NIL | | Rs. NIL | | -11. EARNING PER SHARE -| | Share Price | | Share Price | | -| --- | ----------- | --- | ----------- | --- | -As at -As at -| | JUNE 30, 2005 | | March 31. 2004 | | -| ----------------------------- | ------------- | ----- | -------------- | ----- | -| Nominal value of Equity Share | Rs.10/- | | Rs.10/- | | -| Earning per Share | (Rs. | 0.27) | (Rs. | 0.25) | -12. Related Party Disclosure -Mr. P. V.R. Murthy, Director of the company was covered by the definition of the Key Management -Personnel upto December 2004. Remuneration of Rs 5,52,000/- is paid under his contract of -employment with the company. Additionally, as at 30th June, 2005 there is no other related party -relationships as envisaged in Accounting Standard 18 pertaining to Related Party Disclosures. -13. Current year figures are for 15 months as compared to previous year figures for 12 months. -14. Previous years' figures have been regrouped and rearranged wherever necessary. - -F ARYAMANFINANCIALSERVICES LIMITED -Additional information as required under Part IV of Schedule VI to the Companies Act, 1956 -Balance Sheet Abstract and Company's General Business Profile -I Registration Details -Registration No. 59009 -State Code No. 55 -Balance Sheet Date 30.06.2005 -II Capital Raised During the year (Rs. In Thousands) -Public Issue -Nil -Rights Issue Nil -Bonus Issue -Nil -Private Placement - Equity Shares -Nil -III Position of Mobillisation and Deployment of Funds (Rs. In Thousands) -Total Liabilities 59066 -Total Assets 59066 -Sources of Funds -Paid - up Capital 53619 -Reserves & Surplus -3299 -Secured Loans -Nil -Unsecured Loans -Deferred Tax liability (Net) 8746 -Application of Funds -Net Fixed Assets 29023 -Investments Nill -Net Current Assets 30043 -Misc. Expenditure 0 -Accumulated Losses Nil -IV Performance of the Company (Rs. In Thousands) -Turnover (Gross Receipts) 2767 -Total Expenditure 4024 -Profit (+) / Loss (-) before Tax -12 5 7 -Profit (+) / Loss (-) after Tax -1257 -Earning per share (Rs.) Nil -Dividend (%) Nil -V Generic names of Principal Products , Services of the C ompany -Item Code No. (ITC Code) -Not Applicable -Production Description -1 Fjnanda| Sen/jces -2. Project Consultancy -3. Merchant Banking -Code No. for the services rendered by the Company is not available in the Publication of Indian Trade -classification for ITC Code of Products by Ministry of Commerce, Directorate General of Commercial -Intelligence and Statistics, Calcutta 700 001 -As per our Report of even date -For Thakur Vaidyanath Aiyar & Co. For and on behalf of the Board -Chartered Accountants -C.V. Parameswar Director -Partner -M.No. 11541 Director -Place:- Mumbai -Dated:-01-12-2005 - -P= ARYAMAN FINANCIAL SERVICES LIMITED -ARYAMAN FINANCIAL SERVICES LIMITED -Registered Office : 401, Chiranjiv Towers, 43, Nehru Place, New Delhi 110019 -Corporate office : 208, Maker Chambers-v, Nariman Point, Mumbai 400 021. -ATTENDANCE SLIP -Eleventh Annual General Meeting Scheduled on Thursday, 29th Dec.2005 AT -401, Chirajiv Towers, 43, Nehru Place, New Delhi 110019. at 10,00 am. -PLEASE COMPLETE THIS ATTENDANCES SLIP AND HAND IT OVER AT THE ENTRANCE OF THE MEETING HALL -NAME & ADDRESS OF THE SHAREHOLDER L.F.NO -I, hereby record my presence at the ELEVENTH ANNUAL GENERAL MEETING of the Company -Scheduled on Thursday , 29th, Dec. 2005 at 401, Chiranjiv Towers, New Dilhi at 10. am. -SIGNATURE OF THE SHAREHOLDER OR THE PROXY ATTENDING THE MEETING. -If Shareholders, please sing here if Proxy, please sing here -NOTE : The copy of Annual Report may please be brought at the Meeting Hall -(TEAR HEAR) -ARYAMAN FINANCIAL SERVICES LIMITED -Registered Office : 401, Chiranjiv Towers, 43, Nehru Place, New Delhi 110019. -Corporate Office : 208, Maker Chambers -v, Nariman Point, Mumbai 400 021 -PROXY FORM -L.F. No. -of being a member / members of ARYAMAN FINANCIAL SERVICES LIMITED -hereby appoint ........ - ...... -of --------------------------------------------------------- failing him / her- ----------------------------------------------------o f -or -as my / our proxy to vote fo>- me/ us on behalf at the Eleven Annual General Meeting of the Company to -be held at 401 , Chiranjiv Tower, New Delhi - 110117 on Thursday 29th Dec. 2005 -as witness my/ our hand(s) this ----------------------------- day of - ................... — 2005 -(Signature of the Shareholder) -NOTE :The Proxy From must be returned so as to reach the Registered Office of the Company note less -than 48 hours before the time for holding the aforesaid meeting. - -BOOK-POST -To, -// undelivered please return to .- -Adroit Corporate Services Pvt. Ltd. -Unit, Aryaman Financial Services Ltd. -19, Jaferbhoy Industrial Estate, 1st Floor, -Makwana Road, Marol Naka, -Andheri (W), Mumbai - 400059 \ No newline at end of file +| Prompt4: | Questiononlyevaluation | | | | | | +| ---------- | --------------------------- | ------ | -------------- | -------- | ------------ | --- | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt5: | Retrievalwithentityname | | | | | | +| Entity | name: | $C$ | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt6: | RetrievalwithKBArticle | | | | | | +| Context: | $Art$ | | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt7: | RetrievalwithKBSection | | | | | | +| Context: | $S$ | | | | | | +| Question: | $Q$ | | | | | | +| The answer | is: | | | | | | +| Prompt8: | LensretrievaltogetKBSection | | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | Is | this | fungus edible? | | | | +Text: This compound induces mammalian cells (specifically, the cell line HL60 to differentiate +into granulocyte- or macrophage-like cells. The fungus also contains the mycotoxin muscarine, +and the antifungal metabolite strobilurin D. Despite the presence of these toxins, some guides +| list this | fungus | safe | for human | consumption. | | | +| ---------- | ------ | ------ | --------- | --------------- | ------------- | --- | +| The answer | is: | yes | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | In | which | season | does this plant | give flowers? | | +Text: This cactus has stems about 1/2-1 inch wide with 6-9 edges. Its flowers are white, up +to 30 centimetres in diameter with a scent redolent of vanilla. The flowers open after sundown, +closing and wasting after a few hours. By 9 am the next day they are gone. +| The answer | is: | no | | | | | +| ---------- | ------ | ------ | ----------- | -------- | ------------ | --- | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | What | is | the habitat | of this | animal? | | +Text: X is native to Europe and North Africa through to Central Asia. It is introduced to the +United States and parts of South America. It widespread across the northeastern United States +and eastern Canada, and can be found outside, or more commonly inside houses. It is thought to +| have been | introduced | | into America | from | Europe by English | colonists. | +| ---------- | ---------- | ------ | ------------ | -------- | ----------------- | ---------- | +| The answer | is: | yes | | | | | +| Can the | answer | to the | question | be found | in the text? | | +| Question: | $Q$ | | | | | | +Text: $S$ +| The answer | is: | | | | | | +| ---------- | --- | --- | --- | --- | --- | --- | \ No newline at end of file diff --git a/q188/random_k4/random_2.md b/q188/random_k4/random_2.md index 5c5eba2e1fd980c21e8cbb1cd873801d5dedd049..a334b8d779333a86838d9effd3dd99b5ea050aa5 100644 --- a/q188/random_k4/random_2.md +++ b/q188/random_k4/random_2.md @@ -1,2541 +1,704 @@ - CONTENTS +SOCIAL MEDIA MARKETING -01 -04 - 16 -36 -38 -39 -40 -41 -44 -45 + SOCIAL MEDIA MARKETING -/ About HKU Business School +W H A T T H I S S E S S I O N W I L L C O V E R -/ Explore Your Options +1 -/ Enrich Your Life +2 -/ 2026 JUPAS Admissions +3 -/ 2026 Non-JUPAS Admissions +4 -/ 2026 Dual Degree Admissions - -/ Develop Your Career - -/ Freshmen Sharing - -/ Our Global Presence - -/ Useful Websites - -About HKU Business School - -HKU Business School strives to nurture first-class business leaders and foster academic and relevant -research to serve the needs of Hong Kong, Chinese Mainland and the rest of the world in the new Asia-led -economy. As Asia’s premier international business school, we engage leading scholars from all corners of the -globe and they instil in students global knowledge with an Asian perspective. - -With the highest proportion of non-local undergraduate students across all Faculties at HKU, four of our -undergraduate programmes ranked amongst the University’s Top 10 programmes in 2025#, HKU Business -School attracts the best and brightest students from Hong Kong and beyond. - -HKU Business School is fully accredited by the European Quality Improvement Systems (EQUIS) and the -Association to Advance Collegiate Schools of Business (AACSB). - -# Based on average JUPAS admission scores in the best 5 subjects in HKDSE - -VISION -To be a leading, -globally-impactful academic -institution of business and -economics. - -UNIQUE -PROPOSITION - -Deeply rooted in Hong Kong -Fully engaged with China -Truly international - -MISSION -STATEMENT - -To provide world-leading -business and economics -education that nurtures -dynamic and responsible -leaders. -To advance the frontiers of -knowledge through innovative -and globally-significant -research. -To create positive change in the -world by partnering with -communities. - -01 - - Dean’s Message - -I am delighted to introduce you to the HKU Business -School, a dynamic and forward-thinking academic -community where transformative opportunities and -unparalleled experiences await. - -As a part of the University of Hong Kong, which -proudly ranks 11th globally in the 2026 QS World -University Rankings, HKU Business School stands out -as a beacon of excellence in business and economics -in Asia. This year marks the 25th anniversary of our -Faculty, we continue to march forward as a leading, -impactful business school, championing -globally -research, and -leadership, pioneering -thought -educational excellence. - -of - -dynamic, - -challenging - -responsible, - -Our mission is to inspire and develop the next -generation -and -globally-minded business leaders. We offer a vibrant -and -learning environment where -students are encouraged to push their limits, think -critically, and innovate. Our team of leading scholars -from around the world, and experienced and -dedicated teachers as well as mentors bring a wealth -of knowledge, global perspectives, and personalised -guidance beyond the classroom. Our full spectrum of -top-quality degree programmes, including 11 diverse -undergraduate programmes and four dual bachelor’s -degrees -in partnership with world-renowned -universities, is designed to equip students with -cutting-edge knowledge and technical skills to -in the fast-changing global -navigate and excel -business landscape. - -In response to the transformative power of artificial -intelligence, we are proud to launch the Second Major -and Minor programmes in “Artificial Intelligence in -Business” -to our -(AIB), offered exclusively -undergraduate students. These programmes are -designed to equip future business -leaders with -practical proficiency in AI technologies, strategic -insight to drive cross-sector innovation, and a critical -understanding of the ethical and societal implications -of AI adoption. Through a rigorous curriculum that -integrates business fundamentals with cutting-edge -AI applications, students will be empowered to tackle -complex challenges and create meaningful impact in -a data-driven world. - -Greetings from -Associate Dean (Undergraduate) - -inspiration - -Business education today is an evolving domain that -builds upon learning from all academic disciplines and -draws -successful and -from both -unsuccessful business practices. Our undergraduate -curricula in six core academic areas equip students -with diverse knowledge across business and -economics spheres, and prepare them to be creative -leaders of tomorrow with global perspectives. - -A hallmark of our School that separates us from the -others is our truly multi-disciplinary approach. HKU -Business School’s extensive resources offer students -in -a wide range of choices, combining studies -economics and finance, accounting and finance, -accounting and data analytics, business and law, -business and data analytics, international business -and global management, as well as dual degrees in -collaboration with world-renowned universities. - -Our faculty members will engage you with their -inspiring teaching and how they bring their impactful -research into the classroom. They will open your mind -to diverse perspectives, -ignite your curiosity to -explore innovative ideas, and instil in you an appetite -for acquiring new knowledge and extraordinary -experiences throughout your academic journey and -career development. Our students are among the -brightest and most creative of their peers. Our alumni -have gone on to forge brilliant careers in all sectors of -the -international business -community, and have formed a global network that -you will be a part of for the rest of your life. - -regional and - -local, - -I hope that you will share my enthusiasm for the -programmes and my admiration for the people who -have contributed to their development. I look forward -to having you as a member of the HKU Business -School family and a future leader in the world of -business. - -Warm regards, - -Professor Derek Chan -Associate Dean (Undergraduate) - -Through our extensive partnerships with over 300 -institutions and corporations across more than 40 -countries, as well as our unique network of centres in -Beijing, Shenzhen, Shanghai, Vietnam, and Barcelona, -our students gain unparalleled global experiential -and enrichment opportunities. Whether through -internships, overseas exchange programmes, or -hands-on -these experiences -sharpen practical skills, broaden perspectives, and -prepare students to confidently tackle real-world -challenges. - -industry projects, - -Joining HKU Business School means becoming part of -a community that supports your growth and success. -You will collaborate with brilliant minds, engaging in -stimulating discussions, and working on impactful -projects that make a difference. The friendships, -networks, and experiences you build here will last a -to a myriad of career -lifetime, open doors -opportunities, and empower you to overcome -challenges, pursue your passions with confidence, -and contribute positively to your community and the -world at large. - -If you are ready to embark on an exciting journey of -learning, discovery, and personal development, I -invite you to explore what HKU Business School has to -look forward to welcoming you to our -offer. I -community, where future leaders are nurtured and -dreams take flight. - -Warm regards, - -Professor Hongbin Cai -Dean -Chair of Economics - -0302 Explore Your Options -A Wide Selection of Programmes - -Bachelor of Business Administration [BBA] - -The BBA programme combines strong functional training with intensive exposure -to communication skills, computer applications, and other social science and/or -science subjects. Students can choose to major in the following area(s): - -Entrepreneurship, -Design and - Innovation* - -Finance - -Human Resource -Management - -Information Systems -and Analytics - -Marketing - -BBA Curriculum Structure +5 -Components - -A - -B - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -BBA Major -Faculty Core Courses -Disciplinary Core Courses / Disciplinary Electives -Capstone Course - -C - -Free Electives / Optional Major / Optional Minor(s) - -Total - -Credits - -54 -12 -36 -6 -- - -78 -24 -48 -6 - -108 - -240 - -*Candidates must undergo a selection process arranged by the Programme Coordinator for -declaring the EDI major. - -HKU – Sciences Po Dual Degree Programme - -Students will spend the first two years at Sciences Po in France and the third -and fourth years at HKU. Programmes at Sciences Po are taught in English (no -French requirements for being admitted) with intensive French language -classes for all non-francophone students. - -Upon successful completion of the dual degree programme, students will be -awarded the Bachelor of Business Administration (BBA) degree from HKU and -a Bachelor’s degree from Sciences Po. - -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. - -Find out more: -https://admissions.hku.hk/programmes/university- -collaborative-programmes/hku-sciences-po-dual- -degree-programme - -HKU – University of British Columbia -Dual Degree Programme - -Students will spend the first and final years at HKU and the second and third -years at the University of British Columbia (UBC) in Canada, and graduate with -a Bachelor of Business Administration from HKU and a Bachelor of Commerce -from UBC upon successful completion of the programme. - -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. - -Find out more: -https://admissions.hku.hk/programmes/university- -collaborative-programmes/hku-university-of-british- -columbia-dual-degree-programme - -Find out more: -https://ug.hkubs.hku.hk/programme/bba - -0504 - - Bachelor of Business Administration in Accounting and -Finance [BBA(Acc&Fin)] - -Bachelor of Business Administration in Accounting Data -Analytics [BBA(ADA)] - -The BBA(Acc&Fin) programme prepares students for a professional qualification -in accounting and provides the broad-based education in accounting and finance -which is necessary to progress toward a leadership role in the financial sector. -Students are required to take professional core in accounting and a minor in -finance, with an option to second major in finance. Tailor-made professional -internship opportunities with renowned -preparation programmes and -accounting firms and financial institutions are available for talented students. It -is accredited as an approved accountancy programme by: - -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia - -BBA(Acc&Fin) Curriculum Structure - -Components - -Credits - -A - -B - -C - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -Professional Core in Accounting -Faculty Core Courses -Disciplinary Core Courses -Capstone Course - -Minor in Finance* -Disciplinary Core Courses -Disciplinary Electives - -D. - -Free Electives / Optional Minor - -*Students may opt to pursue a Major in Finance: - -Components - -A - -B - -C - -University Graduation Requirements - -Professional Core in Accounting - -Major in Finance -Disciplinary Core Courses -Disciplinary Electives -Free Electives -Capstone Course - -54 -12 -36 -6 -- - -108 -24 -78 -6 - -36 -12 -24 - -42 - -Total - -240 - -Credits - -54 - -108 - -78 -30 -18 -24 -6 - -Total - -240 - -Similar to the BBA(Acc&Fin) programme, the BBA(ADA) programme prepares -students for a professional qualification in accounting. It provides students -with an option to focus on data analytics aspect in accounting instead of having -a minor in finance. It is an excellent choice for students who are interested in -STEM with accounting and business applications and wish to pursue their -career in the relevant fields. The intensive accounting training enables -BBA(ADA) students to distinguish themselves from students of the other data -analytics/STEM programmes. Students will be equipped with comprehensive -accounting and business knowledge that prepares them to obtain professional -qualifications in accounting. They will also acquire computer programming -and data analytics skills to meet the needs of the accounting and business -professions in today’s digital transformation economy. It is accredited as an -approved accountancy programme by: - -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia - -BBA(ADA) Curriculum Structure - -Components - -A - -B - -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University - -Professional Core in Accounting Data Analytics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course - -C - -Free Electives - -Total - -Credits - -54 -12 -36 -6 -- - -156 -24 -114 -12 6 -30 - -240 +What is social media marketing and why is it important -Find out more: -https://ug.hkubs.hku.hk/programme/bba-acc-and-fin +Social media landscape -Find out more: -https://ug.hkubs.hku.hk/programme/bba-ada +Social media marketing – use cases and benefits -0706 +What does good look like? - Bachelor of Business Administration (Business Analytics) -[BBA(BA)] +How to execute a campaign + a club success story -Bachelor of Business Administration (Law) and Bachelor of -Laws [BBA(Law)&LLB] +Do’s and Don’ts + additional resources -The BBA(BA) programme is crafted to meet the growing industry demand for -talents in the business analytics field. Its curriculum offers students a -comprehensive coverage of both technical and managerial skill sets. Students -will learn a wide spectrum of knowledge from the disciplines of information -technology, data science, business statistics and management. Core courses -include programming language, database management, introduction to -business analytics, management information systems, decision and risk -analysis, etc. They help students build the foundations for advanced topics in -senior years. Students also have the option to choose from analytics technique -courses and business application courses. The aim is to provide students with -state-of-the-art experiences in artificial intelligence, big data, social media, -and other latest development. Such design helps train well-rounded students -who will be more adaptable in their career development. The programme is -suitable for students who have a passion for problem-solving through -analytics, and who aim to develop their career path in analytics. +2 -BBA(BA) Curriculum Structure + WHAT IS SOCIAL MEDIA MARKETING? -Components +D E S C R I P T I O N -Credits +The use of social +networks, content +sharing apps, +messaging platforms, +blogs, and forums to +connect with and enable +meaningful +conversations about +your brand, product or +service, to ultimately +drive your business +objective. -A +3 -B + WHY IS SOCIAL IMPORTANT? -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +C U S T O M E R E X P E C T A T I O N S H A V E C H A N G E D -Major in Business Analytics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives (Analytics Techniques) -Disciplinary Electives (Business Applications) -Capstone Course +2X -C +customers are twice as likely to +share a negative experience with a +business than a positive one -Free Electives / Optional Major / Optional Minor(s) +90% -54 -12 -36 -6 -- - -96 -24 -42 -12 -12 -6 +will look at online +reviews before making +a purchase -90 +74% -Jointly offered by the HKU Business School and the Faculty of Law, the -BBA(Law)&LLB integrated double-degree programme blends two strategic -disciplines into a single programme for the bright and highly motivated. It is an -excellent choice for students seeking a broader education with parallel -concentration on legal and business studies. Graduates of the programme will be -awarded two degrees — BBA(Law) by the HKU Business School, and LLB by the -Faculty of Law. After successful completion of the LLB programme, students -may apply for admission to the Postgraduate Certificate in Laws (PCLL) -programme provided that they meet the relevant admission requirements. +trust suggestions from +”friends” on social media -The programme offers business stream and accounting stream to students in -their business studies. The accounting stream is accredited by: +67% -The Hong Kong Institute of Certified Public Accountants (HKICPA) -The Association of Chartered Certified Accountants (ACCA) -CPA Australia +of people will spend money after +getting recommendations from +their “friends” online -BBA(Law)&LLB Curriculum Structure (Business Stream) +Sources: White House Office of Consumer Affairs, McKinsey, Experian |24|7, Gigya, CEI -Components +4 -Credits + WHY IS SOCIAL IMPORTANT? -A +N E W T R E N D S A R E E M E R G I N G C H A N G I N G T H E U S A G E A N D I M P O R T A N C E O F S O C I A L -B +Understanding the social media space and managing your social presence amplifies your marketing +strategy. Social media delivers invaluable insight into your brand awareness, customer sentiment, +marketplace trends, and your competitor’s actions, whilst enabling you to reach more prospects than any +other marketing channel. -C +Every negative social interaction has a cost -University Graduation Requirements -Language Enhancement Course -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Customers are content creators -BBA Major -Faculty Core Courses -Disciplinary Core Courses / Disciplinary Electives -Capstone Course +Brands are turning online influencers into advocates -Professional Core in Law -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Social is now a prominent point of purchase -D. +Brand focus is shifting to measuring quality of interactions -Free Electives / Optional Minor +Reviews can increase sales and add credibility -Total +Social is multifunctional – used for marketing, listening, +response, customer care, troubleshooting -240 +5 -Total + SOCIAL MEDIA MARKETING LANDSCAPE -BBA(Law)&LLB Curriculum Structure (Accounting Stream) +T H E K E Y P L A Y E R S -Components +The social media landscape is large and +complex and cluttered with competing +platforms, but the most heavily utilised +for marketing are the big 4: -A +Facebook (FB, Instagram, +Messenger) -B +Google (YouTube) -University Graduation Requirements - -Professional Core in Accounting -Faculty Core Courses -Disciplinary Core Courses -Capstone Course - -C - -Professional Core in Law - -Total - -36 -6 -24 -6 -- +Twitter -78 -24 -48 -6 +LinkedIn -156 -108 -42 6 -30 + SOCIAL MEDIA MARKETING LANDSCAPE -300 +D I F F E R E N T C H A N N E L S A R E U S E D F O R D I F F E R E N T P U R P O S E S -Credits +Social networks +Relationship networks help brands and people connect to share +information and ideas. With a large & regular user base, they are a must- +have and the ”mass-market” approach to social marketing. -36 +Use for: customer care, education, daily comms, breaking news -108 -24 -78 -6 +Messaging & chat apps -156 +Most widely used apps that enable messaging, video and voice calls. Business +can conduct e-commerce conversations via live agents or chatbots. -300 +Use for: customer care, conversational commerce, payments, status updates, +to replace SMS -Find out more: -https://ug.hkubs.hku.hk/programme/bba-ba +Media & content sharing +networks -Find out more: -https://ug.hkubs.hku.hk/programme/bba-law-and-llb +Find and share media online, including photos, video, and live video or other +curated content. -0908 +Consumer review +networks +Manage online reviews and reputation. Positive reviews bring social proof to +your value proposition. Negative reviews provide you with an opportunity to +resolve an issue publicly. - Bachelor of Business Administration in International Business -and Global Management [BBA(IBGM)] +Use for: brand awareness, lead generation, audience engagement, targeted +advertising, influencer marketing -Bachelor of Economics [BEcon] +Use for: customer care, reputation management, market research -BBA(IBGM) is an intensive, premier programme in international business and -global management featuring a unique curriculum designed to evolve with and -reflect the ever-changing demands of a globalised world. Exclusive elements -such as overseas exchange studies, a leadership boot camp, an Asia-Pacific -study tour, an international field trip and global enrichment opportunities are -strategically integrated into the curriculum, whilst students are given the -flexibility to pursue an in-depth inquiry into another specialised area of study -as a second major. This cross-functional and personalised approach aims to -foster a selected group of outstanding students to become the world’s future -leaders equipped with global perspectives, cultural sensitivity, social -responsibility and key competencies to succeed in the international business -environment. +Blogs & forums +Publish, find, discuss, and share news, information, and opinions. Usually allow +users to remain anonymous, leading to more honest opinions. -BBA(IBGM) Curriculum Structure +Interest-based networks +Help connect with others around a shared interest or hobby and tend to focus +solely on a single subject and provide a dedicated experience for users. -Components +Use for: market research, influencer marketing, product advertising, SEO +results -A +Use for: targeted marketing, brand awareness, trend watching -B +7 -C + SOCIAL MARKETING USE CASES -D +H O W T O U S E S O C I A L T O M E E T Y O U R O B J E C T I V E S -E +Manage +own page -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Generate +insights -Major in International Business and Global Management -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Nurture community +& reputation -Second Major +Engage with +customers -Free Electives +Grow through +amplification -Compulsory Overseas Exchange +Draft and publish organic +content to owned +page(s) -Total +Define and manage +content strategy and +calendar -Credits +Manage content +approvals -54 -12 -36 -6 -- +Listen/capture social +conversations about LTA, +coaches, or venues and +identify trends -96 -24 -54 -12 -6 +Keep up with competitors +and stay informed of their +social engagement, +content and campaigns -72-90 +Understand target +audience demographics, +sentiment, and channel +preferences -0-18* +Grow your community of +social followers and +supporters -- +Find and nurture +advocates and +influencers who can +credibly promote your +brand to an extended +audience -240 +Identify and manage +potential crises -*Students may use free electives to fulfil the second major where the requirements exceed 72 credits. +Respond to consumer +problems when they +have questions or +concerns -The BEcon programme trains students in economic theories, statistical tools, -the application of theories to solve practical economic problems in trade, -banking and the labour market, and evaluation of the regional economic -performance of Chinese Mainland, Hong Kong and other economies. Students -will be furnished with solid theoretical and empirical economic knowledge to -analyse and understand human behaviour and social interactions. The -programme allows students a high degree of flexibility in planning their future -career specialisation. +Be part of the +conversations about your +venues, coaches or +brand -BEcon Curriculum Structure +React with relevant +content – use what you +see and hear to help +inform future marketing +strategy -Components +Push paid content to +custom audiences to reach +your marketing objectives -A +Identify and amplify +relevant user content for +authentic engagement -B +Social commerce – +increasing number of +businesses are using +social as a payment +channel -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +8 -Major in Economics -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course + THE BENEFITS OF SOCIAL -C +W H A T S O C I A L M E D I A M A R K E T I N G C A N D O F O R Y O U -Free Electives / Optional Major / Optional Minor(s) +1. -Total +Strengthen your +brand reputation -Credits +2. -54 -12 -36 -6 -- +Grow your +business -78 -24 -30 -18 -6 +3. -108 +Care for your +customers -240 +Know & act on what is being said about +you +• Things escalate quickly on social. By -Find out more: -http://ibgm.business.hku.hk +listening to key phrases that relate to your +brand or activity you can be the first to +know when a news story breaks and react +accordingly. -Find out more: -https://ug.hkubs.hku.hk/programme/becon +• Social insights also give you an unbiased -1110 +view of your brand awareness and +popularity. - Bachelor of Economics and Finance [BEcon&Fin] +Identify trends +• Trends appear on social media first. Social -BEcon&Fin is designed to meet the interests of students who are seeking a -balanced study of both economics and finance. The programme recognises -the significance of economics both in its own right and as a foundation for the -study of finance, and the significance of finance as a career path for students -who are keen on joining this profession. Students will receive solid training in -theories and applications of economics and finance. They will be equipped -with the skills to analyse human behaviour and social interaction, as well as to -solve problems in the financial sphere. +can make or break a product. -BEcon&Fin Curriculum Structure +• Listening to industry thought-leadership +supports your research and innovation -Components +Stay ahead of competition +• Your competitors are on social. Analysing -Credits +their social presence delivers unique +insight into their plans and achievements +• Regular benchmarking of your competitors -A +on social helps you keep up with them -B +Amplify owned content +• Social enables you to publish targeted -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +content to channels that could be seen by +large volumes of users -Major in Economics and Finance -Faculty Core Courses -Disciplinary Core Courses -Economics Disciplinary Electives -Finance Disciplinary Electives -Capstone Course +• The power of influencer marketing -C +provides a human voice to your product -Free Electives / Optional Major / Optional Minor(s) +Generate earned media +• Customers are 56% more likely to buy a +product or service after seeing a positive +customer-generated post of it -54 -12 -36 -6 -- +• By reposting user-generated content on +your owned channels, you increase the +trust in your brand and products -96 -24 -42 -12 -12 -6 +Deliver incremental sales +• Advertising on social media is cheaper +than traditional media (such as TV ads) +but with a massive reach -90 +• The main difference is that you will be -Total +able to target finely who sees what advert, +increasing your conversion while keeping +costs down -240 +Create experiences +• Social is more than just advertising, it’s an +educational, emotional and experiential +way to engage your customers -HKU - Peking University “Future -Leaders” Dual Bachelor’s Degree in -Economics, Finance and Management -The programme combines the strengths of two leading business schools in -Hong Kong and Chinese Mainland and is designed to nurture first-class business -leaders and cultivate their global perspective with intimate understanding of -China. Students will spend the first two years at HKU and the third and fourth -years at Peking University (PKU), and be awarded a Bachelor of Economics and -Finance (BEcon&Fin) degree from HKU and a Bachelor of Management [BMgmt] -degree from PKU upon successful completion of the dual degree programme. +• Customers no longer want to phone when -Separate application is required for applying to the dual degree programme, in -addition to the application under the JUPAS or International/Non-JUPAS -Admissions Scheme. +they have an issue, they want an +immediate answer and turn to social. -Find out more: -https://ug.hkubs.hku.hk/admission/degree-program -mes/hku-pku-dual-degree +Reduce cost to serve +• With customised chatbots and the use of +artificial intelligence on the rise, in-app +messaging is becoming the norm for +customer care -Bachelor of Finance in Asset Management and Private -Banking [BFin(AMPB)] -The BFin(AMPB) programme aims to prepare students to enter the rapidly -growing fields of asset management and private wealth management in Hong -Kong and Asia Pacific, or in banking and finance generally. During the first two -years, students complete introductory courses in accounting, economics, -mathematics, statistics, corporate finance, and investments and portfolio -analysis. In subsequent years, students take more specified and industry -related courses including equity valuation and investment in specific asset -classes e.g. private equity, venture capital, hedge funds and digital assets, -understanding of financial statements and doing financial analysis, lending -and credit in private wealth and corporate bank, financial regulations and -compliance, current -institutional -in finance, ESG -consulting and family office. The programme contains practical-oriented -courses taught by teachers with extensive experience in the finance industry -and offers an array of career-oriented enrichment activities and networking -opportunities with industry practitioners. +• Technology solutions enable improved -investment, +customer service through social -issues +Build a community +• Customers trust each other more than +you. Providing them with a platform +where they can interact and share +feedback will increase their trust and +loyalty to your brand, coaches, or venues. -in +9 -BFin(AMPB) Curriculum Structure + WHAT DOES GOOD LOOK LIKE? -Components +H O W T O W I N W I T H S O C I A L M E D I A -Credits +A great social media strategy includes the following pillars: -A +SOCIAL MEDIA +PILLARS -B +Emotional content -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Build rapport with your customers. Forget transactional posts +and focus on posting content that showcases how you will make +a difference in your customers’ life -Major in Asset Management and Private Banking -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +Consistency -C +Your content should match your brand identity and objectives +and be consistent across channels. -Free Electives / Optional Major / Optional Minor(s) +Regular & well- +timed posts -54 -12 -36 -6 -- +Just like any other channel, there is a better time to post content, +and that varies depending on the network and audience. Keep +your customers interested by posting often and sharing varied +content -96 -24 -42 -24 -6 +Two-way +engagement -90 +Social is about more than just pushing out content. Allow your +customers to interact with you and make sure you respond to +create a differentiated experience -Total +Optimisation -240 +Test, learn, adjust, repeat. Iteration is key to success and social +is no different. Plan for A/B testing and track success of +individual campaigns and performance across platforms -Find out more: -https://ug.hkubs.hku.hk/programme/becon-and-fin +Personalisation -Find out more: -https://ug.hkubs.hku.hk/programme/bfin-ampb +Strategic use of +channels -1312 +Use social insights to know who your customers are and adapt +your messaging, including graphics and tone of voice to target +the right customers with the right message at the right time - Bachelor of Science in Marketing Analytics and Technology -[BSc(MAT)] +Not all channels will be relevant to your brand. Whilst a presence +on the Big 4 is a must, focusing your budget and efforts on the +channels where your customers are the most active will deliver +better value for money -Bachelor of Science in Quantitative Finance [BSc(QFin)] +Overall strategy +alignment -Marketing is critical to all businesses. It dictates whether a company will be -successful. Technology, Big Data, and digitization are fundamentally changing -how marketers operate. The BSc(MAT) programme equips students with -quantitative skills, technology know-how, and business marketing acumen to -pursue a career in the digital economy and modern businesses. It targets both -STEM students with an interest in business as well as business students -interested in the technology space. Students will gain a solid foundation in -cutting-edge technical areas including computer programming, data science, -and technology by excelling in courses from the Faculty of Engineering and -Faculty of Science. In addition, students will master digital business, -marketing and leadership through courses at the HKU Business School. This -interdisciplinary approach will enable students to engage in a variety of jobs -and functions such as the design and launch of new technology products, -targeting consumers, digital marketing, forecast and predicting markets, and -building a technology-focused strategy. The programme aims to nurture -future leaders in the field of technology and marketing who are able to use the -latest technology, science, and data analytics to support business and -marketing decisions in the digital world. +Driving revenue and engagement from social does require +investment. Incorporate social as its own channel into your +overall strategy, budget and resources -BSc(MAT) Curriculum Structure +10 -Components + WHAT DOES GOOD LOOK LIKE? -Credits +P U T T I N G S O C I A L M E D I A P I L L A R S I N T O P R A C T I S E -A +Conversational & Engaging Content – +Royal Academy & Aldi -B +Influencer & Interest-based Marketing – +Gymshark & Wilson -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Royal Academy uses relatable and +humourous posts and interactive +campaigns such as their +#RAdailydoodle to keep their 441K +Twitter followers engaged, drawing +increased user participation and +personalised well-crafted responses -Major in Marketing Analytics and Technology -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives (Marketing Technology and Strategy) -Disciplinary Electives (Marketing Analytics) -Capstone Course +Aldi UK celebrated their birthday with +a simple, yet creative invitation to +engage competitors, generating a ton +of interest on Twitter and a positive +response, well-tracked via their +unique hashtag +#Aldi30thBirthdayParty -C +Gymshark sponsors key Instagram +influencers to capitalise on their large +audiences (500k+). They also use paid +social as a primary means to drive +sales, and in 2019 saw a £73m increase +in turnover. In 2017, their Black Friday +posts drew a 6.6x return on ad spend. -Free Electives / Optional Major / Optional Minor(s) +Influencer marketing can also reap +benefits on a smaller scale. Like the +Germany-based Tennis Siblings Alex +& Sofi, who are sponsored by Wilson +and adidas and use their modest +following of 15K to promote products +and local tennis tournaments. -54 -12 -36 -6 -- +11 -96 -24 -42 -12 -12 -6 + EXECUTING A CAMPAIGN - WHERE TO START -90 +G E T T I N G T H E B A S I C S R I G H T -Total +2 -240 +Identify your audience +and grow followers -BSc(QFin) is a premium finance programme. Built on a solid foundation in -economics and finance, the programme equips students with the essential -quant skills, including mathematics, statistics, and programming, that are -required in more advanced and sophisticated analyses, such as valuation -models of structured products, risk management, simulation and data -analysis, financial engineering, and quantitative trading. +Start following your members and +people who “look like” them (e.g. +have similar demo, interests, +location, etc.) -The programme goes beyond merely equipping students with quant skills. -More importantly, it stresses on combining these skills with an understanding of -the financial markets, the macro environment, and the practical situations -where financial managers work, with a balanced mix of quant and business -courses. To nurture students’ soft skills and market acumen, the programme -provides a wide range of value-enhancing activities, such as Round Tables with -Practitioners, Alumni Drinks, Executive Mentorship Scheme, Bloomberg training -workshops, simulation trading workshops, and an international field trip to -other international financial centres such as New York, London and Zurich. +Plan a content calendar and post a +variety of engaging content +including text, photo, and video. -BSc(QFin) Curriculum Structure +Track performance to see what +content resonates best with your +audience. -Components +3 -Credits +Launch and track your +campaign -A +You can begin with a simple +campaign – such as amplifying a +piece of existing content that +performed well with your existing +followers. -B +Set up campaigns by objective +(page likes, clicks, etc.) and set a +daily or monthly budget. -University Graduation Requirements -Language Enhancement Courses -Common Core Curriculum -Artificial Intelligence Literacy Courses -Non-Credit Bearing Courses as required by the University +Use built-in dashboards from the +social platforms to review +performance and optimise by ad +frequency, time of day, ad format, +and more. -Major in Quantitative Finance -Faculty Core Courses -Disciplinary Core Courses -Disciplinary Electives -Capstone Course +1 -C +Set up your account +page(s) -Free Electives / Optional Major / Optional Minor(s) +Establish your presence by +setting up business account / club +page across the relevant social +platforms -54 -12 -36 -6 -- +Include essential business +information for search purposes, +including: Location, Contact info, +and Website URL + +#Tip: use hashtags to +start a conversation with +your members and easily +follow what they say about +you and your club + +4 + +Scale up and add on +to master social + +Once you have your first campaign +under your belt, report on your +performance and measure against +your original business KPIs. + +Refresh your content and launch +new campaigns for different target +audiences. + +Explore and introduce more +complex functionality available to +you, including new data sources, +more robust tracking (e.g. FB pixel), +new ad formats, or new social +platforms. -96 -24 -54 12 -6 - -90 - -Total - -240 - -Find out more: -https://ug.hkubs.hku.hk/programme/bsc-mat - -Find out more: -https://ug.hkubs.hku.hk/programme/bsc-qfin - -1514 - - Enrich Your Life - -initiative and -HKU Business School fosters a culture of student -implementation of ideas, both within and outside the campus, as it is -these real-world engagements that set us apart. Experiences in learning -beyond the classroom, extending into realms outside academia, enable -students to cultivate an understanding of social dynamics, cultural -diversity, and collaborative -intellect. These experiences contribute -significantly to their personal growth, interpersonal skills, and social -development, shaping them into global citizens, lifelong learners, and -future business leaders. - -Overseas Enrichment Experiences - -Business Exchange Programme / -HKUWW Undergraduate Student Exchange Programme -Even as the youngest Faculty at the University, our exchange -activities are nothing but phenomenal. Each year, over 400 -students participate in overseas exchanges at leading universities -to broaden their horizons. - -Business School -Exchange Partners * - -Canada - -McGill University -University of British Columbia -University of Toronto -Western University -Simon Fraser University - -U.K. - -King's College London -University of Manchester -University of Edinburgh -University of Warwick - -The Netherlands - -University of Amsterdam -Erasmus University Rotterdam - -400+ -exchange partners -across 40+ countries - -400+ -Business School -students on exchange - -Denmark - -Copenhagen Business School -Aarhus University - -Germany - -Frankfurt School of Finance & Management -University of Mannheim -Munich Business School - -Belgium - -KU Leuven - -Italy - -Bocconi University - -France - -U.S. - -Toulouse School of Economics -ESSEC Business School - -New York University -University of Southern California -University of Washington -University of Georgia - -Middle East - -Reichman University - -Singapore - -National University of Singapore -Nanyang Technological University - -* This list is non-exclusive. - -International Field Trips -To enhance students' exposure to global businesses, the School organises -overseas field trips for students visiting corporate firms and mingling with -C-suite Executives in some of the world's most prominent financial -centres, e.g. Australia, France, Japan, London, New York City, Singapore, -Spain, and Zurich. - -Visiting Programmes -Through the Yale Visiting International Students Programme, students -are sent to spend a year at Yale University, where they actively engage -themselves in a multicultural society. With the generosity of the donors, -some of our students are selected for the Junior Year Abroad Programme -and Visiting Students Programme to study as visiting students at -Pembroke College and St Peter’s College of University of Oxford -respectively with substantial scholarships. The intellectual ambience of -these top universities is truly inspirational and conducive to learning. - -Beta Gamma Sigma Regional Leadership Conference -Beta Gamma Sigma (BGS) is the international honour society that -serves business programmes accredited by the Association to Advance -Collegiate Schools of Business (AACSB International). With the -establishment of the HKU BGS Chapter, the School registers students -annually for the BGS Regional Leadership Conference held in the U.S., -where they can meet and bond with many other fellow students from the -world's best business schools. The Leadership Summit is eye-opening -and challenging, which leads to a great journey of self-discovery. - -Japan - -Kyoto University - -Australia - -University of Sydney -Monash University -University of New South Wales - -1716 Student Sharing - -Chen, Hsuan-yu -Exchange at National University of Singapore (Singapore), 2024-25 - -As a third-year undergraduate student pursuing a double major in -Accounting and Finance, I consider it a tremendous privilege to have -participated in a one-semester exchange at the National University of -Singapore (NUS) through the Business Exchange Programme. This -invaluable opportunity has not only enriched my academic journey but -also broadened my global perspective through immersive cultural -experiences and meaningful cross-cultural interactions. As I move -forward in my academic and professional journey, I carry with me not -just the knowledge gained from my courses, but the invaluable life -lessons experienced from Exchange. - -Lee, Tsz To Derek -Exchange at King’s College London (United Kingdom), 2024-25 - -I am incredibly grateful to have this valuable opportunity to study -abroad at King’s College London. The programme provided me a chance -to venture outside of my comfort zone. Not only can you learn how to be -more independent, you can also boost interpersonal skills given the -many opportunities to socialise with others, which is essential for your -future career. The experience is remarkable. I think you will regret it if -you miss out on the overseas exchange programme. - -Cheung, Nga Ting Caroline -Exchange at University of New South Wales (Australia), 2024-25 - -This exchange programme has changed me in many ways. It has helped -me become more outgoing, allowing me to make new friends and -explore different cities. Living independently has been a significant part -of this journey. From exploring new cities to forming lasting friendships, -each adventure has enriched my life in ways I never expected. This -opportunity has not only broadened my horizons but also deepened my -appreciation for diversity and connection among people from different -backgrounds. Each memory I have created during this time is a treasure -that I will carry with me forever. - -Ho, Ying Tung Crystal -Exchange at Copenhagen Business School (Denmark), 2024-25 - -In the Fall semester, I participated in a business exchange programme at -Denmark’s Copenhagen Business School (CBS). This experience was -truly transformative, enriching me both academically and personally -while allowing me to forge lifelong memories and friendships. An -exchange experience abroad truly broadens your horizons and offers -insights beyond what textbooks and lectures can provide. You will gain -invaluable perspectives and undergo personal growth like never before. -I wholeheartedly encourage you to muster your courage and embark on -an exchange journey - you won’t regret it! - -Whitfield III, Gerald Bruce Trey -Exchange at New York University (United States), 2024-25 - -I found the exchange experience to be extremely enriching. A major -highlight for me was exploring the lifestyle and professional sides of the -city. Because of my initiative socially, and in groups such as church, I was -able to get well-acquainted with the local rhythms and lifestyle, and -also made many friends that I’d consider friends for a lifetime. I am -deeply appreciative of being given this rewarding experience, which -significantly shaped both my future and the person I am today. - -Ip, Wing Tung Alla -Student Representative for the BGS Regional Leadership -Conference 2025 in Highland Heights, Kentucky - -Attending the BGS Leadership Conference at Northern Kentucky -University proved transformative. I received profound insights into -leadership, communication and cultural intelligence through immersive -workshops, case competitions and networking opportunities. The event -prompted reflections on personal leadership styles and encouraged -collaboration across diverse peers. As I prepare to enter the workforce, -I intend to apply these learned principles of trust-building, cultural -sensitivity and collaborative leadership to navigate challenges so I can -shape my professional journey with purpose and authenticity. - -1918 Corporate Exposure - -HKU Business School has a strong tie with corporate partners. To nurture students' business -and market acumen, enrichment programmes are organised through the joint efforts of -corporate partnerships. - -Enrichment Programmes in collaboration with Corporate Partners -HKU Business School has been collaborating with corporate partners on initiatives to spice up -students' career exploration journey. The partnership with Accenture enables students to -gain exclusive insights into business consulting through the HKU-Accenture Business -Consulting Programme. - -To provide students with a unique opportunity to explore current business trends and -developments in the Greater Bay Area (GBA), our School has co-organised a tailored study tour -with The Association of Chartered Certified Accountants (ACCA). This initiative aims to -enhance students’ readiness for future career development and personal success. Students will -have the chance to connect with seasoned accounting and business professionals in the GBA. - -HKU Business School Mentorship Programme -Leveraging the School's extensive network in the business and finance industry, we have -developed our own mentorship programme to furnish students with opportunities to -enhance their skills and knowledge essential for their future personal and career -development. Our students are empowered to connect to the real business world through -direct and active interaction with seasoned professionals, executives and entrepreneurs. - -Lee, Lok Yiu Angel -HKU Business School Mentorship Programme 2024-25 - -My mentor offered -invaluable career advice, highlighting key -considerations for entering specific fields and preparing for interviews. -Her experiences and mindset during job searches resonated with me as -I navigated something similar. She emphasised the importance of -open-mindedness and eagerness to learn in the workplace. I kept her -words at heart during my internship. I also enjoyed catching up with her -to share my new job experiences. - -Choi, Han Na Hannah -HKUBS x Deloitte China Mentorship Programme 2024-25 - -As my mentor has been working in Deloitte China for more than 10 -years, I was most curious about how he maintains a work-life balance -and how he stays dedicated and professional at work. Through our -lengthy conversations over lunch, he passed on valuable insights on -developing soft skills and adopting an appropriate work attitude at an -early stage. He is more than an audit mentor but also a life coach to me. - -Chan, Chin Ching Jacky -BSc(QFin) Executive Mentorship Programme 2024-25 - -Throughout the mentorship programme, I have received -precious guidance that has greatly influenced my career -path. My mentor highlighted the importance of identifying -a company that aligns with my goals and would provide -opportunities for me to develop the necessary skill sets. He -also shared practical advice on building and managing -professional and personal relationships. These lessons -have broadened my perspective and also equipped me -with the tools to handle real-world situations more -effectively and confidently. - -Wong, Wan Tat -HKUBS x ACCA Greater Bay Area Study Tour 2024-25 - -My greatest takeaway from this study tour was the power of peer -learning and collaboration. Being immersed in an environment with a -diverse group of individuals, all focused on similar goals, created an -incredible opportunity to learn from each other’s experiences and -perspectives. The collaborative projects and informal discussions were -invaluable in broadening my horizon and developing new approaches -to the accounting field in Chinese Mainland. I am leaving with a strong -network of contacts and a renewed appreciation for teamwork. - -Leung, Ho Lun Maverick -HKU-Accenture Business Consulting Programme 2024-25 - -I visited - -The HKU-Accenture Business Consulting Programme immersed me in -an authentic consulting experience through working with a real client – -their distribution centres, conducted market -HKTV. -questionnaires, and presented the final deliverables - all within the -eight-week programme. I was honoured to meet Managing Directors -from Accenture who shared crucial skills in conducting a consulting -project. It’s truly one of the best programmes for anyone considering a -consulting career! - -2120 Experiential Learning Programmes - -Experiential learning is embedded in our course curriculum. The Business Consulting -Practicum offers business and economics students hands-on experience in working with -clients to solve "real-life" problems relevant to their studies, under the guidance of faculty -members and professional mentors. The Business Case Analysis and Presentation enables -students to communicate and seek practical advice from senior executives through in-class -case presentations. The Impact Lab provides students with an excellent opportunity to work -directly with social ventures, again under the guidance of teachers and professional mentors. -Additionally, students can gain an in-depth understanding of the latest trends in economics, -finance, creativity, innovation, and entrepreneurship in the rapidly growing Greater Bay Area -by participating in the Creativity, Innovation & Entrepreneurship in China Programme. - -Chaabane, Adam Andrea -"Creativity, Innovation & Entrepreneurship in China" -Summer Virtual Programme 2025 - -I have learned that prominent businessmen and professionals in Hong -Kong are people like me who were once attending business school. I -most enjoyed the guest -lectures because they allowed direct -interactions with inspiring figures. The virtual experience enabled me to -follow this programme easily from my home country. I definitely -recommend it for a better understanding of Hong Kong and GBA’s -markets and opportunities that arise in cross-border business. It would -be great to learn more about opportunities for foreigners in the region. - -Kaur, Suchleen -Business Consulting Practicum 2024 - -The opportunity for real-time interaction and feedback with the client had -propelled me to join BCP. I was always curious about how a consultant -would address and communicate with the client. BCP has taught me -about proper business etiquette and equipped me the necessary skills to -negotiate conflicting opinions during team discussions. Throughout the -process, our mentor and professors provided great help and insights. -Their invaluable feedback and questions helped nudge us in the right -direction, taking a step back to look at the big picture and most -importantly, prioritise the client’s needs. - -Zhou, Qingzhuo -"Creativity, Innovation & Entrepreneurship in China" -Summer Virtual Programme 2025 - -My greatest takeaway from the CIEC 2025 programme was the unique -window to gain a comprehensive understanding of the Greater Bay -Area’s dynamic business landscape. I explored the latest trends in -business, finance, and innovation from guest talks and virtual lectures -with industry leaders. I particularly appreciate the coverage on a -diverse range of topics and the opportunity to earn HKU credits. A -highlight was the cultural exploration segment which offered an -unparalleled perspective on the Greater Bay Area. I recommend this -programme without hesitation. - -Tsang, Tze Yeung Matthew -Business Consulting Practicum 2024 - -BCP has been a rewarding experience that allowed me to apply my -academic knowledge to real-world business challenges. Working -directly with my mentor and client has pushed me to think critically and -creatively in ways I hadn't experienced in the classroom. I'm confident -that the problem-solving, communication, and client-facing skills I have -developed will serve me well when pursuing my future career in -consulting or another business-oriented field. - -2322 Cheung, Ho Ning Jolie -UW Global Business Case Competition 2025 (Champion) - -in UW GBCC was an - -Competing -incredibly rewarding and -unforgettable experience. We had the opportunity to work with -participants from different universities. -It was a fantastic -opportunity to exchange perspectives and learn from different -problem-solving approaches. Working under tight timeframes -pushed us to think creatively. UW GBCC has not only elevated my -professional capabilities but further solidified my belief in business -as a powerful vehicle for creating positive change in the world. - -Case Competitions - -The School provides ample opportunities to refine students' presentation and analytical skills -through a myriad of local as well as overseas competitions. From training to delicate analysis -of business cases to a variety of business scenarios, students are equipped with different -skillsets to compete with candidates from the world's leading universities. - -Lo, Nikkia -NHH International Case Competition 2024 (Champion) - -My week at NHH International Case Competition has been an -unforgettable rollercoaster of emotions. From the excitement of -exploring a new Nordic city, the exhaustion of our 24-hour case -competition, to the nervous anticipation of pitching on a global stage -with a -livestream, every moment was extraordinary. Meeting -like-minded peers around the world was an incredible experience that -forged new friendships. Thank you, HKU Business School, along with -our professors, coaches and staff for bringing this opportunity to us! - -Xia, Yunchu Kevin -The McDonough Business Strategy Challenge 2025 (First Place) - -Winning first place at MBSC2025 was incredible and deeply rewarding. -My team was highly engaged. We were constantly bouncing ideas off -each other and refining our strategy. There were moments where we hit -roadblocks, but our team rallied, challenged each other constructively -and developed a cohesive solution. Connecting with like-minded -participants, judges and NGOs was also invaluable. MBSC not only -sharpened my analytical and presentation skills but also reaffirmed my -commitment to leveraging business for good. I am so grateful for the -experience and very eager to apply these lessons moving forward! - -2524 Liu, Yunqi Iris -Deloitte Tax Championship 2024 (Outstanding Team Award) - -This competition has been a fantastic learning experience. It -showed me that tax isn’t just about the technical details - it’s also -deeply connected to social trends and global events like climate -change, geopolitical conflicts and national policies. This broader -perspective has enhanced my understanding of the industry and -its importance in today’s world. I want to express my sincere -gratitude to my advisors for their guidance and recognition of my -team’s hard work. I am grateful for the opportunity to learn and -grow with such an amazing group of people. - -Pan, Ho Man Vivian -HKGCC Business Case Competition 2024/25 (Champion) - -in - -the HKGCC Case Competition was a - -Participating -truly -transformative experience that greatly enhanced my strategic -thinking, teamwork and problem-solving skills. The competition -challenged me to analyse intricate business problems, think critically -under pressure and develop innovative solutions. Working closely -with people of various majors allowed me to appreciate differing -opinions that were at times conflicting, find ways to build consensus -and synthesize ideas effectively. - -Students' achievements in case competitions -(2023-2025 awards) - -Overseas / Chinese Mainland - -Asia Pacific Sustainable Marketing Case Competition 2025 - -Third Place - -Bloomberg Global Trading Challenge 2023 - -CBS Case Competition 2023 - -CFA Institute Research Challenge 2025 - Asia Pacific Regional Final - -CFA Institute Research Challenge 2025 - Sub-Regional Final - -CGMA Global Business Challenge 2023 - North Asia Final - -Champions' Trophy Case Competition 2024 - -Grand Prize Winner -Regional Prize Winner: Asia & Oceania - -Winner - -Winner - -Winner - -Champion - -Finalist - -China Mainland, Hong Kong and Macao Digital Economy Innovation and -Entrepreneurship Competition 2024 - -2nd Runner-up (Gold Medal) - -Deloitte Digital Camp 2024 - -Deloitte Digital Camp 2023 - -Second Place - -Champion - -Deloitte Tax Championship 2024 - National Round - -Outstanding Team Award - -Deloitte Tax Championship 2024 - Regional Round - -Deloitte Tax Championship 2023 - National Round - -Deloitte Tax Championship 2023 - Regional Round - -Top 3 teams - -Merit Award - -Top 3 teams - -Family Office Association Hong Kong (FOAHK) 2023 Case Competition - -2nd Runner-up - -HSBC Global Private Banking Case Challenge 2023 - -Investing in the Future University Competition 2023 -by Pictet Asset Management - -John Molson Undergraduate Case Competition 2025 - -Lazaridis International Case Conference 2023 - -Marshall International Case Competition 2024 - -McDonough Business Strategy Challenge 2025 - -NHH International Case Competition 2024 - -Scotiabank International Case Competition 2023 - -Sydney International Business Competition 2025 - -Thammasat Undergraduate Business Challenge 2024 - -The Third KPMG ESG Case Study Competition - -UW Global Business Case Competition 2025 - -UW Global Business Case Competition 2023 - -1st Runner-up - -2nd Runner-up - -Finalist - -1st Runner-up - -Finalist - -First Place - -Champion - -Winner - -Third Place - -1st Runner-up - -Third Prize - -Champion - -Finalists - -27 - -26 Local - -Local - -100WF Hong Kong JumpStart University Investment Hackathon 2023 - -1st Runner-up - -HKSI Institute Case Competition 2024 - -Champion - -ACCA Hong Kong Business Competition 2023 - -Outstanding Team for start-up Filix -Medtech - -Team Excellence Award - -Merit Award for start-up MakeOmnia - -Alvarez & Marsal Hong Kong Business Case Competition 2024 - -Champion & Best Presenter - -HKSI Institute Case Competition 2023 - -Best Performer of Roma Risk Advisory -Limited - -1st Runner-up - -Merit - -Champion - -1st Runner-up - -Merit & Finalist - -Semi-finalist - -Champion - -2nd Runner-up - -Best Presenter - -1st Runner-up - -HKU-Airstar Bank Joint University Case Competition 2024 - -HKU-SaSa Joint University Case Competition 2023 - -Hong Kong MBS Digital Economy Business Simulation Competition 2025 - -Third Place - -HSBC Global Private Banking Case Challenge 2024 - -HSBC Life Insurance Competition 2025 - -HSBC Life Insurance Innovation Competition 2023 - -1st Runner-up - -Top 3 Winner - -1st Runner-up - -Best Use of Technology Award - -International MBS Business Management Simulation Competition 2023 - -Champion - -ISM-HK Student Case Competition 2025 - -ISM-HK Student Case Competition 2024 - -MBS Simulated Business Management Competition 2024 - -New Century Elite Development Programme 2023 - -Venture Capital On Campus (VCOC) - Investment Challenge 2023 - -1st Runner-up - -Champion - -Third Place - -First place - -Best presenter - -Third Place - -Champion - -1st Runner-up - -2nd Runner-up - -Winner Team - -29 - -AmplifyME APAC M&A APAC Challenge 2025 - -Third Place - -BOCHK Challenge 2023 - -Bronze Award for Innovative Design - -CGMA Global Business Challenge 2025 (Regional Final) - -1st Runner-up - -CGMA Global Business Challenge 2023 (Semi-final Round) - -Deloitte USHINE Case Competition 2024 - -ESG Challenge 2023 - -EY Young Tax Professional of the Year 2023 - Hong Kong Region - -Champion - -Finalist - -Champion - -Champion - -1st Runner-up - -2nd Runner-up - -Family Office Association Hong Kong (FOAHK) 2025 Case Competition - -Champion - -HKCGI Corporate Governance Paper Competition and -Presentation Awards 2024 - -HKGCC Business Case Competition 2023/24 - -HKGCC Business Case Competition 2024/25 - -28 - -HKGCC Pitch Perfect Programme 2024 - -HKICPA Business Case Competition 2023 - -HKSI Institute Case Competition 2025 - -1st Runner-up - -1st Runner-up (Paper Writing Competition) - -1st Runner-up (Paper Presentation -Competition) - -Audience’s Favourite Team Award - -Chevalier - Champion - -Chinachem Group - Champion - -Swire Pacific - Champion - -Chevalier - 1st Runner-up - -Chinachem - 1st Runner-up - -HKGCC - Champion - -Jardine Matheson - Champion - -SOCAM - Champion - -Jardine Matheson - 1st Runner-up - -MTR - 2nd Runner-up - -Overall Champion - -Winner - -2nd Runner-up - -Champion - -1st Runner-up - -2nd Runner-up - -Merit - - Chhawchharia, Mihika Amit -HSBC - Summer Analyst program - -internship, - -I collaborated closely with various -Throughout my -departments of the bank and -interacted regularly with senior -management, gaining comprehensive exposure to the bank’s operations. -This experience allowed me to hone my professional skills, including -communication, leadership, financial analysis, and market expertise. I -was honored to be ranked as the top intern in my cohort and secured a -return offer to join HSBC as a full-time graduate analyst in their Global -Banking division. I am deeply grateful for the knowledge and support I -received at HKU, which enabled me to thrive in a prestigious global -institution alongside some of the brightest talents in the industry. - -Internship Experience - -Lee, Chung Hong Calvin -HSBC - Global Risk Management Intern - -Reflecting on my time at HSBC, I realise that I have grown on a -professional and personal level. I have developed a deeper passion for -the banking industry and a greater appreciation for the critical role that -risk management and compliance play. I wore many hats and gained -precious insights into the banking industry. I was responsible for -overseeing compliance risks across 32 markets in Asia Pacific. I am -confident that the lessons I have learned at HSBC will guide me every -step of the way. - -Hui, Matthew -HSBC - International Banking (ISB) Intern - Corporate Coverage - -The comprehensive exposure to Corporate Banking was invaluable. From -analysing credit risk for lending applications to understanding diverse -product offerings for MNC clients, each day brought new learning -opportunities. I'm truly grateful for the experience of shadowing and -supporting different RMs in the team. Together with the sharing and -introductions from product partners, I learned how to identify clients' -needs and gained perspective on effectively reaching out and pitching to -clients. This experience taught me how to view banking relationships -through a strategic lens - understanding not just immediate needs but -trends. The -also anticipating -professional mindset, strategic thinking, and collaborative spirit fostered -during this internship will undoubtedly shape my future career path. - -future requirements and market - -Nagar, Tanvi -HSBC - International Banking Intern - -immense - -My internship experience was nothing short of incredible. Not only did I -grow immensely in my technical and professional skills but also -underwent an -journey of personal growth. I received -unparalleled support from the team and my mentors and gained an -in-depth insight into the field of banking and finance in Hong Kong. I -formed meaningful connections and grew my network. This experience -has truly shaped me and continues to propel my drive to pursue a career -in Banking in the years to come. - -3130 Sharma, Seerat -KPMG - Financial Services Audit Intern -(COOP3000 Cooperative Education Programme) - -I had the privilege of contributing to audit engagements for leading -clients in the banking and insurance sectors, where I conducted -substantive testing, supported risk assessments and navigated audit -methodologies using advanced tools. This role sharpened my analytical -mindset, strengthened my collaboration skills and offered a front-row -view into the dynamic world of financial auditing. It’s been a truly -rewarding journey that has solidified my passion for a career in -professional services. - -Xu, Aijing -CMB International Capital Corporation - Wealth Management Intern - -During my internship, I actively engaged with the product team in -market analysis and client pitching focused on mutual funds. I gained -valuable insights into various asset classes and macroeconomic -trends. Analyzing mutual fund performances and participating in -client meetings allowed me to contribute to investment portfolio -strategies. Collaborating with -team enhanced my -communication skills and understanding of client needs. Overall, this -experience deepened my knowledge of wealth management and -strengthened my passion for finance. - -the sales - -Shi, Yu Cheung Karson -Bank of China - Credit Summer Intern (Risk Management) - -During the internship, the financial techniques and business knowledge -I learned from HKU Business School courses helped a lot in company -financial analysis. By applying knowledge taught in the financial -accounting module, I assisted in evaluating loan demand. Through -cross-departmental collaborations, I also realised how classroom -simulations mirrored real-world risk mitigation challenges. This -experience reinforced the value of integrating academic theories with -dynamic market practices. - -Hu, Wen Qing -PwC Assurance Practice Intern - -As an assurance intern in PwC, I assisted my seniors with sampling -financial statement line items, reconciling balances, and documented -relevant non-financial information. It showed me how demanding the -auditing industry was, and allowed me to set my expectations for a -career in auditing as an accounting student. - -Wirawen, Karen -HKU – Neuberger Berman Private Equity Internship Programme - -You never truly understand private equity until you’re immersed in it -and my internship at Neuberger Berman was deeply enriching. Coming -in with limited knowledge, I gained direct exposure by working on -various deals across sectors and geographies, including Japan, Korea -and Southeast Asia. From drafting memos for the -investment -committee to attending annual general meetings and taking notes on -calls, I progressed from simple tasks to meaningful responsibilities. The -seniors were welcoming, always open to teaching, providing feedback -and offering valuable career advice. - -Chung, Marcus Lok Sum -Morgan Stanley Industrial Placement Intern - -During my internship with Sales & Trading (Institutional Equity -Division), I had the amazing opportunity to rotate across trading -and structuring desks where I was exposed to the fast-paced -environment especially during periods of high market volatility. To -see how traders manage positions and mitigate risks first-hand -was incredibly insightful. I also contributed to constructing trading -strategies and seeing the tangible impact of our work was deeply -rewarding. Morgan Stanley definitely provided me with a broader -perspective and understanding of market dynamics, liquidity, -client interactions and cross-asset opportunities. - -3332 Our Student Scholars - -Chan, Kwan Yin Cheryl -Recipient of Louisa Cheang Inspiration Scholarship 2024-25 - -I would like to express my gratitude for the Louisa Cheang Inspiration -Scholarship, which has significantly contributed to my personal and -professional growth. In the past academic year, this scholarship enabled -in the full-time Registered Visiting Student -me to participate -Programme at St Catherine’s College, University of Oxford. This -immersive experience enhanced my legal studies through engaging -discussions with scholars and participation in mooting competitions, -sharpening my research and advocacy skills. Additionally, I plan to -utilise this scholarship to pursue overseas postgraduate studies, -allowing me to deepen my understanding of the complexities of law in -a global context. - -Hussain, Iman -Recipient of FBE Student Contribution and -Innovation Award 2024-25 / -HKU Business School Future Leader Scholarship 2024-25 - -As someone who believes in learning through real-world experience, -the FBE Student Contribution and Innovation Award and the HKU -Business School Future Leader Scholarship have been life-changing. -They not only enabled me to go on an exchange semester to Cardiff, -where I explored new cultures and perspectives but also shaped my -entrepreneurial journey. - -With this support, I’ve been able to work on building a social -enterprise that aligns with my passion for creating meaningful impact. -I’m incredibly thankful for these opportunities to grow and make a -difference. - -Chow, Yee Shun Eason -Recipient of the Jardine HKU Scholarship 2024-25 - -It is an incredible honour to have received the Jardine HKU Scholarship -which is enriching my university journey! This scholarship allowed me to -pursue my many goals – from attending international training camps as -a handball player to participating in overseas exchanges without -financial worries. I have had the privilege of undertaking a summer -internship at the Jardine Group, learning from seasoned professionals in -my future industry. The Jardine Foundation also hosted exciting -networking activities and site visits which helped me build lasting -connections with fellow scholars in a vibrant community. - -Shih, Pui Hei Iris -Recipient of the Hong Kong Jockey Club Scholarships - -Undergraduate Scholarship 2024-25 - -This scholarship has been transformative to my university life as it has -allowed me to fully embrace and explore the world beyond the -classroom. It provided the financial stability I needed to focus on my -studies and presented opportunities for me to participate in diverse -extracurricular activities. These experiences broadened my perspective -and opened me to different points of view. - -Nurgali, Adilzhan -Recipient of BOC Hong Kong - -One Belt One Road Scholarship 2024-25 - -It means a lot to me to be awarded the BOC Hong Kong - One Belt One -Road Scholarship. It was not -just about financial support but -recognition of the hard work I have put in. As an international student -soon heading for exchange, this award gives me the freedom to explore -new cultures and experiences without extra stress. I am truly grateful to -be part of such an inclusive and supportive community. - -3534 2026 JUPAS Admissions - -Admissions criteria (HKDSE) - -Local candidates applying on the strength of the Hong Kong Diploma of -Secondary Education (HKDSE) are required to fulfil the minimum entrance -requirements for HKU Business School programmes as shown below: - -Level 4 -Level 4 - -EngEng ≥ -Level 4 or above in English Language -ChinChin ≥ -Level 3 or above in Chinese Language1 -MathMath ≥ -Level 3 or above in Mathematics - -Level 3 -Level 3 - -Level 3 -Level 3 - -AA - -CSDCSD = -Grade “Attained (A)” in -Citizenship and Social Development2 - -2x - -Elective -Elective -Subjects -Subjects - -≥ - -Level 3 -Level 3 - -Level 3 or above in two Elective Subjects - -For BBA(BA), BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) -In addition to the minimum entrance requirements above, candidates applying for the BBA(BA), -BBA(Law)&LLB, BBA(IBGM), BSc(MAT) and BSc(QFin) programmes shall also obtain the following results in the -HKDSE examination: - -BBA(BA) / BSc(MAT) -≥ -Level 3 or above in one of the following subjects: - -Level 3 - -Biology - -BAFS - -Chemistry - -Economics - -ICT - -Physics - -BBA(BA) - -BSc(MAT) - -Level 3 or above in Extended Module 1 or 2 of Mathematics is also preferred by BBA(BA) / BSc(MAT) but not -strictly required. - -BBA(Law)&LLB / BBA(IBGM) / BSc(QFin) - -BBA(Law)&LLB - -BBA(IBGM) - -BSc(QFin) - -EngEng ≥ - -Level 5 -Level 5 - -Chin, Math -Chin, Math ≥ - -Level 4 -Level 4 - -EngEng ≥ - -Level 5 -Level 5 - -+ - -+ - -M1/M2 -M1/M2 ≥ - -Level 3 -Level 3 - -MathMath ≥ - -Level 4 -Level 4 - -Application - -For details of the 2026 JUPAS exercise, please visit www.jupas.edu.hk. -Programme Scoring Formula + CLUB SUCCESS STORY -For more information on the Programme Scoring Formula and Subject Weightings, please -visit the Admissions Office’s website at https://admissions.hku.hk/apply/jupas. +N O R B U R Y P A R K L T C -Note 1: Non-Chinese-speaking applicants will be considered on a case-by-case basis.Note 2: The result(s) of Liberal Studies at Level 2 or above achieved in previous sitting(s) will also be considered.Level 5 or above inEnglish LanguageLevel 4 or above inChinese Language1 & MathematicsLevel 5 or above inEnglish LanguageLevel 4 or above inMathematicsLevel 3 or above in Extended Module 1 or 2 of Mathematics3736 2026 Non-JUPAS Admissions +SITUATION -Selection process and entrance requirements +Norbury Park LTC was established in 1889 in the London Borough of Croydon. It +has four floodlit tarmac courts and a small management committee made up +entirely of volunteers. Club membership was at an all-time low so the committee +put together an action plan to significantly increase membership within a year. -Eligibility -(1) Local students applying for admission on international qualifications other than the +APPROACH -HKDSE examination; +The club recruited a volunteer Marketing and Communications Officer and decided +to run a Nature Valley Big Tennis Weekend (open day) to showcase the club’s +facilities to the local community and engage prospects. -(2) Non-local students with international qualifications other than Chinese Mainland's +Using free resources and content available on My Tennis Toolkit, they created a +Facebook event and advert with a call-to-action to sign up for taster sessions. -National Joint College Entrance Examination (NJCEE/Gao Kao). +OUTCOME -Application Procedures -All non-JUPAS applications should be submitted via the online application system by noon of -November 26, 2025 (Hong Kong Time) for the first-round evaluation. Scanned copies of the -following documents should also be uploaded by the application deadline. +£24 +spent on +Facebook +campaign -(1) Official prediction of grades or scores from your current high school (if available); +Reaching +~3000 target +audience (local +and interested in +tennis) -(2) All internal and public examination results during your senior secondary education; and +145 declared +engagements +on FB event +page -(3) Academic reference letter (optional) +Converted to 69 +bookings – +roughly 80% of +all total +attendees! -Late submission of supporting documents will delay the consideration of application. +Drove an overall increase in club membership by 10% -Shortlisted applicants for BSc(QFin) will be interviewed; optional interviews for other -programmes will be arranged if necessary. +13 -University and Programme Entrance Requirements -Due to the complex combination of education background of applicants, each application will -be assessed individually on its own merits. Non-JUPAS applicants are advised to refer to the -standards and requirements on the website of the Admissions Office of the University before -making an application: + MEASURING PERFORMANCE -(1) Admissions standards (International Qualifications or National Qualifications); +A Q U I C K L O O K A T F A C E B O O K A D M A N A G E R -(2) Programme specific subject requirements; and +#Tip: Don’t be intimidated by complexity. +There are tons of available resources online +that offer step-by-step instructions for setting +up a campaign. -(3) English language requirement +Source: Adspresso by Hootsuiite -In particular, please note that the expected lower boundary for admissions listed on the -website is provided for reference purposes only. Admissions is competitive in nature and -achieving the expected lower boundary does not guarantee admission to the programme. -The actual admission grade or score will be dependent on the overall quality of applications -received in the 2026 admissions exercise. +14 -Announcement of Result -Offer results will be released from December 2025 onwards. All application status will be -updated in the online application system once an admissions decision has been made. -Applicants should check their application status online regularly. + DO’S AND DON'TS -More Details -For further information about the application procedures and admissions -standards, -at -https://admissions.hku.hk/apply/international-qualifications. +F O R S O C I A L M E D I A M A R K E T I N G -the Admissions Office’s website +DO -please +Have a social media presence! Today’s consumers expect to be able to find +brands on social media -visit +Pick the right social platforms that best align with your brand, audience and your +marketing goals – whether it’s Instagram, YouTube, or Twitter -2026 Dual Degree Admissions +Take advantage of free resources and training available to you – including My +Tennis Toolkit, Facebook Blueprint, and Google Marketing Platform -Separate application is required for applying to the dual degree programmes, in addition to the -application under the JUPAS or International/Non-JUPAS Admissions Scheme. +Follow your competitors. See what others are doing well, and find out how you +can differentiate yourself and appeal to those audiences -For details of the dual degrees and the application procedures, please visit -the Admissions Office’s website at -https://admissions.hku.hk/programmes/university-collaborative-programmes. +Get creative! Use social media to find influencers or other unique ways to connect +and engage with your audience -3938 Develop Your Career +DON’T -Freshmen Sharing +…forget to publish content regularly. By maintaining your presence you’ll stay top of +mind with members when they are looking for a coach or venue -Upon graduation, over 82% of our students begin their careers in the commerce and industry sector. Others either -move on to further studies or explore opportunities with the civil service, education institutions and -community/social services. +…be afraid to launch a social media marketing campaign. You can start small and +build and refine as you learn -Career Prospects +15 -Companies/organisations that have offered full-time positions to our graduates include but not limited to the -following: + ADDITIONAL RESOURCES -Accenture Company Limited -Bank of China (Hong Kong) Limited -BDO International Limited -Bloomberg L.P. -Citibank (Hong Kong) Limited -Crédit Agricole Corporate and Investment Bank -Credit Suisse Group AG -Deloitte Advisory (Hong Kong) Limited -Deloitte Touche Tohmatsu Limited -Deutsche Bank -Ernst & Young -Goldman Sachs (Asia) LLC -Hang Seng Bank Limited -HKSAR Government -Hong Kong Monetary Authority -JPMorgan Chase & Co. -KPMG -Morgan Stanley -PricewaterhouseCoopers Limited -Standard Chartered Bank (Hong Kong) Limited -The Hongkong and Shanghai Banking Corporation -Limited (HSBC) -UBS Group AG - -Full employment -for the 19th - -consecutive year - -98.7% - -was the average salary of -our 2024 graduates - -$36,770 - -of our 2024 graduates have -received at least 2 job offers -upon graduation - -57% - -Kantbek, Yerassyl -BBA - -I chose HKU Business School because Hong Kong is -one of the world’s most competitive financial centres, -offering a dynamic environment to grow. HKU -combines -real-world -rigorous academics with -experiences, such as Asia-focused case studies and -competitions like HICS, which I plan to join in Year 2. I -am particularly drawn to programs in marketing and -finance, as well as opportunities for entrepreneurship -through the Techno-Entrepreneurship Core and -iDendron incubator. With a diverse student body and -global exchange options, I look forward to learning -from peers, connecting with mentors, and developing -skills to make a meaningful impact in business. - -The main reason I chose HKU, particularly HKU -Business School, lies in its global connections and -strong partnerships with other institutions and -respected organizations, which I believe will be -advantageous for future employment opportunities. -I expect to embrace challenges over the course of my -four years at the university, as these will be vital for -my intellectual and social maturity. The first time I -visited the campus, I was amazed by the diversity of -the students and -facilities, which -demonstrated that the university is truly a place for -everyone. One of the many great opportunities -offered here is the chance to travel for studies, which -I would rank among my highest interests. - -the open - -Pernice, Audrey Yiwen -BEcon&Fin - -4140 The main reason I chose BBA(IBGM) is its flexibility in allowing -students to choose a second major, as well as the numerous -overseas exchange opportunities. I am also fortunate to have -been admitted as a member of the HKU International Case -Society (HICS), where I get to know people from other programs -who are also interested in the realm of business. HKU Business -School provides me with the perfect platform to strike a balance -between my work and life. - -Ho, Lik Hang Gordon -BBA(IBGM) - -Lo, Hoi Tung Reacker -BSc(MAT) - -I chose HKU Business School for its professional background and -diverse opportunities. The multidisciplinary programme allows me to -pursue Marketing Analytics and Technology, aligning with my varied -interests in business and STEM. As a freshman, I am eager to embrace -exciting experiences such as internships and exchanges. I look -forward to gaining insights from fellow students and professionals -through upcoming career events. With the strong support from HKU, -I am committed to developing my entrepreneurial skills and pursuing -my dreams without regrets! - -Drawn by HKU’s global prestige and the excellence of HKU Business -School, I chose to pursue Quantitative Finance at HKU to deepen my -passion for quantitative trading and data-driven decision-making. -The school’s flexible curriculum empowers me to explore -complementary fields such as computer science, statistics, and risk -management. As a target school for top investment banks, HKU -also offers unparalleled career exposure. HKU provides the ideal -environment to grow academically and professionally, and I am -excited to explore the many opportunities ahead. - -Tsang, Chun Shing Patrick -BSc(QFin) - -Fung, Yat Hang Rex -BFin(AMPB) - -My aspiration for BFin(AMPB) was ignited early in high school by its -unparalleled, practitioner-focused design, a curriculum that truly -bridges finance theory with industry practice. This conviction was -solidified through the heartfelt guidance of our Programme -Director, Professor Anna Wong, who mentored me even before I -officially joined HKU, while I was an apprentice at a US bank. - -Now, experiencing BFin(AMPB) firsthand, the transition from the -professional world to the classroom feels seamless. It is this -powerful integration of theory and real-world practice that makes -AMPB incomparable. HKUBS has instantly instilled in me a sense of -ownership, this is not just a school, but a launchpad for impact, and -I am thrilled to begin the next chapter of my growth here. - -I was drawn to BBA(IBGM) because of HKU’s -world-renowned academic reputation and its vibrant, -international community. I believe a truly “global -business” education extends beyond the classroom, -through experiential learning opportunities, daily -interactions with local and non-local peers, and -participation in hall and student society cultures, I will -gain a more dynamic and holistic multidisciplinary -perspective on the business world. Over the coming -four years, I am eager to harness these opportunities -to become a well-rounded leader and effective -problem-solver for my future career on the global -business stage. - -Yan, Sung Hei Christopher -BBA(IBGM) - -4342 Our Global Presence - -Israel - -Tel Aviv - -Beijing - -Shenzhen - -Hong Kong - -Vietnam - -Ho Chi Minh City - -Useful Websites -The University of Hong Kong (HKU) - -Undergraduate Admissions -https://admissions.hku.hk/ +H E L P F U L L I N K S T O G E T Y O U S T A R T E D -HKU – Entrance Requirements +LTA Marketing Guides and materials: +https://www.lta.org.uk/workforce-venues/tennis-venue-support/membership-growth-and- +retention/membership-growth/ -For JUPAS Candidates -https://admissions.hku.hk/apply/jupas +My Tennis Toolkit (includes creative assets for social media): +https://www.lta.org.uk/globalassets/venue/my-tennis-toolkit-usage-guide.pdf -For non-JUPAS Candidates -https://admissions.hku.hk/apply/international-qualifications +Facebook Blueprint - free online courses for business marketing: +https://www.facebook.com/business/learn -HKU Business School +Instagram Advertising: +https://business.instagram.com/advertising -Overview of Undergraduate Programmes -https://ug.hkubs.hku.hk/programme +Twitter Advertising: +https://business.twitter.com/en/advertising.html +https://blog.hootsuite.com/twitter-ads/ -Enrichment Programmes -https://ug.hkubs.hku.hk/student-enrichment +LinkedIn Advertising: +https://business.linkedin.com/marketing-solutions/ads +https://blog.hootsuite.com/linkedin-ads-guide/ -Other HKU Websites +Measurement / Tracking: +Google Analytics Academy: https://analytics.google.com/analytics/academy/ +Google Ad Manager: https://skillshop.exceedlms.com/student/path/34897-drive-advertising-revenue- +with-google-ad-manager -Mainland Affairs Office (MAO) -https://mainlandaffairs.hku.hk/ +16 -Centre of Development and Resources for -Students (CEDARS) -https://www.cedars.hku.hk/ + ANY QUESTIONS? -44 +THANK YOU -45 +17 diff --git a/q188/random_k4/random_3.md b/q188/random_k4/random_3.md index d3a8827a2c9c1661d68231635f74293c020f1a02..e93939cddbb26da056c6d4cb5d86060af787292b 100644 --- a/q188/random_k4/random_3.md +++ b/q188/random_k4/random_3.md @@ -1,1353 +1,979 @@ -CHAPTER 42 +INTRODUCING THE WINDOWS AZURE +PLATFORM -FORMWORK +DAVID CHAPPELL -42 FORMWORK +OCTOBER 2010 -Definitions +SPONSORED BY MICROSOFT CORPORATION -The following definitions are used in the forming -industry. Some terms may be used by other trades -as well, but their meanings may be different from -these depending upon the application. + CONTENTS -Falsework, in relation to a form or structure, means -the structural supports and bracing used to support -all or part of the form or structure until the concrete -is poured and is strong enough to support loads. +An Overview of the Windows Azure Platform ....................................................................................... 3 -Flying formwork is a designed floor formwork -system that can be hoisted between levels as a unit. +Windows Azure ........................................................................................................................................ 4 -Forms are the moulds into which concrete or -another material is poured. +SQL Azure ................................................................................................................................................. 6 -Formwork is a system of forms connected -together. +Windows Azure AppFabric....................................................................................................................... 8 -Gangforms are large panels designed to be hoisted -as a unit, and to be erected, stripped, and re-used. +Windows Azure MarketPlace .................................................................................................................. 9 -Knock-down forms are traditional formwork -supported by falsework and shoring, assembled -from bulk materials, used once, and then -dismantled. +A Closer Look at the Technologies ....................................................................................................... 10 -Panels are sections of form intended to be -connected together. +Windows Azure ...................................................................................................................................... 10 -Sheathing is the material directly supported by -wales, and against which concrete is to be placed. +Compute ............................................................................................................................................ 11 -Specialty formwork is designed specifically for a -particular structure or placing technique. +Storage .............................................................................................................................................. 12 -Struts are vertical members of shoring that directly -resist pressure from wales. +Fabric Controller ................................................................................................................................ 14 -Wales are horizontal members of shoring that are -placed against sheathing to directly resist pressure -from the sheathing. +Content Delivery Network ................................................................................................................. 14 -Introduction +Connect ............................................................................................................................................. 15 -In most cases, the formwork required for concrete -construction is built by carpenters. Shoring and -bracing support the forms that contain the wet -concrete. Formwork must also support the temporary -weight of material such as bundles of reinforcing -steel and live loads of workers and equipment. +SQL Azure ............................................................................................................................................... 15 -There are three stages in formwork operations: +Database ........................................................................................................................................... 15 -1. Assembly and erection +Reporting ........................................................................................................................................... 17 -2. Concrete placement +Data Sync .......................................................................................................................................... 17 -3. Stripping and dismantling. +Windows Azure AppFabric..................................................................................................................... 19 -To be done safely, each of these jobs requires -planning, knowledge, and skill from both -supervisors and workers. Design and planning -are a supervisory function that may also legally -require a professional engineer’s involvement. -Small construction and renovation jobs, however, -sometimes call for typical design work (e.g., -specific heights and sizes from the manufacturer), -which can be done on site by workers. +Service Bus ........................................................................................................................................ 19 -Where design drawings are provided, it is -important to construct the formwork as designed. -Any confusion regarding the design should be -approved by the designer. +Access Control ................................................................................................................................... 21 -If site conditions require changes or the design -does not seem to suit the situation, changes -and approval should also be obtained from the -designer. Formwork failures frequently involve -deviations from the original design that were done -without consulting the designer. They may also -involve human error. For these reasons, formwork -and shoring must always be inspected before -concrete is placed. +Caching .............................................................................................................................................. 23 -All large formwork installations in Ontario must be -designed by a professional engineer. But there are -always smaller jobs of moderate height or depth— -basements, footings, stairs—that may include -formwork with typical design arrangements from -a form tie manufacturer, which is installed and -constructed on the site. +Windows Azure Marketplace ................................................................................................................ 24 -Every carpenter should therefore know the type -of formwork needed and how to build, install, and -dismantle it safely. +Looking Ahead .................................................................................................................................... 26 -Formwork must always be constructed according -to good, safe, and sound carpentry practice. It must -have the following: +Conclusions ......................................................................................................................................... 26 -• Adequate braces and supports +About the Author ................................................................................................................................ 27 -• Reliable bearing surfaces, especially where wood +2 -structures are involved + AN OVERVIEW OF THE WINDOWS AZURE PLATFORM -• Adequate ties, bolts, or bracing to prevent +Using computers in the cloud can make lots of sense. Rather than buying and maintaining your own +machines, why not exploit the acres of Internet-accessible servers on offer today? For some applications, +both code and data might live in the cloud, where somebody else manages and maintains the systems +they use. Alternatively, applications that run inside an organization—on-premises applications—might +store data in the cloud or rely on other cloud infrastructure services. However it’s done, exploiting the +cloud’s capabilities can improve our world. -movement or bulging. +But whether an application runs in the cloud, uses services provided by the cloud, or both, some kind of +application platform is required. Viewed broadly, an application platform can be thought of as anything +that provides developer-accessible services for creating applications or storing data. In the on-premises +Windows world, this includes technologies such as Windows Server and SQL Server. To let applications +exploit the cloud, a cloud application platform must also exist. -Because wood is relatively soft, it will crush under -heavy loads such as concrete when the bearing -surface of joists on stringers, or studs on wales, is -not adequate. +This is exactly what the Microsoft Windows Azure platform provides. It’s a group of cloud technologies, +each providing a specific set of services to application developers. Figure 1 shows its components. -Crushing can be avoided by increasing the bearing -area between members, using spreader washers -(Figure 42-1), or increasing the number of joists -or studs. - -Figure 42-1: Spreader Washer on -Wooden Wale System - -Tools and Techniques - -42-1 - - FORMWORK - -Hazards - -The following are the main hazard areas in -formwork operations. - -• Falls – They are the major hazard because they are -potentially fatal. Cramped work areas, inadequate -access, failure to install guardrails, failure to use fall -arrest systems, tools or material left underfoot, and -surfaces slippery from form oil can all lead to falls. -Ladders are also frequently involved in falls. All -workers in Ontario who may be exposed to a fall -hazard and who may use a fall protection system -must have working at heights training that has been -approved by the Chief Prevention Officer. (See -Chapter 18: Guardrails and Chapter 19: Personal -Fall Protection) - -• Dust and concrete – Blowing dust and flying - -concrete particles during the chipping or cleaning -of formwork can injure unprotected eyes. - -• Access equipment – Access equipment such as -ladders and scaffolds is involved not only in falls -but in slips, trips, and other accidents. Hazards -include ladders not tied off, workers carrying -materials while climbing, ladders obstructed at -top or bottom, scaffolds not completely decked -in, and scaffolds erected or dismantled without -fall protection. - -• Lighting – Inadequate lighting can create or - -aggravate hazards when workers install or strip -forms in dark areas or place concrete at night. - -• Materials handling – The activity most frequently - -Injuries - -connected with injury. Improper or excessive -materials handling can result in sprains, strains, -and overexertion in shoulders, arms, and back, as -well as bruises, abrasions, and crushed fingers. - -• Struck against – Common because formwork - -operations are constantly changing and involve -the movement of heavy, awkward, and pointed -components. Wales, beams, panels, snap-ties, -nails, bolts, and rebar can cause punctures, cuts, -contusions, and abrasions. - -• Struck by – Another common cause of injury. -Rebar, formwork panels, concrete buckets, -and other material hoisted overhead can strike -workers. Struck-by injuries can also be caused by -hammers, pry bars, stakes, wedges, and material -such as joists and panels during stripping. - -• Electrical contact – Power tools, extension cords, -and temporary supply and wiring systems, used -under less-than-ideal conditions – mud, ground -water, wet excavations, fresh concrete – can lead -to ground faults, shortcircuits, and shock hazards. -Ground fault circuit interrupters are legally -required for portable tools used outdoors or in -wet locations. - -• Collapses – Even with advanced methods of - -design and installation, there is always the risk -that formwork, slabforms, wall forms, and other -large components can come loose, slip out of -place, or fall over, striking or crushing workers -underneath. - -• Health hazards – The spraying of form oils and - -curing compounds can irritate the lungs. Contact -with these chemicals can irritate the skin, leading -to redness, inflammation, or dermatitis. The -same conditions can result from the abrasive/ -corrosive effect of skin contact with concrete or -cement, especially when inadvertently left inside -boots all day. - -• Environmental conditions – Ice, snow, and rain -create slippery conditions. Wind can be a major -hazard. Handling sheets of plywood becomes -more difficult, panels may require more bracing, -and hoisting gets harder, especially with large -panels or tables. - -Formwork hazards can lead to the injuries -described below. Follow the safety tips to prevent -such injuries. - -• Eye injuries – These are quite common in - -formwork operations. Most result from particles -of wood or concrete that fall or are blown into -the eye during chipping and cleaning. The injuries -may not be severe but most can be prevented -by wearing eye protection. It is strongly -recommended that everyone on site wear eye -protection at all times. - -• Cuts, scrapes, punctures – The manhandling - -necessary to install and strip formwork can lead -to cut hands, arms, and legs, as well as pinched -or crushed fingers. Gloves help to prevent -injuries from rough or sharp edges on formwork -components. But workers must also have the -knowledge, skill, and physical ability necessary -for safe materials handling. That means knowing -your limitations and asking for help when needed. -Formwork involves protruding objects such as -nails, snap ties, conduit, and bolts that can give -you cuts and punctures. Where possible, these -objects should not be left sticking out or should -be covered over. - -• Back injuries – These injuries are frequently - -related to materials handling. The most important -preventive measure is back care. Exercise -programs, warm-ups before work, and knowing -your limitations can help to prevent sprains and -strains. Wherever necessary, get help or use -dollies, carts, or other mechanical devices. - -• Ankle sprains and fractures – Working in -close quarters, stepping over debris and -material, climbing into excavations, turning with -awkward loads, jumping down from scaffolds -or benches—these can lead to ankle and other -foot or leg injuries. Prevention starts with proper -housekeeping and materials handling. - -• Bruises and contusions – Handling formwork - -under rushed, cramped, or slippery conditions or -beyond a person's physical limitations can lead to -bruises. Bruises and contusions also result from -contact with protruding formwork components. - -42-2 - -Construction Health and Safety Manual FORMWORK - -More serious are contusions from falling formwork -materials. Formwork must be braced to ensure -stability, especially under windy conditions. Try -to avoid areas where work such as hoisting or -stripping is being done overhead. - -• Fall injuries – All of the injuries above, and many -others, can result from falls. Most falls are caused -by missing or inadequate guardrails, failure to use -fall-arrest equipment, failure to completely plank -scaffolds and other work platforms, and standing -or climbing on surfaces not meant to be used as -such – the tops of wall forms or 2 x 4 wales, for -example. Installing and stripping formwork often -requires the use of a fall-arrest system. - -• Falls also result from holes left unguarded or - -uncovered in formwork. These should be covered -up or fitted with guardrails as quickly as possible. -Where this cannot be done, the area should -be roped off and posted with warning signs to -prevent unauthorized entry. - -Planning - -Planning is the first and most important step in -reducing hazards and preventing injuries. - -Because formwork operations must often be -carried out in congested areas where other trades -are also working, planning is essential in making the -most of the time and space available to improve -safety and efficiency. - -Planning is a must for fall protection, work -platforms, material staging areas, housekeeping, -and material handling and movement. - -Planning should take place at every level from -manager to supervisor to worker. Planning labour, -materials, equipment, and work schedules to -meet design requirements is the responsibility of -management and supervision. But supervisors and -workers should come together to plan the details -of their assigned tasks based on the most effective -work methods and safety measures. - -For example, the normal weight of liquid concrete -is 150 lb per cubic foot (2,403 kg per cubic metre). -So a form designed to withstand 600 lb per square -feet (2929 kg per square metre) of pressure will be -able to support liquid concrete up to a height of 4 -feet (1.2 metres). - -600 lb/ft2 ÷ 150 lb/ft3 = 4 ft - -2929 kg/m2 ÷ 2403 kg/m3 = 1.2 m - -Other factors determine how long concrete -will remain liquid, such as temperature, slump, -vibration, and admixtures. For example, concrete -will set much more quickly in hot summer weather -than in cold winter weather. As a result, the same -form filled at the same pour rate may be subjected -to greater pressure in winter than in summer. - -Concrete pumping may cause additional pressure, -as well as vibration, on forms and must be -considered at the design stage. The action of the -pump sends surges of pressure through the piping -system which are often transmitted directly to -the forms, especially for narrow walls or columns. -Vibration may move the forms or loosen bracing, -ties, or spreaders. - -Pressure acts perpendicular to formwork surfaces -(Figure 42-2). This causes an outward thrust for -typical wall or column forms. However, it can also -cause uplift for battered or sloping forms. These -require hold-down anchors or tie-down braces. The -anchors will prevent the forms from lifting up or -floating on the concrete. - -Expansion Anchors Holding -Down Battered Form - -Design - -Safety and economy are the main factors in design. -Both have to be considered because adjustments in -one affect the other. - -For example, reducing the support structure for -wall forms in expectation of reduced pouring rates -should not be considered if the rate of pour is not -going to be controlled on the job. - -Fresh concrete exerts a pressure on formwork -similar to liquids. Concrete starts to set when -poured. So if the pour rate is slow, the maximum -formwork design pressure can be reduced, since -concrete at the bottom will be set before concrete -at the top is poured. Similarly, if the forms are -filled to the top immediately, they must be able to -withstand the pressure of the full liquid head. - -Figure 42-2: Pressure of Concrete on -Vertical (left) and Battered Formwork - -Consider using bracing systems and spreaders for -wall forms. Concrete filling the bottom of the form -may cause forces at the top to push the two sides -together unless they are properly braced and/ -or separated with spreaders. Formwork has to be -designed to resist such forces. During pouring, -ensure that spreaders are not removed until concrete -has reached at least two-thirds of the form height. - -Where box forms are used—for instance, on one- -piece covers for open-cut tunnels—you must use -bracing against the side thrusts caused by the -uneven pouring rates of the walls. Resisting these -forces requires that the system be tied together -and securely braced (Figure 42-3). - -Tools and Techniques - -42-3 - - FORMWORK - -Figure 42-3: Open-Cut Tunnel Formwork with -Bracing and Spreaders on Each Side - -Formwork should be designed and constructed -with stripping and removing as well as pouring in -mind. On wooden forms, crush plates or filler strips -should be used at corners such as slab-and-column -or slab-and-wall intersections (Figure 42-4). The -plates or strips are easily removed with a wrecking -bar and, once removed, make the stripping of -adjacent panels much easier. - -Crush Plates - -Figure 42-5: Formwork Panel - -A formwork panel or wall form to be lifted as a -single unit must be designed to withstand the -loads and forces exerted by hoisting (Figure -42-6). In most cases, this means designing a more -substantial structure. Fastening components may -also need more attention at the design stage. For -example, simple nailing may not be enough to hold -plywood sheets. - -Figure 42-4: Crush Plates - -The strips should be big enough to leave space at -the edges of the panels to accommodate wrecking -bars. - -When formwork has to be manhandled during -assembly or dismantling, the design should ensure -that the components are manageable. Formwork -panels are not only heavy but awkward (Figure -42-5). Realistic design demands consideration of -the size as well as the weight of panels. - -Figure 42-6: Formwork Lifted as Single Unit - -Special attention must also be applied to the design, -construction, and use of pick points for hoisting. The -strongbacks and wales must be securely attached -to the formwork. The pick points must be located -so that the panel hangs properly during installation, -concrete placement, and removal. - -42-4 +Figure 1: The Windows Azure platform supports applications, data, and infrastructure in the cloud, +together with a cloud marketplace. -Construction Health and Safety Manual FORMWORK +The Windows Azure platform today has four parts: -Types of Formwork + Windows Azure: A Windows environment for running applications and storing data on computers in -Below Grade +Microsoft data centers. -The first concern with formwork below grade is -the stability of the excavation walls. Walls must be -either shored or sloped according to soil type as -defined by section 226 of the Construction Projects -regulation (O.Reg. 213/91). Figure 42-7 shows -typical slopes. - -Good Soil - -Fairly -Good Soil - -Max 1.2m +SQL Azure: Relational data services in the cloud based on SQL Server. -(4 ft.) + Windows Azure AppFabric: Cloud-based infrastructure services for applications running in the cloud -Type 1 & 2 Soil +or on premises. -Type 3 Soil +3 -Poor Soil + Windows Azure Marketplace: An online service for purchasing cloud-based data and applications. -Type 4 Soil - -Figure 42-7: Excavation Slopes for Soil Types - -In most cases the shoring must be designed by an -engineer. Engineers may also specify slopes for -excavations. In both instances the design drawings -must be kept on the project. +All four of these components run in Microsoft data centers located around the world: two in North +America, two in Europe, and two in Asia. Developers using the platform can control which data center +runs their applications and stores their data, giving them the ability to place both closer to their users. -Excavations should be kept essentially dry. Water -should be pumped out. Mud should be cleared off -and replaced by compacted granular material in -work areas and on surfaces where concrete will be -placed. Mud presents slipping hazards and can lead -to inferior construction if not removed or replaced. +Each part of the Windows Azure platform has its own role to play. This overview describes all four, first at +a high level, then in a bit more detail. The goal is to provide a big-picture introduction to this cloud +platform. -Since mud has to be removed before concrete -is placed, it might as well be removed before -formwork is constructed, thereby reducing slipping -hazards at both stages. +WINDOWS AZURE -Water and mud also contribute to electrical -hazards. Grounding and insulation must be -effective and intact. Ground fault circuit -interrupters (GFCIs) are required by law on all -portable tools used outdoors or in wet locations. +At a high level, Windows Azure is simple to understand: It runs Windows applications and stores data in +the cloud. Figure 2 shows its components. -Formwork for footings and grade walls frequently -begins before excavation in the area is complete. -Trucks and excavating equipment put workers on -foot at the risk of being struck down or run over. +Figure 2: Windows Azure provides compute and storage services in the cloud. -Wherever possible, formwork operations should -be roped off from other work such as excavation -or pile-driving (Figure 42-8). Separate access -ramps for vehicles and workers are strongly -recommended. Stairs are an even better alternative -for personnel on foot. +The five parts of Windows Azure today are the following: -Figure 42-8: Formwork Roped-Off -from Other Operations + Compute: The Windows Azure compute service runs applications on a Windows Server foundation. +These applications can be created using the .NET Framework in languages such as C# and Visual Basic, +or they can be built without .NET in C++, Java, and other languages. Developers can use Visual Studio +or other development tools, and they’re free to use technologies such as ASP.NET, Windows +Communication Foundation (WCF), and PHP. -Mud sills must be used to support any shoring or -bracing that rests on soil in the excavation (Figure -42-9). The sill must bear on the soil throughout its -length. Sills should not be used to bridge holes or -irregular surfaces. To ensure uniform bearing, soil -should be levelled before sills are set in position. +Storage: This service allows storing binary large objects (blobs), provides queues for communication +between components of Windows Azure applications, and even offers a form of tables with a simple +query language. (Windows Azure applications that need traditional relational storage can also use -Figure 42-9: Mud Sills Under Shoring Frames +4 -The soil must have the capacity to bear whatever -loads are applied (Table 42-1). This information -may or may not be on the design drawings. + SQL Azure.) Both Windows Azure applications and on-premises applications can access the Windows +Azure storage service, and both do it in the same way: using a RESTful approach. -Table 42-1: In Situ Bearing -Pressure for Dry Soil Conditions -(Conservative Estimates) +Fabric controller: As the figure suggests, Windows Azure runs on a large number of machines. The +fabric controller’s job is to knit the machines in a single Windows Azure data center into a cohesive +whole. The Windows Azure compute and storage services are then built on top of this pool of +processing power. - Silt and clay - Sands - Gravelly sands - Gravel + Content delivery network (CDN): Caching frequently accessed data closer to its users speeds up +access to that data. The Windows Azure CDN can do this for blobs, maintaining cached copies at sites +around the world. -1,200 lb/ft2 (5,859 kg/m2) -4,000 lb/ft2 (19,530 kg/m2) -6,000 lb/ft2 (29,295 kg/m2) -8,000 lb/ft2 (39,059 kg/m2) + Connect: It’s often useful for organizations to interact with cloud applications as if they were inside +the organization’s own firewall. Windows Azure Connect allows this, making it easier for, say, a +Windows Azure application to access an on-premises database. -Tools and Techniques +Running applications and storing data in the cloud can have clear benefits. Rather than buying, installing, +and operating its own systems, for example, an organization can rely on a cloud provider to do this for +them. Also, customers pay just for the computing and storage they use, rather than maintaining a large +set of servers only for peak loads. And applications written for Windows Azure can scale better, be more +reliable, and require less administration than those written using the traditional Windows Server +programming model. -42-5 +To create, configure, and monitor applications, Windows Azure customers can use a browser-accessible +portal. A customer logs in with a Windows Live ID, then chooses whether to create a hosting account for +running applications, a storage account for storing data, or both. Microsoft then charges each customer +based on how much compute time, storage, and bandwidth that customer uses. How an application +charges its own customers—if it charges them at all—is entirely up to the people who create that +application. - FORMWORK +Windows Azure is a general platform that can be used in a broad set of scenarios. Here are a few +examples: -Soil that supports bracing or shoring should -be compacted and qualify as good soil at least -(cohesive, hard, with no water). Professional advice -from a geotechnical engineer may be required -for heavy structures such as elevated equipment -supports shored at or below grade. + An independent software vendor (ISV) creating a software-as-a-service (SaaS) version of an existing +on-premises Windows application might choose to build it on Windows Azure. Because Windows +Azure mostly provides a standard Windows environment, moving the application’s business logic to +this cloud platform won’t typically pose many problems. And once again, building on an existing +platform lets the ISV focus on business logic—the thing that makes them money—rather than +spending time on infrastructure. -Formwork in these situations is frequently built in -place. Planning is required to store material and -equipment out of the way, dispose of scrap and -debris, and ensure safe, efficient access (Figure -42-10). Because conditions are often cramped and -scrap accumulates quickly, it is important to clean -up as work proceeds. + An enterprise creating an application for its customers or employees might choose to build it on +Windows Azure. Because Windows Azure supports .NET, developers with the right skills aren’t +difficult to find, nor are they prohibitively expensive. Running the application in Microsoft data +centers frees the enterprise from the responsibility and expense of managing its own servers, turning +capital expenses into operating expenses. And especially if the application has spikes in usage— +maybe it’s an on-line flower store that must handle the Mother’s Day rush—letting Microsoft +maintain the large server base required for this can make economic sense. -Starting the Form +5 -Setting up the first form is always hard, heavy, -manual work. It requires enough workers to do -the job without overexertion or injury. Whenever -possible, use mechanical assistance such as a boom, -truck, or crane for this operation. It will reduce the -risk of injury to workers and increase productivity. + A start-up creating a new Web site—the next Facebook, say—could build its application on Windows +Azure. Because this platform supports both Web-facing services and background processes, the +application can provide an +interface as well as executing work for users +asynchronously. Rather than spending time and money worrying about infrastructure, the start-up +can instead focus solely on creating code that provides value to its customers and investors. The +company can also start small, incurring low costs while its application has only a few users. If the +application catches on and usage increases, Windows Azure can scale the application as needed. -Temporary bracing (Figure 42-12) is needed from the -start and at every step to prevent wind loads from -toppling the forms—historically the cause of many -serious injuries. A wind speed as low as 30 km/h can -create a dangerous wind load on formwork. - -Figure 42-10: Well-Planned Storage, -Access, and Setup +interactive user -Wall Forms - -Wall forms built in place are hazardous to -construct. Hazards include +These three examples illustrate the kinds of things organizations might do with Windows Azure, but +they’re not an exhaustive list. As interest in cloud computing continues to grow, expect to see a variety of +applications created for this cloud platform. -• Dowels sticking up from concrete slabs or +SQL AZURE -footings +Along with running applications, another attractive way to use the cloud is for storing data. SQL Azure +addresses this area, offering cloud-based services for relational data. As Figure 3 shows, SQL Azure today +includes three components. -• Unstable work surfaces and access created by +Figure 3: SQL Azure provides relational database services in the cloud. + +The components of SQL Azure today are the following: -poor planning +SQL Azure Database provides a cloud-based database management system (DBMS). This technology +lets on-premises and cloud applications store relational data on Microsoft servers in Microsoft data -• Manual handling of heavy material such as +6 + + centers. As with other cloud technologies, an organization pays only for what it uses, increasing and +decreasing usage (and cost) as the organization’s needs change. Using a cloud database also allows +converting what would be capital expenses, such as investments in disks and DBMS software, into +operating expenses. + +SQL Azure Reporting is a version of SQL Server Reporting Services (SSRS) that runs in the cloud. +Intended primarily for use with SQL Azure Database, it allows creating and publishing standard SSRS +reports on cloud data. + +SQL Azure Data Sync allows synchronizing data between SQL Azure Database and on-premises SQL +Server databases. It can also be used to synchronize data across different SQL Azure databases in +different Microsoft data centers. + +SQL Azure is built on Microsoft SQL Server. As with SQL Server, developers can create indexes and views, +use stored procedures, define triggers, and more. Applications can access SQL Azure data using Entity +Framework, ADO.NET, and other Windows data access interfaces. In fact, applications that today access +SQL Server locally will largely work unchanged with data in SQL Azure. Customers can also use on- +premises software such as SQL Server Analysis Services to work with their cloud-based data. + +While applications can use SQL Azure much as they do a local DBMS, the management requirements are +significantly reduced. Rather than worry about mechanics, such as monitoring disk usage and servicing log +files, a SQL Azure customer can focus on their data; Microsoft handles the operational details. And as with +other components of this cloud platform, customers use the common Windows Azure platform portal to +access its services. + +Applications might use SQL Azure in a variety of ways. Here are some examples: + + A Windows Azure application can store its data in SQL Azure. While Windows Azure provides its own +storage, relational tables aren’t among the options it offers. Since many existing applications use +relational storage and many developers know how to work with it, a significant number of Windows +Azure applications rely on SQL Azure to work with data in this familiar way. For example, a SaaS +application built on Windows Azure might create a separate SQL Azure database for each customer, +providing an intrinsically multi-tenant design. And to improve performance, customers can specify +that a particular Windows Azure application must run in the same data center in which SQL Azure +Database stores that application’s information. + + An application in a small business or a department of a larger organization might rely on SQL Azure. +Rather than storing its data in a SQL Server or Access database running on a computer under +somebody’s desk, the application can instead take advantage of the reliability and availability of cloud +storage. It can also create reports on this data using either SQL Azure Reporting or SSRS on-premises. +If the organization wishes to maintain an on-premises copy of the data as well, it can use SQL Azure +Data Sync to synchronize the cloud and on-premises databases. -plywood sheets, panels, wales, and buckets of -snap-ties, wedges, and plates +Suppose a manufacturer wishes to make product information available both to its dealer network +and directly to customers. Putting this data in SQL Azure would let it be accessed by applications +running at the dealers and by a customer-facing Web application run by the manufacturer itself. -• Slippery surfaces at and below grade - -• Inadequate design +7 -• Improper construction. + Whether it’s for supporting a Windows Azure application, making data more accessible, or other reasons, +data services in the cloud can be attractive. The goal of SQL Azure is to provide these services in a familiar, +usable way for cloud and on-premises applications. -The best protection against dowels is a wood -cover built of lumber at least 1 1/2 inches thick and -wired in place or protective caps placed over the -exposed ends of rebar (Figure 42-11). +WINDOWS AZURE APPFABRIC -Figure 42-12: Temporary Bracing at -Start of Form Construction +Running applications and storing data in the cloud are both important aspects of cloud computing. +They’re far from the whole story, however. It’s also possible to provide cloud-based infrastructure +services. Filling this gap is the goal of Windows Azure AppFabric. -Access to wall forms is not always given enough -thought. Forms more than 2 metres high will -require access platforms for workers involved in -concrete placing. The platforms can also be used -to complete the upper portion of the formwork. +The functions provided by AppFabric today address common challenges in building distributed +applications. Figure 4 shows its components. -An alternative is a frame scaffold, which can also -be used to install reinforcing steel (Figure 42-13). +Figure 4: Windows Azure AppFabric provides Windows Azure-based infrastructure services that can be +used by both cloud and on-premises applications. -Figure 42-11: Protective Covers -Over Dowels and Rebar +As the figure suggests, all of the components of Windows Azure AppFabric are built on Windows Azure +(although they don’t all provide services solely to Windows Azure applications). Those components are +the following: -42-6 +Service Bus: Exposing an application’s services on the Internet is harder than it might seem. The goal +of Service Bus is to make this simpler by letting an application expose endpoints in the cloud that can +be accessed by other applications, whether on-premises or in the cloud. Each exposed endpoint is +assigned a URI, which clients can use to locate and access the service. Service Bus also handles the +challenges of dealing with network address translation and getting through firewalls without opening +new ports for exposed applications. -Figure 42-13: Scaffolding for Access + Access Control: There are many ways for a user to get a digital identity today. The options include +Active Directory, Windows Live ID, Google Accounts, Facebook, and more. If an application wants to +let users log in with any of these, the application’s creator faces the daunting challenge of supporting -Construction Health and Safety Manual Fall arrest systems or scaffolds with guardrails -must be used where workers may fall more than -3 m (10 ft), or onto hazards such as projecting -dowels (Figure 42-14). In some circumstances, you -must use fall protection when the height is 2.4 m -(8 ft) or more. (See chapters on Guardrails and -Personal Fall Protection.) +8 -FORMWORK + this plethora of approaches. Access Control simplifies this by providing built-in support for all of them +(and more). It also provides a single place for defining rules to control what each user is allowed to +access. -Figure 42-15: Formwork Properly Tied and Braced + Caching: It’s common for applications to access the same data over and over. One way to speed up +this kind of application is to cache frequently accessed information, reducing the number of times +that application must query a database. The Caching service provides this—and the performance +boost it brings—for Windows Azure applications. -Recommended design pressures for various pour -rates and environmental conditions are set out -in CAN/CSA S269.1-16: Falsework and Formwork. -The standard defines a number of other design -considerations and should be consulted by field staff. +Microsoft has announced plans to add more services under the Windows Azure AppFabric banner, so +expect this list to grow in the not-too-distant future. -Slab Forms or Falsework Built in Place +As with Windows Azure and SQL Azure, customers use a browser-accessible portal to sign up for +AppFabric. Once this has been done, these services can be used in a variety of ways. Here are some of the +possibilities: -With slab forms built in place the major hazard is -falls. Injuries are also connected with the manual -handling of heavy materials and components. +Suppose an enterprise wished to let software at its trading partners access one of its applications. It +could expose this application’s functions through SOAP or RESTful services created using WCF, then +register those service endpoints with Service Bus. Its trading partners could then use Service Bus to +find these endpoints and access the services. -Forms built in place usually have to be taken -down in place after the concrete has hardened -sufficiently. This should be considered at the -construction stage. Stability may also be a -consideration where the structure is high, carries -heavy loads, and is placed on grade as in bridge -and overpass construction. +Imagine that the creator of this same application needs to let trading partners log in with a variety of +different identities. Rather than implementing support for these identities himself, he could use the +Access Control service to hide this complexity. -Fall protection is difficult to provide for workers -building slab forms in place. That’s why planning is -essential in the design and erection procedure. + A Windows Azure application created using ASP.NET might use the Session object to store per-client +state. By changing only a configuration setting, the application can cause this data to be kept in the +Caching service rather than, say, Windows Azure Storage tables. Doing this is likely to make the +application faster and more scalable. -Workers should wear a safety harness with the -lanyard tied off to the structure of the formwork -(Figure 42-16). This means tying off to the support -structure where shoring frame structures are being -constructed, tying off to a lifeline when placing -plywood panels at a leading edge, constructing -a guardrail at the edge of the formwork, or tying -off to the support structure when connecting it -together with tube and clamp. Don’t wait for the -structure to be completed before tying-off. +Along with the cloud-based services of Windows Azure AppFabric, Microsoft also provides an analogous +technology known as Windows Server AppFabric. As its name suggests, the services it provides run on +Windows Server—they support on-premises applications—rather than in the cloud. The on-premises +services aren’t exactly the same today as those in Windows Azure AppFabric (although Microsoft’s +announced plan is make the two congruent). Don’t be confused; throughout this paper, the name +“AppFabric” is used to refer to the cloud-based services. Also, don’t confuse Windows Azure AppFabric +with the Windows Azure fabric controller. Even though both contain the term “fabric”, they’re wholly +separate technologies addressing quite distinct problems. -Figure 42-14: Workers Wear Fall Arrest While -Attaching Wales to Threaded Rods +WINDOWS AZURE MARKETPLACE -Materials should be distributed along the work -location to minimize further handling. But traffic -and work areas must be kept clear for the safe -movement and installation of material. +In the on-premises world, many applications are purchased rather than custom-built. Many organizations +also rely on datasets from commercial providers. With the rise of the cloud, why not let customers find +and buy cloud applications and cloud-accessible data? Making this possible is the goal of Windows Azure +Marketplace. Figure 5 shows its two components. -Form Construction +9 -Wall forms must be constructed as designed. -The design must indicate clearly what is required. -Some wall forms are designed for specific concrete -placement rates expressed in metres of height per -hour (m/hr). A wall form in which concrete is -placed to a height of one metre in one hour would -have a placement rate of 1 m/hr. Slower pouring -rates result in lower formwork pressure because -the bottom concrete has started to set. + Figure 5: Windows Azure Marketplace provides an online marketplace for cloud applications and data. -Note: Using a concrete vibrator will increase the -liquid (hydraulic) pressure on the form in the -immediate area of the vibration. Over-vibrating the -concrete while increasing the pour rate can cause -the formwork to fail. +The two parts of Windows Azure Marketplace are: -Ensure that ties and braces are installed where -indicated on design drawings. Ties should be -snugged up. Braces should be securely fastened to -forms and wedged or fastened to a support that -will not settle or deform under load (Figure 42-15). + DataMarket (originally codenamed “Dallas”), providing a way for content providers to make datasets +available. Customers can browse the offerings, then purchase whatever they find useful. Applications, +both custom and off-the-shelf (such as Microsoft Excel), can then access this data through RESTful +requests or the OData protocol. -Formwork platforms must be: + AppMarket, providing a way for creators of cloud applications to expose those applications to -• Capable of bearing at least 50 lb/ft2 +potential customers. AppMarket will be available sometime after DataMarket. -• Adequately supported +Using the cloud to find, evaluate, and purchase data and applications makes sense. The goal of Windows +Azure Marketplace is to make these things easier to do. -• Equipped with guardrails +A CLOSER LOOK AT THE TECHNOLOGIES -• Secured at the level or levels where work such as +Having a broad understanding of the Windows Azure platform is an important first step. Making good +decisions require a deeper understanding of these technologies, however. This section takes a slightly +more in-depth look at each member of the family. -pouring and stripping will be done. +WINDOWS AZURE -Figure 42-16: Workers Tie Off to -Shoring Frames While Placing Stringers +Windows Azure provides what’s commonly called Platform as a Service (PaaS). Rather than offering a +cloud replica of the on-premises world, it offers a higher level of service that’s meant to make life easier +for both developers and administrators. What follows walks through the five components that +collectively provide this service. -Tools and Techniques +10 -42-7 + Compute - FORMWORK +An application built on the Windows Azure compute service is structured as one or more roles. When it +executes, the application typically runs two or more instances of each role, with each instance running as +its own virtual machine (VM). Figure 6 shows how this looks. -Make sure you have fall protection at all stages of -formwork construction. Wherever possible, cranes or -other equipment should be used to move material, -thereby reducing the amount of manual carrying, -lifting, and handling. +Figure 6: A running Windows Azure application consists of any combination of Web role instances, +Worker role instances, and VM role instances. -Shoring towers require special consideration. +A Windows Azure application today can be created using three kinds of roles: -• Towers must remain stable during construction -and dismantling. Guys or other bracing and -supports may be necessary to maintain stability -(Figure 42-17). + Web roles, intended primarily for running Web-based applications. Each Web role instance has +Internet Information Services (IIS) 7 pre-configured to run inside it, so creating applications using +ASP.NET, WCF, or other Web technologies is straightforward. It’s also possible to create applications +using PHP, Java, and other non-Microsoft technologies. -• If towers are to be tied together and braced -horizontally, this should be done as work -progresses (Figure 42-18). + Worker roles, designed to run a variety of code. A Worker role might run a simulation, for example, +or video processing or nearly anything else. It’s common for an application to interact with users +through a Web role, then hand tasks off to a Worker role for processing. -• Shoring towers and shores should be installed so + VM roles, which can run a user-provided Windows Server 2008 R2 image. A VM role can be the right -they are plumb to within 1/8 inch in 3 feet. +choice for moving some on-premises Windows Server applications to Windows Azure. -• Shoring towers should be snugged up under the +When a developer gives Windows Azure an application to run, she submits configuration information +along with it. Among other things, this information tells the platform how many instances of each role to +run. The Windows Azure fabric controller then creates a VM for each instance, running the code for the -stringers with adjustable base plates and U-heads -(Figure 42-19). +11 -• If frames do not ride tightly on top of one + appropriate role in each VM. And as Figure 6 indicates, requests from the outside world are load balanced +across all instances of a role. -another after tightening, one or more are out of -square and should be replaced (Figure 42-20). +This has an important implication: To be scalable, Windows Azure role instances shouldn’t maintain their +state themselves between requests. Because the load balancer doesn’t allow creating an affinity with a +particular role instance—there’s no support for sticky sessions—there’s no way to guarantee that multiple +requests from the same user will be sent to the same instance. Instead, any client-specific state should be +written to Windows Azure storage, stored in SQL Azure Database, or maintained externally in some other +way. -• With single-post shores, provide adequate lateral -bracing (Figure 42-21). Stairwells and balconies -are places where horizontal bracing for single- -post shoring systems may be required. +For role instances, Windows Azure lets developers choose from several VM sizes, each with a specific +number of processor cores and memory. Since each instance can be assigned one or more cores, +applications have predictable performance. And to increase the load an application can handle, its owner +or the application itself can request an increase in the number of running instances for one or more roles. +The Windows Azure fabric controller will then spin up new VMs for these instances and start them +running. It’s also possible to explicitly decrease the number of instances for a role, letting an application +grow and shrink as needed to handle changing loads. -Figure 42-19: Typical Shoring Tower with -Stringers, Adjustable Base Plates, and U-Heads +For developers, building a Windows Azure application looks much like building a traditional Windows +application. Microsoft provides Visual Studio project templates, for instance, to help developers create +Windows Azure applications. Also, the Windows Azure software development kit includes a version of the +Windows Azure environment that runs on the developer’s machine. Known as the Windows Azure +development fabric, it implements a local replica of the Windows Azure compute and storage services. A +developer can create and debug his application using this local simulacrum, then deploy it to Windows +Azure in the cloud when it’s ready. Windows Azure also provides other services for developers, such as +information about a running application’s CPU consumption, incoming and outgoing bandwidth, and +storage usage. -Replace this frame. +Storage -Figure 42-17: High Towers with Guys for Support +Applications work with data in many different ways. Sometimes, all that’s required are simple blobs, while +other situations call for a more structured way to store information. And in some cases, all that’s really +needed is a way to exchange data between different parts of an application. Windows Azure storage +addresses all three of these requirements, as Figure 7 shows. -Figure 42-20: Frame Bent Out of Shape +12 -Figure 42-18: Towers Tied -Together and Braced Horizontally + Figure 7: Windows Azure provides blobs, tables, and queues, all accessed in a RESTful style via HTTP, +HTTPS, or OData. -Figure 42-21: Single-Post -Shores with Lateral Bracing +The simplest way to store data in Windows Azure storage is to use blobs. As Figure 7 suggests, there’s a +simple hierarchy: A storage account can have one or more containers, each of which holds one or more +blobs. Blobs can be big—as large as a terabyte each—and to make transferring large blobs more efficient, +they can potentially be subdivided into blocks. If a failure occurs, retransmission can resume with the +most recent block rather than sending the entire blob again. Blobs can also have associated metadata, +such as information about where a JPEG photograph was taken or who the composer is for an MP3 file. + +Another way to use blobs is through Windows Azure drives, which can be mounted by a role instance. The +underlying storage for a drive is a blob, and so once a drive is mounted, the instance can read and write +file system data that gets stored persistently in a blob. + +Blobs are just right for some kinds of data, but they’re too unstructured for many situations. To allow +applications to work with data in a more fine-grained way, Windows Azure storage provides tables. Don’t +be misled by the name: These aren’t relational tables. In fact, even though they’re called “tables”, the +data they contain is actually stored in a set of entities with properties. A table has no defined schema; +instead, properties can have various types, such as int, string, Bool, or DateTime. And rather than using +SQL, an application can access a table’s data using the simple query language defined by OData. A single +table can be quite large, with billions of entities holding terabytes of data, and Windows Azure storage +can partition it across many servers if necessary to improve performance. -42-8 +Blobs and tables are both focused on storing data. The third option in Windows Azure storage, queues, +has a somewhat different purpose. A primary use of queues is to let Web role instances communicate +with Worker role instances. For example, a user might submit a request to perform some compute- + +13 + + intensive task via a Web page implemented by a Windows Azure Web role. The Web role instance that +receives this request can write a message into a queue describing the work to be done. A Worker role +instance that’s waiting on this queue can then read the message and carry out the task it specifies. Any +results can be returned via another queue or handled in some other way. + +Regardless of how it’s stored—in blobs, tables, or queues—all data held in Windows Azure storage is +replicated three times. This replication allows fault tolerance, since losing a copy isn’t fatal. The system +guarantees consistency, however, so an application that reads data it has just written will get what it +expects. + +Windows Azure storage can be accessed either by a Windows Azure application or by an application +running somewhere else. In both cases, all three Windows Azure storage styles use the conventions of +REST (and the OData protocol for tables) to identify and expose data. Everything is named using URIs and +accessed with standard HTTP operations, so clients can be created using .NET, Java, or other familiar +technologies. + +The Windows Azure platform charges independently for compute and storage resources. This means that +an on-premises application can use just Windows Azure storage, accessing its data in the RESTful way just +described. For example, a Windows Server application running in an enterprise data center might choose +to store backups in Windows storage blobs. + +Fabric Controller + +Even though a Windows Azure application runs in VMs, a developer doesn’t explicitly create and manage +those VMs. Instead, as described earlier, she just tells the platform how many instances the application +needs, and Windows Azure silently creates the necessary VMs and runs her application. This magic is +performed by the fabric controller, a fundamental aspect of Windows Azure. + +Along with creating VMs and starting the applications that run in them, the fabric controller also monitors +all running instances. A role instance might fail for many reasons; maybe the code threw an exception or +the VM crashed or the physical server it’s running on went down. Whatever the reason, the fabric +controller will start a new instance to maintain the total number of instances required for this application. + +The fabric controller also handles housekeeping tasks like patching the operating system and other +system software, which minimizes the need for administrators. These are all examples of why Windows +Azure is viewed as PaaS: It provides a platform that lets customers run applications without worrying +about administering the environment they run in. + +Content Delivery Network + +One common use of blobs is to store information that will be accessed from many different places. Think +of an application that serves up videos, for example, to Flash, Silverlight, or HTML 5 clients around the +world. To improve performance in situations like this, Windows Azure provides a content delivery +network. The CDN stores copies of a blob at sites closer to the clients that use it. This speeds up delivery +of frequently accessed content, improving performance for users of this information. + +14 + + Connect + +Windows Azure applications can interact with the outside world via HTTP, HTTPS, or TCP. But suppose +you’d like to connect a role in an application to a machine outside the cloud at the IP level. For example, +suppose an organization decides to move an existing ASP.NET application to Windows Azure, but wishes +to keep the application’s data in an on-premises SQL Server database. Windows Azure Connect allows +this. + +Using this option requires running Windows Azure Connect software on the on-premises machine that +contains SQL Server, and it also requires some straightforward configuration. (This configuration can be +done by the developer—there’s no need to get a network administrator involved.) Once this is done, all of +the instances in the Web role can behave as if they’re on the same IP network as the SQL Server machine. +In fact, they can use the same SQL Server connection string as when the application was running on +premises—there’s no difference. + +Note that this isn’t a full-fledged virtual private network (VPN). While Microsoft has announced plans to +let customers connect to Windows Azure via a VPN, this isn’t currently supported. It is possible to use +Windows Azure Connect to domain-join a Windows Azure application to an on-premises Active Directory, +however. Doing this allows single sign-on by on-premises users to the cloud application, and it also lets +the application use existing Active Directory accounts and groups for access control. + +The goal of application platforms, whether on-premises or in the cloud, is to support applications and +data. Windows Azure provides a home for both of these things. Going forward, expect to see a substantial +share of what would have been on-premises Windows applications instead running on this new cloud +platform. + +SQL AZURE + +A DBMS in the cloud is attractive for many reasons. Letting a specialized service provider ensure reliability +and perform essential management functions can make sense, especially for organizations that find these +things hard to do on their own. Data in the cloud can also be accessed by applications running anywhere, +even on mobile devices. And given the economies of scale that a service provider enjoys, using a cloud +database may well be cheaper than doing it yourself. + +SQL Azure provides these things, together with cloud-based reporting and data synchronization services. +And while all of them are important, the place to start is with the foundation: SQL Azure Database. + +Database + +SQL Azure Database is simple to understand. As seen by an application, it provides the core database +functions of SQL Server as a cloud service. Figure 8 shows the basics of this technology. + +15 + + Figure 8: Applications access data in SQL Azure Database through Microsoft’s TDS protocol or via OData. + +An application using SQL Azure Database might run on Windows Azure, in an enterprise’s data center, on +a mobile device, or somewhere else. Wherever it runs, the application typically accesses data via a +protocol called Tabular Data Stream (TDS). This is the same protocol used to access a local SQL Server +database, and so a SQL Azure Database application can use any existing SQL Server client. This includes +Entity Framework, ADO.NET, ODBC, PHP and others. And because SQL Azure Database looks like an +ordinary SQL Server system, standard tools can also be used, including SQL Server Management Studio, +SQL Server Integration Services, and BCP for bulk data copy. Also, as Figure 8 shows, applications can +optionally access SQL Azure Database using OData. + +Each SQL Azure account can have one or more logical servers. (Note that these provide a way to organize +the data and billing for an account—they aren’t actual instances of SQL Server.) Each server can then have +multiple databases, each of which can be up to 50 gigabytes in size. A user is free to use multiple +databases if required, spreading data across them for better performance or for other reasons. It’s also +possible to store a snapshot of one SQL Azure database into another, providing a simple backup +mechanism. + +For the most part, an application using SQL Azure Database sees a familiar SQL Server environment. A few +things are omitted in the technology’s current release, however, such as the SQL Common Language +Runtime (CLR) and support for full-text search. (Microsoft says that both will be available in a future +version.) Also, because administration is handled by Microsoft, the service doesn’t expose physical +administrative functions. A customer can’t shut down the system, for example, or interact directly with +the hardware it runs on. And as you’d expect in a shared environment, a query can run for only a limited +time—no single request can take up more than a pre-defined amount of resources. + +16 + + Yet while the environment looks quite standard, the service an application gets is more robust than what +a single instance of SQL Server provides. As in Windows Azure storage, all data stored in SQL Azure +Database is replicated three times for high availability. Also like Windows Azure storage, the service +provides strong consistency: When a write returns, the data has been made persistent. The goal is to +provide reliable data storage even in the face of system and network failures. + +Whether an application needs multiple databases or just one, SQL Azure Database can help developers +address a range of scenarios. Whatever problem is being solved, the technology’s fundamental goal +remains the same: to provide a familiar, reliable, and low-cost cloud database for many kinds of +applications. + +Reporting + +Storing data in SQL Azure Database can be useful. But as soon as that data exists, there’s likely to be a +demand for reports based on that data. Meeting this demand is the purpose of SQL Azure Reporting. +Based on SQL Server Reporting Services (SSRS), this technology provides a cloud-based way to create +reports. + +Given that SQL Azure Database looks like SQL Server to an application, it’s always possible to use SSRS on- +premises with the data it contains. So why bother to provide a cloud reporting service as well? SQL Azure +Reporting today targets two main scenarios: + + Reports created using SQL Azure Reporting can be published to a SQL Azure Reporting portal, letting + +users access it there, or made accessible directly via a URL. + + An ISV can embed reports published to the SQL Azure Reporting portal in any application, including +Windows Azure applications. This lets users access those reports without leaving the application. To +do this, the ISV can use the standard ReportViewer controls in Visual Studio—it’s no different from +embedding on-premises reports into applications. + +SQL Azure Reporting is designed to work with data stored in SQL Azure Database. Reports used with SQL +Azure Reporting are created on-premises, however, with Business Intelligence Developer Studio, the +Visual Studio-hosted tool that’s also used to create SSRS reports. In fact, SQL Azure Reporting relies on the +same reporting format, expressed in the Report Definition Language (RDL), used with SSRS. + +Like other aspects of SQL Azure, SQL Azure Reporting doesn’t provide all of the functions offered by SSRS +on premises. There’s currently no support for scheduling and subscriptions, for example, that let a report +be run and delivered at regular intervals. Still, cloud-based reporting services can be useful, and they’re an +important part of the SQL Azure technology family. + +Data Sync + +Storing data in SQL Azure Database can make it accessible to any application with an Internet connection. +Still, there are often situations where it makes sense to maintain a copy of this cloud-based data in some +other place. Suppose an organization needs to have an on-premises copy of the same information, for +example, for performance reasons or to ensure access if the network fails. For scenarios like this, the +ability to synchronize data in SQL Azure Database with other data stores is useful. + +17 + + It’s always possible to write synchronization code yourself using the Microsoft Sync Framework. To make +this easier, Microsoft provides SQL Azure Data Sync. Rather than requiring its user to write code, this +technology is entirely configuration-driven (although unsurprisingly, it’s built on the Microsoft Sync +Framework). As Figure 9 shows, SQL Azure Data Sync supports two options today. + +Figure 9: SQL Azure Data Sync can synchronize data between SQL Azure databases or between a SQL +Azure database and on-premises SQL Server databases. + +The two synchronization options are the following: + +Synchronizing data between a SQL Azure database and an on-premises SQL Server database. Having +an on-premises copy of cloud data might make sense for a variety of reasons. Maybe an organization +wants to make sure that this data is always available, for instance, even in the event of network +failures, or perhaps regulations require that a copy of the data always be available within a country’s +borders. And even though SQL Azure’s built-in data replication guards against hardware failures, the +data’s owner might choose to maintain an on-premises backup copy in case of administrative errors, +such as incorrect table deletions. + +Synchronizing data between SQL Azure databases in different Microsoft data centers. For example, +suppose an ISV or a global enterprise creates an application that’s used by people around the world. +To provide good performance for all of them, its creator might choose to run this application in three +different Windows Azure data centers: one each in North America, Europe, and Asia. If the +application stores data in SQL Azure Database, it might rely on SQL Azure Data Sync to keep +information synchronized across these three data centers. + +SQL Azure Data Sync uses a hub-and-spoke model. All changes are copied first to the SQL Azure database +hub, then out to the spokes (referred to as members). The members might be other SQL Azure databases + +18 + + or on-premises SQL Server databases. In both cases, the technology provides synchronization of an entire +database or just specific tables, with changes made to any copy propagated to all of the others. And while +it’s possible to initiate synchronization manually, SQL Azure Data Sync also includes a scheduling service. +This lets users schedule, say, hourly synchronization between a pair of databases. However it’s used, the +purpose is the same: providing a straightforward way to synchronize data between a SQL Azure database +in a Microsoft data center and one or more databases located somewhere else. + + WINDOWS AZURE APPFABRIC + +Applications exist because their business logic has value. Yet applications rely on infrastructure to provide +a foundation for that logic. A good application platform provides this infrastructure; application +developers shouldn’t need to write it themselves. + +Windows Azure AppFabric provides infrastructure for applications. The people who create applications +can benefit from different kinds of infrastructure, and so AppFabric contains various parts. This section +takes a closer look at Service Bus, Access Control, and Caching, the three components of Windows Azure +AppFabric today. + +Service Bus + +Suppose you have an application running inside your organization that exposes a Web service built with +Windows Communication Foundation (WCF). Suppose further that you’d like to connect this service +through the Internet to software running outside your organization. This client software might be running +on a cloud platform, such as Windows Azure, or inside another organization. + +At first glance, this can seem like a simple problem. Since your application provides its functionality +through Web services (either RESTful or SOAP-based), you can just make those Web services visible to the +outside world. When you actually try to do this, though, some problems appear. + +First, how can clients in other organizations find endpoints they can connect to for your service? It would +be nice to have some kind of registry where others could locate your application. And once they’ve found +it, how can requests from software in other organizations get through to your service? Network address +translation (NAT) is very common, so an application frequently doesn’t have a fixed IP address to expose +externally. And even if NAT isn’t being used, how can requests get through your firewall? It’s possible to +open firewall ports to allow access to your application, but network administrators frown on this. + +Service Bus addresses these challenges. Figure 10 shows how. + +19 + + Figure 10: A WCF service can register endpoints with Service Bus, then have clients discover and use +those endpoints to access the service. + +To begin, your WCF service registers one or more endpoints with Service Bus (step 1). For each registered +endpoint, Service Bus exposes its own corresponding endpoint (step 2). Service Bus assigns your +organization a URI root, below which you’re free to create any naming hierarchy you like. This allows your +endpoints to be assigned specific, discoverable URIs. + +When a client running in the cloud or on-premises at some other organization wishes to access your +service, it contacts the Service Bus registry (step 3) to find the endpoint, providing the endpoint’s URI. This +request uses the Atom Publishing Protocol, and it returns an AtomPub service document with references +to the endpoints Service Bus exposes on behalf of your application. Once it has these, the client can +invoke operations on the services exposed through these endpoints (step 4). For each request Service Bus +receives, it invokes the corresponding operation in the endpoint exposed by your WCF service (step 5). +(Although it’s not shown in the figure, Service Bus establishes a direct connection between an application +and its client whenever possible, making their communication more efficient.) + +There’s an obvious question here: How exactly does step 5 work? How does the request from Service Bus +back to your service deal with the challenges of NAT and firewalls? The answer is that in step 1, your +service opened a TCP connection with Service Bus for this exposed endpoint. Service Bus holds this +connection open, which solves two problems. First, NAT is no longer an issue, since traffic on the open +connection with Service Bus will always be routed to your application. Second, because the connection +was initiated from inside the firewall, there’s no problem passing information back to the application via +this connection—the firewall won’t block this traffic. + +20 + + Along with making communication easier, Service Bus can also improve security. Because clients now see +only an IP address provided by Service Bus, there’s no need to expose any IP addresses from within your +organization. This effectively makes your application anonymous, since the outside world can’t see it. +Service Bus acts as an external DMZ, providing a layer of indirection to deter attackers. + +While an application that exposes its services via Service Bus is typically implemented using WCF, clients +can be built with WCF or other technologies, such as Java. However they’re created, these clients can +make requests via TCP, HTTP, or HTTPS. Applications are also free to use their own security mechanisms, +such as encryption, to shield their communication from attackers. + +Service Bus provides a few more useful features, including these: + +Support for message buffers, which act like simple queues. Rather than require a client to make +direct calls to a service, the client can place a message of up to 256 kilobytes into a message buffer. +The message is persisted to disk—it’s durable—and the service can then read this message at some +later time. And as usual in the Windows Azure platform, the persisted messages are replicated to +guard against failures. + + The ability for multiple WCF services to listen on the same URI. Service Bus will then randomly spread +client requests across all of the listening services. The goal is to provide both load balancing and fault +tolerance for the WCF services. + +Exposing applications to the outside world isn’t as simple as it might seem. The intent of Service Bus is to +make implementing these interactions as straightforward as possible. + +Access Control + +Working with identity is a fundamental part of most distributed applications. The modern approach to +this, called claims-based identity, lets a user send a token full of claims that contain identity information. +One claim might contain the user’s name, for example, while another contains her age or a group she +belongs to. An application can use the claims in a token to decide what the user is allowed to do or in +other ways. + +In a claims-based world, tokens are issued by identity providers (IdPs). Some IdPs, such as Active Directory +Federation Services (AD FS) 2.0, exist inside organizations. Others, such as Windows Live ID and Google +Accounts, are accessible to anyone via the Internet. Each application can decide which IdPs it trusts and +thus which tokens it’s willing to accept. For example, an application running inside your company might +only accept tokens issued by your own AD FS server, while an application running on the Internet might +accept tokens issued by, say, Google and Facebook. + +Yet different IdPs use different token formats, and they represent claims in different ways. An application +that directly accepts identities from Google, Facebook, and Yahoo, for instance, would need to handle +these differences itself. But why do this? Why not instead create an intermediary that can generate a +single token format with a common representation of claims? Doing this would make life simpler for the +people who create applications, since they now need to handle only one kind of token. + +The Access Control service does exactly this, providing an intermediary in the cloud for working with +claims-based identity. Figure 11 shows how it works. + +21 + + Figure 11: The Access Control service makes it easier for applications to accept identity information +issued by different identity providers. + +As the figure shows, an application that relies on Access Control can run either on-premises or in the +cloud. In either case, the process begins when a user attempts to access this application via a browser +(step 1). The application then redirects the browser to an IdP whose token this application will accept. The +user authenticates herself with this IdP, such as by entering a username and password, and the IdP +returns a token containing claims about her (step 2). + +Next, the user’s browser sends the IdP token to Access Control (step 3). Access Control validates the +token, making sure that it really was issued by this IdP, then creates a new token according to whatever +rules have been defined for this application (step 4). Access Control contains a rules engine, allowing each +application’s administrator to define how tokens from various IdPs should be transformed into an Access +Control token. For example, if different IdPs use different formats for representing usernames, Access +Control rules can transform all of these into a common username string. Access Control then sends this +new token back to the browser (step 5), which submits it to the application (step 6). Once it has the +Access Control token, the application verifies that it really was issued by Access Control, then uses the +claims it contains (step 7). + +While this process might seem a little complicated, it actually makes life significantly simpler for the +creator of the application. Rather than handle diverse tokens containing various claims, the application +can accept identities issued by multiple identity providers while still receiving only a single token with +familiar claims. And rather than require each application to be configured to trust various IdPs, these trust +relationships are instead maintained by Access Control—an application need only trust it. + +22 + + As Figure 11 shows, Access Control has built-in support for several identity providers, including AD FS 2.0, +Windows Live ID, Google, Yahoo, and Facebook. It can also work with any IdP that supports OpenID. +Browsers and other clients can request Access Control tokens using either OAuth 2 or WS-Trust, and these +tokens can have various formats, including SAML 1.1, SAML 2.0, and Simple Web Token (SWT). And to +create applications that accept Access Control tokens, Windows developers can use Windows Identity +Foundation (WIF). (It’s worth pointing out, however, that nothing about Access Control is tied to +Windows—it could just as well be used by a Linux application that accepted only Google and Facebook +identities.) + +Working with identity is important in nearly every distributed application. The goal of Access Control is to +make it easier for developers to create secure applications that accept identities from diverse providers. +By putting this service in the cloud, Microsoft has made it available to any application running on any +platform. + +Caching + +For many applications, one of the most effective ways to improve performance is by caching frequently +accessed data. Since applications tend to use the same information over and over, making this +information more readily accessible can make the application faster. Doing this is the purpose of Windows +Azure AppFabric Caching. Figure 12 illustrates the idea. + +Figure 12: Windows Azure AppFabric Caching speeds up access to frequently accessed data for +Windows Azure applications. + +23 + + The Caching service provides a distributed cache for Windows Azure applications, along with a library to +access this cache. As Figure 12 suggests, the service includes a local cache that can keep a copy of recently +accessed data items in each of an application’s role instances. If a data item the application needs isn’t +found in the local cache, the cache library automatically contacts the shared cache provided by the +Caching service. As the figure shows, this cache is spread across a number of Windows Azure instances, +each holding different cached data. This diversity isn’t visible to applications using the cache, however. An +application just requests a data item, then lets the cache find it (if it’s in the cache) and return it from +whatever instance contains it. + +Recently accessed data isn’t automatically cached, however. Instead, the application explicitly inserts data +items in the cache, such as by using the Caching API. It’s also possible to configure an ASP.NET application +running on Windows Azure to store Session object data in the Caching service, speeding it up without +changing any of its code. + +Windows Server AppFabric, the on-premises analog to Windows Azure AppFabric, also provides a Caching +service. In fact, the two are very similar. The biggest different is that unlike its on-premises counterpart, +Windows Azure AppFabric Caching is a service—there’s no need to configure servers and administer the +cache. Instead, all of this is handled automatically by the service itself. And the cloud Caching service is +multi-tenant, so each application using it gets its own instance. Because the application must authenticate +itself to this instance, data held in the Caching service isn’t accessible to other applications. + +Caching can make applications faster and more scalable with little effort from developers, and so using it +makes sense. By providing caching as a service, Windows Azure AppFabric Caching makes this easier to +do. + +WINDOWS AZURE MARKETPLACE + +Cloud platforms are useful, but they’re just a means to an end. The real goal is offering applications and +data that provide value for people. Since the Windows Azure platform provides a foundation for both, +why shouldn’t it also include a way for people to find the applications and data that they need? + +The Windows Azure Marketplace is designed to address this need. Its two components, AppMarket and +DataMarket, let users find, try, and purchase applications and data, respectively. While both are +important, DataMarket will appear first, and so what follows focuses on this aspect of the Marketplace. + +Buying applications is common today—nearly every organization does it. Buying data is less common, but +it’s no less important. Many firms sell many different kinds of data, including demographic information, +financial information, legal information, and much more. Yet using purchased data typically requires +figuring out whether the data you need exists, finding a firm that offers it, then determining whether their +data meets your needs. -Construction Health and Safety Manual FORMWORK +All of this is more difficult than it needs to be. Why not create one place where customers can find all +kinds of data from all kinds of content providers? Why not let them examine the data to make sure it +meets their needs, then purchase what they need right then and there? The DataMarket component of +Windows Azure Marketplace was created to do these things. Figure 13 shows its main components. + +24 + + Figure 13: Built on Windows Azure, the DataMarket allows access to content partner datasets stored in +Windows Azure, SQL Azure, or externally. + +As the figure suggests, both people and applications can access information through DataMarket. Using a +Windows Azure-based application called the Service Explorer, a user can see what datasets are available, +then purchase what she needs. Once this is done, applications can access that data using RESTful or OData +requests. The datasets made available by DataMarket can be stored on the Windows Azure platform itself +using Windows Azure Storage or SQL Azure Database. They can also be stored externally, such as in a data +center owned by the content provider. There’s no requirement to put everything in the cloud. + +For customers, DataMarket provides a single place to find, buy, and access a variety of commercial +datasets. For content providers—the owners of those data sets—DataMarket offers a chance to expose +their offerings to more customers through Microsoft’s cloud platform. While content providers set prices +for their data, DataMarket provides a billing service that frees them from dealing directly with customers. +Microsoft will also vet the quality of the content providers, limiting DataMarket initially to just the top five +providers in a particular industry. -Using forms designed for typical floors in non- -typical situations has resulted in serious accidents. -Before using any flying form under non-typical -conditions, consult the designer or manufacturer. -Wall forms should not be extended in height or -width, for instance, or slab panels cantilevered -without the review and approval of a professional -engineer. Such situations usually occur with -penthouses or mechanical rooms where wall and -ceiling heights are greater than for typical floors. +Applications are free to use purchased data in any way that the content provider’s license allows. Users of +Microsoft Excel 2010, for example, can access DataMarket’s information directly using an Excel add-in or +use the PowerPivot for Excel support for OData to access information from DataMarket for data analysis. +It’s also possible to combine data purchased from DataMarket with your own data, such as for reports +created using SQL Server Reporting Services. -Other hazards with flying forms include the -following. +25 -• Stability during initial fabrication + Getting business value from information technology requires the right applications, but it also requires +the right data. By making it easier to discover, evaluate, and buy commercially available datasets, +DataMarket intends to make it simpler for organizations to find and use the information they need. -• Fill-in work between slab panels +LOOKING AHEAD -• Wind and fall hazards during flying operations +Microsoft has announced a number of updates that it plans to add to the Windows Azure platform in the +near future. They include the following: -• Stripping, flying, and re-setting. + The Windows Azure Platform Appliance: Microsoft has announced plans to offer a Windows Azure +Platform Appliance, allowing hosters and enterprises to run Windows Azure in their own data +centers. Because Windows Azure requires a specific hardware configuration—it’s not just software— +this appliance will include servers, networking, and storage, all running Windows Azure. And while it +will initially be targeted only at large organizations, Microsoft’s goal is eventually to make the +Windows Azure Platform Appliance useful for smaller organizations as well. -Although a flying form is designed to be stable -when complete, it may not be stable during -fabrication or erection. Temporary bracing or -temporary support by a crane may be necessary -to ensure stability during certain phases of the -operation (Figure 42-24). + CDN dynamic content caching: Today, the Windows Azure CDN works only with blob data. This +forthcoming functionality will let the CDN also cache content created dynamically by a Windows +Azure application. -Figure 42-22: Two Rows of Shoring Frames with -Row of Shoring Posts in Centre + VM role snapshotting: In its first release, the Windows Azure VM role doesn’t save any changes made +to the OS volume while it’s running. Snapshotting will change this, providing a way to periodically +save the state of this volume to persistent storage. -Frequently, supports for built-in-place forms are -deliberately left out to allow other work to be -done. One example might be a row of single-post -shores left out until work below is complete. Or, -an area might be supported temporarily during -construction by a few single-post shores that will -be replaced later by a shoring tower. + Better Java support: While Windows Azure can run Java applications today, Microsoft plans to make +this better. The coming improvements include better Java performance, stronger support for Eclipse- +based tools, and a more complete set of Java libraries for Windows Azure. -In these and other instances of incomplete -formwork, heavy temporary loads such as bundles -of rebar or stacks of plywood should not be placed -on the structure. Even on completed formwork, -make sure that landed material will not overload -the structure. +Support for composite applications: To make it easier to build Windows Azure applications by +combining existing components and services, Windows Azure AppFabric will add a composition +model. Accompanied by runtime services and a Visual Studio-based designer, this model will let +developers work with the various parts of an application as a single logical entity throughout its +lifecycle. -Before placing concrete, the formwork must be -inspected and approved by a professional engineer -or their designate. Keep a copy of this inspection -and approval on the site in case an inspector asks -to see it. + All of these changes target the same goal: making the Windows Azure platform useful in a broader range +of scenarios. -Flying Forms +CONCLUSIONS -Flying forms must always be designed by a -professional engineer and constructed, hoisted, -moved, and set strictly according to the -instructions of the designer or manufacturer. +The truth is evident: Cloud computing is here. For developers, taking advantage of the cloud means using +cloud platforms. With the Windows Azure platform, Microsoft presents a range of options addressing a +variety of needs: -Figure 42-24: Temporary Support of Shoring -System for Flying Wall Form + Windows Azure provides a computing and storage environment in the cloud. -One example is setting up trusses for a flying slab -formwork table. The trusses must be held upright -to be connected or disconnected. If not adequately -supported, they can fall over on workers. +SQL Azure provides a relational DBMS in the cloud, together with reporting and data synchronization. -Trusses and wall panels have also been blown over -by wind during fabrication and dismantling. Set- -up procedures should indicate the maximum wind -speed where the flying form operation can be done -safely. This wind speed should be determined by the -professional engineer who designed the system. +26 -Work with flying forms requires adequate space -for stacking materials and components. Working in -cramped quarters is not only difficult but hazardous. + Windows Azure AppFabric offers cloud-based infrastructure supporting both cloud and on-premises -Figure 42-23: Flying Forms +applications. -Tools and Techniques + Windows Azure Marketplace is an online store for finding and purchasing datasets from content -42-9 +providers, together with a forthcoming store for cloud applications. - FORMWORK +These approaches address a variety of requirements, and not every developer will use all of them. Yet +whether you work for an ISV, a systems integrator, or an enterprise, some cloud platform services are +likely to be useful for applications your organization creates. A new world is unfolding; prepare to be part +of it. -Fall Protection +ABOUT THE AUTHOR -A fall arrest system should be used by any worker -who is at risk of falling when: +David Chappell is Principal of Chappell & Associates (www.davidchappell.com) in San Francisco, California. +Through his speaking, writing, and consulting, he helps people around the world understand, use, and +make better decisions about new technologies. -• Installing a panel - -• Pushing a panel out toward the slab edge - -• Pulling a panel in from the slab edge (Figure - -42-24) - -• Helping other workers attach rigging hardware - -such as slings - -• Getting on and off - -• Bolting and unbolting wall forms for exterior walls - -and elevator shafts (Figure 42-25) - -• Stepping onto a panel to attach slings to pick - -points - -Figure 42-24: Fall Protection Required when -Receiving a Flying Form - -Figure 42-25: Attaching Chain Sling to Wall Form - -Each worker’s fall-arrest system must be attached -to an individual anchor independent of the flying -form. Contractors can provide for anchorage by -casting rebar anchors in columns or other areas to -be covered over or filled in later (Figure 42-26). -This type of anchorage must be approved by a -professional engineer. - -Figure 42-26: Rebar Anchor Cast in Concrete - -Safety Below Flying Forms - -The previous section covered the safety of workers -flying the forms. But precautions must also be -taken to protect workers below the hoisting -operation and the public at large, since forms are -often swung out over sidewalks and streets. The -Construction Projects regulation (213/91, s. 103(2)) -prohibits hoisting a load over the head of a person -who is not receiving the load or sinking a shaft. - -The most efficient protection for workers is to -fence or rope off the area below the flying form to -prevent anyone from entering the area. Pedestrian -traffic on sidewalks, as well as vehicle traffic if -necessary, should be detoured around the area -while hoisting is under way. - -Communication - -Flying forms are heavy, large, and awkward to -move. Hoisting and moving them safely requires -clear reliable communication. While hand signals -are often necessary and are still used, direct -radio or cell phone communication between the -signaller and the crane operator is more accurate -and effective. Relying on hand signals alone is not -recommended. - -Stripping - -Formwork stripping is probably the most -hazardous operation in concrete construction. -Hazards include the following. - -• Falling material - -• Waste material and equipment underfoot - -• Manual handling of heavy or awkward forms, - -panels, and other components - -• Prying forms loose from concrete presents risk of - -overexertion, lost balance, and slips and falls. - -42-10 - -Construction Health and Safety Manual Hazards can be reduced by - -• Planning and providing for stripping when - -designing and constructing formwork - -• Stripping as soon as it is permitted by the - -engineer - -• Supplying facilities and equipment for removing - -materials as they are stripped - -• Providing proper tools and adequate access for - -the stripping crew - -• Training personnel properly for this and other - -aspects of formwork. - -Forms can be designed with crush plates or filler -strips to facilitate removal at difficult intersections -of columns, beams, and wall forms. Later, formwork -oils should be used liberally to make stripping -easier. - -Wherever possible, waste materials, plywood, and -debris should be removed from the area as work -proceeds. This will reduce the need to walk over or -work around things left on floor or ground. - -Providing carts or cradles can help the crew -remove material and reduce the need for lifting -and carrying. Material on a cart can be rolled away. -Material in cradles can be hoisted off by a crane. - -Climbing partially stripped formwork is not -only hazardous but unnecessary. Safe access -such as rolling scaffolds or powered elevating -work platforms should be provided for stripping -formwork at elevated locations. - -Poor lighting may be a hazard in formwork -stripping in the early morning or late in the day -during certain times of the year. Mobile light -stands are probably the best solution, since pigtail -stringers can easily be knocked down and damaged -during stripping. - -Wherever possible, stripping crews should be -small. This is especially important with knock-down -systems. In small crews, the supervisor and each -worker can more easily keep track of what the -others are doing. Workers are not as likely to create -hazards for each other. Crews of two or three are -recommended for knock-down systems. If more -workers are required, they can still be divided into -small crews working in separate areas. - -Other trades and operations should not be allowed -in areas where stripping is under way. Given the -many hazards involved, the area should be fenced -or roped off and warning signs posted. - -Knock-Down Slab Systems - -Stripping these forms is physically difficult -because much of the work is overhead. The -usual arrangement involves shoring frames or a -combination of shoring frames and jacks. - -Wherever possible, the work should proceed -progressively from one side. That means taking out -one row of formwork supported by a row of stringers -on shoring frames. - -FORMWORK - -The first step is to back off the adjustable base plates -and U-heads in one area, which will in turn lower the -stringers, joists, and plywood (Figure 42-27). - -Figure 42-27: Lowering Base Plates -and U-Heads to Release Slab Form - -In practice, however, the plywood will stick to -the underside of the concrete, especially around -beams, column caps, and similar points. Wherever -possible, stuck sheets should be loosened and -removed before the shoring structure is dismantled. - -Stripping should proceed in reverse order to -erection. Plywood should be removed first, -followed by joists and stringers. The last items to -be removed are the shoring frames. - -When scaffold or shoring frames are used as a -work platform, it should be completely decked -in with planks (Figure 42-28). Otherwise planks -can shift and slide as workers pry or pull at stuck -pieces of formwork, lose their balance, and fall. This -has been a frequent cause of injuries. Temporary -guardrails may be required near the edge of the -structure or workers will need to use fall protection -during this operation. - -Figure 42-28: Shoring Frame -Fully Planked for Access - -Tools and Techniques - -42-11 - - FORMWORK - -The area where stripping starts should be planned -to allow access for taking away material as the -form is dismantled. - -Columns - -Check the following: - -Sound training, effective supervision, well- -designed safe access facilities, and immediate and -continuous cleanup can help reduce hazards in -stripping knock-down slab forms. - -Built-in-Place Wall Forms - -These forms are frequently of only moderate height. -Taller types usually make use of large panels erected -and removed by crane rather than hand. Built- -in-place wall forms are usually a stud-and-wale -system using some type of formwork ties. - -Where workers cannot reach the top of the wall, -scaffolding should be provided for removing wales -on the upper level. Safe access is essential for the -dismantling and manhandling of wales that are -frequently long, heavy, and waterlogged. - -Material should then be removed immediately to a -staging area. - -Inspection - -Before concrete placing begins, formwork must be -inspected and signed off by a professional engineer -or a competent person designated by the engineer -to ensure that it has been constructed to provide -for worker safety and to meet job specifications. - -A report must be completed by the person doing -the inspection stating whether the formwork has -been constructed according to the design. Any -discrepancies must be approved by the design -engineer before concrete placing proceeds. - -It is in everyone’s best interest to ensure that the -formwork has been inspected by a competent -person for workmanship, stability, and adherence -to design drawings and specifications. Inspection -should begin when the forms are being constructed -and continue until concrete placing is complete. - -Checking line and grade is best carried out while the -formwork is being constructed. Shoring structures -should be within the alignment limits specified on -the design drawings. Line and grade should also -be checked during the pour to determine whether -formwork is shifting or deflecting. - -Dimensions of special features like beams, column -capitals, and inserts are best checked at the time of -construction. If inspection is delayed until formwork -is completed, some details may be covered up or -become more difficult to check. - -• The proper size and type of materials are used - -• Column ties or column clamps are spaced - -according to design drawings - -• The spacing of ties or clamps is based on a - -sound assessment of concrete pressure (generally -columns are designed for a full liquid head of -150 lb per cubic foot (2,403 kg per cubic metre) -times height) - -• Columns are adequately braced where they are - -not tied in to a slab-form structure. - -NOTE: For more information on column formwork -pressures, refer to CAN/CSA S269.1-16: Falsework -and Formwork or the American Concrete Institute -(ACI) standard Formwork for Concrete (SP-4). - -Wall Forms - -Check the following: - -• Materials and any manufactured components are as -specified in design drawings (size and spacing of -studs, wales, and ties are crucial to safety) - -• Ties are snugged up before concrete is placed - -• Wedges in wedged systems are tight - -• Nuts in threaded systems are tight - -• Bracing conforms to design drawings - -• Free-standing formwork is braced to ensure stability - -and resistance to loads during concrete placing - -• Specified pour rates are not exceeded (wall - -forms are often designed for specific pour rates; -exceeding these rates can cause failure or collapse). - -Slab Forms - -From a safety perspective, this is the most critical -type of formwork. The collapse of slab forms has -caused many injuries and deaths, whether from -flawed design (e.g., not considering the effect -of wind loading on the design), unauthorized -modifications in the field, or the failure to -inspect, which would have located and corrected -deficiencies. - -Competent inspection demands knowledge, -experience, and the ability to - -1) Distinguish between similar but different - -materials and shoring equipment - -2) Read and interpret design drawings - -3) - -Identify and clear up with the designer any -apparent or real discrepancies in components -such as shoring frames. - -42-12 - -Construction Health and Safety Manual FORMWORK - -Check the following: - -Watch for the following warning signs: - -• Grade beams or mud sills supporting shoring are - -• Lateral displacement or movement of single-post - -properly sized and located. - -shoring for slab forms - -• Hazardous soil conditions such as excessive -moisture, freezing, and uncompacted soil -conditions are reported and discussed with the -professional engineer. - -• Shoring frames and jacks are located and aligned - -within tolerances specified on the drawings. - -• Shoring frames and jacks are out of plumb no - -more than 1/8 inch every 3 feet. - -• Adjustable base plates for shoring frames and - -jacks are snugged up. - -• U-heads are wedged in place. - -• Stringers are the specified size and number, with - -supports properly spaced. - -• Aluminum stringers have no bent flanges or other - -damage. - -• Movement or deflection of lateral bracing for - -single-post shores - -• Movement of stringers on U-heads - -• Crushing of wooden stringers on U-heads - -(Figure 42-29) - -• Gaps or shoring that is not snugged up under - -stringers - -• Deflection of stringers between supports - -(Figure 42-30) - -• Deflection of wales or strongbacks on wall forms - -• Wall forms that are bulging or cracking - -• Crushing of wales or strongbacks at washers - -for ties - -• Movement of wall forms - -• Joists are the specified size and properly spaced. - -• Uplifting of battered forms - -• Support structures and shoring for beam bottoms -and column capitals are constructed according to -design. - -• Lateral bracing is provided where required (for -instance, on freestanding formwork for bridges -and overpasses). - -• The bearing surface for lateral bracing is -adequate—that is, stable footings or well- -compacted soil. - -• Temporary loads such as rebar are not - -overloading the system. - -Concrete Placing - -Forms should continue being observed during -concrete placing. Any signs of movement, crushing, -cracking, or deflection are cause for alarm. Placing -concrete should be suspended until the situation is -corrected. - -• Pour rates that exceed design specifications. - -Figure 42-29: Crushing of Stringer Over U-Head - -Figure 42-30: Deflection -of Stringers Between Supports - -Tools and Techniques - -42-13 +27 diff --git a/q188/random_k4/random_4.md b/q188/random_k4/random_4.md index e9de94a4bbd08f28025507ab54711ae0934d7254..2a0f41fa3f2088cf2adce6fed1fbcc051440cda4 100644 --- a/q188/random_k4/random_4.md +++ b/q188/random_k4/random_4.md @@ -1,558 +1,2106 @@ -Becoming a -data-driven -organization - -Joe Chung, Enterprise Strategist and -Evangelist at Amazon Web Services - - Every company has -a data problem - -Imagine this... - -The weekly Excel report comes out and is delivered to your - -email. As you review it, you see an anomaly in the financial - -data that you don’t understand, despite the pivot table - -provided in the report that allows you to drill down to at - -least some level of detail. You ask your operations analyst - -what’s going on. To which your analyst responds, “I’m not - -sure. Let me find out.” - -The next day, the analyst tells you that the reason - -for the anomaly is productivity was way down at the - -manufacturing plant. - -“That doesn’t make sense,” you say. “Can you ask -HR if sick days are impacting the productivity -numbers? Or could it be that there was an issue -with the time capture application at the plant?” - -“It will take a week to get at that data and merge -it with the financial data,” your analyst says. - -“Can’t you just send me a dump of the data from -the ERP and time application, and I’ll work with -it myself?” - -The analyst responds, “I don’t have access to the -data, and it will take a few days to submit the -right tickets to get access to it.” - -If this scenario seems all too familiar to you, your -organization has a big data problem. - - Your first reaction may be this is a business -intelligence process and tools challenge that -has plagued organizations since time began, -not a true big data problem. Without getting -into a religious debate about what analytics -vs. reporting vs. business intelligence are, -the key fact is that every organization has a -big data problem. With artificial intelligence -and machine learning capabilities starting -to come to fruition, it’s now more important -than ever for enterprises to get a better -handle on the data they have, and how to -really harness its power to become a -data-driven organization. - -The mental model I have for becoming a -data-driven organization is the nervous -system of the human body. Nerve endings -extend throughout our bodies sending sensory -signals to the spinal cord to be processed by -our brains and acted upon. It’s a model that’s -being emulated through data architectures -that have the ability to receive, process and -store data real time from anywhere inside -and outside the enterprise. The signals are -processed in real time and acted upon by -machine learning algorithms. Unfortunately, -far too many organizations believe these -new capabilities are nice-to-haves, only -applicable to specialized data scenarios, or -they try to relabel legacy business intelligence -platforms as “data lakes.” - -Data dysfunctions - -Most of us think of big data problems as -being about volume. The truth is every -organization has data issues beyond volume -that have been masked in a number of -different ways. Here are a few common data -dysfunctions that I’ve observed: - -Lonely and discarded data - -First, many organizations don’t realize that -a lot of interesting data is thrown away or is -just not accessible. Some examples include -data like user activity in the application -(and how they might use the application in -relation to other applications); the telemetry -of the infrastructure hosting the application; -or older versions of data that are no longer -compatible with current table schemas. - -Second, data is siloed across many -applications and data warehouses. While -no one application may be “big,” together, -they are big. So, when the business needs -to analyze data across multiple sources, it -becomes very challenging. That’s because -siloed data poses an access problem. Every -place where data is stored comes with its -own access roles, rules, and ceremonies to be -adhered to that can make it challenging to -access the data. - -Every organization -has a big data -issue that has -been masked in -different ways. - - Data-driven -decisions -necessitate -access to many -disparate types -of information. - -Low-fidelity data - -Legacy enterprise systems generally process -and capture end states and typically only -report small snapshots of time. In addition, -data is processed in batch rather than in -real time. Data can change a lot between -batch windows, but older systems are often -designed to discard interim changes because -they cannot handle the speed in which data -can change. - -Round data in a square table - -Many enterprises realize there are troves -of data that don’t neatly fit into traditional -database storage technologies (for example, -images, sensor data, and so on). There is also -a lot of variety in how you can analyze and -consume insights from the data. For example, -when launching a new analytics initiative, -you may realize that no single reporting -or visualization solution can fulfill all your -users’ needs. You may need to consider -delivering insights processed by algorithms -through APIs, within applications using -custom visualization widgets using JavaScript -frameworks like D3.js, and through business -intelligence portals leveraging Tableau and -other visualization solutions. - -Messy data - -Enterprise systems don’t like messy data, -which is why there are forms, rules, and - -other validations to make sure data is as -clean as possible before it’s stored. But -some of the most interesting data may not -be so clean. When you start tapping into -unstructured or object-based data, there’s -going to be noise. Just like in electronic -noise, there are filtering, enhancement, -and amplification mechanisms you can use -to get at the data you want. One use case -that many enterprises are concerned with -is the spiraling costs of sending data into -proprietary log aggregation, security, or -monitoring tools. In most of these cases, -there’s an opportunity to lower costs by -filtering out a vast portion of the log data -that isn’t useful. - -If you’ve struggled or resonate with any of -the points above, it’s time your organization -revisits your analytics approach and -architecture. Every enterprise has the -opportunity to deploy fit-for-purpose -analytics solutions (storage, processing, -querying, analysis, presentation, and so on) to -meet their existing business and IT challenges. - -Modern analytics platforms enable -critical business insight - -Once you’re ready to tackle your big data -problem, what can you reasonably expect to -accomplish with a modern analytics platform? -Here’s what’s possible and how it’s being done -today, from a technical perspective. - - such as NoSQL/non-relational technology. -Amazon then evolved and leveraged Dynamo -to create services such as Amazon S3 and -Amazon DynamoDB. Amazon S3 is attractive -to enterprises looking to create data lakes -due to its ability to store many different data -types and its low storage cost. There are, of -course, other technical solutions, including -Hadoop, but an important characteristic of all -data lake solutions is their ability to store all -types of data at petabyte scale and at -low cost. - -Responsive to change - -Business systems and data change all the -time, but often the systems which report or -share that information end up being the last -to change. How many times have you been -told it will take six months or more for data -to be remediated in data warehouses and -reports? Or that data changes from source -systems have not yet flowed to reporting -systems, and that it takes several days for -those changes to make their way down due -to the batch processing? The speed at which -data is available dictates the speed at -which decisions can be made. Therefore, we -should expect that modern analytics systems -be able to process and report data in near -real time and be responsive to changes to -upstream data sources. - -Access to any data I want - -Data-driven decisions necessitate access to -many disparate types of information. -A pilot relies on the gauges in the airplane -to understand information critical to flight— -like altitude, air speed, fuel consumption. -But imagine if the pilot didn’t have those -gauges all in one place. Perhaps they have -to walk to the back cabin or radio in for -the information or worse yet, have to -ask permission for the data. Unfortunately, -this is a daily reality in today’s enterprise -environment. - -Forward-leaning organizations have flipped -this standard on its head by pulling data out -of the systems in which it exists and storing it -in one place (that is, a data lake). While there -are many instances of companies storing large -amounts of one type of data, more and more -companies are creating enterprise-wide data -lakes that contain multiple data types from -different sources. - -Internet-scale companies like Amazon, Yahoo, -and Facebook started to see in the early -2000s that relational database technologies -had reached their limits in terms of scalability -and performance. Amazon responded with a -technology called Dynamo, which is a highly -available and scalable key value store, - -The speed at which -data is available -dictates the speed -at which decisions -can be made. - - We need to stop -making users slog -through terrible -experiences to -get the data and -insights they need. - -The first key enabler is the nature of how data is -stored in big data technologies like Amazon S3 -or Hadoop. One of the big inhibitors to changing -a relational database is modifying the schema -or definition of how data should be stored. Until -the schema is modified, data can’t land in the -database or it will break. File or object-based -technologies like Amazon S3 don’t care how -data is structured—data can come as it is versus -the “you need to fit my structure” approach. - -The other challenge is that only one schema is -active at any given time. While I’m sure we’ve -all seen database tables named “2015” and -“2016,” it’s not ideal. Big data technologies have -a schema on read-based approach, meaning the -structure of the data is applied when you grab it -and not inferred based on how it’s stored. What -that means for businesses is that data changes -from source systems aren’t a big deal. - -The second enabler is streaming technologies -like Amazon Kinesis and Apache Spark. Most -enterprises move data around in big batches; -typically, this occurs once a day. Streaming -technologies allow data to be ingested in -smaller pieces at a massive scale. For example, -SONOS, the speaker manufacturer, processes 1 -billion events per week using Amazon Kinesis. -You should never have to wait for the daily -batch to be completed to understand where -your business stands. - -Interactive insights where and how -I want them - -Today’s enterprise users jump through lots -of hoops to understand the information -being presented to them. Maybe it’s digging -through their inbox to find the report that -was attached. Or logging into the reporting -system to download a PDF only to find they -have to copy and paste the data into Excel -to make sense of what is there. We need -to stop making users slog through terrible -experiences to get the data and insights they -need. The rallying cry for users should be: -Bring the data in the right form with the right -tools at the right time. - -Software like Tableau, Amazon QuickSight, -and others have made things better by -considering user experience when interacting -with data. However, I have found that at -most enterprises it requires the use of -many tools to meet users’ needs. It could -be Amazon QuickSight embedded in a -business intelligence portal to a Tableau -workbook sent in an email. AWS provides -the diversity of data storage and business -intelligence tools delivered through a pay- -as-you-go model. This allows organizations -to experiment with many different business -intelligence tools without making large -investments in infrastructure and licensing. - - One persona you should not ignore is -the data scientists in your organization. -Jupyter notebooks have really taken off in -the data science community; they are part -content management, part code execution, -and part visualization. It’s a very powerful -tool to share knowledge and document -and execute machine learning algorithms. -Amazon SageMaker is a managed notebook -environment that takes care of all the heavy -lifting for you and your data scientists - -Intelligence embedded in the business - -Artificial intelligence and machine learning -are all the rage these days, and rightly so. -Advances in machine learning frameworks -coupled with the use of specialized servers -utilizing graphics processing units (GPUs) -are enabling all kinds of new capabilities, -like autonomous driving. Of course, in -order to train machine learning models -vast amounts of data are required (thus -the data lake points I discussed above). -Organizations are already starting to take -advantage of these AI/ML capabilities to -drive new outcomes not previously possible, -such as being able to better predict health -outcomes based on retinal imaging, or -predict outages or hardware breakdowns in - -the field. Businesses can strengthen their -organizational AI/ML muscles by letting -AWS take care of the heavy lifting, as this is -not the stuff of science fiction but operating -in production today. - -One final point I would like to emphasize -is that the best algorithm in the world is -worthless unless it can be integrated with -business processes. Often getting the -insight or data science model created is the -easy part—getting it integrated into your -insurance policy engine or retail platform -is the hard work, as these systems do -not typically have the ability to integrate -outside data sources or APIs. This is a great -opportunity to consider moving these -systems to the cloud to take advantage of -all the services available to help modernize -or re-architect them. - -Organizing for insights - -Building an advanced analytics capability -at your company is not just about the -technology. Often, the biggest challenge -organizations face has to do with the -organization—its processes, governance -and people. So, what do you need to keep -in mind as you organize your analytics -investment for success? - -The best algorithm -in the world is -worthless unless -it can be integrated -with business -processes. - - It’s not just about -having the latest -tools in the belt; -it’s about making -it easy for your -customers to get -what they need. - -Start with an analytics Center of Excellence - -To go from strategy and intent to meaningful -progress, one of the first steps is to identify -and choose a leader and a team to spearhead -the change, and to set up an analytics Center -of Excellence (COE). The team typically starts -small, with a few cross-functional roles to -bootstrap it; the team will then grow to -service more and more needs. - -Many large enterprises already have -established shared services organizations -that handle business intelligence or -reporting. Those organizations can seed -the analytics COE with technical and -business roles. Similar to IT infrastructure -organizations, they should not only supply -the talent, but also be a key driver and -sponsor for the effort. Because over time, -these reporting shared services organizations -will need to evolve to either adapt or become -part of the analytics COE. The starter roles -are often data engineers and architects, -business intelligence analysts, and data -scientists. The group needs to be led by -someone who can work across multiple -organizations, business units, and back-office -groups, such as Finance and IT. - -Serve all your customers’ needs - -One of the primary mental shifts that -organizations need to make is moving from -an attitude of “you must use our reporting - -solution and you will like it” to “what are your -analytics needs and how can we help enable -you?” Shared services reporting organizations -are often report pushers and are not positioned -to answer the difficult questions that come -from employees, business leads, and customers. - -Therefore, when establishing a new analytics -COE, it’s important to establish tenets for the -group which will set expectations for how the -group acts and makes decisions. - -The analytics COE will need to service -two types of customers: - -• The data and analytics consumers: The - -decision makers, data scientists, business -intelligence (BI) analysts, and developers. -These customers typically care about -the ability to quickly access insights and -data and the quality of the tools and -services available to them to process and -present data. - -• The data producers: The owners of - -applications, infrastructure, and devices -who will supply the data into the platform. -These customers need services like the -ability to easily publish their data into -the analytics platform and define a data -contract. This includes the domain model -of the data, frequency of refresh, and -definition of policies, for example, -a security policy outlining who can access -their data. - - The analytics capability and the platform -need to service both types of customers— -if their needs are not met then the analytics -effort will not deliver the business value. -Therefore, it is critical to have a mechanism -to capture the needs of these two types of -customers across a potentially very large and -diverse set of business units and personas. -Some organizations set up advisory boards or -work with a few key stakeholders to drive the -need. There is no single correct answer, but -having a mechanism to capture the voice of -the customer(s) and prioritizing their needs -is critical. - -Rethink the COE - -An analytics COE services and presents a -specialized set of cloud services focused -on meeting analytics needs. In the past, -reporting and BI organizations often -provided one solution to fit everyone’s -needs (a one-size-fits-all strategy). In this -era of rapidly evolving technologies in big -data, rich visualizations, automated decision -making, artificial intelligence, and machine -learning it’s just not possible to have a single -technology stack. It’s not just about having -the latest tools in the belt; it’s about making -it easy for your customers (producers or -consumers) to get what they need. - -COEs run the risk of becoming concierge -services. That can be fine for certain types -of requests, but the COE can quickly -get overwhelmed and backlogged with -requests if there’s no scalable, self-service -mechanisms and transparent prioritization -and governance processes in place. Analytics -COEs need to engineer and architect a self- -service, secure, operable, and scalable data -platform with an ever-evolving ecosystem -of technologies to process, analyze, and -present insights. - -While becoming a data-driven -organization won’t happen -overnight, pinpointing your data -challenges, organizing a plan for -how you will serve your customers, -and empowering your teams to -deliver the right value at the right -time are steps in the right direction. - -About the Author - -Joe Chung is an Enterprise -Strategist and Evangelist -with Amazon Web Services. +User Guide + +Lenovo ThinkBook 14 Gen 4 and Lenovo ThinkBook 15 Gen 4 + + Read this first + +Before using this documentation and the product it supports, ensure that you read and understand the +following: + +• Generic Safety and Compliance Notices + +• Safety and Warranty Guide + +• Setup Guide + +First Edition (March 2022) + +© Copyright Lenovo 2022. + +LIMITED AND RESTRICTED RIGHTS NOTICE: If data or software is delivered pursuant to a General Services +Administration “GSA” contract, use, reproduction, or disclosure is subject to restrictions set forth in Contract No. GS- +35F-05925. + + Contents + +About this guide . . . . . . . . . . . . . iii + +. +. +. +. +. + +. +. +. +. +. + +. +. +. +. +. + +. . +. . +. . +. . +. . + +Chapter 1. Meet your computer . . . . . 1 +. . +. +. . +. 1 +Front . +. +. . +. +. . +. 2 +Base . +. +. . +. +. 4 +. . +Left +. +. +. 6 +. . +. +Right . +. . +. +. 7 +Bottom . . +. . +. +. +. 8 +Features and specifications . . +. +. +. 9 +. 10 + +. . +. . +. . +. . +. . +. . +Statement on USB transfer rate . . +Operating environment . . +. . +Avoid constant body contact with specific hot +. +sections . + +. . +. . +. . +. . +. . +. . +. . +. . + +. +. +. +. +. +. +. +. + +. +. +. +. +. +. +. +. + +. +. +. +. +. +. + +. 10 + +. . + +. . + +. . + +. . + +. + +. + +. + +. + +. + +. + +. + +. + +. +. + +. . +. . + +. . +Windows help information . + +Chapter 2. Get started with your +computer . . . . . . . . . . . . . . . . 11 +. 11 +Work with Windows . +. 12 +. 12 +. 12 +. 12 +. 13 +. 13 +. 14 + +. +. +Lenovo Vantage and Lenovo PC Manager +. +The Novo Button menu . +. +. +. +. + +. +. +. +. +. +. +. +Keyboard hotkeys . +Numeric keypad . +. +. +Multi-touch gestures for touch-enabled +. +. +screens . + +. +Open the Novo Button menu . +. +. +. + +Interact with your computer . . +. . +. . + +. . +. . +. . +. . +. . +. . +. . +. . + +. . +. . +. . +. . +. . + +. . + +. 14 + +. . + +. . + +. . + +. . + +. +. + +. + +. + +. + +. + +. + +Chapter 3. Explore your computer . . 17 +. 17 +Manage power . +. 17 +. 17 + +Check the battery status +Charge the battery . + +. . +. +. . + +. . +. . +. . + +. . +. . +. . + +. +. +. + +. +. +. + +. +. +. + +. . + +. + +. + +. + +. + +. . + +. +. +. + +. +. +. + +. +. +. +. +. +. +. +. +. +. +. +. + +Set power button behavior. . +. . +A power plan . +. +. . +Set performance mode . + +. +What is UEFI/BIOS setup utility . +Open the UEFI/BIOS setup utility . +. +Change the startup sequence . +. +. +Change hotkey mode . +. . +. +. +Enable or disable always-on . + +. . +. . +. . +Change settings in UEFI/BIOS setup utility . +. . +. . +. . +. . +. . +Set passwords in UEFI/BIOS setup utility . . +. . +. . + +. +. +. +. +. +. +. +. +. +. +. +Password types . +. . +Set administrator password . +. +Change or remove administrator password . +. . +. +Set user password . . +. . +. +. . +Enable power-on password . +. +Set hard disk password . +. . +. . +. +Change or remove hard disk password . + +. +. +. +. + +. +. +. + +. +. +. + +. . + +. +. + +. +. + +. + +. + +. + +. 17 +. 17 +. 18 +. 18 +. 18 +. 18 +. 18 +. 18 +. 19 +. 19 +. 19 +. 19 +. 20 +. 20 +. 20 +. 20 +. 21 + +. . +. . + +Chapter 4. Help and support +. . +. +Frequently asked questions . +. . +. +. +Self-help resources . +. . +. +What is a CRU? . +. +. +. . +. +CRUs for your product model +. . +. +. +. +Call Lenovo . . +. +Before you contact Lenovo . +Lenovo Customer Support Center . +. + +Purchase additional services. + +. +. +. +. +. +. + +. . + +. . + +. + +. + +. + +. . . . . 23 +. 23 +. +. . +. 23 +. +. . +. 24 +. +. . +. 25 +. +. . +. 25 +. +. . +. 25 +. +. . +. 25 +. +. . +. 26 +. +. . + +. +. +. +. +. +. +. +. + +Appendix A. Notices and +trademarks . . . . . . . . . . . . . . . 27 + +© Copyright Lenovo 2022 + +i + + ii + +User Guide + + About this guide + +• This guide applies to Lenovo product model(s) listed below. Illustrations in this guide may look slightly + +different from your product model. + +Model name + +ThinkBook 14 G4 IAP + +ThinkBook 14 G4 ABA + +ThinkBook 15 G4 IAP + +ThinkBook 15 G4 ABA + +Machine type (MT) + +21DH + +21DK + +21DJ + +21DL + +• For further compliance information, refer to the Generic Safety and Compliance Notices at https:// + +pcsupport.lenovo.com/docs/generic_notices. + +• This guide may contain information about accessories, features, and software that are not available on all + +models. + +• This guide contains instructions that are based on the Windows operating system. These instructions are + +not applicable if you install and use other operating systems. + +• Microsoft® makes periodic feature changes to the Windows® operating system through Windows Update. +As a result, the operating system related instructions may become outdated. Refer to Microsoft resources +for the latest information. + +• The content of the guide is subject to change without notice. To obtain the latest version, go to https:// + +support.lenovo.com. + +© Copyright Lenovo 2022 + +iii + + iv + +User Guide + + Chapter 1. Meet your computer + +Front + +1. Microphones + +2. Camera + +3. Camera switch + +4. Camera light + +5. Screen + +Capture or record sound and voice. + +Capture still and moving images for taking photos, recording videos, +and video chatting. + +Turn the camera on or off. + +If the light is on, the camera is in use. + +Display text, graphics, and videos. Selected models support multi- +touch function. + +© Copyright Lenovo 2022 + +1 + + Base + +Figure 1. Base view—14-inch models + +Figure 2. Base view—15-inch models + +2 + +User Guide + + 1. Wireless antennas + +2. Power button (with power light) / +Fingerprint reader (with Fingerprint reader +light) * + +3. Numeric keypad + +4. Touchpad + +5. Keyboard + +* for selected models + +Send and receive radio waves for the built-in wireless LAN (local area +network) and Bluetooth module. +Notes: + +• The antennas are not visible from the outside of the computer. + +• If the location of the antennas is close to the user’s body (< 20 cm), +you can find the Specific Absorption Rate (SAR) values in the Setup +Guide that comes with your computer. + +• Press the button to turn on the computer or put the computer into + +sleep mode. + +Note: You can turn on the computer by flipping open the LCD +screen if Flip to Boot is enabled. + +• Enroll and read your fingerprint for user authentication. + +• The light shows the power status of the computer. If the computer +includes a fingerprint reader, it can also show the status of the +reader. + +– Power light + +– On: The computer is turned on. + +– Off: The computer is turned off or in hibernation mode. + +– Blinking: The computer is in sleep mode. + +– Fingerprint reader light + +– Blinking and then solid green: The fingerprint reader is + +working. + +To enable or disable the numeric keypad, press the Num Lock key. + +Perform finger touch and all the functions of a traditional mouse. +Note: The touchpad also supports multi-touch function. + +Enter characters and interact with programs. +Note: The keyboard also includes hotkeys and function keys for +changing settings and performing tasks quickly. + +Chapter 1. Meet your computer + +3 + + Left + +1. USB-C® connector/Power connector + +• Connect to power with the included power cord and ac power + +adapter. + +• Connect to an external display, storage, or data devices. + +Note: An appropriate adapter may be needed for connecting +external devices. + +2. Charging light + +Indicate whether the computer is connected to ac power. + +3. Thunderbolt™ 4 connector * + +4. HDMI™ connector + +5. USB (3.2 Gen 1) Type-A connector +(always-on) + +6. Combo audio jack + +• Solid white: Connected to ac power; battery capacity 91%–100% + +• Solid amber: Connected to ac power; battery capacity 1%–90% + +• Off: Not connected to ac power + +Connect to an external display, storage, or data devices. +Note: An appropriate adapter may be needed for connecting to +external devices. + +Connect to a compatible digital audio device or video monitor, such +as an HDTV. + +The always-on connector can charge external devices when the +computer is turned off, in sleep or hibernation mode. The always-on +function can be turned on or off in the UEFI/BIOS setup utility. +Note: If Lenovo Vantage is pre-installed on your computer, you can +also set always-on function within that software. + +Connect to headphones or headsets with a 3.5 mm (0.14 inch), 4-pole +plug. +Note: This jack does not support standalone external microphones. If +you use headsets, choose one with a single plug. + +4 + +User Guide + + 7. Novo button hole + +Press the Novo button to open the Novo Button menu. From the +menu, you can choose to open the UEFI/BIOS setup utility, the Boot +menu, or the Windows startup options screen. + +8. Battery light + +Indicate the battery capacity of the computer. + +• Connected to ac power + +– Solid white: Battery capacity 80%–100% + +– Blinking white: Battery capacity 20%–80% + +– Blinking amber: Battery capacity 1%–20% + +• Not connected to ac power + +– Off: Battery capacity 20%–100% + +– Solid amber: Battery capacity 5%–20% + +– Blinking amber: Battery capacity 1%–5% + +* for ThinkBook 14 G4 IAP and ThinkBook 15 G4 IAP only + +Chapter 1. Meet your computer + +5 + + Right + +1. SD card slot + +Insert an SD card or a MultiMediaCard (MMC) into the slot for data +transfer or storage. + +2. USB (3.2 Gen 1) Type-A connector + +Connect to a USB-compatible device, such as a USB keyboard, a +USB mouse, a USB storage device, or a USB printer. + +3. Ethernet connector + +Connect to a local area network (LAN). + +4. Nano security-lock slot + +Lock your computer to a desk, table, or other fixtures through a +compatible security cable lock. + +6 + +User Guide + + Bottom + +Figure 3. Bottom view—14-inch models + +Figure 4. Bottom view—15-inch models + +1. Ventilation slots (inlet) + +Maintain the incoming airflow to dissipate internal heat. +Note: Do not block the ventilation slots. Otherwise, the computer +might overheat. + +2. Speakers + +Produce sound. + +Chapter 1. Meet your computer + +7 + + Features and specifications + +Dimensions + +• ThinkBook 14 G4 IAP and ThinkBook 14 G4 ABA + +– Width: 323.9 mm (12.75 inches) + +– Depth: 218 mm (8.58 inches) + +– Thickness: 17.9 mm (0.70 inches) + +• ThinkBook 15 G4 IAP and ThinkBook 15 G4 ABA + +– Width: 357 mm (14.06 inches) + +– Depth: 235 mm (9.25 inches) + +– Thickness: + +– With metal D cover: 17.9 mm (0.70 inches) + +– With plastic D cover: 18.9 mm (0.74 inches) + +• Input: 100 V ac–240 V ac, 50 Hz–60 Hz + +• Maximum output: 20 V dc, 3.25 A/20 V dc, 5 A + +• Maximum power: 65 W/100 W + +ac power adapter + +Battery pack + +• Capacity: 45 Wh/60 Wh + +• Number of cells: 3/4 + +Note: The battery capacity is the typical or average capacity as measured in a specific test +environment. Capacities measured in other environments may differ but are no lower than the +rated capacity (see product label). + +Microprocessor + +To view the microprocessor information of your computer, type system information in the +Windows search box and then press Enter. + +Memory + +Storage device + +• Type: DDR4 + +• Number of physical slots: 1 + +• Type: solid-state drive + +• Form factor: M.2 (2242/2280) + +• Bus: + +– PCIe Gen4x4 + +– PCIe Gen3x4 + +Screen + +• Size: + +– ThinkBook 14 G4 IAP and ThinkBook 14 G4 ABA: 355.60 mm (14 inches) + +– ThinkBook 15 G4 IAP and ThinkBook 15 G4 ABA: 396.24 mm (15.6 inches) + +• Display resolution: 1920 × 1080 pixels (FHD model) + +Keyboard + +• Function keys + +• Hotkeys + +• Numeric keypad + +• Keyboard backlight * + +8 + +User Guide + + Connectors and slots + +• Combo audio jack + +• Ethernet connector + +• USB Type-A connector × 2 (one connector with Always-On function) + +• ThinkBook 14 G4 ABA and ThinkBook 15 G4 ABA + +– USB-C connector × 2 + +– Power input up to 100 W using the included power adapter + +– Power output up to 5 V and 3 A + +– USB data transfer rate up to 10 Gbps for USB 3.2 Gen 2 + +– DisplayPort 1.4. Supports maximum output resolution of 5120 × 3200, at 60 Hz frame + +rate and color depth of 24 bpp (bits per pixel). + +• ThinkBook 14 G4 IAP and ThinkBook 15 G4 IAP + +– Thunderbolt 4 connector + +– Power output up to 5 V and 3 A + +– DisplayPort 1.4. Supports maximum output resolution of 5120 × 3200, at 60 Hz frame + +rate and color depth of 24 bpp (bits per pixel). + +– Up to 40 Gbps bi-directional data transfer rate in Thunderbolt mode. + +– USB-C connector + +– Power input up to 100 W using the included power adapter + +– Power output up to 5 V and 3 A + +– USB data transfer rate up to 10 Gbps for USB 3.2 Gen 2 + +– DisplayPort 1.4. Supports maximum output resolution of 5120 × 3200, at 60 Hz frame + +rate and color depth of 24 bpp (bits per pixel). + +• HDMI 2.0 connector + +• SD card slot + +Note: Data rates and performance ratings are dependent on connected devices and cables if +they are used. For DisplayPort connection through a USB-C port, the listed maximum output +capacity is only available on external displays with a DisplayPort, a Mini DisplayPort, or a +USB-C port that supports DisplayPort Alternate Mode. For connections using a converter or +an adapter, the actual output resolution may be lower. + +Security + +• Fingerprint reader * + +• Nano security-lock slot + +• UEFI/BIOS passwords + +– Administrator password + +– User password + +– Master hard disk password + +– User hard disk password + +Network + +• Ethernet (10/100/1000 Mbps) network adapter + +• Bluetooth and Wi-Fi combo adapter + +* for selected models + +Statement on USB transfer rate + +Depending on many factors such as the processing capability of the host and peripheral devices, file +attributes, and other factors related to system configuration and operating environments, the actual transfer + +Chapter 1. Meet your computer + +9 + + rate using the various USB connectors on this device will vary and will be slower than the data rate listed +below for each corresponding device. + +USB device + +3.2 Gen 1 + +3.2 Gen 2 + +Data rate (Gbit/s) + +5 + +10 + +Operating environment + +Maximum altitude (without pressurization) + +3048 m (10 000 ft) + +Temperature + +• At altitudes up to 2438 m (8000 ft) + +– Operating: 5°C to 35°C (41°F to 95°F) + +– Storage: 5°C to 43°C (41°F to 109°F) + +• At altitudes above 2438 m (8000 ft) + +– Maximum temperature when operating under the unpressurized condition: 31.3°C (88°F) + +Note: When you charge the battery, its temperature must be no lower than 10°C (50°F). + +Relative humidity + +• Operating: 8% to 95% at wet-bulb temperature 23°C (73°F) + +• Storage: 5% to 95% at wet-bulb temperature 27°C (81°F) + +Avoid constant body contact with specific hot sections +CAUTION: +When the computer is operating, it should be placed on a hard and flat surface with its bottom area +not in contact with user’s bare skin. Under normal operating conditions, the temperature of the +bottom surface will remain within an acceptable range as defined in IEC 62368-1, but such +temperatures can still be high enough to cause discomfort or harm to the user if directly touched for +over one minute at a time. As such, it is recommended that users avoid prolonged direct contact with +the bottom of the computer. + +10 + +User Guide + + Chapter 2. Get started with your computer + +Work with Windows + +The following table contains Windows frequently used settings. You can learn the basics and start working +with Windows right away. + +To configure settings, type the corresponding keywords in the Windows search box and select the best +match. Follow the on-screen instructions to customize settings. + +Table 1. Basic settings + +Functions + +Descriptions + +Control Panel + +View or change Windows settings, including hardware and software setup +and configuration. + +Connect to Wi-Fi +networks + +For models with a wireless LAN module, you can connect your computer to +a Wi-Fi® network. Click the network icon in the Windows notification area, +and then select a network for connection. +Note: The wireless LAN module on your computer may support different +standards. For some countries or regions, use of 802.11ax may be disabled +according to local regulations. + +Search by +keywords + +Control Panel + +Wi-Fi + +Connect to wired +networks + +For models with an Ethernet (RJ-45) connector, you can use an Ethernet +cable to connect your computer to a wired network. + +— + +Airplane mode + +Airplane mode is a convenient setting to turn all wireless communications of +your computer on and off. You may need to turn it on when boarding an +airplane. + +Airplane mode + +Night light mode + +Adjust color +temperature + +Windows +Updates + +Night light is a switch in Windows that you can turn on and off. When it is +turned on, your screen shows warmer colors and the amount of emitted +blue light is reduced. Enabling night light reduces the chances of developing +eye fatigue or eye strain. +Note: For more guidance on reducing eye fatigue or eye strain, go to +https://www.lenovo.com/us/en/compliance/visual-fatigue. + +Night light + +If night light mode is turned on, you can adjust the color temperature of the +screen. +Note: Selected Lenovo computers are low blue light certified. These +computers are tested with night light turned on and color temperature set to +the default value of 48. + +Night light + +Microsoft periodically releases feature and security updates to the Windows +operating system. Updates that are applicable to your Windows version are +downloaded automatically when your computer is connected to the +Internet. When updates are downloaded, you are prompted to restart the +computer to install these updates. You can also manually check if there are +available updates for the installed version of Windows. +Attention: Only use Windows Update to download and install updates. +Updates downloaded from other sources may include security risks. + +Windows Updates + +© Copyright Lenovo 2022 + +11 + + Table 2. Optional settings + +Functions + +Descriptions + +Enroll your +fingerprint + +For models with a fingerprint reader, you can enroll your fingerprint for +Windows sign-in. + +Search by +keywords + +Sign-in options + +Windows help information + +If the on-screen instructions cannot solve your problem, refer to the following to get the online Windows help +information. + +• Type Get Help or Tips in the Windows search box and then press Enter. When the app opens, type the + +problem description and select the matched result. + +• Visit Microsoft support Web site: https://support.microsoft.com. Enter what you are looking for in the + +search box and get search results. + +Lenovo Vantage and Lenovo PC Manager + +They are both apps developed by Lenovo. Using either of them, you can: + +• View product information and system status + +• Manage and change device settings + +• Check for and perform system update + +Note: This feature is only available on Lenovo Vantage. In Lenovo PC Manager, you can check for and +perform driver update. + +Depending on the country or region in which your computer is purchased, one of the apps may be pre- +installed on your computer. To open it, type its name in the Windows Search box and select the matched +result. If neither of the apps is pre-installed, you may install one on your own. Lenovo Vantage is available for +download free of charge from the Microsoft Store. + +Note: App features are subject to change with updates and specific features may not be available on all +Lenovo products. + +The Novo Button menu + +The Novo Button menu can be displayed before the operating system starts. From the menu, you can +choose to + +• Open the BIOS/UEFI setup utility + +• Open the boot device selection menu + +• Open the Windows startup options screen + +Note: From the Windows startup options screen, you can then choose to + +Start your computer using a recovery drive + +Reset your computer + +Open advanced options screen + +Open the Novo Button menu + +For Lenovo computers with a Novo button, you can press the button to open the Novo Button menu. + +12 + +User Guide + + Step 1. Turn off the computer. + +Step 2. Open the LCD screen and press the Novo button. + +Note: Alternatively, turn off the computer. Press Fn and the power button to open the Novo Button menu. + +Interact with your computer + +Keyboard hotkeys + +Some keys on the keyboard are printed with icons. These keys are called hotkeys and can be pressed alone +or in combination with the Fn key to open certain apps or change certain settings quickly. The functions of +the hotkeys are symbolized by the icons printed on them. + +Hotkey + +Function + +Mute/Unmute sound. + +Decrease/Increase system volume. + +Mute/Unmute the microphone. + +Decrease/Increase screen brightness. + +Switch display devices. + +Turn airplane mode on and off. + +Service key. + +Answer the Microsoft Teams calls. +Notes: + +• Supported on select versions of Microsoft Teams. + +• Install Microsoft Teams manually. + +Reject the Microsoft Teams calls. +Notes: + +• Supported on select versions of Microsoft Teams. + +• Install Microsoft Teams manually. + +Open the Windows Calculator app. + +Open Lenovo Vantage, Lenovo PC Manager, or an app launch bar. + +Open the Windows snipping tool. + +Play or pause playback of media. + +Stop playback of media. + +Go the next media file in the playlist or folder. + +Go to the previous media file in the playlist or folder. + +Turn on/off or adjust the brightness of the keyboard backlight. +Note: This hotkey must be used with the Fn key. + +Chapter 2. Get started with your computer + +13 + +* + +* + +* + +* + +* for 15-inch models only + + Hotkey mode +Some hotkeys share keys with function keys (F1 to F12). The hotkey mode is a UEFI/BIOS setting that +changes how hotkeys (or function keys) are used. + +Hotkey mode setting + +How to use hotkeys + +How to use function keys + +Disabled + +Enabled + +Hold down the Fn key and press one +of the hotkeys. + +Press the function keys directly. + +Press the hotkeys directly. + +Hold down the Fn key and press one +of the function keys. + +Note: Hotkeys that do not share keys with function keys are not affected by the hotkey mode setting. They +should always be used with the Fn key. + +Numeric keypad + +Some Lenovo computers includes a dedicated numeric keypad on the far right of the keyboard. The keypad +is used for entering numbers and operators quickly. + +Press the NumLock key to enable or disable the numeric keypad. + +Multi-touch gestures for touch-enabled screens + +Selected Lenovo computers include a touch-enabled screen. For such a computer, you can use your fingers +to interact with the operating system and apps that are gesture aware. + +Note: Not all gestures are supported by all apps. + +Function + +Selecting an item + +Double-clicking an item + +Right-clicking an item + +Gesture + +Tap + +Tap twice quickly + +Tap and hold + +14 + +User Guide + + Gesture + +Pinch + +Spread + +Tap from the left edge + +Tap and hold + +Function + +Zooming out + +Zooming in + +Viewing all open apps + +Displaying the Action Center of Windows + +Chapter 2. Get started with your computer + +15 + + 16 + +User Guide + + Chapter 3. Explore your computer + +Manage power + +Use the information in this section to achieve the best balance between performance and power efficiency. + +Check the battery status + +The battery status icon +the current power plan, and access battery settings quickly. + +or + +is in the Windows notification area. You can check the battery status, view + +Click the battery status icon to display the percentage of battery power remaining and change the power +mode. A warning message is displayed when the battery is low. + +Charge the battery + +When the remaining battery power is low, charge your battery by connecting your computer to ac power. + +The battery is fully charged in about two to four hours. The actual charging time depends on the battery +capacity, the physical environment, and whether you are using the computer. + +Battery charging is also affected by its temperature. The recommended temperature range for charging the +battery is between 10°C (50°F) and 35°C (95°F). + +Note: + +You can check the battery temperature in Lenovo Vantage or Lenovo PC Manager. + +To maximize the life of the battery, once the battery is fully charged, it must discharge to 94% or lower +before it will be allowed to recharge again. + +To extend your battery’s lifespan, the battery may optimize its full charge capacity based on your usage. +After prolonged periods of limited use, full battery capacity may not be available until you discharge to as +low as 20% and recharge completely. For more information, refer to the power section of Lenovo +Vantage. + +Set power button behavior + +By default, pressing the power button puts the computer to sleep mode. However, you can change the +power button behavior in Windows Control Panel. + +Step 1. Type Control Panel in the Windows search box and then press Enter. Open the control panel and + +view by large or small icons. + +Step 2. Select the power options and then click choose what the power button does. + +A power plan + +A power plan is a collection of power-saving settings made available by an operating system. With a power +plan, you can set idle time-outs for different hardware components to enter a low-power state. The default +power plan and some of its idle time-out settings are listed below for computer models pre-installed with +Windows. + +The listed settings are active when the computer is connected to an electrical outlet. If your computer include +a built-in battery pack, a different collection of time-out settings is set to take effect when the computer is +operating on battery power. + +© Copyright Lenovo 2022 + +17 + + • Default power plan: Balanced + +• Turn off the display: after 10 minutes + +• Put the computer to sleep: after 20 minutes + +Note: To wake the computer from the sleep state, press the power button or any key on the keyboard. + +Change or customize a power plan +This operation is applicable to computers with pre-installed Windows. + +Step 1. Type power plan in the Windows search box and then press Enter. + +Step 2. Customize a power plan of your preference. + +Set performance mode + +Your computer can work in different performance modes. The performance and power consumption of the +computer vary in different performance modes. Press Fn + Q to cycle through different performance modes. + +Change settings in UEFI/BIOS setup utility + +This section introduces what is UEFI/BIOS and the operations you can perform in its setup utility. + +What is UEFI/BIOS setup utility + +UEFI/BIOS is the first program that runs when a computer starts. UEFI/BIOS initializes hardware components +and loads the operating system and other programs. Your computer may include a setup program (setup +utility) with which you can change certain UEFI/BIOS settings. + +Open the UEFI/BIOS setup utility + +Step 1. Turn on or restart the computer. + +Step 2. When the Lenovo logo appears on the screen, press F1 repeatedly. + +Note: For computers with hotkey mode enabled, press Fn + F1. + +Change the startup sequence + +You may need to change the startup sequence to allow the computer to start from another device or a +network location. + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Navigate to the Boot menu. + +Step 3. Follow on-screen instructions to change the order of devices under Boot Priority Order. + +Note: You can also change startup sequence temporarily without opening the setup utility. To do so, start +the computer. When the Lenovo logo appears, press F12 (or Fn + F12) repeatedly. + +Change hotkey mode + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Configuration ➙ Hotkey Mode and press Enter. + +Step 3. Change the setting to Disabled or Enabled. + +Step 4. Select Exit ➙ Exit Saving Changes. + +18 + +User Guide + + Enable or disable always-on + +For some Lenovo computers with always-on connectors, the always-on function can be enabled or disabled +in the UEFI/BIOS setup utility. + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Configuration ➙ Always On USB and press Enter. + +Step 3. Change the setting to Disabled or Enabled. + +Step 4. Select Exit ➙ Exit Saving Changes. + +Set passwords in UEFI/BIOS setup utility + +This section introduces the types of passwords that you can set in the UEFI (Unified Extensible Firmware +Interface) or BIOS (Basic Input/Output System) setup utility. + +Password types + +You can set various types of passwords in the UEFI/BIOS setup utility. + +Password type + +Pre-requisite + +Usage + +Administrator password + +No + +You must enter it to start the setup +utility. + +User password + +The administrator password must be +set. + +You can use the user password to +start the setup utility. + +Master hard disk password + +No + +User hard disk password + +The master hard disk password must +be set. + +You must enter it to start the +operating system. + +You can use the user hard disk +password to start the operating +system. + +Notes: + +• All passwords set in the setup utility consist of alphanumeric characters only. + +• If you start the setup utility using the user password, you can only change a few settings. + +Set administrator password + +You set the administrator password to prevent unauthorized access to the UEFI/BIOS setup utility. + +Attention: If you forget the administrator password, a Lenovo authorized service personnel cannot reset +your password. You must take your computer to a Lenovo authorized service personnel to have the system +board replaced. Proof of purchase is required and a fee will be charged for parts and service. + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Security ➙ Set Administrator Password and press Enter. + +Step 3. Enter a password string that contains only letters and numbers and then press Enter + +Step 4. Enter the password again and press Enter. + +Step 5. Select Exit ➙ Exit Saving Changes. + +Chapter 3. Explore your computer + +19 + + Next time you start the computer, you must enter the administrator password to open the setup utility. If +Power on Password is enabled, you must enter the administrator password or the user password to start +the computer. + +Change or remove administrator password + +Only the administrator can change or remove the administrator password. + +Step 1. Open the UEFI/BIOS setup utility using the administrator password. + +Step 2. Select Security ➙ Set Administrator Password and press Enter. + +Step 3. Enter the current password. + +Step 4. + +In the Enter New Password text box, enter the new password. + +Step 5. + +In the Confirm New Password text box, enter the new password again. + +Note: If you want to remove the password, press Enter in both text boxes without entering any +character. + +Step 6. Select Exit ➙ Exit Saving Changes. + +If you remove the administrator password, the user password is also removed. + +Set user password + +You must set the administrator password before you can set the user password. + +The administrator of the setup utility might need to set a user password for use by others. + +Step 1. Open the UEFI/BIOS setup utility using the administrator password. + +Step 2. Select Security ➙ Set User Password and press Enter. + +Step 3. Enter a password string that contains only letters and numbers and then press Enter. The user + +password must be different from the administrator password. + +Step 4. Enter the password again and press Enter. + +Step 5. Select Exit ➙ Exit Saving Changes. + +Enable power-on password + +If the administrator password has been set, you can enable power-on password to enforce greater security. + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Security ➙ Power on Password and press Enter. + +Note: The administrator password must be set in advance. + +Step 3. Change the setting to Enabled. + +Step 4. Select Exit ➙ Exit Saving Changes. + +If power-on password is enabled, a prompt appears on the screen every time you turn on the computer. You +must enter the administrator or user password to start the computer. + +Set hard disk password + +You can set a hard disk password in the setup utility to prevent unauthorized access to your data. + +20 + +User Guide + + Attention: Be extremely careful when setting a hard disk password. If you forget the hard disk password, a +Lenovo authorized service personnel cannot reset your password or recover data from the hard disk. You +must take your computer to a Lenovo authorized service personnel to have the hard disk drive replaced. +Proof of purchase is required and a fee will be charged for parts and service. + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Security ➙ Set Hard Disk Password and press Enter. + +Note: If you start the setup utility using the user password, you cannot set hard disk password. + +Step 3. Follow on-screen instructions to set both master and user passwords. + +Note: The master and user hard disk passwords must be set at the same time. + +Step 4. Select Exit ➙ Exit Saving Changes. + +If the hard disk password is set, you must provide the correct password to start the operating system. + +Change or remove hard disk password + +Step 1. Open the UEFI/BIOS setup utility. + +Step 2. Select Security. + +Step 3. Change or remove the hard disk password. + +To change or remove master password, select Change Master Password and press Enter. + +Note: If you remove the master hard disk password, the user hard disk password is also removed. + +To change user password, select Change User Password and press Enter. + +Note: The user hard disk password cannot be removed separately. + +Step 4. Select Exit ➙ Exit Saving Changes. + +Chapter 3. Explore your computer + +21 + + 22 + +User Guide + + Chapter 4. Help and support + +Frequently asked questions + +How do I partition my storage +drive? + +Refer to https://support.lenovo.com/solutions/ht503851 + +What should I do if my computer +stops responding. + +Press and hold the power button until the computer turns off. Then restart the +computer. + +What should I do if I spill liquid on +the computer? + +1. Carefully unplug the ac power adapter and turn off the computer immediately. +The more quickly you stop the current from passing through the computer the +more likely you will reduce damage from short circuits. + +Attention: Although you might lose some data or work by turning off the +computer immediately, leaving the computer on might make your computer +unusable. + +2. Wait until you are certain that all the liquid is dry before turning on your + +computer. + +CAUTION: +Do not try to drain out the liquid by turning over the computer. If your +computer has keyboard drainage holes on the bottom, the liquid will be +drained out through the holes. + +Where can I get the latest device +drivers and UEFI/BIOS? + +• Lenovo Vantage or Lenovo PC Manager + +• Lenovo Support Web site at https://support.lenovo.com + +Self-help resources + +Use the following self-help resources to learn more about the computer and troubleshoot problems. + +Resources + +How to access? + +Troubleshooting and frequently asked questions + +• https://www.lenovo.com/tips + +• https://forums.lenovo.com + +Accessibility information + +https://www.lenovo.com/accessibility + +Reset or restore Windows + +• Use Lenovo recovery options. + +1. Go to https://support.lenovo.com/ +HowToCreateLenovoRecovery. + +2. Follow the on-screen instructions. + +• Use Windows recovery options. + +1. Go to https://pcsupport.lenovo.com. + +2. Detect your computer or manually select your + +computer model. + +3. Click Diagnostics ➙ Operating System + +Diagnostics and then follow the on-screen +instructions. + +© Copyright Lenovo 2022 + +23 + + Resources + +How to access? + +Use Lenovo Vantage or Lenovo PC Manager to: + +• Download and install the latest drivers and firmware. + +• Configure hardware settings + +Use Windows Search. + +• Diagnose computer hardware problems. + +• Check the computer warranty status. + +Product documentation: + +• Generic Safety and Compliance Notices + +• Safety and Warranty Guide + +• Setup Guide + +• This User Guide + +• Regulatory Notice + +Lenovo Support Web site with the latest support +information of the following: + +• Drivers and software + +• Diagnostic solutions + +• Product and service warranty + +• Product and parts details + +• Knowledge base and frequently asked questions + +Windows help information + +What is a CRU? + +1. Go to https://support.lenovo.com. + +2. Detect your computer or select computer model + +manually. + +3. Select Documentation and filter out the + +documentation you want. + +Visit https://support.lenovo.com + +• Use Get Help or Tips. + +• Use Windows Search or the Cortana® personal + +assistant. + +• Microsoft support Web site: https:// + +support.microsoft.com + +Customer replaceable units (CRUs) are parts that can be upgraded or replaced by the customer. A Lenovo +computer may contain the following types of CRUs: + +Self-service CRU + +Optional-service CRU + +Parts that can be installed or replaced easily by +customer themselves or by trained service +technicians at an additional cost. + +Parts that can be installed or replaced by customers +with a greater skill level. Trained service technicians +can also provide service to install or replace the +parts under the type of warranty designated for the +customer’s machine. + +If you intend to install a CRU, Lenovo will ship the CRU to you. You might be required to return the defective +part that is replaced by the CRU. When return is required: (1) return instructions, a prepaid shipping label, +and a container will be included with the replacement CRU; and (2) you might be charged for the +replacement CRU if Lenovo does not receive the defective CRU within thirty (30) days of your receipt of the +replacement CRU. For full details, see the Lenovo Limited Warranty at https://www.lenovo.com/warranty/llw_ +02. + +24 + +User Guide + + CRUs for your product model + +The table below lists the CRUs and CRU types that are defined for your product model. + +Part + +Power cord * + +ac power adapter + +* for selected models + +Notes: + +Self-service CRU + +Optional-service CRU + +X + +X + +• CRU replacement instruction is provided in one or more of the following publications and are available + +from Lenovo at any time upon your request. + +the product User Guide + +the printed publications that came with the product + +• Replacement of any parts not listed above, including the built-in rechargeable battery, must be done by a +Lenovo-authorized repair facility or technician. Go to https://support.lenovo.com/partnerlocation for more +information. + +Call Lenovo + +If you have tried to correct the problem yourself and still need help, you can call Lenovo Customer Support +Center. + +Before you contact Lenovo + +Record product information and problem details before you contact Lenovo. + +Product information + +• Product name + +Problem symptoms and details + +• What is the problem? Is it continuous or intermittent? + +• Machine type and serial number + +• Any error message or error code? + +• What operating system are you using? Which version? + +• Which software applications were running at the time + +of the problem? + +• Can the problem be reproduced? If so, how? + +Note: The product name and serial number can usually be found on the bottom of the computer, either +printed on a label or etched on the cover. + +Lenovo Customer Support Center + +During the warranty period, you can call Lenovo Customer Support Center for help. + +Telephone numbers + +For a list of the Lenovo Support phone numbers for your country or region, go to https:// +pcsupport.lenovo.com/supportphonelist. + +Chapter 4. Help and support + +25 + + Note: Phone numbers are subject to change without notice. If the number for your country or region is not +provided, contact your Lenovo reseller or Lenovo marketing representative. + +Services available during the warranty period + +• Problem determination - Trained personnel are available to assist you with determining if you have a + +hardware problem and deciding what action is necessary to fix the problem. + +• Lenovo hardware repair - If the problem is determined to be caused by Lenovo hardware under warranty, + +trained service personnel are available to provide the applicable level of service. + +• Engineering change management - Occasionally, there might be changes that are required after a product +has been sold. Lenovo or your reseller, if authorized by Lenovo, will make selected Engineering Changes +(ECs) that apply to your hardware available. + +Services not covered + +• Replacement or use of parts not manufactured for or by Lenovo or nonwarranted parts +• Identification of software problem sources +• Configuration of UEFI/BIOS as part of an installation or upgrade +• Changes, modifications, or upgrades to device drivers +• Installation and maintenance of network operating systems (NOS) +• Installation and maintenance of programs + +For the terms and conditions of the Lenovo Limited Warranty that apply to your Lenovo hardware product, +see “Warranty information” in the Safety and Warranty Guide that comes with your computer. + +Purchase additional services + +During and after the warranty period, you can purchase additional services from Lenovo at https:// +pcsupport.lenovo.com/warrantyupgrade. + +Service availability and service name might vary by country or region. + +26 + +User Guide + + Appendix A. Notices and trademarks + +Notices + +Lenovo may not offer the products, services, or features discussed in this document in all countries. Consult +your local Lenovo representative for information on the products and services currently available in your +area. Any reference to a Lenovo product, program, or service is not intended to state or imply that only that +Lenovo product, program, or service may be used. Any functionally equivalent product, program, or service +that does not infringe any Lenovo intellectual property right may be used instead. However, it is the user's +responsibility to evaluate and verify the operation of any other product, program, or service. + +Lenovo may have patents or pending patent programs covering subject matter described in this document. +The furnishing of this document does not give you any license to these patents. You can send license +inquiries, in writing, to: +Lenovo (United States), Inc. +8001 Development Drive +Morrisville, NC 27560 +U.S.A. +Attention: Lenovo Director of Licensing + +LENOVO PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS +OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, +MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some jurisdictions do not allow +disclaimer of express or implied warranties in certain transactions, therefore, this statement may not apply to +you. + +Changes are made periodically to the information herein; these changes will be incorporated in new editions +of the publication. To provide better service, Lenovo reserves the right to improve and/or modify the +products and software programs described in the manuals included with your computer, and the content of +the manual, at any time without additional notice. + +The software interface and function and hardware configuration described in the manuals included with your +computer might not match exactly the actual configuration of the computer that you purchase. For the +configuration of the product, refer to the related contract (if any) or product packing list, or consult the +distributor for the product sales. Lenovo may use or distribute any of the information you supply in any way it +believes appropriate without incurring any obligation to you. + +The products described in this document are not intended for use in implantation or other life support +applications where malfunction may result in injury or death to persons. The information contained in this +document does not affect or change Lenovo product specifications or warranties. Nothing in this document +shall operate as an express or implied license or indemnity under the intellectual property rights of Lenovo or +third parties. All information contained in this document was obtained in specific environments and is +presented as an illustration. The result obtained in other operating environments may vary. + +Lenovo may use or distribute any of the information you supply in any way it believes appropriate without +incurring any obligation to you. + +Any references in this publication to non-Lenovo Web sites are provided for convenience only and do not in +any manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the +materials for this Lenovo product, and use of those Web sites is at your own risk. + +© Copyright Lenovo 2022 + +27 + + Any performance data contained herein was determined in a controlled environment. Therefore, the result +obtained in other operating environments may vary significantly. Some measurements may have been made +on development-level systems and there is no guarantee that these measurements will be the same on +generally available systems. Furthermore, some measurements may have been estimated through +extrapolation. Actual results may vary. Users of this document should verify the applicable data for their +specific environment. + +This document is copyrighted by Lenovo and is not covered by any open source license, including any Linux +agreement(s) which may accompany software included with this product. Lenovo may update this document +at any time without notice. + +For the latest information or any questions or comments, contact or visit the Lenovo Web site: + +https://support.lenovo.com + +Trademarks + +Lenovo and the Lenovo logo are trademarks of Lenovo. Thunderbolt is a trademark of Intel Corporation or its +subsidiaries. Microsoft, Windows, and Cortana are trademarks of the Microsoft group of companies. +DisplayPort is a trademark of the Video Electronics Standards Association. The terms HDMI and HDMI High- +Definition Multimedia Interface, and the HDMI logo are trademarks or registered trademarks of HDMI +Licensing Administrator, Inc. in the United States and other countries. Wi-Fi is a registered trademark of Wi- +Fi Alliance. USB-C is a trademark of USB Implementers Forum. All other trademarks are the property of their +respective owners. + +28 + +User Guide diff --git a/q192/random_k2/question.json b/q192/random_k2/question.json index 7858851cff87179a016711d7b7a2297fc7d91be5..8bc095b6513a3ef909c76e59ab18974f1fa26db7 100644 --- a/q192/random_k2/question.json +++ b/q192/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "Guide-for-international-students-web.pdf", - "web_7795b10bf9b6a434.pdf", - "web_ca39b8721f6d1255.pdf" + "web_f3420c2dcb7612a6.pdf", + "web_e5f79db0bc92a9fa.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q192/random_k2/random_1.md b/q192/random_k2/random_1.md index 4c601abeb6f180fb2f07694cada010de6f7dd010..c628675fbe529d89838de1ae7c364d79b8d4ce01 100644 --- a/q192/random_k2/random_1.md +++ b/q192/random_k2/random_1.md @@ -1,462 +1,1223 @@ -Proposal for the Theme on Big Data -Analytics +Online BSc Degree +and Diploma Program +PROGRAMMING AND DATA SCIENCE -Qiang Yang, HKUST +Batch 1: Starts January 2021 -Jiannong Cao, PolyU +For latest updates, visit - www.onlinedegree.iitm.ac.in -Qi-man Shao, CUHK +Brochure Last Updated: 2nd November 2020 -May 2015 + Contents - Motivation +Introduction -• +Why learn Programming & Data Science -The world's technological per-capita capacity to +About IIT Madras -store information doubled every 40 months +Our Experience in Online Education -– +Academics -As of 2012, 2.5 exabytes (2.5×1018) of data/day -Relational database management systems and -• -desktop statistics and visualization packages often -have difficulty handling big data. +Overall Structure of the programme -– +Term Structure -Big Data: new driver for digital economy&society +Course Structure - +Assessments - + Online Assignments - + Three Quizzes -Gartner: hundreds of billions of GDP by 2020. -Intangible factor after labor and capital -Data Science: The fourth paradigm + End Term Exam -2 + Pass Criteria for Each Course - The Power of Big Data + Learners Who Do Not Pass a Course -• +Exam Cities -• +Fee Structure -• +Foundational Level -• +Diploma Level -Big Data can bring “big values” to our life in -almost every aspects. -Technologically, Big Data is bringing about changes in our lives because it -allows diverse and heterogeneous data to be fully integrated and -analyzed to help us make decisions. -Today, with the Big Data technology, thousands of data from seemingly -unrelated areas can help support important decisions. This is the -power of Big Data. -Areas of Applications + Diploma in Programming -– + Diploma in Data Science -– +Degree Level -– +Admissions -– +Important Dates -– +Two Admission Paths -Health and Well being + Regular Entry -Policy making and public opinions + Diploma Only Entry -Smart cities and more efficient society +Mandatory Requirements -New online educational models: MOOC and Student-Teacher modeling +Regular Entry -Robotics and human-robot interaction + Eligibility to Apply -• + Application Process -Much of this power hinges on Research on Analytics + Qualifier Process + + Qualifier Exam and Pass Criteria + +Contact Us 3 - Hong Kong needs Big Data -Research +3 + +4 + +4 + +5 + +5 + +6 + +6 + +7 + +7 + +7 + +7 + +8 + +8 + +8 + +9 + +10 + +10 + +11 + +11 + +12 + +13 + +13 + +13 + +13 + +13 + +13 + +14 + +14 + +14 + +14 + +15 + +16 + + Introduction + +IIT Madras, India’s top technical institute, welcomes you to the world’s first online BSc +Degree program in Programming and Data Science. For the first time, you can work +towards an undergraduate degree / diploma from an IIT regardless of your age or +location, and with a wide range of academic backgrounds. + +“ + +IIT Madras is pioneering a completely online program to +produce graduates who will be employable in the sector +that presently has high demand - Programming and Data +Science. The program uses an innovative combination of +online learning and in-person assessment and is flexible and +affordable. I believe the program will be a gamechanger +in the field of education in India in the coming years and +inspire other top-ranking institutions to launch many such +online initiatives. + +- Prof. Bhaskar Ramamurthi, Director, IIT Madras + +Why Programming & Data Science? + +Decision making is increasingly becoming +data driven within the commercial sector +and the government. There is a large +quantity of data readily available for this +at present. Smart devices connected to +the internet bring more input sources, +apart from those generated by humans +as they interact with services and one +another. Data driven decision making +involves the analysis of large volumes +of data to identify patterns and build +predictive models. This requires a +combination of skills ranging from +computing, statistics and mathematics, +and is broadly labelled data science. + +INCREASING DEMAND + +BETTER JOB OPPORTUNITIES + +UPSKILL + +As formal academic programs in data +science are only just emerging, there +is a huge gap between the demand for +data scientists and the supply of suitable +qualified applicants in the job market. +The throughput of traditional classroom +programs is limited and will not be able +to provide the numbers needed to meet +the upcoming requirements. Online +teaching has the potential to train much +larger numbers, but delivering scale +without diluting quality has proven to be +a challenge until recently. + +Our program therefore aims to create +a thoroughly trained programmer, +proficient in application development, +data sciences and machine learning. + +3 + + About IIT Madras + +Indian Institute of Technology Madras is one of the +foremost institutes of national importance for higher +technical education and research. + +In 1956, the German Government offered +technical assistance in establishing +an institute of higher education in +engineering in India. The first Indo- +German agreement in Bonn, West +Germany for the establishment of the +Indian Institute of Technology at Madras +was signed in 1959. The Institute with its +campus in Chennai, previously known +as Madras, was formally inaugurated +in 1959 by Prof. Humayun Kabir, Union +Minister for Scientific Research and +Cultural Affairs. + +The Institute today has sixteen academic +departments and several advanced +research centres in various disciplines +of engineering and pure sciences. A +faculty of international repute, a brilliant +student community, excellent technical +& supporting staff and an effective +administration have all contributed to the +pre-eminent status of IIT Madras. + +Our Experience in Online Education + +IITM is well-equipped to provide an +online degree course. We have been +ranked number 1 Engineering Institute +for the past 5 years and we have been +ranked number 1 Overall for the past 2 +years by NIRF. + +Furthermore, IITM began the NPTEL +program, the country’s first online +content portal as an inter-IIT consortium +project in 2001 and nptel.ac.in  is one +of the largest online course portals in +the world. In 2014, IITM also created +NPTEL online courses - the country’s first +Massive Open Online Course (MOOC). +Today, NPTEL is considered the largest +MOOC provider in India. + +MHRD launched the SWAYAM MOOC +platform in 2017 and NPTEL has been +the largest contributor to it. NPTEL +also manages swayam.gov.in  - the +national MOOCs portal, as well as the +network of more than 3800 SWAYAM +local chapter colleges. With IITM +continuing to lead the NPTEL project, it +makes them the perfect choice to offer +an online degree program. + +2001 +NPTEL +Program + +2014 +NPTEL +Online +Courses + +2017 +SWAYAM +MOOC + +2020 +IITM +Online +Degree + +4 + + Overall Structure + +There are THREE LEVELS in our Online +Degree program and to get a BSc +Degree in Programming and Data +Science from IIT Madras, a learner +has to successfully complete all three +levels. 1. +FOUNDATIONAL LEVEL 2. +DIPLOMA LEVEL -to develop state-of-the-art Big -Data platform in research, -education and industrial -applications, and open it to the -Hong Kong society and the world -at large, and +3. +DEGREE LEVEL -to make a difference in Smart -Cities, Health and Well-being -(including supporting aging -populations), and modernizing -Finance, Education and Logistics -in Hong Kong. +There is also the flexibility to exit at +any level. Depending on the courses +completed, the learner can receive +Foundational Certificate from Centre +for Continuing Education, IIT Madras +or Diploma(s) from IIT Madras. (Pg 13 +for Admissions related information) -4 +We are currently formulating a +Diploma Only Entry (Pg 13) option for +those who are interested in doing only +Diploma Level courses and exiting +with one or both Diploma(s) from IIT +Madras. + +Total Courses: 31 +116 credits + +Completion time: 3 - 6 years +based on preferred pace & performance - Big Data Analytics Objectives +Fees: Based on courses registered +*Refer Fee Structure (Pg 9). + +Online Courses & Assignments +12 weeks of coursework for each course + +Quizzes and Exams +3 Quizzes and one End Term Exam for +each course - in person and invigilated. + +ACADEMICS 5 - Relation to Smart Cities and IoT + Term Structure + +Every year is divided into three +terms - January Term, May Term and +September Term. + +Before progressing to the next level, +the learner has to clear all the courses +in the current level. + +In a term, a learner may register +for a course only if they have +already successfully completed the +prerequisites for that course. + +The learner may be allowed to +register for upto 4 courses in a term +depending on their performance in +the previous exam(s). + +A learner has to clear all Foundational +level courses successfully, before +registering for any Diploma Level +courses. Similarly, learners have +to clear Diploma Level courses +successfully before registering for any +Degree Level courses. + +Course Structure + +The duration of the core courses +in Foundational, Diploma and +Degree Level is 12 weeks and the +courses carry 4 credits each. +The Skill Enhancement courses +and Electives in Degree Level +have varying credits and may +be of different duration. + +The fee for each term depends on the +total credits one registers for in that +term. Refer Fee Structure (Pg 9). + +The recorded video content released +every week will comprise a series of +lectures, sample problems and step +by step instructions on how to solve +them. + +The duration of each video would be +15 to 30 minutes amounting to a total +of about 2 - 3 hours of video content +each week in each course. + +Weekly assignment(s), which will be +released along with the videos, will +have to be completed and submitted +online in the specified period of time. + +There will be 3 quizzes for each +course, one each at the end of weeks +4, 7 and 10. Each course further +culminates in an End Term Exam. All +the quizzes and the end term exams +have to be completed by the learner, +in person, at designated exam centres, +held under invigilation. + +All in-person, invigilated quizzes and +exams will be scheduled for weekends, +subject to availability of dates with +exam partner. -• +6 -• + Assessments -• +There are 3 types of assessments for each course - online weekly assignments, +monthly in-person invigilated quizzes and an in-person invigilated end term exam. -World economic forum -ranking HK’s -infrastructure: #1 -– +If the Average Assignment Score for a +course is less than 40/100, the learner +will not be eligible to write the end +term exam for that course. -Maintain the lead in IT +Online Assignments -Infrastructure +A 12-week course will have at least one +assignment every week. -East Kowloon Project: -Energizing Hong Kong -via Smart Cities -Big Data: -– +Each assignment has to be submitted +online within the specified due date. -IoT provides the +From the 12 weekly assignment scores +for a course, a learner’s best 8 scores +will be considered to calculate the +Average Assignment Score. -infrastructure for +Three Quizzes +IN PERSON, INVIGILATED -collecting the data +There will be 3 quizzes for each course, one each at the end of Weeks 4, 7 and 10. +Every quiz will be based on content covered up to that point in the course. These +quizzes have to be attempted in person, in the specified exam city (Pg 9). -Smart Cities as +Each quiz will be conducted in a single +session and will consist of sections of +45 minutes duration for every course +chosen. -important application +It is mandatory to appear for at least +one out of the three quizzes to be +eligible to write the end term exam. -goal +The best two out of the three quiz +scores will be considered to calculate +the Average Quiz Score (Q) out of 100. -– +40% of Q will be used to calculate Total +Course Score T and course grade. -6 +End Term Exam - Research Objectives +IN PERSON, INVIGILATED -• +At the end of a term, there will be an +end term exam of 3 hours duration for +each course. -• +This exam score (out of 100) is called +End Term Score (E) which accounts for +60% of the Total Course Score (T). -• +To be eligible to write the End Term +Exam, a learner’s average assignment +score should be >= 40/100 AND they +should have appeared for at least one +out of the three quizzes. -• +7 -Big Data Analytics: data mining and machine learning -– + Pass Criteria for Each Course -Large-scale machine learning, data mining and data visualization +Total Course Score (T) = 40% of Average quiz score (Q) + 60% of End Term Score (E) -Big Data Computing: data center support for Analytics -– +A learner is deemed to have passed a course only if the below mentioned two +conditions are met: -Big data collection and transformation, integration and distributed -data management and computing +Total Course Score (T) >= 50/100 AND -Big data sampling and statistical theory, Big data security and privacy +End Term exam score (E) >= 40/100 -Big Data Theory, Privacy&Security issues on Analytics -– -Big Data Science: 4th Paradigm – Analytics for Science and -Engineering -– +These criteria are the same for all learners participating in the course. -Big Data and Multi-disciplines (Bio, Chemistry, Engineering, Social) +An absolute grading system, as described in the below table, will be followed: -7 +Condition satisfied by ‘T’ - Total Course Score (Out Of 100) +and ‘E’ - End Semester Score (Out Of 60) + +Letter Grade + +(T ≥ 90) AND (E ≥ 24) - Why Hong Kong is Ready for the Theme +(90 > T ≥ 80) AND (E ≥ 24) -• +(80 > T ≥ 70) AND (E ≥ 24) -We have the best researchers in machine learning, data mining, data -management, sensor networks, statistics, and multidisciplinary -research such as bioinformatics +(70 > T ≥ 60) AND (E ≥ 24) -– +(60 > T ≥ 50) AND (E ≥ 24) -– +S -– +A -China National 973 Projects on Big Data +B -IEEE Transactions on Big Data: EiC +C -ACM KDD Conferences: PC and Conference Chairs +D -Winner of Big Data related international competitions +(T < 50) OR (E < 24) OR Average assignment score <40/100 OR No quiz attended -– -New industries based on lots of data +U (Fail) -– +Learners Who Do Not Pass a Course -Financial industry, logistics industry, education sector, government -services, etc. +For any learner, if Total Course Score (T) < 50/100 or average assignment +score < 40/100 or no quiz attended, they have to repeat the entire course. This +includes submission of assignments and redoing all the subsequent quizzes and +end term exam. Fees for repeating the course is the same as the course fees. -We have many potential collaborators and partners -– +If Total Course Score (T) >= 50/100 but End Term Score (E) < 40/100, the learner has +2 options: -Huawei, Tencent, Baidu, Alibaba, Google, Microsoft, etc. +Attend only the end term exam whenever it is conducted in the subsequent +term. The fee to repeat an end term exam will be ₹1000 for foundational and +₹2000 for diploma / degree level courses. -• +They can repeat the entire course by paying the full course fees and +completing all assignments, invigilated quizzes and the end term exam. -• +Exam Cities + +The Invigilated Quizzes and End Term +exams are scheduled to be conducted +in a limited number of cities across +India. Learners can select 3 options +among the cities, and they will be +expected to attend the exam in the +City / Centre that has been allotted +to them. The allotted city will be from +these 3 options. + +The list of exam cities where +invigilated exams will be scheduled +may change later depending on the +location of majority of our learners. +Here is the link to view the current +exam cities . 8 - Big Data Analytics Workflow - - 9 - -Data -Extraction - -Data -Integration - -• HKUST -CUHK -• -Baptist -• -HSBC -• -Astri -• - -• HKUST -HKU -• -City U -• -Alibaba -• -Astri -• - -Applications - -• Biology and Genetics -• -• -• -• -• -• -• -• - -Chemistry -Physics -Government Policies -Social Sciences -General Health -Logistics -Finance -Business - -Data Mining & -Visualization - -• HKUST -CUHK -• -City U -• -Huawei -• -Tencent -• - -Big Data -Computing &Data -Management - -• CUHK -• -• -• -• -• - -HKUST -HKU -PolyU -City U -Baptist U - - Multi-disciplinary Big-data Analytics - -• - -Objectives: -• - -Interdisciplinary, -research -technological big data analytics - -heterogeneous - -multi-university, - -on - -multi-team -and - -scientific + Fee Structure + +For details about application fees, check Application Process (Pg 14) section. + +Each term, pay only for the courses you register for! + +Foundational Level - ₹4000 per course. +Skill Enhancement courses in Diploma / Degree Level - ₹5,000 per course. +All other courses in Diploma / Degree Level - ₹10,000 per course. + +Fee waivers depend on category of learner and family income: + +Family Income > 5LPA + +Family Income > 1LPA +and <=5LPA + +Family Income <=1LPA + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +General + +Full Fee + +OBC + +Full Fee + +- + +- + +50% waiver + +EWS + Family +Income + +75% waiver + +EWS + Family +Income + +50%waiver + +OBC-NCL + +Family Income + +75% waiver + +OBC-NCL + +Family Income + +SC / ST + +50% waiver + +SC / ST + +50% waiver + +SC / ST + +75% waiver + +PwD + +50% waiver + +PwD + +50% waiver + +PwD + +75% waiver + +SC / ST + Family +Income + +EWS / OBC-NCL + +PwD + Family Income + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +The term family income for the purpose of availing fee waivers includes the income +of the candidate, the income of his/her parents and spouse, also the income of his/ +her siblings and children below the age of 18 years. + +OBC-NCL / EWS / Family Income certificates need to be obtained in the prescribed +format. Only those who clear Qualifier Exam and wish to avail fee waiver based on +family income need to submit the Family Income certificate. (Note that Family Income +certificate is valid only for one year and will need to be submitted afresh each year to +continue availing corresponding fee waiver.) +Format for OBC-NCL certificate  +Format for EWS certificate  +Format for Family Income certificate  + +Tentative Fee for Each Term, depending on your end Goal and possible paths to +reach the goal: + + Foundational Certificate  + + Diploma in Programming  | Diploma in Data Science  | Both Diplomas  + + BSc in Programming and Data Science  + +*Note: If a learner does not pass a course in the term they registered for it - refer +section about Learners who do not pass a course for details on whether they will +need to repeat entire course or only repeat the end term exam. Fee to repeat an entire +course will be the same as the full course fee. Fee to only repeat the end term exam is +₹1000 for foundational level courses and ₹2000 for diploma / degree level courses. + +9 + + Foundational Level + +The Foundational Level comprises +courses in Mathematics, Statistics, Basics +of Programing and Python, and English. +These courses have been chosen to +ensure that the learner who passes these +successfully is well prepared to proceed +to the Diploma Level courses. + +8 courses +32 credits + +1 - 3 years +10 hrs/course/week + +₹32,000* +*Refer Fee Structure (Pg 9). + +Requirements for registration + +The learner should apply for and clear the Qualifier Process (Pg 14). + +Options on successful completion + +Learners have the following two options when they successfully complete all 8 +Foundational Level courses: + +Exit: The learner may exit with a Foundational Certificate from Centre for +Continuing Education, IIT Madras. + +Proceed to next level: The learner can join the Diploma Level (Pg 10). + +Courses in Foundational Level + +Mathematics for Data Science I + +Statistics for Data Science I + +Mathematics for Data Science II + +Statistics for Data Science II + +Computational Thinking + +Programming in Python + +English I + +English II + +Diploma Level + +There are two sections in the Diploma +Level with courses for Diploma in +Programming and courses for Diploma +in Data Science. Each of these diplomas +comprise 5 core courses and 1 skill +enhancement course. These Diploma +Level courses assume that the learner is +well-versed with fundamentals. + +Requirements for registration + +6 + 6 courses +22 + 22 credits + +1 - 3 years +10 hrs/course/week + +₹55,000* + ₹55,000* +*Refer Fee Structure (Pg 9). + +The learner should have cleared all 8 Foundational Level (Pg 10) courses or should +have joined via Diploma Only Entry (Pg 13). 10 - Why Big Data needs Team Work? + Options on successful completion + +Learners have the following options based on the courses completed in this level: + +They may exit with a Diploma in Programming from IIT Madras. + +They may exit with a Diploma in Data Science from IIT Madras. + +They may exit with both Diplomas from IIT Madras. + +The learner can join the Degree Level (Pg 11) if they have completed all 12 +Diploma Level courses AND have not entered the program via Diploma Only +Entry option. + +Diploma in Programming + +The Diploma in Programming lays a +sturdy foundation in Databases and +Programming concepts with Data +Structures and Algorithms. The learner +goes on to apply these in the building +of a web application by the end of the +diploma. + +Courses for Diploma in Programming + +6 courses +22 credits + +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Database Management Systems + +Modern Application Development I + +Programming, Data Structures and +Algorithms using Python + +Modern Application Development II -• +Programming Concepts using Java -Big data analytics is necessarily a joint effort by -researchers from academic institutions, -government and society and industry. -– +Skill Enhancement I -The government and industry are sources of Big -Data, and providers of problems and challenges, -The academic researchers are solution +Diploma in Data Science -providers. +The Diploma in Data Science exposes +the learner to the holistic approach of +gathering, analysing, and interpreting +data for a variety of problems. The +courses on Business Data lays down the +context and the need for the data, while +the Machine Learning courses equip +the learner to use and analyse this data +towards impactful conclusions. -When it comes to package the solutions from -university labs to transfer to the real world, -universities and industry must work together to -build scalable and robust solutions. +Courses for Diploma in Data Science -– +6 courses +22 credits -– +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Machine Learning Foundations + +Business Data Management + +Machine Learning Theory + +Business Analytics + +Machine Learning Practice + +Skill Enhancement II 11 - Expected Outcomes + Degree Level + +The Degree Level has courses that +complete the requirements for the BSc +degree. + +Requirements for registration +The learner should have come in via +Regular Entry (Pg 14), cleared all 8 +Foundational Level (Pg 11) courses and +all 12 Diploma Level (Pg 11) courses. + +11 courses +40 credits + +1 - 3 years +10 hrs/course/week + +₹1,00,000* +*Refer Fee Structure (Pg 9). + +Exit: Once the learner successfully completes 40 credits from Degree Level courses, +they can exit with a BSc Degree in Programming & Data Science from +IIT Madras. + +Credit Requirements +Learners have to complete a total of 40 credits at the degree level: + +4 credits from mandatory course titled “Strategies for Professional Growth” +4 credits from 2 Skill Enhancement Courses of 2 credits each. These are to be +completed on NPTEL and the credits transfered to this degree program. +8 credits from 2 Core Courses of 4 credits each. There are 3 pairs of core course +options to pick from. +24 credits from Elective Courses - 6 courses of 4 credits each or more courses of +lesser credits. + +Core Courses +Learner has to do a pair of core courses where each course is of 4 credits. Following +are the options for Core Courses: + +Core Course Option I + +Core Course Option II -• +Core Course Option III -• +Operating Systems -• +Software Testing -• +Computer Architecture -New methodologies and solutions for Big Data -research -New applications that impact the society and -industry in Hong Kong and beyond, and -New digital economies created based on big -data -New educational programs for students; -cultivating leaders for the Big Data society and -industry. +Software Engineering + +AI: Search Methods for +Problem Solving + +Deep Learning + +Elective Courses +A total of 24 credits need to be obtained from Elective Courses. Listed below are 9 +courses of 4 credits each, including the 3 pairs of core courses. Excluding the 2 courses +that the learner has already picked as core courses, he / she can opt for the elective +courses in the following two ways: + +a. Obtain all 24 credits by picking 6 of the remaining 7 courses listed below. OR + +b. Obtain upto 12 credits from NPTEL (accepted courses list will be provided soon) and +obtain remaining credits by picking from a subset of the 7 courses below. + +Operating Systems + +Software Testing + +Computer Architecture + +Software Engineering + +Topics in Application +Development + +Topics in Machine +Learning + +AI: Search Methods for +Problem Solving + +Deep Learning + +Topics in Data Science 12 - Breakthroughs Expected + ADMISSIONS + +Important Dates + +Applications Closed (Regular Entry) + +- 15 September 2020 + +Weekly release of Term 1 Qualifier courses +content starts -• +- 5 October 2020 -• +Qualifier Exam Hall Ticket (only for those who +get minimum required marks in Assignments) -• +- 2 November 2020 -• +Qualifier Exam (only for those with Hall Ticket)* -New algorithms, methodologies, systems and -applications in Big Data -New knowledge from Big data applications in -Science, Engineering, and Societal Problems -New insights into Big data practices in real -world -New ways to protect security and privacy of big +- 22 November, 2020* -data relevant to individuals and organizations +Qualifier Exam Results + +- 7 December, 2020 + +Registration for Term 1 (only for those who clear +Qualifier Exam) + +- 7 to 27 December, 2020 + +Foundational Level Batch 1 starts + +- 4 January, 2021 + +*Dates for in-person exams are subject to government restrictions prevalent at that time. + +Two Admission Paths +1. Regular Entry + +Anyone joining the program via Regular Entry will join at the Foundational Level. + +The Foundational Level courses allow for interested applicants with a variety of +educational backgrounds to learn the basics (Math, Statistics, Computational Thinking, +Python programming and English) which help them take up the Diploma Level and +then the Degree Level courses. + +There are three exit options for those who enter the program via Regular Entry: +Exit 1. Foundational Level certificate from Centre for Continuing Education, IIT Madras. +Exit 2. Diploma in Programming AND/OR Diploma in Data Science from IIT Madras +Exit 3. BSc Degree in Programming and Data Science from IIT Madras + +View Regular Entry (Pg 14) for details on Eligibility, Application Process, Qualifier +Process, Qualifier Exam and Passing Criteria. + +2. Diploma Only Entry + +The Diploma Only Entry is primarily aimed at working professionals or learners who +already possess the fundamentals taught in Foundational Level courses and wish to +exit with one or two Diplomas (Diploma in Programming AND/OR Diploma in Data +Science) from IIT Madras. + +Learners entering via Diploma Only Entry can only do Diploma Level courses and +cannot proceed to the Degree Level courses. + +We have plans to open applications for Diploma Only Entry in May / June 2021. All +details related to Diploma Only Entry will be provided soon on our website . 13 - Call for Proposals in Big Data Analytics + Mandatory Requirements + +Access to good internet connection as +well as a laptop/desktop device will be +a key requirement to learn effectively. + +Learner should be able to travel to +assigned exam centres for quizzes and +exams, each term. + +Regular Entry +Eligibility to Apply + +Eligibility for those who cleared class 12 or equivalent in 2018 or earlier: + +1. + +2. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2018 or earlier. + +Eligibility for those who cleared class 12 or equivalent in 2019 or 2020: + +1. + +2. + +3. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2019 or 2020. + +The candidate should be: +(i) currently enrolled / previously enrolled in any Bachelor’s Degree program in +campus mode OR +(ii) completed 1 year Diploma (campus mode only) OR +(iii) completed CA Inter + +Note that applicants who cleared class 12 or equivalent in 2019 or 2020 may apply for the +qualfier process with eligibility criteria 1 & 2 alone, but will be required to fulfil criterion 3 to +join the foundational courses after clearing the qualifying exam. + +Application Process -• +Anyone who is eligible may apply by filling in the application form, uploading +required documents and paying the application fee. Check out Application +Process  section on our website for the latest updates and to start applying. -• +Application fee - ₹3,000 (For applicants belonging to PwD (>= 40% disability) / SC / ST +category - ₹1,500; For applicants belonging to PwD (>= 40% disability) and also SC / +ST category - ₹750) -Foundations in Big Data Analytics Research: -– +Qualifier Process -developing and studying fundamental theories, -algorithms, techniques, methodologies, -technologies to address the effectiveness and -efficiency issues to enable the applicability of Big -Data problems; +All applicants will have to go through a Qualifier Process, like a trial month, +wherein they will get access to 4 weeks of content for the four foundational level +courses - English I, Mathematics for Data Science I, Statistics for Data Science I and +Computational Thinking. -Innovative Applications in Big Data Analytics: -developing techniques, methodologies and -– -technologies of key importance to a Big Data -problem that requires the seamless cooperation -of domain scientists with big data researchers. +Video content of 2 - 3 hours duration will be released for each course on a weekly +basis. Along with the content, the first three weeks will also include an assignment +for each course that need to be completed and submitted online within the given +deadline. + +Average assignment score in each course will be calculated. Only those who get +minimum required average assignment scores in all four courses will be allowed to +write the Qualifier Exam (refer next page). 14 - Benefit for the Hong Kong Society + Minimum Average Assignment +Score required in each course + +General Learner + +SC / ST / PwD with 40% disability + +PwD with 40% disability & SC / ST + +OBC-NCL / EWS + +40% + +30% + +30% + +35% + +Qualifier Exam and Passing Criteria + +Only learners who get minimum required average assignment scores in all four +courses will be eligible for the in-person invigilated Qualifier Exam conducted at the +end of 4 weeks of coursework. Only these candidates will receive hall tickets. + +The Qualifier Exam is of 3 hours duration and covers all 4 subjects with 45 minutes +allotted per course. + +To pass the Qualifier Exam, the learner has to get a minimum Total Qualifier Exam +Score and a minimum Qualifier Exam Score in each course individually. Refer table +below. + +Only those who pass the Qualifier Exam will be eligible to register for Foundational +Level courses. + +Minimum required +Qualifier Exam Score +in each course + +Minimum required +Total Qualifier +Exam Score + +General Learner + +SC / ST / PwD with 40% disability -• +PwD with 40% disability & SC / ST -• +OBC-NCL / EWS -• +40% -• +30% -• +30% -Can Hong Kong transform into a more data- -driven society and maintain -leadership -globally? +35% -its +50% -How can companies in Hong Kong become -more competitive with Big Data technology? +40% -Can Data Science in Hong Kong’s research -fields benefit from strong foundations on big data? +40% -Can the government become more efficient with -big data driven methodologies in decision -making? +45% -Can education, finance, logistics and health -benefit from the ever increasing data? +Note: Relaxations in pass criteria indicated for various categories of learners, is +applicable ONLY for the qualifier process. There will be no relaxations in terms of +grades / pass criteria once registered into the program. + +Though everyone who passes the Qualifier Exam will be allowed to register for +Foundational Level, there will be limitations on the number of courses a learner will +be allowed to register for in the first two terms based on the Total Qualifier Exam +Score (S). + +Total Qualifier Exam Score (S) + +Number of courses allowed to +register for in first two terms + +Minimum required <= S < 50% + +up to 2 courses + +50% <= S < 70% + +up to 3 courses + +S >= 70% + +up to 4 course + +Maximum number of courses a learner will be allowed to register for in later terms +will depend similarly on the average of course scores in the previous two terms. 15 + Contact Us + +Online Degree Office, 3rd Floor, ICSR Building, + +IIT Madras, Chennai - 600036 + +7850-999966 (Mon-Fri 9am-6pm) + +support@onlinedegree.iitm.ac.in + +www.onlinedegree.iitm.ac.in  + +iitmadrasonline + +16 + diff --git a/q192/random_k2/random_2.md b/q192/random_k2/random_2.md index 6bfd67b67b9eb8baf6562a416c94e48f3b69ec3d..4f0fbb2c84e93fd7c3f72a0ecb704f9505f5e8f3 100644 --- a/q192/random_k2/random_2.md +++ b/q192/random_k2/random_2.md @@ -1,253 +1,1571 @@ -EBay Annual Report 2024 +Brookfield Asset +Management -Form 10-K (NASDAQ:EBAY) +Investor Presentation -Published: May 29th, 2024 +May 2025 -PDF generated by stocklight.com + Brookfield Asset Management is a leading pure-play global asset manager - UNITED STATES -SECURITIES AND EXCHANGE COMMISSION -Washington, D.C. 20549 +1. Strong fundraising capabilities -Form 10-K/A -Amendment No. 1 +2. Positioned in fast-growing sectors with long-term tailwinds -☒ +3. Best-in-class transaction execution and operational capabilities -☐ +4. Track record of delivering attractive risk-adjusted returns -ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934 +5. Continuously innovating and scaling to deliver long-term growth -For the fiscal year ended December 31, 2023. +2 + + Company Overview + +3 + + Brookfield Asset Management today + +$1T+ + +Assets Under Management + +A leading global asset manager + +NYSE / TSX: BAM + +~$90B market cap + +$2.6B LTM fee-related earnings + +$549B fee-bearing capital + +Over 2,500 asset management employees +globally and ~250,000 operating employees + +125-year history as an owner-operator + +Best-in-class track record of generating strong returns -OR +All figures in this presentation are as of March 31, 2025, unless otherwise stated +See Notice to Recipients -TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934 +4 + + Brookfield Asset Management – Structure + +Brookfield Asset Management (BAM) +Manages over $1 trillion of assets within long-term private funds, perpetual strategies, +permanent capital vehicles, and liquid products, across five key verticals: + +Renewable Power +& Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit -For the Transition Period from to . +Benefits of the Brookfield Model + +Focused, asset-light model + +Long-term, consistent fee stream + +Combines sophisticated asset manager with access to Brookfield’s over $170B perpetual capital + +Transparent disclosures explain fund management, fee structure and transactions + +We are continuously working to improve our business model for investors and shareholders + +5 -Commission file number 001-37713 + Our business model is simple and effective +Our structure provides access to all major pools of capital and provides a range of investors +with a tailored approach to meet their objectives -eBay Inc. -(Exact name of registrant as specified in its charter) +Fundraising channels -Delaware -(State or other jurisdiction of -incorporation or organization) +• Institutional -2025 Hamilton Avenue -San Jose , California -(Address of principal executive offices) +• Insurance -77-0430924 -(I.R.S. Employer -Identification No.) +• Private Wealth -95125 -(Zip Code) +• Public Markets -Registrant’s telephone number, including area code: -(408) 376-7108 +Capital +raising -Securities registered pursuant to Section 12(b) of the Act: +Brookfield +Asset +Management -Title of each class -Common stock +Investing -Trading symbol -EBAY +Investing in attractive, +durable asset classes -Name of exchange on which registered -The Nasdaq Global Select Market +Transition -Securities registered pursuant to Section 12(g) of the Act: None +Infrastructure -Indicate by check mark if the registrant is a well-known seasoned issuer, as defined in Rule 405 of the Securities Act. Yes ☒ No ☐ +Private Equity -Indicate by check mark if the registrant is not required to file reports pursuant to Section 13 or Section 15(d) of the Act. Yes ☐ No ☒ +Real Estate -Indicate by check mark whether the registrant (1) has filed all reports required to be filed by Section 13 or 15(d) of the Securities Exchange Act of 1934 during the preceding 12 months (or for such shorter period that the -registrant was required to file such reports), and (2) has been subject to such filing requirements for the past 90 days. Yes ☒ No ☐ +Credit -Indicate by check mark whether the registrant has submitted electronically every Interactive Data File required to be submitted pursuant to Rule 405 of Regulation S-T (§ 232.405 of this chapter) during the preceding 12 -months (or for such shorter period that the registrant was required to submit such files). Yes ☒ No ☐ +6 -Indicate by check mark whether the registrant is a large accelerated filer, an accelerated filer, a non-accelerated filer, a smaller reporting company, or an emerging growth company. See the definitions of “large accelerated -filer,” “accelerated filer,” “smaller reporting company” and “emerging growth company” in Rule 12b-2 of the Exchange Act. + Strong culture built on Brookfield’s 125-year history as an owner-operator -Large accelerated filer +Our values -Non-accelerated filer +Our approach -☒ -☐ +Disciplined -Accelerated filer +Entrepreneurial -Smaller reporting company +Collaborative -Emerging growth company +Owner-Operators -☐ -☐ -☐ +Long Term View -If an emerging growth company, indicate by check mark if the registrant has elected not to use the extended transition period for complying with any new or revised financial accounting standards provided pursuant to -Section 13(a) of the Exchange Act. ☐ +Focus on Operational +Excellence - Indicate by check mark whether the registrant has filed a report on and attestation to its management’s assessment of the effectiveness of its internal control over financial reporting under Section 404(b) of the Sarbanes- -Oxley Act (15 U.S.C. 7262(b)) by the registered public accounting firm that prepared or issued its audit report. ☒ +7 -If securities are registered pursuant to Section 12(b) of the Act, indicate by check mark whether the financial statements of the registrant included in the filing reflect the correction of an error to previously issued financial -statements. ☐ + The Brookfield Ecosystem provides unparalleled insights -Indicate by check mark whether any of those error corrections are restatements that required a recovery analysis of incentive-based compensation received by any of the registrant’s executive officers during the relevant -recovery period pursuant to §240.10D-1(b). ☐ +30+ -Indicate by check mark whether the registrant is a shell company (as defined in Rule 12b-2 of the Act). Yes ☐ No ☒ +Countries -As of June 30, 2023, the aggregate market value of the registrant’s common stock held by non-affiliates of the registrant was $ -Market. +300+ -23,784,458,062 based on the closing sale price as reported on The Nasdaq Global Select +Portfolio companies -502 million shares of common stock issued and outstanding as of May 24, 2024. +Benefits of the +Brookfield Ecosystem -Auditor Name: +Helps us identify themes and trends, spot +pockets of value and source attractive +investment opportunities -PricewaterhouseCoopers LLP +Enhances intelligence and improves accuracy +in underwriting -Auditor Location: +Drives operational improvements and +boosts returns -San Jose, California +Hospitality -Auditor Firm ID: +Logistics & +Self-Storage -238 +Energy +Transition + +Housing + +Mixed Use, Retail +& Entertainment + +Business +Services + +Transport + +Infrastructure +Services + +$1+ trillion +AUM + +Utilities + +Solar + +Hydro + +Office + +Wind + +Technology Services + +Data + +Enables us to execute at scale with conviction, +regardless of market sentiment + +Midstream + +Industrials + +Healthcare + +Science & Innovation + +See Notice to Recipients + +8 + + Sustainable Growth Model + +9 + + Brookfield Asset Management’s sustainable growth model + +1 + +Pick good +sectors + + Well-positioned +in fast growing +alternative investment +management + + Focused on the +Backbone of the +Global Economy + ++ 2 - EXPLANATORY NOTE +Raise +capital + + Longstanding -This Amendment No. 1 on Form 10-K/A (this “Amendment”) amends Part IV, Item 15 of the Annual Report on Form 10-K for the fiscal year ended -December 31, 2023 of eBay Inc. (the “Company”), as filed with the U.S. Securities and Exchange Commission (the “SEC”) on February 28, 2024 (the -“Form 10-K”). The purpose of this Amendment is to provide financial statements for Adevinta ASA (“Adevinta”), pursuant to Rule 3-09 of Regulation S-X -for the years ended December 31, 2023 (audited), 2022 (unaudited) and 2021 (audited) and as of December 31, 2023 (audited) and 2022 (unaudited). In -accordance with Rule 3-09(b)(1), Adevinta’s financial statements are being filed herewith as Exhibit 99.01 within six months after the end of Adevinta’s -fiscal year. In addition, the Company is filing the consent of the independent auditors of Adevinta as Exhibit 23.02 and new certifications by the -Company’s Chief Executive Officer and Chief Financial Officer as Exhibits 31.03, 31.04, 32.03 and 32.04, respectively, as required by Rule 12b-15 under -the Securities Exchange Act of 1934, as amended. +partnerships with +institutional investors -Except as otherwise expressly noted, this Amendment does not modify or update in any way (i) the consolidated financial position, the results of -operations or cash flows of the Company, or (ii) the Form 10-K; nor does it reflect events occurring after the filing of the Form 10-K. Among other things, -forward-looking statements made in the Form 10-K have not been revised to reflect events that occurred or facts that became known to us after the filing -of the Form 10-K, and such forward-looking statements should be read in their historical context. Furthermore, this Amendment should be read in -conjunction with the Form 10-K and any subsequent filings with the SEC. + Globally diversified -Unless the context requires otherwise, all references to “we,” “our,” “us” or “eBay” mean eBay Inc., a Delaware corporation, and its consolidated -subsidiaries. +reach + + Access to Brookfield’s +permanent capital +and insurance +business + ++ 3 - ITEM 15: EXHIBITS AND FINANCIAL STATEMENT SCHEDULE +Invest +wisely -a. The following documents are filed as part of this report: + Assess opportunities -PART IV +to build or buy -1. Consolidated Financial Statements: + Leverage the -The consolidated financial statements of the Company, as listed in Item 15 of the Form 10-K, are included in Item 15 of the Form 10-K. +Brookfield Ecosystem -2. Financial Statement Schedule: + Scale through organic +and inorganic growth -The financial statement schedule of the Company, as listed in Item 15 of the Form 10-K, is included in Item 15 of the Form 10-K. -The financial statements of Adevinta required by Rule 3-09 of Regulation S-X are provided as Exhibit 99.01 to this Amendment. ++ -3. Exhibits Required by Item 601 of Regulation S-K: +4 -The exhibits listed in the Exhibit Index of the Form 10-K and this Amendment are filed with, or incorporated by reference in, this report. +Operate +well -b. The information required by this Item is set forth in the exhibits listed in the Exhibit Index of the Form 10-K and this Amendment filed with, or + Invest with an -incorporated by reference in, this report. +Owner-operator +mindset -c. Financial Statement Schedule and Separate Financial Statements of Subsidiaries Not Consolidated and Fifty Percent or Less Owned Persons. + Enabled by an -Adevinta was deemed a significant equity investee under Rule 3-09 of Regulation S-X for the fiscal years ended December 31, 2023 and -December 31, 2021 (though not for the fiscal year ended December 31, 2022). As such, financial statements of Adevinta for the fiscal years ended -December 31, 2023, 2022 and 2021, respectively, are required to be filed by amendment to this Annual Report on Form 10-K within six months of -Adevinta’s fiscal year end. Accordingly, Adevinta’s financial statements for its fiscal year ended December 31, 2023 are provided as Exhibit 99.01 -to this Amendment. +Owner-operator +structure -4 += - No. +Generating +excellent +risk-adjusted +returns -23.02 +Results propel a +virtuous cycle +supporting the entire +growth model -31.03 +Significant tailwinds driven by AUM growth, alternatives growth, and favorable sectors -31.04 +10 -32.03 + The past 20 years have seen major tailwinds for alternatives… +and there is much more to come -32.04 +2002 -99.01 +Today -104 +$25T -INDEX TO EXHIBITS +Alternatives AUM -Exhibit Description +~15% -Filed or Furnished with -this 10-K/A +$150T+ +Global AUM -Incorporated by Reference +$2T -Form +Alternatives AUM -File No. +~5% -Date Filed +$40T +Global AUM -Consent of Ernst & Young AS related to Exhibit 99.01. +2032E -Certification of Registrant’s Chief Executive Officer, as required by -Section 302 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +$60T+ -Certification of Registrant’s Chief Financial Officer, as required by -Section 302 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +Alternatives AUM -Certification of Registrant’s Chief Executive Officer, as required by -Section 906 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +20%+ -Certification of Registrant’s Chief Financial Officer, as required by -Section 906 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +$300T+ -Consolidated financial statements of Adevinta ASA for the years ended -December 31, 2023 (audited), 2022 (unaudited) and 2021 (audited) and -as of December 31, 2023 (audited) and 2022 (unaudited). +Global AUM -Cover Page Interactive Data File (formatted as Inline XBRL and -contained in Exhibit 101). +Key growth drivers -X +• Global AUM growth -X +• Institutional investors -X +increasing allocations to +alternatives -X +• Investors consolidating +their exposure to the +largest and most +diversified managers -X +Why investors are +choosing to invest their +growing capital base in +alternative assets -X +Excess returns -X +Diversification -5 +Less volatility + +Predictable cash flow + +Sources: Preqin; Bain analysis; PwC; Company estimates + +11 + + We are focused on the Backbone of the Global Economy + +Digitalization + +Decarbonization + +Deglobalization + +Our assets live at the epicenter of these decades-long megatrends +Trillions of dollars will need to be invested as the world adapts to paradigm shifts + +12 + + We are growing our sources of capital while expanding into new ones + +New Sources of Capital: Tremendous Growth + +• + +Insurance Solutions + +• Private Wealth + +• Manager Acquisitions + +• Strategic Partnerships + +Existing Sources of Capital: More to Do + +• + +Institutional Investors + +• Public Markets + +13 + + Unique and diverse pools of capital support our fundraising growth + +Institutional Investors + +300+ Client & Marketing Professionals +2,200+ Institutional Relationships + +Insurance Solutions + +$15B+ Annual inflows from Brookfield +Wealth Solutions + +150 Existing Third-Party Insurance Relationships + +Private Wealth + +150 dedicated professionals +160+ Wealth Management Platform Relationships + +Public Markets + +See Notice to Recipients + +3 permanent capital vehicles +~$60B Market Value +~2,500 Institutional Shareholders + +BAM’s fundraising +differentiators + +Breadth and specialization – +at the same time + +Operational expert, not +financial engineers + +Global capabilities with local +mindset + +Partnership mentality with +360° engagement – investors +are more than “just an LP” + +Access to co-investment, +partnership opportunities +and customized solutions + +14 + + Our leadership positions set us up to outgrow the market +We have positioned ourselves as a leader within our sectors of focus +across the Brookfield Ecosystem + +Renewable Power +& Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit + +Largest global +transition platform + +Leading infrastructure +strategy + +Best-in-class PE +flagship platform + +One of the largest + real estate investors + +Unmatched credit +capabilities across +funds and partnerships + +$125B +AUM + +$214B +AUM + +$147B +AUM + +$272B +AUM + +$321B +AUM + +See Notice to Recipients + +15 + + Competitive advantages across all our business groups + +Scale and Global +Footprint + +Sourcing +Capabilities + +Owner-Operator +History + +Access to +Debt Capital + +Leveraging the +Brookfield Ecosystem + +16 + + We are consistently expanding our capabilities + +Sometimes we build + +Sometimes we buy + +Example: Renewables expertise... + +Renewable Power & Transition FBC +($B) + +Example: Expanding our credit platform by +adding Oaktree... + +25% +CAGR + +$21 + +2018 + +$64 + +2024 + +…growing into energy transition +• Flagship Transition Fund +• Catalytic Transition Fund +• Renewables Sidecar +• Credit Opportunities + +…enabled further growth + +Insurance Platform + +• +• Private Wealth +• Other Partnerships + +Note: 2024 FBC includes targeted fund size for BGTF II, which is currently fundraising. + +17 + + We partner with leading managers to expand our capabilities + +2021 + +2023 + +2025 + +2019 + +2022 + +2024 + +1. Angel Oak acquisition announced in April 2025. Closing is subject to customary conditions + +1 + +18 + + Business Group Overview + +19 + + Renewable Power & Transition – Overview +Brookfield is a leading owner, operator and developer of renewable energy +and decarbonization technology +$125B + +$58B + +substantial deal pipeline + +93 + + Large footprint, extensive experience and + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals + + Unsurpassed capabilities, relationships and + +access to scale capital + +Key Sectors + +Products + +Hydro +8,300 MW + +Wind +16,900 MW + +Solar +11,700 MW + +Flagship Fund + +Complementary Strategies + +Global Transition (BGTF Series) + +Catalytic Transition Fund (CTF) + +Asia Renewable Fund (BCRF) + +DG, Storage & +Sustainable Solutions +4,900 MW + +Permanent Capital Vehicle + +Brookfield Renewable Partners (BEP / BEPC) + +Key drivers: + +Accelerating demand for power from digitalization, electrification and AI, and renewables position +as the lowest cost source of bulk power globally + +Global goals of net-zero emissions, energy security and independence, alongside corporate +demand and support from favorable legislation + +20 + + Infrastructure – Overview +Brookfield is one of the world’s largest infrastructure investors, owners and operators + +$214B + +Assets Under Management + +$96B + +Fee-Bearing Capital + +127 + +Investment Professionals + + Ideally positioned at the epicenter of the global secular + +trends of deglobalization, decarbonization and digitalization + + Operational expertise and institutional knowledge + +Key Sectors + +Products + +Utilities +$66B + +Transport +$48B + +Data +$65B + +Midstream +$25B + +Flagship Fund + +Complementary Strategies + +Infrastructure Core Plus (BIF Series) + +Brookfield Supercore Infrastructure Partners (BSIP) + +Brookfield Infrastructure Structured Solutions (BISS) + +Permanent Capital Vehicle + +Brookfield Infrastructure Partners (BIP / BIPC) + +Key drivers: + +$94 trillion of infrastructure investment required by 2040 + +AI infrastructure emerging as a new digitalization-themed asset class representing +a potential market opportunity of more than $8 trillion over the next five years + +21 + + Private Equity – Overview +Brookfield is one of the most experienced private equity investors globally + +$147B + +Assets Under Management + +$43B + +Fee-Bearing Capital + +171 + +Investment Professionals + + Global sourcing engine + Experienced team with local operating expertise + Value creation through operational improvement + +Key Sectors + +Products + +Industrials +$42B + +Infrastructure Services +$22B + +Flagship Fund + +Complementary Strategies + +Opportunistic Private Equity (BCP Series) + +Brookfield Special Investments (BSI) + +Business Services +$52B + +Directly Held & Other +$31B + +Permanent Capital Vehicle + +Brookfield Middle East Partners (BMEP) + +Brookfield Business Partners (BBU / BBUC) + +Brookfield Sponsor Solutions (BSS) + +Brookfield Financial Infrastructure Partners (BFIP) + +Key drivers: + +Convergence of technology and industrial landscape reawakening displaced industries + +Revitalization of industrial sector requires significant capital and operational expertise + +22 + + Real Estate – Overview +Brookfield is one of the largest real estate investors with a diversified portfolio +in the world’s most well-established markets +295 +$272B + +$100B + + Scale, relationships, operating expertise + + Assets well positioned to weather market cycles and + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals + +deliver long term value + +Key Sectors + +Logistics, +Storage & NNN +160M+ sf logistics + +Mixed Use, Retail +& Entertainment +~130M combined sq. ft. + +Housing +110K+ units + +Office +~190M sq. ft. + +Products + +Hospitality +40K+ keys + +Science & +Innovation +5M+ sq. ft. + +Flagship Fund + +Complementary Strategies + +Opportunistic Real Estate (BSREP Series) + +Brookfield Premier Real Estate Partners (BPREP) + +Permanent Capital Vehicle + +Brookfield Single Family Rental (BSFR) + +Brookfield Property Group (BPG) + +Brookfield Non-Traded REIT + +Key drivers: + +Demographic shifts and market cycle dynamics driving increasing real estate needs + +Digitalization & Decarbonization rapidly transforming the logistics & hospitality industries + +23 + + Credit – Overview +Brookfield’s credit strategies include our longstanding private credit and direct lending funds +and our partnerships with leading credit managers + +$321B + +$252B + +647 + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals1 + + Marquee partnerships with leading credit managers + Cycle-tested platform + Extensive lending capabilities + Longstanding relationships with financial institutions + +Brookfield Credit Capabilities + +Partner Managers + +Brookfield Wealth Solutions (BWS) and 3rd Party Capabilities +- Private Credit (SMAs) + +Real Estate + +Asset Based + +- Private Funds + +Infrastructure & Renewable + +- Liquid Credit + +Corporate + +Liquid + +2 + +Key drivers: + +Equity integration – access to the Brookfield Ecosystem allows us to meet the evolving needs of +clients globally and provide investment opportunities across the full capital stack + +Origination capabilities – robust network of proprietary origination platforms and strategic +partnerships with specialized credit managers + +1. +2. + +Investment professionals include Brookfield Credit Group, Brookfield Wealth Solutions, Public Securities Group (PSG), and partner manager professionals. +Angel Oak acquisition announced in April 2025. Closing is subject to customary conditions + +24 + + Financial Performance and +Growth Targets + +25 + + Our results demonstrate the stability, predictability and +growth of our earnings + +16% +CAGR + +$2,108 + +$418 + +$2,241 + +$457 + +$2,456 + +$539 + +$1,376 + +$312 + +$1,827 + +$364 + +$2,000 + +$1,500 + +$1,000 + +$500 + +$0 + +2020 + +2021 + +2022 + +2023 + +2024 + +Full year periods as of December 31 + +FBC ($B) + +FRE ($M) + +…and should continue to drive earnings growth, giving us confidence in delivering +15%+ long term annual dividend growth + +See Notice to Recipients + +$500 + +$400 + +$300 + +$200 + +$100 + +$0 + +26 + + We have delivered strong results over the long term + +Fund + +Gross IRR + +Net IRR + +Infrastructure / Renewable Power + +Private Equity + +Real Estate + +Opportunistic Credit + +Transition + +15% + +26% + +20% + +22% + +n.m. + +12% + +20% + +16% + +16% + +n.m. + +~$225 billion of profits generated for investors over the past 30 years + +See Notice to Recipients + +27 + + How we will generate future growth + +Strategies + +Sectors + +Enhancing +Fundraising Channels + +Diversifying & Scaling +Product Offerings + +Building & Buying +New Businesses + +Energy Transition + +AI Infrastructure + +Private Credit + +28 + + Growth is driven across all our business groups, as our long-term or +permanent capital base continues to grow + +Q1-25 + +$58 + +$252 + +87% + +Long Term or +Permanent +Capital + +$96 + +$43 + +$100 + +$549B +Fee-Bearing Capital + +Growing +To + +2029 + +$110 + +$590 + +92% + +Long Term or +Permanent +Capital + +$160 + +$90 + +$140 + +$1.1T +Fee-Bearing Capital + +Renewable Power and Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit + +See Notice to Recipients + +29 + + Growth in fee-bearing capital and carried interest will propel earnings + +2024 + +17% +CAGR + +2029 + +2024 + +18% +CAGR + +2029 + +FRE per share + +DE per share + +With a projected ~95% payout ratio, the plan supports 15%+ annual dividend growth + +30 + + With further upside in carried interest +We are steadily growing our carried interest base with realizations beginning +to impact earnings in 2029, creating the second leg of growth + +Gross Carried Interest to BAM + +~$2B + +Estimated Realizations +in 2029 + +Growing +To + +~$7B + +Estimated Realizations +in 2034 + +See Notice to Recipients + +31 + + Robust Liquidity Position to Drive Growth + +Our strong liquidity and new debt capacity allow us to capitalize on opportunities and defend +against downside risks + +$1.4B + +Corporate Liquidity + +$750M + +Inaugural Debt Offering + +A / A- + +Investment Grade +Corporate Ratings(1) +(Fitch / S&P) + +We are strategically positioned to support our enhance liquidity through a variety of +fund-raising channels + +(1) + +A security rating is not a recommendation to buy, sell or hold securities and may be subject to revision or withdrawal at any time + +32 + + The best is yet to come for BAM + +1. Leading pure-play global asset manager + +Long-term growth targets + +2. Strong fundraising capabilities + +3. Positioned in fast-growing sectors with + +long-term tailwinds + +4. Best-in-class transaction execution and + +operational capabilities + +5. Track record of delivering attractive returns + +6. Continuously innovating and scaling to + +deliver growth + +See Notice to Recipients + +5-year CAGR + +By 2029 + +FBC + +16% $1.1T + +FRE + +17% + +$5.0B + +DE + +18% + +$5.1B + +33 + + Appendix + +34 + + Senior management team + +Bruce Flatt +CEO, Brookfield Asset +Management + +Connor Teskey +President, Brookfield Asset +Management; CEO Renewable +Power & Transition + +Justin Beber +Chief Operating Officer + +Joined Brookfield: 1990 + +2012 + +2007 + +Hadley Peer Marshall +Chief Financial Officer, +Brookfield Asset Management +and Managing Partner, +Infrastructure +2015 + +Brian Kingston +CEO, Real Estate + +Anuj Ranjan +CEO, Private Equity + +Craig Noble +CEO, Credit + +Sam Pollock +CEO, Infrastructure + +2001 + +2006 + +2004 + +1994 + +Brookfield’s senior management team possesses deep experience in our target regions, sectors and asset classes, +as well as decades of investment expertise built on managing assets and businesses across market cycles + +35 + + Notice to Recipients + +Brookfield Asset Management Ltd. (“BAM,” “Brookfield Asset +Management” or “Brookfield”) is not making any offer or +invitation of any kind by communication of this presentation and +under no circumstance is it to be construed as a prospectus or +an advertisement. + +Except where otherwise indicated herein, the information +provided herein is based on matters as they exist as of the +date of this presentation and not as of any future date, is +subject to change, and, unless required by law, will not be +that +updated or otherwise revised +subsequently becomes available or circumstances existing or +changes occurring after the date hereof. + +information + +to reflect + +Unless otherwise noted, all references to “$” or “Dollars” are to +U.S. Dollars. This presentation +financial +information as of and for the last twelve months ended +2024. +December + +includes certain + +31, + +Cautionary Note Concerning Forward-Looking Statements + +information within + +include statements + +This presentation contains forward-looking statements within +the meaning of Section 27A of the Securities Act of 1933, as +amended (the “Securities Act”), Section 21E of the Securities +Exchange Act of 1934, as amended (the “Exchange Act”) and +forward-looking +the meaning of other +relevant securities legislation, including applicable securities +laws in Canada, which reflect our current views with respect to, +among other things, our operations and financial performance. +Forward-looking statements +that are +predictive in nature, depend upon or refer to future results, +events or conditions, and include, but are not limited to, +statements which reflect management’s current estimates, +beliefs and assumptions regarding the operations, business, +financial condition, expected financial results, performance, +prospects, opportunities, priorities, targets, goals, ongoing +objectives, strategies, capital management and outlook of +Brookfield Asset Management Ltd., as well as the outlook for +North American and international economies for the current +fiscal year and subsequent periods, and which in turn are +based on our experience and perception of historical trends, +current conditions and expected future developments, as well +as other factors management believes are appropriate in the +circumstances. The estimates, beliefs and assumptions of +Brookfield Asset Management Ltd. are inherently subject to +significant business, economic, competitive and other + +uncertainties and contingencies regarding future events and as +such, are subject to change. Forward-looking statements are +typically identified by words such as “expect”, “anticipate”, +“believe”, “foresee”, “could”, “estimate”, “goal”, “intend”, “plan”, +“seek”, “strive”, “will”, “aim”, “may” and “should” and similar +expressions. +In particular, the forward-looking statements +contained in this presentation include statements referring to +Brookfield Asset Management Ltd.’s growth prospects and +liquidity, and the impact of acquisitions and dispositions on our +business. + +Although we believe that such forward-looking statements are +based upon reasonable estimates, beliefs and assumptions, +actual results may differ materially from the forward-looking +statements. Factors that could cause actual results to differ +materially from those contemplated or implied by forward- +looking statements include, but are not limited to: (i) returns +that are lower than target; (ii) the impact or unanticipated +impact of general economic, political and market factors in the +countries in which we do business; (iii) the behavior of financial +markets, including fluctuations in interest and foreign exchange +rates and heightened inflationary pressures; (iv) global equity +and capital markets and the availability of equity and debt +financing and refinancing within these markets; (v) strategic +actions including acquisitions and dispositions; the ability to +complete and effectively integrate acquisitions into existing +operations and the ability to attain expected benefits; (vi) +changes in accounting policies and methods used to report +financial condition (including uncertainties associated with +critical accounting assumptions and estimates); (vii) the ability +to appropriately manage human capital; (viii) the effect of +applying future accounting changes; (ix) business competition; +(x) operational and reputational risks; (xi) technological change; +(xii) changes in government regulation and legislation within the +(xiii) governmental +countries +investigations and sanctions; (xiv) litigation; (xv) changes in tax +laws; (xvi) ability to collect amounts owed; (xvii) catastrophic +and +events, +as +of +epidemics/pandemics; +international conflicts and other developments, +including +introduction, +terrorist acts and cyberterrorism; +withdrawal, success and timing of business initiatives and +strategies; (xx) the failure of effective disclosure controls and +procedures and internal controls over financial reporting and +other risks; (xxi) health, safety and environmental risks; (xxii) +the maintenance of adequate insurance coverage; (xxiii) the +existence of information barriers between certain businesses + +in which we operate; + +earthquakes, + +hurricanes + +possible + +impact + +(xviii) + +such + +(xix) + +the + +the + +within our asset management operations; (xxiv) risks related to +our renewable power and transition, infrastructure, private +equity, real estate and credit strategies; and (xxv) factors +detailed from time to time in our documents filed with the +securities regulators in the United States and Canada. + +We caution that the foregoing list of important factors that may +affect future results is not exhaustive, and other factors could +also adversely affect future results. Readers are urged to +consider these risks, as well as other uncertainties, factors and +forward-looking +assumptions carefully +statements and are cautioned not to place undue reliance on +such forward-looking statements, which are based only on +information available to us as of the date of this presentation. +Except as required by law, we undertake no obligation to +publicly update or revise any forward-looking statements, +whether written or oral, that may be as a result of new +information, future events or otherwise. + +in evaluating + +the + +Past performance is not indicative nor a guarantee of future +results. There can be no assurance that comparable results will +be achieved in the future, that future investments will be similar +to historic investments discussed herein, that targeted returns, +growth objectives, diversification or asset allocations will be +met or that an investment strategy or investment objectives will +be achieved (because of economic conditions, the availability +of appropriate opportunities or otherwise). + +36 + + Notice to Recipients + +in + +forth + +(including changes + +Target returns and growth objectives set +this +presentation are for illustrative and informational purposes only +and have been presented based on various assumptions made +by Brookfield Asset Management Ltd. in relation to the +investment strategies being pursued, any of which may prove +to be incorrect. There can be no assurance that targeted +returns or growth objectives will be achieved. Due to various +risks, uncertainties and changes +in +economic, operational, political or other circumstances) +beyond our control, the actual performance of the business +could differ materially from the target returns and growth +objectives set forth herein. In addition, industry experts may +disagree with the assumptions used in presenting the target +returns and growth objectives. No assurance, representation or +warranty is made by any person that the target returns or +growth objectives will be achieved, and undue reliance should +not be put on them. Prior performance is not indicative of +future results and there can be no guarantee that we will +achieve the target returns or growth objectives or be able to +avoid losses. + +The information in this presentation does not take into account +an investor’s investment objectives, financial situation or +particular needs and nothing contained herein should be +construed as legal, business or tax advice. Each prospective +investor should consult its own attorney, business adviser and +tax advisor as to legal, business, tax and related matters +concerning the information contained herein. + +Non-GAAP Measures + +for + +including + +business, + +We prepare our financial statements in conformity with the +accounting principles generally accepted in the United States +of America (“U.S. GAAP”). This presentation discloses a +number of non-GAAP financial and supplemental financial +in monitoring our asset +measures which are utilized +management +performance +measurement, capital allocation and valuation purposes. We +believe that providing these performance measures is helpful +to investors in assessing the overall performance of our asset +management business. These non-GAAP financial measures +should not be considered as the sole measure of our business’ +performance and should not be considered in isolation from, or +as a substitute for, similar financial measures calculated in +conformity with U.S. GAAP financial measures. These non- +GAAP +financial measures are not standardized financial +measures and may not be comparable to similar financial +issuers. Non-GAAP measures +measures used by other + +include, but are not limited to: (i) distributable earnings +(“Distributable Earnings”), (ii) fee revenues (“Fee Revenues”) +and (iii) fee-related earnings (“Fee Related Earnings” or +“FRE”). These non-GAAP measures are not standardized +financial measures and may not be comparable to similar +financial measures used by other issuers. + +Supplemental financial measures + +financial measures +(“AUM”), + +include assets under +Supplemental +management +(“Fee-Bearing +Capital”) and uncalled fund commitments (“Uncalled Fund +Commitments”). + +fee-bearing capital + +Securities Law + +This presentation does not constitute an offer to sell or a +solicitation of an offer to buy any securities of Brookfield Asset +Management Ltd. + +37 diff --git a/q192/random_k4/question.json b/q192/random_k4/question.json index 0c64b26913cc45cfffc572794f288ce51c75229b..111aa3af162990698559012f9df8c876dbe92533 100644 --- a/q192/random_k4/question.json +++ b/q192/random_k4/question.json @@ -17,10 +17,10 @@ ], "original_filenames": [ "Guide-for-international-students-web.pdf", - "web_7795b10bf9b6a434.pdf", - "web_ca39b8721f6d1255.pdf", - "web_2c0d954cf4efd312.pdf", - "web_18e98cd838edeb2e.pdf" + "web_f3420c2dcb7612a6.pdf", + "web_e5f79db0bc92a9fa.pdf", + "web_f539c13fe18de9dd.pdf", + "resize_r1_39.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q192/random_k4/random_1.md b/q192/random_k4/random_1.md index 4c601abeb6f180fb2f07694cada010de6f7dd010..c628675fbe529d89838de1ae7c364d79b8d4ce01 100644 --- a/q192/random_k4/random_1.md +++ b/q192/random_k4/random_1.md @@ -1,462 +1,1223 @@ -Proposal for the Theme on Big Data -Analytics +Online BSc Degree +and Diploma Program +PROGRAMMING AND DATA SCIENCE -Qiang Yang, HKUST +Batch 1: Starts January 2021 -Jiannong Cao, PolyU +For latest updates, visit - www.onlinedegree.iitm.ac.in -Qi-man Shao, CUHK +Brochure Last Updated: 2nd November 2020 -May 2015 + Contents - Motivation +Introduction -• +Why learn Programming & Data Science -The world's technological per-capita capacity to +About IIT Madras -store information doubled every 40 months +Our Experience in Online Education -– +Academics -As of 2012, 2.5 exabytes (2.5×1018) of data/day -Relational database management systems and -• -desktop statistics and visualization packages often -have difficulty handling big data. +Overall Structure of the programme -– +Term Structure -Big Data: new driver for digital economy&society +Course Structure - +Assessments - + Online Assignments - + Three Quizzes -Gartner: hundreds of billions of GDP by 2020. -Intangible factor after labor and capital -Data Science: The fourth paradigm + End Term Exam -2 + Pass Criteria for Each Course - The Power of Big Data + Learners Who Do Not Pass a Course -• +Exam Cities -• +Fee Structure -• +Foundational Level -• +Diploma Level -Big Data can bring “big values” to our life in -almost every aspects. -Technologically, Big Data is bringing about changes in our lives because it -allows diverse and heterogeneous data to be fully integrated and -analyzed to help us make decisions. -Today, with the Big Data technology, thousands of data from seemingly -unrelated areas can help support important decisions. This is the -power of Big Data. -Areas of Applications + Diploma in Programming -– + Diploma in Data Science -– +Degree Level -– +Admissions -– +Important Dates -– +Two Admission Paths -Health and Well being + Regular Entry -Policy making and public opinions + Diploma Only Entry -Smart cities and more efficient society +Mandatory Requirements -New online educational models: MOOC and Student-Teacher modeling +Regular Entry -Robotics and human-robot interaction + Eligibility to Apply -• + Application Process -Much of this power hinges on Research on Analytics + Qualifier Process + + Qualifier Exam and Pass Criteria + +Contact Us 3 - Hong Kong needs Big Data -Research +3 + +4 + +4 + +5 + +5 + +6 + +6 + +7 + +7 + +7 + +7 + +8 + +8 + +8 + +9 + +10 + +10 + +11 + +11 + +12 + +13 + +13 + +13 + +13 + +13 + +13 + +14 + +14 + +14 + +14 + +15 + +16 + + Introduction + +IIT Madras, India’s top technical institute, welcomes you to the world’s first online BSc +Degree program in Programming and Data Science. For the first time, you can work +towards an undergraduate degree / diploma from an IIT regardless of your age or +location, and with a wide range of academic backgrounds. + +“ + +IIT Madras is pioneering a completely online program to +produce graduates who will be employable in the sector +that presently has high demand - Programming and Data +Science. The program uses an innovative combination of +online learning and in-person assessment and is flexible and +affordable. I believe the program will be a gamechanger +in the field of education in India in the coming years and +inspire other top-ranking institutions to launch many such +online initiatives. + +- Prof. Bhaskar Ramamurthi, Director, IIT Madras + +Why Programming & Data Science? + +Decision making is increasingly becoming +data driven within the commercial sector +and the government. There is a large +quantity of data readily available for this +at present. Smart devices connected to +the internet bring more input sources, +apart from those generated by humans +as they interact with services and one +another. Data driven decision making +involves the analysis of large volumes +of data to identify patterns and build +predictive models. This requires a +combination of skills ranging from +computing, statistics and mathematics, +and is broadly labelled data science. + +INCREASING DEMAND + +BETTER JOB OPPORTUNITIES + +UPSKILL + +As formal academic programs in data +science are only just emerging, there +is a huge gap between the demand for +data scientists and the supply of suitable +qualified applicants in the job market. +The throughput of traditional classroom +programs is limited and will not be able +to provide the numbers needed to meet +the upcoming requirements. Online +teaching has the potential to train much +larger numbers, but delivering scale +without diluting quality has proven to be +a challenge until recently. + +Our program therefore aims to create +a thoroughly trained programmer, +proficient in application development, +data sciences and machine learning. + +3 + + About IIT Madras + +Indian Institute of Technology Madras is one of the +foremost institutes of national importance for higher +technical education and research. + +In 1956, the German Government offered +technical assistance in establishing +an institute of higher education in +engineering in India. The first Indo- +German agreement in Bonn, West +Germany for the establishment of the +Indian Institute of Technology at Madras +was signed in 1959. The Institute with its +campus in Chennai, previously known +as Madras, was formally inaugurated +in 1959 by Prof. Humayun Kabir, Union +Minister for Scientific Research and +Cultural Affairs. + +The Institute today has sixteen academic +departments and several advanced +research centres in various disciplines +of engineering and pure sciences. A +faculty of international repute, a brilliant +student community, excellent technical +& supporting staff and an effective +administration have all contributed to the +pre-eminent status of IIT Madras. + +Our Experience in Online Education + +IITM is well-equipped to provide an +online degree course. We have been +ranked number 1 Engineering Institute +for the past 5 years and we have been +ranked number 1 Overall for the past 2 +years by NIRF. + +Furthermore, IITM began the NPTEL +program, the country’s first online +content portal as an inter-IIT consortium +project in 2001 and nptel.ac.in  is one +of the largest online course portals in +the world. In 2014, IITM also created +NPTEL online courses - the country’s first +Massive Open Online Course (MOOC). +Today, NPTEL is considered the largest +MOOC provider in India. + +MHRD launched the SWAYAM MOOC +platform in 2017 and NPTEL has been +the largest contributor to it. NPTEL +also manages swayam.gov.in  - the +national MOOCs portal, as well as the +network of more than 3800 SWAYAM +local chapter colleges. With IITM +continuing to lead the NPTEL project, it +makes them the perfect choice to offer +an online degree program. + +2001 +NPTEL +Program + +2014 +NPTEL +Online +Courses + +2017 +SWAYAM +MOOC + +2020 +IITM +Online +Degree + +4 + + Overall Structure + +There are THREE LEVELS in our Online +Degree program and to get a BSc +Degree in Programming and Data +Science from IIT Madras, a learner +has to successfully complete all three +levels. 1. +FOUNDATIONAL LEVEL 2. +DIPLOMA LEVEL -to develop state-of-the-art Big -Data platform in research, -education and industrial -applications, and open it to the -Hong Kong society and the world -at large, and +3. +DEGREE LEVEL -to make a difference in Smart -Cities, Health and Well-being -(including supporting aging -populations), and modernizing -Finance, Education and Logistics -in Hong Kong. +There is also the flexibility to exit at +any level. Depending on the courses +completed, the learner can receive +Foundational Certificate from Centre +for Continuing Education, IIT Madras +or Diploma(s) from IIT Madras. (Pg 13 +for Admissions related information) -4 +We are currently formulating a +Diploma Only Entry (Pg 13) option for +those who are interested in doing only +Diploma Level courses and exiting +with one or both Diploma(s) from IIT +Madras. + +Total Courses: 31 +116 credits + +Completion time: 3 - 6 years +based on preferred pace & performance - Big Data Analytics Objectives +Fees: Based on courses registered +*Refer Fee Structure (Pg 9). + +Online Courses & Assignments +12 weeks of coursework for each course + +Quizzes and Exams +3 Quizzes and one End Term Exam for +each course - in person and invigilated. + +ACADEMICS 5 - Relation to Smart Cities and IoT + Term Structure + +Every year is divided into three +terms - January Term, May Term and +September Term. + +Before progressing to the next level, +the learner has to clear all the courses +in the current level. + +In a term, a learner may register +for a course only if they have +already successfully completed the +prerequisites for that course. + +The learner may be allowed to +register for upto 4 courses in a term +depending on their performance in +the previous exam(s). + +A learner has to clear all Foundational +level courses successfully, before +registering for any Diploma Level +courses. Similarly, learners have +to clear Diploma Level courses +successfully before registering for any +Degree Level courses. + +Course Structure + +The duration of the core courses +in Foundational, Diploma and +Degree Level is 12 weeks and the +courses carry 4 credits each. +The Skill Enhancement courses +and Electives in Degree Level +have varying credits and may +be of different duration. + +The fee for each term depends on the +total credits one registers for in that +term. Refer Fee Structure (Pg 9). + +The recorded video content released +every week will comprise a series of +lectures, sample problems and step +by step instructions on how to solve +them. + +The duration of each video would be +15 to 30 minutes amounting to a total +of about 2 - 3 hours of video content +each week in each course. + +Weekly assignment(s), which will be +released along with the videos, will +have to be completed and submitted +online in the specified period of time. + +There will be 3 quizzes for each +course, one each at the end of weeks +4, 7 and 10. Each course further +culminates in an End Term Exam. All +the quizzes and the end term exams +have to be completed by the learner, +in person, at designated exam centres, +held under invigilation. + +All in-person, invigilated quizzes and +exams will be scheduled for weekends, +subject to availability of dates with +exam partner. -• +6 -• + Assessments -• +There are 3 types of assessments for each course - online weekly assignments, +monthly in-person invigilated quizzes and an in-person invigilated end term exam. -World economic forum -ranking HK’s -infrastructure: #1 -– +If the Average Assignment Score for a +course is less than 40/100, the learner +will not be eligible to write the end +term exam for that course. -Maintain the lead in IT +Online Assignments -Infrastructure +A 12-week course will have at least one +assignment every week. -East Kowloon Project: -Energizing Hong Kong -via Smart Cities -Big Data: -– +Each assignment has to be submitted +online within the specified due date. -IoT provides the +From the 12 weekly assignment scores +for a course, a learner’s best 8 scores +will be considered to calculate the +Average Assignment Score. -infrastructure for +Three Quizzes +IN PERSON, INVIGILATED -collecting the data +There will be 3 quizzes for each course, one each at the end of Weeks 4, 7 and 10. +Every quiz will be based on content covered up to that point in the course. These +quizzes have to be attempted in person, in the specified exam city (Pg 9). -Smart Cities as +Each quiz will be conducted in a single +session and will consist of sections of +45 minutes duration for every course +chosen. -important application +It is mandatory to appear for at least +one out of the three quizzes to be +eligible to write the end term exam. -goal +The best two out of the three quiz +scores will be considered to calculate +the Average Quiz Score (Q) out of 100. -– +40% of Q will be used to calculate Total +Course Score T and course grade. -6 +End Term Exam - Research Objectives +IN PERSON, INVIGILATED -• +At the end of a term, there will be an +end term exam of 3 hours duration for +each course. -• +This exam score (out of 100) is called +End Term Score (E) which accounts for +60% of the Total Course Score (T). -• +To be eligible to write the End Term +Exam, a learner’s average assignment +score should be >= 40/100 AND they +should have appeared for at least one +out of the three quizzes. -• +7 -Big Data Analytics: data mining and machine learning -– + Pass Criteria for Each Course -Large-scale machine learning, data mining and data visualization +Total Course Score (T) = 40% of Average quiz score (Q) + 60% of End Term Score (E) -Big Data Computing: data center support for Analytics -– +A learner is deemed to have passed a course only if the below mentioned two +conditions are met: -Big data collection and transformation, integration and distributed -data management and computing +Total Course Score (T) >= 50/100 AND -Big data sampling and statistical theory, Big data security and privacy +End Term exam score (E) >= 40/100 -Big Data Theory, Privacy&Security issues on Analytics -– -Big Data Science: 4th Paradigm – Analytics for Science and -Engineering -– +These criteria are the same for all learners participating in the course. -Big Data and Multi-disciplines (Bio, Chemistry, Engineering, Social) +An absolute grading system, as described in the below table, will be followed: -7 +Condition satisfied by ‘T’ - Total Course Score (Out Of 100) +and ‘E’ - End Semester Score (Out Of 60) + +Letter Grade + +(T ≥ 90) AND (E ≥ 24) - Why Hong Kong is Ready for the Theme +(90 > T ≥ 80) AND (E ≥ 24) -• +(80 > T ≥ 70) AND (E ≥ 24) -We have the best researchers in machine learning, data mining, data -management, sensor networks, statistics, and multidisciplinary -research such as bioinformatics +(70 > T ≥ 60) AND (E ≥ 24) -– +(60 > T ≥ 50) AND (E ≥ 24) -– +S -– +A -China National 973 Projects on Big Data +B -IEEE Transactions on Big Data: EiC +C -ACM KDD Conferences: PC and Conference Chairs +D -Winner of Big Data related international competitions +(T < 50) OR (E < 24) OR Average assignment score <40/100 OR No quiz attended -– -New industries based on lots of data +U (Fail) -– +Learners Who Do Not Pass a Course -Financial industry, logistics industry, education sector, government -services, etc. +For any learner, if Total Course Score (T) < 50/100 or average assignment +score < 40/100 or no quiz attended, they have to repeat the entire course. This +includes submission of assignments and redoing all the subsequent quizzes and +end term exam. Fees for repeating the course is the same as the course fees. -We have many potential collaborators and partners -– +If Total Course Score (T) >= 50/100 but End Term Score (E) < 40/100, the learner has +2 options: -Huawei, Tencent, Baidu, Alibaba, Google, Microsoft, etc. +Attend only the end term exam whenever it is conducted in the subsequent +term. The fee to repeat an end term exam will be ₹1000 for foundational and +₹2000 for diploma / degree level courses. -• +They can repeat the entire course by paying the full course fees and +completing all assignments, invigilated quizzes and the end term exam. -• +Exam Cities + +The Invigilated Quizzes and End Term +exams are scheduled to be conducted +in a limited number of cities across +India. Learners can select 3 options +among the cities, and they will be +expected to attend the exam in the +City / Centre that has been allotted +to them. The allotted city will be from +these 3 options. + +The list of exam cities where +invigilated exams will be scheduled +may change later depending on the +location of majority of our learners. +Here is the link to view the current +exam cities . 8 - Big Data Analytics Workflow - - 9 - -Data -Extraction - -Data -Integration - -• HKUST -CUHK -• -Baptist -• -HSBC -• -Astri -• - -• HKUST -HKU -• -City U -• -Alibaba -• -Astri -• - -Applications - -• Biology and Genetics -• -• -• -• -• -• -• -• - -Chemistry -Physics -Government Policies -Social Sciences -General Health -Logistics -Finance -Business - -Data Mining & -Visualization - -• HKUST -CUHK -• -City U -• -Huawei -• -Tencent -• - -Big Data -Computing &Data -Management - -• CUHK -• -• -• -• -• - -HKUST -HKU -PolyU -City U -Baptist U - - Multi-disciplinary Big-data Analytics - -• - -Objectives: -• - -Interdisciplinary, -research -technological big data analytics - -heterogeneous - -multi-university, - -on - -multi-team -and - -scientific + Fee Structure + +For details about application fees, check Application Process (Pg 14) section. + +Each term, pay only for the courses you register for! + +Foundational Level - ₹4000 per course. +Skill Enhancement courses in Diploma / Degree Level - ₹5,000 per course. +All other courses in Diploma / Degree Level - ₹10,000 per course. + +Fee waivers depend on category of learner and family income: + +Family Income > 5LPA + +Family Income > 1LPA +and <=5LPA + +Family Income <=1LPA + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +Applicable +Fees + +Certificates +Required + +General + +Full Fee + +OBC + +Full Fee + +- + +- + +50% waiver + +EWS + Family +Income + +75% waiver + +EWS + Family +Income + +50%waiver + +OBC-NCL + +Family Income + +75% waiver + +OBC-NCL + +Family Income + +SC / ST + +50% waiver + +SC / ST + +50% waiver + +SC / ST + +75% waiver + +PwD + +50% waiver + +PwD + +50% waiver + +PwD + +75% waiver + +SC / ST + Family +Income + +EWS / OBC-NCL + +PwD + Family Income + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +75% waiver + +SC / ST + PwD + +The term family income for the purpose of availing fee waivers includes the income +of the candidate, the income of his/her parents and spouse, also the income of his/ +her siblings and children below the age of 18 years. + +OBC-NCL / EWS / Family Income certificates need to be obtained in the prescribed +format. Only those who clear Qualifier Exam and wish to avail fee waiver based on +family income need to submit the Family Income certificate. (Note that Family Income +certificate is valid only for one year and will need to be submitted afresh each year to +continue availing corresponding fee waiver.) +Format for OBC-NCL certificate  +Format for EWS certificate  +Format for Family Income certificate  + +Tentative Fee for Each Term, depending on your end Goal and possible paths to +reach the goal: + + Foundational Certificate  + + Diploma in Programming  | Diploma in Data Science  | Both Diplomas  + + BSc in Programming and Data Science  + +*Note: If a learner does not pass a course in the term they registered for it - refer +section about Learners who do not pass a course for details on whether they will +need to repeat entire course or only repeat the end term exam. Fee to repeat an entire +course will be the same as the full course fee. Fee to only repeat the end term exam is +₹1000 for foundational level courses and ₹2000 for diploma / degree level courses. + +9 + + Foundational Level + +The Foundational Level comprises +courses in Mathematics, Statistics, Basics +of Programing and Python, and English. +These courses have been chosen to +ensure that the learner who passes these +successfully is well prepared to proceed +to the Diploma Level courses. + +8 courses +32 credits + +1 - 3 years +10 hrs/course/week + +₹32,000* +*Refer Fee Structure (Pg 9). + +Requirements for registration + +The learner should apply for and clear the Qualifier Process (Pg 14). + +Options on successful completion + +Learners have the following two options when they successfully complete all 8 +Foundational Level courses: + +Exit: The learner may exit with a Foundational Certificate from Centre for +Continuing Education, IIT Madras. + +Proceed to next level: The learner can join the Diploma Level (Pg 10). + +Courses in Foundational Level + +Mathematics for Data Science I + +Statistics for Data Science I + +Mathematics for Data Science II + +Statistics for Data Science II + +Computational Thinking + +Programming in Python + +English I + +English II + +Diploma Level + +There are two sections in the Diploma +Level with courses for Diploma in +Programming and courses for Diploma +in Data Science. Each of these diplomas +comprise 5 core courses and 1 skill +enhancement course. These Diploma +Level courses assume that the learner is +well-versed with fundamentals. + +Requirements for registration + +6 + 6 courses +22 + 22 credits + +1 - 3 years +10 hrs/course/week + +₹55,000* + ₹55,000* +*Refer Fee Structure (Pg 9). + +The learner should have cleared all 8 Foundational Level (Pg 10) courses or should +have joined via Diploma Only Entry (Pg 13). 10 - Why Big Data needs Team Work? + Options on successful completion + +Learners have the following options based on the courses completed in this level: + +They may exit with a Diploma in Programming from IIT Madras. + +They may exit with a Diploma in Data Science from IIT Madras. + +They may exit with both Diplomas from IIT Madras. + +The learner can join the Degree Level (Pg 11) if they have completed all 12 +Diploma Level courses AND have not entered the program via Diploma Only +Entry option. + +Diploma in Programming + +The Diploma in Programming lays a +sturdy foundation in Databases and +Programming concepts with Data +Structures and Algorithms. The learner +goes on to apply these in the building +of a web application by the end of the +diploma. + +Courses for Diploma in Programming + +6 courses +22 credits + +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Database Management Systems + +Modern Application Development I + +Programming, Data Structures and +Algorithms using Python + +Modern Application Development II -• +Programming Concepts using Java -Big data analytics is necessarily a joint effort by -researchers from academic institutions, -government and society and industry. -– +Skill Enhancement I -The government and industry are sources of Big -Data, and providers of problems and challenges, -The academic researchers are solution +Diploma in Data Science -providers. +The Diploma in Data Science exposes +the learner to the holistic approach of +gathering, analysing, and interpreting +data for a variety of problems. The +courses on Business Data lays down the +context and the need for the data, while +the Machine Learning courses equip +the learner to use and analyse this data +towards impactful conclusions. -When it comes to package the solutions from -university labs to transfer to the real world, -universities and industry must work together to -build scalable and robust solutions. +Courses for Diploma in Data Science -– +6 courses +22 credits -– +1 - 2 years +10 hrs/course/week + +₹55,000* +*Refer Fee Structure (Pg 9). + +Machine Learning Foundations + +Business Data Management + +Machine Learning Theory + +Business Analytics + +Machine Learning Practice + +Skill Enhancement II 11 - Expected Outcomes + Degree Level + +The Degree Level has courses that +complete the requirements for the BSc +degree. + +Requirements for registration +The learner should have come in via +Regular Entry (Pg 14), cleared all 8 +Foundational Level (Pg 11) courses and +all 12 Diploma Level (Pg 11) courses. + +11 courses +40 credits + +1 - 3 years +10 hrs/course/week + +₹1,00,000* +*Refer Fee Structure (Pg 9). + +Exit: Once the learner successfully completes 40 credits from Degree Level courses, +they can exit with a BSc Degree in Programming & Data Science from +IIT Madras. + +Credit Requirements +Learners have to complete a total of 40 credits at the degree level: + +4 credits from mandatory course titled “Strategies for Professional Growth” +4 credits from 2 Skill Enhancement Courses of 2 credits each. These are to be +completed on NPTEL and the credits transfered to this degree program. +8 credits from 2 Core Courses of 4 credits each. There are 3 pairs of core course +options to pick from. +24 credits from Elective Courses - 6 courses of 4 credits each or more courses of +lesser credits. + +Core Courses +Learner has to do a pair of core courses where each course is of 4 credits. Following +are the options for Core Courses: + +Core Course Option I + +Core Course Option II -• +Core Course Option III -• +Operating Systems -• +Software Testing -• +Computer Architecture -New methodologies and solutions for Big Data -research -New applications that impact the society and -industry in Hong Kong and beyond, and -New digital economies created based on big -data -New educational programs for students; -cultivating leaders for the Big Data society and -industry. +Software Engineering + +AI: Search Methods for +Problem Solving + +Deep Learning + +Elective Courses +A total of 24 credits need to be obtained from Elective Courses. Listed below are 9 +courses of 4 credits each, including the 3 pairs of core courses. Excluding the 2 courses +that the learner has already picked as core courses, he / she can opt for the elective +courses in the following two ways: + +a. Obtain all 24 credits by picking 6 of the remaining 7 courses listed below. OR + +b. Obtain upto 12 credits from NPTEL (accepted courses list will be provided soon) and +obtain remaining credits by picking from a subset of the 7 courses below. + +Operating Systems + +Software Testing + +Computer Architecture + +Software Engineering + +Topics in Application +Development + +Topics in Machine +Learning + +AI: Search Methods for +Problem Solving + +Deep Learning + +Topics in Data Science 12 - Breakthroughs Expected + ADMISSIONS + +Important Dates + +Applications Closed (Regular Entry) + +- 15 September 2020 + +Weekly release of Term 1 Qualifier courses +content starts -• +- 5 October 2020 -• +Qualifier Exam Hall Ticket (only for those who +get minimum required marks in Assignments) -• +- 2 November 2020 -• +Qualifier Exam (only for those with Hall Ticket)* -New algorithms, methodologies, systems and -applications in Big Data -New knowledge from Big data applications in -Science, Engineering, and Societal Problems -New insights into Big data practices in real -world -New ways to protect security and privacy of big +- 22 November, 2020* -data relevant to individuals and organizations +Qualifier Exam Results + +- 7 December, 2020 + +Registration for Term 1 (only for those who clear +Qualifier Exam) + +- 7 to 27 December, 2020 + +Foundational Level Batch 1 starts + +- 4 January, 2021 + +*Dates for in-person exams are subject to government restrictions prevalent at that time. + +Two Admission Paths +1. Regular Entry + +Anyone joining the program via Regular Entry will join at the Foundational Level. + +The Foundational Level courses allow for interested applicants with a variety of +educational backgrounds to learn the basics (Math, Statistics, Computational Thinking, +Python programming and English) which help them take up the Diploma Level and +then the Degree Level courses. + +There are three exit options for those who enter the program via Regular Entry: +Exit 1. Foundational Level certificate from Centre for Continuing Education, IIT Madras. +Exit 2. Diploma in Programming AND/OR Diploma in Data Science from IIT Madras +Exit 3. BSc Degree in Programming and Data Science from IIT Madras + +View Regular Entry (Pg 14) for details on Eligibility, Application Process, Qualifier +Process, Qualifier Exam and Passing Criteria. + +2. Diploma Only Entry + +The Diploma Only Entry is primarily aimed at working professionals or learners who +already possess the fundamentals taught in Foundational Level courses and wish to +exit with one or two Diplomas (Diploma in Programming AND/OR Diploma in Data +Science) from IIT Madras. + +Learners entering via Diploma Only Entry can only do Diploma Level courses and +cannot proceed to the Degree Level courses. + +We have plans to open applications for Diploma Only Entry in May / June 2021. All +details related to Diploma Only Entry will be provided soon on our website . 13 - Call for Proposals in Big Data Analytics + Mandatory Requirements + +Access to good internet connection as +well as a laptop/desktop device will be +a key requirement to learn effectively. + +Learner should be able to travel to +assigned exam centres for quizzes and +exams, each term. + +Regular Entry +Eligibility to Apply + +Eligibility for those who cleared class 12 or equivalent in 2018 or earlier: + +1. + +2. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2018 or earlier. + +Eligibility for those who cleared class 12 or equivalent in 2019 or 2020: + +1. + +2. + +3. + +The candidate must have studied Mathematics and English in class 10. + +They should have cleared class 12 or equivalent in 2019 or 2020. + +The candidate should be: +(i) currently enrolled / previously enrolled in any Bachelor’s Degree program in +campus mode OR +(ii) completed 1 year Diploma (campus mode only) OR +(iii) completed CA Inter + +Note that applicants who cleared class 12 or equivalent in 2019 or 2020 may apply for the +qualfier process with eligibility criteria 1 & 2 alone, but will be required to fulfil criterion 3 to +join the foundational courses after clearing the qualifying exam. + +Application Process -• +Anyone who is eligible may apply by filling in the application form, uploading +required documents and paying the application fee. Check out Application +Process  section on our website for the latest updates and to start applying. -• +Application fee - ₹3,000 (For applicants belonging to PwD (>= 40% disability) / SC / ST +category - ₹1,500; For applicants belonging to PwD (>= 40% disability) and also SC / +ST category - ₹750) -Foundations in Big Data Analytics Research: -– +Qualifier Process -developing and studying fundamental theories, -algorithms, techniques, methodologies, -technologies to address the effectiveness and -efficiency issues to enable the applicability of Big -Data problems; +All applicants will have to go through a Qualifier Process, like a trial month, +wherein they will get access to 4 weeks of content for the four foundational level +courses - English I, Mathematics for Data Science I, Statistics for Data Science I and +Computational Thinking. -Innovative Applications in Big Data Analytics: -developing techniques, methodologies and -– -technologies of key importance to a Big Data -problem that requires the seamless cooperation -of domain scientists with big data researchers. +Video content of 2 - 3 hours duration will be released for each course on a weekly +basis. Along with the content, the first three weeks will also include an assignment +for each course that need to be completed and submitted online within the given +deadline. + +Average assignment score in each course will be calculated. Only those who get +minimum required average assignment scores in all four courses will be allowed to +write the Qualifier Exam (refer next page). 14 - Benefit for the Hong Kong Society + Minimum Average Assignment +Score required in each course + +General Learner + +SC / ST / PwD with 40% disability + +PwD with 40% disability & SC / ST + +OBC-NCL / EWS + +40% + +30% + +30% + +35% + +Qualifier Exam and Passing Criteria + +Only learners who get minimum required average assignment scores in all four +courses will be eligible for the in-person invigilated Qualifier Exam conducted at the +end of 4 weeks of coursework. Only these candidates will receive hall tickets. + +The Qualifier Exam is of 3 hours duration and covers all 4 subjects with 45 minutes +allotted per course. + +To pass the Qualifier Exam, the learner has to get a minimum Total Qualifier Exam +Score and a minimum Qualifier Exam Score in each course individually. Refer table +below. + +Only those who pass the Qualifier Exam will be eligible to register for Foundational +Level courses. + +Minimum required +Qualifier Exam Score +in each course + +Minimum required +Total Qualifier +Exam Score + +General Learner + +SC / ST / PwD with 40% disability -• +PwD with 40% disability & SC / ST -• +OBC-NCL / EWS -• +40% -• +30% -• +30% -Can Hong Kong transform into a more data- -driven society and maintain -leadership -globally? +35% -its +50% -How can companies in Hong Kong become -more competitive with Big Data technology? +40% -Can Data Science in Hong Kong’s research -fields benefit from strong foundations on big data? +40% -Can the government become more efficient with -big data driven methodologies in decision -making? +45% -Can education, finance, logistics and health -benefit from the ever increasing data? +Note: Relaxations in pass criteria indicated for various categories of learners, is +applicable ONLY for the qualifier process. There will be no relaxations in terms of +grades / pass criteria once registered into the program. + +Though everyone who passes the Qualifier Exam will be allowed to register for +Foundational Level, there will be limitations on the number of courses a learner will +be allowed to register for in the first two terms based on the Total Qualifier Exam +Score (S). + +Total Qualifier Exam Score (S) + +Number of courses allowed to +register for in first two terms + +Minimum required <= S < 50% + +up to 2 courses + +50% <= S < 70% + +up to 3 courses + +S >= 70% + +up to 4 course + +Maximum number of courses a learner will be allowed to register for in later terms +will depend similarly on the average of course scores in the previous two terms. 15 + Contact Us + +Online Degree Office, 3rd Floor, ICSR Building, + +IIT Madras, Chennai - 600036 + +7850-999966 (Mon-Fri 9am-6pm) + +support@onlinedegree.iitm.ac.in + +www.onlinedegree.iitm.ac.in  + +iitmadrasonline + +16 + diff --git a/q192/random_k4/random_2.md b/q192/random_k4/random_2.md index 6bfd67b67b9eb8baf6562a416c94e48f3b69ec3d..4f0fbb2c84e93fd7c3f72a0ecb704f9505f5e8f3 100644 --- a/q192/random_k4/random_2.md +++ b/q192/random_k4/random_2.md @@ -1,253 +1,1571 @@ -EBay Annual Report 2024 +Brookfield Asset +Management -Form 10-K (NASDAQ:EBAY) +Investor Presentation -Published: May 29th, 2024 +May 2025 -PDF generated by stocklight.com + Brookfield Asset Management is a leading pure-play global asset manager - UNITED STATES -SECURITIES AND EXCHANGE COMMISSION -Washington, D.C. 20549 +1. Strong fundraising capabilities -Form 10-K/A -Amendment No. 1 +2. Positioned in fast-growing sectors with long-term tailwinds -☒ +3. Best-in-class transaction execution and operational capabilities -☐ +4. Track record of delivering attractive risk-adjusted returns -ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934 +5. Continuously innovating and scaling to deliver long-term growth -For the fiscal year ended December 31, 2023. +2 + + Company Overview + +3 + + Brookfield Asset Management today + +$1T+ + +Assets Under Management + +A leading global asset manager + +NYSE / TSX: BAM + +~$90B market cap + +$2.6B LTM fee-related earnings + +$549B fee-bearing capital + +Over 2,500 asset management employees +globally and ~250,000 operating employees + +125-year history as an owner-operator + +Best-in-class track record of generating strong returns -OR +All figures in this presentation are as of March 31, 2025, unless otherwise stated +See Notice to Recipients -TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934 +4 + + Brookfield Asset Management – Structure + +Brookfield Asset Management (BAM) +Manages over $1 trillion of assets within long-term private funds, perpetual strategies, +permanent capital vehicles, and liquid products, across five key verticals: + +Renewable Power +& Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit -For the Transition Period from to . +Benefits of the Brookfield Model + +Focused, asset-light model + +Long-term, consistent fee stream + +Combines sophisticated asset manager with access to Brookfield’s over $170B perpetual capital + +Transparent disclosures explain fund management, fee structure and transactions + +We are continuously working to improve our business model for investors and shareholders + +5 -Commission file number 001-37713 + Our business model is simple and effective +Our structure provides access to all major pools of capital and provides a range of investors +with a tailored approach to meet their objectives -eBay Inc. -(Exact name of registrant as specified in its charter) +Fundraising channels -Delaware -(State or other jurisdiction of -incorporation or organization) +• Institutional -2025 Hamilton Avenue -San Jose , California -(Address of principal executive offices) +• Insurance -77-0430924 -(I.R.S. Employer -Identification No.) +• Private Wealth -95125 -(Zip Code) +• Public Markets -Registrant’s telephone number, including area code: -(408) 376-7108 +Capital +raising -Securities registered pursuant to Section 12(b) of the Act: +Brookfield +Asset +Management -Title of each class -Common stock +Investing -Trading symbol -EBAY +Investing in attractive, +durable asset classes -Name of exchange on which registered -The Nasdaq Global Select Market +Transition -Securities registered pursuant to Section 12(g) of the Act: None +Infrastructure -Indicate by check mark if the registrant is a well-known seasoned issuer, as defined in Rule 405 of the Securities Act. Yes ☒ No ☐ +Private Equity -Indicate by check mark if the registrant is not required to file reports pursuant to Section 13 or Section 15(d) of the Act. Yes ☐ No ☒ +Real Estate -Indicate by check mark whether the registrant (1) has filed all reports required to be filed by Section 13 or 15(d) of the Securities Exchange Act of 1934 during the preceding 12 months (or for such shorter period that the -registrant was required to file such reports), and (2) has been subject to such filing requirements for the past 90 days. Yes ☒ No ☐ +Credit -Indicate by check mark whether the registrant has submitted electronically every Interactive Data File required to be submitted pursuant to Rule 405 of Regulation S-T (§ 232.405 of this chapter) during the preceding 12 -months (or for such shorter period that the registrant was required to submit such files). Yes ☒ No ☐ +6 -Indicate by check mark whether the registrant is a large accelerated filer, an accelerated filer, a non-accelerated filer, a smaller reporting company, or an emerging growth company. See the definitions of “large accelerated -filer,” “accelerated filer,” “smaller reporting company” and “emerging growth company” in Rule 12b-2 of the Exchange Act. + Strong culture built on Brookfield’s 125-year history as an owner-operator -Large accelerated filer +Our values -Non-accelerated filer +Our approach -☒ -☐ +Disciplined -Accelerated filer +Entrepreneurial -Smaller reporting company +Collaborative -Emerging growth company +Owner-Operators -☐ -☐ -☐ +Long Term View -If an emerging growth company, indicate by check mark if the registrant has elected not to use the extended transition period for complying with any new or revised financial accounting standards provided pursuant to -Section 13(a) of the Exchange Act. ☐ +Focus on Operational +Excellence - Indicate by check mark whether the registrant has filed a report on and attestation to its management’s assessment of the effectiveness of its internal control over financial reporting under Section 404(b) of the Sarbanes- -Oxley Act (15 U.S.C. 7262(b)) by the registered public accounting firm that prepared or issued its audit report. ☒ +7 -If securities are registered pursuant to Section 12(b) of the Act, indicate by check mark whether the financial statements of the registrant included in the filing reflect the correction of an error to previously issued financial -statements. ☐ + The Brookfield Ecosystem provides unparalleled insights -Indicate by check mark whether any of those error corrections are restatements that required a recovery analysis of incentive-based compensation received by any of the registrant’s executive officers during the relevant -recovery period pursuant to §240.10D-1(b). ☐ +30+ -Indicate by check mark whether the registrant is a shell company (as defined in Rule 12b-2 of the Act). Yes ☐ No ☒ +Countries -As of June 30, 2023, the aggregate market value of the registrant’s common stock held by non-affiliates of the registrant was $ -Market. +300+ -23,784,458,062 based on the closing sale price as reported on The Nasdaq Global Select +Portfolio companies -502 million shares of common stock issued and outstanding as of May 24, 2024. +Benefits of the +Brookfield Ecosystem -Auditor Name: +Helps us identify themes and trends, spot +pockets of value and source attractive +investment opportunities -PricewaterhouseCoopers LLP +Enhances intelligence and improves accuracy +in underwriting -Auditor Location: +Drives operational improvements and +boosts returns -San Jose, California +Hospitality -Auditor Firm ID: +Logistics & +Self-Storage -238 +Energy +Transition + +Housing + +Mixed Use, Retail +& Entertainment + +Business +Services + +Transport + +Infrastructure +Services + +$1+ trillion +AUM + +Utilities + +Solar + +Hydro + +Office + +Wind + +Technology Services + +Data + +Enables us to execute at scale with conviction, +regardless of market sentiment + +Midstream + +Industrials + +Healthcare + +Science & Innovation + +See Notice to Recipients + +8 + + Sustainable Growth Model + +9 + + Brookfield Asset Management’s sustainable growth model + +1 + +Pick good +sectors + + Well-positioned +in fast growing +alternative investment +management + + Focused on the +Backbone of the +Global Economy + ++ 2 - EXPLANATORY NOTE +Raise +capital + + Longstanding -This Amendment No. 1 on Form 10-K/A (this “Amendment”) amends Part IV, Item 15 of the Annual Report on Form 10-K for the fiscal year ended -December 31, 2023 of eBay Inc. (the “Company”), as filed with the U.S. Securities and Exchange Commission (the “SEC”) on February 28, 2024 (the -“Form 10-K”). The purpose of this Amendment is to provide financial statements for Adevinta ASA (“Adevinta”), pursuant to Rule 3-09 of Regulation S-X -for the years ended December 31, 2023 (audited), 2022 (unaudited) and 2021 (audited) and as of December 31, 2023 (audited) and 2022 (unaudited). In -accordance with Rule 3-09(b)(1), Adevinta’s financial statements are being filed herewith as Exhibit 99.01 within six months after the end of Adevinta’s -fiscal year. In addition, the Company is filing the consent of the independent auditors of Adevinta as Exhibit 23.02 and new certifications by the -Company’s Chief Executive Officer and Chief Financial Officer as Exhibits 31.03, 31.04, 32.03 and 32.04, respectively, as required by Rule 12b-15 under -the Securities Exchange Act of 1934, as amended. +partnerships with +institutional investors -Except as otherwise expressly noted, this Amendment does not modify or update in any way (i) the consolidated financial position, the results of -operations or cash flows of the Company, or (ii) the Form 10-K; nor does it reflect events occurring after the filing of the Form 10-K. Among other things, -forward-looking statements made in the Form 10-K have not been revised to reflect events that occurred or facts that became known to us after the filing -of the Form 10-K, and such forward-looking statements should be read in their historical context. Furthermore, this Amendment should be read in -conjunction with the Form 10-K and any subsequent filings with the SEC. + Globally diversified -Unless the context requires otherwise, all references to “we,” “our,” “us” or “eBay” mean eBay Inc., a Delaware corporation, and its consolidated -subsidiaries. +reach + + Access to Brookfield’s +permanent capital +and insurance +business + ++ 3 - ITEM 15: EXHIBITS AND FINANCIAL STATEMENT SCHEDULE +Invest +wisely -a. The following documents are filed as part of this report: + Assess opportunities -PART IV +to build or buy -1. Consolidated Financial Statements: + Leverage the -The consolidated financial statements of the Company, as listed in Item 15 of the Form 10-K, are included in Item 15 of the Form 10-K. +Brookfield Ecosystem -2. Financial Statement Schedule: + Scale through organic +and inorganic growth -The financial statement schedule of the Company, as listed in Item 15 of the Form 10-K, is included in Item 15 of the Form 10-K. -The financial statements of Adevinta required by Rule 3-09 of Regulation S-X are provided as Exhibit 99.01 to this Amendment. ++ -3. Exhibits Required by Item 601 of Regulation S-K: +4 -The exhibits listed in the Exhibit Index of the Form 10-K and this Amendment are filed with, or incorporated by reference in, this report. +Operate +well -b. The information required by this Item is set forth in the exhibits listed in the Exhibit Index of the Form 10-K and this Amendment filed with, or + Invest with an -incorporated by reference in, this report. +Owner-operator +mindset -c. Financial Statement Schedule and Separate Financial Statements of Subsidiaries Not Consolidated and Fifty Percent or Less Owned Persons. + Enabled by an -Adevinta was deemed a significant equity investee under Rule 3-09 of Regulation S-X for the fiscal years ended December 31, 2023 and -December 31, 2021 (though not for the fiscal year ended December 31, 2022). As such, financial statements of Adevinta for the fiscal years ended -December 31, 2023, 2022 and 2021, respectively, are required to be filed by amendment to this Annual Report on Form 10-K within six months of -Adevinta’s fiscal year end. Accordingly, Adevinta’s financial statements for its fiscal year ended December 31, 2023 are provided as Exhibit 99.01 -to this Amendment. +Owner-operator +structure -4 += - No. +Generating +excellent +risk-adjusted +returns -23.02 +Results propel a +virtuous cycle +supporting the entire +growth model -31.03 +Significant tailwinds driven by AUM growth, alternatives growth, and favorable sectors -31.04 +10 -32.03 + The past 20 years have seen major tailwinds for alternatives… +and there is much more to come -32.04 +2002 -99.01 +Today -104 +$25T -INDEX TO EXHIBITS +Alternatives AUM -Exhibit Description +~15% -Filed or Furnished with -this 10-K/A +$150T+ +Global AUM -Incorporated by Reference +$2T -Form +Alternatives AUM -File No. +~5% -Date Filed +$40T +Global AUM -Consent of Ernst & Young AS related to Exhibit 99.01. +2032E -Certification of Registrant’s Chief Executive Officer, as required by -Section 302 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +$60T+ -Certification of Registrant’s Chief Financial Officer, as required by -Section 302 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +Alternatives AUM -Certification of Registrant’s Chief Executive Officer, as required by -Section 906 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +20%+ -Certification of Registrant’s Chief Financial Officer, as required by -Section 906 of the Sarbanes-Oxley Act of 2002, dated May 28, 2024. +$300T+ -Consolidated financial statements of Adevinta ASA for the years ended -December 31, 2023 (audited), 2022 (unaudited) and 2021 (audited) and -as of December 31, 2023 (audited) and 2022 (unaudited). +Global AUM -Cover Page Interactive Data File (formatted as Inline XBRL and -contained in Exhibit 101). +Key growth drivers -X +• Global AUM growth -X +• Institutional investors -X +increasing allocations to +alternatives -X +• Investors consolidating +their exposure to the +largest and most +diversified managers -X +Why investors are +choosing to invest their +growing capital base in +alternative assets -X +Excess returns -X +Diversification -5 +Less volatility + +Predictable cash flow + +Sources: Preqin; Bain analysis; PwC; Company estimates + +11 + + We are focused on the Backbone of the Global Economy + +Digitalization + +Decarbonization + +Deglobalization + +Our assets live at the epicenter of these decades-long megatrends +Trillions of dollars will need to be invested as the world adapts to paradigm shifts + +12 + + We are growing our sources of capital while expanding into new ones + +New Sources of Capital: Tremendous Growth + +• + +Insurance Solutions + +• Private Wealth + +• Manager Acquisitions + +• Strategic Partnerships + +Existing Sources of Capital: More to Do + +• + +Institutional Investors + +• Public Markets + +13 + + Unique and diverse pools of capital support our fundraising growth + +Institutional Investors + +300+ Client & Marketing Professionals +2,200+ Institutional Relationships + +Insurance Solutions + +$15B+ Annual inflows from Brookfield +Wealth Solutions + +150 Existing Third-Party Insurance Relationships + +Private Wealth + +150 dedicated professionals +160+ Wealth Management Platform Relationships + +Public Markets + +See Notice to Recipients + +3 permanent capital vehicles +~$60B Market Value +~2,500 Institutional Shareholders + +BAM’s fundraising +differentiators + +Breadth and specialization – +at the same time + +Operational expert, not +financial engineers + +Global capabilities with local +mindset + +Partnership mentality with +360° engagement – investors +are more than “just an LP” + +Access to co-investment, +partnership opportunities +and customized solutions + +14 + + Our leadership positions set us up to outgrow the market +We have positioned ourselves as a leader within our sectors of focus +across the Brookfield Ecosystem + +Renewable Power +& Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit + +Largest global +transition platform + +Leading infrastructure +strategy + +Best-in-class PE +flagship platform + +One of the largest + real estate investors + +Unmatched credit +capabilities across +funds and partnerships + +$125B +AUM + +$214B +AUM + +$147B +AUM + +$272B +AUM + +$321B +AUM + +See Notice to Recipients + +15 + + Competitive advantages across all our business groups + +Scale and Global +Footprint + +Sourcing +Capabilities + +Owner-Operator +History + +Access to +Debt Capital + +Leveraging the +Brookfield Ecosystem + +16 + + We are consistently expanding our capabilities + +Sometimes we build + +Sometimes we buy + +Example: Renewables expertise... + +Renewable Power & Transition FBC +($B) + +Example: Expanding our credit platform by +adding Oaktree... + +25% +CAGR + +$21 + +2018 + +$64 + +2024 + +…growing into energy transition +• Flagship Transition Fund +• Catalytic Transition Fund +• Renewables Sidecar +• Credit Opportunities + +…enabled further growth + +Insurance Platform + +• +• Private Wealth +• Other Partnerships + +Note: 2024 FBC includes targeted fund size for BGTF II, which is currently fundraising. + +17 + + We partner with leading managers to expand our capabilities + +2021 + +2023 + +2025 + +2019 + +2022 + +2024 + +1. Angel Oak acquisition announced in April 2025. Closing is subject to customary conditions + +1 + +18 + + Business Group Overview + +19 + + Renewable Power & Transition – Overview +Brookfield is a leading owner, operator and developer of renewable energy +and decarbonization technology +$125B + +$58B + +substantial deal pipeline + +93 + + Large footprint, extensive experience and + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals + + Unsurpassed capabilities, relationships and + +access to scale capital + +Key Sectors + +Products + +Hydro +8,300 MW + +Wind +16,900 MW + +Solar +11,700 MW + +Flagship Fund + +Complementary Strategies + +Global Transition (BGTF Series) + +Catalytic Transition Fund (CTF) + +Asia Renewable Fund (BCRF) + +DG, Storage & +Sustainable Solutions +4,900 MW + +Permanent Capital Vehicle + +Brookfield Renewable Partners (BEP / BEPC) + +Key drivers: + +Accelerating demand for power from digitalization, electrification and AI, and renewables position +as the lowest cost source of bulk power globally + +Global goals of net-zero emissions, energy security and independence, alongside corporate +demand and support from favorable legislation + +20 + + Infrastructure – Overview +Brookfield is one of the world’s largest infrastructure investors, owners and operators + +$214B + +Assets Under Management + +$96B + +Fee-Bearing Capital + +127 + +Investment Professionals + + Ideally positioned at the epicenter of the global secular + +trends of deglobalization, decarbonization and digitalization + + Operational expertise and institutional knowledge + +Key Sectors + +Products + +Utilities +$66B + +Transport +$48B + +Data +$65B + +Midstream +$25B + +Flagship Fund + +Complementary Strategies + +Infrastructure Core Plus (BIF Series) + +Brookfield Supercore Infrastructure Partners (BSIP) + +Brookfield Infrastructure Structured Solutions (BISS) + +Permanent Capital Vehicle + +Brookfield Infrastructure Partners (BIP / BIPC) + +Key drivers: + +$94 trillion of infrastructure investment required by 2040 + +AI infrastructure emerging as a new digitalization-themed asset class representing +a potential market opportunity of more than $8 trillion over the next five years + +21 + + Private Equity – Overview +Brookfield is one of the most experienced private equity investors globally + +$147B + +Assets Under Management + +$43B + +Fee-Bearing Capital + +171 + +Investment Professionals + + Global sourcing engine + Experienced team with local operating expertise + Value creation through operational improvement + +Key Sectors + +Products + +Industrials +$42B + +Infrastructure Services +$22B + +Flagship Fund + +Complementary Strategies + +Opportunistic Private Equity (BCP Series) + +Brookfield Special Investments (BSI) + +Business Services +$52B + +Directly Held & Other +$31B + +Permanent Capital Vehicle + +Brookfield Middle East Partners (BMEP) + +Brookfield Business Partners (BBU / BBUC) + +Brookfield Sponsor Solutions (BSS) + +Brookfield Financial Infrastructure Partners (BFIP) + +Key drivers: + +Convergence of technology and industrial landscape reawakening displaced industries + +Revitalization of industrial sector requires significant capital and operational expertise + +22 + + Real Estate – Overview +Brookfield is one of the largest real estate investors with a diversified portfolio +in the world’s most well-established markets +295 +$272B + +$100B + + Scale, relationships, operating expertise + + Assets well positioned to weather market cycles and + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals + +deliver long term value + +Key Sectors + +Logistics, +Storage & NNN +160M+ sf logistics + +Mixed Use, Retail +& Entertainment +~130M combined sq. ft. + +Housing +110K+ units + +Office +~190M sq. ft. + +Products + +Hospitality +40K+ keys + +Science & +Innovation +5M+ sq. ft. + +Flagship Fund + +Complementary Strategies + +Opportunistic Real Estate (BSREP Series) + +Brookfield Premier Real Estate Partners (BPREP) + +Permanent Capital Vehicle + +Brookfield Single Family Rental (BSFR) + +Brookfield Property Group (BPG) + +Brookfield Non-Traded REIT + +Key drivers: + +Demographic shifts and market cycle dynamics driving increasing real estate needs + +Digitalization & Decarbonization rapidly transforming the logistics & hospitality industries + +23 + + Credit – Overview +Brookfield’s credit strategies include our longstanding private credit and direct lending funds +and our partnerships with leading credit managers + +$321B + +$252B + +647 + +Assets Under Management + +Fee-Bearing Capital + +Investment Professionals1 + + Marquee partnerships with leading credit managers + Cycle-tested platform + Extensive lending capabilities + Longstanding relationships with financial institutions + +Brookfield Credit Capabilities + +Partner Managers + +Brookfield Wealth Solutions (BWS) and 3rd Party Capabilities +- Private Credit (SMAs) + +Real Estate + +Asset Based + +- Private Funds + +Infrastructure & Renewable + +- Liquid Credit + +Corporate + +Liquid + +2 + +Key drivers: + +Equity integration – access to the Brookfield Ecosystem allows us to meet the evolving needs of +clients globally and provide investment opportunities across the full capital stack + +Origination capabilities – robust network of proprietary origination platforms and strategic +partnerships with specialized credit managers + +1. +2. + +Investment professionals include Brookfield Credit Group, Brookfield Wealth Solutions, Public Securities Group (PSG), and partner manager professionals. +Angel Oak acquisition announced in April 2025. Closing is subject to customary conditions + +24 + + Financial Performance and +Growth Targets + +25 + + Our results demonstrate the stability, predictability and +growth of our earnings + +16% +CAGR + +$2,108 + +$418 + +$2,241 + +$457 + +$2,456 + +$539 + +$1,376 + +$312 + +$1,827 + +$364 + +$2,000 + +$1,500 + +$1,000 + +$500 + +$0 + +2020 + +2021 + +2022 + +2023 + +2024 + +Full year periods as of December 31 + +FBC ($B) + +FRE ($M) + +…and should continue to drive earnings growth, giving us confidence in delivering +15%+ long term annual dividend growth + +See Notice to Recipients + +$500 + +$400 + +$300 + +$200 + +$100 + +$0 + +26 + + We have delivered strong results over the long term + +Fund + +Gross IRR + +Net IRR + +Infrastructure / Renewable Power + +Private Equity + +Real Estate + +Opportunistic Credit + +Transition + +15% + +26% + +20% + +22% + +n.m. + +12% + +20% + +16% + +16% + +n.m. + +~$225 billion of profits generated for investors over the past 30 years + +See Notice to Recipients + +27 + + How we will generate future growth + +Strategies + +Sectors + +Enhancing +Fundraising Channels + +Diversifying & Scaling +Product Offerings + +Building & Buying +New Businesses + +Energy Transition + +AI Infrastructure + +Private Credit + +28 + + Growth is driven across all our business groups, as our long-term or +permanent capital base continues to grow + +Q1-25 + +$58 + +$252 + +87% + +Long Term or +Permanent +Capital + +$96 + +$43 + +$100 + +$549B +Fee-Bearing Capital + +Growing +To + +2029 + +$110 + +$590 + +92% + +Long Term or +Permanent +Capital + +$160 + +$90 + +$140 + +$1.1T +Fee-Bearing Capital + +Renewable Power and Transition + +Infrastructure + +Private Equity + +Real Estate + +Credit + +See Notice to Recipients + +29 + + Growth in fee-bearing capital and carried interest will propel earnings + +2024 + +17% +CAGR + +2029 + +2024 + +18% +CAGR + +2029 + +FRE per share + +DE per share + +With a projected ~95% payout ratio, the plan supports 15%+ annual dividend growth + +30 + + With further upside in carried interest +We are steadily growing our carried interest base with realizations beginning +to impact earnings in 2029, creating the second leg of growth + +Gross Carried Interest to BAM + +~$2B + +Estimated Realizations +in 2029 + +Growing +To + +~$7B + +Estimated Realizations +in 2034 + +See Notice to Recipients + +31 + + Robust Liquidity Position to Drive Growth + +Our strong liquidity and new debt capacity allow us to capitalize on opportunities and defend +against downside risks + +$1.4B + +Corporate Liquidity + +$750M + +Inaugural Debt Offering + +A / A- + +Investment Grade +Corporate Ratings(1) +(Fitch / S&P) + +We are strategically positioned to support our enhance liquidity through a variety of +fund-raising channels + +(1) + +A security rating is not a recommendation to buy, sell or hold securities and may be subject to revision or withdrawal at any time + +32 + + The best is yet to come for BAM + +1. Leading pure-play global asset manager + +Long-term growth targets + +2. Strong fundraising capabilities + +3. Positioned in fast-growing sectors with + +long-term tailwinds + +4. Best-in-class transaction execution and + +operational capabilities + +5. Track record of delivering attractive returns + +6. Continuously innovating and scaling to + +deliver growth + +See Notice to Recipients + +5-year CAGR + +By 2029 + +FBC + +16% $1.1T + +FRE + +17% + +$5.0B + +DE + +18% + +$5.1B + +33 + + Appendix + +34 + + Senior management team + +Bruce Flatt +CEO, Brookfield Asset +Management + +Connor Teskey +President, Brookfield Asset +Management; CEO Renewable +Power & Transition + +Justin Beber +Chief Operating Officer + +Joined Brookfield: 1990 + +2012 + +2007 + +Hadley Peer Marshall +Chief Financial Officer, +Brookfield Asset Management +and Managing Partner, +Infrastructure +2015 + +Brian Kingston +CEO, Real Estate + +Anuj Ranjan +CEO, Private Equity + +Craig Noble +CEO, Credit + +Sam Pollock +CEO, Infrastructure + +2001 + +2006 + +2004 + +1994 + +Brookfield’s senior management team possesses deep experience in our target regions, sectors and asset classes, +as well as decades of investment expertise built on managing assets and businesses across market cycles + +35 + + Notice to Recipients + +Brookfield Asset Management Ltd. (“BAM,” “Brookfield Asset +Management” or “Brookfield”) is not making any offer or +invitation of any kind by communication of this presentation and +under no circumstance is it to be construed as a prospectus or +an advertisement. + +Except where otherwise indicated herein, the information +provided herein is based on matters as they exist as of the +date of this presentation and not as of any future date, is +subject to change, and, unless required by law, will not be +that +updated or otherwise revised +subsequently becomes available or circumstances existing or +changes occurring after the date hereof. + +information + +to reflect + +Unless otherwise noted, all references to “$” or “Dollars” are to +U.S. Dollars. This presentation +financial +information as of and for the last twelve months ended +2024. +December + +includes certain + +31, + +Cautionary Note Concerning Forward-Looking Statements + +information within + +include statements + +This presentation contains forward-looking statements within +the meaning of Section 27A of the Securities Act of 1933, as +amended (the “Securities Act”), Section 21E of the Securities +Exchange Act of 1934, as amended (the “Exchange Act”) and +forward-looking +the meaning of other +relevant securities legislation, including applicable securities +laws in Canada, which reflect our current views with respect to, +among other things, our operations and financial performance. +Forward-looking statements +that are +predictive in nature, depend upon or refer to future results, +events or conditions, and include, but are not limited to, +statements which reflect management’s current estimates, +beliefs and assumptions regarding the operations, business, +financial condition, expected financial results, performance, +prospects, opportunities, priorities, targets, goals, ongoing +objectives, strategies, capital management and outlook of +Brookfield Asset Management Ltd., as well as the outlook for +North American and international economies for the current +fiscal year and subsequent periods, and which in turn are +based on our experience and perception of historical trends, +current conditions and expected future developments, as well +as other factors management believes are appropriate in the +circumstances. The estimates, beliefs and assumptions of +Brookfield Asset Management Ltd. are inherently subject to +significant business, economic, competitive and other + +uncertainties and contingencies regarding future events and as +such, are subject to change. Forward-looking statements are +typically identified by words such as “expect”, “anticipate”, +“believe”, “foresee”, “could”, “estimate”, “goal”, “intend”, “plan”, +“seek”, “strive”, “will”, “aim”, “may” and “should” and similar +expressions. +In particular, the forward-looking statements +contained in this presentation include statements referring to +Brookfield Asset Management Ltd.’s growth prospects and +liquidity, and the impact of acquisitions and dispositions on our +business. + +Although we believe that such forward-looking statements are +based upon reasonable estimates, beliefs and assumptions, +actual results may differ materially from the forward-looking +statements. Factors that could cause actual results to differ +materially from those contemplated or implied by forward- +looking statements include, but are not limited to: (i) returns +that are lower than target; (ii) the impact or unanticipated +impact of general economic, political and market factors in the +countries in which we do business; (iii) the behavior of financial +markets, including fluctuations in interest and foreign exchange +rates and heightened inflationary pressures; (iv) global equity +and capital markets and the availability of equity and debt +financing and refinancing within these markets; (v) strategic +actions including acquisitions and dispositions; the ability to +complete and effectively integrate acquisitions into existing +operations and the ability to attain expected benefits; (vi) +changes in accounting policies and methods used to report +financial condition (including uncertainties associated with +critical accounting assumptions and estimates); (vii) the ability +to appropriately manage human capital; (viii) the effect of +applying future accounting changes; (ix) business competition; +(x) operational and reputational risks; (xi) technological change; +(xii) changes in government regulation and legislation within the +(xiii) governmental +countries +investigations and sanctions; (xiv) litigation; (xv) changes in tax +laws; (xvi) ability to collect amounts owed; (xvii) catastrophic +and +events, +as +of +epidemics/pandemics; +international conflicts and other developments, +including +introduction, +terrorist acts and cyberterrorism; +withdrawal, success and timing of business initiatives and +strategies; (xx) the failure of effective disclosure controls and +procedures and internal controls over financial reporting and +other risks; (xxi) health, safety and environmental risks; (xxii) +the maintenance of adequate insurance coverage; (xxiii) the +existence of information barriers between certain businesses + +in which we operate; + +earthquakes, + +hurricanes + +possible + +impact + +(xviii) + +such + +(xix) + +the + +the + +within our asset management operations; (xxiv) risks related to +our renewable power and transition, infrastructure, private +equity, real estate and credit strategies; and (xxv) factors +detailed from time to time in our documents filed with the +securities regulators in the United States and Canada. + +We caution that the foregoing list of important factors that may +affect future results is not exhaustive, and other factors could +also adversely affect future results. Readers are urged to +consider these risks, as well as other uncertainties, factors and +forward-looking +assumptions carefully +statements and are cautioned not to place undue reliance on +such forward-looking statements, which are based only on +information available to us as of the date of this presentation. +Except as required by law, we undertake no obligation to +publicly update or revise any forward-looking statements, +whether written or oral, that may be as a result of new +information, future events or otherwise. + +in evaluating + +the + +Past performance is not indicative nor a guarantee of future +results. There can be no assurance that comparable results will +be achieved in the future, that future investments will be similar +to historic investments discussed herein, that targeted returns, +growth objectives, diversification or asset allocations will be +met or that an investment strategy or investment objectives will +be achieved (because of economic conditions, the availability +of appropriate opportunities or otherwise). + +36 + + Notice to Recipients + +in + +forth + +(including changes + +Target returns and growth objectives set +this +presentation are for illustrative and informational purposes only +and have been presented based on various assumptions made +by Brookfield Asset Management Ltd. in relation to the +investment strategies being pursued, any of which may prove +to be incorrect. There can be no assurance that targeted +returns or growth objectives will be achieved. Due to various +risks, uncertainties and changes +in +economic, operational, political or other circumstances) +beyond our control, the actual performance of the business +could differ materially from the target returns and growth +objectives set forth herein. In addition, industry experts may +disagree with the assumptions used in presenting the target +returns and growth objectives. No assurance, representation or +warranty is made by any person that the target returns or +growth objectives will be achieved, and undue reliance should +not be put on them. Prior performance is not indicative of +future results and there can be no guarantee that we will +achieve the target returns or growth objectives or be able to +avoid losses. + +The information in this presentation does not take into account +an investor’s investment objectives, financial situation or +particular needs and nothing contained herein should be +construed as legal, business or tax advice. Each prospective +investor should consult its own attorney, business adviser and +tax advisor as to legal, business, tax and related matters +concerning the information contained herein. + +Non-GAAP Measures + +for + +including + +business, + +We prepare our financial statements in conformity with the +accounting principles generally accepted in the United States +of America (“U.S. GAAP”). This presentation discloses a +number of non-GAAP financial and supplemental financial +in monitoring our asset +measures which are utilized +management +performance +measurement, capital allocation and valuation purposes. We +believe that providing these performance measures is helpful +to investors in assessing the overall performance of our asset +management business. These non-GAAP financial measures +should not be considered as the sole measure of our business’ +performance and should not be considered in isolation from, or +as a substitute for, similar financial measures calculated in +conformity with U.S. GAAP financial measures. These non- +GAAP +financial measures are not standardized financial +measures and may not be comparable to similar financial +issuers. Non-GAAP measures +measures used by other + +include, but are not limited to: (i) distributable earnings +(“Distributable Earnings”), (ii) fee revenues (“Fee Revenues”) +and (iii) fee-related earnings (“Fee Related Earnings” or +“FRE”). These non-GAAP measures are not standardized +financial measures and may not be comparable to similar +financial measures used by other issuers. + +Supplemental financial measures + +financial measures +(“AUM”), + +include assets under +Supplemental +management +(“Fee-Bearing +Capital”) and uncalled fund commitments (“Uncalled Fund +Commitments”). + +fee-bearing capital + +Securities Law + +This presentation does not constitute an offer to sell or a +solicitation of an offer to buy any securities of Brookfield Asset +Management Ltd. + +37 diff --git a/q192/random_k4/random_3.md b/q192/random_k4/random_3.md index 82b82c26456cb0f85a7ad988e31ecf2fa7ec88b3..f28bcc1c1e50866e33c7f4488c244d56c08f6e52 100644 --- a/q192/random_k4/random_3.md +++ b/q192/random_k4/random_3.md @@ -1,39 +1,452 @@ -E17889 -First Edition / June 2021 +G R M I N G M O N I T O R -User Guide +U S E R G U I D E -MyASUS FAQ +-1- - Top View +LCD Monitor User's Manual Copyright -NOTE: The keyboard's layout may vary per region or country. The top view may also -vary in appearance depending on the Notebook PC model. +© 2020 GIGA-BYTE TECHNOLOGY CO., LTD. All rights reserved. -Array microphones +The trademarks mentioned in this manual are legally registered to their +respective owners. -Camera indicator +Disclaimer -Camera/Webcam shield -(built-in) +Information in this manual is protected by copyright laws and is the +property of GIGABYTE. +Changes to the specifications and features in this manual may be made +by GIGABYTE without prior notice. +No part of this manual may be reproduced, copied, translated, +transmitted, or published in any form or by any means without +GIGABYTE’s prior written permission. -360º-adjustable -touch screen panel +• -Power button/ -Fingerprint sensor +• -Keyboard + In order to assist in the use of this monitor, carefully read the User +Guide. + For more information, check on our website at: https://www. +gigabyte.com + +-2- + +LCD Monitor User's Manual SafETy INfORMaTION + +Regulatory Notices + +United States of america, federal Communications +Commission Statement + +Supplier’s Declaration of Conformity +47 CfR § 2.1077 Compliance Information + +Product Name: LCD monitor +Trade Name: GIGABYTE +Model Number: G27F + +Responsible Party – U.S. Contact Information: G.B.T. Inc. +Address: 17358 Railroad street, City Of Industry, CA91748 +Tel.: 1-626-854-9338 +Internet contact information: https://www.gigabyte.com + +fCC Compliance Statement: +This device complies with Part 15 of the FCC Rules, Subpart B, Unin- +tentional Radiators. +Operation is subject to the following two conditions: (1) This device +may not cause harmful interference, and (2) this device must accept +any interference received, including interference that may cause un- +desired operation. + +This equipment has been tested and found to comply with the limits +for a Class B digital device, pursuant to Part 15 of the FCC Rules. These +limits are designed to provide reasonable protection against harmful +interference in a residential installation. This equipment generates, uses +and can radiate radio frequency energy and, if not installed and used in +accordance with manufacturer’s instructions, may cause harmful inter- +ference to radio communications. However, there is no guarantee that +interference will not occur in a particular installation. If this equipment +does cause harmful interference to radio or television reception, which +can be determined by turning the equipment o and on, the user is en- +couraged to try to correct the interference by one or more of the follow- +ing measures: +• Reorient or relocate the receiving antenna. +• + +Increase the separation between the equipment and receiver. -Touchpad/NumberPad +-3- -Function key lock indicator +LCD Monitor User's Manual • -Capital lock indicator + Connect the equipment to an outlet on a circuit different from that +to which the receiver is connected. -2 +• Consult the dealer or an experienced radio/TV technician for help. + +Canadian Department of Communications Statement +This digital apparatus does not exceed the Class B limits for radio noise +emissions from digital apparatus set out in the Radio Interference +Regulations of the Canadian Department of Communications. This class +B digital apparatus complies with Canadian ICES-003. + +avis de conformité à la réglementation d’ Industrie Canada +Cet appareil numérique de la classe B est conforme à la norme NMB-003 +du Canada. + +European Union (EU) CE Declaration of Conformity +This device complies with the following directives: Electromagnetic +Compatibility Directive 2014/30/EU, Low-voltage Directive 2014/35/ +EU, ErP Directive 2009/125/EC, RoHS directive (recast) 2011/65/EU & the +2015/863 Statement. This product has been tested and found to comply +with all essential requirements of the Directives. + +European Union (EU) RoHS (recast) Directive 2011/65/EU & +the European Commission Delegated Directive (EU) 2015/863 +Statement +GIGABYTE products have not intended to add and safe from hazardous +substances (Cd, Pb, Hg, Cr+6, PBDE, PBB, DEHP, BBP, DBP and DIBP). +The parts and components have been carefully selected to meet RoHS +requirement. Moreover, we at GIGABYTE are continuing our efforts +to develop products that do not use internationally banned toxic +chemicals. + +European Union (EU) Community Waste Electrical & Electronic +Equipment (WEEE) Directive Statement +GIGABYTE will fulfill the national laws as interpreted from the 2012/19/ +EU WEEE (Waste Electrical and Electronic Equipment) (recast) directive. +The WEEE Directive specifies the treatment, collection, recycling and +disposal of electric and electronic devices and their components. Under +the Directive, used equipment must be marked, collected separately, +and disposed of properly. + +-4- + +LCD Monitor User's Manual WEEE Symbol Statement + +The symbol shown below is on the product or on its pack- +aging, which indicates that this product must not be dis- +posed of with other waste. Instead, the device should be +taken to the waste collection centers for activation of the +treatment, collection, recycling and disposal procedure. + +For more information about where you can drop off your waste equip- +ment for recycling, please contact your local government office, your +household waste disposal service or where you purchased the product +for details of environmentally safe recycling. + +End of Life Directives-Recycling + +The symbol shown below is on the product or on its +packaging, which indicates that this product must not be +disposed of with other waste. Instead, the device should be +taken to the waste collection centers for activation of the +treatment, collection, recycling and disposal procedure. + +Déclaration de Conformité aux Directives de l’Union +européenne (UE) +Cet appareil portant la marque CE est conforme aux directives de l’UE +suivantes: directive Compatibilité Electromagnétique 2014/30/UE, +directive Basse Tension 2014/35/UE, directive 2009/125/CE en matière +d’écoconception, la directive RoHS II 2011/65/UE & la déclaration +2015/863. +La conformité à ces directives est évaluée sur la base des normes +européennes harmonisées applicables. + +European Union (EU) CE-Konformitätserklärung +Dieses Produkte mit CE-Kennzeichnung erfüllen folgenden +EURichtlinien: +EMV-Richtlinie 2014/30/EU, Niederspannungsrichtlinie +2014/30/EU, Ökodesign-Richtlinie 2009/125/EC, RoHS-Richtlinie +2011/65/EU erfüllt und die 2015/863 Erklärung. +Die Konformität mit diesen Richtlinien wird unter Verwendung der +entsprechenden Standards zurEuropäischen Normierung beurteilt. + +-5- + +LCD Monitor User's Manual CE declaração de conformidade +Este produto com a marcação CE estão em conformidade com das +seguintes Diretivas UE: Diretiva Baixa Tensão 2014/35/EU; Diretiva CEM +2014/30/EU; Diretiva Conceção Ecológica 2009/125/CE; Diretiva RSP +2011/65/UE e a declaração 2015/863. +A conformidade com estas diretivas é verificada utilizando as normas +europeias harmonizadas. + +CE Declaración de conformidad +Este producto que llevan la marca CE cumplen con las siguientes +Directivas de la Unión Europea: Directiva EMC (2014/30/EU), Directiva +de bajo voltaje (2014/35/EU), Directiva de Ecodiseño (2009/125/EC), +Directiva RoHS (recast) (2011/65/EU) y la Declaración 2015/863. +El cumplimiento de estas directivas se evalúa mediante las normas +europeas armonizadas. + +Japan Class 01 Equipment Statement + +安全にご使用いただくために +接地接続は必ず電源プラグを電源につなぐ前に行って下さい。 +また、接地接続を外す場合は、必ず電源プラグを電源から切り離してから +行って下さい。 + +Japan VCCI Class B Statement + +クラス B VCCI 基準について + +この装置は、クラス B 情報技術装置です。 この装置は、家庭環境で使 +用することを目的としていますが、この装置がラジオやテレビジョン +受信機に近接して使用されると、受信障害を引き起こすことがありま +す。 +取扱説明書に従って正しい取り扱いをして下さい。 + +VCCI-B + +-6- + +LCD Monitor User's Manual China RoHS Compliance Statement +中国《废弃电器电子产品回收处理管理条例》提示性说明 +为了更好地关爱及保护地球，当用户不再需要此产品或产品寿命终 +止时，请遵守国家废弃电器电子产品回收处理相关法律法规，将其 +交给当地具有国家认可的回收处理资质的厂商进行回收处理。 + +环保使用期限 +Environment-friendly Use Period + +此标识指期限（十年），电子电气产品中含有的有害物质 +不会发生外泄或突变、电子电气产品用户正常使用该电子 +电气产品不会对环境造成严重 污染或对其人身、财产造 +成严重损害的期限。 + +产品中有害物质的名称及含量 : + +部件名称 + +铅 +(Pb) + +汞 +(Hg) + +镉 +(Cd) + +六价铬 +(Cr+6) + +多溴联苯 +(PBB) + +多溴二苯醚 +(PBDE) + +有害物质 + +印刷电路板及其电 +子组件 + +液晶面板 + +外壳 + +外部信号连接头及 +线材 + +X ○ ○ + +X ○ ○ + +○ ○ ○ + +X ○ ○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +本表格依据 SJ/T 11364 的规定编制。 +○ : 表示该有害物质在该部件所有均质材料中的含量均在 GB/T 26572 规 + +X: + +定的限量要求以下。 + 表示该有害物质至少在该部件的某一均质材料中的含量超出 GB/T +26572 规定的限量要求。 + +HDMI HDMI 高清晰度多媒体接口以及 HDMI 标志是 HDMI Licensing +Administrator, Inc. 在美国和其他国家的商标或注册商标。 + +-7- + +LCD Monitor User's Manual 限用物質含有情況標示聲明書 +Declaration of the Presence Condition of the Restricted Substances +Marking + +設備名稱 : +Equipment name: + +液晶顯示器型號 ( 型式 ): G27F +Typedesignation (Tape): -Notebook PC User Guide Using the NumberPad (on selected models) +限用物質及其化學符號 +Restricted substances and its chemical symbols + +單元 +Un it + +铅 +Lead +(Pb) + +汞 Mer- +cury +(Hg) + +镉 +Cadmium +(Cd) + +六价铬 +Hexavalent +chromium +(Cr+6) + +多溴联苯 Poly- +brominated +biphenyls (PBB) + +多溴二苯醚 Poly- +brominated +diphenylethers +(PBDE) + +塑料外框 + +○ ○ + +後殼 + +○ ○ + +液晶面板 + +－ ○ + +電路板組件 － ○ + +底座 + +電源線 + +○ ○ + +－ ○ + +其他線材 + +－ ○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +○ + +備考 1.〝超出 0.1 wt %〞及〝超出 0.01 wt %〞係指限用物質之百分比含量超出百 + +分比含量基準值。 + +Note 1: 〝Exceeding 0.1 wt %〞and〝exceeding 0.01 wt %〞 indicate that the per- + +centage content of the restricted substance exceeds the reference percentage +value of presence condition . + +備考 2.〝 ○ 〞係指該項限用物質之百分比含量未超出百分比含量基準值。 + +Note 2: 〝○ 〞indicates that the percentage content of the restricted substance does + +not exceed the percentage of reference value of presence . + +備考 3.〝－〞係指該項限用物質為排除項目。 + +Note 3: 〝－〞indicates that the restricted substance corresponds to the exemption + +-8- + +LCD Monitor User's Manual Safety Precautions + +Read through the following Safety Precautions before using the +monitor. + +• Only use the accessories that are provided with the monitor or that • @@ -41,618 +454,1548 @@ Notebook PC User Guide Using the NumberPad (on selected models) • -Press and hold +which are recommended by the manufacturer. +Keep the plastic packaging bag for the product in a place that +cannot be reached by children. +Before connecting the monitor to the power outlet, make sure that +the voltage rating of the power cable is compatible with the power +specification in the country where you are located. +The power cord plug must be connected to a properly wired and +grounded power outlet. - for a second to toggle between number keys or touchpad. +• Do not touch the plug with wet hands, otherwise easily cause -Tap +electric shock. +Place the monitor in a stable and well-ventilated place. - to adjust brightness for the NumberPad. +• +• Do not place the monitor near any heat sources such as electric -Press and hold +• - then slide anywhere on the NumberPad to launch Windows +radiators or direct sunlight. +The holes or openings on the monitor are for ventilation. Do not +cover or block the ventilation holes with any objects. -Calculator. +• Do not use the monitor near water, drinks, or all types of liquids. +Failure to do so may result in electric shock or damage to the +monitor. -IMPORTANT! Please note that all versions of NumberPad are only supported on the -latest Microsoft Windows® operating system. +• Make sure to unplug the monitor from the power outlet before -NOTE: To use the % and = function, set the input language to English. +cleaning. -3 +• As the screen surface is easy to be scratched, avoid touching the -Notebook PC User Guide I/O ports and slots +surface with any hard or sharp object. -MicroSD card slot +• Use a soft lint-free cloth instead of a tissue to wipe the screen. You +may use a glass cleaner to clean the monitor if required. However, +never spray the cleaner directly onto the screen. -HDMI output port +• Disconnect the power cable if the monitor is not being used for a -Headphone/Headset/Microphone jack +long period of time. -Thunderbolt™ 4 port with -Power Delivery +• Do not attempt to disassemble or repair the monitor yourself. -USB 3.2 Gen 2 port +-9- -IMPORTANT! -To prevent any damage, use only power sources rated 20V/5A to charge your Notebook PC with -the USB Power Delivery combo port. For more information, consult an ASUS service center for -assistance. +LCD Monitor User's Manual CONTENTS -ASUS Pen (on selected models) +Safety Information ..................................................................................................................... 3 +Regulatory Notices .......................................................................................................... 3 +Safety Precautions ........................................................................................................... 9 -Right-click button +INTRODUCTION ........................................................................................................................11 +Unpacking ........................................................................................................................11 +Package Contents ..........................................................................................................13 +Product Overview ..........................................................................................................14 -Eraser button +GETTING STARTED ...................................................................................................................16 +Installing the Monitor Base ........................................................................................16 +Adjusting the Viewing Angle .....................................................................................17 +Installing a Wall-Mount Bracket (Optional) ..........................................................18 +Making Connections ....................................................................................................19 + +USING THE DEVICE ...................................................................................................................20 +Turning the Power On/Off ..........................................................................................20 +User Comfort Recommendations ............................................................................21 +Selecting the Input Source .........................................................................................22 + +OPERATIONS ..............................................................................................................................23 +Quick Menu ......................................................................................................................23 +Configure the Device Settings ..................................................................................28 + +APPENDICES ..............................................................................................................................37 +Specifications ..................................................................................................................37 +Supported Timing List..................................................................................................38 +Troubleshooting .............................................................................................................39 +Basic Care ..........................................................................................................................40 +Notes on USB Charging ...............................................................................................40 +GIGABYTE Service Information .................................................................................40 + +-10- + +LCD Monitor User's Manual INTRODUCTION + +Unpacking + +1. Open the upper cover of the packaging box. Then observe the + +unpacking label attached on the inner cover. + +2. Making sure the box is in the correct orientation, carefully lay the + +box on the stable surface. + +UP + +3. Open the inner cover. + +4. Grab the EPS foam assembly (through the openings) to pull it out of + +the box. + +Opening + +-11- + +LCD Monitor User's Manual 5. Remove the items from the upper EPS foam. + +6. Remove the upper EPS foam. Then you can remove the monitor + +from the lower EPS foam. + +-12- + +LCD Monitor User's Manual Package Contents + +The following items come with your packaging box. If any of them is +missing, please contact your local dealer. + +Monitor + +Monitor Base + +Stand + +Power Cable + +HDMI Cable + +USB Cable + +DP Cable + +Adapter + +Quick Start Guide + +Multilingual Installation Guide + +G27F + +Read this manual carefully before you use this machine and keep it handy for future +reference. For safe and correct use, be sure to read the Safety Precautions in the User’s +Manual before using the machine. + +Warranty Card + +Warranty Card + +Note: Please keep the packaging box and packing materials for future +transportation of the monitor. + +-13- + +LCD Monitor User's Manual Product Overview + +front View + +Power LED + +-14- + +LCD Monitor User's Manual Rear View + +7 + +8 + +1 + +2 + +3 4 -Notebook PC User Guide Getting started +5 + +6 + +1. USB upstream port +2. USB 3.0 ports (x2) +3. DC IN jack +4. HDMI ports (x2) + +5. DisplayPort +6. Headphone jack +7. Control button +8. Kensington lock + +-15- + +LCD Monitor User's Manual GETTING STaRTED +Installing the Monitor Base +1. Place the monitor on the lower EPS foam, with the screen facing + +down. + +2. Align the monitor base with the two guide pins underneath the + +stand. Then install the base onto the stand. + +3. Lift the monitor to an upright position and place it on a table. The + +monitor's height may be adjusted after item 1 and item 2 shown in +the figure are removed. -IMPORTANT! -Do not use this Notebook PC for cryptocurrency mining (consuming a vast amount of electricity -and time to gain convertible virtual currency) and/or related activities. +2 + +1 + +-16- + +LCD Monitor User's Manual adjusting the Viewing angle +Note: Hold the monitor base so that the screen does not topple when +you make the adjustment. + +adjusting the Tilt angle +Tilt the screen forward or backward to the desired viewing angle (-5° to +20°). + +20° + +5° + +adjusting the Screen Orientation +Raise the screen to the maximum height (130mm). + +130 mm + +-17- + +LCD Monitor User's Manual Installing a Wall-Mount Bracket (Optional) +Use only the 100 x 100 mm wall mount kit recommended by the manu- +facturer. +1. Place the monitor on the lower EPS foam, with the screen facing + +down. + +2. Push the locking clips inward to detach the stand from the monitor. + +3. Attach the mounting bracket to the mounting holes at the rear of +the monitor. Then use the screws to secure the bracket in place. + +100 mm + +100 mm -1. Charge your Notebook PC +Note: -A. Connect the DC power connector into your Notebook PC’s power (DC) input port. + To mount the monitor on the wall, refer to the installation guide that is +included in the wall-mount bracket kit. -B. Plug the AC power adapter into a 100V~240V power source. + The monitor head may be installed on wall, rotary brackets or other brackets. +This monitor supports 100 mm mounting holes that comply with VESA indus- +trial standard. To install any third-party solution to the monitor, four φ4 mm +screws with pitch of 0.7 mm and length of 10 mm are required. If any longer +screw is used, the monitor may be damaged. Check if the bracket provided +by the manufacturer complies with VESA standard and is able to support the +monitor body. Please use the power cable and video cable attached to the +monitor, in order to ensure the optimal performance. -IMPORTANT! Use only the bundled power adapter to charge the battery pack and -supply power to your Notebook PC. +-18- -NOTE: The power adapter may vary in appearance, depending on models and your -region. +LCD Monitor User's Manual + Making Connections -2. Lift to open the display panel +21 -3. Press the power button +Headphone -Charge the Notebook PC -for 3 hours before using -it in battery mode for the -first time. +2 + +3 + +4 + +Computer 5 -Notebook PC User Guide +1 Headphone cable +2 HDMI cable +3 DisplayPort cable - Safety notices for your Notebook PC -WARNING! -Your Notebook PC can get warm to hot while in use or while charging the battery pack. Do not -leave your Notebook PC on your lap or near any part of your body to prevent injury from heat. -When working on your Notebook PC, do not place it on surfaces that can block the vents. +Keyboard -CAUTION! -• This Notebook PC should only be used in environments with ambient temperatures between +Mouse -5°C (41°F) and 35°C (95°F). +Computer -• Refer to the rating label on the bottom of your Notebook PC and ensure that your power +USB +disk +drive -adapter complies with this rating. +4 USB (A-Male to B-Male) cable +5 USB cable -• The power adapter may become warm to hot while in use. Do not cover the adapter and keep +-19- -it away from your body while it is connected to a power source. +LCD Monitor User's Manual USING THE DEVICE +Turning the Power On/Off -IMPORTANT! -• Ensure that your Notebook PC is connected to the power adapter before turning it on for the -first time. Always plug the power cord into a wall socket without using any extension cords. -For your safety, connect this device to a properly grounded electrical outlet only. +Power On -• When using your Notebook PC on power adapter mode, the socket outlet must be near to the +1. Connect one end of the power cable to the adapter and connect +the output cable of the adapter to the DC IN jack of the monitor. -unit and easily accessible. +2. Connect other end of the the power cable to a power outlet. +3. Press the Control button to turn on the monitor. The Power LED will -• Locate the input/output rating label on your Notebook PC and ensure that it matches the -input/output rating information on your power adapter. Some Notebook PC models may -have multiple rating output currents based on the available SKU. +light in white, indicating the monitor is ready to use. -• Power adapter information: +Power Off +Press the Control button for 2 seconds to turn off the monitor. +Note: When the Main menu appears on the screen, you can also turn off +). Refer to page 25. +the monitor by moving the Control button down ( -- Input voltage: 100-240Vac +-20- -- Input frequency: 50-60Hz +LCD Monitor User's Manual User Comfort Recommendations -- Rating output current: 5A (100W) +The following are some tips for comfortable viewing of your monitor: +• Optimal viewing distances for monitors range from approximately -- Rating output voltage: 20V +510 mm to 760 mm (20” to 30”). -WARNING! -Read the following precautions for your Notebook PC’s battery: +• A general guideline is to position the monitor such that the top +of the screen is at or slightly below your eye-height when you are +comfortably seated. -• Only ASUS-authorized technicians should -remove the battery inside the device -(for non-removable battery only). +• Use adequate lighting for the type of work you are performing. +• -• The battery used in this device may present -a risk of fire or chemical burn if removed or -disassembled. +Take regular and frequent breaks (at least for 10 minutes) every half +hour. +Be sure to periodically look away from your monitor screen and +focus on a far object for at least 20 seconds during the breaks. +Eye exercise can help reduce eye strain. Repeat these exercises +frequently: (1) look up and down (2) slowly roll your eye (3) move +your eyes diagonally. -• Follow the warning labels for your personal +• -safety. +• -• Risk of explosion if battery is replaced by an +-21- -incorrect type. +LCD Monitor User's Manual Selecting the Input Source -• Do not dispose of in fire. +1. Move the Control button right +) to enter the the Input -• Never attempt to short-circuit your +( +menu. -Notebook PC’s battery. +2. Move the Control button up/ +) to select the desired +down ( +input source. Then press the +Control button to confirm. -• Never attempt to disassemble and +Input -reassemble the battery -(for non-removable battery only). +Select -• Discontinue usage if leakage is found. +Exit -• This battery and its components must be +-22- -recycled or disposed of properly. +LCD Monitor User's Manual OPERaTIONS +Quick Menu -• Keep the battery and other small -components away from children. +Hot Key +By default, the Control button has been assigned with the specific +function. +Note: To change the preset hot key function, refer to the “Quick Switch” section +on page 34. -6 +To access the hot key function, do the following: +• Move the Control button up -Notebook PC User Guide +) to enter the Black Equalizer - Avis concernant les batteries remplaçables +( +menu. + Move the Control button +) to adjust +up/down ( +the setting and press the +Control button to confirm. -• La batterie de l’appareil peut présenter un risque d’incendie ou de brûlure si celle-ci est +Black Equalizer -retirée ou désassemblée. +adjust -• La batterie et ses composants doivent être recyclés de façon appropriée. +Exit -Copyright Information -You acknowledge that all rights of this Manual remain with ASUS. Any and all rights, including without -limitation, in the Manual or website, are and shall remain the exclusive property of ASUS and/or its -licensors. Nothing in this Manual intends to transfer any such rights, or to vest any such rights to you. +-23- -ASUS PROVIDES THIS MANUAL “AS IS” WITHOUT WARRANTY OF ANY KIND. SPECIFICATIONS AND -INFORMATION CONTAINED IN THIS MANUAL ARE FURNISHED FOR INFORMATIONAL USE ONLY, AND -ARE SUBJECT TO CHANGE AT ANY TIME WITHOUT NOTICE, AND SHOULD NOT BE CONSTRUED AS A -COMMITMENT BY ASUS. +LCD Monitor User's Manual + • Move the Control button down ( -Copyright © 2021 ASUSTeK COMPUTER INC. All Rights Reserved. +) to enter the Picture Mode -Limitation of Liability -Circumstances may arise where because of a default on ASUS’ part or other liability, you are entitled to -recover damages from ASUS. In each such instance, regardless of the basis on which you are entitled to -claim damages from ASUS, ASUS is liable for no more than damages for bodily injury (including death) -and damage to real property and tangible personal property; or any other actual and direct damages -resulted from omission or failure of performing legal duties under this Warranty Statement, up to the -listed contract price of each product. +menu. + Move the Control button + ) to select the +up/down ( +desired option and press +the Control button to con- +firm. -ASUS will only be responsible for or indemnify you for loss, damages or claims based in contract, tort or -infringement under this Warranty Statement. +Picture Mode -This limit also applies to ASUS’ suppliers and its reseller. It is the maximum for which ASUS, its suppliers, -and your reseller are collectively responsible. +Standard -UNDER NO CIRCUMSTANCES IS ASUS LIABLE FOR ANY OF THE FOLLOWING: (1) THIRD-PARTY CLAIMS -AGAINST YOU FOR DAMAGES; (2) LOSS OF, OR DAMAGE TO, YOUR RECORDS OR DATA; OR (3) SPECIAL, -INCIDENTAL, OR INDIRECT DAMAGES OR FOR ANY ECONOMIC CONSEQUENTIAL DAMAGES (INCLUDING -LOST PROFITS OR SAVINGS), EVEN IF ASUS, ITS SUPPLIERS OR YOUR RESELLER IS INFORMED OF THEIR -POSSIBILITY. +fPS -Service and Support -For complete E-Manual version, refer to our multi-language website at: -https://www.asus.com/support/ +RTS/RPG -MyASUS offers a variety of support features including troubleshooting, products performance -optimization, ASUS software integration, and helps you to organize personal desktop and increase -storage space. For more details, please visit https://www.asus.com/support/FAQ/1038301/. +Movie -7 +Reader -Notebook PC User Guide FCC Radio Frequency (RF) Exposure Caution -Statement +sRGB -WARNING! Any changes or modifications not expressly approved by the party -responsible for compliance could void the user’s authority to operate this equipment. +Select -This equipment complies with FCC radiation exposure limits set forth for an uncontrolled -environment. To maintain compliance with FCC RF exposure compliance requirements, please -avoid direct contact to the transmitting antenna during transmitting. End users must follow the -specific operating instructions for satisfying RF exposure compliance. +Exit -UL Safety Notices +• Move the Control button left -• DO NOT use the Notebook PC near water, for example, near a bath tub, wash bowl, kitchen +( -sink or laundry tub, in a wet basement or near a swimming pool. + ) to enter the Volume menu. -• DO NOT use the Notebook PC during an electrical storm. There may be a remote risk of +Volume -electric shock from lightning. + Move the Control button +up/down ( +) to adjust +the headphone volume +level and press the Control +button to confirm. -• DO NOT use the Notebook PC in the vicinity of a gas leak. +adjust -• DO NOT dispose the Notebook PC battery pack in a fire, as they may explode. Check with local -codes for possible special disposal instructions to reduce the risk of injury to persons due to -fire or explosion. +Exit -• DO NOT use power adapters or batteries from other devices to reduce the risk of injury to +• Move the Control button right ( -persons due to fire or explosion. Use only UL certified power adapters or batteries supplied -by the manufacturer or authorized retailers. +) to enter the the Input menu. -Coating Notice +Refer to the “Selecting the Input Source” section. -IMPORTANT! To provide electrical insulation and maintain electrical safety, a coating -is applied to insulate the device except on the areas where the I/O ports are located. +Note: To close the menu, move the Control button left ( -Prevention of Hearing Loss -To prevent possible hearing damage, do not listen at high volume levels for long -periods. + ). -Power Safety Requirement -Products with electrical current ratings up to 6A and weighing more than 3Kg must use -approved power cords greater than or equal to: H05VV-F, 3G, 0.75mm2 or H05VV-F, 2G, 0.75mm2. +-24- -8 +LCD Monitor User's Manual -Notebook PC User Guide Declaration of Compliance for Product Environmental -Regulation -ASUS follows the green design concept to design and manufacture our products, and -makes sure that each stage of the product life cycle of ASUS product is in line with global -environmental regulations. In addition, ASUS disclose the relevant information based on -regulation requirements. Please refer to http://csr.asus.com/Compliance.htm for information -disclosure based on regulation requirements ASUS is complied with. - -EU REACH and Article 33 -Complying with the REACH (Registration, Evaluation, Authorization, and Restriction of -Chemicals) regulatory framework, we publish the chemical substances in our products at ASUS -REACH website at http://csr.asus.com/english/REACH.htm. - -EU RoHS -This product complies with the EU RoHS Directive. For more details, see -http://csr.asus.com/english/article.aspx?id=35. - -Japan JIS-C-0950 Material Declarations -Information on Japan RoHS (JIS-C-0950) chemical disclosures is available on -http://csr.asus.com/english/article.aspx?id=19. - -India RoHS -This product complies with the “India E-Waste (Management) Rules, 2016” and prohibits use of -lead, mercury, hexavalent chromium, polybrominated biphenyls (PBBs) and polybrominated -diphenyl ethers (PBDEs) in concentrations exceeding 0.1% by weight in homogenous materials -and 0.01% by weight in homogenous materials for cadmium, except for the exemptions listed -in Schedule II of the Rule. - -Vietnam RoHS -ASUS products sold in Vietnam, on or after September 23, 2011, meet the requirements of the -Vietnam Circular 30/2011/TT-BCT. - -Các sản phẩm ASUS bán tại Việt Nam, vào ngày 23 tháng 9 năm2011 trở về sau, đều phải đáp -ứng các yêu cầu của Thông tư 30/2011/TT-BCT của Việt Nam. - -9 - -Notebook PC User Guide ASUS Recycling/Takeback Services -ASUS recycling and takeback programs come from our commitment to the highest standards for -protecting our environment. We believe in providing solutions for you to be able to responsibly -recycle our products, batteries, other components as well as the packaging materials. Please -go to http://csr.asus.com/english/Takeback.htm for detailed recycling information in different -regions. - -Ecodesign Directive -European Union announced a framework for the setting of ecodesign requirements for energy- -related products (2009/125/EC). Specific Implementing Measures are aimed at improving -environmental performance of specific products or across multiple product types. ASUS -provides product information on the CSR website. Further information could be found at -https://csr.asus.com/english/article.aspx?id=1555. - -EPEAT Registered Products -The public disclosure of key environmental information for ASUS EPEAT (Electronic Product -Environmental Assessment Tool) registered products is available at -https://csr.asus.com/english/article.aspx?id=41. More information about EPEAT program and -purchase guidance can be found at www.epeat.net. - -Regional notice for Singapore - -Complies with -IMDA Standards -DB103778 - -This ASUS product complies with IMDA Standards. - -RF Exposure Information (SAR) -This device meets the government’s requirements for exposure to radio waves. This device is -designed and manufactured not to exceed the emission limits for exposure to radio frequency -(RF) energy set by the Federal Communications Commission of the U.S. Government. - -The exposure standard for wireless device employs a unit of measurement known as the Specific -Absorption Rate, or SAR. The SAR limit set by the FCC is 1.6 W/kg. Tests for SAR are conducted -using standard operating positions accepted by the FCC with the device transmitting at its -highest certified power level in all tested frequency bands. + function Key Guide -10 +Press the Control button to display the Main menu. -Notebook PC User Guide Simplified EU Declaration of Conformity -ASUSTek Computer Inc. hereby declares that this device is in compliance with the essential -requirements and other relevant provisions of Directive 2014/53/EU. Full text of EU declaration -of conformity is available at https://www.asus.com/support/. - -The WiFi operating in the band 5150-5350 MHz shall be restricted to indoor use for countries -listed in the table below: - -AT -DE -LV -NO -FI - -BE -IS -LI -PL -SE - -BG -IE -LT -PT -CH - -CZ -IT -LU -RO -HR - -DK -EL -HU -SI -UK(NI) - -EE -ES -MT -SK - -FR -CY -NL -TR - -Simplified UKCA Declaration of Conformity -ASUSTek Computer Inc. hereby declares that this device is in compliance with the essential -requirements and other relevant provisions of The Radio Equipment Regulations 2017 (S.I. -2017/1206). Full text of UKCA declaration of conformity is available at -https://www.asus.com/support/. - -The WiFi operating in the band 5150-5350 MHz shall be restricted to indoor use for the country -listed below: - -11 - -Notebook PC User Guide RF Exposure information (SAR) -This device meets the EU requirements (2014/53/EU) on the limitation of exposure of the -general public to electromagnetic fields by way of health protection. - -The limits are part of extensive recommendations for the protection of the general public. These -recommendations have been developed and checked by independent scientific organizations -through regular and thorough evaluations of scientific studies. The unit of measurement for the -European Council’s recommended limit for mobile devices is the “Specific Absorption Rate” (SAR), -and the SAR limit is 2.0 W/Kg averaged over 10 gram of body tissue. It meets the requirements -of the International Commission on Non-Ionizing Radiation Protection (ICNIRP). - -For next-to-body operation, this device has been tested and meets the ICNIRP exposure -guidelines and the European Standard EN 50566 and EN 62209-2. SAR is measured with the -device directly contacted to the body while transmitting at the highest certified output power -level in all frequency bands of the mobile device. - -A minimum separation distance of 1 cm must be maintained between the user’s body and the -device, including the antenna during body-worn operation to comply with the RF exposure -requirements in Europe. - -12 - -Notebook PC User Guide Federal Communications Commission Interference -Statement -This equipment has been tested and found to comply with the limits for a Class B digital -device, pursuant to part 15 of the FCC Rules. These limits are designed to provide reasonable -protection against harmful interference in a residential installation. This equipment generates, -uses and can radiate radio frequency energy and, if not installed and used in accordance with -the instructions, may cause harmful interference to radio communications. However, there is -no guarantee that interference will not occur in a particular installation. If this equipment does -cause harmful interference to radio or television reception, which can be determined by turning -the equipment off and on, the user is encouraged to try to correct the interference by one or -more of the following measures: +Settings -• Reorient or relocate the receiving antenna. +Dashboard + +GameAssist + +Exit + +Power Off + +Then use the Control button to select the desired function and +configure the related settings. + +Dashboard +When the Main menu appears on the screen, move the Control button +left ( + +) to enter the Dashboard menu. + +OFF + +Dashboard + +function + +ON + +Dashboard +Location + +Off + +Select + +Exit + +CPU Temperature + +CPU frequency + +CPU fan Speed + +CPU Usage Rate + +GPU Temperature + +GPU frequency + +GPU fan Speed • -Increase the separation between the equipment and receiver. +Function: Enable/Disable the function. When enabled, you can +select the desired option(s) to be shown on the screen. -• Connect the equipment into an outlet on a circuit different from that to which the receiver is +• Dashboard Location: Specify the location of the Dashboard -connected. +information to be shown on the screen. -• Consult the dealer or an experienced radio/TV technician for help. +-25- - FCC COMPLIANCE INFORMATION +LCD Monitor User's Manual Once the Dashboard configuration is complete, it will transfer the +system data to the SOC through the USB port and display the value of +the selected function(s) on the screen. -Per FCC Part 2 Section 2.1077 +CPU Temperature -Responsible Party: Asus Computer International +CPU frequency -Address: +CPU fan Speed -48720 Kato Rd., Fremont, CA 94538 +CPU Usage Rate -Phone/Fax No: +GPU Temperature -(510)739-3777/(510)608-4555 +GPU frequency -hereby declares that the product +GPU fan Speed -Product Name : ZenBook +GPU Usage Rate -Model Number : UP5401E, RP5401E +aORUS Mouse DPI -compliance statement: +OFF -This device complies with part 15 of the FCC Rules. Operation is subject to the -following two conditions: (1) This device may not cause harmful interference, -and (2) this device must accept any interference received, including interference -that may cause undesired operation. +Dashboard - Ver. 180620 +function -13 +ON -Notebook PC User Guide +Dashboard +Location - CE RED RF Output table (Directive 2014/53/EU) +Off -MT7921 +Select -Function +Exit -WiFi +CPU Temperature -Bluetooth +CPU frequency -* Receiver category 1 +CPU fan Speed -Intel AX201NGW +CPU Usage Rate -Function +GPU Temperature -WiFi +GPU frequency -Bluetooth +GPU fan Speed -* Receiver category 2 +Note: Make sure the USB cable is properly connected to the USB upstream port +of the monitor and the USB port of your computer +Gameassist +When the Main menu appears on the screen, move the Control button +right ( -Frequency -2412 - 2472 MHz +) to enter the GameAssist menu. -5150 - 5350 MHz +Gameassist -5470 - 5725 MHz +GaME INfO -5725 - 5850 MHz +Gaming Timer -2402 - 2480 MHz +Gaming Counter -Frequency -2412 - 2472 MHz +Crosshair -5150 - 5350 MHz +Refresh Rate -5470 - 5725 MHz +Info Location -5725 - 5850 MHz +Display alignment -2402 - 2480 MHz +Select -Maximum Output Power (EIRP) -19 dBm +Exit -22 dBm +-26- -20 dBm +LCD Monitor User's Manual • Game Info: Configure the FPS (First Person Shooting) game related -13.97 dBm +settings. +√ Select Gaming Timer to select countdown timer mode. To -14 dBm +disable this function, set the setting to OFF. -Maximum Output Power (EIRP) -19 dBm +√ Select Gaming Counter or Refresh Rate to enable/disable the -22 dBm +gaming counter or real-time refresh rate setting. -23 dBm +√ Select Info Location to specify the location of the information to -13.97 dBm +be shown on the screen. -13 dBm +Gameassist -14 +GaME INfO -Notebook PC User Guide UKCA RF Output table -(The Radio Equipment Regulations 2017) +Gaming Timer -MT7921 +Gaming Counter -Function +Crosshair -WiFi +Refresh Rate -Bluetooth +Info Location -* Receiver category 1 +Display alignment -Intel AX201NGW +Select -Function +Exit -WiFi +• Crosshair: Select the desired crosshair type to suit your gaming -Bluetooth +environment. It makes aiming much easier. -* Receiver category 2 +• Display alignment: When the function is enabled, it displays the +alignment lines on the four sides of the screen, providing handy +tool for you to line up multiple monitors perfectly. -Frequency -2412 - 2472 MHz +-27- -5150 - 5350 MHz +LCD Monitor User's Manual Configure the Device Settings -5470 - 5725 MHz +When the Main menu appears on the screen, move the Control button +up ( -5725 - 5850 MHz + ) to enter the Settings menu. -2402 - 2480 MHz +Note: Use the Control button to navigate through the menu and make +adjustments. -Frequency -2412 - 2472 MHz +Exit -5150 - 5350 MHz +Move/adjust -5470 - 5725 MHz +Enter/Confirm -5725 - 5850 MHz +Button Guide + +Gaming +Configure the game related settings. + +Resolution +1902x1080 + +Black Equalizer +10 -2402 - 2480 MHz +aim Stabilizer +Off -Maximum Output Power (EIRP) -19 dBm +Overdrive +Balance -22 dBm +Refresh Rate +60Hz + +freeSync +ON + +Picture Mode + +Standard + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +ON + +Off + +aim Stabilizer + +Black Equalizer + +Super Resolution + +Display Mode + +Overdrive + +freeSync + +Exit + +Move/adjust + +Enter/Confirm + +Item +Aim Stabilizer + +Description +When this function is enabled, it reduces the motion +blur in a fast moving games. +Note: This function is not available under the following +conditions: +-when the frequency setting is lower than 100Hz. +-when the FreeSync function is activated. + +-28- + +LCD Monitor User's Manual Item +Black Equalizer +Super Resolution +Display Mode + +Overdrive +FreeSync + +Description +Adjust the brightness of the black areas. +Sharpen the low resolution images. +Select the aspect ratio of the screen. +• Full: Scale the input image to fill the screen. + +Ideal for 16:9 aspect images. + +• Aspect: Display the input image without geo- +metric distortion filling as much of the screen +as possible. + +Improve the response time of the LCD monitor. +When this function is enabled, it eliminates the +screen lag and tearing while playing games. + +-29- + +LCD Monitor User's Manual Picture + +Configure the image related settings. + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Standard + +fPS + +RTS/RPG + +Movie + +Reader + +sRGB + +Custom 1 + +Brightness + +Contrast Ratio + +Color Vibrance + +Sharpness + +Gamma + +Color Temperature + +Low Blue Light + +Exit + +Move/adjust + +Enter/Confirm + +Select the one of the preset picture mode. +• +• +• + +Standard: For editing documents or browsing websites. +FPS: For playing FPS (First Person Shooting) games. +RTS/RPG: For playing RTS (Real-Time Strategy) or RPG (Role-Playing +Game) games. + +Reader: For viewing documents. +sRGB: For viewing photos and graphics on the computer. + +• Movie: For watching movies. +• +• +• Custom 1: Customized picture mode settings. +• Custom 2: Customized picture mode settings. +• Custom 3: Customized picture mode settings. +Then you can configure the following image settings. + +Item +Brightness +Contrast +Color Vibrance +Sharpness + +Description +Adjust the brightness of the image. +Adjust the contrast of the image. +Adjust the intensity of the more muted colors. +Adjust the sharpness of the image. + +-30- + +LCD Monitor User's Manual Item +Gamma +Color Temperature Select the color temperature. + +Description +Adjust the middle level of luminance. + +Low Blue Light + +Note: Select User Define to customize the color +temperature by adjusting the red (R), green (G), or +blue (B) level according to your preference. +Reduce the amount of blue light exposure from +the screen. +• Level 0: No change. +• Level 1~10: The higher the level, the more blue + +light will be reduced. + +Note: + + This is the Low Blue Light Level 10 under +the normal image mode with normal col- +or temperature. It is compliance with TUV +Low Blue Light Certification + +Dynamic Contrast Adjust the black level of the image to achieve the + +Senseye Demo + +Reset Picture + +optimum contrast. +Split the screen in half (2 windows). +The image of the selected mode with its default +settings will appear on the left window and the +adjusted image with the new settings will appear +on the right window. +Reset all Picture settings to the default settings. + +-31- + +LCD Monitor User's Manual Display + +Configure the display related settings. + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +Input + +HDMI RGB PC Range + +Overscan + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Exit + +Move/adjust + +Enter/Confirm + +Item +Input +HDMI RGB +PC Range + +Overscan + +Description +Select the input source. +Select an appropriate RGB range setting or let the +monitor to detect it automatically. +Note: This option is only available for HDMI input. +When this function is enabled, it slightly enlarges +the input image to hide the outermost edges of +the image. +Note: This option is only available for HDMI input. + +-32- + +LCD Monitor User's Manual System + +Configure the system related settings. + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Item +Audio + +OSD Settings + +audio + +OSD Settings + +Quick Switch + +Other Settings + +Volume + +Mute + +Exit + +Move/adjust + +Enter/Confirm + +Description +Configure the audio settings. +• Volume: Adjust the volume level. +• Mute: Enable/Disable the mute function. +Configure the on-screen display (OSD) menu +related settings. +• Display Time: Set the length of time the OSD + +menu remains on the screen. + +• OSD Transparency: Adjust the OSD menu + +transparency. + +• OSD Lock: When this function is enabled, no + +OSD adjustment is allowed. +Note: To disable the OSD lock function, press +the Control key. When the message appears on +the screen, select Yes to confirm. + +The button is locked + +Confirm to unlock +OSD. + +No + +yes + +-33- + +LCD Monitor User's Manual Item +Quick Switch + +Other Settings + +Description +Assign the function of the hot keys. +Available options: Aim Stabilizer, Black Equalizer, +Low Blue Light, Volume, Input, Contrast, Bright- +ness, and Picture Mode. +The default setting of the hot keys as below: + +D Up( + +):Black Equalizer + +D Down( + +):Picture Mode + +D Right( + +):Input + +D Left( +• Resolution Notice: When this function is en- + +):Volume + +abled, the recommended resolution message +will appear on the screen when the monitor +switches to another input source. +Input Auto Switch: When this function is en- +abled, it automatically switches to an available +input source. + +• + +• Auto Power OFF: When this function is enabled, +the monitor will automatically turn off after a +certain amount of time. + +• DisplayPort Ver.: Set the DisplayPort version to + +1.1, 1.2. + +Note: +• + + Make sure to configure this option based +on the DisplayPort version supported by +your graphics card. + +-34- + +LCD Monitor User's Manual Language + +Language-related settings + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +English +English + +Deutsch + +Français + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Exit + +Move/adjust + +Enter/Confirm + +Item +Language + +Description +Select an available language for the OSD menu. + +-35- + +LCD Monitor User's Manual Save Settings + +Set the personalized OSD configurations for Setting1, Setting2, or +Setting3. + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +Setting1 + +Setting1 + +Setting1 + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Exit + +Move/adjust + +Enter/Confirm + +Item +Save +Load + +Reset all + +Description +Save the customized settings. +Load the saved settings. + +Restore the monitor to its factory default settings. + +Resolution +1902x1080 + +Brightness +85 + +Contrast +50 + +Sharpness +5 + +Gamma +Gamma 3 + +Color Temp. Normal + +Picture Mode + +Standard + +Gaming + +Picture + +Display + +System + +Language + +Save Settings + +Reset all + +Exit + +Move/adjust + +Enter/Confirm + +-36- + +LCD Monitor User's Manual aPPENDICES +Specifications + +Item +Specifications +Panel Size +27-inch +Aspect Ratio +16:9 +Resolution +1920 x 1080 +Refresh Rate +144Hz +LED Life +30000 hours (min.) +Pixel Pitch +0.31125 x 0.31125 +Response Time +1ms (MPRT) +Brightness +300nits (typ.) / 250nits (min.) +1000:1 (typ.) +Contrast Ratio +Dynamic Contrast Ratio 12M:1 (typ.) +Color Depth +Active Display area +Viewing Angle +Frequency +Input/Output Terminals • 2x HDMI 1.4 + +16. 7M +597.6(H) x 336.15 (V) +178° (H) / 178° (V) +48~144Hz (gaming mode) + +• 2x USB 3.0 (downstream) + +• 1x DP1.2 +• 1x Headphone +• 1x USB upstream + +Support USB BC1.2 +specifications +Single-port output: 5V/1.5A +Double-port output: 5V/1A + +port + +Power Supply/Input +Power Supply/Model +Power Consumptions + +AC100~240V@ 50/60Hz +Asian Power Devices Inc., NB-65B19 +• On: 60W (Max.) +• Standby: 0.5W +• Off: 0.3W + +Dimensions (W x H x D) 619.04 x 518.88 x 202. 97mm (with stand) +6.5±0.5 kgs (Net) +Weight +0°C - 40°C +Operating conditions: +temperature range + +Note: Specifications subject to change without notice. + +-37- + +LCD Monitor User's Manual Supported Timing List + +Timing List +640x480@60Hz +640x480@75Hz +720x480@60Hz +800x600@60Hz +800x600@75Hz +1024x768@60Hz +1024x768@75Hz +1280x720@60Hz +1280x720@120Hz +1280x720@144Hz +1920x1080@60Hz +1920x1080@120Hz +1920x1080@144Hz + +HDMI +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ + +DP +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ + +-38- + +LCD Monitor User's Manual Troubleshooting + +Problem + +No power + +No image appears +on the screen + +The image does +not fill the entire +screen + +The colors are +distorted + +No sound or +volume is low + +Possible Solution(s) + +• Make sure the power cable is properly connected + +to the power source and the monitor. + +• Check the cable and plug to make sure they are + +not broken. + +• Make sure the monitor is turned on. +• Make sure the adapter LED is on. + +• Make sure the monitor and the computer are + +properly connected and both devices are turned +on. + +• Make sure you have selected the correct input + +source. Refer to page 22 or 32. + +• Adjust the Brightness and Contrast settings. + +Refer to page 30. + +• Try different Display Mode (aspect ratio) setting. + +Refer to page 29. + +• Make sure the input signal cable is properly + +connected. + +• Configure the image related settings. Refer to + +the “Picture” section. + +• Make sure the headphone cable is properly + +connected. + +• Adjust the volume level. Refer to page 24 or 33. +• Adjust your computer sound related settings. + +Note: If the problem persists, please contact our customer service for +further assistance. + +-39- + +LCD Monitor User's Manual Basic Care + +When the screen becomes dirty, do the following to clean the screen: +• Wipe the screen with your soft, lint-free cloth to remove any residual + +• -20 dBm +dust. +Spray or apply the cleaning fluid onto a soft cloth. Then wipe the +screen gently with the damp cloth. -13.97 dBm +Note: +• Make sure that the cloth is damp, but not wet. +• When wiping, avoid applying excessive pressure on the screen. +• Acidic or abrasive cleaners can damage the screen. -14 dBm +Notes on USB Charging -Maximum Output Power (EIRP) -19 dBm +The following devices are not compatible with BC 1.2 of CDP mode. +When the Hub is connected to a PC, these devices cannot be charged +using the USB cable. -22 dBm +Vendor +Apple +Samsung +Samsung -23 dBm +Device +IPad +Galaxy Tab2 10.1 +GALAXY TabPro -13.97 dBm +PID/VID +PID_129A/VID_05AC +PID_6860/VID_04E8 +PID_6860/VID_04E8 -13 dBm +GIGaByTE Service Information -15 +More service information, please visit GIGABYTE official website: +https://www.gigabyte.com/ -Notebook PC User Guide 16 +-40- -Notebook PC User Guide \ No newline at end of file +LCD Monitor User's Manual \ No newline at end of file diff --git a/q192/random_k4/random_4.md b/q192/random_k4/random_4.md index c346d7f5a27964259c5644ba47c0ac6027670487..b916f8fadcb4f3d92a3b808c9955a8afa494b778 100644 --- a/q192/random_k4/random_4.md +++ b/q192/random_k4/random_4.md @@ -1,1018 +1,362 @@ -USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 1 of 27 +Mutation + + A mutation is a heritable change in the base sequence of the nucleic acid in +the genome of an organism. strain carrying such a change is called a mutant. +A mutant by definition differs from its parental strain in genotype, the +nucleotide sequence of the genome. + +Mutations can be either spontaneous or induced. Spontaneous mutations can +occur as a result of exposure to natural radiation (cosmic rays, and so on) +that alters the structure of bases in the DNA , Also, oxygen radicals can +affect DNA structure by chemically modifying DNA. + +Mutagens + +Mutagen is a physical or chemical agent that changes the genetic material, +usually DNA, of an organism and thus increases the frequency of mutations +above the natural background level. As many mutations can cause cancer, +mutagens are therefore also likely to be carcinogens. Not all mutations are +caused by mutagens: so-called "spontaneous mutations" occur due to +spontaneous hydrolysis, errors +repair and +recombination. + Mutagens cause changes to the DNA that can affect the transcription and +replication of the DNA, which in severe cases can lead to cell death. The +mutagen produces mutations in the DNA, and deleterious mutation can +result in aberrant, impaired or loss of function for a particular gene, and +accumulation of mutations may lead to cancer. + +replication, + +in DNA + +Mutagens may also modify the DNA sequence; the changes in nucleic acid +sequences by mutations include substitution of nucleotide base-pairs and +insertions and deletions of one or more nucleotides in DNA sequences. +Although some of these mutations are lethal or cause serious disease, many +have minor effects as they do not result in residue changes that have +significant effect on the structure and function of the proteins. Many +mutations are silent mutations, causing no visible effects at all, either +because they occur in non-coding or non-functional sequences, or they do +not change the amino-acid sequence due to the redundancy of codons. -[PUBLISH] +1 + + Types of mutagens + +Mutagens may be of physical, chemical or biological origin. They may act +directly on the DNA, causing direct damage to the DNA, and most often +result in replication error. Some however may act on the replication +mechanism and chromosomal partition. Many mutagens are not mutagenic +by themselves, but can form mutagenic metabolites through cellular +processes. Such mutagens are called promutagens. + +Physical mutagens + + -In the -United States Court of Appeals -For the Eleventh Circuit +Ionizing radiations such as X-rays, gamma rays and alpha particles +may cause DNA breakage and other damages. The most common +sources include cobalt-60 and cesium-137. -____________________ + Ultraviolet radiations with wavelength above 260 nm are absorbed +strongly by bases, producing pyrimidine dimers, which can cause +error in replication if left uncorrected. -No. 22-13963 + Radioactive decay, such as 14C in DNA which decays into nitrogen. -____________________ +Chemical mutagens -UNITED STATES OF AMERICA, +A large number of chemicals may interact directly with DNA. -versus + Reactive oxygen species (ROS) – These may be superoxide, hydroxyl -SHADON DEMETRIC EDWARDS, +radicals and hydrogen peroxide, - Plaintiff-Appellee, + Deaminating agents, for example nitrous acid which can cause - Defendant-Appellant. +transition mutations by converting cytosine to uracil. -____________________ + Polycyclic aromatic hydrocarbon (PAH), when activated to diol- -Appeal from the United States District Court -for the Southern District of Florida -D.C. Docket No. 0:19-cr-60373-AHS-1 -____________________ +epoxides can bind to DNA and form adducts. - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 2 of 27 + Alkylating agents such as ethylnitrosourea. The compounds transfer +methyl or ethyl group to bases or the backbone phosphate groups. +Guanine when alkylated may be mispaired with thymine. Some may +cause DNA crosslinking and breakages. Nitrosamines are an +important group of mutagens found in tobacco, and may also be +formed in smoked meats and fish via the interaction of amines in food 2 -Opinion of the Court + with nitrites added as preservatives. Other alkylating agents include +mustard gas and vinyl chloride. -22-13963 + Aromatic amines and amides have been associated with -Before ROSENBAUM, NEWSOM, and MARCUS, Circuit Judges. +carcinogenesis since 1895 . -NEWSOM, Circuit Judge: + Alkaloid from plants, such as those from Vinca species may be +converted by metabolic processes into the active mutagen or +carcinogen. -Shadon Edwards pleaded guilty to possessing a firearm and -ammunition despite having previously been convicted of a crime -punishable by more than a year in prison, in violation of 18 U.S.C. -§ 922(g). On appeal, Edwards challenges his § 922(g) conviction -and 180-month sentence on three grounds. First, he contends that -his guilty plea wasn’t knowing and voluntary because the district -court failed to inform him that, by pleading, he would waive his -right to appeal the denial of an earlier suppression motion. Second, -he asserts that the district court wrongly enhanced his sentence un- -der the Armed Career Criminal Act because the government didn’t -prove beyond a reasonable doubt, as required by Erlinger v. United -States, 602 U.S. 821 (2024), that he had earlier been convicted of -three crimes that occurred on different occasions. Third, he argues -that § 922(g) exceeds Congress’s Commerce Clause authority. + Bromine and some compounds that contain bromine in their chemical -After careful review, and with the benefit of oral argument, +structure. -we affirm the district court’s judgment. + Sodium azide, an azide salt that is a common reagent in organic -I +synthesis and a component in many car airbag systems -A + Psoralen combined with ultraviolet radiation causes DNA cross- -Shadon Edwards was suspected of being a high-level organ- -izer in a drug-trafficking organization responsible for distributing -illegal narcotics throughout south Florida. After obtaining an ar- -rest warrant for Edwards, law-enforcement officials observed him -driving around Pompano Beach. The officers followed and, after +linking and hence chromosome breakage. - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 3 of 27 + Benzene, an industrial solvent and precursor in the production of -22-13963 +drugs, plastics, synthetic rubber and dyes. -Opinion of the Court +Base analogs -3 + Base analog, which can substitute for DNA bases during replication -seeing Edwards park, approached and took him into custody. A -search incident to Edwards’s arrest revealed that he had a Glock 43 -nine-millimeter semi-automatic pistol, loaded with six rounds of -ammunition. - -B - -A grand jury indicted Edwards on a single count: the know- -ing possession of a firearm and ammunition by one who had pre- -viously been convicted of a crime punishable by more than a year -in prison, in violation of 18 U.S.C. § 922(g)(1).1 Edwards filed a -motion to suppress the gun and ammunition found in his car, -which the district court denied. He then pleaded guilty pursuant -to a written plea agreement that neither expressly preserved nor -expressly waived his right to appeal. In connection with his plea, -Edwards also signed a factual proffer describing his offense conduct -and some of his prior criminal history, including, as particularly rel- -evant here, convictions “for trafficking -in oxycodone (10- -022573CF10A), aggravated battery with a deadly weapon (10- -022679CF10A), and robbery with a deadly weapon/aggravated bat- -tery (97-000465CF10B).” Factual Proffer ¶ 2, Dkt. No. 81. - -At the change-of-plea hearing, the district judge engaged in -a colloquy with Edwards to assess several matters, including -whether he was competent to plead guilty and whether he under- -stood the consequences of doing so. At the hearing, there was no - -1 The indictment also mentioned the Armed Career Criminal Act, 18 U.S.C. -§ 924(e)(1), albeit without further elaboration. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 4 of 27 +and cause transition mutations. -4 + Intercalating agents -Opinion of the Court - -22-13963 - -discussion—one way or the other—about whether, by pleading, -Edwards would forfeit his right to appeal the earlier denial of his -suppression motion. Edwards pleaded guilty, and the judge found -that his plea was “entered into freely and voluntarily with a know- -ing, intelligent waiver of [his] rights.” Tr. of Change of Plea at -18:20–22, Dkt. No. 105. - -In computing Edwards’s offense level for sentencing pur- -poses, the presentence investigation report applied an enhance- -ment pursuant to U.S.S.G. § 4B1.4(a), which states that “[a] defend- -ant who is subject to an enhanced sentence under the provisions of -18 U.S.C. § 924(e) is an armed career criminal.” Section 924(e), the -Armed Career Criminal Act, in turn prescribes a mandatory mini- -mum 15-year sentence for anyone who violates § 922(g) and “has -three previous convictions” for “violent felon[ies] or . . . serious -drug offense[s], or both, committed on occasions different from -one another.” 18 U.S.C. § 924(e). In particular, Edwards’s PSI -listed the same three predicate convictions as had his factual prof- -fer: - -The defendant is an armed career criminal because he -is subject to an enhanced sentence under the provi- -sions of 18 U.S.C. § 924(e), pursuant to § 4B1.4(a). -The defendant has the prior felony convictions as -noted in docket numbers 10-022573CF110A [i.e., for -trafficking in oxycodone], 10-022679CF10A [i.e., for -aggravated battery with a deadly weapon], and 97- -000465CF10B [i.e., -for robbery with a deadly -weapon/aggravated battery]. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 5 of 27 - -22-13963 - -Opinion of the Court + -5 +Intercalating agents, such as ethidium bromide and proflavine, are +molecules that may insert between bases in DNA, causing frameshift +mutation during replication. Some such as daunorubicin may block +transcription and +to +proliferating cells. -PSI ¶ 20. With the ACCA enhancement, Edwards’s statutory min- -imum term of imprisonment was 15 years, so the PSI increased the -low end of Edwards’s range from 168 months to 180 months in -prison. Edwards didn’t object to the PSI’s recommended sen- -tence.2 - -At the sentencing hearing, the district judge noted that the -PSI’s recommended sentence reflected an ACCA-enhanced 15-year -mandatory minimum, and Edwards agreed to it. The judge there- -fore imposed a sentence of 180 months in prison, followed by a -two-year term of supervised release. Again, Edwards didn’t object. - -During the sentencing hearing, Edwards’s lawyer asked the -judge to appoint separate counsel to appeal the earlier denial of the -motion to suppress the handgun evidence. Seemingly in an effort -to recall the nature of Edwards’s plea, which Edwards had entered -several months earlier, the judge asked the lawyers whether there -was a formal agreement. After Edwards’s counsel explained that -he didn’t “remember there being an appellate waiver,” the judge -suggested that because “the plea agreement . . . does not mention -an appellate waiver,” Edwards had “the right . . . to take an appeal” - -2 Edwards made two factual objections to the PSI—both are unrelated to this -appeal. See Tr. of Sentencing Hr’g at 3:20–25, Dkt. No. 106 (“[Defense Coun- -sel]: There were two very minor—they’re not even—they don’t even affect -his guidelines—two minor objections, I would say, to the PSI, and they’re mi- -nor in and of itself. That’s about it. Other than that, we’re just asking for the -Court to sentence him to 15 years and we all go about our business.”). - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 6 of 27 +replication, making -6 +them highly -Opinion of the Court +toxic -22-13963 + Metals -of the suppression ruling. Tr. of Sentencing Hr’g at 24:18–22, Dkt. -No. 106. + Many metals, such as arsenic, cadmium, chromium, nickel and their +compounds may be mutagenic, they may however act via a number of +different mechanisms.. -Edwards subsequently filed a notice of appeal indicating his -intent to seek review of the denial of the suppression motion. The -district court granted Edwards’s lawyer’s motion to withdraw and -assigned new counsel, who filed an amended notice of appeal. +3 -II + Biological agents + + Transposon, a section of DNA that undergoes autonomous fragment +relocation/multiplication. Its insertion into chromosomal DNA disrupt +functional elements of the genes. Transposons have the ability to +move (transpose) from one site to another. + +Elements known as Insertion Sequences (IS) have a specific ability to +insert into other DNA sequences, thus generating insertion mutations. +A substantial proportion of spontaneous mutations may be due to +inactivation of genes by insertion of a copy of an IS element rather +than by replication errors. + + Virus , Virus DNA may be inserted into the genome and disrupts +genetic function. Infectious agents have been suggested to cause +cancer as early as 1908 by Vilhelm Ellermann and Oluf Bang. + +In summary mutagens can be divided into Three classes based on the ways +they cause mutation: + + * Base analogs. Examples: ( 2-aminopurine, ) (5-bromouracil.) +These base analogs are incorporated into DNA where they mispair with +other bases. 5BU can pair with adenine and guanine. These result in +transition mutations. + + * Base modifications causing mispairs Examples: ethyl methane +sulfonate (EMS). These mutagens modify bases on DNA such that they +mispair. EMS alkylate the O6 of guanine, which is highly mutagenic and +causes mispairing with thymine ,and show preference for GC to AT +transitions. However, they also alkylate bases at many positions with other +effects. Nitrous acid and hydroxylamine deaminate cytosine to yield uracil +(see deamination above) resulting in transition mutations. + + * Base modifications which destroy pairing; SOS-dependent +mutagens. Examples: UV light, benzo(a)pyrene, aflatoxin B1 (i.e. most +carcinogens) These mutagens or their metabolites modify DNA so that no +specific pairing is possible; replication cannot proceed past the lesion. +Unrepaired AP sites also elicit this response. -Before us, Edwards challenges his conviction and 180-month -sentence on three grounds: He argues (1) that his guilty plea wasn’t -knowing and voluntary because he didn’t realize that by pleading -he would waive his right to appeal the district court’s denial of his -suppression motion; (2) that the district court erred in enhancing -his sentence under ACCA because the government didn’t prove, as -required by Erlinger v. United States, 602 U.S. 821 (2024), that his -predicate offenses occurred on different occasions; and (3) that -§ 922(g)(1) exceeds Congress’s Commerce Clause power. We will -address each issue in turn. +4 -A + Types of mutations + + Point mutations +There are many ways in which the structure of the genetic material may +change. In point mutations the sequence of the DNA has been altered at a +single position. Where this change consists of replacing one nucleotide by +another, it is known as a base substitution. The consequence of such a +change depends both on the nature of the change and its location. If the +change is within the coding region of a gene (i.e. the region which ultimately +is translated into protein), it may cause an alteration of the amino acid +sequence which may affect the function of the protein. The alteration may of +course have little or no effect, either because the changed triplet still codes +for the same amino acid or because the new amino acid is sufficiently +similar to the original one for the function of the protein to remain +unaffected. +For example, the triplet UUA codes for leucine; a single base change in the +DNA can give rise to one of nine other codons. Two of the possible changes +(CUA , UUG) are completely silent, as the resulting codons still code for +leucine. These are known as synonymous codons. Two further changes +(AUA and GUA) may well have little effect on the protein since the +substituted amino acids (isoleucine and valine respectively) are similar to the +original leucine (they are all hydrophobic amino acids). Phenylalanine (UUC +or UUU codons) is also hydrophobic but is more likely to cause a significant +change in the structure of the protein at that point. The significance of the +change to UCA, resulting in the substitution of serine (which is considerably +different) for leucine will depend on the role played by that amino acid (and +its neighbours) in the overall function or conformation of the protein. +The final two changes (UGA , UAA) are referred to as stop or termination +codons (as is a third codon, UAG), since they result in termination of +translation; there is normally no tRNA molecule with the corresponding +anticodon. + +Types of point mutations + +• A nucleotide-pair substitution: replaces one nucleotide and its partner +with another pair of nucleotides -First up, the voluntariness of Edwards’s plea. When, as here, -a defendant fails to object during his plea colloquy, we review the -district court’s determination that his plea was knowing and volun- -tary only for plain error. See United States v. Monroe, 353 F.3d 1346, -1349 (11th Cir. 2003) (citing United States v. Vonn, 535 U.S. 55, 59 -(2002)). “To establish plain error, a defendant must show that (1) -there was an error, (2) the error was plain, and (3) the error affects -substantial rights.” United States v. Roosevelt Coats, 8 F.4th 1228, 1235 +5 - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 7 of 27 + • Silent mutations have no effect on the amino acid produced by a codon +because of redundancy in the genetic code +• Missense mutations still code for an amino acid, but not the correct amino +acid +• Nonsense mutations change an amino acid codon into a stop codon, +nearly always leading to a nonfunctional protein + +A different kind of mutation still involving a change at a single position, +consists of the deletion or addition of a single nucleotide (or of any number +other than a multiple of three). This is known as a frameshift mutation, +since it results in the reading frame being altered for the remainder of the +gene. Since the message is read in triplets, with no punctuation marks (the +reading frame being determined solely by the translation start codon), an +alteration in the reading frame will result in the synthesis of a totally +different protein from that point on. + +Figure 13: Nucleotide –pair substitution : silent [Pearson Education +inc,2011]. -22-13963 +6 -Opinion of the Court + Figure 14: Nucleotide –pair substitution : missense [Pearson Education +inc,2011]. 7 -(11th Cir. 2021). An error about the voluntariness of a plea affects -a defendant’s “substantial rights” if he shows a reasonable proba- -bility that the error influenced his “‘decision to plead.’” United -States v. Steiger, 99 F.4th 1316, 1326 (11th Cir. 2024) (en banc) (quot- -ing United States v. Dominguez Benitez, 542 U.S. 74, 84 (2004)). If the -defendant makes the first three showings, we have the “discretion -to recognize an unpreserved error but only if (4) the error seriously -affects the fairness, integrity or public reputation of judicial pro- -ceedings.” Roosevelt Coats, 8 F.4th at 1235 (citation modified). - -Before a district judge accepts a guilty plea, he must “address -the defendant personally in open court and determine that the plea -is voluntary.” Fed. R. Crim. P. 11(b)(2). In particular, the judge -must ensure that the defendant understands certain facts, including -“the terms of any plea-agreement provision waiving the right to -appeal or to collaterally attack the sentence.” Id. 11(b)(1)(N); see -also, e.g., United States v. Lejarde-Rada, 319 F.3d 1288, 1289 (11th Cir. -2003) (noting that the Rule 11 colloquy aims to ensure, among -other things, that the defendant “know[s] and understand[s] the -consequences of his guilty plea” (citation modified)). - -Importantly for present purposes, a knowing and voluntary -guilty plea “waives all non-jurisdictional defects,” United States v. -Pierre, 120 F.3d 1153, 1155 (11th Cir. 1997) (citation modified), in- -cluding, as relevant here, the allegedly erroneous denial of a mo- -tion to suppress evidence. Accordingly, a “defendant who wishes -to preserve appellate review of a non-jurisdictional defect while at -the same time pleading guilty can do so only by entering a - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 8 of 27 + Figure 15: Nucleotide –pair substitution : nonsense [Pearson Education +inc,2011]. 8 -Opinion of the Court - -22-13963 - -‘conditional plea’ in accordance with Fed. R. Crim. P. 11(a)(2).” Id. -All here agree that Edwards’s plea was “unconditional.” Br. of Ap- -pellant at 31. - -Edwards insists, though, that his guilty plea wasn’t knowing -and voluntary because he didn’t understand one of the conse- -quences of pleading—namely, that he would lose the right to ap- -peal the denial of his motion to suppress the handgun evidence. -He asserts that the district court failed to inform him of that fact -during the Rule 11 colloquy and, indeed, even suggested at his later -sentencing hearing that he could pursue the appeal. Edwards relies -principally on our decision in Pierre to argue that because he be- -lieved that his plea agreement preserved the right to appeal the de- -nial of the suppression motion—but in fact didn’t—his plea wasn’t -knowing and voluntary. - -In Pierre, we considered the appeal of a defendant who, hav- -ing pleaded guilty, sought to argue that his speedy-trial rights had -been violated. 120 F.3d at 1154. We vacated the conviction upon -determining that, at the time he pleaded guilty, the defendant didn’t -understand the consequences of doing so: “Because Pierre en- -tered—and the district court accepted—this guilty plea only on the -reasonable (but mistaken) belief that [he] had preserved the speedy -trial issues for appeal, his plea was, as a matter of law, not knowing -and voluntary.” Id. at 1156. We based our decision on a review of -“[t]he plea colloquy,” which “unequivocally indicate[d] that Pierre -intended to plead guilty on the condition” that he retained his ap- -pellate rights. Id. at 1155. For example, we cited statements at the - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 9 of 27 - -22-13963 - -Opinion of the Court + Figure 16: Nucleotide –pair insertion or deletion: immediate nonsense +[Pearson Education inc,2011]. 9 -plea hearing demonstrating that Pierre had pleaded guilty “only af- -ter having been assured by the district court judge that [] he had -preserved the speedy trial issues for appeal.” Id. In particular, the -district judge had said: “So I’m going to deny your renewed motion -to dismiss. Now, if you’re ready to go to trial, we have a jury com- -ing up. If you feel that you wish to plead guilty and preserve these -legal issues, you may do so.” Id. We also cited statements demon- -strating that “Pierre’s counsel [had] explicitly tried to preserve the -speedy trial issue.” Id. Specifically, Pierre’s counsel had stated: “As -long as my client is assured by the court, which you have done, that -these issues are protected for purposes of appeal, [Pierre] may be -willing to wish to do that in the future. But for the purposes of -trial, he is entering a plea. It is in his best interest at this time.” Id. - -Here, Edwards hasn’t pointed to any similar evidence from -the plea hearing—no statement by him, his lawyer, or the district -judge—indicating that he didn’t understand the consequences of -pleading guilty. To be sure, Edwards now says that once he lost -the suppression issue, he wanted to plead and “go onto the sentenc- -ing part” so that he could appeal the denial of the motion. See Br. -of Appellant at 29–30 (quoting Tr. of Change of Plea at 10:4–12). -But at the plea hearing, the suppression issue was mentioned only -once, in passing. The district judge asked Edwards’s counsel how -he “got ready for today’s hearing.” Tr. of Change of Plea at 8:5–6. -Edwards’s lawyer responded: “[I]n anticipation of today’s hearing, -[my client] and I had a scheduled visit last week where we went -over the entire plea agreement and . . . the entire factual proffer - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 10 of 27 + Figure 17: Nucleotide –pair insertion or deletion: extensive missense +[Pearson Education inc,2011]. -10 +Conditional mutants: +There are many genes that do not affect resistance to antibiotics or +bacteriophages, biosynthesis of essential metabolites or utilization of carbon +sources. Some of these genes are indispensable and any mutants defective in +those activities would die (or fail to grow). -Opinion of the Court +DNA modification repair pathways -22-13963 + Repair pathways that remove DNA modifications have Three basic +mechanisms: direct repair, base-excision repair and nucleotide-excision +repair. -together.” Id. at 8:18–21. Edwards’s lawyer then added the follow- -ing regarding the suppression motion—its only mention: - -[Defense Counsel]: As the Court noted earlier, we -also had a motion to suppress in this case. The facts -are rather simple. I mean, he was pulled over and he -was arrested and he was in possession of a firearm -and he was a convicted felon. It’s not a—this is not -factually difficult. It was just whether or not they had -requisite legal rights to effectuate that stop. That was -the whole issue. And this Court, unfortunately, ruled -against us, but that’s the nature of the case. - -Id. at 8:22–9:4. - -No one suggested at that time that Edwards wanted to—or -thought he could—appeal the denial of the suppression motion. -And Edwards’s counsel certainly didn’t tie the entry of his plea to -the preservation of the right to appeal, as the lawyer had in Pierre. -To the contrary, Edwards’s lawyer seemed resigned to the district -court’s ruling. +10 -The preservation issue didn’t resurface for another three -months, at the sentencing hearing. Edwards asserts before us that -a colloquy at that hearing shows that he intended all along to ap- -peal the denial of the motion to suppress. Br. of Appellant at 30– -31. We disagree. + A. Direct chemical reversal. -Here’s the relevant sentencing-hearing exchange, in full: + 1. Photolyase, AKA photo-reactivation. UV +the +formation of pyrimidine dimers between adjacent C or T bases in DNA. The +photolyase enzyme break the cyclobutane dipyrimidine bond. To do so, the +enzyme must absorb visible light, hence the name photo-reactivation. E. coli +and the yeast Saccharomyces cerevisiae have such an enzyme. -Defense Counsel: He did tell me, Judge, he wants me -to try to get counsel appointed to just appeal the +induces - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 11 of 27 +light -22-13963 + 2. Methyltransferase. The methyl groups from mutagenic O6- +methylguanine (O6-MeG is particularly mutagenic) and O4-methylthymine +can be removed directly by this enzyme. -Opinion of the Court + B. base-excision: -11 +Glycosylase + AP endonuclease. Many modified bases are recognized by +specific N-glycosylases that cleave the modified base. The resulting AP site +is repaired by AP endonucleases. Examples in E. coli: hypoxanthine-DNA +glycosylase, +formamidopyrimidine +glycosylase, hydroxymethyl glycosylase. -ruling on the motion to suppress. So I told him that I -would file the notice. And if the Court at some point -needs to have a hearing as to whether he is in forma -pauperis and is entitled to counsel for that appeal, I -guess we’ll have a hearing. But that’s what he re- -quested of me. +3-methyladenine -The Court: Sure. On this case did we have a formal -plea agreement? I can’t remember now. +glycosylase -Defense Counsel: I don’t remember – – +. -Government: No, I think he pled open, Your Honor. + c. Nucleotide excision repair. -Defense Counsel: Yeah. Yeah, I don’t remember -there being an appellate waiver. I don’t recall. +In this form of DNA repair, the damaged bases are removed from DNA as +an oligonucleotide and the resulting gap is repaired by resynthesis. This +pathway is used to remove many bulky adducts in DNA, including cross- +links and UV-induced pyrimidine dimers. -The Court: Right. I just wanted to advise him if there -was. So the plea agreement at docket entry 80 does -not mention an appellate waiver. Certainly, sir, you -do have the right, if you want, to take an appeal of -that. Just don’t miss your timelines on that. Okay? +11 -Mr. Edwards: All right. Thank you, sir. + Figure 17: Nucleotide excision repair = enzymes that function to cut out and replace +DNA damage [Pearson Education inc,2004] -Tr. of Sentencing Hr’g at 24:7–25:1. +In both E. coli , yeast and human cells, there seems to be a mechanism that +targets excision repair to lesions on transcribed strand (i. e. the template +strand for transcription) of the damaged gene. -Unsurprisingly, Edwards focuses on the district court’s state- -ment that he “ha[d] the right, if [he] want[ed], to take an appeal” -from the order denying the suppression motion. Id. at 24:23–24. -But that remark is both too little and too late. As for the too-little -part, the context makes clear that in the months that had passed -since the plea hearing, everyone (including the district judge) had +SOS response : a number of genes involved with DNA repair and related +functions are induced by the presence of unrepaired DNA. An alternative +strategy, when faced with overwhelming levels of DNA damage preventing +normal replication, is to temporarily modify or abolish the specificity of the +DNA polymerase. +Mutation Rate - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 12 of 27 +Mutation rate is a probability that gene will mutate when cell divides . A +mutation rate has evolved in cells that is very low yet detectable. This allows +organisms to balance the need for genetic stability with that for evolutionary +improvement. 12 -Opinion of the Court - -22-13963 - -lost track of the details of Edwards’s plea—whether it was a formal -plea agreement, whether it contained an appeal waiver, etc. At the -sentencing hearing, the judge assumed that because the agreement -didn’t contain an appeal waiver, Edwards could appeal the suppres- -sion order. But the judge hardly made any guarantees. - -And as for the too-late: All of the evidence to which Ed- -wards now points comes from the sentencing hearing, which, -again, took place some three months after he pleaded guilty. That -makes this case quite different from Pierre, in which the defendant -could—and did—point to strong indications from the plea hearing -itself that he misunderstood, at the time, what appellate rights he -was waiving. There is simply no basis in the record here for imput- -ing backwards, as it were, whatever was said at sentencing to show -genuine confusion at the earlier plea hearing.3 - -Edwards has failed to show that the district court plainly -erred in accepting his plea as knowing and voluntary.4 We there- -fore affirm the district court’s judgment in that respect. - -3 Edwards separately complains that the district judge failed to explicitly warn -him that he couldn’t appeal the denial of his motion to suppress. But nothing -in Rule 11 or our caselaw requires a judge to proactively inform a defendant -that by entering an unconditional plea he waives his right to appeal all non- -jurisdictional matters. Edwards ultimately concedes as much, saying only that -one circuit court has called this the “preferred practice.” Reply Br. at 3–4 (cit- -ing United States v. Lampazianie, 251 F.3d 519, 526 (5th Cir. 2001)). Perhaps, -but that is no basis for finding plain error. - -4 Unrelatedly, Edwards argues that his plea wasn’t knowingly and voluntarily -entered “on the basis of Erlinger as well.” Letter Br. of Appellant at 5 n.4, Dkt. -No. 37. In a footnote in a supplemental brief, Edwards asserts that had he - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 13 of 27 - -22-13963 - -Opinion of the Court + The fact that organisms as phylogenetically disparate as hyper thermophilic +Archaea and Escherichia coli have about the same mutation rate might make +one believe that evolutionary pressure has selected for organisms with the +lowest possible mutation rates. However, this is not so. The mutation rate in +an organismis subject to change. For example, mutants of some organisms +have been selected in the laboratory that are hyperaccurate in DNA +replication and repair. +However, in these strains, the improved proofreading repair mechanisms has +a significant metabolic cost; thus, a hyperaccurate mutant might actually be +at a disadvantage in its natural environment. On the other hand, some +organisms seem to benefit from a hyperaccurate phenotype that enables +them to occupy particular niches in nature. A good example is the bacterium +Deinococcus radiodurans . This organism is 20 times more resistant to UV +radiation and 200 times more resistant to ionizing radiation than E. coli. 13 -B - -1 - -We next consider the Armed Career Criminal Act enhance- -ment. Because Edwards never objected to the enhancement, we -review only for plain error. See United States v. Jones, 743 F.3d 826, -828 (11th Cir. 2014); Fed. R. Crim. P. 52(b). While a conviction un- -der § 922(g) ordinarily yields a maximum sentence of 10 years in -prison, 18 U.S.C. § 924(a)(2), ACCA mandates a minimum 15-year -sentence for an offender who has three prior convictions for “vio- -lent felon[ies]” or “serious drug offense[s]” that were “committed -on occasions different from one another,” id. § 924(e)(1). Edwards -contends that the district court erred in enhancing his sentence un- -der ACCA because the government failed to prove beyond a rea- -sonable doubt that his three predicate convictions occurred on dif- -ferent occasions. - -In Wooden v. United States, the Supreme Court explained that -ACCA “contains both a three-offense requirement and a three-occa- -sion requirement.” 595 U.S. 360, 368 (2022). Although ACCA - -“known that he was entitled to a jury determination as to whether his offenses -were committed on different occasions, he may not have pled guilty—or he -may have pled guilty to the 18 U.S.C. § 922(g)(1) charge, and sought a trial on -the occasions issue.” Id. We’ll address the merits of Edwards’s Erlinger argu- -ment in the next Part, but we decline to consider it vis-à-vis the voluntariness -of his plea. “We have long held that an appellant abandons a claim when he -either makes only passing references to it or raises it in a perfunctory manner -without supporting arguments and authority.” Sapuppo v. Allstate Floridian Ins. -Co., 739 F.3d 678, 680–81 (11th Cir. 2014). - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 14 of 27 - -14 - -Opinion of the Court - -22-13963 - -initially focused only on the number of offenses, Congress later -amended it to specify that qualifying offenses must also occur on -different occasions. See Anti-Drug Abuse Act of 1988, Pub. L. 100– -690, § 7056 (codified as amended at 18 U.S.C. § 924(e)(1)) (inserting -“committed on occasions different from one another,” after “or -both”). Congress “added the occasions clause only after a court -applied ACCA to an offender . . . convicted of multiple counts of -robbery arising from a single criminal episode.” Wooden, 595 U.S.at -371–73 (citing Petty v. United States, 481 U.S. 1034, 1034–1035 -(1987)). The Wooden Court emphasized that while each ACCA- -qualifying offense often occurs separately, “multiple crimes may -occur on one occasion even if not at the same moment.” Id. at 366. -“[T]reating each temporally distinct offense as its own occasion” as -a per se matter, the Court explained, would improperly “collaps[e] -two separate statutory conditions.” Id. at 368. - -To operationalize the different-occasions requirement, the -Wooden Court described a “range of circumstances [that] may be -relevant to identifying episodes of criminal activity”—e.g., the of- -fenses’ timing, location, and character. Id. at 369. Applying those -considerations—which we’ll explore in detail soon enough—the -Court concluded that Wooden’s ten burglaries didn’t occur on dif- -ferent occasions because (1) he committed them “in a single unin- -terrupted course of conduct,” (2) he committed them at “one loca- -tion,” and (3) the offenses were “essentially identical” and “inter- -twined.” Id. at 370. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 15 of 27 - -22-13963 - -Opinion of the Court - -15 - -Approximately four months after Wooden was decided, Ed- -wards entered his unconditional guilty plea in this case. His written -plea agreement recites that “[t]he defendant . . . understands and -acknowledges that the Court must impose a statutory mandatory -minimum term of imprisonment of fifteen (15) years.” Plea Agree- -ment ¶ 3, Dkt. No. 80. The agreement was accompanied by a -signed factual proffer, which recited that “[t]he defendant has at -least seven (7) felony convictions” and specifically enumerated -three of them—“for trafficking in oxycodone (10-022573CF10A), -aggravated battery with a deadly weapon (10-022679CF10A), and -(97- -robbery with a deadly weapon/aggravated battery -000465CF10B).” Factual Proffer ¶ 2. The same three convictions -were cited in Edwards’s presentence investigation report as -grounds for enhancing his sentence under ACCA. See PSI ¶ 20 -(“The defendant is an armed career criminal because he is subject -to an enhanced sentence under the provisions of 18 U.S.C. § 924(e), -pursuant to § 4B1.4(a). The defendant has the prior felony convic- -in docket numbers 10-022573CF110A, 10- -tions as noted -022679CF10A, and 97-000465CF10B.”). And the district court in -fact used those three convictions to enhance Edwards’s sentence. -The judge explained that “looking at the offense level computa- -tion” in the PSI, the “Armed Career Criminal enhancement gets -[the offense level] up to 33.” Tr. of Sentencing Hr’g at 4:19–23 (cit- -ing PSI ¶ 20). - -Edwards didn’t object to the ACCA enhancement on any -ground, including that the offenses underlying the ACCA predi- -cates weren’t committed on different occasions. He didn’t object - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 16 of 27 - -16 - -Opinion of the Court - -22-13963 - -to the PSI’s ACCA-enhanced recommended sentence, see supra at 5 -n.2, nor did he object to the enhancement during the sentencing -hearing. To the contrary, Edwards’s counsel suggested at the sen- -tencing hearing that the ACCA-enhanced 15-year sentence was ap- -propriate: - -[Defense Counsel]: My client has been aware for -months that he’s facing a statutory minimum sen- -tence of 15 years, which is 180 months in federal -prison. He’s been aware of it. He has not gone to -trial on this particular case. He has not testified in this -particular case. He got acceptance of responsibility, -even though it doesn’t really affect him. So he’s just -asking the Court to accept both parties’ recommen- -dations, sentence him to 15 years, and he’ll move on -with his life. - -Tr. of Sentencing Hr’g at 16:5–13; see also id. at 21:14–17 (“Court: -[N]ow that sentence has been imposed, do either of you object to -the Court’s finding of fact or the manner in which the sentence was -pronounced? [Defense Counsel]: No, Judge.”). Hearing no objec- -tion to the applicability of the ACCA enhancement based on Ed- -wards’s three prior convictions, the district court sentenced him to -the 15-year statutory minimum. - -Then came Erlinger, in which the Supreme Court held that -because the determination that a defendant’s prior crimes were -committed on different occasions “‘increase[s] the prescribed range -of penalties to which a criminal defendant is exposed,’” it “must be -resolved by a unanimous jury beyond a reasonable doubt (or freely - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 17 of 27 - -22-13963 - -Opinion of the Court - -17 - -admitted in a guilty plea).” 602 U.S. at 834 (quoting Apprendi v. New -Jersey, 530 U.S. 466, 490 (2000)).5 In light of Erlinger, Edwards con- -tends on appeal that because a jury didn’t find beyond a reasonable -doubt that his three ACCA predicate offenses were committed on -different occasions, the district court committed plain error when -it imposed an enhanced 15-year minimum-mandatory sentence. - -2 - -The government concedes that in light of Erlinger, Edwards -has satisfied the first two prongs of the plain-error standard—an -“error” that is “plain”—because, it says, the different-occasions re- -quirement was neither proven to a jury beyond a reasonable doubt -nor expressly admitted by Edwards in his plea. See Br. of Appellee -at 28.6 - -5 To be clear, Erlinger didn’t add a different-occasions element to the statute. -As already explained, Congress codified the different-occasions requirement -some 35 years ago. And almost four months before Edwards entered his plea, -and eight months before his sentencing hearing, the Supreme Court in Wooden -reiterated the significance of the different-occasions element and clarified the -circumstances under which offenses will (and won’t) satisfy it. Erlinger, rather, -resolved the “who decides” question—namely, by holding that the different- -occasions issue must be resolved either by a jury beyond a reasonable doubt -or by the defendant himself in a guilty plea. See 602 U.S. at 834. District courts, -the Erlinger Court held, “may not assume the jury’s factfinding function for -themselves” to conclude that offenses occurred on different occasions. Id. - -6 The second half of that disjunction isn’t self-evidently correct. Here, Ed- -wards at least arguably admitted in his plea that his three convictions satisfied -ACCA’s different-occasions requirement. As already explained, Edwards ad- -mitted that the three convictions enumerated in both the factual proffer and -the PSI qualified him for the ACCA enhancement. And by pleading to the - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 18 of 27 - -18 - -Opinion of the Court - -22-13963 - -Still, to prevail on plain error, Edwards must show, at step -three, that the Erlinger error affected his “substantial rights.”� -“[U]nlike harmless error—where the government carries the bur- -den—the onus of establishing prejudice under plain error rests with -the defendant.” United States v. Margarita Garcia, 906 F.3d 1255, -1267 (11th Cir. 2018). That burden is “heavy”—and, in particular, -“the burden of showing prejudice to meet the third-prong require- -ment is anything but easy.” Id. (citation modified). - -To demonstrate prejudice, a defendant must “show a rea- -sonable probability that, but for the error, the outcome of the pro- -ceeding would have been different.” United States v. Corbett, 921 F.3d -1032, 1037 (11th Cir. 2019) (quoting Molina-Martinez v. United States, -578 U.S. 189, 194 (2016)). “[W]e define the ‘outcome’ that must -have been affected by considering the nature of the rule violated -by the district court and the role that rule plays in the judicial pro- -cess.” Steiger, 99 F.4th at 1326. When, as here, “the proceeding is -sentencing, we ask whether the sentence would be different but for -erroneous sentencing enhancements or an inaccurate sentencing -guidelines range.” Id. (citation modified). Edwards thus “has the -difficult burden of showing there is a ‘reasonable probability’ that -he would have received a lighter sentence but for the error.” Jones, - -whole—i.e., ACCA applicability, writ large, based on those three convic- -tions—one could argue that he implicitly pleaded to the constituent parts, in- -cluding the applicability of the different-occasions requirement. But the gov- -ernment hasn’t pressed that argument, so we won’t pursue it. - -� - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 19 of 27 - -22-13963 - -Opinion of the Court - -19 - -743 F.3d at 830.7 In other words, a showing that Edwards would -have been entitled to a jury determination on the separate-occa- -sions inquiry is not, in and of itself, enough to establish that the -error affected his “substantial rights” under the plain-error frame- -work. Rather, to prevail, Edwards must further show a reasonable -probability that a jury would have concluded that he committed -the three predicate crimes on fewer than three occasions. - -As already explained, the district court enhanced Edwards’s -sentence under ACCA on the basis of three prior convictions: -(1) for a December 20, 1996, robbery with a deadly weapon and - -7 Both parties seem to misunderstand—albeit in different ways—how the sub- -stantial-rights prong operates in this case. For the first time at oral argument, -Edwards asserted that under Greer v. United States, 593 U.S. 508 (2021), he can -show prejudice if there is a reasonable probability that had he known about -the Erlinger rule, he wouldn’t have pleaded guilty and would have instead in- -sisted on going to trial. See Oral Arg. at 7:46, 28:22. That is incorrect. True, -when a defendant challenges the voluntariness of his plea, the reasonable- -probability standard is appropriately focused on whether he would have -pleaded guilty. Greer, 593 U.S. at 508. Accordingly, that formulation would -have governed Edwards’s voluntariness challenge had we found the first two -prongs of the plain-error standard satisfied. See supra at 7. But it doesn’t gov- -ern his challenge to the applicability of the ACCA enhancement. - -For its part, the government contends that the requisite reasonable -probability doesn’t exist here because Edwards hasn’t shown either “that the -three qualifying offenses were not committed on different occasions” or that -“the government would have been unable to prove that fact had he made a -timely objection.” Br. of Appellee at 31. But that formulation erects too high -a bar. To be sure, a defendant may seek to demonstrate prejudice in those -ways—and they would likely be persuasive—but he isn’t required to do so in -order to show an effect on his substantial rights. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 20 of 27 - -20 - -Opinion of the Court - -22-13963 - -aggravated battery, for which the arrest occurred on April 1, 1997; -(2) for an undated instance of drug trafficking, for which the arrest -occurred on December 28, 2010; and (3) for an October 1, 2010, -aggravated battery with a deadly weapon, for which the arrest oc- -curred on December 30, 2010. Edwards concedes that it is fair to -assume that the 1996 robbery was committed on a different occa- -sion than the other two—the nearly 14-year gap is hard to ignore. -But, he maintains, the government didn’t prove that the drug-traf- -ficking and aggravated-battery offenses, both with arrest dates in -2010, were committed on different occasions. - -The PSI contains no details about the drug-trafficking of- -fense other than the arrest date. Although “[t]he circumstances of -this arrest were requested from the Broward County Clerk of -Court,” no report was ever produced. Letter Br. of Appellant at 7, -Dkt. No. 37 (citing PSI ¶ 34). Edwards contends that, “based on the -lack of facts in the record concerning the 2010 drug offense,” a jury -likely would have found that the drug-trafficking and aggravated- -battery offenses were committed on the same occasion and, there- -fore, that he wasn’t subject to the ACCA enhancement.8 Id. at 8. -We cannot agree. - -8 Edwards’s other arguments improperly employ a harmless-error-like stand- -ard, pursuant to which the government bears the burden. See Letter Br. of -Appellant at 5 (“[B]ecause the government did not prove that his offenses oc- -curred on three separate occasions,” “Edwards is entitled to be resentenced -without the ACCA enhancement.”); see also id. at 6 (“[T]he ‘facts’ in the record -here—even the ‘unobjected to’ facts in the PSI—are insufficient for a jury to - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 21 of 27 - -22-13963 - -Opinion of the Court - -21 - -In our review for plain error, we consider “the whole rec- -ord” to determine “the effect of any error on [the defendant’s] sub- -stantial rights,” including “proceedings that both precede and post- -date the errors about which [the defendant] complains.” United -States v. Reed, 941 F.3d 1018, 1021 (11th Cir. 2019) (citation modi- -fied); see also Greer v. United States, 593 U.S. 503, 511 (2021) (explain- -ing that “an appellate court conducting plain-error review may -consider the entire record—not just the record from the particular -proceeding where the error occurred”). So we will examine the en- -tire record for evidence bearing on what the Wooden Court called -the “range of circumstances” that “may be relevant to identifying -episodes of criminal activity.” Wooden, 595 U.S. at 369; see also Er- -linger, 602 U.S. at 828 (noting the “intensely factual nature of this -inquiry”). The Court hoped that, “[f]or the most part,” applying -the different-occasions requirement would be “straightforward and -intuitive.” Wooden, 595 U.S. at 369. But for the tough cases, the -Court offered guidance to help determine whether offenses consti- -tute single or multiple episodes of criminal activity. In particular, -the Court emphasized considerations such as “[1] whether the de- -fendant’s past offenses were ‘committed close in time,’ [2] whether -they were committed near to or far from one another, and -[3] whether the offenses were ‘similar or intertwined’ in purpose -and character.” Erlinger, 602 U.S. at 828 (quoting Wooden, 595 U.S. -at 369). - -conclude beyond a reasonable doubt that the offenses occurred on different -occasions.”). - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 22 of 27 - -22 - -Opinion of the Court - -22-13963 - -Candidly, the record doesn’t shed much light on the first two -Wooden factors. With respect to timing, while the PSI establishes -that the aggravated-battery offense occurred on October 1, 2010, it -doesn’t specify the date of the drug-trafficking offense. Accord- -ingly, we can’t make any firm judgments about whether those two -crimes were “committed close in time, in an uninterrupted course -of conduct” or if they were “separated by substantial gaps in time -or significant intervening events.” Wooden, 595 U.S. at 369. The -record likewise reveals nothing about the “[p]roximity of location” -that would permit an assessment of whether the two crimes oc- -curred far from one another, which would of course make it “less -likely they are components of the same criminal event.” Id. - -That leaves the offenses’ “purpose and character”—and -there, we have more to go on. Edwards’s drug-related offense in- -volved trafficking in oxycodone, possession of cocaine with intent -to deliver or sell, and possession of drug paraphernalia. By con- -trast, his aggravated-battery arose out of domestic dispute with a -former romantic partner. According to the arrest affidavit— - -[O]n October 1, 2010, the defendant picked up a small -flowerpot, forcefully grabbed the victim by her hair, -and then smashed the pot against the victim’s head. -The victim sustained a laceration to the left side of -her head. The victim and the defendant had been in- -volved in a romantic relationship for two years and -lived together for seven months. The victim was -transported to the hospital where she received multi- -ple staples on her head. The defendant was later ar- -rested on December 30, 2010. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 23 of 27 - -22-13963 - -Opinion of the Court - -23 - -PSI ¶ 35. At least on the face of things, Edwards’s drug-trafficking -and aggravated battery offenses seem to have nothing in com- -mon—to be totally unrelated. And there’s certainly nothing in the -record to suggest that the two crimes “share a common scheme or -purpose,” such that they are likely to “compose one occasion.” -Wooden, 595 U.S. at 369. Accordingly, we think the third, “purpose -and character” factor significantly diminishes the probability that -the drug-trafficking and aggravated-battery offenses occurred on -the same occasion. - -Other facts likewise indicate, on balance, that those crimes -occurred on different occasions. First, Edwards was arrested for -the offenses two days apart—for the drug-trafficking crime on De- -cember 28, 2010, and for the aggravated-battery crime on Decem- -ber 30, 2010. Second, the criminal cases bear non-consecutive -docket numbers—indicating, at the very least, that the charges -were filed on different dates. And third, the two crimes were adju- -dicated almost three years apart—the aggravated battery on July -21, 2011, and the drug trafficking on April 29, 2014. We think it at -least relevant—and certainly not irrelevant—that the two offenses -were never combined into a single arrest or case, despite ample op- -portunities to do so. - -We also note the conspicuous absence of any evidence to -the contrary. Even on appeal, and with an opportunity to supple- -ment the record, Edwards hasn’t presented any evidence that -would demonstrate, or even strongly suggest, that the drug-traf- -ficking and aggravated-battery offenses were committed on the - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 24 of 27 - -24 - -Opinion of the Court - -22-13963 - -same occasion. He doesn’t assert, for instance, that the two crimes -took place on the same date, were committed in the same location, -or involved the same victims. Rather, Edwards simply says that -there isn’t sufficient evidence in the record for the government to -prove beyond a reasonable doubt that the ACCA enhancement ap- -plies. But on plain-error review, where the burden rests with Ed- -wards, and with the balance of the record evidence against him, -that’s not enough. - -To be sure, Edwards isn’t required to show by a preponder- -ance of the evidence that, but for the Erlinger error, the result -would have been different. See Dominguez Benitez, 542 U.S. at 83 -n.9 (“The reasonable-probability standard is not the same as, and -should not be confused with, a requirement that a defendant prove -by a preponderance of the evidence that but for error things would -have been different.”). And we acknowledge Edwards’s argument -that the lack of smoking-gun evidence about the locations and -dates of the crimes makes it possible that a jury could have found -the different-occasions requirement not satisfied. But a mere pos- -sibility isn’t a reasonable probability. See United States v. Marcus, -560 U.S. 258, 263 (2010) (deeming a standard pegged to “any possi- -bility, no matter how unlikely,” of a different outcome “irreconcilable -with our ‘plain error’ precedent” (citation modified)). And again, -the record evidence that we do have—the drug-trafficking and ag- -gravated-battery offenses’ divergent “purpose[s] and character[s],” -the different arrest and adjudication dates, the non-consecutive -docket numbers, etc.—indicates that they were indeed “committed -on occasions different from one another.” 18 U.S.C. § 924(e)(1). - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 25 of 27 - -22-13963 - -Opinion of the Court - -25 - -Finally, we note that Edwards hasn’t explained how there -could be a “reasonable probability”—more than a mere possibil- -ity—of a different outcome here. As we’ve observed, see supra at -24, Edwards hasn’t put forward any theory that pushes this case -over the reasonable-probability hurdle. Far from showing a rea- -sonable probability, Edwards hasn’t even speculated to how the Er- -linger error might have affected the outcome. For example, he -hasn’t argued that the aggravated-battery offense occurred in the -course of a dispute over the drug trafficking, such that a jury could -have found that those two offenses occurred on the same occasion. -And “[w]here the effect of an alleged error is so uncertain, a defend- -ant cannot meet his burden of showing that the error actually af- -fected his substantial rights.” Jones v. United States, 527 U.S. 373, -394–95 (1999). - -We therefore hold that Edwards hasn’t met his plain-error -burden to show that his “substantial rights” were affected, and we -affirm the district court’s decision to apply the ACCA enhance- -ment. - -C - -Finally, we address Edwards’s argument that Congress ex- -ceeded its Commerce Clause authority when it enacted 18 U.S.C. -§ 922(g).9 - -9 We review a district court’s determination regarding the constitutionality of -a statute for plain error when, as here, the issue wasn’t presented below. See -United States v. Valois, 915 F.3d 717, 729 n.7 (11th Cir. 2019). - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 26 of 27 - -26 - -Opinion of the Court - -22-13963 - -We have repeatedly held that § 922(g) is facially constitu- -tional under the Commerce Clause because it contains an express -jurisdictional requirement. See, e.g., United States v. Stancil, 4 F.4th -1193, 1200 (11th Cir. 2021); United States v. Jordan, 635 F.3d 1181, -1189 (11th Cir. 2011); United States v. Scott, 263 F.3d 1270, 1273–74 -(11th Cir. 2001) (“[T]he jurisdictional element of the statute, i.e., -the requirement that the felon ‘possess in or affecting commerce, -any firearm or ammunition,’ immunizes § 922(g)(1) from [a] facial -constitutional attack.”); United States v. Dupree, 258 F.3d 1258, 1259– -60 (11th Cir. 2001); United States v. McAllister, 77 F.3d 387, 390 (11th -Cir. 1996). - -We have also rejected as-applied challenges to § 922(g), -holding that the government demonstrates the required “minimal -nexus” to interstate commerce when it proves that the firearms -were manufactured outside the state where the offense took place -and, thus, necessarily traveled in interstate commerce. See, e.g., Jor- -dan, 635 F.3d at 1189 (holding that § 922(g) is constitutional as ap- -plied to “a defendant who possessed a firearm only intrastate” be- -cause the firearm had moved in interstate commerce); see also -United States v. Wright, 607 F.3d 708, 715–16 (11th Cir. 2010). Here, -Edwards admitted in his factual proffer that the firearm and ammu- -nition that he possessed in Florida were manufactured out of state -and had therefore moved in interstate commerce prior to his pos- -session of them. The government thus satisfied the minimal-nexus -requirement here. - - USCA11 Case: 22-13963 Document: 56-1 Date Filed: 06/27/2025 Page: 27 of 27 - -22-13963 - -Opinion of the Court - -27 - -Under the prior-panel-precedent rule, we’re obliged to fol- -low our existing decisions unless overruled by this Court en banc -or abrogated by the Supreme Court. United States v. White, 837 -F.3d 1225, 1228 (11th Cir. 2016). Because neither has occurred -here, we affirm the judgment of the district court that § 922(g) -doesn’t violate the Commerce Clause. - -III - -For the foregoing reasons, we AFFIRM the district court’s - -judgment. - -AFFIRMED. - diff --git a/q195/random_k2/question.json b/q195/random_k2/question.json index eb01f21c2f226bf2722b449b632158228807d797..13993614cc0bc392043aed29e788c5923ff52b5a 100644 --- a/q195/random_k2/question.json +++ b/q195/random_k2/question.json @@ -15,8 +15,8 @@ ], "original_filenames": [ "2021-Apple-Catalog.pdf", - "web_f77c9a6a0c40e754.pdf", - "web_5e3d260ea31a243f.pdf" + "web_a5b939c47368269e.pdf", + "web_0d779de9374a69ef.pdf" ], "modality": "markdown" } \ No newline at end of file diff --git a/q195/random_k2/random_1.md b/q195/random_k2/random_1.md index e93939cddbb26da056c6d4cb5d86060af787292b..868ced3168a91ab2e22fce714cd7ae12c9159778 100644 --- a/q195/random_k2/random_1.md +++ b/q195/random_k2/random_1.md @@ -1,979 +1,2046 @@ -INTRODUCING THE WINDOWS AZURE -PLATFORM - -DAVID CHAPPELL - -OCTOBER 2010 - -SPONSORED BY MICROSOFT CORPORATION - - CONTENTS - -An Overview of the Windows Azure Platform ....................................................................................... 3 - -Windows Azure ........................................................................................................................................ 4 - -SQL Azure ................................................................................................................................................. 6 - -Windows Azure AppFabric....................................................................................................................... 8 - -Windows Azure MarketPlace .................................................................................................................. 9 - -A Closer Look at the Technologies ....................................................................................................... 10 - -Windows Azure ...................................................................................................................................... 10 - -Compute ............................................................................................................................................ 11 - -Storage .............................................................................................................................................. 12 - -Fabric Controller ................................................................................................................................ 14 - -Content Delivery Network ................................................................................................................. 14 - -Connect ............................................................................................................................................. 15 - -SQL Azure ............................................................................................................................................... 15 - -Database ........................................................................................................................................... 15 - -Reporting ........................................................................................................................................... 17 - -Data Sync .......................................................................................................................................... 17 - -Windows Azure AppFabric..................................................................................................................... 19 - -Service Bus ........................................................................................................................................ 19 - -Access Control ................................................................................................................................... 21 - -Caching .............................................................................................................................................. 23 - -Windows Azure Marketplace ................................................................................................................ 24 - -Looking Ahead .................................................................................................................................... 26 - -Conclusions ......................................................................................................................................... 26 - -About the Author ................................................................................................................................ 27 - -2 - - AN OVERVIEW OF THE WINDOWS AZURE PLATFORM - -Using computers in the cloud can make lots of sense. Rather than buying and maintaining your own -machines, why not exploit the acres of Internet-accessible servers on offer today? For some applications, -both code and data might live in the cloud, where somebody else manages and maintains the systems -they use. Alternatively, applications that run inside an organization—on-premises applications—might -store data in the cloud or rely on other cloud infrastructure services. However it’s done, exploiting the -cloud’s capabilities can improve our world. - -But whether an application runs in the cloud, uses services provided by the cloud, or both, some kind of -application platform is required. Viewed broadly, an application platform can be thought of as anything -that provides developer-accessible services for creating applications or storing data. In the on-premises -Windows world, this includes technologies such as Windows Server and SQL Server. To let applications -exploit the cloud, a cloud application platform must also exist. - -This is exactly what the Microsoft Windows Azure platform provides. It’s a group of cloud technologies, -each providing a specific set of services to application developers. Figure 1 shows its components. - -Figure 1: The Windows Azure platform supports applications, data, and infrastructure in the cloud, -together with a cloud marketplace. - -The Windows Azure platform today has four parts: - - Windows Azure: A Windows environment for running applications and storing data on computers in - -Microsoft data centers. - -SQL Azure: Relational data services in the cloud based on SQL Server. - - Windows Azure AppFabric: Cloud-based infrastructure services for applications running in the cloud - -or on premises. - +Enabling Large Language Models to Generate Text with Citations +| | | TianyuGao | | | HowardYen | | JiatongYu | | | DanqiChen | | | +| --- | --- | --------- | --- | --- | --------- | --- | --------- | --- | --- | --------- | --- | --- | +DepartmentofComputerScience&PrincetonLanguageandIntelligence +PrincetonUniversity +{tianyug,hyen,jiatongy,danqic}@cs.princeton.edu +| | | Abstract | | | | | | Question | | | | | +| --- | --- | -------- | --- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | +When did the US break away from England? +Largelanguagemodels(LLMs)haveemerged +Model output +The US took the first step towards gaining independence from +asawidely-usedtoolforinformationseeking, +GB when it declared independence on July 2, 1776 (although the +buttheirgeneratedoutputsarepronetohalluci- LLM event is now commemorated on July 4, 1776, the date when the +3202 tcO 13 ]LC.sc[ 2v72641.5032:viXra +nation. Inthiswork,ouraimistoallowLLMs eveirteR Declaration of Independence was officially adopted by Congress) +[1][2]. The Treaty of Paris was later signed on September 3, 1783, +togeneratetextwithcitations,improvingtheir formally separating the United States from the British Empire [3]. +| factualcorrectnessandverifiability. | | | | | Existing | | | | | | | | +| ----------------------------------- | --- | --- | --- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +work mainly relies on commercial search en- [1] Decolonization of [2] Decolonization: … [3] American Revolu- +| | | | | | | | | the Americas: … in- | | It was officially adop- | tion: … The Treaty of | … | +| --- | --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | ----------------------- | ---------------------- | --- | +gines and human evaluation, making it chal- Corpus dependence from GB ted by Congress on Paris was signed +| | | | | | | | | on July 2, 1776 … | | July 4, 1776 … | September 3, 1783 ... | | +| ------- | ------------ | --- | --- | ------- | --------- | --- | --- | ----------------- | --- | -------------- | --------------------- | --- | +| lenging | to reproduce | | and | compare | different | | | | | | | | +modelingapproaches. WeproposeALCE,the Figure1: ThetasksetupofALCE.Givenaquestion, +firstbenchmarkforAutomaticLLMs’Citation the system generates text while providing citing pas- +| Evaluation. | | ALCE | collects | a diverse | set | of | | | | | | | +| ----------- | --- | ---- | -------- | --------- | --- | --- | ------------------------------- | --- | --- | --- | ---------------- | --- | +| | | | | | | | sagesfromalargeretrievalcorpus. | | | | Eachstatementmay | | +questions and retrieval corpora and requires containmultiplecitations(e.g.,[1][2]). +| building | end-to-end | | systems | to retrieve | | sup- | | | | | | | +| -------- | ---------- | --- | -------- | ----------- | --- | ---- | --- | --- | --- | --- | --- | --- | +| porting | evidence | and | generate | answers | | with | | | | | | | +toprovidecitationstooneorafewtextpassages +| citations. | Wedevelopautomaticmetricsalong | | | | | | | | | | | | +| ---------- | ------------------------------ | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +threedimensions—fluency,correctness,andci- foranystatementtheygenerate(Figure1). Incor- +tation quality—and demonstrate their strong poratingcitationsbringsseveralbenefits: (1)users +| correlationwithhumanjudgements. | | | | | Ourexper- | | | | | | | | +| ------------------------------- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | --- | --- | +caneasilyverifyLLMs’claimswiththeprovided +iments with state-of-the-art LLMs and novel citations;(2)LLMscangeneratetextthatfaithfully +promptingstrategiesshowthatcurrentsystems +| | | | | | | | follows | | cited passages, | | which has the promise | to | +| --- | --- | --- | --- | --- | --- | --- | ------- | --- | --------------- | --- | --------------------- | --- | +haveconsiderableroomforimprovement—For +improvecorrectnessandalleviatehallucination. +| example, | on | the ELI5 | dataset, | even | the | best | | | | | | | +| -------- | --- | -------- | -------- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | +modelslackcompletecitationsupport50%of Multiplecommercialsystemshaveadoptedthis +BingChat2andperplexity.ai3respondto +| thetime.Ouranalysesfurtherhighlightpromis- | | | | | | | paradigm: | | | | | | +| ------------------------------------------ | --- | --- | --- | --- | --- | --- | --------- | --- | --- | --- | --- | --- | +ingfuturedirections,includingdevelopingbet- userquestionsinnaturallanguagewithreferences +| ter retrievers, | | advancing | long-context | | LLMs, | | | | | | | | +| --------------- | --- | --------- | ------------ | --- | ----- | --- | ----------- | --- | --- | ----------------------------- | --- | --- | +| | | | | | | | toWebpages. | | | Nakanoetal.(2021);Menicketal. | | | +andimprovingtheabilitytosynthesizeinfor- +(2022)shareasimilarmotivation,buttheymainly +mationfrommultiplesources.1 +| | | | | | | | experiment | | with | commercial | search engines | and | +| --- | --- | --- | --- | --- | --- | --- | ---------- | --- | ---- | ---------- | -------------- | --- | +1 Introduction closed-sourcemodels,makingtheirresultsdifficult +| | | | | | | | toevaluate. | | Retrieval-augmentedLMs(Borgeaud | | | | +| --- | --- | --- | --- | --- | --- | --- | ----------- | --- | ------------------------------- | --- | --- | --- | +Largelanguagemodels(LLMs;Brownetal.,2020; +| | | | | | | | et | al., 2022; | Izacard | et | al., 2022) incorporate | re- | +| --- | --- | --- | --- | --- | --- | --- | --- | ---------- | ------- | --- | ---------------------- | --- | +OpenAI,2023)havegainedincreasingpopularity +| | | | | | | | trieved | | passages | during | both training | and infer- | +| ----------------------------- | --- | --- | --- | --------------- | --- | --- | ------- | --- | -------- | ------ | ------------- | ---------- | +| asatoolforinformationseeking. | | | | Whiletheygener- | | | | | | | | | +ence,butdonotguaranteefaithfulnesstoretrieved +ateengagingandcoherentresponses,theiroutputs +| | | | | | | | passagesorexplicitlyprovidecitations. | | | | | Addition- | +| ----------------------------------------- | ---------------- | --- | --- | ----- | ------- | ---- | ------------------------------------- | -------- | ------- | ------------- | ------------- | ---------- | +| are prone | to hallucination | | and | often | contain | fac- | | | | | | | +| | | | | | | | ally, | previous | studies | mostly | rely on human | eval- | +| tuallyincorrectinformation(Jietal.,2023). | | | | | | This | | | | | | | +| | | | | | | | uation | (Nakano | | et al., 2021; | Menick et | al., 2022; | +makesitharderforuserstotrustandverifyLLM- +Liuetal.,2023),whichisexpensiveanddifficultto +generatedoutputswithoutanysupportingevidence. +| | | | | | | | reproduce. | | Wearguethattheabsenceofautomated | | | | +| ------- | ----- | --- | ----- | ----- | ---------- | --- | ---------- | --- | -------------------------------- | --- | --- | --- | +| In this | work, | we | study | a new | generation | | | | | | | | +evaluationhinderstheadvancesofsuchsystems. +| paradigm | for LLMs, | in | which | we | require | LLMs | | | | | | | +| -------- | --------- | --- | ----- | --- | ------- | ---- | --- | --- | --- | --- | --- | --- | +2https://www.bing.com/new +| 1Our code | and | data are | available | at https://github. | | | | | | | | | +| ----------------------- | --- | -------- | --------- | ------------------ | --- | --- | --- | -------------------------- | --- | --- | --- | --- | +| com/princeton-nlp/ALCE. | | | | | | | | 3https://www.perplexity.ai | | | | | + +| Dataset | Corpus(#passages) | | Questiontype | Example | | | | | +| ------- | ----------------- | --- | ------------ | ------- | --- | --- | --- | --- | +ASQA Wikipedia(21M) Factoid(ambiguous) Q:WhendidtheUSbreakawayfromEngland? +A:TheUSdeclaredindependenceonJuly2,1776[1][2]...The +TreatyofPariswaslatersignedonSeptember3,1783[3]. +QAMPARI Wikipedia(21M) Factoid(list) Q:WhichfilmshaveGongLiasamemberoftheircast? +A:TheStoryofQiuJu[1],FarewellMyConcubine[2],The +MonkeyKing2[3],Mulan[3],SaturdayFiction[3]... +ELI5 Sphere(899M) Why/How/What Q:Howdostudentloansaffectgettingamortgage? +A:Studentloanscanaffectthedebttoincomeratio[1],whichis +akeyfactorindeterminingtheamountthat...[2][3] +Table1: ThethreedatasetsusedinourALCEbenchmark. Thesedatasetscoverawiderangeofquestiontypesand +thecorrespondingcorporaspanfromWikipediatoWeb-scaledocumentcollection. +WepresentALCE,thefirstreproduciblebench- withpost-hoccitingachievesgoodcorrectnessbut +mark for automatically evaluating LLMs’ gener- muchworsecitationquality;(2)althoughinterac- +ations with citations. ALCE assumes a natural- tiveretrievalapproaches(Yaoetal.,2023;Schick +language question and a retrieval corpus, and re- et al., 2023) offer more flexibility in when/what +quiresbuildingend-to-endsystemstoretrieverel- to retrieve, they do not improve the performance +evant passages from the corpus, generate a re- on this challenging benchmark; (3) summarizing +sponsetothequestion,andcitecorrespondingsup- the retrieved passages in a shorter text improves +porting passages. We compile three datasets that correctnessbutnotcitationquality;(4)reranking +cover different types of questions and corpora— multiple generations boosts citation quality mea- +ASQA (Stelmakh et al., 2022), QAMPARI (Ru- suredbyhumanevaluation;(5)incorporatingmore +binetal.,2022),andELI5(Fanetal.,2019)—as retrieved passages in context does not help Chat- +showninTable1. Differentfrompreviousbench- GPTbutimprovesGPT-4performance. +marks(Leeetal.,2019;Bohnetetal.,2022),ALCE Our extensive analyses highlight three major +evaluates long-text generation, focusing on auto- challengesofbuildingLLMstogeneratetextwith +matically evaluating citation quality, and allows citations: (1)theretrievalqualityiscrucialtothe +citingmultiplepassagesforindividualstatements. +finalperformanceandhassubstantialroomforim- +provement;(2)LLMs’limitedcontextwindowre- +Wedesignautomaticevaluationmethodsinthree +dimensions: fluency, correctness, and citation strictsthenumberofpassagestheycanincorporate; +quality. Specifically, we use MAUVE (Pillutla (3) current LLMs struggle to synthesize multiple +documentsincontextwithoutbeingdistractedby +| etal., 2021)to | measurefluency, | | propose tailored | | | | | | +| -------------- | --------------- | --- | ---------------- | --- | --- | --- | --- | --- | +irrelevantones,althoughbetterinstructiontuning +| correctness | metrics | for each | dataset, and adopt | a | | | | | +| ----------- | ------- | -------- | ------------------ | --- | --- | --- | --- | --- | +naturallanguageinference(NLI)model(Honovich bringssignificantimprovement. Thesechallenges +etal.,2022)tomeasurecitationquality. Weshow- posepromisingresearchdirectionsfordeveloping +bettersystemsintegratingretrievalandLLMs. +casehowthethreedimensionstogethercontribute +toarobustevaluation,preventingsystemsfromex- +| | | | | 2 TaskSetupandDatasets | | | | | +| ------------------ | --------------------------- | --- | -------------------- | ---------------------- | ------------- | ----------- | ------- | ----- | +| ploitingshortcuts. | Additionally,weconducthuman | | | | | | | | +| evaluation | and demonstrate | | a strong correlation | | | | | | +| | | | | Our task | is formalized | as follows: | Given a | query | +withourautomaticmetrics. q and a corpus of text passages D, the system +Weexperimentonmultiplesystemswithstate- is required to return an output S, which consists +| of-the-art | LLMs and | retrievers | and also propose | | | | | | +| --------------------------------------------- | -------- | ---------- | ---------------- | --------------------- | --- | -------- | ------------------ | --- | +| | | | | of n statements | s , | ..., s , | and each statement | s | +| | | | | | 1 | n | | i | +| novelpromptingstrategiestosynthesizeretrieved | | | | | | | ,...}4,where | | +| | | | | citesalistofpassagesC | | i = {c | i,1 ,c i,2 | | +textintotextgeneration. Althoughallsystemsare c ∈ D. Inthiswork,wesegmentLLMs’output +i,j +capableofprovidingfluentandcoherentresponses, into statements by sentence boundaries.5 While +thereremainssubstantialroomforimprovementin +LLMsmayincludesentencesthatdonotrequirea +termsofcorrectnessandcitationquality: Forexam- citation,suchas“I’mhappytohelp”,weobserve +ple,ontheELI5dataset,around50%generations thatalmostallsentencesthatLLMsoutputprovide +ofourChatGPTandGPT-4baselinesarenotfully +4Inpractice,weallowatmost3citationsforeachstate- +| supportedbythecitedpassages. | | | Additionally,we | | | | | | +| ---------------------------- | --- | --- | --------------- | --- | --- | --- | --- | --- | +mentasmorecitationsusuallydonothelp. +| find that | (1) a closed-book | model | (generating | an- | | | | | +| --------- | ----------------- | ----- | ----------- | --- | --- | --- | --- | --- | +5QAMPARIrequiresalistastheanswer,andwechoose +swerswithoutaccessinganyretrieveddocuments) +eachentityinthegeneratedlistasastatement. + +| | | | | | | | | 3 AutomaticEvaluation | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | --- | +valuableinformationandrequirecitations,similar +| tofindingsinLiuetal.(2023). | | | | | Inthiswork,cita- | | | | | | | +| --------------------------- | --- | --- | --- | --- | ---------------- | --- | --- | ------------- | -------- | ------------- | --------- | +| | | | | | | | | Our benchmark | measures | the following | three di- | +tionsareenclosedbyboxbracketssuchas[1][2]. +mensionsofsystemresponses: +| WedividethecorpusD | | | | into100-wordpassages | | | | | | | | +| ------------------ | --- | --- | --- | -------------------- | --- | --- | --- | ---------- | -------------------------------- | --- | --- | +| | | | | | | | | • Fluency: | whetherthemodel’sgeneratedtextis | | | +followingpreviousworksonopen-domainquestion +fluentandcoherent. +| answering | | (Karpukhin | | et al., | 2020; | Petroni | et al., | | | | | +| ------------- | --- | ---------- | ---------- | ------- | -------- | --------- | ------- | -------------- | ------- | ---------- | ----------- | +| | | | | | | | | • Correctness: | whether | the answer | is accurate | +| 2021;Piktuset | | | al.,2021), | in | contrast | tocommer- | | | | | | +andcoversallaspectsofinterest. +cialsystemslikeBingChat,whichciteentireWeb +| | | | | | | | | • Citation | quality: whether | the answer | is well | +| ------ | --- | ---- | -------- | -------- | --- | ------- | ----- | ---------- | ---------------- | ---------- | ------- | +| pages. | We | take | 100-word | passages | | because | it is | | | | | +supportedbythecitedpassagesandnoirrelevant +| easier | for | humans | to verify, | | and allows | for | more | | | | | +| -------------------------------------------- | --- | ------ | ---------- | --- | ---------- | --- | ---- | ----------------- | --- | --- | --- | +| retrievedpassagestofitinLLMs’limitedcontext. | | | | | | | | passagesarecited. | | | | +Inthefollowing,wepresentautomaticmetricsfor +WechooseQAdatasetssothat(1)theycontain +eachdimensionanddiscusswhythecombination +| factual | questions, | | in which | references | | are | impor- | | | | | +| ------- | ---------- | --- | -------- | ---------- | --- | --- | ------ | --- | --- | --- | --- | +ofthethreemetricsprovidesarobustevaluation. +| tant; | (2) questions | | require | long-text | | answers | that | | | | | +| --------------------------------------------- | ------------- | --- | ------- | --------- | --- | -------- | ---- | ----------- | --- | --- | --- | +| covermultipleaspects;(3)answeringthequestions | | | | | | | | 3.1 Fluency | | | | +| requiressynthesizingmultiplesources. | | | | | | Weselect | | | | | | +WeuseMAUVE(Pillutlaetal.,2021)toevaluate +threedatasets(Table1)andintroducethembelow. +thefluencyoftheoutput(§C).WedeployMAUVE +See§Bforadditionalstatistics. +forASQAandELI5andomititforQAMPARI,as +ASQA(Stelmakhetal.,2022)isalong-formfac- +QAMPARIonlyrequiresalistofshortanswersas +| toiddataset. | | AsshowninFigure1,eachquestionis | | | | | | | | | | +| ------------ | --- | ------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +theresponseandLLMsconsistentlyadheretothe +anambiguousquestionfromAmbigQA(Minetal., +| | | | | | | | | formatinourexperiments. | | AsMAUVEissensitive | | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | ------------------ | --- | +2020)thatrequiresmultipleshortanswerstocover +tooutputlengthandtextstyle,andmostLLMsare +differentaspects,andthedatasetprovidesalong- +capableofproducingfluenttext,wemainlyemploy +| form | answer | that | covers | all short | answers. | | Since | | | | | +| ---- | ------ | ---- | ------ | --------- | -------- | --- | ----- | --- | --- | --- | --- | +itasasanitycheckaslongastheMAUVEscores +mostquestionscanbeansweredbyWikipedia,we +arehighenough. +usethe2018-12-20WikipediasnapshotasD. +| QAMPARI(Rubinetal.,2022)isafactoidQA | | | | | | | | 3.2 Correctness | | | | +| ------------------------------------ | --- | --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --- | +datasetconstructedfromWikipedia,wherethean- Ourobjectiveistomeasuretheinformativenessand +swerisalistofentitiesthataredrawnfromdiffer- utilityofthegenerationtothequestion. Liuetal. +| entpassages. | | SameasASQA,weusethe2018-12- | | | | | | | | | | +| ------------ | --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +(2023)proposetodirectlyevaluateperceivedutility +20Wikipediaasthecorpus. byhumans,aprocessdifficulttoautomate. There- +ELI5 (Fan et al., 2019) is a long-form QA fore,weusecorrectness—whethertheresponseis +datasetbuiltontheRedditforum“ExplainLikeI’m accuratecomparedtoagroundtruthanswer—asa +| Five”.6 | | | | | | | | proxy. Evaluatingthecorrectnessoflong-formgen- | | | | +| ------- | ---- | ---- | --------- | --- | ---------------- | --- | --- | ---------------------------------------------- | --- | --- | --- | +| | Most | ELI5 | questions | | are how/why/what | | | | | | | +questions that require long answers and multiple erationisachallengingtask(Krishnaetal.,2021), +passagesasevidence. Duetothediversetopicsdis- and we describe our strategy for each dataset be- +cussedinthequestions,weuseSphere(Piktusetal., low. Figure2illustratesthemetricsandweinclude +2021)—afilteredversionofCommonCrawl7—as additionalimplementationdetailsin§C. +the corpus. The ELI5 dataset is widely used in For ASQA, we follow Stelmakh et al. (2022) +relatedworkduetoitschallengingnature(Nakano andcalculatetherecallofcorrectshortanswersby +checkingwhethertheshortanswers(providedby +etal.,2021;Menicketal.,2022;Liuetal.,2023). +thedataset)areexactsubstringsofthegeneration +| We | randomly | | select | 1,000 | examples | from | the | | | | | +| --- | -------- | --- | ------ | ----- | -------- | ---- | --- | --- | --- | --- | --- | +(exactmatchrecall;EMrecall). +| development | | set | of each | dataset | for | ALCE. | Our | | | | | +| ----------- | --- | --- | ------- | ------- | --- | ----- | --- | --- | --- | --- | --- | +benchmarkprimarilyassessesthecitationcapabili- ForQAMPARI,wefollowRubinetal.(2022) +andcalculatetheprecisionandrecallofthemodel +tiesofexistingLLMsanddoesnotprovidetraining +prediction,bycheckingtheexactmatchtothegold +| data, | as there | are | no available | | examples | | that pro- | | | | | +| ----- | -------- | --- | ------------ | --- | -------- | --- | --------- | --- | --- | --- | --- | +videsupervisionforcitationsinthesedatasets. answer list. We add one additional adjustment: +| | | | | | | | | considering | that users often | want to know | only a | +| --- | --- | --- | --- | --- | --- | --- | --- | ----------- | ---------------- | ------------ | ------ | +fewexampleanswersofthequestion,ourevalua- +6https://www.reddit.com/r/explainlikeimfive/ +tionconsidersrecalltobe100%iftheprediction +7https://commoncrawl.org.WealsofilteroutanyWeb +includesatleast5correctanswers(recall-5). +pagesfromReddit. + +Question +ASQA Exact Match Recall +When did the US break away from England? +Question +| When did the US break away from England? | | | | | | Model output | | | | | | +| ---------------------------------------- | --- | --- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | +{statement 1} [1][2]. {statement 2} [3].{statement 3} [2][4][5]. +Model output +… it declared independence on July 3, 1776 … The Treaty of +Paris was signed on September 3, 1783 … +Citation Recall +Short answers (from the dataset) +| | | | | | String exact match | Recall = 1 if the concatenation of all cited passages fully supports the segment. | | | | | | +| --- | --- | --- | --- | --- | ------------------ | --------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | +July 2, 1776 July 4, 1776 September 3, 1783 Recall=33.3% We use an NLI model to determine “fully support”. +{statement 1} [1][2] +| | | | | | | | Premise: passage [1][2] | | NLI | | | +| --- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | +QAMPARI Precision, Recall Hypothesis: {statement 1} model “entailment” +{statement 2} [3] +Question +| Which films have Gong Li as a member of their cast? | | | | | | {statement 3} [2][4][5] | | | | | | +| --------------------------------------------------- | --- | --- | --- | --- | --- | --------------------------- | --- | --- | --- | --- | --- | +Model output For this question, citation recall = 2 / 3 = 66% +Saturday Fiction, The Great Wall, Mulan, Shadow. +| Short answers (from the dataset) | | | | String exact match | | | | | | | | +| -------------------------------- | --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | +Citation Precision +| The Story of Qiu Ju | | Mulan Saturday Fiction | | Recall=66.6% | | | | | | | | +| ------------------- | --- | ---------------------- | --- | ------------- | --- | --------------------------------------------------------------------------------- | --- | --- | --- | --- | --- | +| | | | | Precision=50% | | Detect “irrelevant” citation: one citation alone does not support the claim, and | | | | | | +removing it does not affect other citations combined to support the claim. +ELI5 Claim Recall +| Question | | | | | | {statement 1} [1][2] : | | | | | | +| -------- | --- | --- | --- | --- | --- | ---------------------------- | ------------------------------------ | --- | --- | --- | --- | +| | | | | | | | [1] [2] Precision = 1 | | | | | +How do student loans affect getting a mortgage? +[2] [ 1 ] Precision = 1 +Model output +They do not appear in credit history and do not affect debt to {statement 2} [3] If recall = 0, then precision = 0 +income ratio. Thus they do not affect getting a mortgage. +{statement 3} [2][4][5] : +| Claims (generated by text-davinci-003 based on gold answers) | | | | | | | | | | | | +| ------------------------------------------------------------ | --- | --- | --- | --- | --- | ------------------------- | ------------------------- | ----------------------------------- | --- | --- | --- | +| | | | | | NLI | | [2] [4][5] | [2] is “irrelevant’’. Precision = 0 | | | | +Student loan does not appear in credit history. Claim recall=33.3% [4] [2][5] Precision = 1 +Student loan can affect the debt to income ratio. [5] [2][4] Precision = 1 +Debt to income ratio affects mortgage applications. +For this question, citation precision = 4 / 6 = 66% +Figure2: Evaluationofcorrectness(detailsin§3.2). Figure3: Evaluationofcitationquality(detailsin§3.3). +WeuseanNLImodeltoverifywhetherastatementis +UnlikeASQAandQAMPARI,theELI5dataset +supportedbyitscitations. +| does not | provide | short | entity | answers. | Fan et al. | | | | | | | +| ----------- | --------- | ----------- | --------------- | ------------- | ---------- | --------- | ------------------------------------ | --- | --- | --- | --- | +| (2019) | use ROUGE | | for evaluation, | | which does | | | | | | | +| | | | | | | passages. | Figure3providesanillustratedexample. | | | | | +| not reflect | the | correctness | | well (Krishna | et al., | | | | | | | +WeusetheNLImodelTRUE(Honovichetal., +2021; §A). Inspired by works in summariza- 2022)againtoautomaticallyexaminewhetherthe +tion evaluation (Zhang and Bansal, 2021; Kamoi citedpassagesentailthemodelgeneration. Wecon- +| et al., 2023; | Wang | et | al., 2020), | we | use Instruct- | | | | | | | +| ------------- | ---- | --- | ----------- | --- | ------------- | --- | --- | --- | --- | --- | --- | +ducthumanevaluation(§6)todemonstratestrong +GPT (text-davinci-003; Ouyang et al., 2022) humancorrelationofourmetric. +| to generate | three | “sub-claims”. | | Then | we use | | | | | | | +| ----------- | ----- | ------------- | --- | ---- | ------ | --- | --- | --- | --- | --- | --- | +TRUE8 (Honovich et al., 2022), a T5-11B (Raf- Citationrecall. Wecalculatethecitationrecallof +eachstatement(0or1)andaverageoverallstate- +feletal.,2020)modelfine-tunedonacollectionof +| | | | | | | mentsinthemodelresponse. | | | Foreachstatements | | , | +| --------------------------------------------- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | ----------------- | --- | --- | +| naturallanguageinference(NLI)datasets,tocheck | | | | | | | | | | | i | +itscitationrecallis1ifandonlyifthereisatleast +| whether | the model | output | entails | the | sub-claims | | | | | | | +| ------- | --------- | ------ | ------- | --- | ---------- | --- | --- | --- | --- | --- | --- | +(claim recall). TRUE targets factual correctness one citation (C i ̸= ∅) and ϕ(concat(C i ),s i ) = 1, +| | | | | | | where ϕ(premise,hypothesis) | | | is the | NLI model | | +| ---------------------------------------- | ---- | ------- | -------- | ----- | ---------- | --------------------------- | -------- | ------- | ------- | ----------- | --- | +| and has | been | used by | previous | works | in similar | | | | | | | +| | | | | | | that outputs | 1 if the | premise | entails | the hypoth- | | +| context(Bohnetetal.,2022;Gaoetal.,2023). | | | | | We | | | | | | | +demonstratethatclaimrecallprovidesamoreac- esis,and0otherwise;concat(C i )concatenatesall +| | | | | | | passages | in C together | (details | in §C). | The | NLI | +| --------------------------------------------- | --- | --- | --- | --- | --- | -------- | ------------- | -------- | ------- | --- | --- | +| curatemeasureofcorrectnessthanexistingmetrics | | | | | | | i | | | | | +evaluationisinaccordancewiththeattributableto +(moredetailsin§A). +identifiedsources(AIS)framework(Rashkinetal., +| | | | | | | ϕ(concat(C | ),s | ) = | 1impliesthats | | | +| ------------------------------------------- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ------------- | -------- | --- | +| 3.3 CitationQuality | | | | | | 2023): | i | i | | i istrue | | +| | | | | | | basedsolelyonconcat(C | | ). | | | | +| Weevaluatecitationqualitiesusingtwometrics: | | | | | (1) | | | i | | | | +citationrecall,whichdeterminesiftheoutputisen- Citationprecision. Ourcitationprecisionevalua- +tirelysupportedbycitedpassages,and(2)citation tiondetectscitationsthatareirrelevant,butitdoes +precision,whichidentifiesanyirrelevantcitations. +| | | | | | | notrequirecitingaminimalset. | | | Wefollowthisde- | | | +| --- | --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --------------- | --- | --- | +Althoughweprioritizecitationrecallasitentails +signbecausehumanwritingoftencitesredundant +a well-supported and truthful answer, enhancing sourcestoenhancecredibility;humanreadersmay +precisioniscrucialforbetterusersatisfaction,re- alsoappreciatemultiplecitations,especiallywhen +| ducing | the need | for human | | review of | extraneous | | | | | | | +| ------ | -------- | --------- | --- | --------- | ---------- | --- | --- | --- | --- | --- | --- | +itpertainstocriticalclaimssuchasmedicaladvice. +| | | | | | | We calculate | the | citation | precision | for each | ci- | +| --- | --- | --- | --- | --- | --- | ------------ | --- | -------- | --------- | -------- | --- | +8https://huggingface.co/google/t5_xxl_true_ +tation(0or1)andaverageoverallcitationsinthe +nli_mixture.Detailsin§C. + +Instruction: Write an accurate, engaging, and et al., 2022) and DPR (Karpukhin et al., 2020); +concise answer for ... we use BM25 for Sphere. For each question, we +| | | | | retrievethetop-100passages. | | | | | | +| --------- | ---------- | --- | --------------- | --------------- | --- | --- | --------------------------- | --- | --- | --- | --- | --- | +| Document | [1](Title: | | American | Decolonization) | | | | | | | | | +| ... | | | | | | | 4.2 Synthesis | | | | | | +| Document | [2](Title: | | Decolonization) | | ... | | | | | | | | +Document [3](Title: American Revolution) ... We focus on how to prompt an LLM to interact +... +| | | | | | | | with the | retriever, | and synthesize | | and cite | the ev- | +| --------- | ---- | --- | -------- | ---- | ---- | -------- | --------------- | ---------- | -------------- | -------- | ------------ | ------- | +| Question: | When | did | US break | away | from | England? | | | | | | | +| | | | | | | | idence (without | | fine-tuning | internal | parameters). | | +Answer: The United States took the first step One noteworthy challenge is that existing LLMs +| towards | gaining | independence | | ... | [1][2]. | The | | | | | | | +| ------- | ------- | ------------ | --- | --- | ------- | --- | --- | --- | --- | --- | --- | --- | +allhavelimitedcontextwindowandthuscanonly +| Treaty | of Paris | was | later | signed | ... | [3]. | | | | | | | +| ------ | -------- | --- | ----- | ------ | --- | ---- | --- | --- | --- | --- | --- | --- | +fitahandfulofpassages. +| Table2: | AnexampleofourVANILLAmethod. | | | | | Differ- | | | | | | | +| ------------------------------ | ---------------------------- | --- | --- | --- | -------------- | ------- | -------- | ------------------------------ | --- | --- | --- | --- | +| | | | | | | | VANILLA. | Wesimplyprovidethemodelwiththe | | | | | +| entcolorsrepresentprompt,model | | | | | generation,and | | | | | | | | +top-k9 +. Wealsoprovidetwoin-contextdemonstra- passagesandinstructthemodeltociteac- +| | | | | | | | cordingly(Table2). | | Wealsousein-contextlearn- | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------ | --- | ------------------------- | --- | --- | --- | +tionsbeforethetestexample. +ing(Brownetal.,2020)andprependtwodemon- +| response. | We | first | define | if a citation | | is “irrele- | | | | | | | +| ----------------------------- | --- | ----- | ------ | ----------------------- | --- | ----------- | ---------- | ---------------------------------- | --- | --- | --- | --- | +| | | | | | | | strations. | ThecompleteinstructionisinTable23. | | | | | +| vant”. Intuitively,acitationc | | | | i,j is“irrelevant”if(a) | | | | | | | | | +c itself cannot support s and (b) removing c SUMM/SNIPPET. Witha4Kcontextwindow,we +| i,j | | | | i | | i,j | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | --- | ---------- | --------- | --- | +| | | | | | | | canatmostsafelyfitk | | = | 5passages. | Asshownin | | +doesnotaffecttherestofthecitationstosupport +Figure4,top-5retrievedpassagescanonlycover +| s . Formally,c | | is“irrelevant”ifandonlyif | | | | | | | | | | | +| -------------- | --- | ------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| i | | i,j | | | | | | | | | | | +56.8%percentoftheanswersinASQA. +| (a) ϕ(c | ,s | ) = 0,AND | | | | | | | | | | | +| ------- | ----- | --------- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | +| | i,j i | | | | | | Totacklethislimitation,weproposetoprovide | | | | | | +(b) ϕ(concat(C \{c }),s ) = 1. summaries or snippets of passages instead of the +| | | i | i,j | i | | | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +fulltext(summariesareabstractivebutsnippetsare +| c has | a precision | | of 1 if | s has | recall=1 | and | | | | | | | +| ----- | --------------- | --- | ------- | ------- | -------- | --- | ------------------- | --- | --------------------- | --- | --- | --- | +| i,j | | | | i | | | | | | | | | +| | | | | | | | spansfrompassages). | | Weacquiresummariesand | | | | +| c is | not irrelevant. | | For | example | (Figure | 3), | | | | | | | +i,j +snippetsbypromptingChatGPTwithinstructions +| when s | 3 cites | three | references | | [2][4][5] | and | | | | | | | +| ----------------------------------- | ------- | ----- | ---------- | --- | --------- | -------- | -------------------------- | --- | --- | ------------- | --- | --- | +| | | | | | | | (promptsinTable25and26).10 | | | Thenwereplace | | | +| recall=1,[2]is“irrelevant”ifϕ([2],s | | | | | | ) = 0and | | | | | | | +3 +| | | | | | | | allpassageswithsummaries/snippets. | | | | Summaries | | +| ---------- | --- | ------ | ------------------------ | --- | --- | --- | ---------------------------------- | --- | --- | --- | --------- | --- | +| ϕ([4][5],s | | ) = 1. | Forcondition(b)towork,we | | | | | | | | | | +3 +orsnippetssignificantlyreducethepassagelength, +| set recall=1 | | as a prerequisite | | for | precision= | 1. | | | | | | | +| ------------ | ---- | ----------------- | --------- | --- | ---------- | -------- | -------- | -------- | -------- | ------ | ------- | ----- | +| | | | | | | | allowing | for more | passages | to fit | in: for | ASQA, | +| Note that | this | algorithm | overlooks | | the | scenario | | | | | | | +theyreducepassagelengthby6×onaverage. +whenonecitationpartiallysupportsthestatement. +| | | | | | | | Though | SUMM/SNIPPET | | allows | for more | re- | +| --- | --- | --- | --- | --- | --- | --- | ------ | ------------ | --- | ------ | -------- | --- | +Wediscussthedetailsin§E. +| | | | | | | | trievedpassages,theyarelossycompressions. | | | | | To | +| --- | --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | +3.4 ALCEisRobusttoShortcutCases +| | | | | | | | alleviatethisproblem,wepropose | | | | INTERACT,an | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | ----------- | --- | +interactivepromptingschemetoallowthemodelto +WeshowcasehowtheALCEevaluationisrobust +| | | | | | | | checkthefulltextofcertainpassages. | | | | Ateachstep, | | +| --- | --- | --- | --- | --- | --- | --- | ---------------------------------- | --- | --- | --- | ----------- | --- | +totwopossibleshortcutsin§D:(1)usingthetop-1 +retrievedpassageastheresponseandcitingitself, the model can execute one of three actions: (1) +| | | | | | | | “Check: | Document | [1][2]”tocheckthefulltext | | | | +| -------- | ----- | --------- | ---- | -------------- | --- | --------- | ---------------------------------------- | -------- | ------------------------- | --- | --- | --- | +| and (2) | using | the first | two | sentences | of | the top-1 | | | | | | | +| | | | | | | | ofthecorrespondingdocuments;(2)“Output:” | | | | | to | +| passage. | Both | cases | have | almost-perfect | | citation | | | | | | | +scores,but(1)haslowfluencyduetoitsunnaturally outputastatementoftheanswer;(3)“End.” toend +| | | | | | | | thegeneration. | §Cprovidesmoredetails. | | | | | +| --------------------------------- | --- | --- | --- | --- | --- | ------ | -------------- | ---------------------- | --- | --- | --- | --- | +| longlengthcomparedtohumananswers, | | | | | | and(2) | | | | | | | +haslowcorrectnessduetolowcoverage. +| | | | | | | | INLINESEARCH. | | The | above methods | | all dis- | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ------- | --- | -------------- | --- | -------- | +| | | | | | | | play retrieval | results | at | the beginning. | In | INLI- | +4 Modeling +NESEARCH,weallowLLMstocall“search”dur- +In this section, we discuss three major modeling ingthegenerationprocess(Yaoetal.,2023;Press +componentsforanALCEsystem—retrieval,syn- et al., 2022; Jiang et al., 2023). At each step, the +thesis,andpost-editing. modelcanexecuteoneofthreeactions: “Search: +| 4.1 Retrieval | | | | | | | 9Wecanfitatmostk | | | | | | +| ------------- | --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ---------------------- | --- | --- | +| | | | | | | | | | = | 3formodelswith2Kwindow | | | +andatmostk=5formodelswith4Kcontextwindow. +| Weexploresimple,off-the-shelfretrievers. | | | | | | Weuse | | | | | | | +| ---------------------------------------- | --- | --- | --- | --- | --- | ----- | --- | --- | --- | --- | --- | --- | +10WealsoqueryChatGPTwhetherthepassageisrelevant +denseretrieversforWikipedia,includingGTR(Ni +tothequestion,andfilteroutpassagesthatare“irrelevant”. + +| Instruction: | | ... | | | | | | Fluency | Correct. | Citation | | +| ------------ | ---- | -------- | ------ | --------- | -------- | --- | --- | ------- | -------- | -------- | ----- | +| | | | | | (MAUVE) | (EMRec.) | Rec. | Prec. | +| Question: | When | did US | break | away from | England? | | | | ChatGPT | | | +Search: Declaration of Independence VANILLA(5-psg) 66.6 40.4 73.6 72.5 + w/RERANK 77.0 40.2 84.8 81.6 +Document [1](Title: ...) ... SUMM(10-psg) 70.0 43.3 68.9 61.8 +Output: The United States ... [1]. w/INTERACT 69.0 39.1 73.4 66.5 + SNIPPET(10-psg) 69.8 41.4 65.3 57.4 +| Search: | Treaty | of Paris | | | | INLINESEARCH | | | 58.7 32.4 | 58.3 | 58.2 | +| ------- | ------ | -------- | --- | --- | --- | ------------ | --- | --- | --------- | ---- | ---- | + CLOSEDBOOK 52.7 38.3 26.7 26.7 +| Document | [3](Title: | | ...) ... | | | | GPT-4(VANILLAprompting) | | | | | +| -------- | ---------- | --- | -------- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | +Output: The Treaty of Paris ... [3]. GPT-4(5-psg) 67.1 41.3 68.5 75.6 + GPT-4(20-psg) 64.9 44.4 73.0 76.5 +End. +LLaMA(VANILLAprompting) +| | | | | | | LLaMA-13B(3-psg) | | | 68.4 26.9 | 10.6 | 15.4 | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | --------- | ---- | ---- | +Table3: AnexampleofINLINESEARCH. Vicuna-13B(3-psg) 82.6 31.9 51.1 50.1 +| | | | | | | Chat-13B(5-psg) | | | 72.4 35.2 | 38.4 | 39.4 | +| --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --------- | ---- | ---- | +{query}”tosearchamongthetop-100passages11 +| | | | | | | Chat-70B(5-psg) | | | 88.3 41.5 | 62.9 | 61.3 | +| --- | --- | --- | --- | --- | --- | --------------- | --- | --- | --------- | ---- | ---- | +byusingGTR;the“Output”and“End”actionsare +| | | | | | | Table4: | ExperimentsonASQA.For | | | CLOSEDBOOK, | | +| --------- | --------- | --- | ---------------------- | --- | --- | ------- | --------------------- | --- | --- | ----------- | --- | +| thesameas | INTERACT. | | Foreach“Search”action, | | | | | | | | | +wedisplaythebestretrievedpassageinthecontext. weusePOSTCITEtogetcitations. k-psg: puttingtop- +| | | | | | | k passages | from | the retrieval | results | into the | context. | +| ----------- | --- | ------- | ----- | ---------- | ------- | ---------- | ---- | ------------- | ------- | -------- | -------- | +| The passage | is | removed | after | one action | to save | | | | | | | +Chat-13BandChat-70BrefertoLLaMA-2-Chat. +| contextspace. | | Table3showsanexample. | | | | | | | | | | +| ------------- | --- | --------------------- | ---- | ------------ | ------- | ---------------------- | --- | --- | ------------------ | --- | --- | +| | | | | | | Oasst(Köpfetal.,2023). | | | Theyallhavea2Kcon- | | | +| CLOSEDBOOK. | | We | also | add a simple | closed- | | | | | | | +bookbaseline,wherethemodelisonlyprompted textwindow. WeuseshortinstructionsforLLaMA +| | | | | | | (Table | 24) to | save context | budget. | Additionally, | | +| --- | --- | --- | --- | --- | --- | ------ | ------ | ------------ | ------- | ------------- | --- | +withtheinstructionandthequestion,withoutany +wetestLLaMA-2-Chat,whichwerealsotrainedto +| retrieved | passages | provided. | | Consequently, | this | | | | | | | +| --------- | -------- | --------- | --- | ------------- | ---- | --- | --- | --- | --- | --- | --- | +variantdoesnotciteanyevidences. followinstructions(Touvronetal.,2023b). These +modelshaveacontextwindowof4Ktokens,which +| 4.3 Post-editing | | | | | | allowsfor5passagesperquestion. | | | | | | +| ---------------- | --- | --- | --- | --- | --- | ------------------------------ | --- | --- | --- | --- | --- | +Inthissectionwediscusstwostrategiesforrefining +5.1 MainResults +theoutputtofurtherimproveitsquality. +| | | | | | | We present | the | main | results on three | datasets | in | +| ------- | -------- | --------- | ------ | ---------- | ------------ | --------------------------------------------- | --- | ---- | ---------------- | -------- | --- | +| RERANK. | We | randomly | sample | n | = 4 re- | | | | | | | +| | | | | sample | | Table4,5,and6respectively(fullresultsin§G.6). | | | | | | +| sponses | for each | question, | | and select | the best re- | | | | | | | +Wefirstnotethatallmodelsachievegoodfluency +| sponse | using the | automatic | | citation recall | score. | | | | | | | +| ------ | --------- | --------- | --- | --------------- | ------ | --- | --- | --- | --- | --- | --- | +scores(exceptsomemodelsonELI5mainlydueto +weexpectRERANKtoimprovethecitationquality. +| | | | | | | theirlongergenerations). | | | Wesummarizethemain | | | +| --- | --- | --- | --- | --- | --- | ------------------------ | --- | --- | ------------------ | --- | --- | +POSTCITE. Foreachstatement,wefindthebest takeawaysfromtheexperimentsbelow. +matchingpassageamongthetop-100retrievedpas- +| | | | | | | VANILLAachievesstrongperformance. | | | | | Despite | +| ----------------------- | --- | --- | --- | ----------------- | --- | --------------------------------- | --- | --- | --- | --- | ------- | +| sagesusingGTRandciteit. | | | | Wecombinethiswith | | | | | | | | +itssimplicity,VANILLA(puttingretrievedpassages +CLOSEDBOOKinourexperiments. +incontext)achievesclose-to-the-bestperformance +amongallpromptingstrategies. +5 Experiments +Usingsummariesorsnippetsimprovescorrect- +Wedescribeexperimentdetailsin§C.WeuseChat- +| | | | | | | ness. Weseeauniversaltrendthat | | | | SUMMor | SNIP- | +| ------------------------ | --- | --- | --- | ------ | ---------- | ------------------------------ | --- | ------------ | ------ | ------- | ----- | +| GPT (gpt-3.5-turbo-0301) | | | | with a | 4K context | | | | | | | +| | | | | | | PET improves | | correctness, | though | on ASQA | and | +windowformostmainexperimentsandablations. +| | | | | | | ELI5, | such an | improvement | comes | at a | cost of | +| ------------------- | -------- | ------- | ---- | ------------------ | ------- | ---------------------------------------- | -------- | ----------- | ------------ | ------- | ------- | +| We also | report | results | with | ChatGPT-16K | (gpt- | | | | | | | +| | | | | | | citationqualityduetothelossycompression. | | | | | Com- | +| 3.5-turbo-16k-0613) | | | and | GPT-4 (gpt-4-0613; | | | | | | | | +| | | | | | | bining | INTERACT | with | SUMM/SNIPPET | doesnot | | +| 8K context | window). | | For | open-source | models, | | | | | | | +bringimprovement,andwehypothesizethatcheck- +| we test | LLaMA | (Touvron | | et al., 2023a) | and its | | | | | | | +| ------- | ----- | -------- | --- | -------------- | ------- | --- | --- | --- | --- | --- | --- | +ingthefullpassagesofferslimitedbenefitandcur- +instruction-tunedversions,includingAlpaca(Taori +rentLLMsarenotproficientinaninteractiveusage. +| et al., 2023), | | Vicuna | (Chiang | et al., | 2023), and | | | | | | | +| -------------- | --- | ------ | ------- | ------- | ---------- | --- | --- | --- | --- | --- | --- | +Retrievingtextontheflydoesnotimproveper- +11Wedonotsearchovertheentirecorpusbecause{query} +| | | | | | | formance. | All | datasets | show that | VANILLA | out- | +| --- | --- | --- | --- | --- | --- | --------- | --- | -------- | --------- | ------- | ---- | +mayleaveoutcertaincontextinthequestionandsearching +| | | | | | | performs | INLINESEARCHoncitationquality(and | | | | | +| --- | --- | --- | --- | --- | --- | -------- | --------------------------------- | --- | --- | --- | --- | +amongthealready-retrievedpassagesgivesbetterresults. + +| | | Correctness | | | Citation | | | | Fluency | | Correct. | Citation | | +| -------------- | --- | ----------- | ----- | ---- | -------- | ----- | -------------- | --- | ------- | ---- | -------- | -------- | ----- | +| | | Rec.-5 | Prec. | | Rec. | Prec. | | | (MAUVE) | | (Claim) | Rec. | Prec. | +| | | ChatGPT | | | | | | | ChatGPT | | | | | +| | | | | | | | VANILLA(5-psg) | | | 57.2 | 12.0 | 51.1 | 50.0 | +| VANILLA(5-psg) | | 20.8 | | 20.8 | 20.5 | 20.9 | | | | | | | | +| w/RERANK | | 22.8 | | 21.4 | 21.2 | 21.4 | w/RERANK | | | 56.1 | 11.4 | 69.3 | 67.8 | +SUMM(10-psg) 23.6 21.2 23.6 25.7 SUMM(10-psg) 40.3 12.5 51.5 48.2 +SNIPPET(10-psg) 24.5 21.5 22.9 24.9 SNIPPET(10-psg) 62.9 14.3 50.4 45.0 +w/INTERACT 21.9 23.0 21.9 23.4 w/INTERACT 68.0 13.3 47.8 45.0 +INLINESEARCH 17.2 20.4 14.9 14.9 INLINESEARCH 49.7 13.4 45.6 43.7 +CLOSEDBOOK 32.9 19.8 10.0 10.0 CLOSEDBOOK 32.6 18.6 15.5 15.5 +| | GPT-4(VANILLAprompting) | | | | | | | GPT-4(VANILLAprompting) | | | | | | +| --- | ----------------------- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | +GPT-4(5-psg) 22.2 25.0 25.9 27.0 GPT-4(5-psg) 38.4 14.2 44.0 50.1 +GPT-4(20-psg) 29.6 26.2 27.4 28.5 GPT-4(20-psg) 41.5 18.3 48.5 53.4 +| | LLaMA(VANILLAprompting) | | | | | | | LLaMA(VANILLAprompting) | | | | | | +| --- | ----------------------- | --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | +LLaMA-13B(3-psg) 9.7 9.1 6.7 7.1 LLaMA-13B(3-psg) 50.0 3.9 3.1 5.3 +Vicuna-13B(5-psg) 14.0 15.9 12.5 13.4 Vicuna-13B(3-psg) 58.2 10.0 15.6 19.6 +Chat-13B(5-psg) 21.1 18.2 9.6 9.7 Chat-13B(5-psg) 34.7 13.4 17.3 15.8 +Chat-70B(5-psg) 21.8 18.4 15.1 15.6 Chat-70B(5-psg) 38.6 12.8 38.3 37.9 +Table5: ExperimentsonQAMPARI.“Rec.-5”: weset Table 6: Experiments on ELI5. We use claim recall +therecalltobe100%ifthepredictionincludesatleast forthecorrectnessevaluation. Chat-13BandChat-70B +| 5correctanswers. | | | | | | | refertoLLaMA-2-Chat. | | | | | | | +| ---------------- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | +GPT-4bringslimitedimprovementbutisbetter +| oncorrectnessforASQAandELI5). | | | | | Bymanually | | | | | | | | | +| ----------------------------- | --- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +examiningtheexamples,wefindthatitischalleng- at using long context. We evaluate GPT-4 with +VANILLAanddifferentnumbersofpassages(more +| ing to ask | detailed | questions | | without | seeing | any | | | | | | | | +| ---------- | ---------- | ------------- | --- | ------- | ------ | --- | ----------------- | --- | ----------------------------- | ------------ | --- | --------- | ---- | +| | | | | | | | resultsin§G.6). | | GPT-4bringsconsistent(butlim- | | | | | +| passages. | To improve | INLINESEARCH, | | | one | may | | | | | | | | +| | | | | | | | ited) improvement | | on | correctness, | | but often | at a | +needtoprovidemorecontextaboutthequestions +inadvanceorencouragethemodeltocallretrievers costofcitationquality. GPT-4canalsoincorporate +| | | | | | | | more passages | | due to | its longer | context | | window, | +| --- | --- | --- | --- | --- | --- | --- | ------------- | --- | ------ | ---------- | ------- | --- | ------- | +withmoredetailedanddiversequeries. +| | | | | | | | which boosts | both | correctness | | and | citation | qual- | +| ---------------------------- | --- | --- | --- | ------------- | --- | --- | -------------------------------------------- | ---- | ----------- | --- | --- | -------- | ----- | +| RERANKboostscitationquality. | | | | Weobservethat | | | | | | | | | | +| | | | | | | | ity. Onthecontrary,includingmorepassageswith | | | | | | | +RERANK leadstoconsistentimprovementincita- ChatGPT-16K does not improve the results (Ta- +tionquality(onASQAandELI5). Astheautomatic ble 7), suggesting that processing more passages +scoresmaybebiasedinRERANK,wealsoconduct +| | | | | | | | is non-trivial | and | GPT-4 | is | better | at synthesizing | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | --- | ----- | --- | ------ | --------------- | --- | +humanevaluation(§6)andverifyitseffectiveness. +informationfromitslongcontextthanChatGPT. +CLOSEDBOOK+POSTCITEdeliversstrongcor- +| | | | | | | | 5.2 ComparisonofDifferentLLMs | | | | | | | +| ---------------- | -------- | -------- | -------- | -------------- | ------- | --- | ----------------------------- | -------- | --------- | --- | ---- | --- | ---- | +| rectness | but poor | citation | quality. | | CLOSED- | | | | | | | | | +| | | | | | | | Table 7 | compares | different | | LLMs | on | ASQA | +| BOOK outperforms | | VANILLA | | in correctness | | on | | | | | | | | +ELI5 and QAMPARI, and has only a 2% gap on using VANILLA (more results in §G.6). No- +ASQA.However, CLOSEDBOOK cannotprovide tably, instruction-tuned models (Vicuna-13B and +anycitation;whencombinedwithPOSTCITE,the LLaMA-2-Chat)outperformtheoriginalLLaMA +| | | | | | | | models in | correctness | | and | considerably | | enhance | +| --------------------------------- | --- | --- | --- | --- | ------------ | --- | --------- | ----------- | --- | --- | ------------ | --- | ------- | +| citationqualityremainsinadequate. | | | | | Forinstance, | | | | | | | | | +citation recall of CLOSEDBOOK+POSTCITE is thecitationquality. Weobservethatwhiletheorig- +lowerthan VANILLAby47%onASQA. inal LLaMA models are able to copy facts from +thecontext,theystrugglewithaccuratelycitingthe +| Tounderstandwhy | | CLOSEDBOOKachievesbet- | | | | | | | | | | | | +| --------------- | --- | ---------------------- | --- | --- | --- | --- | ---------- | ------ | --- | --------- | -------- | --- | -------- | +| | | | | | | | sources or | simply | do | not cite. | Notably, | | the best | +tercorrectnessandwhyPOSTCITEcannotdeliver +open-sourcemodel,LLaMA-2-70B-Chat,achieves +| satisfying | citation | quality, | we | manually | examine | | | | | | | | | +| ---------- | -------- | -------- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | --- | --- | +model outputs and find that: (1) open-book mod- comparablecorrectnessscoreastheOpenAImod- +els,butstilllagsbehindincitationquality. +elsareeasilydistractedbyirrelevantpassagesand +generateresponseswithlowercorrectness,aphe- +| | | | | | | | 5.3 RetrievalAnalysis | | | | | | | +| ------- | ------------- | --- | ------ | ------ | ------- | --- | --------------------- | --- | --- | --- | --- | --- | --- | +| nomenon | also observed | | by Shi | et al. | (2023); | (2) | | | | | | | | +CLOSEDBOOK often generates texts that are cor- Theretrievalresultsplayacrucialroletothecor- +rectbutnotsimilartoanyretrievedpassages,mak- rectnessandthecitationquality. Figure4presents +ingitdifficulttomatchacitationpost-hoc. the retrieval recall@k with different datasets and + +80 +70 +60 +50 +40 +30 +5 20 50 100 +k +k@llaceR +ASQA QAMPARI ELI5 +60 30 Fluency Correct. Citation +(MAUVE) (EM) Rec. Prec. +Oracle +40 Oracle 20 ChatGPT(VANILLA) +Oracle DPR(5-psg) 61.8 36.1 65.0 65.6 +Vanilla (5-psg) Vanilla (5-psg) Vanilla (5-psg) GTR(1-psg) 69.5 38.4 56.0 64.0 20 10 +DPR DPR GTR(3-psg) 66.6 39.6 72.8 73.9 +GTR GTR BM25 GTR(5-psg) 66.6 40.4 73.6 72.5 +0 +5 20 50 100 5 20 50 100 +k k +Figure4: Retrievalrecall@konASQA(EMrecall),QAMPARI(recall-5),andELI5(claimrecall). Retrievalrecall +servesasanupperboundformodelperformance,andwecomparethemwithtwomodels’correctnessresultsinthe +figure(dashedlines): “Vanilla(5-psg)”isChatGPTVANILLAwithtop-5passagesincontext;“Oracle”isthesame +modelexceptthatituses5goldpassages(§G.1),whoserecallmatchesRecall@100onallthreedatasets. +Fluency Correct. Citation improvement for ChatGPT. Specifically, correct- +(MAUVE) (EMRec.) Rec. Prec. ness plateaus at top-1 passage and citation qual- +Open-source(max#tokens=2K-4K) ity plateaus at top-3. GPT-4 (Table 7) exhibits +LLaMA-13B(3-psg) 68.4 26.9 10.6 15.4 an increasing trend with more passages, but the +Vicuna-13B(3-psg) 82.6 31.9 51.1 50.1 +improvement is not proportional to the retrieval +Chat-13B(5-psg) 72.4 35.2 38.4 39.4 +Chat-70B(5-psg) 88.3 41.5 62.9 61.3 performance. Thisindicatesthelimitedabilityof +ChatGPT(max#tokens=4K) LLMsinutilizingmultiplepassageswithincontext. +ChatGPT(3-psg) 66.6 39.6 72.8 73.9 +ChatGPT(5-psg) 66.6 40.4 73.6 72.5 5.4 OtherAblations +ChatGPT-16K(max#tokens=16K) +Weprovideadditionalablationsin§G.Insummary, +ChatGPT(5-psg) 60.3 36.1 76.2 76.5 +ChatGPT(10-psg) 56.3 36.7 75.3 75.0 wefindthat(1)usingcomprehensiveinstructions +ChatGPT(20-psg) 56.7 36.1 73.7 73.5 enhancesthecitationqualityofinstruction-tuned +GPT-4(max#tokens=8K) +models (§G.2); (2) including at least one demon- +GPT-4(5-psg) 67.1 41.3 68.5 75.6 +strationimprovestheperformance(§G.3);(3)fine- +GPT-4(10-psg) 71.5 43.1 72.0 75.5 +GPT-4(20-psg) 64.9 44.4 73.0 76.5 tunedmodels(FiD;IzacardandGrave,2021)with +POSTCITE lag behind LLMs in both correctness +Table 7: Comparison of different LLMs on ASQA +andcitationqualityandfailtogeneralize(§G.4). +(GTR+VANILLA). LLaMA-13BandVicuna-13Bhave +acontextlimitof2,048tokens,andthuscanonlyusea +6 HumanEvaluation +shortversionofinstructionsandatmosttop-3passages. +Chat-13BandChat-70BrefertoLLaMA-2-Chat. Toverifythatourautomaticevaluationcorrelates +with human judgement, we conduct human eval- +retrievers. Asthenumberofpassagesincreases,re- uationonselectedmodelsandrequestworkersto +trievalrecallsteadilyimproves. Additionally,Fig- judgemodelgenerationsonthreedimensionssimi- +ure 4 shows the correctness performance of two lartoLiuetal.(2023)—(1)utility: a1-to-5score +models: (1) ChatGPT VANILLA with top-5 pas- indicatingwhetherthegenerationhelpsanswerthe +sages (our primary baseline); (2) an oracle ver- question;(2)citationrecall: theannotatorisgiven +sionofthesamemodelemploying5goldpassages asentenceandallpassagesthatthesentencecited, +(§G.1; the 5gold passages matchthe retrieval re- and is asked to judge whether the passages fully +call@100). Notably,bothmodels’correctnesslags supportthesentence;(3)citationprecision: given +behind the corresponding retrieval recall (except a sentence and one of its citations, the annotator +forELI5top-5). Thediscrepancysuggeststhatde- is asked to judge whether the citation “fully sup- +spitethepresenceofaccurateanswersincontext, ports”,“partiallysupports”,or“doesnotsupport” +LLMsstruggletoutilizethemintheiroutputs. thesentence. Eachcitationgetsaprecisionscore +1 if the output sentence has a citation recall of 1 +We compare the impact of different retrievers +andthiscitationatleast“partiallysupports”it. See +anddifferentnumbersofpassagestoLLMs. Fig- +AppendixFformoredetails. +ure 4 (right) shows that GTR outperforms DPR +inbothcorrectnessandcitationquality,emphasiz- Model outputs score high utility. The utility +ing the importance of deploying better retrievers. scoresdonotdiffersignificantlybetweenmodels, +Contrary to the retrieval recall trend in Figure 4, ranging 3.7-3.9 for ASQA and 3.5-3.6 for ELI5. +morepassagesincontextdonotyieldsubstantial Uponinspection,alltestedmodelsaremostlyable + +Humanscores ALCEscores model is provided the papers-to-cite and context +| | | | | | | andisrequiredtorecoverthecitingtext. | | | | Itisdiffer- | | +| --- | --- | ---- | ----- | ---- | ----- | ------------------------------------ | --- | --- | --- | ----------- | --- | +| | | Rec. | Prec. | Rec. | Prec. | | | | | | | +entfromALCEasallcitationsareprovidedandthe +| ChatGPTVANILLA | | 74.7 | 76.6 | 75.3 | 74.4 | | | | | | | +| -------------- | --- | ---- | ---- | ---- | ---- | --- | --- | --- | --- | --- | --- | +w/RERANK 79.3 81.9 83.9 80.8 modelonlyneedstoperformthesummarization. +| Vicuna-13BVANILLA | | 51.6 | 51.5 | 50.3 | 50.1 | | | | | | | +| ----------------- | --- | ---- | ---- | ---- | ---- | ----------------------- | --- | --- | ------------------ | --- | --- | +| | | | | | | Retrieval-augmentedLMs. | | | Manystudieshaveex- | | | +ploredaugmentingLMswithexternallyretrieved +| Table8: Humancitationqualityevaluationvs. | | | | | ALCE | | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | +citationqualityevaluationonASQA. information. Guu et al. (2020); Borgeaud et al. +| | | | | | | (2022); Izacard | | et al. (2022) | pre-train | language | | +| --- | --- | ----------- | --- | ---------- | --- | --------------- | --------- | ------------- | --------- | -------------- | --- | +| | | Humanscores | | ALCEscores | | | | | | | | +| | | | | | | models with | retrieved | passages, | | while Khandel- | | +Rec. Prec. Rec. Prec. wal et al. (2020); Zhong et al. (2022) augment +ChatGPTVANILLA 50.8 52.4 52.8 50.4 LLMs’outputbyinterpolatingitwithakNNmod- +w/RERANK 59.7 60.6 63.0 60.6 ule; though none of them explicitly provide cita- +Vicuna-13BVANILLA 13.4 19.2 13.6 18.1 tionstotheretrievedsources. Otherworksprompt +orfine-tuneLLMsto“retrieveon-the-fly”(Parisi +| Table9: Humancitationqualityevaluationvs. | | | | | ALCE | | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | ---- | --- | --- | --- | --- | --- | --- | +etal.,2022;Schicketal.,2023;Shusteretal.,2022; +citationqualityevaluationonELI5. +| | | | | | | Jiang et | al., 2023; | Yao et | al., 2023; | Press | et al., | +| --- | --- | --- | --- | --- | --- | -------- | ---------- | ------ | ---------- | ----- | ------- | +2022),whichoffersflexibilityofwhenandwhatto +tooutputfluentanswersthatarerelatedtotheques- +tion,despitedifferencesinfactualcorrectness. search. Gaoetal.(2023);Heetal.(2022)propose +tofirstgeneratetextwithoutaccessingexternaldoc- +| Our automatic | evaluation | | of | citation | quality | | | | | | | +| ------------- | ---------- | --- | --- | -------- | ------- | --- | --- | --- | --- | --- | --- | +umentsandthenretrieverelevantdocumentsand +| stronglycorrelateswithhumanjudgements. | | | | | As | | | | | | | +| -------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +revisethegenerationtobeconsistent. +showninTable8(ASQA)andTable9(ELI5),the +| | | | | | | Among | previous | explorations, | | Nakano | et al. | +| --- | --- | --- | --- | --- | --- | ----- | -------- | ------------- | --- | ------ | ------ | +relativerankingsinducedbyhumanandourauto- (2021);Menicketal.(2022)aretheclosesttoour +| maticmetricsareconsistent. | | | Theabsolutecitation | | | | | | | | | +| -------------------------- | --- | --- | ------------------- | --- | --- | -------------- | --- | ---------------- | --- | --------- | ----- | +| | | | | | | setting, where | | LLMs are trained | | to answer | ques- | +scoresfromhumanandALCEareverycloseexcept +| | | | | | | tionswhileprovidingcitations. | | | However,theydo | | | +| ---------- | ------ | ---- | ------------- | --- | -------- | ----------------------------- | --- | --- | -------------- | --- | --- | +| for RERANK | (which | uses | the automated | | citation | | | | | | | +notexploreretrievalstrategiesandsimplyusecom- +| recall for | reranking). | This | suggests | | that an im- | | | | | | | +| ---------- | ----------- | ---- | -------- | --- | ----------- | --- | --- | --- | --- | --- | --- | +mercialsearchengines,whicharenotreproducible, +provementonALCEcitationmetricstranslatesto +| | | | | | | and their | models | and training | data | are closed- | | +| ------------------------------ | --- | --- | --- | ------------ | --- | -------------------------------- | ------ | ------------ | ---- | ----------- | --- | +| improvementonhumanpreferences. | | | | Furthermore, | | | | | | | | +| | | | | | | source. Tothebestofourknowledge, | | | | wearethe | | +theCohen’skappacoefficientbetweenhumanand firsttoimplementend-to-endsystemsthatretrieve, +ALCEsuggestssubstantialagreementforcitation +synthesize,andcitedocumentswithLLMs. +recall(0.698)andmoderateagreementforcitation +| precision(0.525). | | Wealsoshowin§G.5thatour | | | | 8 Conclusion | | | | | | +| ----------------- | --- | ----------------------- | --- | --- | --- | ------------ | --- | --- | --- | --- | --- | +automaticevaluationachieveshighaccuracywhen +WeproposeALCE,thefirstautomaticbenchmark +treatinghumanannotationsasgoldlabels(85.1% +| | | | | | | forevaluatingLLMgenerationswithcitations. | | | | | We | +| --- | --- | --- | --- | --- | --- | ----------------------------------------- | --- | --- | --- | --- | --- | +forcitationrecalland77.6%forcitationprecision). +deployautomaticmetricstomeasurefluency,cor- +7 RelatedWork rectness,andcitationquality,andverifytheireffi- +| | | | | | | cacyviahumanevaluation. | | | Weexploreavarietyof | | | +| --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | ------------------- | --- | --- | +Evaluating citations. Generating text with ci- strategiesforincorporatingcitationsinLLMsand +tations is closely related to attribution. Rashkin demonstratethatcurrentsystemshaveconsiderable +et al. (2023) define the “attributable to identified roomforimprovementonALCE. +sources”(AIS)scoretomeasurehowfaithfulagen- Ourexperimentshighlightanumberofpromis- +eratedtextistoitssources. Bohnetetal.(2022)ap- ingresearchdirections,including(1)enhancingre- +plyAISscoresonasingle-documentshort-answer +trievalandrefiningretrievalintegrationsinLLMs, +QA dataset. Honovich et al. (2022); Yue et al. (2)developinglong-contextLLMs,and(3)advanc- +(2023) study automatic evaluations for the AIS ingLLMs’abilitytosynthesizemultiplesources. +score. Aconcurrentwork(Liuetal.,2023)conduct +What’sevenmoreintriguingisthattheseresearch +humanevaluationoncommercialgenerativesearch proposalsextendbeyondtheALCEsetup(forex- +enginestoexaminetheircitationqualities. ample,long-contextLLMshavenumerousexciting +Scientific citation text generation (Funkquist applications), and ALCE can serve as a valuable +et al., 2022) is a related task to ALCE where the testbedfortheirdevelopment. + +Limitations +| | | | | Lespiau, | Bogdan | Damoc, | | Aidan | Clark, | Diego | +| --- | --- | --- | --- | -------- | ------ | ------ | --- | ----- | ------ | ----- | +DeLasCasas,AureliaGuy,JacobMenick,Roman +Ourevaluationstillhasroomforimprovement: (1) Ring, Tom Hennigan, Saffron Huang, Loren Mag- +| | | | | giore, | Chris | Jones, | Albin | Cassirer, | Andy | Brock, | +| --- | --- | --- | --- | ------ | ----- | ------ | ----- | --------- | ---- | ------ | +MAUVEisfoundtobesensitivetooutputlength +andmayprovideunstableresults;(2)fortheELI5’s Michela Paganini, Geoffrey Irving, Oriol Vinyals, +SimonOsindero,KarenSimonyan,JackRae,Erich +correctnessevaluation,theautomaticallygenerated +| | | | | Elsen,andLaurentSifre.2022. | | | | Improvinglanguage | | | +| --- | --- | --- | --- | --------------------------- | --- | --- | --- | ----------------- | --- | --- | +claimsmaynotcoverallpossibleanswersdueto modelsbyretrievingfromtrillionsoftokens. InInter- +theopen-endednatureofthequestions;(3)ourci- nationalConferenceonMachineLearning(ICML), +tationqualityevaluationislimitedbytheaccuracy volume162,pages2206–2240. +oftheNLImodel; forcitationprecision, theNLI SamuelBowman,GaborAngeli,ChristopherPotts,and +modelcannotdetectthecaseof“partiallysupport” +| | | | | Christopher | D | Manning. | 2015. | A | large | annotated | +| --- | --- | --- | --- | ----------- | --- | -------- | ----- | --- | ----- | --------- | +and thus leads to a lower citation precision score corpus for learning natural language inference. In +EmpiricalMethodsinNaturalLanguageProcessing +thanthehumanevaluation. +(EMNLP). +Althoughwebelieveourcurateddatasetsclosely +resemblethedistributionofreal-worlduserques- TomBBrown,BenjaminMann,NickRyder,Melanie +| | | | | Subbiah, | Jared | Kaplan, | Prafulla | Dhariwal, | | Arvind | +| --- | --- | --- | --- | -------- | ----- | ------- | -------- | --------- | --- | ------ | +tions,weacknowledgethattheydonotcovermore +Neelakantan,PranavShyam,GirishSastry,Amanda +challengingscenarios,suchasmulti-hopreasoning, +| | | | | Askell,etal.2020. | | Languagemodelsarefew-shot | | | | | +| --- | --- | --- | --- | ----------------- | --- | ------------------------- | --- | --- | --- | --- | +mathreasoning,andcodecompletion. +| | | | | learners. | InAdvancesinNeuralInformationProcess- | | | | | | +| --- | --- | --- | --- | --------- | ------------------------------------- | --- | --- | --- | --- | --- | +In our experiments, we focus on prompting ingSystems(NeurIPS). +| LLMswithoutupdatingtheirmodelweights. | | | Train- | | | | | | | | +| ------------------------------------- | --- | --- | ------ | --------------- | --- | ------- | --- | ------- | ---- | ------ | +| | | | | Wei-Lin Chiang, | | Zhuohan | Li, | Zi Lin, | Ying | Sheng, | +ing a model directly to incorporate citations re- ZhanghaoWu,HaoZhang,LianminZheng,Siyuan +mains challenging due to the lack of supervised Zhuang,YonghaoZhuang,JosephE.Gonzalez,Ion +data. However, we observe that certain human- Stoica, and Eric P. Xing. 2023. Vicuna: An open- +sourcechatbotimpressinggpt-4with90%*chatgpt +instructiondatasetscontainexamplessimilartoour +quality. +| task setup. | We leave | the exploration | of training | | | | | | | | +| ----------- | -------- | --------------- | ----------- | --- | --- | --- | --- | --- | --- | --- | +LLMstogeneratecitationsforfuturework. AngelaFan,YacineJernite,EthanPerez,DavidGrang- +| | | | | ier, Jason | Weston, | and | Michael | Auli. | 2019. | ELI5: | +| --- | --- | --- | --- | -------------------------- | ------- | --- | ------- | ---------------- | ----- | ----- | +| | | | | Longformquestionanswering. | | | | InAssociationfor | | | +Acknowledgments +ComputationalLinguistics(ACL),pages3558–3567. +Weappreciatethehelpfulfeedbackfromthemem- +| | | | | Martin Funkquist, | | Ilia | Kuznetsov, | Yufang | | Hou, and | +| ----------------------------------------- | ----------- | -------------- | -------- | --------------------------------- | ------ | ---- | ---------- | -------- | ------------- | -------- | +| bersofthePrincetonNLPgroup. | | WethankAlexan- | | | | | | | | | +| | | | | IrynaGurevych.2022. | | | CiteBench: | | Abenchmarkfor | | +| der Wettig, | Nelson Liu, | Tianyi Zhang, | Yu Meng, | | | | | | | | +| | | | | ScientificCitationTextGeneration. | | | | | arXivpreprint | | +| SadhikaMalladi,YangsiboHuang,ZhiyuanZeng, | | | | arXiv:2212.09577. | | | | | | | +| andDanFriedmanforthevaluablediscussion. | | | We | | | | | | | | +| | | | | Luyu Gao, | Zhuyun | Dai, | Panupong | Pasupat, | | Anthony | +thankSurgeAI(especiallyAnnaFolinskyandEd- Chen, Arun Tejasvi Chaganty, Yicheng Fan, Vin- +win Chen) for their support with the human eval- centYZhao,NiLao,HongraeLee,Da-ChengJuan, +| | | | | etal.2023. | RARR:Researchingandrevisingwhat | | | | | | +| --- | --- | --- | --- | ---------- | ------------------------------- | --- | --- | --- | --- | --- | +uation. TianyuGaoissupportedbyanIBMPhD +| | | | | language | models | say, | using | language | models. | In | +| ----------- | ------------------------------ | --- | --- | -------- | ------ | ---- | ----- | -------- | ------- | --- | +| Fellowship. | ThisresearchissupportedbyanNSF | | | | | | | | | | +AssociationforComputationalLinguistics(ACL). +CAREERaward(IIS-2239290),aSloanResearch +Fellowship, andMicrosoftAzurecreditsthrough KelvinGuu,KentonLee,ZoraTung,PanupongPasupat, +| | | | | and Ming-Wei | | Chang. | 2020. | REALM: | | Retrieval- | +| --- | --- | --- | --- | ------------ | --- | ------ | ----- | ------ | --- | ---------- | +the“AccelerateFoundationModelsAcademicRe- +| | | | | augmentedlanguagemodelpre-training. | | | | | | InInterna- | +| --- | --- | --- | --- | ----------------------------------- | --- | --- | --- | --- | --- | ---------- | +search”Initiative. +tionalConferenceonMachineLearning(ICML). +HangfengHe,HongmingZhang,andDanRoth.2022. +| | | | | Rethinking | with | retrieval: | Faithful | | large | language | +| --- | --- | --- | --- | ---------- | ---- | ---------- | -------- | --- | ----- | -------- | +References +| | | | | modelinference. | | arXivpreprintarXiv:2301.00303. | | | | | +| --- | --- | --- | --- | --------------- | --- | ------------------------------ | --- | --- | --- | --- | +BerndBohnet,VinhQTran,PatVerga,RoeeAharoni, +DanielAndor,LivioBaldiniSoares,JacobEisenstein, AriHoltzman,JanBuys,LiDu,MaxwellForbes,and +KuzmanGanchev, JonathanHerzig, KaiHui, etal. YejinChoi.2020. Thecuriouscaseofneuraltextde- +2022. Attributedquestionanswering:Evaluationand generation. InInternationalConferenceonLearning +modelingforattributedlargelanguagemodels. arXiv Representations(ICLR). +preprintarXiv:2212.08037. +| | | | | OrHonovich, | RoeeAharoni, | | JonathanHerzig, | | | Hagai | +| --- | --- | --- | --- | ----------- | ------------ | --- | --------------- | --- | --- | ----- | +Sebastian Borgeaud, Arthur Mensch, Jordan Hoff- Taitelbaum,DoronKukliansy,VeredCohen,Thomas +mann, Trevor Cai, Eliza Rutherford, Katie Milli- Scialom, Idan Szpektor, Avinatan Hassidim, and +can,GeorgeBmVanDenDriessche,Jean-Baptiste Yossi Matias. 2022. TRUE: Re-evaluating factual + +consistencyevaluation. InNorthAmericanChapter Andreas Köpf, Yannic Kilcher, Dimitri von Rütte, +oftheAssociationforComputationalLinguistics:Hu- Sotiris Anagnostidis, Zhi-Rui Tam, Keith Stevens, +manLanguageTechnologies(NAACL-HLT),pages Abdullah Barhoum, Nguyen Minh Duc, Oliver +| 3905–3920. | | | | | | | Stanley, | Richárd Nagyfi, | Shahul | ES, | Sameer Suri, | +| ---------- | --- | --- | --- | --- | --- | --- | -------- | --------------- | ------ | --- | ------------ | +DavidGlushkov,ArnavDantuluri,AndrewMaguire, +GautierIzacardandEdouardGrave.2021. Leveraging ChristophSchuhmann,HuuNguyen,andAlexander +passageretrievalwithgenerativemodelsforopendo- Mattick. 2023. Openassistant conversations – de- +mainquestionanswering. InProceedingsofthe16th mocratizinglargelanguagemodelalignment. arXiv +preprintarXiv:2304.07327. +ConferenceoftheEuropeanChapteroftheAssoci- +| ationforComputationalLinguistics: | | | | | MainVolume, | | | | | | | +| --------------------------------- | --- | --- | --- | --- | ----------- | --- | --- | --- | --- | --- | --- | +KentonLee,Ming-WeiChang,andKristinaToutanova. +| pages 874–880, | | Online. | Association | | for Computa- | | | | | | | +| -------------- | --- | ------- | ----------- | --- | ------------ | --- | ------------ | --------- | ---------- | ---------- | ---- | +| | | | | | | | 2019. Latent | retrieval | for weakly | supervised | open | +tionalLinguistics. +| | | | | | | | domainquestionanswering. | | InAssociationforCom- | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------------ | --- | -------------------- | --- | --- | +putationalLinguistics(ACL),pages6086–6096. +| Gautier Izacard, | | Patrick | Lewis, | Maria | Lomeli, | Lucas | | | | | | +| ---------------- | --- | ------- | ------ | ----- | ------- | ----- | --- | --- | --- | --- | --- | +Hosseini,FabioPetroni,TimoSchick,JaneDwivedi- +| | | | | | | | Nelson F Liu, | Tianyi Zhang, | and | Percy | Liang. 2023. | +| --- | --- | --- | --- | --- | --- | --- | ------------- | ------------- | --- | ----- | ------------ | +Yu,ArmandJoulin,SebastianRiedel,andEdouard +Evaluatingverifiabilityingenerativesearchengines. +| Grave. | 2022. | Atlas: | Few-shot | learning | | with re- | | | | | | +| ------ | ----- | ------ | -------- | -------- | --- | -------- | --- | --- | --- | --- | --- | +arXivpreprintarXiv:2304.09848. +| trievalaugmentedlanguagemodels. | | | | | arXivpreprint | | | | | | | +| ------------------------------- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | +arXiv:2208.03299. Jacob Menick, Maja Trebacz, Vladimir Mikulik, +| | | | | | | | John Aslanides, | Francis | Song, | Martin | Chadwick, | +| --- | --- | --- | --- | --- | --- | --- | --------------- | ------- | ----- | ------ | --------- | +ZiweiJi,NayeonLee,RitaFrieske,TiezhengYu,Dan Mia Glaese, Susannah Young, Lucy Campbell- +| Su, Yan | Xu, | Etsuko | Ishii, Ye | Jin | Bang, | Andrea | | | | | | +| ------- | --- | ------ | --------- | --- | ----- | ------ | ------------------------------------ | --- | --- | --- | -------- | +| | | | | | | | Gillingham,GeoffreyIrving,etal.2022. | | | | Teaching | +Madotto,andPascaleFung.2023. Surveyofhalluci- language models to support answers with verified +nationinnaturallanguagegeneration. ACMComput- quotes. arXivpreprintarXiv:2203.11147. +ingSurveys,55(12):1–38. +SewonMin,JulianMichael,HannanehHajishirzi,and +| | | | | | | | LukeZettlemoyer.2020. | | AmbigQA:Answeringam- | | | +| --------- | ------ | ----- | ----------- | ---- | ------ | ------- | ---------------------------- | --- | -------------------- | --------------- | --- | +| Zhengbao | Jiang, | Frank | F Xu, | Luyu | Gao, | Zhiqing | | | | | | +| | | | | | | | biguousopen-domainquestions. | | | InProceedingsof | | +| Sun, Qian | Liu, | Jane | Dwivedi-Yu, | | Yiming | Yang, | | | | | | +the2020ConferenceonEmpiricalMethodsinNat- +| Jamie | Callan, | and Graham | | Neubig. | 2023. | Ac- | | | | | | +| ----- | ------- | ---------- | --- | ------- | ----- | --- | --- | --- | --- | --- | --- | +tiveretrievalaugmentedgeneration. arXivpreprint ural Language Processing (EMNLP), pages 5783– +arXiv:2305.06983. 5797, Online. Association for Computational Lin- +guistics. +| Ryo Kamoi, | Tanya | Goyal, | Juan | Diego | Rodriguez, | | | | | | | +| ---------- | ----- | ------ | ---- | ----- | ---------- | --- | --- | --- | --- | --- | --- | +ReiichiroNakano,JacobHilton,SuchirBalaji,JeffWu, +| and Greg | Durrett. | 2023. | WiCE: | Real-World | | En- | | | | | | +| -------- | ---------- | ----- | ---------- | ---------- | ----- | -------- | ------------ | --------- | ---- | ----------- | ------ | +| | | | | | | | Long Ouyang, | Christina | Kim, | Christopher | Hesse, | +| tailment | for Claims | in | Wikipedia. | | arXiv | preprint | | | | | | +ShantanuJain,VineetKosaraju,WilliamSaunders, +arXiv:2303.01432. +| | | | | | | | et al. 2021. | WebGPT: | Browser-assisted | | question- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | ---------- | ---------------- | ----- | --------- | +| | | | | | | | answering | with human | feedback. | arXiv | preprint | +VladimirKarpukhin,BarlasOguz,SewonMin,Patrick +arXiv:2112.09332. +Lewis,LedellWu,SergeyEdunov,DanqiChen,and +Wen-tauYih.2020. Densepassageretrievalforopen- Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo +domainquestionanswering. InProceedingsofthe HernandezAbrego, JiMa, VincentZhao, YiLuan, +2020ConferenceonEmpiricalMethodsinNatural KeithHall,Ming-WeiChang,andYinfeiYang.2022. +LanguageProcessing(EMNLP),pages6769–6781, Largedualencodersaregeneralizableretrievers. In +Online.AssociationforComputationalLinguistics. +EmpiricalMethodsinNaturalLanguageProcessing +(EMNLP),pages9844–9855. +UrvashiKhandelwal,OmerLevy,DanJurafsky,Luke +Zettlemoyer,andMikeLewis.2020. Generalization OpenAI.2023. GPT-4TechnicalReport. +| throughmemorization: | | | Nearestneighborlanguage | | | | | | | | | +| -------------------- | --- | --- | ----------------------- | --- | --- | --- | --- | --- | --- | --- | --- | +International Conference on Learning LongOuyang,JeffreyWu,XuJiang,DiogoAlmeida, +| models. | In | | | | | | | | | | | +| ------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +CarrollWainwright,PamelaMishkin,ChongZhang, +Representations(ICLR). +SandhiniAgarwal,KatarinaSlama,AlexRay,etal. +| | | | | | | | 2022. Training | languagemodelsto | | followinstruc- | | +| --- | --- | --- | --- | --- | --- | --- | -------------- | ---------------- | --- | -------------- | --- | +TusharKhot,AshishSabharwal,andPeterClark.2018. +| | | | | | | | tionswithhumanfeedback. | | AdvancesinNeuralIn- | | | +| -------- | --------- | ---------- | --- | ------- | ---- | ------- | ----------------------- | --- | ------------------- | --- | --- | +| Scitail: | A textual | entailment | | dataset | from | science | | | | | | +formationProcessingSystems(NeurIPS),35:27730– +| questionanswering. | | InConferenceonArtificialIn- | | | | | | | | | | +| ------------------ | --- | --------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +27744. +telligence(AAAI),volume32. +| | | | | | | | AaronParisi,YaoZhao,andNoahFiedel.2022. | | | | TALM: | +| ---------------------------------------------- | --- | ----- | -------- | ----- | ------ | ----- | --------------------------------------- | -------- | ------- | ----- | -------- | +| Kalpesh Krishna, | | Aurko | Roy, and | Mohit | Iyyer. | 2021. | | | | | | +| | | | | | | | Tool augmented | language | models. | arXiv | preprint | +| Hurdlestoprogressinlong-formquestionanswering. | | | | | | | arXiv:2205.12255. | | | | | +InProceedingsofthe2021ConferenceoftheNorth +AmericanChapteroftheAssociationforComputa- FabioPetroni,AleksandraPiktus,AngelaFan,Patrick +tionalLinguistics: HumanLanguageTechnologies, Lewis,MajidYazdani,NicolaDeCao,JamesThorne, +pages4940–4957,Online.AssociationforComputa- YacineJernite,VladimirKarpukhin,JeanMaillard, +tionalLinguistics. VassilisPlachouras,TimRocktäschel,andSebastian + +Riedel. 2021. KILT: a benchmark for knowledge Freda Shi, Xinyun Chen, Kanishka Misra, Nathan +intensivelanguagetasks. InProceedingsofthe2021 Scales, David Dohan, Ed Chi, Nathanael Schärli, +Conference of the North American Chapter of the andDennyZhou.2023. Largelanguagemodelscan +AssociationforComputationalLinguistics: Human beeasilydistractedbyirrelevantcontext. InInterna- +LanguageTechnologies,pages2523–2544,Online. tionalConferenceonMachineLearning(ICML). +AssociationforComputationalLinguistics. +| | | | | | Kurt Shuster, | Mojtaba | | Komeili, | Leonard | Adolphs, | +| --- | --- | --- | --- | --- | ------------- | ------- | --- | -------- | ------- | -------- | +AleksandraPiktus,FabioPetroni,VladimirKarpukhin, Stephen Roller, Arthur Szlam, and Jason Weston. +DmytroOkhonko,SamuelBroscheit,GautierIzac- 2022. Language models that seek for knowledge: +ard, Patrick Lewis, Barlas Og˘uz, Edouard Grave, Modular search & generation for dialogue and +Wen-tauYih,etal.2021. TheWebIsYourOyster– promptcompletion. InFindingsoftheAssociation +| | | | | | forComputationalLinguistics: | | | | EMNLP2022,pages | | +| --- | --- | --- | --- | --- | ---------------------------- | --- | --- | --- | --------------- | --- | +Knowledge-IntensiveNLPagainstaVeryLargeWeb +| Corpus. | arXivpreprintarXiv:2112.09924. | | | | 373–393. | | | | | | +| ------- | ------------------------------ | --- | --- | --- | -------- | --- | --- | --- | --- | --- | +KrishnaPillutla,SwabhaSwayamdipta,RowanZellers, IvanStelmakh,YiLuan,BhuwanDhingra,andMing- +JohnThickstun,SeanWelleck,YejinChoi,andZaid Wei Chang. 2022. ASQA: Factoid questions meet +Harchaoui.2021. MAUVE:Measuringthegapbe- long-formanswers. InProceedingsofthe2022Con- +ferenceonEmpiricalMethodsinNaturalLanguage +tweenneuraltextandhumantextusingdivergence +| | | | | | Processing, | pages | 8273–8288, | | Abu Dhabi, | United | +| --------------- | ----------- | --- | ------------------ | ---- | -------------------------------------------- | ----- | ---------- | --- | ---------- | ------ | +| frontiers. | In Advances | in | Neural Information | Pro- | | | | | | | +| cessingSystems. | | | | | ArabEmirates.AssociationforComputationalLin- | | | | | | +guistics. +OfirPress,MuruZhang,SewonMin,LudwigSchmidt, +NoahASmith,andMikeLewis.2022. Measuring Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann +andnarrowingthecompositionalitygapinlanguage Dubois,XuechenLi,CarlosGuestrin,PercyLiang, +| | | | | | andTatsunoriB.Hashimoto.2023. | | | | StanfordAlpaca: | | +| ------- | ------------------------------ | --- | --- | --- | ----------------------------- | --- | --- | --- | --------------- | --- | +| models. | arXivpreprintarXiv:2210.03350. | | | | | | | | | | +AnInstruction-followingLLaMAmodel. +ColinRaffel,NoamShazeer,AdamRoberts,Katherine +| | | | | | James Thorne, | | Andreas | | Vlachos, | Christos | +| --- | --- | --- | --- | --- | ------------- | --- | ------- | --- | -------- | -------- | +Lee,SharanNarang,MichaelMatena,YanqiZhou, +WeiLi,andPeterJLiu.2020. Exploringthelimits Christodoulopoulos, and Arpit Mittal. 2018. +oftransferlearningwithaunifiedtext-to-textTrans- FEVER: a large-scale dataset for fact extraction +former. TheJournalofMachineLearningResearch and VERification. In North American Chapter +| | | | | | of the Association | | for | Computational | | Linguistics: | +| --- | --- | --- | --- | --- | ------------------ | --- | --- | ------------- | --- | ------------ | +(JMLR),21(140). +| | | | | | Human | Language | Technologies | | (NAACL-HLT), | | +| ----------------------------------------------- | ----- | ------ | ------------------ | --- | ------------- | -------- | ------------ | --- | ------------ | --- | +| PranavRajpurkar,JianZhang,KonstantinLopyrev,and | | | | | pages809–819. | | | | | | +| Percy Liang. | 2016. | SQuAD: | 100,000+ questions | | | | | | | | +HugoTouvron,ThibautLavril,GautierIzacard,Xavier +| for machine | comprehension | | of text. In Empirical | | | | | | | | +| ----------- | ------------- | --- | --------------------- | --- | --- | --- | --- | --- | --- | --- | +MethodsinNaturalLanguageProcessing(EMNLP), Martinet,Marie-AnneLachaux,TimothéeLacroix, +BaptisteRozière,NamanGoyal,EricHambro,Faisal +pages2383–2392. +| | | | | | Azhar, | et al. | 2023a. | LLaMA: | Open | and Effi- | +| --------------- | --- | ---------------- | ------- | ----- | ------------------------------ | ------ | ------ | ------ | ---- | ------------- | +| | | | | | cientFoundationLanguageModels. | | | | | arXivpreprint | +| Hannah Rashkin, | | Vitaly Nikolaev, | Matthew | Lamm, | | | | | | | +arXiv:2302.13971. +| Lora Aroyo, | Michael | Collins, | Dipanjan Das, | Slav | | | | | | | +| ----------- | ------- | -------- | ------------- | ---- | --- | --- | --- | --- | --- | --- | +Petrov,GauravSinghTomar,IuliaTurc,andDavid +| | | | | | Hugo Touvron, | Louis | Martin, | Kevin | Stone, | Peter Al- | +| ------------- | --------------------------------- | --- | --- | --- | ------------- | ----- | ------- | ----- | ------ | --------- | +| Reitter.2023. | MeasuringAttributioninNaturalLan- | | | | | | | | | | +guageGenerationModels. ComputationalLinguis- bert, Amjad Almahairi, Yasmine Babaei, Nikolay +Bashlykov,SoumyaBatra,PrajjwalBhargava,Shruti +tics,pages1–64. +Bhosale,DanBikel,LukasBlecher,CristianCanton +Ferrer,MoyaChen,GuillemCucurull,DavidEsiobu, +SamuelJosephAmouyalOhadRubin,OriYoran,Tomer +JudeFernandes,JeremyFu,WenyinFu,BrianFuller, +| Wolfson, | Jonathan | Herzig, | and Jonathan | Berant. | | | | | | | +| -------- | -------- | ------- | ------------ | ------- | --- | --- | --- | --- | --- | --- | +2022. QAMPARI:AnOpen-domainQuestionAn- CynthiaGao,VedanujGoswami,NamanGoyal,An- +swering Benchmark for Questions with Many An- thonyHartshorn,SagharHosseini,RuiHou,Hakan +swers from Multiple Paragraphs. arXiv preprint Inan,MarcinKardas,ViktorKerkez,MadianKhabsa, +IsabelKloumann,ArtemKorenev,PunitSinghKoura, +arXiv:2205.12665. +Marie-AnneLachaux,ThibautLavril,JenyaLee,Di- +TimoSchick,JaneDwivedi-Yu,RobertoDessì,Roberta anaLiskovich,YinghaiLu,YuningMao,XavierMar- +Raileanu,MariaLomeli,LukeZettlemoyer,Nicola tinet,TodorMihaylov,PushkarMishra,IgorMoly- +Cancedda,andThomasScialom.2023. Toolformer: bog, Yixin Nie, Andrew Poulton, Jeremy Reizen- +Languagemodelscanteachthemselvestousetools. stein,RashiRungta,KalyanSaladi,AlanSchelten, +| | | | | | Ruan Silva, | Eric | Michael | Smith, | Ranjan | Subrama- | +| --- | --- | --- | --- | --- | ----------- | ---- | ------- | ------ | ------ | -------- | +arXivpreprintarXiv:2302.04761. +| | | | | | nian, Xiaoqing | | Ellen | Tan, Binh | Tang, | Ross Tay- | +| --- | --- | --- | --- | --- | -------------- | --- | ----- | --------- | ----- | --------- | +TalSchuster,AdamFisch,andReginaBarzilay.2021. lor, Adina Williams, Jian Xiang Kuan, Puxin Xu, +GetyourvitaminC!robustfactverificationwithcon- ZhengYan,IliyanZarov,YuchenZhang,AngelaFan, +trastiveevidence. InNorthAmericanChapterofthe Melanie Kambadur, Sharan Narang, Aurelien Ro- +AssociationforComputationalLinguistics: Human driguez,RobertStojnic,SergeyEdunov,andThomas +LanguageTechnologies(NAACL-HLT),pages624– Scialom. 2023b. Llama 2: Open foundation and +| 643. | | | | | fine-tunedchatmodels. | | | | | | +| ---- | --- | --- | --- | --- | --------------------- | --- | --- | --- | --- | --- | + +AlexWang, KyunghyunCho, andMikeLewis.2020. +Askingandansweringquestionstoevaluatethefac- +tualconsistencyofsummaries. InProceedingsofthe +58thAnnualMeetingoftheAssociationforCompu- +tationalLinguistics,pages5008–5020,Online.Asso- +ciationforComputationalLinguistics. +AdinaWilliams,NikitaNangia,andSamuelBowman. +2018. A broad-coverage challenge corpus for sen- +tence understanding through inference. In North +AmericanChapteroftheAssociationforComputa- +tionalLinguistics: HumanLanguageTechnologies +(NAACL-HLT). +Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak +Shafran,KarthikRNarasimhan,andYuanCao.2023. +React: Synergizingreasoningandactinginlanguage +models. In International Conference on Learning +Representations(ICLR). +XiangYue,BoshiWang,KaiZhang,ZiruChen,YuSu, +and Huan Sun. 2023. Automatic evaluation of at- +tributionbylargelanguagemodels. arXivpreprint +arXiv:2305.06311. +ShiyueZhangandMohitBansal.2021. Findingabal- +anceddegreeofautomationforsummaryevaluation. +InEmpiricalMethodsinNaturalLanguageProcess- +ing(EMNLP),pages6617–6632. +Yuan Zhang, Jason Baldridge, and Luheng He. 2019. +PAWS:Paraphraseadversariesfromwordscrambling. +In North American Chapter of the Association for +ComputationalLinguistics: HumanLanguageTech- +nologies(NAACL-HLT),pages1298–1308. +ZexuanZhong,TaoLei,andDanqiChen.2022. Train- +inglanguagemodelswithmemoryaugmentation. In +EmpiricalMethodsinNaturalLanguageProcessing +(EMNLP),pages5657–5673. + +A GeneratingClaimsforELI5 +| | | | | | | NLI model | accurately | | predicts | | the entailment | | +| --- | --- | --- | --- | --- | --- | ------------- | ---------- | ---------------------------- | -------- | --- | -------------- | --- | +| | | | | | | ofsub-claims. | | Wefurtheranalyzeoursub-claim | | | | | +We elect not to use ROUGE-L as our main cor- evaluation metrics by checking the error rate of +| rectness metrics | since | it does | not | account | for the | | | | | | | | +| ---------------- | ----- | ------- | --- | ------- | ------- | -------------------------------- | --- | --- | --- | --- | ---------- | --- | +| | | | | | | thefinalpredictionoftheNLImodel. | | | | | Tothisend, | | +differentwaysofexpressingthesameanswerandit we first manually annotate the entailment scores +canbeeasilygamed(Krishnaetal.,2021). Wefur- between40outputsandtheirsub-claims(intotal +| therillustratethisissueinTable10. | | | | Asystemcan | | | | | | | | | +| --------------------------------- | --- | --- | --- | ---------- | --- | --- | --- | --- | --- | --- | --- | --- | +of120pairs;thesearethesamequestionsfromthe +easilyachievehighROUGE-Lscorebyretrieving +| | | | | | | previous | analysis). | We | then | use | the NLI | model | +| --- | --- | --- | --- | --- | --- | -------- | ---------- | --- | ---- | --- | ------- | ----- | +andreturningthetoppassagefromaBM25index. toobtaintheentailmentscoresfortheoutputand +However, the claims evaluation metric does not sub-claims. Using the human annotations as the +rewardthisapproachsincetheoutputoftenlacks +| | | | | | | ground | truth label, | we | found | that | the NLI | model | +| ----------------------------- | --- | --- | --- | --- | --- | -------------------------- | ------------ | --- | ----- | ---- | ------- | ----- | +| differentaspectsoftheanswers. | | | | | | achievedanaccuracyof80.0%. | | | | | | | +B DatasetStatistics +| | | ROUGE-L | | Claimrecall | | | | | | | | | +| -------------- | --- | ------- | ---- | ----------- | ---- | --- | --- | --- | --- | --- | --- | --- | +| ChatGPTVANILLA | | | 20.6 | | 12.0 | | | | | | | | +ForASQA,humananswershaveanaveragelength +| ChatGPTORACLE | | | 21.2 | | 21.3 | | | | | | | | +| ------------- | --- | --- | ---- | --- | ---- | --- | --- | --- | --- | --- | --- | --- | +LLaMa-13BVANILLA 16.2 3.9 of 65 words. For QAMPARI, each question has +Top-1passage 19.1 3.0 onaverage13answers. ForELI5,humananswers +haveanaveragelengthof131words. +| Table10: ComparisonbetweenROUGE-Landclaim | | | | | | | | | | | | | +| ----------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +recallscoresonELI5. +C ImplementationDetails +Instead,weleveragetheoriginalanswerstogen- +| | | | | | | NLI model. | | We use | the | version | of | TRUE | +| --- | --- | --- | --- | --- | --- | ---------- | --- | ------ | --- | ------- | --- | ---- | +eratesub-claimsandusethemtoserveasanesti- +| | | | | | | model from | https://huggingface.co/google/ | | | | | | +| --- | --- | --- | --- | --- | --- | ---------- | ------------------------------ | --- | --- | --- | --- | --- | +mate of the different aspects of the answers that t5_xxl_true_nli_mixture, which is trained on +| we expect | the model | to cover. | | This approach | is | | | | | | | | +| --------- | --------- | --------- | --- | ------------- | --- | ------------ | --- | ------- | ------ | ---- | --------- | --- | +| | | | | | | SNLI (Bowman | | et al., | 2015), | MNLI | (Williams | | +inspiredbyworksinsummarizationevaluationand +| | | | | | | et al., 2018), | Fever | | (Thorne | et | al., 2018), | Sci- | +| --- | --- | --- | --- | --- | --- | -------------- | ----- | --- | ------- | --- | ----------- | ---- | +claimverification(ZhangandBansal,2021;Kamoi tail(Khotetal.,2018),PAWS(Zhangetal.,2019), +etal.,2023;Wangetal.,2020). +| | | | | | | andVitaminC(Schusteretal.,2021). | | | | | Thismodel | | +| --- | --- | --- | --- | --- | --- | -------------------------------- | --- | --- | --- | --- | --------- | --- | +Specifically,weusetext-davinci-003togen- +| | | | | | | usesthefollowingprompt: | | | “premise: | | {PREMISE} | | +| --- | --- | --- | --- | --- | --- | ----------------------- | --- | --- | --------- | --- | --------- | --- | +erate the sub-claims. We first manually annotate hypothesis: {}”andoutputs“1”ifthepremise +threequestionandanswerpairsfromtheoriginal entails the hypothesis. We format each passage +ELI5 training set with 3 sub-claims each. Then, (whenusedaspremise)bytheformatof“Title: +weprompttext-davinci-003withthesepairsas +| | | | | | | {TITLE}\n{TEXT}” | | and | concatenate | | all | passages | +| --- | --- | --- | --- | --- | --- | ---------------- | --- | --- | ----------- | --- | --- | -------- | +demonstrations. Thefullpromptwithanexample with“\n”asaseparator. +isshowninTable22. +MAUVE.WhenrunningMAUVE,weconcatenate +InstructGPT generates coherent and faithful the question and the model output (or human an- +sub-claims. To ensure that the generated sub- swer) by space. We truncate both the references +claims are of good quality, we manually inspect and the model generations to 100 words, as we +arandomsampleof40answersandtheirgenerated found MAUVE results are unstable beyond this +sub-claims(totalingto120sub-claims). Foreach lengthforELI5(thisisduetothatELI5hasalot +sub-claim,weassignascoreof1ifitisrelevantto +ofextremelylonghumananswers). +thequestionandfaithfultothefactspresentedin +| | | | | | | Exact match | for | ASQA | and | QAMPARI. | | Both | +| ----------------------------- | --- | --- | --- | ----------- | --- | ----------- | ------- | ---- | ------- | -------- | --- | --------- | +| thegroundtruth,and0otherwise. | | | | Wefoundthat | | | | | | | | | +| | | | | | | ASQA and | QAMPARI | | provide | aliases | | for their | +112outofthe120(93.33%)sub-claimsreceiveda +| | | | | | | shortanswers. | Wenormalizetheresponseandthe | | | | | | +| --- | --- | --- | --- | --- | --- | ------------- | ---------------------------- | --- | --- | --- | --- | --- | +scoreof1,meaningthatourgeneratedsub-claims +shortanswerssimilarlytoRajpurkaretal.(2016) +areofhighqualityandfaithfultothegroundtruth. +andreportthescorewiththebest-matchingaliases. +Furthermore,theaveragenumberofwordsinthe +ForASQA,Stelmakhetal.(2022)alsoproposea +| generated | sub-claims | is 14 | words, | and | they are | | | | | | | | +| --------- | ---------- | ----- | ------ | --- | -------- | --- | --- | --- | --- | --- | --- | --- | +QA-basedevaluationwhichwefoundtobenotas +| typically | just one | sentence | long. | This | is aligned | | | | | | | | +| --------- | -------- | -------- | ----- | ---- | ---------- | --- | --- | --- | --- | --- | --- | --- | +stable,andthuswedonotreportitinourpaper. +withtheintentbehindthemetric—tocaptureshort +factualclaimsmadebytheoriginalanswer. Output truncation. Before evaluation, we trun- + +catemodeloutputbynewlines,asnon-instruction- [2]willbecountedas“irrelevant”whileitshould +tunedmodelsmaygeneratemorecontentafternew not be penalized. Liu et al. (2023) conduct hu- +linesthatareirrelevant. manevaluationoncitationprecisioninadifferent +way: Foreachcitation,theyaskannotatorstojudge +INTERACT. Empirically, we found that models +whetherthecitation(1)fullysupport,(2)partially +tendtoexecutetoomanyconsecutive“check”ac- +support,or(3)doesnotsupports . Onecitationc +tions, so we force the model to always “output” i i,j +ispreciseif(a)c fullysupportss or(b)C fully +aftereach“check”. Welimitthemaximumnumber i,j i i +supportss ,c partiallysupportss ,andnoc ∈ C +of passages to check as 3 to avoid exceeding the i i,j i i +alonefullysupportss . Thisevaluationsolvedthe +lengthlimit. Thefullpassagesareremovedfrom i +cornercasewementionedinthemainpaper(one +thecontextafteroneactiontosavecontextspace. +citation partially supports the claim but is identi- +Table27providesanexamplefor INTERACT. +fiedas“irrelevant”). However,itischallengingto +Main experiments. For all experiments except conductsuchevaluationautomatically,asthereis +ChatGPTRERANK,weruneachmodelthreetimes no existing model that can judge whether a cita- +withdifferentseedsandeachtimewesampletwo tion“partially”supportsaclaim. Wealsoexplore +demonstrationsfromapooloffour. Wereportthe promptingChatGPTtoconductsuchatask,which +averaged scores for all experiments in the main yields poor results. We defer it to future work to +paperandwereportthestandarddeviationsinAp- collectsuperviseddatatotrainabetterϕthatcan +pendixG.6. detect“partialsupport”. +Decodingmethods. Basedonpreliminaryexperi- +F HumanEvaluation +mentswechoosethefollowingdecodingmethods: +For ChatGPT and GPT-4, we use sampling with We employ Surge AI (https://www.surgehq. +temperature 0.5; for all open-source models, we ai/)forourhumanevaluation. Theaveragepayto +useNucleussampling(Holtzmanetal.,2020)and +workersis20USDperhour. Werandomlysample +settop_p = 0.95. +100examplesfromASQAandELI5andannotate +outputs of selected models: ChatGPT VANILLA, +D ALCECatchesShortcutCases +ChatGPT RERANK,andVicuna-13B VANILLA. +Fluency Correct. Citation F.1 Utility +(MAUVE) (EMRec.) Rec. Prec. To check if the model output is useful to down- +ChatGPT 66.6 40.4 73.6 63.0 streamusers,wemeasuretheutilityoftheresponse +Top-1passage 20.8 35.1 99.4 99.4 S. Wefirstshowthequeryq andmodelresponse +First2sents 67.2 18.9 98.7 98.7 +S to the worker and ask them to rate their agree- +mentwiththestatement"Theresponseisahelpful +Table11: ASQAcheatingcases. “ChatGPT”:theChat- +andinformativeanswertothequery"onaLikert +GPT VANILLA model with GTR-retrieved top-5 pas- +sages. “Top-1passage”: usethetop-1retrievedpassage scaleof1-5, correspondingtoStronglyDisagree, +astheresponse. “First2sents”: usethefirst2sentences Disagree,Neutral,Agree,andStronglyAgree. +ofthetop-1retrievedpassage. +F.2 CitationRecall +Table11demonstratestheexperimentstoshow Theannotatorsareshownthequestionq,thestate- +that ALCE is robust to shortcut cases. Using the ments ,andallofitscitationsC ,andtheyrateif +i i +top-1 passages or first two sentences of the top-1 thejointsetofcitationsfullysupportthestatement +passages induces almost perfect citation quality, (recall=1) or if they do not support all the claims +butfluencyandcorrectnessaredramaticallylower. (recall=0). We calculate the overall recall score +for the generation by taking an average of all the +E CitationRecallDiscussion +statements’recallscores. +Our citation precision evaluation cannot detect a +F.3 CitationPrecision +citation that partially supports the statement and +hence will falsely penalize it. Consider a state- Weshowthequestionq andapairofastatement +ments anditscitations[2][4][5]: if[2]entails s andoneofitscitationc ∈ C totheannotator. +3 i i,j i +partialinformationofs that[4][5]alsoentails, Weasktheannotatorifthecitationfullysupports, 3 - Windows Azure Marketplace: An online service for purchasing cloud-based data and applications. - -All four of these components run in Microsoft data centers located around the world: two in North -America, two in Europe, and two in Asia. Developers using the platform can control which data center -runs their applications and stores their data, giving them the ability to place both closer to their users. - -Each part of the Windows Azure platform has its own role to play. This overview describes all four, first at -a high level, then in a bit more detail. The goal is to provide a big-picture introduction to this cloud -platform. - -WINDOWS AZURE - -At a high level, Windows Azure is simple to understand: It runs Windows applications and stores data in -the cloud. Figure 2 shows its components. - -Figure 2: Windows Azure provides compute and storage services in the cloud. - -The five parts of Windows Azure today are the following: - - Compute: The Windows Azure compute service runs applications on a Windows Server foundation. -These applications can be created using the .NET Framework in languages such as C# and Visual Basic, -or they can be built without .NET in C++, Java, and other languages. Developers can use Visual Studio -or other development tools, and they’re free to use technologies such as ASP.NET, Windows -Communication Foundation (WCF), and PHP. - -Storage: This service allows storing binary large objects (blobs), provides queues for communication -between components of Windows Azure applications, and even offers a form of tables with a simple -query language. (Windows Azure applications that need traditional relational storage can also use - -4 - - SQL Azure.) Both Windows Azure applications and on-premises applications can access the Windows -Azure storage service, and both do it in the same way: using a RESTful approach. - -Fabric controller: As the figure suggests, Windows Azure runs on a large number of machines. The -fabric controller’s job is to knit the machines in a single Windows Azure data center into a cohesive -whole. The Windows Azure compute and storage services are then built on top of this pool of -processing power. - - Content delivery network (CDN): Caching frequently accessed data closer to its users speeds up -access to that data. The Windows Azure CDN can do this for blobs, maintaining cached copies at sites -around the world. - - Connect: It’s often useful for organizations to interact with cloud applications as if they were inside -the organization’s own firewall. Windows Azure Connect allows this, making it easier for, say, a -Windows Azure application to access an on-premises database. - -Running applications and storing data in the cloud can have clear benefits. Rather than buying, installing, -and operating its own systems, for example, an organization can rely on a cloud provider to do this for -them. Also, customers pay just for the computing and storage they use, rather than maintaining a large -set of servers only for peak loads. And applications written for Windows Azure can scale better, be more -reliable, and require less administration than those written using the traditional Windows Server -programming model. - -To create, configure, and monitor applications, Windows Azure customers can use a browser-accessible -portal. A customer logs in with a Windows Live ID, then chooses whether to create a hosting account for -running applications, a storage account for storing data, or both. Microsoft then charges each customer -based on how much compute time, storage, and bandwidth that customer uses. How an application -charges its own customers—if it charges them at all—is entirely up to the people who create that -application. - -Windows Azure is a general platform that can be used in a broad set of scenarios. Here are a few -examples: - - An independent software vendor (ISV) creating a software-as-a-service (SaaS) version of an existing -on-premises Windows application might choose to build it on Windows Azure. Because Windows -Azure mostly provides a standard Windows environment, moving the application’s business logic to -this cloud platform won’t typically pose many problems. And once again, building on an existing -platform lets the ISV focus on business logic—the thing that makes them money—rather than -spending time on infrastructure. - - An enterprise creating an application for its customers or employees might choose to build it on -Windows Azure. Because Windows Azure supports .NET, developers with the right skills aren’t -difficult to find, nor are they prohibitively expensive. Running the application in Microsoft data -centers frees the enterprise from the responsibility and expense of managing its own servers, turning -capital expenses into operating expenses. And especially if the application has spikes in usage— -maybe it’s an on-line flower store that must handle the Mother’s Day rush—letting Microsoft -maintain the large server base required for this can make economic sense. - -5 - - A start-up creating a new Web site—the next Facebook, say—could build its application on Windows -Azure. Because this platform supports both Web-facing services and background processes, the -application can provide an -interface as well as executing work for users -asynchronously. Rather than spending time and money worrying about infrastructure, the start-up -can instead focus solely on creating code that provides value to its customers and investors. The -company can also start small, incurring low costs while its application has only a few users. If the -application catches on and usage increases, Windows Azure can scale the application as needed. - -interactive user - -These three examples illustrate the kinds of things organizations might do with Windows Azure, but -they’re not an exhaustive list. As interest in cloud computing continues to grow, expect to see a variety of -applications created for this cloud platform. - -SQL AZURE - -Along with running applications, another attractive way to use the cloud is for storing data. SQL Azure -addresses this area, offering cloud-based services for relational data. As Figure 3 shows, SQL Azure today -includes three components. - -Figure 3: SQL Azure provides relational database services in the cloud. - -The components of SQL Azure today are the following: - -SQL Azure Database provides a cloud-based database management system (DBMS). This technology -lets on-premises and cloud applications store relational data on Microsoft servers in Microsoft data - -6 - - centers. As with other cloud technologies, an organization pays only for what it uses, increasing and -decreasing usage (and cost) as the organization’s needs change. Using a cloud database also allows -converting what would be capital expenses, such as investments in disks and DBMS software, into -operating expenses. - -SQL Azure Reporting is a version of SQL Server Reporting Services (SSRS) that runs in the cloud. -Intended primarily for use with SQL Azure Database, it allows creating and publishing standard SSRS -reports on cloud data. - -SQL Azure Data Sync allows synchronizing data between SQL Azure Database and on-premises SQL -Server databases. It can also be used to synchronize data across different SQL Azure databases in -different Microsoft data centers. - -SQL Azure is built on Microsoft SQL Server. As with SQL Server, developers can create indexes and views, -use stored procedures, define triggers, and more. Applications can access SQL Azure data using Entity -Framework, ADO.NET, and other Windows data access interfaces. In fact, applications that today access -SQL Server locally will largely work unchanged with data in SQL Azure. Customers can also use on- -premises software such as SQL Server Analysis Services to work with their cloud-based data. - -While applications can use SQL Azure much as they do a local DBMS, the management requirements are -significantly reduced. Rather than worry about mechanics, such as monitoring disk usage and servicing log -files, a SQL Azure customer can focus on their data; Microsoft handles the operational details. And as with -other components of this cloud platform, customers use the common Windows Azure platform portal to -access its services. - -Applications might use SQL Azure in a variety of ways. Here are some examples: - - A Windows Azure application can store its data in SQL Azure. While Windows Azure provides its own -storage, relational tables aren’t among the options it offers. Since many existing applications use -relational storage and many developers know how to work with it, a significant number of Windows -Azure applications rely on SQL Azure to work with data in this familiar way. For example, a SaaS -application built on Windows Azure might create a separate SQL Azure database for each customer, -providing an intrinsically multi-tenant design. And to improve performance, customers can specify -that a particular Windows Azure application must run in the same data center in which SQL Azure -Database stores that application’s information. - - An application in a small business or a department of a larger organization might rely on SQL Azure. -Rather than storing its data in a SQL Server or Access database running on a computer under -somebody’s desk, the application can instead take advantage of the reliability and availability of cloud -storage. It can also create reports on this data using either SQL Azure Reporting or SSRS on-premises. -If the organization wishes to maintain an on-premises copy of the data as well, it can use SQL Azure -Data Sync to synchronize the cloud and on-premises databases. - -Suppose a manufacturer wishes to make product information available both to its dealer network -and directly to customers. Putting this data in SQL Azure would let it be accessed by applications -running at the dealers and by a customer-facing Web application run by the manufacturer itself. - -7 - - Whether it’s for supporting a Windows Azure application, making data more accessible, or other reasons, -data services in the cloud can be attractive. The goal of SQL Azure is to provide these services in a familiar, -usable way for cloud and on-premises applications. - -WINDOWS AZURE APPFABRIC - -Running applications and storing data in the cloud are both important aspects of cloud computing. -They’re far from the whole story, however. It’s also possible to provide cloud-based infrastructure -services. Filling this gap is the goal of Windows Azure AppFabric. - -The functions provided by AppFabric today address common challenges in building distributed -applications. Figure 4 shows its components. - -Figure 4: Windows Azure AppFabric provides Windows Azure-based infrastructure services that can be -used by both cloud and on-premises applications. - -As the figure suggests, all of the components of Windows Azure AppFabric are built on Windows Azure -(although they don’t all provide services solely to Windows Azure applications). Those components are -the following: - -Service Bus: Exposing an application’s services on the Internet is harder than it might seem. The goal -of Service Bus is to make this simpler by letting an application expose endpoints in the cloud that can -be accessed by other applications, whether on-premises or in the cloud. Each exposed endpoint is -assigned a URI, which clients can use to locate and access the service. Service Bus also handles the -challenges of dealing with network address translation and getting through firewalls without opening -new ports for exposed applications. - - Access Control: There are many ways for a user to get a digital identity today. The options include -Active Directory, Windows Live ID, Google Accounts, Facebook, and more. If an application wants to -let users log in with any of these, the application’s creator faces the daunting challenge of supporting - -8 - - this plethora of approaches. Access Control simplifies this by providing built-in support for all of them -(and more). It also provides a single place for defining rules to control what each user is allowed to -access. - - Caching: It’s common for applications to access the same data over and over. One way to speed up -this kind of application is to cache frequently accessed information, reducing the number of times -that application must query a database. The Caching service provides this—and the performance -boost it brings—for Windows Azure applications. - -Microsoft has announced plans to add more services under the Windows Azure AppFabric banner, so -expect this list to grow in the not-too-distant future. - -As with Windows Azure and SQL Azure, customers use a browser-accessible portal to sign up for -AppFabric. Once this has been done, these services can be used in a variety of ways. Here are some of the -possibilities: - -Suppose an enterprise wished to let software at its trading partners access one of its applications. It -could expose this application’s functions through SOAP or RESTful services created using WCF, then -register those service endpoints with Service Bus. Its trading partners could then use Service Bus to -find these endpoints and access the services. - -Imagine that the creator of this same application needs to let trading partners log in with a variety of -different identities. Rather than implementing support for these identities himself, he could use the -Access Control service to hide this complexity. - - A Windows Azure application created using ASP.NET might use the Session object to store per-client -state. By changing only a configuration setting, the application can cause this data to be kept in the -Caching service rather than, say, Windows Azure Storage tables. Doing this is likely to make the -application faster and more scalable. - -Along with the cloud-based services of Windows Azure AppFabric, Microsoft also provides an analogous -technology known as Windows Server AppFabric. As its name suggests, the services it provides run on -Windows Server—they support on-premises applications—rather than in the cloud. The on-premises -services aren’t exactly the same today as those in Windows Azure AppFabric (although Microsoft’s -announced plan is make the two congruent). Don’t be confused; throughout this paper, the name -“AppFabric” is used to refer to the cloud-based services. Also, don’t confuse Windows Azure AppFabric -with the Windows Azure fabric controller. Even though both contain the term “fabric”, they’re wholly -separate technologies addressing quite distinct problems. - -WINDOWS AZURE MARKETPLACE - -In the on-premises world, many applications are purchased rather than custom-built. Many organizations -also rely on datasets from commercial providers. With the rise of the cloud, why not let customers find -and buy cloud applications and cloud-accessible data? Making this possible is the goal of Windows Azure -Marketplace. Figure 5 shows its two components. - -9 - - Figure 5: Windows Azure Marketplace provides an online marketplace for cloud applications and data. - -The two parts of Windows Azure Marketplace are: - - DataMarket (originally codenamed “Dallas”), providing a way for content providers to make datasets -available. Customers can browse the offerings, then purchase whatever they find useful. Applications, -both custom and off-the-shelf (such as Microsoft Excel), can then access this data through RESTful -requests or the OData protocol. - - AppMarket, providing a way for creators of cloud applications to expose those applications to - -potential customers. AppMarket will be available sometime after DataMarket. - -Using the cloud to find, evaluate, and purchase data and applications makes sense. The goal of Windows -Azure Marketplace is to make these things easier to do. - -A CLOSER LOOK AT THE TECHNOLOGIES - -Having a broad understanding of the Windows Azure platform is an important first step. Making good -decisions require a deeper understanding of these technologies, however. This section takes a slightly -more in-depth look at each member of the family. - -WINDOWS AZURE - -Windows Azure provides what’s commonly called Platform as a Service (PaaS). Rather than offering a -cloud replica of the on-premises world, it offers a higher level of service that’s meant to make life easier -for both developers and administrators. What follows walks through the five components that -collectively provide this service. - -10 - - Compute - -An application built on the Windows Azure compute service is structured as one or more roles. When it -executes, the application typically runs two or more instances of each role, with each instance running as -its own virtual machine (VM). Figure 6 shows how this looks. - -Figure 6: A running Windows Azure application consists of any combination of Web role instances, -Worker role instances, and VM role instances. - -A Windows Azure application today can be created using three kinds of roles: - - Web roles, intended primarily for running Web-based applications. Each Web role instance has -Internet Information Services (IIS) 7 pre-configured to run inside it, so creating applications using -ASP.NET, WCF, or other Web technologies is straightforward. It’s also possible to create applications -using PHP, Java, and other non-Microsoft technologies. - - Worker roles, designed to run a variety of code. A Worker role might run a simulation, for example, -or video processing or nearly anything else. It’s common for an application to interact with users -through a Web role, then hand tasks off to a Worker role for processing. - - VM roles, which can run a user-provided Windows Server 2008 R2 image. A VM role can be the right - -choice for moving some on-premises Windows Server applications to Windows Azure. - -When a developer gives Windows Azure an application to run, she submits configuration information -along with it. Among other things, this information tells the platform how many instances of each role to -run. The Windows Azure fabric controller then creates a VM for each instance, running the code for the - -11 - - appropriate role in each VM. And as Figure 6 indicates, requests from the outside world are load balanced -across all instances of a role. - -This has an important implication: To be scalable, Windows Azure role instances shouldn’t maintain their -state themselves between requests. Because the load balancer doesn’t allow creating an affinity with a -particular role instance—there’s no support for sticky sessions—there’s no way to guarantee that multiple -requests from the same user will be sent to the same instance. Instead, any client-specific state should be -written to Windows Azure storage, stored in SQL Azure Database, or maintained externally in some other -way. - -For role instances, Windows Azure lets developers choose from several VM sizes, each with a specific -number of processor cores and memory. Since each instance can be assigned one or more cores, -applications have predictable performance. And to increase the load an application can handle, its owner -or the application itself can request an increase in the number of running instances for one or more roles. -The Windows Azure fabric controller will then spin up new VMs for these instances and start them -running. It’s also possible to explicitly decrease the number of instances for a role, letting an application -grow and shrink as needed to handle changing loads. - -For developers, building a Windows Azure application looks much like building a traditional Windows -application. Microsoft provides Visual Studio project templates, for instance, to help developers create -Windows Azure applications. Also, the Windows Azure software development kit includes a version of the -Windows Azure environment that runs on the developer’s machine. Known as the Windows Azure -development fabric, it implements a local replica of the Windows Azure compute and storage services. A -developer can create and debug his application using this local simulacrum, then deploy it to Windows -Azure in the cloud when it’s ready. Windows Azure also provides other services for developers, such as -information about a running application’s CPU consumption, incoming and outgoing bandwidth, and -storage usage. - -Storage - -Applications work with data in many different ways. Sometimes, all that’s required are simple blobs, while -other situations call for a more structured way to store information. And in some cases, all that’s really -needed is a way to exchange data between different parts of an application. Windows Azure storage -addresses all three of these requirements, as Figure 7 shows. - -12 - - Figure 7: Windows Azure provides blobs, tables, and queues, all accessed in a RESTful style via HTTP, -HTTPS, or OData. - -The simplest way to store data in Windows Azure storage is to use blobs. As Figure 7 suggests, there’s a -simple hierarchy: A storage account can have one or more containers, each of which holds one or more -blobs. Blobs can be big—as large as a terabyte each—and to make transferring large blobs more efficient, -they can potentially be subdivided into blocks. If a failure occurs, retransmission can resume with the -most recent block rather than sending the entire blob again. Blobs can also have associated metadata, -such as information about where a JPEG photograph was taken or who the composer is for an MP3 file. - -Another way to use blobs is through Windows Azure drives, which can be mounted by a role instance. The -underlying storage for a drive is a blob, and so once a drive is mounted, the instance can read and write -file system data that gets stored persistently in a blob. - -Blobs are just right for some kinds of data, but they’re too unstructured for many situations. To allow -applications to work with data in a more fine-grained way, Windows Azure storage provides tables. Don’t -be misled by the name: These aren’t relational tables. In fact, even though they’re called “tables”, the -data they contain is actually stored in a set of entities with properties. A table has no defined schema; -instead, properties can have various types, such as int, string, Bool, or DateTime. And rather than using -SQL, an application can access a table’s data using the simple query language defined by OData. A single -table can be quite large, with billions of entities holding terabytes of data, and Windows Azure storage -can partition it across many servers if necessary to improve performance. - -Blobs and tables are both focused on storing data. The third option in Windows Azure storage, queues, -has a somewhat different purpose. A primary use of queues is to let Web role instances communicate -with Worker role instances. For example, a user might submit a request to perform some compute- - -13 - - intensive task via a Web page implemented by a Windows Azure Web role. The Web role instance that -receives this request can write a message into a queue describing the work to be done. A Worker role -instance that’s waiting on this queue can then read the message and carry out the task it specifies. Any -results can be returned via another queue or handled in some other way. - -Regardless of how it’s stored—in blobs, tables, or queues—all data held in Windows Azure storage is -replicated three times. This replication allows fault tolerance, since losing a copy isn’t fatal. The system -guarantees consistency, however, so an application that reads data it has just written will get what it -expects. - -Windows Azure storage can be accessed either by a Windows Azure application or by an application -running somewhere else. In both cases, all three Windows Azure storage styles use the conventions of -REST (and the OData protocol for tables) to identify and expose data. Everything is named using URIs and -accessed with standard HTTP operations, so clients can be created using .NET, Java, or other familiar -technologies. - -The Windows Azure platform charges independently for compute and storage resources. This means that -an on-premises application can use just Windows Azure storage, accessing its data in the RESTful way just -described. For example, a Windows Server application running in an enterprise data center might choose -to store backups in Windows storage blobs. - -Fabric Controller - -Even though a Windows Azure application runs in VMs, a developer doesn’t explicitly create and manage -those VMs. Instead, as described earlier, she just tells the platform how many instances the application -needs, and Windows Azure silently creates the necessary VMs and runs her application. This magic is -performed by the fabric controller, a fundamental aspect of Windows Azure. - -Along with creating VMs and starting the applications that run in them, the fabric controller also monitors -all running instances. A role instance might fail for many reasons; maybe the code threw an exception or -the VM crashed or the physical server it’s running on went down. Whatever the reason, the fabric -controller will start a new instance to maintain the total number of instances required for this application. - -The fabric controller also handles housekeeping tasks like patching the operating system and other -system software, which minimizes the need for administrators. These are all examples of why Windows -Azure is viewed as PaaS: It provides a platform that lets customers run applications without worrying -about administering the environment they run in. - -Content Delivery Network - -One common use of blobs is to store information that will be accessed from many different places. Think -of an application that serves up videos, for example, to Flash, Silverlight, or HTML 5 clients around the -world. To improve performance in situations like this, Windows Azure provides a content delivery -network. The CDN stores copies of a blob at sites closer to the clients that use it. This speeds up delivery -of frequently accessed content, improving performance for users of this information. - -14 - - Connect - -Windows Azure applications can interact with the outside world via HTTP, HTTPS, or TCP. But suppose -you’d like to connect a role in an application to a machine outside the cloud at the IP level. For example, -suppose an organization decides to move an existing ASP.NET application to Windows Azure, but wishes -to keep the application’s data in an on-premises SQL Server database. Windows Azure Connect allows -this. - -Using this option requires running Windows Azure Connect software on the on-premises machine that -contains SQL Server, and it also requires some straightforward configuration. (This configuration can be -done by the developer—there’s no need to get a network administrator involved.) Once this is done, all of -the instances in the Web role can behave as if they’re on the same IP network as the SQL Server machine. -In fact, they can use the same SQL Server connection string as when the application was running on -premises—there’s no difference. - -Note that this isn’t a full-fledged virtual private network (VPN). While Microsoft has announced plans to -let customers connect to Windows Azure via a VPN, this isn’t currently supported. It is possible to use -Windows Azure Connect to domain-join a Windows Azure application to an on-premises Active Directory, -however. Doing this allows single sign-on by on-premises users to the cloud application, and it also lets -the application use existing Active Directory accounts and groups for access control. - -The goal of application platforms, whether on-premises or in the cloud, is to support applications and -data. Windows Azure provides a home for both of these things. Going forward, expect to see a substantial -share of what would have been on-premises Windows applications instead running on this new cloud -platform. - -SQL AZURE - -A DBMS in the cloud is attractive for many reasons. Letting a specialized service provider ensure reliability -and perform essential management functions can make sense, especially for organizations that find these -things hard to do on their own. Data in the cloud can also be accessed by applications running anywhere, -even on mobile devices. And given the economies of scale that a service provider enjoys, using a cloud -database may well be cheaper than doing it yourself. - -SQL Azure provides these things, together with cloud-based reporting and data synchronization services. -And while all of them are important, the place to start is with the foundation: SQL Azure Database. - -Database - -SQL Azure Database is simple to understand. As seen by an application, it provides the core database -functions of SQL Server as a cloud service. Figure 8 shows the basics of this technology. - -15 - - Figure 8: Applications access data in SQL Azure Database through Microsoft’s TDS protocol or via OData. - -An application using SQL Azure Database might run on Windows Azure, in an enterprise’s data center, on -a mobile device, or somewhere else. Wherever it runs, the application typically accesses data via a -protocol called Tabular Data Stream (TDS). This is the same protocol used to access a local SQL Server -database, and so a SQL Azure Database application can use any existing SQL Server client. This includes -Entity Framework, ADO.NET, ODBC, PHP and others. And because SQL Azure Database looks like an -ordinary SQL Server system, standard tools can also be used, including SQL Server Management Studio, -SQL Server Integration Services, and BCP for bulk data copy. Also, as Figure 8 shows, applications can -optionally access SQL Azure Database using OData. - -Each SQL Azure account can have one or more logical servers. (Note that these provide a way to organize -the data and billing for an account—they aren’t actual instances of SQL Server.) Each server can then have -multiple databases, each of which can be up to 50 gigabytes in size. A user is free to use multiple -databases if required, spreading data across them for better performance or for other reasons. It’s also -possible to store a snapshot of one SQL Azure database into another, providing a simple backup -mechanism. - -For the most part, an application using SQL Azure Database sees a familiar SQL Server environment. A few -things are omitted in the technology’s current release, however, such as the SQL Common Language -Runtime (CLR) and support for full-text search. (Microsoft says that both will be available in a future -version.) Also, because administration is handled by Microsoft, the service doesn’t expose physical -administrative functions. A customer can’t shut down the system, for example, or interact directly with -the hardware it runs on. And as you’d expect in a shared environment, a query can run for only a limited -time—no single request can take up more than a pre-defined amount of resources. - -16 - - Yet while the environment looks quite standard, the service an application gets is more robust than what -a single instance of SQL Server provides. As in Windows Azure storage, all data stored in SQL Azure -Database is replicated three times for high availability. Also like Windows Azure storage, the service -provides strong consistency: When a write returns, the data has been made persistent. The goal is to -provide reliable data storage even in the face of system and network failures. - -Whether an application needs multiple databases or just one, SQL Azure Database can help developers -address a range of scenarios. Whatever problem is being solved, the technology’s fundamental goal -remains the same: to provide a familiar, reliable, and low-cost cloud database for many kinds of -applications. - -Reporting - -Storing data in SQL Azure Database can be useful. But as soon as that data exists, there’s likely to be a -demand for reports based on that data. Meeting this demand is the purpose of SQL Azure Reporting. -Based on SQL Server Reporting Services (SSRS), this technology provides a cloud-based way to create -reports. - -Given that SQL Azure Database looks like SQL Server to an application, it’s always possible to use SSRS on- -premises with the data it contains. So why bother to provide a cloud reporting service as well? SQL Azure -Reporting today targets two main scenarios: - - Reports created using SQL Azure Reporting can be published to a SQL Azure Reporting portal, letting - -users access it there, or made accessible directly via a URL. - - An ISV can embed reports published to the SQL Azure Reporting portal in any application, including -Windows Azure applications. This lets users access those reports without leaving the application. To -do this, the ISV can use the standard ReportViewer controls in Visual Studio—it’s no different from -embedding on-premises reports into applications. - -SQL Azure Reporting is designed to work with data stored in SQL Azure Database. Reports used with SQL -Azure Reporting are created on-premises, however, with Business Intelligence Developer Studio, the -Visual Studio-hosted tool that’s also used to create SSRS reports. In fact, SQL Azure Reporting relies on the -same reporting format, expressed in the Report Definition Language (RDL), used with SSRS. - -Like other aspects of SQL Azure, SQL Azure Reporting doesn’t provide all of the functions offered by SSRS -on premises. There’s currently no support for scheduling and subscriptions, for example, that let a report -be run and delivered at regular intervals. Still, cloud-based reporting services can be useful, and they’re an -important part of the SQL Azure technology family. - -Data Sync - -Storing data in SQL Azure Database can make it accessible to any application with an Internet connection. -Still, there are often situations where it makes sense to maintain a copy of this cloud-based data in some -other place. Suppose an organization needs to have an on-premises copy of the same information, for -example, for performance reasons or to ensure access if the network fails. For scenarios like this, the -ability to synchronize data in SQL Azure Database with other data stores is useful. - -17 - - It’s always possible to write synchronization code yourself using the Microsoft Sync Framework. To make -this easier, Microsoft provides SQL Azure Data Sync. Rather than requiring its user to write code, this -technology is entirely configuration-driven (although unsurprisingly, it’s built on the Microsoft Sync -Framework). As Figure 9 shows, SQL Azure Data Sync supports two options today. - -Figure 9: SQL Azure Data Sync can synchronize data between SQL Azure databases or between a SQL -Azure database and on-premises SQL Server databases. - -The two synchronization options are the following: - -Synchronizing data between a SQL Azure database and an on-premises SQL Server database. Having -an on-premises copy of cloud data might make sense for a variety of reasons. Maybe an organization -wants to make sure that this data is always available, for instance, even in the event of network -failures, or perhaps regulations require that a copy of the data always be available within a country’s -borders. And even though SQL Azure’s built-in data replication guards against hardware failures, the -data’s owner might choose to maintain an on-premises backup copy in case of administrative errors, -such as incorrect table deletions. - -Synchronizing data between SQL Azure databases in different Microsoft data centers. For example, -suppose an ISV or a global enterprise creates an application that’s used by people around the world. -To provide good performance for all of them, its creator might choose to run this application in three -different Windows Azure data centers: one each in North America, Europe, and Asia. If the -application stores data in SQL Azure Database, it might rely on SQL Azure Data Sync to keep -information synchronized across these three data centers. - -SQL Azure Data Sync uses a hub-and-spoke model. All changes are copied first to the SQL Azure database -hub, then out to the spokes (referred to as members). The members might be other SQL Azure databases - -18 - - or on-premises SQL Server databases. In both cases, the technology provides synchronization of an entire -database or just specific tables, with changes made to any copy propagated to all of the others. And while -it’s possible to initiate synchronization manually, SQL Azure Data Sync also includes a scheduling service. -This lets users schedule, say, hourly synchronization between a pair of databases. However it’s used, the -purpose is the same: providing a straightforward way to synchronize data between a SQL Azure database -in a Microsoft data center and one or more databases located somewhere else. - - WINDOWS AZURE APPFABRIC - -Applications exist because their business logic has value. Yet applications rely on infrastructure to provide -a foundation for that logic. A good application platform provides this infrastructure; application -developers shouldn’t need to write it themselves. - -Windows Azure AppFabric provides infrastructure for applications. The people who create applications -can benefit from different kinds of infrastructure, and so AppFabric contains various parts. This section -takes a closer look at Service Bus, Access Control, and Caching, the three components of Windows Azure -AppFabric today. - -Service Bus - -Suppose you have an application running inside your organization that exposes a Web service built with -Windows Communication Foundation (WCF). Suppose further that you’d like to connect this service -through the Internet to software running outside your organization. This client software might be running -on a cloud platform, such as Windows Azure, or inside another organization. - -At first glance, this can seem like a simple problem. Since your application provides its functionality -through Web services (either RESTful or SOAP-based), you can just make those Web services visible to the -outside world. When you actually try to do this, though, some problems appear. - -First, how can clients in other organizations find endpoints they can connect to for your service? It would -be nice to have some kind of registry where others could locate your application. And once they’ve found -it, how can requests from software in other organizations get through to your service? Network address -translation (NAT) is very common, so an application frequently doesn’t have a fixed IP address to expose -externally. And even if NAT isn’t being used, how can requests get through your firewall? It’s possible to -open firewall ports to allow access to your application, but network administrators frown on this. - -Service Bus addresses these challenges. Figure 10 shows how. - -19 - - Figure 10: A WCF service can register endpoints with Service Bus, then have clients discover and use -those endpoints to access the service. - -To begin, your WCF service registers one or more endpoints with Service Bus (step 1). For each registered -endpoint, Service Bus exposes its own corresponding endpoint (step 2). Service Bus assigns your -organization a URI root, below which you’re free to create any naming hierarchy you like. This allows your -endpoints to be assigned specific, discoverable URIs. - -When a client running in the cloud or on-premises at some other organization wishes to access your -service, it contacts the Service Bus registry (step 3) to find the endpoint, providing the endpoint’s URI. This -request uses the Atom Publishing Protocol, and it returns an AtomPub service document with references -to the endpoints Service Bus exposes on behalf of your application. Once it has these, the client can -invoke operations on the services exposed through these endpoints (step 4). For each request Service Bus -receives, it invokes the corresponding operation in the endpoint exposed by your WCF service (step 5). -(Although it’s not shown in the figure, Service Bus establishes a direct connection between an application -and its client whenever possible, making their communication more efficient.) - -There’s an obvious question here: How exactly does step 5 work? How does the request from Service Bus -back to your service deal with the challenges of NAT and firewalls? The answer is that in step 1, your -service opened a TCP connection with Service Bus for this exposed endpoint. Service Bus holds this -connection open, which solves two problems. First, NAT is no longer an issue, since traffic on the open -connection with Service Bus will always be routed to your application. Second, because the connection -was initiated from inside the firewall, there’s no problem passing information back to the application via -this connection—the firewall won’t block this traffic. - -20 - - Along with making communication easier, Service Bus can also improve security. Because clients now see -only an IP address provided by Service Bus, there’s no need to expose any IP addresses from within your -organization. This effectively makes your application anonymous, since the outside world can’t see it. -Service Bus acts as an external DMZ, providing a layer of indirection to deter attackers. - -While an application that exposes its services via Service Bus is typically implemented using WCF, clients -can be built with WCF or other technologies, such as Java. However they’re created, these clients can -make requests via TCP, HTTP, or HTTPS. Applications are also free to use their own security mechanisms, -such as encryption, to shield their communication from attackers. - -Service Bus provides a few more useful features, including these: - -Support for message buffers, which act like simple queues. Rather than require a client to make -direct calls to a service, the client can place a message of up to 256 kilobytes into a message buffer. -The message is persisted to disk—it’s durable—and the service can then read this message at some -later time. And as usual in the Windows Azure platform, the persisted messages are replicated to -guard against failures. - - The ability for multiple WCF services to listen on the same URI. Service Bus will then randomly spread -client requests across all of the listening services. The goal is to provide both load balancing and fault -tolerance for the WCF services. - -Exposing applications to the outside world isn’t as simple as it might seem. The intent of Service Bus is to -make implementing these interactions as straightforward as possible. - -Access Control - -Working with identity is a fundamental part of most distributed applications. The modern approach to -this, called claims-based identity, lets a user send a token full of claims that contain identity information. -One claim might contain the user’s name, for example, while another contains her age or a group she -belongs to. An application can use the claims in a token to decide what the user is allowed to do or in -other ways. - -In a claims-based world, tokens are issued by identity providers (IdPs). Some IdPs, such as Active Directory -Federation Services (AD FS) 2.0, exist inside organizations. Others, such as Windows Live ID and Google -Accounts, are accessible to anyone via the Internet. Each application can decide which IdPs it trusts and -thus which tokens it’s willing to accept. For example, an application running inside your company might -only accept tokens issued by your own AD FS server, while an application running on the Internet might -accept tokens issued by, say, Google and Facebook. - -Yet different IdPs use different token formats, and they represent claims in different ways. An application -that directly accepts identities from Google, Facebook, and Yahoo, for instance, would need to handle -these differences itself. But why do this? Why not instead create an intermediary that can generate a -single token format with a common representation of claims? Doing this would make life simpler for the -people who create applications, since they now need to handle only one kind of token. - -The Access Control service does exactly this, providing an intermediary in the cloud for working with -claims-based identity. Figure 11 shows how it works. - -21 - - Figure 11: The Access Control service makes it easier for applications to accept identity information -issued by different identity providers. - -As the figure shows, an application that relies on Access Control can run either on-premises or in the -cloud. In either case, the process begins when a user attempts to access this application via a browser -(step 1). The application then redirects the browser to an IdP whose token this application will accept. The -user authenticates herself with this IdP, such as by entering a username and password, and the IdP -returns a token containing claims about her (step 2). - -Next, the user’s browser sends the IdP token to Access Control (step 3). Access Control validates the -token, making sure that it really was issued by this IdP, then creates a new token according to whatever -rules have been defined for this application (step 4). Access Control contains a rules engine, allowing each -application’s administrator to define how tokens from various IdPs should be transformed into an Access -Control token. For example, if different IdPs use different formats for representing usernames, Access -Control rules can transform all of these into a common username string. Access Control then sends this -new token back to the browser (step 5), which submits it to the application (step 6). Once it has the -Access Control token, the application verifies that it really was issued by Access Control, then uses the -claims it contains (step 7). - -While this process might seem a little complicated, it actually makes life significantly simpler for the -creator of the application. Rather than handle diverse tokens containing various claims, the application -can accept identities issued by multiple identity providers while still receiving only a single token with -familiar claims. And rather than require each application to be configured to trust various IdPs, these trust -relationships are instead maintained by Access Control—an application need only trust it. - -22 - - As Figure 11 shows, Access Control has built-in support for several identity providers, including AD FS 2.0, -Windows Live ID, Google, Yahoo, and Facebook. It can also work with any IdP that supports OpenID. -Browsers and other clients can request Access Control tokens using either OAuth 2 or WS-Trust, and these -tokens can have various formats, including SAML 1.1, SAML 2.0, and Simple Web Token (SWT). And to -create applications that accept Access Control tokens, Windows developers can use Windows Identity -Foundation (WIF). (It’s worth pointing out, however, that nothing about Access Control is tied to -Windows—it could just as well be used by a Linux application that accepted only Google and Facebook -identities.) - -Working with identity is important in nearly every distributed application. The goal of Access Control is to -make it easier for developers to create secure applications that accept identities from diverse providers. -By putting this service in the cloud, Microsoft has made it available to any application running on any -platform. - -Caching - -For many applications, one of the most effective ways to improve performance is by caching frequently -accessed data. Since applications tend to use the same information over and over, making this -information more readily accessible can make the application faster. Doing this is the purpose of Windows -Azure AppFabric Caching. Figure 12 illustrates the idea. - -Figure 12: Windows Azure AppFabric Caching speeds up access to frequently accessed data for -Windows Azure applications. - -23 - - The Caching service provides a distributed cache for Windows Azure applications, along with a library to -access this cache. As Figure 12 suggests, the service includes a local cache that can keep a copy of recently -accessed data items in each of an application’s role instances. If a data item the application needs isn’t -found in the local cache, the cache library automatically contacts the shared cache provided by the -Caching service. As the figure shows, this cache is spread across a number of Windows Azure instances, -each holding different cached data. This diversity isn’t visible to applications using the cache, however. An -application just requests a data item, then lets the cache find it (if it’s in the cache) and return it from -whatever instance contains it. - -Recently accessed data isn’t automatically cached, however. Instead, the application explicitly inserts data -items in the cache, such as by using the Caching API. It’s also possible to configure an ASP.NET application -running on Windows Azure to store Session object data in the Caching service, speeding it up without -changing any of its code. - -Windows Server AppFabric, the on-premises analog to Windows Azure AppFabric, also provides a Caching -service. In fact, the two are very similar. The biggest different is that unlike its on-premises counterpart, -Windows Azure AppFabric Caching is a service—there’s no need to configure servers and administer the -cache. Instead, all of this is handled automatically by the service itself. And the cloud Caching service is -multi-tenant, so each application using it gets its own instance. Because the application must authenticate -itself to this instance, data held in the Caching service isn’t accessible to other applications. - -Caching can make applications faster and more scalable with little effort from developers, and so using it -makes sense. By providing caching as a service, Windows Azure AppFabric Caching makes this easier to -do. - -WINDOWS AZURE MARKETPLACE - -Cloud platforms are useful, but they’re just a means to an end. The real goal is offering applications and -data that provide value for people. Since the Windows Azure platform provides a foundation for both, -why shouldn’t it also include a way for people to find the applications and data that they need? - -The Windows Azure Marketplace is designed to address this need. Its two components, AppMarket and -DataMarket, let users find, try, and purchase applications and data, respectively. While both are -important, DataMarket will appear first, and so what follows focuses on this aspect of the Marketplace. - -Buying applications is common today—nearly every organization does it. Buying data is less common, but -it’s no less important. Many firms sell many different kinds of data, including demographic information, -financial information, legal information, and much more. Yet using purchased data typically requires -figuring out whether the data you need exists, finding a firm that offers it, then determining whether their -data meets your needs. - -All of this is more difficult than it needs to be. Why not create one place where customers can find all -kinds of data from all kinds of content providers? Why not let them examine the data to make sure it -meets their needs, then purchase what they need right then and there? The DataMarket component of -Windows Azure Marketplace was created to do these things. Figure 13 shows its main components. - -24 - - Figure 13: Built on Windows Azure, the DataMarket allows access to content partner datasets stored in -Windows Azure, SQL Azure, or externally. - -As the figure suggests, both people and applications can access information through DataMarket. Using a -Windows Azure-based application called the Service Explorer, a user can see what datasets are available, -then purchase what she needs. Once this is done, applications can access that data using RESTful or OData -requests. The datasets made available by DataMarket can be stored on the Windows Azure platform itself -using Windows Azure Storage or SQL Azure Database. They can also be stored externally, such as in a data -center owned by the content provider. There’s no requirement to put everything in the cloud. - -For customers, DataMarket provides a single place to find, buy, and access a variety of commercial -datasets. For content providers—the owners of those data sets—DataMarket offers a chance to expose -their offerings to more customers through Microsoft’s cloud platform. While content providers set prices -for their data, DataMarket provides a billing service that frees them from dealing directly with customers. -Microsoft will also vet the quality of the content providers, limiting DataMarket initially to just the top five -providers in a particular industry. - -Applications are free to use purchased data in any way that the content provider’s license allows. Users of -Microsoft Excel 2010, for example, can access DataMarket’s information directly using an Excel add-in or -use the PowerPivot for Excel support for OData to access information from DataMarket for data analysis. -It’s also possible to combine data purchased from DataMarket with your own data, such as for reports -created using SQL Server Reporting Services. - -25 - - Getting business value from information technology requires the right applications, but it also requires -the right data. By making it easier to discover, evaluate, and buy commercially available datasets, -DataMarket intends to make it simpler for organizations to find and use the information they need. - -LOOKING AHEAD - -Microsoft has announced a number of updates that it plans to add to the Windows Azure platform in the -near future. They include the following: - - The Windows Azure Platform Appliance: Microsoft has announced plans to offer a Windows Azure -Platform Appliance, allowing hosters and enterprises to run Windows Azure in their own data -centers. Because Windows Azure requires a specific hardware configuration—it’s not just software— -this appliance will include servers, networking, and storage, all running Windows Azure. And while it -will initially be targeted only at large organizations, Microsoft’s goal is eventually to make the -Windows Azure Platform Appliance useful for smaller organizations as well. - - CDN dynamic content caching: Today, the Windows Azure CDN works only with blob data. This -forthcoming functionality will let the CDN also cache content created dynamically by a Windows -Azure application. - - VM role snapshotting: In its first release, the Windows Azure VM role doesn’t save any changes made -to the OS volume while it’s running. Snapshotting will change this, providing a way to periodically -save the state of this volume to persistent storage. - - Better Java support: While Windows Azure can run Java applications today, Microsoft plans to make -this better. The coming improvements include better Java performance, stronger support for Eclipse- -based tools, and a more complete set of Java libraries for Windows Azure. - -Support for composite applications: To make it easier to build Windows Azure applications by -combining existing components and services, Windows Azure AppFabric will add a composition -model. Accompanied by runtime services and a Visual Studio-based designer, this model will let -developers work with the various parts of an application as a single logical entity throughout its -lifecycle. - - All of these changes target the same goal: making the Windows Azure platform useful in a broader range -of scenarios. - -CONCLUSIONS - -The truth is evident: Cloud computing is here. For developers, taking advantage of the cloud means using -cloud platforms. With the Windows Azure platform, Microsoft presents a range of options addressing a -variety of needs: - - Windows Azure provides a computing and storage environment in the cloud. - -SQL Azure provides a relational DBMS in the cloud, together with reporting and data synchronization. - -26 - - Windows Azure AppFabric offers cloud-based infrastructure supporting both cloud and on-premises - -applications. - - Windows Azure Marketplace is an online store for finding and purchasing datasets from content - -providers, together with a forthcoming store for cloud applications. - -These approaches address a variety of requirements, and not every developer will use all of them. Yet -whether you work for an ISV, a systems integrator, or an enterprise, some cloud platform services are -likely to be useful for applications your organization creates. A new world is unfolding; prepare to be part -of it. - -ABOUT THE AUTHOR - -David Chappell is Principal of Chappell & Associates (www.davidchappell.com) in San Francisco, California. -Through his speaking, writing, and consulting, he helps people around the world understand, use, and -make better decisions about new technologies. - -27 - +| | R@1 | R@3 | R@5 | R@20 | R@100 | | | R@1 | R@3 | R@5 | R@20 | | R@100 | +| -------- | ---------------------------------- | ---- | ---- | ---- | ----- | --- | -------- | ------------------------------------- | ------- | ---- | -------- | -------- | ----- | +| DPR | 29.6 | 44.5 | 51.5 | 64.6 | 74.1 | | BM25 | 3.0 | 6.6 | 9.6 | 19.3 | | 31.8 | +| GTR | 35.1 | 50.7 | 56.8 | 70.3 | 78.4 | | Oracle | 25.3 | 29.7 | 31.8 | | - | - | +| Oracle | 63.8 | 72.8 | 78.4 | - | | - | | | | | | | | +| | | | | | | | Table14: | RetrievalresultsforELI5(claimrecall). | | | | | | +| Table12: | RetrievalresultsforASQA(EMrecall). | | | | | | | | | | | | | +| | | | | | | | | | Fluency | | Correct. | Citation | | +| | R@1 | R@3 | R@5 | R@20 | R@100 | | | | | | | | | +| | | | | | | | | | (MAUVE) | | (EMRec.) | Rec. | Prec. | +| DPR | 6.7 | 13.5 | 17.6 | 30.4 | 47.6 | | | | | | | | | +| GTR | 14.6 | 24.7 | 31.6 | 49.7 | 65.6 | | | ChatGPT(VANILLA,5-doc) | | | | | | +Oracle 44.3 58.7 65.6 - - Shortinstruction 64.1 39.5 69.6 73.2 +| | | | | | | | Fullinstruction | | 66.6 | | 40.4 | 73.6 | 72.5 | +| -------- | ------------------------------------- | --- | --- | --- | --- | --- | --------------- | ------------------------------------ | ---- | --- | ---- | ---- | ---- | +| Table13: | RetrievalresultsforQAMPARI(recall-5). | | | | | | | | | | | | | +| | | | | | | | Table15: | EffectofdifferentinstructionsonASQA. | | | | | | +partiallysupports,ordoesnotsupportthefactual +| | | | | | | | G.3 EffectofDemonstrations | | | | | | | +| --------- | ------------------ | ------------- | --------------------- | ---------------- | --- | ------ | ------------------------------------------ | --------------- | ------ | ------ | --------- | ------- | -------- | +| claimsins | . Citationc | | hasacitationprecision | | | | | | | | | | | +| | i | | i,j | | | | | | | | | | | +| of1ifs | hasarecallof1,andc | | | fullyorpartially | | | | | | | | | | +| | i | | | i,j | | | Table16showsresultsoneffectofdifferentnum- | | | | | | | +| supports | s . Finally, | | we take | an average | of | preci- | | | | | | | | +| | i | | | | | | bers of | demonstrations. | | We | see that | numbers | of | +| sion | scores of | all citations | | in the statement | | S to | | | | | | | | +| | | | | | | | demonstrations | | do not | affect | ChatGPT’s | | correct- | +obtainthecitationprecisionscore. +nessbutusingatleastonedemonstrationensures +| | | | | | | | highcitationrecall. | | FortheoriginalLLaMAmodel, | | | | | +| --- | --- | --- | --- | --- | --- | --- | ------------------- | --- | ------------------------- | --- | --- | --- | --- | +G MoreExperiments +Table16showsthetrendthatmoredemonstrations +leadtobetterperformance. +G.1 RetrievalAnalysis +| | | | | | | | G.4 Fine-tunedModels | | | | | | | +| --- | --- | --- | --- | --- | --- | --- | -------------------- | --- | --- | --- | --- | --- | --- | +Oracle. Sincetheoriginaldatasetsdonotcontain +goldpassagesatthesamegranularitylevelasour +Tobetterunderstandthedifferencesbetweenfine- +setting(100-wordpassages),weapproximategold +| | | | | | | | tuned models | | and prompted | | large | language | mod- | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | ------------ | --- | ----- | -------- | ---- | +passagesbyrunningthefollowingalgorithmonthe els, we train state-of-the-art question answer- +| top-100retrievedpassages. | | | | Wefirstcalculatethe | | | | | | | | | | +| ------------------------- | --------- | ---- | -------- | ------------------- | ------- | --- | ---------- | ----------------- | --- | --- | ----- | ------- | --- | +| | | | | | | | ing model, | Fusion-in-Decoder | | | (FiD; | Izacard | and | +| recall | score for | each | passage. | Then, | we sort | the | | | | | | | | +Grave(2021)),andevaluateitinconjunctionwith +passagesusingtheirrecallscoreandtakethetop5 POSTCITE. Due to the lack of training data with +| passagesasourinitialoracleset. | | | | Finally,weiterate | | | | | | | | | | +| ------------------------------ | --- | --- | --- | ----------------- | --- | --- | -------- | ----------- | --- | ----- | ----- | --------- | --- | +| | | | | | | | citation | annotation, | we | first | train | a T5-base | FiD | +throughallpassagesthatwerenotinitiallyintheor- +modelfor5epochsontheASQAtrainingsetwith +aclesetandtrytoreplacethepassagesintheoracle abatchsizeof64andalearningrateof1e-4. Dur- +setinagreedyfashion: wecalculatethechangein ingevaluation,weuse POSTCITE toaddcitations +therecallscoreoftheoraclesetforeverypossible +| | | | | | | | totheoutput. | | Wealsousek | | = 5passagesduring | | | +| --- | --- | --- | --- | --- | --- | --- | ------------ | --- | ---------- | --- | ----------------- | --- | --- | +replacementandproceedwiththereplacementthat +bothtrainingandevaluationoftheFiDmodel. +| resultsinthelargestrecallimprovement. | | | | | | Theset | | | | | | | | +| ------------------------------------- | --- | --- | --- | --- | --- | ------ | ----- | ----------- | --- | ---------- | --- | ------- | ---- | +| | | | | | | | Then, | we evaluate | | this model | | on both | ASQA | +of5oraclepassageswereabletomatchtherecall +(in-domain)andELI5(out-of-domain),andthere- +scoresofthetop-100retrievedpassages. +| | | | | | | | sultscanbefoundinTables17and18. | | | | | | Notethat | +| --- | --- | --- | --- | --- | --- | --- | ------------------------------- | --- | --- | --- | --- | --- | -------- | +Detailed retrieval results. We show detailed re- thisisnotadirectcomparison,asALCEassumes +trievalresultsinTables12,13,and14. only evaluation data available and uses only few- +| | | | | | | | shotdataforprompting. | | | Astheresultsshow, | | | the | +| --- | --- | --- | --- | --- | --- | --- | --------------------- | --- | --- | ----------------- | --- | --- | --- | +FiDbaselinestillsignificantlylagsbehindprompt- +G.2 EffectofInstructions +ingChatGPTinbothcorrectnessandcitationqual- +Table 15 shows results of using a full instruction ity(eventhoughitistrainedon4000+examples). +(Table 23) and a short version of the instruction When tested on another dataset (ELI5), FiD per- +(Table 24). We see that the full version induces forms even worse, showing that it is challenging +strongercorrectnessandcitationrecall,whilethe to solve the problem by fine-tuning a small pre- +| twoinstructionsleadtosimilarcitationprecision. | | | | | | | trainedmodel. | | | | | | | +| ---------------------------------------------- | --- | --- | --- | --- | --- | --- | ------------- | --- | --- | --- | --- | --- | --- | + +| | | Fluency | Correct. | | Citation | | | | Fluency | Correct. | Citation | +| --- | --- | ------- | -------- | ---- | -------- | ----- | ------- | --- | ------- | -------- | ---------- | +| | | (MAUVE) | | (EM) | Rec. | Prec. | | | (MAUVE) | (Claim) | Rec. Prec. | +| | | | | | | | ChatGPT | | 57.2 | 12.0 | 51.1 50.0 | +ChatGPT(VANILLA) +| | | | | | | | FiD+POSTCITE | | 25.2 | 4.4 | 39.3 39.3 | +| ------- | --- | ---- | --- | ---- | ---- | ---- | ------------ | --- | ---- | --- | --------- | +| #demo=0 | | 74.5 | | 41.9 | 69.3 | 73.4 | | | | | | +| #demo=1 | | 68.9 | | 39.8 | 74.6 | 73.2 | | | | | | +Table18: ComparisonofFusion-in-DecoderwithChat- +| #demo=2 | | 66.6 | | 40.4 | 73.6 | 72.5 | | | | | | +| -------- | ------------------------------ | ---- | --- | ---- | ---- | ---- | ---------------- | ------------------------------ | ------------ | -------- | ----------- | +| | | | | | | | GPTonELI5. | Bothmodelsusetop-5GTRpassages. | | | | +| Table16: | DifferentdemonstrationsonASQA. | | | | | | | | | | | +| | | | | | | | both correctness | | and citation | quality, | the largest | +Fluency Correct. Citation instruction-followingmodels(i.e. LLaMA-2-70B- +ChatandStableBeluga2)cansometimesachieve +| | | (MAUVE) | | (EMRec.) | Rec. | Prec. | | | | | | +| --- | --- | ------- | --- | -------- | ---- | ----- | --- | --- | --- | --- | --- | +competitivecorrectnesstotheSoTA. +| ChatGPT | | 66.6 | | 40.4 | 73.6 | 72.5 | | | | | | +| ------- | --- | ---- | --- | ---- | ---- | ---- | --- | --- | --- | --- | --- | +FiD+POSTCITE 75.8 28.4 58.1 58.0 Furthermore, we found that the open-source +| | | | | | | | models follow | a | similar trend | between | different | +| -------- | -------------------------------------- | --- | --- | --- | --- | --- | -------------------------------------- | --- | ------------- | ------- | --------- | +| Table17: | ComparisonofFusion-in-DecoderwithChat- | | | | | | | | | | | +| | | | | | | | approachesasChatGPT.Specifically,using | | | | SUMM | +GPTonASQA.Bothmodelsusetop-5GTRpassages. +| | | | | | | | and SNIPPET | improvescorrectnessand | | | RERANK | +| --- | --- | --- | --- | --- | --- | --- | ----------- | ---------------------- | --- | --- | ------ | +boostscitationquality. +G.5 MoreHumanEvaluation +H Prompts +Weevaluatetheaccuracyofourautomaticmetrics +by treating the human annotations as gold labels. We show detailed prompts used in our paper in +Tables23,24,25,26,27,28,and29. +Forcitationrecall,ALCEachievesanaccuracyof +85.1%;forcitationprecision,ALCEhasanaccu- +I Examples +| racy of | 77.6%. | Regarding | | detecting | insufficient | | | | | | | +| ------- | ------ | --------- | --- | --------- | ------------ | --- | --- | --- | --- | --- | --- | +citations, ALCE has a recall of 82.3% and a pre- In Tables 30 and 31 we show some examples of +| cision of | 84.2%; | regarding | | detecting | “irrelevant” | | | | | | | +| --------- | ------ | --------- | --- | --------- | ------------ | --- | --- | --- | --- | --- | --- | +questionsandmodelgeneratedoutputs. +| citations, | ALCE | has | a recall | of 75.6% | | and a pre- | | | | | | +| ------------ | ---------- | --- | -------- | --------- | -------------- | ---------- | --- | --- | --- | --- | --- | +| cision of | 66.1%—ALCE | | is | effective | in | detecting | | | | | | +| “irrelevant” | citations, | | but due | to | the limitation | of | | | | | | +theNLImodel(cannotdetect“partialsupport”),it +hasarelativelyhighfalsepositiverate. +G.6 MainResults +Weshowfullresultsofourexperimentsalongwith +| thestandarddeviationinTables19,20,and21. | | | | | | We | | | | | | +| ---------------------------------------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +repeatallexperimentswiththreedifferentrandom +| seeds. However, | | for | ChatGPT | RERANK, | | we use | | | | | | +| --------------- | ------ | --- | ------- | ------- | ----------- | ------ | --- | --- | --- | --- | --- | +| only one | seeded | run | since | each | run repeats | the | | | | | | +generationstepfourtimes,andmoreexperiments +| wouldincursignificantcosts. | | | | Similarly,duetothe | | | | | | | | +| --------------------------- | --- | --- | --- | ------------------ | --- | --- | --- | --- | --- | --- | --- | +costofrunningChatGPT-16KandGPT-4,weonly +useoneseededrunforeachmodel. +G.7 Open-sourceModels +Inadditiontotheopen-sourcemodelsdiscussedin +themaintext,wealsoshowtheresultsofLLaMA- +| 7B, Alpaca-7B, | | Vicuna-7B, | | LLaMA-33B, | | Oasst- | | | | | | +| -------------- | --- | ---------- | --- | ---------- | --- | ------ | --- | --- | --- | --- | --- | +33B,andStableBeluga2intheTables19,20,and +21. Forselectedmodels,wealsotestedthemusing +| approachesbeyond | | VANILLA. | | | | | | | | | | +| ---------------- | --- | -------- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +Althoughopen-sourcemodelsgenerallylagbe- +| hind the | state-of-the-art | | models | | (i.e. | GPT-4) in | | | | | | +| -------- | ---------------- | --- | ------ | --- | ----- | --------- | --- | --- | --- | --- | --- | + +| | Fluency | Correct. | Citation | | | | +| --- | ------- | -------- | -------- | ----- | ------- | ------ | +| | (MAUVE) | (EMRec.) | Rec. | Prec. | ROUGE-L | Length | +ChatGPT +| VANILLA(5-psg) | 66.8 | 40.4 | 73.6 | 72.5 | 37.0 | 40.0 | +| --------------- | ----- | ----- | ----- | ----- | ----- | ----- | +| | (2.0) | (0.6) | (1.1) | (1.8) | (0.4) | (3.1) | +| w/RERANK | 77.0 | 40.2 | 84.8 | 81.6 | 36.9 | 40.8 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| SUMM(10-psg) | 70.0 | 43.3 | 68.8 | 61.8 | 36.9 | 49.8 | +| | (1.2) | (0.8) | (0.6) | (1.1) | (0.2) | (4.3) | +| w/INTERACT | 69.0 | 39.1 | 73.4 | 66.5 | 35.7 | 34.0 | +| | (2.7) | (0.5) | (0.2) | (4.9) | (0.2) | (0.9) | +| SNIPPET(10-psg) | 69.8 | 41.4 | 65.3 | 57.4 | 36.4 | 43.0 | +| | (2.5) | (0.6) | (0.6) | (0.9) | (0.4) | (3.5) | +| INLINESEARCH | 58.7 | 32.4 | 58.3 | 58.3 | 58.2 | 23.7 | +| | (1.3) | (0.6) | (1.3) | (1.3) | (1.1) | (1.1) | +| CLOSEDBOOK | 52.7 | 38.2 | 26.7 | 26.7 | 37.1 | 61.1 | +| | (4.9) | (0.1) | (1.1) | (1.1) | (0.3) | (4.5) | +| ORACLE(5-psg) | 64.4 | 48.9 | 74.5 | 72.7 | 38.2 | 37.4 | +| | (0.6) | (1.2) | (0.6) | (1.0) | (1.0) | (3.0) | +ChatGPT-16K +VANILLA(5-psg) 60.3 (−) 36.1 (−) 76.2 (−) 76.5 (−) 36.2 (−) 24.7 (−) +VANILLA(10-psg) 56.3 (−) 36.7 (−) 75.3 (−) 75.0 (−) 35.6 (−) 23.5 (−) +VANILLA(20-psg) 56.7 (−) 36.1 (−) 73.7 (−) 73.5 (−) 35.5 (−) 23.1 (−) +GPT-4 +| VANILLA(5-psg) | 67.1 | 41.3 | 68.5 | 75.6 | 39.2 | 31.8 | +| --------------- | ---- | ---- | ---- | ---- | ---- | ---- | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(10-psg) | 71.5 | 43.1 | 72.0 | 75.5 | 39.7 | 33.8 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(20-psg) | 64.9 | 44.4 | 73.0 | 76.5 | 40.1 | 34.3 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +Open-source +| LLaMA-7BVANILLA(3-psg) | 69.8 | 22.6 | 6.2 | 9.2 | 29.1 | 61.3 | +| ------------------------- | ------ | ----- | ----- | ----- | ----- | ------ | +| | (2.0) | (0.9) | (2.7) | (2.9) | (0.2) | (14.3) | +| Alpaca-7BVANILLA(3-psg) | 84.2 | 32.1 | 12.3 | 14.1 | 33.1 | 51.7 | +| | (2.7) | (1.7) | (7.2) | (7.0) | (0.8) | (12.8) | +| Vicuna-7BVANILLA(3-psg) | 82.9 | 34.6 | 40.3 | 42.6 | 35.9 | 48.9 | +| | (5.0) | (0.7) | (0.5) | (1.0) | (0.7) | (6.6) | +| LLaMA-13BVANILLA(3-psg) | 68.4 | 26.9 | 10.6 | 15.4 | 29.8 | 67.1 | +| | (6.4) | (0.4) | (4.7) | (5.2) | (0.5) | (19.1) | +| w/RERANK | 60.9 | 25.2 | 28.1 | 37.0 | 27.9 | 50.5 | +| | (14.5) | (2.5) | (9.3) | (7.2) | (2.4) | (14.3) | +| LLaMA-13BSUMM(10-psg) | 76.8 | 33.3 | 19.6 | 23.7 | 32.1 | 54.4 | +| | (4.7) | (0.7) | (3.9) | (4.7) | (0.3) | (1.5) | +| LLaMA-13BSNIPPET(10-psg) | 72.0 | 31.3 | 18.2 | 21.1 | 30.8 | 50.5 | +| | (0.8) | (1.1) | (3.1) | (3.6) | (0.4) | (4.5) | +| LLaMA-13BORACLE(3-psg) | 69.5 | 34.3 | 10.8 | 15.8 | 30.6 | 67.3 | +| | (11.4) | (0.9) | (4.9) | (5.9) | (0.1) | (17.9) | +| Vicuna-13BVANILLA(3-psg) | 82.6 | 31.9 | 51.1 | 50.1 | 34.9 | 39.1 | +| | (9.4) | (3.9) | (1.4) | (2.5) | (1.3) | (6.6) | +| w/RERANK | 73.5 | 32.9 | 71.9 | 65.4 | 34.6 | 35.7 | +| | (2.1) | (1.3) | (1.9) | (1.5) | (0.3) | (4.2) | +| Vicuna-13BSUMM(10-psg) | 67.7 | 43.2 | 52.7 | 50.0 | 36.7 | 66.0 | +| | (0.3) | (0.1) | (2.6) | (2.1) | (0.2) | (1.2) | +| Vicuna-13BSNIPPET(10-psg) | 81.4 | 42.1 | 53.4 | 48.7 | 36.9 | 61.2 | +| | (3.0) | (1.2) | (1.9) | (1.6) | (0.4) | (7.4) | +| Vicuna-13BORACLE(3-psg) | 72.9 | 42.5 | 52.2 | 50.7 | 36.5 | 38.7 | +| | (3.5) | (1.6) | (0.8) | (1.6) | (0.9) | (3.5) | +LLaMA-33BVANILLA(3-psg) 83.7 (5.4) 31.0 (0.8) 19.5 (5.3) 23.0 (5.3) 32.3 (0.6) 44.1 (9.3) +w/RERANK 82.1 (3.0) 31.3 (1.1) 41.3 (6.4) 44.7 (5.5) 32.5 (0.9) 39.4 (8.0) +LLaMA-33BSUMM(10-psg) 72.0 (3.0) 33.1 (1.9) 34.7 (5.8) 35.2 (6.0) 31.1 (0.8) 43.7 (5.0) +LLaMA-33BSNIPPET(10-psg) 70.8 (3.1) 30.9 (1.4) 31.4 (4.2) 31.5 (5.3) 30.1 (0.7) 42.8 (3.6) +LLaMA-33BORACLE(3-psg) 82.6 (7.1) 39.3 (2.9) 20.2 (6.2) 23.9 (6.3) 33.1 (0.9) 42.0 (9.3) +Oasst-33BVANILLA(3-psg) 82.9 (2.7) 34.8 (1.5) 36.2 (1.7) 38.3 (2.7) 35.5 (0.7) 45.2 (6.3) +w/RERANK 83.2 (2.4) 35.1 (1.4) 66.7 (0.2) 64.3 (1.0) 35.0 (0.6) 41.8 (6.0) +Oasst-33BSUMM(10-psg) 74.3 (4.6) 40.9 (1.1) 45.5 (1.9) 44.0 (2.9) 35.8 (0.6) 54.3 (4.8) +Oasst-33BSNIPPET(10-psg) 79.3 (1.0) 40.1 (0.9) 45.0 (1.3) 43.3 (2.2) 35.8 (0.2) 50.9 (4.1) +Oasst-33BORACLE(3-psg) 85.1 (2.8) 44.3 (2.4) 37.0 (1.0) 39.6 (1.5) 36.5 (1.1) 44.2 (5.8) +LLaMA-2-7B-ChatVANILLA(5-psg) 80.1 (6.5) 33.9 (2.1) 50.9 (4.5) 47.5 (3.7) 35.1 (0.9) 42.3 (10.1) +LLaMA-2-13B-ChatVANILLA(5-psg) 72.4 (6.3) 35.2 (1.2) 38.4 (5.9) 39.4 (4.8) 35.8 (0.9) 38.0 (6.4) +LLaMA-2-70B-ChatVANILLA(5-psg) 88.3 (4.1) 41.5 (0.8) 62.9 (1.4) 61.3 (2.1) 37.1 (0.4) 52.9 (9.5) +| StableBeluga2VANILLA(5-psg) | 59.7 | 37.2 | 63.6 | 63.5 | 35.3 | 23.8 | +| --------------------------- | -------- | ---------------- | ----- | ----- | ----- | ----- | +| | (3.3) | (1.0) | (0.7) | (0.1) | (0.7) | (3.2) | +| | Table19: | ASQAfullresults. | | | | | + +| | Correctness | Citation | | +| --- | ------------ | ---------- | -------- | +| | Rec.-5 Prec. | Rec. Prec. | NumPred. | +ChatGPT +| VANILLA(5-psg) | 20.8 20.8 | 20.5 20.9 | 5.0 | +| --------------- | ----------- | ----------- | ----- | +| | (2.2) (0.2) | (0.7) (0.7) | (0.5) | +| w/RERANK | 22.8 21.4 | 21.2 21.4 | 5.4 | +| | (−) (−) | (−) (−) | (−) | +| SUMM(10-psg) | 23.6 21.2 | 23.6 25.7 | 6.7 | +| | (0.9) (0.5) | (0.7) (0.8) | (0.4) | +| SNIPPET(10-psg) | 24.5 21.5 | 22.9 24.9 | 7.2 | +| | (1.4) (1.8) | (1.6) (0.4) | (0.9) | +| w/INTERACT | 21.9 23.0 | 21.9 23.4 | 6.7 | +| | (0.9) (0.4) | (1.2) (0.9) | (0.4) | +| INLINESEARCH | 17.2 20.4 | 14.9 14.9 | 6.7 | +| | (1.1) (0.8) | (0.8) (0.8) | (0.2) | +| CLOSEDBOOK | 32.9 19.8 | 10.0 10.0 | 17.0 | +| | (1.1) (1.6) | (0.4) (0.4) | (2.9) | +| ORACLE | 37.0 36.9 | 24.1 24.6 | 5.3 | +| | (3.1) (0.6) | (1.2) (1.3) | (0.6) | +ChatGPT-16K +| VANILLA(5-psg) | 21.1 22.0 | 20.7 21.2 | 4.9 | +| --------------- | --------- | --------- | --- | +| | (−) (−) | (−) (−) | (−) | +| VANILLA(10-psg) | 23.4 21.9 | 21.6 22.0 | 5.7 | +| | (−) (−) | (−) (−) | (−) | +| VANILLA(20-psg) | 26.4 21.1 | 19.4 19.7 | 7.6 | +| | (−) (−) | (−) (−) | (−) | +GPT-4 +| VANILLA(5-psg) | 22.2 25.0 | 25.9 27.0 | 4.4 | +| --------------- | --------- | --------- | --- | +| | (−) (−) | (−) (−) | (−) | +| VANILLA(10-psg) | 26.8 25.1 | 26.2 27.2 | 5.7 | +| | (−) (−) | (−) (−) | (−) | +| VANILLA(20-psg) | 29.6 26.2 | 27.4 28.5 | 6.8 | +| | (−) (−) | (−) (−) | (−) | +Open-source +| LLaMA-7BVANILLA(3-psg) | 7.8 7.4 | 5.1 5.7 | 5.7 | +| ------------------------------ | ----------- | ----------- | ----- | +| | (3.4) (2.7) | (0.5) (0.8) | (0.6) | +| Alpaca-7BVANILLA(3-psg) | 9.4 9.5 | 6.4 6.8 | 5.1 | +| | (3.7) (3.6) | (0.5) (0.5) | (0.1) | +| Vicuna-7BVANILLA(3-psg) | 11.3 13.3 | 10.1 10.9 | 3.9 | +| | (1.4) (2.3) | (0.6) (0.5) | (0.3) | +| LLaMA-13BVANILLA(3-psg) | 9.7 9.1 | 6.7 7.1 | 5.9 | +| | (3.6) (3.1) | (0.9) (0.9) | (0.6) | +| w/RERANK | 10.0 10.7 | 9.9 10.2 | 5.4 | +| | (3.3) (3.3) | (1.2) (1.1) | (0.5) | +| LLaMA-13BSUMM(10-psg) | 14.8 12.6 | 7.4 8.0 | 8.1 | +| | (2.5) (1.5) | (0.5) (0.6) | (0.9) | +| LLaMA-13BSNIPPET(10-psg) | 17.7 15.7 | 8.8 9.9 | 8.2 | +| | (1.4) (0.9) | (0.7) (0.6) | (0.4) | +| LLaMA-13BORACLE(3-psg) | 16.8 15.4 | 7.7 8.3 | 5.7 | +| | (6.6) (5.6) | (1.0) (1.1) | (0.7) | +| Vicuna-13BVANILLA(3-psg) | 14.0 15.9 | 12.5 13.4 | 4.7 | +| | (0.6) (1.7) | (0.8) (0.7) | (0.3) | +| w/RERANK | 13.0 17.2 | 17.3 17.7 | 4.4 | +| | (0.7) (2.2) | (0.8) (0.6) | (0.3) | +| Vicuna-13BSUMM(10-psg) | 21.1 17.1 | 15.7 17.8 | 6.9 | +| | (1.4) (0.3) | (0.2) (0.1) | (0.7) | +| Vicuna-13BSNIPPET(10-psg) | 21.9 18.2 | 16.8 19.7 | 7.5 | +| | (0.8) (0.3) | (0.3) (0.6) | (0.4) | +| Vicuna-13BORACLE(3-psg) | 25.9 28.4 | 15.8 16.8 | 4.9 | +| | (1.6) (2.6) | (1.4) (1.4) | (0.5) | +| LLaMA-33BVANILLA(3-psg) | 14.7 12.0 | 7.9 8.3 | 7.2 | +| | (3.3) (2.2) | (0.7) (0.6) | (0.7) | +| w/RERANK | 14.0 13.9 | 10.7 11.1 | 6.4 | +| | (3.4) (2.6) | (0.6) (0.5) | (0.7) | +| LLaMA-33BSUMM(10-psg) | 19.0 14.8 | 12.5 15.0 | 7.6 | +| | (1.9) (0.8) | (0.2) (0.3) | (0.6) | +| LLaMA-33BSNIPPET(10-psg) | 19.6 15.7 | 12.8 15.2 | 7.8 | +| | (1.1) (0.1) | (1.1) (1.2) | (0.5) | +| LLaMA-33BORACLE(3-psg) | 23.9 20.3 | 9.8 10.4 | 6.8 | +| | (6.9) (5.2) | (1.2) (1.2) | (0.9) | +| Oasst-33BVANILLA(3-psg) | 15.5 14.9 | 9.0 10.1 | 5.6 | +| | (1.5) (1.4) | (1.6) (1.8) | (0.3) | +| w/RERANK | 14.1 15.8 | 15.0 15.9 | 4.7 | +| | (1.1) (1.0) | (1.6) (1.6) | (0.3) | +| Oasst-33BSUMM(10-psg) | 21.0 17.5 | 12.9 16.6 | 7.1 | +| | (0.6) (1.0) | (1.2) (1.2) | (0.4) | +| Oasst-33BSNIPPET(10-psg) | 22.0 17.4 | 13.6 17.7 | 7.5 | +| | (0.4) (0.3) | (1.7) (1.6) | (0.1) | +| Oasst-33BORACLE(3-psg) | 26.9 26.0 | 11.7 12.9 | 5.6 | +| | (3.7) (3.3) | (1.0) (1.2) | (0.4) | +| LLaMA-2-7B-ChatVANILLA(5-psg) | 16.2 15.3 | 10.6 10.9 | 5.5 | +| | (1.3) (1.6) | (0.9) (1.0) | (0.0) | +| LLaMA-2-13B-ChatVANILLA(5-psg) | 21.1 18.2 | 9.6 9.7 | 6.5 | +| | (0.9) (0.5) | (1.5) (1.5) | (0.3) | +| LLaMA-2-70B-ChatVANILLA(5-psg) | 21.8 18.4 | 15.1 15.6 | 7.1 | +| | (0.7) (0.1) | (1.2) (1.3) | (0.2) | +| StableBeluga2 | 19.7 21.2 | 18.6 20.7 | 5.1 | +| | (0.7) (1.4) | (1.2) (1.1) | (0.3) | +Table20: QAMPARIfullresults. + +| | Fluency | Correct. | Citation | | | | +| --- | ------- | -------- | -------- | ----- | ------- | ------ | +| | (MAUVE) | (Claim) | Rec. | Prec. | ROUGE-L | Length | +ChatGPT +| VANILLA(5-psg) | 57.2 | 12.0 | 51.1 | 50.0 | 20.6 | 91.5 | +| --------------- | ----- | ----- | ----- | ----- | ----- | ------ | +| | (1.6) | (0.6) | (4.2) | (4.8) | (0.2) | (6.5) | +| w/RERANK | 56.1 | 11.4 | 69.3 | 67.8 | 20.3 | 103.4 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| SUMM(10-psg) | 40.2 | 12.5 | 51.5 | 48.2 | 20.3 | 90.0 | +| | (1.2) | (0.2) | (1.1) | (2.0) | (0.1) | (6.6) | +| SNIPPET(10-psg) | 62.9 | 14.3 | 50.4 | 45.0 | 21.0 | 100.0 | +| | (2.2) | (0.1) | (1.1) | (2.w) | (0.1) | (6.8) | +| w/INTERACT | 68.0 | 13.3 | 47.8 | 45.0 | 20.1 | 99.8 | +| | (5.8) | (0.5) | (3.3) | (3.1) | (0.2) | (6.1) | +| INLINESEARCH | 49.7 | 13.4 | 45.6 | 43.7 | 20.4 | 103.0 | +| | (4.6) | (1.1) | (2.5) | (3.9) | (0.3) | (18.1) | +| CLOSEDBOOK | 32.6 | 18.6 | 15.4 | 15.4 | 22.8 | 108.3 | +| | (1.1) | (0.5) | (0.3) | (0.3) | (0.1) | (8.9) | +| ORACLE(5-psg) | 59.4 | 21.3 | 57.8 | 56.0 | 21.2 | 93.0 | +| | (4.1) | (0.2) | (3.7) | (3.8) | (0.3) | (7.8) | +ChatGPT-16K +| VANILLA(5-psg) | 31.6 | 14.4 | 44.6 | 44.1 | 21.4 | 87.6 | +| --------------- | ---- | ---- | ---- | ---- | ---- | ---- | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(10-psg) | 26.6 | 14.4 | 45.5 | 43.3 | 21.5 | 87.5 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(20-psg) | 31.6 | 15.9 | 43.4 | 40.9 | 21.7 | 92.6 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +GPT-4 +| VANILLA(5-psg) | 38.4 | 14.2 | 44.0 | 50.1 | 20.6 | 79.6 | +| --------------- | ---- | ---- | ---- | ---- | ---- | ---- | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(10-psg) | 39.9 | 15.7 | 49.5 | 54.2 | 21.2 | 88.2 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +| VANILLA(20-psg) | 41.5 | 18.3 | 48.5 | 53.4 | 22.2 | 97.0 | +| | (−) | (−) | (−) | (−) | (−) | (−) | +Open-source +| LLaMA-7BVANILLA(3-psg) | 28.6 | 1.6 | 1.2 | 2.7 | 12.2 | 46.9 | +| ------------------------------ | ------ | ----- | ----- | ----- | ----- | ----- | +| | (17.9) | (0.9) | (0.0) | (0.1) | (1.3) | (1.2) | +| Alpaca-7BVANILLA(3-psg) | 45.9 | 9.2 | 4.5 | 5.2 | 18.8 | 67.1 | +| | (5.3) | (0.1) | (1.6) | (1.9) | (0.3) | (1.2) | +| Vicuna-7BVANILLA(3-psg) | 43.2 | 10.0 | 12.6 | 16.3 | 19.1 | 68.7 | +| | (3.9) | (0.5) | (2.3) | (2.6) | (0.4) | (2.0) | +| LLaMA-13BVANILLA(3-psg) | 50.0 | 3.9 | 3.1 | 5.3 | 16.1 | 63.3 | +| | (2.0) | (0.4) | (0.9) | (1.3) | (0.5) | (2.0) | +| w/RERANK | 46.7 | 4.3 | 9.7 | 15.0 | 16.1 | 63.0 | +| | (2.9) | (0.4) | (2.1) | (2.2) | (0.7) | (2.3) | +| LLaMA-13BSUMM(10-psg) | 28.6 | 2.9 | 2.5 | 3.8 | 8.5 | 33.1 | +| | (1.8) | (0.1) | (0.8) | (0.8) | (0.3) | (0.6) | +| LLaMA-13BSNIPPET(10-psg) | 48.4 | 5.7 | 5.8 | 7.6 | 15.1 | 60.2 | +| | (3.1) | (0.9) | (0.6) | (0.9) | (1.1) | (3.2) | +| LLaMA-13BORACLE(3-psg) | 49.5 | 6.4 | 3.7 | 6.5 | 16.8 | 64.5 | +| | (2.4) | (0.6) | (0.7) | (1.0) | (0.5) | (1.7) | +| Vicuna-13BVANILLA(3-psg) | 58.2 | 10.0 | 15.6 | 19.6 | 19.1 | 69.6 | +| | (25.1) | (0.3) | (2.2) | (2.0) | (0.3) | (0.6) | +| w/RERANK | 45.9 | 9.2 | 31.7 | 38.2 | 18.6 | 69.7 | +| | (4.3) | (0.0) | (2.9) | (1.6) | (0.5) | (1.0) | +| Vicuna-13BSUMM(10-psg) | 22.4 | 4.9 | 9.7 | 12.2 | 9.3 | 33.0 | +| | (3.0) | (0.1) | (1.3) | (1.2) | (0.4) | (3.7) | +| Vicuna-13BSNIPPET(10-psg) | 48.1 | 11.2 | 27.2 | 27.9 | 18.4 | 76.8 | +| | (5.3) | (1.4) | (3.6) | (1.9) | (1.9) | (8.7) | +| Vicuna-13BORACLE(3-psg) | 41.6 | 17.1 | 20.2 | 26.5 | 20.0 | 72.0 | +| | (3.1) | (0.4) | (3.0) | (3.0) | (0.3) | (0.3) | +| LLaMA-33BVANILLA(3-psg) | 58.8 | 6.2 | 9.3 | 12.1 | 16.9 | 60.0 | +| | (4.3) | (0.0) | (3.0) | (4.2) | (0.2) | (1.3) | +| w/RERANK | 65.9 | 6.0 | 22.5 | 26.1 | 17.5 | 61.0 | +| | (2.5) | (0.7) | (5.2) | (6.9) | (0.4) | (1.2) | +| LLaMA-33BSUMM(10-psg) | 23.3 | 3.0 | 6.2 | 8.2 | 7.5 | 26.2 | +| | (2.0) | (0.2) | (0.5) | (0.7) | (0.4) | (2.3) | +| LLaMA-33BSNIPPET(10-psg) | 53.2 | 7.4 | 13.7 | 15.1 | 14.4 | 53.3 | +| | (4.0) | (1.3) | (0.5) | (0.4) | (1.7) | (8.5) | +| LLaMA-33BORACLE(3-psg) | 63.7 | 11.4 | 11.9 | 15.4 | 17.9 | 61.7 | +| | (2.8) | (0.5) | (2.6) | (3.6) | (0.2) | (2.6) | +| Oasst-33BVANILLA(3-psg) | 46.8 | 9.5 | 16.0 | 21.6 | 18.6 | 67.8 | +| | (7.6) | (0.2) | (2.5) | (3.5) | (0.3) | (1.5) | +| w/RERANK | 52.1 | 8.5 | 34.4 | 41.5 | 18.2 | 67.0 | +| | (6.1) | (0.5) | (2.9) | (2.5) | (0.3) | (1.5) | +| Oasst-33BSUMM(10-psg) | 24.8 | 3.9 | 12.3 | 16.3 | 9.1 | 31.6 | +| | (2.8) | (0.3) | (0.2) | (0.3) | (0.3) | (1.5) | +| Oasst-33BSNIPPET(10-psg) | 50.7 | 10.7 | 25.8 | 26.7 | 17.8 | 69.6 | +| | (4.6) | (1.2) | (3.3) | (2.3) | (1.8) | (8.6) | +| Oasst-33BORACLE(3-psg) | 50.7 | 15.8 | 20.8 | 28.0 | 19.4 | 70.3 | +| | (12.1) | (0.1) | (2.8) | (3.2) | (0.1) | (1.1) | +| LLaMA-2-7B-ChatVANILLA(5-psg) | 27.8 | 10.9 | 19.8 | 15.0 | 20.5 | 87.8 | +| | (3.0) | (0.2) | (1.2) | (1.4) | (0.2) | (8.1) | +| LLaMA-2-13B-ChatVANILLA(5-psg) | 34.7 | 13.4 | 17.3 | 15.8 | 20.9 | 88.3 | +| | (1.5) | (0.4) | (1.3) | (1.4) | (0.2) | (6.3) | +LLaMA-2-70B-ChatVANILLA(5-psg) 38.6 12.8 38.3 37.9 21.3 110.8 +| | (4.8) | (1.0) | (2.4) | (1.9) | (0.1) | (5.6) | +| ------------- | -------- | ---------------- | ----- | ----- | ----- | ----- | +| StableBeluga2 | 33.0 | 14.0 | 27.9 | 29.0 | 20.6 | 75.6 | +| | (4.4) | (0.5) | (2.7) | (2.7) | (0.2) | (5.0) | +| | Table21: | ELI5fullresults. | | | | | + +Read the original question and passage, and generate 3 additional claims that are supported by the +| passage and | answer the question. | | | | +| ----------- | -------------------- | -------------- | ------------ | ---------------- | +| Original | question: What’s | the difference | between Shia | vs. Sunni Islam? | +Passage: The main difference between Shia and Sunni Muslim is related to ideological heritage and +issues of leadership. This difference is first formed after the death of the Prophet Muhammad in 632 +A.D. The ideological practice of the Sunni branch strictly follows Prophet Muhammad and his +teachings, while the Shia branch follows Prophet Muhammad’s son-in-law Ali. Nowadays, Sunni and +| Shia are | the major branches | of Islam. | | | +| -------- | ------------------ | ----------- | ------------- | ----- | +| Claim 1: | The major branches | of Islam | are Sunni and | Shia. | +| Claim 2: | Prophet Muhammad | died in 632 | A.D. | | +Claim 3: The ideological practice of the Sunni branch strictly follows Prophet Muhammad and his +teachings. +| Original | question: What | causes Bi-polar | disorder? | | +| -------- | -------------- | --------------- | --------- | --- | +Passage: Bipolar disorder is an emotional disorder that causes extreme mood swings between +excitement and depression. The spectrum of mood swing may span from days to months. We are still not +certain of the exact factors that cause such disorder, but genetics is considered a major factor. +Claim 1: One symptom of Bi-polar disorder is extreme mood swings between excitement and depression. +Claim 2: Genetics could be one of the major factors that causes Bi-polar disorder. +| Claim 3: | The mood swing | from Bi-polar | disorder can | last days to months. | +| -------- | -------------- | ------------- | ------------ | -------------------- | +Original question: How do we hear differences in sound besides volume and pitch? +Passage: Pitch refers to the frequency of soundwave, and volumn refers to the amplitude of the +soundwave. Besides volumn and pitch, we can also tell the difference between sounds based on the +tone of sound. For example, we can differentiate the sound of different instruments based on the +| tone of the | sounds. | | | | +| ----------- | ---------------------- | ---------------- | ----------------- | --- | +| Claim 1: | Volume of sound | is the amplitude | of the soundwave. | | +| Claim 2: | Pitch is the frequency | of soundwave. | | | +Claim 3: We can use the tone of the sounds to differentiate the sound of different instruments. +Original question: How are we able to discern whether a sound is coming from in front of us or +behind us? +Passage: There are multiple explanations for why we can localize sounds. One explanation is that +sounds travelling to the corresponding side of one’s ear will be slightly louder. Another +explanation is that there is a slight difference in the hitting time to one’s left and right ear +based on the sound’s direction. However, these explanation means that when a sound is exactly in +front of someone or exactly behind someone, he or she can not tell the difference. +Claim 1: We can localize sounds by recognizing that the sound travelling to the corresponding side +| of one’s | ear will be slightly | louder. | | | +| -------- | -------------------- | ------- | --- | --- | +Claim 2: We can also localize sounds by recognizing the difference in hitting time to one’s left and +| right ear | based on the sound’s | direction. | | | +| --------- | -------------------- | ---------- | --- | --- | +Claim 3: We cannot tell the difference between a sound that is exactly in front of us or exactly +behind us. +Table22: Promptusedtogeneratethesub-claimsforELI5questions. Bluetextismodelgeneration. Browntext +istheELI5examplethatwewanttogeneratesub-claimsfor. Weconstructthepromptbymanuallywritingthe +sub-claimsforthreequestionsfromthetrainingset. +Instruction: Write an accurate, engaging, and concise answer for the given question using only +the provided search results (some of which might be irrelevant) and cite them properly. Use an +unbiased and journalistic tone. Always cite for any factual claim. When citing several search +results, use [1][2][3]. Cite at least one document and at most three documents in each sentence. +If multiple documents support the sentence, only cite a minimum sufficient subset of the +documents. +| | | Table23: | InstructionforVANILLA. | | +| --- | --- | -------- | ---------------------- | --- | +Instruction: Write a high-quality answer for the given question using only the provided search +| results | and cite them properly | using | [1][2][3]. | | +| ------- | ---------------------- | -------- | --------------------------- | --- | +| | | Table24: | ShortinstructionforVANILLA. | | + +Summarize the following document within 50 words with the question of interest "{QUESTION}" +Return "irrelevant" if the document is irrelevant to the question. Try to keep all the important +| dates, numbers, | | and names. | | | | +| --------------- | --- | ---------- | --- | --- | --- | +Title: {TITLE} +Text: {TEXT} +Summary: +| | | | | Table25: | PromptsforSUMM. | +| --- | --- | --- | --- | -------- | --------------- | +Given the follow passage and the question "{QUESTION}", extract a useful span from the passage +that can answer the question. Resolve all the coreference issues to make the extracted span +understandable standalone. If the passage is not helpful for answering the question, return +"irrelevant". +Title: {TITLE} +Text: {TEXT} +| Extracted | span: | | | | | +| --------- | ----- | --- | --- | -------- | ------------------ | +| | | | | Table26: | PromptsforSNIPPET. | +Instruction: Write an accurate, engaging, and concise answer for the given question using only +the provided search results and cite them properly. Use an unbiased and journalistic tone. +| Always cite | for | any factual | claim. | | | +| ----------- | --- | ----------- | ------ | --- | --- | +You are provided summaries/snippets of the search results. You can use "Check: Document [1][2]" +to check the corresponding full documents (you should only check relevant documents and you can +at most check 3 documents at a time) and use "Output:" to output a sentence in the answer. In the +answer, cite properly by using [1][2][3]. Cite at least one document and at most three documents +in each sentence. If multiple documents support the sentence, only cite a minimum sufficient +| subset of | the documents. | | Use "End" | to | end the generation. | +| --------- | -------------- | --- | --------- | --- | ------------------- | +| | | | + +| Document | [1](Title: | ...) | {SUMMARY | OR SNIPPET} | | +| -------- | ---------- | ---- | -------- | ----------- | --- | +... +| Question: | When did | US | break away | from | England? | +| --------------- | ---------- | ------ | ----------- | ------------- | -------- | +| Check: Document | | [1][2] | | | | +| Document | [1] {FULL | TEXT} | | | | +| Document | [2] {FULL | TEXT} | | | | +| Output: | The United | States | ... | [1] ... | [2]. | +| | | +| Check: Document | | [3] | | | | +| Document | [3] {FULL | TEXT} | | | | +| Output: | The Treaty | of | Paris ... | [3]. | | +| | | +End. +| | | | | Table27: | AnexampleforINTERACT. | +| --- | --- | --- | --- | -------- | --------------------- | +Instruction: Write an accurate, engaging, and concise answer for the given question using only +the provided search results and cite them properly. Use an unbiased and journalistic tone. +| Always cite | for | any factual | claim. | | | +| ----------- | --- | ----------- | ------ | --- | --- | +You can use "Search: key words" to check the most relevant document’s full text and use +"Output:" to output a sentence in the answer. In the answer, cite properly by using [1][2][3]. +Cite at least one document and at most three documents in each sentence. If multiple documents +support the sentence, only cite a minimum sufficient subset of the documents. Use "End" to end +the generation. +| | | | Table28: | | InstructionforINLINESEARCH. | +| --- | --- | --- | -------- | --- | --------------------------- | +Instruction: Write an accurate, engaging, and concise answer for the given question. Use an +| unbiased | and journalistic | | tone. | | | +| -------- | ---------------- | --- | ----- | -------- | ------------------------- | +| | | | | Table29: | InstructionforCLOSEDBOOK. | + +| Instruction: | Write an accurate, | engaging, | and | concise | answer for | ... | +| ------------ | ------------------ | --------- | --- | ------- | ---------- | --- | +Document [1](Title: How to Treat and Prevent Food Poisoning - MsPrepper) just a typical gastro +upset. Salmonella is most commonly caused by eating undercooked or raw foods like eggs or meat. +You know how your mom always warned you not to eat raw cookie dough? This is why. Most people do +eat cookie dough and they are fine, but salmonella is a risk. If you do contract salmonella, you +could start to feel bad within in a couple of hours after eating contaminated food, and sometimes +it could take a day or two. Common symptoms are nausea and vomiting, loose stools (sometimes +| bloody), | flu like symptoms, | and stomach | cramps. | To treat | | | +| -------- | ------------------ | ----------- | ------- | -------- | --- | --- | +Document [2](Title: FDA Issues Warning About Eating Raw Cookie Dough, But Not For Salmonella +Risks) FDA Issues Warning About Eating Raw Cookie Dough, But Not For Salmonella Risks Used to +licking the spoon or placating yourself with full-on chunks of raw cookie dough? The Food and +Drug Administration issued a warning on Tuesday that strongly advises against continuing the +habit. The agency asserted that consuming raw batter of any kind, whether for bread, cookies or +pizza, could make a person sick. While you may have been warned in the past against eating raw +dough due to the risk of contracting salmonella from raw eggs, the FDA is citing raw flour as the +| culprit for | a | | | | | | +| ----------- | --- | --- | --- | --- | --- | --- | +Document [3](Title: It’s Probably OK to Eat Raw Cookie Dough — As Long As You’re Smart About It - +The Crux - Very Top Secret Information) First, when most people think about health risks and +cookie dough, they think about raw egg. Eggs can be contaminated with salmonella bacteria, and +food safety recommendations encourage people to cook eggs until the white and yolk are firm in +order to kill any bacteria. However, anyone making cookies can do things to reduce this risk by +using pasteurized egg products. When my kids and I make cookie dough, we never use regular eggs. +Instead, we use shell eggs that have been pasteurized to kill any harmful bacteria without +| actually | cooking the egg | itself. (A | great public | health | innovation, | if | +| -------- | --------------- | ---------- | ------------ | ------ | ----------- | --- | +Document [4](Title: How Dangerous Is It to Eat Raw Cookie Dough? | Men’s Health) Can Eating Raw +Cookie Dough Really Make You Sick? Scientists reveal the truth about this supposedly dangerous +delicacy By Katherine Dempsey There are few things more tempting in life than eyeing a bowl of +cookie dough and deciding whether or not to stick your finger in for a scoop. It’s a bit like +playing Russian roulette. You could get lucky and enjoy the delicious dough without conseqence, +but there’s always the risk of getting serously sick with a food-borne illness. That’s because +multiple ingredients within the dough could be contaminated by pathogens such as Salmonella and +| E. coli, | says Soohyoun | | | | | | +| -------- | ------------- | --- | --- | --- | --- | --- | +Document [5](Title: How Dangerous Is It to Eat Raw Cookie Dough? | Men’s Health) is usually +pasteurized, so it’s not likely that the egg would make you sick. However, other ingredients in +the dough could potentially harbor pathogens. Experts say that a prime suspect in a 2009 E. coli +outbreak linked to prepackaged cookie dough was actually the flour. Nuts and chocolate have also +been linked to Salmonella outbreaks. Bottom line: You’re better off skipping raw cookie dough. +At least you’ll have killer cookies as a consolation. The article How Bad Is It To Eat Raw Cookie +Dough? originally ran on Prevention.com The 9 Foods Most Likely to Make You Sick Mmm, Just Don’t +... +Question: Why is it bad to eat cookie dough for risk of salmonella but things like Cookie Dough +| Bites are | ok? | | | | | | +| --------- | --- | --- | --- | --- | --- | --- | +Answer: Raw cookie dough is not recommended to be eaten due to the risk of salmonella [1][2]. +Eating raw flour is also a risk for food poisoning [2]. The egg in cookie dough is usually +pasteurized, so it’s not likely that the egg would make you sick, but other ingredients in the +dough could potentially harbor pathogens [4][5]. However, prepackaged cookie dough like Cookie +Dough Bites is safe to eat because the dough is made with pasteurized egg products and +| heat-treated | flour [2][3].. | | | | | | +| ------------- | ---------------- | ----------- | ----------- | ---------- | ---------- | ----- | +| Ground truth | claims: | | | | | | +| 1. Cookie | Dough Bites are | safe to eat | since they | do not | contain | eggs. | +| 2. Salmonella | is a bacteria | found in | eggs. | | | | +| 3. Eating | raw cookie dough | carries | the risk of | salmonella | infection. | | +Table30: ELI5example1. + +Instruction: Write an accurate, engaging, and concise answer for ... +Document [1](Title: Is Snapchat really worth $19 billion? - CSMonitor.com) reporting that the Los +Angeles-based company is aiming to raise $500 million at a valuation of $16 billion to $19 billion, +making it the third most highly valued tech start-up backed by venture capitalists. The Chinese +handset maker Xiaomi is valued at $45 billion, while Uber is estimated to be valued at about $40 +billion, according to data from CB Insights. Read MoreVC investment hits $86B thanks to Uber, Xiaomi +Snapchat was valued at $10 billion in August, according to a Dow Jones report. Some of its investors +from previous rounds include Benchmark, Lightspeed Venture Partners and Kleiner Perkins Caufield +Document [2](Title: What Are Venture Capital Investments? – DollarsAndSense.my) Ever wondered how +highly valued technology giants like Google and Facebook were able to grow so fast and pay their +employees so well in such a short amount of time, or how still growing start-ups like Uber are able +to lose 1.2 billion US dollars in just the first half of this year alone and still command a +valuation upwards of 50 billion US dollars? The answer lies with a special category of investment +activity known as venture capital. Venture capitalists are professional investors who invest in a +number of highly scalable high-risk technology ventures hoping to make a multi-fold +Document [3](Title: Opinion | What Dara Khosrowshahi Must Do to Save Uber - The New York Times) at a +discount. These are troubling signs. Every start-up must one day fulfill the market’s demand that +it turn a profit, but Uber has never figured out how to do that. While ride sharing in some form +will probably survive, it’s more likely that without some drastic changes, Uber won’t be around in +three to five years. Mr. Khosrowshahi must avoid the mistakes of his predecessor by accepting that +“pivots” (Silicon Valley-speak for the desperate changes troubled companies make to reassure their +venture capitalist funders) are not the answer. None of the pivots Mr. Kalanick tried — like +on-demand delivery +Document [4](Title: Snapchat raising funding round at $19 billion valuation: Report) Snapchat +raising funding round at $19 billion valuation: Report The mobile messaging app Snapchat is said to +be seeking funding at a $19 billion valuation, according to a report, citing a person with knowledge +of the matter. Bloomberg is reporting that the Los Angeles-based company is aiming to raise $500 +million at a valuation of $16 billion to $19 billion, making it the third most highly valued tech +start-up backed by venture capitalists. The Chinese handset maker Xiaomi is valued at $45 billion, +while Uber is estimated to be valued at about $40 billion, according to data from CB Insights. +Document [5](Title: Unicorns And Wall Street | MoneyTips) not start showing profits soon. Even Uber, +with its massive $62-billion-plus valuation, will test investor’s patience if it does not start +showing signs of a profit in the next few years. Cracks are starting to show in terms of pre-IPO +devaluations. Dropbox and Snapchat have seen a nearly one-quarter drop in their pre-IPO valuations, +while Square’s pre-IPO value has dropped by approximately one-third. Larger investors and venture +capitalists are sending a message to the unicorns: Outline a solid path to profits and start +producing them soon, or expect your valuation to be slashed. Diving ahead with an IPO is not +... +Question: How are firms like snapchat, uber etc valued so highly while still not making a profit? Do +venture capitalists not expect some form of repayment within a number of years? +Answer: Venture capitalists invest in a number of highly scalable high-risk technology ventures +hoping to make a multi-fold return on their investment in a short amount of time [2]. Highly valued +companies like Snapchat, Uber, and Xiaomi are able to grow so fast and command high valuations +because they have the potential to generate significant revenue in the future [2][4]. Venture +capitalists expect a return on their investment within a number of years, but they also invest in the +long-term potential of the company [2]. It is important to note that every start-up must eventually +turn a profit, but some start-ups like Uber have not yet figured out how to do that [3][5]. +Ground truth claims: +1. Firms like Snapchat and Uber need to establish their brand and amass users before introducing +ads. +2. Introducing ads too early can deter potential users. +3. Uber is reinvesting a lot of money to make their service better. +Table31: ELI5example2. \ No newline at end of file diff --git a/q195/random_k2/random_2.md b/q195/random_k2/random_2.md index f94127163277f669a14f3b0a86d9bfd3ebbd9e38..3c145277c1fd495efe98de73b305517bed6fbacd 100644 --- a/q195/random_k2/random_2.md +++ b/q195/random_k2/random_2.md @@ -1,6514 +1,739 @@ -C H A P T E R 4 - -Climate Change Impacts -on Human Health - -Song Liang1,5, Kristina Kintziger2, Phyllis Reaves3, and Sadie J. Ryan4,5 - -1Department of Environmental and Global Health, College of Public Health and Health Professions, -University of Florida, Gainesville, FL; 2Department of Public Health, College of Education, Health & -Human Sciences, The University of Tennessee, Knoxville, TN; 3Division of Physical Therapy, School of -Allied Health Sciences, Florida Agricultural & Mechanical University, Tallahassee, FL; 4Department of -Geography, University of Florida, Gainesville, FL; 5Emerging Pathogens Institute, University of Florida, -Gainesville, FL - -Climate change poses major challenges to human society and to Earth systems, influencing the -functioning of many ecosystems and thereby affecting human health. Many climate change/variability- -and extreme weather-associated events, such as sea level rise, hurricanes, and storm surge, as well as -other weather extremes, including excessive precipitation and heatwaves, have direct and/or indirect -impacts on human health. These impacts include death/injury, cardiovascular and respiratory diseases, -environmentally-mediated infectious diseases, and mental health, among others. Due to its unique -geography, Florida is particularly vulnerable to these environmental impacts, which have important -health implications for the state’s more than 20 million residents. In this chapter, we review the health -impacts of climate change and associated weather events, with an emphasis on those relevant to Florida, -and environmental hazards, including hurricanes and storms, lightning, sea level rise, excessive -precipitation, extreme heat, and drought. There is clear evidence for significant climate-sensitive hazards -and human health impacts in the state, despite uncertainties associated with the assessment of some -effects. To address health impacts and challenges, policies focused on mitigation and adaptation -strategies, health surveillance, and research that could close knowledge gaps on human exposures to the -climate-sensitive hazards and health impacts are needed. - -Key Messages - -• Florida is highly vulnerable to climate-sensitive hazards (e.g. sea level rise, heat waves, storm - -surge, and hurricanes), which have a wide range of human health effects. - -• The health effects can be direct, such as storm/temperature related illnesses, injuries, and -deaths; or indirect, such as waterborne, food-borne, and vector-borne diseases; or take social -and economic pathways, such as stress and mental illness. - -• The health effects exhibit substantial regional disparities across the state. -• Policies focused on health surveillance and research on knowledge gaps between human - -exposure to the hazards and health effects are much needed. - -Keywords - -Climate change; Environmental hazards; Human health; Florida - -125 - - 1 2 6 • S O N G L I A N G E T A L . - -Introduction - -C - -limate change is well-characterized at the global level using metrics such as the global -mean temperature and sea level. But the manifestations of global climate change and -variability occur at varying geographical scales, causing a variety of weather events, -such as excessive rainfall, drought, severe storm/flooding, sea level rise, and heat waves (Field -et al. 2014), all of which are likely to have direct and/or indirect human health impacts -(McMichael and Haines 1997; Colwell et al. 1998; Frumkin et al. 2008). Indeed, there is clear -and increasing evidence that many health outcomes (most adverse) exhibit high sensitivities to -these varying weather events. The interface between the climate change-associated weather -events and human health exhibits a complex web of relationships involving both natural and -social environments through direct or indirect impact pathways, social institutional disruption, -or a combinations of these things (Fig. 4.1). Globally, substantial disease burdens are attributable -to hazards that are associated with climate change and related weather events. According to the -World Health Organization’s 2008 estimate, climate change-associated hazards were responsible -for more than 150,000 deaths and 5,517,000 disability-adjusted life years, which were highly -likely to be underestimated as the estimates were only based on selected risk factors and illnesses -associated with climate change and related weather events (WHO 2008). The actual disease -burden might be greater and the adverse health impacts are likely to increase in the years to come, -as the Intergovernmental Panel on Climate Change (IPCC) and US Global Change Research -Program have clearly indicated that the accelerated changing climate poses a substantial threat -to global human health and the risk will continue to become severe if no remediation action is -taken (Field et al. 2014; Balbus et al. 2016). - -Due to its unique geography, Florida is highly vulnerable to a variety of hazards associated -with climate change and variability, and related weather events. Sea level rise, hurricane and -storm surge, excessive precipitation, and heatwaves all pose threats to Florida’s agriculture, -ecosystems, tourism, and public health. For example, Florida has experienced a greater number -of hurricane landfalls than other states in the country (Knight and Davis 2009), and more intense -hurricanes are expected in the future (Knutson et al. 2013). In addition, rising sea levels are -influencing and expected to continue to have significant impacts on communities and residents -along Florida’s coastal areas, particularly in the southern part of the state, making Floridians -exceptionally vulnerable to these environmental hazards (Elsner et al. 2008). - -This chapter is organized into sections addressing impact pathways and specific -environmental hazards of particular concern in Florida (Fig. 4.1). For direct impacts, we focus -on extreme heat, flooding, storms and lightning, and sea level rise. For indirect impacts on public -health, we focus on water, food, and vector-borne diseases. And lastly, we conclude with our -discussion of impact pathways through social disruption on mental and community health impact -associated with climate change and related weather events. Throughout this chapter, health - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 7 - -impacts and implications, particularly those pertinent to Florida, are reviewed primarily based -on the Building Resilience Against Climate Effects (BRACE) project reports from the Florida -Department of Health (FDH) (FDH 2015b, a) and other relevant publications. - -Figure 4.1. Impact pathways by which climate change and variability and related weather events affect -human health. - -Direct Impact Pathways – Environmental Hazards -and Human Health - -A number of human health impacts come from direct exposure to environmental hazards -associated with climate change and related weather events. Among those of particular concern -in Florida are extreme heat, flooding, storms and lightning, and sea level rise. - - 1 2 8 • S O N G L I A N G E T A L . - -Extreme Heat and Health Impacts - -Extreme climate and weather events (e.g., heat, cold, storms, and floods) are expected to occur -more frequently worldwide due to climate change. These extreme events, such as heatwaves and -cold spells (e.g. frequent very hot days and possibly fewer very cold days), have a direct impact -on human health by compromising the body’s ability to regulate its internal temperature. -Increased human morbidity and mortality have been associated with extreme temperatures, both -hot and cold, and documented in both developing and developed countries worldwide (Center -for Disease Control and Prevention 1994; Keatinge et al. 2000a; Keatinge et al. 2000b; Basu and -Samet 2002; Medina-Ramon et al. 2006; Kovats and Hajat 2008; Bandyopadhyay et al. 2012). -The mechanistic effects of extreme temperature on death and illness are briefly summarized -below (Sarofim et al. 2016): -• Extreme heat may induce heat cramps, heat exhaustion, heatstroke, and hyperthermia. -• Extreme cold may induce hypothermia and frostbite. -• Extreme heat/cold may exacerbate chronic conditions related to cardiovascular disease, - -respiratory disease, cerebrovascular disease, and diabetes-related illnesses. - -• Extended exposure to high temperatures may induce increased hospital admissions for - -cardiovascular, kidney, and respiratory disorders. - -• Extreme heat may induce or exacerbate mental health and behavioral disorders. - -A number of studies have reported associations between high temperatures and human deaths -or illnesses around the world. For example, a large number of heat-related deaths have been -reported in the United States (Center for Disease Control and Prevention 1994; Curriero et al. -2002; Hoshiko et al. 2010), Europe (Keatinge et al. 2000a; Keatinge et al. 2000b; Huynen et al. -2001), and Asia (Qian et al. 2008; Chung et al. 2009). A recent study by Berko et al. examined -deaths attributed to extreme weather events from 2006 to 2010 in the United States and found -that each year about 2,000 U.S. residents died from weather-related causes, and among those -31% died due to exposure to excessive heat, heat stroke, and/or sun stroke (Berko et al. 2014). -Mounting evidence has suggested that substantial increases in mortality from respiratory and -cardiovascular diseases, in particular among the elderly and youth, are associated with high -temperatures in the United States (Basu and Samet 2002; Medina-Ramon et al. 2006; Kovats and -Hajat 2008). During the 2006 California heat wave, 16,166 excess emergency department visits -and 1,182 excess hospitalizations were reported statewide. This included a significant increase -in emergency department visits for acute renal failure, cardiovascular diseases, and diabetes -(Knowlton et al. 2009). The association between heat and respiratory diseases has also been -reported in many studies. In New York state, a significant number of respiratory hospitalizations -were attributable to excessive heat, and a projection suggested that excess respiratory admissions -in the state due to extreme heat will be two to six times higher in 2080–2099 than what was seen -in 1991–2004 (Lin et al. 2012). In another study conducted in Greater London, UK, an analysis - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 2 9 - -of historical data suggested that heat-related increases in emergency admissions for respiratory -and renal diseases had been observed in children under five, and for respiratory disease in the -75+ age group (Kovats et al. 2004). - -Extreme Heat and Public Health in Florida - -Florida has a warm subtropical and tropical climate. Due to continuous summer heat and extreme -weather events, temperatures around the state can reach levels potentially harmful to human -health. Through analyses of historical data (for the periods 1895–2009 and 1970–2009), a -significant trend in Florida’s mean temperature has been found (Martinez et al. 2012). The -average summer temperature (June–August) has been showing an overall increasing trend, -particularly after the 1980s, which shows a greater upward trend (FDH 2015a), likely due to a -variety of factors ranging from rapid urban development (Martinez et al. 2012; FDH 2015a) to -climate change (Ji et al. 2014). Within the state, substantial spatial heterogeneities in mean, -minimum, and maximum temperature have been reported (Martinez et al. 2012). For example, -the distribution of annual mean number of days with a maximum temperature greater than or -equal to 95 °F and greater than or equal to 75 °F showed that more days with hottest daytime -temperatures occurred in the northern and interior portions of the state (FDH 2015a). - -Using statewide information on specific health outcome-related emergency department visits -that occurred May to October from 2005–2012, the BRACE project examined the effects of daily -maximum temperature or daily maximum heat index on emergency department visits through a -two-stage analysis—the first stage involved the analysis on regional (National Weather Service -regions) scales using the Poisson regression model; the second stage used a meta-analysis -technique to integrate regional estimates into a statewide estimate (FDH 2015a). Rate ratios, -based on the comparison between the exposure (e.g., hotter days) and the reference (e.g. 88 °F -for temperature and 94 °F for heat index), were used to compare emergency department visits -for specific health conditions (FDH 2015a). -a) Heat and heat-related illness. A strong association between heat-related illness and heat -exposure was observed; the significant increase in emergency department visit rates for heat- -related illness was associated with increasing temperatures and a strong dose-response -relationship across all regions in Florida was found (FDH 2015a). - -b) Heat and cardiovascular disease. Two types of cardiovascular disease were considered— -myocardial infarction and ischemic stroke. No statistically significant association was found -between temperature and ischemic stroke at the state level; however, significant regional -variations were identified. For example, the Tampa region showed a significant positive -association between heat index (above the reference level at 94 °F) and stroke (FDH 2015a). -Furthermore, a positive, statistically significant relationship was observed between -myocardial infarction and heat index despite remarkable geographical variations (FDH -2015a). - - 1 3 0 • S O N G L I A N G E T A L . - -c) Heat and respiratory disease. Overall, increases in the number of emergency department -visits for asthma were associated with higher temperatures. Although no significant -association was found at the state level between temperature and emergency department visits -for asthma, the regional analysis suggested that all regions except Tallahassee showed a -statistically significant positive association between temperature and asthma (FDH 2015a). - -Flooding, Storms, Lightning, and Sea Level Rise - -Excessive precipitation, hurricanes, coastal storms, sea level rise, and thunderstorm-related -lightning, which are all typically accompanied by coastal and inland flooding, have the potential -for substantial direct human health impacts (e.g. injury and death, maternal and child health, and -mental health issues). Natural disasters related to flooding and storms are a significant cause of -mortality and morbidity. For example, according to the International Federation of Red Cross -and Red Crescent Societies (IFRCRCS), more than 3,448 climatic, hydrological and -meteorological disasters (including 1,751 major floods and 988 storms) resulting in 339,710 -deaths (59,092 and 177,685 deaths specifically tied to flooding and storms) were reported -worldwide between 2005 and 2014 (IFRCRCS 2015). Lightning strike is the second leading -cause of weather-related mortality at the global scale, with an estimated 0.2-1.7 deaths per million -people, and those who survive lightning strikes often suffer from significant injuries (Aslar et al. -2001; Ritenour et al. 2008). The magnitude and impacts of these extreme weather events, their -severity, and the extent of the effects are influenced by many different factors. For example, -short- and long-term averages and variability of weather conditions and physical impacts of -associated extreme events, as well as social environmental factors (e.g. infrastructure, social and -individual vulnerability), are considered important (Bouma et al. 1997, Kovats 2000, Kovats et -al. 2003, Miranda 2004). For projections of future flooding under different climate change -scenarios, climate models have consistently suggested that episodes of severe flooding may -become more frequent in inland river systems including flood plains (Christensen and -Christensen 2003, Booij 2005), urban and coastal environments in various parts of the world -(Schreider et al. 2000; Douglas et al. 2008; Kirshen et al. 2008; Thompson et al. 2009, Diez et -al. 2011; Lyle and Mills 2016). Meanwhile, climate models also predict positive correlations -between lightning and global temperatures, suggesting the likelihood of a greater number of and -more severe lightning episodes (Price and Rind 1994; Reeve and Toumi 1999; Kochtubajda et -al. 2006). These episodes are expected to increase the risk of extreme environmental hazards and -adverse public health impacts. -a) Natural disasters in Florida. Due to its geography (extensive coastline and a peninsular -shape) and its tropical/subtropical climate, Florida is particularly vulnerable to tropical -storms, hurricanes, and lightning strikes. Historically the occurrence of hurricanes in the US -has been clustered in Florida and along the Atlantic coastline (Ellis et al. 2015). A total of 67 -known Florida hurricanes have occurred over the 108-year period (1900-2007) with four - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 1 - -different hurricanes occurring in one year (2004) in Florida. The climate change model -suggests a 46% chance that Florida will be hit by at least one hurricane each year in the future -(Malmstadt et al. 2009), and recent experience seems to support this. Fig. 4.2 illustrates the -distribution of average return time of hurricane landfalls and populations along the coastal -areas of Florida, showing spatial heterogeneities in vulnerability to hurricanes. - -b) Public health impacts. Key direct public health impacts associated with extreme weather -events include injury and death due to trauma, drowning, destructive forces of wind, -collapsed building and trees, and lightning strike and carbon monoxide poisoning related to -power outages. Historically, Florida experienced significantly higher mortality in the early -20th century; approximately 3,000 deaths were attributed to extreme weather-related events -(Winsberg 2003) and the largest number of deaths due to lightning strike in the US were -reported during this time (Duclos et al. 1990; Ritenour et al. 2008). During 2004 and 2005, -eight hurricanes hit Florida resulting in 213 deaths, over half of which were caused by trauma, -followed by drowning, other injury, electrocution, and carbon monoxide poisoning (Ragan et -al. 2008). Injury is common cause of hurricane-related morbidity and mortality. For example, -in the aftermath of Hurricane Katrina between September 8 and October 14, 2005, 7,543 non- -fatal injuries among residents and relief workers were recorded in the surveillance system -(Sullivent et al. 2006). During a 2006 flood event in El Paso County in Texas, 43% of -individuals (out of 475 surveyed) were reported having physical health issues related to the -flooding episode (Collins et al. 2013). - -c) Injury. The BRACE study compared injury-related emergency department visits and -hospitalizations during impact periods (periods covering tropical cyclone landfall) vs. control -periods (periods with no such impact) between 2004 and 2012. The study found that -emergency department visit and hospitalization rates were significantly higher during the -impact periods, with a rate ratio of 1.03 (95% confidence interval (CI): 1.02, 1.05) for the -emergency department visit rate and 1.4 (95% CI: 1.02, 1.05) for the hospitalization rate. The -most common types of injuries during the impact periods included falls, being struck by an -object, being cut or pierced, and motor vehicle transport accidents (FDH 2015b). - -d) Carbon monoxide poisoning. Carbon monoxide (CO) is a colorless, odorless, poisonous gas -that can be harmful when inhaled in a large amount. Inhaled CO enters the blood stream and -reduces the delivery of oxygen to the body’s critical organs, such as the heart and brain. The -gas is primarily generated through incomplete combustion. The greatest sources of CO -outdoors are usually related to motor vehicles, machinery that burn fossil fuels. Indoors, some -major sources include unvented kerosene and gas space heaters, leaking chimneys and -furnaces, and gas stoves. Exposure to CO may cause weakness, headache, dizziness, nausea, -shortness of breath, confusion, and even death at very high concentration levels, which is -more likely in indoor environments. A recent systematic review of the health impacts of -power outages due to extreme events indicated that CO poisoning is an important health -concern (Klinger et al. 2014). In Florida, the BRACE study reported that during the study - - 1 3 2 • S O N G L I A N G E T A L . - -period 2004–2012, the rate of CO exposure calls during impact periods was 6.59 times (95% -CI: 4.48, 9.7) the rate of CO exposure calls to poison control centers during the control -periods. The study also found the rate of CO-poisoning related emergency department visits -was significantly higher for the impact periods than during the control periods with the rate -ratio 3.44 (95% CI: 2.07, 5.72). Finally, the BRACE study revealed that rates of CO -poisoning-related hospitalizations were also significantly higher during the impact periods -with the rate ratio 4.0 (95% CI: 2.9, 5.51) (FDH 2015b). - -Figure 4.2. Historical distribution of hurricane landfalls in Florida. - -Indirect Impact Pathways–Infectious Diseases - -Climate change and variability and associated variable weather conditions can affect human -health through indirect pathways (Fig. 4.1). These indirect impacts are modulated through -biological and ecological processes that can influence infectious diseases (in particular, some -water/food- and vector-borne diseases) and air quality. In this section, we focus on water/food-, -and vector-borne diseases. - -Disease Transmission Groupings - -The impact of climate change and variability and associated weather events on infectious -diseases has been receiving much attention in the past several decades. Across the globe, high - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 3 - -human mortality rates and disability have been attributed to infectious diseases; climate change -is intensifying and will continue to exacerbate transmission of infectious diseases, poses -increasing concerns (Patz et al. 2005). Indeed, the transmission of many infectious diseases is -closely linked to physical environmental factors, such as water and temperature. Many water- -associated and zoonotic pathogens, for instance, which account for more than 75% of known -pathogens causing human diseases, are associated with hydrological process (WHO 2003). From -a transmission perspective, infectious diseases can be broadly classified into two groups: 1) -directly transmitted (i.e., those spread directly from person to person via direct contact or droplet -exposure, such as influenza and tuberculosis); and 2) indirectly transmitted, including those -mediated through an intermediate vector organisms (vector-borne) or through environmental -media such as water and soil (environmentally-mediated) (Eisenberg et al. 2007). Because the -diseases in the indirectly transmitted group are either directly or indirectly related to water, this -group can be further broken down into the following sub-groups based on water’s role in disease -transmission: waterborne (e.g., cholera typically via fecal-oral transmission), water-based (e.g. -part of the pathogen life cycle requires aquatic environment, such as schistosomiasis), and water- -related (e.g., malaria and dengue, which need water for breeding of insect vectors to fulfill the -transmission cycle) (Yang et al. 2012; Mordecai et al. 2013; Ryan et al. 2015; Johnson et al. -2015; Mordecai et al., 2017) . - -Climate-Sensitivity of Microbial Agents and Their Associated Diseases - -Table 4.1 lists some of the major pathogens and their associated diseases that are climate sensitive -and associated with water. These disease-causing pathogens, either through infection or toxin- -generating (e.g. caused by protozoa, bacteria, virus, helminth, or algae), and their associated -vectors (e.g., snails, mosquitoes, sand flies) are very small in size and lack thermostatic -mechanisms (i.e. lack of function to maintain body temperatures). Their biological (e.g., survival, -reproduction, development/growth, and infectivity) and environmental (e.g., contamination and -movement) processes are determined by environmental conditions, such as rainfall, storm runoff, -temperature, and sunlight. Climate change and variability and associated weather events are -expected to affect both fresh and marine water environments thereby changing humans’ exposure -to these water-related contaminants or pathogens that cause ill health. - -a) Waterborne Diseases -Waterborne diseases remain an important contributor to the global burden of disease (Pruss et al. -2002; Yang et al. 2012; Pruss-Ustun et al. 2014). Many different viral, bacterial, and parasitic -diseases have been associated with waterborne transmission (Table 4.1). Human exposure to -waterborne pathogens is primarily through drinking water, recreational water (e.g. via accidental -drinking and, for some pathogens, via dermal contact), or foods. The distribution of these -pathogens exhibits significant geographical variations, depending on both physical - - 1 3 4 • S O N G L I A N G E T A L . - -environmental and socio-economic conditions (Yang et al. 2012). For example, waterborne -cholera, viral hepatitis, and some other water-based and water-related diseases (e.g., -schistosomiasis and onchocerciasis) are largely confined to certain tropical areas where the -prevalence of such diseases is largely due to lack of access to improved drinking water, sanitation -and hygiene (Pruss et al. 2002), while some pathogens, such as Cryptosporidium parvum, -Giardia duodenalis, and Campylobacter spp. have a much wider geographical distribution (Yang -et al. 2012). Regardless of the geographical region, heavy rainfall, flooding, and temperature are -among the most important factors associated with transmission and outbreaks of these -waterborne diseases. - -There is mounting evidence that weather events are often an important factor triggering -waterborne outbreaks. In developed countries, extreme weather events such as excessive rainfall -and flooding can overwhelm water treatment plants and/or increase runoff into recreational -water, leading to water contamination that causes outbreaks (Kistemann et al. 2002). Some recent -studies have clearly indicated that excessive rainfall has been a significant contributor to -waterborne outbreaks. An analysis of historical records of 548 waterborne outbreaks in the US -reported between 1948 and 1994 suggested a significant association between the outbreaks and -rainfall—51% of waterborne disease outbreaks were preceded by precipitation events above the -90th percentile and 68% by events above the 80th percentile; the strongest association was -between the outbreaks and surface water contamination due to extreme precipitation (Curriero et -al. 2001). In Canada, Thomas et al. examined extreme rainfall and spring snowmelt in relation to -92 Canadian waterborne disease outbreaks that occurred between 1975 and 2001. They found -that warmer temperatures and extreme rainfall were significant contributing factors to waterborne -disease outbreaks in Canada (Thomas et al. 2006). Similar studies have been conducted in -England (Nichols et al. 2009), the Netherlands (Schijven and de Roda Husman 2005), Finland -(Miettinen et al. 2001), Denmark (Laursen et al. 1994), and Taiwan (Chen et al. 2012). In many -developing countries, lack of access to improved drinking water and sanitation exacerbates the -impact of these extreme precipitation weather events on waterborne diseases. Diarrhea, caused -by many waterborne pathogens, remains a top killer of children under five in the developing -world, particularly in Africa (Pruss-Ustun et al. 2014). A large number of studies have been -carried out in the developing world. For example, cholera outbreaks caused by toxigenic Vibrio -cholerae have been consistently shown to be correlated with excessive rainfall, flooding, and -high temperatures in the epidemic areas of West Africa and Bangladesh (Mhalu et al. 1987; -Hashizume et al. 2008; Luque Fernandez et al. 2009; Ngwa et al. 2016), although the exact -mechanistic relationships that result in this correlation remain elusive (Colwell et al. 1998). -Significant (positive) relationships have been reported between excessive rainfall and/or high -temperatures and diarrhea in the Pacific Islands (Singh et al. 2001), Ecuador (Carlton et al. 2014), -and Sub-Saharan Africa (Bandyopadhyay et al. 2012), while evidence also suggests that low -rainfall and even drought can also be associated with diarrhea, as reported in Bangladesh -(Hashizume et al. 2008) and Denmark (Senhorst and Zwolsman 2005). - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 5 - -b) Waterborne Diseases and Climate Impact in Florida -Across the US, climate change and variability and associated weather events are affecting and -are expected to continue to affect both marine and freshwater resources. These effects extend to -some water-associated pathogens and related diseases. In Florida, waterborne and foodborne -diseases of major public health concerns are highlighted in Table 1. Enteric bacteria, protozoan -parasites, enteric viruses including Salmonella enterica, Campylobacter spp., toxigenic -Escherichia coli, Vibrio bacteria species, Cryptosporidium and Giardia enteroviruses, -rotaviruses, noroviruses, and hepatitis A and E, are among those closely related to drinking and -recreational waters, and shellfish. These pathogens, while in environmental stages, are sensitive -to temperature, precipitation, and water flow. Several important toxin producers of water sources -are of particular public health importance in Florida (Table 4.1) including toxins from harmful -algal blooms, toxigenic marine species of Alexandrium (causing paralytic shellfish poisoning), -Karenia brevis (causing neurotoxic shellfish poisoning), and Gambierdiscus spp. (causing -ciguatera fish poisoning). Human exposure pathways are primarily through the consumption of -contaminated shellfish and fish, and in recreational waters. Cyanobacteria, consisting of multiple -species, can produce toxins (including microcystin), and the primary human exposure is from -drinking water and recreational water. For all of these harmful algal blooms species, temperature, -and precipitation are among the important factors affecting their reproduction, growth, and -distribution. - -Fig. 4.3 shows yearly distribution of total reported outbreaks and cases of food-borne and -waterborne diseases in Florida from 1989 to 2011 and Fig. 4.4 shows the monthly distribution of -reported outbreaks and cases for 2011, exhibiting marked annual variations and distinct seasonal -distribution. According to the BRACE report for the period 2004–2012, approximately 1,150 -cases of campylobacteriosis were reported annually, with the majority of cases reported between -May and September. From that same report, ~426 cases of cryptosporidiosis were reported -annually, with the majority of cases reported between June and October. Approximately 940 -cases of giardiasis were reported annually, with an annual incidence rate of 5.1 cases per 100,000 -population and the majority of cases reported between May and October. Salmonellosis -accounted for the greatest number of foodborne illnesses reported, with an average of 5,438 cases -reported each year and ~70% reported between June and November. Finally, ~101 cases of -vibriosis were reported annually, with 81% of the cases reported between April and October -(FDH 2015b). Focusing on the five diseases—campylobacteriosis. cryptosporidiosis, giardiasis, -salmonellosis, and vibriosis—the BRACE study identified a total of 1,231 follow-up days of -interest including 775 control days and 456 impact days—42 associated with hurricanes and 414 -with tropical storms (FDH 2015b); the result suggested that the occurrences of cryptosporidiosis -and salmonellosis were significantly associated with the tropical cyclones, with the risk ratio -being 1.26 (95% CI, 1.04, 1.52) and 1.35 (95% CI, 1.29, 1.42), respectively (FDH 2015b). - - g -n -i -t -i - -m -o -v - -, -a -e -h -r -r -a -i -d - -, -h -s -a -r - -, -e -h -c -a -d -a -e -h - -, -e -u -g -i -t -a -f - -, -r -e -v -e -F - -d -e -t -a -n -i -m -a -t -n -o -c - -, -g -u -b - -e -n -i -m -o -t -a -i -r -t - -y -b - -e -t -i - -B - -e -s -a -e -s -i -d -s -a -g -a -h -C - -1 3 6 • S O N G L I A N G E T A L . - -t -n -a -n -g -e -r -p -n -i - -e -g -a -m -a -d -l -a -t -e -f - -, -e -y -e - -, -n -i -a -r -b -n -o -e -g -a -m -a -d -, -r -e -v -e -F - -l -a -t -i -n -e -g -n -o -c - -, -) -F - -, - -D - -( - -r -e -t -a - -W - -s -i -s -o -m -s -a -l -p -o -x -o -T - -i -i -d -n -o -g -a -m -s -a -l -p -o -x -o -T - -, -e -u -g -i -t -a -f - -, -e -h -c -a -d -a -e -h -, -r -e -v -e -f - -, -s -s -e -n -k -c -i -s - -g -n -i -p -e -e -l -s - -, -e -r -c -n -a -h -C - -a -e -s -u -a -n - -, -e -u -g -i -t -a -f - -, -s -t -a -e -w -s - -, -e -h -c -a -d -a -e -h - -, -s -l -l -i -h -c - -, -r -e -v -e -F - -l -a -t -a -f - -e -b - -n -a -c - -n -a -m -o -w - -s -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -a -i -r -a -l -a - -M - -y -l -f - -e -s -t -e -s -t -y -b -e -t -i - -B - -s -i -s -a -i -m -o -s -o -n -a -p -y -r -t -n -a -c -i -r -f - -A - -P -P -S -a -m -o -s -o -n -a -p -y -r -T - -P -P -S -m -u -i -d -o -m -s -a -l -P - -r -o -f - -l -a -t -a -f - -d -n -a - -, -s -s -o -l - -t -h -g -i -e -w - -, -n -i -a -p -h -c -a -m -o -t -s - -, -a -e -h -r -r -a -i -D - -s -s -o -l - -t -h -g -i -e -w - -, -n -i -a -p -h -c -a -m -o -t -s - -, -a -e -h -r -r -a -i -D - -s -s -o -l - -t -h -g -i -e -w - -, -n -i -a -p -h -c -a -m -o -t -s - -, -a -e -h -r -r -a -i -D - -y -r -e -t -n -e -s -y -d -, -n -i -a -p -h -c -a -m -o -t -s - -, -a -e -h -r -r -a -i -D - -e -l -p -o -e -p -d -e -s -i -m -o -r -p -m -o -c -o -n -u -m -m - -i - -) - -D - -( - -r -e -t -a -w -a -i -v -l -a -r -o -- -l -a -c -e -F - -s -i -s -a -i -b -e -m -A - -a -c -i -t -y -l -o -t -s -i -h -a -b -e -o -m -a -t -n -E - -) - -, - -R -D - -( - -) - -, - -R -D - -( - -) - -D - -( - -r -e -t -a -w -a -i -v -l -a -r -o -- -l -a -c -e -F - -s -i -s -o -i -d -i -r -o -p -s -o -t -p -y -r -C - -m -u -v -r -a -p -m -u -i -d -i -r -o -p -s -o -t -p -y -r -C - -r -e -t -a -w -a -i -v -l -a -r -o -- -l -a -c -e -F - -r -e -t -a -w -a -i -v -l -a -r -o -- -l -a -c -e -F - -s -i -s -a -i -d -r -a -i -G - -s -i -l -a -n -e -d -o -u -d - -a -i -d -r -a -i -G - -s -i -s -a -i -r -o -p -s -o -l -c -y -C - -s -i -s -n -e -n -a -t -e -y -a -c -a -r -o -p -s -o -l -c -y -C - -e -n -r -o -b -r -e -t -a - -W - -a -o -z -o -t -o -r -P - -g -n -i -p -u -o -r -G - -s - -m -o -t -p -m -y -S -& - -s -s -e -n -l -l -I -n -i -a -M - -s -y -a -w -h -t -a -P -e -r -u -s -o -p -x -E -n -i -a -M - -e -s -a -e -s -i -D - -m - -s -i -n -a -g -r -O - -n -o -i -s -s -i - -m -s -n -a -r -T - -. -s -e -s -s -e -n -l -l -i - -d -e -t -a -l -e -r - -d -n -a - -s -t -n -e -g -a - -d -e -t -a -i -c -o -s -s -a -- -r -e -t -a -w - -, -e -v -i -t -i -s -n -e -s -- -e -t -a -m - -i -l -c - -f -o - -t -s -i -l - -A - -. -1 -. -4 - -e -l -b -a -T - -e -n -o -b -d -n -a - -e -l -c -s -u -m - -, -t -n -i -o -j - -e -r -e -v -e -s - -, -e -h -c -a -d -a -e -h -e -r -e -v -e -s - -, -r -e -v -e -F - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -c -i -g -a -h -r -r -o -m -e -h -e -u -g -n -e -D - -g -n -i -d -e -e -l -b - -, -h -s -a -r - -, -n -i -a -p - -) -F -H -D - -( - -r -e -v -e -f - -s -i -t -i -g -n -i -n -e -m - -r -o - -, -s -s -e -n -i -z -z -i -d - -, -e -h -c -a -d -a -e -h - -, -r -e -v -e -f - -, -c -i -t -a -m -o -t -p -m -y -s -a - -s -e -s -a -c - -t -s -o -M - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -s -i -t -i -l -a -h -p -e -c -n -e - -s -i -u -o -L - -. -t -S - -s -i -t -i -l -a -h -p -e -c -n -e - -s -i -u -o -L - -. -t -S - -h -s -a -r - -, -n -i -a -p -e -l -c -s -u -m - -, -g -n -i -l -l -e -w -s -d -n -a -n -i -a -p -t -n -i -o -j - -, -r -e -v -e -F - -d -n -a - -, -s -l -l -i -h -c - -, -r -e -v -e -f - -h -g -i -h - -, -e -h -c -a -d -a -e -h -, -s -n -a -m -u -h - -n -i - -e -r -a -R - -s -e -s -a -c - -e -r -e -v -e -s - -n -i - -g -n -i -t -i - -m -o -v - -e -s -i -a -l -a -m -d -n -a - -, -a -e -s -u -a -n - -o -t -i -u -q -s -o -m -y -b -e -t -i - -B - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -e -n -i -u -q -e -n -r -e -t -s -a -E - -a -y -n -u -g -n -u -k -i -h -C - -s -i -t -i -l -a -h -p -e -c -n -e - -s -u -r -i -v -a -y -n -u -g -n -u -k -i -h -C - -s -u -r -i -v - -s -i -t -i -l -a -h -p -e -c -n -e - -e -n -i -u -q -e -n -r -e -t -s -a -E - -s -u -r -i -v - -, -a -e -s -u -a -n - -, -e -h -c -a -d -a -e -h - -, -r -e -v -e -f - -, -c -i -t -a -m -o -t -p -m -y -s -a - -s -e -s -a -c - -t -s -o -M - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -s -i -t -i -l -a -h -p -e -c -n -e - -e -s -s -o -r -C -a -L - -s -i -t -i -l -a -h -p -e -c -n -e - -e -s -s -o -r -C -a -L - -s -s -e -n -d -e -r -i -t - -d -n -a - -, -g -n -i -t -i - -m -o -v - -s -u -r -i -v - -s -i -t -i -l -a -h -p -e -c -n -e - -, -s -s -e -n -l -l -i - -e -l -i -r -b -e -f - -, -c -i -t -a -m -o -t -p -m -y -s -a - -s -e -s -a -c - -t -s -o -M - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -e -s -a -e -s -i -d -s -u -r -i -v - -e -l -i - -N - -t -s -e - -W - -s -u -r -i -v - -e -l -i - -N - -t -s -e - -W - -) -e -n -r -o -b -- -r -o -t -c -e -V - -( - -, -r -e -d -r -o -s -i -d - -t -n -e -m -e -v -o -m - -, -g -n -i -t -i - -m -o -v -, -e -h -c -a -d -a -e -h - -, -r -e -v -e -F - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -s -i -t -i -l -a -h -p -e -c -n -e - -e -s -e -n -a -p -a -J - -s -i -t -i -l -a -h -p -e -c -n -e - -e -s -e -n -a -p -a -J - -d -e -t -a -l -e -r -- -r -e -t -a - -W - -s -e -r -u -z -i -e -s - -s -u -r -i -v - -) -s -u -o -e -n -a -t -u -c -( - -s -e -r -o -s -n -i -k -s - -, -) -l -a -r -e -c -s -i -v -( - -t -n -e -m -e -g -r -a -l -n -e - -r -e -v -i -l -/ -n -e -e -l -p -s - -, -s -s -o -l - -t -h -g -i -e -w - -, -r -e -v -e -F - -y -l -f -d -n -a -s -y -b -e -t -i - -B - -s -r -e -h -t -o -d -n -a - -, -) -F -( - -a -e -h -r -r -a -i -d -d -n -a - -g -n -i -t -i - -m -o -V - -s -i -t -i -t -a -p -e -H - -) -F - -, - -R - -, - -D - -( - -r -e -t -a -w -d -e -t -a -n -i -m -a -t -n -o -C - -) -F -R - -, - -, - -D - -( - -r -e -t -a -w -d -e -t -a -n -i -m -a -t -n -o -C - -s -u -o -i -r -a -V - -t -c -a -t -n -o -c - -, - -, -) -F -R -D - -, - -( - -r -e -t -a -w -d -e -t -a -n -i -m -a -t -n -o -C - -s -i -s -a -i -n -a -m -h -s -i -e -L - -P -P -S -a -i -n -a -m -h -s -i -e -L - -s -i -t -i -t -a -p -e -h - -l -a -r -i - -V - -s -u -r -i -v -E - -, - -A -s -i -t -i -t -a -p -e -H - -s -e -s -u -r -i -V - -s -i -t -i -r -e -t -n -e -o -r -t -s -a -g - -l -a -r -i - -V - -s -u -o -i -r -a -V - -4 - -, -3 -, -2 -, -1 -V -N -E -D - -s -e -s -u -r -i -v -o -r -e -t -n -E - -s -u -r -i -v -o -r -o -N - -e -n -r -o -b -r -e -t -a - -W - -i -z -u -r -c - -. - -T - -) -e -n -r -o -b -- -r -o -t -c -e -V - -( - -d -e -t -a -l -e -r -- -r -e -t -a - -W - - 7 -3 -1 - -• - -H -T -L -A -E -H - -N -A -M -U -H - -N -O - -S -T -C -A -P -M - -I - -E -G -N -A -H -C - -E -T -A -M - -I -L -C - -s - -m -o -t -p -m -y -S -& - -s -s -e -n -l -l -I -n -i -a -M - -s -y -a -w -h -t -a -P -e -r -u -s -o -p -x -E -n -i -a -M - -e -s -a -e -s -i -D - -m - -s -i -n -a -g -r -O - -n -o -i -s -s -i - -m -s -n -a -r -T - -C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 7 - -e -l -c -s -u -m - -, -a -e -s -u -a -n -, -r -e -v -e -f - -, -g -n -i -p -m -a -r -c - -l -a -n -i -m -o -d -b -a - -, -a -e -h -r -r -a -i -D - -l -a -t -a -f - -e -b -n -a -c - -, -n -i -a -p -l -a -n -i -m -o -d -b -a - -, -e -s -i -a -l -a -m - -, -r -e -v -e -F - -s -p -m -a -r -c -h -c -a -m -o -t -s -d -n -a - -, -r -e -v -e -f - -, -a -e -h -r -r -a -i -D - -s -p -m -a -r -c -h -c -a -m -o -t -s - -d -n -a - -, -r -e -v -e -f - -, -a -e -h -r -r -a -i -D - -s -e -h -c -a - -e -m -o -r -d -n -y -s - -c -i -m -e -a -r -u -- -c -i -t -y -l -o -m -e -a -h - -d -n -a - -, -a -e -h -r -r -a -i -d -y -d -o -o -l -B - -n -i -a -p -l -a -n -i -m -o -d -b -a - -, -r -e -v -e -f - -, -a -e -h -r -r -a -i -D - -c -i -n -o -m -u -e -n -p -g -n -i -n -e -t -a -e -r -h -t -- -e -f -i -l - -o -t - -n -o -i -s -e -l - -n -i -k -s - -d -l -i - -M - -h -t -i - -w - -t -c -a -t -n -o -c - -, -y -l -f - -r -e -e -d -r -o - -k -c -i -t - -f -o - -s -e -t -i - -B - -, -r -i -a -h -g -u -o -r -h -t - -e -b -n -a -c - -, -s -l -a -m -i -n -a -d -e -t -c -e -f -n -i - -r -e -t -a -w - -) -F -D - -, - -( -d -e -t -a -n -i -m -a -t -n -o -C - -s -p -m -a -r -c - -e -l -c -s -u -m - -, -g -n -i -t -i - -m -o -v -, -a -e -h -r -r -a -i -d - -e -r -e -v -e -S - -) -F - -, - -D - -( - -d -e -t -a -n -i -m -a -t -n -o -C - -r -e -v -e -f - -, -n -i -a -p -l -a -n -i -m -o -d -b -a - -, -a -e -h -r -r -a -i -D - -) - -D -s -s -e -l - -, - -F -( -d -e -t -a -n -i -m -a -t -n -o -C - -a -i -m -e -c -i -t -p -e -s - -, -a -e -h -r -r -a -i -D - -d -n -u -o -w -n -e -p -o -, -h -s -i -f -l -l -e -h -S -d -e -t -a -n -i -m -a -t -n -o -C - -) -F -D - -, - -( - -) -F -D - -, - -( - -d -e -t -a -n -i -m -a -t -n -o -C - -s -s -e -n -l -l -i - -l -a -e -h -r -r -a -i -D - -i -l -o -c - -. - -E -c -i -n -e -g -i -x -o -t -o -r -e -t -n -E - -d -e -t -a -n -i -m -a -t -n -o -C - -s -s -e -n -l -l -i - -l -a -e -h -r -r -a -i -D - -. - -E -c -i -g -a -h -r -r -o -m -e -a -h -o -r -e -t -n -E - -) - -R - -, - -F - -) -F - -) -F - -, - -D - -( -d -e -t -a -n -i -m -a -t -n -o -C - -, - -D - -( - -d -e -t -a -n -i -m -a -t -n -o -C - -, - -D - -( -d -e -t -a -n -i -m -a -t -n -o -C - -s -i -s -e -l -l -e -n -o -m -l -a -S - -s -i -s -o -l -l -e -g -i -h -S - -d -i -o -h -p -y -T - -s -i -s -o -i -r -b -i -V - -a -r -e -l -o -h -C - -e -a -r -e -l -o -h -c -o -i -r -b -i -V - -P -P -S -o -i -r -b -i -V - -. -p -p -s - -a -l -l -e -n -o -m -l -a -S - -i -h -p -y -t - -a -l -l -e -n -o -m -l -a -S - -. -p -p -s - -a -l -l -e -g -i -h -S - -a -i -r -e -t -c -a -B - -s -i -s -o -i -r -e -t -c -a -b -o -l -y -p -m -a -C - -. -p -p -s - -r -e -t -c -a -b -o -l -y -p -m -a -C - -i - -g -n -p -u -o -r -G - -e -n -r -o -b -r -e -t -a - -W - -s -i -s -o -i -n -i -s -r -e -Y - -a -i -m -e -a -r -a -l -u -T - -s -i -s -n -e -r -a -l -u -t - -a -l -l -e -s -i -c -n -a -r -F - -. -p -p -s -a -i -n -i -s -r -e -Y - -i -l -o -c - -) -e -n -r -o -b -- -r -o -t -c -e -V - -( - -d -e -t -a -l -e -r -- -r -e -t -a - -W - -n -i -k -s - -e -h -t - -r -e -d -n -u -s -e -l -u -d -o -n -d -n -a - -, -e -s -a -e -s -i -d -e -y -e - -, -h -s -a -r -n -i -k -S - -m -r -o -w -e -y -e - -, -s -e -u -g -i -t -a -f - -, -n -i -a -p -t -n -i -o -j - -d -n -a - -e -l -c -s -u -m - -, -y -h -c -t -I - -y -r -a -n -o -m -l -u -p -, -a -m -e -d -e -h -p -m -y -l - -, -c -i -t -a -m -o -t -p -m -y -s -a - -t -s -o -M - -n -o -i -t -a -d -r -a -t -e -r -h -t -w -o -r -g -, -l -o -o -t -s -y -r -e -t -a -w -d -n -a -y -d -o -o -l -B - -a -i -l -i -h -p -o -n -i -s -o -e - -n -e -r -d -l -i -h -c - -n -i - -n -o -i -t -a -d -r -a -t -e -r -h -t -w -o -r -g -, -t -r -o -f -m -o -c -s -i -d -l -a -n -i -m -o -d -b -A - -) -F - -, - -D - -( - -) -F -D - -, - -( - -l -a -r -o -- -l -a -c -e -F - -l -a -r -o -- -l -a -c -e -F - -o -t -i -u -q -s -o -m -y -b - -e -t -i - -B - -y -l -f -k -c -a -l -b -y -b -e -t -i - -B - -y -l -f -r -e -e -d -y -b -e -t -i - -B - -s -i -s -a -i -o -L - -s -u -o -i -r -a -V - -a -i -r -e -t -c -a -b -o -n -a -y -C - -a -o -l - -a -o -L - -e -a -g -l -A - -s -i -r -a -c -s -A - -s -e -d -i -o -c -i -r -b -m -u -l - -s -i -r -a -c -s -A - -s -u -n -a -c -i -r -e -m -a - -r -o -t -a -c -e -N -d -n -a - -n -o -i -t -c -e -f -n -i - -m -r -o -w -p -i -h -W - -s -i -s -a -i -r -a -l -i -f - -c -i -t -a -h -p -m -y -L - -a -r -u -i -h -c -i -r -t - -s -i -r -u -h -c -i -r -T - -i -t -f -o -r -c -n -a -b - -a -i -r -e -r -e -h -c -u -W - -d -e -t -a -l -e -r -- -r -e -t -a - -W - -s -i -s -a -i -c -r -e -c -o -h -c -n -O - -s -u -l -u -v -l -o -v - -a -c -r -e -c -o -h -c -n -O - -) -e -n -r -o -b -- -r -o -t -c -e -V - -( - -c -i -t -o -r -u -e -n - -, -e -g -a -m -a -d - -r -e -v -i -l - -, -s -s -e -n -l -l -i - -l -a -n -i -t -s -e -t -n -i - -, -s -i -t -i -t -a -m -r -e -D - -t -c -e -r -i -d -d -n -a -D -a -i -v -s -n -i -x -o -t -o -t - -e -r -u -s -o -p -x -E - -s -n -o -i -t -c -a -e -r - -t -c -a -t -n -o -c - -y -r -o -t -a -r -i -p -s -e -r - -, -a -i -s -e -h -t -s -e -r -a -p - -, -s -m -o -t -p -m -y -s - -l -a -n -i -t -s -e -t -n -i -o -r -t -s -a -G - -n -i -x -o -t -y -b -) -F -( -d -e -t -a -n -i -m -a -t -n -o -C - -h -s -i -f -l -l -e -h -s - -c -i -x -o -t -o -r -u -e -N - -s -i -v -e -r -b - -a -i -n -e -r -a -K - -n -o -i -t -a -t -i -r -r -i - -e -y -e - -d -n -a - -g -n -i -n -o -s -i -o -p - -a -i -s -e -h -t -s -e -r -a -p - -y -t -i - -m -e -r -t -x -e - -, -e -s -i -a -l -a -m - -, -g -n -i -t -i - -m -o -v -, -a -e -h -r -r -a -i -D - -n -i -x -o -t -y -b -) -F -( -d -e -t -a -n -i -m -a -t -n -o -C - -g -n -i -n -o -s -i -o -p -h -s -i -f - -a -r -e -t -a -u -g -i -C - -p -p -s - -s -u -c -s -i -d -r -e -i -b -m -a -G - -s -i -s -y -l -a -r -a -p - -g -n -i -n -o -s -i -o -p - -y -r -o -t -a -r -i -p -s -e -r - -, -a -i -s -e -h -t -s -e -r -a -p - -y -t -i - -m -e -r -t -x -e -d -n -a - -l -a -r -o -m -u -c -r -i - -C - -n -i -x -o -t - -y -b -) -F -( - -d -e -t -a -n -i -m -a -t -n -o -C - -h -s -i -f -l -l -e -h -s - -c -i -t -y -l -a -r -a -P - -p -p -s -m -u -i -r -d -n -a -x -e -l -A - -d -e -t -a -l -e -r -- -r -e -t -a - -W - -. -a -d -i -r -o -l -F -n -i - -e -c -n -a -c -i -f -i -n -g -i -s -h -t -l -a -e -h -c -i -l -b -u -p -r -o -j -a -m - -f -o -s -e -s -a -e -s -i -d -/ -s -n -e -g -o -h -t -a -p - -s -e -t -a -c -i -d -n -i -d -e -r -n -i - -d -e -t -h -g -i -l -h -g -i -H - -. -2 -; -) -F -( - -s -d -o -o -f - -d -n -a -, -) -R -( - -r -e -t -a -w - -l -a -n -o -i -t -a -e -r -c -e -r - -, -) - -D - -( - -r -e -t -a -w -g -n -i -k -n -i -r -d -: -s -y -a -w -h -t -a -p -e -r -u -s -o -p -x -e - -n -i -a -m -r -o -f - -s -n -o -i -t -a -i -v -e -r -b -b -A - -. - -1 -: -e -t -o -N - -e -g -a -m -a -d - -l -a -n -i -t -s -e -t -n -i - -d -n -a -y -r -a -n -i -r -u -, -a -e -h -r -r -a -i -D - -r -e -t -a -w -h -t -i - -w - -t -c -a -t -n -o -c - -l -a -m -r -e -D - -s -i -s -a -i -m -o -s -o -t -s -i -h -c -S - -. -p -p -s - -a -m -o -s -o -t -s -i -h -c -S - -s -h -t -n -i -m -l -e -H - -a -i -m -e -n -a - -d -n -a - -, -e -u -g -i -t -a -f - -, -e -t -i -t -e -p -p -a - -f -o -s -s -o -l - -, -a -e -h -r -r -a -i -D - -l -i -o -s -h -t -i - -w - -t -c -a -t -n -o -c - -l -a -m -r -e -D - -n -o -i -t -c -e -f -n -i - -m -r -o -w -k -o -o -H - -e -l -a -n -e -d -o -u -d -a -m -o -t -s -o -l -y -c -n -A - -n -o -i -t -a -m -m -a -l -f -n -i -d -n -a - -r -e -c -l -u -l -u -f -n -i -a -P - -) - -D - -( -d -e -t -a -n -i -m -a -t -n -o -C - -s -i -s -a -i -l -u -c -n -u -c -a -r -D - -s -i -s -n -e -n -i -d -e -m -s -u -l -u -c -n -u -c -a -r -D - -d -e -s -a -b -- -r -e -t -a - -W - - 1 3 8 • S O N G L I A N G E T A L . - -Figure 4.3. Reported outbreaks (bar) and human cases (line) of food-borne and waterborne diseases in -Florida each year from 1989 to 2011 (Source: Florida Department of Health). Toxigenic V. cholerae -infections are highly unusual in Florida and the first V. cholerae O75 outbreak was detected and reported -in Florida between March 23 and April 13, 2011. - -Figure 4.4. Monthly distribution of reported outbreaks and cases of food-borne and waterborne diseases in -Florida, 2011 (Source: Florida Department of Health) - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 3 9 - -c) Vector-Borne Diseases -Vector-borne diseases pose serious public health threats throughout the world. According to the -World Health Organization, vector-borne diseases account for more than 17% of infectious -diseases, causing more than one million deaths annually from diseases such as malaria, dengue -fever, yellow fever, Japanese encephalitis, and schistosomiasis (WHO 2016). Vector-borne -diseases are transmitted through vector organisms including mosquitoes, ticks, flies, fleas, and -snails. The transmission of vector-borne diseases takes two primary transmission pathways. The -first is the human-vector pathway, with transmission through a vector biting a host; pathogens -survive outside hosts in arthropod vectors and humans are typically the only host (Eisenberg et -al. 2007). Diseases in this group include malaria, dengue fever, onchocerciasis, trypanosomiasis, -and filariasis (Table 4.1). The second pathway involves zoonotic transmission, with a vector -biting a nonhuman host as the transmission route and humans as the dead-end host (Eisenberg et -al. 2007). Diseases in this group include Lyme disease, Yellow fever, West Nile virus, and -Japanese encephalitis (Table 4.1). The transmission and spread of vector-borne diseases are -determined by complex interactions between the host (either human or nonhuman), the vector -(e.g., mosquitos, flies, snails, and ticks), and various pathogens ranging from protozoans, viruses, -bacteria, and helminthes (Table 4.1). Important biological properties underlying the transmission -of vector-borne diseases include survival, development, and reproduction of the vector; also the -pathogens in the vector, the vector’s biting rate, and the behavior of hosts (human and -nonhuman), all of which are associated with climatic conditions. The following sections discuss -some specific aspects of the effects of rainfall, temperature, and flooding with regard to vector- -borne diseases. - -Temperature Sensitivity -Through the modulating effects on physiological activities and tolerance limits of vectors and -infectivity of pathogens, temperature can affect distribution of the vector and transmission of the -pathogens between vectors and hosts. Some key mechanistic effects of temperature on vectors, -vector-borne pathogens and their transmission are summarized by Gulber et al. (2001). -• Change in survival of vectors (e.g. increase or decrease depending on species) -• Change in susceptibility of vectors to pathogens -• Change in vector population growth -• Change in feeding behavior and host contact -• Change in incubation period of pathogens in vectors (e.g. decreased extrinsic incubation - -period at higher temperatures) - -• Change in pathogen replication and infectivity -• Change in transmission season - -A large number of studies on vector-borne diseases at different scales have provided -empirical evidence and predictions on the climatic factors and vector-borne disease links. For - - 1 4 0 • S O N G L I A N G E T A L . - -example, studies have shown significant relationships between ambient temperature and malaria -transmission for both mean temperature (Loevinsohn 1994; Lindblade et al. 2000; Thomson et -al. 2005; Tian et al. 2008; Wardrop et al. 2013) and its daily fluctuations (Paaijmans et al. 2010). -Laboratory studies have shown clear temperature-dependent effects on dengue virus infection in -mosquitoes (Alto and Bettinardi 2013). Fluctuations at a low mean temperature accelerate -dengue virus transmission in Aedes aegypti mosquitoes (Carrington et al. 2013). A recent -systematic review and meta-analysis based on 137 studies on the relationship between dengue -risk and temperature suggests that the dengue transmission and risk are sensitive to temperature, -with a positive relationship between them for a minimum range (18.1–24.2 °C) to a maximum -temperature range (28.0–34.5 °C) (Fan et al. 2014). A regional predictive study reported that -increased temperature increases the likelihood of many vector-borne diseases, in particular -malaria, dengue fever, and Lyme disease in new areas (Githeko et al. 2000). Similar findings are -also reported for West Nile virus (Soverow et al. 2009). These empirical findings have been used -to parameterize mechanistic models of temperature driven transmission for vector-borne -diseases, which can be coupled to projected climate scenarios to anticipate shifting geographies -of transmission risk, e.g. for malaria (Mordecai et al. 2013; Johnson et al. 2014; Johnson et al. -2015; Ryan et al. 2015; Ryan et al. 2015) and for Aedes spp transmitted diseases of concern for -Florida: dengue, chikugunya, and zika viruses (Moedecai et al. 2017; Ryan et al. 2017). - -Rainfall -Rainfall can supply water to both transient and persistent environments that can serve as habitats -or breeding sites for vectors and/or for part of pathogens’ lifecycles, critical for transmission of -vector-borne diseases. Meanwhile, rainfall can also have negative impacts on the habitats (for -example, washing away), thereby reducing the transmission risk. Variations in rainfall may have -direct and indirect impacts on the distribution and timing of transmission of vector-borne -diseases. Some key mechanistic effects of rainfall on vectors, vector-borne pathogens, and their -transmission are summarized by Gubler et al. (2001). -• -• Excess rain (e.g. causing flooding) may eliminate habitat for both vector and hosts. -• -• -• - -Increased humidity may increase vector survival and biting rate. -Increased rainfall may facilitate transport of pathogens for certain distance (Liang et al. 2007). -Increased rain may increase vertebrate host size through increased food availability (e.g. -vegetation). - -Increased rain may increase larval habitat and vector population. - -• Flooding may decrease vertebrate host size (e.g. reduced food availability) but increase - -interactions (e.g. contacts) with humans. - -Such mechanistic links have been reported by many studies including for diseases such as -malaria (van der Hoek et al. 1997; Thomson et al. 2005; Galardo et al. 2009; Gao et al. 2012; -Yang et al. 2012), dengue fever (Li et al. 1985; Su 2008; Johansson et al. 2009; Banu et al. 2011; - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 1 - -Hii et al. 2012), West Nile virus (Shaman et al. 2005; Soverow et al. 2009), and Lyme disease -(Ogden et al. 2006; Ostfeld et al. 2006). - -Vector-Borne Diseases in Florida -In the United States, 14 vector-borne diseases are of primary public health concern. Table 4.2 -lists key notifiable vector-borne diseases and reported cases between 2013 and 2015 in Florida. -Major mosquito-borne diseases include the St. Louis encephalitis virus, West Nile virus, Eastern -equine encephalitis virus, Western equine encephalitis virus, Venezuelan equine encephalitis -virus, Everglades virus, and the California serogroup viruses including La Crosse encephalitis -virus, all of which are transmitted by mosquitoes. More recently, concern about Aedes spp -transmitted flavivirus and alphaviral disesaes have become more prevalent: dengue (DENV), -chikungunya (CHIKV), and zika (ZIKV). The 2015-2016 global public health emergency for -Zika declared by the WHO focused attention on Florida as an introduction and establishment -gateway for the U.S. - -Table 4.2. A summary of reported case counts of notifiable vector-borne diseases in Florida. (Source: -Florida Department of Health). - -2013-2015 Reported Cases Number of Counties Reported - -Diseases -Tick-borne -Lyme disease -Spotted Fever Rickettsia -Anaplasmosis/Ehrlichiosis -Tularemia -Mosquito-Borne -West Nile virus -Malaria -Dengue -California serogroup viruses 2 -3 -Eastern equine encephalitis -2 -St. Louis encephalitis - -459 -74 -82 -2 - -37 -146 -331 - -46 -18 -30 -2 - -14 -24 -31 -2 -3 -1 - -St. Louis Encephalitis. The St. Louis encephalitis virus, a flavivirus, was the most common -mosquito-transmitted human pathogen in the U.S. prior to the introduction of the West Nile virus -in 1999 (FDH 2016). In Florida, the principal vector is Culex nigripalpus, a ubiquitous species -throughout the state. No human case has been reported in Florida since 2003. However, sentinel -chickens testing positive for antibodies to the St. Louis encephalitis virus have been reported in -several counties in Central and South Florida, suggesting the potential for a possible resurgence -(FDH 2014, 2016). - -West Nile virus. The peak period of transmission in Florida for the West Nile virus is July -through September. The natural cycle of West Nile virus involves Culex mosquitoes and wild -birds. Between 2001 and 2013, 318 human West Nile virus cases were reported in Florida and - - 1 4 2 • S O N G L I A N G E T A L . - -West Nile virus activities have been consistently reported in many counties around the state -(FDH 2014). - -Eastern Equine Encephalitis. Eastern equine encephalitis occurs in natural cycles involving -birds and Culiseta melanura in freshwater swampy areas, with a peak in transmission occurring -between May and August. Historical and current evidence indicates limited human Eastern -equine encephalitis epidemic potential in Florida (FDH 2014, 2016). - -Dengue Fever. In Florida, a historical dengue fever epidemic occurred in 1934-1935 and then -it ceased. It re-emerged in 2009, and since then a small number of cases have been reported each -year among individuals who had previous travel to dengue-endemic countries. In the summer of -2009, local dengue transmission was identified in Key West, Florida. Since then, sporadic local -transmission has been identified in other Central and South Florida counties during 2010, 2011, -2012, and 2013 (FDH 2014, 2016). The presence of the Aedes spp mosquito, both Aedes aegypti, -also known as the Yellow Fever mosquito, which has been a persistent invasive mosquito in -Florida for a couple of centuries, and the more recently introduced Asian Tiger Mosquito, Aedes -albopictus, maintain transmission potential in Florida. Aedes mosquitoes are capable of breeding -in as little as a teaspoon of water, and are able to adapt well to urban environments, and exhibit -transmission of dengue, chikungunya, and zika, at temperature ranges found almost year-round -in most of Florida (Mordecai et al. 2017). The optimal transmission temperature range for Aedes -aegypti, in particular, is higher than that for Aedes albopictus, suggesting that future climates -will promote Ae aegypti transmission in Florida, before it becomes too hot for transmission (Ryan -et al. 2017). - -Malaria. Malaria is caused by the mosquito-borne parasite Plasmodium falciparum. Endemic -malaria was eliminated in Florida in the late 1940s. However, imported cases—either from -travelers returning to the state from malaria-endemic regions or tourists carrying the pathogen— -have been reported. The Anopheles mosquitoes responsible for transmitting the malaria parasite -to humans are common in the state so establishment of local transmission is still possible (FDH -2014, 2016). - -Tick-Borne Diseases. The most common tick-borne diseases are ehrlichiosis, anaplasmosis, -Lyme disease, Rocky Mountain spotted fever, and other spotted fever illnesses. Lyme disease is -the most commonly reported vector-borne disease in the United States. It is caused by a bacteria -known as Borrelia burgdorferi. Lyme disease is reported in Florida year-round. An estimate -suggests that about 23% of cases were acquired in Florida and about 77% were acquired while -travelling to other states or countries (FDH 2016). Ehrlichiosis and Anaplasmosis—several -pathogenic species in the genus Ehrlichia and Anaplasma—can cause human illness. The human -illness caused by Ehrlichia chaffeensis is called Human Monocytic Ehrlichiosis, and the illness -caused by Anaplasma phagocytophilum is called Anaplasmosis or Human Granulocytotropic -Anaplasmosis. In Florida, the majority of Human Monocytic Ehrlichiosis cases (73%) are -acquired in Florida, primarily in the northern and central parts of the state, and Human -Granulocytotropic Anaplasmosis cases are even more likely to be acquired in Florida. Cases are - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 3 - -reported year-round, with a peak occurring during the spring and summer months (FDH 2016). -Rocky Mountain Spotted Fever is a disease is caused by the bacterium Rickettsia richerrsii and, -in Florida, it is transmitted primarily by the American dog tick. It can also be caused by -Dermacentor variabilis and cases of this transmission are reported year-round with more than -70% acquired in Florida, and the majority of cases are reported in the northern and central regions -of the state (FDH 2016). - -Drought -Drought is considered a meteorological anomaly characterized by a prolonged and abnormal -moisture deficiency. Climate change and variability are progressively increasing the severity and -frequency of drought events. Drought has many possible public health implications (Kalis et al. -2009; Kalis and Curtiss 2016). Drought may: -• Compromise quantity and quality of water for drinking, sanitation, and hygiene -• Compromise air quality (e.g. increased particulate matters in the air) -• Reduce crop yield, food availability and nutrition -• - -Increase or decrease risks to infectious pathogens (e.g. waterborne, air-borne, and vector- -borne) -Increase risk of non-communicable illness (e.g. respiratory infections and/or illness) - -• - -Drought has been linked to deterioration of water quality (Golladay and Battle 2002; Al- -Kharabsheh and Ta'any 2003) and air quality (Taylor and Davies 1990; Field et al. 2009) in -various settings in both the developed and developing world. Drought has been implicated for -infectious diseases—waterborne (Lipp et al. 2002; Schuster et al. 2005; Senhorst and Zwolsman -2005), vector-borne (Chretien et al. 2007; Erlanger et al. 2009; Medlock and Leach 2015), and -airborne (Polymenakou et al. 2008). Drought conditions have also been linked to increased -incidence of respiratory diseases (Smith et al. 2014). - -In Florida, the BRACE study examined associations between monthly drought conditions -(using the Standardized Precipitation Index), disease rates and emergency department visits for -specific health conditions (e.g. allergic rhinitis, asthma, and all respiratory diseases excluding -asthma) among Florida residents from 2005 to 2012. The results suggested that drought was -significantly associated with emergency department visits for respiratory diseases. It is worth -noting that the relationship between the most extreme drought conditions and emergency -department visits tended to be protective (e.g. less emergency department visits for respiratory -illness), while more moderate drought appeared to be associated with an increase in emergency -department visits for the health conditions. A similar pattern was also observed for asthma (FDH -2014b). - - 1 4 4 • S O N G L I A N G E T A L . - -Indirect Impact Pathway – Mental Health, Well-Being, -and Community Health - -Mental health disorders and well-being are an important aspect of the human health impacts -associated with climate change and variability, and associated weather events. Many mental -health consequences, ranging from simple stress and distress symptoms to some severe -conditions including anxiety, distress, posttraumatic stress disorder (PTSD), and even suicidal -tendency have been the focus of much of the research and public health responses in the past few -decades. Here are some key mental disorders and community health issues that may be related -to climate change and associated weather events, based on the U.S. Global Change Research -Program report (Crimmins et al. 2016). - -Experience with extreme weather events, such as flooding, drought, and hurricanes, may - -cause the following mental disorders: -• -• -• -• - -Increased stress, anxiety, depression, grief, and even suicidality; -Increased tensions on social stress and relationships; -Increased substance abuse; and -Increased risk of PTSD. -At the community level (Crimmins et al. 2016), potential impacts may include: -• -Increased interpersonal aggression, -• -Increased violence and crime, -• -Increased social instability, and -• Decreased community cohesion. - -Assessing the impacts of natural disasters on mental or community health and well-being of -affected populations, as well as mediation techniques, has been a focus of much research, -community and public health responses. In Florida, studies have suggested that anxiety disorder, -major depressive episodes, and PTSD among affected populations could be significantly -associated with storm exposure and displacement during the occurrence of Hurricane Andrew in -1992 (David et al. 1996), as well as a number of hurricanes in 2004 (Acierno et al. 2007) and -2012 (Neria and Shultz 2012). Studies have also indicated that social support would help to -alleviate such negative impacts (Acierno et al. 2007). Similar impacts have been reported for -community and public health workers who responded to these weather events. For example, -during the 2004 Florida hurricane season, four hurricanes (Charley, Frances, Ivan, and Jean) as -well as one tropical storm (Bonnie) hit Florida during a period of seven weeks. Increased PTSD, -other mental health issues, and substance abuse were reported among public health workers -during and after these events (Fullerton et al. 2013; Fullerton et al. 2015). The studies also -suggested that these mental health outcomes were influenced by multiple community -characteristics including the collective efficacy of neighborhood populations in the community. - - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 5 - -Significantly lower depressive symptoms were associated with communities that had sufficient -resources and received social support (Acierno et al. 2007, Fullerton et al. 2013, Fullerton et al. -2015). - -Impacts on children following natural disasters are of particular concern. For example, -Hurricane Andrew was a devastating category 5 Atlantic hurricane that struck South Florida in -1992. Studies showed significant impacts of hurricane exposure, stressors occurring during the -hurricane, and recovery periods on children’s persistent posttraumatic stress. In the Miami-Dade -County area, 35% to 60% of children surveyed reported moderate to very severe levels of -posttraumatic stress symptoms (Vernberg et al. 1996; La Greca et al. 2010), with hurricane- -related stressors influencing children’s persistent posttraumatic stress symptoms and other life -events in later stages of children’s life (La Greca et al. 2010; Weems et al. 2010; La Greca et al. -2013). These findings have important community health implications for identifying and -potentially helping youth in the aftermath of natural disasters. - -Conclusion - -The current available evidence has clearly indicated increases in average temperature, total -annual precipitation, frequency of extreme temperature conditions and heavy precipitation in the -United States in the past five decades, as well as increases in the tropical cyclone activity in -regions along the Atlantic Ocean, the Caribbean, and the Gulf of Mexico. Similar patterns are -expected in the future if no remediation is conducted (USEPA 2016). Florida is likely to face -even greater effects than the rest of the US, particularly effects associated with sea level rise, heat -waves, storm surge, hurricanes, and others (FOCC 2010) that affect human health in a variety of -different ways. Some effects are through direct exposure pathways, such as hurricane/storm and -temperature-related illnesses, injuries, and deaths. Some are through indirect exposure pathways -such as waterborne and vector-borne diseases. And still others are through social and economic -pathways, such as stress and mental illness. In addition to these direct and indirect effects, we -see coupled and compounded effects of climate change, which are anticipated to be exacerbated -in the future. For example, Florida is home to a large elderly population; the warm climate and -housing development capacity made this ideal. However, the elderly are more vulnerable to the -effects of heatwaves, less mobile in the event of hurricanes, and suffer greater impacts of the -symptomology of febrile vectorborne diseases. This puts the state in a position higher public -health urgency by virtue of an interaction of demographic profile and increased impact of climate -change due to geography. Given the increase in hurricances and storm impacts expected under -climate change, we are likely to also see the added impact of vectorborne disease risk anticipated -in the aftermath of such disasters. For example, the additional potential vector breeding sites -provided by debris and damaged infrastructure after hurricanes have been noted as concerns for -Zika transmission in the 2016-2017 hurricane season. An additional component of this is the - - 1 4 6 • S O N G L I A N G E T A L . - -unanticipated exposures post-disaster, in which both displaced residents and rescue workers have -radically increased outdoor exposure, reduced access to air conditioned spaces, and to basic -protection measures such as insect repellent and appropriate protective clothing. These are just -two of the compounded scenarios of public health concern that the state of Florida faces in a -changing climate. Despite the comprehensive information reviewed in the chapter, many things -are still unknown or are not well understood about the health impacts of Florida’s changing -climate. For instance, what are the impacts of sea level rise, flooding, and other extreme weather -events on water contamination that will ultimately influence human health? How will future -warming trends and other climate-sensitive hazards influence the transmission and spread of -existing and/or new waterborne and vector-borne diseases? What are the disease burdens and -their distributions attributable to known climate-sensitive hazards in the state? To address these -challenges, policies focused on mitigation, adaptation strategies, health surveillance, and -research that could close knowledge gaps on human exposures to these climate-sensitive hazards -and associated health impacts are much needed. - -Acknowledgments +Gilmanton High School -The authors thank Melissa Jordan for her instrumental discussion on this chapter. Song Liang was supported -in part by grants from NSF/EEID (EF-1015908) and NSF/WSC-Category 3 (1360330). Sadie J. Ryan was -supported in part by grants from NSF/EEID (DEB-1518681), NSF RAPID (DEB-1641145), and an Early -Career Fellowship from the Florida Climate Institute. +Personal Finance +Semester Syllabus -References +Part 1: Course Information -Acierno, R., K. J. Ruggiero, S. Galea, H. S. Resnick, K. Koenen, J. Roitzsch, M. de Arellano, J. Boyle, and -D. G. Kilpatrick. 2007. Psychological sequelae resulting from the 2004 Florida hurricanes: implications -for postdisaster intervention. Am J Public Health 97 Suppl 1:S103-108. +Instructor Information -Al-Kharabsheh, A., and R. Ta'any. 2003. Influence of urbanization on water quality deterioration during +Instructor: Julie Kacures +E-mail: jkacures@ghs.k12.wi.us -drought periods at South Jordan. Journal of Arid Environments 53:619-630. +Course Description -Alto, B. W., and D. Bettinardi. 2013. Temperature and dengue virus infection in mosquitoes: independent +Students need financial knowledge to ensure they make the best +financial decisions. Whether it is completing transactions for daily use +or analyzing investments for their future, students must understand +how to use resources to develop economic growth. Topics will include +banking activities, investments, budgeting, credit, taxes, consumer +rights, and insurance. -effects on the immature and adult stages. Am J Trop Med Hyg 88:497-505. + This course is a graduation requirement. -Aslar, A. K., A. Soran, Y. Yildiz, and Y. Isik. 2001. Epidemiology, morbidity, mortality and treatment of -lightning injuries in a Turkish burns units. International Journal of Clinical Practice 55:502-504. -Balbus, J., A. Crimmins, J. L. Gamble, D. R. Easterling, K. E. Kunkel, S. Saha, and M. C. Sarofim. 2016. -Ch. 1: Introduction: Climate Change and Human Health. Pages 25–42 The Impacts of Climate Change -on Human Health in the United States: A Scientific Assessment. U.S. Global Change Research -Program, Washington, DC. +Textbook & Course Materials -Bandyopadhyay, S., S. Kanji, and L. M. Wang. 2012. The impact of rainfall and temperature variation on + Managing Your Personal Finances. Ryan, Joan S. South- -diarrheal prevalence in Sub-Saharan Africa. Applied Geography 33:63-72. +Western Publishing. ISBN 0-538-69965-5. -Banu, S., W. Hu, C. Hurst, and S. Tong. 2011. Dengue transmission in the Asia-Pacific region: impact of + Foundations in Personal Finance. Ramsey, Dave. Ramsey -climate change and socio-environmental factors. Trop Med Int Health 16:598-607. +Solutions. ISBN 9 7891936-948123. -Basu, R., and J. M. Samet. 2002. Relation between elevated ambient temperature and mortality: A review + Building Your Future. The Actuarial Foundation. Modules 1-4. -of the epidemiologic evidence. Epidemiologic Reviews 24:190-202. + Take Charge Today. University of Arizona (Family Economics and -Berko, J., D. D. Ingram, S. Saha, and J. D. Parker. 2014. Deaths attributed to heat, cold, and other weather +Financial Education -- FEFE). Online resource. +https://takechargetoday.arizona.edu/. -events in the United States, 2006-2010. Natl Health Stat Report:1-15. + Practical Money Skills. VISA and Federal Reserve Bank of -Booij, M. J. 2005. Impact of climate change on river flooding assessed with different spatial model +Chicago. Online resource. https://www.practicalmoneyskills.com. -resolutions. Journal of Hydrology 303:176-198. + Security First Bank Simulation. South-Western Publishing. -Bouma, M. J., R. S. Kovats, S. A. Goubet, J. S. Cox, and A. Haines. 1997. Global assessment of El Nino's +ISBN 0-538-43187-3 -disaster burden. Lancet 350:1435-1438. +Page 1 - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 7 + Personal Finance -Carlton, E. J., J. N. S. Eisenberg, J. Goldstick, W. Cevallos, J. Trostle, and K. Levy. 2014. Heavy Rainfall -Events and Diarrhea Incidence: The Role of Social and Environmental Factors. American Journal of -Epidemiology 179:344-352. +Syllabus -Carrington, L. B., M. V. Armijos, L. Lambrechts, and T. W. Scott. 2013. Fluctuations at a Low Mean -Temperature Accelerate Dengue Virus Transmission by Aedes aegypti. Plos Neglected Tropical -Diseases 7. +Grading -Center for Disease Control, and Prevention. 1994. Heat-related deaths--Philadelphia and United States, +50% -1993-1994. MMWR Morb Mortal Wkly Rep 43:453-455. +Homework and Assignments -Chen, M. J., C. Y. Lin, Y. T. Wu, P. C. Wu, S. C. Lung, and H. J. Su. 2012. Effects of extreme precipitation +50% -to the distribution of infectious diseases in Taiwan, 1994-2008. PLoS One 7:e34651. +Tests / Quizzes -Chretien, J. P., A. Anyamba, S. A. Bedno, R. F. Breiman, R. Sang, K. Sergon, A. M. Powers, C. O. Onyango, -J. Small, C. J. Tucker, and K. J. Linthicum. 2007. Drought-associated chikungunya emergence along -coastal East Africa. Am J Trop Med Hyg 76:405-407. +School Homework Policy -Christensen, J. H., and O. B. Christensen. 2003. Climate modelling: Severe summertime flooding in Europe. +After 2 episodes of late homework (grace period), homework that is +one day late will be graded at 80%. Any work that is late 2 days or +more will receive 66%. -Nature 421:805-806. +Required Materials -Chung, J. Y., Y. Honda, Y. C. Hong, X. C. Pan, Y. L. Guo, and H. Kim. 2009. Ambient temperature and -mortality: an international study in four capital cities of East Asia. Sci Total Environ 408:390-396. -Collins, T. W., A. M. Jimenez, and S. E. Grineski. 2013. Hispanic health disparities after a flood disaster: -results of a population-based survey of individuals experiencing home site damage in El Paso (Texas, -USA). J Immigr Minor Health 15:415-426. + Notebook or loose leaf paper + Folder + Writing utensil + Calculator -Colwell, R. R., P. R. Epstein, D. Gubler, N. Maynard, A. J. McMichael, J. A. Patz, R. B. Sack, and R. +Mobile Device Policy -Shope. 1998. Climate change and human health. Science 279:968-969. + Any cell phone or other mobile device seen or heard in this classroom +during the school day will be confiscated by the teacher. The student +may pick up the device at the end of the day. -Crimmins, A., J. Balbus, J. L. Gamble, C. B. Beard, J. E. Bell, D. Dodgen, R. J. Eisen, N. Fann, M. D. -Hawkins, S. C. Herring, L. Jantarasami, D. M. Mills, S. Saha, M. C. Sarofim, J. Trtanj, and L. Ziska. -2016. Executive Summary. Pages 1–24 The Impacts of Climate Change on Human Health in the United -States: A Scientific Assessment. U.S. Global Change Research Program, Washington, DC. + After repeat offenses, a parent must come in to pick up the phone -Curriero, F. C., K. S. Heiner, J. M. Samet, S. L. Zeger, L. Strug, and J. A. Patz. 2002. Temperature and +after school. -mortality in 11 cities of the eastern United States. Am J Epidemiol 155:80-87. +School Tardy Policy -Curriero, F. C., J. A. Patz, J. B. Rose, and S. Lele. 2001. The association between extreme precipitation and -waterborne disease outbreaks in the United States, 1948-1994. American Journal of Public Health -91:1194-1199. + A student is considered tardy if he/she is not in his/her seat when the -David, D., T. A. Mellman, L. M. Mendoza, R. Kulick-Bell, G. Ironson, and N. Schneiderman. 1996. +second bell has rung. -Psychiatric morbidity following Hurricane Andrew. J Trauma Stress 9:607-612. + Students who are tardy for class will lose a portion of their class -Diez, J. J., M. D. Esteban, R. Paz, J. S. Lopez-Gutierrez, V. Negro, and J. V. Monnot. 2011. Urban Coastal +participation or classwork points. -Flooding and Climate Change. Journal of Coastal Research:205-209. + The teacher will deal with the first three tardies each quarter as a +classroom discipline problem. After three (3) tardies to a class, a +detention is assigned. -Douglas, I., K. Alam, M. Maghenda, Y. McDonnell, L. McLean, and J. Campbell. 2008. Unjust waters: -climate change, flooding and the urban poor in Africa. Environment and Urbanization 20:187-205. -Duclos, P. J., L. M. Sanderson, and K. C. Klontz. 1990. Lightning-related mortality and morbidity in + Tardies reset or start at zero at the beginning of each quarter. -Florida. Public Health Rep 105:276-282. +Page 2 -Eisenberg, J. N., M. A. Desai, K. Levy, S. J. Bates, S. Liang, K. Naumoff, and J. C. Scott. 2007. -Environmental determinants of infectious disease: a framework for tracking causal links and guiding -public health research. Environ Health Perspect 115:1216-1223. + Personal Finance -Ellis, K. N., L. M. Sylvester, and J. C. Trepanier. 2015. Spatiotemporal patterns of extreme hurricanes +Syllabus -impacting US coastal cities. Natural Hazards 75:2733-2749. +Part 2: Student Learning Outcomes -Elsner, James B., James P. Kossin, and Thomas H. Jagger. "The increasing intensity of the strongest tropical +1. -cyclones." Nature 455.7209 (2008): 92. +Competency: Relate the foundations of financial planning -Erlanger, T. E., S. Weiss, J. Keiser, J. Utzinger, and K. Wiedenmayer. 2009. Past, Present, and Future of +Objectives: -Japanese Encephalitis. Emerging Infectious Diseases 15:1-7. + Define key terms in the financial planning environment -Fan, J., W. Wei, Z. Bai, C. Fan, S. Li, Q. Liu, and K. Yang. 2014. A systematic review and meta-analysis + Define components of balance sheet, income and expense -of dengue risk with temperature change. Int J Environ Res Public Health 12:1-15. +statement, and personal budget -FDH. 2014a. Surveillance and control of selected mosquito-borne diseases in Florida. Florida Department + Compute equations using time value of money concepts -of Health, Tallahassee, Florida. + Compile amounts in a balance sheet -FDH. 2014b. Health effects of precipitation abundance and deficits in Florida. Tallahassee, Florida. -FDH. 2015a. Health effects of summer heat in Florida. Florida Department of Health, Tallahassee, Florida. -FDH. 2015b. Health effects of tropical storms and hurricanes in Florida. Florida Department of Health, + Describe a balance sheet template to show personal financial -Tallahassee, Florida. +position - 1 4 8 • S O N G L I A N G E T A L . + Describe a budget template to reflect personal financial position -FDH. 2016. Diseases and Conditions, Florida Department of Health, Tallahassee, Florida. -Field, C. B., V. R. Barros, and Intergovernmental Panel on Climate Change. Working Group II. 2014. -Climate change 2014: impacts, adaptation, and vulnerability: Working Group II contribution to the fifth -assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, -New York, NY. + Relate the concept of limited resources to personal financial -Field, R. D., G. R. van der Werf, and S. S. P. Shen. 2009. Human amplification of drought-induced biomass +position -burning in Indonesia since 1960. Nature Geoscience 2:185-188. +Assessments: -FOCC. 2010. Climate change and sea-level rise in Florida. An update of the effects of climate change on -Florida's ocean & coastal resources. Page 36 in T. F. O. a. C. Council, editor, Tallahassee, Florida -Frumkin, H., A. J. McMichael, and J. J. Hess. 2008. Climate change and the health of the public. Am J Prev + Create a list of personal financial goals -Med 35:401-402. + Create personal financial statements -Fullerton, C. S., J. B. A. McKibben, D. B. Reissman, T. Scharf, K. M. Kowalski-Trakofler, J. M. Shultz, -and R. J. Ursano. 2013. Posttraumatic Stress Disorder, Depression, and Alcohol and Tobacco Use in -Public Health Workers After the 2004 Florida Hurricanes. Disaster Medicine and Public Health -Preparedness 7:89-95. + Create a personal budget -Fullerton, C. S., R. J. Ursano, X. Liu, J. B. McKibben, L. Wang, and D. B. Reissman. 2015. Depressive -Symptom Severity and Community Collective Efficacy following the 2004 Florida Hurricanes. PLoS -One 10:e0130863. +Criteria. You will meet the objectives listed above through a +combination of the following activities in this course: -Galardo, A. K. R., R. H. Zimmerman, L. P. Lounibos, L. J. Young, C. D. Galardo, M. Arruda, and A. A. R. -D. Couto. 2009. Seasonal abundance of anopheline mosquitoes and their association with rainfall and -malaria along the Matapi River, Amapi, Brazil. Medical and Veterinary Entomology 23:335-349. -Gao, H. W., L. P. Wang, S. Liang, Y. X. Liu, S. L. Tong, J. J. Wang, Y. P. Li, X. F. Wang, H. Yang, J. Q. -Ma, L. Q. Fang, and W. C. Cao. 2012. Change in rainfall drives malaria re-emergence in Anhui -Province, China. PLoS One 7:e43686. + Create a list of five personal financial goals -Githeko, A. K., S. W. Lindsay, U. E. Confalonieri, and J. A. Patz. 2000. Climate change and vector-borne + Set a target date to meet those goals -diseases: a regional analysis. Bull World Health Organ 78:1136-1147. + Research the financial impact of each of those goals (in today’s -Golladay, S. W., and J. Battle. 2002. Effects of flooding and drought on water quality in gulf coastal plain +dollars) -streams in Georgia. Journal of Environmental Quality 31:1266-1272. + Calculate the annual savings requirement starting today in order to -Gubler, D. J., P. Reiter, K. L. Ebi, W. Yap, R. Nasci, and J. A. Patz. 2001. Climate variability and change -in the United States: potential impacts on vector- and rodent-borne diseases. Environ Health Perspect -109 Suppl 2:223-233. +reach the goals by the target date -Hashizume, M., B. Armstrong, S. Hajat, Y. Wagatsuma, A. S. Faruque, T. Hayashi, and D. A. Sack. 2008. + Compile a journal (minimum of 2 pages) of entries that reflect upon +what you learned from your listing of goals, what lifestyle changes +you may need to make to reach those goals, and whether your +goals changed because of the exercise. Following prompts, include +in your journal: target dates for each goal, dollar amounts of the +impact of each goal, and annual savings requirement -The effect of rainfall on the incidence of cholera in Bangladesh. Epidemiology 19:103-110. + Prepare a balance sheet -Hii, Y. L., H. P. Zhu, N. Ng, L. C. Ng, and J. Rocklov. 2012. Forecast of Dengue Incidence Using + Prepare a statement of income and expenses -Temperature and Rainfall. Plos Neglected Tropical Diseases 6. + Compile a journal (minimum of 2 pages) of entries that reflect on -Hoshiko, S., P. English, D. Smith, and R. Trent. 2010. A simple method for estimating excess mortality due -to heat waves, as applied to the 2006 California heat wave. International Journal of Public Health -55:133-137. +Page 3 -Huynen, M. M., P. Martens, D. Schram, M. P. Weijenberg, and A. E. Kunst. 2001. The impact of heat -waves and cold spells on mortality rates in the Dutch population. Environ Health Perspect 109:463-470. -IFRCRCS. 2015. World disasters report: focus on local actors, the key to humanitarian effectiveness + Personal Finance -International Federation of Red Cross and Red Crescent Societies. +Syllabus -Ji, F., Z. H. Wu, J. P. Huang, and E. P. Chassignet. 2014. Evolution of land surface air temperature trend. +the financial reports. Include a balanced balance sheet, an income +and expense statement, a reflection of your experience in creating +the reports. -Nature Climate Change 4:462-466. + Given a simulated family situation, analyze current income and -Johansson, M. A., D. A. T. Cummings, and G. E. Glass. 2009. Multiyear Climate Variability and Dengue- -El Nino Southern Oscillation, Weather, and Dengue Incidence in Puerto Rico, Mexico, and Thailand: -A Longitudinal Data Analysis. Plos Medicine 6. +expenses for a normal month. -Johnson, L. R., K.D. Lafferty, A. McNally, E. Mordecai, K. Paaijmans, S. Pawar, S.J. Ryan. 2014. -"Mapping the distribution of Malaria: Current Approaches and Future Directions". in Analyzing and -Modeling Spatial and Temporal Dynamics of Infectious Diseases. D. Chen, B. Moulin, J. Wu (Eds). -John Wiley & Sons. pp 189-209. + Create a budget using an Excel spreadsheet for the same simulated -Johnson, L.R. Ben-Horin, T., Lafferty, K.D., McNally, A., Mordecai, E., Paaijmans, K.P., Pawar, S., Ryan, -S.J. 2015. Understanding uncertainty in temperature effects on vector-borne disease: A Bayesian -approach. Ecology 96(1): 203-213. +family situation. -Kalis, M., and E. Curtiss. 2016. CDC's Drought Guidance: Your Public Health Resource for Understanding + Compile a journal (minimum of 2 pages) of entries that reflect on +the customized budget, the process of creating the budget, and +your analysis of how you will what you learned about budgeting to +reach your goals. -and Preparing for Drought in Your Community. J Environ Health 78:34-35. +2. - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 4 9 +Competency: Evaluate asset management components -Kalis, M. A., M. D. Miller, and R. J. Wilson. 2009. Public health and drought. J Environ Health 72:10-11. -Keatinge, W. R., G. C. Donaldson, K. Bucher, G. Jendritzky, E. Cordioli, M. Martinelli, K. Katsouyanni, -A. E. Kunst, C. McDonald, S. Nayha, I. Vuori, and G. Eurowinter. 2000a. Winter mortality in relation -to climate. Int J Circumpolar Health 59:154-159. +Objectives: -Keatinge, W. R., G. C. Donaldson, E. Cordioli, M. Martinelli, A. E. Kunst, J. P. Mackenbach, S. Nayha, -and I. Vuori. 2000b. Heat related mortality in warm and cold regions of Europe: observational study. -BMJ 321:670-673. + Define bank account reconciliation -Kirshen, P., C. Watson, E. Douglas, A. Gontz, J. Lee, and Y. Tian. 2008. Coastal flooding in the -Northeastern United States due to climate change. Mitigation and Adaptation Strategies for Global -Change 13:437-451. + -Kistemann, T., T. Classen, C. Koch, F. Dangendorf, R. Fischeder, J. Gebel, V. Vacata, and M. Exner. 2002. -Microbial load of drinking water reservoir tributaries during extreme rainfall and runoff. Appl Environ -Microbiol 68:2188-2197. + -Klinger, C., O. Landeg, and V. Murray. 2014. Power outages, extreme events and health: a systematic + -review of the literature from 2011-2012. PLoS Curr 6. + -Knight, D. B., and R. E. Davis. 2009. Contribution of tropical cyclones to extreme rainfall events in the + -southeastern United States. Journal of Geophysical Research-Atmospheres 114. +Identify the purpose for reconciling a bank account -Knowlton, K., M. Rotkin-Ellman, G. King, H. G. Margolis, D. Smith, G. Solomon, R. Trent, and P. English. -2009. The 2006 California heat wave: impacts on hospitalizations and emergency department visits. -Environ Health Perspect 117:61-67. +Identify the steps for completing a bank account reconciliation -Knutson, T. R., J. J. Sirutis, G. A. Vecchi, S. Garner, M. Zhao, H. S. Kim, M. Bender, R. E. Tuleya, I. M. -Held, and G. Villarini. 2013. Dynamical Downscaling Projections of Twenty-First-Century Atlantic -Hurricane Activity: CMIP3 and CMIP5 Model-Based Scenarios. Journal of Climate 26:6591-6617. -Kochtubajda, B., M. D. Flannigan, J. R. Gyakum, R. E. Stewart, K. A. Logan, and T. V. Nguyen. 2006. -Lightning and fires in the Northwest Territories and responses to future climate change. Arctic 59:211- -221. +Identify cash management products -Kovats, R. S. 2000. El Nino and human health. Bull World Health Organ 78:1127-1135. -Kovats, R. S., M. J. Bouma, S. Hajat, E. Worrall, and A. Haines. 2003. El Nino and health. Lancet 362:1481- +Identify the costs of home ownership -1489. +Identify the costs of car ownership -Kovats, R. S., and S. Hajat. 2008. Heat stress and public health: a critical review. Annu Rev Public Health + Apply the time value of money concepts -29:41-55. + Analyze a personal situation to establish a cash and savings -Kovats, R. S., S. Hajat, and P. Wilkinson. 2004. Contrasting patterns of mortality and hospital admissions +program -during hot weather and heat waves in Greater London, UK. Occup Environ Med 61:893-898. + Apply the cost/benefit analysis of owning versus renting a home to -La Greca, A. M., B. S. Lai, M. M. Llabre, W. K. Silverman, E. M. Vernberg, and M. J. Prinstein. 2013. -Children's Postdisaster Trajectories of PTS Symptoms: Predicting Chronic Distress. Child Youth Care -Forum 42:351-369. +a personal financial position -La Greca, A. M., W. K. Silverman, B. Lai, and J. Jaccard. 2010. Hurricane-Related Exposure Experiences -and Stressors, Other Life Events, and Social Support: Concurrent and Prospective Impact on Children's -Persistent Posttraumatic Stress Symptoms. Journal of Consulting and Clinical Psychology 78:794-805. -Laursen, E., O. Mygind, B. Rasmussen, and T. Ronne. 1994. Gastroenteritis: a waterborne outbreak + Calculate the amount of house a person can afford based on 30% -affecting 1600 people in a small Danish town. J Epidemiol Community Health 48:453-458. +of income -Li, C. F., T. W. Lim, L. L. Han, and R. Fang. 1985. Rainfall, abundance of Aedes aegypti and dengue + Apply the principle of living within your means. -infection in Selangor, Malaysia. Southeast Asian J Trop Med Public Health 16:560-568. +Assessments: -Liang, S., E. Y. Seto, J. V. Remais, B. Zhong, C. Yang, A. Hubbard, G. M. Davis, X. Gu, D. Qiu, and R. -C. Spear. 2007. Environmental effects on parasitic disease transmission exemplified by schistosomiasis -in western China. Proc Natl Acad Sci U S A 104:7110-7115. + Complete a bank account reconciliation -Lin, S., W. H. Hsu, A. R. Van Zutphen, S. Saha, G. Luber, and S. A. Hwang. 2012. Excessive heat and -respiratory hospitalizations in New York State: estimating current and future public health burden -related to climate change. Environ Health Perspect 120:1571-1577. + Develop a personal savings plan -Lindblade, K. A., E. D. Walker, A. W. Onapa, J. Katungu, and M. L. Wilson. 2000. Land use change alters -malaria transmission parameters by modifying temperature in a highland area of Uganda. Tropical -Medicine & International Health 5:263-274. + Analyze a home purchasing experience -Lipp, E. K., A. Huq, and R. R. Colwell. 2002. Effects of global climate on infectious disease: the cholera +Criteria. You will meet the objectives listed above through a -model. Clin Microbiol Rev 15:757-770. +Page 4 -Loevinsohn, M. E. 1994. Climatic warming and increased malaria incidence in Rwanda. Lancet 343:714- + Personal Finance -718. +Syllabus - 1 5 0 • S O N G L I A N G E T A L . +combination of the following activities in this course: -Luque Fernandez, M. A., A. Bauernfeind, J. D. Jimenez, C. L. Gil, N. El Omeiri, and D. H. Guibert. 2009. -Influence of temperature and rainfall on the evolution of cholera epidemics in Lusaka, Zambia, 2003- -2006: analysis of a time series. Trans R Soc Trop Med Hyg 103:137-143. + Describe the various parts of a bank reconciliation and what -Lyle, T. S., and T. Mills. 2016. Assessing coastal flood risk in a changing climate for the City of Vancouver. +information goes in each part of the reconciliation. -Canadian Water Resources Journal 41:343-352. + -Malmstadt, J., K. Scheltin, and J. Elsner. 2009. Florida hurricanes and damage costs. Southeastern +Identify the reasons for the difference between the ending bank +balance and the ending checkbook balance and correctly adjust +each balance. -Geographer 49:108-131. + Successfully complete a reconciliation for a simulated scenario. -Martinez, C. J., J. J. Maleski, and M. F. Miller. 2012. Trends in precipitation and temperature in Florida, + -USA. Journal of Hydrology 452:259-281. + -McMichael, A. J., and A. Haines. 1997. Global climate change: the potential effects on health. BMJ +Identify steps to take to reach the desired amount of liquid assets +in a portfolio. -315:805-809. +Identify the type of account to be used to help reach the savings +goals and the interest rates on each account -Medina-Ramon, M., A. Zanobetti, D. P. Cavanagh, and J. Schwartz. 2006. Extreme temperatures and -mortality: assessing effect modification by personal characteristics and specific cause of death in a -multi-city case-only analysis. Environ Health Perspect 114:1331-1336. + Differentiate the advantages and disadvantages between owning +and renting a home, and locate an advertisement for a house for +sale as well as a home for rent -Medlock, J. M., and S. A. Leach. 2015. Effect of climate change on vector-borne disease risk in the UK. + Analyze whether a monthly mortgage or rent “fits” into a current -Lancet Infect Dis 15:721-730. +financial situation. -Mhalu, F. S., W. M. Mntenga, and F. D. Mtango. 1987. Seasonality of cholera in Tanzania: possible role of +3. -rainfall in disease transmission. East Afr Med J 64:378-387. +Competency: Evaluate credit management components -Miettinen, I. T., O. Zacheus, C. H. von Bonsdorff, and T. Vartiainen. 2001. Waterborne epidemics in +Objectives: -Finland in 1998-1999. Water Sci Technol 43:67-71. -Miranda, J. J. 2004. El Nino and health. Lancet 363:247-248. -Mordecai, E.A., Cohen, J., Evans, M.V., Gudapati, P., Johnson, L.R., Lippi, C.A., Miazgowicz, K., -Murdock, C.C., Rohr, J.R., Ryan, S.J., Savage, V., Shocket, M., Stewart Ibarra, A.M., Thomas, M.B., -Weikel, D.P. 2017. Detecting the impact of temperature on transmission of Zika, dengue, and -chikungunya using mechanistic models. PLOS NTDs. 11(4): e0005568 + Explore the various types of debt -Mordecai, E.A., Paaijmans, K.P., Johnson, L.R., Balzer, C.H., Ben-Horin, T., deMoor, E., McNally, A., -Pawar, S., Ryan, S.J., Smith, T.C., Lafferty, K.D., 2013. Physiological constraints dramatically lower -the expected temperature for peak malaria transmission. Ecology Letters 16(1):22-30 + -Neria, Y., and J. M. Shultz. 2012. Mental health effects of Hurricane Sandy: characteristics, potential +Identify the cost of using credit -aftermath, and response. JAMA 308:2571-2572. + Apply the time value of money concepts -Ngwa, M. C., S. Liang, I. T. Kracalik, L. Morris, J. K. Blackburn, L. M. Mbam, S. F. Ba Pouth, A. Teboh, -Y. Yang, M. Arabi, J. D. Sugimoto, and J. G. Morris, Jr. 2016. Cholera in Cameroon, 2000-2012: Spatial -and Temporal Analysis at the Operational (Health District) and Sub Climate Levels. PLoS Negl Trop -Dis 10:e0005105. + Calculate the interest costs of using credit -Nichols, G., C. Lane, N. Asgari, N. Q. Verlander, and A. Charlett. 2009. Rainfall and outbreaks of drinking + Prepare a loan amortization schedule -water related disease and in England and Wales. J Water Health 7:1-8. + Apply the principle of living within your means -Ogden, N. H., A. Maarouf, I. K. Barker, M. Bigras-Poulin, L. R. Lindsay, M. G. Morshed, C. J. O'Callaghan, -F. Ramay, D. Waltner-Toews, and D. F. Charron. 2006. Climate change and the potential for range -expansion of the Lyme disease vector Ixodes scapularis in Canada. International Journal for -Parasitology 36:63-70. +Assessments: -Ostfeld, R. S., C. D. Canham, K. Oggenfuss, R. J. Winchcombe, and F. Keesing. 2006. Climate, deer, -rodents, and acorns as determinants of variation in Lyme-disease risk. Plos Biology 4:1058-1068. -Paaijmans, K. P., S. Blanford, A. S. Bell, J. I. Blanford, A. F. Read, and M. B. Thomas. 2010. Influence of -climate on malaria transmission depends on daily temperature variation. Proc Natl Acad Sci U S A -107:15135-15139. + Complete a personal credit usage reflection -Polymenakou, P. N., M. Mandalakis, E. G. Stephanou, and A. Tselepides. 2008. Particle size distribution -of airborne microorganisms and pathogens during an Intense African dust event in the eastern -Mediterranean. Environ Health Perspect 116:292-296. + Evaluate a consumer loan scenario -Price, C., and D. Rind. 1994. Possible Implications of Global Climate-Change on Global Lightning +Criteria. You will meet the objectives listed above through a +combination of the following activities in this course: -Distributions and Frequencies. Journal of Geophysical Research-Atmospheres 99:10823-10831. + Distinguish different forms of open account credit -Pruss-Ustun, A., J. Bartram, T. Clasen, J. M. Colford, O. Cumming, V. Curtis, S. Bonjour, A. D. Dangour, -J. De France, L. Fewtrell, M. C. Freeman, B. Gordon, P. R. Hunter, R. B. Johnston, C. Mathers, D. -Mausezahl, K. Medlicott, M. Neira, M. Stocks, J. Wolf, and S. Cairncross. 2014. Burden of disease -from inadequate water, sanitation and hygiene in low- and middle-income settings: a retrospective -analysis of data from 145 countries. Tropical Medicine & International Health 19:894-905. + - C L I M A T E C H A N G E I M P A C T S O N H U M A N H E A L T H • 1 5 1 +Identify the risks and benefits of using credit -Pruss, A., D. Kay, L. Fewtrell, and J. Bartram. 2002. Estimating the burden of disease from water, sanitation, + Consider the level of personal credit usage that “fits” into a -and hygiene at a global level. Environmental Health Perspectives 110:537-542. +personal financial position to avoid credit problems -Qian, Z. M., Q. C. He, H. M. Lin, L. L. Kong, C. M. Bentley, W. S. Liu, and D. J. Zhou. 2008. High -temperatures enhanced acute mortality effects of ambient particle pollution in the "Oven" city of -Wuhan, China. Environmental Health Perspectives 116:1172-1178. + -Ragan, P., J. Schulte, S. J. Nelson, and K. T. Jones. 2008. Mortality surveillance: 2004 to 2005 Florida + -hurricane-related deaths. Am J Forensic Med Pathol 29:148-153. +Identify the difference between major types of consumer loans -Reeve, N., and R. Toumi. 1999. Lightning activity as an indicator of climate change. Quarterly Journal of +Identify the sources of various consumer loans -the Royal Meteorological Society 125:893-903. +Page 5 -Ritenour, A. E., M. J. Morton, J. G. McManus, D. J. Barillo, and L. C. Cancio. 2008. Lightning injury: a + Personal Finance -review. Burns 34:585-594. +Syllabus -Ryan, S.J., McNally, A., Johnson, L.R., Mordecai, E.A., Ben-Horin, T., Paaijmans, K.P., Lafferty, K.D. -2015. Mapping physiological suitability limits of malaria in Africa under climate change. Journal of -vector borne and zoonotic diseases 15(12): 817-725. + Calculate the finance charges on various types of loans -Ryan, S.J., Ben-Horin, T., Johnson, L.R. 2015. Malaria control and senescence: the importance of + Given a scenario, identify the appropriate loan to recommend -accounting for the pace and shape of ageing in wild mosquitoes. Ecosphere 6(9):170. +4. -Ryan, S.J. et al. 2017. Climate change drives uncertain global shifts in potential distribution and seasonal +Competency: Complete an insurance needs analysis -risk of Aedes-transmitted viruses. BioRXiv doi: https://doi.org/10.1101/172221. +Objectives: -Sarofim, M. C., S. Saha, M. D. Hawkins, D. M. Mills, J. Hess, R. Horton, P. Kinney, J. Schwartz, and A. -St. Juliana. 2016. Ch. 2: Temperature-Related Death and Illness. Pages 43-68 The Impacts of Climate -Change on Human Health in the United States: A + Define insurance terms -Scientific Assessment. U.S. Global Change Research Program, Washington, DC. -Schijven, J. F., and A. M. de Roda Husman. 2005. Effect of climate changes on waterborne disease in The + Explore the different types of life, health, disability, automobile, and -Netherlands. Water Sci Technol 51:79-87. +property insurance -Schreider, S. Y., D. I. Smith, and A. J. Jakeman. 2000. Climate change impacts on urban flooding. Climatic + -Change 47:91-115. + -Schuster, C. J., A. G. Ellis, W. J. Robertson, D. F. Charron, J. J. Aramini, B. J. Marshall, and D. T. Medeiros. -2005. Infectious disease outbreaks related to drinking water in Canada, 1974-2001. Can J Public Health -96:254-258. +Identify components of lifestyle choices and health history that +impact insurance costs -Senhorst, H. A., and J. J. Zwolsman. 2005. Climate change and effects on water quality: a first impression. +Identify the coverage from different types of insurance – life, +health, disability, automobile, and property insurance -Water Sci Technol 51:53-59. + Calculate the amount of insurance needed to cover life, health, and -Shaman, J., J. F. Day, and M. Stieglitz. 2005. Drought-induced amplification and epidemic transmission of +property -West Nile virus in southern Florida. J Med Entomol 42:134-141. +Assessment: -Singh, R. B., S. Hales, N. de Wet, R. Raj, M. Hearnden, and P. Weinstein. 2001. The influence of climate -variation and change on diarrheal disease in the Pacific Islands. Environ Health Perspect 109:155-159. -Soverow, J. E., G. A. Wellenius, D. N. Fisman, and M. A. Mittleman. 2009. Infectious Disease in a -Warming World: How Weather Influenced West Nile Virus in the United States (2001-2005). -Environmental Health Perspectives 117:1049-1052. + Complete an insurance needs analysis -Su, G. L. S. 2008. Correlation of climatic factors and dengue incidence in Metro Manila, Philippines. Ambio +Criteria. You will meet the objectives listed above through a +combination of the following activities in this course: -37:292-294. + -Sullivent, E. E., 3rd, C. A. West, R. S. Noe, K. E. Thomas, L. J. Wallace, and R. T. Leeb. 2006. Nonfatal +Identify the different types of life insurance -injuries following Hurricane Katrina--New Orleans, Louisiana, 2005. J Safety Res 37:213-217. + Consider the advantages and disadvantages of the various types of -Taylor, G., and W. J. Davies. 1990. Root-Growth of Fagus-Sylvatica - Impact of Air-Quality and Drought +life insurance -at a Site in Southern Britain. New Phytologist 116:457-464. + Calculate the amount of life insurance needed -Thomas, K. M., D. F. Charron, D. Waltner-Toews, C. Schuster, A. R. Maarouf, and J. D. Holt. 2006. A role -of high impact weather events in waterborne disease outbreaks in Canada, 1975 - 2001. Int J Environ -Health Res 16:167-180. + -Thompson, K. R., N. B. Bernier, and P. Chan. 2009. Extreme sea levels, coastal flooding and climate change +Identify the various types of health insurance plans -with a focus on Atlantic Canada. Natural Hazards 51:139-150. + Consider the features of disability income insurance and how to -Thomson, M. C., S. J. Mason, T. Phindela, and S. J. Connor. 2005. Use of rainfall and sea surface -temperature monitoring for malaria early warning in Botswana. American Journal of Tropical Medicine -and Hygiene 73:214-221. +determine the need for this insurance -Tian, L., Y. Bi, S. C. Ho, W. Liu, S. Liang, W. B. Goggins, E. Y. Chan, S. Zhou, and J. J. Sung. 2008. One- -year delayed effect of fog on malaria transmission: a time-series analysis in the rain forest area of -Mengla County, south-west China. Malar J 7:110. + -USEPA. 2016. Climate change indicators in the United States. Page 96 in U. S. E. P. Agency, editor. +Identify the various types of property and liability insurance - 1 5 2 • S O N G L I A N G E T A L . + Calculate the coverage provided by various insurance policies -van der Hoek, W., F. Konradsen, D. Perera, P. H. Amerasinghe, and F. P. Amerasinghe. 1997. Correlation -between rainfall and malaria in the dry zone of Sri Lanka. Annals of Tropical Medicine and Parasitology -91:945-949. +5. -Vernberg, E. M., A. M. LaGreca, W. K. Silverman, and M. J. Prinstein. 1996. Prediction of posttraumatic -stress symptoms in children after Hurricane Andrew. Journal of Abnormal Psychology 105:237-248. -Wardrop, N. A., A. G. Barnett, J. A. Atkinson, and A. C. A. Clements. 2013. Plasmodium vivax malaria -incidence over time and its association with temperature and rainfall in four counties of Yunnan -Province, China. Malaria Journal 12. +Competency: Evaluate investment management strategies -Weems, C. F., L. K. Taylor, M. F. Cannon, R. C. Marino, D. M. Romano, B. G. Scott, A. M. Perry, and V. -Triplett. 2010. Post Traumatic Stress, Context, and the Lingering Effects of the Hurricane Katrina -Disaster among Ethnic Minority Youth. Journal of Abnormal Child Psychology 38:49-56. +Objectives: -WHO. 2008. Global Burden of Disease 2004 update. World Health Organization. -Winsberg, M. D. 2003. Florida weather. 2nd edition. University Press of Florida, Gainesville, Fla. -Yang, K., J. LeJeune, D. Alsdorf, B. Lu, C. K. Shum, and S. Liang. 2012. Global distribution of outbreaks + Describe how an investing strategy can be incorporated into any -of water-associated infectious diseases. PLoS Negl Trop Dis 6:e1483. +lifestyle + + Explain the difference between primary and secondary markets as + +well as between broker and dealer markets + + Describe the process of buying and selling securities + + Explore the impact that technology has on personal investment + +portfolios + +Page 6 + + Personal Finance + +Syllabus + + Examine the various types of risk assumed by investors + + Correlate the level of risk with the level of return on an investment + + Explore the different types of stocks and bonds + + Examine performance measurements of stocks + + Calculate different measures of yield on bonds + + Describe the basic features and operating characteristics of a + +mutual fund + + Explain the different types of mutual funds and the variables that + +should be considered when selecting a fund + + Compute the rate of return earned on an investment in a mutual + +fund + + Explore the role that real estate can play in a diversified investment + +portfolio + +Assessments: + + Complete an investment plan + +Criteria. You will meet the objectives listed above through a +combination of the following activities in this course: + + + +Identify the components to an investment portfolio that would help +reach financial goals + + + +Identify the risk associated with investing + + Distinguish the different types of stocks and bonds and their + +characteristics + + Consider the advantages and disadvantages of stocks and bonds + + Explore how mutual funds can help diversify a portfolio + + Consider the role that real estate can play in a well-diversified + +portfolio + +6. + +Competency: Explore retirement and estate planning +strategies + +Objectives: + + Recognize the importance of retirement planning + + + +Identify the biggest pitfalls to good planning + + Estimate your income needs in retirement and your retirement + +income + + Explore eligibility requirements and benefits of the Social Security + +program + +Page 7 + + Personal Finance + +Syllabus + + Explore various pension plans + + Describe the various types of annuities + + Describe the role of estate planning in personal financial planning + + Recognize the importance of preparing a will and other documents + +to protect you and your estate + + Explore trusts and their use in estate planning + + Describe effective estate planning techniques to minimize estate + +taxes + +Assessments: + + Complete a personal retirement plan + +Criteria. You will meet the objectives listed above through a +combination of the following activities in this course: + + Understand the need to start planning for retirement before age 20 + + Consider a person’s financial needs during retirement + + Calculate the income needed during your retirement + + + +Identify which annuity is right for a retirement plan + + Consider the steps a person needs to take to protect an estate + +Page 8 + + Personal Finance + +Syllabus + +Part 3: Course Outline/Schedule + + Week 1: Importance and Relevancy of Personal Financial + +Management; Setting Financial Goals + +o Why Study Personal Finance?; iGrad College Student Video; + +“Personal Finance Intro” PowerPoint (1 day) + +o Pre-Test to Personal Finance Concepts: Jeopardy Game (1 + +day) + +o Key Components of Financial Planning: Dave Ramsey + +Chapter 1 and Video (1 days) + +o Setting Financial Goals: Dave Ramsey Chapter 11 and Video + +(2 days) + + Weeks 2-4: Pay, Benefits, and Working Conditions + +o Gross Pay, Benefits, Incentives, Deductions, Net Pay: + +Chapter 6.1; Module 4 (6 days) + +o Work Arrangements, Organizations, and Pensions: Chapter + +6.2 (3 days) + +o Summative Exam + +o Filing Federal and State Income Taxes: Chapter 7, IRS + +Website, Wisconsin Department of Revenue Website; Tax +Forms; Module 1 (6 days) + +o Summative Assessment + + Weeks 5-7: Budgeting + +o Budgeting Terminology, Personal Records, and Personal + +Property Inventories: Chapter 8.1 + +o Life in the United States Budgeting Simulation (University of + +Arizona – FEFE). Simulate: (10 days) + + Purchasing and Financing a Car, Module 2 + + Renting vs. Owning a Home; Homeowners’ and + +Renters’ Insurance, Module 2 + + Health, Life, and Disability Insurance, Module 2; Dave + +Ramsey Video and Chapter 9 + + Utilities and Living Expenses + + Costs of Raising Children + +o Zero-Based Budgeting: Dave Ramsey Video and Chapter 3 + +(2 days) + +o Budgeting Reflection Activity (1 day) + +o Summative Assessment + +Page 9 + + Personal Finance + +Syllabus + + Weeks 8-10: Banking + +o Banking Services, Parties to a Check, Types of Accounts, + +Opening Accounts: Security First Banking Simulation Lesson +1; Module 1 (2 days) + +o Using Checking Accounts, Check Registers, Writing Checks, + +Endorsements, Fees, Reconciling Bank Statements: Security +First Banking Simulation Lesson 2; Module 1 (6 days) + +o Computing Simple Interest: Practical Money Skills + +Worksheets (1 day) + +o Simulating Bank Transactions by Recording, Writing Checks, + +Depositing, Using Debit Cards and Reconciling Bank +Statements for Two Months of Transactions: Security First +Banking Simulation Lesson 3 (6 days) + +o Summative Assessment + + Weeks 11-13: Credit Management + +o Credit in America: Power Point “The Good Life” from + +Consumer Jungle; Chapter 16.1 (1 day) + +o Credit Cards: Practice Money Skills Worksheets (2 days) + +o Credit Card Statements: Take Charge America (FEFE) (1 + +day) + +o Credit Records, Laws, and Bureaus: Chapter 17.2; Take + +Charge America (FEFE); Practical Money Skills Worksheets (3 +days) + +o Credit Worthiness: Chapter 17.1 (1 day) + +o Establishing a Good Credit History and Advantages of Credit: +Power Point “Credit Wisdom” from Practice Money Skills; +Chapter 17.2; Practice Money Skills Worksheets (3 days) + +o Shopping for a Credit Card: Practical Money Skills Handouts + +(1 day) + +o Problems with Credit and Disadvantages of Credit: Chapter + +19; Dave Ramsey Video and Chapter 4 (3 days) + +o Pay Day Loans: Take Charge America (FEFE) (1 day) + +o Summative Assessment + + Weeks 14-16: Saving, Investing and Planning for Retirement + +o Saving vs. Investing: Chapter 10; Dave Ramsey Videos and + +Chapter 8 (2 days) + +o Compounding Interest, Time Value of Money, Inflation, and + +Rule of 72: Module 3; Take Charge America (FEFE); Practical +Money Skills; Power Point – Rule of 72s + +Page 10 + + Personal Finance + +Syllabus + +o Stocks: Chapter 12; Module 3; How the Market Works Stock + +Simulation (3 days) + +o Bonds: Chapter 13; Module 3 (3 days) + +o Mutual Funds: Chapter 14; Module 3 (3 days) + +o Retirement Planning: Chapter 15; Module 4 (3 days) + +o Summative Assessment + + Weeks 17-18: Identity Theft and Protecting Yourself; + +Consumer Awareness + +o Consumer Awareness: Dave Ramsey Videos, Chapter 6, and + +Worksheets (1 day) + +o Bargain Shopping: Dave Ramsey Video and Chapter 7 + +o Money and Relationships: Dave Ramsey Videos, Chapter 10, + +and Worksheets (1 days) + +o Identity Theft and Consumer Scams: Teacher Power Point; +Students Create Prezi Describing the Scam and Tips for +Preventing It (3 days) + + Week 18: Review and Final Exam + +Page 11